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"
23 #include "exec/exec-all.h"
25 #include "qemu/atomic.h"
26 #include "sysemu/qtest.h"
27 #include "qemu/timer.h"
28 #include "exec/address-spaces.h"
30 #include "exec/tb-hash.h"
31 #include "exec/tb-lookup.h"
33 #include "qemu/main-loop.h"
34 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
35 #include "hw/i386/apic.h"
37 #include "sysemu/cpus.h"
38 #include "sysemu/replay.h"
40 /* -icount align implementation. */
42 typedef struct SyncClocks
{
44 int64_t last_cpu_icount
;
45 int64_t realtime_clock
;
48 #if !defined(CONFIG_USER_ONLY)
49 /* Allow the guest to have a max 3ms advance.
50 * The difference between the 2 clocks could therefore
53 #define VM_CLOCK_ADVANCE 3000000
54 #define THRESHOLD_REDUCE 1.5
55 #define MAX_DELAY_PRINT_RATE 2000000000LL
56 #define MAX_NB_PRINTS 100
58 static void align_clocks(SyncClocks
*sc
, const CPUState
*cpu
)
62 if (!icount_align_option
) {
66 cpu_icount
= cpu
->icount_extra
+ cpu
->icount_decr
.u16
.low
;
67 sc
->diff_clk
+= cpu_icount_to_ns(sc
->last_cpu_icount
- cpu_icount
);
68 sc
->last_cpu_icount
= cpu_icount
;
70 if (sc
->diff_clk
> VM_CLOCK_ADVANCE
) {
72 struct timespec sleep_delay
, rem_delay
;
73 sleep_delay
.tv_sec
= sc
->diff_clk
/ 1000000000LL;
74 sleep_delay
.tv_nsec
= sc
->diff_clk
% 1000000000LL;
75 if (nanosleep(&sleep_delay
, &rem_delay
) < 0) {
76 sc
->diff_clk
= rem_delay
.tv_sec
* 1000000000LL + rem_delay
.tv_nsec
;
81 Sleep(sc
->diff_clk
/ SCALE_MS
);
87 static void print_delay(const SyncClocks
*sc
)
89 static float threshold_delay
;
90 static int64_t last_realtime_clock
;
93 if (icount_align_option
&&
94 sc
->realtime_clock
- last_realtime_clock
>= MAX_DELAY_PRINT_RATE
&&
95 nb_prints
< MAX_NB_PRINTS
) {
96 if ((-sc
->diff_clk
/ (float)1000000000LL > threshold_delay
) ||
97 (-sc
->diff_clk
/ (float)1000000000LL <
98 (threshold_delay
- THRESHOLD_REDUCE
))) {
99 threshold_delay
= (-sc
->diff_clk
/ 1000000000LL) + 1;
100 printf("Warning: The guest is now late by %.1f to %.1f seconds\n",
104 last_realtime_clock
= sc
->realtime_clock
;
109 static void init_delay_params(SyncClocks
*sc
,
112 if (!icount_align_option
) {
115 sc
->realtime_clock
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL_RT
);
116 sc
->diff_clk
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
) - sc
->realtime_clock
;
117 sc
->last_cpu_icount
= cpu
->icount_extra
+ cpu
->icount_decr
.u16
.low
;
118 if (sc
->diff_clk
< max_delay
) {
119 max_delay
= sc
->diff_clk
;
121 if (sc
->diff_clk
> max_advance
) {
122 max_advance
= sc
->diff_clk
;
125 /* Print every 2s max if the guest is late. We limit the number
126 of printed messages to NB_PRINT_MAX(currently 100) */
130 static void align_clocks(SyncClocks
*sc
, const CPUState
*cpu
)
134 static void init_delay_params(SyncClocks
*sc
, const CPUState
*cpu
)
137 #endif /* CONFIG USER ONLY */
139 /* Execute a TB, and fix up the CPU state afterwards if necessary */
