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     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 158     if (!qemu_aio_context) {
 159         error_propagate(errp, local_error);
 160         return -EMFILE;
 161     }
 162     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 163     src = aio_get_g_source(qemu_aio_context);
 164     g_source_attach(src, NULL);
 165     g_source_unref(src);
 166     src = iohandler_get_g_source();
 167     g_source_attach(src, NULL);
 168     g_source_unref(src);
 169     return 0;
 170 }
 171
 172 static int max_priority;
 173
 174 #ifndef _WIN32
 175 static int glib_pollfds_idx;
 176 static int glib_n_poll_fds;
 177
 178 static void glib_pollfds_fill(int64_t *cur_timeout)
 179 {
 180     GMainContext *context = g_main_context_default();
 181     int timeout = 0;
 182     int64_t timeout_ns;
 183     int n;
 184
 185     g_main_context_prepare(context, &max_priority);
 186
 187     glib_pollfds_idx = gpollfds->len;
 188     n = glib_n_poll_fds;
 189     do {
 190         GPollFD *pfds;
 191         glib_n_poll_fds = n;
 192         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 193         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 194         n = g_main_context_query(context, max_priority, &timeout, pfds,
 195                                  glib_n_poll_fds);
 196     } while (n != glib_n_poll_fds);
 197
 198     if (timeout < 0) {
 199         timeout_ns = -1;
 200     } else {
 201         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 202     }
 203
 204     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 205 }
 206
 207 static void glib_pollfds_poll(void)
 208 {
 209     GMainContext *context = g_main_context_default();
 210     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 211
 212     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 213         g_main_context_dispatch(context);
 214     }
 215 }
 216
 217 #define MAX_MAIN_LOOP_SPIN (1000)
 218
 219 static int os_host_main_loop_wait(int64_t timeout)
 220 {
 221     int ret;
 222     static int spin_counter;
 223
 224     glib_pollfds_fill(&timeout);
 225
 226     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 227      * the I/O thread, this can lead to starvation of the BQL such that the
 228      * VCPU threads never run.  To make sure we can detect the later case,
 229      * print a message to the screen.  If we run into this condition, create
 230      * a fake timeout in order to give the VCPU threads a chance to run.
 231      */
 232     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 233         static bool notified;
 234
 235         if (!notified && !qtest_driver()) {
 236             fprintf(stderr,
 237                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 238                     MAX_MAIN_LOOP_SPIN);
 239             notified = true;
 240         }
 241
 242         timeout = SCALE_MS;
 243     }
 244
 245     if (timeout) {
 246         spin_counter = 0;
 247         qemu_mutex_unlock_iothread();
 248     } else {
 249         spin_counter++;
 250     }
 251
 252     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 253
 254     if (timeout) {
 255         qemu_mutex_lock_iothread();
 256     }
 257
 258     glib_pollfds_poll();
 259     return ret;
 260 }
 261 #else
 262 /***********************************************************/
 263 /* Polling handling */
 264
 265 typedef struct PollingEntry {
 266     PollingFunc *func;
 267     void *opaque;
 268     struct PollingEntry *next;
 269 } PollingEntry;
 270
 271 static PollingEntry *first_polling_entry;
 272
 273 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 274 {
 275     PollingEntry **ppe, *pe;
 276     pe = g_malloc0(sizeof(PollingEntry));
 277     pe->func = func;
 278     pe->opaque = opaque;
 279     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 280     *ppe = pe;
 281     return 0;
 282 }
 283
 284 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 285 {
 286     PollingEntry **ppe, *pe;
 287     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 288         pe = *ppe;
 289         if (pe->func == func && pe->opaque == opaque) {
 290             *ppe = pe->next;
 291             g_free(pe);
 292             break;
 293         }
 294     }
 295 }
 296
 297 /***********************************************************/
 298 /* Wait objects support */
 299 typedef struct WaitObjects {
 300     int num;
 301     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 302     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 303     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 304     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 305 } WaitObjects;
 306
 307 static WaitObjects wait_objects = {0};
 308
 309 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 310 {
 311     WaitObjects *w = &wait_objects;
 312     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 313         return -1;
 314     }
 315     w->events[w->num] = handle;
 316     w->func[w->num] = func;
 317     w->opaque[w->num] = opaque;
 318     w->revents[w->num] = 0;
 319     w->num++;
 320     return 0;
 321 }
 322
 323 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 324 {
 325     int i, found;
 326     WaitObjects *w = &wait_objects;
 327
 328     found = 0;
 329     for (i = 0; i < w->num; i++) {
 330         if (w->events[i] == handle) {
 331             found = 1;
 332         }
 333         if (found) {
 334             w->events[i] = w->events[i + 1];
 335             w->func[i] = w->func[i + 1];
 336             w->opaque[i] = w->opaque[i + 1];
 337             w->revents[i] = w->revents[i + 1];
