[LibOS] test program for sigsuspend and alarm

[yamahata]

This test program shows race condition between thread_sleep() and
thread_wakeup() and DkThreadResume(thread->pal_handle).

With graphene, sigsuspend doesn't return or results in assert.

assert failed shim_syscalls.c:260 preempt == get_cur_preempt() (value:0)

Signed-off-by: Isaku Yamahata isaku.yamahata@gmail.com

Please fill in the following form before submitting this PR:

• Affected components

  • README and global configurations
  • Linux PAL
  • SGX PAL
  • FreeBSD PAL
  • Library OS (i.e., SHIM), including GLIBC

• A brief description of this PR (describe the reasons and measures)

• How to test this PR?

  • Documentation-only; no need to test
    no need to test because this is a test program

Please follow the coding style guidelines. Do not submit PRs which violate these rules.

• Comments and commit messages: no spelling or grammatical errors
• 4 spaces per indentation level, at most 100 chars per line
• Asterisks (*) next to types: int* pointer; (one pointer each line)
• Braces ({) begin in the same line as function names, if, else, while, do, for, switch, union, and struct keywords.
• Naming: Macros, constants - NAMED_THIS_WAY; global variables - g_named_this_way; others - named_this_way.
• Other styling issues may be pointed out by reviewers.

Please preserve the following checklist for reviewing:

• Pass all CI unit tests
• Resolve all discussions/requests from reviewers
• Reviewer approval (select one):
  • 3 approving reviews
  • 2 approving reviews and 5 days since the PR was created
  • The PR is from a maintainer; 1 approving review and 10 days since the PR was created

---

This change is 

[therealoscarbot]

Can one of the admins verify this patch?

[therealoscarbot]

Can one of the admins verify this patch?

[therealoscarbot]

Can one of the admins verify this patch?

[yamahata]

#453

[dimakuv]

It is unclear how to incorporate this test into regression suite: the data race is non-deterministic (it takes up to several minutes on my machine to trigger the race). I suggest to move this test case under LibOS/shim/test/native, so interested people can play with it.

Note: This data race is visible only in non-debug mode in manifest: loader.debug_type = none.

Reviewed 1 of 1 files at r1.
Reviewable status: all files reviewed, 1 unresolved discussion, not enough approvals from maintainers (2 more required), not enough approvals from different teams (2 more required, approved so far: ) (waiting on @yamahata)

---

LibOS/shim/test/regression/sigsuspend.c, line 52 at r1 (raw file):

        struct timespec t = {
            .tv_sec = 1,

If we change from 1 to say 5, the data race disappears but Graphene-SGX hangs deterministically. This behavior is explained in #453 (comment). We can implement a regression test by changing this line. This bug will require a separate fix (again, see the link).

[donporter]

Thanks for the test case. This also needs a python wrapper that can automatically run this test as part of 'make regression' and tell if there is a failure or success. Since this is a separate PR from the fix, my suggestion would be to go ahead and write the test and have it fail, but mask the error code until the fix lands.

After reading @dimakuv's comments, I think he is advocating we do some other fixes first.

Reviewable status: all files reviewed, 1 unresolved discussion, not enough approvals from maintainers (2 more required), not enough approvals from different teams (2 more required, approved so far: ) (waiting on @yamahata)

[stefanberger]

Would we still need this test case now since we have LTP's rt_sigsuspend01 and sigsuspend01?

[dimakuv]

I would suggest to close this PR. @yamahata What do you think?

[dimakuv]

This PR is outdated, the tests work correctly on current master.