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