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"
22 #include "disas/disas.h"
24 #include "qemu/atomic.h"
25 #include "sysemu/qtest.h"
26 #include "qemu/timer.h"
27 #include "exec/address-spaces.h"
29 #include "exec/tb-hash.h"
31 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
32 #include "hw/i386/apic.h"
34 #include "sysemu/replay.h"
36 /* -icount align implementation. */
38 typedef struct SyncClocks
{
40 int64_t last_cpu_icount
;
41 int64_t realtime_clock
;
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
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
54 static void align_clocks(SyncClocks
*sc
, const CPUState
*cpu
)
58 if (!icount_align_option
) {
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
;
66 if (sc
->diff_clk
> VM_CLOCK_ADVANCE
) {
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
;
77 Sleep(sc
->diff_clk
/ SCALE_MS
);
83 static void print_delay(const SyncClocks
*sc
)
85 static float threshold_delay
;
86 static int64_t last_realtime_clock
;
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",
100 last_realtime_clock
= sc
->realtime_clock
;
105 static void init_delay_params(SyncClocks
*sc
,
108 if (!icount_align_option
) {
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
;
117 if (sc
->diff_clk
> max_advance
) {
118 max_advance
= sc
->diff_clk
;
121 /* Print every 2s max if the guest is late. We limit the number
122 of printed messages to NB_PRINT_MAX(currently 100) */
126 static void align_clocks(SyncClocks
*sc
, const CPUState
*cpu
)
130 static void init_delay_params(SyncClocks
*sc
, const CPUState
*cpu
)
133 #endif /* CONFIG USER ONLY */
135 /* Execute a TB, and fix up the CPU state afterwards if necessary */
136 static inline tcg_target_ulong
cpu_tb_exec(CPUState
*cpu
, uint8_t *tb_ptr
)
138 CPUArchState
*env
= cpu
->env_ptr
;
141 #if defined(DEBUG_DISAS)
142 if (qemu_loglevel_mask(CPU_LOG_TB_CPU
)) {
143 #if defined(TARGET_I386)
144 log_cpu_state(cpu
, CPU_DUMP_CCOP
);
145 #elif defined(TARGET_M68K)
146 /* ??? Should not modify env state for dumping. */
147 cpu_m68k_flush_flags(env
, env
->cc_op
);
148 env
->cc_op
= CC_OP_FLAGS
;
149 env
->sr
= (env
->sr
& 0xffe0) | env
->cc_dest
| (env
->cc_x
<< 4);
150 log_cpu_state(cpu
, 0);
152 log_cpu_state(cpu
, 0);
155 #endif /* DEBUG_DISAS */
157 cpu
->can_do_io
= !use_icount
;
158 next_tb
= tcg_qemu_tb_exec(env
, tb_ptr
);
160 trace_exec_tb_exit((void *) (next_tb
& ~TB_EXIT_MASK
),
161 next_tb
& TB_EXIT_MASK
);
163 if ((next_tb
& TB_EXIT_MASK
) > TB_EXIT_IDX1
) {
164 /* We didn't start executing this TB (eg because the instruction
165 * counter hit zero); we must restore the guest PC to the address
166 * of the start of the TB.
168 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
169 TranslationBlock
*tb
= (TranslationBlock
*)(next_tb
& ~TB_EXIT_MASK
);
170 if (cc
->synchronize_from_tb
) {
171 cc
->synchronize_from_tb(cpu
, tb
);
174 cc
->set_pc(cpu
, tb
->pc
);
177 if ((next_tb
& TB_EXIT_MASK
) == TB_EXIT_REQUESTED
) {
178 /* We were asked to stop executing TBs (probably a pending
179 * interrupt. We've now stopped, so clear the flag.
181 cpu
->tcg_exit_req
= 0;
186 /* Execute the code without caching the generated code. An interpreter
187 could be used if available. */
188 static void cpu_exec_nocache(CPUState
*cpu
, int max_cycles
,
189 TranslationBlock
*orig_tb
, bool ignore_icount
)
191 TranslationBlock
*tb
;
193 /* Should never happen.
