Re: [PATCH] perf pmu: Consider raw events as legacy events

[Namhyung Kim <namhyung@kernel.org>]

Hi Ian,

On Wed, May 07, 2025 at 08:42:27AM -0700, Ian Rogers wrote:
> On Tue, May 6, 2025 at 3:23 PM Namhyung Kim <namhyung@kernel.org> wrote:
> >
> > When it finds a matching pmu for a legacy event, it should look for
> > core pmus.  The raw events also refers to core events so it should be
> > handled together.
> >
> > Signed-off-by: Namhyung Kim <namhyung@kernel.org>
> > ---
> >  tools/perf/util/pmus.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/tools/perf/util/pmus.c b/tools/perf/util/pmus.c
> > index b99292de76693dbb..0134321fc520b1fc 100644
> > --- a/tools/perf/util/pmus.c
> > +++ b/tools/perf/util/pmus.c
> > @@ -726,7 +726,8 @@ struct perf_pmu *evsel__find_pmu(const struct evsel *evsel)
> >         pmu = perf_pmus__find_by_type(evsel->core.attr.type);
> >         legacy_core_type =
> >                 evsel->core.attr.type == PERF_TYPE_HARDWARE ||
> > -               evsel->core.attr.type == PERF_TYPE_HW_CACHE;
> > +               evsel->core.attr.type == PERF_TYPE_HW_CACHE ||
> > +               evsel->core.attr.type == PERF_TYPE_RAW;
> 
> I don't think this is right. The PERF_TYPE_RAW isn't a legacy type as
> we map sysfs/json names to the format encoding. More than this, the
> extended type information will not be present for a raw type and so
> this change could cause an invalid PMU to be looked up on
> hybrid/BIG.little systems. On x86 the cpu or cpu_core PMU will have
> type PERF_TYPE_RAW (4). As is common ARM decided to do things
> differently, meaning there is no type 4 to a PMU mapping. I think we
> need a change like:
> ```
> diff --git a/tools/perf/util/pmus.c b/tools/perf/util/pmus.c
> index b99292de7669..6632209c6664 100644
> --- a/tools/perf/util/pmus.c
> +++ b/tools/perf/util/pmus.c
> @@ -727,14 +727,21 @@ struct perf_pmu *evsel__find_pmu(const struct
> evsel *evsel)
>        legacy_core_type =
>                evsel->core.attr.type == PERF_TYPE_HARDWARE ||
>                evsel->core.attr.type == PERF_TYPE_HW_CACHE;
> -       if (!pmu && legacy_core_type) {
> -               if (perf_pmus__supports_extended_type()) {
> -                       u32 type = evsel->core.attr.config >>
> PERF_PMU_TYPE_SHIFT;
> +       if (!pmu && legacy_core_type && perf_pmus__supports_extended_type()) {
> +               u32 type = evsel->core.attr.config >> PERF_PMU_TYPE_SHIFT;
> 
> -                       pmu = perf_pmus__find_by_type(type);
> -               } else {
> -                       pmu = perf_pmus__find_core_pmu();
> -               }
> +               pmu = perf_pmus__find_by_type(type);
> +       }
> +       if (!pmu && (legacy_core_type || evsel->core.attr.type ==
> PERF_TYPE_RAW)) {
> +               /*
> +                * For legacy events, if there was no extended type info then
> +                * assume the PMU is the first core PMU.
> +                *
> +                * On architectures like ARM there is no sysfs PMU with type
> +                * PERF_TYPE_RAW, assume the RAW events are going to be handled
> +                * by the first core PMU.
> +                */
> +               pmu = perf_pmus__find_core_pmu();
>        }
>        ((struct evsel *)evsel)->pmu = pmu;
>        return pmu;
> ```
> which handles both the case that the extended type was missing on
> hybrid/BIG.little and the ARM raw type problem.

Looks good to me.  Thanks for the suggestion!

Thanks,
Namhyung