qemu_ram_free: Implement it
[qemu/kraxel.git] / cpus.c
blobeef0bb8a0a1b26a99de0aa1cda956b0bef3cd4ac
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"
33 #include "exec-all.h"
35 #include "cpus.h"
37 #ifdef SIGRTMIN
38 #define SIG_IPI (SIGRTMIN+4)
39 #else
40 #define SIG_IPI SIGUSR1
41 #endif
43 static CPUState *cur_cpu;
44 static CPUState *next_cpu;
46 /***********************************************************/
47 void hw_error(const char *fmt, ...)
49 va_list ap;
50 CPUState *env;
52 va_start(ap, fmt);
53 fprintf(stderr, "qemu: hardware error: ");
54 vfprintf(stderr, fmt, ap);
55 fprintf(stderr, "\n");
56 for(env = first_cpu; env != NULL; env = env->next_cpu) {
57 fprintf(stderr, "CPU #%d:\n", env->cpu_index);
58 #ifdef TARGET_I386
59 cpu_dump_state(env, stderr, fprintf, X86_DUMP_FPU);
60 #else
61 cpu_dump_state(env, stderr, fprintf, 0);
62 #endif
64 va_end(ap);
65 abort();
68 void cpu_synchronize_all_states(void)
70 CPUState *cpu;
72 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
73 cpu_synchronize_state(cpu);
77 void cpu_synchronize_all_post_reset(void)
79 CPUState *cpu;
81 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
82 cpu_synchronize_post_reset(cpu);
86 void cpu_synchronize_all_post_init(void)
88 CPUState *cpu;
90 for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
91 cpu_synchronize_post_init(cpu);
95 int cpu_is_stopped(CPUState *env)
97 return !vm_running || env->stopped;
100 static void do_vm_stop(int reason)
102 if (vm_running) {
103 cpu_disable_ticks();
104 vm_running = 0;
105 pause_all_vcpus();
106 vm_state_notify(0, reason);
107 monitor_protocol_event(QEVENT_STOP, NULL);
111 static int cpu_can_run(CPUState *env)
113 if (env->stop)
114 return 0;
115 if (env->stopped || !vm_running)
116 return 0;
117 return 1;
120 static int cpu_has_work(CPUState *env)
122 if (env->stop)
123 return 1;
124 if (env->queued_work_first)
125 return 1;
126 if (env->stopped || !vm_running)
127 return 0;
128 if (!env->halted)
129 return 1;
130 if (qemu_cpu_has_work(env))
131 return 1;
132 return 0;
135 static int tcg_has_work(void)
137 CPUState *env;
139 for (env = first_cpu; env != NULL; env = env->next_cpu)
140 if (cpu_has_work(env))
141 return 1;
142 return 0;
145 #ifndef _WIN32
146 static int io_thread_fd = -1;
148 static void qemu_event_increment(void)
150 /* Write 8 bytes to be compatible with eventfd. */
151 static const uint64_t val = 1;
152 ssize_t ret;
154 if (io_thread_fd == -1)
155 return;
157 do {
158 ret = write(io_thread_fd, &val, sizeof(val));
159 } while (ret < 0 && errno == EINTR);
161 /* EAGAIN is fine, a read must be pending. */
162 if (ret < 0 && errno != EAGAIN) {
163 fprintf(stderr, "qemu_event_increment: write() filed: %s\n",
164 strerror(errno));
165 exit (1);
169 static void qemu_event_read(void *opaque)
171 int fd = (unsigned long)opaque;
172 ssize_t len;
173 char buffer[512];
175 /* Drain the notify pipe. For eventfd, only 8 bytes will be read. */
176 do {
177 len = read(fd, buffer, sizeof(buffer));
178 } while ((len == -1 && errno == EINTR) || len == sizeof(buffer));
181 static int qemu_event_init(void)
183 int err;
184 int fds[2];
186 err = qemu_eventfd(fds);
187 if (err == -1)
188 return -errno;
190 err = fcntl_setfl(fds[0], O_NONBLOCK);
191 if (err < 0)
192 goto fail;
194 err = fcntl_setfl(fds[1], O_NONBLOCK);
195 if (err < 0)
196 goto fail;
198 qemu_set_fd_handler2(fds[0], NULL, qemu_event_read, NULL,
199 (void *)(unsigned long)fds[0]);
201 io_thread_fd = fds[1];
202 return 0;
204 fail:
205 close(fds[0]);
206 close(fds[1]);
207 return err;
209 #else
210 HANDLE qemu_event_handle;
212 static void dummy_event_handler(void *opaque)
216 static int qemu_event_init(void)
218 qemu_event_handle = CreateEvent(NULL, FALSE, FALSE, NULL);
