nbd: Use CoQueue for free_sema instead of CoMutex
[qemu.git] / cpu-exec.c
blob3e408865a85f5d618de47331c76531d101fab437
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 #else
156 log_cpu_state(cpu, 0);
157 #endif
159 #endif /* DEBUG_DISAS */
161 cpu->can_do_io = !use_icount;
162 ret = tcg_qemu_tb_exec(env, tb_ptr);
163 cpu->can_do_io = 1;
164 last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
165 tb_exit = ret & TB_EXIT_MASK;
166 trace_exec_tb_exit(last_tb, tb_exit);
168 if (tb_exit > TB_EXIT_IDX1) {
169 /* We didn't start executing this TB (eg because the instruction
170 * counter hit zero); we must restore the guest PC to the address
171 * of the start of the TB.
173 CPUClass *cc = CPU_GET_CLASS(cpu);
174 qemu_log_mask_and_addr(CPU_LOG_EXEC, last_tb->pc,
175 "Stopped execution of TB chain before %p ["
176 TARGET_FMT_lx "] %s\n",
177 last_tb->tc_ptr, last_tb->pc,
178 lookup_symbol(last_tb->pc));
179 if (cc->synchronize_from_tb) {
180 cc->synchronize_from_tb(cpu, last_tb);
181 } else {
182 assert(cc->set_pc);
183 cc->set_pc(cpu, last_tb->pc);
186 if (tb_exit == TB_EXIT_REQUESTED) {
187 /* We were asked to stop executing TBs (probably a pending
188 * interrupt. We've now stopped, so clear the flag.
190 atomic_set(&cpu->tcg_exit_req, 0);
192 return ret;
195 #ifndef CONFIG_USER_ONLY
196 /* Execute the code without caching the generated code. An interpreter
197 could be used if available. */
198 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
199 TranslationBlock *orig_tb, bool ignore_icount)
201 TranslationBlock *tb;
203 /* Should never happen.
204 We only end up here when an existing TB is too long. */
205 if (max_cycles > CF_COUNT_MASK)
206 max_cycles = CF_COUNT_MASK;
208 tb_lock();
209 tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
210 max_cycles | CF_NOCACHE
211 | (ignore_icount ? CF_IGNORE_ICOUNT : 0));
212 tb->orig_tb = orig_tb;
213 tb_unlock();
215 /* execute the generated code */
216 trace_exec_tb_nocache(tb, tb->pc);
217 cpu_tb_exec(cpu, tb);
219 tb_lock();
220 tb_phys_invalidate(tb, -1);
221 tb_free(tb);
222 tb_unlock();
224 #endif
226 static void cpu_exec_step(CPUState *cpu)
228 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
229 TranslationBlock *tb;
230 target_ulong cs_base, pc;
231 uint32_t flags;
233 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
234 tb = tb_gen_code(cpu, pc, cs_base, flags,
235 1 | CF_NOCACHE | CF_IGNORE_ICOUNT);
236 tb->orig_tb = NULL;
237 /* execute the generated code */
238 trace_exec_tb_nocache(tb, pc);
239 cpu_tb_exec(cpu, tb);
240 tb_phys_invalidate(tb, -1);
241 tb_free(tb);
244 void cpu_exec_step_atomic(CPUState *cpu)
246 start_exclusive();
248 /* Since we got here, we know that parallel_cpus must be true. */
249 parallel_cpus = false;
250 cpu_exec_step(cpu);
251 parallel_cpus = true;
253 end_exclusive();
256 struct tb_desc {
257 target_ulong pc;
258 target_ulong cs_base;
259 CPUArchState *env;
260 tb_page_addr_t phys_page1;
261 uint32_t flags;
264 static bool tb_cmp(const void *p, const void *d)
266 const TranslationBlock *tb = p;
267 const struct tb_desc *desc = d;
269 if (tb->pc == desc->pc &&
270 tb->page_addr[0] == desc->phys_page1 &&
271 tb->cs_base == desc->cs_base &&
272 tb->flags == desc->flags &&
273 !atomic_read(&tb->invalid)) {
274 /* check next page if needed */
275 if (tb->page_addr[1] == -1) {
276 return true;
277 } else {
278 tb_page_addr_t phys_page2;
279 target_ulong virt_page2;
281 virt_page2 = (desc->pc & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
282 phys_page2 = get_page_addr_code(desc->env, virt_page2);
283 if (tb->page_addr[1] == phys_page2) {
284 return true;
288 return false;
291 static TranslationBlock *tb_htable_lookup(CPUState *cpu,
292 target_ulong pc,
293 target_ulong cs_base,
