AArch64ExternalSymbolizer.cpp revision dce4a407a24b04eebc6a376f8e62b41aaa7b071f
1//===- AArch64ExternalSymbolizer.cpp - Symbolizer for AArch64 ---*- C++ -*-===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file is distributed under the University of Illinois Open Source 6// License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9 10#include "AArch64ExternalSymbolizer.h" 11#include "AArch64Subtarget.h" 12#include "MCTargetDesc/AArch64AddressingModes.h" 13#include "Utils/AArch64BaseInfo.h" 14#include "llvm/MC/MCContext.h" 15#include "llvm/MC/MCExpr.h" 16#include "llvm/MC/MCInst.h" 17#include "llvm/Support/Format.h" 18#include "llvm/Support/raw_ostream.h" 19 20using namespace llvm; 21 22#define DEBUG_TYPE "aarch64-disassembler" 23 24static MCSymbolRefExpr::VariantKind 25getVariant(uint64_t LLVMDisassembler_VariantKind) { 26 switch (LLVMDisassembler_VariantKind) { 27 case LLVMDisassembler_VariantKind_None: 28 return MCSymbolRefExpr::VK_None; 29 case LLVMDisassembler_VariantKind_ARM64_PAGE: 30 return MCSymbolRefExpr::VK_PAGE; 31 case LLVMDisassembler_VariantKind_ARM64_PAGEOFF: 32 return MCSymbolRefExpr::VK_PAGEOFF; 33 case LLVMDisassembler_VariantKind_ARM64_GOTPAGE: 34 return MCSymbolRefExpr::VK_GOTPAGE; 35 case LLVMDisassembler_VariantKind_ARM64_GOTPAGEOFF: 36 return MCSymbolRefExpr::VK_GOTPAGEOFF; 37 case LLVMDisassembler_VariantKind_ARM64_TLVP: 38 case LLVMDisassembler_VariantKind_ARM64_TLVOFF: 39 default: 40 assert(0 && "bad LLVMDisassembler_VariantKind"); 41 return MCSymbolRefExpr::VK_None; 42 } 43} 44 45/// tryAddingSymbolicOperand - tryAddingSymbolicOperand trys to add a symbolic 46/// operand in place of the immediate Value in the MCInst. The immediate 47/// Value has not had any PC adjustment made by the caller. If the instruction 48/// is a branch that adds the PC to the immediate Value then isBranch is 49/// Success, else Fail. If GetOpInfo is non-null, then it is called to get any 50/// symbolic information at the Address for this instrution. If that returns 51/// non-zero then the symbolic information it returns is used to create an 52/// MCExpr and that is added as an operand to the MCInst. If GetOpInfo() 53/// returns zero and isBranch is Success then a symbol look up for 54/// Address + Value is done and if a symbol is found an MCExpr is created with 55/// that, else an MCExpr with Address + Value is created. If GetOpInfo() 56/// returns zero and isBranch is Fail then the the Opcode of the MCInst is 57/// tested and for ADRP an other instructions that help to load of pointers 58/// a symbol look up is done to see it is returns a specific reference type 59/// to add to the comment stream. This function returns Success if it adds 60/// an operand to the MCInst and Fail otherwise. 61bool AArch64ExternalSymbolizer::tryAddingSymbolicOperand( 62 MCInst &MI, raw_ostream &CommentStream, int64_t Value, uint64_t Address, 63 bool IsBranch, uint64_t Offset, uint64_t InstSize) { 64 // FIXME: This method shares a lot of code with 65 // MCExternalSymbolizer::tryAddingSymbolicOperand. It may be possible 66 // refactor the MCExternalSymbolizer interface to allow more of this 67 // implementation to be shared. 