]>
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 "qemu/osdep.h" | |
27 | #include "qemu-common.h" | |
28 | #include "qemu/config-file.h" | |
29 | #include "cpu.h" | |
30 | #include "monitor/monitor.h" | |
31 | #include "qapi/qmp/qerror.h" | |
32 | #include "qemu/error-report.h" | |
33 | #include "sysemu/sysemu.h" | |
34 | #include "sysemu/block-backend.h" | |
35 | #include "exec/gdbstub.h" | |
36 | #include "sysemu/dma.h" | |
37 | #include "sysemu/hw_accel.h" | |
38 | #include "sysemu/kvm.h" | |
39 | #include "sysemu/hax.h" | |
40 | #include "qmp-commands.h" | |
41 | #include "exec/exec-all.h" | |
42 | ||
43 | #include "qemu/thread.h" | |
44 | #include "sysemu/cpus.h" | |
45 | #include "sysemu/qtest.h" | |
46 | #include "qemu/main-loop.h" | |
47 | #include "qemu/bitmap.h" | |
48 | #include "qemu/seqlock.h" | |
49 | #include "tcg.h" | |
50 | #include "qapi-event.h" | |
51 | #include "hw/nmi.h" | |
52 | #include "sysemu/replay.h" | |
53 | #include "hw/boards.h" | |
54 | ||
55 | #ifdef CONFIG_LINUX | |
56 | ||
57 | #include <sys/prctl.h> | |
58 | ||
59 | #ifndef PR_MCE_KILL | |
60 | #define PR_MCE_KILL 33 | |
61 | #endif | |
62 | ||
63 | #ifndef PR_MCE_KILL_SET | |
64 | #define PR_MCE_KILL_SET 1 | |
65 | #endif | |
66 | ||
67 | #ifndef PR_MCE_KILL_EARLY | |
68 | #define PR_MCE_KILL_EARLY 1 | |
69 | #endif | |
70 | ||
71 | #endif /* CONFIG_LINUX */ | |
72 | ||
73 | int64_t max_delay; | |
74 | int64_t max_advance; | |
75 | ||
76 | /* vcpu throttling controls */ | |
77 | static QEMUTimer *throttle_timer; | |
78 | static unsigned int throttle_percentage; | |
79 | ||
80 | #define CPU_THROTTLE_PCT_MIN 1 | |
81 | #define CPU_THROTTLE_PCT_MAX 99 | |
82 | #define CPU_THROTTLE_TIMESLICE_NS 10000000 | |
83 | ||
84 | bool cpu_is_stopped(CPUState *cpu) | |
85 | { | |
86 | return cpu->stopped || !runstate_is_running(); | |
87 | } | |
88 | ||
89 | static bool cpu_thread_is_idle(CPUState *cpu) | |
90 | { | |
91 | if (cpu->stop || cpu->queued_work_first) { | |
92 | return false; | |
93 | } | |
94 | if (cpu_is_stopped(cpu)) { | |
95 | return true; | |
96 | } | |
97 | if (!cpu->halted || cpu_has_work(cpu) || | |
98 | kvm_halt_in_kernel()) { | |
99 | return false; | |
100 | } | |
101 | return true; | |
102 | } | |
103 | ||
104 | static bool all_cpu_threads_idle(void) | |
105 | { | |
106 | CPUState *cpu; | |
107 | ||
108 | CPU_FOREACH(cpu) { | |
109 | if (!cpu_thread_is_idle(cpu)) { | |
110 | return false; | |
111 | } | |
112 | } | |
113 | return true; | |
114 | } | |
115 | ||
116 | /***********************************************************/ | |
117 | /* guest cycle counter */ | |
118 | ||
119 | /* Protected by TimersState seqlock */ | |
120 | ||
121 | static bool icount_sleep = true; | |
122 | static int64_t vm_clock_warp_start = -1; | |
123 | /* Conversion factor from emulated instructions to virtual clock ticks. */ | |
124 | static int icount_time_shift; | |
125 | /* Arbitrarily pick 1MIPS as the minimum allowable speed. */ | |
126 | #define MAX_ICOUNT_SHIFT 10 | |
127 | ||
128 | static QEMUTimer *icount_rt_timer; | |
129 | static QEMUTimer *icount_vm_timer; | |
130 | static QEMUTimer *icount_warp_timer; | |
131 | ||
132 | typedef struct TimersState { | |
133 | /* Protected by BQL. */ | |
134 | int64_t cpu_ticks_prev; | |
135 | int64_t cpu_ticks_offset; | |
136 | ||
137 | /* cpu_clock_offset can be read out of BQL, so protect it with | |
138 | * this lock. | |
139 | */ | |
140 | QemuSeqLock vm_clock_seqlock; | |
141 | int64_t cpu_clock_offset; | |
142 | int32_t cpu_ticks_enabled; | |
143 | int64_t dummy; | |
144 | ||
145 | /* Compensate for varying guest execution speed. */ | |
146 | int64_t qemu_icount_bias; | |
147 | /* Only written by TCG thread */ | |
148 | int64_t qemu_icount; | |
149 | } TimersState; | |
150 | ||
151 | static TimersState timers_state; | |
152 | bool mttcg_enabled; | |
153 | ||
154 | /* | |
155 | * We default to false if we know other options have been enabled | |
156 | * which are currently incompatible with MTTCG. Otherwise when each | |
157 | * guest (target) has been updated to support: | |
158 | * - atomic instructions | |
159 | * - memory ordering primitives (barriers) | |
160 | * they can set the appropriate CONFIG flags in ${target}-softmmu.mak | |
161 | * | |
162 | * Once a guest architecture has been converted to the new primitives | |
163 | * there are two remaining limitations to check. | |
164 | * | |
165 | * - The guest can't be oversized (e.g. 64 bit guest on 32 bit host) | |
166 | * - The host must have a stronger memory order than the guest | |
167 | * | |
168 | * It may be possible in future to support strong guests on weak hosts | |
169 | * but that will require tagging all load/stores in a guest with their | |
170 | * implicit memory order requirements which would likely slow things | |
171 | * down a lot. | |
172 | */ | |
173 | ||
174 | static bool check_tcg_memory_orders_compatible(void) | |
175 | { | |
176 | #if defined(TCG_GUEST_DEFAULT_MO) && defined(TCG_TARGET_DEFAULT_MO) | |
177 | return (TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO) == 0; | |
178 | #else | |
179 | return false; | |
180 | #endif | |
181 | } | |
182 | ||
183 | static bool default_mttcg_enabled(void) | |
184 | { | |
185 | if (use_icount || TCG_OVERSIZED_GUEST) { | |
186 | return false; | |
187 | } else { | |
188 | #ifdef TARGET_SUPPORTS_MTTCG | |
189 | return check_tcg_memory_orders_compatible(); | |
190 | #else | |
191 | return false; | |
192 | #endif | |
193 | } | |
194 | } | |
195 | ||
196 | void qemu_tcg_configure(QemuOpts *opts, Error **errp) | |
197 | { | |
198 | const char *t = qemu_opt_get(opts, "thread"); | |
199 | if (t) { | |
200 | if (strcmp(t, "multi") == 0) { | |
201 | if (TCG_OVERSIZED_GUEST) { | |
202 | error_setg(errp, "No MTTCG when guest word size > hosts"); | |
203 | } else if (use_icount) { | |
204 | error_setg(errp, "No MTTCG when icount is enabled"); | |
205 | } else { | |
206 | #ifndef TARGET_SUPPORTS_MTTCG | |
207 | error_report("Guest not yet converted to MTTCG - " | |
208 | "you may get unexpected results"); | |
209 | #endif | |
210 | if (!check_tcg_memory_orders_compatible()) { | |
211 | error_report("Guest expects a stronger memory ordering " | |
212 | "than the host provides"); | |
213 | error_printf("This may cause strange/hard to debug errors\n"); | |
214 | } | |
215 | mttcg_enabled = true; | |
216 | } | |
217 | } else if (strcmp(t, "single") == 0) { | |
218 | mttcg_enabled = false; | |
219 | } else { | |
220 | error_setg(errp, "Invalid 'thread' setting %s", t); | |
221 | } | |
222 | } else { | |
223 | mttcg_enabled = default_mttcg_enabled(); | |
224 | } | |
225 | } | |
226 | ||
227 | /* The current number of executed instructions is based on what we | |
228 | * originally budgeted minus the current state of the decrementing | |
229 | * icount counters in extra/u16.low. | |
230 | */ | |
231 | static int64_t cpu_get_icount_executed(CPUState *cpu) | |
232 | { | |
233 | return cpu->icount_budget - (cpu->icount_decr.u16.low + cpu->icount_extra); | |
234 | } | |
235 | ||
236 | /* | |
237 | * Update the global shared timer_state.qemu_icount to take into | |
238 | * account executed instructions. This is done by the TCG vCPU | |
239 | * thread so the main-loop can see time has moved forward. | |
240 | */ | |
241 | void cpu_update_icount(CPUState *cpu) | |
242 | { | |
243 | int64_t executed = cpu_get_icount_executed(cpu); | |
244 | cpu->icount_budget -= executed; | |
245 | ||
246 | #ifdef CONFIG_ATOMIC64 | |
247 | atomic_set__nocheck(&timers_state.qemu_icount, | |
248 | atomic_read__nocheck(&timers_state.qemu_icount) + | |
249 | executed); | |
250 | #else /* FIXME: we need 64bit atomics to do this safely */ | |
251 | timers_state.qemu_icount += executed; | |
252 | #endif | |
253 | } | |
254 | ||
255 | int64_t cpu_get_icount_raw(void) | |
256 | { | |
257 | CPUState *cpu = current_cpu; | |
258 | ||
259 | if (cpu && cpu->running) { | |
260 | if (!cpu->can_do_io) { | |
261 | fprintf(stderr, "Bad icount read\n"); | |
262 | exit(1); | |
263 | } | |
264 | /* Take into account what has run */ | |
265 | cpu_update_icount(cpu); | |
266 | } | |
267 | #ifdef CONFIG_ATOMIC64 | |
268 | return atomic_read__nocheck(&timers_state.qemu_icount); | |
269 | #else /* FIXME: we need 64bit atomics to do this safely */ | |
270 | return timers_state.qemu_icount; | |
