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     src = iohandler_get_g_source();
 165     g_source_attach(src, NULL);
 166     g_source_unref(src);
 167     return 0;
 168 }
 169
 170 static int max_priority;
 171
 172 #ifndef _WIN32
 173 static int glib_pollfds_idx;
 174 static int glib_n_poll_fds;
 175
 176 static void glib_pollfds_fill(int64_t *cur_timeout)
 177 {
 178     GMainContext *context = g_main_context_default();
 179     int timeout = 0;
 180     int64_t timeout_ns;
 181     int n;
 182
 183     g_main_context_prepare(context, &max_priority);
 184
 185     glib_pollfds_idx = gpollfds->len;
 186     n = glib_n_poll_fds;
 187     do {
 188         GPollFD *pfds;
 189         glib_n_poll_fds = n;
 190         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 191         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 192         n = g_main_context_query(context, max_priority, &timeout, pfds,
 193                                  glib_n_poll_fds);
 194     } while (n != glib_n_poll_fds);
 195
 196     if (timeout < 0) {
 197         timeout_ns = -1;
 198     } else {
 199         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 200     }
 201
 202     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 203 }
 204
 205 static void glib_pollfds_poll(void)
 206 {
 207     GMainContext *context = g_main_context_default();
 208     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 209
 210     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 211         g_main_context_dispatch(context);
 212     }
 213 }
 214
 215 #define MAX_MAIN_LOOP_SPIN (1000)
 216
 217 static int os_host_main_loop_wait(int64_t timeout)
 218 {
 219     int ret;
 220     static int spin_counter;
 221
 222     glib_pollfds_fill(&timeout);
 223
 224     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 225      * the I/O thread, this can lead to starvation of the BQL such that the
 226      * VCPU threads never run.  To make sure we can detect the later case,
 227      * print a message to the screen.  If we run into this condition, create
 228      * a fake timeout in order to give the VCPU threads a chance to run.
 229      */
 230     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 231         static bool notified;
 232
 233         if (!notified && !qtest_enabled()) {
 234             fprintf(stderr,
 235                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 236                     MAX_MAIN_LOOP_SPIN);
 237             notified = true;
 238         }
 239
 240         timeout = SCALE_MS;
 241     }
 242
 243     if (timeout) {
 244         spin_counter = 0;
 245         qemu_mutex_unlock_iothread();
 246     } else {
 247         spin_counter++;
 248     }
 249
 250     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 251
 252     if (timeout) {
 253         qemu_mutex_lock_iothread();
 254     }
 255
 256     glib_pollfds_poll();
 257     return ret;
 258 }
 259 #else
 260 /***********************************************************/
 261 /* Polling handling */
 262
 263 typedef struct PollingEntry {
 264     PollingFunc *func;
 265     void *opaque;
 266     struct PollingEntry *next;
 267 } PollingEntry;
 268
 269 static PollingEntry *first_polling_entry;
 270
 271 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 272 {
 273     PollingEntry **ppe, *pe;
 274     pe = g_malloc0(sizeof(PollingEntry));
 275     pe->func = func;
 276     pe->opaque = opaque;
 277     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 278     *ppe = pe;
 279     return 0;
 280 }
 281
 282 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 283 {
 284     PollingEntry **ppe, *pe;
 285     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 286         pe = *ppe;
 287         if (pe->func == func && pe->opaque == opaque) {
 288             *ppe = pe->next;
 289             g_free(pe);
 290             break;
 291         }
 292     }
 293 }
 294
 295 /***********************************************************/
 296 /* Wait objects support */
 297 typedef struct WaitObjects {
 298     int num;
 299     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 300     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 301     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 302     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 303 } WaitObjects;
 304
 305 static WaitObjects wait_objects = {0};
 306
 307 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 308 {
 309     WaitObjects *w = &wait_objects;
 310     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 311         return -1;
 312     }
 313     w->events[w->num] = handle;
 314     w->func[w->num] = func;
 315     w->opaque[w->num] = opaque;
 316     w->revents[w->num] = 0;
 317     w->num++;
 318     return 0;
 319 }
 320
 321 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 322 {
 323     int i, found;
 324     WaitObjects *w = &wait_objects;
 325
 326     found = 0;
 327     for (i = 0; i < w->num; i++) {
 328         if (w->events[i] == handle) {
 329             found = 1;
 330         }
 331         if (found) {
 332             w->events[i] = w->events[i + 1];
 333             w->func[i] = w->func[i + 1];
 334             w->opaque[i] = w->opaque[i + 1];
 335             w->revents[i] = w->revents[i + 1];
 336         }
 337     }
 338     if (found) {
 339         w->num--;
 340     }
 341 }
 342
 343 void qemu_fd_register(int fd)
 344 {
