tcg: move tcg_exec_all and helpers above thread fn
[qemu/ar7.git] / cpu-exec.c
blobe9b50a6b307ed8724780357e70dde545037e4bf6
1 /*
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"
20 #include "cpu.h"
21 #include "trace.h"
22 #include "disas/disas.h"
23 #include "exec/exec-all.h"
24 #include "tcg.h"
25 #include "qemu/atomic.h"
26 #include "sysemu/qtest.h"
27 #include "qemu/timer.h"
28 #include "exec/address-spaces.h"
29 #include "qemu/rcu.h"
30 #include "exec/tb-hash.h"
31 #include "exec/log.h"
32 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
33 #include "hw/i386/apic.h"
34 #endif
35 #include "sysemu/replay.h"
37 /* -icount align implementation. */
39 typedef struct SyncClocks {
40 int64_t diff_clk;
41 int64_t last_cpu_icount;
42 int64_t realtime_clock;
43 } SyncClocks;
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
48 * oscillate around 0.
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)
57 int64_t cpu_icount;
59 if (!icount_align_option) {
60 return;
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) {
68 #ifndef _WIN32
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;
74 } else {
75 sc->diff_clk = 0;
77 #else
78 Sleep(sc->diff_clk / SCALE_MS);
79 sc->diff_clk = 0;
80 #endif
84 static void print_delay(const SyncClocks *sc)
86 static float threshold_delay;
87 static int64_t last_realtime_clock;
88 static int nb_prints;
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",
98 threshold_delay - 1,
99 threshold_delay);
100 nb_prints++;
101 last_realtime_clock = sc->realtime_clock;
106 static void init_delay_params(SyncClocks *sc,
107 const CPUState *cpu)
109 if (!icount_align_option) {
110 return;
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) */
124 print_delay(sc);
126 #else
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;
140 uintptr_t ret;
141 TranslationBlock *last_tb;
142 int tb_exit;
143 uint8_t *tb_ptr = itb->tc_ptr;
145 qemu_log_mask_and_addr(CPU_LOG_EXEC, itb->pc,
146 "Trace %p [%d: " TARGET_FMT_lx "] %s\n",
147 itb->tc_ptr, cpu->cpu_index, itb->pc,
148 lookup_symbol(itb->pc));
150 #if defined(DEBUG_DISAS)
151 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)
152 && qemu_log_in_addr_range(itb->pc)) {
153 #if defined(TARGET_I386)
154 log_cpu_state(cpu, CPU_DUMP_CCOP);
155 #elif defined(TARGET_M68K)
156 /* ??? Should not modify env state for dumping. */
157 cpu_m68k_flush_flags(env, env->cc_op);
158 env->cc_op = CC_OP_FLAGS;
159 env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
160 log_cpu_state(cpu, 0);
161 #else
162 log_cpu_state(cpu, 0);
163 #endif
165 #endif /* DEBUG_DISAS */
167 cpu->can_do_io = !use_icount;
168 ret = tcg_qemu_tb_exec(env, tb_ptr);
169 cpu->can_do_io = 1;
170 last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
171 tb_exit = ret & TB_EXIT_MASK;
172 trace_exec_tb_exit(last_tb, tb_exit);
174 if (tb_exit > TB_EXIT_IDX1) {
175 /* We didn't start executing this TB (eg because the instruction
176 * counter hit zero); we must restore the guest PC to the address
177 * of the start of the TB.
179 CPUClass *cc = CPU_GET_CLASS(cpu);
180 qemu_log_mask_and_addr(CPU_LOG_EXEC, last_tb->pc,
181 "Stopped execution of TB chain before %p ["
182 TARGET_FMT_lx "] %s\n",
183 last_tb->tc_ptr, last_tb->pc,
184 lookup_symbol(last_tb->pc));
185 if (cc->synchronize_from_tb) {
186 cc->synchronize_from_tb(cpu, last_tb);
187 } else {
188 assert(cc->set_pc);
189 cc->set_pc(cpu, last_tb->pc);
192 if (tb_exit == TB_EXIT_REQUESTED) {
193 /* We were asked to stop executing TBs (probably a pending
194 * interrupt. We've now stopped, so clear the flag.
