Re: [PATCH v2 6/7] perf pmu-events: Remember the perf_events_map for a PMU

[Yang Jihong]

Hello,

On 2023/10/13 1:56, Ian Rogers wrote:
> strcmp_cpuid_str performs regular expression comparisons and so per
> CPUID linear searches over the perf_events_map are expensive. Add a
> helper function called map_for_pmu that does the search but also
> caches the map specific to a PMU. As the PMU may differ, also cache
> the CPUID string so that PMUs with the same CPUID string don't require
> the linear search and regular expression comparisons. This speeds
> loading PMUs as the search is done once per PMU to find the
> appropriate tables.
>
> Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
> ---
>   tools/perf/pmu-events/jevents.py | 109 ++++++++++++++++++++-----------
>   1 file changed, 70 insertions(+), 39 deletions(-)
>
> diff --git a/tools/perf/pmu-events/jevents.py b/tools/perf/pmu-events/jevents.py
> index 96dc74c90b20..3c091ab75305 100755
> --- a/tools/perf/pmu-events/jevents.py
> +++ b/tools/perf/pmu-events/jevents.py
> @@ -976,68 +976,99 @@ int pmu_metrics_table__for_each_metric(const struct pmu_metrics_table *table,
>           return 0;
>   }
>   
> -const struct pmu_events_table *perf_pmu__find_events_table(struct perf_pmu *pmu)
> +static const struct pmu_events_map *map_for_pmu(struct perf_pmu *pmu)
>   {
> -        const struct pmu_events_table *table = NULL;
> -        char *cpuid = perf_pmu__getcpuid(pmu);
> +        static struct {
> +                const struct pmu_events_map *map;
> +                struct perf_pmu *pmu;
> +        } last_result;
> +        static struct {
> +                const struct pmu_events_map *map;
> +                char *cpuid;
> +        } last_map_search;
> +        static bool has_last_result, has_last_map_search;
> +        const struct pmu_events_map *map = NULL;
> +        char *cpuid = NULL;
>           size_t i;
>   
> -        /* on some platforms which uses cpus map, cpuid can be NULL for
> +        if (has_last_result && last_result.pmu == pmu)
> +                return last_result.map;
> +
> +        cpuid = perf_pmu__getcpuid(pmu);
For the software pmu, we do not need to look for the events table.
It seems that the software pmu can be filtered out in perf_pmu__lookup() 
to reduce unnecessary perf_pmu__find_events_table() calls.

I tried to submit a patch, please see if it helps:
https://lore.kernel.org/all/20231016093309.726436-1-yangjihong1@xxxxxxxxxx/

Thanks,
Yang