2 * emulator main execution loop
4 * Copyright (c) 2003-2005 Fabrice Bellard
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.
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.
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/>.
19 #include "qemu/osdep.h"
21 #include "trace-root.h"
22 #include "disas/disas.h"
23 #include "exec/exec-all.h"
25 #include "qemu/atomic.h"
26 #include "sysemu/qtest.h"
27 #include "qemu/timer.h"
28 #include "exec/address-spaces.h"
30 #include "exec/tb-hash.h"
32 #include "qemu/main-loop.h"
33 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
34 #include "hw/i386/apic.h"
36 #include "sysemu/replay.h"
38 /* -icount align implementation. */
40 typedef struct SyncClocks {
42 int64_t last_cpu_icount;
43 int64_t realtime_clock;
46 #if !defined(CONFIG_USER_ONLY)
47 /* Allow the guest to have a max 3ms advance.
48 * The difference between the 2 clocks could therefore
51 #define VM_CLOCK_ADVANCE 3000000
52 #define THRESHOLD_REDUCE 1.5
53 #define MAX_DELAY_PRINT_RATE 2000000000LL
54 #define MAX_NB_PRINTS 100
56 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
60 if (!icount_align_option) {
64 cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
65 sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
66 sc->last_cpu_icount = cpu_icount;
68 if (sc->diff_clk > VM_CLOCK_ADVANCE) {
70 struct timespec sleep_delay, rem_delay;
71 sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
72 sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
73 if (nanosleep(&sleep_delay, &rem_delay) < 0) {
74 sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
79 Sleep(sc->diff_clk / SCALE_MS);
85 static void print_delay(const SyncClocks *sc)
87 static float threshold_delay;
88 static int64_t last_realtime_clock;
91 if (icount_align_option &&
92 sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
93 nb_prints < MAX_NB_PRINTS) {
94 if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
95 (-sc->diff_clk / (float)1000000000LL <
96 (threshold_delay - THRESHOLD_REDUCE))) {
97 threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
98 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
102 last_realtime_clock = sc->realtime_clock;
107 static void init_delay_params(SyncClocks *sc,
110 if (!icount_align_option) {
113 sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
114 sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
115 sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
116 if (sc->diff_clk < max_delay) {
117 max_delay = sc->diff_clk;
119 if (sc->diff_clk > max_advance) {
120 max_advance = sc->diff_clk;
123 /* Print every 2s max if the guest is late. We limit the number
124 of printed messages to NB_PRINT_MAX(currently 100) */
128 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
132 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
135 #endif /* CONFIG USER ONLY */
137 /* Execute a TB, and fix up the CPU state afterwards if necessary */
138 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, TranslationBlock *itb)
140 CPUArchState *env = cpu->env_ptr;
142 TranslationBlock *last_tb;
144 uint8_t *tb_ptr = itb->tc_ptr;
146 qemu_log_mask_and_addr(CPU_LOG_EXEC, itb->pc,
147 "Trace %p [%d: " TARGET_FMT_lx "] %s\n",
148 itb->tc_ptr, cpu->cpu_index, itb->pc,
149 lookup_symbol(itb->pc));
151 #if defined(DEBUG_DISAS)
152 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)
153 && qemu_log_in_addr_range(itb->pc)) {
155 #if defined(TARGET_I386)
156 log_cpu_state(cpu, CPU_DUMP_CCOP);
158 log_cpu_state(cpu, 0);
162 #endif /* DEBUG_DISAS */
164 cpu->can_do_io = !use_icount;
165 ret = tcg_qemu_tb_exec(env, tb_ptr);
167 last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
168 tb_exit = ret & TB_EXIT_MASK;
169 trace_exec_tb_exit(last_tb, tb_exit);
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.
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);
186 cc->set_pc(cpu, last_tb->pc);
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.
193 atomic_set(&cpu->tcg_exit_req, 0);
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)
204 TranslationBlock *tb;
206 /* Should never happen.
