diff options
Diffstat (limited to 'lib/Target/ARM/Disassembler')
| -rw-r--r-- | lib/Target/ARM/Disassembler/ARMDisassembler.cpp | 61 | ||||
| -rw-r--r-- | lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp | 825 | ||||
| -rw-r--r-- | lib/Target/ARM/Disassembler/ARMDisassemblerCore.h | 76 | ||||
| -rw-r--r-- | lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h | 398 |
4 files changed, 1130 insertions, 230 deletions
diff --git a/lib/Target/ARM/Disassembler/ARMDisassembler.cpp b/lib/Target/ARM/Disassembler/ARMDisassembler.cpp index 78d73d3..bdce2c4 100644 --- a/lib/Target/ARM/Disassembler/ARMDisassembler.cpp +++ b/lib/Target/ARM/Disassembler/ARMDisassembler.cpp @@ -18,6 +18,7 @@ #include "ARMDisassembler.h" #include "ARMDisassemblerCore.h" +#include "llvm/ADT/OwningPtr.h" #include "llvm/MC/EDInstInfo.h" #include "llvm/MC/MCInst.h" #include "llvm/Target/TargetRegistry.h" @@ -94,6 +95,9 @@ static unsigned decodeARMInstruction(uint32_t &insn) { // As a result, the decoder fails to deocode USAT properly. if (slice(insn, 27, 21) == 0x37 && slice(insn, 5, 4) == 1) return ARM::USAT; + // As a result, the decoder fails to deocode UQADD16 properly. + if (slice(insn, 27, 20) == 0x66 && slice(insn, 7, 4) == 1) + return ARM::UQADD16; // Ditto for ADDSrs, which is a super-instruction for A8.6.7 & A8.6.8. // As a result, the decoder fails to decode UMULL properly. @@ -280,6 +284,24 @@ static unsigned T2Morph2LoadLiteral(unsigned Opcode) { } } +// Helper function for special case handling of PLD (literal) and friends. +// See A8.6.117 T1 & T2 and friends for why we morphed the opcode +// before returning it. +static unsigned T2Morph2PLDLiteral(unsigned Opcode) { + switch (Opcode) { + default: + return Opcode; // Return unmorphed opcode. + + case ARM::t2PLDi8: case ARM::t2PLDs: + case ARM::t2PLDWi12: case ARM::t2PLDWi8: + case ARM::t2PLDWs: + return ARM::t2PLDi12; + + case ARM::t2PLIi8: case ARM::t2PLIs: + return ARM::t2PLIi12; + } +} + /// decodeThumbSideEffect is a decorator function which can potentially twiddle /// the instruction or morph the returned opcode under Thumb2. /// @@ -330,12 +352,27 @@ static unsigned decodeThumbSideEffect(bool IsThumb2, unsigned &insn) { } // --------- Transform End Marker --------- + unsigned unmorphed = decodeThumbInstruction(insn); + // See, for example, A6.3.7 Load word: Table A6-18 Load word. // See A8.6.57 T3, T4 & A8.6.60 T2 and friends for why we morphed the opcode // before returning it to our caller. if (op1 == 3 && slice(op2, 6, 5) == 0 && slice(op2, 0, 0) == 1 - && slice(insn, 19, 16) == 15) - return T2Morph2LoadLiteral(decodeThumbInstruction(insn)); + && slice(insn, 19, 16) == 15) { + unsigned morphed = T2Morph2LoadLiteral(unmorphed); + if (morphed != unmorphed) + return morphed; + } + + // See, for example, A8.6.117 PLD,PLDW (immediate) T1 & T2, and friends for + // why we morphed the opcode before returning it to our caller. + if (slice(insn, 31, 25) == 0x7C && slice(insn, 15, 12) == 0xF + && slice(insn, 22, 22) == 0 && slice(insn, 20, 20) == 1 + && slice(insn, 19, 16) == 15) { + unsigned morphed = T2Morph2PLDLiteral(unmorphed); + if (morphed != unmorphed) + return morphed; + } // One last check for NEON/VFP instructions. if ((op1 == 1 || op1 == 3) && slice(op2, 6, 6) == 1) @@ -375,21 +412,23 @@ bool ARMDisassembler::getInstruction(MCInst &MI, Size = 4; DEBUG({ - errs() << "Opcode=" << Opcode << " Name=" << ARMUtils::OpcodeName(Opcode) + errs() << "\nOpcode=" << Opcode << " Name=" <<ARMUtils::OpcodeName(Opcode) << " Format=" << stringForARMFormat(Format) << '(' << (int)Format << ")\n"; showBitVector(errs(), insn); }); - ARMBasicMCBuilder *Builder = CreateMCBuilder(Opcode, Format); + OwningPtr<ARMBasicMCBuilder> Builder(CreateMCBuilder(Opcode, Format)); if (!Builder) return false; + Builder->setupBuilderForSymbolicDisassembly(getLLVMOpInfoCallback(), + getDisInfoBlock(), getMCContext(), + Address); + if (!Builder->Build(MI, insn)) return false; - delete Builder; - return true; } @@ -398,7 +437,7 @@ bool ThumbDisassembler::getInstruction(MCInst &MI, const MemoryObject &Region, uint64_t Address, raw_ostream &os) const { - // The Thumb instruction stream is a sequence of halhwords. + // The Thumb instruction stream is a sequence of halfwords. // This represents the first halfword as well as the machine instruction // passed to decodeThumbInstruction(). For 16-bit Thumb instruction, the top @@ -463,17 +502,19 @@ bool ThumbDisassembler::getInstruction(MCInst &MI, showBitVector(errs(), insn); }); - ARMBasicMCBuilder *Builder = CreateMCBuilder(Opcode, Format); + OwningPtr<ARMBasicMCBuilder> Builder(CreateMCBuilder(Opcode, Format)); if (!Builder) return false; Builder->SetSession(const_cast<Session *>(&SO)); + Builder->setupBuilderForSymbolicDisassembly(getLLVMOpInfoCallback(), + getDisInfoBlock(), getMCContext(), + Address); + if (!Builder->Build(MI, insn)) return false; - delete Builder; - return true; } diff --git a/lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp b/lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp index bac68dd..642829c 100644 --- a/lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp +++ b/lib/Target/ARM/Disassembler/ARMDisassemblerCore.cpp @@ -17,6 +17,7 @@ #include "ARMDisassemblerCore.h" #include "ARMAddressingModes.h" +#include "ARMMCExpr.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" @@ -82,10 +83,28 @@ const char *ARMUtils::OpcodeName(unsigned Opcode) { // FIXME: Auto-gened? static unsigned getRegisterEnum(BO B, unsigned RegClassID, unsigned RawRegister) { - // For this purpose, we can treat rGPR as if it were GPR. - if (RegClassID == ARM::rGPRRegClassID) RegClassID = ARM::GPRRegClassID; + if (RegClassID == ARM::rGPRRegClassID) { + // Check for The register numbers 13 and 15 that are not permitted for many + // Thumb register specifiers. + if (RawRegister == 13 || RawRegister == 15) { + B->SetErr(-1); + return 0; + } + // For this purpose, we can treat rGPR as if it were GPR. + RegClassID = ARM::GPRRegClassID; + } // See also decodeNEONRd(), decodeNEONRn(), decodeNEONRm(). + // A7.3 register encoding + // Qd -> bit[12] == 0 + // Qn -> bit[16] == 0 + // Qm -> bit[0] == 0 + // + // If one of these bits is 1, the instruction is UNDEFINED. + if (RegClassID == ARM::QPRRegClassID && slice(RawRegister, 0, 0) == 1) { + B->SetErr(-1); + return 0; + } unsigned RegNum = RegClassID == ARM::QPRRegClassID ? RawRegister >> 1 : RawRegister; @@ -497,14 +516,66 @@ static bool DisassemblePseudo(MCInst &MI, unsigned Opcode, uint32_t insn, return false; } +// A8.6.94 MLA +// if d == 15 || n == 15 || m == 15 || a == 15 then UNPREDICTABLE; +// +// A8.6.105 MUL +// if d == 15 || n == 15 || m == 15 then UNPREDICTABLE; +// +// A8.6.246 UMULL +// if dLo == 15 || dHi == 15 || n == 15 || m == 15 then UNPREDICTABLE; +// if dHi == dLo then UNPREDICTABLE; +static bool BadRegsMulFrm(unsigned Opcode, uint32_t insn) { + unsigned R19_16 = slice(insn, 19, 16); + unsigned R15_12 = slice(insn, 15, 12); + unsigned R11_8 = slice(insn, 11, 8); + unsigned R3_0 = slice(insn, 3, 0); + switch (Opcode) { + default: + // Did we miss an opcode? + DEBUG(errs() << "BadRegsMulFrm: unexpected opcode!"); + return false; + case ARM::MLA: case ARM::MLS: case ARM::SMLABB: case ARM::SMLABT: + case ARM::SMLATB: case ARM::SMLATT: case ARM::SMLAWB: case ARM::SMLAWT: + case ARM::SMMLA: case ARM::SMMLAR: case ARM::SMMLS: case ARM::SMMLSR: + case ARM::USADA8: + if (R19_16 == 15 || R15_12 == 15 || R11_8 == 15 || R3_0 == 15) + return true; + return false; + case ARM::MUL: case ARM::SMMUL: case ARM::SMMULR: + case ARM::SMULBB: case ARM::SMULBT: case ARM::SMULTB: case ARM::SMULTT: + case ARM::SMULWB: case ARM::SMULWT: case ARM::SMUAD: case ARM::SMUADX: + // A8.6.167 SMLAD & A8.6.172 SMLSD + case ARM::SMLAD: case ARM::SMLADX: case ARM::SMLSD: case ARM::SMLSDX: + case ARM::USAD8: + if (R19_16 == 15 || R11_8 == 15 || R3_0 == 15) + return true; + return false; + case ARM::SMLAL: case ARM::SMULL: case ARM::UMAAL: case ARM::UMLAL: + case ARM::UMULL: + case ARM::SMLALBB: case ARM::SMLALBT: case ARM::SMLALTB: case ARM::SMLALTT: + case ARM::SMLALD: case ARM::SMLALDX: case ARM::SMLSLD: case ARM::SMLSLDX: + if (R19_16 == 15 || R15_12 == 15 || R11_8 == 15 || R3_0 == 15) + return true; + if (R19_16 == R15_12) + return true; + return false;; + } +} + // Multiply Instructions. -// MLA, MLS, SMLABB, SMLABT, SMLATB, SMLATT, SMLAWB, SMLAWT, SMMLA, SMMLS: +// MLA, MLS, SMLABB, SMLABT, SMLATB, SMLATT, SMLAWB, SMLAWT, SMMLA, SMMLAR, +// SMMLS, SMMLAR, SMLAD, SMLADX, SMLSD, SMLSDX, and USADA8 (for convenience): // Rd{19-16} Rn{3-0} Rm{11-8} Ra{15-12} +// But note that register checking for {SMLAD, SMLADX, SMLSD, SMLSDX} is +// only for {d, n, m}. // -// MUL, SMMUL, SMULBB, SMULBT, SMULTB, SMULTT, SMULWB, SMULWT: +// MUL, SMMUL, SMMULR, SMULBB, SMULBT, SMULTB, SMULTT, SMULWB, SMULWT, SMUAD, +// SMUADX, and USAD8 (for convenience): // Rd{19-16} Rn{3-0} Rm{11-8} // -// SMLAL, SMULL, UMAAL, UMLAL, UMULL, SMLALBB, SMLALBT, SMLALTB, SMLALTT: +// SMLAL, SMULL, UMAAL, UMLAL, UMULL, SMLALBB, SMLALBT, SMLALTB, SMLALTT, +// SMLALD, SMLADLX, SMLSLD, SMLSLDX: // RdLo{15-12} RdHi{19-16} Rn{3-0} Rm{11-8} // // The mapping of the multiply registers to the "regular" ARM registers, where @@ -531,6 +602,10 @@ static bool DisassembleMulFrm(MCInst &MI, unsigned Opcode, uint32_t insn, && OpInfo[2].RegClass == ARM::GPRRegClassID && "Expect three register operands"); + // Sanity check for the register encodings. + if (BadRegsMulFrm(Opcode, insn)) + return false; + // Instructions with two destination registers have RdLo{15-12} first. if (NumDefs == 2) { assert(NumOps >= 4 && OpInfo[3].RegClass == ARM::GPRRegClassID && @@ -618,18 +693,38 @@ static inline unsigned GetCopOpc(uint32_t insn) { static bool