]>
Commit | Line | Data |
---|---|---|
296af7c9 BS |
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. */ | |
7b31bbc2 | 26 | #include "qemu/osdep.h" |
33c11879 PB |
27 | #include "qemu-common.h" |
28 | #include "cpu.h" | |
83c9089e | 29 | #include "monitor/monitor.h" |
a4e15de9 | 30 | #include "qapi/qmp/qerror.h" |
d49b6836 | 31 | #include "qemu/error-report.h" |
9c17d615 | 32 | #include "sysemu/sysemu.h" |
da31d594 | 33 | #include "sysemu/block-backend.h" |
022c62cb | 34 | #include "exec/gdbstub.h" |
9c17d615 PB |
35 | #include "sysemu/dma.h" |
36 | #include "sysemu/kvm.h" | |
de0b36b6 | 37 | #include "qmp-commands.h" |
63c91552 | 38 | #include "exec/exec-all.h" |
296af7c9 | 39 | |
1de7afc9 | 40 | #include "qemu/thread.h" |
9c17d615 PB |
41 | #include "sysemu/cpus.h" |
42 | #include "sysemu/qtest.h" | |
1de7afc9 PB |
43 | #include "qemu/main-loop.h" |
44 | #include "qemu/bitmap.h" | |
cb365646 | 45 | #include "qemu/seqlock.h" |
a4e15de9 | 46 | #include "qapi-event.h" |
9cb805fd | 47 | #include "hw/nmi.h" |
8b427044 | 48 | #include "sysemu/replay.h" |
0ff0fc19 JK |
49 | |
50 | #ifndef _WIN32 | |
1de7afc9 | 51 | #include "qemu/compatfd.h" |
0ff0fc19 | 52 | #endif |
296af7c9 | 53 | |
6d9cb73c JK |
54 | #ifdef CONFIG_LINUX |
55 | ||
56 | #include <sys/prctl.h> | |
57 | ||
c0532a76 MT |
58 | #ifndef PR_MCE_KILL |
59 | #define PR_MCE_KILL 33 | |
60 | #endif | |
61 | ||
6d9cb73c JK |
62 | #ifndef PR_MCE_KILL_SET |
63 | #define PR_MCE_KILL_SET 1 | |
64 | #endif | |
65 | ||
66 | #ifndef PR_MCE_KILL_EARLY | |
67 | #define PR_MCE_KILL_EARLY 1 | |
68 | #endif | |
69 | ||
70 | #endif /* CONFIG_LINUX */ | |
71 | ||
182735ef | 72 | static CPUState *next_cpu; |
27498bef ST |
73 | int64_t max_delay; |
74 | int64_t max_advance; | |
296af7c9 | 75 | |
2adcc85d JH |
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 | ||
321bc0b2 TC |
84 | bool cpu_is_stopped(CPUState *cpu) |
85 | { | |
86 | return cpu->stopped || !runstate_is_running(); | |
87 | } | |
88 | ||
a98ae1d8 | 89 | static bool cpu_thread_is_idle(CPUState *cpu) |
ac873f1e | 90 | { |
c64ca814 | 91 | if (cpu->stop || cpu->queued_work_first) { |
ac873f1e PM |
92 | return false; |
93 | } | |
321bc0b2 | 94 | if (cpu_is_stopped(cpu)) { |
ac873f1e PM |
95 | return true; |
96 | } | |
8c2e1b00 | 97 | if (!cpu->halted || cpu_has_work(cpu) || |
215e79c0 | 98 | kvm_halt_in_kernel()) { |
ac873f1e PM |
99 | return false; |
100 | } | |
101 | return true; | |
102 | } | |
103 | ||
104 | static bool all_cpu_threads_idle(void) | |
105 | { | |
182735ef | 106 | CPUState *cpu; |
ac873f1e | 107 | |
bdc44640 | 108 | CPU_FOREACH(cpu) { |
182735ef | 109 | if (!cpu_thread_is_idle(cpu)) { |
ac873f1e PM |
110 | return false; |
111 | } | |
112 | } | |
113 | return true; | |
114 | } | |
115 | ||
946fb27c PB |
116 | /***********************************************************/ |
117 | /* guest cycle counter */ | |
118 | ||
a3270e19 PB |
119 | /* Protected by TimersState seqlock */ |
120 | ||
5045e9d9 | 121 | static bool icount_sleep = true; |
71468395 | 122 | static int64_t vm_clock_warp_start = -1; |
946fb27c PB |
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 | |
a3270e19 | 127 | |
946fb27c PB |
128 | static QEMUTimer *icount_rt_timer; |
129 | static QEMUTimer *icount_vm_timer; | |
130 | static QEMUTimer *icount_warp_timer; | |
946fb27c PB |
131 | |
132 | typedef struct TimersState { | |
cb365646 | 133 | /* Protected by BQL. */ |
946fb27c PB |
134 | int64_t cpu_ticks_prev; |
135 | int64_t cpu_ticks_offset; | |
cb365646 LPF |
136 | |
137 | /* cpu_clock_offset can be read out of BQL, so protect it with | |
138 | * this lock. | |
139 | */ | |
140 | QemuSeqLock vm_clock_seqlock; | |
946fb27c PB |
141 | int64_t cpu_clock_offset; |
142 | int32_t cpu_ticks_enabled; | |
143 | int64_t dummy; | |
c96778bb FK |
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; | |
946fb27c PB |
149 | } TimersState; |
150 | ||
d9cd4007 | 151 | static TimersState timers_state; |
946fb27c | 152 | |
2a62914b | 153 | int64_t cpu_get_icount_raw(void) |
946fb27c PB |
154 | { |
155 | int64_t icount; | |
4917cf44 | 156 | CPUState *cpu = current_cpu; |
946fb27c | 157 | |
c96778bb | 158 | icount = timers_state.qemu_icount; |
4917cf44 | 159 | if (cpu) { |
414b15c9 | 160 | if (!cpu->can_do_io) { |
2a62914b PD |
161 | fprintf(stderr, "Bad icount read\n"); |
162 | exit(1); | |
946fb27c | 163 | } |
28ecfd7a | 164 | icount -= (cpu->icount_decr.u16.low + cpu->icount_extra); |
946fb27c | 165 | } |
2a62914b PD |
166 | return icount; |
167 | } | |
168 | ||
169 | /* Return the virtual CPU time, based on the instruction counter. */ | |
170 | static int64_t cpu_get_icount_locked(void) | |
171 | { | |
172 | int64_t icount = cpu_get_icount_raw(); | |
3f031313 | 173 | return timers_state.qemu_icount_bias + cpu_icount_to_ns(icount); |
946fb27c PB |
174 | } |
175 | ||
17a15f1b PB |
176 | int64_t cpu_get_icount(void) |
177 | { | |
178 | int64_t icount; | |
179 | unsigned start; | |
180 | ||
181 | do { | |
182 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
183 | icount = cpu_get_icount_locked(); | |
184 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
185 | ||
186 | return icount; | |
187 | } | |
188 | ||
3f031313 FK |
189 | int64_t cpu_icount_to_ns(int64_t icount) |
190 | { | |
191 | return icount << icount_time_shift; | |
192 | } | |
193 | ||
d90f3cca C |
194 | /* return the time elapsed in VM between vm_start and vm_stop. Unless |
195 | * icount is active, cpu_get_ticks() uses units of the host CPU cycle | |
196 | * counter. | |
197 | * | |
198 | * Caller must hold the BQL | |
199 | */ | |
946fb27c PB |
200 | int64_t cpu_get_ticks(void) |
201 | { | |
5f3e3101 PB |
202 | int64_t ticks; |
203 | ||
946fb27c PB |
204 | if (use_icount) { |
205 | return cpu_get_icount(); | |
206 | } | |
5f3e3101 PB |
207 | |
208 | ticks = timers_state.cpu_ticks_offset; | |
209 | if (timers_state.cpu_ticks_enabled) { | |
4a7428c5 | 210 | ticks += cpu_get_host_ticks(); |
5f3e3101 PB |
211 | } |
212 | ||
213 | if (timers_state.cpu_ticks_prev > ticks) { | |
214 | /* Note: non increasing ticks may happen if the host uses | |
215 | software suspend */ | |
216 | timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks; | |
217 | ticks = timers_state.cpu_ticks_prev; | |
946fb27c | 218 | } |
5f3e3101 PB |
219 | |
220 | timers_state.cpu_ticks_prev = ticks; | |
221 | return ticks; | |
946fb27c PB |
222 | } |
223 | ||
cb365646 | 224 | static int64_t cpu_get_clock_locked(void) |
946fb27c | 225 | { |
1d45cea5 | 226 | int64_t time; |
cb365646 | 227 | |
1d45cea5 | 228 | time = timers_state.cpu_clock_offset; |
5f3e3101 | 229 | if (timers_state.cpu_ticks_enabled) { |
1d45cea5 | 230 | time += get_clock(); |
946fb27c | 231 | } |
cb365646 | 232 | |
1d45cea5 | 233 | return time; |
cb365646 LPF |
234 | } |
235 | ||
d90f3cca | 236 | /* Return the monotonic time elapsed in VM, i.e., |
8212ff86 PM |
237 | * the time between vm_start and vm_stop |
238 | */ | |
cb365646 LPF |
239 | int64_t cpu_get_clock(void) |
240 | { | |
241 | int64_t ti; | |
242 | unsigned start; | |
243 | ||
244 | do { | |
245 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
246 | ti = cpu_get_clock_locked(); | |
247 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
248 | ||
249 | return ti; | |
946fb27c PB |
250 | } |
251 | ||
cb365646 | 252 | /* enable cpu_get_ticks() |
3224e878 | 253 | * Caller must hold BQL which serves as mutex for vm_clock_seqlock. |
cb365646 | 254 | */ |
946fb27c PB |
255 | void cpu_enable_ticks(void) |
256 | { | |
cb365646 | 257 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ |
03719e44 | 258 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
946fb27c | 259 | if (!timers_state.cpu_ticks_enabled) { |
4a7428c5 | 260 | timers_state.cpu_ticks_offset -= cpu_get_host_ticks(); |
946fb27c PB |
261 | timers_state.cpu_clock_offset -= get_clock(); |
