migration: do not search dirty pages in bulk stage
[qemu-kvm.git] / main-loop.c
blobeb80ff369f9ef9e327a2170201086702ba9b4db2
1 /*
2 * QEMU System Emulator
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
22 * THE SOFTWARE.
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"
31 #ifndef _WIN32
33 #include "qemu/compatfd.h"
35 /* If we have signalfd, we mask out the signals we want to handle and then
36 * use signalfd to listen for them. We rely on whatever the current signal
37 * handler is to dispatch the signals when we receive them.
39 static void sigfd_handler(void *opaque)
41 int fd = (intptr_t)opaque;
42 struct qemu_signalfd_siginfo info;
43 struct sigaction action;
44 ssize_t len;
46 while (1) {
47 do {
48 len = read(fd, &info, sizeof(info));
49 } while (len == -1 && errno == EINTR);
51 if (len == -1 && errno == EAGAIN) {
52 break;
55 if (len != sizeof(info)) {
56 printf("read from sigfd returned %zd: %m\n", len);
57 return;
60 sigaction(info.ssi_signo, NULL, &action);
61 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
62 action.sa_sigaction(info.ssi_signo,
63 (siginfo_t *)&info, NULL);
64 } else if (action.sa_handler) {
65 action.sa_handler(info.ssi_signo);
70 static int qemu_signal_init(void)
72 int sigfd;
73 sigset_t set;
76 * SIG_IPI must be blocked in the main thread and must not be caught
77 * by sigwait() in the signal thread. Otherwise, the cpu thread will
78 * not catch it reliably.
80 sigemptyset(&set);
81 sigaddset(&set, SIG_IPI);
82 sigaddset(&set, SIGIO);
83 sigaddset(&set, SIGALRM);
84 sigaddset(&set, SIGBUS);
85 pthread_sigmask(SIG_BLOCK, &set, NULL);
87 sigdelset(&set, SIG_IPI);
88 sigfd = qemu_signalfd(&set);
89 if (sigfd == -1) {
90 fprintf(stderr, "failed to create signalfd\n");
91 return -errno;
94 fcntl_setfl(sigfd, O_NONBLOCK);
96 qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
97 (void *)(intptr_t)sigfd);
99 return 0;
102 #else /* _WIN32 */
104 static int qemu_signal_init(void)
106 return 0;
108 #endif
110 static AioContext *qemu_aio_context;
112 AioContext *qemu_get_aio_context(void)
114 return qemu_aio_context;
117 void qemu_notify_event(void)
119 if (!qemu_aio_context) {
120 return;
122 aio_notify(qemu_aio_context);
125 static GArray *gpollfds;
127 int qemu_init_main_loop(void)
129 int ret;
130 GSource *src;
132 init_clocks();
133 if (init_timer_alarm() < 0) {
134 fprintf(stderr, "could not initialize alarm timer\n");
135 exit(1);
138 ret = qemu_signal_init();
139 if (ret) {
140 return ret;
143 gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
144 qemu_aio_context = aio_context_new();
145 src = aio_get_g_source(qemu_aio_context);
146 g_source_attach(src, NULL);
147 g_source_unref(src);
148 return 0;
151 static int max_priority;
153 #ifndef _WIN32
154 static int glib_pollfds_idx;
155 static int glib_n_poll_fds;
157 static void glib_pollfds_fill(uint32_t *cur_timeout)
159 GMainContext *context = g_main_context_default();
160 int timeout = 0;
161 int n;
163 g_main_context_prepare(context, &max_priority);
165 glib_pollfds_idx = gpollfds->len;
166 n = glib_n_poll_fds;
167 do {
168 GPollFD *pfds;
169 glib_n_poll_fds = n;
170 g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
171 pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
172 n = g_main_context_query(context, max_priority, &timeout, pfds,
173 glib_n_poll_fds);
174 } while (n != glib_n_poll_fds);
176 if (timeout >= 0 && timeout < *cur_timeout) {
177 *cur_timeout = timeout;
181 static void glib_pollfds_poll(void)
183 GMainContext *context = g_main_context_default();
184 GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
186 if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
187 g_main_context_dispatch(context);
191 static int os_host_main_loop_wait(uint32_t timeout)
193 int ret;
195 glib_pollfds_fill(&timeout);
197 if (timeout > 0) {
198 qemu_mutex_unlock_iothread();
201 ret = g_poll((GPollFD *)gpollfds->data, gpollfds->len, timeout);
203 if (timeout > 0) {
204 qemu_mutex_lock_iothread();
207 glib_pollfds_poll();
208 return ret;
210 #else
211 /***********************************************************/
212 /* Polling handling */
214 typedef struct PollingEntry {
215 PollingFunc *func;
216 void *opaque;
217 struct PollingEntry *next;
218 } PollingEntry;
220 static PollingEntry *first_polling_entry;
222 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
224 PollingEntry **ppe, *pe;
225 pe = g_malloc0(sizeof(PollingEntry));
226 pe->func = func;
227 pe->opaque = opaque;
228 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
229 *ppe = pe;
230 return 0;
233 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
235 PollingEntry **ppe, *pe;
236 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
237 pe = *ppe;
238 if (pe->func == func && pe->opaque == opaque) {
239 *ppe = pe->next;
240 g_free(pe);
241 break;
246 /***********************************************************/
247 /* Wait objects support */
248 typedef struct WaitObjects {
249 int num;
250 int revents[MAXIMUM_WAIT_OBJECTS + 1];
251 HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
252 WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
253 void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
254 } WaitObjects;
256 static WaitObjects wait_objects = {0};
