]>
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 JK |
53 | |
54 | #ifndef _WIN32 | |
1de7afc9 | 55 | #include "qemu/compatfd.h" |
0ff0fc19 | 56 | #endif |
296af7c9 | 57 | |
6d9cb73c JK |
58 | #ifdef CONFIG_LINUX |
59 | ||
60 | #include <sys/prctl.h> | |
61 | ||
c0532a76 MT |
62 | #ifndef PR_MCE_KILL |
63 | #define PR_MCE_KILL 33 | |
64 | #endif | |
65 | ||
6d9cb73c JK |
66 | #ifndef PR_MCE_KILL_SET |
67 | #define PR_MCE_KILL_SET 1 | |
68 | #endif | |
69 | ||
70 | #ifndef PR_MCE_KILL_EARLY | |
71 | #define PR_MCE_KILL_EARLY 1 | |
72 | #endif | |
73 | ||
74 | #endif /* CONFIG_LINUX */ | |
75 | ||
27498bef ST |
76 | int64_t max_delay; |
77 | int64_t max_advance; | |
296af7c9 | 78 | |
2adcc85d JH |
79 | /* vcpu throttling controls */ |
80 | static QEMUTimer *throttle_timer; | |
81 | static unsigned int throttle_percentage; | |
82 | ||
83 | #define CPU_THROTTLE_PCT_MIN 1 | |
84 | #define CPU_THROTTLE_PCT_MAX 99 | |
85 | #define CPU_THROTTLE_TIMESLICE_NS 10000000 | |
86 | ||
321bc0b2 TC |
87 | bool cpu_is_stopped(CPUState *cpu) |
88 | { | |
89 | return cpu->stopped || !runstate_is_running(); | |
90 | } | |
91 | ||
a98ae1d8 | 92 | static bool cpu_thread_is_idle(CPUState *cpu) |
ac873f1e | 93 | { |
c64ca814 | 94 | if (cpu->stop || cpu->queued_work_first) { |
ac873f1e PM |
95 | return false; |
96 | } | |
321bc0b2 | 97 | if (cpu_is_stopped(cpu)) { |
ac873f1e PM |
98 | return true; |
99 | } | |
8c2e1b00 | 100 | if (!cpu->halted || cpu_has_work(cpu) || |
215e79c0 | 101 | kvm_halt_in_kernel()) { |
ac873f1e PM |
102 | return false; |
103 | } | |
104 | return true; | |
105 | } | |
106 | ||
107 | static bool all_cpu_threads_idle(void) | |
108 | { | |
182735ef | 109 | CPUState *cpu; |
ac873f1e | 110 | |
bdc44640 | 111 | CPU_FOREACH(cpu) { |
182735ef | 112 | if (!cpu_thread_is_idle(cpu)) { |
ac873f1e PM |
113 | return false; |
114 | } | |
115 | } | |
116 | return true; | |
117 | } | |
118 | ||
946fb27c PB |
119 | /***********************************************************/ |
120 | /* guest cycle counter */ | |
121 | ||
a3270e19 PB |
122 | /* Protected by TimersState seqlock */ |
123 | ||
5045e9d9 | 124 | static bool icount_sleep = true; |
71468395 | 125 | static int64_t vm_clock_warp_start = -1; |
946fb27c PB |
126 | /* Conversion factor from emulated instructions to virtual clock ticks. */ |
127 | static int icount_time_shift; | |
128 | /* Arbitrarily pick 1MIPS as the minimum allowable speed. */ | |
129 | #define MAX_ICOUNT_SHIFT 10 | |
a3270e19 | 130 | |
946fb27c PB |
131 | static QEMUTimer *icount_rt_timer; |
132 | static QEMUTimer *icount_vm_timer; | |
133 | static QEMUTimer *icount_warp_timer; | |
946fb27c PB |
134 | |
135 | typedef struct TimersState { | |
cb365646 | 136 | /* Protected by BQL. */ |
946fb27c PB |
137 | int64_t cpu_ticks_prev; |
138 | int64_t cpu_ticks_offset; | |
cb365646 LPF |
139 | |
140 | /* cpu_clock_offset can be read out of BQL, so protect it with | |
141 | * this lock. | |
142 | */ | |
143 | QemuSeqLock vm_clock_seqlock; | |
946fb27c PB |
144 | int64_t cpu_clock_offset; |
145 | int32_t cpu_ticks_enabled; | |
146 | int64_t dummy; | |
c96778bb FK |
147 | |
148 | /* Compensate for varying guest execution speed. */ | |
149 | int64_t qemu_icount_bias; | |
150 | /* Only written by TCG thread */ | |
151 | int64_t qemu_icount; | |
946fb27c PB |
152 | } TimersState; |
153 | ||
d9cd4007 | 154 | static TimersState timers_state; |
8d4e9146 FK |
155 | bool mttcg_enabled; |
156 | ||
157 | /* | |
158 | * We default to false if we know other options have been enabled | |
159 | * which are currently incompatible with MTTCG. Otherwise when each | |
160 | * guest (target) has been updated to support: | |
161 | * - atomic instructions | |
162 | * - memory ordering primitives (barriers) | |
163 | * they can set the appropriate CONFIG flags in ${target}-softmmu.mak | |
164 | * | |
165 | * Once a guest architecture has been converted to the new primitives | |
166 | * there are two remaining limitations to check. | |
167 | * | |
168 | * - The guest can't be oversized (e.g. 64 bit guest on 32 bit host) | |
169 | * - The host must have a stronger memory order than the guest | |
170 | * | |
171 | * It may be possible in future to support strong guests on weak hosts | |
172 | * but that will require tagging all load/stores in a guest with their | |
173 | * implicit memory order requirements which would likely slow things | |
174 | * down a lot. | |
175 | */ | |
176 | ||
177 | static bool check_tcg_memory_orders_compatible(void) | |
178 | { | |
179 | #if defined(TCG_GUEST_DEFAULT_MO) && defined(TCG_TARGET_DEFAULT_MO) | |
180 | return (TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO) == 0; | |
181 | #else | |
182 | return false; | |
183 | #endif | |
184 | } | |
185 | ||
186 | static bool default_mttcg_enabled(void) | |
187 | { | |
188 | QemuOpts *icount_opts = qemu_find_opts_singleton("icount"); | |
189 | const char *rr = qemu_opt_get(icount_opts, "rr"); | |
190 | ||
191 | if (rr || TCG_OVERSIZED_GUEST) { | |
192 | return false; | |
193 | } else { | |
194 | #ifdef TARGET_SUPPORTS_MTTCG | |
195 | return check_tcg_memory_orders_compatible(); | |
196 | #else | |
197 | return false; | |
198 | #endif | |
199 | } | |
200 | } | |
201 | ||
202 | void qemu_tcg_configure(QemuOpts *opts, Error **errp) | |
203 | { | |
204 | const char *t = qemu_opt_get(opts, "thread"); | |
205 | if (t) { | |
206 | if (strcmp(t, "multi") == 0) { | |
207 | if (TCG_OVERSIZED_GUEST) { | |
208 | error_setg(errp, "No MTTCG when guest word size > hosts"); | |
209 | } else { | |
210 | if (!check_tcg_memory_orders_compatible()) { | |
211 | error_report("Guest expects a stronger memory ordering " | |
212 | "than the host provides"); | |
213 | error_printf("This may cause strange/hard to debug errors"); | |
214 | } | |
215 | mttcg_enabled = true; | |
216 | } | |
217 | } else if (strcmp(t, "single") == 0) { | |
218 | mttcg_enabled = false; | |
219 | } else { | |
220 | error_setg(errp, "Invalid 'thread' setting %s", t); | |
221 | } | |
222 | } else { | |
223 | mttcg_enabled = default_mttcg_enabled(); | |
224 | } | |
225 | } | |
946fb27c | 226 | |
2a62914b | 227 | int64_t cpu_get_icount_raw(void) |
946fb27c PB |
228 | { |
229 | int64_t icount; | |
4917cf44 | 230 | CPUState *cpu = current_cpu; |
946fb27c | 231 | |
c96778bb | 232 | icount = timers_state.qemu_icount; |
4917cf44 | 233 | if (cpu) { |
414b15c9 | 234 | if (!cpu->can_do_io) { |
2a62914b PD |
235 | fprintf(stderr, "Bad icount read\n"); |
236 | exit(1); | |
946fb27c | 237 | } |
28ecfd7a | 238 | icount -= (cpu->icount_decr.u16.low + cpu->icount_extra); |
946fb27c | 239 | } |
2a62914b PD |
240 | return icount; |
241 | } | |
242 | ||
243 | /* Return the virtual CPU time, based on the instruction counter. */ | |
244 | static int64_t cpu_get_icount_locked(void) | |
245 | { | |
246 | int64_t icount = cpu_get_icount_raw(); | |
3f031313 | 247 | return timers_state.qemu_icount_bias + cpu_icount_to_ns(icount); |
946fb27c PB |
248 | } |
249 | ||
17a15f1b PB |
250 | int64_t cpu_get_icount(void) |
251 | { | |
252 | int64_t icount; | |
253 | unsigned start; | |
254 | ||
255 | do { | |
256 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
257 | icount = cpu_get_icount_locked(); | |
258 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
259 | ||
260 | return icount; | |
261 | } | |
262 | ||
3f031313 FK |
263 | int64_t cpu_icount_to_ns(int64_t icount) |
264 | { | |
265 | return icount << icount_time_shift; | |
266 | } | |
267 | ||
d90f3cca C |
268 | /* return the time elapsed in VM between vm_start and vm_stop. Unless |
269 | * icount is active, cpu_get_ticks() uses units of the host CPU cycle | |
270 | * counter. | |
271 | * | |
272 | * Caller must hold the BQL | |
273 | */ | |
946fb27c PB |
274 | int64_t cpu_get_ticks(void) |
275 | { | |
5f3e3101 PB |
276 | int64_t ticks; |
277 | ||
946fb27c PB |
278 | if (use_icount) { |
279 | return cpu_get_icount(); | |
280 | } | |
5f3e3101 PB |
281 | |
282 | ticks = timers_state.cpu_ticks_offset; | |
283 | if (timers_state.cpu_ticks_enabled) { | |
4a7428c5 | 284 | ticks += cpu_get_host_ticks(); |
5f3e3101 PB |
285 | } |
286 | ||
287 | if (timers_state.cpu_ticks_prev > ticks) { | |
288 | /* Note: non increasing ticks may happen if the host uses | |
289 | software suspend */ | |
290 | timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks; | |
291 | ticks = timers_state.cpu_ticks_prev; | |
946fb27c | 292 | } |
5f3e3101 PB |
293 | |
294 | timers_state.cpu_ticks_prev = ticks; | |
295 | return ticks; | |
946fb27c PB |
296 | } |
297 | ||
cb365646 | 298 | static int64_t cpu_get_clock_locked(void) |
