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 handler 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 static void glib_pollfds_fill(int64_t *cur_timeout)
 256 {
 257     GMainContext *context = g_main_context_default();
 258     int timeout = 0;
 259     int64_t timeout_ns;
 260     int n;
 261
 262     g_main_context_prepare(context, &max_priority);
 263
 264     glib_pollfds_idx = gpollfds->len;
 265     n = glib_n_poll_fds;
 266     do {
 267         GPollFD *pfds;
 268         glib_n_poll_fds = n;
 269         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 270         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 271         n = g_main_context_query(context, max_priority, &timeout, pfds,
 272                                  glib_n_poll_fds);
 273     } while (n != glib_n_poll_fds);
 274
 275     if (timeout < 0) {
 276         timeout_ns = -1;
 277     } else {
 278         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 279     }
 280
 281     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 282 }
 283
 284 static void glib_pollfds_poll(void)
 285 {
 286     GMainContext *context = g_main_context_default();
 287     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 288
 289     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 290         g_main_context_dispatch(context);
 291     }
 292 }
 293
 294 #define MAX_MAIN_LOOP_SPIN (1000)
 295
 296 static int os_host_main_loop_wait(int64_t timeout)
 297 {
 298     GMainContext *context = g_main_context_default();
 299     int ret;
 300
 301     g_main_context_acquire(context);
 302
 303     glib_pollfds_fill(&timeout);
 304
 305     qemu_mutex_unlock_iothread();
 306     replay_mutex_unlock();
 307
 308     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 309
 310     replay_mutex_lock();
 311     qemu_mutex_lock_iothread();
 312
 313     glib_pollfds_poll();
 314
 315     g_main_context_release(context);
 316
 317     return ret;
 318 }
 319 #else
 320 /***********************************************************/
 321 /* Polling handling */
 322
 323 typedef struct PollingEntry {
 324     PollingFunc *func;
 325     void *opaque;
 326     struct PollingEntry *next;
 327 } PollingEntry;
 328
 329 static PollingEntry *first_polling_entry;
 330
 331 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 332 {
 333     PollingEntry **ppe, *pe;
 334     pe = g_new0(PollingEntry, 1);
 335     pe->func = func;
 336     pe->opaque = opaque;
 337     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 338     *ppe = pe;
 339     return 0;
 340 }
 341
 342 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 343 {
 344     PollingEntry **ppe, *pe;
 345     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 346         pe = *ppe;
 347         if (pe->func == func && pe->opaque == opaque) {
 348             *ppe = pe->next;
 349             g_free(pe);
 350             break;
 351         }
 352     }
 353 }
 354
 355 /***********************************************************/
 356 /* Wait objects support */
 357 typedef struct WaitObjects {
 358     int num;
 359     int revents[MAXIMUM_WAIT_OBJECTS];
 360     HANDLE events[MAXIMUM_WAIT_OBJECTS];
 361     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS];
 362     void *opaque[MAXIMUM_WAIT_OBJECTS];
 363 } WaitObjects;
 364
 365 static WaitObjects wait_objects = {0};
 366
 367 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 368 {
 369     int i;
 370     WaitObjects *w = &wait_objects;
 371
 372     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 373         return -1;
 374     }
 375
 376     for (i = 0; i < w->num; i++) {
 377         /* check if the same handle is added twice */
 378         if (w->events[i] == handle) {
 379             return -1;
 380         }
 381     }
 382
 383     w->events[w->num] = handle;
 384     w->func[w->num] = func;
 385     w->opaque[w->num] = opaque;
 386     w->revents[w->num] = 0;
 387     w->num++;
 388     return 0;
 389 }
 390
 391 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 392 {
 393     int i, found;
 394     WaitObjects *w = &wait_objects;
 395
 396     found = 0;
 397     for (i = 0; i < w->num; i++) {
 398         if (w->events[i] == handle) {
 399             found = 1;
 400         }
 401         if (found && i < (MAXIMUM_WAIT_OBJECTS - 1)) {
 402             w->events[i] = w->events[i + 1];
 403             w->func[i] = w->func[i + 1];
 404             w->opaque[i] = w->opaque[i + 1];
 405             w->revents[i] = w->revents[i + 1];
 406         }
 407     }
 408     if (found) {
 409         w->num--;
 410     }
 411 }
 412
 413 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 414                         fd_set *xfds)
 415 {
 416     int nfds = -1;
 417     int i;
 418
 419     for (i = 0; i < pollfds->len; i++) {
 420         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 421         int fd = pfd->fd;
 422         int events = pfd->events;
 423         if (events & G_IO_IN) {
 424             FD_SET(fd, rfds);
 425             nfds = MAX(nfds, fd);
 426         }
 427         if (events & G_IO_OUT) {
 428             FD_SET(fd, wfds);
 429             nfds = MAX(nfds, fd);
 430         }
 431         if (events & G_IO_PRI) {
 432             FD_SET(fd, xfds);
 433             nfds = MAX(nfds, fd);
 434         }
 435     }
 436     return nfds;
 437 }
 438
 439 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 440                          fd_set *wfds, fd_set *xfds)
 441 {
 442     int i;
 443
 444     for (i = 0; i < pollfds->len; i++) {
 445         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 446         int fd = pfd->fd;
 447         int revents = 0;
 448
 449         if (FD_ISSET(fd, rfds)) {
 450             revents |= G_IO_IN;
 451         }
 452         if (FD_ISSET(fd, wfds)) {
 453             revents |= G_IO_OUT;
 454         }
 455         if (FD_ISSET(fd, xfds)) {
 456             revents |= G_IO_PRI;
 457         }
 458         pfd->revents = revents & pfd->events;
 459     }
 460 }
 461
 462 static int os_host_main_loop_wait(int64_t timeout)
