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/cpu-timers.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 #include "qemu/compiler.h"
  37
  38 #ifndef _WIN32
  39 #include <sys/wait.h>
  40 #endif
  41
  42 #ifndef _WIN32
  43
  44 /* If we have signalfd, we mask out the signals we want to handle and then
  45  * use signalfd to listen for them.  We rely on whatever the current signal
  46  * handler is to dispatch the signals when we receive them.
  47  */
  48 /*
  49  * Disable CFI checks.
  50  * We are going to call a signal hander directly. Such handler may or may not
  51  * have been defined in our binary, so there's no guarantee that the pointer
  52  * used to set the handler is a cfi-valid pointer. Since the handlers are
  53  * stored in kernel memory, changing the handler to an attacker-defined
  54  * function requires being able to call a sigaction() syscall,
  55  * which is not as easy as overwriting a pointer in memory.
  56  */
  57 QEMU_DISABLE_CFI
  58 static void sigfd_handler(void *opaque)
  59 {
  60     int fd = (intptr_t)opaque;
  61     struct qemu_signalfd_siginfo info;
  62     struct sigaction action;
  63     ssize_t len;
  64
  65     while (1) {
  66         do {
  67             len = read(fd, &info, sizeof(info));
  68         } while (len == -1 && errno == EINTR);
  69
  70         if (len == -1 && errno == EAGAIN) {
  71             break;
  72         }
  73
  74         if (len != sizeof(info)) {
  75             error_report("read from sigfd returned %zd: %s", len,
  76                          g_strerror(errno));
  77             return;
  78         }
  79
  80         sigaction(info.ssi_signo, NULL, &action);
  81         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  82             sigaction_invoke(&action, &info);
  83         } else if (action.sa_handler) {
  84             action.sa_handler(info.ssi_signo);
  85         }
  86     }
  87 }
  88
  89 static int qemu_signal_init(Error **errp)
  90 {
  91     int sigfd;
  92     sigset_t set;
  93
  94     /*
  95      * SIG_IPI must be blocked in the main thread and must not be caught
  96      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  97      * not catch it reliably.
  98      */
  99     sigemptyset(&set);
 100     sigaddset(&set, SIG_IPI);
 101     sigaddset(&set, SIGIO);
 102     sigaddset(&set, SIGALRM);
 103     sigaddset(&set, SIGBUS);
 104     /* SIGINT cannot be handled via signalfd, so that ^C can be used
 105      * to interrupt QEMU when it is being run under gdb.  SIGHUP and
 106      * SIGTERM are also handled asynchronously, even though it is not
 107      * strictly necessary, because they use the same handler as SIGINT.
 108      */
 109     pthread_sigmask(SIG_BLOCK, &set, NULL);
 110
 111     sigdelset(&set, SIG_IPI);
 112     sigfd = qemu_signalfd(&set);
 113     if (sigfd == -1) {
 114         error_setg_errno(errp, errno, "failed to create signalfd");
 115         return -errno;
 116     }
 117
 118     fcntl_setfl(sigfd, O_NONBLOCK);
 119
 120     qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
 121
 122     return 0;
 123 }
 124
 125 #else /* _WIN32 */
 126
 127 static int qemu_signal_init(Error **errp)
 128 {
 129     return 0;
 130 }
 131 #endif
 132
 133 static AioContext *qemu_aio_context;
 134 static QEMUBH *qemu_notify_bh;
 135
 136 static void notify_event_cb(void *opaque)
 137 {
 138     /* No need to do anything; this bottom half is only used to
 139      * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
 140      */
 141 }
 142
 143 AioContext *qemu_get_aio_context(void)
 144 {
 145     return qemu_aio_context;
 146 }
 147
 148 void qemu_notify_event(void)
 149 {
 150     if (!qemu_aio_context) {
 151         return;
 152     }
 153     qemu_bh_schedule(qemu_notify_bh);
 154 }
 155
 156 static GArray *gpollfds;
 157
 158 int qemu_init_main_loop(Error **errp)
 159 {
 160     int ret;
 161     GSource *src;
 162
 163     init_clocks(qemu_timer_notify_cb);
 164
 165     ret = qemu_signal_init(errp);
 166     if (ret) {
 167         return ret;
 168     }
 169
 170     qemu_aio_context = aio_context_new(errp);
 171     if (!qemu_aio_context) {
 172         return -EMFILE;
 173     }
 174     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 175     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 176     src = aio_get_g_source(qemu_aio_context);
 177     g_source_set_name(src, "aio-context");
 178     g_source_attach(src, NULL);
 179     g_source_unref(src);
 180     src = iohandler_get_g_source();
 181     g_source_set_name(src, "io-handler");
 182     g_source_attach(src, NULL);
 183     g_source_unref(src);
 184     return 0;
 185 }
 186
 187 static int max_priority;
 188
 189 #ifndef _WIN32
 190 static int glib_pollfds_idx;
 191 static int glib_n_poll_fds;
 192
 193 void qemu_fd_register(int fd)
 194 {
 195 }
 196
 197 static void glib_pollfds_fill(int64_t *cur_timeout)
 198 {
 199     GMainContext *context = g_main_context_default();
 200     int timeout = 0;
 201     int64_t timeout_ns;
 202     int n;
 203
 204     g_main_context_prepare(context, &max_priority);
 205
 206     glib_pollfds_idx = gpollfds->len;
 207     n = glib_n_poll_fds;