207 We only end up here when an existing TB is too long. */
208 if (max_cycles > CF_COUNT_MASK)
209 max_cycles = CF_COUNT_MASK;
212 tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
213 max_cycles | CF_NOCACHE
214 | (ignore_icount ? CF_IGNORE_ICOUNT : 0));
215 tb->orig_tb = orig_tb;
218 /* execute the generated code */
219 trace_exec_tb_nocache(tb, tb->pc);
220 cpu_tb_exec(cpu, tb);
223 tb_phys_invalidate(tb, -1);
229 static void cpu_exec_step(CPUState *cpu)
231 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
232 TranslationBlock *tb;
233 target_ulong cs_base, pc;
236 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
238 tb = tb_gen_code(cpu, pc, cs_base, flags,
239 1 | CF_NOCACHE | CF_IGNORE_ICOUNT);
242 /* execute the generated code */
243 trace_exec_tb_nocache(tb, pc);
244 cpu_tb_exec(cpu, tb);
246 tb_phys_invalidate(tb, -1);
251 void cpu_exec_step_atomic(CPUState *cpu)
255 /* Since we got here, we know that parallel_cpus must be true. */
256 parallel_cpus = false;
258 parallel_cpus = true;
265 target_ulong cs_base;
267 tb_page_addr_t phys_page1;
271 static bool tb_cmp(const void *p, const void *d)
273 const TranslationBlock *tb = p;
274 const struct tb_desc *desc = d;
276 if (tb->pc == desc->pc &&
277 tb->page_addr[0] == desc->phys_page1 &&
278 tb->cs_base == desc->cs_base &&
279 tb->flags == desc->flags &&
280 !atomic_read(&tb->invalid)) {
281 /* check next page if needed */
282 if (tb->page_addr[1] == -1) {
285 tb_page_addr_t phys_page2;
286 target_ulong virt_page2;
288 virt_page2 = (desc->pc & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
289 phys_page2 = get_page_addr_code(desc->env, virt_page2);
290 if (tb->page_addr[1] == phys_page2) {
298 static TranslationBlock *tb_htable_lookup(CPUState *cpu,
300 target_ulong cs_base,
303 tb_page_addr_t phys_pc;
307 desc.env = (CPUArchState *)cpu->env_ptr;
308 desc.cs_base = cs_base;
311 phys_pc = get_page_addr_code(desc.env, pc);
312 desc.phys_page1 = phys_pc & TARGET_PAGE_MASK;
313 h = tb_hash_func(phys_pc, pc, flags);
314 return qht_lookup(&tcg_ctx.tb_ctx.htable, tb_cmp, &desc, h);
317 static inline TranslationBlock *tb_find(CPUState *cpu,
318 TranslationBlock *last_tb,
321 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
322 TranslationBlock *tb;
323 target_ulong cs_base, pc;
325 bool have_tb_lock = false;
327 /* we record a subset of the CPU state. It will
328 always be the same before a given translated block
330 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
331 tb = atomic_rcu_read(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)]);
332 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
333 tb->flags != flags)) {
334 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
337 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
338 * taken outside tb_lock. As system emulation is currently
339 * single threaded the locks are NOPs.
345 /* There's a chance that our desired tb has been translated while
346 * taking the locks so we check again inside the lock.
348 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
350 /* if no translated code available, then translate it now */
351 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
357 /* We add the TB in the virtual pc hash table for the fast lookup */
358 atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb);
360 #ifndef CONFIG_USER_ONLY
361 /* We don't take care of direct jumps when address mapping changes in
362 * system emulation. So it's not safe to make a direct jump to a TB
363 * spanning two pages because the mapping for the second page can change.
