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 "sysemu/cpus.h"
  32 #include "slirp/libslirp.h"
  33 #include "qemu/main-loop.h"
  34 #include "block/aio.h"
  35 #include "qemu/error-report.h"
  36
  37 #ifndef _WIN32
  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             sigaction_invoke(&action, &info);
  67         } else if (action.sa_handler) {
  68             action.sa_handler(info.ssi_signo);
  69         }
  70     }
  71 }
  72
  73 static int qemu_signal_init(void)
  74 {
  75     int sigfd;
  76     sigset_t set;
  77
  78     /*
  79      * SIG_IPI must be blocked in the main thread and must not be caught
  80      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  81      * not catch it reliably.
  82      */
  83     sigemptyset(&set);
  84     sigaddset(&set, SIG_IPI);
  85     sigaddset(&set, SIGIO);
  86     sigaddset(&set, SIGALRM);
  87     sigaddset(&set, SIGBUS);
  88     /* SIGINT cannot be handled via signalfd, so that ^C can be used
  89      * to interrupt QEMU when it is being run under gdb.  SIGHUP and
  90      * SIGTERM are also handled asynchronously, even though it is not
  91      * strictly necessary, because they use the same handler as SIGINT.
  92      */
  93     pthread_sigmask(SIG_BLOCK, &set, NULL);
  94
  95     sigdelset(&set, SIG_IPI);
  96     sigfd = qemu_signalfd(&set);
  97     if (sigfd == -1) {
  98         fprintf(stderr, "failed to create signalfd\n");
  99         return -errno;
 100     }
 101
 102     fcntl_setfl(sigfd, O_NONBLOCK);
 103
 104     qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
 105
 106     return 0;
 107 }
 108
 109 #else /* _WIN32 */
 110
 111 static int qemu_signal_init(void)
 112 {
 113     return 0;
 114 }
 115 #endif
 116
 117 static AioContext *qemu_aio_context;
 118 static QEMUBH *qemu_notify_bh;
 119
 120 static void notify_event_cb(void *opaque)
 121 {
 122     /* No need to do anything; this bottom half is only used to
 123      * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
 124      */
 125 }
 126
 127 AioContext *qemu_get_aio_context(void)
 128 {
 129     return qemu_aio_context;
 130 }
 131
 132 void qemu_notify_event(void)
 133 {
 134     if (!qemu_aio_context) {
 135         return;
 136     }
 137     qemu_bh_schedule(qemu_notify_bh);
 138 }
 139
 140 static GArray *gpollfds;
 141
 142 int qemu_init_main_loop(Error **errp)
 143 {
 144     int ret;
 145     GSource *src;
 146     Error *local_error = NULL;
 147
 148     init_clocks(qemu_timer_notify_cb);
 149
 150     ret = qemu_signal_init();
 151     if (ret) {
 152         return ret;
 153     }
 154
 155     qemu_aio_context = aio_context_new(&local_error);
 156     if (!qemu_aio_context) {
 157         error_propagate(errp, local_error);
 158         return -EMFILE;
 159     }
 160     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 161     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 162     src = aio_get_g_source(qemu_aio_context);
 163     g_source_set_name(src, "aio-context");
 164     g_source_attach(src, NULL);
 165     g_source_unref(src);
 166     src = iohandler_get_g_source();
 167     g_source_set_name(src, "io-handler");
 168     g_source_attach(src, NULL);
 169     g_source_unref(src);
 170     return 0;
 171 }
 172
 173 static int max_priority;
 174
 175 #ifndef _WIN32
 176 static int glib_pollfds_idx;
 177 static int glib_n_poll_fds;
 178
 179 static void glib_pollfds_fill(int64_t *cur_timeout)
 180 {
 181     GMainContext *context = g_main_context_default();
 182     int timeout = 0;
 183     int64_t timeout_ns;
 184     int n;
 185
 186     g_main_context_prepare(context, &max_priority);
 187
 188     glib_pollfds_idx = gpollfds->len;
 189     n = glib_n_poll_fds;
 190     do {
 191         GPollFD *pfds;
 192         glib_n_poll_fds = n;
 193         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 194         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 195         n = g_main_context_query(context, max_priority, &timeout, pfds,
 196                                  glib_n_poll_fds);
 197     } while (n != glib_n_poll_fds);
 198
 199     if (timeout < 0) {
 200         timeout_ns = -1;
 201     } else {
 202         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 203     }
 204
 205     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 206 }
 207
 208 static void glib_pollfds_poll(void)
 209 {
 210     GMainContext *context = g_main_context_default();
 211     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 212
 213     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 214         g_main_context_dispatch(context);
 215     }
 216 }
 217
 218 #define MAX_MAIN_LOOP_SPIN (1000)
 219
 220 static int os_host_main_loop_wait(int64_t timeout)
 221 {
 222     GMainContext *context = g_main_context_default();
 223     int ret;
 224     static int spin_counter;
 225
 226     g_main_context_acquire(context);
 227
 228     glib_pollfds_fill(&timeout);
 229
 230     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 231      * the I/O thread, this can lead to starvation of the BQL such that the
 232      * VCPU threads never run.  To make sure we can detect the later case,
 233      * print a message to the screen.  If we run into this condition, create
 234      * a fake timeout in order to give the VCPU threads a chance to run.