258 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
260 WaitObjects *w = &wait_objects;
261 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
262 return -1;
264 w->events[w->num] = handle;
265 w->func[w->num] = func;
266 w->opaque[w->num] = opaque;
267 w->revents[w->num] = 0;
268 w->num++;
269 return 0;
272 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
274 int i, found;
275 WaitObjects *w = &wait_objects;
277 found = 0;
278 for (i = 0; i < w->num; i++) {
279 if (w->events[i] == handle) {
280 found = 1;
282 if (found) {
283 w->events[i] = w->events[i + 1];
284 w->func[i] = w->func[i + 1];
285 w->opaque[i] = w->opaque[i + 1];
286 w->revents[i] = w->revents[i + 1];
289 if (found) {
290 w->num--;
294 void qemu_fd_register(int fd)
296 WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
297 FD_READ | FD_ACCEPT | FD_CLOSE |
298 FD_CONNECT | FD_WRITE | FD_OOB);
301 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
302 fd_set *xfds)
304 int nfds = -1;
305 int i;
307 for (i = 0; i < pollfds->len; i++) {
308 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
309 int fd = pfd->fd;
310 int events = pfd->events;
311 if (events & (G_IO_IN | G_IO_HUP | G_IO_ERR)) {
312 FD_SET(fd, rfds);
313 nfds = MAX(nfds, fd);
315 if (events & (G_IO_OUT | G_IO_ERR)) {
316 FD_SET(fd, wfds);
317 nfds = MAX(nfds, fd);
319 if (events & G_IO_PRI) {
320 FD_SET(fd, xfds);
321 nfds = MAX(nfds, fd);
324 return nfds;
327 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
328 fd_set *wfds, fd_set *xfds)
330 int i;
332 for (i = 0; i < pollfds->len; i++) {
333 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
334 int fd = pfd->fd;
335 int revents = 0;
337 if (FD_ISSET(fd, rfds)) {
338 revents |= G_IO_IN | G_IO_HUP | G_IO_ERR;
340 if (FD_ISSET(fd, wfds)) {
341 revents |= G_IO_OUT | G_IO_ERR;
343 if (FD_ISSET(fd, xfds)) {
344 revents |= G_IO_PRI;
346 pfd->revents = revents & pfd->events;
350 static int os_host_main_loop_wait(uint32_t timeout)
352 GMainContext *context = g_main_context_default();
353 GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
354 int select_ret = 0;
355 int g_poll_ret, ret, i, n_poll_fds;
356 PollingEntry *pe;
357 WaitObjects *w = &wait_objects;
358 gint poll_timeout;
359 static struct timeval tv0;
360 fd_set rfds, wfds, xfds;
361 int nfds;
363 /* XXX: need to suppress polling by better using win32 events */
364 ret = 0;
365 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
366 ret |= pe->func(pe->opaque);
368 if (ret != 0) {
369 return ret;
372 g_main_context_prepare(context, &max_priority);
373 n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
374 poll_fds, ARRAY_SIZE(poll_fds));
375 g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
377 for (i = 0; i < w->num; i++) {
378 poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
379 poll_fds[n_poll_fds + i].events = G_IO_IN;
382 if (poll_timeout < 0 || timeout < poll_timeout) {
383 poll_timeout = timeout;
386 qemu_mutex_unlock_iothread();
387 g_poll_ret = g_poll(poll_fds, n_poll_fds + w->num, poll_timeout);
388 qemu_mutex_lock_iothread();
389 if (g_poll_ret > 0) {
390 for (i = 0; i < w->num; i++) {
391 w->revents[i] = poll_fds[n_poll_fds + i].revents;
393 for (i = 0; i < w->num; i++) {
394 if (w->revents[i] && w->func[i]) {
395 w->func[i](w->opaque[i]);
400 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
401 g_main_context_dispatch(context);
404 /* Call select after g_poll to avoid a useless iteration and therefore
405 * improve socket latency.
408 FD_ZERO(&rfds);
409 FD_ZERO(&wfds);
410 FD_ZERO(&xfds);
411 nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
412 if (nfds >= 0) {
413 select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
414 if (select_ret != 0) {
415 timeout = 0;
417 if (select_ret > 0) {
418 pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
422 return select_ret || g_poll_ret;
424 #endif
426 int main_loop_wait(int nonblocking)
428 int ret;
429 uint32_t timeout = UINT32_MAX;
431 if (nonblocking) {
432 timeout = 0;
435 /* poll any events */
436 g_array_set_size(gpollfds, 0); /* reset for new iteration */
437 /* XXX: separate device handlers from system ones */
438 #ifdef CONFIG_SLIRP
439 slirp_update_timeout(&timeout);
440 slirp_pollfds_fill(gpollfds);
441 #endif
442 qemu_iohandler_fill(gpollfds);
443 ret = os_host_main_loop_wait(timeout);
444 qemu_iohandler_poll(gpollfds, ret);
445 #ifdef CONFIG_SLIRP
446 slirp_pollfds_poll(gpollfds, (ret < 0));
447 #endif
449 qemu_run_all_timers();
451 return ret;
454 /* Functions to operate on the main QEMU AioContext. */
456 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
458 return aio_bh_new(qemu_aio_context, cb, opaque);
461 bool qemu_aio_wait(void)
463 return aio_poll(qemu_aio_context, true);
466 #ifdef CONFIG_POSIX
467 void qemu_aio_set_fd_handler(int fd,
468 IOHandler *io_read,
469 IOHandler *io_write,
470 AioFlushHandler *io_flush,
471 void *opaque)
473 aio_set_fd_handler(qemu_aio_context, fd, io_read, io_write, io_flush,
474 opaque);
476 #endif
478 void qemu_aio_set_event_notifier(EventNotifier *notifier,
479 EventNotifierHandler *io_read,
480 AioFlushEventNotifierHandler *io_flush)
482 aio_set_event_notifier(qemu_aio_context, notifier, io_read, io_flush);