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"
38 #include <sys/prctl.h>
42 #define SIG_IPI (SIGRTMIN+4)
44 #define SIG_IPI SIGUSR1
48 #define PR_MCE_KILL 33
51 static CPUState
*next_cpu
;
53 /***********************************************************/
54 void hw_error(const char *fmt
, ...)
60 fprintf(stderr
, "qemu: hardware error: ");
61 vfprintf(stderr
, fmt
, ap
);
62 fprintf(stderr
, "\n");
63 for(env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
64 fprintf(stderr
, "CPU #%d:\n", env
->cpu_index
);
66 cpu_dump_state(env
, stderr
, fprintf
, X86_DUMP_FPU
);
68 cpu_dump_state(env
, stderr
, fprintf
, 0);
75 void cpu_synchronize_all_states(void)
79 for (cpu
= first_cpu
; cpu
; cpu
= cpu
->next_cpu
) {
80 cpu_synchronize_state(cpu
);
84 void cpu_synchronize_all_post_reset(void)
88 for (cpu
= first_cpu
; cpu
; cpu
= cpu
->next_cpu
) {
89 cpu_synchronize_post_reset(cpu
);
93 void cpu_synchronize_all_post_init(void)
97 for (cpu
= first_cpu
; cpu
; cpu
= cpu
->next_cpu
) {
98 cpu_synchronize_post_init(cpu
);
102 int cpu_is_stopped(CPUState
*env
)
104 return !vm_running
|| env
->stopped
;
107 static void do_vm_stop(int reason
)
113 vm_state_notify(0, reason
);
116 monitor_protocol_event(QEVENT_STOP
, NULL
);
120 static int cpu_can_run(CPUState
*env
)
124 if (env
->stopped
|| !vm_running
)
129 static int cpu_has_work(CPUState
*env
)
133 if (env
->queued_work_first
)
135 if (env
->stopped
|| !vm_running
)
139 if (qemu_cpu_has_work(env
))
144 static int any_cpu_has_work(void)
148 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
)
149 if (cpu_has_work(env
))
154 static void cpu_debug_handler(CPUState
*env
)
156 gdb_set_stop_cpu(env
);
157 debug_requested
= EXCP_DEBUG
;
162 static int io_thread_fd
= -1;
164 static void qemu_event_increment(void)
166 /* Write 8 bytes to be compatible with eventfd. */
167 static const uint64_t val
= 1;
170 if (io_thread_fd
== -1)
174 ret
= write(io_thread_fd
, &val
, sizeof(val
));
175 } while (ret
< 0 && errno
== EINTR
);
177 /* EAGAIN is fine, a read must be pending. */
178 if (ret
< 0 && errno
!= EAGAIN
) {
179 fprintf(stderr
, "qemu_event_increment: write() filed: %s\n",
185 static void qemu_event_read(void *opaque
)
187 int fd
= (unsigned long)opaque
;
191 /* Drain the notify pipe. For eventfd, only 8 bytes will be read. */
193 len
= read(fd
, buffer
, sizeof(buffer
));
194 } while ((len
== -1 && errno
== EINTR
) || len
== sizeof(buffer
));
197 static int qemu_event_init(void)
202 err
= qemu_eventfd(fds
);
206 err
= fcntl_setfl(fds
[0], O_NONBLOCK
);
210 err
= fcntl_setfl(fds
[1], O_NONBLOCK
);
214 qemu_set_fd_handler2(fds
[0], NULL
, qemu_event_read
, NULL
,
215 (void *)(unsigned long)fds
[0]);
217 io_thread_fd
= fds
[1];
226 HANDLE qemu_event_handle
;
228 static void dummy_event_handler(void *opaque
)
232 static int qemu_event_init(void)
234 qemu_event_handle
= CreateEvent(NULL
, FALSE
, FALSE
, NULL
);
235 if (!qemu_event_handle
) {
236 fprintf(stderr
, "Failed CreateEvent: %ld\n", GetLastError());
239 qemu_add_wait_object(qemu_event_handle
, dummy_event_handler
, NULL
);
243 static void qemu_event_increment(void)
245 if (!SetEvent(qemu_event_handle
)) {
246 fprintf(stderr
, "qemu_event_increment: SetEvent failed: %ld\n",
253 #ifndef CONFIG_IOTHREAD
254 int qemu_init_main_loop(void)
256 cpu_set_debug_excp_handler(cpu_debug_handler
);
258 return qemu_event_init();
261 void qemu_main_loop_start(void)
265 void qemu_init_vcpu(void *_env
)
267 CPUState
*env
= _env
;
269 env
->nr_cores
= smp_cores
;
270 env
->nr_threads
= smp_threads
;
276 int qemu_cpu_self(void *env
)
281 void run_on_cpu(CPUState
*env
, void (*func
)(void *data
), void *data
)
286 void resume_all_vcpus(void)
290 void pause_all_vcpus(void)
294 void qemu_cpu_kick(void *env
)
299 void qemu_notify_event(void)
