system/qdev-monitor: move drain_call_rcu call under if (!dev) in qmp_device_add()
[qemu/ar7.git] / system / cpus.c
blob68d161d96b746e1e24e357e4a89b485bfdb09f84
1 /*
2 * QEMU System Emulator
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
22 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "monitor/monitor.h"
27 #include "qemu/coroutine-tls.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/cpu-common.h"
36 #include "qemu/thread.h"
37 #include "qemu/main-loop.h"
38 #include "qemu/plugin.h"
39 #include "sysemu/cpus.h"
40 #include "qemu/guest-random.h"
41 #include "hw/nmi.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"
47 #include "hw/hw.h"
48 #include "trace.h"
50 #ifdef CONFIG_LINUX
52 #include <sys/prctl.h>
54 #ifndef PR_MCE_KILL
55 #define PR_MCE_KILL 33
56 #endif
58 #ifndef PR_MCE_KILL_SET
59 #define PR_MCE_KILL_SET 1
60 #endif
62 #ifndef PR_MCE_KILL_EARLY
63 #define PR_MCE_KILL_EARLY 1
64 #endif
66 #endif /* CONFIG_LINUX */
68 /* The Big QEMU Lock (BQL) */
69 static QemuMutex bql;
72 * The chosen accelerator is supposed to register this.
74 static const AccelOpsClass *cpus_accel;
76 bool cpu_is_stopped(CPUState *cpu)
78 return cpu->stopped || !runstate_is_running();
81 bool cpu_work_list_empty(CPUState *cpu)
83 return QSIMPLEQ_EMPTY_ATOMIC(&cpu->work_list);
86 bool cpu_thread_is_idle(CPUState *cpu)
88 if (cpu->stop || !cpu_work_list_empty(cpu)) {
89 return false;
91 if (cpu_is_stopped(cpu)) {
92 return true;
94 if (!cpu->halted || cpu_has_work(cpu)) {
95 return false;
97 if (cpus_accel->cpu_thread_is_idle) {
98 return cpus_accel->cpu_thread_is_idle(cpu);
100 return true;
103 bool all_cpu_threads_idle(void)
105 CPUState *cpu;
107 CPU_FOREACH(cpu) {
108 if (!cpu_thread_is_idle(cpu)) {
109 return false;
112 return true;
115 /***********************************************************/
116 void hw_error(const char *fmt, ...)
118 va_list ap;
119 CPUState *cpu;
121 va_start(ap, fmt);
122 fprintf(stderr, "qemu: hardware error: ");
123 vfprintf(stderr, fmt, ap);
124 fprintf(stderr, "\n");
125 CPU_FOREACH(cpu) {
126 fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
127 cpu_dump_state(cpu, stderr, CPU_DUMP_FPU);
129 va_end(ap);
130 abort();
133 void cpu_synchronize_all_states(void)
135 CPUState *cpu;
137 CPU_FOREACH(cpu) {
138 cpu_synchronize_state(cpu);
142 void cpu_synchronize_all_post_reset(void)
144 CPUState *cpu;
146 CPU_FOREACH(cpu) {
147 cpu_synchronize_post_reset(cpu);
151 void cpu_synchronize_all_post_init(void)
153 CPUState *cpu;
155 CPU_FOREACH(cpu) {
156 cpu_synchronize_post_init(cpu);
160 void cpu_synchronize_all_pre_loadvm(void)
162 CPUState *cpu;
164 CPU_FOREACH(cpu) {
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 bool cpus_are_resettable(void)
199 if (cpus_accel->cpus_are_resettable) {
200 return cpus_accel->cpus_are_resettable();
202 return true;
205 void cpu_exec_reset_hold(CPUState *cpu)
207 if (cpus_accel->cpu_reset_hold) {
208 cpus_accel->cpu_reset_hold(cpu);
212 int64_t cpus_get_virtual_clock(void)
215 * XXX
217 * need to check that cpus_accel is not NULL, because qcow2 calls
218 * qemu_get_clock_ns(CLOCK_VIRTUAL) without any accel initialized and
219 * with ticks disabled in some io-tests:
220 * 030 040 041 060 099 120 127 140 156 161 172 181 191 192 195 203 229 249 256 267
222 * is this expected?
