view src/share/vm/gc_implementation/parallelScavenge/psPromotionManager.cpp @ 453:c96030fff130

6684579: SoftReference processing can be made more efficient Summary: For current soft-ref clearing policies, we can decide at marking time if a soft-reference will definitely not be cleared, postponing the decision of whether it will definitely be cleared to the final reference processing phase. This can be especially beneficial in the case of concurrent collectors where the marking is usually concurrent but reference processing is usually not. Reviewed-by: jmasa
author ysr
date Thu, 20 Nov 2008 16:56:09 -0800
parents d1605aabd0a1
children 7d7a7c599c17
line wrap: on
line source

/*
 * Copyright 2002-2008 Sun Microsystems, Inc.  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
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
 * CA 95054 USA or visit www.sun.com if you need additional information or
 * have any questions.
 *
 */

#include "incls/_precompiled.incl"
#include "incls/_psPromotionManager.cpp.incl"

PSPromotionManager**         PSPromotionManager::_manager_array = NULL;
OopStarTaskQueueSet*         PSPromotionManager::_stack_array_depth = NULL;
OopTaskQueueSet*             PSPromotionManager::_stack_array_breadth = NULL;
PSOldGen*                    PSPromotionManager::_old_gen = NULL;
MutableSpace*                PSPromotionManager::_young_space = NULL;

void PSPromotionManager::initialize() {
  ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
  assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");

  _old_gen = heap->old_gen();
  _young_space = heap->young_gen()->to_space();

  assert(_manager_array == NULL, "Attempt to initialize twice");
  _manager_array = NEW_C_HEAP_ARRAY(PSPromotionManager*, ParallelGCThreads+1 );
  guarantee(_manager_array != NULL, "Could not initialize promotion manager");

  if (UseDepthFirstScavengeOrder) {
    _stack_array_depth = new OopStarTaskQueueSet(ParallelGCThreads);
    guarantee(_stack_array_depth != NULL, "Count not initialize promotion manager");
  } else {
    _stack_array_breadth = new OopTaskQueueSet(ParallelGCThreads);
    guarantee(_stack_array_breadth != NULL, "Count not initialize promotion manager");
  }

  // Create and register the PSPromotionManager(s) for the worker threads.
  for(uint i=0; i<ParallelGCThreads; i++) {
    _manager_array[i] = new PSPromotionManager();
    guarantee(_manager_array[i] != NULL, "Could not create PSPromotionManager");
    if (UseDepthFirstScavengeOrder) {
      stack_array_depth()->register_queue(i, _manager_array[i]->claimed_stack_depth());
    } else {
      stack_array_breadth()->register_queue(i, _manager_array[i]->claimed_stack_breadth());
    }
  }

  // The VMThread gets its own PSPromotionManager, which is not available
  // for work stealing.
  _manager_array[ParallelGCThreads] = new PSPromotionManager();
  guarantee(_manager_array[ParallelGCThreads] != NULL, "Could not create PSPromotionManager");
}

PSPromotionManager* PSPromotionManager::gc_thread_promotion_manager(int index) {
  assert(index >= 0 && index < (int)ParallelGCThreads, "index out of range");
  assert(_manager_array != NULL, "Sanity");
  return _manager_array[index];
}

PSPromotionManager* PSPromotionManager::vm_thread_promotion_manager() {
  assert(_manager_array != NULL, "Sanity");
  return _manager_array[ParallelGCThreads];
}

void PSPromotionManager::pre_scavenge() {
  ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
  assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");

  _young_space = heap->young_gen()->to_space();

  for(uint i=0; i<ParallelGCThreads+1; i++) {
    manager_array(i)->reset();
  }
}

void PSPromotionManager::post_scavenge() {
#if PS_PM_STATS
  print_stats();
#endif // PS_PM_STATS

  for(uint i=0; i<ParallelGCThreads+1; i++) {
    PSPromotionManager* manager = manager_array(i);

    // the guarantees are a bit gratuitous but, if one fires, we'll
    // have a better idea of what went wrong
    if (i < ParallelGCThreads) {
      guarantee((!UseDepthFirstScavengeOrder ||
                 manager->overflow_stack_depth()->length() <= 0),
                "promotion manager overflow stack must be empty");
      guarantee((UseDepthFirstScavengeOrder ||
                 manager->overflow_stack_breadth()->length() <= 0),
                "promotion manager overflow stack must be empty");

