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 "slirp/libslirp.h"
  29 #include "qemu/main-loop.h"
  30 #include "block/aio.h"
  31
  32 #ifndef _WIN32
  33
  34 #include "qemu/compatfd.h"
  35
  36 /* If we have signalfd, we mask out the signals we want to handle and then
  37  * use signalfd to listen for them.  We rely on whatever the current signal
  38  * handler is to dispatch the signals when we receive them.
  39  */
  40 static void sigfd_handler(void *opaque)
  41 {
  42     int fd = (intptr_t)opaque;
  43     struct qemu_signalfd_siginfo info;
  44     struct sigaction action;
  45     ssize_t len;
  46
  47     while (1) {
  48         do {
  49             len = read(fd, &info, sizeof(info));
  50         } while (len == -1 && errno == EINTR);
  51
  52         if (len == -1 && errno == EAGAIN) {
  53             break;
  54         }
  55
  56         if (len != sizeof(info)) {
  57             printf("read from sigfd returned %zd: %m\n", len);
  58             return;
  59         }
  60
  61         sigaction(info.ssi_signo, NULL, &action);
  62         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  63             action.sa_sigaction(info.ssi_signo,
  64                                 (siginfo_t *)&info, NULL);
  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     pthread_sigmask(SIG_BLOCK, &set, NULL);
  87
  88     sigdelset(&set, SIG_IPI);
  89     sigfd = qemu_signalfd(&set);
  90     if (sigfd == -1) {
  91         fprintf(stderr, "failed to create signalfd\n");
  92         return -errno;
  93     }
  94
  95     fcntl_setfl(sigfd, O_NONBLOCK);
  96
  97     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
  98                          (void *)(intptr_t)sigfd);
  99
 100     return 0;
 101 }
 102
 103 #else /* _WIN32 */
 104
 105 static int qemu_signal_init(void)
 106 {
 107     return 0;
 108 }
 109 #endif
 110
 111 static AioContext *qemu_aio_context;
 112
 113 AioContext *qemu_get_aio_context(void)
 114 {
 115     return qemu_aio_context;
 116 }
 117
 118 void qemu_notify_event(void)
 119 {
 120     if (!qemu_aio_context) {
 121         return;
 122     }
 123     aio_notify(qemu_aio_context);
 124 }
 125
 126 static GArray *gpollfds;
 127
 128 int qemu_init_main_loop(void)
 129 {
 130     int ret;
 131     GSource *src;
 132
 133     init_clocks();
 134     if (init_timer_alarm() < 0) {
 135         fprintf(stderr, "could not initialize alarm timer\n");
 136         exit(1);
 137     }
 138
 139     ret = qemu_signal_init();
 140     if (ret) {
 141         return ret;
 142     }
 143
 144     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 145     qemu_aio_context = aio_context_new();
 146     src = aio_get_g_source(qemu_aio_context);
 147     g_source_attach(src, NULL);
 148     g_source_unref(src);
 149     return 0;
 150 }
 151
 152 static int max_priority;
 153
 154 #ifndef _WIN32
 155 static int glib_pollfds_idx;
 156 static int glib_n_poll_fds;
 157
 158 static void glib_pollfds_fill(uint32_t *cur_timeout)
 159 {
 160     GMainContext *context = g_main_context_default();
 161     int timeout = 0;
 162     int n;
 163
 164     g_main_context_prepare(context, &max_priority);
 165
 166     glib_pollfds_idx = gpollfds->len;
 167     n = glib_n_poll_fds;
 168     do {
 169         GPollFD *pfds;
 170         glib_n_poll_fds = n;
 171         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 172         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 173         n = g_main_context_query(context, max_priority, &timeout, pfds,
 174                                  glib_n_poll_fds);
 175     } while (n != glib_n_poll_fds);
 176
 177     if (timeout >= 0 && timeout < *cur_timeout) {
 178         *cur_timeout = timeout;
 179     }
 180 }
 181
 182 static void glib_pollfds_poll(void)
 183 {
 184     GMainContext *context = g_main_context_default();
 185     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 186
 187     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 188         g_main_context_dispatch(context);
 189     }
 190 }
 191
 192 #define MAX_MAIN_LOOP_SPIN (1000)
 193
 194 static int os_host_main_loop_wait(uint32_t timeout)
 195 {
 196     int ret;
 197     static int spin_counter;
 198
 199     glib_pollfds_fill(&timeout);
 200
 201     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 202      * the I/O thread, this can lead to starvation of the BQL such that the
 203      * VCPU threads never run.  To make sure we can detect the later case,
 204      * print a message to the screen.  If we run into this condition, create
 205      * a fake timeout in order to give the VCPU threads a chance to run.
