一生一芯学习：pa2.1 RTFM

前面的内容后面再补，因为近期要C考核了，所以先准备下C考核所需的内容。

RTFSC(2)
整理一条指令在NEMU中的执行结果

从函数中跳转，宏嵌套中慢慢进入最终的代码，人肉gdb一下，
比如在sdb中执行一次si，那么根据cmd_si代码那我们会跳转到cpu_exec()中

static int cmd_si(char *args) {if (args == NULL) {cpu_exec(1);return 0;}int i = atoi(args);if (i <= 0) {printf("Invalid argument '%s'\n", args);} else {cpu_exec(i);}return 0;
}

在cpu_exec()中我们进入execute()函数中。

/* Simulate how the CPU works. */
void cpu_exec(uint64_t n) {g_print_step = (n < MAX_INST_TO_PRINT);//一次执行太多步就不打印了，bool类型的gprintstep就赋值为false//printf("%d\n",nemu_state.state);switch (nemu_state.state) {case NEMU_END: case NEMU_ABORT: case NEMU_QUIT:printf("Program execution has ended. To restart the program, exit NEMU and run again.\n");return;//如果状态是结束了，出错了，退出了就打印“退出nemu”。default: nemu_state.state = NEMU_RUNNING;//默认running}uint64_t timer_start = get_time();//获取执行指令前的时间execute(n);uint64_t timer_end = get_time();//获取执行指令后的时间g_timer += timer_end - timer_start; //看执行了多久。switch (nemu_state.state) { case NEMU_RUNNING: nemu_state.state = NEMU_STOP; break;case NEMU_END: case NEMU_ABORT:Log("nemu: %s at pc = " FMT_WORD,//nemu出错或者异常退出就用红色打印，正常退出就绿色打印。  (nemu_state.state == NEMU_ABORT ? ANSI_FMT("ABORT", ANSI_FG_RED) :(nemu_state.halt_ret == 0 ? ANSI_FMT("HIT GOOD TRAP", ANSI_FG_GREEN) :ANSI_FMT("HIT BAD TRAP", ANSI_FG_RED))),nemu_state.halt_pc);// fall throughcase NEMU_QUIT: statistic();}
}

在execute()中跳转到exec_once()中

static void execute(uint64_t n) {Decode s;initBuffer(&cb); // 初始化环形缓冲区，大小为BUFFER_SIZEfor (;n > 0; n --) {exec_once(&s, cpu.pc);g_nr_guest_inst ++;trace_and_difftest(&s, cpu.pc);if (nemu_state.state != NEMU_RUNNING) {break;}IFDEF(CONFIG_DEVICE, device_update());}/*条件编译宏，如果CONFIG_DEVICE被定义，则调用device_update函数，如果 CONFIG_DEVICE 没有被定义，这一行什么都不会生成（等价于被注释掉）。*/printBuffer(&cb);
}

从exec_once()中跳转到isa_exec_once()中。

static void exec_once(Decode *s, vaddr_t pc) {s->pc = pc;//当前指令地址s->snpc = pc;//静态下一条指令地址，默认为pc+4// if(s->pc == 0x80001480){//   printf("找到一场入口地址\n");//   nemu_state.state = NEMU_STOP;// }isa_exec_once(s);cpu.pc = s->dnpc;//动态下一条指令，可能跳转或者分支改变
#ifdef CONFIG_ITRACE//如果启用了 CONFIG_ITRACE，会记录指令的详细信息到日志缓冲区 s->logbuf：char *p = s->logbuf;p += snprintf(p, sizeof(s->logbuf), FMT_WORD ":", s->pc);//FMT_WORD：格式化字符串（如 "0x%08x"），用于输出 PC 地址。//printf("0x%08x\n",s->pc);//printf("0x%08x\n",s->snpc);int ilen = s->snpc - s->pc; //计算指令长度int i;uint8_t *inst = (uint8_t *)&s->isa.inst;// printf("inst = 0x%08x\n",s->isa.inst);// printf("inst ***= 0x%08x\n", *inst);
#ifdef CONFIG_ISA_x86for (i = 0; i < ilen; i ++) { //x86是小段，从低地址开始打印
#elsefor (i = ilen - 1; i >= 0; i --) {//riscv是大段，从高地址开始打印
#endifp += snprintf(p, 4, " %02x", inst[i]); //把指令打印出来}int ilen_max = MUXDEF(CONFIG_ISA_x86, 8, 4); //不是x86ilenmax就是4int space_len = ilen_max - ilen;   //计算需要填充的空格数if (space_len < 0) space_len = 0; //space_len = space_len * 3 + 1;memset(p, ' ', space_len);p += space_len;void disassemble(char *str, int size, uint64_t pc, uint8_t *code, int nbyte);//反汇编指令disassemble(p, s->logbuf + sizeof(s->logbuf) - p,   //将反汇编指令出来后传到logbuf里面MUXDEF(CONFIG_ISA_x86, s->snpc, s->pc), (uint8_t *)&s->isa.inst, ilen);//muxdef，有点像  ？：，enqueue(&cb, s->logbuf);#endif
}

