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"
28 /* -icount align implementation. */
30 typedef struct SyncClocks {
32 int64_t last_cpu_icount;
33 int64_t realtime_clock;
36 #if !defined(CONFIG_USER_ONLY)
37 /* Allow the guest to have a max 3ms advance.
38 * The difference between the 2 clocks could therefore
41 #define VM_CLOCK_ADVANCE 3000000
42 #define THRESHOLD_REDUCE 1.5
43 #define MAX_DELAY_PRINT_RATE 2000000000LL
44 #define MAX_NB_PRINTS 100
46 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
50 if (!icount_align_option) {
54 cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
55 sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
56 sc->last_cpu_icount = cpu_icount;
58 if (sc->diff_clk > VM_CLOCK_ADVANCE) {
60 struct timespec sleep_delay, rem_delay;
61 sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
62 sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
63 if (nanosleep(&sleep_delay, &rem_delay) < 0) {
64 sc->diff_clk -= (sleep_delay.tv_sec - rem_delay.tv_sec) * 1000000000LL;
65 sc->diff_clk -= sleep_delay.tv_nsec - rem_delay.tv_nsec;
70 Sleep(sc->diff_clk / SCALE_MS);
76 static void print_delay(const SyncClocks *sc)
78 static float threshold_delay;
79 static int64_t last_realtime_clock;
82 if (icount_align_option &&
83 sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
84 nb_prints < MAX_NB_PRINTS) {
85 if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
86 (-sc->diff_clk / (float)1000000000LL <
87 (threshold_delay - THRESHOLD_REDUCE))) {
88 threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
89 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
93 last_realtime_clock = sc->realtime_clock;
98 static void init_delay_params(SyncClocks *sc,
101 if (!icount_align_option) {
104 sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
105 sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) -
107 cpu_get_clock_offset();
108 sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
109 if (sc->diff_clk < max_delay) {
110 max_delay = sc->diff_clk;
112 if (sc->diff_clk > max_advance) {
113 max_advance = sc->diff_clk;
116 /* Print every 2s max if the guest is late. We limit the number
117 of printed messages to NB_PRINT_MAX(currently 100) */
121 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
125 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
128 #endif /* CONFIG USER ONLY */
130 void cpu_loop_exit(CPUState *cpu)
132 cpu->current_tb = NULL;
133 siglongjmp(cpu->jmp_env, 1);
136 /* exit the current TB from a signal handler. The host registers are
137 restored in a state compatible with the CPU emulator
139 #if defined(CONFIG_SOFTMMU)
140 void cpu_resume_from_signal(CPUState *cpu, void *puc)
142 /* XXX: restore cpu registers saved in host registers */
144 cpu->exception_index = -1;
145 siglongjmp(cpu->jmp_env, 1);
149 /* Execute a TB, and fix up the CPU state afterwards if necessary */
150 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, uint8_t *tb_ptr)
152 CPUArchState *env = cpu->env_ptr;
155 #if defined(DEBUG_DISAS)
156 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
157 #if defined(TARGET_I386)
158 log_cpu_state(cpu, CPU_DUMP_CCOP);
159 #elif defined(TARGET_M68K)
160 /* ??? Should not modify env state for dumping. */
161 cpu_m68k_flush_flags(env, env->cc_op);
162 env->cc_op = CC_OP_FLAGS;
163 env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
164 log_cpu_state(cpu, 0);
166 log_cpu_state(cpu, 0);
169 #endif /* DEBUG_DISAS */
171 next_tb = tcg_qemu_tb_exec(env, tb_ptr);
172 trace_exec_tb_exit((void *) (next_tb & ~TB_EXIT_MASK),
173 next_tb & TB_EXIT_MASK);
175 if ((next_tb & TB_EXIT_MASK) > TB_EXIT_IDX1) {
176 /* We didn't start executing this TB (eg because the instruction
177 * counter hit zero); we must restore the guest PC to the address
178 * of the start of the TB.
180 CPUClass *cc = CPU_GET_CLASS(cpu);
181 TranslationBlock *tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
182 if (cc->synchronize_from_tb) {
183 cc->synchronize_from_tb(cpu, tb);
186 cc->set_pc(cpu, tb->pc);
189 if ((next_tb & TB_EXIT_MASK) == 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 cpu->tcg_exit_req = 0;
198 /* Execute the code without caching the generated code. An interpreter
199 could be used if available. */
200 static void cpu_exec_nocache(CPUArchState *env, int max_cycles,
201 TranslationBlock *orig_tb)
203 CPUState *cpu = ENV_GET_CPU(env);
204 TranslationBlock *tb;
205 target_ulong pc = orig_tb->pc;
206 target_ulong cs_base = orig_tb->cs_base;
207 uint64_t flags = orig_tb->flags;
209 /* Should never happen.
