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