]>
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 | 27 | #include "qemu-common.h" |
8d4e9146 | 28 | #include "qemu/config-file.h" |
33c11879 | 29 | #include "cpu.h" |
83c9089e | 30 | #include "monitor/monitor.h" |
a4e15de9 | 31 | #include "qapi/qmp/qerror.h" |
d49b6836 | 32 | #include "qemu/error-report.h" |
9c17d615 | 33 | #include "sysemu/sysemu.h" |
da31d594 | 34 | #include "sysemu/block-backend.h" |
022c62cb | 35 | #include "exec/gdbstub.h" |
9c17d615 | 36 | #include "sysemu/dma.h" |
b3946626 | 37 | #include "sysemu/hw_accel.h" |
9c17d615 | 38 | #include "sysemu/kvm.h" |
b0cb0a66 | 39 | #include "sysemu/hax.h" |
de0b36b6 | 40 | #include "qmp-commands.h" |
63c91552 | 41 | #include "exec/exec-all.h" |
296af7c9 | 42 | |
1de7afc9 | 43 | #include "qemu/thread.h" |
9c17d615 PB |
44 | #include "sysemu/cpus.h" |
45 | #include "sysemu/qtest.h" | |
1de7afc9 PB |
46 | #include "qemu/main-loop.h" |
47 | #include "qemu/bitmap.h" | |
cb365646 | 48 | #include "qemu/seqlock.h" |
8d4e9146 | 49 | #include "tcg.h" |
a4e15de9 | 50 | #include "qapi-event.h" |
9cb805fd | 51 | #include "hw/nmi.h" |
8b427044 | 52 | #include "sysemu/replay.h" |
0ff0fc19 | 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 | ||
27498bef ST |
72 | int64_t max_delay; |
73 | int64_t max_advance; | |
296af7c9 | 74 | |
2adcc85d JH |
75 | /* vcpu throttling controls */ |
76 | static QEMUTimer *throttle_timer; | |
77 | static unsigned int throttle_percentage; | |
78 | ||
79 | #define CPU_THROTTLE_PCT_MIN 1 | |
80 | #define CPU_THROTTLE_PCT_MAX 99 | |
81 | #define CPU_THROTTLE_TIMESLICE_NS 10000000 | |
82 | ||
321bc0b2 TC |
83 | bool cpu_is_stopped(CPUState *cpu) |
84 | { | |
85 | return cpu->stopped || !runstate_is_running(); | |
86 | } | |
87 | ||
a98ae1d8 | 88 | static bool cpu_thread_is_idle(CPUState *cpu) |
ac873f1e | 89 | { |
c64ca814 | 90 | if (cpu->stop || cpu->queued_work_first) { |
ac873f1e PM |
91 | return false; |
92 | } | |
321bc0b2 | 93 | if (cpu_is_stopped(cpu)) { |
ac873f1e PM |
94 | return true; |
95 | } | |
8c2e1b00 | 96 | if (!cpu->halted || cpu_has_work(cpu) || |
215e79c0 | 97 | kvm_halt_in_kernel()) { |
ac873f1e PM |
98 | return false; |
99 | } | |
100 | return true; | |
101 | } | |
102 | ||
103 | static bool all_cpu_threads_idle(void) | |
104 | { | |
182735ef | 105 | CPUState *cpu; |
ac873f1e | 106 | |
bdc44640 | 107 | CPU_FOREACH(cpu) { |
182735ef | 108 | if (!cpu_thread_is_idle(cpu)) { |
ac873f1e PM |
109 | return false; |
110 | } | |
111 | } | |
112 | return true; | |
113 | } | |
114 | ||
946fb27c PB |
115 | /***********************************************************/ |
116 | /* guest cycle counter */ | |
117 | ||
a3270e19 PB |
118 | /* Protected by TimersState seqlock */ |
119 | ||
5045e9d9 | 120 | static bool icount_sleep = true; |
71468395 | 121 | static int64_t vm_clock_warp_start = -1; |
946fb27c PB |
122 | /* Conversion factor from emulated instructions to virtual clock ticks. */ |
123 | static int icount_time_shift; | |
124 | /* Arbitrarily pick 1MIPS as the minimum allowable speed. */ | |
125 | #define MAX_ICOUNT_SHIFT 10 | |
a3270e19 | 126 | |
946fb27c PB |
127 | static QEMUTimer *icount_rt_timer; |
128 | static QEMUTimer *icount_vm_timer; | |
129 | static QEMUTimer *icount_warp_timer; | |
946fb27c PB |
130 | |
131 | typedef struct TimersState { | |
cb365646 | 132 | /* Protected by BQL. */ |
946fb27c PB |
133 | int64_t cpu_ticks_prev; |
134 | int64_t cpu_ticks_offset; | |
cb365646 LPF |
135 | |
136 | /* cpu_clock_offset can be read out of BQL, so protect it with | |
137 | * this lock. | |
138 | */ | |
139 | QemuSeqLock vm_clock_seqlock; | |
946fb27c PB |
140 | int64_t cpu_clock_offset; |
141 | int32_t cpu_ticks_enabled; | |
142 | int64_t dummy; | |
c96778bb FK |
143 | |
144 | /* Compensate for varying guest execution speed. */ | |
145 | int64_t qemu_icount_bias; | |
146 | /* Only written by TCG thread */ | |
147 | int64_t qemu_icount; | |
946fb27c PB |
148 | } TimersState; |
149 | ||
d9cd4007 | 150 | static TimersState timers_state; |
8d4e9146 FK |
151 | bool mttcg_enabled; |
152 | ||
153 | /* | |
154 | * We default to false if we know other options have been enabled | |
155 | * which are currently incompatible with MTTCG. Otherwise when each | |
156 | * guest (target) has been updated to support: | |
157 | * - atomic instructions | |
158 | * - memory ordering primitives (barriers) | |
159 | * they can set the appropriate CONFIG flags in ${target}-softmmu.mak | |
160 | * | |
161 | * Once a guest architecture has been converted to the new primitives | |
162 | * there are two remaining limitations to check. | |
163 | * | |
164 | * - The guest can't be oversized (e.g. 64 bit guest on 32 bit host) | |
165 | * - The host must have a stronger memory order than the guest | |
166 | * | |
167 | * It may be possible in future to support strong guests on weak hosts | |
168 | * but that will require tagging all load/stores in a guest with their | |
169 | * implicit memory order requirements which would likely slow things | |
170 | * down a lot. | |
171 | */ | |
172 | ||
173 | static bool check_tcg_memory_orders_compatible(void) | |
174 | { | |
175 | #if defined(TCG_GUEST_DEFAULT_MO) && defined(TCG_TARGET_DEFAULT_MO) | |
176 | return (TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO) == 0; | |
177 | #else | |
178 | return false; | |
179 | #endif | |
180 | } | |
181 | ||
182 | static bool default_mttcg_enabled(void) | |
183 | { | |
83fd9629 | 184 | if (use_icount || TCG_OVERSIZED_GUEST) { |
8d4e9146 FK |
185 | return false; |
186 | } else { | |
187 | #ifdef TARGET_SUPPORTS_MTTCG | |
188 | return check_tcg_memory_orders_compatible(); | |
189 | #else | |
190 | return false; | |
191 | #endif | |
192 | } | |
193 | } | |
194 | ||
195 | void qemu_tcg_configure(QemuOpts *opts, Error **errp) | |
196 | { | |
197 | const char *t = qemu_opt_get(opts, "thread"); | |
198 | if (t) { | |
199 | if (strcmp(t, "multi") == 0) { | |
200 | if (TCG_OVERSIZED_GUEST) { | |
201 | error_setg(errp, "No MTTCG when guest word size > hosts"); | |
83fd9629 AB |
202 | } else if (use_icount) { |
203 | error_setg(errp, "No MTTCG when icount is enabled"); | |
8d4e9146 | 204 | } else { |
c34c7620 AB |
205 | #ifndef TARGET_SUPPORT_MTTCG |
206 | error_report("Guest not yet converted to MTTCG - " | |
207 | "you may get unexpected results"); | |
208 | #endif | |
8d4e9146 FK |
209 | if (!check_tcg_memory_orders_compatible()) { |
210 | error_report("Guest expects a stronger memory ordering " | |
211 | "than the host provides"); | |
212 | error_printf("This may cause strange/hard to debug errors"); | |
213 | } | |
214 | mttcg_enabled = true; | |
215 | } | |
216 | } else if (strcmp(t, "single") == 0) { | |
217 | mttcg_enabled = false; | |
218 | } else { | |
219 | error_setg(errp, "Invalid 'thread' setting %s", t); | |
220 | } | |
221 | } else { | |
222 | mttcg_enabled = default_mttcg_enabled(); | |
223 | } | |
224 | } | |
946fb27c | 225 | |
2a62914b | 226 | int64_t cpu_get_icount_raw(void) |
946fb27c PB |
227 | { |
228 | int64_t icount; | |
4917cf44 | 229 | CPUState *cpu = current_cpu; |
946fb27c | 230 | |
c96778bb | 231 | icount = timers_state.qemu_icount; |
4917cf44 | 232 | if (cpu) { |
414b15c9 | 233 | if (!cpu->can_do_io) { |
2a62914b PD |
234 | fprintf(stderr, "Bad icount read\n"); |
235 | exit(1); | |
946fb27c | 236 | } |
28ecfd7a | 237 | icount -= (cpu->icount_decr.u16.low + cpu->icount_extra); |
946fb27c | 238 | } |
2a62914b PD |
239 | return icount; |
240 | } | |
241 | ||
242 | /* Return the virtual CPU time, based on the instruction counter. */ | |
243 | static int64_t cpu_get_icount_locked(void) | |
244 | { | |
245 | int64_t icount = cpu_get_icount_raw(); | |
3f031313 | 246 | return timers_state.qemu_icount_bias + cpu_icount_to_ns(icount); |
946fb27c PB |
247 | } |
248 | ||
17a15f1b PB |
249 | int64_t cpu_get_icount(void) |
250 | { | |
251 | int64_t icount; | |
252 | unsigned start; | |
253 | ||
254 | do { | |
255 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
256 | icount = cpu_get_icount_locked(); | |
257 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
258 | ||
259 | return icount; | |
260 | } | |
261 | ||
3f031313 FK |
262 | int64_t cpu_icount_to_ns(int64_t icount) |
263 | { | |
264 | return icount << icount_time_shift; | |
265 | } | |
266 | ||
d90f3cca C |
267 | /* return the time elapsed in VM between vm_start and vm_stop. Unless |
268 | * icount is active, cpu_get_ticks() uses units of the host CPU cycle | |
269 | * counter. | |
270 | * | |
271 | * Caller must hold the BQL | |
272 | */ | |
946fb27c PB |
273 | int64_t cpu_get_ticks(void) |
274 | { | |
5f3e3101 PB |
275 | int64_t ticks; |
276 | ||
946fb27c PB |
277 | if (use_icount) { |
278 | return cpu_get_icount(); | |
