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/cpu-timers.h"
  30 #include "sysemu/replay.h"
  31 #include "qemu/main-loop.h"
  32 #include "block/aio.h"
  33 #include "block/thread-pool.h"
  34 #include "qemu/error-report.h"
  35 #include "qemu/queue.h"
  36 #include "qemu/compiler.h"
  37 #include "qom/object.h"
  38
  39 #ifndef _WIN32
  40 #include <sys/wait.h>
  41 #endif
  42
  43 #ifndef _WIN32
  44
  45 /* If we have signalfd, we mask out the signals we want to handle and then
  46  * use signalfd to listen for them.  We rely on whatever the current signal
  47  * handler is to dispatch the signals when we receive them.
  48  */
  49 /*
  50  * Disable CFI checks.
  51  * We are going to call a signal hander directly. Such handler may or may not
  52  * have been defined in our binary, so there's no guarantee that the pointer
  53  * used to set the handler is a cfi-valid pointer. Since the handlers are
  54  * stored in kernel memory, changing the handler to an attacker-defined
  55  * function requires being able to call a sigaction() syscall,
  56  * which is not as easy as overwriting a pointer in memory.
  57  */
  58 QEMU_DISABLE_CFI
  59 static void sigfd_handler(void *opaque)
  60 {
  61     int fd = (intptr_t)opaque;
  62     struct qemu_signalfd_siginfo info;
  63     struct sigaction action;
  64     ssize_t len;
  65
  66     while (1) {
  67         do {
  68             len = read(fd, &info, sizeof(info));
  69         } while (len == -1 && errno == EINTR);
  70
  71         if (len == -1 && errno == EAGAIN) {
  72             break;
  73         }
  74
  75         if (len != sizeof(info)) {
  76             error_report("read from sigfd returned %zd: %s", len,
  77                          g_strerror(errno));
  78             return;
  79         }
  80
  81         sigaction(info.ssi_signo, NULL, &action);
  82         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
  83             sigaction_invoke(&action, &info);
  84         } else if (action.sa_handler) {
  85             action.sa_handler(info.ssi_signo);
  86         }
  87     }
  88 }
  89
  90 static int qemu_signal_init(Error **errp)
  91 {
  92     int sigfd;
  93     sigset_t set;
  94
  95     /*
  96      * SIG_IPI must be blocked in the main thread and must not be caught
  97      * by sigwait() in the signal thread. Otherwise, the cpu thread will
  98      * not catch it reliably.
  99      */
 100     sigemptyset(&set);
 101     sigaddset(&set, SIG_IPI);
 102     sigaddset(&set, SIGIO);
 103     sigaddset(&set, SIGALRM);
 104     sigaddset(&set, SIGBUS);
 105     /* SIGINT cannot be handled via signalfd, so that ^C can be used
 106      * to interrupt QEMU when it is being run under gdb.  SIGHUP and
 107      * SIGTERM are also handled asynchronously, even though it is not
 108      * strictly necessary, because they use the same handler as SIGINT.
