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"
29 #include "exec/tb-hash.h"
30 #include "exec/tb-lookup.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/cpus.h"
37 #include "sysemu/replay.h"
39 /* -icount align implementation. */
41 typedef struct SyncClocks {
43 int64_t last_cpu_icount;
44 int64_t realtime_clock;
47 #if !defined(CONFIG_USER_ONLY)
48 /* Allow the guest to have a max 3ms advance.
49 * The difference between the 2 clocks could therefore
52 #define VM_CLOCK_ADVANCE 3000000
53 #define THRESHOLD_REDUCE 1.5
54 #define MAX_DELAY_PRINT_RATE 2000000000LL
55 #define MAX_NB_PRINTS 100
57 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
61 if (!icount_align_option) {
65 cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
66 sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
67 sc->last_cpu_icount = cpu_icount;
69 if (sc->diff_clk > VM_CLOCK_ADVANCE) {
71 struct timespec sleep_delay, rem_delay;
72 sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
73 sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
74 if (nanosleep(&sleep_delay, &rem_delay) < 0) {
75 sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
80 Sleep(sc->diff_clk / SCALE_MS);
86 static void print_delay(const SyncClocks *sc)
88 static float threshold_delay;
89 static int64_t last_realtime_clock;
92 if (icount_align_option &&
93 sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
94 nb_prints < MAX_NB_PRINTS) {
95 if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
96 (-sc->diff_clk / (float)1000000000LL <
97 (threshold_delay - THRESHOLD_REDUCE))) {
98 threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
99 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
103 last_realtime_clock = sc->realtime_clock;
108 static void init_delay_params(SyncClocks *sc,
111 if (!icount_align_option) {
114 sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
115 sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
116 sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
117 if (sc->diff_clk < max_delay) {
118 max_delay = sc->diff_clk;
120 if (sc->diff_clk > max_advance) {
121 max_advance = sc->diff_clk;
124 /* Print every 2s max if the guest is late. We limit the number
125 of printed messages to NB_PRINT_MAX(currently 100) */
129 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
133 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
136 #endif /* CONFIG USER ONLY */
138 /* Execute a TB, and fix up the CPU state afterwards if necessary */
139 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, TranslationBlock *itb)
141 CPUArchState *env = cpu->env_ptr;
143 TranslationBlock *last_tb;
145 uint8_t *tb_ptr = itb->tc.ptr;
147 qemu_log_mask_and_addr(CPU_LOG_EXEC, itb->pc,
149 TARGET_FMT_lx "/" TARGET_FMT_lx "/%#x] %s\n",
150 cpu->cpu_index, itb->tc.ptr,
151 itb->cs_base, itb->pc, itb->flags,
152 lookup_symbol(itb->pc));
154 #if defined(DEBUG_DISAS)
155 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)
156 && qemu_log_in_addr_range(itb->pc)) {
159 if (qemu_loglevel_mask(CPU_LOG_TB_FPU)) {
160 flags |= CPU_DUMP_FPU;
162 #if defined(TARGET_I386)
163 flags |= CPU_DUMP_CCOP;
165 log_cpu_state(cpu, flags);
168 #endif /* DEBUG_DISAS */
170 cpu->can_do_io = !use_icount;
171 ret = tcg_qemu_tb_exec(env, tb_ptr);
173 last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
174 tb_exit = ret & TB_EXIT_MASK;
175 trace_exec_tb_exit(last_tb, tb_exit);
177 if (tb_exit > TB_EXIT_IDX1) {
178 /* We didn't start executing this TB (eg because the instruction
179 * counter hit zero); we must restore the guest PC to the address
180 * of the start of the TB.
