Skip to contentSkip to conversation

perf stat: Separate bperf from bpf_profiler

RFC PATCH In review Dmitrii Dolgov 3 participants 5 messages Apr 12, 2023 – May 5, 2023 perf events

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 perf stat: Separate bperf from bpf_profiler

It seems that perf stat -b <prog id> doesn't produce any results:

    $ perf stat -e cycles -b 4 -I 10000 -vvv
    Control descriptor is not initialized
    cycles: 0 0 0
                time        counts unit      events
	10.007641640    <not supported>      cycles

Looks like this happens because fentry/fexit progs are getting loaded, but the corresponding perf event is not enabled and not added into the events bpf map. I think there is some mixing up between two type of bpf support, one for bperf and one for bpf_profiler. Both are identified via evsel__is_bpf, based on which perf events are enabled, but for the latter (bpf_profiler) a perf event is required. Using evsel__is_bperf to check only bperf produces expected results:

    $ perf stat -e cycles -b 4 -I 10000 -vvv
    Control descriptor is not initialized
    ------------------------------------------------------------
    perf_event_attr:
      size                             136
      sample_type                      IDENTIFIER
      read_format                      TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING
      disabled                         1
      exclude_guest                    1
    ------------------------------------------------------------
    sys_perf_event_open: pid -1  cpu 0  group_fd -1  flags 0x8 = 3
    ------------------------------------------------------------
    [...perf_event_attr for other CPUs...]
    ------------------------------------------------------------
    cycles: 309426 169009 169009
		time             counts unit events
	10.010091271             309426      cycles

The final numbers correspond (at least in the level of magnitude) to the same metric obtained via bpftool.

---
 tools/perf/builtin-stat.c | 4 ++--
 tools/perf/util/evsel.h   | 5 +++++
 2 files changed, 7 insertions(+), 2 deletions(-)
tools/perf/builtin-stat.c +2 -2
index d3cbee7460f..23b8d684ca6 100644
@@ -712,7 +712,7 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
 		counter->reset_group = false;
 		if (bpf_counter__load(counter, &target))
 			return -1;
-		if (!evsel__is_bpf(counter))
+		if (!(evsel__is_bperf(counter)))
 			all_counters_use_bpf = false;
 	}
 
@@ -728,7 +728,7 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
 
 		if (counter->reset_group || counter->errored)
 			continue;
-		if (evsel__is_bpf(counter))
+		if (evsel__is_bperf(counter))
 			continue;
 try_again:
 		if (create_perf_stat_counter(counter, &stat_config, &target,
tools/perf/util/evsel.h +5 -0
index 68072ec655c..f0aa17dcf3d 100644
@@ -269,6 +269,11 @@ static inline bool evsel__is_bpf(struct evsel *evsel)
 	return evsel->bpf_counter_ops != NULL;
 }
 
+static inline bool evsel__is_bperf(struct evsel *evsel)
+{
+	return evsel->bpf_counter_ops != NULL && list_empty(&evsel->bpf_counter_list);
+}
+
 #define EVSEL__MAX_ALIASES 8
 
 extern const char *const evsel__hw_cache[PERF_COUNT_HW_CACHE_MAX][EVSEL__MAX_ALIASES];
Signature · 1 line
2.32.0
#1 lore raw
Quoted 9 lines · Dmitrii Dolgov

It seems that perf stat -b <prog id> doesn't produce any results:

    $ perf stat -e cycles -b 4 -I 10000 -vvv
    Control descriptor is not initialized
    cycles: 0 0 0
                time        counts unit      events
	10.007641640    <not supported>      cycles
Quoted 6 lines

Looks like this happens because fentry/fexit progs are getting loaded, but the corresponding perf event is not enabled and not added into the events bpf map. I think there is some mixing up between two type of bpf support, one for bperf and one for bpf_profiler. Both are identified via evsel__is_bpf, based on which perf events are enabled, but for the latter (bpf_profiler) a perf event is required. Using evsel__is_bperf to check only bperf produces expected results:

Any thoughts on this? I would appreciate clarifications if I'm missing something.

#2 lore raw

Em Fri, Apr 21, 2023 at 10:56:10PM +0200, Dmitry Dolgov escreveu:

Quoted 16 lines
Quoted 9 lines · Dmitrii Dolgov

It seems that perf stat -b <prog id> doesn't produce any results:

    $ perf stat -e cycles -b 4 -I 10000 -vvv
    Control descriptor is not initialized
    cycles: 0 0 0
                time        counts unit      events
	10.007641640    <not supported>      cycles
Quoted 6 lines

Looks like this happens because fentry/fexit progs are getting loaded, but the corresponding perf event is not enabled and not added into the events bpf map. I think there is some mixing up between two type of bpf support, one for bperf and one for bpf_profiler. Both are identified via evsel__is_bpf, based on which perf events are enabled, but for the latter (bpf_profiler) a perf event is required. Using evsel__is_bperf to check only bperf produces expected results:

Quoted 2 lines

Any thoughts on this? I would appreciate clarifications if I'm missing something.

Namhyung, Song, can you please take a look at this?

- Arnaldo

#3 lore raw
Quoted 18 lines · Arnaldo Carvalho de Melo

Em Fri, Apr 21, 2023 at 10:56:10PM +0200, Dmitry Dolgov escreveu:

Quoted 16 lines
Quoted 15 lines · Dmitrii Dolgov

It seems that perf stat -b <prog id> doesn't produce any results:

    $ perf stat -e cycles -b 4 -I 10000 -vvv
    Control descriptor is not initialized
    cycles: 0 0 0
                time        counts unit      events
    10.007641640    <not supported>      cycles

Looks like this happens because fentry/fexit progs are getting loaded, but the corresponding perf event is not enabled and not added into the events bpf map. I think there is some mixing up between two type of bpf support, one for bperf and one for bpf_profiler. Both are identified via evsel__is_bpf, based on which perf events are enabled, but for the latter (bpf_profiler) a perf event is required. Using evsel__is_bperf to check only bperf produces expected results:

Quoted 4 lines
Quoted 2 lines

Any thoughts on this? I would appreciate clarifications if I'm missing something.

Namhyung, Song, can you please take a look at this?

Sorry for the late response. The fix looks good to me and worked well in my test.

Reviewed-and-tested-by Song Liu <[email protected]>

I guess we also need:

Fixes 112cb56164bc2 ("perf stat: Introduce config stat.bpf-counter-events")

Thanks for the fix!
Song

#4 lore raw

Em Thu, May 04, 2023 at 07:01:04PM -0700, Song Liu escreveu:

Quoted 24 lines · Arnaldo Carvalho de Melo
Quoted 18 lines

Em Fri, Apr 21, 2023 at 10:56:10PM +0200, Dmitry Dolgov escreveu:

Quoted 15 lines
Quoted 15 lines · Dmitrii Dolgov

It seems that perf stat -b <prog id> doesn't produce any results:

    $ perf stat -e cycles -b 4 -I 10000 -vvv
    Control descriptor is not initialized
    cycles: 0 0 0
                time        counts unit      events
    10.007641640    <not supported>      cycles

Looks like this happens because fentry/fexit progs are getting loaded, but the corresponding perf event is not enabled and not added into the events bpf map. I think there is some mixing up between two type of bpf support, one for bperf and one for bpf_profiler. Both are identified via evsel__is_bpf, based on which perf events are enabled, but for the latter (bpf_profiler) a perf event is required. Using evsel__is_bperf to check only bperf produces expected results:

Quoted 4 lines
Quoted 2 lines

Any thoughts on this? I would appreciate clarifications if I'm missing something.

Namhyung, Song, can you please take a look at this?

Quoted 8 lines

Sorry for the late response. The fix looks good to me and worked well in my test.

Reviewed-and-tested-by Song Liu <[email protected]>

I guess we also need:

Fixes 112cb56164bc2 ("perf stat: Introduce config stat.bpf-counter-events")

Thanks a lot, applied, and this is relevant in the current situation, where we're trying to have Linux v6.4 perf tools building BPF skels by default.

- Arnaldo