kvm: Enable XSAVE live migration support
[qemu/stefanha.git] / cpus.c
blobfcd0f09b423bf64a2d588047c8c916d942251257
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 /* Needed early for CONFIG_BSD etc. */
26 #include "config-host.h"
28 #include "monitor.h"
29 #include "sysemu.h"
30 #include "gdbstub.h"
31 #include "dma.h"
32 #include "kvm.h"
34 #include "cpus.h"
36 #ifdef SIGRTMIN
37 #define SIG_IPI (SIGRTMIN+4)
38 #else
39 #define SIG_IPI SIGUSR1
40 #endif
42 static CPUState *cur_cpu;
43 static CPUState *next_cpu;
45 /***********************************************************/
46 void hw_error(const char *fmt, ...)
48 va_list ap;
49 CPUState *env;
51 va_start(ap, fmt);
52 fprintf(stderr, "qemu: hardware error: ");
53 vfprintf(stderr, fmt, ap);
54 fprintf(stderr, "\n");
55 for(env = first_cpu; env != NULL; env = env->next_cpu) {
56 fprintf(stderr, "CPU #%d:\n", env->cpu_index);
57 #ifdef TARGET_I386
58 cpu_dump_state(env, stderr, fprintf, X86_DUMP_FPU);
59 #else
60 cpu_dump_state(env, stderr, fprintf, 0);
61 #endif
63 va_end(ap);
64 abort();
67 void cpu_synchronize_all_states(void)
69 CPUState *cpu;
71 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
72 cpu_synchronize_state(cpu);
76 void cpu_synchronize_all_post_reset(void)
78 CPUState *cpu;
80 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
81 cpu_synchronize_post_reset(cpu);
85 void cpu_synchronize_all_post_init(void)
87 CPUState *cpu;
89 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
90 cpu_synchronize_post_init(cpu);
94 int cpu_is_stopped(CPUState *env)
96 return !vm_running || env->stopped;
99 static void do_vm_stop(int reason)
101 if (vm_running) {
102 cpu_disable_ticks();
103 vm_running = 0;
104 pause_all_vcpus();
105 vm_state_notify(0, reason);
106 monitor_protocol_event(QEVENT_STOP, NULL);
110 static int cpu_can_run(CPUState *env)
112 if (env->stop)
113 return 0;
114 if (env->stopped || !vm_running)
115 return 0;
116 return 1;
119 static int cpu_has_work(CPUState *env)
121 if (env->stop)
122 return 1;
123 if (env->queued_work_first)
124 return 1;
125 if (env->stopped || !vm_running)
126 return 0;
127 if (!env->halted)
128 return 1;
129 if (qemu_cpu_has_work(env))
130 return 1;
131 return 0;
134 static int tcg_has_work(void)
136 CPUState *env;
138 for (env = first_cpu; env != NULL; env = env->next_cpu)
139 if (cpu_has_work(env))
140 return 1;
141 return 0;
144 #ifndef _WIN32
145 static int io_thread_fd = -1;
147 static void qemu_event_increment(void)
149 /* Write 8 bytes to be compatible with eventfd. */
150 static const uint64_t val = 1;
151 ssize_t ret;
153 if (io_thread_fd == -1)
154 return;
156 do {
157 ret = write(io_thread_fd, &val, sizeof(val));
158 } while (ret < 0 && errno == EINTR);
160 /* EAGAIN is fine, a read must be pending. */
161 if (ret < 0 && errno != EAGAIN) {
162 fprintf(stderr, "qemu_event_increment: write() filed: %s\n",
163 strerror(errno));
164 exit (1);
168 static void qemu_event_read(void *opaque)
170 int fd = (unsigned long)opaque;
171 ssize_t len;
172 char buffer[512];
174 /* Drain the notify pipe. For eventfd, only 8 bytes will be read. */
175 do {
176 len = read(fd, buffer, sizeof(buffer));
177 } while ((len == -1 && errno == EINTR) || len == sizeof(buffer));
180 static int qemu_event_init(void)
182 int err;
183 int fds[2];
185 err = qemu_eventfd(fds);
186 if (err == -1)
187 return -errno;
189 err = fcntl_setfl(fds[0], O_NONBLOCK);
190 if (err < 0)
191 goto fail;
193 err = fcntl_setfl(fds[1], O_NONBLOCK);
194 if (err < 0)
195 goto fail;
197 qemu_set_fd_handler2(fds[0], NULL, qemu_event_read, NULL,
198 (void *)(unsigned long)fds[0]);
200 io_thread_fd = fds[1];
201 return 0;
203 fail:
204 close(fds[0]);
205 close(fds[1]);
206 return err;
208 #else
209 HANDLE qemu_event_handle;
211 static void dummy_event_handler(void *opaque)
