]>
Commit | Line | Data |
---|---|---|
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 | #include "qemu/seqlock.h" | |
41 | ||
42 | #ifndef _WIN32 | |
43 | #include "qemu/compatfd.h" | |
44 | #endif | |
45 | ||
46 | #ifdef CONFIG_LINUX | |
47 | ||
48 | #include <sys/prctl.h> | |
49 | ||
50 | #ifndef PR_MCE_KILL | |
51 | #define PR_MCE_KILL 33 | |
52 | #endif | |
53 | ||
54 | #ifndef PR_MCE_KILL_SET | |
55 | #define PR_MCE_KILL_SET 1 | |
56 | #endif | |
57 | ||
58 | #ifndef PR_MCE_KILL_EARLY | |
59 | #define PR_MCE_KILL_EARLY 1 | |
60 | #endif | |
61 | ||
62 | #endif /* CONFIG_LINUX */ | |
63 | ||
64 | static CPUState *next_cpu; | |
65 | ||
66 | bool cpu_is_stopped(CPUState *cpu) | |
67 | { | |
68 | return cpu->stopped || !runstate_is_running(); | |
69 | } | |
70 | ||
71 | static bool cpu_thread_is_idle(CPUState *cpu) | |
72 | { | |
73 | if (cpu->stop || cpu->queued_work_first) { | |
74 | return false; | |
75 | } | |
76 | if (cpu_is_stopped(cpu)) { | |
77 | return true; | |
78 | } | |
79 | if (!cpu->halted || cpu_has_work(cpu) || | |
80 | kvm_halt_in_kernel()) { | |
81 | return false; | |
82 | } | |
83 | return true; | |
84 | } | |
85 | ||
86 | static bool all_cpu_threads_idle(void) | |
87 | { | |
88 | CPUState *cpu; | |
89 | ||
90 | CPU_FOREACH(cpu) { | |
91 | if (!cpu_thread_is_idle(cpu)) { | |
92 | return false; | |
93 | } | |
94 | } | |
95 | return true; | |
96 | } | |
97 | ||
98 | /***********************************************************/ | |
99 | /* guest cycle counter */ | |
100 | ||
101 | /* Protected by TimersState seqlock */ | |
102 | ||
103 | /* Compensate for varying guest execution speed. */ | |
104 | static int64_t qemu_icount_bias; | |
105 | static int64_t vm_clock_warp_start; | |
106 | /* Conversion factor from emulated instructions to virtual clock ticks. */ | |
107 | static int icount_time_shift; | |
108 | /* Arbitrarily pick 1MIPS as the minimum allowable speed. */ | |
109 | #define MAX_ICOUNT_SHIFT 10 | |
110 | ||
111 | /* Only written by TCG thread */ | |
112 | static int64_t qemu_icount; | |
113 | ||
114 | static QEMUTimer *icount_rt_timer; | |
115 | static QEMUTimer *icount_vm_timer; | |
116 | static QEMUTimer *icount_warp_timer; | |
117 | ||
118 | typedef struct TimersState { | |
119 | /* Protected by BQL. */ | |
120 | int64_t cpu_ticks_prev; | |
121 | int64_t cpu_ticks_offset; | |
122 | ||
123 | /* cpu_clock_offset can be read out of BQL, so protect it with | |
124 | * this lock. | |
125 | */ | |
126 | QemuSeqLock vm_clock_seqlock; | |
127 | int64_t cpu_clock_offset; | |
128 | int32_t cpu_ticks_enabled; | |
129 | int64_t dummy; | |
130 | } TimersState; | |
131 | ||
132 | static TimersState timers_state; | |
133 | ||
134 | /* Return the virtual CPU time, based on the instruction counter. */ | |
135 | static int64_t cpu_get_icount_locked(void) | |
136 | { | |
137 | int64_t icount; | |
138 | CPUState *cpu = current_cpu; | |
139 | ||
140 | icount = qemu_icount; | |
141 | if (cpu) { | |
142 | if (!cpu_can_do_io(cpu)) { | |
143 | fprintf(stderr, "Bad clock read\n"); | |
144 | } | |
145 | icount -= (cpu->icount_decr.u16.low + cpu->icount_extra); | |
146 | } | |
147 | return qemu_icount_bias + (icount << icount_time_shift); | |
148 | } | |
149 | ||
150 | int64_t cpu_get_icount(void) | |
151 | { | |
152 | int64_t icount; | |
153 | unsigned start; | |
154 | ||
155 | do { | |
156 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
157 | icount = cpu_get_icount_locked(); | |
158 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
159 | ||
160 | return icount; | |
161 | } | |
162 | ||
163 | /* return the host CPU cycle counter and handle stop/restart */ | |
164 | /* Caller must hold the BQL */ | |
165 | int64_t cpu_get_ticks(void) | |
166 | { | |
167 | int64_t ticks; | |
168 | ||
169 | if (use_icount) { | |
170 | return cpu_get_icount(); | |
171 | } | |
172 | ||
173 | ticks = timers_state.cpu_ticks_offset; | |
174 | if (timers_state.cpu_ticks_enabled) { | |
175 | ticks += cpu_get_real_ticks(); | |
176 | } | |
177 | ||
178 | if (timers_state.cpu_ticks_prev > ticks) { | |
179 | /* Note: non increasing ticks may happen if the host uses | |
180 | software suspend */ | |
181 | timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks; | |
182 | ticks = timers_state.cpu_ticks_prev; | |
183 | } | |
184 | ||
185 | timers_state.cpu_ticks_prev = ticks; | |
186 | return ticks; | |
187 | } | |
188 | ||
189 | static int64_t cpu_get_clock_locked(void) | |
190 | { | |
191 | int64_t ticks; | |
192 | ||
193 | ticks = timers_state.cpu_clock_offset; | |
194 | if (timers_state.cpu_ticks_enabled) { | |
195 | ticks += get_clock(); | |
196 | } | |
197 | ||
198 | return ticks; | |
199 | } | |
200 | ||
201 | /* return the host CPU monotonic timer and handle stop/restart */ | |
202 | int64_t cpu_get_clock(void) | |
203 | { | |
204 | int64_t ti; | |
205 | unsigned start; | |
206 | ||
207 | do { | |
208 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
209 | ti = cpu_get_clock_locked(); | |
210 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
211 | ||
212 | return ti; | |
213 | } | |
214 | ||
215 | /* enable cpu_get_ticks() | |
216 | * Caller must hold BQL which server as mutex for vm_clock_seqlock. | |
217 | */ | |
218 | void cpu_enable_ticks(void) | |
219 | { | |
220 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ | |
221 | seqlock_write_lock(&timers_state.vm_clock_seqlock); | |
222 | if (!timers_state.cpu_ticks_enabled) { | |
223 | timers_state.cpu_ticks_offset -= cpu_get_real_ticks(); | |
224 | timers_state.cpu_clock_offset -= get_clock(); | |
225 | timers_state.cpu_ticks_enabled = 1; | |
226 | } | |
227 | seqlock_write_unlock(&timers_state.vm_clock_seqlock); | |
