Skip to contentSkip to conversation

perf stat: Ensure metrics are displayed even with failed events

PATCH Applied Chun-Tse Shao 3 participants 6 messages Feb 3 – Feb 8 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf stat: Ensure metrics are displayed even with failed events

Currently, perf stat skips or hides metrics when the underlying hardware events cannot be counted (e.g., due to insufficient permissions or unsupported events). In --metric-only mode, this often results in missing columns or blank spaces, making the output difficult to parse.

Modify the logic to ensure metrics are consistently displayed by propagating NAN (Not a Number) through the expression evaluator. Specifically:

1. Update prepare_metric() in stat-shadow.c to treat uncounted events (where run == 0) as NAN. This leverages the existing math in expr.y to propagate NAN through metric expressions.
2. Remove the early return in the display logic's printout() function that was previously skipping metrics in --metric-only mode for failed events.
3. Simplify perf_stat__skip_metric_event() to no longer depend on event runtime.

Tested:
1. perf all metrics test did not crash while paranoid is 2.
2. Multiple combinations with CPUs_utilized while paranoid is 2.

  $ ./perf stat -M CPUs_utilized -a -- sleep 1

Performance counter stats for 'system wide':

     <not supported> msec cpu-clock:u                      #      nan CPUs  CPUs_utilized
       1,006,356,120      duration_time

         1.004375550 seconds time elapsed
  $ ./perf stat -M CPUs_utilized -a -j -- sleep 1
  {"counter-value" : "<not supported>", "unit" : "msec", "event" : "cpu-clock:u", "event-runtime" : 0, "pcnt-running" : 100.00, "metric-value" : "nan", "metric-unit" : "CPUs  CPUs_utilized"}
  {"counter-value" : "1006642462.000000", "unit" : "", "event" : "duration_time", "event-runtime" : 1, "pcnt-running" : 100.00}
  $ ./perf stat -M CPUs_utilized -a --metric-only -- sleep 1

Performance counter stats for 'system wide':

    CPUs  CPUs_utilized
                      nan

         1.004424652 seconds time elapsed
  $ ./perf stat -M CPUs_utilized -a --metric-only -j -- sleep 1
  {"CPUs  CPUs_utilized" : "none"}
Signed-off-by [email protected]
---
 tools/perf/util/stat-display.c | 59 +++++++++++++++-------------------
 tools/perf/util/stat-shadow.c  |  8 ++---
 tools/perf/util/stat.h         |  2 +-
 3 files changed, 29 insertions(+), 40 deletions(-)
tools/perf/util/stat-display.c +26 -33
index 2ce0602974a1..dc2b66855f6c 100644
@@ -820,12 +820,6 @@ static void printout(struct perf_stat_config *config, struct outstate *os,
 	}
 
 	if (run == 0 || ena == 0 || counter->counts->scaled == -1) {
-		if (config->metric_only) {
-			pm(config, os, METRIC_THRESHOLD_UNKNOWN, /*format=*/NULL,
-			   /*unit=*/NULL, /*val=*/0);
-			return;
-		}
-
 		ok = false;
 
 		if (counter->supported) {
@@ -848,33 +842,32 @@ static void printout(struct perf_stat_config *config, struct outstate *os,
 		print_running(config, os, run, ena, /*before_metric=*/true);
 	}
 
