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 #include "qemu/osdep.h"
26 #include "qemu-common.h"
27 #include "monitor/monitor.h"
28 #include "qemu/coroutine-tls.h"
29 #include "qapi/error.h"
30 #include "qapi/qapi-commands-machine.h"
31 #include "qapi/qapi-commands-misc.h"
32 #include "qapi/qapi-events-run-state.h"
33 #include "qapi/qmp/qerror.h"
34 #include "exec/gdbstub.h"
35 #include "sysemu/hw_accel.h"
36 #include "exec/exec-all.h"
37 #include "qemu/thread.h"
38 #include "qemu/plugin.h"
39 #include "sysemu/cpus.h"
40 #include "qemu/guest-random.h"
42 #include "sysemu/replay.h"
43 #include "sysemu/runstate.h"
44 #include "sysemu/cpu-timers.h"
45 #include "sysemu/whpx.h"
46 #include "hw/boards.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 QemuMutex qemu_global_mutex
;
70 bool cpu_is_stopped(CPUState
*cpu
)
72 return cpu
->stopped
|| !runstate_is_running();
75 bool cpu_work_list_empty(CPUState
*cpu
)
77 return QSIMPLEQ_EMPTY_ATOMIC(&cpu
->work_list
);
80 bool cpu_thread_is_idle(CPUState
*cpu
)
82 if (cpu
->stop
|| !cpu_work_list_empty(cpu
)) {
85 if (cpu_is_stopped(cpu
)) {
88 if (!cpu
->halted
|| cpu_has_work(cpu
) ||
89 kvm_halt_in_kernel() || whpx_apic_in_platform()) {
95 bool all_cpu_threads_idle(void)
100 if (!cpu_thread_is_idle(cpu
)) {
107 /***********************************************************/
108 void hw_error(const char *fmt
, ...)
114 fprintf(stderr
, "qemu: hardware error: ");
115 vfprintf(stderr
, fmt
, ap
);
116 fprintf(stderr
, "\n");
118 fprintf(stderr
, "CPU #%d:\n", cpu
->cpu_index
);
119 cpu_dump_state(cpu
, stderr
, CPU_DUMP_FPU
);
126 * The chosen accelerator is supposed to register this.
128 static const AccelOpsClass
*cpus_accel
;
130 void cpu_synchronize_all_states(void)
135 cpu_synchronize_state(cpu
);
139 void cpu_synchronize_all_post_reset(void)
144 cpu_synchronize_post_reset(cpu
);
148 void cpu_synchronize_all_post_init(void)
153 cpu_synchronize_post_init(cpu
);
157 void cpu_synchronize_all_pre_loadvm(void)
162 cpu_synchronize_pre_loadvm(cpu
);
166 void cpu_synchronize_state(CPUState
*cpu
)
168 if (cpus_accel
->synchronize_state
) {
169 cpus_accel
->synchronize_state(cpu
);
173 void cpu_synchronize_post_reset(CPUState
*cpu
)
175 if (cpus_accel
->synchronize_post_reset
) {
176 cpus_accel
->synchronize_post_reset(cpu
);
180 void cpu_synchronize_post_init(CPUState
*cpu
)
182 if (cpus_accel
->synchronize_post_init
) {
183 cpus_accel
->synchronize_post_init(cpu
);
187 void cpu_synchronize_pre_loadvm(CPUState
*cpu
)
189 if (cpus_accel
->synchronize_pre_loadvm
) {
190 cpus_accel
->synchronize_pre_loadvm(cpu
);
194 bool cpus_are_resettable(void)
196 return cpu_check_are_resettable();
199 int64_t cpus_get_virtual_clock(void)
204 * need to check that cpus_accel is not NULL, because qcow2 calls
205 * qemu_get_clock_ns(CLOCK_VIRTUAL) without any accel initialized and
206 * with ticks disabled in some io-tests:
207 * 030 040 041 060 099 120 127 140 156 161 172 181 191 192 195 203 229 249 256 267
213 if (cpus_accel
&& cpus_accel
->get_virtual_clock
) {
214 return cpus_accel
->get_virtual_clock();
216 return cpu_get_clock();
220 * return the time elapsed in VM between vm_start and vm_stop. Unless
221 * icount is active, cpus_get_elapsed_ticks() uses units of the host CPU cycle
224 int64_t cpus_get_elapsed_ticks(void)
226 if (cpus_accel
->get_elapsed_ticks
) {
227 return cpus_accel
->get_elapsed_ticks();
229 return cpu_get_ticks();
232 static void generic_handle_interrupt(CPUState
*cpu
, int mask
)
234 cpu
->interrupt_request
|= mask
;
236 if (!qemu_cpu_is_self(cpu
)) {
241 void cpu_interrupt(CPUState
*cpu
, int mask
)
243 if (cpus_accel
->handle_interrupt
) {
244 cpus_accel
->handle_interrupt(cpu
, mask
);
246 generic_handle_interrupt(cpu
, mask
);
250 static int do_vm_stop(RunState state
, bool send_stop
)
254 if (runstate_is_running()) {
258 vm_state_notify(0, state
);
260 qapi_event_send_stop();
265 ret
= bdrv_flush_all();
266 trace_vm_stop_flush_all(ret
);
271 /* Special vm_stop() variant for terminating the process. Historically clients
272 * did not expect a QMP STOP event and so we need to retain compatibility.
