]> git.proxmox.com Git - mirror_qemu.git/blob - cpu-exec.c
qemu-log: Improve the "exec" TB execution logging
[mirror_qemu.git] / cpu-exec.c
1 /*
2 * emulator main execution loop
3 *
4 * Copyright (c) 2003-2005 Fabrice Bellard
5 *
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.
10 *
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.
15 *
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/>.
18 */
19 #include "qemu/osdep.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 #include "exec/log.h"
31 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
32 #include "hw/i386/apic.h"
33 #endif
34 #include "sysemu/replay.h"
35
36 /* -icount align implementation. */
37
38 typedef struct SyncClocks {
39 int64_t diff_clk;
40 int64_t last_cpu_icount;
41 int64_t realtime_clock;
42 } SyncClocks;
43
44 #if !defined(CONFIG_USER_ONLY)
45 /* Allow the guest to have a max 3ms advance.
46 * The difference between the 2 clocks could therefore
47 * oscillate around 0.
48 */
49 #define VM_CLOCK_ADVANCE 3000000
50 #define THRESHOLD_REDUCE 1.5
51 #define MAX_DELAY_PRINT_RATE 2000000000LL
52 #define MAX_NB_PRINTS 100
53
54 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
55 {
56 int64_t cpu_icount;
57
58 if (!icount_align_option) {
59 return;
60 }
61
62 cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
63 sc->diff_clk += cpu_icount_to_ns(sc->last_cpu_icount - cpu_icount);
64 sc->last_cpu_icount = cpu_icount;
65
66 if (sc->diff_clk > VM_CLOCK_ADVANCE) {
67 #ifndef _WIN32
68 struct timespec sleep_delay, rem_delay;
69 sleep_delay.tv_sec = sc->diff_clk / 1000000000LL;
70 sleep_delay.tv_nsec = sc->diff_clk % 1000000000LL;
71 if (nanosleep(&sleep_delay, &rem_delay) < 0) {
72 sc->diff_clk = rem_delay.tv_sec * 1000000000LL + rem_delay.tv_nsec;
73 } else {
74 sc->diff_clk = 0;
75 }
76 #else
77 Sleep(sc->diff_clk / SCALE_MS);
78 sc->diff_clk = 0;
79 #endif
80 }
81 }
82
83 static void print_delay(const SyncClocks *sc)
84 {
85 static float threshold_delay;
86 static int64_t last_realtime_clock;
87 static int nb_prints;
88
89 if (icount_align_option &&
90 sc->realtime_clock - last_realtime_clock >= MAX_DELAY_PRINT_RATE &&
91 nb_prints < MAX_NB_PRINTS) {
92 if ((-sc->diff_clk / (float)1000000000LL > threshold_delay) ||
93 (-sc->diff_clk / (float)1000000000LL <
94 (threshold_delay - THRESHOLD_REDUCE))) {
95 threshold_delay = (-sc->diff_clk / 1000000000LL) + 1;
96 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
97 threshold_delay - 1,
98 threshold_delay);
99 nb_prints++;
100 last_realtime_clock = sc->realtime_clock;
101 }
102 }
103 }
104
105 static void init_delay_params(SyncClocks *sc,
106 const CPUState *cpu)
107 {
108 if (!icount_align_option) {
109 return;
110 }
111 sc->realtime_clock = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT);
112 sc->diff_clk = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) - sc->realtime_clock;
113 sc->last_cpu_icount = cpu->icount_extra + cpu->icount_decr.u16.low;
114 if (sc->diff_clk < max_delay) {
115 max_delay = sc->diff_clk;
116 }
117 if (sc->diff_clk > max_advance) {
118 max_advance = sc->diff_clk;
119 }
120
121 /* Print every 2s max if the guest is late. We limit the number
122 of printed messages to NB_PRINT_MAX(currently 100) */
123 print_delay(sc);
124 }
125 #else
126 static void align_clocks(SyncClocks *sc, const CPUState *cpu)
127 {
128 }
129
130 static void init_delay_params(SyncClocks *sc, const CPUState *cpu)
131 {
132 }
133 #endif /* CONFIG USER ONLY */
134
135 /* Execute a TB, and fix up the CPU state afterwards if necessary */