从isa_exec_once中在跳转到decode_exec()函数中

int isa_exec_once(Decode *s) {s->isa.inst = inst_fetch(&s->snpc, 4);//return一个0回去，但是现在并没有使用这个返回值，可以忽略他。return decode_exec(s);
}

inst_fetch()是在snpc中获取一条长度为4字节的指令赋值给s->inst，然后执行一遍decode_exec()并返回decode_exec()的返回值。

先一次性将decode_exec()代码全部贴出，很长

static int decode_exec(Decode *s) {int rd = 0;word_t src1 = 0, src2 = 0, imm = 0;s->dnpc = s->snpc;#define INSTPAT_INST(s) ((s)->isa.inst)
#define INSTPAT_MATCH(s, name, type, ... /* execute body */ ) { \decode_operand(s, &rd, &src1, &src2, &imm, concat(TYPE_, type)); \__VA_ARGS__ ; \
}INSTPAT_START();//INSTPAT(模式字符串, 指令名称, 指令类型, 指令执行操作);INSTPAT("??????? ????? ????? ??? ????? 00101 11", auipc  , U, R(rd) = s->pc + imm); INSTPAT("??????? ????? ????? ??? ????? 01101 11", lui    , U, R(rd) = imm);     INSTPAT("0000000 ????? ????? 101 ????? 00100 11", srli   , I, R(rd) = src1 >> BITS(imm, 5, 0)); INSTPAT("0000000 ????? ????? 001 ????? 00100 11", slli   , I, R(rd) = src1 << BITS(imm, 5, 0));INSTPAT("0100000 ????? ????? 101 ????? 00100 11", srai   , I, R(rd) = (int32_t)src1 >> BITS(imm , 4 , 0) ); INSTPAT("??????? ????? ????? 100 ????? 00000 11", lbu    , I, R(rd) = Mr(src1 + imm, 1));INSTPAT("??????? ????? ????? 000 ????? 00100 11", addi   , I, R(rd) = src1 + imm);INSTPAT("??????? ????? ????? 011 ????? 00100 11", sltiu  , I, R(rd) = (src1 < imm) ? 1 : 0); INSTPAT("??????? ????? ????? 010 ????? 00100 11", slti   , I, R(rd) = ((int32_t)src1 < ((int32_t)imm)) ? 1 : 0); INSTPAT("??????? ????? ????? 000 ????? 00000 11", lb     , I, R(rd) = SEXT(Mr(src1 + imm, 2),8));INSTPAT("??????? ????? ????? 001 ????? 00000 11", lh     , I, R(rd) = SEXT(Mr(src1 + imm, 2),16));INSTPAT("??????? ????? ????? 101 ????? 00000 11", lhu    , I, R(rd) = Mr(src1 + imm, 2));INSTPAT("??????? ????? ????? 010 ????? 00000 11", lw     , I, R(rd) = Mr(src1 + imm, 4)); INSTPAT("??????? ????? ????? 111 ????? 00100 11", andi   , I, R(rd) = src1 & imm); INSTPAT("??????? ????? ????? 100 ????? 00100 11", xori   , I, R(rd) = src1 ^ imm); INSTPAT("??????? ????? ????? 110 ????? 00100 11", ori    , I, R(rd) = src1 | imm);//CSR寄存器INSTPAT("??????? ????? ????? 001 ????? 11100 11", csrrw  , I,  if(imm == 0x305){  //mtvecR(rd) = cpu.mtvec;cpu.mtvec =  src1;};if(imm == 0x300){ //mstatusR(rd) = cpu.mstatus;cpu.mstatus =  src1;};if(imm == 0x341){ //mepcR(rd) = cpu.mepc;cpu.mepc =  src1;};if(imm == 0x342){ //mcauseR(rd) = cpu.mcause;cpu.mcause =  src1;};
