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 /* Needed early for CONFIG_BSD etc. */
26 #include "config-host.h"
39 #define SIG_IPI (SIGRTMIN+4)
41 #define SIG_IPI SIGUSR1
44 static CPUState *next_cpu;
46 /***********************************************************/
47 void hw_error(const char *fmt, ...)
53 fprintf(stderr, "qemu: hardware error: ");
54 vfprintf(stderr, fmt, ap);
55 fprintf(stderr, "\n");
56 for(env = first_cpu; env != NULL; env = env->next_cpu) {
57 fprintf(stderr, "CPU #%d:\n", env->cpu_index);
59 cpu_dump_state(env, stderr, fprintf, X86_DUMP_FPU);
61 cpu_dump_state(env, stderr, fprintf, 0);
68 void cpu_synchronize_all_states(void)
72 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
73 cpu_synchronize_state(cpu);
77 void cpu_synchronize_all_post_reset(void)
81 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
82 cpu_synchronize_post_reset(cpu);
86 void cpu_synchronize_all_post_init(void)
90 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
91 cpu_synchronize_post_init(cpu);
95 int cpu_is_stopped(CPUState *env)
97 return !vm_running || env->stopped;
100 static void do_vm_stop(int reason)
106 vm_state_notify(0, reason);
107 monitor_protocol_event(QEVENT_STOP, NULL);
111 static int cpu_can_run(CPUState *env)
115 if (env->stopped || !vm_running)
120 static int cpu_has_work(CPUState *env)
124 if (env->queued_work_first)
126 if (env->stopped || !vm_running)
130 if (qemu_cpu_has_work(env))
135 static int any_cpu_has_work(void)
139 for (env = first_cpu; env != NULL; env = env->next_cpu)
140 if (cpu_has_work(env))
145 static void cpu_debug_handler(CPUState *env)
147 gdb_set_stop_cpu(env);
148 debug_requested = EXCP_DEBUG;
153 static int io_thread_fd = -1;
155 static void qemu_event_increment(void)
157 /* Write 8 bytes to be compatible with eventfd. */
158 static const uint64_t val = 1;
161 if (io_thread_fd == -1)
165 ret = write(io_thread_fd, &val, sizeof(val));
166 } while (ret < 0 && errno == EINTR);
168 /* EAGAIN is fine, a read must be pending. */
169 if (ret < 0 && errno != EAGAIN) {
170 fprintf(stderr, "qemu_event_increment: write() filed: %s\n",
176 static void qemu_event_read(void *opaque)
178 int fd = (unsigned long)opaque;
182 /* Drain the notify pipe. For eventfd, only 8 bytes will be read. */
184 len = read(fd, buffer, sizeof(buffer));
185 } while ((len == -1 && errno == EINTR) || len == sizeof(buffer));
188 static int qemu_event_init(void)
193 err = qemu_eventfd(fds);
197 err = fcntl_setfl(fds[0], O_NONBLOCK);
201 err = fcntl_setfl(fds[1], O_NONBLOCK);
205 qemu_set_fd_handler2(fds[0], NULL, qemu_event_read, NULL,
206 (void *)(unsigned long)fds[0]);
208 io_thread_fd = fds[1];
217 HANDLE qemu_event_handle;
219 static void dummy_event_handler(void *opaque)
223 static int qemu_event_init(void)
225 qemu_event_handle = CreateEvent(NULL, FALSE, FALSE, NULL);
226 if (!qemu_event_handle) {
227 fprintf(stderr, "Failed CreateEvent: %ld\n", GetLastError());
230 qemu_add_wait_object(qemu_event_handle, dummy_event_handler, NULL);
234 static void qemu_event_increment(void)
236 if (!SetEvent(qemu_event_handle)) {
237 fprintf(stderr, "qemu_event_increment: SetEvent failed: %ld\n",
244 #ifndef CONFIG_IOTHREAD
245 int qemu_init_main_loop(void)
247 cpu_set_debug_excp_handler(cpu_debug_handler);
249 return qemu_event_init();
252 void qemu_main_loop_start(void)
256 void qemu_init_vcpu(void *_env)
258 CPUState *env = _env;
260 env->nr_cores = smp_cores;
261 env->nr_threads = smp_threads;
267 int qemu_cpu_self(void *env)
272 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
277 void resume_all_vcpus(void)
281 void pause_all_vcpus(void)
285 void qemu_cpu_kick(void *env)
290 void qemu_notify_event(void)
292 CPUState *env = cpu_single_env;
294 qemu_event_increment ();
298 if (next_cpu && env != next_cpu) {
303 void qemu_mutex_lock_iothread(void) {}
304 void qemu_mutex_unlock_iothread(void) {}
306 void vm_stop(int reason)
311 #else /* CONFIG_IOTHREAD */
313 #include "qemu-thread.h"
315 QemuMutex qemu_global_mutex;
316 static QemuMutex qemu_fair_mutex;
318 static QemuThread io_thread;
320 static QemuThread *tcg_cpu_thread;
321 static QemuCond *tcg_halt_cond;
323 static int qemu_system_ready;
325 static QemuCond qemu_cpu_cond;
327 static QemuCond qemu_system_cond;
328 static QemuCond qemu_pause_cond;
329 static QemuCond qemu_work_cond;
331 static void tcg_init_ipi(void);
332 static void kvm_init_ipi(CPUState *env);
333 static sigset_t block_io_signals(void);
335 /* If we have signalfd, we mask out the signals we want to handle and then
336 * use signalfd to listen for them. We rely on whatever the current signal
337 * handler is to dispatch the signals when we receive them.
