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-common.h"
26 #include "qemu/timer.h"
27 #include "qemu/sockets.h" // struct in_addr needed for libslirp.h
28 #include "sysemu/qtest.h"
29 #include "slirp/libslirp.h"
30 #include "qemu/main-loop.h"
31 #include "block/aio.h"
35 #include "qemu/compatfd.h"
37 /* If we have signalfd, we mask out the signals we want to handle and then
38 * use signalfd to listen for them. We rely on whatever the current signal
39 * handler is to dispatch the signals when we receive them.
41 static void sigfd_handler(void *opaque)
43 int fd = (intptr_t)opaque;
44 struct qemu_signalfd_siginfo info;
45 struct sigaction action;
50 len = read(fd, &info, sizeof(info));
51 } while (len == -1 && errno == EINTR);
53 if (len == -1 && errno == EAGAIN) {
57 if (len != sizeof(info)) {
58 printf("read from sigfd returned %zd: %m\n", len);
62 sigaction(info.ssi_signo, NULL, &action);
63 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
64 action.sa_sigaction(info.ssi_signo,
65 (siginfo_t *)&info, NULL);
66 } else if (action.sa_handler) {
67 action.sa_handler(info.ssi_signo);
72 static int qemu_signal_init(void)
78 * SIG_IPI must be blocked in the main thread and must not be caught
79 * by sigwait() in the signal thread. Otherwise, the cpu thread will
80 * not catch it reliably.
83 sigaddset(&set, SIG_IPI);
84 sigaddset(&set, SIGIO);
85 sigaddset(&set, SIGALRM);
86 sigaddset(&set, SIGBUS);
87 sigaddset(&set, SIGINT);
88 sigaddset(&set, SIGHUP);
89 sigaddset(&set, SIGTERM);
90 pthread_sigmask(SIG_BLOCK, &set, NULL);
92 sigdelset(&set, SIG_IPI);
93 sigfd = qemu_signalfd(&set);
95 fprintf(stderr, "failed to create signalfd\n");
99 fcntl_setfl(sigfd, O_NONBLOCK);
101 qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
102 (void *)(intptr_t)sigfd);
109 static int qemu_signal_init(void)
115 static AioContext *qemu_aio_context;
117 AioContext *qemu_get_aio_context(void)
119 return qemu_aio_context;
122 void qemu_notify_event(void)
124 if (!qemu_aio_context) {
127 aio_notify(qemu_aio_context);
130 static GArray *gpollfds;
132 int qemu_init_main_loop(Error **errp)
136 Error *local_error = NULL;
140 ret = qemu_signal_init();
145 qemu_aio_context = aio_context_new(&local_error);
146 if (!qemu_aio_context) {
147 error_propagate(errp, local_error);
150 gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
151 src = aio_get_g_source(qemu_aio_context);
152 g_source_attach(src, NULL);
157 static int max_priority;
160 static int glib_pollfds_idx;
161 static int glib_n_poll_fds;
163 static void glib_pollfds_fill(int64_t *cur_timeout)
165 GMainContext *context = g_main_context_default();
170 g_main_context_prepare(context, &max_priority);
172 glib_pollfds_idx = gpollfds->len;
177 g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
178 pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
179 n = g_main_context_query(context, max_priority, &timeout, pfds,
181 } while (n != glib_n_poll_fds);
186 timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
189 *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
192 static void glib_pollfds_poll(void)
194 GMainContext *context = g_main_context_default();
195 GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
197 if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
198 g_main_context_dispatch(context);
202 #define MAX_MAIN_LOOP_SPIN (1000)
204 static int os_host_main_loop_wait(int64_t timeout)
207 static int spin_counter;
209 glib_pollfds_fill(&timeout);
211 /* If the I/O thread is very busy or we are incorrectly busy waiting in
212 * the I/O thread, this can lead to starvation of the BQL such that the
213 * VCPU threads never run. To make sure we can detect the later case,
214 * print a message to the screen. If we run into this condition, create
215 * a fake timeout in order to give the VCPU threads a chance to run.