);INSTPAT("0000000 00000 00000 000 00000 11100 11", ecall  , I, s->dnpc = isa_raise_intr(11,s->pc);etrace());INSTPAT("??????? ????? ????? 010 ????? 11100 11", csrrs  , I, if(imm == 0x305){  //mtvecR(rd) = cpu.mtvec;cpu.mtvec |=  src1;};if(imm == 0x300){ //mstatusR(rd) = cpu.mstatus;cpu.mstatus |=  src1;};if(imm == 0x341){ //mepcR(rd) = cpu.mepc;cpu.mepc |=  src1;};if(imm == 0x342){ //mcause// printf("??????????????????????????*****\n");R(rd) = cpu.mcause;cpu.mcause |=  src1;};
);INSTPAT("??????? ????? ????? 010 ????? 01000 11", sw     , S, Mw(src1 + imm, 4, src2));INSTPAT("??????? ????? ????? 001 ????? 01000 11", sh     , S, Mw(src1 + imm, 2, src2)); INSTPAT("??????? ????? ????? 000 ????? 01000 11", sb     , S, Mw(src1 + imm, 1, src2)); INSTPAT("??????? ????? ????? 011 ????? 01000 11", sd     , S, Mw(src1 + imm, 8, src2)); INSTPAT("??????? ????? ????? ??? ????? 11011 11", jal    , J, R(rd) = s->pc + 4;s->dnpc = s->pc + imm;IFDEF(CONFIG_FTRACE, {if (rd == 1) {call_trace(s->pc, s->dnpc);}}));INSTPAT("??????? ????? ????? 000 ????? 11001 11", jalr   , I, R(rd) = s->pc + 4;s->dnpc = (src1 + imm) & (~1);IFDEF(CONFIG_FTRACE,{if (s->isa.inst == 0x00008067)ret_trace(s->pc);else if (rd == 1) {call_trace(s->pc, s->dnpc);} else if (rd == 0 && imm == 0) {call_trace(s->pc, s->dnpc);}}));INSTPAT("0000000 ????? ????? 101 ????? 01100 11", srl    , R, R(rd) = src1 >> BITS(src2, 4, 0));INSTPAT("0000000 ????? ????? 000 ????? 01100 11", add    , R, R(rd) = src1 + src2); INSTPAT("0000000 ????? ????? 001 ????? 01100 11", sll    , R, R(rd) = src1 <<  BITS(src2 , 4 , 0)); INSTPAT("0000000 ????? ????? 010 ????? 01100 11", slt    , R, R(rd) = ((int32_t)src1 < (int32_t)src2) ? 1 : 0);INSTPAT("0000000 ????? ????? 011 ????? 01100 11", sltu   , R, R(rd) = src1 < src2 ? 1 : 0); INSTPAT("0000000 ????? ????? 100 ????? 01100 11", xor    , R, R(rd) = src1 ^ src2); INSTPAT("0000000 ????? ????? 110 ????? 01100 11", or     , R, R(rd) = src1 | src2); INSTPAT("0000000 ????? ????? 111 ????? 01100 11", and    , R, R(rd) = src1 & src2); INSTPAT("0100000 ????? ????? 101 ????? 01100 11", sra    , R, R(rd) = (int32_t)src1 >> BITS(src2 , 4 , 0)); INSTPAT("0100000 ????? ????? 000 ????? 01100 11", sub    , R, R(rd) = src1 - src2); INSTPAT("0000001 ????? ????? 000 ????? 01100 11", mul    , R, R(rd) = (unsigned)src1 * (unsigned)src2);//INSTPAT("0000001 ????? ????? 100 ????? 01100 11", div    , R, R(rd) = src1 / src2);//INSTPAT("0000001 ????? ????? 110 ????? 01100 11", rem    , R, R(rd) = src1 % src2);//INSTPAT("0000001 ????? ????? 111 ????? 01100 11", remu   , R, R(rd) = (unsigned)src1 % (unsigned)src2);//INSTPAT("0000001 ????? ????? 101 ????? 01100 11", divu   , R, R(rd) = (unsigned)src1 / (unsigned)src2);INSTPAT("0000001 ????? ????? 001 ????? 01100 11", mulh   , R, R(rd) = ((int64_t)(int32_t)src1 * (int64_t)(int32_t)src2) >> 32;);INSTPAT("0000001 ????? ????? 010 ????? 