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 "sysemu/qtest.h"
  30 #include "sysemu/cpus.h"
  31 #include "sysemu/replay.h"
  32 #include "qemu/main-loop.h"
  33 #include "block/aio.h"
  34 #include "qemu/error-report.h"
  35 #include "qemu/queue.h"
  36
  37 #ifndef _WIN32
  38 #include <sys/wait.h>
  39 #endif
  40
  41 #ifndef _WIN32
  42
  43 /* If we have signalfd, we mask out the signals we want to handle and then
  44  * use signalfd to listen for them.  We rely on whatever the current signal
  45  * handler is to dispatch the signals when we receive them.
  46  */
  47 static void sigfd_handler(void *opaque)
  48 {
  49     int fd = (intptr_t)opaque;
  50     struct qemu_signalfd_siginfo info;
  51     struct sigaction action;
  52     ssize_t len;
  53
  54     while (1) {
  55         do {
  56             len = read(fd, &info, sizeof(info));
  57         } while (len == -1 && errno == EINTR);
  58
  59         if (len == -1 && errno == EAGAIN) {
  60             break;
  61         }
  62
  63         if (len != sizeof(info)) {
  64             error_report("read from sigfd returned %zd: %s", len,
  65                          g_strerror(errno));
  66             return;
  67         }
  68
  69         sigaction(info.ssi_signo, NULL, &action);
  70         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  71             sigaction_invoke(&action, &info);
  72         } else if (action.sa_handler) {
  73             action.sa_handler(info.ssi_signo);
  74         }
  75     }
  76 }
  77
  78 static int qemu_signal_init(Error **errp)
  79 {
  80     int sigfd;
  81     sigset_t set;
  82
  83     /*
  84      * SIG_IPI must be blocked in the main thread and must not be caught
  85      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  86      * not catch it reliably.
  87      */
  88     sigemptyset(&set);
  89     sigaddset(&set, SIG_IPI);
  90     sigaddset(&set, SIGIO);
  91     sigaddset(&set, SIGALRM);
  92     sigaddset(&set, SIGBUS);
  93     /* SIGINT cannot be handled via signalfd, so that ^C can be used
  94      * to interrupt QEMU when it is being run under gdb.  SIGHUP and
  95      * SIGTERM are also handled asynchronously, even though it is not
  96      * strictly necessary, because they use the same handler as SIGINT.
  97      */
  98     pthread_sigmask(SIG_BLOCK, &set, NULL);
  99
 100     sigdelset(&set, SIG_IPI);
 101     sigfd = qemu_signalfd(&set);
 102     if (sigfd == -1) {
 103         error_setg_errno(errp, errno, "failed to create signalfd");
 104         return -errno;
 105     }
 106
 107     fcntl_setfl(sigfd, O_NONBLOCK);
 108
 109     qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
 110
 111     return 0;
 112 }
 113
 114 #else /* _WIN32 */
 115
 116 static int qemu_signal_init(Error **errp)
 117 {
 118     return 0;
 119 }
 120 #endif
 121
 122 static AioContext *qemu_aio_context;
 123 static QEMUBH *qemu_notify_bh;
 124
 125 static void notify_event_cb(void *opaque)
 126 {
 127     /* No need to do anything; this bottom half is only used to
 128      * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
 129      */
 130 }
 131
 132 AioContext *qemu_get_aio_context(void)
 133 {
 134     return qemu_aio_context;
 135 }
 136
 137 void qemu_notify_event(void)
 138 {
 139     if (!qemu_aio_context) {
 140         return;
 141     }
 142     qemu_bh_schedule(qemu_notify_bh);
 143 }
 144
 145 static GArray *gpollfds;
 146
 147 int qemu_init_main_loop(Error **errp)
 148 {
 149     int ret;
 150     GSource *src;
 151
 152     init_clocks(qemu_timer_notify_cb);
 153
 154     ret = qemu_signal_init(errp);
 155     if (ret) {
 156         return ret;
 157     }
 158
 159     qemu_aio_context = aio_context_new(errp);
 160     if (!qemu_aio_context) {
 161         return -EMFILE;
 162     }
 163     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 164     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 165     src = aio_get_g_source(qemu_aio_context);
 166     g_source_set_name(src, "aio-context");
 167     g_source_attach(src, NULL);
 168     g_source_unref(src);
 169     src = iohandler_get_g_source();
 170     g_source_set_name(src, "io-handler");
 171     g_source_attach(src, NULL);
 172     g_source_unref(src);
 173     return 0;
 174 }
 175
 176 static int max_priority;
 177
 178 #ifndef _WIN32
 179 static int glib_pollfds_idx;
 180 static int glib_n_poll_fds;
 181
 182 static void glib_pollfds_fill(int64_t *cur_timeout)
 183 {
 184     GMainContext *context = g_main_context_default();
 185     int timeout = 0;
 186     int64_t timeout_ns;
 187     int n;
 188
 189     g_main_context_prepare(context, &max_priority);
 190
 191     glib_pollfds_idx = gpollfds->len;
 192     n = glib_n_poll_fds;
 193     do {
 194         GPollFD *pfds;
 195         glib_n_poll_fds = n;
 196         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 197         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 198         n = g_main_context_query(context, max_priority, &timeout, pfds,
