1 | //
|
---|
2 | // Cforall Version 1.0.0 Copyright (C) 2021 University of Waterloo
|
---|
3 | //
|
---|
4 | // The contents of this file are covered under the licence agreement in the
|
---|
5 | // file "LICENCE" distributed with Cforall.
|
---|
6 | //
|
---|
7 | // locks.hfa -- PUBLIC
|
---|
8 | // Runtime locks that used with the runtime thread system.
|
---|
9 | //
|
---|
10 | // Author : Colby Alexander Parsons
|
---|
11 | // Created On : Thu Jan 21 19:46:50 2021
|
---|
12 | // Last Modified By :
|
---|
13 | // Last Modified On :
|
---|
14 | // Update Count :
|
---|
15 | //
|
---|
16 |
|
---|
17 | #pragma once
|
---|
18 |
|
---|
19 | #include <stdbool.h>
|
---|
20 | #include <stdio.h>
|
---|
21 |
|
---|
22 | #include "bits/weakso_locks.hfa"
|
---|
23 | #include "containers/queueLockFree.hfa"
|
---|
24 | #include "containers/list.hfa"
|
---|
25 |
|
---|
26 | #include "limits.hfa"
|
---|
27 | #include "thread.hfa"
|
---|
28 |
|
---|
29 | #include "time_t.hfa"
|
---|
30 | #include "time.hfa"
|
---|
31 |
|
---|
32 | //-----------------------------------------------------------------------------
|
---|
33 | // Semaphores
|
---|
34 |
|
---|
35 | // '0-nary' semaphore
|
---|
36 | // Similar to a counting semaphore except the value of one is never reached
|
---|
37 | // as a consequence, a V() that would bring the value to 1 *spins* until
|
---|
38 | // a P consumes it
|
---|
39 | struct Semaphore0nary {
|
---|
40 | __spinlock_t lock; // needed to protect
|
---|
41 | mpsc_queue(thread$) queue;
|
---|
42 | };
|
---|
43 |
|
---|
44 | static inline bool P(Semaphore0nary & this, thread$ * thrd) {
|
---|
45 | /* paranoid */ verify(!thrd`next);
|
---|
46 | /* paranoid */ verify(!(&(*thrd)`next));
|
---|
47 |
|
---|
48 | push(this.queue, thrd);
|
---|
49 | return true;
|
---|
50 | }
|
---|
51 |
|
---|
52 | static inline bool P(Semaphore0nary & this) {
|
---|
53 | thread$ * thrd = active_thread();
|
---|
54 | P(this, thrd);
|
---|
55 | park();
|
---|
56 | return true;
|
---|
57 | }
|
---|
58 |
|
---|
59 | static inline thread$ * V(Semaphore0nary & this, bool doUnpark = true) {
|
---|
60 | thread$ * next;
|
---|
61 | lock(this.lock __cfaabi_dbg_ctx2);
|
---|
62 | for (;;) {
|
---|
63 | next = pop(this.queue);
|
---|
64 | if (next) break;
|
---|
65 | Pause();
|
---|
66 | }
|
---|
67 | unlock(this.lock);
|
---|
68 |
|
---|
69 | if (doUnpark) unpark(next);
|
---|
70 | return next;
|
---|
71 | }
|
---|
72 |
|
---|
73 | // Wrapper used on top of any sempahore to avoid potential locking
|
---|
74 | struct BinaryBenaphore {
|
---|
75 | volatile ssize_t counter;
|
---|
76 | };
|
---|
77 |
|
---|
78 | static inline {
|
---|
79 | void ?{}(BinaryBenaphore & this) { this.counter = 0; }
|
---|
80 | void ?{}(BinaryBenaphore & this, zero_t) { this.counter = 0; }
|
---|
81 | void ?{}(BinaryBenaphore & this, one_t ) { this.counter = 1; }
|
---|
82 |
|
---|
83 | // returns true if no blocking needed
|
---|
84 | bool P(BinaryBenaphore & this) {
|
---|
85 | return __atomic_fetch_sub(&this.counter, 1, __ATOMIC_SEQ_CST) > 0;
|
---|
86 | }
|
---|
87 |
|
---|