262 | timers_state.cpu_ticks_enabled = 1; | |
263 | } | |
03719e44 | 264 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
946fb27c PB |
265 | } |
266 | ||
267 | /* disable cpu_get_ticks() : the clock is stopped. You must not call | |
cb365646 | 268 | * cpu_get_ticks() after that. |
3224e878 | 269 | * Caller must hold BQL which serves as mutex for vm_clock_seqlock. |
cb365646 | 270 | */ |
946fb27c PB |
271 | void cpu_disable_ticks(void) |
272 | { | |
cb365646 | 273 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ |
03719e44 | 274 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
946fb27c | 275 | if (timers_state.cpu_ticks_enabled) { |
4a7428c5 | 276 | timers_state.cpu_ticks_offset += cpu_get_host_ticks(); |
cb365646 | 277 | timers_state.cpu_clock_offset = cpu_get_clock_locked(); |
946fb27c PB |
278 | timers_state.cpu_ticks_enabled = 0; |
279 | } | |
03719e44 | 280 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
946fb27c PB |
281 | } |
282 | ||
283 | /* Correlation between real and virtual time is always going to be | |
284 | fairly approximate, so ignore small variation. | |
285 | When the guest is idle real and virtual time will be aligned in | |
286 | the IO wait loop. */ | |
73bcb24d | 287 | #define ICOUNT_WOBBLE (NANOSECONDS_PER_SECOND / 10) |
946fb27c PB |
288 | |
289 | static void icount_adjust(void) | |
290 | { | |
291 | int64_t cur_time; | |
292 | int64_t cur_icount; | |
293 | int64_t delta; | |
a3270e19 PB |
294 | |
295 | /* Protected by TimersState mutex. */ | |
946fb27c | 296 | static int64_t last_delta; |
468cc7cf | 297 | |
946fb27c PB |
298 | /* If the VM is not running, then do nothing. */ |
299 | if (!runstate_is_running()) { | |
300 | return; | |
301 | } | |
468cc7cf | 302 | |
03719e44 | 303 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
17a15f1b PB |
304 | cur_time = cpu_get_clock_locked(); |
305 | cur_icount = cpu_get_icount_locked(); | |
468cc7cf | 306 | |
946fb27c PB |
307 | delta = cur_icount - cur_time; |
308 | /* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */ | |
309 | if (delta > 0 | |
310 | && last_delta + ICOUNT_WOBBLE < delta * 2 | |
311 | && icount_time_shift > 0) { | |
312 | /* The guest is getting too far ahead. Slow time down. */ | |
313 | icount_time_shift--; | |
314 | } | |
315 | if (delta < 0 | |
316 | && last_delta - ICOUNT_WOBBLE > delta * 2 | |
317 | && icount_time_shift < MAX_ICOUNT_SHIFT) { | |
318 | /* The guest is getting too far behind. Speed time up. */ | |
319 | icount_time_shift++; | |
320 | } | |
321 | last_delta = delta; | |
c96778bb FK |
322 | timers_state.qemu_icount_bias = cur_icount |
323 | - (timers_state.qemu_icount << icount_time_shift); | |
03719e44 | 324 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
946fb27c PB |
325 | } |
326 | ||
327 | static void icount_adjust_rt(void *opaque) | |
328 | { | |
40daca54 | 329 | timer_mod(icount_rt_timer, |
1979b908 | 330 | qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000); |
946fb27c PB |
331 | icount_adjust(); |
332 | } | |
333 | ||
334 | static void icount_adjust_vm(void *opaque) | |
335 | { | |
40daca54 AB |
336 | timer_mod(icount_vm_timer, |
337 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
73bcb24d | 338 | NANOSECONDS_PER_SECOND / 10); |
946fb27c PB |
339 | icount_adjust(); |
340 | } | |
341 | ||
342 | static int64_t qemu_icount_round(int64_t count) | |
343 | { | |
344 | return (count + (1 << icount_time_shift) - 1) >> icount_time_shift; | |
345 | } | |
346 | ||
efab87cf | 347 | static void icount_warp_rt(void) |
946fb27c | 348 | { |
ccffff48 AB |
349 | unsigned seq; |
350 | int64_t warp_start; | |
351 | ||
17a15f1b PB |
352 | /* The icount_warp_timer is rescheduled soon after vm_clock_warp_start |
353 | * changes from -1 to another value, so the race here is okay. | |
354 | */ | |
ccffff48 AB |
355 | do { |
356 | seq = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
357 | warp_start = vm_clock_warp_start; | |
358 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, seq)); | |
359 | ||
360 | if (warp_start == -1) { | |
946fb27c PB |
361 | return; |
362 | } | |
363 | ||
03719e44 | 364 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
946fb27c | 365 | if (runstate_is_running()) { |
8eda206e PD |
366 | int64_t clock = REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT, |
367 | cpu_get_clock_locked()); | |
8ed961d9 PB |
368 | int64_t warp_delta; |
369 | ||
370 | warp_delta = clock - vm_clock_warp_start; | |
371 | if (use_icount == 2) { | |
946fb27c | 372 | /* |
40daca54 | 373 | * In adaptive mode, do not let QEMU_CLOCK_VIRTUAL run too |
946fb27c PB |
374 | * far ahead of real time. |
375 | */ | |
17a15f1b | 376 | int64_t cur_icount = cpu_get_icount_locked(); |
bf2a7ddb | 377 | int64_t delta = clock - cur_icount; |
8ed961d9 | 378 | warp_delta = MIN(warp_delta, delta); |
946fb27c | 379 | } |
c96778bb | 380 | timers_state.qemu_icount_bias += warp_delta; |
946fb27c PB |
381 | } |
382 | vm_clock_warp_start = -1; | |
03719e44 | 383 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
8ed961d9 PB |
384 | |
385 | if (qemu_clock_expired(QEMU_CLOCK_VIRTUAL)) { | |
386 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
387 | } | |
946fb27c PB |
388 | } |
389 | ||
e76d1798 | 390 | static void icount_timer_cb(void *opaque) |
efab87cf | 391 | { |
e76d1798 PD |
392 | /* No need for a checkpoint because the timer already synchronizes |
393 | * with CHECKPOINT_CLOCK_VIRTUAL_RT. | |
394 | */ | |
395 | icount_warp_rt(); | |
efab87cf PD |
396 | } |
397 | ||
8156be56 PB |
398 | void qtest_clock_warp(int64_t dest) |
399 | { | |
40daca54 | 400 | int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); |
efef88b3 | 401 | AioContext *aio_context; |
8156be56 | 402 | assert(qtest_enabled()); |
efef88b3 | 403 | aio_context = qemu_get_aio_context(); |
8156be56 | 404 | while (clock < dest) { |
40daca54 | 405 | int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); |
c9299e2f | 406 | int64_t warp = qemu_soonest_timeout(dest - clock, deadline); |
efef88b3 | 407 | |
03719e44 | 408 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
c96778bb | 409 | timers_state.qemu_icount_bias += warp; |
03719e44 | 410 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
17a15f1b | 411 | |
40daca54 | 412 | qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL); |
efef88b3 | 413 | timerlist_run_timers(aio_context->tlg.tl[QEMU_CLOCK_VIRTUAL]); |
40daca54 | 414 | clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); |
8156be56 | 415 | } |
40daca54 | 416 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
8156be56 PB |
417 | } |
418 | ||
e76d1798 | 419 | void qemu_start_warp_timer(void) |
946fb27c | 420 | { |
ce78d18c | 421 | int64_t clock; |
946fb27c PB |
422 | int64_t deadline; |
423 | ||
e76d1798 | 424 | if (!use_icount) { |
946fb27c PB |
425 | return; |
426 | } | |
427 | ||
8bd7f71d PD |
428 | /* Nothing to do if the VM is stopped: QEMU_CLOCK_VIRTUAL timers |
429 | * do not fire, so computing the deadline does not make sense. | |
430 | */ | |
431 | if (!runstate_is_running()) { | |
432 | return; | |
433 | } | |
434 | ||
435 | /* warp clock deterministically in record/replay mode */ | |
e76d1798 | 436 | if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_START)) { |
8bd7f71d PD |
437 | return; |
438 | } | |
439 | ||
ce78d18c | 440 | if (!all_cpu_threads_idle()) { |
946fb27c PB |
441 | return; |
442 | } | |
443 | ||
8156be56 PB |
444 | if (qtest_enabled()) { |
445 | /* When testing, qtest commands advance icount. */ | |
e76d1798 | 446 | return; |
8156be56 PB |
447 | } |
448 | ||
ac70aafc | 449 | /* We want to use the earliest deadline from ALL vm_clocks */ |
bf2a7ddb | 450 | clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT); |
40daca54 | 451 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); |
ce78d18c | 452 | if (deadline < 0) { |
d7a0f71d VC |
453 | static bool notified; |