      guarantee((!UseDepthFirstScavengeOrder ||
                 manager->claimed_stack_depth()->size() <= 0),
                "promotion manager claimed stack must be empty");
      guarantee((UseDepthFirstScavengeOrder ||
                 manager->claimed_stack_breadth()->size() <= 0),
                "promotion manager claimed stack must be empty");
    } else {
      guarantee((!UseDepthFirstScavengeOrder ||
                 manager->overflow_stack_depth()->length() <= 0),
                "VM Thread promotion manager overflow stack "
                "must be empty");
      guarantee((UseDepthFirstScavengeOrder ||
                 manager->overflow_stack_breadth()->length() <= 0),
                "VM Thread promotion manager overflow stack "
                "must be empty");

      guarantee((!UseDepthFirstScavengeOrder ||
                 manager->claimed_stack_depth()->size() <= 0),
                "VM Thread promotion manager claimed stack "
                "must be empty");
      guarantee((UseDepthFirstScavengeOrder ||
                 manager->claimed_stack_breadth()->size() <= 0),
                "VM Thread promotion manager claimed stack "
                "must be empty");
    }

    manager->flush_labs();
  }
}

#if PS_PM_STATS

void
PSPromotionManager::print_stats(uint i) {
  tty->print_cr("---- GC Worker %2d Stats", i);
  tty->print_cr("    total pushes            %8d", _total_pushes);
  tty->print_cr("    masked pushes           %8d", _masked_pushes);
  tty->print_cr("    overflow pushes         %8d", _overflow_pushes);
  tty->print_cr("    max overflow length     %8d", _max_overflow_length);
  tty->print_cr("");
  tty->print_cr("    arrays chunked          %8d", _arrays_chunked);
  tty->print_cr("    array chunks processed  %8d", _array_chunks_processed);
  tty->print_cr("");
  tty->print_cr("    total steals            %8d", _total_steals);
  tty->print_cr("    masked steals           %8d", _masked_steals);
  tty->print_cr("");
}

void
PSPromotionManager::print_stats() {
  tty->print_cr("== GC Tasks Stats (%s), GC %3d",
                (UseDepthFirstScavengeOrder) ? "Depth-First" : "Breadth-First",
                Universe::heap()->total_collections());

  for (uint i = 0; i < ParallelGCThreads+1; ++i) {
    PSPromotionManager* manager = manager_array(i);
    manager->print_stats(i);
  }
}

#endif // PS_PM_STATS

PSPromotionManager::PSPromotionManager() {
  ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
  assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");
  _depth_first = UseDepthFirstScavengeOrder;

  // We set the old lab's start array.
  _old_lab.set_start_array(old_gen()->start_array());

  uint queue_size;
  if (depth_first()) {
    claimed_stack_depth()->initialize();
    queue_size = claimed_stack_depth()->max_elems();
    // We want the overflow stack to be permanent
    _overflow_stack_depth = new (ResourceObj::C_HEAP) GrowableArray<StarTask>(10, true);
    _overflow_stack_breadth = NULL;
  } else {
    claimed_stack_breadth()->initialize();
    queue_size = claimed_stack_breadth()->max_elems();
    // We want the overflow stack to be permanent
    _overflow_stack_breadth = new (ResourceObj::C_HEAP) GrowableArray<oop>(10, true);
    _overflow_stack_depth = NULL;
  }

  _totally_drain = (ParallelGCThreads == 1) || (GCDrainStackTargetSize == 0);
  if (_totally_drain) {
    _target_stack_size = 0;
  } else {
    // don't let the target stack size to be more than 1/4 of the entries
    _target_stack_size = (uint) MIN2((uint) GCDrainStackTargetSize,
                                     (uint) (queue_size / 4));
  }

  _array_chunk_size = ParGCArrayScanChunk;
  // let's choose 1.5x the chunk size
  _min_array_size_for_chunking = 3 * _array_chunk_size / 2;

  reset();
}

void PSPromotionManager::reset() {
  assert(claimed_stack_empty(), "reset of non-empty claimed stack");
  assert(overflow_stack_empty(), "reset of non-empty overflow stack");

  // We need to get an assert in here to make sure the labs are always flushed.

  ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
  assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");

  // Do not prefill the LAB's, save heap wastage!
  HeapWord* lab_base = young_space()->top();
  _young_lab.initialize(MemRegion(lab_base, (size_t)0));
  _young_gen_is_full = false;

  lab_base = old_gen()->object_space()->top();
  _old_lab.initialize(MemRegion(lab_base, (size_t)0));
  _old_gen_is_full = false;

  _prefetch_queue.clear();

#if PS_PM_STATS
  _total_pushes = 0;
  _masked_pushes = 0;
  _overflow_pushes = 0;
  _max_overflow_length = 0;
  _arrays_chunked = 0;
  _array_chunks_processed = 0;
  _total_steals = 0;
  _masked_steals = 0;
#endif // PS_PM_STATS
}


void PSPromotionManager::drain_stacks_depth(bool totally_drain) {
  assert(depth_first(), "invariant");
  assert(overflow_stack_depth() != NULL, "invariant");
  totally_drain = totally_drain || _totally_drain;

#ifdef ASSERT
  ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
  assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");
  MutableSpace* to_space = heap->young_gen()->to_space();
  MutableSpace* old_space = heap->old_gen()->object_space();
  MutableSpace* perm_space = heap->perm_gen()->object_space();
#endif /* ASSERT */

  do {
    StarTask p;

    // Drain overflow stack first, so other threads can steal from
    // claimed stack while we work.
    while(!overflow_stack_depth()->is_empty()) {
      // linux compiler wants different overloaded operator= in taskqueue to
      // assign to p that the other compilers don't like.
      StarTask ptr = overflow_stack_depth()->pop();
      process_popped_location_depth(ptr);
    }

    if (totally_drain) {
      while (claimed_stack_depth()->pop_local(p)) {
        process_popped_location_depth(p);
      }
    } else {
      while (claimed_stack_depth()->size() > _target_stack_size &&
             claimed_stack_depth()->pop_local(p)) {
        process_popped_location_depth(p);
      }
    }
  } while( (totally_drain && claimed_stack_depth()->size() > 0) ||
           (overflow_stack_depth()->length() > 0) );

  assert(!totally_drain || claimed_stack_empty(), "Sanity");
  assert(totally_drain ||
         claimed_stack_depth()->size() <= _target_stack_size,
         "Sanity");
  assert(overflow_stack_empty(), "Sanity");
}

void PSPromotionManager::drain_stacks_breadth(bool totally_drain) {
  assert(!depth_first(), "invariant");
  assert(overflow_stack_breadth() != NULL, "invariant");
  totally_drain = totally_drain || _totally_drain;

#ifdef ASSERT
  ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
  assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");
  MutableSpace* to_space = heap->young_gen()->to_space();
  MutableSpace* old_space = heap->old_gen()->object_space();
  MutableSpace* perm_space = heap->perm_gen()->object_space();
#endif /* ASSERT */

  do {
    oop obj;

    // Drain overflow stack first, so other threads can steal from
    // claimed stack while we work.
    while(!overflow_stack_breadth()->is_empty()) {
      obj = overflow_stack_breadth()->pop();
      obj->copy_contents(this);
    }

    if (totally_drain) {
      // obj is a reference!!!
      while (claimed_stack_breadth()->pop_local(obj)) {
        // It would be nice to assert about the type of objects we might
        // pop, but they can come from anywhere, unfortunately.
        obj->copy_contents(this);
      }
    } else {
      // obj is a reference!!!
      while (claimed_stack_breadth()->size() > _target_stack_size &&
             claimed_stack_breadth()->pop_local(obj)) {
        // It would be nice to assert about the type of objects we might
        // pop, but they can come from anywhere, unfortunately.
        obj->copy_contents(this);
      }
    }

    // If we could not find any other work, flush the prefetch queue
    if (claimed_stack_breadth()->size() == 0 &&
        (overflow_stack_breadth()->length() == 0)) {
      flush_prefetch_queue();
    }
  } while((totally_drain && claimed_stack_breadth()->size() > 0) ||
          (overflow_stack_breadth()->length() > 0));

  assert(!totally_drain || claimed_stack_empty(), "Sanity");
  assert(totally_drain ||
         claimed_stack_breadth()->size() <= _target_stack_size,
         "Sanity");
  assert(overflow_stack_empty(), "Sanity");
}

void PSPromotionManager::flush_labs() {
  assert(claimed_stack_empty(), "Attempt to flush lab with live stack");
  assert(overflow_stack_empty(), "Attempt to flush lab with live overflow stack");

  // If either promotion lab fills up, we can flush the
  // lab but not refill it, so check first.
  assert(!_young_lab.is_flushed() || _young_gen_is_full, "Sanity");
  if (!_young_lab.is_flushed())
    _young_lab.flush();

  assert(!_old_lab.is_flushed() || _old_gen_is_full, "Sanity");
  if (!_old_lab.is_flushed())
    _old_lab.flush();

  // Let PSScavenge know if we overflowed
  if (_young_gen_is_full) {
    PSScavenge::set_survivor_overflow(true);
  }
}

//
// This method is pretty bulky. It would be nice to split it up
// into smaller submethods, but we need to be careful not to hurt
// performance.
//

oop PSPromotionManager::copy_to_survivor_space(oop o, bool depth_first) {
  assert(PSScavenge::should_scavenge(&o), "Sanity");

  oop new_obj = NULL;

