1 /* 2 * coroutine queues and locks 3 * 4 * Copyright (c) 2011 Kevin Wolf <kwolf@redhat.com> 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 * 24 * The lock-free mutex implementation is based on OSv 25 * (core/lfmutex.cc, include/lockfree/mutex.hh). 26 * Copyright (C) 2013 Cloudius Systems, Ltd. 27 */ 28 29 #include "qemu/osdep.h" 30 #include "qemu/coroutine_int.h" 31 #include "qemu/processor.h" 32 #include "qemu/queue.h" 33 #include "block/aio.h" 34 #include "trace.h" 35 36 void qemu_co_queue_init(CoQueue *queue) 37 { 38 QSIMPLEQ_INIT(&queue->entries); 39 } 40 41 void coroutine_fn qemu_co_queue_wait_impl(CoQueue *queue, QemuLockable *lock, 42 CoQueueWaitFlags flags) 43 { 44 Coroutine *self = qemu_coroutine_self(); 45 if (flags & CO_QUEUE_WAIT_FRONT) { 46 QSIMPLEQ_INSERT_HEAD(&queue->entries, self, co_queue_next); 47 } else { 48 QSIMPLEQ_INSERT_TAIL(&queue->entries, self, co_queue_next); 49 } 50 51 if (lock) { 52 qemu_lockable_unlock(lock); 53 } 54 55 /* There is no race condition here. Other threads will call 56 * aio_co_schedule on our AioContext, which can reenter this 57 * coroutine but only after this yield and after the main loop 58 * has gone through the next iteration. 59 */ 60 qemu_coroutine_yield(); 61 assert(qemu_in_coroutine()); 62 63 /* TODO: OSv implements wait morphing here, where the wakeup 64 * primitive automatically places the woken coroutine on the 65 * mutex's queue. This avoids the thundering herd effect. 66 * This could be implemented for CoMutexes, but not really for 67 * other cases of QemuLockable. 68 */ 69 if (lock) { 70 qemu_lockable_lock(lock); 71 } 72 } 73 74 bool qemu_co_enter_next_impl(CoQueue *queue, QemuLockable *lock) 75 { 76 Coroutine *next; 77 78 next = QSIMPLEQ_FIRST(&queue->entries); 79 if (!next) { 80 return false; 81 } 82 83 QSIMPLEQ_REMOVE_HEAD(&queue->entries, co_queue_next); 84 if (lock) { 85 qemu_lockable_unlock(lock); 86 } 87 aio_co_wake(next); 88 if (lock) { 89 qemu_lockable_lock(lock); 90 } 91 return true; 92 } 93 94 bool coroutine_fn qemu_co_queue_next(CoQueue *queue) 95 { 96 /* No unlock/lock needed in coroutine context. */ 97 return qemu_co_enter_next_impl(queue, NULL); 98 } 99 100 void qemu_co_enter_all_impl(CoQueue *queue, QemuLockable *lock) 101 { 102 while (qemu_co_enter_next_impl(queue, lock)) { 103 /* just loop */ 104 } 105 } 106 107 void coroutine_fn qemu_co_queue_restart_all(CoQueue *queue) 108 { 109 /* No unlock/lock needed in coroutine context. */ 110 qemu_co_enter_all_impl(queue, NULL); 111 } 112 113 bool qemu_co_queue_empty(CoQueue *queue) 114 { 115 return QSIMPLEQ_FIRST(&queue->entries) == NULL; 116 } 117 118 /* The wait records are handled with a multiple-producer, single-consumer 119 * lock-free queue. There cannot be two concurrent pop_waiter() calls 120 * because pop_waiter() can only be called while mutex->handoff is zero. 121 * This can happen in three cases: 122 * - in qemu_co_mutex_unlock, before the hand-off protocol has started. 123 * In this case, qemu_co_mutex_lock will see mutex->handoff == 0 and 124 * not take part in the handoff. 