140 static inline tcg_target_ulong
cpu_tb_exec(CPUState
*cpu
, TranslationBlock
*itb
)
142 CPUArchState
*env
= cpu
->env_ptr
;
144 TranslationBlock
*last_tb
;
146 uint8_t *tb_ptr
= itb
->tc
.ptr
;
148 qemu_log_mask_and_addr(CPU_LOG_EXEC
, itb
->pc
,
149 "Trace %p [%d: " TARGET_FMT_lx
"] %s\n",
150 itb
->tc
.ptr
, cpu
->cpu_index
, itb
->pc
,
151 lookup_symbol(itb
->pc
));
153 #if defined(DEBUG_DISAS)
154 if (qemu_loglevel_mask(CPU_LOG_TB_CPU
)
155 && qemu_log_in_addr_range(itb
->pc
)) {
157 #if defined(TARGET_I386)
158 log_cpu_state(cpu
, CPU_DUMP_CCOP
);
160 log_cpu_state(cpu
, 0);
164 #endif /* DEBUG_DISAS */
166 cpu
->can_do_io
= !use_icount
;
167 ret
= tcg_qemu_tb_exec(env
, tb_ptr
);
169 last_tb
= (TranslationBlock
*)(ret
& ~TB_EXIT_MASK
);
170 tb_exit
= ret
& TB_EXIT_MASK
;
171 trace_exec_tb_exit(last_tb
, tb_exit
);
173 if (tb_exit
> TB_EXIT_IDX1
) {
174 /* We didn't start executing this TB (eg because the instruction
175 * counter hit zero); we must restore the guest PC to the address
176 * of the start of the TB.
178 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
179 qemu_log_mask_and_addr(CPU_LOG_EXEC
, last_tb
->pc
,
180 "Stopped execution of TB chain before %p ["
181 TARGET_FMT_lx
"] %s\n",
182 last_tb
->tc
.ptr
, last_tb
->pc
,
183 lookup_symbol(last_tb
->pc
));
184 if (cc
->synchronize_from_tb
) {
185 cc
->synchronize_from_tb(cpu
, last_tb
);
188 cc
->set_pc(cpu
, last_tb
->pc
);
194 #ifndef CONFIG_USER_ONLY
195 /* Execute the code without caching the generated code. An interpreter
196 could be used if available. */
197 static void cpu_exec_nocache(CPUState
*cpu
, int max_cycles
,
198 TranslationBlock
*orig_tb
, bool ignore_icount
)
200 TranslationBlock
*tb
;
202 /* Should never happen.
203 We only end up here when an existing TB is too long. */
204 if (max_cycles
> CF_COUNT_MASK
)
205 max_cycles
= CF_COUNT_MASK
;
208 tb
= tb_gen_code(cpu
, orig_tb
->pc
, orig_tb
->cs_base
, orig_tb
->flags
,
209 max_cycles
| CF_NOCACHE
210 | (ignore_icount
? CF_IGNORE_ICOUNT
: 0));
211 tb
->orig_tb
= orig_tb
;
214 /* execute the generated code */
215 trace_exec_tb_nocache(tb
, tb
->pc
);
216 cpu_tb_exec(cpu
, tb
);
219 tb_phys_invalidate(tb
, -1);
225 static void cpu_exec_step(CPUState
*cpu
)
227 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
228 CPUArchState
*env
= (CPUArchState
*)cpu
->env_ptr
;
229 TranslationBlock
*tb
;
230 target_ulong cs_base
, pc
;
233 cpu_get_tb_cpu_state(env
, &pc
, &cs_base
, &flags
);
234 if (sigsetjmp(cpu
->jmp_env
, 0) == 0) {
237 tb
= tb_gen_code(cpu
, pc
, cs_base
, flags
,
238 1 | CF_NOCACHE
| CF_IGNORE_ICOUNT
);
243 cc
->cpu_exec_enter(cpu
);
244 /* execute the generated code */
245 trace_exec_tb_nocache(tb
, pc
);
246 cpu_tb_exec(cpu
, tb
);
247 cc
->cpu_exec_exit(cpu
);
250 tb_phys_invalidate(tb
, -1);
254 /* We may have exited due to another problem here, so we need
255 * to reset any tb_locks we may have taken but didn't release.