 235      */
 236     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 237         static bool notified;
 238
 239         if (!notified && !qtest_enabled() && !qtest_driver()) {
 240             warn_report("I/O thread spun for %d iterations",
 241                         MAX_MAIN_LOOP_SPIN);
 242             notified = true;
 243         }
 244
 245         timeout = SCALE_MS;
 246     }
 247
 248     if (timeout) {
 249         spin_counter = 0;
 250         qemu_mutex_unlock_iothread();
 251     } else {
 252         spin_counter++;
 253     }
 254
 255     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 256
 257     if (timeout) {
 258         qemu_mutex_lock_iothread();
 259     }
 260
 261     glib_pollfds_poll();
 262
 263     g_main_context_release(context);
 264
 265     return ret;
 266 }
 267 #else
 268 /***********************************************************/
 269 /* Polling handling */
 270
 271 typedef struct PollingEntry {
 272     PollingFunc *func;
 273     void *opaque;
 274     struct PollingEntry *next;
 275 } PollingEntry;
 276
 277 static PollingEntry *first_polling_entry;
 278
 279 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 280 {
 281     PollingEntry **ppe, *pe;
 282     pe = g_malloc0(sizeof(PollingEntry));
 283     pe->func = func;
 284     pe->opaque = opaque;
 285     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 286     *ppe = pe;
 287     return 0;
 288 }
 289
 290 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 291 {
 292     PollingEntry **ppe, *pe;
 293     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 294         pe = *ppe;
 295         if (pe->func == func && pe->opaque == opaque) {
 296             *ppe = pe->next;
 297             g_free(pe);
 298             break;
 299         }
 300     }
 301 }
 302
 303 /***********************************************************/
 304 /* Wait objects support */
 305 typedef struct WaitObjects {
 306     int num;
 307     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 308     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 309     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 310     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 311 } WaitObjects;
 312
 313 static WaitObjects wait_objects = {0};
 314
 315 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 316 {
 317     WaitObjects *w = &wait_objects;
 318     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 319         return -1;
 320     }
 321     w->events[w->num] = handle;
 322     w->func[w->num] = func;
 323     w->opaque[w->num] = opaque;
 324     w->revents[w->num] = 0;
 325     w->num++;
 326     return 0;
 327 }
 328
 329 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 330 {
 331     int i, found;
 332     WaitObjects *w = &wait_objects;
 333
 334     found = 0;
 335     for (i = 0; i < w->num; i++) {
 336         if (w->events[i] == handle) {
 337             found = 1;
 338         }
 339         if (found) {
 340             w->events[i] = w->events[i + 1];
 341             w->func[i] = w->func[i + 1];
 342             w->opaque[i] = w->opaque[i + 1];
 343             w->revents[i] = w->revents[i + 1];
 344         }
 345     }
 346     if (found) {
 347         w->num--;
 348     }
 349 }
 350
 351 void qemu_fd_register(int fd)
 352 {
 353     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 354                    FD_READ | FD_ACCEPT | FD_CLOSE |
 355                    FD_CONNECT | FD_WRITE | FD_OOB);
 356 }
 357
 358 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 359                         fd_set *xfds)
 360 {
 361     int nfds = -1;
 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 events = pfd->events;
 368         if (events & G_IO_IN) {
 369             FD_SET(fd, rfds);
 370             nfds = MAX(nfds, fd);
 371         }
 372         if (events & G_IO_OUT) {
 373             FD_SET(fd, wfds);
 374             nfds = MAX(nfds, fd);
 375         }
 376         if (events & G_IO_PRI) {
 377             FD_SET(fd, xfds);
 378             nfds = MAX(nfds, fd);
 379         }
 380     }
 381     return nfds;
 382 }
 383
 384 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 385                          fd_set *wfds, fd_set *xfds)
 386 {
 387     int i;
 388
 389     for (i = 0; i < pollfds->len; i++) {
 390         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 391         int fd = pfd->fd;
