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 "qemu/error-report.h"
31 #include "sysemu/sysemu.h"
32 #include "exec/gdbstub.h"
33 #include "sysemu/dma.h"
34 #include "sysemu/kvm.h"
35 #include "qmp-commands.h"
37 #include "qemu/thread.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/qtest.h"
40 #include "qemu/main-loop.h"
41 #include "qemu/bitmap.h"
42 #include "qemu/seqlock.h"
43 #include "qapi-event.h"
47 #include "qemu/compatfd.h"
52 #include <sys/prctl.h>
55 #define PR_MCE_KILL 33
58 #ifndef PR_MCE_KILL_SET
59 #define PR_MCE_KILL_SET 1
62 #ifndef PR_MCE_KILL_EARLY
63 #define PR_MCE_KILL_EARLY 1
66 #endif /* CONFIG_LINUX */
68 static CPUState *next_cpu;
72 bool cpu_is_stopped(CPUState *cpu)
74 return cpu->stopped || !runstate_is_running();
77 static bool cpu_thread_is_idle(CPUState *cpu)
79 if (cpu->stop || cpu->queued_work_first) {
82 if (cpu_is_stopped(cpu)) {
85 if (!cpu->halted || cpu_has_work(cpu) ||
86 kvm_halt_in_kernel()) {
92 static bool all_cpu_threads_idle(void)
97 if (!cpu_thread_is_idle(cpu)) {
104 /***********************************************************/
105 /* guest cycle counter */
107 /* Protected by TimersState seqlock */
109 static bool icount_sleep = true;
110 static int64_t vm_clock_warp_start = -1;
111 /* Conversion factor from emulated instructions to virtual clock ticks. */
112 static int icount_time_shift;
113 /* Arbitrarily pick 1MIPS as the minimum allowable speed. */
114 #define MAX_ICOUNT_SHIFT 10
116 static QEMUTimer *icount_rt_timer;
117 static QEMUTimer *icount_vm_timer;
118 static QEMUTimer *icount_warp_timer;
120 typedef struct TimersState {
121 /* Protected by BQL. */
122 int64_t cpu_ticks_prev;
123 int64_t cpu_ticks_offset;
125 /* cpu_clock_offset can be read out of BQL, so protect it with
128 QemuSeqLock vm_clock_seqlock;
129 int64_t cpu_clock_offset;
130 int32_t cpu_ticks_enabled;
133 /* Compensate for varying guest execution speed. */
134 int64_t qemu_icount_bias;
135 /* Only written by TCG thread */
139 static TimersState timers_state;
141 int64_t cpu_get_icount_raw(void)
144 CPUState *cpu = current_cpu;
146 icount = timers_state.qemu_icount;
148 if (!cpu->can_do_io) {
149 fprintf(stderr, "Bad icount read\n");
152 icount -= (cpu->icount_decr.u16.low + cpu->icount_extra);
157 /* Return the virtual CPU time, based on the instruction counter. */
158 static int64_t cpu_get_icount_locked(void)
160 int64_t icount = cpu_get_icount_raw();
161 return timers_state.qemu_icount_bias + cpu_icount_to_ns(icount);
164 int64_t cpu_get_icount(void)
170 start = seqlock_read_begin(&timers_state.vm_clock_seqlock);
171 icount = cpu_get_icount_locked();
172 } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start));
177 int64_t cpu_icount_to_ns(int64_t icount)
179 return icount << icount_time_shift;
182 /* return the host CPU cycle counter and handle stop/restart */
183 /* Caller must hold the BQL */
184 int64_t cpu_get_ticks(void)
189 return cpu_get_icount();
192 ticks = timers_state.cpu_ticks_offset;
193 if (timers_state.cpu_ticks_enabled) {
194 ticks += cpu_get_real_ticks();
197 if (timers_state.cpu_ticks_prev > ticks) {
198 /* Note: non increasing ticks may happen if the host uses
200 timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks;
201 ticks = timers_state.cpu_ticks_prev;
204 timers_state.cpu_ticks_prev = ticks;
208 static int64_t cpu_get_clock_locked(void)
212 ticks = timers_state.cpu_clock_offset;
213 if (timers_state.cpu_ticks_enabled) {
214 ticks += get_clock();
220 /* return the host CPU monotonic timer and handle stop/restart */
221 int64_t cpu_get_clock(void)
227 start = seqlock_read_begin(&timers_state.vm_clock_seqlock);
228 ti = cpu_get_clock_locked();
229 } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start));
234 /* enable cpu_get_ticks()
235 * Caller must hold BQL which server as mutex for vm_clock_seqlock.
