[ACCEPTED]-Get size of assembly instructions-machine-code
@AlexisWilke's response is right: this is 14 messy. He provides the right insights and 13 references to do the work, too.
I have done 12 this work in C. The code follows; this 11 is used in production contexts.
Caveats: It 10 does a good part of the traditional x86 9 instruction set, but not all, in particular 8 none of the instructions involving the vector 7 register sets. And it contains decoding 6 for a few "virtual" instructions that we 5 happen to use in our code. I don't think 4 extending this to x86-64 would be difficult, but 3 it would get messier. Lastly, this is lifted 2 directly, but I don't make any guarantees 1 this will compile out-of-the box.
/* (C) Copyright 2012-2014 Semantic Designs, Inc.
You may freely use this code provided you retain this copyright message
*/
typedef unsigned int natural;
natural InstructionLength(BYTE* pc)
{ // returns length of instruction at PC
natural length=0;
natural opcode, opcode2;
natural modrm;
natural sib;
BYTE* p=pc;
while (true)
{ // scan across prefix bytes
opcode=*p++;
switch (opcode)
{ case 0x64: case 0x65: // FS: GS: prefixes
case 0x36: // SS: prefix
case 0x66: case 0x67: // operand size overrides
case 0xF0: case 0xF2: // LOCK, REPNE prefixes
length++;
break;
case 0x2E: // CS: prefix, used as HNT prefix on jumps
case 0x3E: // DS: prefix, used as HT prefix on jumps
length++;
// goto process relative jmp // tighter check possible here
break;
default:
goto process_instruction_body;
}
}
process_instruction_body:
switch(opcode) // switch on main opcode
{
// ONE BYTE OPCODE, move to next opcode without remark
case 0x27: case 0x2F:
case 0x37: case 0x3F:
case 0x40: case 0x41: case 0x42: case 0x43: case 0x44: case 0x45: case 0x46: case 0x47:
case 0x48: case 0x49: case 0x4A: case 0x4B: case 0x4C: case 0x4D: case 0x4E: case 0x4F:
case 0x50: case 0x51: case 0x52: case 0x53: case 0x54: case 0x55: case 0x56: case 0x57:
case 0x58: case 0x59: case 0x5A: case 0x5B: case 0x5C: case 0x5D: case 0x5E: case 0x5F:
case 0x90: // nop
case 0x91: case 0x92: case 0x93: case 0x94: case 0x95: case 0x96: case 0x97: // xchg
case 0x98: case 0x99:
case 0x9C: case 0x9D: case 0x9E: case 0x9F:
case 0xA4: case 0xA5: case 0xA6: case 0xA7: case 0xAA: case 0xAB: // string operators
case 0xAC: case 0xAD: case 0xAE: case 0xAF:
/* case 0xC3: // RET handled elsewhere */
case 0xC9:
case 0xCC: // int3
case 0xF5: case 0xF8: case 0xF9: case 0xFC: case 0xFD:
return length+1; // include opcode
case 0xC3: // RET
if (*p++ != 0xCC)
return length+1;
if (*p++ != 0xCC)
return length+2;
if (*p++ == 0xCC
&& *p++ == 0xCC)
return length+5;
goto error;
// TWO BYTE INSTRUCTION
case 0x04: case 0x0C: case 0x14: case 0x1C: case 0x24: case 0x2C: case 0x34: case 0x3C:
case 0x6A:
case 0xB0: case 0xB1: case 0xB2: case 0xB3: case 0xB4: case 0xB5: case 0xB6: case 0xB7:
case 0xC2:
return length+2;
// TWO BYTE RELATIVE BRANCH
case 0x70: case 0x71: case 0x72: case 0x73: case 0x74: case 0x75: case 0x76: case 0x77:
case 0x78: case 0x79: case 0x7A: case 0x7B: case 0x7C: case 0x7D: case 0x7E: case 0x7F:
case 0xE0: case 0xE1: case 0xE2: case 0xE3: case 0xEB:
return length+2;
// THREE BYTE INSTRUCTION (NONE!)
