[libfabric-users] Race in GNI provider?

[Sung-Eun Choi]

Hi Thomas,

No this doesn't look familiar, but we've done a bit of a shake up with
our VC progress code lately and possible it hasn't see enough action
yet.  Do you have a small reproducer?

Actually this mail is reminding me that none of us ever responded to
John's mail about locks in the GNI provider (Sorry John!).  We were
actually meeting face-to-face at the time, and talked a bit about it.
We'll follow up.

-- Sung

On Wed, Mar 22, 2017 at 10:03:52PM +0100, Thomas Heller wrote:
> Hi List,
> 
> I am currently looking into some of the issues related to our new libfabric 
> message layer that John initially wrote. I am running into segmentation fault
> out of libfabric with a backtrace like this:
> https://gist.github.com/sithhell/72092260662a5ffe64e2e682543fe166
> Other threads have backtraces calling into that function and spinning on 
> locks, for example here:
> https://gist.github.com/sithhell/cc18f4cd2ef6f35c279b99bc78b3623e
> 
> Does this ring a bell for anyone? It doesn't happen when using just one 
> thread...
> I am using the latest HEAD from here:
> https://github.com/ofi-cray/libfabric-cray
> Commit: 001186332f19a37e60da2dd41f4baab5972e145e
> 
> Any help is appreciated!
> 
> Thanks,
> Thomas
> _______________________________________________
> Libfabric-users mailing list
> Libfabric-users at lists.openfabrics.org
> http://lists.openfabrics.org/mailman/listinfo/libfabric-users