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 "slirp/slirp.h"
  28 #include "qemu/main-loop.h"
  29 #include "block/aio.h"
  30
  31 #ifndef _WIN32
  32
  33 #include "qemu/compatfd.h"
  34
  35 /* If we have signalfd, we mask out the signals we want to handle and then
  36  * use signalfd to listen for them.  We rely on whatever the current signal
  37  * handler is to dispatch the signals when we receive them.
  38  */
  39 static void sigfd_handler(void *opaque)
  40 {
  41     int fd = (intptr_t)opaque;
  42     struct qemu_signalfd_siginfo info;
  43     struct sigaction action;
  44     ssize_t len;
  45
  46     while (1) {
  47         do {
  48             len = read(fd, &info, sizeof(info));
  49         } while (len == -1 && errno == EINTR);
  50
  51         if (len == -1 && errno == EAGAIN) {
  52             break;
  53         }
  54
  55         if (len != sizeof(info)) {
  56             printf("read from sigfd returned %zd: %m\n", len);
  57             return;
  58         }
  59
  60         sigaction(info.ssi_signo, NULL, &action);
  61         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  62             action.sa_sigaction(info.ssi_signo,
  63                                 (siginfo_t *)&info, NULL);
  64         } else if (action.sa_handler) {
  65             action.sa_handler(info.ssi_signo);
  66         }
  67     }
  68 }
  69
  70 static int qemu_signal_init(void)
  71 {
  72     int sigfd;
  73     sigset_t set;
  74
  75     /*
  76      * SIG_IPI must be blocked in the main thread and must not be caught
  77      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  78      * not catch it reliably.
  79      */
  80     sigemptyset(&set);
  81     sigaddset(&set, SIG_IPI);
  82     sigaddset(&set, SIGIO);
  83     sigaddset(&set, SIGALRM);
  84     sigaddset(&set, SIGBUS);
  85     pthread_sigmask(SIG_BLOCK, &set, NULL);
  86
  87     sigdelset(&set, SIG_IPI);
  88     sigfd = qemu_signalfd(&set);
  89     if (sigfd == -1) {
  90         fprintf(stderr, "failed to create signalfd\n");
  91         return -errno;
  92     }
  93
  94     fcntl_setfl(sigfd, O_NONBLOCK);
  95
  96     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
  97                          (void *)(intptr_t)sigfd);
  98
  99     return 0;
 100 }
 101
 102 #else /* _WIN32 */
 103
 104 static int qemu_signal_init(void)
 105 {
 106     return 0;
 107 }
 108 #endif
 109
 110 static AioContext *qemu_aio_context;
 111
 112 void qemu_notify_event(void)
 113 {
 114     if (!qemu_aio_context) {
 115         return;
 116     }
 117     aio_notify(qemu_aio_context);
 118 }
 119
 120 static GArray *gpollfds;
 121
 122 int qemu_init_main_loop(void)
 123 {
 124     int ret;
 125     GSource *src;
 126
 127     init_clocks();
 128     if (init_timer_alarm() < 0) {
 129         fprintf(stderr, "could not initialize alarm timer\n");
 130         exit(1);
 131     }
 132
 133     ret = qemu_signal_init();
 134     if (ret) {
 135         return ret;
 136     }
 137
 138     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 139     qemu_aio_context = aio_context_new();
 140     src = aio_get_g_source(qemu_aio_context);
 141     g_source_attach(src, NULL);
 142     g_source_unref(src);
 143     return 0;
 144 }
 145
 146 static fd_set rfds, wfds, xfds;
 147 static int nfds;
 148 static GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 149 static int n_poll_fds;
 150 static int max_priority;
 151
 152 /* Load rfds/wfds/xfds into gpollfds.  Will be removed a few commits later. */
 153 static void gpollfds_from_select(void)
 154 {
 155     int fd;
 156     for (fd = 0; fd <= nfds; fd++) {
 157         int events = 0;
 158         if (FD_ISSET(fd, &rfds)) {
 159             events |= G_IO_IN | G_IO_HUP | G_IO_ERR;
 160         }
 161         if (FD_ISSET(fd, &wfds)) {
 162             events |= G_IO_OUT | G_IO_ERR;
 163         }
 164         if (FD_ISSET(fd, &xfds)) {
 165             events |= G_IO_PRI;
 166         }
 167         if (events) {
 168             GPollFD pfd = {
 169                 .fd = fd,
 170                 .events = events,
 171             };
 172             g_array_append_val(gpollfds, pfd);
 173         }
 174     }
 175 }
 176
 177 /* Store gpollfds revents into rfds/wfds/xfds.  Will be removed a few commits
 178  * later.
