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"
28 #include "monitor/monitor.h"
29 #include "sysemu/sysemu.h"
30 #include "exec/gdbstub.h"
31 #include "sysemu/dma.h"
32 #include "sysemu/kvm.h"
33 #include "qmp-commands.h"
35 #include "qemu/thread.h"
36 #include "sysemu/cpus.h"
37 #include "sysemu/qtest.h"
38 #include "qemu/main-loop.h"
39 #include "qemu/bitmap.h"
42 #include "qemu/compatfd.h"
47 #include <sys/prctl.h>
50 #define PR_MCE_KILL 33
53 #ifndef PR_MCE_KILL_SET
54 #define PR_MCE_KILL_SET 1
57 #ifndef PR_MCE_KILL_EARLY
58 #define PR_MCE_KILL_EARLY 1
61 #endif /* CONFIG_LINUX */
63 static CPUState *next_cpu;
65 static bool cpu_thread_is_idle(CPUState *cpu)
67 if (cpu->stop || cpu->queued_work_first) {
70 if (cpu->stopped || !runstate_is_running()) {
73 if (!cpu->halted || qemu_cpu_has_work(cpu) ||
74 kvm_halt_in_kernel()) {
80 static bool all_cpu_threads_idle(void)
84 for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
85 if (!cpu_thread_is_idle(cpu)) {
92 /***********************************************************/
93 /* guest cycle counter */
95 /* Conversion factor from emulated instructions to virtual clock ticks. */
96 static int icount_time_shift;
97 /* Arbitrarily pick 1MIPS as the minimum allowable speed. */
98 #define MAX_ICOUNT_SHIFT 10
99 /* Compensate for varying guest execution speed. */
100 static int64_t qemu_icount_bias;
101 static QEMUTimer *icount_rt_timer;
102 static QEMUTimer *icount_vm_timer;
103 static QEMUTimer *icount_warp_timer;
104 static int64_t vm_clock_warp_start;
105 static int64_t qemu_icount;
107 typedef struct TimersState {
108 int64_t cpu_ticks_prev;
109 int64_t cpu_ticks_offset;
110 int64_t cpu_clock_offset;
111 int32_t cpu_ticks_enabled;
115 TimersState timers_state;
117 /* Return the virtual CPU time, based on the instruction counter. */
118 int64_t cpu_get_icount(void)
121 CPUState *cpu = current_cpu;
123 icount = qemu_icount;
125 CPUArchState *env = cpu->env_ptr;
126 if (!can_do_io(env)) {
127 fprintf(stderr, "Bad clock read\n");
129 icount -= (env->icount_decr.u16.low + env->icount_extra);
131 return qemu_icount_bias + (icount << icount_time_shift);
134 /* return the host CPU cycle counter and handle stop/restart */
135 int64_t cpu_get_ticks(void)
138 return cpu_get_icount();
140 if (!timers_state.cpu_ticks_enabled) {
141 return timers_state.cpu_ticks_offset;
144 ticks = cpu_get_real_ticks();
145 if (timers_state.cpu_ticks_prev > ticks) {
146 /* Note: non increasing ticks may happen if the host uses
148 timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks;
150 timers_state.cpu_ticks_prev = ticks;
151 return ticks + timers_state.cpu_ticks_offset;
155 /* return the host CPU monotonic timer and handle stop/restart */
156 int64_t cpu_get_clock(void)
159 if (!timers_state.cpu_ticks_enabled) {
160 return timers_state.cpu_clock_offset;
163 return ti + timers_state.cpu_clock_offset;
167 /* enable cpu_get_ticks() */
168 void cpu_enable_ticks(void)
170 if (!timers_state.cpu_ticks_enabled) {
171 timers_state.cpu_ticks_offset -= cpu_get_real_ticks();
172 timers_state.cpu_clock_offset -= get_clock();
173 timers_state.cpu_ticks_enabled = 1;
177 /* disable cpu_get_ticks() : the clock is stopped. You must not call
178 cpu_get_ticks() after that. */
179 void cpu_disable_ticks(void)
181 if (timers_state.cpu_ticks_enabled) {
182 timers_state.cpu_ticks_offset = cpu_get_ticks();
183 timers_state.cpu_clock_offset = cpu_get_clock();
184 timers_state.cpu_ticks_enabled = 0;
188 /* Correlation between real and virtual time is always going to be
189 fairly approximate, so ignore small variation.