194 We only end up here when an existing TB is too long. */
195 if (max_cycles
> CF_COUNT_MASK
)
196 max_cycles
= CF_COUNT_MASK
;
198 tb
= tb_gen_code(cpu
, orig_tb
->pc
, orig_tb
->cs_base
, orig_tb
->flags
,
199 max_cycles
| CF_NOCACHE
200 | (ignore_icount
? CF_IGNORE_ICOUNT
: 0));
201 tb
->orig_tb
= tcg_ctx
.tb_ctx
.tb_invalidated_flag
? NULL
: orig_tb
;
202 cpu
->current_tb
= tb
;
203 /* execute the generated code */
204 trace_exec_tb_nocache(tb
, tb
->pc
);
205 cpu_tb_exec(cpu
, tb
->tc_ptr
);
206 cpu
->current_tb
= NULL
;
207 tb_phys_invalidate(tb
, -1);
211 static TranslationBlock
*tb_find_physical(CPUState
*cpu
,
213 target_ulong cs_base
,
216 CPUArchState
*env
= (CPUArchState
*)cpu
->env_ptr
;
217 TranslationBlock
*tb
, **ptb1
;
219 tb_page_addr_t phys_pc
, phys_page1
;
220 target_ulong virt_page2
;
222 tcg_ctx
.tb_ctx
.tb_invalidated_flag
= 0;
224 /* find translated block using physical mappings */
225 phys_pc
= get_page_addr_code(env
, pc
);
226 phys_page1
= phys_pc
& TARGET_PAGE_MASK
;
227 h
= tb_phys_hash_func(phys_pc
);
228 ptb1
= &tcg_ctx
.tb_ctx
.tb_phys_hash
[h
];
235 tb
->page_addr
[0] == phys_page1
&&
236 tb
->cs_base
== cs_base
&&
237 tb
->flags
== flags
) {
238 /* check next page if needed */
239 if (tb
->page_addr
[1] != -1) {
240 tb_page_addr_t phys_page2
;
242 virt_page2
= (pc
& TARGET_PAGE_MASK
) +
244 phys_page2
= get_page_addr_code(env
, virt_page2
);
245 if (tb
->page_addr
[1] == phys_page2
) {
252 ptb1
= &tb
->phys_hash_next
;
255 /* Move the TB to the head of the list */
256 *ptb1
= tb
->phys_hash_next
;
257 tb
->phys_hash_next
= tcg_ctx
.tb_ctx
.tb_phys_hash
[h
];
258 tcg_ctx
.tb_ctx
.tb_phys_hash
[h
] = tb
;
262 static TranslationBlock
*tb_find_slow(CPUState
*cpu
,
264 target_ulong cs_base
,
267 TranslationBlock
*tb
;
269 tb
= tb_find_physical(cpu
, pc
, cs_base
, flags
);
274 #ifdef CONFIG_USER_ONLY
275 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
276 * taken outside tb_lock. Since we're momentarily dropping
277 * tb_lock, there's a chance that our desired tb has been
283 tb
= tb_find_physical(cpu
, pc
, cs_base
, flags
);
290 /* if no translated code available, then translate it now */
291 tb
= tb_gen_code(cpu
, pc
, cs_base
, flags
, 0);
293 #ifdef CONFIG_USER_ONLY
298 /* we add the TB in the virtual pc hash table */
299 cpu
->tb_jmp_cache
[tb_jmp_cache_hash_func(pc
)] = tb
;
303 static inline TranslationBlock
*tb_find_fast(CPUState
*cpu
)
305 CPUArchState
*env
= (CPUArchState
*)cpu
->env_ptr
;
306 TranslationBlock
*tb
;
307 target_ulong cs_base
, pc
;
310 /* we record a subset of the CPU state. It will
311 always be the same before a given translated block
313 cpu_get_tb_cpu_state(env
, &pc
, &cs_base
, &flags
);
314 tb
= cpu
->tb_jmp_cache
[tb_jmp_cache_hash_func(pc
)];
315 if (unlikely(!tb
|| tb
->pc
!= pc
|| tb
->cs_base
!= cs_base
||
316 tb
->flags
!= flags
)) {
317 tb
= tb_find_slow(cpu
, pc
, cs_base
, flags
);
322 static void cpu_handle_debug_exception(CPUState
*cpu
)
324 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
327 if (!cpu
->watchpoint_hit
) {
328 QTAILQ_FOREACH(wp
, &cpu
->watchpoints
, entry
) {
329 wp
->flags
&= ~BP_WATCHPOINT_HIT
;
333 cc
->debug_excp_handler(cpu
);
336 /* main execution loop */
338 int cpu_exec(CPUState
*cpu
)
340 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
342 X86CPU
*x86_cpu
= X86_CPU(cpu
);
343 CPUArchState
*env
= &x86_cpu
->env
;
345 int ret
, interrupt_request
;
346 TranslationBlock
*tb
;
351 /* replay_interrupt may need current_cpu */
355 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
356 if ((cpu
->interrupt_request
& CPU_INTERRUPT_POLL
)
357 && replay_interrupt()) {
358 apic_poll_irq(x86_cpu
->apic_state
);
359 cpu_reset_interrupt(cpu
, CPU_INTERRUPT_POLL
);
362 if (!cpu_has_work(cpu
)) {
370 atomic_mb_set(&tcg_current_cpu
, cpu
);
373 if (unlikely(atomic_mb_read(&exit_request
))) {
374 cpu
->exit_request
= 1;
377 cc
->cpu_exec_enter(cpu
);
379 /* Calculate difference between guest clock and host clock.
