1 /*
2 * Copyright (c) 1999, 2018, 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 and allocation_end
42 // fields.
43 // allocation_end contains the real end of the tlab allocation,
44 // whereas end can be set to an arbitrary spot in the tlab to
45 // trip the return and sample the allocation.
46 class ThreadLocalAllocBuffer: public CHeapObj<mtThread> {
47 friend class VMStructs;
48 friend class JVMCIVMStructs;
49 private:
50 HeapWord* _start; // address of TLAB
51 HeapWord* _top; // address after last allocation
52 HeapWord* _pf_top; // allocation prefetch watermark
53 HeapWord* _end; // allocation end (can be the sampling end point or _allocation_end)
54 HeapWord* _allocation_end; // end for allocations (actual TLAB end, excluding alignment_reserve)
55
56 size_t _desired_size; // desired size (including alignment_reserve)
57 size_t _refill_waste_limit; // hold onto tlab if free() is larger than this
58 size_t _allocated_before_last_gc; // total bytes allocated up until the last gc
59 size_t _bytes_since_last_sample_point; // bytes since last sample point.
60
61 static size_t _max_size; // maximum size of any TLAB
62 static int _reserve_for_allocation_prefetch; // Reserve at the end of the TLAB
63 static unsigned _target_refills; // expected number of refills between GCs
64
65 unsigned _number_of_refills;
66 unsigned _fast_refill_waste;
67 unsigned _slow_refill_waste;
68 unsigned _gc_waste;
69 unsigned _slow_allocations;
70 size_t _allocated_size;
71
72 AdaptiveWeightedAverage _allocation_fraction; // fraction of eden allocated in tlabs
73
74 void accumulate_statistics();
75 void initialize_statistics();
76
77 void set_start(HeapWord* start) { _start = start; }
78 void set_end(HeapWord* end) { _end = end; }
79 void set_allocation_end(HeapWord* ptr) { _allocation_end = ptr; }
80 void set_top(HeapWord* top) { _top = top; }
81 void set_pf_top(HeapWord* pf_top) { _pf_top = pf_top; }
82 void set_desired_size(size_t desired_size) { _desired_size = desired_size; }
83 void set_refill_waste_limit(size_t waste) { _refill_waste_limit = waste; }
84
85 size_t initial_refill_waste_limit() { return desired_size() / TLABRefillWasteFraction; }
86
87 static int target_refills() { return _target_refills; }
88 size_t initial_desired_size();
89
90 size_t remaining();
91
92 // Make parsable and release it.
93 void reset();
94
95 void invariants() const { assert(top() >= start() && top() <= end(), "invalid tlab"); }
96
97 void initialize(HeapWord* start, HeapWord* top, HeapWord* end);
98
99 void print_stats(const char* tag);
100
101 Thread* thread();
102
103 // statistics
104
105 int number_of_refills() const { return _number_of_refills; }
106 int fast_refill_waste() const { return _fast_refill_waste; }
107 int slow_refill_waste() const { return _slow_refill_waste; }
108 int gc_waste() const { return _gc_waste; }
109 int slow_allocations() const { return _slow_allocations; }
110
111 static GlobalTLABStats* _global_stats;
112 static GlobalTLABStats* global_stats() { return _global_stats; }
113
114 public:
115 ThreadLocalAllocBuffer() : _allocated_before_last_gc(0), _allocation_fraction(TLABAllocationWeight) {
116 // do nothing. tlabs must be inited by initialize() calls
117 }
118
119 static size_t min_size() { return align_object_size(MinTLABSize / HeapWordSize) + alignment_reserve(); }
120 static size_t max_size() { assert(_max_size != 0, "max_size not set up"); return _max_size; }
121 static size_t max_size_in_bytes() { return max_size() * BytesPerWord; }
122 static void set_max_size(size_t max_size) { _max_size = max_size; }
123
124 HeapWord* start() const { return _start; }
125 HeapWord* end() const { return _end; }
126 HeapWord* top() const { return _top; }
127 HeapWord* hard_end();
128 HeapWord* pf_top() const { return _pf_top; }
129 size_t desired_size() const { return _desired_size; }
130 size_t used() const { return pointer_delta(top(), start()); }
131 size_t used_bytes() const { return pointer_delta(top(), start(), 1); }
132 size_t free() const { return pointer_delta(end(), top()); }
133 // Don't discard tlab if remaining space is larger than this.
134 size_t refill_waste_limit() const { return _refill_waste_limit; }
135 size_t bytes_since_last_sample_point() const { return _bytes_since_last_sample_point; }
136
137 // Allocate size HeapWords. The memory is NOT initialized to zero.
