1 /*===-- X86DisassemblerDecoderCommon.h - Disassembler decoder -----*- 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 * This file is part of the X86 Disassembler. 11 * It contains common definitions used by both the disassembler and the table 12 * generator. 13 * Documentation for the disassembler can be found in X86Disassembler.h. 14 * 15 *===----------------------------------------------------------------------===*/ 16 17 /* 18 * This header file provides those definitions that need to be shared between 19 * the decoder and the table generator in a C-friendly manner. 20 */ 21 22 #ifndef X86DISASSEMBLERDECODERCOMMON_H 23 #define X86DISASSEMBLERDECODERCOMMON_H 24 25 #include "llvm/Support/DataTypes.h" 26 27 #define INSTRUCTIONS_SYM x86DisassemblerInstrSpecifiers 28 #define CONTEXTS_SYM x86DisassemblerContexts 29 #define ONEBYTE_SYM x86DisassemblerOneByteOpcodes 30 #define TWOBYTE_SYM x86DisassemblerTwoByteOpcodes 31 #define THREEBYTE38_SYM x86DisassemblerThreeByte38Opcodes 32 #define THREEBYTE3A_SYM x86DisassemblerThreeByte3AOpcodes 33 #define THREEBYTEA6_SYM x86DisassemblerThreeByteA6Opcodes 34 #define THREEBYTEA7_SYM x86DisassemblerThreeByteA7Opcodes 35 36 #define INSTRUCTIONS_STR "x86DisassemblerInstrSpecifiers" 37 #define CONTEXTS_STR "x86DisassemblerContexts" 38 #define ONEBYTE_STR "x86DisassemblerOneByteOpcodes" 39 #define TWOBYTE_STR "x86DisassemblerTwoByteOpcodes" 40 #define THREEBYTE38_STR "x86DisassemblerThreeByte38Opcodes" 41 #define THREEBYTE3A_STR "x86DisassemblerThreeByte3AOpcodes" 42 #define THREEBYTEA6_STR "x86DisassemblerThreeByteA6Opcodes" 43 #define THREEBYTEA7_STR "x86DisassemblerThreeByteA7Opcodes" 44 45 /* 46 * Attributes of an instruction that must be known before the opcode can be 47 * processed correctly. Most of these indicate the presence of particular 48 * prefixes, but ATTR_64BIT is simply an attribute of the decoding context. 49 */ 50 #define ATTRIBUTE_BITS \ 51 ENUM_ENTRY(ATTR_NONE, 0x00) \ 52 ENUM_ENTRY(ATTR_64BIT, 0x01) \ 53 ENUM_ENTRY(ATTR_XS, 0x02) \ 54 ENUM_ENTRY(ATTR_XD, 0x04) \ 55 ENUM_ENTRY(ATTR_REXW, 0x08) \ 56 ENUM_ENTRY(ATTR_OPSIZE, 0x10) \ 57 ENUM_ENTRY(ATTR_ADSIZE, 0x20) \ 58 ENUM_ENTRY(ATTR_VEX, 0x40) \ 59 ENUM_ENTRY(ATTR_VEXL, 0x80) 60 61 #define ENUM_ENTRY(n, v) n = v, 62 enum attributeBits { 63 ATTRIBUTE_BITS 64 ATTR_max 65 }; 66 #undef ENUM_ENTRY 67 68 /* 69 * Combinations of the above attributes that are relevant to instruction 70 * decode. Although other combinations are possible, they can be reduced to 71 * these without affecting the ultimately decoded instruction. 72 */ 73 74 /* Class name Rank Rationale for rank assignment */ 75 #define INSTRUCTION_CONTEXTS \ 76 ENUM_ENTRY(IC, 0, "says nothing about the instruction") \ 77 ENUM_ENTRY(IC_64BIT, 1, "says the instruction applies in " \ 78 "64-bit mode but no more") \ 79 ENUM_ENTRY(IC_OPSIZE, 3, "requires an OPSIZE prefix, so " \ 80 "operands change width") \ 81 ENUM_ENTRY(IC_ADSIZE, 3, "requires an ADSIZE prefix, so " \ 82 "operands change width") \ 83 ENUM_ENTRY(IC_XD, 