01100 11", mulhsu , R, R(rd) = ((int64_t)(int32_t)src1 * (int64_t)(uint32_t)src2) >> 32;);INSTPAT("0000001 ????? ????? 011 ????? 01100 11", mulhu  , R, R(rd) = ((int64_t)(uint32_t)src1 * (int64_t)(uint32_t)src2) >> 32;);INSTPAT("0000001 ????? ????? 100 ????? 01100 11", div    , R, if (src2 == 0) R(rd) = -1;else if ((int32_t)src1 == INT32_MIN && (int32_t)src2 == -1) R(rd) = INT32_MIN;else R(rd) = (int32_t)src1 / (int32_t)src2;);INSTPAT("0000001 ????? ????? 101 ????? 01100 11", divu   , R, if (src2 == 0) R(rd) = 0xFFFFFFFF;else R(rd) = (uint32_t)src1 / (uint32_t)src2;);INSTPAT("0000001 ????? ????? 110 ????? 01100 11", rem    , R, if (src2 == 0) R(rd) = (int32_t)src1;else if ((int32_t)src1 == INT32_MIN && (int32_t)src2 == -1) R(rd) = 0;else R(rd) = (int32_t)src1 % (int32_t)src2;);INSTPAT("0000001 ????? ????? 111 ????? 01100 11", remu   , R, if (src2 == 0) R(rd) = (uint32_t)src1;else R(rd) = (uint32_t)src1 % (uint32_t)src2;);INSTPAT("0011000 00010 00000 000 0000 011100 11", mret   , R, s->dnpc = cpu.mepc);//div注释：//匹配 div 指令（有符号除法）。//如果除数 src2 为 0，结果规定为 -1。//如果被除数是最小负数（INT32_MIN），除数为 -1，结果规定为 INT32_MIN（防止溢出）。//否则正常做有符号除法。//printf("mulh:%lx\n", (int64_t)(~src1+1) * (int64_t)src2));//正确的a5:0001 1001 1101 0010 1001 1010 1011 1001//INSTPAT("0000001 ????? ????? 001 ????? 01100 11", mulh   , R, R(rd) = SEXT(src1 * src2, 32));//把寄存器 x[rs2]乘到寄存器 x[rs1]上，都视为 2 的补码，将乘积的高位写入 x[rd]。INSTPAT("??????? ????? ????? 000 ????? 11000 11", beq    , B, // if(s->pc == 0x800115c0){// printf("src1 =%d\n",src1);// printf("src2 =%d\n",src2);// printf("pc =0x%08x\n",s->pc);// printf("imm =0x%08x\n",imm);// printf("dnpc =0x%08x\n",s->dnpc);  // }// printf("src1 =%d\n",src1);printf("src2 =%d\n",src2);// printf("pc =0x%08x\n",s->pc);printf("imm =0x%08x\n",imm);// printf("dnpc =0x%08x\n",s->dnpc);if(src1 == src2) s->dnpc = s->pc + imm);INSTPAT("??????? ????? ????? 001 ????? 11000 11", bne    , B, if(src1 != src2) s->dnpc = s->pc + imm);INSTPAT("??????? ????? ????? 100 ????? 11000 11", blt    , B, s->dnpc = ((int32_t)src1< (int32_t)src2) ? s->pc + imm : s->dnpc); INSTPAT("??????? ????? ????? 101 ????? 11000 11", bge    , B, s->dnpc = ((int32_t)src1>=(int32_t)src2) ? s->pc + imm : s->dnpc); INSTPAT("??????? ????? ????? 110 ????? 11000 11", bltu   , B, s->dnpc = (src1< src2) ? s->pc + imm : s->dnpc); INSTPAT("??????? ????? ????? 111 ????? 11000 11", bgeu   , B, s->dnpc = (src1>=src2) ? s->pc + imm : s->dnpc); INSTPAT("0000000 00001 00000 000 00000 11100 11", ebreak , N, NEMUTRAP(s->pc, R(10))); // R(10) is $a0INSTPAT("??????? ????? ????? ??? ????? ????? ??", inv    , N, INV(s->pc));INSTPAT_END();R(0) = 0; // reset $zero to 0return 0;
}

