Hi all,

Attached is a test app that checks three details of the system V 
semaphore implementation:
- is forward progress guaranteed before other tasks are woken up?
- is sem_ctime updated by semctl(SETVAL)?
- is it handled that a "decrease" operation can wake up a task that does 
a wait-for-zero, even if the result of the decrease is not 0?

If I read the available sources correctly, Linux
- implements sem_ctime differently from all other Unices I found.
- tries to implement more (hopefully: all) corner cases with complex 
semops's.

Here is the result of my code review:
http://calculix-rpm.sourceforge.net/sysvsem.html

It would be great if someone with access to e.g. FreeBSD or other Unix 
versions could run it and confirm my review.

Chris, Nick: Btw, Linux is the only implementation that guarantees FIFO 
for semop().

Here is the Linux output:
<<<
   ****************************************
   test1: do a non-trivial sleep
  got semaphore array d0002h.
    SETVAL succeeded.
   IPC_RMID returned 0, errno now 0 (expected: 0, 0).
   Child 0(3636): sys 0.000000, user 0.000000.
   Child 1(3637): sys 0.000000, user 0.000000.
   Child 2(3638): sys 0.000000, user 0.000000.
   Child 3(3639): sys 0.000000, user 0.000000.
   Child 4(3640): sys 0.000000, user 0.000999.
  Total time: sys 0.000999, user 0.000000.
+++ test1 success: No Livelock
  got semaphore array d8002h.
   ****************************************
   test2: does SETVAL update sem_ctime?
    before SETVAL: ctime now 1273832652 (3).
    SETVAL succeeded.
    before SETVAL: ctime now 1273832655 (0).
  INFO: OS updates sem_ctime in semctl(SETVAL)
   ****************************************
   test3: check that kernel wakes up child that waits for non-zero value.
    SETVAL succeeded.
    child completed, before exit: GETVAL now 2.
    parent: after sleep: GETVAL now 2.
    parent: after waitpid: GETVAL now 2.
   +++ test3 success: kernel wakes up child correctly.
   ****************************************
   Done, cleanup.
    IPC_RMID returned 0, errno now 0 (expected: 0, 0).

 >>>

And here opensolaris:
<<<
  ./complexdec
   ****************************************
   test1: do a non-trivial sleep
  got semaphore array 2h.
    SETVAL succeeded.
   IPC_RMID returned 0, errno now 0 (expected: 0, 0).
   Child 0(6815): sys 0.000000, user 4.160000.
   Child 1(6816): sys 0.000000, user 4.160000.
   Child 2(6817): sys 0.000000, user 4.130000.
   Child 3(6818): sys 0.000000, user 4.090000.
   Child 4(6819): sys 0.000000, user 4.080000.
  Total time: sys 20.20620000, user 0.000000.
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
test 1 bad: OS should hide in shame, Livelock detected.
  got semaphore array 3h.
   ****************************************
   test2: does SETVAL update sem_ctime?
    before SETVAL: ctime now 1273830869 (3).
    SETVAL succeeded.
    before SETVAL: ctime now 1273830869 (3).
  INFO: OS performs no update to sem_ctime in semctl(SETVAL)
   ****************************************
   test3: check that kernel wakes up child that waits for non-zero value.
    SETVAL succeeded.
    parent: after sleep: GETVAL now 1.
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
Test 3 bad: got unexpected value (child not woken up?).
   ****************************************
   Done, cleanup.
    IPC_RMID returned 0, errno now 0 (expected: 0, 0).
child: semop failed.

 >>>

--
     Manfred