1 /*
2 * Copyright (c) 2001, 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 #include "precompiled.hpp"
26 #include "gc/shared/taskqueue.hpp"
27 #include "gc/shared/owstTaskTerminator.hpp"
28 #include "oops/oop.inline.hpp"
29 #include "logging/log.hpp"
30 #include "runtime/atomic.hpp"
31 #include "runtime/os.hpp"
32 #include "runtime/thread.inline.hpp"
33 #include "utilities/debug.hpp"
34 #include "utilities/stack.inline.hpp"
35
36 #ifdef TRACESPINNING
37 uint ParallelTaskTerminator::_total_yields = 0;
38 uint ParallelTaskTerminator::_total_spins = 0;
39 uint ParallelTaskTerminator::_total_peeks = 0;
40 #endif
41
42 #if TASKQUEUE_STATS
43 const char * const TaskQueueStats::_names[last_stat_id] = {
44 "qpush", "qpop", "qpop-s", "qattempt", "qsteal", "opush", "omax"
45 };
46
47 TaskQueueStats & TaskQueueStats::operator +=(const TaskQueueStats & addend)
48 {
49 for (unsigned int i = 0; i < last_stat_id; ++i) {
50 _stats[i] += addend._stats[i];
51 }
52 return *this;
53 }
54
55 void TaskQueueStats::print_header(unsigned int line, outputStream* const stream,
56 unsigned int width)
57 {
58 // Use a width w: 1 <= w <= max_width
59 const unsigned int max_width = 40;
60 const unsigned int w = MAX2(MIN2(width, max_width), 1U);
61
62 if (line == 0) { // spaces equal in width to the header
63 const unsigned int hdr_width = w * last_stat_id + last_stat_id - 1;
64 stream->print("%*s", hdr_width, " ");
65 } else if (line == 1) { // labels
66 stream->print("%*s", w, _names[0]);
67 for (unsigned int i = 1; i < last_stat_id; ++i) {
68 stream->print(" %*s", w, _names[i]);
69 }
70 } else if (line == 2) { // dashed lines
71 char dashes[max_width + 1];
72 memset(dashes, '-', w);
73 dashes[w] = '\0';
74 stream->print("%s", dashes);
75 for (unsigned int i = 1; i < last_stat_id; ++i) {
76 stream->print(" %s", dashes);
77 }
78 }
79 }
80
81 void TaskQueueStats::print(outputStream* stream, unsigned int width) const
82 {
83 #define FMT SIZE_FORMAT_W(*)
84 stream->print(FMT, width, _stats[0]);
85 for (unsigned int i = 1; i < last_stat_id; ++i) {
86 stream->print(" " FMT, width, _stats[i]);
87 }
88 #undef FMT
89 }
90
91 #ifdef ASSERT
92 // Invariants which should hold after a TaskQueue has been emptied and is
93 // quiescent; they do not hold at arbitrary times.
94 void TaskQueueStats::verify() const
95 {
96 assert(get(push) == get(pop) + get(steal),
97 "push=" SIZE_FORMAT " pop=" SIZE_FORMAT " steal=" SIZE_FORMAT,
98 get(push), get(pop), get(steal));
99 assert(get(pop_slow) <= get(pop),
100 "pop_slow=" SIZE_FORMAT " pop=" SIZE_FORMAT,
101 get(pop_slow), get(pop));
102 assert(get(steal) <= get(steal_attempt),
103 "steal=" SIZE_FORMAT " steal_attempt=" SIZE_FORMAT,
104 get(steal), get(steal_attempt));
105 assert(get(overflow) == 0 || get(push) != 0,
106 "overflow=" SIZE_FORMAT " push=" SIZE_FORMAT,
107 get(overflow), get(push));
108 assert(get(overflow_max_len) == 0 || get(overflow) != 0,
109 "overflow_max_len=" SIZE_FORMAT " overflow=" SIZE_FORMAT,
110 get(overflow_max_len), get(overflow));
111 }
112 #endif // ASSERT
113 #endif // TASKQUEUE_STATS
114
115 ParallelTaskTerminator::
116 ParallelTaskTerminator(uint n_threads, TaskQueueSetSuper* queue_set) :
117 _n_threads(n_threads),
118 _queue_set(queue_set),
119 _offered_termination(0) {}
120
121 bool ParallelTaskTerminator::peek_in_queue_set() {
122 return _queue_set->peek();
123 }
124
125 void ParallelTaskTerminator::yield() {
126 assert(_offered_termination <= _n_threads, "Invariant");
127 os::naked_yield();
128 }
129
130 void ParallelTaskTerminator::sleep(uint millis) {
131 assert(_offered_termination <= _n_threads, "Invariant");
132 os::sleep(Thread::current(), millis, false);
133 }
134
135 bool
136 ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) {
137 assert(_n_threads > 0, "Initialization is incorrect");
138 assert(_offered_termination < _n_threads, "Invariant");
139 Atomic::inc(&_offered_termination);
140
141 uint yield_count = 0;
142 // Number of hard spin loops done since last yield
143 uint hard_spin_count = 0;
144 // Number of iterations in the hard spin loop.
