Skip to contentSkip to conversation

perf metrics: Fixes stat cmd cannot parse sys metrics when cpus MIDR mismatch

PATCH In review Junhao He 2 participants 3 messages Aug 7, 2024 – Sep 6, 2024 perf events

lore
Share & tools
lore

Junhao He

author
#0 lore raw

PATCH 1/1 perf metrics: Fixes stat cmd cannot parse sys metrics when cpus MIDR mismatch

On some platforms that do not use cpu MIDR mapping, such as hisilicon HIP09 platform. The list cmd can display the CPA metrics, but the stat cmd cannot work well to parse of CPA metrics.

  $ perf list metric
  Metrics:
    cpa_p0_avg_bw
         [Average bandwidth of CPA Port 0]
    cpa_p1_avg_bw
         [Average bandwidth of CPA Port 1]
  $ perf stat -M cpa_p0_avg_bw --timeout 1000 --> No error messages output
  $ echo $?
  234

Currently, the metricgroup__parse_groups() expects to find an cpu metric table, but the hisilicon/hip09 doesn't uses cpus MIDR to map json events and metrics, so pmu_metrics_table__find() will return NULL, than the cmd run failed.
But in metricgroup__add_metric(), the function parse for each sys metric and add it to metric_list, which also will get an valid sys metric table. So, we can ignore the NULL result of pmu_metrics_table__find() and to use the sys metric table.

metricgroup__parse_groups
-> parse_groups

     -> metricgroup__add_metric_list
         -> metricgroup__add_metric
             -> pmu_for_each_sys_metric   --> We've got the sys metric

Testing:
[root@localhost ~]# perf stat -M cpa_p0_avg_bw --no-merge --timeout 1000

Performance counter stats for 'system wide':

   CPU0        1,001,121,239      cpa_cycles [hisi_sicl0_cpa0]     #     0.00 cpa_p0_avg_bw
   CPU0                    0      cpa_p0_wr_dat [hisi_sicl0_cpa0]
   CPU0                    0      cpa_p0_rd_dat_64b [hisi_sicl0_cpa0]
   CPU0                    0      cpa_p0_rd_dat_32b [hisi_sicl0_cpa0]
   CPU0        1,001,094,851      cpa_cycles [hisi_sicl2_cpa0]     #     0.00 cpa_p0_avg_bw
   CPU0                    0      cpa_p0_wr_dat [hisi_sicl2_cpa0]
   CPU0                    0      cpa_p0_rd_dat_64b [hisi_sicl2_cpa0]
   CPU0                    0      cpa_p0_rd_dat_32b [hisi_sicl2_cpa0]
       1.001306160 seconds time elapsed
---
 tools/perf/util/metricgroup.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
tools/perf/util/metricgroup.c +2 -4
index 69f6a46402c3..fa62fb3418b6 100644
@@ -1123,7 +1123,7 @@ static int metricgroup__add_metric_sys_event_iter(const struct pmu_metric *pm,
 
 	ret = add_metric(d->metric_list, pm, d->modifier, d->metric_no_group,
 			 d->metric_no_threshold, d->user_requested_cpu_list,
-			 d->system_wide, d->root_metric, d->visited, d->table);
+			 d->system_wide, d->root_metric, d->visited, d->table ?: table);
 	if (ret)
 		goto out;
 
@@ -1239,7 +1239,7 @@ static int metricgroup__add_metric(const char *pmu, const char *metric_name, con
 	int ret;
 	bool has_match = false;
 
-	{
+	if (table) {
 		struct metricgroup__add_metric_data data = {
 			.list = &list,
 			.pmu = pmu,
@@ -1696,8 +1696,6 @@ int metricgroup__parse_groups(struct evlist *perf_evlist,
 {
 	const struct pmu_metrics_table *table = pmu_metrics_table__find();
 
-	if (!table)
-		return -EINVAL;
 	if (hardware_aware_grouping)
 		pr_debug("Use hardware aware grouping instead of traditional metric grouping method\n");
 
Signature · 1 line
2.33.0
Tested-by on v1 1/1
#1 lore raw
Quoted 43 lines · Junhao He

On some platforms that do not use cpu MIDR mapping, such as hisilicon HIP09 platform. The list cmd can display the CPA metrics, but the stat cmd cannot work well to parse of CPA metrics.