 179  */
 180 static void gpollfds_to_select(int ret)
 181 {
 182     int i;
 183
 184     FD_ZERO(&rfds);
 185     FD_ZERO(&wfds);
 186     FD_ZERO(&xfds);
 187
 188     if (ret <= 0) {
 189         return;
 190     }
 191
 192     for (i = 0; i < gpollfds->len; i++) {
 193         int fd = g_array_index(gpollfds, GPollFD, i).fd;
 194         int revents = g_array_index(gpollfds, GPollFD, i).revents;
 195
 196         if (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) {
 197             FD_SET(fd, &rfds);
 198         }
 199         if (revents & (G_IO_OUT | G_IO_ERR)) {
 200             FD_SET(fd, &wfds);
 201         }
 202         if (revents & G_IO_PRI) {
 203             FD_SET(fd, &xfds);
 204         }
 205     }
 206 }
 207
 208 #ifndef _WIN32
 209 static void glib_select_fill(int *max_fd, fd_set *rfds, fd_set *wfds,
 210                              fd_set *xfds, uint32_t *cur_timeout)
 211 {
 212     GMainContext *context = g_main_context_default();
 213     int i;
 214     int timeout = 0;
 215
 216     g_main_context_prepare(context, &max_priority);
 217
 218     n_poll_fds = g_main_context_query(context, max_priority, &timeout,
 219                                       poll_fds, ARRAY_SIZE(poll_fds));
 220     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 221
 222     for (i = 0; i < n_poll_fds; i++) {
 223         GPollFD *p = &poll_fds[i];
 224
 225         if ((p->events & G_IO_IN)) {
 226             FD_SET(p->fd, rfds);
 227             *max_fd = MAX(*max_fd, p->fd);
 228         }
 229         if ((p->events & G_IO_OUT)) {
 230             FD_SET(p->fd, wfds);
 231             *max_fd = MAX(*max_fd, p->fd);
 232         }
 233         if ((p->events & G_IO_ERR)) {
 234             FD_SET(p->fd, xfds);
 235             *max_fd = MAX(*max_fd, p->fd);
 236         }
 237     }
 238
 239     if (timeout >= 0 && timeout < *cur_timeout) {
 240         *cur_timeout = timeout;
 241     }
 242 }
 243
 244 static void glib_select_poll(fd_set *rfds, fd_set *wfds, fd_set *xfds,
 245                              bool err)
 246 {
 247     GMainContext *context = g_main_context_default();
 248
 249     if (!err) {
 250         int i;
 251
 252         for (i = 0; i < n_poll_fds; i++) {
 253             GPollFD *p = &poll_fds[i];
 254
 255             if ((p->events & G_IO_IN) && FD_ISSET(p->fd, rfds)) {
 256                 p->revents |= G_IO_IN;
 257             }
 258             if ((p->events & G_IO_OUT) && FD_ISSET(p->fd, wfds)) {
 259                 p->revents |= G_IO_OUT;
 260             }
 261             if ((p->events & G_IO_ERR) && FD_ISSET(p->fd, xfds)) {
 262                 p->revents |= G_IO_ERR;
 263             }
 264         }
 265     }
 266
 267     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 268         g_main_context_dispatch(context);
 269     }
 270 }
 271
 272 static int os_host_main_loop_wait(uint32_t timeout)
 273 {
 274     int ret;
 275
 276     glib_select_fill(&nfds, &rfds, &wfds, &xfds, &timeout);
 277
 278     if (timeout > 0) {
 279         qemu_mutex_unlock_iothread();
 280     }
 281
 282     /* We'll eventually drop fd_set completely.  But for now we still have
 283      * *_fill() and *_poll() functions that use rfds/wfds/xfds.