301 CPUState
*env
= cpu_single_env
;
303 qemu_event_increment ();
307 if (next_cpu
&& env
!= next_cpu
) {
313 void qemu_mutex_lock_iothread(void) {}
314 void qemu_mutex_unlock_iothread(void) {}
316 void cpu_stop_current(void)
320 void vm_stop(int reason
)
325 #else /* CONFIG_IOTHREAD */
327 #include "qemu-thread.h"
329 QemuMutex qemu_global_mutex
;
330 static QemuMutex qemu_fair_mutex
;
332 static QemuThread io_thread
;
334 static QemuThread
*tcg_cpu_thread
;
335 static QemuCond
*tcg_halt_cond
;
337 static int qemu_system_ready
;
339 static QemuCond qemu_cpu_cond
;
341 static QemuCond qemu_system_cond
;
342 static QemuCond qemu_pause_cond
;
343 static QemuCond qemu_work_cond
;
345 static void tcg_init_ipi(void);
346 static void kvm_init_ipi(CPUState
*env
);
347 static sigset_t
block_io_signals(void);
349 /* If we have signalfd, we mask out the signals we want to handle and then
350 * use signalfd to listen for them. We rely on whatever the current signal
351 * handler is to dispatch the signals when we receive them.
353 static void sigfd_handler(void *opaque
)
355 int fd
= (unsigned long) opaque
;
356 struct qemu_signalfd_siginfo info
;
357 struct sigaction action
;
362 len
= read(fd
, &info
, sizeof(info
));
363 } while (len
== -1 && errno
== EINTR
);
365 if (len
== -1 && errno
== EAGAIN
) {
369 if (len
!= sizeof(info
)) {
370 printf("read from sigfd returned %zd: %m\n", len
);
374 sigaction(info
.ssi_signo
, NULL
, &action
);
375 if ((action
.sa_flags
& SA_SIGINFO
) && action
.sa_sigaction
) {
376 action
.sa_sigaction(info
.ssi_signo
,
377 (siginfo_t
*)&info
, NULL
);
378 } else if (action
.sa_handler
) {
379 action
.sa_handler(info
.ssi_signo
);
384 static int qemu_signalfd_init(sigset_t mask
)
388 sigfd
= qemu_signalfd(&mask
);
390 fprintf(stderr
, "failed to create signalfd\n");
394 fcntl_setfl(sigfd
, O_NONBLOCK
);
396 qemu_set_fd_handler2(sigfd
, NULL
, sigfd_handler
, NULL
,
397 (void *)(unsigned long) sigfd
);
402 int qemu_init_main_loop(void)
405 sigset_t blocked_signals
;
407 cpu_set_debug_excp_handler(cpu_debug_handler
);
409 blocked_signals
= block_io_signals();
411 ret
= qemu_signalfd_init(blocked_signals
);
415 /* Note eventfd must be drained before signalfd handlers run */
416 ret
= qemu_event_init();
420 qemu_cond_init(&qemu_pause_cond
);
421 qemu_cond_init(&qemu_system_cond
);
422 qemu_mutex_init(&qemu_fair_mutex
);
423 qemu_mutex_init(&qemu_global_mutex
);
424 qemu_mutex_lock(&qemu_global_mutex
);
426 qemu_thread_self(&io_thread
);
431 void qemu_main_loop_start(void)
433 qemu_system_ready
= 1;
434 qemu_cond_broadcast(&qemu_system_cond
);
437 void run_on_cpu(CPUState
*env
, void (*func
)(void *data
), void *data
)
439 struct qemu_work_item wi
;
441 if (qemu_cpu_self(env
)) {
448 if (!env
->queued_work_first
)
449 env
->queued_work_first
= &wi
;
451 env
->queued_work_last
->next
= &wi
;
452 env
->queued_work_last
= &wi
;
458 CPUState
*self_env
= cpu_single_env
;
460 qemu_cond_wait(&qemu_work_cond
, &qemu_global_mutex
);
461 cpu_single_env
= self_env
;
465 static void flush_queued_work(CPUState
*env
)
467 struct qemu_work_item
*wi
;
469 if (!env
->queued_work_first
)
472 while ((wi
= env
->queued_work_first
)) {
473 env
->queued_work_first
= wi
->next
;
477 env
->queued_work_last
= NULL
;
478 qemu_cond_broadcast(&qemu_work_cond
);
481 static void qemu_wait_io_event_common(CPUState
*env
)
486 qemu_cond_signal(&qemu_pause_cond
);
488 flush_queued_work(env
);
489 env
->thread_kicked
= false;
492 static void qemu_tcg_wait_io_event(void)
496 while (!any_cpu_has_work())
497 qemu_cond_timedwait(tcg_halt_cond
, &qemu_global_mutex
, 1000);
499 qemu_mutex_unlock(&qemu_global_mutex
);
502 * Users of qemu_global_mutex can be starved, having no chance
503 * to acquire it since this path will get to it first.
