pliden@6413: /*
pliden@6413:  * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
pliden@6413:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
pliden@6413:  *
pliden@6413:  * This code is free software; you can redistribute it and/or modify it
pliden@6413:  * under the terms of the GNU General Public License version 2 only, as
pliden@6413:  * published by the Free Software Foundation.
pliden@6413:  *
pliden@6413:  * This code is distributed in the hope that it will be useful, but WITHOUT
pliden@6413:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
pliden@6413:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
pliden@6413:  * version 2 for more details (a copy is included in the LICENSE file that
pliden@6413:  * accompanied this code).
pliden@6413:  *
pliden@6413:  * You should have received a copy of the GNU General Public License version
pliden@6413:  * 2 along with this work; if not, write to the Free Software Foundation,
pliden@6413:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
pliden@6413:  *
pliden@6413:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
pliden@6413:  * or visit www.oracle.com if you need additional information or have any
pliden@6413:  * questions.
pliden@6413:  *
pliden@6413:  */
pliden@6413: 
pliden@6413: #ifndef SHARE_VM_GC_IMPLEMENTATION_G1_G1STRINGDEDUP_HPP
pliden@6413: #define SHARE_VM_GC_IMPLEMENTATION_G1_G1STRINGDEDUP_HPP
pliden@6413: 
pliden@6413: //
pliden@6413: // String Deduplication
pliden@6413: //
pliden@6413: // String deduplication aims to reduce the heap live-set by deduplicating identical
pliden@6413: // instances of String so that they share the same backing character array.
pliden@6413: //
pliden@6413: // The deduplication process is divided in two main parts, 1) finding the objects to
pliden@6413: // deduplicate, and 2) deduplicating those objects. The first part is done as part of
pliden@6413: // a normal GC cycle when objects are marked or evacuated. At this time a check is
pliden@6413: // applied on each object to check if it is a candidate for deduplication. If so, the
pliden@6413: // object is placed on the deduplication queue for later processing. The second part,
pliden@6413: // processing the objects on the deduplication queue, is a concurrent phase which
pliden@6413: // starts right after the stop-the-wold marking/evacuation phase. This phase is
pliden@6413: // executed by the deduplication thread, which pulls deduplication candidates of the
pliden@6413: // deduplication queue and tries to deduplicate them.
pliden@6413: //
pliden@6413: // A deduplication hashtable is used to keep track of all unique character arrays
pliden@6413: // used by String objects. When deduplicating, a lookup is made in this table to see
pliden@6413: // if there is already an identical character array somewhere on the heap. If so, the
pliden@6413: // String object is adjusted to point to that character array, releasing the reference
pliden@6413: // to the original array allowing it to eventually be garbage collected. If the lookup
pliden@6413: // fails the character array is instead inserted into the hashtable so that this array
pliden@6413: // can be shared at some point in the future.
pliden@6413: //
pliden@6413: // Candidate selection
pliden@6413: //
pliden@6413: // An object is considered a deduplication candidate if all of the following
pliden@6413: // statements are true:
pliden@6413: //
pliden@6413: // - The object is an instance of java.lang.String
pliden@6413: //
pliden@6413: // - The object is being evacuated from a young heap region
pliden@6413: //
pliden@6413: // - The object is being evacuated to a young/survivor heap region and the
pliden@6413: //   object's age is equal to the deduplication age threshold
pliden@6413: //
pliden@6413: //   or
pliden@6413: //
pliden@6413: //   The object is being evacuated to an old heap region and the object's age is
pliden@6413: //   less than the deduplication age threshold
pliden@6413: //
pliden@6413: // Once an string object has been promoted to an old region, or its age is higher
pliden@6413: // than the deduplication age threshold, is will never become a candidate again.
pliden@6413: // This approach avoids making the same object a candidate more than once.
pliden@6413: //
pliden@6413: // Interned strings are a bit special. They are explicitly deduplicated just before
pliden@6413: // being inserted into the StringTable (to avoid counteracting C2 optimizations done
pliden@6413: // on string literals), then they also become deduplication candidates if they reach
pliden@6413: // the deduplication age threshold or are evacuated to an old heap region. The second
pliden@6413: // attempt to deduplicate such strings will be in vain, but we have no fast way of
pliden@6413: // filtering them out. This has not shown to be a problem, as the number of interned
pliden@6413: // strings is usually dwarfed by the number of normal (non-interned) strings.
pliden@6413: //
pliden@6413: // For additional information on string deduplication, please see JEP 192,
pliden@6413: // http://openjdk.java.net/jeps/192
pliden@6413: //
pliden@6413: 
pliden@6413: #include "memory/allocation.hpp"
pliden@6413: #include "oops/oop.hpp"
pliden@6413: 
pliden@6413: class OopClosure;
pliden@6413: class BoolObjectClosure;
pliden@6413: class ThreadClosure;
pliden@6413: class outputStream;
pliden@6413: class G1StringDedupTable;
brutisso@7658: class G1GCPhaseTimes;
pliden@6413: 
pliden@6413: //
pliden@6413: // Main interface for interacting with string deduplication.
pliden@6413: //
pliden@6413: class G1StringDedup : public AllStatic {
pliden@6413: private:
pliden@6413:   // Single state for checking if both G1 and string deduplication is enabled.
