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