504 * So use another lock to provide fairness.
506 qemu_mutex_lock(&qemu_fair_mutex
);
507 qemu_mutex_unlock(&qemu_fair_mutex
);
509 qemu_mutex_lock(&qemu_global_mutex
);
511 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
512 qemu_wait_io_event_common(env
);
516 static void sigbus_reraise(void)
519 struct sigaction action
;
521 memset(&action
, 0, sizeof(action
));
522 action
.sa_handler
= SIG_DFL
;
523 if (!sigaction(SIGBUS
, &action
, NULL
)) {
526 sigaddset(&set
, SIGBUS
);
527 sigprocmask(SIG_UNBLOCK
, &set
, NULL
);
529 perror("Failed to re-raise SIGBUS!\n");
533 static void sigbus_handler(int n
, struct qemu_signalfd_siginfo
*siginfo
,
536 #if defined(TARGET_I386)
537 if (kvm_on_sigbus(siginfo
->ssi_code
, (void *)(intptr_t)siginfo
->ssi_addr
))
542 static void qemu_kvm_eat_signal(CPUState
*env
, int timeout
)
550 ts
.tv_sec
= timeout
/ 1000;
551 ts
.tv_nsec
= (timeout
% 1000) * 1000000;
553 sigemptyset(&waitset
);
554 sigaddset(&waitset
, SIG_IPI
);
555 sigaddset(&waitset
, SIGBUS
);
558 qemu_mutex_unlock(&qemu_global_mutex
);
560 r
= sigtimedwait(&waitset
, &siginfo
, &ts
);
563 qemu_mutex_lock(&qemu_global_mutex
);
565 if (r
== -1 && !(e
== EAGAIN
|| e
== EINTR
)) {
566 fprintf(stderr
, "sigtimedwait: %s\n", strerror(e
));
573 if (kvm_on_sigbus_vcpu(env
, siginfo
.si_code
, siginfo
.si_addr
))
581 r
= sigpending(&chkset
);
583 fprintf(stderr
, "sigpending: %s\n", strerror(e
));
586 } while (sigismember(&chkset
, SIG_IPI
) || sigismember(&chkset
, SIGBUS
));
589 static void qemu_kvm_wait_io_event(CPUState
*env
)
591 while (!cpu_has_work(env
))
592 qemu_cond_timedwait(env
->halt_cond
, &qemu_global_mutex
, 1000);
594 qemu_kvm_eat_signal(env
, 0);
595 qemu_wait_io_event_common(env
);
598 static int qemu_cpu_exec(CPUState
*env
);
600 static void *kvm_cpu_thread_fn(void *arg
)
604 qemu_mutex_lock(&qemu_global_mutex
);
605 qemu_thread_self(env
->thread
);
611 /* signal CPU creation */
613 qemu_cond_signal(&qemu_cpu_cond
);
615 /* and wait for machine initialization */
616 while (!qemu_system_ready
)
617 qemu_cond_timedwait(&qemu_system_cond
, &qemu_global_mutex
, 100);
620 if (cpu_can_run(env
))
622 qemu_kvm_wait_io_event(env
);
628 static void *tcg_cpu_thread_fn(void *arg
)
633 qemu_thread_self(env
->thread
);
635 /* signal CPU creation */
636 qemu_mutex_lock(&qemu_global_mutex
);
637 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
)
639 qemu_cond_signal(&qemu_cpu_cond
);
641 /* and wait for machine initialization */
642 while (!qemu_system_ready
)
643 qemu_cond_timedwait(&qemu_system_cond
, &qemu_global_mutex
, 100);
647 qemu_tcg_wait_io_event();
653 void qemu_cpu_kick(void *_env
)
655 CPUState
*env
= _env
;
656 qemu_cond_broadcast(env
->halt_cond
);
657 if (!env
->thread_kicked
) {
658 qemu_thread_signal(env
->thread
, SIG_IPI
);
659 env
->thread_kicked
= true;
663 int qemu_cpu_self(void *_env
)
665 CPUState
*env
= _env
;
668 qemu_thread_self(&this);
670 return qemu_thread_equal(&this, env
->thread
);
673 static void cpu_signal(int sig
)
676 cpu_exit(cpu_single_env
);
680 static void tcg_init_ipi(void)
683 struct sigaction sigact
;
685 memset(&sigact
, 0, sizeof(sigact
));
686 sigact
.sa_handler
= cpu_signal
;
687 sigaction(SIG_IPI
, &sigact
, NULL
);
690 sigaddset(&set
, SIG_IPI
);
691 pthread_sigmask(SIG_UNBLOCK
, &set
, NULL
);
694 static void dummy_signal(int sig
)
698 static void kvm_init_ipi(CPUState
*env
)
702 struct sigaction sigact
;
