[mvapich-discuss] SIGSEGV in MPIDI_CH3I_MRAIL_Parse_header

[Jian Lin]

Hi, Johannes,

Thanks for your notes! However, we didn't observe such issue in our
tests. Will you please provide more detailed information on your
environment, including platform, network adapter, MVAPICH2 version,
configuration options, and the runtime command-line? Also, will you
please provide a reproducer that can trigger the issue? With those, 
we can further explore it. Thank you!

On Wed, 24 Jun 2015 22:58:46 +0200
Johannes Ziegenbalg <johannes.ziegenbalg at zih.tu-dresden.de> wrote:

> Hello.
> 
> I'm a student at Dresden University, working on my diploma thesis. I
> have a MPI client-server app (MPI_Comm_connect/accept) and the server
> receives a SIGSEGV in the function MPIDI_CH3I_MRAIL_Parse_header, when
> the client connects. 
> After a little research I found out that it occurs in the macro
> SET_CREDIT in the file
> mpid/ch3/channels/mrail/src/gen2/ibv_recv.c:382 And a little more
> research later it has emerged that the array vc->mrail.srp.credits is
> still unallocated. One later "version" of a MPIDI_CH3_VC, on an other
> Thread, gets a correct initialization (with MRAILI_Init_vc), but it's
> not used (see attachment).
> 
> The debug output and the back-traces from GDB are attached.
> 
> My configure call is: ./configure --enable-threads=multiple --enable
> -fortran=all --with-pm=slurm --with-pmi=pmi1 --with-rdma=gen2 --enable
> -hybrid --with-device=ch3:mrail --enable-g=dbg,log,most -q --enable
> -debuginfo --disable-fast
> 
> Thanks, in advance, for your help!
> 
> Regards,
> Johannes Ziegenbalg


-- 
Jian Lin
http://linjian.org