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 "sysemu/qtest.h"
  29 #include "slirp/libslirp.h"
  30 #include "qemu/main-loop.h"
  31 #include "block/aio.h"
  32
  33 #ifndef _WIN32
  34
  35 #include "qemu/compatfd.h"
  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             action.sa_sigaction(info.ssi_signo,
  65                                 (siginfo_t *)&info, NULL);
  66         } else if (action.sa_handler) {
  67             action.sa_handler(info.ssi_signo);
  68         }
  69     }
  70 }
  71
  72 static int qemu_signal_init(void)
  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     pthread_sigmask(SIG_BLOCK, &set, NULL);
  88
  89     sigdelset(&set, SIG_IPI);
  90     sigfd = qemu_signalfd(&set);
  91     if (sigfd == -1) {
  92         fprintf(stderr, "failed to create signalfd\n");
  93         return -errno;
  94     }
  95
  96     fcntl_setfl(sigfd, O_NONBLOCK);
  97
  98     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
  99                          (void *)(intptr_t)sigfd);
 100
 101     return 0;
 102 }
 103
 104 #else /* _WIN32 */
 105
 106 static int qemu_signal_init(void)
 107 {
 108     return 0;
 109 }
 110 #endif
 111
 112 static AioContext *qemu_aio_context;
 113
 114 AioContext *qemu_get_aio_context(void)
 115 {
 116     return qemu_aio_context;
 117 }
 118
 119 void qemu_notify_event(void)
 120 {
 121     if (!qemu_aio_context) {
 122         return;
 123     }
 124     aio_notify(qemu_aio_context);
 125 }
 126
 127 static GArray *gpollfds;
 128
 129 int qemu_init_main_loop(void)
 130 {
 131     int ret;
 132     GSource *src;
 133
 134     init_clocks();
 135
 136     ret = qemu_signal_init();
 137     if (ret) {
 138         return ret;
 139     }
 140
 141     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 142     qemu_aio_context = aio_context_new();
 143     src = aio_get_g_source(qemu_aio_context);
 144     g_source_attach(src, NULL);
 145     g_source_unref(src);
 146     return 0;
 147 }
 148
 149 static int max_priority;
 150
 151 #ifndef _WIN32
 152 static int glib_pollfds_idx;
 153 static int glib_n_poll_fds;
 154
 155 static void glib_pollfds_fill(int64_t *cur_timeout)
 156 {
 157     GMainContext *context = g_main_context_default();
 158     int timeout = 0;
 159     int64_t timeout_ns;
 160     int n;
 161
 162     g_main_context_prepare(context, &max_priority);
 163
 164     glib_pollfds_idx = gpollfds->len;
 165     n = glib_n_poll_fds;
 166     do {
 167         GPollFD *pfds;
 168         glib_n_poll_fds = n;
 169         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 170         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 171         n = g_main_context_query(context, max_priority, &timeout, pfds,
 172                                  glib_n_poll_fds);
 173     } while (n != glib_n_poll_fds);
 174
 175     if (timeout < 0) {
 176         timeout_ns = -1;
 177     } else {
 178         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 179     }
 180
 181     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 182 }
 183
 184 static void glib_pollfds_poll(void)
 185 {
 186     GMainContext *context = g_main_context_default();
 187     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 188
 189     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 190         g_main_context_dispatch(context);
 191     }
 192 }
 193
 194 #define MAX_MAIN_LOOP_SPIN (1000)
 195
 196 static int os_host_main_loop_wait(int64_t timeout)
 197 {
 198     int ret;
 199     static int spin_counter;
 200
 201     glib_pollfds_fill(&timeout);
 202
 203     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 204      * the I/O thread, this can lead to starvation of the BQL such that the
 205      * VCPU threads never run.  To make sure we can detect the later case,
 206      * print a message to the screen.  If we run into this condition, create
 207      * a fake timeout in order to give the VCPU threads a chance to run.
