duke@435: /*
xdono@1014:  * Copyright 2001-2009 Sun Microsystems, Inc.  All Rights Reserved.
duke@435:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435:  *
duke@435:  * This code is free software; you can redistribute it and/or modify it
duke@435:  * under the terms of the GNU General Public License version 2 only, as
duke@435:  * published by the Free Software Foundation.
duke@435:  *
duke@435:  * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
duke@435:  * version 2 for more details (a copy is included in the LICENSE file that
duke@435:  * accompanied this code).
duke@435:  *
duke@435:  * You should have received a copy of the GNU General Public License version
duke@435:  * 2 along with this work; if not, write to the Free Software Foundation,
duke@435:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435:  *
duke@435:  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@435:  * CA 95054 USA or visit www.sun.com if you need additional information or
duke@435:  * have any questions.
duke@435:  *
duke@435:  */
duke@435: 
duke@435: class ChunkArray;
duke@435: class ParScanWithoutBarrierClosure;
duke@435: class ParScanWithBarrierClosure;
duke@435: class ParRootScanWithoutBarrierClosure;
duke@435: class ParRootScanWithBarrierTwoGensClosure;
duke@435: class ParEvacuateFollowersClosure;
duke@435: 
duke@435: // It would be better if these types could be kept local to the .cpp file,
duke@435: // but they must be here to allow ParScanClosure::do_oop_work to be defined
duke@435: // in genOopClosures.inline.hpp.
duke@435: 
duke@435: typedef OopTaskQueue    ObjToScanQueue;
duke@435: typedef OopTaskQueueSet ObjToScanQueueSet;
duke@435: 
duke@435: // Enable this to get push/pop/steal stats.
duke@435: const int PAR_STATS_ENABLED = 0;
duke@435: 
duke@435: class ParKeepAliveClosure: public DefNewGeneration::KeepAliveClosure {
coleenp@548:  private:
duke@435:   ParScanWeakRefClosure* _par_cl;
coleenp@548:  protected:
coleenp@548:   template <class T> void do_oop_work(T* p);
duke@435:  public:
duke@435:   ParKeepAliveClosure(ParScanWeakRefClosure* cl);
coleenp@548:   virtual void do_oop(oop* p);
coleenp@548:   virtual void do_oop(narrowOop* p);
duke@435: };
duke@435: 
duke@435: // The state needed by thread performing parallel young-gen collection.
duke@435: class ParScanThreadState {
duke@435:   friend class ParScanThreadStateSet;
coleenp@548:  private:
duke@435:   ObjToScanQueue *_work_queue;
ysr@1114:   GrowableArray<oop>* _overflow_stack;
duke@435: 
duke@435:   ParGCAllocBuffer _to_space_alloc_buffer;
duke@435: 
duke@435:   ParScanWithoutBarrierClosure         _to_space_closure; // scan_without_gc_barrier
duke@435:   ParScanWithBarrierClosure            _old_gen_closure; // scan_with_gc_barrier
duke@435:   ParRootScanWithoutBarrierClosure     _to_space_root_closure; // scan_root_without_gc_barrier
duke@435:   // One of these two will be passed to process_strong_roots, which will
duke@435:   // set its generation.  The first is for two-gen configs where the
duke@435:   // old gen collects the perm gen; the second is for arbitrary configs.
duke@435:   // The second isn't used right now (it used to be used for the train, an
duke@435:   // incremental collector) but the declaration has been left as a reminder.
duke@435:   ParRootScanWithBarrierTwoGensClosure _older_gen_closure;
duke@435:   // This closure will always be bound to the old gen; it will be used
duke@435:   // in evacuate_followers.