215 static int qemu_event_init(void)
217 qemu_event_handle = CreateEvent(NULL, FALSE, FALSE, NULL);
218 if (!qemu_event_handle) {
219 fprintf(stderr, "Failed CreateEvent: %ld\n", GetLastError());
220 return -1;
222 qemu_add_wait_object(qemu_event_handle, dummy_event_handler, NULL);
223 return 0;
226 static void qemu_event_increment(void)
228 if (!SetEvent(qemu_event_handle)) {
229 fprintf(stderr, "qemu_event_increment: SetEvent failed: %ld\n",
230 GetLastError());
231 exit (1);
234 #endif
236 #ifndef CONFIG_IOTHREAD
237 int qemu_init_main_loop(void)
239 return qemu_event_init();
242 void qemu_main_loop_start(void)
246 void qemu_init_vcpu(void *_env)
248 CPUState *env = _env;
250 env->nr_cores = smp_cores;
251 env->nr_threads = smp_threads;
252 if (kvm_enabled())
253 kvm_init_vcpu(env);
254 return;
257 int qemu_cpu_self(void *env)
259 return 1;
262 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
264 func(data);
267 void resume_all_vcpus(void)
271 void pause_all_vcpus(void)
275 void qemu_cpu_kick(void *env)
277 return;
280 void qemu_notify_event(void)
282 CPUState *env = cpu_single_env;
284 qemu_event_increment ();
285 if (env) {
286 cpu_exit(env);
288 if (next_cpu && env != next_cpu) {
289 cpu_exit(next_cpu);
293 void qemu_mutex_lock_iothread(void) {}
294 void qemu_mutex_unlock_iothread(void) {}
296 void vm_stop(int reason)
298 do_vm_stop(reason);
301 #else /* CONFIG_IOTHREAD */
303 #include "qemu-thread.h"
305 QemuMutex qemu_global_mutex;
306 static QemuMutex qemu_fair_mutex;
308 static QemuThread io_thread;
310 static QemuThread *tcg_cpu_thread;
311 static QemuCond *tcg_halt_cond;
313 static int qemu_system_ready;
314 /* cpu creation */
315 static QemuCond qemu_cpu_cond;
316 /* system init */
317 static QemuCond qemu_system_cond;
318 static QemuCond qemu_pause_cond;
319 static QemuCond qemu_work_cond;
321 static void tcg_init_ipi(void);
322 static void kvm_init_ipi(CPUState *env);
323 static void unblock_io_signals(void);
325 int qemu_init_main_loop(void)
327 int ret;
329 ret = qemu_event_init();
330 if (ret)
331 return ret;
333 qemu_cond_init(&qemu_pause_cond);
334 qemu_mutex_init(&qemu_fair_mutex);
335 qemu_mutex_init(&qemu_global_mutex);
336 qemu_mutex_lock(&qemu_global_mutex);
338 unblock_io_signals();
339 qemu_thread_self(&io_thread);
341 return 0;
344 void qemu_main_loop_start(void)
346 qemu_system_ready = 1;
347 qemu_cond_broadcast(&qemu_system_cond);
350 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
352 struct qemu_work_item wi;
354 if (qemu_cpu_self(env)) {
355 func(data);
356 return;
359 wi.func = func;
360 wi.data = data;
361 if (!env->queued_work_first)
362 env->queued_work_first = &wi;
363 else
364 env->queued_work_last->next = &wi;
365 env->queued_work_last = &wi;
366 wi.next = NULL;
367 wi.done = false;
369 qemu_cpu_kick(env);
370 while (!wi.done) {
371 CPUState *self_env = cpu_single_env;
373 qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
374 cpu_single_env = self_env;
378 static void flush_queued_work(CPUState *env)
380 struct qemu_work_item *wi;
382 if (!env->queued_work_first)
383 return;
385 while ((wi = env->queued_work_first)) {
386 env->queued_work_first = wi->next;
387 wi->func(wi->data);
388 wi->done = true;
390 env->queued_work_last = NULL;
391 qemu_cond_broadcast(&qemu_work_cond);
394 static void qemu_wait_io_event_common(CPUState *env)
396 if (env->stop) {
397 env->stop = 0;
398 env->stopped = 1;
399 qemu_cond_signal(&qemu_pause_cond);
401 flush_queued_work(env);
404 static void qemu_wait_io_event(CPUState *env)
406 while (!tcg_has_work())
407 qemu_cond_timedwait(env->halt_cond, &qemu_global_mutex, 1000);
409 qemu_mutex_unlock(&qemu_global_mutex);
412 * Users of qemu_global_mutex can be starved, having no chance
413 * to acquire it since this path will get to it first.
