1 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
6 * Implementation of MiniDisassembler.
9 #include "mini_disassembler.h"
13 MiniDisassembler::MiniDisassembler(bool operand_default_is_32_bits
,
14 bool address_default_is_32_bits
)
15 : operand_default_is_32_bits_(operand_default_is_32_bits
),
16 address_default_is_32_bits_(address_default_is_32_bits
) {
20 MiniDisassembler::MiniDisassembler()
21 : operand_default_is_32_bits_(true),
22 address_default_is_32_bits_(true) {
26 InstructionType
MiniDisassembler::Disassemble(
27 unsigned char* start_byte
,
28 unsigned int& instruction_bytes
) {
29 // Clean up any state from previous invocations.
32 // Start by processing any prefixes.
33 unsigned char* current_byte
= start_byte
;
34 unsigned int size
= 0;
35 InstructionType instruction_type
= ProcessPrefixes(current_byte
, size
);
37 if (IT_UNKNOWN
== instruction_type
)
38 return instruction_type
;
43 // Invariant: We have stripped all prefixes, and the operand_is_32_bits_
44 // and address_is_32_bits_ flags are correctly set.
46 instruction_type
= ProcessOpcode(current_byte
, 0, size
);
48 // Check for error processing instruction
49 if ((IT_UNKNOWN
== instruction_type_
) || (IT_UNUSED
== instruction_type_
)) {
55 // Invariant: operand_bytes_ indicates the total size of operands
56 // specified by the opcode and/or ModR/M byte and/or SIB byte.
57 // pCurrentByte points to the first byte after the ModR/M byte, or after
58 // the SIB byte if it is present (i.e. the first byte of any operands
59 // encoded in the instruction).
61 // We get the total length of any prefixes, the opcode, and the ModR/M and
62 // SIB bytes if present, by taking the difference of the original starting
63 // address and the current byte (which points to the first byte of the
64 // operands if present, or to the first byte of the next instruction if
65 // they are not). Adding the count of bytes in the operands encoded in
66 // the instruction gives us the full length of the instruction in bytes.
67 instruction_bytes
+= operand_bytes_
+ (current_byte
- start_byte
);
69 // Return the instruction type, which was set by ProcessOpcode().
70 return instruction_type_
;
73 void MiniDisassembler::Initialize() {
74 operand_is_32_bits_
= operand_default_is_32_bits_
;
75 address_is_32_bits_
= address_default_is_32_bits_
;
78 should_decode_modrm_
= false;
79 instruction_type_
= IT_UNKNOWN
;
80 got_f2_prefix_
= false;
81 got_f3_prefix_
= false;
82 got_66_prefix_
= false;
85 InstructionType
MiniDisassembler::ProcessPrefixes(unsigned char* start_byte
,
87 InstructionType instruction_type
= IT_GENERIC
;
88 const Opcode
& opcode
= s_ia32_opcode_map_
[0].table_
[*start_byte
];
90 switch (opcode
.type_
) {
91 case IT_PREFIX_ADDRESS
:
92 address_is_32_bits_
= !address_default_is_32_bits_
;
94 case IT_PREFIX_OPERAND
:
95 operand_is_32_bits_
= !operand_default_is_32_bits_
;
99 if (0xF2 == (*start_byte
))
100 got_f2_prefix_
= true;
101 else if (0xF3 == (*start_byte
))
102 got_f3_prefix_
= true;
103 else if (0x66 == (*start_byte
))
104 got_66_prefix_
= true;
106 instruction_type
= opcode
.type_
;
108 // we got a prefix, so add one and check next byte
109 ProcessPrefixes(start_byte
+ 1, size
);
111 break; // not a prefix byte
114 return instruction_type
;
117 InstructionType
MiniDisassembler::ProcessOpcode(unsigned char* start_byte
,
118 unsigned int table_index
,
119 unsigned int& size
) {
120 const OpcodeTable
& table
= s_ia32_opcode_map_
[table_index
]; // Get our table
121 unsigned char current_byte
= (*start_byte
) >> table
.shift_
;
122 current_byte
= current_byte
& table
.mask_
; // Mask out the bits we will use
124 // Check whether the byte we have is inside the table we have.
125 if (current_byte
< table
.min_lim_
|| current_byte
> table
.max_lim_
) {
126 instruction_type_
= IT_UNKNOWN
;
127 return instruction_type_
;
130 const Opcode
& opcode
= table
.table_
[current_byte
];
131 if (IT_UNUSED
== opcode
.type_
) {
132 // This instruction is not used by the IA-32 ISA, so we indicate
133 // this to the user. Probably means that we were pointed to
134 // a byte in memory that was not the start of an instruction.