 208     do {
 209         GPollFD *pfds;
 210         glib_n_poll_fds = n;
 211         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 212         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 213         n = g_main_context_query(context, max_priority, &timeout, pfds,
 214                                  glib_n_poll_fds);
 215     } while (n != glib_n_poll_fds);
 216
 217     if (timeout < 0) {
 218         timeout_ns = -1;
 219     } else {
 220         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 221     }
 222
 223     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 224 }
 225
 226 static void glib_pollfds_poll(void)
 227 {
 228     GMainContext *context = g_main_context_default();
 229     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 230
 231     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 232         g_main_context_dispatch(context);
 233     }
 234 }
 235
 236 #define MAX_MAIN_LOOP_SPIN (1000)
 237
 238 static int os_host_main_loop_wait(int64_t timeout)
 239 {
 240     GMainContext *context = g_main_context_default();
 241     int ret;
 242
 243     g_main_context_acquire(context);
 244
 245     glib_pollfds_fill(&timeout);
 246
 247     qemu_mutex_unlock_iothread();
 248     replay_mutex_unlock();
 249
 250     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 251
 252     replay_mutex_lock();
 253     qemu_mutex_lock_iothread();
 254
 255     glib_pollfds_poll();
 256
 257     g_main_context_release(context);
 258
 259     return ret;
 260 }
 261 #else
 262 /***********************************************************/
 263 /* Polling handling */
 264
 265 typedef struct PollingEntry {
 266     PollingFunc *func;
 267     void *opaque;
 268     struct PollingEntry *next;
 269 } PollingEntry;
 270
 271 static PollingEntry *first_polling_entry;
 272
 273 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 274 {
 275     PollingEntry **ppe, *pe;
 276     pe = g_malloc0(sizeof(PollingEntry));
 277     pe->func = func;
 278     pe->opaque = opaque;
 279     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 280     *ppe = pe;
 281     return 0;
 282 }
 283
 284 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 285 {
 286     PollingEntry **ppe, *pe;
 287     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 288         pe = *ppe;
 289         if (pe->func == func && pe->opaque == opaque) {
 290             *ppe = pe->next;
 291             g_free(pe);
 292             break;
 293         }
 294     }
 295 }
 296
 297 /***********************************************************/
 298 /* Wait objects support */
 299 typedef struct WaitObjects {
 300     int num;
 301     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 302     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 303     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 304     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 305 } WaitObjects;
 306
 307 static WaitObjects wait_objects = {0};
 308
 309 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 310 {
 311     WaitObjects *w = &wait_objects;
 312     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 313         return -1;
 314     }
 315     w->events[w->num] = handle;
 316     w->func[w->num] = func;
 317     w->opaque[w->num] = opaque;
 318     w->revents[w->num] = 0;
 319     w->num++;
 320     return 0;
 321 }
 322
 323 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 324 {
 325     int i, found;
 326     WaitObjects *w = &wait_objects;
 327
 328     found = 0;
 329     for (i = 0; i < w->num; i++) {
 330         if (w->events[i] == handle) {
 331             found = 1;
 332         }
 333         if (found) {
 334             w->events[i] = w->events[i + 1];
 335             w->func[i] = w->func[i + 1];
 336             w->opaque[i] = w->opaque[i + 1];
 337             w->revents[i] = w->revents[i + 1];
 338         }
 339     }
 340     if (found) {
 341         w->num--;
 342     }
 343 }
 344
 345 void qemu_fd_register(int fd)
 346 {
 347     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 348                    FD_READ | FD_ACCEPT | FD_CLOSE |
 349                    FD_CONNECT | FD_WRITE | FD_OOB);
 350 }
 351
 352 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 353                         fd_set *xfds)
 354 {
 355     int nfds = -1;
 356     int i;
 357
 358     for (i = 0; i < pollfds->len; i++) {
 359         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 360         int fd = pfd->fd;
 361         int events = pfd->events;
 362         if (events & G_IO_IN) {
 363             FD_SET(fd, rfds);
 364             nfds = MAX(nfds, fd);
 365         }
 366         if (events & G_IO_OUT) {
 367             FD_SET(fd, wfds);
 368             nfds = MAX(nfds, fd);
 369         }
 370         if (events & G_IO_PRI) {
 371             FD_SET(fd, xfds);
 372             nfds = MAX(nfds, fd);
 373         }
 374     }
 375     return nfds;
 376 }
 377
 378 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 379                          fd_set *wfds, fd_set *xfds)
 380 {
 381     int i;
 382
 383     for (i = 0; i < pollfds->len; i++) {
 384         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 385         int fd = pfd->fd;
