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 "slirp/libslirp.h"
  29 #include "qemu/main-loop.h"
  30 #include "block/aio.h"
  31
  32 #ifndef _WIN32
  33
  34 #include "qemu/compatfd.h"
  35
  36 /* If we have signalfd, we mask out the signals we want to handle and then
  37  * use signalfd to listen for them.  We rely on whatever the current signal
  38  * handler is to dispatch the signals when we receive them.
  39  */
  40 static void sigfd_handler(void *opaque)
  41 {
  42     int fd = (intptr_t)opaque;
  43     struct qemu_signalfd_siginfo info;
  44     struct sigaction action;
  45     ssize_t len;
  46
  47     while (1) {
  48         do {
  49             len = read(fd, &info, sizeof(info));
  50         } while (len == -1 && errno == EINTR);
  51
  52         if (len == -1 && errno == EAGAIN) {
  53             break;
  54         }
  55
  56         if (len != sizeof(info)) {
  57             printf("read from sigfd returned %zd: %m\n", len);
  58             return;
  59         }
  60
  61         sigaction(info.ssi_signo, NULL, &action);
  62         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  63             action.sa_sigaction(info.ssi_signo,
  64                                 (siginfo_t *)&info, NULL);
  65         } else if (action.sa_handler) {
  66             action.sa_handler(info.ssi_signo);
  67         }
  68     }
  69 }
  70
  71 static int qemu_signal_init(void)
  72 {
  73     int sigfd;
  74     sigset_t set;
  75
  76     /*
  77      * SIG_IPI must be blocked in the main thread and must not be caught
  78      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  79      * not catch it reliably.
  80      */
  81     sigemptyset(&set);
  82     sigaddset(&set, SIG_IPI);
  83     sigaddset(&set, SIGIO);
  84     sigaddset(&set, SIGALRM);
  85     sigaddset(&set, SIGBUS);
  86     pthread_sigmask(SIG_BLOCK, &set, NULL);
  87
  88     sigdelset(&set, SIG_IPI);
  89     sigfd = qemu_signalfd(&set);
  90     if (sigfd == -1) {
  91         fprintf(stderr, "failed to create signalfd\n");
  92         return -errno;
  93     }
  94
  95     fcntl_setfl(sigfd, O_NONBLOCK);
  96
  97     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
  98                          (void *)(intptr_t)sigfd);
  99
 100     return 0;
 101 }
 102
 103 #else /* _WIN32 */
 104
 105 static int qemu_signal_init(void)
 106 {
 107     return 0;
 108 }
 109 #endif
 110
 111 static AioContext *qemu_aio_context;
 112
 113 AioContext *qemu_get_aio_context(void)
 114 {
 115     return qemu_aio_context;
 116 }
 117
 118 void qemu_notify_event(void)
 119 {
 120     if (!qemu_aio_context) {
 121         return;
 122     }
 123     aio_notify(qemu_aio_context);
 124 }
 125
 126 static GArray *gpollfds;
 127
 128 int qemu_init_main_loop(void)
 129 {
 130     int ret;
 131     GSource *src;
 132
 133     init_clocks();
 134
 135     ret = qemu_signal_init();
 136     if (ret) {
 137         return ret;
 138     }
 139
 140     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 141     qemu_aio_context = aio_context_new();
 142     src = aio_get_g_source(qemu_aio_context);
 143     g_source_attach(src, NULL);
 144     g_source_unref(src);
 145     return 0;
 146 }
 147
 148 static int max_priority;
 149
 150 #ifndef _WIN32
 151 static int glib_pollfds_idx;
 152 static int glib_n_poll_fds;
 153
 154 static void glib_pollfds_fill(int64_t *cur_timeout)
 155 {
 156     GMainContext *context = g_main_context_default();
 157     int timeout = 0;
 158     int64_t timeout_ns;
 159     int n;
 160
 161     g_main_context_prepare(context, &max_priority);
 162
 163     glib_pollfds_idx = gpollfds->len;
 164     n = glib_n_poll_fds;
 165     do {
 166         GPollFD *pfds;
 167         glib_n_poll_fds = n;
 168         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 169         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 170         n = g_main_context_query(context, max_priority, &timeout, pfds,
 171                                  glib_n_poll_fds);
 172     } while (n != glib_n_poll_fds);
 173
 174     if (timeout < 0) {
 175         timeout_ns = -1;
 176     } else {
 177         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 178     }
 179
 180     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 181 }
 182
 183 static void glib_pollfds_poll(void)
 184 {
 185     GMainContext *context = g_main_context_default();
 186     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 187
 188     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 189         g_main_context_dispatch(context);
 190     }
 191 }
 192
 193 #define MAX_MAIN_LOOP_SPIN (1000)
 194
 195 static int os_host_main_loop_wait(int64_t timeout)
 196 {
 197     int ret;
 198     static int spin_counter;
 199
 200     glib_pollfds_fill(&timeout);
 201
 202     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 203      * the I/O thread, this can lead to starvation of the BQL such that the
 204      * VCPU threads never run.  To make sure we can detect the later case,
 205      * print a message to the screen.  If we run into this condition, create
 206      * a fake timeout in order to give the VCPU threads a chance to run.
