| 1 | // Copyright (c) 2012, the Dart project authors. Please see the AUTHORS file |
| 2 | // for details. All rights reserved. Use of this source code is governed by a |
| 3 | // BSD-style license that can be found in the LICENSE file. |
| 4 | |
| 5 | // A combined disassembler for IA32 and X64. |
| 6 | |
| 7 | #include "vm/globals.h" // Needed here to get TARGET_ARCH_xxx. |
| 8 | #if defined(TARGET_ARCH_X64) || defined(TARGET_ARCH_IA32) |
| 9 | |
| 10 | #include "vm/compiler/assembler/disassembler.h" |
| 11 | |
| 12 | #include "platform/unaligned.h" |
| 13 | #include "platform/utils.h" |
| 14 | #include "vm/allocation.h" |
| 15 | #include "vm/constants_x86.h" |
| 16 | #include "vm/heap/heap.h" |
| 17 | #include "vm/instructions.h" |
| 18 | #include "vm/os.h" |
| 19 | #include "vm/stack_frame.h" |
| 20 | #include "vm/stub_code.h" |
| 21 | |
| 22 | namespace dart { |
| 23 | |
| 24 | #if !defined(PRODUCT) || defined(FORCE_INCLUDE_DISASSEMBLER) |
| 25 | |
| 26 | enum OperandType { |
| 27 | UNSET_OP_ORDER = 0, |
| 28 | // Operand size decides between 16, 32 and 64 bit operands. |
| 29 | REG_OPER_OP_ORDER = 1, // Register destination, operand source. |
| 30 | OPER_REG_OP_ORDER = 2, // Operand destination, register source. |
| 31 | // Fixed 8-bit operands. |
| 32 | BYTE_SIZE_OPERAND_FLAG = 4, |
| 33 | BYTE_REG_OPER_OP_ORDER = REG_OPER_OP_ORDER | BYTE_SIZE_OPERAND_FLAG, |
| 34 | BYTE_OPER_REG_OP_ORDER = OPER_REG_OP_ORDER | BYTE_SIZE_OPERAND_FLAG |
| 35 | }; |
| 36 | |
| 37 | //------------------------------------------------------------------ |
| 38 | // Tables |
| 39 | //------------------------------------------------------------------ |
| 40 | struct ByteMnemonic { |
| 41 | int b; // -1 terminates, otherwise must be in range (0..255) |
| 42 | OperandType op_order_; |
| 43 | const char* mnem; |
| 44 | }; |
| 45 | |
| 46 | #define ALU_ENTRY(name, code) \ |
| 47 | {code * 8 + 0, BYTE_OPER_REG_OP_ORDER, #name}, \ |
| 48 | {code * 8 + 1, OPER_REG_OP_ORDER, #name}, \ |
| 49 | {code * 8 + 2, BYTE_REG_OPER_OP_ORDER, #name}, \ |
| 50 | {code * 8 + 3, REG_OPER_OP_ORDER, #name}, |
| 51 | static const ByteMnemonic two_operands_instr[] = { |
| 52 | X86_ALU_CODES(ALU_ENTRY){0x63, REG_OPER_OP_ORDER, "movsxd" }, |
| 53 | {0x84, BYTE_REG_OPER_OP_ORDER, "test" }, |
| 54 | {0x85, REG_OPER_OP_ORDER, "test" }, |
| 55 | {0x86, BYTE_REG_OPER_OP_ORDER, "xchg" }, |
| 56 | {0x87, REG_OPER_OP_ORDER, "xchg" }, |
| 57 | {0x88, BYTE_OPER_REG_OP_ORDER, "mov" }, |
| 58 | {0x89, OPER_REG_OP_ORDER, "mov" }, |
| 59 | {0x8A, BYTE_REG_OPER_OP_ORDER, "mov" }, |
| 60 | {0x8B, REG_OPER_OP_ORDER, "mov" }, |
| 61 | {0x8D, REG_OPER_OP_ORDER, "lea" }, |
| 62 | {-1, UNSET_OP_ORDER, "" }}; |
| 63 | |
| 64 | #define ZERO_OPERAND_ENTRY(name, opcode) {opcode, UNSET_OP_ORDER, #name}, |
| 65 | static const ByteMnemonic zero_operands_instr[] = { |
| 66 | X86_ZERO_OPERAND_1_BYTE_INSTRUCTIONS(ZERO_OPERAND_ENTRY){-1, UNSET_OP_ORDER, |
| 67 | "" }}; |
| 68 | |
| 69 | static const ByteMnemonic call_jump_instr[] = {{0xE8, UNSET_OP_ORDER, "call" }, |
| 70 | {0xE9, UNSET_OP_ORDER, "jmp" }, |
| 71 | {-1, UNSET_OP_ORDER, "" }}; |
| 72 | |
| 73 | #define SHORT_IMMEDIATE_ENTRY(name, code) {code * 8 + 5, UNSET_OP_ORDER, #name}, |
| 74 | static const ByteMnemonic short_immediate_instr[] = { |
| 75 | X86_ALU_CODES(SHORT_IMMEDIATE_ENTRY){-1, UNSET_OP_ORDER, "" }}; |
| 76 | |
| 77 | static const char* const conditional_code_suffix[] = { |
| 78 | #define STRINGIFY(name, number) #name, |
| 79 | X86_CONDITIONAL_SUFFIXES(STRINGIFY) |
| 80 | #undef STRINGIFY |
| 81 | }; |
| 82 | |
| 83 | #define STRINGIFY_NAME(name, code) #name, |
| 84 | static const char* const xmm_conditional_code_suffix[] = { |
| 85 | XMM_CONDITIONAL_CODES(STRINGIFY_NAME)}; |
| 86 | #undef STRINGIFY_NAME |
| 87 | |
| 88 | enum InstructionType { |
| 89 | NO_INSTR, |
| 90 | ZERO_OPERANDS_INSTR, |
| 91 | TWO_OPERANDS_INSTR, |
| 92 | JUMP_CONDITIONAL_SHORT_INSTR, |
| 93 | REGISTER_INSTR, |
| 94 | PUSHPOP_INSTR, // Has implicit 64-bit operand size. |
| 95 | MOVE_REG_INSTR, |
| 96 | CALL_JUMP_INSTR, |
| 97 | SHORT_IMMEDIATE_INSTR |
| 98 | }; |
| 99 | |
| 100 | enum Prefixes { |
| 101 | ESCAPE_PREFIX = 0x0F, |
| 102 | OPERAND_SIZE_OVERRIDE_PREFIX = 0x66, |
| 103 | ADDRESS_SIZE_OVERRIDE_PREFIX = 0x67, |
| 104 | REPNE_PREFIX = 0xF2, |
| 105 | REP_PREFIX = 0xF3, |
| 106 | REPEQ_PREFIX = REP_PREFIX |
| 107 | }; |
| 108 | |
| 109 | struct XmmMnemonic { |
| 110 | const char* ps_name; |
| 111 | const char* pd_name; |
| 112 | const char* ss_name; |
| 113 | const char* sd_name; |
| 114 | }; |
| 115 | |
| 116 | #define XMM_INSTRUCTION_ENTRY(name, code) \ |
| 117 | {#name "ps", #name "pd", #name "ss", #name "sd"}, |
| 118 | static const XmmMnemonic xmm_instructions[] = { |
| 119 | XMM_ALU_CODES(XMM_INSTRUCTION_ENTRY)}; |
| 120 | |
| 121 | struct InstructionDesc { |
| 122 | const char* mnem; |
| 123 | InstructionType type; |
| 124 | OperandType op_order_; |
| 125 | bool byte_size_operation; // Fixed 8-bit operation. |
| 126 | }; |
| 127 | |
| 128 | class InstructionTable : public ValueObject { |
| 129 | public: |
| 130 | InstructionTable(); |
| 131 | const InstructionDesc& Get(uint8_t x) const { return instructions_[x]; } |
| 132 | |
| 133 | private: |
| 134 | InstructionDesc instructions_[256]; |
| 135 | void Clear(); |
| 136 | void Init(); |
| 137 | void CopyTable(const ByteMnemonic bm[], InstructionType type); |
| 138 | void SetTableRange(InstructionType type, |
| 139 | uint8_t start, |
| 140 | uint8_t end, |
| 141 | bool byte_size, |
| 142 | const char* mnem); |
| 143 | void AddJumpConditionalShort(); |
| 144 | |
| 145 | DISALLOW_COPY_AND_ASSIGN(InstructionTable); |
| 146 | }; |
| 147 | |
| 148 | InstructionTable::InstructionTable() { |
| 149 | Clear(); |
| 150 | Init(); |
| 151 | } |
| 152 | |
| 153 | void InstructionTable::Clear() { |
| 154 | for (int i = 0; i < 256; i++) { |
| 155 | instructions_[i].mnem = "(bad)" ; |
| 156 | instructions_[i].type = NO_INSTR; |
| 157 | instructions_[i].op_order_ = UNSET_OP_ORDER; |
| 158 | instructions_[i].byte_size_operation = false; |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | void InstructionTable::Init() { |
| 163 | CopyTable(two_operands_instr, TWO_OPERANDS_INSTR); |
| 164 | CopyTable(zero_operands_instr, ZERO_OPERANDS_INSTR); |
| 165 | CopyTable(call_jump_instr, CALL_JUMP_INSTR); |
| 166 | CopyTable(short_immediate_instr, SHORT_IMMEDIATE_INSTR); |
| 167 | AddJumpConditionalShort(); |
| 168 | SetTableRange(PUSHPOP_INSTR, 0x50, 0x57, false, "push" ); |
| 169 | SetTableRange(PUSHPOP_INSTR, 0x58, 0x5F, false, "pop" ); |
| 170 | SetTableRange(MOVE_REG_INSTR, 0xB8, 0xBF, false, "mov" ); |
| 171 | } |
| 172 | |
| 173 | void InstructionTable::CopyTable(const ByteMnemonic bm[], |
| 174 | InstructionType type) { |
| 175 | for (int i = 0; bm[i].b >= 0; i++) { |
| 176 | InstructionDesc* id = &instructions_[bm[i].b]; |
| 177 | id->mnem = bm[i].mnem; |
| 178 | OperandType op_order = bm[i].op_order_; |
| 179 | id->op_order_ = |
| 180 | static_cast<OperandType>(op_order & ~BYTE_SIZE_OPERAND_FLAG); |
| 181 | ASSERT(NO_INSTR == id->type); // Information not already entered |
| 182 | id->type = type; |
| 183 | id->byte_size_operation = ((op_order & BYTE_SIZE_OPERAND_FLAG) != 0); |
| 184 | } |
| 185 | } |
| 186 | |
| 187 | void InstructionTable::SetTableRange(InstructionType type, |
| 188 | uint8_t start, |
| 189 | uint8_t end, |
| 190 | bool byte_size, |
| 191 | const char* mnem) { |
| 192 | for (uint8_t b = start; b <= end; b++) { |
| 193 | InstructionDesc* id = &instructions_[b]; |
| 194 | ASSERT(NO_INSTR == id->type); // Information not already entered |
| 195 | id->mnem = mnem; |
| 196 | id->type = type; |
| 197 | id->byte_size_operation = byte_size; |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | void InstructionTable::AddJumpConditionalShort() { |
| 202 | for (uint8_t b = 0x70; b <= 0x7F; b++) { |
| 203 | InstructionDesc* id = &instructions_[b]; |
| 204 | ASSERT(NO_INSTR == id->type); // Information not already entered |
| 205 | id->mnem = NULL; // Computed depending on condition code. |
| 206 | id->type = JUMP_CONDITIONAL_SHORT_INSTR; |
| 207 | } |
| 208 | } |
| 209 | |
| 210 | static InstructionTable instruction_table; |
| 211 | |
| 212 | static InstructionDesc cmov_instructions[16] = { |
| 213 | {"cmovo" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 214 | {"cmovno" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 215 | {"cmovc" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 216 | {"cmovnc" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 217 | {"cmovz" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 218 | {"cmovnz" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 219 | {"cmovna" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 220 | {"cmova" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 221 | {"cmovs" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 222 | {"cmovns" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 223 | {"cmovpe" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 224 | {"cmovpo" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 225 | {"cmovl" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 226 | {"cmovge" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 227 | {"cmovle" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}, |
