1 //===- llvm/MC/MCDisassembler.h - Disassembler interface --------*- C++ -*-===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 9 #ifndef LLVM_MC_MCDISASSEMBLER_MCDISASSEMBLER_H 10 #define LLVM_MC_MCDISASSEMBLER_MCDISASSEMBLER_H 11 12 #include "llvm/ADT/StringRef.h" 13 #include "llvm/BinaryFormat/XCOFF.h" 14 #include "llvm/MC/MCDisassembler/MCSymbolizer.h" 15 #include <cstdint> 16 #include <memory> 17 #include <vector> 18 19 namespace llvm { 20 21 struct XCOFFSymbolInfoTy { 22 std::optional<XCOFF::StorageMappingClass> StorageMappingClass; 23 std::optional<uint32_t> Index; 24 bool IsLabel = false; 25 bool operator<(const XCOFFSymbolInfoTy &SymInfo) const; 26 }; 27 28 struct SymbolInfoTy { 29 uint64_t Addr; 30 StringRef Name; 31 // XCOFF uses XCOFFSymInfo. Other targets use Type. 32 XCOFFSymbolInfoTy XCOFFSymInfo; 33 uint8_t Type; 34 35 private: 36 bool IsXCOFF; 37 bool HasType; 38 39 public: 40 SymbolInfoTy(uint64_t Addr, StringRef Name, 41 std::optional<XCOFF::StorageMappingClass> Smc, 42 std::optional<uint32_t> Idx, bool Label) 43 : Addr(Addr), Name(Name), XCOFFSymInfo{Smc, Idx, Label}, Type(0), 44 IsXCOFF(true), HasType(false) {} 45 SymbolInfoTy(uint64_t Addr, StringRef Name, uint8_t Type, 46 bool IsXCOFF = false) 47 : Addr(Addr), Name(Name), Type(Type), IsXCOFF(IsXCOFF), HasType(true) {} 48 bool isXCOFF() const { return IsXCOFF; } 49 50 private: 51 friend bool operator<(const SymbolInfoTy &P1, const SymbolInfoTy &P2) { 52 assert((P1.IsXCOFF == P2.IsXCOFF && P1.HasType == P2.HasType) && 53 "The value of IsXCOFF and HasType in P1 and P2 should be the same " 54 "respectively."); 55 56 if (P1.IsXCOFF && P1.HasType) 57 return std::tie(P1.Addr, P1.Type, P1.Name) < 58 std::tie(P2.Addr, P2.Type, P2.Name); 59 60 if (P1.IsXCOFF) 61 return std::tie(P1.Addr, P1.XCOFFSymInfo, P1.Name) < 62 std::tie(P2.Addr, P2.XCOFFSymInfo, P2.Name); 63 64 return std::tie(P1.Addr, P1.Name, P1.Type) < 65 std::tie(P2.Addr, P2.Name, P2.Type); 66 } 67 }; 68 69 using SectionSymbolsTy = std::vector<SymbolInfoTy>; 70 71 template <typename T> class ArrayRef; 72 class MCContext; 73 class MCInst; 74 class MCSubtargetInfo; 75 class raw_ostream; 76 77 /// Superclass for all disassemblers. Consumes a memory region and provides an 78 /// array of assembly instructions. 79 class MCDisassembler { 80 public: 81 /// Ternary decode status. Most backends will just use Fail and 82 /// Success, however some have a concept of an instruction with 83 /// understandable semantics but which is architecturally 84 /// incorrect. An example of this is ARM UNPREDICTABLE instructions 85 /// which are disassemblable but cause undefined behaviour. 86 /// 87 /// Because it makes sense to disassemble these instructions, there 88 /// is a "soft fail" failure mode that indicates the MCInst& is 89 /// valid but architecturally incorrect. 90 /// 91 /// The enum numbers are deliberately chosen such that reduction 92 /// from Success->SoftFail ->Fail can be done with a simple 93 /// bitwise-AND: 94 /// 95 /// LEFT & TOP = | Success Unpredictable Fail 96 /// --------------+----------------------------------- 97 /// Success | Success Unpredictable Fail 98 /// Unpredictable | Unpredictable Unpredictable Fail 99 /// Fail | Fail Fail Fail 100 /// 101 /// An easy way of encoding this is as 0b11, 0b01, 0b00 for 102 /// Success, SoftFail, Fail respectively. 103 enum DecodeStatus { 104 Fail = 0, 105 SoftFail = 1, 106 Success = 3 107 }; 108 109 MCDisassembler(const MCSubtargetInfo &STI, MCContext &Ctx) 110 : Ctx(Ctx), STI(STI) {} 111 112 virtual ~MCDisassembler(); 113 114 /// Returns the disassembly of a single instruction. 115 /// 116 /// \param Instr - An MCInst to populate with the contents of the 117 /// instruction. 118 /// \param Size - A value to populate with the size of the instruction, or 119 /// the number of bytes consumed while attempting to decode 120 /// an invalid instruction. 121 /// \param Address - The address, in the memory space of region, of the first 122 /// byte of the instruction. 123 /// \param Bytes - A reference to the actual bytes of the instruction. 124 /// \param CStream - The stream to print comments and annotations on. 125 /// \return - MCDisassembler::Success if the instruction is valid, 126 /// MCDisassembler::SoftFail if the instruction was 127 /// disassemblable but invalid, 128 /// MCDisassembler::Fail if the instruction was invalid. 