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 fast_path_end/allocation_end
42 // fields.
43 // allocation_end contains the real end of the tlab allocation,
44 // whereas fast_path_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* _fast_path_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
71 AdaptiveWeightedAverage _allocation_fraction; // fraction of eden allocated in tlabs
72
73 void accumulate_statistics();
74 void initialize_statistics();
75
76 void set_start(HeapWord* start) { _start = start; }
77 void set_fast_path_end(HeapWord* fast_path_end){ _fast_path_end = fast_path_end; }
78 void set_allocation_end(HeapWord* ptr) { _allocation_end = ptr; }
79 void set_top(HeapWord* top) { _top = top; }
80 void set_pf_top(HeapWord* pf_top) { _pf_top = pf_top; }
81 void set_desired_size(size_t desired_size) { _desired_size = desired_size; }
82 void set_refill_waste_limit(size_t waste) { _refill_waste_limit = waste; }
83
84 size_t initial_refill_waste_limit() { return desired_size() / TLABRefillWasteFraction; }
85
86 static int target_refills() { return _target_refills; }
87 size_t initial_desired_size();
88
89 size_t remaining();
90
91 // Make parsable and release it.
92 void reset();
93
94 // Resize based on amount of allocation, etc.
95 void resize();
96
97 void invariants() const { assert(top() >= start() && top() <= fast_path_end(), "invalid tlab"); }
98
99 void initialize(HeapWord* start, HeapWord* top, HeapWord* end);
100
101 void print_stats(const char* tag);
102
103 Thread* myThread();
104
105 // statistics
106
107 int number_of_refills() const { return _number_of_refills; }
108 int fast_refill_waste() const { return _fast_refill_waste; }
109 int slow_refill_waste() const { return _slow_refill_waste; }
110 int gc_waste() const { return _gc_waste; }
111 int slow_allocations() const { return _slow_allocations; }
112
113 static GlobalTLABStats* _global_stats;
114 static GlobalTLABStats* global_stats() { return _global_stats; }
115
116 public:
117 ThreadLocalAllocBuffer() : _allocation_fraction(TLABAllocationWeight), _allocated_before_last_gc(0) {
118 // do nothing. tlabs must be inited by initialize() calls
119 }
120
121 static size_t min_size() { return align_object_size(MinTLABSize / HeapWordSize) + alignment_reserve(); }
122 static size_t max_size() { assert(_max_size != 0, "max_size not set up"); return _max_size; }
123 static size_t max_size_in_bytes() { return max_size() * BytesPerWord; }
124 static void set_max_size(size_t max_size) { _max_size = max_size; }
125
126 HeapWord* start() const { return _start; }
127 HeapWord* fast_path_end() const { return _fast_path_end; }
128 HeapWord* top() const { return _top; }
129 HeapWord* hard_end();
130 HeapWord* pf_top() const { return _pf_top; }
131 size_t desired_size() const { return _desired_size; }
132 size_t used() const { return pointer_delta(top(), start()); }
133 size_t used_bytes() const { return pointer_delta(top(), start(), 1); }
134 size_t free() const { return pointer_delta(fast_path_end(), top()); }
135 // Don't discard tlab if remaining space is larger than this.
136 size_t refill_waste_limit() const { return _refill_waste_limit; }
137 size_t bytes_since_last_sample_point() const { return _bytes_since_last_sample_point; }
138
139 // Allocate size HeapWords. The memory is NOT initialized to zero.
140 inline HeapWord* allocate(size_t size);
141 HeapWord* allocate_sampled_object(size_t size);
142
143 // Reserve space at the end of TLAB
144 static size_t end_reserve() {
145 int reserve_size = typeArrayOopDesc::header_size(T_INT);
146 return MAX2(reserve_size, _reserve_for_allocation_prefetch);
147 }
148 static size_t alignment_reserve() { return align_object_size(end_reserve()); }
149 static size_t alignment_reserve_in_bytes() { return alignment_reserve() * HeapWordSize; }
150
151 // Return tlab size or remaining space in eden such that the
152 // space is large enough to hold obj_size and necessary fill space.
