8252103: Parallel heap inspection for ParallelScavengeHeap

Reviewed-by: sjohanss, tschatzl

Author: Lin Zang

src/hotspot/share/gc/parallel/parallelScavengeHeap.cpp

@@ -539,6 +539,71 @@ void ParallelScavengeHeap::object_iterate(ObjectClosure* cl) {
   old_gen()->object_iterate(cl);
 }
 
+// The HeapBlockClaimer is used during parallel iteration over the heap,
+// allowing workers to claim heap areas ("blocks"), gaining exclusive rights to these.
+// The eden and survivor spaces are treated as single blocks as it is hard to divide
+// these spaces.
+// The old space is divided into fixed-size blocks.
+class HeapBlockClaimer : public StackObj {
+  size_t _claimed_index;
+
+public:
+  static const size_t InvalidIndex = SIZE_MAX;
+  static const size_t EdenIndex = 0;
+  static const size_t SurvivorIndex = 1;
+  static const size_t NumNonOldGenClaims = 2;
+
+  HeapBlockClaimer() : _claimed_index(EdenIndex) { }
+  // Claim the block and get the block index.
+  size_t claim_and_get_block() {
+    size_t block_index;
+    block_index = Atomic::fetch_and_add(&_claimed_index, 1u);
+
+    PSOldGen* old_gen = ParallelScavengeHeap::heap()->old_gen();
+    size_t num_claims = old_gen->num_iterable_blocks() + NumNonOldGenClaims;
+
+    return block_index < num_claims ? block_index : InvalidIndex;
+  }
+};
+
+void ParallelScavengeHeap::object_iterate_parallel(ObjectClosure* cl,
+                                                   HeapBlockClaimer* claimer) {
+  size_t block_index = claimer->claim_and_get_block();
+  // Iterate until all blocks are claimed
+  if (block_index == HeapBlockClaimer::EdenIndex) {
+    young_gen()->eden_space()->object_iterate(cl);
+    block_index = claimer->claim_and_get_block();
+  }
+  if (block_index == HeapBlockClaimer::SurvivorIndex) {
+    young_gen()->from_space()->object_iterate(cl);
+    young_gen()->to_space()->object_iterate(cl);
+    block_index = claimer->claim_and_get_block();
+  }
+  while (block_index != HeapBlockClaimer::InvalidIndex) {
+    old_gen()->object_iterate_block(cl, block_index - HeapBlockClaimer::NumNonOldGenClaims);
+    block_index = claimer->claim_and_get_block();
+  }
+}
+
+class PSScavengeParallelObjectIterator : public ParallelObjectIterator {
+private:
+  ParallelScavengeHeap*  _heap;
+  HeapBlockClaimer      _claimer;
+
+public:
+  PSScavengeParallelObjectIterator() :
+      _heap(ParallelScavengeHeap::heap()),
+      _claimer() {}
+
+  virtual void object_iterate(ObjectClosure* cl, uint worker_id) {
+    _heap->object_iterate_parallel(cl, &_claimer);
+  }
+};
+
+ParallelObjectIterator* ParallelScavengeHeap::parallel_object_iterator(uint thread_num) {
+  return new PSScavengeParallelObjectIterator();
+}
+
 HeapWord* ParallelScavengeHeap::block_start(const void* addr) const {
   if (young_gen()->is_in_reserved(addr)) {
     assert(young_gen()->is_in(addr),

src/hotspot/share/gc/parallel/parallelScavengeHeap.hpp

@@ -45,6 +45,7 @@
 
 class AdjoiningGenerations;
 class GCHeapSummary;
+class HeapBlockClaimer;
 class MemoryManager;
 class MemoryPool;
 class PSAdaptiveSizePolicy;
@@ -207,6 +208,8 @@ class ParallelScavengeHeap : public CollectedHeap {
   size_t unsafe_max_tlab_alloc(Thread* thr) const;
 
   void object_iterate(ObjectClosure* cl);
+  void object_iterate_parallel(ObjectClosure* cl, HeapBlockClaimer* claimer);
+  virtual ParallelObjectIterator* parallel_object_iterator(uint thread_num);
 
   HeapWord* block_start(const void* addr) const;
   bool block_is_obj(const HeapWord* addr) const;

src/hotspot/share/gc/parallel/psOldGen.cpp

@@ -173,6 +173,38 @@ HeapWord* PSOldGen::allocate(size_t word_size) {
   return res;
 }
 
+size_t PSOldGen::num_iterable_blocks() const {
+  return (object_space()->used_in_bytes() + IterateBlockSize - 1) / IterateBlockSize;
+}
+
+void PSOldGen::object_iterate_block(ObjectClosure* cl, size_t block_index) {
+  size_t block_word_size = IterateBlockSize / HeapWordSize;
+  assert((block_word_size % (ObjectStartArray::block_size)) == 0,
+         "Block size not a multiple of start_array block");
+
+  MutableSpace *space = object_space();
+
+  HeapWord* begin = space->bottom() + block_index * block_word_size;
+  HeapWord* end = MIN2(space->top(), begin + block_word_size);
+
+  if (!start_array()->object_starts_in_range(begin, end)) {
+    return;
+  }
+
+  // Get object starting at or reaching into this block.
+  HeapWord* start = start_array()->object_start(begin);
+  if (start < begin) {
+    start += oop(start)->size();
+  }
+  assert(start >= begin,
+         "Object address" PTR_FORMAT " must be larger or equal to block address at " PTR_FORMAT,
+         p2i(start), p2i(begin));
+  // Iterate all objects until the end.
+  for (HeapWord* p = start; p < end; p += oop(p)->size()) {
+    cl->do_object(oop(p));
+  }
+}
+
 HeapWord* PSOldGen::expand_and_allocate(size_t word_size) {
   expand(word_size*HeapWordSize);
   if (GCExpandToAllocateDelayMillis > 0) {

src/hotspot/share/gc/parallel/psOldGen.hpp

@@ -52,6 +52,9 @@ class PSOldGen : public CHeapObj<mtGC> {
   const size_t _min_gen_size;
   const size_t _max_gen_size;
 
+  // Block size for parallel iteration
+  static const size_t IterateBlockSize = 1024 * 1024;
+
 #ifdef ASSERT
   void assert_block_in_covered_region(MemRegion new_memregion) {
     // Explictly capture current covered_region in a local
@@ -163,6 +166,14 @@ class PSOldGen : public CHeapObj<mtGC> {
   void oop_iterate(OopIterateClosure* cl) { object_space()->oop_iterate(cl); }
   void object_iterate(ObjectClosure* cl) { object_space()->object_iterate(cl); }
 
+  // Number of blocks to be iterated over in the used part of old gen.
+  size_t num_iterable_blocks() const;
+  // Iterate the objects starting in block block_index within [bottom, top) of the
+  // old gen. The object just reaching into this block is not iterated over.
+  // A block is an evenly sized non-overlapping part of the old gen of
+  // IterateBlockSize bytes.
+  void object_iterate_block(ObjectClosure* cl, size_t block_index);
+
   // Debugging - do not use for time critical operations
   void print() const;
   virtual void print_on(outputStream* st) const;