| 228 | {"cmovg" , TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}}; |
| 229 | |
| 230 | //------------------------------------------------- |
| 231 | // DisassemblerX64 implementation. |
| 232 | |
| 233 | static const int kMaxXmmRegisters = 16; |
| 234 | static const char* xmm_regs[kMaxXmmRegisters] = { |
| 235 | "xmm0" , "xmm1" , "xmm2" , "xmm3" , "xmm4" , "xmm5" , "xmm6" , "xmm7" , |
| 236 | "xmm8" , "xmm9" , "xmm10" , "xmm11" , "xmm12" , "xmm13" , "xmm14" , "xmm15" }; |
| 237 | |
| 238 | class DisassemblerX64 : public ValueObject { |
| 239 | public: |
| 240 | DisassemblerX64(char* buffer, intptr_t buffer_size) |
| 241 | : buffer_(buffer), |
| 242 | buffer_size_(buffer_size), |
| 243 | buffer_pos_(0), |
| 244 | rex_(0), |
| 245 | operand_size_(0), |
| 246 | group_1_prefix_(0), |
| 247 | byte_size_operand_(false) { |
| 248 | buffer_[buffer_pos_] = '\0'; |
| 249 | } |
| 250 | |
| 251 | virtual ~DisassemblerX64() {} |
| 252 | |
| 253 | int InstructionDecode(uword pc); |
| 254 | |
| 255 | private: |
| 256 | enum OperandSize { |
| 257 | BYTE_SIZE = 0, |
| 258 | WORD_SIZE = 1, |
| 259 | DOUBLEWORD_SIZE = 2, |
| 260 | QUADWORD_SIZE = 3 |
| 261 | }; |
| 262 | |
| 263 | void setRex(uint8_t rex) { |
| 264 | ASSERT(0x40 == (rex & 0xF0)); |
| 265 | rex_ = rex; |
| 266 | } |
| 267 | |
| 268 | bool rex() { return rex_ != 0; } |
| 269 | |
| 270 | bool rex_b() { return (rex_ & 0x01) != 0; } |
| 271 | |
| 272 | // Actual number of base register given the low bits and the rex.b state. |
| 273 | int base_reg(int low_bits) { return low_bits | ((rex_ & 0x01) << 3); } |
| 274 | |
| 275 | bool rex_x() { return (rex_ & 0x02) != 0; } |
| 276 | |
| 277 | bool rex_r() { return (rex_ & 0x04) != 0; } |
| 278 | |
| 279 | bool rex_w() { return (rex_ & 0x08) != 0; } |
| 280 | |
| 281 | OperandSize operand_size() { |
| 282 | if (byte_size_operand_) return BYTE_SIZE; |
| 283 | if (rex_w()) return QUADWORD_SIZE; |
| 284 | if (operand_size_ != 0) return WORD_SIZE; |
| 285 | return DOUBLEWORD_SIZE; |
| 286 | } |
| 287 | |
| 288 | const char* operand_size_code() { |
| 289 | #if defined(TARGET_ARCH_X64) |
| 290 | return &"b\0w\0l\0q\0" [2 * operand_size()]; |
| 291 | #else |
| 292 | // We omit the 'l' suffix on IA32. |
| 293 | return &"b\0w\0\0\0q\0" [2 * operand_size()]; |
| 294 | #endif |
| 295 | } |
| 296 | |
| 297 | // Disassembler helper functions. |
| 298 | void get_modrm(uint8_t data, int* mod, int* regop, int* rm) { |
| 299 | *mod = (data >> 6) & 3; |
| 300 | *regop = ((data & 0x38) >> 3) | (rex_r() ? 8 : 0); |
| 301 | *rm = (data & 7) | (rex_b() ? 8 : 0); |
| 302 | ASSERT(*rm < kNumberOfCpuRegisters); |
| 303 | } |
| 304 | |
| 305 | void get_sib(uint8_t data, int* scale, int* index, int* base) { |
| 306 | *scale = (data >> 6) & 3; |
| 307 | *index = ((data >> 3) & 7) | (rex_x() ? 8 : 0); |
| 308 | *base = (data & 7) | (rex_b() ? 8 : 0); |
| 309 | ASSERT(*base < kNumberOfCpuRegisters); |
| 310 | } |
| 311 | |
| 312 | const char* NameOfCPURegister(int reg) const { |
| 313 | return RegisterNames::RegisterName(static_cast<Register>(reg)); |
| 314 | } |
| 315 | |
| 316 | const char* NameOfByteCPURegister(int reg) const { |
| 317 | return NameOfCPURegister(reg); |
| 318 | } |
| 319 | |
| 320 | // A way to get rax or eax's name. |
| 321 | const char* Rax() const { return NameOfCPURegister(0); } |
| 322 | |
| 323 | const char* NameOfXMMRegister(int reg) const { |
| 324 | ASSERT((0 <= reg) && (reg < kMaxXmmRegisters)); |
| 325 | return xmm_regs[reg]; |
| 326 | } |
| 327 | |
| 328 | void Print(const char* format, ...) PRINTF_ATTRIBUTE(2, 3); |
| 329 | void PrintJump(uint8_t* pc, int32_t disp); |
| 330 | void PrintAddress(uint8_t* addr); |
| 331 | |
| 332 | int PrintOperands(const char* mnem, OperandType op_order, uint8_t* data); |
| 333 | |
| 334 | typedef const char* (DisassemblerX64::*RegisterNameMapping)(int reg) const; |
| 335 | |
| 336 | int PrintRightOperandHelper(uint8_t* modrmp, |
| 337 | RegisterNameMapping register_name); |
| 338 | int PrintRightOperand(uint8_t* modrmp); |
| 339 | int PrintRightByteOperand(uint8_t* modrmp); |
| 340 | int PrintRightXMMOperand(uint8_t* modrmp); |
| 341 | void PrintDisp(int disp, const char* after); |
| 342 | int PrintImmediate(uint8_t* data, OperandSize size, bool sign_extend = false); |
| 343 | void PrintImmediateValue(int64_t value, |
| 344 | bool signed_value = false, |
| 345 | int byte_count = -1); |
| 346 | int PrintImmediateOp(uint8_t* data); |
| 347 | const char* TwoByteMnemonic(uint8_t opcode); |
| 348 | int TwoByteOpcodeInstruction(uint8_t* data); |
| 349 | int Print660F38Instruction(uint8_t* data); |
| 350 | |
| 351 | int F6F7Instruction(uint8_t* data); |
| 352 | int ShiftInstruction(uint8_t* data); |
| 353 | int JumpShort(uint8_t* data); |
| 354 | int JumpConditional(uint8_t* data); |
| 355 | int JumpConditionalShort(uint8_t* data); |
| 356 | int SetCC(uint8_t* data); |
| 357 | int FPUInstruction(uint8_t* data); |
| 358 | int MemoryFPUInstruction(int escape_opcode, int regop, uint8_t* modrm_start); |
| 359 | int RegisterFPUInstruction(int escape_opcode, uint8_t modrm_byte); |
| 360 | |
| 361 | bool DecodeInstructionType(uint8_t** data); |
| 362 | |
| 363 | void UnimplementedInstruction() { UNREACHABLE(); } |
| 364 | |
| 365 | char* buffer_; // Decode instructions into this buffer. |
| 366 | intptr_t buffer_size_; // The size of the buffer_. |
| 367 | intptr_t buffer_pos_; // Current character position in the buffer_. |
| 368 | |
| 369 | // Prefixes parsed |
| 370 | uint8_t rex_; |
| 371 | uint8_t operand_size_; // 0x66 or (if no group 3 prefix is present) 0x0. |
| 372 | // 0xF2, 0xF3, or (if no group 1 prefix is present) 0. |
| 373 | uint8_t group_1_prefix_; |
| 374 | // Byte size operand override. |
| 375 | bool byte_size_operand_; |
| 376 | |
| 377 | DISALLOW_COPY_AND_ASSIGN(DisassemblerX64); |
| 378 | }; |
| 379 | |
| 380 | // Append the str to the output buffer. |
| 381 | void DisassemblerX64::Print(const char* format, ...) { |
| 382 | intptr_t available = buffer_size_ - buffer_pos_; |
| 383 | if (available <= 1) { |
| 384 | ASSERT(buffer_[buffer_pos_] == '\0'); |
| 385 | return; |
| 386 | } |
| 387 | char* buf = buffer_ + buffer_pos_; |
| 388 | va_list args; |
| 389 | va_start(args, format); |
| 390 | int length = Utils::VSNPrint(buf, available, format, args); |
| 391 | va_end(args); |
| 392 | buffer_pos_ = |
| 393 | (length >= available) ? (buffer_size_ - 1) : (buffer_pos_ + length); |
| 394 | ASSERT(buffer_pos_ < buffer_size_); |
| 395 | } |
| 396 | |
| 397 | int DisassemblerX64::PrintRightOperandHelper( |
| 398 | uint8_t* modrmp, |
| 399 | RegisterNameMapping direct_register_name) { |
| 400 | int mod, regop, rm; |
| 401 | get_modrm(*modrmp, &mod, ®op, &rm); |
| 402 | RegisterNameMapping register_name = |
| 403 | (mod == 3) ? direct_register_name : &DisassemblerX64::NameOfCPURegister; |
| 404 | switch (mod) { |
| 405 | case 0: |
| 406 | if ((rm & 7) == 5) { |
| 407 | int32_t disp = LoadUnaligned(reinterpret_cast<int32_t*>(modrmp + 1)); |
| 408 | Print("[rip" ); |
| 409 | PrintDisp(disp, "]" ); |
| 410 | return 5; |
| 411 | } else if ((rm & 7) == 4) { |
| 412 | // Codes for SIB byte. |
| 413 | uint8_t sib = *(modrmp + 1); |
| 414 | int scale, index, base; |
| 415 | get_sib(sib, &scale, &index, &base); |
| 416 | if (index == 4 && (base & 7) == 4 && scale == 0 /*times_1*/) { |
| 417 | // index == rsp means no index. Only use sib byte with no index for |
| 418 | // rsp and r12 base. |
| 419 | Print("[%s]" , NameOfCPURegister(base)); |
| 420 | return 2; |
| 421 | } else if (base == 5) { |
| 422 | // base == rbp means no base register (when mod == 0). |
| 423 | int32_t disp = LoadUnaligned(reinterpret_cast<int32_t*>(modrmp + 2)); |
| 424 | Print("[%s*%d" , NameOfCPURegister(index), 1 << scale); |
| 425 | PrintDisp(disp, "]" ); |
| 426 | return 6; |
| 427 | } else if (index != 4 && base != 5) { |
| 428 | // [base+index*scale] |
| 429 | Print("[%s+%s*%d]" , NameOfCPURegister(base), NameOfCPURegister(index), |
| 430 | 1 << scale); |
| 431 | return 2; |
| 432 | } else { |
| 433 | UnimplementedInstruction(); |
| 434 | return 1; |
| 435 | } |
| 436 | } else { |
| 437 | Print("[%s]" , NameOfCPURegister(rm)); |
| 438 | return 1; |
| 439 | } |
| 440 | break; |
| 441 | case 1: |
| 442 | FALL_THROUGH; |
| 443 | case 2: |
| 444 | if ((rm & 7) == 4) { |
| 445 | uint8_t sib = *(modrmp + 1); |
| 446 | int scale, index, base; |
| 447 | get_sib(sib, &scale, &index, &base); |
| 448 | int disp = (mod == 2) |
| 449 | ? LoadUnaligned(reinterpret_cast<int32_t*>(modrmp + 2)) |
| 450 | : *reinterpret_cast<int8_t*>(modrmp + 2); |