224 * XXX
226 if (cpus_accel && cpus_accel->get_virtual_clock) {
227 return cpus_accel->get_virtual_clock();
229 return cpu_get_clock();
233 * return the time elapsed in VM between vm_start and vm_stop. Unless
234 * icount is active, cpus_get_elapsed_ticks() uses units of the host CPU cycle
235 * counter.
237 int64_t cpus_get_elapsed_ticks(void)
239 if (cpus_accel->get_elapsed_ticks) {
240 return cpus_accel->get_elapsed_ticks();
242 return cpu_get_ticks();
245 static void generic_handle_interrupt(CPUState *cpu, int mask)
247 cpu->interrupt_request |= mask;
249 if (!qemu_cpu_is_self(cpu)) {
250 qemu_cpu_kick(cpu);
254 void cpu_interrupt(CPUState *cpu, int mask)
256 if (cpus_accel->handle_interrupt) {
257 cpus_accel->handle_interrupt(cpu, mask);
258 } else {
259 generic_handle_interrupt(cpu, mask);
264 * True if the vm was previously suspended, and has not been woken or reset.
266 static int vm_was_suspended;
268 void vm_set_suspended(bool suspended)
270 vm_was_suspended = suspended;
273 bool vm_get_suspended(void)
275 return vm_was_suspended;
278 static int do_vm_stop(RunState state, bool send_stop)
280 int ret = 0;
281 RunState oldstate = runstate_get();
283 if (runstate_is_live(oldstate)) {
284 vm_was_suspended = (oldstate == RUN_STATE_SUSPENDED);
285 runstate_set(state);
286 cpu_disable_ticks();
287 if (oldstate == RUN_STATE_RUNNING) {
288 pause_all_vcpus();
290 vm_state_notify(0, state);
291 if (send_stop) {
292 qapi_event_send_stop();
296 bdrv_drain_all();
297 ret = bdrv_flush_all();
298 trace_vm_stop_flush_all(ret);
300 return ret;
303 /* Special vm_stop() variant for terminating the process. Historically clients
304 * did not expect a QMP STOP event and so we need to retain compatibility.
306 int vm_shutdown(void)
308 return do_vm_stop(RUN_STATE_SHUTDOWN, false);
311 bool cpu_can_run(CPUState *cpu)
313 if (cpu->stop) {
314 return false;
316 if (cpu_is_stopped(cpu)) {
317 return false;
319 return true;
322 void cpu_handle_guest_debug(CPUState *cpu)
324 if (replay_running_debug()) {
325 if (!cpu->singlestep_enabled) {
327 * Report about the breakpoint and
328 * make a single step to skip it
330 replay_breakpoint();
331 cpu_single_step(cpu, SSTEP_ENABLE);
332 } else {
333 cpu_single_step(cpu, 0);
335 } else {
336 gdb_set_stop_cpu(cpu);
337 qemu_system_debug_request();
338 cpu->stopped = true;
342 #ifdef CONFIG_LINUX
343 static void sigbus_reraise(void)
345 sigset_t set;
346 struct sigaction action;
348 memset(&action, 0, sizeof(action));
349 action.sa_handler = SIG_DFL;
350 if (!sigaction(SIGBUS, &action, NULL)) {
351 raise(SIGBUS);
352 sigemptyset(&set);
353 sigaddset(&set, SIGBUS);
354 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
356 perror("Failed to re-raise SIGBUS!");
357 abort();
360 static void sigbus_handler(int n, siginfo_t *siginfo, void *ctx)
362 if (siginfo->si_code != BUS_MCEERR_AO && siginfo->si_code != BUS_MCEERR_AR) {
363 sigbus_reraise();
366 if (current_cpu) {
367 /* Called asynchronously in VCPU thread. */
368 if (kvm_on_sigbus_vcpu(current_cpu, siginfo->si_code, siginfo->si_addr)) {
369 sigbus_reraise();
371 } else {
372 /* Called synchronously (via signalfd) in main thread. */
373 if (kvm_on_sigbus(siginfo->si_code, siginfo->si_addr)) {
374 sigbus_reraise();
379 static void qemu_init_sigbus(void)
381 struct sigaction action;
384 * ALERT: when modifying this, take care that SIGBUS forwarding in
385 * qemu_prealloc_mem() will continue working as expected.
