On Thu, Feb 26, 2009 at 4:44 PM, Brice Goglin <Brice.Goglin@...> wrote:
> stephane eranian wrote:
>> The panic message you sent out does not seem to have any perfmon related
>> calls in it.
> Right, but I have never seen any CPU lockup on a non-perfmon-patched
> kernel :)
>> Are you doing actual measurements when this happens and if so what
>> kind of measurement?
> Running 2 processes doing a MPI pingpong with MPICH2. papiex -aU was
> attached to one of them. MPICH2 used vmsplice/readv across a pipe for
> transferring data between them in this case. If I switch to MPICH2 using
> the usual mmap'ed user-space shared buffered, it seems to work fine.
What if you use pfmon instead of papiex with similar options?