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();
665 static void sigbus_reraise(void)
668 struct sigaction action;
670 memset(&action, 0, sizeof(action));
671 action.sa_handler = SIG_DFL;
672 if (!sigaction(SIGBUS, &action, NULL)) {
675 sigaddset(&set, SIGBUS);
676 sigprocmask(SIG_UNBLOCK, &set, NULL);
678 perror("Failed to re-raise SIGBUS!\n");
682 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
685 if (kvm_on_sigbus(siginfo->ssi_code,
686 (void *)(intptr_t)siginfo->ssi_addr)) {
691 static void qemu_init_sigbus(void)
693 struct sigaction action;
695 memset(&action, 0, sizeof(action));
696 action.sa_flags = SA_SIGINFO;
697 action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
698 sigaction(SIGBUS, &action, NULL);
700 prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
703 static void qemu_kvm_eat_signals(CPUState *cpu)
705 struct timespec ts = { 0, 0 };
711 sigemptyset(&waitset);
712 sigaddset(&waitset, SIG_IPI);
713 sigaddset(&waitset, SIGBUS);
716 r = sigtimedwait(&waitset, &siginfo, &ts);
717 if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
718 perror("sigtimedwait");
724 if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) {
732 r = sigpending(&chkset);
734 perror("sigpending");
737 } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
740 #else /* !CONFIG_LINUX */
742 static void qemu_init_sigbus(void)
746 static void qemu_kvm_eat_signals(CPUState *cpu)
749 #endif /* !CONFIG_LINUX */
752 static void dummy_signal(int sig)
756 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
760 struct sigaction sigact;
762 memset(&sigact, 0, sizeof(sigact));
763 sigact.sa_handler = dummy_signal;
764 sigaction(SIG_IPI, &sigact, NULL);
766 pthread_sigmask(SIG_BLOCK, NULL, &set);
767 sigdelset(&set, SIG_IPI);
768 sigdelset(&set, SIGBUS);
769 r = kvm_set_signal_mask(cpu, &set);
771 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
777 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
783 static QemuMutex qemu_global_mutex;
784 static QemuCond qemu_io_proceeded_cond;
785 static unsigned iothread_requesting_mutex;
787 static QemuThread io_thread;
789 static QemuThread *tcg_cpu_thread;
790 static QemuCond *tcg_halt_cond;
793 static QemuCond qemu_cpu_cond;
795 static QemuCond qemu_pause_cond;
796 static QemuCond qemu_work_cond;
798 void qemu_init_cpu_loop(void)
801 qemu_cond_init(&qemu_cpu_cond);
802 qemu_cond_init(&qemu_pause_cond);
803 qemu_cond_init(&qemu_work_cond);
804 qemu_cond_init(&qemu_io_proceeded_cond);
805 qemu_mutex_init(&qemu_global_mutex);
807 qemu_thread_get_self(&io_thread);
810 void run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
812 struct qemu_work_item wi;
814 if (qemu_cpu_is_self(cpu)) {
823 qemu_mutex_lock(&cpu->work_mutex);
824 if (cpu->queued_work_first == NULL) {
825 cpu->queued_work_first = &wi;
827 cpu->queued_work_last->next = &wi;
829 cpu->queued_work_last = &wi;
832 qemu_mutex_unlock(&cpu->work_mutex);
835 while (!atomic_mb_read(&wi.done)) {
836 CPUState *self_cpu = current_cpu;
838 qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
839 current_cpu = self_cpu;
843 void async_run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
845 struct qemu_work_item *wi;
847 if (qemu_cpu_is_self(cpu)) {
852 wi = g_malloc0(sizeof(struct qemu_work_item));
857 qemu_mutex_lock(&cpu->work_mutex);
858 if (cpu->queued_work_first == NULL) {
859 cpu->queued_work_first = wi;
861 cpu->queued_work_last->next = wi;
863 cpu->queued_work_last = wi;
866 qemu_mutex_unlock(&cpu->work_mutex);
871 static void flush_queued_work(CPUState *cpu)
873 struct qemu_work_item *wi;
875 if (cpu->queued_work_first == NULL) {
879 qemu_mutex_lock(&cpu->work_mutex);
880 while (cpu->queued_work_first != NULL) {
881 wi = cpu->queued_work_first;
882 cpu->queued_work_first = wi->next;
883 if (!cpu->queued_work_first) {
884 cpu->queued_work_last = NULL;
886 qemu_mutex_unlock(&cpu->work_mutex);
888 qemu_mutex_lock(&cpu->work_mutex);
892 atomic_mb_set(&wi->done, true);
895 qemu_mutex_unlock(&cpu->work_mutex);
896 qemu_cond_broadcast(&qemu_work_cond);
899 static void qemu_wait_io_event_common(CPUState *cpu)
904 qemu_cond_signal(&qemu_pause_cond);
906 flush_queued_work(cpu);
907 cpu->thread_kicked = false;
910 static void qemu_tcg_wait_io_event(void)
914 while (all_cpu_threads_idle()) {
915 /* Start accounting real time to the virtual clock if the CPUs
917 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
918 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
921 while (iothread_requesting_mutex) {
922 qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex);
926 qemu_wait_io_event_common(cpu);
930 static void qemu_kvm_wait_io_event(CPUState *cpu)
932 while (cpu_thread_is_idle(cpu)) {
933 qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
936 qemu_kvm_eat_signals(cpu);
937 qemu_wait_io_event_common(cpu);
940 static void *qemu_kvm_cpu_thread_fn(void *arg)
945 rcu_register_thread();
947 qemu_mutex_lock_iothread();
948 qemu_thread_get_self(cpu->thread);
949 cpu->thread_id = qemu_get_thread_id();
953 r = kvm_init_vcpu(cpu);
955 fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
959 qemu_kvm_init_cpu_signals(cpu);
961 /* signal CPU creation */
963 qemu_cond_signal(&qemu_cpu_cond);
966 if (cpu_can_run(cpu)) {
967 r = kvm_cpu_exec(cpu);
968 if (r == EXCP_DEBUG) {
969 cpu_handle_guest_debug(cpu);
972 qemu_kvm_wait_io_event(cpu);
978 static void *qemu_dummy_cpu_thread_fn(void *arg)
981 fprintf(stderr, "qtest is not supported under Windows\n");
988 rcu_register_thread();
990 qemu_mutex_lock_iothread();
991 qemu_thread_get_self(cpu->thread);
992 cpu->thread_id = qemu_get_thread_id();
995 sigemptyset(&waitset);
996 sigaddset(&waitset, SIG_IPI);
998 /* signal CPU creation */
1000 qemu_cond_signal(&qemu_cpu_cond);
1005 qemu_mutex_unlock_iothread();
1008 r = sigwait(&waitset, &sig);
1009 } while (r == -1 && (errno == EAGAIN || errno == EINTR));
1014 qemu_mutex_lock_iothread();
1016 qemu_wait_io_event_common(cpu);
1023 static void tcg_exec_all(void);
1025 static void *qemu_tcg_cpu_thread_fn(void *arg)
1027 CPUState *cpu = arg;
1029 rcu_register_thread();
1031 qemu_mutex_lock_iothread();
1032 qemu_thread_get_self(cpu->thread);
1035 cpu->thread_id = qemu_get_thread_id();
1036 cpu->created = true;
1039 qemu_cond_signal(&qemu_cpu_cond);
1041 /* wait for initial kick-off after machine start */
1042 while (first_cpu->stopped) {
1043 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
1045 /* process any pending work */
1047 qemu_wait_io_event_common(cpu);
1051 /* process any pending work */
1052 atomic_mb_set(&exit_request, 1);
1058 int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
1060 if (deadline == 0) {
1061 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
1064 qemu_tcg_wait_io_event();
1070 static void qemu_cpu_kick_thread(CPUState *cpu)
1075 if (cpu->thread_kicked) {
1078 cpu->thread_kicked = true;
1079 err = pthread_kill(cpu->thread->thread, SIG_IPI);
1081 fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
1089 static void qemu_cpu_kick_no_halt(void)
1092 /* Ensure whatever caused the exit has reached the CPU threads before
1093 * writing exit_request.
1095 atomic_mb_set(&exit_request, 1);
1096 cpu = atomic_mb_read(&tcg_current_cpu);
1102 void qemu_cpu_kick(CPUState *cpu)
1104 qemu_cond_broadcast(cpu->halt_cond);
1105 if (tcg_enabled()) {
1106 qemu_cpu_kick_no_halt();
1108 qemu_cpu_kick_thread(cpu);
1112 void qemu_cpu_kick_self(void)
1114 assert(current_cpu);
1115 qemu_cpu_kick_thread(current_cpu);
1118 bool qemu_cpu_is_self(CPUState *cpu)
1120 return qemu_thread_is_self(cpu->thread);
1123 bool qemu_in_vcpu_thread(void)
1125 return current_cpu && qemu_cpu_is_self(current_cpu);
1128 static __thread bool iothread_locked = false;
1130 bool qemu_mutex_iothread_locked(void)
1132 return iothread_locked;
1135 void qemu_mutex_lock_iothread(void)
1137 atomic_inc(&iothread_requesting_mutex);
1138 /* In the simple case there is no need to bump the VCPU thread out of
1139 * TCG code execution.
1141 if (!tcg_enabled() || qemu_in_vcpu_thread() ||
1142 !first_cpu || !first_cpu->created) {
1143 qemu_mutex_lock(&qemu_global_mutex);
1144 atomic_dec(&iothread_requesting_mutex);
1146 if (qemu_mutex_trylock(&qemu_global_mutex)) {
1147 qemu_cpu_kick_no_halt();
1148 qemu_mutex_lock(&qemu_global_mutex);
1150 atomic_dec(&iothread_requesting_mutex);
1151 qemu_cond_broadcast(&qemu_io_proceeded_cond);
1153 iothread_locked = true;
1156 void qemu_mutex_unlock_iothread(void)
1158 iothread_locked = false;
1159 qemu_mutex_unlock(&qemu_global_mutex);
1162 static int all_vcpus_paused(void)
1167 if (!cpu->stopped) {
1175 void pause_all_vcpus(void)
1179 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false);
1185 if (qemu_in_vcpu_thread()) {
1187 if (!kvm_enabled()) {
1190 cpu->stopped = true;
1196 while (!all_vcpus_paused()) {
1197 qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
1204 void cpu_resume(CPUState *cpu)
1207 cpu->stopped = false;
1211 void resume_all_vcpus(void)
1215 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true);
1221 /* For temporary buffers for forming a name */
1222 #define VCPU_THREAD_NAME_SIZE 16
1224 static void qemu_tcg_init_vcpu(CPUState *cpu)
1226 char thread_name[VCPU_THREAD_NAME_SIZE];
1228 tcg_cpu_address_space_init(cpu, cpu->as);
1230 /* share a single thread for all cpus with TCG */
1231 if (!tcg_cpu_thread) {
1232 cpu->thread = g_malloc0(sizeof(QemuThread));
1233 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1234 qemu_cond_init(cpu->halt_cond);
1235 tcg_halt_cond = cpu->halt_cond;
1236 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG",
1238 qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn,
1239 cpu, QEMU_THREAD_JOINABLE);
1241 cpu->hThread = qemu_thread_get_handle(cpu->thread);
1243 while (!cpu->created) {
1244 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1246 tcg_cpu_thread = cpu->thread;
1248 cpu->thread = tcg_cpu_thread;
1249 cpu->halt_cond = tcg_halt_cond;
1253 static void qemu_kvm_start_vcpu(CPUState *cpu)
1255 char thread_name[VCPU_THREAD_NAME_SIZE];
1257 cpu->thread = g_malloc0(sizeof(QemuThread));
1258 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1259 qemu_cond_init(cpu->halt_cond);
1260 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/KVM",
1262 qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn,
1263 cpu, QEMU_THREAD_JOINABLE);
1264 while (!cpu->created) {
1265 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1269 static void qemu_dummy_start_vcpu(CPUState *cpu)
1271 char thread_name[VCPU_THREAD_NAME_SIZE];
1273 cpu->thread = g_malloc0(sizeof(QemuThread));
1274 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1275 qemu_cond_init(cpu->halt_cond);
1276 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/DUMMY",
1278 qemu_thread_create(cpu->thread, thread_name, qemu_dummy_cpu_thread_fn, cpu,
1279 QEMU_THREAD_JOINABLE);
1280 while (!cpu->created) {
1281 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1285 void qemu_init_vcpu(CPUState *cpu)
1287 cpu->nr_cores = smp_cores;
1288 cpu->nr_threads = smp_threads;
1289 cpu->stopped = true;
1290 if (kvm_enabled()) {
1291 qemu_kvm_start_vcpu(cpu);
1292 } else if (tcg_enabled()) {
1293 qemu_tcg_init_vcpu(cpu);
1295 qemu_dummy_start_vcpu(cpu);
1299 void cpu_stop_current(void)
1302 current_cpu->stop = false;
1303 current_cpu->stopped = true;
1304 cpu_exit(current_cpu);
1305 qemu_cond_signal(&qemu_pause_cond);
1309 int vm_stop(RunState state)
1311 if (qemu_in_vcpu_thread()) {
1312 qemu_system_vmstop_request_prepare();
1313 qemu_system_vmstop_request(state);
1315 * FIXME: should not return to device code in case
1316 * vm_stop() has been requested.
1322 return do_vm_stop(state);
1325 /* does a state transition even if the VM is already stopped,
1326 current state is forgotten forever */
1327 int vm_stop_force_state(RunState state)
1329 if (runstate_is_running()) {
1330 return vm_stop(state);
1332 runstate_set(state);
1333 /* Make sure to return an error if the flush in a previous vm_stop()
1335 return bdrv_flush_all();
1339 static int tcg_cpu_exec(CPUState *cpu)
1342 #ifdef CONFIG_PROFILER
1346 #ifdef CONFIG_PROFILER
1347 ti = profile_getclock();
1353 timers_state.qemu_icount -= (cpu->icount_decr.u16.low
1354 + cpu->icount_extra);
1355 cpu->icount_decr.u16.low = 0;
1356 cpu->icount_extra = 0;
1357 deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
1359 /* Maintain prior (possibly buggy) behaviour where if no deadline
1360 * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than
1361 * INT32_MAX nanoseconds ahead, we still use INT32_MAX
1364 if ((deadline < 0) || (deadline > INT32_MAX)) {
1365 deadline = INT32_MAX;
1368 count = qemu_icount_round(deadline);
1369 timers_state.qemu_icount += count;
1370 decr = (count > 0xffff) ? 0xffff : count;
1372 cpu->icount_decr.u16.low = decr;
1373 cpu->icount_extra = count;
1375 ret = cpu_exec(cpu);
1376 #ifdef CONFIG_PROFILER
1377 tcg_time += profile_getclock() - ti;
1380 /* Fold pending instructions back into the
1381 instruction counter, and clear the interrupt flag. */
1382 timers_state.qemu_icount -= (cpu->icount_decr.u16.low
1383 + cpu->icount_extra);
1384 cpu->icount_decr.u32 = 0;
1385 cpu->icount_extra = 0;
1390 static void tcg_exec_all(void)
1394 /* Account partial waits to QEMU_CLOCK_VIRTUAL. */
1395 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
1397 if (next_cpu == NULL) {
1398 next_cpu = first_cpu;
1400 for (; next_cpu != NULL && !exit_request; next_cpu = CPU_NEXT(next_cpu)) {
1401 CPUState *cpu = next_cpu;
1403 qemu_clock_enable(QEMU_CLOCK_VIRTUAL,
1404 (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0);
1406 if (cpu_can_run(cpu)) {
1407 r = tcg_cpu_exec(cpu);
1408 if (r == EXCP_DEBUG) {
1409 cpu_handle_guest_debug(cpu);
1412 } else if (cpu->stop || cpu->stopped) {
1417 /* Pairs with smp_wmb in qemu_cpu_kick. */
1418 atomic_mb_set(&exit_request, 0);
1421 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1423 /* XXX: implement xxx_cpu_list for targets that still miss it */
1424 #if defined(cpu_list)
1425 cpu_list(f, cpu_fprintf);
1429 CpuInfoList *qmp_query_cpus(Error **errp)
1431 CpuInfoList *head = NULL, *cur_item = NULL;
1436 #if defined(TARGET_I386)
1437 X86CPU *x86_cpu = X86_CPU(cpu);
1438 CPUX86State *env = &x86_cpu->env;
1439 #elif defined(TARGET_PPC)
1440 PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu);
1441 CPUPPCState *env = &ppc_cpu->env;
1442 #elif defined(TARGET_SPARC)
1443 SPARCCPU *sparc_cpu = SPARC_CPU(cpu);
1444 CPUSPARCState *env = &sparc_cpu->env;
1445 #elif defined(TARGET_MIPS)
1446 MIPSCPU *mips_cpu = MIPS_CPU(cpu);
1447 CPUMIPSState *env = &mips_cpu->env;
1448 #elif defined(TARGET_TRICORE)
1449 TriCoreCPU *tricore_cpu = TRICORE_CPU(cpu);
1450 CPUTriCoreState *env = &tricore_cpu->env;
1453 cpu_synchronize_state(cpu);
1455 info = g_malloc0(sizeof(*info));
1456 info->value = g_malloc0(sizeof(*info->value));
1457 info->value->CPU = cpu->cpu_index;
1458 info->value->current = (cpu == first_cpu);
1459 info->value->halted = cpu->halted;
1460 info->value->qom_path = object_get_canonical_path(OBJECT(cpu));
1461 info->value->thread_id = cpu->thread_id;
1462 #if defined(TARGET_I386)
1463 info->value->has_pc = true;
1464 info->value->pc = env->eip + env->segs[R_CS].base;
1465 #elif defined(TARGET_PPC)
1466 info->value->has_nip = true;
1467 info->value->nip = env->nip;
1468 #elif defined(TARGET_SPARC)
1469 info->value->has_pc = true;
1470 info->value->pc = env->pc;
1471 info->value->has_npc = true;
1472 info->value->npc = env->npc;
1473 #elif defined(TARGET_MIPS)
1474 info->value->has_PC = true;
1475 info->value->PC = env->active_tc.PC;
1476 #elif defined(TARGET_TRICORE)
1477 info->value->has_PC = true;
1478 info->value->PC = env->PC;
1481 /* XXX: waiting for the qapi to support GSList */
1483 head = cur_item = info;
1485 cur_item->next = info;
1493 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
1494 bool has_cpu, int64_t cpu_index, Error **errp)
1500 int64_t orig_addr = addr, orig_size = size;
1506 cpu = qemu_get_cpu(cpu_index);
1508 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
1513 f = fopen(filename, "wb");
1515 error_setg_file_open(errp, errno, filename);
1523 if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) {
1524 error_setg(errp, "Invalid addr 0x%016" PRIx64 "/size %" PRId64
1525 " specified", orig_addr, orig_size);
1528 if (fwrite(buf, 1, l, f) != l) {
1529 error_setg(errp, QERR_IO_ERROR);
1540 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
1547 f = fopen(filename, "wb");
1549 error_setg_file_open(errp, errno, filename);
1557 cpu_physical_memory_read(addr, buf, l);
1558 if (fwrite(buf, 1, l, f) != l) {
1559 error_setg(errp, QERR_IO_ERROR);
1570 void qmp_inject_nmi(Error **errp)
1572 #if defined(TARGET_I386)
1576 X86CPU *cpu = X86_CPU(cs);
1578 if (!cpu->apic_state) {
1579 cpu_interrupt(cs, CPU_INTERRUPT_NMI);
1581 apic_deliver_nmi(cpu->apic_state);
1585 nmi_monitor_handle(monitor_get_cpu_index(), errp);
1589 void dump_drift_info(FILE *f, fprintf_function cpu_fprintf)
1595 cpu_fprintf(f, "Host - Guest clock %"PRIi64" ms\n",
1596 (cpu_get_clock() - cpu_get_icount())/SCALE_MS);
1597 if (icount_align_option) {
1598 cpu_fprintf(f, "Max guest delay %"PRIi64" ms\n", -max_delay/SCALE_MS);
1599 cpu_fprintf(f, "Max guest advance %"PRIi64" ms\n", max_advance/SCALE_MS);
1601 cpu_fprintf(f, "Max guest delay NA\n");
1602 cpu_fprintf(f, "Max guest advance NA\n");