perf test: Fix ftrace test with regex patterns

During the parallel testing, I've noticed some ftrace test failures.  It
seems the regex pattern checks 100 msec of nanosleep with the error
range of 10 msec.  But sometimes it's affected by other processes and
resulted in more time in the syscall.

The following output shows that it took more than 120 msec and failed.
Let's update the regex pattern so that it can allow more drifts.

  perf ftrace profile test
  # Total (us)   Avg (us)   Max (us)      Count   Function
    121279.500 121279.500 121279.500          1   __x64_sys_clock_nanosleep
    121278.400 121278.400 121278.400          1   common_nsleep
    121277.800 121277.800 121277.800          1   hrtimer_nanosleep
    121277.100 121277.100 121277.100          1   do_nanosleep
    341760.289  56960.048 121273.400          6   schedule
       176.200     25.171     31.616          7   scheduler_tick
         0.923      0.923      0.923          1   native_smp_send_reschedule
    345522.360  69104.472 345320.600          5   __x64_sys_execve
    345486.585  69097.317 345312.700          5   do_execveat_common.isra.0
    340730.300 340730.300 340730.300          1   bprm_execve
         1.758      0.879      0.883          2   sched_mm_cid_before_execve
         1.112      1.112      1.112          1   sched_mm_cid_after_execve
  ---- end(-1) ----
   81: perf ftrace tests                                               : FAILED!

Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20241102231702.2262258-1-namhyung@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
This commit is contained in:
Namhyung Kim 2024-11-02 16:17:02 -07:00
parent a52143aa21
commit 29bf07bc9a

View File

@ -72,7 +72,7 @@ test_ftrace_profile() {
grep sleep "${output}"
grep schedule "${output}"
grep execve "${output}"
time_re="[[:space:]]+10[[:digit:]]{4}\.[[:digit:]]{3}"
time_re="[[:space:]]+1[[:digit:]]{5}\.[[:digit:]]{3}"
# 100283.000 100283.000 100283.000 1 __x64_sys_clock_nanosleep
# Check for one *clock_nanosleep line with a Count of just 1 that takes a bit more than 0.1 seconds
# Strip the _x64_sys part to work with other architectures