387 memset(&action, 0, sizeof(action));
388 action.sa_flags = SA_SIGINFO;
389 action.sa_sigaction = sigbus_handler;
390 sigaction(SIGBUS, &action, NULL);
392 prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
394 #else /* !CONFIG_LINUX */
395 static void qemu_init_sigbus(void)
398 #endif /* !CONFIG_LINUX */
400 static QemuThread io_thread;
402 /* cpu creation */
403 static QemuCond qemu_cpu_cond;
404 /* system init */
405 static QemuCond qemu_pause_cond;
407 void qemu_init_cpu_loop(void)
409 qemu_init_sigbus();
410 qemu_cond_init(&qemu_cpu_cond);
411 qemu_cond_init(&qemu_pause_cond);
412 qemu_mutex_init(&bql);
414 qemu_thread_get_self(&io_thread);
417 void run_on_cpu(CPUState *cpu, run_on_cpu_func func, run_on_cpu_data data)
419 do_run_on_cpu(cpu, func, data, &bql);
422 static void qemu_cpu_stop(CPUState *cpu, bool exit)
424 g_assert(qemu_cpu_is_self(cpu));
425 cpu->stop = false;
426 cpu->stopped = true;
427 if (exit) {
428 cpu_exit(cpu);
430 qemu_cond_broadcast(&qemu_pause_cond);
433 void qemu_wait_io_event_common(CPUState *cpu)
435 qatomic_set_mb(&cpu->thread_kicked, false);
436 if (cpu->stop) {
437 qemu_cpu_stop(cpu, false);
439 process_queued_cpu_work(cpu);
442 void qemu_wait_io_event(CPUState *cpu)
444 bool slept = false;
446 while (cpu_thread_is_idle(cpu)) {
447 if (!slept) {
448 slept = true;
449 qemu_plugin_vcpu_idle_cb(cpu);
451 qemu_cond_wait(cpu->halt_cond, &bql);
453 if (slept) {
454 qemu_plugin_vcpu_resume_cb(cpu);
457 qemu_wait_io_event_common(cpu);
460 void cpus_kick_thread(CPUState *cpu)
462 if (cpu->thread_kicked) {
463 return;
465 cpu->thread_kicked = true;
467 #ifndef _WIN32
468 int err = pthread_kill(cpu->thread->thread, SIG_IPI);
469 if (err && err != ESRCH) {
470 fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
471 exit(1);
473 #else
474 qemu_sem_post(&cpu->sem);
475 #endif
478 void qemu_cpu_kick(CPUState *cpu)
480 qemu_cond_broadcast(cpu->halt_cond);
481 if (cpus_accel->kick_vcpu_thread) {
482 cpus_accel->kick_vcpu_thread(cpu);
483 } else { /* default */
484 cpus_kick_thread(cpu);
488 void qemu_cpu_kick_self(void)
490 assert(current_cpu);
491 cpus_kick_thread(current_cpu);
494 bool qemu_cpu_is_self(CPUState *cpu)
496 return qemu_thread_is_self(cpu->thread);
499 bool qemu_in_vcpu_thread(void)
501 return current_cpu && qemu_cpu_is_self(current_cpu);
504 QEMU_DEFINE_STATIC_CO_TLS(bool, bql_locked)
506 bool bql_locked(void)
508 return get_bql_locked();
511 bool qemu_in_main_thread(void)
513 return bql_locked();
517 * The BQL is taken from so many places that it is worth profiling the
518 * callers directly, instead of funneling them all through a single function.
