Skip to contentSkip to conversation

perf stat: Use field separator in the metric header

PATCH In review Namhyung Kim 3 participants 8 messages Jun 27, 2024 – Feb 24, 2025 perf events

lore
Share & tools
lore

Namhyung Kim

author maintainer
#0 lore raw

PATCH 1/2 perf stat: Use field separator in the metric header

It didn't use the passed field separator (using -x option) when it prints the metric headers and always put "," between the fields.

Before:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core,cpus,%  tma_core_bound:     <<<--- here: "core,cpus," but ":" expected
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:14.8:
  S0-D0-C2:2:9.9:
  S0-D0-C3:2:13.2:

After:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core:cpus:%  tma_core_bound:
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:15.0:
  S0-D0-C2:2:16.5:
  S0-D0-C3:2:12.5:
---
 tools/perf/util/stat-display.c | 37 ++++++++++++++++++++++++++--------
 1 file changed, 29 insertions(+), 8 deletions(-)
tools/perf/util/stat-display.c +29 -8
index 91d2f7f65df7..e8673c9f6b49 100644
@@ -47,16 +47,27 @@ static int aggr_header_lens[] = {
 };
 
 static const char *aggr_header_csv[] = {
-	[AGGR_CORE] 	= 	"core,cpus,",
-	[AGGR_CACHE]	= 	"cache,cpus,",
-	[AGGR_DIE] 	= 	"die,cpus,",
-	[AGGR_SOCKET] 	= 	"socket,cpus,",
-	[AGGR_NONE] 	= 	"cpu,",
-	[AGGR_THREAD] 	= 	"comm-pid,",
-	[AGGR_NODE] 	= 	"node,",
+	[AGGR_CORE] 	= 	"core%scpus%s",
+	[AGGR_CACHE]	= 	"cache%scpus%s",
+	[AGGR_DIE] 	= 	"die%scpus%s",
+	[AGGR_SOCKET] 	= 	"socket%scpus%s",
+	[AGGR_NONE] 	= 	"cpu%s",
+	[AGGR_THREAD] 	= 	"comm-pid%s",
+	[AGGR_NODE] 	= 	"node%s",
 	[AGGR_GLOBAL] 	=	""
 };
 
+static int aggr_header_num[] = {
+	[AGGR_CORE] 	= 	2,
+	[AGGR_CACHE]	= 	2,
+	[AGGR_DIE] 	= 	2,
+	[AGGR_SOCKET] 	= 	2,
+	[AGGR_NONE] 	= 	1,
+	[AGGR_THREAD] 	= 	1,
+	[AGGR_NODE] 	= 	1,
+	[AGGR_GLOBAL] 	=	0,
+};
+
 static const char *aggr_header_std[] = {
 	[AGGR_CORE] 	= 	"core",
 	[AGGR_CACHE] 	= 	"cache",
@@ -1185,8 +1196,18 @@ static void print_metric_headers_csv(struct perf_stat_config *config,
 {
 	if (config->interval)
 		fputs("time,", config->output);
-	if (!config->iostat_run)
+	if (config->iostat_run)
+		return;
+
+	if (aggr_header_num[config->aggr_mode] == 1) {
+		fprintf(config->output, aggr_header_csv[config->aggr_mode],
+			config->csv_sep);
+	} else if (aggr_header_num[config->aggr_mode] == 2) {
+		fprintf(config->output, aggr_header_csv[config->aggr_mode],
+			config->csv_sep, config->csv_sep);
+	} else {
 		fputs(aggr_header_csv[config->aggr_mode], config->output);
+	}
 }
 
 static void print_metric_headers_json(struct perf_stat_config *config __maybe_unused,
Signature · 1 line
2.45.2.803.g4e1b14247a-goog
author maintainer
#1 lore raw

PATCH 2/2 perf stat: Fix a segfault with --per-cluster --metric-only

The new --per-cluster option was added recently but it forgot to update the aggr_header fields which are used for --metric-only option. And it resulted in a segfault due to NULL string in fputs().