145 uint hard_spin_limit = WorkStealingHardSpins;
146
147 // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done.
148 // If it is greater than 0, then start with a small number
149 // of spins and increase number with each turn at spinning until
150 // the count of hard spins exceeds WorkStealingSpinToYieldRatio.
151 // Then do a yield() call and start spinning afresh.
152 if (WorkStealingSpinToYieldRatio > 0) {
153 hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio;
154 hard_spin_limit = MAX2(hard_spin_limit, 1U);
155 }
156 // Remember the initial spin limit.
157 uint hard_spin_start = hard_spin_limit;
158
159 // Loop waiting for all threads to offer termination or
160 // more work.
161 while (true) {
162 assert(_offered_termination <= _n_threads, "Invariant");
163 // Are all threads offering termination?
164 if (_offered_termination == _n_threads) {
165 return true;
166 } else {
167 // Look for more work.
168 // Periodically sleep() instead of yield() to give threads
169 // waiting on the cores the chance to grab this code
170 if (yield_count <= WorkStealingYieldsBeforeSleep) {
171 // Do a yield or hardspin. For purposes of deciding whether
172 // to sleep, count this as a yield.
173 yield_count++;
174
175 // Periodically call yield() instead spinning
176 // After WorkStealingSpinToYieldRatio spins, do a yield() call
177 // and reset the counts and starting limit.
178 if (hard_spin_count > WorkStealingSpinToYieldRatio) {
179 yield();
180 hard_spin_count = 0;
181 hard_spin_limit = hard_spin_start;
182 #ifdef TRACESPINNING
183 _total_yields++;
184 #endif
185 } else {
186 // Hard spin this time
187 // Increase the hard spinning period but only up to a limit.
188 hard_spin_limit = MIN2(2*hard_spin_limit,
189 (uint) WorkStealingHardSpins);
190 for (uint j = 0; j < hard_spin_limit; j++) {
191 SpinPause();
192 }
193 hard_spin_count++;
194 #ifdef TRACESPINNING
195 _total_spins++;
196 #endif
197 }
198 } else {
199 log_develop_trace(gc, task)("ParallelTaskTerminator::offer_termination() thread " PTR_FORMAT " sleeps after %u yields",
200 p2i(Thread::current()), yield_count);
201 yield_count = 0;
202 // A sleep will cause this processor to seek work on another processor's
203 // runqueue, if it has nothing else to run (as opposed to the yield
204 // which may only move the thread to the end of the this processor's
205 // runqueue).
206 sleep(WorkStealingSleepMillis);
207 }
208
209 #ifdef TRACESPINNING
210 _total_peeks++;
211 #endif
212 if (peek_in_queue_set() ||
213 (terminator != NULL && terminator->should_exit_termination())) {
214 Atomic::dec(&_offered_termination);
215 assert(_offered_termination < _n_threads, "Invariant");
216 return false;
217 }
218 }
219 }
220 }
221
222 #ifdef TRACESPINNING
223 void ParallelTaskTerminator::print_termination_counts() {
224 log_trace(gc, task)("ParallelTaskTerminator Total yields: %u"
225 " Total spins: %u Total peeks: %u",
226 total_yields(),
227 total_spins(),
228 total_peeks());
229 }
230 #endif
231
232 void ParallelTaskTerminator::reset_for_reuse() {
233 if (_offered_termination != 0) {
234 assert(_offered_termination == _n_threads,
235 "Terminator may still be in use");
236 _offered_termination = 0;
237 }
238 }
239
240 #ifdef ASSERT
241 bool ObjArrayTask::is_valid() const {
242 return _obj != NULL && _obj->is_objArray() && _index >= 0 &&
243 _index < objArrayOop(_obj)->length();
244 }
245 #endif // ASSERT
246
247 void ParallelTaskTerminator::reset_for_reuse(uint n_threads) {
248 reset_for_reuse();
249 _n_threads = n_threads;
250 }
251
252 TaskTerminator::TaskTerminator(uint n_threads, TaskQueueSetSuper* queue_set) :
253 _terminator(UseOWSTTaskTerminator ? new OWSTTaskTerminator(n_threads, queue_set)
254 : new ParallelTaskTerminator(n_threads, queue_set)) {
255 }
256
257 TaskTerminator::~TaskTerminator() {
258 if (_terminator != NULL) {
259 delete _terminator;
260 }
261 }
262
263 // Move assignment
264 TaskTerminator& TaskTerminator::operator=(const TaskTerminator& o) {
265 if (_terminator != NULL) {
266 delete _terminator;
267 }
268 _terminator = o.terminator();
269 const_cast<TaskTerminator&>(o)._terminator = NULL;
270 return *this;
271 }
272