 207      */
 208     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 209         static bool notified;
 210
 211         if (!notified) {
 212             fprintf(stderr,
 213                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 214                     MAX_MAIN_LOOP_SPIN);
 215             notified = true;
 216         }
 217
 218         timeout = SCALE_MS;
 219     }
 220
 221     if (timeout) {
 222         spin_counter = 0;
 223         qemu_mutex_unlock_iothread();
 224     } else {
 225         spin_counter++;
 226     }
 227
 228     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 229
 230     if (timeout) {
 231         qemu_mutex_lock_iothread();
 232     }
 233
 234     glib_pollfds_poll();
 235     return ret;
 236 }
 237 #else
 238 /***********************************************************/
 239 /* Polling handling */
 240
 241 typedef struct PollingEntry {
 242     PollingFunc *func;
 243     void *opaque;
 244     struct PollingEntry *next;
 245 } PollingEntry;
 246
 247 static PollingEntry *first_polling_entry;
 248
 249 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 250 {
 251     PollingEntry **ppe, *pe;
 252     pe = g_malloc0(sizeof(PollingEntry));
 253     pe->func = func;
 254     pe->opaque = opaque;
 255     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 256     *ppe = pe;
 257     return 0;
 258 }
 259
 260 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 261 {
 262     PollingEntry **ppe, *pe;
 263     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 264         pe = *ppe;
 265         if (pe->func == func && pe->opaque == opaque) {
 266             *ppe = pe->next;
 267             g_free(pe);
 268             break;
 269         }
 270     }
 271 }
 272
 273 /***********************************************************/
 274 /* Wait objects support */
 275 typedef struct WaitObjects {
 276     int num;
 277     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 278     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 279     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 280     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 281 } WaitObjects;
 282
 283 static WaitObjects wait_objects = {0};
 284
 285 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 286 {
 287     WaitObjects *w = &wait_objects;
 288     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 289         return -1;
 290     }
 291     w->events[w->num] = handle;
 292     w->func[w->num] = func;
 293     w->opaque[w->num] = opaque;
 294     w->revents[w->num] = 0;
 295     w->num++;
 296     return 0;
 297 }
 298
 299 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 300 {
 301     int i, found;
 302     WaitObjects *w = &wait_objects;
 303
 304     found = 0;
 305     for (i = 0; i < w->num; i++) {
 306         if (w->events[i] == handle) {
 307             found = 1;
 308         }
 309         if (found) {
 310             w->events[i] = w->events[i + 1];
 311             w->func[i] = w->func[i + 1];
 312             w->opaque[i] = w->opaque[i + 1];
 313             w->revents[i] = w->revents[i + 1];
 314         }
 315     }
 316     if (found) {
 317         w->num--;
 318     }
 319 }
 320
 321 void qemu_fd_register(int fd)
 322 {
 323     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 324                    FD_READ | FD_ACCEPT | FD_CLOSE |
 325                    FD_CONNECT | FD_WRITE | FD_OOB);
 326 }
 327
 328 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 329                         fd_set *xfds)
 330 {
 331     int nfds = -1;
 332     int i;
 333
 334     for (i = 0; i < pollfds->len; i++) {
 335         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 336         int fd = pfd->fd;
 337         int events = pfd->events;
 338         if (events & G_IO_IN) {
 339             FD_SET(fd, rfds);
 340             nfds = MAX(nfds, fd);
 341         }
 342         if (events & G_IO_OUT) {
 343             FD_SET(fd, wfds);
 344             nfds = MAX(nfds, fd);
 345         }
 346         if (events & G_IO_PRI) {
 347             FD_SET(fd, xfds);
 348             nfds = MAX(nfds, fd);
 349         }
 350     }
 351     return nfds;
 352 }
 353
 354 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 355                          fd_set *wfds, fd_set *xfds)
 356 {
 357     int i;
 358
 359     for (i = 0; i < pollfds->len; i++) {
 360         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 361         int fd = pfd->fd;
 362         int revents = 0;
 363
 364         if (FD_ISSET(fd, rfds)) {
 365             revents |= G_IO_IN;
 366         }
 367         if (FD_ISSET(fd, wfds)) {
 368             revents |= G_IO_OUT;
 369         }