 199                                  glib_n_poll_fds);
 200     } while (n != glib_n_poll_fds);
 201
 202     if (timeout < 0) {
 203         timeout_ns = -1;
 204     } else {
 205         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 206     }
 207
 208     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 209 }
 210
 211 static void glib_pollfds_poll(void)
 212 {
 213     GMainContext *context = g_main_context_default();
 214     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 215
 216     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 217         g_main_context_dispatch(context);
 218     }
 219 }
 220
 221 #define MAX_MAIN_LOOP_SPIN (1000)
 222
 223 static int os_host_main_loop_wait(int64_t timeout)
 224 {
 225     GMainContext *context = g_main_context_default();
 226     int ret;
 227
 228     g_main_context_acquire(context);
 229
 230     glib_pollfds_fill(&timeout);
 231
 232     qemu_mutex_unlock_iothread();
 233     replay_mutex_unlock();
 234
 235     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 236
 237     replay_mutex_lock();
 238     qemu_mutex_lock_iothread();
 239
 240     glib_pollfds_poll();
 241
 242     g_main_context_release(context);
 243
 244     return ret;
 245 }
 246 #else
 247 /***********************************************************/
 248 /* Polling handling */
 249
 250 typedef struct PollingEntry {
 251     PollingFunc *func;
 252     void *opaque;
 253     struct PollingEntry *next;
 254 } PollingEntry;
 255
 256 static PollingEntry *first_polling_entry;
 257
 258 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 259 {
 260     PollingEntry **ppe, *pe;
 261     pe = g_malloc0(sizeof(PollingEntry));
 262     pe->func = func;
 263     pe->opaque = opaque;
 264     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 265     *ppe = pe;
 266     return 0;
 267 }
 268
 269 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 270 {
 271     PollingEntry **ppe, *pe;
 272     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 273         pe = *ppe;
 274         if (pe->func == func && pe->opaque == opaque) {
 275             *ppe = pe->next;
 276             g_free(pe);
 277             break;
 278         }
 279     }
 280 }
 281
 282 /***********************************************************/
 283 /* Wait objects support */
 284 typedef struct WaitObjects {
 285     int num;
 286     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 287     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 288     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 289     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 290 } WaitObjects;
 291
 292 static WaitObjects wait_objects = {0};
 293
 294 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 295 {
 296     WaitObjects *w = &wait_objects;
 297     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 298         return -1;
 299     }
 300     w->events[w->num] = handle;
 301     w->func[w->num] = func;
 302     w->opaque[w->num] = opaque;
 303     w->revents[w->num] = 0;
 304     w->num++;
 305     return 0;
 306 }
 307
 308 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 309 {
 310     int i, found;
 311     WaitObjects *w = &wait_objects;
 312
 313     found = 0;
 314     for (i = 0; i < w->num; i++) {
 315         if (w->events[i] == handle) {
 316             found = 1;
 317         }
 318         if (found) {
 319             w->events[i] = w->events[i + 1];
 320             w->func[i] = w->func[i + 1];
 321             w->opaque[i] = w->opaque[i + 1];
 322             w->revents[i] = w->revents[i + 1];
 323         }
 324     }
 325     if (found) {
 326         w->num--;
 327     }
 328 }
 329
 330 void qemu_fd_register(int fd)
 331 {
 332     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 333                    FD_READ | FD_ACCEPT | FD_CLOSE |
 334                    FD_CONNECT | FD_WRITE | FD_OOB);
 335 }
 336
 337 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 338                         fd_set *xfds)
 339 {
 340     int nfds = -1;
 341     int i;
 342
 343     for (i = 0; i < pollfds->len; i++) {
 344         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 345         int fd = pfd->fd;
 346         int events = pfd->events;
 347         if (events & G_IO_IN) {
 348             FD_SET(fd, rfds);
 349             nfds = MAX(nfds, fd);
 350         }
 351         if (events & G_IO_OUT) {
 352             FD_SET(fd, wfds);
 353             nfds = MAX(nfds, fd);
 354         }
 355         if (events & G_IO_PRI) {
 356             FD_SET(fd, xfds);
 357             nfds = MAX(nfds, fd);
 358         }
 359     }
 360     return nfds;
 361 }
 362
 363 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 364                          fd_set *wfds, fd_set *xfds)
 365 {
 366     int i;
 367
 368     for (i = 0; i < pollfds->len; i++) {
 369         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 370         int fd = pfd->fd;
 371         int revents = 0;
 372
 373         if (FD_ISSET(fd, rfds)) {
