util/main-loop.c

   1 /*
   2  * QEMU System Emulator
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   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
  25 #include "qemu/osdep.h"
  26 #include "qapi/error.h"
  27 #include "qemu/cutils.h"
  28 #include "qemu/timer.h"
  29 #include "sysemu/cpu-timers.h"
  30 #include "sysemu/replay.h"
  31 #include "qemu/main-loop.h"
  32 #include "block/aio.h"
  33 #include "block/thread-pool.h"
  34 #include "qemu/error-report.h"
  35 #include "qemu/queue.h"
  36 #include "qemu/compiler.h"
  37 #include "qom/object.h"
  38
  39 #ifndef _WIN32
  40 #include <sys/wait.h>
  41 #endif
  42
  43 #ifndef _WIN32
  44
  45 /* If we have signalfd, we mask out the signals we want to handle and then
  46  * use signalfd to listen for them.  We rely on whatever the current signal
  47  * handler is to dispatch the signals when we receive them.
  48  */
  49 /*
  50  * Disable CFI checks.
  51  * We are going to call a signal hander directly. Such handler may or may not
  52  * have been defined in our binary, so there's no guarantee that the pointer
  53  * used to set the handler is a cfi-valid pointer. Since the handlers are
  54  * stored in kernel memory, changing the handler to an attacker-defined
  55  * function requires being able to call a sigaction() syscall,
  56  * which is not as easy as overwriting a pointer in memory.
  57  */
  58 QEMU_DISABLE_CFI
  59 static void sigfd_handler(void *opaque)
  60 {
  61     int fd = (intptr_t)opaque;
  62     struct qemu_signalfd_siginfo info;
  63     struct sigaction action;
  64     ssize_t len;
  65
  66     while (1) {
  67         len = RETRY_ON_EINTR(read(fd, &info, sizeof(info)));
  68
  69         if (len == -1 && errno == EAGAIN) {
  70             break;
  71         }
  72
  73         if (len != sizeof(info)) {
  74             error_report("read from sigfd returned %zd: %s", len,
  75                          g_strerror(errno));
  76             return;
  77         }
  78
  79         sigaction(info.ssi_signo, NULL, &action);
  80         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  81             sigaction_invoke(&action, &info);
  82         } else if (action.sa_handler) {
  83             action.sa_handler(info.ssi_signo);
  84         }
  85     }
  86 }
  87
  88 static int qemu_signal_init(Error **errp)
  89 {
  90     int sigfd;
  91     sigset_t set;
  92
  93     /*
  94      * SIG_IPI must be blocked in the main thread and must not be caught
  95      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  96      * not catch it reliably.
  97      */
  98     sigemptyset(&set);
  99     sigaddset(&set, SIG_IPI);
 100     sigaddset(&set, SIGIO);
 101     sigaddset(&set, SIGALRM);
 102     sigaddset(&set, SIGBUS);
 103     /* SIGINT cannot be handled via signalfd, so that ^C can be used
 104      * to interrupt QEMU when it is being run under gdb.  SIGHUP and
 105      * SIGTERM are also handled asynchronously, even though it is not
 106      * strictly necessary, because they use the same handler as SIGINT.
 107      */
 108     pthread_sigmask(SIG_BLOCK, &set, NULL);
 109
 110     sigdelset(&set, SIG_IPI);
 111     sigfd = qemu_signalfd(&set);
 112     if (sigfd == -1) {
 113         error_setg_errno(errp, errno, "failed to create signalfd");
 114         return -errno;
 115     }
 116
 117     g_unix_set_fd_nonblocking(sigfd, true, NULL);
 118
 119     qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
 120
 121     return 0;
 122 }
 123
 124 #else /* _WIN32 */
 125
 126 static int qemu_signal_init(Error **errp)
 127 {
 128     return 0;
 129 }
 130 #endif
 131
 132 static AioContext *qemu_aio_context;
 133 static QEMUBH *qemu_notify_bh;
 134
 135 static void notify_event_cb(void *opaque)
 136 {
 137     /* No need to do anything; this bottom half is only used to
 138      * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
 139      */
 140 }
 141
 142 AioContext *qemu_get_aio_context(void)
 143 {
 144     return qemu_aio_context;
 145 }
 146
 147 void qemu_notify_event(void)
 148 {
 149     if (!qemu_aio_context) {
 150         return;
 151     }
 152     qemu_bh_schedule(qemu_notify_bh);
 153 }
 154
 155 static GArray *gpollfds;
 156
 157 int qemu_init_main_loop(Error **errp)
 158 {
 159     int ret;
 160     GSource *src;
 161
 162     init_clocks(qemu_timer_notify_cb);
 163
 164     ret = qemu_signal_init(errp);
 165     if (ret) {
 166         return ret;
 167     }
 168
 169     qemu_aio_context = aio_context_new(errp);
 170     if (!qemu_aio_context) {
 171         return -EMFILE;
 172     }
 173     qemu_set_current_aio_context(qemu_aio_context);
 174     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 175     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 176     src = aio_get_g_source(qemu_aio_context);
