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 "qemu/sockets.h" // struct in_addr needed for libslirp.h
30 #include "sysemu/qtest.h"
31 #include "sysemu/cpus.h"
32 #include "sysemu/replay.h"
33 #include "slirp/libslirp.h"
34 #include "qemu/main-loop.h"
35 #include "block/aio.h"
36 #include "qemu/error-report.h"
40 /* If we have signalfd, we mask out the signals we want to handle and then
41 * use signalfd to listen for them. We rely on whatever the current signal
42 * handler is to dispatch the signals when we receive them.
44 static void sigfd_handler(void *opaque)
46 int fd = (intptr_t)opaque;
47 struct qemu_signalfd_siginfo info;
48 struct sigaction action;
53 len = read(fd, &info, sizeof(info));
54 } while (len == -1 && errno == EINTR);
56 if (len == -1 && errno == EAGAIN) {
60 if (len != sizeof(info)) {
61 printf("read from sigfd returned %zd: %m\n", len);
65 sigaction(info.ssi_signo, NULL, &action);
66 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
67 sigaction_invoke(&action, &info);
68 } else if (action.sa_handler) {
69 action.sa_handler(info.ssi_signo);
74 static int qemu_signal_init(void)
80 * SIG_IPI must be blocked in the main thread and must not be caught
81 * by sigwait() in the signal thread. Otherwise, the cpu thread will
82 * not catch it reliably.
85 sigaddset(&set, SIG_IPI);
86 sigaddset(&set, SIGIO);
87 sigaddset(&set, SIGALRM);
88 sigaddset(&set, SIGBUS);
89 /* SIGINT cannot be handled via signalfd, so that ^C can be used
90 * to interrupt QEMU when it is being run under gdb. SIGHUP and
91 * SIGTERM are also handled asynchronously, even though it is not
92 * strictly necessary, because they use the same handler as SIGINT.
94 pthread_sigmask(SIG_BLOCK, &set, NULL);
96 sigdelset(&set, SIG_IPI);
97 sigfd = qemu_signalfd(&set);
99 fprintf(stderr, "failed to create signalfd\n");
103 fcntl_setfl(sigfd, O_NONBLOCK);
105 qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
112 static int qemu_signal_init(void)
118 static AioContext *qemu_aio_context;
119 static QEMUBH *qemu_notify_bh;
121 static void notify_event_cb(void *opaque)
123 /* No need to do anything; this bottom half is only used to
124 * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
128 AioContext *qemu_get_aio_context(void)
130 return qemu_aio_context;
133 void qemu_notify_event(void)
135 if (!qemu_aio_context) {
138 qemu_bh_schedule(qemu_notify_bh);
141 static GArray *gpollfds;
143 int qemu_init_main_loop(Error **errp)
147 Error *local_error = NULL;
149 init_clocks(qemu_timer_notify_cb);
151 ret = qemu_signal_init();
156 qemu_aio_context = aio_context_new(&local_error);
157 if (!qemu_aio_context) {
158 error_propagate(errp, local_error);
161 qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
162 gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
163 src = aio_get_g_source(qemu_aio_context);
164 g_source_set_name(src, "aio-context");
165 g_source_attach(src, NULL);
167 src = iohandler_get_g_source();
168 g_source_set_name(src, "io-handler");
169 g_source_attach(src, NULL);
174 static int max_priority;
177 static int glib_pollfds_idx;
178 static int glib_n_poll_fds;
180 static void glib_pollfds_fill(int64_t *cur_timeout)
182 GMainContext *context = g_main_context_default();
187 g_main_context_prepare(context, &max_priority);
189 glib_pollfds_idx = gpollfds->len;
194 g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
195 pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
196 n = g_main_context_query(context, max_priority, &timeout, pfds,
198 } while (n != glib_n_poll_fds);
203 timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
206 *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
209 static void glib_pollfds_poll(void)
211 GMainContext *context = g_main_context_default();
212 GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
214 if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
215 g_main_context_dispatch(context);
219 #define MAX_MAIN_LOOP_SPIN (1000)
221 static int os_host_main_loop_wait(int64_t timeout)
223 GMainContext *context = g_main_context_default();
225 static int spin_counter;
227 g_main_context_acquire(context);
229 glib_pollfds_fill(&timeout);
231 /* If the I/O thread is very busy or we are incorrectly busy waiting in
232 * the I/O thread, this can lead to starvation of the BQL such that the
233 * VCPU threads never run. To make sure we can detect the later case,
234 * print a message to the screen. If we run into this condition, create
235 * a fake timeout in order to give the VCPU threads a chance to run.
237 if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
238 static bool notified;
240 if (!notified && !qtest_enabled() && !qtest_driver()) {
241 warn_report("I/O thread spun for %d iterations",
255 qemu_mutex_unlock_iothread();
256 replay_mutex_unlock();
258 ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
261 qemu_mutex_lock_iothread();
265 g_main_context_release(context);
270 /***********************************************************/
271 /* Polling handling */
273 typedef struct PollingEntry {
276 struct PollingEntry *next;
279 static PollingEntry *first_polling_entry;
281 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
283 PollingEntry **ppe, *pe;
284 pe = g_malloc0(sizeof(PollingEntry));
287 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
292 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
294 PollingEntry **ppe, *pe;
295 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
297 if (pe->func == func && pe->opaque == opaque) {
305 /***********************************************************/
306 /* Wait objects support */
307 typedef struct WaitObjects {
309 int revents[MAXIMUM_WAIT_OBJECTS + 1];
310 HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
311 WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
312 void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
315 static WaitObjects wait_objects = {0};
317 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
319 WaitObjects *w = &wait_objects;
320 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
323 w->events[w->num] = handle;
324 w->func[w->num] = func;
325 w->opaque[w->num] = opaque;
326 w->revents[w->num] = 0;
331 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
334 WaitObjects *w = &wait_objects;
337 for (i = 0; i < w->num; i++) {
338 if (w->events[i] == handle) {
342 w->events[i] = w->events[i + 1];
343 w->func[i] = w->func[i + 1];
344 w->opaque[i] = w->opaque[i + 1];
345 w->revents[i] = w->revents[i + 1];
353 void qemu_fd_register(int fd)
355 WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
356 FD_READ | FD_ACCEPT | FD_CLOSE |
357 FD_CONNECT | FD_WRITE | FD_OOB);
360 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
366 for (i = 0; i < pollfds->len; i++) {
367 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
369 int events = pfd->events;
370 if (events & G_IO_IN) {
372 nfds = MAX(nfds, fd);
374 if (events & G_IO_OUT) {
376 nfds = MAX(nfds, fd);
378 if (events & G_IO_PRI) {
380 nfds = MAX(nfds, fd);
386 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
387 fd_set *wfds, fd_set *xfds)
391 for (i = 0; i < pollfds->len; i++) {
392 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
396 if (FD_ISSET(fd, rfds)) {
399 if (FD_ISSET(fd, wfds)) {
402 if (FD_ISSET(fd, xfds)) {
405 pfd->revents = revents & pfd->events;
409 static int os_host_main_loop_wait(int64_t timeout)
411 GMainContext *context = g_main_context_default();
412 GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
414 int g_poll_ret, ret, i, n_poll_fds;
416 WaitObjects *w = &wait_objects;
418 int64_t poll_timeout_ns;
419 static struct timeval tv0;
420 fd_set rfds, wfds, xfds;
423 g_main_context_acquire(context);
425 /* XXX: need to suppress polling by better using win32 events */
427 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
428 ret |= pe->func(pe->opaque);
431 g_main_context_release(context);
438 nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
440 select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
441 if (select_ret != 0) {
444 if (select_ret > 0) {
445 pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
449 g_main_context_prepare(context, &max_priority);
450 n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
451 poll_fds, ARRAY_SIZE(poll_fds));
452 g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
454 for (i = 0; i < w->num; i++) {
455 poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
456 poll_fds[n_poll_fds + i].events = G_IO_IN;
459 if (poll_timeout < 0) {
460 poll_timeout_ns = -1;
462 poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
465 poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
467 qemu_mutex_unlock_iothread();
469 replay_mutex_unlock();
471 g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
475 qemu_mutex_lock_iothread();
476 if (g_poll_ret > 0) {
477 for (i = 0; i < w->num; i++) {
478 w->revents[i] = poll_fds[n_poll_fds + i].revents;
480 for (i = 0; i < w->num; i++) {
481 if (w->revents[i] && w->func[i]) {
482 w->func[i](w->opaque[i]);
487 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
488 g_main_context_dispatch(context);
491 g_main_context_release(context);
493 return select_ret || g_poll_ret;
497 void main_loop_wait(int nonblocking)
500 uint32_t timeout = UINT32_MAX;
507 /* poll any events */
508 g_array_set_size(gpollfds, 0); /* reset for new iteration */
509 /* XXX: separate device handlers from system ones */
510 slirp_pollfds_fill(gpollfds, &timeout);
512 if (timeout == UINT32_MAX) {
515 timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
518 timeout_ns = qemu_soonest_timeout(timeout_ns,
519 timerlistgroup_deadline_ns(
522 ret = os_host_main_loop_wait(timeout_ns);
523 slirp_pollfds_poll(gpollfds, (ret < 0));
525 /* CPU thread can infinitely wait for event after
527 qemu_start_warp_timer();
528 qemu_clock_run_all_timers();
531 /* Functions to operate on the main QEMU AioContext. */
533 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
535 return aio_bh_new(qemu_aio_context, cb, opaque);