2, "may say something about the opcode " \ 84 "but not the operands") \ 85 ENUM_ENTRY(IC_XS, 2, "may say something about the opcode " \ 86 "but not the operands") \ 87 ENUM_ENTRY(IC_XD_OPSIZE, 3, "requires an OPSIZE prefix, so " \ 88 "operands change width") \ 89 ENUM_ENTRY(IC_XS_OPSIZE, 3, "requires an OPSIZE prefix, so " \ 90 "operands change width") \ 91 ENUM_ENTRY(IC_64BIT_REXW, 4, "requires a REX.W prefix, so operands "\ 92 "change width; overrides IC_OPSIZE") \ 93 ENUM_ENTRY(IC_64BIT_OPSIZE, 3, "Just as meaningful as IC_OPSIZE") \ 94 ENUM_ENTRY(IC_64BIT_ADSIZE, 3, "Just as meaningful as IC_ADSIZE") \ 95 ENUM_ENTRY(IC_64BIT_XD, 5, "XD instructions are SSE; REX.W is " \ 96 "secondary") \ 97 ENUM_ENTRY(IC_64BIT_XS, 5, "Just as meaningful as IC_64BIT_XD") \ 98 ENUM_ENTRY(IC_64BIT_XD_OPSIZE, 3, "Just as meaningful as IC_XD_OPSIZE") \ 99 ENUM_ENTRY(IC_64BIT_XS_OPSIZE, 3, "Just as meaningful as IC_XS_OPSIZE") \ 100 ENUM_ENTRY(IC_64BIT_REXW_XS, 6, "OPSIZE could mean a different " \ 101 "opcode") \ 102 ENUM_ENTRY(IC_64BIT_REXW_XD, 6, "Just as meaningful as " \ 103 "IC_64BIT_REXW_XS") \ 104 ENUM_ENTRY(IC_64BIT_REXW_OPSIZE, 7, "The Dynamic Duo! Prefer over all " \ 105 "else because this changes most " \ 106 "operands' meaning") \ 107 ENUM_ENTRY(IC_VEX, 1, "requires a VEX prefix") \ 108 ENUM_ENTRY(IC_VEX_XS, 2, "requires VEX and the XS prefix") \ 109 ENUM_ENTRY(IC_VEX_XD, 2, "requires VEX and the XD prefix") \ 110 ENUM_ENTRY(IC_VEX_OPSIZE, 2, "requires VEX and the OpSize prefix") \ 111 ENUM_ENTRY(IC_VEX_W, 3, "requires VEX and the W prefix") \ 112 ENUM_ENTRY(IC_VEX_W_XS, 4, "requires VEX, W, and XS prefix") \ 113 ENUM_ENTRY(IC_VEX_W_XD, 4, "requires VEX, W, and XD prefix") \ 114 ENUM_ENTRY(IC_VEX_W_OPSIZE, 4, "requires VEX, W, and OpSize") \ 115 ENUM_ENTRY(IC_VEX_L, 3, "requires VEX and the L prefix") \ 116 ENUM_ENTRY(IC_VEX_L_XS, 4, "requires VEX and the L and XS prefix")\ 117 ENUM_ENTRY(IC_VEX_L_XD, 4, "requires VEX and the L and XD prefix")\ 118 ENUM_ENTRY(IC_VEX_L_OPSIZE, 4, "requires VEX, L, and OpSize") \ 119 ENUM_ENTRY(IC_VEX_L_W_OPSIZE, 5, "requires VEX, L, W and OpSize") 120 121 122 #define ENUM_ENTRY(n, r, d) n, 123 typedef enum { 124 INSTRUCTION_CONTEXTS 125 IC_max 126 } InstructionContext; 127 #undef ENUM_ENTRY 128 129 /* 130 * Opcode types, which determine which decode table to use, both in the Intel 131 * manual and also for the decoder. 132 */ 133 typedef enum { 134 ONEBYTE = 0, 135 TWOBYTE = 1, 136 THREEBYTE_38 = 2, 137 THREEBYTE_3A = 3, 138 THREEBYTE_A6 = 4, 139 THREEBYTE_A7 = 5 140 } OpcodeType; 141 142 /* 143 * The following structs are used for the hierarchical decode table. After 144 * determining the instruction's class (i.e., which IC_* constant applies to 145 * it), the decoder reads the opcode. Some instructions require specific 146 * values of the ModR/M byte, so the ModR/M byte indexes into the final table. 147 * 148 * If a ModR/M byte is not required, "required" is left unset, and the values 149 * for each instructionID are identical. 