| 1 | //===- DWARFCFIProgram.h ----------------------------------------*- 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_DEBUGINFO_DWARF_DWARFCFIPROGRAM_H |
| 10 | #define LLVM_DEBUGINFO_DWARF_DWARFCFIPROGRAM_H |
| 11 | |
| 12 | #include "llvm/ADT/ArrayRef.h" |
| 13 | #include "llvm/ADT/SmallString.h" |
| 14 | #include "llvm/ADT/iterator.h" |
| 15 | #include "llvm/DebugInfo/DWARF/LowLevel/DWARFDataExtractorSimple.h" |
| 16 | #include "llvm/DebugInfo/DWARF/LowLevel/DWARFExpression.h" |
| 17 | #include "llvm/Support/Compiler.h" |
| 18 | #include "llvm/Support/Error.h" |
| 19 | #include "llvm/TargetParser/Triple.h" |
| 20 | #include <map> |
| 21 | #include <memory> |
| 22 | #include <vector> |
| 23 | |
| 24 | namespace llvm { |
| 25 | |
| 26 | namespace dwarf { |
| 27 | |
| 28 | /// Represent a sequence of Call Frame Information instructions that, when read |
| 29 | /// in order, construct a table mapping PC to frame state. This can also be |
| 30 | /// referred to as "CFI rules" in DWARF literature to avoid confusion with |
| 31 | /// computer programs in the broader sense, and in this context each instruction |
| 32 | /// would be a rule to establish the mapping. Refer to pg. 172 in the DWARF5 |
| 33 | /// manual, "6.4.1 Structure of Call Frame Information". |
| 34 | class CFIProgram { |
| 35 | public: |
| 36 | static constexpr size_t MaxOperands = 3; |
| 37 | typedef SmallVector<uint64_t, MaxOperands> Operands; |
| 38 | |
| 39 | /// An instruction consists of a DWARF CFI opcode and an optional sequence of |
| 40 | /// operands. If it refers to an expression, then this expression has its own |
| 41 | /// sequence of operations and operands handled separately by DWARFExpression. |
| 42 | struct Instruction { |
| 43 | Instruction(uint8_t Opcode) : Opcode(Opcode) {} |
| 44 | |
| 45 | uint8_t Opcode; |
| 46 | Operands Ops; |
| 47 | // Associated DWARF expression in case this instruction refers to one |
| 48 | std::optional<DWARFExpression> Expression; |
| 49 | |
| 50 | LLVM_ABI Expected<uint64_t> getOperandAsUnsigned(const CFIProgram &CFIP, |
| 51 | uint32_t OperandIdx) const; |
| 52 | |
| 53 | LLVM_ABI Expected<int64_t> getOperandAsSigned(const CFIProgram &CFIP, |
| 54 | uint32_t OperandIdx) const; |
| 55 | }; |
| 56 | |
| 57 | using InstrList = std::vector<Instruction>; |
| 58 | using iterator = InstrList::iterator; |
| 59 | using const_iterator = InstrList::const_iterator; |
| 60 | |
| 61 | iterator begin() { return Instructions.begin(); } |
| 62 | const_iterator begin() const { return Instructions.begin(); } |
| 63 | iterator end() { return Instructions.end(); } |
| 64 | const_iterator end() const { return Instructions.end(); } |
| 65 | |
| 66 | unsigned size() const { return (unsigned)Instructions.size(); } |
| 67 | bool empty() const { return Instructions.empty(); } |
| 68 | uint64_t codeAlign() const { return CodeAlignmentFactor; } |
| 69 | int64_t dataAlign() const { return DataAlignmentFactor; } |
| 70 | Triple::ArchType triple() const { return Arch; } |
| 71 | |
| 72 | CFIProgram(uint64_t CodeAlignmentFactor, int64_t DataAlignmentFactor, |