-	if (ok) {
-		if (!config->metric_only && counter->default_metricgroup && !counter->default_show_events) {
-			void *from = NULL;
-
-			aggr_printout(config, os, os->evsel, os->id, os->aggr_nr);
-			/* Print out all the metricgroup with the same metric event. */
-			do {
-				int num = 0;
-
-				/* Print out the new line for the next new metricgroup. */
-				if (from) {
-					if (config->json_output)
-						new_line_json(config, (void *)os);
-					else
-						__new_line_std_csv(config, os);
-				}
-
-				print_noise(config, os, counter, noise, /*before_metric=*/true);
-				print_running(config, os, run, ena, /*before_metric=*/true);
-				from = perf_stat__print_shadow_stats_metricgroup(config, counter, aggr_idx,
-										 &num, from, &out);
-			} while (from != NULL);
-		} else {
-			perf_stat__print_shadow_stats(config, counter, aggr_idx, &out);
-		}
+	if (!config->metric_only && counter->default_metricgroup &&
+	    !counter->default_show_events) {
+		void *from = NULL;
+
+		aggr_printout(config, os, os->evsel, os->id, os->aggr_nr);
+		/* Print out all the metricgroup with the same metric event. */
+		do {
+			int num = 0;
+
+			/* Print out the new line for the next new metricgroup. */
+			if (from) {
+				if (config->json_output)
+					new_line_json(config, (void *)os);
+				else
+					__new_line_std_csv(config, os);
+			}
+
+			print_noise(config, os, counter, noise,
+				    /*before_metric=*/true);
+			print_running(config, os, run, ena,
+				      /*before_metric=*/true);
+			from = perf_stat__print_shadow_stats_metricgroup(
+				config, counter, aggr_idx, &num, from, &out);
+		} while (from != NULL);
 	} else {
-		pm(config, os, METRIC_THRESHOLD_UNKNOWN, /*format=*/NULL, /*unit=*/NULL, /*val=*/0);
+		perf_stat__print_shadow_stats(config, counter, aggr_idx, &out);
 	}
 
 	if (!config->metric_only) {
@@ -987,7 +980,7 @@ static void print_counter_aggrdata(struct perf_stat_config *config,
 	ena = aggr->counts.ena;
 	run = aggr->counts.run;
 
-	if (perf_stat__skip_metric_event(counter, ena, run))
+	if (perf_stat__skip_metric_event(counter))
 		return;
 
 	if (val == 0 && should_skip_zero_counter(config, counter, &id))
tools/perf/util/stat-shadow.c +2 -6
index 9c83f7d96caa..5d8d09e0e6ae 100644
@@ -83,7 +83,7 @@ static int prepare_metric(struct perf_stat_config *config,
 		}
 		/* Time events are always on CPU0, the first aggregation index. */
 		aggr = &ps->aggr[is_tool_time ? tool_aggr_idx : aggr_idx];
-		if (!aggr || !metric_events[i]->supported) {
+		if (!aggr || !metric_events[i]->supported || aggr->counts.run == 0) {
 			/*
 			 * Not supported events will have a count of 0, which
 			 * can be confusing in a metric. Explicitly set the
@@ -335,14 +335,10 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
  * perf_stat__skip_metric_event - Skip the evsel in the Default metricgroup,
  *				  if it's not running or not the metric event.
  */
-bool perf_stat__skip_metric_event(struct evsel *evsel,
-				  u64 ena, u64 run)
+bool perf_stat__skip_metric_event(struct evsel *evsel)
 {
 	if (!evsel->default_metricgroup)
 		return false;
 
-	if (!ena || !run)
-		return true;
-
 	return !metricgroup__lookup(&evsel->evlist->metric_events, evsel, false);
 }
tools/perf/util/stat.h +1 -1
index f986911c9296..4bced233d2fc 100644
@@ -163,7 +163,7 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
 				   struct evsel *evsel,
 				   int aggr_idx,
 				   struct perf_stat_output_ctx *out);
-bool perf_stat__skip_metric_event(struct evsel *evsel, u64 ena, u64 run);
+bool perf_stat__skip_metric_event(struct evsel *evsel);
 void *perf_stat__print_shadow_stats_metricgroup(struct perf_stat_config *config,
 						struct evsel *evsel,
 						int aggr_idx,
Signature · 1 line
2.53.0.rc2.204.g2597b5adb4-goog
reviewer Reviewed-by on v1 1/1
#1 lore raw
Quoted 45 lines · Chun-Tse Shao

Currently, perf stat skips or hides metrics when the underlying hardware events cannot be counted (e.g., due to insufficient permissions or unsupported events). In --metric-only mode, this often results in missing columns or blank spaces, making the output difficult to parse.

