4 * Copyright (c) 2003-2008 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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
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 "qom/object.h"
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.
50 * We are going to call a signal handler 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.
58 static void sigfd_handler(void *opaque)
60 int fd = (intptr_t)opaque;
61 struct qemu_signalfd_siginfo info;
62 struct sigaction action;
66 len = RETRY_ON_EINTR(read(fd, &info, sizeof(info)));
68 if (len == -1 && errno == EAGAIN) {
72 if (len != sizeof(info)) {
73 error_report("read from sigfd returned %zd: %s", len,
78 sigaction(info.ssi_signo, NULL, &action);
79 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
80 sigaction_invoke(&action, &info);
81 } else if (action.sa_handler) {
82 action.sa_handler(info.ssi_signo);
87 static int qemu_signal_init(Error **errp)
93 * SIG_IPI must be blocked in the main thread and must not be caught
94 * by sigwait() in the signal thread. Otherwise, the cpu thread will
95 * not catch it reliably.
98 sigaddset(&set, SIG_IPI);
99 sigaddset(&set, SIGIO);
100 sigaddset(&set, SIGALRM);
101 sigaddset(&set, SIGBUS);
102 /* SIGINT cannot be handled via signalfd, so that ^C can be used
103 * to interrupt QEMU when it is being run under gdb. SIGHUP and
104 * SIGTERM are also handled asynchronously, even though it is not
105 * strictly necessary, because they use the same handler as SIGINT.
107 pthread_sigmask(SIG_BLOCK, &set, NULL);
109 sigdelset(&set, SIG_IPI);
110 sigfd = qemu_signalfd(&set);
112 error_setg_errno(errp, errno, "failed to create signalfd");
116 g_unix_set_fd_nonblocking(sigfd, true, NULL);
118 qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
125 static int qemu_signal_init(Error **errp)
131 static AioContext *qemu_aio_context;
132 static QEMUBH *qemu_notify_bh;
134 static void notify_event_cb(void *opaque)
136 /* No need to do anything; this bottom half is only used to
137 * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
141 AioContext *qemu_get_aio_context(void)
143 return qemu_aio_context;
146 void qemu_notify_event(void)
148 if (!qemu_aio_context) {
151 qemu_bh_schedule(qemu_notify_bh);
154 static GArray *gpollfds;
156 int qemu_init_main_loop(Error **errp)
161 init_clocks(qemu_timer_notify_cb);
163 ret = qemu_signal_init(errp);
168 qemu_aio_context = aio_context_new(errp);
169 if (!qemu_aio_context) {
172 qemu_set_current_aio_context(qemu_aio_context);
173 qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
174 gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
175 src = aio_get_g_source(qemu_aio_context);
176 g_source_set_name(src, "aio-context");
177 g_source_attach(src, NULL);
179 src = iohandler_get_g_source();
180 g_source_set_name(src, "io-handler");
181 g_source_attach(src, NULL);
186 static void main_loop_update_params(EventLoopBase *base, Error **errp)
190 if (!qemu_aio_context) {
191 error_setg(errp, "qemu aio context not ready");
195 aio_context_set_aio_params(qemu_aio_context, base->aio_max_batch, errp);
200 aio_context_set_thread_pool_params(qemu_aio_context, base->thread_pool_min,
201 base->thread_pool_max, errp);
206 static void main_loop_init(EventLoopBase *base, Error **errp)
208 MainLoop *m = MAIN_LOOP(base);
211 error_setg(errp, "only one main-loop instance allowed");
215 main_loop_update_params(base, errp);
221 static bool main_loop_can_be_deleted(EventLoopBase *base)
226 static void main_loop_class_init(ObjectClass *oc, void *class_data)
228 EventLoopBaseClass *bc = EVENT_LOOP_BASE_CLASS(oc);
230 bc->init = main_loop_init;
231 bc->update_params = main_loop_update_params;
232 bc->can_be_deleted = main_loop_can_be_deleted;
235 static const TypeInfo main_loop_info = {
236 .name = TYPE_MAIN_LOOP,
237 .parent = TYPE_EVENT_LOOP_BASE,
238 .class_init = main_loop_class_init,
239 .instance_size = sizeof(MainLoop),
