ipxe: update to 20161108 snapshot
[qemu.git] / cpu-exec.c
blob4188fed3c62c66545f515cb41401441360164588
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 qemu_log_lock();
154 #if defined(TARGET_I386)
155 log_cpu_state(cpu, CPU_DUMP_CCOP);
156 #else
157 log_cpu_state(cpu, 0);
158 #endif
159 qemu_log_unlock();
161 #endif /* DEBUG_DISAS */
163 cpu->can_do_io = !use_icount;
164 ret = tcg_qemu_tb_exec(env, tb_ptr);
165 cpu->can_do_io = 1;
166 last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
167 tb_exit = ret & TB_EXIT_MASK;
168 trace_exec_tb_exit(last_tb, tb_exit);
170 if (tb_exit > TB_EXIT_IDX1) {
171 /* We didn't start executing this TB (eg because the instruction
172 * counter hit zero); we must restore the guest PC to the address
173 * of the start of the TB.
175 CPUClass *cc = CPU_GET_CLASS(cpu);
176 qemu_log_mask_and_addr(CPU_LOG_EXEC, last_tb->pc,
177 "Stopped execution of TB chain before %p ["
178 TARGET_FMT_lx "] %s\n",
179 last_tb->tc_ptr, last_tb->pc,
180 lookup_symbol(last_tb->pc));
181 if (cc->synchronize_from_tb) {
182 cc->synchronize_from_tb(cpu, last_tb);
183 } else {
184 assert(cc->set_pc);
185 cc->set_pc(cpu, last_tb->pc);
188 if (tb_exit == TB_EXIT_REQUESTED) {
189 /* We were asked to stop executing TBs (probably a pending
190 * interrupt. We've now stopped, so clear the flag.
192 atomic_set(&cpu->tcg_exit_req, 0);
194 return ret;
197 #ifndef CONFIG_USER_ONLY
198 /* Execute the code without caching the generated code. An interpreter
199 could be used if available. */
200 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
201 TranslationBlock *orig_tb, bool ignore_icount)
203 TranslationBlock *tb;
205 /* Should never happen.
206 We only end up here when an existing TB is too long. */
207 if (max_cycles > CF_COUNT_MASK)
208 max_cycles = CF_COUNT_MASK;
210 tb_lock();
211 tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
212 max_cycles | CF_NOCACHE
213 | (ignore_icount ? CF_IGNORE_ICOUNT : 0));
214 tb->orig_tb = orig_tb;
215 tb_unlock();
217 /* execute the generated code */
218 trace_exec_tb_nocache(tb, tb->pc);
219 cpu_tb_exec(cpu, tb);
221 tb_lock();
222 tb_phys_invalidate(tb, -1);
223 tb_free(tb);
224 tb_unlock();
226 #endif
228 static void cpu_exec_step(CPUState *cpu)
230 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
231 TranslationBlock *tb;
232 target_ulong cs_base, pc;
233 uint32_t flags;
235 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
236 tb = tb_gen_code(cpu, pc, cs_base, flags,
237 1 | CF_NOCACHE | CF_IGNORE_ICOUNT);
238 tb->orig_tb = NULL;
239 /* execute the generated code */
240 trace_exec_tb_nocache(tb, pc);
241 cpu_tb_exec(cpu, tb);
242 tb_phys_invalidate(tb, -1);
243 tb_free(tb);
246 void cpu_exec_step_atomic(CPUState *cpu)
248 start_exclusive();
250 /* Since we got here, we know that parallel_cpus must be true. */
251 parallel_cpus = false;
252 cpu_exec_step(cpu);
253 parallel_cpus = true;
255 end_exclusive();
258 struct tb_desc {
259 target_ulong pc;
260 target_ulong cs_base;
261 CPUArchState *env;
262 tb_page_addr_t phys_page1;
263 uint32_t flags;
266 static bool tb_cmp(const void *p, const void *d)
268 const TranslationBlock *tb = p;
269 const struct tb_desc *desc = d;
271 if (tb->pc == desc->pc &&
272 tb->page_addr[0] == desc->phys_page1 &&
273 tb->cs_base == desc->cs_base &&
274 tb->flags == desc->flags &&
275 !atomic_read(&tb->invalid)) {
276 /* check next page if needed */
277 if (tb->page_addr[1] == -1) {
278 return true;
279 } else {
280 tb_page_addr_t phys_page2;
281 target_ulong virt_page2;
283 virt_page2 = (desc->pc & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE;
284 phys_page2 = get_page_addr_code(desc->env, virt_page2);
285 if (tb->page_addr[1] == phys_page2) {
286 return true;
290 return false;
293 static TranslationBlock *tb_htable_lookup(CPUState *cpu,
