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_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
 104
 105     return 0;
 106 }
 107
 108 #else /* _WIN32 */
 109
 110 static int qemu_signal_init(void)
 111 {
 112     return 0;
 113 }
 114 #endif
 115
 116 static AioContext *qemu_aio_context;
 117 static QEMUBH *qemu_notify_bh;
 118
 119 static void notify_event_cb(void *opaque)
 120 {
 121     /* No need to do anything; this bottom half is only used to
 122      * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
 123      */
 124 }
 125
 126 AioContext *qemu_get_aio_context(void)
 127 {
 128     return qemu_aio_context;
 129 }
 130
 131 void qemu_notify_event(void)
 132 {
 133     if (!qemu_aio_context) {
 134         return;
 135     }
 136     qemu_bh_schedule(qemu_notify_bh);
 137 }
 138
 139 static GArray *gpollfds;
 140
 141 int qemu_init_main_loop(Error **errp)
 142 {
 143     int ret;
 144     GSource *src;
 145     Error *local_error = NULL;
 146
 147     init_clocks();
 148
 149     ret = qemu_signal_init();
 150     if (ret) {
 151         return ret;
 152     }
 153
 154     qemu_aio_context = aio_context_new(&local_error);
 155     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 156     if (!qemu_aio_context) {
 157         error_propagate(errp, local_error);
 158         return -EMFILE;
 159     }
 160     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 161     src = aio_get_g_source(qemu_aio_context);
 162     g_source_attach(src, NULL);
 163     g_source_unref(src);
 164     return 0;
 165 }
 166
 167 static int max_priority;
 168
 169 #ifndef _WIN32
 170 static int glib_pollfds_idx;
 171 static int glib_n_poll_fds;
 172
 173 static void glib_pollfds_fill(int64_t *cur_timeout)
 174 {
 175     GMainContext *context = g_main_context_default();
 176     int timeout = 0;
 177     int64_t timeout_ns;
 178     int n;
 179
 180     g_main_context_prepare(context, &max_priority);
 181
 182     glib_pollfds_idx = gpollfds->len;
 183     n = glib_n_poll_fds;
 184     do {
 185         GPollFD *pfds;
 186         glib_n_poll_fds = n;
 187         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 188         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 189         n = g_main_context_query(context, max_priority, &timeout, pfds,
 190                                  glib_n_poll_fds);
 191     } while (n != glib_n_poll_fds);
 192
 193     if (timeout < 0) {
 194         timeout_ns = -1;
 195     } else {
 196         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 197     }
 198
 199     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 200 }
 201
 202 static void glib_pollfds_poll(void)
 203 {
 204     GMainContext *context = g_main_context_default();
 205     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 206
 207     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 208         g_main_context_dispatch(context);
 209     }
 210 }
 211
 212 #define MAX_MAIN_LOOP_SPIN (1000)
 213
 214 static int os_host_main_loop_wait(int64_t timeout)
 215 {
 216     int ret;
 217     static int spin_counter;
 218
 219     glib_pollfds_fill(&timeout);
 220
 221     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 222      * the I/O thread, this can lead to starvation of the BQL such that the
 223      * VCPU threads never run.  To make sure we can detect the later case,
 224      * print a message to the screen.  If we run into this condition, create
 225      * a fake timeout in order to give the VCPU threads a chance to run.
 226      */
 227     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 228         static bool notified;
 229
 230         if (!notified && !qtest_enabled()) {
 231             fprintf(stderr,
 232                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 233                     MAX_MAIN_LOOP_SPIN);
 234             notified = true;
 235         }
 236
 237         timeout = SCALE_MS;
 238     }
 239
 240     if (timeout) {
 241         spin_counter = 0;
 242         qemu_mutex_unlock_iothread();
 243     } else {
 244         spin_counter++;
 245     }
 246
 247     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 248
 249     if (timeout) {
 250         qemu_mutex_lock_iothread();
 251     }
 252
 253     glib_pollfds_poll();
 254     return ret;
 255 }
 256 #else
 257 /***********************************************************/
 258 /* Polling handling */
 259
 260 typedef struct PollingEntry {
 261     PollingFunc *func;
 262     void *opaque;
 263     struct PollingEntry *next;
 264 } PollingEntry;
 265
 266 static PollingEntry *first_polling_entry;
 267
 268 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 269 {
 270     PollingEntry **ppe, *pe;
 271     pe = g_malloc0(sizeof(PollingEntry));
 272     pe->func = func;
 273     pe->opaque = opaque;
 274     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 275     *ppe = pe;
 276     return 0;
 277 }
 278
 279 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 280 {
 281     PollingEntry **ppe, *pe;
 282     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 283         pe = *ppe;
 284         if (pe->func == func && pe->opaque == opaque) {
 285             *ppe = pe->next;
 286             g_free(pe);
 287             break;
 288         }
 289     }
 290 }
 291
 292 /***********************************************************/
 293 /* Wait objects support */
 294 typedef struct WaitObjects {
 295     int num;
 296     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 297     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 298     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 299     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 300 } WaitObjects;
 301
 302 static WaitObjects wait_objects = {0};
 303
 304 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 305 {
 306     WaitObjects *w = &wait_objects;
 307     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 308         return -1;
 309     }
 310     w->events[w->num] = handle;
 311     w->func[w->num] = func;
 312     w->opaque[w->num] = opaque;
 313     w->revents[w->num] = 0;
 314     w->num++;
