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/osdep.h"
  26 #include "qapi/error.h"
  27 #include "qemu/cutils.h"
  28 #include "qemu/timer.h"
  29 #include "qemu/sockets.h"       // struct in_addr needed for libslirp.h
  30 #include "sysemu/qtest.h"
  31 #include "slirp/libslirp.h"
  32 #include "qemu/main-loop.h"
  33 #include "block/aio.h"
  34
  35 #ifndef _WIN32
  36
  37 #include "qemu/compatfd.h"
  38
  39 /* If we have signalfd, we mask out the signals we want to handle and then
  40  * use signalfd to listen for them.  We rely on whatever the current signal
  41  * handler is to dispatch the signals when we receive them.
  42  */
  43 static void sigfd_handler(void *opaque)
  44 {
  45     int fd = (intptr_t)opaque;
  46     struct qemu_signalfd_siginfo info;
  47     struct sigaction action;
  48     ssize_t len;
  49
  50     while (1) {
  51         do {
  52             len = read(fd, &info, sizeof(info));
  53         } while (len == -1 && errno == EINTR);
  54
  55         if (len == -1 && errno == EAGAIN) {
  56             break;
  57         }
  58
  59         if (len != sizeof(info)) {
  60             printf("read from sigfd returned %zd: %m\n", len);
  61             return;
  62         }
  63
  64         sigaction(info.ssi_signo, NULL, &action);
  65         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  66             action.sa_sigaction(info.ssi_signo,
  67                                 (siginfo_t *)&info, NULL);
  68         } else if (action.sa_handler) {
  69             action.sa_handler(info.ssi_signo);
  70         }
  71     }
  72 }
  73
  74 static int qemu_signal_init(void)
  75 {
  76     int sigfd;
  77     sigset_t set;
  78
  79     /*
  80      * SIG_IPI must be blocked in the main thread and must not be caught
  81      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  82      * not catch it reliably.
  83      */
  84     sigemptyset(&set);
  85     sigaddset(&set, SIG_IPI);
  86     sigaddset(&set, SIGIO);
  87     sigaddset(&set, SIGALRM);
  88     sigaddset(&set, SIGBUS);
  89     /* SIGINT cannot be handled via signalfd, so that ^C can be used
  90      * to interrupt QEMU when it is being run under gdb.  SIGHUP and
  91      * SIGTERM are also handled asynchronously, even though it is not
  92      * strictly necessary, because they use the same handler as SIGINT.
  93      */
  94     pthread_sigmask(SIG_BLOCK, &set, NULL);
  95
  96     sigdelset(&set, SIG_IPI);
  97     sigfd = qemu_signalfd(&set);
  98     if (sigfd == -1) {
  99         fprintf(stderr, "failed to create signalfd\n");
 100         return -errno;
 101     }
 102
 103     fcntl_setfl(sigfd, O_NONBLOCK);
 104
 105     qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
 106
 107     return 0;
 108 }
 109
 110 #else /* _WIN32 */
 111
 112 static int qemu_signal_init(void)
 113 {
 114     return 0;
 115 }
 116 #endif
 117
 118 static AioContext *qemu_aio_context;
 119 static QEMUBH *qemu_notify_bh;
 120
 121 static void notify_event_cb(void *opaque)
 122 {
 123     /* No need to do anything; this bottom half is only used to
 124      * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
 125      */
 126 }
 127
 128 AioContext *qemu_get_aio_context(void)
 129 {
 130     return qemu_aio_context;
 131 }
 132
 133 void qemu_notify_event(void)
 134 {
 135     if (!qemu_aio_context) {
 136         return;
 137     }
 138     qemu_bh_schedule(qemu_notify_bh);
 139 }
 140
 141 static GArray *gpollfds;
 142
 143 int qemu_init_main_loop(Error **errp)
 144 {
 145     int ret;
 146     GSource *src;
 147     Error *local_error = NULL;
 148
 149     init_clocks();
 150
 151     ret = qemu_signal_init();
 152     if (ret) {
 153         return ret;
 154     }
 155
 156     qemu_aio_context = aio_context_new(&local_error);
 157     if (!qemu_aio_context) {
 158         error_propagate(errp, local_error);
 159         return -EMFILE;
 160     }
 161     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 162     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 163     src = aio_get_g_source(qemu_aio_context);
 164     g_source_set_name(src, "aio-context");
 165     g_source_attach(src, NULL);
 166     g_source_unref(src);
 167     src = iohandler_get_g_source();
 168     g_source_set_name(src, "io-handler");
 169     g_source_attach(src, NULL);
 170     g_source_unref(src);
 171     return 0;
 172 }
 173
 174 static int max_priority;
 175
 176 #ifndef _WIN32
 177 static int glib_pollfds_idx;
 178 static int glib_n_poll_fds;
 179
 180 static void glib_pollfds_fill(int64_t *cur_timeout)
 181 {
 182     GMainContext *context = g_main_context_default();
 183     int timeout = 0;
 184     int64_t timeout_ns;
 185     int n;
 186
 187     g_main_context_prepare(context, &max_priority);
 188
 189     glib_pollfds_idx = gpollfds->len;
 190     n = glib_n_poll_fds;
 191     do {
 192         GPollFD *pfds;
 193         glib_n_poll_fds = n;
 194         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 195         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 196         n = g_main_context_query(context, max_priority, &timeout, pfds,