237 void cpu_enable_ticks(void)
239 /* Here, the really thing protected by seqlock is cpu_clock_offset. */
240 seqlock_write_lock(&timers_state.vm_clock_seqlock);
241 if (!timers_state.cpu_ticks_enabled) {
242 timers_state.cpu_ticks_offset -= cpu_get_real_ticks();
243 timers_state.cpu_clock_offset -= get_clock();
244 timers_state.cpu_ticks_enabled = 1;
246 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
249 /* disable cpu_get_ticks() : the clock is stopped. You must not call
250 * cpu_get_ticks() after that.
251 * Caller must hold BQL which server as mutex for vm_clock_seqlock.
253 void cpu_disable_ticks(void)
255 /* Here, the really thing protected by seqlock is cpu_clock_offset. */
256 seqlock_write_lock(&timers_state.vm_clock_seqlock);
257 if (timers_state.cpu_ticks_enabled) {
258 timers_state.cpu_ticks_offset += cpu_get_real_ticks();
259 timers_state.cpu_clock_offset = cpu_get_clock_locked();
260 timers_state.cpu_ticks_enabled = 0;
262 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
265 /* Correlation between real and virtual time is always going to be
266 fairly approximate, so ignore small variation.
267 When the guest is idle real and virtual time will be aligned in
269 #define ICOUNT_WOBBLE (get_ticks_per_sec() / 10)
271 static void icount_adjust(void)
277 /* Protected by TimersState mutex. */
278 static int64_t last_delta;
280 /* If the VM is not running, then do nothing. */
281 if (!runstate_is_running()) {
285 seqlock_write_lock(&timers_state.vm_clock_seqlock);
286 cur_time = cpu_get_clock_locked();
287 cur_icount = cpu_get_icount_locked();
289 delta = cur_icount - cur_time;
290 /* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */
292 && last_delta + ICOUNT_WOBBLE < delta * 2
293 && icount_time_shift > 0) {
294 /* The guest is getting too far ahead. Slow time down. */
298 && last_delta - ICOUNT_WOBBLE > delta * 2
299 && icount_time_shift < MAX_ICOUNT_SHIFT) {
300 /* The guest is getting too far behind. Speed time up. */
304 timers_state.qemu_icount_bias = cur_icount
305 - (timers_state.qemu_icount << icount_time_shift);
306 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
309 static void icount_adjust_rt(void *opaque)
311 timer_mod(icount_rt_timer,
312 qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000);
316 static void icount_adjust_vm(void *opaque)
318 timer_mod(icount_vm_timer,
319 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
320 get_ticks_per_sec() / 10);
324 static int64_t qemu_icount_round(int64_t count)
326 return (count + (1 << icount_time_shift) - 1) >> icount_time_shift;
329 static void icount_warp_rt(void *opaque)
331 /* The icount_warp_timer is rescheduled soon after vm_clock_warp_start
332 * changes from -1 to another value, so the race here is okay.
334 if (atomic_read(&vm_clock_warp_start) == -1) {
338 seqlock_write_lock(&timers_state.vm_clock_seqlock);
339 if (runstate_is_running()) {
340 int64_t clock = cpu_get_clock_locked();
343 warp_delta = clock - vm_clock_warp_start;
344 if (use_icount == 2) {
346 * In adaptive mode, do not let QEMU_CLOCK_VIRTUAL run too
347 * far ahead of real time.
349 int64_t cur_icount = cpu_get_icount_locked();
350 int64_t delta = clock - cur_icount;
351 warp_delta = MIN(warp_delta, delta);
353 timers_state.qemu_icount_bias += warp_delta;
355 vm_clock_warp_start = -1;
356 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
358 if (qemu_clock_expired(QEMU_CLOCK_VIRTUAL)) {
359 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
363 void qtest_clock_warp(int64_t dest)
365 int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
366 AioContext *aio_context;
367 assert(qtest_enabled());
368 aio_context = qemu_get_aio_context();
369 while (clock < dest) {
370 int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
371 int64_t warp = qemu_soonest_timeout(dest - clock, deadline);
373 seqlock_write_lock(&timers_state.vm_clock_seqlock);
374 timers_state.qemu_icount_bias += warp;
375 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
377 qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL);
378 timerlist_run_timers(aio_context->tlg.tl[QEMU_CLOCK_VIRTUAL]);
379 clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
381 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
384 void qemu_clock_warp(QEMUClockType type)
390 * There are too many global variables to make the "warp" behavior
391 * applicable to other clocks. But a clock argument removes the
392 * need for if statements all over the place.