294 target_ulong pc,
295 target_ulong cs_base,
296 uint32_t flags)
298 tb_page_addr_t phys_pc;
299 struct tb_desc desc;
300 uint32_t h;
302 desc.env = (CPUArchState *)cpu->env_ptr;
303 desc.cs_base = cs_base;
304 desc.flags = flags;
305 desc.pc = pc;
306 phys_pc = get_page_addr_code(desc.env, pc);
307 desc.phys_page1 = phys_pc & TARGET_PAGE_MASK;
308 h = tb_hash_func(phys_pc, pc, flags);
309 return qht_lookup(&tcg_ctx.tb_ctx.htable, tb_cmp, &desc, h);
312 static inline TranslationBlock *tb_find(CPUState *cpu,
313 TranslationBlock *last_tb,
314 int tb_exit)
316 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
317 TranslationBlock *tb;
318 target_ulong cs_base, pc;
319 uint32_t flags;
320 bool have_tb_lock = false;
322 /* we record a subset of the CPU state. It will
323 always be the same before a given translated block
324 is executed. */
325 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
326 tb = atomic_rcu_read(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)]);
327 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
328 tb->flags != flags)) {
329 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
330 if (!tb) {
332 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
333 * taken outside tb_lock. As system emulation is currently
334 * single threaded the locks are NOPs.
336 mmap_lock();
337 tb_lock();
338 have_tb_lock = true;
340 /* There's a chance that our desired tb has been translated while
341 * taking the locks so we check again inside the lock.
343 tb = tb_htable_lookup(cpu, pc, cs_base, flags);
344 if (!tb) {
345 /* if no translated code available, then translate it now */
346 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
349 mmap_unlock();
352 /* We add the TB in the virtual pc hash table for the fast lookup */
353 atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)], tb);
355 #ifndef CONFIG_USER_ONLY
356 /* We don't take care of direct jumps when address mapping changes in
357 * system emulation. So it's not safe to make a direct jump to a TB
358 * spanning two pages because the mapping for the second page can change.
360 if (tb->page_addr[1] != -1) {
361 last_tb = NULL;
363 #endif
364 /* See if we can patch the calling TB. */
365 if (last_tb && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
366 if (!have_tb_lock) {
367 tb_lock();
368 have_tb_lock = true;
370 if (!tb->invalid) {
371 tb_add_jump(last_tb, tb_exit, tb);
374 if (have_tb_lock) {
375 tb_unlock();
377 return tb;
380 static inline bool cpu_handle_halt(CPUState *cpu)
382 if (cpu->halted) {
383 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
384 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
385 && replay_interrupt()) {
386 X86CPU *x86_cpu = X86_CPU(cpu);
387 apic_poll_irq(x86_cpu->apic_state);
388 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
390 #endif
391 if (!cpu_has_work(cpu)) {
392 current_cpu = NULL;
393 return true;
396 cpu->halted = 0;
399 return false;
402 static inline void cpu_handle_debug_exception(CPUState *cpu)
404 CPUClass *cc = CPU_GET_CLASS(cpu);
405 CPUWatchpoint *wp;
407 if (!cpu->watchpoint_hit) {
408 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
409 wp->flags &= ~BP_WATCHPOINT_HIT;
413 cc->debug_excp_handler(cpu);
416 static inline bool cpu_handle_exception(CPUState *cpu, int *ret)
418 if (cpu->exception_index >= 0) {
419 if (cpu->exception_index >= EXCP_INTERRUPT) {
420 /* exit request from the cpu execution loop */
421 *ret = cpu->exception_index;
422 if (*ret == EXCP_DEBUG) {
423 cpu_handle_debug_exception(cpu);
425 cpu->exception_index = -1;
426 return true;
427 } else {
428 #if defined(CONFIG_USER_ONLY)
429 /* if user mode only, we simulate a fake exception
430 which will be handled outside the cpu execution
431 loop */
432 #if defined(TARGET_I386)
433 CPUClass *cc = CPU_GET_CLASS(cpu);