| 451 | if (index == 4 && (base & 7) == 4 && scale == 0 /*times_1*/) { |
| 452 | Print("[%s" , NameOfCPURegister(base)); |
| 453 | PrintDisp(disp, "]" ); |
| 454 | } else { |
| 455 | Print("[%s+%s*%d" , NameOfCPURegister(base), NameOfCPURegister(index), |
| 456 | 1 << scale); |
| 457 | PrintDisp(disp, "]" ); |
| 458 | } |
| 459 | return mod == 2 ? 6 : 3; |
| 460 | } else { |
| 461 | // No sib. |
| 462 | int disp = (mod == 2) |
| 463 | ? LoadUnaligned(reinterpret_cast<int32_t*>(modrmp + 1)) |
| 464 | : *reinterpret_cast<int8_t*>(modrmp + 1); |
| 465 | Print("[%s" , NameOfCPURegister(rm)); |
| 466 | PrintDisp(disp, "]" ); |
| 467 | return (mod == 2) ? 5 : 2; |
| 468 | } |
| 469 | break; |
| 470 | case 3: |
| 471 | Print("%s" , (this->*register_name)(rm)); |
| 472 | return 1; |
| 473 | default: |
| 474 | UnimplementedInstruction(); |
| 475 | return 1; |
| 476 | } |
| 477 | UNREACHABLE(); |
| 478 | } |
| 479 | |
| 480 | int DisassemblerX64::PrintImmediate(uint8_t* data, |
| 481 | OperandSize size, |
| 482 | bool sign_extend) { |
| 483 | int64_t value; |
| 484 | int count; |
| 485 | switch (size) { |
| 486 | case BYTE_SIZE: |
| 487 | if (sign_extend) { |
| 488 | value = *reinterpret_cast<int8_t*>(data); |
| 489 | } else { |
| 490 | value = *data; |
| 491 | } |
| 492 | count = 1; |
| 493 | break; |
| 494 | case WORD_SIZE: |
| 495 | if (sign_extend) { |
| 496 | value = LoadUnaligned(reinterpret_cast<int16_t*>(data)); |
| 497 | } else { |
| 498 | value = LoadUnaligned(reinterpret_cast<uint16_t*>(data)); |
| 499 | } |
| 500 | count = 2; |
| 501 | break; |
| 502 | case DOUBLEWORD_SIZE: |
| 503 | case QUADWORD_SIZE: |
| 504 | if (sign_extend) { |
| 505 | value = LoadUnaligned(reinterpret_cast<int32_t*>(data)); |
| 506 | } else { |
| 507 | value = LoadUnaligned(reinterpret_cast<uint32_t*>(data)); |
| 508 | } |
| 509 | count = 4; |
| 510 | break; |
| 511 | default: |
| 512 | UNREACHABLE(); |
| 513 | value = 0; // Initialize variables on all paths to satisfy the compiler. |
| 514 | count = 0; |
| 515 | } |
| 516 | PrintImmediateValue(value, sign_extend, count); |
| 517 | return count; |
| 518 | } |
| 519 | |
| 520 | void DisassemblerX64::PrintImmediateValue(int64_t value, |
| 521 | bool signed_value, |
| 522 | int byte_count) { |
| 523 | if ((value >= 0) && (value <= 9)) { |
| 524 | Print("%" Pd64, value); |
| 525 | } else if (signed_value && (value < 0) && (value >= -9)) { |
| 526 | Print("-%" Pd64, -value); |
| 527 | } else { |
| 528 | if (byte_count == 1) { |
| 529 | int8_t v8 = static_cast<int8_t>(value); |
| 530 | if (v8 < 0 && signed_value) { |
| 531 | Print("-%#" Px32, -static_cast<uint8_t>(v8)); |
| 532 | } else { |
| 533 | Print("%#" Px32, static_cast<uint8_t>(v8)); |
| 534 | } |
| 535 | } else if (byte_count == 2) { |
| 536 | int16_t v16 = static_cast<int16_t>(value); |
| 537 | if (v16 < 0 && signed_value) { |
| 538 | Print("-%#" Px32, -static_cast<uint16_t>(v16)); |
| 539 | } else { |
| 540 | Print("%#" Px32, static_cast<uint16_t>(v16)); |
| 541 | } |
| 542 | } else if (byte_count == 4) { |
| 543 | int32_t v32 = static_cast<int32_t>(value); |
| 544 | if (v32 < 0 && signed_value) { |
| 545 | Print("-%#010" Px32, -static_cast<uint32_t>(v32)); |
| 546 | } else { |
| 547 | if (v32 > 0xffff) { |
| 548 | Print("%#010" Px32, v32); |
| 549 | } else { |
| 550 | Print("%#" Px32, v32); |
| 551 | } |
| 552 | } |
| 553 | } else if (byte_count == 8) { |
| 554 | int64_t v64 = static_cast<int64_t>(value); |
| 555 | if (v64 < 0 && signed_value) { |
| 556 | Print("-%#018" Px64, -static_cast<uint64_t>(v64)); |
| 557 | } else { |
| 558 | if (v64 > 0xffffffffll) { |
| 559 | Print("%#018" Px64, v64); |
| 560 | } else { |
| 561 | Print("%#" Px64, v64); |
| 562 | } |
| 563 | } |
| 564 | } else { |
| 565 | // Natural-sized immediates. |
| 566 | #if defined(TARGET_ARCH_X64) |
| 567 | if (value < 0 && signed_value) { |
| 568 | Print("-%#" Px64, -value); |
| 569 | } else { |
| 570 | Print("%#" Px64, value); |
| 571 | } |
| 572 | #else |
| 573 | int32_t v32 = static_cast<int32_t>(value); |
| 574 | if (v32 < 0 && signed_value) { |
| 575 | Print("-%#" Px32, -v32); |
| 576 | } else { |
| 577 | Print("%#" Px32, v32); |
| 578 | } |
| 579 | #endif |
| 580 | } |
| 581 | } |
| 582 | } |
| 583 | |
| 584 | void DisassemblerX64::PrintDisp(int disp, const char* after) { |
| 585 | if (-disp > 0) { |
| 586 | Print("-%#x" , -disp); |
| 587 | } else { |
| 588 | Print("+%#x" , disp); |
| 589 | } |
| 590 | if (after != NULL) Print("%s" , after); |
| 591 | } |
| 592 | |
| 593 | // Returns number of bytes used by machine instruction, including *data byte. |
| 594 | // Writes immediate instructions to 'tmp_buffer_'. |
| 595 | int DisassemblerX64::PrintImmediateOp(uint8_t* data) { |
| 596 | bool byte_size_immediate = (*data & 0x03) != 1; |
| 597 | uint8_t modrm = *(data + 1); |
| 598 | int mod, regop, rm; |
| 599 | get_modrm(modrm, &mod, ®op, &rm); |
| 600 | const char* mnem = "Imm???" ; |
| 601 | switch (regop) { |
| 602 | case 0: |
| 603 | mnem = "add" ; |
| 604 | break; |
| 605 | case 1: |
| 606 | mnem = "or" ; |
| 607 | break; |
| 608 | case 2: |
| 609 | mnem = "adc" ; |
| 610 | break; |
| 611 | case 3: |
| 612 | mnem = "sbb" ; |
| 613 | break; |
| 614 | case 4: |
| 615 | mnem = "and" ; |
| 616 | break; |
| 617 | case 5: |
| 618 | mnem = "sub" ; |
| 619 | break; |
| 620 | case 6: |
| 621 | mnem = "xor" ; |
| 622 | break; |
| 623 | case 7: |
| 624 | mnem = "cmp" ; |
| 625 | break; |
| 626 | default: |
| 627 | UnimplementedInstruction(); |
| 628 | } |
| 629 | Print("%s%s " , mnem, operand_size_code()); |
| 630 | int count = PrintRightOperand(data + 1); |
| 631 | Print("," ); |
| 632 | OperandSize immediate_size = byte_size_immediate ? BYTE_SIZE : operand_size(); |
| 633 | count += |
| 634 | PrintImmediate(data + 1 + count, immediate_size, byte_size_immediate); |
| 635 | return 1 + count; |
| 636 | } |
| 637 | |
| 638 | // Returns number of bytes used, including *data. |
| 639 | int DisassemblerX64::F6F7Instruction(uint8_t* data) { |
| 640 | ASSERT(*data == 0xF7 || *data == 0xF6); |
| 641 | uint8_t modrm = *(data + 1); |
| 642 | int mod, regop, rm; |
| 643 | get_modrm(modrm, &mod, ®op, &rm); |
| 644 | static const char* mnemonics[] = {"test" , NULL, "not" , "neg" , |
| 645 | "mul" , "imul" , "div" , "idiv" }; |
| 646 | const char* mnem = mnemonics[regop]; |
| 647 | if (mod == 3 && regop != 0) { |
| 648 | if (regop > 3) { |
| 649 | // These are instructions like idiv that implicitly use EAX and EDX as a |
| 650 | // source and destination. We make this explicit in the disassembly. |
| 651 | Print("%s%s (%s,%s),%s" , mnem, operand_size_code(), Rax(), |
| 652 | NameOfCPURegister(2), NameOfCPURegister(rm)); |
| 653 | } else { |
| 654 | Print("%s%s %s" , mnem, operand_size_code(), NameOfCPURegister(rm)); |
| 655 | } |
| 656 | return 2; |
| 657 | } else if (regop == 0) { |
| 658 | Print("test%s " , operand_size_code()); |
| 659 | int count = PrintRightOperand(data + 1); // Use name of 64-bit register. |
| 660 | Print("," ); |
| 661 | count += PrintImmediate(data + 1 + count, operand_size()); |
| 662 | return 1 + count; |
| 663 | } else if (regop >= 4) { |
| 664 | Print("%s%s (%s,%s)," , mnem, operand_size_code(), Rax(), |
| 665 | NameOfCPURegister(2)); |
| 666 | return 1 + PrintRightOperand(data + 1); |
| 667 | } else { |
| 668 | UnimplementedInstruction(); |
| 669 | return 2; |
| 670 | } |
| 671 | } |
| 672 | |
| 673 | int DisassemblerX64::ShiftInstruction(uint8_t* data) { |
| 674 | // C0/C1: Shift Imm8 |
| 675 | // D0/D1: Shift 1 |
| 676 | // D2/D3: Shift CL |
| 677 | uint8_t op = *data & (~1); |
| 678 | if (op != 0xD0 && op != 0xD2 && op != 0xC0) { |
| 679 | UnimplementedInstruction(); |
| 680 | return 1; |
| 681 | } |
| 682 | uint8_t* modrm = data + 1; |
| 683 | int mod, regop, rm; |
| 684 | get_modrm(*modrm, &mod, ®op, &rm); |
| 685 | regop &= 0x7; // The REX.R bit does not affect the operation. |
| 686 | int num_bytes = 1; |
| 687 | const char* mnem = NULL; |
| 688 | switch (regop) { |
| 689 | case 0: |
| 690 | mnem = "rol" ; |
| 691 | break; |
| 692 | case 1: |
| 693 | mnem = "ror" ; |
| 694 | break; |
| 695 | case 2: |
| 696 | mnem = "rcl" ; |
| 697 | break; |
| 698 | case 3: |
| 699 | mnem = "rcr" ; |
| 700 | break; |
| 701 | case 4: |
| 702 | mnem = "shl" ; |
| 703 | break; |
| 704 | case 5: |
| 705 | mnem = "shr" ; |
| 706 | break; |
| 707 | case 7: |
| 708 | mnem = "sar" ; |
| 709 | break; |
| 710 | default: |
| 711 | UnimplementedInstruction(); |
| 712 | return num_bytes; |
| 713 | } |
| 714 | ASSERT(NULL != mnem); |
| 715 | Print("%s%s " , mnem, operand_size_code()); |
| 716 | if (byte_size_operand_) { |
| 717 | num_bytes += PrintRightByteOperand(modrm); |
| 718 | } else { |
| 719 | num_bytes += PrintRightOperand(modrm); |
| 720 | } |
| 721 | |
| 722 | if (op == 0xD0) { |
| 723 | Print(",1" ); |
| 724 | } else if (op == 0xC0) { |
| 725 | uint8_t imm8 = *(data + num_bytes); |
| 726 | Print(",%d" , imm8); |
| 727 | num_bytes++; |
| 728 | } else { |