125 * - in qemu_co_mutex_lock, if it steals the hand-off responsibility from 126 * qemu_co_mutex_unlock. In this case, qemu_co_mutex_unlock will fail 127 * the cmpxchg (it will see either 0 or the next sequence value) and 128 * exit. The next hand-off cannot begin until qemu_co_mutex_lock has 129 * woken up someone. 130 * - in qemu_co_mutex_unlock, if it takes the hand-off token itself. 131 * In this case another iteration starts with mutex->handoff == 0; 132 * a concurrent qemu_co_mutex_lock will fail the cmpxchg, and 133 * qemu_co_mutex_unlock will go back to case (1). 134 * 135 * The following functions manage this queue. 136 */ 137 typedef struct CoWaitRecord { 138 Coroutine *co; 139 QSLIST_ENTRY(CoWaitRecord) next; 140 } CoWaitRecord; 141 142 static void coroutine_fn push_waiter(CoMutex *mutex, CoWaitRecord *w) 143 { 144 w->co = qemu_coroutine_self(); 145 QSLIST_INSERT_HEAD_ATOMIC(&mutex->from_push, w, next); 146 } 147 148 static void move_waiters(CoMutex *mutex) 149 { 150 QSLIST_HEAD(, CoWaitRecord) reversed; 151 QSLIST_MOVE_ATOMIC(&reversed, &mutex->from_push); 152 while (!QSLIST_EMPTY(&reversed)) { 153 CoWaitRecord *w = QSLIST_FIRST(&reversed); 154 QSLIST_REMOVE_HEAD(&reversed, next); 155 QSLIST_INSERT_HEAD(&mutex->to_pop, w, next); 156 } 157 } 158 159 static CoWaitRecord *pop_waiter(CoMutex *mutex) 160 { 161 CoWaitRecord *w; 162 163 if (QSLIST_EMPTY(&mutex->to_pop)) { 164 move_waiters(mutex); 165 if (QSLIST_EMPTY(&mutex->to_pop)) { 166 return NULL; 167 } 168 } 169 w = QSLIST_FIRST(&mutex->to_pop); 170 QSLIST_REMOVE_HEAD(&mutex->to_pop, next); 171 return w; 172 } 173 174 static bool has_waiters(CoMutex *mutex) 175 { 176 return QSLIST_EMPTY(&mutex->to_pop) || QSLIST_EMPTY(&mutex->from_push); 177 } 178 179 void qemu_co_mutex_init(CoMutex *mutex) 180 { 181 memset(mutex, 0, sizeof(*mutex)); 182 } 183 184 static void coroutine_fn qemu_co_mutex_wake(CoMutex *mutex, Coroutine *co) 185 { 186 /* Read co before co->ctx; pairs with smp_wmb() in 187 * qemu_coroutine_enter(). 188 */ 189 smp_read_barrier_depends(); 190 mutex->ctx = co->ctx; 191 aio_co_wake(co); 192 } 193 194 static void coroutine_fn qemu_co_mutex_lock_slowpath(AioContext *ctx, 195 CoMutex *mutex) 196 { 197 Coroutine *self = qemu_coroutine_self(); 198 CoWaitRecord w; 199 unsigned old_handoff; 200 201 trace_qemu_co_mutex_lock_entry(mutex, self); 202 push_waiter(mutex, &w); 203 204 /* 205 * Add waiter before reading mutex->handoff. Pairs with qatomic_mb_set 206 * in qemu_co_mutex_unlock. 207 */ 208 smp_mb__after_rmw(); 209 210 /* This is the "Responsibility Hand-Off" protocol; a lock() picks from 211 * a concurrent unlock() the responsibility of waking somebody up. 212 */ 213 old_handoff = qatomic_read(&mutex->handoff); 214 if (old_handoff && 215 has_waiters(mutex) && 216 qatomic_cmpxchg(&mutex->handoff, old_handoff, 0) == old_handoff) { 217 /* There can be no concurrent pops, because there can be only 218 * one active handoff at a time. 219 */ 220 CoWaitRecord *to_wake = pop_waiter(mutex); 221 Coroutine *co = to_wake->co; 222 if (co == self) { 223 /* We got the lock ourselves! */ 224 assert(to_wake == &w); 225 mutex->ctx = ctx; 226 return; 227 } 228 229 qemu_co_mutex_wake(mutex, co); 230 } 231 232 qemu_coroutine_yield(); 233 trace_qemu_co_mutex_lock_return(mutex, self); 