| 1 |
// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- |
| 2 |
/* Copyright (c) 2007, Google Inc. |
| 3 |
* All rights reserved. |
| 4 |
* |
| 5 |
* Redistribution and use in source and binary forms, with or without |
| 6 |
* modification, are permitted provided that the following conditions are |
| 7 |
* met: |
| 8 |
* |
| 9 |
* * Redistributions of source code must retain the above copyright |
| 10 |
* notice, this list of conditions and the following disclaimer. |
| 11 |
* * Redistributions in binary form must reproduce the above |
| 12 |
* copyright notice, this list of conditions and the following disclaimer |
| 13 |
* in the documentation and/or other materials provided with the |
| 14 |
* distribution. |
| 15 |
* * Neither the name of Google Inc. nor the names of its |
| 16 |
* contributors may be used to endorse or promote products derived from |
| 17 |
* this software without specific prior written permission. |
| 18 |
* |
| 19 |
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 20 |
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 21 |
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 22 |
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 23 |
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 24 |
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 25 |
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 26 |
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 27 |
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 28 |
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 29 |
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 30 |
* |
| 31 |
* --- |
| 32 |
* Author: Joi Sigurdsson |
| 33 |
* |
| 34 |
* Implementation of MiniDisassembler. |
| 35 |
*/ |
| 36 |
|
| 37 |
#include "mini_disassembler.h" |
| 38 |
|
| 39 |
namespace sidestep { |
| 40 |
|
| 41 |
MiniDisassembler::MiniDisassembler(bool operand_default_is_32_bits, |
| 42 |
bool address_default_is_32_bits) |
| 43 |
: operand_default_is_32_bits_(operand_default_is_32_bits), |
| 44 |
address_default_is_32_bits_(address_default_is_32_bits) { |
| 45 |
Initialize(); |
| 46 |
} |
| 47 |
|
| 48 |
MiniDisassembler::MiniDisassembler() |
| 49 |
: operand_default_is_32_bits_(true), |
| 50 |
address_default_is_32_bits_(true) { |
| 51 |
Initialize(); |
| 52 |
} |
| 53 |
|
| 54 |
InstructionType MiniDisassembler::Disassemble( |
| 55 |
unsigned char* start_byte, |
| 56 |
unsigned int& instruction_bytes) { |
| 57 |
// Clean up any state from previous invocations. |
| 58 |
Initialize(); |
| 59 |
|
| 60 |
// Start by processing any prefixes. |
| 61 |
unsigned char* current_byte = start_byte; |
| 62 |
unsigned int size = 0; |
| 63 |
InstructionType instruction_type = ProcessPrefixes(current_byte, size); |
| 64 |
|
| 65 |
if (IT_UNKNOWN == instruction_type) |
| 66 |
return instruction_type; |
| 67 |
|
| 68 |
current_byte += size; |
| 69 |
size = 0; |
| 70 |
|
| 71 |
// Invariant: We have stripped all prefixes, and the operand_is_32_bits_ |
| 72 |
// and address_is_32_bits_ flags are correctly set. |
| 73 |
|
| 74 |
instruction_type = ProcessOpcode(current_byte, 0, size); |
| 75 |
|
| 76 |
// Check for error processing instruction |
| 77 |
if ((IT_UNKNOWN == instruction_type_) || (IT_UNUSED == instruction_type_)) { |
| 78 |
return IT_UNKNOWN; |
| 79 |
} |
| 80 |
|
| 81 |
current_byte += size; |
| 82 |
|
| 83 |
// Invariant: operand_bytes_ indicates the total size of operands |
| 84 |
// specified by the opcode and/or ModR/M byte and/or SIB byte. |
| 85 |
// pCurrentByte points to the first byte after the ModR/M byte, or after |
| 86 |