414 * So use another lock to provide fairness.
416 qemu_mutex_lock(&qemu_fair_mutex);
417 qemu_mutex_unlock(&qemu_fair_mutex);
419 qemu_mutex_lock(&qemu_global_mutex);
420 qemu_wait_io_event_common(env);
423 static void qemu_kvm_eat_signal(CPUState *env, int timeout)
425 struct timespec ts;
426 int r, e;
427 siginfo_t siginfo;
428 sigset_t waitset;
430 ts.tv_sec = timeout / 1000;
431 ts.tv_nsec = (timeout % 1000) * 1000000;
433 sigemptyset(&waitset);
434 sigaddset(&waitset, SIG_IPI);
436 qemu_mutex_unlock(&qemu_global_mutex);
437 r = sigtimedwait(&waitset, &siginfo, &ts);
438 e = errno;
439 qemu_mutex_lock(&qemu_global_mutex);
441 if (r == -1 && !(e == EAGAIN || e == EINTR)) {
442 fprintf(stderr, "sigtimedwait: %s\n", strerror(e));
443 exit(1);
447 static void qemu_kvm_wait_io_event(CPUState *env)
449 while (!cpu_has_work(env))
450 qemu_cond_timedwait(env->halt_cond, &qemu_global_mutex, 1000);
452 qemu_kvm_eat_signal(env, 0);
453 qemu_wait_io_event_common(env);
456 static int qemu_cpu_exec(CPUState *env);
458 static void *kvm_cpu_thread_fn(void *arg)
460 CPUState *env = arg;
462 qemu_mutex_lock(&qemu_global_mutex);
463 qemu_thread_self(env->thread);
464 if (kvm_enabled())
465 kvm_init_vcpu(env);
467 kvm_init_ipi(env);
469 /* signal CPU creation */
470 env->created = 1;
471 qemu_cond_signal(&qemu_cpu_cond);
473 /* and wait for machine initialization */
474 while (!qemu_system_ready)
475 qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
477 while (1) {
478 if (cpu_can_run(env))
479 qemu_cpu_exec(env);
480 qemu_kvm_wait_io_event(env);
483 return NULL;
486 static void *tcg_cpu_thread_fn(void *arg)
488 CPUState *env = arg;
490 tcg_init_ipi();
491 qemu_thread_self(env->thread);
493 /* signal CPU creation */
494 qemu_mutex_lock(&qemu_global_mutex);
495 for (env = first_cpu; env != NULL; env = env->next_cpu)
496 env->created = 1;
497 qemu_cond_signal(&qemu_cpu_cond);
499 /* and wait for machine initialization */
500 while (!qemu_system_ready)
501 qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
503 while (1) {
504 tcg_cpu_exec();
505 qemu_wait_io_event(cur_cpu);
508 return NULL;
511 void qemu_cpu_kick(void *_env)
513 CPUState *env = _env;
514 qemu_cond_broadcast(env->halt_cond);
515 qemu_thread_signal(env->thread, SIG_IPI);
518 int qemu_cpu_self(void *_env)
520 CPUState *env = _env;
521 QemuThread this;
523 qemu_thread_self(&this);
525 return qemu_thread_equal(&this, env->thread);
528 static void cpu_signal(int sig)
530 if (cpu_single_env)
531 cpu_exit(cpu_single_env);
532 exit_request = 1;
535 static void tcg_init_ipi(void)
537 sigset_t set;
538 struct sigaction sigact;
540 memset(&sigact, 0, sizeof(sigact));
541 sigact.sa_handler = cpu_signal;
542 sigaction(SIG_IPI, &sigact, NULL);
544 sigemptyset(&set);
545 sigaddset(&set, SIG_IPI);
546 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
549 static void dummy_signal(int sig)
553 static void kvm_init_ipi(CPUState *env)
555 int r;
556 sigset_t set;
557 struct sigaction sigact;
559 memset(&sigact, 0, sizeof(sigact));
560 sigact.sa_handler = dummy_signal;
561 sigaction(SIG_IPI, &sigact, NULL);
563 pthread_sigmask(SIG_BLOCK, NULL, &set);
564 sigdelset(&set, SIG_IPI);
565 r = kvm_set_signal_mask(env, &set);
