Skip to contentSkip to conversation

perf pmu: Recognize default_core as a core PMU in more places

PATCH Applied Ian Rogers 3 participants 6 messages Jun 12 – Jul 1 perf events

lore
Share & tools
lore

Ian Rogers

author reviewer
#0 lore raw

PATCH 1/1 perf pmu: Recognize default_core as a core PMU in more places

The python metrics code used in places like ilist.py passes a pmu-filter of "default_core" on non-hybrid x86/ARM/.. systems. As a PMU like "cpu" isn't a literal name match then no PMU matches "default_core" and the events fail to parse for the metric. Fix the name matching and PMU lookup for "default_core" and check that it fixes ilist.py.

Fixes 74e2dbe7be50 ("perf tools: Add --pmu-filter option for filtering PMUs")

---
Note: this bug is in 7.1 but it is a little too late to send for the
release. It should get picked up via the fixes tag.

---
 tools/perf/util/pmu.c  | 6 +++++-
 tools/perf/util/pmus.c | 2 ++
 2 files changed, 7 insertions(+), 1 deletion(-)
tools/perf/util/pmu.c +5 -1
index a550f030b85d..836e3b5615cd 100644
@@ -2660,8 +2660,12 @@ bool perf_pmu__wildcard_match(const struct perf_pmu *pmu, const char *wildcard_t
 		pmu->name,
 		pmu->alias_name,
 	};
-	bool need_fnmatch = strisglob(wildcard_to_match);
+	bool need_fnmatch;
 
+	if (pmu->is_core && !strcmp(wildcard_to_match, "default_core"))
+		return true;
+
+	need_fnmatch = strisglob(wildcard_to_match);
 	if (!strncmp(wildcard_to_match, "uncore_", 7))
 		wildcard_to_match += 7;
 
tools/perf/util/pmus.c +2 -0
index 5e3f571450fe..e0a4cb2428ca 100644
@@ -150,6 +150,8 @@ struct perf_pmu *perf_pmus__find(const char *name)
 	bool core_pmu;
 	unsigned int to_read_pmus = 0;
 
+	if (!strcmp(name, "default_core"))
+		return perf_pmus__find_core_pmu();
 	/*
 	 * Once PMU is loaded it stays in the list,
 	 * so we keep us from multiple reading/parsing
Signature · 1 line
2.54.0.1136.gdb2ca164c4-goog
#1 lore raw
Quoted 40 lines · Ian Rogers

The python metrics code used in places like ilist.py passes a pmu-filter of "default_core" on non-hybrid x86/ARM/.. systems. As a PMU like "cpu" isn't a literal name match then no PMU matches "default_core" and the events fail to parse for the metric. Fix the name matching and PMU lookup for "default_core" and check that it fixes ilist.py.

Fixes 74e2dbe7be50 ("perf tools: Add --pmu-filter option for filtering PMUs")

---
Note: this bug is in 7.1 but it is a little too late to send for the
release. It should get picked up via the fixes tag.

---
  tools/perf/util/pmu.c  | 6 +++++-
  tools/perf/util/pmus.c | 2 ++
  2 files changed, 7 insertions(+), 1 deletion(-)
tools/perf/util/pmu.c +5 -1
index a550f030b85d..836e3b5615cd 100644
@@ -2660,8 +2660,12 @@ bool perf_pmu__wildcard_match(const struct perf_pmu *pmu, const char *wildcard_t
  		pmu->name,
  		pmu->alias_name,
  	};
-	bool need_fnmatch = strisglob(wildcard_to_match);
+	bool need_fnmatch;
 
+	if (pmu->is_core && !strcmp(wildcard_to_match, "default_core"))
+		return true;
+
+	need_fnmatch = strisglob(wildcard_to_match);
  	if (!strncmp(wildcard_to_match, "uncore_", 7))
  		wildcard_to_match += 7;
 
tools/perf/util/pmus.c +0 -0
index 5e3f571450fe..e0a4cb2428ca 100644

tools/perf/util/pmus.c struct perf_pmu *perf_pmus__find(const char *name) · line 157

Quoted 8 lines
tools/perf/util/pmus.c +2 -0
@@ -150,6 +150,8 @@ struct perf_pmu *perf_pmus__find(const char *name)
  	bool core_pmu;
  	unsigned int to_read_pmus = 0;
 
+	if (!strcmp(name, "default_core"))
+		return perf_pmus__find_core_pmu();
  	/*
  	 * Once PMU is loaded it stays in the list,
  	 * so we keep us from multiple reading/parsing

Thank you for the fix, Ian.
In the previous patch, I overlooked the handling of the "default_core" alias. Also, perhaps we can leave need_fnmatch as-is, since it doesn't seem to affect anything here.