// FIVE BYTE INSTRUCTION:
case 0x05: case 0x0D: case 0x15: case 0x1D:
case 0x25: case 0x2D: case 0x35: case 0x3D:
case 0x68:
case 0xA9:
case 0xB8: case 0xB9: case 0xBA: case 0xBB: case 0xBC: case 0xBD: case 0xBE: case 0xBF:
return length+5;
// FIVE BYTE RELATIVE CALL
case 0xE8:
return length+5;
// FIVE BYTE RELATIVE BRANCH
case 0xE9:
if (p[4]==0xCC)
return length+6; // <jmp near ptr ... int 3>
return length+5; // plain <jmp near ptr>
// FIVE BYTE DIRECT ADDRESS
case 0xA1: case 0xA2: case 0xA3: // MOV AL,AX,EAX moffset...
return length+5;
break;
// ModR/M with no immediate operand
case 0x00: case 0x01: case 0x02: case 0x03: case 0x08: case 0x09: case 0x0A: case 0x0B:
case 0x10: case 0x11: case 0x12: case 0x13: case 0x18: case 0x19: case 0x1A: case 0x1B:
case 0x20: case 0x21: case 0x22: case 0x23: case 0x28: case 0x29: case 0x2A: case 0x2B:
case 0x30: case 0x31: case 0x32: case 0x33: case 0x38: case 0x39: case 0x3A: case 0x3B:
case 0x84: case 0x85: case 0x86: case 0x87: case 0x88: case 0x89: case 0x8A: case 0x8B: case 0x8D: case 0x8F:
case 0xD1: case 0xD2: case 0xD3:
case 0xFE: case 0xFF: // misinterprets JMP far and CALL far, not worth fixing
length++; // count opcode
goto modrm;
// ModR/M with immediate 8 bit value
case 0x80: case 0x82: case 0x83:
case 0xC0: case 0xC1:
case 0xC6: // with r=0?
length+=2; // count opcode and immediate byte
goto modrm;
// ModR/M with immediate 32 bit value
case 0x81:
case 0xC7: // with r=0?
length+=5; // count opcode and immediate byte
goto modrm;
case 0x9B: // FSTSW AX = 9B DF E0
if (*p++==0xDF)
{ if (*p++==0xE0)
return length+3;
printf("InstructionLength: Unimplemented 0x9B tertiary opcode %2x at %x\n",*p,p);
goto error;
}
else { printf("InstructionLength: Unimplemented 0x9B secondary opcode %2x at %x\n",*p,p);
goto error;
}
case 0xD9: // various FP instructions
modrm=*p++;
length++; // account for FP prefix
switch (modrm)
{ case 0xC9: case 0xD0:
case 0xE0: case 0xE1: case 0xE4: case 0xE5:
case 0xE8: case 0xE9: case 0xEA: case 0xEB: case 0xEC: case 0xED: case 0xEE:
case 0xF8: case 0xF9: case 0xFA: case 0xFB: case 0xFC: case 0xFD: case 0xFE: case 0xFF:
return length+1;
default: // r bits matter if not one of the above specific opcodes
switch((modrm&0x38)>>3)
{ case 0: goto modrm_fetched; // fld
case 1: return length+1; // fxch
case 2: goto modrm_fetched; // fst
case 3: goto modrm_fetched; // fstp
case 4: goto modrm_fetched; // fldenv
case 5: goto modrm_fetched; // fldcw
case 6: goto modrm_fetched; // fnstenv
case 7: goto modrm_fetched; // fnstcw
}
goto error; // unrecognized 2nd byte
}
case 0xDB: // various FP instructions
modrm=*p++;
length++; // account for FP prefix
switch (modrm)
{ case 0xE3:
return length+1;
default: // r bits matter if not one of the above specific opcodes
#if 0
switch((modrm&0x38)>>3)
{ case 0: goto modrm_fetched; // fld
case 1: return length+1; // fxch
case 2: goto modrm_fetched; // fst
case 3: goto modrm_fetched; // fstp
case 4: goto modrm_fetched; // fldenv
case 5: goto modrm_fetched; // fldcw
case 6: goto modrm_fetched; // fnstenv
case 7: goto modrm_fetched; // fnstcw
}
#endif
goto error; // unrecognized 2nd byte
}
case 0xDD: // various FP instructions
modrm=*p++;
length++; // account for FP prefix
switch (modrm)
{ case 0xE1: case 0xE9:
return length+1;
default: // r bits matter if not one of the above specific opcodes