Modify the logic to ensure metrics are consistently displayed by propagating NAN (Not a Number) through the expression evaluator. Specifically:

1. Update prepare_metric() in stat-shadow.c to treat uncounted events (where run == 0) as NAN. This leverages the existing math in expr.y to propagate NAN through metric expressions.
2. Remove the early return in the display logic's printout() function that was previously skipping metrics in --metric-only mode for failed events.
3. Simplify perf_stat__skip_metric_event() to no longer depend on event runtime.

Tested:
1. perf all metrics test did not crash while paranoid is 2.
2. Multiple combinations with CPUs_utilized while paranoid is 2.

  $ ./perf stat -M CPUs_utilized -a -- sleep 1

Performance counter stats for 'system wide':

     <not supported> msec cpu-clock:u                      #      nan CPUs  CPUs_utilized
       1,006,356,120      duration_time

         1.004375550 seconds time elapsed
  $ ./perf stat -M CPUs_utilized -a -j -- sleep 1
  {"counter-value" : "<not supported>", "unit" : "msec", "event" : "cpu-clock:u", "event-runtime" : 0, "pcnt-running" : 100.00, "metric-value" : "nan", "metric-unit" : "CPUs  CPUs_utilized"}
  {"counter-value" : "1006642462.000000", "unit" : "", "event" : "duration_time", "event-runtime" : 1, "pcnt-running" : 100.00}
  $ ./perf stat -M CPUs_utilized -a --metric-only -- sleep 1

Performance counter stats for 'system wide':

    CPUs  CPUs_utilized
                      nan

         1.004424652 seconds time elapsed
Quoted 4 lines
  $ ./perf stat -M CPUs_utilized -a --metric-only -j -- sleep 1
  {"CPUs  CPUs_utilized" : "none"}
Signed-off-by [email protected]
Quoted 96 lines
---
 tools/perf/util/stat-display.c | 59 +++++++++++++++-------------------
 tools/perf/util/stat-shadow.c  |  8 ++---
 tools/perf/util/stat.h         |  2 +-
 3 files changed, 29 insertions(+), 40 deletions(-)
tools/perf/util/stat-display.c +26 -33
index 2ce0602974a1..dc2b66855f6c 100644
@@ -820,12 +820,6 @@ static void printout(struct perf_stat_config *config, struct outstate *os,
        }
 
        if (run == 0 || ena == 0 || counter->counts->scaled == -1) {
-               if (config->metric_only) {
-                       pm(config, os, METRIC_THRESHOLD_UNKNOWN, /*format=*/NULL,
-                          /*unit=*/NULL, /*val=*/0);
-                       return;
-               }
-
                ok = false;
 
                if (counter->supported) {
@@ -848,33 +842,32 @@ static void printout(struct perf_stat_config *config, struct outstate *os,
                print_running(config, os, run, ena, /*before_metric=*/true);
        }
 
-       if (ok) {
-               if (!config->metric_only && counter->default_metricgroup && !counter->default_show_events) {
-                       void *from = NULL;
-
-                       aggr_printout(config, os, os->evsel, os->id, os->aggr_nr);
-                       /* Print out all the metricgroup with the same metric event. */
-                       do {
-                               int num = 0;
-
-                               /* Print out the new line for the next new metricgroup. */
-                               if (from) {
-                                       if (config->json_output)
-                                               new_line_json(config, (void *)os);
-                                       else
-                                               __new_line_std_csv(config, os);
-                               }
-
-                               print_noise(config, os, counter, noise, /*before_metric=*/true);
-                               print_running(config, os, run, ena, /*before_metric=*/true);
-                               from = perf_stat__print_shadow_stats_metricgroup(config, counter, aggr_idx,
-                                                                                &num, from, &out);
-                       } while (from != NULL);
-               } else {
-                       perf_stat__print_shadow_stats(config, counter, aggr_idx, &out);
-               }
+       if (!config->metric_only && counter->default_metricgroup &&
+           !counter->default_show_events) {
+               void *from = NULL;
+
+               aggr_printout(config, os, os->evsel, os->id, os->aggr_nr);
+               /* Print out all the metricgroup with the same metric event. */
+               do {
+                       int num = 0;
+
+                       /* Print out the new line for the next new metricgroup. */
+                       if (from) {
+                               if (config->json_output)
+                                       new_line_json(config, (void *)os);
+                               else
+                                       __new_line_std_csv(config, os);
+                       }
+
+                       print_noise(config, os, counter, noise,
+                                   /*before_metric=*/true);
+                       print_running(config, os, run, ena,
+                                     /*before_metric=*/true);
+                       from = perf_stat__print_shadow_stats_metricgroup(
+                               config, counter, aggr_idx, &num, from, &out);
+               } while (from != NULL);
        } else {
-               pm(config, os, METRIC_THRESHOLD_UNKNOWN, /*format=*/NULL, /*unit=*/NULL, /*val=*/0);
+               perf_stat__print_shadow_stats(config, counter, aggr_idx, &out);
        }
 