339 static void sigfd_handler(void *opaque)
341 int fd = (unsigned long) opaque;
342 struct qemu_signalfd_siginfo info;
343 struct sigaction action;
348 len = read(fd, &info, sizeof(info));
349 } while (len == -1 && errno == EINTR);
351 if (len == -1 && errno == EAGAIN) {
355 if (len != sizeof(info)) {
356 printf("read from sigfd returned %zd: %m\n", len);
360 sigaction(info.ssi_signo, NULL, &action);
361 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
362 action.sa_sigaction(info.ssi_signo,
363 (siginfo_t *)&info, NULL);
364 } else if (action.sa_handler) {
365 action.sa_handler(info.ssi_signo);
370 static int qemu_signalfd_init(sigset_t mask)
374 sigfd = qemu_signalfd(&mask);
376 fprintf(stderr, "failed to create signalfd\n");
380 fcntl_setfl(sigfd, O_NONBLOCK);
382 qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
383 (void *)(unsigned long) sigfd);
388 int qemu_init_main_loop(void)
391 sigset_t blocked_signals;
393 cpu_set_debug_excp_handler(cpu_debug_handler);
395 blocked_signals = block_io_signals();
397 ret = qemu_signalfd_init(blocked_signals);
401 /* Note eventfd must be drained before signalfd handlers run */
402 ret = qemu_event_init();
406 qemu_cond_init(&qemu_pause_cond);
407 qemu_cond_init(&qemu_system_cond);
408 qemu_mutex_init(&qemu_fair_mutex);
409 qemu_mutex_init(&qemu_global_mutex);
410 qemu_mutex_lock(&qemu_global_mutex);
412 qemu_thread_self(&io_thread);
417 void qemu_main_loop_start(void)
419 qemu_system_ready = 1;
420 qemu_cond_broadcast(&qemu_system_cond);
423 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
425 struct qemu_work_item wi;
427 if (qemu_cpu_self(env)) {
434 if (!env->queued_work_first)
435 env->queued_work_first = &wi;
437 env->queued_work_last->next = &wi;
438 env->queued_work_last = &wi;
444 CPUState *self_env = cpu_single_env;
446 qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
447 cpu_single_env = self_env;
451 static void flush_queued_work(CPUState *env)
453 struct qemu_work_item *wi;
455 if (!env->queued_work_first)
458 while ((wi = env->queued_work_first)) {
459 env->queued_work_first = wi->next;
463 env->queued_work_last = NULL;
464 qemu_cond_broadcast(&qemu_work_cond);
467 static void qemu_wait_io_event_common(CPUState *env)
472 qemu_cond_signal(&qemu_pause_cond);
474 flush_queued_work(env);
477 static void qemu_tcg_wait_io_event(void)
481 while (!any_cpu_has_work())
482 qemu_cond_timedwait(tcg_halt_cond, &qemu_global_mutex, 1000);
484 qemu_mutex_unlock(&qemu_global_mutex);
487 * Users of qemu_global_mutex can be starved, having no chance
488 * to acquire it since this path will get to it first.
489 * So use another lock to provide fairness.