271 | #endif | |
272 | } | |
273 | ||
274 | /* Return the virtual CPU time, based on the instruction counter. */ | |
275 | static int64_t cpu_get_icount_locked(void) | |
276 | { | |
277 | int64_t icount = cpu_get_icount_raw(); | |
278 | return timers_state.qemu_icount_bias + cpu_icount_to_ns(icount); | |
279 | } | |
280 | ||
281 | int64_t cpu_get_icount(void) | |
282 | { | |
283 | int64_t icount; | |
284 | unsigned start; | |
285 | ||
286 | do { | |
287 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
288 | icount = cpu_get_icount_locked(); | |
289 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
290 | ||
291 | return icount; | |
292 | } | |
293 | ||
294 | int64_t cpu_icount_to_ns(int64_t icount) | |
295 | { | |
296 | return icount << icount_time_shift; | |
297 | } | |
298 | ||
299 | /* return the time elapsed in VM between vm_start and vm_stop. Unless | |
300 | * icount is active, cpu_get_ticks() uses units of the host CPU cycle | |
301 | * counter. | |
302 | * | |
303 | * Caller must hold the BQL | |
304 | */ | |
305 | int64_t cpu_get_ticks(void) | |
306 | { | |
307 | int64_t ticks; | |
308 | ||
309 | if (use_icount) { | |
310 | return cpu_get_icount(); | |
311 | } | |
312 | ||
313 | ticks = timers_state.cpu_ticks_offset; | |
314 | if (timers_state.cpu_ticks_enabled) { | |
315 | ticks += cpu_get_host_ticks(); | |
316 | } | |
317 | ||
318 | if (timers_state.cpu_ticks_prev > ticks) { | |
319 | /* Note: non increasing ticks may happen if the host uses | |
320 | software suspend */ | |
321 | timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks; | |
322 | ticks = timers_state.cpu_ticks_prev; | |
323 | } | |
324 | ||
325 | timers_state.cpu_ticks_prev = ticks; | |
326 | return ticks; | |
327 | } | |
328 | ||
329 | static int64_t cpu_get_clock_locked(void) | |
330 | { | |
331 | int64_t time; | |
332 | ||
333 | time = timers_state.cpu_clock_offset; | |
334 | if (timers_state.cpu_ticks_enabled) { | |
335 | time += get_clock(); | |
336 | } | |
337 | ||
338 | return time; | |
339 | } | |
340 | ||
341 | /* Return the monotonic time elapsed in VM, i.e., | |
342 | * the time between vm_start and vm_stop | |
343 | */ | |
344 | int64_t cpu_get_clock(void) | |
345 | { | |
346 | int64_t ti; | |
347 | unsigned start; | |
348 | ||
349 | do { | |
350 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
351 | ti = cpu_get_clock_locked(); | |
352 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
353 | ||
354 | return ti; | |
355 | } | |
356 | ||
357 | /* enable cpu_get_ticks() | |
358 | * Caller must hold BQL which serves as mutex for vm_clock_seqlock. | |
359 | */ | |
360 | void cpu_enable_ticks(void) | |
361 | { | |
362 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ | |
363 | seqlock_write_begin(&timers_state.vm_clock_seqlock); | |
364 | if (!timers_state.cpu_ticks_enabled) { | |
365 | timers_state.cpu_ticks_offset -= cpu_get_host_ticks(); | |
366 | timers_state.cpu_clock_offset -= get_clock(); | |
367 | timers_state.cpu_ticks_enabled = 1; | |
368 | } | |
369 | seqlock_write_end(&timers_state.vm_clock_seqlock); | |
370 | } | |
371 | ||
372 | /* disable cpu_get_ticks() : the clock is stopped. You must not call | |
373 | * cpu_get_ticks() after that. | |
374 | * Caller must hold BQL which serves as mutex for vm_clock_seqlock. | |
375 | */ | |
376 | void cpu_disable_ticks(void) | |
377 | { | |
378 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ | |
379 | seqlock_write_begin(&timers_state.vm_clock_seqlock); | |
380 | if (timers_state.cpu_ticks_enabled) { | |
381 | timers_state.cpu_ticks_offset += cpu_get_host_ticks(); | |
382 | timers_state.cpu_clock_offset = cpu_get_clock_locked(); | |
383 | timers_state.cpu_ticks_enabled = 0; | |
384 | } | |
385 | seqlock_write_end(&timers_state.vm_clock_seqlock); | |
386 | } | |
387 | ||
388 | /* Correlation between real and virtual time is always going to be | |
389 | fairly approximate, so ignore small variation. | |
390 | When the guest is idle real and virtual time will be aligned in | |
391 | the IO wait loop. */ | |
392 | #define ICOUNT_WOBBLE (NANOSECONDS_PER_SECOND / 10) | |
393 | ||
394 | static void icount_adjust(void) | |
395 | { | |
396 | int64_t cur_time; | |
397 | int64_t cur_icount; | |
398 | int64_t delta; | |
399 | ||
400 | /* Protected by TimersState mutex. */ | |
401 | static int64_t last_delta; | |
402 | ||
403 | /* If the VM is not running, then do nothing. */ | |
404 | if (!runstate_is_running()) { | |
405 | return; | |
406 | } | |
407 | ||
408 | seqlock_write_begin(&timers_state.vm_clock_seqlock); | |
409 | cur_time = cpu_get_clock_locked(); | |
410 | cur_icount = cpu_get_icount_locked(); | |
411 | ||
412 | delta = cur_icount - cur_time; | |
413 | /* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */ | |
414 | if (delta > 0 | |
415 | && last_delta + ICOUNT_WOBBLE < delta * 2 | |
416 | && icount_time_shift > 0) { | |
417 | /* The guest is getting too far ahead. Slow time down. */ | |
418 | icount_time_shift--; | |
419 | } | |
420 | if (delta < 0 | |
421 | && last_delta - ICOUNT_WOBBLE > delta * 2 | |
422 | && icount_time_shift < MAX_ICOUNT_SHIFT) { | |
423 | /* The guest is getting too far behind. Speed time up. */ | |
424 | icount_time_shift++; | |
425 | } | |
426 | last_delta = delta; | |
427 | timers_state.qemu_icount_bias = cur_icount | |
428 | - (timers_state.qemu_icount << icount_time_shift); | |
429 | seqlock_write_end(&timers_state.vm_clock_seqlock); | |
430 | } | |
431 | ||
432 | static void icount_adjust_rt(void *opaque) | |
433 | { | |
434 | timer_mod(icount_rt_timer, | |
435 | qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000); | |
436 | icount_adjust(); | |
437 | } | |
438 | ||
439 | static void icount_adjust_vm(void *opaque) | |
440 | { | |
441 | timer_mod(icount_vm_timer, | |
442 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
443 | NANOSECONDS_PER_SECOND / 10); | |
444 | icount_adjust(); | |
445 | } | |
446 | ||
447 | static int64_t qemu_icount_round(int64_t count) | |
448 | { | |
449 | return (count + (1 << icount_time_shift) - 1) >> icount_time_shift; | |
450 | } | |
451 | ||
452 | static void icount_warp_rt(void) | |
453 | { | |
454 | unsigned seq; | |
455 | int64_t warp_start; | |
456 | ||
457 | /* The icount_warp_timer is rescheduled soon after vm_clock_warp_start | |
458 | * changes from -1 to another value, so the race here is okay. | |
459 | */ | |
460 | do { | |
461 | seq = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
462 | warp_start = vm_clock_warp_start; | |
463 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, seq)); | |
464 | ||
465 | if (warp_start == -1) { | |
466 | return; | |
467 | } | |
468 | ||
469 | seqlock_write_begin(&timers_state.vm_clock_seqlock); | |
470 | if (runstate_is_running()) { | |
471 | int64_t clock = REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT, | |
472 | cpu_get_clock_locked()); | |
473 | int64_t warp_delta; | |
474 | ||
475 | warp_delta = clock - vm_clock_warp_start; | |
476 | if (use_icount == 2) { | |
477 | /* | |
478 | * In adaptive mode, do not let QEMU_CLOCK_VIRTUAL run too | |
479 | * far ahead of real time. | |
480 | */ | |
481 | int64_t cur_icount = cpu_get_icount_locked(); | |
482 | int64_t delta = clock - cur_icount; | |
483 | warp_delta = MIN(warp_delta, delta); | |
484 | } | |
485 | timers_state.qemu_icount_bias += warp_delta; | |
486 | } | |
487 | vm_clock_warp_start = -1; | |
488 | seqlock_write_end(&timers_state.vm_clock_seqlock); | |
489 | ||
490 | if (qemu_clock_expired(QEMU_CLOCK_VIRTUAL)) { | |
491 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
492 | } | |
493 | } | |
494 | ||
495 | static void icount_timer_cb(void *opaque) | |
496 | { | |
497 | /* No need for a checkpoint because the timer already synchronizes | |
498 | * with CHECKPOINT_CLOCK_VIRTUAL_RT. | |
499 | */ | |
500 | icount_warp_rt(); | |
501 | } | |
502 | ||
503 | void qtest_clock_warp(int64_t dest) | |
504 | { | |
505 | int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); | |
506 | AioContext *aio_context; | |
507 | assert(qtest_enabled()); | |
508 | aio_context = qemu_get_aio_context(); | |
509 | while (clock < dest) { | |
510 | int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
511 | int64_t warp = qemu_soonest_timeout(dest - clock, deadline); | |
512 | ||