88 | bool tryP(BinaryBenaphore & this) {
|
---|
89 | ssize_t c = this.counter;
|
---|
90 | /* paranoid */ verify( c > MIN );
|
---|
91 | return (c >= 1) && __atomic_compare_exchange_n(&this.counter, &c, c-1, false, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED);
|
---|
92 | }
|
---|
93 |
|
---|
94 | // returns true if notify needed
|
---|
95 | bool V(BinaryBenaphore & this) {
|
---|
96 | ssize_t c = 0;
|
---|
97 | for () {
|
---|
98 | /* paranoid */ verify( this.counter < MAX );
|
---|
99 | if (__atomic_compare_exchange_n(&this.counter, &c, c+1, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) {
|
---|
100 | if (c == 0) return true;
|
---|
101 | /* paranoid */ verify(c < 0);
|
---|
102 | return false;
|
---|
103 | } else {
|
---|
104 | if (c == 1) return true;
|
---|
105 | /* paranoid */ verify(c < 1);
|
---|
106 | Pause();
|
---|
107 | }
|
---|
108 | }
|
---|
109 | }
|
---|
110 | }
|
---|
111 |
|
---|
112 | // Binary Semaphore based on the BinaryBenaphore on top of the 0-nary Semaphore
|
---|
113 | struct ThreadBenaphore {
|
---|
114 | BinaryBenaphore ben;
|
---|
115 | Semaphore0nary sem;
|
---|
116 | };
|
---|
117 |
|
---|
118 | static inline void ?{}(ThreadBenaphore & this) {}
|
---|
119 | static inline void ?{}(ThreadBenaphore & this, zero_t) { (this.ben){ 0 }; }
|
---|
120 | static inline void ?{}(ThreadBenaphore & this, one_t ) { (this.ben){ 1 }; }
|
---|
121 |
|
---|
122 | static inline bool P(ThreadBenaphore & this) { return P(this.ben) ? false : P(this.sem); }
|
---|
123 | static inline bool tryP(ThreadBenaphore & this) { return tryP(this.ben); }
|
---|
124 | static inline bool P(ThreadBenaphore & this, bool wait) { return wait ? P(this) : tryP(this); }
|
---|
125 |
|
---|
126 | static inline thread$ * V(ThreadBenaphore & this, bool doUnpark = true) {
|
---|
127 | if (V(this.ben)) return 0p;
|
---|
128 | return V(this.sem, doUnpark);
|
---|
129 | }
|
---|
130 |
|
---|
131 | //-----------------------------------------------------------------------------
|
---|
132 | // Semaphore
|
---|
133 | struct semaphore {
|
---|
134 | __spinlock_t lock;
|
---|
135 | int count;
|
---|
136 | __queue_t(thread$) waiting;
|
---|
137 | };
|
---|
138 |
|
---|
139 | void ?{}(semaphore & this, int count = 1);
|
---|
140 | void ^?{}(semaphore & this);
|
---|
141 | bool P (semaphore & this);
|
---|
142 | bool V (semaphore & this);
|
---|
143 | bool V (semaphore & this, unsigned count);
|
---|
144 | thread$ * V (semaphore & this, bool );
|
---|
145 |
|
---|
146 | //----------
|
---|
147 | struct single_acquisition_lock {
|
---|
148 | inline blocking_lock;
|
---|
149 | };
|
---|
150 |
|
---|
151 | static inline void ?{}( single_acquisition_lock & this ) {((blocking_lock &)this){ false, false };}
|
---|
152 | static inline void ^?{}( single_acquisition_lock & this ) {}
|
---|
153 | static inline void lock ( single_acquisition_lock & this ) { lock ( (blocking_lock &)this ); }
|
---|
154 | static inline bool try_lock ( single_acquisition_lock & this ) { return try_lock( (blocking_lock &)this ); }
|
---|
