Board-level Linux integration, U-Boot, device trees, kernel drivers and peripheral bring-up. Separate power and hardware checks from software adaptation, and define a recoverable boot path.
Why can a top-level Yocto configuration fragment appear in list-fragments without affecting the build, or remain active after a setup command is no longer in the shell history?
How can cleaning recipe-native delete the source archive still expected by the target recipe even though that target’s do_fetch stamp survives?
Why does a native Wayland HMI receive unconfigured_buffer or invalid_serial when it attaches its first image immediately after creating xdg_toplevel?
Why is a Wayland frame callback insufficient evidence that an HMI update was displayed or that its buffer is reusable?
Why can a Qt HMI that works with several desktop OpenGL windows terminate or become fullscreen when launched through EGLFS?
Why can an exported DMA-BUF fence complete while another graphics context has already submitted newer conflicting work, and what must serialize export, submit and import?
Why can two embedded graphics components agree on a FourCC yet disagree on the buffer layout, and why is an omitted modifier different from explicitly linear?
Why can an embedded device write a temporary core file while its coredump policy says Storage=none?
How can one fsync report a delayed writeback error and a later call return zero without proving the affected application data was repaired?
Why can a service’s configured cpuset stay unchanged while the CPUs actually granted to it change or fall back to its parent’s available set?