135 instruction_type_
= IT_UNUSED
;
136 return instruction_type_
;
137 } else if (IT_REFERENCE
== opcode
.type_
) {
138 // We are looking at an opcode that has more bytes (or is continued
139 // in the ModR/M byte). Recursively find the opcode definition in
140 // the table for the opcode's next byte.
142 ProcessOpcode(start_byte
+ 1, opcode
.table_index_
, size
);
143 return instruction_type_
;
146 const SpecificOpcode
* specific_opcode
= (SpecificOpcode
*)&opcode
;
147 if (opcode
.is_prefix_dependent_
) {
148 if (got_f2_prefix_
&& opcode
.opcode_if_f2_prefix_
.mnemonic_
!= 0) {
149 specific_opcode
= &opcode
.opcode_if_f2_prefix_
;
150 } else if (got_f3_prefix_
&& opcode
.opcode_if_f3_prefix_
.mnemonic_
!= 0) {
151 specific_opcode
= &opcode
.opcode_if_f3_prefix_
;
152 } else if (got_66_prefix_
&& opcode
.opcode_if_66_prefix_
.mnemonic_
!= 0) {
153 specific_opcode
= &opcode
.opcode_if_66_prefix_
;
157 // Inv: The opcode type is known.
158 instruction_type_
= specific_opcode
->type_
;
160 // Let's process the operand types to see if we have any immediate
161 // operands, and/or a ModR/M byte.
163 ProcessOperand(specific_opcode
->flag_dest_
);
164 ProcessOperand(specific_opcode
->flag_source_
);
165 ProcessOperand(specific_opcode
->flag_aux_
);
167 // Inv: We have processed the opcode and incremented operand_bytes_
168 // by the number of bytes of any operands specified by the opcode
169 // that are stored in the instruction (not registers etc.). Now
170 // we need to return the total number of bytes for the opcode and
171 // for the ModR/M or SIB bytes if they are present.
173 if (table
.mask_
!= 0xff) {
175 // we're looking at a ModR/M byte so we're not going to
176 // count that into the opcode size
177 ProcessModrm(start_byte
, size
);
180 // need to count the ModR/M byte even if it's just being
181 // used for opcode extension
187 // The ModR/M byte is the next byte.
189 ProcessModrm(start_byte
+ 1, size
);
198 bool MiniDisassembler::ProcessOperand(int flag_operand
) {
199 bool succeeded
= true;
200 if (AM_NOT_USED
== flag_operand
)
203 // Decide what to do based on the addressing mode.
204 switch (flag_operand
& AM_MASK
) {
205 // No ModR/M byte indicated by these addressing modes, and no
206 // additional (e.g. immediate) parameters.
207 case AM_A
: // Direct address
208 case AM_F
: // EFLAGS register
209 case AM_X
: // Memory addressed by the DS:SI register pair
210 case AM_Y
: // Memory addressed by the ES:DI register pair
211 case AM_IMPLICIT
: // Parameter is implicit, occupies no space in
215 // There is a ModR/M byte but it does not necessarily need
217 case AM_C
: // reg field of ModR/M selects a control register
218 case AM_D
: // reg field of ModR/M selects a debug register
219 case AM_G
: // reg field of ModR/M selects a general register
220 case AM_P
: // reg field of ModR/M selects an MMX register
221 case AM_R
: // mod field of ModR/M may refer only to a general register
222 case AM_S
: // reg field of ModR/M selects a segment register
223 case AM_T
: // reg field of ModR/M selects a test register
224 case AM_V
: // reg field of ModR/M selects a 128-bit XMM register
228 // In these addressing modes, there is a ModR/M byte and it needs to be
229 // decoded. No other (e.g. immediate) params than indicated in ModR/M.
230 case AM_E
: // Operand is either a general-purpose register or memory,
231 // specified by ModR/M byte
232 case AM_M
: // ModR/M byte will refer only to memory
233 case AM_Q
: // Operand is either an MMX register or memory (complex
234 // evaluation), specified by ModR/M byte
235 case AM_W
: // Operand is either a 128-bit XMM register or memory (complex
236 // eval), specified by ModR/M byte
238 should_decode_modrm_
= true;
241 // These addressing modes specify an immediate or an offset value
242 // directly, so we need to look at the operand type to see how many
244 case AM_I
: // Immediate data.
245 case AM_J
: // Jump to offset.
246 case AM_O
: // Operand is at offset.
247 switch (flag_operand
& OT_MASK
) {
248 case OT_B
: // Byte regardless of operand-size attribute.
249 operand_bytes_
+= OS_BYTE
;
251 case OT_C
: // Byte or word, depending on operand-size attribute.