279 | } | |
5f3e3101 PB |
280 | |
281 | ticks = timers_state.cpu_ticks_offset; | |
282 | if (timers_state.cpu_ticks_enabled) { | |
4a7428c5 | 283 | ticks += cpu_get_host_ticks(); |
5f3e3101 PB |
284 | } |
285 | ||
286 | if (timers_state.cpu_ticks_prev > ticks) { | |
287 | /* Note: non increasing ticks may happen if the host uses | |
288 | software suspend */ | |
289 | timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks; | |
290 | ticks = timers_state.cpu_ticks_prev; | |
946fb27c | 291 | } |
5f3e3101 PB |
292 | |
293 | timers_state.cpu_ticks_prev = ticks; | |
294 | return ticks; | |
946fb27c PB |
295 | } |
296 | ||
cb365646 | 297 | static int64_t cpu_get_clock_locked(void) |
946fb27c | 298 | { |
1d45cea5 | 299 | int64_t time; |
cb365646 | 300 | |
1d45cea5 | 301 | time = timers_state.cpu_clock_offset; |
5f3e3101 | 302 | if (timers_state.cpu_ticks_enabled) { |
1d45cea5 | 303 | time += get_clock(); |
946fb27c | 304 | } |
cb365646 | 305 | |
1d45cea5 | 306 | return time; |
cb365646 LPF |
307 | } |
308 | ||
d90f3cca | 309 | /* Return the monotonic time elapsed in VM, i.e., |
8212ff86 PM |
310 | * the time between vm_start and vm_stop |
311 | */ | |
cb365646 LPF |
312 | int64_t cpu_get_clock(void) |
313 | { | |
314 | int64_t ti; | |
315 | unsigned start; | |
316 | ||
317 | do { | |
318 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
319 | ti = cpu_get_clock_locked(); | |
320 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
321 | ||
322 | return ti; | |
946fb27c PB |
323 | } |
324 | ||
cb365646 | 325 | /* enable cpu_get_ticks() |
3224e878 | 326 | * Caller must hold BQL which serves as mutex for vm_clock_seqlock. |
cb365646 | 327 | */ |
946fb27c PB |
328 | void cpu_enable_ticks(void) |
329 | { | |
cb365646 | 330 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ |
03719e44 | 331 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
946fb27c | 332 | if (!timers_state.cpu_ticks_enabled) { |
4a7428c5 | 333 | timers_state.cpu_ticks_offset -= cpu_get_host_ticks(); |
946fb27c PB |
334 | timers_state.cpu_clock_offset -= get_clock(); |
335 | timers_state.cpu_ticks_enabled = 1; | |
336 | } | |
03719e44 | 337 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
946fb27c PB |
338 | } |
339 | ||
340 | /* disable cpu_get_ticks() : the clock is stopped. You must not call | |
cb365646 | 341 | * cpu_get_ticks() after that. |
3224e878 | 342 | * Caller must hold BQL which serves as mutex for vm_clock_seqlock. |
cb365646 | 343 | */ |
946fb27c PB |
344 | void cpu_disable_ticks(void) |
345 | { | |
cb365646 | 346 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ |
03719e44 | 347 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
946fb27c | 348 | if (timers_state.cpu_ticks_enabled) { |
4a7428c5 | 349 | timers_state.cpu_ticks_offset += cpu_get_host_ticks(); |
cb365646 | 350 | timers_state.cpu_clock_offset = cpu_get_clock_locked(); |
946fb27c PB |
351 | timers_state.cpu_ticks_enabled = 0; |
352 | } | |
03719e44 | 353 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
946fb27c PB |
354 | } |
355 | ||
356 | /* Correlation between real and virtual time is always going to be | |
357 | fairly approximate, so ignore small variation. | |
358 | When the guest is idle real and virtual time will be aligned in | |
359 | the IO wait loop. */ | |
73bcb24d | 360 | #define ICOUNT_WOBBLE (NANOSECONDS_PER_SECOND / 10) |
946fb27c PB |
361 | |
362 | static void icount_adjust(void) | |
363 | { | |
364 | int64_t cur_time; | |
365 | int64_t cur_icount; | |
366 | int64_t delta; | |
a3270e19 PB |
367 | |
368 | /* Protected by TimersState mutex. */ | |
946fb27c | 369 | static int64_t last_delta; |
468cc7cf | 370 | |
946fb27c PB |
371 | /* If the VM is not running, then do nothing. */ |
372 | if (!runstate_is_running()) { | |
373 | return; | |
374 | } | |
468cc7cf | 375 | |
03719e44 | 376 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
17a15f1b PB |
377 | cur_time = cpu_get_clock_locked(); |
378 | cur_icount = cpu_get_icount_locked(); | |
468cc7cf | 379 | |
946fb27c PB |
380 | delta = cur_icount - cur_time; |
381 | /* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */ | |
382 | if (delta > 0 | |
383 | && last_delta + ICOUNT_WOBBLE < delta * 2 | |
384 | && icount_time_shift > 0) { | |
385 | /* The guest is getting too far ahead. Slow time down. */ | |
386 | icount_time_shift--; | |
387 | } | |
388 | if (delta < 0 | |
389 | && last_delta - ICOUNT_WOBBLE > delta * 2 | |
390 | && icount_time_shift < MAX_ICOUNT_SHIFT) { | |
391 | /* The guest is getting too far behind. Speed time up. */ | |
392 | icount_time_shift++; | |
393 | } | |
394 | last_delta = delta; | |
c96778bb FK |
395 | timers_state.qemu_icount_bias = cur_icount |
396 | - (timers_state.qemu_icount << icount_time_shift); | |
03719e44 | 397 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
946fb27c PB |
398 | } |
399 | ||
400 | static void icount_adjust_rt(void *opaque) | |
401 | { | |
40daca54 | 402 | timer_mod(icount_rt_timer, |
1979b908 | 403 | qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000); |
946fb27c PB |
404 | icount_adjust(); |
405 | } | |
406 | ||
407 | static void icount_adjust_vm(void *opaque) | |
408 | { | |
40daca54 AB |
409 | timer_mod(icount_vm_timer, |
410 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
73bcb24d | 411 | NANOSECONDS_PER_SECOND / 10); |
946fb27c PB |
412 | icount_adjust(); |
413 | } | |
414 | ||
415 | static int64_t qemu_icount_round(int64_t count) | |
416 | { | |
417 | return (count + (1 << icount_time_shift) - 1) >> icount_time_shift; | |
418 | } | |
419 | ||
efab87cf | 420 | static void icount_warp_rt(void) |
946fb27c | 421 | { |
ccffff48 AB |
422 | unsigned seq; |
423 | int64_t warp_start; | |
424 | ||
17a15f1b PB |
425 | /* The icount_warp_timer is rescheduled soon after vm_clock_warp_start |
426 | * changes from -1 to another value, so the race here is okay. | |
427 | */ | |
ccffff48 AB |
428 | do { |
429 | seq = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
430 | warp_start = vm_clock_warp_start; | |
431 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, seq)); | |
432 | ||
433 | if (warp_start == -1) { | |
946fb27c PB |
434 | return; |
435 | } | |
436 | ||
03719e44 | 437 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
946fb27c | 438 | if (runstate_is_running()) { |
8eda206e PD |
439 | int64_t clock = REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT, |
440 | cpu_get_clock_locked()); | |
8ed961d9 PB |
441 | int64_t warp_delta; |
442 | ||
443 | warp_delta = clock - vm_clock_warp_start; | |
444 | if (use_icount == 2) { | |
946fb27c | 445 | /* |
40daca54 | 446 | * In adaptive mode, do not let QEMU_CLOCK_VIRTUAL run too |
946fb27c PB |
447 | * far ahead of real time. |
448 | */ | |
17a15f1b | 449 | int64_t cur_icount = cpu_get_icount_locked(); |
bf2a7ddb | 450 | int64_t delta = clock - cur_icount; |
8ed961d9 | 451 | warp_delta = MIN(warp_delta, delta); |
946fb27c | 452 | } |
c96778bb | 453 | timers_state.qemu_icount_bias += warp_delta; |
946fb27c PB |
454 | } |
455 | vm_clock_warp_start = -1; | |
03719e44 | 456 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
8ed961d9 PB |
457 | |
458 | if (qemu_clock_expired(QEMU_CLOCK_VIRTUAL)) { | |
459 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
460 | } | |
946fb27c PB |
461 | } |
462 | ||
e76d1798 | 463 | static void icount_timer_cb(void *opaque) |
efab87cf | 464 | { |
e76d1798 PD |
465 | /* No need for a checkpoint because the timer already synchronizes |
466 | * with CHECKPOINT_CLOCK_VIRTUAL_RT. | |
467 | */ | |
468 | icount_warp_rt(); | |
efab87cf PD |
469 | } |
470 | ||
8156be56 PB |
471 | void qtest_clock_warp(int64_t dest) |
472 | { | |
40daca54 | 473 | int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); |
efef88b3 | 474 | AioContext *aio_context; |
8156be56 | 475 | assert(qtest_enabled()); |
efef88b3 | 476 | aio_context = qemu_get_aio_context(); |
8156be56 | 477 | while (clock < dest) { |
40daca54 | 478 | int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); |
c9299e2f | 479 | int64_t warp = qemu_soonest_timeout(dest - clock, deadline); |
efef88b3 | 480 | |
03719e44 | 481 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
c96778bb | 482 | timers_state.qemu_icount_bias += warp; |
03719e44 | 483 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
17a15f1b | 484 | |
40daca54 | 485 | qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL); |
efef88b3 | 486 | timerlist_run_timers(aio_context->tlg.tl[QEMU_CLOCK_VIRTUAL]); |
40daca54 | 487 | clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); |
8156be56 | 488 | } |
40daca54 | 489 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
8156be56 PB |
490 | } |
491 | ||