566 if (r) {
567 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(r));
568 exit(1);
572 static void unblock_io_signals(void)
574 sigset_t set;
576 sigemptyset(&set);
577 sigaddset(&set, SIGUSR2);
578 sigaddset(&set, SIGIO);
579 sigaddset(&set, SIGALRM);
580 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
582 sigemptyset(&set);
583 sigaddset(&set, SIG_IPI);
584 pthread_sigmask(SIG_BLOCK, &set, NULL);
587 void qemu_mutex_lock_iothread(void)
589 if (kvm_enabled()) {
590 qemu_mutex_lock(&qemu_fair_mutex);
591 qemu_mutex_lock(&qemu_global_mutex);
592 qemu_mutex_unlock(&qemu_fair_mutex);
593 } else {
594 qemu_mutex_lock(&qemu_fair_mutex);
595 if (qemu_mutex_trylock(&qemu_global_mutex)) {
596 qemu_thread_signal(tcg_cpu_thread, SIG_IPI);
597 qemu_mutex_lock(&qemu_global_mutex);
599 qemu_mutex_unlock(&qemu_fair_mutex);
603 void qemu_mutex_unlock_iothread(void)
605 qemu_mutex_unlock(&qemu_global_mutex);
608 static int all_vcpus_paused(void)
610 CPUState *penv = first_cpu;
612 while (penv) {
613 if (!penv->stopped)
614 return 0;
615 penv = (CPUState *)penv->next_cpu;
618 return 1;
621 void pause_all_vcpus(void)
623 CPUState *penv = first_cpu;
625 while (penv) {
626 penv->stop = 1;
627 qemu_cpu_kick(penv);
628 penv = (CPUState *)penv->next_cpu;
631 while (!all_vcpus_paused()) {
632 qemu_cond_timedwait(&qemu_pause_cond, &qemu_global_mutex, 100);
633 penv = first_cpu;
634 while (penv) {
635 qemu_cpu_kick(penv);
636 penv = (CPUState *)penv->next_cpu;
641 void resume_all_vcpus(void)
643 CPUState *penv = first_cpu;
645 while (penv) {
646 penv->stop = 0;
647 penv->stopped = 0;
648 qemu_cpu_kick(penv);
649 penv = (CPUState *)penv->next_cpu;
653 static void tcg_init_vcpu(void *_env)
655 CPUState *env = _env;
656 /* share a single thread for all cpus with TCG */
657 if (!tcg_cpu_thread) {
658 env->thread = qemu_mallocz(sizeof(QemuThread));
659 env->halt_cond = qemu_mallocz(sizeof(QemuCond));
660 qemu_cond_init(env->halt_cond);
661 qemu_thread_create(env->thread, tcg_cpu_thread_fn, env);
662 while (env->created == 0)
663 qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
664 tcg_cpu_thread = env->thread;
665 tcg_halt_cond = env->halt_cond;
666 } else {
667 env->thread = tcg_cpu_thread;
668 env->halt_cond = tcg_halt_cond;
672 static void kvm_start_vcpu(CPUState *env)
674 env->thread = qemu_mallocz(sizeof(QemuThread));
675 env->halt_cond = qemu_mallocz(sizeof(QemuCond));
676 qemu_cond_init(env->halt_cond);
677 qemu_thread_create(env->thread, kvm_cpu_thread_fn, env);
678 while (env->created == 0)
679 qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
682 void qemu_init_vcpu(void *_env)
684 CPUState *env = _env;
686 env->nr_cores = smp_cores;
687 env->nr_threads = smp_threads;
688 if (kvm_enabled())
689 kvm_start_vcpu(env);
690 else
691 tcg_init_vcpu(env);
694 void qemu_notify_event(void)
696 qemu_event_increment();
699 static void qemu_system_vmstop_request(int reason)
701 vmstop_requested = reason;
702 qemu_notify_event();
705 void vm_stop(int reason)
707 QemuThread me;
708 qemu_thread_self(&me);
710 if (!qemu_thread_equal(&me, &io_thread)) {
711 qemu_system_vmstop_request(reason);
713 * FIXME: should not return to device code in case
714 * vm_stop() has been requested.
