linux-user: add open() hijack infrastructure
[qemu.git] / main-loop.c
1 /*
2 * QEMU System Emulator
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 *
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:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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.
23 */
24
25 #include "qemu-common.h"
26 #include "qemu-timer.h"
27 #include "slirp/slirp.h"
28 #include "main-loop.h"
29
30 #ifndef _WIN32
31
32 #include "compatfd.h"
33
34 static int io_thread_fd = -1;
35
36 void qemu_notify_event(void)
37 {
38 /* Write 8 bytes to be compatible with eventfd. */
39 static const uint64_t val = 1;
40 ssize_t ret;
41
42 if (io_thread_fd == -1) {
43 return;
44 }
45 do {
46 ret = write(io_thread_fd, &val, sizeof(val));
47 } while (ret < 0 && errno == EINTR);
48
49 /* EAGAIN is fine, a read must be pending. */
50 if (ret < 0 && errno != EAGAIN) {
51 fprintf(stderr, "qemu_notify_event: write() failed: %s\n",
52 strerror(errno));
53 exit(1);
54 }
55 }
56
57 static void qemu_event_read(void *opaque)
58 {
59 int fd = (intptr_t)opaque;
60 ssize_t len;
61 char buffer[512];
62
63 /* Drain the notify pipe. For eventfd, only 8 bytes will be read. */
64 do {
65 len = read(fd, buffer, sizeof(buffer));
66 } while ((len == -1 && errno == EINTR) || len == sizeof(buffer));
67 }
68
69 static int qemu_event_init(void)
70 {
71 int err;
72 int fds[2];
73
74 err = qemu_eventfd(fds);
75 if (err == -1) {
76 return -errno;
77 }
78 err = fcntl_setfl(fds[0], O_NONBLOCK);
79 if (err < 0) {
80 goto fail;
81 }
82 err = fcntl_setfl(fds[1], O_NONBLOCK);
83 if (err < 0) {
84 goto fail;
85 }
86 qemu_set_fd_handler2(fds[0], NULL, qemu_event_read, NULL,
87 (void *)(intptr_t)fds[0]);
88
89 io_thread_fd = fds[1];
90 return 0;
91
92 fail:
93 close(fds[0]);
94 close(fds[1]);
95 return err;
96 }
97
98 /* If we have signalfd, we mask out the signals we want to handle and then
99 * use signalfd to listen for them. We rely on whatever the current signal
100 * handler is to dispatch the signals when we receive them.
101 */
102 static void sigfd_handler(void *opaque)
103 {
104 int fd = (intptr_t)opaque;
105 struct qemu_signalfd_siginfo info;
106 struct sigaction action;
107 ssize_t len;
108
109 while (1) {
110 do {
111 len = read(fd, &info, sizeof(info));
112 } while (len == -1 && errno == EINTR);
113
114 if (len == -1 && errno == EAGAIN) {
115 break;
116 }
117
118 if (len != sizeof(info)) {
119 printf("read from sigfd returned %zd: %m\n", len);
120 return;
121 }
122
123 sigaction(info.ssi_signo, NULL, &action);
124 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
125 action.sa_sigaction(info.ssi_signo,
126 (siginfo_t *)&info, NULL);
127 } else if (action.sa_handler) {
128 action.sa_handler(info.ssi_signo);
129 }
130 }
131 }
132
133 static int qemu_signal_init(void)
134 {
135 int sigfd;
136 sigset_t set;
137
138 /*
139 * SIG_IPI must be blocked in the main thread and must not be caught
140 * by sigwait() in the signal thread. Otherwise, the cpu thread will
141 * not catch it reliably.
