diff --git a/native/gc.c b/native/gc.c
index f7c079c703..c166cd934a 100644
--- a/native/gc.c
+++ b/native/gc.c
@@ -10,6 +10,14 @@ CELL init_zone(ZONE *z, CELL size, CELL base)
 	return z->limit;
 }
 
+/* update this global variable. since it is stored in a non-volatile register,
+we need to save its contents and re-initialize it when entering a callback,
+and restore its contents when leaving the callback. see stack.c */
+void update_cards_offset(void)
+{
+	cards_offset = (CELL)cards - (heap_start >> CARD_BITS);
+}
+
 /* input parameters must be 8 byte aligned */
 /* the heap layout is important:
 - two semispaces: tenured and prior
@@ -35,7 +43,7 @@ void init_arena(CELL gens, CELL young_size, CELL aging_size)
 
 	cards = safe_malloc(cards_size);
 	cards_end = cards + cards_size;
-	cards_offset = (CELL)cards - (heap_start >> CARD_BITS);
+	update_cards_offset();
 
 	alloter = heap_start;
 
diff --git a/native/gc.h b/native/gc.h
index a7295fc458..07482fcbdb 100644
--- a/native/gc.h
+++ b/native/gc.h
@@ -118,6 +118,7 @@ INLINE void* allot_object(CELL type, CELL length)
 	return object;
 }
 
+void update_cards_offset(void);
 CELL collect_next(CELL scan);
 void garbage_collection(CELL gen);
 void primitive_gc(void);
diff --git a/native/stack.c b/native/stack.c
index c453fa532e..ca3f925cad 100644
--- a/native/stack.c
+++ b/native/stack.c
@@ -56,9 +56,11 @@ void nest_stacks(void)
 	new_stacks->call_region = alloc_bounded_block(cs_size);
 	new_stacks->next = stack_chain;
 	stack_chain = new_stacks;
+
 	callframe = F;
 	reset_datastack();
 	reset_callstack();
+	update_cards_offset();
 }
 
 /* called when leaving a compiled callback */