DisassembleCoprocessor(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - assert(NumOps >= 5 && "Num of operands >= 5 for coprocessor instr"); + assert(NumOps >= 4 && "Num of operands >= 4 for coprocessor instr"); unsigned &OpIdx = NumOpsAdded; + // A8.6.92 + // if coproc == '101x' then SEE "Advanced SIMD and VFP" + // But since the special instructions have more explicit encoding bits + // specified, if coproc == 10 or 11, we should reject it as invalid. + unsigned coproc = GetCoprocessor(insn); + if ((Opcode == ARM::MCR || Opcode == ARM::MCRR || + Opcode == ARM::MRC || Opcode == ARM::MRRC) && + (coproc == 10 || coproc == 11)) { + DEBUG(errs() << "Encoding error: coproc == 10 or 11 for MCR[R]/MR[R]C\n"); + return false; + } + bool OneCopOpc = (Opcode == ARM::MCRR || Opcode == ARM::MCRR2 || Opcode == ARM::MRRC || Opcode == ARM::MRRC2); + // CDP/CDP2 has no GPR operand; the opc1 operand is also wider (Inst{23-20}). bool NoGPR = (Opcode == ARM::CDP || Opcode == ARM::CDP2); bool LdStCop = LdStCopOpcode(Opcode); + bool RtOut = (Opcode == ARM::MRC || Opcode == ARM::MRC2); OpIdx = 0; - MI.addOperand(MCOperand::CreateImm(GetCoprocessor(insn))); + if (RtOut) { + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, + decodeRd(insn)))); + ++OpIdx; + } + MI.addOperand(MCOperand::CreateImm(coproc)); + ++OpIdx; if (LdStCop) { // Unindex if P:W = 0b00 --> _OPTION variant @@ -639,26 +734,34 @@ static bool DisassembleCoprocessor(MCInst &MI, unsigned Opcode, uint32_t insn, MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRn(insn)))); + OpIdx += 2; if (PW) { MI.addOperand(MCOperand::CreateReg(0)); ARM_AM::AddrOpc AddrOpcode = getUBit(insn) ? ARM_AM::add : ARM_AM::sub; + const TargetInstrDesc &TID = ARMInsts[Opcode]; + unsigned IndexMode = + (TID.TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift; unsigned Offset = ARM_AM::getAM2Opc(AddrOpcode, slice(insn, 7, 0) << 2, - ARM_AM::no_shift); + ARM_AM::no_shift, IndexMode); MI.addOperand(MCOperand::CreateImm(Offset)); - OpIdx = 5; + OpIdx += 2; } else { MI.addOperand(MCOperand::CreateImm(slice(insn, 7, 0))); - OpIdx = 4; + ++OpIdx; } } else { MI.addOperand(MCOperand::CreateImm(OneCopOpc ? GetCopOpc(insn) : GetCopOpc1(insn, NoGPR))); + ++OpIdx; - MI.addOperand(NoGPR ? MCOperand::CreateImm(decodeRd(insn)) - : MCOperand::CreateReg( - getRegisterEnum(B, ARM::GPRRegClassID, - decodeRd(insn)))); + if (!RtOut) { + MI.addOperand(NoGPR ? MCOperand::CreateImm(decodeRd(insn)) + : MCOperand::CreateReg( + getRegisterEnum(B, ARM::GPRRegClassID, + decodeRd(insn)))); + ++OpIdx; + } MI.addOperand(OneCopOpc ? MCOperand::CreateReg( getRegisterEnum(B, ARM::GPRRegClassID, @@ -667,7 +770,7 @@ static bool DisassembleCoprocessor(MCInst &MI, unsigned Opcode, uint32_t insn, MI.addOperand(MCOperand::CreateImm(decodeRm(insn))); - OpIdx = 5; + OpIdx += 2; if (!OneCopOpc) { MI.addOperand(MCOperand::CreateImm(GetCopOpc2(insn))); @@ -679,8 +782,8 @@ static bool DisassembleCoprocessor(MCInst &MI, unsigned Opcode, uint32_t insn, } // Branch Instructions. -// BLr9: SignExtend(Imm24:'00', 32) -// Bcc, BLr9_pred: SignExtend(Imm24:'00', 32) Pred0 Pred1 +// BL: SignExtend(Imm24:'00', 32) +// Bcc, BL_pred: SignExtend(Imm24:'00', 32) Pred0 Pred1 // SMC: ZeroExtend(imm4, 32) // SVC: ZeroExtend(Imm24, 32) // @@ -735,6 +838,11 @@ static bool DisassembleBrFrm(MCInst &MI, unsigned Opcode, uint32_t insn, // MSRi take a mask, followed by one so_imm operand. The mask contains the // R Bit in bit 4, and the special register fields in bits 3-0. if (Opcode == ARM::MSRi) { + // A5.2.11 MSR (immediate), and hints & B6.1.6 MSR (immediate) + // The hints instructions have more specific encodings, so if mask == 0, + // we should reject this as an invalid instruction. + if (slice(insn, 19, 16) == 0) + return false; MI.addOperand(MCOperand::CreateImm(slice(insn, 22, 22) << 4 /* R Bit */ | slice(insn, 19, 16) /* Special Reg */ )); // SOImm is 4-bit rotate amount in bits 11-8 with 8-bit imm in bits 7-0. @@ -760,11 +868,11 @@ static bool DisassembleBrFrm(MCInst &MI, unsigned Opcode, uint32_t insn, return true; } - assert((Opcode == ARM::Bcc || Opcode == ARM::BLr9 || Opcode == ARM::BLr9_pred + assert((Opcode == ARM::Bcc || Opcode == ARM::BL || Opcode == ARM::BL_pred || Opcode == ARM::SMC || Opcode == ARM::SVC) && "Unexpected Opcode"); - assert(NumOps >= 1 && OpInfo[0].RegClass < 0 && "Reg operand expected"); + assert(NumOps >= 1 && OpInfo[0].RegClass < 0 && "Imm operand expected"); int Imm32 = 0; if (Opcode == ARM::SMC) { @@ -778,12 +886,6 @@ static bool DisassembleBrFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned Imm26 = slice(insn, 23, 0) << 2; //Imm32 = signextend<signed int, 26>(Imm26); Imm32 = SignExtend32<26>(Imm26); - - // When executing an ARM instruction, PC reads as the address of the current - // instruction plus 8. The assembler subtracts 8 from the difference - // between the branch instruction and the target address, disassembler has - // to add 8 to compensate. - Imm32 += 8; } MI.addOperand(MCOperand::CreateImm(Imm32)); @@ -793,7 +895,7 @@ static bool DisassembleBrFrm(MCInst &MI, unsigned Opcode, uint32_t insn, } // Misc. Branch Instructions. -// BLXr9, BXr9 +// BLX, BLXi, BX // BX, BX_RET static bool DisassembleBrMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { @@ -809,8 +911,9 @@ static bool DisassembleBrMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, if (Opcode == ARM::BX_RET || Opcode == ARM::MOVPCLR) return true; - // BLXr9 and BX take one GPR reg. - if (Opcode == ARM::BLXr9 || Opcode == ARM::BX) { + // BLX and BX take one GPR reg. + if (Opcode == ARM::BLX || Opcode == ARM::BLX_pred || + Opcode == ARM::BX) { assert(NumOps >= 1 && OpInfo[OpIdx].RegClass == ARM::GPRRegClassID && "Reg operand expected"); MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, @@ -819,6 +922,17 @@ static bool DisassembleBrMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, return true; } + // BLXi takes imm32 (the PC offset). + if (Opcode == ARM::BLXi) { + assert(NumOps >= 1 && OpInfo[0].RegClass < 0 && "Imm operand expected"); + // SignExtend(imm24:H:'0', 32) where imm24 = Inst{23-0} and H = Inst{24}. + unsigned Imm26 = slice(insn, 23, 0) << 2 | slice(insn, 24, 24) << 1; + int Imm32 = SignExtend32<26>(Imm26); + MI.addOperand(MCOperand::CreateImm(Imm32)); + OpIdx = 1; + return true; + } + return false; } @@ -837,6 +951,24 @@ static inline bool getBFCInvMask(uint32_t insn, uint32_t &mask) { return true; } +// Standard data-processing instructions allow PC as a register specifier, +// but we should reject other DPFrm instructions with PC as registers. +static bool BadRegsDPFrm(unsigned Opcode, uint32_t insn) { + switch (Opcode) { + default: + // Did we miss an opcode? + if (decodeRd(insn) == 15 || decodeRn(insn) == 15 || decodeRm(insn) == 15) { + DEBUG(errs() << "DPFrm with bad reg specifier(s)\n"); + return true; + } + case ARM::ADCrr: case ARM::ADDSrr: case ARM::ADDrr: case ARM::ANDrr: + case ARM::BICrr: case ARM::CMNzrr: case ARM::CMPrr: case ARM::EORrr: + case ARM::ORRrr: case ARM::RSBrr: case ARM::RSCrr: case ARM::SBCrr: + case ARM::SUBSrr: case ARM::SUBrr: case ARM::TEQrr: case ARM::TSTrr: + return false; + } +} + // A major complication is the fact that some of the saturating add/subtract // operations have Rd Rm Rn, instead of the "normal" Rd Rn Rm. // They are QADD, QDADD, QDSUB, and QSUB. @@ -864,6 +996,10 @@ static bool DisassembleDPFrm(MCInst &MI, unsigned Opcode, uint32_t insn, // Special-case handling of BFC/BFI/SBFX/UBFX. if (Opcode == ARM::BFC || Opcode == ARM::BFI) { + // A8.6.17 BFC & A8.6.18 BFI + // Sanity check Rd. + if (decodeRd(insn) == 15) + return false; MI.addOperand(MCOperand::CreateReg(0)); if (Opcode == ARM::BFI) { MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, @@ -879,6 +1015,9 @@ static bool DisassembleDPFrm(MCInst &MI, unsigned Opcode, uint32_t insn, return true; } if (Opcode == ARM::SBFX || Opcode == ARM::UBFX) { + // Sanity check Rd and Rm. + if (decodeRd(insn) == 15 || decodeRm(insn) == 15) + return false; MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRm(insn)))); MI.addOperand(MCOperand::CreateImm(slice(insn, 11, 7))); @@ -915,15 +1054,21 @@ static bool DisassembleDPFrm(MCInst &MI, unsigned Opcode, uint32_t insn, // Assert disabled because saturating operations, e.g., A8.6.127 QASX, are // routed here as well. // assert(getIBit(insn) == 0 && "I_Bit != '0' reg/reg form"); + if (BadRegsDPFrm(Opcode, insn)) + return false; MI.addOperand(MCOperand::CreateReg( getRegisterEnum(B, ARM::GPRRegClassID, RmRn? decodeRn(insn) : decodeRm(insn)))); ++OpIdx; } else if (Opcode == ARM::MOVi16 || Opcode == ARM::MOVTi16) { + // These two instructions don't allow d as 15. + if (decodeRd(insn) == 15) + return false; // We have an imm16 = imm4:imm12 (imm4=Inst{19:16}, imm12 = Inst{11:0}). assert(getIBit(insn) == 1 && "I_Bit != '1' reg/imm form"); unsigned Imm16 = slice(insn, 19, 16) << 12 | slice(insn, 11, 0); - MI.addOperand(MCOperand::CreateImm(Imm16)); + if (!B->tryAddingSymbolicOperand(Imm16, 4, MI)) + MI.addOperand(MCOperand::CreateImm(Imm16)); ++OpIdx; } else { // We have a reg/imm form. @@ -992,6 +1137,21 @@ static bool DisassembleDPSoRegFrm(MCInst &MI, unsigned Opcode, uint32_t insn, MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRm(insn)))); if (Rs) { + // If Inst{7} != 0, we should reject this insn as an invalid encoding. + if (slice(insn, 7, 7)) + return false; + + // A8.6.3 ADC (register-shifted register) + // if d == 15 || n == 15 || m == 15 || s == 15 then UNPREDICTABLE; + // + // This also accounts for shift instructions (register) where, fortunately, + // Inst{19-16} = 0b0000. + // A8.6.89 LSL (register) + // if d == 15 || n == 15 || m == 15 then UNPREDICTABLE; + if (decodeRd(insn) == 15 || decodeRn(insn) == 15 || + decodeRm(insn) == 15 || decodeRs(insn) == 15) + return false; + // Register-controlled shifts: [Rm, Rs, shift]. MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRs(insn)))); @@ -1015,6 +1175,71 @@ static bool DisassembleDPSoRegFrm(MCInst &MI, unsigned Opcode, uint32_t insn, return true; } +static bool BadRegsLdStFrm(unsigned Opcode, uint32_t insn, bool Store, bool WBack, + bool Imm) { + const StringRef Name = ARMInsts[Opcode].Name; + unsigned Rt = decodeRd(insn); + unsigned Rn = decodeRn(insn); + unsigned Rm = decodeRm(insn); + unsigned P = getPBit(insn); + unsigned W = getWBit(insn); + + if (Store) { + // Only STR (immediate, register) allows PC as the source. + if (Name.startswith("STRB") && Rt == 15) { + DEBUG(errs() << "if t == 15 then UNPREDICTABLE\n"); + return true; + } + if (WBack && (Rn == 15 || Rn == Rt)) { + DEBUG(errs() << "if wback && (n == 15 || n == t) then UNPREDICTABLE\n"); + return true; + } + if (!Imm && Rm == 15) { + DEBUG(errs() << "if m == 15 then UNPREDICTABLE\n"); + return true; + } + } else { + // Only LDR (immediate, register) allows PC as the destination. + if (Name.startswith("LDRB") && Rt == 15) { + DEBUG(errs() << "if t == 15 then UNPREDICTABLE\n"); + return true; + } + if (Imm) { + // Immediate + if (Rn == 15) { + // The literal form must be in offset mode; it's an encoding error + // otherwise. + if (!