Bug 1263655 - PSM support missing
Summary: PSM support missing
Keywords:
Status: CLOSED ERRATA
Alias: None
Product: Fedora
Classification: Fedora
Component: openmpi
Version: 24
Hardware: Unspecified
OS: Unspecified
unspecified
unspecified
Target Milestone: ---
Assignee: Michal Schmidt
QA Contact: Fedora Extras Quality Assurance
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2015-09-16 11:14 UTC by Dave Love
Modified: 2016-10-31 07:36 UTC (History)
3 users (show)

Fixed In Version: openmpi-1.10.4-3 openmpi-1.10.4-4.fc25
Doc Type: Bug Fix
Doc Text:
Clone Of:
Environment:
Last Closed: 2016-10-31 07:36:29 UTC
Type: Bug
Embargoed:


Attachments (Terms of Use)

Description Dave Love 2015-09-16 11:14:06 UTC
Description of problem:

I notice there's no support for the PSM MTL needed for "the other"
Infiniband hardware from Intel (whatever it's called this year).

It is provided in RHEL as the openmpi-psm package, depending on
infinipath-psm.  infinipath-psm isn't in Fedora, but the el6 source
package builds directly on f22.

Infinipath packaging would also be relevant to libfabric, but I
don't know if building openmpi-1.10 against libfabric would be equivalent
to the PSM MTL.

Comment 1 Orion Poplawski 2015-09-19 03:47:24 UTC
Personally, I don't have any time/interest to work on this.  But if psm/psm2/whatever were packaged for Fedora, it would be trivial to enable openmpi support.

Comment 2 Jan Kurik 2016-02-24 13:45:31 UTC
This bug appears to have been reported against 'rawhide' during the Fedora 24 development cycle.
Changing version to '24'.

More information and reason for this action is here:
https://fedoraproject.org/wiki/Fedora_Program_Management/HouseKeeping/Fedora24#Rawhide_Rebase

Comment 3 Dave Love 2016-05-05 15:35:00 UTC
FWIW, I noticed infinipath-psm is recently packaged.

Comment 4 Michal Schmidt 2016-06-15 13:13:33 UTC
libpsm2 is available in Rawhide (but not in F24). To build openmpi with PSM2 support it should be sufficient to add libpsm2-devel to BuildRequires.

However, openmpi does not build in Rawhide, due to glibc finally poisoning the __malloc_initialize_hook symbol:
https://sourceware.org/bugzilla/show_bug.cgi?id=19564
https://sourceware.org/git/gitweb.cgi?p=glibc.git;a=commitdiff;h=2ba3cfa1607c36613f3b30fb1ae4ec530245ce64

In file included from btl_openib_component.c:58:0:
../../../../opal/mca/memory/linux/malloc.h:233:29: error: attempt to use poisoned "__malloc_initialize_hook"
 OPAL_DECLSPEC extern void (*__malloc_initialize_hook) __MALLOC_PMT ((void));
                             ^

Comment 5 Michal Schmidt 2016-06-15 14:25:10 UTC
v1.10.3 (upstream release likely to happen today) will build against the new glibc, because ./configure added a check for "__malloc_initialize_hook", so it will disable MCA component memory:linux.

https://github.com/open-mpi/ompi-release/commit/c2798a8bd6dddf68b69608ee04500261a0eda842

Comment 6 Orion Poplawski 2016-10-20 22:28:01 UTC
Built for rawhide.  Hope to move to older releases in a bit.

Comment 7 Fedora Update System 2016-10-26 04:24:04 UTC
openmpi-1.10.4-4.fc25 has been pushed to the Fedora 25 testing repository. If problems still persist, please make note of it in this bug report.
See https://fedoraproject.org/wiki/QA:Updates_Testing for
instructions on how to install test updates.
You can provide feedback for this update here: https://bodhi.fedoraproject.org/updates/FEDORA-2016-9b130deecb

Comment 8 Fedora Update System 2016-10-31 07:36:29 UTC
openmpi-1.10.4-4.fc25 has been pushed to the Fedora 25 stable repository. If problems still persist, please make note of it in this bug report.


Note You need to log in before you can comment on or make changes to this bug.