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 "MCTargetDesc/WebAssemblyInstPrinter.h"
18 #include "MCTargetDesc/WebAssemblyMCTargetDesc.h"
19 #include "TargetInfo/WebAssemblyTargetInfo.h"
20 #include "llvm/MC/MCContext.h"
21 #include "llvm/MC/MCDisassembler/MCDisassembler.h"
22 #include "llvm/MC/MCFixedLenDisassembler.h"
23 #include "llvm/MC/MCInst.h"
24 #include "llvm/MC/MCInstrInfo.h"
25 #include "llvm/MC/MCSubtargetInfo.h"
26 #include "llvm/MC/MCSymbol.h"
27 #include "llvm/MC/MCSymbolWasm.h"
28 #include "llvm/Support/Endian.h"
29 #include "llvm/Support/LEB128.h"
30 #include "llvm/Support/TargetRegistry.h"
31 
32 using namespace llvm;
33 
34 #define DEBUG_TYPE "wasm-disassembler"
35 
36 using DecodeStatus = MCDisassembler::DecodeStatus;
37 
38 #include "WebAssemblyGenDisassemblerTables.inc"
39 
40 namespace {
41 static constexpr int WebAssemblyInstructionTableSize = 256;
42 
43 class WebAssemblyDisassembler final : public MCDisassembler {
44   std::unique_ptr<const MCInstrInfo> MCII;
45 
46   DecodeStatus getInstruction(MCInst &Instr, uint64_t &Size,
47                               ArrayRef<uint8_t> Bytes, uint64_t Address,
48                               raw_ostream &CStream) const override;
49   Optional<DecodeStatus> onSymbolStart(SymbolInfoTy &Symbol, uint64_t &Size,
50                                        ArrayRef<uint8_t> Bytes,
51                                        uint64_t Address,
52                                        raw_ostream &CStream) const override;
53 
54 public:
55   WebAssemblyDisassembler(const MCSubtargetInfo &STI, MCContext &Ctx,
56                           std::unique_ptr<const MCInstrInfo> MCII)
57       : MCDisassembler(STI, Ctx), MCII(std::move(MCII)) {}
58 };
59 } // end anonymous namespace
60 
61 static MCDisassembler *createWebAssemblyDisassembler(const Target &T,
62                                                      const MCSubtargetInfo &STI,
63                                                      MCContext &Ctx) {
64   std::unique_ptr<const MCInstrInfo> MCII(T.createMCInstrInfo());
65   return new WebAssemblyDisassembler(STI, Ctx, std::move(MCII));
66 }
67 
68 extern "C" LLVM_EXTERNAL_VISIBILITY void
69 LLVMInitializeWebAssemblyDisassembler() {
70   // Register the disassembler for each target.
71   TargetRegistry::RegisterMCDisassembler(getTheWebAssemblyTarget32(),
72                                          createWebAssemblyDisassembler);
73   TargetRegistry::RegisterMCDisassembler(getTheWebAssemblyTarget64(),
74                                          createWebAssemblyDisassembler);
75 }
76 
77 static int nextByte(ArrayRef<uint8_t> Bytes, uint64_t &Size) {
78   if (Size >= Bytes.size())
79     return -1;
80   auto V = Bytes[Size];
81   Size++;
82   return V;
83 }
84 
85 static bool nextLEB(int64_t &Val, ArrayRef<uint8_t> Bytes, uint64_t &Size,
86                     bool Signed) {
87   unsigned N = 0;
88   const char *Error = nullptr;
89   Val = Signed ? decodeSLEB128(Bytes.data() + Size, &N,
90                                Bytes.data() + Bytes.size(), &Error)
91                : static_cast<int64_t>(decodeULEB128(Bytes.data() + Size, &N,
92                                                     Bytes.data() + Bytes.size(),
93                                                     &Error));
94   if (Error)
95     return false;
96   Size += N;
97   return true;
98 }
99 
100 static bool parseLEBImmediate(MCInst &MI, uint64_t &Size,
101                               ArrayRef<uint8_t> Bytes, bool Signed) {
102   int64_t Val;
103   if (!nextLEB(Val, Bytes, Size, Signed))
104     return false;
105   MI.addOperand(MCOperand::createImm(Val));
106   return true;
107 }
108 
109 template <typename T>
110 bool parseImmediate(MCInst &MI, uint64_t &Size, ArrayRef<uint8_t> Bytes) {
111   if (Size + sizeof(T) > Bytes.size())
112     return false;
113   T Val = support::endian::read<T, support::endianness::little, 1>(
114       Bytes.data() + Size);
115   Size += sizeof(T);
116   if (std::is_floating_point<T>::value) {
117     MI.addOperand(
118         MCOperand::createDFPImm(bit_cast<uint64_t>(static_cast<double>(Val))));
119   } else {
120     MI.addOperand(MCOperand::createImm(static_cast<int64_t>(Val)));
121   }
122   return true;
123 }
124 
125 Optional<MCDisassembler::DecodeStatus> WebAssemblyDisassembler::onSymbolStart(
126     SymbolInfoTy &Symbol, uint64_t &Size, ArrayRef<uint8_t> Bytes,
127     uint64_t Address, raw_ostream &CStream) const {
128   Size = 0;
129   if (Address == 0) {
130     // Start of a code section: we're parsing only the function count.
131     int64_t FunctionCount;
132     if (!nextLEB(FunctionCount, Bytes, Size, false))
133       return None;
134     outs() << "        # " << FunctionCount << " functions in section.";
135   } else {
136     // Parse the start of a single function.
137     int64_t BodySize, LocalEntryCount;
138     if (!nextLEB(BodySize, Bytes, Size, false) ||
139         !nextLEB(LocalEntryCount, Bytes, Size, false))
140       return None;
141     if (LocalEntryCount) {
142       outs() << "        .local ";
143       for (int64_t I = 0; I < LocalEntryCount; I++) {
144         int64_t Count, Type;
145         if (!nextLEB(Count, Bytes, Size, false) ||
146             !nextLEB(Type, Bytes, Size, false))
147           return None;
148         for (int64_t J = 0; J < Count; J++) {
149           if (I || J)
150             outs() << ", ";
151           outs() << WebAssembly::anyTypeToString(Type);
152         }
153       }
154     }
155   }
156   outs() << "\n";
157   return MCDisassembler::Success;
158 }
159 
160 MCDisassembler::DecodeStatus WebAssemblyDisassembler::getInstruction(
161     MCInst &MI, uint64_t &Size, ArrayRef<uint8_t> Bytes, uint64_t /*Address*/,
162     raw_ostream &CS) const {
163   CommentStream = &CS;
164   Size = 0;
165   int Opc = nextByte(Bytes, Size);
166   if (Opc < 0)
167     return MCDisassembler::Fail;
168   const auto *WasmInst = &InstructionTable0[Opc];
169   // If this is a prefix byte, indirect to another table.
170   if (WasmInst->ET == ET_Prefix) {
171     WasmInst = nullptr;
172     // Linear search, so far only 2 entries.
173     for (auto PT = PrefixTable; PT->Table; PT++) {
174       if (PT->Prefix == Opc) {
175         WasmInst = PT->Table;
176         break;
177       }
178     }
179     if (!WasmInst)
180       return MCDisassembler::Fail;
181     int64_t PrefixedOpc;
182     if (!nextLEB(PrefixedOpc, Bytes, Size, false))
183       return MCDisassembler::Fail;
184     if (PrefixedOpc < 0 || PrefixedOpc >= WebAssemblyInstructionTableSize)
185       return MCDisassembler::Fail;
186     WasmInst += PrefixedOpc;
187   }
188   if (WasmInst->ET == ET_Unused)
189     return MCDisassembler::Fail;
190   // At this point we must have a valid instruction to decode.
191   assert(WasmInst->ET == ET_Instruction);
192   MI.setOpcode(WasmInst->Opcode);
193   // Parse any operands.
194   for (uint8_t OPI = 0; OPI < WasmInst->NumOperands; OPI++) {
195     auto OT = OperandTable[WasmInst->OperandStart + OPI];
196     switch (OT) {
197     // ULEB operands:
198     case WebAssembly::OPERAND_BASIC_BLOCK:
199     case WebAssembly::OPERAND_LOCAL:
200     case WebAssembly::OPERAND_GLOBAL:
201     case WebAssembly::OPERAND_FUNCTION32:
202     case WebAssembly::OPERAND_TABLE:
203     case WebAssembly::OPERAND_OFFSET32:
204     case WebAssembly::OPERAND_OFFSET64:
205     case WebAssembly::OPERAND_P2ALIGN:
206     case WebAssembly::OPERAND_TYPEINDEX:
207     case WebAssembly::OPERAND_EVENT:
208     case MCOI::OPERAND_IMMEDIATE: {
209       if (!parseLEBImmediate(MI, Size, Bytes, false))
210         return MCDisassembler::Fail;
211       break;
212     }
213     // SLEB operands:
214     case WebAssembly::OPERAND_I32IMM:
215     case WebAssembly::OPERAND_I64IMM: {
216       if (!parseLEBImmediate(MI, Size, Bytes, true))
217         return MCDisassembler::Fail;
218       break;
219     }
220     // block_type operands:
221     case WebAssembly::OPERAND_SIGNATURE: {
222       int64_t Val;
223       uint64_t PrevSize = Size;
224       if (!nextLEB(Val, Bytes, Size, true))
225         return MCDisassembler::Fail;
226       if (Val < 0) {
227         // Negative values are single septet value types or empty types
228         if (Size != PrevSize + 1) {
229           MI.addOperand(
230               MCOperand::createImm(int64_t(WebAssembly::BlockType::Invalid)));
231         } else {
232           MI.addOperand(MCOperand::createImm(Val & 0x7f));
233         }
234       } else {
235         // We don't have access to the signature, so create a symbol without one
236         MCSymbol *Sym = getContext().createTempSymbol("typeindex", true);
237         auto *WasmSym = cast<MCSymbolWasm>(Sym);
238         WasmSym->setType(wasm::WASM_SYMBOL_TYPE_FUNCTION);
239         const MCExpr *Expr = MCSymbolRefExpr::create(
240             WasmSym, MCSymbolRefExpr::VK_WASM_TYPEINDEX, getContext());
241         MI.addOperand(MCOperand::createExpr(Expr));
242       }
243       break;
244     }
245     // heap_type operands, for e.g. ref.null:
246     case WebAssembly::OPERAND_HEAPTYPE: {
247       int64_t Val;
248       uint64_t PrevSize = Size;
249       if (!nextLEB(Val, Bytes, Size, true))
250         return MCDisassembler::Fail;
251       if (Val < 0 && Size == PrevSize + 1) {
252         // The HeapType encoding is like BlockType, in that encodings that
253         // decode as negative values indicate ValTypes.  In practice we expect
254         // either wasm::ValType::EXTERNREF or wasm::ValType::FUNCREF here.
255         //
256         // The positive SLEB values are reserved for future expansion and are
257         // expected to be type indices in the typed function references
258         // proposal, and should disassemble as MCSymbolRefExpr as in BlockType
259         // above.
260         MI.addOperand(MCOperand::createImm(Val & 0x7f));
261       } else {
262         MI.addOperand(
263             MCOperand::createImm(int64_t(WebAssembly::HeapType::Invalid)));
264       }
265       break;
266     }
267     // FP operands.
268     case WebAssembly::OPERAND_F32IMM: {
269       if (!parseImmediate<float>(MI, Size, Bytes))
270         return MCDisassembler::Fail;
271       break;
272     }
273     case WebAssembly::OPERAND_F64IMM: {
274       if (!parseImmediate<double>(MI, Size, Bytes))
275         return MCDisassembler::Fail;
276       break;
277     }
278     // Vector lane operands (not LEB encoded).
279     case WebAssembly::OPERAND_VEC_I8IMM: {
280       if (!parseImmediate<uint8_t>(MI, Size, Bytes))
281         return MCDisassembler::Fail;
282       break;
283     }
284     case WebAssembly::OPERAND_VEC_I16IMM: {
285       if (!parseImmediate<uint16_t>(MI, Size, Bytes))
286         return MCDisassembler::Fail;
287       break;
288     }
289     case WebAssembly::OPERAND_VEC_I32IMM: {
290       if (!parseImmediate<uint32_t>(MI, Size, Bytes))
291         return MCDisassembler::Fail;
292       break;
293     }
294     case WebAssembly::OPERAND_VEC_I64IMM: {
295       if (!parseImmediate<uint64_t>(MI, Size, Bytes))
296         return MCDisassembler::Fail;
297       break;
298     }
299     case WebAssembly::OPERAND_BRLIST: {
300       int64_t TargetTableLen;
301       if (!nextLEB(TargetTableLen, Bytes, Size, false))
302         return MCDisassembler::Fail;
303       for (int64_t I = 0; I < TargetTableLen; I++) {
304         if (!parseLEBImmediate(MI, Size, Bytes, false))
305           return MCDisassembler::Fail;
306       }
307       // Default case.
308       if (!parseLEBImmediate(MI, Size, Bytes, false))
309         return MCDisassembler::Fail;
310       break;
311     }
312     case MCOI::OPERAND_REGISTER:
313       // The tablegen header currently does not have any register operands since
314       // we use only the stack (_S) instructions.
315       // If you hit this that probably means a bad instruction definition in
316       // tablegen.
317       llvm_unreachable("Register operand in WebAssemblyDisassembler");
318     default:
319       llvm_unreachable("Unknown operand type in WebAssemblyDisassembler");
320     }
321   }
322   return MCDisassembler::Success;
323 }
324