Chinaunix首页 | 论坛 | 博客
  • 博客访问: 99976
  • 博文数量: 28
  • 博客积分: 1435
  • 博客等级: 上尉
  • 技术积分: 265
  • 用 户 组: 普通用户
  • 注册时间: 2010-04-26 11:40
文章分类

全部博文(28)

文章存档

2017年(1)

2012年(1)

2011年(6)

2010年(20)

我的朋友

分类: LINUX

2010-09-08 15:44:49

Main loop qemu

cpu_exec.c

cpu_exec()
 {
     for(;;) {
         if (setjmp(env->jmp_env) == 0) {
             exception handling(2) if requested through the loop break;
             for (;;) {
                 exception handling(1) ARM FIQ case
 
                 //PCto find the code corresponding to the converted. Without conversion
                 tb = tb_find_fast(); 
 
                 //Execute the code converted
                 next_tb = tcg_qemu_tb_exec(tb->tc_ptr);
 
             }
         } else {
             Longjmpto the top of the loop through here then
         }
     }
     
 }

JIT code conversion part is easy to think and understand the instruction cache. If it’s running a cache hit, if you don’t miss cache fill.

cpu_exec call stack

 main   (vl.c)
   main_loop   (vl.c)
     tcg_cpu_exec   (vl.c)
       qemu_cpu_exec   (vl.c)
         cpu_exec      (cpu_exec.c)

tcg_cpu_exec to simulate a loop in the multi-core. Separate the execution time on each core as TSS. Multithreading is not.

tb_find_fast inside

exec.c

 tb_find_fast   (cpu-exec.c)
   tb_find_slow    (cpu-exec.c)
     not_found:
       tb = tb_gen_code( ... );    (exec.c)

Transformed tb (TranslationBlock) looking for a hash table to find first, then find a physical address from the map, if not find, tb_gen_code will be called to convert the code.

QEMU TCG

Tb_gen_code function inside

 tb_gen_code  (exec.c) tb_gen_code (exec.c)
   cpu_gen_code  (translate-all.c) cpu_gen_code (translate-all.c)
     gen_intermediate_code  (target-arm/translate.c)
 tcg_gen_code (tcg / tcg.c)

 

The intermediate code to generate the target code instruction gen_intermediate_code() is to generate code from intermediate code instruction of the host tcg_gen_code(). This two –step generation has always continue to do.

Intermediate code is always generated in the same fixed array (gen_opc_buf[]), so it will be overwritten the next time. Not saved.

 

Host instruction codes are stored in code_gen_buffer[]. The default buffer size is 32MB.(Ram_size/4, while ram_size default value is 128MB).

 Code_gen_buffer [] can’t detect the overflow issue.

 

These sizes can be changed by qemu command line options (-M, -tb-size).

 

Gen_intermediate_code function inside

Target-arm/translate.c

 

   gen_intermediate_code

        gen_intermediate_code_internal

          disas_asm_insn

          disas_thumb_insn

 

TCG Structure of the intermediate code

 

l         Operation code

 

uint_16 gen_opc_buf[OPC_BUF_SIZE];   //OPC_BUF_SIZE = 512 uint_16 gen_opc_buf [OPC_BUF_SIZE]; / / OPC_BUF_SIZE = 512
 uint_16 *gen_opc_ptr; uint_16 * gen_opc_ptr;

 

l         Operand

TCGARG gen_opparam_buff[OPPARAM_BUF_SIZE]; //OPC_OPPAERAM_BUF_SIZE = 512 * 10 TCGARG gen_opparam_buff [OPPARAM_BUF_SIZE]; / / OPC_OPPAERAM_BUF_SIZE = 512 * 10
 TCGARG *gen_opparam_ptr; TCGARG * gen_opparam_ptr;
阅读(1031) | 评论(0) | 转发(0) |
给主人留下些什么吧!~~