]> git.proxmox.com Git - qemu.git/blob - cpus.c
Merge remote-tracking branch 'kiszka/queues/slirp' into staging
[qemu.git] / cpus.c
1 /*
2 * QEMU System Emulator
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 *
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:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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.
23 */
24
25 /* Needed early for CONFIG_BSD etc. */
26 #include "config-host.h"
27
28 #include "monitor/monitor.h"
29 #include "sysemu/sysemu.h"
30 #include "exec/gdbstub.h"
31 #include "sysemu/dma.h"
32 #include "sysemu/kvm.h"
33 #include "qmp-commands.h"
34
35 #include "qemu/thread.h"
36 #include "sysemu/cpus.h"
37 #include "sysemu/qtest.h"
38 #include "qemu/main-loop.h"
39 #include "qemu/bitmap.h"
40
41 #ifndef _WIN32
42 #include "qemu/compatfd.h"
43 #endif
44
45 #ifdef CONFIG_LINUX
46
47 #include <sys/prctl.h>
48
49 #ifndef PR_MCE_KILL
50 #define PR_MCE_KILL 33
51 #endif
52
53 #ifndef PR_MCE_KILL_SET
54 #define PR_MCE_KILL_SET 1
55 #endif
56
57 #ifndef PR_MCE_KILL_EARLY
58 #define PR_MCE_KILL_EARLY 1
59 #endif
60
61 #endif /* CONFIG_LINUX */
62
63 static CPUState *next_cpu;
64
65 bool cpu_is_stopped(CPUState *cpu)
66 {
67 return cpu->stopped || !runstate_is_running();
68 }
69
70 static bool cpu_thread_is_idle(CPUState *cpu)
71 {
72 if (cpu->stop || cpu->queued_work_first) {
73 return false;
74 }
75 if (cpu_is_stopped(cpu)) {
76 return true;
77 }
78 if (!cpu->halted || qemu_cpu_has_work(cpu) ||
79 kvm_halt_in_kernel()) {
80 return false;
81 }
82 return true;
83 }
84
85 static bool all_cpu_threads_idle(void)
86 {
87 CPUState *cpu;
88
89 CPU_FOREACH(cpu) {
90 if (!cpu_thread_is_idle(cpu)) {
91 return false;
92 }
93 }
94 return true;
95 }
96
97 /***********************************************************/
98 /* guest cycle counter */
99
100 /* Conversion factor from emulated instructions to virtual clock ticks. */
101 static int icount_time_shift;
102 /* Arbitrarily pick 1MIPS as the minimum allowable speed. */
103 #define MAX_ICOUNT_SHIFT 10
104 /* Compensate for varying guest execution speed. */
105 static int64_t qemu_icount_bias;
106 static QEMUTimer *icount_rt_timer;
107 static QEMUTimer *icount_vm_timer;
108 static QEMUTimer *icount_warp_timer;
109 static int64_t vm_clock_warp_start;
110 static int64_t qemu_icount;
111
112 typedef struct TimersState {
113 int64_t cpu_ticks_prev;
114 int64_t cpu_ticks_offset;
115 int64_t cpu_clock_offset;
116 int32_t cpu_ticks_enabled;
117 int64_t dummy;
118 } TimersState;
119
120 static TimersState timers_state;
121
122 /* Return the virtual CPU time, based on the instruction counter. */
123 int64_t cpu_get_icount(void)
124 {
125 int64_t icount;
126 CPUState *cpu = current_cpu;
127
128 icount = qemu_icount;
129 if (cpu) {
130 CPUArchState *env = cpu->env_ptr;
131 if (!can_do_io(env)) {
132 fprintf(stderr, "Bad clock read\n");
133 }
134 icount -= (env->icount_decr.u16.low + env->icount_extra);
135 }
136 return qemu_icount_bias + (icount << icount_time_shift);
137 }
138
139 /* return the host CPU cycle counter and handle stop/restart */
140 int64_t cpu_get_ticks(void)
141 {
142 if (use_icount) {
143 return cpu_get_icount();
144 }
145 if (!timers_state.cpu_ticks_enabled) {
146 return timers_state.cpu_ticks_offset;
147 } else {
148 int64_t ticks;
149 ticks = cpu_get_real_ticks();
150 if (timers_state.cpu_ticks_prev > ticks) {
151 /* Note: non increasing ticks may happen if the host uses
152 software suspend */
153 timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks;
154 }
155 timers_state.cpu_ticks_prev = ticks;
156 return ticks + timers_state.cpu_ticks_offset;
157 }
158 }
159
160 /* return the host CPU monotonic timer and handle stop/restart */
161 int64_t cpu_get_clock(void)
162 {
163 int64_t ti;
164 if (!timers_state.cpu_ticks_enabled) {
165 return timers_state.cpu_clock_offset;
166 } else {
167 ti = get_clock();
168 return ti + timers_state.cpu_clock_offset;
169 }
170 }
171
172 /* enable cpu_get_ticks() */
173 void cpu_enable_ticks(void)
174 {
175 if (!timers_state.cpu_ticks_enabled) {
176 timers_state.cpu_ticks_offset -= cpu_get_real_ticks();
177 timers_state.cpu_clock_offset -= get_clock();
178 timers_state.cpu_ticks_enabled = 1;
179 }
180 }
181
182 /* disable cpu_get_ticks() : the clock is stopped. You must not call
183 cpu_get_ticks() after that. */
184 void cpu_disable_ticks(void)
185 {
186 if (timers_state.cpu_ticks_enabled) {
187 timers_state.cpu_ticks_offset = cpu_get_ticks();
188 timers_state.cpu_clock_offset = cpu_get_clock();
189 timers_state.cpu_ticks_enabled = 0;
190 }
191 }
192
193 /* Correlation between real and virtual time is always going to be
194 fairly approximate, so ignore small variation.