Fixes cbc917a1b03b ("perf stat: Support per-cluster aggregation")
Cc Yicong Yang <[email protected]>
---
 tools/perf/util/stat-display.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/stat-display.c +3 -0
index e8673c9f6b49..462227f663cb 100644
@@ -38,6 +38,7 @@
 static int aggr_header_lens[] = {
 	[AGGR_CORE] 	= 18,
 	[AGGR_CACHE]	= 22,
+	[AGGR_CLUSTER]	= 20,
 	[AGGR_DIE] 	= 12,
 	[AGGR_SOCKET] 	= 6,
 	[AGGR_NODE] 	= 6,
@@ -49,6 +50,7 @@ static int aggr_header_lens[] = {
 static const char *aggr_header_csv[] = {
 	[AGGR_CORE] 	= 	"core%scpus%s",
 	[AGGR_CACHE]	= 	"cache%scpus%s",
+	[AGGR_CLUSTER]	= 	"cluster%scpus%s",
 	[AGGR_DIE] 	= 	"die%scpus%s",
 	[AGGR_SOCKET] 	= 	"socket%scpus%s",
 	[AGGR_NONE] 	= 	"cpu%s",
@@ -60,6 +62,7 @@ static const char *aggr_header_csv[] = {
 static int aggr_header_num[] = {
 	[AGGR_CORE] 	= 	2,
 	[AGGR_CACHE]	= 	2,
+	[AGGR_CLUSTER]	= 	2,
 	[AGGR_DIE] 	= 	2,
 	[AGGR_SOCKET] 	= 	2,
 	[AGGR_NONE] 	= 	1,
Signature · 1 line
2.45.2.803.g4e1b14247a-goog
maintainer Tested-by on v1 2/2
#2 lore raw

Re: [PATCH 2/2] perf stat: Fix a segfault with --per-cluster --metric-only

Quoted 3 lines · Namhyung Kim

The new --per-cluster option was added recently but it forgot to update the aggr_header fields which are used for --metric-only option. And it resulted in a segfault due to NULL string in fputs().

Before:

acme@number:~$ sudo ~acme/bin/perf stat -a -x : --per-cluster -M tma_core_bound --metric-only true
Segmentation fault
acme@number:~$ 
acme@number:~$ sudo su -
root@number:~# gdb perf
(gdb) run stat -a -x : --per-cluster -M tma_core_bound --metric-only true
Starting program: /root/bin/perf stat -a -x : --per-cluster -M tma_core_bound --metric-only true

Program received signal SIGSEGV, Segmentation fault. 0x00007ffff6f7f8dd in __strlen_avx2 () from /lib64/libc.so.6 (gdb) bt

#0  0x00007ffff6f7f8dd in __strlen_avx2 () from /lib64/libc.so.6
#1  0x00007ffff6e97a3a in fputs () from /lib64/libc.so.6
#2  0x000000000056b805 in print_metric_headers ()
#3  0x000000000056e084 in evlist.print_counters ()
#4  0x0000000000432513 in cmd_stat ()
#5  0x00000000004c5fb9 in run_builtin ()
#6  0x00000000004c62c9 in handle_internal_command ()
#7  0x0000000000410e57 in main ()

(gdb)

After:

acme@number:~$ sudo ~acme/bin/perf stat -a -x : --per-cluster -M tma_core_bound --metric-only true
cluster:cpus:%  tma_core_bound:%  tma_core_bound:
S0-D0-CLS0:2:18.2::::
S0-D0-CLS8:2:26.7::::
S0-D0-CLS16:2:14.2::::
S0-D0-CLS24:2:10.6::::
S0-D0-CLS32:2:0.6::::
S0-D0-CLS40:2:42.5::::
S0-D0-CLS48:2:21.1::::
S0-D0-CLS56:2:36.8::::
S0-D0-CLS64:0:::::::1.0:
S0-D0-CLS72:0:::::::0.8:
S0-D0-CLS80:0:::::::1.0:
acme@number:~$

- Arnaldo

Quoted 37 lines
Fixes cbc917a1b03b ("perf stat: Support per-cluster aggregation")
Cc Yicong Yang <[email protected]>
---
 tools/perf/util/stat-display.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/stat-display.c +3 -0
index e8673c9f6b49..462227f663cb 100644
@@ -38,6 +38,7 @@
 static int aggr_header_lens[] = {
 	[AGGR_CORE] 	= 18,
 	[AGGR_CACHE]	= 22,
+	[AGGR_CLUSTER]	= 20,
 	[AGGR_DIE] 	= 12,
 	[AGGR_SOCKET] 	= 6,
 	[AGGR_NODE] 	= 6,
@@ -49,6 +50,7 @@ static int aggr_header_lens[] = {
 static const char *aggr_header_csv[] = {
 	[AGGR_CORE] 	= 	"core%scpus%s",
 	[AGGR_CACHE]	= 	"cache%scpus%s",
+	[AGGR_CLUSTER]	= 	"cluster%scpus%s",
 	[AGGR_DIE] 	= 	"die%scpus%s",
 	[AGGR_SOCKET] 	= 	"socket%scpus%s",
 	[AGGR_NONE] 	= 	"cpu%s",
@@ -60,6 +62,7 @@ static const char *aggr_header_csv[] = {
 static int aggr_header_num[] = {
 	[AGGR_CORE] 	= 	2,
 	[AGGR_CACHE]	= 	2,
+	[AGGR_CLUSTER]	= 	2,
 	[AGGR_DIE] 	= 	2,
 	[AGGR_SOCKET] 	= 	2,
 	[AGGR_NONE] 	= 	1,
Signature · 1 line
2.45.2.803.g4e1b14247a-goog
reviewer
#3 lore raw
In reply to Namhyung Kim
Quoted 65 lines · Namhyung Kim

