1 /*
2 * Copyright (c) 2001, 2010, 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 // Forward decl
26 class ConcurrentG1RefineThread;
27 class G1RemSet;
28
29 class ConcurrentG1Refine: public CHeapObj {
30 ConcurrentG1RefineThread** _threads;
31 int _n_threads;
32 int _n_worker_threads;
33 /*
34 * The value of the update buffer queue length falls into one of 3 zones:
35 * green, yellow, red. If the value is in [0, green) nothing is
36 * done, the buffers are left unprocessed to enable the caching effect of the
37 * dirtied cards. In the yellow zone [green, yellow) the concurrent refinement
38 * threads are gradually activated. In [yellow, red) all threads are
39 * running. If the length becomes red (max queue length) the mutators start
40 * processing the buffers.
41 *
42 * There are some interesting cases (when G1UseAdaptiveConcRefinement
43 * is turned off):
44 * 1) green = yellow = red = 0. In this case the mutator will process all
45 * buffers. Except for those that are created by the deferred updates
46 * machinery during a collection.
47 * 2) green = 0. Means no caching. Can be a good way to minimize the
48 * amount of time spent updating rsets during a collection.
49 */
50 int _green_zone;
51 int _yellow_zone;
52 int _red_zone;
53
54 int _thread_threshold_step;
55
56 // Reset the threshold step value based of the current zone boundaries.
57 void reset_threshold_step();
58
59 // The cache for card refinement.
60 bool _use_cache;
61 bool _def_use_cache;
62
63 size_t _n_periods; // Used as clearing epoch
64
65 // An evicting cache of the number of times each card
66 // is accessed. Reduces, but does not eliminate, the amount
67 // of duplicated processing of dirty cards.
68
69 enum SomePrivateConstants {
70 epoch_bits = 32,
71 card_num_shift = epoch_bits,
72 epoch_mask = AllBits,
73 card_num_mask = AllBits,
74
75 // The initial cache size is approximately this fraction
76 // of a maximal cache (i.e. the size needed for all cards
77 // in the heap)
78 InitialCacheFraction = 512
79 };
80
81 const static julong card_num_mask_in_place =
82 (julong) card_num_mask << card_num_shift;
83
84 typedef struct {
85 julong _value; // | card_num | epoch |
86 } CardEpochCacheEntry;
87
88 julong make_epoch_entry(unsigned int card_num, unsigned int epoch) {
89 assert(0 <= card_num && card_num < _max_n_card_counts, "Bounds");
90 assert(0 <= epoch && epoch <= _n_periods, "must be");
91
92 return ((julong) card_num << card_num_shift) | epoch;
93 }
94
95 unsigned int extract_epoch(julong v) {
96 return (v & epoch_mask);
97 }
98
99 unsigned int extract_card_num(julong v) {
100 return (v & card_num_mask_in_place) >> card_num_shift;
101 }
102
103 typedef struct {
104 unsigned char _count;
105 unsigned char _evict_count;
106 } CardCountCacheEntry;
107
108 CardCountCacheEntry* _card_counts;
109 CardEpochCacheEntry* _card_epochs;
110
111 // The current number of buckets in the card count cache
112 unsigned _n_card_counts;
113
114 // The max number of buckets required for the number of
115 // cards for the entire reserved heap
116 unsigned _max_n_card_counts;
117
118 // Possible sizes of the cache: odd primes that roughly double in size.
119 // (See jvmtiTagMap.cpp).
120 static int _cc_cache_sizes[];
121
122 // The index in _cc_cache_sizes corresponding to the size of
123 // _card_counts.
124 int _cache_size_index;
125
126 bool _expand_card_counts;
127
128 const jbyte* _ct_bot;
129
130 jbyte** _hot_cache;
131 int _hot_cache_size;
132 int _n_hot;
133 int _hot_cache_idx;
134
135 int _hot_cache_par_chunk_size;
136 volatile int _hot_cache_par_claimed_idx;
137
138 // Needed to workaround 6817995
139 CardTableModRefBS* _ct_bs;
140 G1CollectedHeap* _g1h;
141
142 // Expands the array that holds the card counts to the next size up
143 void expand_card_count_cache();
144
145 // hash a given key (index of card_ptr) with the specified size
146 static unsigned int hash(size_t key, int size) {
147 return (unsigned int) key % size;
148 }
149
150 // hash a given key (index of card_ptr)
151 unsigned int hash(size_t key) {
152 return hash(key, _n_card_counts);
153 }
154
155 unsigned ptr_2_card_num(jbyte* card_ptr) {
156 return (unsigned) (card_ptr - _ct_bot);
157 }
158
159 jbyte* card_num_2_ptr(unsigned card_num) {
160 return (jbyte*) (_ct_bot + card_num);
161 }
162
163 // Returns the count of this card after incrementing it.
164 jbyte* add_card_count(jbyte* card_ptr, int* count, bool* defer);
165
166 // Returns true if this card is in a young region
167 bool is_young_card(jbyte* card_ptr);
168
169 public:
170 ConcurrentG1Refine();
171 ~ConcurrentG1Refine();
172
173 void init(); // Accomplish some initialization that has to wait.
174 void stop();
175
176 void reinitialize_threads();
177
178 // Iterate over the conc refine threads
179 void threads_do(ThreadClosure *tc);
180
181 // If this is the first entry for the slot, writes into the cache and
182 // returns NULL. If it causes an eviction, returns the evicted pointer.
183 // Otherwise, its a cache hit, and returns NULL.
184 jbyte* cache_insert(jbyte* card_ptr, bool* defer);
185
186 // Process the cached entries.
187 void clean_up_cache(int worker_i, G1RemSet* g1rs, DirtyCardQueue* into_cset_dcq);
188
189 // Set up for parallel processing of the cards in the hot cache
190 void clear_hot_cache_claimed_index() {
191 _hot_cache_par_claimed_idx = 0;
192 }
193
194 // Discard entries in the hot cache.
195 void clear_hot_cache() {
196 _hot_cache_idx = 0; _n_hot = 0;
197 }
198
199 bool hot_cache_is_empty() { return _n_hot == 0; }
200
201 bool use_cache() { return _use_cache; }
202 void set_use_cache(bool b) {
203 if (b) _use_cache = _def_use_cache;
204 else _use_cache = false;
205 }
206
207 void clear_and_record_card_counts();
208
209 static int thread_num();
210
211 void print_worker_threads_on(outputStream* st) const;
212
213 void set_green_zone(int x) { _green_zone = x; }
214 void set_yellow_zone(int x) { _yellow_zone = x; }
215 void set_red_zone(int x) { _red_zone = x; }
216
217 int green_zone() const { return _green_zone; }
218 int yellow_zone() const { return _yellow_zone; }
219 int red_zone() const { return _red_zone; }
220
221 int total_thread_num() const { return _n_threads; }
222 int worker_thread_num() const { return _n_worker_threads; }
223
224 int thread_threshold_step() const { return _thread_threshold_step; }
225 };