duke@435:   ParRootScanWithBarrierTwoGensClosure _old_gen_root_closure; // scan_old_root_with_gc_barrier
duke@435:   ParEvacuateFollowersClosure          _evacuate_followers;
duke@435:   DefNewGeneration::IsAliveClosure     _is_alive_closure;
duke@435:   ParScanWeakRefClosure                _scan_weak_ref_closure;
duke@435:   ParKeepAliveClosure                  _keep_alive_closure;
duke@435: 
duke@435: 
duke@435:   Space* _to_space;
duke@435:   Space* to_space() { return _to_space; }
duke@435: 
ysr@1114:   ParNewGeneration* _young_gen;
ysr@1114:   ParNewGeneration* young_gen() const { return _young_gen; }
ysr@1114: 
duke@435:   Generation* _old_gen;
duke@435:   Generation* old_gen() { return _old_gen; }
duke@435: 
duke@435:   HeapWord *_young_old_boundary;
duke@435: 
duke@435:   int _hash_seed;
duke@435:   int _thread_num;
duke@435:   ageTable _ageTable;
duke@435: 
duke@435:   bool _to_space_full;
duke@435: 
duke@435:   int _pushes, _pops, _steals, _steal_attempts, _term_attempts;
duke@435:   int _overflow_pushes, _overflow_refills, _overflow_refill_objs;
duke@435: 
duke@435:   // Timing numbers.
duke@435:   double _start;
duke@435:   double _start_strong_roots;
duke@435:   double _strong_roots_time;
duke@435:   double _start_term;
duke@435:   double _term_time;
duke@435: 
duke@435:   // Helper for trim_queues. Scans subset of an array and makes
duke@435:   // remainder available for work stealing.
duke@435:   void scan_partial_array_and_push_remainder(oop obj);
duke@435: 
duke@435:   // In support of CMS' parallel rescan of survivor space.
duke@435:   ChunkArray* _survivor_chunk_array;
duke@435:   ChunkArray* survivor_chunk_array() { return _survivor_chunk_array; }
duke@435: 
duke@435:   void record_survivor_plab(HeapWord* plab_start, size_t plab_word_size);
duke@435: 
duke@435:   ParScanThreadState(Space* to_space_, ParNewGeneration* gen_,
duke@435:                      Generation* old_gen_, int thread_num_,
duke@435:                      ObjToScanQueueSet* work_queue_set_, size_t desired_plab_sz_,
duke@435:                      ParallelTaskTerminator& term_);
duke@435: 
coleenp@548:  public:
duke@435:   ageTable* age_table() {return &_ageTable;}
duke@435: 
duke@435:   ObjToScanQueue* work_queue() { return _work_queue; }
duke@435: 
duke@435:   ParGCAllocBuffer* to_space_alloc_buffer() {
duke@435:     return &_to_space_alloc_buffer;
duke@435:   }
duke@435: 
duke@435:   ParEvacuateFollowersClosure&      evacuate_followers_closure() { return _evacuate_followers; }
duke@435:   DefNewGeneration::IsAliveClosure& is_alive_closure() { return _is_alive_closure; }
duke@435:   ParScanWeakRefClosure&            scan_weak_ref_closure() { return _scan_weak_ref_closure; }
duke@435:   ParKeepAliveClosure&              keep_alive_closure() { return _keep_alive_closure; }
duke@435:   ParScanClosure&                   older_gen_closure() { return _older_gen_closure; }
duke@435:   ParRootScanWithoutBarrierClosure& to_space_root_closure() { return _to_space_root_closure; };
duke@435: 
duke@435:   // Decrease queue size below "max_size".
duke@435:   void trim_queues(int max_size);
duke@435: 
ysr@1114:   // Private overflow stack usage
ysr@1114:   GrowableArray<oop>* overflow_stack() { return _overflow_stack; }
ysr@1114:   bool take_from_overflow_stack();
ysr@1114:   void push_on_overflow_stack(oop p);
ysr@1114: 
duke@435:   // Is new_obj a candidate for scan_partial_array_and_push_remainder method.
duke@435:   inline bool should_be_partially_scanned(oop new_obj, oop old_obj) const;
duke@435: 
duke@435:   int* hash_seed()  { return &_hash_seed; }
duke@435:   int  thread_num() { return _thread_num; }
duke@435: 
duke@435:   // Allocate a to-space block of size "sz", or else return NULL.
duke@435:   HeapWord* alloc_in_to_space_slow(size_t word_sz);
duke@435: 
duke@435:   HeapWord* alloc_in_to_space(size_t word_sz) {
duke@435:     HeapWord* obj = to_space_alloc_buffer()->allocate(word_sz);
duke@435:     if (obj != NULL) return obj;
duke@435:     else return alloc_in_to_space_slow(word_sz);
duke@435:   }
duke@435: 
duke@435:   HeapWord* young_old_boundary() { return _young_old_boundary; }
duke@435: 
duke@435:   void set_young_old_boundary(HeapWord *boundary) {
duke@435:     _young_old_boundary = boundary;
duke@435:   }
duke@435: 
duke@435:   // Undo the most recent allocation ("obj", of "word_sz").