454 | if (!icount_sleep && !notified) { | |
455 | error_report("WARNING: icount sleep disabled and no active timers"); | |
456 | notified = true; | |
457 | } | |
ce78d18c | 458 | return; |
ac70aafc AB |
459 | } |
460 | ||
946fb27c PB |
461 | if (deadline > 0) { |
462 | /* | |
40daca54 | 463 | * Ensure QEMU_CLOCK_VIRTUAL proceeds even when the virtual CPU goes to |
946fb27c PB |
464 | * sleep. Otherwise, the CPU might be waiting for a future timer |
465 | * interrupt to wake it up, but the interrupt never comes because | |
466 | * the vCPU isn't running any insns and thus doesn't advance the | |
40daca54 | 467 | * QEMU_CLOCK_VIRTUAL. |
946fb27c | 468 | */ |
5045e9d9 VC |
469 | if (!icount_sleep) { |
470 | /* | |
471 | * We never let VCPUs sleep in no sleep icount mode. | |
472 | * If there is a pending QEMU_CLOCK_VIRTUAL timer we just advance | |
473 | * to the next QEMU_CLOCK_VIRTUAL event and notify it. | |
474 | * It is useful when we want a deterministic execution time, | |
475 | * isolated from host latencies. | |
476 | */ | |
03719e44 | 477 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
5045e9d9 | 478 | timers_state.qemu_icount_bias += deadline; |
03719e44 | 479 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
5045e9d9 VC |
480 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
481 | } else { | |
482 | /* | |
483 | * We do stop VCPUs and only advance QEMU_CLOCK_VIRTUAL after some | |
484 | * "real" time, (related to the time left until the next event) has | |
485 | * passed. The QEMU_CLOCK_VIRTUAL_RT clock will do this. | |
486 | * This avoids that the warps are visible externally; for example, | |
487 | * you will not be sending network packets continuously instead of | |
488 | * every 100ms. | |
489 | */ | |
03719e44 | 490 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
5045e9d9 VC |
491 | if (vm_clock_warp_start == -1 || vm_clock_warp_start > clock) { |
492 | vm_clock_warp_start = clock; | |
493 | } | |
03719e44 | 494 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
5045e9d9 | 495 | timer_mod_anticipate(icount_warp_timer, clock + deadline); |
ce78d18c | 496 | } |
ac70aafc | 497 | } else if (deadline == 0) { |
40daca54 | 498 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
946fb27c PB |
499 | } |
500 | } | |
501 | ||
e76d1798 PD |
502 | static void qemu_account_warp_timer(void) |
503 | { | |
504 | if (!use_icount || !icount_sleep) { | |
505 | return; | |
506 | } | |
507 | ||
508 | /* Nothing to do if the VM is stopped: QEMU_CLOCK_VIRTUAL timers | |
509 | * do not fire, so computing the deadline does not make sense. | |
510 | */ | |
511 | if (!runstate_is_running()) { | |
512 | return; | |
513 | } | |
514 | ||
515 | /* warp clock deterministically in record/replay mode */ | |
516 | if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_ACCOUNT)) { | |
517 | return; | |
518 | } | |
519 | ||
520 | timer_del(icount_warp_timer); | |
521 | icount_warp_rt(); | |
522 | } | |
523 | ||
d09eae37 FK |
524 | static bool icount_state_needed(void *opaque) |
525 | { | |
526 | return use_icount; | |
527 | } | |
528 | ||
529 | /* | |
530 | * This is a subsection for icount migration. | |
531 | */ | |
532 | static const VMStateDescription icount_vmstate_timers = { | |
533 | .name = "timer/icount", | |
534 | .version_id = 1, | |
535 | .minimum_version_id = 1, | |
5cd8cada | 536 | .needed = icount_state_needed, |
d09eae37 FK |
537 | .fields = (VMStateField[]) { |
538 | VMSTATE_INT64(qemu_icount_bias, TimersState), | |
539 | VMSTATE_INT64(qemu_icount, TimersState), | |
540 | VMSTATE_END_OF_LIST() | |
541 | } | |
542 | }; | |
543 | ||
946fb27c PB |
544 | static const VMStateDescription vmstate_timers = { |
545 | .name = "timer", | |
546 | .version_id = 2, | |
547 | .minimum_version_id = 1, | |
35d08458 | 548 | .fields = (VMStateField[]) { |
946fb27c PB |
549 | VMSTATE_INT64(cpu_ticks_offset, TimersState), |
550 | VMSTATE_INT64(dummy, TimersState), | |
551 | VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2), | |
552 | VMSTATE_END_OF_LIST() | |
d09eae37 | 553 | }, |
5cd8cada JQ |
554 | .subsections = (const VMStateDescription*[]) { |
555 | &icount_vmstate_timers, | |
556 | NULL | |
946fb27c PB |
557 | } |
558 | }; | |
559 | ||
e0eeb4a2 | 560 | static void cpu_throttle_thread(CPUState *cpu, void *opaque) |
2adcc85d | 561 | { |
2adcc85d JH |
562 | double pct; |
563 | double throttle_ratio; | |
564 | long sleeptime_ns; | |
565 | ||
566 | if (!cpu_throttle_get_percentage()) { | |
567 | return; | |
568 | } | |
569 | ||
570 | pct = (double)cpu_throttle_get_percentage()/100; | |
571 | throttle_ratio = pct / (1 - pct); | |
572 | sleeptime_ns = (long)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS); | |
573 | ||
574 | qemu_mutex_unlock_iothread(); | |
575 | atomic_set(&cpu->throttle_thread_scheduled, 0); | |
576 | g_usleep(sleeptime_ns / 1000); /* Convert ns to us for usleep call */ | |
577 | qemu_mutex_lock_iothread(); | |
578 | } | |
579 | ||
580 | static void cpu_throttle_timer_tick(void *opaque) | |
581 | { | |
582 | CPUState *cpu; | |
583 | double pct; | |
584 | ||
585 | /* Stop the timer if needed */ | |
586 | if (!cpu_throttle_get_percentage()) { | |
587 | return; | |
588 | } | |
589 | CPU_FOREACH(cpu) { | |
590 | if (!atomic_xchg(&cpu->throttle_thread_scheduled, 1)) { | |
e0eeb4a2 | 591 | async_run_on_cpu(cpu, cpu_throttle_thread, NULL); |
2adcc85d JH |
592 | } |
593 | } | |
594 | ||
595 | pct = (double)cpu_throttle_get_percentage()/100; | |
596 | timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) + | |
597 | CPU_THROTTLE_TIMESLICE_NS / (1-pct)); | |
598 | } | |
599 | ||
600 | void cpu_throttle_set(int new_throttle_pct) | |
601 | { | |
602 | /* Ensure throttle percentage is within valid range */ | |
603 | new_throttle_pct = MIN(new_throttle_pct, CPU_THROTTLE_PCT_MAX); | |
604 | new_throttle_pct = MAX(new_throttle_pct, CPU_THROTTLE_PCT_MIN); | |
605 | ||
606 | atomic_set(&throttle_percentage, new_throttle_pct); | |
607 | ||
608 | timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) + | |
609 | CPU_THROTTLE_TIMESLICE_NS); | |
610 | } | |
611 | ||
612 | void cpu_throttle_stop(void) | |
613 | { | |
614 | atomic_set(&throttle_percentage, 0); | |
615 | } | |
616 | ||
617 | bool cpu_throttle_active(void) | |
618 | { | |
619 | return (cpu_throttle_get_percentage() != 0); | |
620 | } | |
621 | ||
622 | int cpu_throttle_get_percentage(void) | |
623 | { | |
624 | return atomic_read(&throttle_percentage); | |
625 | } | |
626 | ||
4603ea01 PD |
627 | void cpu_ticks_init(void) |
628 | { | |
ccdb3c1f | 629 | seqlock_init(&timers_state.vm_clock_seqlock); |
4603ea01 | 630 | vmstate_register(NULL, 0, &vmstate_timers, &timers_state); |
2adcc85d JH |
631 | throttle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT, |
632 | cpu_throttle_timer_tick, NULL); | |
4603ea01 PD |
633 | } |
634 | ||
1ad9580b | 635 | void configure_icount(QemuOpts *opts, Error **errp) |
946fb27c | 636 | { |
1ad9580b | 637 | const char *option; |
a8bfac37 | 638 | char *rem_str = NULL; |
1ad9580b | 639 | |
1ad9580b | 640 | option = qemu_opt_get(opts, "shift"); |
946fb27c | 641 | if (!option) { |
a8bfac37 ST |
642 | if (qemu_opt_get(opts, "align") != NULL) { |
643 | error_setg(errp, "Please specify shift option when using align"); | |
644 | } | |
946fb27c PB |
645 | return; |
646 | } | |
f1f4b57e VC |
647 | |
648 | icount_sleep = qemu_opt_get_bool(opts, "sleep", true); | |
5045e9d9 VC |
649 | if (icount_sleep) { |
650 | icount_warp_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT, | |
e76d1798 | 651 | icount_timer_cb, NULL); |
5045e9d9 | 652 | } |
f1f4b57e | 653 | |
a8bfac37 | 654 | icount_align_option = qemu_opt_get_bool(opts, "align", false); |
f1f4b57e VC |
655 | |
656 | if (icount_align_option && !icount_sleep) { | |
778d9f9b | 657 | error_setg(errp, "align=on and sleep=off are incompatible"); |