 208      */
 209     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 210         static bool notified;
 211
 212         if (!notified && !qtest_enabled()) {
 213             fprintf(stderr,
 214                     "main-loop: WARNING: I/O thread spun for %d iterations\n",
 215                     MAX_MAIN_LOOP_SPIN);
 216             notified = true;
 217         }
 218
 219         timeout = SCALE_MS;
 220     }
 221
 222     if (timeout) {
 223         spin_counter = 0;
 224         qemu_mutex_unlock_iothread();
 225     } else {
 226         spin_counter++;
 227     }
 228
 229     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 230
 231     if (timeout) {
 232         qemu_mutex_lock_iothread();
 233     }
 234
 235     glib_pollfds_poll();
 236     return ret;
 237 }
 238 #else
 239 /***********************************************************/
 240 /* Polling handling */
 241
 242 typedef struct PollingEntry {
 243     PollingFunc *func;
 244     void *opaque;
 245     struct PollingEntry *next;
 246 } PollingEntry;
 247
 248 static PollingEntry *first_polling_entry;
 249
 250 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 251 {
 252     PollingEntry **ppe, *pe;
 253     pe = g_malloc0(sizeof(PollingEntry));
 254     pe->func = func;
 255     pe->opaque = opaque;
 256     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 257     *ppe = pe;
 258     return 0;
 259 }
 260
 261 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 262 {
 263     PollingEntry **ppe, *pe;
 264     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 265         pe = *ppe;
 266         if (pe->func == func && pe->opaque == opaque) {
 267             *ppe = pe->next;
 268             g_free(pe);
 269             break;
 270         }
 271     }
 272 }
 273
 274 /***********************************************************/
 275 /* Wait objects support */
 276 typedef struct WaitObjects {
 277     int num;
 278     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 279     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 280     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 281     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 282 } WaitObjects;
 283
 284 static WaitObjects wait_objects = {0};
 285
 286 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 287 {
 288     WaitObjects *w = &wait_objects;
 289     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 290         return -1;
 291     }
 292     w->events[w->num] = handle;
 293     w->func[w->num] = func;
 294     w->opaque[w->num] = opaque;
 295     w->revents[w->num] = 0;
 296     w->num++;
 297     return 0;
 298 }
 299
 300 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 301 {
 302     int i, found;
 303     WaitObjects *w = &wait_objects;
 304
 305     found = 0;
 306     for (i = 0; i < w->num; i++) {
 307         if (w->events[i] == handle) {
 308             found = 1;
 309         }
 310         if (found) {
 311             w->events[i] = w->events[i + 1];
 312             w->func[i] = w->func[i + 1];
 313             w->opaque[i] = w->opaque[i + 1];
 314             w->revents[i] = w->revents[i + 1];
 315         }
 316     }
 317     if (found) {
 318         w->num--;
 319     }
 320 }
 321
 322 void qemu_fd_register(int fd)
 323 {
 324     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 325                    FD_READ | FD_ACCEPT | FD_CLOSE |
 326                    FD_CONNECT | FD_WRITE | FD_OOB);
 327 }
 328
 329 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 330                         fd_set *xfds)
 331 {
 332     int nfds = -1;
 333     int i;
 334
 335     for (i = 0; i < pollfds->len; i++) {
 336         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 337         int fd = pfd->fd;
 338         int events = pfd->events;
 339         if (events & G_IO_IN) {
 340             FD_SET(fd, rfds);
 341             nfds = MAX(nfds, fd);
 342         }
 343         if (events & G_IO_OUT) {
 344             FD_SET(fd, wfds);
 345             nfds = MAX(nfds, fd);
 346         }
 347         if (events & G_IO_PRI) {
 348             FD_SET(fd, xfds);
 349             nfds = MAX(nfds, fd);
 350         }
 351     }
 352     return nfds;
 353 }
 354
 355 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 356                          fd_set *wfds, fd_set *xfds)
 357 {
 358     int i;
 359
 360     for (i = 0; i < pollfds->len; i++) {
 361         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 362         int fd = pfd->fd;
 363         int revents = 0;
 364
 365         if (FD_ISSET(fd, rfds)) {
 366             revents |= G_IO_IN;
 367         }
 368         if (FD_ISSET(fd, wfds)) {
 369             revents |= G_IO_OUT;