It didn't use the passed field separator (using -x option) when it prints the metric headers and always put "," between the fields.

Before:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core,cpus,%  tma_core_bound:     <<<--- here: "core,cpus," but ":" expected
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:14.8:
  S0-D0-C2:2:9.9:
  S0-D0-C3:2:13.2:

After:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core:cpus:%  tma_core_bound:
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:15.0:
  S0-D0-C2:2:16.5:
  S0-D0-C3:2:12.5:
---
 tools/perf/util/stat-display.c | 37 ++++++++++++++++++++++++++--------
 1 file changed, 29 insertions(+), 8 deletions(-)
tools/perf/util/stat-display.c +18 -7
index 91d2f7f65df7..e8673c9f6b49 100644
@@ -47,16 +47,27 @@ static int aggr_header_lens[] = {
 };
 
 static const char *aggr_header_csv[] = {
-       [AGGR_CORE]     =       "core,cpus,",
-       [AGGR_CACHE]    =       "cache,cpus,",
-       [AGGR_DIE]      =       "die,cpus,",
-       [AGGR_SOCKET]   =       "socket,cpus,",
-       [AGGR_NONE]     =       "cpu,",
-       [AGGR_THREAD]   =       "comm-pid,",
-       [AGGR_NODE]     =       "node,",
+       [AGGR_CORE]     =       "core%scpus%s",
+       [AGGR_CACHE]    =       "cache%scpus%s",
+       [AGGR_DIE]      =       "die%scpus%s",
+       [AGGR_SOCKET]   =       "socket%scpus%s",
+       [AGGR_NONE]     =       "cpu%s",
+       [AGGR_THREAD]   =       "comm-pid%s",
+       [AGGR_NODE]     =       "node%s",
        [AGGR_GLOBAL]   =       ""
 };
 
+static int aggr_header_num[] = {
+       [AGGR_CORE]     =       2,
+       [AGGR_CACHE]    =       2,
+       [AGGR_DIE]      =       2,
+       [AGGR_SOCKET]   =       2,
+       [AGGR_NONE]     =       1,
+       [AGGR_THREAD]   =       1,
+       [AGGR_NODE]     =       1,
+       [AGGR_GLOBAL]   =       0,
+};
+
 static const char *aggr_header_std[] = {
        [AGGR_CORE]     =       "core",
        [AGGR_CACHE]    =       "cache",

tools/perf/util/stat-display.c static void print_metric_headers_csv(struct perf_stat_config *config, · line 1201

Quoted 7 lines
tools/perf/util/stat-display.c +3 -1
@@ -1185,8 +1196,18 @@ static void print_metric_headers_csv(struct perf_stat_config *config,
 {
        if (config->interval)
                fputs("time,", config->output);
-       if (!config->iostat_run)
+       if (config->iostat_run)
+               return;
+

