linux-user: support netlink protocol NETLINK_KOBJECT_UEVENT
[qemu/kevin.git] / cpu-exec.c
blobf7c642f4a9135e05a89c07be85248474421699f0
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 [" TARGET_FMT_lx "] %s\n",
147 itb->tc_ptr, itb->pc, lookup_symbol(itb->pc));
149 #if defined(DEBUG_DISAS)
150 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
151 #if defined(TARGET_I386)
152 log_cpu_state(cpu, CPU_DUMP_CCOP);
153 #elif defined(TARGET_M68K)
154 /* ??? Should not modify env state for dumping. */
155 cpu_m68k_flush_flags(env, env->cc_op);
156 env->cc_op = CC_OP_FLAGS;
157 env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
158 log_cpu_state(cpu, 0);
159 #else
160 log_cpu_state(cpu, 0);
161 #endif
163 #endif /* DEBUG_DISAS */
165 cpu->can_do_io = !use_icount;
166 ret = tcg_qemu_tb_exec(env, tb_ptr);
167 cpu->can_do_io = 1;
168 last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
169 tb_exit = ret & TB_EXIT_MASK;
170 trace_exec_tb_exit(last_tb, tb_exit);
172 if (tb_exit > TB_EXIT_IDX1) {
173 /* We didn't start executing this TB (eg because the instruction
174 * counter hit zero); we must restore the guest PC to the address
175 * of the start of the TB.
177 CPUClass *cc = CPU_GET_CLASS(cpu);
178 qemu_log_mask_and_addr(CPU_LOG_EXEC, last_tb->pc,
179 "Stopped execution of TB chain before %p ["
180 TARGET_FMT_lx "] %s\n",
181 last_tb->tc_ptr, last_tb->pc,
182 lookup_symbol(last_tb->pc));
183 if (cc->synchronize_from_tb) {
184 cc->synchronize_from_tb(cpu, last_tb);
185 } else {
186 assert(cc->set_pc);
187 cc->set_pc(cpu, last_tb->pc);
190 if (tb_exit == TB_EXIT_REQUESTED) {
191 /* We were asked to stop executing TBs (probably a pending
192 * interrupt. We've now stopped, so clear the flag.
194 cpu->tcg_exit_req = 0;
196 return ret;
199 #ifndef CONFIG_USER_ONLY
200 /* Execute the code without caching the generated code. An interpreter
201 could be used if available. */
202 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
203 TranslationBlock *orig_tb, bool ignore_icount)
205 TranslationBlock *tb;
206 bool old_tb_flushed;
208 /* Should never happen.
209 We only end up here when an existing TB is too long. */
210 if (max_cycles > CF_COUNT_MASK)
211 max_cycles = CF_COUNT_MASK;
213 old_tb_flushed = cpu->tb_flushed;
214 cpu->tb_flushed = false;
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 = cpu->tb_flushed ? NULL : orig_tb;
219 cpu->tb_flushed |= old_tb_flushed;
220 /* execute the generated code */
221 trace_exec_tb_nocache(tb, tb->pc);
222 cpu_tb_exec(cpu, tb);
223 tb_phys_invalidate(tb, -1);
224 tb_free(tb);
226 #endif
228 static TranslationBlock *tb_find_physical(CPUState *cpu,
229 target_ulong pc,
230 target_ulong cs_base,
231 uint32_t flags)
233 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
234 TranslationBlock *tb, **tb_hash_head, **ptb1;
235 unsigned int h;
236 tb_page_addr_t phys_pc, phys_page1;
238 /* find translated block using physical mappings */
239 phys_pc = get_page_addr_code(env, pc);
240 phys_page1 = phys_pc & TARGET_PAGE_MASK;
241 h = tb_phys_hash_func(phys_pc);
243 /* Start at head of the hash entry */
244 ptb1 = tb_hash_head = &tcg_ctx.tb_ctx.tb_phys_hash[h];
245 tb = *ptb1;
247 while (tb) {
248 if (tb->pc == pc &&
249 tb->page_addr[0] == phys_page1 &&
250 tb->cs_base == cs_base &&
251 tb->flags == flags) {
253 if (tb->page_addr[1] == -1) {
254 /* done, we have a match */
255 break;
256 } else {
257 /* check next page if needed */
258 target_ulong virt_page2 = (pc & TARGET_PAGE_MASK) +
259 TARGET_PAGE_SIZE;
260 tb_page_addr_t phys_page2 = get_page_addr_code(env, virt_page2);
262 if (tb->page_addr[1] == phys_page2) {
263 break;
268 ptb1 = &tb->phys_hash_next;
269 tb = *ptb1;
272 if (tb) {
273 /* Move the TB to the head of the list */