217 if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
218 static bool notified;
220 if (!notified && !qtest_enabled()) {
222 "main-loop: WARNING: I/O thread spun for %d iterations\n",
232 qemu_mutex_unlock_iothread();
237 ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
240 qemu_mutex_lock_iothread();
247 /***********************************************************/
248 /* Polling handling */
250 typedef struct PollingEntry {
253 struct PollingEntry *next;
256 static PollingEntry *first_polling_entry;
258 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
260 PollingEntry **ppe, *pe;
261 pe = g_malloc0(sizeof(PollingEntry));
264 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
269 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
271 PollingEntry **ppe, *pe;
272 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
274 if (pe->func == func && pe->opaque == opaque) {
282 /***********************************************************/
283 /* Wait objects support */
284 typedef struct WaitObjects {
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];
292 static WaitObjects wait_objects = {0};
294 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
296 WaitObjects *w = &wait_objects;
297 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
300 w->events[w->num] = handle;
301 w->func[w->num] = func;
302 w->opaque[w->num] = opaque;
303 w->revents[w->num] = 0;
308 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
311 WaitObjects *w = &wait_objects;
314 for (i = 0; i < w->num; i++) {
315 if (w->events[i] == handle) {
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];
330 void qemu_fd_register(int fd)
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);
337 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
343 for (i = 0; i < pollfds->len; i++) {
344 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
346 int events = pfd->events;
347 if (events & G_IO_IN) {
349 nfds = MAX(nfds, fd);
351 if (events & G_IO_OUT) {
353 nfds = MAX(nfds, fd);
355 if (events & G_IO_PRI) {
357 nfds = MAX(nfds, fd);
363 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
364 fd_set *wfds, fd_set *xfds)
368 for (i = 0; i < pollfds->len; i++) {
369 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
373 if (FD_ISSET(fd, rfds)) {
376 if (FD_ISSET(fd, wfds)) {
379 if (FD_ISSET(fd, xfds)) {
382 pfd->revents = revents & pfd->events;
386 static int os_host_main_loop_wait(int64_t timeout)
388 GMainContext *context = g_main_context_default();
389 GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
391 int g_poll_ret, ret, i, n_poll_fds;
393 WaitObjects *w = &wait_objects;
395 int64_t poll_timeout_ns;
396 static struct timeval tv0;
397 fd_set rfds, wfds, xfds;
400 /* XXX: need to suppress polling by better using win32 events */
402 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
403 ret |= pe->func(pe->opaque);
412 nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
414 select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
415 if (select_ret != 0) {
418 if (select_ret > 0) {
419 pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
423 g_main_context_prepare(context, &max_priority);
424 n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
425 poll_fds, ARRAY_SIZE(poll_fds));
426 g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
428 for (i = 0; i < w->num; i++) {
429 poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
430 poll_fds[n_poll_fds + i].events = G_IO_IN;
433 if (poll_timeout < 0) {
434 poll_timeout_ns = -1;
436 poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
439 poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
441 qemu_mutex_unlock_iothread();
442 g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
444 qemu_mutex_lock_iothread();
445 if (g_poll_ret > 0) {
446 for (i = 0; i < w->num; i++) {
447 w->revents[i] = poll_fds[n_poll_fds + i].revents;
449 for (i = 0; i < w->num; i++) {
450 if (w->revents[i] && w->func[i]) {
451 w->func[i](w->opaque[i]);
456 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
457 g_main_context_dispatch(context);
460 return select_ret || g_poll_ret;
464 int main_loop_wait(int nonblocking)
467 uint32_t timeout = UINT32_MAX;
474 /* poll any events */
475 g_array_set_size(gpollfds, 0); /* reset for new iteration */
476 /* XXX: separate device handlers from system ones */
478 slirp_pollfds_fill(gpollfds, &timeout);
480 qemu_iohandler_fill(gpollfds);
482 if (timeout == UINT32_MAX) {
485 timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
488 timeout_ns = qemu_soonest_timeout(timeout_ns,
489 timerlistgroup_deadline_ns(
492 ret = os_host_main_loop_wait(timeout_ns);
493 qemu_iohandler_poll(gpollfds, ret);
495 slirp_pollfds_poll(gpollfds, (ret < 0));
498 qemu_clock_run_all_timers();
503 /* Functions to operate on the main QEMU AioContext. */
505 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
507 return aio_bh_new(qemu_aio_context, cb, opaque);