256 * The mmap_lock is dropped by tb_gen_code if it runs out of
259 #ifndef CONFIG_SOFTMMU
260 tcg_debug_assert(!have_mmap_lock());
266 void cpu_exec_step_atomic(CPUState
*cpu
)
270 /* Since we got here, we know that parallel_cpus must be true. */
271 parallel_cpus
= false;
273 parallel_cpus
= true;
280 target_ulong cs_base
;
282 tb_page_addr_t phys_page1
;
284 uint32_t trace_vcpu_dstate
;
287 static bool tb_cmp(const void *p
, const void *d
)
289 const TranslationBlock
*tb
= p
;
290 const struct tb_desc
*desc
= d
;
292 if (tb
->pc
== desc
->pc
&&
293 tb
->page_addr
[0] == desc
->phys_page1
&&
294 tb
->cs_base
== desc
->cs_base
&&
295 tb
->flags
== desc
->flags
&&
296 tb
->trace_vcpu_dstate
== desc
->trace_vcpu_dstate
&&
297 !(atomic_read(&tb
->cflags
) & CF_INVALID
)) {
298 /* check next page if needed */
299 if (tb
->page_addr
[1] == -1) {
302 tb_page_addr_t phys_page2
;
303 target_ulong virt_page2
;
305 virt_page2
= (desc
->pc
& TARGET_PAGE_MASK
) + TARGET_PAGE_SIZE
;
306 phys_page2
= get_page_addr_code(desc
->env
, virt_page2
);
307 if (tb
->page_addr
[1] == phys_page2
) {
315 TranslationBlock
*tb_htable_lookup(CPUState
*cpu
, target_ulong pc
,
316 target_ulong cs_base
, uint32_t flags
)
318 tb_page_addr_t phys_pc
;
322 desc
.env
= (CPUArchState
*)cpu
->env_ptr
;
323 desc
.cs_base
= cs_base
;
325 desc
.trace_vcpu_dstate
= *cpu
->trace_dstate
;
327 phys_pc
= get_page_addr_code(desc
.env
, pc
);
328 desc
.phys_page1
= phys_pc
& TARGET_PAGE_MASK
;
329 h
= tb_hash_func(phys_pc
, pc
, flags
, *cpu
->trace_dstate
);
330 return qht_lookup(&tcg_ctx
.tb_ctx
.htable
, tb_cmp
, &desc
, h
);
333 void tb_set_jmp_target(TranslationBlock
*tb
, int n
, uintptr_t addr
)
335 if (TCG_TARGET_HAS_direct_jump
) {
336 uintptr_t offset
= tb
->jmp_target_arg
[n
];
337 uintptr_t tc_ptr
= (uintptr_t)tb
->tc
.ptr
;
338 tb_target_set_jmp_target(tc_ptr
, tc_ptr
+ offset
, addr
);
340 tb
->jmp_target_arg
[n
] = addr
;
344 /* Called with tb_lock held. */
345 static inline void tb_add_jump(TranslationBlock
*tb
, int n
,
346 TranslationBlock
*tb_next
)
348 assert(n
< ARRAY_SIZE(tb
->jmp_list_next
));
349 if (tb
->jmp_list_next
[n
]) {
350 /* Another thread has already done this while we were
351 * outside of the lock; nothing to do in this case */
354 qemu_log_mask_and_addr(CPU_LOG_EXEC
, tb
->pc
,
355 "Linking TBs %p [" TARGET_FMT_lx
356 "] index %d -> %p [" TARGET_FMT_lx
"]\n",
357 tb
->tc
.ptr
, tb
->pc
, n
,
358 tb_next
->tc
.ptr
, tb_next
->pc
);
360 /* patch the native jump address */
361 tb_set_jmp_target(tb
, n
, (uintptr_t)tb_next
->tc
.ptr
);
363 /* add in TB jmp circular list */
364 tb
->jmp_list_next
[n
] = tb_next
->jmp_list_first
;
365 tb_next
->jmp_list_first
= (uintptr_t)tb
| n
;
368 static inline TranslationBlock
*tb_find(CPUState
*cpu
,
369 TranslationBlock
*last_tb
,
372 TranslationBlock
*tb
;
373 target_ulong cs_base
, pc
;
375 bool acquired_tb_lock
= false;
377 tb
= tb_lookup__cpu_state(cpu
, &pc
, &cs_base
, &flags
);
379 /* mmap_lock is needed by tb_gen_code, and mmap_lock must be
380 * taken outside tb_lock. As system emulation is currently
381 * single threaded the locks are NOPs.
