1 /* 2 * QEMU aio implementation 3 * 4 * Copyright IBM, Corp. 2008 5 * 6 * Authors: 7 * Anthony Liguori <aliguori@us.ibm.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2. See 10 * the COPYING file in the top-level directory. 11 * 12 * Contributions after 2012-01-13 are licensed under the terms of the 13 * GNU GPL, version 2 or (at your option) any later version. 14 */ 15 16 #include "qemu/osdep.h" 17 #include "qemu-common.h" 18 #include "block/block.h" 19 #include "qemu/rcu_queue.h" 20 #include "qemu/sockets.h" 21 #include "qemu/cutils.h" 22 #include "trace.h" 23 #ifdef CONFIG_EPOLL_CREATE1 24 #include <sys/epoll.h> 25 #endif 26 27 struct AioHandler 28 { 29 GPollFD pfd; 30 IOHandler *io_read; 31 IOHandler *io_write; 32 AioPollFn *io_poll; 33 IOHandler *io_poll_begin; 34 IOHandler *io_poll_end; 35 int deleted; 36 void *opaque; 37 bool is_external; 38 QLIST_ENTRY(AioHandler) node; 39 }; 40 41 #ifdef CONFIG_EPOLL_CREATE1 42 43 /* The fd number threashold to switch to epoll */ 44 #define EPOLL_ENABLE_THRESHOLD 64 45 46 static void aio_epoll_disable(AioContext *ctx) 47 { 48 ctx->epoll_available = false; 49 if (!ctx->epoll_enabled) { 50 return; 51 } 52 ctx->epoll_enabled = false; 53 close(ctx->epollfd); 54 } 55 56 static inline int epoll_events_from_pfd(int pfd_events) 57 { 58 return (pfd_events & G_IO_IN ? EPOLLIN : 0) | 59 (pfd_events & G_IO_OUT ? EPOLLOUT : 0) | 60 (pfd_events & G_IO_HUP ? EPOLLHUP : 0) | 61 (pfd_events & G_IO_ERR ? EPOLLERR : 0); 62 } 63 64 static bool aio_epoll_try_enable(AioContext *ctx) 65 { 66 AioHandler *node; 67 struct epoll_event event; 68 69 QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) { 70 int r; 71 if (node->deleted || !node->pfd.events) { 72 continue; 73 } 74 event.events = epoll_events_from_pfd(node->pfd.events); 75 event.data.ptr = node; 76 r = epoll_ctl(ctx->epollfd, EPOLL_CTL_ADD, node->pfd.fd, &event); 77 if (r) { 78 return false; 79 } 80 } 81 ctx->epoll_enabled = true; 82 return true; 83 } 84 85 static void aio_epoll_update(AioContext *ctx, AioHandler *node, bool is_new) 86 { 87 struct epoll_event event; 88 int r; 89 int ctl; 90 91 if (!ctx->epoll_enabled) { 92 return; 93 } 94 if (!node->pfd.events) { 95 ctl = EPOLL_CTL_DEL; 96 } else { 97 event.data.ptr = node; 98 event.events = epoll_events_from_pfd(node->pfd.events); 99 ctl = is_new ? EPOLL_CTL_ADD : EPOLL_CTL_MOD; 100 } 101 102 r = epoll_ctl(ctx->epollfd, ctl, node->pfd.fd, &event); 103 if (r) { 104 aio_epoll_disable(ctx); 105 } 106 } 107 108 static int aio_epoll(AioContext *ctx, GPollFD *pfds, 109 unsigned npfd, int64_t timeout) 110 { 111 AioHandler *node; 112 int i, ret = 0; 113 struct epoll_event events[128]; 114 115 assert(npfd == 1); 116 assert(pfds[0].fd == ctx->epollfd); 117 if (timeout > 0) { 118 ret = qemu_poll_ns(pfds, npfd, timeout); 119 } 120 if (timeout <= 0 || ret > 0) { 121 ret = epoll_wait(ctx->epollfd, events, 122 sizeof(events) / sizeof(events[0]), 123 timeout); 124 if (ret <= 0) { 125 goto out; 126 } 127 for (i = 0; i < ret; i++) { 128 int ev = events[i].events; 129 node = events[i].data.ptr; 130 node->pfd.revents = (ev & EPOLLIN ? G_IO_IN : 0) | 131 (ev & EPOLLOUT ? G_IO_OUT : 0) | 132 (ev & EPOLLHUP ? G_IO_HUP : 0) | 133 (ev & EPOLLERR ? G_IO_ERR : 0); 134 } 135 } 136 out: 137 return ret; 138 } 139 140 static bool aio_epoll_enabled(AioContext *ctx) 141 { 142 /* Fall back to ppoll when external clients are disabled. */ 143 return !aio_external_disabled(ctx) && ctx->epoll_enabled; 144 } 145 146 static bool aio_epoll_check_poll(AioContext *ctx, GPollFD *pfds, 147 unsigned npfd, int64_t timeout) 148 { 149 if (!ctx->epoll_available) { 150 return false; 151 } 152 if (aio_epoll_enabled(ctx)) { 153 return true; 154 } 155 if (npfd >= EPOLL_ENABLE_THRESHOLD) { 156 if (aio_epoll_try_enable(ctx)) { 157 return true; 158 } else { 159 aio_epoll_disable(ctx); 160 } 161 } 162 return false; 163 } 164 165 #else 166 167 static void aio_epoll_update(AioContext *ctx, AioHandler *node, bool is_new) 168 { 169 } 170 171 static int aio_epoll(AioContext *ctx, GPollFD *pfds, 172 unsigned npfd, int64_t timeout) 173 { 174 assert(false); 175 } 176 177 static bool aio_epoll_enabled(AioContext *ctx) 178 { 179 return false; 180 } 181 182 static bool aio_epoll_check_poll(AioContext *ctx, GPollFD *pfds, 183 unsigned npfd, int64_t timeout) 184 { 185 return false; 186 } 187 188 #endif 189 190 static AioHandler *find_aio_handler(AioContext *ctx, int fd) 191 { 192 AioHandler *node; 193 194 QLIST_FOREACH(node, &ctx->aio_handlers, node) { 195 if (node->pfd.fd == fd) 196 if (!node->deleted) 197 return node; 198 } 199 200 return NULL; 201 } 202 203 void aio_set_fd_handler(AioContext *ctx, 204 int fd, 205 bool is_external, 206 IOHandler *io_read, 207 IOHandler *io_write, 208 AioPollFn *io_poll, 209 void *opaque) 210 { 211 AioHandler *node; 212 bool is_new = false; 213 bool deleted = false; 214 215 qemu_lockcnt_lock(&ctx->list_lock); 216 217 node = find_aio_handler(ctx, fd); 218 219 /* Are we deleting the fd handler? */ 220 if (!io_read && !io_write && !io_poll) { 221 if (node == NULL) { 222 qemu_lockcnt_unlock(&ctx->list_lock); 223 return; 224 } 225 226 /* If the GSource is in the process of being destroyed then 227 * g_source_remove_poll() causes an assertion failure. Skip 228 * removal in that case, because glib cleans up its state during 229 * destruction anyway. 230 */ 231 if (!g_source_is_destroyed(&ctx->source)) { 232 g_source_remove_poll(&ctx->source, &node->pfd); 233 } 234 235 /* If the lock is held, just mark the node as deleted */ 236 if (qemu_lockcnt_count(&ctx->list_lock)) { 237 node->deleted = 1; 238 node->pfd.revents = 0; 239 } else { 240 /* Otherwise, delete it for real. We can't just mark it as 241 * deleted because deleted nodes are only cleaned up while 242 * no one is walking the handlers list. 243 */ 244 QLIST_REMOVE(node, node); 245 deleted = true; 246 } 247 248 if (!node->io_poll) { 249 ctx->poll_disable_cnt--; 250 } 251 } else { 252 if (node == NULL) { 253 /* Alloc and insert if it's not already there */ 254 node = g_new0(AioHandler, 1); 255 node->pfd.fd = fd; 256 QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node); 257 258 g_source_add_poll(&ctx->source, &node->pfd); 259 is_new = true; 260 