1 //==- WebAssemblyDisassembler.cpp - Disassembler for WebAssembly -*- 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 /// \file 10 /// This file is part of the WebAssembly Disassembler. 11 /// 12 /// It contains code to translate the data produced by the decoder into 13 /// MCInsts. 14 /// 15 //===----------------------------------------------------------------------===// 16 17 #include "InstPrinter/WebAssemblyInstPrinter.h" 18 #include "MCTargetDesc/WebAssemblyMCTargetDesc.h" 19 #include "llvm/MC/MCContext.h" 20 #include "llvm/MC/MCDisassembler/MCDisassembler.h" 21 #include "llvm/MC/MCFixedLenDisassembler.h" 22 #include "llvm/MC/MCInst.h" 23 #include "llvm/MC/MCInstrInfo.h" 24 #include "llvm/MC/MCSubtargetInfo.h" 25 #include "llvm/MC/MCSymbol.h" 26 #include "llvm/Support/Endian.h" 27 #include "llvm/Support/LEB128.h" 28 #include "llvm/Support/TargetRegistry.h" 29 30 using namespace llvm; 31 32 #define DEBUG_TYPE "wasm-disassembler" 33 34 using DecodeStatus = MCDisassembler::DecodeStatus; 35 36 #include "WebAssemblyGenDisassemblerTables.inc" 37 38 namespace { 39 static constexpr int WebAssemblyInstructionTableSize = 256; 40 41 class WebAssemblyDisassembler final : public MCDisassembler { 42 std::unique_ptr<const MCInstrInfo> MCII; 43 44 DecodeStatus getInstruction(MCInst &Instr, uint64_t &Size, 45 ArrayRef<uint8_t> Bytes, uint64_t Address, 46 raw_ostream &VStream, 47 raw_ostream &CStream) const override; 48 DecodeStatus onSymbolStart(StringRef Name, uint64_t &Size, 49 ArrayRef<uint8_t> Bytes, uint64_t Address, 50 raw_ostream &VStream, 51 raw_ostream &CStream) const override; 52 53 public: 54 WebAssemblyDisassembler(const MCSubtargetInfo &STI, MCContext &Ctx, 55 std::unique_ptr<const MCInstrInfo> MCII) 56 : MCDisassembler(STI, Ctx), MCII(std::move(MCII)) {} 57 }; 58 } // end anonymous namespace 59 60 static MCDisassembler *createWebAssemblyDisassembler(const Target &T, 61 const MCSubtargetInfo &STI, 62 MCContext &Ctx) { 63 std::unique_ptr<const MCInstrInfo> MCII(T.createMCInstrInfo()); 64 return new WebAssemblyDisassembler(STI, Ctx, std::move(MCII)); 65 } 66 67 extern "C" void LLVMInitializeWebAssemblyDisassembler() { 68 // Register the disassembler for each target. 69 TargetRegistry::RegisterMCDisassembler(getTheWebAssemblyTarget32(), 70 createWebAssemblyDisassembler); 71 TargetRegistry::RegisterMCDisassembler(getTheWebAssemblyTarget64(), 72 createWebAssemblyDisassembler); 73 } 74 75 static int nextByte(ArrayRef<uint8_t> Bytes, uint64_t &Size) { 76 if (Size >= Bytes.size()) 77 return -1; 78 auto V = Bytes[Size]; 79 Size++; 80 return V; 81 } 82 83 static bool nextLEB(int64_t &Val, ArrayRef<uint8_t> Bytes, uint64_t &Size, 84 bool Signed) { 85 unsigned N = 0; 86 const char *Error = nullptr; 87 Val = Signed ? decodeSLEB128(Bytes.data() + Size, &N, 88 Bytes.data() + Bytes.size(), &Error) 89 : static_cast<int64_t>(decodeULEB128(Bytes.data() + Size, &N, 90 Bytes.data() + Bytes.size(), 91 &Error)); 92 if (Error) 93 return false; 94 Size += N; 95 return true; 96 } 97 98 static bool parseLEBImmediate(MCInst &MI, uint64_t &Size, 99 ArrayRef<uint8_t> Bytes, bool Signed) { 100 int64_t Val; 101 if (!nextLEB(Val, Bytes, Size, Signed)) 102 return false; 103 MI.addOperand(MCOperand::createImm(Val)); 104 return true; 105 } 106 107 template <typename T> 108 bool parseImmediate(MCInst &MI, uint64_t &Size, ArrayRef<uint8_t> Bytes) { 109 if (Size + sizeof(T) > Bytes.size()) 110 return false; 111 T Val; 112 memcpy(&Val, Bytes.data() + Size, sizeof(T)); 113 support::endian::byte_swap<T, support::endianness::little>(Val); 114 Size += sizeof(T); 115 if (std::is_floating_point<T>::value) { 116 MI.addOperand(MCOperand::createFPImm(static_cast<double>(Val))); 117 } else { 118 MI.addOperand(MCOperand::createImm(static_cast<int64_t>(Val))); 119 } 120 return true; 121 } 122 123 MCDisassembler::DecodeStatus WebAssemblyDisassembler::onSymbolStart( 124 StringRef Name, uint64_t &Size, ArrayRef<uint8_t> Bytes, uint64_t Address, 125 raw_ostream &VStream, raw_ostream &CStream) const { 126 Size = 0; 127 if (Address == 0) { 128 // Start of a code section: we're parsing only the function count. 129 int64_t FunctionCount; 130 if (!nextLEB(FunctionCount, Bytes, Size, false)) 131 return MCDisassembler::Fail; 132 outs() << " # " << FunctionCount << " functions in section."; 133 } else { 134 // Parse the start of a single function. 135 int64_t BodySize, LocalEntryCount; 136 if (!nextLEB(BodySize, Bytes, Size, false) || 137 !nextLEB(LocalEntryCount, Bytes, Size, false)) 138 return MCDisassembler::Fail; 139 if (LocalEntryCount) { 140 outs() << " .local "; 141 for (int64_t I = 0; I < LocalEntryCount; I++) { 142 int64_t Count, Type; 143 if (!nextLEB(Count, Bytes, Size, false) || 144 !nextLEB(Type, Bytes, Size, false)) 145 return MCDisassembler::Fail; 146 for (int64_t J = 0; J < Count; J++) { 147 if (I || J) 148 outs() << ", "; 149 outs() << WebAssembly::anyTypeToString(Type); 150 } 151 } 152 } 153 } 154 outs() << "\n"; 155 return MCDisassembler::Success; 156 } 157 158 MCDisassembler::DecodeStatus WebAssemblyDisassembler::getInstruction( 159 MCInst &MI, uint64_t &Size, ArrayRef<uint8_t> Bytes, uint64_t /*Address*/, 160 raw_ostream & /*OS*/, raw_ostream &CS) const { 161 CommentStream = &CS; 162 Size = 0; 163 int Opc = nextByte(Bytes, Size); 164 if (Opc < 0) 165 return MCDisassembler::Fail; 166 const auto *WasmInst = &InstructionTable0[Opc]; 167 // If this is a prefix byte, indirect to another table. 168 if (WasmInst->ET == ET_Prefix) { 169 WasmInst = nullptr; 170 // Linear search, so far only 2 entries. 171 for (auto PT = PrefixTable; PT->Table; PT++) { 172 if (PT->Prefix == Opc) { 173 WasmInst = PT->Table; 174 break; 175 } 176 } 177 if (!WasmInst) 178 return MCDisassembler::Fail; 179 int64_t PrefixedOpc; 180 if (!nextLEB(PrefixedOpc, Bytes, Size, false)) 181 return MCDisassembler::Fail; 182 if (PrefixedOpc < 0 || PrefixedOpc >= WebAssemblyInstructionTableSize) 183 return MCDisassembler::Fail; 184 WasmInst += PrefixedOpc; 185 } 186 if (WasmInst->ET == ET_Unused) 187 return MCDisassembler::Fail; 188 // At this point we must have a valid instruction to decode. 189 assert(WasmInst->ET == ET_Instruction); 190 MI.setOpcode(WasmInst->Opcode); 191 // Parse any operands. 192 for (uint8_t OPI = 0; OPI < WasmInst->NumOperands; OPI++) { 193 auto OT = OperandTable[WasmInst->OperandStart + OPI]; 194 switch (OT) { 195 // ULEB operands: 196 case WebAssembly::OPERAND_BASIC_BLOCK: 197 case WebAssembly::OPERAND_LOCAL: 198 case WebAssembly::OPERAND_GLOBAL: 199 case WebAssembly::OPERAND_FUNCTION32: 200 case WebAssembly::OPERAND_OFFSET32: 201 case WebAssembly::OPERAND_P2ALIGN: 202 case WebAssembly::OPERAND_TYPEINDEX: 203 case MCOI::OPERAND_IMMEDIATE: { 204 if (!parseLEBImmediate(MI, Size, Bytes, false)) 205 return MCDisassembler::Fail; 206 break; 207 } 208 // SLEB operands: 209 case WebAssembly::OPERAND_I32IMM: 210 case WebAssembly::OPERAND_I64IMM: { 211 if (!parseLEBImmediate(MI, Size, Bytes, true)) 212 return MCDisassembler::Fail; 213 break; 214 } 215 // block_type operands (uint8_t). 216 case WebAssembly::OPERAND_SIGNATURE: { 217 if (!parseImmediate<uint8_t>(MI, Size, Bytes)) 218 return MCDisassembler::Fail; 219 break; 220 } 221 // FP operands. 222 case WebAssembly::OPERAND_F32IMM: { 223 if (!parseImmediate<float>(MI, Size, Bytes)) 224 return MCDisassembler::Fail; 225 break; 226 } 227 case WebAssembly::OPERAND_F64IMM: { 228 if (!parseImmediate<double>(MI, Size, Bytes)) 229 return MCDisassembler::Fail; 230 break; 231 } 232 // Vector lane operands (not LEB encoded). 233 case WebAssembly::OPERAND_VEC_I8IMM: { 234 if (!parseImmediate<uint8_t>(MI, Size, Bytes)) 235 return MCDisassembler::Fail; 236 break; 237 } 238 case WebAssembly::OPERAND_VEC_I16IMM: { 239 if (!parseImmediate<uint16_t>(MI, Size, Bytes)) 240 return MCDisassembler::Fail; 241 break; 242 } 243 case WebAssembly::OPERAND_VEC_I32IMM: { 244 if (!parseImmediate<uint32_t>(MI, Size, Bytes)) 245 return MCDisassembler::Fail; 246 break; 247 } 248 case WebAssembly::OPERAND_VEC_I64IMM: { 249 if (!parseImmediate<uint64_t>(MI, Size, Bytes)) 250 return MCDisassembler::Fail; 251 break; 252 } 253 case WebAssembly::OPERAND_BRLIST: { 254 int64_t TargetTableLen; 255 if (!nextLEB(TargetTableLen, Bytes, Size, false)) 256 return MCDisassembler::Fail; 257 for (int64_t I = 0; I < TargetTableLen; I++) { 258 if (!parseLEBImmediate(MI, Size, Bytes, false)) 259 return MCDisassembler::Fail; 260 } 261 // Default case. 262 if (!parseLEBImmediate(MI, Size, Bytes, false)) 263 return MCDisassembler::Fail; 264 break; 265 } 266 case MCOI::OPERAND_REGISTER: 267 // The tablegen header currently does not have any register operands since 268 // we use only the stack (_S) instructions. 269 // If you hit this that probably means a bad instruction definition in 270 // tablegen. 271 llvm_unreachable("Register operand in WebAssemblyDisassembler"); 272 default: 273 llvm_unreachable("Unknown operand type in WebAssemblyDisassembler"); 274 } 275 } 276 return MCDisassembler::Success; 277 } 278