234 } 235 236 void coroutine_fn qemu_co_mutex_lock(CoMutex *mutex) 237 { 238 AioContext *ctx = qemu_get_current_aio_context(); 239 Coroutine *self = qemu_coroutine_self(); 240 int waiters, i; 241 242 /* Running a very small critical section on pthread_mutex_t and CoMutex 243 * shows that pthread_mutex_t is much faster because it doesn't actually 244 * go to sleep. What happens is that the critical section is shorter 245 * than the latency of entering the kernel and thus FUTEX_WAIT always 246 * fails. With CoMutex there is no such latency but you still want to 247 * avoid wait and wakeup. So introduce it artificially. 248 */ 249 i = 0; 250 retry_fast_path: 251 waiters = qatomic_cmpxchg(&mutex->locked, 0, 1); 252 if (waiters != 0) { 253 while (waiters == 1 && ++i < 1000) { 254 if (qatomic_read(&mutex->ctx) == ctx) { 255 break; 256 } 257 if (qatomic_read(&mutex->locked) == 0) { 258 goto retry_fast_path; 259 } 260 cpu_relax(); 261 } 262 waiters = qatomic_fetch_inc(&mutex->locked); 263 } 264 265 if (waiters == 0) { 266 /* Uncontended. */ 267 trace_qemu_co_mutex_lock_uncontended(mutex, self); 268 mutex->ctx = ctx; 269 } else { 270 qemu_co_mutex_lock_slowpath(ctx, mutex); 271 } 272 mutex->holder = self; 273 self->locks_held++; 274 } 275 276 void coroutine_fn qemu_co_mutex_unlock(CoMutex *mutex) 277 { 278 Coroutine *self = qemu_coroutine_self(); 279 280 trace_qemu_co_mutex_unlock_entry(mutex, self); 281 282 assert(mutex->locked); 283 assert(mutex->holder == self); 284 assert(qemu_in_coroutine()); 285 286 mutex->ctx = NULL; 287 mutex->holder = NULL; 288 self->locks_held--; 289 if (qatomic_fetch_dec(&mutex->locked) == 1) { 290 /* No waiting qemu_co_mutex_lock(). Pfew, that was easy! */ 291 return; 292 } 293 294 for (;;) { 295 CoWaitRecord *to_wake = pop_waiter(mutex); 296 unsigned our_handoff; 297 298 if (to_wake) { 299 qemu_co_mutex_wake(mutex, to_wake->co); 300 break; 301 } 302 303 /* Some concurrent lock() is in progress (we know this because 304 * mutex->locked was >1) but it hasn't yet put itself on the wait 305 * queue. Pick a sequence number for the handoff protocol (not 0). 306 */ 307 if (++mutex->sequence == 0) { 308 mutex->sequence = 1; 309 } 310 311 our_handoff = mutex->sequence; 312 /* Set handoff before checking for waiters. */ 313 qatomic_mb_set(&mutex->handoff, our_handoff); 314 if (!has_waiters(mutex)) { 315 /* The concurrent lock has not added itself yet, so it 316 * will be able to pick our handoff. 317 */ 318 break; 319 } 320 321 /* Try to do the handoff protocol ourselves; if somebody else has 322 * already taken it, however, we're done and they're responsible. 323 */ 324 if (qatomic_cmpxchg(&mutex->handoff, our_handoff, 0) != our_handoff) { 325 break; 326 } 327 } 328 329 trace_qemu_co_mutex_unlock_return(mutex, self); 330 } 331 332 struct CoRwTicket { 333 bool read; 334 Coroutine *co; 335 QSIMPLEQ_ENTRY(CoRwTicket) next; 336 }; 337 338 void qemu_co_rwlock_init(CoRwlock *lock) 339 { 340 qemu_co_mutex_init(&lock->mutex); 341 lock->owners = 0; 342 QSIMPLEQ_INIT(&lock->tickets); 343 } 344 345 /* Releases the internal CoMutex. */ 346 static void coroutine_fn qemu_co_rwlock_maybe_wake_one(CoRwlock *lock) 347 { 348 