[Linux/Nobara/AMD] ChatGPT 26.818.21641 repeatedly SIGSEGVs in GPU, renderer, and bundled Codex processes
Nobody has claimed this yet.
- Dominant language
- Rust
- Stars
- 125k
- Forks
- 19.4k
- PR merge metrics
- PR metrics pending
Description
Codex app version
26.818.21641 (RPM), with bundled codex-cli 0.148.0-alpha.21
ChatGPT plan
Unknown / not relevant to the native crash
Platform
Nobara 44 (Fedora 44), Linux 7.1.4-200.nobara.fc44.x86_64, KDE Wayland, AMD Navi 10 GPU, Mesa 26.2.0
What happened?
The Linux Codex app crashed repeatedly during normal use on August 20. systemd-coredump retained eight SIGSEGVs from /usr/lib/chatgpt/ChatGPT across GPU, renderer, and utility processes, plus three SIGSEGVs from the bundled resources/codex binary.
One GPU-process core has this top frame:
amdgpu_bo_slab_free (libgallium-26.2.0.so + 0xbaddc9)
pb_slabs_reclaim_locked
pb_slab_alloc_reclaimed
amdgpu_bo_create
si_alloc_resource
The bundled Codex crashes had top offsets +0x438b251, +0xc2205ea, and +0xc21f5b4. The last two were also logged by the kernel as general protection faults in tokio-rt-worker.
Steps to reproduce
I do not have a deterministic minimal reproduction. Run the app normally on this AMD/Mesa system; the crashes occurred across several process types during one session.
Expected behavior
The app and its bundled Codex process should remain running without native faults.
Additional information
This host also showed cross-process memory-corruption symptoms that day, including an unrelated compiler Bad page state and a kernel Oops in lruvec_stat_mod_folio. NVMe SMART is clean and a full Btrfs scrub found no errors. Because of that, hardware or kernel corruption may be contributing; I am reporting the repeatable app-specific signatures without claiming every event is an independent Codex defect.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the retained systemd-coredump reports for /usr/lib/chatgpt/ChatGPT and the bundled resources/codex binary, including the listed native offsets and tokio-rt-worker faults. Reproduce normal use on Nobara with the AMD Navi 10 GPU, KDE Wayland, and Mesa 26.2.0 while checking whether the app-specific crashes recur. Done means the relevant native faults are characterized and the app and bundled Codex process remain running without them.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- linux, rust
- Domain
- computer-graphics, operating-systems
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100