[Pkg-openmpi-maintainers] Bug#790540: /usr/lib/libmpi.so.1: Segfault in MPI_Init for simplest MPI program

[Tom Epperly]

Package: libopenmpi1.6
Version: 1.6.5-10
Severity: normal
File: /usr/lib/libmpi.so.1

Dear Maintainer,


   * What led up to the situation?

I was trying to install QBox http://eslab.ucdavis.edu/software/qbox/index.htm.
I compiled it, but it segfaults when I run it with mpirun -np 1 ./qb
qb needs a variety of packages that require MPI, so openmpi was installed
using apt-get by installing the dependencies for QBox.

Oddly enough, QBox doesn't segfault when I run it without mpirun.

I made a simple program to reproduce the behavior.

# cat mpitest.c
#include <mpi.h>

int
main(int argc, char **argv)
{
  MPI_Init(&argc, &argv);
  MPI_Finalize();
}


   * What exactly did you do (or not do) that was effective (or
     ineffective)?

Here are the steps needed to reproduce the behavior.
faerun:/tmp# mpicc -o mpitest mpitest.c
faerun:/tmp# mpirun -np 1 /tmp/mpitest
[faerun:20716] *** Process received signal ***
[faerun:20716] Signal: Segmentation fault (11)
[faerun:20716] Signal code:  (128)
[faerun:20716] Failing at address: (nil)
[faerun:20716] [ 0] /lib/x86_64-linux-gnu/libpthread.so.0(+0xf8d0) [0x7f7b39bef8d0]
[faerun:20716] [ 1] /usr/lib/x86_64-linux-gnu/libhwloc.so.5(+0x244c5) [0x7f7b39e244c5]
[faerun:20716] [ 2] /usr/lib/x86_64-linux-gnu/libhwloc.so.5(+0xc1c7) [0x7f7b39e0c1c7]
[faerun:20716] [ 3] /usr/lib/x86_64-linux-gnu/libhwloc.so.5(hwloc_topology_load+0x1ab) [0x7f7b39e0d15a]
[faerun:20716] [ 4] /usr/lib/libmpi.so.1(opal_hwloc_unpack+0xc7) [0x7f7b3a34e107]
[faerun:20716] [ 5] /usr/lib/libmpi.so.1(opal_dss_unpack_buffer+0x6e) [0x7f7b3a3497ce]
[faerun:20716] [ 6] /usr/lib/libmpi.so.1(opal_dss_unpack+0xae) [0x7f7b3a34a6de]
[faerun:20716] [ 7] /usr/lib/libmpi.so.1(orte_util_nidmap_init+0x2b1) [0x7f7b3a315591]
[faerun:20716] [ 8] /usr/lib/openmpi/lib/openmpi/mca_ess_env.so(+0x1b29) [0x7f7b380b9b29]
[faerun:20716] [ 9] /usr/lib/libmpi.so.1(orte_init+0x193) [0x7f7b3a2fdc63]
[faerun:20716] [10] /usr/lib/libmpi.so.1(ompi_mpi_init+0x166) [0x7f7b3a2b81d6]
[faerun:20716] [11] /usr/lib/libmpi.so.1(MPI_Init+0x16a) [0x7f7b3a2d095a]
[faerun:20716] [12] /tmp/mpitest() [0x400708]
[faerun:20716] [13] /lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xf5) [0x7f7b39851b45]
[faerun:20716] [14] /tmp/mpitest() [0x400619]
[faerun:20716] *** End of error message ***
--------------------------------------------------------------------------
mpirun noticed that process rank 0 with PID 20716 on node faerun exited on signal 11 (Segmentation fault).
--------------------------------------------------------------------------
faerun:/tmp# /tmp/mpitest
faerun:/tmp# 

   * What was the outcome of this action?

Note the segfault listed above.

   * What outcome did you expect instead?

I expected mpitest to run without segfaulting. Note, running mpitest
without mpirun doesn't segfault. Using mpirun should allow me to run
using multiple threads.

faerun:/tmp# dpkg --list '*openmpi*'
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
||/ Name                  Version         Architecture    Description
+++-=====================-===============-===============-===============================================
ii  libblacs-openmpi1     1.1-33          amd64           Basic Linear Algebra Comm. Subprograms - Shared
ii  libopenmpi-dev        1.6.5-10        amd64           high performance message passing library -- hea
un  libopenmpi1           <none>          <none>          (no description available)
rc  libopenmpi1.3         1.4.5-1         amd64           high performance message passing library -- sha
ii  libopenmpi1.6         1.6.5-10        amd64           high performance message passing library -- sha
un  libopenmpi2           <none>          <none>          (no description available)
ii  libscalapack-openmpi1 1.8.0-12.1      amd64           Scalable Linear Algebra Package - Shared libs. 
un  octave-openmpi-ext    <none>          <none>          (no description available)
ii  openmpi-bin           1.6.5-10        amd64           high performance message passing library -- bin
ii  openmpi-checkpoint    1.6.5-10        amd64           high performance message passing library -- che
ii  openmpi-common        1.6.5-10        all             high performance message passing library -- com
un  openmpi-dev           <none>          <none>          (no description available)
un  openmpi-libs0         <none>          <none>          (no description available)
un  yorick-mpy-openmpi    <none>          <none>          (no description available)


-- System Information:
Debian Release: stretch/sid
  APT prefers unstable
  APT policy: (500, 'unstable'), (1, 'experimental')
Architecture: amd64 (x86_64)

Kernel: Linux 3.10-2-amd64 (SMP w/4 CPU cores)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/bash
Init: systemd (via /run/systemd/system)

Versions of packages libopenmpi1.6 depends on:
ii  libc6         2.19-18
ii  libcr0        0.8.5-2.2
ii  libgcc1       1:5.1.1-9
ii  libgfortran3  5.1.1-9
ii  libhwloc5     1.11.0-1
ii  libibverbs1   1.1.8-1.1
ii  libltdl7      2.4.2-1.11
ii  libquadmath0  5.1.1-9
ii  libstdc++6    5.1.1-9

libopenmpi1.6 recommends no packages.

libopenmpi1.6 suggests no packages.

-- no debconf information