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 "sysemu/replay.h"
  33 #include "slirp/libslirp.h"
  34 #include "qemu/main-loop.h"
  35 #include "block/aio.h"
  36 #include "qemu/error-report.h"
  37
  38 #ifndef _WIN32
  39
  40 /* If we have signalfd, we mask out the signals we want to handle and then
  41  * use signalfd to listen for them.  We rely on whatever the current signal
  42  * handler is to dispatch the signals when we receive them.
  43  */
  44 static void sigfd_handler(void *opaque)
  45 {
  46     int fd = (intptr_t)opaque;
  47     struct qemu_signalfd_siginfo info;
  48     struct sigaction action;
  49     ssize_t len;
  50
  51     while (1) {
  52         do {
  53             len = read(fd, &info, sizeof(info));
  54         } while (len == -1 && errno == EINTR);
  55
  56         if (len == -1 && errno == EAGAIN) {
  57             break;
  58         }
  59
  60         if (len != sizeof(info)) {
  61             printf("read from sigfd returned %zd: %m\n", len);
  62             return;
  63         }
  64
  65         sigaction(info.ssi_signo, NULL, &action);
  66         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  67             sigaction_invoke(&action, &info);
  68         } else if (action.sa_handler) {
  69             action.sa_handler(info.ssi_signo);
  70         }
  71     }
  72 }
  73
  74 static int qemu_signal_init(void)
  75 {
  76     int sigfd;
  77     sigset_t set;
  78
  79     /*
  80      * SIG_IPI must be blocked in the main thread and must not be caught
  81      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  82      * not catch it reliably.
  83      */
  84     sigemptyset(&set);
  85     sigaddset(&set, SIG_IPI);
  86     sigaddset(&set, SIGIO);
  87     sigaddset(&set, SIGALRM);
  88     sigaddset(&set, SIGBUS);
  89     /* SIGINT cannot be handled via signalfd, so that ^C can be used
  90      * to interrupt QEMU when it is being run under gdb.  SIGHUP and
  91      * SIGTERM are also handled asynchronously, even though it is not
  92      * strictly necessary, because they use the same handler as SIGINT.
  93      */
  94     pthread_sigmask(SIG_BLOCK, &set, NULL);
  95
  96     sigdelset(&set, SIG_IPI);
  97     sigfd = qemu_signalfd(&set);
  98     if (sigfd == -1) {
  99         fprintf(stderr, "failed to create signalfd\n");
 100         return -errno;
 101     }
 102
 103     fcntl_setfl(sigfd, O_NONBLOCK);
 104
 105     qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
 106
 107     return 0;
 108 }
 109
 110 #else /* _WIN32 */
 111
 112 static int qemu_signal_init(void)
 113 {
 114     return 0;
 115 }
 116 #endif
 117
 118 static AioContext *qemu_aio_context;
 119 static QEMUBH *qemu_notify_bh;
 120
 121 static void notify_event_cb(void *opaque)
 122 {
 123     /* No need to do anything; this bottom half is only used to
 124      * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
 125      */
 126 }
 127
 128 AioContext *qemu_get_aio_context(void)
 129 {
 130     return qemu_aio_context;
 131 }
 132
 133 void qemu_notify_event(void)
 134 {
 135     if (!qemu_aio_context) {
 136         return;
 137     }
 138     qemu_bh_schedule(qemu_notify_bh);
 139 }
 140
 141 static GArray *gpollfds;
 142
 143 int qemu_init_main_loop(Error **errp)
 144 {
 145     int ret;
 146     GSource *src;
 147     Error *local_error = NULL;
 148
 149     init_clocks(qemu_timer_notify_cb);
 150
 151     ret = qemu_signal_init();
 152     if (ret) {
 153         return ret;
 154     }
 155
 156     qemu_aio_context = aio_context_new(&local_error);
 157     if (!qemu_aio_context) {
 158         error_propagate(errp, local_error);
 159         return -EMFILE;
 160     }
 161     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 162     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 163     src = aio_get_g_source(qemu_aio_context);
 164     g_source_set_name(src, "aio-context");
 165     g_source_attach(src, NULL);
 166     g_source_unref(src);
 167     src = iohandler_get_g_source();
 168     g_source_set_name(src, "io-handler");
 169     g_source_attach(src, NULL);
 170     g_source_unref(src);
 171     return 0;
 172 }
 173
 174 static int max_priority;
 175
 176 #ifndef _WIN32
 177 static int glib_pollfds_idx;
 178 static int glib_n_poll_fds;
 179
 180 static void glib_pollfds_fill(int64_t *cur_timeout)
 181 {
 182     GMainContext *context = g_main_context_default();
 183     int timeout = 0;
 184     int64_t timeout_ns;
 185     int n;
 186
 187     g_main_context_prepare(context, &max_priority);
 188
 189     glib_pollfds_idx = gpollfds->len;
 190     n = glib_n_poll_fds;
 191     do {
 192         GPollFD *pfds;
 193         glib_n_poll_fds = n;
 194         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 195         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 196         n = g_main_context_query(context, max_priority, &timeout, pfds,
 197                                  glib_n_poll_fds);
 198     } while (n != glib_n_poll_fds);
 199
 200     if (timeout < 0) {
 201         timeout_ns = -1;
 202     } else {
