Skip to contentSkip to conversation

perf tests: build-test coverage for NO_JEVENTS=1

PATCH Applied Ian Rogers 4 participants 17 messages Feb 5 – Feb 24 perf events

lore
Share & tools
lore

Ian Rogers

author reviewer
#0 lore raw

PATCH 1/2 perf tests: build-test coverage for NO_JEVENTS=1

Leo reported the NO_JEVENTS=1 build being broken and this highlighted it being missing from the build-test, add it.

---
 tools/perf/tests/make | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/tests/make +2 -0
index eb41516c0562..6587dc326d1b 100644
@@ -70,6 +70,7 @@ make_python_perf_so := $(python_perf_so)
 make_debug          := DEBUG=1
 make_nondistro      := BUILD_NONDISTRO=1
 make_extra_tests    := EXTRA_TESTS=1
+make_no_jevents     := NO_JEVENTS=1
 make_jevents_all    := JEVENTS_ARCH=all
 make_no_bpf_skel    := BUILD_BPF_SKEL=0
 make_gen_vmlinux_h  := GEN_VMLINUX_H=1
@@ -144,6 +145,7 @@ ifneq ($(new_libbfd),)
 run += make_nondistro
 endif
 run += make_extra_tests
+run += make_no_jevents
 run += make_jevents_all
 run += make_no_bpf_skel
 run += make_gen_vmlinux_h
Signature · 1 line
2.53.0.rc2.204.g2597b5adb4-goog
author reviewer
#1 lore raw

PATCH 2/2 perf metricgroup: Don't early exit if no CPUID table exists

The failure to find a table of metrics with a CPUID shouldn't early exit as the metric code will now also consider the default table. When searching for a metric or metric group, pmu_metrics_table__for_each_metric considers all tables and so the caller doesn't need to switch the table to do this.

Fixes c7adeb0974f1 ("perf jevents: Add set of common metrics based on default ones")
---
 tools/perf/util/metricgroup.c | 18 +++++-------------
 1 file changed, 5 insertions(+), 13 deletions(-)
tools/perf/util/metricgroup.c +5 -13
index 40a1e14de418..46bf4dfeebc8 100644
@@ -1562,8 +1562,6 @@ 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");
 
@@ -1601,22 +1599,16 @@ static int metricgroup__has_metric_or_groups_callback(const struct pmu_metric *p
 
 bool metricgroup__has_metric_or_groups(const char *pmu, const char *metric_or_groups)
 {
-	const struct pmu_metrics_table *tables[2] = {
-		pmu_metrics_table__find(),
-		pmu_metrics_table__default(),
-	};
+	const struct pmu_metrics_table *table = pmu_metrics_table__find();
 	struct metricgroup__has_metric_data data = {
 		.pmu = pmu,
 		.metric_or_groups = metric_or_groups,
 	};
 
-	for (size_t i = 0; i < ARRAY_SIZE(tables); i++) {
-		if (pmu_metrics_table__for_each_metric(tables[i],
-							metricgroup__has_metric_or_groups_callback,
-							&data))
-			return true;
-	}
-	return false;
+	return pmu_metrics_table__for_each_metric(table,
+						  metricgroup__has_metric_or_groups_callback,
+						  &data)
+		? true : false;
 }
 
 static int metricgroup__topdown_max_level_callback(const struct pmu_metric *pm,
Signature · 1 line
2.53.0.rc2.204.g2597b5adb4-goog
Tested-by on v1 2/2
#2 lore raw

Re: [PATCH v1 2/2] perf metricgroup: Don't early exit if no CPUID table exists

Quoted 26 lines · Ian Rogers

The failure to find a table of metrics with a CPUID shouldn't early exit as the metric code will now also consider the default table. When searching for a metric or metric group, pmu_metrics_table__for_each_metric considers all tables and so the caller doesn't need to switch the table to do this.