pliden@6413:   static bool _enabled;
pliden@6413: 
pliden@6413:   // Candidate selection policies, returns true if the given object is
pliden@6413:   // candidate for string deduplication.
pliden@6413:   static bool is_candidate_from_mark(oop obj);
pliden@6413:   static bool is_candidate_from_evacuation(bool from_young, bool to_young, oop obj);
pliden@6413: 
pliden@6413: public:
pliden@6413:   // Returns true if both G1 and string deduplication is enabled.
pliden@6413:   static bool is_enabled() {
pliden@6413:     return _enabled;
pliden@6413:   }
pliden@6413: 
pliden@6690:   // Initialize string deduplication.
pliden@6413:   static void initialize();
pliden@6413: 
pliden@6690:   // Stop the deduplication thread.
pliden@6690:   static void stop();
pliden@6690: 
pliden@6413:   // Immediately deduplicates the given String object, bypassing the
pliden@6413:   // the deduplication queue.
pliden@6413:   static void deduplicate(oop java_string);
pliden@6413: 
pliden@6413:   // Enqueues a deduplication candidate for later processing by the deduplication
pliden@6413:   // thread. Before enqueuing, these functions apply the appropriate candidate
pliden@6413:   // selection policy to filters out non-candidates.
pliden@6413:   static void enqueue_from_mark(oop java_string);
pliden@6413:   static void enqueue_from_evacuation(bool from_young, bool to_young,
pliden@6413:                                       unsigned int queue, oop java_string);
pliden@6413: 
pliden@6413:   static void oops_do(OopClosure* keep_alive);
pliden@6413:   static void unlink(BoolObjectClosure* is_alive);
pliden@6413:   static void unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* keep_alive,
brutisso@7658:                                 bool allow_resize_and_rehash, G1GCPhaseTimes* phase_times = NULL);
pliden@6413: 
pliden@6413:   static void threads_do(ThreadClosure* tc);
pliden@6413:   static void print_worker_threads_on(outputStream* st);
pliden@6413:   static void verify();
pliden@6413: };
pliden@6413: 
pliden@6413: //
pliden@6413: // This closure encapsulates the state and the closures needed when scanning
pliden@6413: // the deduplication queue and table during the unlink_or_oops_do() operation.
pliden@6413: // A single instance of this closure is created and then shared by all worker
pliden@6413: // threads participating in the scan. The _next_queue and _next_bucket fields
pliden@6413: // provide a simple mechanism for GC workers to claim exclusive access to a
pliden@6413: // queue or a table partition.
pliden@6413: //
pliden@6413: class G1StringDedupUnlinkOrOopsDoClosure : public StackObj {
pliden@6413: private:
pliden@6413:   BoolObjectClosure*  _is_alive;
pliden@6413:   OopClosure*         _keep_alive;
pliden@6413:   G1StringDedupTable* _resized_table;
pliden@6413:   G1StringDedupTable* _rehashed_table;
pliden@6413:   size_t              _next_queue;
pliden@6413:   size_t              _next_bucket;
pliden@6413: 
pliden@6413: public:
pliden@6413:   G1StringDedupUnlinkOrOopsDoClosure(BoolObjectClosure* is_alive,
pliden@6413:                                      OopClosure* keep_alive,
pliden@6413:                                      bool allow_resize_and_rehash);
pliden@6413:   ~G1StringDedupUnlinkOrOopsDoClosure();
pliden@6413: 
pliden@6413:   bool is_resizing() {
pliden@6413:     return _resized_table != NULL;
pliden@6413:   }
pliden@6413: 
pliden@6413:   G1StringDedupTable* resized_table() {
pliden@6413:     return _resized_table;
pliden@6413:   }
pliden@6413: 
pliden@6413:   bool is_rehashing() {
pliden@6413:     return _rehashed_table != NULL;
pliden@6413:   }
pliden@6413: 
pliden@6413:   // Atomically claims the next available queue for exclusive access by
pliden@6413:   // the current thread. Returns the queue number of the claimed queue.
pliden@6413:   size_t claim_queue() {
pliden@6413:     return (size_t)Atomic::add_ptr(1, &_next_queue) - 1;
pliden@6413:   }
pliden@6413: 
pliden@6413:   // Atomically claims the next available table partition for exclusive
pliden@6413:   // access by the current thread. Returns the table bucket number where
pliden@6413:   // the claimed partition starts.
pliden@6413:   size_t claim_table_partition(size_t partition_size) {
pliden@6413:     return (size_t)Atomic::add_ptr(partition_size, &_next_bucket) - partition_size;
pliden@6413:   }
pliden@6413: 
pliden@6413:   // Applies and returns the result from the is_alive closure, or
pliden@6413:   // returns true if no such closure was provided.
pliden@6413:   bool is_alive(oop o) {
pliden@6413:     if (_is_alive != NULL) {
pliden@6413:       return _is_alive->do_object_b(o);
pliden@6413:     }
pliden@6413:     return true;
pliden@6413:   }
pliden@6413: 
pliden@6413:   // Applies the keep_alive closure, or does nothing if no such
pliden@6413:   // closure was provided.
pliden@6413:   void keep_alive(oop* p) {
pliden@6413:     if (_keep_alive != NULL) {
pliden@6413:       _keep_alive->do_oop(p);
pliden@6413:     }
pliden@6413:   }
pliden@6413: };
pliden@6413: 
pliden@6413: #endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1STRINGDEDUP_HPP