hgn / machine-code-analyzer

Machine Code Analyzer for X86(_64) ISA

Machine Code Analyzer

A two pass code analyzer for x86 and x86_64, requires debug symbols for full analyze capabilities.

Modules

Function Local Branch and Jump Analyzer

Analyze local jump instructions. "Local" refers to jumps within the function space - function calls and long jumps are not analyzed by this module.

Following data calcalated:

• Jump analyzer
  • Number of jumps
  • Average jump distance in bytes
  • Minimal jump distance in bytes
  • Maximal jump distance in bytes
  • Histogram of jumps
• Forward jump analyzer
  • Number of jumps
  • Average jump distance in bytes
  • Minimal jump distance in bytes
  • Maximal jump distance in bytes
  • Histogram of jumps
• Number of backward jumps per function
  • Number of jumps
  • Average jump distance in bytes
  • Minimal jump distance in bytes
  • Maximal jump distance in bytes
  • Histogram of jumps

Usage

# instruction-layout-analyzer --function-branch-jump <binary>

Function Anatomy Analyzer

Started as an internal module for other modules this module provides information about the function start and end addresses, the size of a function and the number of functions.

Usage

# instruction-layout-analyzer --function-anatomy <binary>

FAQ

Sometimes instruction analyzer and readelf differs from the code size - why?

Consider the following end of an function:

405021:       5d                      pop    %rbp
405022:       c3                      retq
405023:       66 2e 0f 1f 84 00 00    nopw   %cs:0x0(%rax,%rax,1)
40502a:       00 00 00
40502d:       0f 1f 00                nopl   (%rax)

What you see here are alignmend padding in form of NOP instructions (13 byte). The code actual function end at retq instruction. The subsequent code is now aligned at 0x405030, better for cachelines, etc. The question now is: should the instruction analayzer account the padding instructions or not? Feel free and verify code with

readelf -s <binary>

NOPs are mostly generated and added by GAS (GNU assembler if using GCC). There are other ways as well to decode NOPs: xchg %ax,%ax (two byte), leal 0(%esi),%esi (three byte).

Last note: the assembler should always pad with as less instruction as possible to relax CPU prefetcher and subsequent CPU logic.

JSON to Chart.js Converter

To generate charts a helper script comes bundled with instruction-layout-analyzer.

ToDo

Function Gap Analysis

Size in bytes from function end to next function start. Function alignment characteristics and the resulting effect of "wasted" memory is of interest. Especially for small and tiny functions