434 cc->do_interrupt(cpu);
435 #endif
436 *ret = cpu->exception_index;
437 cpu->exception_index = -1;
438 return true;
439 #else
440 if (replay_exception()) {
441 CPUClass *cc = CPU_GET_CLASS(cpu);
442 cc->do_interrupt(cpu);
443 cpu->exception_index = -1;
444 } else if (!replay_has_interrupt()) {
445 /* give a chance to iothread in replay mode */
446 *ret = EXCP_INTERRUPT;
447 return true;
449 #endif
451 #ifndef CONFIG_USER_ONLY
452 } else if (replay_has_exception()
453 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
454 /* try to cause an exception pending in the log */
455 cpu_exec_nocache(cpu, 1, tb_find(cpu, NULL, 0), true);
456 *ret = -1;
457 return true;
458 #endif
461 return false;
464 static inline void cpu_handle_interrupt(CPUState *cpu,
465 TranslationBlock **last_tb)
467 CPUClass *cc = CPU_GET_CLASS(cpu);
468 int interrupt_request = cpu->interrupt_request;
470 if (unlikely(interrupt_request)) {
471 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
472 /* Mask out external interrupts for this step. */
473 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
475 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
476 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
477 cpu->exception_index = EXCP_DEBUG;
478 cpu_loop_exit(cpu);
480 if (replay_mode == REPLAY_MODE_PLAY && !replay_has_interrupt()) {
481 /* Do nothing */
482 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
483 replay_interrupt();
484 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
485 cpu->halted = 1;
486 cpu->exception_index = EXCP_HLT;
487 cpu_loop_exit(cpu);
489 #if defined(TARGET_I386)
490 else if (interrupt_request & CPU_INTERRUPT_INIT) {
491 X86CPU *x86_cpu = X86_CPU(cpu);
492 CPUArchState *env = &x86_cpu->env;
493 replay_interrupt();
494 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
495 do_cpu_init(x86_cpu);
496 cpu->exception_index = EXCP_HALTED;
497 cpu_loop_exit(cpu);
499 #else
500 else if (interrupt_request & CPU_INTERRUPT_RESET) {
501 replay_interrupt();
502 cpu_reset(cpu);
503 cpu_loop_exit(cpu);
505 #endif
506 /* The target hook has 3 exit conditions:
507 False when the interrupt isn't processed,
508 True when it is, and we should restart on a new TB,
509 and via longjmp via cpu_loop_exit. */
510 else {
511 replay_interrupt();
512 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
513 *last_tb = NULL;
515 /* The target hook may have updated the 'cpu->interrupt_request';
516 * reload the 'interrupt_request' value */
517 interrupt_request = cpu->interrupt_request;
519 if (interrupt_request & CPU_INTERRUPT_EXITTB) {
520 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
521 /* ensure that no TB jump will be modified as
522 the program flow was changed */
523 *last_tb = NULL;
526 if (unlikely(atomic_read(&cpu->exit_request) || replay_has_interrupt())) {
527 atomic_set(&cpu->exit_request, 0);
528 cpu->exception_index = EXCP_INTERRUPT;
529 cpu_loop_exit(cpu);
533 static inline void cpu_loop_exec_tb(CPUState *cpu, TranslationBlock *tb,
534 TranslationBlock **last_tb, int *tb_exit,
535 SyncClocks *sc)
537 uintptr_t ret;
539 if (unlikely(atomic_read(&cpu->exit_request))) {
540 return;
543 trace_exec_tb(tb, tb->pc);
544 ret = cpu_tb_exec(cpu, tb);
545 *last_tb = (TranslationBlock *)(ret & ~TB_EXIT_MASK);
546 *tb_exit = ret & TB_EXIT_MASK;
547 switch (*tb_exit) {
548 case TB_EXIT_REQUESTED:
549 /* Something asked us to stop executing
550 * chained TBs; just continue round the main
551 * loop. Whatever requested the exit will also
552 * have set something else (eg exit_request or
553 * interrupt_request) which we will handle
554 * next time around the loop. But we need to
555 * ensure the tcg_exit_req read in generated code
556 * comes before the next read of cpu->exit_request
557 * or cpu->interrupt_request.