duke@435:   void undo_alloc_in_to_space(HeapWord* obj, size_t word_sz);
duke@435: 
duke@435:   int pushes() { return _pushes; }
duke@435:   int pops()   { return _pops; }
duke@435:   int steals() { return _steals; }
duke@435:   int steal_attempts() { return _steal_attempts; }
duke@435:   int term_attempts()  { return _term_attempts; }
duke@435:   int overflow_pushes() { return _overflow_pushes; }
duke@435:   int overflow_refills() { return _overflow_refills; }
duke@435:   int overflow_refill_objs() { return _overflow_refill_objs; }
duke@435: 
duke@435:   void note_push()  { if (PAR_STATS_ENABLED) _pushes++; }
duke@435:   void note_pop()   { if (PAR_STATS_ENABLED) _pops++; }
duke@435:   void note_steal() { if (PAR_STATS_ENABLED) _steals++; }
duke@435:   void note_steal_attempt() { if (PAR_STATS_ENABLED) _steal_attempts++; }
duke@435:   void note_term_attempt()  { if (PAR_STATS_ENABLED) _term_attempts++; }
duke@435:   void note_overflow_push() { if (PAR_STATS_ENABLED) _overflow_pushes++; }
duke@435:   void note_overflow_refill(int objs) {
duke@435:     if (PAR_STATS_ENABLED) {
duke@435:       _overflow_refills++;
duke@435:       _overflow_refill_objs += objs;
duke@435:     }
duke@435:   }
duke@435: 
duke@435:   void start_strong_roots() {
duke@435:     _start_strong_roots = os::elapsedTime();
duke@435:   }
duke@435:   void end_strong_roots() {
duke@435:     _strong_roots_time += (os::elapsedTime() - _start_strong_roots);
duke@435:   }
duke@435:   double strong_roots_time() { return _strong_roots_time; }
duke@435:   void start_term_time() {
duke@435:     note_term_attempt();
duke@435:     _start_term = os::elapsedTime();
duke@435:   }
duke@435:   void end_term_time() {
duke@435:     _term_time += (os::elapsedTime() - _start_term);
duke@435:   }
duke@435:   double term_time() { return _term_time; }
duke@435: 
duke@435:   double elapsed() {
duke@435:     return os::elapsedTime() - _start;
duke@435:   }
duke@435: };
duke@435: 
duke@435: class ParNewGenTask: public AbstractGangTask {
coleenp@548:  private:
coleenp@548:   ParNewGeneration*            _gen;
coleenp@548:   Generation*                  _next_gen;
coleenp@548:   HeapWord*                    _young_old_boundary;
duke@435:   class ParScanThreadStateSet* _state_set;
duke@435: 
duke@435: public:
duke@435:   ParNewGenTask(ParNewGeneration*      gen,
duke@435:                 Generation*            next_gen,
duke@435:                 HeapWord*              young_old_boundary,
duke@435:                 ParScanThreadStateSet* state_set);
duke@435: 
duke@435:   HeapWord* young_old_boundary() { return _young_old_boundary; }
duke@435: 
duke@435:   void work(int i);
duke@435: };
duke@435: 
duke@435: class KeepAliveClosure: public DefNewGeneration::KeepAliveClosure {
coleenp@548:  protected:
coleenp@548:   template <class T> void do_oop_work(T* p);
duke@435:  public:
duke@435:   KeepAliveClosure(ScanWeakRefClosure* cl);
coleenp@548:   virtual void do_oop(oop* p);
coleenp@548:   virtual void do_oop(narrowOop* p);
duke@435: };
duke@435: 
duke@435: class EvacuateFollowersClosureGeneral: public VoidClosure {
coleenp@548:  private:
coleenp@548:   GenCollectedHeap* _gch;
coleenp@548:   int               _level;
coleenp@548:   OopsInGenClosure* _scan_cur_or_nonheap;
coleenp@548:   OopsInGenClosure* _scan_older;
coleenp@548:  public:
coleenp@548:   EvacuateFollowersClosureGeneral(GenCollectedHeap* gch, int level,
coleenp@548:                                   OopsInGenClosure* cur,
coleenp@548:                                   OopsInGenClosure* older);
coleenp@548:   virtual void do_void();
duke@435: };
duke@435: 
duke@435: // Closure for scanning ParNewGeneration.
