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-common.h"
  26 #include "qemu/timer.h"
  27 #include "slirp/slirp.h"
  28 #include "qemu/main-loop.h"
  29 #include "block/aio.h"
  30
  31 #ifndef _WIN32
  32
  33 #include "qemu/compatfd.h"
  34
  35 /* If we have signalfd, we mask out the signals we want to handle and then
  36  * use signalfd to listen for them.  We rely on whatever the current signal
  37  * handler is to dispatch the signals when we receive them.
  38  */
  39 static void sigfd_handler(void *opaque)
  40 {
  41     int fd = (intptr_t)opaque;
  42     struct qemu_signalfd_siginfo info;
  43     struct sigaction action;
  44     ssize_t len;
  45
  46     while (1) {
  47         do {
  48             len = read(fd, &info, sizeof(info));
  49         } while (len == -1 && errno == EINTR);
  50
  51         if (len == -1 && errno == EAGAIN) {
  52             break;
  53         }
  54
  55         if (len != sizeof(info)) {
  56             printf("read from sigfd returned %zd: %m\n", len);
  57             return;
  58         }
  59
  60         sigaction(info.ssi_signo, NULL, &action);
  61         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  62             action.sa_sigaction(info.ssi_signo,
  63                                 (siginfo_t *)&info, NULL);
  64         } else if (action.sa_handler) {
  65             action.sa_handler(info.ssi_signo);
  66         }
  67     }
  68 }
  69
  70 static int qemu_signal_init(void)
  71 {
  72     int sigfd;
  73     sigset_t set;
  74
  75     /*
  76      * SIG_IPI must be blocked in the main thread and must not be caught
  77      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  78      * not catch it reliably.
  79      */
  80     sigemptyset(&set);
  81     sigaddset(&set, SIG_IPI);
  82     sigaddset(&set, SIGIO);
  83     sigaddset(&set, SIGALRM);
  84     sigaddset(&set, SIGBUS);
  85     pthread_sigmask(SIG_BLOCK, &set, NULL);
  86
  87     sigdelset(&set, SIG_IPI);
  88     sigfd = qemu_signalfd(&set);
  89     if (sigfd == -1) {
  90         fprintf(stderr, "failed to create signalfd\n");
  91         return -errno;
  92     }
  93
  94     fcntl_setfl(sigfd, O_NONBLOCK);
  95
  96     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
  97                          (void *)(intptr_t)sigfd);
  98
  99     return 0;
 100 }
 101
 102 #else /* _WIN32 */
 103
 104 static int qemu_signal_init(void)
 105 {
 106     return 0;
 107 }
 108 #endif
 109
 110 static AioContext *qemu_aio_context;
 111
 112 AioContext *qemu_get_aio_context(void)
 113 {
 114     return qemu_aio_context;
 115 }
 116
 117 void qemu_notify_event(void)
 118 {
 119     if (!qemu_aio_context) {
 120         return;
 121     }
 122     aio_notify(qemu_aio_context);
 123 }
 124
 125 static GArray *gpollfds;
 126
 127 int qemu_init_main_loop(void)
 128 {
 129     int ret;
 130     GSource *src;
 131
 132     init_clocks();
 133     if (init_timer_alarm() < 0) {
 134         fprintf(stderr, "could not initialize alarm timer\n");
 135         exit(1);
 136     }
 137
 138     ret = qemu_signal_init();
 139     if (ret) {
 140         return ret;
 141     }
 142
 143     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 144     qemu_aio_context = aio_context_new();
 145     src = aio_get_g_source(qemu_aio_context);
 146     g_source_attach(src, NULL);
 147     g_source_unref(src);
 148     return 0;
 149 }
 150
 151 static int max_priority;
 152
 153 #ifndef _WIN32
 154 static int glib_pollfds_idx;
 155 static int glib_n_poll_fds;
 156
 157 static void glib_pollfds_fill(uint32_t *cur_timeout)
 158 {
 159     GMainContext *context = g_main_context_default();
 160     int timeout = 0;
 161     int n;
 162
 163     g_main_context_prepare(context, &max_priority);
 164
 165     glib_pollfds_idx = gpollfds->len;
 166     n = glib_n_poll_fds;
 167     do {
 168         GPollFD *pfds;
 169         glib_n_poll_fds = n;
 170         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 171         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 172         n = g_main_context_query(context, max_priority, &timeout, pfds,
 173                                  glib_n_poll_fds);
 174     } while (n != glib_n_poll_fds);
 175
 176     if (timeout >= 0 && timeout < *cur_timeout) {
 177         *cur_timeout = timeout;
 178     }
 179 }
 180
 181 static void glib_pollfds_poll(void)
 182 {
 183     GMainContext *context = g_main_context_default();
 184     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 185
 186     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 187         g_main_context_dispatch(context);
 188     }
 189 }
 190
 191 #define MAX_MAIN_LOOP_SPIN (1000)
 192
 193 static int os_host_main_loop_wait(uint32_t timeout)
 194 {
 195     int ret;
 196     static int spin_counter;
 197
 198     glib_pollfds_fill(&timeout);
 199
 200     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 201      * the I/O thread, this can lead to starvation of the BQL such that the
 202      * VCPU threads never run.  To make sure we can detect the later case,
 203      * print a message to the screen.  If we run into this condition, create
 204      * a fake timeout in order to give the VCPU threads a chance to run.
