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 //
41 // Heap sampling is performed via the end/actual_end fields.
42 // actual_end contains the real end of the tlab allocation,
43 // whereas end can be set to an arbitrary spot in the tlab to
44 // trip the return and sample the allocation.
45 // slow_path_end is used to track if a fast tlab refill occured
46 // between slowpath calls.
47 class ThreadLocalAllocBuffer: public CHeapObj<mtThread> {
48 friend class VMStructs;
49 friend class JVMCIVMStructs;
50 private:
51 HeapWord* _start; // address of TLAB
52 HeapWord* _top; // address after last allocation
53 HeapWord* _pf_top; // allocation prefetch watermark
54 HeapWord* _end; // allocation end (can be the sampling end point or
55 // the actual TLAB end, excluding alignment_reserve)
56 HeapWord* _actual_end; // allocation actual_end (actual TLAB end, excluding alignment_reserve)
57 HeapWord* _slow_path_end; // remember the end in case a fast refill occurs.
58
59 size_t _desired_size; // desired size (including alignment_reserve)
60 size_t _refill_waste_limit; // hold onto tlab if free() is larger than this
61 size_t _allocated_before_last_gc; // total bytes allocated up until the last gc
62 size_t _bytes_until_sample; // bytes until sample.
63
64 static size_t _max_size; // maximum size of any TLAB
65 static int _reserve_for_allocation_prefetch; // Reserve at the end of the TLAB
66 static unsigned _target_refills; // expected number of refills between GCs
67
68 unsigned _number_of_refills;
69 unsigned _fast_refill_waste;
70 unsigned _slow_refill_waste;
71 unsigned _gc_waste;
72 unsigned _slow_allocations;
73
74 AdaptiveWeightedAverage _allocation_fraction; // fraction of eden allocated in tlabs
75
76 void accumulate_statistics();
77 void initialize_statistics();
78
79 void set_start(HeapWord* start) { _start = start; }
80 void set_end(HeapWord* end) { _end = end; }
81 void set_actual_end(HeapWord* actual_end) { _actual_end = actual_end; }
82 void set_slow_path_end(HeapWord* slow_path_end) { _slow_path_end = slow_path_end; }
83 void set_top(HeapWord* top) { _top = top; }
84 void set_pf_top(HeapWord* pf_top) { _pf_top = pf_top; }
85 void set_desired_size(size_t desired_size) { _desired_size = desired_size; }
86 void set_refill_waste_limit(size_t waste) { _refill_waste_limit = waste; }
87 void set_bytes_until_sample(size_t bytes) { _bytes_until_sample = bytes; }
88
89 size_t initial_refill_waste_limit() { return desired_size() / TLABRefillWasteFraction; }
90
91 static int target_refills() { return _target_refills; }
92 size_t initial_desired_size();
93
94 size_t remaining() { return end() == NULL ? 0 : pointer_delta(hard_end(), top()); }
95
96 // Make parsable and release it.
97 void reset();
98
99 // Resize based on amount of allocation, etc.
100 void resize();
101
102 void invariants() const { assert(top() >= start() && top() <= end(), "invalid tlab"); }
103
104 void initialize(HeapWord* start, HeapWord* top, HeapWord* end);
105
106 void print_stats(const char* tag);
107
108 Thread* myThread();
109
110 // statistics
111
112 int number_of_refills() const { return _number_of_refills; }
113 int fast_refill_waste() const { return _fast_refill_waste; }
114 int slow_refill_waste() const { return _slow_refill_waste; }
115 int gc_waste() const { return _gc_waste; }
116 int slow_allocations() const { return _slow_allocations; }
117
118 static GlobalTLABStats* _global_stats;
119 static GlobalTLABStats* global_stats() { return _global_stats; }
120
121 public:
122 ThreadLocalAllocBuffer() : _allocation_fraction(TLABAllocationWeight), _allocated_before_last_gc(0) {
123 // do nothing. tlabs must be inited by initialize() calls
124 }
125
126 static size_t min_size() { return align_object_size(MinTLABSize / HeapWordSize) + alignment_reserve(); }
127 static size_t max_size() { assert(_max_size != 0, "max_size not set up"); return _max_size; }
128 static size_t max_size_in_bytes() { return max_size() * BytesPerWord; }
129 static void set_max_size(size_t max_size) { _max_size = max_size; }
130
131 HeapWord* start() const { return _start; }
132 HeapWord* end() const { return _end; }
133 HeapWord* slow_path_end() const { return _slow_path_end; }
134 HeapWord* actual_end() const { return _actual_end; }
135
136 HeapWord* hard_end();
137 HeapWord* top() const { return _top; }
138 HeapWord* pf_top() const { return _pf_top; }
139 size_t desired_size() const { return _desired_size; }
140 size_t used() const { return pointer_delta(top(), start()); }
141 size_t used_bytes() const { return pointer_delta(top(), start(), 1); }
142 size_t free() const { return pointer_delta(end(), top()); }
143 // Don't discard tlab if remaining space is larger than this.
144 size_t refill_waste_limit() const { return _refill_waste_limit; }
145
146 // Allocate size HeapWords. The memory is NOT initialized to zero.
