perf stat: Separate bperf from bpf_profiler
authorDmitrii Dolgov <9erthalion6@gmail.com>
Wed, 12 Apr 2023 18:23:16 +0000 (20:23 +0200)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 5 May 2023 20:30:05 +0000 (17:30 -0300)
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.

Fixes: 112cb56164bc2108 ("perf stat: Introduce config stat.bpf-counter-events")
Reviewed-by: Song Liu <song@kernel.org>
Signed-off-by: Dmitrii Dolgov <9erthalion6@gmail.com>
Tested-by: Song Liu <song@kernel.org>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20230412182316.11628-1-9erthalion6@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/builtin-stat.c
tools/perf/util/evsel.h

index be9677a..cc9fa48 100644 (file)
@@ -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,
index 35805dc..d575390 100644 (file)
@@ -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];