 109      */
 110     pthread_sigmask(SIG_BLOCK, &set, NULL);
 111
 112     sigdelset(&set, SIG_IPI);
 113     sigfd = qemu_signalfd(&set);
 114     if (sigfd == -1) {
 115         error_setg_errno(errp, errno, "failed to create signalfd");
 116         return -errno;
 117     }
 118
 119     g_unix_set_fd_nonblocking(sigfd, true, NULL);
 120
 121     qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
 122
 123     return 0;
 124 }
 125
 126 #else /* _WIN32 */
 127
 128 static int qemu_signal_init(Error **errp)
 129 {
 130     return 0;
 131 }
 132 #endif
 133
 134 static AioContext *qemu_aio_context;
 135 static QEMUBH *qemu_notify_bh;
 136
 137 static void notify_event_cb(void *opaque)
 138 {
 139     /* No need to do anything; this bottom half is only used to
 140      * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
 141      */
 142 }
 143
 144 AioContext *qemu_get_aio_context(void)
 145 {
 146     return qemu_aio_context;
 147 }
 148
 149 void qemu_notify_event(void)
 150 {
 151     if (!qemu_aio_context) {
 152         return;
 153     }
 154     qemu_bh_schedule(qemu_notify_bh);
 155 }
 156
 157 static GArray *gpollfds;
 158
 159 int qemu_init_main_loop(Error **errp)
 160 {
 161     int ret;
 162     GSource *src;
 163
 164     init_clocks(qemu_timer_notify_cb);
 165
 166     ret = qemu_signal_init(errp);
 167     if (ret) {
 168         return ret;
 169     }
 170
 171     qemu_aio_context = aio_context_new(errp);
 172     if (!qemu_aio_context) {
 173         return -EMFILE;
 174     }
 175     qemu_set_current_aio_context(qemu_aio_context);
 176     qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
 177     gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
 178     src = aio_get_g_source(qemu_aio_context);
 179     g_source_set_name(src, "aio-context");
 180     g_source_attach(src, NULL);
 181     g_source_unref(src);
 182     src = iohandler_get_g_source();
 183     g_source_set_name(src, "io-handler");
 184     g_source_attach(src, NULL);
 185     g_source_unref(src);
 186     return 0;
 187 }
 188
 189 static void main_loop_update_params(EventLoopBase *base, Error **errp)
 190 {
 191     ERRP_GUARD();
 192
 193     if (!qemu_aio_context) {
 194         error_setg(errp, "qemu aio context not ready");
 195         return;
 196     }
 197
 198     aio_context_set_aio_params(qemu_aio_context, base->aio_max_batch, errp);
 199     if (*errp) {
 200         return;
 201     }
 202
 203     aio_context_set_thread_pool_params(qemu_aio_context, base->thread_pool_min,
 204                                        base->thread_pool_max, errp);
 205 }
 206
 207 MainLoop *mloop;
 208
 209 static void main_loop_init(EventLoopBase *base, Error **errp)
 210 {
 211     MainLoop *m = MAIN_LOOP(base);
 212
 213     if (mloop) {
 214         error_setg(errp, "only one main-loop instance allowed");
 215         return;
 216     }
 217
 218     main_loop_update_params(base, errp);
 219
 220     mloop = m;
 221     return;
 222 }
 223
 224 static bool main_loop_can_be_deleted(EventLoopBase *base)
 225 {
 226     return false;
 227 }
 228
 229 static void main_loop_class_init(ObjectClass *oc, void *class_data)
 230 {
 231     EventLoopBaseClass *bc = EVENT_LOOP_BASE_CLASS(oc);
 232
 233     bc->init = main_loop_init;
 234     bc->update_params = main_loop_update_params;
 235     bc->can_be_deleted = main_loop_can_be_deleted;
 236 }
 237
 238 static const TypeInfo main_loop_info = {
 239     .name = TYPE_MAIN_LOOP,
 240     .parent = TYPE_EVENT_LOOP_BASE,
 241     .class_init = main_loop_class_init,
 242     .instance_size = sizeof(MainLoop),
 243 };
 244
 245 static void main_loop_register_types(void)
 246 {
 247     type_register_static(&main_loop_info);
 248 }
 249
 250 type_init(main_loop_register_types)
 251
 252 static int max_priority;
 253
 254 #ifndef _WIN32
 255 static int glib_pollfds_idx;
 256 static int glib_n_poll_fds;
 257
 258 void qemu_fd_register(int fd)
 259 {
 260 }
 261
 262 static void glib_pollfds_fill(int64_t *cur_timeout)
 263 {
 264     GMainContext *context = g_main_context_default();
 265     int timeout = 0;
 266     int64_t timeout_ns;
 267     int n;
 268
 269     g_main_context_prepare(context, &max_priority);
 270
 271     glib_pollfds_idx = gpollfds->len;
 272     n = glib_n_poll_fds;
 273     do {
 274         GPollFD *pfds;
 275         glib_n_poll_fds = n;
 276         g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
 277         pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 278         n = g_main_context_query(context, max_priority, &timeout, pfds,
 279                                  glib_n_poll_fds);
 280     } while (n != glib_n_poll_fds);
 281
 282     if (timeout < 0) {
 283         timeout_ns = -1;
 284     } else {
 285         timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
 286     }
 287
 288     *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
 289 }
 290
 291 static void glib_pollfds_poll(void)
 292 {
 293     GMainContext *context = g_main_context_default();
 294     GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
 295
 296     if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
 297         g_main_context_dispatch(context);
 298     }
 299 }
 300
 301 #define MAX_MAIN_LOOP_SPIN (1000)
 302
 303 static int os_host_main_loop_wait(int64_t timeout)
 304 {
 305     GMainContext *context = g_main_context_default();
 306     int ret;
 307
 308     g_main_context_acquire(context);
 309
 310     glib_pollfds_fill(&timeout);
 311
 312     qemu_mutex_unlock_iothread();
 313     replay_mutex_unlock();
 314
 315     ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
 316
 317     replay_mutex_lock();
 318     qemu_mutex_lock_iothread();
 319
 320     glib_pollfds_poll();
 321
 322     g_main_context_release(context);
 323
 324     return ret;
 325 }
 326 #else
 327 /***********************************************************/
 328 /* Polling handling */
 329
 330 typedef struct PollingEntry {
 331     PollingFunc *func;
 332     void *opaque;
 333     struct PollingEntry *next;
 334 } PollingEntry;
 335
 336 static PollingEntry *first_polling_entry;
 337
 338 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
 339 {
 340     PollingEntry **ppe, *pe;
 341     pe = g_new0(PollingEntry, 1);
 342     pe->func = func;
 343     pe->opaque = opaque;
 344     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
 345     *ppe = pe;
 346     return 0;
 347 }
 348
 349 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
 350 {
 351     PollingEntry **ppe, *pe;
 352     for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
 353         pe = *ppe;
 354         if (pe->func == func && pe->opaque == opaque) {
 355             *ppe = pe->next;
 356             g_free(pe);
 357             break;
 358         }
 359     }
 360 }
 361
 362 /***********************************************************/
 363 /* Wait objects support */
 364 typedef struct WaitObjects {
 365     int num;
 366     int revents[MAXIMUM_WAIT_OBJECTS];
 367     HANDLE events[MAXIMUM_WAIT_OBJECTS];
 368     WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS];
 369     void *opaque[MAXIMUM_WAIT_OBJECTS];
 370 } WaitObjects;
 371
 372 static WaitObjects wait_objects = {0};
 373
 374 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 375 {
 376     int i;
 377     WaitObjects *w = &wait_objects;
 378
 379     if (w->num >= MAXIMUM_WAIT_OBJECTS) {
 380         return -1;
 381     }
 382
 383     for (i = 0; i < w->num; i++) {
 384         /* check if the same handle is added twice */
 385         if (w->events[i] == handle) {
 386             return -1;
 387         }
 388     }
 389
 390     w->events[w->num] = handle;
 391     w->func[w->num] = func;
 392     w->opaque[w->num] = opaque;
 393     w->revents[w->num] = 0;
 394     w->num++;
 395     return 0;
 396 }
 397
 398 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
 399 {
 400     int i, found;
 401     WaitObjects *w = &wait_objects;
 402
 403     found = 0;
 404     for (i = 0; i < w->num; i++) {
 405         if (w->events[i] == handle) {
 406             found = 1;
 407         }
 408         if (found && i < (MAXIMUM_WAIT_OBJECTS - 1)) {
 409             w->events[i] = w->events[i + 1];
 410             w->func[i] = w->func[i + 1];
 411             w->opaque[i] = w->opaque[i + 1];
 412             w->revents[i] = w->revents[i + 1];
 413         }
 414     }
 415     if (found) {
 416         w->num--;
 417     }
 418 }
 419
 420 void qemu_fd_register(int fd)
 421 {
 422     WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
 423                    FD_READ | FD_ACCEPT | FD_CLOSE |
 424                    FD_CONNECT | FD_WRITE | FD_OOB);
 425 }
 426
 427 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
 428                         fd_set *xfds)
 429 {
 430     int nfds = -1;
 431     int i;
 432
 433     for (i = 0; i < pollfds->len; i++) {
 434         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 435         int fd = pfd->fd;
 436         int events = pfd->events;
 437         if (events & G_IO_IN) {
 438             FD_SET(fd, rfds);
 439             nfds = MAX(nfds, fd);
 440         }
 441         if (events & G_IO_OUT) {
 442             FD_SET(fd, wfds);
 443             nfds = MAX(nfds, fd);
 444         }
 445         if (events & G_IO_PRI) {
 446             FD_SET(fd, xfds);
 447             nfds = MAX(nfds, fd);
 448         }
 449     }
 450     return nfds;
 451 }
 452
 453 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
 454                          fd_set *wfds, fd_set *xfds)
 455 {
 456     int i;
 457
 458     for (i = 0; i < pollfds->len; i++) {
 459         GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
 460         int fd = pfd->fd;
 461         int revents = 0;
 462
 463         if (FD_ISSET(fd, rfds)) {
 464             revents |= G_IO_IN;
 465         }
 466         if (FD_ISSET(fd, wfds)) {
 467             revents |= G_IO_OUT;
 468         }
 469         if (FD_ISSET(fd, xfds)) {
 470             revents |= G_IO_PRI;
 471         }
 472         pfd->revents = revents & pfd->events;
 473     }
 474 }
 475
 476 static int os_host_main_loop_wait(int64_t timeout)
 477 {
 478     GMainContext *context = g_main_context_default();
 479     GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
 480     int select_ret = 0;
 481     int g_poll_ret, ret, i, n_poll_fds;
 482     PollingEntry *pe;
 483     WaitObjects *w = &wait_objects;
 484     gint poll_timeout;
 485     int64_t poll_timeout_ns;
 486     static struct timeval tv0;
 487     fd_set rfds, wfds, xfds;
 488     int nfds;
 489
 490     g_main_context_acquire(context);
 491
 492     /* XXX: need to suppress polling by better using win32 events */
 493     ret = 0;
 494     for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
 495         ret |= pe->func(pe->opaque);
 496     }
 497     if (ret != 0) {
 498         g_main_context_release(context);
 499         return ret;
 500     }
 501
 502     FD_ZERO(&rfds);
 503     FD_ZERO(&wfds);
 504     FD_ZERO(&xfds);
 505     nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
 506     if (nfds >= 0) {
 507         select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
 508         if (select_ret != 0) {
 509             timeout = 0;
 510         }
 511         if (select_ret > 0) {
 512             pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
 513         }
 514     }
 515
 516     g_main_context_prepare(context, &max_priority);
 517     n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
 518                                       poll_fds, ARRAY_SIZE(poll_fds));
 519     g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
 520
 521     for (i = 0; i < w->num; i++) {
 522         poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
 523         poll_fds[n_poll_fds + i].events = G_IO_IN;
 524     }
 525
 526     if (poll_timeout < 0) {
 527         poll_timeout_ns = -1;
 528     } else {
 529         poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
 530     }
 531
 532     poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
 533
 534     qemu_mutex_unlock_iothread();
 535