274 int vm_shutdown(void)
276 return do_vm_stop(RUN_STATE_SHUTDOWN
, false);
279 bool cpu_can_run(CPUState
*cpu
)
284 if (cpu_is_stopped(cpu
)) {
290 void cpu_handle_guest_debug(CPUState
*cpu
)
292 if (replay_running_debug()) {
293 if (!cpu
->singlestep_enabled
) {
295 * Report about the breakpoint and
296 * make a single step to skip it
299 cpu_single_step(cpu
, SSTEP_ENABLE
);
301 cpu_single_step(cpu
, 0);
304 gdb_set_stop_cpu(cpu
);
305 qemu_system_debug_request();
311 static void sigbus_reraise(void)
314 struct sigaction action
;
316 memset(&action
, 0, sizeof(action
));
317 action
.sa_handler
= SIG_DFL
;
318 if (!sigaction(SIGBUS
, &action
, NULL
)) {
321 sigaddset(&set
, SIGBUS
);
322 pthread_sigmask(SIG_UNBLOCK
, &set
, NULL
);
324 perror("Failed to re-raise SIGBUS!");
328 static void sigbus_handler(int n
, siginfo_t
*siginfo
, void *ctx
)
330 if (siginfo
->si_code
!= BUS_MCEERR_AO
&& siginfo
->si_code
!= BUS_MCEERR_AR
) {
335 /* Called asynchronously in VCPU thread. */
336 if (kvm_on_sigbus_vcpu(current_cpu
, siginfo
->si_code
, siginfo
->si_addr
)) {
340 /* Called synchronously (via signalfd) in main thread. */
341 if (kvm_on_sigbus(siginfo
->si_code
, siginfo
->si_addr
)) {
347 static void qemu_init_sigbus(void)
349 struct sigaction action
;
352 * ALERT: when modifying this, take care that SIGBUS forwarding in
353 * os_mem_prealloc() will continue working as expected.
355 memset(&action
, 0, sizeof(action
));
356 action
.sa_flags
= SA_SIGINFO
;
357 action
.sa_sigaction
= sigbus_handler
;
358 sigaction(SIGBUS
, &action
, NULL
);
360 prctl(PR_MCE_KILL
, PR_MCE_KILL_SET
, PR_MCE_KILL_EARLY
, 0, 0);
362 #else /* !CONFIG_LINUX */
363 static void qemu_init_sigbus(void)
366 #endif /* !CONFIG_LINUX */
368 static QemuThread io_thread
;
371 static QemuCond qemu_cpu_cond
;
373 static QemuCond qemu_pause_cond
;
375 void qemu_init_cpu_loop(void)
378 qemu_cond_init(&qemu_cpu_cond
);
379 qemu_cond_init(&qemu_pause_cond
);
380 qemu_mutex_init(&qemu_global_mutex
);
382 qemu_thread_get_self(&io_thread
);
385 void run_on_cpu(CPUState
*cpu
, run_on_cpu_func func
, run_on_cpu_data data
)
387 do_run_on_cpu(cpu
, func
, data
, &qemu_global_mutex
);
390 static void qemu_cpu_stop(CPUState
*cpu
, bool exit
)
392 g_assert(qemu_cpu_is_self(cpu
));
398 qemu_cond_broadcast(&qemu_pause_cond
);
401 void qemu_wait_io_event_common(CPUState
*cpu
)
403 qatomic_mb_set(&cpu
->thread_kicked
, false);
405 qemu_cpu_stop(cpu
, false);
407 process_queued_cpu_work(cpu
);
410 void qemu_wait_io_event(CPUState
*cpu
)
414 while (cpu_thread_is_idle(cpu
)) {
417 qemu_plugin_vcpu_idle_cb(cpu
);
419 qemu_cond_wait(cpu
->halt_cond
, &qemu_global_mutex
);
422 qemu_plugin_vcpu_resume_cb(cpu
);
426 /* Eat dummy APC queued by cpus_kick_thread. */
431 qemu_wait_io_event_common(cpu
);
434 void cpus_kick_thread(CPUState
*cpu
)
439 if (cpu
->thread_kicked
) {
442 cpu
->thread_kicked
= true;
443 err
= pthread_kill(cpu
->thread
->thread
, SIG_IPI
);
444 if (err
&& err
!= ESRCH
) {