e76d1798 | 492 | void qemu_start_warp_timer(void) |
946fb27c | 493 | { |
ce78d18c | 494 | int64_t clock; |
946fb27c PB |
495 | int64_t deadline; |
496 | ||
e76d1798 | 497 | if (!use_icount) { |
946fb27c PB |
498 | return; |
499 | } | |
500 | ||
8bd7f71d PD |
501 | /* Nothing to do if the VM is stopped: QEMU_CLOCK_VIRTUAL timers |
502 | * do not fire, so computing the deadline does not make sense. | |
503 | */ | |
504 | if (!runstate_is_running()) { | |
505 | return; | |
506 | } | |
507 | ||
508 | /* warp clock deterministically in record/replay mode */ | |
e76d1798 | 509 | if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_START)) { |
8bd7f71d PD |
510 | return; |
511 | } | |
512 | ||
ce78d18c | 513 | if (!all_cpu_threads_idle()) { |
946fb27c PB |
514 | return; |
515 | } | |
516 | ||
8156be56 PB |
517 | if (qtest_enabled()) { |
518 | /* When testing, qtest commands advance icount. */ | |
e76d1798 | 519 | return; |
8156be56 PB |
520 | } |
521 | ||
ac70aafc | 522 | /* We want to use the earliest deadline from ALL vm_clocks */ |
bf2a7ddb | 523 | clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT); |
40daca54 | 524 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); |
ce78d18c | 525 | if (deadline < 0) { |
d7a0f71d VC |
526 | static bool notified; |
527 | if (!icount_sleep && !notified) { | |
528 | error_report("WARNING: icount sleep disabled and no active timers"); | |
529 | notified = true; | |
530 | } | |
ce78d18c | 531 | return; |
ac70aafc AB |
532 | } |
533 | ||
946fb27c PB |
534 | if (deadline > 0) { |
535 | /* | |
40daca54 | 536 | * Ensure QEMU_CLOCK_VIRTUAL proceeds even when the virtual CPU goes to |
946fb27c PB |
537 | * sleep. Otherwise, the CPU might be waiting for a future timer |
538 | * interrupt to wake it up, but the interrupt never comes because | |
539 | * the vCPU isn't running any insns and thus doesn't advance the | |
40daca54 | 540 | * QEMU_CLOCK_VIRTUAL. |
946fb27c | 541 | */ |
5045e9d9 VC |
542 | if (!icount_sleep) { |
543 | /* | |
544 | * We never let VCPUs sleep in no sleep icount mode. | |
545 | * If there is a pending QEMU_CLOCK_VIRTUAL timer we just advance | |
546 | * to the next QEMU_CLOCK_VIRTUAL event and notify it. | |
547 | * It is useful when we want a deterministic execution time, | |
548 | * isolated from host latencies. | |
549 | */ | |
03719e44 | 550 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
5045e9d9 | 551 | timers_state.qemu_icount_bias += deadline; |
03719e44 | 552 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
5045e9d9 VC |
553 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
554 | } else { | |
555 | /* | |
556 | * We do stop VCPUs and only advance QEMU_CLOCK_VIRTUAL after some | |
557 | * "real" time, (related to the time left until the next event) has | |
558 | * passed. The QEMU_CLOCK_VIRTUAL_RT clock will do this. | |
559 | * This avoids that the warps are visible externally; for example, | |
560 | * you will not be sending network packets continuously instead of | |
561 | * every 100ms. | |
562 | */ | |
03719e44 | 563 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
5045e9d9 VC |
564 | if (vm_clock_warp_start == -1 || vm_clock_warp_start > clock) { |
565 | vm_clock_warp_start = clock; | |
566 | } | |
03719e44 | 567 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
5045e9d9 | 568 | timer_mod_anticipate(icount_warp_timer, clock + deadline); |
ce78d18c | 569 | } |
ac70aafc | 570 | } else if (deadline == 0) { |
40daca54 | 571 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
946fb27c PB |
572 | } |
573 | } | |
574 | ||
e76d1798 PD |
575 | static void qemu_account_warp_timer(void) |
576 | { | |
577 | if (!use_icount || !icount_sleep) { | |
578 | return; | |
579 | } | |
580 | ||
581 | /* Nothing to do if the VM is stopped: QEMU_CLOCK_VIRTUAL timers | |
582 | * do not fire, so computing the deadline does not make sense. | |
583 | */ | |
584 | if (!runstate_is_running()) { | |
585 | return; | |
586 | } | |
587 | ||
588 | /* warp clock deterministically in record/replay mode */ | |
589 | if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_ACCOUNT)) { | |
590 | return; | |
591 | } | |
592 | ||
593 | timer_del(icount_warp_timer); | |
594 | icount_warp_rt(); | |
595 | } | |
596 | ||
d09eae37 FK |
597 | static bool icount_state_needed(void *opaque) |
598 | { | |
599 | return use_icount; | |
600 | } | |
601 | ||
602 | /* | |
603 | * This is a subsection for icount migration. | |
604 | */ | |
605 | static const VMStateDescription icount_vmstate_timers = { | |
606 | .name = "timer/icount", | |
607 | .version_id = 1, | |
608 | .minimum_version_id = 1, | |
5cd8cada | 609 | .needed = icount_state_needed, |
d09eae37 FK |
610 | .fields = (VMStateField[]) { |
611 | VMSTATE_INT64(qemu_icount_bias, TimersState), | |
612 | VMSTATE_INT64(qemu_icount, TimersState), | |
613 | VMSTATE_END_OF_LIST() | |
614 | } | |
615 | }; | |
616 | ||
946fb27c PB |
617 | static const VMStateDescription vmstate_timers = { |
618 | .name = "timer", | |
619 | .version_id = 2, | |
620 | .minimum_version_id = 1, | |
35d08458 | 621 | .fields = (VMStateField[]) { |
946fb27c PB |
622 | VMSTATE_INT64(cpu_ticks_offset, TimersState), |
623 | VMSTATE_INT64(dummy, TimersState), | |
624 | VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2), | |
625 | VMSTATE_END_OF_LIST() | |
d09eae37 | 626 | }, |
5cd8cada JQ |
627 | .subsections = (const VMStateDescription*[]) { |
628 | &icount_vmstate_timers, | |
629 | NULL | |
946fb27c PB |
630 | } |
631 | }; | |
632 | ||
14e6fe12 | 633 | static void cpu_throttle_thread(CPUState *cpu, run_on_cpu_data opaque) |
2adcc85d | 634 | { |
2adcc85d JH |
635 | double pct; |
636 | double throttle_ratio; | |
637 | long sleeptime_ns; | |
638 | ||
639 | if (!cpu_throttle_get_percentage()) { | |
640 | return; | |
641 | } | |
642 | ||
643 | pct = (double)cpu_throttle_get_percentage()/100; | |
644 | throttle_ratio = pct / (1 - pct); | |
645 | sleeptime_ns = (long)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS); | |
646 | ||
647 | qemu_mutex_unlock_iothread(); | |
648 | atomic_set(&cpu->throttle_thread_scheduled, 0); | |
649 | g_usleep(sleeptime_ns / 1000); /* Convert ns to us for usleep call */ | |
650 | qemu_mutex_lock_iothread(); | |
651 | } | |
652 | ||
653 | static void cpu_throttle_timer_tick(void *opaque) | |
654 | { | |
655 | CPUState *cpu; | |
656 | double pct; | |
657 | ||
658 | /* Stop the timer if needed */ | |
659 | if (!cpu_throttle_get_percentage()) { | |
660 | return; | |
661 | } | |
662 | CPU_FOREACH(cpu) { | |
663 | if (!atomic_xchg(&cpu->throttle_thread_scheduled, 1)) { | |
14e6fe12 PB |
664 | async_run_on_cpu(cpu, cpu_throttle_thread, |
665 | RUN_ON_CPU_NULL); | |
2adcc85d JH |
666 | } |
667 | } | |
668 | ||
669 | pct = (double)cpu_throttle_get_percentage()/100; | |
670 | timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) + | |
671 | CPU_THROTTLE_TIMESLICE_NS / (1-pct)); | |
672 | } | |
673 | ||
674 | void cpu_throttle_set(int new_throttle_pct) | |
675 | { | |
676 | /* Ensure throttle percentage is within valid range */ | |
677 | new_throttle_pct = MIN(new_throttle_pct, CPU_THROTTLE_PCT_MAX); | |
678 | new_throttle_pct = MAX(new_throttle_pct, CPU_THROTTLE_PCT_MIN); | |
679 | ||
680 | atomic_set(&throttle_percentage, new_throttle_pct); | |
681 | ||
682 | timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) + | |
683 | CPU_THROTTLE_TIMESLICE_NS); | |
684 | } | |
685 | ||
686 | void cpu_throttle_stop(void) | |
687 | { | |
688 | atomic_set(&throttle_percentage, 0); | |
689 | } | |
690 | ||
691 | bool cpu_throttle_active(void) | |
692 | { | |
693 | return (cpu_throttle_get_percentage() != 0); | |
694 | } | |
695 | ||
696 | int cpu_throttle_get_percentage(void) | |
697 | { | |
698 | return atomic_read(&throttle_percentage); | |
699 | } | |
700 | ||
4603ea01 PD |
701 | void cpu_ticks_init(void) |
702 | { | |
ccdb3c1f | 703 | seqlock_init(&timers_state.vm_clock_seqlock); |
4603ea01 | 704 | vmstate_register(NULL, 0, &vmstate_timers, &timers_state); |
2adcc85d JH |
705 | throttle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT, |
706 | cpu_throttle_timer_tick, NULL); | |
4603ea01 PD |
707 | } |
708 | ||
1ad9580b | 709 | void configure_icount(QemuOpts *opts, Error **errp) |
946fb27c | 710 | { |
1ad9580b | 711 | const char *option; |
a8bfac37 | 712 | char *rem_str = NULL; |
1ad9580b | 713 | |
1ad9580b | 714 | option = qemu_opt_get(opts, "shift"); |
946fb27c | 715 | if (!option) { |
a8bfac37 ST |
716 | if (qemu_opt_get(opts, "align") != NULL) { |
717 | error_setg(errp, "Please specify shift option when using align"); | |
718 | } | |
946fb27c PB |
719 | return; |
720 | } | |
f1f4b57e VC |
721 | |
722 | icount_sleep = qemu_opt_get_bool(opts, "sleep", true); | |