195 When the guest is idle real and virtual time will be aligned in
196 the IO wait loop. */
197 #define ICOUNT_WOBBLE (get_ticks_per_sec() / 10)
198
199 static void icount_adjust(void)
200 {
201 int64_t cur_time;
202 int64_t cur_icount;
203 int64_t delta;
204 static int64_t last_delta;
205 /* If the VM is not running, then do nothing. */
206 if (!runstate_is_running()) {
207 return;
208 }
209 cur_time = cpu_get_clock();
210 cur_icount = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
211 delta = cur_icount - cur_time;
212 /* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */
213 if (delta > 0
214 && last_delta + ICOUNT_WOBBLE < delta * 2
215 && icount_time_shift > 0) {
216 /* The guest is getting too far ahead. Slow time down. */
217 icount_time_shift--;
218 }
219 if (delta < 0
220 && last_delta - ICOUNT_WOBBLE > delta * 2
221 && icount_time_shift < MAX_ICOUNT_SHIFT) {
222 /* The guest is getting too far behind. Speed time up. */
223 icount_time_shift++;
224 }
225 last_delta = delta;
226 qemu_icount_bias = cur_icount - (qemu_icount << icount_time_shift);
227 }
228
229 static void icount_adjust_rt(void *opaque)
230 {
231 timer_mod(icount_rt_timer,
232 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + 1000);
233 icount_adjust();
234 }
235
236 static void icount_adjust_vm(void *opaque)
237 {
238 timer_mod(icount_vm_timer,
239 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
240 get_ticks_per_sec() / 10);
241 icount_adjust();
242 }
243
244 static int64_t qemu_icount_round(int64_t count)
245 {
246 return (count + (1 << icount_time_shift) - 1) >> icount_time_shift;
247 }
248
249 static void icount_warp_rt(void *opaque)
250 {
251 if (vm_clock_warp_start == -1) {
252 return;
253 }
254
255 if (runstate_is_running()) {
256 int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
257 int64_t warp_delta = clock - vm_clock_warp_start;
258 if (use_icount == 1) {
259 qemu_icount_bias += warp_delta;
260 } else {
261 /*
262 * In adaptive mode, do not let QEMU_CLOCK_VIRTUAL run too
263 * far ahead of real time.
264 */
265 int64_t cur_time = cpu_get_clock();
266 int64_t cur_icount = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
267 int64_t delta = cur_time - cur_icount;
268 qemu_icount_bias += MIN(warp_delta, delta);
269 }
270 if (qemu_clock_expired(QEMU_CLOCK_VIRTUAL)) {
271 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
272 }
273 }
274 vm_clock_warp_start = -1;
275 }
276
277 void qtest_clock_warp(int64_t dest)
278 {
279 int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
280 assert(qtest_enabled());
281 while (clock < dest) {
282 int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
283 int64_t warp = MIN(dest - clock, deadline);
284 qemu_icount_bias += warp;
285 qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL);
286 clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
287 }
288 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
289 }
290
291 void qemu_clock_warp(QEMUClockType type)
292 {
293 int64_t deadline;
294
295 /*
296 * There are too many global variables to make the "warp" behavior
297 * applicable to other clocks. But a clock argument removes the
298 * need for if statements all over the place.
299 */
300 if (type != QEMU_CLOCK_VIRTUAL || !use_icount) {
301 return;
302 }
303
304 /*
305 * If the CPUs have been sleeping, advance QEMU_CLOCK_VIRTUAL timer now.
306 * This ensures that the deadline for the timer is computed correctly below.
307 * This also makes sure that the insn counter is synchronized before the
308 * CPU starts running, in case the CPU is woken by an event other than
309 * the earliest QEMU_CLOCK_VIRTUAL timer.
310 */
311 icount_warp_rt(NULL);
312 if (!all_cpu_threads_idle() || !qemu_clock_has_timers(QEMU_CLOCK_VIRTUAL)) {
313 timer_del(icount_warp_timer);
314 return;
315 }
316
317 if (qtest_enabled()) {
318 /* When testing, qtest commands advance icount. */
319 return;
320 }
321
322 vm_clock_warp_start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
323 /* We want to use the earliest deadline from ALL vm_clocks */
324 deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
325
326 /* Maintain prior (possibly buggy) behaviour where if no deadline
327 * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than
328 * INT32_MAX nanoseconds ahead, we still use INT32_MAX
329 * nanoseconds.
330 */
331 if ((deadline < 0) || (deadline > INT32_MAX)) {
332 deadline = INT32_MAX;
333 }
334
335 if (deadline > 0) {
336 /*
337 * Ensure QEMU_CLOCK_VIRTUAL proceeds even when the virtual CPU goes to
338 * sleep. Otherwise, the CPU might be waiting for a future timer
339 * interrupt to wake it up, but the interrupt never comes because
340 * the vCPU isn't running any insns and thus doesn't advance the
341 * QEMU_CLOCK_VIRTUAL.