// the SIB byte if it is present (i.e. the first byte of any operands |
| 87 |
// encoded in the instruction). |
| 88 |
|
| 89 |
// We get the total length of any prefixes, the opcode, and the ModR/M and |
| 90 |
// SIB bytes if present, by taking the difference of the original starting |
| 91 |
// address and the current byte (which points to the first byte of the |
| 92 |
// operands if present, or to the first byte of the next instruction if |
| 93 |
// they are not). Adding the count of bytes in the operands encoded in |
| 94 |
// the instruction gives us the full length of the instruction in bytes. |
| 95 |
instruction_bytes += operand_bytes_ + (current_byte - start_byte); |
| 96 |
|
| 97 |
// Return the instruction type, which was set by ProcessOpcode(). |
| 98 |
return instruction_type_; |
| 99 |
} |
| 100 |
|
| 101 |
void MiniDisassembler::Initialize() { |
| 102 |
operand_is_32_bits_ = operand_default_is_32_bits_; |
| 103 |
address_is_32_bits_ = address_default_is_32_bits_; |
| 104 |
#ifdef _M_X64 |
| 105 |
operand_default_support_64_bits_ = true; |
| 106 |
#else |
| 107 |
operand_default_support_64_bits_ = false; |
| 108 |
#endif |
| 109 |
operand_is_64_bits_ = false; |
| 110 |
operand_bytes_ = 0; |
| 111 |
have_modrm_ = false; |
| 112 |
should_decode_modrm_ = false; |
| 113 |
instruction_type_ = IT_UNKNOWN; |
| 114 |
got_f2_prefix_ = false; |
| 115 |
got_f3_prefix_ = false; |
| 116 |
got_66_prefix_ = false; |
| 117 |
} |
| 118 |
|
| 119 |
InstructionType MiniDisassembler::ProcessPrefixes(unsigned char* start_byte, |
| 120 |
unsigned int& size) { |
| 121 |
InstructionType instruction_type = IT_GENERIC; |
| 122 |
const Opcode& opcode = s_ia32_opcode_map_[0].table_[*start_byte]; |
| 123 |
|
| 124 |
switch (opcode.type_) { |
| 125 |
case IT_PREFIX_ADDRESS: |
| 126 |
address_is_32_bits_ = !address_default_is_32_bits_; |
| 127 |
goto nochangeoperand; |
| 128 |
case IT_PREFIX_OPERAND: |
| 129 |
operand_is_32_bits_ = !operand_default_is_32_bits_; |
| 130 |
nochangeoperand: |
| 131 |
case IT_PREFIX: |
| 132 |
|
| 133 |
if (0xF2 == (*start_byte)) |
| 134 |
got_f2_prefix_ = true; |
| 135 |
else if (0xF3 == (*start_byte)) |
| 136 |
got_f3_prefix_ = true; |
| 137 |
else if (0x66 == (*start_byte)) |
| 138 |
got_66_prefix_ = true; |
| 139 |
else if (operand_default_support_64_bits_ && (*start_byte) & 0x48) |
| 140 |
operand_is_64_bits_ = true; |
| 141 |
|
| 142 |
instruction_type = opcode.type_; |
| 143 |
size ++; |
| 144 |
// we got a prefix, so add one and check next byte |
| 145 |
ProcessPrefixes(start_byte + 1, size); |
| 146 |
default: |
| 147 |
break; // not a prefix byte |
| 148 |
} |
| 149 |
|
| 150 |
return instruction_type; |
| 151 |
} |
| 152 |
|
| 153 |
InstructionType MiniDisassembler::ProcessOpcode(unsigned char* start_byte, |
| 154 |
unsigned int table_index, |
| 155 |
unsigned int& size) { |
| 156 |
const OpcodeTable& table = s_ia32_opcode_map_[table_index]; // Get our table |
| 157 |
unsigned char current_byte = (*start_byte) >> table.shift_; |
| 158 |
current_byte = current_byte & table.mask_; // Mask out the bits we will use |
| 159 |
|
| 160 |
// Check whether the byte we have is inside the table we have. |
| 161 |
if (current_byte < table.min_lim_ || current_byte > table.max_lim_) { |
| 162 |
instruction_type_ = IT_UNKNOWN; |
| 163 |
return instruction_type_; |
| 164 |
} |
| 165 |
|
| 166 |
const Opcode& opcode = table.table_[current_byte]; |
| 167 |
if (IT_UNUSED == opcode.type_) { |
| 168 |
// This instruction is not used by the IA-32 ISA, so we indicate |
| 169 |
// this to the user. Probably means that we were pointed to |
| 170 |
