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 "slirp/libslirp.h"
32 #include "qemu/main-loop.h"
33 #include "block/aio.h"
35 #include "sysemu/hax.h"
39 #include "qemu/compatfd.h"
41 /* If we have signalfd, we mask out the signals we want to handle and then
42 * use signalfd to listen for them. We rely on whatever the current signal
43 * handler is to dispatch the signals when we receive them.
45 static void sigfd_handler(void *opaque)
47 int fd = (intptr_t)opaque;
48 struct qemu_signalfd_siginfo info;
49 struct sigaction action;
54 len = read(fd, &info, sizeof(info));
55 } while (len == -1 && errno == EINTR);
57 if (len == -1 && errno == EAGAIN) {
61 if (len != sizeof(info)) {
62 printf("read from sigfd returned %zd: %m\n", len);
66 sigaction(info.ssi_signo, NULL, &action);
67 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
68 action.sa_sigaction(info.ssi_signo,
69 (siginfo_t *)&info, NULL);
70 } else if (action.sa_handler) {
71 action.sa_handler(info.ssi_signo);
76 static int qemu_signal_init(void)
82 * SIG_IPI must be blocked in the main thread and must not be caught
83 * by sigwait() in the signal thread. Otherwise, the cpu thread will
84 * not catch it reliably.
87 sigaddset(&set, SIG_IPI);
88 sigaddset(&set, SIGIO);
89 sigaddset(&set, SIGALRM);
90 sigaddset(&set, SIGBUS);
91 /* SIGINT cannot be handled via signalfd, so that ^C can be used
92 * to interrupt QEMU when it is being run under gdb. SIGHUP and
93 * SIGTERM are also handled asynchronously, even though it is not
94 * strictly necessary, because they use the same handler as SIGINT.
96 pthread_sigmask(SIG_BLOCK, &set, NULL);
98 sigdelset(&set, SIG_IPI);
99 sigfd = qemu_signalfd(&set);
101 fprintf(stderr, "failed to create signalfd\n");
105 fcntl_setfl(sigfd, O_NONBLOCK);
107 qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
114 static int qemu_signal_init(void)
120 static AioContext *qemu_aio_context;
121 static QEMUBH *qemu_notify_bh;
123 static void notify_event_cb(void *opaque)
125 /* No need to do anything; this bottom half is only used to
126 * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
130 AioContext *qemu_get_aio_context(void)
132 return qemu_aio_context;
135 void qemu_notify_event(void)
137 if (!qemu_aio_context) {
140 qemu_bh_schedule(qemu_notify_bh);
143 static GArray *gpollfds;
145 int qemu_init_main_loop(Error **errp)
149 Error *local_error = NULL;
153 ret = qemu_signal_init();
158 qemu_aio_context = aio_context_new(&local_error);
159 if (!qemu_aio_context) {
160 error_propagate(errp, local_error);
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);
169 src = iohandler_get_g_source();
170 g_source_set_name(src, "io-handler");
171 g_source_attach(src, NULL);
176 static int max_priority;
179 static int glib_pollfds_idx;
180 static int glib_n_poll_fds;
182 static void glib_pollfds_fill(int64_t *cur_timeout)
184 GMainContext *context = g_main_context_default();
189 g_main_context_prepare(context, &max_priority);
191 glib_pollfds_idx = gpollfds->len;
196 g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
197 pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
198 n = g_main_context_query(context, max_priority, &timeout, pfds,
200 } while (n != glib_n_poll_fds);
205 timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
208 *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
211 static void glib_pollfds_poll(void)
213 GMainContext *context = g_main_context_default();
214 GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
216 if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
217 g_main_context_dispatch(context);
221 #define MAX_MAIN_LOOP_SPIN (1000)
223 static int os_host_main_loop_wait(int64_t timeout)
226 static int spin_counter;
228 glib_pollfds_fill(&timeout);
230 /* If the I/O thread is very busy or we are incorrectly busy waiting in
231 * the I/O thread, this can lead to starvation of the BQL such that the
232 * VCPU threads never run. To make sure we can detect the later case,
233 * print a message to the screen. If we run into this condition, create
234 * a fake timeout in order to give the VCPU threads a chance to run.
236 if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
237 static bool notified;
239 if (!notified && !qtest_enabled() && !qtest_driver()) {
241 "main-loop: WARNING: I/O thread spun for %d iterations\n",
251 qemu_mutex_unlock_iothread();
256 ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
259 qemu_mutex_lock_iothread();
266 /***********************************************************/
267 /* Polling handling */
269 typedef struct PollingEntry {
272 struct PollingEntry *next;
275 static PollingEntry *first_polling_entry;
277 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
279 PollingEntry **ppe, *pe;
280 pe = g_malloc0(sizeof(PollingEntry));
283 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
288 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
290 PollingEntry **ppe, *pe;
291 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
293 if (pe->func == func && pe->opaque == opaque) {
301 /***********************************************************/
302 /* Wait objects support */
303 typedef struct WaitObjects {
305 int revents[MAXIMUM_WAIT_OBJECTS + 1];
306 HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
307 WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
308 void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
311 static WaitObjects wait_objects = {0};
313 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
315 WaitObjects *w = &wait_objects;
316 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
319 w->events[w->num] = handle;
320 w->func[w->num] = func;
321 w->opaque[w->num] = opaque;
322 w->revents[w->num] = 0;
327 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
330 WaitObjects *w = &wait_objects;
333 for (i = 0; i < w->num; i++) {
334 if (w->events[i] == handle) {
338 w->events[i] = w->events[i + 1];
339 w->func[i] = w->func[i + 1];
340 w->opaque[i] = w->opaque[i + 1];
341 w->revents[i] = w->revents[i + 1];
349 void qemu_fd_register(int fd)
351 WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
352 FD_READ | FD_ACCEPT | FD_CLOSE |
353 FD_CONNECT | FD_WRITE | FD_OOB);
356 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
362 for (i = 0; i < pollfds->len; i++) {
363 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
365 int events = pfd->events;
366 if (events & G_IO_IN) {
368 nfds = MAX(nfds, fd);
370 if (events & G_IO_OUT) {
372 nfds = MAX(nfds, fd);
374 if (events & G_IO_PRI) {
376 nfds = MAX(nfds, fd);
382 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
383 fd_set *wfds, fd_set *xfds)
387 for (i = 0; i < pollfds->len; i++) {
388 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
392 if (FD_ISSET(fd, rfds)) {
395 if (FD_ISSET(fd, wfds)) {
398 if (FD_ISSET(fd, xfds)) {
401 pfd->revents = revents & pfd->events;
405 static int os_host_main_loop_wait(int64_t timeout)
407 GMainContext *context = g_main_context_default();
408 GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
410 int g_poll_ret, ret, i, n_poll_fds;
412 WaitObjects *w = &wait_objects;
414 int64_t poll_timeout_ns;
415 static struct timeval tv0;
416 fd_set rfds, wfds, xfds;
419 /* XXX: need to suppress polling by better using win32 events */
421 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
422 ret |= pe->func(pe->opaque);
431 nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
433 select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
434 if (select_ret != 0) {
437 if (select_ret > 0) {
438 pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
442 g_main_context_prepare(context, &max_priority);
443 n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
444 poll_fds, ARRAY_SIZE(poll_fds));
445 g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
447 for (i = 0; i < w->num; i++) {
448 poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
449 poll_fds[n_poll_fds + i].events = G_IO_IN;
452 if (poll_timeout < 0) {
453 poll_timeout_ns = -1;
455 poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
458 poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
460 qemu_mutex_unlock_iothread();
461 g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
463 qemu_mutex_lock_iothread();
464 if (g_poll_ret > 0) {
465 for (i = 0; i < w->num; i++) {
466 w->revents[i] = poll_fds[n_poll_fds + i].revents;
468 for (i = 0; i < w->num; i++) {
469 if (w->revents[i] && w->func[i]) {
470 w->func[i](w->opaque[i]);
475 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
476 g_main_context_dispatch(context);
479 return select_ret || g_poll_ret;
483 int main_loop_wait(int nonblocking)
486 uint32_t timeout = UINT32_MAX;
493 /* poll any events */
494 g_array_set_size(gpollfds, 0); /* reset for new iteration */
495 /* XXX: separate device handlers from system ones */
497 slirp_pollfds_fill(gpollfds, &timeout);
500 if (timeout == UINT32_MAX) {
503 timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
506 timeout_ns = qemu_soonest_timeout(timeout_ns,
507 timerlistgroup_deadline_ns(
510 ret = os_host_main_loop_wait(timeout_ns);
512 slirp_pollfds_poll(gpollfds, (ret < 0));
515 /* CPU thread can infinitely wait for event after
517 qemu_start_warp_timer();
518 qemu_clock_run_all_timers();
523 /* Functions to operate on the main QEMU AioContext. */
525 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
527 return aio_bh_new(qemu_aio_context, cb, opaque);