394 if (type != QEMU_CLOCK_VIRTUAL || !use_icount) {
400 * If the CPUs have been sleeping, advance QEMU_CLOCK_VIRTUAL timer now.
401 * This ensures that the deadline for the timer is computed correctly
403 * This also makes sure that the insn counter is synchronized before
404 * the CPU starts running, in case the CPU is woken by an event other
405 * than the earliest QEMU_CLOCK_VIRTUAL timer.
407 icount_warp_rt(NULL);
408 timer_del(icount_warp_timer);
410 if (!all_cpu_threads_idle()) {
414 if (qtest_enabled()) {
415 /* When testing, qtest commands advance icount. */
419 /* We want to use the earliest deadline from ALL vm_clocks */
420 clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
421 deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
423 static bool notified;
424 if (!icount_sleep && !notified) {
425 error_report("WARNING: icount sleep disabled and no active timers");
433 * Ensure QEMU_CLOCK_VIRTUAL proceeds even when the virtual CPU goes to
434 * sleep. Otherwise, the CPU might be waiting for a future timer
435 * interrupt to wake it up, but the interrupt never comes because
436 * the vCPU isn't running any insns and thus doesn't advance the
437 * QEMU_CLOCK_VIRTUAL.
441 * We never let VCPUs sleep in no sleep icount mode.
442 * If there is a pending QEMU_CLOCK_VIRTUAL timer we just advance
443 * to the next QEMU_CLOCK_VIRTUAL event and notify it.
444 * It is useful when we want a deterministic execution time,
445 * isolated from host latencies.
447 seqlock_write_lock(&timers_state.vm_clock_seqlock);
448 timers_state.qemu_icount_bias += deadline;
449 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
450 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
453 * We do stop VCPUs and only advance QEMU_CLOCK_VIRTUAL after some
454 * "real" time, (related to the time left until the next event) has
455 * passed. The QEMU_CLOCK_VIRTUAL_RT clock will do this.
456 * This avoids that the warps are visible externally; for example,
457 * you will not be sending network packets continuously instead of
460 seqlock_write_lock(&timers_state.vm_clock_seqlock);
461 if (vm_clock_warp_start == -1 || vm_clock_warp_start > clock) {
462 vm_clock_warp_start = clock;
464 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
465 timer_mod_anticipate(icount_warp_timer, clock + deadline);
467 } else if (deadline == 0) {
468 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
472 static bool icount_state_needed(void *opaque)
478 * This is a subsection for icount migration.
480 static const VMStateDescription icount_vmstate_timers = {
481 .name = "timer/icount",
483 .minimum_version_id = 1,
484 .needed = icount_state_needed,
485 .fields = (VMStateField[]) {
486 VMSTATE_INT64(qemu_icount_bias, TimersState),
487 VMSTATE_INT64(qemu_icount, TimersState),
488 VMSTATE_END_OF_LIST()
492 static const VMStateDescription vmstate_timers = {
495 .minimum_version_id = 1,
496 .fields = (VMStateField[]) {
497 VMSTATE_INT64(cpu_ticks_offset, TimersState),
498 VMSTATE_INT64(dummy, TimersState),
499 VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
500 VMSTATE_END_OF_LIST()
502 .subsections = (const VMStateDescription*[]) {
503 &icount_vmstate_timers,
508 void cpu_ticks_init(void)
510 seqlock_init(&timers_state.vm_clock_seqlock, NULL);
511 vmstate_register(NULL, 0, &vmstate_timers, &timers_state);
514 void configure_icount(QemuOpts *opts, Error **errp)
517 char *rem_str = NULL;
519 option = qemu_opt_get(opts, "shift");
521 if (qemu_opt_get(opts, "align") != NULL) {
522 error_setg(errp, "Please specify shift option when using align");
527 icount_sleep = qemu_opt_get_bool(opts, "sleep", true);
529 icount_warp_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT,
530 icount_warp_rt, NULL);
533 icount_align_option = qemu_opt_get_bool(opts, "align", false);
535 if (icount_align_option && !icount_sleep) {
536 error_setg(errp, "align=on and sleep=no are incompatible");
538 if (strcmp(option, "auto") != 0) {
540 icount_time_shift = strtol(option, &rem_str, 0);
541 if (errno != 0 || *rem_str != '\0' || !strlen(option)) {
542 error_setg(errp, "icount: Invalid shift value");
546 } else if (icount_align_option) {
547 error_setg(errp, "shift=auto and align=on are incompatible");
548 } else if (!icount_sleep) {
549 error_setg(errp, "shift=auto and sleep=no are incompatible");
554 /* 125MIPS seems a reasonable initial guess at the guest speed.