| 729 | ASSERT(op == 0xD2); |
| 730 | Print(",cl" ); |
| 731 | } |
| 732 | return num_bytes; |
| 733 | } |
| 734 | |
| 735 | int DisassemblerX64::PrintRightOperand(uint8_t* modrmp) { |
| 736 | return PrintRightOperandHelper(modrmp, &DisassemblerX64::NameOfCPURegister); |
| 737 | } |
| 738 | |
| 739 | int DisassemblerX64::PrintRightByteOperand(uint8_t* modrmp) { |
| 740 | return PrintRightOperandHelper(modrmp, |
| 741 | &DisassemblerX64::NameOfByteCPURegister); |
| 742 | } |
| 743 | |
| 744 | int DisassemblerX64::PrintRightXMMOperand(uint8_t* modrmp) { |
| 745 | return PrintRightOperandHelper(modrmp, &DisassemblerX64::NameOfXMMRegister); |
| 746 | } |
| 747 | |
| 748 | // Returns number of bytes used including the current *data. |
| 749 | // Writes instruction's mnemonic, left and right operands to 'tmp_buffer_'. |
| 750 | int DisassemblerX64::PrintOperands(const char* mnem, |
| 751 | OperandType op_order, |
| 752 | uint8_t* data) { |
| 753 | uint8_t modrm = *data; |
| 754 | int mod, regop, rm; |
| 755 | get_modrm(modrm, &mod, ®op, &rm); |
| 756 | int advance = 0; |
| 757 | const char* register_name = byte_size_operand_ ? NameOfByteCPURegister(regop) |
| 758 | : NameOfCPURegister(regop); |
| 759 | switch (op_order) { |
| 760 | case REG_OPER_OP_ORDER: { |
| 761 | Print("%s%s %s," , mnem, operand_size_code(), register_name); |
| 762 | advance = byte_size_operand_ ? PrintRightByteOperand(data) |
| 763 | : PrintRightOperand(data); |
| 764 | break; |
| 765 | } |
| 766 | case OPER_REG_OP_ORDER: { |
| 767 | Print("%s%s " , mnem, operand_size_code()); |
| 768 | advance = byte_size_operand_ ? PrintRightByteOperand(data) |
| 769 | : PrintRightOperand(data); |
| 770 | Print(",%s" , register_name); |
| 771 | break; |
| 772 | } |
| 773 | default: |
| 774 | UNREACHABLE(); |
| 775 | break; |
| 776 | } |
| 777 | return advance; |
| 778 | } |
| 779 | |
| 780 | void DisassemblerX64::PrintJump(uint8_t* pc, int32_t disp) { |
| 781 | if (FLAG_disassemble_relative) { |
| 782 | Print("%+d" , disp); |
| 783 | } else { |
| 784 | PrintAddress( |
| 785 | reinterpret_cast<uint8_t*>(reinterpret_cast<uintptr_t>(pc) + disp)); |
| 786 | } |
| 787 | } |
| 788 | |
| 789 | void DisassemblerX64::PrintAddress(uint8_t* addr_byte_ptr) { |
| 790 | #if defined(TARGET_ARCH_X64) |
| 791 | Print("%#018" Px64 "" , reinterpret_cast<uint64_t>(addr_byte_ptr)); |
| 792 | #else |
| 793 | Print("%#010" Px32 "" , reinterpret_cast<uint32_t>(addr_byte_ptr)); |
| 794 | #endif |
| 795 | |
| 796 | // Try to print as stub name. |
| 797 | uword addr = reinterpret_cast<uword>(addr_byte_ptr); |
| 798 | const char* name_of_stub = StubCode::NameOfStub(addr); |
| 799 | if (name_of_stub != NULL) { |
| 800 | Print(" [stub: %s]" , name_of_stub); |
| 801 | } |
| 802 | } |
| 803 | // Returns number of bytes used, including *data. |
| 804 | int DisassemblerX64::JumpShort(uint8_t* data) { |
| 805 | ASSERT(0xEB == *data); |
| 806 | uint8_t b = *(data + 1); |
| 807 | int32_t disp = static_cast<int8_t>(b) + 2; |
| 808 | Print("jmp " ); |
| 809 | PrintJump(data, disp); |
| 810 | return 2; |
| 811 | } |
| 812 | |
| 813 | // Returns number of bytes used, including *data. |
| 814 | int DisassemblerX64::JumpConditional(uint8_t* data) { |
| 815 | ASSERT(0x0F == *data); |
| 816 | uint8_t cond = *(data + 1) & 0x0F; |
| 817 | int32_t disp = LoadUnaligned(reinterpret_cast<int32_t*>(data + 2)) + 6; |
| 818 | const char* mnem = conditional_code_suffix[cond]; |
| 819 | Print("j%s " , mnem); |
| 820 | PrintJump(data, disp); |
| 821 | return 6; // includes 0x0F |
| 822 | } |
| 823 | |
| 824 | // Returns number of bytes used, including *data. |
| 825 | int DisassemblerX64::JumpConditionalShort(uint8_t* data) { |
| 826 | uint8_t cond = *data & 0x0F; |
| 827 | uint8_t b = *(data + 1); |
| 828 | int32_t disp = static_cast<int8_t>(b) + 2; |
| 829 | const char* mnem = conditional_code_suffix[cond]; |
| 830 | Print("j%s " , mnem); |
| 831 | PrintJump(data, disp); |
| 832 | return 2; |
| 833 | } |
| 834 | |
| 835 | // Returns number of bytes used, including *data. |
| 836 | int DisassemblerX64::SetCC(uint8_t* data) { |
| 837 | ASSERT(0x0F == *data); |
| 838 | uint8_t cond = *(data + 1) & 0x0F; |
| 839 | const char* mnem = conditional_code_suffix[cond]; |
| 840 | Print("set%s%s " , mnem, operand_size_code()); |
| 841 | PrintRightByteOperand(data + 2); |
| 842 | return 3; // includes 0x0F |
| 843 | } |
| 844 | |
| 845 | // Returns number of bytes used, including *data. |
| 846 | int DisassemblerX64::FPUInstruction(uint8_t* data) { |
| 847 | uint8_t escape_opcode = *data; |
| 848 | ASSERT(0xD8 == (escape_opcode & 0xF8)); |
| 849 | uint8_t modrm_byte = *(data + 1); |
| 850 | |
| 851 | if (modrm_byte >= 0xC0) { |
| 852 | return RegisterFPUInstruction(escape_opcode, modrm_byte); |
| 853 | } else { |
| 854 | return MemoryFPUInstruction(escape_opcode, modrm_byte, data + 1); |
| 855 | } |
| 856 | } |
| 857 | |
| 858 | int DisassemblerX64::MemoryFPUInstruction(int escape_opcode, |
| 859 | int modrm_byte, |
| 860 | uint8_t* modrm_start) { |
| 861 | const char* mnem = "?" ; |
| 862 | int regop = (modrm_byte >> 3) & 0x7; // reg/op field of modrm byte. |
| 863 | switch (escape_opcode) { |
| 864 | case 0xD9: |
| 865 | switch (regop) { |
| 866 | case 0: |
| 867 | mnem = "fld_s" ; |
| 868 | break; |
| 869 | case 3: |
| 870 | mnem = "fstp_s" ; |
| 871 | break; |
| 872 | case 5: |
| 873 | mnem = "fldcw" ; |
| 874 | break; |
| 875 | case 7: |
| 876 | mnem = "fnstcw" ; |
| 877 | break; |
| 878 | default: |
| 879 | UnimplementedInstruction(); |
| 880 | } |
| 881 | break; |
| 882 | |
| 883 | case 0xDB: |
| 884 | switch (regop) { |
| 885 | case 0: |
| 886 | mnem = "fild_s" ; |
| 887 | break; |
| 888 | case 1: |
| 889 | mnem = "fisttp_s" ; |
| 890 | break; |
| 891 | case 2: |
| 892 | mnem = "fist_s" ; |
| 893 | break; |
| 894 | case 3: |
| 895 | mnem = "fistp_s" ; |
| 896 | break; |
| 897 | default: |
| 898 | UnimplementedInstruction(); |
| 899 | } |
| 900 | break; |
| 901 | |
| 902 | case 0xDD: |
| 903 | switch (regop) { |
| 904 | case 0: |
| 905 | mnem = "fld_d" ; |
| 906 | break; |
| 907 | case 3: |
| 908 | mnem = "fstp_d" ; |
| 909 | break; |
| 910 | default: |
| 911 | UnimplementedInstruction(); |
| 912 | } |
| 913 | break; |
| 914 | |
| 915 | case 0xDF: |
| 916 | switch (regop) { |
| 917 | case 5: |
| 918 | mnem = "fild_d" ; |
| 919 | break; |
| 920 | case 7: |
| 921 | mnem = "fistp_d" ; |
| 922 | break; |
| 923 | default: |
| 924 | UnimplementedInstruction(); |
| 925 | } |
| 926 | break; |
| 927 | |
| 928 | default: |
| 929 | UnimplementedInstruction(); |
| 930 | } |
| 931 | Print("%s " , mnem); |
| 932 | int count = PrintRightOperand(modrm_start); |
| 933 | return count + 1; |
| 934 | } |
| 935 | |
| 936 | int DisassemblerX64::RegisterFPUInstruction(int escape_opcode, |
| 937 | uint8_t modrm_byte) { |
| 938 | bool has_register = false; // Is the FPU register encoded in modrm_byte? |
| 939 | const char* mnem = "?" ; |
| 940 | |
| 941 | switch (escape_opcode) { |
| 942 | case 0xD8: |
| 943 | UnimplementedInstruction(); |
| 944 | break; |
| 945 | |
| 946 | case 0xD9: |
| 947 | switch (modrm_byte & 0xF8) { |
| 948 | case 0xC0: |
| 949 | mnem = "fld" ; |
| 950 | has_register = true; |
| 951 | break; |
| 952 | case 0xC8: |
| 953 | mnem = "fxch" ; |
| 954 | has_register = true; |
| 955 | break; |
| 956 | default: |
| 957 | switch (modrm_byte) { |
| 958 | case 0xE0: |
| 959 | mnem = "fchs" ; |
| 960 | break; |
| 961 | case 0xE1: |
| 962 | mnem = "fabs" ; |
| 963 | break; |
| 964 | case 0xE3: |
| 965 | mnem = "fninit" ; |
| 966 | break; |
| 967 | case 0xE4: |
| 968 | mnem = "ftst" ; |
| 969 | break; |
| 970 | case 0xE8: |
| 971 | mnem = "fld1" ; |
| 972 | break; |
| 973 | case 0xEB: |
| 974 | mnem = "fldpi" ; |
| 975 | break; |
| 976 | case 0xED: |
| 977 | mnem = "fldln2" ; |
| 978 | break; |
| 979 | case 0xEE: |
| 980 | mnem = "fldz" ; |
| 981 | break; |
| 982 | case 0xF0: |
| 983 | mnem = "f2xm1" ; |
| 984 | break; |
| 985 | case 0xF1: |
| 986 | mnem = "fyl2x" ; |
| 987 | break; |
| 988 | case 0xF2: |
| 989 | mnem = "fptan" ; |
| 990 | break; |
| 991 | case 0xF5: |
| 992 | mnem = "fprem1" ; |
| 993 | break; |
| 994 | case 0xF7: |
| 995 | mnem = "fincstp" ; |
| 996 | break; |
| 997 | case 0xF8: |
| 998 | mnem = "fprem" ; |
| 999 | break; |
| 1000 | case 0xFB: |
| 1001 | mnem = "fsincos" ; |
| 1002 | break; |
| 1003 | case 0xFD: |
| 1004 | mnem = "fscale" ; |
| 1005 | break; |
| 1006 | case 0xFE: |
| 1007 | mnem = "fsin" ; |
| 1008 | break; |
| 1009 | case 0xFF: |
| 1010 | mnem = "fcos" ; |
| 1011 | break; |
| 1012 | default: |
| 1013 | UnimplementedInstruction(); |
| 1014 | } |
| 1015 | } |
| 1016 | break; |
| 1017 | |
| 1018 | case 0xDA: |
| 1019 | if (modrm_byte == 0xE9) { |
| 1020 | mnem = "fucompp" ; |
| 1021 | } else { |
| 1022 | UnimplementedInstruction(); |
| 1023 | } |
| 1024 | break; |
| 1025 | |
| 1026 | case 0xDB: |
| 1027 | if ((modrm_byte & 0xF8) == 0xE8) { |
| 1028 | mnem = "fucomi" ; |
| 1029 | has_register = true; |
| 1030 | } else if (modrm_byte == 0xE2) { |
| 1031 | mnem = "fclex" ; |
| 1032 | } else { |
| 1033 | UnimplementedInstruction(); |
| 1034 | } |
| 1035 | break; |
| 1036 | |
| 1037 | case 0xDC: |
| 1038 | has_register = true; |
| 1039 | switch (modrm_byte & 0xF8) { |
| 1040 | case 0xC0: |
| 1041 | mnem = "fadd" ; |
| 1042 | break; |
| 1043 | case 0xE8: |
| 1044 | mnem = "fsub" ; |
| 1045 | break; |
| 1046 | case 0xC8: |
| 1047 | mnem = "fmul" ; |
| 1048 | break; |
| 1049 | case 0xF8: |
| 1050 | mnem = "fdiv" ; |
| 1051 | break; |
| 1052 | default: |
| 1053 | UnimplementedInstruction(); |
| 1054 | } |
| 1055 | break; |
| 1056 | |
| 1057 | case 0xDD: |
| 1058 | has_register = true; |
| 1059 | switch (modrm_byte & 0xF8) { |
| 1060 | case 0xC0: |
| 1061 | mnem = "ffree" ; |
| 1062 | break; |
| 1063 | case 0xD8: |
| 1064 | mnem = "fstp" ; |
| 1065 | break; |
| 1066 | default: |
| 1067 | UnimplementedInstruction(); |
| 1068 | } |
| 1069 | break; |
| 1070 | |
| 1071 | case 0xDE: |
| 1072 | if (modrm_byte == 0xD9) { |
| 1073 | mnem = "fcompp" ; |
| 1074 | } else { |
| 1075 | has_register = true; |
| 1076 | switch (modrm_byte & 0xF8) { |
| 1077 | case 0xC0: |
| 1078 | mnem = "faddp" ; |
| 1079 | break; |
| 1080 | case 0xE8: |
| 1081 | mnem = "fsubp" ; |
| 1082 | break; |
| 1083 | case 0xC8: |
| 1084 | mnem = "fmulp" ; |
| 1085 | break; |
| 1086 | case 0xF8: |
| 1087 | mnem = "fdivp" ; |
| 1088 | break; |
| 1089 | default: |
| 1090 | UnimplementedInstruction(); |
| 1091 | } |
| 1092 | } |
| 1093 | break; |
| 1094 | |
| 1095 | case 0xDF: |
| 1096 | if (modrm_byte == 0xE0) { |
| 1097 | mnem = "fnstsw_ax" ; |
| 1098 | } else if ((modrm_byte & 0xF8) == 0xE8) { |
| 1099 | mnem = "fucomip" ; |
| 1100 | has_register = true; |
| 1101 | } |
| 1102 | break; |
| 1103 | |
| 1104 | default: |
| 1105 | UnimplementedInstruction(); |
| 1106 | } |
| 1107 | |
| 1108 | if (has_register) { |
| 1109 | Print("%s st%d" , mnem, modrm_byte & 0x7); |
| 1110 | } else { |
| 1111 | Print("%s" , mnem); |
| 1112 | } |
| 1113 | return 2; |
| 1114 | } |
| 1115 | |
| 1116 | // TODO(srdjan): Should we add a branch hint argument? |
| 1117 | bool DisassemblerX64::DecodeInstructionType(uint8_t** data) { |
| 1118 | uint8_t current; |
| 1119 | |
| 1120 | // Scan for prefixes. |
| 1121 | while (true) { |
| 1122 | current = **data; |
| 1123 | if (current == OPERAND_SIZE_OVERRIDE_PREFIX) { // Group 3 prefix. |
| 1124 | operand_size_ = current; |
| 1125 | #if defined(TARGET_ARCH_X64) |
| 1126 | } else if ((current & 0xF0) == 0x40) { |
| 1127 | // REX prefix. |
| 1128 | setRex(current); |
| 1129 | // TODO(srdjan): Should we enable printing of REX.W? |
| 1130 | // if (rex_w()) Print("REX.W "); |
| 1131 | #endif |
| 1132 | } else if ((current & 0xFE) == 0xF2) { // Group 1 prefix (0xF2 or 0xF3). |
| 1133 | group_1_prefix_ = current; |
| 1134 | } else if (current == 0xF0) { |
| 1135 | Print("lock " ); |
| 1136 | } else { // Not a prefix - an opcode. |
| 1137 | break; |
| 1138 | } |
| 1139 | (*data)++; |
| 1140 | } |
| 1141 | |
| 1142 | const InstructionDesc& idesc = instruction_table.Get(current); |
| 1143 | byte_size_operand_ = idesc.byte_size_operation; |
| 1144 | |
| 1145 | switch (idesc.type) { |
| 1146 | case ZERO_OPERANDS_INSTR: |
| 1147 | if (current >= 0xA4 && current <= 0xA7) { |
| 1148 | // String move or compare operations. |
| 1149 | if (group_1_prefix_ == REP_PREFIX) { |
| 1150 | // REP. |
| 1151 | Print("rep " ); |
| 1152 | } |
| 1153 | if ((current & 0x01) == 0x01) { |
| 1154 | // Operation size: word, dword or qword |
| 1155 | switch (operand_size()) { |
| 1156 | case WORD_SIZE: |
| 1157 | Print("%sw" , idesc.mnem); |
| 1158 | break; |
| 1159 | case DOUBLEWORD_SIZE: |
| 1160 | Print("%sl" , idesc.mnem); |
| 1161 | break; |
| 1162 | case QUADWORD_SIZE: |
| 1163 | Print("%sq" , idesc.mnem); |
| 1164 | break; |
| 1165 | default: |
| 1166 | UNREACHABLE(); |
| 1167 | } |
| 1168 | } else { |
| 1169 | // Operation size: byte |
| 1170 | Print("%s" , idesc.mnem); |
| 1171 | } |
| 1172 | } else if (current == 0x99 && rex_w()) { |
| 1173 | Print("cqo" ); // Cdql is called cdq and cdqq is called cqo. |
| 1174 | } else { |
| 1175 | Print("%s" , idesc.mnem); |
| 1176 | } |
| 1177 | (*data)++; |
| 1178 | break; |
| 1179 | |
| 1180 | case TWO_OPERANDS_INSTR: |
| 1181 | (*data)++; |
| 1182 | (*data) += PrintOperands(idesc.mnem, idesc.op_order_, *data); |
| 1183 | break; |
| 1184 | |
| 1185 | case JUMP_CONDITIONAL_SHORT_INSTR: |
| 1186 | (*data) += JumpConditionalShort(*data); |
| 1187 | break; |
| 1188 | |
| 1189 | case REGISTER_INSTR: |
| 1190 | Print("%s%s %s" , idesc.mnem, operand_size_code(), |
| 1191 | NameOfCPURegister(base_reg(current & 0x07))); |
| 1192 | (*data)++; |
| 1193 | break; |
| 1194 | case PUSHPOP_INSTR: |
| 1195 | Print("%s %s" , idesc.mnem, NameOfCPURegister(base_reg(current & 0x07))); |
| 1196 | (*data)++; |
| 1197 | break; |
| 1198 | case MOVE_REG_INSTR: { |
| 1199 | intptr_t addr = 0; |
| 1200 | int imm_bytes = 0; |
| 1201 | switch (operand_size()) { |
| 1202 | case WORD_SIZE: |
| 1203 | addr = LoadUnaligned(reinterpret_cast<int16_t*>(*data + 1)); |
| 1204 | imm_bytes = 2; |
| 1205 | break; |
| 1206 | case DOUBLEWORD_SIZE: |
| 1207 | addr = LoadUnaligned(reinterpret_cast<int32_t*>(*data + 1)); |
| 1208 | imm_bytes = 4; |
| 1209 | break; |
| 1210 | case QUADWORD_SIZE: |
| 1211 | addr = LoadUnaligned(reinterpret_cast<int64_t*>(*data + 1)); |
| 1212 | imm_bytes = 8; |
| 1213 | break; |
| 1214 | default: |
| 1215 | UNREACHABLE(); |
| 1216 | } |
| 1217 | (*data) += 1 + imm_bytes; |
| 1218 | Print("mov%s %s," , operand_size_code(), |
| 1219 | NameOfCPURegister(base_reg(current & 0x07))); |
| 1220 | PrintImmediateValue(addr, /* signed = */ false, imm_bytes); |
| 1221 | break; |
| 1222 | } |
| 1223 | |
| 1224 | case CALL_JUMP_INSTR: { |
| 1225 | int32_t disp = LoadUnaligned(reinterpret_cast<int32_t*>(*data + 1)) + 5; |
| 1226 | Print("%s " , idesc.mnem); |
| 1227 | PrintJump(*data, disp); |
| 1228 | (*data) += 5; |
| 1229 | break; |
| 1230 | } |
| 1231 | |
| 1232 | case SHORT_IMMEDIATE_INSTR: { |
| 1233 | Print("%s%s %s," , idesc.mnem, operand_size_code(), Rax()); |
| 1234 | PrintImmediate(*data + 1, DOUBLEWORD_SIZE); |
| 1235 | (*data) += 5; |
| 1236 | break; |
| 1237 | } |
| 1238 | |
| 1239 | case NO_INSTR: |
| 1240 | return false; |
| 1241 | |
| 1242 | default: |
| 1243 | UNIMPLEMENTED(); // This type is not implemented. |
| 1244 | } |
| 1245 | return true; |
| 1246 | } |
| 1247 | |
| 1248 | int DisassemblerX64::Print660F38Instruction(uint8_t* current) { |
| 1249 | int mod, regop, rm; |
| 1250 | if (*current == 0x25) { |
| 1251 | get_modrm(*(current + 1), &mod, ®op, &rm); |
| 1252 | Print("pmovsxdq %s," , NameOfXMMRegister(regop)); |
| 1253 | return 1 + PrintRightXMMOperand(current + 1); |
| 1254 | } else if (*current == 0x29) { |
| 1255 | get_modrm(*(current + 1), &mod, ®op, &rm); |
| 1256 | Print("pcmpeqq %s," , NameOfXMMRegister(regop)); |
| 1257 | return 1 + PrintRightXMMOperand(current + 1); |
| 1258 | } else { |
| 1259 | UnimplementedInstruction(); |
| 1260 | return 1; |
| 1261 | } |
| 1262 | } |
| 1263 | |
| 1264 | // Handle all two-byte opcodes, which start with 0x0F. |
| 1265 | // These instructions may be affected by an 0x66, 0xF2, or 0xF3 prefix. |
| 1266 | // We do not use any three-byte opcodes, which start with 0x0F38 or 0x0F3A. |
| 1267 | int DisassemblerX64::TwoByteOpcodeInstruction(uint8_t* data) { |
| 1268 | uint8_t opcode = *(data + 1); |
| 1269 | uint8_t* current = data + 2; |
| 1270 | // At return, "current" points to the start of the next instruction. |
| 1271 | const char* mnemonic = TwoByteMnemonic(opcode); |
| 1272 | if (operand_size_ == 0x66) { |
| 1273 | // 0x66 0x0F prefix. |
| 1274 | int mod, regop, rm; |
| 1275 | if (opcode == 0xC6) { |
| 1276 | int mod, regop, rm; |
| 1277 | get_modrm(*current, &mod, ®op, &rm); |
| 1278 | Print("shufpd %s, " , NameOfXMMRegister(regop)); |
| 1279 | current += PrintRightXMMOperand(current); |
| 1280 | Print(" [%x]" , *current); |
| 1281 | current++; |
| 1282 | } else if (opcode == 0x3A) { |
| 1283 | uint8_t third_byte = *current; |
| 1284 | current = data + 3; |
| 1285 | if (third_byte == 0x16) { |
| 1286 | get_modrm(*current, &mod, ®op, &rm); |
| 1287 | Print("pextrd " ); // reg/m32, xmm, imm8 |
| 1288 | current += PrintRightOperand(current); |
| 1289 | Print(",%s,%d" , NameOfXMMRegister(regop), (*current) & 7); |
| 1290 | current += 1; |
| 1291 | } else if (third_byte == 0x17) { |
| 1292 | get_modrm(*current, &mod, ®op, &rm); |
| 1293 | Print("extractps " ); // reg/m32, xmm, imm8 |
| 1294 | current += PrintRightOperand(current); |
| 1295 | Print(", %s, %d" , NameOfCPURegister(regop), (*current) & 3); |
| 1296 | current += 1; |
| 1297 | } else if (third_byte == 0x0b) { |
| 1298 | get_modrm(*current, &mod, ®op, &rm); |
| 1299 | // roundsd xmm, xmm/m64, imm8 |
| 1300 | Print("roundsd %s, " , NameOfCPURegister(regop)); |