 370         if (FD_ISSET(fd, xfds)) {
 371             revents |= G_IO_PRI;
 372         }
 373         pfd->revents = revents & pfd->events;
 374     }
 375 }
 376
 377 static int os_host_main_loop_wait(int64_t timeout)
 378 {
 379     GMainContext *context = g_main_context_default();
 380     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 381     int select_ret = 0;
 382     int g_poll_ret, ret, i, n_poll_fds;
 383     PollingEntry *pe;
 384     WaitObjects *w = &wait_objects;
 385     gint poll_timeout;
 386     int64_t poll_timeout_ns;
 387     static struct timeval tv0;
 388     fd_set rfds, wfds, xfds;
 389     int nfds;
 390
 391     /* XXX: need to suppress polling by better using win32 events */
 392     ret = 0;
 393     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 394         ret |= pe->func(pe->opaque);
 395     }
 396     if (ret != 0) {
 397         return ret;
 398     }
 399
 400     FD_ZERO(&rfds);
 401     FD_ZERO(&wfds);
 402     FD_ZERO(&xfds);
 403     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 404     if (nfds >= 0) {
 405         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 406         if (select_ret != 0) {
 407             timeout = 0;
 408         }
 409         if (select_ret > 0) {
 410             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 411         }
 412     }
 413
 414     g_main_context_prepare(context, &max_priority);
 415     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 416                                       poll_fds, ARRAY_SIZE(poll_fds));
 417     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 418
 419     for (i = 0; i < w->num; i++) {
 420         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 421         poll_fds[n_poll_fds + i].events = G_IO_IN;
 422     }
 423
 424     if (poll_timeout < 0) {
 425         poll_timeout_ns = -1;
 426     } else {
 427         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 428     }
 429
 430     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 431
 432     qemu_mutex_unlock_iothread();
 433     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 434
 435     qemu_mutex_lock_iothread();
 436     if (g_poll_ret > 0) {
 437         for (i = 0; i < w->num; i++) {
 438             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 439         }
 440         for (i = 0; i < w->num; i++) {
 441             if (w->revents[i] && w->func[i]) {
 442                 w->func[i](w->opaque[i]);
 443             }
 444         }
 445     }
 446
 447     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 448         g_main_context_dispatch(context);
 449     }
 450
 451     return select_ret || g_poll_ret;
 452 }
 453 #endif
 454
 455 int main_loop_wait(int nonblocking)
 456 {
 457     int ret;
 458     uint32_t timeout = UINT32_MAX;
 459     int64_t timeout_ns;
 460
 461     if (nonblocking) {
 462         timeout = 0;
 463     }
 464
 465     /* poll any events */
 466     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 467     /* XXX: separate device handlers from system ones */
 468 #ifdef CONFIG_SLIRP
 469     slirp_update_timeout(&timeout);
 470     slirp_pollfds_fill(gpollfds);
 471 #endif
 472     qemu_iohandler_fill(gpollfds);
 473
 474     if (timeout == UINT32_MAX) {
 475         timeout_ns = -1;
 476     } else {
 477         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 478     }
 479
 480     timeout_ns = qemu_soonest_timeout(timeout_ns,
 481                                       timerlistgroup_deadline_ns(
 482                                           &main_loop_tlg));
 483
 484     ret = os_host_main_loop_wait(timeout_ns);
 485     qemu_iohandler_poll(gpollfds, ret);
 486 #ifdef CONFIG_SLIRP
 487     slirp_pollfds_poll(gpollfds, (ret < 0));
 488 #endif
 489
 490     qemu_clock_run_all_timers();
 491
 492     return ret;
 493 }
 494
 495 /* Functions to operate on the main QEMU AioContext.  */
 496
 497 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 498 {
 499     return aio_bh_new(qemu_aio_context, cb, opaque);
 500 }
 501
 502 bool qemu_aio_wait(void)
 503 {
 504     return aio_poll(qemu_aio_context, true);
 505 }
 506
 507 #ifdef CONFIG_POSIX
 508 void qemu_aio_set_fd_handler(int fd,
 509                              IOHandler *io_read,
 510                              IOHandler *io_write,
 511                              void *opaque)
 512 {
 513     aio_set_fd_handler(qemu_aio_context, fd, io_read, io_write, opaque);
 514 }
 515 #endif
 516
 517 void qemu_aio_set_event_notifier(EventNotifier *notifier,
 518                                  EventNotifierHandler *io_read)
 519 {
 520     aio_set_event_notifier(qemu_aio_context, notifier, io_read);
 521 }