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/>.
22 #include "disas/disas.h"
24 #include "qemu/atomic.h"
25 #include "sysemu/qtest.h"
26 #include "qemu/timer.h"
27 #include "exec/address-spaces.h"
28 #include "exec/memory-internal.h"
30 /* -icount align implementation. */
32 typedef struct SyncClocks {
34 int64_t last_cpu_icount;
35 int64_t realtime_clock;
38 #if !defined(CONFIG_USER_ONLY)
39 /* Allow the guest to have a max 3ms advance.
40 * The difference between the 2 clocks could therefore
43 #define VM_CLOCK_ADVANCE 3000000
44 #define THRESHOLD_REDUCE 1.5
45 #define MAX_DELAY_PRINT_RATE 2000000000LL
46 #define MAX_NB_PRINTS 100
48 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
52 if (!icount_align_option) {
56 cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
57 sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
58 sc->last_cpu_icount = cpu_icount;
60 if (sc->diff_clk > VM_CLOCK_ADVANCE) {
62 struct timespec sleep_delay, rem_delay;
63 sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
64 sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
65 if (nanosleep(&sleep_delay, &rem_delay) < 0) {
66 sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
71 Sleep(sc->diff_clk / SCALE_MS);
77 static void print_delay(const SyncClocks *sc)
79 static float threshold_delay;
80 static int64_t last_realtime_clock;
83 if (icount_align_option &&
84 sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
85 nb_prints < MAX_NB_PRINTS) {
86 if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
87 (-sc->diff_clk / (float)1000000000LL <
88 (threshold_delay - THRESHOLD_REDUCE))) {
89 threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
90 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
94 last_realtime_clock = sc->realtime_clock;
99 static void init_delay_params(SyncClocks *sc,
102 if (!icount_align_option) {
105 sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
106 sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
107 sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
108 if (sc->diff_clk < max_delay) {
109 max_delay = sc->diff_clk;
111 if (sc->diff_clk > max_advance) {
112 max_advance = sc->diff_clk;
115 /* Print every 2s max if the guest is late. We limit the number
116 of printed messages to NB_PRINT_MAX(currently 100) */
120 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
124 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
127 #endif /* CONFIG USER ONLY */
129 void cpu_loop_exit(CPUState *cpu)
131 cpu->current_tb = NULL;
132 siglongjmp(cpu->jmp_env, 1);
135 /* exit the current TB from a signal handler. The host registers are
136 restored in a state compatible with the CPU emulator
138 #if defined(CONFIG_SOFTMMU)
139 void cpu_resume_from_signal(CPUState *cpu, void *puc)
141 /* XXX: restore cpu registers saved in host registers */
143 cpu->exception_index = -1;
144 siglongjmp(cpu->jmp_env, 1);
147 void cpu_reload_memory_map(CPUState *cpu)
149 /* The CPU and TLB are protected by the iothread lock. */
150 AddressSpaceDispatch *d = cpu->as->dispatch;
151 cpu->memory_dispatch = d;
156 /* Execute a TB, and fix up the CPU state afterwards if necessary */
157 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, uint8_t *tb_ptr)
159 CPUArchState *env = cpu->env_ptr;
162 #if defined(DEBUG_DISAS)
163 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
164 #if defined(TARGET_I386)
165 log_cpu_state(cpu, CPU_DUMP_CCOP);
166 #elif defined(TARGET_M68K)
167 /* ??? Should not modify env state for dumping. */
168 cpu_m68k_flush_flags(env, env->cc_op);
169 env->cc_op = CC_OP_FLAGS;
170 env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
171 log_cpu_state(cpu, 0);
173 log_cpu_state(cpu, 0);
176 #endif /* DEBUG_DISAS */
179 next_tb = tcg_qemu_tb_exec(env, tb_ptr);
181 trace_exec_tb_exit((void *) (next_tb & ~TB_EXIT_MASK),
182 next_tb & TB_EXIT_MASK);
184 if ((next_tb & TB_EXIT_MASK) > TB_EXIT_IDX1) {
185 /* We didn't start executing this TB (eg because the instruction
186 * counter hit zero); we must restore the guest PC to the address
187 * of the start of the TB.
189 CPUClass *cc = CPU_GET_CLASS(cpu);
190 TranslationBlock *tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
191 if (cc->synchronize_from_tb) {
192 cc->synchronize_from_tb(cpu, tb);
195 cc->set_pc(cpu, tb->pc);
198 if ((next_tb & TB_EXIT_MASK) == TB_EXIT_REQUESTED) {
199 /* We were asked to stop executing TBs (probably a pending
200 * interrupt. We've now stopped, so clear the flag.
202 cpu->tcg_exit_req = 0;
207 /* Execute the code without caching the generated code. An interpreter
208 could be used if available. */
209 static void cpu_exec_nocache(CPUArchState *env, int max_cycles,
210 TranslationBlock *orig_tb)
212 CPUState *cpu = ENV_GET_CPU(env);
213 TranslationBlock *tb;
214 target_ulong pc = orig_tb->pc;
215 target_ulong cs_base = orig_tb->cs_base;
216 uint64_t flags = orig_tb->flags;
218 /* Should never happen.