190 When the guest is idle real and virtual time will be aligned in
192 #define ICOUNT_WOBBLE (get_ticks_per_sec() / 10)
194 static void icount_adjust(void)
199 static int64_t last_delta;
200 /* If the VM is not running, then do nothing. */
201 if (!runstate_is_running()) {
204 cur_time = cpu_get_clock();
205 cur_icount = qemu_get_clock_ns(vm_clock);
206 delta = cur_icount - cur_time;
207 /* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */
209 && last_delta + ICOUNT_WOBBLE < delta * 2
210 && icount_time_shift > 0) {
211 /* The guest is getting too far ahead. Slow time down. */
215 && last_delta - ICOUNT_WOBBLE > delta * 2
216 && icount_time_shift < MAX_ICOUNT_SHIFT) {
217 /* The guest is getting too far behind. Speed time up. */
221 qemu_icount_bias = cur_icount - (qemu_icount << icount_time_shift);
224 static void icount_adjust_rt(void *opaque)
226 qemu_mod_timer(icount_rt_timer,
227 qemu_get_clock_ms(rt_clock) + 1000);
231 static void icount_adjust_vm(void *opaque)
233 qemu_mod_timer(icount_vm_timer,
234 qemu_get_clock_ns(vm_clock) + get_ticks_per_sec() / 10);
238 static int64_t qemu_icount_round(int64_t count)
240 return (count + (1 << icount_time_shift) - 1) >> icount_time_shift;
243 static void icount_warp_rt(void *opaque)
245 if (vm_clock_warp_start == -1) {
249 if (runstate_is_running()) {
250 int64_t clock = qemu_get_clock_ns(rt_clock);
251 int64_t warp_delta = clock - vm_clock_warp_start;
252 if (use_icount == 1) {
253 qemu_icount_bias += warp_delta;
256 * In adaptive mode, do not let the vm_clock run too
257 * far ahead of real time.
259 int64_t cur_time = cpu_get_clock();
260 int64_t cur_icount = qemu_get_clock_ns(vm_clock);
261 int64_t delta = cur_time - cur_icount;
262 qemu_icount_bias += MIN(warp_delta, delta);
264 if (qemu_clock_expired(vm_clock)) {
268 vm_clock_warp_start = -1;
271 void qtest_clock_warp(int64_t dest)
273 int64_t clock = qemu_get_clock_ns(vm_clock);
274 assert(qtest_enabled());
275 while (clock < dest) {
276 int64_t deadline = qemu_clock_deadline(vm_clock);
277 int64_t warp = MIN(dest - clock, deadline);
278 qemu_icount_bias += warp;
279 qemu_run_timers(vm_clock);
280 clock = qemu_get_clock_ns(vm_clock);
285 void qemu_clock_warp(QEMUClock *clock)
290 * There are too many global variables to make the "warp" behavior
291 * applicable to other clocks. But a clock argument removes the
292 * need for if statements all over the place.
294 if (clock != vm_clock || !use_icount) {
299 * If the CPUs have been sleeping, advance the vm_clock timer now. This
300 * ensures that the deadline for the timer is computed correctly below.
301 * This also makes sure that the insn counter is synchronized before the
302 * CPU starts running, in case the CPU is woken by an event other than
303 * the earliest vm_clock timer.
305 icount_warp_rt(NULL);
306 if (!all_cpu_threads_idle() || !qemu_clock_has_timers(vm_clock)) {
307 qemu_del_timer(icount_warp_timer);
311 if (qtest_enabled()) {
312 /* When testing, qtest commands advance icount. */
316 vm_clock_warp_start = qemu_get_clock_ns(rt_clock);
317 deadline = qemu_clock_deadline(vm_clock);
320 * Ensure the vm_clock proceeds even when the virtual CPU goes to
321 * sleep. Otherwise, the CPU might be waiting for a future timer
322 * interrupt to wake it up, but the interrupt never comes because
323 * the vCPU isn't running any insns and thus doesn't advance the
326 * An extreme solution for this problem would be to never let VCPUs
327 * sleep in icount mode if there is a pending vm_clock timer; rather
328 * time could just advance to the next vm_clock event. Instead, we
329 * do stop VCPUs and only advance vm_clock after some "real" time,
330 * (related to the time left until the next event) has passed. This
331 * rt_clock timer will do this. This avoids that the warps are too
332 * visible externally---for example, you will not be sending network
333 * packets continuously instead of every 100ms.