138 inline HeapWord* allocate(size_t size);
139
140 // Reserve space at the end of TLAB
141 static size_t end_reserve() {
142 int reserve_size = typeArrayOopDesc::header_size(T_INT);
143 return MAX2(reserve_size, _reserve_for_allocation_prefetch);
144 }
145 static size_t alignment_reserve() { return align_object_size(end_reserve()); }
146 static size_t alignment_reserve_in_bytes() { return alignment_reserve() * HeapWordSize; }
147
148 // Return tlab size or remaining space in eden such that the
149 // space is large enough to hold obj_size and necessary fill space.
150 // Otherwise return 0;
151 inline size_t compute_size(size_t obj_size);
152
153 // Compute the minimal needed tlab size for the given object size.
154 static inline size_t compute_min_size(size_t obj_size);
155
156 // Record slow allocation
157 inline void record_slow_allocation(size_t obj_size);
158
159 // Initialization at startup
160 static void startup_initialization();
161
162 // Make an in-use tlab parsable, optionally retiring and/or zapping it.
163 void make_parsable(bool retire, bool zap = true);
164
165 // Retire in-use tlab before allocation of a new tlab
166 void clear_before_allocation();
167
168 // Resize based on amount of allocation, etc.
169 void resize();
170
171 // Accumulate statistics across all tlabs before gc
172 static void accumulate_statistics_before_gc();
173
174 void fill(HeapWord* start, HeapWord* top, size_t new_size);
175 void initialize();
176
177 void set_back_allocation_end();
178 void set_sample_end();
179
180 static size_t refill_waste_limit_increment() { return TLABWasteIncrement; }
181
182 template <typename T> void addresses_do(T f) {
183 f(&_start);
184 f(&_top);
185 f(&_pf_top);
186 f(&_end);
187 f(&_allocation_end);
188 }
189
190 // Code generation support
191 static ByteSize start_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _start); }
192 static ByteSize end_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _end); }
193 static ByteSize top_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _top); }
194 static ByteSize pf_top_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _pf_top); }
195
196 void verify();
197 };
198
199 class GlobalTLABStats: public CHeapObj<mtThread> {
200 private:
201
202 // Accumulate perfdata in private variables because
203 // PerfData should be write-only for security reasons
204 // (see perfData.hpp)
205 unsigned _allocating_threads;
206 unsigned _total_refills;
207 unsigned _max_refills;
208 size_t _total_allocation;
209 size_t _total_gc_waste;
210 size_t _max_gc_waste;
211 size_t _total_slow_refill_waste;
212 size_t _max_slow_refill_waste;
213 size_t _total_fast_refill_waste;
214 size_t _max_fast_refill_waste;
215 unsigned _total_slow_allocations;
216 unsigned _max_slow_allocations;
217
218 PerfVariable* _perf_allocating_threads;
219 PerfVariable* _perf_total_refills;
220 PerfVariable* _perf_max_refills;
221 PerfVariable* _perf_allocation;
222 PerfVariable* _perf_gc_waste;
223 PerfVariable* _perf_max_gc_waste;
224 PerfVariable* _perf_slow_refill_waste;
225 PerfVariable* _perf_max_slow_refill_waste;
226 PerfVariable* _perf_fast_refill_waste;
227 PerfVariable* _perf_max_fast_refill_waste;
228 PerfVariable* _perf_slow_allocations;
229 PerfVariable* _perf_max_slow_allocations;
230
231 AdaptiveWeightedAverage _allocating_threads_avg;
232
233 public:
234 GlobalTLABStats();
235
236 // Initialize all counters
237 void initialize();
238
239 // Write all perf counters to the perf_counters
240 void publish();
241
242 void print();
243
244 // Accessors
245 unsigned allocating_threads_avg() {
246 return MAX2((unsigned)(_allocating_threads_avg.average() + 0.5), 1U);
247 }
248
249 size_t allocation() {
250 return _total_allocation;
251 }
252
253 // Update methods
254
255 void update_allocating_threads() {
256 _allocating_threads++;
257 }
258 void update_number_of_refills(unsigned value) {
259 _total_refills += value;
260 _max_refills = MAX2(_max_refills, value);
261 }
262 void update_allocation(size_t value) {
263 _total_allocation += value;
264 }
265 void update_gc_waste(size_t value) {
266 _total_gc_waste += value;
267 _max_gc_waste = MAX2(_max_gc_waste, value);
268 }
269 void update_fast_refill_waste(size_t value) {
270 _total_fast_refill_waste += value;
271 _max_fast_refill_waste = MAX2(_max_fast_refill_waste, value);
272 }
273 void update_slow_refill_waste(size_t value) {
274 _total_slow_refill_waste += value;
275 _max_slow_refill_waste = MAX2(_max_slow_refill_waste, value);
276 }
277 void update_slow_allocations(unsigned value) {
278 _total_slow_allocations += value;
279 _max_slow_allocations = MAX2(_max_slow_allocations, value);
280 }
281 };
282
283 #endif // SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP