1 /* |
|
2 * Copyright (c) 2005, 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 #include "precompiled.hpp" |
|
26 #include "gc/cms/yieldingWorkgroup.hpp" |
|
27 #include "gc/shared/gcId.hpp" |
|
28 #include "utilities/macros.hpp" |
|
29 |
|
30 YieldingFlexibleGangWorker::YieldingFlexibleGangWorker(YieldingFlexibleWorkGang* gang, int id) |
|
31 : AbstractGangWorker(gang, id) {} |
|
32 |
|
33 YieldingFlexibleWorkGang::YieldingFlexibleWorkGang( |
|
34 const char* name, uint workers, bool are_GC_task_threads) : |
|
35 AbstractWorkGang(name, workers, are_GC_task_threads, false), |
|
36 _yielded_workers(0), |
|
37 _started_workers(0), |
|
38 _finished_workers(0), |
|
39 _sequence_number(0), |
|
40 _task(NULL) { |
|
41 |
|
42 // Other initialization. |
|
43 _monitor = new Monitor(/* priority */ Mutex::leaf, |
|
44 /* name */ "WorkGroup monitor", |
|
45 /* allow_vm_block */ are_GC_task_threads, |
|
46 Monitor::_safepoint_check_never); |
|
47 |
|
48 assert(monitor() != NULL, "Failed to allocate monitor"); |
|
49 } |
|
50 |
|
51 AbstractGangWorker* YieldingFlexibleWorkGang::allocate_worker(uint which) { |
|
52 return new YieldingFlexibleGangWorker(this, which); |
|
53 } |
|
54 |
|
55 void YieldingFlexibleWorkGang::internal_worker_poll(YieldingWorkData* data) const { |
|
56 assert(data != NULL, "worker data is null"); |
|
57 data->set_task(task()); |
|
58 data->set_sequence_number(sequence_number()); |
|
59 } |
|
60 |
|
61 void YieldingFlexibleWorkGang::internal_note_start() { |
|
62 assert(monitor()->owned_by_self(), "note_finish is an internal method"); |
|
63 _started_workers += 1; |
|
64 } |
|
65 |
|
66 void YieldingFlexibleWorkGang::internal_note_finish() { |
|
67 assert(monitor()->owned_by_self(), "note_finish is an internal method"); |
|
68 _finished_workers += 1; |
|
69 } |
|
70 |
|
71 // Run a task; returns when the task is done, or the workers yield, |
|
72 // or the task is aborted. |
|
73 // A task that has been yielded can be continued via this interface |
|
74 // by using the same task repeatedly as the argument to the call. |
|
75 // It is expected that the YieldingFlexibleGangTask carries the appropriate |
|
76 // continuation information used by workers to continue the task |
|
77 // from its last yield point. Thus, a completed task will return |
|
78 // immediately with no actual work having been done by the workers. |
|
79 ///////////////////// |
|
80 // Implementatiuon notes: remove before checking XXX |
|
81 /* |
|
82 Each gang is working on a task at a certain time. |
|
83 Some subset of workers may have yielded and some may |
|
84 have finished their quota of work. Until this task has |
|
85 been completed, the workers are bound to that task. |
|
86 Once the task has been completed, the gang unbounds |
|
87 itself from the task. |
|
88 |
|
89 The yielding work gang thus exports two invokation |
|
90 interfaces: run_task() and continue_task(). The |
|
91 first is used to initiate a new task and bind it |
|
92 to the workers; the second is used to continue an |
|
93 already bound task that has yielded. Upon completion |
|
94 the binding is released and a new binding may be |
|
95 created. |
|
96 |
|
97 The shape of a yielding work gang is as follows: |
|
98 |
|
99 Overseer invokes run_task(*task). |
|
100 Lock gang monitor |
|
101 Check that there is no existing binding for the gang |
|
102 If so, abort with an error |
|
103 Else, create a new binding of this gang to the given task |
|
104 Set number of active workers (as asked) |
|
105 Notify workers that work is ready to be done |
|
106 [the requisite # workers would then start up |
|
107 and do the task] |
|
108 Wait on the monitor until either |
|
109 all work is completed or the task has yielded |
|
110 -- this is normally done through |
|
111 yielded + completed == active |
|
112 [completed workers are rest to idle state by overseer?] |
|
113 return appropriate status to caller |
|
114 |
|
115 Overseer invokes continue_task(*task), |
|
116 Lock gang monitor |
|
117 Check that task is the same as current binding |
|
118 If not, abort with an error |
|
119 Else, set the number of active workers as requested? |
|
120 Notify workers that they can continue from yield points |
|
121 New workers can also start up as required |
|
122 while satisfying the constraint that |
|
123 active + yielded does not exceed required number |
|
124 Wait (as above). |
|
125 |
|
126 NOTE: In the above, for simplicity in a first iteration |
|
127 our gangs will be of fixed population and will not |
|
128 therefore be flexible work gangs, just yielding work |
|
129 gangs. Once this works well, we will in a second |
|
130 iteration.refinement introduce flexibility into |
|
131 the work gang. |
|
132 |
|
133 NOTE: we can always create a new gang per each iteration |
|
134 in order to get the flexibility, but we will for now |
|
135 desist that simplified route. |
|
136 |
|
137 */ |
|
138 ///////////////////// |
|
139 void YieldingFlexibleWorkGang::start_task(YieldingFlexibleGangTask* new_task) { |
|
140 MutexLocker ml(monitor(), Mutex::_no_safepoint_check_flag); |
|
141 assert(task() == NULL, "Gang currently tied to a task"); |
|
142 assert(new_task != NULL, "Null task"); |
|
143 // Bind task to gang |
|
144 _task = new_task; |
|
145 new_task->set_gang(this); // Establish 2-way binding to support yielding |
|
146 _sequence_number++; |
|
147 |
|
148 uint requested_size = new_task->requested_size(); |
|
149 if (requested_size != 0) { |
|
150 _active_workers = MIN2(requested_size, total_workers()); |
|
151 } else { |
|
152 _active_workers = active_workers(); |
|
153 } |
|
154 new_task->set_actual_size(_active_workers); |
|
155 new_task->set_for_termination(_active_workers); |
|
156 |
|
157 assert(_started_workers == 0, "Tabula rasa non"); |
|
158 assert(_finished_workers == 0, "Tabula rasa non"); |
|
159 assert(_yielded_workers == 0, "Tabula rasa non"); |
|
160 yielding_task()->set_status(ACTIVE); |
|
161 |
|
162 // Wake up all the workers, the first few will get to work, |
|
163 // and the rest will go back to sleep |
|
164 monitor()->notify_all(); |
|
165 wait_for_gang(); |
|
166 } |
|
167 |
|
168 void YieldingFlexibleWorkGang::wait_for_gang() { |
|
169 |
|
170 assert(monitor()->owned_by_self(), "Data race"); |
|
171 // Wait for task to complete or yield |
|
172 for (Status status = yielding_task()->status(); |
|
173 status != COMPLETED && status != YIELDED && status != ABORTED; |
|
174 status = yielding_task()->status()) { |
|
175 assert(started_workers() <= active_workers(), "invariant"); |
|
176 assert(finished_workers() <= active_workers(), "invariant"); |
|
177 assert(yielded_workers() <= active_workers(), "invariant"); |
|
178 monitor()->wait_without_safepoint_check(); |
|
179 } |
|
180 switch (yielding_task()->status()) { |
|
181 case COMPLETED: |
|
182 case ABORTED: { |
|
183 assert(finished_workers() == active_workers(), "Inconsistent status"); |
|
184 assert(yielded_workers() == 0, "Invariant"); |
|
185 reset(); // for next task; gang<->task binding released |
|
186 break; |
|
187 } |
|
188 case YIELDED: { |
|
189 assert(yielded_workers() > 0, "Invariant"); |
|
190 assert(yielded_workers() + finished_workers() == active_workers(), |
|
191 "Inconsistent counts"); |
|
192 break; |
|
193 } |
|
194 case ACTIVE: |
|
195 case INACTIVE: |
|
196 case COMPLETING: |
|
197 case YIELDING: |
|
198 case ABORTING: |
|
199 default: |
|
200 ShouldNotReachHere(); |
|
201 } |
|
202 } |
|
203 |
|
204 void YieldingFlexibleWorkGang::continue_task( |
|
205 YieldingFlexibleGangTask* gang_task) { |
|
206 |
|
207 MutexLocker ml(monitor(), Mutex::_no_safepoint_check_flag); |
|
208 assert(task() != NULL && task() == gang_task, "Incorrect usage"); |
|
209 assert(_started_workers == _active_workers, "Precondition"); |
|
210 assert(_yielded_workers > 0 && yielding_task()->status() == YIELDED, |
|
211 "Else why are we calling continue_task()"); |
|
212 // Restart the yielded gang workers |
|
213 yielding_task()->set_status(ACTIVE); |
|
214 monitor()->notify_all(); |
|
215 wait_for_gang(); |
|
216 } |
|
217 |
|
218 void YieldingFlexibleWorkGang::reset() { |
|
219 _started_workers = 0; |
|
220 _finished_workers = 0; |
|
221 yielding_task()->set_gang(NULL); |
|
222 _task = NULL; // unbind gang from task |
|
223 } |
|
224 |
|
225 void YieldingFlexibleWorkGang::yield() { |
|
226 assert(task() != NULL, "Inconsistency; should have task binding"); |
|
227 MutexLocker ml(monitor(), Mutex::_no_safepoint_check_flag); |
|
228 assert(yielded_workers() < active_workers(), "Consistency check"); |
|
229 if (yielding_task()->status() == ABORTING) { |
|
230 // Do not yield; we need to abort as soon as possible |
|
231 // XXX NOTE: This can cause a performance pathology in the |
|
232 // current implementation in Mustang, as of today, and |
|
233 // pre-Mustang in that as soon as an overflow occurs, |
|
234 // yields will not be honoured. The right way to proceed |
|
235 // of course is to fix bug # TBF, so that abort's cause |
|
236 // us to return at each potential yield point. |
|
237 return; |
|
238 } |
|
239 if (++_yielded_workers + finished_workers() == active_workers()) { |
|
240 yielding_task()->set_status(YIELDED); |
|
241 monitor()->notify_all(); |
|
242 } else { |
|
243 yielding_task()->set_status(YIELDING); |
|
244 } |
|
245 |
|
246 while (true) { |
|
247 switch (yielding_task()->status()) { |
|
248 case YIELDING: |
|
249 case YIELDED: { |
|
250 monitor()->wait_without_safepoint_check(); |
|
251 break; // from switch |
|
252 } |
|
253 case ACTIVE: |
|
254 case ABORTING: |
|
255 case COMPLETING: { |
|
256 assert(_yielded_workers > 0, "Else why am i here?"); |
|
257 _yielded_workers--; |
|
258 return; |
|
259 } |
|
260 case INACTIVE: |
|
261 case ABORTED: |
|
262 case COMPLETED: |
|
263 default: { |
|
264 ShouldNotReachHere(); |
|
265 } |
|
266 } |
|
267 } |
|
268 // Only return is from inside switch statement above |
|
269 ShouldNotReachHere(); |
|
270 } |
|
271 |
|
272 void YieldingFlexibleWorkGang::abort() { |
|
273 assert(task() != NULL, "Inconsistency; should have task binding"); |
|
274 MutexLocker ml(monitor(), Mutex::_no_safepoint_check_flag); |
|
275 assert(yielded_workers() < active_workers(), "Consistency check"); |
|
276 #ifndef PRODUCT |
|
277 switch (yielding_task()->status()) { |
|
278 // allowed states |
|
279 case ACTIVE: |
|
280 case ABORTING: |
|
281 case COMPLETING: |
|
282 case YIELDING: |
|
283 break; |
|
284 // not allowed states |
|
285 case INACTIVE: |
|
286 case ABORTED: |
|
287 case COMPLETED: |
|
288 case YIELDED: |
|
289 default: |
|
290 ShouldNotReachHere(); |
|
291 } |
|
292 #endif // !PRODUCT |
|
293 Status prev_status = yielding_task()->status(); |
|
294 yielding_task()->set_status(ABORTING); |
|
295 if (prev_status == YIELDING) { |
|
296 assert(yielded_workers() > 0, "Inconsistency"); |
|
297 // At least one thread has yielded, wake it up |
|
298 // so it can go back to waiting stations ASAP. |
|
299 monitor()->notify_all(); |
|
300 } |
|
301 } |
|
302 |
|
303 /////////////////////////////// |
|
304 // YieldingFlexibleGangTask |
|
305 /////////////////////////////// |
|
306 void YieldingFlexibleGangTask::yield() { |
|
307 assert(gang() != NULL, "No gang to signal"); |
|
308 gang()->yield(); |
|
309 } |
|
310 |
|
311 void YieldingFlexibleGangTask::abort() { |
|
312 assert(gang() != NULL, "No gang to signal"); |
|
313 gang()->abort(); |
|
314 } |
|
315 |
|
316 /////////////////////////////// |
|
317 // YieldingFlexibleGangWorker |
|
318 /////////////////////////////// |
|
319 void YieldingFlexibleGangWorker::loop() { |
|
320 int previous_sequence_number = 0; |
|
321 Monitor* gang_monitor = yf_gang()->monitor(); |
|
322 MutexLocker ml(gang_monitor, Mutex::_no_safepoint_check_flag); |
|
323 YieldingWorkData data; |
|
324 int id; |
|
325 while (true) { |
|
326 // Check if there is work to do. |
|
327 yf_gang()->internal_worker_poll(&data); |
|
328 if (data.task() != NULL && data.sequence_number() != previous_sequence_number) { |
|
329 // There is work to be done. |
|
330 // First check if we need to become active or if there |
|
331 // are already the requisite number of workers |
|
332 if (yf_gang()->started_workers() == yf_gang()->active_workers()) { |
|
333 // There are already enough workers, we do not need to |
|
334 // to run; fall through and wait on monitor. |
|
335 } else { |
|
336 // We need to pitch in and do the work. |
|
337 assert(yf_gang()->started_workers() < yf_gang()->active_workers(), |
|
338 "Unexpected state"); |
|
339 id = yf_gang()->started_workers(); |
|
340 yf_gang()->internal_note_start(); |
|
341 // Now, release the gang mutex and do the work. |
|
342 { |
|
343 MutexUnlocker mul(gang_monitor, Mutex::_no_safepoint_check_flag); |
|
344 GCIdMark gc_id_mark(data.task()->gc_id()); |
|
345 data.task()->work(id); // This might include yielding |
|
346 } |
|
347 // Reacquire monitor and note completion of this worker |
|
348 yf_gang()->internal_note_finish(); |
|
349 // Update status of task based on whether all workers have |
|
350 // finished or some have yielded |
|
351 assert(data.task() == yf_gang()->task(), "Confused task binding"); |
|
352 if (yf_gang()->finished_workers() == yf_gang()->active_workers()) { |
|
353 switch (data.yf_task()->status()) { |
|
354 case ABORTING: { |
|
355 data.yf_task()->set_status(ABORTED); |
|
356 break; |
|
357 } |
|
358 case ACTIVE: |
|
359 case COMPLETING: { |
|
360 data.yf_task()->set_status(COMPLETED); |
|
361 break; |
|
362 } |
|
363 default: |
|
364 ShouldNotReachHere(); |
|
365 } |
|
366 gang_monitor->notify_all(); // Notify overseer |
|
367 } else { // at least one worker is still working or yielded |
|
368 assert(yf_gang()->finished_workers() < yf_gang()->active_workers(), |
|
369 "Counts inconsistent"); |
|
370 switch (data.yf_task()->status()) { |
|
371 case ACTIVE: { |
|
372 // first, but not only thread to complete |
|
373 data.yf_task()->set_status(COMPLETING); |
|
374 break; |
|
375 } |
|
376 case YIELDING: { |
|
377 if (yf_gang()->finished_workers() + yf_gang()->yielded_workers() |
|
378 == yf_gang()->active_workers()) { |
|
379 data.yf_task()->set_status(YIELDED); |
|
380 gang_monitor->notify_all(); // notify overseer |
|
381 } |
|
382 break; |
|
383 } |
|
384 case ABORTING: |
|
385 case COMPLETING: { |
|
386 break; // nothing to do |
|
387 } |
|
388 default: // everything else: INACTIVE, YIELDED, ABORTED, COMPLETED |
|
389 ShouldNotReachHere(); |
|
390 } |
|
391 } |
|
392 } |
|
393 } |
|
394 // Remember the sequence number |
|
395 previous_sequence_number = data.sequence_number(); |
|
396 // Wait for more work |
|
397 gang_monitor->wait_without_safepoint_check(); |
|
398 } |
|
399 } |
|