385 acquired_tb_lock
= true;
387 /* There's a chance that our desired tb has been translated while
388 * taking the locks so we check again inside the lock.
390 tb
= tb_htable_lookup(cpu
, pc
, cs_base
, flags
);
391 if (likely(tb
== NULL
)) {
392 /* if no translated code available, then translate it now */
393 tb
= tb_gen_code(cpu
, pc
, cs_base
, flags
, 0);
397 /* We add the TB in the virtual pc hash table for the fast lookup */
398 atomic_set(&cpu
->tb_jmp_cache
[tb_jmp_cache_hash_func(pc
)], tb
);
400 #ifndef CONFIG_USER_ONLY
401 /* We don't take care of direct jumps when address mapping changes in
402 * system emulation. So it's not safe to make a direct jump to a TB
403 * spanning two pages because the mapping for the second page can change.
405 if (tb
->page_addr
[1] != -1) {
409 /* See if we can patch the calling TB. */
410 if (last_tb
&& !qemu_loglevel_mask(CPU_LOG_TB_NOCHAIN
)) {
411 if (!acquired_tb_lock
) {
413 acquired_tb_lock
= true;
415 if (!(tb
->cflags
& CF_INVALID
)) {
416 tb_add_jump(last_tb
, tb_exit
, tb
);
419 if (acquired_tb_lock
) {
425 static inline bool cpu_handle_halt(CPUState
*cpu
)
428 #if defined(TARGET_I386) && !defined(CONFIG_USER_ONLY)
429 if ((cpu
->interrupt_request
& CPU_INTERRUPT_POLL
)
430 && replay_interrupt()) {
431 X86CPU
*x86_cpu
= X86_CPU(cpu
);
432 qemu_mutex_lock_iothread();
433 apic_poll_irq(x86_cpu
->apic_state
);
434 cpu_reset_interrupt(cpu
, CPU_INTERRUPT_POLL
);
435 qemu_mutex_unlock_iothread();
438 if (!cpu_has_work(cpu
)) {
448 static inline void cpu_handle_debug_exception(CPUState
*cpu
)
450 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
453 if (!cpu
->watchpoint_hit
) {
454 QTAILQ_FOREACH(wp
, &cpu
->watchpoints
, entry
) {
455 wp
->flags
&= ~BP_WATCHPOINT_HIT
;
459 cc
->debug_excp_handler(cpu
);
462 static inline bool cpu_handle_exception(CPUState
*cpu
, int *ret
)
464 if (cpu
->exception_index
>= 0) {
465 if (cpu
->exception_index
>= EXCP_INTERRUPT
) {
466 /* exit request from the cpu execution loop */
467 *ret
= cpu
->exception_index
;
468 if (*ret
== EXCP_DEBUG
) {
469 cpu_handle_debug_exception(cpu
);
471 cpu
->exception_index
= -1;
474 #if defined(CONFIG_USER_ONLY)
475 /* if user mode only, we simulate a fake exception
476 which will be handled outside the cpu execution
478 #if defined(TARGET_I386)
479 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
480 cc
->do_interrupt(cpu
);
482 *ret
= cpu
->exception_index
;
483 cpu
->exception_index
= -1;
486 if (replay_exception()) {
487 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
488 qemu_mutex_lock_iothread();
489 cc
->do_interrupt(cpu
);
490 qemu_mutex_unlock_iothread();
491 cpu
->exception_index
= -1;
492 } else if (!replay_has_interrupt()) {
493 /* give a chance to iothread in replay mode */
494 *ret
= EXCP_INTERRUPT
;
499 #ifndef CONFIG_USER_ONLY
500 } else if (replay_has_exception()
501 && cpu
->icount_decr
.u16
.low
+ cpu
->icount_extra
== 0) {
502 /* try to cause an exception pending in the log */
503 cpu_exec_nocache(cpu
, 1, tb_find(cpu
, NULL
, 0), true);
512 static inline bool cpu_handle_interrupt(CPUState
*cpu
,
513 TranslationBlock