Signature · 2 lines
Thanks,
Qinxin
author reviewer
#2 lore raw
Quoted 47 lines · Qinxin Xia
Quoted 43 lines · Ian Rogers

The python metrics code used in places like ilist.py passes a pmu-filter of "default_core" on non-hybrid x86/ARM/.. systems. As a PMU like "cpu" isn't a literal name match then no PMU matches "default_core" and the events fail to parse for the metric. Fix the name matching and PMU lookup for "default_core" and check that it fixes ilist.py.

Fixes 74e2dbe7be50 ("perf tools: Add --pmu-filter option for filtering PMUs")

---
Note: this bug is in 7.1 but it is a little too late to send for the
release. It should get picked up via the fixes tag.

---
  tools/perf/util/pmu.c  | 6 +++++-
  tools/perf/util/pmus.c | 2 ++
  2 files changed, 7 insertions(+), 1 deletion(-)
tools/perf/util/pmu.c +5 -1
index a550f030b85d..836e3b5615cd 100644
@@ -2660,8 +2660,12 @@ bool perf_pmu__wildcard_match(const struct perf_pmu *pmu, const char *wildcard_t
              pmu->name,
              pmu->alias_name,
      };
-     bool need_fnmatch = strisglob(wildcard_to_match);
+     bool need_fnmatch;
 
+     if (pmu->is_core && !strcmp(wildcard_to_match, "default_core"))
+             return true;
+
+     need_fnmatch = strisglob(wildcard_to_match);
      if (!strncmp(wildcard_to_match, "uncore_", 7))
              wildcard_to_match += 7;
 
tools/perf/util/pmus.c +0 -0
index 5e3f571450fe..e0a4cb2428ca 100644
@@ -150,6 +150,8 @@ struct perf_pmu *perf_pmus__find(const char *name)
      bool core_pmu;
      unsigned int to_read_pmus = 0;
 
Quoted 10 lines

tools/perf/util/pmus.c struct perf_pmu *perf_pmus__find(const char *name) · line 157

Quoted 5 lines
tools/perf/util/pmus.c +2 -0
+     if (!strcmp(name, "default_core"))
+             return perf_pmus__find_core_pmu();
      /*
       * Once PMU is loaded it stays in the list,
       * so we keep us from multiple reading/parsing

Thank you for the fix, Ian.
In the previous patch, I overlooked the handling of the "default_core" alias. Also, perhaps we can leave need_fnmatch as-is, since it doesn't seem to affect anything here.

Thanks. I moved need_fnmatch's initialization just to save some work in the "default_core" case where the value is unused. While leaving it in place would reduce the diff, I think the optimization is worth it as we would otherwise call strisglob when say iterating all metrics.

Would it be okay to supply a Reviewed-by tag? I notice this fix is missing from perf-tools-next.

Thanks!
Ian

Quoted 4 lines

--
Thanks,
Qinxin

#3 lore raw
Quoted 56 lines · Ian Rogers
Quoted 46 lines · Qinxin Xia
Quoted 43 lines · Ian Rogers

The python metrics code used in places like ilist.py passes a pmu-filter of "default_core" on non-hybrid x86/ARM/.. systems. As a PMU like "cpu" isn't a literal name match then no PMU matches "default_core" and the events fail to parse for the metric. Fix the name matching and PMU lookup for "default_core" and check that it fixes ilist.py.

Fixes 74e2dbe7be50 ("perf tools: Add --pmu-filter option for filtering PMUs")

---
Note: this bug is in 7.1 but it is a little too late to send for the
release. It should get picked up via the fixes tag.

---
   tools/perf/util/pmu.c  | 6 +++++-
   tools/perf/util/pmus.c | 2 ++
   2 files changed, 7 insertions(+), 1 deletion(-)
tools/perf/util/pmu.c +5 -1
index a550f030b85d..836e3b5615cd 100644
@@ -2660,8 +2660,12 @@ bool perf_pmu__wildcard_match(const struct perf_pmu *pmu, const char *wildcard_t
               pmu->name,
               pmu->alias_name,
       };
-     bool need_fnmatch = strisglob(wildcard_to_match);
+     bool need_fnmatch;
 
+     if (pmu->is_core && !strcmp(wildcard_to_match, "default_core"))
+             return true;
+
+     need_fnmatch = strisglob(wildcard_to_match);
       if (!strncmp(wildcard_to_match, "uncore_", 7))
               wildcard_to_match += 7;
 
tools/perf/util/pmus.c +0 -0
index 5e3f571450fe..e0a4cb2428ca 100644
@@ -150,6 +150,8 @@ struct perf_pmu *perf_pmus__find(const char *name)
       bool core_pmu;
       unsigned int to_read_pmus = 0;
 
Quoted 9 lines

tools/perf/util/pmus.c struct perf_pmu *perf_pmus__find(const char *name) · line 157

Quoted 5 lines
tools/perf/util/pmus.c +2 -0
+     if (!strcmp(name, "default_core"))
+             return perf_pmus__find_core_pmu();
       /*
        * Once PMU is loaded it stays in the list,
        * so we keep us from multiple reading/parsing

Thank you for the fix, Ian.
In the previous patch, I overlooked the handling of the "default_core" alias. Also, perhaps we can leave need_fnmatch as-is, since it doesn't seem to affect anything here.