946fb27c | 299 | { |
1d45cea5 | 300 | int64_t time; |
cb365646 | 301 | |
1d45cea5 | 302 | time = timers_state.cpu_clock_offset; |
5f3e3101 | 303 | if (timers_state.cpu_ticks_enabled) { |
1d45cea5 | 304 | time += get_clock(); |
946fb27c | 305 | } |
cb365646 | 306 | |
1d45cea5 | 307 | return time; |
cb365646 LPF |
308 | } |
309 | ||
d90f3cca | 310 | /* Return the monotonic time elapsed in VM, i.e., |
8212ff86 PM |
311 | * the time between vm_start and vm_stop |
312 | */ | |
cb365646 LPF |
313 | int64_t cpu_get_clock(void) |
314 | { | |
315 | int64_t ti; | |
316 | unsigned start; | |
317 | ||
318 | do { | |
319 | start = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
320 | ti = cpu_get_clock_locked(); | |
321 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start)); | |
322 | ||
323 | return ti; | |
946fb27c PB |
324 | } |
325 | ||
cb365646 | 326 | /* enable cpu_get_ticks() |
3224e878 | 327 | * Caller must hold BQL which serves as mutex for vm_clock_seqlock. |
cb365646 | 328 | */ |
946fb27c PB |
329 | void cpu_enable_ticks(void) |
330 | { | |
cb365646 | 331 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ |
03719e44 | 332 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
946fb27c | 333 | if (!timers_state.cpu_ticks_enabled) { |
4a7428c5 | 334 | timers_state.cpu_ticks_offset -= cpu_get_host_ticks(); |
946fb27c PB |
335 | timers_state.cpu_clock_offset -= get_clock(); |
336 | timers_state.cpu_ticks_enabled = 1; | |
337 | } | |
03719e44 | 338 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
946fb27c PB |
339 | } |
340 | ||
341 | /* disable cpu_get_ticks() : the clock is stopped. You must not call | |
cb365646 | 342 | * cpu_get_ticks() after that. |
3224e878 | 343 | * Caller must hold BQL which serves as mutex for vm_clock_seqlock. |
cb365646 | 344 | */ |
946fb27c PB |
345 | void cpu_disable_ticks(void) |
346 | { | |
cb365646 | 347 | /* Here, the really thing protected by seqlock is cpu_clock_offset. */ |
03719e44 | 348 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
946fb27c | 349 | if (timers_state.cpu_ticks_enabled) { |
4a7428c5 | 350 | timers_state.cpu_ticks_offset += cpu_get_host_ticks(); |
cb365646 | 351 | timers_state.cpu_clock_offset = cpu_get_clock_locked(); |
946fb27c PB |
352 | timers_state.cpu_ticks_enabled = 0; |
353 | } | |
03719e44 | 354 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
946fb27c PB |
355 | } |
356 | ||
357 | /* Correlation between real and virtual time is always going to be | |
358 | fairly approximate, so ignore small variation. | |
359 | When the guest is idle real and virtual time will be aligned in | |
360 | the IO wait loop. */ | |
73bcb24d | 361 | #define ICOUNT_WOBBLE (NANOSECONDS_PER_SECOND / 10) |
946fb27c PB |
362 | |
363 | static void icount_adjust(void) | |
364 | { | |
365 | int64_t cur_time; | |
366 | int64_t cur_icount; | |
367 | int64_t delta; | |
a3270e19 PB |
368 | |
369 | /* Protected by TimersState mutex. */ | |
946fb27c | 370 | static int64_t last_delta; |
468cc7cf | 371 | |
946fb27c PB |
372 | /* If the VM is not running, then do nothing. */ |
373 | if (!runstate_is_running()) { | |
374 | return; | |
375 | } | |
468cc7cf | 376 | |
03719e44 | 377 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
17a15f1b PB |
378 | cur_time = cpu_get_clock_locked(); |
379 | cur_icount = cpu_get_icount_locked(); | |
468cc7cf | 380 | |
946fb27c PB |
381 | delta = cur_icount - cur_time; |
382 | /* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */ | |
383 | if (delta > 0 | |
384 | && last_delta + ICOUNT_WOBBLE < delta * 2 | |
385 | && icount_time_shift > 0) { | |
386 | /* The guest is getting too far ahead. Slow time down. */ | |
387 | icount_time_shift--; | |
388 | } | |
389 | if (delta < 0 | |
390 | && last_delta - ICOUNT_WOBBLE > delta * 2 | |
391 | && icount_time_shift < MAX_ICOUNT_SHIFT) { | |
392 | /* The guest is getting too far behind. Speed time up. */ | |
393 | icount_time_shift++; | |
394 | } | |
395 | last_delta = delta; | |
c96778bb FK |
396 | timers_state.qemu_icount_bias = cur_icount |
397 | - (timers_state.qemu_icount << icount_time_shift); | |
03719e44 | 398 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
946fb27c PB |
399 | } |
400 | ||
401 | static void icount_adjust_rt(void *opaque) | |
402 | { | |
40daca54 | 403 | timer_mod(icount_rt_timer, |
1979b908 | 404 | qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000); |
946fb27c PB |
405 | icount_adjust(); |
406 | } | |
407 | ||
408 | static void icount_adjust_vm(void *opaque) | |
409 | { | |
40daca54 AB |
410 | timer_mod(icount_vm_timer, |
411 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
73bcb24d | 412 | NANOSECONDS_PER_SECOND / 10); |
946fb27c PB |
413 | icount_adjust(); |
414 | } | |
415 | ||
416 | static int64_t qemu_icount_round(int64_t count) | |
417 | { | |
418 | return (count + (1 << icount_time_shift) - 1) >> icount_time_shift; | |
419 | } | |
420 | ||
efab87cf | 421 | static void icount_warp_rt(void) |
946fb27c | 422 | { |
ccffff48 AB |
423 | unsigned seq; |
424 | int64_t warp_start; | |
425 | ||
17a15f1b PB |
426 | /* The icount_warp_timer is rescheduled soon after vm_clock_warp_start |
427 | * changes from -1 to another value, so the race here is okay. | |
428 | */ | |
ccffff48 AB |
429 | do { |
430 | seq = seqlock_read_begin(&timers_state.vm_clock_seqlock); | |
431 | warp_start = vm_clock_warp_start; | |
432 | } while (seqlock_read_retry(&timers_state.vm_clock_seqlock, seq)); | |
433 | ||
434 | if (warp_start == -1) { | |
946fb27c PB |
435 | return; |
436 | } | |
437 | ||
03719e44 | 438 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
946fb27c | 439 | if (runstate_is_running()) { |
8eda206e PD |
440 | int64_t clock = REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT, |
441 | cpu_get_clock_locked()); | |
8ed961d9 PB |
442 | int64_t warp_delta; |
443 | ||
444 | warp_delta = clock - vm_clock_warp_start; | |
445 | if (use_icount == 2) { | |
946fb27c | 446 | /* |
40daca54 | 447 | * In adaptive mode, do not let QEMU_CLOCK_VIRTUAL run too |
946fb27c PB |
448 | * far ahead of real time. |
449 | */ | |
17a15f1b | 450 | int64_t cur_icount = cpu_get_icount_locked(); |
bf2a7ddb | 451 | int64_t delta = clock - cur_icount; |
8ed961d9 | 452 | warp_delta = MIN(warp_delta, delta); |
946fb27c | 453 | } |
c96778bb | 454 | timers_state.qemu_icount_bias += warp_delta; |
946fb27c PB |
455 | } |
456 | vm_clock_warp_start = -1; | |
03719e44 | 457 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
8ed961d9 PB |
458 | |
459 | if (qemu_clock_expired(QEMU_CLOCK_VIRTUAL)) { | |
460 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
461 | } | |
946fb27c PB |
462 | } |
463 | ||
e76d1798 | 464 | static void icount_timer_cb(void *opaque) |
efab87cf | 465 | { |
e76d1798 PD |
466 | /* No need for a checkpoint because the timer already synchronizes |
467 | * with CHECKPOINT_CLOCK_VIRTUAL_RT. | |
468 | */ | |
469 | icount_warp_rt(); | |
efab87cf PD |
470 | } |
471 | ||
8156be56 PB |
472 | void qtest_clock_warp(int64_t dest) |
473 | { | |
40daca54 | 474 | int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); |
efef88b3 | 475 | AioContext *aio_context; |
8156be56 | 476 | assert(qtest_enabled()); |
efef88b3 | 477 | aio_context = qemu_get_aio_context(); |
8156be56 | 478 | while (clock < dest) { |
40daca54 | 479 | int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); |
c9299e2f | 480 | int64_t warp = qemu_soonest_timeout(dest - clock, deadline); |
efef88b3 | 481 | |
03719e44 | 482 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
c96778bb | 483 | timers_state.qemu_icount_bias += warp; |
03719e44 | 484 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
17a15f1b | 485 | |
40daca54 | 486 | qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL); |
efef88b3 | 487 | timerlist_run_timers(aio_context->tlg.tl[QEMU_CLOCK_VIRTUAL]); |
40daca54 | 488 | clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); |
8156be56 | 489 | } |
40daca54 | 490 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
8156be56 PB |
491 | } |
492 | ||
e76d1798 | 493 | void qemu_start_warp_timer(void) |
946fb27c | 494 | { |
ce78d18c | 495 | int64_t clock; |
946fb27c PB |
496 | int64_t deadline; |
497 | ||
e76d1798 | 498 | if (!use_icount) { |
946fb27c PB |
499 | return; |
500 | } | |
501 | ||
8bd7f71d PD |
502 | /* Nothing to do if the VM is stopped: QEMU_CLOCK_VIRTUAL timers |
503 | * do not fire, so computing the deadline does not make sense. | |
504 | */ | |
505 | if (!runstate_is_running()) { | |
506 | return; | |
507 | } | |
508 | ||
509 | /* warp clock deterministically in record/replay mode */ | |
