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 "sysemu/qtest.h"
  30 #include "sysemu/cpus.h"
  31 #include "sysemu/replay.h"
  32 #include "qemu/main-loop.h"
  33 #include "block/aio.h"
  34 #include "qemu/error-report.h"
  35
  36 #ifndef _WIN32
  37
  38 /* If we have signalfd, we mask out the signals we want to handle and then
  39  * use signalfd to listen for them.  We rely on whatever the current signal
  40  * handler is to dispatch the signals when we receive them.
  41  */
  42 static void sigfd_handler(void *opaque)
  43 {
  44     int fd = (intptr_t)opaque;
  45     struct qemu_signalfd_siginfo info;
  46     struct sigaction action;
  47     ssize_t len;
  48
  49     while (1) {
  50         do {
  51             len = read(fd, &info, sizeof(info));
  52         } while (len == -1 && errno == EINTR);
  53
  54         if (len == -1 && errno == EAGAIN) {
  55             break;
  56         }
  57
  58         if (len != sizeof(info)) {
  59             printf("read from sigfd returned %zd: %m\n", len);
  60             return;
  61         }
  62
  63         sigaction(info.ssi_signo, NULL, &action);
  64         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  65             sigaction_invoke(&action, &info);
  66         } else if (action.sa_handler) {
  67             action.sa_handler(info.ssi_signo);
  68         }
  69     }
  70 }
  71
  72 static int qemu_signal_init(Error **errp)
  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         error_setg_errno(errp, errno, "failed to create signalfd");
  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(Error **errp)
 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(qemu_timer_notify_cb);
 148
 149     ret = qemu_signal_init(errp);
 150     if (ret) {
 151         return ret;
 152     }
 153
 154     qemu_aio_context = aio_context_new(&local_error);
 155     if (!qemu_aio_context) {
 156         error_propagate(errp, local_error);
 157         return -EMFILE;
 158     }
 159     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 160     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 161     src = aio_get_g_source(qemu_aio_context);
 162     g_source_set_name(src, "aio-context");
 163     g_source_attach(src, NULL);
 164     g_source_unref(src);
 165     src = iohandler_get_g_source();
 166     g_source_set_name(src, "io-handler");
 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     GMainContext *context = g_main_context_default();
 222     int ret;
 223
 224     g_main_context_acquire(context);
 225
 226     glib_pollfds_fill(&timeout);
 227
 228     qemu_mutex_unlock_iothread();
 229     replay_mutex_unlock();
 230
 231     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 232
 233     replay_mutex_lock();
 234     qemu_mutex_lock_iothread();
 235
 236     glib_pollfds_poll();
 237
 238     g_main_context_release(context);
 239
 240     return ret;
 241 }
 242 #else
 243 /***********************************************************/
 244 /* Polling handling */
 245
 246 typedef struct PollingEntry {
 247     PollingFunc *func;
 248     void *opaque;
 249     struct PollingEntry *next;
 250 } PollingEntry;
 251
 252 static PollingEntry *first_polling_entry;
 253
 254 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 255 {
 256     PollingEntry **ppe, *pe;
 257     pe = g_malloc0(sizeof(PollingEntry));
 258     pe->func = func;
 259     pe->opaque = opaque;
 260     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 261     *ppe = pe;
 262     return 0;
 263 }
 264
 265 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 266 {
 267     PollingEntry **ppe, *pe;
 268     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 269         pe = *ppe;
 270         if (pe->func == func && pe->opaque == opaque) {
 271             *ppe = pe->next;
 272             g_free(pe);
 273             break;
 274         }
 275     }
 276 }
 277
 278 /***********************************************************/
 279 /* Wait objects support */
 280 typedef struct WaitObjects {
 281     int num;
 282     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 283     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 284     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 285     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 286 } WaitObjects;
 287
 288 static WaitObjects wait_objects = {0};
 289
 290 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 291 {
 292     WaitObjects *w = &wait_objects;
 293     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 294         return -1;
 295     }
 296     w->events[w->num] = handle;
 297     w->func[w->num] = func;
 298     w->opaque[w->num] = opaque;
 299     w->revents[w->num] = 0;
 300     w->num++;
 301     return 0;
 302 }
 303
 304 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 305 {
 306     int i, found;
 307     WaitObjects *w = &wait_objects;
 308
 309     found = 0;
 310     for (i = 0; i < w->num; i++) {
 311         if (w->events[i] == handle) {
 312             found = 1;
 313         }
 314         if (found) {
 315             w->events[i] = w->events[i + 1];
 316             w->func[i] = w->func[i + 1];
 317             w->opaque[i] = w->opaque[i + 1];
 318             w->revents[i] = w->revents[i + 1];
 319         }
 320     }
 321     if (found) {
 322         w->num--;
 323     }
 324 }
 325
 326 void qemu_fd_register(int fd)
 327 {
 328     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 329                    FD_READ | FD_ACCEPT | FD_CLOSE |
 330                    FD_CONNECT | FD_WRITE | FD_OOB);
 331 }
 332
 333 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 334                         fd_set *xfds)
 335 {
 336     int nfds = -1;
 337     int i;
 338
 339     for (i = 0; i < pollfds->len; i++) {
