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 "qapi/qmp/qerror.h"
30 #include "sysemu/sysemu.h"
31 #include "exec/gdbstub.h"
32 #include "sysemu/dma.h"
33 #include "sysemu/kvm.h"
34 #include "qmp-commands.h"
36 #include "qemu/thread.h"
37 #include "sysemu/cpus.h"
38 #include "sysemu/qtest.h"
39 #include "qemu/main-loop.h"
40 #include "qemu/bitmap.h"
41 #include "qemu/seqlock.h"
42 #include "qapi-event.h"
46 #include "qemu/compatfd.h"
51 #include <sys/prctl.h>
54 #define PR_MCE_KILL 33
57 #ifndef PR_MCE_KILL_SET
58 #define PR_MCE_KILL_SET 1
61 #ifndef PR_MCE_KILL_EARLY
62 #define PR_MCE_KILL_EARLY 1
65 #endif /* CONFIG_LINUX */
67 static CPUState *next_cpu;
71 bool cpu_is_stopped(CPUState *cpu)
73 return cpu->stopped || !runstate_is_running();
76 static bool cpu_thread_is_idle(CPUState *cpu)
78 if (cpu->stop || cpu->queued_work_first) {
81 if (cpu_is_stopped(cpu)) {
84 if (!cpu->halted || cpu_has_work(cpu) ||
85 kvm_halt_in_kernel()) {
91 static bool all_cpu_threads_idle(void)
96 if (!cpu_thread_is_idle(cpu)) {
103 /***********************************************************/
104 /* guest cycle counter */
106 /* Protected by TimersState seqlock */
108 static int64_t vm_clock_warp_start = -1;
109 /* Conversion factor from emulated instructions to virtual clock ticks. */
110 static int icount_time_shift;
111 /* Arbitrarily pick 1MIPS as the minimum allowable speed. */
112 #define MAX_ICOUNT_SHIFT 10
114 static QEMUTimer *icount_rt_timer;
115 static QEMUTimer *icount_vm_timer;
116 static QEMUTimer *icount_warp_timer;
118 typedef struct TimersState {
119 /* Protected by BQL. */
120 int64_t cpu_ticks_prev;
121 int64_t cpu_ticks_offset;
123 /* cpu_clock_offset can be read out of BQL, so protect it with
126 QemuSeqLock vm_clock_seqlock;
127 int64_t cpu_clock_offset;
128 int32_t cpu_ticks_enabled;
131 /* Compensate for varying guest execution speed. */
132 int64_t qemu_icount_bias;
133 /* Only written by TCG thread */
137 static TimersState timers_state;
139 /* Return the virtual CPU time, based on the instruction counter. */
140 static int64_t cpu_get_icount_locked(void)
143 CPUState *cpu = current_cpu;
145 icount = timers_state.qemu_icount;
147 if (!cpu_can_do_io(cpu)) {
148 fprintf(stderr, "Bad clock read\n");
150 icount -= (cpu->icount_decr.u16.low + cpu->icount_extra);
152 return timers_state.qemu_icount_bias + cpu_icount_to_ns(icount);
155 int64_t cpu_get_icount(void)
161 start = seqlock_read_begin(&timers_state.vm_clock_seqlock);
162 icount = cpu_get_icount_locked();
163 } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start));
168 int64_t cpu_icount_to_ns(int64_t icount)
170 return icount << icount_time_shift;
173 /* return the host CPU cycle counter and handle stop/restart */
174 /* Caller must hold the BQL */
175 int64_t cpu_get_ticks(void)
180 return cpu_get_icount();
183 ticks = timers_state.cpu_ticks_offset;
184 if (timers_state.cpu_ticks_enabled) {
185 ticks += cpu_get_real_ticks();
188 if (timers_state.cpu_ticks_prev > ticks) {
189 /* Note: non increasing ticks may happen if the host uses
191 timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks;
192 ticks = timers_state.cpu_ticks_prev;
195 timers_state.cpu_ticks_prev = ticks;
199 static int64_t cpu_get_clock_locked(void)
203 ticks = timers_state.cpu_clock_offset;
204 if (timers_state.cpu_ticks_enabled) {
205 ticks += get_clock();
211 /* return the host CPU monotonic timer and handle stop/restart */
212 int64_t cpu_get_clock(void)
218 start = seqlock_read_begin(&timers_state.vm_clock_seqlock);
219 ti = cpu_get_clock_locked();
220 } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start));
225 /* return the offset between the host clock and virtual CPU clock */
226 int64_t cpu_get_clock_offset(void)
232 start = seqlock_read_begin(&timers_state.vm_clock_seqlock);
233 ti = timers_state.cpu_clock_offset;
234 if (!timers_state.cpu_ticks_enabled) {
237 } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start));
242 /* enable cpu_get_ticks()
243 * Caller must hold BQL which server as mutex for vm_clock_seqlock.
