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