520 void bql_lock_impl(const char *file, int line)
522 QemuMutexLockFunc bql_lock_fn = qatomic_read(&bql_mutex_lock_func);
524 g_assert(!bql_locked());
525 bql_lock_fn(&bql, file, line);
526 set_bql_locked(true);
529 void bql_unlock(void)
531 g_assert(bql_locked());
532 set_bql_locked(false);
533 qemu_mutex_unlock(&bql);
536 void qemu_cond_wait_bql(QemuCond *cond)
538 qemu_cond_wait(cond, &bql);
541 void qemu_cond_timedwait_bql(QemuCond *cond, int ms)
543 qemu_cond_timedwait(cond, &bql, ms);
546 /* signal CPU creation */
547 void cpu_thread_signal_created(CPUState *cpu)
549 cpu->created = true;
550 qemu_cond_signal(&qemu_cpu_cond);
553 /* signal CPU destruction */
554 void cpu_thread_signal_destroyed(CPUState *cpu)
556 cpu->created = false;
557 qemu_cond_signal(&qemu_cpu_cond);
561 static bool all_vcpus_paused(void)
563 CPUState *cpu;
565 CPU_FOREACH(cpu) {
566 if (!cpu->stopped) {
567 return false;
571 return true;
574 void pause_all_vcpus(void)
576 CPUState *cpu;
578 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false);
579 CPU_FOREACH(cpu) {
580 if (qemu_cpu_is_self(cpu)) {
581 qemu_cpu_stop(cpu, true);
582 } else {
583 cpu->stop = true;
584 qemu_cpu_kick(cpu);
588 /* We need to drop the replay_lock so any vCPU threads woken up
589 * can finish their replay tasks
591 replay_mutex_unlock();
593 while (!all_vcpus_paused()) {
594 qemu_cond_wait(&qemu_pause_cond, &bql);
595 CPU_FOREACH(cpu) {
596 qemu_cpu_kick(cpu);
600 bql_unlock();
601 replay_mutex_lock();
602 bql_lock();
605 void cpu_resume(CPUState *cpu)
607 cpu->stop = false;
608 cpu->stopped = false;
609 qemu_cpu_kick(cpu);
612 void resume_all_vcpus(void)
614 CPUState *cpu;
616 if (!runstate_is_running()) {
617 return;
620 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true);
621 CPU_FOREACH(cpu) {
622 cpu_resume(cpu);
626 void cpu_remove_sync(CPUState *cpu)
628 cpu->stop = true;
629 cpu->unplug = true;
630 qemu_cpu_kick(cpu);
631 bql_unlock();
632 qemu_thread_join(cpu->thread);
633 bql_lock();
636 void cpus_register_accel(const AccelOpsClass *ops)
638 assert(ops != NULL);
639 assert(ops->create_vcpu_thread != NULL); /* mandatory */
640 cpus_accel = ops;
643 const AccelOpsClass *cpus_get_accel(void)
645 /* broken if we call this early */
646 assert(cpus_accel);
647 return cpus_accel;
650 void qemu_init_vcpu(CPUState *cpu)
652 MachineState *ms = MACHINE(qdev_get_machine());
654 cpu->nr_cores = machine_topo_get_cores_per_socket(ms);
655 cpu->nr_threads = ms->smp.threads;
656 cpu->stopped = true;
657 cpu->random_seed = qemu_guest_random_seed_thread_part1();
659 if (!cpu->as) {
660 /* If the target cpu hasn't set up any address spaces itself,
661 * give it the default one.
663 cpu->num_ases = 1;
664 cpu_address_space_init(cpu, 0, "cpu-memory", cpu->memory);
667 /* accelerators all implement the AccelOpsClass */
668 g_assert(cpus_accel != NULL && cpus_accel->create_vcpu_thread != NULL);
669 cpus_accel->create_vcpu_thread(cpu);
671 while (!cpu->created) {
672 qemu_cond_wait(&qemu_cpu_cond, &bql);
676 void cpu_stop_current(void)
678 if (current_cpu) {
679 current_cpu->stop = true;
680 cpu_exit(current_cpu);
684 int vm_stop(RunState state)
686 if (qemu_in_vcpu_thread()) {
687 qemu_system_vmstop_request_prepare();
688 qemu_system_vmstop_request(state);
690 * FIXME: should not return to device code in case
691 * vm_stop() has been requested.