274 *ptb1 = tb->phys_hash_next;
275 tb->phys_hash_next = *tb_hash_head;
276 *tb_hash_head = tb;
278 return tb;
281 static TranslationBlock *tb_find_slow(CPUState *cpu,
282 target_ulong pc,
283 target_ulong cs_base,
284 uint32_t flags)
286 TranslationBlock *tb;
288 tb = tb_find_physical(cpu, pc, cs_base, flags);
289 if (tb) {
290 goto found;
293 #ifdef CONFIG_USER_ONLY
294 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
295 * taken outside tb_lock. Since we're momentarily dropping
296 * tb_lock, there's a chance that our desired tb has been
297 * translated.
299 tb_unlock();
300 mmap_lock();
301 tb_lock();
302 tb = tb_find_physical(cpu, pc, cs_base, flags);
303 if (tb) {
304 mmap_unlock();
305 goto found;
307 #endif
309 /* if no translated code available, then translate it now */
310 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
312 #ifdef CONFIG_USER_ONLY
313 mmap_unlock();
314 #endif
316 found:
317 /* we add the TB in the virtual pc hash table */
318 cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
319 return tb;
322 static inline TranslationBlock *tb_find_fast(CPUState *cpu,
323 TranslationBlock **last_tb,
324 int tb_exit)
326 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
327 TranslationBlock *tb;
328 target_ulong cs_base, pc;
329 uint32_t flags;
331 /* we record a subset of the CPU state. It will
332 always be the same before a given translated block
333 is executed. */
334 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
335 tb_lock();
336 tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
337 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
338 tb->flags != flags)) {
339 tb = tb_find_slow(cpu, pc, cs_base, flags);
341 if (cpu->tb_flushed) {
342 /* Ensure that no TB jump will be modified as the
343 * translation buffer has been flushed.
345 *last_tb = NULL;
346 cpu->tb_flushed = false;
348 #ifndef CONFIG_USER_ONLY
349 /* We don't take care of direct jumps when address mapping changes in
350 * system emulation. So it's not safe to make a direct jump to a TB
351 * spanning two pages because the mapping for the second page can change.
353 if (tb->page_addr[1] != -1) {
354 *last_tb = NULL;
356 #endif
357 /* See if we can patch the calling TB. */
358 if (*last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
359 tb_add_jump(*last_tb, tb_exit, tb);
361 tb_unlock();
362 return tb;
365 static inline bool cpu_handle_halt(CPUState *cpu)
367 if (cpu->halted) {
368 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
369 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
370 && replay_interrupt()) {
371 X86CPU *x86_cpu = X86_CPU(cpu);
372 apic_poll_irq(x86_cpu->apic_state);
373 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
375 #endif
376 if (!cpu_has_work(cpu)) {
377 current_cpu = NULL;
378 return true;
381 cpu->halted = 0;
384 return false;
387 static inline void cpu_handle_debug_exception(CPUState *cpu)
389 CPUClass *cc = CPU_GET_CLASS(cpu);
390 CPUWatchpoint *wp;
392 if (!cpu->watchpoint_hit) {
393 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
394 wp->flags &= ~BP_WATCHPOINT_HIT;
398 cc->debug_excp_handler(cpu);
401 static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
403 if (cpu->exception_index >= 0) {
404 if (cpu->exception_index >= EXCP_INTERRUPT) {
405 /* exit request from the cpu execution loop */
406 *ret = cpu->exception_index;
407 if (*ret == EXCP_DEBUG) {
408 cpu_handle_debug_exception(cpu);
410 cpu->exception_index = -1;
411 return true;
412 } else {
413 #if defined(CONFIG_USER_ONLY)
414 /* if user mode only, we simulate a fake exception
415 which will be handled outside the cpu execution
416 loop */
417 #if defined(TARGET_I386)
418 CPUClass *cc = CPU_GET_CLASS(cpu);
419 cc->do_interrupt(cpu);
420 #endif
421 *ret = cpu->exception_index;
422 cpu->exception_index = -1;
423 return true;
424 #else
425 if (replay_exception()) {
426 CPUClass *cc = CPU_GET_CLASS(cpu);
427 cc->do_interrupt(cpu);
428 cpu->exception_index = -1;
429 } else if (!replay_has_interrupt()) {
430 /* give a chance to iothread in replay mode */
431 *ret = EXCP_INTERRUPT;
432 return true;
434 #endif
436 #ifndef CONFIG_USER_ONLY
437 } else if (replay_has_exception()
438 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
439 /* try to cause an exception pending in the log */
440 TranslationBlock *last_tb = NULL; /* Avoid chaining TBs */
441 cpu_exec_nocache(cpu, 1, tb_find_fast(cpu, &last_tb, 0), true);
442 *ret = -1;
443 return true;
444 #endif
447 return false;
450 static inline void cpu_handle_interrupt(CPUState *cpu,
451 TranslationBlock **last_tb)
453 CPUClass *cc = CPU_GET_CLASS(cpu);
454 int interrupt_request = cpu->interrupt_request;
456 if (unlikely(interrupt_request)) {
457 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
458 /* Mask out external interrupts for this step. */
459 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
461 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
462 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
463 cpu->exception_index = EXCP_DEBUG;
464 cpu_loop_exit(cpu);
466 if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
467 /* Do nothing */
468 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
469 replay_interrupt();
470 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
471 cpu->halted = 1;
472 cpu->exception_index = EXCP_HLT;
473 cpu_loop_exit(cpu);
475 #if defined(TARGET_I386)
476 else if (interrupt_request & CPU_INTERRUPT_INIT) {
477 X86CPU *x86_cpu = X86_CPU(cpu);
478 CPUArchState *env = &x86_cpu->env;
479 replay_interrupt();
480 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
481 do_cpu_init(x86_cpu);
482 cpu->exception_index = EXCP_HALTED;
483 cpu_loop_exit(cpu);
485 #else
486 else if (interrupt_request & CPU_INTERRUPT_RESET) {
487 replay_interrupt();
488 cpu_reset(cpu);
489 cpu_loop_exit(cpu);
491 #endif
492 /* The target hook has 3 exit conditions:
493 False when the interrupt isn't processed,
494 True when it is, and we should restart on a new TB,
495 and via longjmp via cpu_loop_exit. */
496 else {
497 replay_interrupt();
498 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
499 *last_tb = NULL;
501 /* The target hook may have updated the 'cpu->interrupt_request';
502 * reload the 'interrupt_request' value */
503 interrupt_request = cpu->interrupt_request;
505 if (interrupt_request & CPU_INTERRUPT_EXITTB) {
506 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
507 /* ensure that no TB jump will be modified as
508 the program flow was changed */
509 *last_tb = NULL;
512 if (unlikely(cpu->exit_request || replay_has_interrupt())) {
513 cpu->exit_request = 0;
514 cpu->exception_index = EXCP_INTERRUPT;
515 cpu_loop_exit(cpu);
519 static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
520 TranslationBlock **last_tb, int *tb_exit,
521 SyncClocks *sc)
523 uintptr_t ret;
525 if (unlikely(cpu->exit_request)) {
526 return;
529 trace_exec_tb(tb, tb->pc);
530 ret = cpu_tb_exec(cpu, tb);
531 *last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
532 *tb_exit = ret & TB_EXIT_MASK;
533 switch (*tb_exit) {
534 case TB_EXIT_REQUESTED:
535 /* Something asked us to stop executing
536 * chained TBs; just continue round the main
537 * loop. Whatever requested the exit will also
538 * have set something else (eg exit_request or
539 * interrupt_request) which we will handle
540 * next time around the loop. But we need to
541 * ensure the tcg_exit_req read in generated code
542 * comes before the next read of cpu->exit_request
543 * or cpu->interrupt_request.