380 * This delay includes the delay of the last cycle, so
381 * what we have to do is sleep until it is 0. As for the
382 * advance/delay we gain here, we try to fix it next time.
384 init_delay_params(&sc
, cpu
);
386 /* prepare setjmp context for exception handling */
388 if (sigsetjmp(cpu
->jmp_env
, 0) == 0) {
389 /* if an exception is pending, we execute it here */
390 if (cpu
->exception_index
>= 0) {
391 if (cpu
->exception_index
>= EXCP_INTERRUPT
) {
392 /* exit request from the cpu execution loop */
393 ret
= cpu
->exception_index
;
394 if (ret
== EXCP_DEBUG
) {
395 cpu_handle_debug_exception(cpu
);
397 cpu
->exception_index
= -1;
400 #if defined(CONFIG_USER_ONLY)
401 /* if user mode only, we simulate a fake exception
402 which will be handled outside the cpu execution
404 #if defined(TARGET_I386)
405 cc
->do_interrupt(cpu
);
407 ret
= cpu
->exception_index
;
408 cpu
->exception_index
= -1;
411 if (replay_exception()) {
412 cc
->do_interrupt(cpu
);
413 cpu
->exception_index
= -1;
414 } else if (!replay_has_interrupt()) {
415 /* give a chance to iothread in replay mode */
416 ret
= EXCP_INTERRUPT
;
421 } else if (replay_has_exception()
422 && cpu
->icount_decr
.u16
.low
+ cpu
->icount_extra
== 0) {
423 /* try to cause an exception pending in the log */
424 cpu_exec_nocache(cpu
, 1, tb_find_fast(cpu
), true);
429 next_tb
= 0; /* force lookup of first TB */
431 interrupt_request
= cpu
->interrupt_request
;
432 if (unlikely(interrupt_request
)) {
433 if (unlikely(cpu
->singlestep_enabled
& SSTEP_NOIRQ
)) {
434 /* Mask out external interrupts for this step. */
435 interrupt_request
&= ~CPU_INTERRUPT_SSTEP_MASK
;
437 if (interrupt_request
& CPU_INTERRUPT_DEBUG
) {
438 cpu
->interrupt_request
&= ~CPU_INTERRUPT_DEBUG
;
439 cpu
->exception_index
= EXCP_DEBUG
;
442 if (replay_mode
== REPLAY_MODE_PLAY
443 && !replay_has_interrupt()) {
445 } else if (interrupt_request
& CPU_INTERRUPT_HALT
) {
447 cpu
->interrupt_request
&= ~CPU_INTERRUPT_HALT
;
449 cpu
->exception_index
= EXCP_HLT
;
452 #if defined(TARGET_I386)
453 else if (interrupt_request
& CPU_INTERRUPT_INIT
) {
455 cpu_svm_check_intercept_param(env
, SVM_EXIT_INIT
, 0);
456 do_cpu_init(x86_cpu
);
457 cpu
->exception_index
= EXCP_HALTED
;
461 else if (interrupt_request
& CPU_INTERRUPT_RESET
) {
467 /* The target hook has 3 exit conditions:
468 False when the interrupt isn't processed,
469 True when it is, and we should restart on a new TB,
470 and via longjmp via cpu_loop_exit. */
473 if (cc
->cpu_exec_interrupt(cpu
, interrupt_request
)) {
477 /* Don't use the cached interrupt_request value,
478 do_interrupt may have updated the EXITTB flag. */
479 if (cpu
->interrupt_request
& CPU_INTERRUPT_EXITTB
) {
480 cpu
->interrupt_request
&= ~CPU_INTERRUPT_EXITTB
;
481 /* ensure that no TB jump will be modified as
482 the program flow was changed */
486 if (unlikely(cpu
->exit_request
487 || replay_has_interrupt())) {
488 cpu
->exit_request
= 0;
489 cpu
->exception_index
= EXCP_INTERRUPT
;
493 tb
= tb_find_fast(cpu
);
494 /* Note: we do it here to avoid a gcc bug on Mac OS X when
495 doing it in tb_find_slow */
496 if (tcg_ctx