 463 {
 464     GMainContext *context = g_main_context_default();
 465     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 466     int select_ret = 0;
 467     int g_poll_ret, ret, i, n_poll_fds;
 468     PollingEntry *pe;
 469     WaitObjects *w = &wait_objects;
 470     gint poll_timeout;
 471     int64_t poll_timeout_ns;
 472     static struct timeval tv0;
 473     fd_set rfds, wfds, xfds;
 474     int nfds;
 475
 476     g_main_context_acquire(context);
 477
 478     /* XXX: need to suppress polling by better using win32 events */
 479     ret = 0;
 480     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 481         ret |= pe->func(pe->opaque);
 482     }
 483     if (ret != 0) {
 484         g_main_context_release(context);
 485         return ret;
 486     }
 487
 488     FD_ZERO(&rfds);
 489     FD_ZERO(&wfds);
 490     FD_ZERO(&xfds);
 491     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 492     if (nfds >= 0) {
 493         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 494         if (select_ret != 0) {
 495             timeout = 0;
 496         }
 497         if (select_ret > 0) {
 498             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 499         }
 500     }
 501
 502     g_main_context_prepare(context, &max_priority);
 503     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 504                                       poll_fds, ARRAY_SIZE(poll_fds));
 505     g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
 506
 507     for (i = 0; i < w->num; i++) {
 508         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 509         poll_fds[n_poll_fds + i].events = G_IO_IN;
 510     }
 511
 512     if (poll_timeout < 0) {
 513         poll_timeout_ns = -1;
 514     } else {
 515         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 516     }
 517
 518     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 519
 520     qemu_mutex_unlock_iothread();
 521
 522     replay_mutex_unlock();
 523
 524     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 525
 526     replay_mutex_lock();
 527
 528     qemu_mutex_lock_iothread();
 529     if (g_poll_ret > 0) {
 530         for (i = 0; i < w->num; i++) {
 531             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 532         }
 533         for (i = 0; i < w->num; i++) {
 534             if (w->revents[i] && w->func[i]) {
 535                 w->func[i](w->opaque[i]);
 536             }
 537         }
 538     }
 539
 540     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 541         g_main_context_dispatch(context);
 542     }
 543
 544     g_main_context_release(context);
 545
 546     return select_ret || g_poll_ret;
 547 }
 548 #endif
 549
 550 static NotifierList main_loop_poll_notifiers =
 551     NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
 552
 553 void main_loop_poll_add_notifier(Notifier *notify)
 554 {
 555     notifier_list_add(&main_loop_poll_notifiers, notify);
 556 }
 557
 558 void main_loop_poll_remove_notifier(Notifier *notify)
 559 {
 560     notifier_remove(notify);
 561 }
 562
 563 void main_loop_wait(int nonblocking)
 564 {
 565     MainLoopPoll mlpoll = {
 566         .state = MAIN_LOOP_POLL_FILL,
 567         .timeout = UINT32_MAX,
 568         .pollfds = gpollfds,
 569     };
 570     int ret;
 571     int64_t timeout_ns;
 572
 573     if (nonblocking) {
 574         mlpoll.timeout = 0;
 575     }
 576
 577     /* poll any events */
 578     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 579     /* XXX: separate device handlers from system ones */
 580     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 581
 582     if (mlpoll.timeout == UINT32_MAX) {
 583         timeout_ns = -1;
 584     } else {
 585         timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
 586     }
 587
 588     timeout_ns = qemu_soonest_timeout(timeout_ns,
 589                                       timerlistgroup_deadline_ns(
 590                                           &main_loop_tlg));
 591
 592     ret = os_host_main_loop_wait(timeout_ns);
 593     mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
 594     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 595
 596     if (icount_enabled()) {
 597         /*
 598          * CPU thread can infinitely wait for event after
 599          * missing the warp
 600          */
 601         icount_start_warp_timer();
 602     }
 603     qemu_clock_run_all_timers();
 604 }
 605
 606 /* Functions to operate on the main QEMU AioContext.  */
 607
 608 QEMUBH *qemu_bh_new_full(QEMUBHFunc *cb, void *opaque, const char *name,
 609                          MemReentrancyGuard *reentrancy_guard)
 610 {
 611     return aio_bh_new_full(qemu_aio_context, cb, opaque, name,
 612                            reentrancy_guard);
 613 }
 614
 615 /*
 616  * Functions to operate on the I/O handler AioContext.
 617  * This context runs on top of main loop. We can't reuse qemu_aio_context
 618  * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
 619  */
 620 static AioContext *iohandler_ctx;
 621
 622 static void iohandler_init(void)
 623 {
 624     if (!iohandler_ctx) {
 625         iohandler_ctx = aio_context_new(&error_abort);
 626     }
 627 }
 628
 629 AioContext *iohandler_get_aio_context(void)
 630 {
 631     iohandler_init();
 632     return iohandler_ctx;
 633 }
 634
 635 GSource *iohandler_get_g_source(void)
 636 {
 637     iohandler_init();
 638     return aio_get_g_source(iohandler_ctx);
 639 }
 640
 641 void qemu_set_fd_handler(int fd,
 642                          IOHandler *fd_read,
 643                          IOHandler *fd_write,
 644                          void *opaque)
 645 {
 646     iohandler_init();
 647     aio_set_fd_handler(iohandler_ctx, fd, fd_read, fd_write, NULL, NULL,
 648                        opaque);
 649 }
 650
 651 void event_notifier_set_handler(EventNotifier *e,
 652                                 EventNotifierHandler *handler)
 653 {
 654     iohandler_init();
 655     aio_set_event_notifier(iohandler_ctx, e, handler, NULL, NULL);
 656 }