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 "qapi/error.h"
29 #include "qapi/qapi-commands-machine.h"
30 #include "qapi/qapi-commands-misc.h"
31 #include "qapi/qapi-events-run-state.h"
32 #include "qapi/qmp/qerror.h"
33 #include "exec/gdbstub.h"
34 #include "sysemu/hw_accel.h"
35 #include "exec/exec-all.h"
36 #include "qemu/thread.h"
37 #include "qemu/plugin.h"
38 #include "sysemu/cpus.h"
39 #include "qemu/guest-random.h"
41 #include "sysemu/replay.h"
42 #include "sysemu/runstate.h"
43 #include "sysemu/cpu-timers.h"
44 #include "sysemu/whpx.h"
45 #include "hw/boards.h"
50 #include <sys/prctl.h>
53 #define PR_MCE_KILL 33
56 #ifndef PR_MCE_KILL_SET
57 #define PR_MCE_KILL_SET 1
60 #ifndef PR_MCE_KILL_EARLY
61 #define PR_MCE_KILL_EARLY 1
64 #endif /* CONFIG_LINUX */
66 static QemuMutex qemu_global_mutex
;
68 bool cpu_is_stopped(CPUState
*cpu
)
70 return cpu
->stopped
|| !runstate_is_running();
73 bool cpu_work_list_empty(CPUState
*cpu
)
77 qemu_mutex_lock(&cpu
->work_mutex
);
78 ret
= QSIMPLEQ_EMPTY(&cpu
->work_list
);
79 qemu_mutex_unlock(&cpu
->work_mutex
);
83 bool cpu_thread_is_idle(CPUState
*cpu
)
85 if (cpu
->stop
|| !cpu_work_list_empty(cpu
)) {
88 if (cpu_is_stopped(cpu
)) {
91 if (!cpu
->halted
|| cpu_has_work(cpu
) ||
92 kvm_halt_in_kernel() || whpx_apic_in_platform()) {
98 bool all_cpu_threads_idle(void)
103 if (!cpu_thread_is_idle(cpu
)) {
110 /***********************************************************/
111 void hw_error(const char *fmt
, ...)
117 fprintf(stderr
, "qemu: hardware error: ");
118 vfprintf(stderr
, fmt
, ap
);
119 fprintf(stderr
, "\n");
121 fprintf(stderr
, "CPU #%d:\n", cpu
->cpu_index
);
122 cpu_dump_state(cpu
, stderr
, CPU_DUMP_FPU
);
129 * The chosen accelerator is supposed to register this.
131 static const AccelOpsClass
*cpus_accel
;
133 void cpu_synchronize_all_states(void)
138 cpu_synchronize_state(cpu
);
142 void cpu_synchronize_all_post_reset(void)
147 cpu_synchronize_post_reset(cpu
);
151 void cpu_synchronize_all_post_init(void)
156 cpu_synchronize_post_init(cpu
);
160 void cpu_synchronize_all_pre_loadvm(void)
165 cpu_synchronize_pre_loadvm(cpu
);
169 void cpu_synchronize_state(CPUState
*cpu
)
171 if (cpus_accel
->synchronize_state
) {
172 cpus_accel
->synchronize_state(cpu
);
176 void cpu_synchronize_post_reset(CPUState
*cpu
)
178 if (cpus_accel
->synchronize_post_reset
) {
179 cpus_accel
->synchronize_post_reset(cpu
);
183 void cpu_synchronize_post_init(CPUState
*cpu
)
185 if (cpus_accel
->synchronize_post_init
) {
186 cpus_accel
->synchronize_post_init(cpu
);
190 void cpu_synchronize_pre_loadvm(CPUState
*cpu
)
192 if (cpus_accel
->synchronize_pre_loadvm
) {
193 cpus_accel
->synchronize_pre_loadvm(cpu
);
197 int64_t cpus_get_virtual_clock(void)
202 * need to check that cpus_accel is not NULL, because qcow2 calls
203 * qemu_get_clock_ns(CLOCK_VIRTUAL) without any accel initialized and
204 * with ticks disabled in some io-tests:
205 * 030 040 041 060 099 120 127 140 156 161 172 181 191 192 195 203 229 249 256 267
211 if (cpus_accel
&& cpus_accel
->get_virtual_clock
) {
212 return cpus_accel
->get_virtual_clock();
214 return cpu_get_clock();
218 * return the time elapsed in VM between vm_start and vm_stop. Unless
219 * icount is active, cpus_get_elapsed_ticks() uses units of the host CPU cycle
222 int64_t cpus_get_elapsed_ticks(void)
224 if (cpus_accel
->get_elapsed_ticks
) {
225 return cpus_accel
->get_elapsed_ticks();
227 return cpu_get_ticks();
230 static void generic_handle_interrupt(CPUState
*cpu
, int mask
)
232 cpu
->interrupt_request
|= mask
;
234 if (!qemu_cpu_is_self(cpu
)) {
239 void cpu_interrupt(CPUState
*cpu
, int mask
)
241 if (cpus_accel
->handle_interrupt
) {
242 cpus_accel
->handle_interrupt(cpu
, mask
);
244 generic_handle_interrupt(cpu
, mask
);
248 static int do_vm_stop(RunState state
, bool send_stop
)
252 if (runstate_is_running()) {
256 vm_state_notify(0, state
);
258 qapi_event_send_stop();
263 ret
= bdrv_flush_all();
268 /* Special vm_stop() variant for terminating the process. Historically clients
269 * did not expect a QMP STOP event and so we need to retain compatibility.
