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 "qemu/sockets.h"       // struct in_addr needed for libslirp.h
  28 #include "sysemu/qtest.h"
  29 #include "slirp/libslirp.h"
  30 #include "qemu/main-loop.h"
  31 #include "block/aio.h"
  32
  33 #ifndef _WIN32
  34
  35 #include "qemu/compatfd.h"
  36
  37 /* If we have signalfd, we mask out the signals we want to handle and then
  38  * use signalfd to listen for them.  We rely on whatever the current signal
  39  * handler is to dispatch the signals when we receive them.
  40  */
  41 static void sigfd_handler(void *opaque)
  42 {
  43     int fd = (intptr_t)opaque;
  44     struct qemu_signalfd_siginfo info;
  45     struct sigaction action;
  46     ssize_t len;
  47
  48     while (1) {
  49         do {
  50             len = read(fd, &info, sizeof(info));
  51         } while (len == -1 && errno == EINTR);
  52
  53         if (len == -1 && errno == EAGAIN) {
  54             break;
  55         }
  56
  57         if (len != sizeof(info)) {
  58             printf("read from sigfd returned %zd: %m\n", len);
  59             return;
  60         }
  61
  62         sigaction(info.ssi_signo, NULL, &action);
  63         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  64             action.sa_sigaction(info.ssi_signo,
  65                                 (siginfo_t *)&info, NULL);
  66         } else if (action.sa_handler) {
  67             action.sa_handler(info.ssi_signo);
  68         }
  69     }
  70 }
  71
  72 static int qemu_signal_init(void)
  73 {
  74     int sigfd;
  75     sigset_t set;
  76
  77     /*
  78      * SIG_IPI must be blocked in the main thread and must not be caught
  79      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  80      * not catch it reliably.
  81      */
  82     sigemptyset(&set);
  83     sigaddset(&set, SIG_IPI);
  84     sigaddset(&set, SIGIO);
  85     sigaddset(&set, SIGALRM);
  86     sigaddset(&set, SIGBUS);
  87     /* SIGINT cannot be handled via signalfd, so that ^C can be used
  88      * to interrupt QEMU when it is being run under gdb.  SIGHUP and
  89      * SIGTERM are also handled asynchronously, even though it is not
  90      * strictly necessary, because they use the same handler as SIGINT.
  91      */
  92     pthread_sigmask(SIG_BLOCK, &set, NULL);
  93
  94     sigdelset(&set, SIG_IPI);
  95     sigfd = qemu_signalfd(&set);
  96     if (sigfd == -1) {
  97         fprintf(stderr, "failed to create signalfd\n");
  98         return -errno;
  99     }
 100
 101     fcntl_setfl(sigfd, O_NONBLOCK);
 102
 103     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
 104                          (void *)(intptr_t)sigfd);
 105
 106     return 0;
 107 }
 108
 109 #else /* _WIN32 */
 110
 111 static int qemu_signal_init(void)
 112 {
 113     return 0;
 114 }
 115 #endif
 116
 117 static AioContext *qemu_aio_context;
 118
 119 AioContext *qemu_get_aio_context(void)
 120 {
 121     return qemu_aio_context;
 122 }
 123
 124 void qemu_notify_event(void)
 125 {
 126     if (!qemu_aio_context) {
 127         return;
 128     }
 129     aio_notify(qemu_aio_context);
 130 }
 131
 132 static GArray *gpollfds;
 133
 134 int qemu_init_main_loop(Error **errp)
 135 {
 136     int ret;
 137     GSource *src;
 138     Error *local_error = NULL;
 139
 140     init_clocks();
 141
 142     ret = qemu_signal_init();
 143     if (ret) {
 144         return ret;
 145     }
 146
 147     qemu_aio_context = aio_context_new(&local_error);
 148     if (!qemu_aio_context) {
 149         error_propagate(errp, local_error);
 150         return -EMFILE;
 151     }
 152     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 153     src = aio_get_g_source(qemu_aio_context);
 154     g_source_attach(src, NULL);
 155     g_source_unref(src);
 156     return 0;
 157 }
 158
 159 static int max_priority;
 160
 161 #ifndef _WIN32
 162 static int glib_pollfds_idx;
 163 static int glib_n_poll_fds;
 164
 165 static void glib_pollfds_fill(int64_t *cur_timeout)
 166 {
 167     GMainContext *context = g_main_context_default();
 168     int timeout = 0;
 169     int64_t timeout_ns;
 170     int n;
 171
 172     g_main_context_prepare(context, &max_priority);
 173
 174     glib_pollfds_idx = gpollfds->len;
 175     n = glib_n_poll_fds;
 176     do {
 177         GPollFD *pfds;
 178         glib_n_poll_fds = n;
 179         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 180         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 181         n = g_main_context_query(context, max_priority, &timeout, pfds,
 182                                  glib_n_poll_fds);
 183     } while (n != glib_n_poll_fds);
 184
 185     if (timeout < 0) {
