1 /* 2 * Data plane event loop 3 * 4 * Copyright (c) 2003-2008 Fabrice Bellard 5 * Copyright (c) 2009-2017 QEMU contributors 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a copy 8 * of this software and associated documentation files (the "Software"), to deal 9 * in the Software without restriction, including without limitation the rights 10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11 * copies of the Software, and to permit persons to whom the Software is 12 * furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23 * THE SOFTWARE. 24 */ 25 26 #include "qemu/osdep.h" 27 #include "qapi/error.h" 28 #include "block/aio.h" 29 #include "block/thread-pool.h" 30 #include "qemu/main-loop.h" 31 #include "qemu/atomic.h" 32 #include "qemu/rcu_queue.h" 33 #include "block/raw-aio.h" 34 #include "qemu/coroutine_int.h" 35 #include "trace.h" 36 37 /***********************************************************/ 38 /* bottom halves (can be seen as timers which expire ASAP) */ 39 40 /* QEMUBH::flags values */ 41 enum { 42 /* Already enqueued and waiting for aio_bh_poll() */ 43 BH_PENDING = (1 << 0), 44 45 /* Invoke the callback */ 46 BH_SCHEDULED = (1 << 1), 47 48 /* Delete without invoking callback */ 49 BH_DELETED = (1 << 2), 50 51 /* Delete after invoking callback */ 52 BH_ONESHOT = (1 << 3), 53 54 /* Schedule periodically when the event loop is idle */ 55 BH_IDLE = (1 << 4), 56 }; 57 58 struct QEMUBH { 59 AioContext *ctx; 60 QEMUBHFunc *cb; 61 void *opaque; 62 QSLIST_ENTRY(QEMUBH) next; 63 unsigned flags; 64 }; 65 66 /* Called concurrently from any thread */ 67 static void aio_bh_enqueue(QEMUBH *bh, unsigned new_flags) 68 { 69 AioContext *ctx = bh->ctx; 70 unsigned old_flags; 71 72 /* 73 * The memory barrier implicit in atomic_fetch_or makes sure that: 74 * 1. idle & any writes needed by the callback are done before the 75 * locations are read in the aio_bh_poll. 76 * 2. ctx is loaded before the callback has a chance to execute and bh 77 * could be freed. 78 */ 79 old_flags = atomic_fetch_or(&bh->flags, BH_PENDING | new_flags); 80 if (!(old_flags & BH_PENDING)) { 81 QSLIST_INSERT_HEAD_ATOMIC(&ctx->bh_list, bh, next); 82 } 83 84 aio_notify(ctx); 85 } 86 87 /* Only called from aio_bh_poll() and aio_ctx_finalize() */ 88 static QEMUBH *aio_bh_dequeue(BHList *head, unsigned *flags) 89 { 90 QEMUBH *bh = QSLIST_FIRST_RCU(head); 91 92 if (!bh) { 93 return NULL; 94 } 95 96 QSLIST_REMOVE_HEAD(head, next); 97 98 /* 99 * The atomic_and is paired with aio_bh_enqueue(). The implicit memory 100 * barrier ensures that the callback sees all writes done by the scheduling 101 * thread. It also ensures that the scheduling thread sees the cleared 102 * flag before bh->cb has run, and thus will call aio_notify again if 103 * necessary. 