Quoted 5 lines

Thanks. I moved need_fnmatch's initialization just to save some work in the "default_core" case where the value is unused. While leaving it in place would reduce the diff, I think the optimization is worth it as we would otherwise call strisglob when say iterating all metrics.

Perhaps we can do it this way:

	if (pmu->is_core && !strcmp(wildcard_to_match, "default_core"))
             	return true;

	bool need_fnmatch = strisglob(wildcard_to_match);
Quoted 3 lines

Would it be okay to supply a Reviewed-by tag? I notice this fix is missing from perf-tools-next.

Sure, add my Reviewed‑by.:-)

Quoted 6 lines

Thanks!
Ian

Quoted 3 lines

--
Thanks,
Qinxin

Signature · 2 lines
Thanks,
Qinxin
author reviewer
#4 lore raw
Quoted 65 lines · Qinxin Xia
Quoted 56 lines · Ian Rogers
Quoted 46 lines · Qinxin Xia
Quoted 43 lines · Ian Rogers

The python metrics code used in places like ilist.py passes a pmu-filter of "default_core" on non-hybrid x86/ARM/.. systems. As a PMU like "cpu" isn't a literal name match then no PMU matches "default_core" and the events fail to parse for the metric. Fix the name matching and PMU lookup for "default_core" and check that it fixes ilist.py.

Fixes 74e2dbe7be50 ("perf tools: Add --pmu-filter option for filtering PMUs")

---
Note: this bug is in 7.1 but it is a little too late to send for the
release. It should get picked up via the fixes tag.

---
   tools/perf/util/pmu.c  | 6 +++++-
   tools/perf/util/pmus.c | 2 ++
   2 files changed, 7 insertions(+), 1 deletion(-)
tools/perf/util/pmu.c +5 -1
index a550f030b85d..836e3b5615cd 100644
@@ -2660,8 +2660,12 @@ bool perf_pmu__wildcard_match(const struct perf_pmu *pmu, const char *wildcard_t
               pmu->name,
               pmu->alias_name,
       };
-     bool need_fnmatch = strisglob(wildcard_to_match);
+     bool need_fnmatch;
 
+     if (pmu->is_core && !strcmp(wildcard_to_match, "default_core"))
+             return true;
+
+     need_fnmatch = strisglob(wildcard_to_match);
       if (!strncmp(wildcard_to_match, "uncore_", 7))
               wildcard_to_match += 7;
 
tools/perf/util/pmus.c +0 -0
index 5e3f571450fe..e0a4cb2428ca 100644
@@ -150,6 +150,8 @@ struct perf_pmu *perf_pmus__find(const char *name)
       bool core_pmu;
       unsigned int to_read_pmus = 0;
 
Quoted 9 lines

tools/perf/util/pmus.c struct perf_pmu *perf_pmus__find(const char *name) · line 157

Quoted 5 lines
tools/perf/util/pmus.c +2 -0
+     if (!strcmp(name, "default_core"))
+             return perf_pmus__find_core_pmu();
       /*
        * Once PMU is loaded it stays in the list,
        * so we keep us from multiple reading/parsing

Thank you for the fix, Ian.
In the previous patch, I overlooked the handling of the "default_core" alias. Also, perhaps we can leave need_fnmatch as-is, since it doesn't seem to affect anything here.

Quoted 5 lines

Thanks. I moved need_fnmatch's initialization just to save some work in the "default_core" case where the value is unused. While leaving it in place would reduce the diff, I think the optimization is worth it as we would otherwise call strisglob when say iterating all metrics.

Quoted 5 lines

Perhaps we can do it this way:

        if (pmu->is_core && !strcmp(wildcard_to_match, "default_core"))
                return true;

        bool need_fnmatch = strisglob(wildcard_to_match);

Unfortunately we require variables to be declared at the top of blocks/functions.

Quoted 4 lines
Quoted 3 lines

Would it be okay to supply a Reviewed-by tag? I notice this fix is missing from perf-tools-next.

Sure, add my Reviewed‑by.:-)

Thanks,
Ian

Quoted 12 lines
Quoted 7 lines

Thanks!
Ian

Quoted 3 lines

--
Thanks,
Qinxin

--
Thanks,
Qinxin

maintainer Applied
#5 lore raw
In reply to Ian Rogers
Quoted 8 lines · Ian Rogers

The python metrics code used in places like ilist.py passes a pmu-filter of "default_core" on non-hybrid x86/ARM/.. systems. As a PMU like "cpu" isn't a literal name match then no PMU matches "default_core" and the events fail to parse for the metric. Fix the name matching and PMU lookup for "default_core" and check that it fixes ilist.py.

[...]

Applied to perf-tools-next, thanks!

Best regards,
Namhyung