242 static void main_loop_register_types(void)
244 type_register_static(&main_loop_info);
247 type_init(main_loop_register_types)
249 static int max_priority;
252 static int glib_pollfds_idx;
253 static int glib_n_poll_fds;
255 static void glib_pollfds_fill(int64_t *cur_timeout)
257 GMainContext *context = g_main_context_default();
262 g_main_context_prepare(context, &max_priority);
264 glib_pollfds_idx = gpollfds->len;
269 g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
270 pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
271 n = g_main_context_query(context, max_priority, &timeout, pfds,
273 } while (n != glib_n_poll_fds);
278 timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
281 *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
284 static void glib_pollfds_poll(void)
286 GMainContext *context = g_main_context_default();
287 GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
289 if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
290 g_main_context_dispatch(context);
294 #define MAX_MAIN_LOOP_SPIN (1000)
296 static int os_host_main_loop_wait(int64_t timeout)
298 GMainContext *context = g_main_context_default();
301 g_main_context_acquire(context);
303 glib_pollfds_fill(&timeout);
305 qemu_mutex_unlock_iothread();
306 replay_mutex_unlock();
308 ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
311 qemu_mutex_lock_iothread();
315 g_main_context_release(context);
320 /***********************************************************/
321 /* Polling handling */
323 typedef struct PollingEntry {
326 struct PollingEntry *next;
329 static PollingEntry *first_polling_entry;
331 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
333 PollingEntry **ppe, *pe;
334 pe = g_new0(PollingEntry, 1);
337 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
342 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
344 PollingEntry **ppe, *pe;
345 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
347 if (pe->func == func && pe->opaque == opaque) {
355 /***********************************************************/
356 /* Wait objects support */
357 typedef struct WaitObjects {
359 int revents[MAXIMUM_WAIT_OBJECTS];
360 HANDLE events[MAXIMUM_WAIT_OBJECTS];
361 WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS];
362 void *opaque[MAXIMUM_WAIT_OBJECTS];
365 static WaitObjects wait_objects = {0};
367 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
370 WaitObjects *w = &wait_objects;
372 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
376 for (i = 0; i < w->num; i++) {
377 /* check if the same handle is added twice */
378 if (w->events[i] == handle) {
383 w->events[w->num] = handle;
384 w->func[w->num] = func;
385 w->opaque[w->num] = opaque;
386 w->revents[w->num] = 0;
391 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
394 WaitObjects *w = &wait_objects;
397 for (i = 0; i < w->num; i++) {
398 if (w->events[i] == handle) {
401 if (found && i < (MAXIMUM_WAIT_OBJECTS - 1)) {
402 w->events[i] = w->events[i + 1];
403 w->func[i] = w->func[i + 1];
404 w->opaque[i] = w->opaque[i + 1];
405 w->revents[i] = w->revents[i + 1];
413 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
419 for (i = 0; i < pollfds->len; i++) {
420 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
422 int events = pfd->events;
423 if (events & G_IO_IN) {
425 nfds = MAX(nfds, fd);
427 if (events & G_IO_OUT) {
429 nfds = MAX(nfds, fd);
431 if (events & G_IO_PRI) {
433 nfds = MAX(nfds, fd);
439 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
440 fd_set *wfds, fd_set *xfds)
444 for (i = 0; i < pollfds->len; i++) {
445 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
449 if (FD_ISSET(fd, rfds)) {
452 if (FD_ISSET(fd, wfds)) {
455 if (FD_ISSET(fd, xfds)) {
458 pfd->revents = revents & pfd->events;
462 static int os_host_main_loop_wait(int64_t timeout)
464 GMainContext *context = g_main_context_default();