142 */
143 sigemptyset(&set);
144 sigaddset(&set, SIG_IPI);
145 sigaddset(&set, SIGIO);
146 sigaddset(&set, SIGALRM);
147 sigaddset(&set, SIGBUS);
148 pthread_sigmask(SIG_BLOCK, &set, NULL);
149
150 sigdelset(&set, SIG_IPI);
151 sigfd = qemu_signalfd(&set);
152 if (sigfd == -1) {
153 fprintf(stderr, "failed to create signalfd\n");
154 return -errno;
155 }
156
157 fcntl_setfl(sigfd, O_NONBLOCK);
158
159 qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
160 (void *)(intptr_t)sigfd);
161
162 return 0;
163 }
164
165 #else /* _WIN32 */
166
167 HANDLE qemu_event_handle = NULL;
168
169 static void dummy_event_handler(void *opaque)
170 {
171 }
172
173 static int qemu_event_init(void)
174 {
175 qemu_event_handle = CreateEvent(NULL, FALSE, FALSE, NULL);
176 if (!qemu_event_handle) {
177 fprintf(stderr, "Failed CreateEvent: %ld\n", GetLastError());
178 return -1;
179 }
180 qemu_add_wait_object(qemu_event_handle, dummy_event_handler, NULL);
181 return 0;
182 }
183
184 void qemu_notify_event(void)
185 {
186 if (!qemu_event_handle) {
187 return;
188 }
189 if (!SetEvent(qemu_event_handle)) {
190 fprintf(stderr, "qemu_notify_event: SetEvent failed: %ld\n",
191 GetLastError());
192 exit(1);
193 }
194 }
195
196 static int qemu_signal_init(void)
197 {
198 return 0;
199 }
200 #endif
201
202 int main_loop_init(void)
203 {
204 int ret;
205
206 qemu_mutex_lock_iothread();
207 ret = qemu_signal_init();
208 if (ret) {
209 return ret;
210 }
211
212 /* Note eventfd must be drained before signalfd handlers run */
213 ret = qemu_event_init();
214 if (ret) {
215 return ret;
216 }
217
218 return 0;
219 }
220
221
222 static GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
223 static int n_poll_fds;
224 static int max_priority;
225
226 static void glib_select_fill(int *max_fd, fd_set *rfds, fd_set *wfds,
227 fd_set *xfds, struct timeval *tv)
228 {
229 GMainContext *context = g_main_context_default();
230 int i;
231 int timeout = 0, cur_timeout;
232
233 g_main_context_prepare(context, &max_priority);
234
235 n_poll_fds = g_main_context_query(context, max_priority, &timeout,
236 poll_fds, ARRAY_SIZE(poll_fds));
237 g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
238
239 for (i = 0; i < n_poll_fds; i++) {
240 GPollFD *p = &poll_fds[i];
241
242 if ((p->events & G_IO_IN)) {
243 FD_SET(p->fd, rfds);
244 *max_fd = MAX(*max_fd, p->fd);
245 }
246 if ((p->events & G_IO_OUT)) {
247 FD_SET(p->fd, wfds);
248 *max_fd = MAX(*max_fd, p->fd);
249 }
250 if ((p->events & G_IO_ERR)) {
251 FD_SET(p->fd, xfds);
252 *max_fd = MAX(*max_fd, p->fd);
253 }
254 }
255
256 cur_timeout = (tv->tv_sec * 1000) + ((tv->tv_usec + 500) / 1000);
257 if (timeout >= 0 && timeout < cur_timeout) {
258 tv->tv_sec = timeout / 1000;
259 tv->tv_usec = (timeout % 1000) * 1000;
260 }
261 }
262
263 static void glib_select_poll(fd_set *rfds, fd_set *wfds, fd_set *xfds,
264 bool err)
265 {
266 GMainContext *context = g_main_context_default();
267
268 if (!err) {
269 int i;
270
271 for (i = 0; i < n_poll_fds; i++) {
272 GPollFD *p = &poll_fds[i];
273
274 if ((p->events & G_IO_IN) && FD_ISSET(p->fd, rfds)) {
275 p->revents |= G_IO_IN;
276 }
277 if ((p->events & G_IO_OUT) && FD_ISSET(p->fd, wfds)) {
278 p->revents |= G_IO_OUT;
279 }
280 if ((p->events & G_IO_ERR) && FD_ISSET(p->fd, xfds)) {
281 p->revents |= G_IO_ERR;
282 }
283 }
284 }
285
286 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
287 g_main_context_dispatch(context);
288 }
289 }
290
291 #ifdef _WIN32
292 /***********************************************************/
293 /* Polling handling */
294
295 typedef struct PollingEntry {
296 PollingFunc *func;
297 void *opaque;
298 struct PollingEntry *next;
299 } PollingEntry;
300
301 static PollingEntry *first_polling_entry;
302
303 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
304 {
305 PollingEntry **ppe, *pe;
306 pe = g_malloc0(sizeof(PollingEntry));
307 pe->func = func;
308 pe->opaque = opaque;
309 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
310 *ppe = pe;
311 return 0;
312 }
313
314 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
