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"
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 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
33 #include "hw/i386/apic.h"
35 #include "sysemu/replay.h"
37 /* -icount align implementation. */
39 typedef struct SyncClocks {
41 int64_t last_cpu_icount;
42 int64_t realtime_clock;
45 #if !defined(CONFIG_USER_ONLY)
46 /* Allow the guest to have a max 3ms advance.
47 * The difference between the 2 clocks could therefore
50 #define VM_CLOCK_ADVANCE 3000000
51 #define THRESHOLD_REDUCE 1.5
52 #define MAX_DELAY_PRINT_RATE 2000000000LL
53 #define MAX_NB_PRINTS 100
55 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
59 if (!icount_align_option) {
63 cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
64 sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
65 sc->last_cpu_icount = cpu_icount;
67 if (sc->diff_clk > VM_CLOCK_ADVANCE) {
69 struct timespec sleep_delay, rem_delay;
70 sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
71 sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
72 if (nanosleep(&sleep_delay, &rem_delay) < 0) {
73 sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
78 Sleep(sc->diff_clk / SCALE_MS);
84 static void print_delay(const SyncClocks *sc)
86 static float threshold_delay;
87 static int64_t last_realtime_clock;
90 if (icount_align_option &&
91 sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
92 nb_prints < MAX_NB_PRINTS) {
93 if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
94 (-sc->diff_clk / (float)1000000000LL <
95 (threshold_delay - THRESHOLD_REDUCE))) {
96 threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
97 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
101 last_realtime_clock = sc->realtime_clock;
106 static void init_delay_params(SyncClocks *sc,
109 if (!icount_align_option) {
112 sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
113 sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
114 sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
115 if (sc->diff_clk < max_delay) {
116 max_delay = sc->diff_clk;
118 if (sc->diff_clk > max_advance) {
119 max_advance = sc->diff_clk;
122 /* Print every 2s max if the guest is late. We limit the number
123 of printed messages to NB_PRINT_MAX(currently 100) */
127 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
131 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
134 #endif /* CONFIG USER ONLY */
136 /* Execute a TB, and fix up the CPU state afterwards if necessary */
137 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, TranslationBlock *itb)
139 CPUArchState *env = cpu->env_ptr;
141 TranslationBlock *last_tb;
143 uint8_t *tb_ptr = itb->tc_ptr;
145 qemu_log_mask_and_addr(CPU_LOG_EXEC, itb->pc,
146 "Trace %p [" TARGET_FMT_lx "] %s\n",
147 itb->tc_ptr, itb->pc, lookup_symbol(itb->pc));
149 #if defined(DEBUG_DISAS)
150 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
151 #if defined(TARGET_I386)
152 log_cpu_state(cpu, CPU_DUMP_CCOP);
153 #elif defined(TARGET_M68K)
154 /* ??? Should not modify env state for dumping. */
155 cpu_m68k_flush_flags(env, env->cc_op);
156 env->cc_op = CC_OP_FLAGS;
157 env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
158 log_cpu_state(cpu, 0);
160 log_cpu_state(cpu, 0);
163 #endif /* DEBUG_DISAS */
165 cpu->can_do_io = !use_icount;
166 ret = tcg_qemu_tb_exec(env, tb_ptr);
168 last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
169 tb_exit = ret & TB_EXIT_MASK;
170 trace_exec_tb_exit(last_tb, tb_exit);
172 if (tb_exit > TB_EXIT_IDX1) {
173 /* We didn't start executing this TB (eg because the instruction
174 * counter hit zero); we must restore the guest PC to the address
175 * of the start of the TB.
177 CPUClass *cc = CPU_GET_CLASS(cpu);
178 qemu_log_mask_and_addr(CPU_LOG_EXEC, last_tb->pc,
179 "Stopped execution of TB chain before %p ["
180 TARGET_FMT_lx "] %s\n",
181 last_tb->tc_ptr, last_tb->pc,
182 lookup_symbol(last_tb->pc));
183 if (cc->synchronize_from_tb) {
184 cc->synchronize_from_tb(cpu, last_tb);
187 cc->set_pc(cpu, last_tb->pc);
190 if (tb_exit == TB_EXIT_REQUESTED) {
191 /* We were asked to stop executing TBs (probably a pending
192 * interrupt. We've now stopped, so clear the flag.
194 cpu->tcg_exit_req = 0;
199 #ifndef CONFIG_USER_ONLY
200 /* Execute the code without caching the generated code. An interpreter
201 could be used if available. */
202 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
203 TranslationBlock *orig_tb, bool ignore_icount)
205 TranslationBlock *tb;
208 /* Should never happen.
209 We only end up here when an existing TB is too long. */
210 if (max_cycles > CF_COUNT_MASK)
211 max_cycles = CF_COUNT_MASK;
213 old_tb_flushed = cpu->tb_flushed;
214 cpu->tb_flushed = false;
215 tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
216 max_cycles | CF_NOCACHE
217 | (ignore_icount ? CF_IGNORE_ICOUNT : 0));
218 tb->orig_tb = cpu->tb_flushed ? NULL : orig_tb;
219 cpu->tb_flushed |= old_tb_flushed;
220 /* execute the generated code */
221 trace_exec_tb_nocache(tb, tb->pc);
222 cpu_tb_exec(cpu, tb);
223 tb_phys_invalidate(tb, -1);
230 target_ulong cs_base;
232 tb_page_addr_t phys_page1;
236 static bool tb_cmp(const void *p, const void *d)
238 const TranslationBlock *tb = p;
239 const struct tb_desc *desc = d;
241 if (tb->pc == desc->pc &&
242 tb->page_addr[0] == desc->phys_page1 &&
243 tb->cs_base == desc->cs_base &&
244 tb->flags == desc->flags &&
245 !atomic_read(&tb->invalid)) {
246 /* check next page if needed */
247 if (tb->page_addr[1] == -1) {
250 tb_page_addr_t phys_page2;
251 target_ulong virt_page2;
253 virt_page2 = (desc->pc & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
254 phys_page2 = get_page_addr_code(desc->env, virt_page2);
255 if (tb->page_addr[1] == phys_page2) {
263 static TranslationBlock *tb_htable_lookup(CPUState *cpu,
265 target_ulong cs_base,
268 tb_page_addr_t phys_pc;
272 desc.env = (CPUArchState *)cpu->env_ptr;
273 desc.cs_base = cs_base;
276 phys_pc = get_page_addr_code(desc.env, pc);
277 desc.phys_page1 = phys_pc & TARGET_PAGE_MASK;
278 h = tb_hash_func(phys_pc, pc, flags);
279 return qht_lookup(&tcg_ctx.tb_ctx.htable, tb_cmp, &desc, h);
282 static inline TranslationBlock *tb_find(CPUState *cpu,
283 TranslationBlock *last_tb,
286 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
287 TranslationBlock *tb;
288 target_ulong cs_base, pc;
290 bool have_tb_lock = false;
292 /* we record a subset of the CPU state. It will
293 always be the same before a given translated block
295 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
296 tb = atomic_rcu_read(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)]);
297 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
298 tb->flags != flags)) {
299 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
302 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
303 * taken outside tb_lock. As system emulation is currently
304 * single threaded the locks are NOPs.
310 /* There's a chance that our desired tb has been translated while
311 * taking the locks so we check again inside the lock.
313 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
315 /* if no translated code available, then translate it now */
316 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
322 /* We add the TB in the virtual pc hash table for the fast lookup */
323 atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb);
325 #ifndef CONFIG_USER_ONLY
326 /* We don't take care of direct jumps when address mapping changes in
327 * system emulation. So it's not safe to make a direct jump to a TB
328 * spanning two pages because the mapping for the second page can change.
330 if (tb->page_addr[1] != -1) {
334 /* See if we can patch the calling TB. */
335 if (last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
340 /* Check if translation buffer has been flushed */
341 if (cpu->tb_flushed) {
342 cpu->tb_flushed = false;
343 } else if (!tb->invalid) {
344 tb_add_jump(last_tb, tb_exit, tb);
353 static inline bool cpu_handle_halt(CPUState *cpu)
356 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
357 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
358 && replay_interrupt()) {
359 X86CPU *x86_cpu = X86_CPU(cpu);
360 apic_poll_irq(x86_cpu->apic_state);
361 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
364 if (!cpu_has_work(cpu)) {
375 static inline void cpu_handle_debug_exception(CPUState *cpu)
377 CPUClass *cc = CPU_GET_CLASS(cpu);
380 if (!cpu->watchpoint_hit) {
381 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
382 wp->flags &= ~BP_WATCHPOINT_HIT;
386 cc->debug_excp_handler(cpu);
389 static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
391 if (cpu->exception_index >= 0) {
392 if (cpu->exception_index >= EXCP_INTERRUPT) {
393 /* exit request from the cpu execution loop */
394 *ret = cpu->exception_index;
395 if (*ret == EXCP_DEBUG) {
396 cpu_handle_debug_exception(cpu);
398 cpu->exception_index = -1;
401 #if defined(CONFIG_USER_ONLY)
402 /* if user mode only, we simulate a fake exception
403 which will be handled outside the cpu execution
405 #if defined(TARGET_I386)
406 CPUClass *cc = CPU_GET_CLASS(cpu);
407 cc->do_interrupt(cpu);
409 *ret = cpu->exception_index;
410 cpu->exception_index = -1;
413 if (replay_exception()) {
414 CPUClass *cc = CPU_GET_CLASS(cpu);
415 cc->do_interrupt(cpu);
416 cpu->exception_index = -1;
417 } else if (!replay_has_interrupt()) {
418 /* give a chance to iothread in replay mode */
419 *ret = EXCP_INTERRUPT;
424 #ifndef CONFIG_USER_ONLY
425 } else if (replay_has_exception()
426 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
427 /* try to cause an exception pending in the log */
428 cpu_exec_nocache(cpu, 1, tb_find(cpu, NULL, 0), true);