 205      */
 206     if (spin_counter > MAX_MAIN_LOOP_SPIN) {
 207         static bool notified;
 208
 209         if (!notified) {
 210             fprintf(stderr,
 211                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 212                     MAX_MAIN_LOOP_SPIN);
 213             notified = true;
 214         }
 215
 216         timeout = 1;
 217     }
 218
 219     if (timeout > 0) {
 220         spin_counter = 0;
 221         qemu_mutex_unlock_iothread();
 222     } else {
 223         spin_counter++;
 224     }
 225
 226     ret = g_poll((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 227
 228     if (timeout > 0) {
 229         qemu_mutex_lock_iothread();
 230     }
 231
 232     glib_pollfds_poll();
 233     return ret;
 234 }
 235 #else
 236 /***********************************************************/
 237 /* Polling handling */
 238
 239 typedef struct PollingEntry {
 240     PollingFunc *func;
 241     void *opaque;
 242     struct PollingEntry *next;
 243 } PollingEntry;
 244
 245 static PollingEntry *first_polling_entry;
 246
 247 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 248 {
 249     PollingEntry **ppe, *pe;
 250     pe = g_malloc0(sizeof(PollingEntry));
 251     pe->func = func;
 252     pe->opaque = opaque;
 253     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 254     *ppe = pe;
 255     return 0;
 256 }
 257
 258 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 259 {
 260     PollingEntry **ppe, *pe;
 261     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 262         pe = *ppe;
 263         if (pe->func == func && pe->opaque == opaque) {
 264             *ppe = pe->next;
 265             g_free(pe);
 266             break;
 267         }
 268     }
 269 }
 270
 271 /***********************************************************/
 272 /* Wait objects support */
 273 typedef struct WaitObjects {
 274     int num;
 275     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 276     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 277     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 278     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 279 } WaitObjects;
 280
 281 static WaitObjects wait_objects = {0};
 282
 283 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 284 {
 285     WaitObjects *w = &wait_objects;
 286     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 287         return -1;
 288     }
 289     w->events[w->num] = handle;
 290     w->func[w->num] = func;
 291     w->opaque[w->num] = opaque;
 292     w->revents[w->num] = 0;
 293     w->num++;
 294     return 0;
 295 }
 296
 297 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 298 {
 299     int i, found;
 300     WaitObjects *w = &wait_objects;
 301
 302     found = 0;
 303     for (i = 0; i < w->num; i++) {
 304         if (w->events[i] == handle) {
 305             found = 1;
 306         }
 307         if (found) {
 308             w->events[i] = w->events[i + 1];
 309             w->func[i] = w->func[i + 1];
 310             w->opaque[i] = w->opaque[i + 1];
 311             w->revents[i] = w->revents[i + 1];
 312         }
 313     }
 314     if (found) {
 315         w->num--;
 316     }
 317 }
 318
 319 void qemu_fd_register(int fd)
 320 {
 321     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 322                    FD_READ | FD_ACCEPT | FD_CLOSE |
 323                    FD_CONNECT | FD_WRITE | FD_OOB);
 324 }
 325
 326 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 327                         fd_set *xfds)
 328 {
 329     int nfds = -1;
 330     int i;
 331
 332     for (i = 0; i < pollfds->len; i++) {
 333         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 334         int fd = pfd->fd;
 335         int events = pfd->events;
 336         if (events & (G_IO_IN | G_IO_HUP | G_IO_ERR)) {
 337             FD_SET(fd, rfds);
 338             nfds = MAX(nfds, fd);
 339         }
 340         if (events & (G_IO_OUT | G_IO_ERR)) {
 341             FD_SET(fd, wfds);
 342             nfds = MAX(nfds, fd);
 343         }
 344         if (events & G_IO_PRI) {
 345             FD_SET(fd, xfds);
 346             nfds = MAX(nfds, fd);
 347         }
 348     }
 349     return nfds;
 350 }
 351
 352 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 353                          fd_set *wfds, fd_set *xfds)
 354 {
 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 revents = 0;
 361
 362         if (FD_ISSET(fd, rfds)) {
 363             revents |= G_IO_IN | G_IO_HUP | G_IO_ERR;
 364         }