 186         timeout_ns = -1;
 187     } else {
 188         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 189     }
 190
 191     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 192 }
 193
 194 static void glib_pollfds_poll(void)
 195 {
 196     GMainContext *context = g_main_context_default();
 197     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 198
 199     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 200         g_main_context_dispatch(context);
 201     }
 202 }
 203
 204 #define MAX_MAIN_LOOP_SPIN (1000)
 205
 206 static int os_host_main_loop_wait(int64_t timeout)
 207 {
 208     int ret;
 209     static int spin_counter;
 210
 211     glib_pollfds_fill(&timeout);
 212
 213     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 214      * the I/O thread, this can lead to starvation of the BQL such that the
 215      * VCPU threads never run.  To make sure we can detect the later case,
 216      * print a message to the screen.  If we run into this condition, create
 217      * a fake timeout in order to give the VCPU threads a chance to run.
 218      */
 219     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 220         static bool notified;
 221
 222         if (!notified && !qtest_enabled()) {
 223             fprintf(stderr,
 224                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 225                     MAX_MAIN_LOOP_SPIN);
 226             notified = true;
 227         }
 228
 229         timeout = SCALE_MS;
 230     }
 231
 232     if (timeout) {
 233         spin_counter = 0;
 234         qemu_mutex_unlock_iothread();
 235     } else {
 236         spin_counter++;
 237     }
 238
 239     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 240
 241     if (timeout) {
 242         qemu_mutex_lock_iothread();
 243     }
 244
 245     glib_pollfds_poll();
 246     return ret;
 247 }
 248 #else
 249 /***********************************************************/
 250 /* Polling handling */
 251
 252 typedef struct PollingEntry {
 253     PollingFunc *func;
 254     void *opaque;
 255     struct PollingEntry *next;
 256 } PollingEntry;
 257
 258 static PollingEntry *first_polling_entry;
 259
 260 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 261 {
 262     PollingEntry **ppe, *pe;
 263     pe = g_malloc0(sizeof(PollingEntry));
 264     pe->func = func;
 265     pe->opaque = opaque;
 266     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 267     *ppe = pe;
 268     return 0;
 269 }
 270
 271 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 272 {
 273     PollingEntry **ppe, *pe;
 274     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 275         pe = *ppe;
 276         if (pe->func == func && pe->opaque == opaque) {
 277             *ppe = pe->next;
 278             g_free(pe);
 279             break;
 280         }
 281     }
 282 }
 283
 284 /***********************************************************/
 285 /* Wait objects support */
 286 typedef struct WaitObjects {
 287     int num;
 288     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 289     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 290     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 291     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 292 } WaitObjects;
 293
 294 static WaitObjects wait_objects = {0};
 295
 296 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 297 {
 298     WaitObjects *w = &wait_objects;
 299     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 300         return -1;
 301     }
 302     w->events[w->num] = handle;
 303     w->func[w->num] = func;
 304     w->opaque[w->num] = opaque;
 305     w->revents[w->num] = 0;
 306     w->num++;
 307     return 0;
 308 }
 309
 310 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 311 {
 312     int i, found;
 313     WaitObjects *w = &wait_objects;
 314
 315     found = 0;
 316     for (i = 0; i < w->num; i++) {
 317         if (w->events[i] == handle) {
 318             found = 1;
 319         }
 320         if (found) {
 321             w->events[i] = w->events[i + 1];
 322             w->func[i] = w->func[i + 1];
 323             w->opaque[i] = w->opaque[i + 1];
 324             w->revents[i] = w->revents[i + 1];
 325         }
 326     }
 327     if (found) {
 328         w->num--;
 329     }
 330 }
 331
 332 void qemu_fd_register(int fd)
 333 {
 334     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 335                    FD_READ | FD_ACCEPT | FD_CLOSE |
 336                    FD_CONNECT | FD_WRITE | FD_OOB);
 337 }
 338
 339 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 340                         fd_set *xfds)
 341 {
 342     int nfds = -1;
