diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index ece66c261e7..0c87bfff3c5 100644
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,3 +1,8 @@
+2014-05-20  Doug Evans  <dje@google.com>
+
+	* python/py-progspace.c (py_free_pspace): Call target_gdbarch
+	instead of get_current_arch.
+
 2014-05-20  Pedro Alves  <palves@redhat.com>
 
 	* NEWS: Mention that compare-sections now works with all targets.
diff --git a/gdb/python/py-progspace.c b/gdb/python/py-progspace.c
index cda5a86ff91..db4c564ac22 100644
--- a/gdb/python/py-progspace.c
+++ b/gdb/python/py-progspace.c
@@ -241,7 +241,16 @@ py_free_pspace (struct program_space *pspace, void *datum)
 {
   struct cleanup *cleanup;
   pspace_object *object = datum;
-  struct gdbarch *arch = get_current_arch ();
+  /* This is a fiction, but we're in a nasty spot: The pspace is in the
+     process of being deleted, we can't rely on anything in it.  Plus
+     this is one time when the current program space and current inferior
+     are not in sync: All inferiors that use PSPACE may no longer exist.
+     We don't need to do much here, and since "there is always an inferior"
+     using target_gdbarch suffices.
+     Note: We cannot call get_current_arch because it may try to access
+     the target, which may involve accessing data in the pspace currently
+     being deleted.  */
+  struct gdbarch *arch = target_gdbarch ();
 
   cleanup = ensure_python_env (arch, current_language);
   object->pspace = NULL;