从编译器源码中提取ARMv8的指令编码

问题1：提取ARMv8的指令编码

答：ARMv8指令的opcode可以在支持ARMv8的编译器binutils中找到，该工具能在网上找到，见参考资料[1]，或者在[1]中搜索aarch64_opcode_table。

为了说明指令编码细节，现将网页中部分内容摘录如下：

1208 struct aarch64_opcode aarch64_opcode_table[] =

1209 {

1210  /* Add/subtract (with carry).  */

1211  {"adc", 0x1a000000, 0x7fe0fc00, addsub_carry, 0, CORE, OP3 (Rd, Rn, Rm), QL_I3SAMER, F_SF},

1212  {"adcs", 0x3a000000, 0x7fe0fc00, addsub_carry, 0, CORE, OP3 (Rd, Rn, Rm), QL_I3SAMER, F_SF},

以上结构体中的每一项对应某条指令的操作码等信息，其中aarch64_opcode结构体（见参考文献[2]）在，内容摘录如下：

structaarch64_opcode

{

  /*The name of the mnemonic.  */

 const char *name;

  /*The opcode itself.  Those bits which willbe filled in with operands are zeroes. */

 aarch64_insn opcode;

  /*The opcode mask.  This is used by thedisassembler.  This is a mask containingones indicating those bits which must match the opcode field, and zeroesindicating those bits which need not match (and are presumably filled in byoperands).  */

 aarch64_insn mask;

  /* Instruction class.  */

 enum aarch64_insn_class iclass;

  /*Enumerator identifier.  */

 enum aarch64_op op;

  /*Which architecture variant provides this instruction.  */

 const aarch64_feature_set *avariant;

  /*An array of operand codes.  Each code isan index into the operand table.  Theyappear in the order which the operands must appear in assembly code, and are terminatedby a zero.  */

 enum aarch64_opnd operands[AARCH64_MAX_OPND_NUM];

  /*A list of operand qualifier code sequence. Each operand qualifier code qualifies the corresponding operandcode.  Each operand qualifier sequencespecifies a valid opcode variant and related constraint on operands.  */

 aarch64_opnd_qualifier_seq_t qualifiers_list[AARCH64_MAX_QLF_SEQ_NUM];

  /*Flags providing information about this instruction */

 uint32_t flags;