 386         int revents = 0;
 387
 388         if (FD_ISSET(fd, rfds)) {
 389             revents |= G_IO_IN;
 390         }
 391         if (FD_ISSET(fd, wfds)) {
 392             revents |= G_IO_OUT;
 393         }
 394         if (FD_ISSET(fd, xfds)) {
 395             revents |= G_IO_PRI;
 396         }
 397         pfd->revents = revents & pfd->events;
 398     }
 399 }
 400
 401 static int os_host_main_loop_wait(int64_t timeout)
 402 {
 403     GMainContext *context = g_main_context_default();
 404     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 405     int select_ret = 0;
 406     int g_poll_ret, ret, i, n_poll_fds;
 407     PollingEntry *pe;
 408     WaitObjects *w = &wait_objects;
 409     gint poll_timeout;
 410     int64_t poll_timeout_ns;
 411     static struct timeval tv0;
 412     fd_set rfds, wfds, xfds;
 413     int nfds;
 414
 415     g_main_context_acquire(context);
 416
 417     /* XXX: need to suppress polling by better using win32 events */
 418     ret = 0;
 419     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 420         ret |= pe->func(pe->opaque);
 421     }
 422     if (ret != 0) {
 423         g_main_context_release(context);
 424         return ret;
 425     }
 426
 427     FD_ZERO(&rfds);
 428     FD_ZERO(&wfds);
 429     FD_ZERO(&xfds);
 430     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 431     if (nfds >= 0) {
 432         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 433         if (select_ret != 0) {
 434             timeout = 0;
 435         }
 436         if (select_ret > 0) {
 437             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 438         }
 439     }
 440
 441     g_main_context_prepare(context, &max_priority);
 442     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 443                                       poll_fds, ARRAY_SIZE(poll_fds));
 444     g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
 445
 446     for (i = 0; i < w->num; i++) {
 447         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 448         poll_fds[n_poll_fds + i].events = G_IO_IN;
 449     }
 450
 451     if (poll_timeout < 0) {
 452         poll_timeout_ns = -1;
 453     } else {
 454         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 455     }
 456
 457     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 458
 459     qemu_mutex_unlock_iothread();
 460
 461     replay_mutex_unlock();
 462
 463     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 464
 465     replay_mutex_lock();
 466
 467     qemu_mutex_lock_iothread();
 468     if (g_poll_ret > 0) {
 469         for (i = 0; i < w->num; i++) {
 470             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 471         }
 472         for (i = 0; i < w->num; i++) {
 473             if (w->revents[i] && w->func[i]) {
 474                 w->func[i](w->opaque[i]);
 475             }
 476         }
 477     }
 478
 479     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 480         g_main_context_dispatch(context);
 481     }
 482
 483     g_main_context_release(context);
 484
 485     return select_ret || g_poll_ret;
 486 }
 487 #endif
 488
 489 static NotifierList main_loop_poll_notifiers =
 490     NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
 491
 492 void main_loop_poll_add_notifier(Notifier *notify)
 493 {
 494     notifier_list_add(&main_loop_poll_notifiers, notify);
 495 }
 496
 497 void main_loop_poll_remove_notifier(Notifier *notify)
 498 {
 499     notifier_remove(notify);
 500 }
 501
 502 void main_loop_wait(int nonblocking)
 503 {
 504     MainLoopPoll mlpoll = {
 505         .state = MAIN_LOOP_POLL_FILL,
 506         .timeout = UINT32_MAX,
 507         .pollfds = gpollfds,
 508     };
 509     int ret;
 510     int64_t timeout_ns;
 511
 512     if (nonblocking) {
 513         mlpoll.timeout = 0;
 514     }
 515
 516     /* poll any events */
 517     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 518     /* XXX: separate device handlers from system ones */
 519     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 520
 521     if (mlpoll.timeout == UINT32_MAX) {
 522         timeout_ns = -1;
 523     } else {
 524         timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
 525     }
 526
 527     timeout_ns = qemu_soonest_timeout(timeout_ns,
 528                                       timerlistgroup_deadline_ns(
 529                                           &main_loop_tlg));
 530
 531     ret = os_host_main_loop_wait(timeout_ns);
 532     mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
 533     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 534
 535     if (icount_enabled()) {
 536         /*
 537          * CPU thread can infinitely wait for event after
 538          * missing the warp
 539          */
 540         icount_start_warp_timer();
 541     }
 542     qemu_clock_run_all_timers();
 543 }
 544
 545 /* Functions to operate on the main QEMU AioContext.  */
 546
 547 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 548 {
 549     return aio_bh_new(qemu_aio_context, cb, opaque);
 550 }
 551
 552 /*
 553  * Functions to operate on the I/O handler AioContext.