68 // 69 struct LLVMOpInfo1 SymbolicOp; 70 memset(&SymbolicOp, '\0', sizeof(struct LLVMOpInfo1)); 71 SymbolicOp.Value = Value; 72 uint64_t ReferenceType; 73 const char *ReferenceName; 74 if (!GetOpInfo || 75 !GetOpInfo(DisInfo, Address, 0 /* Offset */, InstSize, 1, &SymbolicOp)) { 76 if (IsBranch) { 77 ReferenceType = LLVMDisassembler_ReferenceType_In_Branch; 78 const char *Name = SymbolLookUp(DisInfo, Address + Value, &ReferenceType, 79 Address, &ReferenceName); 80 if (Name) { 81 SymbolicOp.AddSymbol.Name = Name; 82 SymbolicOp.AddSymbol.Present = true; 83 SymbolicOp.Value = 0; 84 } else { 85 SymbolicOp.Value = Address + Value; 86 } 87 if (ReferenceType == LLVMDisassembler_ReferenceType_Out_SymbolStub) 88 CommentStream << "symbol stub for: " << ReferenceName; 89 else if (ReferenceType == 90 LLVMDisassembler_ReferenceType_Out_Objc_Message) 91 CommentStream << "Objc message: " << ReferenceName; 92 } else if (MI.getOpcode() == AArch64::ADRP) { 93 ReferenceType = LLVMDisassembler_ReferenceType_In_ARM64_ADRP; 94 // otool expects the fully encoded ADRP instruction to be passed in as 95 // the value here, so reconstruct it: 96 const MCRegisterInfo &MCRI = *Ctx.getRegisterInfo(); 97 uint32_t EncodedInst = 0x90000000; 98 EncodedInst |= (Value & 0x3) << 29; // immlo 99 EncodedInst |= ((Value >> 2) & 0x7FFFF) << 5; // immhi 100 EncodedInst |= MCRI.getEncodingValue(MI.getOperand(0).getReg()); // reg 101 SymbolLookUp(DisInfo, EncodedInst, &ReferenceType, Address, 102 &ReferenceName); 103 CommentStream << format("0x%llx", 104 0xfffffffffffff000LL & (Address + Value)); 105 } else if (MI.getOpcode() == AArch64::ADDXri || 106 MI.getOpcode() == AArch64::LDRXui || 107 MI.getOpcode() == AArch64::LDRXl || 108 MI.getOpcode() == AArch64::ADR) { 109 if (MI.getOpcode() == AArch64::ADDXri) 110 ReferenceType = LLVMDisassembler_ReferenceType_In_ARM64_ADDXri; 111 else if (MI.getOpcode() == AArch64::LDRXui) 112 ReferenceType = LLVMDisassembler_ReferenceType_In_ARM64_LDRXui; 113 if (MI.getOpcode() == AArch64::LDRXl) { 114 ReferenceType = LLVMDisassembler_ReferenceType_In_ARM64_LDRXl; 115 SymbolLookUp(DisInfo, Address + Value, &ReferenceType, Address, 116 &ReferenceName); 117 } else if (MI.getOpcode() == AArch64::ADR) { 118 ReferenceType = LLVMDisassembler_ReferenceType_In_ARM64_ADR; 119 SymbolLookUp(DisInfo, Address + Value, &ReferenceType, Address, 120 &ReferenceName); 121 } else { 122 const MCRegisterInfo &MCRI = *Ctx.getRegisterInfo(); 123 // otool expects the fully encoded ADD/LDR instruction to be passed in 124 // as the value here, so reconstruct it: 125 unsigned EncodedInst = 126 MI.getOpcode() == AArch64::ADDXri ? 