261 ctx->poll_disable_cnt += !io_poll; 262 } else { 263 ctx->poll_disable_cnt += !io_poll - !node->io_poll; 264 } 265 266 /* Update handler with latest information */ 267 node->io_read = io_read; 268 node->io_write = io_write; 269 node->io_poll = io_poll; 270 node->opaque = opaque; 271 node->is_external = is_external; 272 273 node->pfd.events = (io_read ? G_IO_IN | G_IO_HUP | G_IO_ERR : 0); 274 node->pfd.events |= (io_write ? G_IO_OUT | G_IO_ERR : 0); 275 } 276 277 aio_epoll_update(ctx, node, is_new); 278 qemu_lockcnt_unlock(&ctx->list_lock); 279 aio_notify(ctx); 280 281 if (deleted) { 282 g_free(node); 283 } 284 } 285 286 void aio_set_fd_poll(AioContext *ctx, int fd, 287 IOHandler *io_poll_begin, 288 IOHandler *io_poll_end) 289 { 290 AioHandler *node = find_aio_handler(ctx, fd); 291 292 if (!node) { 293 return; 294 } 295 296 node->io_poll_begin = io_poll_begin; 297 node->io_poll_end = io_poll_end; 298 } 299 300 void aio_set_event_notifier(AioContext *ctx, 301 EventNotifier *notifier, 302 bool is_external, 303 EventNotifierHandler *io_read, 304 AioPollFn *io_poll) 305 { 306 aio_set_fd_handler(ctx, event_notifier_get_fd(notifier), is_external, 307 (IOHandler *)io_read, NULL, io_poll, notifier); 308 } 309 310 void aio_set_event_notifier_poll(AioContext *ctx, 311 EventNotifier *notifier, 312 EventNotifierHandler *io_poll_begin, 313 EventNotifierHandler *io_poll_end) 314 { 315 aio_set_fd_poll(ctx, event_notifier_get_fd(notifier), 316 (IOHandler *)io_poll_begin, 317 (IOHandler *)io_poll_end); 318 } 319 320 static void poll_set_started(AioContext *ctx, bool started) 321 { 322 AioHandler *node; 323 324 if (started == ctx->poll_started) { 325 return; 326 } 327 328 ctx->poll_started = started; 329 330 qemu_lockcnt_inc(&ctx->list_lock); 331 QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) { 332 IOHandler *fn; 333 334 if (node->deleted) { 335 continue; 336 } 337 338 if (started) { 339 fn = node->io_poll_begin; 340 } else { 341 fn = node->io_poll_end; 342 } 343 344 if (fn) { 345 fn(node->opaque); 346 } 347 } 348 qemu_lockcnt_dec(&ctx->list_lock); 349 } 350 351 352 bool aio_prepare(AioContext *ctx) 353 { 354 /* Poll mode cannot be used with glib's event loop, disable it. */ 355 poll_set_started(ctx, false); 356 357 return false; 358 } 359 360 bool aio_pending(AioContext *ctx) 361 { 362 AioHandler *node; 363 bool result = false; 364 365 /* 366 * We have to walk very carefully in case aio_set_fd_handler is 367 * called while we're walking. 368 */ 369 qemu_lockcnt_inc(&ctx->list_lock); 370 371 QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) { 372 int revents; 373 374 revents = node->pfd.revents & node->pfd.events; 375 if (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR) && node->io_read && 376 aio_node_check(ctx, node->is_external)) { 377 result = true; 378 break; 379 } 380 if (revents & (G_IO_OUT | G_IO_ERR) && node->io_write && 381 aio_node_check(ctx, node->is_external)) { 382 result = true; 383 break; 384 } 385 } 386 qemu_lockcnt_dec(&ctx->list_lock); 387 388 return result; 389 } 390 391 static bool aio_dispatch_handlers(AioContext *ctx) 392 { 393 