 554  * This context runs on top of main loop. We can't reuse qemu_aio_context
 555  * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
 556  */
 557 static AioContext *iohandler_ctx;
 558
 559 static void iohandler_init(void)
 560 {
 561     if (!iohandler_ctx) {
 562         iohandler_ctx = aio_context_new(&error_abort);
 563     }
 564 }
 565
 566 AioContext *iohandler_get_aio_context(void)
 567 {
 568     iohandler_init();
 569     return iohandler_ctx;
 570 }
 571
 572 GSource *iohandler_get_g_source(void)
 573 {
 574     iohandler_init();
 575     return aio_get_g_source(iohandler_ctx);
 576 }
 577
 578 void qemu_set_fd_handler(int fd,
 579                          IOHandler *fd_read,
 580                          IOHandler *fd_write,
 581                          void *opaque)
 582 {
 583     iohandler_init();
 584     aio_set_fd_handler(iohandler_ctx, fd, false,
 585                        fd_read, fd_write, NULL, opaque);
 586 }
 587
 588 void event_notifier_set_handler(EventNotifier *e,
 589                                 EventNotifierHandler *handler)
 590 {
 591     iohandler_init();
 592     aio_set_event_notifier(iohandler_ctx, e, false,
 593                            handler, NULL);
 594 }
 595
 596 /* reaping of zombies.  right now we're not passing the status to
 597    anyone, but it would be possible to add a callback.  */
 598 #ifndef _WIN32
 599 typedef struct ChildProcessRecord {
 600     int pid;
 601     QLIST_ENTRY(ChildProcessRecord) next;
 602 } ChildProcessRecord;
 603
 604 static QLIST_HEAD(, ChildProcessRecord) child_watches =
 605     QLIST_HEAD_INITIALIZER(child_watches);
 606
 607 static QEMUBH *sigchld_bh;
 608
 609 static void sigchld_handler(int signal)
 610 {
 611     qemu_bh_schedule(sigchld_bh);
 612 }
 613
 614 static void sigchld_bh_handler(void *opaque)
 615 {
 616     ChildProcessRecord *rec, *next;
 617
 618     QLIST_FOREACH_SAFE(rec, &child_watches, next, next) {
 619         if (waitpid(rec->pid, NULL, WNOHANG) == rec->pid) {
 620             QLIST_REMOVE(rec, next);
 621             g_free(rec);
 622         }
 623     }
 624 }
 625
 626 static void qemu_init_child_watch(void)
 627 {
 628     struct sigaction act;
 629     sigchld_bh = qemu_bh_new(sigchld_bh_handler, NULL);
 630
 631     memset(&act, 0, sizeof(act));
 632     act.sa_handler = sigchld_handler;
 633     act.sa_flags = SA_NOCLDSTOP;
 634     sigaction(SIGCHLD, &act, NULL);
 635 }
 636
 637 int qemu_add_child_watch(pid_t pid)
 638 {
 639     ChildProcessRecord *rec;
 640
 641     if (!sigchld_bh) {
 642         qemu_init_child_watch();
 643     }
 644
 645     QLIST_FOREACH(rec, &child_watches, next) {
 646         if (rec->pid == pid) {
 647             return 1;
 648         }
 649     }
 650     rec = g_malloc0(sizeof(ChildProcessRecord));
 651     rec->pid = pid;
 652     QLIST_INSERT_HEAD(&child_watches, rec, next);
 653     return 0;
 654 }
 655 #endif