nptl01 test in LTP fails for the 20070415 release on sparc

[Marc Tardif]

The nptl01 test is for detecting a bug found in the pthread_cond_timedwait() system call of the Native POSIX Thread Library (NPTL) library code. There was an error path in the system call, where the sequence counters were getting updated w/o holding the internal condvar lock. A FAIL is indicated by the test hanging and not completing execution.

The nptl01 test hung for the 20070415 release installed on both T1000 and T2000 hardware.

[Marc Tardif (cr3)]

The nptl01 test is part of the ltp-kernel-test package. For your convenience, I have attached the relevant file from the source package.

[Fabio Massimo Di Nitto (fabbione)]

this is another hang in FUTEX_WAIT but at least we have a much smaller test case then building openoffice.

Fabio

[Marc Tardif (cr3)]

I tried rebooting from the linux-image-2.6.20-15-sparc64 kernel and then ran the same test 1000 times:

  export PATH="$PATH:/usr/lib/debian-test/tools:/usr/lib/debian-test/tests/linux/testcases/bin"
  export LTPROOT="/usr/lib/debian-test/tests/linux";
  for i in `seq 1 1000`; do nptl01; done

That never hung, so that seems to confirm the bug is SMP related.

[Fabio Massimo Di Nitto (fabbione)]

Patch has been made available. Note that this bug affects all arches and not just sparc. Diff is that sparc did show it quickly due to the amount of CPUs locking/unlocking.

Fabio

[Cesare Tirabassi (norsetto)]

I hope I'm not injecting unwanted noise here.
It seems to me that bug #107093 is also due to a hang in FUTEX_WAIT, or am I totally wrong?

[Fabio Massimo Di Nitto (fabbione)]

this was fixed with the last feisty kernel upload to -security