Having a static count of commas seems somewhat error prone, perhaps:

```
const char *header = aggr_header_csv[config->aggr_mode];
if (config->csv_sep == ',' || !strchr(header, ',')) {
  fputs(config->output, header);
} else {
  char *tmp = strdup(header);
  char *p = tmp;
   while (p && *p) {
      if (p == ',')
        *p = config->csv_sep;
     p++;
   }
  fputs(config->output, tmp);
  free(tmp);
}
```
I'm somewhat surprised that we have no metric tests in the stat output
tests like tools/perf/tests/shell/stat+csv_output.sh. Perhaps we can
add the following:
```
diff --git a/tools/perf/tests/shell/lib/stat_output.sh
b/tools/perf/tests/shell/lib/stat_output.sh
index 9a176ceae4a3..a920b2d78abb 100644
--- a/tools/perf/tests/shell/lib/stat_output.sh
+++ b/tools/perf/tests/shell/lib/stat_output.sh
@@ -148,6 +148,14 @@ check_per_socket()
       echo "[Success]"
}

+check_metric_only()
+{
+        echo -n "Checking $1 output: metric only "
+        perf stat --metric-only $2 -e instructions,cycles true
+        commachecker --metric-only
+        echo "[Success]"
+}
+
# The perf stat options for per-socket, per-core, per-die
# and -A ( no_aggr mode ) uses the info fetched from this
# directory: "/sys/devices/system/cpu/cpu*/topology". For
diff --git a/tools/perf/tests/shell/stat+csv_output.sh
b/tools/perf/tests/shell/stat+csv_output.sh
index fc2d8cc6e5e0..d6807dbab931 100755
--- a/tools/perf/tests/shell/stat+csv_output.sh
+++ b/tools/perf/tests/shell/stat+csv_output.sh
@@ -44,6 +44,7 @@ function commachecker()
       ;; "--per-die")         exp=8
       ;; "--per-cluster")     exp=8
       ;; "--per-cache")       exp=8
+        ;; "--metric-only")     exp=2
       esac

       while read line
@@ -83,6 +84,7 @@ then
       check_per_cluster "CSV" "$perf_cmd"
       check_per_die "CSV" "$perf_cmd"
       check_per_socket "CSV" "$perf_cmd"
+        check_metric_only "CSV" "$perf_cmd"
else
       echo "[Skip] Skipping tests for system_wide_no_aggr, per_core,
per_die and per_socket since
socket id exposed via topology is invalid"
fi
```

It is using the hard coded metrics and it looks like the header printing for that is broken, but this is so often the case for stat output :-(

Thanks,
Ian

Quoted 15 lines
tools/perf/util/stat-display.c +8 -1
+       if (aggr_header_num[config->aggr_mode] == 1) {
+               fprintf(config->output, aggr_header_csv[config->aggr_mode],
+                       config->csv_sep);
+       } else if (aggr_header_num[config->aggr_mode] == 2) {
+               fprintf(config->output, aggr_header_csv[config->aggr_mode],
+                       config->csv_sep, config->csv_sep);
+       } else {
                fputs(aggr_header_csv[config->aggr_mode], config->output);
+       }
 }
 
 static void print_metric_headers_json(struct perf_stat_config *config __maybe_unused,
--

2.45.2.803.g4e1b14247a-goog

author maintainer
#4 lore raw

Hi Ian,

Quoted 76 lines · Ian Rogers
Quoted 65 lines · Namhyung Kim

It didn't use the passed field separator (using -x option) when it prints the metric headers and always put "," between the fields.

Before:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core,cpus,%  tma_core_bound:     <<<--- here: "core,cpus," but ":" expected
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:14.8:
  S0-D0-C2:2:9.9:
  S0-D0-C3:2:13.2:

After:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core:cpus:%  tma_core_bound:
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:15.0:
  S0-D0-C2:2:16.5:
  S0-D0-C3:2:12.5:
---
 tools/perf/util/stat-display.c | 37 ++++++++++++++++++++++++++--------
 1 file changed, 29 insertions(+), 8 deletions(-)
tools/perf/util/stat-display.c +18 -7
index 91d2f7f65df7..e8673c9f6b49 100644
@@ -47,16 +47,27 @@ static int aggr_header_lens[] = {
 };
 
 static const char *aggr_header_csv[] = {
-       [AGGR_CORE]     =       "core,cpus,",
-       [AGGR_CACHE]    =       "cache,cpus,",
-       [AGGR_DIE]      =       "die,cpus,",
-       [AGGR_SOCKET]   =       "socket,cpus,",
-       [AGGR_NONE]     =       "cpu,",
-       [AGGR_THREAD]   =       "comm-pid,",
-       [AGGR_NODE]     =       "node,",
+       [AGGR_CORE]     =       "core%scpus%s",
+       [AGGR_CACHE]    =       "cache%scpus%s",
+       [AGGR_DIE]      =       "die%scpus%s",
+       [AGGR_SOCKET]   =       "socket%scpus%s",
+       [AGGR_NONE]     =       "cpu%s",
+       [AGGR_THREAD]   =       "comm-pid%s",
+       [AGGR_NODE]     =       "node%s",
        [AGGR_GLOBAL]   =       ""
 };
 