f1f4b57e | 658 | } |
946fb27c | 659 | if (strcmp(option, "auto") != 0) { |
a8bfac37 ST |
660 | errno = 0; |
661 | icount_time_shift = strtol(option, &rem_str, 0); | |
662 | if (errno != 0 || *rem_str != '\0' || !strlen(option)) { | |
663 | error_setg(errp, "icount: Invalid shift value"); | |
664 | } | |
946fb27c PB |
665 | use_icount = 1; |
666 | return; | |
a8bfac37 ST |
667 | } else if (icount_align_option) { |
668 | error_setg(errp, "shift=auto and align=on are incompatible"); | |
f1f4b57e | 669 | } else if (!icount_sleep) { |
778d9f9b | 670 | error_setg(errp, "shift=auto and sleep=off are incompatible"); |
946fb27c PB |
671 | } |
672 | ||
673 | use_icount = 2; | |
674 | ||
675 | /* 125MIPS seems a reasonable initial guess at the guest speed. | |
676 | It will be corrected fairly quickly anyway. */ | |
677 | icount_time_shift = 3; | |
678 | ||
679 | /* Have both realtime and virtual time triggers for speed adjustment. | |
680 | The realtime trigger catches emulated time passing too slowly, | |
681 | the virtual time trigger catches emulated time passing too fast. | |
682 | Realtime triggers occur even when idle, so use them less frequently | |
683 | than VM triggers. */ | |
bf2a7ddb PD |
684 | icount_rt_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL_RT, |
685 | icount_adjust_rt, NULL); | |
40daca54 | 686 | timer_mod(icount_rt_timer, |
bf2a7ddb | 687 | qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000); |
40daca54 AB |
688 | icount_vm_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, |
689 | icount_adjust_vm, NULL); | |
690 | timer_mod(icount_vm_timer, | |
691 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
73bcb24d | 692 | NANOSECONDS_PER_SECOND / 10); |
946fb27c PB |
693 | } |
694 | ||
296af7c9 BS |
695 | /***********************************************************/ |
696 | void hw_error(const char *fmt, ...) | |
697 | { | |
698 | va_list ap; | |
55e5c285 | 699 | CPUState *cpu; |
296af7c9 BS |
700 | |
701 | va_start(ap, fmt); | |
702 | fprintf(stderr, "qemu: hardware error: "); | |
703 | vfprintf(stderr, fmt, ap); | |
704 | fprintf(stderr, "\n"); | |
bdc44640 | 705 | CPU_FOREACH(cpu) { |
55e5c285 | 706 | fprintf(stderr, "CPU #%d:\n", cpu->cpu_index); |
878096ee | 707 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU); |
296af7c9 BS |
708 | } |
709 | va_end(ap); | |
710 | abort(); | |
711 | } | |
712 | ||
713 | void cpu_synchronize_all_states(void) | |
714 | { | |
182735ef | 715 | CPUState *cpu; |
296af7c9 | 716 | |
bdc44640 | 717 | CPU_FOREACH(cpu) { |
182735ef | 718 | cpu_synchronize_state(cpu); |
296af7c9 BS |
719 | } |
720 | } | |
721 | ||
722 | void cpu_synchronize_all_post_reset(void) | |
723 | { | |
182735ef | 724 | CPUState *cpu; |
296af7c9 | 725 | |
bdc44640 | 726 | CPU_FOREACH(cpu) { |
182735ef | 727 | cpu_synchronize_post_reset(cpu); |
296af7c9 BS |
728 | } |
729 | } | |
730 | ||
731 | void cpu_synchronize_all_post_init(void) | |
732 | { | |
182735ef | 733 | CPUState *cpu; |
296af7c9 | 734 | |
bdc44640 | 735 | CPU_FOREACH(cpu) { |
182735ef | 736 | cpu_synchronize_post_init(cpu); |
296af7c9 BS |
737 | } |
738 | } | |
739 | ||
56983463 | 740 | static int do_vm_stop(RunState state) |
296af7c9 | 741 | { |
56983463 KW |
742 | int ret = 0; |
743 | ||
1354869c | 744 | if (runstate_is_running()) { |
296af7c9 | 745 | cpu_disable_ticks(); |
296af7c9 | 746 | pause_all_vcpus(); |
f5bbfba1 | 747 | runstate_set(state); |
1dfb4dd9 | 748 | vm_state_notify(0, state); |
a4e15de9 | 749 | qapi_event_send_stop(&error_abort); |
296af7c9 | 750 | } |
56983463 | 751 | |
594a45ce | 752 | bdrv_drain_all(); |
da31d594 | 753 | ret = blk_flush_all(); |
594a45ce | 754 | |
56983463 | 755 | return ret; |
296af7c9 BS |
756 | } |
757 | ||
a1fcaa73 | 758 | static bool cpu_can_run(CPUState *cpu) |
296af7c9 | 759 | { |
4fdeee7c | 760 | if (cpu->stop) { |
a1fcaa73 | 761 | return false; |
0ab07c62 | 762 | } |
321bc0b2 | 763 | if (cpu_is_stopped(cpu)) { |
a1fcaa73 | 764 | return false; |
0ab07c62 | 765 | } |
a1fcaa73 | 766 | return true; |
296af7c9 BS |
767 | } |
768 | ||
91325046 | 769 | static void cpu_handle_guest_debug(CPUState *cpu) |
83f338f7 | 770 | { |
64f6b346 | 771 | gdb_set_stop_cpu(cpu); |
8cf71710 | 772 | qemu_system_debug_request(); |
f324e766 | 773 | cpu->stopped = true; |
3c638d06 JK |
774 | } |
775 | ||
6d9cb73c JK |
776 | #ifdef CONFIG_LINUX |
777 | static void sigbus_reraise(void) | |
778 | { | |
779 | sigset_t set; | |
780 | struct sigaction action; | |
781 | ||
782 | memset(&action, 0, sizeof(action)); | |
783 | action.sa_handler = SIG_DFL; | |
784 | if (!sigaction(SIGBUS, &action, NULL)) { | |
785 | raise(SIGBUS); | |
786 | sigemptyset(&set); | |
787 | sigaddset(&set, SIGBUS); | |
a2d1761d | 788 | pthread_sigmask(SIG_UNBLOCK, &set, NULL); |
6d9cb73c JK |
789 | } |
790 | perror("Failed to re-raise SIGBUS!\n"); | |
791 | abort(); | |
792 | } | |
793 | ||
794 | static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo, | |
795 | void *ctx) | |
796 | { | |
797 | if (kvm_on_sigbus(siginfo->ssi_code, | |
798 | (void *)(intptr_t)siginfo->ssi_addr)) { | |
799 | sigbus_reraise(); | |
800 | } | |
801 | } | |
802 | ||
803 | static void qemu_init_sigbus(void) | |
804 | { | |
805 | struct sigaction action; | |
806 | ||
807 | memset(&action, 0, sizeof(action)); | |
808 | action.sa_flags = SA_SIGINFO; | |
809 | action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler; | |
810 | sigaction(SIGBUS, &action, NULL); | |
811 | ||
812 | prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0); | |
813 | } | |
814 | ||
290adf38 | 815 | static void qemu_kvm_eat_signals(CPUState *cpu) |
1ab3c6c0 JK |
816 | { |
817 | struct timespec ts = { 0, 0 }; | |
818 | siginfo_t siginfo; | |
819 | sigset_t waitset; | |
820 | sigset_t chkset; | |
821 | int r; | |
822 | ||
823 | sigemptyset(&waitset); | |
824 | sigaddset(&waitset, SIG_IPI); | |
825 | sigaddset(&waitset, SIGBUS); | |
826 | ||
827 | do { | |
828 | r = sigtimedwait(&waitset, &siginfo, &ts); | |
829 | if (r == -1 && !(errno == EAGAIN || errno == EINTR)) { | |
830 | perror("sigtimedwait"); | |
831 | exit(1); | |
832 | } | |
833 | ||
834 | switch (r) { | |
835 | case SIGBUS: | |
290adf38 | 836 | if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) { |
1ab3c6c0 JK |
837 | sigbus_reraise(); |
838 | } | |
839 | break; | |
840 | default: | |
841 | break; | |
842 | } | |
843 | ||
844 | r = sigpending(&chkset); | |
845 | if (r == -1) { | |
846 | perror("sigpending"); | |
847 | exit(1); | |
848 | } | |
849 | } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS)); | |
1ab3c6c0 JK |
850 | } |
851 | ||
6d9cb73c JK |
852 | #else /* !CONFIG_LINUX */ |
853 | ||
854 | static void qemu_init_sigbus(void) | |
855 | { | |
856 | } | |
1ab3c6c0 | 857 | |
290adf38 | 858 | static void qemu_kvm_eat_signals(CPUState *cpu) |
1ab3c6c0 JK |
859 | { |
860 | } | |
6d9cb73c JK |
861 | #endif /* !CONFIG_LINUX */ |
862 | ||
296af7c9 | 863 | #ifndef _WIN32 |
55f8d6ac JK |
864 | static void dummy_signal(int sig) |
865 | { | |
866 | } | |
55f8d6ac | 867 | |
13618e05 | 868 | static void qemu_kvm_init_cpu_signals(CPUState *cpu) |
714bd040 PB |
869 | { |
870 | int r; | |
871 | sigset_t set; | |
872 | struct sigaction sigact; | |
873 | ||
874 | memset(&sigact, 0, sizeof(sigact)); | |
875 | sigact.sa_handler = dummy_signal; | |
876 | sigaction(SIG_IPI, &sigact, NULL); | |
877 | ||
714bd040 PB |
878 | pthread_sigmask(SIG_BLOCK, NULL, &set); |
879 | sigdelset(&set, SIG_IPI); | |
714bd040 | 880 | sigdelset(&set, SIGBUS); |
491d6e80 | 881 | r = kvm_set_signal_mask(cpu, &set); |
714bd040 PB |
882 | if (r) { |
883 | fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r)); | |
884 | exit(1); | |
885 | } | |
886 | } | |
887 | ||
55f8d6ac | 888 | #else /* _WIN32 */ |
13618e05 | 889 | static void qemu_kvm_init_cpu_signals(CPUState *cpu) |
ff48eb5f | 890 | { |
714bd040 PB |