155 | static inline void unlock ( single_acquisition_lock & this ) { unlock ( (blocking_lock &)this ); }
|
---|
156 | static inline size_t on_wait ( single_acquisition_lock & this ) { return on_wait ( (blocking_lock &)this ); }
|
---|
157 | static inline void on_wakeup( single_acquisition_lock & this, size_t v ) { on_wakeup ( (blocking_lock &)this, v ); }
|
---|
158 | static inline void on_notify( single_acquisition_lock & this, struct thread$ * t ) { on_notify( (blocking_lock &)this, t ); }
|
---|
159 |
|
---|
160 | //----------
|
---|
161 | struct owner_lock {
|
---|
162 | inline blocking_lock;
|
---|
163 | };
|
---|
164 |
|
---|
165 | static inline void ?{}( owner_lock & this ) {((blocking_lock &)this){ true, true };}
|
---|
166 | static inline void ^?{}( owner_lock & this ) {}
|
---|
167 | static inline void lock ( owner_lock & this ) { lock ( (blocking_lock &)this ); }
|
---|
168 | static inline bool try_lock ( owner_lock & this ) { return try_lock( (blocking_lock &)this ); }
|
---|
169 | static inline void unlock ( owner_lock & this ) { unlock ( (blocking_lock &)this ); }
|
---|
170 | static inline size_t on_wait ( owner_lock & this ) { return on_wait ( (blocking_lock &)this ); }
|
---|
171 | static inline void on_wakeup( owner_lock & this, size_t v ) { on_wakeup ( (blocking_lock &)this, v ); }
|
---|
172 | static inline void on_notify( owner_lock & this, struct thread$ * t ) { on_notify( (blocking_lock &)this, t ); }
|
---|
173 |
|
---|
174 | struct fast_lock {
|
---|
175 | thread$ * volatile owner;
|
---|
176 | ThreadBenaphore sem;
|
---|
177 | };
|
---|
178 |
|
---|
179 | static inline void ?{}(fast_lock & this) { this.owner = 0p; }
|
---|
180 |
|
---|
181 | static inline bool $try_lock(fast_lock & this, thread$ * thrd) {
|
---|
182 | thread$ * exp = 0p;
|
---|
183 | return __atomic_compare_exchange_n(&this.owner, &exp, thrd, false, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED);
|
---|
184 | }
|
---|
185 |
|
---|
186 | static inline void lock( fast_lock & this ) __attribute__((artificial));
|
---|
187 | static inline void lock( fast_lock & this ) {
|
---|
188 | thread$ * thrd = active_thread();
|
---|
189 | /* paranoid */verify(thrd != this.owner);
|
---|
190 |
|
---|
191 | for (;;) {
|
---|
192 | if ($try_lock(this, thrd)) return;
|
---|
193 | P(this.sem);
|
---|
194 | }
|
---|
195 | }
|
---|
196 |
|
---|
197 | static inline bool try_lock( fast_lock & this ) __attribute__((artificial));
|
---|
198 | static inline bool try_lock ( fast_lock & this ) {
|
---|
199 | thread$ * thrd = active_thread();
|
---|
200 | /* paranoid */ verify(thrd != this.owner);
|
---|
201 | return $try_lock(this, thrd);
|
---|
202 | }
|
---|
203 |
|
---|
204 | static inline thread$ * unlock( fast_lock & this ) __attribute__((artificial));
|
---|
205 | static inline thread$ * unlock( fast_lock & this ) {
|
---|
206 | /* paranoid */ verify(active_thread() == this.owner);
|
---|
207 |
|
---|
208 | // open 'owner' before unlocking anyone
|
---|
209 | // so new and unlocked threads don't park incorrectly.
|
---|
210 | // This may require additional fencing on ARM.