219 if (!qemu_event_handle) {
220 fprintf(stderr, "Failed CreateEvent: %ld\n", GetLastError());
221 return -1;
223 qemu_add_wait_object(qemu_event_handle, dummy_event_handler, NULL);
224 return 0;
227 static void qemu_event_increment(void)
229 if (!SetEvent(qemu_event_handle)) {
230 fprintf(stderr, "qemu_event_increment: SetEvent failed: %ld\n",
231 GetLastError());
232 exit (1);
235 #endif
237 #ifndef CONFIG_IOTHREAD
238 int qemu_init_main_loop(void)
240 return qemu_event_init();
243 void qemu_main_loop_start(void)
247 void qemu_init_vcpu(void *_env)
249 CPUState *env = _env;
251 env->nr_cores = smp_cores;
252 env->nr_threads = smp_threads;
253 if (kvm_enabled())
254 kvm_init_vcpu(env);
255 return;
258 int qemu_cpu_self(void *env)
260 return 1;
263 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
265 func(data);
268 void resume_all_vcpus(void)
272 void pause_all_vcpus(void)
276 void qemu_cpu_kick(void *env)
278 return;
281 void qemu_notify_event(void)
283 CPUState *env = cpu_single_env;
285 qemu_event_increment ();
286 if (env) {
287 cpu_exit(env);
289 if (next_cpu && env != next_cpu) {
290 cpu_exit(next_cpu);
294 void qemu_mutex_lock_iothread(void) {}
295 void qemu_mutex_unlock_iothread(void) {}
297 void vm_stop(int reason)
299 do_vm_stop(reason);
302 #else /* CONFIG_IOTHREAD */
304 #include "qemu-thread.h"
306 QemuMutex qemu_global_mutex;
307 static QemuMutex qemu_fair_mutex;
309 static QemuThread io_thread;
311 static QemuThread *tcg_cpu_thread;
312 static QemuCond *tcg_halt_cond;
314 static int qemu_system_ready;
315 /* cpu creation */
316 static QemuCond qemu_cpu_cond;
317 /* system init */
318 static QemuCond qemu_system_cond;
319 static QemuCond qemu_pause_cond;
320 static QemuCond qemu_work_cond;
322 static void tcg_init_ipi(void);
323 static void kvm_init_ipi(CPUState *env);
324 static void unblock_io_signals(void);
326 int qemu_init_main_loop(void)
328 int ret;
330 ret = qemu_event_init();
331 if (ret)
332 return ret;
334 qemu_cond_init(&qemu_pause_cond);
335 qemu_mutex_init(&qemu_fair_mutex);
336 qemu_mutex_init(&qemu_global_mutex);
337 qemu_mutex_lock(&qemu_global_mutex);
339 unblock_io_signals();
340 qemu_thread_self(&io_thread);
342 return 0;
345 void qemu_main_loop_start(void)
347 qemu_system_ready = 1;
348 qemu_cond_broadcast(&qemu_system_cond);
351 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
353 struct qemu_work_item wi;
355 if (qemu_cpu_self(env)) {
356 func(data);
357 return;
360 wi.func = func;
361 wi.data = data;
362 if (!env->queued_work_first)
363 env->queued_work_first = &wi;
364 else
365 env->queued_work_last->next = &wi;
366 env->queued_work_last = &wi;
367 wi.next = NULL;
368 wi.done = false;
370 qemu_cpu_kick(env);
371 while (!wi.done) {
372 CPUState *self_env = cpu_single_env;
374 qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
375 cpu_single_env = self_env;
379 static void flush_queued_work(CPUState *env)
381 struct qemu_work_item *wi;
383 if (!env->queued_work_first)
384 return;
386 while ((wi = env->queued_work_first)) {
387 env->queued_work_first = wi->next;
388 wi->func(wi->data);
389 wi->done = true;
391 env->queued_work_last = NULL;
392 qemu_cond_broadcast(&qemu_work_cond);
395 static void qemu_wait_io_event_common(CPUState *env)
397 if (env->stop) {
398 env->stop = 0;
399 env->stopped = 1;
400 qemu_cond_signal(&qemu_pause_cond);
402 flush_queued_work(env);
405 static void qemu_wait_io_event(CPUState *env)
407 while (!tcg_has_work())
408 qemu_cond_timedwait(env->halt_cond, &qemu_global_mutex, 1000);
410 qemu_mutex_unlock(&qemu_global_mutex);
413 * Users of qemu_global_mutex can be starved, having no chance
414 * to acquire it since this path will get to it first.