+static int aggr_header_num[] = {
+       [AGGR_CORE]     =       2,
+       [AGGR_CACHE]    =       2,
+       [AGGR_DIE]      =       2,
+       [AGGR_SOCKET]   =       2,
+       [AGGR_NONE]     =       1,
+       [AGGR_THREAD]   =       1,
+       [AGGR_NODE]     =       1,
+       [AGGR_GLOBAL]   =       0,
+};
+
 static const char *aggr_header_std[] = {
        [AGGR_CORE]     =       "core",
        [AGGR_CACHE]    =       "cache",

tools/perf/util/stat-display.c static void print_metric_headers_csv(struct perf_stat_config *config, · line 1201

Quoted 7 lines
tools/perf/util/stat-display.c +3 -1
@@ -1185,8 +1196,18 @@ static void print_metric_headers_csv(struct perf_stat_config *config,
 {
        if (config->interval)
                fputs("time,", config->output);
-       if (!config->iostat_run)
+       if (config->iostat_run)
+               return;
+

Having a static count of commas seems somewhat error prone, perhaps:

Quoted 16 lines
```
const char *header = aggr_header_csv[config->aggr_mode];
if (config->csv_sep == ',' || !strchr(header, ',')) {
  fputs(config->output, header);
} else {
  char *tmp = strdup(header);
  char *p = tmp;
   while (p && *p) {
      if (p == ',')
        *p = config->csv_sep;
     p++;
   }
  fputs(config->output, tmp);
  free(tmp);
}
```

Looks good. But I think we should handle longer separators like -x ":::". Will do in v2.

Quoted 43 lines

I'm somewhat surprised that we have no metric tests in the stat output tests like tools/perf/tests/shell/stat+csv_output.sh. Perhaps we can add the following:
```

tools/perf/tests/shell/lib/stat_output.sh +8 -0
index 9a176ceae4a3..a920b2d78abb 100644
@@ -148,6 +148,14 @@ check_per_socket()
       echo "[Success]"
 }
 
+check_metric_only()
+{
+        echo -n "Checking $1 output: metric only "
+        perf stat --metric-only $2 -e instructions,cycles true
+        commachecker --metric-only
+        echo "[Success]"
+}
+
 # The perf stat options for per-socket, per-core, per-die
 # and -A ( no_aggr mode ) uses the info fetched from this
 # directory: "/sys/devices/system/cpu/cpu*/topology". For
tools/perf/tests/shell/stat+csv_output.sh +2 -0
index fc2d8cc6e5e0..d6807dbab931 100755
@@ -44,6 +44,7 @@ function commachecker()
       ;; "--per-die")         exp=8
       ;; "--per-cluster")     exp=8
       ;; "--per-cache")       exp=8
+        ;; "--metric-only")     exp=2
       esac
 
       while read line
@@ -83,6 +84,7 @@ then
       check_per_cluster "CSV" "$perf_cmd"
       check_per_die "CSV" "$perf_cmd"
       check_per_socket "CSV" "$perf_cmd"
+        check_metric_only "CSV" "$perf_cmd"

else

Quoted 8 lines

echo "[Skip] Skipping tests for system_wide_no_aggr, per_core,
per_die and per_socket since
socket id exposed via topology is invalid"
fi
```
It is using the hard coded metrics and it looks like the header printing for that is broken, but this is so often the case for stat output :-(

Right, I also noticed something in the header. One more work item to the list.

Anyway, I'll add it to the test case!

Thanks,
Namhyung

Quoted 16 lines
Quoted 16 lines
tools/perf/util/stat-display.c +8 -1
+       if (aggr_header_num[config->aggr_mode] == 1) {
+               fprintf(config->output, aggr_header_csv[config->aggr_mode],
+                       config->csv_sep);
+       } else if (aggr_header_num[config->aggr_mode] == 2) {
+               fprintf(config->output, aggr_header_csv[config->aggr_mode],
+                       config->csv_sep, config->csv_sep);
+       } else {
                fputs(aggr_header_csv[config->aggr_mode], config->output);
+       }
 }
 
 static void print_metric_headers_json(struct perf_stat_config *config __maybe_unused,
--

2.45.2.803.g4e1b14247a-goog

reviewer
#5 lore raw
Quoted 97 lines · Namhyung Kim

Hi Ian,

Quoted 76 lines · Ian Rogers
Quoted 65 lines · Namhyung Kim

It didn't use the passed field separator (using -x option) when it prints the metric headers and always put "," between the fields.