e76d1798 | 510 | if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_START)) { |
8bd7f71d PD |
511 | return; |
512 | } | |
513 | ||
ce78d18c | 514 | if (!all_cpu_threads_idle()) { |
946fb27c PB |
515 | return; |
516 | } | |
517 | ||
8156be56 PB |
518 | if (qtest_enabled()) { |
519 | /* When testing, qtest commands advance icount. */ | |
e76d1798 | 520 | return; |
8156be56 PB |
521 | } |
522 | ||
ac70aafc | 523 | /* We want to use the earliest deadline from ALL vm_clocks */ |
bf2a7ddb | 524 | clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT); |
40daca54 | 525 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); |
ce78d18c | 526 | if (deadline < 0) { |
d7a0f71d VC |
527 | static bool notified; |
528 | if (!icount_sleep && !notified) { | |
529 | error_report("WARNING: icount sleep disabled and no active timers"); | |
530 | notified = true; | |
531 | } | |
ce78d18c | 532 | return; |
ac70aafc AB |
533 | } |
534 | ||
946fb27c PB |
535 | if (deadline > 0) { |
536 | /* | |
40daca54 | 537 | * Ensure QEMU_CLOCK_VIRTUAL proceeds even when the virtual CPU goes to |
946fb27c PB |
538 | * sleep. Otherwise, the CPU might be waiting for a future timer |
539 | * interrupt to wake it up, but the interrupt never comes because | |
540 | * the vCPU isn't running any insns and thus doesn't advance the | |
40daca54 | 541 | * QEMU_CLOCK_VIRTUAL. |
946fb27c | 542 | */ |
5045e9d9 VC |
543 | if (!icount_sleep) { |
544 | /* | |
545 | * We never let VCPUs sleep in no sleep icount mode. | |
546 | * If there is a pending QEMU_CLOCK_VIRTUAL timer we just advance | |
547 | * to the next QEMU_CLOCK_VIRTUAL event and notify it. | |
548 | * It is useful when we want a deterministic execution time, | |
549 | * isolated from host latencies. | |
550 | */ | |
03719e44 | 551 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
5045e9d9 | 552 | timers_state.qemu_icount_bias += deadline; |
03719e44 | 553 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
5045e9d9 VC |
554 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
555 | } else { | |
556 | /* | |
557 | * We do stop VCPUs and only advance QEMU_CLOCK_VIRTUAL after some | |
558 | * "real" time, (related to the time left until the next event) has | |
559 | * passed. The QEMU_CLOCK_VIRTUAL_RT clock will do this. | |
560 | * This avoids that the warps are visible externally; for example, | |
561 | * you will not be sending network packets continuously instead of | |
562 | * every 100ms. | |
563 | */ | |
03719e44 | 564 | seqlock_write_begin(&timers_state.vm_clock_seqlock); |
5045e9d9 VC |
565 | if (vm_clock_warp_start == -1 || vm_clock_warp_start > clock) { |
566 | vm_clock_warp_start = clock; | |
567 | } | |
03719e44 | 568 | seqlock_write_end(&timers_state.vm_clock_seqlock); |
5045e9d9 | 569 | timer_mod_anticipate(icount_warp_timer, clock + deadline); |
ce78d18c | 570 | } |
ac70aafc | 571 | } else if (deadline == 0) { |
40daca54 | 572 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); |
946fb27c PB |
573 | } |
574 | } | |
575 | ||
e76d1798 PD |
576 | static void qemu_account_warp_timer(void) |
577 | { | |
578 | if (!use_icount || !icount_sleep) { | |
579 | return; | |
580 | } | |
581 | ||
582 | /* Nothing to do if the VM is stopped: QEMU_CLOCK_VIRTUAL timers | |
583 | * do not fire, so computing the deadline does not make sense. | |
584 | */ | |
585 | if (!runstate_is_running()) { | |
586 | return; | |
587 | } | |
588 | ||
589 | /* warp clock deterministically in record/replay mode */ | |
590 | if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_ACCOUNT)) { | |
591 | return; | |
592 | } | |
593 | ||
594 | timer_del(icount_warp_timer); | |
595 | icount_warp_rt(); | |
596 | } | |
597 | ||
d09eae37 FK |
598 | static bool icount_state_needed(void *opaque) |
599 | { | |
600 | return use_icount; | |
601 | } | |
602 | ||
603 | /* | |
604 | * This is a subsection for icount migration. | |
605 | */ | |
606 | static const VMStateDescription icount_vmstate_timers = { | |
607 | .name = "timer/icount", | |
608 | .version_id = 1, | |
609 | .minimum_version_id = 1, | |
5cd8cada | 610 | .needed = icount_state_needed, |
d09eae37 FK |
611 | .fields = (VMStateField[]) { |
612 | VMSTATE_INT64(qemu_icount_bias, TimersState), | |
613 | VMSTATE_INT64(qemu_icount, TimersState), | |
614 | VMSTATE_END_OF_LIST() | |
615 | } | |
616 | }; | |
617 | ||
946fb27c PB |
618 | static const VMStateDescription vmstate_timers = { |
619 | .name = "timer", | |
620 | .version_id = 2, | |
621 | .minimum_version_id = 1, | |
35d08458 | 622 | .fields = (VMStateField[]) { |
946fb27c PB |
623 | VMSTATE_INT64(cpu_ticks_offset, TimersState), |
624 | VMSTATE_INT64(dummy, TimersState), | |
625 | VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2), | |
626 | VMSTATE_END_OF_LIST() | |
d09eae37 | 627 | }, |
5cd8cada JQ |
628 | .subsections = (const VMStateDescription*[]) { |
629 | &icount_vmstate_timers, | |
630 | NULL | |
946fb27c PB |
631 | } |
632 | }; | |
633 | ||
14e6fe12 | 634 | static void cpu_throttle_thread(CPUState *cpu, run_on_cpu_data opaque) |
2adcc85d | 635 | { |
2adcc85d JH |
636 | double pct; |
637 | double throttle_ratio; | |
638 | long sleeptime_ns; | |
639 | ||
640 | if (!cpu_throttle_get_percentage()) { | |
641 | return; | |
642 | } | |
643 | ||
644 | pct = (double)cpu_throttle_get_percentage()/100; | |
645 | throttle_ratio = pct / (1 - pct); | |
646 | sleeptime_ns = (long)(throttle_ratio * CPU_THROTTLE_TIMESLICE_NS); | |
647 | ||
648 | qemu_mutex_unlock_iothread(); | |
649 | atomic_set(&cpu->throttle_thread_scheduled, 0); | |
650 | g_usleep(sleeptime_ns / 1000); /* Convert ns to us for usleep call */ | |
651 | qemu_mutex_lock_iothread(); | |
652 | } | |
653 | ||
654 | static void cpu_throttle_timer_tick(void *opaque) | |
655 | { | |
656 | CPUState *cpu; | |
657 | double pct; | |
658 | ||
659 | /* Stop the timer if needed */ | |
660 | if (!cpu_throttle_get_percentage()) { | |
661 | return; | |
662 | } | |
663 | CPU_FOREACH(cpu) { | |
664 | if (!atomic_xchg(&cpu->throttle_thread_scheduled, 1)) { | |
14e6fe12 PB |
665 | async_run_on_cpu(cpu, cpu_throttle_thread, |
666 | RUN_ON_CPU_NULL); | |
2adcc85d JH |
667 | } |
668 | } | |
669 | ||
670 | pct = (double)cpu_throttle_get_percentage()/100; | |
671 | timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) + | |
672 | CPU_THROTTLE_TIMESLICE_NS / (1-pct)); | |
673 | } | |
674 | ||
675 | void cpu_throttle_set(int new_throttle_pct) | |
676 | { | |
677 | /* Ensure throttle percentage is within valid range */ | |
678 | new_throttle_pct = MIN(new_throttle_pct, CPU_THROTTLE_PCT_MAX); | |
679 | new_throttle_pct = MAX(new_throttle_pct, CPU_THROTTLE_PCT_MIN); | |
680 | ||
681 | atomic_set(&throttle_percentage, new_throttle_pct); | |
682 | ||
683 | timer_mod(throttle_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT) + | |
684 | CPU_THROTTLE_TIMESLICE_NS); | |
685 | } | |
686 | ||
687 | void cpu_throttle_stop(void) | |
688 | { | |
689 | atomic_set(&throttle_percentage, 0); | |
690 | } | |
691 | ||
692 | bool cpu_throttle_active(void) | |
693 | { | |
694 | return (cpu_throttle_get_percentage() != 0); | |
695 | } | |
696 | ||
697 | int cpu_throttle_get_percentage(void) | |
698 | { | |
699 | return atomic_read(&throttle_percentage); | |
700 | } | |
701 | ||
4603ea01 PD |
702 | void cpu_ticks_init(void) |
703 | { | |
ccdb3c1f | 704 | seqlock_init(&timers_state.vm_clock_seqlock); |
4603ea01 | 705 | vmstate_register(NULL, 0, &vmstate_timers, &timers_state); |
2adcc85d JH |
706 | throttle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT, |
707 | cpu_throttle_timer_tick, NULL); | |
4603ea01 PD |
708 | } |
709 | ||
1ad9580b | 710 | void configure_icount(QemuOpts *opts, Error **errp) |
946fb27c | 711 | { |
1ad9580b | 712 | const char *option; |
a8bfac37 | 713 | char *rem_str = NULL; |
1ad9580b | 714 | |
1ad9580b | 715 | option = qemu_opt_get(opts, "shift"); |
946fb27c | 716 | if (!option) { |
a8bfac37 ST |
717 | if (qemu_opt_get(opts, "align") != NULL) { |
718 | error_setg(errp, "Please specify shift option when using align"); | |
719 | } | |
946fb27c PB |
720 | return; |
721 | } | |
f1f4b57e VC |
722 | |
723 | icount_sleep = qemu_opt_get_bool(opts, "sleep", true); | |
5045e9d9 VC |
724 | if (icount_sleep) { |
725 | icount_warp_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT, | |
e76d1798 | 726 | icount_timer_cb, NULL); |
5045e9d9 | 727 | } |
f1f4b57e | 728 | |
a8bfac37 | 729 | icount_align_option = qemu_opt_get_bool(opts, "align", false); |
f1f4b57e VC |
730 | |
731 | if (icount_align_option && !icount_sleep) { | |
778d9f9b | 732 | error_setg(errp, "align=on and sleep=off are incompatible"); |