228 | } | |
229 | ||
230 | /* disable cpu_get_ticks() : the clock is stopped. You must not call | |
231 | * cpu_get_ticks() after that. | |
232 | * Caller must hold BQL which server as mutex for vm_clock_seqlock. | |
233 | */ | |
234 | void cpu_disable_ticks(void) | |
235 | { | |
236 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ | |
237 | seqlock_write_lock(&timers_state.vm_clock_seqlock); | |
238 | if (timers_state.cpu_ticks_enabled) { | |
239 | timers_state.cpu_ticks_offset += cpu_get_real_ticks(); | |
240 | timers_state.cpu_clock_offset = cpu_get_clock_locked(); | |
241 | timers_state.cpu_ticks_enabled = 0; | |
242 | } | |
243 | seqlock_write_unlock(&timers_state.vm_clock_seqlock); | |
244 | } | |
245 | ||
246 | /* Correlation between real and virtual time is always going to be | |
247 | fairly approximate, so ignore small variation. | |
248 | When the guest is idle real and virtual time will be aligned in | |
249 | the IO wait loop. */ | |
250 | #define ICOUNT_WOBBLE (get_ticks_per_sec() / 10) | |
251 | ||
252 | static void icount_adjust(void) | |
253 | { | |
254 | int64_t cur_time; | |
255 | int64_t cur_icount; | |
256 | int64_t delta; | |
257 | ||
258 | /* Protected by TimersState mutex. */ | |
259 | static int64_t last_delta; | |
260 | ||
261 | /* If the VM is not running, then do nothing. */ | |
262 | if (!runstate_is_running()) { | |
263 | return; | |
264 | } | |
265 | ||
266 | seqlock_write_lock(&timers_state.vm_clock_seqlock); | |
267 | cur_time = cpu_get_clock_locked(); | |
268 | cur_icount = cpu_get_icount_locked(); | |
269 | ||
270 | delta = cur_icount - cur_time; | |
271 | /* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */ | |
272 | if (delta > 0 | |
273 | && last_delta + ICOUNT_WOBBLE < delta * 2 | |
274 | && icount_time_shift > 0) { | |
275 | /* The guest is getting too far ahead. Slow time down. */ | |
276 | icount_time_shift--; | |
277 | } | |
278 | if (delta < 0 | |
279 | && last_delta - ICOUNT_WOBBLE > delta * 2 | |
280 | && icount_time_shift < MAX_ICOUNT_SHIFT) { | |
281 | /* The guest is getting too far behind. Speed time up. */ | |
282 | icount_time_shift++; | |
283 | } | |
284 | last_delta = delta; | |
285 | qemu_icount_bias = cur_icount - (qemu_icount << icount_time_shift); | |
286 | seqlock_write_unlock(&timers_state.vm_clock_seqlock); | |
287 | } | |
288 | ||
289 | static void icount_adjust_rt(void *opaque) | |
290 | { | |
291 | timer_mod(icount_rt_timer, | |
292 | qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + 1000); | |
293 | icount_adjust(); | |
294 | } | |
295 | ||
296 | static void icount_adjust_vm(void *opaque) | |
297 | { | |
298 | timer_mod(icount_vm_timer, | |
299 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
300 | get_ticks_per_sec() / 10); | |
301 | icount_adjust(); | |
302 | } | |
303 | ||
304 | static int64_t qemu_icount_round(int64_t count) | |
305 | { | |
306 | return (count + (1 << icount_time_shift) - 1) >> icount_time_shift; | |
307 | } | |
308 | ||
309 | static void icount_warp_rt(void *opaque) | |
310 | { | |
311 | /* The icount_warp_timer is rescheduled soon after vm_clock_warp_start | |
312 | * changes from -1 to another value, so the race here is okay. | |
313 | */ | |
314 | if (atomic_read(&vm_clock_warp_start) == -1) { | |
315 | return; | |
316 | } | |
317 | ||
318 | seqlock_write_lock(&timers_state.vm_clock_seqlock); | |
319 | if (runstate_is_running()) { | |
320 | int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
321 | int64_t warp_delta; | |
322 | ||
323 | warp_delta = clock - vm_clock_warp_start; | |
324 | if (use_icount == 2) { | |
325 | /* | |
326 | * In adaptive mode, do not let QEMU_CLOCK_VIRTUAL run too | |
327 | * far ahead of real time. | |
328 | */ | |
329 | int64_t cur_time = cpu_get_clock_locked(); | |
330 | int64_t cur_icount = cpu_get_icount_locked(); | |
331 | int64_t delta = cur_time - cur_icount; | |
332 | warp_delta = MIN(warp_delta, delta); | |
333 | } | |
334 | qemu_icount_bias += warp_delta; | |
335 | } | |
336 | vm_clock_warp_start = -1; | |
337 | seqlock_write_unlock(&timers_state.vm_clock_seqlock); | |
338 | ||
339 | if (qemu_clock_expired(QEMU_CLOCK_VIRTUAL)) { | |
340 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
341 | } | |
342 | } | |
343 | ||
344 | void qtest_clock_warp(int64_t dest) | |
345 | { | |
346 | int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); | |
347 | assert(qtest_enabled()); | |
348 | while (clock < dest) { | |
349 | int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
350 | int64_t warp = MIN(dest - clock, deadline); | |
351 | seqlock_write_lock(&timers_state.vm_clock_seqlock); | |
352 | qemu_icount_bias += warp; | |
353 | seqlock_write_unlock(&timers_state.vm_clock_seqlock); | |
354 | ||
355 | qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL); | |
356 | clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); | |
357 | } | |
358 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
359 | } | |
360 | ||
361 | void qemu_clock_warp(QEMUClockType type) | |
362 | { | |
363 | int64_t clock; | |
364 | int64_t deadline; | |
365 | ||
366 | /* | |
367 | * There are too many global variables to make the "warp" behavior | |
368 | * applicable to other clocks. But a clock argument removes the | |
369 | * need for if statements all over the place. | |
370 | */ | |
371 | if (type != QEMU_CLOCK_VIRTUAL || !use_icount) { | |
372 | return; | |
373 | } | |
374 | ||
375 | /* | |
376 | * If the CPUs have been sleeping, advance QEMU_CLOCK_VIRTUAL timer now. | |
377 | * This ensures that the deadline for the timer is computed correctly below. | |
378 | * This also makes sure that the insn counter is synchronized before the | |
