Merge remote-tracking branch 'remotes/otubo/tags/pull-seccomp-20151116' into staging
[qemu/hpoussin.git] / cpu-exec.c
blobc88d0ffdcdb619422999bbfb6c91215f953c087f
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 "config.h"
20 #include "cpu.h"
21 #include "trace.h"
22 #include "disas/disas.h"
23 #include "tcg.h"
24 #include "qemu/atomic.h"
25 #include "sysemu/qtest.h"
26 #include "qemu/timer.h"
27 #include "exec/address-spaces.h"
28 #include "qemu/rcu.h"
29 #include "exec/tb-hash.h"
30 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
31 #include "hw/i386/apic.h"
32 #endif
33 #include "sysemu/replay.h"
35 /* -icount align implementation. */
37 typedef struct SyncClocks {
38 int64_t diff_clk;
39 int64_t last_cpu_icount;
40 int64_t realtime_clock;
41 } SyncClocks;
43 #if !defined(CONFIG_USER_ONLY)
44 /* Allow the guest to have a max 3ms advance.
45 * The difference between the 2 clocks could therefore
46 * oscillate around 0.
48 #define VM_CLOCK_ADVANCE 3000000
49 #define THRESHOLD_REDUCE 1.5
50 #define MAX_DELAY_PRINT_RATE 2000000000LL
51 #define MAX_NB_PRINTS 100
53 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
55 int64_t cpu_icount;
57 if (!icount_align_option) {
58 return;
61 cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
62 sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
63 sc->last_cpu_icount = cpu_icount;
65 if (sc->diff_clk > VM_CLOCK_ADVANCE) {
66 #ifndef _WIN32
67 struct timespec sleep_delay, rem_delay;
68 sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
69 sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
70 if (nanosleep(&sleep_delay, &rem_delay) < 0) {
71 sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
72 } else {
73 sc->diff_clk = 0;
75 #else
76 Sleep(sc->diff_clk / SCALE_MS);
77 sc->diff_clk = 0;
78 #endif
82 static void print_delay(const SyncClocks *sc)
84 static float threshold_delay;
85 static int64_t last_realtime_clock;
86 static int nb_prints;
88 if (icount_align_option &&
89 sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
90 nb_prints < MAX_NB_PRINTS) {
91 if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
92 (-sc->diff_clk / (float)1000000000LL <
93 (threshold_delay - THRESHOLD_REDUCE))) {
94 threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
95 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
96 threshold_delay - 1,
97 threshold_delay);
98 nb_prints++;
99 last_realtime_clock = sc->realtime_clock;
104 static void init_delay_params(SyncClocks *sc,
105 const CPUState *cpu)
107 if (!icount_align_option) {
108 return;
110 sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
111 sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
112 sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
113 if (sc->diff_clk < max_delay) {
114 max_delay = sc->diff_clk;
116 if (sc->diff_clk > max_advance) {
117 max_advance = sc->diff_clk;
120 /* Print every 2s max if the guest is late. We limit the number
121 of printed messages to NB_PRINT_MAX(currently 100) */
122 print_delay(sc);
124 #else
125 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
129 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
132 #endif /* CONFIG USER ONLY */
134 /* Execute a TB, and fix up the CPU state afterwards if necessary */
135 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, uint8_t *tb_ptr)
137 CPUArchState *env = cpu->env_ptr;
138 uintptr_t next_tb;
140 #if defined(DEBUG_DISAS)
141 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
142 #if defined(TARGET_I386)
143 log_cpu_state(cpu, CPU_DUMP_CCOP);
144 #elif defined(TARGET_M68K)
145 /* ??? Should not modify env state for dumping. */
146 cpu_m68k_flush_flags(env, env->cc_op);
147 env->cc_op = CC_OP_FLAGS;
148 env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
149 log_cpu_state(cpu, 0);
150 #else
151 log_cpu_state(cpu, 0);
152 #endif
154 #endif /* DEBUG_DISAS */
156 cpu->can_do_io = !use_icount;
157 next_tb = tcg_qemu_tb_exec(env, tb_ptr);
158 cpu->can_do_io = 1;
159 trace_exec_tb_exit((void *) (next_tb & ~TB_EXIT_MASK),
160 next_tb & TB_EXIT_MASK);
162 if ((next_tb & TB_EXIT_MASK) > TB_EXIT_IDX1) {
163 /* We didn't start executing this TB (eg because the instruction
164 * counter hit zero); we must restore the guest PC to the address
165 * of the start of the TB.