// a byte in memory that was not the start of an instruction. |
| 171 |
instruction_type_ = IT_UNUSED; |
| 172 |
return instruction_type_; |
| 173 |
} else if (IT_REFERENCE == opcode.type_) { |
| 174 |
// We are looking at an opcode that has more bytes (or is continued |
| 175 |
// in the ModR/M byte). Recursively find the opcode definition in |
| 176 |
// the table for the opcode's next byte. |
| 177 |
size++; |
| 178 |
ProcessOpcode(start_byte + 1, opcode.table_index_, size); |
| 179 |
return instruction_type_; |
| 180 |
} |
| 181 |
|
| 182 |
const SpecificOpcode* specific_opcode = (SpecificOpcode*)&opcode; |
| 183 |
if (opcode.is_prefix_dependent_) { |
| 184 |
if (got_f2_prefix_ && opcode.opcode_if_f2_prefix_.mnemonic_ != 0) { |
| 185 |
specific_opcode = &opcode.opcode_if_f2_prefix_; |
| 186 |
} else if (got_f3_prefix_ && opcode.opcode_if_f3_prefix_.mnemonic_ != 0) { |
| 187 |
specific_opcode = &opcode.opcode_if_f3_prefix_; |
| 188 |
} else if (got_66_prefix_ && opcode.opcode_if_66_prefix_.mnemonic_ != 0) { |
| 189 |
specific_opcode = &opcode.opcode_if_66_prefix_; |
| 190 |
} |
| 191 |
} |
| 192 |
|
| 193 |
// Inv: The opcode type is known. |
| 194 |
instruction_type_ = specific_opcode->type_; |
| 195 |
|
| 196 |
// Let's process the operand types to see if we have any immediate |
| 197 |
// operands, and/or a ModR/M byte. |
| 198 |
|
| 199 |
ProcessOperand(specific_opcode->flag_dest_); |
| 200 |
ProcessOperand(specific_opcode->flag_source_); |
| 201 |
ProcessOperand(specific_opcode->flag_aux_); |
| 202 |
|
| 203 |
// Inv: We have processed the opcode and incremented operand_bytes_ |
| 204 |
// by the number of bytes of any operands specified by the opcode |
| 205 |
// that are stored in the instruction (not registers etc.). Now |
| 206 |
// we need to return the total number of bytes for the opcode and |
| 207 |
// for the ModR/M or SIB bytes if they are present. |
| 208 |
|
| 209 |
if (table.mask_ != 0xff) { |
| 210 |
if (have_modrm_) { |
| 211 |
// we're looking at a ModR/M byte so we're not going to |
| 212 |
// count that into the opcode size |
| 213 |
ProcessModrm(start_byte, size); |
| 214 |
return IT_GENERIC; |
| 215 |
} else { |
| 216 |
// need to count the ModR/M byte even if it's just being |
| 217 |
// used for opcode extension |
| 218 |
size++; |
| 219 |
return IT_GENERIC; |
| 220 |
} |
| 221 |
} else { |
| 222 |
if (have_modrm_) { |
| 223 |
// The ModR/M byte is the next byte. |
| 224 |
size++; |
| 225 |
ProcessModrm(start_byte + 1, size); |
| 226 |
return IT_GENERIC; |
| 227 |
} else { |
| 228 |
size++; |
| 229 |
return IT_GENERIC; |
| 230 |
} |
| 231 |
} |
| 232 |
} |
| 233 |
|
| 234 |
bool MiniDisassembler::ProcessOperand(int flag_operand) { |
| 235 |
bool succeeded = true; |
| 236 |
if (AM_NOT_USED == flag_operand) |
| 237 |
return succeeded; |
| 238 |
|
| 239 |
// Decide what to do based on the addressing mode. |
| 240 |
switch (flag_operand & AM_MASK) { |
| 241 |
// No ModR/M byte indicated by these addressing modes, and no |
| 242 |
// additional (e.g. immediate) parameters. |
| 243 |
case AM_A: // Direct address |
| 244 |
case AM_F: // EFLAGS register |
| 245 |
case AM_X: // Memory addressed by the DS:SI register pair |
| 246 |
case AM_Y: // Memory addressed by the ES:DI register pair |
| 247 |
case AM_IMPLICIT: // Parameter is implicit, occupies no space in |
| 248 |
// instruction |
| 249 |
break; |
| 250 |
|
| 251 |
// There is a ModR/M byte but it does not necessarily need |
| 252 |
// to be decoded. |
| 253 |
case AM_C: // reg field of ModR/M selects a control register |
| 254 |
case AM_D: // reg field of ModR/M selects a debug register |