AioHandler *node, *tmp; 394 bool progress = false; 395 396 QLIST_FOREACH_SAFE_RCU(node, &ctx->aio_handlers, node, tmp) { 397 int revents; 398 399 revents = node->pfd.revents & node->pfd.events; 400 node->pfd.revents = 0; 401 402 if (!node->deleted && 403 (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) && 404 aio_node_check(ctx, node->is_external) && 405 node->io_read) { 406 node->io_read(node->opaque); 407 408 /* aio_notify() does not count as progress */ 409 if (node->opaque != &ctx->notifier) { 410 progress = true; 411 } 412 } 413 if (!node->deleted && 414 (revents & (G_IO_OUT | G_IO_ERR)) && 415 aio_node_check(ctx, node->is_external) && 416 node->io_write) { 417 node->io_write(node->opaque); 418 progress = true; 419 } 420 421 if (node->deleted) { 422 if (qemu_lockcnt_dec_if_lock(&ctx->list_lock)) { 423 QLIST_REMOVE(node, node); 424 g_free(node); 425 qemu_lockcnt_inc_and_unlock(&ctx->list_lock); 426 } 427 } 428 } 429 430 return progress; 431 } 432 433 void aio_dispatch(AioContext *ctx) 434 { 435 qemu_lockcnt_inc(&ctx->list_lock); 436 aio_bh_poll(ctx); 437 aio_dispatch_handlers(ctx); 438 qemu_lockcnt_dec(&ctx->list_lock); 439 440 timerlistgroup_run_timers(&ctx->tlg); 441 } 442 443 /* These thread-local variables are used only in a small part of aio_poll 444 * around the call to the poll() system call. In particular they are not 445 * used while aio_poll is performing callbacks, which makes it much easier 446 * to think about reentrancy! 447 * 448 * Stack-allocated arrays would be perfect but they have size limitations; 449 * heap allocation is expensive enough that we want to reuse arrays across 450 * calls to aio_poll(). And because poll() has to be called without holding 451 * any lock, the arrays cannot be stored in AioContext. Thread-local data 452 * has none of the disadvantages of these three options. 453 */ 454 static __thread GPollFD *pollfds; 455 static __thread AioHandler **nodes; 456 static __thread unsigned npfd, nalloc; 457 static __thread Notifier pollfds_cleanup_notifier; 458 459 static void pollfds_cleanup(Notifier *n, void *unused) 460 { 461 g_assert(npfd == 0); 462 g_free(pollfds); 463 g_free(nodes); 464 nalloc = 0; 465 } 466 467 static void add_pollfd(AioHandler *node) 468 { 469 if (npfd == nalloc) { 470 if (nalloc == 0) { 471 pollfds_cleanup_notifier.notify = pollfds_cleanup; 472 qemu_thread_atexit_add(&pollfds_cleanup_notifier); 473 nalloc = 8; 474 } else { 475 g_assert(nalloc <= INT_MAX); 476 nalloc *= 2; 477 } 478 pollfds = g_renew(GPollFD, pollfds, nalloc); 479 nodes = g_renew(AioHandler *, nodes, nalloc); 480 } 481 nodes[npfd] = node; 482 pollfds[npfd] = (GPollFD) { 483 .fd = node->pfd.fd, 484 .events = node->pfd.events, 485 }; 486 npfd++; 487 } 488 489 static bool run_poll_handlers_once(AioContext *ctx) 490 { 491 bool progress = false; 492 AioHandler *node; 493 494 QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) { 495 if (!node->deleted && node->io_poll && 496 aio_node_check(ctx, node->is_external) && 497 node->io_poll(node->opaque)) { 498 progress = true; 499 } 500 501 /* Caller handles freeing deleted nodes. Don't do it here. */ 502 } 503 504 return progress; 505 } 506 507 /* run_poll_handlers: 508 * @ctx: the AioContext 509 * @max_ns: maximum time to poll for, in nanoseconds 510 * 511 * Polls for a given time. 512 * 513 * Note that ctx->notify_me must be non-zero so this function can detect 514 * aio_notify(). 