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/osdep.h"
26#include "qapi/error.h"
27#include "qemu/cutils.h"
28#include "qemu/timer.h"
29#include "sysemu/qtest.h"
30#include "sysemu/cpus.h"
31#include "sysemu/replay.h"
32#include "qemu/main-loop.h"
33#include "block/aio.h"
34#include "qemu/error-report.h"
35#include "qemu/queue.h"
36
37#ifndef _WIN32
38#include <sys/wait.h>
39#endif
40
41#ifndef _WIN32
42
43/* If we have signalfd, we mask out the signals we want to handle and then
44 * use signalfd to listen for them. We rely on whatever the current signal
45 * handler is to dispatch the signals when we receive them.
46 */
47static void sigfd_handler(void *opaque)
48{
49 int fd = (intptr_t)opaque;
50 struct qemu_signalfd_siginfo info;
51 struct sigaction action;
52 ssize_t len;
53
54 while (1) {
55 do {
56 len = read(fd, &info, sizeof(info));
57 } while (len == -1 && errno == EINTR);
58
59 if (len == -1 && errno == EAGAIN) {
60 break;
61 }
62
63 if (len != sizeof(info)) {
64 printf("read from sigfd returned %zd: %m\n", len);
65 return;
66 }
67
68 sigaction(info.ssi_signo, NULL, &action);
69 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
70 sigaction_invoke(&action, &info);
71 } else if (action.sa_handler) {
72 action.sa_handler(info.ssi_signo);
73 }
74 }
75}
76
77static int qemu_signal_init(Error **errp)
78{
79 int sigfd;
80 sigset_t set;
81
82 /*
83 * SIG_IPI must be blocked in the main thread and must not be caught
84 * by sigwait() in the signal thread. Otherwise, the cpu thread will
85 * not catch it reliably.
86 */
87 sigemptyset(&set);
88 sigaddset(&set, SIG_IPI);
89 sigaddset(&set, SIGIO);
90 sigaddset(&set, SIGALRM);
91 sigaddset(&set, SIGBUS);
92 /* SIGINT cannot be handled via signalfd, so that ^C can be used
93 * to interrupt QEMU when it is being run under gdb. SIGHUP and
94 * SIGTERM are also handled asynchronously, even though it is not
95 * strictly necessary, because they use the same handler as SIGINT.
96 */
97 pthread_sigmask(SIG_BLOCK, &set, NULL);
98
99 sigdelset(&set, SIG_IPI);
100 sigfd = qemu_signalfd(&set);
101 if (sigfd == -1) {
102 error_setg_errno(errp, errno, "failed to create signalfd");
103 return -errno;
104 }
105
106 fcntl_setfl(sigfd, O_NONBLOCK);
107
108 qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
109
110 return 0;
111}
112
113#else /* _WIN32 */
114
115static int qemu_signal_init(Error **errp)
116{
117 return 0;
118}
119#endif
120
121static AioContext *qemu_aio_context;
122static QEMUBH *qemu_notify_bh;
123
124static void notify_event_cb(void *opaque)
125{
126 /* No need to do anything; this bottom half is only used to
127 * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
128 */
129}
130
131AioContext *qemu_get_aio_context(void)
132{
133 return qemu_aio_context;
134}
135
136void qemu_notify_event(void)
137{
138 if (!qemu_aio_context) {
139 return;
140 }
141 qemu_bh_schedule(qemu_notify_bh);
142}
143
144static GArray *gpollfds;
145
146int qemu_init_main_loop(Error **errp)
147{
148 int ret;
149 GSource *src;
150 Error *local_error = NULL;
151
152 init_clocks(qemu_timer_notify_cb);
153
154 ret = qemu_signal_init(errp);
155 if (ret) {
156 return ret;
157 }
158
159 qemu_aio_context = aio_context_new(&local_error);
160 if (!qemu_aio_context) {
161 error_propagate(errp, local_error);
162 return -EMFILE;
163 }
164 qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
165 gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
166 src = aio_get_g_source(qemu_aio_context);
167 g_source_set_name(src, "aio-context");
168 g_source_attach(src, NULL);
169 g_source_unref(src);
170 src = iohandler_get_g_source();
171 g_source_set_name(src, "io-handler");
172 g_source_attach(src, NULL);
173 g_source_unref(src);
174 return 0;
175}
176
177static int max_priority;
178
179#ifndef _WIN32
180static int glib_pollfds_idx;
181static int glib_n_poll_fds;
182
183static void glib_pollfds_fill(int64_t *cur_timeout)
184{
185 GMainContext *context = g_main_context_default();
186 int timeout = 0;
187 int64_t timeout_ns;
188 int n;
189
190 g_main_context_prepare(context, &max_priority);
191
192 glib_pollfds_idx = gpollfds->len;
193 n = glib_n_poll_fds;
194 do {
195 GPollFD *pfds;
196 glib_n_poll_fds = n;