136 static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, TranslationBlock *itb)
137 {
138 CPUArchState *env = cpu->env_ptr;
139 uintptr_t next_tb;
140 uint8_t *tb_ptr = itb->tc_ptr;
141
142 qemu_log_mask(CPU_LOG_EXEC, "Trace %p [" TARGET_FMT_lx "] %s\n",
143 itb->tc_ptr, itb->pc, lookup_symbol(itb->pc));
144
145 #if defined(DEBUG_DISAS)
146 if (qemu_loglevel_mask(CPU_LOG_TB_CPU)) {
147 #if defined(TARGET_I386)
148 log_cpu_state(cpu, CPU_DUMP_CCOP);
149 #elif defined(TARGET_M68K)
150 /* ??? Should not modify env state for dumping. */
151 cpu_m68k_flush_flags(env, env->cc_op);
152 env->cc_op = CC_OP_FLAGS;
153 env->sr = (env->sr & 0xffe0) | env->cc_dest | (env->cc_x << 4);
154 log_cpu_state(cpu, 0);
155 #else
156 log_cpu_state(cpu, 0);
157 #endif
158 }
159 #endif /* DEBUG_DISAS */
160
161 cpu->can_do_io = !use_icount;
162 next_tb = tcg_qemu_tb_exec(env, tb_ptr);
163 cpu->can_do_io = 1;
164 trace_exec_tb_exit((void *) (next_tb & ~TB_EXIT_MASK),
165 next_tb & TB_EXIT_MASK);
166
167 if ((next_tb & TB_EXIT_MASK) > TB_EXIT_IDX1) {
168 /* We didn't start executing this TB (eg because the instruction
169 * counter hit zero); we must restore the guest PC to the address
170 * of the start of the TB.
171 */
172 CPUClass *cc = CPU_GET_CLASS(cpu);
173 TranslationBlock *tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
174 qemu_log_mask(CPU_LOG_EXEC,
175 "Stopped execution of TB chain before %p ["
176 TARGET_FMT_lx "] %s\n",
177 itb->tc_ptr, itb->pc, lookup_symbol(itb->pc));
178 if (cc->synchronize_from_tb) {
179 cc->synchronize_from_tb(cpu, tb);
180 } else {
181 assert(cc->set_pc);
182 cc->set_pc(cpu, tb->pc);
183 }
184 }
185 if ((next_tb & TB_EXIT_MASK) == TB_EXIT_REQUESTED) {
186 /* We were asked to stop executing TBs (probably a pending
187 * interrupt. We've now stopped, so clear the flag.
188 */
189 cpu->tcg_exit_req = 0;
190 }
191 return next_tb;
192 }
193
194 /* Execute the code without caching the generated code. An interpreter
195 could be used if available. */
196 static void cpu_exec_nocache(CPUState *cpu, int max_cycles,
197 TranslationBlock *orig_tb, bool ignore_icount)
198 {
199 TranslationBlock *tb;
200
201 /* Should never happen.
202 We only end up here when an existing TB is too long. */
203 if (max_cycles > CF_COUNT_MASK)
204 max_cycles = CF_COUNT_MASK;
205
206 tb = tb_gen_code(cpu, orig_tb->pc, orig_tb->cs_base, orig_tb->flags,
207 max_cycles | CF_NOCACHE
208 | (ignore_icount ? CF_IGNORE_ICOUNT : 0));
209 tb->orig_tb = tcg_ctx.tb_ctx.tb_invalidated_flag ? NULL : orig_tb;
210 cpu->current_tb = tb;
211 /* execute the generated code */
212 trace_exec_tb_nocache(tb, tb->pc);
213 cpu_tb_exec(cpu, tb);
214 cpu->current_tb = NULL;
215 tb_phys_invalidate(tb, -1);
216 tb_free(tb);
217 }
218
219 static TranslationBlock *tb_find_physical(CPUState *cpu,
220 target_ulong pc,
221 target_ulong cs_base,
222 uint64_t flags)
223 {
224 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
225 TranslationBlock *tb, **ptb1;
226 unsigned int h;
227 tb_page_addr_t phys_pc, phys_page1;
228 target_ulong virt_page2;
229
230 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
231
232 /* find translated block using physical mappings */
233 phys_pc = get_page_addr_code(env, pc);
234 phys_page1 = phys_pc & TARGET_PAGE_MASK;
235 h = tb_phys_hash_func(phys_pc);
236 ptb1 = &tcg_ctx.tb_ctx.tb_phys_hash[h];
237 for(;;) {
238 tb = *ptb1;
239 if (!tb) {
240 return NULL;
241 }
242 if (tb->pc == pc &&
243 tb->page_addr[0] == phys_page1 &&
244 tb->cs_base == cs_base &&
245 tb->flags == flags) {