        if (!config->metric_only) {
@@ -987,7 +980,7 @@ static void print_counter_aggrdata(struct perf_stat_config *config,
        ena = aggr->counts.ena;
        run = aggr->counts.run;
 
-       if (perf_stat__skip_metric_event(counter, ena, run))
+       if (perf_stat__skip_metric_event(counter))
                return;
 
        if (val == 0 && should_skip_zero_counter(config, counter, &id))
tools/perf/util/stat-shadow.c +0 -0
index 9c83f7d96caa..5d8d09e0e6ae 100644

tools/perf/util/stat-shadow.c static int prepare_metric(struct perf_stat_config *config, · line 86

Quoted 5 lines
tools/perf/util/stat-shadow.c +1 -1
@@ -83,7 +83,7 @@ static int prepare_metric(struct perf_stat_config *config,
                }
                /* Time events are always on CPU0, the first aggregation index. */
                aggr = &ps->aggr[is_tool_time ? tool_aggr_idx : aggr_idx];
-               if (!aggr || !metric_events[i]->supported) {
+               if (!aggr || !metric_events[i]->supported || aggr->counts.run == 0) {

This will conflict with:
https://lore.kernel.org/linux-perf-users/[email protected]/
but the same addition just needs adding in this place too.

Thanks,
Ian

Quoted 35 lines
tools/perf/util/stat-shadow.c +1 -5
                        /*
                         * Not supported events will have a count of 0, which
                         * can be confusing in a metric. Explicitly set the
@@ -335,14 +335,10 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
  * perf_stat__skip_metric_event - Skip the evsel in the Default metricgroup,
  *                               if it's not running or not the metric event.
  */
-bool perf_stat__skip_metric_event(struct evsel *evsel,
-                                 u64 ena, u64 run)
+bool perf_stat__skip_metric_event(struct evsel *evsel)
 {
        if (!evsel->default_metricgroup)
                return false;
 
-       if (!ena || !run)
-               return true;
-
        return !metricgroup__lookup(&evsel->evlist->metric_events, evsel, false);
 }
tools/perf/util/stat.h +1 -2
index f986911c9296..4bced233d2fc 100644
@@ -163,7 +163,7 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
                                   struct evsel *evsel,
                                   int aggr_idx,
                                   struct perf_stat_output_ctx *out);
-bool perf_stat__skip_metric_event(struct evsel *evsel, u64 ena, u64 run);
+bool perf_stat__skip_metric_event(struct evsel *evsel);
 void *perf_stat__print_shadow_stats_metricgroup(struct perf_stat_config *config,
                                                struct evsel *evsel,
                                                int aggr_idx,
--

2.53.0.rc2.204.g2597b5adb4-goog

reviewer
#3 lore raw
Quoted 4 lines · Arnaldo Carvalho de Melo
Quoted 1 line · Ian Rogers

On Tue, Feb 3, 2026 at 3:07 PM Chun-Tse Shao <[email protected]> wrote:

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
  $ ./perf stat -M CPUs_utilized -a --metric-only -j -- sleep 1
  {"CPUs  CPUs_utilized" : "none"}
Signed-off-by [email protected]

