perf test task_exit: No need for a cycles event to check if we get an PERF_RECORD_EXIT

The intent of this test is to check we get a PERF_RECORD_EXIT as asked
for by setting perf_event_attr.task=1.

When the test was written we didn't had the "dummy" event so we went
with the default event, "cycles".

There were reports of this test failing sometimes, one of these reports
was with a PREEMPT_RT_FULL, but I noticed it failing sometimes with an
aarch64 Firefly board.

In the kernel the call to perf_event_task_output(), that generates the
PERF_RECORD_EXIT may fail when there is not enough memory in the ring
buffer, if the ring buffer is paused, etc.

So switch to using the "dummy" event to use the ring buffer just for
what the test was designed for, avoiding uneeded PERF_RECORD_SAMPLEs.

Acked-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/lkml/ZLGXmMuNRpx1ubFm@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This commit is contained in:
Arnaldo Carvalho de Melo 2023-07-14 15:29:31 -03:00
parent 963293ff05
commit 2480232c61
1 changed files with 2 additions and 2 deletions

View File

@ -58,9 +58,9 @@ static int test__task_exit(struct test_suite *test __maybe_unused, int subtest _
signal(SIGCHLD, sig_handler);
evlist = evlist__new_default();
evlist = evlist__new_dummy();
if (evlist == NULL) {
pr_debug("evlist__new_default\n");
pr_debug("evlist__new_dummy\n");
return -1;
}