 374             revents |= G_IO_IN;
 375         }
 376         if (FD_ISSET(fd, wfds)) {
 377             revents |= G_IO_OUT;
 378         }
 379         if (FD_ISSET(fd, xfds)) {
 380             revents |= G_IO_PRI;
 381         }
 382         pfd->revents = revents & pfd->events;
 383     }
 384 }
 385
 386 static int os_host_main_loop_wait(int64_t timeout)
 387 {
 388     GMainContext *context = g_main_context_default();
 389     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 390     int select_ret = 0;
 391     int g_poll_ret, ret, i, n_poll_fds;
 392     PollingEntry *pe;
 393     WaitObjects *w = &wait_objects;
 394     gint poll_timeout;
 395     int64_t poll_timeout_ns;
 396     static struct timeval tv0;
 397     fd_set rfds, wfds, xfds;
 398     int nfds;
 399
 400     g_main_context_acquire(context);
 401
 402     /* XXX: need to suppress polling by better using win32 events */
 403     ret = 0;
 404     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 405         ret |= pe->func(pe->opaque);
 406     }
 407     if (ret != 0) {
 408         g_main_context_release(context);
 409         return ret;
 410     }
 411
 412     FD_ZERO(&rfds);
 413     FD_ZERO(&wfds);
 414     FD_ZERO(&xfds);
 415     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 416     if (nfds >= 0) {
 417         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 418         if (select_ret != 0) {
 419             timeout = 0;
 420         }
 421         if (select_ret > 0) {
 422             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 423         }
 424     }
 425
 426     g_main_context_prepare(context, &max_priority);
 427     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 428                                       poll_fds, ARRAY_SIZE(poll_fds));
 429     g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
 430
 431     for (i = 0; i < w->num; i++) {
 432         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 433         poll_fds[n_poll_fds + i].events = G_IO_IN;
 434     }
 435
 436     if (poll_timeout < 0) {
 437         poll_timeout_ns = -1;
 438     } else {
 439         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 440     }
 441
 442     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 443
 444     qemu_mutex_unlock_iothread();
 445
 446     replay_mutex_unlock();
 447
 448     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 449
 450     replay_mutex_lock();
 451
 452     qemu_mutex_lock_iothread();
 453     if (g_poll_ret > 0) {
 454         for (i = 0; i < w->num; i++) {
 455             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 456         }
 457         for (i = 0; i < w->num; i++) {
 458             if (w->revents[i] && w->func[i]) {
 459                 w->func[i](w->opaque[i]);
 460             }
 461         }
 462     }
 463
 464     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 465         g_main_context_dispatch(context);
 466     }
 467
 468     g_main_context_release(context);
 469
 470     return select_ret || g_poll_ret;
 471 }
 472 #endif
 473
 474 static NotifierList main_loop_poll_notifiers =
 475     NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
 476
 477 void main_loop_poll_add_notifier(Notifier *notify)
 478 {
 479     notifier_list_add(&main_loop_poll_notifiers, notify);
 480 }
 481
 482 void main_loop_poll_remove_notifier(Notifier *notify)
 483 {
 484     notifier_remove(notify);
 485 }
 486
 487 void main_loop_wait(int nonblocking)
 488 {
 489     MainLoopPoll mlpoll = {
 490         .state = MAIN_LOOP_POLL_FILL,
 491         .timeout = UINT32_MAX,
 492         .pollfds = gpollfds,
 493     };
 494     int ret;
 495     int64_t timeout_ns;
 496
 497     if (nonblocking) {
 498         mlpoll.timeout = 0;
 499     }
 500
 501     /* poll any events */
 502     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 503     /* XXX: separate device handlers from system ones */
 504     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 505
 506     if (mlpoll.timeout == UINT32_MAX) {
 507         timeout_ns = -1;
 508     } else {
 509         timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
 510     }
 511
 512     timeout_ns = qemu_soonest_timeout(timeout_ns,
 513                                       timerlistgroup_deadline_ns(
 514                                           &main_loop_tlg));
 515
 516     ret = os_host_main_loop_wait(timeout_ns);
 517     mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
 518     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 519
 520     /* CPU thread can infinitely wait for event after
 521        missing the warp */
 522     qemu_start_warp_timer();
 523     qemu_clock_run_all_timers();
 524 }
 525
 526 /* Functions to operate on the main QEMU AioContext.  */
 527
 528 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 529 {
 530     return aio_bh_new(qemu_aio_context, cb, opaque);
 531 }
 532
 533 /*
 534  * Functions to operate on the I/O handler AioContext.
