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 "main-loop.h"
  29 #include "qemu-aio.h"
  30
  31 #ifndef _WIN32
  32
  33 #include "compatfd.h"
  34
  35 static int io_thread_fd = -1;
  36
  37 void qemu_notify_event(void)
  38 {
  39     /* Write 8 bytes to be compatible with eventfd.  */
  40     static const uint64_t val = 1;
  41     ssize_t ret;
  42
  43     if (io_thread_fd == -1) {
  44         return;
  45     }
  46     do {
  47         ret = write(io_thread_fd, &val, sizeof(val));
  48     } while (ret < 0 && errno == EINTR);
  49
  50     /* EAGAIN is fine, a read must be pending.  */
  51     if (ret < 0 && errno != EAGAIN) {
  52         fprintf(stderr, "qemu_notify_event: write() failed: %s\n",
  53                 strerror(errno));
  54         exit(1);
  55     }
  56 }
  57
  58 static void qemu_event_read(void *opaque)
  59 {
  60     int fd = (intptr_t)opaque;
  61     ssize_t len;
  62     char buffer[512];
  63
  64     /* Drain the notify pipe.  For eventfd, only 8 bytes will be read.  */
  65     do {
  66         len = read(fd, buffer, sizeof(buffer));
  67     } while ((len == -1 && errno == EINTR) || len == sizeof(buffer));
  68 }
  69
  70 static int qemu_event_init(void)
  71 {
  72     int err;
  73     int fds[2];
  74
  75     err = qemu_eventfd(fds);
  76     if (err == -1) {
  77         return -errno;
  78     }
  79     err = fcntl_setfl(fds[0], O_NONBLOCK);
  80     if (err < 0) {
  81         goto fail;
  82     }
  83     err = fcntl_setfl(fds[1], O_NONBLOCK);
  84     if (err < 0) {
  85         goto fail;
  86     }
  87     qemu_set_fd_handler2(fds[0], NULL, qemu_event_read, NULL,
  88                          (void *)(intptr_t)fds[0]);
  89
  90     io_thread_fd = fds[1];
  91     return 0;
  92
  93 fail:
  94     close(fds[0]);
  95     close(fds[1]);
  96     return err;
  97 }
  98
  99 /* If we have signalfd, we mask out the signals we want to handle and then
 100  * use signalfd to listen for them.  We rely on whatever the current signal
 101  * handler is to dispatch the signals when we receive them.
 102  */
 103 static void sigfd_handler(void *opaque)
 104 {
 105     int fd = (intptr_t)opaque;
 106     struct qemu_signalfd_siginfo info;
 107     struct sigaction action;
 108     ssize_t len;
 109
 110     while (1) {
 111         do {
 112             len = read(fd, &info, sizeof(info));
 113         } while (len == -1 && errno == EINTR);
 114
 115         if (len == -1 && errno == EAGAIN) {
 116             break;
 117         }
 118
 119         if (len != sizeof(info)) {
 120             printf("read from sigfd returned %zd: %m\n", len);
 121             return;
 122         }
 123
 124         sigaction(info.ssi_signo, NULL, &action);
 125         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
 126             action.sa_sigaction(info.ssi_signo,
 127                                 (siginfo_t *)&info, NULL);
 128         } else if (action.sa_handler) {
 129             action.sa_handler(info.ssi_signo);
 130         }
 131     }
 132 }
 133
 134 static int qemu_signal_init(void)
 135 {
 136     int sigfd;
 137     sigset_t set;
 138
 139     /*
 140      * SIG_IPI must be blocked in the main thread and must not be caught
 141      * by sigwait() in the signal thread. Otherwise, the cpu thread will
 142      * not catch it reliably.