246 /* check next page if needed */
247 if (tb->page_addr[1] != -1) {
248 tb_page_addr_t phys_page2;
249
250 virt_page2 = (pc & TARGET_PAGE_MASK) +
251 TARGET_PAGE_SIZE;
252 phys_page2 = get_page_addr_code(env, virt_page2);
253 if (tb->page_addr[1] == phys_page2) {
254 break;
255 }
256 } else {
257 break;
258 }
259 }
260 ptb1 = &tb->phys_hash_next;
261 }
262
263 /* Move the TB to the head of the list */
264 *ptb1 = tb->phys_hash_next;
265 tb->phys_hash_next = tcg_ctx.tb_ctx.tb_phys_hash[h];
266 tcg_ctx.tb_ctx.tb_phys_hash[h] = tb;
267 return tb;
268 }
269
270 static TranslationBlock *tb_find_slow(CPUState *cpu,
271 target_ulong pc,
272 target_ulong cs_base,
273 uint64_t flags)
274 {
275 TranslationBlock *tb;
276
277 tb = tb_find_physical(cpu, pc, cs_base, flags);
278 if (tb) {
279 goto found;
280 }
281
282 #ifdef CONFIG_USER_ONLY
283 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
284 * taken outside tb_lock. Since we're momentarily dropping
285 * tb_lock, there's a chance that our desired tb has been
286 * translated.
287 */
288 tb_unlock();
289 mmap_lock();
290 tb_lock();
291 tb = tb_find_physical(cpu, pc, cs_base, flags);
292 if (tb) {
293 mmap_unlock();
294 goto found;
295 }
296 #endif
297
298 /* if no translated code available, then translate it now */
299 tb = tb_gen_code(cpu, pc, cs_base, flags, 0);
300
301 #ifdef CONFIG_USER_ONLY
302 mmap_unlock();
303 #endif
304
305 found:
306 /* we add the TB in the virtual pc hash table */
307 cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)] = tb;
308 return tb;
309 }
310
311 static inline TranslationBlock *tb_find_fast(CPUState *cpu)
312 {
313 CPUArchState *env = (CPUArchState *)cpu->env_ptr;
314 TranslationBlock *tb;
315 target_ulong cs_base, pc;
316 int flags;
317
318 /* we record a subset of the CPU state. It will
319 always be the same before a given translated block
320 is executed. */
321 cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
322 tb = cpu->tb_jmp_cache[tb_jmp_cache_hash_func(pc)];
323 if (unlikely(!tb || tb->pc != pc || tb->cs_base != cs_base ||
324 tb->flags != flags)) {
325 tb = tb_find_slow(cpu, pc, cs_base, flags);
326 }
327 return tb;
328 }
329
330 static void cpu_handle_debug_exception(CPUState *cpu)
331 {
332 CPUClass *cc = CPU_GET_CLASS(cpu);
333 CPUWatchpoint *wp;
334
335 if (!cpu->watchpoint_hit) {
336 QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) {
337 wp->flags &= ~BP_WATCHPOINT_HIT;
338 }
339 }
340
341 cc->debug_excp_handler(cpu);
342 }
343
344 /* main execution loop */
345
346 int cpu_exec(CPUState *cpu)
347 {
348 CPUClass *cc = CPU_GET_CLASS(cpu);
349 #ifdef TARGET_I386
350 X86CPU *x86_cpu = X86_CPU(cpu);
351 CPUArchState *env = &x86_cpu->env;
352 #endif
353 int ret, interrupt_request;
354 TranslationBlock *tb;
355 uintptr_t next_tb;
356 SyncClocks sc;
357
358 /* replay_interrupt may need current_cpu */
359 current_cpu = cpu;
360
361 if (cpu->halted) {
362 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
363 if ((cpu->interrupt_request & CPU_INTERRUPT_POLL)
364 && replay_interrupt()) {
365 apic_poll_irq(x86_cpu->apic_state);
366 cpu_reset_interrupt(cpu, CPU_INTERRUPT_POLL);
367 }
368 #endif
369 if (!cpu_has_work(cpu)) {
370 current_cpu = NULL;
371 return EXCP_HALTED;
372 }
373
374 cpu->halted = 0;
375 }
376
377 atomic_mb_set(&tcg_current_cpu, cpu);
378 rcu_read_lock();
379
380 if (unlikely(atomic_mb_read(&exit_request))) {
381 cpu->exit_request = 1;
382 }
383
384 cc->cpu_exec_enter(cpu);
385
386 /* Calculate difference between guest clock and host clock.