716 if (cpu_single_env) {
717 cpu_exit(cpu_single_env);
718 cpu_single_env->stop = 1;
720 return;
722 do_vm_stop(reason);
725 #endif
727 static int qemu_cpu_exec(CPUState *env)
729 int ret;
730 #ifdef CONFIG_PROFILER
731 int64_t ti;
732 #endif
734 #ifdef CONFIG_PROFILER
735 ti = profile_getclock();
736 #endif
737 if (use_icount) {
738 int64_t count;
739 int decr;
740 qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
741 env->icount_decr.u16.low = 0;
742 env->icount_extra = 0;
743 count = qemu_icount_round (qemu_next_deadline());
744 qemu_icount += count;
745 decr = (count > 0xffff) ? 0xffff : count;
746 count -= decr;
747 env->icount_decr.u16.low = decr;
748 env->icount_extra = count;
750 ret = cpu_exec(env);
751 #ifdef CONFIG_PROFILER
752 qemu_time += profile_getclock() - ti;
753 #endif
754 if (use_icount) {
755 /* Fold pending instructions back into the
756 instruction counter, and clear the interrupt flag. */
757 qemu_icount -= (env->icount_decr.u16.low
758 + env->icount_extra);
759 env->icount_decr.u32 = 0;
760 env->icount_extra = 0;
762 return ret;
765 bool tcg_cpu_exec(void)
767 int ret = 0;
769 if (next_cpu == NULL)
770 next_cpu = first_cpu;
771 for (; next_cpu != NULL; next_cpu = next_cpu->next_cpu) {
772 CPUState *env = cur_cpu = next_cpu;
774 qemu_clock_enable(vm_clock,
775 (cur_cpu->singlestep_enabled & SSTEP_NOTIMER) == 0);
777 if (qemu_alarm_pending())
778 break;
779 if (cpu_can_run(env))
780 ret = qemu_cpu_exec(env);
781 else if (env->stop)
782 break;
784 if (ret == EXCP_DEBUG) {
785 gdb_set_stop_cpu(env);
786 debug_requested = EXCP_DEBUG;
787 break;
790 return tcg_has_work();
793 void set_numa_modes(void)
795 CPUState *env;
796 int i;
798 for (env = first_cpu; env != NULL; env = env->next_cpu) {
799 for (i = 0; i < nb_numa_nodes; i++) {
800 if (node_cpumask[i] & (1 << env->cpu_index)) {
801 env->numa_node = i;
807 void set_cpu_log(const char *optarg)
809 int mask;
810 const CPULogItem *item;
812 mask = cpu_str_to_log_mask(optarg);
813 if (!mask) {
814 printf("Log items (comma separated):\n");
815 for (item = cpu_log_items; item->mask != 0; item++) {
816 printf("%-10s %s\n", item->name, item->help);
818 exit(1);
820 cpu_set_log(mask);
823 /* Return the virtual CPU time, based on the instruction counter. */
824 int64_t cpu_get_icount(void)
826 int64_t icount;
827 CPUState *env = cpu_single_env;;
829 icount = qemu_icount;
830 if (env) {
831 if (!can_do_io(env)) {
832 fprintf(stderr, "Bad clock read\n");
834 icount -= (env->icount_decr.u16.low + env->icount_extra);
836 return qemu_icount_bias + (icount << icount_time_shift);
839 void list_cpus(FILE *f, int (*cpu_fprintf)(FILE *f, const char *fmt, ...),
840 const char *optarg)
842 /* XXX: implement xxx_cpu_list for targets that still miss it */
843 #if defined(cpu_list_id)
844 cpu_list_id(f, cpu_fprintf, optarg);
845 #elif defined(cpu_list)
846 cpu_list(f, cpu_fprintf); /* deprecated */
847 #endif