(P == 1 && W == 0)) { + DEBUG(errs() << "Ld literal form with !(P == 1 && W == 0)\n"); + return true; + } + // LDRB (literal) does not allow PC as the destination. + if (Opcode != ARM::LDRi12 && Rt == 15) { + DEBUG(errs() << "if t == 15 then UNPREDICTABLE\n"); + return true; + } + } else { + // Write back while Rn == Rt does not make sense. + if (WBack && (Rn == Rt)) { + DEBUG(errs() << "if wback && n == t then UNPREDICTABLE\n"); + return true; + } + } + } else { + // Register + if (Rm == 15) { + DEBUG(errs() << "if m == 15 then UNPREDICTABLE\n"); + return true; + } + if (WBack && (Rn == 15 || Rn == Rt)) { + DEBUG(errs() << "if wback && (n == 15 || n == t) then UNPREDICTABLE\n"); + return true; + } + } + } + return false; +} + static bool DisassembleLdStFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, bool isStore, BO B) { @@ -1077,19 +1302,41 @@ static bool DisassembleLdStFrm(MCInst &MI, unsigned Opcode, uint32_t insn, if (OpIdx + 1 >= NumOps) return false; - assert((OpInfo[OpIdx].RegClass == ARM::GPRRegClassID) && - (OpInfo[OpIdx+1].RegClass < 0) && - "Expect 1 reg operand followed by 1 imm operand"); + if (BadRegsLdStFrm(Opcode, insn, isStore, isPrePost, getIBit(insn)==0)) + return false; ARM_AM::AddrOpc AddrOpcode = getUBit(insn) ? ARM_AM::add : ARM_AM::sub; + unsigned IndexMode = + (TID.TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift; if (getIBit(insn) == 0) { - MI.addOperand(MCOperand::CreateReg(0)); + // For pre- and post-indexed case, add a reg0 operand (Addressing Mode #2). + // Otherwise, skip the reg operand since for addrmode_imm12, Rn has already + // been populated. + if (isPrePost) { + MI.addOperand(MCOperand::CreateReg(0)); + OpIdx += 1; + } - // Disassemble the 12-bit immediate offset. unsigned Imm12 = slice(insn, 11, 0); - unsigned Offset = ARM_AM::getAM2Opc(AddrOpcode, Imm12, ARM_AM::no_shift); - MI.addOperand(MCOperand::CreateImm(Offset)); + if (Opcode == ARM::LDRBi12 || Opcode == ARM::LDRi12 || + Opcode == ARM::STRBi12 || Opcode == ARM::STRi12) { + // Disassemble the 12-bit immediate offset, which is the second operand in + // $addrmode_imm12 => (ops GPR:$base, i32imm:$offsimm). + int Offset = AddrOpcode == ARM_AM::add ? 1 * Imm12 : -1 * Imm12; + MI.addOperand(MCOperand::CreateImm(Offset)); + } else { + // Disassemble the 12-bit immediate offset, which is the second operand in + // $am2offset => (ops GPR, i32imm). + unsigned Offset = ARM_AM::getAM2Opc(AddrOpcode, Imm12, ARM_AM::no_shift, + IndexMode); + MI.addOperand(MCOperand::CreateImm(Offset)); + } + OpIdx += 1; } else { + // If Inst{25} = 1 and Inst{4} != 0, we should reject this as invalid. + if (slice(insn,4,4) == 1) + return false; + // Disassemble the offset reg (Rm), shift type, and immediate shift length. MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRm(insn)))); @@ -1101,9 +1348,9 @@ static bool DisassembleLdStFrm(MCInst &MI, unsigned Opcode, uint32_t insn, // A8.4.1. Possible rrx or shift amount of 32... getImmShiftSE(ShOp, ShImm); MI.addOperand(MCOperand::CreateImm( - ARM_AM::getAM2Opc(AddrOpcode, ShImm, ShOp))); + ARM_AM::getAM2Opc(AddrOpcode, ShImm, ShOp, IndexMode))); + OpIdx += 2; } - OpIdx += 2; return true; } @@ -1125,7 +1372,7 @@ static bool HasDualReg(unsigned Opcode) { case ARM::LDRD: case ARM::LDRD_PRE: case ARM::LDRD_POST: case ARM::STRD: case ARM::STRD_PRE: case ARM::STRD_POST: return true; - } + } } static bool DisassembleLdStMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, @@ -1153,8 +1400,6 @@ static bool DisassembleLdStMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, ++OpIdx; } - bool DualReg = HasDualReg(Opcode); - // Disassemble the dst/src operand. if (OpIdx >= NumOps) return false; @@ -1165,8 +1410,8 @@ static bool DisassembleLdStMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, decodeRd(insn)))); ++OpIdx; - // Fill in LDRD and STRD's second operand. - if (DualReg) { + // Fill in LDRD and STRD's second operand Rt operand. + if (HasDualReg(Opcode)) { MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRd(insn) + 1))); ++OpIdx; @@ -1188,7 +1433,7 @@ static bool DisassembleLdStMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, assert(OpInfo[OpIdx].RegClass == ARM::GPRRegClassID && "Reg operand expected"); assert((!isPrePost || (TID.getOperandConstraint(OpIdx, TOI::TIED_TO) != -1)) - && "Index mode or tied_to operand expected"); + && "Offset mode or tied_to operand expected"); MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRn(insn)))); ++OpIdx; @@ -1204,19 +1449,22 @@ static bool DisassembleLdStMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, "Expect 1 reg operand followed by 1 imm operand"); ARM_AM::AddrOpc AddrOpcode = getUBit(insn) ? ARM_AM::add : ARM_AM::sub; + unsigned IndexMode = + (TID.TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift; if (getAM3IBit(insn) == 1) { MI.addOperand(MCOperand::CreateReg(0)); // Disassemble the 8-bit immediate offset. unsigned Imm4H = (insn >> ARMII::ImmHiShift) & 0xF; unsigned Imm4L = insn & 0xF; - unsigned Offset = ARM_AM::getAM3Opc(AddrOpcode, (Imm4H << 4) | Imm4L); + unsigned Offset = ARM_AM::getAM3Opc(AddrOpcode, (Imm4H << 4) | Imm4L, + IndexMode); MI.addOperand(MCOperand::CreateImm(Offset)); } else { // Disassemble the offset reg (Rm). MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRm(insn)))); - unsigned Offset = ARM_AM::getAM3Opc(AddrOpcode, 0); + unsigned Offset = ARM_AM::getAM3Opc(AddrOpcode, 0, IndexMode); MI.addOperand(MCOperand::CreateImm(Offset)); } OpIdx += 2; @@ -1236,13 +1484,13 @@ static bool DisassembleStMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, } // The algorithm for disassembly of LdStMulFrm is different from others because -// it explicitly populates the two predicate operands after operand 0 (the base) -// and operand 1 (the AM4 mode imm). After operand 3, we need to populate the -// reglist with each affected register encoded as an MCOperand. +// it explicitly populates the two predicate operands after the base register. +// After that, we need to populate the reglist with each affected register +// encoded as an MCOperand. static bool DisassembleLdStMulFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - assert(NumOps >= 5 && "LdStMulFrm expects NumOps >= 5"); + assert(NumOps >= 4 && "LdStMulFrm expects NumOps >= 4"); NumOpsAdded = 0; unsigned Base = getRegisterEnum(B, ARM::GPRRegClassID, decodeRn(insn)); @@ -1260,8 +1508,10 @@ static bool DisassembleLdStMulFrm(MCInst &MI, unsigned Opcode, uint32_t insn, MI.addOperand(MCOperand::CreateReg(Base)); // Handling the two predicate operands before the reglist. - int64_t CondVal = insn >> ARMII::CondShift; - MI.addOperand(MCOperand::CreateImm(CondVal == 0xF ? 0xE : CondVal)); + int64_t CondVal = getCondField(insn); + if (CondVal == 0xF) + return false; + MI.addOperand(MCOperand::CreateImm(CondVal)); MI.addOperand(MCOperand::CreateReg(ARM::CPSR)); NumOpsAdded += 3; @@ -1352,6 +1602,12 @@ static bool DisassembleArithMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, bool ThreeReg = NumOps > 2 && OpInfo[2].RegClass == ARM::GPRRegClassID; + // Sanity check the registers, which should not be 15. + if (decodeRd(insn) == 15 || decodeRm(insn) == 15) + return false; + if (ThreeReg && decodeRn(insn) == 15) + return false; + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRd(insn)))); ++OpIdx; @@ -1376,7 +1632,7 @@ static bool DisassembleArithMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, ARM_AM::ShiftOpc Opc = ARM_AM::no_shift; if (Opcode == ARM::PKHBT) Opc = ARM_AM::lsl; - else if (Opcode == ARM::PKHBT) + else if (Opcode == ARM::PKHTB) Opc = ARM_AM::asr; getImmShiftSE(Opc, ShiftAmt); MI.addOperand(MCOperand::CreateImm(ARM_AM::getSORegOpc(Opc, ShiftAmt))); @@ -1391,6 +1647,11 @@ static bool DisassembleArithMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleSatFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { + // A8.6.183 SSAT + // if d == 15 || n == 15 then UNPREDICTABLE; + if (decodeRd(insn) == 15 || decodeRm(insn) == 15) + return false; + const TargetInstrDesc &TID = ARMInsts[Opcode]; NumOpsAdded = TID.getNumOperands() - 2; // ignore predicate operands @@ -1429,6 +1690,11 @@ static bool DisassembleSatFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleExtFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { + // A8.6.220 SXTAB + // if d == 15 || m == 15 then UNPREDICTABLE; + if (decodeRd(insn) == 15 || decodeRm(insn) == 15) + return false; + const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; unsigned &OpIdx = NumOpsAdded; @@ -1611,7 +1877,7 @@ static bool DisassembleVFPBinaryFrm(MCInst &MI, unsigned Opcode, uint32_t insn, // A8.6.295 vcvt (floating-point <-> integer) // Int to FP: VSITOD, VSITOS, VUITOD, VUITOS // FP to Int: VTOSI[Z|R]D, VTOSI[Z|R]S, VTOUI[Z|R]D, VTOUI[Z|R]S -// +// // A8.6.297 vcvt (floating-point and fixed-point) // Dd|Sd Dd|Sd(TIED_TO) #fbits(= 16|32 - UInt(imm4:i)) static bool DisassembleVFPConv1Frm(MCInst &MI, unsigned Opcode, uint32_t insn, @@ -1800,15 +2066,14 @@ static bool DisassembleVFPLdStFrm(MCInst &MI, unsigned Opcode, uint32_t insn, } // VFP Load/Store Multiple Instructions. -// This is similar to the algorithm for LDM/STM in that operand 0 (the base) and -// operand 1 (the AM4 mode imm) is followed by two predicate operands. It is -// followed by a reglist of either DPR(s) or SPR(s). +// We have an optional write back reg, the base, and two predicate operands. +// It is then followed by a reglist of either DPR(s) or SPR(s). // // VLDMD[_UPD], VLDMS[_UPD], VSTMD[_UPD], VSTMS[_UPD] static bool DisassembleVFPLdStMulFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - assert(NumOps >= 5 && "VFPLdStMulFrm expects NumOps >= 5"); + assert(NumOps >= 4 && "VFPLdStMulFrm expects NumOps >= 4"); unsigned &OpIdx = NumOpsAdded; @@ -1827,25 +2092,18 @@ static bool DisassembleVFPLdStMulFrm(MCInst &MI, unsigned Opcode, uint32_t insn, MI.addOperand(MCOperand::CreateReg(Base)); - // Next comes the AM4 Opcode. - ARM_AM::AMSubMode SubMode = getAMSubModeForBits(getPUBits(insn)); - // Must be either "ia" or "db" submode. - if (SubMode != ARM_AM::ia && SubMode != ARM_AM::db) { - DEBUG(errs() << "Illegal addressing mode 4 sub-mode!\n"); - return false; - } - MI.addOperand(MCOperand::CreateImm(ARM_AM::getAM4ModeImm(SubMode))); - // Handling the two predicate operands before the reglist. - int64_t CondVal = insn >> ARMII::CondShift; - MI.addOperand(MCOperand::CreateImm(CondVal == 0xF ? 0xE : CondVal)); + int64_t CondVal = getCondField(insn); + if (CondVal == 0xF) + return false; + MI.addOperand(MCOperand::CreateImm(CondVal)); MI.addOperand(MCOperand::CreateReg(ARM::CPSR)); - OpIdx += 4; + OpIdx += 3; - bool isSPVFP = (Opcode == ARM::VLDMSIA || Opcode == ARM::VLDMSDB || + bool isSPVFP = (Opcode == ARM::VLDMSIA || Opcode == ARM::VLDMSIA_UPD || Opcode == ARM::VLDMSDB_UPD || - Opcode == ARM::VSTMSIA || Opcode == ARM::VSTMSDB || + Opcode == ARM::VSTMSIA || Opcode == ARM::VSTMSIA_UPD || Opcode == ARM::VSTMSDB_UPD); unsigned RegClassID = isSPVFP ? ARM::SPRRegClassID : ARM::DPRRegClassID; @@ -1855,6 +2113,11 @@ static bool DisassembleVFPLdStMulFrm(MCInst &MI, unsigned Opcode, uint32_t insn, // Fill the variadic part of reglist. unsigned char Imm8 = insn & 0xFF; unsigned Regs = isSPVFP ? Imm8 : Imm8/2; + + // Apply some sanity checks before proceeding. + if (Regs == 0 || (RegD + Regs) > 32 || (!isSPVFP && Regs > 16)) + return false; + for (unsigned i = 0; i < Regs; ++i) { MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, RegClassID, RegD + i))); @@ -2136,7 +2399,7 @@ static unsigned decodeN3VImm(uint32_t insn) { // Correctly set VLD*/VST*'s TIED_TO GPR, as the asm printer needs it. static bool DisassembleNLdSt0(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, bool Store, bool DblSpaced, - BO B) { + unsigned alignment, BO B) { const TargetInstrDesc &TID = ARMInsts[Opcode]; const TargetOperandInfo *OpInfo = TID.OpInfo; @@ -2180,9 +2443,10 @@ static bool DisassembleNLdSt0(MCInst &MI, unsigned Opcode, uint32_t insn, assert((OpIdx+1) < NumOps && OpInfo[OpIdx].RegClass == ARM::GPRRegClassID && OpInfo[OpIdx + 1].RegClass < 0 && "Addrmode #6 Operands expected"); + // addrmode6 := (ops GPR:$addr, i32imm) MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, Rn))); - MI.addOperand(MCOperand::CreateImm(0)); // Alignment ignored? + MI.addOperand(MCOperand::CreateImm(alignment)); // Alignment OpIdx += 2; if (WB) { @@ -2230,9 +2494,10 @@ static bool DisassembleNLdSt0(MCInst &MI, unsigned Opcode, uint32_t insn, assert((OpIdx+1) < NumOps && OpInfo[OpIdx].RegClass == ARM::GPRRegClassID && OpInfo[OpIdx + 1].RegClass < 0 && "Addrmode #6 Operands expected"); + // addrmode6 := (ops GPR:$addr, i32imm) MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, Rn))); - MI.addOperand(MCOperand::CreateImm(0)); // Alignment ignored? + MI.addOperand(MCOperand::CreateImm(alignment)); // Alignment OpIdx += 2; if (WB) { @@ -2263,6 +2528,92 @@ static bool DisassembleNLdSt0(MCInst &MI, unsigned Opcode, uint32_t insn, return true; } +// A8.6.308, A8.6.311, A8.6.314, A8.6.317. +static bool Align4OneLaneInst(unsigned elem, unsigned size, + unsigned index_align, unsigned & alignment) { + unsigned bits = 0; + switch (elem) { + default: + return false; + case 1: + // A8.6.308 + if (size == 0) + return slice(index_align, 0, 0) == 0; + else if (size == 1) { + bits = slice(index_align, 1, 0); + if (bits != 0 && bits != 1) + return false; + if (bits == 1) + alignment = 16; + return true; + } else if (size == 2) { + bits = slice(index_align, 2, 0); + if (bits != 0 && bits != 3) + return false; + if (bits == 3) + alignment = 32; + return true;; + } + return true; + case 2: + // A8.6.311 + if (size == 0) { + if (slice(index_align, 0, 0) == 1) + alignment = 16; + return true; + } if (size == 1) { + if (slice(index_align, 0, 0) == 1) + alignment = 32; + return true; + } else if (size == 2) { + if (slice(index_align, 1, 1) != 0) + return false; + if (slice(index_align, 0, 0) == 1) + alignment = 64; + return true;; + } + return true; + case 3: + // A8.6.314 + if (size == 0) { + if (slice(index_align, 0, 0) != 0) + return false; + return true; + } if (size == 1) { + if (slice(index_align, 0, 0) != 0) + return false; + return true; + return true; + } else if (size == 2) { + if (slice(index_align, 1, 0) != 0) + return false; + return true;; + } + return true; + case 4: + // A8.6.317 + if (size == 0) { + if (slice(index_align, 0, 0) == 1) + alignment = 32; + return true; + } if (size == 1) { + if (slice(index_align, 0, 0) == 1) + alignment = 64; + return true; + } else if (size == 2) { + bits = slice(index_align, 1, 0); + if (bits == 3) + return false; + if (bits == 1) + alignment = 64; + else if (bits == 2) + alignment = 128; + return true;; + } + return true; + } +} + // A7.7 // If L (Inst{21}) == 0, store instructions. // Find out about double-spaced-ness of the Opcode and pass it on to @@ -2272,11 +2623,33 @@ static bool DisassembleNLdSt(MCInst &MI, unsigned Opcode, uint32_t insn, const StringRef Name = ARMInsts[Opcode].Name; bool DblSpaced = false; + // 0 represents standard alignment, i.e., unaligned data access. + unsigned alignment = 0; + + unsigned elem = 0; // legal values: {1, 2, 3, 4} + if (Name.startswith("VST1") || Name.startswith("VLD1")) + elem = 1; + + if (Name.startswith("VST2") || Name.startswith("VLD2")) + elem = 2; + + if (Name.startswith("VST3") || Name.startswith("VLD3")) + elem = 3; + + if (Name.startswith("VST4") || Name.startswith("VLD4")) + elem = 4; if (Name.find("LN") != std::string::npos) { // To one lane instructions. // See, for example, 8.6.317 VLD4 (single 4-element structure to one lane). + // Utility function takes number of elements, size, and index_align. + if (!Align4OneLaneInst(elem, + slice(insn, 11, 10), + slice(insn, 7, 4), + alignment)) + return false; + // <size> == 16 && Inst{5} == 1 --> DblSpaced = true if (Name.endswith("16") || Name.endswith("16_UPD")) DblSpaced = slice(insn, 5, 5) == 1; @@ -2284,30 +2657,102 @@ static bool DisassembleNLdSt(MCInst &MI, unsigned Opcode, uint32_t insn, // <size> == 32 && Inst{6} == 1 --> DblSpaced = true if (Name.endswith("32") || Name.endswith("32_UPD")) DblSpaced = slice(insn, 6, 6) == 1; - + } else if (Name.find("DUP") != std::string::npos) { + // Single element (or structure) to all lanes. + // Inst{9-8} encodes the number of element(s) in the structure, with: + // 0b00 (VLD1DUP) (for this, a bit makes sense only for data size 16 and 32. + // 0b01 (VLD2DUP) + // 0b10 (VLD3DUP) (for this, a bit must be encoded as 0) + // 0b11 (VLD4DUP) + // + // Inst{7-6} encodes the data size, with: + // 0b00 => 8, 0b01 => 16, 0b10 => 32 + // + // Inst{4} (the a bit) encodes the align action (0: standard alignment) + unsigned elem = slice(insn, 9, 8) + 1; + unsigned a = slice(insn, 4, 4); + if (elem != 3) { + // 0b11 is not a valid encoding for Inst{7-6}. + if (slice(insn, 7, 6) == 3) + return false; + unsigned data_size = 8 << slice(insn, 7, 6); + // For VLD1DUP, a bit makes sense only for data size of 16 and 32. + if (a && data_size == 8) + return false; + + // Now we can calculate the alignment! + if (a) + alignment = elem * data_size; + } else { + if (a) { + // A8.6.315 VLD3 (single 3-element structure to all lanes) + // The a bit must be encoded as 0. + return false; + } + } } else { // Multiple n-element structures with type encoded as Inst{11-8}. // See, for example, A8.6.316 VLD4 (multiple 4-element structures). - // n == 2 && type == 0b1001 -> DblSpaced = true - if (Name.startswith("VST2") || Name.startswith("VLD2")) - DblSpaced = slice(insn, 11, 8) == 9; - - // n == 3 && type == 0b0101 -> DblSpaced = true - if (Name.startswith("VST3") || Name.startswith("VLD3")) - DblSpaced = slice(insn, 11, 8) == 5; - - // n == 4 && type == 0b0001 -> DblSpaced = true - if (Name.startswith("VST4") || Name.startswith("VLD4")) - DblSpaced = slice(insn, 11, 8) == 1; - + // Inst{5-4} encodes alignment. + unsigned align = slice(insn, 5, 4); + switch (align) { + default: + break; + case 1: + alignment = 64; break; + case 2: + alignment = 128; break; + case 3: + alignment = 256; break; + } + + unsigned type = slice(insn, 11, 8); + // Reject UNDEFINED