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