342 *
343 * An extreme solution for this problem would be to never let VCPUs
344 * sleep in icount mode if there is a pending QEMU_CLOCK_VIRTUAL
345 * timer; rather time could just advance to the next QEMU_CLOCK_VIRTUAL
346 * event. Instead, we do stop VCPUs and only advance QEMU_CLOCK_VIRTUAL
347 * after some e"real" time, (related to the time left until the next
348 * event) has passed. The QEMU_CLOCK_REALTIME timer will do this.
349 * This avoids that the warps are visible externally; for example,
350 * you will not be sending network packets continuously instead of
351 * every 100ms.
352 */
353 timer_mod(icount_warp_timer, vm_clock_warp_start + deadline);
354 } else if (deadline == 0) {
355 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
356 }
357 }
358
359 static const VMStateDescription vmstate_timers = {
360 .name = "timer",
361 .version_id = 2,
362 .minimum_version_id = 1,
363 .minimum_version_id_old = 1,
364 .fields = (VMStateField[]) {
365 VMSTATE_INT64(cpu_ticks_offset, TimersState),
366 VMSTATE_INT64(dummy, TimersState),
367 VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
368 VMSTATE_END_OF_LIST()
369 }
370 };
371
372 void configure_icount(const char *option)
373 {
374 vmstate_register(NULL, 0, &vmstate_timers, &timers_state);
375 if (!option) {
376 return;
377 }
378
379 icount_warp_timer = timer_new_ns(QEMU_CLOCK_REALTIME,
380 icount_warp_rt, NULL);
381 if (strcmp(option, "auto") != 0) {
382 icount_time_shift = strtol(option, NULL, 0);
383 use_icount = 1;
384 return;
385 }
386
387 use_icount = 2;
388
389 /* 125MIPS seems a reasonable initial guess at the guest speed.
390 It will be corrected fairly quickly anyway. */
391 icount_time_shift = 3;
392
393 /* Have both realtime and virtual time triggers for speed adjustment.
394 The realtime trigger catches emulated time passing too slowly,
395 the virtual time trigger catches emulated time passing too fast.
396 Realtime triggers occur even when idle, so use them less frequently
397 than VM triggers. */
398 icount_rt_timer = timer_new_ms(QEMU_CLOCK_REALTIME,
399 icount_adjust_rt, NULL);
400 timer_mod(icount_rt_timer,
401 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + 1000);
402 icount_vm_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
403 icount_adjust_vm, NULL);
404 timer_mod(icount_vm_timer,
405 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) +
406 get_ticks_per_sec() / 10);
407 }
408
409 /***********************************************************/
410 void hw_error(const char *fmt, ...)
411 {
412 va_list ap;
413 CPUState *cpu;
414
415 va_start(ap, fmt);
416 fprintf(stderr, "qemu: hardware error: ");
417 vfprintf(stderr, fmt, ap);
418 fprintf(stderr, "\n");
419 CPU_FOREACH(cpu) {
420 fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
421 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU);
422 }
423 va_end(ap);
424 abort();
425 }
426
427 void cpu_synchronize_all_states(void)
428 {
429 CPUState *cpu;
430
431 CPU_FOREACH(cpu) {
432 cpu_synchronize_state(cpu);
433 }
434 }
435
436 void cpu_synchronize_all_post_reset(void)
437 {
438 CPUState *cpu;
439
440 CPU_FOREACH(cpu) {
441 cpu_synchronize_post_reset(cpu);
442 }
443 }
444
445 void cpu_synchronize_all_post_init(void)
446 {
447 CPUState *cpu;
448
449 CPU_FOREACH(cpu) {
450 cpu_synchronize_post_init(cpu);
451 }
452 }
453
454 static int do_vm_stop(RunState state)
455 {
456 int ret = 0;
457
458 if (runstate_is_running()) {
459 cpu_disable_ticks();
460 pause_all_vcpus();
461 runstate_set(state);
462 vm_state_notify(0, state);
463 monitor_protocol_event(QEVENT_STOP, NULL);
464 }
465
466 bdrv_drain_all();
467 ret = bdrv_flush_all();
468
469 return ret;
470 }
471
472 static bool cpu_can_run(CPUState *cpu)
473 {
474 if (cpu->stop) {
475 return false;
476 }
477 if (cpu_is_stopped(cpu)) {
478 return false;
479 }
480 return true;
481 }
482
483 static void cpu_handle_guest_debug(CPUState *cpu)
484 {
485 gdb_set_stop_cpu(cpu);
486 qemu_system_debug_request();
487 cpu->stopped = true;
488 }
489
490 static void cpu_signal(int sig)
491 {
492 if (current_cpu) {
493 cpu_exit(current_cpu);
494 }
495 exit_request = 1;
496 }
497
498 #ifdef CONFIG_LINUX
499 static void sigbus_reraise(void)
500 {
501 sigset_t set;
502 struct sigaction action;
503
504 memset(&action, 0, sizeof(action));
505 action.sa_handler = SIG_DFL;
506 if (!sigaction(SIGBUS, &action, NULL)) {
507 raise(SIGBUS);
508 sigemptyset(&set);
509 sigaddset(&set, SIGBUS);
510 sigprocmask(SIG_UNBLOCK, &set, NULL);
511 }
512 perror("Failed to re-raise SIGBUS!\n");
513 abort();
