[Devel] [PATCH hci-8.1 v3 0/7] selftests: fix sticky failure in runners

[Konstantin Khorenko]

Hi Eva,

thank you for the series, I'm taking it, but with the following changes.

Code:

- [2/7] selftests/memfd: in a container /proc/meminfo has no
  HugePages_Free line by default. freepgs stays empty, the
  "[ $freepgs -lt $hpages_test ]" check fails with a syntax error, the
  skip is not taken and memfd_test aborts. Before the series that abort
  was a false "ok", with the patch it would be a false "not ok". Added
  a skip for that case:

  +if [ -z "$freepgs" ]; then
  +	echo "No HugePages_Free in /proc/meminfo"
  +	exit $ksft_skip
  +fi

- [4/7] selftests/zram: when comp_algorithm cannot be read, $algs is
  empty, so after the FAIL line the loop printed "SKIP 'lzo' is not
  supported by this kernel" as well. Return right after the FAIL:

   		echo "FAIL can't read /sys/block/zram${i}/comp_algorithm"
   		ERR_CODE=-1
  +		return
   	fi

- [7/7] selftests/futex: error() from logging.h only prints, so after
  a failed pthread_create() join_waiter() would call pthread_join() on
  an uninitialized pthread_t (private case) or on an already joined
  one (shared case). That is undefined behaviour and can crash the
  test before it prints its result. Fail the test right there instead:

  -		error("pthread_create failed\n", errno);
  +		ksft_exit_fail_msg("pthread_create failed\n");

Authorship:

- [1/7] ms/selftests/bpf: the author is Ricardo B. Marlière
  <rbm at suse.com>, as in upstream 2a912258c90e, not you. Please keep
  the original author when you cherry-pick, as you did in 5/7.

Commit messages:

- [2/7] Dropped "The hugetlbfs F_SEAL_EXEC bug went unnoticed that way
  until the abort was seen by eye." How a bug was found does not
  belong in the message. Added a paragraph about the container skip.
- [4/7] The first sentence chained three facts with commas, turned it
  into a list.
- [7/7] Split the sentence joined with ";" into two paragraphs, added
  a sentence about the pthread_create() failure.

Cover letter:

- Added a paragraph saying what the series does and why.
- Replaced "Blocked on VSTOR-150731" with what it means: 1/7 makes
  test_xsk.sh fail on the xsk subtest failures fixed there, and those
  fixes are already in the base.
- Fixed the author list and the diffstat.

Two issues found during review are not part of this series, I have
filed tasks for them:

- VSTOR-152656: zram_load() derives the device ids from the number of
  /dev/zram* nodes, while hot_add takes the lowest free id. With a gap
  in the ids the test works on and resets a zram device it did not
  create. The same code is in mainline; VSTOR-152657 tracks sending
  the fix there.
- VSTOR-152660: the five negative checks in futex_waitv.c compare
  res == EINVAL, which is never true, so they always pass. Upstream
  fixed this in a894f6f403320; it needs a port, see the task.

--
Best regards,
Konstantin Khorenko

-- 
Konstantin Khorenko,
Virtuozzo Linux Kernel Team