167 CPUClass *cc = CPU_GET_CLASS(cpu);
168 TranslationBlock *tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
169 if (cc->synchronize_from_tb) {
170 cc->synchronize_from_tb(cpu, tb);
171 } else {
172 assert(cc->set_pc);
173 cc->set_pc(cpu, tb->pc);
176 if ((next_tb & TB_EXIT_MASK) == TB_EXIT_REQUESTED) {
177 /* We were asked to stop executing TBs (probably a pending
178 * interrupt. We've now stopped, so clear the flag.
180 cpu->tcg_exit_req = 0;
182 return next_tb;
185 /* Execute the code without caching the generated code. An interpreter
186 could be used if available. */
187 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
188 TranslationBlock *orig_tb, bool ignore_icount)
190 TranslationBlock *tb;
192 /* Should never happen.
193 We only end up here when an existing TB is too long. */
194 if (max_cycles > CF_COUNT_MASK)
195 max_cycles = CF_COUNT_MASK;
197 tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
198 max_cycles | CF_NOCACHE
199 | (ignore_icount ? CF_IGNORE_ICOUNT : 0));
200 tb->orig_tb = tcg_ctx.tb_ctx.tb_invalidated_flag ? NULL : orig_tb;
201 cpu->current_tb = tb;
202 /* execute the generated code */
203 trace_exec_tb_nocache(tb, tb->pc);
204 cpu_tb_exec(cpu, tb->tc_ptr);
205 cpu->current_tb = NULL;
206 tb_phys_invalidate(tb, -1);
207 tb_free(tb);
210 static TranslationBlock *tb_find_physical(CPUState *cpu,
211 target_ulong pc,
212 target_ulong cs_base,
213 uint64_t flags)
215 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
216 TranslationBlock *tb, **ptb1;
217 unsigned int h;
218 tb_page_addr_t phys_pc, phys_page1;
219 target_ulong virt_page2;
221 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
223 /* find translated block using physical mappings */
224 phys_pc = get_page_addr_code(env, pc);
225 phys_page1 = phys_pc & TARGET_PAGE_MASK;
226 h = tb_phys_hash_func(phys_pc);
227 ptb1 = &tcg_ctx.tb_ctx.tb_phys_hash[h];
228 for(;;) {
229 tb = *ptb1;
230 if (!tb) {
231 return NULL;
233 if (tb->pc == pc &&
234 tb->page_addr[0] == phys_page1 &&
235 tb->cs_base == cs_base &&
236 tb->flags == flags) {
237 /* check next page if needed */
238 if (tb->page_addr[1] != -1) {
239 tb_page_addr_t phys_page2;
241 virt_page2 = (pc & TARGET_PAGE_MASK) +
242 TARGET_PAGE_SIZE;
243 phys_page2 = get_page_addr_code(env, virt_page2);
244 if (tb->page_addr[1] == phys_page2) {
245 break;
247 } else {
248 break;
251 ptb1 = &tb->phys_hash_next;
254 /* Move the TB to the head of the list */
255 *ptb1 = tb->phys_hash_next;
256 tb->phys_hash_next = tcg_ctx.tb_ctx.tb_phys_hash[h];
257 tcg_ctx.tb_ctx.tb_phys_hash[h] = tb;
258 return tb;
261 static TranslationBlock *tb_find_slow(CPUState *cpu,
262 target_ulong pc,
263 target_ulong cs_base,
264 uint64_t flags)
266 TranslationBlock *tb;
268 tb = tb_find_physical(cpu, pc, cs_base, flags);
269 if (tb) {
270 goto found;
273 #ifdef CONFIG_USER_ONLY
274 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
275 * taken outside tb_lock. Since we're momentarily dropping
276 * tb_lock, there's a chance that our desired tb has been
277 * translated.
