On 18/11/2020 13:21, Namhyung Kim wrote:
> Hello,
>
> On Tue, Nov 17, 2020 at 11:49 PM James Clark <james.cl...@arm.com> wrote:
>>
>> Improve the topology test to check all aggregation
>> types. This is to lock down the behaviour before
>> 'id' is changed into a struct in later commits.
>>
>> Signed-off-by: James Clark <james.cl...@arm.com>
>> Cc: Peter Zijlstra <pet...@infradead.org>
>> Cc: Ingo Molnar <mi...@redhat.com>
>> Cc: Arnaldo Carvalho de Melo <a...@kernel.org>
>> Cc: Mark Rutland <mark.rutl...@arm.com>
>> Cc: Alexander Shishkin <alexander.shish...@linux.intel.com>
>> Cc: Jiri Olsa <jo...@redhat.com>
>> Cc: Namhyung Kim <namhy...@kernel.org>
>> Cc: Thomas Richter <tmri...@linux.ibm.com>
>> Cc: John Garry <john.ga...@huawei.com>
>> ---
>> tools/perf/tests/topology.c | 53 ++++++++++++++++++++++++++++++++-----
>> 1 file changed, 46 insertions(+), 7 deletions(-)
>>
>> diff --git a/tools/perf/tests/topology.c b/tools/perf/tests/topology.c
>> index 22daf2bdf5fa..7bd8848d36b6 100644
>> --- a/tools/perf/tests/topology.c
>> +++ b/tools/perf/tests/topology.c
>> @@ -64,10 +64,11 @@ static int check_cpu_topology(char *path, struct
>> perf_cpu_map *map)
>> .path = path,
>> .mode = PERF_DATA_MODE_READ,
>> };
>> - int i;
>> + int i, id;
>>
>> session = perf_session__new(&data, false, NULL);
>> TEST_ASSERT_VAL("can't get session", !IS_ERR(session));
>> + cpu__setup_cpunode_map();
>>
>> /* On platforms with large numbers of CPUs process_cpu_topology()
>> * might issue an error while reading the perf.data file section
>> @@ -85,11 +86,18 @@ static int check_cpu_topology(char *path, struct
>> perf_cpu_map *map)
>> * "socket_id number is too big. You may need to upgrade the
>> * perf tool."
>> *
>> - * This is the reason why this test might be skipped.
>> + * This is the reason why this test might be skipped. aarch64 and
>> + * s390 always write this part of the header, even when the above
>> + * condition is true (see do_core_id_test in header.c). So always
>> + * run this test on those platforms.
>> */
>> - if (!session->header.env.cpu)
>> + if (!session->header.env.cpu
>> + && strncmp(session->header.env.arch, "s390", 4)
>> + && strncmp(session->header.env.arch, "aarch64", 7))
>> return TEST_SKIP;
>>
>> + TEST_ASSERT_VAL("Session header CPU map not set",
>> session->header.env.cpu);
>> +
>> for (i = 0; i < session->header.env.nr_cpus_avail; i++) {
>> if (!cpu_map__has(map, i))
>> continue;
>> @@ -98,14 +106,45 @@ static int check_cpu_topology(char *path, struct
>> perf_cpu_map *map)
>> session->header.env.cpu[i].socket_id);
>> }
>>
>> + // Test that core ID contains socket, die and core
>> + for (i = 0; i < map->nr; i++) {
>> + id = cpu_map__get_core(map, i, NULL);
>> + TEST_ASSERT_VAL("Core map - Core ID doesn't match",
>> + session->header.env.cpu[map->map[i]].core_id ==
>> cpu_map__id_to_cpu(id));
>> +
>> + TEST_ASSERT_VAL("Core map - Socket ID doesn't match",
>> + session->header.env.cpu[map->map[i]].socket_id ==
>> + cpu_map__id_to_socket(id));
>> +
>> + TEST_ASSERT_VAL("Core map - Die ID doesn't match",
>> + session->header.env.cpu[map->map[i]].die_id ==
>> cpu_map__id_to_die(id));
>> + }
>> +
>> + // Test that die ID contains socket and die
>> for (i = 0; i < map->nr; i++) {
>> - TEST_ASSERT_VAL("Core ID doesn't match",
>> - (session->header.env.cpu[map->map[i]].core_id ==
>> (cpu_map__get_core(map, i, NULL) & 0xffff)));
>> + id = cpu_map__get_die(map, i, NULL);
>> + TEST_ASSERT_VAL("Die map - Socket ID doesn't match",
>> + session->header.env.cpu[map->map[i]].socket_id ==
>> + cpu_map__id_to_socket(id));
>
> I'm not sure it works. It seems cpu_map__get_die() returns
> 16 bit id (socket | die) but cpu_map__id_to_socket() takes
> 32 bit id (socket | die | core), right?
Hi Namhyung,
Yes you are right. I assumed the cpu_map__id_to_...() etc functions applied in
all cases. Actually
they only work in the per core aggregation mode. In stat-display.c the id is
shifted when in die
mode to account for this:
case AGGR_DIE:
fprintf(config->output, "S%d-D%*d%s%*d%s",
cpu_map__id_to_socket(id << 16),
config->csv_output ? 0 : -8,
cpu_map__id_to_die(id << 16),
I've updated the test to match this in patchset v6. When running the test on a
multi socket machine it did fail,
but now it passes. The reason I didn't see the issue is because I only tested
the last patchset in the
series which doesn't have this issue.
Thanks
James
>
>>
>> - TEST_ASSERT_VAL("Socket ID doesn't match",
>> - (session->header.env.cpu[map->map[i]].socket_id ==
>> cpu_map__get_socket(map, i, NULL)));
>> + TEST_ASSERT_VAL("Die map - Die ID doesn't match",
>> + session->header.env.cpu[map->map[i]].die_id ==
>> cpu_map__id_to_die(id));
>> }
>>
>> + // Test that socket ID contains only socket
>> + for (i = 0; i < map->nr; i++) {
>> + id = cpu_map__get_socket(map, i, NULL);
>> + TEST_ASSERT_VAL("Socket map - Socket ID doesn't match",
>> + session->header.env.cpu[map->map[i]].socket_id ==
>> + cpu_map__id_to_socket(id));
>
> Same here.
>
> Thanks,
> Namhyung
>
>
>> + }
>> +
>> + // Test that node ID contains only node
>> + for (i = 0; i < map->nr; i++) {
>> + id = cpu_map__get_node(map, i, NULL);
>> + TEST_ASSERT_VAL("Node map - Node ID doesn't match",
>> + cpu__get_node(map->map[i]) == id);
>> + }
>> perf_session__delete(session);
>>
>> return 0;
>> --
>> 2.28.0
>>