387 * This delay includes the delay of the last cycle, so
388 * what we have to do is sleep until it is 0. As for the
389 * advance/delay we gain here, we try to fix it next time.
390 */
391 init_delay_params(&sc, cpu);
392
393 /* prepare setjmp context for exception handling */
394 for(;;) {
395 if (sigsetjmp(cpu->jmp_env, 0) == 0) {
396 /* if an exception is pending, we execute it here */
397 if (cpu->exception_index >= 0) {
398 if (cpu->exception_index >= EXCP_INTERRUPT) {
399 /* exit request from the cpu execution loop */
400 ret = cpu->exception_index;
401 if (ret == EXCP_DEBUG) {
402 cpu_handle_debug_exception(cpu);
403 }
404 cpu->exception_index = -1;
405 break;
406 } else {
407 #if defined(CONFIG_USER_ONLY)
408 /* if user mode only, we simulate a fake exception
409 which will be handled outside the cpu execution
410 loop */
411 #if defined(TARGET_I386)
412 cc->do_interrupt(cpu);
413 #endif
414 ret = cpu->exception_index;
415 cpu->exception_index = -1;
416 break;
417 #else
418 if (replay_exception()) {
419 cc->do_interrupt(cpu);
420 cpu->exception_index = -1;
421 } else if (!replay_has_interrupt()) {
422 /* give a chance to iothread in replay mode */
423 ret = EXCP_INTERRUPT;
424 break;
425 }
426 #endif
427 }
428 } else if (replay_has_exception()
429 && cpu->icount_decr.u16.low + cpu->icount_extra == 0) {
430 /* try to cause an exception pending in the log */
431 cpu_exec_nocache(cpu, 1, tb_find_fast(cpu), true);
432 ret = -1;
433 break;
434 }
435
436 next_tb = 0; /* force lookup of first TB */
437 for(;;) {
438 interrupt_request = cpu->interrupt_request;
439 if (unlikely(interrupt_request)) {
440 if (unlikely(cpu->singlestep_enabled & SSTEP_NOIRQ)) {
441 /* Mask out external interrupts for this step. */
442 interrupt_request &= ~CPU_INTERRUPT_SSTEP_MASK;
443 }
444 if (interrupt_request & CPU_INTERRUPT_DEBUG) {
445 cpu->interrupt_request &= ~CPU_INTERRUPT_DEBUG;
446 cpu->exception_index = EXCP_DEBUG;
447 cpu_loop_exit(cpu);
448 }
449 if (replay_mode == REPLAY_MODE_PLAY
450 && !replay_has_interrupt()) {
451 /* Do nothing */
452 } else if (interrupt_request & CPU_INTERRUPT_HALT) {
453 replay_interrupt();
454 cpu->interrupt_request &= ~CPU_INTERRUPT_HALT;
455 cpu->halted = 1;
456 cpu->exception_index = EXCP_HLT;
457 cpu_loop_exit(cpu);
458 }
459 #if defined(TARGET_I386)
460 else if (interrupt_request & CPU_INTERRUPT_INIT) {
461 replay_interrupt();
462 cpu_svm_check_intercept_param(env, SVM_EXIT_INIT, 0);
463 do_cpu_init(x86_cpu);
464 cpu->exception_index = EXCP_HALTED;
465 cpu_loop_exit(cpu);
466 }
467 #else
468 else if (interrupt_request & CPU_INTERRUPT_RESET) {
469 replay_interrupt();
470 cpu_reset(cpu);
471 cpu_loop_exit(cpu);
472 }
473 #endif
474 /* The target hook has 3 exit conditions:
475 False when the interrupt isn't processed,
476 True when it is, and we should restart on a new TB,
477 and via longjmp via cpu_loop_exit. */
478 else {
479 replay_interrupt();
480 if (cc->cpu_exec_interrupt(cpu, interrupt_request)) {
481 next_tb = 0;
482 }
483 }
484 /* Don't use the cached interrupt_request value,
485 do_interrupt may have updated the EXITTB flag. */
486 if (cpu->interrupt_request & CPU_INTERRUPT_EXITTB) {
487 cpu->interrupt_request &= ~CPU_INTERRUPT_EXITTB;
488 /* ensure that no TB jump will be modified as
489 the program flow was changed */
490 next_tb = 0;
491 }
492 }
493 if (unlikely(cpu->exit_request
494 || replay_has_interrupt())) {
495 cpu->exit_request = 0;
496 cpu->exception_index = EXCP_INTERRUPT;
497 cpu_loop_exit(cpu);
498 }
499 tb_lock();
500 tb = tb_find_fast(cpu);