271 int vm_shutdown(void)
273 return do_vm_stop(RUN_STATE_SHUTDOWN
, false);
276 bool cpu_can_run(CPUState
*cpu
)
281 if (cpu_is_stopped(cpu
)) {
287 void cpu_handle_guest_debug(CPUState
*cpu
)
289 if (replay_running_debug()) {
290 if (!cpu
->singlestep_enabled
) {
292 * Report about the breakpoint and
293 * make a single step to skip it
296 cpu_single_step(cpu
, SSTEP_ENABLE
);
298 cpu_single_step(cpu
, 0);
301 gdb_set_stop_cpu(cpu
);
302 qemu_system_debug_request();
308 static void sigbus_reraise(void)
311 struct sigaction action
;
313 memset(&action
, 0, sizeof(action
));
314 action
.sa_handler
= SIG_DFL
;
315 if (!sigaction(SIGBUS
, &action
, NULL
)) {
318 sigaddset(&set
, SIGBUS
);
319 pthread_sigmask(SIG_UNBLOCK
, &set
, NULL
);
321 perror("Failed to re-raise SIGBUS!\n");
325 static void sigbus_handler(int n
, siginfo_t
*siginfo
, void *ctx
)
327 if (siginfo
->si_code
!= BUS_MCEERR_AO
&& siginfo
->si_code
!= BUS_MCEERR_AR
) {
332 /* Called asynchronously in VCPU thread. */
333 if (kvm_on_sigbus_vcpu(current_cpu
, siginfo
->si_code
, siginfo
->si_addr
)) {
337 /* Called synchronously (via signalfd) in main thread. */
338 if (kvm_on_sigbus(siginfo
->si_code
, siginfo
->si_addr
)) {
344 static void qemu_init_sigbus(void)
346 struct sigaction action
;
348 memset(&action
, 0, sizeof(action
));
349 action
.sa_flags
= SA_SIGINFO
;
350 action
.sa_sigaction
= sigbus_handler
;
351 sigaction(SIGBUS
, &action
, NULL
);
353 prctl(PR_MCE_KILL
, PR_MCE_KILL_SET
, PR_MCE_KILL_EARLY
, 0, 0);
355 #else /* !CONFIG_LINUX */
356 static void qemu_init_sigbus(void)
359 #endif /* !CONFIG_LINUX */
361 static QemuThread io_thread
;
364 static QemuCond qemu_cpu_cond
;
366 static QemuCond qemu_pause_cond
;
368 void qemu_init_cpu_loop(void)
371 qemu_cond_init(&qemu_cpu_cond
);
372 qemu_cond_init(&qemu_pause_cond
);
373 qemu_mutex_init(&qemu_global_mutex
);
375 qemu_thread_get_self(&io_thread
);
378 void run_on_cpu(CPUState
*cpu
, run_on_cpu_func func
, run_on_cpu_data data
)
380 do_run_on_cpu(cpu
, func
, data
, &qemu_global_mutex
);
383 static void qemu_cpu_stop(CPUState
*cpu
, bool exit
)
385 g_assert(qemu_cpu_is_self(cpu
));
391 qemu_cond_broadcast(&qemu_pause_cond
);
394 void qemu_wait_io_event_common(CPUState
*cpu
)
396 qatomic_mb_set(&cpu
->thread_kicked
, false);
398 qemu_cpu_stop(cpu
, false);
400 process_queued_cpu_work(cpu
);
403 void qemu_wait_io_event(CPUState
*cpu
)
407 while (cpu_thread_is_idle(cpu
)) {
410 qemu_plugin_vcpu_idle_cb(cpu
);
412 qemu_cond_wait(cpu
->halt_cond
, &qemu_global_mutex
);
415 qemu_plugin_vcpu_resume_cb(cpu
);
419 /* Eat dummy APC queued by cpus_kick_thread. */
424 qemu_wait_io_event_common(cpu
);
427 void cpus_kick_thread(CPUState
*cpu
)
432 if (cpu
->thread_kicked
) {
435 cpu
->thread_kicked
= true;
436 err
= pthread_kill(cpu
->thread
->thread
, SIG_IPI
);
437 if (err
&& err
!= ESRCH
) {
438 fprintf(stderr
, "qemu:%s: %s", __func__
, strerror(err
));