279 tb_unlock();
280 mmap_lock();
281 tb_lock();
282 tb = tb_find_physical(cpu, pc, cs_base, flags);
283 if (tb) {
284 mmap_unlock();
285 goto found;
287 #endif
289 /* if no translated code available, then translate it now */
290 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
292 #ifdef CONFIG_USER_ONLY
293 mmap_unlock();
294 #endif
296 found:
297 /* we add the TB in the virtual pc hash table */
298 cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
299 return tb;
302 static inline TranslationBlock *tb_find_fast(CPUState *cpu)
304 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
305 TranslationBlock *tb;
306 target_ulong cs_base, pc;
307 int flags;
309 /* we record a subset of the CPU state. It will
310 always be the same before a given translated block
311 is executed. */
312 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
313 tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
314 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
315 tb->flags != flags)) {
316 tb = tb_find_slow(cpu, pc, cs_base, flags);
318 return tb;
321 static void cpu_handle_debug_exception(CPUState *cpu)
323 CPUClass *cc = CPU_GET_CLASS(cpu);
324 CPUWatchpoint *wp;
326 if (!cpu->watchpoint_hit) {
327 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
328 wp->flags &= ~BP_WATCHPOINT_HIT;
332 cc->debug_excp_handler(cpu);
335 /* main execution loop */
337 int cpu_exec(CPUState *cpu)
339 CPUClass *cc = CPU_GET_CLASS(cpu);
340 #ifdef TARGET_I386
341 X86CPU *x86_cpu = X86_CPU(cpu);
342 CPUArchState *env = &x86_cpu->env;
343 #endif
344 int ret, interrupt_request;
345 TranslationBlock *tb;
346 uint8_t *tc_ptr;
347 uintptr_t next_tb;
348 SyncClocks sc;
350 /* replay_interrupt may need current_cpu */
351 current_cpu = cpu;
353 if (cpu->halted) {
354 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
355 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
356 && replay_interrupt()) {
357 apic_poll_irq(x86_cpu->apic_state);
358 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
360 #endif
361 if (!cpu_has_work(cpu)) {
362 current_cpu = NULL;
363 return EXCP_HALTED;
366 cpu->halted = 0;
369 atomic_mb_set(&tcg_current_cpu, cpu);
370 rcu_read_lock();
372 if (unlikely(atomic_mb_read(&exit_request))) {
373 cpu->exit_request = 1;
376 cc->cpu_exec_enter(cpu);
378 /* Calculate difference between guest clock and host clock.
379 * This delay includes the delay of the last cycle, so
380 * what we have to do is sleep until it is 0. As for the
381 * advance/delay we gain here, we try to fix it next time.
383 init_delay_params(&sc, cpu);
385 /* prepare setjmp context for exception handling */
386 for(;;) {
387 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
388 /* if an exception is pending, we execute it here */
389 if (cpu->exception_index >= 0) {
390 if (cpu->exception_index >= EXCP_INTERRUPT) {
391 /* exit request from the cpu execution loop */
392 ret = cpu->exception_index;
393 if (ret == EXCP_DEBUG) {
394 cpu_handle_debug_exception(cpu);
396 cpu->exception_index = -1;
397 break;
398 } else {
399 #if defined(CONFIG_USER_ONLY)
400 /* if user mode only, we simulate a fake exception
401 which will be handled outside the cpu execution
402 loop */
403 #if defined(TARGET_I386)
404 cc->do_interrupt(cpu);
405 #endif
406 ret = cpu->exception_index;
407 cpu->exception_index = -1;
408 break;
409 #else
410 if (replay_exception()) {
411 cc->do_interrupt(cpu);
412 cpu->exception_index = -1;
413 } else if (!replay_has_interrupt()) {
414 /* give a chance to iothread in replay mode */
415 ret = EXCP_INTERRUPT;
416 break;
418 #endif
420 } else if (replay_has_exception()
421 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
422 /* try to cause an exception pending in the log */
423 cpu_exec_nocache(cpu, 1, tb_find_fast(cpu), true);
424 ret = -1;
425 break;
428 next_tb = 0; /* force lookup of first TB */
429 for(;;) {
430 interrupt_request = cpu->interrupt_request;
431 if (unlikely(interrupt_request)) {
432 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
433 /* Mask out external interrupts for this step. */
434 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
436 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
437 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
438 cpu->exception_index = EXCP_DEBUG;
439 cpu_loop_exit(cpu);
441 if (replay_mode == REPLAY_MODE_PLAY
442 && !replay_has_interrupt()) {
443 /* Do nothing */
444 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
445 replay_interrupt();
446 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
447 cpu->halted = 1;
448 cpu->exception_index = EXCP_HLT;
449 cpu_loop_exit(cpu);
451 #if defined(TARGET_I386)
452 else if (interrupt_request & CPU_INTERRUPT_INIT) {
453 replay_interrupt();