|
---|
211 | this.owner = 0p;
|
---|
212 |
|
---|
213 | return V(this.sem);
|
---|
214 | }
|
---|
215 |
|
---|
216 | static inline size_t on_wait( fast_lock & this ) { unlock(this); return 0; }
|
---|
217 | static inline void on_wakeup( fast_lock & this, size_t ) { lock(this); }
|
---|
218 | static inline void on_notify( fast_lock &, struct thread$ * t ) { unpark(t); }
|
---|
219 |
|
---|
220 | struct mcs_node {
|
---|
221 | mcs_node * volatile next;
|
---|
222 | single_sem sem;
|
---|
223 | };
|
---|
224 |
|
---|
225 | static inline void ?{}(mcs_node & this) { this.next = 0p; }
|
---|
226 |
|
---|
227 | static inline mcs_node * volatile & ?`next ( mcs_node * node ) {
|
---|
228 | return node->next;
|
---|
229 | }
|
---|
230 |
|
---|
231 | struct mcs_lock {
|
---|
232 | mcs_queue(mcs_node) queue;
|
---|
233 | };
|
---|
234 |
|
---|
235 | static inline void lock(mcs_lock & l, mcs_node & n) {
|
---|
236 | if(push(l.queue, &n))
|
---|
237 | wait(n.sem);
|
---|
238 | }
|
---|
239 |
|
---|
240 | static inline void unlock(mcs_lock & l, mcs_node & n) {
|
---|
241 | mcs_node * next = advance(l.queue, &n);
|
---|
242 | if(next) post(next->sem);
|
---|
243 | }
|
---|
244 |
|
---|
245 | struct linear_backoff_then_block_lock {
|
---|
246 | // Spin lock used for mutual exclusion
|
---|
247 | __spinlock_t spinlock;
|
---|
248 |
|
---|
249 | // Current thread owning the lock
|
---|
250 | struct thread$ * owner;
|
---|
251 |
|
---|
252 | // List of blocked threads
|
---|
253 | dlist( thread$ ) blocked_threads;
|
---|
254 |
|
---|
255 | // Used for comparing and exchanging
|
---|
256 | volatile size_t lock_value;
|
---|
257 |
|
---|
258 | // used for linear backoff spinning
|
---|
259 | int spin_start;
|
---|
260 | int spin_end;
|
---|
261 | int spin_count;
|
---|
262 |
|
---|
263 | // after unsuccessful linear backoff yield this many times
|
---|
264 | int yield_count;
|
---|
265 | };
|
---|
266 |
|
---|
267 | static inline void ?{}( linear_backoff_then_block_lock & this, int spin_start, int spin_end, int spin_count, int yield_count ) {
|
---|
268 | this.spinlock{};
|
---|
269 | this.blocked_threads{};
|
---|
270 | this.lock_value = 0;
|
---|
271 | this.spin_start = spin_start;
|
---|
272 | this.spin_end = spin_end;
|
---|
273 | this.spin_count = spin_count;
|
---|
274 | this.yield_count = yield_count;
|
---|
275 | }
|
---|
276 | static inline void ?{}( linear_backoff_then_block_lock & this ) { this{4, 1024, 16, 0}; }
|
---|
277 | static inline void ^?{}( linear_backoff_then_block_lock & this ) {}
|
---|
278 | static inline void ?{}( linear_backoff_then_block_lock & this, linear_backoff_then_block_lock this2 ) = void;
|
---|
279 | static inline void ?=?( linear_backoff_then_block_lock & this, linear_backoff_then_block_lock this2 ) = void;
|
---|
280 |
|
---|
281 | static inline bool internal_try_lock(linear_backoff_then_block_lock & this, size_t & compare_val) with(this) {
|
---|
282 | if (__atomic_compare_exchange_n(&lock_value, &compare_val, 1, false, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED)) {
|
---|
283 | owner = active_thread();
|
---|
284 | return true;
|
---|
285 | }
|
---|
286 | return false;
|
---|
287 | }
|
---|
288 |
|
---|
289 | static inline bool try_lock(linear_backoff_then_block_lock & this) { size_t compare_val = 0; return internal_try_lock(this, compare_val); }
|
---|
290 |
|
---|
291 | static inline bool try_lock_contention(linear_backoff_then_block_lock & this) with(this) {
|
---|
292 | if (__atomic_exchange_n(&lock_value, 2, __ATOMIC_ACQUIRE) == 0) {
|
---|
293 | owner = active_thread();
|
---|
294 | return true;
|
---|
295 | }
|
---|
