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 void qemu_notify_event(void)
 113 {
 114     if (!qemu_aio_context) {
 115         return;
 116     }
 117     aio_notify(qemu_aio_context);
 118 }
 119
 120 static GArray *gpollfds;
 121
 122 int qemu_init_main_loop(void)
 123 {
 124     int ret;
 125     GSource *src;
 126
 127     init_clocks();
 128     if (init_timer_alarm() < 0) {
 129         fprintf(stderr, "could not initialize alarm timer\n");
 130         exit(1);
 131     }
 132
 133     ret = qemu_signal_init();
 134     if (ret) {
 135         return ret;
 136     }
 137
 138     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 139     qemu_aio_context = aio_context_new();
 140     src = aio_get_g_source(qemu_aio_context);
 141     g_source_attach(src, NULL);
 142     g_source_unref(src);
 143     return 0;
 144 }
 145
 146 static fd_set rfds, wfds, xfds;
 147 static int nfds;
 148 static int max_priority;
 149
 150 /* Load rfds/wfds/xfds into gpollfds.  Will be removed a few commits later. */
 151 static void gpollfds_from_select(void)
 152 {
 153     int fd;
 154     for (fd = 0; fd <= nfds; fd++) {
 155         int events = 0;
 156         if (FD_ISSET(fd, &rfds)) {
 157             events |= G_IO_IN | G_IO_HUP | G_IO_ERR;
 158         }
 159         if (FD_ISSET(fd, &wfds)) {
 160             events |= G_IO_OUT | G_IO_ERR;
 161         }
 162         if (FD_ISSET(fd, &xfds)) {
 163             events |= G_IO_PRI;
 164         }
 165         if (events) {
 166             GPollFD pfd = {
 167                 .fd = fd,
 168                 .events = events,
 169             };
 170             g_array_append_val(gpollfds, pfd);
 171         }
 172     }
 173 }
 174
 175 /* Store gpollfds revents into rfds/wfds/xfds.  Will be removed a few commits
 176  * later.
 177  */
 178 static void gpollfds_to_select(int ret)
 179 {
 180     int i;
 181
 182     FD_ZERO(&rfds);
 183     FD_ZERO(&wfds);
 184     FD_ZERO(&xfds);
 185
 186     if (ret <= 0) {
 187         return;
 188     }
 189
 190     for (i = 0; i < gpollfds->len; i++) {
 191         int fd = g_array_index(gpollfds, GPollFD, i).fd;
 192         int revents = g_array_index(gpollfds, GPollFD, i).revents;
 193
 194         if (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) {
 195             FD_SET(fd, &rfds);
 196         }
 197         if (revents & (G_IO_OUT | G_IO_ERR)) {
 198             FD_SET(fd, &wfds);
 199         }
 200         if (revents & G_IO_PRI) {
 201             FD_SET(fd, &xfds);
 202         }
 203     }
 204 }
 205
 206 #ifndef _WIN32
 207 static int glib_pollfds_idx;
 208 static int glib_n_poll_fds;
 209
 210 static void glib_pollfds_fill(uint32_t *cur_timeout)
 211 {
 212     GMainContext *context = g_main_context_default();
 213     int timeout = 0;
 214     int n;
 215
 216     g_main_context_prepare(context, &max_priority);
 217
 218     glib_pollfds_idx = gpollfds->len;
 219     n = glib_n_poll_fds;
 220     do {
 221         GPollFD *pfds;
 222         glib_n_poll_fds = n;
 223         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 224         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 225         n = g_main_context_query(context, max_priority, &timeout, pfds,
 226                                  glib_n_poll_fds);
 227     } while (n != glib_n_poll_fds);
 228
 229     if (timeout >= 0 && timeout < *cur_timeout) {
 230         *cur_timeout = timeout;
 231     }
 232 }
 233
 234 static void glib_pollfds_poll(void)
 235 {
 236     GMainContext *context = g_main_context_default();
 237     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 238
 239     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 240         g_main_context_dispatch(context);
 241     }
 242 }
 243
 244 static int os_host_main_loop_wait(uint32_t timeout)
 245 {
 246     int ret;
 247
 248     glib_pollfds_fill(&timeout);
 249
 250     if (timeout > 0) {
 251         qemu_mutex_unlock_iothread();
 252     }
 253
 254     /* We'll eventually drop fd_set completely.  But for now we still have
 255      * *_fill() and *_poll() functions that use rfds/wfds/xfds.