f1f4b57e | 733 | } |
946fb27c | 734 | if (strcmp(option, "auto") != 0) { |
a8bfac37 ST |
735 | errno = 0; |
736 | icount_time_shift = strtol(option, &rem_str, 0); | |
737 | if (errno != 0 || *rem_str != '\0' || !strlen(option)) { | |
738 | error_setg(errp, "icount: Invalid shift value"); | |
739 | } | |
946fb27c PB |
740 | use_icount = 1; |
741 | return; | |
a8bfac37 ST |
742 | } else if (icount_align_option) { |
743 | error_setg(errp, "shift=auto and align=on are incompatible"); | |
f1f4b57e | 744 | } else if (!icount_sleep) { |
778d9f9b | 745 | error_setg(errp, "shift=auto and sleep=off are incompatible"); |
946fb27c PB |
746 | } |
747 | ||
748 | use_icount = 2; | |
749 | ||
750 | /* 125MIPS seems a reasonable initial guess at the guest speed. | |
751 | It will be corrected fairly quickly anyway. */ | |
752 | icount_time_shift = 3; | |
753 | ||
754 | /* Have both realtime and virtual time triggers for speed adjustment. | |
755 | The realtime trigger catches emulated time passing too slowly, | |
756 | the virtual time trigger catches emulated time passing too fast. | |
757 | Realtime triggers occur even when idle, so use them less frequently | |
758 | than VM triggers. */ | |
bf2a7ddb PD |
759 | icount_rt_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL_RT, |
760 | icount_adjust_rt, NULL); | |
40daca54 | 761 | timer_mod(icount_rt_timer, |
bf2a7ddb | 762 | qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL_RT) + 1000); |
40daca54 AB |
763 | icount_vm_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, |
764 | icount_adjust_vm, NULL); | |
765 | timer_mod(icount_vm_timer, | |
766 | qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + | |
73bcb24d | 767 | NANOSECONDS_PER_SECOND / 10); |
946fb27c PB |
768 | } |
769 | ||
6546706d AB |
770 | /***********************************************************/ |
771 | /* TCG vCPU kick timer | |
772 | * | |
773 | * The kick timer is responsible for moving single threaded vCPU | |
774 | * emulation on to the next vCPU. If more than one vCPU is running a | |
775 | * timer event with force a cpu->exit so the next vCPU can get | |
776 | * scheduled. | |
777 | * | |
778 | * The timer is removed if all vCPUs are idle and restarted again once | |
779 | * idleness is complete. | |
780 | */ | |
781 | ||
782 | static QEMUTimer *tcg_kick_vcpu_timer; | |
791158d9 | 783 | static CPUState *tcg_current_rr_cpu; |
6546706d AB |
784 | |
785 | #define TCG_KICK_PERIOD (NANOSECONDS_PER_SECOND / 10) | |
786 | ||
787 | static inline int64_t qemu_tcg_next_kick(void) | |
788 | { | |
789 | return qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + TCG_KICK_PERIOD; | |
790 | } | |
791 | ||
791158d9 AB |
792 | /* Kick the currently round-robin scheduled vCPU */ |
793 | static void qemu_cpu_kick_rr_cpu(void) | |
794 | { | |
795 | CPUState *cpu; | |
796 | atomic_mb_set(&exit_request, 1); | |
797 | do { | |
798 | cpu = atomic_mb_read(&tcg_current_rr_cpu); | |
799 | if (cpu) { | |
800 | cpu_exit(cpu); | |
801 | } | |
802 | } while (cpu != atomic_mb_read(&tcg_current_rr_cpu)); | |
803 | } | |
804 | ||
6546706d AB |
805 | static void kick_tcg_thread(void *opaque) |
806 | { | |
807 | timer_mod(tcg_kick_vcpu_timer, qemu_tcg_next_kick()); | |
791158d9 | 808 | qemu_cpu_kick_rr_cpu(); |
6546706d AB |
809 | } |
810 | ||
811 | static void start_tcg_kick_timer(void) | |
812 | { | |
813 | if (!tcg_kick_vcpu_timer && CPU_NEXT(first_cpu)) { | |
814 | tcg_kick_vcpu_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, | |
815 | kick_tcg_thread, NULL); | |
816 | timer_mod(tcg_kick_vcpu_timer, qemu_tcg_next_kick()); | |
817 | } | |
818 | } | |
819 | ||
820 | static void stop_tcg_kick_timer(void) | |
821 | { | |
822 | if (tcg_kick_vcpu_timer) { | |
823 | timer_del(tcg_kick_vcpu_timer); | |
824 | tcg_kick_vcpu_timer = NULL; | |
825 | } | |
826 | } | |
827 | ||
296af7c9 BS |
828 | /***********************************************************/ |
829 | void hw_error(const char *fmt, ...) | |
830 | { | |
831 | va_list ap; | |
55e5c285 | 832 | CPUState *cpu; |
296af7c9 BS |
833 | |
834 | va_start(ap, fmt); | |
835 | fprintf(stderr, "qemu: hardware error: "); | |
836 | vfprintf(stderr, fmt, ap); | |
837 | fprintf(stderr, "\n"); | |
bdc44640 | 838 | CPU_FOREACH(cpu) { |
55e5c285 | 839 | fprintf(stderr, "CPU #%d:\n", cpu->cpu_index); |
878096ee | 840 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU); |
296af7c9 BS |
841 | } |
842 | va_end(ap); | |
843 | abort(); | |
844 | } | |
845 | ||
846 | void cpu_synchronize_all_states(void) | |
847 | { | |
182735ef | 848 | CPUState *cpu; |
296af7c9 | 849 | |
bdc44640 | 850 | CPU_FOREACH(cpu) { |
182735ef | 851 | cpu_synchronize_state(cpu); |
296af7c9 BS |
852 | } |
853 | } | |
854 | ||
855 | void cpu_synchronize_all_post_reset(void) | |
856 | { | |
182735ef | 857 | CPUState *cpu; |
296af7c9 | 858 | |
bdc44640 | 859 | CPU_FOREACH(cpu) { |
182735ef | 860 | cpu_synchronize_post_reset(cpu); |
296af7c9 BS |
861 | } |
862 | } | |
863 | ||
864 | void cpu_synchronize_all_post_init(void) | |
865 | { | |
182735ef | 866 | CPUState *cpu; |
296af7c9 | 867 | |
bdc44640 | 868 | CPU_FOREACH(cpu) { |
182735ef | 869 | cpu_synchronize_post_init(cpu); |
296af7c9 BS |
870 | } |
871 | } | |
872 | ||
56983463 | 873 | static int do_vm_stop(RunState state) |
296af7c9 | 874 | { |
56983463 KW |
875 | int ret = 0; |
876 | ||
1354869c | 877 | if (runstate_is_running()) { |
296af7c9 | 878 | cpu_disable_ticks(); |
296af7c9 | 879 | pause_all_vcpus(); |
f5bbfba1 | 880 | runstate_set(state); |
1dfb4dd9 | 881 | vm_state_notify(0, state); |
a4e15de9 | 882 | qapi_event_send_stop(&error_abort); |
296af7c9 | 883 | } |
56983463 | 884 | |
594a45ce | 885 | bdrv_drain_all(); |
6d0ceb80 | 886 | replay_disable_events(); |
22af08ea | 887 | ret = bdrv_flush_all(); |
594a45ce | 888 | |
56983463 | 889 | return ret; |
296af7c9 BS |
890 | } |
891 | ||
a1fcaa73 | 892 | static bool cpu_can_run(CPUState *cpu) |
296af7c9 | 893 | { |
4fdeee7c | 894 | if (cpu->stop) { |
a1fcaa73 | 895 | return false; |
0ab07c62 | 896 | } |
321bc0b2 | 897 | if (cpu_is_stopped(cpu)) { |
a1fcaa73 | 898 | return false; |
0ab07c62 | 899 | } |
a1fcaa73 | 900 | return true; |
296af7c9 BS |
901 | } |
902 | ||
91325046 | 903 | static void cpu_handle_guest_debug(CPUState *cpu) |
83f338f7 | 904 | { |
64f6b346 | 905 | gdb_set_stop_cpu(cpu); |
8cf71710 | 906 | qemu_system_debug_request(); |
f324e766 | 907 | cpu->stopped = true; |
3c638d06 JK |
908 | } |
909 | ||
6d9cb73c JK |
910 | #ifdef CONFIG_LINUX |
911 | static void sigbus_reraise(void) | |
912 | { | |
913 | sigset_t set; | |
914 | struct sigaction action; | |
915 | ||
916 | memset(&action, 0, sizeof(action)); | |
917 | action.sa_handler = SIG_DFL; | |
918 | if (!sigaction(SIGBUS, &action, NULL)) { | |
919 | raise(SIGBUS); | |
920 | sigemptyset(&set); | |
921 | sigaddset(&set, SIGBUS); | |
a2d1761d | 922 | pthread_sigmask(SIG_UNBLOCK, &set, NULL); |
6d9cb73c JK |
923 | } |
924 | perror("Failed to re-raise SIGBUS!\n"); | |
925 | abort(); | |
926 | } | |
927 | ||
928 | static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo, | |
929 | void *ctx) | |
930 | { | |
931 | if (kvm_on_sigbus(siginfo->ssi_code, | |
932 | (void *)(intptr_t)siginfo->ssi_addr)) { | |
933 | sigbus_reraise(); | |
934 | } | |
935 | } | |
936 | ||
937 | static void qemu_init_sigbus(void) | |
938 | { | |
939 | struct sigaction action; | |
940 | ||
941 | memset(&action, 0, sizeof(action)); | |
942 | action.sa_flags = SA_SIGINFO; | |
943 | action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler; | |
944 | sigaction(SIGBUS, &action, NULL); | |
945 | ||
946 | prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0); | |
947 | } | |
948 | ||
290adf38 | 949 | static void qemu_kvm_eat_signals(CPUState *cpu) |
1ab3c6c0 JK |
950 | { |
951 | struct timespec ts = { 0, 0 }; | |
952 | siginfo_t siginfo; | |
953 | sigset_t waitset; | |
954 | sigset_t chkset; | |
955 | int r; | |
956 | ||
957 | sigemptyset(&waitset); | |
958 | sigaddset(&waitset, SIG_IPI); | |
959 | sigaddset(&waitset, SIGBUS); | |
960 | ||
961 | do { | |
962 | r = sigtimedwait(&waitset, &siginfo, &ts); | |
963 | if (r == -1 && !(errno == EAGAIN || errno == EINTR)) { | |
964 | perror("sigtimedwait"); | |
965 | exit(1); | |
966 | } | |
967 | ||
968 | switch (r) { | |
969 | case SIGBUS: | |
290adf38 | 970 | if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) { |
1ab3c6c0 JK |
971 | sigbus_reraise(); |
972 | } | |
973 | break; | |
974 | default: | |
975 | break; | |
976 | } | |
977 | ||
978 | r = sigpending(&chkset); | |
979 | if (r == -1) { | |
980 | perror("sigpending"); | |
981 | exit(1); | |
982 | } | |
983 | } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS)); | |
1ab3c6c0 JK |
984 | } |
985 | ||
6d9cb73c JK |
986 | #else /* !CONFIG_LINUX */ |
987 | ||
988 | static void qemu_init_sigbus(void) | |
989 | { | |
990 | } | |
1ab3c6c0 | 991 | |
290adf38 | 992 | static void qemu_kvm_eat_signals(CPUState *cpu) |
1ab3c6c0 JK |
993 | { |
994 | } | |
6d9cb73c JK |
995 | #endif /* !CONFIG_LINUX */ |
996 | ||
296af7c9 | 997 | #ifndef _WIN32 |
55f8d6ac JK |
998 | static void dummy_signal(int sig) |
999 | { | |
1000 | } | |
55f8d6ac | 1001 | |
13618e05 | 1002 | static void qemu_kvm_init_cpu_signals(CPUState *cpu) |
714bd040 PB |
1003 | { |
1004 | int r; | |
1005 | sigset_t set; | |
1006 | struct sigaction sigact; | |
1007 | ||
1008 | memset(&sigact, 0, sizeof(sigact)); | |
1009 | sigact.sa_handler = dummy_signal; | |
1010 | sigaction(SIG_IPI, &sigact, NULL); | |
1011 | ||
714bd040 PB |
1012 | pthread_sigmask(SIG_BLOCK, NULL, &set); |
1013 | sigdelset(&set, SIG_IPI); | |
714bd040 | 1014 | sigdelset(&set, SIGBUS); |
491d6e80 | 1015 | r = kvm_set_signal_mask(cpu, &set); |
714bd040 PB |
1016 | if (r) { |
1017 | fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r)); | |
1018 | exit(1); | |
1019 | } | |
1020 | } | |
1021 | ||
55f8d6ac | 1022 | #else /* _WIN32 */ |
13618e05 | 1023 | static void qemu_kvm_init_cpu_signals(CPUState *cpu) |
ff48eb5f | 1024 | { |
714bd040 PB |
1025 | abort(); |
1026 | } | |
714bd040 | 1027 | #endif /* _WIN32 */ |
ff48eb5f | 1028 | |
b2532d88 | 1029 | static QemuMutex qemu_global_mutex; |
46daff13 | 1030 | static QemuCond qemu_io_proceeded_cond; |
6b49809c | 1031 | static unsigned iothread_requesting_mutex; |
296af7c9 BS |
1032 | |
1033 | static QemuThread io_thread; | |
1034 | ||
296af7c9 BS |
1035 | /* cpu creation */ |
1036 | static QemuCond qemu_cpu_cond; | |
1037 | /* system init */ | |
296af7c9 BS |
1038 | static QemuCond qemu_pause_cond; |
1039 | ||
d3b12f5d | 1040 | void qemu_init_cpu_loop(void) |
296af7c9 | 1041 | { |
6d9cb73c | 1042 | qemu_init_sigbus(); |
ed94592b | 1043 | qemu_cond_init(&qemu_cpu_cond); |
ed94592b | 1044 | qemu_cond_init(&qemu_pause_cond); |
46daff13 | 1045 | qemu_cond_init(&qemu_io_proceeded_cond); |
296af7c9 | 1046 | qemu_mutex_init(&qemu_global_mutex); |
296af7c9 | 1047 | |
b7680cb6 | 1048 | qemu_thread_get_self(&io_thread); |
296af7c9 BS |
1049 | } |
1050 | ||
14e6fe12 | 1051 | void run_on_cpu(CPUState *cpu, run_on_cpu_func func, run_on_cpu_data data) |
e82bcec2 | 1052 | { |
d148d90e | 1053 | do_run_on_cpu(cpu, func, data, &qemu_global_mutex); |
3c02270d CV |
1054 | } |
1055 | ||
4c055ab5 GZ |
1056 | static void qemu_kvm_destroy_vcpu(CPUState *cpu) |
1057 | { | |
1058 | if (kvm_destroy_vcpu(cpu) < 0) { | |
1059 | error_report("kvm_destroy_vcpu failed"); | |
1060 | exit(EXIT_FAILURE); | |
1061 | } | |
1062 | } | |
1063 | ||
1064 | static void qemu_tcg_destroy_vcpu(CPUState *cpu) | |
1065 | { | |
1066 | } | |
1067 | ||
509a0d78 | 1068 | static void qemu_wait_io_event_common(CPUState *cpu) |
296af7c9 | 1069 | { |
4fdeee7c AF |
1070 | if (cpu->stop) { |
1071 | cpu->stop = false; | |
f324e766 | 1072 | cpu->stopped = true; |
96bce683 | 1073 | qemu_cond_broadcast(&qemu_pause_cond); |
296af7c9 | 1074 | } |
a5403c69 | 1075 | process_queued_cpu_work(cpu); |
216fc9a4 | 1076 | cpu->thread_kicked = false; |
296af7c9 BS |
1077 | } |
1078 | ||
d5f8d613 | 1079 | static void qemu_tcg_wait_io_event(CPUState *cpu) |
296af7c9 | 1080 | { |
16400322 | 1081 | while (all_cpu_threads_idle()) { |
6546706d | 1082 | stop_tcg_kick_timer(); |
d5f8d613 | 1083 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); |
16400322 | 1084 | } |
296af7c9 | 1085 | |
6546706d AB |
1086 | start_tcg_kick_timer(); |
1087 | ||
46daff13 PB |
1088 | while (iothread_requesting_mutex) { |
1089 | qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex); | |
1090 | } | |
6cabe1f3 | 1091 | |
bdc44640 | 1092 | CPU_FOREACH(cpu) { |
182735ef | 1093 | qemu_wait_io_event_common(cpu); |
6cabe1f3 | 1094 | } |
296af7c9 BS |
1095 | } |
1096 | ||
fd529e8f | 1097 | static void qemu_kvm_wait_io_event(CPUState *cpu) |
296af7c9 | 1098 | { |
a98ae1d8 | 1099 | while (cpu_thread_is_idle(cpu)) { |
f5c121b8 | 1100 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); |
16400322 | 1101 | } |
296af7c9 | 1102 | |
290adf38 | 1103 | qemu_kvm_eat_signals(cpu); |
509a0d78 | 1104 | qemu_wait_io_event_common(cpu); |
296af7c9 BS |
1105 | } |
1106 | ||
7e97cd88 | 1107 | static void *qemu_kvm_cpu_thread_fn(void *arg) |
296af7c9 | 1108 | { |
48a106bd | 1109 | CPUState *cpu = arg; |
84b4915d | 1110 | int r; |
296af7c9 | 1111 | |
ab28bd23 PB |
1112 | rcu_register_thread(); |
1113 | ||
2e7f7a3c | 1114 | qemu_mutex_lock_iothread(); |
814e612e | 1115 | qemu_thread_get_self(cpu->thread); |
9f09e18a | 1116 | cpu->thread_id = qemu_get_thread_id(); |
626cf8f4 | 1117 | cpu->can_do_io = 1; |
4917cf44 | 1118 | current_cpu = cpu; |
296af7c9 | 1119 | |
504134d2 | 1120 | r = kvm_init_vcpu(cpu); |
84b4915d JK |
1121 | if (r < 0) { |
1122 | fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r)); | |
1123 | exit(1); | |
1124 | } | |
296af7c9 | 1125 | |
13618e05 | 1126 | qemu_kvm_init_cpu_signals(cpu); |
296af7c9 BS |
1127 | |
1128 | /* signal CPU creation */ | |
61a46217 | 1129 | cpu->created = true; |
296af7c9 BS |
1130 | qemu_cond_signal(&qemu_cpu_cond); |
1131 | ||
4c055ab5 | 1132 | do { |
a1fcaa73 | 1133 | if (cpu_can_run(cpu)) { |
1458c363 | 1134 | r = kvm_cpu_exec(cpu); |
83f338f7 | 1135 | if (r == EXCP_DEBUG) { |
91325046 | 1136 | cpu_handle_guest_debug(cpu); |
83f338f7 | 1137 | } |
0ab07c62 | 1138 | } |
fd529e8f | 1139 | qemu_kvm_wait_io_event(cpu); |
4c055ab5 | 1140 | } while (!cpu->unplug || cpu_can_run(cpu)); |
296af7c9 | 1141 | |
4c055ab5 | 1142 | qemu_kvm_destroy_vcpu(cpu); |
2c579042 BR |
1143 | cpu->created = false; |
1144 | qemu_cond_signal(&qemu_cpu_cond); | |
4c055ab5 | 1145 | qemu_mutex_unlock_iothread(); |
296af7c9 BS |
1146 | return NULL; |
1147 | } | |
1148 | ||
c7f0f3b1 AL |
1149 | static void *qemu_dummy_cpu_thread_fn(void *arg) |
1150 | { | |
1151 | #ifdef _WIN32 | |
1152 | fprintf(stderr, "qtest is not supported under Windows\n"); | |
1153 | exit(1); | |
1154 | #else | |
10a9021d | 1155 | CPUState *cpu = arg; |
c7f0f3b1 AL |
1156 | sigset_t waitset; |
1157 | int r; | |
1158 | ||
ab28bd23 PB |
1159 | rcu_register_thread(); |
1160 | ||
c7f0f3b1 | 1161 | qemu_mutex_lock_iothread(); |
814e612e | 1162 | qemu_thread_get_self(cpu->thread); |
9f09e18a | 1163 | cpu->thread_id = qemu_get_thread_id(); |
626cf8f4 | 1164 | cpu->can_do_io = 1; |
c7f0f3b1 AL |
1165 | |
1166 | sigemptyset(&waitset); | |
1167 | sigaddset(&waitset, SIG_IPI); | |
1168 | ||
1169 | /* signal CPU creation */ | |
61a46217 | 1170 | cpu->created = true; |
c7f0f3b1 AL |
1171 | qemu_cond_signal(&qemu_cpu_cond); |
1172 | ||
4917cf44 | 1173 | current_cpu = cpu; |
c7f0f3b1 | 1174 | while (1) { |
4917cf44 | 1175 | current_cpu = NULL; |
c7f0f3b1 AL |
1176 | qemu_mutex_unlock_iothread(); |
1177 | do { | |
1178 | int sig; | |
1179 | r = sigwait(&waitset, &sig); | |
1180 | } while (r == -1 && (errno == EAGAIN || errno == EINTR)); | |
1181 | if (r == -1) { | |
1182 | perror("sigwait"); | |
1183 | exit(1); | |
1184 | } | |
1185 | qemu_mutex_lock_iothread(); | |
4917cf44 | 1186 | current_cpu = cpu; |
509a0d78 | 1187 | qemu_wait_io_event_common(cpu); |
c7f0f3b1 AL |
1188 | } |
1189 | ||
1190 | return NULL; | |
1191 | #endif | |
1192 | } | |
1193 | ||
1be7fcb8 AB |
1194 | static int64_t tcg_get_icount_limit(void) |
1195 | { | |
1196 | int64_t deadline; | |
1197 | ||
1198 | if (replay_mode != REPLAY_MODE_PLAY) { | |
1199 | deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
1200 | ||
1201 | /* Maintain prior (possibly buggy) behaviour where if no deadline | |
1202 | * was set (as there is no QEMU_CLOCK_VIRTUAL timer) or it is more than | |
1203 | * INT32_MAX nanoseconds ahead, we still use INT32_MAX | |
1204 | * nanoseconds. | |
1205 | */ | |
1206 | if ((deadline < 0) || (deadline > INT32_MAX)) { | |
1207 | deadline = INT32_MAX; | |
1208 | } | |
1209 | ||
1210 | return qemu_icount_round(deadline); | |
1211 | } else { | |
1212 | return replay_get_instructions(); | |
1213 | } | |
1214 | } | |
1215 | ||
12e9700d AB |
1216 | static void handle_icount_deadline(void) |
1217 | { | |
1218 | if (use_icount) { | |
1219 | int64_t deadline = | |
1220 | qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL); | |
1221 | ||
1222 | if (deadline == 0) { | |
1223 | qemu_clock_notify(QEMU_CLOCK_VIRTUAL); | |
1224 | } | |
1225 | } | |
1226 | } | |
1227 | ||
1be7fcb8 AB |
1228 | static int tcg_cpu_exec(CPUState *cpu) |
1229 | { | |
1230 | int ret; | |
1231 | #ifdef CONFIG_PROFILER | |
1232 | int64_t ti; | |
1233 | #endif | |
1234 | ||
1235 | #ifdef CONFIG_PROFILER | |
1236 | ti = profile_getclock(); | |
1237 | #endif | |
1238 | if (use_icount) { | |
1239 | int64_t count; | |
1240 | int decr; | |
1241 | timers_state.qemu_icount -= (cpu->icount_decr.u16.low | |
1242 | + cpu->icount_extra); | |
1243 | cpu->icount_decr.u16.low = 0; | |
1244 | cpu->icount_extra = 0; | |
1245 | count = tcg_get_icount_limit(); | |
1246 | timers_state.qemu_icount += count; | |
1247 | decr = (count > 0xffff) ? 0xffff : count; | |
1248 | count -= decr; | |
1249 | cpu->icount_decr.u16.low = decr; | |
1250 | cpu->icount_extra = count; | |
1251 | } | |
1252 | cpu_exec_start(cpu); | |
1253 | ret = cpu_exec(cpu); | |
1254 | cpu_exec_end(cpu); | |
1255 | #ifdef CONFIG_PROFILER | |
1256 | tcg_time += profile_getclock() - ti; | |
1257 | #endif | |
1258 | if (use_icount) { | |
1259 | /* Fold pending instructions back into the | |
1260 | instruction counter, and clear the interrupt flag. */ | |
1261 | timers_state.qemu_icount -= (cpu->icount_decr.u16.low | |
1262 | + cpu->icount_extra); | |
1263 | cpu->icount_decr.u32 = 0; | |
1264 | cpu->icount_extra = 0; | |
1265 | replay_account_executed_instructions(); | |
1266 | } | |
1267 | return ret; | |
1268 | } | |
1269 | ||
c93bbbef AB |
1270 | /* Destroy any remaining vCPUs which have been unplugged and have |
1271 | * finished running | |
1272 | */ | |
1273 | static void deal_with_unplugged_cpus(void) | |
1be7fcb8 | 1274 | { |
c93bbbef | 1275 | CPUState *cpu; |
1be7fcb8 | 1276 | |
c93bbbef AB |
1277 | CPU_FOREACH(cpu) { |
1278 | if (cpu->unplug && !cpu_can_run(cpu)) { | |
1279 | qemu_tcg_destroy_vcpu(cpu); | |
1280 | cpu->created = false; | |
1281 | qemu_cond_signal(&qemu_cpu_cond); | |
1be7fcb8 AB |
1282 | break; |
1283 | } | |
1284 | } | |
1be7fcb8 | 1285 | } |
bdb7ca67 | 1286 | |
6546706d AB |
1287 | /* Single-threaded TCG |
1288 | * | |
1289 | * In the single-threaded case each vCPU is simulated in turn. If | |
1290 | * there is more than a single vCPU we create a simple timer to kick | |
1291 | * the vCPU and ensure we don't get stuck in a tight loop in one vCPU. | |
1292 | * This is done explicitly rather than relying on side-effects | |
1293 | * elsewhere. | |
1294 | */ | |
1295 | ||
7e97cd88 | 1296 | static void *qemu_tcg_cpu_thread_fn(void *arg) |
296af7c9 | 1297 | { |
c3586ba7 | 1298 | CPUState *cpu = arg; |
296af7c9 | 1299 | |
ab28bd23 PB |
1300 | rcu_register_thread(); |
1301 | ||
2e7f7a3c | 1302 | qemu_mutex_lock_iothread(); |
814e612e | 1303 | qemu_thread_get_self(cpu->thread); |
296af7c9 | 1304 | |
38fcbd3f AF |
1305 | CPU_FOREACH(cpu) { |
1306 | cpu->thread_id = qemu_get_thread_id(); | |
1307 | cpu->created = true; | |
626cf8f4 | 1308 | cpu->can_do_io = 1; |
38fcbd3f | 1309 | } |
296af7c9 BS |
1310 | qemu_cond_signal(&qemu_cpu_cond); |
1311 | ||
fa7d1867 | 1312 | /* wait for initial kick-off after machine start */ |
c28e399c | 1313 | while (first_cpu->stopped) { |
d5f8d613 | 1314 | qemu_cond_wait(first_cpu->halt_cond, &qemu_global_mutex); |
8e564b4e JK |
1315 | |
1316 | /* process any pending work */ | |
bdc44640 | 1317 | CPU_FOREACH(cpu) { |
182735ef | 1318 | qemu_wait_io_event_common(cpu); |
8e564b4e | 1319 | } |
0ab07c62 | 1320 | } |
296af7c9 | 1321 | |
6546706d AB |
1322 | start_tcg_kick_timer(); |
1323 | ||
21618b3e | 1324 | /* process any pending work */ |
aed807c8 | 1325 | atomic_mb_set(&exit_request, 1); |
21618b3e | 1326 | |
c93bbbef AB |
1327 | cpu = first_cpu; |
1328 | ||
296af7c9 | 1329 | while (1) { |
c93bbbef AB |
1330 | /* Account partial waits to QEMU_CLOCK_VIRTUAL. */ |
1331 | qemu_account_warp_timer(); | |
1332 | ||
1333 | if (!cpu) { | |
1334 | cpu = first_cpu; | |
1335 | } | |
1336 | ||
1337 | for (; cpu != NULL && !exit_request; cpu = CPU_NEXT(cpu)) { | |
791158d9 | 1338 | atomic_mb_set(&tcg_current_rr_cpu, cpu); |
c93bbbef AB |
1339 | |
1340 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, | |
1341 | (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0); | |
1342 | ||
1343 | if (cpu_can_run(cpu)) { | |
1344 | int r; | |
1345 | r = tcg_cpu_exec(cpu); | |
1346 | if (r == EXCP_DEBUG) { | |
1347 | cpu_handle_guest_debug(cpu); | |
1348 | break; | |
1349 | } | |
1350 | } else if (cpu->stop || cpu->stopped) { | |
1351 | if (cpu->unplug) { | |
1352 | cpu = CPU_NEXT(cpu); | |
1353 | } | |
1354 | break; | |
1355 | } | |
1356 | ||
1357 | } /* for cpu.. */ | |
791158d9 AB |
1358 | /* Does not need atomic_mb_set because a spurious wakeup is okay. */ |
1359 | atomic_set(&tcg_current_rr_cpu, NULL); | |
c93bbbef AB |
1360 | |
1361 | /* Pairs with smp_wmb in qemu_cpu_kick. */ | |
1362 | atomic_mb_set(&exit_request, 0); | |
ac70aafc | 1363 | |
12e9700d | 1364 | handle_icount_deadline(); |
ac70aafc | 1365 | |
d5f8d613 | 1366 | qemu_tcg_wait_io_event(QTAILQ_FIRST(&cpus)); |
c93bbbef | 1367 | deal_with_unplugged_cpus(); |
296af7c9 BS |
1368 | } |
1369 | ||
1370 | return NULL; | |
1371 | } | |
1372 | ||
b0cb0a66 VP |
1373 | static void *qemu_hax_cpu_thread_fn(void *arg) |
1374 | { | |
1375 | CPUState *cpu = arg; | |
1376 | int r; | |
1377 | qemu_thread_get_self(cpu->thread); | |
1378 | qemu_mutex_lock(&qemu_global_mutex); | |
1379 | ||
1380 | cpu->thread_id = qemu_get_thread_id(); | |
1381 | cpu->created = true; | |
1382 | cpu->halted = 0; | |
1383 | current_cpu = cpu; | |
1384 | ||
1385 | hax_init_vcpu(cpu); | |
1386 | qemu_cond_signal(&qemu_cpu_cond); | |
1387 | ||
1388 | while (1) { | |
1389 | if (cpu_can_run(cpu)) { | |
1390 | r = hax_smp_cpu_exec(cpu); | |
1391 | if (r == EXCP_DEBUG) { | |
1392 | cpu_handle_guest_debug(cpu); | |
1393 | } | |
1394 | } | |
1395 | ||
1396 | while (cpu_thread_is_idle(cpu)) { | |
1397 | qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex); | |
1398 | } | |
1399 | #ifdef _WIN32 | |
1400 | SleepEx(0, TRUE); | |
1401 | #endif | |
1402 | qemu_wait_io_event_common(cpu); | |
1403 | } | |
1404 | return NULL; | |
1405 | } | |
1406 | ||
1407 | #ifdef _WIN32 | |
1408 | static void CALLBACK dummy_apc_func(ULONG_PTR unused) | |
1409 | { | |
1410 | } | |
1411 | #endif | |
1412 | ||
2ff09a40 | 1413 | static void qemu_cpu_kick_thread(CPUState *cpu) |
cc015e9a PB |
1414 | { |
1415 | #ifndef _WIN32 | |
1416 | int err; | |
1417 | ||
e0c38211 PB |
1418 | if (cpu->thread_kicked) { |
1419 | return; | |
9102deda | 1420 | } |
e0c38211 | 1421 | cpu->thread_kicked = true; |
814e612e | 1422 | err = pthread_kill(cpu->thread->thread, SIG_IPI); |
cc015e9a PB |
1423 | if (err) { |
1424 | fprintf(stderr, "qemu:%s: %s", __func__, strerror(err)); | |
1425 | exit(1); | |
1426 | } | |
1427 | #else /* _WIN32 */ | |
b0cb0a66 VP |
1428 | if (!qemu_cpu_is_self(cpu)) { |
1429 | if (!QueueUserAPC(dummy_apc_func, cpu->hThread, 0)) { | |
1430 | fprintf(stderr, "%s: QueueUserAPC failed with error %lu\n", | |
1431 | __func__, GetLastError()); | |
1432 | exit(1); | |
1433 | } | |
1434 | } | |
e0c38211 PB |
1435 | #endif |
1436 | } | |
ed9164a3 | 1437 | |
c08d7424 | 1438 | void qemu_cpu_kick(CPUState *cpu) |
296af7c9 | 1439 | { |
f5c121b8 | 1440 | qemu_cond_broadcast(cpu->halt_cond); |
e0c38211 | 1441 | if (tcg_enabled()) { |
791158d9 AB |
1442 | cpu_exit(cpu); |
1443 | /* Also ensure current RR cpu is kicked */ | |
1444 | qemu_cpu_kick_rr_cpu(); | |
e0c38211 | 1445 | } else { |
b0cb0a66 VP |
1446 | if (hax_enabled()) { |
1447 | /* | |
1448 | * FIXME: race condition with the exit_request check in | |
1449 | * hax_vcpu_hax_exec | |
1450 | */ | |
1451 | cpu->exit_request = 1; | |
1452 | } | |
e0c38211 PB |
1453 | qemu_cpu_kick_thread(cpu); |
1454 | } | |
296af7c9 BS |
1455 | } |
1456 | ||
46d62fac | 1457 | void qemu_cpu_kick_self(void) |
296af7c9 | 1458 | { |
4917cf44 | 1459 | assert(current_cpu); |
9102deda | 1460 | qemu_cpu_kick_thread(current_cpu); |
296af7c9 BS |
1461 | } |
1462 | ||
60e82579 | 1463 | bool qemu_cpu_is_self(CPUState *cpu) |
296af7c9 | 1464 | { |
814e612e | 1465 | return qemu_thread_is_self(cpu->thread); |
296af7c9 BS |
1466 | } |
1467 | ||
79e2b9ae | 1468 | bool qemu_in_vcpu_thread(void) |
aa723c23 | 1469 | { |
4917cf44 | 1470 | return current_cpu && qemu_cpu_is_self(current_cpu); |
aa723c23 JQ |
1471 | } |
1472 | ||
afbe7053 PB |
1473 | static __thread bool iothread_locked = false; |
1474 | ||
1475 | bool qemu_mutex_iothread_locked(void) | |
1476 | { | |
1477 | return iothread_locked; | |
1478 | } | |
1479 | ||
296af7c9 BS |
1480 | void qemu_mutex_lock_iothread(void) |
1481 | { | |
21618b3e | 1482 | atomic_inc(&iothread_requesting_mutex); |
2e7f7a3c PB |
1483 | /* In the simple case there is no need to bump the VCPU thread out of |
1484 | * TCG code execution. | |
1485 | */ | |
1486 | if (!tcg_enabled() || qemu_in_vcpu_thread() || | |
46036b24 | 1487 | !first_cpu || !first_cpu->created) { |
296af7c9 | 1488 | qemu_mutex_lock(&qemu_global_mutex); |
21618b3e | 1489 | atomic_dec(&iothread_requesting_mutex); |
1a28cac3 | 1490 | } else { |
1a28cac3 | 1491 | if (qemu_mutex_trylock(&qemu_global_mutex)) { |
791158d9 | 1492 | qemu_cpu_kick_rr_cpu(); |
1a28cac3 MT |
1493 | qemu_mutex_lock(&qemu_global_mutex); |
1494 | } | |
6b49809c | 1495 | atomic_dec(&iothread_requesting_mutex); |
46daff13 | 1496 | qemu_cond_broadcast(&qemu_io_proceeded_cond); |
1a28cac3 | 1497 | } |
afbe7053 | 1498 | iothread_locked = true; |
296af7c9 BS |
1499 | } |
1500 | ||
1501 | void qemu_mutex_unlock_iothread(void) | |
1502 | { | |
afbe7053 | 1503 | iothread_locked = false; |
296af7c9 BS |
1504 | qemu_mutex_unlock(&qemu_global_mutex); |
1505 | } | |
1506 | ||
e8faee06 | 1507 | static bool all_vcpus_paused(void) |
296af7c9 | 1508 | { |
bdc44640 | 1509 | CPUState *cpu; |
296af7c9 | 1510 | |
bdc44640 | 1511 | CPU_FOREACH(cpu) { |
182735ef | 1512 | if (!cpu->stopped) { |
e8faee06 | 1513 | return false; |
0ab07c62 | 1514 | } |
296af7c9 BS |
1515 | } |
1516 | ||
e8faee06 | 1517 | return true; |
296af7c9 BS |
1518 | } |
1519 | ||
1520 | void pause_all_vcpus(void) | |
1521 | { | |
bdc44640 | 1522 | CPUState *cpu; |
296af7c9 | 1523 | |
40daca54 | 1524 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, false); |
bdc44640 | 1525 | CPU_FOREACH(cpu) { |
182735ef AF |
1526 | cpu->stop = true; |
1527 | qemu_cpu_kick(cpu); | |
296af7c9 BS |
1528 | } |
1529 | ||
aa723c23 | 1530 | if (qemu_in_vcpu_thread()) { |
d798e974 JK |
1531 | cpu_stop_current(); |
1532 | if (!kvm_enabled()) { | |
bdc44640 | 1533 | CPU_FOREACH(cpu) { |
182735ef AF |
1534 | cpu->stop = false; |
1535 | cpu->stopped = true; | |
d798e974 JK |
1536 | } |
1537 | return; | |
1538 | } | |
1539 | } | |
1540 | ||
296af7c9 | 1541 | while (!all_vcpus_paused()) { |
be7d6c57 | 1542 | qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex); |
bdc44640 | 1543 | CPU_FOREACH(cpu) { |
182735ef | 1544 | qemu_cpu_kick(cpu); |
296af7c9 BS |
1545 | } |
1546 | } | |
1547 | } | |
1548 | ||
2993683b IM |
1549 | void cpu_resume(CPUState *cpu) |
1550 | { | |
1551 | cpu->stop = false; | |
1552 | cpu->stopped = false; | |
1553 | qemu_cpu_kick(cpu); | |
1554 | } | |
1555 | ||
296af7c9 BS |
1556 | void resume_all_vcpus(void) |
1557 | { | |
bdc44640 | 1558 | CPUState *cpu; |
296af7c9 | 1559 | |
40daca54 | 1560 | qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true); |
bdc44640 | 1561 | CPU_FOREACH(cpu) { |
182735ef | 1562 | cpu_resume(cpu); |
296af7c9 BS |
1563 | } |
1564 | } | |
1565 | ||
4c055ab5 GZ |
1566 | void cpu_remove(CPUState *cpu) |
1567 | { | |
1568 | cpu->stop = true; | |
1569 | cpu->unplug = true; | |
1570 | qemu_cpu_kick(cpu); | |
1571 | } | |
1572 | ||
2c579042 BR |
1573 | void cpu_remove_sync(CPUState *cpu) |
1574 | { | |
1575 | cpu_remove(cpu); | |
1576 | while (cpu->created) { | |
1577 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1578 | } | |
1579 | } | |
1580 | ||
4900116e DDAG |
1581 | /* For temporary buffers for forming a name */ |
1582 | #define VCPU_THREAD_NAME_SIZE 16 | |
1583 | ||
e5ab30a2 | 1584 | static void qemu_tcg_init_vcpu(CPUState *cpu) |
296af7c9 | 1585 | { |
4900116e | 1586 | char thread_name[VCPU_THREAD_NAME_SIZE]; |
d5f8d613 FK |
1587 | static QemuCond *tcg_halt_cond; |
1588 | static QemuThread *tcg_cpu_thread; | |
4900116e | 1589 | |
296af7c9 BS |
1590 | /* share a single thread for all cpus with TCG */ |
1591 | if (!tcg_cpu_thread) { | |
814e612e | 1592 | cpu->thread = g_malloc0(sizeof(QemuThread)); |
f5c121b8 AF |
1593 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); |
1594 | qemu_cond_init(cpu->halt_cond); | |
1595 | tcg_halt_cond = cpu->halt_cond; | |
4900116e DDAG |
1596 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG", |
1597 | cpu->cpu_index); | |
1598 | qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn, | |
1599 | cpu, QEMU_THREAD_JOINABLE); | |
1ecf47bf | 1600 | #ifdef _WIN32 |
814e612e | 1601 | cpu->hThread = qemu_thread_get_handle(cpu->thread); |
1ecf47bf | 1602 | #endif |
61a46217 | 1603 | while (!cpu->created) { |
18a85728 | 1604 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); |
0ab07c62 | 1605 | } |
814e612e | 1606 | tcg_cpu_thread = cpu->thread; |
296af7c9 | 1607 | } else { |
814e612e | 1608 | cpu->thread = tcg_cpu_thread; |
f5c121b8 | 1609 | cpu->halt_cond = tcg_halt_cond; |
296af7c9 BS |
1610 | } |
1611 | } | |
1612 | ||
b0cb0a66 VP |
1613 | static void qemu_hax_start_vcpu(CPUState *cpu) |
1614 | { | |
1615 | char thread_name[VCPU_THREAD_NAME_SIZE]; | |
1616 | ||
1617 | cpu->thread = g_malloc0(sizeof(QemuThread)); | |
1618 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); | |
1619 | qemu_cond_init(cpu->halt_cond); | |
1620 | ||
1621 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/HAX", | |
1622 | cpu->cpu_index); | |
1623 | qemu_thread_create(cpu->thread, thread_name, qemu_hax_cpu_thread_fn, | |
1624 | cpu, QEMU_THREAD_JOINABLE); | |
1625 | #ifdef _WIN32 | |
1626 | cpu->hThread = qemu_thread_get_handle(cpu->thread); | |
1627 | #endif | |
1628 | while (!cpu->created) { | |
1629 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); | |
1630 | } | |
1631 | } | |
1632 | ||
48a106bd | 1633 | static void qemu_kvm_start_vcpu(CPUState *cpu) |
296af7c9 | 1634 | { |
4900116e DDAG |
1635 | char thread_name[VCPU_THREAD_NAME_SIZE]; |
1636 | ||
814e612e | 1637 | cpu->thread = g_malloc0(sizeof(QemuThread)); |
f5c121b8 AF |
1638 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); |
1639 | qemu_cond_init(cpu->halt_cond); | |
4900116e DDAG |
1640 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/KVM", |
1641 | cpu->cpu_index); | |
1642 | qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn, | |
1643 | cpu, QEMU_THREAD_JOINABLE); | |
61a46217 | 1644 | while (!cpu->created) { |
18a85728 | 1645 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); |
0ab07c62 | 1646 | } |
296af7c9 BS |
1647 | } |
1648 | ||
10a9021d | 1649 | static void qemu_dummy_start_vcpu(CPUState *cpu) |
c7f0f3b1 | 1650 | { |
4900116e DDAG |
1651 | char thread_name[VCPU_THREAD_NAME_SIZE]; |
1652 | ||
814e612e | 1653 | cpu->thread = g_malloc0(sizeof(QemuThread)); |
f5c121b8 AF |
1654 | cpu->halt_cond = g_malloc0(sizeof(QemuCond)); |
1655 | qemu_cond_init(cpu->halt_cond); | |
4900116e DDAG |
1656 | snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/DUMMY", |
1657 | cpu->cpu_index); | |
1658 | qemu_thread_create(cpu->thread, thread_name, qemu_dummy_cpu_thread_fn, cpu, | |
c7f0f3b1 | 1659 | QEMU_THREAD_JOINABLE); |
61a46217 | 1660 | while (!cpu->created) { |
c7f0f3b1 AL |
1661 | qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); |
1662 | } | |
1663 | } | |
1664 | ||
c643bed9 | 1665 | void qemu_init_vcpu(CPUState *cpu) |
296af7c9 | 1666 | { |
ce3960eb AF |
1667 | cpu->nr_cores = smp_cores; |
1668 | cpu->nr_threads = smp_threads; | |
f324e766 | 1669 | cpu->stopped = true; |
56943e8c PM |
1670 | |
1671 | if (!cpu->as) { | |
1672 | /* If the target cpu hasn't set up any address spaces itself, | |
1673 | * give it the default one. | |
1674 | */ | |
6731d864 PC |
1675 | AddressSpace *as = address_space_init_shareable(cpu->memory, |
1676 | "cpu-memory"); | |
12ebc9a7 | 1677 | cpu->num_ases = 1; |
6731d864 | 1678 | cpu_address_space_init(cpu, as, 0); |
56943e8c PM |
1679 | } |
1680 | ||
0ab07c62 | 1681 | if (kvm_enabled()) { |
48a106bd | 1682 | qemu_kvm_start_vcpu(cpu); |
b0cb0a66 VP |
1683 | } else if (hax_enabled()) { |
1684 | qemu_hax_start_vcpu(cpu); | |
c7f0f3b1 | 1685 | } else if (tcg_enabled()) { |
e5ab30a2 | 1686 | qemu_tcg_init_vcpu(cpu); |
c7f0f3b1 | 1687 | } else { |
10a9021d | 1688 | qemu_dummy_start_vcpu(cpu); |
0ab07c62 | 1689 | } |
296af7c9 BS |
1690 | } |
1691 | ||
b4a3d965 | 1692 | void cpu_stop_current(void) |
296af7c9 | 1693 | { |
4917cf44 AF |
1694 | if (current_cpu) { |
1695 | current_cpu->stop = false; | |
1696 | current_cpu->stopped = true; | |
1697 | cpu_exit(current_cpu); | |
96bce683 | 1698 | qemu_cond_broadcast(&qemu_pause_cond); |
b4a3d965 | 1699 | } |
296af7c9 BS |
1700 | } |
1701 | ||
56983463 | 1702 | int vm_stop(RunState state) |
296af7c9 | 1703 | { |
aa723c23 | 1704 | if (qemu_in_vcpu_thread()) { |
74892d24 | 1705 | qemu_system_vmstop_request_prepare(); |
1dfb4dd9 | 1706 | qemu_system_vmstop_request(state); |
296af7c9 BS |
1707 | /* |
1708 | * FIXME: should not return to device code in case | |
1709 | * vm_stop() has been requested. | |
1710 | */ | |
b4a3d965 | 1711 | cpu_stop_current(); |
56983463 | 1712 | return 0; |
296af7c9 | 1713 | } |
56983463 KW |
1714 | |
1715 | return do_vm_stop(state); | |
296af7c9 BS |
1716 | } |
1717 | ||
2d76e823 CI |
1718 | /** |
1719 | * Prepare for (re)starting the VM. | |
1720 | * Returns -1 if the vCPUs are not to be restarted (e.g. if they are already | |
1721 | * running or in case of an error condition), 0 otherwise. | |
1722 | */ | |
1723 | int vm_prepare_start(void) | |
1724 | { | |
1725 | RunState requested; | |
1726 | int res = 0; | |
1727 | ||
1728 | qemu_vmstop_requested(&requested); | |
1729 | if (runstate_is_running() && requested == RUN_STATE__MAX) { | |
1730 | return -1; | |
1731 | } | |
1732 | ||
1733 | /* Ensure that a STOP/RESUME pair of events is emitted if a | |
1734 | * vmstop request was pending. The BLOCK_IO_ERROR event, for | |
1735 | * example, according to documentation is always followed by | |
1736 | * the STOP event. | |
1737 | */ | |
1738 | if (runstate_is_running()) { | |
1739 | qapi_event_send_stop(&error_abort); | |
1740 | res = -1; | |
1741 | } else { | |
1742 | replay_enable_events(); | |
1743 | cpu_enable_ticks(); | |
1744 | runstate_set(RUN_STATE_RUNNING); | |
1745 | vm_state_notify(1, RUN_STATE_RUNNING); | |
1746 | } | |
1747 | ||
1748 | /* We are sending this now, but the CPUs will be resumed shortly later */ | |
1749 | qapi_event_send_resume(&error_abort); | |
1750 | return res; | |
1751 | } | |
1752 | ||
1753 | void vm_start(void) | |
1754 | { | |
1755 | if (!vm_prepare_start()) { | |
1756 | resume_all_vcpus(); | |
1757 | } | |
1758 | } | |
1759 | ||
8a9236f1 LC |
1760 | /* does a state transition even if the VM is already stopped, |
1761 | current state is forgotten forever */ | |
56983463 | 1762 | int vm_stop_force_state(RunState state) |
8a9236f1 LC |
1763 | { |
1764 | if (runstate_is_running()) { | |
56983463 | 1765 | return vm_stop(state); |
8a9236f1 LC |
1766 | } else { |
1767 | runstate_set(state); | |
b2780d32 WC |
1768 | |
1769 | bdrv_drain_all(); | |
594a45ce KW |
1770 | /* Make sure to return an error if the flush in a previous vm_stop() |
1771 | * failed. */ | |
22af08ea | 1772 | return bdrv_flush_all(); |
8a9236f1 LC |
1773 | } |
1774 | } | |
1775 | ||
9a78eead | 1776 | void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg) |
262353cb BS |
1777 | { |
1778 | /* XXX: implement xxx_cpu_list for targets that still miss it */ | |
e916cbf8 PM |
1779 | #if defined(cpu_list) |
1780 | cpu_list(f, cpu_fprintf); | |
262353cb BS |
1781 | #endif |
1782 | } | |
de0b36b6 LC |
1783 | |
1784 | CpuInfoList *qmp_query_cpus(Error **errp) | |
1785 | { | |
1786 | CpuInfoList *head = NULL, *cur_item = NULL; | |
182735ef | 1787 | CPUState *cpu; |
de0b36b6 | 1788 | |
bdc44640 | 1789 | CPU_FOREACH(cpu) { |
de0b36b6 | 1790 | CpuInfoList *info; |
182735ef AF |
1791 | #if defined(TARGET_I386) |
1792 | X86CPU *x86_cpu = X86_CPU(cpu); | |
1793 | CPUX86State *env = &x86_cpu->env; | |
1794 | #elif defined(TARGET_PPC) | |
1795 | PowerPCCPU *ppc_cpu = POWERPC_CPU(cpu); | |
1796 | CPUPPCState *env = &ppc_cpu->env; | |
1797 | #elif defined(TARGET_SPARC) | |
1798 | SPARCCPU *sparc_cpu = SPARC_CPU(cpu); | |
1799 | CPUSPARCState *env = &sparc_cpu->env; | |
1800 | #elif defined(TARGET_MIPS) | |
1801 | MIPSCPU *mips_cpu = MIPS_CPU(cpu); | |
1802 | CPUMIPSState *env = &mips_cpu->env; | |
48e06fe0 BK |
1803 | #elif defined(TARGET_TRICORE) |
1804 | TriCoreCPU *tricore_cpu = TRICORE_CPU(cpu); | |
1805 | CPUTriCoreState *env = &tricore_cpu->env; | |
182735ef | 1806 | #endif |
de0b36b6 | 1807 | |
cb446eca | 1808 | cpu_synchronize_state(cpu); |
de0b36b6 LC |
1809 | |
1810 | info = g_malloc0(sizeof(*info)); | |
1811 | info->value = g_malloc0(sizeof(*info->value)); | |
55e5c285 | 1812 | info->value->CPU = cpu->cpu_index; |
182735ef | 1813 | info->value->current = (cpu == first_cpu); |
259186a7 | 1814 | info->value->halted = cpu->halted; |
58f88d4b | 1815 | info->value->qom_path = object_get_canonical_path(OBJECT(cpu)); |
9f09e18a | 1816 | info->value->thread_id = cpu->thread_id; |
de0b36b6 | 1817 | #if defined(TARGET_I386) |
86f4b687 | 1818 | info->value->arch = CPU_INFO_ARCH_X86; |
544a3731 | 1819 | info->value->u.x86.pc = env->eip + env->segs[R_CS].base; |
de0b36b6 | 1820 | #elif defined(TARGET_PPC) |
86f4b687 | 1821 | info->value->arch = CPU_INFO_ARCH_PPC; |
544a3731 | 1822 | info->value->u.ppc.nip = env->nip; |
de0b36b6 | 1823 | #elif defined(TARGET_SPARC) |
86f4b687 | 1824 | info->value->arch = CPU_INFO_ARCH_SPARC; |
544a3731 EB |
1825 | info->value->u.q_sparc.pc = env->pc; |
1826 | info->value->u.q_sparc.npc = env->npc; | |
de0b36b6 | 1827 | #elif defined(TARGET_MIPS) |
86f4b687 | 1828 | info->value->arch = CPU_INFO_ARCH_MIPS; |
544a3731 | 1829 | info->value->u.q_mips.PC = env->active_tc.PC; |
48e06fe0 | 1830 | #elif defined(TARGET_TRICORE) |
86f4b687 | 1831 | info->value->arch = CPU_INFO_ARCH_TRICORE; |
544a3731 | 1832 | info->value->u.tricore.PC = env->PC; |
86f4b687 EB |
1833 | #else |
1834 | info->value->arch = CPU_INFO_ARCH_OTHER; | |
de0b36b6 LC |
1835 | #endif |
1836 | ||
1837 | /* XXX: waiting for the qapi to support GSList */ | |
1838 | if (!cur_item) { | |
1839 | head = cur_item = info; | |
1840 | } else { | |
1841 | cur_item->next = info; | |
1842 | cur_item = info; | |
1843 | } | |
1844 | } | |
1845 | ||
1846 | return head; | |
1847 | } | |
0cfd6a9a LC |
1848 | |
1849 | void qmp_memsave(int64_t addr, int64_t size, const char *filename, | |
1850 | bool has_cpu, int64_t cpu_index, Error **errp) | |
1851 | { | |
1852 | FILE *f; | |
1853 | uint32_t l; | |
55e5c285 | 1854 | CPUState *cpu; |
0cfd6a9a | 1855 | uint8_t buf[1024]; |
0dc9daf0 | 1856 | int64_t orig_addr = addr, orig_size = size; |
0cfd6a9a LC |
1857 | |
1858 | if (!has_cpu) { | |
1859 | cpu_index = 0; | |
1860 | } | |
1861 | ||
151d1322 AF |
1862 | cpu = qemu_get_cpu(cpu_index); |
1863 | if (cpu == NULL) { | |
c6bd8c70 MA |
1864 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index", |
1865 | "a CPU number"); | |
0cfd6a9a LC |
1866 | return; |
1867 | } | |
1868 | ||
1869 | f = fopen(filename, "wb"); | |
1870 | if (!f) { | |
618da851 | 1871 | error_setg_file_open(errp, errno, filename); |
0cfd6a9a LC |
1872 | return; |
1873 | } | |
1874 | ||
1875 | while (size != 0) { | |
1876 | l = sizeof(buf); | |
1877 | if (l > size) | |
1878 | l = size; | |
2f4d0f59 | 1879 | if (cpu_memory_rw_debug(cpu, addr, buf, l, 0) != 0) { |
0dc9daf0 BP |
1880 | error_setg(errp, "Invalid addr 0x%016" PRIx64 "/size %" PRId64 |
1881 | " specified", orig_addr, orig_size); | |
2f4d0f59 AK |
1882 | goto exit; |
1883 | } | |
0cfd6a9a | 1884 | if (fwrite(buf, 1, l, f) != l) { |
c6bd8c70 | 1885 | error_setg(errp, QERR_IO_ERROR); |
0cfd6a9a LC |
1886 | goto exit; |
1887 | } | |
1888 | addr += l; | |
1889 | size -= l; | |
1890 | } | |
1891 | ||
1892 | exit: | |
1893 | fclose(f); | |
1894 | } | |
6d3962bf LC |
1895 | |
1896 | void qmp_pmemsave(int64_t addr, int64_t size, const char *filename, | |
1897 | Error **errp) | |
1898 | { | |
1899 | FILE *f; | |
1900 | uint32_t l; | |
1901 | uint8_t buf[1024]; | |
1902 | ||
1903 | f = fopen(filename, "wb"); | |
1904 | if (!f) { | |
618da851 | 1905 | error_setg_file_open(errp, errno, filename); |
6d3962bf LC |
1906 | return; |
1907 | } | |
1908 | ||
1909 | while (size != 0) { | |
1910 | l = sizeof(buf); | |
1911 | if (l > size) | |
1912 | l = size; | |
eb6282f2 | 1913 | cpu_physical_memory_read(addr, buf, l); |
6d3962bf | 1914 | if (fwrite(buf, 1, l, f) != l) { |
c6bd8c70 | 1915 | error_setg(errp, QERR_IO_ERROR); |
6d3962bf LC |
1916 | goto exit; |
1917 | } | |
1918 | addr += l; | |
1919 | size -= l; | |
1920 | } | |
1921 | ||
1922 | exit: | |
1923 | fclose(f); | |
1924 | } | |
ab49ab5c LC |
1925 | |
1926 | void qmp_inject_nmi(Error **errp) | |
1927 | { | |
9cb805fd | 1928 | nmi_monitor_handle(monitor_get_cpu_index(), errp); |
ab49ab5c | 1929 | } |
27498bef ST |
1930 | |
1931 | void dump_drift_info(FILE *f, fprintf_function cpu_fprintf) | |
1932 | { | |
1933 | if (!use_icount) { | |
1934 | return; | |
1935 | } | |
1936 | ||
1937 | cpu_fprintf(f, "Host - Guest clock %"PRIi64" ms\n", | |
1938 | (cpu_get_clock() - cpu_get_icount())/SCALE_MS); | |
1939 | if (icount_align_option) { | |
1940 | cpu_fprintf(f, "Max guest delay %"PRIi64" ms\n", -max_delay/SCALE_MS); | |
1941 | cpu_fprintf(f, "Max guest advance %"PRIi64" ms\n", max_advance/SCALE_MS); | |
1942 | } else { | |
1943 | cpu_fprintf(f, "Max guest delay NA\n"); | |
1944 | cpu_fprintf(f, "Max guest advance NA\n"); | |
1945 | } | |
1946 | } |