245 void cpu_enable_ticks(void)
247 /* Here, the really thing protected by seqlock is cpu_clock_offset. */
248 seqlock_write_lock(&timers_state.vm_clock_seqlock);
249 if (!timers_state.cpu_ticks_enabled) {
250 timers_state.cpu_ticks_offset -= cpu_get_real_ticks();
251 timers_state.cpu_clock_offset -= get_clock();
252 timers_state.cpu_ticks_enabled = 1;
254 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
257 /* disable cpu_get_ticks() : the clock is stopped. You must not call
258 * cpu_get_ticks() after that.
259 * Caller must hold BQL which server as mutex for vm_clock_seqlock.
261 void cpu_disable_ticks(void)
263 /* Here, the really thing protected by seqlock is cpu_clock_offset. */
264 seqlock_write_lock(&timers_state.vm_clock_seqlock);
265 if (timers_state.cpu_ticks_enabled) {
266 timers_state.cpu_ticks_offset += cpu_get_real_ticks();
267 timers_state.cpu_clock_offset = cpu_get_clock_locked();
268 timers_state.cpu_ticks_enabled = 0;
270 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
273 /* Correlation between real and virtual time is always going to be
274 fairly approximate, so ignore small variation.
275 When the guest is idle real and virtual time will be aligned in
277 #define ICOUNT_WOBBLE (get_ticks_per_sec() / 10)
279 static void icount_adjust(void)
285 /* Protected by TimersState mutex. */
286 static int64_t last_delta;
288 /* If the VM is not running, then do nothing. */
289 if (!runstate_is_running()) {
293 seqlock_write_lock(&timers_state.vm_clock_seqlock);
294 cur_time = cpu_get_clock_locked();
295 cur_icount = cpu_get_icount_locked();
297 delta = cur_icount - cur_time;
298 /* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */
300 && last_delta + ICOUNT_WOBBLE < delta * 2
301 && icount_time_shift > 0) {
302 /* The guest is getting too far ahead. Slow time down. */
306 && last_delta - ICOUNT_WOBBLE > delta * 2
307 && icount_time_shift < MAX_ICOUNT_SHIFT) {
308 /* The guest is getting too far behind. Speed time up. */
312 timers_state.qemu_icount_bias = cur_icount
313 - (timers_state.qemu_icount << icount_time_shift);
314 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
317 static void icount_adjust_rt(void *opaque)
319 timer_mod(icount_rt_timer,
320 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + 1000);
324 static void icount_adjust_vm(void *opaque)
326 timer_mod(icount_vm_timer,
327 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
328 get_ticks_per_sec() / 10);
332 static int64_t qemu_icount_round(int64_t count)
334 return (count + (1 << icount_time_shift) - 1) >> icount_time_shift;
337 static void icount_warp_rt(void *opaque)
339 /* The icount_warp_timer is rescheduled soon after vm_clock_warp_start
340 * changes from -1 to another value, so the race here is okay.
342 if (atomic_read(&vm_clock_warp_start) == -1) {
346 seqlock_write_lock(&timers_state.vm_clock_seqlock);
347 if (runstate_is_running()) {
348 int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
351 warp_delta = clock - vm_clock_warp_start;
352 if (use_icount == 2) {
354 * In adaptive mode, do not let QEMU_CLOCK_VIRTUAL run too
355 * far ahead of real time.
357 int64_t cur_time = cpu_get_clock_locked();
358 int64_t cur_icount = cpu_get_icount_locked();
359 int64_t delta = cur_time - cur_icount;
360 warp_delta = MIN(warp_delta, delta);
362 timers_state.qemu_icount_bias += warp_delta;
364 vm_clock_warp_start = -1;
365 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
367 if (qemu_clock_expired(QEMU_CLOCK_VIRTUAL)) {
368 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
372 void qtest_clock_warp(int64_t dest)
374 int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
375 assert(qtest_enabled());
376 while (clock < dest) {
377 int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
378 int64_t warp = qemu_soonest_timeout(dest - clock, deadline);
379 seqlock_write_lock(&timers_state.vm_clock_seqlock);
380 timers_state.qemu_icount_bias += warp;
381 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
383 qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL);
384 clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
386 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
389 void qemu_clock_warp(QEMUClockType type)
395 * There are too many global variables to make the "warp" behavior
396 * applicable to other clocks. But a clock argument removes the
397 * need for if statements all over the place.
