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
  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 void qemu_fd_register(int fd)
 183 {
 184 }
 185
 186 static void glib_pollfds_fill(int64_t *cur_timeout)
 187 {
 188     GMainContext *context = g_main_context_default();
 189     int timeout = 0;
 190     int64_t timeout_ns;
 191     int n;
 192
 193     g_main_context_prepare(context, &max_priority);
 194
 195     glib_pollfds_idx = gpollfds->len;
 196     n = glib_n_poll_fds;
 197     do {
 198         GPollFD *pfds;
 199         glib_n_poll_fds = n;
 200         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 201         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 202         n = g_main_context_query(context, max_priority, &timeout, pfds,
 203                                  glib_n_poll_fds);
 204     } while (n != glib_n_poll_fds);
 205
 206     if (timeout < 0) {
 207         timeout_ns = -1;
 208     } else {
 209         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 210     }
 211
 212     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 213 }
 214
 215 static void glib_pollfds_poll(void)
 216 {
 217     GMainContext *context = g_main_context_default();
 218     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 219
 220     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 221         g_main_context_dispatch(context);
 222     }
 223 }
 224
 225 #define MAX_MAIN_LOOP_SPIN (1000)
 226
 227 static int os_host_main_loop_wait(int64_t timeout)
 228 {
 229     GMainContext *context = g_main_context_default();
 230     int ret;
 231
 232     g_main_context_acquire(context);
 233
 234     glib_pollfds_fill(&timeout);
 235
 236     qemu_mutex_unlock_iothread();
 237     replay_mutex_unlock();
 238
 239     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 240
 241     replay_mutex_lock();
 242     qemu_mutex_lock_iothread();
 243
 244     glib_pollfds_poll();
 245
 246     g_main_context_release(context);
 247
 248     return ret;
 249 }
 250 #else
 251 /***********************************************************/
 252 /* Polling handling */
 253
 254 typedef struct PollingEntry {
 255     PollingFunc *func;
 256     void *opaque;
 257     struct PollingEntry *next;
 258 } PollingEntry;
 259
 260 static PollingEntry *first_polling_entry;
 261
 262 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 263 {
 264     PollingEntry **ppe, *pe;
 265     pe = g_malloc0(sizeof(PollingEntry));
 266     pe->func = func;
 267     pe->opaque = opaque;
 268     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 269     *ppe = pe;
 270     return 0;
 271 }
 272
 273 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 274 {
 275     PollingEntry **ppe, *pe;
 276     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 277         pe = *ppe;
 278         if (pe->func == func && pe->opaque == opaque) {
 279             *ppe = pe->next;
 280             g_free(pe);
 281             break;
 282         }
 283     }
 284 }
 285
 286 /***********************************************************/
 287 /* Wait objects support */
 288 typedef struct WaitObjects {
 289     int num;
 290     int revents[MAXIMUM_WAIT_OBJECTS + 1];
 291     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
 292     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
 293     void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
 294 } WaitObjects;
 295
 296 static WaitObjects wait_objects = {0};
 297
 298 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 299 {
 300     WaitObjects *w = &wait_objects;
 301     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 302         return -1;
 303     }
 304     w->events[w->num] = handle;
 305     w->func[w->num] = func;
 306     w->opaque[w->num] = opaque;
 307     w->revents[w->num] = 0;
 308     w->num++;
 309     return 0;
 310 }
 311
 312 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 313 {
 314     int i, found;
 315     WaitObjects *w = &wait_objects;
 316
 317     found = 0;
 318     for (i = 0; i < w->num; i++) {
 319         if (w->events[i] == handle) {
 320             found = 1;
 321         }
 322         if (found) {
 323             w->events[i] = w->events[i + 1];
 324             w->func[i] = w->func[i + 1];
 325             w->opaque[i] = w->opaque[i + 1];
 326             w->revents[i] = w->revents[i + 1];
 327         }
 328     }
 329     if (found) {
 330         w->num--;
 331     }
 332 }
 333
 334 void qemu_fd_register(int fd)
 335 {
 336     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 337                    FD_READ | FD_ACCEPT | FD_CLOSE |
 338                    FD_CONNECT | FD_WRITE | FD_OOB);
 339 }
 340
 341 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 342                         fd_set *xfds)
 343 {
 344     int nfds = -1;
 345     int i;
 346
 347     for (i = 0; i < pollfds->len; i++) {
 348         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 349         int fd = pfd->fd;
 350         int events = pfd->events;
 351         if (events & G_IO_IN) {
 352             FD_SET(fd, rfds);
 353             nfds = MAX(nfds, fd);
 354         }
 355         if (events & G_IO_OUT) {
 356             FD_SET(fd, wfds);
 357             nfds = MAX(nfds, fd);
 358         }
 359         if (events & G_IO_PRI) {
 360             FD_SET(fd, xfds);
 361             nfds = MAX(nfds, fd);
 362         }
 363     }
 364     return nfds;
 365 }
 366