147 inline HeapWord* allocate(size_t size);
148
149 // Reserve space at the end of TLAB
150 static size_t end_reserve() {
151 int reserve_size = typeArrayOopDesc::header_size(T_INT);
152 return MAX2(reserve_size, _reserve_for_allocation_prefetch);
153 }
154 static size_t alignment_reserve() { return align_object_size(end_reserve()); }
155 static size_t alignment_reserve_in_bytes() { return alignment_reserve() * HeapWordSize; }
156
157 // Return tlab size or remaining space in eden such that the
158 // space is large enough to hold obj_size and necessary fill space.
159 // Otherwise return 0;
160 inline size_t compute_size(size_t obj_size);
161
162 // Record slow allocation
163 inline void record_slow_allocation(size_t obj_size);
164
165 // Initialization at startup
166 static void startup_initialization();
167
168 // Make an in-use tlab parsable, optionally retiring and/or zapping it.
169 void make_parsable(bool retire, bool zap = true);
170
171 // Retire in-use tlab before allocation of a new tlab
172 void clear_before_allocation();
173
174 // Accumulate statistics across all tlabs before gc
175 static void accumulate_statistics_before_gc();
176
177 // Resize tlabs for all threads
178 static void resize_all_tlabs();
179
180 void fill(HeapWord* start, HeapWord* top, size_t new_size);
181 void initialize();
182
183 void pick_next_sample();
184 void set_back_actual_end();
185 void handle_sample(Thread* thread, HeapWord* result, size_t size);
186 size_t bytes_until_sample() { return _bytes_until_sample; }
187 size_t *bytes_until_sample_addr() { return &_bytes_until_sample; }
188 bool should_sample() { return bytes_until_sample() == 0; }
189
190 static size_t refill_waste_limit_increment() { return TLABWasteIncrement; }
191
192 // Code generation support
193 static ByteSize start_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _start); }
194 static ByteSize end_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _end ); }
195 static ByteSize actual_end_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _actual_end ); }
196 static ByteSize top_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _top ); }
197 static ByteSize pf_top_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _pf_top ); }
198 static ByteSize size_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _desired_size ); }
199 static ByteSize refill_waste_limit_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _refill_waste_limit ); }
200
201 static ByteSize number_of_refills_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _number_of_refills ); }
202 static ByteSize fast_refill_waste_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _fast_refill_waste ); }
203 static ByteSize slow_allocations_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _slow_allocations ); }
204
205 void verify();
206 };
207
208 class GlobalTLABStats: public CHeapObj<mtThread> {
209 private:
210
211 // Accumulate perfdata in private variables because
212 // PerfData should be write-only for security reasons
213 // (see perfData.hpp)
214 unsigned _allocating_threads;
215 unsigned _total_refills;
216 unsigned _max_refills;
217 size_t _total_allocation;
218 size_t _total_gc_waste;
219 size_t _max_gc_waste;
220 size_t _total_slow_refill_waste;
221 size_t _max_slow_refill_waste;
222 size_t _total_fast_refill_waste;
223 size_t _max_fast_refill_waste;
224 unsigned _total_slow_allocations;
225 unsigned _max_slow_allocations;
226
227 PerfVariable* _perf_allocating_threads;
228 PerfVariable* _perf_total_refills;
229 PerfVariable* _perf_max_refills;
230 PerfVariable* _perf_allocation;
231 PerfVariable* _perf_gc_waste;
232 PerfVariable* _perf_max_gc_waste;
233 PerfVariable* _perf_slow_refill_waste;
234 PerfVariable* _perf_max_slow_refill_waste;
235 PerfVariable* _perf_fast_refill_waste;
236 PerfVariable* _perf_max_fast_refill_waste;
237 PerfVariable* _perf_slow_allocations;
238 PerfVariable* _perf_max_slow_allocations;
239
240 AdaptiveWeightedAverage _allocating_threads_avg;
241
242 public:
243 GlobalTLABStats();
244
245 // Initialize all counters
246 void initialize();
247
248 // Write all perf counters to the perf_counters
249 void publish();
250
251 void print();
252
253 // Accessors
254 unsigned allocating_threads_avg() {
255 return MAX2((unsigned)(_allocating_threads_avg.average() + 0.5), 1U);
256 }
257
258 size_t allocation() {
259 return _total_allocation;
260 }
261
262 // Update methods
263
264 void update_allocating_threads() {
265 _allocating_threads++;
266 }
267 void update_number_of_refills(unsigned value) {
268 _total_refills += value;
269 _max_refills = MAX2(_max_refills, value);
270 }
271 void update_allocation(size_t value) {
272 _total_allocation += value;
273 }
274 void update_gc_waste(size_t value) {
275 _total_gc_waste += value;
276 _max_gc_waste = MAX2(_max_gc_waste, value);
277 }
278 void update_fast_refill_waste(size_t value) {
279 _total_fast_refill_waste += value;
280 _max_fast_refill_waste = MAX2(_max_fast_refill_waste, value);
281 }
282 void update_slow_refill_waste(size_t value) {
283 _total_slow_refill_waste += value;
284 _max_slow_refill_waste = MAX2(_max_slow_refill_waste, value);
285 }
286 void update_slow_allocations(unsigned value) {
287 _total_slow_allocations += value;
288 _max_slow_allocations = MAX2(_max_slow_allocations, value);
289 }
290 };
291
292 #endif // SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP