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 "slirp/slirp.h"
28 #include "qemu/main-loop.h"
29 #include "block/aio.h"
36 static int io_thread_fd = -1;
38 void qemu_notify_event(void)
40 /* Write 8 bytes to be compatible with eventfd. */
41 static const uint64_t val = 1;
44 if (io_thread_fd == -1) {
48 qemu_notify_hax_event();
51 ret = write(io_thread_fd, &val, sizeof(val));
52 } while (ret < 0 && errno == EINTR);
54 /* EAGAIN is fine, a read must be pending. */
55 if (ret < 0 && errno != EAGAIN) {
56 fprintf(stderr, "qemu_notify_event: write() failed: %s\n",
62 static void qemu_event_read(void *opaque)
64 int fd = (intptr_t)opaque;
68 /* Drain the notify pipe. For eventfd, only 8 bytes will be read. */
70 len = read(fd, buffer, sizeof(buffer));
71 } while ((len == -1 && errno == EINTR) || len == sizeof(buffer));
74 static int qemu_event_init(void)
79 err = qemu_eventfd(fds);
83 err = fcntl_setfl(fds[0], O_NONBLOCK);
87 err = fcntl_setfl(fds[1], O_NONBLOCK);
91 qemu_set_fd_handler2(fds[0], NULL, qemu_event_read, NULL,
92 (void *)(intptr_t)fds[0]);
94 io_thread_fd = fds[1];
103 #include "qemu/compatfd.h"
106 /* If we have signalfd, we mask out the signals we want to handle and then
107 * use signalfd to listen for them. We rely on whatever the current signal
108 * handler is to dispatch the signals when we receive them.
110 static void sigfd_handler(void *opaque)
112 int fd = (intptr_t)opaque;
113 struct qemu_signalfd_siginfo info;
114 struct sigaction action;
119 len = read(fd, &info, sizeof(info));
120 } while (len == -1 && errno == EINTR);
122 if (len == -1 && errno == EAGAIN) {
126 if (len != sizeof(info)) {
127 printf("read from sigfd returned %zd: %m\n", len);
131 sigaction(info.ssi_signo, NULL, &action);
132 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
133 action.sa_sigaction(info.ssi_signo,
134 (siginfo_t *)&info, NULL);
135 } else if (action.sa_handler) {
136 action.sa_handler(info.ssi_signo);
141 static int qemu_signal_init(void)
147 * SIG_IPI must be blocked in the main thread and must not be caught
148 * by sigwait() in the signal thread. Otherwise, the cpu thread will
149 * not catch it reliably.
152 sigaddset(&set, SIG_IPI);
153 sigaddset(&set, SIGIO);
154 sigaddset(&set, SIGALRM);
155 sigaddset(&set, SIGBUS);
156 pthread_sigmask(SIG_BLOCK, &set, NULL);
158 sigdelset(&set, SIG_IPI);
159 sigfd = qemu_signalfd(&set);
161 fprintf(stderr, "failed to create signalfd\n");
165 fcntl_setfl(sigfd, O_NONBLOCK);
167 qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
168 (void *)(intptr_t)sigfd);
175 static int qemu_signal_init(void)
181 static AioContext *qemu_aio_context;
183 AioContext *qemu_get_aio_context(void)
185 return qemu_aio_context;
188 void qemu_notify_event(void)
190 if (!qemu_aio_context) {
195 qemu_notify_hax_event();
197 if (!SetEvent(qemu_event_handle)) {
198 fprintf(stderr, "qemu_notify_event: SetEvent failed: %ld\n",
203 aio_notify(qemu_aio_context);
207 static GArray *gpollfds;
209 int qemu_init_main_loop(void)
215 if (init_timer_alarm() < 0) {
216 fprintf(stderr, "could not initialize alarm timer\n");
220 ret = qemu_signal_init();
225 gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
226 qemu_aio_context = aio_context_new();
227 src = aio_get_g_source(qemu_aio_context);
228 g_source_attach(src, NULL);
233 static int max_priority;
236 static int glib_pollfds_idx;
237 static int glib_n_poll_fds;
239 static void glib_pollfds_fill(uint32_t *cur_timeout)
241 GMainContext *context = g_main_context_default();
245 g_main_context_prepare(context, &max_priority);
247 glib_pollfds_idx = gpollfds->len;
252 g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
253 pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
254 n = g_main_context_query(context, max_priority, &timeout, pfds,
256 } while (n != glib_n_poll_fds);
258 if (timeout >= 0 && timeout < *cur_timeout) {
259 *cur_timeout = timeout;
263 static void glib_pollfds_poll(void)
265 GMainContext *context = g_main_context_default();
266 GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
268 if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
269 g_main_context_dispatch(context);
273 #define MAX_MAIN_LOOP_SPIN (1000)
275 static int os_host_main_loop_wait(uint32_t timeout)
278 static int spin_counter;
280 glib_pollfds_fill(&timeout);
282 /* If the I/O thread is very busy or we are incorrectly busy waiting in
283 * the I/O thread, this can lead to starvation of the BQL such that the
284 * VCPU threads never run. To make sure we can detect the later case,
285 * print a message to the screen. If we run into this condition, create
286 * a fake timeout in order to give the VCPU threads a chance to run.
288 if (spin_counter > MAX_MAIN_LOOP_SPIN) {
289 static bool notified;
293 "main-loop: WARNING: I/O thread spun for %d iterations\n",
303 qemu_mutex_unlock_iothread();
308 ret = g_poll((GPollFD *)gpollfds->data, gpollfds->len, timeout);
311 qemu_mutex_lock_iothread();
318 /***********************************************************/
319 /* Polling handling */
321 typedef struct PollingEntry {
324 struct PollingEntry *next;
327 static PollingEntry *first_polling_entry;
329 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
331 PollingEntry **ppe, *pe;
332 pe = g_malloc0(sizeof(PollingEntry));
335 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
340 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
342 PollingEntry **ppe, *pe;
343 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
345 if (pe->func == func && pe->opaque == opaque) {
353 /***********************************************************/
354 /* Wait objects support */
355 typedef struct WaitObjects {
357 int revents[MAXIMUM_WAIT_OBJECTS + 1];
358 HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
359 WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
360 void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
363 static WaitObjects wait_objects = {0};
365 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
367 WaitObjects *w = &wait_objects;
368 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
371 w->events[w->num] = handle;
372 w->func[w->num] = func;
373 w->opaque[w->num] = opaque;
374 w->revents[w->num] = 0;
379 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
382 WaitObjects *w = &wait_objects;
385 for (i = 0; i < w->num; i++) {
386 if (w->events[i] == handle) {
390 w->events[i] = w->events[i + 1];
391 w->func[i] = w->func[i + 1];
392 w->opaque[i] = w->opaque[i + 1];
393 w->revents[i] = w->revents[i + 1];
401 void qemu_fd_register(int fd)
403 WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
404 FD_READ | FD_ACCEPT | FD_CLOSE |
405 FD_CONNECT | FD_WRITE | FD_OOB);
408 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
414 for (i = 0; i < pollfds->len; i++) {
415 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
417 int events = pfd->events;
418 if (events & G_IO_IN) {
420 nfds = MAX(nfds, fd);
422 if (events & G_IO_OUT) {
424 nfds = MAX(nfds, fd);
426 if (events & G_IO_PRI) {
428 nfds = MAX(nfds, fd);
434 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
435 fd_set *wfds, fd_set *xfds)
439 for (i = 0; i < pollfds->len; i++) {
440 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
444 if (FD_ISSET(fd, rfds)) {
447 if (FD_ISSET(fd, wfds)) {
450 if (FD_ISSET(fd, xfds)) {
453 pfd->revents = revents & pfd->events;
457 static int os_host_main_loop_wait(uint32_t timeout)
459 GMainContext *context = g_main_context_default();
460 GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
462 int g_poll_ret, ret, i, n_poll_fds;
464 WaitObjects *w = &wait_objects;
466 static struct timeval tv0;
467 fd_set rfds, wfds, xfds;
470 /* XXX: need to suppress polling by better using win32 events */
472 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
473 ret |= pe->func(pe->opaque);
482 nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
484 select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
485 if (select_ret != 0) {
488 if (select_ret > 0) {
489 pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
493 g_main_context_prepare(context, &max_priority);
494 n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
495 poll_fds, ARRAY_SIZE(poll_fds));
496 g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
498 for (i = 0; i < w->num; i++) {
499 poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
500 poll_fds[n_poll_fds + i].events = G_IO_IN;
503 if (poll_timeout < 0 || timeout < poll_timeout) {
504 poll_timeout = timeout;
507 qemu_mutex_unlock_iothread();
508 g_poll_ret = g_poll(poll_fds, n_poll_fds + w->num, poll_timeout);
509 qemu_mutex_lock_iothread();
510 if (g_poll_ret > 0) {
511 for (i = 0; i < w->num; i++) {
512 w->revents[i] = poll_fds[n_poll_fds + i].revents;
514 for (i = 0; i < w->num; i++) {
515 if (w->revents[i] && w->func[i]) {
516 w->func[i](w->opaque[i]);
521 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
522 g_main_context_dispatch(context);
525 return select_ret || g_poll_ret;
529 int main_loop_wait(int nonblocking)
532 uint32_t timeout = UINT32_MAX;
538 /* poll any events */
539 g_array_set_size(gpollfds, 0); /* reset for new iteration */
540 /* XXX: separate device handlers from system ones */
542 slirp_update_timeout(&timeout);
543 slirp_pollfds_fill(gpollfds);
545 qemu_iohandler_fill(gpollfds);
546 ret = os_host_main_loop_wait(timeout);
547 qemu_iohandler_poll(gpollfds, ret);
549 slirp_pollfds_poll(gpollfds, (ret < 0));
552 qemu_run_all_timers();
557 /* Functions to operate on the main QEMU AioContext. */
559 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
561 return aio_bh_new(qemu_aio_context, cb, opaque);
564 bool qemu_aio_wait(void)
566 return aio_poll(qemu_aio_context, true);
570 void qemu_aio_set_fd_handler(int fd,
573 AioFlushHandler *io_flush,
576 aio_set_fd_handler(qemu_aio_context, fd, io_read, io_write, io_flush,
581 void qemu_aio_set_event_notifier(EventNotifier *notifier,
582 EventNotifierHandler *io_read,
583 AioFlushEventNotifierHandler *io_flush)
585 aio_set_event_notifier(qemu_aio_context, notifier, io_read, io_flush);