514 }
515
516 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
517 void *ctx)
518 {
519 if (kvm_on_sigbus(siginfo->ssi_code,
520 (void *)(intptr_t)siginfo->ssi_addr)) {
521 sigbus_reraise();
522 }
523 }
524
525 static void qemu_init_sigbus(void)
526 {
527 struct sigaction action;
528
529 memset(&action, 0, sizeof(action));
530 action.sa_flags = SA_SIGINFO;
531 action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
532 sigaction(SIGBUS, &action, NULL);
533
534 prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
535 }
536
537 static void qemu_kvm_eat_signals(CPUState *cpu)
538 {
539 struct timespec ts = { 0, 0 };
540 siginfo_t siginfo;
541 sigset_t waitset;
542 sigset_t chkset;
543 int r;
544
545 sigemptyset(&waitset);
546 sigaddset(&waitset, SIG_IPI);
547 sigaddset(&waitset, SIGBUS);
548
549 do {
550 r = sigtimedwait(&waitset, &siginfo, &ts);
551 if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
552 perror("sigtimedwait");
553 exit(1);
554 }
555
556 switch (r) {
557 case SIGBUS:
558 if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) {
559 sigbus_reraise();
560 }
561 break;
562 default:
563 break;
564 }
565
566 r = sigpending(&chkset);
567 if (r == -1) {
568 perror("sigpending");
569 exit(1);
570 }
571 } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
572 }
573
574 #else /* !CONFIG_LINUX */
575
576 static void qemu_init_sigbus(void)
577 {
578 }
579
580 static void qemu_kvm_eat_signals(CPUState *cpu)
581 {
582 }
583 #endif /* !CONFIG_LINUX */
584
585 #ifndef _WIN32
586 static void dummy_signal(int sig)
587 {
588 }
589
590 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
591 {
592 int r;
593 sigset_t set;
594 struct sigaction sigact;
595
596 memset(&sigact, 0, sizeof(sigact));
597 sigact.sa_handler = dummy_signal;
598 sigaction(SIG_IPI, &sigact, NULL);
599
600 pthread_sigmask(SIG_BLOCK, NULL, &set);
601 sigdelset(&set, SIG_IPI);
602 sigdelset(&set, SIGBUS);
603 r = kvm_set_signal_mask(cpu, &set);
604 if (r) {
605 fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
606 exit(1);
607 }
608 }
609
610 static void qemu_tcg_init_cpu_signals(void)
611 {
612 sigset_t set;
613 struct sigaction sigact;
614
615 memset(&sigact, 0, sizeof(sigact));
616 sigact.sa_handler = cpu_signal;
617 sigaction(SIG_IPI, &sigact, NULL);
618
619 sigemptyset(&set);
620 sigaddset(&set, SIG_IPI);
621 pthread_sigmask(SIG_UNBLOCK, &set, NULL);
622 }
623
624 #else /* _WIN32 */
625 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
626 {
627 abort();
628 }
629
630 static void qemu_tcg_init_cpu_signals(void)
631 {
632 }
633 #endif /* _WIN32 */
634
635 static QemuMutex qemu_global_mutex;
636 static QemuCond qemu_io_proceeded_cond;
637 static bool iothread_requesting_mutex;
638
639 static QemuThread io_thread;
640
641 static QemuThread *tcg_cpu_thread;
642 static QemuCond *tcg_halt_cond;
643
644 /* cpu creation */
645 static QemuCond qemu_cpu_cond;
646 /* system init */
647 static QemuCond qemu_pause_cond;
648 static QemuCond qemu_work_cond;
649
650 void qemu_init_cpu_loop(void)
651 {
652 qemu_init_sigbus();
653 qemu_cond_init(&qemu_cpu_cond);
654 qemu_cond_init(&qemu_pause_cond);
655 qemu_cond_init(&qemu_work_cond);
656 qemu_cond_init(&qemu_io_proceeded_cond);
657 qemu_mutex_init(&qemu_global_mutex);
658
659 qemu_thread_get_self(&io_thread);
660 }
661
662 void run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
663 {
664 struct qemu_work_item wi;
665
666 if (qemu_cpu_is_self(cpu)) {
667 func(data);
668 return;
669 }
670
671 wi.func = func;
672 wi.data = data;
673 wi.free = false;
674 if (cpu->queued_work_first == NULL) {
675 cpu->queued_work_first = &wi;
676 } else {
677 cpu->queued_work_last->next = &wi;
678 }
679 cpu->queued_work_last = &wi;
680 wi.next = NULL;
681 wi.done = false;
682
683 qemu_cpu_kick(cpu);
684 while (!wi.done) {
685 CPUState *self_cpu = current_cpu;
686
687 qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
688 current_cpu = self_cpu;
689 }
690 }
691
692 void async_run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
693 {
694 struct qemu_work_item *wi;
695
696 if (qemu_cpu_is_self(cpu)) {
697 func(data);
698 return;
699 }
700
701 wi = g_malloc0(sizeof(struct qemu_work_item));
702 wi->func = func;
703 wi->data = data;
704 wi->free = true;
705 if (cpu->queued_work_first == NULL) {
706 cpu->queued_work_first = wi;
707 } else {
708 cpu->queued_work_last->next = wi;
709 }
710 cpu->queued_work_last = wi;
711 wi->next = NULL;
712 wi->done = false;
713
714 qemu_cpu_kick(cpu);
715 }
716
717 static void flush_queued_work(CPUState *cpu)
718 {