501 /* Note: we do it here to avoid a gcc bug on Mac OS X when
502 doing it in tb_find_slow */
503 if (tcg_ctx.tb_ctx.tb_invalidated_flag) {
504 /* as some TB could have been invalidated because
505 of memory exceptions while generating the code, we
506 must recompute the hash index here */
507 next_tb = 0;
508 tcg_ctx.tb_ctx.tb_invalidated_flag = 0;
509 }
510 /* see if we can patch the calling TB. When the TB
511 spans two pages, we cannot safely do a direct
512 jump. */
513 if (next_tb != 0 && tb->page_addr[1] == -1
514 && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN)) {
515 tb_add_jump((TranslationBlock *)(next_tb & ~TB_EXIT_MASK),
516 next_tb & TB_EXIT_MASK, tb);
517 }
518 tb_unlock();
519 if (likely(!cpu->exit_request)) {
520 trace_exec_tb(tb, tb->pc);
521 /* execute the generated code */
522 cpu->current_tb = tb;
523 next_tb = cpu_tb_exec(cpu, tb);
524 cpu->current_tb = NULL;
525 switch (next_tb & TB_EXIT_MASK) {
526 case TB_EXIT_REQUESTED:
527 /* Something asked us to stop executing
528 * chained TBs; just continue round the main
529 * loop. Whatever requested the exit will also
530 * have set something else (eg exit_request or
531 * interrupt_request) which we will handle
532 * next time around the loop. But we need to
533 * ensure the tcg_exit_req read in generated code
534 * comes before the next read of cpu->exit_request
535 * or cpu->interrupt_request.
536 */
537 smp_rmb();
538 next_tb = 0;
539 break;
540 case TB_EXIT_ICOUNT_EXPIRED:
541 {
542 /* Instruction counter expired. */
543 int insns_left = cpu->icount_decr.u32;
544 if (cpu->icount_extra && insns_left >= 0) {
545 /* Refill decrementer and continue execution. */
546 cpu->icount_extra += insns_left;
547 insns_left = MIN(0xffff, cpu->icount_extra);
548 cpu->icount_extra -= insns_left;
549 cpu->icount_decr.u16.low = insns_left;
550 } else {
551 if (insns_left > 0) {
552 /* Execute remaining instructions. */
553 tb = (TranslationBlock *)(next_tb & ~TB_EXIT_MASK);
554 cpu_exec_nocache(cpu, insns_left, tb, false);
555 align_clocks(&sc, cpu);
556 }
557 cpu->exception_index = EXCP_INTERRUPT;
558 next_tb = 0;
559 cpu_loop_exit(cpu);
560 }
561 break;
562 }
563 default:
564 break;
565 }
566 }
567 /* Try to align the host and virtual clocks
568 if the guest is in advance */
569 align_clocks(&sc, cpu);
570 /* reset soft MMU for next block (it can currently
571 only be set by a memory fault) */
572 } /* for(;;) */
573 } else {
574 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
575 /* Some compilers wrongly smash all local variables after
576 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
577 * Reload essential local variables here for those compilers.
578 * Newer versions of gcc would complain about this code (-Wclobbered). */
579 cpu = current_cpu;
580 cc = CPU_GET_CLASS(cpu);
581 #ifdef TARGET_I386
582 x86_cpu = X86_CPU(cpu);
583 env = &x86_cpu->env;
584 #endif
585 #else /* buggy compiler */
586 /* Assert that the compiler does not smash local variables. */
587 g_assert(cpu == current_cpu);
588 g_assert(cc == CPU_GET_CLASS(cpu));
589 #ifdef TARGET_I386
590 g_assert(x86_cpu == X86_CPU(cpu));
591 g_assert(env == &x86_cpu->env);
592 #endif
593 #endif /* buggy compiler */
594 cpu->can_do_io = 1;
595 tb_lock_reset();
596 }
597 } /* for(;;) */
598
599 cc->cpu_exec_exit(cpu);
600 rcu_read_unlock();
601
602 /* fail safe : never use current_cpu outside cpu_exec() */
603 current_cpu = NULL;
604
605 /* Does not need atomic_mb_set because a spurious wakeup is okay. */
606 atomic_set(&tcg_current_cpu, NULL);
607 return ret;
608 }