 177     g_source_set_name(src, "aio-context");
 178     g_source_attach(src, NULL);
 179     g_source_unref(src);
 180     src = iohandler_get_g_source();
 181     g_source_set_name(src, "io-handler");
 182     g_source_attach(src, NULL);
 183     g_source_unref(src);
 184     return 0;
 185 }
 186
 187 static void main_loop_update_params(EventLoopBase *base, Error **errp)
 188 {
 189     ERRP_GUARD();
 190
 191     if (!qemu_aio_context) {
 192         error_setg(errp, "qemu aio context not ready");
 193         return;
 194     }
 195
 196     aio_context_set_aio_params(qemu_aio_context, base->aio_max_batch, errp);
 197     if (*errp) {
 198         return;
 199     }
 200
 201     aio_context_set_thread_pool_params(qemu_aio_context, base->thread_pool_min,
 202                                        base->thread_pool_max, errp);
 203 }
 204
 205 MainLoop *mloop;
 206
 207 static void main_loop_init(EventLoopBase *base, Error **errp)
 208 {
 209     MainLoop *m = MAIN_LOOP(base);
 210
 211     if (mloop) {
 212         error_setg(errp, "only one main-loop instance allowed");
 213         return;
 214     }
 215
 216     main_loop_update_params(base, errp);
 217
 218     mloop = m;
 219     return;
 220 }
 221
 222 static bool main_loop_can_be_deleted(EventLoopBase *base)
 223 {
 224     return false;
 225 }
 226
 227 static void main_loop_class_init(ObjectClass *oc, void *class_data)
 228 {
 229     EventLoopBaseClass *bc = EVENT_LOOP_BASE_CLASS(oc);
 230
 231     bc->init = main_loop_init;
 232     bc->update_params = main_loop_update_params;
 233     bc->can_be_deleted = main_loop_can_be_deleted;
 234 }
 235
 236 static const TypeInfo main_loop_info = {
 237     .name = TYPE_MAIN_LOOP,
 238     .parent = TYPE_EVENT_LOOP_BASE,
 239     .class_init = main_loop_class_init,
 240     .instance_size = sizeof(MainLoop),
 241 };
 242
 243 static void main_loop_register_types(void)
 244 {
 245     type_register_static(&main_loop_info);
 246 }
 247
 248 type_init(main_loop_register_types)
 249
 250 static int max_priority;
 251
 252 #ifndef _WIN32
 253 static int glib_pollfds_idx;
 254 static int glib_n_poll_fds;
 255
 256 void qemu_fd_register(int fd)
 257 {
 258 }
 259
 260 static void glib_pollfds_fill(int64_t *cur_timeout)
 261 {
 262     GMainContext *context = g_main_context_default();
 263     int timeout = 0;
 264     int64_t timeout_ns;
 265     int n;
 266
 267     g_main_context_prepare(context, &max_priority);
 268
 269     glib_pollfds_idx = gpollfds->len;
 270     n = glib_n_poll_fds;
 271     do {
 272         GPollFD *pfds;
 273         glib_n_poll_fds = n;
 274         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 275         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 276         n = g_main_context_query(context, max_priority, &timeout, pfds,
 277                                  glib_n_poll_fds);
 278     } while (n != glib_n_poll_fds);
 279
 280     if (timeout < 0) {
 281         timeout_ns = -1;
 282     } else {
 283         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 284     }
 285
 286     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 287 }
 288
 289 static void glib_pollfds_poll(void)
 290 {
 291     GMainContext *context = g_main_context_default();
 292     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 293
 294     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 295         g_main_context_dispatch(context);
 296     }
 297 }
 298
 299 #define MAX_MAIN_LOOP_SPIN (1000)
 300
 301 static int os_host_main_loop_wait(int64_t timeout)
 302 {
 303     GMainContext *context = g_main_context_default();
 304     int ret;
 305
 306     g_main_context_acquire(context);
 307
 308     glib_pollfds_fill(&timeout);
 309
 310     qemu_mutex_unlock_iothread();
 311     replay_mutex_unlock();
 312
 313     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 314
 315     replay_mutex_lock();
 316     qemu_mutex_lock_iothread();
 317
 318     glib_pollfds_poll();
 319
 320     g_main_context_release(context);
 321
 322     return ret;
 323 }
 324 #else
 325 /***********************************************************/
 326 /* Polling handling */
 327
 328 typedef struct PollingEntry {
 329     PollingFunc *func;
 330     void *opaque;
 331     struct PollingEntry *next;
 332 } PollingEntry;
 333
 334 static PollingEntry *first_polling_entry;
 335
 336 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 337 {
 338     PollingEntry **ppe, *pe;
 339     pe = g_new0(PollingEntry, 1);
 340     pe->func = func;
 341     pe->opaque = opaque;
 342     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 343     *ppe = pe;
 344     return 0;
 345 }
 346
 347 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 348 {
 349     PollingEntry **ppe, *pe;