 367 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 368                          fd_set *wfds, fd_set *xfds)
 369 {
 370     int i;
 371
 372     for (i = 0; i < pollfds->len; i++) {
 373         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 374         int fd = pfd->fd;
 375         int revents = 0;
 376
 377         if (FD_ISSET(fd, rfds)) {
 378             revents |= G_IO_IN;
 379         }
 380         if (FD_ISSET(fd, wfds)) {
 381             revents |= G_IO_OUT;
 382         }
 383         if (FD_ISSET(fd, xfds)) {
 384             revents |= G_IO_PRI;
 385         }
 386         pfd->revents = revents & pfd->events;
 387     }
 388 }
 389
 390 static int os_host_main_loop_wait(int64_t timeout)
 391 {
 392     GMainContext *context = g_main_context_default();
 393     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 394     int select_ret = 0;
 395     int g_poll_ret, ret, i, n_poll_fds;
 396     PollingEntry *pe;
 397     WaitObjects *w = &wait_objects;
 398     gint poll_timeout;
 399     int64_t poll_timeout_ns;
 400     static struct timeval tv0;
 401     fd_set rfds, wfds, xfds;
 402     int nfds;
 403
 404     g_main_context_acquire(context);
 405
 406     /* XXX: need to suppress polling by better using win32 events */
 407     ret = 0;
 408     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 409         ret |= pe->func(pe->opaque);
 410     }
 411     if (ret != 0) {
 412         g_main_context_release(context);
 413         return ret;
 414     }
 415
 416     FD_ZERO(&rfds);
 417     FD_ZERO(&wfds);
 418     FD_ZERO(&xfds);
 419     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 420     if (nfds >= 0) {
 421         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 422         if (select_ret != 0) {
 423             timeout = 0;
 424         }
 425         if (select_ret > 0) {
 426             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 427         }
 428     }
 429
 430     g_main_context_prepare(context, &max_priority);
 431     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 432                                       poll_fds, ARRAY_SIZE(poll_fds));
 433     g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
 434
 435     for (i = 0; i < w->num; i++) {
 436         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 437         poll_fds[n_poll_fds + i].events = G_IO_IN;
 438     }
 439
 440     if (poll_timeout < 0) {
 441         poll_timeout_ns = -1;
 442     } else {
 443         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 444     }
 445
 446     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 447
 448     qemu_mutex_unlock_iothread();
 449
 450     replay_mutex_unlock();
 451
 452     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 453
 454     replay_mutex_lock();
 455
 456     qemu_mutex_lock_iothread();
 457     if (g_poll_ret > 0) {
 458         for (i = 0; i < w->num; i++) {
 459             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 460         }
 461         for (i = 0; i < w->num; i++) {
 462             if (w->revents[i] && w->func[i]) {
 463                 w->func[i](w->opaque[i]);
 464             }
 465         }
 466     }
 467
 468     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 469         g_main_context_dispatch(context);
 470     }
 471
 472     g_main_context_release(context);
 473
 474     return select_ret || g_poll_ret;
 475 }
 476 #endif
 477
 478 static NotifierList main_loop_poll_notifiers =
 479     NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
 480
 481 void main_loop_poll_add_notifier(Notifier *notify)
 482 {
 483     notifier_list_add(&main_loop_poll_notifiers, notify);
 484 }
 485
 486 void main_loop_poll_remove_notifier(Notifier *notify)
 487 {
 488     notifier_remove(notify);
 489 }
 490
 491 void main_loop_wait(int nonblocking)
 492 {
 493     MainLoopPoll mlpoll = {
 494         .state = MAIN_LOOP_POLL_FILL,
 495         .timeout = UINT32_MAX,
 496         .pollfds = gpollfds,
 497     };
 498     int ret;
 499     int64_t timeout_ns;
 500
 501     if (nonblocking) {
 502         mlpoll.timeout = 0;
 503     }
 504
 505     /* poll any events */
 506     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 507     /* XXX: separate device handlers from system ones */
 508     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 509
 510     if (mlpoll.timeout == UINT32_MAX) {
 511         timeout_ns = -1;
 512     } else {
 513         timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
 514     }
 515
 516     timeout_ns = qemu_soonest_timeout(timeout_ns,
 517                                       timerlistgroup_deadline_ns(
 518                                           &main_loop_tlg));
 519
 520     ret = os_host_main_loop_wait(timeout_ns);
 521     mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
 522     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 523
 524     if (icount_enabled()) {
 525         /*
 526          * CPU thread can infinitely wait for event after
 527          * missing the warp
 528          */
 529         icount_start_warp_timer();
 530     }
 531     qemu_clock_run_all_timers();
 532 }
 533
 534 /* Functions to operate on the main QEMU AioContext.  */
 535
 536 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
 537 {
 538     return aio_bh_new(qemu_aio_context, cb, opaque);
 539 }
 540
 541 /*
 542  * Functions to operate on the I/O handler AioContext.