379 | * CPU starts running, in case the CPU is woken by an event other than | |
380 | * the earliest QEMU_CLOCK_VIRTUAL timer. | |
381 | */ | |
382 | icount_warp_rt(NULL); | |
383 | timer_del(icount_warp_timer); | |
384 | if (!all_cpu_threads_idle()) { | |
385 | return; | |
386 | } | |
387 | ||
388 | if (qtest_enabled()) { | |
389 | /* When testing, qtest commands advance icount. */ | |
390 | return; | |
391 | } | |
392 | ||
393 | /* We want to use the earliest deadline from ALL vm_clocks */ | |
394 | clock = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
395 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
396 | if (deadline < 0) { | |
397 | return; | |
398 | } | |
399 | ||
400 | if (deadline > 0) { | |
401 | /* | |
402 | * Ensure QEMU_CLOCK_VIRTUAL proceeds even when the virtual CPU goes to | |
403 | * sleep. Otherwise, the CPU might be waiting for a future timer | |
404 | * interrupt to wake it up, but the interrupt never comes because | |
405 | * the vCPU isn't running any insns and thus doesn't advance the | |
406 | * QEMU_CLOCK_VIRTUAL. | |
407 | * | |
408 | * An extreme solution for this problem would be to never let VCPUs | |
409 | * sleep in icount mode if there is a pending QEMU_CLOCK_VIRTUAL | |
410 | * timer; rather time could just advance to the next QEMU_CLOCK_VIRTUAL | |
411 | * event. Instead, we do stop VCPUs and only advance QEMU_CLOCK_VIRTUAL | |
412 | * after some e"real" time, (related to the time left until the next | |
413 | * event) has passed. The QEMU_CLOCK_REALTIME timer will do this. | |
414 | * This avoids that the warps are visible externally; for example, | |
415 | * you will not be sending network packets continuously instead of | |
416 | * every 100ms. | |
417 | */ | |
418 | seqlock_write_lock(&timers_state.vm_clock_seqlock); | |
419 | if (vm_clock_warp_start == -1 || vm_clock_warp_start > clock) { | |
420 | vm_clock_warp_start = clock; | |
421 | } | |
422 | seqlock_write_unlock(&timers_state.vm_clock_seqlock); | |
423 | timer_mod_anticipate(icount_warp_timer, clock + deadline); | |
424 | } else if (deadline == 0) { | |
425 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
426 | } | |
427 | } | |
428 | ||
429 | static const VMStateDescription vmstate_timers = { | |
430 | .name = "timer", | |
431 | .version_id = 2, | |
432 | .minimum_version_id = 1, | |
433 | .minimum_version_id_old = 1, | |
434 | .fields = (VMStateField[]) { | |
435 | VMSTATE_INT64(cpu_ticks_offset, TimersState), | |
436 | VMSTATE_INT64(dummy, TimersState), | |
437 | VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2), | |
438 | VMSTATE_END_OF_LIST() | |
439 | } | |
440 | }; | |
441 | ||
442 | void configure_icount(const char *option) | |
443 | { | |
444 | seqlock_init(&timers_state.vm_clock_seqlock, NULL); | |
445 | vmstate_register(NULL, 0, &vmstate_timers, &timers_state); | |
446 | if (!option) { | |
447 | return; | |
448 | } | |
449 | ||
450 | icount_warp_timer = timer_new_ns(QEMU_CLOCK_REALTIME, | |
451 | icount_warp_rt, NULL); | |
452 | if (strcmp(option, "auto") != 0) { | |
453 | icount_time_shift = strtol(option, NULL, 0); | |
454 | use_icount = 1; | |
455 | return; | |
456 | } | |
457 | ||
458 | use_icount = 2; | |
459 | ||
460 | /* 125MIPS seems a reasonable initial guess at the guest speed. | |
461 | It will be corrected fairly quickly anyway. */ | |
462 | icount_time_shift = 3; | |
463 | ||
464 | /* Have both realtime and virtual time triggers for speed adjustment. | |
465 | The realtime trigger catches emulated time passing too slowly, | |
466 | the virtual time trigger catches emulated time passing too fast. | |
467 | Realtime triggers occur even when idle, so use them less frequently | |
468 | than VM triggers. */ | |
469 | icount_rt_timer = timer_new_ms(QEMU_CLOCK_REALTIME, | |
470 | icount_adjust_rt, NULL); | |
471 | timer_mod(icount_rt_timer, | |
472 | qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + 1000); | |
473 | icount_vm_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, | |
474 | icount_adjust_vm, NULL); | |
475 | timer_mod(icount_vm_timer, | |
476 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
477 | get_ticks_per_sec() / 10); | |
478 | } | |
479 | ||
480 | /***********************************************************/ | |
481 | void hw_error(const char *fmt, ...) | |
482 | { | |
483 | va_list ap; | |
484 | CPUState *cpu; | |
485 | ||
486 | va_start(ap, fmt); | |
487 | fprintf(stderr, "qemu: hardware error: "); | |
488 | vfprintf(stderr, fmt, ap); | |
489 | fprintf(stderr, "\n"); | |
490 | CPU_FOREACH(cpu) { | |
491 | fprintf(stderr, "CPU #%d:\n", cpu->cpu_index); | |
492 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU); | |
493 | } | |
494 | va_end(ap); | |
495 | abort(); | |
496 | } | |
497 | ||
498 | void cpu_synchronize_all_states(void) | |
499 | { | |
500 | CPUState *cpu; | |
501 | ||
502 | CPU_FOREACH(cpu) { | |
503 | cpu_synchronize_state(cpu); | |
504 | } | |
505 | } | |
506 | ||
507 | void cpu_synchronize_all_post_reset(void) | |
508 | { | |
509 | CPUState *cpu; | |
510 | ||
511 | CPU_FOREACH(cpu) { | |
512 | cpu_synchronize_post_reset(cpu); | |
513 | } | |
514 | } | |
515 | ||
516 | void cpu_synchronize_all_post_init(void) | |
517 | { | |
518 | CPUState *cpu; | |
519 | ||
520 | CPU_FOREACH(cpu) { | |
521 | cpu_synchronize_post_init(cpu); | |
522 | } | |
523 | } | |
524 | ||
525 | static int do_vm_stop(RunState state) | |
526 | { | |
527 | int ret = 0; | |
528 | ||
529 | if (runstate_is_running()) { | |
530 | cpu_disable_ticks(); | |
531 | pause_all_vcpus(); | |
532 | runstate_set(state); | |
533 | vm_state_notify(0, state); | |
534 | monitor_protocol_event(QEVENT_STOP, NULL); | |