 206      */
 207     if (spin_counter > MAX_MAIN_LOOP_SPIN) {
 208         static bool notified;
 209
 210         if (!notified) {
 211             fprintf(stderr,
 212                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 213                     MAX_MAIN_LOOP_SPIN);
 214             notified = true;
 215         }
 216
 217         timeout = 1;
 218     }
 219
 220     if (timeout > 0) {
 221         spin_counter = 0;
 222         qemu_mutex_unlock_iothread();
 223     } else {
 224         spin_counter++;
 225     }
 226
 227     ret = g_poll((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 228
 229     if (timeout > 0) {
 230         qemu_mutex_lock_iothread();
 231     }
 232
 233     glib_pollfds_poll();
 234     return ret;
 235 }
 236 #else
 237 /***********************************************************/
 238 /* Polling handling */
 239
 240 typedef struct PollingEntry {
 241     PollingFunc *func;
 242     void *opaque;
 243     struct PollingEntry *next;
 244 } PollingEntry;
 245
 246 static PollingEntry *first_polling_entry;
 247
 248 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 249 {
 250     PollingEntry **ppe, *pe;
 251     pe = g_malloc0(sizeof(PollingEntry));
 252     pe->func = func;
 253     pe->opaque = opaque;
 254     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 255     *ppe = pe;
 256     return 0;
 257 }
 258
 259 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 260 {
 261     PollingEntry **ppe, *pe;
 262     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 263         pe = *ppe;
 264         if (pe->func == func && pe->opaque == opaque) {
 265             *ppe = pe->next;
 266             g_free(pe);
 267             break;
 268         }
 269     }
 270 }
 271
 272 /***********************************************************/
 273 /* Wait objects support */
 274 typedef struct WaitObjects {
 275     int num;
 276     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 277     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 278     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 279     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 280 } WaitObjects;
 281
 282 static WaitObjects wait_objects = {0};
 283
 284 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 285 {
 286     WaitObjects *w = &wait_objects;
 287     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 288         return -1;
 289     }
 290     w->events[w->num] = handle;
 291     w->func[w->num] = func;
 292     w->opaque[w->num] = opaque;
 293     w->revents[w->num] = 0;
 294     w->num++;
 295     return 0;
 296 }
 297
 298 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 299 {
 300     int i, found;
 301     WaitObjects *w = &wait_objects;
 302
 303     found = 0;
 304     for (i = 0; i < w->num; i++) {
 305         if (w->events[i] == handle) {
 306             found = 1;
 307         }
 308         if (found) {
 309             w->events[i] = w->events[i + 1];
 310             w->func[i] = w->func[i + 1];
 311             w->opaque[i] = w->opaque[i + 1];
 312             w->revents[i] = w->revents[i + 1];
 313         }
 314     }
 315     if (found) {
 316         w->num--;
 317     }
 318 }
 319
 320 void qemu_fd_register(int fd)
 321 {
 322     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 323                    FD_READ | FD_ACCEPT | FD_CLOSE |
 324                    FD_CONNECT | FD_WRITE | FD_OOB);
 325 }
 326
 327 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 328                         fd_set *xfds)
 329 {
 330     int nfds = -1;
 331     int i;
 332
 333     for (i = 0; i < pollfds->len; i++) {
 334         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 335         int fd = pfd->fd;
 336         int events = pfd->events;
 337         if (events & G_IO_IN) {
 338             FD_SET(fd, rfds);
 339             nfds = MAX(nfds, fd);
 340         }
 341         if (events & G_IO_OUT) {
 342             FD_SET(fd, wfds);
 343             nfds = MAX(nfds, fd);
 344         }
 345         if (events & G_IO_PRI) {
 346             FD_SET(fd, xfds);
 347             nfds = MAX(nfds, fd);
 348         }
 349     }
 350     return nfds;
 351 }
 352
 353 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 354                          fd_set *wfds, fd_set *xfds)
 355 {
 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 revents = 0;
 362
 363         if (FD_ISSET(fd, rfds)) {
 364             revents |= G_IO_IN;