219 We only end up here when an existing TB is too long. */
220 if (max_cycles > CF_COUNT_MASK)
221 max_cycles = CF_COUNT_MASK;
223 /* tb_gen_code can flush our orig_tb, invalidate it now */
224 tb_phys_invalidate(orig_tb, -1);
225 tb = tb_gen_code(cpu, pc, cs_base, flags,
226 max_cycles | CF_NOCACHE);
227 cpu->current_tb = tb;
228 /* execute the generated code */
229 trace_exec_tb_nocache(tb, tb->pc);
230 cpu_tb_exec(cpu, tb->tc_ptr);
231 cpu->current_tb = NULL;
232 tb_phys_invalidate(tb, -1);
236 static TranslationBlock *tb_find_slow(CPUArchState *env,
238 target_ulong cs_base,
241 CPUState *cpu = ENV_GET_CPU(env);
242 TranslationBlock *tb, **ptb1;
244 tb_page_addr_t phys_pc, phys_page1;
245 target_ulong virt_page2;
247 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
249 /* find translated block using physical mappings */
250 phys_pc = get_page_addr_code(env, pc);
251 phys_page1 = phys_pc & TARGET_PAGE_MASK;
252 h = tb_phys_hash_func(phys_pc);
253 ptb1 = &tcg_ctx.tb_ctx.tb_phys_hash[h];
259 tb->page_addr[0] == phys_page1 &&
260 tb->cs_base == cs_base &&
261 tb->flags == flags) {
262 /* check next page if needed */
263 if (tb->page_addr[1] != -1) {
264 tb_page_addr_t phys_page2;
266 virt_page2 = (pc & TARGET_PAGE_MASK) +
268 phys_page2 = get_page_addr_code(env, virt_page2);
269 if (tb->page_addr[1] == phys_page2)
275 ptb1 = &tb->phys_hash_next;
278 /* if no translated code available, then translate it now */
279 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
282 /* Move the last found TB to the head of the list */
284 *ptb1 = tb->phys_hash_next;
285 tb->phys_hash_next = tcg_ctx.tb_ctx.tb_phys_hash[h];
286 tcg_ctx.tb_ctx.tb_phys_hash[h] = tb;
288 /* we add the TB in the virtual pc hash table */
289 cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
293 static inline TranslationBlock *tb_find_fast(CPUArchState *env)
295 CPUState *cpu = ENV_GET_CPU(env);
296 TranslationBlock *tb;
297 target_ulong cs_base, pc;
300 /* we record a subset of the CPU state. It will
301 always be the same before a given translated block
303 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
304 tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
305 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
306 tb->flags != flags)) {
307 tb = tb_find_slow(env, pc, cs_base, flags);
312 static void cpu_handle_debug_exception(CPUArchState *env)
314 CPUState *cpu = ENV_GET_CPU(env);
315 CPUClass *cc = CPU_GET_CLASS(cpu);
318 if (!cpu->watchpoint_hit) {
319 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
320 wp->flags &= ~BP_WATCHPOINT_HIT;
324 cc->debug_excp_handler(cpu);
327 /* main execution loop */
329 volatile sig_atomic_t exit_request;
331 int cpu_exec(CPUArchState *env)
333 CPUState *cpu = ENV_GET_CPU(env);
334 CPUClass *cc = CPU_GET_CLASS(cpu);
336 X86CPU *x86_cpu = X86_CPU(cpu);
338 int ret, interrupt_request;
339 TranslationBlock *tb;
344 /* This must be volatile so it is not trashed by longjmp() */
345 volatile bool have_tb_lock = false;
348 if (!cpu_has_work(cpu)) {
357 /* As long as current_cpu is null, up to the assignment just above,
358 * requests by other threads to exit the execution loop are expected to
359 * be issued using the exit_request global. We must make sure that our
360 * evaluation of the global value is performed past the current_cpu
361 * value transition point, which requires a memory barrier as well as
362 * an instruction scheduling constraint on modern architectures. */
365 if (unlikely(exit_request)) {
366 cpu->exit_request = 1;
369 cc->cpu_exec_enter(cpu);
371 /* Calculate difference between guest clock and host clock.
372 * This delay includes the delay of the last cycle, so
373 * what we have to do is sleep until it is 0. As for the
374 * advance/delay we gain here, we try to fix it next time.