445 fprintf(stderr
, "qemu:%s: %s", __func__
, strerror(err
));
451 void qemu_cpu_kick(CPUState
*cpu
)
453 qemu_cond_broadcast(cpu
->halt_cond
);
454 if (cpus_accel
->kick_vcpu_thread
) {
455 cpus_accel
->kick_vcpu_thread(cpu
);
456 } else { /* default */
457 cpus_kick_thread(cpu
);
461 void qemu_cpu_kick_self(void)
464 cpus_kick_thread(current_cpu
);
467 bool qemu_cpu_is_self(CPUState
*cpu
)
469 return qemu_thread_is_self(cpu
->thread
);
472 bool qemu_in_vcpu_thread(void)
474 return current_cpu
&& qemu_cpu_is_self(current_cpu
);
477 QEMU_DEFINE_STATIC_CO_TLS(bool, iothread_locked
)
479 bool qemu_mutex_iothread_locked(void)
481 return get_iothread_locked();
485 * The BQL is taken from so many places that it is worth profiling the
486 * callers directly, instead of funneling them all through a single function.
488 void qemu_mutex_lock_iothread_impl(const char *file
, int line
)
490 QemuMutexLockFunc bql_lock
= qatomic_read(&qemu_bql_mutex_lock_func
);
492 g_assert(!qemu_mutex_iothread_locked());
493 bql_lock(&qemu_global_mutex
, file
, line
);
494 set_iothread_locked(true);
497 void qemu_mutex_unlock_iothread(void)
499 g_assert(qemu_mutex_iothread_locked());
500 set_iothread_locked(false);
501 qemu_mutex_unlock(&qemu_global_mutex
);
504 void qemu_cond_wait_iothread(QemuCond
*cond
)
506 qemu_cond_wait(cond
, &qemu_global_mutex
);
509 void qemu_cond_timedwait_iothread(QemuCond
*cond
, int ms
)
511 qemu_cond_timedwait(cond
, &qemu_global_mutex
, ms
);
514 /* signal CPU creation */
515 void cpu_thread_signal_created(CPUState
*cpu
)
518 qemu_cond_signal(&qemu_cpu_cond
);
521 /* signal CPU destruction */
522 void cpu_thread_signal_destroyed(CPUState
*cpu
)
524 cpu
->created
= false;
525 qemu_cond_signal(&qemu_cpu_cond
);
529 static bool all_vcpus_paused(void)
542 void pause_all_vcpus(void)
546 qemu_clock_enable(QEMU_CLOCK_VIRTUAL
, false);
548 if (qemu_cpu_is_self(cpu
)) {
549 qemu_cpu_stop(cpu
, true);
556 /* We need to drop the replay_lock so any vCPU threads woken up
557 * can finish their replay tasks
559 replay_mutex_unlock();
561 while (!all_vcpus_paused()) {
562 qemu_cond_wait(&qemu_pause_cond
, &qemu_global_mutex
);
568 qemu_mutex_unlock_iothread();
570 qemu_mutex_lock_iothread();
573 void cpu_resume(CPUState
*cpu
)
576 cpu
->stopped
= false;
580 void resume_all_vcpus(void)
584 if (!runstate_is_running()) {
588 qemu_clock_enable(QEMU_CLOCK_VIRTUAL
, true);
594 void cpu_remove_sync(CPUState
*cpu
)
599 qemu_mutex_unlock_iothread();
600 qemu_thread_join(cpu
->thread
);
601 qemu_mutex_lock_iothread();
604 void cpus_register_accel(const AccelOpsClass
*ops
)
607 assert(ops
->create_vcpu_thread
!= NULL
); /* mandatory */
611 void qemu_init_vcpu(CPUState
*cpu
)
613 MachineState
*ms
= MACHINE(qdev_get_machine());
615 cpu
->nr_cores
= ms
->smp
.cores
;
616 cpu
->nr_threads
= ms
->smp
.threads
;
618 cpu
->random_seed
= qemu_guest_random_seed_thread_part1();
621 /* If the target cpu hasn't set up any address spaces itself,
622 * give it the default one.
