8055479: TLAB stability

Reviewed-by: brutisso, stefank, ahgross

src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp

@@ -737,7 +737,7 @@ CMSCollector::CMSCollector(ConcurrentMarkSweepGeneration* cmsGen,
   // Support for parallelizing survivor space rescan
   if ((CMSParallelRemarkEnabled && CMSParallelSurvivorRemarkEnabled) || CMSParallelInitialMarkEnabled) {
     const size_t max_plab_samples =
-      ((DefNewGeneration*)_young_gen)->max_survivor_size()/MinTLABSize;
+      ((DefNewGeneration*)_young_gen)->max_survivor_size() / plab_sample_minimum_size();
 
     _survivor_plab_array  = NEW_C_HEAP_ARRAY(ChunkArray, ParallelGCThreads, mtGC);
     _survivor_chunk_array = NEW_C_HEAP_ARRAY(HeapWord*, 2*max_plab_samples, mtGC);
@@ -795,6 +795,12 @@ CMSCollector::CMSCollector(ConcurrentMarkSweepGeneration* cmsGen,
   _inter_sweep_timer.start();  // start of time
 }
 
+size_t CMSCollector::plab_sample_minimum_size() {
+  // The default value of MinTLABSize is 2k, but there is
+  // no way to get the default value if the flag has been overridden.
+  return MAX2(ThreadLocalAllocBuffer::min_size() * HeapWordSize, 2 * K);
+}
+
 const char* ConcurrentMarkSweepGeneration::name() const {
   return "concurrent mark-sweep generation";
 }

src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp

@@ -763,6 +763,10 @@ class CMSCollector: public CHeapObj<mtGC> {
   size_t*    _cursor;
   ChunkArray* _survivor_plab_array;
 
+  // A bounded minimum size of PLABs, should not return too small values since
+  // this will affect the size of the data structures used for parallel young gen rescan
+  size_t plab_sample_minimum_size();
+
   // Support for marking stack overflow handling
   bool take_from_overflow_list(size_t num, CMSMarkStack* to_stack);
   bool par_take_from_overflow_list(size_t num,

src/share/vm/gc_implementation/shared/parGCAllocBuffer.hpp

@@ -62,7 +62,8 @@ public:
   ParGCAllocBuffer(size_t word_sz);
 
   static const size_t min_size() {
-    return ThreadLocalAllocBuffer::min_size();
+    // Make sure that we return something that is larger than AlignmentReserve
+    return align_object_size(MAX2(MinTLABSize / HeapWordSize, (uintx)oopDesc::header_size())) + AlignmentReserve;
   }
 
   static const size_t max_size() {

src/share/vm/memory/threadLocalAllocBuffer.cpp

@@ -235,22 +235,19 @@ void ThreadLocalAllocBuffer::startup_initialization() {
 }
 
 size_t ThreadLocalAllocBuffer::initial_desired_size() {
-  size_t init_sz;
+  size_t init_sz = 0;
 
   if (TLABSize > 0) {
-    init_sz = MIN2(TLABSize / HeapWordSize, max_size());
-  } else if (global_stats() == NULL) {
-    // Startup issue - main thread initialized before heap initialized.
-    init_sz = min_size();
-  } else {
+    init_sz = TLABSize / HeapWordSize;
+  } else if (global_stats() != NULL) {
     // Initial size is a function of the average number of allocating threads.
     unsigned nof_threads = global_stats()->allocating_threads_avg();
 
     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
                       (nof_threads * target_refills());
     init_sz = align_object_size(init_sz);
-    init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
   }
+  init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
   return init_sz;
 }
 

src/share/vm/memory/threadLocalAllocBuffer.hpp

@@ -105,7 +105,7 @@ public:
     // do nothing.  tlabs must be inited by initialize() calls
   }
 
-  static const size_t min_size()                 { return align_object_size(MinTLABSize / HeapWordSize); }
+  static const size_t min_size()                 { return align_object_size(MinTLABSize / HeapWordSize) + alignment_reserve(); }
   static const size_t max_size()                 { assert(_max_size != 0, "max_size not set up"); return _max_size; }
   static void set_max_size(size_t max_size)      { _max_size = max_size; }