535 | } | |
536 | ||
537 | bdrv_drain_all(); | |
538 | ret = bdrv_flush_all(); | |
539 | ||
540 | return ret; | |
541 | } | |
542 | ||
543 | static bool cpu_can_run(CPUState *cpu) | |
544 | { | |
545 | if (cpu->stop) { | |
546 | return false; | |
547 | } | |
548 | if (cpu_is_stopped(cpu)) { | |
549 | return false; | |
550 | } | |
551 | return true; | |
552 | } | |
553 | ||
554 | static void cpu_handle_guest_debug(CPUState *cpu) | |
555 | { | |
556 | gdb_set_stop_cpu(cpu); | |
557 | qemu_system_debug_request(); | |
558 | cpu->stopped = true; | |
559 | } | |
560 | ||
561 | static void cpu_signal(int sig) | |
562 | { | |
563 | if (current_cpu) { | |
564 | cpu_exit(current_cpu); | |
565 | } | |
566 | exit_request = 1; | |
567 | } | |
568 | ||
569 | #ifdef CONFIG_LINUX | |
570 | static void sigbus_reraise(void) | |
571 | { | |
572 | sigset_t set; | |
573 | struct sigaction action; | |
574 | ||
575 | memset(&action, 0, sizeof(action)); | |
576 | action.sa_handler = SIG_DFL; | |
577 | if (!sigaction(SIGBUS, &action, NULL)) { | |
578 | raise(SIGBUS); | |
579 | sigemptyset(&set); | |
580 | sigaddset(&set, SIGBUS); | |
581 | sigprocmask(SIG_UNBLOCK, &set, NULL); | |
582 | } | |
583 | perror("Failed to re-raise SIGBUS!\n"); | |
584 | abort(); | |
585 | } | |
586 | ||
587 | static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo, | |
588 | void *ctx) | |
589 | { | |
590 | if (kvm_on_sigbus(siginfo->ssi_code, | |
591 | (void *)(intptr_t)siginfo->ssi_addr)) { | |
592 | sigbus_reraise(); | |
593 | } | |
594 | } | |
595 | ||
596 | static void qemu_init_sigbus(void) | |
597 | { | |
598 | struct sigaction action; | |
599 | ||
600 | memset(&action, 0, sizeof(action)); | |
601 | action.sa_flags = SA_SIGINFO; | |
602 | action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler; | |
603 | sigaction(SIGBUS, &action, NULL); | |
604 | ||
605 | prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0); | |
606 | } | |
607 | ||
608 | static void qemu_kvm_eat_signals(CPUState *cpu) | |
609 | { | |
610 | struct timespec ts = { 0, 0 }; | |
611 | siginfo_t siginfo; | |
612 | sigset_t waitset; | |
613 | sigset_t chkset; | |
614 | int r; | |
615 | ||
616 | sigemptyset(&waitset); | |
617 | sigaddset(&waitset, SIG_IPI); | |
618 | sigaddset(&waitset, SIGBUS); | |
619 | ||
620 | do { | |
621 | r = sigtimedwait(&waitset, &siginfo, &ts); | |
622 | if (r == -1 && !(errno == EAGAIN || errno == EINTR)) { | |
623 | perror("sigtimedwait"); | |
624 | exit(1); | |
625 | } | |
626 | ||
627 | switch (r) { | |
628 | case SIGBUS: | |
629 | if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) { | |
630 | sigbus_reraise(); | |
631 | } | |
632 | break; | |
633 | default: | |
634 | break; | |
635 | } | |
636 | ||
637 | r = sigpending(&chkset); | |
638 | if (r == -1) { | |
639 | perror("sigpending"); | |
640 | exit(1); | |
641 | } | |
642 | } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS)); | |
643 | } | |
644 | ||
645 | #else /* !CONFIG_LINUX */ | |
646 | ||
647 | static void qemu_init_sigbus(void) | |
648 | { | |
649 | } | |
650 | ||
651 | static void qemu_kvm_eat_signals(CPUState *cpu) | |
652 | { | |
653 | } | |
654 | #endif /* !CONFIG_LINUX */ | |
655 | ||
656 | #ifndef _WIN32 | |
657 | static void dummy_signal(int sig) | |
658 | { | |
659 | } | |
660 | ||
661 | static void qemu_kvm_init_cpu_signals(CPUState *cpu) | |
662 | { | |
663 | int r; | |
664 | sigset_t set; | |
665 | struct sigaction sigact; | |
666 | ||
667 | memset(&sigact, 0, sizeof(sigact)); | |
668 | sigact.sa_handler = dummy_signal; | |
669 | sigaction(SIG_IPI, &sigact, NULL); | |
670 | ||
671 | pthread_sigmask(SIG_BLOCK, NULL, &set); | |
672 | sigdelset(&set, SIG_IPI); | |
673 | sigdelset(&set, SIGBUS); | |
674 | r = kvm_set_signal_mask(cpu, &set); | |
675 | if (r) { | |
676 | fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r)); | |
677 | exit(1); | |
678 | } | |
679 | } | |
680 | ||
681 | static void qemu_tcg_init_cpu_signals(void) | |
682 | { | |
683 | sigset_t set; | |
684 | struct sigaction sigact; | |
685 | ||
686 | memset(&sigact, 0, sizeof(sigact)); | |
687 | sigact.sa_handler = cpu_signal; | |
688 | sigaction(SIG_IPI, &sigact, NULL); | |
689 | ||
690 | sigemptyset(&set); | |
691 | sigaddset(&set, SIG_IPI); | |
692 | pthread_sigmask(SIG_UNBLOCK, &set, NULL); | |
693 | } | |
694 | ||
695 | #else /* _WIN32 */ | |
696 | static void qemu_kvm_init_cpu_signals(CPUState *cpu) | |
697 | { | |
698 | abort(); | |
699 | } | |
700 | ||
701 | static void qemu_tcg_init_cpu_signals(void) | |
702 | { | |
703 | } | |
704 | #endif /* _WIN32 */ | |
705 | ||
706 | static QemuMutex qemu_global_mutex; | |
707 | static QemuCond qemu_io_proceeded_cond; | |
708 | static bool iothread_requesting_mutex; | |
709 | ||
710 | static QemuThread io_thread; | |
711 | ||
712 | static QemuThread *tcg_cpu_thread; | |
713 | static QemuCond *tcg_halt_cond; | |
714 | ||
715 | /* cpu creation */ | |
716 | static QemuCond qemu_cpu_cond; | |
717 | /* system init */ | |
718 | static QemuCond qemu_pause_cond; | |
719 | static QemuCond qemu_work_cond; | |
720 | ||
721 | void qemu_init_cpu_loop(void) | |
722 | { | |
723 | qemu_init_sigbus(); | |
724 | qemu_cond_init(&qemu_cpu_cond); | |
725 | qemu_cond_init(&qemu_pause_cond); | |
726 | qemu_cond_init(&qemu_work_cond); | |
727 | qemu_cond_init(&qemu_io_proceeded_cond); | |
728 | qemu_mutex_init(&qemu_global_mutex); | |
729 | ||
730 | qemu_thread_get_self(&io_thread); | |
731 | } | |
732 | ||
733 | void run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data) | |
734 | { | |
735 | struct qemu_work_item wi; | |
736 | ||
737 | if (qemu_cpu_is_self(cpu)) { | |
738 | func(data); | |
739 | return; | |
740 | } | |