182 CPUClass *cc = CPU_GET_CLASS(cpu);
183 qemu_log_mask_and_addr(CPU_LOG_EXEC, last_tb->pc,
184 "Stopped execution of TB chain before %p ["
185 TARGET_FMT_lx "] %s\n",
186 last_tb->tc.ptr, last_tb->pc,
187 lookup_symbol(last_tb->pc));
188 if (cc->synchronize_from_tb) {
189 cc->synchronize_from_tb(cpu, last_tb);
192 cc->set_pc(cpu, last_tb->pc);
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;
205 uint32_t cflags = curr_cflags() | CF_NOCACHE;
208 cflags &= ~CF_USE_ICOUNT;
211 /* Should never happen.
212 We only end up here when an existing TB is too long. */
213 cflags |= MIN(max_cycles, CF_COUNT_MASK);
216 tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base,
217 orig_tb->flags, cflags);
218 tb->orig_tb = orig_tb;
221 /* execute the generated code */
222 trace_exec_tb_nocache(tb, tb->pc);
223 cpu_tb_exec(cpu, tb);
226 tb_phys_invalidate(tb, -1);
232 void cpu_exec_step_atomic(CPUState *cpu)
234 CPUClass *cc = CPU_GET_CLASS(cpu);
235 TranslationBlock *tb;
236 target_ulong cs_base, pc;
239 uint32_t cf_mask = cflags & CF_HASH_MASK;
240 /* volatile because we modify it between setjmp and longjmp */
241 volatile bool in_exclusive_region = false;
243 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
244 tb = tb_lookup__cpu_state(cpu, &pc, &cs_base, &flags, cf_mask);
248 tb = tb_gen_code(cpu, pc, cs_base, flags, cflags);
255 /* Since we got here, we know that parallel_cpus must be true. */
256 parallel_cpus = false;
257 in_exclusive_region = true;
258 cc->cpu_exec_enter(cpu);
259 /* execute the generated code */
260 trace_exec_tb(tb, pc);
261 cpu_tb_exec(cpu, tb);
262 cc->cpu_exec_exit(cpu);
264 /* We may have exited due to another problem here, so we need
265 * to reset any tb_locks we may have taken but didn't release.
266 * The mmap_lock is dropped by tb_gen_code if it runs out of
269 #ifndef CONFIG_SOFTMMU
270 tcg_debug_assert(!have_mmap_lock());
273 assert_no_pages_locked();
276 if (in_exclusive_region) {
277 /* We might longjump out of either the codegen or the
278 * execution, so must make sure we only end the exclusive
279 * region if we started it.
281 parallel_cpus = true;
288 target_ulong cs_base;
290 tb_page_addr_t phys_page1;
293 uint32_t trace_vcpu_dstate;
296 static bool tb_lookup_cmp(const void *p, const void *d)
298 const TranslationBlock *tb = p;
299 const struct tb_desc *desc = d;
301 if (tb->pc == desc->pc &&
302 tb->page_addr[0] == desc->phys_page1 &&
303 tb->cs_base == desc->cs_base &&
304 tb->flags == desc->flags &&
305 tb->trace_vcpu_dstate == desc->trace_vcpu_dstate &&
306 (tb_cflags(tb) & (CF_HASH_MASK | CF_INVALID)) == desc->cf_mask) {
307 /* check next page if needed */
308 if (tb->page_addr[1] == -1) {
311 tb_page_addr_t phys_page2;