296 | return false;
|
---|
297 | }
|
---|
298 |
|
---|
299 | static inline bool block(linear_backoff_then_block_lock & this) with(this) {
|
---|
300 | lock( spinlock __cfaabi_dbg_ctx2 );
|
---|
301 | if (lock_value != 2) {
|
---|
302 | unlock( spinlock );
|
---|
303 | return true;
|
---|
304 | }
|
---|
305 | insert_last( blocked_threads, *active_thread() );
|
---|
306 | unlock( spinlock );
|
---|
307 | park( );
|
---|
308 | return true;
|
---|
309 | }
|
---|
310 |
|
---|
311 | static inline bool lock(linear_backoff_then_block_lock & this) with(this) {
|
---|
312 | // if owner just return
|
---|
313 | if (active_thread() == owner) return true;
|
---|
314 | size_t compare_val = 0;
|
---|
315 | int spin = spin_start;
|
---|
316 | // linear backoff
|
---|
317 | for( ;; ) {
|
---|
318 | compare_val = 0;
|
---|
319 | if (internal_try_lock(this, compare_val)) return true;
|
---|
320 | if (2 == compare_val) break;
|
---|
321 | for (int i = 0; i < spin; i++) Pause();
|
---|
322 | if (spin >= spin_end) break;
|
---|
323 | spin += spin;
|
---|
324 | }
|
---|
325 |
|
---|
326 | // linear backoff bounded by spin_count
|
---|
327 | spin = spin_start;
|
---|
328 | int spin_counter = 0;
|
---|
329 | int yield_counter = 0;
|
---|
330 | for ( ;; ) {
|
---|
331 | if(try_lock_contention(this)) return true;
|
---|
332 | if(spin_counter < spin_count) {
|
---|
333 | for (int i = 0; i < spin; i++) Pause();
|
---|
334 | if (spin < spin_end) spin += spin;
|
---|
335 | else spin_counter++;
|
---|
336 | } else if (yield_counter < yield_count) {
|
---|
337 | // after linear backoff yield yield_count times
|
---|
338 | yield_counter++;
|
---|
339 | yield();
|
---|
340 | } else { break; }
|
---|
341 | }
|
---|
342 |
|
---|
343 | // block until signalled
|
---|
344 | while (block(this)) if(try_lock_contention(this)) return true;
|
---|
345 |
|
---|
346 | // this should never be reached as block(this) always returns true
|
---|
347 | return false;
|
---|
348 | }
|
---|
349 |
|
---|
350 | static inline bool lock_improved(linear_backoff_then_block_lock & this) with(this) {
|
---|
351 | // if owner just return
|
---|
352 | if (active_thread() == owner) return true;
|
---|
353 | size_t compare_val = 0;
|
---|
354 | int spin = spin_start;
|
---|
355 | // linear backoff
|
---|
356 | for( ;; ) {
|
---|
357 | compare_val = 0;
|
---|
358 | if (internal_try_lock(this, compare_val)) return true;
|
---|
359 | if (2 == compare_val) break;
|
---|
360 | for (int i = 0; i < spin; i++) Pause();
|
---|
361 | if (spin >= spin_end) break;
|
---|
362 | spin += spin;
|
---|
363 | }
|
---|
364 |
|
---|
365 | // linear backoff bounded by spin_count
|
---|
366 | spin = spin_start;
|
---|
367 | int spin_counter = 0;
|
---|
368 | int yield_counter = 0;
|
---|
369 | for ( ;; ) {
|
---|
370 | compare_val = 0;
|
---|
371 | if(internal_try_lock(this, compare_val)) return true;
|
---|
372 | if (2 == compare_val) break;
|
---|
373 | if(spin_counter < spin_count) {
|
---|
374 | for (int i = 0; i < spin; i++) Pause();
|
---|
375 | if (spin < spin_end) spin += spin;
|
---|
376 | else spin_counter++;
|
---|
377 | } else if (yield_counter < yield_count) {
|
---|
378 | // after linear backoff yield yield_count times
|
---|
379 | yield_counter++;
|
---|
380 | yield();
|
---|
381 | } else { break; }
|
---|
382 | }
|
---|
383 |
|
---|
384 | if(2 != compare_val && try_lock_contention(this)) return true;
|
---|
385 | // block until signalled
|
---|
386 | while (block(this)) if(try_lock_contention(this)) return true;
|
---|
387 |
|
---|
388 | // this should never be reached as block(this) always returns true
|
---|
389 | return false;