437 static inline void cpu_handle_interrupt(CPUState *cpu,
438 TranslationBlock **last_tb)
440 CPUClass *cc = CPU_GET_CLASS(cpu);
441 int interrupt_request = cpu->interrupt_request;
443 if (unlikely(interrupt_request)) {
444 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
445 /* Mask out external interrupts for this step. */
446 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
448 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
449 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
450 cpu->exception_index = EXCP_DEBUG;
453 if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
455 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
457 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
459 cpu->exception_index = EXCP_HLT;
462 #if defined(TARGET_I386)
463 else if (interrupt_request & CPU_INTERRUPT_INIT) {
464 X86CPU *x86_cpu = X86_CPU(cpu);
465 CPUArchState *env = &x86_cpu->env;
467 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
468 do_cpu_init(x86_cpu);
469 cpu->exception_index = EXCP_HALTED;
473 else if (interrupt_request & CPU_INTERRUPT_RESET) {
479 /* The target hook has 3 exit conditions:
480 False when the interrupt isn't processed,
481 True when it is, and we should restart on a new TB,
482 and via longjmp via cpu_loop_exit. */
485 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
488 /* The target hook may have updated the 'cpu->interrupt_request';
489 * reload the 'interrupt_request' value */
490 interrupt_request = cpu->interrupt_request;
492 if (interrupt_request & CPU_INTERRUPT_EXITTB) {
493 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
494 /* ensure that no TB jump will be modified as
495 the program flow was changed */
499 if (unlikely(cpu->exit_request || replay_has_interrupt())) {
500 cpu->exit_request = 0;
501 cpu->exception_index = EXCP_INTERRUPT;
506 static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
507 TranslationBlock **last_tb, int *tb_exit,
512 if (unlikely(cpu->exit_request)) {
516 trace_exec_tb(tb, tb->pc);
517 ret = cpu_tb_exec(cpu, tb);
518 *last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
519 *tb_exit = ret & TB_EXIT_MASK;
521 case TB_EXIT_REQUESTED:
522 /* Something asked us to stop executing
523 * chained TBs; just continue round the main
524 * loop. Whatever requested the exit will also
525 * have set something else (eg exit_request or
526 * interrupt_request) which we will handle
527 * next time around the loop. But we need to
528 * ensure the tcg_exit_req read in generated code
529 * comes before the next read of cpu->exit_request
530 * or cpu->interrupt_request.
535 case TB_EXIT_ICOUNT_EXPIRED:
537 /* Instruction counter expired. */
538 #ifdef CONFIG_USER_ONLY
541 int insns_left = cpu->icount_decr.u32;
542 if (cpu->icount_extra && insns_left >= 0) {
543 /* Refill decrementer and continue execution. */
544 cpu->icount_extra += insns_left;
545 insns_left = MIN(0xffff, cpu->icount_extra);
546 cpu->icount_extra -= insns_left;
547 cpu->icount_decr.u16.low = insns_left;
549 if (insns_left > 0) {
550 /* Execute remaining instructions. */
551 cpu_exec_nocache(cpu, insns_left, *last_tb, false);
552 align_clocks(sc, cpu);
554 cpu->exception_index = EXCP_INTERRUPT;
566 /* main execution loop */
568 int cpu_exec(CPUState *cpu)
570 CPUClass *cc = CPU_GET_CLASS(cpu);
574 /* replay_interrupt may need current_cpu */
577 if (cpu_handle_halt(cpu)) {
581 atomic_mb_set(&tcg_current_cpu, cpu);
584 if (unlikely(atomic_mb_read(&exit_request))) {
585 cpu->exit_request = 1;
588 cc->cpu_exec_enter(cpu);
590 /* Calculate difference between guest clock and host clock.
591 * This delay includes the delay of the last cycle, so
592 * what we have to do is sleep until it is 0. As for the
593 * advance/delay we gain here, we try to fix it next time.
595 init_delay_params(&sc, cpu);
598 /* prepare setjmp context for exception handling */
599 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
600 TranslationBlock *tb, *last_tb = NULL;
603 /* if an exception is pending, we execute it here */
604 if (cpu_handle_exception(cpu, &ret)) {
608 atomic_mb_set(&cpu->tb_flushed, false); /* reset before first TB lookup */
610 cpu_handle_interrupt(cpu, &last_tb);
611 tb = tb_find(cpu, last_tb, tb_exit);
612 cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc);
613 /* Try to align the host and virtual clocks
614 if the guest is in advance */
615 align_clocks(&sc, cpu);
618 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
619 /* Some compilers wrongly smash all local variables after
620 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
621 * Reload essential local variables here for those compilers.
622 * Newer versions of gcc would complain about this code (-Wclobbered). */
624 cc = CPU_GET_CLASS(cpu);
625 #else /* buggy compiler */
626 /* Assert that the compiler does not smash local variables. */
627 g_assert(cpu == current_cpu);
628 g_assert(cc == CPU_GET_CLASS(cpu));
629 #endif /* buggy compiler */
635 cc->cpu_exec_exit(cpu);
638 /* fail safe : never use current_cpu outside cpu_exec() */
641 /* Does not need atomic_mb_set because a spurious wakeup is okay. */
642 atomic_set(&tcg_current_cpu, NULL);