其中INSTPAT()意思是
首先我们来解析一下里面的宏函数

1. 先来看一下这个INSTPAT_MATCH()定义了一条模式匹配规则

INSTPAT(模式字符串, 指令名称, 指令类型, 指令执行操作);如果匹配上了，那就直接用C语言的goto跳转到INSTPAT_END表示为

#define INSTPAT_MATCH(s, name, type, ... /* execute body */ ) { \decode_operand(s, &rd, &src1, &src2, &imm, concat(TYPE_, type)); \__VA_ARGS__ ; \
}

其中的s是译码所需的信息比如pc,snpc,dnpc,isa,logbuf(用于ITRACE)
其中的decode_operand

static void decode_operand(Decode *s, int *rd, word_t *src1, word_t *src2, word_t *imm, int type) {uint32_t i = s->isa.inst;int rs1 = BITS(i, 19, 15);int rs2 = BITS(i, 24, 20);*rd     = BITS(i, 11, 7);switch (type) {case TYPE_I: src1R();          immI(); break;case TYPE_U:                   immU(); break;case TYPE_S: src1R(); src2R(); immS(); break;case TYPE_J:                   immJ(); break;case TYPE_B: src1R(); src2R(); immB(); break;case TYPE_R: src1R(); src2R();         break;}
}

理解为译码的预处理，BITS()的用法有点像verilog中的变量[1：0]的味道，也就是取这个区间的位宽，在看riscv手册中提到，指令中的源寄存器，目的寄存器，立即数基本上都有对应的位置。如果检测到一条指令是什么类型的便把需要的值赋给rs1,rs2,*rd,imm即可。此外SEXT()用于符号扩展。在匹配过程最后有一条inv的规则，如果前面所有的匹配模式都没有成功那就把此指令视为非法指令

此时取指、译码功能便实现了，接下来就是执行，访存、回写。
INSTPAT("??????? ????? ????? ??? ????? 00101 11", auipc , U, R(rd) = s->pc + imm);
都在后面的R(rd) = s->pc + imm实现了。

这个操作懂了之后，后面运行更多的程序你就知道怎么做了。

首先程序会被riscv工具链交叉编译之后生成机器码之后将一串数据放到img中，你的nemu就根据img的值不断取指、译码、执行、访存、回写等等。那你此时只需要去看am的cpu-test被汇编成什么指令，比如展示一下dummy.c汇编出来的东西放在了dummy-riscv32-nemu.txt，你就只需看你少了什么指令然后补充上去就可以了。

下一期介绍一下cpu-test一堆例程是如何跑在nemu上运行的，顺便复习一下对AM的理解。