Fixes c7adeb0974f1 ("perf jevents: Add set of common metrics based on default ones")
---
 tools/perf/util/metricgroup.c | 18 +++++-------------
 1 file changed, 5 insertions(+), 13 deletions(-)
tools/perf/util/metricgroup.c +0 -2
index 40a1e14de418..46bf4dfeebc8 100644
@@ -1562,8 +1562,6 @@ 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");
 

tools/perf/util/metricgroup.c static int metricgroup__has_metric_or_groups_callback(const struct pmu_metric *p · line 1602

Quoted 8 lines
tools/perf/util/metricgroup.c +1 -4
@@ -1601,22 +1599,16 @@ static int metricgroup__has_metric_or_groups_callback(const struct pmu_metric *p
 
 bool metricgroup__has_metric_or_groups(const char *pmu, const char *metric_or_groups)
 {
-	const struct pmu_metrics_table *tables[2] = {
-		pmu_metrics_table__find(),
-		pmu_metrics_table__default(),
-	};
+	const struct pmu_metrics_table *table = pmu_metrics_table__find();

Here should be:

  const struct pmu_metrics_table *table = pmu_metrics_table__default();

With this change, the default metrics can work on my side:

Performance counter stats for 'sleep 1':

                 1      context-switches                 #    434.8 cs/sec  cs_per_second     
                 0      cpu-migrations                   #      0.0 migrations/sec  migrations_per_second
                70      page-faults                      #  30439.1 faults/sec  page_faults_per_second
              2.30 msec task-clock                       #      0.0 CPUs  CPUs_utilized       
             17398      armv8_pmuv3_0/branch-misses/     #      4.3 %  branch_miss_rate       
            401746      armv8_pmuv3_0/branches/          #    174.7 M/sec  branch_frequency   
           1778245      armv8_pmuv3_0/cpu-cycles/        #      0.8 GHz  cycles_frequency     
           2341698      armv8_pmuv3_0/instructions/      #      1.3 instructions  insn_per_cycle
            680491      armv8_pmuv3_0/stalled-cycles-frontend/ #     0.38 frontend_cycles_idle      
            380360      armv8_pmuv3_0/stalled-cycles-backend/ #     0.21 backend_cycles_idle       
            380360      armv8_pmuv3_0/stalled-cycles-backend/ #     0.29 stalled_cycles_per_instruction

With above change:

Quoted 23 lines
tools/perf/util/metricgroup.c +4 -7
 	struct metricgroup__has_metric_data data = {
 		.pmu = pmu,
 		.metric_or_groups = metric_or_groups,
 	};
 
-	for (size_t i = 0; i < ARRAY_SIZE(tables); i++) {
-		if (pmu_metrics_table__for_each_metric(tables[i],
-							metricgroup__has_metric_or_groups_callback,
-							&data))
-			return true;
-	}
-	return false;
+	return pmu_metrics_table__for_each_metric(table,
+						  metricgroup__has_metric_or_groups_callback,
+						  &data)
+		? true : false;
 }
 
 static int metricgroup__topdown_max_level_callback(const struct pmu_metric *pm,
Signature · 1 line
2.53.0.rc2.204.g2597b5adb4-goog
#3 lore raw

Re: [PATCH v1 2/2] perf metricgroup: Don't early exit if no CPUID table exists

Quoted 38 lines · Leo Yan
Quoted 26 lines · Ian Rogers

The failure to find a table of metrics with a CPUID shouldn't early exit as the metric code will now also consider the default table. When searching for a metric or metric group, pmu_metrics_table__for_each_metric considers all tables and so the caller doesn't need to switch the table to do this.

Fixes c7adeb0974f1 ("perf jevents: Add set of common metrics based on default ones")
---
 tools/perf/util/metricgroup.c | 18 +++++-------------
 1 file changed, 5 insertions(+), 13 deletions(-)
tools/perf/util/metricgroup.c +0 -2
index 40a1e14de418..46bf4dfeebc8 100644
@@ -1562,8 +1562,6 @@ 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");
 

tools/perf/util/metricgroup.c static int metricgroup__has_metric_or_groups_callback(const struct pmu_metric *p · line 1602

Quoted 8 lines
tools/perf/util/metricgroup.c +1 -4
@@ -1601,22 +1599,16 @@ static int metricgroup__has_metric_or_groups_callback(const struct pmu_metric *p
 
 bool metricgroup__has_metric_or_groups(const char *pmu, const char *metric_or_groups)
 {
-	const struct pmu_metrics_table *tables[2] = {
-		pmu_metrics_table__find(),
-		pmu_metrics_table__default(),
-	};
+	const struct pmu_metrics_table *table = pmu_metrics_table__find();

Here should be:

Quoted 1 line
  const struct pmu_metrics_table *table = pmu_metrics_table__default();

Or, I think you should not change metricgroup__has_metric_or_groups(). The change only in metricgroup__parse_groups() can fix my issue.