294 uint32_t flags)
296 tb_page_addr_t phys_pc;
297 struct tb_desc desc;
298 uint32_t h;
300 desc.env = (CPUArchState *)cpu->env_ptr;
301 desc.cs_base = cs_base;
302 desc.flags = flags;
303 desc.pc = pc;
304 phys_pc = get_page_addr_code(desc.env, pc);
305 desc.phys_page1 = phys_pc & TARGET_PAGE_MASK;
306 h = tb_hash_func(phys_pc, pc, flags);
307 return qht_lookup(&tcg_ctx.tb_ctx.htable, tb_cmp, &desc, h);
310 static inline TranslationBlock *tb_find(CPUState *cpu,
311 TranslationBlock *last_tb,
312 int tb_exit)
314 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
315 TranslationBlock *tb;
316 target_ulong cs_base, pc;
317 uint32_t flags;
318 bool have_tb_lock = false;
320 /* we record a subset of the CPU state. It will
321 always be the same before a given translated block
322 is executed. */
323 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
324 tb = atomic_rcu_read(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)]);
325 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
326 tb->flags != flags)) {
327 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
328 if (!tb) {
330 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
331 * taken outside tb_lock. As system emulation is currently
332 * single threaded the locks are NOPs.
334 mmap_lock();
335 tb_lock();
336 have_tb_lock = true;
338 /* There's a chance that our desired tb has been translated while
339 * taking the locks so we check again inside the lock.
341 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
342 if (!tb) {
343 /* if no translated code available, then translate it now */
344 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
347 mmap_unlock();
350 /* We add the TB in the virtual pc hash table for the fast lookup */
351 atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb);
353 #ifndef CONFIG_USER_ONLY
354 /* We don't take care of direct jumps when address mapping changes in
355 * system emulation. So it's not safe to make a direct jump to a TB
356 * spanning two pages because the mapping for the second page can change.
358 if (tb->page_addr[1] != -1) {
359 last_tb = NULL;
361 #endif
362 /* See if we can patch the calling TB. */
363 if (last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
364 if (!have_tb_lock) {
365 tb_lock();
366 have_tb_lock = true;
368 if (!tb->invalid) {
369 tb_add_jump(last_tb, tb_exit, tb);
372 if (have_tb_lock) {
373 tb_unlock();
375 return tb;
378 static inline bool cpu_handle_halt(CPUState *cpu)
380 if (cpu->halted) {
381 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
382 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
383 && replay_interrupt()) {
384 X86CPU *x86_cpu = X86_CPU(cpu);
385 apic_poll_irq(x86_cpu->apic_state);
386 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
388 #endif
389 if (!cpu_has_work(cpu)) {
390 current_cpu = NULL;
391 return true;
394 cpu->halted = 0;
397 return false;
400 static inline void cpu_handle_debug_exception(CPUState *cpu)
402 CPUClass *cc = CPU_GET_CLASS(cpu);
403 CPUWatchpoint *wp;
405 if (!cpu->watchpoint_hit) {
406 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
407 wp->flags &= ~BP_WATCHPOINT_HIT;
411 cc->debug_excp_handler(cpu);
414 static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
416 if (cpu->exception_index >= 0) {
417 if (cpu->exception_index >= EXCP_INTERRUPT) {
418 /* exit request from the cpu execution loop */
419 *ret = cpu->exception_index;
420 if (*ret == EXCP_DEBUG) {
421 cpu_handle_debug_exception(cpu);
423 cpu->exception_index = -1;
424 return true;
425 } else {
426 #if defined(CONFIG_USER_ONLY)
427 /* if user mode only, we simulate a fake exception
428 which will be handled outside the cpu execution
429 loop */
430 #if defined(TARGET_I386)
431 CPUClass *cc = CPU_GET_CLASS(cpu);