 284      */
 285     gpollfds_from_select();
 286
 287     ret = g_poll((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 288
 289     gpollfds_to_select(ret);
 290
 291     if (timeout > 0) {
 292         qemu_mutex_lock_iothread();
 293     }
 294
 295     glib_select_poll(&rfds, &wfds, &xfds, (ret < 0));
 296     return ret;
 297 }
 298 #else
 299 /***********************************************************/
 300 /* Polling handling */
 301
 302 typedef struct PollingEntry {
 303     PollingFunc *func;
 304     void *opaque;
 305     struct PollingEntry *next;
 306 } PollingEntry;
 307
 308 static PollingEntry *first_polling_entry;
 309
 310 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 311 {
 312     PollingEntry **ppe, *pe;
 313     pe = g_malloc0(sizeof(PollingEntry));
 314     pe->func = func;
 315     pe->opaque = opaque;
 316     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 317     *ppe = pe;
 318     return 0;
 319 }
 320
 321 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 322 {
 323     PollingEntry **ppe, *pe;
 324     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 325         pe = *ppe;
 326         if (pe->func == func && pe->opaque == opaque) {
 327             *ppe = pe->next;
 328             g_free(pe);
 329             break;
 330         }
 331     }
 332 }
 333
 334 /***********************************************************/
 335 /* Wait objects support */
 336 typedef struct WaitObjects {
 337     int num;
 338     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 339     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 340     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 341     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 342 } WaitObjects;
 343
 344 static WaitObjects wait_objects = {0};
 345
 346 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 347 {
 348     WaitObjects *w = &wait_objects;
 349     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 350         return -1;
 351     }
 352     w->events[w->num] = handle;
 353     w->func[w->num] = func;
 354     w->opaque[w->num] = opaque;
 355     w->revents[w->num] = 0;
 356     w->num++;
 357     return 0;
 358 }
 359
 360 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 361 {
 362     int i, found;
 363     WaitObjects *w = &wait_objects;
 364
 365     found = 0;
 366     for (i = 0; i < w->num; i++) {
 367         if (w->events[i] == handle) {
 368             found = 1;
 369         }
 370         if (found) {
 371             w->events[i] = w->events[i + 1];
 372             w->func[i] = w->func[i + 1];
 373             w->opaque[i] = w->opaque[i + 1];
 374             w->revents[i] = w->revents[i + 1];
 375         }
 376     }
 377     if (found) {
 378         w->num--;
 379     }
 380 }
 381
 382 void qemu_fd_register(int fd)
 383 {
 384     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 385                    FD_READ | FD_ACCEPT | FD_CLOSE |
 386                    FD_CONNECT | FD_WRITE | FD_OOB);
 387 }
 388
 389 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 390                         fd_set *xfds)
 391 {
 392     int nfds = -1;
 393     int i;
 394
 395     for (i = 0; i < pollfds->len; i++) {
 396         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 397         int fd = pfd->fd;
 398         int events = pfd->events;
 399         if (events & (G_IO_IN | G_IO_HUP | G_IO_ERR)) {
 400             FD_SET(fd, rfds);
 401             nfds = MAX(nfds, fd);
 402         }
 403         if (events & (G_IO_OUT | G_IO_ERR)) {
 404             FD_SET(fd, wfds);
 405             nfds = MAX(nfds, fd);
 406         }
 407         if (events & G_IO_PRI) {
 408             FD_SET(fd, xfds);
 409             nfds = MAX(nfds, fd);
 410         }
 411     }
 412     return nfds;
 413 }
 414
 415 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 416                          fd_set *wfds, fd_set *xfds)
 417 {
 418     int i;
 419
 420     for (i = 0; i < pollfds->len; i++) {
 421         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 422         int fd = pfd->fd;
 423         int revents = 0;
 424
 425         if (FD_ISSET(fd, rfds)) {
 426             revents |= G_IO_IN | G_IO_HUP | G_IO_ERR;
 427         }
 428         if (FD_ISSET(fd, wfds)) {
 429             revents |= G_IO_OUT | G_IO_ERR;
 430         }
 431         if (FD_ISSET(fd, xfds)) {
 432             revents |= G_IO_PRI;
 433         }
 434         pfd->revents = revents & pfd->events;
 435     }
 436 }
 437