|
---|
390 | }
|
---|
391 |
|
---|
392 | static inline void unlock(linear_backoff_then_block_lock & this) with(this) {
|
---|
393 | verify(lock_value > 0);
|
---|
394 | owner = 0p;
|
---|
395 | if (__atomic_exchange_n(&lock_value, 0, __ATOMIC_RELEASE) == 1) return;
|
---|
396 | lock( spinlock __cfaabi_dbg_ctx2 );
|
---|
397 | thread$ * t = &try_pop_front( blocked_threads );
|
---|
398 | unlock( spinlock );
|
---|
399 | unpark( t );
|
---|
400 | }
|
---|
401 |
|
---|
402 | static inline void on_notify(linear_backoff_then_block_lock & this, struct thread$ * t ) { unpark(t); }
|
---|
403 | static inline size_t on_wait(linear_backoff_then_block_lock & this) { unlock(this); return 0; }
|
---|
404 | static inline void on_wakeup(linear_backoff_then_block_lock & this, size_t recursion ) { lock_improved(this); }
|
---|
405 |
|
---|
406 | //-----------------------------------------------------------------------------
|
---|
407 | // is_blocking_lock
|
---|
408 | trait is_blocking_lock(L & | sized(L)) {
|
---|
409 | // For synchronization locks to use when acquiring
|
---|
410 | void on_notify( L &, struct thread$ * );
|
---|
411 |
|
---|
412 | // For synchronization locks to use when releasing
|
---|
413 | size_t on_wait( L & );
|
---|
414 |
|
---|
415 | // to set recursion count after getting signalled;
|
---|
416 | void on_wakeup( L &, size_t recursion );
|
---|
417 | };
|
---|
418 |
|
---|
419 | //-----------------------------------------------------------------------------
|
---|
420 | // // info_thread
|
---|
421 | // // the info thread is a wrapper around a thread used
|
---|
422 | // // to store extra data for use in the condition variable
|
---|
423 | forall(L & | is_blocking_lock(L)) {
|
---|
424 | struct info_thread;
|
---|
425 |
|
---|
426 | // // for use by sequence
|
---|
427 | // info_thread(L) *& Back( info_thread(L) * this );
|
---|
428 | // info_thread(L) *& Next( info_thread(L) * this );
|
---|
429 | }
|
---|
430 |
|
---|
431 | //-----------------------------------------------------------------------------
|
---|
432 | // Synchronization Locks
|
---|
433 | forall(L & | is_blocking_lock(L)) {
|
---|
434 | struct condition_variable {
|
---|
435 | // Spin lock used for mutual exclusion
|
---|
436 | __spinlock_t lock;
|
---|
437 |
|
---|
438 | // List of blocked threads
|
---|
439 | dlist( info_thread(L) ) blocked_threads;
|
---|
440 |
|
---|
441 | // Count of current blocked threads
|
---|
442 | int count;
|
---|
443 | };
|
---|
444 |
|
---|
445 |
|
---|
446 | void ?{}( condition_variable(L) & this );
|
---|
447 | void ^?{}( condition_variable(L) & this );
|
---|
448 |
|
---|
449 | bool notify_one( condition_variable(L) & this );
|
---|
450 | bool notify_all( condition_variable(L) & this );
|
---|
451 |
|
---|
452 | uintptr_t front( condition_variable(L) & this );
|
---|
453 |
|
---|
454 | bool empty ( condition_variable(L) & this );
|
---|
455 | int counter( condition_variable(L) & this );
|
---|
456 |
|
---|
457 | void wait( condition_variable(L) & this );
|
---|
458 | void wait( condition_variable(L) & this, uintptr_t info );
|
---|
459 | bool wait( condition_variable(L) & this, Duration duration );
|
---|
460 | bool wait( condition_variable(L) & this, uintptr_t info, Duration duration );
|
---|
461 |
|
---|
462 | void wait( condition_variable(L) & this, L & l );
|
---|
463 | void wait( condition_variable(L) & this, L & l, uintptr_t info );
|
---|
464 | bool wait( condition_variable(L) & this, L & l, Duration duration );
|
---|
465 | bool wait( condition_variable(L) & this, L & l, uintptr_t info, Duration duration );
|
---|
466 | }
|
---|