432 cc->do_interrupt(cpu);
433 #endif
434 *ret = cpu->exception_index;
435 cpu->exception_index = -1;
436 return true;
437 #else
438 if (replay_exception()) {
439 CPUClass *cc = CPU_GET_CLASS(cpu);
440 cc->do_interrupt(cpu);
441 cpu->exception_index = -1;
442 } else if (!replay_has_interrupt()) {
443 /* give a chance to iothread in replay mode */
444 *ret = EXCP_INTERRUPT;
445 return true;
447 #endif
449 #ifndef CONFIG_USER_ONLY
450 } else if (replay_has_exception()
451 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
452 /* try to cause an exception pending in the log */
453 cpu_exec_nocache(cpu, 1, tb_find(cpu, NULL, 0), true);
454 *ret = -1;
455 return true;
456 #endif
459 return false;
462 static inline void cpu_handle_interrupt(CPUState *cpu,
463 TranslationBlock **last_tb)
465 CPUClass *cc = CPU_GET_CLASS(cpu);
466 int interrupt_request = cpu->interrupt_request;
468 if (unlikely(interrupt_request)) {
469 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
470 /* Mask out external interrupts for this step. */
471 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
473 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
474 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
475 cpu->exception_index = EXCP_DEBUG;
476 cpu_loop_exit(cpu);
478 if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
479 /* Do nothing */
480 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
481 replay_interrupt();
482 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
483 cpu->halted = 1;
484 cpu->exception_index = EXCP_HLT;
485 cpu_loop_exit(cpu);
487 #if defined(TARGET_I386)
488 else if (interrupt_request & CPU_INTERRUPT_INIT) {
489 X86CPU *x86_cpu = X86_CPU(cpu);
490 CPUArchState *env = &x86_cpu->env;
491 replay_interrupt();
492 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
493 do_cpu_init(x86_cpu);
494 cpu->exception_index = EXCP_HALTED;
495 cpu_loop_exit(cpu);
497 #else
498 else if (interrupt_request & CPU_INTERRUPT_RESET) {
499 replay_interrupt();
500 cpu_reset(cpu);
501 cpu_loop_exit(cpu);
503 #endif
504 /* The target hook has 3 exit conditions:
505 False when the interrupt isn't processed,
506 True when it is, and we should restart on a new TB,
507 and via longjmp via cpu_loop_exit. */
508 else {
509 replay_interrupt();
510 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
511 *last_tb = NULL;
513 /* The target hook may have updated the 'cpu->interrupt_request';
514 * reload the 'interrupt_request' value */
515 interrupt_request = cpu->interrupt_request;
517 if (interrupt_request & CPU_INTERRUPT_EXITTB) {
518 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
519 /* ensure that no TB jump will be modified as
520 the program flow was changed */
521 *last_tb = NULL;
524 if (unlikely(atomic_read(&cpu->exit_request) || replay_has_interrupt())) {
525 atomic_set(&cpu->exit_request, 0);
526 cpu->exception_index = EXCP_INTERRUPT;
527 cpu_loop_exit(cpu);
531 static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
532 TranslationBlock **last_tb, int *tb_exit,
533 SyncClocks *sc)
535 uintptr_t ret;
537 if (unlikely(atomic_read(&cpu->exit_request))) {
538 return;
541 trace_exec_tb(tb, tb->pc);
542 ret = cpu_tb_exec(cpu, tb);
543 *last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
544 *tb_exit = ret & TB_EXIT_MASK;
545 switch (*tb_exit) {
546 case TB_EXIT_REQUESTED:
547 /* Something asked us to stop executing
548 * chained TBs; just continue round the main
549 * loop. Whatever requested the exit will also
550 * have set something else (eg exit_request or
551 * interrupt_request) which we will handle
552 * next time around the loop. But we need to
553 * ensure the tcg_exit_req read in generated code
554 * comes before the next read of cpu->exit_request
555 * or cpu->interrupt_request.