instructions based on type and align. + // Plus set DblSpaced flag where appropriate. + switch (elem) { + default: + break; + case 1: + // n == 1 + // A8.6.307 & A8.6.391 + if ((type == 7 && slice(align, 1, 1) == 1) || + (type == 10 && align == 3) || + (type == 6 && slice(align, 1, 1) == 1)) + return false; + break; + case 2: + // n == 2 && type == 0b1001 -> DblSpaced = true + // A8.6.310 & A8.6.393 + if ((type == 8 || type == 9) && align == 3) + return false; + DblSpaced = (type == 9); + break; + case 3: + // n == 3 && type == 0b0101 -> DblSpaced = true + // A8.6.313 & A8.6.395 + if (slice(insn, 7, 6) == 3 || slice(align, 1, 1) == 1) + return false; + DblSpaced = (type == 5); + break; + case 4: + // n == 4 && type == 0b0001 -> DblSpaced = true + // A8.6.316 & A8.6.397 + if (slice(insn, 7, 6) == 3) + return false; + DblSpaced = (type == 1); + break; + } } return DisassembleNLdSt0(MI, Opcode, insn, NumOps, NumOpsAdded, - slice(insn, 21, 21) == 0, DblSpaced, B); + slice(insn, 21, 21) == 0, DblSpaced, alignment/8, B); } // VMOV (immediate) // Qd/Dd imm +// VBIC (immediate) +// VORR (immediate) +// Qd/Dd imm src(=Qd/Dd) static bool DisassembleN1RegModImmFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { @@ -2334,12 +2779,20 @@ static bool DisassembleN1RegModImmFrm(MCInst &MI, unsigned Opcode, case ARM::VMOVv8i16: case ARM::VMVNv4i16: case ARM::VMVNv8i16: + case ARM::VBICiv4i16: + case ARM::VBICiv8i16: + case ARM::VORRiv4i16: + case ARM::VORRiv8i16: esize = ESize16; break; case ARM::VMOVv2i32: case ARM::VMOVv4i32: case ARM::VMVNv2i32: case ARM::VMVNv4i32: + case ARM::VBICiv2i32: + case ARM::VBICiv4i32: + case ARM::VORRiv2i32: + case ARM::VORRiv4i32: esize = ESize32; break; case ARM::VMOVv1i64: @@ -2347,7 +2800,7 @@ static bool DisassembleN1RegModImmFrm(MCInst &MI, unsigned Opcode, esize = ESize64; break; default: - assert(0 && "Unreachable code!"); + assert(0 && "Unexpected opcode!"); return false; } @@ -2356,6 +2809,16 @@ static bool DisassembleN1RegModImmFrm(MCInst &MI, unsigned Opcode, MI.addOperand(MCOperand::CreateImm(decodeN1VImm(insn, esize))); NumOpsAdded = 2; + + // VBIC/VORRiv*i* variants have an extra $src = $Vd to be filled in. + if (NumOps >= 3 && + (OpInfo[2].RegClass == ARM::DPRRegClassID || + OpInfo[2].RegClass == ARM::QPRRegClassID)) { + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, OpInfo[0].RegClass, + decodeNEONRd(insn)))); + NumOpsAdded += 1; + } + return true; } @@ -2376,7 +2839,7 @@ enum N2VFlag { // // Vector Move Long: // Qd Dm -// +// // Vector Move Narrow: // Dd Qm // @@ -2518,7 +2981,7 @@ static bool DisassembleNVectorShift(MCInst &MI, unsigned Opcode, uint32_t insn, assert(OpInfo[OpIdx].RegClass < 0 && "Imm operand expected"); // Add the imm operand. - + // VSHLL has maximum shift count as the imm, inferred from its size. unsigned Imm; switch (Opcode) { @@ -2631,7 +3094,7 @@ static bool DisassembleNVdVnVmOptImm(MCInst &MI, unsigned Opcode, uint32_t insn, // N3RegFrm. if (Opcode == ARM::VMOVDneon || Opcode == ARM::VMOVQ) return true; - + // Dm = Inst{5:3-0} => NEON Rm // or // Dm is restricted to D0-D7 if size is 16, D0-D15 otherwise @@ -2770,7 +3233,7 @@ static bool DisassembleNGetLnFrm(MCInst &MI, unsigned Opcode, uint32_t insn, ElemSize esize = Opcode == ARM::VGETLNi32 ? ESize32 : ((Opcode == ARM::VGETLNs16 || Opcode == ARM::VGETLNu16) ? ESize16 - : ESize32); + : ESize8); // Rt = Inst{15-12} => ARM Rd MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, @@ -2852,17 +3315,6 @@ static bool DisassembleNDupFrm(MCInst &MI, unsigned Opcode, uint32_t insn, return true; } -// A8.6.41 DMB -// A8.6.42 DSB -// A8.6.49 ISB -static inline bool MemBarrierInstr(uint32_t insn) { - unsigned op7_4 = slice(insn, 7, 4); - if (slice(insn, 31, 8) == 0xf57ff0 && (op7_4 >= 4 && op7_4 <= 6)) - return true; - - return false; -} - static inline bool PreLoadOpcode(unsigned Opcode) { switch(Opcode) { case ARM::PLDi12: case ARM::PLDrs: @@ -2878,8 +3330,8 @@ static bool DisassemblePreLoadFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { // Preload Data/Instruction requires either 2 or 3 operands. - // PLDi, PLDWi, PLIi: addrmode_imm12 - // PLDr[a|m], PLDWr[a|m], PLIr[a|m]: ldst_so_reg + // PLDi12, PLDWi12, PLIi12: addrmode_imm12 + // PLDrs, PLDWrs, PLIrs: ldst_so_reg MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRn(insn)))); @@ -2888,10 +3340,19 @@ static bool DisassemblePreLoadFrm(MCInst &MI, unsigned Opcode, uint32_t insn, || Opcode == ARM::PLIi12) { unsigned Imm12 = slice(insn, 11, 0); bool Negative = getUBit(insn) == 0; + + // A8.6.118 PLD (literal) PLDWi12 with Rn=PC is transformed to PLDi12. + if (Opcode == ARM::PLDWi12 && slice(insn, 19, 16) == 0xF) { + DEBUG(errs() << "Rn == '1111': PLDWi12 morphed to PLDi12\n"); + MI.setOpcode(ARM::PLDi12); + } + // -0 is represented specially. All other values are as normal. + int Offset = Negative ? -1 * Imm12 : Imm12; if (Imm12 == 0 && Negative) - Imm12 = INT32_MIN; - MI.addOperand(MCOperand::CreateImm(Imm12)); + Offset = INT32_MIN; + + MI.addOperand(MCOperand::CreateImm(Offset)); NumOpsAdded = 2; } else { MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, @@ -2917,14 +3378,20 @@ static bool DisassemblePreLoadFrm(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - if (MemBarrierInstr(insn)) { - // DMBsy, DSBsy, and ISBsy instructions have zero operand and are taken care - // of within the generic ARMBasicMCBuilder::BuildIt() method. - // + if (Opcode == ARM::DMB || Opcode == ARM::DSB) { // Inst{3-0} encodes the memory barrier option for the variants. - MI.addOperand(MCOperand::CreateImm(slice(insn, 3, 0))); - NumOpsAdded = 1; - return true; + unsigned opt = slice(insn, 3, 0); + switch (opt) { + case ARM_MB::SY: case ARM_MB::ST: + case ARM_MB::ISH: case ARM_MB::ISHST: + case ARM_MB::NSH: case ARM_MB::NSHST: + case ARM_MB::OSH: case ARM_MB::OSHST: + MI.addOperand(MCOperand::CreateImm(opt)); + NumOpsAdded = 1; + return true; + default: + return false; + } } switch (Opcode) { @@ -2936,6 +3403,11 @@ static bool DisassembleMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, case ARM::WFI: case ARM::SEV: return true; + case ARM::SWP: + case ARM::SWPB: + // SWP, SWPB: Rd Rm Rn + // Delegate to DisassembleLdStExFrm().... + return DisassembleLdStExFrm(MI, Opcode, insn, NumOps, NumOpsAdded, B); default: break; } @@ -2950,20 +3422,32 @@ static bool DisassembleMiscFrm(MCInst &MI, unsigned Opcode, uint32_t insn, // opcodes which match the same real instruction. This is needed since there's // no current handling of optional arguments. Fix here when a better handling // of optional arguments is implemented. - if (Opcode == ARM::CPS3p) { + if (Opcode == ARM::CPS3p) { // M = 1 + // Let's reject these impossible imod values by returning false: + // 1. (imod=0b01) + // + // AsmPrinter cannot handle imod=0b00, plus (imod=0b00,M=1,iflags!=0) is an + // invalid combination, so we just check for imod=0b00 here. + if (slice(insn, 19, 18) == 0 || slice(insn, 19, 18) == 1) + return false; MI.addOperand(MCOperand::CreateImm(slice(insn, 19, 18))); // imod MI.addOperand(MCOperand::CreateImm(slice(insn, 8, 6))); // iflags MI.addOperand(MCOperand::CreateImm(slice(insn, 4, 0))); // mode NumOpsAdded = 3; return true; } - if (Opcode == ARM::CPS2p) { + if (Opcode == ARM::CPS2p) { // mode = 0, M = 0 + // Let's reject these impossible imod values by returning false: + // 1. (imod=0b00,M=0) + // 2. (imod=0b01) + if (slice(insn, 19, 18) == 0 || slice(insn, 19, 18) == 1) + return false; MI.addOperand(MCOperand::CreateImm(slice(insn, 19, 18))); // imod MI.addOperand(MCOperand::CreateImm(slice(insn, 8, 6))); // iflags NumOpsAdded = 2; return true; } - if (Opcode == ARM::CPS1p) { + if (Opcode == ARM::CPS1p) { // imod = 0, iflags = 0, M = 1 MI.addOperand(MCOperand::CreateImm(slice(insn, 4, 0))); // mode NumOpsAdded = 1; return true; @@ -3142,7 +3626,7 @@ bool ARMBasicMCBuilder::DoPredicateOperands(MCInst& MI, unsigned Opcode, return false; } - + /// TryPredicateAndSBitModifier - TryPredicateAndSBitModifier tries to process /// the possible Predicate and SBitModifier, to build the remaining MCOperand /// constituents. @@ -3154,6 +3638,7 @@ bool ARMBasicMCBuilder::TryPredicateAndSBitModifier(MCInst& MI, unsigned Opcode, const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; const std::string &Name = ARMInsts[Opcode].Name; unsigned Idx = MI.getNumOperands(); + uint64_t TSFlags = ARMInsts[Opcode].TSFlags; // First, we check whether this instr specifies the PredicateOperand through // a pair of TargetOperandInfos with isPredicate() property. @@ -3173,14 +3658,23 @@ bool ARMBasicMCBuilder::TryPredicateAndSBitModifier(MCInst& MI, unsigned Opcode, // like ARM. // // A8.6.16 B - if (Name == "t2Bcc") - MI.addOperand(MCOperand::CreateImm(CondCode(slice(insn, 25, 22)))); - else if (Name == "tBcc") - MI.addOperand(MCOperand::CreateImm(CondCode(slice(insn, 11, 8)))); - else + // Check for undefined encodings. + unsigned cond; + if (Name == "t2Bcc") { + if ((cond = slice(insn, 25, 22)) >= 14) + return false; + MI.addOperand(MCOperand::CreateImm(CondCode(cond))); + } else if (Name == "tBcc") { + if ((cond = slice(insn, 11, 8)) == 14) + return false; + MI.addOperand(MCOperand::CreateImm(CondCode(cond))); + } else MI.addOperand(MCOperand::CreateImm(ARMCC::AL)); } else { // ARM instructions get their condition field from Inst{31-28}. + // We should reject Inst{31-28} = 0b1111 as invalid encoding. + if (!isNEONDomain(TSFlags) && getCondField(insn) == 0xF) + return false; MI.addOperand(MCOperand::CreateImm(CondCode(getCondField(insn)))); } } @@ -3243,3 +3737,84 @@ ARMBasicMCBuilder *llvm::CreateMCBuilder(unsigned Opcode, ARMFormat Format) { return new ARMBasicMCBuilder(Opcode, Format, ARMInsts[Opcode].getNumOperands()); } + +/// tryAddingSymbolicOperand - tryAddingSymbolicOperand trys to add a symbolic +/// operand in place of the immediate Value in the MCInst. The immediate +/// Value has had any PC adjustment made by the caller. If the getOpInfo() +/// function was set as part of the setupBuilderForSymbolicDisassembly() call +/// then that function is called to get any symbolic information at the +/// builder's Address for this instrution. If that returns non-zero then the +/// symbolic information it returns is used to create an MCExpr and that is +/// added as an operand to the MCInst. This function returns true if it adds +/// an operand to the MCInst and false otherwise. +bool ARMBasicMCBuilder::tryAddingSymbolicOperand(uint64_t Value, + uint64_t InstSize, + MCInst &MI) { + if (!GetOpInfo) + return false; + + struct LLVMOpInfo1 SymbolicOp; + SymbolicOp.Value = Value; + if (!GetOpInfo(DisInfo, Address, 0 /* Offset */, InstSize, 1, &SymbolicOp)) + return false; + + const MCExpr *Add = NULL; + if (SymbolicOp.AddSymbol.Present) { + if (SymbolicOp.AddSymbol.Name) { + StringRef Name(SymbolicOp.AddSymbol.Name); + MCSymbol *Sym = Ctx->GetOrCreateSymbol(Name); + Add = MCSymbolRefExpr::Create(Sym, *Ctx); + } else { + Add = MCConstantExpr::Create(SymbolicOp.AddSymbol.Value, *Ctx); + } + } + + const MCExpr *Sub = NULL; + if (SymbolicOp.SubtractSymbol.Present) { + if (SymbolicOp.SubtractSymbol.Name) { + StringRef Name(SymbolicOp.SubtractSymbol.Name); + MCSymbol *Sym = Ctx->GetOrCreateSymbol(Name); + Sub = MCSymbolRefExpr::Create(Sym, *Ctx); + } else { + Sub = MCConstantExpr::Create(SymbolicOp.SubtractSymbol.Value, *Ctx); + } + } + + const MCExpr *Off = NULL; + if (SymbolicOp.Value != 0) + Off = MCConstantExpr::Create(SymbolicOp.Value, *Ctx); + + const MCExpr *Expr; + if (Sub) { + const MCExpr *LHS; + if (Add) + LHS = MCBinaryExpr::CreateSub(Add, Sub, *Ctx); + else + LHS = MCUnaryExpr::CreateMinus(Sub, *Ctx); + if (Off != 0) + Expr = MCBinaryExpr::CreateAdd(LHS, Off, *Ctx); + else + Expr = LHS; + } else if (Add) { + if (Off != 0) + Expr = MCBinaryExpr::CreateAdd(Add, Off, *Ctx); + else + Expr = Add; + } else { + if (Off != 0) + Expr = Off; + else + Expr = MCConstantExpr::Create(0, *Ctx); + } + + if (SymbolicOp.VariantKind == LLVMDisassembler_VariantKind_ARM_HI16) + MI.addOperand(MCOperand::CreateExpr(ARMMCExpr::CreateUpper16(Expr, *Ctx))); + else if (SymbolicOp.VariantKind == LLVMDisassembler_VariantKind_ARM_LO16) + MI.addOperand(MCOperand::CreateExpr(ARMMCExpr::CreateLower16(Expr, *Ctx))); + else if (SymbolicOp.VariantKind == LLVMDisassembler_VariantKind_None) + MI.addOperand(MCOperand::CreateExpr(Expr)); + else + assert("bad SymbolicOp.VariantKind"); + + return true; +} diff --git a/lib/Target/ARM/Disassembler/ARMDisassemblerCore.h b/lib/Target/ARM/Disassembler/ARMDisassemblerCore.h index 9c30d33..a7ba141 100644 --- a/lib/Target/ARM/Disassembler/ARMDisassemblerCore.h +++ b/lib/Target/ARM/Disassembler/ARMDisassemblerCore.h @@ -22,12 +22,17 @@ #define ARMDISASSEMBLERCORE_H #include "llvm/MC/MCInst.h" +#include "llvm/MC/MCSymbol.h" +#include "llvm/MC/MCExpr.h" +#include "llvm/MC/MCContext.h" #include "llvm/Target/TargetInstrInfo.h" +#include "llvm-c/Disassembler.h" #include "ARMBaseInstrInfo.h" #include "ARMRegisterInfo.h" #include "ARMDisassembler.h" namespace llvm { +class MCContext; class ARMUtils { public: @@ -134,6 +139,31 @@ static inline void setSlice(unsigned &Bits, unsigned From, unsigned To, Bits |= (Val & Mask) << To; } +// Return an integer result equal to the number of bits of x that are ones. +static inline uint32_t +BitCount (uint64_t x) +{ + // c accumulates the total bits set in x + uint32_t c; + for (c = 0; x; ++c) + { + x &= x - 1; // clear the least significant bit set + } + return c; +} + +static inline bool +BitIsSet (const uint64_t value, const uint64_t bit) +{ + return (value & (1ull << bit)) != 0; +} + +static inline bool +BitIsClear (const uint64_t value, const uint64_t bit) +{ + return (value & (1ull << bit)) == 0; +} + /// Various utilities for checking the target specific flags. /// A unary data processing instruction doesn't have an Rn operand. @@ -141,6 +171,12 @@ static inline bool isUnaryDP(uint64_t TSFlags) { return (TSFlags & ARMII::UnaryDP); } +/// A NEON Domain instruction has cond field (Inst{31-28}) as 0b1111. +static inline bool isNEONDomain(uint64_t TSFlags) { + return (TSFlags & ARMII::DomainNEON) || + (TSFlags & ARMII::DomainNEONA8); +} + /// This four-bit field describes the addressing mode used. /// See also ARMBaseInstrInfo.h. static inline unsigned getAddrMode(uint64_t TSFlags) { @@ -196,7 +232,7 @@ private: public: ARMBasicMCBuilder(ARMBasicMCBuilder &B) : Opcode(B.Opcode), Format(B.Format), NumOps(B.NumOps), Disasm(B.Disasm), - SP(B.SP) { + SP(B.SP), GetOpInfo(0), DisInfo(0), Ctx(0) { Err = 0; } @@ -255,6 +291,44 @@ private: assert(SP); return slice(SP->ITState, 7, 4); } + +private: + // + // Hooks for symbolic disassembly via the public 'C' interface. + // + // The function to get the symbolic information for operands. + LLVMOpInfoCallback GetOpInfo; + // The pointer to the block of symbolic information for above call back. + void *DisInfo; + // The assembly context for creating symbols and MCExprs in place of + // immediate operands when there is symbolic information. + MCContext *Ctx; + // The address of the instruction being disassembled. + uint64_t Address; + +public: + void setupBuilderForSymbolicDisassembly(LLVMOpInfoCallback getOpInfo, + void *disInfo, MCContext *ctx, + uint64_t address) { + GetOpInfo = getOpInfo; + DisInfo = disInfo; + Ctx = ctx; + Address = address; + } + + uint64_t getBuilderAddress() const { return Address; } + + /// tryAddingSymbolicOperand - tryAddingSymbolicOperand trys to add a symbolic + /// operand in place of the immediate Value in the MCInst. The immediate + /// Value has had any PC adjustment made by the caller. If the getOpInfo() + /// function was set as part of the setupBuilderForSymbolicDisassembly() call + /// then that function is called to get any symbolic information at the + /// builder's Address for this instrution. If that returns non-zero then the + /// symbolic information it returns is used to create an MCExpr and that is + /// added as an operand to the MCInst. This function returns true if it adds + /// an operand to the MCInst and false otherwise. + bool tryAddingSymbolicOperand(uint64_t Value, uint64_t InstSize, MCInst &MI); + }; } // namespace llvm diff --git a/lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h b/lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h index 23372e0..8d39982 100644 --- a/lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h +++ b/lib/Target/ARM/Disassembler/ThumbDisassemblerCore.h @@ -108,6 +108,8 @@ static inline bool IsGPR(unsigned RegClass) { // Utilities for 32-bit Thumb instructions. +static inline bool BadReg(uint32_t n) { return n == 13 || n == 15; } + // Extract imm4: Inst{19-16}. static inline unsigned getImm4(uint32_t insn) { return slice(insn, 19, 16); @@ -398,9 +400,17 @@ static bool DisassembleThumb1General(MCInst &MI, unsigned Opcode, uint32_t insn, assert(OpInfo[OpIdx].RegClass < 0 && !OpInfo[OpIdx].isPredicate() && !OpInfo[OpIdx].isOptionalDef() && "Pure imm operand expected"); - MI.addOperand(MCOperand::CreateImm(UseRt ? getT1Imm8(insn) - : (Imm3 ? getT1Imm3(insn) - : getT1Imm5(insn)))); + unsigned Imm = 0; + if (UseRt) + Imm = getT1Imm8(insn); + else if (Imm3) + Imm = getT1Imm3(insn); + else { + Imm = getT1Imm5(insn); + ARM_AM::ShiftOpc ShOp = getShiftOpcForBits(slice(insn, 12, 11)); + getImmShiftSE(ShOp, Imm); + } + MI.addOperand(MCOperand::CreateImm(Imm)); } ++OpIdx; @@ -469,6 +479,7 @@ static bool DisassembleThumb1DP(MCInst &MI, unsigned Opcode, uint32_t insn, // tBX_RET: 0 operand // tBX_RET_vararg: Rm // tBLXr_r9: Rm +// tBRIND: Rm static bool DisassembleThumb1Special(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { @@ -476,11 +487,17 @@ static bool DisassembleThumb1Special(MCInst &MI, unsigned Opcode, uint32_t insn, if (NumOps == 0) return true; - // BX/BLX has 1 reg operand: Rm. - if (NumOps == 1) { + // BX/BLX/tBRIND (indirect branch, i.e, mov pc, Rm) has 1 reg operand: Rm. + if (Opcode==ARM::tBLXr_r9 || Opcode==ARM::tBX_Rm || Opcode==ARM::tBRIND) { + if (Opcode != ARM::tBRIND) { + // Handling the two predicate operands before the reg operand. + if (!B->DoPredicateOperands(MI, Opcode, insn, NumOps)) + return false; + NumOpsAdded += 2; + } MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, getT1Rm(insn)))); - NumOpsAdded = 1; + NumOpsAdded += 1; return true; } @@ -598,7 +615,7 @@ static bool DisassembleThumb2Ldpci(MCInst &MI, unsigned Opcode, // A6.2.4 Load/store single data item // -// Load/Store Register (reg|imm): tRd tRn imm5 tRm +// Load/Store Register (reg|imm): tRd tRn imm5|tRm // Load Register Signed Byte|Halfword: tRd tRn tRm static bool DisassembleThumb1LdSt(unsigned opA, MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { @@ -607,11 +624,6 @@ static bool DisassembleThumb1LdSt(unsigned opA, MCInst &MI, unsigned Opcode, const TargetOperandInfo *OpInfo = TID.OpInfo; unsigned &OpIdx = NumOpsAdded; - // Table A6-5 16-bit Thumb Load/store instructions - // opA = 0b0101 for STR/LDR (register) and friends. - // Otherwise, we have STR/LDR (immediate) and friends. - bool Imm5 = (opA != 5); - assert(NumOps >= 2 && OpInfo[0].RegClass == ARM::tGPRRegClassID && OpInfo[1].RegClass == ARM::tGPRRegClassID @@ -624,28 +636,28 @@ static bool DisassembleThumb1LdSt(unsigned opA, MCInst &MI, unsigned Opcode, getT1tRn(insn)))); OpIdx = 2; - // We have either { imm5, tRm } or { tRm } remaining. - // Process the imm5 first. Note that STR/LDR (register) should skip the imm5 - // offset operand for t_addrmode_s[1|2|4]. + // We have either { imm5 } or { tRm } remaining. + // Note that STR/LDR (register) should skip the imm5 offset operand for + // t_addrmode_s[1|2|4]. assert(OpIdx < NumOps && "More operands expected"); if (OpInfo[OpIdx].RegClass < 0 && !OpInfo[OpIdx].isPredicate() && !OpInfo[OpIdx].isOptionalDef()) { - - MI.addOperand(MCOperand::CreateImm(Imm5 ? getT1Imm5(insn) : 0)); + // Table A6-5 16-bit Thumb Load/store instructions + // opA = 0b0101 for STR/LDR (register) and friends. + // Otherwise, we have STR/LDR (immediate) and friends. + assert(opA != 5 && "Immediate operand expected for this opcode"); + MI.addOperand(MCOperand::CreateImm(getT1Imm5(insn))); + ++OpIdx; + } else { + // The next reg operand is tRm, the offset. + assert(OpIdx < NumOps && OpInfo[OpIdx].RegClass == ARM::tGPRRegClassID + && "Thumb reg operand expected"); + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::tGPRRegClassID, + getT1tRm(insn)))); ++OpIdx; } - - // The next reg operand is tRm, the offset. - assert(OpIdx < NumOps && OpInfo[OpIdx].RegClass == ARM::tGPRRegClassID - && "Thumb reg operand expected"); - MI.addOperand(MCOperand::CreateReg( - Imm5 ? 