Fix a race condition when multiple threads try to create a bml endpoint simultaneously.

ompi/mca/bml/r2/bml_r2.c

@@ -160,7 +160,6 @@ static mca_bml_base_endpoint_t *mca_bml_r2_allocate_endpoint (ompi_proc_t *proc)
     mca_bml_base_btl_array_reserve(&bml_endpoint->btl_rdma,  mca_bml_r2.num_btl_modules);
     bml_endpoint->btl_max_send_size = -1;
     bml_endpoint->btl_proc = proc;
-    proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_BML] = bml_endpoint;
 
     bml_endpoint->btl_flags_or = 0;
     return bml_endpoint;
@@ -430,6 +429,12 @@ static int mca_bml_r2_add_proc (struct ompi_proc_t *proc)
     /* compute metrics for registered btls */
     mca_bml_r2_compute_endpoint_metrics (bml_endpoint);
 
+    /* do it last, for the lazy initialization check in bml_base_get* */
+#if OPAL_ENABLE_THREAD_MULTI
+    opal_atomic_wmb();
+#endif /* OPAL_ENABLE_THREAD_MULTI */
+    proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_BML] = bml_endpoint;
+
     return OMPI_SUCCESS;
 }
 
@@ -523,6 +528,7 @@ static int mca_bml_r2_add_procs( size_t nprocs,
 
             if (NULL == bml_endpoint) {
                 bml_endpoint = mca_bml_r2_allocate_endpoint (proc);
+                proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_BML] = bml_endpoint;
                 if (NULL == bml_endpoint) {
                     free(btl_endpoints);
                     free(new_procs);