}

按照以上信息解释aarch64_opcode_table中的结构体如下表：

1：Aarch64-opc.c中有关于指令位域的定义。

const aarch64_field fields[] =

{

   {  0,  0 },      /*NIL.  */

   {  0,  4 },      /*cond2: condition in truly conditional-executed inst.  */

   {  0,  4 },      /*nzcv: flag bit specifier, encoded in the "nzcv" field.  */

   {  5,  5 },      /*defgh: d:e:f:g:h bits in AdvSIMD modified immediate.  */

    {16,  3 },       /*abc: a:b:c bits in AdvSIMD modified immediate. */

   {  5, 19 },       /* imm19: e.g. in CBZ. */

   {  5, 19 },       /* immhi: e.g. in ADRP. */

    {29,  2 },       /*immlo: e.g. in ADRP.  */

    {22,  2 },       /*size: in most AdvSIMD and floating-point instructions.  */

    {10,  2 },       /*vldst_size: size field in the AdvSIMD load/store inst.  */

    {29,  1 },       /*op: in AdvSIMD modified immediate instructions. */

    {30,  1 },       /*Q: in most AdvSIMD instructions.  */

   {  0,  5 },      /*Rt: in load/store instructions.  */

   {  0,  5 },      /*Rd: in many integer instructions.  */

   {  5,  5 },      /*Rn: in many integer instructions.  */

    {10,  5 },       /*Rt2: in load/store pair instructions.  */

    {10,  5 },       /*Ra: in fp instructions.  */

   {  5,  3 },      /*op2: in the system instructions.  */

   {  8,  4 },      /*CRm: in the system instructions.  */

    {12,  4 },       /*CRn: in the system instructions.  */

    {16,  3 },       /*op1: in the system instructions.  */

    {19,  2 },       /*op0: in the system instructions.  */

    {10,  3 },       /*imm3: in add/sub extended reg instructions. */

    {12,  4 },       /*cond: condition flags as a source operand. */

    {12,  4 },       /*opcode: in advsimd load/store instructions. */

    {12,  4 },       /*cmode: in advsimd modified immediate instructions.  */

    {13,  3 },       /*asisdlso_opcode: opcode in advsimd ld/st single element.  */

    {13,  2 },       /*len: in advsimd tbl/tbx instructions.  */

    {16,  5 },       /*Rm: in ld/st reg offset and some integer inst. */

    {16,  5 },       /*Rs: in load/store exclusive instructions. */

    {13,  3 },       /*option: in ld/st reg offset + add/sub extended reg inst.  */

    {12,  1 },       /*S: in load/store reg offset instructions. */

    {21,  2 },       /*hw: in move wide constant instructions. */

    {22,  2 },       /*opc: in load/store reg offset instructions. */

    {23,  1 },       /*opc1: in load/store reg offset instructions. */

    {22,  2 },       /*shift: in add/sub reg/imm shifted instructions. */

    {22,  2 },       /*type: floating point type field in fp data inst.  */

    {30,  2 },       /*ldst_size: size field in ld/st reg offset inst. */

    {10,  6 },       /*imm6: in add/sub reg shifted instructions. */

    {11,  4 },       /*imm4: in advsimd ext and advsimd ins instructions.  */

    {16,  5 },       /*imm5: in conditional compare (immediate) instructions.  */

    {15,  7 },       /*imm7: in load/store pair pre/post index instructions.  */

    {13,  8 },       /*imm8: in floating-point scalar move immediate inst.  */

    {12,  9 },       /*imm9: in load/store pre/post index instructions.  */

    {10, 12 },         /* imm12: in ld/stunsigned imm or add/sub shifted inst.  */

   {  5, 14 },       /* imm14: in test bit and branch instructions.  */

   {  5, 16 },       /* imm16: in exception instructions.  */

   {  0, 26 },       /* imm26: in unconditional branch instructions.  */

    { 10, 6 },       /* imms: in bitfield andlogical immediate instructions.  */

    {16,  6 },       /*immr: in bitfield and logical immediate instructions.  */

    {16,  3 },       /*immb: in advsimd shift by immediate instructions.  */

    {19,  4 },       /*immh: in advsimd shift by immediate instructions.  */

    {22,  1 },       /*N: in logical (immediate) instructions. */

    {11,  1 },       /*index: in ld/st inst deciding the pre/post-index.  */

    {24,  1 },       /*index2: in ld/st pair inst deciding the pre/post-index.  */

    {31,  1 },       /*sf: in integer data processing instructions. */

    {11,  1 },       /*H: in advsimd scalar x indexed element instructions.  */

    {21,  1 },       /*L: in advsimd scalar x indexed element instructions.  */

    {20,  1 },       /*M: in advsimd scalar x indexed element instructions.  */

    {31,  1 },       /*b5: in the test bit and branch instructions. */

    {19,  5 },       /*b40: in the test bit and branch instructions. */

    {10,  6 },       /*scale: in the fixed-point scalar to fp converting inst.  */

};

2、函数aarch64_opcode_encode用于将操作数插入opcode中

参考资料

[1] ARMv8的编译器binutils，结构体aarch64_opcode_table定义见line 1208

http://kernel.ubuntu.com/git?p=jk/arm64/binutils.git;a=blob;f=opcodes/aarch64-tbl.h;h=d360b1406718257da86050f5b3a760cd02196250;hb=aarch64

[2] 结构体aarch64_opcode定义，line451

http://kernel.ubuntu.com/git?p=jk/arm64/binutils.git;a=blob;f=include/opcode/aarch64.h;h=98529954ea098349eb16572d4915f4edbd2e7b5d;hb=aarch64

[3] 《ARMv8 InstructionSet Overview 》page11