  // NOTE! We must be very careful with any methods that access the mark
  // in o. There may be multiple threads racing on it, and it may be forwarded
  // at any time. Do not use oop methods for accessing the mark!
  markOop test_mark = o->mark();

  // The same test as "o->is_forwarded()"
  if (!test_mark->is_marked()) {
    bool new_obj_is_tenured = false;
    size_t new_obj_size = o->size();

    // Find the objects age, MT safe.
    int age = (test_mark->has_displaced_mark_helper() /* o->has_displaced_mark() */) ?
      test_mark->displaced_mark_helper()->age() : test_mark->age();

    // Try allocating obj in to-space (unless too old)
    if (age < PSScavenge::tenuring_threshold()) {
      new_obj = (oop) _young_lab.allocate(new_obj_size);
      if (new_obj == NULL && !_young_gen_is_full) {
        // Do we allocate directly, or flush and refill?
        if (new_obj_size > (YoungPLABSize / 2)) {
          // Allocate this object directly
          new_obj = (oop)young_space()->cas_allocate(new_obj_size);
        } else {
          // Flush and fill
          _young_lab.flush();

          HeapWord* lab_base = young_space()->cas_allocate(YoungPLABSize);
          if (lab_base != NULL) {
            _young_lab.initialize(MemRegion(lab_base, YoungPLABSize));
            // Try the young lab allocation again.
            new_obj = (oop) _young_lab.allocate(new_obj_size);
          } else {
            _young_gen_is_full = true;
          }
        }
      }
    }

    // Otherwise try allocating obj tenured
    if (new_obj == NULL) {
#ifndef PRODUCT
      if (Universe::heap()->promotion_should_fail()) {
        return oop_promotion_failed(o, test_mark);
      }
#endif  // #ifndef PRODUCT

      new_obj = (oop) _old_lab.allocate(new_obj_size);
      new_obj_is_tenured = true;

      if (new_obj == NULL) {
        if (!_old_gen_is_full) {
          // Do we allocate directly, or flush and refill?
          if (new_obj_size > (OldPLABSize / 2)) {
            // Allocate this object directly
            new_obj = (oop)old_gen()->cas_allocate(new_obj_size);
          } else {
            // Flush and fill
            _old_lab.flush();

            HeapWord* lab_base = old_gen()->cas_allocate(OldPLABSize);
            if(lab_base != NULL) {
              _old_lab.initialize(MemRegion(lab_base, OldPLABSize));
              // Try the old lab allocation again.
              new_obj = (oop) _old_lab.allocate(new_obj_size);
            }
          }
        }

        // This is the promotion failed test, and code handling.
        // The code belongs here for two reasons. It is slightly
        // different thatn the code below, and cannot share the
        // CAS testing code. Keeping the code here also minimizes
        // the impact on the common case fast path code.

        if (new_obj == NULL) {
          _old_gen_is_full = true;
          return oop_promotion_failed(o, test_mark);
        }
      }
    }

    assert(new_obj != NULL, "allocation should have succeeded");

    // Copy obj
    Copy::aligned_disjoint_words((HeapWord*)o, (HeapWord*)new_obj, new_obj_size);

    // Now we have to CAS in the header.
    if (o->cas_forward_to(new_obj, test_mark)) {
      // We won any races, we "own" this object.
      assert(new_obj == o->forwardee(), "Sanity");

      // Increment age if obj still in new generation. Now that
      // we're dealing with a markOop that cannot change, it is
      // okay to use the non mt safe oop methods.
      if (!new_obj_is_tenured) {
        new_obj->incr_age();
        assert(young_space()->contains(new_obj), "Attempt to push non-promoted obj");
      }

      if (depth_first) {
        // Do the size comparison first with new_obj_size, which we
        // already have. Hopefully, only a few objects are larger than
        // _min_array_size_for_chunking, and most of them will be arrays.
        // So, the is->objArray() test would be very infrequent.
        if (new_obj_size > _min_array_size_for_chunking &&
            new_obj->is_objArray() &&
            PSChunkLargeArrays) {
          // we'll chunk it
#if PS_PM_STATS
          ++_arrays_chunked;
#endif // PS_PM_STATS
          oop* const masked_o = mask_chunked_array_oop(o);
          push_depth(masked_o);
#if PS_PM_STATS
          ++_masked_pushes;
#endif // PS_PM_STATS
        } else {
          // we'll just push its contents
          new_obj->push_contents(this);
        }
      } else {
        push_breadth(new_obj);
      }
    }  else {
      // We lost, someone else "owns" this object
      guarantee(o->is_forwarded(), "Object must be forwarded if the cas failed.");