415 * So use another lock to provide fairness.
417 qemu_mutex_lock(&qemu_fair_mutex);
418 qemu_mutex_unlock(&qemu_fair_mutex);
420 qemu_mutex_lock(&qemu_global_mutex);
421 qemu_wait_io_event_common(env);
424 static void qemu_kvm_eat_signal(CPUState *env, int timeout)
426 struct timespec ts;
427 int r, e;
428 siginfo_t siginfo;
429 sigset_t waitset;
431 ts.tv_sec = timeout / 1000;
432 ts.tv_nsec = (timeout % 1000) * 1000000;
434 sigemptyset(&waitset);
435 sigaddset(&waitset, SIG_IPI);
437 qemu_mutex_unlock(&qemu_global_mutex);
438 r = sigtimedwait(&waitset, &siginfo, &ts);
439 e = errno;
440 qemu_mutex_lock(&qemu_global_mutex);
442 if (r == -1 && !(e == EAGAIN || e == EINTR)) {
443 fprintf(stderr, "sigtimedwait: %s\n", strerror(e));
444 exit(1);
448 static void qemu_kvm_wait_io_event(CPUState *env)
450 while (!cpu_has_work(env))
451 qemu_cond_timedwait(env->halt_cond, &qemu_global_mutex, 1000);
453 qemu_kvm_eat_signal(env, 0);
454 qemu_wait_io_event_common(env);
457 static int qemu_cpu_exec(CPUState *env);
459 static void *kvm_cpu_thread_fn(void *arg)
461 CPUState *env = arg;
463 qemu_mutex_lock(&qemu_global_mutex);
464 qemu_thread_self(env->thread);
465 if (kvm_enabled())
466 kvm_init_vcpu(env);
468 kvm_init_ipi(env);
470 /* signal CPU creation */
471 env->created = 1;
472 qemu_cond_signal(&qemu_cpu_cond);
474 /* and wait for machine initialization */
475 while (!qemu_system_ready)
476 qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
478 while (1) {
479 if (cpu_can_run(env))
480 qemu_cpu_exec(env);
481 qemu_kvm_wait_io_event(env);
484 return NULL;
487 static void *tcg_cpu_thread_fn(void *arg)
489 CPUState *env = arg;
491 tcg_init_ipi();
492 qemu_thread_self(env->thread);
494 /* signal CPU creation */
495 qemu_mutex_lock(&qemu_global_mutex);
496 for (env = first_cpu; env != NULL; env = env->next_cpu)
497 env->created = 1;
498 qemu_cond_signal(&qemu_cpu_cond);
500 /* and wait for machine initialization */
501 while (!qemu_system_ready)
502 qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
504 while (1) {
505 tcg_cpu_exec();
506 qemu_wait_io_event(cur_cpu);
509 return NULL;
512 void qemu_cpu_kick(void *_env)
514 CPUState *env = _env;
515 qemu_cond_broadcast(env->halt_cond);
516 qemu_thread_signal(env->thread, SIG_IPI);
519 int qemu_cpu_self(void *_env)
521 CPUState *env = _env;
522 QemuThread this;
524 qemu_thread_self(&this);
526 return qemu_thread_equal(&this, env->thread);
529 static void cpu_signal(int sig)
531 if (cpu_single_env)
532 cpu_exit(cpu_single_env);
533 exit_request = 1;
536 static void tcg_init_ipi(void)
538 sigset_t set;
539 struct sigaction sigact;
541 memset(&sigact, 0, sizeof(sigact));
542 sigact.sa_handler = cpu_signal;
543 sigaction(SIG_IPI, &sigact, NULL);
545 sigemptyset(&set);
546 sigaddset(&set, SIG_IPI);
547 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
550 static void dummy_signal(int sig)
554 static void kvm_init_ipi(CPUState *env)
556 int r;
557 sigset_t set;
558 struct sigaction sigact;
560 memset(&sigact, 0, sizeof(sigact));
561 sigact.sa_handler = dummy_signal;
562 sigaction(SIG_IPI, &sigact, NULL);
564 pthread_sigmask(SIG_BLOCK, NULL, &set);
565 sigdelset(&set, SIG_IPI);
566 r = kvm_set_signal_mask(env, &set);