 143      */
 144     sigemptyset(&set);
 145     sigaddset(&set, SIG_IPI);
 146     sigaddset(&set, SIGIO);
 147     sigaddset(&set, SIGALRM);
 148     sigaddset(&set, SIGBUS);
 149     pthread_sigmask(SIG_BLOCK, &set, NULL);
 150
 151     sigdelset(&set, SIG_IPI);
 152     sigfd = qemu_signalfd(&set);
 153     if (sigfd == -1) {
 154         fprintf(stderr, "failed to create signalfd\n");
 155         return -errno;
 156     }
 157
 158     fcntl_setfl(sigfd, O_NONBLOCK);
 159
 160     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
 161                          (void *)(intptr_t)sigfd);
 162
 163     return 0;
 164 }
 165
 166 #else /* _WIN32 */
 167
 168 static HANDLE qemu_event_handle = NULL;
 169
 170 static void dummy_event_handler(void *opaque)
 171 {
 172 }
 173
 174 static int qemu_event_init(void)
 175 {
 176     qemu_event_handle = CreateEvent(NULL, FALSE, FALSE, NULL);
 177     if (!qemu_event_handle) {
 178         fprintf(stderr, "Failed CreateEvent: %ld\n", GetLastError());
 179         return -1;
 180     }
 181     qemu_add_wait_object(qemu_event_handle, dummy_event_handler, NULL);
 182     return 0;
 183 }
 184
 185 void qemu_notify_event(void)
 186 {
 187     if (!qemu_event_handle) {
 188         return;
 189     }
 190     if (!SetEvent(qemu_event_handle)) {
 191         fprintf(stderr, "qemu_notify_event: SetEvent failed: %ld\n",
 192                 GetLastError());
 193         exit(1);
 194     }
 195 }
 196
 197 static int qemu_signal_init(void)
 198 {
 199     return 0;
 200 }
 201 #endif
 202
 203 static AioContext *qemu_aio_context;
 204
 205 int qemu_init_main_loop(void)
 206 {
 207     int ret;
 208     GSource *src;
 209
 210     init_clocks();
 211     init_timer_alarm();
 212
 213     qemu_mutex_lock_iothread();
 214     ret = qemu_signal_init();
 215     if (ret) {
 216         return ret;
 217     }
 218
 219     /* Note eventfd must be drained before signalfd handlers run */
 220     ret = qemu_event_init();
 221     if (ret) {
 222         return ret;
 223     }
 224
 225     qemu_aio_context = aio_context_new();
 226     src = aio_get_g_source(qemu_aio_context);
 227     g_source_attach(src, NULL);
 228     g_source_unref(src);
 229     return 0;
 230 }
 231
 232 static fd_set rfds, wfds, xfds;
 233 static int nfds;
 234 static GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 235 static int n_poll_fds;
 236 static int max_priority;
 237
 238 #ifndef _WIN32
 239 static void glib_select_fill(int *max_fd, fd_set *rfds, fd_set *wfds,
 240                              fd_set *xfds, uint32_t *cur_timeout)
 241 {
 242     GMainContext *context = g_main_context_default();
 243     int i;
 244     int timeout = 0;
 245
 246     g_main_context_prepare(context, &max_priority);
 247
 248     n_poll_fds = g_main_context_query(context, max_priority, &timeout,
 249                                       poll_fds, ARRAY_SIZE(poll_fds));
 250     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 251
 252     for (i = 0; i < n_poll_fds; i++) {
 253         GPollFD *p = &poll_fds[i];
 254
 255         if ((p->events & G_IO_IN)) {
 256             FD_SET(p->fd, rfds);
 257             *max_fd = MAX(*max_fd, p->fd);
 258         }
 259         if ((p->events & G_IO_OUT)) {
 260             FD_SET(p->fd, wfds);
 261             *max_fd = MAX(*max_fd, p->fd);
 262         }
 263         if ((p->events & G_IO_ERR)) {
 264             FD_SET(p->fd, xfds);
 265             *max_fd = MAX(*max_fd, p->fd);
 266         }
 267     }
 268
 269     if (timeout >= 0 && timeout < *cur_timeout) {
 270         *cur_timeout = timeout;
 271     }
 272 }
 273
 274 static void glib_select_poll(fd_set *rfds, fd_set *wfds, fd_set *xfds,
 275                              bool err)
 276 {
 277     GMainContext *context = g_main_context_default();
 278
 279     if (!err) {
 280         int i;
 281
 282         for (i = 0; i < n_poll_fds; i++) {
 283             GPollFD *p = &poll_fds[i];
 284