399 if (type != QEMU_CLOCK_VIRTUAL || !use_icount) {
404 * If the CPUs have been sleeping, advance QEMU_CLOCK_VIRTUAL timer now.
405 * This ensures that the deadline for the timer is computed correctly below.
406 * This also makes sure that the insn counter is synchronized before the
407 * CPU starts running, in case the CPU is woken by an event other than
408 * the earliest QEMU_CLOCK_VIRTUAL timer.
410 icount_warp_rt(NULL);
411 timer_del(icount_warp_timer);
412 if (!all_cpu_threads_idle()) {
416 if (qtest_enabled()) {
417 /* When testing, qtest commands advance icount. */
421 /* We want to use the earliest deadline from ALL vm_clocks */
422 clock = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
423 deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
430 * Ensure QEMU_CLOCK_VIRTUAL proceeds even when the virtual CPU goes to
431 * sleep. Otherwise, the CPU might be waiting for a future timer
432 * interrupt to wake it up, but the interrupt never comes because
433 * the vCPU isn't running any insns and thus doesn't advance the
434 * QEMU_CLOCK_VIRTUAL.
436 * An extreme solution for this problem would be to never let VCPUs
437 * sleep in icount mode if there is a pending QEMU_CLOCK_VIRTUAL
438 * timer; rather time could just advance to the next QEMU_CLOCK_VIRTUAL
439 * event. Instead, we do stop VCPUs and only advance QEMU_CLOCK_VIRTUAL
440 * after some e"real" time, (related to the time left until the next
441 * event) has passed. The QEMU_CLOCK_REALTIME timer will do this.
442 * This avoids that the warps are visible externally; for example,
443 * you will not be sending network packets continuously instead of
446 seqlock_write_lock(&timers_state.vm_clock_seqlock);
447 if (vm_clock_warp_start == -1 || vm_clock_warp_start > clock) {
448 vm_clock_warp_start = clock;
450 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
451 timer_mod_anticipate(icount_warp_timer, clock + deadline);
452 } else if (deadline == 0) {
453 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
457 static bool icount_state_needed(void *opaque)
463 * This is a subsection for icount migration.
465 static const VMStateDescription icount_vmstate_timers = {
466 .name = "timer/icount",
468 .minimum_version_id = 1,
469 .fields = (VMStateField[]) {
470 VMSTATE_INT64(qemu_icount_bias, TimersState),
471 VMSTATE_INT64(qemu_icount, TimersState),
472 VMSTATE_END_OF_LIST()
476 static const VMStateDescription vmstate_timers = {
479 .minimum_version_id = 1,
480 .fields = (VMStateField[]) {
481 VMSTATE_INT64(cpu_ticks_offset, TimersState),
482 VMSTATE_INT64(dummy, TimersState),
483 VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
484 VMSTATE_END_OF_LIST()
486 .subsections = (VMStateSubsection[]) {
488 .vmsd = &icount_vmstate_timers,
489 .needed = icount_state_needed,
496 void cpu_ticks_init(void)
498 seqlock_init(&timers_state.vm_clock_seqlock, NULL);
499 vmstate_register(NULL, 0, &vmstate_timers, &timers_state);
502 void configure_icount(QemuOpts *opts, Error **errp)
505 char *rem_str = NULL;
507 option = qemu_opt_get(opts, "shift");
509 if (qemu_opt_get(opts, "align") != NULL) {
510 error_setg(errp, "Please specify shift option when using align");
514 icount_align_option = qemu_opt_get_bool(opts, "align", false);
515 icount_warp_timer = timer_new_ns(QEMU_CLOCK_REALTIME,
516 icount_warp_rt, NULL);
517 if (strcmp(option, "auto") != 0) {
519 icount_time_shift = strtol(option, &rem_str, 0);
520 if (errno != 0 || *rem_str != '\0' || !strlen(option)) {
521 error_setg(errp, "icount: Invalid shift value");
525 } else if (icount_align_option) {
526 error_setg(errp, "shift=auto and align=on are incompatible");
531 /* 125MIPS seems a reasonable initial guess at the guest speed.