Thanks,
Leo

author reviewer
#4 lore raw

Re: [PATCH v1 2/2] perf metricgroup: Don't early exit if no CPUID table exists

Quoted 40 lines · Leo Yan
Quoted 38 lines · Leo Yan
Quoted 26 lines · Ian Rogers

The failure to find a table of metrics with a CPUID shouldn't early exit as the metric code will now also consider the default table. When searching for a metric or metric group, pmu_metrics_table__for_each_metric considers all tables and so the caller doesn't need to switch the table to do this.

Fixes c7adeb0974f1 ("perf jevents: Add set of common metrics based on default ones")
---
 tools/perf/util/metricgroup.c | 18 +++++-------------
 1 file changed, 5 insertions(+), 13 deletions(-)
tools/perf/util/metricgroup.c +0 -2
index 40a1e14de418..46bf4dfeebc8 100644
@@ -1562,8 +1562,6 @@ 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");
 

tools/perf/util/metricgroup.c static int metricgroup__has_metric_or_groups_callback(const struct pmu_metric *p · line 1602

Quoted 8 lines
tools/perf/util/metricgroup.c +1 -4
@@ -1601,22 +1599,16 @@ static int metricgroup__has_metric_or_groups_callback(const struct pmu_metric *p
 
 bool metricgroup__has_metric_or_groups(const char *pmu, const char *metric_or_groups)
 {
-   const struct pmu_metrics_table *tables[2] = {
-           pmu_metrics_table__find(),
-           pmu_metrics_table__default(),
-   };
+   const struct pmu_metrics_table *table = pmu_metrics_table__find();

Here should be:

Quoted 4 lines
Quoted 1 line
  const struct pmu_metrics_table *table = pmu_metrics_table__default();

Or, I think you should not change metricgroup__has_metric_or_groups(). The change only in metricgroup__parse_groups() can fix my issue.

This is wrong and the patch I tested I believe to be correct. The table here is CPUID associated table and all the metric code will bottom out in metricgroup__for_each_metric where the CPUID table is the table parameter (its a parameter because of testing where we force the machine/model to "testarch"):
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/metricgroup.c?h=perf-tools-next#n430

```
int metricgroup__for_each_metric(const struct pmu_metrics_table
*table, pmu_metric_iter_fn fn, void *data)
{
...
const struct pmu_metrics_table *tables[2] = {
    table,
    pmu_metrics_table__default(),
};

for (size_t i = 0; i < ARRAY_SIZE(tables); i++) {
    int ret;

    if (!tables[i])
        continue;
```

If you make the table here the default table then you will process the default table twice in metricgroup__for_each_metric which will mean double metric look ups, etc. So I think the patch is correct and passed my testing both with and without NO_JEVENTS=1.

Thanks,
Ian

Quoted 2 lines

Thanks,
Leo

#5 lore raw

Re: [PATCH v1 2/2] perf metricgroup: Don't early exit if no CPUID table exists

On Fri, Feb 06, 2026 at 10:57:54AM -0800, Ian Rogers wrote:

[...]

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines

bool metricgroup__has_metric_or_groups(const char *pmu, const char *metric_or_groups) {

tools/perf/util/metricgroup.c +1 -4
-   const struct pmu_metrics_table *tables[2] = {
-           pmu_metrics_table__find(),
-           pmu_metrics_table__default(),
-   };
+   const struct pmu_metrics_table *table = pmu_metrics_table__find();
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Here should be:

  const struct pmu_metrics_table *table = pmu_metrics_table__default();

Or, I think you should not change metricgroup__has_metric_or_groups(). The change only in metricgroup__parse_groups() can fix my issue.

This is wrong and the patch I tested I believe to be correct.

It is very likely you are not working on the issue same as mine.

Quoted 1 line

The table here is CPUID associated table and all the metric code will

As you said, the key point is CPUID. On my board, as no corresponding json file to support the CPU variant, it fails to find any table based on the CPUID and pmu_metrics_table__find() always returns NULL.

The failure flow is:

add_default_events()
if (!metricgroup__has_metric_or_groups(pmu, default_metricgroup_names[i]))
continue;
^^^
Always return false

As a result, metricgroup__parse_groups() has no chance to be invoked and no any events are added to the ev list.