| 73 | Triple::ArchType Arch) |
| 74 | : CodeAlignmentFactor(CodeAlignmentFactor), |
| 75 | DataAlignmentFactor(DataAlignmentFactor), Arch(Arch) {} |
| 76 | |
| 77 | /// Parse and store a sequence of CFI instructions from Data, |
| 78 | /// starting at *Offset and ending at EndOffset. *Offset is updated |
| 79 | /// to EndOffset upon successful parsing, or indicates the offset |
| 80 | /// where a problem occurred in case an error is returned. |
| 81 | template <typename T> |
| 82 | LLVM_ABI Error (DWARFDataExtractorBase<T> &Data, uint64_t *Offset, |
| 83 | uint64_t EndOffset) { |
| 84 | // See DWARF standard v3, section 7.23 |
| 85 | const uint8_t DWARF_CFI_PRIMARY_OPCODE_MASK = 0xc0; |
| 86 | const uint8_t DWARF_CFI_PRIMARY_OPERAND_MASK = 0x3f; |
| 87 | |
| 88 | DataExtractor::Cursor C(*Offset); |
| 89 | while (C && C.tell() < EndOffset) { |
| 90 | uint8_t Opcode = Data.getRelocatedValue(C, 1); |
| 91 | if (!C) |
| 92 | break; |
| 93 | |
| 94 | // Some instructions have a primary opcode encoded in the top bits. |
| 95 | if (uint8_t Primary = Opcode & DWARF_CFI_PRIMARY_OPCODE_MASK) { |
| 96 | // If it's a primary opcode, the first operand is encoded in the |
| 97 | // bottom bits of the opcode itself. |
| 98 | uint64_t Op1 = Opcode & DWARF_CFI_PRIMARY_OPERAND_MASK; |
| 99 | switch (Primary) { |
| 100 | case DW_CFA_advance_loc: |
| 101 | case DW_CFA_restore: |
| 102 | addInstruction(Opcode: Primary, Operand1: Op1); |
| 103 | break; |
| 104 | case DW_CFA_offset: |
| 105 | addInstruction(Primary, Op1, Data.getULEB128(C)); |
| 106 | break; |
| 107 | default: |
| 108 | llvm_unreachable("invalid primary CFI opcode" ); |
| 109 | } |
| 110 | continue; |
| 111 | } |
| 112 | |
| 113 | // Extended opcode - its value is Opcode itself. |
| 114 | switch (Opcode) { |
| 115 | default: |
| 116 | return createStringError(EC: errc::illegal_byte_sequence, |
| 117 | Fmt: "invalid extended CFI opcode 0x%" PRIx8, |
| 118 | Vals: Opcode); |
| 119 | case DW_CFA_nop: |
| 120 | case DW_CFA_remember_state: |
| 121 | case DW_CFA_restore_state: |
| 122 | case DW_CFA_GNU_window_save: |
| 123 | case DW_CFA_AARCH64_negate_ra_state_with_pc: |
| 124 | // No operands |
| 125 | addInstruction(Opcode); |
| 126 | break; |
| 127 | case DW_CFA_set_loc: |
| 128 | // Operands: Address |
| 129 | addInstruction(Opcode, Data.getRelocatedAddress(C)); |
| 130 | break; |
| 131 | case DW_CFA_advance_loc1: |
| 132 | // Operands: 1-byte delta |
| 133 | addInstruction(Opcode, Data.getRelocatedValue(C, 1)); |
| 134 | break; |
| 135 | case DW_CFA_advance_loc2: |
| 136 | // Operands: 2-byte delta |
| 137 | addInstruction(Opcode, Data.getRelocatedValue(C, 2)); |
| 138 | break; |
| 139 | case DW_CFA_advance_loc4: |
| 140 | // Operands: 4-byte delta |
| 141 | addInstruction(Opcode, Data.getRelocatedValue(C, 4)); |
| 142 | break; |
| 143 | case DW_CFA_restore_extended: |
| 144 | case DW_CFA_undefined: |
| 145 | case DW_CFA_same_value: |
| 146 | case DW_CFA_def_cfa_register: |
| 147 | case DW_CFA_def_cfa_offset: |
| 148 | case DW_CFA_GNU_args_size: |