Thanks, applied to perf-tools-next,

I've started seeing this test failure following this change:

```
$ perf test -v 59
--- start ---
test child forked, pid 2316256
metric expr inst_retired.any / cpu_clk_unhalted.thread for IPC
Using CPUID GenuineIntel-6-8D-1
found event inst_retired.any
found event cpu_clk_unhalted.thread
Parsing metric events
'{inst_retired.any/metric-id=inst_retired.any/,cpu_clk_unhalted.thread/metric-id=cpu_clk_unhalted.thread/}:W'
Matched metric-id inst_retired.any to inst_retired.any
Matched metric-id cpu_clk_unhalted.thread to cpu_clk_unhalted.thread
FAILED tests/parse-metric.c:142 IPC failed, wrong ratio
FAILED tests/parse-metric.c:293 IPC failed
---- end(-1) ----
 59: Parse and process metrics                                       : FAILED!
```

The following fixes it for me:

```
diff --git a/tools/perf/tests/parse-metric.c b/tools/perf/tests/parse-metric.c
index 6bbc209a5c6a..7c7f489a5eb0 100644
--- a/tools/perf/tests/parse-metric.c
+++ b/tools/perf/tests/parse-metric.c
@@ -41,6 +41,8 @@ static void load_runtime_stat(struct evlist *evlist,
struct value *vals)
               count = find_value(evsel->name, vals);
               evsel->supported = true;
               evsel->stats->aggr->counts.val = count;
+               evsel->stats->aggr->counts.ena = 1;
+               evsel->stats->aggr->counts.run = 1;
       }
}

```

I'll mail it out as a fix.

Thanks,
Ian

Quoted 1 line

- Arnaldo

reviewer
#4 lore raw

PATCH 1/1 perf test parse-metric: Ensure aggregate counts appear to have run

Commit bb5a920b9099 ("perf stat: Ensure metrics are displayed even with failed events") made it so that counters which weren't enabled in the kernel were handled as NaN in metrics. This cased the "Parse and process metrics" test to start failing as it wasn't putting a non-zero value in these variables. Add arbitrary values of 1 to fix the test.

Fixes bb5a920b9099 ("perf stat: Ensure metrics are displayed even with failed events")
---
 tools/perf/tests/parse-metric.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/tests/parse-metric.c +2 -0
index 6bbc209a5c6a..7c7f489a5eb0 100644
@@ -41,6 +41,8 @@ static void load_runtime_stat(struct evlist *evlist, struct value *vals)
 		count = find_value(evsel->name, vals);
 		evsel->supported = true;
 		evsel->stats->aggr->counts.val = count;
+		evsel->stats->aggr->counts.ena = 1;
+		evsel->stats->aggr->counts.run = 1;
 	}
 }
 
Signature · 1 line
2.53.0.239.g8d8fc8a987-goog
maintainer Applied
#5 lore raw

Re: [PATCH v1] perf test parse-metric: Ensure aggregate counts appear to have run

Quoted 7 lines · Ian Rogers

Commit bb5a920b9099 ("perf stat: Ensure metrics are displayed even with failed events") made it so that counters which weren't enabled in the kernel were handled as NaN in metrics. This cased the "Parse and process metrics" test to start failing as it wasn't putting a non-zero value in these variables. Add arbitrary values of 1 to fix the test.

Fixes bb5a920b9099 ("perf stat: Ensure metrics are displayed even with failed events")

Thanks, applied to perf-tools-next,

- Arnaldo

Quoted 20 lines
---
 tools/perf/tests/parse-metric.c | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/tests/parse-metric.c +2 -0
index 6bbc209a5c6a..7c7f489a5eb0 100644
@@ -41,6 +41,8 @@ static void load_runtime_stat(struct evlist *evlist, struct value *vals)
 		count = find_value(evsel->name, vals);
 		evsel->supported = true;
 		evsel->stats->aggr->counts.val = count;
+		evsel->stats->aggr->counts.ena = 1;
+		evsel->stats->aggr->counts.run = 1;
 	}
 }
 
Signature · 1 line
2.53.0.239.g8d8fc8a987-goog