Re: [PATCH 2/2] perf evsel: Improve error message for uncore events

[Ian Rogers]

On Sun, Nov 28, 2021 at 8:58 AM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
>
> On Mon, Nov 22, 2021 at 06:03:41PM -0800, Ian Rogers wrote:
> > When a group has multiple events and the leader fails it can yield
> > errors like:
> >
> > $ perf stat -e '{uncore_imc/cas_count_read/},instructions' /bin/true
> > Error:
> > The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (uncore_imc/cas_count_read/).
> > /bin/dmesg | grep -i perf may provide additional information.
> >
> > However, when not the group leader <not supported> is given:
> >
> > $ perf stat -e '{instructions,uncore_imc/cas_count_read/}' /bin/true
> > ...
> >          1,619,057      instructions
> >    <not supported> MiB  uncore_imc/cas_count_read/
> >
> > This is necessary because get_group_fd will fail if the leader fails and
> > is the direct result of the check on line 750 of builtin-stat.c in
> > stat_handle_error that returns COUNTER_SKIP for the latter case.
> >
> > This patch improves the error message to:
> >
> > $ perf stat -e '{uncore_imc/cas_count_read/},instructions' /bin/true
> > Error:
> > Invalid event (uncore_imc/cas_count_read/) in per-thread mode, enable system wide with '-a'.
> >
> > Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
> > ---
> >  tools/perf/util/evsel.c | 5 +++++
> >  1 file changed, 5 insertions(+)
> >
> > diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> > index a59fb2ecb84e..48696ff4bddb 100644
> > --- a/tools/perf/util/evsel.c
> > +++ b/tools/perf/util/evsel.c
> > @@ -2950,6 +2950,11 @@ int evsel__open_strerror(struct evsel *evsel, struct target *target,
> >                       return scnprintf(msg, size, "wrong clockid (%d).", clockid);
> >               if (perf_missing_features.aux_output)
> >                       return scnprintf(msg, size, "The 'aux_output' feature is not supported, update the kernel.");
> > +             if ((evsel__leader(evsel) == evsel) &&
> > +                 (evsel->core.leader->nr_members > 1))
> > +                     return scnprintf(msg, size,
> > +     "Invalid event (%s) in per-thread mode, enable system wide with '-a'.",
> > +                                     evsel__name(evsel));
>
> should we rather check 'target' pointer for the per-thread mode?
> I'm not sure that per-thread mode will always be the case for the failure

Unfortunately the target isn't populated at that point:

gdb --args perf stat -e '{uncore_imc/cas_count_write/},instructions' /bin/true
(gdb) p *target
$2 = {pid = 0x0, tid = 0x0, cpu_list = 0x0, uid_str = 0x0, bpf_str =
0x0, uid = 4294967295, system_wide = false, uses_mmap = false,
  default_per_cpu = false, per_thread = false, use_bpf = false, hybrid
= false, attr_map = 0x0}

#0  evsel__open_strerror (evsel=0x616000015680, target=0x5555586aa140
<target>, err=22, msg=0x7fffffff78d0 "]k\264WUU", size=8192)
    at util/evsel.c:2857
#1  0x00005555561744e8 in stat_handle_error (counter=0x616000015680)
at builtin-stat.c:771
#2  0x0000555556174f05 in __run_perf_stat (argc=1,
argv=0x7fffffffe450, run_idx=0) at builtin-stat.c:852
#3  0x00005555561763e1 in run_perf_stat (argc=1, argv=0x7fffffffe450,
run_idx=0) at builtin-stat.c:1048
#4  0x000055555617df82 in cmd_stat (argc=1, argv=0x7fffffffe450) at
builtin-stat.c:2550
#5  0x00005555562f36b8 in run_builtin (p=0x5555586bad00
<commands+288>, argc=4, argv=0x7fffffffe450) at perf.c:313
#6  0x00005555562f3c11 in handle_internal_command (argc=4,
argv=0x7fffffffe450) at perf.c:365
#7  0x00005555562f3fce in run_argv (argcp=0x7fffffffe230,
argv=0x7fffffffe240) at perf.c:409
#8  0x00005555562f47bd in main (argc=4, argv=0x7fffffffe450) at perf.c:539

Thanks,
Ian

> jirka
>
> >               break;
> >       case ENODATA:
> >               return scnprintf(msg, size, "Cannot collect data source with the load latency event alone. "
> > --
> > 2.34.0.rc2.393.gf8c9666880-goog
> >
>