741 | ||
742 | wi.func = func; | |
743 | wi.data = data; | |
744 | wi.free = false; | |
745 | if (cpu->queued_work_first == NULL) { | |
746 | cpu->queued_work_first = &wi; | |
747 | } else { | |
748 | cpu->queued_work_last->next = &wi; | |
749 | } | |
750 | cpu->queued_work_last = &wi; | |
751 | wi.next = NULL; | |
752 | wi.done = false; | |
753 | ||
754 | qemu_cpu_kick(cpu); | |
755 | while (!wi.done) { | |
756 | CPUState *self_cpu = current_cpu; | |
757 | ||
758 | qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex); | |
759 | current_cpu = self_cpu; | |
760 | } | |
761 | } | |
762 | ||
763 | void async_run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data) | |
764 | { | |
765 | struct qemu_work_item *wi; | |
766 | ||
767 | if (qemu_cpu_is_self(cpu)) { | |
768 | func(data); | |
769 | return; | |
770 | } | |
771 | ||
772 | wi = g_malloc0(sizeof(struct qemu_work_item)); | |
773 | wi->func = func; | |
774 | wi->data = data; | |
775 | wi->free = true; | |
776 | if (cpu->queued_work_first == NULL) { | |
777 | cpu->queued_work_first = wi; | |
778 | } else { | |
779 | cpu->queued_work_last->next = wi; | |
780 | } | |
781 | cpu->queued_work_last = wi; | |
782 | wi->next = NULL; | |
783 | wi->done = false; | |
784 | ||
785 | qemu_cpu_kick(cpu); | |
786 | } | |
787 | ||
788 | static void flush_queued_work(CPUState *cpu) | |
789 | { | |
790 | struct qemu_work_item *wi; | |
791 | ||
792 | if (cpu->queued_work_first == NULL) { | |
793 | return; | |
794 | } | |
795 | ||
796 | while ((wi = cpu->queued_work_first)) { | |
797 | cpu->queued_work_first = wi->next; | |
798 | wi->func(wi->data); | |
799 | wi->done = true; | |
800 | if (wi->free) { | |
801 | g_free(wi); | |
802 | } | |
803 | } | |
804 | cpu->queued_work_last = NULL; | |
805 | qemu_cond_broadcast(&qemu_work_cond); | |
806 | } | |
807 | ||
808 | static void qemu_wait_io_event_common(CPUState *cpu) | |
809 | { | |
810 | if (cpu->stop) { | |
811 | cpu->stop = false; | |
812 | cpu->stopped = true; | |
813 | qemu_cond_signal(&qemu_pause_cond); | |
814 | } | |
815 | flush_queued_work(cpu); | |
816 | cpu->thread_kicked = false; | |
817 | } | |
818 | ||
819 | static void qemu_tcg_wait_io_event(void) | |
820 | { | |
821 | CPUState *cpu; | |
822 | ||
823 | while (all_cpu_threads_idle()) { | |
824 | /* Start accounting real time to the virtual clock if the CPUs | |
825 | are idle. */ | |
826 | qemu_clock_warp(QEMU_CLOCK_VIRTUAL); | |
827 | qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex); | |
828 | } | |
829 | ||
830 | while (iothread_requesting_mutex) { | |
831 | qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex); | |
832 | } | |
833 | ||
834 | CPU_FOREACH(cpu) { | |
835 | qemu_wait_io_event_common(cpu); | |
836 | } | |
837 | } | |
838 | ||
839 | static void qemu_kvm_wait_io_event(CPUState *cpu) | |
840 | { | |
841 | while (cpu_thread_is_idle(cpu)) { | |
842 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); | |
843 | } | |
844 | ||
845 | qemu_kvm_eat_signals(cpu); | |
846 | qemu_wait_io_event_common(cpu); | |
847 | } | |
848 | ||
849 | static void *qemu_kvm_cpu_thread_fn(void *arg) | |
850 | { | |
851 | CPUState *cpu = arg; | |
852 | int r; | |
853 | ||
854 | qemu_mutex_lock(&qemu_global_mutex); | |
855 | qemu_thread_get_self(cpu->thread); | |
856 | cpu->thread_id = qemu_get_thread_id(); | |
857 | current_cpu = cpu; | |
858 | ||
859 | r = kvm_init_vcpu(cpu); | |
860 | if (r < 0) { | |
861 | fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r)); | |
862 | exit(1); | |
863 | } | |
864 | ||
865 | qemu_kvm_init_cpu_signals(cpu); | |
866 | ||
867 | /* signal CPU creation */ | |
868 | cpu->created = true; | |
869 | qemu_cond_signal(&qemu_cpu_cond); | |
870 | ||
871 | while (1) { | |
872 | if (cpu_can_run(cpu)) { | |
873 | r = kvm_cpu_exec(cpu); | |
874 | if (r == EXCP_DEBUG) { | |
875 | cpu_handle_guest_debug(cpu); | |
876 | } | |
877 | } | |
878 | qemu_kvm_wait_io_event(cpu); | |
879 | } | |
880 | ||
881 | return NULL; | |
882 | } | |
883 | ||
884 | static void *qemu_dummy_cpu_thread_fn(void *arg) | |
885 | { | |
886 | #ifdef _WIN32 | |
887 | fprintf(stderr, "qtest is not supported under Windows\n"); | |
888 | exit(1); | |
889 | #else | |
890 | CPUState *cpu = arg; | |
891 | sigset_t waitset; | |
892 | int r; | |
893 | ||
894 | qemu_mutex_lock_iothread(); | |
895 | qemu_thread_get_self(cpu->thread); | |
896 | cpu->thread_id = qemu_get_thread_id(); | |
897 | ||
898 | sigemptyset(&waitset); | |
899 | sigaddset(&waitset, SIG_IPI); | |
900 | ||
901 | /* signal CPU creation */ | |
902 | cpu->created = true; | |
903 | qemu_cond_signal(&qemu_cpu_cond); | |
904 | ||
905 | current_cpu = cpu; | |
906 | while (1) { | |
907 | current_cpu = NULL; | |
908 | qemu_mutex_unlock_iothread(); | |
909 | do { | |
910 | int sig; | |
911 | r = sigwait(&waitset, &sig); | |
912 | } while (r == -1 && (errno == EAGAIN || errno == EINTR)); | |
913 | if (r == -1) { | |
914 | perror("sigwait"); | |
915 | exit(1); | |
916 | } | |
917 | qemu_mutex_lock_iothread(); | |
918 | current_cpu = cpu; | |
919 | qemu_wait_io_event_common(cpu); | |
920 | } | |
921 | ||
922 | return NULL; | |
923 | #endif | |
924 | } | |
925 | ||
926 | static void tcg_exec_all(void); | |
927 | ||
928 | static void *qemu_tcg_cpu_thread_fn(void *arg) | |
929 | { | |
930 | CPUState *cpu = arg; | |
931 | ||
932 | qemu_tcg_init_cpu_signals(); | |
933 | qemu_thread_get_self(cpu->thread); | |
934 | ||
935 | qemu_mutex_lock(&qemu_global_mutex); | |
936 | CPU_FOREACH(cpu) { | |
937 | cpu->thread_id = qemu_get_thread_id(); | |
938 | cpu->created = true; | |
939 | } | |
940 | qemu_cond_signal(&qemu_cpu_cond); | |
941 | ||
942 | /* wait for initial kick-off after machine start */ | |
