May 21, 2026

Created in May 21, 2026

2026   ·   gpu-security   rowhammer   ·   daily-log

Goals

  • Find out more exploitable regions inside the libtriton.so library.

What I worked on

  • It is very time consuming to clone and build the llvm project so instead of cloning the entire repository with its git history intact I just clone a shallow copy of the remote and just git fetch the required commit. This way I have drastically reduced the cloning time but the build time is still the bottleneck as the build is single-threaded, I can increase the jobs to let’s say an emperical number like 7.
  • I have realized the framework I created contains a lot of slop, so I will go ahead and create the whole framework from scratch :)

Key Insights

  • The design of a framework should be seamless.

Questions / Confusions

  • I still need to how predicate in ptx can be manipulated to skip instructions.

Ideas / Hypotheses

  • I can generalize the attack surface of LLVM to different targets like AMDGPU, RISC-V etc.

Papers / Resources

  • None

Next Steps

  • Create the framework from scratch with better or at least easily understandable design.

Introspection

  • Today was a good day I destroyed two fluent but sloppy frameworks.