1 /*
2 * Copyright (c) 2001, 2019, 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_GC_SHARED_PTRQUEUE_HPP
26 #define SHARE_GC_SHARED_PTRQUEUE_HPP
27
28 #include "memory/padded.hpp"
29 #include "utilities/align.hpp"
30 #include "utilities/debug.hpp"
31 #include "utilities/lockFreeStack.hpp"
32 #include "utilities/sizes.hpp"
33
34 class Mutex;
35 class Monitor;
36
37 // There are various techniques that require threads to be able to log
38 // addresses. For example, a generational write barrier might log
39 // the addresses of modified old-generation objects. This type supports
40 // this operation.
41
42 class BufferNode;
43 class PtrQueueSet;
44 class PtrQueue {
45 friend class VMStructs;
46
47 // Noncopyable - not defined.
48 PtrQueue(const PtrQueue&);
49 PtrQueue& operator=(const PtrQueue&);
50
51 // The ptr queue set to which this queue belongs.
52 PtrQueueSet* const _qset;
53
54 // Whether updates should be logged.
55 bool _active;
56
57 // If true, the queue is permanent, and doesn't need to deallocate
58 // its buffer in the destructor (since that obtains a lock which may not
59 // be legally locked by then.
60 const bool _permanent;
61
62 // The (byte) index at which an object was last enqueued. Starts at
63 // capacity_in_bytes (indicating an empty buffer) and goes towards zero.
64 // Value is always pointer-size aligned.
65 size_t _index;
66
67 // Size of the current buffer, in bytes.
68 // Value is always pointer-size aligned.
69 size_t _capacity_in_bytes;
70
71 static const size_t _element_size = sizeof(void*);
72
73 // Get the capacity, in bytes. The capacity must have been set.
74 size_t capacity_in_bytes() const {
75 assert(_capacity_in_bytes > 0, "capacity not set");
76 return _capacity_in_bytes;
77 }
78
79 void set_capacity(size_t entries) {
80 size_t byte_capacity = index_to_byte_index(entries);
81 assert(_capacity_in_bytes == 0 || _capacity_in_bytes == byte_capacity,
82 "changing capacity " SIZE_FORMAT " -> " SIZE_FORMAT,
83 _capacity_in_bytes, byte_capacity);
84 _capacity_in_bytes = byte_capacity;
85 }
86
87 static size_t byte_index_to_index(size_t ind) {
88 assert(is_aligned(ind, _element_size), "precondition");
89 return ind / _element_size;
90 }
91
92 static size_t index_to_byte_index(size_t ind) {
93 return ind * _element_size;
94 }
95
96 protected:
97 // The buffer.
98 void** _buf;
99
100 size_t index() const {
101 return byte_index_to_index(_index);
102 }
103
104 void set_index(size_t new_index) {
105 size_t byte_index = index_to_byte_index(new_index);
106 assert(byte_index <= capacity_in_bytes(), "precondition");
107 _index = byte_index;
108 }
109
110 size_t capacity() const {
111 return byte_index_to_index(capacity_in_bytes());
112 }
113
114 // If there is a lock associated with this buffer, this is that lock.
115 Mutex* _lock;
116
117 PtrQueueSet* qset() { return _qset; }
118 bool is_permanent() const { return _permanent; }
119
120 // Process queue entries and release resources.
121 void flush_impl();
122
123 // Initialize this queue to contain a null buffer, and be part of the
124 // given PtrQueueSet.
125 PtrQueue(PtrQueueSet* qset, bool permanent = false, bool active = false);
126
127 // Requires queue flushed or permanent.
128 ~PtrQueue();
129
130 public:
131
132 // Associate a lock with a ptr queue.
133 void set_lock(Mutex* lock) { _lock = lock; }
134
135 // Forcibly set empty.
136 void reset() {
137 if (_buf != NULL) {
138 _index = capacity_in_bytes();
139 }
140 }
141
142 void enqueue(volatile void* ptr) {
143 enqueue((void*)(ptr));
144 }
145
146 // Enqueues the given "obj".
147 void enqueue(void* ptr) {
148 if (!_active) return;
149 else enqueue_known_active(ptr);
150 }
151
152 // This method is called when we're doing the zero index handling
153 // and gives a chance to the queues to do any pre-enqueueing
154 // processing they might want to do on the buffer. It should return
155 // true if the buffer should be enqueued, or false if enough
156 // entries were cleared from it so that it can be re-used. It should
157 // not return false if the buffer is still full (otherwise we can
158 // get into an infinite loop).
159 virtual bool should_enqueue_buffer() { return true; }
160 void handle_zero_index();
161
162 void enqueue_known_active(void* ptr);
163
164 // Return the size of the in-use region.
165 size_t size() const {
166 size_t result = 0;
167 if (_buf != NULL) {
168 assert(_index <= capacity_in_bytes(), "Invariant");
169 result = byte_index_to_index(capacity_in_bytes() - _index);
170 }
171 return result;
172 }
173
174 bool is_empty() const {
175 return _buf == NULL || capacity_in_bytes() == _index;
176 }
177
178 // Set the "active" property of the queue to "b". An enqueue to an
179 // inactive thread is a no-op. Setting a queue to inactive resets its
180 // log to the empty state.
181 void set_active(bool b) {
182 _active = b;
183 if (!b && _buf != NULL) {
184 reset();
185 } else if (b && _buf != NULL) {
186 assert(index() == capacity(),
187 "invariant: queues are empty when activated.");
188 }
189 }
190
191 bool is_active() const { return _active; }
192
193 // To support compiler.
194
195 protected:
196 template<typename Derived>
197 static ByteSize byte_offset_of_index() {
198 return byte_offset_of(Derived, _index);
199 }
200
201 static ByteSize byte_width_of_index() { return in_ByteSize(sizeof(size_t)); }
202
203 template<typename Derived>
204 static ByteSize byte_offset_of_buf() {
205 return byte_offset_of(Derived, _buf);
206 }
207
208 static ByteSize byte_width_of_buf() { return in_ByteSize(_element_size); }
209
210 template<typename Derived>
211 static ByteSize byte_offset_of_active() {
212 return byte_offset_of(Derived, _active);
213 }
214
215 static ByteSize byte_width_of_active() { return in_ByteSize(sizeof(bool)); }
216
217 };
218
219 class BufferNode {
220 size_t _index;
221 BufferNode* volatile _next;
222 void* _buffer[1]; // Pseudo flexible array member.
223
224 BufferNode() : _index(0), _next(NULL) { }
225 ~BufferNode() { }
226
227 static size_t buffer_offset() {
228 return offset_of(BufferNode, _buffer);
229 }
230
231 static BufferNode* volatile* next_ptr(BufferNode& bn) { return &bn._next; }
232
233 AIX_ONLY(public:) // xlC 12 on AIX doesn't implement C++ DR45.
234 // Allocate a new BufferNode with the "buffer" having size elements.
235 static BufferNode* allocate(size_t size);
236
237 // Free a BufferNode.
238 static void deallocate(BufferNode* node);
239
240 public:
241 typedef LockFreeStack<BufferNode, &next_ptr> Stack;
242
243 BufferNode* next() const { return _next; }
244 void set_next(BufferNode* n) { _next = n; }
245 size_t index() const { return _index; }
246 void set_index(size_t i) { _index = i; }
247
248 // Return the BufferNode containing the buffer, after setting its index.
249 static BufferNode* make_node_from_buffer(void** buffer, size_t index) {
250 BufferNode* node =
251 reinterpret_cast<BufferNode*>(
252 reinterpret_cast<char*>(buffer) - buffer_offset());
253 node->set_index(index);
254 return node;
255 }
256
257 // Return the buffer for node.
258 static void** make_buffer_from_node(BufferNode *node) {
259 // &_buffer[0] might lead to index out of bounds warnings.
260 return reinterpret_cast<void**>(
261 reinterpret_cast<char*>(node) + buffer_offset());
262 }
263
264 class Allocator; // Free-list based allocator.
265 class TestSupport; // Unit test support.
266 };
267
268 // Allocation is based on a lock-free free list of nodes, linked through
269 // BufferNode::_next (see BufferNode::Stack). To solve the ABA problem,
270 // popping a node from the free list is performed within a GlobalCounter
271 // critical section, and pushing nodes onto the free list is done after
272 // a GlobalCounter synchronization associated with the nodes to be pushed.
273 // This is documented behavior so that other parts of the node life-cycle
274 // can depend on and make use of it too.
275 class BufferNode::Allocator {
276 friend class TestSupport;
277
278 // Since we don't expect many instances, and measured >15% speedup
279 // on stress gtest, padding seems like a good tradeoff here.
280 #define DECLARE_PADDED_MEMBER(Id, Type, Name) \
281 Type Name; DEFINE_PAD_MINUS_SIZE(Id, DEFAULT_CACHE_LINE_SIZE, sizeof(Type))
282
283 const size_t _buffer_size;
284 char _name[DEFAULT_CACHE_LINE_SIZE - sizeof(size_t)]; // Use name as padding.
285 DECLARE_PADDED_MEMBER(1, Stack, _pending_list);
286 DECLARE_PADDED_MEMBER(2, Stack, _free_list);
287 DECLARE_PADDED_MEMBER(3, volatile size_t, _pending_count);
288 DECLARE_PADDED_MEMBER(4, volatile size_t, _free_count);
289 DECLARE_PADDED_MEMBER(5, volatile bool, _transfer_lock);
290
291 #undef DECLARE_PADDED_MEMBER
292
293 void delete_list(BufferNode* list);
294 bool try_transfer_pending();
295
296 public:
297 Allocator(const char* name, size_t buffer_size);
298 ~Allocator();
299
300 const char* name() const { return _name; }
301 size_t buffer_size() const { return _buffer_size; }
302 size_t free_count() const;
303 BufferNode* allocate();
304 void release(BufferNode* node);
305
306 // Deallocate some of the available buffers. remove_goal is the target
307 // number to remove. Returns the number actually deallocated, which may
308 // be less than the goal if there were fewer available.
309 size_t reduce_free_list(size_t remove_goal);
310 };
311
312 // A PtrQueueSet represents resources common to a set of pointer queues.
313 // In particular, the individual queues allocate buffers from this shared
314 // set, and return completed buffers to the set.
315 class PtrQueueSet {
316 BufferNode::Allocator* _allocator;
317
318 Monitor* _cbl_mon; // Protects the fields below.
319 BufferNode* _completed_buffers_head;
320 BufferNode* _completed_buffers_tail;
321 size_t _n_completed_buffers;
322
323 size_t _process_completed_buffers_threshold;
324 volatile bool _process_completed_buffers;
325
326 // If true, notify_all on _cbl_mon when the threshold is reached.
327 bool _notify_when_complete;
328
329 // Maximum number of elements allowed on completed queue: after that,
330 // enqueuer does the work itself.
331 size_t _max_completed_buffers;
332 size_t _completed_buffers_padding;
333
334 void assert_completed_buffers_list_len_correct_locked() NOT_DEBUG_RETURN;
335
336 protected:
337 bool _all_active;
338
339 // A mutator thread does the the work of processing a buffer.
340 // Returns "true" iff the work is complete (and the buffer may be
341 // deallocated).
342 virtual bool mut_process_buffer(BufferNode* node) {
343 ShouldNotReachHere();
344 return false;
345 }
346
347 // Create an empty ptr queue set.
348 PtrQueueSet(bool notify_when_complete = false);
349 ~PtrQueueSet();
350
351 // Because of init-order concerns, we can't pass these as constructor
352 // arguments.
353 void initialize(Monitor* cbl_mon, BufferNode::Allocator* allocator);
354
355 // For (unlocked!) iteration over the completed buffers.
356 BufferNode* completed_buffers_head() const { return _completed_buffers_head; }
357
358 // Deallocate all of the completed buffers.
359 void abandon_completed_buffers();
360
361 public:
362
363 // Return the buffer for a BufferNode of size buffer_size().
364 void** allocate_buffer();
365
366 // Return an empty buffer to the free list. The node is required
367 // to have been allocated with a size of buffer_size().
368 void deallocate_buffer(BufferNode* node);
369
370 // A completed buffer is a buffer the mutator is finished with, and
371 // is ready to be processed by the collector. It need not be full.
372
373 // Adds node to the completed buffer list.
374 void enqueue_completed_buffer(BufferNode* node);
375
376 // If the number of completed buffers is > stop_at, then remove and
377 // return a completed buffer from the list. Otherwise, return NULL.
378 BufferNode* get_completed_buffer(size_t stop_at = 0);
379
380 // To be invoked by the mutator.
381 bool process_or_enqueue_completed_buffer(BufferNode* node);
382
383 bool process_completed_buffers() { return _process_completed_buffers; }
384 void set_process_completed_buffers(bool x) { _process_completed_buffers = x; }
385
386 bool is_active() { return _all_active; }
387
388 size_t buffer_size() const {
389 return _allocator->buffer_size();
390 }
391
392 // Get/Set the number of completed buffers that triggers log processing.
393 // Log processing should be done when the number of buffers exceeds the
394 // threshold.
395 void set_process_completed_buffers_threshold(size_t sz) {
396 _process_completed_buffers_threshold = sz;
397 }
398 size_t process_completed_buffers_threshold() const {
399 return _process_completed_buffers_threshold;
400 }
401 static const size_t ProcessCompletedBuffersThresholdNever = ~size_t(0);
402
403 size_t completed_buffers_num() const { return _n_completed_buffers; }
404
405 void merge_bufferlists(PtrQueueSet* src);
406
407 void set_max_completed_buffers(size_t m) {
408 _max_completed_buffers = m;
409 }
410 size_t max_completed_buffers() const {
411 return _max_completed_buffers;
412 }
413 static const size_t MaxCompletedBuffersUnlimited = ~size_t(0);
414
415 void set_completed_buffers_padding(size_t padding) {
416 _completed_buffers_padding = padding;
417 }
418 size_t completed_buffers_padding() const {
419 return _completed_buffers_padding;
420 }
421
422 // Notify the consumer if the number of buffers crossed the threshold
423 void notify_if_necessary();
424 };
425
426 #endif // SHARE_GC_SHARED_PTRQUEUE_HPP