444 void qemu_cpu_kick(CPUState
*cpu
)
446 qemu_cond_broadcast(cpu
->halt_cond
);
447 if (cpus_accel
->kick_vcpu_thread
) {
448 cpus_accel
->kick_vcpu_thread(cpu
);
449 } else { /* default */
450 cpus_kick_thread(cpu
);
454 void qemu_cpu_kick_self(void)
457 cpus_kick_thread(current_cpu
);
460 bool qemu_cpu_is_self(CPUState
*cpu
)
462 return qemu_thread_is_self(cpu
->thread
);
465 bool qemu_in_vcpu_thread(void)
467 return current_cpu
&& qemu_cpu_is_self(current_cpu
);
470 static __thread
bool iothread_locked
= false;
472 bool qemu_mutex_iothread_locked(void)
474 return iothread_locked
;
478 * The BQL is taken from so many places that it is worth profiling the
479 * callers directly, instead of funneling them all through a single function.
481 void qemu_mutex_lock_iothread_impl(const char *file
, int line
)
483 QemuMutexLockFunc bql_lock
= qatomic_read(&qemu_bql_mutex_lock_func
);
485 g_assert(!qemu_mutex_iothread_locked());
486 bql_lock(&qemu_global_mutex
, file
, line
);
487 iothread_locked
= true;
490 void qemu_mutex_unlock_iothread(void)
492 g_assert(qemu_mutex_iothread_locked());
493 iothread_locked
= false;
494 qemu_mutex_unlock(&qemu_global_mutex
);
497 void qemu_cond_wait_iothread(QemuCond
*cond
)
499 qemu_cond_wait(cond
, &qemu_global_mutex
);
502 void qemu_cond_timedwait_iothread(QemuCond
*cond
, int ms
)
504 qemu_cond_timedwait(cond
, &qemu_global_mutex
, ms
);
507 /* signal CPU creation */
508 void cpu_thread_signal_created(CPUState
*cpu
)
511 qemu_cond_signal(&qemu_cpu_cond
);
514 /* signal CPU destruction */
515 void cpu_thread_signal_destroyed(CPUState
*cpu
)
517 cpu
->created
= false;
518 qemu_cond_signal(&qemu_cpu_cond
);
522 static bool all_vcpus_paused(void)
535 void pause_all_vcpus(void)
539 qemu_clock_enable(QEMU_CLOCK_VIRTUAL
, false);
541 if (qemu_cpu_is_self(cpu
)) {
542 qemu_cpu_stop(cpu
, true);
549 /* We need to drop the replay_lock so any vCPU threads woken up
550 * can finish their replay tasks
552 replay_mutex_unlock();
554 while (!all_vcpus_paused()) {
555 qemu_cond_wait(&qemu_pause_cond
, &qemu_global_mutex
);
561 qemu_mutex_unlock_iothread();
563 qemu_mutex_lock_iothread();
566 void cpu_resume(CPUState
*cpu
)
569 cpu
->stopped
= false;
573 void resume_all_vcpus(void)
577 if (!runstate_is_running()) {
581 qemu_clock_enable(QEMU_CLOCK_VIRTUAL
, true);
587 void cpu_remove_sync(CPUState
*cpu
)
592 qemu_mutex_unlock_iothread();
593 qemu_thread_join(cpu
->thread
);
594 qemu_mutex_lock_iothread();
597 void cpus_register_accel(const AccelOpsClass
*ops
)
600 assert(ops
->create_vcpu_thread
!= NULL
); /* mandatory */
604 void qemu_init_vcpu(CPUState
*cpu
)
606 MachineState
*ms
= MACHINE(qdev_get_machine());
608 cpu
->nr_cores
= ms
->smp
.cores
;
609 cpu
->nr_threads
= ms
->smp
.threads
;
611 cpu
->random_seed
= qemu_guest_random_seed_thread_part1();
614 /* If the target cpu hasn't set up any address spaces itself,
615 * give it the default one.
