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 "qemu/error-report.h"
  34 #include "qemu/queue.h"
  35 #include "qemu/compiler.h"
  36
  37 #ifndef _WIN32
  38 #include <sys/wait.h>
  39 #endif
  40
  41 #ifndef _WIN32
  42
  43 /* If we have signalfd, we mask out the signals we want to handle and then
  44  * use signalfd to listen for them.  We rely on whatever the current signal
  45  * handler is to dispatch the signals when we receive them.
  46  */
  47 /*
  48  * Disable CFI checks.
  49  * We are going to call a signal hander directly. Such handler may or may not
  50  * have been defined in our binary, so there's no guarantee that the pointer
  51  * used to set the handler is a cfi-valid pointer. Since the handlers are
  52  * stored in kernel memory, changing the handler to an attacker-defined
  53  * function requires being able to call a sigaction() syscall,
  54  * which is not as easy as overwriting a pointer in memory.
  55  */
  56 QEMU_DISABLE_CFI
  57 static void sigfd_handler(void *opaque)
  58 {
  59     int fd = (intptr_t)opaque;
  60     struct qemu_signalfd_siginfo info;
  61     struct sigaction action;
  62     ssize_t len;
  63
  64     while (1) {
  65         do {
  66             len = read(fd, &info, sizeof(info));
  67         } while (len == -1 && errno == EINTR);
  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     fcntl_setfl(sigfd, O_NONBLOCK);
 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_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 int max_priority;
 187
 188 #ifndef _WIN32
 189 static int glib_pollfds_idx;
 190 static int glib_n_poll_fds;
 191
 192 void qemu_fd_register(int fd)
 193 {
 194 }
 195
 196 static void glib_pollfds_fill(int64_t *cur_timeout)
 197 {
 198     GMainContext *context = g_main_context_default();
 199     int timeout = 0;
 200     int64_t timeout_ns;
 201     int n;
 202
 203     g_main_context_prepare(context, &max_priority);
 204
 205     glib_pollfds_idx = gpollfds->len;
 206     n = glib_n_poll_fds;
 207     do {
 208         GPollFD *pfds;
 209         glib_n_poll_fds = n;
 210         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 211         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 212         n = g_main_context_query(context, max_priority, &timeout, pfds,
 213                                  glib_n_poll_fds);
 214     } while (n != glib_n_poll_fds);
 215
 216     if (timeout < 0) {
 217         timeout_ns = -1;
 218     } else {
 219         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 220     }
 221
 222     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 223 }
 224
 225 static void glib_pollfds_poll(void)
 226 {
 227     GMainContext *context = g_main_context_default();
 228     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 229
 230     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 231         g_main_context_dispatch(context);
 232     }
 233 }
 234
 235 #define MAX_MAIN_LOOP_SPIN (1000)
 236
 237 static int os_host_main_loop_wait(int64_t timeout)
 238 {
 239     GMainContext *context = g_main_context_default();
 240     int ret;
 241
 242     g_main_context_acquire(context);
 243
 244     glib_pollfds_fill(&timeout);
 245
 246     qemu_mutex_unlock_iothread();
 247     replay_mutex_unlock();
 248
 249     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 250
 251     replay_mutex_lock();
 252     qemu_mutex_lock_iothread();
 253
 254     glib_pollfds_poll();
 255
 256     g_main_context_release(context);
 257
 258     return ret;
 259 }
 260 #else
 261 /***********************************************************/
 262 /* Polling handling */
 263
 264 typedef struct PollingEntry {
 265     PollingFunc *func;
 266     void *opaque;
 267     struct PollingEntry *next;
 268 } PollingEntry;
 269
 270 static PollingEntry *first_polling_entry;
 271
 272 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 273 {
 274     PollingEntry **ppe, *pe;
 275     pe = g_malloc0(sizeof(PollingEntry));
 276     pe->func = func;
 277     pe->opaque = opaque;
 278     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 279     *ppe = pe;
 280     return 0;
 281 }
 282
 283 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 284 {
 285     PollingEntry **ppe, *pe;
 286     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 287         pe = *ppe;
 288         if (pe->func == func && pe->opaque == opaque) {
 289             *ppe = pe->next;
 290             g_free(pe);
 291             break;
 292         }
 293     }
 294 }
 295
 296 /***********************************************************/
 297 /* Wait objects support */
 298 typedef struct WaitObjects {
 299     int num;
 300     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 301     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 302     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 303     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 304 } WaitObjects;
 305
 306 static WaitObjects wait_objects = {0};
 307
 308 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 309 {
 310     WaitObjects *w = &wait_objects;