555 It will be corrected fairly quickly anyway. */
556 icount_time_shift = 3;
558 /* Have both realtime and virtual time triggers for speed adjustment.
559 The realtime trigger catches emulated time passing too slowly,
560 the virtual time trigger catches emulated time passing too fast.
561 Realtime triggers occur even when idle, so use them less frequently
563 icount_rt_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL_RT,
564 icount_adjust_rt, NULL);
565 timer_mod(icount_rt_timer,
566 qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000);
567 icount_vm_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
568 icount_adjust_vm, NULL);
569 timer_mod(icount_vm_timer,
570 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
571 get_ticks_per_sec() / 10);
574 /***********************************************************/
575 void hw_error(const char *fmt, ...)
581 fprintf(stderr, "qemu: hardware error: ");
582 vfprintf(stderr, fmt, ap);
583 fprintf(stderr, "\n");
585 fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
586 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU);
592 void cpu_synchronize_all_states(void)
597 cpu_synchronize_state(cpu);
601 void cpu_synchronize_all_post_reset(void)
606 cpu_synchronize_post_reset(cpu);
610 void cpu_synchronize_all_post_init(void)
615 cpu_synchronize_post_init(cpu);
619 void cpu_clean_all_dirty(void)
624 cpu_clean_state(cpu);
628 static int do_vm_stop(RunState state)
632 if (runstate_is_running()) {
636 vm_state_notify(0, state);
637 qapi_event_send_stop(&error_abort);
641 ret = bdrv_flush_all();
646 static bool cpu_can_run(CPUState *cpu)
651 if (cpu_is_stopped(cpu)) {
657 static void cpu_handle_guest_debug(CPUState *cpu)
659 gdb_set_stop_cpu(cpu);
660 qemu_system_debug_request();
664 static void cpu_signal(int sig)
666 CPUState *cpu = atomic_mb_read(&tcg_current_cpu);
674 static void sigbus_reraise(void)
677 struct sigaction action;
679 memset(&action, 0, sizeof(action));
680 action.sa_handler = SIG_DFL;
681 if (!sigaction(SIGBUS, &action, NULL)) {
684 sigaddset(&set, SIGBUS);
685 sigprocmask(SIG_UNBLOCK, &set, NULL);
687 perror("Failed to re-raise SIGBUS!\n");
691 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
694 if (kvm_on_sigbus(siginfo->ssi_code,
695 (void *)(intptr_t)siginfo->ssi_addr)) {
700 static void qemu_init_sigbus(void)
702 struct sigaction action;
704 memset(&action, 0, sizeof(action));
705 action.sa_flags = SA_SIGINFO;
706 action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
707 sigaction(SIGBUS, &action, NULL);
709 prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
712 static void qemu_kvm_eat_signals(CPUState *cpu)
714 struct timespec ts = { 0, 0 };
720 sigemptyset(&waitset);
721 sigaddset(&waitset, SIG_IPI);
722 sigaddset(&waitset, SIGBUS);
725 r = sigtimedwait(&waitset, &siginfo, &ts);
726 if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
727 perror("sigtimedwait");
733 if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) {
741 r = sigpending(&chkset);
743 perror("sigpending");
746 } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
749 #else /* !CONFIG_LINUX */
751 static void qemu_init_sigbus(void)
755 static void qemu_kvm_eat_signals(CPUState *cpu)
758 #endif /* !CONFIG_LINUX */
761 static void dummy_signal(int sig)
765 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
769 struct sigaction sigact;
771 memset(&sigact, 0, sizeof(sigact));
772 sigact.sa_handler = dummy_signal;
773 sigaction(SIG_IPI, &sigact, NULL);
775 pthread_sigmask(SIG_BLOCK, NULL, &set);
776 sigdelset(&set, SIG_IPI);
777 sigdelset(&set, SIGBUS);
778 r = kvm_set_signal_mask(cpu, &set);
780 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
785 static void qemu_tcg_init_cpu_signals(void)
788 struct sigaction sigact;
790 memset(&sigact, 0, sizeof(sigact));
791 sigact.sa_handler = cpu_signal;
792 sigaction(SIG_IPI, &sigact, NULL);
795 sigaddset(&set, SIG_IPI);
796 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
800 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
805 static void qemu_tcg_init_cpu_signals(void)
810 static QemuMutex qemu_global_mutex;
811 static QemuCond qemu_io_proceeded_cond;
812 static unsigned iothread_requesting_mutex;
814 static QemuThread io_thread;
816 static QemuThread *tcg_cpu_thread;
817 static QemuCond *tcg_halt_cond;
820 static QemuCond qemu_cpu_cond;
822 static QemuCond qemu_pause_cond;
823 static QemuCond qemu_work_cond;