 343     int i;
 344
 345     for (i = 0; i < pollfds->len; i++) {
 346         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 347         int fd = pfd->fd;
 348         int events = pfd->events;
 349         if (events & G_IO_IN) {
 350             FD_SET(fd, rfds);
 351             nfds = MAX(nfds, fd);
 352         }
 353         if (events & G_IO_OUT) {
 354             FD_SET(fd, wfds);
 355             nfds = MAX(nfds, fd);
 356         }
 357         if (events & G_IO_PRI) {
 358             FD_SET(fd, xfds);
 359             nfds = MAX(nfds, fd);
 360         }
 361     }
 362     return nfds;
 363 }
 364
 365 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 366                          fd_set *wfds, fd_set *xfds)
 367 {
 368     int i;
 369
 370     for (i = 0; i < pollfds->len; i++) {
 371         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 372         int fd = pfd->fd;
 373         int revents = 0;
 374
 375         if (FD_ISSET(fd, rfds)) {
 376             revents |= G_IO_IN;
 377         }
 378         if (FD_ISSET(fd, wfds)) {
 379             revents |= G_IO_OUT;
 380         }
 381         if (FD_ISSET(fd, xfds)) {
 382             revents |= G_IO_PRI;
 383         }
 384         pfd->revents = revents & pfd->events;
 385     }
 386 }
 387
 388 static int os_host_main_loop_wait(int64_t timeout)
 389 {
 390     GMainContext *context = g_main_context_default();
 391     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 392     int select_ret = 0;
 393     int g_poll_ret, ret, i, n_poll_fds;
 394     PollingEntry *pe;
 395     WaitObjects *w = &wait_objects;
 396     gint poll_timeout;
 397     int64_t poll_timeout_ns;
 398     static struct timeval tv0;
 399     fd_set rfds, wfds, xfds;
 400     int nfds;
 401
 402     /* XXX: need to suppress polling by better using win32 events */
 403     ret = 0;
 404     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 405         ret |= pe->func(pe->opaque);
 406     }
 407     if (ret != 0) {
 408         return ret;
 409     }
 410
 411     FD_ZERO(&rfds);
 412     FD_ZERO(&wfds);
 413     FD_ZERO(&xfds);
 414     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 415     if (nfds >= 0) {
 416         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 417         if (select_ret != 0) {
 418             timeout = 0;
 419         }
 420         if (select_ret > 0) {
 421             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 422         }
 423     }
 424
 425     g_main_context_prepare(context, &max_priority);
 426     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 427                                       poll_fds, ARRAY_SIZE(poll_fds));
 428     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 429
 430     for (i = 0; i < w->num; i++) {
 431         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 432         poll_fds[n_poll_fds + i].events = G_IO_IN;
 433     }
 434
 435     if (poll_timeout < 0) {
 436         poll_timeout_ns = -1;
 437     } else {
 438         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 439     }
 440
 441     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 442
 443     qemu_mutex_unlock_iothread();
 444     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 445
 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     return select_ret || g_poll_ret;
 463 }
 464 #endif
 465
 466 int main_loop_wait(int nonblocking)
 467 {
 468     int ret;
 469     uint32_t timeout = UINT32_MAX;
 470     int64_t timeout_ns;
 471
 472     if (nonblocking) {
 473         timeout = 0;
 474     }
 475
 476     /* poll any events */
 477     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 478     /* XXX: separate device handlers from system ones */
 479 #ifdef CONFIG_SLIRP
 480     slirp_pollfds_fill(gpollfds, &timeout);
 481 #endif
 482     qemu_iohandler_fill(gpollfds);
 483
 484     if (timeout == UINT32_MAX) {
 485         timeout_ns = -1;
 486     } else {
 487         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 488     }
 489
 490     timeout_ns = qemu_soonest_timeout(timeout_ns,
 491                                       timerlistgroup_deadline_ns(
 492                                           &main_loop_tlg));
 493
 494     ret = os_host_main_loop_wait(timeout_ns);
 495     qemu_iohandler_poll(gpollfds, ret);
 496 #ifdef CONFIG_SLIRP
 497     slirp_pollfds_poll(gpollfds, (ret < 0));
 498 #endif
 499
 500     qemu_clock_run_all_timers();
 501
 502     return ret;
 503 }
 504
 505 /* Functions to operate on the main QEMU AioContext.  */
 506
 507 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 508 {
 509     return aio_bh_new(qemu_aio_context, cb, opaque);
 510 }