891 | abort(); |
892 | } | |
714bd040 | 893 | #endif /* _WIN32 */ |
ff48eb5f | 894 | |
b2532d88 | 895 | static QemuMutex qemu_global_mutex; |
46daff13 | 896 | static QemuCond qemu_io_proceeded_cond; |
6b49809c | 897 | static unsigned iothread_requesting_mutex; |
296af7c9 BS |
898 | |
899 | static QemuThread io_thread; | |
900 | ||
296af7c9 BS |
901 | /* cpu creation */ |
902 | static QemuCond qemu_cpu_cond; | |
903 | /* system init */ | |
296af7c9 | 904 | static QemuCond qemu_pause_cond; |
e82bcec2 | 905 | static QemuCond qemu_work_cond; |
296af7c9 | 906 | |
d3b12f5d | 907 | void qemu_init_cpu_loop(void) |
296af7c9 | 908 | { |
6d9cb73c | 909 | qemu_init_sigbus(); |
ed94592b | 910 | qemu_cond_init(&qemu_cpu_cond); |
ed94592b AL |
911 | qemu_cond_init(&qemu_pause_cond); |
912 | qemu_cond_init(&qemu_work_cond); | |
46daff13 | 913 | qemu_cond_init(&qemu_io_proceeded_cond); |
296af7c9 | 914 | qemu_mutex_init(&qemu_global_mutex); |
296af7c9 | 915 | |
b7680cb6 | 916 | qemu_thread_get_self(&io_thread); |
296af7c9 BS |
917 | } |
918 | ||
fd38b251 SF |
919 | static void queue_work_on_cpu(CPUState *cpu, struct qemu_work_item *wi) |
920 | { | |
921 | qemu_mutex_lock(&cpu->work_mutex); | |
922 | if (cpu->queued_work_first == NULL) { | |
923 | cpu->queued_work_first = wi; | |
924 | } else { | |
925 | cpu->queued_work_last->next = wi; | |
926 | } | |
927 | cpu->queued_work_last = wi; | |
928 | wi->next = NULL; | |
929 | wi->done = false; | |
930 | qemu_mutex_unlock(&cpu->work_mutex); | |
931 | ||
932 | qemu_cpu_kick(cpu); | |
933 | } | |
934 | ||
e0eeb4a2 | 935 | void run_on_cpu(CPUState *cpu, run_on_cpu_func func, void *data) |
e82bcec2 MT |
936 | { |
937 | struct qemu_work_item wi; | |
938 | ||
60e82579 | 939 | if (qemu_cpu_is_self(cpu)) { |
e0eeb4a2 | 940 | func(cpu, data); |
e82bcec2 MT |
941 | return; |
942 | } | |
943 | ||
944 | wi.func = func; | |
945 | wi.data = data; | |
3c02270d | 946 | wi.free = false; |
376692b9 | 947 | |
fd38b251 | 948 | queue_work_on_cpu(cpu, &wi); |
376692b9 | 949 | while (!atomic_mb_read(&wi.done)) { |
4917cf44 | 950 | CPUState *self_cpu = current_cpu; |
e82bcec2 MT |
951 | |
952 | qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex); | |
4917cf44 | 953 | current_cpu = self_cpu; |
e82bcec2 MT |
954 | } |
955 | } | |
956 | ||
e0eeb4a2 | 957 | void async_run_on_cpu(CPUState *cpu, run_on_cpu_func func, void *data) |
3c02270d CV |
958 | { |
959 | struct qemu_work_item *wi; | |
960 | ||
961 | if (qemu_cpu_is_self(cpu)) { | |
e0eeb4a2 | 962 | func(cpu, data); |
3c02270d CV |
963 | return; |
964 | } | |
965 | ||
966 | wi = g_malloc0(sizeof(struct qemu_work_item)); | |
967 | wi->func = func; | |
968 | wi->data = data; | |
969 | wi->free = true; | |
376692b9 | 970 | |
fd38b251 | 971 | queue_work_on_cpu(cpu, wi); |
3c02270d CV |
972 | } |
973 | ||
4c055ab5 GZ |
974 | static void qemu_kvm_destroy_vcpu(CPUState *cpu) |
975 | { | |
976 | if (kvm_destroy_vcpu(cpu) < 0) { | |
977 | error_report("kvm_destroy_vcpu failed"); | |
978 | exit(EXIT_FAILURE); | |
979 | } | |
980 | } | |
981 | ||
982 | static void qemu_tcg_destroy_vcpu(CPUState *cpu) | |
983 | { | |
984 | } | |
985 | ||
a5403c69 | 986 | static void process_queued_cpu_work(CPUState *cpu) |
e82bcec2 MT |
987 | { |
988 | struct qemu_work_item *wi; | |
989 | ||
c64ca814 | 990 | if (cpu->queued_work_first == NULL) { |
e82bcec2 | 991 | return; |
0ab07c62 | 992 | } |
e82bcec2 | 993 | |
376692b9 PB |
994 | qemu_mutex_lock(&cpu->work_mutex); |
995 | while (cpu->queued_work_first != NULL) { | |
996 | wi = cpu->queued_work_first; | |
c64ca814 | 997 | cpu->queued_work_first = wi->next; |
376692b9 PB |
998 | if (!cpu->queued_work_first) { |
999 | cpu->queued_work_last = NULL; | |
1000 | } | |
1001 | qemu_mutex_unlock(&cpu->work_mutex); | |
e0eeb4a2 | 1002 | wi->func(cpu, wi->data); |
376692b9 | 1003 | qemu_mutex_lock(&cpu->work_mutex); |
3c02270d CV |
1004 | if (wi->free) { |
1005 | g_free(wi); | |
376692b9 PB |
1006 | } else { |
1007 | atomic_mb_set(&wi->done, true); | |
3c02270d | 1008 | } |
e82bcec2 | 1009 | } |
376692b9 | 1010 | qemu_mutex_unlock(&cpu->work_mutex); |
e82bcec2 MT |
1011 | qemu_cond_broadcast(&qemu_work_cond); |
1012 | } | |
1013 | ||
509a0d78 | 1014 | static void qemu_wait_io_event_common(CPUState *cpu) |
296af7c9 | 1015 | { |
4fdeee7c AF |
1016 | if (cpu->stop) { |
1017 | cpu->stop = false; | |
f324e766 | 1018 | cpu->stopped = true; |
96bce683 | 1019 | qemu_cond_broadcast(&qemu_pause_cond); |
296af7c9 | 1020 | } |
a5403c69 | 1021 | process_queued_cpu_work(cpu); |
216fc9a4 | 1022 | cpu->thread_kicked = false; |
296af7c9 BS |
1023 | } |
1024 | ||
d5f8d613 | 1025 | static void qemu_tcg_wait_io_event(CPUState *cpu) |
296af7c9 | 1026 | { |
16400322 | 1027 | while (all_cpu_threads_idle()) { |
d5f8d613 | 1028 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); |
16400322 | 1029 | } |
296af7c9 | 1030 | |
46daff13 PB |
1031 | while (iothread_requesting_mutex) { |
1032 | qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex); | |
1033 | } | |
6cabe1f3 | 1034 | |
bdc44640 | 1035 | CPU_FOREACH(cpu) { |
182735ef | 1036 | qemu_wait_io_event_common(cpu); |
6cabe1f3 | 1037 | } |
296af7c9 BS |
1038 | } |
1039 | ||
fd529e8f | 1040 | static void qemu_kvm_wait_io_event(CPUState *cpu) |
296af7c9 | 1041 | { |
a98ae1d8 | 1042 | while (cpu_thread_is_idle(cpu)) { |
f5c121b8 | 1043 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); |
16400322 | 1044 | } |
296af7c9 | 1045 | |
290adf38 | 1046 | qemu_kvm_eat_signals(cpu); |
509a0d78 | 1047 | qemu_wait_io_event_common(cpu); |
296af7c9 BS |
1048 | } |
1049 | ||
7e97cd88 | 1050 | static void *qemu_kvm_cpu_thread_fn(void *arg) |
296af7c9 | 1051 | { |
48a106bd | 1052 | CPUState *cpu = arg; |
84b4915d | 1053 | int r; |
296af7c9 | 1054 | |
ab28bd23 PB |
1055 | rcu_register_thread(); |
1056 | ||
2e7f7a3c | 1057 | qemu_mutex_lock_iothread(); |
814e612e | 1058 | qemu_thread_get_self(cpu->thread); |
9f09e18a | 1059 | cpu->thread_id = qemu_get_thread_id(); |
626cf8f4 | 1060 | cpu->can_do_io = 1; |
4917cf44 | 1061 | current_cpu = cpu; |
296af7c9 | 1062 | |
504134d2 | 1063 | r = kvm_init_vcpu(cpu); |
84b4915d JK |
1064 | if (r < 0) { |
1065 | fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r)); | |
1066 | exit(1); | |
1067 | } | |
296af7c9 | 1068 | |
13618e05 | 1069 | qemu_kvm_init_cpu_signals(cpu); |
296af7c9 BS |
1070 | |
1071 | /* signal CPU creation */ | |
61a46217 | 1072 | cpu->created = true; |
296af7c9 BS |
1073 | qemu_cond_signal(&qemu_cpu_cond); |
1074 | ||
4c055ab5 | 1075 | do { |
a1fcaa73 | 1076 | if (cpu_can_run(cpu)) { |
1458c363 | 1077 | r = kvm_cpu_exec(cpu); |
83f338f7 | 1078 | if (r == EXCP_DEBUG) { |
91325046 | 1079 | cpu_handle_guest_debug(cpu); |
83f338f7 | 1080 | } |
0ab07c62 | 1081 | } |
fd529e8f | 1082 | qemu_kvm_wait_io_event(cpu); |
4c055ab5 | 1083 | } while (!cpu->unplug || cpu_can_run(cpu)); |
296af7c9 | 1084 | |
4c055ab5 | 1085 | qemu_kvm_destroy_vcpu(cpu); |
2c579042 BR |
1086 | cpu->created = false; |
1087 | qemu_cond_signal(&qemu_cpu_cond); | |
4c055ab5 | 1088 | qemu_mutex_unlock_iothread(); |
296af7c9 BS |
1089 | return NULL; |
1090 | } | |
1091 | ||
c7f0f3b1 AL |
1092 | static void *qemu_dummy_cpu_thread_fn(void *arg) |
1093 | { | |
1094 | #ifdef _WIN32 | |
1095 | fprintf(stderr, "qtest is not supported under Windows\n"); | |
1096 | exit(1); | |
1097 | #else | |
10a9021d | 1098 | CPUState *cpu = arg; |
c7f0f3b1 AL |
1099 | sigset_t waitset; |
1100 | int r; | |
1101 | ||
ab28bd23 PB |
1102 | rcu_register_thread(); |
1103 | ||
c7f0f3b1 | 1104 | qemu_mutex_lock_iothread(); |
814e612e | 1105 | qemu_thread_get_self(cpu->thread); |
9f09e18a | 1106 | cpu->thread_id = qemu_get_thread_id(); |
626cf8f4 | 1107 | cpu->can_do_io = 1; |
c7f0f3b1 AL |
1108 | |
1109 | sigemptyset(&waitset); | |
1110 | sigaddset(&waitset, SIG_IPI); | |
1111 | ||
1112 | /* signal CPU creation */ | |
61a46217 | 1113 | cpu->created = true; |
c7f0f3b1 AL |
1114 | qemu_cond_signal(&qemu_cpu_cond); |
1115 | ||
4917cf44 | 1116 | current_cpu = cpu; |
c7f0f3b1 | 1117 | while (1) { |
4917cf44 | 1118 | current_cpu = NULL; |
c7f0f3b1 AL |
1119 | qemu_mutex_unlock_iothread(); |
1120 | do { | |
1121 | int sig; | |
1122 | r = sigwait(&waitset, &sig); | |
1123 | } while (r == -1 && (errno == EAGAIN || errno == EINTR)); | |
1124 | if (r == -1) { | |
1125 | perror("sigwait"); | |
1126 | exit(1); | |
1127 | } | |
1128 | qemu_mutex_lock_iothread(); | |
4917cf44 | 1129 | current_cpu = cpu; |
509a0d78 | 1130 | qemu_wait_io_event_common(cpu); |
c7f0f3b1 AL |
1131 | } |
1132 | ||
1133 | return NULL; | |
1134 | #endif | |
1135 | } | |
1136 | ||
bdb7ca67 JK |
1137 | static void tcg_exec_all(void); |
1138 | ||
7e97cd88 | 1139 | static void *qemu_tcg_cpu_thread_fn(void *arg) |
296af7c9 | 1140 | { |
c3586ba7 | 1141 | CPUState *cpu = arg; |
4c055ab5 | 1142 | CPUState *remove_cpu = NULL; |
296af7c9 | 1143 | |
ab28bd23 PB |
1144 | rcu_register_thread(); |
1145 | ||
2e7f7a3c | 1146 | qemu_mutex_lock_iothread(); |
814e612e | 1147 | qemu_thread_get_self(cpu->thread); |
296af7c9 | 1148 | |
38fcbd3f AF |
1149 | CPU_FOREACH(cpu) { |
1150 | cpu->thread_id = qemu_get_thread_id(); | |
1151 | cpu->created = true; | |
626cf8f4 | 1152 | cpu->can_do_io = 1; |
38fcbd3f | 1153 | } |
296af7c9 BS |
1154 | qemu_cond_signal(&qemu_cpu_cond); |
1155 | ||
fa7d1867 | 1156 | /* wait for initial kick-off after machine start */ |
c28e399c | 1157 | while (first_cpu->stopped) { |
d5f8d613 | 1158 | qemu_cond_wait(first_cpu->halt_cond, &qemu_global_mutex); |
8e564b4e JK |
1159 | |
1160 | /* process any pending work */ | |
bdc44640 | 1161 | CPU_FOREACH(cpu) { |
182735ef | 1162 | qemu_wait_io_event_common(cpu); |
8e564b4e | 1163 | } |
0ab07c62 | 1164 | } |
296af7c9 | 1165 | |
21618b3e | 1166 | /* process any pending work */ |
aed807c8 | 1167 | atomic_mb_set(&exit_request, 1); |
21618b3e | 1168 | |
296af7c9 | 1169 | while (1) { |
bdb7ca67 | 1170 | tcg_exec_all(); |
ac70aafc AB |
1171 | |
1172 | if (use_icount) { | |
40daca54 | 1173 | int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); |
ac70aafc AB |
1174 | |
1175 | if (deadline == 0) { | |
40daca54 | 1176 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
ac70aafc | 1177 | } |
3b2319a3 | 1178 | } |
d5f8d613 | 1179 | qemu_tcg_wait_io_event(QTAILQ_FIRST(&cpus)); |
4c055ab5 GZ |
1180 | CPU_FOREACH(cpu) { |
1181 | if (cpu->unplug && !cpu_can_run(cpu)) { | |
1182 | remove_cpu = cpu; | |
1183 | break; | |
1184 | } | |
1185 | } | |
1186 | if (remove_cpu) { | |
1187 | qemu_tcg_destroy_vcpu(remove_cpu); | |
2c579042 BR |
1188 | cpu->created = false; |
1189 | qemu_cond_signal(&qemu_cpu_cond); | |
4c055ab5 GZ |
1190 | remove_cpu = NULL; |
1191 | } | |
296af7c9 BS |
1192 | } |
1193 | ||
1194 | return NULL; | |
1195 | } | |
1196 | ||
2ff09a40 | 1197 | static void qemu_cpu_kick_thread(CPUState *cpu) |
cc015e9a PB |
1198 | { |
1199 | #ifndef _WIN32 | |
1200 | int err; | |
1201 | ||
e0c38211 PB |
1202 | if (cpu->thread_kicked) { |
1203 | return; | |
9102deda | 1204 | } |
e0c38211 | 1205 | cpu->thread_kicked = true; |
814e612e | 1206 | err = pthread_kill(cpu->thread->thread, SIG_IPI); |
cc015e9a PB |
1207 | if (err) { |
1208 | fprintf(stderr, "qemu:%s: %s", __func__, strerror(err)); | |
1209 | exit(1); | |
1210 | } | |
1211 | #else /* _WIN32 */ | |
e0c38211 PB |
1212 | abort(); |
1213 | #endif | |
1214 | } | |
ed9164a3 | 1215 | |
e0c38211 PB |
1216 | static void qemu_cpu_kick_no_halt(void) |
1217 | { | |
1218 | CPUState *cpu; | |
1219 | /* Ensure whatever caused the exit has reached the CPU threads before | |
1220 | * writing exit_request. | |
1221 | */ | |
1222 | atomic_mb_set(&exit_request, 1); | |
1223 | cpu = atomic_mb_read(&tcg_current_cpu); | |
1224 | if (cpu) { | |
1225 | cpu_exit(cpu); | |
cc015e9a | 1226 | } |
cc015e9a PB |
1227 | } |
1228 | ||
c08d7424 | 1229 | void qemu_cpu_kick(CPUState *cpu) |
296af7c9 | 1230 | { |
f5c121b8 | 1231 | qemu_cond_broadcast(cpu->halt_cond); |
e0c38211 PB |
1232 | if (tcg_enabled()) { |
1233 | qemu_cpu_kick_no_halt(); | |
1234 | } else { | |
1235 | qemu_cpu_kick_thread(cpu); | |
1236 | } | |
296af7c9 BS |
1237 | } |
1238 | ||
46d62fac | 1239 | void qemu_cpu_kick_self(void) |
296af7c9 | 1240 | { |
4917cf44 | 1241 | assert(current_cpu); |
9102deda | 1242 | qemu_cpu_kick_thread(current_cpu); |
296af7c9 BS |
1243 | } |
1244 | ||
60e82579 | 1245 | bool qemu_cpu_is_self(CPUState *cpu) |
296af7c9 | 1246 | { |
814e612e | 1247 | return qemu_thread_is_self(cpu->thread); |
296af7c9 BS |
1248 | } |
1249 | ||
79e2b9ae | 1250 | bool qemu_in_vcpu_thread(void) |
aa723c23 | 1251 | { |
4917cf44 | 1252 | return current_cpu && qemu_cpu_is_self(current_cpu); |
aa723c23 JQ |
1253 | } |
1254 | ||
afbe7053 PB |
1255 | static __thread bool iothread_locked = false; |
1256 | ||
1257 | bool qemu_mutex_iothread_locked(void) | |
1258 | { | |
1259 | return iothread_locked; | |
1260 | } | |
1261 | ||
296af7c9 BS |
1262 | void qemu_mutex_lock_iothread(void) |
1263 | { | |
21618b3e | 1264 | atomic_inc(&iothread_requesting_mutex); |
2e7f7a3c PB |
1265 | /* In the simple case there is no need to bump the VCPU thread out of |
1266 | * TCG code execution. | |
1267 | */ | |
1268 | if (!tcg_enabled() || qemu_in_vcpu_thread() || | |
46036b24 | 1269 | !first_cpu || !first_cpu->created) { |
296af7c9 | 1270 | qemu_mutex_lock(&qemu_global_mutex); |
21618b3e | 1271 | atomic_dec(&iothread_requesting_mutex); |
1a28cac3 | 1272 | } else { |
1a28cac3 | 1273 | if (qemu_mutex_trylock(&qemu_global_mutex)) { |
e0c38211 | 1274 | qemu_cpu_kick_no_halt(); |
1a28cac3 MT |
1275 | qemu_mutex_lock(&qemu_global_mutex); |
1276 | } | |
6b49809c | 1277 | atomic_dec(&iothread_requesting_mutex); |
46daff13 | 1278 | qemu_cond_broadcast(&qemu_io_proceeded_cond); |
1a28cac3 | 1279 | } |
afbe7053 | 1280 | iothread_locked = true; |
296af7c9 BS |
1281 | } |
1282 | ||
1283 | void qemu_mutex_unlock_iothread(void) | |
1284 | { | |
afbe7053 | 1285 | iothread_locked = false; |
296af7c9 BS |
1286 | qemu_mutex_unlock(&qemu_global_mutex); |
1287 | } | |
1288 | ||
1289 | static int all_vcpus_paused(void) | |
1290 | { | |
bdc44640 | 1291 | CPUState *cpu; |
296af7c9 | 1292 | |
bdc44640 | 1293 | CPU_FOREACH(cpu) { |
182735ef | 1294 | if (!cpu->stopped) { |
296af7c9 | 1295 | return 0; |
0ab07c62 | 1296 | } |
296af7c9 BS |
1297 | } |
1298 | ||
1299 | return 1; | |
1300 | } | |
1301 | ||
1302 | void pause_all_vcpus(void) | |
1303 | { | |
bdc44640 | 1304 | CPUState *cpu; |
296af7c9 | 1305 | |
40daca54 | 1306 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false); |
bdc44640 | 1307 | CPU_FOREACH(cpu) { |
182735ef AF |
1308 | cpu->stop = true; |
1309 | qemu_cpu_kick(cpu); | |
296af7c9 BS |
1310 | } |
1311 | ||
aa723c23 | 1312 | if (qemu_in_vcpu_thread()) { |
d798e974 JK |
1313 | cpu_stop_current(); |
1314 | if (!kvm_enabled()) { | |
bdc44640 | 1315 | CPU_FOREACH(cpu) { |
182735ef AF |
1316 | cpu->stop = false; |
1317 | cpu->stopped = true; | |
d798e974 JK |
1318 | } |
1319 | return; | |
1320 | } | |
1321 | } | |
1322 | ||
296af7c9 | 1323 | while (!all_vcpus_paused()) { |
be7d6c57 | 1324 | qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex); |
bdc44640 | 1325 | CPU_FOREACH(cpu) { |
182735ef | 1326 | qemu_cpu_kick(cpu); |
296af7c9 BS |
1327 | } |
1328 | } | |
1329 | } | |
1330 | ||
2993683b IM |
1331 | void cpu_resume(CPUState *cpu) |
1332 | { | |
1333 | cpu->stop = false; | |
1334 | cpu->stopped = false; | |
1335 | qemu_cpu_kick(cpu); | |
1336 | } | |
1337 | ||
296af7c9 BS |
1338 | void resume_all_vcpus(void) |
1339 | { | |
bdc44640 | 1340 | CPUState *cpu; |
296af7c9 | 1341 | |
40daca54 | 1342 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true); |
bdc44640 | 1343 | CPU_FOREACH(cpu) { |
182735ef | 1344 | cpu_resume(cpu); |
296af7c9 BS |
1345 | } |
1346 | } | |
1347 | ||
4c055ab5 GZ |
1348 | void cpu_remove(CPUState *cpu) |
1349 | { | |
1350 | cpu->stop = true; | |
1351 | cpu->unplug = true; | |
1352 | qemu_cpu_kick(cpu); | |
1353 | } | |
1354 | ||
2c579042 BR |
1355 | void cpu_remove_sync(CPUState *cpu) |
1356 | { | |
1357 | cpu_remove(cpu); | |
1358 | while (cpu->created) { | |
1359 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1360 | } | |
1361 | } | |
1362 | ||
4900116e DDAG |
1363 | /* For temporary buffers for forming a name */ |
1364 | #define VCPU_THREAD_NAME_SIZE 16 | |
1365 | ||
e5ab30a2 | 1366 | static void qemu_tcg_init_vcpu(CPUState *cpu) |
296af7c9 | 1367 | { |
4900116e | 1368 | char thread_name[VCPU_THREAD_NAME_SIZE]; |
d5f8d613 FK |
1369 | static QemuCond *tcg_halt_cond; |
1370 | static QemuThread *tcg_cpu_thread; | |
4900116e | 1371 | |
296af7c9 BS |
1372 | /* share a single thread for all cpus with TCG */ |
1373 | if (!tcg_cpu_thread) { | |
814e612e | 1374 | cpu->thread = g_malloc0(sizeof(QemuThread)); |
f5c121b8 AF |
1375 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); |
1376 | qemu_cond_init(cpu->halt_cond); | |
1377 | tcg_halt_cond = cpu->halt_cond; | |
4900116e DDAG |
1378 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG", |
1379 | cpu->cpu_index); | |
1380 | qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn, | |
1381 | cpu, QEMU_THREAD_JOINABLE); | |
1ecf47bf | 1382 | #ifdef _WIN32 |
814e612e | 1383 | cpu->hThread = qemu_thread_get_handle(cpu->thread); |
1ecf47bf | 1384 | #endif |
61a46217 | 1385 | while (!cpu->created) { |
18a85728 | 1386 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); |
0ab07c62 | 1387 | } |
814e612e | 1388 | tcg_cpu_thread = cpu->thread; |
296af7c9 | 1389 | } else { |
814e612e | 1390 | cpu->thread = tcg_cpu_thread; |
f5c121b8 | 1391 | cpu->halt_cond = tcg_halt_cond; |
296af7c9 BS |
1392 | } |
1393 | } | |
1394 | ||
48a106bd | 1395 | static void qemu_kvm_start_vcpu(CPUState *cpu) |
296af7c9 | 1396 | { |
4900116e DDAG |
1397 | char thread_name[VCPU_THREAD_NAME_SIZE]; |
1398 | ||
814e612e | 1399 | cpu->thread = g_malloc0(sizeof(QemuThread)); |
f5c121b8 AF |
1400 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); |
1401 | qemu_cond_init(cpu->halt_cond); | |
4900116e DDAG |
1402 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/KVM", |
1403 | cpu->cpu_index); | |
1404 | qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn, | |
1405 | cpu, QEMU_THREAD_JOINABLE); | |
61a46217 | 1406 | while (!cpu->created) { |
18a85728 | 1407 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); |
0ab07c62 | 1408 | } |
296af7c9 BS |
1409 | } |
1410 | ||
10a9021d | 1411 | static void qemu_dummy_start_vcpu(CPUState *cpu) |
c7f0f3b1 | 1412 | { |
4900116e DDAG |
1413 | char thread_name[VCPU_THREAD_NAME_SIZE]; |
1414 | ||
814e612e | 1415 | cpu->thread = g_malloc0(sizeof(QemuThread)); |
f5c121b8 AF |
1416 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); |
1417 | qemu_cond_init(cpu->halt_cond); | |
4900116e DDAG |
1418 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/DUMMY", |
1419 | cpu->cpu_index); | |
1420 | qemu_thread_create(cpu->thread, thread_name, qemu_dummy_cpu_thread_fn, cpu, | |
c7f0f3b1 | 1421 | QEMU_THREAD_JOINABLE); |
61a46217 | 1422 | while (!cpu->created) { |
c7f0f3b1 AL |
1423 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); |
1424 | } | |
1425 | } | |
1426 | ||
c643bed9 | 1427 | void qemu_init_vcpu(CPUState *cpu) |
296af7c9 | 1428 | { |
ce3960eb AF |
1429 | cpu->nr_cores = smp_cores; |
1430 | cpu->nr_threads = smp_threads; | |
f324e766 | 1431 | cpu->stopped = true; |
56943e8c PM |
1432 | |
1433 | if (!cpu->as) { | |
1434 | /* If the target cpu hasn't set up any address spaces itself, | |
1435 | * give it the default one. | |
1436 | */ | |
6731d864 PC |
1437 | AddressSpace *as = address_space_init_shareable(cpu->memory, |
1438 | "cpu-memory"); | |
12ebc9a7 | 1439 | cpu->num_ases = 1; |
6731d864 | 1440 | cpu_address_space_init(cpu, as, 0); |
56943e8c PM |
1441 | } |
1442 | ||
0ab07c62 | 1443 | if (kvm_enabled()) { |
48a106bd | 1444 | qemu_kvm_start_vcpu(cpu); |
c7f0f3b1 | 1445 | } else if (tcg_enabled()) { |
e5ab30a2 | 1446 | qemu_tcg_init_vcpu(cpu); |
c7f0f3b1 | 1447 | } else { |
10a9021d | 1448 | qemu_dummy_start_vcpu(cpu); |
0ab07c62 | 1449 | } |
296af7c9 BS |
1450 | } |
1451 | ||
b4a3d965 | 1452 | void cpu_stop_current(void) |
296af7c9 | 1453 | { |
4917cf44 AF |
1454 | if (current_cpu) { |
1455 | current_cpu->stop = false; | |
1456 | current_cpu->stopped = true; | |
1457 | cpu_exit(current_cpu); | |
96bce683 | 1458 | qemu_cond_broadcast(&qemu_pause_cond); |
b4a3d965 | 1459 | } |
296af7c9 BS |
1460 | } |
1461 | ||
56983463 | 1462 | int vm_stop(RunState state) |
296af7c9 | 1463 | { |
aa723c23 | 1464 | if (qemu_in_vcpu_thread()) { |
74892d24 | 1465 | qemu_system_vmstop_request_prepare(); |
1dfb4dd9 | 1466 | qemu_system_vmstop_request(state); |
296af7c9 BS |
1467 | /* |
1468 | * FIXME: should not return to device code in case | |
1469 | * vm_stop() has been requested. | |
1470 | */ | |
b4a3d965 | 1471 | cpu_stop_current(); |
56983463 | 1472 | return 0; |
296af7c9 | 1473 | } |
56983463 KW |
1474 | |
1475 | return do_vm_stop(state); | |
296af7c9 BS |
1476 | } |
1477 | ||
8a9236f1 LC |
1478 | /* does a state transition even if the VM is already stopped, |
1479 | current state is forgotten forever */ | |
56983463 | 1480 | int vm_stop_force_state(RunState state) |
8a9236f1 LC |
1481 | { |
1482 | if (runstate_is_running()) { | |
56983463 | 1483 | return vm_stop(state); |
8a9236f1 LC |
1484 | } else { |
1485 | runstate_set(state); | |
b2780d32 WC |
1486 | |
1487 | bdrv_drain_all(); | |
594a45ce KW |
1488 | /* Make sure to return an error if the flush in a previous vm_stop() |
1489 | * failed. */ | |
da31d594 | 1490 | return blk_flush_all(); |
8a9236f1 LC |
1491 | } |
1492 | } | |
1493 | ||
8b427044 PD |
1494 | static int64_t tcg_get_icount_limit(void) |
1495 | { | |
1496 | int64_t deadline; | |
1497 | ||
1498 | if (replay_mode != REPLAY_MODE_PLAY) { | |
1499 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
1500 | ||
1501 | /* Maintain prior (possibly buggy) behaviour where if no deadline | |
1502 | * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than | |
1503 | * INT32_MAX nanoseconds ahead, we still use INT32_MAX | |
1504 | * nanoseconds. | |
1505 | */ | |
1506 | if ((deadline < 0) || (deadline > INT32_MAX)) { | |
1507 | deadline = INT32_MAX; | |
1508 | } | |
1509 | ||
1510 | return qemu_icount_round(deadline); | |
1511 | } else { | |
1512 | return replay_get_instructions(); | |
1513 | } | |
1514 | } | |
1515 | ||
3d57f789 | 1516 | static int tcg_cpu_exec(CPUState *cpu) |
296af7c9 BS |
1517 | { |
1518 | int ret; | |
1519 | #ifdef CONFIG_PROFILER | |
1520 | int64_t ti; | |
1521 | #endif | |
1522 | ||
1523 | #ifdef CONFIG_PROFILER | |
1524 | ti = profile_getclock(); | |
1525 | #endif | |
1526 | if (use_icount) { | |
1527 | int64_t count; | |
1528 | int decr; | |
c96778bb FK |
1529 | timers_state.qemu_icount -= (cpu->icount_decr.u16.low |
1530 | + cpu->icount_extra); | |
28ecfd7a | 1531 | cpu->icount_decr.u16.low = 0; |
efee7340 | 1532 | cpu->icount_extra = 0; |
8b427044 | 1533 | count = tcg_get_icount_limit(); |
c96778bb | 1534 | timers_state.qemu_icount += count; |
296af7c9 BS |
1535 | decr = (count > 0xffff) ? 0xffff : count; |
1536 | count -= decr; | |
28ecfd7a | 1537 | cpu->icount_decr.u16.low = decr; |
efee7340 | 1538 | cpu->icount_extra = count; |
296af7c9 | 1539 | } |
ea3e9847 | 1540 | ret = cpu_exec(cpu); |
296af7c9 | 1541 | #ifdef CONFIG_PROFILER |
89d5cbdd | 1542 | tcg_time += profile_getclock() - ti; |
296af7c9 BS |
1543 | #endif |
1544 | if (use_icount) { | |
1545 | /* Fold pending instructions back into the | |
1546 | instruction counter, and clear the interrupt flag. */ | |
c96778bb FK |
1547 | timers_state.qemu_icount -= (cpu->icount_decr.u16.low |
1548 | + cpu->icount_extra); | |
28ecfd7a | 1549 | cpu->icount_decr.u32 = 0; |
efee7340 | 1550 | cpu->icount_extra = 0; |
8b427044 | 1551 | replay_account_executed_instructions(); |
296af7c9 BS |
1552 | } |
1553 | return ret; | |
1554 | } | |
1555 | ||
bdb7ca67 | 1556 | static void tcg_exec_all(void) |
296af7c9 | 1557 | { |
9a36085b JK |
1558 | int r; |
1559 | ||
40daca54 | 1560 | /* Account partial waits to QEMU_CLOCK_VIRTUAL. */ |
e76d1798 | 1561 | qemu_account_warp_timer(); |
ab33fcda | 1562 | |
0ab07c62 | 1563 | if (next_cpu == NULL) { |
296af7c9 | 1564 | next_cpu = first_cpu; |
0ab07c62 | 1565 | } |
bdc44640 | 1566 | for (; next_cpu != NULL && !exit_request; next_cpu = CPU_NEXT(next_cpu)) { |
182735ef | 1567 | CPUState *cpu = next_cpu; |
296af7c9 | 1568 | |
40daca54 | 1569 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, |
ed2803da | 1570 | (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0); |
296af7c9 | 1571 | |
a1fcaa73 | 1572 | if (cpu_can_run(cpu)) { |
3d57f789 | 1573 | r = tcg_cpu_exec(cpu); |
9a36085b | 1574 | if (r == EXCP_DEBUG) { |
91325046 | 1575 | cpu_handle_guest_debug(cpu); |
3c638d06 JK |
1576 | break; |
1577 | } | |
f324e766 | 1578 | } else if (cpu->stop || cpu->stopped) { |
4c055ab5 GZ |
1579 | if (cpu->unplug) { |
1580 | next_cpu = CPU_NEXT(cpu); | |
1581 | } | |
296af7c9 BS |
1582 | break; |
1583 | } | |
1584 | } | |
aed807c8 PB |
1585 | |
1586 | /* Pairs with smp_wmb in qemu_cpu_kick. */ | |
1587 | atomic_mb_set(&exit_request, 0); | |
296af7c9 BS |
1588 | } |
1589 | ||
9a78eead | 1590 | void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg) |
262353cb BS |
1591 | { |
1592 | /* XXX: implement xxx_cpu_list for targets that still miss it */ | |
e916cbf8 PM |
1593 | #if defined(cpu_list) |
1594 | cpu_list(f, cpu_fprintf); | |
262353cb BS |
1595 | #endif |
1596 | } | |
de0b36b6 LC |
1597 | |
1598 | CpuInfoList *qmp_query_cpus(Error **errp) | |
1599 | { | |
1600 | CpuInfoList *head = NULL, *cur_item = NULL; | |
182735ef | 1601 | CPUState *cpu; |
de0b36b6 | 1602 | |
bdc44640 | 1603 | CPU_FOREACH(cpu) { |
de0b36b6 | 1604 | CpuInfoList *info; |
182735ef AF |
1605 | #if defined(TARGET_I386) |
1606 | X86CPU *x86_cpu = X86_CPU(cpu); | |
1607 | CPUX86State *env = &x86_cpu->env; | |
1608 | #elif defined(TARGET_PPC) | |
1609 | PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu); | |
1610 | CPUPPCState *env = &ppc_cpu->env; | |
1611 | #elif defined(TARGET_SPARC) | |
1612 | SPARCCPU *sparc_cpu = SPARC_CPU(cpu); | |
1613 | CPUSPARCState *env = &sparc_cpu->env; | |
1614 | #elif defined(TARGET_MIPS) | |
1615 | MIPSCPU *mips_cpu = MIPS_CPU(cpu); | |
1616 | CPUMIPSState *env = &mips_cpu->env; | |
48e06fe0 BK |
1617 | #elif defined(TARGET_TRICORE) |
1618 | TriCoreCPU *tricore_cpu = TRICORE_CPU(cpu); | |
1619 | CPUTriCoreState *env = &tricore_cpu->env; | |
182735ef | 1620 | #endif |
de0b36b6 | 1621 | |
cb446eca | 1622 | cpu_synchronize_state(cpu); |
de0b36b6 LC |
1623 | |
1624 | info = g_malloc0(sizeof(*info)); | |
1625 | info->value = g_malloc0(sizeof(*info->value)); | |
55e5c285 | 1626 | info->value->CPU = cpu->cpu_index; |
182735ef | 1627 | info->value->current = (cpu == first_cpu); |
259186a7 | 1628 | info->value->halted = cpu->halted; |
58f88d4b | 1629 | info->value->qom_path = object_get_canonical_path(OBJECT(cpu)); |
9f09e18a | 1630 | info->value->thread_id = cpu->thread_id; |
de0b36b6 | 1631 | #if defined(TARGET_I386) |
86f4b687 | 1632 | info->value->arch = CPU_INFO_ARCH_X86; |
544a3731 | 1633 | info->value->u.x86.pc = env->eip + env->segs[R_CS].base; |
de0b36b6 | 1634 | #elif defined(TARGET_PPC) |
86f4b687 | 1635 | info->value->arch = CPU_INFO_ARCH_PPC; |
544a3731 | 1636 | info->value->u.ppc.nip = env->nip; |
de0b36b6 | 1637 | #elif defined(TARGET_SPARC) |
86f4b687 | 1638 | info->value->arch = CPU_INFO_ARCH_SPARC; |
544a3731 EB |
1639 | info->value->u.q_sparc.pc = env->pc; |
1640 | info->value->u.q_sparc.npc = env->npc; | |
de0b36b6 | 1641 | #elif defined(TARGET_MIPS) |
86f4b687 | 1642 | info->value->arch = CPU_INFO_ARCH_MIPS; |
544a3731 | 1643 | info->value->u.q_mips.PC = env->active_tc.PC; |
48e06fe0 | 1644 | #elif defined(TARGET_TRICORE) |
86f4b687 | 1645 | info->value->arch = CPU_INFO_ARCH_TRICORE; |
544a3731 | 1646 | info->value->u.tricore.PC = env->PC; |
86f4b687 EB |
1647 | #else |
1648 | info->value->arch = CPU_INFO_ARCH_OTHER; | |
de0b36b6 LC |
1649 | #endif |
1650 | ||
1651 | /* XXX: waiting for the qapi to support GSList */ | |
1652 | if (!cur_item) { | |
1653 | head = cur_item = info; | |
1654 | } else { | |
1655 | cur_item->next = info; | |
1656 | cur_item = info; | |
1657 | } | |
1658 | } | |
1659 | ||
1660 | return head; | |
1661 | } | |
0cfd6a9a LC |
1662 | |
1663 | void qmp_memsave(int64_t addr, int64_t size, const char *filename, | |
1664 | bool has_cpu, int64_t cpu_index, Error **errp) | |
1665 | { | |
1666 | FILE *f; | |
1667 | uint32_t l; | |
55e5c285 | 1668 | CPUState *cpu; |
0cfd6a9a | 1669 | uint8_t buf[1024]; |
0dc9daf0 | 1670 | int64_t orig_addr = addr, orig_size = size; |
0cfd6a9a LC |
1671 | |
1672 | if (!has_cpu) { | |
1673 | cpu_index = 0; | |
1674 | } | |
1675 | ||
151d1322 AF |
1676 | cpu = qemu_get_cpu(cpu_index); |
1677 | if (cpu == NULL) { | |
c6bd8c70 MA |
1678 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index", |
1679 | "a CPU number"); | |
0cfd6a9a LC |
1680 | return; |
1681 | } | |
1682 | ||
1683 | f = fopen(filename, "wb"); | |
1684 | if (!f) { | |
618da851 | 1685 | error_setg_file_open(errp, errno, filename); |
0cfd6a9a LC |
1686 | return; |
1687 | } | |
1688 | ||
1689 | while (size != 0) { | |
1690 | l = sizeof(buf); | |
1691 | if (l > size) | |
1692 | l = size; | |
2f4d0f59 | 1693 | if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) { |
0dc9daf0 BP |
1694 | error_setg(errp, "Invalid addr 0x%016" PRIx64 "/size %" PRId64 |
1695 | " specified", orig_addr, orig_size); | |
2f4d0f59 AK |
1696 | goto exit; |
1697 | } | |
0cfd6a9a | 1698 | if (fwrite(buf, 1, l, f) != l) { |
c6bd8c70 | 1699 | error_setg(errp, QERR_IO_ERROR); |
0cfd6a9a LC |
1700 | goto exit; |
1701 | } | |
1702 | addr += l; | |
1703 | size -= l; | |
1704 | } | |
1705 | ||
1706 | exit: | |
1707 | fclose(f); | |
1708 | } | |
6d3962bf LC |
1709 | |
1710 | void qmp_pmemsave(int64_t addr, int64_t size, const char *filename, | |
1711 | Error **errp) | |
1712 | { | |
1713 | FILE *f; | |
1714 | uint32_t l; | |
1715 | uint8_t buf[1024]; | |
1716 | ||
1717 | f = fopen(filename, "wb"); | |
1718 | if (!f) { | |
618da851 | 1719 | error_setg_file_open(errp, errno, filename); |
6d3962bf LC |
1720 | return; |
1721 | } | |
1722 | ||
1723 | while (size != 0) { | |
1724 | l = sizeof(buf); | |
1725 | if (l > size) | |
1726 | l = size; | |
eb6282f2 | 1727 | cpu_physical_memory_read(addr, buf, l); |
6d3962bf | 1728 | if (fwrite(buf, 1, l, f) != l) { |
c6bd8c70 | 1729 | error_setg(errp, QERR_IO_ERROR); |
6d3962bf LC |
1730 | goto exit; |
1731 | } | |
1732 | addr += l; | |
1733 | size -= l; | |
1734 | } | |
1735 | ||
1736 | exit: | |
1737 | fclose(f); | |
1738 | } | |
ab49ab5c LC |
1739 | |
1740 | void qmp_inject_nmi(Error **errp) | |
1741 | { | |
9cb805fd | 1742 | nmi_monitor_handle(monitor_get_cpu_index(), errp); |
ab49ab5c | 1743 | } |
27498bef ST |
1744 | |
1745 | void dump_drift_info(FILE *f, fprintf_function cpu_fprintf) | |
1746 | { | |
1747 | if (!use_icount) { | |
1748 | return; | |
1749 | } | |
1750 | ||
1751 | cpu_fprintf(f, "Host - Guest clock %"PRIi64" ms\n", | |
1752 | (cpu_get_clock() - cpu_get_icount())/SCALE_MS); | |
1753 | if (icount_align_option) { | |
1754 | cpu_fprintf(f, "Max guest delay %"PRIi64" ms\n", -max_delay/SCALE_MS); | |
1755 | cpu_fprintf(f, "Max guest advance %"PRIi64" ms\n", max_advance/SCALE_MS); | |
1756 | } else { | |
1757 | cpu_fprintf(f, "Max guest delay NA\n"); | |
1758 | cpu_fprintf(f, "Max guest advance NA\n"); | |
1759 | } | |
1760 | } |