README: known issues (with symbols/error codes for searchability) + roadmap

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
2026-07-18 18:34:54 -04:00
co-authored by Claude Fable 5
parent 359fe4077b
commit 493e4528f3
+46
View File
@@ -53,6 +53,52 @@ Big Picture autostart follows `CONSOLE_KID_USER` in `/etc/bc250-console.conf`.
Modded BIOS (see [mothenjoyer69/bc250-documentation](https://github.com/mothenjoyer69/bc250-documentation)),
then BIOS settings: 512 MB dynamic VRAM, IOMMU disabled, auto power-on.
## Known issues (found the hard way on real hardware)
All of these are worked around in this image; listed here for searchability
and until fixed upstream.
- **GNOME Remote Desktop (RDP/VNC) crashes the whole session** on this GPU:
`gnome-shell` SIGSEGV in `meta_cursor_renderer_get_cursor` via
`meta_screen_cast_monitor_stream_src_set_cursor_metadata` (mutter 50.3,
cyan-skillfish/GFX1013) the moment a client connects and requests
cursor-metadata frames. `MUTTER_DEBUG_DISABLE_HW_CURSORS=1` does **not**
help. *Workaround: Sunshine + Moonlight instead (baked). Upstream report
to GNOME/mutter: pending.*
- **Sunshine crash-loops (SIGSEGV, exit 139) probing the Vulkan H.264
encoder** on a GPU with no video-encode hardware, instead of falling
through to the next encoder. *Workaround: `encoder = software` seeded
before first start (baked). Upstream report to LizardByte: pending.*
- **GeForce NOW's native Linux app cannot run** — error `0xC0F11312`: the
client hard-requires Vulkan video-decode extensions and the BC-250's
silicon has no VCN block at all. Unfixable at the driver level.
*Workaround: GFN via Chromium web app (baked shortcut).*
- **BlueBuild `default-flatpaks@v1` `notify: true` breaks on multi-user
systems**: the post-install notification runs `sudo -u` with all human
users newline-joined into one name, failing the whole install service.
*Workaround: `notify: false`. Upstream report: pending.*
- **Bazzite's `bazzite-steam-bpm` shows "ERROR: code=0; This is
strange..."** at login when Steam lingers from a previous session: a
`waitpid` race on already-gone PIDs plus an `if !` clobbering `$?`.
*Workaround: this image ships its own quit-and-wait wrapper.*
- Cosmetic: `amdgpu` logs a wall of `hpd(rx)`/`dal_irq` errors at every
boot on BC-250 — the display core probes IRQ sources this cut-down chip
lacks. Harmless; every BC-250 does this.
## Roadmap / hoped-for work
- File the four upstream bug reports above (coredumps and stack traces are
in hand).
- In-home game streaming: Sunshine host VM (R730, RTX 3050 passthrough)
serving the baked Moonlight client — self-hosted "GeForce NOW" at LAN
latency. Fortnite stays on real GFN (Epic's anti-cheat policy).
- Thermal soak validation and GPU governor ceiling tuning (see
OPERATIONS.md) once long gaming sessions accumulate.
- Retry GNOME RDP after a fixed mutter ships; retry native GFN if NVIDIA
adds a software-decode fallback.
- Explicit non-goal: the experimental 40-CU unlock — this console stays on
the stable 24-CU configuration.
## Credits
Forked from