This is why I suggested to keep default table in metricgroup__has_metric_or_groups(), it can rallback to find default table when CPUID is not supported.

Quoted 20 lines

bottom out in metricgroup__for_each_metric where the CPUID table is the table parameter (its a parameter because of testing where we force the machine/model to "testarch"):
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/metricgroup.c?h=perf-tools-next#n430

```
int metricgroup__for_each_metric(const struct pmu_metrics_table
*table, pmu_metric_iter_fn fn, void *data)
{
...
const struct pmu_metrics_table *tables[2] = {
    table,
    pmu_metrics_table__default(),
};

for (size_t i = 0; i < ARRAY_SIZE(tables); i++) {
    int ret;

    if (!tables[i])
        continue;
```
Quoted 3 lines

If you make the table here the default table then you will process the default table twice in metricgroup__for_each_metric which will mean double metric look ups, etc.

In my case, it will not have double metric issue, as pmu_metrics_table__find() cannot find any table, a NULL table pointer will be passed to metricgroup__for_each_metric(), then only pmu_metrics_table__default() will be used.

As a side question, using a single CPUID to look up metrics can be error-prone on asymmetric systems (such as Arm big.LITTLE). Since only the first CPU ID is used to search the metrics table, could this cause issues for other CPU variants?

Thanks,
Leo

author reviewer
#6 lore raw

Re: [PATCH v1 2/2] perf metricgroup: Don't early exit if no CPUID table exists

Quoted 69 lines · Leo Yan

On Fri, Feb 06, 2026 at 10:57:54AM -0800, Ian Rogers wrote:

[...]

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines

bool metricgroup__has_metric_or_groups(const char *pmu, const char *metric_or_groups) {

tools/perf/util/metricgroup.c +1 -4
-   const struct pmu_metrics_table *tables[2] = {
-           pmu_metrics_table__find(),
-           pmu_metrics_table__default(),
-   };
+   const struct pmu_metrics_table *table = pmu_metrics_table__find();
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Here should be:

  const struct pmu_metrics_table *table = pmu_metrics_table__default();

Or, I think you should not change metricgroup__has_metric_or_groups(). The change only in metricgroup__parse_groups() can fix my issue.

This is wrong and the patch I tested I believe to be correct.

It is very likely you are not working on the issue same as mine.

Quoted 1 line

The table here is CPUID associated table and all the metric code will

As you said, the key point is CPUID. On my board, as no corresponding json file to support the CPU variant, it fails to find any table based on the CPUID and pmu_metrics_table__find() always returns NULL.

The failure flow is:

add_default_events()
if (!metricgroup__has_metric_or_groups(pmu, default_metricgroup_names[i]))
continue;
^^^
Always return false

As a result, metricgroup__parse_groups() has no chance to be invoked and no any events are added to the ev list.

This is why I suggested to keep default table in metricgroup__has_metric_or_groups(), it can rallback to find default table when CPUID is not supported.

Quoted 20 lines

bottom out in metricgroup__for_each_metric where the CPUID table is the table parameter (its a parameter because of testing where we force the machine/model to "testarch"):
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/metricgroup.c?h=perf-tools-next#n430

```
int metricgroup__for_each_metric(const struct pmu_metrics_table
*table, pmu_metric_iter_fn fn, void *data)
{
...
const struct pmu_metrics_table *tables[2] = {
    table,
    pmu_metrics_table__default(),
};

for (size_t i = 0; i < ARRAY_SIZE(tables); i++) {
    int ret;

    if (!tables[i])
        continue;
```
Quoted 3 lines

If you make the table here the default table then you will process the default table twice in metricgroup__for_each_metric which will mean double metric look ups, etc.

Quoted 4 lines

In my case, it will not have double metric issue, as pmu_metrics_table__find() cannot find any table, a NULL table pointer will be passed to metricgroup__for_each_metric(), then only pmu_metrics_table__default() will be used.

Which is the table you want to be used to such an extent that you are proposing changing this patch to put it twice in the tables array in the snippet of code above. Also note that the code skips the NULL, failed to look up table, which is why this patch just needs to avoid the early return on a missing table to address the problem of the default table not being considered.