 370         }
 371         if (FD_ISSET(fd, xfds)) {
 372             revents |= G_IO_PRI;
 373         }
 374         pfd->revents = revents & pfd->events;
 375     }
 376 }
 377
 378 static int os_host_main_loop_wait(int64_t timeout)
 379 {
 380     GMainContext *context = g_main_context_default();
 381     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 382     int select_ret = 0;
 383     int g_poll_ret, ret, i, n_poll_fds;
 384     PollingEntry *pe;
 385     WaitObjects *w = &wait_objects;
 386     gint poll_timeout;
 387     int64_t poll_timeout_ns;
 388     static struct timeval tv0;
 389     fd_set rfds, wfds, xfds;
 390     int nfds;
 391
 392     /* XXX: need to suppress polling by better using win32 events */
 393     ret = 0;
 394     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 395         ret |= pe->func(pe->opaque);
 396     }
 397     if (ret != 0) {
 398         return ret;
 399     }
 400
 401     FD_ZERO(&rfds);
 402     FD_ZERO(&wfds);
 403     FD_ZERO(&xfds);
 404     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 405     if (nfds >= 0) {
 406         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 407         if (select_ret != 0) {
 408             timeout = 0;
 409         }
 410         if (select_ret > 0) {
 411             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 412         }
 413     }
 414
 415     g_main_context_prepare(context, &max_priority);
 416     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 417                                       poll_fds, ARRAY_SIZE(poll_fds));
 418     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 419
 420     for (i = 0; i < w->num; i++) {
 421         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 422         poll_fds[n_poll_fds + i].events = G_IO_IN;
 423     }
 424
 425     if (poll_timeout < 0) {
 426         poll_timeout_ns = -1;
 427     } else {
 428         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 429     }
 430
 431     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 432
 433     qemu_mutex_unlock_iothread();
 434     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 435
 436     qemu_mutex_lock_iothread();
 437     if (g_poll_ret > 0) {
 438         for (i = 0; i < w->num; i++) {
 439             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 440         }
 441         for (i = 0; i < w->num; i++) {
 442             if (w->revents[i] && w->func[i]) {
 443                 w->func[i](w->opaque[i]);
 444             }
 445         }
 446     }
 447
 448     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 449         g_main_context_dispatch(context);
 450     }
 451
 452     return select_ret || g_poll_ret;
 453 }
 454 #endif
 455
 456 int main_loop_wait(int nonblocking)
 457 {
 458     int ret;
 459     uint32_t timeout = UINT32_MAX;
 460     int64_t timeout_ns;
 461
 462     if (nonblocking) {
 463         timeout = 0;
 464     }
 465
 466     /* poll any events */
 467     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 468     /* XXX: separate device handlers from system ones */
 469 #ifdef CONFIG_SLIRP
 470     slirp_pollfds_fill(gpollfds, &timeout);
 471 #endif
 472     qemu_iohandler_fill(gpollfds);
 473
 474     if (timeout == UINT32_MAX) {
 475         timeout_ns = -1;
 476     } else {
 477         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 478     }
 479
 480     timeout_ns = qemu_soonest_timeout(timeout_ns,
 481                                       timerlistgroup_deadline_ns(
 482                                           &main_loop_tlg));
 483
 484     ret = os_host_main_loop_wait(timeout_ns);
 485     qemu_iohandler_poll(gpollfds, ret);
 486 #ifdef CONFIG_SLIRP
 487     slirp_pollfds_poll(gpollfds, (ret < 0));
 488 #endif
 489
 490     qemu_clock_run_all_timers();
 491
 492     return ret;
 493 }
 494
 495 /* Functions to operate on the main QEMU AioContext.  */
 496
 497 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 498 {
 499     return aio_bh_new(qemu_aio_context, cb, opaque);
 500 }
 501
 502 bool qemu_aio_wait(void)
 503 {
 504     return aio_poll(qemu_aio_context, true);
 505 }
 506
 507 #ifdef CONFIG_POSIX
 508 void qemu_aio_set_fd_handler(int fd,
 509                              IOHandler *io_read,
 510                              IOHandler *io_write,
 511                              void *opaque)
 512 {
 513     aio_set_fd_handler(qemu_aio_context, fd, io_read, io_write, opaque);
 514 }
 515 #endif
 516
 517 void qemu_aio_set_event_notifier(EventNotifier *notifier,
 518                                  EventNotifierHandler *io_read)
 519 {
 520     aio_set_event_notifier(qemu_aio_context, notifier, io_read);
 521 }