1 /*
2 * Copyright (c) 1999, 2015, 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_SHARED_THREADLOCALALLOCBUFFER_HPP
26 #define SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP
27
28 #include "gc/shared/gcUtil.hpp"
29 #include "oops/typeArrayOop.hpp"
30 #include "runtime/perfData.hpp"
31 #include "runtime/vm_version.hpp"
32
33 class GlobalTLABStats;
34
35 // ThreadLocalAllocBuffer: a descriptor for thread-local storage used by
36 // the threads for allocation.
37 // It is thread-private at any time, but maybe multiplexed over
38 // time across multiple threads. The park()/unpark() pair is
39 // used to make it available for such multiplexing.
40 class ThreadLocalAllocBuffer: public CHeapObj<mtThread> {
41 friend class VMStructs;
42 friend class JVMCIVMStructs;
43 private:
44 HeapWord* _start; // address of TLAB
45 HeapWord* _top; // address after last allocation
46 HeapWord* _pf_top; // allocation prefetch watermark
47 HeapWord* _end; // allocation end (excluding alignment_reserve)
48 HeapWord* _actual_end; // allocation actual_end (excluding alignment_reserve)
49 size_t _desired_size; // desired size (including alignment_reserve)
50 size_t _refill_waste_limit; // hold onto tlab if free() is larger than this
51 size_t _allocated_before_last_gc; // total bytes allocated up until the last gc
52
53 static size_t _max_size; // maximum size of any TLAB
54 static int _reserve_for_allocation_prefetch; // Reserve at the end of the TLAB
55 static unsigned _target_refills; // expected number of refills between GCs
56 size_t _bytes_until_sample; // bytes until sample.
57 size_t _extra_space; // extra space due to sampling logic.
58
59 unsigned _number_of_refills;
60 unsigned _fast_refill_waste;
61 unsigned _slow_refill_waste;
62 unsigned _gc_waste;
63 unsigned _slow_allocations;
64
65 AdaptiveWeightedAverage _allocation_fraction; // fraction of eden allocated in tlabs
66
67 void accumulate_statistics();
68 void initialize_statistics();
69
70 void set_start(HeapWord* start) { _start = start; }
71 void set_end(HeapWord* end) { _end = end; }
72 void set_actual_end(HeapWord* actual_end) { _actual_end = actual_end; }
73 void set_top(HeapWord* top) { _top = top; }
74 void set_pf_top(HeapWord* pf_top) { _pf_top = pf_top; }
75 void set_desired_size(size_t desired_size) { _desired_size = desired_size; }
76 void set_refill_waste_limit(size_t waste) { _refill_waste_limit = waste; }
77 void set_bytes_until_sample(size_t bytes) { _bytes_until_sample = bytes; }
78
79 size_t initial_refill_waste_limit() { return desired_size() / TLABRefillWasteFraction; }
80
81 static int target_refills() { return _target_refills; }
82 size_t initial_desired_size();
83
84 size_t remaining() const { return end() == NULL ? 0 : pointer_delta(hard_end(), top()); }
85
86 // Make parsable and release it.
87 void reset();
88
89 // Resize based on amount of allocation, etc.
90 void resize();
91
92 void invariants() const { assert(top() >= start() && top() <= end(), "invalid tlab"); }
93
94 void initialize(HeapWord* start, HeapWord* top, HeapWord* end);
95
96 void print_stats(const char* tag);
97
98 Thread* myThread();
99
100 // statistics
101
102 int number_of_refills() const { return _number_of_refills; }
103 int fast_refill_waste() const { return _fast_refill_waste; }
104 int slow_refill_waste() const { return _slow_refill_waste; }
105 int gc_waste() const { return _gc_waste; }
106 int slow_allocations() const { return _slow_allocations; }
107
108 static GlobalTLABStats* _global_stats;
109 static GlobalTLABStats* global_stats() { return _global_stats; }
110
111 public:
112 ThreadLocalAllocBuffer() : _allocation_fraction(TLABAllocationWeight), _allocated_before_last_gc(0) {
113 // do nothing. tlabs must be inited by initialize() calls
114 }
115
116 static size_t min_size() { return align_object_size(MinTLABSize / HeapWordSize) + alignment_reserve(); }
117 static size_t max_size() { assert(_max_size != 0, "max_size not set up"); return _max_size; }
118 static size_t max_size_in_bytes() { return max_size() * BytesPerWord; }
119 static void set_max_size(size_t max_size) { _max_size = max_size; }
120
121 HeapWord* start() const { return _start; }
122 HeapWord* end() const { return _end; }
123 HeapWord* hard_end() const { return _actual_end + alignment_reserve(); }
124 HeapWord* top() const { return _top; }
125 HeapWord* pf_top() const { return _pf_top; }
126 size_t desired_size() const { return _desired_size; }
127 size_t used() const { return pointer_delta(top(), start()); }
128 size_t used_bytes() const { return pointer_delta(top(), start(), 1); }
129 size_t free() const { return pointer_delta(end(), top()); }
130
131 // Don't discard tlab if remaining space is larger than this.
132 size_t refill_waste_limit() const { return _refill_waste_limit; }
133
134 // Allocate size HeapWords. The memory is NOT initialized to zero.