| 149 | // Operands: ULEB128 |
| 150 | addInstruction(Opcode, Data.getULEB128(C)); |
| 151 | break; |
| 152 | case DW_CFA_def_cfa_offset_sf: |
| 153 | // Operands: SLEB128 |
| 154 | addInstruction(Opcode, Data.getSLEB128(C)); |
| 155 | break; |
| 156 | case DW_CFA_LLVM_def_aspace_cfa: |
| 157 | case DW_CFA_LLVM_def_aspace_cfa_sf: { |
| 158 | auto RegNum = Data.getULEB128(C); |
| 159 | auto CfaOffset = Opcode == DW_CFA_LLVM_def_aspace_cfa |
| 160 | ? Data.getULEB128(C) |
| 161 | : Data.getSLEB128(C); |
| 162 | auto AddressSpace = Data.getULEB128(C); |
| 163 | addInstruction(Opcode, RegNum, CfaOffset, AddressSpace); |
| 164 | break; |
| 165 | } |
| 166 | case DW_CFA_offset_extended: |
| 167 | case DW_CFA_register: |
| 168 | case DW_CFA_def_cfa: |
| 169 | case DW_CFA_val_offset: { |
| 170 | // Operands: ULEB128, ULEB128 |
| 171 | // Note: We can not embed getULEB128 directly into function |
| 172 | // argument list. getULEB128 changes Offset and order of evaluation |
| 173 | // for arguments is unspecified. |
| 174 | uint64_t op1 = Data.getULEB128(C); |
| 175 | uint64_t op2 = Data.getULEB128(C); |
| 176 | addInstruction(Opcode, Operand1: op1, Operand2: op2); |
| 177 | break; |
| 178 | } |
| 179 | case DW_CFA_offset_extended_sf: |
| 180 | case DW_CFA_def_cfa_sf: |
| 181 | case DW_CFA_val_offset_sf: { |
| 182 | // Operands: ULEB128, SLEB128 |
| 183 | // Note: see comment for the previous case |
| 184 | uint64_t op1 = Data.getULEB128(C); |
| 185 | uint64_t op2 = (uint64_t)Data.getSLEB128(C); |
| 186 | addInstruction(Opcode, Operand1: op1, Operand2: op2); |
| 187 | break; |
| 188 | } |
| 189 | case DW_CFA_def_cfa_expression: { |
| 190 | uint64_t ExprLength = Data.getULEB128(C); |
| 191 | addInstruction(Opcode, Operand1: 0); |
| 192 | StringRef Expression = Data.getBytes(C, ExprLength); |
| 193 | |
| 194 | DataExtractor (Expression, Data.isLittleEndian(), |
| 195 | Data.getAddressSize()); |
| 196 | // Note. We do not pass the DWARF format to DWARFExpression, because |
| 197 | // DW_OP_call_ref, the only operation which depends on the format, is |
| 198 | // prohibited in call frame instructions, see sec. 6.4.2 in DWARFv5. |
| 199 | Instructions.back().Expression = |
| 200 | DWARFExpression(Extractor, Data.getAddressSize()); |
| 201 | break; |
| 202 | } |
| 203 | case DW_CFA_expression: |
| 204 | case DW_CFA_val_expression: { |
| 205 | uint64_t RegNum = Data.getULEB128(C); |
| 206 | addInstruction(Opcode, Operand1: RegNum, Operand2: 0); |
| 207 | |
| 208 | uint64_t BlockLength = Data.getULEB128(C); |
| 209 | StringRef Expression = Data.getBytes(C, BlockLength); |
| 210 | DataExtractor (Expression, Data.isLittleEndian(), |
| 211 | Data.getAddressSize()); |
| 212 | // Note. We do not pass the DWARF format to DWARFExpression, because |
| 213 | // DW_OP_call_ref, the only operation which depends on the format, is |
| 214 | // prohibited in call frame instructions, see sec. 6.4.2 in DWARFv5. |
| 215 | Instructions.back().Expression = |
| 216 | DWARFExpression(Extractor, Data.getAddressSize()); |
| 217 | break; |
| 218 | } |
| 219 | } |
| 220 | } |
