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