 350     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 351         pe = *ppe;
 352         if (pe->func == func && pe->opaque == opaque) {
 353             *ppe = pe->next;
 354             g_free(pe);
 355             break;
 356         }
 357     }
 358 }
 359
 360 /***********************************************************/
 361 /* Wait objects support */
 362 typedef struct WaitObjects {
 363     int num;
 364     int revents[MAXIMUM_WAIT_OBJECTS];
 365     HANDLE events[MAXIMUM_WAIT_OBJECTS];
 366     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS];
 367     void *opaque[MAXIMUM_WAIT_OBJECTS];
 368 } WaitObjects;
 369
 370 static WaitObjects wait_objects = {0};
 371
 372 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 373 {
 374     int i;
 375     WaitObjects *w = &wait_objects;
 376
 377     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 378         return -1;
 379     }
 380
 381     for (i = 0; i < w->num; i++) {
 382         /* check if the same handle is added twice */
 383         if (w->events[i] == handle) {
 384             return -1;
 385         }
 386     }
 387
 388     w->events[w->num] = handle;
 389     w->func[w->num] = func;
 390     w->opaque[w->num] = opaque;
 391     w->revents[w->num] = 0;
 392     w->num++;
 393     return 0;
 394 }
 395
 396 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 397 {
 398     int i, found;
 399     WaitObjects *w = &wait_objects;
 400
 401     found = 0;
 402     for (i = 0; i < w->num; i++) {
 403         if (w->events[i] == handle) {
 404             found = 1;
 405         }
 406         if (found && i < (MAXIMUM_WAIT_OBJECTS - 1)) {
 407             w->events[i] = w->events[i + 1];
 408             w->func[i] = w->func[i + 1];
 409             w->opaque[i] = w->opaque[i + 1];
 410             w->revents[i] = w->revents[i + 1];
 411         }
 412     }
 413     if (found) {
 414         w->num--;
 415     }
 416 }
 417
 418 void qemu_fd_register(int fd)
 419 {
 420     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 421                    FD_READ | FD_ACCEPT | FD_CLOSE |
 422                    FD_CONNECT | FD_WRITE | FD_OOB);
 423 }
 424
 425 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 426                         fd_set *xfds)
 427 {
 428     int nfds = -1;
 429     int i;
 430
 431     for (i = 0; i < pollfds->len; i++) {
 432         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 433         int fd = pfd->fd;
 434         int events = pfd->events;
 435         if (events & G_IO_IN) {
 436             FD_SET(fd, rfds);
 437             nfds = MAX(nfds, fd);
 438         }
 439         if (events & G_IO_OUT) {
 440             FD_SET(fd, wfds);
 441             nfds = MAX(nfds, fd);
 442         }
 443         if (events & G_IO_PRI) {
 444             FD_SET(fd, xfds);
 445             nfds = MAX(nfds, fd);
 446         }
 447     }
 448     return nfds;
 449 }
 450
 451 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 452                          fd_set *wfds, fd_set *xfds)
 453 {
 454     int i;
 455
 456     for (i = 0; i < pollfds->len; i++) {
 457         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 458         int fd = pfd->fd;
 459         int revents = 0;
 460
 461         if (FD_ISSET(fd, rfds)) {
 462             revents |= G_IO_IN;
 463         }
 464         if (FD_ISSET(fd, wfds)) {
 465             revents |= G_IO_OUT;
 466         }
 467         if (FD_ISSET(fd, xfds)) {
 468             revents |= G_IO_PRI;
 469         }
 470         pfd->revents = revents & pfd->events;
 471     }
 472 }
 473
 474 static int os_host_main_loop_wait(int64_t timeout)
 475 {
 476     GMainContext *context = g_main_context_default();
 477     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 478     int select_ret = 0;
 479     int g_poll_ret, ret, i, n_poll_fds;
 480     PollingEntry *pe;
 481     WaitObjects *w = &wait_objects;
 482     gint poll_timeout;
 483     int64_t poll_timeout_ns;
 484     static struct timeval tv0;
 485     fd_set rfds, wfds, xfds;
 486     int nfds;
 487
 488     g_main_context_acquire(context);
 489
 490     /* XXX: need to suppress polling by better using win32 events */
 491     ret = 0;
 492     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 493         ret |= pe->func(pe->opaque);
 494     }
 495     if (ret != 0) {
 496         g_main_context_release(context);
 497         return ret;
 498     }
 499
 500     FD_ZERO(&rfds);
 501     FD_ZERO(&wfds);
 502     FD_ZERO(&xfds);
 503     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 504     if (nfds >= 0) {
 505         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 506         if (select_ret != 0) {
 507             timeout = 0;
 508         }
 509         if (select_ret > 0) {
 510             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 511         }
