UCT/CUDA/COPY: Revert qualify cuda_copy for RNDV with peer-failure EH

Author: pentschev

src/uct/cuda/cuda_copy/cuda_copy_iface.c

@@ -104,11 +104,7 @@ static ucs_status_t uct_cuda_copy_iface_query(uct_iface_h tl_iface,
     iface_attr->iface_addr_len          = sizeof(uct_cuda_copy_iface_addr_t);
     iface_attr->device_addr_len         = 0;
     iface_attr->ep_addr_len             = 0;
-    /* UCT_IFACE_FLAG_ERRHANDLE_PEER_FAILURE required for RMA BW wireup
-     * (ucp_wireup_fill_peer_err_criteria) when error handling is requested.
-     * Transfers are local copies; UCP handles invalidation when a peer fails. */
-    iface_attr->cap.flags               = UCT_IFACE_FLAG_ERRHANDLE_PEER_FAILURE |
-                                          UCT_IFACE_FLAG_CONNECT_TO_IFACE |
+    iface_attr->cap.flags               = UCT_IFACE_FLAG_CONNECT_TO_IFACE |
                                           UCT_IFACE_FLAG_GET_SHORT |
                                           UCT_IFACE_FLAG_PUT_SHORT |
                                           UCT_IFACE_FLAG_GET_ZCOPY |

src/uct/cuda/cuda_copy/cuda_copy_md.c

@@ -20,7 +20,6 @@
 #include <ucs/sys/ptr_arith.h>
 #include <uct/cuda/base/cuda_ctx.inl>
 #include <uct/api/v2/uct_v2.h>
-#include <uct/base/uct_iface.h>
 #include <cuda.h>
 #if CUDA_VERSION >= 11070
 #include <cudaTypedefs.h>
@@ -119,14 +118,7 @@ uct_cuda_copy_md_query(uct_md_h uct_md, uct_md_attr_v2_t *md_attr)
     uct_cuda_copy_md_t *md = ucs_derived_of(uct_md, uct_cuda_copy_md_t);
 
     uct_md_base_md_query(md_attr);
-    /* Advertise RMA invalidation so RMA BW lanes qualify for UCP wireup when
-     * UCP_EP_INIT_ERR_MODE_PEER_FAILURE is set with RNDV (see
-     * ucp_wireup_add_rma_bw_lanes). cuda_copy has no remote RMA keys; local
-     * registration teardown is synchronous and satisfies the invalidate
-     * contract via uct_cuda_copy_mem_dereg(). */
-    md_attr->flags            = UCT_MD_FLAG_REG | UCT_MD_FLAG_ALLOC |
-                                UCT_MD_FLAG_INVALIDATE |
-                                UCT_MD_FLAG_INVALIDATE_RMA;
+    md_attr->flags            = UCT_MD_FLAG_REG | UCT_MD_FLAG_ALLOC;
     md_attr->reg_mem_types    = UCS_BIT(UCS_MEMORY_TYPE_HOST) |
                                 UCS_BIT(UCS_MEMORY_TYPE_CUDA) |
                                 UCS_BIT(UCS_MEMORY_TYPE_CUDA_MANAGED);
@@ -188,21 +180,13 @@ UCS_PROFILE_FUNC(ucs_status_t, uct_cuda_copy_mem_dereg,
                  (md, params),
                  uct_md_h md, const uct_md_mem_dereg_params_t *params)
 {
-    unsigned flags;
-
-    UCT_MD_MEM_DEREG_CHECK_PARAMS(params, 1);
+    UCT_MD_MEM_DEREG_CHECK_PARAMS(params, 0);
 
     if (params->memh != &uct_cuda_dummy_memh) {
         UCT_CUDADRV_FUNC(cuMemHostUnregister((void*)params->memh),
                          UCS_LOG_LEVEL_DIAG);
     }
 
-    flags = UCT_MD_MEM_DEREG_FIELD_VALUE(params, flags, FIELD_FLAGS, 0);
-    if (flags & UCT_MD_MEM_DEREG_FLAG_INVALIDATE) {
-        ucs_assert(params->comp != NULL); /* suppress coverity false-positive */
-        uct_invoke_completion(params->comp, UCS_OK);
-    }
-
     return UCS_OK;
 }