--- old/src/hotspot/share/gc/shenandoah/shenandoahConcurrentMark.cpp	2018-06-19 21:10:10.294553141 +0200
+++ new/src/hotspot/share/gc/shenandoah/shenandoahConcurrentMark.cpp	2018-06-19 21:10:10.101555505 +0200
@@ -211,6 +211,29 @@
   }
 };
 
+class ShenandoahSATBThreadsClosure : public ThreadClosure {
+  ShenandoahSATBBufferClosure* _satb_cl;
+  int _thread_parity;
+
+ public:
+  ShenandoahSATBThreadsClosure(ShenandoahSATBBufferClosure* satb_cl) :
+    _satb_cl(satb_cl),
+    _thread_parity(Threads::thread_claim_parity()) {}
+
+  void do_thread(Thread* thread) {
+    if (thread->is_Java_thread()) {
+      if (thread->claim_oops_do(true, _thread_parity)) {
+        JavaThread* jt = (JavaThread*)thread;
+        ShenandoahThreadLocalData::satb_mark_queue(jt).apply_closure_and_empty(_satb_cl);
+      }
+    } else if (thread->is_VM_thread()) {
+      if (thread->claim_oops_do(true, _thread_parity)) {
+        ShenandoahBarrierSet::satb_mark_queue_set().shared_satb_queue()->apply_closure_and_empty(_satb_cl);
+      }
+    }
+  }
+};
+
 class ShenandoahFinalMarkingTask : public AbstractGangTask {
 private:
   ShenandoahConcurrentMark* _cm;
@@ -232,7 +255,12 @@
     // it requires a StrongRootsScope around the task, we need to claim the
     // threads, and performance-wise it doesn't really matter. Adds about 1ms to
     // full-gc.
-    _cm->drain_satb_buffers(worker_id);
+    ShenandoahObjToScanQueue* q = _cm->get_queue(worker_id);
+    ShenandoahSATBBufferClosure cl(q);
+    SATBMarkQueueSet& satb_mq_set = ShenandoahBarrierSet::satb_mark_queue_set();
+    while (satb_mq_set.apply_closure_to_completed_buffer(&cl));
+    ShenandoahSATBThreadsClosure tc(&cl);
+    Threads::threads_do(&tc);
 
     ReferenceProcessor* rp;
     if (_cm->process_references()) {
@@ -474,36 +502,6 @@
 
 }
 
-class ShenandoahSATBThreadsClosure : public ThreadClosure {
-  ShenandoahSATBBufferClosure* _satb_cl;
-  int _thread_parity;
-
- public:
-  ShenandoahSATBThreadsClosure(ShenandoahSATBBufferClosure* satb_cl) :
-    _satb_cl(satb_cl),
-    _thread_parity(Threads::thread_claim_parity()) {}
-
-  void do_thread(Thread* thread) {
-    if (thread->is_Java_thread()) {
-      if (thread->claim_oops_do(true, _thread_parity)) {
-        JavaThread* jt = (JavaThread*)thread;
-        ShenandoahThreadLocalData::satb_mark_queue(jt).apply_closure_and_empty(_satb_cl);
-      }
-    } else if (thread->is_VM_thread()) {
-      if (thread->claim_oops_do(true, _thread_parity)) {
-        ShenandoahBarrierSet::satb_mark_queue_set().shared_satb_queue()->apply_closure_and_empty(_satb_cl);
-      }
-    }
-  }
-};
-
-void ShenandoahConcurrentMark::drain_satb_buffers(uint worker_id) {
-  ShenandoahObjToScanQueue* q = get_queue(worker_id);
-  ShenandoahSATBBufferClosure cl(q);
-  ShenandoahSATBThreadsClosure tc(&cl);
-  Threads::threads_do(&tc);
-}
-
 #if TASKQUEUE_STATS
 void ShenandoahConcurrentMark::print_taskqueue_stats_hdr(outputStream* const st) {
   st->print_raw_cr("GC Task Stats");
--- old/src/hotspot/share/gc/shenandoah/shenandoahConcurrentMark.hpp	2018-06-19 21:10:10.871546073 +0200
+++ new/src/hotspot/share/gc/shenandoah/shenandoahConcurrentMark.hpp	2018-06-19 21:10:10.686548339 +0200
@@ -118,7 +118,6 @@
   ShenandoahObjToScanQueue* get_queue(uint worker_id);
   void clear_queue(ShenandoahObjToScanQueue *q);
 
-  void drain_satb_buffers(uint worker_id);
   ShenandoahObjToScanQueueSet* task_queues() { return _task_queues;}
 
   jushort* get_liveness(uint worker_id);
--- old/src/hotspot/share/gc/shenandoah/shenandoahTraversalGC.cpp	2018-06-19 21:10:11.436539153 +0200
+++ new/src/hotspot/share/gc/shenandoah/shenandoahTraversalGC.cpp	2018-06-19 21:10:11.242541529 +0200
@@ -243,8 +243,17 @@
       rp = _heap->ref_processor();
     }
 
-    // Step 1: Process ordinary GC roots.
+    // Step 0: Drain outstanding SATB queues.
+    // NOTE: we piggy-back draining of remaining thread SATB buffers on the final root scan below.
     ShenandoahTraversalSATBBufferClosure satb_cl(q);
+    {
+      // Process remaining finished SATB buffers.
+      SATBMarkQueueSet& satb_mq_set = ShenandoahBarrierSet::satb_mark_queue_set();
+      while (satb_mq_set.apply_closure_to_completed_buffer(&satb_cl));
+      // Process remaining threads SATB buffers below.
+    }
+
+    // Step 1: Process ordinary GC roots.
     if (!_heap->is_degenerated_gc_in_progress()) {
       ShenandoahTraversalClosure roots_cl(q, rp);
       CLDToOopClosure cld_cl(&roots_cl);