**last_tb
)
515 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
517 if (unlikely(atomic_read(&cpu
->interrupt_request
))) {
518 int interrupt_request
;
519 qemu_mutex_lock_iothread();
520 interrupt_request
= cpu
->interrupt_request
;
521 if (unlikely(cpu
->singlestep_enabled
& SSTEP_NOIRQ
)) {
522 /* Mask out external interrupts for this step. */
523 interrupt_request
&= ~CPU_INTERRUPT_SSTEP_MASK
;
525 if (interrupt_request
& CPU_INTERRUPT_DEBUG
) {
526 cpu
->interrupt_request
&= ~CPU_INTERRUPT_DEBUG
;
527 cpu
->exception_index
= EXCP_DEBUG
;
528 qemu_mutex_unlock_iothread();
531 if (replay_mode
== REPLAY_MODE_PLAY
&& !replay_has_interrupt()) {
533 } else if (interrupt_request
& CPU_INTERRUPT_HALT
) {
535 cpu
->interrupt_request
&= ~CPU_INTERRUPT_HALT
;
537 cpu
->exception_index
= EXCP_HLT
;
538 qemu_mutex_unlock_iothread();
541 #if defined(TARGET_I386)
542 else if (interrupt_request
& CPU_INTERRUPT_INIT
) {
543 X86CPU
*x86_cpu
= X86_CPU(cpu
);
544 CPUArchState
*env
= &x86_cpu
->env
;
546 cpu_svm_check_intercept_param(env
, SVM_EXIT_INIT
, 0, 0);
547 do_cpu_init(x86_cpu
);
548 cpu
->exception_index
= EXCP_HALTED
;
549 qemu_mutex_unlock_iothread();
553 else if (interrupt_request
& CPU_INTERRUPT_RESET
) {
556 qemu_mutex_unlock_iothread();
560 /* The target hook has 3 exit conditions:
561 False when the interrupt isn't processed,
562 True when it is, and we should restart on a new TB,
563 and via longjmp via cpu_loop_exit. */
565 if (cc
->cpu_exec_interrupt(cpu
, interrupt_request
)) {
569 /* The target hook may have updated the 'cpu->interrupt_request';
570 * reload the 'interrupt_request' value */
571 interrupt_request
= cpu
->interrupt_request
;
573 if (interrupt_request
& CPU_INTERRUPT_EXITTB
) {
574 cpu
->interrupt_request
&= ~CPU_INTERRUPT_EXITTB
;
575 /* ensure that no TB jump will be modified as
576 the program flow was changed */
580 /* If we exit via cpu_loop_exit/longjmp it is reset in cpu_exec */
581 qemu_mutex_unlock_iothread();
584 /* Finally, check if we need to exit to the main loop. */
585 if (unlikely(atomic_read(&cpu
->exit_request
)
586 || (use_icount
&& cpu
->icount_decr
.u16
.low
+ cpu
->icount_extra
== 0))) {
587 atomic_set(&cpu
->exit_request
, 0);
588 cpu
->exception_index
= EXCP_INTERRUPT
;
595 static inline void cpu_loop_exec_tb(CPUState
*cpu
, TranslationBlock
*tb
,
596 TranslationBlock
**last_tb
, int *tb_exit
)
601 trace_exec_tb(tb
, tb
->pc
);
602 ret
= cpu_tb_exec(cpu
, tb
);
603 tb
= (TranslationBlock
*)(ret
& ~TB_EXIT_MASK
);
604 *tb_exit
= ret
& TB_EXIT_MASK
;
605 if (*tb_exit
!= TB_EXIT_REQUESTED
) {
611 insns_left
= atomic_read(&cpu
->icount_decr
.u32
);
612 atomic_set(&cpu
->icount_decr
.u16
.high
, 0);
613 if (insns_left
< 0) {
614 /* Something asked us to stop executing chained TBs; just
615 * continue round the main loop. Whatever requested the exit
616 * will also have set something else (eg exit_request or
617 * interrupt_request) which we will handle next time around
618 * the loop. But we need to ensure the zeroing of icount_decr
619 * comes before the next read of cpu->exit_request
620 * or cpu->interrupt_request.
626 /* Instruction counter expired. */
628 #ifndef CONFIG_USER_ONLY
629 /* Ensure global icount has gone forward */
630 cpu_update_icount(cpu
);
631 /* Refill decrementer and continue execution. */
632 insns_left
= MIN(0xffff, cpu
->icount_budget
);
633 cpu
->icount_decr
.u16
.low
= insns_left
;
634 cpu
->icount_extra
= cpu
->icount_budget
- insns_left
;
635 if (!cpu
->icount_extra
) {
636 /* Execute any remaining instructions, then let the main loop
637 * handle the next event.
639 if (insns_left
> 0) {
640 cpu_exec_nocache(cpu
, insns_left
, tb
, false);
646 /* main execution loop */
648 int cpu_exec(CPUState
*cpu
)
650 CPUClass
*cc
= CPU_GET_CLASS(cpu
);
652 SyncClocks sc
= { 0 };
654 /* replay_interrupt may need current_cpu */
657 if (cpu_handle_halt(cpu
)) {
663 cc
->cpu_exec_enter(cpu
);
665 /* Calculate difference between guest clock and host clock.
666 * This delay includes the delay of the last cycle, so
667 * what we have to do is sleep until it is 0. As for the
668 * advance/delay we gain here, we try to fix it next time.
670 init_delay_params(&sc
, cpu
);
672 /* prepare setjmp context for exception handling */
673 if (sigsetjmp(cpu
->jmp_env
, 0) != 0) {
674 #if defined(__clang__) || !QEMU_GNUC_PREREQ(4, 6)
675 /* Some compilers wrongly smash all local variables after
676 * siglongjmp. There were bug reports for gcc 4.5.0 and clang.
677 * Reload essential local variables here for those compilers.
678 * Newer versions of gcc would complain about this code (-Wclobbered). */
680 cc
= CPU_GET_CLASS(cpu
);
681 #else /* buggy compiler */
682 /* Assert that the compiler does not smash local variables. */
683 g_assert(cpu
== current_cpu
);
684 g_assert(cc
== CPU_GET_CLASS(cpu
));
685 #endif /* buggy compiler */
688 if (qemu_mutex_iothread_locked()) {
689 qemu_mutex_unlock_iothread();
693 /* if an exception is pending, we execute it here */
694 while (!cpu_handle_exception(cpu
, &ret
)) {
695 TranslationBlock
*last_tb
= NULL
;
698 while (!cpu_handle_interrupt(cpu
, &last_tb
)) {
699 TranslationBlock
*tb
= tb_find(cpu
, last_tb
, tb_exit
);
700 cpu_loop_exec_tb(cpu
, tb
, &last_tb
, &tb_exit
);
701 /* Try to align the host and virtual clocks
702 if the guest is in advance */
703 align_clocks(&sc
, cpu
);
707 cc
->cpu_exec_exit(cpu
);