150 */ 151 152 typedef uint16_t InstrUID; 153 154 /* 155 * ModRMDecisionType - describes the type of ModR/M decision, allowing the 156 * consumer to determine the number of entries in it. 157 * 158 * MODRM_ONEENTRY - No matter what the value of the ModR/M byte is, the decoded 159 * instruction is the same. 160 * MODRM_SPLITRM - If the ModR/M byte is between 0x00 and 0xbf, the opcode 161 * corresponds to one instruction; otherwise, it corresponds to 162 * a different instruction. 163 * MODRM_SPLITMISC- If the ModR/M byte is between 0x00 and 0xbf, ModR/M byte 164 * divided by 8 is used to select instruction; otherwise, each 165 * value of the ModR/M byte could correspond to a different 166 * instruction. 167 * MODRM_SPLITREG - ModR/M byte divided by 8 is used to select instruction. This 168 corresponds to instructions that use reg field as opcode 169 * MODRM_FULL - Potentially, each value of the ModR/M byte could correspond 170 * to a different instruction. 171 */ 172 173 #define MODRMTYPES \ 174 ENUM_ENTRY(MODRM_ONEENTRY) \ 175 ENUM_ENTRY(MODRM_SPLITRM) \ 176 ENUM_ENTRY(MODRM_SPLITMISC) \ 177 ENUM_ENTRY(MODRM_SPLITREG) \ 178 ENUM_ENTRY(MODRM_FULL) 179 180 #define ENUM_ENTRY(n) n, 181 typedef enum { 182 MODRMTYPES 183 MODRM_max 184 } ModRMDecisionType; 185 #undef ENUM_ENTRY 186 187 /* 188 * ModRMDecision - Specifies whether a ModR/M byte is needed and (if so) which 189 * instruction each possible value of the ModR/M byte corresponds to. Once 190 * this information is known, we have narrowed down to a single instruction. 191 */ 192 struct ModRMDecision { 193 uint8_t modrm_type; 194 195 /* The macro below must be defined wherever this file is included. */ 196 INSTRUCTION_IDS 197 }; 198 199 /* 200 * OpcodeDecision - Specifies which set of ModR/M->instruction tables to look at 201 * given a particular opcode. 202 */ 203 struct OpcodeDecision { 204 struct ModRMDecision modRMDecisions[256]; 205 }; 206 207 /* 208 * ContextDecision - Specifies which opcode->instruction tables to look at given 209 * a particular context (set of attributes). Since there are many possible 210 * contexts, the decoder first uses CONTEXTS_SYM to determine which context 211 * applies given a specific set of attributes. Hence there are only IC_max 212 * entries in this table, rather than 2^(ATTR_max). 213 */ 214 struct ContextDecision { 215 struct OpcodeDecision opcodeDecisions[IC_max]; 216 }; 217 218 /* 219 * Physical encodings of instruction operands. 220 */ 221 222 #define ENCODINGS \ 223 ENUM_ENTRY(ENCODING_NONE, "") \ 224 ENUM_ENTRY(ENCODING_REG, "Register operand in ModR/M byte.") \ 225 ENUM_ENTRY(ENCODING_RM, "R/M operand in ModR/M byte.") \ 226 ENUM_ENTRY(ENCODING_VVVV, "Register operand in VEX.vvvv byte.") \ 227 ENUM_ENTRY(ENCODING_CB, "1-byte code offset (possible new CS value)") \ 228 ENUM_ENTRY(ENCODING_CW, "2-byte") \ 229 ENUM_ENTRY(ENCODING_CD, "4-byte") \ 230 ENUM_ENTRY(ENCODING_CP, "6-byte") \ 231 ENUM_ENTRY(ENCODING_CO, "8-byte") \ 232 ENUM_ENTRY(ENCODING_CT, "10-byte") \ 233 ENUM_ENTRY(ENCODING_IB, "1-byte immediate") \ 234 