567 if (r) {
568 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(r));
569 exit(1);
573 static void unblock_io_signals(void)
575 sigset_t set;
577 sigemptyset(&set);
578 sigaddset(&set, SIGUSR2);
579 sigaddset(&set, SIGIO);
580 sigaddset(&set, SIGALRM);
581 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
583 sigemptyset(&set);
584 sigaddset(&set, SIG_IPI);
585 pthread_sigmask(SIG_BLOCK, &set, NULL);
588 void qemu_mutex_lock_iothread(void)
590 if (kvm_enabled()) {
591 qemu_mutex_lock(&qemu_fair_mutex);
592 qemu_mutex_lock(&qemu_global_mutex);
593 qemu_mutex_unlock(&qemu_fair_mutex);
594 } else {
595 qemu_mutex_lock(&qemu_fair_mutex);
596 if (qemu_mutex_trylock(&qemu_global_mutex)) {
597 qemu_thread_signal(tcg_cpu_thread, SIG_IPI);
598 qemu_mutex_lock(&qemu_global_mutex);
600 qemu_mutex_unlock(&qemu_fair_mutex);
604 void qemu_mutex_unlock_iothread(void)
606 qemu_mutex_unlock(&qemu_global_mutex);
609 static int all_vcpus_paused(void)
611 CPUState *penv = first_cpu;
613 while (penv) {
614 if (!penv->stopped)
615 return 0;
616 penv = (CPUState *)penv->next_cpu;
619 return 1;
622 void pause_all_vcpus(void)
624 CPUState *penv = first_cpu;
626 while (penv) {
627 penv->stop = 1;
628 qemu_cpu_kick(penv);
629 penv = (CPUState *)penv->next_cpu;
632 while (!all_vcpus_paused()) {
633 qemu_cond_timedwait(&qemu_pause_cond, &qemu_global_mutex, 100);
634 penv = first_cpu;
635 while (penv) {
636 qemu_cpu_kick(penv);
637 penv = (CPUState *)penv->next_cpu;
642 void resume_all_vcpus(void)
644 CPUState *penv = first_cpu;
646 while (penv) {
647 penv->stop = 0;
648 penv->stopped = 0;
649 qemu_cpu_kick(penv);
650 penv = (CPUState *)penv->next_cpu;
654 static void tcg_init_vcpu(void *_env)
656 CPUState *env = _env;
657 /* share a single thread for all cpus with TCG */
658 if (!tcg_cpu_thread) {
659 env->thread = qemu_mallocz(sizeof(QemuThread));
660 env->halt_cond = qemu_mallocz(sizeof(QemuCond));
661 qemu_cond_init(env->halt_cond);
662 qemu_thread_create(env->thread, tcg_cpu_thread_fn, env);
663 while (env->created == 0)
664 qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
665 tcg_cpu_thread = env->thread;
666 tcg_halt_cond = env->halt_cond;
667 } else {
668 env->thread = tcg_cpu_thread;
669 env->halt_cond = tcg_halt_cond;
673 static void kvm_start_vcpu(CPUState *env)
675 env->thread = qemu_mallocz(sizeof(QemuThread));
676 env->halt_cond = qemu_mallocz(sizeof(QemuCond));
677 qemu_cond_init(env->halt_cond);
678 qemu_thread_create(env->thread, kvm_cpu_thread_fn, env);
679 while (env->created == 0)
680 qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
683 void qemu_init_vcpu(void *_env)
685 CPUState *env = _env;
687 env->nr_cores = smp_cores;
688 env->nr_threads = smp_threads;
689 if (kvm_enabled())
690 kvm_start_vcpu(env);
691 else
692 tcg_init_vcpu(env);
695 void qemu_notify_event(void)
697 qemu_event_increment();
700 static void qemu_system_vmstop_request(int reason)
702 vmstop_requested = reason;
703 qemu_notify_event();
706 void vm_stop(int reason)
708 QemuThread me;
709 qemu_thread_self(&me);
711 if (!qemu_thread_equal(&me, &io_thread)) {
712 qemu_system_vmstop_request(reason);
714 * FIXME: should not return to device code in case
715 * vm_stop() has been requested.