252 if (operand_is_32_bits_
)
253 operand_bytes_
+= OS_WORD
;
255 operand_bytes_
+= OS_BYTE
;
257 case OT_D
: // Doubleword, regardless of operand-size attribute.
258 operand_bytes_
+= OS_DOUBLE_WORD
;
260 case OT_DQ
: // Double-quadword, regardless of operand-size attribute.
261 operand_bytes_
+= OS_DOUBLE_QUAD_WORD
;
263 case OT_P
: // 32-bit or 48-bit pointer, depending on operand-size
265 if (operand_is_32_bits_
)
266 operand_bytes_
+= OS_48_BIT_POINTER
;
268 operand_bytes_
+= OS_32_BIT_POINTER
;
270 case OT_PS
: // 128-bit packed single-precision floating-point data.
271 operand_bytes_
+= OS_128_BIT_PACKED_SINGLE_PRECISION_FLOATING
;
273 case OT_Q
: // Quadword, regardless of operand-size attribute.
274 operand_bytes_
+= OS_QUAD_WORD
;
276 case OT_S
: // 6-byte pseudo-descriptor.
277 operand_bytes_
+= OS_PSEUDO_DESCRIPTOR
;
279 case OT_SD
: // Scalar Double-Precision Floating-Point Value
280 case OT_PD
: // Unaligned packed double-precision floating point value
281 operand_bytes_
+= OS_DOUBLE_PRECISION_FLOATING
;
284 // Scalar element of a 128-bit packed single-precision
286 // We simply return enItUnknown since we don't have to support
290 case OT_V
: // Word or doubleword, depending on operand-size attribute.
291 if (operand_is_32_bits_
)
292 operand_bytes_
+= OS_DOUBLE_WORD
;
294 operand_bytes_
+= OS_WORD
;
296 case OT_W
: // Word, regardless of operand-size attribute.
297 operand_bytes_
+= OS_WORD
;
300 // Can safely ignore these.
301 case OT_A
: // Two one-word operands in memory or two double-word
302 // operands in memory
303 case OT_PI
: // Quadword MMX technology register (e.g. mm0)
304 case OT_SI
: // Doubleword integer register (e.g., eax)
319 bool MiniDisassembler::ProcessModrm(unsigned char* start_byte
,
320 unsigned int& size
) {
321 // If we don't need to decode, we just return the size of the ModR/M
322 // byte (there is never a SIB byte in this case).
323 if (!should_decode_modrm_
) {
328 // We never care about the reg field, only the combination of the mod
329 // and r/m fields, so let's start by packing those fields together into
331 unsigned char modrm
= (*start_byte
);
332 unsigned char mod
= modrm
& 0xC0; // mask out top two bits to get mod field
333 modrm
= modrm
& 0x07; // mask out bottom 3 bits to get r/m field
334 mod
= mod
>> 3; // shift the mod field to the right place
335 modrm
= mod
| modrm
; // combine the r/m and mod fields as discussed
336 mod
= mod
>> 3; // shift the mod field to bits 2..0
338 // Invariant: modrm contains the mod field in bits 4..3 and the r/m field
339 // in bits 2..0, and mod contains the mod field in bits 2..0
341 const ModrmEntry
* modrm_entry
= 0;
342 if (address_is_32_bits_
)
343 modrm_entry
= &s_ia32_modrm_map_
[modrm
];
345 modrm_entry
= &s_ia16_modrm_map_
[modrm
];
347 // Invariant: modrm_entry points to information that we need to decode
350 // Add to the count of operand bytes, if the ModR/M byte indicates
351 // that some operands are encoded in the instruction.
352 if (modrm_entry
->is_encoded_in_instruction_
)
353 operand_bytes_
+= modrm_entry
->operand_size_
;
355 // Process the SIB byte if necessary, and return the count
356 // of ModR/M and SIB bytes.
357 if (modrm_entry
->use_sib_byte_
) {
359 return ProcessSib(start_byte
+ 1, mod
, size
);
366 bool MiniDisassembler::ProcessSib(unsigned char* start_byte
,
368 unsigned int& size
) {
369 // get the mod field from the 2..0 bits of the SIB byte
370 unsigned char sib_base
= (*start_byte
) & 0x07;
371 if (0x05 == sib_base
) {
373 case 0x00: // mod == 00
374 case 0x02: // mod == 10
375 operand_bytes_
+= OS_DOUBLE_WORD
;
377 case 0x01: // mod == 01
378 operand_bytes_
+= OS_BYTE
;
380 case 0x03: // mod == 11
381 // According to the IA-32 docs, there does not seem to be a disp
382 // value for this value of mod
392 }; // namespace sidestep