      // Unallocate the space used. NOTE! We may have directly allocated
      // the object. If so, we cannot deallocate it, so we have to test!
      if (new_obj_is_tenured) {
        if (!_old_lab.unallocate_object(new_obj)) {
          // The promotion lab failed to unallocate the object.
          // We need to overwrite the object with a filler that
          // contains no interior pointers.
          MemRegion mr((HeapWord*)new_obj, new_obj_size);
          // Clean this up and move to oopFactory (see bug 4718422)
          SharedHeap::fill_region_with_object(mr);
        }
      } else {
        if (!_young_lab.unallocate_object(new_obj)) {
          // The promotion lab failed to unallocate the object.
          // We need to overwrite the object with a filler that
          // contains no interior pointers.
          MemRegion mr((HeapWord*)new_obj, new_obj_size);
          // Clean this up and move to oopFactory (see bug 4718422)
          SharedHeap::fill_region_with_object(mr);
        }
      }

      // don't update this before the unallocation!
      new_obj = o->forwardee();
    }
  } else {
    assert(o->is_forwarded(), "Sanity");
    new_obj = o->forwardee();
  }

#ifdef DEBUG
  // This code must come after the CAS test, or it will print incorrect
  // information.
  if (TraceScavenge) {
    gclog_or_tty->print_cr("{%s %s " PTR_FORMAT " -> " PTR_FORMAT " (" SIZE_FORMAT ")}",
       PSScavenge::should_scavenge(&new_obj) ? "copying" : "tenuring",
       new_obj->blueprint()->internal_name(), o, new_obj, new_obj->size());
  }
#endif

  return new_obj;
}

template <class T> void PSPromotionManager::process_array_chunk_work(
                                                 oop obj,
                                                 int start, int end) {
  assert(start < end, "invariant");
  T* const base      = (T*)objArrayOop(obj)->base();
  T* p               = base + start;
  T* const chunk_end = base + end;
  while (p < chunk_end) {
    if (PSScavenge::should_scavenge(p)) {
      claim_or_forward_depth(p);
    }
    ++p;
  }
}

void PSPromotionManager::process_array_chunk(oop old) {
  assert(PSChunkLargeArrays, "invariant");
  assert(old->is_objArray(), "invariant");
  assert(old->is_forwarded(), "invariant");

#if PS_PM_STATS
  ++_array_chunks_processed;
#endif // PS_PM_STATS

  oop const obj = old->forwardee();

  int start;
  int const end = arrayOop(old)->length();
  if (end > (int) _min_array_size_for_chunking) {
    // we'll chunk more
    start = end - _array_chunk_size;
    assert(start > 0, "invariant");
    arrayOop(old)->set_length(start);
    push_depth(mask_chunked_array_oop(old));
#if PS_PM_STATS
    ++_masked_pushes;
#endif // PS_PM_STATS
  } else {
    // this is the final chunk for this array
    start = 0;
    int const actual_length = arrayOop(obj)->length();
    arrayOop(old)->set_length(actual_length);
  }

  if (UseCompressedOops) {
    process_array_chunk_work<narrowOop>(obj, start, end);
  } else {
    process_array_chunk_work<oop>(obj, start, end);
  }
}

oop PSPromotionManager::oop_promotion_failed(oop obj, markOop obj_mark) {
  assert(_old_gen_is_full || PromotionFailureALot, "Sanity");

  // Attempt to CAS in the header.
  // This tests if the header is still the same as when
  // this started.  If it is the same (i.e., no forwarding
  // pointer has been installed), then this thread owns
  // it.
  if (obj->cas_forward_to(obj, obj_mark)) {
    // We won any races, we "own" this object.
    assert(obj == obj->forwardee(), "Sanity");

    if (depth_first()) {
      obj->push_contents(this);
    } else {
      // Don't bother incrementing the age, just push
      // onto the claimed_stack..
      push_breadth(obj);
    }

    // Save the mark if needed
    PSScavenge::oop_promotion_failed(obj, obj_mark);
  }  else {
    // We lost, someone else "owns" this object
    guarantee(obj->is_forwarded(), "Object must be forwarded if the cas failed.");

    // No unallocation to worry about.
    obj = obj->forwardee();
  }

#ifdef DEBUG
  if (TraceScavenge) {
    gclog_or_tty->print_cr("{%s %s 0x%x (%d)}",
                           "promotion-failure",
                           obj->blueprint()->internal_name(),
                           obj, obj->size());

  }
#endif

  return obj;
}