135 inline HeapWord* allocate(size_t size);
136
137 // Reserve space at the end of TLAB
138 static size_t end_reserve() {
139 int reserve_size = typeArrayOopDesc::header_size(T_INT);
140 return MAX2(reserve_size, _reserve_for_allocation_prefetch);
141 }
142 static size_t alignment_reserve() { return align_object_size(end_reserve()); }
143 static size_t alignment_reserve_in_bytes() { return alignment_reserve() * HeapWordSize; }
144
145 // Return tlab size or remaining space in eden such that the
146 // space is large enough to hold obj_size and necessary fill space.
147 // Otherwise return 0;
148 inline size_t compute_size(size_t obj_size);
149
150 // Record slow allocation
151 inline void record_slow_allocation(size_t obj_size);
152
153 // Initialization at startup
154 static void startup_initialization();
155
156 // Make an in-use tlab parsable, optionally retiring and/or zapping it.
157 void make_parsable(bool retire, bool zap = true);
158
159 // Retire in-use tlab before allocation of a new tlab
160 void clear_before_allocation();
161
162 // Accumulate statistics across all tlabs before gc
163 static void accumulate_statistics_before_gc();
164
165 // Resize tlabs for all threads
166 static void resize_all_tlabs();
167
168 void set_back_actual_end() { _end = _actual_end; }
169 void fill(HeapWord* start, HeapWord* top, size_t new_size);
170 void initialize();
171
172 void pick_next_sample();
173 void handle_sample(Thread* thread, HeapWord* result, size_t size);
174 size_t bytes_until_sample() { return _bytes_until_sample; }
175 size_t *bytes_until_sample_addr() { return &_bytes_until_sample; }
176 bool should_sample() { return bytes_until_sample() == 0; }
177
178 static size_t refill_waste_limit_increment() { return TLABWasteIncrement; }
179
180 // Code generation support
181 static ByteSize start_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _start); }
182 static ByteSize end_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _end ); }
183 static ByteSize actual_end_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _actual_end ); }
184 static ByteSize top_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _top ); }
185 static ByteSize pf_top_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _pf_top ); }
186 static ByteSize size_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _desired_size ); }
187 static ByteSize refill_waste_limit_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _refill_waste_limit ); }
188
189 static ByteSize number_of_refills_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _number_of_refills ); }
190 static ByteSize fast_refill_waste_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _fast_refill_waste ); }
191 static ByteSize slow_allocations_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _slow_allocations ); }
192
193 void verify();
194 };
195
196 class GlobalTLABStats: public CHeapObj<mtThread> {
197 private:
198
199 // Accumulate perfdata in private variables because
200 // PerfData should be write-only for security reasons
201 // (see perfData.hpp)
202 unsigned _allocating_threads;
203 unsigned _total_refills;
204 unsigned _max_refills;
205 size_t _total_allocation;
206 size_t _total_gc_waste;
207 size_t _max_gc_waste;
208 size_t _total_slow_refill_waste;
209 size_t _max_slow_refill_waste;
210 size_t _total_fast_refill_waste;
211 size_t _max_fast_refill_waste;
212 unsigned _total_slow_allocations;
213 unsigned _max_slow_allocations;
214
215 PerfVariable* _perf_allocating_threads;
216 PerfVariable* _perf_total_refills;
217 PerfVariable* _perf_max_refills;
218 PerfVariable* _perf_allocation;
219 PerfVariable* _perf_gc_waste;
220 PerfVariable* _perf_max_gc_waste;
221 PerfVariable* _perf_slow_refill_waste;
222 PerfVariable* _perf_max_slow_refill_waste;
223 PerfVariable* _perf_fast_refill_waste;
224 PerfVariable* _perf_max_fast_refill_waste;
225 PerfVariable* _perf_slow_allocations;
226 PerfVariable* _perf_max_slow_allocations;
227
228 AdaptiveWeightedAverage _allocating_threads_avg;
229
230 public:
231 GlobalTLABStats();
232
233 // Initialize all counters
234 void initialize();
235
236 // Write all perf counters to the perf_counters
237 void publish();
238
239 void print();
240
241 // Accessors
242 unsigned allocating_threads_avg() {
243 return MAX2((unsigned)(_allocating_threads_avg.average() + 0.5), 1U);
244 }
245
246 size_t allocation() {
247 return _total_allocation;
248 }
249
250 // Update methods
251
252 void update_allocating_threads() {
253 _allocating_threads++;
254 }
255 void update_number_of_refills(unsigned value) {
256 _total_refills += value;
257 _max_refills = MAX2(_max_refills, value);
258 }
259 void update_allocation(size_t value) {
260 _total_allocation += value;
261 }
262 void update_gc_waste(size_t value) {
263 _total_gc_waste += value;
264 _max_gc_waste = MAX2(_max_gc_waste, value);
265 }
266 void update_fast_refill_waste(size_t value) {
267 _total_fast_refill_waste += value;
268 _max_fast_refill_waste = MAX2(_max_fast_refill_waste, value);
269 }
270 void update_slow_refill_waste(size_t value) {
271 _total_slow_refill_waste += value;
272 _max_slow_refill_waste = MAX2(_max_slow_refill_waste, value);
273 }
274 void update_slow_allocations(unsigned value) {
275 _total_slow_allocations += value;
276 _max_slow_allocations = MAX2(_max_slow_allocations, value);
277 }
278 };
279
280 #endif // SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP