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 "TargetInfo/WebAssemblyTargetInfo.h"
18 #include "Utils/WebAssemblyTypeUtilities.h"
19 #include "llvm/MC/MCContext.h"
20 #include "llvm/MC/MCDecoderOps.h"
21 #include "llvm/MC/MCDisassembler/MCDisassembler.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/MC/MCSymbolWasm.h"
27 #include "llvm/MC/TargetRegistry.h"
28 #include "llvm/Support/Endian.h"
29 #include "llvm/Support/LEB128.h"
30 
31 using namespace llvm;
32 
33 #define DEBUG_TYPE "wasm-disassembler"
34 
35 using DecodeStatus = MCDisassembler::DecodeStatus;
36 
37 #include "WebAssemblyGenDisassemblerTables.inc"
38 
39 namespace {
40 static constexpr int WebAssemblyInstructionTableSize = 256;
41 
42 class WebAssemblyDisassembler final : public MCDisassembler {
43   std::unique_ptr<const MCInstrInfo> MCII;
44 
45   DecodeStatus getInstruction(MCInst &Instr, uint64_t &Size,
46                               ArrayRef<uint8_t> Bytes, uint64_t Address,
47                               raw_ostream &CStream) const override;
48   Optional<DecodeStatus> onSymbolStart(SymbolInfoTy &Symbol, uint64_t &Size,
49                                        ArrayRef<uint8_t> Bytes,
50                                        uint64_t Address,
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" LLVM_EXTERNAL_VISIBILITY void
68 LLVMInitializeWebAssemblyDisassembler() {
69   // Register the disassembler for each target.
70   TargetRegistry::RegisterMCDisassembler(getTheWebAssemblyTarget32(),
71                                          createWebAssemblyDisassembler);
72   TargetRegistry::RegisterMCDisassembler(getTheWebAssemblyTarget64(),
73                                          createWebAssemblyDisassembler);
74 }
75 
76 static int nextByte(ArrayRef<uint8_t> Bytes, uint64_t &Size) {
77   if (Size >= Bytes.size())
78     return -1;
79   auto V = Bytes[Size];
80   Size++;
81   return V;
82 }
83 
84 static bool nextLEB(int64_t &Val, ArrayRef<uint8_t> Bytes, uint64_t &Size,
85                     bool Signed) {
86   unsigned N = 0;
87   const char *Error = nullptr;
88   Val = Signed ? decodeSLEB128(Bytes.data() + Size, &N,
89                                Bytes.data() + Bytes.size(), &Error)
90                : static_cast<int64_t>(decodeULEB128(Bytes.data() + Size, &N,
91                                                     Bytes.data() + Bytes.size(),
92                                                     &Error));
93   if (Error)
94     return false;
95   Size += N;
96   return true;
97 }
98 
99 static bool parseLEBImmediate(MCInst &MI, uint64_t &Size,
100                               ArrayRef<uint8_t> Bytes, bool Signed) {
101   int64_t Val;
102   if (!nextLEB(Val, Bytes, Size, Signed))
103     return false;
104   MI.addOperand(MCOperand::createImm(Val));
105   return true;
106 }
107 
108 template <typename T>
109 bool parseImmediate(MCInst &MI, uint64_t &Size, ArrayRef<uint8_t> Bytes) {
110   if (Size + sizeof(T) > Bytes.size())
111     return false;
112   T Val = support::endian::read<T, support::endianness::little, 1>(
113       Bytes.data() + Size);
114   Size += sizeof(T);
115   if (std::is_floating_point<T>::value) {
116     MI.addOperand(
117         MCOperand::createDFPImm(bit_cast<uint64_t>(static_cast<double>(Val))));
118   } else {
119     MI.addOperand(MCOperand::createImm(static_cast<int64_t>(Val)));
120   }
121   return true;
122 }
123 
124 Optional<MCDisassembler::DecodeStatus> WebAssemblyDisassembler::onSymbolStart(
125     SymbolInfoTy &Symbol, uint64_t &Size, ArrayRef<uint8_t> Bytes,
126     uint64_t Address, raw_ostream &CStream) const {
127   Size = 0;
128   if (Address == 0) {
129     // Start of a code section: we're parsing only the function count.
130     int64_t FunctionCount;
131     if (!nextLEB(FunctionCount, Bytes, Size, false))
132       return None;
133     outs() << "        # " << FunctionCount << " functions in section.";
134   } else {
135     // Parse the start of a single function.
136     int64_t BodySize, LocalEntryCount;
137     if (!nextLEB(BodySize, Bytes, Size, false) ||
138         !nextLEB(LocalEntryCount, Bytes, Size, false))
139       return None;
140     if (LocalEntryCount) {
141       outs() << "        .local ";
142       for (int64_t I = 0; I < LocalEntryCount; I++) {
143         int64_t Count, Type;
144         if (!nextLEB(Count, Bytes, Size, false) ||
145             !nextLEB(Type, Bytes, Size, false))
146           return None;
147         for (int64_t J = 0; J < Count; J++) {
148           if (I || J)
149             outs() << ", ";
150           outs() << WebAssembly::anyTypeToString(Type);
151         }
152       }
153     }
154   }
155   outs() << "\n";
156   return MCDisassembler::Success;
157 }
158 
159 MCDisassembler::DecodeStatus WebAssemblyDisassembler::getInstruction(
160     MCInst &MI, uint64_t &Size, ArrayRef<uint8_t> Bytes, uint64_t /*Address*/,
161     raw_ostream &CS) const {
162   CommentStream = &CS;
163   Size = 0;
164   int Opc = nextByte(Bytes, Size);
165   if (Opc < 0)
166     return MCDisassembler::Fail;
167   const auto *WasmInst = &InstructionTable0[Opc];
168   // If this is a prefix byte, indirect to another table.
169   if (WasmInst->ET == ET_Prefix) {
170     WasmInst = nullptr;
171     // Linear search, so far only 2 entries.
172     for (auto PT = PrefixTable; PT->Table; PT++) {
173       if (PT->Prefix == Opc) {
174         WasmInst = PT->Table;
175         break;
176       }
177     }
178     if (!WasmInst)
179       return MCDisassembler::Fail;
180     int64_t PrefixedOpc;
181     if (!nextLEB(PrefixedOpc, Bytes, Size, false))
182       return MCDisassembler::Fail;
183     if (PrefixedOpc < 0 || PrefixedOpc >= WebAssemblyInstructionTableSize)
184       return MCDisassembler::Fail;
185     WasmInst += PrefixedOpc;
186   }
187   if (WasmInst->ET == ET_Unused)
188     return MCDisassembler::Fail;
189   // At this point we must have a valid instruction to decode.
190   assert(WasmInst->ET == ET_Instruction);
191   MI.setOpcode(WasmInst->Opcode);
192   // Parse any operands.
193   for (uint8_t OPI = 0; OPI < WasmInst->NumOperands; OPI++) {
194     auto OT = OperandTable[WasmInst->OperandStart + OPI];
195     switch (OT) {
196     // ULEB operands:
197     case WebAssembly::OPERAND_BASIC_BLOCK:
198     case WebAssembly::OPERAND_LOCAL:
199     case WebAssembly::OPERAND_GLOBAL:
200     case WebAssembly::OPERAND_FUNCTION32:
201     case WebAssembly::OPERAND_TABLE:
202     case WebAssembly::OPERAND_OFFSET32:
203     case WebAssembly::OPERAND_OFFSET64:
204     case WebAssembly::OPERAND_P2ALIGN:
205     case WebAssembly::OPERAND_TYPEINDEX:
206     case WebAssembly::OPERAND_TAG:
207     case MCOI::OPERAND_IMMEDIATE: {
208       if (!parseLEBImmediate(MI, Size, Bytes, false))
209         return MCDisassembler::Fail;
210       break;
211     }
212     // SLEB operands:
213     case WebAssembly::OPERAND_I32IMM:
214     case WebAssembly::OPERAND_I64IMM: {
215       if (!parseLEBImmediate(MI, Size, Bytes, true))
216         return MCDisassembler::Fail;
217       break;
218     }
219     // block_type operands:
220     case WebAssembly::OPERAND_SIGNATURE: {
221       int64_t Val;
222       uint64_t PrevSize = Size;
223       if (!nextLEB(Val, Bytes, Size, true))
224         return MCDisassembler::Fail;
225       if (Val < 0) {
226         // Negative values are single septet value types or empty types
227         if (Size != PrevSize + 1) {
228           MI.addOperand(
229               MCOperand::createImm(int64_t(WebAssembly::BlockType::Invalid)));
230         } else {
231           MI.addOperand(MCOperand::createImm(Val & 0x7f));
232         }
233       } else {
234         // We don't have access to the signature, so create a symbol without one
235         MCSymbol *Sym = getContext().createTempSymbol("typeindex", true);
236         auto *WasmSym = cast<MCSymbolWasm>(Sym);
237         WasmSym->setType(wasm::WASM_SYMBOL_TYPE_FUNCTION);
238         const MCExpr *Expr = MCSymbolRefExpr::create(
239             WasmSym, MCSymbolRefExpr::VK_WASM_TYPEINDEX, getContext());
240         MI.addOperand(MCOperand::createExpr(Expr));
241       }
242       break;
243     }
244     // FP operands.
245     case WebAssembly::OPERAND_F32IMM: {
246       if (!parseImmediate<float>(MI, Size, Bytes))
247         return MCDisassembler::Fail;
248       break;
249     }
250     case WebAssembly::OPERAND_F64IMM: {
251       if (!parseImmediate<double>(MI, Size, Bytes))
252         return MCDisassembler::Fail;
253       break;
254     }
255     // Vector lane operands (not LEB encoded).
256     case WebAssembly::OPERAND_VEC_I8IMM: {
257       if (!parseImmediate<uint8_t>(MI, Size, Bytes))
258         return MCDisassembler::Fail;
259       break;
260     }
261     case WebAssembly::OPERAND_VEC_I16IMM: {
262       if (!parseImmediate<uint16_t>(MI, Size, Bytes))
263         return MCDisassembler::Fail;
264       break;
265     }
266     case WebAssembly::OPERAND_VEC_I32IMM: {
267       if (!parseImmediate<uint32_t>(MI, Size, Bytes))
268         return MCDisassembler::Fail;
269       break;
270     }
271     case WebAssembly::OPERAND_VEC_I64IMM: {
272       if (!parseImmediate<uint64_t>(MI, Size, Bytes))
273         return MCDisassembler::Fail;
274       break;
275     }
276     case WebAssembly::OPERAND_BRLIST: {
277       int64_t TargetTableLen;
278       if (!nextLEB(TargetTableLen, Bytes, Size, false))
279         return MCDisassembler::Fail;
280       for (int64_t I = 0; I < TargetTableLen; I++) {
281         if (!parseLEBImmediate(MI, Size, Bytes, false))
282           return MCDisassembler::Fail;
283       }
284       // Default case.
285       if (!parseLEBImmediate(MI, Size, Bytes, false))
286         return MCDisassembler::Fail;
287       break;
288     }
289     case MCOI::OPERAND_REGISTER:
290       // The tablegen header currently does not have any register operands since
291       // we use only the stack (_S) instructions.
292       // If you hit this that probably means a bad instruction definition in
293       // tablegen.
294       llvm_unreachable("Register operand in WebAssemblyDisassembler");
295     default:
296       llvm_unreachable("Unknown operand type in WebAssemblyDisassembler");
297     }
298   }
299   return MCDisassembler::Success;
300 }
301