 311     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 312         return -1;
 313     }
 314     w->events[w->num] = handle;
 315     w->func[w->num] = func;
 316     w->opaque[w->num] = opaque;
 317     w->revents[w->num] = 0;
 318     w->num++;
 319     return 0;
 320 }
 321
 322 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 323 {
 324     int i, found;
 325     WaitObjects *w = &wait_objects;
 326
 327     found = 0;
 328     for (i = 0; i < w->num; i++) {
 329         if (w->events[i] == handle) {
 330             found = 1;
 331         }
 332         if (found) {
 333             w->events[i] = w->events[i + 1];
 334             w->func[i] = w->func[i + 1];
 335             w->opaque[i] = w->opaque[i + 1];
 336             w->revents[i] = w->revents[i + 1];
 337         }
 338     }
 339     if (found) {
 340         w->num--;
 341     }
 342 }
 343
 344 void qemu_fd_register(int fd)
 345 {
 346     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 347                    FD_READ | FD_ACCEPT | FD_CLOSE |
 348                    FD_CONNECT | FD_WRITE | FD_OOB);
 349 }
 350
 351 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 352                         fd_set *xfds)
 353 {
 354     int nfds = -1;
 355     int i;
 356
 357     for (i = 0; i < pollfds->len; i++) {
 358         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 359         int fd = pfd->fd;
 360         int events = pfd->events;
 361         if (events & G_IO_IN) {
 362             FD_SET(fd, rfds);
 363             nfds = MAX(nfds, fd);
 364         }
 365         if (events & G_IO_OUT) {
 366             FD_SET(fd, wfds);
 367             nfds = MAX(nfds, fd);
 368         }
 369         if (events & G_IO_PRI) {
 370             FD_SET(fd, xfds);
 371             nfds = MAX(nfds, fd);
 372         }
 373     }
 374     return nfds;
 375 }
 376
 377 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 378                          fd_set *wfds, fd_set *xfds)
 379 {
 380     int i;
 381
 382     for (i = 0; i < pollfds->len; i++) {
 383         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 384         int fd = pfd->fd;
 385         int revents = 0;
 386
 387         if (FD_ISSET(fd, rfds)) {
 388             revents |= G_IO_IN;
 389         }
 390         if (FD_ISSET(fd, wfds)) {
 391             revents |= G_IO_OUT;
 392         }
 393         if (FD_ISSET(fd, xfds)) {
 394             revents |= G_IO_PRI;
 395         }
 396         pfd->revents = revents & pfd->events;
 397     }
 398 }
 399
 400 static int os_host_main_loop_wait(int64_t timeout)
 401 {
 402     GMainContext *context = g_main_context_default();
 403     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 404     int select_ret = 0;
 405     int g_poll_ret, ret, i, n_poll_fds;
 406     PollingEntry *pe;
 407     WaitObjects *w = &wait_objects;
 408     gint poll_timeout;
 409     int64_t poll_timeout_ns;
 410     static struct timeval tv0;
 411     fd_set rfds, wfds, xfds;
 412     int nfds;
 413
 414     g_main_context_acquire(context);
 415
 416     /* XXX: need to suppress polling by better using win32 events */
 417     ret = 0;
 418     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 419         ret |= pe->func(pe->opaque);
 420     }
 421     if (ret != 0) {
 422         g_main_context_release(context);
 423         return ret;
 424     }
 425
 426     FD_ZERO(&rfds);
 427     FD_ZERO(&wfds);
 428     FD_ZERO(&xfds);
 429     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 430     if (nfds >= 0) {
 431         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 432         if (select_ret != 0) {
 433             timeout = 0;
 434         }
 435         if (select_ret > 0) {
 436             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 437         }
 438     }
 439
 440     g_main_context_prepare(context, &max_priority);
 441     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 442                                       poll_fds, ARRAY_SIZE(poll_fds));
 443     g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
 444
 445     for (i = 0; i < w->num; i++) {
 446         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 447         poll_fds[n_poll_fds + i].events = G_IO_IN;
 448     }
 449
 450     if (poll_timeout < 0) {
 451         poll_timeout_ns = -1;
 452     } else {
 453         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 454     }
 455
 456     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 457
 458     qemu_mutex_unlock_iothread();
 459
 460     replay_mutex_unlock();
 461
 462     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 463
 464     replay_mutex_lock();
 465
 466     qemu_mutex_lock_iothread();
 467     if (g_poll_ret > 0) {
 468         for (i = 0; i < w->num; i++) {
 469             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 470         }
 471         for (i = 0; i < w->num; i++) {
 472             if (w->revents[i] && w->func[i]) {
 473                 w->func[i](w->opaque[i]);
 474             }
 475         }
 476     }
 477
