kvm: dirty-ring: Fix race with vcpu creation
[qemu/ar7.git] / softmmu / cpus.c
blob9cbc8172b5f20d6c55acdc852020143ad8d80781
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 static QemuMutex qemu_global_mutex;
71 * The chosen accelerator is supposed to register this.
73 static const AccelOpsClass *cpus_accel;
75 bool cpu_is_stopped(CPUState *cpu)
77 return cpu->stopped || !runstate_is_running();
80 bool cpu_work_list_empty(CPUState *cpu)
82 return QSIMPLEQ_EMPTY_ATOMIC(&cpu->work_list);
85 bool cpu_thread_is_idle(CPUState *cpu)
87 if (cpu->stop || !cpu_work_list_empty(cpu)) {
88 return false;
90 if (cpu_is_stopped(cpu)) {
91 return true;
93 if (!cpu->halted || cpu_has_work(cpu)) {
94 return false;
96 if (cpus_accel->cpu_thread_is_idle) {
97 return cpus_accel->cpu_thread_is_idle(cpu);
99 return true;
102 bool all_cpu_threads_idle(void)
104 CPUState *cpu;
106 CPU_FOREACH(cpu) {
107 if (!cpu_thread_is_idle(cpu)) {
108 return false;
111 return true;
114 /***********************************************************/
115 void hw_error(const char *fmt, ...)
117 va_list ap;
118 CPUState *cpu;
120 va_start(ap, fmt);
121 fprintf(stderr, "qemu: hardware error: ");
122 vfprintf(stderr, fmt, ap);
123 fprintf(stderr, "\n");
124 CPU_FOREACH(cpu) {
125 fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
126 cpu_dump_state(cpu, stderr, CPU_DUMP_FPU);
128 va_end(ap);
129 abort();
132 void cpu_synchronize_all_states(void)
134 CPUState *cpu;
136 CPU_FOREACH(cpu) {
137 cpu_synchronize_state(cpu);
141 void cpu_synchronize_all_post_reset(void)
143 CPUState *cpu;
145 CPU_FOREACH(cpu) {
146 cpu_synchronize_post_reset(cpu);
150 void cpu_synchronize_all_post_init(void)
152 CPUState *cpu;
154 CPU_FOREACH(cpu) {
155 cpu_synchronize_post_init(cpu);
159 void cpu_synchronize_all_pre_loadvm(void)
161 CPUState *cpu;
163 CPU_FOREACH(cpu) {
164 cpu_synchronize_pre_loadvm(cpu);
168 void cpu_synchronize_state(CPUState *cpu)
170 if (cpus_accel->synchronize_state) {
171 cpus_accel->synchronize_state(cpu);
175 void cpu_synchronize_post_reset(CPUState *cpu)
177 if (cpus_accel->synchronize_post_reset) {
178 cpus_accel->synchronize_post_reset(cpu);
182 void cpu_synchronize_post_init(CPUState *cpu)
184 if (cpus_accel->synchronize_post_init) {
185 cpus_accel->synchronize_post_init(cpu);
189 void cpu_synchronize_pre_loadvm(CPUState *cpu)
191 if (cpus_accel->synchronize_pre_loadvm) {
192 cpus_accel->synchronize_pre_loadvm(cpu);
196 bool cpus_are_resettable(void)
198 if (cpus_accel->cpus_are_resettable) {
199 return cpus_accel->cpus_are_resettable();
201 return true;
204 int64_t cpus_get_virtual_clock(void)
207 * XXX
209 * need to check that cpus_accel is not NULL, because qcow2 calls
210 * qemu_get_clock_ns(CLOCK_VIRTUAL) without any accel initialized and
211 * with ticks disabled in some io-tests:
212 * 030 040 041 060 099 120 127 140 156 161 172 181 191 192 195 203 229 249 256 267
214 * is this expected?