365 if (tb->page_addr[1] != -1) {
369 /* See if we can patch the calling TB. */
370 if (last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
376 tb_add_jump(last_tb, tb_exit, tb);
385 static inline bool cpu_handle_halt(CPUState *cpu)
388 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
389 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
390 && replay_interrupt()) {
391 X86CPU *x86_cpu = X86_CPU(cpu);
392 qemu_mutex_lock_iothread();
393 apic_poll_irq(x86_cpu->apic_state);
394 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
395 qemu_mutex_unlock_iothread();
398 if (!cpu_has_work(cpu)) {
409 static inline void cpu_handle_debug_exception(CPUState *cpu)
411 CPUClass *cc = CPU_GET_CLASS(cpu);
414 if (!cpu->watchpoint_hit) {
415 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
416 wp->flags &= ~BP_WATCHPOINT_HIT;
420 cc->debug_excp_handler(cpu);
423 static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
425 if (cpu->exception_index >= 0) {
426 if (cpu->exception_index >= EXCP_INTERRUPT) {
427 /* exit request from the cpu execution loop */
428 *ret = cpu->exception_index;
429 if (*ret == EXCP_DEBUG) {
430 cpu_handle_debug_exception(cpu);
432 cpu->exception_index = -1;
435 #if defined(CONFIG_USER_ONLY)
436 /* if user mode only, we simulate a fake exception
437 which will be handled outside the cpu execution
439 #if defined(TARGET_I386)
440 CPUClass *cc = CPU_GET_CLASS(cpu);
441 cc->do_interrupt(cpu);
443 *ret = cpu->exception_index;
444 cpu->exception_index = -1;
447 if (replay_exception()) {
448 CPUClass *cc = CPU_GET_CLASS(cpu);
449 qemu_mutex_lock_iothread();
450 cc->do_interrupt(cpu);
451 qemu_mutex_unlock_iothread();
452 cpu->exception_index = -1;
453 } else if (!replay_has_interrupt()) {
454 /* give a chance to iothread in replay mode */
455 *ret = EXCP_INTERRUPT;
460 #ifndef CONFIG_USER_ONLY
461 } else if (replay_has_exception()
462 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
463 /* try to cause an exception pending in the log */
464 cpu_exec_nocache(cpu, 1, tb_find(cpu, NULL, 0), true);
473 static inline bool cpu_handle_interrupt(CPUState *cpu,
474 TranslationBlock **last_tb)
476 CPUClass *cc = CPU_GET_CLASS(cpu);
478 if (unlikely(atomic_read(&cpu->interrupt_request))) {
479 int interrupt_request;
480 qemu_mutex_lock_iothread();
481 interrupt_request = cpu->interrupt_request;
482 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
483 /* Mask out external interrupts for this step. */
484 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
486 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
487 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
488 cpu->exception_index = EXCP_DEBUG;
489 qemu_mutex_unlock_iothread();
492 if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
494 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
496 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
498 cpu->exception_index = EXCP_HLT;
499 qemu_mutex_unlock_iothread();
502 #if defined(TARGET_I386)
503 else if (interrupt_request & CPU_INTERRUPT_INIT) {
504 X86CPU *x86_cpu = X86_CPU(cpu);
505 CPUArchState *env = &x86_cpu->env;
507 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0, 0);
508 do_cpu_init(x86_cpu);
509 cpu->exception_index = EXCP_HALTED;
510 qemu_mutex_unlock_iothread();
514 else if (interrupt_request & CPU_INTERRUPT_RESET) {
517 qemu_mutex_unlock_iothread();
521 /* The target hook has 3 exit conditions:
522 False when the interrupt isn't processed,
523 True when it is, and we should restart on a new TB,
524 and via longjmp via cpu_loop_exit. */
526 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
530 /* The target hook may have updated the 'cpu->interrupt_request';
531 * reload the 'interrupt_request' value */
532 interrupt_request = cpu->interrupt_request;
534 if (interrupt_request & CPU_INTERRUPT_EXITTB) {
535 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
536 /* ensure that no TB jump will be modified as
537 the program flow was changed */
541 /* If we exit via cpu_loop_exit/longjmp it is reset in cpu_exec */
542 qemu_mutex_unlock_iothread();
546 if (unlikely(atomic_read(&cpu->exit_request) || replay_has_interrupt())) {
547 atomic_set(&cpu->exit_request, 0);
548 cpu->exception_index = EXCP_INTERRUPT;
555 static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
556 TranslationBlock **last_tb, int *tb_exit,
561 if (unlikely(atomic_read(&cpu->exit_request))) {
565 trace_exec_tb(tb, tb->pc);
566 ret = cpu_tb_exec(cpu, tb);
567 tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
568 *tb_exit = ret & TB_EXIT_MASK;
570 case TB_EXIT_REQUESTED:
571 /* Something asked us to stop executing
572 * chained TBs; just continue round the main
573 * loop. Whatever requested the exit will also
574 * have set something else (eg exit_request or
575 * interrupt_request) which we will handle
576 * next time around the loop. But we need to
577 * ensure the zeroing of tcg_exit_req (see cpu_tb_exec)
578 * comes before the next read of cpu->exit_request
579 * or cpu->interrupt_request.