491 qemu_mutex_lock(&qemu_fair_mutex);
492 qemu_mutex_unlock(&qemu_fair_mutex);
494 qemu_mutex_lock(&qemu_global_mutex);
496 for (env = first_cpu; env != NULL; env = env->next_cpu) {
497 qemu_wait_io_event_common(env);
501 static void qemu_kvm_eat_signal(CPUState *env, int timeout)
508 ts.tv_sec = timeout / 1000;
509 ts.tv_nsec = (timeout % 1000) * 1000000;
511 sigemptyset(&waitset);
512 sigaddset(&waitset, SIG_IPI);
514 qemu_mutex_unlock(&qemu_global_mutex);
515 r = sigtimedwait(&waitset, &siginfo, &ts);
517 qemu_mutex_lock(&qemu_global_mutex);
519 if (r == -1 && !(e == EAGAIN || e == EINTR)) {
520 fprintf(stderr, "sigtimedwait: %s\n", strerror(e));
525 static void qemu_kvm_wait_io_event(CPUState *env)
527 while (!cpu_has_work(env))
528 qemu_cond_timedwait(env->halt_cond, &qemu_global_mutex, 1000);
530 qemu_kvm_eat_signal(env, 0);
531 qemu_wait_io_event_common(env);
534 static int qemu_cpu_exec(CPUState *env);
536 static void *kvm_cpu_thread_fn(void *arg)
540 qemu_mutex_lock(&qemu_global_mutex);
541 qemu_thread_self(env->thread);
547 /* signal CPU creation */
549 qemu_cond_signal(&qemu_cpu_cond);
551 /* and wait for machine initialization */
552 while (!qemu_system_ready)
553 qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
556 if (cpu_can_run(env))
558 qemu_kvm_wait_io_event(env);
564 static void *tcg_cpu_thread_fn(void *arg)
569 qemu_thread_self(env->thread);
571 /* signal CPU creation */
572 qemu_mutex_lock(&qemu_global_mutex);
573 for (env = first_cpu; env != NULL; env = env->next_cpu)
575 qemu_cond_signal(&qemu_cpu_cond);
577 /* and wait for machine initialization */
578 while (!qemu_system_ready)
579 qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
583 qemu_tcg_wait_io_event();
589 void qemu_cpu_kick(void *_env)
591 CPUState *env = _env;
592 qemu_cond_broadcast(env->halt_cond);
593 qemu_thread_signal(env->thread, SIG_IPI);
596 int qemu_cpu_self(void *_env)
598 CPUState *env = _env;
601 qemu_thread_self(&this);
603 return qemu_thread_equal(&this, env->thread);
606 static void cpu_signal(int sig)
609 cpu_exit(cpu_single_env);
613 static void tcg_init_ipi(void)
616 struct sigaction sigact;
618 memset(&sigact, 0, sizeof(sigact));
619 sigact.sa_handler = cpu_signal;
620 sigaction(SIG_IPI, &sigact, NULL);
623 sigaddset(&set, SIG_IPI);
624 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
627 static void dummy_signal(int sig)
631 static void kvm_init_ipi(CPUState *env)
635 struct sigaction sigact;
637 memset(&sigact, 0, sizeof(sigact));
638 sigact.sa_handler = dummy_signal;
639 sigaction(SIG_IPI, &sigact, NULL);
641 pthread_sigmask(SIG_BLOCK, NULL, &set);
642 sigdelset(&set, SIG_IPI);
643 r = kvm_set_signal_mask(env, &set);
645 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(r));
650 static sigset_t block_io_signals(void)
654 /* SIGUSR2 used by posix-aio-compat.c */
656 sigaddset(&set, SIGUSR2);
657 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
660 sigaddset(&set, SIGIO);
661 sigaddset(&set, SIGALRM);
662 sigaddset(&set, SIG_IPI);
663 pthread_sigmask(SIG_BLOCK, &set, NULL);
668 void qemu_mutex_lock_iothread(void)
671 qemu_mutex_lock(&qemu_fair_mutex);
672 qemu_mutex_lock(&qemu_global_mutex);
673 qemu_mutex_unlock(&qemu_fair_mutex);
675 qemu_mutex_lock(&qemu_fair_mutex);
676 if (qemu_mutex_trylock(&qemu_global_mutex)) {
677 qemu_thread_signal(tcg_cpu_thread, SIG_IPI);
678 qemu_mutex_lock(&qemu_global_mutex);
680 qemu_mutex_unlock(&qemu_fair_mutex);
684 void qemu_mutex_unlock_iothread(void)
686 qemu_mutex_unlock(&qemu_global_mutex);
689 static int all_vcpus_paused(void)
691 CPUState *penv = first_cpu;
696 penv = (CPUState *)penv->next_cpu;
702 void pause_all_vcpus(void)
704 CPUState *penv = first_cpu;
709 penv = (CPUState *)penv->next_cpu;
712 while (!all_vcpus_paused()) {
713 qemu_cond_timedwait(&qemu_pause_cond, &qemu_global_mutex, 100);
717 penv = (CPUState *)penv->next_cpu;