 536     replay_mutex_unlock();
 537
 538     g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
 539
 540     replay_mutex_lock();
 541
 542     qemu_mutex_lock_iothread();
 543     if (g_poll_ret > 0) {
 544         for (i = 0; i < w->num; i++) {
 545             w->revents[i] = poll_fds[n_poll_fds + i].revents;
 546         }
 547         for (i = 0; i < w->num; i++) {
 548             if (w->revents[i] && w->func[i]) {
 549                 w->func[i](w->opaque[i]);
 550             }
 551         }
 552     }
 553
 554     if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
 555         g_main_context_dispatch(context);
 556     }
 557
 558     g_main_context_release(context);
 559
 560     return select_ret || g_poll_ret;
 561 }
 562 #endif
 563
 564 static NotifierList main_loop_poll_notifiers =
 565     NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
 566
 567 void main_loop_poll_add_notifier(Notifier *notify)
 568 {
 569     notifier_list_add(&main_loop_poll_notifiers, notify);
 570 }
 571
 572 void main_loop_poll_remove_notifier(Notifier *notify)
 573 {
 574     notifier_remove(notify);
 575 }
 576
 577 void main_loop_wait(int nonblocking)
 578 {
 579     MainLoopPoll mlpoll = {
 580         .state = MAIN_LOOP_POLL_FILL,
 581         .timeout = UINT32_MAX,
 582         .pollfds = gpollfds,
 583     };
 584     int ret;
 585     int64_t timeout_ns;
 586
 587     if (nonblocking) {
 588         mlpoll.timeout = 0;
 589     }
 590
 591     /* poll any events */
 592     g_array_set_size(gpollfds, 0); /* reset for new iteration */
 593     /* XXX: separate device handlers from system ones */
 594     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 595
 596     if (mlpoll.timeout == UINT32_MAX) {
 597         timeout_ns = -1;
 598     } else {
 599         timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
 600     }
 601
 602     timeout_ns = qemu_soonest_timeout(timeout_ns,
 603                                       timerlistgroup_deadline_ns(
 604                                           &main_loop_tlg));
 605
 606     ret = os_host_main_loop_wait(timeout_ns);
 607     mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
 608     notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
 609
 610     if (icount_enabled()) {
 611         /*
 612          * CPU thread can infinitely wait for event after
 613          * missing the warp
 614          */
 615         icount_start_warp_timer();
 616     }
 617     qemu_clock_run_all_timers();
 618 }
 619
 620 /* Functions to operate on the main QEMU AioContext.  */
 621
 622 QEMUBH *qemu_bh_new_full(QEMUBHFunc *cb, void *opaque, const char *name)
 623 {
 624     return aio_bh_new_full(qemu_aio_context, cb, opaque, name);
 625 }
 626
 627 /*
 628  * Functions to operate on the I/O handler AioContext.
 629  * This context runs on top of main loop. We can't reuse qemu_aio_context
 630  * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
 631  */
 632 static AioContext *iohandler_ctx;
 633
 634 static void iohandler_init(void)
 635 {
 636     if (!iohandler_ctx) {
 637         iohandler_ctx = aio_context_new(&error_abort);
 638     }
 639 }
 640
 641 AioContext *iohandler_get_aio_context(void)
 642 {
 643     iohandler_init();
 644     return iohandler_ctx;
 645 }
 646
 647 GSource *iohandler_get_g_source(void)
 648 {
 649     iohandler_init();
 650     return aio_get_g_source(iohandler_ctx);
 651 }
 652
 653 void qemu_set_fd_handler(int fd,
 654                          IOHandler *fd_read,
 655                          IOHandler *fd_write,
 656                          void *opaque)
 657 {
 658     iohandler_init();
 659     aio_set_fd_handler(iohandler_ctx, fd, false,
 660                        fd_read, fd_write, NULL, NULL, opaque);
 661 }
 662
 663 void event_notifier_set_handler(EventNotifier *e,
 664                                 EventNotifierHandler *handler)
 665 {
 666     iohandler_init();
 667     aio_set_event_notifier(iohandler_ctx, e, false,
 668                            handler, NULL, NULL);
 669 }