duke@435: // Same as ScanClosure, except does parallel GC barrier.
duke@435: class ScanClosureWithParBarrier: public ScanClosure {
coleenp@548:  protected:
coleenp@548:   template <class T> void do_oop_work(T* p);
coleenp@548:  public:
duke@435:   ScanClosureWithParBarrier(ParNewGeneration* g, bool gc_barrier);
coleenp@548:   virtual void do_oop(oop* p);
coleenp@548:   virtual void do_oop(narrowOop* p);
duke@435: };
duke@435: 
duke@435: // Implements AbstractRefProcTaskExecutor for ParNew.
duke@435: class ParNewRefProcTaskExecutor: public AbstractRefProcTaskExecutor {
coleenp@548:  private:
coleenp@548:   ParNewGeneration&      _generation;
coleenp@548:   ParScanThreadStateSet& _state_set;
coleenp@548:  public:
duke@435:   ParNewRefProcTaskExecutor(ParNewGeneration& generation,
duke@435:                             ParScanThreadStateSet& state_set)
duke@435:     : _generation(generation), _state_set(state_set)
duke@435:   { }
duke@435: 
duke@435:   // Executes a task using worker threads.
duke@435:   virtual void execute(ProcessTask& task);
duke@435:   virtual void execute(EnqueueTask& task);
duke@435:   // Switch to single threaded mode.
duke@435:   virtual void set_single_threaded_mode();
duke@435: };
duke@435: 
duke@435: 
duke@435: // A Generation that does parallel young-gen collection.
duke@435: 
duke@435: class ParNewGeneration: public DefNewGeneration {
duke@435:   friend class ParNewGenTask;
duke@435:   friend class ParNewRefProcTask;
duke@435:   friend class ParNewRefProcTaskExecutor;
duke@435:   friend class ParScanThreadStateSet;
ysr@969:   friend class ParEvacuateFollowersClosure;
duke@435: 
coleenp@548:  private:
duke@435:   // XXX use a global constant instead of 64!
duke@435:   struct ObjToScanQueuePadded {
duke@435:         ObjToScanQueue work_queue;
duke@435:         char pad[64 - sizeof(ObjToScanQueue)];  // prevent false sharing
duke@435:   };
duke@435: 
duke@435:   // The per-thread work queues, available here for stealing.
duke@435:   ObjToScanQueueSet* _task_queues;
duke@435: 
duke@435:   // Desired size of survivor space plab's
duke@435:   PLABStats _plab_stats;
duke@435: 
duke@435:   // A list of from-space images of to-be-scanned objects, threaded through
duke@435:   // klass-pointers (klass information already copied to the forwarded
duke@435:   // image.)  Manipulated with CAS.
duke@435:   oop _overflow_list;
ysr@969:   NOT_PRODUCT(ssize_t _num_par_pushes;)
duke@435: 
duke@435:   // If true, older generation does not support promotion undo, so avoid.
duke@435:   static bool _avoid_promotion_undo;
duke@435: 
duke@435:   // This closure is used by the reference processor to filter out
duke@435:   // references to live referent.
duke@435:   DefNewGeneration::IsAliveClosure _is_alive_closure;
duke@435: 
duke@435:   static oop real_forwardee_slow(oop obj);
duke@435:   static void waste_some_time();
duke@435: 
duke@435:   // Preserve the mark of "obj", if necessary, in preparation for its mark
duke@435:   // word being overwritten with a self-forwarding-pointer.
duke@435:   void preserve_mark_if_necessary(oop obj, markOop m);
duke@435: 
duke@435:  protected:
duke@435: 
duke@435:   bool _survivor_overflow;
duke@435: 
duke@435:   bool avoid_promotion_undo() { return _avoid_promotion_undo; }
duke@435:   void set_avoid_promotion_undo(bool v) { _avoid_promotion_undo = v; }
duke@435: 
duke@435:   bool survivor_overflow() { return _survivor_overflow; }
duke@435:   void set_survivor_overflow(bool v) { _survivor_overflow = v; }
duke@435: 
duke@435:   // Adjust the tenuring threshold.  See the implementation for
duke@435:   // the details of the policy.