335 qemu_mod_timer(icount_warp_timer, vm_clock_warp_start + deadline);
341 static const VMStateDescription vmstate_timers = {
344 .minimum_version_id = 1,
345 .minimum_version_id_old = 1,
346 .fields = (VMStateField[]) {
347 VMSTATE_INT64(cpu_ticks_offset, TimersState),
348 VMSTATE_INT64(dummy, TimersState),
349 VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
350 VMSTATE_END_OF_LIST()
354 void configure_icount(const char *option)
356 vmstate_register(NULL, 0, &vmstate_timers, &timers_state);
361 icount_warp_timer = qemu_new_timer_ns(rt_clock, icount_warp_rt, NULL);
362 if (strcmp(option, "auto") != 0) {
363 icount_time_shift = strtol(option, NULL, 0);
370 /* 125MIPS seems a reasonable initial guess at the guest speed.
371 It will be corrected fairly quickly anyway. */
372 icount_time_shift = 3;
374 /* Have both realtime and virtual time triggers for speed adjustment.
375 The realtime trigger catches emulated time passing too slowly,
376 the virtual time trigger catches emulated time passing too fast.
377 Realtime triggers occur even when idle, so use them less frequently
379 icount_rt_timer = qemu_new_timer_ms(rt_clock, icount_adjust_rt, NULL);
380 qemu_mod_timer(icount_rt_timer,
381 qemu_get_clock_ms(rt_clock) + 1000);
382 icount_vm_timer = qemu_new_timer_ns(vm_clock, icount_adjust_vm, NULL);
383 qemu_mod_timer(icount_vm_timer,
384 qemu_get_clock_ns(vm_clock) + get_ticks_per_sec() / 10);
387 /***********************************************************/
388 void hw_error(const char *fmt, ...)
394 fprintf(stderr, "qemu: hardware error: ");
395 vfprintf(stderr, fmt, ap);
396 fprintf(stderr, "\n");
397 for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
398 fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
399 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU);
405 void cpu_synchronize_all_states(void)
409 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
410 cpu_synchronize_state(cpu);
414 void cpu_synchronize_all_post_reset(void)
418 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
419 cpu_synchronize_post_reset(cpu);
423 void cpu_synchronize_all_post_init(void)
427 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
428 cpu_synchronize_post_init(cpu);
432 bool cpu_is_stopped(CPUState *cpu)
434 return !runstate_is_running() || cpu->stopped;
437 static void do_vm_stop(RunState state)
439 if (runstate_is_running()) {
443 vm_state_notify(0, state);
446 monitor_protocol_event(QEVENT_STOP, NULL);
450 static bool cpu_can_run(CPUState *cpu)
455 if (cpu->stopped || !runstate_is_running()) {
461 static void cpu_handle_guest_debug(CPUState *cpu)
463 gdb_set_stop_cpu(cpu);
464 qemu_system_debug_request();
468 static void cpu_signal(int sig)
471 cpu_exit(current_cpu);
477 static void sigbus_reraise(void)
480 struct sigaction action;
482 memset(&action, 0, sizeof(action));
483 action.sa_handler = SIG_DFL;
484 if (!sigaction(SIGBUS, &action, NULL)) {
487 sigaddset(&set, SIGBUS);
488 sigprocmask(SIG_UNBLOCK, &set, NULL);
490 perror("Failed to re-raise SIGBUS!\n");
494 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
497 if (kvm_on_sigbus(siginfo->ssi_code,
498 (void *)(intptr_t)siginfo->ssi_addr)) {
503 static void qemu_init_sigbus(void)
505 struct sigaction action;
507 memset(&action, 0, sizeof(action));
508 action.sa_flags = SA_SIGINFO;
509 action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
510 sigaction(SIGBUS, &action, NULL);
512 prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
515 static void qemu_kvm_eat_signals(CPUState *cpu)
517 struct timespec ts = { 0, 0 };
523 sigemptyset(&waitset);