197 g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
198 pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
199 n = g_main_context_query(context, max_priority, &timeout, pfds,
200 glib_n_poll_fds);
201 } while (n != glib_n_poll_fds);
202
203 if (timeout < 0) {
204 timeout_ns = -1;
205 } else {
206 timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
207 }
208
209 *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
210}
211
212static void glib_pollfds_poll(void)
213{
214 GMainContext *context = g_main_context_default();
215 GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
216
217 if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
218 g_main_context_dispatch(context);
219 }
220}
221
222#define MAX_MAIN_LOOP_SPIN (1000)
223
224static int os_host_main_loop_wait(int64_t timeout)
225{
226 GMainContext *context = g_main_context_default();
227 int ret;
228
229 g_main_context_acquire(context);
230
231 glib_pollfds_fill(&timeout);
232
233 qemu_mutex_unlock_iothread();
234 replay_mutex_unlock();
235
236 ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
237
238 replay_mutex_lock();
239 qemu_mutex_lock_iothread();
240
241 glib_pollfds_poll();
242
243 g_main_context_release(context);
244
245 return ret;
246}
247#else
248/***********************************************************/
249/* Polling handling */
250
251typedef struct PollingEntry {
252 PollingFunc *func;
253 void *opaque;
254 struct PollingEntry *next;
255} PollingEntry;
256
257static PollingEntry *first_polling_entry;
258
259int qemu_add_polling_cb(PollingFunc *func, void *opaque)
260{
261 PollingEntry **ppe, *pe;
262 pe = g_malloc0(sizeof(PollingEntry));
263 pe->func = func;
264 pe->opaque = opaque;
265 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
266 *ppe = pe;
267 return 0;
268}
269
270void qemu_del_polling_cb(PollingFunc *func, void *opaque)
271{
272 PollingEntry **ppe, *pe;
273 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
274 pe = *ppe;
275 if (pe->func == func && pe->opaque == opaque) {
276 *ppe = pe->next;
277 g_free(pe);
278 break;
279 }
280 }
281}
282
283/***********************************************************/
284/* Wait objects support */
285typedef struct WaitObjects {
286 int num;
287 int revents[MAXIMUM_WAIT_OBJECTS + 1];
288 HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
289 WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
290 void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
291} WaitObjects;
292
293static WaitObjects wait_objects = {0};
294
295int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
296{
297 WaitObjects *w = &wait_objects;
298 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
299 return -1;
300 }
301 w->events[w->num] = handle;
302 w->func[w->num] = func;
303 w->opaque[w->num] = opaque;
304 w->revents[w->num] = 0;
305 w->num++;
306 return 0;
307}
308
309void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
310{
311 int i, found;
312 WaitObjects *w = &wait_objects;
313
314 found = 0;
315 for (i = 0; i < w->num; i++) {
316 if (w->events[i] == handle) {
317 found = 1;
318 }
319 if (found) {
320 w->events[i] = w->events[i + 1];
321 w->func[i] = w->func[i + 1];
322 w->opaque[i] = w->opaque[i + 1];
323 w->revents[i] = w->revents[i + 1];
324 }
325 }
326 if (found) {
327 w->num--;
328 }
329}
330
331void qemu_fd_register(int fd)
332{
333 WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
334 FD_READ | FD_ACCEPT | FD_CLOSE |
335 FD_CONNECT | FD_WRITE | FD_OOB);
336}
337
338static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
339 fd_set *xfds)
340{
341 int nfds = -1;
342 int i;
343
344 for (i = 0; i < pollfds->len; i++) {
345 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
346 int fd = pfd->fd;
347 int events = pfd->events;
348 if (events & G_IO_IN) {
349 FD_SET(fd, rfds);
350 nfds = MAX(nfds, fd);
351 }
352 if (events & G_IO_OUT) {
353 FD_SET(fd, wfds);
354 nfds = MAX(nfds, fd);
355 }
356 if (events & G_IO_PRI) {
357 FD_SET(fd, xfds);
358 nfds = MAX(nfds, fd);
359 }
360 }
361 return nfds;
362}
363
364static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
365 fd_set *wfds, fd_set *xfds)
366{
367 int i;
368
369 for (i = 0; i < pollfds->len; i++) {
370 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