0 - : getRegisterEnum(B, ARM::tGPRRegClassID, - getT1tRm(insn)))); - ++OpIdx; - return true; } @@ -895,6 +907,10 @@ static bool DisassembleThumb1LdStMul(bool Ld, MCInst &MI, unsigned Opcode, } unsigned RegListBits = slice(insn, 7, 0); + if (BitCount(RegListBits) < 1) { + DEBUG(errs() << "if BitCount(registers) < 1 then UNPREDICTABLE\n"); + return false; + } // Fill the variadic part of reglist. for (unsigned i = 0; i < 8; ++i) @@ -945,6 +961,11 @@ static bool DisassembleThumb1CondBr(MCInst &MI, unsigned Opcode, uint32_t insn, : (int)Imm8)); // Predicate operands by ARMBasicMCBuilder::TryPredicateAndSBitModifier(). + // But note that for tBcc, if cond = '1110' then UNDEFINED. + if (Opcode == ARM::tBcc && slice(insn, 11, 8) == 14) { + DEBUG(errs() << "if cond = '1110' then UNDEFINED\n"); + return false; + } NumOpsAdded = 1; return true; @@ -965,11 +986,7 @@ static bool DisassembleThumb1Br(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned Imm11 = getT1Imm11(insn); - // When executing a Thumb instruction, PC reads as the address of the current - // instruction plus 4. The assembler subtracts 4 from the difference between - // the branch instruction and the target address, disassembler has to add 4 to - // to compensate. - MI.addOperand(MCOperand::CreateImm(SignExtend32<12>(Imm11 << 1) + 4)); + MI.addOperand(MCOperand::CreateImm(SignExtend32<12>(Imm11 << 1))); NumOpsAdded = 1; @@ -1129,8 +1146,12 @@ static bool DisassembleThumb2SRS(MCInst &MI, unsigned Opcode, uint32_t insn, // t2RFE[IA|DB]W/t2RFE[IA|DB]: Rn static bool DisassembleThumb2RFE(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, - decodeRn(insn)))); + unsigned Rn = decodeRn(insn); + if (Rn == 15) { + DEBUG(errs() << "if n == 15 then UNPREDICTABLE\n"); + return false; + } + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B,ARM::GPRRegClassID,Rn))); NumOpsAdded = 1; return true; } @@ -1149,7 +1170,7 @@ static bool DisassembleThumb2LdStMul(MCInst &MI, unsigned Opcode, uint32_t insn, Opcode == ARM::t2STMIA || Opcode == ARM::t2STMIA_UPD || Opcode == ARM::t2STMDB || Opcode == ARM::t2STMDB_UPD) && "Unexpected opcode"); - assert(NumOps >= 5 && "Thumb2 LdStMul expects NumOps >= 5"); + assert(NumOps >= 4 && "Thumb2 LdStMul expects NumOps >= 4"); NumOpsAdded = 0; @@ -1203,45 +1224,79 @@ static bool DisassembleThumb2LdStEx(MCInst &MI, unsigned Opcode, uint32_t insn, OpIdx = 0; assert(NumOps >= 2 - && OpInfo[0].RegClass == ARM::GPRRegClassID - && OpInfo[1].RegClass == ARM::GPRRegClassID + && OpInfo[0].RegClass > 0 + && OpInfo[1].RegClass > 0 && "Expect >=2 operands and first two as reg operands"); bool isStore = (ARM::t2STREX <= Opcode && Opcode <= ARM::t2STREXH); bool isSW = (Opcode == ARM::t2LDREX || Opcode == ARM::t2STREX); bool isDW = (Opcode == ARM::t2LDREXD || Opcode == ARM::t2STREXD); + unsigned Rt = decodeRd(insn); + unsigned Rt2 = decodeRs(insn); // But note that this is Rd for t2STREX. + unsigned Rd = decodeRm(insn); + unsigned Rn = decodeRn(insn); + + // Some sanity checking first. + if (isStore) { + // if d == n || d == t then UNPREDICTABLE + // if d == n || d == t || d == t2 then UNPREDICTABLE + if (isDW) { + if (Rd == Rn || Rd == Rt || Rd == Rt2) { + DEBUG(errs() << "if d == n || d == t || d == t2 then UNPREDICTABLE\n"); + return false; + } + } else { + if (isSW) { + if (Rt2 == Rn || Rt2 == Rt) { + DEBUG(errs() << "if d == n || d == t then UNPREDICTABLE\n"); + return false; + } + } else { + if (Rd == Rn || Rd == Rt) { + DEBUG(errs() << "if d == n || d == t then UNPREDICTABLE\n"); + return false; + } + } + } + } else { + // Load + // A8.6.71 LDREXD + // if t == t2 then UNPREDICTABLE + if (isDW && Rt == Rt2) { + DEBUG(errs() << "if t == t2 then UNPREDICTABLE\n"); + return false; + } + } + // Add the destination operand for store. if (isStore) { MI.addOperand(MCOperand::CreateReg( - getRegisterEnum(B, ARM::GPRRegClassID, - isSW ? decodeRs(insn) : decodeRm(insn)))); + getRegisterEnum(B, OpInfo[OpIdx].RegClass, + isSW ? Rt2 : Rd))); ++OpIdx; } // Source operand for store and destination operand for load. - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, - decodeRd(insn)))); + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, OpInfo[OpIdx].RegClass, + Rt))); ++OpIdx; // Thumb2 doubleword complication: with an extra source/destination operand. if (isDW) { - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, - decodeRs(insn)))); + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B,OpInfo[OpIdx].RegClass, + Rt2))); ++OpIdx; } // Finally add the pointer operand. - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, - decodeRn(insn)))); + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, OpInfo[OpIdx].RegClass, + Rn))); ++OpIdx; return true; } -// LLVM, as of Jan-05-2010, does not output <Rt2>, i.e., Rs, in the asm. -// Whereas the ARM Arch. Manual does not require that t2 = t+1 like in ARM ISA. -// // t2LDRDi8: Rd Rs Rn imm8s4 (offset mode) // t2LDRDpci: Rd Rs imm8s4 (Not decoded, prefer the generic t2LDRDi8 version) // t2STRDi8: Rd Rs Rn imm8s4 (offset mode) @@ -1255,18 +1310,50 @@ static bool DisassembleThumb2LdStDual(MCInst &MI, unsigned Opcode, if (!OpInfo) return false; assert(NumOps >= 4 - && OpInfo[0].RegClass == ARM::GPRRegClassID - && OpInfo[1].RegClass == ARM::GPRRegClassID - && OpInfo[2].RegClass == ARM::GPRRegClassID + && OpInfo[0].RegClass > 0 + && OpInfo[0].RegClass == OpInfo[1].RegClass + && OpInfo[2].RegClass > 0 && OpInfo[3].RegClass < 0 && "Expect >= 4 operands and first 3 as reg operands"); + // Thumnb allows for specifying Rt and Rt2, unlike ARM (which has Rt2==Rt+1). + unsigned Rt = decodeRd(insn); + unsigned Rt2 = decodeRs(insn); + unsigned Rn = decodeRn(insn); + + // Some sanity checking first. + + // A8.6.67 LDRD (literal) has its W bit as (0). + if (Opcode == ARM::t2LDRDi8 || Opcode == ARM::t2LDRD_PRE || Opcode == ARM::t2LDRD_POST) { + if (Rn == 15 && slice(insn, 21, 21) != 0) + return false; + } else { + // For Dual Store, PC cannot be used as the base register. + if (Rn == 15) { + DEBUG(errs() << "if n == 15 then UNPREDICTABLE\n"); + return false; + } + } + if (Rt == Rt2) { + DEBUG(errs() << "if t == t2 then UNPREDICTABLE\n"); + return false; + } + if (Opcode != ARM::t2LDRDi8 && Opcode != ARM::t2STRDi8) { + if (Rn == Rt || Rn == Rt2) { + DEBUG(errs() << "if wback && (n == t || n == t2) then UNPREDICTABLE\n"); + return false; + } + } + // Add the <Rt> <Rt2> operands. - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, + unsigned RegClassPair = OpInfo[0].RegClass; + unsigned RegClassBase = OpInfo[2].RegClass; + + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, RegClassPair, decodeRd(insn)))); - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, RegClassPair, decodeRs(insn)))); - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, RegClassBase, decodeRn(insn)))); // Finally add (+/-)imm8*4, depending on the U bit. @@ -1394,9 +1481,12 @@ static bool DisassembleThumb2DPSoReg(MCInst &MI, unsigned Opcode, uint32_t insn, if (OpInfo[OpIdx].RegClass < 0 && !OpInfo[OpIdx].isPredicate() && !OpInfo[OpIdx].isOptionalDef()) { - if (Thumb2ShiftOpcode(Opcode)) - MI.addOperand(MCOperand::CreateImm(getShiftAmtBits(insn))); - else { + if (Thumb2ShiftOpcode(Opcode)) { + unsigned Imm = getShiftAmtBits(insn); + ARM_AM::ShiftOpc ShOp = getShiftOpcForBits(slice(insn, 5, 4)); + getImmShiftSE(ShOp, Imm); + MI.addOperand(MCOperand::CreateImm(Imm)); + } else { // Build the constant shift specifier operand. unsigned bits2 = getShiftTypeBits(insn); unsigned imm5 = getShiftAmtBits(insn); @@ -1421,7 +1511,8 @@ static bool DisassembleThumb2DPSoReg(MCInst &MI, unsigned Opcode, uint32_t insn, static bool DisassembleThumb2DPModImm(MCInst &MI, unsigned Opcode, uint32_t insn, unsigned short NumOps, unsigned &NumOpsAdded, BO B) { - const TargetOperandInfo *OpInfo = ARMInsts[Opcode].OpInfo; + const TargetInstrDesc &TID = ARMInsts[Opcode]; + const TargetOperandInfo *OpInfo = TID.OpInfo; unsigned &OpIdx = NumOpsAdded; OpIdx = 0; @@ -1448,8 +1539,15 @@ static bool DisassembleThumb2DPModImm(MCInst &MI, unsigned Opcode, DEBUG(errs()<<"Thumb2 encoding error: d==15 for DPModImm 2-reg instr.