8253971: ZGC: Flush mark stacks after processing concurrent roots

Reviewed-by: ayang, eosterlund

src/hotspot/share/gc/z/zMark.cpp

@@ -643,13 +643,15 @@ class ZMarkConcurrentRootsIteratorClosure : public ZRootsIteratorClosure {
 
 class ZMarkConcurrentRootsTask : public ZTask {
 private:
+  ZMark* const                        _mark;
   SuspendibleThreadSetJoiner          _sts_joiner;
   ZConcurrentRootsIteratorClaimStrong _roots;
   ZMarkConcurrentRootsIteratorClosure _cl;
 
 public:
   ZMarkConcurrentRootsTask(ZMark* mark) :
       ZTask("ZMarkConcurrentRootsTask"),
+      _mark(mark),
       _sts_joiner(),
       _roots(),
       _cl() {
@@ -662,6 +664,12 @@ class ZMarkConcurrentRootsTask : public ZTask {
 
   virtual void work() {
     _roots.oops_do(&_cl);
+
+    // Flush and free worker stacks. Needed here since
+    // the set of workers executing during root scanning
+    // can be different from the set of workers executing
+    // during mark.
+    _mark->flush_and_free();
   }
 };