216 * XXX
218 if (cpus_accel && cpus_accel->get_virtual_clock) {
219 return cpus_accel->get_virtual_clock();
221 return cpu_get_clock();
225 * return the time elapsed in VM between vm_start and vm_stop. Unless
226 * icount is active, cpus_get_elapsed_ticks() uses units of the host CPU cycle
227 * counter.
229 int64_t cpus_get_elapsed_ticks(void)
231 if (cpus_accel->get_elapsed_ticks) {
232 return cpus_accel->get_elapsed_ticks();
234 return cpu_get_ticks();
237 static void generic_handle_interrupt(CPUState *cpu, int mask)
239 cpu->interrupt_request |= mask;
241 if (!qemu_cpu_is_self(cpu)) {
242 qemu_cpu_kick(cpu);
246 void cpu_interrupt(CPUState *cpu, int mask)
248 if (cpus_accel->handle_interrupt) {
249 cpus_accel->handle_interrupt(cpu, mask);
250 } else {
251 generic_handle_interrupt(cpu, mask);
255 static int do_vm_stop(RunState state, bool send_stop)
257 int ret = 0;
259 if (runstate_is_running()) {
260 runstate_set(state);
261 cpu_disable_ticks();
262 pause_all_vcpus();
263 vm_state_notify(0, state);
264 if (send_stop) {
265 qapi_event_send_stop();
269 bdrv_drain_all();
270 ret = bdrv_flush_all();
271 trace_vm_stop_flush_all(ret);
273 return ret;
276 /* Special vm_stop() variant for terminating the process. Historically clients
277 * did not expect a QMP STOP event and so we need to retain compatibility.
279 int vm_shutdown(void)
281 return do_vm_stop(RUN_STATE_SHUTDOWN, false);
284 bool cpu_can_run(CPUState *cpu)
286 if (cpu->stop) {
287 return false;
289 if (cpu_is_stopped(cpu)) {
290 return false;
292 return true;
295 void cpu_handle_guest_debug(CPUState *cpu)
297 if (replay_running_debug()) {
298 if (!cpu->singlestep_enabled) {
300 * Report about the breakpoint and
301 * make a single step to skip it
303 replay_breakpoint();
304 cpu_single_step(cpu, SSTEP_ENABLE);
305 } else {
306 cpu_single_step(cpu, 0);
308 } else {
309 gdb_set_stop_cpu(cpu);
310 qemu_system_debug_request();
311 cpu->stopped = true;
315 #ifdef CONFIG_LINUX
316 static void sigbus_reraise(void)
318 sigset_t set;
319 struct sigaction action;
321 memset(&action, 0, sizeof(action));
322 action.sa_handler = SIG_DFL;
323 if (!sigaction(SIGBUS, &action, NULL)) {
324 raise(SIGBUS);
325 sigemptyset(&set);
326 sigaddset(&set, SIGBUS);
327 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
329 perror("Failed to re-raise SIGBUS!");
330 abort();
333 static void sigbus_handler(int n, siginfo_t *siginfo, void *ctx)
335 if (siginfo->si_code != BUS_MCEERR_AO && siginfo->si_code != BUS_MCEERR_AR) {
336 sigbus_reraise();
339 if (current_cpu) {
340 /* Called asynchronously in VCPU thread. */
341 if (kvm_on_sigbus_vcpu(current_cpu, siginfo->si_code, siginfo->si_addr)) {
342 sigbus_reraise();
344 } else {
345 /* Called synchronously (via signalfd) in main thread. */
346 if (kvm_on_sigbus(siginfo->si_code, siginfo->si_addr)) {
347 sigbus_reraise();
352 static void qemu_init_sigbus(void)
354 struct sigaction action;
357 * ALERT: when modifying this, take care that SIGBUS forwarding in
358 * qemu_prealloc_mem() will continue working as expected.