| 255 |
case AM_G: // reg field of ModR/M selects a general register |
| 256 |
case AM_P: // reg field of ModR/M selects an MMX register |
| 257 |
case AM_R: // mod field of ModR/M may refer only to a general register |
| 258 |
case AM_S: // reg field of ModR/M selects a segment register |
| 259 |
case AM_T: // reg field of ModR/M selects a test register |
| 260 |
case AM_V: // reg field of ModR/M selects a 128-bit XMM register |
| 261 |
have_modrm_ = true; |
| 262 |
break; |
| 263 |
|
| 264 |
// In these addressing modes, there is a ModR/M byte and it needs to be |
| 265 |
// decoded. No other (e.g. immediate) params than indicated in ModR/M. |
| 266 |
case AM_E: // Operand is either a general-purpose register or memory, |
| 267 |
// specified by ModR/M byte |
| 268 |
case AM_M: // ModR/M byte will refer only to memory |
| 269 |
case AM_Q: // Operand is either an MMX register or memory (complex |
| 270 |
// evaluation), specified by ModR/M byte |
| 271 |
case AM_W: // Operand is either a 128-bit XMM register or memory (complex |
| 272 |
// eval), specified by ModR/M byte |
| 273 |
have_modrm_ = true; |
| 274 |
should_decode_modrm_ = true; |
| 275 |
break; |
| 276 |
|
| 277 |
// These addressing modes specify an immediate or an offset value |
| 278 |
// directly, so we need to look at the operand type to see how many |
| 279 |
// bytes. |
| 280 |
case AM_I: // Immediate data. |
| 281 |
case AM_J: // Jump to offset. |
| 282 |
case AM_O: // Operand is at offset. |
| 283 |
switch (flag_operand & OT_MASK) { |
| 284 |
case OT_B: // Byte regardless of operand-size attribute. |
| 285 |
operand_bytes_ += OS_BYTE; |
| 286 |
break; |
| 287 |
case OT_C: // Byte or word, depending on operand-size attribute. |
| 288 |
if (operand_is_32_bits_) |
| 289 |
operand_bytes_ += OS_WORD; |
| 290 |
else |
| 291 |
operand_bytes_ += OS_BYTE; |
| 292 |
break; |
| 293 |
case OT_D: // Doubleword, regardless of operand-size attribute. |
| 294 |
operand_bytes_ += OS_DOUBLE_WORD; |
| 295 |
break; |
| 296 |
case OT_DQ: // Double-quadword, regardless of operand-size attribute. |
| 297 |
operand_bytes_ += OS_DOUBLE_QUAD_WORD; |
| 298 |
break; |
| 299 |
case OT_P: // 32-bit or 48-bit pointer, depending on operand-size |
| 300 |
// attribute. |
| 301 |
if (operand_is_32_bits_) |
| 302 |
operand_bytes_ += OS_48_BIT_POINTER; |
| 303 |
else |
| 304 |
operand_bytes_ += OS_32_BIT_POINTER; |
| 305 |
break; |
| 306 |
case OT_PS: // 128-bit packed single-precision floating-point data. |
| 307 |
operand_bytes_ += OS_128_BIT_PACKED_SINGLE_PRECISION_FLOATING; |
| 308 |
break; |
| 309 |
case OT_Q: // Quadword, regardless of operand-size attribute. |
| 310 |
operand_bytes_ += OS_QUAD_WORD; |
| 311 |
break; |
| 312 |
case OT_S: // 6-byte pseudo-descriptor. |
| 313 |
operand_bytes_ += OS_PSEUDO_DESCRIPTOR; |
| 314 |
break; |
| 315 |
case OT_SD: // Scalar Double-Precision Floating-Point Value |
| 316 |
case OT_PD: // Unaligned packed double-precision floating point value |
| 317 |
operand_bytes_ += OS_DOUBLE_PRECISION_FLOATING; |
| 318 |
break; |
| 319 |
case OT_SS: |
| 320 |
// Scalar element of a 128-bit packed single-precision |
| 321 |
// floating data. |
| 322 |
// We simply return enItUnknown since we don't have to support |
| 323 |
// floating point |
| 324 |
succeeded = false; |
| 325 |
break; |
| 326 |
case OT_V: // Word, doubleword or quadword, depending on operand-size |
| 327 |
// attribute. |
| 328 |
if (operand_is_64_bits_ && flag_operand & AM_I && |
| 329 |
flag_operand & IOS_64) |
| 330 |
operand_bytes_ += OS_QUAD_WORD; |
| 331 |
else if (operand_is_32_bits_) |