0x91000000: 0xF9400000; 127 EncodedInst |= Value << 10; // imm12 [+ shift:2 for ADD] 128 EncodedInst |= 129 MCRI.getEncodingValue(MI.getOperand(1).getReg()) << 5; // Rn 130 EncodedInst |= MCRI.getEncodingValue(MI.getOperand(0).getReg()); // Rd 131 132 SymbolLookUp(DisInfo, EncodedInst, &ReferenceType, Address, 133 &ReferenceName); 134 } 135 if (ReferenceType == LLVMDisassembler_ReferenceType_Out_LitPool_SymAddr) 136 CommentStream << "literal pool symbol address: " << ReferenceName; 137 else if (ReferenceType == 138 LLVMDisassembler_ReferenceType_Out_LitPool_CstrAddr) 139 CommentStream << "literal pool for: \"" << ReferenceName << "\""; 140 else if (ReferenceType == 141 LLVMDisassembler_ReferenceType_Out_Objc_CFString_Ref) 142 CommentStream << "Objc cfstring ref: @\"" << ReferenceName << "\""; 143 else if (ReferenceType == 144 LLVMDisassembler_ReferenceType_Out_Objc_Message) 145 CommentStream << "Objc message: " << ReferenceName; 146 else if (ReferenceType == 147 LLVMDisassembler_ReferenceType_Out_Objc_Message_Ref) 148 CommentStream << "Objc message ref: " << ReferenceName; 149 else if (ReferenceType == 150 LLVMDisassembler_ReferenceType_Out_Objc_Selector_Ref) 151 CommentStream << "Objc selector ref: " << ReferenceName; 152 else if (ReferenceType == 153 LLVMDisassembler_ReferenceType_Out_Objc_Class_Ref) 154 CommentStream << "Objc class ref: " << ReferenceName; 155 // For these instructions, the SymbolLookUp() above is just to get the 156 // ReferenceType and ReferenceName. We want to make sure not to 157 // fall through so we don't build an MCExpr to leave the disassembly 158 // of the immediate values of these instructions to the InstPrinter. 159 return false; 160 } else { 161 return false; 162 } 163 } 164 165 const MCExpr *Add = nullptr; 166 if (SymbolicOp.AddSymbol.Present) { 167 if (SymbolicOp.AddSymbol.Name) { 168 StringRef Name(SymbolicOp.AddSymbol.Name); 169 MCSymbol *Sym = Ctx.GetOrCreateSymbol(Name); 170 MCSymbolRefExpr::VariantKind Variant = getVariant(SymbolicOp.VariantKind); 171 if (Variant != MCSymbolRefExpr::VK_None) 172 Add = MCSymbolRefExpr::Create(Sym, Variant, Ctx); 173 else 174 Add = MCSymbolRefExpr::Create(Sym, Ctx); 175 } else { 176 Add = MCConstantExpr::Create(SymbolicOp.AddSymbol.Value, Ctx); 177 } 178 } 179 180 const MCExpr *Sub = nullptr; 181 if (SymbolicOp.SubtractSymbol.Present) { 182 if (SymbolicOp.SubtractSymbol.Name) { 183 StringRef Name(SymbolicOp.SubtractSymbol.Name); 184 MCSymbol *Sym = Ctx.GetOrCreateSymbol(Name); 185 Sub = MCSymbolRefExpr::Create(Sym, Ctx); 186 } else { 187 Sub = MCConstantExpr::Create(SymbolicOp.SubtractSymbol.Value, Ctx); 188 } 189 } 190 191 const MCExpr *Off = nullptr; 192 if (SymbolicOp.Value != 0) 193 Off = MCConstantExpr::Create(SymbolicOp.Value, Ctx); 194 195 const MCExpr *Expr; 196 if (Sub) { 197 const MCExpr *LHS; 198 if (Add) 199 LHS = MCBinaryExpr::CreateSub(Add, Sub, Ctx); 200 else 201 LHS = MCUnaryExpr::CreateMinus(Sub, Ctx); 202 if (Off) 203 Expr = MCBinaryExpr::CreateAdd(LHS, Off, Ctx); 204 else 205 Expr = LHS; 206 } else if (Add) { 207 if (Off) 208 Expr = MCBinaryExpr::CreateAdd(Add, Off, Ctx); 209 else 210 Expr = Add; 211 } else { 212 if (Off) 213 Expr = Off; 214 else 215 Expr = MCConstantExpr::Create(0, Ctx); 216 } 217 218 MI.addOperand(MCOperand::CreateExpr(Expr)); 219 220 return true; 221} 222