mpxyd: MPI IMB scatter on 12 ranks, 2 KNCs + 2 Hosts, fails

multiple QP's processing multiple completions
hit bug in mix_dto_event when copying multiple
cq wc entries into a single dat_mix_dto_comp_t msg.

Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>

diff --git a/dapl/svc/mpxyd.c b/dapl/svc/mpxyd.c
index 375efadc27f88787cc60858886166b818206866e..754b4b7d5f2187c4a1233403c4dbd0c5986e860b 100644 (file)
--- a/dapl/svc/mpxyd.c
+++ b/dapl/svc/mpxyd.c
@@ -2619,7 +2619,7 @@ static void mix_dto_event(struct mcm_cq *m_cq, struct ibv_wc *wc, int nc)
        msg.wc_cnt = nc;
 
        for (i=0; i < nc; i++) {
-               memcpy(&msg.wc, &wc[i], sizeof(*wc));
+               memcpy(&msg.wc[i], &wc[i], sizeof(*wc));
 
                if (msg.wc[i].status != IBV_WC_SUCCESS) {
                        mlog(0, " ERROR (ep=%d): cq %p id %d ctx %p stat %d"