| 1301 | current += PrintRightOperand(current); |
| 1302 | Print(", %d" , (*current) & 3); |
| 1303 | current += 1; |
| 1304 | } else { |
| 1305 | UnimplementedInstruction(); |
| 1306 | } |
| 1307 | } else { |
| 1308 | get_modrm(*current, &mod, ®op, &rm); |
| 1309 | if (opcode == 0x1f) { |
| 1310 | current++; |
| 1311 | if (rm == 4) { // SIB byte present. |
| 1312 | current++; |
| 1313 | } |
| 1314 | if (mod == 1) { // Byte displacement. |
| 1315 | current += 1; |
| 1316 | } else if (mod == 2) { // 32-bit displacement. |
| 1317 | current += 4; |
| 1318 | } // else no immediate displacement. |
| 1319 | Print("nop" ); |
| 1320 | } else if (opcode == 0x28) { |
| 1321 | Print("movapd %s, " , NameOfXMMRegister(regop)); |
| 1322 | current += PrintRightXMMOperand(current); |
| 1323 | } else if (opcode == 0x29) { |
| 1324 | Print("movapd " ); |
| 1325 | current += PrintRightXMMOperand(current); |
| 1326 | Print(", %s" , NameOfXMMRegister(regop)); |
| 1327 | } else if (opcode == 0x38) { |
| 1328 | current += Print660F38Instruction(current); |
| 1329 | } else if (opcode == 0x6E) { |
| 1330 | Print("mov%c %s," , rex_w() ? 'q' : 'd', NameOfXMMRegister(regop)); |
| 1331 | current += PrintRightOperand(current); |
| 1332 | } else if (opcode == 0x6F) { |
| 1333 | Print("movdqa %s," , NameOfXMMRegister(regop)); |
| 1334 | current += PrintRightXMMOperand(current); |
| 1335 | } else if (opcode == 0x7E) { |
| 1336 | Print("mov%c " , rex_w() ? 'q' : 'd'); |
| 1337 | current += PrintRightOperand(current); |
| 1338 | Print(",%s" , NameOfXMMRegister(regop)); |
| 1339 | } else if (opcode == 0x7F) { |
| 1340 | Print("movdqa " ); |
| 1341 | current += PrintRightXMMOperand(current); |
| 1342 | Print(",%s" , NameOfXMMRegister(regop)); |
| 1343 | } else if (opcode == 0xD6) { |
| 1344 | Print("movq " ); |
| 1345 | current += PrintRightXMMOperand(current); |
| 1346 | Print(",%s" , NameOfXMMRegister(regop)); |
| 1347 | } else if (opcode == 0x50) { |
| 1348 | Print("movmskpd %s," , NameOfCPURegister(regop)); |
| 1349 | current += PrintRightXMMOperand(current); |
| 1350 | } else if (opcode == 0xD7) { |
| 1351 | Print("pmovmskb %s," , NameOfCPURegister(regop)); |
| 1352 | current += PrintRightXMMOperand(current); |
| 1353 | } else { |
| 1354 | const char* mnemonic = "?" ; |
| 1355 | if (opcode == 0x5A) { |
| 1356 | mnemonic = "cvtpd2ps" ; |
| 1357 | } else if (0x51 <= opcode && opcode <= 0x5F) { |
| 1358 | mnemonic = xmm_instructions[opcode & 0xF].pd_name; |
| 1359 | } else if (opcode == 0x14) { |
| 1360 | mnemonic = "unpcklpd" ; |
| 1361 | } else if (opcode == 0x15) { |
| 1362 | mnemonic = "unpckhpd" ; |
| 1363 | } else if (opcode == 0x2E) { |
| 1364 | mnemonic = "ucomisd" ; |
| 1365 | } else if (opcode == 0x2F) { |
| 1366 | mnemonic = "comisd" ; |
| 1367 | } else if (opcode == 0xFE) { |
| 1368 | mnemonic = "paddd" ; |
| 1369 | } else if (opcode == 0xFA) { |
| 1370 | mnemonic = "psubd" ; |
| 1371 | } else if (opcode == 0xEF) { |
| 1372 | mnemonic = "pxor" ; |
| 1373 | } else { |
| 1374 | UnimplementedInstruction(); |
| 1375 | } |
| 1376 | Print("%s %s," , mnemonic, NameOfXMMRegister(regop)); |
| 1377 | current += PrintRightXMMOperand(current); |
| 1378 | } |
| 1379 | } |
| 1380 | } else if (group_1_prefix_ == 0xF2) { |
| 1381 | // Beginning of instructions with prefix 0xF2. |
| 1382 | |
| 1383 | if (opcode == 0x11 || opcode == 0x10) { |
| 1384 | // MOVSD: Move scalar double-precision fp to/from/between XMM registers. |
| 1385 | Print("movsd " ); |
| 1386 | int mod, regop, rm; |
| 1387 | get_modrm(*current, &mod, ®op, &rm); |
| 1388 | if (opcode == 0x11) { |
| 1389 | current += PrintRightXMMOperand(current); |
| 1390 | Print(",%s" , NameOfXMMRegister(regop)); |
| 1391 | } else { |
| 1392 | Print("%s," , NameOfXMMRegister(regop)); |
| 1393 | current += PrintRightXMMOperand(current); |
| 1394 | } |
| 1395 | } else if (opcode == 0x2A) { |
| 1396 | // CVTSI2SD: integer to XMM double conversion. |
| 1397 | int mod, regop, rm; |
| 1398 | get_modrm(*current, &mod, ®op, &rm); |
| 1399 | Print("%sd %s," , mnemonic, NameOfXMMRegister(regop)); |
| 1400 | current += PrintRightOperand(current); |
| 1401 | } else if (opcode == 0x2C) { |
| 1402 | // CVTTSD2SI: |
| 1403 | // Convert with truncation scalar double-precision FP to integer. |
| 1404 | int mod, regop, rm; |
| 1405 | get_modrm(*current, &mod, ®op, &rm); |
| 1406 | Print("cvttsd2si%s %s," , operand_size_code(), NameOfCPURegister(regop)); |
| 1407 | current += PrintRightXMMOperand(current); |
| 1408 | } else if (opcode == 0x2D) { |
| 1409 | // CVTSD2SI: Convert scalar double-precision FP to integer. |
| 1410 | int mod, regop, rm; |
| 1411 | get_modrm(*current, &mod, ®op, &rm); |
| 1412 | Print("cvtsd2si%s %s," , operand_size_code(), NameOfCPURegister(regop)); |
| 1413 | current += PrintRightXMMOperand(current); |
| 1414 | } else if (0x51 <= opcode && opcode <= 0x5F) { |
| 1415 | // XMM arithmetic. Get the F2 0F prefix version of the mnemonic. |
| 1416 | int mod, regop, rm; |
| 1417 | get_modrm(*current, &mod, ®op, &rm); |
| 1418 | const char* mnemonic = |
| 1419 | opcode == 0x5A ? "cvtsd2ss" : xmm_instructions[opcode & 0xF].sd_name; |
| 1420 | Print("%s %s," , mnemonic, NameOfXMMRegister(regop)); |
| 1421 | current += PrintRightXMMOperand(current); |
| 1422 | } else { |
| 1423 | UnimplementedInstruction(); |
| 1424 | } |
| 1425 | } else if (group_1_prefix_ == 0xF3) { |
| 1426 | // Instructions with prefix 0xF3. |
| 1427 | if (opcode == 0x11 || opcode == 0x10) { |
| 1428 | // MOVSS: Move scalar double-precision fp to/from/between XMM registers. |
| 1429 | Print("movss " ); |
| 1430 | int mod, regop, rm; |
| 1431 | get_modrm(*current, &mod, ®op, &rm); |
| 1432 | if (opcode == 0x11) { |
| 1433 | current += PrintRightOperand(current); |
| 1434 | Print(",%s" , NameOfXMMRegister(regop)); |
| 1435 | } else { |
| 1436 | Print("%s," , NameOfXMMRegister(regop)); |
| 1437 | current += PrintRightOperand(current); |
| 1438 | } |
| 1439 | } else if (opcode == 0x2A) { |
| 1440 | // CVTSI2SS: integer to XMM single conversion. |
| 1441 | int mod, regop, rm; |
| 1442 | get_modrm(*current, &mod, ®op, &rm); |
| 1443 | Print("%ss %s," , mnemonic, NameOfXMMRegister(regop)); |
| 1444 | current += PrintRightOperand(current); |
| 1445 | } else if (opcode == 0x2C || opcode == 0x2D) { |
| 1446 | bool truncating = (opcode & 1) == 0; |
| 1447 | // CVTTSS2SI/CVTSS2SI: |
| 1448 | // Convert (with truncation) scalar single-precision FP to dword integer. |
| 1449 | int mod, regop, rm; |
| 1450 | get_modrm(*current, &mod, ®op, &rm); |
| 1451 | Print("cvt%sss2si%s %s," , truncating ? "t" : "" , operand_size_code(), |
| 1452 | NameOfCPURegister(regop)); |
| 1453 | current += PrintRightXMMOperand(current); |
| 1454 | } else if (0x51 <= opcode && opcode <= 0x5F) { |
| 1455 | int mod, regop, rm; |
| 1456 | get_modrm(*current, &mod, ®op, &rm); |
| 1457 | const char* mnemonic = |
| 1458 | opcode == 0x5A ? "cvtss2sd" : xmm_instructions[opcode & 0xF].ss_name; |
| 1459 | Print("%s %s," , mnemonic, NameOfXMMRegister(regop)); |
| 1460 | current += PrintRightXMMOperand(current); |
| 1461 | } else if (opcode == 0x7E) { |
| 1462 | int mod, regop, rm; |
| 1463 | get_modrm(*current, &mod, ®op, &rm); |
| 1464 | Print("movq %s, " , NameOfXMMRegister(regop)); |
| 1465 | current += PrintRightXMMOperand(current); |
| 1466 | } else if (opcode == 0xE6) { |
| 1467 | int mod, regop, rm; |
| 1468 | get_modrm(*current, &mod, ®op, &rm); |
| 1469 | Print("cvtdq2pd %s," , NameOfXMMRegister(regop)); |
| 1470 | current += PrintRightXMMOperand(current); |
| 1471 | } else if (opcode == 0xB8) { |
| 1472 | // POPCNT. |
| 1473 | current += PrintOperands(mnemonic, REG_OPER_OP_ORDER, current); |
| 1474 | } else if (opcode == 0xBD) { |
| 1475 | // LZCNT (rep BSR encoding). |
| 1476 | current += PrintOperands("lzcnt" , REG_OPER_OP_ORDER, current); |
| 1477 | } else { |
| 1478 | UnimplementedInstruction(); |
| 1479 | } |
| 1480 | } else if (opcode == 0x1F) { |
| 1481 | // NOP |
| 1482 | int mod, regop, rm; |
| 1483 | get_modrm(*current, &mod, ®op, &rm); |
| 1484 | current++; |
| 1485 | if (rm == 4) { // SIB byte present. |
| 1486 | current++; |
| 1487 | } |
| 1488 | if (mod == 1) { // Byte displacement. |
| 1489 | current += 1; |
| 1490 | } else if (mod == 2) { // 32-bit displacement. |
| 1491 | current += 4; |
| 1492 | } // else no immediate displacement. |
| 1493 | Print("nop" ); |
| 1494 | |
| 1495 | } else if (opcode == 0x28 || opcode == 0x2f) { |
| 1496 | // ...s xmm, xmm/m128 |
| 1497 | int mod, regop, rm; |
| 1498 | get_modrm(*current, &mod, ®op, &rm); |
| 1499 | const char* mnemonic = opcode == 0x28 ? "movaps" : "comiss" ; |
| 1500 | Print("%s %s," , mnemonic, NameOfXMMRegister(regop)); |
| 1501 | current += PrintRightXMMOperand(current); |
| 1502 | } else if (opcode == 0x29) { |
| 1503 | // movaps xmm/m128, xmm |
| 1504 | int mod, regop, rm; |
| 1505 | get_modrm(*current, &mod, ®op, &rm); |
| 1506 | Print("movaps " ); |
| 1507 | current += PrintRightXMMOperand(current); |