duke@435:   virtual void adjust_desired_tenuring_threshold();
duke@435: 
coleenp@548:  public:
duke@435:   ParNewGeneration(ReservedSpace rs, size_t initial_byte_size, int level);
duke@435: 
duke@435:   ~ParNewGeneration() {
duke@435:     for (uint i = 0; i < ParallelGCThreads; i++)
duke@435:         delete _task_queues->queue(i);
duke@435: 
duke@435:     delete _task_queues;
duke@435:   }
duke@435: 
duke@435:   virtual void ref_processor_init();
duke@435:   virtual Generation::Name kind()        { return Generation::ParNew; }
duke@435:   virtual const char* name() const;
duke@435:   virtual const char* short_name() const { return "ParNew"; }
duke@435: 
duke@435:   // override
duke@435:   virtual bool refs_discovery_is_mt()     const {
duke@435:     assert(UseParNewGC, "ParNewGeneration only when UseParNewGC");
duke@435:     return ParallelGCThreads > 1;
duke@435:   }
duke@435: 
duke@435:   // Make the collection virtual.
duke@435:   virtual void collect(bool   full,
duke@435:                        bool   clear_all_soft_refs,
duke@435:                        size_t size,
duke@435:                        bool   is_tlab);
duke@435: 
duke@435:   // This needs to be visible to the closure function.
duke@435:   // "obj" is the object to be copied, "m" is a recent value of its mark
duke@435:   // that must not contain a forwarding pointer (though one might be
duke@435:   // inserted in "obj"s mark word by a parallel thread).
duke@435:   inline oop copy_to_survivor_space(ParScanThreadState* par_scan_state,
duke@435:                              oop obj, size_t obj_sz, markOop m) {
duke@435:     if (_avoid_promotion_undo) {
duke@435:        return copy_to_survivor_space_avoiding_promotion_undo(par_scan_state,
duke@435:                                                              obj, obj_sz, m);
duke@435:     }
duke@435: 
duke@435:     return copy_to_survivor_space_with_undo(par_scan_state, obj, obj_sz, m);
duke@435:   }
duke@435: 
duke@435:   oop copy_to_survivor_space_avoiding_promotion_undo(ParScanThreadState* par_scan_state,
duke@435:                              oop obj, size_t obj_sz, markOop m);
duke@435: 
duke@435:   oop copy_to_survivor_space_with_undo(ParScanThreadState* par_scan_state,
duke@435:                              oop obj, size_t obj_sz, markOop m);
duke@435: 
ysr@969:   // in support of testing overflow code
ysr@969:   NOT_PRODUCT(int _overflow_counter;)
ysr@969:   NOT_PRODUCT(bool should_simulate_overflow();)
ysr@969: 
ysr@1114:   // Accessor for overflow list
ysr@1114:   oop overflow_list() { return _overflow_list; }
ysr@1114: 
duke@435:   // Push the given (from-space) object on the global overflow list.
ysr@969:   void push_on_overflow_list(oop from_space_obj, ParScanThreadState* par_scan_state);
duke@435: 
duke@435:   // If the global overflow list is non-empty, move some tasks from it
ysr@1114:   // onto "work_q" (which need not be empty).  No more than 1/4 of the
ysr@1114:   // available space on "work_q" is used.
duke@435:   bool take_from_overflow_list(ParScanThreadState* par_scan_state);
ysr@1114:   bool take_from_overflow_list_work(ParScanThreadState* par_scan_state);
duke@435: 
duke@435:   // The task queues to be used by parallel GC threads.
duke@435:   ObjToScanQueueSet* task_queues() {
duke@435:     return _task_queues;
duke@435:   }
duke@435: 
duke@435:   PLABStats* plab_stats() {
duke@435:     return &_plab_stats;
duke@435:   }
duke@435: 
duke@435:   size_t desired_plab_sz() {
duke@435:     return _plab_stats.desired_plab_sz();
duke@435:   }
duke@435: 
duke@435:   static oop real_forwardee(oop obj);
duke@435: 
duke@435:   DEBUG_ONLY(static bool is_legal_forward_ptr(oop p);)
duke@435: };