| 332 |
operand_bytes_ += OS_DOUBLE_WORD; |
| 333 |
else |
| 334 |
operand_bytes_ += OS_WORD; |
| 335 |
break; |
| 336 |
case OT_W: // Word, regardless of operand-size attribute. |
| 337 |
operand_bytes_ += OS_WORD; |
| 338 |
break; |
| 339 |
|
| 340 |
// Can safely ignore these. |
| 341 |
case OT_A: // Two one-word operands in memory or two double-word |
| 342 |
// operands in memory |
| 343 |
case OT_PI: // Quadword MMX technology register (e.g. mm0) |
| 344 |
case OT_SI: // Doubleword integer register (e.g., eax) |
| 345 |
break; |
| 346 |
|
| 347 |
default: |
| 348 |
break; |
| 349 |
} |
| 350 |
break; |
| 351 |
|
| 352 |
default: |
| 353 |
break; |
| 354 |
} |
| 355 |
|
| 356 |
return succeeded; |
| 357 |
} |
| 358 |
|
| 359 |
bool MiniDisassembler::ProcessModrm(unsigned char* start_byte, |
| 360 |
unsigned int& size) { |
| 361 |
// If we don't need to decode, we just return the size of the ModR/M |
| 362 |
// byte (there is never a SIB byte in this case). |
| 363 |
if (!should_decode_modrm_) { |
| 364 |
size++; |
| 365 |
return true; |
| 366 |
} |
| 367 |
|
| 368 |
// We never care about the reg field, only the combination of the mod |
| 369 |
// and r/m fields, so let's start by packing those fields together into |
| 370 |
// 5 bits. |
| 371 |
unsigned char modrm = (*start_byte); |
| 372 |
unsigned char mod = modrm & 0xC0; // mask out top two bits to get mod field |
| 373 |
modrm = modrm & 0x07; // mask out bottom 3 bits to get r/m field |
| 374 |
mod = mod >> 3; // shift the mod field to the right place |
| 375 |
modrm = mod | modrm; // combine the r/m and mod fields as discussed |
| 376 |
mod = mod >> 3; // shift the mod field to bits 2..0 |
| 377 |
|
| 378 |
// Invariant: modrm contains the mod field in bits 4..3 and the r/m field |
| 379 |
// in bits 2..0, and mod contains the mod field in bits 2..0 |
| 380 |
|
| 381 |
const ModrmEntry* modrm_entry = 0; |
| 382 |
if (address_is_32_bits_) |
| 383 |
modrm_entry = &s_ia32_modrm_map_[modrm]; |
| 384 |
else |
| 385 |
modrm_entry = &s_ia16_modrm_map_[modrm]; |
| 386 |
|
| 387 |
// Invariant: modrm_entry points to information that we need to decode |
| 388 |
// the ModR/M byte. |
| 389 |
|
| 390 |
// Add to the count of operand bytes, if the ModR/M byte indicates |
| 391 |
// that some operands are encoded in the instruction. |
| 392 |
if (modrm_entry->is_encoded_in_instruction_) |
| 393 |
operand_bytes_ += modrm_entry->operand_size_; |
| 394 |
|
| 395 |
// Process the SIB byte if necessary, and return the count |
| 396 |
// of ModR/M and SIB bytes. |
| 397 |
if (modrm_entry->use_sib_byte_) { |
| 398 |
size++; |
| 399 |
return ProcessSib(start_byte + 1, mod, size); |
| 400 |
} else { |
| 401 |
size++; |
| 402 |
return true; |
| 403 |
} |
| 404 |
} |
| 405 |
|
| 406 |
bool MiniDisassembler::ProcessSib(unsigned char* start_byte, |
| 407 |
unsigned char mod, |
| 408 |
unsigned int& size) { |
| 409 |
// get the mod field from the 2..0 bits of the SIB byte |
| 410 |
unsigned char sib_base = (*start_byte) & 0x07; |
| 411 |
if (0x05 == sib_base) { |
| 412 |
switch (mod) { |
| 413 |
case 0x00: // mod == 00 |
| 414 |
case 0x02: // mod == 10 |
| 415 |
operand_bytes_ += OS_DOUBLE_WORD; |
| 416 |
break; |
| 417 |
case 0x01: // mod == 01 |
| 418 |
operand_bytes_ += OS_BYTE; |
| 419 |
break; |
| 420 |
case 0x03: // mod == 11 |
| 421 |
// According to the IA-32 docs, there does not seem to be a disp |
| 422 |
// value for this value of mod |
| 423 |
default: |
| 424 |
break; |
| 425 |
} |
| 426 |
} |
| 427 |
|
| 428 |
size++; |
| 429 |
return true; |
| 430 |
} |
| 431 |
|
| 432 |
}; // namespace sidestep |