Merge pull request #6918 from hoopoepg/topic/fixed-hand-on-shmem-finalize

SPML/UCX: fixed hang in SHMEM_FINALIZE

Author: yosefe

opal/mca/common/ucx/common_ucx.c

@@ -187,8 +187,11 @@ static void opal_common_ucx_wait_all_requests(void **reqs, int count, ucp_worker
     }
 }
 
-OPAL_DECLSPEC int opal_common_ucx_del_procs_nofence(opal_common_ucx_del_proc_t *procs, size_t count,
-                                            size_t my_rank, size_t max_disconnect, ucp_worker_h worker) {
+OPAL_DECLSPEC int opal_common_ucx_del_procs_nofence(opal_common_ucx_del_proc_t *procs,
+                                                    size_t count, size_t my_rank,
+                                                    size_t max_disconnect,
+                                                    ucp_worker_h worker)
+{
     size_t num_reqs;
     size_t max_reqs;
     void *dreq, **dreqs;

oshmem/mca/spml/ucx/spml_ucx.c

@@ -124,18 +124,16 @@ int mca_spml_ucx_del_procs(ompi_proc_t** procs, size_t nprocs)
         mca_spml_ucx_ctx_default.ucp_peers[i].ucp_conn = NULL;
     }
 
-    ret = opal_common_ucx_del_procs(del_procs, nprocs, oshmem_my_proc_id(),
-                                    mca_spml_ucx.num_disconnect,
-                                    mca_spml_ucx_ctx_default.ucp_worker);
-
+    ret = opal_common_ucx_del_procs_nofence(del_procs, nprocs, oshmem_my_proc_id(),
+                                            mca_spml_ucx.num_disconnect,
+                                            mca_spml_ucx_ctx_default.ucp_worker);
+    /* No need to barrier here - barrier is called in _shmem_finalize */
     free(del_procs);
     free(mca_spml_ucx.remote_addrs_tbl);
     free(mca_spml_ucx_ctx_default.ucp_peers);
 
     mca_spml_ucx_ctx_default.ucp_peers = NULL;
 
-    opal_common_ucx_mca_proc_added();
-
     return ret;
 }
 
@@ -323,6 +321,8 @@ int mca_spml_ucx_add_procs(ompi_proc_t** procs, size_t nprocs)
     free(wk_roffs);
 
     SPML_UCX_VERBOSE(50, "*** ADDED PROCS ***");
+
+    opal_common_ucx_mca_proc_added();
     return OSHMEM_SUCCESS;
 
 error2: