block: User BdrvChild callback for device name
[qemu.git] / cpu-exec.c
1 /*
2 * emulator main execution loop
3 *
4 * Copyright (c) 2003-2005 Fabrice Bellard
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
18 */
19 #include "qemu/osdep.h"
20 #include "cpu.h"
21 #include "trace.h"
22 #include "disas/disas.h"
23 #include "tcg.h"
24 #include "qemu/atomic.h"
25 #include "sysemu/qtest.h"
26 #include "qemu/timer.h"
27 #include "exec/address-spaces.h"
28 #include "qemu/rcu.h"
29 #include "exec/tb-hash.h"
30 #include "exec/log.h"
31 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
32 #include "hw/i386/apic.h"
33 #endif
34 #include "sysemu/replay.h"
35
36 /* -icount align implementation. */
37
38 typedef struct SyncClocks {
39 int64_t diff_clk;
40 int64_t last_cpu_icount;
41 int64_t realtime_clock;
42 } SyncClocks;
43
44 #if !defined(CONFIG_USER_ONLY)
45 /* Allow the guest to have a max 3ms advance.
46 * The difference between the 2 clocks could therefore
47 * oscillate around 0.
48 */
49 #define VM_CLOCK_ADVANCE 3000000
50 #define THRESHOLD_REDUCE 1.5
51 #define MAX_DELAY_PRINT_RATE 2000000000LL
52 #define MAX_NB_PRINTS 100
53
54 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
55 {
56 int64_t cpu_icount;
57
58 if (!icount_align_option) {
59 return;
60 }
61
62 cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
63 sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
64 sc->last_cpu_icount = cpu_icount;
65
66 if (sc->diff_clk > VM_CLOCK_ADVANCE) {
67 #ifndef _WIN32
68 struct timespec sleep_delay, rem_delay;
69 sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
70 sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
71 if (nanosleep(&sleep_delay, &rem_delay) < 0) {
72 sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
73 } else {
74 sc->diff_clk = 0;
75 }
76 #else
77 Sleep(sc->diff_clk / SCALE_MS);
78 sc->diff_clk = 0;
79 #endif
80 }
81 }
82
83 static void print_delay(const SyncClocks *sc)
84 {
85 static float threshold_delay;
86 static int64_t last_realtime_clock;
87 static int nb_prints;
88
89 if (icount_align_option &&
90 sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
91 nb_prints < MAX_NB_PRINTS) {
92 if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
93 (-sc->diff_clk / (float)1000000000LL <
94 (threshold_delay - THRESHOLD_REDUCE))) {
95 threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
96 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
97 threshold_delay - 1,
98 threshold_delay);
99 nb_prints++;
100 last_realtime_clock = sc->realtime_clock;
101 }
102 }
103 }
104
105 static void init_delay_params(SyncClocks *sc,
106 const CPUState *cpu)
107 {
108 if (!icount_align_option) {
109 return;
110 }
111 sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
112 sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
113 sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
114 if (sc->diff_clk < max_delay) {
115 max_delay = sc->diff_clk;
116 }
117 if (sc->diff_clk > max_advance) {
118 max_advance = sc->diff_clk;
119 }
120
121 /* Print every 2s max if the guest is late. We limit the number
122 of printed messages to NB_PRINT_MAX(currently 100) */
123 print_delay(sc);
124 }
125 #else
126 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
127 {
128 }
129
130 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
131 {
132 }
133 #endif /* CONFIG USER ONLY */
134
135 /* Execute a TB, and fix up the CPU state afterwards if necessary */
136 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, TranslationBlock *itb)
137 {
138 CPUArchState *env = cpu->env_ptr;
139 uintptr_t ret;
140 TranslationBlock *last_tb;
141 int tb_exit;
142 uint8_t *tb_ptr = itb->tc_ptr;
143
144 qemu_log_mask_and_addr(CPU_LOG_EXEC, itb->pc,
145 "Trace %p [" TARGET_FMT_lx "] %s\n",
146 itb->tc_ptr, itb->pc, lookup_symbol(itb->pc));
147
148 #if defined(DEBUG_DISAS)
149 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
150 #if defined(TARGET_I386)
151 log_cpu_state(cpu, CPU_DUMP_CCOP);
152 #elif defined(TARGET_M68K)
153 /* ??? Should not modify env state for dumping. */
154 cpu_m68k_flush_flags(env, env->cc_op);
155 env->cc_op = CC_OP_FLAGS;
156 env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
157 log_cpu_state(cpu, 0);
158 #else
159 log_cpu_state(cpu, 0);
160 #endif
161 }
162 #endif /* DEBUG_DISAS */
163
164 cpu->can_do_io = !use_icount;
165 ret = tcg_qemu_tb_exec(env, tb_ptr);
166 cpu->can_do_io = 1;
167 last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
168 tb_exit = ret & TB_EXIT_MASK;
169 trace_exec_tb_exit(last_tb, tb_exit);
170
171 if (tb_exit > TB_EXIT_IDX1) {
172 /* We didn't start executing this TB (eg because the instruction
173 * counter hit zero); we must restore the guest PC to the address
174 * of the start of the TB.
175 */
176 CPUClass *cc = CPU_GET_CLASS(cpu);
177 qemu_log_mask_and_addr(CPU_LOG_EXEC, last_tb->pc,
178 "Stopped execution of TB chain before %p ["
179 TARGET_FMT_lx "] %s\n",
180 last_tb->tc_ptr, last_tb->pc,
181 lookup_symbol(last_tb->pc));
182 if (cc->synchronize_from_tb) {
183 cc->synchronize_from_tb(cpu, last_tb);
184 } else {
185 assert(cc->set_pc);
186 cc->set_pc(cpu, last_tb->pc);
187 }
188 }
189 if (tb_exit == TB_EXIT_REQUESTED) {
190 /* We were asked to stop executing TBs (probably a pending
191 * interrupt. We've now stopped, so clear the flag.
192 */
193 cpu->tcg_exit_req = 0;
194 }
195 return ret;
196 }
197
198 #ifndef CONFIG_USER_ONLY
199 /* Execute the code without caching the generated code. An interpreter
200 could be used if available. */
201 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
202 TranslationBlock *orig_tb, bool ignore_icount)
203 {
204 TranslationBlock *tb;
205 bool old_tb_flushed;
206
207 /* Should never happen.
208 We only end up here when an existing TB is too long. */
209 if (max_cycles > CF_COUNT_MASK)
210 max_cycles = CF_COUNT_MASK;
211
212 old_tb_flushed = cpu->tb_flushed;
213 cpu->tb_flushed = false;
214 tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
215 max_cycles | CF_NOCACHE
216 | (ignore_icount ? CF_IGNORE_ICOUNT : 0));
217 tb->orig_tb = cpu->tb_flushed ? NULL : orig_tb;
218 cpu->tb_flushed |= old_tb_flushed;
219 /* execute the generated code */
220 trace_exec_tb_nocache(tb, tb->pc);
221 cpu_tb_exec(cpu, tb);
222 tb_phys_invalidate(tb, -1);
223 tb_free(tb);
224 }
225 #endif
226
227 static TranslationBlock *tb_find_physical(CPUState *cpu,
228 target_ulong pc,
229 target_ulong cs_base,
230 uint32_t flags)
231 {
232 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
233 TranslationBlock *tb, **tb_hash_head, **ptb1;
234 unsigned int h;
235 tb_page_addr_t phys_pc, phys_page1;
236
237 /* find translated block using physical mappings */
238 phys_pc = get_page_addr_code(env, pc);
239 phys_page1 = phys_pc & TARGET_PAGE_MASK;
240 h = tb_phys_hash_func(phys_pc);
241
242 /* Start at head of the hash entry */
243 ptb1 = tb_hash_head = &tcg_ctx.tb_ctx.tb_phys_hash[h];
244 tb = *ptb1;
245
246 while (tb) {
247 if (tb->pc == pc &&
248 tb->page_addr[0] == phys_page1 &&
249 tb->cs_base == cs_base &&
250 tb->flags == flags) {
251
252 if (tb->page_addr[1] == -1) {
253 /* done, we have a match */
254 break;
255 } else {
256 /* check next page if needed */
257 target_ulong virt_page2 = (pc & TARGET_PAGE_MASK) +
258 TARGET_PAGE_SIZE;
259 tb_page_addr_t phys_page2 = get_page_addr_code(env, virt_page2);
260
261 if (tb->page_addr[1] == phys_page2) {
262 break;
263 }
264 }
265 }
266
267 ptb1 = &tb->phys_hash_next;
268 tb = *ptb1;
269 }
270
271 if (tb) {
272 /* Move the TB to the head of the list */
273 *ptb1 = tb->phys_hash_next;
274 tb->phys_hash_next = *tb_hash_head;
275 *tb_hash_head = tb;
276 }
277 return tb;
278 }
279
280 static TranslationBlock *tb_find_slow(CPUState *cpu,
281 target_ulong pc,
282 target_ulong cs_base,
283 uint32_t flags)
284 {
285 TranslationBlock *tb;
286
287 tb = tb_find_physical(cpu, pc, cs_base, flags);
288 if (tb) {
289 goto found;
290 }
291
292 #ifdef CONFIG_USER_ONLY
293 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
294 * taken outside tb_lock. Since we're momentarily dropping
295 * tb_lock, there's a chance that our desired tb has been
296 * translated.
297 */
298 tb_unlock();
299 mmap_lock();
300 tb_lock();
301 tb = tb_find_physical(cpu, pc, cs_base, flags);
302 if (tb) {
303 mmap_unlock();
304 goto found;
305 }
306 #endif
307
308 /* if no translated code available, then translate it now */
309 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
310
311 #ifdef CONFIG_USER_ONLY
312 mmap_unlock();
313 #endif
314
315 found:
316 /* we add the TB in the virtual pc hash table */
317 cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
318 return tb;
319 }
320
321 static inline TranslationBlock *tb_find_fast(CPUState *cpu,
322 TranslationBlock **last_tb,
323 int tb_exit)
324 {
325 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
326 TranslationBlock *tb;
327 target_ulong cs_base, pc;
328 uint32_t flags;
329
330 /* we record a subset of the CPU state. It will
331 always be the same before a given translated block
332 is executed. */
333 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
334 tb_lock();
335 tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
336 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
337 tb->flags != flags)) {
338 tb = tb_find_slow(cpu, pc, cs_base, flags);
339 }
340 if (cpu->tb_flushed) {
341 /* Ensure that no TB jump will be modified as the
342 * translation buffer has been flushed.
343 */
344 *last_tb = NULL;
345 cpu->tb_flushed = false;
346 }
347 /* See if we can patch the calling TB. */
348 if (*last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
349 tb_add_jump(*last_tb, tb_exit, tb);
350 }
351 tb_unlock();
352 return tb;
353 }
354
355 static inline bool cpu_handle_halt(CPUState *cpu)
356 {
357 if (cpu->halted) {
358 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
359 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
360 && replay_interrupt()) {
361 X86CPU *x86_cpu = X86_CPU(cpu);
362 apic_poll_irq(x86_cpu->apic_state);
363 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
364 }
365 #endif
366 if (!cpu_has_work(cpu)) {
367 current_cpu = NULL;
368 return true;
369 }
370
371 cpu->halted = 0;
372 }
373
374 return false;
375 }
376
377 static inline void cpu_handle_debug_exception(CPUState *cpu)
378 {
379 CPUClass *cc = CPU_GET_CLASS(cpu);
380 CPUWatchpoint *wp;
381
382 if (!cpu->watchpoint_hit) {
383 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
384 wp->flags &= ~BP_WATCHPOINT_HIT;
385 }
386 }
387
388 cc->debug_excp_handler(cpu);
389 }
390
391 static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
392 {
393 if (cpu->exception_index >= 0) {
394 if (cpu->exception_index >= EXCP_INTERRUPT) {
395 /* exit request from the cpu execution loop */
396 *ret = cpu->exception_index;
397 if (*ret == EXCP_DEBUG) {
398 cpu_handle_debug_exception(cpu);
399 }
400 cpu->exception_index = -1;
401 return true;
402 } else {
403 #if defined(CONFIG_USER_ONLY)
404 /* if user mode only, we simulate a fake exception
405 which will be handled outside the cpu execution
406 loop */
407 #if defined(TARGET_I386)
408 CPUClass *cc = CPU_GET_CLASS(cpu);
409 cc->do_interrupt(cpu);
410 #endif
411 *ret = cpu->exception_index;
412 cpu->exception_index = -1;
413 return true;
414 #else
415 if (replay_exception()) {
416 CPUClass *cc = CPU_GET_CLASS(cpu);
417 cc->do_interrupt(cpu);
418 cpu->exception_index = -1;
419 } else if (!replay_has_interrupt()) {
420 /* give a chance to iothread in replay mode */
421 *ret = EXCP_INTERRUPT;
422 return true;
423 }
424 #endif
425 }
426 #ifndef CONFIG_USER_ONLY
427 } else if (replay_has_exception()
428 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
429 /* try to cause an exception pending in the log */
430 TranslationBlock *last_tb = NULL; /* Avoid chaining TBs */
431 cpu_exec_nocache(cpu, 1, tb_find_fast(cpu, &last_tb, 0), true);
432 *ret = -1;
433 return true;
434 #endif
435 }
436
437 return false;
438 }
439
440 static inline void cpu_handle_interrupt(CPUState *cpu,
441 TranslationBlock **last_tb)
442 {
443 CPUClass *cc = CPU_GET_CLASS(cpu);
444 int interrupt_request = cpu->interrupt_request;
445
446 if (unlikely(interrupt_request)) {
447 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
448 /* Mask out external interrupts for this step. */
449 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
450 }
451 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
452 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
453 cpu->exception_index = EXCP_DEBUG;
454 cpu_loop_exit(cpu);
455 }
456 if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
457 /* Do nothing */
458 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
459 replay_interrupt();
460 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
461 cpu->halted = 1;
462 cpu->exception_index = EXCP_HLT;
463 cpu_loop_exit(cpu);
464 }
465 #if defined(TARGET_I386)
466 else if (interrupt_request & CPU_INTERRUPT_INIT) {
467 X86CPU *x86_cpu = X86_CPU(cpu);
468 CPUArchState *env = &x86_cpu->env;
469 replay_interrupt();
470 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
471 do_cpu_init(x86_cpu);
472 cpu->exception_index = EXCP_HALTED;
473 cpu_loop_exit(cpu);
474 }
475 #else
476 else if (interrupt_request & CPU_INTERRUPT_RESET) {
477 replay_interrupt();
478 cpu_reset(cpu);
479 cpu_loop_exit(cpu);
480 }
481 #endif
482 /* The target hook has 3 exit conditions:
483 False when the interrupt isn't processed,
484 True when it is, and we should restart on a new TB,
485 and via longjmp via cpu_loop_exit. */
486 else {
487 replay_interrupt();
488 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
489 *last_tb = NULL;
490 }
491 /* The target hook may have updated the 'cpu->interrupt_request';
492 * reload the 'interrupt_request' value */
493 interrupt_request = cpu->interrupt_request;
494 }
495 if (interrupt_request & CPU_INTERRUPT_EXITTB) {
496 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
497 /* ensure that no TB jump will be modified as
498 the program flow was changed */
499 *last_tb = NULL;
500 }
501 }
502 if (unlikely(cpu->exit_request || replay_has_interrupt())) {
503 cpu->exit_request = 0;
504 cpu->exception_index = EXCP_INTERRUPT;
505 cpu_loop_exit(cpu);
506 }
507 }
508
509 static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
510 TranslationBlock **last_tb, int *tb_exit,
511 SyncClocks *sc)
512 {
513 uintptr_t ret;
514
515 if (unlikely(cpu->exit_request)) {
516 return;
517 }
518
519 trace_exec_tb(tb, tb->pc);
520 ret = cpu_tb_exec(cpu, tb);
521 *last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
522 *tb_exit = ret & TB_EXIT_MASK;
523 switch (*tb_exit) {
524 case TB_EXIT_REQUESTED:
525 /* Something asked us to stop executing
526 * chained TBs; just continue round the main
527 * loop. Whatever requested the exit will also
528 * have set something else (eg exit_request or
529 * interrupt_request) which we will handle
530 * next time around the loop. But we need to
531 * ensure the tcg_exit_req read in generated code
532 * comes before the next read of cpu->exit_request
533 * or cpu->interrupt_request.
534 */
535 smp_rmb();
536 *last_tb = NULL;
537 break;
538 case TB_EXIT_ICOUNT_EXPIRED:
539 {
540 /* Instruction counter expired. */
541 #ifdef CONFIG_USER_ONLY
542 abort();
543 #else
544 int insns_left = cpu->icount_decr.u32;
545 if (cpu->icount_extra && insns_left >= 0) {
546 /* Refill decrementer and continue execution. */
547 cpu->icount_extra += insns_left;
548 insns_left = MIN(0xffff, cpu->icount_extra);
549 cpu->icount_extra -= insns_left;
550 cpu->icount_decr.u16.low = insns_left;
551 } else {
552 if (insns_left > 0) {
553 /* Execute remaining instructions. */
554 cpu_exec_nocache(cpu, insns_left, *last_tb, false);
555 align_clocks(sc, cpu);
556 }
557 cpu->exception_index = EXCP_INTERRUPT;
558 *last_tb = NULL;
559 cpu_loop_exit(cpu);
560 }
561 break;
562 #endif
563 }
564 default:
565 break;
566 }
567 }
568
569 /* main execution loop */
570
571 int cpu_exec(CPUState *cpu)
572 {
573 CPUClass *cc = CPU_GET_CLASS(cpu);
574 int ret;
575 SyncClocks sc;
576
577 /* replay_interrupt may need current_cpu */
578 current_cpu = cpu;
579
580 if (cpu_handle_halt(cpu)) {
581 return EXCP_HALTED;
582 }
583
584 atomic_mb_set(&tcg_current_cpu, cpu);
585 rcu_read_lock();
586
587 if (unlikely(atomic_mb_read(&exit_request))) {
588 cpu->exit_request = 1;
589 }
590
591 cc->cpu_exec_enter(cpu);
592
593 /* Calculate difference between guest clock and host clock.
594 * This delay includes the delay of the last cycle, so
595 * what we have to do is sleep until it is 0. As for the
596 * advance/delay we gain here, we try to fix it next time.
597 */
598 init_delay_params(&sc, cpu);
599
600 for(;;) {
601 TranslationBlock *tb, *last_tb;
602 int tb_exit = 0;
603
604 /* prepare setjmp context for exception handling */
605 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
606 /* if an exception is pending, we execute it here */
607 if (cpu_handle_exception(cpu, &ret)) {
608 break;
609 }
610
611 last_tb = NULL; /* forget the last executed TB after exception */
612 cpu->tb_flushed = false; /* reset before first TB lookup */
613 for(;;) {
614 cpu_handle_interrupt(cpu, &last_tb);
615 tb = tb_find_fast(cpu, &last_tb, tb_exit);
616 cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc);
617 /* Try to align the host and virtual clocks
618 if the guest is in advance */
619 align_clocks(&sc, cpu);
620 } /* for(;;) */
621 } else {
622 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
623 /* Some compilers wrongly smash all local variables after
624 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
625 * Reload essential local variables here for those compilers.
626 * Newer versions of gcc would complain about this code (-Wclobbered). */
627 cpu = current_cpu;
628 cc = CPU_GET_CLASS(cpu);
629 #else /* buggy compiler */
630 /* Assert that the compiler does not smash local variables. */
631 g_assert(cpu == current_cpu);
632 g_assert(cc == CPU_GET_CLASS(cpu));
633 #endif /* buggy compiler */
634 cpu->can_do_io = 1;
635 tb_lock_reset();
636 }
637 } /* for(;;) */
638
639 cc->cpu_exec_exit(cpu);
640 rcu_read_unlock();
641
642 /* fail safe : never use current_cpu outside cpu_exec() */
643 current_cpu = NULL;
644
645 /* Does not need atomic_mb_set because a spurious wakeup is okay. */
646 atomic_set(&tcg_current_cpu, NULL);
647 return ret;
648 }