557 smp_rmb();
558 *last_tb = NULL;
559 break;
560 case TB_EXIT_ICOUNT_EXPIRED:
562 /* Instruction counter expired. */
563 #ifdef CONFIG_USER_ONLY
564 abort();
565 #else
566 int insns_left = cpu->icount_decr.u32;
567 if (cpu->icount_extra && insns_left >= 0) {
568 /* Refill decrementer and continue execution. */
569 cpu->icount_extra += insns_left;
570 insns_left = MIN(0xffff, cpu->icount_extra);
571 cpu->icount_extra -= insns_left;
572 cpu->icount_decr.u16.low = insns_left;
573 } else {
574 if (insns_left > 0) {
575 /* Execute remaining instructions. */
576 cpu_exec_nocache(cpu, insns_left, *last_tb, false);
577 align_clocks(sc, cpu);
579 cpu->exception_index = EXCP_INTERRUPT;
580 *last_tb = NULL;
581 cpu_loop_exit(cpu);
583 break;
584 #endif
586 default:
587 break;
591 /* main execution loop */
593 int cpu_exec(CPUState *cpu)
595 CPUClass *cc = CPU_GET_CLASS(cpu);
596 int ret;
597 SyncClocks sc;
599 /* replay_interrupt may need current_cpu */
600 current_cpu = cpu;
602 if (cpu_handle_halt(cpu)) {
603 return EXCP_HALTED;
606 atomic_mb_set(&tcg_current_cpu, cpu);
607 rcu_read_lock();
609 if (unlikely(atomic_mb_read(&exit_request))) {
610 cpu->exit_request = 1;
613 cc->cpu_exec_enter(cpu);
615 /* Calculate difference between guest clock and host clock.
616 * This delay includes the delay of the last cycle, so
617 * what we have to do is sleep until it is 0. As for the
618 * advance/delay we gain here, we try to fix it next time.
620 init_delay_params(&sc, cpu);
622 for(;;) {
623 /* prepare setjmp context for exception handling */
624 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
625 TranslationBlock *tb, *last_tb = NULL;
626 int tb_exit = 0;
628 /* if an exception is pending, we execute it here */
629 if (cpu_handle_exception(cpu, &ret)) {
630 break;
633 for(;;) {
634 cpu_handle_interrupt(cpu, &last_tb);
635 tb = tb_find(cpu, last_tb, tb_exit);
636 cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc);
637 /* Try to align the host and virtual clocks
638 if the guest is in advance */
639 align_clocks(&sc, cpu);
640 } /* for(;;) */
641 } else {
642 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
643 /* Some compilers wrongly smash all local variables after
644 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
645 * Reload essential local variables here for those compilers.
646 * Newer versions of gcc would complain about this code (-Wclobbered). */
647 cpu = current_cpu;
648 cc = CPU_GET_CLASS(cpu);
649 #else /* buggy compiler */
650 /* Assert that the compiler does not smash local variables. */
651 g_assert(cpu == current_cpu);
652 g_assert(cc == CPU_GET_CLASS(cpu));
653 #endif /* buggy compiler */
654 cpu->can_do_io = 1;
655 tb_lock_reset();
657 } /* for(;;) */
659 cc->cpu_exec_exit(cpu);
660 rcu_read_unlock();
662 /* fail safe : never use current_cpu outside cpu_exec() */
663 current_cpu = NULL;
665 /* Does not need atomic_mb_set because a spurious wakeup is okay. */
666 atomic_set(&tcg_current_cpu, NULL);
667 return ret;