\n"); return false; } - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, RnRegClassID, - decodeRn(insn)))); + int Idx; + if ((Idx = TID.getOperandConstraint(OpIdx, TOI::TIED_TO)) != -1) { + // The reg operand is tied to the first reg operand. + MI.addOperand(MI.getOperand(Idx)); + } else { + // Add second reg operand. + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, RnRegClassID, + decodeRn(insn)))); + } ++OpIdx; } @@ -1518,7 +1616,7 @@ static bool DisassembleThumb2Sat(MCInst &MI, unsigned Opcode, uint32_t insn, // o t2ADDri12, t2SUBri12: Rs Rn imm12 // o t2LEApcrel (ADR): Rs imm12 // o t2BFC (BFC): Rs Ro(TIED_TO) bf_inv_mask_imm -// o t2BFI (BFI) (Currently not defined in LLVM as of Jan-07-2010) +// o t2BFI (BFI): Rs Ro(TIED_TO) Rn bf_inv_mask_imm // o t2MOVi16: Rs imm16 // o t2MOVTi16: Rs imm16 // o t2SBFX (SBFX): Rs Rn lsb width @@ -1579,9 +1677,10 @@ static bool DisassembleThumb2DPBinImm(MCInst &MI, unsigned Opcode, if (Opcode == ARM::t2ADDri12 || Opcode == ARM::t2SUBri12 || Opcode == ARM::t2LEApcrel) MI.addOperand(MCOperand::CreateImm(getIImm3Imm8(insn))); - else if (Opcode == ARM::t2MOVi16 || Opcode == ARM::t2MOVTi16) - MI.addOperand(MCOperand::CreateImm(getImm16(insn))); - else if (Opcode == ARM::t2BFC || Opcode == ARM::t2BFI) { + else if (Opcode == ARM::t2MOVi16 || Opcode == ARM::t2MOVTi16) { + if (!B->tryAddingSymbolicOperand(getImm16(insn), 4, MI)) + MI.addOperand(MCOperand::CreateImm(getImm16(insn))); + } else if (Opcode == ARM::t2BFC || Opcode == ARM::t2BFI) { uint32_t mask = 0; if (getBitfieldInvMask(insn, mask)) MI.addOperand(MCOperand::CreateImm(mask)); @@ -1625,8 +1724,7 @@ static inline bool t2MiscCtrlInstr(uint32_t insn) { // A8.6.26 // t2BXJ -> Rn // -// Miscellaneous control: t2DMBsy (and its t2DMB variants), -// t2DSBsy (and its t2DSB varianst), t2ISBsy, t2CLREX +// Miscellaneous control: // -> no operand (except pred-imm pred-ccr for CLREX, memory barrier variants) // // Hint: t2NOP, t2YIELD, t2WFE, t2WFI, t2SEV @@ -1643,6 +1741,22 @@ static bool DisassembleThumb2BrMiscCtrl(MCInst &MI, unsigned Opcode, if (NumOps == 0) return true; + if (Opcode == ARM::t2DMB || Opcode == ARM::t2DSB) { + // Inst{3-0} encodes the memory barrier option for the variants. + unsigned opt = slice(insn, 3, 0); + switch (opt) { + case ARM_MB::SY: case ARM_MB::ST: + case ARM_MB::ISH: case ARM_MB::ISHST: + case ARM_MB::NSH: case ARM_MB::NSHST: + case ARM_MB::OSH: case ARM_MB::OSHST: + MI.addOperand(MCOperand::CreateImm(opt)); + NumOpsAdded = 1; + return true; + default: + return false; + } + } + if (t2MiscCtrlInstr(insn)) return true; @@ -1719,6 +1833,17 @@ static bool DisassembleThumb2BrMiscCtrl(MCInst &MI, unsigned Opcode, return true; } + // Some instructions have predicate operands first before the immediate. + if (Opcode == ARM::tBLXi_r9 || Opcode == ARM::tBLr9) { + // Handling the two predicate operands before the imm operand. + if (B->DoPredicateOperands(MI, Opcode, insn, NumOps)) + NumOpsAdded += 2; + else { + DEBUG(errs() << "Expected predicate operands not found.\n"); + return false; + } + } + // Add the imm operand. int Offset = 0; @@ -1739,13 +1864,12 @@ static bool DisassembleThumb2BrMiscCtrl(MCInst &MI, unsigned Opcode, Offset = decodeImm32_BLX(insn); break; } - // When executing a Thumb instruction, PC reads as the address of the current - // instruction plus 4. The assembler subtracts 4 from the difference between - // the branch instruction and the target address, disassembler has to add 4 to - // to compensate. - MI.addOperand(MCOperand::CreateImm(Offset + 4)); - NumOpsAdded = 1; + if (!B->tryAddingSymbolicOperand(Offset + B->getBuilderAddress() + 4, 4, MI)) + MI.addOperand(MCOperand::CreateImm(Offset)); + + // This is an increment as some predicate operands may have been added first. + NumOpsAdded += 1; return true; } @@ -1787,7 +1911,7 @@ static bool DisassembleThumb2PreLoad(MCInst &MI, unsigned Opcode, uint32_t insn, decodeRn(insn)))); ++OpIdx; - if (OpInfo[OpIdx].RegClass == ARM::GPRRegClassID) { + if (OpInfo[OpIdx].RegClass == ARM::rGPRRegClassID) { MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, decodeRm(insn)))); } else { @@ -1795,17 +1919,17 @@ static bool DisassembleThumb2PreLoad(MCInst &MI, unsigned Opcode, uint32_t insn, && !OpInfo[OpIdx].isOptionalDef() && "Pure imm operand expected"); int Offset = 0; - if (slice(insn, 19, 16) == 0xFF) { - bool Negative = slice(insn, 23, 23) == 0; - unsigned Imm12 = getImm12(insn); - Offset = Negative ? -1 - Imm12 : 1 * Imm12; - } else if (Opcode == ARM::t2PLDi8 || Opcode == ARM::t2PLDWi8 || - Opcode == ARM::t2PLIi8) { + if (Opcode == ARM::t2PLDi8 || Opcode == ARM::t2PLDWi8 || + Opcode == ARM::t2PLIi8) { // A8.6.117 Encoding T2: add = FALSE unsigned Imm8 = getImm8(insn); - Offset = -1 - Imm8; - } else // The i12 forms. See, for example, A8.6.117 Encoding T1. + Offset = -1 * Imm8; + } else { + // The i12 forms. See, for example, A8.6.117 Encoding T1. + // Note that currently t2PLDi12 also handles the previously named t2PLDpci + // opcode, that's why we use decodeImm12(insn) which returns +/- imm12. Offset = decodeImm12(insn); + } MI.addOperand(MCOperand::CreateImm(Offset)); } ++OpIdx; @@ -1820,6 +1944,87 @@ static bool DisassembleThumb2PreLoad(MCInst &MI, unsigned Opcode, uint32_t insn, return true; } +static bool BadRegsThumb2LdSt(unsigned Opcode, uint32_t insn, bool Load, + unsigned R0, unsigned R1, unsigned R2, bool UseRm, bool WB) { + + // Inst{22-21} encodes the data item transferred for load/store. + // For single word, it is encoded as ob10. + bool Word = (slice(insn, 22, 21) == 2); + bool Half = (slice(insn, 22, 21) == 1); + bool Byte = (slice(insn, 22, 21) == 0); + + if (UseRm && BadReg(R2)) { + DEBUG(errs() << "if BadReg(m) then UNPREDICTABLE\n"); + return true; + } + + if (Load) { + if (!Word && R0 == 13) { + DEBUG(errs() << "if t == 13 then UNPREDICTABLE\n"); + return true; + } + if (Byte) { + if (WB && R0 == 15 && slice(insn, 10, 8) == 3) { + // A8.6.78 LDRSB (immediate) Encoding T2 (errata markup 8.0) + DEBUG(errs() << "if t == 15 && PUW == '011' then UNPREDICTABLE\n"); + return true; + } + } + // A6.3.8 Load halfword, memory hints + if (Half) { + if (WB) { + if (R0 == R1) { + // A8.6.82 LDRSH (immediate) Encoding T2 + DEBUG(errs() << "if WB && n == t then UNPREDICTABLE\n"); + return true; + } + if (R0 == 15 && slice(insn, 10, 8) == 3) { + // A8.6.82 LDRSH (immediate) Encoding T2 (errata markup 8.0) + DEBUG(errs() << "if t == 15 && PUW == '011' then UNPREDICTABLE\n"); + return true; + } + } else { + if (Opcode == ARM::t2LDRHi8 || Opcode == ARM::t2LDRSHi8) { + if (R0 == 15 && slice(insn, 10, 8) == 4) { + // A8.6.82 LDRSH (immediate) Encoding T2 + DEBUG(errs() << "if Rt == '1111' and PUW == '100' then SEE" + << " \"Unallocated memory hints\"\n"); + return true; + } + } else { + if (R0 == 15) { + // A8.6.82 LDRSH (immediate) Encoding T1 + DEBUG(errs() << "if Rt == '1111' then SEE" + << " \"Unallocated memory hints\"\n"); + return true; + } + } + } + } + } else { + if (WB && R0 == R1) { + DEBUG(errs() << "if wback && n == t then UNPREDICTABLE\n"); + return true; + } + if ((WB && R0 == 15) || (!WB && R1 == 15)) { + DEBUG(errs() << "if Rn == '1111' then UNDEFINED\n"); + return true; + } + if (Word) { + if ((WB && R1 == 15) || (!WB && R0 == 15)) { + DEBUG(errs() << "if t == 15 then UNPREDICTABLE\n"); + return true; + } + } else { + if ((WB && BadReg(R1)) || (!WB && BadReg(R0))) { + DEBUG(errs() << "if BadReg(t) then UNPREDICTABLE\n"); + return true; + } + } + } + return false; +} + // A6.3.10 Store single data item // A6.3.9 Load byte, memory hints // A6.3.8 Load halfword, memory hints @@ -1865,16 +2070,16 @@ static bool DisassembleThumb2LdSt(bool Load, MCInst &MI, unsigned Opcode, OpIdx = 0; assert(NumOps >= 3 && - OpInfo[0].RegClass == ARM::GPRRegClassID && - OpInfo[1].RegClass == ARM::GPRRegClassID && + OpInfo[0].RegClass > 0 && + OpInfo[1].RegClass > 0 && "Expect >= 3 operands and first two as reg operands"); - bool ThreeReg = (OpInfo[2].RegClass == ARM::GPRRegClassID); + bool ThreeReg = (OpInfo[2].RegClass > 0); bool TIED_TO = ThreeReg && TID.getOperandConstraint(2, TOI::TIED_TO) != -1; bool Imm12 = !ThreeReg && slice(insn, 23, 23) == 1; // ARMInstrThumb2.td // Build the register operands, followed by the immediate. - unsigned R0, R1, R2 = 0; + unsigned R0 = 0, R1 = 0, R2 = 0; unsigned Rd = decodeRd(insn); int Imm = 0; @@ -1905,19 +2110,24 @@ static bool DisassembleThumb2LdSt(bool Load, MCInst &MI, unsigned Opcode, Imm = decodeImm8(insn); } - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, OpInfo[OpIdx].RegClass, R0))); ++OpIdx; - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, OpInfo[OpIdx].RegClass, R1))); ++OpIdx; if (ThreeReg) { - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::GPRRegClassID, + // This could be an offset register or a TIED_TO register. + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B,OpInfo[OpIdx].RegClass, R2))); ++OpIdx; } + if (BadRegsThumb2LdSt(Opcode, insn, Load, R0, R1, R2, ThreeReg & !TIED_TO, + TIED_TO)) + return false; + assert(OpInfo[OpIdx].RegClass < 0 && !OpInfo[OpIdx].isPredicate() && !OpInfo[OpIdx].isOptionalDef() && "Pure imm operand expected"); @@ -1947,25 +2157,25 @@ static bool DisassembleThumb2DPReg(MCInst &MI, unsigned Opcode, uint32_t insn, OpIdx = 0; assert(NumOps >= 2 && - OpInfo[0].RegClass == ARM::rGPRRegClassID && - OpInfo[1].RegClass == ARM::rGPRRegClassID && + OpInfo[0].RegClass > 0 && + OpInfo[1].RegClass > 0 && "Expect >= 2 operands and first two as reg operands"); // Build the register operands, followed by the optional rotation amount. - bool ThreeReg = NumOps > 2 && OpInfo[2].RegClass == ARM::rGPRRegClassID; + bool ThreeReg = NumOps > 2 && OpInfo[2].RegClass > 0; - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::rGPRRegClassID, + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, OpInfo[OpIdx].RegClass, decodeRs(insn)))); ++OpIdx; if (ThreeReg) { - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::rGPRRegClassID, + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B,OpInfo[OpIdx].RegClass, decodeRn(insn)))); ++OpIdx; } - MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, ARM::rGPRRegClassID, + MI.addOperand(MCOperand::CreateReg(getRegisterEnum(B, OpInfo[OpIdx].RegClass, decodeRm(insn)))); ++OpIdx; |