 543  * This context runs on top of main loop. We can't reuse qemu_aio_context
 544  * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
 545  */
 546 static AioContext *iohandler_ctx;
 547
 548 static void iohandler_init(void)
 549 {
 550     if (!iohandler_ctx) {
 551         iohandler_ctx = aio_context_new(&error_abort);
 552     }
 553 }
 554
 555 AioContext *iohandler_get_aio_context(void)
 556 {
 557     iohandler_init();
 558     return iohandler_ctx;
 559 }
 560
 561 GSource *iohandler_get_g_source(void)
 562 {
 563     iohandler_init();
 564     return aio_get_g_source(iohandler_ctx);
 565 }
 566
 567 void qemu_set_fd_handler(int fd,
 568                          IOHandler *fd_read,
 569                          IOHandler *fd_write,
 570                          void *opaque)
 571 {
 572     iohandler_init();
 573     aio_set_fd_handler(iohandler_ctx, fd, false,
 574                        fd_read, fd_write, NULL, opaque);
 575 }
 576
 577 void event_notifier_set_handler(EventNotifier *e,
 578                                 EventNotifierHandler *handler)
 579 {
 580     iohandler_init();
 581     aio_set_event_notifier(iohandler_ctx, e, false,
 582                            handler, NULL);
 583 }
 584
 585 /* reaping of zombies.  right now we're not passing the status to
 586    anyone, but it would be possible to add a callback.  */
 587 #ifndef _WIN32
 588 typedef struct ChildProcessRecord {
 589     int pid;
 590     QLIST_ENTRY(ChildProcessRecord) next;
 591 } ChildProcessRecord;
 592
 593 static QLIST_HEAD(, ChildProcessRecord) child_watches =
 594     QLIST_HEAD_INITIALIZER(child_watches);
 595
 596 static QEMUBH *sigchld_bh;
 597
 598 static void sigchld_handler(int signal)
 599 {
 600     qemu_bh_schedule(sigchld_bh);
 601 }
 602
 603 static void sigchld_bh_handler(void *opaque)
 604 {
 605     ChildProcessRecord *rec, *next;
 606
 607     QLIST_FOREACH_SAFE(rec, &child_watches, next, next) {
 608         if (waitpid(rec->pid, NULL, WNOHANG) == rec->pid) {
 609             QLIST_REMOVE(rec, next);
 610             g_free(rec);
 611         }
 612     }
 613 }
 614
 615 static void qemu_init_child_watch(void)
 616 {
 617     struct sigaction act;
 618     sigchld_bh = qemu_bh_new(sigchld_bh_handler, NULL);
 619
 620     memset(&act, 0, sizeof(act));
 621     act.sa_handler = sigchld_handler;
 622     act.sa_flags = SA_NOCLDSTOP;
 623     sigaction(SIGCHLD, &act, NULL);
 624 }
 625
 626 int qemu_add_child_watch(pid_t pid)
 627 {
 628     ChildProcessRecord *rec;
 629
 630     if (!sigchld_bh) {
 631         qemu_init_child_watch();
 632     }
 633
 634     QLIST_FOREACH(rec, &child_watches, next) {
 635         if (rec->pid == pid) {
 636             return 1;
 637         }
 638     }
 639     rec = g_malloc0(sizeof(ChildProcessRecord));
 640     rec->pid = pid;
 641     QLIST_INSERT_HEAD(&child_watches, rec, next);
 642     return 0;
 643 }
 644 #endif