 512     }
 513
 514     g_main_context_prepare(context, &max_priority);
 515     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 516                                       poll_fds, ARRAY_SIZE(poll_fds));
 517     g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
 518
 519     for (i = 0; i < w->num; i++) {
 520         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 521         poll_fds[n_poll_fds + i].events = G_IO_IN;
 522     }
 523
 524     if (poll_timeout < 0) {
 525         poll_timeout_ns = -1;
 526     } else {
 527         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 528     }
 529
 530     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 531
 532     qemu_mutex_unlock_iothread();
 533
 534     replay_mutex_unlock();
 535
 536     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 537
 538     replay_mutex_lock();
 539
 540     qemu_mutex_lock_iothread();
 541     if (g_poll_ret > 0) {
 542         for (i = 0; i < w->num; i++) {
 543             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 544         }
 545         for (i = 0; i < w->num; i++) {
 546             if (w->revents[i] && w->func[i]) {
 547                 w->func[i](w->opaque[i]);
 548             }
 549         }
 550     }
 551
 552     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 553         g_main_context_dispatch(context);
 554     }
 555
 556     g_main_context_release(context);
 557
 558     return select_ret || g_poll_ret;
 559 }
 560 #endif
 561
 562 static NotifierList main_loop_poll_notifiers =
 563     NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
 564
 565 void main_loop_poll_add_notifier(Notifier *notify)
 566 {
 567     notifier_list_add(&main_loop_poll_notifiers, notify);
 568 }
 569
 570 void main_loop_poll_remove_notifier(Notifier *notify)
 571 {
 572     notifier_remove(notify);
 573 }
 574
 575 void main_loop_wait(int nonblocking)
 576 {
 577     MainLoopPoll mlpoll = {
 578         .state = MAIN_LOOP_POLL_FILL,
 579         .timeout = UINT32_MAX,
 580         .pollfds = gpollfds,
 581     };
 582     int ret;
 583     int64_t timeout_ns;
 584
 585     if (nonblocking) {
 586         mlpoll.timeout = 0;
 587     }
 588
 589     /* poll any events */
 590     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 591     /* XXX: separate device handlers from system ones */
 592     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 593
 594     if (mlpoll.timeout == UINT32_MAX) {
 595         timeout_ns = -1;
 596     } else {
 597         timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
 598     }
 599
 600     timeout_ns = qemu_soonest_timeout(timeout_ns,
 601                                       timerlistgroup_deadline_ns(
 602                                           &main_loop_tlg));
 603
 604     ret = os_host_main_loop_wait(timeout_ns);
 605     mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
 606     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 607
 608     if (icount_enabled()) {
 609         /*
 610          * CPU thread can infinitely wait for event after
 611          * missing the warp
 612          */
 613         icount_start_warp_timer();
 614     }
 615     qemu_clock_run_all_timers();
 616 }
 617
 618 /* Functions to operate on the main QEMU AioContext.  */
 619
 620 QEMUBH *qemu_bh_new_full(QEMUBHFunc *cb, void *opaque, const char *name)
 621 {
 622     return aio_bh_new_full(qemu_aio_context, cb, opaque, name);
 623 }
 624
 625 /*
 626  * Functions to operate on the I/O handler AioContext.
 627  * This context runs on top of main loop. We can't reuse qemu_aio_context
 628  * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
 629  */
 630 static AioContext *iohandler_ctx;
 631
 632 static void iohandler_init(void)
 633 {
 634     if (!iohandler_ctx) {
 635         iohandler_ctx = aio_context_new(&error_abort);
 636     }
 637 }
 638
 639 AioContext *iohandler_get_aio_context(void)
 640 {
 641     iohandler_init();
 642     return iohandler_ctx;
 643 }
 644
 645 GSource *iohandler_get_g_source(void)
 646 {
 647     iohandler_init();
 648     return aio_get_g_source(iohandler_ctx);
 649 }
 650
 651 void qemu_set_fd_handler(int fd,
 652                          IOHandler *fd_read,
 653                          IOHandler *fd_write,
 654                          void *opaque)
 655 {
 656     iohandler_init();
 657     aio_set_fd_handler(iohandler_ctx, fd, false,
 658                        fd_read, fd_write, NULL, NULL, opaque);
 659 }
 660
 661 void event_notifier_set_handler(EventNotifier *e,
 662                                 EventNotifierHandler *handler)
 663 {
 664     iohandler_init();
 665     aio_set_event_notifier(iohandler_ctx, e, false,
 666                            handler, NULL, NULL);
 667 }