943 | while (QTAILQ_FIRST(&cpus)->stopped) { | |
944 | qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex); | |
945 | ||
946 | /* process any pending work */ | |
947 | CPU_FOREACH(cpu) { | |
948 | qemu_wait_io_event_common(cpu); | |
949 | } | |
950 | } | |
951 | ||
952 | while (1) { | |
953 | tcg_exec_all(); | |
954 | ||
955 | if (use_icount) { | |
956 | int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
957 | ||
958 | if (deadline == 0) { | |
959 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
960 | } | |
961 | } | |
962 | qemu_tcg_wait_io_event(); | |
963 | } | |
964 | ||
965 | return NULL; | |
966 | } | |
967 | ||
968 | static void qemu_cpu_kick_thread(CPUState *cpu) | |
969 | { | |
970 | #ifndef _WIN32 | |
971 | int err; | |
972 | ||
973 | err = pthread_kill(cpu->thread->thread, SIG_IPI); | |
974 | if (err) { | |
975 | fprintf(stderr, "qemu:%s: %s", __func__, strerror(err)); | |
976 | exit(1); | |
977 | } | |
978 | #else /* _WIN32 */ | |
979 | if (!qemu_cpu_is_self(cpu)) { | |
980 | CONTEXT tcgContext; | |
981 | ||
982 | if (SuspendThread(cpu->hThread) == (DWORD)-1) { | |
983 | fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__, | |
984 | GetLastError()); | |
985 | exit(1); | |
986 | } | |
987 | ||
988 | /* On multi-core systems, we are not sure that the thread is actually | |
989 | * suspended until we can get the context. | |
990 | */ | |
991 | tcgContext.ContextFlags = CONTEXT_CONTROL; | |
992 | while (GetThreadContext(cpu->hThread, &tcgContext) != 0) { | |
993 | continue; | |
994 | } | |
995 | ||
996 | cpu_signal(0); | |
997 | ||
998 | if (ResumeThread(cpu->hThread) == (DWORD)-1) { | |
999 | fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__, | |
1000 | GetLastError()); | |
1001 | exit(1); | |
1002 | } | |
1003 | } | |
1004 | #endif | |
1005 | } | |
1006 | ||
1007 | void qemu_cpu_kick(CPUState *cpu) | |
1008 | { | |
1009 | qemu_cond_broadcast(cpu->halt_cond); | |
1010 | if (!tcg_enabled() && !cpu->thread_kicked) { | |
1011 | qemu_cpu_kick_thread(cpu); | |
1012 | cpu->thread_kicked = true; | |
1013 | } | |
1014 | } | |
1015 | ||
1016 | void qemu_cpu_kick_self(void) | |
1017 | { | |
1018 | #ifndef _WIN32 | |
1019 | assert(current_cpu); | |
1020 | ||
1021 | if (!current_cpu->thread_kicked) { | |
1022 | qemu_cpu_kick_thread(current_cpu); | |
1023 | current_cpu->thread_kicked = true; | |
1024 | } | |
1025 | #else | |
1026 | abort(); | |
1027 | #endif | |
1028 | } | |
1029 | ||
1030 | bool qemu_cpu_is_self(CPUState *cpu) | |
1031 | { | |
1032 | return qemu_thread_is_self(cpu->thread); | |
1033 | } | |
1034 | ||
1035 | static bool qemu_in_vcpu_thread(void) | |
1036 | { | |
1037 | return current_cpu && qemu_cpu_is_self(current_cpu); | |
1038 | } | |
1039 | ||
1040 | void qemu_mutex_lock_iothread(void) | |
1041 | { | |
1042 | if (!tcg_enabled()) { | |
1043 | qemu_mutex_lock(&qemu_global_mutex); | |
1044 | } else { | |
1045 | iothread_requesting_mutex = true; | |
1046 | if (qemu_mutex_trylock(&qemu_global_mutex)) { | |
1047 | qemu_cpu_kick_thread(first_cpu); | |
1048 | qemu_mutex_lock(&qemu_global_mutex); | |
1049 | } | |
1050 | iothread_requesting_mutex = false; | |
1051 | qemu_cond_broadcast(&qemu_io_proceeded_cond); | |
1052 | } | |
1053 | } | |
1054 | ||
1055 | void qemu_mutex_unlock_iothread(void) | |
1056 | { | |
1057 | qemu_mutex_unlock(&qemu_global_mutex); | |
1058 | } | |
1059 | ||
1060 | static int all_vcpus_paused(void) | |
1061 | { | |
1062 | CPUState *cpu; | |
1063 | ||
1064 | CPU_FOREACH(cpu) { | |
1065 | if (!cpu->stopped) { | |
1066 | return 0; | |
1067 | } | |
1068 | } | |
1069 | ||
1070 | return 1; | |
1071 | } | |
1072 | ||
1073 | void pause_all_vcpus(void) | |
1074 | { | |
1075 | CPUState *cpu; | |
1076 | ||
1077 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false); | |
1078 | CPU_FOREACH(cpu) { | |
1079 | cpu->stop = true; | |
1080 | qemu_cpu_kick(cpu); | |
1081 | } | |
1082 | ||
1083 | if (qemu_in_vcpu_thread()) { | |
1084 | cpu_stop_current(); | |
1085 | if (!kvm_enabled()) { | |
1086 | CPU_FOREACH(cpu) { | |
1087 | cpu->stop = false; | |
1088 | cpu->stopped = true; | |
1089 | } | |
1090 | return; | |
1091 | } | |
1092 | } | |
1093 | ||
1094 | while (!all_vcpus_paused()) { | |
1095 | qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex); | |
1096 | CPU_FOREACH(cpu) { | |
1097 | qemu_cpu_kick(cpu); | |
1098 | } | |
1099 | } | |
1100 | } | |
1101 | ||
1102 | void cpu_resume(CPUState *cpu) | |
1103 | { | |
1104 | cpu->stop = false; | |
1105 | cpu->stopped = false; | |
1106 | qemu_cpu_kick(cpu); | |
1107 | } | |
1108 | ||
1109 | void resume_all_vcpus(void) | |
1110 | { | |
1111 | CPUState *cpu; | |
1112 | ||
1113 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true); | |
1114 | CPU_FOREACH(cpu) { | |
1115 | cpu_resume(cpu); | |
1116 | } | |
1117 | } | |
1118 | ||
1119 | /* For temporary buffers for forming a name */ | |
1120 | #define VCPU_THREAD_NAME_SIZE 16 | |
1121 | ||
1122 | static void qemu_tcg_init_vcpu(CPUState *cpu) | |
1123 | { | |
1124 | char thread_name[VCPU_THREAD_NAME_SIZE]; | |
1125 | ||
1126 | tcg_cpu_address_space_init(cpu, cpu->as); | |
1127 | ||
1128 | /* share a single thread for all cpus with TCG */ | |
1129 | if (!tcg_cpu_thread) { | |
1130 | cpu->thread = g_malloc0(sizeof(QemuThread)); | |
1131 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); | |
1132 | qemu_cond_init(cpu->halt_cond); | |
1133 | tcg_halt_cond = cpu->halt_cond; | |
1134 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG", | |
1135 | cpu->cpu_index); | |
1136 | qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn, | |
1137 | cpu, QEMU_THREAD_JOINABLE); | |
1138 | #ifdef _WIN32 | |
