1 /*
   2  * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #ifndef SHARE_VM_GC_IMPLEMENTATION_G1_G1STRINGDEDUP_HPP
  26 #define SHARE_VM_GC_IMPLEMENTATION_G1_G1STRINGDEDUP_HPP
  27 
  28 //
  29 // String Deduplication
  30 //
  31 // String deduplication aims to reduce the heap live-set by deduplicating identical
  32 // instances of String so that they share the same backing character array.
  33 //
  34 // The deduplication process is divided in two main parts, 1) finding the objects to
  35 // deduplicate, and 2) deduplicating those objects. The first part is done as part of
  36 // a normal GC cycle when objects are marked or evacuated. At this time a check is
  37 // applied on each object to check if it is a candidate for deduplication. If so, the
  38 // object is placed on the deduplication queue for later processing. The second part,
  39 // processing the objects on the deduplication queue, is a concurrent phase which
  40 // starts right after the stop-the-wold marking/evacuation phase. This phase is
  41 // executed by the deduplication thread, which pulls deduplication candidates of the
  42 // deduplication queue and tries to deduplicate them.
  43 //
  44 // A deduplication hashtable is used to keep track of all unique character arrays
  45 // used by String objects. When deduplicating, a lookup is made in this table to see
  46 // if there is already an identical character array somewhere on the heap. If so, the
  47 // String object is adjusted to point to that character array, releasing the reference
  48 // to the original array allowing it to eventually be garbage collected. If the lookup
  49 // fails the character array is instead inserted into the hashtable so that this array
  50 // can be shared at some point in the future.
  51 //
  52 // Candidate selection
  53 //
  54 // An object is considered a deduplication candidate if all of the following
  55 // statements are true:
  56 //
  57 // - The object is an instance of java.lang.String
  58 //
  59 // - The object is being evacuated from a young heap region
  60 //
  61 // - The object is being evacuated to a young/survivor heap region and the
  62 //   object's age is equal to the deduplication age threshold
  63 //
  64 //   or
  65 //
  66 //   The object is being evacuated to an old heap region and the object's age is
  67 //   less than the deduplication age threshold
  68 //
  69 // Once an string object has been promoted to an old region, or its age is higher
  70 // than the deduplication age threshold, is will never become a candidate again.
  71 // This approach avoids making the same object a candidate more than once.
  72 //
  73 // Interned strings are a bit special. They are explicitly deduplicated just before
  74 // being inserted into the StringTable (to avoid counteracting C2 optimizations done
  75 // on string literals), then they also become deduplication candidates if they reach
  76 // the deduplication age threshold or is evacuated to an old heap region. The second
  77 // appempt to deduplicate such strings will be in vain, but we have no fast way of
  78 // filtering them out. This has not shown to be a problem, as the number of interned
  79 // strings is usually dwarfed by the number of normal (non-interned) strings.
  80 //
  81 // For additional information on string deduplication, please see JEP 192,
  82 // http://openjdk.java.net/jeps/192
  83 //
  84 
  85 #include "memory/allocation.hpp"
  86 #include "oops/oop.hpp"
  87 
  88 class OopClosure;
  89 class BoolObjectClosure;
  90 class ThreadClosure;
  91 class outputStream;
  92 class G1StringDedupTable;
  93 
  94 //
  95 // Main interface for interacting with string deduplication.
  96 //
  97 class G1StringDedup : public AllStatic {
  98 private:
  99   // Single state for checking if both G1 and string deduplication is enabled.
 100   static bool _enabled;
 101 
 102   // Candidate selection policies, returns true if the given object is
 103   // candidate for string deduplication.
 104   static bool is_candidate_from_mark(oop obj);
 105   static bool is_candidate_from_evacuation(bool from_young, bool to_young, oop obj);
 106 
 107 public:
 108   // Returns true if both G1 and string deduplication is enabled.
 109   static bool is_enabled() {
 110     return _enabled;
 111   }
 112 
 113   static void initialize();
 114 
 115   // Immediately deduplicates the given String object, bypassing the
 116   // the deduplication queue.
 117   static void deduplicate(oop java_string);
 118 
 119   // Enqueues a deduplication candidate for later processing by the deduplication
 120   // thread. Before enqueuing, these functions apply the approapriate candidate
 121   // selection policy to filters out non-candidates.
 122   static void enqueue_from_mark(oop java_string);
 123   static void enqueue_from_evacuation(bool from_young, bool to_young,
 124                                       unsigned int queue, oop java_string);
 125 
 126   static void oops_do(OopClosure* keep_alive);
 127   static void unlink(BoolObjectClosure* is_alive);
 128   static void unlink_or_oops_do(BoolObjectClosure* is_alive, OopClosure* keep_alive,
 129                                 bool allow_resize_and_rehash = true);
 130 
 131   static void threads_do(ThreadClosure* tc);
 132   static void print_worker_threads_on(outputStream* st);
 133   static void verify();
 134 };
 135 
 136 //
 137 // This closure encapsulates the state and the closures needed when scanning
 138 // the deduplication queue and table during the unlink_or_oops_do() operation.
 139 // A single instance of this closure is create and then shared by all worker
 140 // threads participating in the scan. The _next_queue and _next_bucket fields
 141 // provide a simple mechanism for GC workers to claim exclusing access to a
 142 // queue or a table partition.
 143 //
 144 class G1StringDedupUnlinkOrOopsDoClosure : public StackObj {
 145 private:
 146   BoolObjectClosure*  _is_alive;
 147   OopClosure*         _keep_alive;
 148   G1StringDedupTable* _resized_table;
 149   G1StringDedupTable* _rehashed_table;
 150   size_t              _next_queue;
 151   size_t              _next_bucket;
 152 
 153 public:
 154   G1StringDedupUnlinkOrOopsDoClosure(BoolObjectClosure* is_alive,
 155                                      OopClosure* keep_alive,
 156                                      bool allow_resize_and_rehash);
 157   ~G1StringDedupUnlinkOrOopsDoClosure();
 158 
 159   bool is_resizing() {
 160     return _resized_table != NULL;
 161   }
 162 
 163   G1StringDedupTable* resized_table() {
 164     return _resized_table;
 165   }
 166 
 167   bool is_rehashing() {
 168     return _rehashed_table != NULL;
 169   }
 170 
 171   // Atomically claims the next available queue for exclusive access by
 172   // the current thread. Returns the queue number of the claimed queue.
 173   size_t claim_queue() {
 174     return (size_t)Atomic::add_ptr(1, &_next_queue) - 1;
 175   }
 176 
 177   // Atomically claims the next available table partition for exclusive
 178   // access by the current thread. Returns the table bucket number where
 179   // the claimed parition starts.
 180   size_t claim_table_partition(size_t partition_size) {
 181     return (size_t)Atomic::add_ptr(partition_size, &_next_bucket) - partition_size;
 182   }
 183 
 184   // Applies and returns the result from the is_alive closure, or
 185   // returns true if no such closure was provided.
 186   bool is_alive(oop o) {
 187     if (_is_alive != NULL) {
 188       return _is_alive->do_object_b(o);
 189     }
 190     return true;
 191   }
 192 
 193   // Applies the keep_alive closure, or does nothing if no such
 194   // closure was provided.
 195   void keep_alive(oop* p) {
 196     if (_keep_alive != NULL) {
 197       _keep_alive->do_oop(p);
 198     }
 199   }
 200 };
 201 
 202 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1STRINGDEDUP_HPP