switch((modrm&0x38)>>3)
{ case 0: goto modrm_fetched; // fld
// case 1: return length+1; // fisttp
case 2: goto modrm_fetched; // fst
case 3: goto modrm_fetched; // fstp
case 4: return length+1; // frstor
case 5: return length+1; // fucomp
case 6: goto modrm_fetched; // fnsav
case 7: goto modrm_fetched; // fnstsw
}
goto error; // unrecognized 2nd byte
}
case 0xF3: // funny prefix REPE
opcode2=*p++; // get second opcode byte
switch (opcode2)
{ case 0x90: // == PAUSE
case 0xA4: case 0xA5: case 0xA6: case 0xA7: case 0xAA: case 0xAB: // string operators
return length+2;
case 0xC3: // (REP) RET
if (*p++ != 0xCC)
return length+2; // only (REP) RET
if (*p++ != 0xCC)
goto error;
if (*p++ == 0xCC)
return length+5; // (REP) RET CLONE IS LONG JUMP RELATIVE
goto error;
case 0x66: // operand size override (32->16 bits)
if (*p++ == 0xA5) // "rep movsw"
return length+3;
goto error;
default: goto error;
}
case 0xF6: // funny subblock of opcodes
modrm=*p++;
if ((modrm & 0x20) == 0)
length++; // 8 bit immediate operand
goto modrm_fetched;
case 0xF7: // funny subblock of opcodes
modrm=*p++;
if ((modrm & 0x30) == 0)
length+=4; // 32 bit immediate operand
goto modrm_fetched;
// Intel's special prefix opcode
case 0x0F:
length+=2; // add one for special prefix, and one for following opcode
opcode2=*p++;
switch(opcode2)
{ case 0x31: // RDTSC
return length;
// CMOVxx
case 0x40: case 0x41: case 0x42: case 0x43: case 0x44: case 0x45: case 0x46: case 0x47:
case 0x48: case 0x49: case 0x4A: case 0x4B: case 0x4C: case 0x4D: case 0x4E: case 0x4F:
goto modrm;
// JC relative 32 bits
case 0x80: case 0x81: case 0x82: case 0x83: case 0x84: case 0x85: case 0x86: case 0x87:
case 0x88: case 0x89: case 0x8A: case 0x8B: case 0x8C: case 0x8D: case 0x8E: case 0x8F:
return length+4; // account for subopcode and displacement
// SETxx rm32
case 0x90: case 0x91: case 0x92: case 0x93: case 0x94: case 0x95: case 0x96: case 0x97:
case 0x98: case 0x99: case 0x9A: case 0x9B: case 0x9C: case 0x9D: case 0x9E: case 0x9F:
goto modrm;
case 0xA2: // CPUID
return length+2;
case 0xAE: // LFENCE, SFENCE, MFENCE
opcode2=*p++;
switch (opcode2)
{ case 0xE8: // LFENCE
case 0xF0: // MFENCE
case 0xF8: // SFENCE
return length+1;
default:
printf("InstructionLength: Unimplemented 0x0F, 0xAE tertiary opcode in clone %2x at %x\n",opcode2,p-1);
goto error;
}
case 0xAF: // imul
case 0xB0: // cmpxchg 8 bits
goto error;
case 0xB1: // cmpxchg 32 bits
case 0xB6: case 0xB7: // movzx
case 0xBC: /* bsf */ case 0xBD: // bsr
// case 0xBE: case 0xBF: // movsx
case 0xC1: // xadd
case 0xC7: // cmpxchg8b
goto modrm;
default:
printf("InstructionLength: Unimplemented 0x0F secondary opcode in clone %2x at %x\n",opcode,p-1);
goto error;
} // switch
// ALL THE THE REST OF THE INSTRUCTIONS; these are instructions that runtime system shouldn't ever use
default:
/* case 0x26: case 0x36: // ES: SS: prefixes
case 0x9A:
case 0xC8: case 0xCA: case 0xCB: case 0xCD: case 0xCE: case 0xCF:
case 0xD6: case 0xD7:
case 0xE4: case 0xE5: case 0xE6: case 0xE7: case 0xEA: case 0xEB: case 0xEC: case 0xED: case 0xEF:
case 0xF4: case 0xFA: case 0xFB:
*/
printf("InstructionLength: Unexpected opcode %2x\n",opcode);
goto error;
}
modrm:
modrm=*p++;
modrm_fetched:
if (trace_clone_checking)
printf("InstructionLength: ModR/M byte %x %2x\n",pc,modrm);
if (modrm >= 0xC0)
return length+1; // account for modrm opcode