ENUM_ENTRY(ENCODING_IW, "2-byte") \ 235 ENUM_ENTRY(ENCODING_ID, "4-byte") \ 236 ENUM_ENTRY(ENCODING_IO, "8-byte") \ 237 ENUM_ENTRY(ENCODING_RB, "(AL..DIL, R8L..R15L) Register code added to " \ 238 "the opcode byte") \ 239 ENUM_ENTRY(ENCODING_RW, "(AX..DI, R8W..R15W)") \ 240 ENUM_ENTRY(ENCODING_RD, "(EAX..EDI, R8D..R15D)") \ 241 ENUM_ENTRY(ENCODING_RO, "(RAX..RDI, R8..R15)") \ 242 ENUM_ENTRY(ENCODING_I, "Position on floating-point stack added to the " \ 243 "opcode byte") \ 244 \ 245 ENUM_ENTRY(ENCODING_Iv, "Immediate of operand size") \ 246 ENUM_ENTRY(ENCODING_Ia, "Immediate of address size") \ 247 ENUM_ENTRY(ENCODING_Rv, "Register code of operand size added to the " \ 248 "opcode byte") \ 249 ENUM_ENTRY(ENCODING_DUP, "Duplicate of another operand; ID is encoded " \ 250 "in type") 251 252 #define ENUM_ENTRY(n, d) n, 253 typedef enum { 254 ENCODINGS 255 ENCODING_max 256 } OperandEncoding; 257 #undef ENUM_ENTRY 258 259 /* 260 * Semantic interpretations of instruction operands. 261 */ 262 263 #define TYPES \ 264 ENUM_ENTRY(TYPE_NONE, "") \ 265 ENUM_ENTRY(TYPE_REL8, "1-byte immediate address") \ 266 ENUM_ENTRY(TYPE_REL16, "2-byte") \ 267 ENUM_ENTRY(TYPE_REL32, "4-byte") \ 268 ENUM_ENTRY(TYPE_REL64, "8-byte") \ 269 ENUM_ENTRY(TYPE_PTR1616, "2+2-byte segment+offset address") \ 270 ENUM_ENTRY(TYPE_PTR1632, "2+4-byte") \ 271 ENUM_ENTRY(TYPE_PTR1664, "2+8-byte") \ 272 ENUM_ENTRY(TYPE_R8, "1-byte register operand") \ 273 ENUM_ENTRY(TYPE_R16, "2-byte") \ 274 ENUM_ENTRY(TYPE_R32, "4-byte") \ 275 ENUM_ENTRY(TYPE_R64, "8-byte") \ 276 ENUM_ENTRY(TYPE_IMM8, "1-byte immediate operand") \ 277 ENUM_ENTRY(TYPE_IMM16, "2-byte") \ 278 ENUM_ENTRY(TYPE_IMM32, "4-byte") \ 279 ENUM_ENTRY(TYPE_IMM64, "8-byte") \ 280 ENUM_ENTRY(TYPE_IMM3, "1-byte immediate operand between 0 and 7") \ 281 ENUM_ENTRY(TYPE_IMM5, "1-byte immediate operand between 0 and 31") \ 282 ENUM_ENTRY(TYPE_RM8, "1-byte register or memory operand") \ 283 ENUM_ENTRY(TYPE_RM16, "2-byte") \ 284 ENUM_ENTRY(TYPE_RM32, "4-byte") \ 285 ENUM_ENTRY(TYPE_RM64, "8-byte") \ 286 ENUM_ENTRY(TYPE_M, "Memory operand") \ 287 ENUM_ENTRY(TYPE_M8, "1-byte") \ 288 ENUM_ENTRY(TYPE_M16, "2-byte") \ 289 ENUM_ENTRY(TYPE_M32, "4-byte") \ 290 ENUM_ENTRY(TYPE_M64, "8-byte") \ 291 ENUM_ENTRY(TYPE_LEA, "Effective address") \ 292 ENUM_ENTRY(TYPE_M128, "16-byte (SSE/SSE2)") \ 293 ENUM_ENTRY(TYPE_M256, "256-byte (AVX)") \ 294 ENUM_ENTRY(TYPE_M1616, "2+2-byte segment+offset address") \ 295 ENUM_ENTRY(TYPE_M1632, "2+4-byte") \ 296 ENUM_ENTRY(TYPE_M1664, "2+8-byte") \ 297 ENUM_ENTRY(TYPE_M16_32, "2+4-byte two-part memory operand (LIDT, LGDT)") \ 298 ENUM_ENTRY(TYPE_M16_16, "2+2-byte (BOUND)") \ 299 ENUM_ENTRY(TYPE_M32_32, "4+4-byte (BOUND)") \ 300 ENUM_ENTRY(TYPE_M16_64, "2+8-byte (LIDT, LGDT)") \ 301 ENUM_ENTRY(TYPE_MOFFS8, "1-byte memory offset (relative to segment " \ 302 "base)") \ 303 ENUM_ENTRY(TYPE_MOFFS16, "2-byte") \ 304 ENUM_ENTRY(TYPE_MOFFS32, "4-byte") \ 305 ENUM_ENTRY(TYPE_MOFFS64, "8-byte") \ 306 ENUM_ENTRY(TYPE_SREG, "Byte with single bit set: 0 = ES, 1 = CS, " \ 