196 atomic_set(&cpu->tcg_exit_req, 0);
198 return ret;
201 #ifndef CONFIG_USER_ONLY
202 /* Execute the code without caching the generated code. An interpreter
203 could be used if available. */
204 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
205 TranslationBlock *orig_tb, bool ignore_icount)
207 TranslationBlock *tb;
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_lock();
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 = orig_tb;
219 tb_unlock();
221 /* execute the generated code */
222 trace_exec_tb_nocache(tb, tb->pc);
223 cpu_tb_exec(cpu, tb);
225 tb_lock();
226 tb_phys_invalidate(tb, -1);
227 tb_free(tb);
228 tb_unlock();
230 #endif
232 static void cpu_exec_step(CPUState *cpu)
234 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
235 TranslationBlock *tb;
236 target_ulong cs_base, pc;
237 uint32_t flags;
239 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
240 tb = tb_gen_code(cpu, pc, cs_base, flags,
241 1 | CF_NOCACHE | CF_IGNORE_ICOUNT);
242 tb->orig_tb = NULL;
243 /* execute the generated code */
244 trace_exec_tb_nocache(tb, pc);
245 cpu_tb_exec(cpu, tb);
246 tb_phys_invalidate(tb, -1);
247 tb_free(tb);
250 void cpu_exec_step_atomic(CPUState *cpu)
252 start_exclusive();
254 /* Since we got here, we know that parallel_cpus must be true. */
255 parallel_cpus = false;
256 cpu_exec_step(cpu);
257 parallel_cpus = true;
259 end_exclusive();
262 struct tb_desc {
263 target_ulong pc;
264 target_ulong cs_base;
265 CPUArchState *env;
266 tb_page_addr_t phys_page1;
267 uint32_t flags;
270 static bool tb_cmp(const void *p, const void *d)
272 const TranslationBlock *tb = p;
273 const struct tb_desc *desc = d;
275 if (tb->pc == desc->pc &&
276 tb->page_addr[0] == desc->phys_page1 &&
277 tb->cs_base == desc->cs_base &&
278 tb->flags == desc->flags &&
279 !atomic_read(&tb->invalid)) {
280 /* check next page if needed */
281 if (tb->page_addr[1] == -1) {
282 return true;
283 } else {
284 tb_page_addr_t phys_page2;
285 target_ulong virt_page2;
287 virt_page2 = (desc->pc & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
288 phys_page2 = get_page_addr_code(desc->env, virt_page2);
289 if (tb->page_addr[1] == phys_page2) {
290 return true;
294 return false;
297 static TranslationBlock *tb_htable_lookup(CPUState *cpu,
298 target_ulong pc,
299 target_ulong cs_base,
300 uint32_t flags)
302 tb_page_addr_t phys_pc;
303 struct tb_desc desc;
304 uint32_t h;
306 desc.env = (CPUArchState *)cpu->env_ptr;
307 desc.cs_base = cs_base;
308 desc.flags = flags;
309 desc.pc = pc;
310 phys_pc = get_page_addr_code(desc.env, pc);
311 desc.phys_page1 = phys_pc & TARGET_PAGE_MASK;
312 h = tb_hash_func(phys_pc, pc, flags);
313 return qht_lookup(&tcg_ctx.tb_ctx.htable, tb_cmp, &desc, h);
316 static inline TranslationBlock *tb_find(CPUState *cpu,
317 TranslationBlock *last_tb,
318 int tb_exit)
320 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
321 TranslationBlock *tb;
322 target_ulong cs_base, pc;
323 uint32_t flags;
324 bool have_tb_lock = false;
326 /* we record a subset of the CPU state. It will
327 always be the same before a given translated block
328 is executed. */
329 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
330 tb = atomic_rcu_read(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)]);
331 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
332 tb->flags != flags)) {
333 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
334 if (!tb) {
336 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
337 * taken outside tb_lock. As system emulation is currently
338 * single threaded the locks are NOPs.
340 mmap_lock();
341 tb_lock();
342 have_tb_lock = true;
344 /* There's a chance that our desired tb has been translated while
345 * taking the locks so we check again inside the lock.
347 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
348 if (!tb) {
349 /* if no translated code available, then translate it now */
350 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
353 mmap_unlock();
356 /* We add the TB in the virtual pc hash table for the fast lookup */
357 atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb);
359 #ifndef CONFIG_USER_ONLY
360 /* We don't take care of direct jumps when address mapping changes in
361 * system emulation. So it's not safe to make a direct jump to a TB
362 * spanning two pages because the mapping for the second page can change.