 197                                  glib_n_poll_fds);
 198     } while (n != glib_n_poll_fds);
 199
 200     if (timeout < 0) {
 201         timeout_ns = -1;
 202     } else {
 203         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 204     }
 205
 206     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 207 }
 208
 209 static void glib_pollfds_poll(void)
 210 {
 211     GMainContext *context = g_main_context_default();
 212     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 213
 214     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 215         g_main_context_dispatch(context);
 216     }
 217 }
 218
 219 #define MAX_MAIN_LOOP_SPIN (1000)
 220
 221 static int os_host_main_loop_wait(int64_t timeout)
 222 {
 223     int ret;
 224     static int spin_counter;
 225
 226     glib_pollfds_fill(&timeout);
 227
 228     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 229      * the I/O thread, this can lead to starvation of the BQL such that the
 230      * VCPU threads never run.  To make sure we can detect the later case,
 231      * print a message to the screen.  If we run into this condition, create
 232      * a fake timeout in order to give the VCPU threads a chance to run.
 233      */
 234     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 235         static bool notified;
 236
 237         if (!notified && !qtest_driver()) {
 238             fprintf(stderr,
 239                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 240                     MAX_MAIN_LOOP_SPIN);
 241             notified = true;
 242         }
 243
 244         timeout = SCALE_MS;
 245     }
 246
 247     if (timeout) {
 248         spin_counter = 0;
 249         qemu_mutex_unlock_iothread();
 250     } else {
 251         spin_counter++;
 252     }
 253
 254     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 255
 256     if (timeout) {
 257         qemu_mutex_lock_iothread();
 258     }
 259
 260     glib_pollfds_poll();
 261     return ret;
 262 }
 263 #else
 264 /***********************************************************/
 265 /* Polling handling */
 266
 267 typedef struct PollingEntry {
 268     PollingFunc *func;
 269     void *opaque;
 270     struct PollingEntry *next;
 271 } PollingEntry;
 272
 273 static PollingEntry *first_polling_entry;
 274
 275 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 276 {
 277     PollingEntry **ppe, *pe;
 278     pe = g_malloc0(sizeof(PollingEntry));
 279     pe->func = func;
 280     pe->opaque = opaque;
 281     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 282     *ppe = pe;
 283     return 0;
 284 }
 285
 286 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 287 {
 288     PollingEntry **ppe, *pe;
 289     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 290         pe = *ppe;
 291         if (pe->func == func && pe->opaque == opaque) {
 292             *ppe = pe->next;
 293             g_free(pe);
 294             break;
 295         }
 296     }
 297 }
 298
 299 /***********************************************************/
 300 /* Wait objects support */
 301 typedef struct WaitObjects {
 302     int num;
 303     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 304     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 305     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 306     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 307 } WaitObjects;
 308
 309 static WaitObjects wait_objects = {0};
 310
 311 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 312 {
 313     WaitObjects *w = &wait_objects;
 314     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 315         return -1;
 316     }
 317     w->events[w->num] = handle;
 318     w->func[w->num] = func;
 319     w->opaque[w->num] = opaque;
 320     w->revents[w->num] = 0;
 321     w->num++;
 322     return 0;
 323 }
 324
 325 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 326 {
 327     int i, found;
 328     WaitObjects *w = &wait_objects;
 329
 330     found = 0;
 331     for (i = 0; i < w->num; i++) {
 332         if (w->events[i] == handle) {
 333             found = 1;
 334         }
 335         if (found) {
 336             w->events[i] = w->events[i + 1];
 337             w->func[i] = w->func[i + 1];
 338             w->opaque[i] = w->opaque[i + 1];
 339             w->revents[i] = w->revents[i + 1];
 340         }
 341     }
 342     if (found) {
 343         w->num--;
 344     }
 345 }
 346
 347 void qemu_fd_register(int fd)
 348 {
 349     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 350                    FD_READ | FD_ACCEPT | FD_CLOSE |
 351                    FD_CONNECT | FD_WRITE | FD_OOB);
 352 }
 353
 354 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 355                         fd_set *xfds)
 356 {
 357     int nfds = -1;
 358     int i;
 359
 360     for (i = 0; i < pollfds->len; i++) {
 361         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 362         int fd = pfd->fd;