  $ perf list metric
  Metrics:
    cpa_p0_avg_bw
         [Average bandwidth of CPA Port 0]
    cpa_p1_avg_bw
         [Average bandwidth of CPA Port 1]
  $ perf stat -M cpa_p0_avg_bw --timeout 1000 --> No error messages output
  $ echo $?
  234

Currently, the metricgroup__parse_groups() expects to find an cpu metric table, but the hisilicon/hip09 doesn't uses cpus MIDR to map json events and metrics, so pmu_metrics_table__find() will return NULL, than the cmd run failed.
But in metricgroup__add_metric(), the function parse for each sys metric and add it to metric_list, which also will get an valid sys metric table. So, we can ignore the NULL result of pmu_metrics_table__find() and to use the sys metric table.

metricgroup__parse_groups
-> parse_groups

     -> metricgroup__add_metric_list
         -> metricgroup__add_metric
             -> pmu_for_each_sys_metric   --> We've got the sys metric

Testing:
[root@localhost ~]# perf stat -M cpa_p0_avg_bw --no-merge --timeout 1000

Performance counter stats for 'system wide':

   CPU0        1,001,121,239      cpa_cycles [hisi_sicl0_cpa0]     #     0.00 cpa_p0_avg_bw
   CPU0                    0      cpa_p0_wr_dat [hisi_sicl0_cpa0]
   CPU0                    0      cpa_p0_rd_dat_64b [hisi_sicl0_cpa0]
   CPU0                    0      cpa_p0_rd_dat_32b [hisi_sicl0_cpa0]
   CPU0        1,001,094,851      cpa_cycles [hisi_sicl2_cpa0]     #     0.00 cpa_p0_avg_bw
   CPU0                    0      cpa_p0_wr_dat [hisi_sicl2_cpa0]
   CPU0                    0      cpa_p0_rd_dat_64b [hisi_sicl2_cpa0]
   CPU0                    0      cpa_p0_rd_dat_32b [hisi_sicl2_cpa0]
Quoted 3 lines
       1.001306160 seconds time elapsed

The changes looks fine to me, I'd prefer to add some documents for this. See below.

Quoted 18 lines
---
 tools/perf/util/metricgroup.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
tools/perf/util/metricgroup.c +1 -1
index 69f6a46402c3..fa62fb3418b6 100644
@@ -1123,7 +1123,7 @@ static int metricgroup__add_metric_sys_event_iter(const struct pmu_metric *pm,
 
 	ret = add_metric(d->metric_list, pm, d->modifier, d->metric_no_group,
 			 d->metric_no_threshold, d->user_requested_cpu_list,
-			 d->system_wide, d->root_metric, d->visited, d->table);
+			 d->system_wide, d->root_metric, d->visited, d->table ?: table);
 	if (ret)
 		goto out;
 

tools/perf/util/metricgroup.c static int metricgroup__add_metric(const char *pmu, const char *metric_name, con · line 1242

Quoted 5 lines
tools/perf/util/metricgroup.c +1 -1
@@ -1239,7 +1239,7 @@ static int metricgroup__add_metric(const char *pmu, const char *metric_name, con
 	int ret;
 	bool has_match = false;
 
-	{
+	if (table) {

may add a comment here. If !table then the metrics may belong to a system PMU.

tools/perf/util/metricgroup.c int metricgroup__parse_groups(struct evlist *perf_evlist, · line 1700

Quoted 9 lines
tools/perf/util/metricgroup.c +0 -2
 		struct metricgroup__add_metric_data data = {
 			.list = &list,
 			.pmu = pmu,
@@ -1696,8 +1696,6 @@ int metricgroup__parse_groups(struct evlist *perf_evlist,
 {
 	const struct pmu_metrics_table *table = pmu_metrics_table__find();
 
-	if (!table)
-		return -EINVAL;

may add a debug message to mention we're parsing metrics of system PMUs.

Thanks.

Quoted 4 lines
tools/perf/util/metricgroup.c +0 -0
 	if (hardware_aware_grouping)
 		pr_debug("Use hardware aware grouping instead of traditional metric grouping method\n");
 
author
#2 lore raw

Hi, Yicong

Thanks for your comments.

Quoted 68 lines · Yicong Yang
Quoted 43 lines · Junhao He

On some platforms that do not use cpu MIDR mapping, such as hisilicon HIP09 platform. The list cmd can display the CPA metrics, but the stat cmd cannot work well to parse of CPA metrics.