Before:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core,cpus,%  tma_core_bound:     <<<--- here: "core,cpus," but ":" expected
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:14.8:
  S0-D0-C2:2:9.9:
  S0-D0-C3:2:13.2:

After:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core:cpus:%  tma_core_bound:
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:15.0:
  S0-D0-C2:2:16.5:
  S0-D0-C3:2:12.5:
---
 tools/perf/util/stat-display.c | 37 ++++++++++++++++++++++++++--------
 1 file changed, 29 insertions(+), 8 deletions(-)
tools/perf/util/stat-display.c +18 -7
index 91d2f7f65df7..e8673c9f6b49 100644
@@ -47,16 +47,27 @@ static int aggr_header_lens[] = {
 };
 
 static const char *aggr_header_csv[] = {
-       [AGGR_CORE]     =       "core,cpus,",
-       [AGGR_CACHE]    =       "cache,cpus,",
-       [AGGR_DIE]      =       "die,cpus,",
-       [AGGR_SOCKET]   =       "socket,cpus,",
-       [AGGR_NONE]     =       "cpu,",
-       [AGGR_THREAD]   =       "comm-pid,",
-       [AGGR_NODE]     =       "node,",
+       [AGGR_CORE]     =       "core%scpus%s",
+       [AGGR_CACHE]    =       "cache%scpus%s",
+       [AGGR_DIE]      =       "die%scpus%s",
+       [AGGR_SOCKET]   =       "socket%scpus%s",
+       [AGGR_NONE]     =       "cpu%s",
+       [AGGR_THREAD]   =       "comm-pid%s",
+       [AGGR_NODE]     =       "node%s",
        [AGGR_GLOBAL]   =       ""
 };
 
+static int aggr_header_num[] = {
+       [AGGR_CORE]     =       2,
+       [AGGR_CACHE]    =       2,
+       [AGGR_DIE]      =       2,
+       [AGGR_SOCKET]   =       2,
+       [AGGR_NONE]     =       1,
+       [AGGR_THREAD]   =       1,
+       [AGGR_NODE]     =       1,
+       [AGGR_GLOBAL]   =       0,
+};
+
 static const char *aggr_header_std[] = {
        [AGGR_CORE]     =       "core",
        [AGGR_CACHE]    =       "cache",

tools/perf/util/stat-display.c static void print_metric_headers_csv(struct perf_stat_config *config, · line 1201

Quoted 7 lines
tools/perf/util/stat-display.c +3 -1
@@ -1185,8 +1196,18 @@ static void print_metric_headers_csv(struct perf_stat_config *config,
 {
        if (config->interval)
                fputs("time,", config->output);
-       if (!config->iostat_run)
+       if (config->iostat_run)
+               return;
+

Having a static count of commas seems somewhat error prone, perhaps:

Quoted 16 lines
```
const char *header = aggr_header_csv[config->aggr_mode];
if (config->csv_sep == ',' || !strchr(header, ',')) {
  fputs(config->output, header);
} else {
  char *tmp = strdup(header);
  char *p = tmp;
   while (p && *p) {
      if (p == ',')
        *p = config->csv_sep;
     p++;
   }
  fputs(config->output, tmp);
  free(tmp);
}
```
Quoted 2 lines

Looks good. But I think we should handle longer separators like -x ":::". Will do in v2.

Hi Namhyung,

It looks like this has been forgotten. Did you have a v2?

Thanks,
Ian

Quoted 76 lines
Quoted 43 lines

I'm somewhat surprised that we have no metric tests in the stat output tests like tools/perf/tests/shell/stat+csv_output.sh. Perhaps we can add the following:
```

tools/perf/tests/shell/lib/stat_output.sh +8 -0
index 9a176ceae4a3..a920b2d78abb 100644
@@ -148,6 +148,14 @@ check_per_socket()
       echo "[Success]"
 }
 
+check_metric_only()
+{
+        echo -n "Checking $1 output: metric only "
+        perf stat --metric-only $2 -e instructions,cycles true
+        commachecker --metric-only
+        echo "[Success]"
+}
+
 # The perf stat options for per-socket, per-core, per-die
 # and -A ( no_aggr mode ) uses the info fetched from this
 # directory: "/sys/devices/system/cpu/cpu*/topology". For
tools/perf/tests/shell/stat+csv_output.sh +2 -0
index fc2d8cc6e5e0..d6807dbab931 100755
@@ -44,6 +44,7 @@ function commachecker()
       ;; "--per-die")         exp=8
       ;; "--per-cluster")     exp=8
       ;; "--per-cache")       exp=8
+        ;; "--metric-only")     exp=2
       esac
 
       while read line
@@ -83,6 +84,7 @@ then
       check_per_cluster "CSV" "$perf_cmd"
       check_per_die "CSV" "$perf_cmd"
       check_per_socket "CSV" "$perf_cmd"
+        check_metric_only "CSV" "$perf_cmd"

else

Quoted 8 lines

echo "[Skip] Skipping tests for system_wide_no_aggr, per_core,
per_die and per_socket since
socket id exposed via topology is invalid"
fi
```
It is using the hard coded metrics and it looks like the header printing for that is broken, but this is so often the case for stat output :-(

Right, I also noticed something in the header. One more work item to the list.

