Re: [PATCH v2] perf evsel amd: Fix IBS error message

[Namhyung Kim <namhyung@kernel.org>]

Hi Ravi,

On Mon, Jun 26, 2023 at 3:40 AM Ravi Bangoria <ravi.bangoria@amd.com> wrote:
>
> AMD IBS can do per-process profiling[1] and is no longer restricted to
> per-cpu or systemwide only. Remove stale error message. Also, checking
> just exclude_kernel is not sufficient since IBS does not support any
> privilege filters. So include all exclude_* checks. And finally, move
> these checks under tools/perf/arch/x86/ from generic code.
>
> Before:
>   $ sudo ./perf record -e ibs_op//k -C 0
>   Error:
>   AMD IBS may only be available in system-wide/per-cpu mode.  Try
>   using -a, or -C and workload affinity
>
> After:
>   $ sudo ./perf record -e ibs_op//k -C 0
>   Error:
>   AMD IBS doesn't support privilege filtering. Try again with
>   exclude_{kernel|user|hv|idle|host|guest}=0.

Can we have more user-friendly messages like below?

 "Try again without the privilege modifiers like 'k' at the end."

>
> Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com>
> ---
> v1: https://lore.kernel.org/r/20230621062359.201-1-ravi.bangoria@amd.com
> v1->v2:
>   - Check all exclude_* flags not just exclude_kernel
>   - Move AMD specific checks under tools/perf/arch/x86/
>
>  tools/perf/arch/x86/util/evsel.c | 25 +++++++++++++++++++++++++
>  tools/perf/util/evsel.c          | 30 +++++++++---------------------
>  tools/perf/util/evsel.h          |  1 +
>  3 files changed, 35 insertions(+), 21 deletions(-)
>
> diff --git a/tools/perf/arch/x86/util/evsel.c b/tools/perf/arch/x86/util/evsel.c
> index 512c2d885d24..9a7141c5a4ea 100644
> --- a/tools/perf/arch/x86/util/evsel.c
> +++ b/tools/perf/arch/x86/util/evsel.c
> @@ -102,3 +102,28 @@ void arch__post_evsel_config(struct evsel *evsel, struct perf_event_attr *attr)
>                 }
>         }
>  }
> +
> +int arch_evsel__open_strerror(struct evsel *evsel, char *msg, size_t size)
> +{
> +       if (!x86__is_amd_cpu())
> +               return 0;
> +
> +       if (!evsel->core.attr.precise_ip &&
> +           !(evsel->pmu_name && !strncmp(evsel->pmu_name, "ibs", 3)))
> +               return 0;
> +
> +       /* More verbose IBS errors. */
> +       if (evsel->core.attr.exclude_kernel || evsel->core.attr.exclude_user ||
> +           evsel->core.attr.exclude_hv || evsel->core.attr.exclude_idle ||
> +           evsel->core.attr.exclude_host || evsel->core.attr.exclude_guest) {
> +               return scnprintf(msg, size, "AMD IBS doesn't support privilege filtering. Try "
> +                                "again with exclude_{kernel|user|hv|idle|host|guest}=0.");
> +       }
> +
> +       if (!evsel->core.attr.sample_period) {
> +               return scnprintf(msg, size, "AMD IBS doesn't support counting mode. Try "
> +                                "again with sample_{period|freq} with non-zero value.");

Is this for perf stat?   We don't allow zero period for perf record.

  $ perf record -F 0 true
  frequency and count are zero, aborting

Then maybe it can say: "IBS doesn't support perf stat, Use perf record."

Thanks,
Namhyung


> +       }
> +
> +       return 0;
> +}
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index f607b5bddc76..762e2b2634a5 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -2924,25 +2924,19 @@ static bool find_process(const char *name)
>         return ret ? false : true;
>  }
>
> -static bool is_amd(const char *arch, const char *cpuid)
> +int __weak arch_evsel__open_strerror(struct evsel *evsel __maybe_unused,
> +                                    char *msg __maybe_unused,
> +                                    size_t size __maybe_unused)
>  {
> -       return arch && !strcmp("x86", arch) && cpuid && strstarts(cpuid, "AuthenticAMD");
> -}
> -
> -static bool is_amd_ibs(struct evsel *evsel)
> -{
> -       return evsel->core.attr.precise_ip
> -           || (evsel->pmu_name && !strncmp(evsel->pmu_name, "ibs", 3));
> +       return 0;
>  }
>
>  int evsel__open_strerror(struct evsel *evsel, struct target *target,
>                          int err, char *msg, size_t size)
>  {
> -       struct perf_env *env = evsel__env(evsel);
> -       const char *arch = perf_env__arch(env);
> -       const char *cpuid = perf_env__cpuid(env);
>         char sbuf[STRERR_BUFSIZE];
>         int printed = 0, enforced = 0;
> +       int ret;
>
>         switch (err) {
>         case EPERM:
> @@ -3044,16 +3038,6 @@ int evsel__open_strerror(struct evsel *evsel, struct target *target,
>                         return scnprintf(msg, size,
>         "Invalid event (%s) in per-thread mode, enable system wide with '-a'.",
>                                         evsel__name(evsel));
> -               if (is_amd(arch, cpuid)) {
> -                       if (is_amd_ibs(evsel)) {
> -                               if (evsel->core.attr.exclude_kernel)
> -                                       return scnprintf(msg, size,
> -       "AMD IBS can't exclude kernel events.  Try running at a higher privilege level.");
> -                               if (!evsel->core.system_wide)
> -                                       return scnprintf(msg, size,
> -       "AMD IBS may only be available in system-wide/per-cpu mode.  Try using -a, or -C and workload affinity");
> -                       }
> -               }
>
>                 break;
>         case ENODATA:
> @@ -3063,6 +3047,10 @@ int evsel__open_strerror(struct evsel *evsel, struct target *target,
>                 break;
>         }
>
> +       ret = arch_evsel__open_strerror(evsel, msg, size);
> +       if (ret)
> +               return ret;
> +
>         return scnprintf(msg, size,
>         "The sys_perf_event_open() syscall returned with %d (%s) for event (%s).\n"
>         "/bin/dmesg | grep -i perf may provide additional information.\n",
> diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
> index 9f06d6cd5379..848534ec74fa 100644
> --- a/tools/perf/util/evsel.h
> +++ b/tools/perf/util/evsel.h
> @@ -311,6 +311,7 @@ void evsel__set_sample_id(struct evsel *evsel, bool use_sample_identifier);
>
>  void arch_evsel__set_sample_weight(struct evsel *evsel);
>  void arch__post_evsel_config(struct evsel *evsel, struct perf_event_attr *attr);
> +int arch_evsel__open_strerror(struct evsel *evsel, char *msg, size_t size);
>
>  int evsel__set_filter(struct evsel *evsel, const char *filter);
>  int evsel__append_tp_filter(struct evsel *evsel, const char *filter);
> --
> 2.41.0
>