 256      */
 257     gpollfds_from_select();
 258
 259     ret = g_poll((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 260
 261     gpollfds_to_select(ret);
 262
 263     if (timeout > 0) {
 264         qemu_mutex_lock_iothread();
 265     }
 266
 267     glib_pollfds_poll();
 268     return ret;
 269 }
 270 #else
 271 /***********************************************************/
 272 /* Polling handling */
 273
 274 typedef struct PollingEntry {
 275     PollingFunc *func;
 276     void *opaque;
 277     struct PollingEntry *next;
 278 } PollingEntry;
 279
 280 static PollingEntry *first_polling_entry;
 281
 282 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 283 {
 284     PollingEntry **ppe, *pe;
 285     pe = g_malloc0(sizeof(PollingEntry));
 286     pe->func = func;
 287     pe->opaque = opaque;
 288     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 289     *ppe = pe;
 290     return 0;
 291 }
 292
 293 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 294 {
 295     PollingEntry **ppe, *pe;
 296     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 297         pe = *ppe;
 298         if (pe->func == func && pe->opaque == opaque) {
 299             *ppe = pe->next;
 300             g_free(pe);
 301             break;
 302         }
 303     }
 304 }
 305
 306 /***********************************************************/
 307 /* Wait objects support */
 308 typedef struct WaitObjects {
 309     int num;
 310     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 311     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 312     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 313     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 314 } WaitObjects;
 315
 316 static WaitObjects wait_objects = {0};
 317
 318 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 319 {
 320     WaitObjects *w = &wait_objects;
 321     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 322         return -1;
 323     }
 324     w->events[w->num] = handle;
 325     w->func[w->num] = func;
 326     w->opaque[w->num] = opaque;
 327     w->revents[w->num] = 0;
 328     w->num++;
 329     return 0;
 330 }
 331
 332 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 333 {
 334     int i, found;
 335     WaitObjects *w = &wait_objects;
 336
 337     found = 0;
 338     for (i = 0; i < w->num; i++) {
 339         if (w->events[i] == handle) {
 340             found = 1;
 341         }
 342         if (found) {
 343             w->events[i] = w->events[i + 1];
 344             w->func[i] = w->func[i + 1];
 345             w->opaque[i] = w->opaque[i + 1];
 346             w->revents[i] = w->revents[i + 1];
 347         }
 348     }
 349     if (found) {
 350         w->num--;
 351     }
 352 }
 353
 354 void qemu_fd_register(int fd)
 355 {
 356     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 357                    FD_READ | FD_ACCEPT | FD_CLOSE |
 358                    FD_CONNECT | FD_WRITE | FD_OOB);
 359 }
 360
 361 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 362                         fd_set *xfds)
 363 {
 364     int nfds = -1;
 365     int i;
 366
 367     for (i = 0; i < pollfds->len; i++) {
 368         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 369         int fd = pfd->fd;
 370         int events = pfd->events;
 371         if (events & (G_IO_IN | G_IO_HUP | G_IO_ERR)) {
 372             FD_SET(fd, rfds);
 373             nfds = MAX(nfds, fd);
 374         }
 375         if (events & (G_IO_OUT | G_IO_ERR)) {
 376             FD_SET(fd, wfds);
 377             nfds = MAX(nfds, fd);
 378         }
 379         if (events & G_IO_PRI) {
 380             FD_SET(fd, xfds);
 381             nfds = MAX(nfds, fd);
 382         }
 383     }
 384     return nfds;
 385 }
 386
 387 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 388                          fd_set *wfds, fd_set *xfds)
 389 {
 390     int i;
 391
 392     for (i = 0; i < pollfds->len; i++) {
 393         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 394         int fd = pfd->fd;
 395         int revents = 0;
 396
 397         if (FD_ISSET(fd, rfds)) {
 398             revents |= G_IO_IN | G_IO_HUP | G_IO_ERR;
 399         }
 400         if (FD_ISSET(fd, wfds)) {
 401             revents |= G_IO_OUT | G_IO_ERR;
 402         }
 403         if (FD_ISSET(fd, xfds)) {
 404             revents |= G_IO_PRI;
 405         }
 406         pfd->revents = revents & pfd->events;
 407     }
 408 }
 409
 410 static int os_host_main_loop_wait(uint32_t timeout)
 411 {