| 221 | |
| 222 | *Offset = C.tell(); |
| 223 | return C.takeError(); |
| 224 | } |
| 225 | |
| 226 | void addInstruction(const Instruction &I) { Instructions.push_back(x: I); } |
| 227 | |
| 228 | /// Get a DWARF CFI call frame string for the given DW_CFA opcode. |
| 229 | LLVM_ABI StringRef callFrameString(unsigned Opcode) const; |
| 230 | |
| 231 | /// Types of operands to CFI instructions |
| 232 | /// In DWARF, this type is implicitly tied to a CFI instruction opcode and |
| 233 | /// thus this type doesn't need to be explicitly written to the file (this is |
| 234 | /// not a DWARF encoding). The relationship of instrs to operand types can |
| 235 | /// be obtained from getOperandTypes() and is only used to simplify |
| 236 | /// instruction printing and error messages. |
| 237 | enum OperandType { |
| 238 | OT_Unset, |
| 239 | OT_None, |
| 240 | OT_Address, |
| 241 | OT_Offset, |
| 242 | OT_FactoredCodeOffset, |
| 243 | OT_SignedFactDataOffset, |
| 244 | OT_UnsignedFactDataOffset, |
| 245 | OT_Register, |
| 246 | OT_AddressSpace, |
| 247 | OT_Expression |
| 248 | }; |
| 249 | |
| 250 | /// Get the OperandType as a "const char *". |
| 251 | LLVM_ABI static const char *operandTypeString(OperandType OT); |
| 252 | |
| 253 | /// Retrieve the array describing the types of operands according to the enum |
| 254 | /// above. This is indexed by opcode. |
| 255 | LLVM_ABI static ArrayRef<OperandType[MaxOperands]> getOperandTypes(); |
| 256 | |
| 257 | /// Convenience method to add a new instruction with the given opcode. |
| 258 | void addInstruction(uint8_t Opcode) { |
| 259 | Instructions.push_back(x: Instruction(Opcode)); |
| 260 | } |
| 261 | |
| 262 | /// Add a new single-operand instruction. |
| 263 | void addInstruction(uint8_t Opcode, uint64_t Operand1) { |
| 264 | Instructions.push_back(x: Instruction(Opcode)); |
| 265 | Instructions.back().Ops.push_back(Elt: Operand1); |
| 266 | } |
| 267 | |
| 268 | /// Add a new instruction that has two operands. |
| 269 | void addInstruction(uint8_t Opcode, uint64_t Operand1, uint64_t Operand2) { |
| 270 | Instructions.push_back(x: Instruction(Opcode)); |
| 271 | Instructions.back().Ops.push_back(Elt: Operand1); |
| 272 | Instructions.back().Ops.push_back(Elt: Operand2); |
| 273 | } |
| 274 | |
| 275 | /// Add a new instruction that has three operands. |
| 276 | void addInstruction(uint8_t Opcode, uint64_t Operand1, uint64_t Operand2, |
| 277 | uint64_t Operand3) { |
| 278 | Instructions.push_back(x: Instruction(Opcode)); |
| 279 | Instructions.back().Ops.push_back(Elt: Operand1); |
| 280 | Instructions.back().Ops.push_back(Elt: Operand2); |
| 281 | Instructions.back().Ops.push_back(Elt: Operand3); |
| 282 | } |
| 283 | |
| 284 | private: |
| 285 | std::vector<Instruction> Instructions; |
| 286 | const uint64_t CodeAlignmentFactor; |
| 287 | const int64_t DataAlignmentFactor; |
| 288 | Triple::ArchType Arch; |
| 289 | }; |
| 290 | |
| 291 | } // end namespace dwarf |
| 292 | |
| 293 | } // end namespace llvm |
| 294 | |
| 295 | #endif // LLVM_DEBUGINFO_DWARF_DWARFCFIPROGRAM_H |
| 296 | |