104 */ 105 *flags = atomic_fetch_and(&bh->flags, 106 ~(BH_PENDING | BH_SCHEDULED | BH_IDLE)); 107 return bh; 108 } 109 110 void aio_bh_schedule_oneshot(AioContext *ctx, QEMUBHFunc *cb, void *opaque) 111 { 112 QEMUBH *bh; 113 bh = g_new(QEMUBH, 1); 114 *bh = (QEMUBH){ 115 .ctx = ctx, 116 .cb = cb, 117 .opaque = opaque, 118 }; 119 aio_bh_enqueue(bh, BH_SCHEDULED | BH_ONESHOT); 120 } 121 122 QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque) 123 { 124 QEMUBH *bh; 125 bh = g_new(QEMUBH, 1); 126 *bh = (QEMUBH){ 127 .ctx = ctx, 128 .cb = cb, 129 .opaque = opaque, 130 }; 131 return bh; 132 } 133 134 void aio_bh_call(QEMUBH *bh) 135 { 136 bh->cb(bh->opaque); 137 } 138 139 /* Multiple occurrences of aio_bh_poll cannot be called concurrently. */ 140 int aio_bh_poll(AioContext *ctx) 141 { 142 BHListSlice slice; 143 BHListSlice *s; 144 int ret = 0; 145 146 QSLIST_MOVE_ATOMIC(&slice.bh_list, &ctx->bh_list); 147 QSIMPLEQ_INSERT_TAIL(&ctx->bh_slice_list, &slice, next); 148 149 while ((s = QSIMPLEQ_FIRST(&ctx->bh_slice_list))) { 150 QEMUBH *bh; 151 unsigned flags; 152 153 bh = aio_bh_dequeue(&s->bh_list, &flags); 154 if (!bh) { 155 QSIMPLEQ_REMOVE_HEAD(&ctx->bh_slice_list, next); 156 continue; 157 } 158 159 if ((flags & (BH_SCHEDULED | BH_DELETED)) == BH_SCHEDULED) { 160 /* Idle BHs don't count as progress */ 161 if (!(flags & BH_IDLE)) { 162 ret = 1; 163 } 164 aio_bh_call(bh); 165 } 166 if (flags & (BH_DELETED | BH_ONESHOT)) { 167 g_free(bh); 168 } 169 } 170 171 return ret; 172 } 173 174 void qemu_bh_schedule_idle(QEMUBH *bh) 175 { 176 aio_bh_enqueue(bh, BH_SCHEDULED | BH_IDLE); 177 } 178 179 void qemu_bh_schedule(QEMUBH *bh) 180 { 181 aio_bh_enqueue(bh, BH_SCHEDULED); 182 } 183 184 /* This func is async. 185 */ 186 void qemu_bh_cancel(QEMUBH *bh) 187 { 188 atomic_and(&bh->flags, ~BH_SCHEDULED); 189 } 190 191 /* This func is async.The bottom half will do the delete action at the finial 192 * end. 193 */ 194 void qemu_bh_delete(QEMUBH *bh) 195 { 196 aio_bh_enqueue(bh, BH_DELETED); 197 } 198 199 static int64_t aio_compute_bh_timeout(BHList *head, int timeout) 200 { 201 QEMUBH *bh; 202 203 QSLIST_FOREACH_RCU(bh, head, next) { 204 if ((bh->flags & (BH_SCHEDULED | BH_DELETED)) == BH_SCHEDULED) { 205 if (bh->flags & BH_IDLE) { 206 /* idle bottom halves will be polled at least 207 * every 10ms */ 208 timeout = 10000000; 209 } else { 210 /* non-idle bottom halves will be executed 211 * immediately */ 212 return 0; 213 } 214 } 215 } 216 217 return timeout; 218 } 219 220 int64_t 221 aio_compute_timeout(AioContext *ctx) 222 { 223 BHListSlice *s; 224 int64_t deadline; 225 int timeout = -1; 226 227 timeout = aio_compute_bh_timeout(&ctx->bh_list, timeout); 228 if (timeout == 0) { 229 return 0; 230 } 231 232 QSIMPLEQ_FOREACH(s, &ctx->bh_slice_list, next) { 233 timeout = aio_compute_bh_timeout(&s->bh_list, timeout); 234 if (timeout == 0) { 235 return 0; 236 } 237 } 238 239 deadline = timerlistgroup_deadline_ns(&ctx->tlg); 240 if (deadline == 0) { 241 return 0; 242 } else { 243 return qemu_soonest_timeout(timeout, deadline); 244 } 245 } 246 247 static gboolean 248 aio_ctx_prepare(GSource *source, gint *timeout) 249 { 250 AioContext *ctx = (AioContext *) source; 251 252 atomic_set(&ctx->notify_me, atomic_read(&ctx->notify_me) | 1); 253 254 /* 255 * Write ctx->notify_me before computing the timeout 256 * (reading bottom half flags, etc.). Pairs with 257 * smp_mb in aio_notify(). 