825 void qemu_init_cpu_loop(void)
828 qemu_cond_init(&qemu_cpu_cond);
829 qemu_cond_init(&qemu_pause_cond);
830 qemu_cond_init(&qemu_work_cond);
831 qemu_cond_init(&qemu_io_proceeded_cond);
832 qemu_mutex_init(&qemu_global_mutex);
834 qemu_thread_get_self(&io_thread);
837 void run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
839 struct qemu_work_item wi;
841 if (qemu_cpu_is_self(cpu)) {
849 if (cpu->queued_work_first == NULL) {
850 cpu->queued_work_first = &wi;
852 cpu->queued_work_last->next = &wi;
854 cpu->queued_work_last = &wi;
860 CPUState *self_cpu = current_cpu;
862 qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
863 current_cpu = self_cpu;
867 void async_run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
869 struct qemu_work_item *wi;
871 if (qemu_cpu_is_self(cpu)) {
876 wi = g_malloc0(sizeof(struct qemu_work_item));
880 if (cpu->queued_work_first == NULL) {
881 cpu->queued_work_first = wi;
883 cpu->queued_work_last->next = wi;
885 cpu->queued_work_last = wi;
892 static void flush_queued_work(CPUState *cpu)
894 struct qemu_work_item *wi;
896 if (cpu->queued_work_first == NULL) {
900 while ((wi = cpu->queued_work_first)) {
901 cpu->queued_work_first = wi->next;
908 cpu->queued_work_last = NULL;
909 qemu_cond_broadcast(&qemu_work_cond);
912 static void qemu_wait_io_event_common(CPUState *cpu)
917 qemu_cond_signal(&qemu_pause_cond);
919 flush_queued_work(cpu);
920 cpu->thread_kicked = false;
923 static void qemu_tcg_wait_io_event(void)
927 while (all_cpu_threads_idle()) {
928 /* Start accounting real time to the virtual clock if the CPUs
930 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
931 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
934 while (iothread_requesting_mutex) {
935 qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex);
939 qemu_wait_io_event_common(cpu);
943 static void qemu_kvm_wait_io_event(CPUState *cpu)
945 while (cpu_thread_is_idle(cpu)) {
946 qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
949 qemu_kvm_eat_signals(cpu);
950 qemu_wait_io_event_common(cpu);
953 static void *qemu_kvm_cpu_thread_fn(void *arg)
958 rcu_register_thread();
960 qemu_mutex_lock_iothread();
961 qemu_thread_get_self(cpu->thread);
962 cpu->thread_id = qemu_get_thread_id();
966 r = kvm_init_vcpu(cpu);
968 fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
972 qemu_kvm_init_cpu_signals(cpu);
974 /* signal CPU creation */
976 qemu_cond_signal(&qemu_cpu_cond);
979 if (cpu_can_run(cpu)) {
980 r = kvm_cpu_exec(cpu);
981 if (r == EXCP_DEBUG) {
982 cpu_handle_guest_debug(cpu);
985 qemu_kvm_wait_io_event(cpu);
991 static void *qemu_dummy_cpu_thread_fn(void *arg)
994 fprintf(stderr, "qtest is not supported under Windows\n");
1001 rcu_register_thread();
1003 qemu_mutex_lock_iothread();
1004 qemu_thread_get_self(cpu->thread);
1005 cpu->thread_id = qemu_get_thread_id();
1008 sigemptyset(&waitset);
1009 sigaddset(&waitset, SIG_IPI);
1011 /* signal CPU creation */
1012 cpu->created = true;
1013 qemu_cond_signal(&qemu_cpu_cond);
1018 qemu_mutex_unlock_iothread();
1021 r = sigwait(&waitset, &sig);
1022 } while (r == -1 && (errno == EAGAIN || errno == EINTR));
1027 qemu_mutex_lock_iothread();
1029 qemu_wait_io_event_common(cpu);
1036 static void tcg_exec_all(void);
1038 static void *qemu_tcg_cpu_thread_fn(void *arg)
1040 CPUState *cpu = arg;
1042 rcu_register_thread();
1044 qemu_mutex_lock_iothread();
1045 qemu_tcg_init_cpu_signals();
1046 qemu_thread_get_self(cpu->thread);
1049 cpu->thread_id = qemu_get_thread_id();
1050 cpu->created = true;
1053 qemu_cond_signal(&qemu_cpu_cond);
1055 /* wait for initial kick-off after machine start */
1056 while (first_cpu->stopped) {
1057 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
1059 /* process any pending work */
1061 qemu_wait_io_event_common(cpu);
1065 /* process any pending work */
1072 int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
1074 if (deadline == 0) {
1075 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
1078 qemu_tcg_wait_io_event();
1084 static void qemu_cpu_kick_thread(CPUState *cpu)
1089 err = pthread_kill(cpu->thread->thread, SIG_IPI);
1091 fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
1095 if (!qemu_cpu_is_self(cpu)) {
1098 if (SuspendThread(cpu->hThread) == (DWORD)-1) {
1099 fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
1104 /* On multi-core systems, we are not sure that the thread is actually
1105 * suspended until we can get the context.
1107 tcgContext.ContextFlags = CONTEXT_CONTROL;
1108 while (GetThreadContext(cpu->hThread, &tcgContext) != 0) {
1114 if (ResumeThread(cpu->hThread) == (DWORD)-1) {
1115 fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
1123 void qemu_cpu_kick(CPUState *cpu)
1125 qemu_cond_broadcast(cpu->halt_cond);
1126 if (!tcg_enabled() && !cpu->thread_kicked) {
1127 qemu_cpu_kick_thread(cpu);
1128 cpu->thread_kicked = true;
1132 void qemu_cpu_kick_self(void)
1135 assert(current_cpu);
1137 if (!current_cpu->thread_kicked) {
1138 qemu_cpu_kick_thread(current_cpu);
1139 current_cpu->thread_kicked = true;
1146 bool qemu_cpu_is_self(CPUState *cpu)
1148 return qemu_thread_is_self(cpu->thread);
1151 bool qemu_in_vcpu_thread(void)
1153 return current_cpu && qemu_cpu_is_self(current_cpu);
1156 static __thread bool iothread_locked = false;
1158 bool qemu_mutex_iothread_locked(void)
1160 return iothread_locked;
1163 void qemu_mutex_lock_iothread(void)
1165 atomic_inc(&iothread_requesting_mutex);
1166 /* In the simple case there is no need to bump the VCPU thread out of
1167 * TCG code execution.
1169 if (!tcg_enabled() || qemu_in_vcpu_thread() ||
1170 !first_cpu || !first_cpu->created) {
1171 qemu_mutex_lock(&qemu_global_mutex);
1172 atomic_dec(&iothread_requesting_mutex);
1174 if (qemu_mutex_trylock(&qemu_global_mutex)) {
1175 qemu_cpu_kick_thread(first_cpu);
1176 qemu_mutex_lock(&qemu_global_mutex);
1178 atomic_dec(&iothread_requesting_mutex);
1179 qemu_cond_broadcast(&qemu_io_proceeded_cond);
1181 iothread_locked = true;
1184 void qemu_mutex_unlock_iothread(void)
1186 iothread_locked = false;
1187 qemu_mutex_unlock(&qemu_global_mutex);
1190 static int all_vcpus_paused(void)
1195 if (!cpu->stopped) {
1203 void pause_all_vcpus(void)
1207 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false);
1213 if (qemu_in_vcpu_thread()) {
1215 if (!kvm_enabled()) {
1218 cpu->stopped = true;
1224 while (!all_vcpus_paused()) {
1225 qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
1232 void cpu_resume(CPUState *cpu)
1235 cpu->stopped = false;
1239 void resume_all_vcpus(void)
1243 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true);
1249 /* For temporary buffers for forming a name */
1250 #define VCPU_THREAD_NAME_SIZE 16
1252 static void qemu_tcg_init_vcpu(CPUState *cpu)
1254 char thread_name[VCPU_THREAD_NAME_SIZE];
1256 tcg_cpu_address_space_init(cpu, cpu->as);
1258 /* share a single thread for all cpus with TCG */
1259 if (!tcg_cpu_thread) {
1260 cpu->thread = g_malloc0(sizeof(QemuThread));
1261 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1262 qemu_cond_init(cpu->halt_cond);
1263 tcg_halt_cond = cpu->halt_cond;
1264 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG",
1266 qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn,
1267 cpu, QEMU_THREAD_JOINABLE);
1269 cpu->hThread = qemu_thread_get_handle(cpu->thread);
1271 while (!cpu->created) {
1272 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1274 tcg_cpu_thread = cpu->thread;
1276 cpu->thread = tcg_cpu_thread;
1277 cpu->halt_cond = tcg_halt_cond;
1281 static void qemu_kvm_start_vcpu(CPUState *cpu)
1283 char thread_name[VCPU_THREAD_NAME_SIZE];
1285 cpu->thread = g_malloc0(sizeof(QemuThread));
1286 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1287 qemu_cond_init(cpu->halt_cond);
1288 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/KVM",
1290 qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn,
1291 cpu, QEMU_THREAD_JOINABLE);
1292 while (!cpu->created) {
1293 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1297 static void qemu_dummy_start_vcpu(CPUState *cpu)
1299 char thread_name[VCPU_THREAD_NAME_SIZE];
1301 cpu->thread = g_malloc0(sizeof(QemuThread));
1302 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1303 qemu_cond_init(cpu->halt_cond);
1304 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/DUMMY",
1306 qemu_thread_create(cpu->thread, thread_name, qemu_dummy_cpu_thread_fn, cpu,
1307 QEMU_THREAD_JOINABLE);
1308 while (!cpu->created) {
1309 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1313 void qemu_init_vcpu(CPUState *cpu)
1315 cpu->nr_cores = smp_cores;
1316 cpu->nr_threads = smp_threads;
1317 cpu->stopped = true;
1318 if (kvm_enabled()) {
1319 qemu_kvm_start_vcpu(cpu);
1320 } else if (tcg_enabled()) {
1321 qemu_tcg_init_vcpu(cpu);
1323 qemu_dummy_start_vcpu(cpu);
1327 void cpu_stop_current(void)
1330 current_cpu->stop = false;
1331 current_cpu->stopped = true;
1332 cpu_exit(current_cpu);
1333 qemu_cond_signal(&qemu_pause_cond);
1337 int vm_stop(RunState state)
1339 if (qemu_in_vcpu_thread()) {
1340 qemu_system_vmstop_request_prepare();
1341 qemu_system_vmstop_request(state);
1343 * FIXME: should not return to device code in case
1344 * vm_stop() has been requested.
1350 return do_vm_stop(state);
1353 /* does a state transition even if the VM is already stopped,
1354 current state is forgotten forever */
1355 int vm_stop_force_state(RunState state)
1357 if (runstate_is_running()) {
1358 return vm_stop(state);
1360 runstate_set(state);
1361 /* Make sure to return an error if the flush in a previous vm_stop()
1363 return bdrv_flush_all();
1367 static int tcg_cpu_exec(CPUState *cpu)
1370 #ifdef CONFIG_PROFILER
1374 #ifdef CONFIG_PROFILER
1375 ti = profile_getclock();
1381 timers_state.qemu_icount -= (cpu->icount_decr.u16.low
1382 + cpu->icount_extra);
1383 cpu->icount_decr.u16.low = 0;
1384 cpu->icount_extra = 0;
1385 deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
1387 /* Maintain prior (possibly buggy) behaviour where if no deadline
1388 * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than
1389 * INT32_MAX nanoseconds ahead, we still use INT32_MAX
1392 if ((deadline < 0) || (deadline > INT32_MAX)) {
1393 deadline = INT32_MAX;
1396 count = qemu_icount_round(deadline);
1397 timers_state.qemu_icount += count;
1398 decr = (count > 0xffff) ? 0xffff : count;
1400 cpu->icount_decr.u16.low = decr;
1401 cpu->icount_extra = count;
1403 ret = cpu_exec(cpu);
1404 #ifdef CONFIG_PROFILER
1405 tcg_time += profile_getclock() - ti;
1408 /* Fold pending instructions back into the
1409 instruction counter, and clear the interrupt flag. */
1410 timers_state.qemu_icount -= (cpu->icount_decr.u16.low
1411 + cpu->icount_extra);
1412 cpu->icount_decr.u32 = 0;
1413 cpu->icount_extra = 0;
1418 static void tcg_exec_all(void)
1422 /* Account partial waits to QEMU_CLOCK_VIRTUAL. */
1423 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
1425 if (next_cpu == NULL) {
1426 next_cpu = first_cpu;
1428 for (; next_cpu != NULL && !exit_request; next_cpu = CPU_NEXT(next_cpu)) {
1429 CPUState *cpu = next_cpu;
1431 qemu_clock_enable(QEMU_CLOCK_VIRTUAL,
1432 (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0);
1434 if (cpu_can_run(cpu)) {
1435 r = tcg_cpu_exec(cpu);
1436 if (r == EXCP_DEBUG) {
1437 cpu_handle_guest_debug(cpu);
1440 } else if (cpu->stop || cpu->stopped) {
1447 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1449 /* XXX: implement xxx_cpu_list for targets that still miss it */
1450 #if defined(cpu_list)
1451 cpu_list(f, cpu_fprintf);
1455 CpuInfoList *qmp_query_cpus(Error **errp)
1457 CpuInfoList *head = NULL, *cur_item = NULL;
1462 #if defined(TARGET_I386)
1463 X86CPU *x86_cpu = X86_CPU(cpu);
1464 CPUX86State *env = &x86_cpu->env;
1465 #elif defined(TARGET_PPC)
1466 PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu);
1467 CPUPPCState *env = &ppc_cpu->env;
1468 #elif defined(TARGET_SPARC)
1469 SPARCCPU *sparc_cpu = SPARC_CPU(cpu);
1470 CPUSPARCState *env = &sparc_cpu->env;
1471 #elif defined(TARGET_MIPS)
1472 MIPSCPU *mips_cpu = MIPS_CPU(cpu);
1473 CPUMIPSState *env = &mips_cpu->env;
1474 #elif defined(TARGET_TRICORE)
1475 TriCoreCPU *tricore_cpu = TRICORE_CPU(cpu);
1476 CPUTriCoreState *env = &tricore_cpu->env;
1479 cpu_synchronize_state(cpu);
1481 info = g_malloc0(sizeof(*info));
1482 info->value = g_malloc0(sizeof(*info->value));
1483 info->value->CPU = cpu->cpu_index;
1484 info->value->current = (cpu == first_cpu);
1485 info->value->halted = cpu->halted;
1486 info->value->qom_path = object_get_canonical_path(OBJECT(cpu));
1487 info->value->thread_id = cpu->thread_id;
1488 #if defined(TARGET_I386)
1489 info->value->has_pc = true;
1490 info->value->pc = env->eip + env->segs[R_CS].base;
1491 #elif defined(TARGET_PPC)
1492 info->value->has_nip = true;
1493 info->value->nip = env->nip;
1494 #elif defined(TARGET_SPARC)
1495 info->value->has_pc = true;
1496 info->value->pc = env->pc;
1497 info->value->has_npc = true;
1498 info->value->npc = env->npc;
1499 #elif defined(TARGET_MIPS)
1500 info->value->has_PC = true;
1501 info->value->PC = env->active_tc.PC;
1502 #elif defined(TARGET_TRICORE)
1503 info->value->has_PC = true;
1504 info->value->PC = env->PC;
1507 /* XXX: waiting for the qapi to support GSList */
1509 head = cur_item = info;
1511 cur_item->next = info;
1519 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
1520 bool has_cpu, int64_t cpu_index, Error **errp)
1526 int64_t orig_addr = addr, orig_size = size;
1532 cpu = qemu_get_cpu(cpu_index);
1534 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
1539 f = fopen(filename, "wb");
1541 error_setg_file_open(errp, errno, filename);
1549 if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) {
1550 error_setg(errp, "Invalid addr 0x%016" PRIx64 "/size %" PRId64
1551 " specified", orig_addr, orig_size);
1554 if (fwrite(buf, 1, l, f) != l) {
1555 error_setg(errp, QERR_IO_ERROR);
1566 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
1573 f = fopen(filename, "wb");
1575 error_setg_file_open(errp, errno, filename);
1583 cpu_physical_memory_read(addr, buf, l);
1584 if (fwrite(buf, 1, l, f) != l) {
1585 error_setg(errp, QERR_IO_ERROR);
1596 void qmp_inject_nmi(Error **errp)
1598 #if defined(TARGET_I386)
1602 X86CPU *cpu = X86_CPU(cs);
1604 if (!cpu->apic_state) {
1605 cpu_interrupt(cs, CPU_INTERRUPT_NMI);
1607 apic_deliver_nmi(cpu->apic_state);
1611 nmi_monitor_handle(monitor_get_cpu_index(), errp);
1615 void dump_drift_info(FILE *f, fprintf_function cpu_fprintf)
1621 cpu_fprintf(f, "Host - Guest clock %"PRIi64" ms\n",
1622 (cpu_get_clock() - cpu_get_icount())/SCALE_MS);
1623 if (icount_align_option) {
1624 cpu_fprintf(f, "Max guest delay %"PRIi64" ms\n", -max_delay/SCALE_MS);
1625 cpu_fprintf(f, "Max guest advance %"PRIi64" ms\n", max_advance/SCALE_MS);
1627 cpu_fprintf(f, "Max guest delay NA\n");
1628 cpu_fprintf(f, "Max guest advance NA\n");