Skip to contentSkip to conversation

perf header: Switch "cpu" for find_core_pmu in caps feature writing

PATCH Applied Ian Rogers 5 participants 11 messages Nov 14, 2025 – Nov 27, 2025 perf events

lore
Share & tools
lore

Ian Rogers

author reviewer
#0 lore raw

PATCH 1/3 perf header: Switch "cpu" for find_core_pmu in caps feature writing

Writing currently fails on non-x86 and hybrid CPUs. Switch to the more regular find_core_pmu that is normally used in this case. Tested on hybrid alderlake system.

---
 tools/perf/util/header.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/header.c +1 -1
index a2f808841b33..e69b271f8073 100644
@@ -1560,7 +1560,7 @@ static int __write_pmu_caps(struct feat_fd *ff, struct perf_pmu *pmu,
 static int write_cpu_pmu_caps(struct feat_fd *ff,
 			      struct evlist *evlist __maybe_unused)
 {
-	struct perf_pmu *cpu_pmu = perf_pmus__find("cpu");
+	struct perf_pmu *cpu_pmu = perf_pmus__find_core_pmu();
 	int ret;
 
 	if (!cpu_pmu)
Signature · 1 line
2.52.0.rc1.455.g30608eb744-goog
author reviewer
#1 lore raw

PATCH 2/3 perf pmu: Add PMU kind to simplify differentiating

Rather than perf_pmu__is_xxx calls, and a notion of kind so that a single call can be used.

---
 tools/perf/util/pmu.h | 34 ++++++++++++++++++++++++++++++++++
 1 file changed, 34 insertions(+)
tools/perf/util/pmu.h +34 -0
index 1ebcf0242af8..cfcaba3e2e75 100644
@@ -37,6 +37,19 @@ struct perf_pmu_caps {
 	struct list_head list;
 };
 
+enum pmu_kind {
+	/* A perf event syscall PMU. */
+	PERF_PMU_KIND_PE,
+	/* A perf tool provided DRM PMU. */
+	PERF_PMU_KIND_DRM,
+	/* A perf tool provided HWMON PMU. */
+	PERF_PMU_KIND_HWMON,
+	/* Perf tool provided PMU for tool events like time. */
+	PERF_PMU_KIND_TOOL,
+	/* A testing PMU kind. */
+	PERF_PMU_KIND_FAKE
+};
+
 enum {
 	PERF_PMU_TYPE_PE_START    = 0,
 	PERF_PMU_TYPE_PE_END      = 0xFFFDFFFF,
@@ -306,4 +319,25 @@ void perf_pmu__delete(struct perf_pmu *pmu);
 const char *perf_pmu__name_from_config(struct perf_pmu *pmu, u64 config);
 bool perf_pmu__is_fake(const struct perf_pmu *pmu);
 
+static inline enum pmu_kind perf_pmu__kind(const struct perf_pmu *pmu)
+{
+	__u32 type;
+
+	if (!pmu)
+		return PERF_PMU_KIND_PE;
+
+	type = pmu->type;
+	if (type <= PERF_PMU_TYPE_PE_END)
+		return PERF_PMU_KIND_PE;
+	if (type <= PERF_PMU_TYPE_DRM_END)
+		return PERF_PMU_KIND_DRM;
+	if (type <= PERF_PMU_TYPE_DRM_END)
+		return PERF_PMU_KIND_DRM;
+	if (type <= PERF_PMU_TYPE_HWMON_END)
+		return PERF_PMU_KIND_HWMON;
+	if (type == PERF_PMU_TYPE_TOOL)
+		return PERF_PMU_KIND_TOOL;
+	return PERF_PMU_KIND_FAKE;
+}
+
 #endif /* __PMU_H */
Signature · 1 line
2.52.0.rc1.455.g30608eb744-goog
#2 lore raw

Re: [PATCH v1 2/3] perf pmu: Add PMU kind to simplify differentiating

Quoted 11 lines · Ian Rogers
tools/perf/util/pmu.h +11 -0
+static inline enum pmu_kind perf_pmu__kind(const struct perf_pmu *pmu)
+{
+       __u32 type;
+
+       if (!pmu)
+               return PERF_PMU_KIND_PE;
+
+       type = pmu->type;
+       if (type <= PERF_PMU_TYPE_PE_END)
+               return PERF_PMU_KIND_PE;
+       if (type <= PERF_PMU_TYPE_DRM_END)

tools/perf/util/pmu.h void perf_pmu__delete(struct perf_pmu *pmu); · line 342

Quoted 10 lines
tools/perf/util/pmu.h +10 -0
+               return PERF_PMU_KIND_DRM;
+       if (type <= PERF_PMU_TYPE_DRM_END)
+               return PERF_PMU_KIND_DRM;
+       if (type <= PERF_PMU_TYPE_HWMON_END)
+               return PERF_PMU_KIND_HWMON;
+       if (type == PERF_PMU_TYPE_TOOL)
+               return PERF_PMU_KIND_TOOL;
+       return PERF_PMU_KIND_FAKE;
+}
+

Looks like there's a duplicate check for PERF_PMU_TYPE_DRM_END.

maintainer
#3 lore raw

Re: [PATCH v1 2/3] perf pmu: Add PMU kind to simplify differentiating

Hello,

Quoted 16 lines · Anubhav Shelat
Quoted 15 lines · Ian Rogers
tools/perf/util/pmu.h +15 -0
+static inline enum pmu_kind perf_pmu__kind(const struct perf_pmu *pmu)
+{
+       __u32 type;
+
+       if (!pmu)
+               return PERF_PMU_KIND_PE;
+
+       type = pmu->type;
+       if (type <= PERF_PMU_TYPE_PE_END)
+               return PERF_PMU_KIND_PE;
+       if (type <= PERF_PMU_TYPE_DRM_END)
+               return PERF_PMU_KIND_DRM;
+       if (type <= PERF_PMU_TYPE_DRM_END)
+               return PERF_PMU_KIND_DRM;
+       if (type <= PERF_PMU_TYPE_HWMON_END)
Quoted 8 lines

tools/perf/util/pmu.h void perf_pmu__delete(struct perf_pmu *pmu); · line 342

Quoted 6 lines
tools/perf/util/pmu.h +6 -0
+               return PERF_PMU_KIND_HWMON;
+       if (type == PERF_PMU_TYPE_TOOL)
+               return PERF_PMU_KIND_TOOL;
+       return PERF_PMU_KIND_FAKE;
+}
+

Looks like there's a duplicate check for PERF_PMU_TYPE_DRM_END.

Right, can you please send a fix?

Thanks,
Namhyung

#4 lore raw

PATCH 1/1 perf pmu: fix duplicate conditional statement

In reply to Ian Rogers

Remove duplicate check for PERF_PMU_TYPE_DRM_END in perf_pmu__kind.

Fixes cedf1f4852ae.

---
 tools/perf/util/pmu.h | 2 --
 1 file changed, 2 deletions(-)
tools/perf/util/pmu.h +0 -2
index cfcaba3e2e75..723a67cb41f4 100644
@@ -331,8 +331,6 @@ static inline enum pmu_kind perf_pmu__kind(const struct perf_pmu *pmu)
 		return PERF_PMU_KIND_PE;
 	if (type <= PERF_PMU_TYPE_DRM_END)
 		return PERF_PMU_KIND_DRM;
-	if (type <= PERF_PMU_TYPE_DRM_END)
-		return PERF_PMU_KIND_DRM;
 	if (type <= PERF_PMU_TYPE_HWMON_END)
 		return PERF_PMU_KIND_HWMON;
 	if (type == PERF_PMU_TYPE_TOOL)
Signature · 1 line
2.51.1
#5 lore raw

Re: [PATCH] perf pmu: fix duplicate conditional statement

Quoted 3 lines · Anubhav Shelat

Remove duplicate check for PERF_PMU_TYPE_DRM_END in perf_pmu__kind.

Fixes cedf1f4852ae.

The code change looks good to me. But the offensive commit hash ID seems incorrect.

The change comes from this commit.

f0feb21e0a10 ("perf pmu: Add PMU kind to simplify differentiating")

Better add a "Fixes" tag for this change.

Quoted 19 lines
---
 tools/perf/util/pmu.h | 2 --
 1 file changed, 2 deletions(-)
tools/perf/util/pmu.h +0 -2
index cfcaba3e2e75..723a67cb41f4 100644
@@ -331,8 +331,6 @@ static inline enum pmu_kind perf_pmu__kind(const struct perf_pmu *pmu)
 		return PERF_PMU_KIND_PE;
 	if (type <= PERF_PMU_TYPE_DRM_END)
 		return PERF_PMU_KIND_DRM;
-	if (type <= PERF_PMU_TYPE_DRM_END)
-		return PERF_PMU_KIND_DRM;
 	if (type <= PERF_PMU_TYPE_HWMON_END)
 		return PERF_PMU_KIND_HWMON;
 	if (type == PERF_PMU_TYPE_TOOL)
author reviewer Reviewed-by
#6 lore raw

Re: [PATCH] perf pmu: fix duplicate conditional statement

Quoted 7 lines · Dapeng
Quoted 3 lines · Anubhav Shelat

Remove duplicate check for PERF_PMU_TYPE_DRM_END in perf_pmu__kind.

Fixes cedf1f4852ae.

Quoted 8 lines

The code change looks good to me. But the offensive commit hash ID seems incorrect.

The change comes from this commit.

f0feb21e0a10 ("perf pmu: Add PMU kind to simplify differentiating")

Better add a "Fixes" tag for this change.

Thanks! I think the fixes tag should be:

Fixes f0feb21e0a10 ("perf pmu: Add PMU kind to simplify differentiating")

and there could also be:

as well as:

Thanks,
Ian

Quoted 19 lines
Quoted 19 lines
---
 tools/perf/util/pmu.h | 2 --
 1 file changed, 2 deletions(-)
tools/perf/util/pmu.h +0 -2
index cfcaba3e2e75..723a67cb41f4 100644
@@ -331,8 +331,6 @@ static inline enum pmu_kind perf_pmu__kind(const struct perf_pmu *pmu)
              return PERF_PMU_KIND_PE;
      if (type <= PERF_PMU_TYPE_DRM_END)
              return PERF_PMU_KIND_DRM;
-     if (type <= PERF_PMU_TYPE_DRM_END)
-             return PERF_PMU_KIND_DRM;
      if (type <= PERF_PMU_TYPE_HWMON_END)
              return PERF_PMU_KIND_HWMON;
      if (type == PERF_PMU_TYPE_TOOL)
maintainer Applied
#7 lore raw

Re: [PATCH] perf pmu: fix duplicate conditional statement

In reply to Anubhav Shelat
Quoted 5 lines · Anubhav Shelat

Remove duplicate check for PERF_PMU_TYPE_DRM_END in perf_pmu__kind.

Fixes cedf1f4852ae.

Applied to perf-tools-next, thanks!

Best regards,
Namhyung

author reviewer
#8 lore raw

PATCH 3/3 perf evsel: Skip store_evsel_ids for non-perf-event PMUs

In reply to Ian Rogers

The IDs are associated with perf events and not applicable to non-perf event PMUs. The failure to generate the ids was causing perf stat record to fail.

```
$ perf stat record -a sleep 1

 Performance counter stats for 'system wide':

            47,941      context-switches                 #      nan cs/sec  cs_per_second
              0.00 msec cpu-clock                        #      0.0 CPUs  CPUs_utilized
             3,261      cpu-migrations                   #      nan migrations/sec  migrations_per_second
               516      page-faults                      #      nan faults/sec  page_faults_per_second
         7,525,483      cpu_core/branch-misses/          #      2.3 %  branch_miss_rate
       322,069,004      cpu_core/branches/               #      nan M/sec  branch_frequency
     1,895,684,291      cpu_core/cpu-cycles/             #      nan GHz  cycles_frequency
     2,789,777,426      cpu_core/instructions/           #      1.5 instructions  insn_per_cycle
         7,074,765      cpu_atom/branch-misses/          #      3.2 %  branch_miss_rate         (49.89%)
       224,225,412      cpu_atom/branches/               #      nan M/sec  branch_frequency     (50.29%)
     2,061,679,981      cpu_atom/cpu-cycles/             #      nan GHz  cycles_frequency       (50.33%)
     2,011,242,533      cpu_atom/instructions/           #      1.0 instructions  insn_per_cycle  (50.33%)
             TopdownL1 (cpu_core)                        #      9.0 %  tma_bad_speculation
                                                         #     28.3 %  tma_frontend_bound
                                                         #     35.2 %  tma_backend_bound
                                                         #     27.5 %  tma_retiring
             TopdownL1 (cpu_atom)                        #     36.8 %  tma_backend_bound        (59.65%)
                                                         #     22.8 %  tma_frontend_bound       (59.60%)
                                                         #     11.6 %  tma_bad_speculation
                                                         #     28.8 %  tma_retiring             (59.59%)

       1.006777519 seconds time elapsed

$ perf stat report

 Performance counter stats for 'perf':

     1,013,376,154      duration_time
     <not counted>      duration_time
     <not counted>      duration_time
     <not counted>      duration_time
     <not counted>      duration_time
     <not counted>      duration_time
            47,941      context-switches
              0.00 msec cpu-clock
             3,261      cpu-migrations
               516      page-faults
         7,525,483      cpu_core/branch-misses/
       322,069,814      cpu_core/branches/
       322,069,004      cpu_core/branches/
     1,895,684,291      cpu_core/cpu-cycles/
     1,895,679,209      cpu_core/cpu-cycles/
     2,789,777,426      cpu_core/instructions/
     <not counted>      cpu_core/cpu-cycles/
     <not counted>      cpu_core/stalled-cycles-frontend/
     <not counted>      cpu_core/cpu-cycles/
     <not counted>      cpu_core/stalled-cycles-backend/
     <not counted>      cpu_core/stalled-cycles-backend/
     <not counted>      cpu_core/instructions/
     <not counted>      cpu_core/stalled-cycles-frontend/
         7,074,765      cpu_atom/branch-misses/                                                 (49.89%)
       221,679,088      cpu_atom/branches/                                                      (49.89%)
       224,225,412      cpu_atom/branches/                                                      (50.29%)
     2,061,679,981      cpu_atom/cpu-cycles/                                                    (50.33%)
     2,016,259,567      cpu_atom/cpu-cycles/                                                    (50.33%)
     2,011,242,533      cpu_atom/instructions/                                                  (50.33%)
     <not counted>      cpu_atom/cpu-cycles/
     <not counted>      cpu_atom/stalled-cycles-frontend/
     <not counted>      cpu_atom/cpu-cycles/
     <not counted>      cpu_atom/stalled-cycles-backend/
     <not counted>      cpu_atom/stalled-cycles-backend/
     <not counted>      cpu_atom/instructions/
     <not counted>      cpu_atom/stalled-cycles-frontend/
        17,145,113      cpu_core/INT_MISC.UOP_DROPPING/
    10,594,226,100      cpu_core/TOPDOWN.SLOTS/
     2,919,021,401      cpu_core/topdown-retiring/
       943,101,838      cpu_core/topdown-bad-spec/
     3,031,152,533      cpu_core/topdown-fe-bound/
     3,739,756,791      cpu_core/topdown-be-bound/
     1,909,501,648      cpu_atom/CPU_CLK_UNHALTED.CORE/                                         (60.04%)
     3,516,608,359      cpu_atom/TOPDOWN_BE_BOUND.ALL/                                          (59.65%)
     2,179,403,876      cpu_atom/TOPDOWN_FE_BOUND.ALL/                                          (59.60%)
     2,745,732,458      cpu_atom/TOPDOWN_RETIRING.ALL/                                          (59.59%)

       1.006777519 seconds time elapsed

Some events weren't counted. Try disabling the NMI watchdog:
        echo 0 > /proc/sys/kernel/nmi_watchdog
        perf stat ...
        echo 1 > /proc/sys/kernel/nmi_watchdog
```

---
I looked into adding metrics to perf stat report but there would be a merge conflict with:
https://lore.kernel.org/lkml/[email protected]/
so holding off for now.

---
 tools/perf/util/evsel.c | 3 +++
 1 file changed, 3 insertions(+)
tools/perf/util/evsel.c +3 -0
index 989c56d4a23f..aee42666e882 100644
@@ -3974,6 +3974,9 @@ static int store_evsel_ids(struct evsel *evsel, struct evlist *evlist)
 	if (evsel__is_retire_lat(evsel))
 		return 0;
 
+	if (perf_pmu__kind(evsel->pmu) != PERF_PMU_KIND_PE)
+		return 0;
+
 	for (cpu_map_idx = 0; cpu_map_idx < xyarray__max_x(evsel->core.fd); cpu_map_idx++) {
 		for (thread = 0; thread < xyarray__max_y(evsel->core.fd);
 		     thread++) {
Signature · 1 line
2.52.0.rc1.455.g30608eb744-goog
Tested-by on v1 3/3
#9 lore raw

Re: [PATCH v1 3/3] perf evsel: Skip store_evsel_ids for non-perf-event PMUs

Quoted 4 lines · Ian Rogers

The IDs are associated with perf events and not applicable to non-perf event PMUs. The failure to generate the ids was causing perf stat record to fail.

Hi Ian, looks good to me.

Here's the output on my alder lake:

% sudo ./perf stat record -a sleep 1

Performance counter stats for 'system wide':

             3,485      context-switches                 #    144.8 cs/sec  cs_per_second     
         24,075.55 msec cpu-clock                        #     24.0    CPUs  CPUs_utilized       
               206      cpu-migrations                   #      8.6 migrations/sec  migrations_per_second
               678      page-faults                      #     28.2 faults/sec  page_faults_per_second
         1,508,292      cpu_core/branch-misses/          #      2.1 %  branch_miss_rate       
        73,298,298      cpu_core/branches/               #      3.0 M/sec  branch_frequency   
       319,787,502      cpu_core/cpu-cycles/             #      0.0 GHz  cycles_frequency     
       366,691,216      cpu_core/instructions/           #      1.1 instructions  insn_per_cycle
           455,948      cpu_atom/branch-misses/          #      1.6 %  branch_miss_rate         (49.87%)
        28,573,057      cpu_atom/branches/               #      1.2 M/sec  branch_frequency     (49.98%)
       235,791,714      cpu_atom/cpu-cycles/             #      0.0 GHz  cycles_frequency       (50.07%)
       158,014,230      cpu_atom/instructions/           #      0.7 instructions  insn_per_cycle  (50.15%)
             TopdownL1 (cpu_core)                        #      8.1 %  tma_bad_speculation    
                                                         #     37.0 %  tma_frontend_bound     
                                                         #     36.6 %  tma_backend_bound      
                                                         #     18.2 %  tma_retiring           
             TopdownL1 (cpu_atom)                        #     51.0 %  tma_backend_bound        (59.99%)
                                                         #     21.4 %  tma_frontend_bound       (59.90%)
                                                         #     11.5 %  tma_bad_speculation    
                                                         #     16.2 %  tma_retiring             (59.82%)
       1.003087466 seconds time elapsed

% sudo ./perf stat report

Performance counter stats for '/home/tfalcon/perf-tools-next/tools/perf/perf':

     1,005,135,862      duration_time                                                         
     <not counted>      duration_time                                                         
     <not counted>      duration_time                                                         
     <not counted>      duration_time                                                         
     <not counted>      duration_time                                                         
     <not counted>      duration_time                                                         
             3,485      context-switches                                                      

24,075.55 msec cpu-clock
206 cpu-migrations
678 page-faults
1,508,292 cpu_core/branch-misses/
73,299,004 cpu_core/branches/
73,298,298 cpu_core/branches/
319,787,502 cpu_core/cpu-cycles/
319,799,050 cpu_core/cpu-cycles/
366,691,216 cpu_core/instructions/
<not counted> cpu_core/cpu-cycles/
<not counted> cpu_core/stalled-cycles-frontend/
<not counted> cpu_core/cpu-cycles/
<not counted> cpu_core/stalled-cycles-backend/
<not counted> cpu_core/stalled-cycles-backend/
<not counted> cpu_core/instructions/
<not counted> cpu_core/stalled-cycles-frontend/
455,948 cpu_atom/branch-misses/ (49.87%)
29,378,879 cpu_atom/branches/ (49.87%)
28,573,057 cpu_atom/branches/ (49.98%)
235,791,714 cpu_atom/cpu-cycles/ (50.07%)
231,878,974 cpu_atom/cpu-cycles/ (50.15%)
158,014,230 cpu_atom/instructions/ (50.15%)
<not counted> cpu_atom/cpu-cycles/
<not counted> cpu_atom/stalled-cycles-frontend/
<not counted> cpu_atom/cpu-cycles/
<not counted> cpu_atom/stalled-cycles-backend/
<not counted> cpu_atom/stalled-cycles-backend/
<not counted> cpu_atom/instructions/
<not counted> cpu_atom/stalled-cycles-frontend/
2,082,641 cpu_core/INT_MISC.UOP_DROPPING/
1,895,277,552 cpu_core/TOPDOWN.SLOTS/
345,024,206 cpu_core/topdown-retiring/
152,096,310 cpu_core/topdown-bad-spec/
704,850,131 cpu_core/topdown-fe-bound/
695,054,658 cpu_core/topdown-be-bound/
231,791,063 cpu_atom/CPU_CLK_UNHALTED.CORE/ (60.09%)
590,930,101 cpu_atom/TOPDOWN_BE_BOUND.ALL/ (59.99%)
247,501,143 cpu_atom/TOPDOWN_FE_BOUND.ALL/ (59.90%)
187,767,093 cpu_atom/TOPDOWN_RETIRING.ALL/ (59.82%)

1.003087466 seconds time elapsed

Some events weren't counted. Try disabling the NMI watchdog:
echo 0 > /proc/sys/kernel/nmi_watchdog perf stat ...
echo 1 > /proc/sys/kernel/nmi_watchdog

Thanks,
Tom

Quoted 113 lines
```
$ perf stat record -a sleep 1

 Performance counter stats for 'system wide':

            47,941      context-switches                 #      nan cs/sec  cs_per_second
              0.00 msec cpu-clock                        #      0.0 CPUs  CPUs_utilized
             3,261      cpu-migrations                   #      nan migrations/sec  migrations_per_second
               516      page-faults                      #      nan faults/sec  page_faults_per_second
         7,525,483      cpu_core/branch-misses/          #      2.3 %  branch_miss_rate
       322,069,004      cpu_core/branches/               #      nan M/sec  branch_frequency
     1,895,684,291      cpu_core/cpu-cycles/             #      nan GHz  cycles_frequency
     2,789,777,426      cpu_core/instructions/           #      1.5 instructions  insn_per_cycle
         7,074,765      cpu_atom/branch-misses/          #      3.2 %  branch_miss_rate         (49.89%)
       224,225,412      cpu_atom/branches/               #      nan M/sec  branch_frequency     (50.29%)
     2,061,679,981      cpu_atom/cpu-cycles/             #      nan GHz  cycles_frequency       (50.33%)
     2,011,242,533      cpu_atom/instructions/           #      1.0 instructions  insn_per_cycle  (50.33%)
             TopdownL1 (cpu_core)                        #      9.0 %  tma_bad_speculation
                                                         #     28.3 %  tma_frontend_bound
                                                         #     35.2 %  tma_backend_bound
                                                         #     27.5 %  tma_retiring
             TopdownL1 (cpu_atom)                        #     36.8 %  tma_backend_bound        (59.65%)
                                                         #     22.8 %  tma_frontend_bound       (59.60%)
                                                         #     11.6 %  tma_bad_speculation
                                                         #     28.8 %  tma_retiring             (59.59%)

       1.006777519 seconds time elapsed

$ perf stat report

 Performance counter stats for 'perf':

     1,013,376,154      duration_time
     <not counted>      duration_time
     <not counted>      duration_time
     <not counted>      duration_time
     <not counted>      duration_time
     <not counted>      duration_time
            47,941      context-switches
              0.00 msec cpu-clock
             3,261      cpu-migrations
               516      page-faults
         7,525,483      cpu_core/branch-misses/
       322,069,814      cpu_core/branches/
       322,069,004      cpu_core/branches/
     1,895,684,291      cpu_core/cpu-cycles/
     1,895,679,209      cpu_core/cpu-cycles/
     2,789,777,426      cpu_core/instructions/
     <not counted>      cpu_core/cpu-cycles/
     <not counted>      cpu_core/stalled-cycles-frontend/
     <not counted>      cpu_core/cpu-cycles/
     <not counted>      cpu_core/stalled-cycles-backend/
     <not counted>      cpu_core/stalled-cycles-backend/
     <not counted>      cpu_core/instructions/
     <not counted>      cpu_core/stalled-cycles-frontend/
         7,074,765      cpu_atom/branch-misses/                                                 (49.89%)
       221,679,088      cpu_atom/branches/                                                      (49.89%)
       224,225,412      cpu_atom/branches/                                                      (50.29%)
     2,061,679,981      cpu_atom/cpu-cycles/                                                    (50.33%)
     2,016,259,567      cpu_atom/cpu-cycles/                                                    (50.33%)
     2,011,242,533      cpu_atom/instructions/                                                  (50.33%)
     <not counted>      cpu_atom/cpu-cycles/
     <not counted>      cpu_atom/stalled-cycles-frontend/
     <not counted>      cpu_atom/cpu-cycles/
     <not counted>      cpu_atom/stalled-cycles-backend/
     <not counted>      cpu_atom/stalled-cycles-backend/
     <not counted>      cpu_atom/instructions/
     <not counted>      cpu_atom/stalled-cycles-frontend/
        17,145,113      cpu_core/INT_MISC.UOP_DROPPING/
    10,594,226,100      cpu_core/TOPDOWN.SLOTS/
     2,919,021,401      cpu_core/topdown-retiring/
       943,101,838      cpu_core/topdown-bad-spec/
     3,031,152,533      cpu_core/topdown-fe-bound/
     3,739,756,791      cpu_core/topdown-be-bound/
     1,909,501,648      cpu_atom/CPU_CLK_UNHALTED.CORE/                                         (60.04%)
     3,516,608,359      cpu_atom/TOPDOWN_BE_BOUND.ALL/                                          (59.65%)
     2,179,403,876      cpu_atom/TOPDOWN_FE_BOUND.ALL/                                          (59.60%)
     2,745,732,458      cpu_atom/TOPDOWN_RETIRING.ALL/                                          (59.59%)

       1.006777519 seconds time elapsed

Some events weren't counted. Try disabling the NMI watchdog:
        echo 0 > /proc/sys/kernel/nmi_watchdog
        perf stat ...
        echo 1 > /proc/sys/kernel/nmi_watchdog
```

---
I looked into adding metrics to perf stat report but there would be a merge conflict with:
https://lore.kernel.org/lkml/[email protected]/
so holding off for now.
---
 tools/perf/util/evsel.c | 3 +++

 1 file changed, 3 insertions(+)
tools/perf/util/evsel.c +0 -0
index 989c56d4a23f..aee42666e882 100644
@@ -3974,6 +3974,9 @@ static int store_evsel_ids(struct evsel *evsel, struct evlist *evlist)
  	if (evsel__is_retire_lat(evsel))
  		return 0;

tools/perf/util/evsel.c +3 -0
+	if (perf_pmu__kind(evsel->pmu) != PERF_PMU_KIND_PE)
+		return 0;
+

  for (cpu_map_idx = 0; cpu_map_idx < xyarray__max_x(evsel->core.fd); cpu_map_idx++) { for (thread = 0; thread < xyarray__max_y(evsel->core.fd); thread++) {

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

Writing currently fails on non-x86 and hybrid CPUs. Switch to the more regular find_core_pmu that is normally used in this case. Tested on hybrid alderlake system.

Applied to perf-tools-next, thanks!

Best regards,
Namhyung