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