 392         int revents = 0;
 393
 394         if (FD_ISSET(fd, rfds)) {
 395             revents |= G_IO_IN;
 396         }
 397         if (FD_ISSET(fd, wfds)) {
 398             revents |= G_IO_OUT;
 399         }
 400         if (FD_ISSET(fd, xfds)) {
 401             revents |= G_IO_PRI;
 402         }
 403         pfd->revents = revents & pfd->events;
 404     }
 405 }
 406
 407 static int os_host_main_loop_wait(int64_t timeout)
 408 {
 409     GMainContext *context = g_main_context_default();
 410     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 411     int select_ret = 0;
 412     int g_poll_ret, ret, i, n_poll_fds;
 413     PollingEntry *pe;
 414     WaitObjects *w = &wait_objects;
 415     gint poll_timeout;
 416     int64_t poll_timeout_ns;
 417     static struct timeval tv0;
 418     fd_set rfds, wfds, xfds;
 419     int nfds;
 420
 421     g_main_context_acquire(context);
 422
 423     /* XXX: need to suppress polling by better using win32 events */
 424     ret = 0;
 425     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 426         ret |= pe->func(pe->opaque);
 427     }
 428     if (ret != 0) {
 429         g_main_context_release(context);
 430         return ret;
 431     }
 432
 433     FD_ZERO(&rfds);
 434     FD_ZERO(&wfds);
 435     FD_ZERO(&xfds);
 436     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 437     if (nfds >= 0) {
 438         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 439         if (select_ret != 0) {
 440             timeout = 0;
 441         }
 442         if (select_ret > 0) {
 443             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 444         }
 445     }
 446
 447     g_main_context_prepare(context, &max_priority);
 448     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 449                                       poll_fds, ARRAY_SIZE(poll_fds));
 450     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 451
 452     for (i = 0; i < w->num; i++) {
 453         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 454         poll_fds[n_poll_fds + i].events = G_IO_IN;
 455     }
 456
 457     if (poll_timeout < 0) {
 458         poll_timeout_ns = -1;
 459     } else {
 460         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 461     }
 462
 463     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 464
 465     qemu_mutex_unlock_iothread();
 466     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 467
 468     qemu_mutex_lock_iothread();
 469     if (g_poll_ret > 0) {
 470         for (i = 0; i < w->num; i++) {
 471             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 472         }
 473         for (i = 0; i < w->num; i++) {
 474             if (w->revents[i] && w->func[i]) {
 475                 w->func[i](w->opaque[i]);
 476             }
 477         }
 478     }
 479
 480     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 481         g_main_context_dispatch(context);
 482     }
 483
 484     g_main_context_release(context);
 485
 486     return select_ret || g_poll_ret;
 487 }
 488 #endif
 489
 490 void main_loop_wait(int nonblocking)
 491 {
 492     int ret;
 493     uint32_t timeout = UINT32_MAX;
 494     int64_t timeout_ns;
 495
 496     if (nonblocking) {
 497         timeout = 0;
 498     }
 499
 500     /* poll any events */
 501     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 502     /* XXX: separate device handlers from system ones */
 503     slirp_pollfds_fill(gpollfds, &timeout);
 504
 505     if (timeout == UINT32_MAX) {
 506         timeout_ns = -1;
 507     } else {
 508         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 509     }
 510
 511     timeout_ns = qemu_soonest_timeout(timeout_ns,
 512                                       timerlistgroup_deadline_ns(
 513                                           &main_loop_tlg));
 514
 515     ret = os_host_main_loop_wait(timeout_ns);
 516     slirp_pollfds_poll(gpollfds, (ret < 0));
 517
 518     /* CPU thread can infinitely wait for event after
 519        missing the warp */
 520     qemu_start_warp_timer();
 521     qemu_clock_run_all_timers();
 522 }
 523
 524 /* Functions to operate on the main QEMU AioContext.  */
 525
 526 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 527 {
 528     return aio_bh_new(qemu_aio_context, cb, opaque);
 529 }