else
{ /* memory access */
if ((modrm & 0x7) == 0x04)
{ /* instruction with SIB byte */
length++; // account for SIB byte
sib=*p++; // fetch the sib byte
if ((sib & 0x7) == 0x05)
{ if ((modrm & 0xC0) == 0x40)
return length+1+1; // account for MOD + byte displacment
else return length+1+4; // account for MOD + dword displacement
}
}
switch(modrm & 0xC0)
{ case 0x0:
if ( (modrm & 0x07) == 0x05)
return length+5; // 4 byte displacement
else return length+1; // zero length offset
case 0x80:
return length+5; // 4 byte offset
default:
return length+2; // one byte offset
}
}
error:
{ printf("InstructionLength: unhandled opcode at %8x with opcode %2x\n",pc,opcode);
}
return 0; // can't actually execute this
}
Decoding instructions is not that complicated. However, because 56 the Intel family of processors are CISC, it 55 makes the task rather daunting.
First of 54 all, you should not write it in assembler, because 53 it's going to take you a year or two, but 52 maybe you have the time to do that. Since 51 you only need to scan the code, not print 50 out the results, you can do the work much 49 faster than an actual disassembler would 48 do. That being said you'll bump in the same 47 main problems.
First of all, the manuals 46 are there:
I suggest this one:
Then, all you have to do is read 45 one byte and understand it. You have a table 44 on page 770 that shows you the encoding 43 from the op-code to the instruction.
So for 42 example, 0x33 represents an XOR with Gb,Ev 41 as parameters. G means General register 40 defined in the following ModR/M. Then the 39 b is the size (byte). The E means that you 38 have a ModR/M after that one byte (same 37 byte for G and E). So you will have to read 36 that one byte to determine the addressing 35 mode and from that you can determine the 34 register (Can be ignored) and the address 33 size. The address (Ev) may be another register 32 (then no extra byte), it could be immediate 31 data (1, 2, 4, 8 bytes) or it could be an 30 address (again 1, 2, 4, 8 bytes). Pretty 29 simple, right? Note that ALL instructions 28 use the exact same ModR/M so you have to 27 implement that just once. Also the order 26 in which bytes are added after the instruction 25 code is always exactly the same.
Before the 24 address or immediate (if I'm correct) comes 23 the extra Mod for 64 bit instructions. That 22 one defines additional modes and support 21 for the extended registers. All of that 20 is described in detail in the document I 19 mentioned earlier.
More or less, you need 18 your parser to understand the ModR/M, SIB, prefixes, and 17 voilà. It's not that complicated. Then the 16 first byte tells you the instruction (first 15 2 bytes if the first byte is 0x0F...)
Some 14 instructions also support prefixes to tweak 13 the size of the operands and other similar 12 things. As far as I know, only the 0x66 11 (op size) and 0x67 (addr size) have an effect 10 on the size of the address and immediate 9 data. The other prefixes will not affect 8 the number of bytes used by the instruction 7 so you can simply ignore them (well count 6 them, but no need to know what follows).
All 5 of that said, using the LLVM library (As 4 someone mentioned in the comments) is probably 3 a better/easier option, although it may 2 be much bigger than what you'd need if your 1 stuff is limited.
