fix(viewer): restart to recover /dev/fb0 hotplugged after a headless boot

[vpetersson]

Problem

On linuxfb boards (32-bit Pi 1/2/3, Qt5), wait_for_framebuffer in bin/start_viewer.sh waits for /dev/fb0 before launching the viewer — this is the #2959 guard that stops the headless crash-loop.

But the (privileged) viewer container's /dev is a devtmpfs snapshot taken at container start. If the container boots headless and a display is plugged in later, the host gains /dev/fb0 but it never appears inside the container's stale /dev — so the wait loop blocks forever and the viewer never starts. The comment claimed the wait "self-heals on hotplug"; in a running container it silently does not.

Reproduced live on the Pi 2 testbed: plugging a monitor into a headless-booted board left the host with /dev/fb0 while the container saw only /sys/class/graphics/fb0, and the viewer stayed parked (still no /dev/fb0 after 6540s; waiting for a display) until a manual docker restart.

Fix

sysfs is bind-through from the host and stays truthful across the stale snapshot. When /sys/class/graphics/fb0 exists but the container's /dev/fb0 does not, exit non-zero so the restart: always policy recreates the container with a fresh /dev that includes fb0 — the same exit-to-restart recovery the OOM and viewer-death paths already rely on. Genuinely headless devices (no sysfs fb0 either) keep waiting quietly, so the #2959 behavior is preserved.

Note /dev/fb0 and /sys/class/graphics/fb0 are the same fbdev object (created/removed together by register_framebuffer/unregister_framebuffer), so a real headless box has neither.

Validation (live on the Pi 2, DEVICE_TYPE=pi2, linuxfb/Qt5)

Exercised the real wait_for_framebuffer against each state:

State	Setup	Result
Display present	/dev/fb0 present	returns 0, proceeds
Hotplug-stale	/dev/fb0 removed, /sys/.../fb0 present	prints message, exit 1
Headless	neither present	keeps waiting, never exits

bash -n clean.

[Copilot]

Copilot wasn't able to review any files in this pull request.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[Copilot]

Copilot wasn't able to review any files in this pull request.

[vpetersson]

Folded into #3075 — the two fixes are the pre-launch and post-launch bookends of the same linuxfb HDMI-hotplug problem and belong together. The wait_for_framebuffer headless-boot recovery is now commit daa7f8d on #3075's branch.