Anyway, I'll add it to the test case!

Thanks,
Namhyung

Quoted 16 lines
Quoted 16 lines
tools/perf/util/stat-display.c +8 -1
+       if (aggr_header_num[config->aggr_mode] == 1) {
+               fprintf(config->output, aggr_header_csv[config->aggr_mode],
+                       config->csv_sep);
+       } else if (aggr_header_num[config->aggr_mode] == 2) {
+               fprintf(config->output, aggr_header_csv[config->aggr_mode],
+                       config->csv_sep, config->csv_sep);
+       } else {
                fputs(aggr_header_csv[config->aggr_mode], config->output);
+       }
 }
 
 static void print_metric_headers_json(struct perf_stat_config *config __maybe_unused,
--

2.45.2.803.g4e1b14247a-goog

author maintainer
#6 lore raw
Quoted 98 lines · Ian Rogers
Quoted 97 lines · Namhyung Kim

Hi Ian,

Quoted 92 lines · Ian Rogers
Quoted 65 lines · Namhyung Kim

It didn't use the passed field separator (using -x option) when it prints the metric headers and always put "," between the fields.

Before:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core,cpus,%  tma_core_bound:     <<<--- here: "core,cpus," but ":" expected
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:14.8:
  S0-D0-C2:2:9.9:
  S0-D0-C3:2:13.2:

After:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core:cpus:%  tma_core_bound:
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:15.0:
  S0-D0-C2:2:16.5:
  S0-D0-C3:2:12.5:
---
 tools/perf/util/stat-display.c | 37 ++++++++++++++++++++++++++--------
 1 file changed, 29 insertions(+), 8 deletions(-)
tools/perf/util/stat-display.c +18 -7
index 91d2f7f65df7..e8673c9f6b49 100644
@@ -47,16 +47,27 @@ static int aggr_header_lens[] = {
 };
 
 static const char *aggr_header_csv[] = {
-       [AGGR_CORE]     =       "core,cpus,",
-       [AGGR_CACHE]    =       "cache,cpus,",
-       [AGGR_DIE]      =       "die,cpus,",
-       [AGGR_SOCKET]   =       "socket,cpus,",
-       [AGGR_NONE]     =       "cpu,",
-       [AGGR_THREAD]   =       "comm-pid,",
-       [AGGR_NODE]     =       "node,",
+       [AGGR_CORE]     =       "core%scpus%s",
+       [AGGR_CACHE]    =       "cache%scpus%s",
+       [AGGR_DIE]      =       "die%scpus%s",
+       [AGGR_SOCKET]   =       "socket%scpus%s",
+       [AGGR_NONE]     =       "cpu%s",
+       [AGGR_THREAD]   =       "comm-pid%s",
+       [AGGR_NODE]     =       "node%s",
        [AGGR_GLOBAL]   =       ""
 };
 
+static int aggr_header_num[] = {
+       [AGGR_CORE]     =       2,
+       [AGGR_CACHE]    =       2,
+       [AGGR_DIE]      =       2,
+       [AGGR_SOCKET]   =       2,
+       [AGGR_NONE]     =       1,
+       [AGGR_THREAD]   =       1,
+       [AGGR_NODE]     =       1,
+       [AGGR_GLOBAL]   =       0,
+};
+
 static const char *aggr_header_std[] = {
        [AGGR_CORE]     =       "core",
        [AGGR_CACHE]    =       "cache",

tools/perf/util/stat-display.c static void print_metric_headers_csv(struct perf_stat_config *config, · line 1201

Quoted 7 lines
tools/perf/util/stat-display.c +3 -1
@@ -1185,8 +1196,18 @@ static void print_metric_headers_csv(struct perf_stat_config *config,
 {
        if (config->interval)
                fputs("time,", config->output);
-       if (!config->iostat_run)
+       if (config->iostat_run)
+               return;
+

Having a static count of commas seems somewhat error prone, perhaps:

```
const char *header = aggr_header_csv[config->aggr_mode];
if (config->csv_sep == ',' || !strchr(header, ',')) {
  fputs(config->output, header);
} else {
  char *tmp = strdup(header);
  char *p = tmp;
   while (p && *p) {
      if (p == ',')
        *p = config->csv_sep;
     p++;
   }
  fputs(config->output, tmp);
  free(tmp);
}
```
Quoted 6 lines
Quoted 2 lines

Looks good. But I think we should handle longer separators like -x ":::". Will do in v2.