532 It will be corrected fairly quickly anyway. */
533 icount_time_shift = 3;
535 /* Have both realtime and virtual time triggers for speed adjustment.
536 The realtime trigger catches emulated time passing too slowly,
537 the virtual time trigger catches emulated time passing too fast.
538 Realtime triggers occur even when idle, so use them less frequently
540 icount_rt_timer = timer_new_ms(QEMU_CLOCK_REALTIME,
541 icount_adjust_rt, NULL);
542 timer_mod(icount_rt_timer,
543 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + 1000);
544 icount_vm_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
545 icount_adjust_vm, NULL);
546 timer_mod(icount_vm_timer,
547 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
548 get_ticks_per_sec() / 10);
551 /***********************************************************/
552 void hw_error(const char *fmt, ...)
558 fprintf(stderr, "qemu: hardware error: ");
559 vfprintf(stderr, fmt, ap);
560 fprintf(stderr, "\n");
562 fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
563 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU);
569 void cpu_synchronize_all_states(void)
574 cpu_synchronize_state(cpu);
578 void cpu_synchronize_all_post_reset(void)
583 cpu_synchronize_post_reset(cpu);
587 void cpu_synchronize_all_post_init(void)
592 cpu_synchronize_post_init(cpu);
596 static int do_vm_stop(RunState state)
600 if (runstate_is_running()) {
604 vm_state_notify(0, state);
605 qapi_event_send_stop(&error_abort);
609 ret = bdrv_flush_all();
614 static bool cpu_can_run(CPUState *cpu)
619 if (cpu_is_stopped(cpu)) {
625 static void cpu_handle_guest_debug(CPUState *cpu)
627 gdb_set_stop_cpu(cpu);
628 qemu_system_debug_request();
632 static void cpu_signal(int sig)
635 cpu_exit(current_cpu);
641 static void sigbus_reraise(void)
644 struct sigaction action;
646 memset(&action, 0, sizeof(action));
647 action.sa_handler = SIG_DFL;
648 if (!sigaction(SIGBUS, &action, NULL)) {
651 sigaddset(&set, SIGBUS);
652 sigprocmask(SIG_UNBLOCK, &set, NULL);
654 perror("Failed to re-raise SIGBUS!\n");
658 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
661 if (kvm_on_sigbus(siginfo->ssi_code,
662 (void *)(intptr_t)siginfo->ssi_addr)) {
667 static void qemu_init_sigbus(void)
669 struct sigaction action;
671 memset(&action, 0, sizeof(action));
672 action.sa_flags = SA_SIGINFO;
673 action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
674 sigaction(SIGBUS, &action, NULL);
676 prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
679 static void qemu_kvm_eat_signals(CPUState *cpu)
681 struct timespec ts = { 0, 0 };
687 sigemptyset(&waitset);
688 sigaddset(&waitset, SIG_IPI);
689 sigaddset(&waitset, SIGBUS);
692 r = sigtimedwait(&waitset, &siginfo, &ts);
693 if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
694 perror("sigtimedwait");
700 if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) {
708 r = sigpending(&chkset);
710 perror("sigpending");
713 } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
716 #else /* !CONFIG_LINUX */
718 static void qemu_init_sigbus(void)
722 static void qemu_kvm_eat_signals(CPUState *cpu)
725 #endif /* !CONFIG_LINUX */
728 static void dummy_signal(int sig)
732 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
736 struct sigaction sigact;
738 memset(&sigact, 0, sizeof(sigact));
739 sigact.sa_handler = dummy_signal;
740 sigaction(SIG_IPI, &sigact, NULL);
742 pthread_sigmask(SIG_BLOCK, NULL, &set);
743 sigdelset(&set, SIG_IPI);
744 sigdelset(&set, SIGBUS);
745 r = kvm_set_signal_mask(cpu, &set);
747 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
752 static void qemu_tcg_init_cpu_signals(void)
755 struct sigaction sigact;
757 memset(&sigact, 0, sizeof(sigact));
758 sigact.sa_handler = cpu_signal;
759 sigaction(SIG_IPI, &sigact, NULL);
762 sigaddset(&set, SIG_IPI);
763 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
767 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
772 static void qemu_tcg_init_cpu_signals(void)
777 static QemuMutex qemu_global_mutex;
778 static QemuCond qemu_io_proceeded_cond;
779 static bool iothread_requesting_mutex;
781 static QemuThread io_thread;
783 static QemuThread *tcg_cpu_thread;
784 static QemuCond *tcg_halt_cond;
787 static QemuCond qemu_cpu_cond;
789 static QemuCond qemu_pause_cond;
790 static QemuCond qemu_work_cond;
792 void qemu_init_cpu_loop(void)
795 qemu_cond_init(&qemu_cpu_cond);
796 qemu_cond_init(&qemu_pause_cond);
797 qemu_cond_init(&qemu_work_cond);
798 qemu_cond_init(&qemu_io_proceeded_cond);
799 qemu_mutex_init(&qemu_global_mutex);
801 qemu_thread_get_self(&io_thread);
804 void run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
806 struct qemu_work_item wi;
808 if (qemu_cpu_is_self(cpu)) {
816 if (cpu->queued_work_first == NULL) {
817 cpu->queued_work_first = &wi;
819 cpu->queued_work_last->next = &wi;
821 cpu->queued_work_last = &wi;
827 CPUState *self_cpu = current_cpu;
829 qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
830 current_cpu = self_cpu;
834 void async_run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
836 struct qemu_work_item *wi;
838 if (qemu_cpu_is_self(cpu)) {
843 wi = g_malloc0(sizeof(struct qemu_work_item));
847 if (cpu->queued_work_first == NULL) {
848 cpu->queued_work_first = wi;
850 cpu->queued_work_last->next = wi;
852 cpu->queued_work_last = wi;
859 static void flush_queued_work(CPUState *cpu)
861 struct qemu_work_item *wi;
863 if (cpu->queued_work_first == NULL) {
867 while ((wi = cpu->queued_work_first)) {
868 cpu->queued_work_first = wi->next;
875 cpu->queued_work_last = NULL;
876 qemu_cond_broadcast(&qemu_work_cond);
879 static void qemu_wait_io_event_common(CPUState *cpu)
884 qemu_cond_signal(&qemu_pause_cond);
886 flush_queued_work(cpu);
887 cpu->thread_kicked = false;
890 static void qemu_tcg_wait_io_event(void)
894 while (all_cpu_threads_idle()) {
895 /* Start accounting real time to the virtual clock if the CPUs
897 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
898 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
901 while (iothread_requesting_mutex) {
902 qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex);
906 qemu_wait_io_event_common(cpu);
910 static void qemu_kvm_wait_io_event(CPUState *cpu)
912 while (cpu_thread_is_idle(cpu)) {
913 qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
916 qemu_kvm_eat_signals(cpu);
917 qemu_wait_io_event_common(cpu);
920 static void *qemu_kvm_cpu_thread_fn(void *arg)
925 qemu_mutex_lock(&qemu_global_mutex);
926 qemu_thread_get_self(cpu->thread);
927 cpu->thread_id = qemu_get_thread_id();
930 r = kvm_init_vcpu(cpu);
932 fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
936 qemu_kvm_init_cpu_signals(cpu);
938 /* signal CPU creation */
940 qemu_cond_signal(&qemu_cpu_cond);
943 if (cpu_can_run(cpu)) {
944 r = kvm_cpu_exec(cpu);
945 if (r == EXCP_DEBUG) {
946 cpu_handle_guest_debug(cpu);
949 qemu_kvm_wait_io_event(cpu);
955 static void *qemu_dummy_cpu_thread_fn(void *arg)
958 fprintf(stderr, "qtest is not supported under Windows\n");
965 qemu_mutex_lock_iothread();
966 qemu_thread_get_self(cpu->thread);
967 cpu->thread_id = qemu_get_thread_id();
969 sigemptyset(&waitset);
970 sigaddset(&waitset, SIG_IPI);
972 /* signal CPU creation */
974 qemu_cond_signal(&qemu_cpu_cond);
979 qemu_mutex_unlock_iothread();
982 r = sigwait(&waitset, &sig);
983 } while (r == -1 && (errno == EAGAIN || errno == EINTR));
988 qemu_mutex_lock_iothread();
990 qemu_wait_io_event_common(cpu);
997 static void tcg_exec_all(void);
999 static void *qemu_tcg_cpu_thread_fn(void *arg)
1001 CPUState *cpu = arg;
1003 qemu_tcg_init_cpu_signals();
1004 qemu_thread_get_self(cpu->thread);
1006 qemu_mutex_lock(&qemu_global_mutex);
1008 cpu->thread_id = qemu_get_thread_id();
1009 cpu->created = true;
1011 qemu_cond_signal(&qemu_cpu_cond);
1013 /* wait for initial kick-off after machine start */
1014 while (QTAILQ_FIRST(&cpus)->stopped) {
1015 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
1017 /* process any pending work */
1019 qemu_wait_io_event_common(cpu);
1027 int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
1029 if (deadline == 0) {
1030 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
1033 qemu_tcg_wait_io_event();
1039 static void qemu_cpu_kick_thread(CPUState *cpu)
1044 err = pthread_kill(cpu->thread->thread, SIG_IPI);
1046 fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
1050 if (!qemu_cpu_is_self(cpu)) {
1053 if (SuspendThread(cpu->hThread) == (DWORD)-1) {
1054 fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
1059 /* On multi-core systems, we are not sure that the thread is actually
1060 * suspended until we can get the context.