 412     GMainContext *context = g_main_context_default();
 413     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 414     int select_ret = 0;
 415     int g_poll_ret, ret, i, n_poll_fds;
 416     PollingEntry *pe;
 417     WaitObjects *w = &wait_objects;
 418     gint poll_timeout;
 419     static struct timeval tv0;
 420
 421     /* XXX: need to suppress polling by better using win32 events */
 422     ret = 0;
 423     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 424         ret |= pe->func(pe->opaque);
 425     }
 426     if (ret != 0) {
 427         return ret;
 428     }
 429
 430     g_main_context_prepare(context, &max_priority);
 431     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 432                                       poll_fds, ARRAY_SIZE(poll_fds));
 433     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 434
 435     for (i = 0; i < w->num; i++) {
 436         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 437         poll_fds[n_poll_fds + i].events = G_IO_IN;
 438     }
 439
 440     if (poll_timeout < 0 || timeout < poll_timeout) {
 441         poll_timeout = timeout;
 442     }
 443
 444     qemu_mutex_unlock_iothread();
 445     g_poll_ret = g_poll(poll_fds, n_poll_fds + w->num, poll_timeout);
 446     qemu_mutex_lock_iothread();
 447     if (g_poll_ret > 0) {
 448         for (i = 0; i < w->num; i++) {
 449             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 450         }
 451         for (i = 0; i < w->num; i++) {
 452             if (w->revents[i] && w->func[i]) {
 453                 w->func[i](w->opaque[i]);
 454             }
 455         }
 456     }
 457
 458     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 459         g_main_context_dispatch(context);
 460     }
 461
 462     /* Call select after g_poll to avoid a useless iteration and therefore
 463      * improve socket latency.
 464      */
 465
 466     /* This back-and-forth between GPollFDs and select(2) is temporary.  We'll
 467      * drop it in a couple of patches, I promise :).
 468      */
 469     gpollfds_from_select();
 470     FD_ZERO(&rfds);
 471     FD_ZERO(&wfds);
 472     FD_ZERO(&xfds);
 473     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 474     if (nfds >= 0) {
 475         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 476         if (select_ret != 0) {
 477             timeout = 0;
 478         }
 479         if (select_ret > 0) {
 480             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 481         }
 482     }
 483     gpollfds_to_select(select_ret);
 484
 485     return select_ret || g_poll_ret;
 486 }
 487 #endif
 488
 489 int main_loop_wait(int nonblocking)
 490 {
 491     int ret;
 492     uint32_t timeout = UINT32_MAX;
 493
 494     if (nonblocking) {
 495         timeout = 0;
 496     }
 497
 498     /* poll any events */
 499     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 500     /* XXX: separate device handlers from system ones */
 501     nfds = -1;
 502     FD_ZERO(&rfds);
 503     FD_ZERO(&wfds);
 504     FD_ZERO(&xfds);
 505
 506 #ifdef CONFIG_SLIRP
 507     slirp_update_timeout(&timeout);
 508     slirp_select_fill(&nfds, &rfds, &wfds, &xfds);
 509 #endif
 510     qemu_iohandler_fill(&nfds, &rfds, &wfds, &xfds);
 511     ret = os_host_main_loop_wait(timeout);
 512     qemu_iohandler_poll(&rfds, &wfds, &xfds, ret);
 513 #ifdef CONFIG_SLIRP
 514     slirp_select_poll(&rfds, &wfds, &xfds, (ret < 0));
 515 #endif
 516
 517     qemu_run_all_timers();
 518
 519     return ret;
 520 }
 521
 522 /* Functions to operate on the main QEMU AioContext.  */
 523
 524 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 525 {
 526     return aio_bh_new(qemu_aio_context, cb, opaque);
 527 }
 528
 529 bool qemu_aio_wait(void)
 530 {
 531     return aio_poll(qemu_aio_context, true);
 532 }
 533
 534 #ifdef CONFIG_POSIX
 535 void qemu_aio_set_fd_handler(int fd,
 536                              IOHandler *io_read,
 537                              IOHandler *io_write,
 538                              AioFlushHandler *io_flush,
 539                              void *opaque)
 540 {
 541     aio_set_fd_handler(qemu_aio_context, fd, io_read, io_write, io_flush,
 542                        opaque);
 543 }
 544 #endif
 545
 546 void qemu_aio_set_event_notifier(EventNotifier *notifier,
 547                                  EventNotifierHandler *io_read,
 548                                  AioFlushEventNotifierHandler *io_flush)
 549 {
 550     aio_set_event_notifier(qemu_aio_context, notifier, io_read, io_flush);
 551 }