CoRwTicket *tkt = QSIMPLEQ_FIRST(&lock->tickets); 349 Coroutine *co = NULL; 350 351 /* 352 * Setting lock->owners here prevents rdlock and wrlock from 353 * sneaking in between unlock and wake. 354 */ 355 356 if (tkt) { 357 if (tkt->read) { 358 if (lock->owners >= 0) { 359 lock->owners++; 360 co = tkt->co; 361 } 362 } else { 363 if (lock->owners == 0) { 364 lock->owners = -1; 365 co = tkt->co; 366 } 367 } 368 } 369 370 if (co) { 371 QSIMPLEQ_REMOVE_HEAD(&lock->tickets, next); 372 qemu_co_mutex_unlock(&lock->mutex); 373 aio_co_wake(co); 374 } else { 375 qemu_co_mutex_unlock(&lock->mutex); 376 } 377 } 378 379 void coroutine_fn qemu_co_rwlock_rdlock(CoRwlock *lock) 380 { 381 Coroutine *self = qemu_coroutine_self(); 382 383 qemu_co_mutex_lock(&lock->mutex); 384 /* For fairness, wait if a writer is in line. */ 385 if (lock->owners == 0 || (lock->owners > 0 && QSIMPLEQ_EMPTY(&lock->tickets))) { 386 lock->owners++; 387 qemu_co_mutex_unlock(&lock->mutex); 388 } else { 389 CoRwTicket my_ticket = { true, self }; 390 391 QSIMPLEQ_INSERT_TAIL(&lock->tickets, &my_ticket, next); 392 qemu_co_mutex_unlock(&lock->mutex); 393 qemu_coroutine_yield(); 394 assert(lock->owners >= 1); 395 396 /* Possibly wake another reader, which will wake the next in line. */ 397 qemu_co_mutex_lock(&lock->mutex); 398 qemu_co_rwlock_maybe_wake_one(lock); 399 } 400 401 self->locks_held++; 402 } 403 404 void coroutine_fn qemu_co_rwlock_unlock(CoRwlock *lock) 405 { 406 Coroutine *self = qemu_coroutine_self(); 407 408 assert(qemu_in_coroutine()); 409 self->locks_held--; 410 411 qemu_co_mutex_lock(&lock->mutex); 412 if (lock->owners > 0) { 413 lock->owners--; 414 } else { 415 assert(lock->owners == -1); 416 lock->owners = 0; 417 } 418 419 qemu_co_rwlock_maybe_wake_one(lock); 420 } 421 422 void coroutine_fn qemu_co_rwlock_downgrade(CoRwlock *lock) 423 { 424 qemu_co_mutex_lock(&lock->mutex); 425 assert(lock->owners == -1); 426 lock->owners = 1; 427 428 /* Possibly wake another reader, which will wake the next in line. */ 429 qemu_co_rwlock_maybe_wake_one(lock); 430 } 431 432 void coroutine_fn qemu_co_rwlock_wrlock(CoRwlock *lock) 433 { 434 Coroutine *self = qemu_coroutine_self(); 435 436 qemu_co_mutex_lock(&lock->mutex); 437 if (lock->owners == 0) { 438 lock->owners = -1; 439 qemu_co_mutex_unlock(&lock->mutex); 440 } else { 441 CoRwTicket my_ticket = { false, qemu_coroutine_self() }; 442 443 QSIMPLEQ_INSERT_TAIL(&lock->tickets, &my_ticket, next); 444 qemu_co_mutex_unlock(&lock->mutex); 445 qemu_coroutine_yield(); 446 assert(lock->owners == -1); 447 } 448 449 self->locks_held++; 450 } 451 452 void coroutine_fn qemu_co_rwlock_upgrade(CoRwlock *lock) 453 { 454 qemu_co_mutex_lock(&lock->mutex); 455 assert(lock->owners > 0); 456 /* For fairness, wait if a writer is in line. */ 457 if (lock->owners == 1 && QSIMPLEQ_EMPTY(&lock->tickets)) { 458 lock->owners = -1; 459 qemu_co_mutex_unlock(&lock->mutex); 460 } else { 461 CoRwTicket my_ticket = { false, qemu_coroutine_self() }; 462 463 lock->owners--; 464 QSIMPLEQ_INSERT_TAIL(&lock->tickets, &my_ticket, next); 465 qemu_co_rwlock_maybe_wake_one(lock); 466 qemu_coroutine_yield(); 467 assert(lock->owners == -1); 468 } 469 } 470