454 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
455 do_cpu_init(x86_cpu);
456 cpu->exception_index = EXCP_HALTED;
457 cpu_loop_exit(cpu);
459 #else
460 else if (interrupt_request & CPU_INTERRUPT_RESET) {
461 replay_interrupt();
462 cpu_reset(cpu);
463 cpu_loop_exit(cpu);
465 #endif
466 /* The target hook has 3 exit conditions:
467 False when the interrupt isn't processed,
468 True when it is, and we should restart on a new TB,
469 and via longjmp via cpu_loop_exit. */
470 else {
471 replay_interrupt();
472 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
473 next_tb = 0;
476 /* Don't use the cached interrupt_request value,
477 do_interrupt may have updated the EXITTB flag. */
478 if (cpu->interrupt_request & CPU_INTERRUPT_EXITTB) {
479 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
480 /* ensure that no TB jump will be modified as
481 the program flow was changed */
482 next_tb = 0;
485 if (unlikely(cpu->exit_request
486 || replay_has_interrupt())) {
487 cpu->exit_request = 0;
488 cpu->exception_index = EXCP_INTERRUPT;
489 cpu_loop_exit(cpu);
491 tb_lock();
492 tb = tb_find_fast(cpu);
493 /* Note: we do it here to avoid a gcc bug on Mac OS X when
494 doing it in tb_find_slow */
495 if (tcg_ctx.tb_ctx.tb_invalidated_flag) {
496 /* as some TB could have been invalidated because
497 of memory exceptions while generating the code, we
498 must recompute the hash index here */
499 next_tb = 0;
500 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
502 if (qemu_loglevel_mask(CPU_LOG_EXEC)) {
503 qemu_log("Trace %p [" TARGET_FMT_lx "] %s\n",
504 tb->tc_ptr, tb->pc, lookup_symbol(tb->pc));
506 /* see if we can patch the calling TB. When the TB
507 spans two pages, we cannot safely do a direct
508 jump. */
509 if (next_tb != 0 && tb->page_addr[1] == -1
510 && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
511 tb_add_jump((TranslationBlock *)(next_tb & ~TB_EXIT_MASK),
512 next_tb & TB_EXIT_MASK, tb);
514 tb_unlock();
515 if (likely(!cpu->exit_request)) {
516 trace_exec_tb(tb, tb->pc);
517 tc_ptr = tb->tc_ptr;
518 /* execute the generated code */
519 cpu->current_tb = tb;
520 next_tb = cpu_tb_exec(cpu, tc_ptr);
521 cpu->current_tb = NULL;
522 switch (next_tb & TB_EXIT_MASK) {
523 case TB_EXIT_REQUESTED:
524 /* Something asked us to stop executing
525 * chained TBs; just continue round the main
526 * loop. Whatever requested the exit will also
527 * have set something else (eg exit_request or
528 * interrupt_request) which we will handle
529 * next time around the loop. But we need to
530 * ensure the tcg_exit_req read in generated code
531 * comes before the next read of cpu->exit_request
532 * or cpu->interrupt_request.
534 smp_rmb();
535 next_tb = 0;
536 break;
537 case TB_EXIT_ICOUNT_EXPIRED:
539 /* Instruction counter expired. */
540 int insns_left = cpu->icount_decr.u32;
541 if (cpu->icount_extra && insns_left >= 0) {
542 /* Refill decrementer and continue execution. */
543 cpu->icount_extra += insns_left;
544 insns_left = MIN(0xffff, cpu->icount_extra);
545 cpu->icount_extra -= insns_left;
546 cpu->icount_decr.u16.low = insns_left;
547 } else {
548 if (insns_left > 0) {
549 /* Execute remaining instructions. */
550 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
551 cpu_exec_nocache(cpu, insns_left, tb, false);
552 align_clocks(&sc, cpu);
554 cpu->exception_index = EXCP_INTERRUPT;
555 next_tb = 0;
556 cpu_loop_exit(cpu);
558 break;
560 default:
561 break;
564 /* Try to align the host and virtual clocks
565 if the guest is in advance */
566 align_clocks(&sc, cpu);
567 /* reset soft MMU for next block (it can currently
568 only be set by a memory fault) */
569 } /* for(;;) */
570 } else {
571 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
572 /* Some compilers wrongly smash all local variables after
573 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
574 * Reload essential local variables here for those compilers.
575 * Newer versions of gcc would complain about this code (-Wclobbered). */
576 cpu = current_cpu;
577 cc = CPU_GET_CLASS(cpu);
578 #ifdef TARGET_I386
579 x86_cpu = X86_CPU(cpu);
580 env = &x86_cpu->env;
581 #endif
582 #else /* buggy compiler */
583 /* Assert that the compiler does not smash local variables. */
584 g_assert(cpu == current_cpu);
585 g_assert(cc == CPU_GET_CLASS(cpu));
586 #ifdef TARGET_I386
587 g_assert(x86_cpu == X86_CPU(cpu));
588 g_assert(env == &x86_cpu->env);
589 #endif
590 #endif /* buggy compiler */
591 cpu->can_do_io = 1;
592 tb_lock_reset();
594 } /* for(;;) */
596 cc->cpu_exec_exit(cpu);
597 rcu_read_unlock();
599 /* fail safe : never use current_cpu outside cpu_exec() */
600 current_cpu = NULL;
602 /* Does not need atomic_mb_set because a spurious wakeup is okay. */
603 atomic_set(&tcg_current_cpu, NULL);
604 return ret;