 363         int events = pfd->events;
 364         if (events & G_IO_IN) {
 365             FD_SET(fd, rfds);
 366             nfds = MAX(nfds, fd);
 367         }
 368         if (events & G_IO_OUT) {
 369             FD_SET(fd, wfds);
 370             nfds = MAX(nfds, fd);
 371         }
 372         if (events & G_IO_PRI) {
 373             FD_SET(fd, xfds);
 374             nfds = MAX(nfds, fd);
 375         }
 376     }
 377     return nfds;
 378 }
 379
 380 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 381                          fd_set *wfds, fd_set *xfds)
 382 {
 383     int i;
 384
 385     for (i = 0; i < pollfds->len; i++) {
 386         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 387         int fd = pfd->fd;
 388         int revents = 0;
 389
 390         if (FD_ISSET(fd, rfds)) {
 391             revents |= G_IO_IN;
 392         }
 393         if (FD_ISSET(fd, wfds)) {
 394             revents |= G_IO_OUT;
 395         }
 396         if (FD_ISSET(fd, xfds)) {
 397             revents |= G_IO_PRI;
 398         }
 399         pfd->revents = revents & pfd->events;
 400     }
 401 }
 402
 403 static int os_host_main_loop_wait(int64_t timeout)
 404 {
 405     GMainContext *context = g_main_context_default();
 406     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 407     int select_ret = 0;
 408     int g_poll_ret, ret, i, n_poll_fds;
 409     PollingEntry *pe;
 410     WaitObjects *w = &wait_objects;
 411     gint poll_timeout;
 412     int64_t poll_timeout_ns;
 413     static struct timeval tv0;
 414     fd_set rfds, wfds, xfds;
 415     int nfds;
 416
 417     /* XXX: need to suppress polling by better using win32 events */
 418     ret = 0;
 419     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 420         ret |= pe->func(pe->opaque);
 421     }
 422     if (ret != 0) {
 423         return ret;
 424     }
 425
 426     FD_ZERO(&rfds);
 427     FD_ZERO(&wfds);
 428     FD_ZERO(&xfds);
 429     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 430     if (nfds >= 0) {
 431         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 432         if (select_ret != 0) {
 433             timeout = 0;
 434         }
 435         if (select_ret > 0) {
 436             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 437         }
 438     }
 439
 440     g_main_context_prepare(context, &max_priority);
 441     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 442                                       poll_fds, ARRAY_SIZE(poll_fds));
 443     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 444
 445     for (i = 0; i < w->num; i++) {
 446         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 447         poll_fds[n_poll_fds + i].events = G_IO_IN;
 448     }
 449
 450     if (poll_timeout < 0) {
 451         poll_timeout_ns = -1;
 452     } else {
 453         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 454     }
 455
 456     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 457
 458     qemu_mutex_unlock_iothread();
 459     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 460
 461     qemu_mutex_lock_iothread();
 462     if (g_poll_ret > 0) {
 463         for (i = 0; i < w->num; i++) {
 464             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 465         }
 466         for (i = 0; i < w->num; i++) {
 467             if (w->revents[i] && w->func[i]) {
 468                 w->func[i](w->opaque[i]);
 469             }
 470         }
 471     }
 472
 473     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 474         g_main_context_dispatch(context);
 475     }
 476
 477     return select_ret || g_poll_ret;
 478 }
 479 #endif
 480
 481 int main_loop_wait(int nonblocking)
 482 {
 483     int ret;
 484     uint32_t timeout = UINT32_MAX;
 485     int64_t timeout_ns;
 486
 487     if (nonblocking) {
 488         timeout = 0;
 489     }
 490
 491     /* poll any events */
 492     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 493     /* XXX: separate device handlers from system ones */
 494 #ifdef CONFIG_SLIRP
 495     slirp_pollfds_fill(gpollfds, &timeout);
 496 #endif
 497
 498     if (timeout == UINT32_MAX) {
 499         timeout_ns = -1;
 500     } else {
 501         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 502     }
 503
 504     timeout_ns = qemu_soonest_timeout(timeout_ns,
 505                                       timerlistgroup_deadline_ns(
 506                                           &main_loop_tlg));
 507
 508     ret = os_host_main_loop_wait(timeout_ns);
 509 #ifdef CONFIG_SLIRP
 510     slirp_pollfds_poll(gpollfds, (ret < 0));
 511 #endif
 512
 513     /* CPU thread can infinitely wait for event after
 514        missing the warp */
 515     qemu_start_warp_timer();
 516     qemu_clock_run_all_timers();
 517
 518     return ret;
 519 }
 520
 521 /* Functions to operate on the main QEMU AioContext.  */
 522
 523 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 524 {
 525     return aio_bh_new(qemu_aio_context, cb, opaque);
 526 }