360 memset(&action, 0, sizeof(action));
361 action.sa_flags = SA_SIGINFO;
362 action.sa_sigaction = sigbus_handler;
363 sigaction(SIGBUS, &action, NULL);
365 prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
367 #else /* !CONFIG_LINUX */
368 static void qemu_init_sigbus(void)
371 #endif /* !CONFIG_LINUX */
373 static QemuThread io_thread;
375 /* cpu creation */
376 static QemuCond qemu_cpu_cond;
377 /* system init */
378 static QemuCond qemu_pause_cond;
380 void qemu_init_cpu_loop(void)
382 qemu_init_sigbus();
383 qemu_cond_init(&qemu_cpu_cond);
384 qemu_cond_init(&qemu_pause_cond);
385 qemu_mutex_init(&qemu_global_mutex);
387 qemu_thread_get_self(&io_thread);
390 void run_on_cpu(CPUState *cpu, run_on_cpu_func func, run_on_cpu_data data)
392 do_run_on_cpu(cpu, func, data, &qemu_global_mutex);
395 static void qemu_cpu_stop(CPUState *cpu, bool exit)
397 g_assert(qemu_cpu_is_self(cpu));
398 cpu->stop = false;
399 cpu->stopped = true;
400 if (exit) {
401 cpu_exit(cpu);
403 qemu_cond_broadcast(&qemu_pause_cond);
406 void qemu_wait_io_event_common(CPUState *cpu)
408 qatomic_mb_set(&cpu->thread_kicked, false);
409 if (cpu->stop) {
410 qemu_cpu_stop(cpu, false);
412 process_queued_cpu_work(cpu);
415 void qemu_wait_io_event(CPUState *cpu)
417 bool slept = false;
419 while (cpu_thread_is_idle(cpu)) {
420 if (!slept) {
421 slept = true;
422 qemu_plugin_vcpu_idle_cb(cpu);
424 qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
426 if (slept) {
427 qemu_plugin_vcpu_resume_cb(cpu);
430 #ifdef _WIN32
431 /* Eat dummy APC queued by cpus_kick_thread. */
432 if (hax_enabled()) {
433 SleepEx(0, TRUE);
435 #endif
436 qemu_wait_io_event_common(cpu);
439 void cpus_kick_thread(CPUState *cpu)
441 if (cpu->thread_kicked) {
442 return;
444 cpu->thread_kicked = true;
446 #ifndef _WIN32
447 int err = pthread_kill(cpu->thread->thread, SIG_IPI);
448 if (err && err != ESRCH) {
449 fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
450 exit(1);
452 #else
453 qemu_sem_post(&cpu->sem);
454 #endif
457 void qemu_cpu_kick(CPUState *cpu)
459 qemu_cond_broadcast(cpu->halt_cond);
460 if (cpus_accel->kick_vcpu_thread) {
461 cpus_accel->kick_vcpu_thread(cpu);
462 } else { /* default */
463 cpus_kick_thread(cpu);
467 void qemu_cpu_kick_self(void)
469 assert(current_cpu);
470 cpus_kick_thread(current_cpu);
473 bool qemu_cpu_is_self(CPUState *cpu)
475 return qemu_thread_is_self(cpu->thread);
478 bool qemu_in_vcpu_thread(void)
480 return current_cpu && qemu_cpu_is_self(current_cpu);
483 QEMU_DEFINE_STATIC_CO_TLS(bool, iothread_locked)
485 bool qemu_mutex_iothread_locked(void)
487 return get_iothread_locked();
490 bool qemu_in_main_thread(void)
492 return qemu_mutex_iothread_locked();
496 * The BQL is taken from so many places that it is worth profiling the
497 * callers directly, instead of funneling them all through a single function.