 438 static int os_host_main_loop_wait(uint32_t timeout)
 439 {
 440     GMainContext *context = g_main_context_default();
 441     int select_ret = 0;
 442     int g_poll_ret, ret, i;
 443     PollingEntry *pe;
 444     WaitObjects *w = &wait_objects;
 445     gint poll_timeout;
 446     static struct timeval tv0;
 447
 448     /* XXX: need to suppress polling by better using win32 events */
 449     ret = 0;
 450     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 451         ret |= pe->func(pe->opaque);
 452     }
 453     if (ret != 0) {
 454         return ret;
 455     }
 456
 457     g_main_context_prepare(context, &max_priority);
 458     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 459                                       poll_fds, ARRAY_SIZE(poll_fds));
 460     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 461
 462     for (i = 0; i < w->num; i++) {
 463         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 464         poll_fds[n_poll_fds + i].events = G_IO_IN;
 465     }
 466
 467     if (poll_timeout < 0 || timeout < poll_timeout) {
 468         poll_timeout = timeout;
 469     }
 470
 471     qemu_mutex_unlock_iothread();
 472     g_poll_ret = g_poll(poll_fds, n_poll_fds + w->num, poll_timeout);
 473     qemu_mutex_lock_iothread();
 474     if (g_poll_ret > 0) {
 475         for (i = 0; i < w->num; i++) {
 476             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 477         }
 478         for (i = 0; i < w->num; i++) {
 479             if (w->revents[i] && w->func[i]) {
 480                 w->func[i](w->opaque[i]);
 481             }
 482         }
 483     }
 484
 485     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 486         g_main_context_dispatch(context);
 487     }
 488
 489     /* Call select after g_poll to avoid a useless iteration and therefore
 490      * improve socket latency.
 491      */
 492
 493     /* This back-and-forth between GPollFDs and select(2) is temporary.  We'll
 494      * drop it in a couple of patches, I promise :).
 495      */
 496     gpollfds_from_select();
 497     FD_ZERO(&rfds);
 498     FD_ZERO(&wfds);
 499     FD_ZERO(&xfds);
 500     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 501     if (nfds >= 0) {
 502         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 503         if (select_ret != 0) {
 504             timeout = 0;
 505         }
 506         if (select_ret > 0) {
 507             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 508         }
 509     }
 510     gpollfds_to_select(select_ret);
 511
 512     return select_ret || g_poll_ret;
 513 }
 514 #endif
 515
 516 int main_loop_wait(int nonblocking)
 517 {
 518     int ret;
 519     uint32_t timeout = UINT32_MAX;
 520
 521     if (nonblocking) {
 522         timeout = 0;
 523     }
 524
 525     /* poll any events */
 526     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 527     /* XXX: separate device handlers from system ones */
 528     nfds = -1;
 529     FD_ZERO(&rfds);
 530     FD_ZERO(&wfds);
 531     FD_ZERO(&xfds);
 532
 533 #ifdef CONFIG_SLIRP
 534     slirp_update_timeout(&timeout);
 535     slirp_select_fill(&nfds, &rfds, &wfds, &xfds);
 536 #endif
 537     qemu_iohandler_fill(&nfds, &rfds, &wfds, &xfds);
 538     ret = os_host_main_loop_wait(timeout);
 539     qemu_iohandler_poll(&rfds, &wfds, &xfds, ret);
 540 #ifdef CONFIG_SLIRP
 541     slirp_select_poll(&rfds, &wfds, &xfds, (ret < 0));
 542 #endif
 543
 544     qemu_run_all_timers();
 545
 546     return ret;
 547 }
 548
 549 /* Functions to operate on the main QEMU AioContext.  */
 550
 551 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 552 {
 553     return aio_bh_new(qemu_aio_context, cb, opaque);
 554 }
 555
 556 bool qemu_aio_wait(void)
 557 {
 558     return aio_poll(qemu_aio_context, true);
 559 }
 560
 561 #ifdef CONFIG_POSIX
 562 void qemu_aio_set_fd_handler(int fd,
 563                              IOHandler *io_read,
 564                              IOHandler *io_write,
 565                              AioFlushHandler *io_flush,
 566                              void *opaque)
 567 {
 568     aio_set_fd_handler(qemu_aio_context, fd, io_read, io_write, io_flush,
 569                        opaque);
 570 }
 571 #endif
 572
 573 void qemu_aio_set_event_notifier(EventNotifier *notifier,
 574                                  EventNotifierHandler *io_read,
 575                                  AioFlushEventNotifierHandler *io_flush)
 576 {
 577     aio_set_event_notifier(qemu_aio_context, notifier, io_read, io_flush);
 578 }