 285             if ((p->events & G_IO_IN) && FD_ISSET(p->fd, rfds)) {
 286                 p->revents |= G_IO_IN;
 287             }
 288             if ((p->events & G_IO_OUT) && FD_ISSET(p->fd, wfds)) {
 289                 p->revents |= G_IO_OUT;
 290             }
 291             if ((p->events & G_IO_ERR) && FD_ISSET(p->fd, xfds)) {
 292                 p->revents |= G_IO_ERR;
 293             }
 294         }
 295     }
 296
 297     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 298         g_main_context_dispatch(context);
 299     }
 300 }
 301
 302 static int os_host_main_loop_wait(uint32_t timeout)
 303 {
 304     struct timeval tv, *tvarg = NULL;
 305     int ret;
 306
 307     glib_select_fill(&nfds, &rfds, &wfds, &xfds, &timeout);
 308
 309     if (timeout < UINT32_MAX) {
 310         tvarg = &tv;
 311         tv.tv_sec = timeout / 1000;
 312         tv.tv_usec = (timeout % 1000) * 1000;
 313     }
 314
 315     if (timeout > 0) {
 316         qemu_mutex_unlock_iothread();
 317     }
 318
 319     ret = select(nfds + 1, &rfds, &wfds, &xfds, tvarg);
 320
 321     if (timeout > 0) {
 322         qemu_mutex_lock_iothread();
 323     }
 324
 325     glib_select_poll(&rfds, &wfds, &xfds, (ret < 0));
 326     return ret;
 327 }
 328 #else
 329 /***********************************************************/
 330 /* Polling handling */
 331
 332 typedef struct PollingEntry {
 333     PollingFunc *func;
 334     void *opaque;
 335     struct PollingEntry *next;
 336 } PollingEntry;
 337
 338 static PollingEntry *first_polling_entry;
 339
 340 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 341 {
 342     PollingEntry **ppe, *pe;
 343     pe = g_malloc0(sizeof(PollingEntry));
 344     pe->func = func;
 345     pe->opaque = opaque;
 346     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 347     *ppe = pe;
 348     return 0;
 349 }
 350
 351 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 352 {
 353     PollingEntry **ppe, *pe;
 354     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 355         pe = *ppe;
 356         if (pe->func == func && pe->opaque == opaque) {
 357             *ppe = pe->next;
 358             g_free(pe);
 359             break;
 360         }
 361     }
 362 }
 363
 364 /***********************************************************/
 365 /* Wait objects support */
 366 typedef struct WaitObjects {
 367     int num;
 368     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 369     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 370     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 371     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 372 } WaitObjects;
 373
 374 static WaitObjects wait_objects = {0};
 375
 376 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 377 {
 378     WaitObjects *w = &wait_objects;
 379     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 380         return -1;
 381     }
 382     w->events[w->num] = handle;
 383     w->func[w->num] = func;
 384     w->opaque[w->num] = opaque;
 385     w->revents[w->num] = 0;
 386     w->num++;
 387     return 0;
 388 }
 389
 390 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 391 {
 392     int i, found;
 393     WaitObjects *w = &wait_objects;
 394
 395     found = 0;
 396     for (i = 0; i < w->num; i++) {
 397         if (w->events[i] == handle) {
 398             found = 1;
 399         }
 400         if (found) {
 401             w->events[i] = w->events[i + 1];
 402             w->func[i] = w->func[i + 1];
 403             w->opaque[i] = w->opaque[i + 1];
 404             w->revents[i] = w->revents[i + 1];
 405         }
 406     }
 407     if (found) {
 408         w->num--;
 409     }
 410 }
 411
 412 void qemu_fd_register(int fd)
 413 {
 414     WSAEventSelect(fd, qemu_event_handle, FD_READ | FD_ACCEPT | FD_CLOSE |
 415                    FD_CONNECT | FD_WRITE | FD_OOB);
 416 }
 417
 418 static int os_host_main_loop_wait(uint32_t timeout)
 419 {
 420     GMainContext *context = g_main_context_default();