Quoted 4 lines

As a side question, using a single CPUID to look up metrics can be error-prone on asymmetric systems (such as Arm big.LITTLE). Since only the first CPU ID is used to search the metrics table, could this cause issues for other CPU variants?

There's logic to use the PMU for the cpumask and then to get the CPUID for the ARM CPU associated with the CPUs in the cpumask. It is likely bitrotten, but the problem has had some work on it in the past. It is untested, to my knowledge.

Thanks,
Ian

Quoted 2 lines

Thanks,
Leo

author reviewer
#7 lore raw

Re: [PATCH v1 2/2] perf metricgroup: Don't early exit if no CPUID table exists

Quoted 88 lines · Ian Rogers
Quoted 69 lines · Leo Yan

On Fri, Feb 06, 2026 at 10:57:54AM -0800, Ian Rogers wrote:

[...]

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines

bool metricgroup__has_metric_or_groups(const char *pmu, const char *metric_or_groups) {

tools/perf/util/metricgroup.c +1 -4
-   const struct pmu_metrics_table *tables[2] = {
-           pmu_metrics_table__find(),
-           pmu_metrics_table__default(),
-   };
+   const struct pmu_metrics_table *table = pmu_metrics_table__find();
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Here should be:

  const struct pmu_metrics_table *table = pmu_metrics_table__default();

Or, I think you should not change metricgroup__has_metric_or_groups(). The change only in metricgroup__parse_groups() can fix my issue.

This is wrong and the patch I tested I believe to be correct.

It is very likely you are not working on the issue same as mine.

Quoted 1 line

The table here is CPUID associated table and all the metric code will

As you said, the key point is CPUID. On my board, as no corresponding json file to support the CPU variant, it fails to find any table based on the CPUID and pmu_metrics_table__find() always returns NULL.

The failure flow is:

add_default_events()
if (!metricgroup__has_metric_or_groups(pmu, default_metricgroup_names[i]))
continue;
^^^
Always return false

As a result, metricgroup__parse_groups() has no chance to be invoked and no any events are added to the ev list.

This is why I suggested to keep default table in metricgroup__has_metric_or_groups(), it can rallback to find default table when CPUID is not supported.

Quoted 20 lines

bottom out in metricgroup__for_each_metric where the CPUID table is the table parameter (its a parameter because of testing where we force the machine/model to "testarch"):
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/tools/perf/util/metricgroup.c?h=perf-tools-next#n430

```
int metricgroup__for_each_metric(const struct pmu_metrics_table
*table, pmu_metric_iter_fn fn, void *data)
{
...
const struct pmu_metrics_table *tables[2] = {
    table,
    pmu_metrics_table__default(),
};

for (size_t i = 0; i < ARRAY_SIZE(tables); i++) {
    int ret;

    if (!tables[i])
        continue;
```
Quoted 3 lines

If you make the table here the default table then you will process the default table twice in metricgroup__for_each_metric which will mean double metric look ups, etc.

Quoted 4 lines

In my case, it will not have double metric issue, as pmu_metrics_table__find() cannot find any table, a NULL table pointer will be passed to metricgroup__for_each_metric(), then only pmu_metrics_table__default() will be used.

Which is the table you want to be used to such an extent that you are proposing changing this patch to put it twice in the tables array in the snippet of code above. Also note that the code skips the NULL, failed to look up table, which is why this patch just needs to avoid the early return on a missing table to address the problem of the default table not being considered.

Quoted 4 lines

As a side question, using a single CPUID to look up metrics can be error-prone on asymmetric systems (such as Arm big.LITTLE). Since only the first CPU ID is used to search the metrics table, could this cause issues for other CPU variants?

Quoted 4 lines

There's logic to use the PMU for the cpumask and then to get the CPUID for the ARM CPU associated with the CPUs in the cpumask. It is likely bitrotten, but the problem has had some work on it in the past. It is untested, to my knowledge.

Ah, the issue is that metricgroup__has_metric_or_groups is using pmu_metrics_table__for_each_metric and not metricgroup__for_each_metric:

 ```
-       return pmu_metrics_table__for_each_metric(table,
-
metricgroup__has_metric_or_groups_callback,
-                                                 &data)
+       return metricgroup__for_each_metric(table,
+
metricgroup__has_metric_or_groups_callback,
+
```

I'll send this out.