 365         }
 366         if (FD_ISSET(fd, wfds)) {
 367             revents |= G_IO_OUT;
 368         }
 369         if (FD_ISSET(fd, xfds)) {
 370             revents |= G_IO_PRI;
 371         }
 372         pfd->revents = revents & pfd->events;
 373     }
 374 }
 375
 376 static int os_host_main_loop_wait(uint32_t timeout)
 377 {
 378     GMainContext *context = g_main_context_default();
 379     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 380     int select_ret = 0;
 381     int g_poll_ret, ret, i, n_poll_fds;
 382     PollingEntry *pe;
 383     WaitObjects *w = &wait_objects;
 384     gint poll_timeout;
 385     static struct timeval tv0;
 386     fd_set rfds, wfds, xfds;
 387     int nfds;
 388
 389     /* XXX: need to suppress polling by better using win32 events */
 390     ret = 0;
 391     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 392         ret |= pe->func(pe->opaque);
 393     }
 394     if (ret != 0) {
 395         return ret;
 396     }
 397
 398     FD_ZERO(&rfds);
 399     FD_ZERO(&wfds);
 400     FD_ZERO(&xfds);
 401     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 402     if (nfds >= 0) {
 403         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 404         if (select_ret != 0) {
 405             timeout = 0;
 406         }
 407         if (select_ret > 0) {
 408             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 409         }
 410     }
 411
 412     g_main_context_prepare(context, &max_priority);
 413     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 414                                       poll_fds, ARRAY_SIZE(poll_fds));
 415     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 416
 417     for (i = 0; i < w->num; i++) {
 418         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 419         poll_fds[n_poll_fds + i].events = G_IO_IN;
 420     }
 421
 422     if (poll_timeout < 0 || timeout < poll_timeout) {
 423         poll_timeout = timeout;
 424     }
 425
 426     qemu_mutex_unlock_iothread();
 427     g_poll_ret = g_poll(poll_fds, n_poll_fds + w->num, poll_timeout);
 428     qemu_mutex_lock_iothread();
 429     if (g_poll_ret > 0) {
 430         for (i = 0; i < w->num; i++) {
 431             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 432         }
 433         for (i = 0; i < w->num; i++) {
 434             if (w->revents[i] && w->func[i]) {
 435                 w->func[i](w->opaque[i]);
 436             }
 437         }
 438     }
 439
 440     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 441         g_main_context_dispatch(context);
 442     }
 443
 444     return select_ret || g_poll_ret;
 445 }
 446 #endif
 447
 448 int main_loop_wait(int nonblocking)
 449 {
 450     int ret;
 451     uint32_t timeout = UINT32_MAX;
 452
 453     if (nonblocking) {
 454         timeout = 0;
 455     }
 456
 457     /* poll any events */
 458     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 459     /* XXX: separate device handlers from system ones */
 460 #ifdef CONFIG_SLIRP
 461     slirp_update_timeout(&timeout);
 462     slirp_pollfds_fill(gpollfds);
 463 #endif
 464     qemu_iohandler_fill(gpollfds);
 465     ret = os_host_main_loop_wait(timeout);
 466     qemu_iohandler_poll(gpollfds, ret);
 467 #ifdef CONFIG_SLIRP
 468     slirp_pollfds_poll(gpollfds, (ret < 0));
 469 #endif
 470
 471     qemu_run_all_timers();
 472
 473     return ret;
 474 }
 475
 476 /* Functions to operate on the main QEMU AioContext.  */
 477
 478 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 479 {
 480     return aio_bh_new(qemu_aio_context, cb, opaque);
 481 }
 482
 483 bool qemu_aio_wait(void)
 484 {
 485     return aio_poll(qemu_aio_context, true);
 486 }
 487
 488 #ifdef CONFIG_POSIX
 489 void qemu_aio_set_fd_handler(int fd,
 490                              IOHandler *io_read,
 491                              IOHandler *io_write,
 492                              AioFlushHandler *io_flush,
 493                              void *opaque)
 494 {
 495     aio_set_fd_handler(qemu_aio_context, fd, io_read, io_write, io_flush,
 496                        opaque);
 497 }
 498 #endif
 499
 500 void qemu_aio_set_event_notifier(EventNotifier *notifier,
 501                                  EventNotifierHandler *io_read,
 502                                  AioFlushEventNotifierHandler *io_flush)
 503 {
 504     aio_set_event_notifier(qemu_aio_context, notifier, io_read, io_flush);
 505 }