618 cpu_address_space_init(cpu
, 0, "cpu-memory", cpu
->memory
);
621 /* accelerators all implement the AccelOpsClass */
622 g_assert(cpus_accel
!= NULL
&& cpus_accel
->create_vcpu_thread
!= NULL
);
623 cpus_accel
->create_vcpu_thread(cpu
);
625 while (!cpu
->created
) {
626 qemu_cond_wait(&qemu_cpu_cond
, &qemu_global_mutex
);
630 void cpu_stop_current(void)
633 current_cpu
->stop
= true;
634 cpu_exit(current_cpu
);
638 int vm_stop(RunState state
)
640 if (qemu_in_vcpu_thread()) {
641 qemu_system_vmstop_request_prepare();
642 qemu_system_vmstop_request(state
);
644 * FIXME: should not return to device code in case
645 * vm_stop() has been requested.
651 return do_vm_stop(state
, true);
655 * Prepare for (re)starting the VM.
656 * Returns -1 if the vCPUs are not to be restarted (e.g. if they are already
657 * running or in case of an error condition), 0 otherwise.
659 int vm_prepare_start(void)
663 qemu_vmstop_requested(&requested
);
664 if (runstate_is_running() && requested
== RUN_STATE__MAX
) {
668 /* Ensure that a STOP/RESUME pair of events is emitted if a
669 * vmstop request was pending. The BLOCK_IO_ERROR event, for
670 * example, according to documentation is always followed by
673 if (runstate_is_running()) {
674 qapi_event_send_stop();
675 qapi_event_send_resume();
679 /* We are sending this now, but the CPUs will be resumed shortly later */
680 qapi_event_send_resume();
683 runstate_set(RUN_STATE_RUNNING
);
684 vm_state_notify(1, RUN_STATE_RUNNING
);
690 if (!vm_prepare_start()) {
695 /* does a state transition even if the VM is already stopped,
696 current state is forgotten forever */
697 int vm_stop_force_state(RunState state
)
699 if (runstate_is_running()) {
700 return vm_stop(state
);
705 /* Make sure to return an error if the flush in a previous vm_stop()
707 return bdrv_flush_all();
711 void list_cpus(const char *optarg
)
713 /* XXX: implement xxx_cpu_list for targets that still miss it */
714 #if defined(cpu_list)
719 void qmp_memsave(int64_t addr
, int64_t size
, const char *filename
,
720 bool has_cpu
, int64_t cpu_index
, Error
**errp
)
726 int64_t orig_addr
= addr
, orig_size
= size
;
732 cpu
= qemu_get_cpu(cpu_index
);
734 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cpu-index",
739 f
= fopen(filename
, "wb");
741 error_setg_file_open(errp
, errno
, filename
);
749 if (cpu_memory_rw_debug(cpu
, addr
, buf
, l
, 0) != 0) {
750 error_setg(errp
, "Invalid addr 0x%016" PRIx64
"/size %" PRId64
751 " specified", orig_addr
, orig_size
);
754 if (fwrite(buf
, 1, l
, f
) != l
) {
755 error_setg(errp
, QERR_IO_ERROR
);
766 void qmp_pmemsave(int64_t addr
, int64_t size
, const char *filename
,
773 f
= fopen(filename
, "wb");
775 error_setg_file_open(errp
, errno
, filename
);
783 cpu_physical_memory_read(addr
, buf
, l
);
784 if (fwrite(buf
, 1, l
, f
) != l
) {
785 error_setg(errp
, QERR_IO_ERROR
);
796 void qmp_inject_nmi(Error
**errp
)
798 nmi_monitor_handle(monitor_get_cpu_index(monitor_cur()), errp
);