1062 tcgContext.ContextFlags = CONTEXT_CONTROL;
1063 while (GetThreadContext(cpu->hThread, &tcgContext) != 0) {
1069 if (ResumeThread(cpu->hThread) == (DWORD)-1) {
1070 fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
1078 void qemu_cpu_kick(CPUState *cpu)
1080 qemu_cond_broadcast(cpu->halt_cond);
1081 if (!tcg_enabled() && !cpu->thread_kicked) {
1082 qemu_cpu_kick_thread(cpu);
1083 cpu->thread_kicked = true;
1087 void qemu_cpu_kick_self(void)
1090 assert(current_cpu);
1092 if (!current_cpu->thread_kicked) {
1093 qemu_cpu_kick_thread(current_cpu);
1094 current_cpu->thread_kicked = true;
1101 bool qemu_cpu_is_self(CPUState *cpu)
1103 return qemu_thread_is_self(cpu->thread);
1106 static bool qemu_in_vcpu_thread(void)
1108 return current_cpu && qemu_cpu_is_self(current_cpu);
1111 void qemu_mutex_lock_iothread(void)
1113 if (!tcg_enabled()) {
1114 qemu_mutex_lock(&qemu_global_mutex);
1116 iothread_requesting_mutex = true;
1117 if (qemu_mutex_trylock(&qemu_global_mutex)) {
1118 qemu_cpu_kick_thread(first_cpu);
1119 qemu_mutex_lock(&qemu_global_mutex);
1121 iothread_requesting_mutex = false;
1122 qemu_cond_broadcast(&qemu_io_proceeded_cond);
1126 void qemu_mutex_unlock_iothread(void)
1128 qemu_mutex_unlock(&qemu_global_mutex);
1131 static int all_vcpus_paused(void)
1136 if (!cpu->stopped) {
1144 void pause_all_vcpus(void)
1148 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false);
1154 if (qemu_in_vcpu_thread()) {
1156 if (!kvm_enabled()) {
1159 cpu->stopped = true;
1165 while (!all_vcpus_paused()) {
1166 qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
1173 void cpu_resume(CPUState *cpu)
1176 cpu->stopped = false;
1180 void resume_all_vcpus(void)
1184 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true);
1190 /* For temporary buffers for forming a name */
1191 #define VCPU_THREAD_NAME_SIZE 16
1193 static void qemu_tcg_init_vcpu(CPUState *cpu)
1195 char thread_name[VCPU_THREAD_NAME_SIZE];
1197 tcg_cpu_address_space_init(cpu, cpu->as);
1199 /* share a single thread for all cpus with TCG */
1200 if (!tcg_cpu_thread) {
1201 cpu->thread = g_malloc0(sizeof(QemuThread));
1202 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1203 qemu_cond_init(cpu->halt_cond);
1204 tcg_halt_cond = cpu->halt_cond;
1205 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG",
1207 qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn,
1208 cpu, QEMU_THREAD_JOINABLE);
1210 cpu->hThread = qemu_thread_get_handle(cpu->thread);
1212 while (!cpu->created) {
1213 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1215 tcg_cpu_thread = cpu->thread;
1217 cpu->thread = tcg_cpu_thread;
1218 cpu->halt_cond = tcg_halt_cond;
1222 static void qemu_kvm_start_vcpu(CPUState *cpu)
1224 char thread_name[VCPU_THREAD_NAME_SIZE];
1226 cpu->thread = g_malloc0(sizeof(QemuThread));
1227 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1228 qemu_cond_init(cpu->halt_cond);
1229 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/KVM",
1231 qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn,
1232 cpu, QEMU_THREAD_JOINABLE);
1233 while (!cpu->created) {
1234 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1238 static void qemu_dummy_start_vcpu(CPUState *cpu)
1240 char thread_name[VCPU_THREAD_NAME_SIZE];
1242 cpu->thread = g_malloc0(sizeof(QemuThread));
1243 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1244 qemu_cond_init(cpu->halt_cond);
1245 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/DUMMY",
1247 qemu_thread_create(cpu->thread, thread_name, qemu_dummy_cpu_thread_fn, cpu,
1248 QEMU_THREAD_JOINABLE);
1249 while (!cpu->created) {
1250 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1254 void qemu_init_vcpu(CPUState *cpu)
1256 cpu->nr_cores = smp_cores;
1257 cpu->nr_threads = smp_threads;
1258 cpu->stopped = true;
1259 if (kvm_enabled()) {
1260 qemu_kvm_start_vcpu(cpu);
1261 } else if (tcg_enabled()) {
1262 qemu_tcg_init_vcpu(cpu);
1264 qemu_dummy_start_vcpu(cpu);
1268 void cpu_stop_current(void)
1271 current_cpu->stop = false;
1272 current_cpu->stopped = true;
1273 cpu_exit(current_cpu);
1274 qemu_cond_signal(&qemu_pause_cond);
1278 int vm_stop(RunState state)
1280 if (qemu_in_vcpu_thread()) {
1281 qemu_system_vmstop_request_prepare();
1282 qemu_system_vmstop_request(state);
1284 * FIXME: should not return to device code in case
1285 * vm_stop() has been requested.
1291 return do_vm_stop(state);
1294 /* does a state transition even if the VM is already stopped,
1295 current state is forgotten forever */
1296 int vm_stop_force_state(RunState state)
1298 if (runstate_is_running()) {
1299 return vm_stop(state);
1301 runstate_set(state);
1302 /* Make sure to return an error if the flush in a previous vm_stop()
1304 return bdrv_flush_all();
1308 static int tcg_cpu_exec(CPUArchState *env)
1310 CPUState *cpu = ENV_GET_CPU(env);
1312 #ifdef CONFIG_PROFILER
1316 #ifdef CONFIG_PROFILER
1317 ti = profile_getclock();
1323 timers_state.qemu_icount -= (cpu->icount_decr.u16.low
1324 + cpu->icount_extra);
1325 cpu->icount_decr.u16.low = 0;
1326 cpu->icount_extra = 0;
1327 deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
1329 /* Maintain prior (possibly buggy) behaviour where if no deadline
1330 * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than
1331 * INT32_MAX nanoseconds ahead, we still use INT32_MAX
1334 if ((deadline < 0) || (deadline > INT32_MAX)) {
1335 deadline = INT32_MAX;
1338 count = qemu_icount_round(deadline);
1339 timers_state.qemu_icount += count;
1340 decr = (count > 0xffff) ? 0xffff : count;
1342 cpu->icount_decr.u16.low = decr;
1343 cpu->icount_extra = count;
1345 ret = cpu_exec(env);
1346 #ifdef CONFIG_PROFILER
1347 qemu_time += profile_getclock() - ti;
1350 /* Fold pending instructions back into the
1351 instruction counter, and clear the interrupt flag. */
1352 timers_state.qemu_icount -= (cpu->icount_decr.u16.low
1353 + cpu->icount_extra);
1354 cpu->icount_decr.u32 = 0;
1355 cpu->icount_extra = 0;
1360 static void tcg_exec_all(void)
1364 /* Account partial waits to QEMU_CLOCK_VIRTUAL. */
1365 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
1367 if (next_cpu == NULL) {
1368 next_cpu = first_cpu;
1370 for (; next_cpu != NULL && !exit_request; next_cpu = CPU_NEXT(next_cpu)) {
1371 CPUState *cpu = next_cpu;
1372 CPUArchState *env = cpu->env_ptr;
1374 qemu_clock_enable(QEMU_CLOCK_VIRTUAL,
1375 (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0);
1377 if (cpu_can_run(cpu)) {
1378 r = tcg_cpu_exec(env);
1379 if (r == EXCP_DEBUG) {
1380 cpu_handle_guest_debug(cpu);
1383 } else if (cpu->stop || cpu->stopped) {
1390 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1392 /* XXX: implement xxx_cpu_list for targets that still miss it */
1393 #if defined(cpu_list)
1394 cpu_list(f, cpu_fprintf);
1398 CpuInfoList *qmp_query_cpus(Error **errp)
1400 CpuInfoList *head = NULL, *cur_item = NULL;
1405 #if defined(TARGET_I386)
1406 X86CPU *x86_cpu = X86_CPU(cpu);
1407 CPUX86State *env = &x86_cpu->env;
1408 #elif defined(TARGET_PPC)
1409 PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu);
1410 CPUPPCState *env = &ppc_cpu->env;
1411 #elif defined(TARGET_SPARC)
1412 SPARCCPU *sparc_cpu = SPARC_CPU(cpu);
1413 CPUSPARCState *env = &sparc_cpu->env;
1414 #elif defined(TARGET_MIPS)
1415 MIPSCPU *mips_cpu = MIPS_CPU(cpu);
1416 CPUMIPSState *env = &mips_cpu->env;
1417 #elif defined(TARGET_TRICORE)
1418 TriCoreCPU *tricore_cpu = TRICORE_CPU(cpu);
1419 CPUTriCoreState *env = &tricore_cpu->env;
1422 cpu_synchronize_state(cpu);
1424 info = g_malloc0(sizeof(*info));
1425 info->value = g_malloc0(sizeof(*info->value));
1426 info->value->CPU = cpu->cpu_index;
1427 info->value->current = (cpu == first_cpu);
1428 info->value->halted = cpu->halted;
1429 info->value->thread_id = cpu->thread_id;
1430 #if defined(TARGET_I386)
1431 info->value->has_pc = true;
1432 info->value->pc = env->eip + env->segs[R_CS].base;
1433 #elif defined(TARGET_PPC)
1434 info->value->has_nip = true;
1435 info->value->nip = env->nip;
1436 #elif defined(TARGET_SPARC)
1437 info->value->has_pc = true;
1438 info->value->pc = env->pc;
1439 info->value->has_npc = true;
1440 info->value->npc = env->npc;
1441 #elif defined(TARGET_MIPS)
1442 info->value->has_PC = true;
1443 info->value->PC = env->active_tc.PC;
1444 #elif defined(TARGET_TRICORE)
1445 info->value->has_PC = true;
1446 info->value->PC = env->PC;
1449 /* XXX: waiting for the qapi to support GSList */
1451 head = cur_item = info;
1453 cur_item->next = info;
1461 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
1462 bool has_cpu, int64_t cpu_index, Error **errp)
1473 cpu = qemu_get_cpu(cpu_index);
1475 error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
1480 f = fopen(filename, "wb");
1482 error_setg_file_open(errp, errno, filename);
1490 if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) {
1491 error_setg(errp, "Invalid addr 0x%016" PRIx64 "specified", addr);
1494 if (fwrite(buf, 1, l, f) != l) {
1495 error_set(errp, QERR_IO_ERROR);
1506 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
1513 f = fopen(filename, "wb");
1515 error_setg_file_open(errp, errno, filename);
1523 cpu_physical_memory_read(addr, buf, l);
1524 if (fwrite(buf, 1, l, f) != l) {
1525 error_set(errp, QERR_IO_ERROR);
1536 void qmp_inject_nmi(Error **errp)
1538 #if defined(TARGET_I386)
1542 X86CPU *cpu = X86_CPU(cs);
1544 if (!cpu->apic_state) {
1545 cpu_interrupt(cs, CPU_INTERRUPT_NMI);
1547 apic_deliver_nmi(cpu->apic_state);
1551 nmi_monitor_handle(monitor_get_cpu_index(), errp);
1555 void dump_drift_info(FILE *f, fprintf_function cpu_fprintf)
1561 cpu_fprintf(f, "Host - Guest clock %"PRIi64" ms\n",
1562 (cpu_get_clock() - cpu_get_icount())/SCALE_MS);
1563 if (icount_align_option) {
1564 cpu_fprintf(f, "Max guest delay %"PRIi64" ms\n", -max_delay/SCALE_MS);
1565 cpu_fprintf(f, "Max guest advance %"PRIi64" ms\n", max_advance/SCALE_MS);
1567 cpu_fprintf(f, "Max guest delay NA\n");
1568 cpu_fprintf(f, "Max guest advance NA\n");