Merge pull request #7505 from hkuno/john.l.byrne/btl_ofi

Fix btl ofi clean-up logic

opal/mca/btl/ofi/btl_ofi_component.c

@@ -581,6 +581,12 @@ static int mca_btl_ofi_init_device(struct fi_info *info)
 fail:
     /* clean up */
 
+    /* close basic ep before closing av */
+    if (NULL != ep && !module->is_scalable_ep) {
+        fi_close(&ep->fid);
+        ep = NULL;
+    }
+
     /* if the contexts have not been initiated, num_contexts should
      * be zero and we skip this. */
     for (int i=0; i < module->num_contexts; i++) {
@@ -588,14 +594,16 @@ fail:
     }
     free(module->contexts);
 
-    if (NULL != av) {
-        fi_close(&av->fid);
-    }
-
+    /* check for NULL ep to avoid double-close */
     if (NULL != ep) {
         fi_close(&ep->fid);
     }
 
+    /* close av after closing basic ep */
+    if (NULL != av) {
+        fi_close(&av->fid);
+    }
+
     if (NULL != domain) {
         fi_close(&domain->fid);
     }

opal/mca/btl/ofi/btl_ofi_module.c

@@ -277,20 +277,33 @@ int mca_btl_ofi_finalize (mca_btl_base_module_t* btl)
 
     assert(btl);
 
+    /* clear the rcache */
+    if (ofi_btl->rcache) {
+        mca_rcache_base_module_destroy (ofi_btl->rcache);
+        ofi_btl->rcache = NULL;
+    }
+
+    /* Close basic ep before closing its attached resources. */
+    if (NULL != ofi_btl->ofi_endpoint && !ofi_btl->is_scalable_ep) {
+        fi_close(&ofi_btl->ofi_endpoint->fid);
+        ofi_btl->ofi_endpoint = NULL;
+    }
+
     /* loop over all the contexts */
     for (i=0; i < ofi_btl->num_contexts; i++) {
         mca_btl_ofi_context_finalize(&ofi_btl->contexts[i], ofi_btl->is_scalable_ep);
     }
     free(ofi_btl->contexts);
 
-    if (NULL != ofi_btl->av) {
-        fi_close(&ofi_btl->av->fid);
-    }
-
     if (NULL != ofi_btl->ofi_endpoint) {
         fi_close(&ofi_btl->ofi_endpoint->fid);
     }
 
+    /* close ep before closing av */
+    if (NULL != ofi_btl->av) {
+        fi_close(&ofi_btl->av->fid);
+    }
+
     if (NULL != ofi_btl->domain) {
         fi_close(&ofi_btl->domain->fid);
     }
@@ -313,11 +326,6 @@ int mca_btl_ofi_finalize (mca_btl_base_module_t* btl)
     OBJ_DESTRUCT(&ofi_btl->id_to_endpoint);
     OBJ_DESTRUCT(&ofi_btl->module_lock);
 
-    if (ofi_btl->rcache) {
-        mca_rcache_base_module_destroy (ofi_btl->rcache);
-        ofi_btl->rcache = NULL;
-    }
-
     free (btl);
 
     return OPAL_SUCCESS;