 340         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 341         int fd = pfd->fd;
 342         int events = pfd->events;
 343         if (events & G_IO_IN) {
 344             FD_SET(fd, rfds);
 345             nfds = MAX(nfds, fd);
 346         }
 347         if (events & G_IO_OUT) {
 348             FD_SET(fd, wfds);
 349             nfds = MAX(nfds, fd);
 350         }
 351         if (events & G_IO_PRI) {
 352             FD_SET(fd, xfds);
 353             nfds = MAX(nfds, fd);
 354         }
 355     }
 356     return nfds;
 357 }
 358
 359 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 360                          fd_set *wfds, fd_set *xfds)
 361 {
 362     int i;
 363
 364     for (i = 0; i < pollfds->len; i++) {
 365         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 366         int fd = pfd->fd;
 367         int revents = 0;
 368
 369         if (FD_ISSET(fd, rfds)) {
 370             revents |= G_IO_IN;
 371         }
 372         if (FD_ISSET(fd, wfds)) {
 373             revents |= G_IO_OUT;
 374         }
 375         if (FD_ISSET(fd, xfds)) {
 376             revents |= G_IO_PRI;
 377         }
 378         pfd->revents = revents & pfd->events;
 379     }
 380 }
 381
 382 static int os_host_main_loop_wait(int64_t timeout)
 383 {
 384     GMainContext *context = g_main_context_default();
 385     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 386     int select_ret = 0;
 387     int g_poll_ret, ret, i, n_poll_fds;
 388     PollingEntry *pe;
 389     WaitObjects *w = &wait_objects;
 390     gint poll_timeout;
 391     int64_t poll_timeout_ns;
 392     static struct timeval tv0;
 393     fd_set rfds, wfds, xfds;
 394     int nfds;
 395
 396     g_main_context_acquire(context);
 397
 398     /* XXX: need to suppress polling by better using win32 events */
 399     ret = 0;
 400     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 401         ret |= pe->func(pe->opaque);
 402     }
 403     if (ret != 0) {
 404         g_main_context_release(context);
 405         return ret;
 406     }
 407
 408     FD_ZERO(&rfds);
 409     FD_ZERO(&wfds);
 410     FD_ZERO(&xfds);
 411     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 412     if (nfds >= 0) {
 413         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 414         if (select_ret != 0) {
 415             timeout = 0;
 416         }
 417         if (select_ret > 0) {
 418             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 419         }
 420     }
 421
 422     g_main_context_prepare(context, &max_priority);
 423     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 424                                       poll_fds, ARRAY_SIZE(poll_fds));
 425     g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
 426
 427     for (i = 0; i < w->num; i++) {
 428         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 429         poll_fds[n_poll_fds + i].events = G_IO_IN;
 430     }
 431
 432     if (poll_timeout < 0) {
 433         poll_timeout_ns = -1;
 434     } else {
 435         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 436     }
 437
 438     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 439
 440     qemu_mutex_unlock_iothread();
 441
 442     replay_mutex_unlock();
 443
 444     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 445
 446     replay_mutex_lock();
 447
 448     qemu_mutex_lock_iothread();
 449     if (g_poll_ret > 0) {
 450         for (i = 0; i < w->num; i++) {
 451             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 452         }
 453         for (i = 0; i < w->num; i++) {
 454             if (w->revents[i] && w->func[i]) {
 455                 w->func[i](w->opaque[i]);
 456             }
 457         }
 458     }
 459
 460     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 461         g_main_context_dispatch(context);
 462     }
 463
 464     g_main_context_release(context);
 465
 466     return select_ret || g_poll_ret;
 467 }
 468 #endif
 469
 470 static NotifierList main_loop_poll_notifiers =
 471     NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
 472
 473 void main_loop_poll_add_notifier(Notifier *notify)
 474 {
 475     notifier_list_add(&main_loop_poll_notifiers, notify);
 476 }
 477
 478 void main_loop_poll_remove_notifier(Notifier *notify)
 479 {
 480     notifier_remove(notify);
 481 }
 482
 483 void main_loop_wait(int nonblocking)
 484 {
 485     MainLoopPoll mlpoll = {
 486         .state = MAIN_LOOP_POLL_FILL,
 487         .timeout = UINT32_MAX,
 488         .pollfds = gpollfds,
 489     };
 490     int ret;
 491     int64_t timeout_ns;
 492
 493     if (nonblocking) {
 494         mlpoll.timeout = 0;
 495     }
 496
 497     /* poll any events */
 498     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 499     /* XXX: separate device handlers from system ones */
 500     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 501
 502     if (mlpoll.timeout == UINT32_MAX) {
 503         timeout_ns = -1;
 504     } else {
 505         timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
 506     }
 507
 508     timeout_ns = qemu_soonest_timeout(timeout_ns,
 509                                       timerlistgroup_deadline_ns(
 510                                           &main_loop_tlg));
 511
 512     ret = os_host_main_loop_wait(timeout_ns);
 513     mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
 514     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 515
 516     /* CPU thread can infinitely wait for event after
 517        missing the warp */
 518     qemu_start_warp_timer();
 519     qemu_clock_run_all_timers();
 520 }
 521
 522 /* Functions to operate on the main QEMU AioContext.  */
 523
 524 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 525 {
 526     return aio_bh_new(qemu_aio_context, cb, opaque);
 527 }