   $ perf list metric
   Metrics:
     cpa_p0_avg_bw
          [Average bandwidth of CPA Port 0]
     cpa_p1_avg_bw
          [Average bandwidth of CPA Port 1]
   $ perf stat -M cpa_p0_avg_bw --timeout 1000 --> No error messages output
   $ echo $?
   234

Currently, the metricgroup__parse_groups() expects to find an cpu metric table, but the hisilicon/hip09 doesn't uses cpus MIDR to map json events and metrics, so pmu_metrics_table__find() will return NULL, than the cmd run failed.
But in metricgroup__add_metric(), the function parse for each sys metric and add it to metric_list, which also will get an valid sys metric table. So, we can ignore the NULL result of pmu_metrics_table__find() and to use the sys metric table.

metricgroup__parse_groups
-> parse_groups

      -> metricgroup__add_metric_list
          -> metricgroup__add_metric
              -> pmu_for_each_sys_metric   --> We've got the sys metric

Testing:
[root@localhost ~]# perf stat -M cpa_p0_avg_bw --no-merge --timeout 1000

Performance counter stats for 'system wide':

CPU0 1,001,121,239 cpa_cycles [hisi_sicl0_cpa0] # 0.00 cpa_p0_avg_bw
CPU0 0 cpa_p0_wr_dat [hisi_sicl0_cpa0]
CPU0 0 cpa_p0_rd_dat_64b [hisi_sicl0_cpa0]
CPU0 0 cpa_p0_rd_dat_32b [hisi_sicl0_cpa0]
CPU0 1,001,094,851 cpa_cycles [hisi_sicl2_cpa0] # 0.00 cpa_p0_avg_bw
CPU0 0 cpa_p0_wr_dat [hisi_sicl2_cpa0]
CPU0 0 cpa_p0_rd_dat_64b [hisi_sicl2_cpa0]
CPU0 0 cpa_p0_rd_dat_32b [hisi_sicl2_cpa0]

Quoted 3 lines

1.001306160 seconds time elapsed

The changes looks fine to me, I'd prefer to add some documents for this. See below.

Quoted 17 lines
---
  tools/perf/util/metricgroup.c | 6 ++----
  1 file changed, 2 insertions(+), 4 deletions(-)
tools/perf/util/metricgroup.c +1 -1
index 69f6a46402c3..fa62fb3418b6 100644
@@ -1123,7 +1123,7 @@ static int metricgroup__add_metric_sys_event_iter(const struct pmu_metric *pm,
 
  	ret = add_metric(d->metric_list, pm, d->modifier, d->metric_no_group,
  			 d->metric_no_threshold, d->user_requested_cpu_list,
-			 d->system_wide, d->root_metric, d->visited, d->table);
+			 d->system_wide, d->root_metric, d->visited, d->table ?: table);
  	if (ret)
  		goto out;
 
Quoted 7 lines

tools/perf/util/metricgroup.c static int metricgroup__add_metric(const char *pmu, const char *metric_name, con · line 1242

Quoted 6 lines
tools/perf/util/metricgroup.c +1 -1
@@ -1239,7 +1239,7 @@ static int metricgroup__add_metric(const char *pmu, const char *metric_name, con
  	int ret;
  	bool has_match = false;
 
-	{
+	if (table) {

may add a comment here. If !table then the metrics may belong to a system PMU.

Yes, I will do that.

Quoted 7 lines
Quoted 7 lines
tools/perf/util/metricgroup.c +0 -0
  		struct metricgroup__add_metric_data data = {
  			.list = &list,
  			.pmu = pmu,
@@ -1696,8 +1696,6 @@ int metricgroup__parse_groups(struct evlist *perf_evlist,
  {
  	const struct pmu_metrics_table *table = pmu_metrics_table__find();
 
Quoted 5 lines

tools/perf/util/metricgroup.c int metricgroup__parse_groups(struct evlist *perf_evlist, · line 1700

Quoted 2 lines
tools/perf/util/metricgroup.c +0 -2
-	if (!table)
-		return -EINVAL;

may add a debug message to mention we're parsing metrics of system PMUs.

Thanks.

Yes, I will do that.

Best regards,
Junhao.

Quoted 7 lines

tools/perf/util/metricgroup.c int metricgroup__parse_groups(struct evlist *perf_evlist, · line 1701

Quoted 5 lines
tools/perf/util/metricgroup.c +0 -0
  	if (hardware_aware_grouping)
  		pr_debug("Use hardware aware grouping instead of traditional metric grouping method\n");
 

.