719 struct qemu_work_item *wi;
720
721 if (cpu->queued_work_first == NULL) {
722 return;
723 }
724
725 while ((wi = cpu->queued_work_first)) {
726 cpu->queued_work_first = wi->next;
727 wi->func(wi->data);
728 wi->done = true;
729 if (wi->free) {
730 g_free(wi);
731 }
732 }
733 cpu->queued_work_last = NULL;
734 qemu_cond_broadcast(&qemu_work_cond);
735 }
736
737 static void qemu_wait_io_event_common(CPUState *cpu)
738 {
739 if (cpu->stop) {
740 cpu->stop = false;
741 cpu->stopped = true;
742 qemu_cond_signal(&qemu_pause_cond);
743 }
744 flush_queued_work(cpu);
745 cpu->thread_kicked = false;
746 }
747
748 static void qemu_tcg_wait_io_event(void)
749 {
750 CPUState *cpu;
751
752 while (all_cpu_threads_idle()) {
753 /* Start accounting real time to the virtual clock if the CPUs
754 are idle. */
755 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
756 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
757 }
758
759 while (iothread_requesting_mutex) {
760 qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex);
761 }
762
763 CPU_FOREACH(cpu) {
764 qemu_wait_io_event_common(cpu);
765 }
766 }
767
768 static void qemu_kvm_wait_io_event(CPUState *cpu)
769 {
770 while (cpu_thread_is_idle(cpu)) {
771 qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
772 }
773
774 qemu_kvm_eat_signals(cpu);
775 qemu_wait_io_event_common(cpu);
776 }
777
778 static void *qemu_kvm_cpu_thread_fn(void *arg)
779 {
780 CPUState *cpu = arg;
781 int r;
782
783 qemu_mutex_lock(&qemu_global_mutex);
784 qemu_thread_get_self(cpu->thread);
785 cpu->thread_id = qemu_get_thread_id();
786 current_cpu = cpu;
787
788 r = kvm_init_vcpu(cpu);
789 if (r < 0) {
790 fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
791 exit(1);
792 }
793
794 qemu_kvm_init_cpu_signals(cpu);
795
796 /* signal CPU creation */
797 cpu->created = true;
798 qemu_cond_signal(&qemu_cpu_cond);
799
800 while (1) {
801 if (cpu_can_run(cpu)) {
802 r = kvm_cpu_exec(cpu);
803 if (r == EXCP_DEBUG) {
804 cpu_handle_guest_debug(cpu);
805 }
806 }
807 qemu_kvm_wait_io_event(cpu);
808 }
809
810 return NULL;
811 }
812
813 static void *qemu_dummy_cpu_thread_fn(void *arg)
814 {
815 #ifdef _WIN32
816 fprintf(stderr, "qtest is not supported under Windows\n");
817 exit(1);
818 #else
819 CPUState *cpu = arg;
820 sigset_t waitset;
821 int r;
822
823 qemu_mutex_lock_iothread();
824 qemu_thread_get_self(cpu->thread);
825 cpu->thread_id = qemu_get_thread_id();
826
827 sigemptyset(&waitset);
828 sigaddset(&waitset, SIG_IPI);
829
830 /* signal CPU creation */
831 cpu->created = true;
832 qemu_cond_signal(&qemu_cpu_cond);
833
834 current_cpu = cpu;
835 while (1) {
836 current_cpu = NULL;
837 qemu_mutex_unlock_iothread();
838 do {
839 int sig;
840 r = sigwait(&waitset, &sig);
841 } while (r == -1 && (errno == EAGAIN || errno == EINTR));
842 if (r == -1) {
843 perror("sigwait");
844 exit(1);
845 }
846 qemu_mutex_lock_iothread();
847 current_cpu = cpu;
848 qemu_wait_io_event_common(cpu);
849 }
850
851 return NULL;
852 #endif
853 }
854
855 static void tcg_exec_all(void);
856
857 static void *qemu_tcg_cpu_thread_fn(void *arg)
858 {
859 CPUState *cpu = arg;
860
861 qemu_tcg_init_cpu_signals();
862 qemu_thread_get_self(cpu->thread);
863
864 qemu_mutex_lock(&qemu_global_mutex);
865 CPU_FOREACH(cpu) {
866 cpu->thread_id = qemu_get_thread_id();
867 cpu->created = true;
868 }
869 qemu_cond_signal(&qemu_cpu_cond);
870
871 /* wait for initial kick-off after machine start */
872 while (QTAILQ_FIRST(&cpus)->stopped) {
873 qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
874
875 /* process any pending work */
876 CPU_FOREACH(cpu) {
877 qemu_wait_io_event_common(cpu);
878 }
879 }
880
881 while (1) {
882 tcg_exec_all();
883
884 if (use_icount) {
885 int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
886
887 if (deadline == 0) {
888 qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
889 }
890 }
891 qemu_tcg_wait_io_event();
892 }
893
894 return NULL;
895 }
896
897 static void qemu_cpu_kick_thread(CPUState *cpu)
898 {
899 #ifndef _WIN32
900 int err;
901
902 err = pthread_kill(cpu->thread->thread, SIG_IPI);
903 if (err) {
904 fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
905 exit(1);
906 }
907 #else /* _WIN32 */
908 if (!qemu_cpu_is_self(cpu)) {
909 CONTEXT tcgContext;
910
911 if (SuspendThread(cpu->hThread) == (DWORD)-1) {
912 fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
913 GetLastError());
914 exit(1);
915 }
916
917 /* On multi-core systems, we are not sure that the thread is actually
918 * suspended until we can get the context.
