Codex VS Code sidebar stuck on logo on AMD Wayland; disabling hardware acceleration fixes it
Environment
- VS Code: 1.127.0
- Codex IDE extension (
openai.chatgpt): 26.707.41301 - Platform: Linux x64, GNOME Wayland
- GPU: AMD Phoenix1 using
amdgpu
Issue
After an automatic extension update, the Codex settings page stopped opening. After trying a rollback and reinstall, the Codex sidebar became permanently stuck on the logo/loading screen.
The extension installation and signature verification succeeded. The extension activated, spawned codex app-server, and received the initialize request, but the WebView renderer did not mount.
The system journal contained a large GPU error storm (about 479,000 occurrences):
ContextResult::kFatalFailure: TransferBuffer::Initialize() failed
ContextResult::kFatalFailure: CommandBufferHelper::AllocateRingBuffer() failed
Reinstalling or changing the extension version did not resolve the problem.
Workaround
Enabling VS Code software rendering in argv.json fixed the issue:
"disable-hardware-acceleration": true
After restarting VS Code:
- Codex logged
app routes mounted - Codex logged
ready provider mounted - the sidebar loaded normally
- no new GPU fatal failures occurred
Expected behavior
The Codex WebView should render normally with VS Code hardware acceleration enabled, or fail with a useful diagnostic instead of remaining on the logo indefinitely.
This may be related to #32041, but this report adds a reproducible AMD/Wayland GPU failure signature and a confirmed software-rendering workaround.
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