312 target_ulong virt_page2;
314 virt_page2 = (desc->pc & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
315 phys_page2 = get_page_addr_code(desc->env, virt_page2);
316 if (tb->page_addr[1] == phys_page2) {
324 TranslationBlock *tb_htable_lookup(CPUState *cpu, target_ulong pc,
325 target_ulong cs_base, uint32_t flags,
328 tb_page_addr_t phys_pc;
332 desc.env = (CPUArchState *)cpu->env_ptr;
333 desc.cs_base = cs_base;
335 desc.cf_mask = cf_mask;
336 desc.trace_vcpu_dstate = *cpu->trace_dstate;
338 phys_pc = get_page_addr_code(desc.env, pc);
339 desc.phys_page1 = phys_pc & TARGET_PAGE_MASK;
340 h = tb_hash_func(phys_pc, pc, flags, cf_mask, *cpu->trace_dstate);
341 return qht_lookup_custom(&tb_ctx.htable, &desc, h, tb_lookup_cmp);
344 void tb_set_jmp_target(TranslationBlock *tb, int n, uintptr_t addr)
346 if (TCG_TARGET_HAS_direct_jump) {
347 uintptr_t offset = tb->jmp_target_arg[n];
348 uintptr_t tc_ptr = (uintptr_t)tb->tc.ptr;
349 tb_target_set_jmp_target(tc_ptr, tc_ptr + offset, addr);
351 tb->jmp_target_arg[n] = addr;
355 /* Called with tb_lock held. */
356 static inline void tb_add_jump(TranslationBlock *tb, int n,
357 TranslationBlock *tb_next)
359 assert(n < ARRAY_SIZE(tb->jmp_list_next));
360 if (tb->jmp_list_next[n]) {
361 /* Another thread has already done this while we were
362 * outside of the lock; nothing to do in this case */
365 qemu_log_mask_and_addr(CPU_LOG_EXEC, tb->pc,
366 "Linking TBs %p [" TARGET_FMT_lx
367 "] index %d -> %p [" TARGET_FMT_lx "]\n",
368 tb->tc.ptr, tb->pc, n,
369 tb_next->tc.ptr, tb_next->pc);
371 /* patch the native jump address */
372 tb_set_jmp_target(tb, n, (uintptr_t)tb_next->tc.ptr);
374 /* add in TB jmp circular list */
375 tb->jmp_list_next[n] = tb_next->jmp_list_first;
376 tb_next->jmp_list_first = (uintptr_t)tb | n;
379 static inline TranslationBlock *tb_find(CPUState *cpu,
380 TranslationBlock *last_tb,
381 int tb_exit, uint32_t cf_mask)
383 TranslationBlock *tb;
384 target_ulong cs_base, pc;
386 bool acquired_tb_lock = false;
388 tb = tb_lookup__cpu_state(cpu, &pc, &cs_base, &flags, cf_mask);
390 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
391 * taken outside tb_lock. As system emulation is currently
392 * single threaded the locks are NOPs.
396 acquired_tb_lock = true;
398 tb = tb_gen_code(cpu, pc, cs_base, flags, cf_mask);
401 /* We add the TB in the virtual pc hash table for the fast lookup */
402 atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb);
404 #ifndef CONFIG_USER_ONLY
405 /* We don't take care of direct jumps when address mapping changes in
406 * system emulation. So it's not safe to make a direct jump to a TB
407 * spanning two pages because the mapping for the second page can change.