210 We only end up here when an existing TB is too long. */
211 if (max_cycles > CF_COUNT_MASK)
212 max_cycles = CF_COUNT_MASK;
214 /* tb_gen_code can flush our orig_tb, invalidate it now */
215 tb_phys_invalidate(orig_tb, -1);
216 tb = tb_gen_code(cpu, pc, cs_base, flags,
218 cpu->current_tb = tb;
219 /* execute the generated code */
220 trace_exec_tb_nocache(tb, tb->pc);
221 cpu_tb_exec(cpu, tb->tc_ptr);
222 cpu->current_tb = NULL;
223 tb_phys_invalidate(tb, -1);
227 static TranslationBlock *tb_find_slow(CPUArchState *env,
229 target_ulong cs_base,
232 CPUState *cpu = ENV_GET_CPU(env);
233 TranslationBlock *tb, **ptb1;
235 tb_page_addr_t phys_pc, phys_page1;
236 target_ulong virt_page2;
238 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
240 /* find translated block using physical mappings */
241 phys_pc = get_page_addr_code(env, pc);
242 phys_page1 = phys_pc & TARGET_PAGE_MASK;
243 h = tb_phys_hash_func(phys_pc);
244 ptb1 = &tcg_ctx.tb_ctx.tb_phys_hash[h];
250 tb->page_addr[0] == phys_page1 &&
251 tb->cs_base == cs_base &&
252 tb->flags == flags) {
253 /* check next page if needed */
254 if (tb->page_addr[1] != -1) {
255 tb_page_addr_t phys_page2;
257 virt_page2 = (pc & TARGET_PAGE_MASK) +
259 phys_page2 = get_page_addr_code(env, virt_page2);
260 if (tb->page_addr[1] == phys_page2)
266 ptb1 = &tb->phys_hash_next;
269 /* if no translated code available, then translate it now */
270 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
273 /* Move the last found TB to the head of the list */
275 *ptb1 = tb->phys_hash_next;
276 tb->phys_hash_next = tcg_ctx.tb_ctx.tb_phys_hash[h];
277 tcg_ctx.tb_ctx.tb_phys_hash[h] = tb;
279 /* we add the TB in the virtual pc hash table */
280 cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
284 static inline TranslationBlock *tb_find_fast(CPUArchState *env)
286 CPUState *cpu = ENV_GET_CPU(env);
287 TranslationBlock *tb;
288 target_ulong cs_base, pc;
291 /* we record a subset of the CPU state. It will
292 always be the same before a given translated block
294 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
295 tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
296 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
297 tb->flags != flags)) {
298 tb = tb_find_slow(env, pc, cs_base, flags);
303 static void cpu_handle_debug_exception(CPUArchState *env)
305 CPUState *cpu = ENV_GET_CPU(env);
306 CPUClass *cc = CPU_GET_CLASS(cpu);
309 if (!cpu->watchpoint_hit) {
310 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
311 wp->flags &= ~BP_WATCHPOINT_HIT;
315 cc->debug_excp_handler(cpu);
318 /* main execution loop */
320 volatile sig_atomic_t exit_request;
322 int cpu_exec(CPUArchState *env)
324 CPUState *cpu = ENV_GET_CPU(env);
325 CPUClass *cc = CPU_GET_CLASS(cpu);
327 X86CPU *x86_cpu = X86_CPU(cpu);
329 int ret, interrupt_request;
330 TranslationBlock *tb;
335 /* This must be volatile so it is not trashed by longjmp() */
336 volatile bool have_tb_lock = false;
339 if (!cpu_has_work(cpu)) {
348 /* As long as current_cpu is null, up to the assignment just above,
349 * requests by other threads to exit the execution loop are expected to
350 * be issued using the exit_request global. We must make sure that our
351 * evaluation of the global value is performed past the current_cpu
352 * value transition point, which requires a memory barrier as well as
353 * an instruction scheduling constraint on modern architectures. */
356 if (unlikely(exit_request)) {
357 cpu->exit_request = 1;
360 cc->cpu_exec_enter(cpu);
361 cpu->exception_index = -1;
363 /* Calculate difference between guest clock and host clock.
364 * This delay includes the delay of the last cycle, so
365 * what we have to do is sleep until it is 0. As for the
366 * advance/delay we gain here, we try to fix it next time.