 315     return 0;
 316 }
 317
 318 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 319 {
 320     int i, found;
 321     WaitObjects *w = &wait_objects;
 322
 323     found = 0;
 324     for (i = 0; i < w->num; i++) {
 325         if (w->events[i] == handle) {
 326             found = 1;
 327         }
 328         if (found) {
 329             w->events[i] = w->events[i + 1];
 330             w->func[i] = w->func[i + 1];
 331             w->opaque[i] = w->opaque[i + 1];
 332             w->revents[i] = w->revents[i + 1];
 333         }
 334     }
 335     if (found) {
 336         w->num--;
 337     }
 338 }
 339
 340 void qemu_fd_register(int fd)
 341 {
 342     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 343                    FD_READ | FD_ACCEPT | FD_CLOSE |
 344                    FD_CONNECT | FD_WRITE | FD_OOB);
 345 }
 346
 347 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 348                         fd_set *xfds)
 349 {
 350     int nfds = -1;
 351     int i;
 352
 353     for (i = 0; i < pollfds->len; i++) {
 354         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 355         int fd = pfd->fd;
 356         int events = pfd->events;
 357         if (events & G_IO_IN) {
 358             FD_SET(fd, rfds);
 359             nfds = MAX(nfds, fd);
 360         }
 361         if (events & G_IO_OUT) {
 362             FD_SET(fd, wfds);
 363             nfds = MAX(nfds, fd);
 364         }
 365         if (events & G_IO_PRI) {
 366             FD_SET(fd, xfds);
 367             nfds = MAX(nfds, fd);
 368         }
 369     }
 370     return nfds;
 371 }
 372
 373 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 374                          fd_set *wfds, fd_set *xfds)
 375 {
 376     int i;
 377
 378     for (i = 0; i < pollfds->len; i++) {
 379         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 380         int fd = pfd->fd;
 381         int revents = 0;
 382
 383         if (FD_ISSET(fd, rfds)) {
 384             revents |= G_IO_IN;
 385         }
 386         if (FD_ISSET(fd, wfds)) {
 387             revents |= G_IO_OUT;
 388         }
 389         if (FD_ISSET(fd, xfds)) {
 390             revents |= G_IO_PRI;
 391         }
 392         pfd->revents = revents & pfd->events;
 393     }
 394 }
 395
 396 static int os_host_main_loop_wait(int64_t timeout)
 397 {
 398     GMainContext *context = g_main_context_default();
 399     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 400     int select_ret = 0;
 401     int g_poll_ret, ret, i, n_poll_fds;
 402     PollingEntry *pe;
 403     WaitObjects *w = &wait_objects;
 404     gint poll_timeout;
 405     int64_t poll_timeout_ns;
 406     static struct timeval tv0;
 407     fd_set rfds, wfds, xfds;
 408     int nfds;
 409
 410     /* XXX: need to suppress polling by better using win32 events */
 411     ret = 0;
 412     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 413         ret |= pe->func(pe->opaque);
 414     }
 415     if (ret != 0) {
 416         return ret;
 417     }
 418
 419     FD_ZERO(&rfds);
 420     FD_ZERO(&wfds);
 421     FD_ZERO(&xfds);
 422     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 423     if (nfds >= 0) {
 424         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 425         if (select_ret != 0) {
 426             timeout = 0;
 427         }
 428         if (select_ret > 0) {
 429             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 430         }
 431     }
 432
 433     g_main_context_prepare(context, &max_priority);
 434     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 435                                       poll_fds, ARRAY_SIZE(poll_fds));
 436     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 437
 438     for (i = 0; i < w->num; i++) {
 439         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 440         poll_fds[n_poll_fds + i].events = G_IO_IN;
 441     }
 442
 443     if (poll_timeout < 0) {
 444         poll_timeout_ns = -1;
 445     } else {
 446         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 447     }
 448
 449     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 450
 451     qemu_mutex_unlock_iothread();
 452     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 453
 454     qemu_mutex_lock_iothread();
 455     if (g_poll_ret > 0) {
 456         for (i = 0; i < w->num; i++) {
 457             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 458         }
 459         for (i = 0; i < w->num; i++) {
 460             if (w->revents[i] && w->func[i]) {
 461                 w->func[i](w->opaque[i]);
 462             }
 463         }
 464     }
 465
 466     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 467         g_main_context_dispatch(context);
 468     }
 469
 470     return select_ret || g_poll_ret;
 471 }
 472 #endif
 473
 474 int main_loop_wait(int nonblocking)
 475 {
 476     int ret;
 477     uint32_t timeout = UINT32_MAX;
 478     int64_t timeout_ns;
 479
 480     if (nonblocking) {
 481         timeout = 0;
 482     }
 483
 484     /* poll any events */
 485     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 486     /* XXX: separate device handlers from system ones */
 487 #ifdef CONFIG_SLIRP
 488     slirp_pollfds_fill(gpollfds, &timeout);
 489 #endif
 490     qemu_iohandler_fill(gpollfds);
 491
 492     if (timeout == UINT32_MAX) {
 493         timeout_ns = -1;
 494     } else {
 495         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 496     }
 497
 498     timeout_ns = qemu_soonest_timeout(timeout_ns,
 499                                       timerlistgroup_deadline_ns(
 500                                           &main_loop_tlg));
 501
 502     ret = os_host_main_loop_wait(timeout_ns);
 503     qemu_iohandler_poll(gpollfds, ret);
 504 #ifdef CONFIG_SLIRP
 505     slirp_pollfds_poll(gpollfds, (ret < 0));
 506 #endif
 507
 508     qemu_clock_run_all_timers();
 509
 510     return ret;
 511 }
 512
 513 /* Functions to operate on the main QEMU AioContext.  */
 514
 515 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 516 {
 517     return aio_bh_new(qemu_aio_context, cb, opaque);
 518 }