545 smp_rmb();
546 *last_tb = NULL;
547 break;
548 case TB_EXIT_ICOUNT_EXPIRED:
550 /* Instruction counter expired. */
551 #ifdef CONFIG_USER_ONLY
552 abort();
553 #else
554 int insns_left = cpu->icount_decr.u32;
555 if (cpu->icount_extra && insns_left >= 0) {
556 /* Refill decrementer and continue execution. */
557 cpu->icount_extra += insns_left;
558 insns_left = MIN(0xffff, cpu->icount_extra);
559 cpu->icount_extra -= insns_left;
560 cpu->icount_decr.u16.low = insns_left;
561 } else {
562 if (insns_left > 0) {
563 /* Execute remaining instructions. */
564 cpu_exec_nocache(cpu, insns_left, *last_tb, false);
565 align_clocks(sc, cpu);
567 cpu->exception_index = EXCP_INTERRUPT;
568 *last_tb = NULL;
569 cpu_loop_exit(cpu);
571 break;
572 #endif
574 default:
575 break;
579 /* main execution loop */
581 int cpu_exec(CPUState *cpu)
583 CPUClass *cc = CPU_GET_CLASS(cpu);
584 int ret;
585 SyncClocks sc;
587 /* replay_interrupt may need current_cpu */
588 current_cpu = cpu;
590 if (cpu_handle_halt(cpu)) {
591 return EXCP_HALTED;
594 atomic_mb_set(&tcg_current_cpu, cpu);
595 rcu_read_lock();
597 if (unlikely(atomic_mb_read(&exit_request))) {
598 cpu->exit_request = 1;
601 cc->cpu_exec_enter(cpu);
603 /* Calculate difference between guest clock and host clock.
604 * This delay includes the delay of the last cycle, so
605 * what we have to do is sleep until it is 0. As for the
606 * advance/delay we gain here, we try to fix it next time.
608 init_delay_params(&sc, cpu);
610 for(;;) {
611 TranslationBlock *tb, *last_tb;
612 int tb_exit = 0;
614 /* prepare setjmp context for exception handling */
615 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
616 /* if an exception is pending, we execute it here */
617 if (cpu_handle_exception(cpu, &ret)) {
618 break;
621 last_tb = NULL; /* forget the last executed TB after exception */
622 cpu->tb_flushed = false; /* reset before first TB lookup */
623 for(;;) {
624 cpu_handle_interrupt(cpu, &last_tb);
625 tb = tb_find_fast(cpu, &last_tb, tb_exit);
626 cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc);
627 /* Try to align the host and virtual clocks
628 if the guest is in advance */
629 align_clocks(&sc, cpu);
630 } /* for(;;) */
631 } else {
632 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
633 /* Some compilers wrongly smash all local variables after
634 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
635 * Reload essential local variables here for those compilers.
636 * Newer versions of gcc would complain about this code (-Wclobbered). */
637 cpu = current_cpu;
638 cc = CPU_GET_CLASS(cpu);
639 #else /* buggy compiler */
640 /* Assert that the compiler does not smash local variables. */
641 g_assert(cpu == current_cpu);
642 g_assert(cc == CPU_GET_CLASS(cpu));
643 #endif /* buggy compiler */
644 cpu->can_do_io = 1;
645 tb_lock_reset();
647 } /* for(;;) */
649 cc->cpu_exec_exit(cpu);
650 rcu_read_unlock();
652 /* fail safe : never use current_cpu outside cpu_exec() */
653 current_cpu = NULL;
655 /* Does not need atomic_mb_set because a spurious wakeup is okay. */
656 atomic_set(&tcg_current_cpu, NULL);
657 return ret;