 338         }
 339     }
 340     if (found) {
 341         w->num--;
 342     }
 343 }
 344
 345 void qemu_fd_register(int fd)
 346 {
 347     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 348                    FD_READ | FD_ACCEPT | FD_CLOSE |
 349                    FD_CONNECT | FD_WRITE | FD_OOB);
 350 }
 351
 352 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 353                         fd_set *xfds)
 354 {
 355     int nfds = -1;
 356     int i;
 357
 358     for (i = 0; i < pollfds->len; i++) {
 359         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 360         int fd = pfd->fd;
 361         int events = pfd->events;
 362         if (events & G_IO_IN) {
 363             FD_SET(fd, rfds);
 364             nfds = MAX(nfds, fd);
 365         }
 366         if (events & G_IO_OUT) {
 367             FD_SET(fd, wfds);
 368             nfds = MAX(nfds, fd);
 369         }
 370         if (events & G_IO_PRI) {
 371             FD_SET(fd, xfds);
 372             nfds = MAX(nfds, fd);
 373         }
 374     }
 375     return nfds;
 376 }
 377
 378 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 379                          fd_set *wfds, fd_set *xfds)
 380 {
 381     int i;
 382
 383     for (i = 0; i < pollfds->len; i++) {
 384         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 385         int fd = pfd->fd;
 386         int revents = 0;
 387
 388         if (FD_ISSET(fd, rfds)) {
 389             revents |= G_IO_IN;
 390         }
 391         if (FD_ISSET(fd, wfds)) {
 392             revents |= G_IO_OUT;
 393         }
 394         if (FD_ISSET(fd, xfds)) {
 395             revents |= G_IO_PRI;
 396         }
 397         pfd->revents = revents & pfd->events;
 398     }
 399 }
 400
 401 static int os_host_main_loop_wait(int64_t timeout)
 402 {
 403     GMainContext *context = g_main_context_default();
 404     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 405     int select_ret = 0;
 406     int g_poll_ret, ret, i, n_poll_fds;
 407     PollingEntry *pe;
 408     WaitObjects *w = &wait_objects;
 409     gint poll_timeout;
 410     int64_t poll_timeout_ns;
 411     static struct timeval tv0;
 412     fd_set rfds, wfds, xfds;
 413     int nfds;
 414
 415     /* XXX: need to suppress polling by better using win32 events */
 416     ret = 0;
 417     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 418         ret |= pe->func(pe->opaque);
 419     }
 420     if (ret != 0) {
 421         return ret;
 422     }
 423
 424     FD_ZERO(&rfds);
 425     FD_ZERO(&wfds);
 426     FD_ZERO(&xfds);
 427     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 428     if (nfds >= 0) {
 429         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 430         if (select_ret != 0) {
 431             timeout = 0;
 432         }
 433         if (select_ret > 0) {
 434             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 435         }
 436     }
 437
 438     g_main_context_prepare(context, &max_priority);
 439     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 440                                       poll_fds, ARRAY_SIZE(poll_fds));
 441     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 442
 443     for (i = 0; i < w->num; i++) {
 444         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 445         poll_fds[n_poll_fds + i].events = G_IO_IN;
 446     }
 447
 448     if (poll_timeout < 0) {
 449         poll_timeout_ns = -1;
 450     } else {
 451         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 452     }
 453
 454     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 455
 456     qemu_mutex_unlock_iothread();
 457     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 458
 459     qemu_mutex_lock_iothread();
 460     if (g_poll_ret > 0) {
 461         for (i = 0; i < w->num; i++) {
 462             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 463         }
 464         for (i = 0; i < w->num; i++) {
 465             if (w->revents[i] && w->func[i]) {
 466                 w->func[i](w->opaque[i]);
 467             }
 468         }
 469     }
 470
 471     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 472         g_main_context_dispatch(context);
 473     }
 474
 475     return select_ret || g_poll_ret;
 476 }
 477 #endif
 478
 479 int main_loop_wait(int nonblocking)
 480 {
 481     int ret;
 482     uint32_t timeout = UINT32_MAX;
 483     int64_t timeout_ns;
 484
 485     if (nonblocking) {
 486         timeout = 0;
 487     }
 488
 489     /* poll any events */
 490     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 491     /* XXX: separate device handlers from system ones */
 492 #ifdef CONFIG_SLIRP
 493     slirp_pollfds_fill(gpollfds, &timeout);
 494 #endif
 495
 496     if (timeout == UINT32_MAX) {
 497         timeout_ns = -1;
 498     } else {
 499         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 500     }
 501
 502     timeout_ns = qemu_soonest_timeout(timeout_ns,
 503                                       timerlistgroup_deadline_ns(
 504                                           &main_loop_tlg));
 505
 506     ret = os_host_main_loop_wait(timeout_ns);
 507 #ifdef CONFIG_SLIRP
 508     slirp_pollfds_poll(gpollfds, (ret < 0));
 509 #endif
 510
 511     /* CPU thread can infinitely wait for event after
 512        missing the warp */
 513     qemu_start_warp_timer();
 514     qemu_clock_run_all_timers();
 515
 516     return ret;
 517 }
 518
 519 /* Functions to operate on the main QEMU AioContext.  */
 520
 521 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 522 {
 523     return aio_bh_new(qemu_aio_context, cb, opaque);
 524 }