704 memset(&sigact
, 0, sizeof(sigact
));
705 sigact
.sa_handler
= dummy_signal
;
706 sigaction(SIG_IPI
, &sigact
, NULL
);
708 pthread_sigmask(SIG_BLOCK
, NULL
, &set
);
709 sigdelset(&set
, SIG_IPI
);
710 sigdelset(&set
, SIGBUS
);
711 r
= kvm_set_signal_mask(env
, &set
);
713 fprintf(stderr
, "kvm_set_signal_mask: %s\n", strerror(r
));
718 static sigset_t
block_io_signals(void)
721 struct sigaction action
;
723 /* SIGUSR2 used by posix-aio-compat.c */
725 sigaddset(&set
, SIGUSR2
);
726 pthread_sigmask(SIG_UNBLOCK
, &set
, NULL
);
729 sigaddset(&set
, SIGIO
);
730 sigaddset(&set
, SIGALRM
);
731 sigaddset(&set
, SIG_IPI
);
732 sigaddset(&set
, SIGBUS
);
733 pthread_sigmask(SIG_BLOCK
, &set
, NULL
);
735 memset(&action
, 0, sizeof(action
));
736 action
.sa_flags
= SA_SIGINFO
;
737 action
.sa_sigaction
= (void (*)(int, siginfo_t
*, void*))sigbus_handler
;
738 sigaction(SIGBUS
, &action
, NULL
);
739 prctl(PR_MCE_KILL
, 1, 1, 0, 0);
744 void qemu_mutex_lock_iothread(void)
747 qemu_mutex_lock(&qemu_global_mutex
);
749 qemu_mutex_lock(&qemu_fair_mutex
);
750 if (qemu_mutex_trylock(&qemu_global_mutex
)) {
751 qemu_thread_signal(tcg_cpu_thread
, SIG_IPI
);
752 qemu_mutex_lock(&qemu_global_mutex
);
754 qemu_mutex_unlock(&qemu_fair_mutex
);
758 void qemu_mutex_unlock_iothread(void)
760 qemu_mutex_unlock(&qemu_global_mutex
);
763 static int all_vcpus_paused(void)
765 CPUState
*penv
= first_cpu
;
770 penv
= (CPUState
*)penv
->next_cpu
;
776 void pause_all_vcpus(void)
778 CPUState
*penv
= first_cpu
;
783 penv
= (CPUState
*)penv
->next_cpu
;
786 while (!all_vcpus_paused()) {
787 qemu_cond_timedwait(&qemu_pause_cond
, &qemu_global_mutex
, 100);
791 penv
= (CPUState
*)penv
->next_cpu
;
796 void resume_all_vcpus(void)
798 CPUState
*penv
= first_cpu
;
804 penv
= (CPUState
*)penv
->next_cpu
;
808 static void tcg_init_vcpu(void *_env
)
810 CPUState
*env
= _env
;
811 /* share a single thread for all cpus with TCG */
812 if (!tcg_cpu_thread
) {
813 env
->thread
= qemu_mallocz(sizeof(QemuThread
));
814 env
->halt_cond
= qemu_mallocz(sizeof(QemuCond
));
815 qemu_cond_init(env
->halt_cond
);
816 qemu_thread_create(env
->thread
, tcg_cpu_thread_fn
, env
);
817 while (env
->created
== 0)
818 qemu_cond_timedwait(&qemu_cpu_cond
, &qemu_global_mutex
, 100);
819 tcg_cpu_thread
= env
->thread
;
820 tcg_halt_cond
= env
->halt_cond
;
822 env
->thread
= tcg_cpu_thread
;
823 env
->halt_cond
= tcg_halt_cond
;
827 static void kvm_start_vcpu(CPUState
*env
)
829 env
->thread
= qemu_mallocz(sizeof(QemuThread
));
830 env
->halt_cond
= qemu_mallocz(sizeof(QemuCond
));
831 qemu_cond_init(env
->halt_cond
);
832 qemu_thread_create(env
->thread
, kvm_cpu_thread_fn
, env
);
833 while (env
->created
== 0)
834 qemu_cond_timedwait(&qemu_cpu_cond
, &qemu_global_mutex
, 100);
837 void qemu_init_vcpu(void *_env
)
839 CPUState
*env
= _env
;
841 env
->nr_cores
= smp_cores
;
842 env
->nr_threads
= smp_threads
;
849 void qemu_notify_event(void)
851 qemu_event_increment();
854 static void qemu_system_vmstop_request(int reason
)
856 vmstop_requested
= reason
;
860 void cpu_stop_current(void)
862 if (cpu_single_env
) {
863 cpu_single_env
->stopped
= 1;
864 cpu_exit(cpu_single_env
);
868 void vm_stop(int reason
)
871 qemu_thread_self(&me
);
873 if (!qemu_thread_equal(&me
, &io_thread
)) {
874 qemu_system_vmstop_request(reason
);
876 * FIXME: should not return to device code in case
877 * vm_stop() has been requested.
887 static int qemu_cpu_exec(CPUState
*env
)
890 #ifdef CONFIG_PROFILER
894 #ifdef CONFIG_PROFILER
895 ti
= profile_getclock();
900 qemu_icount
-= (env
->icount_decr
.u16
.low
+ env
->icount_extra
);
901 env
->icount_decr
.u16
.low
= 0;
902 env
->icount_extra
= 0;
903 count
= qemu_icount_round (qemu_next_deadline());
904 qemu_icount
+= count
;
905 decr
= (count
> 0xffff) ? 0xffff : count
;
907 env
->icount_decr
.u16
.low
= decr
;
908 env
->icount_extra
= count
;
911 #ifdef CONFIG_PROFILER
912 qemu_time
+= profile_getclock() - ti
;
915 /* Fold pending instructions back into the
916 instruction counter, and clear the interrupt flag. */
917 qemu_icount
-= (env
->icount_decr
.u16
.low
918 + env
->icount_extra
);
919 env
->icount_decr
.u32
= 0;
920 env
->icount_extra
= 0;
925 bool cpu_exec_all(void)
927 if (next_cpu
== NULL
)
928 next_cpu
= first_cpu
;
929 for (; next_cpu
!= NULL
&& !exit_request
; next_cpu
= next_cpu
->next_cpu
) {
930 CPUState
*env
= next_cpu
;
932 qemu_clock_enable(vm_clock
,
933 (env
->singlestep_enabled
& SSTEP_NOTIMER
) == 0);
935 if (qemu_alarm_pending())
937 if (cpu_can_run(env
)) {
938 if (qemu_cpu_exec(env
) == EXCP_DEBUG
) {
941 } else if (env
->stop
) {
946 return any_cpu_has_work();
949 void set_numa_modes(void)
954 for (env
= first_cpu
; env
!= NULL
; env
= env
->next_cpu
) {
955 for (i
= 0; i
< nb_numa_nodes
; i
++) {
956 if (node_cpumask
[i
] & (1 << env
->cpu_index
)) {
963 void set_cpu_log(const char *optarg
)
966 const CPULogItem
*item
;
968 mask
= cpu_str_to_log_mask(optarg
);
970 printf("Log items (comma separated):\n");
971 for (item
= cpu_log_items
; item
->mask
!= 0; item
++) {
972 printf("%-10s %s\n", item
->name
, item
->help
);
979 /* Return the virtual CPU time, based on the instruction counter. */
980 int64_t cpu_get_icount(void)
983 CPUState
*env
= cpu_single_env
;;
985 icount
= qemu_icount
;
987 if (!can_do_io(env
)) {
988 fprintf(stderr
, "Bad clock read\n");
990 icount
-= (env
->icount_decr
.u16
.low
+ env
->icount_extra
);
992 return qemu_icount_bias
+ (icount
<< icount_time_shift
);
995 void list_cpus(FILE *f
, fprintf_function cpu_fprintf
, const char *optarg
)
997 /* XXX: implement xxx_cpu_list for targets that still miss it */
998 #if defined(cpu_list_id)
999 cpu_list_id(f
, cpu_fprintf
, optarg
);
1000 #elif defined(cpu_list)
1001 cpu_list(f
, cpu_fprintf
); /* deprecated */