919 */
920 tcgContext.ContextFlags = CONTEXT_CONTROL;
921 while (GetThreadContext(cpu->hThread, &tcgContext) != 0) {
922 continue;
923 }
924
925 cpu_signal(0);
926
927 if (ResumeThread(cpu->hThread) == (DWORD)-1) {
928 fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
929 GetLastError());
930 exit(1);
931 }
932 }
933 #endif
934 }
935
936 void qemu_cpu_kick(CPUState *cpu)
937 {
938 qemu_cond_broadcast(cpu->halt_cond);
939 if (!tcg_enabled() && !cpu->thread_kicked) {
940 qemu_cpu_kick_thread(cpu);
941 cpu->thread_kicked = true;
942 }
943 }
944
945 void qemu_cpu_kick_self(void)
946 {
947 #ifndef _WIN32
948 assert(current_cpu);
949
950 if (!current_cpu->thread_kicked) {
951 qemu_cpu_kick_thread(current_cpu);
952 current_cpu->thread_kicked = true;
953 }
954 #else
955 abort();
956 #endif
957 }
958
959 bool qemu_cpu_is_self(CPUState *cpu)
960 {
961 return qemu_thread_is_self(cpu->thread);
962 }
963
964 static bool qemu_in_vcpu_thread(void)
965 {
966 return current_cpu && qemu_cpu_is_self(current_cpu);
967 }
968
969 void qemu_mutex_lock_iothread(void)
970 {
971 if (!tcg_enabled()) {
972 qemu_mutex_lock(&qemu_global_mutex);
973 } else {
974 iothread_requesting_mutex = true;
975 if (qemu_mutex_trylock(&qemu_global_mutex)) {
976 qemu_cpu_kick_thread(first_cpu);
977 qemu_mutex_lock(&qemu_global_mutex);
978 }
979 iothread_requesting_mutex = false;
980 qemu_cond_broadcast(&qemu_io_proceeded_cond);
981 }
982 }
983
984 void qemu_mutex_unlock_iothread(void)
985 {
986 qemu_mutex_unlock(&qemu_global_mutex);
987 }
988
989 static int all_vcpus_paused(void)
990 {
991 CPUState *cpu;
992
993 CPU_FOREACH(cpu) {
994 if (!cpu->stopped) {
995 return 0;
996 }
997 }
998
999 return 1;
1000 }
1001
1002 void pause_all_vcpus(void)
1003 {
1004 CPUState *cpu;
1005
1006 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false);
1007 CPU_FOREACH(cpu) {
1008 cpu->stop = true;
1009 qemu_cpu_kick(cpu);
1010 }
1011
1012 if (qemu_in_vcpu_thread()) {
1013 cpu_stop_current();
1014 if (!kvm_enabled()) {
1015 CPU_FOREACH(cpu) {
1016 cpu->stop = false;
1017 cpu->stopped = true;
1018 }
1019 return;
1020 }
1021 }
1022
1023 while (!all_vcpus_paused()) {
1024 qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
1025 CPU_FOREACH(cpu) {
1026 qemu_cpu_kick(cpu);
1027 }
1028 }
1029 }
1030
1031 void cpu_resume(CPUState *cpu)
1032 {
1033 cpu->stop = false;
1034 cpu->stopped = false;
1035 qemu_cpu_kick(cpu);
1036 }
1037
1038 void resume_all_vcpus(void)
1039 {
1040 CPUState *cpu;
1041
1042 qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true);
1043 CPU_FOREACH(cpu) {
1044 cpu_resume(cpu);
1045 }
1046 }
1047
1048 static void qemu_tcg_init_vcpu(CPUState *cpu)
1049 {
1050 /* share a single thread for all cpus with TCG */
1051 if (!tcg_cpu_thread) {
1052 cpu->thread = g_malloc0(sizeof(QemuThread));
1053 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1054 qemu_cond_init(cpu->halt_cond);
1055 tcg_halt_cond = cpu->halt_cond;
1056 qemu_thread_create(cpu->thread, qemu_tcg_cpu_thread_fn, cpu,
1057 QEMU_THREAD_JOINABLE);
1058 #ifdef _WIN32
1059 cpu->hThread = qemu_thread_get_handle(cpu->thread);
1060 #endif
1061 while (!cpu->created) {
1062 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1063 }
1064 tcg_cpu_thread = cpu->thread;
1065 } else {
1066 cpu->thread = tcg_cpu_thread;
1067 cpu->halt_cond = tcg_halt_cond;
1068 }
1069 }
1070
1071 static void qemu_kvm_start_vcpu(CPUState *cpu)
1072 {
1073 cpu->thread = g_malloc0(sizeof(QemuThread));
1074 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1075 qemu_cond_init(cpu->halt_cond);
1076 qemu_thread_create(cpu->thread, qemu_kvm_cpu_thread_fn, cpu,
1077 QEMU_THREAD_JOINABLE);
1078 while (!cpu->created) {
1079 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1080 }
1081 }
1082
1083 static void qemu_dummy_start_vcpu(CPUState *cpu)
1084 {
1085 cpu->thread = g_malloc0(sizeof(QemuThread));
1086 cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1087 qemu_cond_init(cpu->halt_cond);
1088 qemu_thread_create(cpu->thread, qemu_dummy_cpu_thread_fn, cpu,
1089 QEMU_THREAD_JOINABLE);
1090 while (!cpu->created) {
1091 qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1092 }
1093 }
1094
1095 void qemu_init_vcpu(CPUState *cpu)
1096 {
1097 cpu->nr_cores = smp_cores;
1098 cpu->nr_threads = smp_threads;
1099 cpu->stopped = true;
1100 if (kvm_enabled()) {
1101 qemu_kvm_start_vcpu(cpu);
1102 } else if (tcg_enabled()) {
1103 qemu_tcg_init_vcpu(cpu);
1104 } else {
1105 qemu_dummy_start_vcpu(cpu);
1106 }
1107 }
1108
1109 void cpu_stop_current(void)
1110 {
1111 if (current_cpu) {
1112 current_cpu->stop = false;
1113 current_cpu->stopped = true;
1114 cpu_exit(current_cpu);
1115 qemu_cond_signal(&qemu_pause_cond);
1116 }
1117 }
1118
1119 int vm_stop(RunState state)
1120 {
1121 if (qemu_in_vcpu_thread()) {
1122 qemu_system_vmstop_request(state);
1123 /*
1124 * FIXME: should not return to device code in case
1125 * vm_stop() has been requested.