 478     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 479         g_main_context_dispatch(context);
 480     }
 481
 482     g_main_context_release(context);
 483
 484     return select_ret || g_poll_ret;
 485 }
 486 #endif
 487
 488 static NotifierList main_loop_poll_notifiers =
 489     NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
 490
 491 void main_loop_poll_add_notifier(Notifier *notify)
 492 {
 493     notifier_list_add(&main_loop_poll_notifiers, notify);
 494 }
 495
 496 void main_loop_poll_remove_notifier(Notifier *notify)
 497 {
 498     notifier_remove(notify);
 499 }
 500
 501 void main_loop_wait(int nonblocking)
 502 {
 503     MainLoopPoll mlpoll = {
 504         .state = MAIN_LOOP_POLL_FILL,
 505         .timeout = UINT32_MAX,
 506         .pollfds = gpollfds,
 507     };
 508     int ret;
 509     int64_t timeout_ns;
 510
 511     if (nonblocking) {
 512         mlpoll.timeout = 0;
 513     }
 514
 515     /* poll any events */
 516     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 517     /* XXX: separate device handlers from system ones */
 518     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 519
 520     if (mlpoll.timeout == UINT32_MAX) {
 521         timeout_ns = -1;
 522     } else {
 523         timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
 524     }
 525
 526     timeout_ns = qemu_soonest_timeout(timeout_ns,
 527                                       timerlistgroup_deadline_ns(
 528                                           &main_loop_tlg));
 529
 530     ret = os_host_main_loop_wait(timeout_ns);
 531     mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
 532     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 533
 534     if (icount_enabled()) {
 535         /*
 536          * CPU thread can infinitely wait for event after
 537          * missing the warp
 538          */
 539         icount_start_warp_timer();
 540     }
 541     qemu_clock_run_all_timers();
 542 }
 543
 544 /* Functions to operate on the main QEMU AioContext.  */
 545
 546 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 547 {
 548     return aio_bh_new(qemu_aio_context, cb, opaque);
 549 }
 550
 551 /*
 552  * Functions to operate on the I/O handler AioContext.
 553  * This context runs on top of main loop. We can't reuse qemu_aio_context
 554  * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
 555  */
 556 static AioContext *iohandler_ctx;
 557
 558 static void iohandler_init(void)
 559 {
 560     if (!iohandler_ctx) {
 561         iohandler_ctx = aio_context_new(&error_abort);
 562     }
 563 }
 564
 565 AioContext *iohandler_get_aio_context(void)
 566 {
 567     iohandler_init();
 568     return iohandler_ctx;
 569 }
 570
 571 GSource *iohandler_get_g_source(void)
 572 {
 573     iohandler_init();
 574     return aio_get_g_source(iohandler_ctx);
 575 }
 576
 577 void qemu_set_fd_handler(int fd,
 578                          IOHandler *fd_read,
 579                          IOHandler *fd_write,
 580                          void *opaque)
 581 {
 582     iohandler_init();
 583     aio_set_fd_handler(iohandler_ctx, fd, false,
 584                        fd_read, fd_write, NULL, opaque);
 585 }
 586
 587 void event_notifier_set_handler(EventNotifier *e,
 588                                 EventNotifierHandler *handler)
 589 {
 590     iohandler_init();
 591     aio_set_event_notifier(iohandler_ctx, e, false,
 592                            handler, NULL);
 593 }
 594
 595 /* reaping of zombies.  right now we're not passing the status to
 596    anyone, but it would be possible to add a callback.  */
 597 #ifndef _WIN32
 598 typedef struct ChildProcessRecord {
 599     int pid;
 600     QLIST_ENTRY(ChildProcessRecord) next;
 601 } ChildProcessRecord;
 602
 603 static QLIST_HEAD(, ChildProcessRecord) child_watches =
 604     QLIST_HEAD_INITIALIZER(child_watches);
 605
 606 static QEMUBH *sigchld_bh;
 607
 608 static void sigchld_handler(int signal)
 609 {
 610     qemu_bh_schedule(sigchld_bh);
 611 }
 612
 613 static void sigchld_bh_handler(void *opaque)
 614 {
 615     ChildProcessRecord *rec, *next;
 616
 617     QLIST_FOREACH_SAFE(rec, &child_watches, next, next) {
 618         if (waitpid(rec->pid, NULL, WNOHANG) == rec->pid) {
 619             QLIST_REMOVE(rec, next);
 620             g_free(rec);
 621         }
 622     }
 623 }
 624
 625 static void qemu_init_child_watch(void)
 626 {
 627     struct sigaction act;
 628     sigchld_bh = qemu_bh_new(sigchld_bh_handler, NULL);
 629
 630     memset(&act, 0, sizeof(act));
 631     act.sa_handler = sigchld_handler;
 632     act.sa_flags = SA_NOCLDSTOP;
 633     sigaction(SIGCHLD, &act, NULL);
 634 }
 635
 636 int qemu_add_child_watch(pid_t pid)
 637 {
 638     ChildProcessRecord *rec;
 639
 640     if (!sigchld_bh) {
 641         qemu_init_child_watch();
 642     }
 643
 644     QLIST_FOREACH(rec, &child_watches, next) {
 645         if (rec->pid == pid) {
 646             return 1;
 647         }
 648     }
 649     rec = g_malloc0(sizeof(ChildProcessRecord));
 650     rec->pid = pid;
 651     QLIST_INSERT_HEAD(&child_watches, rec, next);
 652     return 0;
 653 }
 654 #endif