376 init_delay_params(&sc, cpu);
378 /* prepare setjmp context for exception handling */
380 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
381 /* if an exception is pending, we execute it here */
382 if (cpu->exception_index >= 0) {
383 if (cpu->exception_index >= EXCP_INTERRUPT) {
384 /* exit request from the cpu execution loop */
385 ret = cpu->exception_index;
386 if (ret == EXCP_DEBUG) {
387 cpu_handle_debug_exception(env);
389 cpu->exception_index = -1;
392 #if defined(CONFIG_USER_ONLY)
393 /* if user mode only, we simulate a fake exception
394 which will be handled outside the cpu execution
396 #if defined(TARGET_I386)
397 cc->do_interrupt(cpu);
399 ret = cpu->exception_index;
400 cpu->exception_index = -1;
403 cc->do_interrupt(cpu);
404 cpu->exception_index = -1;
409 next_tb = 0; /* force lookup of first TB */
411 interrupt_request = cpu->interrupt_request;
412 if (unlikely(interrupt_request)) {
413 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
414 /* Mask out external interrupts for this step. */
415 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
417 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
418 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
419 cpu->exception_index = EXCP_DEBUG;
422 if (interrupt_request & CPU_INTERRUPT_HALT) {
423 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
425 cpu->exception_index = EXCP_HLT;
428 #if defined(TARGET_I386)
429 if (interrupt_request & CPU_INTERRUPT_INIT) {
430 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
431 do_cpu_init(x86_cpu);
432 cpu->exception_index = EXCP_HALTED;
436 if (interrupt_request & CPU_INTERRUPT_RESET) {
440 /* The target hook has 3 exit conditions:
441 False when the interrupt isn't processed,
442 True when it is, and we should restart on a new TB,
443 and via longjmp via cpu_loop_exit. */
444 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
447 /* Don't use the cached interrupt_request value,
448 do_interrupt may have updated the EXITTB flag. */
449 if (cpu->interrupt_request & CPU_INTERRUPT_EXITTB) {
450 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
451 /* ensure that no TB jump will be modified as
452 the program flow was changed */
456 if (unlikely(cpu->exit_request)) {
457 cpu->exit_request = 0;
458 cpu->exception_index = EXCP_INTERRUPT;
461 spin_lock(&tcg_ctx.tb_ctx.tb_lock);
463 tb = tb_find_fast(env);
464 /* Note: we do it here to avoid a gcc bug on Mac OS X when
465 doing it in tb_find_slow */
466 if (tcg_ctx.tb_ctx.tb_invalidated_flag) {
467 /* as some TB could have been invalidated because
468 of memory exceptions while generating the code, we
469 must recompute the hash index here */
471 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
473 if (qemu_loglevel_mask(CPU_LOG_EXEC)) {
474 qemu_log("Trace %p [" TARGET_FMT_lx "] %s\n",
475 tb->tc_ptr, tb->pc, lookup_symbol(tb->pc));
477 /* see if we can patch the calling TB. When the TB
478 spans two pages, we cannot safely do a direct
480 if (next_tb != 0 && tb->page_addr[1] == -1) {
481 tb_add_jump((TranslationBlock *)(next_tb & ~TB_EXIT_MASK),
482 next_tb & TB_EXIT_MASK, tb);
484 have_tb_lock = false;
485 spin_unlock(&tcg_ctx.tb_ctx.tb_lock);
487 /* cpu_interrupt might be called while translating the
488 TB, but before it is linked into a potentially
489 infinite loop and becomes env->current_tb. Avoid
490 starting execution if there is a pending interrupt. */
491 cpu->current_tb = tb;
493 if (likely(!cpu->exit_request)) {
494 trace_exec_tb(tb, tb->pc);
496 /* execute the generated code */
497 next_tb = cpu_tb_exec(cpu, tc_ptr);
498 switch (next_tb & TB_EXIT_MASK) {
499 case TB_EXIT_REQUESTED:
500 /* Something asked us to stop executing
501 * chained TBs; just continue round the main
502 * loop. Whatever requested the exit will also
503 * have set something else (eg exit_request or
504 * interrupt_request) which we will handle
505 * next time around the loop.
507 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
510 case TB_EXIT_ICOUNT_EXPIRED:
512 /* Instruction counter expired. */
514 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
515 insns_left = cpu->icount_decr.u32;
516 if (cpu->icount_extra && insns_left >= 0) {
517 /* Refill decrementer and continue execution. */
518 cpu->icount_extra += insns_left;
519 if (cpu->icount_extra > 0xffff) {
522 insns_left = cpu->icount_extra;
524 cpu->icount_extra -= insns_left;
525 cpu->icount_decr.u16.low = insns_left;
527 if (insns_left > 0) {
528 /* Execute remaining instructions. */
529 cpu_exec_nocache(env, insns_left, tb);
530 align_clocks(&sc, cpu);
532 cpu->exception_index = EXCP_INTERRUPT;
542 cpu->current_tb = NULL;
543 /* Try to align the host and virtual clocks
544 if the guest is in advance */
545 align_clocks(&sc, cpu);
546 /* reset soft MMU for next block (it can currently
547 only be set by a memory fault) */
550 /* Reload env after longjmp - the compiler may have smashed all
551 * local variables as longjmp is marked 'noreturn'. */
554 cc = CPU_GET_CLASS(cpu);
557 x86_cpu = X86_CPU(cpu);
560 spin_unlock(&tcg_ctx.tb_ctx.tb_lock);
561 have_tb_lock = false;
566 cc->cpu_exec_exit(cpu);
568 /* fail safe : never use current_cpu outside cpu_exec() */