1126 */
1127 cpu_stop_current();
1128 return 0;
1129 }
1130
1131 return do_vm_stop(state);
1132 }
1133
1134 /* does a state transition even if the VM is already stopped,
1135 current state is forgotten forever */
1136 int vm_stop_force_state(RunState state)
1137 {
1138 if (runstate_is_running()) {
1139 return vm_stop(state);
1140 } else {
1141 runstate_set(state);
1142 /* Make sure to return an error if the flush in a previous vm_stop()
1143 * failed. */
1144 return bdrv_flush_all();
1145 }
1146 }
1147
1148 static int tcg_cpu_exec(CPUArchState *env)
1149 {
1150 int ret;
1151 #ifdef CONFIG_PROFILER
1152 int64_t ti;
1153 #endif
1154
1155 #ifdef CONFIG_PROFILER
1156 ti = profile_getclock();
1157 #endif
1158 if (use_icount) {
1159 int64_t count;
1160 int64_t deadline;
1161 int decr;
1162 qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
1163 env->icount_decr.u16.low = 0;
1164 env->icount_extra = 0;
1165 deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
1166
1167 /* Maintain prior (possibly buggy) behaviour where if no deadline
1168 * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than
1169 * INT32_MAX nanoseconds ahead, we still use INT32_MAX
1170 * nanoseconds.
1171 */
1172 if ((deadline < 0) || (deadline > INT32_MAX)) {
1173 deadline = INT32_MAX;
1174 }
1175
1176 count = qemu_icount_round(deadline);
1177 qemu_icount += count;
1178 decr = (count > 0xffff) ? 0xffff : count;
1179 count -= decr;
1180 env->icount_decr.u16.low = decr;
1181 env->icount_extra = count;
1182 }
1183 ret = cpu_exec(env);
1184 #ifdef CONFIG_PROFILER
1185 qemu_time += profile_getclock() - ti;
1186 #endif
1187 if (use_icount) {
1188 /* Fold pending instructions back into the
1189 instruction counter, and clear the interrupt flag. */
1190 qemu_icount -= (env->icount_decr.u16.low
1191 + env->icount_extra);
1192 env->icount_decr.u32 = 0;
1193 env->icount_extra = 0;
1194 }
1195 return ret;
1196 }
1197
1198 static void tcg_exec_all(void)
1199 {
1200 int r;
1201
1202 /* Account partial waits to QEMU_CLOCK_VIRTUAL. */
1203 qemu_clock_warp(QEMU_CLOCK_VIRTUAL);
1204
1205 if (next_cpu == NULL) {
1206 next_cpu = first_cpu;
1207 }
1208 for (; next_cpu != NULL && !exit_request; next_cpu = CPU_NEXT(next_cpu)) {
1209 CPUState *cpu = next_cpu;
1210 CPUArchState *env = cpu->env_ptr;
1211
1212 qemu_clock_enable(QEMU_CLOCK_VIRTUAL,
1213 (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0);
1214
1215 if (cpu_can_run(cpu)) {
1216 r = tcg_cpu_exec(env);
1217 if (r == EXCP_DEBUG) {
1218 cpu_handle_guest_debug(cpu);
1219 break;
1220 }
1221 } else if (cpu->stop || cpu->stopped) {
1222 break;
1223 }
1224 }
1225 exit_request = 0;
1226 }
1227
1228 void set_numa_modes(void)
1229 {
1230 CPUState *cpu;
1231 int i;
1232
1233 CPU_FOREACH(cpu) {
1234 for (i = 0; i < nb_numa_nodes; i++) {
1235 if (test_bit(cpu->cpu_index, node_cpumask[i])) {
1236 cpu->numa_node = i;
1237 }
1238 }
1239 }
1240 }
1241
1242 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1243 {
1244 /* XXX: implement xxx_cpu_list for targets that still miss it */
1245 #if defined(cpu_list)
1246 cpu_list(f, cpu_fprintf);
1247 #endif
1248 }
1249
1250 CpuInfoList *qmp_query_cpus(Error **errp)
1251 {
1252 CpuInfoList *head = NULL, *cur_item = NULL;
1253 CPUState *cpu;
1254
1255 CPU_FOREACH(cpu) {
1256 CpuInfoList *info;
1257 #if defined(TARGET_I386)
1258 X86CPU *x86_cpu = X86_CPU(cpu);
1259 CPUX86State *env = &x86_cpu->env;
1260 #elif defined(TARGET_PPC)
1261 PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu);