717 if (cpu_single_env) {
718 cpu_exit(cpu_single_env);
719 cpu_single_env->stop = 1;
721 return;
723 do_vm_stop(reason);
726 #endif
728 static int qemu_cpu_exec(CPUState *env)
730 int ret;
731 #ifdef CONFIG_PROFILER
732 int64_t ti;
733 #endif
735 #ifdef CONFIG_PROFILER
736 ti = profile_getclock();
737 #endif
738 if (use_icount) {
739 int64_t count;
740 int decr;
741 qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
742 env->icount_decr.u16.low = 0;
743 env->icount_extra = 0;
744 count = qemu_icount_round (qemu_next_deadline());
745 qemu_icount += count;
746 decr = (count > 0xffff) ? 0xffff : count;
747 count -= decr;
748 env->icount_decr.u16.low = decr;
749 env->icount_extra = count;
751 ret = cpu_exec(env);
752 #ifdef CONFIG_PROFILER
753 qemu_time += profile_getclock() - ti;
754 #endif
755 if (use_icount) {
756 /* Fold pending instructions back into the
757 instruction counter, and clear the interrupt flag. */
758 qemu_icount -= (env->icount_decr.u16.low
759 + env->icount_extra);
760 env->icount_decr.u32 = 0;
761 env->icount_extra = 0;
763 return ret;
766 bool tcg_cpu_exec(void)
768 int ret = 0;
770 if (next_cpu == NULL)
771 next_cpu = first_cpu;
772 for (; next_cpu != NULL; next_cpu = next_cpu->next_cpu) {
773 CPUState *env = cur_cpu = next_cpu;
775 qemu_clock_enable(vm_clock,
776 (cur_cpu->singlestep_enabled & SSTEP_NOTIMER) == 0);
778 if (qemu_alarm_pending())
779 break;
780 if (cpu_can_run(env))
781 ret = qemu_cpu_exec(env);
782 else if (env->stop)
783 break;
785 if (ret == EXCP_DEBUG) {
786 gdb_set_stop_cpu(env);
787 debug_requested = EXCP_DEBUG;
788 break;
791 return tcg_has_work();
794 void set_numa_modes(void)
796 CPUState *env;
797 int i;
799 for (env = first_cpu; env != NULL; env = env->next_cpu) {
800 for (i = 0; i < nb_numa_nodes; i++) {
801 if (node_cpumask[i] & (1 << env->cpu_index)) {
802 env->numa_node = i;
808 void set_cpu_log(const char *optarg)
810 int mask;
811 const CPULogItem *item;
813 mask = cpu_str_to_log_mask(optarg);
814 if (!mask) {
815 printf("Log items (comma separated):\n");
816 for (item = cpu_log_items; item->mask != 0; item++) {
817 printf("%-10s %s\n", item->name, item->help);
819 exit(1);
821 cpu_set_log(mask);
824 /* Return the virtual CPU time, based on the instruction counter. */
825 int64_t cpu_get_icount(void)
827 int64_t icount;
828 CPUState *env = cpu_single_env;;
830 icount = qemu_icount;
831 if (env) {
832 if (!can_do_io(env)) {
833 fprintf(stderr, "Bad clock read\n");
835 icount -= (env->icount_decr.u16.low + env->icount_extra);
837 return qemu_icount_bias + (icount << icount_time_shift);
840 void list_cpus(FILE *f, int (*cpu_fprintf)(FILE *f, const char *fmt, ...),
841 const char *optarg)
843 /* XXX: implement xxx_cpu_list for targets that still miss it */
844 #if defined(cpu_list_id)
845 cpu_list_id(f, cpu_fprintf, optarg);
846 #elif defined(cpu_list)
847 cpu_list(f, cpu_fprintf); /* deprecated */
848 #endif