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