371 int fd = pfd->fd;
372 int revents = 0;
373
374 if (FD_ISSET(fd, rfds)) {
375 revents |= G_IO_IN;
376 }
377 if (FD_ISSET(fd, wfds)) {
378 revents |= G_IO_OUT;
379 }
380 if (FD_ISSET(fd, xfds)) {
381 revents |= G_IO_PRI;
382 }
383 pfd->revents = revents & pfd->events;
384 }
385}
386
387static int os_host_main_loop_wait(int64_t timeout)
388{
389 GMainContext *context = g_main_context_default();
390 GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
391 int select_ret = 0;
392 int g_poll_ret, ret, i, n_poll_fds;
393 PollingEntry *pe;
394 WaitObjects *w = &wait_objects;
395 gint poll_timeout;
396 int64_t poll_timeout_ns;
397 static struct timeval tv0;
398 fd_set rfds, wfds, xfds;
399 int nfds;
400
401 g_main_context_acquire(context);
402
403 /* XXX: need to suppress polling by better using win32 events */
404 ret = 0;
405 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
406 ret |= pe->func(pe->opaque);
407 }
408 if (ret != 0) {
409 g_main_context_release(context);
410 return ret;
411 }
412
413 FD_ZERO(&rfds);
414 FD_ZERO(&wfds);
415 FD_ZERO(&xfds);
416 nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
417 if (nfds >= 0) {
418 select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
419 if (select_ret != 0) {
420 timeout = 0;
421 }
422 if (select_ret > 0) {
423 pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
424 }
425 }
426
427 g_main_context_prepare(context, &max_priority);
428 n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
429 poll_fds, ARRAY_SIZE(poll_fds));
430 g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
431
432 for (i = 0; i < w->num; i++) {
433 poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
434 poll_fds[n_poll_fds + i].events = G_IO_IN;
435 }
436
437 if (poll_timeout < 0) {
438 poll_timeout_ns = -1;
439 } else {
440 poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
441 }
442
443 poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
444
445 qemu_mutex_unlock_iothread();
446
447 replay_mutex_unlock();
448
449 g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
450
451 replay_mutex_lock();
452
453 qemu_mutex_lock_iothread();
454 if (g_poll_ret > 0) {
455 for (i = 0; i < w->num; i++) {
456 w->revents[i] = poll_fds[n_poll_fds + i].revents;
457 }
458 for (i = 0; i < w->num; i++) {
459 if (w->revents[i] && w->func[i]) {
460 w->func[i](w->opaque[i]);
461 }
462 }
463 }
464
465 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
466 g_main_context_dispatch(context);
467 }
468
469 g_main_context_release(context);
470
471 return select_ret || g_poll_ret;
472}
473#endif
474
475static NotifierList main_loop_poll_notifiers =
476 NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
477
478void main_loop_poll_add_notifier(Notifier *notify)
479{
480 notifier_list_add(&main_loop_poll_notifiers, notify);
481}
482
483void main_loop_poll_remove_notifier(Notifier *notify)
484{
485 notifier_remove(notify);
486}
487
488void main_loop_wait(int nonblocking)
489{
490 MainLoopPoll mlpoll = {
491 .state = MAIN_LOOP_POLL_FILL,
492 .timeout = UINT32_MAX,
493 .pollfds = gpollfds,
494 };
495 int ret;
496 int64_t timeout_ns;
497
498 if (nonblocking) {
499 mlpoll.timeout = 0;
500 }
501
502 /* poll any events */
503 g_array_set_size(gpollfds, 0); /* reset for new iteration */
504 /* XXX: separate device handlers from system ones */
505 notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
506
507 if (mlpoll.timeout == UINT32_MAX) {
508 timeout_ns = -1;
509 } else {
510 timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
511 }
512
513 timeout_ns = qemu_soonest_timeout(timeout_ns,
514 timerlistgroup_deadline_ns(
515 &main_loop_tlg));
516
517 ret = os_host_main_loop_wait(timeout_ns);
518 mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
519 notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
520
521 /* CPU thread can infinitely wait for event after
522 missing the warp */
523 qemu_start_warp_timer();
524 qemu_clock_run_all_timers();
525}
526
527/* Functions to operate on the main QEMU AioContext. */
528
529QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
530{
531 return aio_bh_new(qemu_aio_context, cb, opaque);
532}
533
534/*
535 * Functions to operate on the I/O handler AioContext.