368 init_delay_params(&sc, cpu);
370 /* prepare setjmp context for exception handling */
372 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
373 /* if an exception is pending, we execute it here */
374 if (cpu->exception_index >= 0) {
375 if (cpu->exception_index >= EXCP_INTERRUPT) {
376 /* exit request from the cpu execution loop */
377 ret = cpu->exception_index;
378 if (ret == EXCP_DEBUG) {
379 cpu_handle_debug_exception(env);
383 #if defined(CONFIG_USER_ONLY)
384 /* if user mode only, we simulate a fake exception
385 which will be handled outside the cpu execution
387 #if defined(TARGET_I386)
388 cc->do_interrupt(cpu);
390 ret = cpu->exception_index;
393 cc->do_interrupt(cpu);
394 cpu->exception_index = -1;
399 next_tb = 0; /* force lookup of first TB */
401 interrupt_request = cpu->interrupt_request;
402 if (unlikely(interrupt_request)) {
403 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
404 /* Mask out external interrupts for this step. */
405 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
407 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
408 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
409 cpu->exception_index = EXCP_DEBUG;
412 if (interrupt_request & CPU_INTERRUPT_HALT) {
413 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
415 cpu->exception_index = EXCP_HLT;
418 #if defined(TARGET_I386)
419 if (interrupt_request & CPU_INTERRUPT_INIT) {
420 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
421 do_cpu_init(x86_cpu);
422 cpu->exception_index = EXCP_HALTED;
426 if (interrupt_request & CPU_INTERRUPT_RESET) {
430 /* The target hook has 3 exit conditions:
431 False when the interrupt isn't processed,
432 True when it is, and we should restart on a new TB,
433 and via longjmp via cpu_loop_exit. */
434 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
437 /* Don't use the cached interrupt_request value,
438 do_interrupt may have updated the EXITTB flag. */
439 if (cpu->interrupt_request & CPU_INTERRUPT_EXITTB) {
440 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
441 /* ensure that no TB jump will be modified as
442 the program flow was changed */
446 if (unlikely(cpu->exit_request)) {
447 cpu->exit_request = 0;
448 cpu->exception_index = EXCP_INTERRUPT;
451 spin_lock(&tcg_ctx.tb_ctx.tb_lock);
453 tb = tb_find_fast(env);
454 /* Note: we do it here to avoid a gcc bug on Mac OS X when
455 doing it in tb_find_slow */
456 if (tcg_ctx.tb_ctx.tb_invalidated_flag) {
457 /* as some TB could have been invalidated because
458 of memory exceptions while generating the code, we
459 must recompute the hash index here */
461 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
463 if (qemu_loglevel_mask(CPU_LOG_EXEC)) {
464 qemu_log("Trace %p [" TARGET_FMT_lx "] %s\n",
465 tb->tc_ptr, tb->pc, lookup_symbol(tb->pc));
467 /* see if we can patch the calling TB. When the TB
468 spans two pages, we cannot safely do a direct
470 if (next_tb != 0 && tb->page_addr[1] == -1) {
471 tb_add_jump((TranslationBlock *)(next_tb & ~TB_EXIT_MASK),
472 next_tb & TB_EXIT_MASK, tb);
474 have_tb_lock = false;
475 spin_unlock(&tcg_ctx.tb_ctx.tb_lock);
477 /* cpu_interrupt might be called while translating the
478 TB, but before it is linked into a potentially
479 infinite loop and becomes env->current_tb. Avoid
480 starting execution if there is a pending interrupt. */
481 cpu->current_tb = tb;
483 if (likely(!cpu->exit_request)) {
484 trace_exec_tb(tb, tb->pc);
486 /* execute the generated code */
487 next_tb = cpu_tb_exec(cpu, tc_ptr);
488 switch (next_tb & TB_EXIT_MASK) {
489 case TB_EXIT_REQUESTED:
490 /* Something asked us to stop executing
491 * chained TBs; just continue round the main
492 * loop. Whatever requested the exit will also
493 * have set something else (eg exit_request or
494 * interrupt_request) which we will handle
495 * next time around the loop.
497 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
500 case TB_EXIT_ICOUNT_EXPIRED:
502 /* Instruction counter expired. */
504 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
505 insns_left = cpu->icount_decr.u32;
506 if (cpu->icount_extra && insns_left >= 0) {
507 /* Refill decrementer and continue execution. */
508 cpu->icount_extra += insns_left;
509 if (cpu->icount_extra > 0xffff) {
512 insns_left = cpu->icount_extra;
514 cpu->icount_extra -= insns_left;
515 cpu->icount_decr.u16.low = insns_left;
517 if (insns_left > 0) {
518 /* Execute remaining instructions. */
519 cpu_exec_nocache(env, insns_left, tb);
520 align_clocks(&sc, cpu);
522 cpu->exception_index = EXCP_INTERRUPT;
532 cpu->current_tb = NULL;
533 /* Try to align the host and virtual clocks
534 if the guest is in advance */
535 align_clocks(&sc, cpu);
536 /* reset soft MMU for next block (it can currently
537 only be set by a memory fault) */
540 /* Reload env after longjmp - the compiler may have smashed all
541 * local variables as longjmp is marked 'noreturn'. */
544 cc = CPU_GET_CLASS(cpu);
546 x86_cpu = X86_CPU(cpu);
549 spin_unlock(&tcg_ctx.tb_ctx.tb_lock);
550 have_tb_lock = false;
555 cc->cpu_exec_exit(cpu);
557 /* fail safe : never use current_cpu outside cpu_exec() */