465 GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
467 int g_poll_ret, ret, i, n_poll_fds;
469 WaitObjects *w = &wait_objects;
471 int64_t poll_timeout_ns;
472 static struct timeval tv0;
473 fd_set rfds, wfds, xfds;
476 g_main_context_acquire(context);
478 /* XXX: need to suppress polling by better using win32 events */
480 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
481 ret |= pe->func(pe->opaque);
484 g_main_context_release(context);
491 nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
493 select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
494 if (select_ret != 0) {
497 if (select_ret > 0) {
498 pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
502 g_main_context_prepare(context, &max_priority);
503 n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
504 poll_fds, ARRAY_SIZE(poll_fds));
505 g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
507 for (i = 0; i < w->num; i++) {
508 poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
509 poll_fds[n_poll_fds + i].events = G_IO_IN;
512 if (poll_timeout < 0) {
513 poll_timeout_ns = -1;
515 poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
518 poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
520 qemu_mutex_unlock_iothread();
522 replay_mutex_unlock();
524 g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
528 qemu_mutex_lock_iothread();
529 if (g_poll_ret > 0) {
530 for (i = 0; i < w->num; i++) {
531 w->revents[i] = poll_fds[n_poll_fds + i].revents;
533 for (i = 0; i < w->num; i++) {
534 if (w->revents[i] && w->func[i]) {
535 w->func[i](w->opaque[i]);
540 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
541 g_main_context_dispatch(context);
544 g_main_context_release(context);
546 return select_ret || g_poll_ret;
550 static NotifierList main_loop_poll_notifiers =
551 NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
553 void main_loop_poll_add_notifier(Notifier *notify)
555 notifier_list_add(&main_loop_poll_notifiers, notify);
558 void main_loop_poll_remove_notifier(Notifier *notify)
560 notifier_remove(notify);
563 void main_loop_wait(int nonblocking)
565 MainLoopPoll mlpoll = {
566 .state = MAIN_LOOP_POLL_FILL,
567 .timeout = UINT32_MAX,
577 /* poll any events */
578 g_array_set_size(gpollfds, 0); /* reset for new iteration */
579 /* XXX: separate device handlers from system ones */
580 notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
582 if (mlpoll.timeout == UINT32_MAX) {
585 timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
588 timeout_ns = qemu_soonest_timeout(timeout_ns,
589 timerlistgroup_deadline_ns(
592 ret = os_host_main_loop_wait(timeout_ns);
593 mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
594 notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
596 if (icount_enabled()) {
598 * CPU thread can infinitely wait for event after
601 icount_start_warp_timer();
603 qemu_clock_run_all_timers();
606 /* Functions to operate on the main QEMU AioContext. */
608 QEMUBH *qemu_bh_new_full(QEMUBHFunc *cb, void *opaque, const char *name,
609 MemReentrancyGuard *reentrancy_guard)
611 return aio_bh_new_full(qemu_aio_context, cb, opaque, name,
616 * Functions to operate on the I/O handler AioContext.
617 * This context runs on top of main loop. We can't reuse qemu_aio_context
618 * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
620 static AioContext *iohandler_ctx;
622 static void iohandler_init(void)
624 if (!iohandler_ctx) {
625 iohandler_ctx = aio_context_new(&error_abort);
629 AioContext *iohandler_get_aio_context(void)
632 return iohandler_ctx;
635 GSource *iohandler_get_g_source(void)
638 return aio_get_g_source(iohandler_ctx);
641 void qemu_set_fd_handler(int fd,
647 aio_set_fd_handler(iohandler_ctx, fd, fd_read, fd_write, NULL, NULL,
651 void event_notifier_set_handler(EventNotifier *e,
652 EventNotifierHandler *handler)
655 aio_set_event_notifier(iohandler_ctx, e, handler, NULL, NULL);