| 1508 | Print(",%s" , NameOfXMMRegister(regop)); |
| 1509 | } else if (opcode == 0x11) { |
| 1510 | // movups xmm/m128, xmm |
| 1511 | int mod, regop, rm; |
| 1512 | get_modrm(*current, &mod, ®op, &rm); |
| 1513 | Print("movups " ); |
| 1514 | current += PrintRightXMMOperand(current); |
| 1515 | Print(",%s" , NameOfXMMRegister(regop)); |
| 1516 | } else if (opcode == 0x50) { |
| 1517 | int mod, regop, rm; |
| 1518 | get_modrm(*current, &mod, ®op, &rm); |
| 1519 | Print("movmskps %s," , NameOfCPURegister(regop)); |
| 1520 | current += PrintRightXMMOperand(current); |
| 1521 | } else if (opcode == 0xA2 || opcode == 0x31) { |
| 1522 | // RDTSC or CPUID |
| 1523 | Print("%s" , mnemonic); |
| 1524 | } else if ((opcode & 0xF0) == 0x40) { |
| 1525 | // CMOVcc: conditional move. |
| 1526 | int condition = opcode & 0x0F; |
| 1527 | const InstructionDesc& idesc = cmov_instructions[condition]; |
| 1528 | byte_size_operand_ = idesc.byte_size_operation; |
| 1529 | current += PrintOperands(idesc.mnem, idesc.op_order_, current); |
| 1530 | } else if (0x10 <= opcode && opcode <= 0x16) { |
| 1531 | // ...ps xmm, xmm/m128 |
| 1532 | static const char* mnemonics[] = {"movups" , NULL, "movhlps" , NULL, |
| 1533 | "unpcklps" , "unpckhps" , "movlhps" }; |
| 1534 | const char* mnemonic = mnemonics[opcode - 0x10]; |
| 1535 | if (mnemonic == NULL) { |
| 1536 | UnimplementedInstruction(); |
| 1537 | mnemonic = "???" ; |
| 1538 | } |
| 1539 | int mod, regop, rm; |
| 1540 | get_modrm(*current, &mod, ®op, &rm); |
| 1541 | Print("%s %s," , mnemonic, NameOfXMMRegister(regop)); |
| 1542 | current += PrintRightXMMOperand(current); |
| 1543 | } else if (0x51 <= opcode && opcode <= 0x5F) { |
| 1544 | int mod, regop, rm; |
| 1545 | get_modrm(*current, &mod, ®op, &rm); |
| 1546 | const char* mnemonic = |
| 1547 | opcode == 0x5A ? "cvtps2pd" : xmm_instructions[opcode & 0xF].ps_name; |
| 1548 | Print("%s %s," , mnemonic, NameOfXMMRegister(regop)); |
| 1549 | current += PrintRightXMMOperand(current); |
| 1550 | } else if (opcode == 0xC2 || opcode == 0xC6) { |
| 1551 | int mod, regop, rm; |
| 1552 | get_modrm(*current, &mod, ®op, &rm); |
| 1553 | if (opcode == 0xC2) { |
| 1554 | Print("cmpps %s," , NameOfXMMRegister(regop)); |
| 1555 | current += PrintRightXMMOperand(current); |
| 1556 | Print(" [%s]" , xmm_conditional_code_suffix[*current]); |
| 1557 | } else { |
| 1558 | ASSERT(opcode == 0xC6); |
| 1559 | Print("shufps %s," , NameOfXMMRegister(regop)); |
| 1560 | current += PrintRightXMMOperand(current); |
| 1561 | Print(" [%x]" , *current); |
| 1562 | } |
| 1563 | current++; |
| 1564 | } else if ((opcode & 0xF0) == 0x80) { |
| 1565 | // Jcc: Conditional jump (branch). |
| 1566 | current = data + JumpConditional(data); |
| 1567 | |
| 1568 | } else if (opcode == 0xBE || opcode == 0xBF || opcode == 0xB6 || |
| 1569 | opcode == 0xB7 || opcode == 0xAF || opcode == 0xB0 || |
| 1570 | opcode == 0xB1 || opcode == 0xBC || opcode == 0xBD) { |
| 1571 | // Size-extending moves, IMUL, cmpxchg, BSF, BSR. |
| 1572 | current += PrintOperands(mnemonic, REG_OPER_OP_ORDER, current); |
| 1573 | } else if ((opcode & 0xF0) == 0x90) { |
| 1574 | // SETcc: Set byte on condition. Needs pointer to beginning of instruction. |
| 1575 | current = data + SetCC(data); |
| 1576 | } else if (((opcode & 0xFE) == 0xA4) || ((opcode & 0xFE) == 0xAC) || |
| 1577 | (opcode == 0xAB) || (opcode == 0xA3)) { |
| 1578 | // SHLD, SHRD (double-prec. shift), BTS (bit test and set), BT (bit test). |
| 1579 | Print("%s%s " , mnemonic, operand_size_code()); |
| 1580 | int mod, regop, rm; |
| 1581 | get_modrm(*current, &mod, ®op, &rm); |
| 1582 | current += PrintRightOperand(current); |
| 1583 | Print(",%s" , NameOfCPURegister(regop)); |
| 1584 | if ((opcode == 0xAB) || (opcode == 0xA3) || (opcode == 0xBD)) { |
| 1585 | // Done. |
| 1586 | } else if ((opcode == 0xA5) || (opcode == 0xAD)) { |
| 1587 | Print(",cl" ); |
| 1588 | } else { |
| 1589 | Print("," ); |
| 1590 | current += PrintImmediate(current, BYTE_SIZE); |
| 1591 | } |
| 1592 | } else if (opcode == 0xBA && (*current & 0x60) == 0x60) { |
| 1593 | // bt? immediate instruction |
| 1594 | int r = (*current >> 3) & 7; |
| 1595 | static const char* const names[4] = {"bt" , "bts" , "btr" , "btc" }; |
| 1596 | Print("%s " , names[r - 4]); |
| 1597 | current += PrintRightOperand(current); |
| 1598 | uint8_t bit = *current++; |
| 1599 | Print(",%d" , bit); |
| 1600 | } else { |
| 1601 | UnimplementedInstruction(); |
| 1602 | } |
| 1603 | return static_cast<int>(current - data); |
| 1604 | } |
| 1605 | |
| 1606 | // Mnemonics for two-byte opcode instructions starting with 0x0F. |
| 1607 | // The argument is the second byte of the two-byte opcode. |
| 1608 | // Returns NULL if the instruction is not handled here. |
| 1609 | const char* DisassemblerX64::TwoByteMnemonic(uint8_t opcode) { |
| 1610 | if (opcode == 0x5A) { |
| 1611 | return "cvtps2pd" ; |
| 1612 | } else if (0x51 <= opcode && opcode <= 0x5F) { |
| 1613 | return xmm_instructions[opcode & 0xF].ps_name; |
| 1614 | } |
| 1615 | if (0xA2 <= opcode && opcode <= 0xBF) { |
| 1616 | static const char* mnemonics[] = { |
| 1617 | "cpuid" , "bt" , "shld" , "shld" , NULL, NULL, |
| 1618 | NULL, NULL, NULL, "bts" , "shrd" , "shrd" , |
| 1619 | NULL, "imul" , "cmpxchg" , "cmpxchg" , NULL, NULL, |
| 1620 | NULL, NULL, "movzxb" , "movzxw" , "popcnt" , NULL, |
| 1621 | NULL, NULL, "bsf" , "bsr" , "movsxb" , "movsxw" }; |
| 1622 | return mnemonics[opcode - 0xA2]; |
| 1623 | } |
| 1624 | switch (opcode) { |
| 1625 | case 0x12: |
| 1626 | return "movhlps" ; |
| 1627 | case 0x16: |
| 1628 | return "movlhps" ; |
| 1629 | case 0x1F: |
| 1630 | return "nop" ; |
| 1631 | case 0x2A: // F2/F3 prefix. |
| 1632 | return "cvtsi2s" ; |
| 1633 | case 0x31: |
| 1634 | return "rdtsc" ; |
| 1635 | default: |
| 1636 | return NULL; |
| 1637 | } |
| 1638 | } |
| 1639 | |
| 1640 | int DisassemblerX64::InstructionDecode(uword pc) { |
| 1641 | uint8_t* data = reinterpret_cast<uint8_t*>(pc); |
| 1642 | |
| 1643 | const bool processed = DecodeInstructionType(&data); |
| 1644 | |
| 1645 | if (!processed) { |
| 1646 | switch (*data) { |
| 1647 | case 0xC2: |
| 1648 | Print("ret " ); |
| 1649 | PrintImmediateValue(*reinterpret_cast<uint16_t*>(data + 1)); |
| 1650 | data += 3; |
| 1651 | break; |
| 1652 | |
| 1653 | case 0xC8: |
| 1654 | Print("enter %d, %d" , *reinterpret_cast<uint16_t*>(data + 1), data[3]); |
| 1655 | data += 4; |
| 1656 | break; |
| 1657 | |
| 1658 | case 0x69: |
| 1659 | FALL_THROUGH; |
| 1660 | case 0x6B: { |
| 1661 | int mod, regop, rm; |
| 1662 | get_modrm(*(data + 1), &mod, ®op, &rm); |
| 1663 | int32_t imm = *data == 0x6B |
| 1664 | ? *(data + 2) |
| 1665 | : LoadUnaligned(reinterpret_cast<int32_t*>(data + 2)); |
| 1666 | Print("imul%s %s,%s," , operand_size_code(), NameOfCPURegister(regop), |
| 1667 | NameOfCPURegister(rm)); |
| 1668 | PrintImmediateValue(imm); |
| 1669 | data += 2 + (*data == 0x6B ? 1 : 4); |
| 1670 | break; |
| 1671 | } |
| 1672 | |
| 1673 | case 0x81: |
| 1674 | FALL_THROUGH; |
| 1675 | case 0x83: // 0x81 with sign extension bit set |
| 1676 | data += PrintImmediateOp(data); |
| 1677 | break; |
| 1678 | |
| 1679 | case 0x0F: |
| 1680 | data += TwoByteOpcodeInstruction(data); |
| 1681 | break; |
| 1682 | |
| 1683 | case 0x8F: { |
| 1684 | data++; |
| 1685 | int mod, regop, rm; |
| 1686 | get_modrm(*data, &mod, ®op, &rm); |
| 1687 | if (regop == 0) { |
| 1688 | Print("pop " ); |
| 1689 | data += PrintRightOperand(data); |
| 1690 | } |
| 1691 | } break; |
| 1692 | |
| 1693 | case 0xFF: { |
| 1694 | data++; |
| 1695 | int mod, regop, rm; |
| 1696 | get_modrm(*data, &mod, ®op, &rm); |
| 1697 | const char* mnem = NULL; |
| 1698 | switch (regop) { |
| 1699 | case 0: |
| 1700 | mnem = "inc" ; |
| 1701 | break; |
| 1702 | case 1: |
| 1703 | mnem = "dec" ; |
| 1704 | break; |
| 1705 | case 2: |
| 1706 | mnem = "call" ; |
| 1707 | break; |
| 1708 | case 4: |
| 1709 | mnem = "jmp" ; |
| 1710 | break; |
| 1711 | case 6: |
| 1712 | mnem = "push" ; |
| 1713 | break; |
| 1714 | default: |
| 1715 | mnem = "???" ; |
| 1716 | } |
| 1717 | if (regop <= 1) { |
| 1718 | Print("%s%s " , mnem, operand_size_code()); |
| 1719 | } else { |
| 1720 | Print("%s " , mnem); |
| 1721 | } |
| 1722 | data += PrintRightOperand(data); |
| 1723 | } break; |
| 1724 | |
| 1725 | case 0xC7: // imm32, fall through |
| 1726 | case 0xC6: // imm8 |
| 1727 | { |
| 1728 | bool is_byte = *data == 0xC6; |
| 1729 | data++; |
| 1730 | if (is_byte) { |
| 1731 | Print("movb " ); |
| 1732 | data += PrintRightByteOperand(data); |
| 1733 | Print("," ); |
| 1734 | data += PrintImmediate(data, BYTE_SIZE); |
| 1735 | } else { |
| 1736 | Print("mov%s " , operand_size_code()); |
| 1737 | data += PrintRightOperand(data); |
| 1738 | Print("," ); |
| 1739 | data += |
| 1740 | PrintImmediate(data, operand_size(), /* sign extend = */ true); |
| 1741 | } |
| 1742 | } break; |
| 1743 | |
| 1744 | case 0x80: { |