There is XED library from Intel to work 36 with x86/x86_64 instructions: https://github.com/intelxed/xed, and it is 35 the only correct way to work with intel 34 machine codes both in x86 and x86_64 modes. It 33 is used by Intel (and was part of their 32 Pin): https://software.intel.com/en-us/articles/xed-x86-encoder-decoder-software-library
https://software.intel.com/sites/landingpage/pintool/docs/67254/Xed/html/main.html XED User Guide (2014) https://software.intel.com/sites/landingpage/pintool/docs/56759/Xed/html/main.html XED2 User Guide 31 (2011)
xed_decode function will provide you all information 30 about instruction: https://intelxed.github.io/ref-manual/group__DEC.html
https://intelxed.github.io/ref-manual/group__DEC.html#ga9a27c2bb97caf98a6024567b261d0652
And xed_ild_decode will only decode 29 instruction for its length:
https://intelxed.github.io/ref-manual/group__DEC.html#ga4bef6152f61997a47c4e0fe4327a3254
XED_DLL_EXPORT xed_error_enum_t xed_ild_decode ( xed_decoded_inst_t * xedd, const xed_uint8_t * itext, const unsigned int bytes )This function 28 just does instruction length decoding. It 27 does not return a fully decoded instruction.
Parameters
- xedd the decoded instruction of type xed_decoded_inst_t . Mode/state sent in via xedd; See the xed_state_t .
- itext the pointer to the array of instruction text bytes
- bytes the length of the itext input array. 1 to 15 bytes, anything more is ignored.
Returns:
xed_error_enum_t 26 indiciating success (XED_ERROR_NONE) or failure. Only 25 two failure codes are valid for this function: XED_ERROR_BUFFER_TOO_SHORT 24 and XED_ERROR_GENERAL_ERROR. In general this 23 function cannot tell if the instruction 22 is valid or not. For valid instructions, XED 21 can figure out if enough bytes were provided to 20 decode the instruction. If not enough were 19 provided, XED returns XED_ERROR_BUFFER_TOO_SHORT. From 18 this function, the XED_ERROR_GENERAL_ERROR 17 is an indication that XED could not decode 16 the instruction's length because the instruction 15 was so invalid that even its length may 14 across implementations.
To get length from 13 xedd struct, filled by xed_ild_decode, use xed_decoded_inst_get_length: https://intelxed.github.io/ref-manual/group__DEC.html#gad1051f7b86c94d5670f684a6ea79fcdf
static XED_INLINE xed_uint_t xed_decoded_inst_get_length ( const xed_decoded_inst_t * p )Return the length 12 of the decoded instruction in bytes.
Example 11 code ("Apache License, Version 2.0", by 10 Intel 2016): https://github.com/intelxed/xed/blob/master/examples/xed-ex-ild.c
#include "xed/xed-interface.h"
#include <stdio.h>
int main()
{
xed_bool_t long_mode = 1;
xed_decoded_inst_t xedd;
xed_state_t dstate;
unsigned char itext[15] = { 0xf2, 0x2e, 0x4f, 0x0F, 0x85, 0x99,
0x00, 0x00, 0x00 };
xed_tables_init(); // one time per process
if (long_mode)
dstate.mmode=XED_MACHINE_MODE_LONG_64;
else
dstate.mmode=XED_MACHINE_MODE_LEGACY_32;
xed_decoded_inst_zero_set_mode(&xedd, &dstate);
xed_ild_decode(&xedd, itext, XED_MAX_INSTRUCTION_BYTES);
printf("length = %u\n",xed_decoded_inst_get_length(&xedd));
return 0;
}
Any other solution like manual 9 prefix/opcode parsing or using third-party 8 disassembler may give you wrong results 7 for some rare cases. We don't know which 6 library is used inside Intel to verify their 5 hardware instruction decoders, but xed is 4 the library used by their software decoders 3 in various binary tools. The ild decoder 2 of xed has more than 1600 lines of code: https://github.com/intelxed/xed/blob/master/src/dec/xed-ild.c, and 1 should be more precise than any other library.
There's a small disassembly library called 4 udis86: http://udis86.sourceforge.net/.
It's small and has decent documentation. If 3 you set the translator to NULL via ud_set_syntax, then the 2 function ud_disassemble should only decode the instruction 1 and return the number of bytes.
More Related questions
We use cookies to improve the performance of the site. By staying on our site, you agree to the terms of use of cookies.