409 if (tb->page_addr[1] != -1) {
413 /* See if we can patch the calling TB. */
414 if (last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
415 if (!acquired_tb_lock) {
417 acquired_tb_lock = true;
419 if (!(tb->cflags & CF_INVALID)) {
420 tb_add_jump(last_tb, tb_exit, tb);
423 if (acquired_tb_lock) {
429 static inline bool cpu_handle_halt(CPUState *cpu)
432 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
433 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
434 && replay_interrupt()) {
435 X86CPU *x86_cpu = X86_CPU(cpu);
436 qemu_mutex_lock_iothread();
437 apic_poll_irq(x86_cpu->apic_state);
438 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
439 qemu_mutex_unlock_iothread();
442 if (!cpu_has_work(cpu)) {
452 static inline void cpu_handle_debug_exception(CPUState *cpu)
454 CPUClass *cc = CPU_GET_CLASS(cpu);
457 if (!cpu->watchpoint_hit) {
458 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
459 wp->flags &= ~BP_WATCHPOINT_HIT;
463 cc->debug_excp_handler(cpu);
466 static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
468 if (cpu->exception_index < 0) {
469 #ifndef CONFIG_USER_ONLY
470 if (replay_has_exception()
471 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
472 /* try to cause an exception pending in the log */
473 cpu_exec_nocache(cpu, 1, tb_find(cpu, NULL, 0, curr_cflags()), true);
476 if (cpu->exception_index < 0) {
481 if (cpu->exception_index >= EXCP_INTERRUPT) {
482 /* exit request from the cpu execution loop */
483 *ret = cpu->exception_index;
484 if (*ret == EXCP_DEBUG) {
485 cpu_handle_debug_exception(cpu);
487 cpu->exception_index = -1;
490 #if defined(CONFIG_USER_ONLY)
491 /* if user mode only, we simulate a fake exception
492 which will be handled outside the cpu execution
494 #if defined(TARGET_I386)
495 CPUClass *cc = CPU_GET_CLASS(cpu);
496 cc->do_interrupt(cpu);
498 *ret = cpu->exception_index;
499 cpu->exception_index = -1;
502 if (replay_exception()) {
503 CPUClass *cc = CPU_GET_CLASS(cpu);
504 qemu_mutex_lock_iothread();
505 cc->do_interrupt(cpu);
506 qemu_mutex_unlock_iothread();
507 cpu->exception_index = -1;
508 } else if (!replay_has_interrupt()) {
509 /* give a chance to iothread in replay mode */
510 *ret = EXCP_INTERRUPT;
519 static inline bool cpu_handle_interrupt(CPUState *cpu,
520 TranslationBlock **last_tb)
522 CPUClass *cc = CPU_GET_CLASS(cpu);
524 /* Clear the interrupt flag now since we're processing
525 * cpu->interrupt_request and cpu->exit_request.
526 * Ensure zeroing happens before reading cpu->exit_request or
527 * cpu->interrupt_request (see also smp_wmb in cpu_exit())
529 atomic_mb_set(&cpu->icount_decr.u16.high, 0);
531 if (unlikely(atomic_read(&cpu->interrupt_request))) {
532 int interrupt_request;
533 qemu_mutex_lock_iothread();
534 interrupt_request = cpu->interrupt_request;
535 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
536 /* Mask out external interrupts for this step. */
537 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
539 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
540 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
541 cpu->exception_index = EXCP_DEBUG;
542 qemu_mutex_unlock_iothread();
545 if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
547 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
549 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
551 cpu->exception_index = EXCP_HLT;
552 qemu_mutex_unlock_iothread();
555 #if defined(TARGET_I386)
556 else if (interrupt_request & CPU_INTERRUPT_INIT) {
557 X86CPU *x86_cpu = X86_CPU(cpu);
558 CPUArchState *env = &x86_cpu->env;
560 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0, 0);
561 do_cpu_init(x86_cpu);
562 cpu->exception_index = EXCP_HALTED;
563 qemu_mutex_unlock_iothread();
567 else if (interrupt_request & CPU_INTERRUPT_RESET) {
570 qemu_mutex_unlock_iothread();
574 /* The target hook has 3 exit conditions:
575 False when the interrupt isn't processed,
576 True when it is, and we should restart on a new TB,
577 and via longjmp via cpu_loop_exit. */
579 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
581 cpu->exception_index = -1;
584 /* The target hook may have updated the 'cpu->interrupt_request';
585 * reload the 'interrupt_request' value */
586 interrupt_request = cpu->interrupt_request;
588 if (interrupt_request & CPU_INTERRUPT_EXITTB) {
589 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
590 /* ensure that no TB jump will be modified as
591 the program flow was changed */
595 /* If we exit via cpu_loop_exit/longjmp it is reset in cpu_exec */
596 qemu_mutex_unlock_iothread();
599 /* Finally, check if we need to exit to the main loop. */
600 if (unlikely(atomic_read(&cpu->exit_request)
601 || (use_icount && cpu->icount_decr.u16.low + cpu->icount_extra == 0))) {
602 atomic_set(&cpu->exit_request, 0);
603 if (cpu->exception_index == -1) {
604 cpu->exception_index = EXCP_INTERRUPT;
612 static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
613 TranslationBlock **last_tb, int *tb_exit)
618 trace_exec_tb(tb, tb->pc);
619 ret = cpu_tb_exec(cpu, tb);
620 tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
621 *tb_exit = ret & TB_EXIT_MASK;
622 if (*tb_exit != TB_EXIT_REQUESTED) {
628 insns_left = atomic_read(&cpu->icount_decr.u32);
629 if (insns_left < 0) {
630 /* Something asked us to stop executing chained TBs; just
631 * continue round the main loop. Whatever requested the exit
632 * will also have set something else (eg exit_request or
633 * interrupt_request) which will be handled by
634 * cpu_handle_interrupt. cpu_handle_interrupt will also
635 * clear cpu->icount_decr.u16.high.