513 | seqlock_write_begin(&timers_state.vm_clock_seqlock); | |
514 | timers_state.qemu_icount_bias += warp; | |
515 | seqlock_write_end(&timers_state.vm_clock_seqlock); | |
516 | ||
517 | qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL); | |
518 | timerlist_run_timers(aio_context->tlg.tl[QEMU_CLOCK_VIRTUAL]); | |
519 | clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); | |
520 | } | |
521 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
522 | } | |
523 | ||
524 | void qemu_start_warp_timer(void) | |
525 | { | |
526 | int64_t clock; | |
527 | int64_t deadline; | |
528 | ||
529 | if (!use_icount) { | |
530 | return; | |
531 | } | |
532 | ||
533 | /* Nothing to do if the VM is stopped: QEMU_CLOCK_VIRTUAL timers | |
534 | * do not fire, so computing the deadline does not make sense. | |
535 | */ | |
536 | if (!runstate_is_running()) { | |
537 | return; | |
538 | } | |
539 | ||
540 | /* warp clock deterministically in record/replay mode */ | |
541 | if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_START)) { | |
542 | return; | |
543 | } | |
544 | ||
545 | if (!all_cpu_threads_idle()) { | |
546 | return; | |
547 | } | |
548 | ||
549 | if (qtest_enabled()) { | |
550 | /* When testing, qtest commands advance icount. */ | |
551 | return; | |
552 | } | |
553 | ||
554 | /* We want to use the earliest deadline from ALL vm_clocks */ | |
555 | clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT); | |
556 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
557 | if (deadline < 0) { | |
558 | static bool notified; | |
559 | if (!icount_sleep && !notified) { | |
560 | warn_report("icount sleep disabled and no active timers"); | |
561 | notified = true; | |
562 | } | |
563 | return; | |
564 | } | |
565 | ||
566 | if (deadline > 0) { | |
567 | /* | |
568 | * Ensure QEMU_CLOCK_VIRTUAL proceeds even when the virtual CPU goes to | |
569 | * sleep. Otherwise, the CPU might be waiting for a future timer | |
570 | * interrupt to wake it up, but the interrupt never comes because | |
571 | * the vCPU isn't running any insns and thus doesn't advance the | |
572 | * QEMU_CLOCK_VIRTUAL. | |
573 | */ | |
574 | if (!icount_sleep) { | |
575 | /* | |
576 | * We never let VCPUs sleep in no sleep icount mode. | |
577 | * If there is a pending QEMU_CLOCK_VIRTUAL timer we just advance | |
578 | * to the next QEMU_CLOCK_VIRTUAL event and notify it. | |
579 | * It is useful when we want a deterministic execution time, | |
580 | * isolated from host latencies. | |
581 | */ | |
582 | seqlock_write_begin(&timers_state.vm_clock_seqlock); | |
583 | timers_state.qemu_icount_bias += deadline; | |
584 | seqlock_write_end(&timers_state.vm_clock_seqlock); | |
585 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
586 | } else { | |
587 | /* | |
588 | * We do stop VCPUs and only advance QEMU_CLOCK_VIRTUAL after some | |
589 | * "real" time, (related to the time left until the next event) has | |
590 | * passed. The QEMU_CLOCK_VIRTUAL_RT clock will do this. | |
591 | * This avoids that the warps are visible externally; for example, | |
592 | * you will not be sending network packets continuously instead of | |
593 | * every 100ms. | |
594 | */ | |
595 | seqlock_write_begin(&timers_state.vm_clock_seqlock); | |
596 | if (vm_clock_warp_start == -1 || vm_clock_warp_start > clock) { | |
597 | vm_clock_warp_start = clock; | |
598 | } | |
599 | seqlock_write_end(&timers_state.vm_clock_seqlock); | |
600 | timer_mod_anticipate(icount_warp_timer, clock + deadline); | |
601 | } | |
602 | } else if (deadline == 0) { | |
603 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
604 | } | |
605 | } | |
606 | ||
607 | static void qemu_account_warp_timer(void) | |
608 | { | |
609 | if (!use_icount || !icount_sleep) { | |
610 | return; | |
611 | } | |
612 | ||
613 | /* Nothing to do if the VM is stopped: QEMU_CLOCK_VIRTUAL timers | |
614 | * do not fire, so computing the deadline does not make sense. | |
615 | */ | |
616 | if (!runstate_is_running()) { | |
617 | return; | |
618 | } | |
619 | ||
620 | /* warp clock deterministically in record/replay mode */ | |
621 | if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_ACCOUNT)) { | |
622 | return; | |
623 | } | |
624 | ||
625 | timer_del(icount_warp_timer); | |
626 | icount_warp_rt(); | |
627 | } | |
628 | ||
629 | static bool icount_state_needed(void *opaque) | |
630 | { | |
631 | return use_icount; | |
632 | } | |
633 | ||
634 | /* | |
635 | * This is a subsection for icount migration. | |
636 | */ | |
637 | static const VMStateDescription icount_vmstate_timers = { | |
638 | .name = "timer/icount", | |
639 | .version_id = 1, | |
640 | .minimum_version_id = 1, | |
641 | .needed = icount_state_needed, | |
642 | .fields = (VMStateField[]) { | |
643 | VMSTATE_INT64(qemu_icount_bias, TimersState), | |
644 | VMSTATE_INT64(qemu_icount, TimersState), | |
645 | VMSTATE_END_OF_LIST() | |
646 | } | |
647 | }; | |
648 | ||
649 | static const VMStateDescription vmstate_timers = { | |
650 | .name = "timer", | |
651 | .version_id = 2, | |
652 | .minimum_version_id = 1, | |
653 | .fields = (VMStateField[]) { | |
654 | VMSTATE_INT64(cpu_ticks_offset, TimersState), | |
655 | VMSTATE_INT64(dummy, TimersState), | |
656 | VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2), | |
657 | VMSTATE_END_OF_LIST() | |
658 | }, | |
659 | .subsections = (const VMStateDescription*[]) { | |
660 | &icount_vmstate_timers, | |
661 | NULL | |
662 | } | |
663 | }; | |
664 | ||
665 | static void cpu_throttle_thread(CPUState *cpu, run_on_cpu_data opaque) | |
666 | { | |
667 | double pct; | |
668 | double throttle_ratio; | |
669 | long sleeptime_ns; | |
670 | ||
671 | if (!cpu_throttle_get_percentage()) { | |
672 | return; | |
673 | } | |
674 | ||
675 | pct = (double)cpu_throttle_get_percentage()/100; | |
676 | throttle_ratio = pct / (1 - pct); | |
677 | sleeptime_ns = (long)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS); | |
678 | ||
679 | qemu_mutex_unlock_iothread(); | |
680 | g_usleep(sleeptime_ns / 1000); /* Convert ns to us for usleep call */ | |
681 | qemu_mutex_lock_iothread(); | |
682 | atomic_set(&cpu->throttle_thread_scheduled, 0); | |
683 | } | |
684 | ||
685 | static void cpu_throttle_timer_tick(void *opaque) | |
686 | { | |
687 | CPUState *cpu; | |
688 | double pct; | |
689 | ||
690 | /* Stop the timer if needed */ | |
691 | if (!cpu_throttle_get_percentage()) { | |
692 | return; | |
693 | } | |
694 | CPU_FOREACH(cpu) { | |
695 | if (!atomic_xchg(&cpu->throttle_thread_scheduled, 1)) { | |
696 | async_run_on_cpu(cpu, cpu_throttle_thread, | |
697 | RUN_ON_CPU_NULL); | |
698 | } | |
699 | } | |
700 | ||
701 | pct = (double)cpu_throttle_get_percentage()/100; | |
702 | timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) + | |
703 | CPU_THROTTLE_TIMESLICE_NS / (1-pct)); | |
704 | } | |
705 | ||
706 | void cpu_throttle_set(int new_throttle_pct) | |
707 | { | |
708 | /* Ensure throttle percentage is within valid range */ | |
709 | new_throttle_pct = MIN(new_throttle_pct, CPU_THROTTLE_PCT_MAX); | |
710 | new_throttle_pct = MAX(new_throttle_pct, CPU_THROTTLE_PCT_MIN); | |
711 | ||
712 | atomic_set(&throttle_percentage, new_throttle_pct); | |
713 | ||
714 | timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) + | |
715 | CPU_THROTTLE_TIMESLICE_NS); | |
716 | } | |
717 | ||
718 | void cpu_throttle_stop(void) | |
719 | { | |
720 | atomic_set(&throttle_percentage, 0); | |
721 | } | |
722 | ||
723 | bool cpu_throttle_active(void) | |
724 | { | |
725 | return (cpu_throttle_get_percentage() != 0); | |
726 | } | |
727 | ||
728 | int cpu_throttle_get_percentage(void) | |
729 | { | |
730 | return atomic_read(&throttle_percentage); | |
731 | } | |
732 | ||
733 | void cpu_ticks_init(void) | |
734 | { | |
735 | seqlock_init(&timers_state.vm_clock_seqlock); | |
736 | vmstate_register(NULL, 0, &vmstate_timers, &timers_state); | |
737 | throttle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT, | |
738 | cpu_throttle_timer_tick, NULL); | |
739 | } | |
740 | ||
741 | void configure_icount(QemuOpts *opts, Error **errp) | |
742 | { | |
743 | const char *option; | |
744 | char *rem_str = NULL; | |
745 | ||
746 | option = qemu_opt_get(opts, "shift"); | |
747 | if (!option) { | |
748 | if (qemu_opt_get(opts, "align") != NULL) { | |
749 | error_setg(errp, "Please specify shift option when using align"); | |
750 | } | |
751 | return; | |
752 | } | |
753 | ||
754 | icount_sleep = qemu_opt_get_bool(opts, "sleep", true); | |
755 | if (icount_sleep) { | |
756 | icount_warp_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT, | |
757 | icount_timer_cb, NULL); | |
758 | } | |
759 | ||
760 | icount_align_option = qemu_opt_get_bool(opts, "align", false); | |
761 | ||
762 | if (icount_align_option && !icount_sleep) { | |
763 | error_setg(errp, "align=on and sleep=off are incompatible"); | |
764 | } | |
765 | if (strcmp(option, "auto") != 0) { | |
766 | errno = 0; | |
767 | icount_time_shift = strtol(option, &rem_str, 0); | |
768 | if (errno != 0 || *rem_str != '\0' || !strlen(option)) { | |
769 | error_setg(errp, "icount: Invalid shift value"); | |
770 | } | |
771 | use_icount = 1; | |
772 | return; | |
773 | } else if (icount_align_option) { | |
774 | error_setg(errp, "shift=auto and align=on are incompatible"); | |
775 | } else if (!icount_sleep) { | |
776 | error_setg(errp, "shift=auto and sleep=off are incompatible"); | |
777 | } | |
778 | ||
779 | use_icount = 2; | |
780 | ||
781 | /* 125MIPS seems a reasonable initial guess at the guest speed. | |
782 | It will be corrected fairly quickly anyway. */ | |
783 | icount_time_shift = 3; | |
784 | ||
785 | /* Have both realtime and virtual time triggers for speed adjustment. | |
786 | The realtime trigger catches emulated time passing too slowly, | |
787 | the virtual time trigger catches emulated time passing too fast. | |
788 | Realtime triggers occur even when idle, so use them less frequently | |
789 | than VM triggers. */ | |
790 | icount_rt_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL_RT, | |
791 | icount_adjust_rt, NULL); | |
792 | timer_mod(icount_rt_timer, | |
793 | qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000); | |
794 | icount_vm_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, | |
795 | icount_adjust_vm, NULL); | |
796 | timer_mod(icount_vm_timer, | |
797 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
798 | NANOSECONDS_PER_SECOND / 10); | |
799 | } | |
800 | ||
801 | /***********************************************************/ | |
802 | /* TCG vCPU kick timer | |
803 | * | |
804 | * The kick timer is responsible for moving single threaded vCPU | |
805 | * emulation on to the next vCPU. If more than one vCPU is running a | |
806 | * timer event with force a cpu->exit so the next vCPU can get | |
807 | * scheduled. | |
808 | * | |
809 | * The timer is removed if all vCPUs are idle and restarted again once | |
810 | * idleness is complete. | |
811 | */ | |
812 | ||
813 | static QEMUTimer *tcg_kick_vcpu_timer; | |
814 | static CPUState *tcg_current_rr_cpu; | |
815 | ||
816 | #define TCG_KICK_PERIOD (NANOSECONDS_PER_SECOND / 10) | |
817 | ||
818 | static inline int64_t qemu_tcg_next_kick(void) | |
819 | { | |
820 | return qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + TCG_KICK_PERIOD; | |
821 | } | |
822 | ||
823 | /* Kick the currently round-robin scheduled vCPU */ | |
824 | static void qemu_cpu_kick_rr_cpu(void) | |
825 | { | |
826 | CPUState *cpu; | |
827 | do { | |
828 | cpu = atomic_mb_read(&tcg_current_rr_cpu); | |
829 | if (cpu) { | |
830 | cpu_exit(cpu); | |
831 | } | |
832 | } while (cpu != atomic_mb_read(&tcg_current_rr_cpu)); | |
833 | } | |
834 | ||
835 | static void do_nothing(CPUState *cpu, run_on_cpu_data unused) | |
836 | { | |
837 | } | |
838 | ||
839 | void qemu_timer_notify_cb(void *opaque, QEMUClockType type) | |
840 | { | |
841 | if (!use_icount || type != QEMU_CLOCK_VIRTUAL) { | |
842 | qemu_notify_event(); | |
843 | return; | |
844 | } | |
845 | ||
846 | if (!qemu_in_vcpu_thread() && first_cpu) { | |
847 | /* qemu_cpu_kick is not enough to kick a halted CPU out of | |
848 | * qemu_tcg_wait_io_event. async_run_on_cpu, instead, | |
849 | * causes cpu_thread_is_idle to return false. This way, | |
850 | * handle_icount_deadline can run. | |
851 | */ | |
852 | async_run_on_cpu(first_cpu, do_nothing, RUN_ON_CPU_NULL); | |
853 | } | |
854 | } | |
855 | ||
856 | static void kick_tcg_thread(void *opaque) | |
857 | { | |
858 | timer_mod(tcg_kick_vcpu_timer, qemu_tcg_next_kick()); | |
859 | qemu_cpu_kick_rr_cpu(); | |
860 | } | |
861 | ||
862 | static void start_tcg_kick_timer(void) | |
863 | { | |
864 | if (!mttcg_enabled && !tcg_kick_vcpu_timer && CPU_NEXT(first_cpu)) { | |
865 | tcg_kick_vcpu_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, | |
866 | kick_tcg_thread, NULL); | |
867 | timer_mod(tcg_kick_vcpu_timer, qemu_tcg_next_kick()); | |
868 | } | |
869 | } | |
870 | ||
871 | static void stop_tcg_kick_timer(void) | |
872 | { | |
873 | if (tcg_kick_vcpu_timer) { | |
874 | timer_del(tcg_kick_vcpu_timer); | |
875 | tcg_kick_vcpu_timer = NULL; | |
876 | } | |
877 | } | |
878 | ||
879 | /***********************************************************/ | |
880 | void hw_error(const char *fmt, ...) | |
881 | { | |
882 | va_list ap; | |
883 | CPUState *cpu; | |
884 | ||
885 | va_start(ap, fmt); | |
886 | fprintf(stderr, "qemu: hardware error: "); | |
887 | vfprintf(stderr, fmt, ap); | |
888 | fprintf(stderr, "\n"); | |
889 | CPU_FOREACH(cpu) { | |
890 | fprintf(stderr, "CPU #%d:\n", cpu->cpu_index); | |
891 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU); | |
892 | } | |
893 | va_end(ap); | |
894 | abort(); | |
895 | } | |
896 | ||
897 | void cpu_synchronize_all_states(void) | |
898 | { | |
899 | CPUState *cpu; | |
900 | ||
901 | CPU_FOREACH(cpu) { | |
902 | cpu_synchronize_state(cpu); | |
903 | } | |
904 | } | |
905 | ||
906 | void cpu_synchronize_all_post_reset(void) | |
907 | { | |
908 | CPUState *cpu; | |
909 | ||
910 | CPU_FOREACH(cpu) { | |
911 | cpu_synchronize_post_reset(cpu); | |
912 | } | |
913 | } | |
914 | ||
915 | void cpu_synchronize_all_post_init(void) | |
916 | { | |
917 | CPUState *cpu; | |
918 | ||
919 | CPU_FOREACH(cpu) { | |
920 | cpu_synchronize_post_init(cpu); | |
921 | } | |
922 | } | |
923 | ||
924 | void cpu_synchronize_all_pre_loadvm(void) | |
925 | { | |
926 | CPUState *cpu; | |
927 | ||
928 | CPU_FOREACH(cpu) { | |
929 | cpu_synchronize_pre_loadvm(cpu); | |
930 | } | |
931 | } | |
932 | ||
933 | static int do_vm_stop(RunState state) | |
934 | { | |
935 | int ret = 0; | |
936 | ||
937 | if (runstate_is_running()) { | |
938 | cpu_disable_ticks(); | |
939 | pause_all_vcpus(); | |
940 | runstate_set(state); | |
941 | vm_state_notify(0, state); | |
942 | qapi_event_send_stop(&error_abort); | |
943 | } | |
944 | ||
945 | bdrv_drain_all(); | |
946 | replay_disable_events(); | |
947 | ret = bdrv_flush_all(); | |
948 | ||
949 | return ret; | |
950 | } | |
951 | ||
952 | static bool cpu_can_run(CPUState *cpu) | |
953 | { | |
954 | if (cpu->stop) { | |
955 | return false; | |
956 | } | |
957 | if (cpu_is_stopped(cpu)) { | |
958 | return false; | |
959 | } | |
960 | return true; | |
961 | } | |
962 | ||
963 | static void cpu_handle_guest_debug(CPUState *cpu) | |
964 | { | |
965 | gdb_set_stop_cpu(cpu); | |
966 | qemu_system_debug_request(); | |
967 | cpu->stopped = true; | |
968 | } | |
969 | ||
970 | #ifdef CONFIG_LINUX | |
971 | static void sigbus_reraise(void) | |
972 | { | |
973 | sigset_t set; | |
974 | struct sigaction action; | |
975 | ||
976 | memset(&action, 0, sizeof(action)); | |
977 | action.sa_handler = SIG_DFL; | |
978 | if (!sigaction(SIGBUS, &action, NULL)) { | |
979 | raise(SIGBUS); | |
980 | sigemptyset(&set); | |
981 | sigaddset(&set, SIGBUS); | |
982 | pthread_sigmask(SIG_UNBLOCK, &set, NULL); | |
983 | } | |
984 | perror("Failed to re-raise SIGBUS!\n"); | |
985 | abort(); | |
986 | } | |
987 | ||
988 | static void sigbus_handler(int n, siginfo_t *siginfo, void *ctx) | |
989 | { | |
990 | if (siginfo->si_code != BUS_MCEERR_AO && siginfo->si_code != BUS_MCEERR_AR) { | |
991 | sigbus_reraise(); | |
992 | } | |
993 | ||
994 | if (current_cpu) { | |
995 | /* Called asynchronously in VCPU thread. */ | |
996 | if (kvm_on_sigbus_vcpu(current_cpu, siginfo->si_code, siginfo->si_addr)) { | |
997 | sigbus_reraise(); | |
998 | } | |
999 | } else { | |
1000 | /* Called synchronously (via signalfd) in main thread. */ | |
1001 | if (kvm_on_sigbus(siginfo->si_code, siginfo->si_addr)) { | |
1002 | sigbus_reraise(); | |
1003 | } | |
1004 | } | |
1005 | } | |
1006 | ||
1007 | static void qemu_init_sigbus(void) | |
