leal (%eax,%eax), %edx invalid operand. failure parsing operand (%eax,%eax), bad supporting to index addressing

[bbqz007]

leal (%eax,%eax), %edx means edx = eax + eax,

[bbqz007]

operation displacement(base, index, scale) not ok

operation displacement(base, scale) ok

[bbqz007]

X86AsmParser.cpp ::X86AsmParser::ParseMemOperand

  if (getLexer().is(AsmToken::Percent)) {
    SMLoc StartLoc, EndLoc;
    BaseLoc = Parser.getTok().getLoc();
    if (ParseRegister(BaseReg, StartLoc, EndLoc, ErrorCode)) {
        KsError = KS_ERR_ASM_INVALIDOPERAND;
        return nullptr;

X86AsmParser.cpp ::X86AsmParser::ParseRegister

  Parser.Lex(); // Eat identifier token.
  return false;
}

X86AsmParser.cpp ::X86AsmParser::ParseMemOperand

  if (getLexer().is(AsmToken::Comma)) {
    Parser.Lex(); // Eat the comma.
    IndexLoc = Parser.getTok().getLoc();

• the correct logic
  • after BaseReg, enter try to Parse Loc,
  • if not Loc, and is Comma, may be IndexReg
  • after IndexReg, try to Parse Loc
• problem
  • when parse Loc after BaseReg, the failure code path eat the comma then return
  • so skip the IndexReg branch.

[bbqz007]

    // Following the comma we should have either an index register, or a scale
    // value. We don't support the later form, but we want to parse it
    // correctly.
    //
    // Not that even though it would be completely consistent to support syntax
    // like "1(%eax,,1)", the assembler doesn't. Use "eiz" or "riz" for this.
    if (getLexer().is(AsmToken::Percent)) {
      SMLoc L;
      if (ParseRegister(IndexReg, L, L, ErrorCode)) {
          KsError = KS_ERR_ASM_INVALIDOPERAND;
          return nullptr;
      }

      if (ParseRegister(IndexReg, L, L, ErrorCode)) {
        KsError = KS_ERR_ASM_X86_INVALIDOPERAND;
        return nullptr;
      }

there are twice parsing IndexReg.

maybe the origin version is

    if (getLexer().is(AsmToken::Percent)) {
      SMLoc L;
      if (ParseRegister(BaseReg, L, L, ErrorCode)) {
          KsError = KS_ERR_ASM_INVALIDOPERAND;
          return nullptr;
      }

      if (ParseRegister(IndexReg, L, L, ErrorCode)) {
        KsError = KS_ERR_ASM_X86_INVALIDOPERAND;
        return nullptr;
      }

to solve like "1(%eax,,1)",
they move the BaseReg parsing code out of this if condition code path, and leave the un-debug codes where it was.

[bbqz007]

Yes, it is.

the problem is solved.