1139 | cpu->hThread = qemu_thread_get_handle(cpu->thread); | |
1140 | #endif | |
1141 | while (!cpu->created) { | |
1142 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1143 | } | |
1144 | tcg_cpu_thread = cpu->thread; | |
1145 | } else { | |
1146 | cpu->thread = tcg_cpu_thread; | |
1147 | cpu->halt_cond = tcg_halt_cond; | |
1148 | } | |
1149 | } | |
1150 | ||
1151 | static void qemu_kvm_start_vcpu(CPUState *cpu) | |
1152 | { | |
1153 | char thread_name[VCPU_THREAD_NAME_SIZE]; | |
1154 | ||
1155 | cpu->thread = g_malloc0(sizeof(QemuThread)); | |
1156 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); | |
1157 | qemu_cond_init(cpu->halt_cond); | |
1158 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/KVM", | |
1159 | cpu->cpu_index); | |
1160 | qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn, | |
1161 | cpu, QEMU_THREAD_JOINABLE); | |
1162 | while (!cpu->created) { | |
1163 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1164 | } | |
1165 | } | |
1166 | ||
1167 | static void qemu_dummy_start_vcpu(CPUState *cpu) | |
1168 | { | |
1169 | char thread_name[VCPU_THREAD_NAME_SIZE]; | |
1170 | ||
1171 | cpu->thread = g_malloc0(sizeof(QemuThread)); | |
1172 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); | |
1173 | qemu_cond_init(cpu->halt_cond); | |
1174 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/DUMMY", | |
1175 | cpu->cpu_index); | |
1176 | qemu_thread_create(cpu->thread, thread_name, qemu_dummy_cpu_thread_fn, cpu, | |
1177 | QEMU_THREAD_JOINABLE); | |
1178 | while (!cpu->created) { | |
1179 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1180 | } | |
1181 | } | |
1182 | ||
1183 | void qemu_init_vcpu(CPUState *cpu) | |
1184 | { | |
1185 | cpu->nr_cores = smp_cores; | |
1186 | cpu->nr_threads = smp_threads; | |
1187 | cpu->stopped = true; | |
1188 | if (kvm_enabled()) { | |
1189 | qemu_kvm_start_vcpu(cpu); | |
1190 | } else if (tcg_enabled()) { | |
1191 | qemu_tcg_init_vcpu(cpu); | |
1192 | } else { | |
1193 | qemu_dummy_start_vcpu(cpu); | |
1194 | } | |
1195 | } | |
1196 | ||
1197 | void cpu_stop_current(void) | |
1198 | { | |
1199 | if (current_cpu) { | |
1200 | current_cpu->stop = false; | |
1201 | current_cpu->stopped = true; | |
1202 | cpu_exit(current_cpu); | |
1203 | qemu_cond_signal(&qemu_pause_cond); | |
1204 | } | |
1205 | } | |
1206 | ||
1207 | int vm_stop(RunState state) | |
1208 | { | |
1209 | if (qemu_in_vcpu_thread()) { | |
1210 | qemu_system_vmstop_request(state); | |
1211 | /* | |
1212 | * FIXME: should not return to device code in case | |
1213 | * vm_stop() has been requested. | |
1214 | */ | |
1215 | cpu_stop_current(); | |
1216 | return 0; | |
1217 | } | |
1218 | ||
1219 | return do_vm_stop(state); | |
1220 | } | |
1221 | ||
1222 | /* does a state transition even if the VM is already stopped, | |
1223 | current state is forgotten forever */ | |
1224 | int vm_stop_force_state(RunState state) | |
1225 | { | |
1226 | if (runstate_is_running()) { | |
1227 | return vm_stop(state); | |
1228 | } else { | |
1229 | runstate_set(state); | |
1230 | /* Make sure to return an error if the flush in a previous vm_stop() | |
1231 | * failed. */ | |
1232 | return bdrv_flush_all(); | |
1233 | } | |
1234 | } | |
1235 | ||
1236 | static int tcg_cpu_exec(CPUArchState *env) | |
1237 | { | |
1238 | CPUState *cpu = ENV_GET_CPU(env); | |
1239 | int ret; | |
1240 | #ifdef CONFIG_PROFILER | |
1241 | int64_t ti; | |
1242 | #endif | |
1243 | ||
1244 | #ifdef CONFIG_PROFILER | |
1245 | ti = profile_getclock(); | |
1246 | #endif | |
1247 | if (use_icount) { | |
1248 | int64_t count; | |
1249 | int64_t deadline; | |
1250 | int decr; | |
1251 | qemu_icount -= (cpu->icount_decr.u16.low + cpu->icount_extra); | |
1252 | cpu->icount_decr.u16.low = 0; | |
1253 | cpu->icount_extra = 0; | |
1254 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
1255 | ||
1256 | /* Maintain prior (possibly buggy) behaviour where if no deadline | |
1257 | * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than | |
1258 | * INT32_MAX nanoseconds ahead, we still use INT32_MAX | |
1259 | * nanoseconds. | |
1260 | */ | |
1261 | if ((deadline < 0) || (deadline > INT32_MAX)) { | |
1262 | deadline = INT32_MAX; | |
1263 | } | |
1264 | ||
1265 | count = qemu_icount_round(deadline); | |
1266 | qemu_icount += count; | |
1267 | decr = (count > 0xffff) ? 0xffff : count; | |
1268 | count -= decr; | |
1269 | cpu->icount_decr.u16.low = decr; | |
1270 | cpu->icount_extra = count; | |
1271 | } | |
1272 | ret = cpu_exec(env); | |
1273 | #ifdef CONFIG_PROFILER | |
1274 | qemu_time += profile_getclock() - ti; | |
1275 | #endif | |
1276 | if (use_icount) { | |
1277 | /* Fold pending instructions back into the | |
1278 | instruction counter, and clear the interrupt flag. */ | |
1279 | qemu_icount -= (cpu->icount_decr.u16.low + cpu->icount_extra); | |
1280 | cpu->icount_decr.u32 = 0; | |
1281 | cpu->icount_extra = 0; | |
1282 | } | |
1283 | return ret; | |
1284 | } | |
1285 | ||
1286 | static void tcg_exec_all(void) | |
1287 | { | |
1288 | int r; | |
1289 | ||
1290 | /* Account partial waits to QEMU_CLOCK_VIRTUAL. */ | |
1291 | qemu_clock_warp(QEMU_CLOCK_VIRTUAL); | |
1292 | ||
1293 | if (next_cpu == NULL) { | |
1294 | next_cpu = first_cpu; | |
1295 | } | |
1296 | for (; next_cpu != NULL && !exit_request; next_cpu = CPU_NEXT(next_cpu)) { | |
1297 | CPUState *cpu = next_cpu; | |
1298 | CPUArchState *env = cpu->env_ptr; | |
1299 | ||
1300 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, | |
1301 | (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0); | |
1302 | ||
1303 | if (cpu_can_run(cpu)) { | |
1304 | r = tcg_cpu_exec(env); | |
1305 | if (r == EXCP_DEBUG) { | |
1306 | cpu_handle_guest_debug(cpu); | |
1307 | break; | |
1308 | } | |
1309 | } else if (cpu->stop || cpu->stopped) { | |
1310 | break; | |
1311 | } | |
1312 | } | |
1313 | exit_request = 0; | |
1314 | } | |
1315 | ||
1316 | void set_numa_modes(void) | |
1317 | { | |
1318 | CPUState *cpu; | |
1319 | int i; | |
1320 | ||
1321 | CPU_FOREACH(cpu) { | |
1322 | for (i = 0; i < nb_numa_nodes; i++) { | |
1323 | if (test_bit(cpu->cpu_index, node_cpumask[i])) { | |
1324 | cpu->numa_node = i; | |
1325 | } | |
1326 | } | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg) | |
1331 | { | |
1332 | /* XXX: implement xxx_cpu_list for targets that still miss it */ | |
1333 | #if defined(cpu_list) | |
1334 | cpu_list(f, cpu_fprintf); | |
1335 | #endif | |
1336 | } | |
1337 | ||
1338 | CpuInfoList *qmp_query_cpus(Error **errp) | |
1339 | { | |
1340 | CpuInfoList *head = NULL, *cur_item = NULL; | |
1341 | CPUState *cpu; | |
1342 | ||
1343 | CPU_FOREACH(cpu) { | |
1344 | CpuInfoList *info; | |
1345 | #if defined(TARGET_I386) | |
1346 | X86CPU *x86_cpu = X86_CPU(cpu); | |
1347 | CPUX86State *env = &x86_cpu->env; | |
1348 | #elif defined(TARGET_PPC) | |
1349 | PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu); | |
1350 | CPUPPCState *env = &ppc_cpu->env; | |
1351 | #elif defined(TARGET_SPARC) | |
1352 | SPARCCPU *sparc_cpu = SPARC_CPU(cpu); | |
1353 | CPUSPARCState *env = &sparc_cpu->env; | |
1354 | #elif defined(TARGET_MIPS) | |
1355 | MIPSCPU *mips_cpu = MIPS_CPU(cpu); | |
1356 | CPUMIPSState *env = &mips_cpu->env; | |
1357 | #endif | |
1358 | ||
1359 | cpu_synchronize_state(cpu); | |
1360 | ||
1361 | info = g_malloc0(sizeof(*info)); | |
1362 | info->value = g_malloc0(sizeof(*info->value)); | |
1363 | info->value->CPU = cpu->cpu_index; | |
1364 | info->value->current = (cpu == first_cpu); | |
1365 | info->value->halted = cpu->halted; | |
1366 | info->value->thread_id = cpu->thread_id; | |
1367 | #if defined(TARGET_I386) | |
1368 | info->value->has_pc = true; | |
1369 | info->value->pc = env->eip + env->segs[R_CS].base; | |
1370 | #elif defined(TARGET_PPC) | |
1371 | info->value->has_nip = true; | |
1372 | info->value->nip = env->nip; | |
1373 | #elif defined(TARGET_SPARC) | |
1374 | info->value->has_pc = true; | |
1375 | info->value->pc = env->pc; | |
1376 | info->value->has_npc = true; | |
1377 | info->value->npc = env->npc; | |
1378 | #elif defined(TARGET_MIPS) | |
1379 | info->value->has_PC = true; | |
1380 | info->value->PC = env->active_tc.PC; | |
1381 | #endif | |
1382 | ||
1383 | /* XXX: waiting for the qapi to support GSList */ | |
1384 | if (!cur_item) { | |
1385 | head = cur_item = info; | |
1386 | } else { | |
1387 | cur_item->next = info; | |
1388 | cur_item = info; | |
1389 | } | |
1390 | } | |
1391 | ||
1392 | return head; | |
1393 | } | |
1394 | ||
1395 | void qmp_memsave(int64_t addr, int64_t size, const char *filename, | |
1396 | bool has_cpu, int64_t cpu_index, Error **errp) | |
1397 | { | |
1398 | FILE *f; | |
1399 | uint32_t l; | |
1400 | CPUState *cpu; | |
1401 | uint8_t buf[1024]; | |
1402 | ||
1403 | if (!has_cpu) { | |
1404 | cpu_index = 0; | |
1405 | } | |
1406 | ||
1407 | cpu = qemu_get_cpu(cpu_index); | |
1408 | if (cpu == NULL) { | |
1409 | error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index", | |
1410 | "a CPU number"); | |
1411 | return; | |
1412 | } | |
1413 | ||
1414 | f = fopen(filename, "wb"); | |
1415 | if (!f) { | |
1416 | error_setg_file_open(errp, errno, filename); | |
1417 | return; | |
1418 | } | |
1419 | ||
1420 | while (size != 0) { | |
1421 | l = sizeof(buf); | |
1422 | if (l > size) | |
1423 | l = size; | |
1424 | if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) { | |
1425 | error_setg(errp, "Invalid addr 0x%016" PRIx64 "specified", addr); | |
1426 | goto exit; | |
1427 | } | |
1428 | if (fwrite(buf, 1, l, f) != l) { | |
1429 | error_set(errp, QERR_IO_ERROR); | |
1430 | goto exit; | |
1431 | } | |
1432 | addr += l; | |
1433 | size -= l; | |
1434 | } | |
1435 | ||
1436 | exit: | |
1437 | fclose(f); | |
1438 | } | |
1439 | ||
1440 | void qmp_pmemsave(int64_t addr, int64_t size, const char *filename, | |
1441 | Error **errp) | |
1442 | { | |
1443 | FILE *f; | |
1444 | uint32_t l; | |
1445 | uint8_t buf[1024]; | |
1446 | ||
1447 | f = fopen(filename, "wb"); | |
1448 | if (!f) { | |
1449 | error_setg_file_open(errp, errno, filename); | |
1450 | return; | |
1451 | } | |
1452 | ||
1453 | while (size != 0) { | |
1454 | l = sizeof(buf); | |
1455 | if (l > size) | |
1456 | l = size; | |
1457 | cpu_physical_memory_read(addr, buf, l); | |
1458 | if (fwrite(buf, 1, l, f) != l) { | |
1459 | error_set(errp, QERR_IO_ERROR); | |
1460 | goto exit; | |
1461 | } | |
1462 | addr += l; | |
1463 | size -= l; | |
1464 | } | |
1465 | ||
1466 | exit: | |
1467 | fclose(f); | |
1468 | } | |
1469 | ||
1470 | void qmp_inject_nmi(Error **errp) | |
1471 | { | |
1472 | #if defined(TARGET_I386) | |
1473 | CPUState *cs; | |
1474 | ||
1475 | CPU_FOREACH(cs) { | |
1476 | X86CPU *cpu = X86_CPU(cs); | |
1477 | ||
1478 | if (!cpu->apic_state) { | |
1479 | cpu_interrupt(cs, CPU_INTERRUPT_NMI); | |
1480 | } else { | |
1481 | apic_deliver_nmi(cpu->apic_state); | |
1482 | } | |
1483 | } | |
1484 | #elif defined(TARGET_S390X) | |
1485 | CPUState *cs; | |
1486 | S390CPU *cpu; | |
1487 | ||
1488 | CPU_FOREACH(cs) { | |
1489 | cpu = S390_CPU(cs); | |
1490 | if (cpu->env.cpu_num == monitor_get_cpu_index()) { | |
1491 | if (s390_cpu_restart(S390_CPU(cs)) == -1) { | |
1492 | error_set(errp, QERR_UNSUPPORTED); | |
1493 | return; | |
1494 | } | |
1495 | break; | |
1496 | } | |
1497 | } | |
1498 | #else | |
1499 | error_set(errp, QERR_UNSUPPORTED); | |
1500 | #endif | |
1501 | } |