Hi Namhyung,

It looks like this has been forgotten. Did you have a v2?

It's merged to v6.11, please see

https://lore.kernel.org/linux-perf-users/[email protected]/

Thanks,
Namhyung

maintainer Tested-by on v1 1/2
#7 lore raw
In reply to Namhyung Kim
Quoted 12 lines · Namhyung Kim

It didn't use the passed field separator (using -x option) when it prints the metric headers and always put "," between the fields.

Before:

  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core,cpus,%  tma_core_bound:     <<<--- here: "core,cpus," but ":" expected
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:14.8:
  S0-D0-C2:2:9.9:
  S0-D0-C3:2:13.2:

After:

Quoted 8 lines
  $ sudo ./perf stat -a -x : --per-core -M tma_core_bound --metric-only true
  core:cpus:%  tma_core_bound:
  S0-D0-C0:2:10.5:
  S0-D0-C1:2:15.0:
  S0-D0-C2:2:16.5:
  S0-D0-C3:2:12.5:

- Arnaldo

Quoted 65 lines
---
 tools/perf/util/stat-display.c | 37 ++++++++++++++++++++++++++--------
 1 file changed, 29 insertions(+), 8 deletions(-)
tools/perf/util/stat-display.c +29 -8
index 91d2f7f65df7..e8673c9f6b49 100644
@@ -47,16 +47,27 @@ static int aggr_header_lens[] = {
 };
 
 static const char *aggr_header_csv[] = {
-	[AGGR_CORE] 	= 	"core,cpus,",
-	[AGGR_CACHE]	= 	"cache,cpus,",
-	[AGGR_DIE] 	= 	"die,cpus,",
-	[AGGR_SOCKET] 	= 	"socket,cpus,",
-	[AGGR_NONE] 	= 	"cpu,",
-	[AGGR_THREAD] 	= 	"comm-pid,",
-	[AGGR_NODE] 	= 	"node,",
+	[AGGR_CORE] 	= 	"core%scpus%s",
+	[AGGR_CACHE]	= 	"cache%scpus%s",
+	[AGGR_DIE] 	= 	"die%scpus%s",
+	[AGGR_SOCKET] 	= 	"socket%scpus%s",
+	[AGGR_NONE] 	= 	"cpu%s",
+	[AGGR_THREAD] 	= 	"comm-pid%s",
+	[AGGR_NODE] 	= 	"node%s",
 	[AGGR_GLOBAL] 	=	""
 };
 
+static int aggr_header_num[] = {
+	[AGGR_CORE] 	= 	2,
+	[AGGR_CACHE]	= 	2,
+	[AGGR_DIE] 	= 	2,
+	[AGGR_SOCKET] 	= 	2,
+	[AGGR_NONE] 	= 	1,
+	[AGGR_THREAD] 	= 	1,
+	[AGGR_NODE] 	= 	1,
+	[AGGR_GLOBAL] 	=	0,
+};
+
 static const char *aggr_header_std[] = {
 	[AGGR_CORE] 	= 	"core",
 	[AGGR_CACHE] 	= 	"cache",
@@ -1185,8 +1196,18 @@ static void print_metric_headers_csv(struct perf_stat_config *config,
 {
 	if (config->interval)
 		fputs("time,", config->output);
-	if (!config->iostat_run)
+	if (config->iostat_run)
+		return;
+
+	if (aggr_header_num[config->aggr_mode] == 1) {
+		fprintf(config->output, aggr_header_csv[config->aggr_mode],
+			config->csv_sep);
+	} else if (aggr_header_num[config->aggr_mode] == 2) {
+		fprintf(config->output, aggr_header_csv[config->aggr_mode],
+			config->csv_sep, config->csv_sep);
+	} else {
 		fputs(aggr_header_csv[config->aggr_mode], config->output);
+	}
 }
 
 static void print_metric_headers_json(struct perf_stat_config *config __maybe_unused,
Signature · 1 line
2.45.2.803.g4e1b14247a-goog