499 void qemu_mutex_lock_iothread_impl(const char *file, int line)
501 QemuMutexLockFunc bql_lock = qatomic_read(&qemu_bql_mutex_lock_func);
503 g_assert(!qemu_mutex_iothread_locked());
504 bql_lock(&qemu_global_mutex, file, line);
505 set_iothread_locked(true);
508 void qemu_mutex_unlock_iothread(void)
510 g_assert(qemu_mutex_iothread_locked());
511 set_iothread_locked(false);
512 qemu_mutex_unlock(&qemu_global_mutex);
515 void qemu_cond_wait_iothread(QemuCond *cond)
517 qemu_cond_wait(cond, &qemu_global_mutex);
520 void qemu_cond_timedwait_iothread(QemuCond *cond, int ms)
522 qemu_cond_timedwait(cond, &qemu_global_mutex, ms);
525 /* signal CPU creation */
526 void cpu_thread_signal_created(CPUState *cpu)
528 cpu->created = true;
529 qemu_cond_signal(&qemu_cpu_cond);
532 /* signal CPU destruction */
533 void cpu_thread_signal_destroyed(CPUState *cpu)
535 cpu->created = false;
536 qemu_cond_signal(&qemu_cpu_cond);
540 static bool all_vcpus_paused(void)
542 CPUState *cpu;
544 CPU_FOREACH(cpu) {
545 if (!cpu->stopped) {
546 return false;
550 return true;
553 void pause_all_vcpus(void)
555 CPUState *cpu;
557 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false);
558 CPU_FOREACH(cpu) {
559 if (qemu_cpu_is_self(cpu)) {
560 qemu_cpu_stop(cpu, true);
561 } else {
562 cpu->stop = true;
563 qemu_cpu_kick(cpu);
567 /* We need to drop the replay_lock so any vCPU threads woken up
568 * can finish their replay tasks
570 replay_mutex_unlock();
572 while (!all_vcpus_paused()) {
573 qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
574 CPU_FOREACH(cpu) {
575 qemu_cpu_kick(cpu);
579 qemu_mutex_unlock_iothread();
580 replay_mutex_lock();
581 qemu_mutex_lock_iothread();
584 void cpu_resume(CPUState *cpu)
586 cpu->stop = false;
587 cpu->stopped = false;
588 qemu_cpu_kick(cpu);
591 void resume_all_vcpus(void)
593 CPUState *cpu;
595 if (!runstate_is_running()) {
596 return;
599 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true);
600 CPU_FOREACH(cpu) {
601 cpu_resume(cpu);
605 void cpu_remove_sync(CPUState *cpu)
607 cpu->stop = true;
608 cpu->unplug = true;
609 qemu_cpu_kick(cpu);
610 qemu_mutex_unlock_iothread();
611 qemu_thread_join(cpu->thread);
612 qemu_mutex_lock_iothread();
615 void cpus_register_accel(const AccelOpsClass *ops)
617 assert(ops != NULL);
618 assert(ops->create_vcpu_thread != NULL); /* mandatory */
619 cpus_accel = ops;
622 const AccelOpsClass *cpus_get_accel(void)
624 /* broken if we call this early */
625 assert(cpus_accel);
626 return cpus_accel;
629 void qemu_init_vcpu(CPUState *cpu)
631 MachineState *ms = MACHINE(qdev_get_machine());
633 cpu->nr_cores = ms->smp.cores;
634 cpu->nr_threads = ms->smp.threads;
635 cpu->stopped = true;
636 cpu->random_seed = qemu_guest_random_seed_thread_part1();
638 if (!cpu->as) {
639 /* If the target cpu hasn't set up any address spaces itself,
640 * give it the default one.
642 cpu->num_ases = 1;
643 cpu_address_space_init(cpu, 0, "cpu-memory", cpu->memory);
646 /* accelerators all implement the AccelOpsClass */
647 g_assert(cpus_accel != NULL && cpus_accel->create_vcpu_thread != NULL);
648 cpus_accel->create_vcpu_thread(cpu);
650 while (!cpu->created) {
651 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
655 void cpu_stop_current(void)
657 if (current_cpu) {
658 current_cpu->stop = true;
659 cpu_exit(current_cpu);
663 int vm_stop(RunState state)
665 if (qemu_in_vcpu_thread()) {
666 qemu_system_vmstop_request_prepare();
667 qemu_system_vmstop_request(state);
669 * FIXME: should not return to device code in case
670 * vm_stop() has been requested.