625 cpu_address_space_init(cpu
, 0, "cpu-memory", cpu
->memory
);
628 /* accelerators all implement the AccelOpsClass */
629 g_assert(cpus_accel
!= NULL
&& cpus_accel
->create_vcpu_thread
!= NULL
);
630 cpus_accel
->create_vcpu_thread(cpu
);
632 while (!cpu
->created
) {
633 qemu_cond_wait(&qemu_cpu_cond
, &qemu_global_mutex
);
637 void cpu_stop_current(void)
640 current_cpu
->stop
= true;
641 cpu_exit(current_cpu
);
645 int vm_stop(RunState state
)
647 if (qemu_in_vcpu_thread()) {
648 qemu_system_vmstop_request_prepare();
649 qemu_system_vmstop_request(state
);
651 * FIXME: should not return to device code in case
652 * vm_stop() has been requested.
658 return do_vm_stop(state
, true);
662 * Prepare for (re)starting the VM.
663 * Returns -1 if the vCPUs are not to be restarted (e.g. if they are already
664 * running or in case of an error condition), 0 otherwise.
666 int vm_prepare_start(void)
670 qemu_vmstop_requested(&requested
);
671 if (runstate_is_running() && requested
== RUN_STATE__MAX
) {
675 /* Ensure that a STOP/RESUME pair of events is emitted if a
676 * vmstop request was pending. The BLOCK_IO_ERROR event, for
677 * example, according to documentation is always followed by
680 if (runstate_is_running()) {
681 qapi_event_send_stop();
682 qapi_event_send_resume();
686 /* We are sending this now, but the CPUs will be resumed shortly later */
687 qapi_event_send_resume();
690 runstate_set(RUN_STATE_RUNNING
);
691 vm_state_notify(1, RUN_STATE_RUNNING
);
697 if (!vm_prepare_start()) {
702 /* does a state transition even if the VM is already stopped,
703 current state is forgotten forever */
704 int vm_stop_force_state(RunState state
)
706 if (runstate_is_running()) {
707 return vm_stop(state
);
713 /* Make sure to return an error if the flush in a previous vm_stop()
715 ret
= bdrv_flush_all();
716 trace_vm_stop_flush_all(ret
);
721 void list_cpus(const char *optarg
)
723 /* XXX: implement xxx_cpu_list for targets that still miss it */
724 #if defined(cpu_list)
729 void qmp_memsave(int64_t addr
, int64_t size
, const char *filename
,
730 bool has_cpu
, int64_t cpu_index
, Error
**errp
)
736 int64_t orig_addr
= addr
, orig_size
= size
;
742 cpu
= qemu_get_cpu(cpu_index
);
744 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cpu-index",
749 f
= fopen(filename
, "wb");
751 error_setg_file_open(errp
, errno
, filename
);
759 if (cpu_memory_rw_debug(cpu
, addr
, buf
, l
, 0) != 0) {
760 error_setg(errp
, "Invalid addr 0x%016" PRIx64
"/size %" PRId64
761 " specified", orig_addr
, orig_size
);
764 if (fwrite(buf
, 1, l
, f
) != l
) {
765 error_setg(errp
, QERR_IO_ERROR
);
776 void qmp_pmemsave(int64_t addr
, int64_t size
, const char *filename
,
783 f
= fopen(filename
, "wb");
785 error_setg_file_open(errp
, errno
, filename
);
793 cpu_physical_memory_read(addr
, buf
, l
);
794 if (fwrite(buf
, 1, l
, f
) != l
) {
795 error_setg(errp
, QERR_IO_ERROR
);
806 void qmp_inject_nmi(Error
**errp
)
808 nmi_monitor_handle(monitor_get_cpu_index(monitor_cur()), errp
);