Merge pull request #12388 from wenduwan/mtl_ofi_null_err_cxt

mtl/ofi: bail gracefully if completion error context is null

Author: wenduwan

ompi/mca/mtl/ofi/mtl_ofi.h

@@ -158,8 +158,7 @@ ompi_mtl_ofi_context_progress(int ctxt_id)
                                 "%s:%d: Error returned by request (type: %d) event callback: %zd.\n"
                                 "*** The Open MPI OFI MTL is aborting the MPI job (via exit(3)).\n",
                                 __FILE__, __LINE__, req_type, ret);
-                    fflush(stderr);
-                    exit(1);
+                    goto bail;
                 }
             }
         }
@@ -181,16 +180,23 @@ ompi_mtl_ofi_context_progress(int ctxt_id)
              * thread fetches the entry while others get -FI_EAGAIN
              * indicating an empty queue, which is not erroneous.
              */
-            if (ret == -FI_EAGAIN)
+            if (ret == -FI_EAGAIN) {
                 return count;
+            }
             opal_output(0, "%s:%d: Error returned from fi_cq_readerr: %s(%zd).\n"
                            "*** The Open MPI OFI MTL is aborting the MPI job (via exit(3)).\n",
                            __FILE__, __LINE__, fi_strerror(-ret), ret);
-            fflush(stderr);
-            exit(1);
+            goto bail;
+        }
+
+        if (!error.op_context) {
+            opal_output(0, "%s:%d: Error returned from fi_cq_readerr with null context. "
+                           "Completion flags: %016lx\n"
+                           "*** The Open MPI OFI MTL is aborting the MPI job (via exit(3)).\n",
+                           __FILE__, __LINE__, error.flags);
+            goto bail;
         }
 
-        assert(error.op_context);
         ofi_req = TO_OFI_REQ(error.op_context);
         assert(ofi_req);
         req_type = ofi_req->type;
@@ -200,18 +206,20 @@ ompi_mtl_ofi_context_progress(int ctxt_id)
                         "%s:%d: Error returned by request (type: %d) error callback: %zd.\n"
                         "*** The Open MPI OFI MTL is aborting the MPI job (via exit(3)).\n",
                         __FILE__, __LINE__, req_type, ret);
-            fflush(stderr);
-            exit(1);
+            goto bail;
         }
     } else if (ret != -FI_EAGAIN && ret != -EINTR) {
         opal_output(0, "%s:%d: Error returned from fi_cq_read: %s(%zd).\n"
                        "*** The Open MPI OFI MTL is aborting the MPI job (via exit(3)).\n",
                        __FILE__, __LINE__, fi_strerror(-ret), ret);
-        fflush(stderr);
-        exit(1);
+        goto bail;
     }
 
     return count;
+
+bail:
+    fflush(stderr);
+    exit(1);
 }
 
 __opal_attribute_always_inline__ static inline int