Thanks,
Ian

Quoted 5 lines

Thanks,
Ian

Quoted 2 lines

Thanks,
Leo

author reviewer
#8 lore raw

PATCH 1/1 perf metricgroup: Fix metricgroup__has_metric_or_groups

Use metricgroup__for_each_metric rather than pmu_metrics_table__for_each_metric that combines the default metric table with, a potentially empty, CPUID table.

Fixes cee275edcdb1 ("perf metricgroup: Don't early exit if no CPUID table exists")
---
 tools/perf/util/metricgroup.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
tools/perf/util/metricgroup.c +3 -3
index 46bf4dfeebc8..7e39d469111b 100644
@@ -1605,9 +1605,9 @@ bool metricgroup__has_metric_or_groups(const char *pmu, const char *metric_or_gr
 		.metric_or_groups = metric_or_groups,
 	};
 
-	return pmu_metrics_table__for_each_metric(table,
-						  metricgroup__has_metric_or_groups_callback,
-						  &data)
+	return metricgroup__for_each_metric(table,
+					    metricgroup__has_metric_or_groups_callback,
+					    &data)
 		? true : false;
 }
 
Signature · 1 line
2.53.0.239.g8d8fc8a987-goog
Reviewed-by Tested-by on v1 1/1
#9 lore raw

Re: [PATCH v1] perf metricgroup: Fix metricgroup__has_metric_or_groups

Quoted 6 lines · Ian Rogers

Use metricgroup__for_each_metric rather than pmu_metrics_table__for_each_metric that combines the default metric table with, a potentially empty, CPUID table.

Fixes cee275edcdb1 ("perf metricgroup: Don't early exit if no CPUID table exists")
author reviewer
#10 lore raw

Re: [PATCH v1] perf metricgroup: Fix metricgroup__has_metric_or_groups

Quoted 11 lines · Leo Yan
Quoted 6 lines · Ian Rogers

Use metricgroup__for_each_metric rather than pmu_metrics_table__for_each_metric that combines the default metric table with, a potentially empty, CPUID table.

Fixes cee275edcdb1 ("perf metricgroup: Don't early exit if no CPUID table exists")

Thanks! Sorry again for the earlier confusion on this.

Ian

#11 lore raw

Re: [PATCH v1] perf metricgroup: Fix metricgroup__has_metric_or_groups

Quoted 7 lines · Ian Rogers
Quoted 6 lines · Leo Yan
Quoted 4 lines · Ian Rogers

Use metricgroup__for_each_metric rather than pmu_metrics_table__for_each_metric that combines the default metric table with, a potentially empty, CPUID table.

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
Fixes cee275edcdb1 ("perf metricgroup: Don't early exit if no CPUID table exists")

Thanks! Sorry again for the earlier confusion on this.

No worries, thanks for quick fix.

maintainer Applied
#12 lore raw

Re: [PATCH v1] perf metricgroup: Fix metricgroup__has_metric_or_groups

In reply to Ian Rogers
Quoted 5 lines · Ian Rogers

Use metricgroup__for_each_metric rather than pmu_metrics_table__for_each_metric that combines the default metric table with, a potentially empty, CPUID table.

Applied to perf-tools-next, thanks!

Best regards,
Namhyung

Reviewed-by on v1 1/2
#13 lore raw
In reply to Ian Rogers
Quoted 2 lines · Ian Rogers

Leo reported the NO_JEVENTS=1 build being broken and this highlighted it being missing from the build-test, add it.

To be clear, I did not say the build is broken.

The issue I reported is that perf stat is broken, both with NO_JEVENTS=1 enabled and without that configuration.

With updating the commit log, this patch LGTM:

Quoted 30 lines
---
 tools/perf/tests/make | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/tests/make +2 -0
index eb41516c0562..6587dc326d1b 100644
@@ -70,6 +70,7 @@ make_python_perf_so := $(python_perf_so)
 make_debug          := DEBUG=1
 make_nondistro      := BUILD_NONDISTRO=1
 make_extra_tests    := EXTRA_TESTS=1
+make_no_jevents     := NO_JEVENTS=1
 make_jevents_all    := JEVENTS_ARCH=all
 make_no_bpf_skel    := BUILD_BPF_SKEL=0
 make_gen_vmlinux_h  := GEN_VMLINUX_H=1
@@ -144,6 +145,7 @@ ifneq ($(new_libbfd),)
 run += make_nondistro
 endif
 run += make_extra_tests
+run += make_no_jevents
 run += make_jevents_all
 run += make_no_bpf_skel
 run += make_gen_vmlinux_h
Signature · 1 line
2.53.0.rc2.204.g2597b5adb4-goog
maintainer Applied
#14 lore raw
Quoted 4 lines · Leo Yan
Quoted 2 lines · Ian Rogers

Leo reported the NO_JEVENTS=1 build being broken and this highlighted it being missing from the build-test, add it.

Quoted 8 lines

To be clear, I did not say the build is broken.

The issue I reported is that perf stat is broken, both with NO_JEVENTS=1 enabled and without that configuration.

With updating the commit log, this patch LGTM:

Thanks, applied to perf-tools-next,

- Arnaldo

Quoted 30 lines
Quoted 30 lines
---
 tools/perf/tests/make | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/tests/make +2 -0
index eb41516c0562..6587dc326d1b 100644
@@ -70,6 +70,7 @@ make_python_perf_so := $(python_perf_so)
 make_debug          := DEBUG=1
 make_nondistro      := BUILD_NONDISTRO=1
 make_extra_tests    := EXTRA_TESTS=1
+make_no_jevents     := NO_JEVENTS=1
 make_jevents_all    := JEVENTS_ARCH=all
 make_no_bpf_skel    := BUILD_BPF_SKEL=0
 make_gen_vmlinux_h  := GEN_VMLINUX_H=1
@@ -144,6 +145,7 @@ ifneq ($(new_libbfd),)
 run += make_nondistro
 endif
 run += make_extra_tests
+run += make_no_jevents
 run += make_jevents_all
 run += make_no_bpf_skel
 run += make_gen_vmlinux_h
Signature · 1 line
2.53.0.rc2.204.g2597b5adb4-goog
#15 lore raw
In reply to Leo Yan
Quoted 10 lines · Leo Yan
Quoted 2 lines · Ian Rogers

Leo reported the NO_JEVENTS=1 build being broken and this highlighted it being missing from the build-test, add it.

To be clear, I did not say the build is broken.

The issue I reported is that perf stat is broken, both with NO_JEVENTS=1 enabled and without that configuration.

With updating the commit log, this patch LGTM:

I did it and will apply the Reviewed-by to patch 2/2 too, ok?

- Arnaldo

Quoted 32 lines
Quoted 30 lines
---
 tools/perf/tests/make | 2 ++
 1 file changed, 2 insertions(+)
tools/perf/tests/make +2 -0
index eb41516c0562..6587dc326d1b 100644
@@ -70,6 +70,7 @@ make_python_perf_so := $(python_perf_so)
 make_debug          := DEBUG=1
 make_nondistro      := BUILD_NONDISTRO=1
 make_extra_tests    := EXTRA_TESTS=1
+make_no_jevents     := NO_JEVENTS=1
 make_jevents_all    := JEVENTS_ARCH=all
 make_no_bpf_skel    := BUILD_BPF_SKEL=0
 make_gen_vmlinux_h  := GEN_VMLINUX_H=1
@@ -144,6 +145,7 @@ ifneq ($(new_libbfd),)
 run += make_nondistro
 endif
 run += make_extra_tests
+run += make_no_jevents
 run += make_jevents_all
 run += make_no_bpf_skel
 run += make_gen_vmlinux_h
Signature · 1 line
2.53.0.rc2.204.g2597b5adb4-goog
#16 lore raw
Quoted 5 lines · Arnaldo Carvalho de Melo
Quoted 4 lines · Leo Yan
Quoted 2 lines · Ian Rogers

Leo reported the NO_JEVENTS=1 build being broken and this highlighted it being missing from the build-test, add it.

Quoted 8 lines
Quoted 6 lines

To be clear, I did not say the build is broken.

The issue I reported is that perf stat is broken, both with NO_JEVENTS=1 enabled and without that configuration.

With updating the commit log, this patch LGTM:

I did it and will apply the Reviewed-by to patch 2/2 too, ok?

Yes, this is fine for me.

Just remind, in patch 02, based on test, we only need to apply the change in metricgroup__parse_groups() for removing the "!table" check.

The change for metricgroup__has_metric_or_groups() is not quite right, as we should keep the searching metrics with pmu_metrics_table__default().

Thanks,
Leo