524 sigaddset(&waitset, SIG_IPI);
525 sigaddset(&waitset, SIGBUS);
528 r = sigtimedwait(&waitset, &siginfo, &ts);
529 if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
530 perror("sigtimedwait");
536 if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) {
544 r = sigpending(&chkset);
546 perror("sigpending");
549 } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
552 #else /* !CONFIG_LINUX */
554 static void qemu_init_sigbus(void)
558 static void qemu_kvm_eat_signals(CPUState *cpu)
561 #endif /* !CONFIG_LINUX */
564 static void dummy_signal(int sig)
568 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
572 struct sigaction sigact;
574 memset(&sigact, 0, sizeof(sigact));
575 sigact.sa_handler = dummy_signal;
576 sigaction(SIG_IPI, &sigact, NULL);
578 pthread_sigmask(SIG_BLOCK, NULL, &set);
579 sigdelset(&set, SIG_IPI);
580 sigdelset(&set, SIGBUS);
581 r = kvm_set_signal_mask(cpu, &set);
583 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
588 static void qemu_tcg_init_cpu_signals(void)
591 struct sigaction sigact;
593 memset(&sigact, 0, sizeof(sigact));
594 sigact.sa_handler = cpu_signal;
595 sigaction(SIG_IPI, &sigact, NULL);
598 sigaddset(&set, SIG_IPI);
599 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
603 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
608 static void qemu_tcg_init_cpu_signals(void)
613 static QemuMutex qemu_global_mutex;
614 static QemuCond qemu_io_proceeded_cond;
615 static bool iothread_requesting_mutex;
617 static QemuThread io_thread;
619 static QemuThread *tcg_cpu_thread;
620 static QemuCond *tcg_halt_cond;
623 static QemuCond qemu_cpu_cond;
625 static QemuCond qemu_pause_cond;
626 static QemuCond qemu_work_cond;
628 void qemu_init_cpu_loop(void)
631 qemu_cond_init(&qemu_cpu_cond);
632 qemu_cond_init(&qemu_pause_cond);
633 qemu_cond_init(&qemu_work_cond);
634 qemu_cond_init(&qemu_io_proceeded_cond);
635 qemu_mutex_init(&qemu_global_mutex);
637 qemu_thread_get_self(&io_thread);
640 void run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
642 struct qemu_work_item wi;
644 if (qemu_cpu_is_self(cpu)) {
652 if (cpu->queued_work_first == NULL) {
653 cpu->queued_work_first = &wi;
655 cpu->queued_work_last->next = &wi;
657 cpu->queued_work_last = &wi;
663 CPUState *self_cpu = current_cpu;
665 qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
666 current_cpu = self_cpu;
670 void async_run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
672 struct qemu_work_item *wi;
674 if (qemu_cpu_is_self(cpu)) {
679 wi = g_malloc0(sizeof(struct qemu_work_item));
683 if (cpu->queued_work_first == NULL) {
684 cpu->queued_work_first = wi;
686 cpu->queued_work_last->next = wi;
688 cpu->queued_work_last = wi;
695 static void flush_queued_work(CPUState *cpu)
697 struct qemu_work_item *wi;
699 if (cpu->queued_work_first == NULL) {
703 while ((wi = cpu->queued_work_first)) {
704 cpu->queued_work_first = wi->next;
711 cpu->queued_work_last = NULL;
712 qemu_cond_broadcast(&qemu_work_cond);
715 static void qemu_wait_io_event_common(CPUState *cpu)
720 qemu_cond_signal(&qemu_pause_cond);
722 flush_queued_work(cpu);
723 cpu->thread_kicked = false;
726 static void qemu_tcg_wait_io_event(void)
730 while (all_cpu_threads_idle()) {
731 /* Start accounting real time to the virtual clock if the CPUs
733 qemu_clock_warp(vm_clock);
734 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
737 while (iothread_requesting_mutex) {
738 qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex);
741 for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
742 qemu_wait_io_event_common(cpu);
746 static void qemu_kvm_wait_io_event(CPUState *cpu)
748 while (cpu_thread_is_idle(cpu)) {
749 qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
752 qemu_kvm_eat_signals(cpu);
753 qemu_wait_io_event_common(cpu);
756 static void *qemu_kvm_cpu_thread_fn(void *arg)
761 qemu_mutex_lock(&qemu_global_mutex);
762 qemu_thread_get_self(cpu->thread);
763 cpu->thread_id = qemu_get_thread_id();
766 r = kvm_init_vcpu(cpu);
768 fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
772 qemu_kvm_init_cpu_signals(cpu);
774 /* signal CPU creation */
776 qemu_cond_signal(&qemu_cpu_cond);
779 if (cpu_can_run(cpu)) {
780 r = kvm_cpu_exec(cpu);
781 if (r == EXCP_DEBUG) {
782 cpu_handle_guest_debug(cpu);
785 qemu_kvm_wait_io_event(cpu);
791 static void *qemu_dummy_cpu_thread_fn(void *arg)
794 fprintf(stderr, "qtest is not supported under Windows\n");
801 qemu_mutex_lock_iothread();
802 qemu_thread_get_self(cpu->thread);
803 cpu->thread_id = qemu_get_thread_id();
805 sigemptyset(&waitset);
806 sigaddset(&waitset, SIG_IPI);
808 /* signal CPU creation */
810 qemu_cond_signal(&qemu_cpu_cond);
815 qemu_mutex_unlock_iothread();
818 r = sigwait(&waitset, &sig);
819 } while (r == -1 && (errno == EAGAIN || errno == EINTR));
824 qemu_mutex_lock_iothread();
826 qemu_wait_io_event_common(cpu);
833 static void tcg_exec_all(void);
835 static void tcg_signal_cpu_creation(CPUState *cpu, void *data)
837 cpu->thread_id = qemu_get_thread_id();
841 static void *qemu_tcg_cpu_thread_fn(void *arg)
845 qemu_tcg_init_cpu_signals();
846 qemu_thread_get_self(cpu->thread);
848 qemu_mutex_lock(&qemu_global_mutex);
849 qemu_for_each_cpu(tcg_signal_cpu_creation, NULL);
850 qemu_cond_signal(&qemu_cpu_cond);
852 /* wait for initial kick-off after machine start */
853 while (first_cpu->stopped) {
854 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
856 /* process any pending work */
857 for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
858 qemu_wait_io_event_common(cpu);
864 if (use_icount && qemu_clock_deadline(vm_clock) <= 0) {
867 qemu_tcg_wait_io_event();
873 static void qemu_cpu_kick_thread(CPUState *cpu)
878 err = pthread_kill(cpu->thread->thread, SIG_IPI);
880 fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
884 if (!qemu_cpu_is_self(cpu)) {
887 if (SuspendThread(cpu->hThread) == (DWORD)-1) {
888 fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
893 /* On multi-core systems, we are not sure that the thread is actually
894 * suspended until we can get the context.
896 tcgContext.ContextFlags = CONTEXT_CONTROL;
897 while (GetThreadContext(cpu->hThread, &tcgContext) != 0) {
903 if (ResumeThread(cpu->hThread) == (DWORD)-1) {
904 fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
912 void qemu_cpu_kick(CPUState *cpu)
914 qemu_cond_broadcast(cpu->halt_cond);
915 if (!tcg_enabled() && !cpu->thread_kicked) {
916 qemu_cpu_kick_thread(cpu);
917 cpu->thread_kicked = true;
921 void qemu_cpu_kick_self(void)
926 if (!current_cpu->thread_kicked) {
927 qemu_cpu_kick_thread(current_cpu);
928 current_cpu->thread_kicked = true;
935 bool qemu_cpu_is_self(CPUState *cpu)
937 return qemu_thread_is_self(cpu->thread);
940 static bool qemu_in_vcpu_thread(void)
942 return current_cpu && qemu_cpu_is_self(current_cpu);
945 void qemu_mutex_lock_iothread(void)
947 if (!tcg_enabled()) {
948 qemu_mutex_lock(&qemu_global_mutex);
950 iothread_requesting_mutex = true;
951 if (qemu_mutex_trylock(&qemu_global_mutex)) {
952 qemu_cpu_kick_thread(first_cpu);
953 qemu_mutex_lock(&qemu_global_mutex);
955 iothread_requesting_mutex = false;
956 qemu_cond_broadcast(&qemu_io_proceeded_cond);
960 void qemu_mutex_unlock_iothread(void)
962 qemu_mutex_unlock(&qemu_global_mutex);
965 static int all_vcpus_paused(void)
967 CPUState *cpu = first_cpu;
979 void pause_all_vcpus(void)
981 CPUState *cpu = first_cpu;
983 qemu_clock_enable(vm_clock, false);
990 if (qemu_in_vcpu_thread()) {
992 if (!kvm_enabled()) {
1003 while (!all_vcpus_paused()) {
1004 qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
1008 cpu = cpu->next_cpu;
1013 void cpu_resume(CPUState *cpu)
1016 cpu->stopped = false;
1020 void resume_all_vcpus(void)
1022 CPUState *cpu = first_cpu;
1024 qemu_clock_enable(vm_clock, true);
1027 cpu = cpu->next_cpu;
1031 static void qemu_tcg_init_vcpu(CPUState *cpu)
1033 /* share a single thread for all cpus with TCG */
1034 if (!tcg_cpu_thread) {
1035 cpu->thread = g_malloc0(sizeof(QemuThread));
1036 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1037 qemu_cond_init(cpu->halt_cond);
1038 tcg_halt_cond = cpu->halt_cond;
1039 qemu_thread_create(cpu->thread, qemu_tcg_cpu_thread_fn, cpu,
1040 QEMU_THREAD_JOINABLE);
1042 cpu->hThread = qemu_thread_get_handle(cpu->thread);
1044 while (!cpu->created) {
1045 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1047 tcg_cpu_thread = cpu->thread;
1049 cpu->thread = tcg_cpu_thread;
1050 cpu->halt_cond = tcg_halt_cond;
1054 static void qemu_kvm_start_vcpu(CPUState *cpu)
1056 cpu->thread = g_malloc0(sizeof(QemuThread));
1057 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1058 qemu_cond_init(cpu->halt_cond);
1059 qemu_thread_create(cpu->thread, qemu_kvm_cpu_thread_fn, cpu,
1060 QEMU_THREAD_JOINABLE);
1061 while (!cpu->created) {
1062 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1066 static void qemu_dummy_start_vcpu(CPUState *cpu)
1068 cpu->thread = g_malloc0(sizeof(QemuThread));
1069 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1070 qemu_cond_init(cpu->halt_cond);
1071 qemu_thread_create(cpu->thread, qemu_dummy_cpu_thread_fn, cpu,
1072 QEMU_THREAD_JOINABLE);
1073 while (!cpu->created) {
1074 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1078 void qemu_init_vcpu(CPUState *cpu)
1080 cpu->nr_cores = smp_cores;
1081 cpu->nr_threads = smp_threads;
1082 cpu->stopped = true;
1083 if (kvm_enabled()) {
1084 qemu_kvm_start_vcpu(cpu);
1085 } else if (tcg_enabled()) {
1086 qemu_tcg_init_vcpu(cpu);
1088 qemu_dummy_start_vcpu(cpu);
1092 void cpu_stop_current(void)
1095 current_cpu->stop = false;
1096 current_cpu->stopped = true;
1097 cpu_exit(current_cpu);
1098 qemu_cond_signal(&qemu_pause_cond);
1102 void vm_stop(RunState state)
1104 if (qemu_in_vcpu_thread()) {
1105 qemu_system_vmstop_request(state);
1107 * FIXME: should not return to device code in case
1108 * vm_stop() has been requested.
1116 /* does a state transition even if the VM is already stopped,
1117 current state is forgotten forever */
1118 void vm_stop_force_state(RunState state)
1120 if (runstate_is_running()) {
1123 runstate_set(state);
1127 static int tcg_cpu_exec(CPUArchState *env)
1130 #ifdef CONFIG_PROFILER
1134 #ifdef CONFIG_PROFILER
1135 ti = profile_getclock();
1140 qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
1141 env->icount_decr.u16.low = 0;
1142 env->icount_extra = 0;
1143 count = qemu_icount_round(qemu_clock_deadline(vm_clock));
1144 qemu_icount += count;
1145 decr = (count > 0xffff) ? 0xffff : count;
1147 env->icount_decr.u16.low = decr;
1148 env->icount_extra = count;
1150 ret = cpu_exec(env);
1151 #ifdef CONFIG_PROFILER
1152 qemu_time += profile_getclock() - ti;
1155 /* Fold pending instructions back into the
1156 instruction counter, and clear the interrupt flag. */
1157 qemu_icount -= (env->icount_decr.u16.low
1158 + env->icount_extra);
1159 env->icount_decr.u32 = 0;
1160 env->icount_extra = 0;
1165 static void tcg_exec_all(void)
1169 /* Account partial waits to the vm_clock. */
1170 qemu_clock_warp(vm_clock);
1172 if (next_cpu == NULL) {
1173 next_cpu = first_cpu;
1175 for (; next_cpu != NULL && !exit_request; next_cpu = next_cpu->next_cpu) {
1176 CPUState *cpu = next_cpu;
1177 CPUArchState *env = cpu->env_ptr;
1179 qemu_clock_enable(vm_clock,
1180 (env->singlestep_enabled & SSTEP_NOTIMER) == 0);
1182 if (cpu_can_run(cpu)) {
1183 r = tcg_cpu_exec(env);
1184 if (r == EXCP_DEBUG) {
1185 cpu_handle_guest_debug(cpu);
1188 } else if (cpu->stop || cpu->stopped) {
1195 void set_numa_modes(void)
1200 for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
1201 for (i = 0; i < nb_numa_nodes; i++) {
1202 if (test_bit(cpu->cpu_index, node_cpumask[i])) {
1209 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1211 /* XXX: implement xxx_cpu_list for targets that still miss it */
1212 #if defined(cpu_list)
1213 cpu_list(f, cpu_fprintf);
1217 CpuInfoList *qmp_query_cpus(Error **errp)
1219 CpuInfoList *head = NULL, *cur_item = NULL;
1222 for (cpu = first_cpu; cpu != NULL; cpu = cpu->next_cpu) {
1224 #if defined(TARGET_I386)
1225 X86CPU *x86_cpu = X86_CPU(cpu);
1226 CPUX86State *env = &x86_cpu->env;
1227 #elif defined(TARGET_PPC)
1228 PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu);
1229 CPUPPCState *env = &ppc_cpu->env;
1230 #elif defined(TARGET_SPARC)
1231 SPARCCPU *sparc_cpu = SPARC_CPU(cpu);
1232 CPUSPARCState *env = &sparc_cpu->env;
1233 #elif defined(TARGET_MIPS)
1234 MIPSCPU *mips_cpu = MIPS_CPU(cpu);
1235 CPUMIPSState *env = &mips_cpu->env;
1238 cpu_synchronize_state(cpu);
1240 info = g_malloc0(sizeof(*info));
1241 info->value = g_malloc0(sizeof(*info->value));
1242 info->value->CPU = cpu->cpu_index;
1243 info->value->current = (cpu == first_cpu);
1244 info->value->halted = cpu->halted;
1245 info->value->thread_id = cpu->thread_id;
1246 #if defined(TARGET_I386)
1247 info->value->has_pc = true;
1248 info->value->pc = env->eip + env->segs[R_CS].base;
1249 #elif defined(TARGET_PPC)
1250 info->value->has_nip = true;
1251 info->value->nip = env->nip;
1252 #elif defined(TARGET_SPARC)
1253 info->value->has_pc = true;
1254 info->value->pc = env->pc;
1255 info->value->has_npc = true;
1256 info->value->npc = env->npc;
1257 #elif defined(TARGET_MIPS)
1258 info->value->has_PC = true;
1259 info->value->PC = env->active_tc.PC;
1262 /* XXX: waiting for the qapi to support GSList */
1264 head = cur_item = info;
1266 cur_item->next = info;
1274 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
1275 bool has_cpu, int64_t cpu_index, Error **errp)
1287 cpu = qemu_get_cpu(cpu_index);
1289 error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
1295 f = fopen(filename, "wb");
1297 error_setg_file_open(errp, errno, filename);
1305 cpu_memory_rw_debug(env, addr, buf, l, 0);
1306 if (fwrite(buf, 1, l, f) != l) {
1307 error_set(errp, QERR_IO_ERROR);
1318 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
1325 f = fopen(filename, "wb");
1327 error_setg_file_open(errp, errno, filename);
1335 cpu_physical_memory_rw(addr, buf, l, 0);
1336 if (fwrite(buf, 1, l, f) != l) {
1337 error_set(errp, QERR_IO_ERROR);
1348 void qmp_inject_nmi(Error **errp)
1350 #if defined(TARGET_I386)
1353 for (cs = first_cpu; cs != NULL; cs = cs->next_cpu) {
1354 X86CPU *cpu = X86_CPU(cs);
1355 CPUX86State *env = &cpu->env;
1357 if (!env->apic_state) {
1358 cpu_interrupt(cs, CPU_INTERRUPT_NMI);
1360 apic_deliver_nmi(env->apic_state);
1364 error_set(errp, QERR_UNSUPPORTED);