129 virtual DecodeStatus getInstruction(MCInst &Instr, uint64_t &Size, 130 ArrayRef<uint8_t> Bytes, uint64_t Address, 131 raw_ostream &CStream) const = 0; 132 133 /// Used to perform separate target specific disassembly for a particular 134 /// symbol. May parse any prelude that precedes instructions after the 135 /// start of a symbol, or the entire symbol. 136 /// This is used for example by WebAssembly to decode preludes. 137 /// 138 /// Base implementation returns std::nullopt. So all targets by default ignore 139 /// to treat symbols separately. 140 /// 141 /// \param Symbol - The symbol. 142 /// \param Size - The number of bytes consumed. 143 /// \param Address - The address, in the memory space of region, of the first 144 /// byte of the symbol. 145 /// \param Bytes - A reference to the actual bytes at the symbol location. 146 /// \param CStream - The stream to print comments and annotations on. 147 /// \return - MCDisassembler::Success if bytes are decoded 148 /// successfully. Size must hold the number of bytes that 149 /// were decoded. 150 /// - MCDisassembler::Fail if the bytes are invalid. Size 151 /// must hold the number of bytes that were decoded before 152 /// failing. The target must print nothing. This can be 153 /// done by buffering the output if needed. 154 /// - std::nullopt if the target doesn't want to handle the 155 /// symbol separately. Value of Size is ignored in this 156 /// case. 157 virtual std::optional<DecodeStatus> 158 onSymbolStart(SymbolInfoTy &Symbol, uint64_t &Size, ArrayRef<uint8_t> Bytes, 159 uint64_t Address, raw_ostream &CStream) const; 160 // TODO: 161 // Implement similar hooks that can be used at other points during 162 // disassembly. Something along the following lines: 163 // - onBeforeInstructionDecode() 164 // - onAfterInstructionDecode() 165 // - onSymbolEnd() 166 // It should help move much of the target specific code from llvm-objdump to 167 // respective target disassemblers. 168 169 /// Suggest a distance to skip in a buffer of data to find the next 170 /// place to look for the start of an instruction. For example, if 171 /// all instructions have a fixed alignment, this might advance to 172 /// the next multiple of that alignment. 173 /// 174 /// If not overridden, the default is 1. 175 /// 176 /// \param Address - The address, in the memory space of region, of the 177 /// starting point (typically the first byte of something 178 /// that did not decode as a valid instruction at all). 179 /// \param Bytes - A reference to the actual bytes at Address. May be 180 /// needed in order to determine the width of an 181 /// unrecognized instruction (e.g. in Thumb this is a simple 182 /// consistent criterion that doesn't require knowing the 183 /// specific instruction). The caller can pass as much data 184 /// as they have available, and the function is required to 185 /// make a reasonable default choice if not enough data is 186 /// available to make a better one. 187 /// \return - A number of bytes to skip. Must always be greater than 188 /// zero. May be greater than the size of Bytes. 189 virtual uint64_t suggestBytesToSkip(ArrayRef<uint8_t> Bytes, 190 uint64_t Address) const; 191 192 private: 193 MCContext &Ctx; 194 195 protected: 196 // Subtarget information, for instruction decoding predicates if required. 197 const MCSubtargetInfo &STI; 198 std::unique_ptr<MCSymbolizer> Symbolizer; 199 200 public: 201 // Helpers around MCSymbolizer 202 bool tryAddingSymbolicOperand(MCInst &Inst, int64_t Value, uint64_t Address, 203 bool IsBranch, uint64_t Offset, uint64_t OpSize, 204 uint64_t InstSize) const; 205 206 void tryAddingPcLoadReferenceComment(int64_t Value, uint64_t Address) const; 207 208 /// Set \p Symzer as the current symbolizer. 209 /// This takes ownership of \p Symzer, and deletes the previously set one. 210 void setSymbolizer(std::unique_ptr<MCSymbolizer> Symzer); 211 212 MCContext& getContext() const { return Ctx; } 213 214 const MCSubtargetInfo& getSubtargetInfo() const { return STI; } 215 216 // Marked mutable because we cache it inside the disassembler, rather than 217 // having to pass it around as an argument through all the autogenerated code. 218 mutable raw_ostream *CommentStream = nullptr; 219 }; 220 221 } // end namespace llvm 222 223 #endif // LLVM_MC_MCDISASSEMBLER_MCDISASSEMBLER_H 224