 365         if (FD_ISSET(fd, wfds)) {
 366             revents |= G_IO_OUT | G_IO_ERR;
 367         }
 368         if (FD_ISSET(fd, xfds)) {
 369             revents |= G_IO_PRI;
 370         }
 371         pfd->revents = revents & pfd->events;
 372     }
 373 }
 374
 375 static int os_host_main_loop_wait(uint32_t timeout)
 376 {
 377     GMainContext *context = g_main_context_default();
 378     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 379     int select_ret = 0;
 380     int g_poll_ret, ret, i, n_poll_fds;
 381     PollingEntry *pe;
 382     WaitObjects *w = &wait_objects;
 383     gint poll_timeout;
 384     static struct timeval tv0;
 385     fd_set rfds, wfds, xfds;
 386     int nfds;
 387
 388     /* XXX: need to suppress polling by better using win32 events */
 389     ret = 0;
 390     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 391         ret |= pe->func(pe->opaque);
 392     }
 393     if (ret != 0) {
 394         return ret;
 395     }
 396
 397     g_main_context_prepare(context, &max_priority);
 398     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 399                                       poll_fds, ARRAY_SIZE(poll_fds));
 400     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 401
 402     for (i = 0; i < w->num; i++) {
 403         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 404         poll_fds[n_poll_fds + i].events = G_IO_IN;
 405     }
 406
 407     if (poll_timeout < 0 || timeout < poll_timeout) {
 408         poll_timeout = timeout;
 409     }
 410
 411     qemu_mutex_unlock_iothread();
 412     g_poll_ret = g_poll(poll_fds, n_poll_fds + w->num, poll_timeout);
 413     qemu_mutex_lock_iothread();
 414     if (g_poll_ret > 0) {
 415         for (i = 0; i < w->num; i++) {
 416             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 417         }
 418         for (i = 0; i < w->num; i++) {
 419             if (w->revents[i] && w->func[i]) {
 420                 w->func[i](w->opaque[i]);
 421             }
 422         }
 423     }
 424
 425     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 426         g_main_context_dispatch(context);
 427     }
 428
 429     /* Call select after g_poll to avoid a useless iteration and therefore
 430      * improve socket latency.
 431      */
 432
 433     FD_ZERO(&rfds);
 434     FD_ZERO(&wfds);
 435     FD_ZERO(&xfds);
 436     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 437     if (nfds >= 0) {
 438         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 439         if (select_ret != 0) {
 440             timeout = 0;
 441         }
 442         if (select_ret > 0) {
 443             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 444         }
 445     }
 446
 447     return select_ret || g_poll_ret;
 448 }
 449 #endif
 450
 451 int main_loop_wait(int nonblocking)
 452 {
 453     int ret;
 454     uint32_t timeout = UINT32_MAX;
 455
 456     if (nonblocking) {
 457         timeout = 0;
 458     }
 459
 460     /* poll any events */
 461     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 462     /* XXX: separate device handlers from system ones */
 463 #ifdef CONFIG_SLIRP
 464     slirp_update_timeout(&timeout);
 465     slirp_pollfds_fill(gpollfds);
 466 #endif
 467     qemu_iohandler_fill(gpollfds);
 468     ret = os_host_main_loop_wait(timeout);
 469     qemu_iohandler_poll(gpollfds, ret);
 470 #ifdef CONFIG_SLIRP
 471     slirp_pollfds_poll(gpollfds, (ret < 0));
 472 #endif
 473
 474     qemu_run_all_timers();
 475
 476     return ret;
 477 }
 478
 479 /* Functions to operate on the main QEMU AioContext.  */
 480
 481 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 482 {
 483     return aio_bh_new(qemu_aio_context, cb, opaque);
 484 }
 485
 486 bool qemu_aio_wait(void)
 487 {
 488     return aio_poll(qemu_aio_context, true);
 489 }
 490
 491 #ifdef CONFIG_POSIX
 492 void qemu_aio_set_fd_handler(int fd,
 493                              IOHandler *io_read,
 494                              IOHandler *io_write,
 495                              AioFlushHandler *io_flush,
 496                              void *opaque)
 497 {
 498     aio_set_fd_handler(qemu_aio_context, fd, io_read, io_write, io_flush,
 499                        opaque);
 500 }
 501 #endif
 502
 503 void qemu_aio_set_event_notifier(EventNotifier *notifier,
 504                                  EventNotifierHandler *io_read,
 505                                  AioFlushEventNotifierHandler *io_flush)
 506 {
 507     aio_set_event_notifier(qemu_aio_context, notifier, io_read, io_flush);
 508 }