Re: SG2000 display: DSI host + LT8912B HDMI working on v7.2, intent to upstream

[nick@detello.com]

(Resent in plain text; the list rejected the HTML copy.)

On 2026-10-13 02:24, nick@detello.com wrote:
> Hi Chen, Inochi,
> 
> I have the SG2000 display path working on v7.2 with two small kernel
> patches: DISP, DSI MAC and D-PHY driven by a new DRM host driver,
> through the in-tree lontium-lt8912b bridge, to HDMI at 1280x720@60
> over
> two lanes. It's tested on both cores (RISC-V and ARM) on the Nix Badge
> 2.0,
> a custom board around the Milk-V Duo S module: console on HDMI and a
> 15 fps framebuffer animation.
> 
> Two questions before I prepare the submission:
> 
> 1. Is either of you, or anyone you know of, working on the VO/DSI
> block
>    for cv18xx/SG2000? I would rather join than duplicate.
> 2. Any preference on where the host driver lives
>    (drivers/gpu/drm/sophgo/?) and on compatibles before I write the
>    bindings? The branch uses sophgo,sg2000-dsi and a
>    sophgo,sg2000-vip-sys syscon, and adds one clock ID
>    (CLK_DSI_SYNTH_SRC) to the cv1800 binding header.
> 
> Code, as one reviewable PR against the board's NixOS repo:
>   https://github.com/NixVegas/BadgeOS/pull/8
>   Driver:  pkgs/video/sophgo-dsi/src/sophgo-dsi.c (~1,500 lines,
>            register provenance from sophgo/osdrv cited in place)
>   Patches: pkgs/kernel/patches/ (clk-cv18xx: poll DISPPLL's real lock
>            bit and expose the DSI source gate; lt8912b: do the bridge
>            setup in pre_enable, between the host's LP-11 and HS)
>   DT:      pkgs/firmware/dts/sg2000-milkv-duo-s-common.dtsi
> 
> Planned order: the clk fix; display nodes for sg2000.dtsi and for the
> Duo S DTS Joshua Milas's series added in July; the DSI host as an RFC
> to dri-devel; then the lt8912b change, which as written is specific to
> this host (fixed HDMI mode, LVDS and PM paths dropped) and will be
> rebased on Maxime's atomic-callback conversion on drm-misc-next before
> it goes anywhere. Known gaps before the RFC: fixed mode (no EDID, no
> hot-plug; HPD goes to the bridge on this board), no vblank interrupt,
> and register sequences reproduced from the vendor driver rather than
> modelled.
> 
> Happy to share a video, serial logs or register traces. The driver was
> developed with Claude Code assistance; I have read and tested every
> line and will carry the review.
> 
> Thanks,
> Nick DeTello