1008 | { | |
1009 | struct sigaction action; | |
1010 | ||
1011 | memset(&action, 0, sizeof(action)); | |
1012 | action.sa_flags = SA_SIGINFO; | |
1013 | action.sa_sigaction = sigbus_handler; | |
1014 | sigaction(SIGBUS, &action, NULL); | |
1015 | ||
1016 | prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0); | |
1017 | } | |
1018 | #else /* !CONFIG_LINUX */ | |
1019 | static void qemu_init_sigbus(void) | |
1020 | { | |
1021 | } | |
1022 | #endif /* !CONFIG_LINUX */ | |
1023 | ||
1024 | static QemuMutex qemu_global_mutex; | |
1025 | ||
1026 | static QemuThread io_thread; | |
1027 | ||
1028 | /* cpu creation */ | |
1029 | static QemuCond qemu_cpu_cond; | |
1030 | /* system init */ | |
1031 | static QemuCond qemu_pause_cond; | |
1032 | ||
1033 | void qemu_init_cpu_loop(void) | |
1034 | { | |
1035 | qemu_init_sigbus(); | |
1036 | qemu_cond_init(&qemu_cpu_cond); | |
1037 | qemu_cond_init(&qemu_pause_cond); | |
1038 | qemu_mutex_init(&qemu_global_mutex); | |
1039 | ||
1040 | qemu_thread_get_self(&io_thread); | |
1041 | } | |
1042 | ||
1043 | void run_on_cpu(CPUState *cpu, run_on_cpu_func func, run_on_cpu_data data) | |
1044 | { | |
1045 | do_run_on_cpu(cpu, func, data, &qemu_global_mutex); | |
1046 | } | |
1047 | ||
1048 | static void qemu_kvm_destroy_vcpu(CPUState *cpu) | |
1049 | { | |
1050 | if (kvm_destroy_vcpu(cpu) < 0) { | |
1051 | error_report("kvm_destroy_vcpu failed"); | |
1052 | exit(EXIT_FAILURE); | |
1053 | } | |
1054 | } | |
1055 | ||
1056 | static void qemu_tcg_destroy_vcpu(CPUState *cpu) | |
1057 | { | |
1058 | } | |
1059 | ||
1060 | static void qemu_wait_io_event_common(CPUState *cpu) | |
1061 | { | |
1062 | atomic_mb_set(&cpu->thread_kicked, false); | |
1063 | if (cpu->stop) { | |
1064 | cpu->stop = false; | |
1065 | cpu->stopped = true; | |
1066 | qemu_cond_broadcast(&qemu_pause_cond); | |
1067 | } | |
1068 | process_queued_cpu_work(cpu); | |
1069 | } | |
1070 | ||
1071 | static bool qemu_tcg_should_sleep(CPUState *cpu) | |
1072 | { | |
1073 | if (mttcg_enabled) { | |
1074 | return cpu_thread_is_idle(cpu); | |
1075 | } else { | |
1076 | return all_cpu_threads_idle(); | |
1077 | } | |
1078 | } | |
1079 | ||
1080 | static void qemu_tcg_wait_io_event(CPUState *cpu) | |
1081 | { | |
1082 | while (qemu_tcg_should_sleep(cpu)) { | |
1083 | stop_tcg_kick_timer(); | |
1084 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); | |
1085 | } | |
1086 | ||
1087 | start_tcg_kick_timer(); | |
1088 | ||
1089 | qemu_wait_io_event_common(cpu); | |
1090 | } | |
1091 | ||
1092 | static void qemu_kvm_wait_io_event(CPUState *cpu) | |
1093 | { | |
1094 | while (cpu_thread_is_idle(cpu)) { | |
1095 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); | |
1096 | } | |
1097 | ||
1098 | qemu_wait_io_event_common(cpu); | |
1099 | } | |
1100 | ||
1101 | static void *qemu_kvm_cpu_thread_fn(void *arg) | |
1102 | { | |
1103 | CPUState *cpu = arg; | |
1104 | int r; | |
1105 | ||
1106 | rcu_register_thread(); | |
1107 | ||
1108 | qemu_mutex_lock_iothread(); | |
1109 | qemu_thread_get_self(cpu->thread); | |
1110 | cpu->thread_id = qemu_get_thread_id(); | |
1111 | cpu->can_do_io = 1; | |
1112 | current_cpu = cpu; | |
1113 | ||
1114 | r = kvm_init_vcpu(cpu); | |
1115 | if (r < 0) { | |
1116 | fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r)); | |
1117 | exit(1); | |
1118 | } | |
1119 | ||
1120 | kvm_init_cpu_signals(cpu); | |
1121 | ||
1122 | /* signal CPU creation */ | |
1123 | cpu->created = true; | |
1124 | qemu_cond_signal(&qemu_cpu_cond); | |
1125 | ||
1126 | do { | |
1127 | if (cpu_can_run(cpu)) { | |
1128 | r = kvm_cpu_exec(cpu); | |
1129 | if (r == EXCP_DEBUG) { | |
1130 | cpu_handle_guest_debug(cpu); | |
1131 | } | |
1132 | } | |
1133 | qemu_kvm_wait_io_event(cpu); | |
1134 | } while (!cpu->unplug || cpu_can_run(cpu)); | |
1135 | ||
1136 | qemu_kvm_destroy_vcpu(cpu); | |
1137 | cpu->created = false; | |
1138 | qemu_cond_signal(&qemu_cpu_cond); | |
1139 | qemu_mutex_unlock_iothread(); | |
1140 | return NULL; | |
1141 | } | |
1142 | ||
1143 | static void *qemu_dummy_cpu_thread_fn(void *arg) | |
1144 | { | |
1145 | #ifdef _WIN32 | |
1146 | fprintf(stderr, "qtest is not supported under Windows\n"); | |
1147 | exit(1); | |
1148 | #else | |
1149 | CPUState *cpu = arg; | |
1150 | sigset_t waitset; | |
1151 | int r; | |
1152 | ||
1153 | rcu_register_thread(); | |
1154 | ||
1155 | qemu_mutex_lock_iothread(); | |
1156 | qemu_thread_get_self(cpu->thread); | |
1157 | cpu->thread_id = qemu_get_thread_id(); | |
1158 | cpu->can_do_io = 1; | |
1159 | current_cpu = cpu; | |
1160 | ||
1161 | sigemptyset(&waitset); | |
1162 | sigaddset(&waitset, SIG_IPI); | |
1163 | ||
1164 | /* signal CPU creation */ | |
1165 | cpu->created = true; | |
1166 | qemu_cond_signal(&qemu_cpu_cond); | |
1167 | ||
1168 | while (1) { | |
1169 | qemu_mutex_unlock_iothread(); | |
1170 | do { | |
1171 | int sig; | |
1172 | r = sigwait(&waitset, &sig); | |
1173 | } while (r == -1 && (errno == EAGAIN || errno == EINTR)); | |
1174 | if (r == -1) { | |
1175 | perror("sigwait"); | |
1176 | exit(1); | |
1177 | } | |
1178 | qemu_mutex_lock_iothread(); | |
1179 | qemu_wait_io_event_common(cpu); | |
1180 | } | |
1181 | ||
1182 | return NULL; | |
1183 | #endif | |
1184 | } | |
1185 | ||
1186 | static int64_t tcg_get_icount_limit(void) | |
1187 | { | |
1188 | int64_t deadline; | |
1189 | ||
1190 | if (replay_mode != REPLAY_MODE_PLAY) { | |
1191 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
1192 | ||
1193 | /* Maintain prior (possibly buggy) behaviour where if no deadline | |
1194 | * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than | |
1195 | * INT32_MAX nanoseconds ahead, we still use INT32_MAX | |
1196 | * nanoseconds. | |
1197 | */ | |
1198 | if ((deadline < 0) || (deadline > INT32_MAX)) { | |
1199 | deadline = INT32_MAX; | |
1200 | } | |
1201 | ||
1202 | return qemu_icount_round(deadline); | |
1203 | } else { | |
1204 | return replay_get_instructions(); | |
1205 | } | |
1206 | } | |
1207 | ||
1208 | static void handle_icount_deadline(void) | |
1209 | { | |
1210 | assert(qemu_in_vcpu_thread()); | |
1211 | if (use_icount) { | |
1212 | int64_t deadline = | |
1213 | qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
1214 | ||
1215 | if (deadline == 0) { | |
1216 | /* Wake up other AioContexts. */ | |
1217 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
1218 | qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL); | |
1219 | } | |
1220 | } | |
1221 | } | |
1222 | ||
1223 | static void prepare_icount_for_run(CPUState *cpu) | |
1224 | { | |
1225 | if (use_icount) { | |
1226 | int insns_left; | |
1227 | ||
1228 | /* These should always be cleared by process_icount_data after | |
1229 | * each vCPU execution. However u16.high can be raised | |
1230 | * asynchronously by cpu_exit/cpu_interrupt/tcg_handle_interrupt | |
1231 | */ | |
1232 | g_assert(cpu->icount_decr.u16.low == 0); | |
1233 | g_assert(cpu->icount_extra == 0); | |
1234 | ||
1235 | cpu->icount_budget = tcg_get_icount_limit(); | |
1236 | insns_left = MIN(0xffff, cpu->icount_budget); | |
1237 | cpu->icount_decr.u16.low = insns_left; | |
1238 | cpu->icount_extra = cpu->icount_budget - insns_left; | |
1239 | } | |
1240 | } | |
1241 | ||
1242 | static void process_icount_data(CPUState *cpu) | |
1243 | { | |
1244 | if (use_icount) { | |
1245 | /* Account for executed instructions */ | |
1246 | cpu_update_icount(cpu); | |
1247 | ||
1248 | /* Reset the counters */ | |
1249 | cpu->icount_decr.u16.low = 0; | |
1250 | cpu->icount_extra = 0; | |
1251 | cpu->icount_budget = 0; | |
1252 | ||
1253 | replay_account_executed_instructions(); | |
1254 | } | |
1255 | } | |
1256 | ||
1257 | ||
1258 | static int tcg_cpu_exec(CPUState *cpu) | |
1259 | { | |
1260 | int ret; | |
1261 | #ifdef CONFIG_PROFILER | |
1262 | int64_t ti; | |
1263 | #endif | |
1264 | ||
1265 | #ifdef CONFIG_PROFILER | |
1266 | ti = profile_getclock(); | |
1267 | #endif | |
1268 | qemu_mutex_unlock_iothread(); | |
1269 | cpu_exec_start(cpu); | |
1270 | ret = cpu_exec(cpu); | |
1271 | cpu_exec_end(cpu); | |
1272 | qemu_mutex_lock_iothread(); | |
1273 | #ifdef CONFIG_PROFILER | |
1274 | tcg_time += profile_getclock() - ti; | |
1275 | #endif | |
1276 | return ret; | |
1277 | } | |
1278 | ||
1279 | /* Destroy any remaining vCPUs which have been unplugged and have | |
1280 | * finished running | |
