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 ThreadLocalAllocStats;
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 reset_statistics();
75
76 void set_start(HeapWord* start) { _start = start; }
77 void set_end(HeapWord* end) { _end = 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 void invariants() const { assert(top() >= start() && top() <= end(), "invalid tlab"); }
95
96 void initialize(HeapWord* start, HeapWord* top, HeapWord* end);
97
98 void insert_filler();
99
100 void accumulate_and_reset_statistics(ThreadLocalAllocStats* stats);
101
102 void print_stats(const char* tag);
103
104 Thread* thread();
105
106 // statistics
107
108 int number_of_refills() const { return _number_of_refills; }
109 int fast_refill_waste() const { return _fast_refill_waste; }
110 int slow_refill_waste() const { return _slow_refill_waste; }
111 int gc_waste() const { return _gc_waste; }
112 int slow_allocations() const { return _slow_allocations; }
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.
163 void make_parsable();
164
165 // Retire an in-use tlab and optionally collect statistics.
166 void retire(ThreadLocalAllocStats* stats = NULL);
167
168 // Retire in-use tlab before allocation of a new tlab
169 void retire_before_allocation();
170
171 // Resize based on amount of allocation, etc.
172 void resize();
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 ThreadLocalAllocStats : public StackObj {
200 private:
201 static PerfVariable* _perf_allocating_threads;
202 static PerfVariable* _perf_total_refills;
203 static PerfVariable* _perf_max_refills;
204 static PerfVariable* _perf_total_allocations;
205 static PerfVariable* _perf_total_gc_waste;
206 static PerfVariable* _perf_max_gc_waste;
207 static PerfVariable* _perf_total_slow_refill_waste;
208 static PerfVariable* _perf_max_slow_refill_waste;
209 static PerfVariable* _perf_total_fast_refill_waste;
210 static PerfVariable* _perf_max_fast_refill_waste;
211 static PerfVariable* _perf_total_slow_allocations;
212 static PerfVariable* _perf_max_slow_allocations;
213
214 static AdaptiveWeightedAverage _allocating_threads_avg;
215
216 unsigned int _allocating_threads;
217 unsigned int _total_refills;
218 unsigned int _max_refills;
219 size_t _total_allocations;
220 size_t _total_gc_waste;
221 size_t _max_gc_waste;
222 size_t _total_fast_refill_waste;
223 size_t _max_fast_refill_waste;
224 size_t _total_slow_refill_waste;
225 size_t _max_slow_refill_waste;
226 unsigned int _total_slow_allocations;
227 unsigned int _max_slow_allocations;
228
229 public:
230 static void initialize();
231 static unsigned int allocating_threads_avg();
232
233 ThreadLocalAllocStats();
234
235 void update_fast_allocations(unsigned int refills,
236 size_t allocations,
237 size_t gc_waste,
238 size_t fast_refill_waste,
239 size_t slow_refill_waste);
240 void update_slow_allocations(unsigned int allocations);
241 void update(const ThreadLocalAllocStats& other);
242
243 void reset();
244 void publish();
245 };
246
247 #endif // SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP