Re: [PATCH v1] perf stat: Additional verbose details for <not supported> events

[Ian Rogers <irogers@google.com>]

On Wed, Oct 8, 2025 at 1:43 AM Namhyung Kim <namhyung@kernel.org> wrote:
>
> On Sun, Oct 05, 2025 at 11:14:21AM -0700, Ian Rogers wrote:
> > If an event shows as "<not supported>" in perf stat output, in verbose
> > mode add the strerror output to help diagnose the issue.
> >
> > Consider:
> > ```
> > $ perf stat -e cycles,data_read,instructions true
> >
> >  Performance counter stats for 'true':
> >
> >            357,457      cycles:u
> >    <not supported> MiB  data_read:u
> >            156,182      instructions:u                   #    0.44  insn per cycle
> >
> >        0.001250315 seconds time elapsed
> >
> >        0.001283000 seconds user
> >        0.000000000 seconds sys
> > ```
> >
> > To understand why the data_read uncore event failed, with this change:
> > ```
> > $ perf stat -v -e cycles,data_read,instructions true
> > Using CPUID GenuineIntel-6-8D-1
> > cycles -> cpu/cycles/
> > data_read -> uncore_imc_free_running_0/data_read/
> > data_read -> uncore_imc_free_running_1/data_read/
> > instructions -> cpu/instructions/
> > Control descriptor is not initialized
> > Warning:
> > kernel.perf_event_paranoid=2, trying to fall back to excluding kernel and hypervisor  samples
> > Warning:
> > kernel.perf_event_paranoid=2, trying to fall back to excluding kernel and hypervisor  samples
> > Warning:
> > kernel.perf_event_paranoid=2, trying to fall back to excluding kernel and hypervisor  samples
> > Warning:
> > data_read:u event is not supported by the kernel.
> > Invalid event (data_read:u) in per-thread mode, enable system wide with '-a'.
> > Warning:
> > kernel.perf_event_paranoid=2, trying to fall back to excluding kernel and hypervisor  samples
> > Warning:
> > data_read:u event is not supported by the kernel.
> > Invalid event (data_read:u) in per-thread mode, enable system wide with '-a'.
> > cycles:u: 351621 362833 362833
> > failed to read counter data_read:u
> > failed to read counter data_read:u
> > instructions:u: 156184 362833 362833
> >
> >  Performance counter stats for 'true':
> >
> >            351,621      cycles:u
> >    <not supported> MiB  data_read:u
> >            156,184      instructions:u                   #    0.44  insn per cycle
> >
> >        0.001584472 seconds time elapsed
> >
> >        0.001811000 seconds user
> >        0.000000000 seconds sys
> > ```
> > where without this change only "data_read:u event is not supported by
> > the kernel." is shown.
>
> I think what you say is:
>
> Before:
>   data_read:u event is not supported by the kernel.
>
> After:
>   data_read:u event is not supported by the kernel.
>   Invalid event (data_read:u) in per-thread mode, enable system wide with '-a'.

I kept things verbose as unfortunately the
"kernel.perf_event_paranoid=2" is important as is the use of
per-thread mode. Different paranoia levels lead to different errors
and unfortunately a lot of the time the error gets reported as "
data_read:u event is not supported by the kernel." and I'm not sure
all users will get that the key part there is the :u modifier.
However, fixing  evsel__open_strerror wasn't in scope for this change.

Thanks,
Ian


> Off-topic, it'd be great if we reduce the number of the same warning
> messages.  I think the data_read is from two uncore PMUs so the message
> was repeated.  If we can connect the related evsels and show the
> messages once, then the output is more readable.  Maybe we also want to
> show the fallback message just once (globally).
>
> Thanks,
> Namhyung
>
> >
> > Signed-off-by: Ian Rogers <irogers@google.com>
> > ---
> >  tools/perf/builtin-stat.c | 12 +++++++-----
> >  1 file changed, 7 insertions(+), 5 deletions(-)
> >
> > diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
> > index 7006f848f87a..84e06ec09cc2 100644
> > --- a/tools/perf/builtin-stat.c
> > +++ b/tools/perf/builtin-stat.c
> > @@ -624,8 +624,9 @@ static enum counter_recovery stat_handle_error(struct evsel *counter, int err)
> >        */
> >       if (err == EINVAL || err == ENOSYS || err == ENOENT || err == ENXIO) {
> >               if (verbose > 0) {
> > -                     ui__warning("%s event is not supported by the kernel.\n",
> > -                                 evsel__name(counter));
> > +                     evsel__open_strerror(counter, &target, err, msg, sizeof(msg));
> > +                     ui__warning("%s event is not supported by the kernel.\n%s\n",
> > +                                 evsel__name(counter), msg);
> >               }
> >               return COUNTER_SKIP;
> >       }
> > @@ -649,10 +650,11 @@ static enum counter_recovery stat_handle_error(struct evsel *counter, int err)
> >               }
> >       }
> >       if (verbose > 0) {
> > +             evsel__open_strerror(counter, &target, err, msg, sizeof(msg));
> >               ui__warning(err == EOPNOTSUPP
> > -                     ? "%s event is not supported by the kernel.\n"
> > -                     : "skipping event %s that kernel failed to open.\n",
> > -                     evsel__name(counter));
> > +                     ? "%s event is not supported by the kernel.\n%s\n"
> > +                     : "skipping event %s that kernel failed to open.\n%s\n",
> > +                     evsel__name(counter), msg);
> >       }
> >       return COUNTER_SKIP;
> >  }
> > --
> > 2.51.0.618.g983fd99d29-goog
> >