gm 1.5.2-1 + rh AS patch/rh kernel 2.4.20

I recently upgraded the kernel on our beowulf nodes from 2.4.18 to
2.4.20, from redhat.  I had to apply the Advanced Server patch that
somebody from Myricom pointed me to, in order to get it to build
properly.  Since then, one of our users running an mpi code has been
getting the following error when running his job:

shandle is bfffcdd4
shandle cookie is 68
shandle at bfffcdd4
      cookie          = 68
      is_complete     = 40
      start           = 0
      bytes_as_contig = 0
[7] MPI internal Aborting program Bad address in Rendezvous send
[7] Bad address in Rendezvous send

I rebuilt his code to use ethernet, and it seems to run fine.  Anybody
ever seen this ?  Is this an artifact of my kernel upgrade or the gm
patch ?  I know I need to get to 1.6x, but the last time I tried to go
there, it looked like users were going to need to recompile their code,
or I would have to upgrade all of the nodes at once (or something like
that, it's been a long while) so I nixed that idea for the time being.

Thanks,

Todd