Merge pull request #5314 from raffenet/abort-cvar

Add cvar to dump a corefile on abort Approved-by: Hui Zhou <[email protected]>
pmodels · Jun 7, 2021 · 5c99b94 · 5c99b94
2 parents 6618ddb + c44ced2
commit 5c99b94
Show file tree

Hide file tree

Showing 3 changed files with 18 additions and 0 deletions.
diff --git a/src/mpi/init/init_impl.c b/src/mpi/init/init_impl.c
@@ -19,6 +19,15 @@
       scope       : MPI_T_SCOPE_ALL_EQ
       description : Disable printing of abort error message.
 
+    - name        : MPIR_CVAR_COREDUMP_ON_ABORT
+      category    : ERROR_HANDLING
+      type        : boolean
+      default     : false
+      class       : none
+      verbosity   : MPI_T_VERBOSITY_USER_BASIC
+      scope       : MPI_T_SCOPE_ALL_EQ
+      description : Call libc abort() to generate a corefile
+
 === END_MPI_T_CVAR_INFO_BLOCK ===
 */
 

diff --git a/src/mpid/ch3/src/mpid_abort.c b/src/mpid/ch3/src/mpid_abort.c
@@ -67,6 +67,10 @@ int MPID_Abort(MPIR_Comm * comm, int mpi_errno, int exit_code,
     MPL_error_printf("%s\n", error_msg);
     fflush(stderr);
 
+    if (MPIR_CVAR_COREDUMP_ON_ABORT) {
+        abort();
+    }
+
     /* FIXME: What is the scope for PMI_Abort?  Shouldn't it be one or more
        process groups?  Shouldn't abort of a communicator abort either the
        process groups of the communicator or only the current process?

diff --git a/src/mpid/ch4/src/ch4_globals.c b/src/mpid/ch4/src/ch4_globals.c
@@ -121,6 +121,11 @@ int MPID_Abort(MPIR_Comm * comm, int mpi_errno, int exit_code, const char *error
     MPIR_FUNC_VERBOSE_EXIT(MPID_STATE_MPID_ABORT);
     fflush(stderr);
     fflush(stdout);
+
+    if (MPIR_CVAR_COREDUMP_ON_ABORT) {
+        abort();
+    }
+
     if (NULL == comm || (MPIR_Comm_size(comm) == 1 && comm->comm_kind == MPIR_COMM_KIND__INTRACOMM))
         MPL_exit(exit_code);