diff --git a/gc.c b/gc.c
index a15b0f51..8be59a00 100644
--- a/gc.c
+++ b/gc.c
@@ -821,13 +821,12 @@ void gc_mut_update(gc_thread_data *thd, object old_obj, object value)
     gc_mark_gray(thd, old_obj);
     // Check if value is on the heap. If so, mark gray right now,
     // otherwise set it to be marked after moved to heap by next GC
-//    if (gc_is_stack_obj(thd, value)) {
-//      grayed(value) = 1;
-//    } else {
-//      gc_mark_gray(thd, value);
-//    }
+    if (gc_is_stack_obj(thd, value)) {
+      grayed(value) = 1;
+    } else {
+      gc_mark_gray(thd, value);
+    }
     pthread_mutex_unlock(&(thd->lock));
-gc_stack_mark_gray(thd, value); // TODO: this line will be replace with above block
   } else if (stage == STAGE_TRACING) {
 //fprintf(stderr, "DEBUG - GC async tracing marking heap obj %p ", old_obj);
 //Cyc_display(old_obj, stderr);
@@ -877,12 +876,20 @@ gc_stack_mark_gray(thd, value); // TODO: this line will be replace with above bl
 void gc_mut_cooperate(gc_thread_data *thd, int buf_len)
 {
   int i, status = ATOMIC_GET(&gc_status_col);
+
+  // Handle any pending marks from write barrier
+  pthread_mutex_lock(&(thd->lock));
+  thd->last_write += thd->pending_writes;
+  thd->pending_writes = 0;
+  pthread_mutex_unlock(&(thd->lock));
+
   if (thd->gc_status != status) {
     if (thd->gc_status == STATUS_ASYNC) {
       // Async is done, so clean up old mark data from the last collection
       pthread_mutex_lock(&(thd->lock));
       thd->last_write = 0;
       thd->last_read = 0;
+      thd->pending_writes = 0;
       pthread_mutex_unlock(&(thd->lock));
     }
     else if (thd->gc_status == STATUS_SYNC2) {
@@ -941,6 +948,17 @@ void gc_mark_gray(gc_thread_data *thd, object obj)
   }
 }
 
+void gc_mark_gray2(gc_thread_data *thd, object obj)
+{
+  if (is_object_type(obj) && mark(obj) == gc_color_clear) {
+    thd->mark_buffer = vpbuffer_add(thd->mark_buffer, 
+                                    &(thd->mark_buffer_len),
+                                    (thd->last_write + thd->pending_writes),
+                                    obj);
+    thd->pending_writes++;
+  }
+}
+
 // TODO: before doing this, may want to debug a bit and see what is
 // being passed to gc_mut_update. see if those objects tend to have
 // any heap refs. may need to add debug code to do that...
@@ -1003,7 +1021,10 @@ void gc_collector_trace()
       if (clean) {
         pthread_mutex_lock(&(m->lock));
         if (m->last_read < m->last_write) {
-          printf("JAE DEBUG - might have exited trace early\n");
+          fprintf(stderr, "JAE DEBUG - might have exited trace early\n");
+          clean = 0;
+        }
+        else if (m->pending_writes) {
           clean = 0;
         }
         pthread_mutex_unlock(&(m->lock));
diff --git a/include/cyclone/types.h b/include/cyclone/types.h
index 423aae83..ab73bed9 100644
--- a/include/cyclone/types.h
+++ b/include/cyclone/types.h
@@ -56,6 +56,7 @@ struct gc_thread_data_t {
   int gc_status;
   int last_write;
   int last_read;
+  int pending_writes;
   void **mark_buffer;
   int mark_buffer_len;
   pthread_mutex_t lock;
@@ -164,6 +165,7 @@ void gc_mut_cooperate(gc_thread_data *thd, int buf_len);
 void gc_stack_mark_refs_gray(gc_thread_data *thd, object obj, int depth);
 void gc_stack_mark_gray(gc_thread_data *thd, object obj);
 void gc_mark_gray(gc_thread_data *thd, object obj);
+void gc_mark_gray2(gc_thread_data *thd, object obj);
 void gc_collector_trace();
 void gc_mark_black(object obj);
 void gc_collector_mark_gray(object parent, object obj);
diff --git a/runtime.c b/runtime.c
index 93cff642..616eb527 100644
--- a/runtime.c
+++ b/runtime.c
@@ -2452,35 +2452,24 @@ void gc_mark_globals()
   }
 }
 
-char *gc_fixup_moved_obj(gc_thread_data *thd, int *alloci, int *num_grayed, char *obj, object hp)
+char *gc_fixup_moved_obj(gc_thread_data *thd, int *alloci, char *obj, object hp)
 {
+  if (grayed(obj)) {
+    pthread_mutex_lock(&(thd->lock));
+    gc_mark_gray2(thd, hp);
+    pthread_mutex_unlock(&(thd->lock));
+  }
+
   // hp ==> new heap object, point to it from old stack object
   forward(obj) = hp;
   type_of(obj) = forward_tag;
   // keep track of each allocation so we can scan/move 
   // the whole live object 'tree'
   gc_thr_add_to_move_buffer(thd, alloci, hp);
-
-// TODO: if grayed(obj) then - do a deferred gc_mark_gray on hp
-// you know, one possibility would be to add it to the thread mark buffer
-// but now increment the read (write?) variable. instead increment another
-// one and wait until the end of minor GC to set read (write?) to the
-// appropriate value. just need to make sure we have the lock before
-// updating it.
-// that should avoid race conditions and the need for an additional buffer
-
- if (TODO){
-   lock
-   gc_mark_gray - but WITHOUT incrementing write (read?)
-   unlock
-   (*num_grayed)++;
-}
-
-
   return (char *)hp;
 }
 
-char *gc_move(char *obj, gc_thread_data *thd, int *alloci, int *heap_grown, int *num_grayed) {
+char *gc_move(char *obj, gc_thread_data *thd, int *alloci, int *heap_grown) {
   if (!is_object_type(obj)) return obj;
   switch(type_of(obj)){
     case cons_tag: {
@@ -2562,7 +2551,7 @@ char *gc_move(char *obj, gc_thread_data *thd, int *alloci, int *heap_grown, int
   temp = obj; \
   if (check_overflow(low_limit, temp) && \
       check_overflow(temp, high_limit)){ \
-    (obj) = (object) gc_move(temp, (gc_thread_data *)data, &alloci, &heap_grown, &num_grayed); \
+    (obj) = (object) gc_move(temp, (gc_thread_data *)data, &alloci, &heap_grown); \
   } \
 }
 
@@ -2575,7 +2564,6 @@ void GC(void *data, closure cont, object *args, int num_args)
   int i;
   int scani = 0, alloci = 0;
   int heap_grown = 0;
-  int num_grayed = 0;
 
 //fprintf(stdout, "DEBUG, started minor GC\n"); // JAE DEBUG
   // Prevent overrunning buffer