315 {
316 PollingEntry **ppe, *pe;
317 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
318 pe = *ppe;
319 if (pe->func == func && pe->opaque == opaque) {
320 *ppe = pe->next;
321 g_free(pe);
322 break;
323 }
324 }
325 }
326
327 /***********************************************************/
328 /* Wait objects support */
329 typedef struct WaitObjects {
330 int num;
331 HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
332 WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
333 void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
334 } WaitObjects;
335
336 static WaitObjects wait_objects = {0};
337
338 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
339 {
340 WaitObjects *w = &wait_objects;
341 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
342 return -1;
343 }
344 w->events[w->num] = handle;
345 w->func[w->num] = func;
346 w->opaque[w->num] = opaque;
347 w->num++;
348 return 0;
349 }
350
351 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
352 {
353 int i, found;
354 WaitObjects *w = &wait_objects;
355
356 found = 0;
357 for (i = 0; i < w->num; i++) {
358 if (w->events[i] == handle) {
359 found = 1;
360 }
361 if (found) {
362 w->events[i] = w->events[i + 1];
363 w->func[i] = w->func[i + 1];
364 w->opaque[i] = w->opaque[i + 1];
365 }
366 }
367 if (found) {
368 w->num--;
369 }
370 }
371
372 static void os_host_main_loop_wait(int *timeout)
373 {
374 int ret, ret2, i;
375 PollingEntry *pe;
376
377 /* XXX: need to suppress polling by better using win32 events */
378 ret = 0;
379 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
380 ret |= pe->func(pe->opaque);
381 }
382 if (ret == 0) {
383 int err;
384 WaitObjects *w = &wait_objects;
385
386 qemu_mutex_unlock_iothread();
387 ret = WaitForMultipleObjects(w->num, w->events, FALSE, *timeout);
388 qemu_mutex_lock_iothread();
389 if (WAIT_OBJECT_0 + 0 <= ret && ret <= WAIT_OBJECT_0 + w->num - 1) {
390 if (w->func[ret - WAIT_OBJECT_0]) {
391 w->func[ret - WAIT_OBJECT_0](w->opaque[ret - WAIT_OBJECT_0]);
392 }
393
394 /* Check for additional signaled events */
395 for (i = (ret - WAIT_OBJECT_0 + 1); i < w->num; i++) {
396 /* Check if event is signaled */
397 ret2 = WaitForSingleObject(w->events[i], 0);
398 if (ret2 == WAIT_OBJECT_0) {
399 if (w->func[i]) {
400 w->func[i](w->opaque[i]);
401 }
402 } else if (ret2 != WAIT_TIMEOUT) {
403 err = GetLastError();
404 fprintf(stderr, "WaitForSingleObject error %d %d\n", i, err);
405 }
406 }
407 } else if (ret != WAIT_TIMEOUT) {
408 err = GetLastError();
409 fprintf(stderr, "WaitForMultipleObjects error %d %d\n", ret, err);
410 }
411 }
412
413 *timeout = 0;
414 }
415 #else
416 static inline void os_host_main_loop_wait(int *timeout)
417 {
418 }
419 #endif
420
421 int main_loop_wait(int nonblocking)
422 {
423 fd_set rfds, wfds, xfds;
424 int ret, nfds;
425 struct timeval tv;
426 int timeout;
427
428 if (nonblocking) {
429 timeout = 0;
430 } else {
431 timeout = qemu_calculate_timeout();
432 qemu_bh_update_timeout(&timeout);
433 }
434
435 os_host_main_loop_wait(&timeout);
436
437 tv.tv_sec = timeout / 1000;
438 tv.tv_usec = (timeout % 1000) * 1000;
439
440 /* poll any events */
441 /* XXX: separate device handlers from system ones */
442 nfds = -1;
443 FD_ZERO(&rfds);
444 FD_ZERO(&wfds);
445 FD_ZERO(&xfds);
446
447 #ifdef CONFIG_SLIRP
448 slirp_select_fill(&nfds, &rfds, &wfds, &xfds);
449 #endif
450 qemu_iohandler_fill(&nfds, &rfds, &wfds, &xfds);
451 glib_select_fill(&nfds, &rfds, &wfds, &xfds, &tv);
452
453 if (timeout > 0) {
454 qemu_mutex_unlock_iothread();
455 }
456
457 ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv);
458
459 if (timeout > 0) {
460 qemu_mutex_lock_iothread();
461 }
462
463 glib_select_poll(&rfds, &wfds, &xfds, (ret < 0));
464 qemu_iohandler_poll(&rfds, &wfds, &xfds, ret);
465 #ifdef CONFIG_SLIRP
466 slirp_select_poll(&rfds, &wfds, &xfds, (ret < 0));
467 #endif
468
469 qemu_run_all_timers();
470
471 /* Check bottom-halves last in case any of the earlier events triggered
472 them. */
473 qemu_bh_poll();
474
475 return ret;
476 }