153 // Otherwise return 0;
154 inline size_t compute_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 // Accumulate statistics across all tlabs before gc
169 static void accumulate_statistics_before_gc();
170
171 // Resize tlabs for all threads
172 static void resize_all_tlabs();
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 // Code generation support
183 static ByteSize start_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _start); }
184 static ByteSize fast_path_end_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _fast_path_end ); }
185 static ByteSize top_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _top ); }
186 static ByteSize pf_top_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _pf_top ); }
187 static ByteSize size_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _desired_size ); }
188 static ByteSize refill_waste_limit_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _refill_waste_limit ); }
189
190 static ByteSize number_of_refills_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _number_of_refills ); }
191 static ByteSize fast_refill_waste_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _fast_refill_waste ); }
192 static ByteSize slow_allocations_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _slow_allocations ); }
193
194 void verify();
195 };
196
197 class GlobalTLABStats: public CHeapObj<mtThread> {
198 private:
199
200 // Accumulate perfdata in private variables because
201 // PerfData should be write-only for security reasons
202 // (see perfData.hpp)
203 unsigned _allocating_threads;
204 unsigned _total_refills;
205 unsigned _max_refills;
206 size_t _total_allocation;
207 size_t _total_gc_waste;
208 size_t _max_gc_waste;
209 size_t _total_slow_refill_waste;
210 size_t _max_slow_refill_waste;
211 size_t _total_fast_refill_waste;
212 size_t _max_fast_refill_waste;
213 unsigned _total_slow_allocations;
214 unsigned _max_slow_allocations;
215
216 PerfVariable* _perf_allocating_threads;
217 PerfVariable* _perf_total_refills;
218 PerfVariable* _perf_max_refills;
219 PerfVariable* _perf_allocation;
220 PerfVariable* _perf_gc_waste;
221 PerfVariable* _perf_max_gc_waste;
222 PerfVariable* _perf_slow_refill_waste;
223 PerfVariable* _perf_max_slow_refill_waste;
224 PerfVariable* _perf_fast_refill_waste;
225 PerfVariable* _perf_max_fast_refill_waste;
226 PerfVariable* _perf_slow_allocations;
227 PerfVariable* _perf_max_slow_allocations;
228
229 AdaptiveWeightedAverage _allocating_threads_avg;
230
231 public:
232 GlobalTLABStats();
233
234 // Initialize all counters
235 void initialize();
236
237 // Write all perf counters to the perf_counters
238 void publish();
239
240 void print();
241
242 // Accessors
243 unsigned allocating_threads_avg() {
244 return MAX2((unsigned)(_allocating_threads_avg.average() + 0.5), 1U);
245 }
246
247 size_t allocation() {
248 return _total_allocation;
249 }
250
251 // Update methods
252
253 void update_allocating_threads() {
254 _allocating_threads++;
255 }
256 void update_number_of_refills(unsigned value) {
257 _total_refills += value;
258 _max_refills = MAX2(_max_refills, value);
259 }
260 void update_allocation(size_t value) {
261 _total_allocation += value;
262 }
263 void update_gc_waste(size_t value) {
264 _total_gc_waste += value;
265 _max_gc_waste = MAX2(_max_gc_waste, value);
266 }
267 void update_fast_refill_waste(size_t value) {
268 _total_fast_refill_waste += value;
269 _max_fast_refill_waste = MAX2(_max_fast_refill_waste, value);
270 }
271 void update_slow_refill_waste(size_t value) {
272 _total_slow_refill_waste += value;
273 _max_slow_refill_waste = MAX2(_max_slow_refill_waste, value);
274 }
275 void update_slow_allocations(unsigned value) {
276 _total_slow_allocations += value;
277 _max_slow_allocations = MAX2(_max_slow_allocations, value);
278 }
279 };
280
281 #endif // SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP