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 /* vcpu throttling controls */
73 static QEMUTimer *throttle_timer;
74 static unsigned int throttle_percentage;
76 #define CPU_THROTTLE_PCT_MIN 1
77 #define CPU_THROTTLE_PCT_MAX 99
78 #define CPU_THROTTLE_TIMESLICE_NS 10000000
80 bool cpu_is_stopped(CPUState *cpu)
82 return cpu->stopped || !runstate_is_running();
85 static bool cpu_thread_is_idle(CPUState *cpu)
87 if (cpu->stop || cpu->queued_work_first) {
90 if (cpu_is_stopped(cpu)) {
93 if (!cpu->halted || cpu_has_work(cpu) ||
94 kvm_halt_in_kernel()) {
100 static bool all_cpu_threads_idle(void)
105 if (!cpu_thread_is_idle(cpu)) {
112 /***********************************************************/
113 /* guest cycle counter */
115 /* Protected by TimersState seqlock */
117 static bool icount_sleep = true;
118 static int64_t vm_clock_warp_start = -1;
119 /* Conversion factor from emulated instructions to virtual clock ticks. */
120 static int icount_time_shift;
121 /* Arbitrarily pick 1MIPS as the minimum allowable speed. */
122 #define MAX_ICOUNT_SHIFT 10
124 static QEMUTimer *icount_rt_timer;
125 static QEMUTimer *icount_vm_timer;
126 static QEMUTimer *icount_warp_timer;
128 typedef struct TimersState {
129 /* Protected by BQL. */
130 int64_t cpu_ticks_prev;
131 int64_t cpu_ticks_offset;
133 /* cpu_clock_offset can be read out of BQL, so protect it with
136 QemuSeqLock vm_clock_seqlock;
137 int64_t cpu_clock_offset;
138 int32_t cpu_ticks_enabled;
141 /* Compensate for varying guest execution speed. */
142 int64_t qemu_icount_bias;
143 /* Only written by TCG thread */
147 static TimersState timers_state;
149 int64_t cpu_get_icount_raw(void)
152 CPUState *cpu = current_cpu;
154 icount = timers_state.qemu_icount;
156 if (!cpu->can_do_io) {
157 fprintf(stderr, "Bad icount read\n");
160 icount -= (cpu->icount_decr.u16.low + cpu->icount_extra);
165 /* Return the virtual CPU time, based on the instruction counter. */
166 static int64_t cpu_get_icount_locked(void)
168 int64_t icount = cpu_get_icount_raw();
169 return timers_state.qemu_icount_bias + cpu_icount_to_ns(icount);
172 int64_t cpu_get_icount(void)
178 start = seqlock_read_begin(&timers_state.vm_clock_seqlock);
179 icount = cpu_get_icount_locked();
180 } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start));
185 int64_t cpu_icount_to_ns(int64_t icount)
187 return icount << icount_time_shift;
190 /* return the host CPU cycle counter and handle stop/restart */
191 /* Caller must hold the BQL */
192 int64_t cpu_get_ticks(void)
197 return cpu_get_icount();
200 ticks = timers_state.cpu_ticks_offset;
201 if (timers_state.cpu_ticks_enabled) {
202 ticks += cpu_get_host_ticks();
205 if (timers_state.cpu_ticks_prev > ticks) {
206 /* Note: non increasing ticks may happen if the host uses
208 timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks;
209 ticks = timers_state.cpu_ticks_prev;
212 timers_state.cpu_ticks_prev = ticks;
216 static int64_t cpu_get_clock_locked(void)
220 ticks = timers_state.cpu_clock_offset;
221 if (timers_state.cpu_ticks_enabled) {
222 ticks += get_clock();
228 /* return the host CPU monotonic timer and handle stop/restart */
229 int64_t cpu_get_clock(void)
235 start = seqlock_read_begin(&timers_state.vm_clock_seqlock);
236 ti = cpu_get_clock_locked();
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_host_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_host_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_VIRTUAL_RT) + 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 = cpu_get_clock_locked();
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_icount = cpu_get_icount_locked();
358 int64_t delta = clock - cur_icount;
359 warp_delta = MIN(warp_delta, delta);
361 timers_state.qemu_icount_bias += warp_delta;
363 vm_clock_warp_start = -1;
364 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
366 if (qemu_clock_expired(QEMU_CLOCK_VIRTUAL)) {
367 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
371 void qtest_clock_warp(int64_t dest)
373 int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
374 AioContext *aio_context;
375 assert(qtest_enabled());
376 aio_context = qemu_get_aio_context();
377 while (clock < dest) {
378 int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
379 int64_t warp = qemu_soonest_timeout(dest - clock, deadline);
381 seqlock_write_lock(&timers_state.vm_clock_seqlock);
382 timers_state.qemu_icount_bias += warp;
383 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
385 qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL);
386 timerlist_run_timers(aio_context->tlg.tl[QEMU_CLOCK_VIRTUAL]);
387 clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
389 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
392 void qemu_clock_warp(QEMUClockType type)
398 * There are too many global variables to make the "warp" behavior
399 * applicable to other clocks. But a clock argument removes the
400 * need for if statements all over the place.
402 if (type != QEMU_CLOCK_VIRTUAL || !use_icount) {
408 * If the CPUs have been sleeping, advance QEMU_CLOCK_VIRTUAL timer now.
409 * This ensures that the deadline for the timer is computed correctly
411 * This also makes sure that the insn counter is synchronized before
412 * the CPU starts running, in case the CPU is woken by an event other
413 * than the earliest QEMU_CLOCK_VIRTUAL timer.
415 icount_warp_rt(NULL);
416 timer_del(icount_warp_timer);
418 if (!all_cpu_threads_idle()) {
422 if (qtest_enabled()) {
423 /* When testing, qtest commands advance icount. */
427 /* We want to use the earliest deadline from ALL vm_clocks */
428 clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
429 deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
431 static bool notified;
432 if (!icount_sleep && !notified) {
433 error_report("WARNING: icount sleep disabled and no active timers");
441 * Ensure QEMU_CLOCK_VIRTUAL proceeds even when the virtual CPU goes to
442 * sleep. Otherwise, the CPU might be waiting for a future timer
443 * interrupt to wake it up, but the interrupt never comes because
444 * the vCPU isn't running any insns and thus doesn't advance the
445 * QEMU_CLOCK_VIRTUAL.
449 * We never let VCPUs sleep in no sleep icount mode.
450 * If there is a pending QEMU_CLOCK_VIRTUAL timer we just advance
451 * to the next QEMU_CLOCK_VIRTUAL event and notify it.
452 * It is useful when we want a deterministic execution time,
453 * isolated from host latencies.
455 seqlock_write_lock(&timers_state.vm_clock_seqlock);
456 timers_state.qemu_icount_bias += deadline;
457 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
458 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
461 * We do stop VCPUs and only advance QEMU_CLOCK_VIRTUAL after some
462 * "real" time, (related to the time left until the next event) has
463 * passed. The QEMU_CLOCK_VIRTUAL_RT clock will do this.
464 * This avoids that the warps are visible externally; for example,
465 * you will not be sending network packets continuously instead of
468 seqlock_write_lock(&timers_state.vm_clock_seqlock);
469 if (vm_clock_warp_start == -1 || vm_clock_warp_start > clock) {
470 vm_clock_warp_start = clock;
472 seqlock_write_unlock(&timers_state.vm_clock_seqlock);
473 timer_mod_anticipate(icount_warp_timer, clock + deadline);
475 } else if (deadline == 0) {
476 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
480 static bool icount_state_needed(void *opaque)
486 * This is a subsection for icount migration.
488 static const VMStateDescription icount_vmstate_timers = {
489 .name = "timer/icount",
491 .minimum_version_id = 1,
492 .needed = icount_state_needed,
493 .fields = (VMStateField[]) {
494 VMSTATE_INT64(qemu_icount_bias, TimersState),
495 VMSTATE_INT64(qemu_icount, TimersState),
496 VMSTATE_END_OF_LIST()
500 static const VMStateDescription vmstate_timers = {
503 .minimum_version_id = 1,
504 .fields = (VMStateField[]) {
505 VMSTATE_INT64(cpu_ticks_offset, TimersState),
506 VMSTATE_INT64(dummy, TimersState),
507 VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
508 VMSTATE_END_OF_LIST()
510 .subsections = (const VMStateDescription*[]) {
511 &icount_vmstate_timers,
516 static void cpu_throttle_thread(void *opaque)
518 CPUState *cpu = opaque;
520 double throttle_ratio;
523 if (!cpu_throttle_get_percentage()) {
527 pct = (double)cpu_throttle_get_percentage()/100;
528 throttle_ratio = pct / (1 - pct);
529 sleeptime_ns = (long)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS);
531 qemu_mutex_unlock_iothread();
532 atomic_set(&cpu->throttle_thread_scheduled, 0);
533 g_usleep(sleeptime_ns / 1000); /* Convert ns to us for usleep call */
534 qemu_mutex_lock_iothread();
537 static void cpu_throttle_timer_tick(void *opaque)
542 /* Stop the timer if needed */
543 if (!cpu_throttle_get_percentage()) {
547 if (!atomic_xchg(&cpu->throttle_thread_scheduled, 1)) {
548 async_run_on_cpu(cpu, cpu_throttle_thread, cpu);
552 pct = (double)cpu_throttle_get_percentage()/100;
553 timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) +
554 CPU_THROTTLE_TIMESLICE_NS / (1-pct));
557 void cpu_throttle_set(int new_throttle_pct)
559 /* Ensure throttle percentage is within valid range */
560 new_throttle_pct = MIN(new_throttle_pct, CPU_THROTTLE_PCT_MAX);
561 new_throttle_pct = MAX(new_throttle_pct, CPU_THROTTLE_PCT_MIN);
563 atomic_set(&throttle_percentage, new_throttle_pct);
565 timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) +
566 CPU_THROTTLE_TIMESLICE_NS);
569 void cpu_throttle_stop(void)
571 atomic_set(&throttle_percentage, 0);
574 bool cpu_throttle_active(void)
576 return (cpu_throttle_get_percentage() != 0);
579 int cpu_throttle_get_percentage(void)
581 return atomic_read(&throttle_percentage);
584 void cpu_ticks_init(void)
586 seqlock_init(&timers_state.vm_clock_seqlock, NULL);
587 vmstate_register(NULL, 0, &vmstate_timers, &timers_state);
588 throttle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT,
589 cpu_throttle_timer_tick, NULL);
592 void configure_icount(QemuOpts *opts, Error **errp)
595 char *rem_str = NULL;
597 option = qemu_opt_get(opts, "shift");
599 if (qemu_opt_get(opts, "align") != NULL) {
600 error_setg(errp, "Please specify shift option when using align");
605 icount_sleep = qemu_opt_get_bool(opts, "sleep", true);
607 icount_warp_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT,
608 icount_warp_rt, NULL);
611 icount_align_option = qemu_opt_get_bool(opts, "align", false);
613 if (icount_align_option && !icount_sleep) {
614 error_setg(errp, "align=on and sleep=no are incompatible");
616 if (strcmp(option, "auto") != 0) {
618 icount_time_shift = strtol(option, &rem_str, 0);
619 if (errno != 0 || *rem_str != '\0' || !strlen(option)) {
620 error_setg(errp, "icount: Invalid shift value");
624 } else if (icount_align_option) {
625 error_setg(errp, "shift=auto and align=on are incompatible");
626 } else if (!icount_sleep) {
627 error_setg(errp, "shift=auto and sleep=no are incompatible");
632 /* 125MIPS seems a reasonable initial guess at the guest speed.
633 It will be corrected fairly quickly anyway. */
634 icount_time_shift = 3;
636 /* Have both realtime and virtual time triggers for speed adjustment.
637 The realtime trigger catches emulated time passing too slowly,
638 the virtual time trigger catches emulated time passing too fast.
639 Realtime triggers occur even when idle, so use them less frequently
641 icount_rt_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL_RT,
642 icount_adjust_rt, NULL);
643 timer_mod(icount_rt_timer,
644 qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000);
645 icount_vm_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
646 icount_adjust_vm, NULL);
647 timer_mod(icount_vm_timer,
648 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
649 get_ticks_per_sec() / 10);
652 /***********************************************************/
653 void hw_error(const char *fmt, ...)
659 fprintf(stderr, "qemu: hardware error: ");
660 vfprintf(stderr, fmt, ap);
661 fprintf(stderr, "\n");
663 fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
664 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU);
670 void cpu_synchronize_all_states(void)
675 cpu_synchronize_state(cpu);
679 void cpu_synchronize_all_post_reset(void)
684 cpu_synchronize_post_reset(cpu);
688 void cpu_synchronize_all_post_init(void)
693 cpu_synchronize_post_init(cpu);
697 void cpu_clean_all_dirty(void)
702 cpu_clean_state(cpu);
706 static int do_vm_stop(RunState state)
710 if (runstate_is_running()) {
714 vm_state_notify(0, state);
715 qapi_event_send_stop(&error_abort);
719 ret = bdrv_flush_all();
724 static bool cpu_can_run(CPUState *cpu)
729 if (cpu_is_stopped(cpu)) {
735 static void cpu_handle_guest_debug(CPUState *cpu)
737 gdb_set_stop_cpu(cpu);
738 qemu_system_debug_request();
743 static void sigbus_reraise(void)
746 struct sigaction action;
748 memset(&action, 0, sizeof(action));
749 action.sa_handler = SIG_DFL;
750 if (!sigaction(SIGBUS, &action, NULL)) {
753 sigaddset(&set, SIGBUS);
754 sigprocmask(SIG_UNBLOCK, &set, NULL);
756 perror("Failed to re-raise SIGBUS!\n");
760 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
763 if (kvm_on_sigbus(siginfo->ssi_code,
764 (void *)(intptr_t)siginfo->ssi_addr)) {
769 static void qemu_init_sigbus(void)
771 struct sigaction action;
773 memset(&action, 0, sizeof(action));
774 action.sa_flags = SA_SIGINFO;
775 action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
776 sigaction(SIGBUS, &action, NULL);
778 prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
781 static void qemu_kvm_eat_signals(CPUState *cpu)
783 struct timespec ts = { 0, 0 };
789 sigemptyset(&waitset);
790 sigaddset(&waitset, SIG_IPI);
791 sigaddset(&waitset, SIGBUS);
794 r = sigtimedwait(&waitset, &siginfo, &ts);
795 if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
796 perror("sigtimedwait");
802 if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) {
810 r = sigpending(&chkset);
812 perror("sigpending");
815 } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
818 #else /* !CONFIG_LINUX */
820 static void qemu_init_sigbus(void)
824 static void qemu_kvm_eat_signals(CPUState *cpu)
827 #endif /* !CONFIG_LINUX */
830 static void dummy_signal(int sig)
834 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
838 struct sigaction sigact;
840 memset(&sigact, 0, sizeof(sigact));
841 sigact.sa_handler = dummy_signal;
842 sigaction(SIG_IPI, &sigact, NULL);
844 pthread_sigmask(SIG_BLOCK, NULL, &set);
845 sigdelset(&set, SIG_IPI);
846 sigdelset(&set, SIGBUS);
847 r = kvm_set_signal_mask(cpu, &set);
849 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
855 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
861 static QemuMutex qemu_global_mutex;
862 static QemuCond qemu_io_proceeded_cond;
863 static unsigned iothread_requesting_mutex;
865 static QemuThread io_thread;
868 static QemuCond qemu_cpu_cond;
870 static QemuCond qemu_pause_cond;
871 static QemuCond qemu_work_cond;
873 void qemu_init_cpu_loop(void)
876 qemu_cond_init(&qemu_cpu_cond);
877 qemu_cond_init(&qemu_pause_cond);
878 qemu_cond_init(&qemu_work_cond);
879 qemu_cond_init(&qemu_io_proceeded_cond);
880 qemu_mutex_init(&qemu_global_mutex);
882 qemu_thread_get_self(&io_thread);
885 void run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
887 struct qemu_work_item wi;
889 if (qemu_cpu_is_self(cpu)) {
898 qemu_mutex_lock(&cpu->work_mutex);
899 if (cpu->queued_work_first == NULL) {
900 cpu->queued_work_first = &wi;
902 cpu->queued_work_last->next = &wi;
904 cpu->queued_work_last = &wi;
907 qemu_mutex_unlock(&cpu->work_mutex);
910 while (!atomic_mb_read(&wi.done)) {
911 CPUState *self_cpu = current_cpu;
913 qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
914 current_cpu = self_cpu;
918 void async_run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
920 struct qemu_work_item *wi;
922 if (qemu_cpu_is_self(cpu)) {
927 wi = g_malloc0(sizeof(struct qemu_work_item));
932 qemu_mutex_lock(&cpu->work_mutex);
933 if (cpu->queued_work_first == NULL) {
934 cpu->queued_work_first = wi;
936 cpu->queued_work_last->next = wi;
938 cpu->queued_work_last = wi;
941 qemu_mutex_unlock(&cpu->work_mutex);
946 static void flush_queued_work(CPUState *cpu)
948 struct qemu_work_item *wi;
950 if (cpu->queued_work_first == NULL) {
954 qemu_mutex_lock(&cpu->work_mutex);
955 while (cpu->queued_work_first != NULL) {
956 wi = cpu->queued_work_first;
957 cpu->queued_work_first = wi->next;
958 if (!cpu->queued_work_first) {
959 cpu->queued_work_last = NULL;
961 qemu_mutex_unlock(&cpu->work_mutex);
963 qemu_mutex_lock(&cpu->work_mutex);
967 atomic_mb_set(&wi->done, true);
970 qemu_mutex_unlock(&cpu->work_mutex);
971 qemu_cond_broadcast(&qemu_work_cond);
974 static void qemu_wait_io_event_common(CPUState *cpu)
979 qemu_cond_signal(&qemu_pause_cond);
981 flush_queued_work(cpu);
982 cpu->thread_kicked = false;
985 static void qemu_tcg_wait_io_event(CPUState *cpu)
987 while (all_cpu_threads_idle()) {
988 /* Start accounting real time to the virtual clock if the CPUs
990 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
991 qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
994 while (iothread_requesting_mutex) {
995 qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex);
999 qemu_wait_io_event_common(cpu);
1003 static void qemu_kvm_wait_io_event(CPUState *cpu)
1005 while (cpu_thread_is_idle(cpu)) {
1006 qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
1009 qemu_kvm_eat_signals(cpu);
1010 qemu_wait_io_event_common(cpu);
1013 static void *qemu_kvm_cpu_thread_fn(void *arg)
1015 CPUState *cpu = arg;
1018 rcu_register_thread();
1020 qemu_mutex_lock_iothread();
1021 qemu_thread_get_self(cpu->thread);
1022 cpu->thread_id = qemu_get_thread_id();
1026 r = kvm_init_vcpu(cpu);
1028 fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
1032 qemu_kvm_init_cpu_signals(cpu);
1034 /* signal CPU creation */
1035 cpu->created = true;
1036 qemu_cond_signal(&qemu_cpu_cond);
1039 if (cpu_can_run(cpu)) {
1040 r = kvm_cpu_exec(cpu);
1041 if (r == EXCP_DEBUG) {
1042 cpu_handle_guest_debug(cpu);
1045 qemu_kvm_wait_io_event(cpu);
1051 static void *qemu_dummy_cpu_thread_fn(void *arg)
1054 fprintf(stderr, "qtest is not supported under Windows\n");
1057 CPUState *cpu = arg;
1061 rcu_register_thread();
1063 qemu_mutex_lock_iothread();
1064 qemu_thread_get_self(cpu->thread);
1065 cpu->thread_id = qemu_get_thread_id();
1068 sigemptyset(&waitset);
1069 sigaddset(&waitset, SIG_IPI);
1071 /* signal CPU creation */
1072 cpu->created = true;
1073 qemu_cond_signal(&qemu_cpu_cond);
1078 qemu_mutex_unlock_iothread();
1081 r = sigwait(&waitset, &sig);
1082 } while (r == -1 && (errno == EAGAIN || errno == EINTR));
1087 qemu_mutex_lock_iothread();
1089 qemu_wait_io_event_common(cpu);
1096 static void tcg_exec_all(void);
1098 static void *qemu_tcg_cpu_thread_fn(void *arg)
1100 CPUState *cpu = arg;
1102 rcu_register_thread();
1104 qemu_mutex_lock_iothread();
1105 qemu_thread_get_self(cpu->thread);
1108 cpu->thread_id = qemu_get_thread_id();
1109 cpu->created = true;
1112 qemu_cond_signal(&qemu_cpu_cond);
1114 /* wait for initial kick-off after machine start */
1115 while (first_cpu->stopped) {
1116 qemu_cond_wait(first_cpu->halt_cond, &qemu_global_mutex);
1118 /* process any pending work */
1120 qemu_wait_io_event_common(cpu);
1124 /* process any pending work */
1125 atomic_mb_set(&exit_request, 1);
1131 int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
1133 if (deadline == 0) {
1134 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
1137 qemu_tcg_wait_io_event(QTAILQ_FIRST(&cpus));
1143 static void qemu_cpu_kick_thread(CPUState *cpu)
1148 if (cpu->thread_kicked) {
1151 cpu->thread_kicked = true;
1152 err = pthread_kill(cpu->thread->thread, SIG_IPI);
1154 fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
1162 static void qemu_cpu_kick_no_halt(void)
1165 /* Ensure whatever caused the exit has reached the CPU threads before
1166 * writing exit_request.
1168 atomic_mb_set(&exit_request, 1);
1169 cpu = atomic_mb_read(&tcg_current_cpu);
1175 void qemu_cpu_kick(CPUState *cpu)
1177 qemu_cond_broadcast(cpu->halt_cond);
1178 if (tcg_enabled()) {
1179 qemu_cpu_kick_no_halt();
1181 qemu_cpu_kick_thread(cpu);
1185 void qemu_cpu_kick_self(void)
1187 assert(current_cpu);
1188 qemu_cpu_kick_thread(current_cpu);
1191 bool qemu_cpu_is_self(CPUState *cpu)
1193 return qemu_thread_is_self(cpu->thread);
1196 bool qemu_in_vcpu_thread(void)
1198 return current_cpu && qemu_cpu_is_self(current_cpu);
1201 static __thread bool iothread_locked = false;
1203 bool qemu_mutex_iothread_locked(void)
1205 return iothread_locked;
1208 void qemu_mutex_lock_iothread(void)
1210 atomic_inc(&iothread_requesting_mutex);
1211 /* In the simple case there is no need to bump the VCPU thread out of
1212 * TCG code execution.
1214 if (!tcg_enabled() || qemu_in_vcpu_thread() ||
1215 !first_cpu || !first_cpu->created) {
1216 qemu_mutex_lock(&qemu_global_mutex);
1217 atomic_dec(&iothread_requesting_mutex);
1219 if (qemu_mutex_trylock(&qemu_global_mutex)) {
1220 qemu_cpu_kick_no_halt();
1221 qemu_mutex_lock(&qemu_global_mutex);
1223 atomic_dec(&iothread_requesting_mutex);
1224 qemu_cond_broadcast(&qemu_io_proceeded_cond);
1226 iothread_locked = true;
1229 void qemu_mutex_unlock_iothread(void)
1231 iothread_locked = false;
1232 qemu_mutex_unlock(&qemu_global_mutex);
1235 static int all_vcpus_paused(void)
1240 if (!cpu->stopped) {
1248 void pause_all_vcpus(void)
1252 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false);
1258 if (qemu_in_vcpu_thread()) {
1260 if (!kvm_enabled()) {
1263 cpu->stopped = true;
1269 while (!all_vcpus_paused()) {
1270 qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
1277 void cpu_resume(CPUState *cpu)
1280 cpu->stopped = false;
1284 void resume_all_vcpus(void)
1288 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true);
1294 /* For temporary buffers for forming a name */
1295 #define VCPU_THREAD_NAME_SIZE 16
1297 static void qemu_tcg_init_vcpu(CPUState *cpu)
1299 char thread_name[VCPU_THREAD_NAME_SIZE];
1300 static QemuCond *tcg_halt_cond;
1301 static QemuThread *tcg_cpu_thread;
1303 tcg_cpu_address_space_init(cpu, cpu->as);
1305 /* share a single thread for all cpus with TCG */
1306 if (!tcg_cpu_thread) {
1307 cpu->thread = g_malloc0(sizeof(QemuThread));
1308 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1309 qemu_cond_init(cpu->halt_cond);
1310 tcg_halt_cond = cpu->halt_cond;
1311 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG",
1313 qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn,
1314 cpu, QEMU_THREAD_JOINABLE);
1316 cpu->hThread = qemu_thread_get_handle(cpu->thread);
1318 while (!cpu->created) {
1319 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1321 tcg_cpu_thread = cpu->thread;
1323 cpu->thread = tcg_cpu_thread;
1324 cpu->halt_cond = tcg_halt_cond;
1328 static void qemu_kvm_start_vcpu(CPUState *cpu)
1330 char thread_name[VCPU_THREAD_NAME_SIZE];
1332 cpu->thread = g_malloc0(sizeof(QemuThread));
1333 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1334 qemu_cond_init(cpu->halt_cond);
1335 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/KVM",
1337 qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn,
1338 cpu, QEMU_THREAD_JOINABLE);
1339 while (!cpu->created) {
1340 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1344 static void qemu_dummy_start_vcpu(CPUState *cpu)
1346 char thread_name[VCPU_THREAD_NAME_SIZE];
1348 cpu->thread = g_malloc0(sizeof(QemuThread));
1349 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1350 qemu_cond_init(cpu->halt_cond);
1351 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/DUMMY",
1353 qemu_thread_create(cpu->thread, thread_name, qemu_dummy_cpu_thread_fn, cpu,
1354 QEMU_THREAD_JOINABLE);
1355 while (!cpu->created) {
1356 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1360 void qemu_init_vcpu(CPUState *cpu)
1362 cpu->nr_cores = smp_cores;
1363 cpu->nr_threads = smp_threads;
1364 cpu->stopped = true;
1365 if (kvm_enabled()) {
1366 qemu_kvm_start_vcpu(cpu);
1367 } else if (tcg_enabled()) {
1368 qemu_tcg_init_vcpu(cpu);
1370 qemu_dummy_start_vcpu(cpu);
1374 void cpu_stop_current(void)
1377 current_cpu->stop = false;
1378 current_cpu->stopped = true;
1379 cpu_exit(current_cpu);
1380 qemu_cond_signal(&qemu_pause_cond);
1384 int vm_stop(RunState state)
1386 if (qemu_in_vcpu_thread()) {
1387 qemu_system_vmstop_request_prepare();
1388 qemu_system_vmstop_request(state);
1390 * FIXME: should not return to device code in case
1391 * vm_stop() has been requested.
1397 return do_vm_stop(state);
1400 /* does a state transition even if the VM is already stopped,
1401 current state is forgotten forever */
1402 int vm_stop_force_state(RunState state)
1404 if (runstate_is_running()) {
1405 return vm_stop(state);
1407 runstate_set(state);
1408 /* Make sure to return an error if the flush in a previous vm_stop()
1410 return bdrv_flush_all();
1414 static int tcg_cpu_exec(CPUState *cpu)
1417 #ifdef CONFIG_PROFILER
1421 #ifdef CONFIG_PROFILER
1422 ti = profile_getclock();
1428 timers_state.qemu_icount -= (cpu->icount_decr.u16.low
1429 + cpu->icount_extra);
1430 cpu->icount_decr.u16.low = 0;
1431 cpu->icount_extra = 0;
1432 deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
1434 /* Maintain prior (possibly buggy) behaviour where if no deadline
1435 * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than
1436 * INT32_MAX nanoseconds ahead, we still use INT32_MAX
1439 if ((deadline < 0) || (deadline > INT32_MAX)) {
1440 deadline = INT32_MAX;
1443 count = qemu_icount_round(deadline);
1444 timers_state.qemu_icount += count;
1445 decr = (count > 0xffff) ? 0xffff : count;
1447 cpu->icount_decr.u16.low = decr;
1448 cpu->icount_extra = count;
1450 ret = cpu_exec(cpu);
1451 #ifdef CONFIG_PROFILER
1452 tcg_time += profile_getclock() - ti;
1455 /* Fold pending instructions back into the
1456 instruction counter, and clear the interrupt flag. */
1457 timers_state.qemu_icount -= (cpu->icount_decr.u16.low
1458 + cpu->icount_extra);
1459 cpu->icount_decr.u32 = 0;
1460 cpu->icount_extra = 0;
1465 static void tcg_exec_all(void)
1469 /* Account partial waits to QEMU_CLOCK_VIRTUAL. */
1470 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
1472 if (next_cpu == NULL) {
1473 next_cpu = first_cpu;
1475 for (; next_cpu != NULL && !exit_request; next_cpu = CPU_NEXT(next_cpu)) {
1476 CPUState *cpu = next_cpu;
1478 qemu_clock_enable(QEMU_CLOCK_VIRTUAL,
1479 (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0);
1481 if (cpu_can_run(cpu)) {
1482 r = tcg_cpu_exec(cpu);
1483 if (r == EXCP_DEBUG) {
1484 cpu_handle_guest_debug(cpu);
1487 } else if (cpu->stop || cpu->stopped) {
1492 /* Pairs with smp_wmb in qemu_cpu_kick. */
1493 atomic_mb_set(&exit_request, 0);
1496 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1498 /* XXX: implement xxx_cpu_list for targets that still miss it */
1499 #if defined(cpu_list)
1500 cpu_list(f, cpu_fprintf);
1504 CpuInfoList *qmp_query_cpus(Error **errp)
1506 CpuInfoList *head = NULL, *cur_item = NULL;
1511 #if defined(TARGET_I386)
1512 X86CPU *x86_cpu = X86_CPU(cpu);
1513 CPUX86State *env = &x86_cpu->env;
1514 #elif defined(TARGET_PPC)
1515 PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu);
1516 CPUPPCState *env = &ppc_cpu->env;
1517 #elif defined(TARGET_SPARC)
1518 SPARCCPU *sparc_cpu = SPARC_CPU(cpu);
1519 CPUSPARCState *env = &sparc_cpu->env;
1520 #elif defined(TARGET_MIPS)
1521 MIPSCPU *mips_cpu = MIPS_CPU(cpu);
1522 CPUMIPSState *env = &mips_cpu->env;
1523 #elif defined(TARGET_TRICORE)
1524 TriCoreCPU *tricore_cpu = TRICORE_CPU(cpu);
1525 CPUTriCoreState *env = &tricore_cpu->env;
1528 cpu_synchronize_state(cpu);
1530 info = g_malloc0(sizeof(*info));
1531 info->value = g_malloc0(sizeof(*info->value));
1532 info->value->CPU = cpu->cpu_index;
1533 info->value->current = (cpu == first_cpu);
1534 info->value->halted = cpu->halted;
1535 info->value->qom_path = object_get_canonical_path(OBJECT(cpu));
1536 info->value->thread_id = cpu->thread_id;
1537 #if defined(TARGET_I386)
1538 info->value->has_pc = true;
1539 info->value->pc = env->eip + env->segs[R_CS].base;
1540 #elif defined(TARGET_PPC)
1541 info->value->has_nip = true;
1542 info->value->nip = env->nip;
1543 #elif defined(TARGET_SPARC)
1544 info->value->has_pc = true;
1545 info->value->pc = env->pc;
1546 info->value->has_npc = true;
1547 info->value->npc = env->npc;
1548 #elif defined(TARGET_MIPS)
1549 info->value->has_PC = true;
1550 info->value->PC = env->active_tc.PC;
1551 #elif defined(TARGET_TRICORE)
1552 info->value->has_PC = true;
1553 info->value->PC = env->PC;
1556 /* XXX: waiting for the qapi to support GSList */
1558 head = cur_item = info;
1560 cur_item->next = info;
1568 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
1569 bool has_cpu, int64_t cpu_index, Error **errp)
1575 int64_t orig_addr = addr, orig_size = size;
1581 cpu = qemu_get_cpu(cpu_index);
1583 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
1588 f = fopen(filename, "wb");
1590 error_setg_file_open(errp, errno, filename);
1598 if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) {
1599 error_setg(errp, "Invalid addr 0x%016" PRIx64 "/size %" PRId64
1600 " specified", orig_addr, orig_size);
1603 if (fwrite(buf, 1, l, f) != l) {
1604 error_setg(errp, QERR_IO_ERROR);
1615 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
1622 f = fopen(filename, "wb");
1624 error_setg_file_open(errp, errno, filename);
1632 cpu_physical_memory_read(addr, buf, l);
1633 if (fwrite(buf, 1, l, f) != l) {
1634 error_setg(errp, QERR_IO_ERROR);
1645 void qmp_inject_nmi(Error **errp)
1647 #if defined(TARGET_I386)
1651 X86CPU *cpu = X86_CPU(cs);
1653 if (!cpu->apic_state) {
1654 cpu_interrupt(cs, CPU_INTERRUPT_NMI);
1656 apic_deliver_nmi(cpu->apic_state);
1660 nmi_monitor_handle(monitor_get_cpu_index(), errp);
1664 void dump_drift_info(FILE *f, fprintf_function cpu_fprintf)
1670 cpu_fprintf(f, "Host - Guest clock %"PRIi64" ms\n",
1671 (cpu_get_clock() - cpu_get_icount())/SCALE_MS);
1672 if (icount_align_option) {
1673 cpu_fprintf(f, "Max guest delay %"PRIi64" ms\n", -max_delay/SCALE_MS);
1674 cpu_fprintf(f, "Max guest advance %"PRIi64" ms\n", max_advance/SCALE_MS);
1676 cpu_fprintf(f, "Max guest delay NA\n");
1677 cpu_fprintf(f, "Max guest advance NA\n");