Re: [PATCH v5 4/6] perf cpumap: Handle dummy maps as empty in subset

[Adrian Hunter <adrian.hunter@intel.com>]

On 3/05/22 17:03, Ian Rogers wrote:
> On Tue, May 3, 2022 at 12:43 AM Adrian Hunter <adrian.hunter@intel.com> wrote:
>>
>> On 3/05/22 07:17, Ian Rogers wrote:
>>> perf_cpu_map__empty is true for empty and dummy maps. Make is_subset
>>> respect that.
>>
>> As I wrote before, I am not keen on this because it prevents -1, as a
>> valid 3rd parameter to perf_event_open(), from being represented
>> in merged evsel cpu maps.
>>
>> Why do you want this?
> 
> Thanks Adrian, could you give me a test case (command line) where the
> differing dummy and empty behavior matters?

perf record --per-thread -e intel_pt// uname

With patchset "perf intel-pt: Better support for perf record --cpu"
the above will have (assuming 8-CPUs):
	user_requested_cpus = {-1}
	intel_pt evsel->cpus = {-1}
	text_poke dummy evsel->cpus = {0-7}
which when merged would result in:
	before this patch: all_cpus = {-1-7}
	after this patch:  all_cpus = {0-7}

The absence of -1 will mean that the intel_pt event does not get
mmapped.

>                                             Normally cpus/own_cpus are
> set to null during parsing. They may get replaced with
> user_requested_cpus:
> https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/tree/tools/lib/perf/evlist.c?h=perf/core#n44
> https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/tree/tools/lib/perf/evlist.c?h=perf/core#n45
> (should it be on line 45 that !empty is expected?)
> 
> During merge the null/empty all_cpus drops this value, which doesn't
> matter as the behavior with empty is the same as dummy:
> https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/tree/tools/lib/perf/evsel.c?h=perf/core#n119
> 
> What's concerning me is the definition of empty:
> https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/tree/tools/lib/perf/cpumap.c?h=perf/core#n279
> ```
> return map ? map->map[0].cpu == -1 : true;
> ```
> If the first entry can be -1 and there can be other CPUs merged after
> then that cpu map will be empty by the definition above. Perhaps it
> should be:
> ```
> return map ? (map->nr == 1 && map->map[0].cpu == -1) : true;
> ```
> but it seems you prefer:
> ```
> return (map == NULL) ? true : false;
> ```
> 
> You'd asked what the behavior with a dummy is and clearly it is
> somewhat muddy. That is what this patch and unit test is trying to
> clean up.
> 
> Thanks,
> Ian
> 
>>>
>>> Signed-off-by: Ian Rogers <irogers@google.com>
>>> ---
>>>  tools/lib/perf/cpumap.c   |  4 ++--
>>>  tools/perf/tests/cpumap.c | 10 +++++++++-
>>>  2 files changed, 11 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/tools/lib/perf/cpumap.c b/tools/lib/perf/cpumap.c
>>> index 384d5e076ee4..9c83675788c2 100644
>>> --- a/tools/lib/perf/cpumap.c
>>> +++ b/tools/lib/perf/cpumap.c
>>> @@ -322,9 +322,9 @@ struct perf_cpu perf_cpu_map__max(struct perf_cpu_map *map)
>>>  /** Is 'b' a subset of 'a'. */
>>>  bool perf_cpu_map__is_subset(const struct perf_cpu_map *a, const struct perf_cpu_map *b)
>>>  {
>>> -     if (a == b || !b)
>>> +     if (a == b || perf_cpu_map__empty(b))
>>>               return true;
>>> -     if (!a || b->nr > a->nr)
>>> +     if (perf_cpu_map__empty(a) || b->nr > a->nr)
>>>               return false;
>>>
>>>       for (int i = 0, j = 0; i < a->nr; i++) {
>>> diff --git a/tools/perf/tests/cpumap.c b/tools/perf/tests/cpumap.c
>>> index f94929ebb54b..d52b58395385 100644
>>> --- a/tools/perf/tests/cpumap.c
>>> +++ b/tools/perf/tests/cpumap.c
>>> @@ -128,13 +128,21 @@ static int test__cpu_map_merge(struct test_suite *test __maybe_unused, int subte
>>>       struct perf_cpu_map *a = perf_cpu_map__new("4,2,1");
>>>       struct perf_cpu_map *b = perf_cpu_map__new("4,5,7");
>>>       struct perf_cpu_map *c = perf_cpu_map__merge(a, b);
>>> +     struct perf_cpu_map *d = perf_cpu_map__dummy_new();
>>> +     struct perf_cpu_map *e = perf_cpu_map__merge(b, d);
>>>       char buf[100];
>>>
>>>       TEST_ASSERT_VAL("failed to merge map: bad nr", perf_cpu_map__nr(c) == 5);
>>>       cpu_map__snprint(c, buf, sizeof(buf));
>>>       TEST_ASSERT_VAL("failed to merge map: bad result", !strcmp(buf, "1-2,4-5,7"));
>>> -     perf_cpu_map__put(b);
>>> +
>>> +     TEST_ASSERT_VAL("failed to merge map: bad nr", perf_cpu_map__nr(e) == 3);
>>> +     cpu_map__snprint(e, buf, sizeof(buf));
>>> +     TEST_ASSERT_VAL("failed to merge map: bad result", !strcmp(buf, "4-5,7"));
>>> +
>>>       perf_cpu_map__put(c);
>>> +     perf_cpu_map__put(d);
>>> +     perf_cpu_map__put(e);
>>>       return 0;
>>>  }
>>>
>>