364 if (tb->page_addr[1] != -1) {
365 last_tb = NULL;
367 #endif
368 /* See if we can patch the calling TB. */
369 if (last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
370 if (!have_tb_lock) {
371 tb_lock();
372 have_tb_lock = true;
374 if (!tb->invalid) {
375 tb_add_jump(last_tb, tb_exit, tb);
378 if (have_tb_lock) {
379 tb_unlock();
381 return tb;
384 static inline bool cpu_handle_halt(CPUState *cpu)
386 if (cpu->halted) {
387 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
388 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
389 && replay_interrupt()) {
390 X86CPU *x86_cpu = X86_CPU(cpu);
391 apic_poll_irq(x86_cpu->apic_state);
392 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
394 #endif
395 if (!cpu_has_work(cpu)) {
396 current_cpu = NULL;
397 return true;
400 cpu->halted = 0;
403 return false;
406 static inline void cpu_handle_debug_exception(CPUState *cpu)
408 CPUClass *cc = CPU_GET_CLASS(cpu);
409 CPUWatchpoint *wp;
411 if (!cpu->watchpoint_hit) {
412 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
413 wp->flags &= ~BP_WATCHPOINT_HIT;
417 cc->debug_excp_handler(cpu);
420 static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
422 if (cpu->exception_index >= 0) {
423 if (cpu->exception_index >= EXCP_INTERRUPT) {
424 /* exit request from the cpu execution loop */
425 *ret = cpu->exception_index;
426 if (*ret == EXCP_DEBUG) {
427 cpu_handle_debug_exception(cpu);
429 cpu->exception_index = -1;
430 return true;
431 } else {
432 #if defined(CONFIG_USER_ONLY)
433 /* if user mode only, we simulate a fake exception
434 which will be handled outside the cpu execution
435 loop */
436 #if defined(TARGET_I386)
437 CPUClass *cc = CPU_GET_CLASS(cpu);
438 cc->do_interrupt(cpu);
439 #endif
440 *ret = cpu->exception_index;
441 cpu->exception_index = -1;
442 return true;
443 #else
444 if (replay_exception()) {
445 CPUClass *cc = CPU_GET_CLASS(cpu);
446 cc->do_interrupt(cpu);
447 cpu->exception_index = -1;
448 } else if (!replay_has_interrupt()) {
449 /* give a chance to iothread in replay mode */
450 *ret = EXCP_INTERRUPT;
451 return true;
453 #endif
455 #ifndef CONFIG_USER_ONLY
456 } else if (replay_has_exception()
457 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
458 /* try to cause an exception pending in the log */
459 cpu_exec_nocache(cpu, 1, tb_find(cpu, NULL, 0), true);
460 *ret = -1;
461 return true;
462 #endif
465 return false;
468 static inline void cpu_handle_interrupt(CPUState *cpu,
469 TranslationBlock **last_tb)
471 CPUClass *cc = CPU_GET_CLASS(cpu);
472 int interrupt_request = cpu->interrupt_request;
474 if (unlikely(interrupt_request)) {
475 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
476 /* Mask out external interrupts for this step. */
477 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
479 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
480 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
481 cpu->exception_index = EXCP_DEBUG;
482 cpu_loop_exit(cpu);
484 if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
485 /* Do nothing */
486 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
487 replay_interrupt();
488 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
489 cpu->halted = 1;
490 cpu->exception_index = EXCP_HLT;
491 cpu_loop_exit(cpu);
493 #if defined(TARGET_I386)
494 else if (interrupt_request & CPU_INTERRUPT_INIT) {
495 X86CPU *x86_cpu = X86_CPU(cpu);
496 CPUArchState *env = &x86_cpu->env;
497 replay_interrupt();
498 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
499 do_cpu_init(x86_cpu);
500 cpu->exception_index = EXCP_HALTED;
501 cpu_loop_exit(cpu);
503 #else
504 else if (interrupt_request & CPU_INTERRUPT_RESET) {
505 replay_interrupt();
506 cpu_reset(cpu);
507 cpu_loop_exit(cpu);
509 #endif
510 /* The target hook has 3 exit conditions:
511 False when the interrupt isn't processed,
512 True when it is, and we should restart on a new TB,
513 and via longjmp via cpu_loop_exit. */
514 else {
515 replay_interrupt();
516 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