515 * 516 * Note that the caller must have incremented ctx->list_lock. 517 * 518 * Returns: true if progress was made, false otherwise 519 */ 520 static bool run_poll_handlers(AioContext *ctx, int64_t max_ns) 521 { 522 bool progress; 523 int64_t end_time; 524 525 assert(ctx->notify_me); 526 assert(qemu_lockcnt_count(&ctx->list_lock) > 0); 527 assert(ctx->poll_disable_cnt == 0); 528 529 trace_run_poll_handlers_begin(ctx, max_ns); 530 531 end_time = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + max_ns; 532 533 do { 534 progress = run_poll_handlers_once(ctx); 535 } while (!progress && qemu_clock_get_ns(QEMU_CLOCK_REALTIME) < end_time); 536 537 trace_run_poll_handlers_end(ctx, progress); 538 539 return progress; 540 } 541 542 /* try_poll_mode: 543 * @ctx: the AioContext 544 * @blocking: busy polling is only attempted when blocking is true 545 * 546 * ctx->notify_me must be non-zero so this function can detect aio_notify(). 547 * 548 * Note that the caller must have incremented ctx->list_lock. 549 * 550 * Returns: true if progress was made, false otherwise 551 */ 552 static bool try_poll_mode(AioContext *ctx, bool blocking) 553 { 554 if (blocking && ctx->poll_max_ns && ctx->poll_disable_cnt == 0) { 555 /* See qemu_soonest_timeout() uint64_t hack */ 556 int64_t max_ns = MIN((uint64_t)aio_compute_timeout(ctx), 557 (uint64_t)ctx->poll_ns); 558 559 if (max_ns) { 560 poll_set_started(ctx, true); 561 562 if (run_poll_handlers(ctx, max_ns)) { 563 return true; 564 } 565 } 566 } 567 568 poll_set_started(ctx, false); 569 570 /* Even if we don't run busy polling, try polling once in case it can make 571 * progress and the caller will be able to avoid ppoll(2)/epoll_wait(2). 572 */ 573 return run_poll_handlers_once(ctx); 574 } 575 576 bool aio_poll(AioContext *ctx, bool blocking) 577 { 578 AioHandler *node; 579 int i; 580 int ret = 0; 581 bool progress; 582 int64_t timeout; 583 int64_t start = 0; 584 585 /* aio_notify can avoid the expensive event_notifier_set if 586 * everything (file descriptors, bottom halves, timers) will 587 * be re-evaluated before the next blocking poll(). This is 588 * already true when aio_poll is called with blocking == false; 589 * if blocking == true, it is only true after poll() returns, 590 * so disable the optimization now. 591 */ 592 if (blocking) { 593 atomic_add(&ctx->notify_me, 2); 594 } 595 596 qemu_lockcnt_inc(&ctx->list_lock); 597 598 if (ctx->poll_max_ns) { 599 start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 600 } 601 602 progress = try_poll_mode(ctx, blocking); 603 if (!progress) { 604 assert(npfd == 0); 605 606 /* fill pollfds */ 607 608 if (!aio_epoll_enabled(ctx)) { 609 QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) { 610 if (!node->deleted && node->pfd.events 611 && aio_node_check(ctx, node->is_external)) { 