584 case TB_EXIT_ICOUNT_EXPIRED:
586 /* Instruction counter expired. */
587 #ifdef CONFIG_USER_ONLY
590 int insns_left = cpu->icount_decr.u32;
592 if (cpu->icount_extra && insns_left >= 0) {
593 /* Refill decrementer and continue execution. */
594 cpu->icount_extra += insns_left;
595 insns_left = MIN(0xffff, cpu->icount_extra);
596 cpu->icount_extra -= insns_left;
597 cpu->icount_decr.u16.low = insns_left;
599 if (insns_left > 0) {
600 /* Execute remaining instructions. */
601 cpu_exec_nocache(cpu, insns_left, tb, false);
602 align_clocks(sc, cpu);
604 cpu->exception_index = EXCP_INTERRUPT;
616 /* main execution loop */
618 int cpu_exec(CPUState *cpu)
620 CPUClass *cc = CPU_GET_CLASS(cpu);
624 /* replay_interrupt may need current_cpu */
627 if (cpu_handle_halt(cpu)) {
633 if (unlikely(atomic_mb_read(&exit_request))) {
634 cpu->exit_request = 1;
637 cc->cpu_exec_enter(cpu);
639 /* Calculate difference between guest clock and host clock.
640 * This delay includes the delay of the last cycle, so
641 * what we have to do is sleep until it is 0. As for the
642 * advance/delay we gain here, we try to fix it next time.
644 init_delay_params(&sc, cpu);
646 /* prepare setjmp context for exception handling */
647 if (sigsetjmp(cpu->jmp_env, 0) != 0) {
648 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
649 /* Some compilers wrongly smash all local variables after
650 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
651 * Reload essential local variables here for those compilers.
652 * Newer versions of gcc would complain about this code (-Wclobbered). */
654 cc = CPU_GET_CLASS(cpu);
655 #else /* buggy compiler */
656 /* Assert that the compiler does not smash local variables. */
657 g_assert(cpu == current_cpu);
658 g_assert(cc == CPU_GET_CLASS(cpu));
659 #endif /* buggy compiler */
662 if (qemu_mutex_iothread_locked()) {
663 qemu_mutex_unlock_iothread();
667 /* if an exception is pending, we execute it here */
668 while (!cpu_handle_exception(cpu, &ret)) {
669 TranslationBlock *last_tb = NULL;
672 while (!cpu_handle_interrupt(cpu, &last_tb)) {
673 TranslationBlock *tb = tb_find(cpu, last_tb, tb_exit);
674 cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc);
675 /* Try to align the host and virtual clocks
676 if the guest is in advance */
677 align_clocks(&sc, cpu);
681 cc->cpu_exec_exit(cpu);
684 /* fail safe : never use current_cpu outside cpu_exec() */