307 "2 = SS, 3 = DS, 4 = FS, 5 = GS") \ 308 ENUM_ENTRY(TYPE_M32FP, "32-bit IEE754 memory floating-point operand") \ 309 ENUM_ENTRY(TYPE_M64FP, "64-bit") \ 310 ENUM_ENTRY(TYPE_M80FP, "80-bit extended") \ 311 ENUM_ENTRY(TYPE_M16INT, "2-byte memory integer operand for use in " \ 312 "floating-point instructions") \ 313 ENUM_ENTRY(TYPE_M32INT, "4-byte") \ 314 ENUM_ENTRY(TYPE_M64INT, "8-byte") \ 315 ENUM_ENTRY(TYPE_ST, "Position on the floating-point stack") \ 316 ENUM_ENTRY(TYPE_MM, "MMX register operand") \ 317 ENUM_ENTRY(TYPE_MM32, "4-byte MMX register or memory operand") \ 318 ENUM_ENTRY(TYPE_MM64, "8-byte") \ 319 ENUM_ENTRY(TYPE_XMM, "XMM register operand") \ 320 ENUM_ENTRY(TYPE_XMM32, "4-byte XMM register or memory operand") \ 321 ENUM_ENTRY(TYPE_XMM64, "8-byte") \ 322 ENUM_ENTRY(TYPE_XMM128, "16-byte") \ 323 ENUM_ENTRY(TYPE_XMM256, "32-byte") \ 324 ENUM_ENTRY(TYPE_XMM0, "Implicit use of XMM0") \ 325 ENUM_ENTRY(TYPE_SEGMENTREG, "Segment register operand") \ 326 ENUM_ENTRY(TYPE_DEBUGREG, "Debug register operand") \ 327 ENUM_ENTRY(TYPE_CONTROLREG, "Control register operand") \ 328 \ 329 ENUM_ENTRY(TYPE_Mv, "Memory operand of operand size") \ 330 ENUM_ENTRY(TYPE_Rv, "Register operand of operand size") \ 331 ENUM_ENTRY(TYPE_IMMv, "Immediate operand of operand size") \ 332 ENUM_ENTRY(TYPE_RELv, "Immediate address of operand size") \ 333 ENUM_ENTRY(TYPE_DUP0, "Duplicate of operand 0") \ 334 ENUM_ENTRY(TYPE_DUP1, "operand 1") \ 335 ENUM_ENTRY(TYPE_DUP2, "operand 2") \ 336 ENUM_ENTRY(TYPE_DUP3, "operand 3") \ 337 ENUM_ENTRY(TYPE_DUP4, "operand 4") \ 338 ENUM_ENTRY(TYPE_M512, "512-bit FPU/MMX/XMM/MXCSR state") 339 340 #define ENUM_ENTRY(n, d) n, 341 typedef enum { 342 TYPES 343 TYPE_max 344 } OperandType; 345 #undef ENUM_ENTRY 346 347 /* 348 * OperandSpecifier - The specification for how to extract and interpret one 349 * operand. 350 */ 351 struct OperandSpecifier { 352 uint8_t encoding; 353 uint8_t type; 354 }; 355 356 /* 357 * Indicates where the opcode modifier (if any) is to be found. Extended 358 * opcodes with AddRegFrm have the opcode modifier in the ModR/M byte. 359 */ 360 361 #define MODIFIER_TYPES \ 362 ENUM_ENTRY(MODIFIER_NONE) \ 363 ENUM_ENTRY(MODIFIER_OPCODE) \ 364 ENUM_ENTRY(MODIFIER_MODRM) 365 366 #define ENUM_ENTRY(n) n, 367 typedef enum { 368 MODIFIER_TYPES 369 MODIFIER_max 370 } ModifierType; 371 #undef ENUM_ENTRY 372 373 #define X86_MAX_OPERANDS 5 374 375 /* 376 * The specification for how to extract and interpret a full instruction and 377 * its operands. 378 */ 379 struct InstructionSpecifier { 380 uint8_t modifierType; 381 uint8_t modifierBase; 382 383 /* The macro below must be defined wherever this file is included. */ 384 INSTRUCTION_SPECIFIER_FIELDS 385 }; 386 387 /* 388 * Decoding mode for the Intel disassembler. 16-bit, 32-bit, and 64-bit mode 389 * are supported, and represent real mode, IA-32e, and IA-32e in 64-bit mode, 390 * respectively. 391 */ 392 typedef enum { 393 MODE_16BIT, 394 MODE_32BIT, 395 MODE_64BIT 396 } DisassemblerMode; 397 398 #endif 399