258 */ 259 smp_mb(); 260 261 /* We assume there is no timeout already supplied */ 262 *timeout = qemu_timeout_ns_to_ms(aio_compute_timeout(ctx)); 263 264 if (aio_prepare(ctx)) { 265 *timeout = 0; 266 } 267 268 return *timeout == 0; 269 } 270 271 static gboolean 272 aio_ctx_check(GSource *source) 273 { 274 AioContext *ctx = (AioContext *) source; 275 QEMUBH *bh; 276 BHListSlice *s; 277 278 /* Finish computing the timeout before clearing the flag. */ 279 atomic_store_release(&ctx->notify_me, atomic_read(&ctx->notify_me) & ~1); 280 aio_notify_accept(ctx); 281 282 QSLIST_FOREACH_RCU(bh, &ctx->bh_list, next) { 283 if ((bh->flags & (BH_SCHEDULED | BH_DELETED)) == BH_SCHEDULED) { 284 return true; 285 } 286 } 287 288 QSIMPLEQ_FOREACH(s, &ctx->bh_slice_list, next) { 289 QSLIST_FOREACH_RCU(bh, &s->bh_list, next) { 290 if ((bh->flags & (BH_SCHEDULED | BH_DELETED)) == BH_SCHEDULED) { 291 return true; 292 } 293 } 294 } 295 return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0); 296 } 297 298 static gboolean 299 aio_ctx_dispatch(GSource *source, 300 GSourceFunc callback, 301 gpointer user_data) 302 { 303 AioContext *ctx = (AioContext *) source; 304 305 assert(callback == NULL); 306 aio_dispatch(ctx); 307 return true; 308 } 309 310 static void 311 aio_ctx_finalize(GSource *source) 312 { 313 AioContext *ctx = (AioContext *) source; 314 QEMUBH *bh; 315 unsigned flags; 316 317 thread_pool_free(ctx->thread_pool); 318 319 #ifdef CONFIG_LINUX_AIO 320 if (ctx->linux_aio) { 321 laio_detach_aio_context(ctx->linux_aio, ctx); 322 laio_cleanup(ctx->linux_aio); 323 ctx->linux_aio = NULL; 324 } 325 #endif 326 327 #ifdef CONFIG_LINUX_IO_URING 328 if (ctx->linux_io_uring) { 329 luring_detach_aio_context(ctx->linux_io_uring, ctx); 330 luring_cleanup(ctx->linux_io_uring); 331 ctx->linux_io_uring = NULL; 332 } 333 #endif 334 335 assert(QSLIST_EMPTY(&ctx->scheduled_coroutines)); 336 qemu_bh_delete(ctx->co_schedule_bh); 337 338 /* There must be no aio_bh_poll() calls going on */ 339 assert(QSIMPLEQ_EMPTY(&ctx->bh_slice_list)); 340 341 while ((bh = aio_bh_dequeue(&ctx->bh_list, &flags))) { 342 /* qemu_bh_delete() must have been called on BHs in this AioContext */ 343 assert(flags & BH_DELETED); 344 345 g_free(bh); 346 } 347 348 aio_set_event_notifier(ctx, &ctx->notifier, false, NULL, NULL); 349 event_notifier_cleanup(&ctx->notifier); 350 qemu_rec_mutex_destroy(&ctx->lock); 351 qemu_lockcnt_destroy(&ctx->list_lock); 352 timerlistgroup_deinit(&ctx->tlg); 353 aio_context_destroy(ctx); 354 } 355 356 static GSourceFuncs aio_source_funcs = { 357 aio_ctx_prepare, 358 aio_ctx_check, 359 aio_ctx_dispatch, 360 aio_ctx_finalize 361 }; 362 363 GSource *aio_get_g_source(AioContext *ctx) 364 { 365 aio_context_use_g_source(ctx); 366 g_source_ref(&ctx->source); 367 return &ctx->source; 368 } 369 370 ThreadPool *aio_get_thread_pool(AioContext *ctx) 371 { 372 if (!ctx->thread_pool) { 373 ctx->thread_pool = thread_pool_new(ctx); 374 } 375 return ctx->thread_pool; 376 } 377 378 #ifdef CONFIG_LINUX_AIO 379 LinuxAioState *aio_setup_linux_aio(AioContext *ctx, Error **errp) 380 { 381 if (!ctx->linux_aio) { 382 ctx->linux_aio = laio_init(errp); 383 if (ctx->linux_aio) { 384 laio_attach_aio_context(ctx->linux_aio, ctx); 385 } 386 } 387 return ctx->linux_aio; 388 } 389 390 LinuxAioState *aio_get_linux_aio(AioContext *ctx) 391 { 392 assert(ctx->linux_aio); 393 return ctx->linux_aio; 394 } 395 #endif 396 397 #ifdef CONFIG_LINUX_IO_URING 398 LuringState *aio_setup_linux_io_uring(AioContext *ctx, Error **errp) 399 { 400 if (ctx->linux_io_uring) { 401 return ctx->linux_io_uring; 402 } 403 404 ctx->linux_io_uring = luring_init(errp); 405 if (!ctx->linux_io_uring) { 406 return NULL; 407 } 408 409 luring_attach_aio_context(ctx->linux_io_uring, ctx); 410 return ctx->linux_io_uring; 411 } 412 413 LuringState *aio_get_linux_io_uring(AioContext *ctx) 414 { 415 assert(ctx->linux_io_uring); 416 return ctx->linux_io_uring; 417 } 418 #endif 419 420 void aio_notify(AioContext *ctx) 421 { 422 /* Write e.g. bh->scheduled before reading ctx->notify_me. Pairs 423 * with smp_mb in aio_ctx_prepare or aio_poll. 424 */ 425 smp_mb(); 426 if (atomic_read(&ctx->notify_me)) { 427 event_notifier_set(&ctx->notifier); 428 atomic_mb_set(&ctx->notified, true); 429 } 430 } 431 432 void aio_notify_accept(AioContext *ctx) 433 { 434 if (atomic_xchg(&ctx->notified, false) 435 #ifdef WIN32 436 || true 437 #endif 438 ) { 439 event_notifier_test_and_clear(&ctx->notifier); 440 } 441 } 442 443 static void aio_timerlist_notify(void *opaque, QEMUClockType type) 444 { 445 aio_notify(opaque); 446 } 447 448 static void event_notifier_dummy_cb(EventNotifier *e) 449 { 450 } 451 452 /* Returns true if aio_notify() was called (e.g. a BH was scheduled) */ 453 static bool event_notifier_poll(void *opaque) 454 { 455 EventNotifier *e = opaque; 456 AioContext *ctx = container_of(e, AioContext, notifier); 457 458 return atomic_read(&ctx->notified); 459 } 460 461 static void co_schedule_bh_cb(void *opaque) 462 { 463 AioContext *ctx = opaque; 464 QSLIST_HEAD(, Coroutine) straight, reversed; 465 466 QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines); 467 QSLIST_INIT(&straight); 468 469 while (!QSLIST_EMPTY(&reversed)) { 470 Coroutine *co = QSLIST_FIRST(&reversed); 471 QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next); 472 QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next); 473 } 474 475 while (!QSLIST_EMPTY(&straight)) { 476 Coroutine *co = QSLIST_FIRST(&straight); 477 QSLIST_REMOVE_HEAD(&straight, co_scheduled_next); 478 trace_aio_co_schedule_bh_cb(ctx, co); 479 aio_context_acquire(ctx); 480 481 /* Protected by write barrier in qemu_aio_coroutine_enter */ 482 atomic_set(&co->scheduled, NULL); 483 qemu_aio_coroutine_enter(ctx, co); 484 aio_context_release(ctx); 485 } 486 } 487 488 AioContext *aio_context_new(Error **errp) 489 { 490 int ret; 491 AioContext *ctx; 492 493 ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext)); 494 QSLIST_INIT(&ctx->bh_list); 495 QSIMPLEQ_INIT(&ctx->bh_slice_list); 496 aio_context_setup(ctx); 497 498 ret = event_notifier_init(&ctx->notifier, false); 499 if (ret < 0) { 500 error_setg_errno(errp, -ret, "Failed to initialize event notifier"); 501 goto fail; 502 } 503 g_source_set_can_recurse(&ctx->source, true); 504 qemu_lockcnt_init(&ctx->list_lock); 505 506 ctx->co_schedule_bh = aio_bh_new(ctx, co_schedule_bh_cb, ctx); 507 QSLIST_INIT(&ctx->scheduled_coroutines); 508 509 aio_set_event_notifier(ctx, &ctx->notifier, 510 false, 511 event_notifier_dummy_cb, 512 event_notifier_poll); 513 #ifdef CONFIG_LINUX_AIO 514 ctx->linux_aio = NULL; 515 #endif 516 517 #ifdef CONFIG_LINUX_IO_URING 518 ctx->linux_io_uring = NULL; 519 #endif 520 521 ctx->thread_pool = NULL; 522 qemu_rec_mutex_init(&ctx->lock); 523 timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx); 524 525 ctx->poll_ns = 0; 526 ctx->poll_max_ns = 0; 527 ctx->poll_grow = 0; 528 ctx->poll_shrink = 0; 529 530 return ctx; 531 fail: 532 g_source_destroy(&ctx->source); 533 return NULL; 534 } 535 536 void aio_co_schedule(AioContext *ctx, Coroutine *co) 537 { 538 trace_aio_co_schedule(ctx, co); 539 const char *scheduled = atomic_cmpxchg(&co->scheduled, NULL, 540 __func__); 541 542 if (scheduled) { 543 fprintf(stderr, 544 "%s: Co-routine was already scheduled in '%s'\n", 545 __func__, scheduled); 546 abort(); 547 } 548 549 /* The coroutine might run and release the last ctx reference before we 550 * invoke qemu_bh_schedule(). Take a reference to keep ctx alive until 551 * we're done. 552 */ 553 aio_context_ref(ctx); 554 555 QSLIST_INSERT_HEAD_ATOMIC(&ctx->scheduled_coroutines, 556 co, co_scheduled_next); 557 qemu_bh_schedule(ctx->co_schedule_bh); 558 559 aio_context_unref(ctx); 560 } 561 562 void aio_co_wake(struct Coroutine *co) 563 { 564 AioContext *ctx; 565 566 /* Read coroutine before co->ctx. Matches smp_wmb in 567 * qemu_coroutine_enter. 568 */ 569 smp_read_barrier_depends(); 570 ctx = atomic_read(&co->ctx); 571 572 aio_co_enter(ctx, co); 573 } 574 575 void aio_co_enter(AioContext *ctx, struct Coroutine *co) 576 { 577 if (ctx != qemu_get_current_aio_context()) { 578 aio_co_schedule(ctx, co); 579 return; 580 } 581 582 if (qemu_in_coroutine()) { 583 Coroutine *self = qemu_coroutine_self(); 584 assert(self != co); 585 QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, co, co_queue_next); 586 } else { 587 aio_context_acquire(ctx); 588 qemu_aio_coroutine_enter(ctx, co); 589 aio_context_release(ctx); 590 } 591 } 592 593 void aio_context_ref(AioContext *ctx) 594 { 595 g_source_ref(&ctx->source); 596 } 597 598 void aio_context_unref(AioContext *ctx) 599 { 600 g_source_unref(&ctx->source); 601 } 602 603 void aio_context_acquire(AioContext *ctx) 604 { 605 qemu_rec_mutex_lock(&ctx->lock); 606 } 607 608 void aio_context_release(AioContext *ctx) 609 { 610 qemu_rec_mutex_unlock(&ctx->lock); 611 } 612