1262 CPUPPCState *env = &ppc_cpu->env;
1263 #elif defined(TARGET_SPARC)
1264 SPARCCPU *sparc_cpu = SPARC_CPU(cpu);
1265 CPUSPARCState *env = &sparc_cpu->env;
1266 #elif defined(TARGET_MIPS)
1267 MIPSCPU *mips_cpu = MIPS_CPU(cpu);
1268 CPUMIPSState *env = &mips_cpu->env;
1269 #endif
1270
1271 cpu_synchronize_state(cpu);
1272
1273 info = g_malloc0(sizeof(*info));
1274 info->value = g_malloc0(sizeof(*info->value));
1275 info->value->CPU = cpu->cpu_index;
1276 info->value->current = (cpu == first_cpu);
1277 info->value->halted = cpu->halted;
1278 info->value->thread_id = cpu->thread_id;
1279 #if defined(TARGET_I386)
1280 info->value->has_pc = true;
1281 info->value->pc = env->eip + env->segs[R_CS].base;
1282 #elif defined(TARGET_PPC)
1283 info->value->has_nip = true;
1284 info->value->nip = env->nip;
1285 #elif defined(TARGET_SPARC)
1286 info->value->has_pc = true;
1287 info->value->pc = env->pc;
1288 info->value->has_npc = true;
1289 info->value->npc = env->npc;
1290 #elif defined(TARGET_MIPS)
1291 info->value->has_PC = true;
1292 info->value->PC = env->active_tc.PC;
1293 #endif
1294
1295 /* XXX: waiting for the qapi to support GSList */
1296 if (!cur_item) {
1297 head = cur_item = info;
1298 } else {
1299 cur_item->next = info;
1300 cur_item = info;
1301 }
1302 }
1303
1304 return head;
1305 }
1306
1307 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
1308 bool has_cpu, int64_t cpu_index, Error **errp)
1309 {
1310 FILE *f;
1311 uint32_t l;
1312 CPUState *cpu;
1313 uint8_t buf[1024];
1314
1315 if (!has_cpu) {
1316 cpu_index = 0;
1317 }
1318
1319 cpu = qemu_get_cpu(cpu_index);
1320 if (cpu == NULL) {
1321 error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
1322 "a CPU number");
1323 return;
1324 }
1325
1326 f = fopen(filename, "wb");
1327 if (!f) {
1328 error_setg_file_open(errp, errno, filename);
1329 return;
1330 }
1331
1332 while (size != 0) {
1333 l = sizeof(buf);
1334 if (l > size)
1335 l = size;
1336 cpu_memory_rw_debug(cpu, addr, buf, l, 0);
1337 if (fwrite(buf, 1, l, f) != l) {
1338 error_set(errp, QERR_IO_ERROR);
1339 goto exit;
1340 }
1341 addr += l;
1342 size -= l;
1343 }
1344
1345 exit:
1346 fclose(f);
1347 }
1348
1349 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
1350 Error **errp)
1351 {
1352 FILE *f;
1353 uint32_t l;
1354 uint8_t buf[1024];
1355
1356 f = fopen(filename, "wb");
1357 if (!f) {
1358 error_setg_file_open(errp, errno, filename);
1359 return;
1360 }
1361
1362 while (size != 0) {
1363 l = sizeof(buf);
1364 if (l > size)
1365 l = size;
1366 cpu_physical_memory_rw(addr, buf, l, 0);
1367 if (fwrite(buf, 1, l, f) != l) {
1368 error_set(errp, QERR_IO_ERROR);
1369 goto exit;
1370 }
1371 addr += l;
1372 size -= l;
1373 }
1374
1375 exit:
1376 fclose(f);
1377 }
1378
1379 void qmp_inject_nmi(Error **errp)
1380 {
1381 #if defined(TARGET_I386)
1382 CPUState *cs;
1383
1384 CPU_FOREACH(cs) {
1385 X86CPU *cpu = X86_CPU(cs);
1386 CPUX86State *env = &cpu->env;
1387
1388 if (!env->apic_state) {
1389 cpu_interrupt(cs, CPU_INTERRUPT_NMI);
1390 } else {
1391 apic_deliver_nmi(env->apic_state);
1392 }
1393 }
1394 #elif defined(TARGET_S390X)
1395 CPUState *cs;
1396 S390CPU *cpu;
1397
1398 CPU_FOREACH(cs) {
1399 cpu = S390_CPU(cs);
1400 if (cpu->env.cpu_num == monitor_get_cpu_index()) {
1401 if (s390_cpu_restart(S390_CPU(cs)) == -1) {
1402 error_set(errp, QERR_UNSUPPORTED);
1403 return;
1404 }
1405 break;
1406 }
1407 }
1408 #else
1409 error_set(errp, QERR_UNSUPPORTED);
1410 #endif
1411 }