 203         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 204     }
 205
 206     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 207 }
 208
 209 static void glib_pollfds_poll(void)
 210 {
 211     GMainContext *context = g_main_context_default();
 212     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 213
 214     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 215         g_main_context_dispatch(context);
 216     }
 217 }
 218
 219 #define MAX_MAIN_LOOP_SPIN (1000)
 220
 221 static int os_host_main_loop_wait(int64_t timeout)
 222 {
 223     GMainContext *context = g_main_context_default();
 224     int ret;
 225     static int spin_counter;
 226
 227     g_main_context_acquire(context);
 228
 229     glib_pollfds_fill(&timeout);
 230
 231     /* If the I/O thread is very busy or we are incorrectly busy waiting in
 232      * the I/O thread, this can lead to starvation of the BQL such that the
 233      * VCPU threads never run.  To make sure we can detect the later case,
 234      * print a message to the screen.  If we run into this condition, create
 235      * a fake timeout in order to give the VCPU threads a chance to run.
 236      */
 237     if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
 238         static bool notified;
 239
 240         if (!notified && !qtest_enabled() && !qtest_driver()) {
 241             warn_report("I/O thread spun for %d iterations",
 242                         MAX_MAIN_LOOP_SPIN);
 243             notified = true;
 244         }
 245
 246         timeout = SCALE_MS;
 247     }
 248
 249
 250     if (timeout) {
 251         spin_counter = 0;
 252     } else {
 253         spin_counter++;
 254     }
 255     qemu_mutex_unlock_iothread();
 256     replay_mutex_unlock();
 257
 258     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 259
 260     replay_mutex_lock();
 261     qemu_mutex_lock_iothread();
 262
 263     glib_pollfds_poll();
 264
 265     g_main_context_release(context);
 266
 267     return ret;
 268 }
 269 #else
 270 /***********************************************************/
 271 /* Polling handling */
 272
 273 typedef struct PollingEntry {
 274     PollingFunc *func;
 275     void *opaque;
 276     struct PollingEntry *next;
 277 } PollingEntry;
 278
 279 static PollingEntry *first_polling_entry;
 280
 281 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 282 {
 283     PollingEntry **ppe, *pe;
 284     pe = g_malloc0(sizeof(PollingEntry));
 285     pe->func = func;
 286     pe->opaque = opaque;
 287     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 288     *ppe = pe;
 289     return 0;
 290 }
 291
 292 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 293 {
 294     PollingEntry **ppe, *pe;
 295     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 296         pe = *ppe;
 297         if (pe->func == func && pe->opaque == opaque) {
 298             *ppe = pe->next;
 299             g_free(pe);
 300             break;
 301         }
 302     }
 303 }
 304
 305 /***********************************************************/
 306 /* Wait objects support */
 307 typedef struct WaitObjects {
 308     int num;
 309     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 310     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 311     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 312     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 313 } WaitObjects;
 314
 315 static WaitObjects wait_objects = {0};
 316
 317 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 318 {
 319     WaitObjects *w = &wait_objects;
 320     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 321         return -1;
 322     }
 323     w->events[w->num] = handle;
 324     w->func[w->num] = func;
 325     w->opaque[w->num] = opaque;
 326     w->revents[w->num] = 0;
 327     w->num++;
 328     return 0;
 329 }
 330
 331 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 332 {
 333     int i, found;
 334     WaitObjects *w = &wait_objects;
 335
 336     found = 0;
 337     for (i = 0; i < w->num; i++) {
 338         if (w->events[i] == handle) {
 339             found = 1;
 340         }
 341         if (found) {
 342             w->events[i] = w->events[i + 1];
 343             w->func[i] = w->func[i + 1];
 344             w->opaque[i] = w->opaque[i + 1];
 345             w->revents[i] = w->revents[i + 1];
 346         }
 347     }
 348     if (found) {
 349         w->num--;
 350     }
 351 }
 352
 353 void qemu_fd_register(int fd)
 354 {
 355     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 356                    FD_READ | FD_ACCEPT | FD_CLOSE |
 357                    FD_CONNECT | FD_WRITE | FD_OOB);
 358 }
 359
 360 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 361                         fd_set *xfds)
 362 {
 363     int nfds = -1;
 364     int i;
 365
 366     for (i = 0; i < pollfds->len; i++) {
 367         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 368         int fd = pfd->fd;
 369         int events = pfd->events;
 370         if (events & G_IO_IN) {
