A panel is a byte string and eleven numbers

the contract

Every property the driver reads

Where it reads each one, and what it becomes. The two rows at the bottom are the interesting ones: the settings the driver does not read, because it hardcodes them.

the blob

Reading the init sequence

The blob is a flat stream of three-byte headers, each followed by its payload. No terminator and no count: the parser walks it twice, once to count commands and once to build them, and rejects the whole property if the lengths do not land exactly on the end.

20 commands, in order, drawn in proportion to payload 0xB4 · 45 B and 0xE0 · 55 B in violet — a third of the blob

The vendor also issues two DCS reads after the sequence, of the error register and the pixel format, apparently for debugging. The panel-init-sequence binding is write-only, so there is nowhere to put them. One further command, a solid blue test pattern, is sent only when the vendor’s test_mode_en is set, and is not part of normal initialisation.

lifecycle

When the reset line actually moves

This is the part that will cost someone an afternoon. The reset pulse happens in probe, once. The reset code inside prepare — the callback DRM invokes every time the panel comes back — is present in the source and entirely commented out.

comparison

The two panels, number for number

Left column is what the reference .dtsi contains. Right column is what the vendor’s connector table says about the RM69A10 — the same numbers the shipped RT-Smart firmware runs on, which makes them observations of working firmware rather than datasheet claims.

the real thing

The fragment we wrote

An earlier draft of this note carried a fragment that compiled and was structurally wrong: the panel sat at the root of the tree rather than inside &dsi, the lane count used the driver’s internal field name instead of the binding, the panel node carried its own reset-gpios, and both the endpoint graph and &vo { vth_line } were missing. Compiling in isolation proved the syntax parsed; it said nothing about whether the driver would ever see the node. This is the corrected file.

// nix/dts/display-rm69a10-568x1232.dtsi (abridged: the 20-command blob
// is shown with its four largest payloads elided for width)

&vo {
    vth_line = <9>;            // intr_line from the vendor connector table; ST7701 uses 10
};

&dsi {
    ports {
        port@1 {
            reg = <1>;
            dsi_out_rm69a10: endpoint { remote-endpoint = <&rm69a10_in>; };
        };
    };

    lcd: panel@0 {                // a DSI child, and labelled so the board can reach it
        compatible = "canaan,universal";
        reg = <0>;

        // DIVERGENCE: the reference puts PIXEL COUNTS in these. These are real
        // millimetres, computed from the stated 4.1" diagonal — not measured.
        panel-width-mm  = <44>;
        panel-height-mm = <95>;

        panel-dsi-lane  = <2>;

        // framing: data_type, delay_ms, payload_length, payload…
        // delay is a u8 of MILLIseconds. The vendor waits 300 us after 0x11
        // and 100 us after 0x29; neither is expressible, so both become 1 ms,
        // which is longer than the vendor waits.
        panel-init-sequence = [
            39 00 04 b9 ff 83 99
            15 00 02 d2 aa
            39 00 10 b1 02 04 71 91 01 32 33 11 11 ab 4d 56 73 02 02
            39 00 10 b2 00 80 80 ae 05 07 5a 11 00 00 10 1e 70 03 d4
            39 00 2d b4 00 ff 02 c0 02 c0 …          45 bytes
            39 00 22 d3 00 00 00 00 00 00 …          34 bytes
            39 00 21 d5 18 18 19 19 18 18 …          33 bytes
            39 00 21 d6 18 18 19 19 40 40 …          33 bytes
            39 00 11 d8 a2 aa 02 a0 a2 a8 02 a0 b0 00 00 00 b0 00 00 00
            15 00 02 bd 01
            39 00 11 d8 b0 00 00 00 b0 00 00 00 e2 aa 03 f0 e2 aa 03 f0
            15 00 02 bd 02
            39 00 09 d8 e2 aa 03 f0 e2 aa 03 f0
            15 00 02 bd 00
            39 00 03 b6 8d 8d
            15 00 02 cc 04
            39 00 03 c6 ff f9
            39 00 37 e0 00 12 1f 1a 40 4a …          55 bytes — the gamma table
            05 01 01 11                              exit sleep, then 1 ms
            05 01 01 29                              display on, then 1 ms
        ];

        display-timings {
            timing-0 {
                clock-frequency = <39600000>;
                hactive = <568>;   vactive = <1232>;
                hfront-porch = <40>;  hback-porch = <140>;  hsync-len = <40>;
                vfront-porch = <16>;  vback-porch = <16>;   vsync-len = <4>;
            };
        };

        ports {                     // the other half of the endpoint pair
            #address-cells = <1>;  #size-cells = <0>;
            port@0 {
                reg = <0>;
                rm69a10_in: endpoint { remote-endpoint = <&dsi_out_rm69a10>; };
            };
        };
    };
};

// NOT HERE, deliberately: no reset-gpios and no backlight GPIO on the panel
//   node. The board .dts attaches both to the 'lcd' label, which is how the
//   reference splits panel from board.

open

What still has to be settled