.tb_ctx
.tb_invalidated_flag
) {
497 /* as some TB could have been invalidated because
498 of memory exceptions while generating the code, we
499 must recompute the hash index here */
501 tcg_ctx
.tb_ctx
.tb_invalidated_flag
= 0;
503 if (qemu_loglevel_mask(CPU_LOG_EXEC
)) {
504 qemu_log("Trace %p [" TARGET_FMT_lx
"] %s\n",
505 tb
->tc_ptr
, tb
->pc
, lookup_symbol(tb
->pc
));
507 /* see if we can patch the calling TB. When the TB
508 spans two pages, we cannot safely do a direct
510 if (next_tb
!= 0 && tb
->page_addr
[1] == -1
511 && !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN
)) {
512 tb_add_jump((TranslationBlock
*)(next_tb
& ~TB_EXIT_MASK
),
513 next_tb
& TB_EXIT_MASK
, tb
);
516 if (likely(!cpu
->exit_request
)) {
517 trace_exec_tb(tb
, tb
->pc
);
519 /* execute the generated code */
520 cpu
->current_tb
= tb
;
521 next_tb
= cpu_tb_exec(cpu
, tc_ptr
);
522 cpu
->current_tb
= NULL
;
523 switch (next_tb
& TB_EXIT_MASK
) {
524 case TB_EXIT_REQUESTED
:
525 /* Something asked us to stop executing
526 * chained TBs; just continue round the main
527 * loop. Whatever requested the exit will also
528 * have set something else (eg exit_request or
529 * interrupt_request) which we will handle
530 * next time around the loop. But we need to
531 * ensure the tcg_exit_req read in generated code
532 * comes before the next read of cpu->exit_request
533 * or cpu->interrupt_request.
538 case TB_EXIT_ICOUNT_EXPIRED
:
540 /* Instruction counter expired. */
541 int insns_left
= cpu
->icount_decr
.u32
;
542 if (cpu
->icount_extra
&& insns_left
>= 0) {
543 /* Refill decrementer and continue execution. */
544 cpu
->icount_extra
+= insns_left
;
545 insns_left
= MIN(0xffff, cpu
->icount_extra
);
546 cpu
->icount_extra
-= insns_left
;
547 cpu
->icount_decr
.u16
.low
= insns_left
;
549 if (insns_left
> 0) {
550 /* Execute remaining instructions. */
551 tb
= (TranslationBlock
*)(next_tb
& ~TB_EXIT_MASK
);
552 cpu_exec_nocache(cpu
, insns_left
, tb
, false);
553 align_clocks(&sc
, cpu
);
555 cpu
->exception_index
= EXCP_INTERRUPT
;
565 /* Try to align the host and virtual clocks
566 if the guest is in advance */
567 align_clocks(&sc
, cpu
);
568 /* reset soft MMU for next block (it can currently
569 only be set by a memory fault) */
572 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
573 /* Some compilers wrongly smash all local variables after
574 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
575 * Reload essential local variables here for those compilers.
576 * Newer versions of gcc would complain about this code (-Wclobbered). */
578 cc
= CPU_GET_CLASS(cpu
);
580 x86_cpu
= X86_CPU(cpu
);
583 #else /* buggy compiler */
584 /* Assert that the compiler does not smash local variables. */
585 g_assert(cpu
== current_cpu
);
586 g_assert(cc
== CPU_GET_CLASS(cpu
));
588 g_assert(x86_cpu
== X86_CPU(cpu
));
589 g_assert(env
== &x86_cpu
->env
);
591 #endif /* buggy compiler */
597 cc
->cpu_exec_exit(cpu
);
600 /* fail safe : never use current_cpu outside cpu_exec() */
603 /* Does not need atomic_mb_set because a spurious wakeup is okay. */
604 atomic_set(&tcg_current_cpu
, NULL
);