--- a/src/hotspot/share/gc/g1/g1ConcurrentRefine.cpp	Mon Jan 20 18:05:07 2020 +0100
+++ b/src/hotspot/share/gc/g1/g1ConcurrentRefine.cpp	Mon Jan 20 14:31:20 2020 -0500
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2019, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -190,6 +190,10 @@
 // For logging zone values, ensuring consistency of level and tags.
 #define LOG_ZONES(...) log_debug( CTRL_TAGS )(__VA_ARGS__)
 
+static size_t buffers_to_cards(size_t value) {
+  return value * G1UpdateBufferSize;
+}
+
 // Package for pair of refinement thread activation and deactivation
 // thresholds.  The activation and deactivation levels are resp. the first
 // and second values of the pair.
@@ -207,8 +211,9 @@
     // available buffers near green_zone value.  When yellow_size is
     // large we don't want to allow a full step to accumulate before
     // doing any processing, as that might lead to significantly more
-    // than green_zone buffers to be processed during scanning.
-    step = MIN2(step, ParallelGCThreads / 2.0);
+    // than green_zone buffers to be processed during pause.  So limit
+    // to an extra half buffer per pause-time processing thread.
+    step = MIN2(step, buffers_to_cards(ParallelGCThreads) / 2.0);
   }
   size_t activate_offset = static_cast<size_t>(ceil(step * (worker_id + 1)));
   size_t deactivate_offset = static_cast<size_t>(floor(step * worker_id));
@@ -233,10 +238,6 @@
   return _thread_control.initialize(this, max_num_threads());
 }
 
-static size_t buffers_to_cards(size_t value) {
-  return value * G1UpdateBufferSize;
-}
-
 static size_t calc_min_yellow_zone_size() {
   size_t step = buffers_to_cards(G1ConcRefinementThresholdStep);
   uint n_workers = G1ConcurrentRefine::max_num_threads();
@@ -443,8 +444,8 @@
   return G1DirtyCardQueueSet::num_par_ids();
 }
 
-void G1ConcurrentRefine::maybe_activate_more_threads(uint worker_id, size_t num_cur_buffers) {
-  if (num_cur_buffers > activation_threshold(worker_id + 1)) {
+void G1ConcurrentRefine::maybe_activate_more_threads(uint worker_id, size_t num_cur_cards) {
+  if (num_cur_cards > activation_threshold(worker_id + 1)) {
     _thread_control.maybe_activate_next(worker_id);
   }
 }