 421     int ret, i;
 422     PollingEntry *pe;
 423     WaitObjects *w = &wait_objects;
 424     gint poll_timeout;
 425     static struct timeval tv0;
 426
 427     /* XXX: need to suppress polling by better using win32 events */
 428     ret = 0;
 429     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 430         ret |= pe->func(pe->opaque);
 431     }
 432     if (ret != 0) {
 433         return ret;
 434     }
 435
 436     if (nfds >= 0) {
 437         ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 438         if (ret != 0) {
 439             timeout = 0;
 440         }
 441     }
 442
 443     g_main_context_prepare(context, &max_priority);
 444     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 445                                       poll_fds, ARRAY_SIZE(poll_fds));
 446     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 447
 448     for (i = 0; i < w->num; i++) {
 449         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 450         poll_fds[n_poll_fds + i].events = G_IO_IN;
 451     }
 452
 453     if (poll_timeout < 0 || timeout < poll_timeout) {
 454         poll_timeout = timeout;
 455     }
 456
 457     qemu_mutex_unlock_iothread();
 458     ret = g_poll(poll_fds, n_poll_fds + w->num, poll_timeout);
 459     qemu_mutex_lock_iothread();
 460     if (ret > 0) {
 461         for (i = 0; i < w->num; i++) {
 462             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 463         }
 464         for (i = 0; i < w->num; i++) {
 465             if (w->revents[i] && w->func[i]) {
 466                 w->func[i](w->opaque[i]);
 467             }
 468         }
 469     }
 470
 471     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 472         g_main_context_dispatch(context);
 473     }
 474
 475     /* If an edge-triggered socket event occurred, select will return a
 476      * positive result on the next iteration.  We do not need to do anything
 477      * here.
 478      */
 479
 480     return ret;
 481 }
 482 #endif
 483
 484 int main_loop_wait(int nonblocking)
 485 {
 486     int ret;
 487     uint32_t timeout = UINT32_MAX;
 488
 489     if (nonblocking) {
 490         timeout = 0;
 491     }
 492
 493     /* poll any events */
 494     /* XXX: separate device handlers from system ones */
 495     nfds = -1;
 496     FD_ZERO(&rfds);
 497     FD_ZERO(&wfds);
 498     FD_ZERO(&xfds);
 499
 500 #ifdef CONFIG_SLIRP
 501     slirp_update_timeout(&timeout);
 502     slirp_select_fill(&nfds, &rfds, &wfds, &xfds);
 503 #endif
 504     qemu_iohandler_fill(&nfds, &rfds, &wfds, &xfds);
 505     ret = os_host_main_loop_wait(timeout);
 506     qemu_iohandler_poll(&rfds, &wfds, &xfds, ret);
 507 #ifdef CONFIG_SLIRP
 508     slirp_select_poll(&rfds, &wfds, &xfds, (ret < 0));
 509 #endif
 510
 511     qemu_run_all_timers();
 512
 513     return ret;
 514 }
 515
 516 /* Functions to operate on the main QEMU AioContext.  */
 517
 518 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 519 {
 520     return aio_bh_new(qemu_aio_context, cb, opaque);
 521 }
 522
 523 void qemu_aio_flush(void)
 524 {
 525     aio_flush(qemu_aio_context);
 526 }
 527
 528 bool qemu_aio_wait(void)
 529 {
 530     return aio_poll(qemu_aio_context, true);
 531 }
 532
 533 #ifdef CONFIG_POSIX
 534 void qemu_aio_set_fd_handler(int fd,
 535                              IOHandler *io_read,
 536                              IOHandler *io_write,
 537                              AioFlushHandler *io_flush,
 538                              void *opaque)
 539 {
 540     aio_set_fd_handler(qemu_aio_context, fd, io_read, io_write, io_flush,
 541                        opaque);
 542 }
 543 #endif
 544
 545 void qemu_aio_set_event_notifier(EventNotifier *notifier,
 546                                  EventNotifierHandler *io_read,
 547                                  AioFlushEventNotifierHandler *io_flush)
 548 {
 549     aio_set_event_notifier(qemu_aio_context, notifier, io_read, io_flush);
 550 }