517 *last_tb = NULL;
519 /* The target hook may have updated the 'cpu->interrupt_request';
520 * reload the 'interrupt_request' value */
521 interrupt_request = cpu->interrupt_request;
523 if (interrupt_request & CPU_INTERRUPT_EXITTB) {
524 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
525 /* ensure that no TB jump will be modified as
526 the program flow was changed */
527 *last_tb = NULL;
530 if (unlikely(atomic_read(&cpu->exit_request) || replay_has_interrupt())) {
531 atomic_set(&cpu->exit_request, 0);
532 cpu->exception_index = EXCP_INTERRUPT;
533 cpu_loop_exit(cpu);
537 static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
538 TranslationBlock **last_tb, int *tb_exit,
539 SyncClocks *sc)
541 uintptr_t ret;
543 if (unlikely(atomic_read(&cpu->exit_request))) {
544 return;
547 trace_exec_tb(tb, tb->pc);
548 ret = cpu_tb_exec(cpu, tb);
549 *last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
550 *tb_exit = ret & TB_EXIT_MASK;
551 switch (*tb_exit) {
552 case TB_EXIT_REQUESTED:
553 /* Something asked us to stop executing
554 * chained TBs; just continue round the main
555 * loop. Whatever requested the exit will also
556 * have set something else (eg exit_request or
557 * interrupt_request) which we will handle
558 * next time around the loop. But we need to
559 * ensure the tcg_exit_req read in generated code
560 * comes before the next read of cpu->exit_request
561 * or cpu->interrupt_request.
563 smp_rmb();
564 *last_tb = NULL;
565 break;
566 case TB_EXIT_ICOUNT_EXPIRED:
568 /* Instruction counter expired. */
569 #ifdef CONFIG_USER_ONLY
570 abort();
571 #else
572 int insns_left = cpu->icount_decr.u32;
573 if (cpu->icount_extra && insns_left >= 0) {
574 /* Refill decrementer and continue execution. */
575 cpu->icount_extra += insns_left;
576 insns_left = MIN(0xffff, cpu->icount_extra);
577 cpu->icount_extra -= insns_left;
578 cpu->icount_decr.u16.low = insns_left;
579 } else {
580 if (insns_left > 0) {
581 /* Execute remaining instructions. */
582 cpu_exec_nocache(cpu, insns_left, *last_tb, false);
583 align_clocks(sc, cpu);
585 cpu->exception_index = EXCP_INTERRUPT;
586 *last_tb = NULL;
587 cpu_loop_exit(cpu);
589 break;
590 #endif
592 default:
593 break;
597 /* main execution loop */
599 int cpu_exec(CPUState *cpu)
601 CPUClass *cc = CPU_GET_CLASS(cpu);
602 int ret;
603 SyncClocks sc;
605 /* replay_interrupt may need current_cpu */
606 current_cpu = cpu;
608 if (cpu_handle_halt(cpu)) {
609 return EXCP_HALTED;
612 atomic_mb_set(&tcg_current_cpu, cpu);
613 rcu_read_lock();
615 if (unlikely(atomic_mb_read(&exit_request))) {
616 cpu->exit_request = 1;
619 cc->cpu_exec_enter(cpu);
621 /* Calculate difference between guest clock and host clock.
622 * This delay includes the delay of the last cycle, so
623 * what we have to do is sleep until it is 0. As for the
624 * advance/delay we gain here, we try to fix it next time.
626 init_delay_params(&sc, cpu);
628 for(;;) {
629 /* prepare setjmp context for exception handling */
630 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
631 TranslationBlock *tb, *last_tb = NULL;
632 int tb_exit = 0;
634 /* if an exception is pending, we execute it here */
635 if (cpu_handle_exception(cpu, &ret)) {
636 break;
639 for(;;) {
640 cpu_handle_interrupt(cpu, &last_tb);
641 tb = tb_find(cpu, last_tb, tb_exit);
642 cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc);
643 /* Try to align the host and virtual clocks
644 if the guest is in advance */
645 align_clocks(&sc, cpu);
646 } /* for(;;) */
647 } else {
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). */
653 cpu = current_cpu;
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 */
660 cpu->can_do_io = 1;
661 tb_lock_reset();
663 } /* for(;;) */
665 cc->cpu_exec_exit(cpu);
666 rcu_read_unlock();
668 /* fail safe : never use current_cpu outside cpu_exec() */
669 current_cpu = NULL;
671 /* Does not need atomic_mb_set because a spurious wakeup is okay. */
672 atomic_set(&tcg_current_cpu, NULL);
673 return ret;