1281 | */ | |
1282 | static void deal_with_unplugged_cpus(void) | |
1283 | { | |
1284 | CPUState *cpu; | |
1285 | ||
1286 | CPU_FOREACH(cpu) { | |
1287 | if (cpu->unplug && !cpu_can_run(cpu)) { | |
1288 | qemu_tcg_destroy_vcpu(cpu); | |
1289 | cpu->created = false; | |
1290 | qemu_cond_signal(&qemu_cpu_cond); | |
1291 | break; | |
1292 | } | |
1293 | } | |
1294 | } | |
1295 | ||
1296 | /* Single-threaded TCG | |
1297 | * | |
1298 | * In the single-threaded case each vCPU is simulated in turn. If | |
1299 | * there is more than a single vCPU we create a simple timer to kick | |
1300 | * the vCPU and ensure we don't get stuck in a tight loop in one vCPU. | |
1301 | * This is done explicitly rather than relying on side-effects | |
1302 | * elsewhere. | |
1303 | */ | |
1304 | ||
1305 | static void *qemu_tcg_rr_cpu_thread_fn(void *arg) | |
1306 | { | |
1307 | CPUState *cpu = arg; | |
1308 | ||
1309 | rcu_register_thread(); | |
1310 | tcg_register_thread(); | |
1311 | ||
1312 | qemu_mutex_lock_iothread(); | |
1313 | qemu_thread_get_self(cpu->thread); | |
1314 | ||
1315 | CPU_FOREACH(cpu) { | |
1316 | cpu->thread_id = qemu_get_thread_id(); | |
1317 | cpu->created = true; | |
1318 | cpu->can_do_io = 1; | |
1319 | } | |
1320 | qemu_cond_signal(&qemu_cpu_cond); | |
1321 | ||
1322 | /* wait for initial kick-off after machine start */ | |
1323 | while (first_cpu->stopped) { | |
1324 | qemu_cond_wait(first_cpu->halt_cond, &qemu_global_mutex); | |
1325 | ||
1326 | /* process any pending work */ | |
1327 | CPU_FOREACH(cpu) { | |
1328 | current_cpu = cpu; | |
1329 | qemu_wait_io_event_common(cpu); | |
1330 | } | |
1331 | } | |
1332 | ||
1333 | start_tcg_kick_timer(); | |
1334 | ||
1335 | cpu = first_cpu; | |
1336 | ||
1337 | /* process any pending work */ | |
1338 | cpu->exit_request = 1; | |
1339 | ||
1340 | while (1) { | |
1341 | /* Account partial waits to QEMU_CLOCK_VIRTUAL. */ | |
1342 | qemu_account_warp_timer(); | |
1343 | ||
1344 | /* Run the timers here. This is much more efficient than | |
1345 | * waking up the I/O thread and waiting for completion. | |
1346 | */ | |
1347 | handle_icount_deadline(); | |
1348 | ||
1349 | if (!cpu) { | |
1350 | cpu = first_cpu; | |
1351 | } | |
1352 | ||
1353 | while (cpu && !cpu->queued_work_first && !cpu->exit_request) { | |
1354 | ||
1355 | atomic_mb_set(&tcg_current_rr_cpu, cpu); | |
1356 | current_cpu = cpu; | |
1357 | ||
1358 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, | |
1359 | (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0); | |
1360 | ||
1361 | if (cpu_can_run(cpu)) { | |
1362 | int r; | |
1363 | ||
1364 | prepare_icount_for_run(cpu); | |
1365 | ||
1366 | r = tcg_cpu_exec(cpu); | |
1367 | ||
1368 | process_icount_data(cpu); | |
1369 | ||
1370 | if (r == EXCP_DEBUG) { | |
1371 | cpu_handle_guest_debug(cpu); | |
1372 | break; | |
1373 | } else if (r == EXCP_ATOMIC) { | |
1374 | qemu_mutex_unlock_iothread(); | |
1375 | cpu_exec_step_atomic(cpu); | |
1376 | qemu_mutex_lock_iothread(); | |
1377 | break; | |
1378 | } | |
1379 | } else if (cpu->stop) { | |
1380 | if (cpu->unplug) { | |
1381 | cpu = CPU_NEXT(cpu); | |
1382 | } | |
1383 | break; | |
1384 | } | |
1385 | ||
1386 | cpu = CPU_NEXT(cpu); | |
1387 | } /* while (cpu && !cpu->exit_request).. */ | |
1388 | ||
1389 | /* Does not need atomic_mb_set because a spurious wakeup is okay. */ | |
1390 | atomic_set(&tcg_current_rr_cpu, NULL); | |
1391 | ||
1392 | if (cpu && cpu->exit_request) { | |
1393 | atomic_mb_set(&cpu->exit_request, 0); | |
1394 | } | |
1395 | ||
1396 | qemu_tcg_wait_io_event(cpu ? cpu : QTAILQ_FIRST(&cpus)); | |
1397 | deal_with_unplugged_cpus(); | |
1398 | } | |
1399 | ||
1400 | return NULL; | |
1401 | } | |
1402 | ||
1403 | static void *qemu_hax_cpu_thread_fn(void *arg) | |
1404 | { | |
1405 | CPUState *cpu = arg; | |
1406 | int r; | |
1407 | ||
1408 | qemu_mutex_lock_iothread(); | |
1409 | qemu_thread_get_self(cpu->thread); | |
1410 | ||
1411 | cpu->thread_id = qemu_get_thread_id(); | |
1412 | cpu->created = true; | |
1413 | cpu->halted = 0; | |
1414 | current_cpu = cpu; | |
1415 | ||
1416 | hax_init_vcpu(cpu); | |
1417 | qemu_cond_signal(&qemu_cpu_cond); | |
1418 | ||
1419 | while (1) { | |
1420 | if (cpu_can_run(cpu)) { | |
1421 | r = hax_smp_cpu_exec(cpu); | |
1422 | if (r == EXCP_DEBUG) { | |
1423 | cpu_handle_guest_debug(cpu); | |
1424 | } | |
1425 | } | |
1426 | ||
1427 | while (cpu_thread_is_idle(cpu)) { | |
1428 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); | |
1429 | } | |
1430 | #ifdef _WIN32 | |
1431 | SleepEx(0, TRUE); | |
1432 | #endif | |
1433 | qemu_wait_io_event_common(cpu); | |
1434 | } | |
1435 | return NULL; | |
1436 | } | |
1437 | ||
1438 | #ifdef _WIN32 | |
1439 | static void CALLBACK dummy_apc_func(ULONG_PTR unused) | |
1440 | { | |
1441 | } | |
1442 | #endif | |
1443 | ||
1444 | /* Multi-threaded TCG | |
1445 | * | |
1446 | * In the multi-threaded case each vCPU has its own thread. The TLS | |
1447 | * variable current_cpu can be used deep in the code to find the | |
1448 | * current CPUState for a given thread. | |
1449 | */ | |
1450 | ||
1451 | static void *qemu_tcg_cpu_thread_fn(void *arg) | |
1452 | { | |
1453 | CPUState *cpu = arg; | |
1454 | ||
1455 | g_assert(!use_icount); | |
1456 | ||
1457 | rcu_register_thread(); | |
1458 | tcg_register_thread(); | |
1459 | ||
1460 | qemu_mutex_lock_iothread(); | |
1461 | qemu_thread_get_self(cpu->thread); | |
1462 | ||
1463 | cpu->thread_id = qemu_get_thread_id(); | |
1464 | cpu->created = true; | |
1465 | cpu->can_do_io = 1; | |
1466 | current_cpu = cpu; | |
1467 | qemu_cond_signal(&qemu_cpu_cond); | |
1468 | ||
1469 | /* process any pending work */ | |
1470 | cpu->exit_request = 1; | |
1471 | ||
1472 | while (1) { | |
1473 | if (cpu_can_run(cpu)) { | |
1474 | int r; | |
1475 | r = tcg_cpu_exec(cpu); | |
1476 | switch (r) { | |
1477 | case EXCP_DEBUG: | |
1478 | cpu_handle_guest_debug(cpu); | |
1479 | break; | |
1480 | case EXCP_HALTED: | |
1481 | /* during start-up the vCPU is reset and the thread is | |
1482 | * kicked several times. If we don't ensure we go back | |
1483 | * to sleep in the halted state we won't cleanly | |
1484 | * start-up when the vCPU is enabled. | |
1485 | * | |
1486 | * cpu->halted should ensure we sleep in wait_io_event | |
1487 | */ | |
1488 | g_assert(cpu->halted); | |
1489 | break; | |
1490 | case EXCP_ATOMIC: | |
1491 | qemu_mutex_unlock_iothread(); | |
1492 | cpu_exec_step_atomic(cpu); | |
1493 | qemu_mutex_lock_iothread(); | |
1494 | default: | |
1495 | /* Ignore everything else? */ | |
1496 | break; | |
1497 | } | |
1498 | } else if (cpu->unplug) { | |
1499 | qemu_tcg_destroy_vcpu(cpu); | |
1500 | cpu->created = false; | |
1501 | qemu_cond_signal(&qemu_cpu_cond); | |
1502 | qemu_mutex_unlock_iothread(); | |
1503 | return NULL; | |
1504 | } | |
1505 | ||
1506 | atomic_mb_set(&cpu->exit_request, 0); | |
1507 | qemu_tcg_wait_io_event(cpu); | |
1508 | } | |
1509 | ||
1510 | return NULL; | |
1511 | } | |
1512 | ||
1513 | static void qemu_cpu_kick_thread(CPUState *cpu) | |
1514 | { | |
1515 | #ifndef _WIN32 | |
1516 | int err; | |
1517 | ||
1518 | if (cpu->thread_kicked) { | |
1519 | return; | |
1520 | } | |
1521 | cpu->thread_kicked = true; | |
1522 | err = pthread_kill(cpu->thread->thread, SIG_IPI); | |
1523 | if (err) { | |
1524 | fprintf(stderr, "qemu:%s: %s", __func__, strerror(err)); | |
1525 | exit(1); | |
1526 | } | |
1527 | #else /* _WIN32 */ | |
1528 | if (!qemu_cpu_is_self(cpu)) { | |
1529 | if (!QueueUserAPC(dummy_apc_func, cpu->hThread, 0)) { | |
1530 | fprintf(stderr, "%s: QueueUserAPC failed with error %lu\n", | |
1531 | __func__, GetLastError()); | |
1532 | exit(1); | |
1533 | } | |
1534 | } | |
1535 | #endif | |
1536 | } | |
1537 | ||
1538 | void qemu_cpu_kick(CPUState *cpu) | |
1539 | { | |
1540 | qemu_cond_broadcast(cpu->halt_cond); | |
1541 | if (tcg_enabled()) { | |
1542 | cpu_exit(cpu); | |
1543 | /* NOP unless doing single-thread RR */ | |
1544 | qemu_cpu_kick_rr_cpu(); | |
1545 | } else { | |
1546 | if (hax_enabled()) { | |
1547 | /* | |
1548 | * FIXME: race condition with the exit_request check in | |
1549 | * hax_vcpu_hax_exec | |
1550 | */ | |
1551 | cpu->exit_request = 1; | |
1552 | } | |
1553 | qemu_cpu_kick_thread(cpu); | |
1554 | } | |
1555 | } | |
1556 | ||
1557 | void qemu_cpu_kick_self(void) | |
1558 | { | |
1559 | assert(current_cpu); | |
1560 | qemu_cpu_kick_thread(current_cpu); | |
1561 | } | |
1562 | ||
1563 | bool qemu_cpu_is_self(CPUState *cpu) | |
1564 | { | |
1565 | return qemu_thread_is_self(cpu->thread); | |
1566 | } | |
1567 | ||
1568 | bool qemu_in_vcpu_thread(void) | |
1569 | { | |
1570 | return current_cpu && qemu_cpu_is_self(current_cpu); | |
1571 | } | |
1572 | ||
1573 | static __thread bool iothread_locked = false; | |
1574 | ||
1575 | bool qemu_mutex_iothread_locked(void) | |
1576 | { | |
1577 | return iothread_locked; | |
1578 | } | |
1579 | ||
1580 | void qemu_mutex_lock_iothread(void) | |
1581 | { | |
1582 | g_assert(!qemu_mutex_iothread_locked()); | |
1583 | qemu_mutex_lock(&qemu_global_mutex); | |
1584 | iothread_locked = true; | |
1585 | } | |
1586 | ||
1587 | void qemu_mutex_unlock_iothread(void) | |
1588 | { | |
1589 | g_assert(qemu_mutex_iothread_locked()); | |
1590 | iothread_locked = false; | |
1591 | qemu_mutex_unlock(&qemu_global_mutex); | |
1592 | } | |
1593 | ||
1594 | static bool all_vcpus_paused(void) | |
1595 | { | |
1596 | CPUState *cpu; | |
1597 | ||
1598 | CPU_FOREACH(cpu) { | |
1599 | if (!cpu->stopped) { | |
1600 | return false; | |
1601 | } | |
1602 | } | |
1603 | ||
1604 | return true; | |
1605 | } | |
1606 | ||
1607 | void pause_all_vcpus(void) | |
1608 | { | |
1609 | CPUState *cpu; | |
1610 | ||
1611 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false); | |
1612 | CPU_FOREACH(cpu) { | |
1613 | cpu->stop = true; | |
1614 | qemu_cpu_kick(cpu); | |
1615 | } | |
1616 | ||
1617 | if (qemu_in_vcpu_thread()) { | |
1618 | cpu_stop_current(); | |
1619 | } | |
1620 | ||
1621 | while (!all_vcpus_paused()) { | |
1622 | qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex); | |
1623 | CPU_FOREACH(cpu) { | |
1624 | qemu_cpu_kick(cpu); | |
1625 | } | |
1626 | } | |
1627 | } | |
1628 | ||
1629 | void cpu_resume(CPUState *cpu) | |
1630 | { | |
1631 | cpu->stop = false; | |
1632 | cpu->stopped = false; | |
1633 | qemu_cpu_kick(cpu); | |
1634 | } | |
1635 | ||
1636 | void resume_all_vcpus(void) | |
1637 | { | |
1638 | CPUState *cpu; | |
1639 | ||
1640 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true); | |
1641 | CPU_FOREACH(cpu) { | |
1642 | cpu_resume(cpu); | |
1643 | } | |
1644 | } | |
1645 | ||
1646 | void cpu_remove(CPUState *cpu) | |
1647 | { | |
1648 | cpu->stop = true; | |
1649 | cpu->unplug = true; | |
1650 | qemu_cpu_kick(cpu); | |
1651 | } | |
1652 | ||
1653 | void cpu_remove_sync(CPUState *cpu) | |
1654 | { | |
1655 | cpu_remove(cpu); | |
1656 | while (cpu->created) { | |
1657 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1658 | } | |
1659 | } | |
1660 | ||
1661 | /* For temporary buffers for forming a name */ | |
1662 | #define VCPU_THREAD_NAME_SIZE 16 | |
1663 | ||
1664 | static void qemu_tcg_init_vcpu(CPUState *cpu) | |
1665 | { | |
1666 | char thread_name[VCPU_THREAD_NAME_SIZE]; | |
1667 | static QemuCond *single_tcg_halt_cond; | |
1668 | static QemuThread *single_tcg_cpu_thread; | |
1669 | static int tcg_region_inited; | |
1670 | ||
1671 | /* | |
1672 | * Initialize TCG regions--once. Now is a good time, because: | |
1673 | * (1) TCG's init context, prologue and target globals have been set up. | |
1674 | * (2) qemu_tcg_mttcg_enabled() works now (TCG init code runs before the | |
1675 | * -accel flag is processed, so the check doesn't work then). | |
1676 | */ | |
1677 | if (!tcg_region_inited) { | |
1678 | tcg_region_inited = 1; | |
1679 | tcg_region_init(); | |
1680 | } | |
1681 | ||
1682 | if (qemu_tcg_mttcg_enabled() || !single_tcg_cpu_thread) { | |
1683 | cpu->thread = g_malloc0(sizeof(QemuThread)); | |
1684 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); | |
1685 | qemu_cond_init(cpu->halt_cond); | |
1686 | ||
1687 | if (qemu_tcg_mttcg_enabled()) { | |
1688 | /* create a thread per vCPU with TCG (MTTCG) */ | |
1689 | parallel_cpus = true; | |
1690 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG", | |
1691 | cpu->cpu_index); | |
1692 | ||
1693 | qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn, | |
1694 | cpu, QEMU_THREAD_JOINABLE); | |
1695 | ||
1696 | } else { | |
1697 | /* share a single thread for all cpus with TCG */ | |
1698 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "ALL CPUs/TCG"); | |
1699 | qemu_thread_create(cpu->thread, thread_name, | |
1700 | qemu_tcg_rr_cpu_thread_fn, | |
1701 | cpu, QEMU_THREAD_JOINABLE); | |
1702 | ||
1703 | single_tcg_halt_cond = cpu->halt_cond; | |
1704 | single_tcg_cpu_thread = cpu->thread; | |
1705 | } | |
1706 | #ifdef _WIN32 | |
1707 | cpu->hThread = qemu_thread_get_handle(cpu->thread); | |
1708 | #endif | |
1709 | while (!cpu->created) { | |
1710 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1711 | } | |
1712 | } else { | |
1713 | /* For non-MTTCG cases we share the thread */ | |
1714 | cpu->thread = single_tcg_cpu_thread; | |
1715 | cpu->halt_cond = single_tcg_halt_cond; | |
1716 | } | |
1717 | } | |
1718 | ||
1719 | static void qemu_hax_start_vcpu(CPUState *cpu) | |
1720 | { | |
1721 | char thread_name[VCPU_THREAD_NAME_SIZE]; | |
1722 | ||
1723 | cpu->thread = g_malloc0(sizeof(QemuThread)); | |
1724 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); | |
1725 | qemu_cond_init(cpu->halt_cond); | |
1726 | ||
1727 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/HAX", | |
1728 | cpu->cpu_index); | |
1729 | qemu_thread_create(cpu->thread, thread_name, qemu_hax_cpu_thread_fn, | |
1730 | cpu, QEMU_THREAD_JOINABLE); | |
1731 | #ifdef _WIN32 | |
1732 | cpu->hThread = qemu_thread_get_handle(cpu->thread); | |
1733 | #endif | |
1734 | while (!cpu->created) { | |
1735 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1736 | } | |
1737 | } | |
1738 | ||
1739 | static void qemu_kvm_start_vcpu(CPUState *cpu) | |
1740 | { | |
1741 | char thread_name[VCPU_THREAD_NAME_SIZE]; | |
1742 | ||
1743 | cpu->thread = g_malloc0(sizeof(QemuThread)); | |
1744 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); | |
1745 | qemu_cond_init(cpu->halt_cond); | |
1746 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/KVM", | |
1747 | cpu->cpu_index); | |
1748 | qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn, | |
1749 | cpu, QEMU_THREAD_JOINABLE); | |
1750 | while (!cpu->created) { | |
1751 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1752 | } | |
1753 | } | |
1754 | ||
1755 | static void qemu_dummy_start_vcpu(CPUState *cpu) | |
1756 | { | |
1757 | char thread_name[VCPU_THREAD_NAME_SIZE]; | |
1758 | ||
1759 | cpu->thread = g_malloc0(sizeof(QemuThread)); | |
1760 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); | |
1761 | qemu_cond_init(cpu->halt_cond); | |
1762 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/DUMMY", | |
1763 | cpu->cpu_index); | |
1764 | qemu_thread_create(cpu->thread, thread_name, qemu_dummy_cpu_thread_fn, cpu, | |
1765 | QEMU_THREAD_JOINABLE); | |
1766 | while (!cpu->created) { | |
1767 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1768 | } | |
1769 | } | |
1770 | ||
1771 | void qemu_init_vcpu(CPUState *cpu) | |
1772 | { | |
1773 | cpu->nr_cores = smp_cores; | |
1774 | cpu->nr_threads = smp_threads; | |
1775 | cpu->stopped = true; | |
1776 | ||
1777 | if (!cpu->as) { | |
1778 | /* If the target cpu hasn't set up any address spaces itself, | |
1779 | * give it the default one. | |
1780 | */ | |
1781 | AddressSpace *as = g_new0(AddressSpace, 1); | |
1782 | ||
1783 | address_space_init(as, cpu->memory, "cpu-memory"); | |
1784 | cpu->num_ases = 1; | |
1785 | cpu_address_space_init(cpu, as, 0); | |
1786 | } | |
1787 | ||
1788 | if (kvm_enabled()) { | |
1789 | qemu_kvm_start_vcpu(cpu); | |
1790 | } else if (hax_enabled()) { | |
1791 | qemu_hax_start_vcpu(cpu); | |
1792 | } else if (tcg_enabled()) { | |
1793 | qemu_tcg_init_vcpu(cpu); | |
1794 | } else { | |
1795 | qemu_dummy_start_vcpu(cpu); | |
1796 | } | |
1797 | } | |
1798 | ||
1799 | void cpu_stop_current(void) | |
1800 | { | |
1801 | if (current_cpu) { | |
1802 | current_cpu->stop = false; | |
1803 | current_cpu->stopped = true; | |
1804 | cpu_exit(current_cpu); | |
1805 | qemu_cond_broadcast(&qemu_pause_cond); | |
1806 | } | |
1807 | } | |
1808 | ||
1809 | int vm_stop(RunState state) | |
1810 | { | |
1811 | if (qemu_in_vcpu_thread()) { | |
1812 | qemu_system_vmstop_request_prepare(); | |
1813 | qemu_system_vmstop_request(state); | |
1814 | /* | |
1815 | * FIXME: should not return to device code in case | |
1816 | * vm_stop() has been requested. | |
1817 | */ | |
1818 | cpu_stop_current(); | |
1819 | return 0; | |
1820 | } | |
1821 | ||
1822 | return do_vm_stop(state); | |
1823 | } | |
1824 | ||
1825 | /** | |
1826 | * Prepare for (re)starting the VM. | |
1827 | * Returns -1 if the vCPUs are not to be restarted (e.g. if they are already | |
1828 | * running or in case of an error condition), 0 otherwise. | |
1829 | */ | |
1830 | int vm_prepare_start(void) | |
1831 | { | |
1832 | RunState requested; | |
1833 | int res = 0; | |
1834 | ||
1835 | qemu_vmstop_requested(&requested); | |
1836 | if (runstate_is_running() && requested == RUN_STATE__MAX) { | |
1837 | return -1; | |
1838 | } | |
1839 | ||
1840 | /* Ensure that a STOP/RESUME pair of events is emitted if a | |
1841 | * vmstop request was pending. The BLOCK_IO_ERROR event, for | |
1842 | * example, according to documentation is always followed by | |
1843 | * the STOP event. | |
1844 | */ | |
1845 | if (runstate_is_running()) { | |
1846 | qapi_event_send_stop(&error_abort); | |
1847 | res = -1; | |
1848 | } else { | |
1849 | replay_enable_events(); | |
1850 | cpu_enable_ticks(); | |
1851 | runstate_set(RUN_STATE_RUNNING); | |
1852 | vm_state_notify(1, RUN_STATE_RUNNING); | |
1853 | } | |
1854 | ||
1855 | /* We are sending this now, but the CPUs will be resumed shortly later */ | |
1856 | qapi_event_send_resume(&error_abort); | |
1857 | return res; | |
1858 | } | |
1859 | ||
1860 | void vm_start(void) | |
1861 | { | |
1862 | if (!vm_prepare_start()) { | |
1863 | resume_all_vcpus(); | |
1864 | } | |
1865 | } | |
1866 | ||
1867 | /* does a state transition even if the VM is already stopped, | |
1868 | current state is forgotten forever */ | |
1869 | int vm_stop_force_state(RunState state) | |
1870 | { | |
1871 | if (runstate_is_running()) { | |
1872 | return vm_stop(state); | |
1873 | } else { | |
1874 | runstate_set(state); | |
1875 | ||
1876 | bdrv_drain_all(); | |
1877 | /* Make sure to return an error if the flush in a previous vm_stop() | |
1878 | * failed. */ | |
1879 | return bdrv_flush_all(); | |
1880 | } | |
1881 | } | |
1882 | ||
1883 | void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg) | |
1884 | { | |
1885 | /* XXX: implement xxx_cpu_list for targets that still miss it */ | |
1886 | #if defined(cpu_list) | |
1887 | cpu_list(f, cpu_fprintf); | |
1888 | #endif | |
1889 | } | |
1890 | ||
1891 | CpuInfoList *qmp_query_cpus(Error **errp) | |
1892 | { | |
1893 | MachineState *ms = MACHINE(qdev_get_machine()); | |
1894 | MachineClass *mc = MACHINE_GET_CLASS(ms); | |
1895 | CpuInfoList *head = NULL, *cur_item = NULL; | |
1896 | CPUState *cpu; | |
1897 | ||
1898 | CPU_FOREACH(cpu) { | |
1899 | CpuInfoList *info; | |
1900 | #if defined(TARGET_I386) | |
1901 | X86CPU *x86_cpu = X86_CPU(cpu); | |
1902 | CPUX86State *env = &x86_cpu->env; | |
1903 | #elif defined(TARGET_PPC) | |
1904 | PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu); | |
1905 | CPUPPCState *env = &ppc_cpu->env; | |
1906 | #elif defined(TARGET_SPARC) | |
1907 | SPARCCPU *sparc_cpu = SPARC_CPU(cpu); | |
1908 | CPUSPARCState *env = &sparc_cpu->env; | |
1909 | #elif defined(TARGET_MIPS) | |
1910 | MIPSCPU *mips_cpu = MIPS_CPU(cpu); | |
1911 | CPUMIPSState *env = &mips_cpu->env; | |
1912 | #elif defined(TARGET_TRICORE) | |
1913 | TriCoreCPU *tricore_cpu = TRICORE_CPU(cpu); | |
1914 | CPUTriCoreState *env = &tricore_cpu->env; | |
1915 | #endif | |
1916 | ||
1917 | cpu_synchronize_state(cpu); | |
1918 | ||
1919 | info = g_malloc0(sizeof(*info)); | |
1920 | info->value = g_malloc0(sizeof(*info->value)); | |
1921 | info->value->CPU = cpu->cpu_index; | |
1922 | info->value->current = (cpu == first_cpu); | |
1923 | info->value->halted = cpu->halted; | |
1924 | info->value->qom_path = object_get_canonical_path(OBJECT(cpu)); | |
1925 | info->value->thread_id = cpu->thread_id; | |
1926 | #if defined(TARGET_I386) | |
1927 | info->value->arch = CPU_INFO_ARCH_X86; | |
1928 | info->value->u.x86.pc = env->eip + env->segs[R_CS].base; | |
1929 | #elif defined(TARGET_PPC) | |
1930 | info->value->arch = CPU_INFO_ARCH_PPC; | |
1931 | info->value->u.ppc.nip = env->nip; | |
1932 | #elif defined(TARGET_SPARC) | |
1933 | info->value->arch = CPU_INFO_ARCH_SPARC; | |
1934 | info->value->u.q_sparc.pc = env->pc; | |
1935 | info->value->u.q_sparc.npc = env->npc; | |
1936 | #elif defined(TARGET_MIPS) | |
1937 | info->value->arch = CPU_INFO_ARCH_MIPS; | |
1938 | info->value->u.q_mips.PC = env->active_tc.PC; | |
1939 | #elif defined(TARGET_TRICORE) | |
1940 | info->value->arch = CPU_INFO_ARCH_TRICORE; | |
1941 | info->value->u.tricore.PC = env->PC; | |
1942 | #else | |
1943 | info->value->arch = CPU_INFO_ARCH_OTHER; | |
1944 | #endif | |
1945 | info->value->has_props = !!mc->cpu_index_to_instance_props; | |
1946 | if (info->value->has_props) { | |
1947 | CpuInstanceProperties *props; | |
1948 | props = g_malloc0(sizeof(*props)); | |
1949 | *props = mc->cpu_index_to_instance_props(ms, cpu->cpu_index); | |
1950 | info->value->props = props; | |
1951 | } | |
1952 | ||
1953 | /* XXX: waiting for the qapi to support GSList */ | |
1954 | if (!cur_item) { | |
1955 | head = cur_item = info; | |
1956 | } else { | |
1957 | cur_item->next = info; | |
1958 | cur_item = info; | |
1959 | } | |
1960 | } | |
1961 | ||
1962 | return head; | |
1963 | } | |
1964 | ||
1965 | void qmp_memsave(int64_t addr, int64_t size, const char *filename, | |
1966 | bool has_cpu, int64_t cpu_index, Error **errp) | |
1967 | { | |
1968 | FILE *f; | |
1969 | uint32_t l; | |
1970 | CPUState *cpu; | |
1971 | uint8_t buf[1024]; | |
1972 | int64_t orig_addr = addr, orig_size = size; | |
1973 | ||
1974 | if (!has_cpu) { | |
1975 | cpu_index = 0; | |
1976 | } | |
1977 | ||
1978 | cpu = qemu_get_cpu(cpu_index); | |
1979 | if (cpu == NULL) { | |
1980 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index", | |
1981 | "a CPU number"); | |
1982 | return; | |
1983 | } | |
1984 | ||
1985 | f = fopen(filename, "wb"); | |
1986 | if (!f) { | |
1987 | error_setg_file_open(errp, errno, filename); | |
1988 | return; | |
1989 | } | |
1990 | ||
1991 | while (size != 0) { | |
1992 | l = sizeof(buf); | |
1993 | if (l > size) | |
1994 | l = size; | |
1995 | if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) { | |
1996 | error_setg(errp, "Invalid addr 0x%016" PRIx64 "/size %" PRId64 | |
1997 | " specified", orig_addr, orig_size); | |
1998 | goto exit; | |
1999 | } | |
2000 | if (fwrite(buf, 1, l, f) != l) { | |
2001 | error_setg(errp, QERR_IO_ERROR); | |
2002 | goto exit; | |
2003 | } | |
2004 | addr += l; | |
2005 | size -= l; | |
2006 | } | |
2007 | ||
2008 | exit: | |
2009 | fclose(f); | |
2010 | } | |
2011 | ||
2012 | void qmp_pmemsave(int64_t addr, int64_t size, const char *filename, | |
2013 | Error **errp) | |
2014 | { | |
2015 | FILE *f; | |
2016 | uint32_t l; | |
2017 | uint8_t buf[1024]; | |
2018 | ||
2019 | f = fopen(filename, "wb"); | |
2020 | if (!f) { | |
2021 | error_setg_file_open(errp, errno, filename); | |
2022 | return; | |
2023 | } | |
2024 | ||
2025 | while (size != 0) { | |
2026 | l = sizeof(buf); | |
2027 | if (l > size) | |
2028 | l = size; | |
2029 | cpu_physical_memory_read(addr, buf, l); | |
2030 | if (fwrite(buf, 1, l, f) != l) { | |
2031 | error_setg(errp, QERR_IO_ERROR); | |
2032 | goto exit; | |
2033 | } | |
2034 | addr += l; | |
2035 | size -= l; | |
2036 | } | |
2037 | ||
2038 | exit: | |
2039 | fclose(f); | |
2040 | } | |
2041 | ||
2042 | void qmp_inject_nmi(Error **errp) | |
2043 | { | |
2044 | nmi_monitor_handle(monitor_get_cpu_index(), errp); | |
2045 | } | |
2046 | ||
2047 | void dump_drift_info(FILE *f, fprintf_function cpu_fprintf) | |
2048 | { | |
2049 | if (!use_icount) { | |
2050 | return; | |
2051 | } | |
2052 | ||
2053 | cpu_fprintf(f, "Host - Guest clock %"PRIi64" ms\n", | |
2054 | (cpu_get_clock() - cpu_get_icount())/SCALE_MS); | |
2055 | if (icount_align_option) { | |
2056 | cpu_fprintf(f, "Max guest delay %"PRIi64" ms\n", -max_delay/SCALE_MS); | |
2057 | cpu_fprintf(f, "Max guest advance %"PRIi64" ms\n", max_advance/SCALE_MS); | |
2058 | } else { | |
2059 | cpu_fprintf(f, "Max guest delay NA\n"); | |
2060 | cpu_fprintf(f, "Max guest advance NA\n"); | |
2061 | } | |
2062 | } |