672 cpu_stop_current();
673 return 0;
676 return do_vm_stop(state, true);
680 * Prepare for (re)starting the VM.
681 * Returns -1 if the vCPUs are not to be restarted (e.g. if they are already
682 * running or in case of an error condition), 0 otherwise.
684 int vm_prepare_start(bool step_pending)
686 RunState requested;
688 qemu_vmstop_requested(&requested);
689 if (runstate_is_running() && requested == RUN_STATE__MAX) {
690 return -1;
693 /* Ensure that a STOP/RESUME pair of events is emitted if a
694 * vmstop request was pending. The BLOCK_IO_ERROR event, for
695 * example, according to documentation is always followed by
696 * the STOP event.
698 if (runstate_is_running()) {
699 qapi_event_send_stop();
700 qapi_event_send_resume();
701 return -1;
705 * WHPX accelerator needs to know whether we are going to step
706 * any CPUs, before starting the first one.
708 if (cpus_accel->synchronize_pre_resume) {
709 cpus_accel->synchronize_pre_resume(step_pending);
712 /* We are sending this now, but the CPUs will be resumed shortly later */
713 qapi_event_send_resume();
715 cpu_enable_ticks();
716 runstate_set(RUN_STATE_RUNNING);
717 vm_state_notify(1, RUN_STATE_RUNNING);
718 return 0;
721 void vm_start(void)
723 if (!vm_prepare_start(false)) {
724 resume_all_vcpus();
728 /* does a state transition even if the VM is already stopped,
729 current state is forgotten forever */
730 int vm_stop_force_state(RunState state)
732 if (runstate_is_running()) {
733 return vm_stop(state);
734 } else {
735 int ret;
736 runstate_set(state);
738 bdrv_drain_all();
739 /* Make sure to return an error if the flush in a previous vm_stop()
740 * failed. */
741 ret = bdrv_flush_all();
742 trace_vm_stop_flush_all(ret);
743 return ret;
747 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
748 bool has_cpu, int64_t cpu_index, Error **errp)
750 FILE *f;
751 uint32_t l;
752 CPUState *cpu;
753 uint8_t buf[1024];
754 int64_t orig_addr = addr, orig_size = size;
756 if (!has_cpu) {
757 cpu_index = 0;
760 cpu = qemu_get_cpu(cpu_index);
761 if (cpu == NULL) {
762 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
763 "a CPU number");
764 return;
767 f = fopen(filename, "wb");
768 if (!f) {
769 error_setg_file_open(errp, errno, filename);
770 return;
773 while (size != 0) {
774 l = sizeof(buf);
775 if (l > size)
776 l = size;
777 if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) {
778 error_setg(errp, "Invalid addr 0x%016" PRIx64 "/size %" PRId64
779 " specified", orig_addr, orig_size);
780 goto exit;
782 if (fwrite(buf, 1, l, f) != l) {
783 error_setg(errp, QERR_IO_ERROR);
784 goto exit;
786 addr += l;
787 size -= l;
790 exit:
791 fclose(f);
794 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
795 Error **errp)
797 FILE *f;
798 uint32_t l;
799 uint8_t buf[1024];
801 f = fopen(filename, "wb");
802 if (!f) {
803 error_setg_file_open(errp, errno, filename);
804 return;
807 while (size != 0) {
808 l = sizeof(buf);
809 if (l > size)
810 l = size;
811 cpu_physical_memory_read(addr, buf, l);
812 if (fwrite(buf, 1, l, f) != l) {
813 error_setg(errp, QERR_IO_ERROR);
814 goto exit;
816 addr += l;
817 size -= l;
820 exit:
821 fclose(f);
824 void qmp_inject_nmi(Error **errp)
826 nmi_monitor_handle(monitor_get_cpu_index(monitor_cur()), errp);