[PATCH V2] perf tools: adding support for address filters

[Mathieu Poirier]

On 31 August 2016 at 00:57, Jiri Olsa <jolsa at redhat.com> wrote:
> On Tue, Aug 30, 2016 at 10:10:15AM -0600, Mathieu Poirier wrote:
>> This patch makes it possible to use the current filter
>> framework with address filters.  That way address filters for
>> HW tracers such as CoreSight and IntelPT can be communicated
>> to the kernel drivers.
>>
>> CC: Alexander Shishkin <alexander.shishkin at linux.intel.com>
>> Signed-off-by: Mathieu Poirier <mathieu.poirier at linaro.org>
>>
>> ---
>> Changes for V2:
>>  - Rebased to v4.8-rc4.
>>  - Revisited error path.
>> ---
>>
>>  tools/perf/util/parse-events.c | 30 +++++++++++++++++++++++++-----
>>  1 file changed, 25 insertions(+), 5 deletions(-)
>>
>> diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
>> index 2eb8b1ed4cc8..1df413fbf7f8 100644
>> --- a/tools/perf/util/parse-events.c
>> +++ b/tools/perf/util/parse-events.c
>> @@ -1760,12 +1760,26 @@ foreach_evsel_in_last_glob(struct perf_evlist *evlist,
>>  static int set_filter(struct perf_evsel *evsel, const void *arg)
>>  {
>>       const char *str = arg;
>> +     bool found = false;
>> +     int nr_addr_filters = 0;
>> +     struct perf_pmu *pmu = NULL;
>>
>> -     if (evsel == NULL || evsel->attr.type != PERF_TYPE_TRACEPOINT) {
>> -             fprintf(stderr,
>> -                     "--filter option should follow a -e tracepoint option\n");
>> -             return -1;
>> -     }
>> +     if (evsel == NULL)
>> +             goto err;
>> +
>> +     while ((pmu = perf_pmu__scan(pmu)) != NULL)
>> +             if (pmu->type == evsel->attr.type) {
>> +                     found = true;
>> +                     break;
>> +             }
>> +
>> +     if (found)
>> +             perf_pmu__scan_file(pmu, "nr_addr_filters",
>> +                                 "%d", &nr_addr_filters);
>> +
>> +
>> +     if (evsel->attr.type != PERF_TYPE_TRACEPOINT && !nr_addr_filters)
>> +             goto err;
>
> should we display another error message for !nr_addr_filters case?
> the one below is misleading in this case

The condition where nr_addr_filters == 0 is valid when we are dealing
with a tracepoint event.  Since we don't have an attribute type for HW
tracers the best we can do is improve on the error message.  What
would you think of:

"--filter option should follow an event that supports filtering"

Thanks,
Mathieu

>
> jirka
>
>>
>>       if (perf_evsel__append_filter(evsel, "&&", str) < 0) {
>>               fprintf(stderr,
>> @@ -1774,6 +1788,12 @@ static int set_filter(struct perf_evsel *evsel, const void *arg)
>>       }
>>
>>       return 0;
>> +
>> +err:
>> +     fprintf(stderr,
>> +             "--filter option should follow a -e tracepoint or HW tracer option\n");
>> +
>> +     return -1;
>>  }
>>
>>  int parse_filter(const struct option *opt, const char *str,
>> --
>> 2.7.4
>>