693 cpu_stop_current();
694 return 0;
697 return do_vm_stop(state, true);
701 * Prepare for (re)starting the VM.
702 * Returns 0 if the vCPUs should be restarted, -1 on an error condition,
703 * and 1 otherwise.
705 int vm_prepare_start(bool step_pending)
707 int ret = vm_was_suspended ? 1 : 0;
708 RunState state = vm_was_suspended ? RUN_STATE_SUSPENDED : RUN_STATE_RUNNING;
709 RunState requested;
711 qemu_vmstop_requested(&requested);
712 if (runstate_is_running() && requested == RUN_STATE__MAX) {
713 return -1;
716 /* Ensure that a STOP/RESUME pair of events is emitted if a
717 * vmstop request was pending. The BLOCK_IO_ERROR event, for
718 * example, according to documentation is always followed by
719 * the STOP event.
721 if (runstate_is_running()) {
722 qapi_event_send_stop();
723 qapi_event_send_resume();
724 return -1;
728 * WHPX accelerator needs to know whether we are going to step
729 * any CPUs, before starting the first one.
731 if (cpus_accel->synchronize_pre_resume) {
732 cpus_accel->synchronize_pre_resume(step_pending);
735 /* We are sending this now, but the CPUs will be resumed shortly later */
736 qapi_event_send_resume();
738 cpu_enable_ticks();
739 runstate_set(state);
740 vm_state_notify(1, state);
741 vm_was_suspended = false;
742 return ret;
745 void vm_start(void)
747 if (!vm_prepare_start(false)) {
748 resume_all_vcpus();
752 void vm_resume(RunState state)
754 if (runstate_is_live(state)) {
755 vm_start();
756 } else {
757 runstate_set(state);
761 /* does a state transition even if the VM is already stopped,
762 current state is forgotten forever */
763 int vm_stop_force_state(RunState state)
765 if (runstate_is_live(runstate_get())) {
766 return vm_stop(state);
767 } else {
768 int ret;
769 runstate_set(state);
771 bdrv_drain_all();
772 /* Make sure to return an error if the flush in a previous vm_stop()
773 * failed. */
774 ret = bdrv_flush_all();
775 trace_vm_stop_flush_all(ret);
776 return ret;
780 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
781 bool has_cpu, int64_t cpu_index, Error **errp)
783 FILE *f;
784 uint32_t l;
785 CPUState *cpu;
786 uint8_t buf[1024];
787 int64_t orig_addr = addr, orig_size = size;
789 if (!has_cpu) {
790 cpu_index = 0;
793 cpu = qemu_get_cpu(cpu_index);
794 if (cpu == NULL) {
795 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
796 "a CPU number");
797 return;
800 f = fopen(filename, "wb");
801 if (!f) {
802 error_setg_file_open(errp, errno, filename);
803 return;
806 while (size != 0) {
807 l = sizeof(buf);
808 if (l > size)
809 l = size;
810 if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) {
811 error_setg(errp, "Invalid addr 0x%016" PRIx64 "/size %" PRId64
812 " specified", orig_addr, orig_size);
813 goto exit;
815 if (fwrite(buf, 1, l, f) != l) {
816 error_setg(errp, QERR_IO_ERROR);
817 goto exit;
819 addr += l;
820 size -= l;
823 exit:
824 fclose(f);
827 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
828 Error **errp)
830 FILE *f;
831 uint32_t l;
832 uint8_t buf[1024];
834 f = fopen(filename, "wb");
835 if (!f) {
836 error_setg_file_open(errp, errno, filename);
837 return;
840 while (size != 0) {
841 l = sizeof(buf);
842 if (l > size)
843 l = size;
844 cpu_physical_memory_read(addr, buf, l);
845 if (fwrite(buf, 1, l, f) != l) {
846 error_setg(errp, QERR_IO_ERROR);
847 goto exit;
849 addr += l;
850 size -= l;
853 exit:
854 fclose(f);
857 void qmp_inject_nmi(Error **errp)
859 nmi_monitor_handle(monitor_get_cpu_index(monitor_cur()), errp);