 345     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 346                    FD_READ | FD_ACCEPT | FD_CLOSE |
 347                    FD_CONNECT | FD_WRITE | FD_OOB);
 348 }
 349
 350 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 351                         fd_set *xfds)
 352 {
 353     int nfds = -1;
 354     int i;
 355
 356     for (i = 0; i < pollfds->len; i++) {
 357         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 358         int fd = pfd->fd;
 359         int events = pfd->events;
 360         if (events & G_IO_IN) {
 361             FD_SET(fd, rfds);
 362             nfds = MAX(nfds, fd);
 363         }
 364         if (events & G_IO_OUT) {
 365             FD_SET(fd, wfds);
 366             nfds = MAX(nfds, fd);
 367         }
 368         if (events & G_IO_PRI) {
 369             FD_SET(fd, xfds);
 370             nfds = MAX(nfds, fd);
 371         }
 372     }
 373     return nfds;
 374 }
 375
 376 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 377                          fd_set *wfds, fd_set *xfds)
 378 {
 379     int i;
 380
 381     for (i = 0; i < pollfds->len; i++) {
 382         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 383         int fd = pfd->fd;
 384         int revents = 0;
 385
 386         if (FD_ISSET(fd, rfds)) {
 387             revents |= G_IO_IN;
 388         }
 389         if (FD_ISSET(fd, wfds)) {
 390             revents |= G_IO_OUT;
 391         }
 392         if (FD_ISSET(fd, xfds)) {
 393             revents |= G_IO_PRI;
 394         }
 395         pfd->revents = revents & pfd->events;
 396     }
 397 }
 398
 399 static int os_host_main_loop_wait(int64_t timeout)
 400 {
 401     GMainContext *context = g_main_context_default();
 402     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 403     int select_ret = 0;
 404     int g_poll_ret, ret, i, n_poll_fds;
 405     PollingEntry *pe;
 406     WaitObjects *w = &wait_objects;
 407     gint poll_timeout;
 408     int64_t poll_timeout_ns;
 409     static struct timeval tv0;
 410     fd_set rfds, wfds, xfds;
 411     int nfds;
 412
 413     /* XXX: need to suppress polling by better using win32 events */
 414     ret = 0;
 415     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 416         ret |= pe->func(pe->opaque);
 417     }
 418     if (ret != 0) {
 419         return ret;
 420     }
 421
 422     FD_ZERO(&rfds);
 423     FD_ZERO(&wfds);
 424     FD_ZERO(&xfds);
 425     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 426     if (nfds >= 0) {
 427         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 428         if (select_ret != 0) {
 429             timeout = 0;
 430         }
 431         if (select_ret > 0) {
 432             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 433         }
 434     }
 435
 436     g_main_context_prepare(context, &max_priority);
 437     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 438                                       poll_fds, ARRAY_SIZE(poll_fds));
 439     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 440
 441     for (i = 0; i < w->num; i++) {
 442         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 443         poll_fds[n_poll_fds + i].events = G_IO_IN;
 444     }
 445
 446     if (poll_timeout < 0) {
 447         poll_timeout_ns = -1;
 448     } else {
 449         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 450     }
 451
 452     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 453
 454     qemu_mutex_unlock_iothread();
 455     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 456
 457     qemu_mutex_lock_iothread();
 458     if (g_poll_ret > 0) {
 459         for (i = 0; i < w->num; i++) {
 460             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 461         }
 462         for (i = 0; i < w->num; i++) {
 463             if (w->revents[i] && w->func[i]) {
 464                 w->func[i](w->opaque[i]);
 465             }
 466         }
 467     }
 468
 469     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 470         g_main_context_dispatch(context);
 471     }
 472
 473     return select_ret || g_poll_ret;
 474 }
 475 #endif
 476
 477 int main_loop_wait(int nonblocking)
 478 {
 479     int ret;
 480     uint32_t timeout = UINT32_MAX;
 481     int64_t timeout_ns;
 482
 483     if (nonblocking) {
 484         timeout = 0;
 485     }
 486
 487     /* poll any events */
 488     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 489     /* XXX: separate device handlers from system ones */
 490 #ifdef CONFIG_SLIRP
 491     slirp_pollfds_fill(gpollfds, &timeout);
 492 #endif
 493
 494     if (timeout == UINT32_MAX) {
 495         timeout_ns = -1;
 496     } else {
 497         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 498     }
 499
 500     timeout_ns = qemu_soonest_timeout(timeout_ns,
 501                                       timerlistgroup_deadline_ns(
 502                                           &main_loop_tlg));
 503
 504     ret = os_host_main_loop_wait(timeout_ns);
 505 #ifdef CONFIG_SLIRP
 506     slirp_pollfds_poll(gpollfds, (ret < 0));
 507 #endif
 508
 509     /* CPU thread can infinitely wait for event after
 510        missing the warp */
 511     qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
 512     qemu_clock_run_all_timers();
 513
 514     return ret;
 515 }
 516
 517 /* Functions to operate on the main QEMU AioContext.  */
 518
 519 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 520 {
 521     return aio_bh_new(qemu_aio_context, cb, opaque);
 522 }