612 add_pollfd(node); 613 } 614 } 615 } 616 617 timeout = blocking ? aio_compute_timeout(ctx) : 0; 618 619 /* wait until next event */ 620 if (aio_epoll_check_poll(ctx, pollfds, npfd, timeout)) { 621 AioHandler epoll_handler; 622 623 epoll_handler.pfd.fd = ctx->epollfd; 624 epoll_handler.pfd.events = G_IO_IN | G_IO_OUT | G_IO_HUP | G_IO_ERR; 625 npfd = 0; 626 add_pollfd(&epoll_handler); 627 ret = aio_epoll(ctx, pollfds, npfd, timeout); 628 } else { 629 ret = qemu_poll_ns(pollfds, npfd, timeout); 630 } 631 } 632 633 if (blocking) { 634 atomic_sub(&ctx->notify_me, 2); 635 } 636 637 /* Adjust polling time */ 638 if (ctx->poll_max_ns) { 639 int64_t block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start; 640 641 if (block_ns <= ctx->poll_ns) { 642 /* This is the sweet spot, no adjustment needed */ 643 } else if (block_ns > ctx->poll_max_ns) { 644 /* We'd have to poll for too long, poll less */ 645 int64_t old = ctx->poll_ns; 646 647 if (ctx->poll_shrink) { 648 ctx->poll_ns /= ctx->poll_shrink; 649 } else { 650 ctx->poll_ns = 0; 651 } 652 653 trace_poll_shrink(ctx, old, ctx->poll_ns); 654 } else if (ctx->poll_ns < ctx->poll_max_ns && 655 block_ns < ctx->poll_max_ns) { 656 /* There is room to grow, poll longer */ 657 int64_t old = ctx->poll_ns; 658 int64_t grow = ctx->poll_grow; 659 660 if (grow == 0) { 661 grow = 2; 662 } 663 664 if (ctx->poll_ns) { 665 ctx->poll_ns *= grow; 666 } else { 667 ctx->poll_ns = 4000; /* start polling at 4 microseconds */ 668 } 669 670 if (ctx->poll_ns > ctx->poll_max_ns) { 671 ctx->poll_ns = ctx->poll_max_ns; 672 } 673 674 trace_poll_grow(ctx, old, ctx->poll_ns); 675 } 676 } 677 678 aio_notify_accept(ctx); 679 680 /* if we have any readable fds, dispatch event */ 681 if (ret > 0) { 682 for (i = 0; i < npfd; i++) { 683 nodes[i]->pfd.revents = pollfds[i].revents; 684 } 685 } 686 687 npfd = 0; 688 689 progress |= aio_bh_poll(ctx); 690 691 if (ret > 0) { 692 progress |= aio_dispatch_handlers(ctx); 693 } 694 695 qemu_lockcnt_dec(&ctx->list_lock); 696 697 progress |= timerlistgroup_run_timers(&ctx->tlg); 698 699 return progress; 700 } 701 702 void aio_context_setup(AioContext *ctx) 703 { 704 /* TODO remove this in final patch submission */ 705 if (getenv("QEMU_AIO_POLL_MAX_NS")) { 706 fprintf(stderr, "The QEMU_AIO_POLL_MAX_NS environment variable has " 707 "been replaced with -object iothread,poll-max-ns=NUM\n"); 708 exit(1); 709 } 710 711 #ifdef CONFIG_EPOLL_CREATE1 712 assert(!ctx->epollfd); 713 ctx->epollfd = epoll_create1(EPOLL_CLOEXEC); 714 if (ctx->epollfd == -1) { 715 fprintf(stderr, "Failed to create epoll instance: %s", strerror(errno)); 716 ctx->epoll_available = false; 717 } else { 718 ctx->epoll_available = true; 719 } 720 #endif 721 } 722 723 void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns, 724 int64_t grow, int64_t shrink, Error **errp) 725 { 726 /* No thread synchronization here, it doesn't matter if an incorrect value 727 * is used once. 728 */ 729 ctx->poll_max_ns = max_ns; 730 ctx->poll_ns = 0; 731 ctx->poll_grow = grow; 732 ctx->poll_shrink = shrink; 733 734 aio_notify(ctx); 735 } 736