640 /* Instruction counter expired. */
642 #ifndef CONFIG_USER_ONLY
643 /* Ensure global icount has gone forward */
644 cpu_update_icount(cpu);
645 /* Refill decrementer and continue execution. */
646 insns_left = MIN(0xffff, cpu->icount_budget);
647 cpu->icount_decr.u16.low = insns_left;
648 cpu->icount_extra = cpu->icount_budget - insns_left;
649 if (!cpu->icount_extra) {
650 /* Execute any remaining instructions, then let the main loop
651 * handle the next event.
653 if (insns_left > 0) {
654 cpu_exec_nocache(cpu, insns_left, tb, false);
660 /* main execution loop */
662 int cpu_exec(CPUState *cpu)
664 CPUClass *cc = CPU_GET_CLASS(cpu);
666 SyncClocks sc = { 0 };
668 /* replay_interrupt may need current_cpu */
671 if (cpu_handle_halt(cpu)) {
677 cc->cpu_exec_enter(cpu);
679 /* Calculate difference between guest clock and host clock.
680 * This delay includes the delay of the last cycle, so
681 * what we have to do is sleep until it is 0. As for the
682 * advance/delay we gain here, we try to fix it next time.
684 init_delay_params(&sc, cpu);
686 /* prepare setjmp context for exception handling */
687 if (sigsetjmp(cpu->jmp_env, 0) != 0) {
688 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
689 /* Some compilers wrongly smash all local variables after
690 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
691 * Reload essential local variables here for those compilers.
692 * Newer versions of gcc would complain about this code (-Wclobbered). */
694 cc = CPU_GET_CLASS(cpu);
695 #else /* buggy compiler */
696 /* Assert that the compiler does not smash local variables. */
697 g_assert(cpu == current_cpu);
698 g_assert(cc == CPU_GET_CLASS(cpu));
699 #endif /* buggy compiler */
701 if (qemu_mutex_iothread_locked()) {
702 qemu_mutex_unlock_iothread();
706 /* if an exception is pending, we execute it here */
707 while (!cpu_handle_exception(cpu, &ret)) {
708 TranslationBlock *last_tb = NULL;
711 while (!cpu_handle_interrupt(cpu, &last_tb)) {
712 uint32_t cflags = cpu->cflags_next_tb;
713 TranslationBlock *tb;
715 /* When requested, use an exact setting for cflags for the next
716 execution. This is used for icount, precise smc, and stop-
717 after-access watchpoints. Since this request should never
718 have CF_INVALID set, -1 is a convenient invalid value that
719 does not require tcg headers for cpu_common_reset. */
721 cflags = curr_cflags();
723 cpu->cflags_next_tb = -1;
726 tb = tb_find(cpu, last_tb, tb_exit, cflags);
727 cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit);
728 /* Try to align the host and virtual clocks
729 if the guest is in advance */
730 align_clocks(&sc, cpu);
734 cc->cpu_exec_exit(cpu);