 535  * This context runs on top of main loop. We can't reuse qemu_aio_context
 536  * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
 537  */
 538 static AioContext *iohandler_ctx;
 539
 540 static void iohandler_init(void)
 541 {
 542     if (!iohandler_ctx) {
 543         iohandler_ctx = aio_context_new(&error_abort);
 544     }
 545 }
 546
 547 AioContext *iohandler_get_aio_context(void)
 548 {
 549     iohandler_init();
 550     return iohandler_ctx;
 551 }
 552
 553 GSource *iohandler_get_g_source(void)
 554 {
 555     iohandler_init();
 556     return aio_get_g_source(iohandler_ctx);
 557 }
 558
 559 void qemu_set_fd_handler(int fd,
 560                          IOHandler *fd_read,
 561                          IOHandler *fd_write,
 562                          void *opaque)
 563 {
 564     iohandler_init();
 565     aio_set_fd_handler(iohandler_ctx, fd, false,
 566                        fd_read, fd_write, NULL, opaque);
 567 }
 568
 569 void event_notifier_set_handler(EventNotifier *e,
 570                                 EventNotifierHandler *handler)
 571 {
 572     iohandler_init();
 573     aio_set_event_notifier(iohandler_ctx, e, false,
 574                            handler, NULL);
 575 }
 576
 577 /* reaping of zombies.  right now we're not passing the status to
 578    anyone, but it would be possible to add a callback.  */
 579 #ifndef _WIN32
 580 typedef struct ChildProcessRecord {
 581     int pid;
 582     QLIST_ENTRY(ChildProcessRecord) next;
 583 } ChildProcessRecord;
 584
 585 static QLIST_HEAD(, ChildProcessRecord) child_watches =
 586     QLIST_HEAD_INITIALIZER(child_watches);
 587
 588 static QEMUBH *sigchld_bh;
 589
 590 static void sigchld_handler(int signal)
 591 {
 592     qemu_bh_schedule(sigchld_bh);
 593 }
 594
 595 static void sigchld_bh_handler(void *opaque)
 596 {
 597     ChildProcessRecord *rec, *next;
 598
 599     QLIST_FOREACH_SAFE(rec, &child_watches, next, next) {
 600         if (waitpid(rec->pid, NULL, WNOHANG) == rec->pid) {
 601             QLIST_REMOVE(rec, next);
 602             g_free(rec);
 603         }
 604     }
 605 }
 606
 607 static void qemu_init_child_watch(void)
 608 {
 609     struct sigaction act;
 610     sigchld_bh = qemu_bh_new(sigchld_bh_handler, NULL);
 611
 612     memset(&act, 0, sizeof(act));
 613     act.sa_handler = sigchld_handler;
 614     act.sa_flags = SA_NOCLDSTOP;
 615     sigaction(SIGCHLD, &act, NULL);
 616 }
 617
 618 int qemu_add_child_watch(pid_t pid)
 619 {
 620     ChildProcessRecord *rec;
 621
 622     if (!sigchld_bh) {
 623         qemu_init_child_watch();
 624     }
 625
 626     QLIST_FOREACH(rec, &child_watches, next) {
 627         if (rec->pid == pid) {
 628             return 1;
 629         }
 630     }
 631     rec = g_malloc0(sizeof(ChildProcessRecord));
 632     rec->pid = pid;
 633     QLIST_INSERT_HEAD(&child_watches, rec, next);
 634     return 0;
 635 }
 636 #endif