5045e9d9 VC |
723 | if (icount_sleep) { |
724 | icount_warp_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT, | |
e76d1798 | 725 | icount_timer_cb, NULL); |
5045e9d9 | 726 | } |
f1f4b57e | 727 | |
a8bfac37 | 728 | icount_align_option = qemu_opt_get_bool(opts, "align", false); |
f1f4b57e VC |
729 | |
730 | if (icount_align_option && !icount_sleep) { | |
778d9f9b | 731 | error_setg(errp, "align=on and sleep=off are incompatible"); |
f1f4b57e | 732 | } |
946fb27c | 733 | if (strcmp(option, "auto") != 0) { |
a8bfac37 ST |
734 | errno = 0; |
735 | icount_time_shift = strtol(option, &rem_str, 0); | |
736 | if (errno != 0 || *rem_str != '\0' || !strlen(option)) { | |
737 | error_setg(errp, "icount: Invalid shift value"); | |
738 | } | |
946fb27c PB |
739 | use_icount = 1; |
740 | return; | |
a8bfac37 ST |
741 | } else if (icount_align_option) { |
742 | error_setg(errp, "shift=auto and align=on are incompatible"); | |
f1f4b57e | 743 | } else if (!icount_sleep) { |
778d9f9b | 744 | error_setg(errp, "shift=auto and sleep=off are incompatible"); |
946fb27c PB |
745 | } |
746 | ||
747 | use_icount = 2; | |
748 | ||
749 | /* 125MIPS seems a reasonable initial guess at the guest speed. | |
750 | It will be corrected fairly quickly anyway. */ | |
751 | icount_time_shift = 3; | |
752 | ||
753 | /* Have both realtime and virtual time triggers for speed adjustment. | |
754 | The realtime trigger catches emulated time passing too slowly, | |
755 | the virtual time trigger catches emulated time passing too fast. | |
756 | Realtime triggers occur even when idle, so use them less frequently | |
757 | than VM triggers. */ | |
bf2a7ddb PD |
758 | icount_rt_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL_RT, |
759 | icount_adjust_rt, NULL); | |
40daca54 | 760 | timer_mod(icount_rt_timer, |
bf2a7ddb | 761 | qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000); |
40daca54 AB |
762 | icount_vm_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, |
763 | icount_adjust_vm, NULL); | |
764 | timer_mod(icount_vm_timer, | |
765 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
73bcb24d | 766 | NANOSECONDS_PER_SECOND / 10); |
946fb27c PB |
767 | } |
768 | ||
6546706d AB |
769 | /***********************************************************/ |
770 | /* TCG vCPU kick timer | |
771 | * | |
772 | * The kick timer is responsible for moving single threaded vCPU | |
773 | * emulation on to the next vCPU. If more than one vCPU is running a | |
774 | * timer event with force a cpu->exit so the next vCPU can get | |
775 | * scheduled. | |
776 | * | |
777 | * The timer is removed if all vCPUs are idle and restarted again once | |
778 | * idleness is complete. | |
779 | */ | |
780 | ||
781 | static QEMUTimer *tcg_kick_vcpu_timer; | |
791158d9 | 782 | static CPUState *tcg_current_rr_cpu; |
6546706d AB |
783 | |
784 | #define TCG_KICK_PERIOD (NANOSECONDS_PER_SECOND / 10) | |
785 | ||
786 | static inline int64_t qemu_tcg_next_kick(void) | |
787 | { | |
788 | return qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + TCG_KICK_PERIOD; | |
789 | } | |
790 | ||
791158d9 AB |
791 | /* Kick the currently round-robin scheduled vCPU */ |
792 | static void qemu_cpu_kick_rr_cpu(void) | |
793 | { | |
794 | CPUState *cpu; | |
791158d9 AB |
795 | do { |
796 | cpu = atomic_mb_read(&tcg_current_rr_cpu); | |
797 | if (cpu) { | |
798 | cpu_exit(cpu); | |
799 | } | |
800 | } while (cpu != atomic_mb_read(&tcg_current_rr_cpu)); | |
801 | } | |
802 | ||
6546706d AB |
803 | static void kick_tcg_thread(void *opaque) |
804 | { | |
805 | timer_mod(tcg_kick_vcpu_timer, qemu_tcg_next_kick()); | |
791158d9 | 806 | qemu_cpu_kick_rr_cpu(); |
6546706d AB |
807 | } |
808 | ||
809 | static void start_tcg_kick_timer(void) | |
810 | { | |
37257942 | 811 | if (!mttcg_enabled && !tcg_kick_vcpu_timer && CPU_NEXT(first_cpu)) { |
6546706d AB |
812 | tcg_kick_vcpu_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, |
813 | kick_tcg_thread, NULL); | |
814 | timer_mod(tcg_kick_vcpu_timer, qemu_tcg_next_kick()); | |
815 | } | |
816 | } | |
817 | ||
818 | static void stop_tcg_kick_timer(void) | |
819 | { | |
820 | if (tcg_kick_vcpu_timer) { | |
821 | timer_del(tcg_kick_vcpu_timer); | |
822 | tcg_kick_vcpu_timer = NULL; | |
823 | } | |
824 | } | |
825 | ||
296af7c9 BS |
826 | /***********************************************************/ |
827 | void hw_error(const char *fmt, ...) | |
828 | { | |
829 | va_list ap; | |
55e5c285 | 830 | CPUState *cpu; |
296af7c9 BS |
831 | |
832 | va_start(ap, fmt); | |
833 | fprintf(stderr, "qemu: hardware error: "); | |
834 | vfprintf(stderr, fmt, ap); | |
835 | fprintf(stderr, "\n"); | |
bdc44640 | 836 | CPU_FOREACH(cpu) { |
55e5c285 | 837 | fprintf(stderr, "CPU #%d:\n", cpu->cpu_index); |
878096ee | 838 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU); |
296af7c9 BS |
839 | } |
840 | va_end(ap); | |
841 | abort(); | |
842 | } | |
843 | ||
844 | void cpu_synchronize_all_states(void) | |
845 | { | |
182735ef | 846 | CPUState *cpu; |
296af7c9 | 847 | |
bdc44640 | 848 | CPU_FOREACH(cpu) { |
182735ef | 849 | cpu_synchronize_state(cpu); |
296af7c9 BS |
850 | } |
851 | } | |
852 | ||
853 | void cpu_synchronize_all_post_reset(void) | |
854 | { | |
182735ef | 855 | CPUState *cpu; |
296af7c9 | 856 | |
bdc44640 | 857 | CPU_FOREACH(cpu) { |
182735ef | 858 | cpu_synchronize_post_reset(cpu); |
296af7c9 BS |
859 | } |
860 | } | |
861 | ||
862 | void cpu_synchronize_all_post_init(void) | |
863 | { | |
182735ef | 864 | CPUState *cpu; |
296af7c9 | 865 | |
bdc44640 | 866 | CPU_FOREACH(cpu) { |
182735ef | 867 | cpu_synchronize_post_init(cpu); |
296af7c9 BS |
868 | } |
869 | } | |
870 | ||
56983463 | 871 | static int do_vm_stop(RunState state) |
296af7c9 | 872 | { |
56983463 KW |
873 | int ret = 0; |
874 | ||
1354869c | 875 | if (runstate_is_running()) { |
296af7c9 | 876 | cpu_disable_ticks(); |
296af7c9 | 877 | pause_all_vcpus(); |
f5bbfba1 | 878 | runstate_set(state); |
1dfb4dd9 | 879 | vm_state_notify(0, state); |
a4e15de9 | 880 | qapi_event_send_stop(&error_abort); |
296af7c9 | 881 | } |
56983463 | 882 | |
594a45ce | 883 | bdrv_drain_all(); |
6d0ceb80 | 884 | replay_disable_events(); |
22af08ea | 885 | ret = bdrv_flush_all(); |
594a45ce | 886 | |
56983463 | 887 | return ret; |
296af7c9 BS |
888 | } |
889 | ||
a1fcaa73 | 890 | static bool cpu_can_run(CPUState *cpu) |
296af7c9 | 891 | { |
4fdeee7c | 892 | if (cpu->stop) { |
a1fcaa73 | 893 | return false; |
0ab07c62 | 894 | } |
321bc0b2 | 895 | if (cpu_is_stopped(cpu)) { |
a1fcaa73 | 896 | return false; |
0ab07c62 | 897 | } |
a1fcaa73 | 898 | return true; |
296af7c9 BS |
899 | } |
900 | ||
91325046 | 901 | static void cpu_handle_guest_debug(CPUState *cpu) |
83f338f7 | 902 | { |
64f6b346 | 903 | gdb_set_stop_cpu(cpu); |
8cf71710 | 904 | qemu_system_debug_request(); |
f324e766 | 905 | cpu->stopped = true; |
3c638d06 JK |
906 | } |
907 | ||
6d9cb73c JK |
908 | #ifdef CONFIG_LINUX |
909 | static void sigbus_reraise(void) | |
910 | { | |
911 | sigset_t set; | |
912 | struct sigaction action; | |
913 | ||
914 | memset(&action, 0, sizeof(action)); | |
915 | action.sa_handler = SIG_DFL; | |
916 | if (!sigaction(SIGBUS, &action, NULL)) { | |
917 | raise(SIGBUS); | |
918 | sigemptyset(&set); | |
919 | sigaddset(&set, SIGBUS); | |
a2d1761d | 920 | pthread_sigmask(SIG_UNBLOCK, &set, NULL); |
6d9cb73c JK |
921 | } |
922 | perror("Failed to re-raise SIGBUS!\n"); | |
923 | abort(); | |
924 | } | |
925 | ||
d98d4072 | 926 | static void sigbus_handler(int n, siginfo_t *siginfo, void *ctx) |
6d9cb73c | 927 | { |
a16fc07e PB |
928 | if (siginfo->si_code != BUS_MCEERR_AO && siginfo->si_code != BUS_MCEERR_AR) { |
929 | sigbus_reraise(); | |
930 | } | |
931 | ||
2ae41db2 PB |
932 | if (current_cpu) { |
933 | /* Called asynchronously in VCPU thread. */ | |
934 | if (kvm_on_sigbus_vcpu(current_cpu, siginfo->si_code, siginfo->si_addr)) { | |
935 | sigbus_reraise(); | |
936 | } | |
937 | } else { | |
938 | /* Called synchronously (via signalfd) in main thread. */ | |
939 | if (kvm_on_sigbus(siginfo->si_code, siginfo->si_addr)) { | |
940 | sigbus_reraise(); | |
941 | } | |
6d9cb73c JK |
942 | } |
943 | } | |
944 | ||
945 | static void qemu_init_sigbus(void) | |
946 | { | |
947 | struct sigaction action; | |
948 | ||
949 | memset(&action, 0, sizeof(action)); | |
950 | action.sa_flags = SA_SIGINFO; | |
d98d4072 | 951 | action.sa_sigaction = sigbus_handler; |
6d9cb73c JK |
952 | sigaction(SIGBUS, &action, NULL); |
953 | ||
954 | prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0); | |
955 | } | |
6d9cb73c | 956 | #else /* !CONFIG_LINUX */ |
6d9cb73c JK |
957 | static void qemu_init_sigbus(void) |
958 | { | |
959 | } | |
a16fc07e | 960 | #endif /* !CONFIG_LINUX */ |
ff48eb5f | 961 | |
b2532d88 | 962 | static QemuMutex qemu_global_mutex; |
296af7c9 BS |
963 | |
964 | static QemuThread io_thread; | |
965 | ||
296af7c9 BS |
966 | /* cpu creation */ |
967 | static QemuCond qemu_cpu_cond; | |
968 | /* system init */ | |
296af7c9 BS |
969 | static QemuCond qemu_pause_cond; |
970 | ||
d3b12f5d | 971 | void qemu_init_cpu_loop(void) |
296af7c9 | 972 | { |
6d9cb73c | 973 | qemu_init_sigbus(); |
ed94592b | 974 | qemu_cond_init(&qemu_cpu_cond); |
ed94592b | 975 | qemu_cond_init(&qemu_pause_cond); |
296af7c9 | 976 | qemu_mutex_init(&qemu_global_mutex); |
296af7c9 | 977 | |
b7680cb6 | 978 | qemu_thread_get_self(&io_thread); |
296af7c9 BS |
979 | } |
980 | ||
14e6fe12 | 981 | void run_on_cpu(CPUState *cpu, run_on_cpu_func func, run_on_cpu_data data) |
e82bcec2 | 982 | { |
d148d90e | 983 | do_run_on_cpu(cpu, func, data, &qemu_global_mutex); |
3c02270d CV |
984 | } |
985 | ||
4c055ab5 GZ |
986 | static void qemu_kvm_destroy_vcpu(CPUState *cpu) |
987 | { | |
988 | if (kvm_destroy_vcpu(cpu) < 0) { | |
989 | error_report("kvm_destroy_vcpu failed"); | |
990 | exit(EXIT_FAILURE); | |
991 | } | |
992 | } | |
993 | ||
994 | static void qemu_tcg_destroy_vcpu(CPUState *cpu) | |
995 | { | |
996 | } | |
997 | ||
509a0d78 | 998 | static void qemu_wait_io_event_common(CPUState *cpu) |
296af7c9 | 999 | { |
37257942 | 1000 | atomic_mb_set(&cpu->thread_kicked, false); |
4fdeee7c AF |
1001 | if (cpu->stop) { |
1002 | cpu->stop = false; | |
f324e766 | 1003 | cpu->stopped = true; |
96bce683 | 1004 | qemu_cond_broadcast(&qemu_pause_cond); |
296af7c9 | 1005 | } |
a5403c69 | 1006 | process_queued_cpu_work(cpu); |
37257942 AB |
1007 | } |
1008 | ||
1009 | static bool qemu_tcg_should_sleep(CPUState *cpu) | |
1010 | { | |
1011 | if (mttcg_enabled) { | |
1012 | return cpu_thread_is_idle(cpu); | |
1013 | } else { | |
1014 | return all_cpu_threads_idle(); | |
1015 | } | |
296af7c9 BS |
1016 | } |
1017 | ||
d5f8d613 | 1018 | static void qemu_tcg_wait_io_event(CPUState *cpu) |
296af7c9 | 1019 | { |
37257942 | 1020 | while (qemu_tcg_should_sleep(cpu)) { |
6546706d | 1021 | stop_tcg_kick_timer(); |
d5f8d613 | 1022 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); |
16400322 | 1023 | } |
296af7c9 | 1024 | |
6546706d AB |
1025 | start_tcg_kick_timer(); |
1026 | ||
37257942 | 1027 | qemu_wait_io_event_common(cpu); |
296af7c9 BS |
1028 | } |
1029 | ||
fd529e8f | 1030 | static void qemu_kvm_wait_io_event(CPUState *cpu) |
296af7c9 | 1031 | { |
a98ae1d8 | 1032 | while (cpu_thread_is_idle(cpu)) { |
f5c121b8 | 1033 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); |
16400322 | 1034 | } |
296af7c9 | 1035 | |
509a0d78 | 1036 | qemu_wait_io_event_common(cpu); |
296af7c9 BS |
1037 | } |
1038 | ||
7e97cd88 | 1039 | static void *qemu_kvm_cpu_thread_fn(void *arg) |
296af7c9 | 1040 | { |
48a106bd | 1041 | CPUState *cpu = arg; |
84b4915d | 1042 | int r; |
296af7c9 | 1043 | |
ab28bd23 PB |
1044 | rcu_register_thread(); |
1045 | ||
2e7f7a3c | 1046 | qemu_mutex_lock_iothread(); |
814e612e | 1047 | qemu_thread_get_self(cpu->thread); |
9f09e18a | 1048 | cpu->thread_id = qemu_get_thread_id(); |
626cf8f4 | 1049 | cpu->can_do_io = 1; |
4917cf44 | 1050 | current_cpu = cpu; |
296af7c9 | 1051 | |
504134d2 | 1052 | r = kvm_init_vcpu(cpu); |
84b4915d JK |
1053 | if (r < 0) { |
1054 | fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r)); | |
1055 | exit(1); | |
1056 | } | |
296af7c9 | 1057 | |
18268b60 | 1058 | kvm_init_cpu_signals(cpu); |
296af7c9 BS |
1059 | |
1060 | /* signal CPU creation */ | |
61a46217 | 1061 | cpu->created = true; |
296af7c9 BS |
1062 | qemu_cond_signal(&qemu_cpu_cond); |
1063 | ||
4c055ab5 | 1064 | do { |
a1fcaa73 | 1065 | if (cpu_can_run(cpu)) { |
1458c363 | 1066 | r = kvm_cpu_exec(cpu); |
83f338f7 | 1067 | if (r == EXCP_DEBUG) { |
91325046 | 1068 | cpu_handle_guest_debug(cpu); |
83f338f7 | 1069 | } |
0ab07c62 | 1070 | } |
fd529e8f | 1071 | qemu_kvm_wait_io_event(cpu); |
4c055ab5 | 1072 | } while (!cpu->unplug || cpu_can_run(cpu)); |
296af7c9 | 1073 | |
4c055ab5 | 1074 | qemu_kvm_destroy_vcpu(cpu); |
2c579042 BR |
1075 | cpu->created = false; |
1076 | qemu_cond_signal(&qemu_cpu_cond); | |
4c055ab5 | 1077 | qemu_mutex_unlock_iothread(); |
296af7c9 BS |
1078 | return NULL; |
1079 | } | |
1080 | ||
c7f0f3b1 AL |
1081 | static void *qemu_dummy_cpu_thread_fn(void *arg) |
1082 | { | |
1083 | #ifdef _WIN32 | |
1084 | fprintf(stderr, "qtest is not supported under Windows\n"); | |
1085 | exit(1); | |
1086 | #else | |
10a9021d | 1087 | CPUState *cpu = arg; |
c7f0f3b1 AL |
1088 | sigset_t waitset; |
1089 | int r; | |
1090 | ||
ab28bd23 PB |
1091 | rcu_register_thread(); |
1092 | ||
c7f0f3b1 | 1093 | qemu_mutex_lock_iothread(); |
814e612e | 1094 | qemu_thread_get_self(cpu->thread); |
9f09e18a | 1095 | cpu->thread_id = qemu_get_thread_id(); |
626cf8f4 | 1096 | cpu->can_do_io = 1; |
37257942 | 1097 | current_cpu = cpu; |
c7f0f3b1 AL |
1098 | |
1099 | sigemptyset(&waitset); | |
1100 | sigaddset(&waitset, SIG_IPI); | |
1101 | ||
1102 | /* signal CPU creation */ | |
61a46217 | 1103 | cpu->created = true; |
c7f0f3b1 AL |
1104 | qemu_cond_signal(&qemu_cpu_cond); |
1105 | ||
c7f0f3b1 | 1106 | while (1) { |
c7f0f3b1 AL |
1107 | qemu_mutex_unlock_iothread(); |
1108 | do { | |
1109 | int sig; | |
1110 | r = sigwait(&waitset, &sig); | |
1111 | } while (r == -1 && (errno == EAGAIN || errno == EINTR)); | |
1112 | if (r == -1) { | |
1113 | perror("sigwait"); | |
1114 | exit(1); | |
1115 | } | |
1116 | qemu_mutex_lock_iothread(); | |
509a0d78 | 1117 | qemu_wait_io_event_common(cpu); |
c7f0f3b1 AL |
1118 | } |
1119 | ||
1120 | return NULL; | |
1121 | #endif | |
1122 | } | |
1123 | ||
1be7fcb8 AB |
1124 | static int64_t tcg_get_icount_limit(void) |
1125 | { | |
1126 | int64_t deadline; | |
1127 | ||
1128 | if (replay_mode != REPLAY_MODE_PLAY) { | |
1129 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
1130 | ||
1131 | /* Maintain prior (possibly buggy) behaviour where if no deadline | |
1132 | * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than | |
1133 | * INT32_MAX nanoseconds ahead, we still use INT32_MAX | |
1134 | * nanoseconds. | |
1135 | */ | |
1136 | if ((deadline < 0) || (deadline > INT32_MAX)) { | |
1137 | deadline = INT32_MAX; | |
1138 | } | |
1139 | ||
1140 | return qemu_icount_round(deadline); | |
1141 | } else { | |
1142 | return replay_get_instructions(); | |
1143 | } | |
1144 | } | |
1145 | ||
12e9700d AB |
1146 | static void handle_icount_deadline(void) |
1147 | { | |
1148 | if (use_icount) { | |
1149 | int64_t deadline = | |
1150 | qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
1151 | ||
1152 | if (deadline == 0) { | |
1153 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
1154 | } | |
1155 | } | |
1156 | } | |
1157 | ||
1be7fcb8 AB |
1158 | static int tcg_cpu_exec(CPUState *cpu) |
1159 | { | |
1160 | int ret; | |
1161 | #ifdef CONFIG_PROFILER | |
1162 | int64_t ti; | |
1163 | #endif | |
1164 | ||
1165 | #ifdef CONFIG_PROFILER | |
1166 | ti = profile_getclock(); | |
1167 | #endif | |
1168 | if (use_icount) { | |
1169 | int64_t count; | |
1170 | int decr; | |
1171 | timers_state.qemu_icount -= (cpu->icount_decr.u16.low | |
1172 | + cpu->icount_extra); | |
1173 | cpu->icount_decr.u16.low = 0; | |
1174 | cpu->icount_extra = 0; | |
1175 | count = tcg_get_icount_limit(); | |
1176 | timers_state.qemu_icount += count; | |
1177 | decr = (count > 0xffff) ? 0xffff : count; | |
1178 | count -= decr; | |
1179 | cpu->icount_decr.u16.low = decr; | |
1180 | cpu->icount_extra = count; | |
1181 | } | |
8d04fb55 | 1182 | qemu_mutex_unlock_iothread(); |
1be7fcb8 AB |
1183 | cpu_exec_start(cpu); |
1184 | ret = cpu_exec(cpu); | |
1185 | cpu_exec_end(cpu); | |
8d04fb55 | 1186 | qemu_mutex_lock_iothread(); |
1be7fcb8 AB |
1187 | #ifdef CONFIG_PROFILER |
1188 | tcg_time += profile_getclock() - ti; | |
1189 | #endif | |
1190 | if (use_icount) { | |
1191 | /* Fold pending instructions back into the | |
1192 | instruction counter, and clear the interrupt flag. */ | |
1193 | timers_state.qemu_icount -= (cpu->icount_decr.u16.low | |
1194 | + cpu->icount_extra); | |
1195 | cpu->icount_decr.u32 = 0; | |
1196 | cpu->icount_extra = 0; | |
1197 | replay_account_executed_instructions(); | |
1198 | } | |
1199 | return ret; | |
1200 | } | |
1201 | ||
c93bbbef AB |
1202 | /* Destroy any remaining vCPUs which have been unplugged and have |
1203 | * finished running | |
1204 | */ | |
1205 | static void deal_with_unplugged_cpus(void) | |
1be7fcb8 | 1206 | { |
c93bbbef | 1207 | CPUState *cpu; |
1be7fcb8 | 1208 | |
c93bbbef AB |
1209 | CPU_FOREACH(cpu) { |
1210 | if (cpu->unplug && !cpu_can_run(cpu)) { | |
1211 | qemu_tcg_destroy_vcpu(cpu); | |
1212 | cpu->created = false; | |
1213 | qemu_cond_signal(&qemu_cpu_cond); | |
1be7fcb8 AB |
1214 | break; |
1215 | } | |
1216 | } | |
1be7fcb8 | 1217 | } |
bdb7ca67 | 1218 | |
6546706d AB |
1219 | /* Single-threaded TCG |
1220 | * | |
1221 | * In the single-threaded case each vCPU is simulated in turn. If | |
1222 | * there is more than a single vCPU we create a simple timer to kick | |
1223 | * the vCPU and ensure we don't get stuck in a tight loop in one vCPU. | |
1224 | * This is done explicitly rather than relying on side-effects | |
1225 | * elsewhere. | |
1226 | */ | |
1227 | ||
37257942 | 1228 | static void *qemu_tcg_rr_cpu_thread_fn(void *arg) |
296af7c9 | 1229 | { |
c3586ba7 | 1230 | CPUState *cpu = arg; |
296af7c9 | 1231 | |
ab28bd23 PB |
1232 | rcu_register_thread(); |
1233 | ||
2e7f7a3c | 1234 | qemu_mutex_lock_iothread(); |
814e612e | 1235 | qemu_thread_get_self(cpu->thread); |
296af7c9 | 1236 | |
38fcbd3f AF |
1237 | CPU_FOREACH(cpu) { |
1238 | cpu->thread_id = qemu_get_thread_id(); | |
1239 | cpu->created = true; | |
626cf8f4 | 1240 | cpu->can_do_io = 1; |
38fcbd3f | 1241 | } |
296af7c9 BS |
1242 | qemu_cond_signal(&qemu_cpu_cond); |
1243 | ||
fa7d1867 | 1244 | /* wait for initial kick-off after machine start */ |
c28e399c | 1245 | while (first_cpu->stopped) { |
d5f8d613 | 1246 | qemu_cond_wait(first_cpu->halt_cond, &qemu_global_mutex); |
8e564b4e JK |
1247 | |
1248 | /* process any pending work */ | |
bdc44640 | 1249 | CPU_FOREACH(cpu) { |
37257942 | 1250 | current_cpu = cpu; |
182735ef | 1251 | qemu_wait_io_event_common(cpu); |
8e564b4e | 1252 | } |
0ab07c62 | 1253 | } |
296af7c9 | 1254 | |
6546706d AB |
1255 | start_tcg_kick_timer(); |
1256 | ||
c93bbbef AB |
1257 | cpu = first_cpu; |
1258 | ||
e5143e30 AB |
1259 | /* process any pending work */ |
1260 | cpu->exit_request = 1; | |
1261 | ||
296af7c9 | 1262 | while (1) { |
c93bbbef AB |
1263 | /* Account partial waits to QEMU_CLOCK_VIRTUAL. */ |
1264 | qemu_account_warp_timer(); | |
1265 | ||
1266 | if (!cpu) { | |
1267 | cpu = first_cpu; | |
1268 | } | |
1269 | ||
e5143e30 AB |
1270 | while (cpu && !cpu->queued_work_first && !cpu->exit_request) { |
1271 | ||
791158d9 | 1272 | atomic_mb_set(&tcg_current_rr_cpu, cpu); |
37257942 | 1273 | current_cpu = cpu; |
c93bbbef AB |
1274 | |
1275 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, | |
1276 | (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0); | |
1277 | ||
1278 | if (cpu_can_run(cpu)) { | |
1279 | int r; | |
1280 | r = tcg_cpu_exec(cpu); | |
1281 | if (r == EXCP_DEBUG) { | |
1282 | cpu_handle_guest_debug(cpu); | |
1283 | break; | |
08e73c48 PK |
1284 | } else if (r == EXCP_ATOMIC) { |
1285 | qemu_mutex_unlock_iothread(); | |
1286 | cpu_exec_step_atomic(cpu); | |
1287 | qemu_mutex_lock_iothread(); | |
1288 | break; | |
c93bbbef | 1289 | } |
37257942 | 1290 | } else if (cpu->stop) { |
c93bbbef AB |
1291 | if (cpu->unplug) { |
1292 | cpu = CPU_NEXT(cpu); | |
1293 | } | |
1294 | break; | |
1295 | } | |
1296 | ||
e5143e30 AB |
1297 | cpu = CPU_NEXT(cpu); |
1298 | } /* while (cpu && !cpu->exit_request).. */ | |
1299 | ||
791158d9 AB |
1300 | /* Does not need atomic_mb_set because a spurious wakeup is okay. */ |
1301 | atomic_set(&tcg_current_rr_cpu, NULL); | |
c93bbbef | 1302 | |
e5143e30 AB |
1303 | if (cpu && cpu->exit_request) { |
1304 | atomic_mb_set(&cpu->exit_request, 0); | |
1305 | } | |
ac70aafc | 1306 | |
12e9700d | 1307 | handle_icount_deadline(); |
ac70aafc | 1308 | |
37257942 | 1309 | qemu_tcg_wait_io_event(cpu ? cpu : QTAILQ_FIRST(&cpus)); |
c93bbbef | 1310 | deal_with_unplugged_cpus(); |
296af7c9 BS |
1311 | } |
1312 | ||
1313 | return NULL; | |
1314 | } | |
1315 | ||
b0cb0a66 VP |
1316 | static void *qemu_hax_cpu_thread_fn(void *arg) |
1317 | { | |
1318 | CPUState *cpu = arg; | |
1319 | int r; | |
1320 | qemu_thread_get_self(cpu->thread); | |
1321 | qemu_mutex_lock(&qemu_global_mutex); | |
1322 | ||
1323 | cpu->thread_id = qemu_get_thread_id(); | |
1324 | cpu->created = true; | |
1325 | cpu->halted = 0; | |
1326 | current_cpu = cpu; | |
1327 | ||
1328 | hax_init_vcpu(cpu); | |
1329 | qemu_cond_signal(&qemu_cpu_cond); | |
1330 | ||
1331 | while (1) { | |
1332 | if (cpu_can_run(cpu)) { | |
1333 | r = hax_smp_cpu_exec(cpu); | |
1334 | if (r == EXCP_DEBUG) { | |
1335 | cpu_handle_guest_debug(cpu); | |
1336 | } | |
1337 | } | |
1338 | ||
1339 | while (cpu_thread_is_idle(cpu)) { | |
1340 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); | |
1341 | } | |
1342 | #ifdef _WIN32 | |
1343 | SleepEx(0, TRUE); | |
1344 | #endif | |
1345 | qemu_wait_io_event_common(cpu); | |
1346 | } | |
1347 | return NULL; | |
1348 | } | |
1349 | ||
1350 | #ifdef _WIN32 | |
1351 | static void CALLBACK dummy_apc_func(ULONG_PTR unused) | |
1352 | { | |
1353 | } | |
1354 | #endif | |
1355 | ||
37257942 AB |
1356 | /* Multi-threaded TCG |
1357 | * | |
1358 | * In the multi-threaded case each vCPU has its own thread. The TLS | |
1359 | * variable current_cpu can be used deep in the code to find the | |
1360 | * current CPUState for a given thread. | |
1361 | */ | |
1362 | ||
1363 | static void *qemu_tcg_cpu_thread_fn(void *arg) | |
1364 | { | |
1365 | CPUState *cpu = arg; | |
1366 | ||
1367 | rcu_register_thread(); | |
1368 | ||
1369 | qemu_mutex_lock_iothread(); | |
1370 | qemu_thread_get_self(cpu->thread); | |
1371 | ||
1372 | cpu->thread_id = qemu_get_thread_id(); | |
1373 | cpu->created = true; | |
1374 | cpu->can_do_io = 1; | |
1375 | current_cpu = cpu; | |
1376 | qemu_cond_signal(&qemu_cpu_cond); | |
1377 | ||
1378 | /* process any pending work */ | |
1379 | cpu->exit_request = 1; | |
1380 | ||
1381 | while (1) { | |
1382 | if (cpu_can_run(cpu)) { | |
1383 | int r; | |
1384 | r = tcg_cpu_exec(cpu); | |
1385 | switch (r) { | |
1386 | case EXCP_DEBUG: | |
1387 | cpu_handle_guest_debug(cpu); | |
1388 | break; | |
1389 | case EXCP_HALTED: | |
1390 | /* during start-up the vCPU is reset and the thread is | |
1391 | * kicked several times. If we don't ensure we go back | |
1392 | * to sleep in the halted state we won't cleanly | |
1393 | * start-up when the vCPU is enabled. | |
1394 | * | |
1395 | * cpu->halted should ensure we sleep in wait_io_event | |
1396 | */ | |
1397 | g_assert(cpu->halted); | |
1398 | break; | |
08e73c48 PK |
1399 | case EXCP_ATOMIC: |
1400 | qemu_mutex_unlock_iothread(); | |
1401 | cpu_exec_step_atomic(cpu); | |
1402 | qemu_mutex_lock_iothread(); | |
37257942 AB |
1403 | default: |
1404 | /* Ignore everything else? */ | |
1405 | break; | |
1406 | } | |
1407 | } | |
1408 | ||
1409 | handle_icount_deadline(); | |
1410 | ||
1411 | atomic_mb_set(&cpu->exit_request, 0); | |
1412 | qemu_tcg_wait_io_event(cpu); | |
1413 | } | |
1414 | ||
1415 | return NULL; | |
1416 | } | |
1417 | ||
2ff09a40 | 1418 | static void qemu_cpu_kick_thread(CPUState *cpu) |
cc015e9a PB |
1419 | { |
1420 | #ifndef _WIN32 | |
1421 | int err; | |
1422 | ||
e0c38211 PB |
1423 | if (cpu->thread_kicked) { |
1424 | return; | |
9102deda | 1425 | } |
e0c38211 | 1426 | cpu->thread_kicked = true; |
814e612e | 1427 | err = pthread_kill(cpu->thread->thread, SIG_IPI); |
cc015e9a PB |
1428 | if (err) { |
1429 | fprintf(stderr, "qemu:%s: %s", __func__, strerror(err)); | |
1430 | exit(1); | |
1431 | } | |
1432 | #else /* _WIN32 */ | |
b0cb0a66 VP |
1433 | if (!qemu_cpu_is_self(cpu)) { |
1434 | if (!QueueUserAPC(dummy_apc_func, cpu->hThread, 0)) { | |
1435 | fprintf(stderr, "%s: QueueUserAPC failed with error %lu\n", | |
1436 | __func__, GetLastError()); | |
1437 | exit(1); | |
1438 | } | |
1439 | } | |
e0c38211 PB |
1440 | #endif |
1441 | } | |
ed9164a3 | 1442 | |
c08d7424 | 1443 | void qemu_cpu_kick(CPUState *cpu) |
296af7c9 | 1444 | { |
f5c121b8 | 1445 | qemu_cond_broadcast(cpu->halt_cond); |
e0c38211 | 1446 | if (tcg_enabled()) { |
791158d9 | 1447 | cpu_exit(cpu); |
37257942 | 1448 | /* NOP unless doing single-thread RR */ |
791158d9 | 1449 | qemu_cpu_kick_rr_cpu(); |
e0c38211 | 1450 | } else { |
b0cb0a66 VP |
1451 | if (hax_enabled()) { |
1452 | /* | |
1453 | * FIXME: race condition with the exit_request check in | |
1454 | * hax_vcpu_hax_exec | |
1455 | */ | |
1456 | cpu->exit_request = 1; | |
1457 | } | |
e0c38211 PB |
1458 | qemu_cpu_kick_thread(cpu); |
1459 | } | |
296af7c9 BS |
1460 | } |
1461 | ||
46d62fac | 1462 | void qemu_cpu_kick_self(void) |
296af7c9 | 1463 | { |
4917cf44 | 1464 | assert(current_cpu); |
9102deda | 1465 | qemu_cpu_kick_thread(current_cpu); |
296af7c9 BS |
1466 | } |
1467 | ||
60e82579 | 1468 | bool qemu_cpu_is_self(CPUState *cpu) |
296af7c9 | 1469 | { |
814e612e | 1470 | return qemu_thread_is_self(cpu->thread); |
296af7c9 BS |
1471 | } |
1472 | ||
79e2b9ae | 1473 | bool qemu_in_vcpu_thread(void) |
aa723c23 | 1474 | { |
4917cf44 | 1475 | return current_cpu && qemu_cpu_is_self(current_cpu); |
aa723c23 JQ |
1476 | } |
1477 | ||
afbe7053 PB |
1478 | static __thread bool iothread_locked = false; |
1479 | ||
1480 | bool qemu_mutex_iothread_locked(void) | |
1481 | { | |
1482 | return iothread_locked; | |
1483 | } | |
1484 | ||
296af7c9 BS |
1485 | void qemu_mutex_lock_iothread(void) |
1486 | { | |
8d04fb55 JK |
1487 | g_assert(!qemu_mutex_iothread_locked()); |
1488 | qemu_mutex_lock(&qemu_global_mutex); | |
afbe7053 | 1489 | iothread_locked = true; |
296af7c9 BS |
1490 | } |
1491 | ||
1492 | void qemu_mutex_unlock_iothread(void) | |
1493 | { | |
8d04fb55 | 1494 | g_assert(qemu_mutex_iothread_locked()); |
afbe7053 | 1495 | iothread_locked = false; |
296af7c9 BS |
1496 | qemu_mutex_unlock(&qemu_global_mutex); |
1497 | } | |
1498 | ||
e8faee06 | 1499 | static bool all_vcpus_paused(void) |
296af7c9 | 1500 | { |
bdc44640 | 1501 | CPUState *cpu; |
296af7c9 | 1502 | |
bdc44640 | 1503 | CPU_FOREACH(cpu) { |
182735ef | 1504 | if (!cpu->stopped) { |
e8faee06 | 1505 | return false; |
0ab07c62 | 1506 | } |
296af7c9 BS |
1507 | } |
1508 | ||
e8faee06 | 1509 | return true; |
296af7c9 BS |
1510 | } |
1511 | ||
1512 | void pause_all_vcpus(void) | |
1513 | { | |
bdc44640 | 1514 | CPUState *cpu; |
296af7c9 | 1515 | |
40daca54 | 1516 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false); |
bdc44640 | 1517 | CPU_FOREACH(cpu) { |
182735ef AF |
1518 | cpu->stop = true; |
1519 | qemu_cpu_kick(cpu); | |
296af7c9 BS |
1520 | } |
1521 | ||
aa723c23 | 1522 | if (qemu_in_vcpu_thread()) { |
d798e974 | 1523 | cpu_stop_current(); |
d798e974 JK |
1524 | } |
1525 | ||
296af7c9 | 1526 | while (!all_vcpus_paused()) { |
be7d6c57 | 1527 | qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex); |
bdc44640 | 1528 | CPU_FOREACH(cpu) { |
182735ef | 1529 | qemu_cpu_kick(cpu); |
296af7c9 BS |
1530 | } |
1531 | } | |
1532 | } | |
1533 | ||
2993683b IM |
1534 | void cpu_resume(CPUState *cpu) |
1535 | { | |
1536 | cpu->stop = false; | |
1537 | cpu->stopped = false; | |
1538 | qemu_cpu_kick(cpu); | |
1539 | } | |
1540 | ||
296af7c9 BS |
1541 | void resume_all_vcpus(void) |
1542 | { | |
bdc44640 | 1543 | CPUState *cpu; |
296af7c9 | 1544 | |
40daca54 | 1545 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true); |
bdc44640 | 1546 | CPU_FOREACH(cpu) { |
182735ef | 1547 | cpu_resume(cpu); |
296af7c9 BS |
1548 | } |
1549 | } | |
1550 | ||
4c055ab5 GZ |
1551 | void cpu_remove(CPUState *cpu) |
1552 | { | |
1553 | cpu->stop = true; | |
1554 | cpu->unplug = true; | |
1555 | qemu_cpu_kick(cpu); | |
1556 | } | |
1557 | ||
2c579042 BR |
1558 | void cpu_remove_sync(CPUState *cpu) |
1559 | { | |
1560 | cpu_remove(cpu); | |
1561 | while (cpu->created) { | |
1562 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1563 | } | |
1564 | } | |
1565 | ||
4900116e DDAG |
1566 | /* For temporary buffers for forming a name */ |
1567 | #define VCPU_THREAD_NAME_SIZE 16 | |
1568 | ||
e5ab30a2 | 1569 | static void qemu_tcg_init_vcpu(CPUState *cpu) |
296af7c9 | 1570 | { |
4900116e | 1571 | char thread_name[VCPU_THREAD_NAME_SIZE]; |
37257942 AB |
1572 | static QemuCond *single_tcg_halt_cond; |
1573 | static QemuThread *single_tcg_cpu_thread; | |
4900116e | 1574 | |
37257942 | 1575 | if (qemu_tcg_mttcg_enabled() || !single_tcg_cpu_thread) { |
814e612e | 1576 | cpu->thread = g_malloc0(sizeof(QemuThread)); |
f5c121b8 AF |
1577 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); |
1578 | qemu_cond_init(cpu->halt_cond); | |
37257942 AB |
1579 | |
1580 | if (qemu_tcg_mttcg_enabled()) { | |
1581 | /* create a thread per vCPU with TCG (MTTCG) */ | |
1582 | parallel_cpus = true; | |
1583 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG", | |
4900116e | 1584 | cpu->cpu_index); |
37257942 AB |
1585 | |
1586 | qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn, | |
1587 | cpu, QEMU_THREAD_JOINABLE); | |
1588 | ||
1589 | } else { | |
1590 | /* share a single thread for all cpus with TCG */ | |
1591 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "ALL CPUs/TCG"); | |
1592 | qemu_thread_create(cpu->thread, thread_name, | |
1593 | qemu_tcg_rr_cpu_thread_fn, | |
1594 | cpu, QEMU_THREAD_JOINABLE); | |
1595 | ||
1596 | single_tcg_halt_cond = cpu->halt_cond; | |
1597 | single_tcg_cpu_thread = cpu->thread; | |
1598 | } | |
1ecf47bf | 1599 | #ifdef _WIN32 |
814e612e | 1600 | cpu->hThread = qemu_thread_get_handle(cpu->thread); |
1ecf47bf | 1601 | #endif |
61a46217 | 1602 | while (!cpu->created) { |
18a85728 | 1603 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); |
0ab07c62 | 1604 | } |
296af7c9 | 1605 | } else { |
37257942 AB |
1606 | /* For non-MTTCG cases we share the thread */ |
1607 | cpu->thread = single_tcg_cpu_thread; | |
1608 | cpu->halt_cond = single_tcg_halt_cond; | |
296af7c9 BS |
1609 | } |
1610 | } | |
1611 | ||
b0cb0a66 VP |
1612 | static void qemu_hax_start_vcpu(CPUState *cpu) |
1613 | { | |
1614 | char thread_name[VCPU_THREAD_NAME_SIZE]; | |
1615 | ||
1616 | cpu->thread = g_malloc0(sizeof(QemuThread)); | |
1617 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); | |
1618 | qemu_cond_init(cpu->halt_cond); | |
1619 | ||
1620 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/HAX", | |
1621 | cpu->cpu_index); | |
1622 | qemu_thread_create(cpu->thread, thread_name, qemu_hax_cpu_thread_fn, | |
1623 | cpu, QEMU_THREAD_JOINABLE); | |
1624 | #ifdef _WIN32 | |
1625 | cpu->hThread = qemu_thread_get_handle(cpu->thread); | |
1626 | #endif | |
1627 | while (!cpu->created) { | |
1628 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1629 | } | |
1630 | } | |
1631 | ||
48a106bd | 1632 | static void qemu_kvm_start_vcpu(CPUState *cpu) |
296af7c9 | 1633 | { |
4900116e DDAG |
1634 | char thread_name[VCPU_THREAD_NAME_SIZE]; |
1635 | ||
814e612e | 1636 | cpu->thread = g_malloc0(sizeof(QemuThread)); |
f5c121b8 AF |
1637 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); |
1638 | qemu_cond_init(cpu->halt_cond); | |
4900116e DDAG |
1639 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/KVM", |
1640 | cpu->cpu_index); | |
1641 | qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn, | |
1642 | cpu, QEMU_THREAD_JOINABLE); | |
61a46217 | 1643 | while (!cpu->created) { |
18a85728 | 1644 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); |
0ab07c62 | 1645 | } |
296af7c9 BS |
1646 | } |
1647 | ||
10a9021d | 1648 | static void qemu_dummy_start_vcpu(CPUState *cpu) |
c7f0f3b1 | 1649 | { |
4900116e DDAG |
1650 | char thread_name[VCPU_THREAD_NAME_SIZE]; |
1651 | ||
814e612e | 1652 | cpu->thread = g_malloc0(sizeof(QemuThread)); |
f5c121b8 AF |
1653 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); |
1654 | qemu_cond_init(cpu->halt_cond); | |
4900116e DDAG |
1655 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/DUMMY", |
1656 | cpu->cpu_index); | |
1657 | qemu_thread_create(cpu->thread, thread_name, qemu_dummy_cpu_thread_fn, cpu, | |
c7f0f3b1 | 1658 | QEMU_THREAD_JOINABLE); |
61a46217 | 1659 | while (!cpu->created) { |
c7f0f3b1 AL |
1660 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); |
1661 | } | |
1662 | } | |
1663 | ||
c643bed9 | 1664 | void qemu_init_vcpu(CPUState *cpu) |
296af7c9 | 1665 | { |
ce3960eb AF |
1666 | cpu->nr_cores = smp_cores; |
1667 | cpu->nr_threads = smp_threads; | |
f324e766 | 1668 | cpu->stopped = true; |
56943e8c PM |
1669 | |
1670 | if (!cpu->as) { | |
1671 | /* If the target cpu hasn't set up any address spaces itself, | |
1672 | * give it the default one. | |
1673 | */ | |
6731d864 PC |
1674 | AddressSpace *as = address_space_init_shareable(cpu->memory, |
1675 | "cpu-memory"); | |
12ebc9a7 | 1676 | cpu->num_ases = 1; |
6731d864 | 1677 | cpu_address_space_init(cpu, as, 0); |
56943e8c PM |
1678 | } |
1679 | ||
0ab07c62 | 1680 | if (kvm_enabled()) { |
48a106bd | 1681 | qemu_kvm_start_vcpu(cpu); |
b0cb0a66 VP |
1682 | } else if (hax_enabled()) { |
1683 | qemu_hax_start_vcpu(cpu); | |
c7f0f3b1 | 1684 | } else if (tcg_enabled()) { |
e5ab30a2 | 1685 | qemu_tcg_init_vcpu(cpu); |
c7f0f3b1 | 1686 | } else { |
10a9021d | 1687 | qemu_dummy_start_vcpu(cpu); |
0ab07c62 | 1688 | } |
296af7c9 BS |
1689 | } |
1690 | ||
b4a3d965 | 1691 | void cpu_stop_current(void) |
296af7c9 | 1692 | { |
4917cf44 AF |
1693 | if (current_cpu) { |
1694 | current_cpu->stop = false; | |
1695 | current_cpu->stopped = true; | |
1696 | cpu_exit(current_cpu); | |
96bce683 | 1697 | qemu_cond_broadcast(&qemu_pause_cond); |
b4a3d965 | 1698 | } |
296af7c9 BS |
1699 | } |
1700 | ||
56983463 | 1701 | int vm_stop(RunState state) |
296af7c9 | 1702 | { |
aa723c23 | 1703 | if (qemu_in_vcpu_thread()) { |
74892d24 | 1704 | qemu_system_vmstop_request_prepare(); |
1dfb4dd9 | 1705 | qemu_system_vmstop_request(state); |
296af7c9 BS |
1706 | /* |
1707 | * FIXME: should not return to device code in case | |
1708 | * vm_stop() has been requested. | |
1709 | */ | |
b4a3d965 | 1710 | cpu_stop_current(); |
56983463 | 1711 | return 0; |
296af7c9 | 1712 | } |
56983463 KW |
1713 | |
1714 | return do_vm_stop(state); | |
296af7c9 BS |
1715 | } |
1716 | ||
2d76e823 CI |
1717 | /** |
1718 | * Prepare for (re)starting the VM. | |
1719 | * Returns -1 if the vCPUs are not to be restarted (e.g. if they are already | |
1720 | * running or in case of an error condition), 0 otherwise. | |
1721 | */ | |
1722 | int vm_prepare_start(void) | |
1723 | { | |
1724 | RunState requested; | |
1725 | int res = 0; | |
1726 | ||
1727 | qemu_vmstop_requested(&requested); | |
1728 | if (runstate_is_running() && requested == RUN_STATE__MAX) { | |
1729 | return -1; | |
1730 | } | |
1731 | ||
1732 | /* Ensure that a STOP/RESUME pair of events is emitted if a | |
1733 | * vmstop request was pending. The BLOCK_IO_ERROR event, for | |
1734 | * example, according to documentation is always followed by | |
1735 | * the STOP event. | |
1736 | */ | |
1737 | if (runstate_is_running()) { | |
1738 | qapi_event_send_stop(&error_abort); | |
1739 | res = -1; | |
1740 | } else { | |
1741 | replay_enable_events(); | |
1742 | cpu_enable_ticks(); | |
1743 | runstate_set(RUN_STATE_RUNNING); | |
1744 | vm_state_notify(1, RUN_STATE_RUNNING); | |
1745 | } | |
1746 | ||
1747 | /* We are sending this now, but the CPUs will be resumed shortly later */ | |
1748 | qapi_event_send_resume(&error_abort); | |
1749 | return res; | |
1750 | } | |
1751 | ||
1752 | void vm_start(void) | |
1753 | { | |
1754 | if (!vm_prepare_start()) { | |
1755 | resume_all_vcpus(); | |
1756 | } | |
1757 | } | |
1758 | ||
8a9236f1 LC |
1759 | /* does a state transition even if the VM is already stopped, |
1760 | current state is forgotten forever */ | |
56983463 | 1761 | int vm_stop_force_state(RunState state) |
8a9236f1 LC |
1762 | { |
1763 | if (runstate_is_running()) { | |
56983463 | 1764 | return vm_stop(state); |
8a9236f1 LC |
1765 | } else { |
1766 | runstate_set(state); | |
b2780d32 WC |
1767 | |
1768 | bdrv_drain_all(); | |
594a45ce KW |
1769 | /* Make sure to return an error if the flush in a previous vm_stop() |
1770 | * failed. */ | |
22af08ea | 1771 | return bdrv_flush_all(); |
8a9236f1 LC |
1772 | } |
1773 | } | |
1774 | ||
9a78eead | 1775 | void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg) |
262353cb BS |
1776 | { |
1777 | /* XXX: implement xxx_cpu_list for targets that still miss it */ | |
e916cbf8 PM |
1778 | #if defined(cpu_list) |
1779 | cpu_list(f, cpu_fprintf); | |
262353cb BS |
1780 | #endif |
1781 | } | |
de0b36b6 LC |
1782 | |
1783 | CpuInfoList *qmp_query_cpus(Error **errp) | |
1784 | { | |
1785 | CpuInfoList *head = NULL, *cur_item = NULL; | |
182735ef | 1786 | CPUState *cpu; |
de0b36b6 | 1787 | |
bdc44640 | 1788 | CPU_FOREACH(cpu) { |
de0b36b6 | 1789 | CpuInfoList *info; |
182735ef AF |
1790 | #if defined(TARGET_I386) |
1791 | X86CPU *x86_cpu = X86_CPU(cpu); | |
1792 | CPUX86State *env = &x86_cpu->env; | |
1793 | #elif defined(TARGET_PPC) | |
1794 | PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu); | |
1795 | CPUPPCState *env = &ppc_cpu->env; | |
1796 | #elif defined(TARGET_SPARC) | |
1797 | SPARCCPU *sparc_cpu = SPARC_CPU(cpu); | |
1798 | CPUSPARCState *env = &sparc_cpu->env; | |
1799 | #elif defined(TARGET_MIPS) | |
1800 | MIPSCPU *mips_cpu = MIPS_CPU(cpu); | |
1801 | CPUMIPSState *env = &mips_cpu->env; | |
48e06fe0 BK |
1802 | #elif defined(TARGET_TRICORE) |
1803 | TriCoreCPU *tricore_cpu = TRICORE_CPU(cpu); | |
1804 | CPUTriCoreState *env = &tricore_cpu->env; | |
182735ef | 1805 | #endif |
de0b36b6 | 1806 | |
cb446eca | 1807 | cpu_synchronize_state(cpu); |
de0b36b6 LC |
1808 | |
1809 | info = g_malloc0(sizeof(*info)); | |
1810 | info->value = g_malloc0(sizeof(*info->value)); | |
55e5c285 | 1811 | info->value->CPU = cpu->cpu_index; |
182735ef | 1812 | info->value->current = (cpu == first_cpu); |
259186a7 | 1813 | info->value->halted = cpu->halted; |
58f88d4b | 1814 | info->value->qom_path = object_get_canonical_path(OBJECT(cpu)); |
9f09e18a | 1815 | info->value->thread_id = cpu->thread_id; |
de0b36b6 | 1816 | #if defined(TARGET_I386) |
86f4b687 | 1817 | info->value->arch = CPU_INFO_ARCH_X86; |
544a3731 | 1818 | info->value->u.x86.pc = env->eip + env->segs[R_CS].base; |
de0b36b6 | 1819 | #elif defined(TARGET_PPC) |
86f4b687 | 1820 | info->value->arch = CPU_INFO_ARCH_PPC; |
544a3731 | 1821 | info->value->u.ppc.nip = env->nip; |
de0b36b6 | 1822 | #elif defined(TARGET_SPARC) |
86f4b687 | 1823 | info->value->arch = CPU_INFO_ARCH_SPARC; |
544a3731 EB |
1824 | info->value->u.q_sparc.pc = env->pc; |
1825 | info->value->u.q_sparc.npc = env->npc; | |
de0b36b6 | 1826 | #elif defined(TARGET_MIPS) |
86f4b687 | 1827 | info->value->arch = CPU_INFO_ARCH_MIPS; |
544a3731 | 1828 | info->value->u.q_mips.PC = env->active_tc.PC; |
48e06fe0 | 1829 | #elif defined(TARGET_TRICORE) |
86f4b687 | 1830 | info->value->arch = CPU_INFO_ARCH_TRICORE; |
544a3731 | 1831 | info->value->u.tricore.PC = env->PC; |
86f4b687 EB |
1832 | #else |
1833 | info->value->arch = CPU_INFO_ARCH_OTHER; | |
de0b36b6 LC |
1834 | #endif |
1835 | ||
1836 | /* XXX: waiting for the qapi to support GSList */ | |
1837 | if (!cur_item) { | |
1838 | head = cur_item = info; | |
1839 | } else { | |
1840 | cur_item->next = info; | |
1841 | cur_item = info; | |
1842 | } | |
1843 | } | |
1844 | ||
1845 | return head; | |
1846 | } | |
0cfd6a9a LC |
1847 | |
1848 | void qmp_memsave(int64_t addr, int64_t size, const char *filename, | |
1849 | bool has_cpu, int64_t cpu_index, Error **errp) | |
1850 | { | |
1851 | FILE *f; | |
1852 | uint32_t l; | |
55e5c285 | 1853 | CPUState *cpu; |
0cfd6a9a | 1854 | uint8_t buf[1024]; |
0dc9daf0 | 1855 | int64_t orig_addr = addr, orig_size = size; |
0cfd6a9a LC |
1856 | |
1857 | if (!has_cpu) { | |
1858 | cpu_index = 0; | |
1859 | } | |
1860 | ||
151d1322 AF |
1861 | cpu = qemu_get_cpu(cpu_index); |
1862 | if (cpu == NULL) { | |
c6bd8c70 MA |
1863 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index", |
1864 | "a CPU number"); | |
0cfd6a9a LC |
1865 | return; |
1866 | } | |
1867 | ||
1868 | f = fopen(filename, "wb"); | |
1869 | if (!f) { | |
618da851 | 1870 | error_setg_file_open(errp, errno, filename); |
0cfd6a9a LC |
1871 | return; |
1872 | } | |
1873 | ||
1874 | while (size != 0) { | |
1875 | l = sizeof(buf); | |
1876 | if (l > size) | |
1877 | l = size; | |
2f4d0f59 | 1878 | if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) { |
0dc9daf0 BP |
1879 | error_setg(errp, "Invalid addr 0x%016" PRIx64 "/size %" PRId64 |
1880 | " specified", orig_addr, orig_size); | |
2f4d0f59 AK |
1881 | goto exit; |
1882 | } | |
0cfd6a9a | 1883 | if (fwrite(buf, 1, l, f) != l) { |
c6bd8c70 | 1884 | error_setg(errp, QERR_IO_ERROR); |
0cfd6a9a LC |
1885 | goto exit; |
1886 | } | |
1887 | addr += l; | |
1888 | size -= l; | |
1889 | } | |
1890 | ||
1891 | exit: | |
1892 | fclose(f); | |
1893 | } | |
6d3962bf LC |
1894 | |
1895 | void qmp_pmemsave(int64_t addr, int64_t size, const char *filename, | |
1896 | Error **errp) | |
1897 | { | |
1898 | FILE *f; | |
1899 | uint32_t l; | |
1900 | uint8_t buf[1024]; | |
1901 | ||
1902 | f = fopen(filename, "wb"); | |
1903 | if (!f) { | |
618da851 | 1904 | error_setg_file_open(errp, errno, filename); |
6d3962bf LC |
1905 | return; |
1906 | } | |
1907 | ||
1908 | while (size != 0) { | |
1909 | l = sizeof(buf); | |
1910 | if (l > size) | |
1911 | l = size; | |
eb6282f2 | 1912 | cpu_physical_memory_read(addr, buf, l); |
6d3962bf | 1913 | if (fwrite(buf, 1, l, f) != l) { |
c6bd8c70 | 1914 | error_setg(errp, QERR_IO_ERROR); |
6d3962bf LC |
1915 | goto exit; |
1916 | } | |
1917 | addr += l; | |
1918 | size -= l; | |
1919 | } | |
1920 | ||
1921 | exit: | |
1922 | fclose(f); | |
1923 | } | |
ab49ab5c LC |
1924 | |
1925 | void qmp_inject_nmi(Error **errp) | |
1926 | { | |
9cb805fd | 1927 | nmi_monitor_handle(monitor_get_cpu_index(), errp); |
ab49ab5c | 1928 | } |
27498bef ST |
1929 | |
1930 | void dump_drift_info(FILE *f, fprintf_function cpu_fprintf) | |
1931 | { | |
1932 | if (!use_icount) { | |
1933 | return; | |
1934 | } | |
1935 | ||
1936 | cpu_fprintf(f, "Host - Guest clock %"PRIi64" ms\n", | |
1937 | (cpu_get_clock() - cpu_get_icount())/SCALE_MS); | |
1938 | if (icount_align_option) { | |
1939 | cpu_fprintf(f, "Max guest delay %"PRIi64" ms\n", -max_delay/SCALE_MS); | |
1940 | cpu_fprintf(f, "Max guest advance %"PRIi64" ms\n", max_advance/SCALE_MS); | |
1941 | } else { | |
1942 | cpu_fprintf(f, "Max guest delay NA\n"); | |
1943 | cpu_fprintf(f, "Max guest advance NA\n"); | |
1944 | } | |
1945 | } |