First physical U-Boot splash trial

On 2026-09-22, the source-built U-Boot displayed the K230 asset on the physical panel while stopped at K230#, before any Linux load. This proves stage-1 rendering. It does not prove a seamless Linux handoff.

Artifacts and scope

• Photograph extracted at video t=25 s.
• Raw 40 s camera recording and capture metadata.
• Serial transcript held at U-Boot and host scheduling timestamps.
• Subsequent Linux boot, late shell checks, and transition recording.
• Guarded installer and verified readback.
• USB host and UMS regression test: PASS before and after UMS, expected exported card, candidate bytes read back, Linux login reached, helper exit status 0. This tests enumeration and recovery, not Ethernet packet connectivity.

The trial replaced the U-Boot slot, DTB and logo on the identified running card. It did not flash a second card or either full-image console variant; OpenSpec task 3.6 remains unchecked because its prescribed second-card procedure has not been completed. SPL, environments, OpenSBI, kernel and rootfs were retained. The previous full U-Boot slot and DTB were backed up on the board and verified before writing. The host also retains the complete GC-rooted USB/portrait recovery image documented in recovery-usb-portrait-build.txt.

The trial used the built 0001–0004 U-Boot patch series from source cc3c37c; the later runtime handoff flag patch 0005 was not in this binary.

What the board showed

U-Boot read exactly 2799104 bytes, printed RM69A10 direct XRGB8888 logo.xrgb full-screen OSD4, and stopped at its prompt. The t=25 s frame clearly shows K230. No Starting kernel occurs in that held transcript; Linux was launched in the separate subsequent capture. This is a serial-initiated warm reboot, not a battery-only cold boot or a physical-touch test. The photograph is rotated 180 degrees for presentation; the original camera video is retained unchanged.

Linux recorded the new 0x10000000..0x103fffff 4096 KiB no-map reservation and the same 512 MiB CMA allocation at 0x1e000000. The existing shell service returned active with one online CPU. The late command used rg, which is absent on the board; its final filtering command failed. The reservation and CMA claims come from the full boot transcript, not that failed command.

Handoff defect remains open

The native compositor screenshot shows the expected Apps, Windows, Keyboard, System row and blue title strip. The physical panel after the splash shows a horizontally wrapped row and changed colors. The first recording’s pre-reboot frames show the correct shell on the same camera, so this is a real trial regression to investigate, not a successful clean handoff.

Possible retained VO/DSI state is a hypothesis, not an established cause. The current kernel has no runtime splash-preservation changes. A comparison boot with the logo renamed out of the loader’s path is recorded in without-logo-boot.txt. Leave splash handoff tasks open until the physical presentation is corrected and measured. No ArUco stability measurement was made in this trial.

The no-logo control boot reached Linux and the physical comparison photograph shows the correct Apps, Windows, Keyboard, System order and blue title strip again. U-Boot explicitly logged logo.xrgb unavailable or unsafe, skip U-Boot logo. Only the logo was disabled; the new U-Boot and reserved-memory DTB remained. The board is left with the asset saved as /boot/logo.xrgb.disabled-first-trial until the handoff defect is resolved.

Presentation clips and approximate timeline

Rotated U-Boot clip and rotated Linux transition clip are 960x540 H.264/yuv420p exports with faststart. The originals above retain full camera resolution and orientation. These are feature evidence assets, not a claim that the continuous-splash feature is ready to ship.

Numerically ordered samples show the previous shell at t=0–6 s, shutdown text at 8–10 s, a dark panel at 12 s and the K230 logo by 15 s, still visible at 39 s in the first clip. The separate transition clip shows the logo at 0–10 s and boot text by about 12 s. It ends before the shell appears; the later still records the shell. These are coarse video-relative observations, not power-on latency or calibrated refresh measurements. Camera glare and focus limit readability. See the U-Boot timeline and Linux timeline.

The board configuration now explicitly sets k230.panelConsole = true so subsequent daily shell images omit logo.xrgb. This carries the observed no-logo workaround into reproducible image builds. The splash package and runtime handoff work remain available for an explicit experimental build; changing the default back requires physical handoff evidence.

Mainline splash on the board, 2026-10-12

the-mainline-kernel-keeps-the-boot-splash tasks 2.1 and 2.2, recorded in docs/evidence/mainline-boot-splash/board/README.md (warm reboots, in place on the board’s one card, by operator decision):

• 2.1: mainline-boot-splash/board/splash-at-uboot-before-kernel.jpg, taken 8 s before bootm/Starting kernel in mainline-boot-splash/board/idle-kernel-serial-excerpt.txt.
• 2.2 measurement (U-Boot target at K230#, {3, 97, 0x17, 0x87}, camera px, unregistered as the 0x87 rows were):

The daily mainline system now defaults to the splash (k230-mainline-drm-shell sets k230.panelConsole = false); the vendor-kernel k230 configuration keeps the explicit true described above.