| 1745 | byte_size_operand_ = true; |
| 1746 | data += PrintImmediateOp(data); |
| 1747 | } break; |
| 1748 | |
| 1749 | case 0x88: // 8bit, fall through |
| 1750 | case 0x89: // 32bit |
| 1751 | { |
| 1752 | bool is_byte = *data == 0x88; |
| 1753 | int mod, regop, rm; |
| 1754 | data++; |
| 1755 | get_modrm(*data, &mod, ®op, &rm); |
| 1756 | if (is_byte) { |
| 1757 | Print("movb " ); |
| 1758 | data += PrintRightByteOperand(data); |
| 1759 | Print(",%s" , NameOfByteCPURegister(regop)); |
| 1760 | } else { |
| 1761 | Print("mov%s " , operand_size_code()); |
| 1762 | data += PrintRightOperand(data); |
| 1763 | Print(",%s" , NameOfCPURegister(regop)); |
| 1764 | } |
| 1765 | } break; |
| 1766 | |
| 1767 | case 0x90: |
| 1768 | case 0x91: |
| 1769 | case 0x92: |
| 1770 | case 0x93: |
| 1771 | case 0x94: |
| 1772 | case 0x95: |
| 1773 | case 0x96: |
| 1774 | case 0x97: { |
| 1775 | int reg = (*data & 0x7) | (rex_b() ? 8 : 0); |
| 1776 | if (reg == 0) { |
| 1777 | Print("nop" ); // Common name for xchg rax,rax. |
| 1778 | } else { |
| 1779 | Print("xchg%s %s, %s" , operand_size_code(), Rax(), |
| 1780 | NameOfCPURegister(reg)); |
| 1781 | } |
| 1782 | data++; |
| 1783 | } break; |
| 1784 | case 0xB0: |
| 1785 | case 0xB1: |
| 1786 | case 0xB2: |
| 1787 | case 0xB3: |
| 1788 | case 0xB4: |
| 1789 | case 0xB5: |
| 1790 | case 0xB6: |
| 1791 | case 0xB7: |
| 1792 | case 0xB8: |
| 1793 | case 0xB9: |
| 1794 | case 0xBA: |
| 1795 | case 0xBB: |
| 1796 | case 0xBC: |
| 1797 | case 0xBD: |
| 1798 | case 0xBE: |
| 1799 | case 0xBF: { |
| 1800 | // mov reg8,imm8 or mov reg32,imm32 |
| 1801 | uint8_t opcode = *data; |
| 1802 | data++; |
| 1803 | const bool is_not_8bit = opcode >= 0xB8; |
| 1804 | int reg = (opcode & 0x7) | (rex_b() ? 8 : 0); |
| 1805 | if (is_not_8bit) { |
| 1806 | Print("mov%s %s," , operand_size_code(), NameOfCPURegister(reg)); |
| 1807 | data += |
| 1808 | PrintImmediate(data, operand_size(), /* sign extend = */ false); |
| 1809 | } else { |
| 1810 | Print("movb %s," , NameOfByteCPURegister(reg)); |
| 1811 | data += PrintImmediate(data, BYTE_SIZE, /* sign extend = */ false); |
| 1812 | } |
| 1813 | break; |
| 1814 | } |
| 1815 | case 0xFE: { |
| 1816 | data++; |
| 1817 | int mod, regop, rm; |
| 1818 | get_modrm(*data, &mod, ®op, &rm); |
| 1819 | if (regop == 1) { |
| 1820 | Print("decb " ); |
| 1821 | data += PrintRightByteOperand(data); |
| 1822 | } else { |
| 1823 | UnimplementedInstruction(); |
| 1824 | } |
| 1825 | break; |
| 1826 | } |
| 1827 | case 0x68: |
| 1828 | Print("push " ); |
| 1829 | PrintImmediateValue( |
| 1830 | LoadUnaligned(reinterpret_cast<int32_t*>(data + 1))); |
| 1831 | data += 5; |
| 1832 | break; |
| 1833 | |
| 1834 | case 0x6A: |
| 1835 | Print("push " ); |
| 1836 | PrintImmediateValue(*reinterpret_cast<int8_t*>(data + 1)); |
| 1837 | data += 2; |
| 1838 | break; |
| 1839 | |
| 1840 | case 0xA1: |
| 1841 | FALL_THROUGH; |
| 1842 | case 0xA3: |
| 1843 | switch (operand_size()) { |
| 1844 | case DOUBLEWORD_SIZE: { |
| 1845 | PrintAddress(reinterpret_cast<uint8_t*>( |
| 1846 | *reinterpret_cast<int32_t*>(data + 1))); |
| 1847 | if (*data == 0xA1) { // Opcode 0xA1 |
| 1848 | Print("movzxlq %s,(" , Rax()); |
| 1849 | PrintAddress(reinterpret_cast<uint8_t*>( |
| 1850 | *reinterpret_cast<int32_t*>(data + 1))); |
| 1851 | Print(")" ); |
| 1852 | } else { // Opcode 0xA3 |
| 1853 | Print("movzxlq (" ); |
| 1854 | PrintAddress(reinterpret_cast<uint8_t*>( |
| 1855 | *reinterpret_cast<int32_t*>(data + 1))); |
| 1856 | Print("),%s" , Rax()); |
| 1857 | } |
| 1858 | data += 5; |
| 1859 | break; |
| 1860 | } |
| 1861 | case QUADWORD_SIZE: { |
| 1862 | // New x64 instruction mov rax,(imm_64). |
| 1863 | if (*data == 0xA1) { // Opcode 0xA1 |
| 1864 | Print("movq %s,(" , Rax()); |
| 1865 | PrintAddress(*reinterpret_cast<uint8_t**>(data + 1)); |
| 1866 | Print(")" ); |
| 1867 | } else { // Opcode 0xA3 |
| 1868 | Print("movq (" ); |
| 1869 | PrintAddress(*reinterpret_cast<uint8_t**>(data + 1)); |
| 1870 | Print("),%s" , Rax()); |
| 1871 | } |
| 1872 | data += 9; |
| 1873 | break; |
| 1874 | } |
| 1875 | default: |
| 1876 | UnimplementedInstruction(); |
| 1877 | data += 2; |
| 1878 | } |
| 1879 | break; |
| 1880 | |
| 1881 | case 0xA8: |
| 1882 | Print("test al," ); |
| 1883 | PrintImmediateValue(*reinterpret_cast<uint8_t*>(data + 1)); |
| 1884 | data += 2; |
| 1885 | break; |
| 1886 | |
| 1887 | case 0xA9: { |
| 1888 | data++; |
| 1889 | Print("test%s %s," , operand_size_code(), Rax()); |
| 1890 | data += PrintImmediate(data, operand_size()); |
| 1891 | break; |
| 1892 | } |
| 1893 | case 0xD1: |
| 1894 | FALL_THROUGH; |
| 1895 | case 0xD3: |
| 1896 | FALL_THROUGH; |
| 1897 | case 0xC1: |
| 1898 | data += ShiftInstruction(data); |
| 1899 | break; |
| 1900 | case 0xD0: |
| 1901 | FALL_THROUGH; |
| 1902 | case 0xD2: |
| 1903 | FALL_THROUGH; |
| 1904 | case 0xC0: |
| 1905 | byte_size_operand_ = true; |
| 1906 | data += ShiftInstruction(data); |
| 1907 | break; |
| 1908 | |
| 1909 | case 0xD9: |
| 1910 | FALL_THROUGH; |
| 1911 | case 0xDA: |
| 1912 | FALL_THROUGH; |
| 1913 | case 0xDB: |
| 1914 | FALL_THROUGH; |
| 1915 | case 0xDC: |
| 1916 | FALL_THROUGH; |
| 1917 | case 0xDD: |
| 1918 | FALL_THROUGH; |
| 1919 | case 0xDE: |
| 1920 | FALL_THROUGH; |
| 1921 | case 0xDF: |
| 1922 | data += FPUInstruction(data); |
| 1923 | break; |
| 1924 | |
| 1925 | case 0xEB: |
| 1926 | data += JumpShort(data); |
| 1927 | break; |
| 1928 | |
| 1929 | case 0xF6: |
| 1930 | byte_size_operand_ = true; |
| 1931 | FALL_THROUGH; |
| 1932 | case 0xF7: |
| 1933 | data += F6F7Instruction(data); |
| 1934 | break; |
| 1935 | |
| 1936 | // These encodings for inc and dec are IA32 only, but we don't get here |
| 1937 | // on X64 - the REX prefix recoginizer catches them earlier. |
| 1938 | case 0x40: |
| 1939 | case 0x41: |
| 1940 | case 0x42: |
| 1941 | case 0x43: |
| 1942 | case 0x44: |
| 1943 | case 0x45: |
| 1944 | case 0x46: |
| 1945 | case 0x47: |
| 1946 | Print("inc %s" , NameOfCPURegister(*data & 7)); |
| 1947 | data += 1; |
| 1948 | break; |
| 1949 | |
| 1950 | case 0x48: |
| 1951 | case 0x49: |
| 1952 | case 0x4a: |
| 1953 | case 0x4b: |
| 1954 | case 0x4c: |
| 1955 | case 0x4d: |
| 1956 | case 0x4e: |
| 1957 | case 0x4f: |
| 1958 | Print("dec %s" , NameOfCPURegister(*data & 7)); |
| 1959 | data += 1; |
| 1960 | break; |
| 1961 | |
| 1962 | #if defined(TARGET_ARCH_IA32) |
| 1963 | case 0x61: |
| 1964 | Print("popad" ); |
| 1965 | break; |
| 1966 | |
| 1967 | case 0x60: |
| 1968 | Print("pushad" ); |
| 1969 | break; |
| 1970 | #endif |
| 1971 | |
| 1972 | default: |
| 1973 | UnimplementedInstruction(); |
| 1974 | data += 1; |
| 1975 | } |
| 1976 | } // !processed |
| 1977 | |
| 1978 | ASSERT(buffer_[buffer_pos_] == '\0'); |
| 1979 | |
| 1980 | int instr_len = data - reinterpret_cast<uint8_t*>(pc); |
| 1981 | ASSERT(instr_len > 0); // Ensure progress. |
| 1982 | |
| 1983 | return instr_len; |
| 1984 | } |
| 1985 | |
| 1986 | void Disassembler::DecodeInstruction(char* hex_buffer, |
| 1987 | intptr_t hex_size, |
| 1988 | char* human_buffer, |
| 1989 | intptr_t human_size, |
| 1990 | int* out_instr_len, |
| 1991 | const Code& code, |
| 1992 | Object** object, |
| 1993 | uword pc) { |
| 1994 | ASSERT(hex_size > 0); |
| 1995 | ASSERT(human_size > 0); |
| 1996 | DisassemblerX64 decoder(human_buffer, human_size); |
| 1997 | int instruction_length = decoder.InstructionDecode(pc); |
| 1998 | uint8_t* pc_ptr = reinterpret_cast<uint8_t*>(pc); |
| 1999 | int hex_index = 0; |
| 2000 | int remaining_size = hex_size - hex_index; |
| 2001 | for (int i = 0; (i < instruction_length) && (remaining_size > 2); ++i) { |
| 2002 | Utils::SNPrint(&hex_buffer[hex_index], remaining_size, "%02x" , pc_ptr[i]); |
| 2003 | hex_index += 2; |
| 2004 | remaining_size -= 2; |
| 2005 | } |
| 2006 | hex_buffer[hex_index] = '\0'; |
| 2007 | if (out_instr_len != nullptr) { |
| 2008 | *out_instr_len = instruction_length; |
| 2009 | } |
| 2010 | |
| 2011 | *object = NULL; |
| 2012 | #if defined(TARGET_ARCH_X64) |
| 2013 | if (!code.IsNull()) { |
| 2014 | *object = &Object::Handle(); |
| 2015 | if (!DecodeLoadObjectFromPoolOrThread(pc, code, *object)) { |
| 2016 | *object = NULL; |
| 2017 | } |
| 2018 | } |
| 2019 | #else |
| 2020 | if (!code.IsNull() && code.is_alive()) { |
| 2021 | intptr_t offsets_length = code.pointer_offsets_length(); |
| 2022 | for (intptr_t i = 0; i < offsets_length; i++) { |
| 2023 | uword addr = code.GetPointerOffsetAt(i) + code.PayloadStart(); |
| 2024 | if ((pc <= addr) && (addr < (pc + instruction_length))) { |
| 2025 | *object = |
| 2026 | &Object::Handle(LoadUnaligned(reinterpret_cast<ObjectPtr*>(addr))); |
| 2027 | break; |
| 2028 | } |
| 2029 | } |
| 2030 | } |
| 2031 | #endif |
| 2032 | } |
| 2033 | |
| 2034 | #endif // !defined(PRODUCT) || defined(FORCE_INCLUDE_DISASSEMBLER) |
| 2035 | |
| 2036 | } // namespace dart |
| 2037 | |
| 2038 | #endif // defined(TARGET_ARCH_X64) || defined (TARGET_ARCH_IA32) |
| 2039 | |