559 smp_rmb();
560 *last_tb = NULL;
561 break;
562 case TB_EXIT_ICOUNT_EXPIRED:
564 /* Instruction counter expired. */
565 #ifdef CONFIG_USER_ONLY
566 abort();
567 #else
568 int insns_left = cpu->icount_decr.u32;
569 if (cpu->icount_extra && insns_left >= 0) {
570 /* Refill decrementer and continue execution. */
571 cpu->icount_extra += insns_left;
572 insns_left = MIN(0xffff, cpu->icount_extra);
573 cpu->icount_extra -= insns_left;
574 cpu->icount_decr.u16.low = insns_left;
575 } else {
576 if (insns_left > 0) {
577 /* Execute remaining instructions. */
578 cpu_exec_nocache(cpu, insns_left, *last_tb, false);
579 align_clocks(sc, cpu);
581 cpu->exception_index = EXCP_INTERRUPT;
582 *last_tb = NULL;
583 cpu_loop_exit(cpu);
585 break;
586 #endif
588 default:
589 break;
593 /* main execution loop */
595 int cpu_exec(CPUState *cpu)
597 CPUClass *cc = CPU_GET_CLASS(cpu);
598 int ret;
599 SyncClocks sc;
601 /* replay_interrupt may need current_cpu */
602 current_cpu = cpu;
604 if (cpu_handle_halt(cpu)) {
605 return EXCP_HALTED;
608 atomic_mb_set(&tcg_current_cpu, cpu);
609 rcu_read_lock();
611 if (unlikely(atomic_mb_read(&exit_request))) {
612 cpu->exit_request = 1;
615 cc->cpu_exec_enter(cpu);
617 /* Calculate difference between guest clock and host clock.
618 * This delay includes the delay of the last cycle, so
619 * what we have to do is sleep until it is 0. As for the
620 * advance/delay we gain here, we try to fix it next time.
622 init_delay_params(&sc, cpu);
624 for(;;) {
625 /* prepare setjmp context for exception handling */
626 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
627 TranslationBlock *tb, *last_tb = NULL;
628 int tb_exit = 0;
630 /* if an exception is pending, we execute it here */
631 if (cpu_handle_exception(cpu, &ret)) {
632 break;
635 for(;;) {
636 cpu_handle_interrupt(cpu, &last_tb);
637 tb = tb_find(cpu, last_tb, tb_exit);
638 cpu_loop_exec_tb(cpu, tb, &last_tb, &tb_exit, &sc);
639 /* Try to align the host and virtual clocks
640 if the guest is in advance */
641 align_clocks(&sc, cpu);
642 } /* for(;;) */
643 } else {
644 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
645 /* Some compilers wrongly smash all local variables after
646 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
647 * Reload essential local variables here for those compilers.
648 * Newer versions of gcc would complain about this code (-Wclobbered). */
649 cpu = current_cpu;
650 cc = CPU_GET_CLASS(cpu);
651 #else /* buggy compiler */
652 /* Assert that the compiler does not smash local variables. */
653 g_assert(cpu == current_cpu);
654 g_assert(cc == CPU_GET_CLASS(cpu));
655 #endif /* buggy compiler */
656 cpu->can_do_io = 1;
657 tb_lock_reset();
659 } /* for(;;) */
661 cc->cpu_exec_exit(cpu);
662 rcu_read_unlock();
664 /* fail safe : never use current_cpu outside cpu_exec() */
665 current_cpu = NULL;
667 /* Does not need atomic_mb_set because a spurious wakeup is okay. */
668 atomic_set(&tcg_current_cpu, NULL);
669 return ret;