 371             FD_SET(fd, rfds);
 372             nfds = MAX(nfds, fd);
 373         }
 374         if (events & G_IO_OUT) {
 375             FD_SET(fd, wfds);
 376             nfds = MAX(nfds, fd);
 377         }
 378         if (events & G_IO_PRI) {
 379             FD_SET(fd, xfds);
 380             nfds = MAX(nfds, fd);
 381         }
 382     }
 383     return nfds;
 384 }
 385
 386 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 387                          fd_set *wfds, fd_set *xfds)
 388 {
 389     int i;
 390
 391     for (i = 0; i < pollfds->len; i++) {
 392         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 393         int fd = pfd->fd;
 394         int revents = 0;
 395
 396         if (FD_ISSET(fd, rfds)) {
 397             revents |= G_IO_IN;
 398         }
 399         if (FD_ISSET(fd, wfds)) {
 400             revents |= G_IO_OUT;
 401         }
 402         if (FD_ISSET(fd, xfds)) {
 403             revents |= G_IO_PRI;
 404         }
 405         pfd->revents = revents & pfd->events;
 406     }
 407 }
 408
 409 static int os_host_main_loop_wait(int64_t timeout)
 410 {
 411     GMainContext *context = g_main_context_default();
 412     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 413     int select_ret = 0;
 414     int g_poll_ret, ret, i, n_poll_fds;
 415     PollingEntry *pe;
 416     WaitObjects *w = &wait_objects;
 417     gint poll_timeout;
 418     int64_t poll_timeout_ns;
 419     static struct timeval tv0;
 420     fd_set rfds, wfds, xfds;
 421     int nfds;
 422
 423     g_main_context_acquire(context);
 424
 425     /* XXX: need to suppress polling by better using win32 events */
 426     ret = 0;
 427     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 428         ret |= pe->func(pe->opaque);
 429     }
 430     if (ret != 0) {
 431         g_main_context_release(context);
 432         return ret;
 433     }
 434
 435     FD_ZERO(&rfds);
 436     FD_ZERO(&wfds);
 437     FD_ZERO(&xfds);
 438     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 439     if (nfds >= 0) {
 440         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 441         if (select_ret != 0) {
 442             timeout = 0;
 443         }
 444         if (select_ret > 0) {
 445             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 446         }
 447     }
 448
 449     g_main_context_prepare(context, &max_priority);
 450     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 451                                       poll_fds, ARRAY_SIZE(poll_fds));
 452     g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
 453
 454     for (i = 0; i < w->num; i++) {
 455         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 456         poll_fds[n_poll_fds + i].events = G_IO_IN;
 457     }
 458
 459     if (poll_timeout < 0) {
 460         poll_timeout_ns = -1;
 461     } else {
 462         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 463     }
 464
 465     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 466
 467     qemu_mutex_unlock_iothread();
 468
 469     replay_mutex_unlock();
 470
 471     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 472
 473     replay_mutex_lock();
 474
 475     qemu_mutex_lock_iothread();
 476     if (g_poll_ret > 0) {
 477         for (i = 0; i < w->num; i++) {
 478             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 479         }
 480         for (i = 0; i < w->num; i++) {
 481             if (w->revents[i] && w->func[i]) {
 482                 w->func[i](w->opaque[i]);
 483             }
 484         }
 485     }
 486
 487     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 488         g_main_context_dispatch(context);
 489     }
 490
 491     g_main_context_release(context);
 492
 493     return select_ret || g_poll_ret;
 494 }
 495 #endif
 496
 497 void main_loop_wait(int nonblocking)
 498 {
 499     int ret;
 500     uint32_t timeout = UINT32_MAX;
 501     int64_t timeout_ns;
 502
 503     if (nonblocking) {
 504         timeout = 0;
 505     }
 506
 507     /* poll any events */
 508     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 509     /* XXX: separate device handlers from system ones */
 510     slirp_pollfds_fill(gpollfds, &timeout);
 511
 512     if (timeout == UINT32_MAX) {
 513         timeout_ns = -1;
 514     } else {
 515         timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
 516     }
 517
 518     timeout_ns = qemu_soonest_timeout(timeout_ns,
 519                                       timerlistgroup_deadline_ns(
 520                                           &main_loop_tlg));
 521
 522     ret = os_host_main_loop_wait(timeout_ns);
 523     slirp_pollfds_poll(gpollfds, (ret < 0));
 524
 525     /* CPU thread can infinitely wait for event after
 526        missing the warp */
 527     qemu_start_warp_timer();
 528     qemu_clock_run_all_timers();
 529 }
 530
 531 /* Functions to operate on the main QEMU AioContext.  */
 532
 533 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 534 {
 535     return aio_bh_new(qemu_aio_context, cb, opaque);
 536 }