So I recently read [1] which evaluates several disassemblers. The truth/test binaries are generated by the SPEC CPU 2006 Benchmark suite. The authors provide detailed build instructions for the ELF set inside a VirtualBox.

However, reconstructing the Windows binaries seems impossible for me, as the authors didn't provide the SPEC configuration files for building those binaries. So building the binaries with Visual Studio 2015, which actually match with the provided mapping files [2] didn't work (bruteforcing/guessing the compilation settings).

For example, most of the samples "SPEC/vs15-32/C++" and "SPEC/vs15-32/C" have the following prologue of the .text base (see appendix for properties description):

@0x0000000000401000:  [FBIC]
@0x0000000000401001:  [IC]C
@0x0000000000401003:  [IC]CC
@0x0000000000401006:  [IC]
@0x0000000000401007:  [IC]
@0x0000000000401008:  [IC]CC
@0x000000000040100b:  [IC]C
@0x000000000040100d:  [JIC]C

All of the 64 bit samples have following prologue:

<Segment .text, vaddr 0x0000000000001000, size 1600112, flag [RX]>
@0x0000000140001000:  [FBIC]CCCC
@0x0000000140001005:  [IC]CCCC
@0x000000014000100a:  [IC]
@0x000000014000100b:  [IC]CCC
@0x000000014000100f:  [IC]CC
@0x0000000140001012:  [IC]CC
@0x0000000140001015:  [IC]CC
@0x0000000140001018:  [JIC]C

I tried several configurations for the SPEC build instructions with VS2015 Compiler and intel compiler. However, inspecting all of the resulting binaries don't match with the provided ground truth of [1].

So two concrete questions:

  1. When all samples share similar .text prologues, they all share similar libraries?

  2. Could somone infer the compilation details by the repeating prologues?

Properties descriptions:

  d - data
  c - code
  i - instruction boundary
    Note that if a byte is an instruction boundary (start of an instruction),
    this implies that it is a code byte
  o - instruction boundary (start of overlapping instruction)
  b - basic block start
    Basic block boundaries are not always explicitly listed, as they can usually
    be found by parsing the instruction/function listing into a control-flow graph
  f - function start
  e - program/binary entry point
  r - function end (return, tail call, etc.)
  j - control-flow instruction (jmp, call, ret, ...)
  x - crossref/call instruction
  n - NOP or other function padding

[1] An In-Depth Analysis of Disassembly on Full-Scale x86/x64 Binaries by Dennis Andriesse:

Zeus is a family of credential-stealing trojans which originally appeared in 2007. The first two variants of Zeus are based on centralized command servers. These command servers are now routinely tracked and blocked by the security community. In an apparent effort to withstand these routine countermeasures, the second version of Zeus was forked into a peer-to-peer variant in September 2011. Compared to earlier versions of Zeus, this peer-to-peer variant is fundamentally more difficult to disable. Through a detailed analysis of this new Zeus variant, we demonstrate the high resilience of state of the art peer-to-peer botnets in general, and of peer-to-peer Zeus in particular.

1 Answer 1


We implemented a generator for creating such ground truth mappings and recently published it.

You can find further details within the github repo:


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.