536 * This context runs on top of main loop. We can't reuse qemu_aio_context
537 * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
538 */
539static AioContext *iohandler_ctx;
540
541static void iohandler_init(void)
542{
543 if (!iohandler_ctx) {
544 iohandler_ctx = aio_context_new(&error_abort);
545 }
546}
547
548AioContext *iohandler_get_aio_context(void)
549{
550 iohandler_init();
551 return iohandler_ctx;
552}
553
554GSource *iohandler_get_g_source(void)
555{
556 iohandler_init();
557 return aio_get_g_source(iohandler_ctx);
558}
559
560void qemu_set_fd_handler(int fd,
561 IOHandler *fd_read,
562 IOHandler *fd_write,
563 void *opaque)
564{
565 iohandler_init();
566 aio_set_fd_handler(iohandler_ctx, fd, false,
567 fd_read, fd_write, NULL, opaque);
568}
569
570void event_notifier_set_handler(EventNotifier *e,
571 EventNotifierHandler *handler)
572{
573 iohandler_init();
574 aio_set_event_notifier(iohandler_ctx, e, false,
575 handler, NULL);
576}
577
578/* reaping of zombies. right now we're not passing the status to
579 anyone, but it would be possible to add a callback. */
580#ifndef _WIN32
581typedef struct ChildProcessRecord {
582 int pid;
583 QLIST_ENTRY(ChildProcessRecord) next;
584} ChildProcessRecord;
585
586static QLIST_HEAD(, ChildProcessRecord) child_watches =
587 QLIST_HEAD_INITIALIZER(child_watches);
588
589static QEMUBH *sigchld_bh;
590
591static void sigchld_handler(int signal)
592{
593 qemu_bh_schedule(sigchld_bh);
594}
595
596static void sigchld_bh_handler(void *opaque)
597{
598 ChildProcessRecord *rec, *next;
599
600 QLIST_FOREACH_SAFE(rec, &child_watches, next, next) {
601 if (waitpid(rec->pid, NULL, WNOHANG) == rec->pid) {
602 QLIST_REMOVE(rec, next);
603 g_free(rec);
604 }
605 }
606}
607
608static void qemu_init_child_watch(void)
609{
610 struct sigaction act;
611 sigchld_bh = qemu_bh_new(sigchld_bh_handler, NULL);
612
613 memset(&act, 0, sizeof(act));
614 act.sa_handler = sigchld_handler;
615 act.sa_flags = SA_NOCLDSTOP;
616 sigaction(SIGCHLD, &act, NULL);
617}
618
619int qemu_add_child_watch(pid_t pid)
620{
621 ChildProcessRecord *rec;
622
623 if (!sigchld_bh) {
624 qemu_init_child_watch();
625 }
626
627 QLIST_FOREACH(rec, &child_watches, next) {
628 if (rec->pid == pid) {
629 return 1;
630 }
631 }
632 rec = g_malloc0(sizeof(ChildProcessRecord));
633 rec->pid = pid;
634 QLIST_INSERT_HEAD(&child_watches, rec, next);
635 return 0;
636}
637#endif
638