722 void resume_all_vcpus(void)
724 CPUState *penv = first_cpu;
730 penv = (CPUState *)penv->next_cpu;
734 static void tcg_init_vcpu(void *_env)
736 CPUState *env = _env;
737 /* share a single thread for all cpus with TCG */
738 if (!tcg_cpu_thread) {
739 env->thread = qemu_mallocz(sizeof(QemuThread));
740 env->halt_cond = qemu_mallocz(sizeof(QemuCond));
741 qemu_cond_init(env->halt_cond);
742 qemu_thread_create(env->thread, tcg_cpu_thread_fn, env);
743 while (env->created == 0)
744 qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
745 tcg_cpu_thread = env->thread;
746 tcg_halt_cond = env->halt_cond;
748 env->thread = tcg_cpu_thread;
749 env->halt_cond = tcg_halt_cond;
753 static void kvm_start_vcpu(CPUState *env)
755 env->thread = qemu_mallocz(sizeof(QemuThread));
756 env->halt_cond = qemu_mallocz(sizeof(QemuCond));
757 qemu_cond_init(env->halt_cond);
758 qemu_thread_create(env->thread, kvm_cpu_thread_fn, env);
759 while (env->created == 0)
760 qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
763 void qemu_init_vcpu(void *_env)
765 CPUState *env = _env;
767 env->nr_cores = smp_cores;
768 env->nr_threads = smp_threads;
775 void qemu_notify_event(void)
777 qemu_event_increment();
780 static void qemu_system_vmstop_request(int reason)
782 vmstop_requested = reason;
786 void vm_stop(int reason)
789 qemu_thread_self(&me);
791 if (!qemu_thread_equal(&me, &io_thread)) {
792 qemu_system_vmstop_request(reason);
794 * FIXME: should not return to device code in case
795 * vm_stop() has been requested.
797 if (cpu_single_env) {
798 cpu_exit(cpu_single_env);
799 cpu_single_env->stop = 1;
808 static int qemu_cpu_exec(CPUState *env)
811 #ifdef CONFIG_PROFILER
815 #ifdef CONFIG_PROFILER
816 ti = profile_getclock();
821 qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
822 env->icount_decr.u16.low = 0;
823 env->icount_extra = 0;
824 count = qemu_icount_round (qemu_next_deadline());
825 qemu_icount += count;
826 decr = (count > 0xffff) ? 0xffff : count;
828 env->icount_decr.u16.low = decr;
829 env->icount_extra = count;
832 #ifdef CONFIG_PROFILER
833 qemu_time += profile_getclock() - ti;
836 /* Fold pending instructions back into the
837 instruction counter, and clear the interrupt flag. */
838 qemu_icount -= (env->icount_decr.u16.low
839 + env->icount_extra);
840 env->icount_decr.u32 = 0;
841 env->icount_extra = 0;
846 bool cpu_exec_all(void)
848 if (next_cpu == NULL)
849 next_cpu = first_cpu;
850 for (; next_cpu != NULL && !exit_request; next_cpu = next_cpu->next_cpu) {
851 CPUState *env = next_cpu;
853 qemu_clock_enable(vm_clock,
854 (env->singlestep_enabled & SSTEP_NOTIMER) == 0);
856 if (qemu_alarm_pending())
858 if (cpu_can_run(env)) {
859 if (qemu_cpu_exec(env) == EXCP_DEBUG) {
862 } else if (env->stop) {
867 return any_cpu_has_work();
870 void set_numa_modes(void)
875 for (env = first_cpu; env != NULL; env = env->next_cpu) {
876 for (i = 0; i < nb_numa_nodes; i++) {
877 if (node_cpumask[i] & (1 << env->cpu_index)) {
884 void set_cpu_log(const char *optarg)
887 const CPULogItem *item;
889 mask = cpu_str_to_log_mask(optarg);
891 printf("Log items (comma separated):\n");
892 for (item = cpu_log_items; item->mask != 0; item++) {
893 printf("%-10s %s\n", item->name, item->help);
900 /* Return the virtual CPU time, based on the instruction counter. */
901 int64_t cpu_get_icount(void)
904 CPUState *env = cpu_single_env;;
906 icount = qemu_icount;
908 if (!can_do_io(env)) {
909 fprintf(stderr, "Bad clock read\n");
911 icount -= (env->icount_decr.u16.low + env->icount_extra);
913 return qemu_icount_bias + (icount << icount_time_shift);
916 void list_cpus(FILE *f, int (*cpu_fprintf)(FILE *f, const char *fmt, ...),
919 /* XXX: implement xxx_cpu_list for targets that still miss it */
920 #if defined(cpu_list_id)
921 cpu_list_id(f, cpu_fprintf, optarg);
922 #elif defined(cpu_list)
923 cpu_list(f, cpu_fprintf); /* deprecated */