Hi Olivier,

> Am 09.11.2026 um 18:46 schrieb Olivier Louvignes <[email protected]>:
> 
> The display controller of the Ingenic X1600 is not the one ingenic-drm
> drives: it fetches frames through descriptors and keeps its interface and
> timings in register blocks of their own. It is the X2000's DPU without
> the composer. This series adds a binding and a driver for it, then the
> panel of the HiBy R1, an X1600 audio player: 480x800 parallel RGB, set up
> over a 3-wire SPI bus.
> 
> The driver covers what the R1 uses and what could be tested: parallel RGB
> panels on a 24-bit bus and one full-screen RGB565 plane. The composer,
> smart panels and other pixel formats are left out.
> 
> Two things come from the vendor kernel rather than the manual. A flip
> also writes SRD_CHAIN_START, which the manual doesn't describe for a
> running controller; flips without it were not tried. The panel's set-up
> sequence is the vendor's too: HiBy calls the panel LG35583, its maker and
> controller are unknown, hence the hiby prefix.
> 
> Changes in v4:
> - 3/5: the clock runs from probe to unbind (devm_clk_get_enabled()), as
>  in ingenic-drm, instead of being gated with the CRTC (Paul). The
>  controller only takes register writes while clocked, so the timings
>  and the output format can now be set in the encoder's
>  atomic_mode_set(), and the bus_flags and color_order fields go. The
>  burst length is set once in probe, and atomic_disable() no longer has
>  a failed clock start to handle.
> - 3/5: probe brings the clock down to at most a 720p pixel clock right
>  after enabling it. After a reset it runs at its parent's rate, 1.4 GHz
>  on the R1, and kept that fast until the first modeset the controller
>  hung the R1 at every cold start.
> - 3/5: atomic_enable() takes the mode from the new CRTC state, there is
>  no writel(readl()) nesting, and drm_mode_config_create_initial_state()
>  replaces drm_mode_config_reset() (Paul).
> - 3/5: mode_valid() still checks the pixel count against 1280x720 rather
>  than 720 per side, which would refuse the R1's 480x800 panel (Paul
>  agreed).
> - 3/5: the file is ingenic-dpu-x1600.c, and the module and driver are
>  ingenic-dpu-x1600 (Nikolaus).
> - 5/5: unprepare() turns vcc off before vccio, the reverse of prepare()
>  (Sashiko).
> - Rebased on drm-misc-next.
> 
> Changes in v3:
> - The compatible is ingenic,x1600-dpu, and the binding and the driver say
>  "display controller" rather than "LCD controller" (Nikolaus). The X2000
>  and X2600 manuals call this block the DPU, and the X1600 manual does too
>  in its register and clock names. There is no generic ingenic,dpu
>  fallback, as writing-bindings.rst asks for SoC compatibles: an X2000
>  binding can fall back to the X1600's. The driver keeps its file name;
>  splitting it for the X2000's composer and interfaces belongs with
>  X2000 support.
> - 2/5: no $nodename and no clock-names; the driver takes its one clock
>  unnamed (Krzysztof). The example's node is display-controller@.
> - 2/5: there is no IPU in front of this controller (Krzysztof asked on v2,
>  Paul answered): it fetches its own frames, so the output port is its
>  only link.
> - 4/5: reset-gpios, backlight and port dropped, panel-common has them
>  (Krzysztof). The example's endpoint points at the controller's new
>  label.
> - Krzysztof's Acked-by on 1/5 and Reviewed-by on 4/5.
> - Rebased on drm-misc-next after -rc1.
> 
> Changes in v2, all in 3/5 and all found by the Sashiko review of v1:
> - Commits wait for flip_done instead of a vblank before the planes are
>  cleaned up. A flip's event waits until the controller reports reading
>  the new descriptor, which can be a frame after the vblank that v1's
>  default commit_tail waited for, so the old framebuffer could be freed
>  while still scanned out.
> - A CRTC being enabled sends its event from atomic_enable(), once the
>  controller runs; v1 sent it from atomic_flush(), which runs first.
> - A plane that is not visible is rejected: the controller always scans
>  out a whole frame, so v1 would have read past the end of a small
>  framebuffer placed off-screen.
> - atomic_disable() returns early when atomic_enable() failed to start
>  the pixel clock, instead of touching an unclocked controller.
> 
> Tested on a HiBy R1: the fbdev console and a player drawing through
> fbdev, blanking and unblanking, 60 page flips in 0.97 s at about 61 Hz,
> and the colour order (RGB order swaps red and blue on this panel). For
> v2 also: 120 fbdev pans in 1.95 s (one flip per frame, none timing out),
> three blank/unblank cycles, and a 64x64 framebuffer set on the primary
> plane just past the right edge through SETPLANE, which v1 accepted and
> v2 refuses with -EINVAL, while a full-screen one is still accepted. v3
> was run on the R1 with its node changed to match: the driver binds to
> ingenic,x1600-dpu with its clock unnamed, and the player draws. For v4:
> four blank/unblank cycles with the clock left running and the timing
> registers kept, 60 pans in 0.98 s and 120 in 1.97 s with none failing,
> an unbind and bind of the driver, which turns the clock off and then
> writes the timings again from atomic_mode_set() on the first modeset,
> and five cold starts with the kernel loaded into RAM through the boot
> ROM's USB mode and started as the boot loader does.
> 
> The X1600 has no device tree upstream yet; its support is being posted
> one subsystem at a time. The R1's backlight uses the X1600 PWM driver
> posted separately, now at v3:
> https://lore.kernel.org/r/[email protected]
> 
> v3: https://lore.kernel.org/r/[email protected]
> v2: https://lore.kernel.org/r/[email protected]
> v1: https://lore.kernel.org/r/[email protected]
> 
> Olivier Louvignes (5):
>  dt-bindings: vendor-prefixes: Add HiBy
>  dt-bindings: display: Add the Ingenic X1600 display controller
>  drm/ingenic: Add a driver for the X1600 display controller
>  dt-bindings: display: panel: Add the HiBy LG35583
>  drm/panel: Add the HiBy LG35583
> 
> .../bindings/display/ingenic,x1600-dpu.yaml   |  61 ++
> .../bindings/display/panel/hiby,lg35583.yaml  |  67 ++
> .../devicetree/bindings/vendor-prefixes.yaml  |   2 +
> MAINTAINERS                                   |  12 +
> drivers/gpu/drm/Makefile                      |   1 +
> drivers/gpu/drm/ingenic/Kconfig               |  22 +
> drivers/gpu/drm/ingenic/Makefile              |   1 +
> drivers/gpu/drm/ingenic/ingenic-dpu-x1600.c   | 739 ++++++++++++++++++
> drivers/gpu/drm/panel/Kconfig                 |  10 +
> drivers/gpu/drm/panel/Makefile                |   1 +
> drivers/gpu/drm/panel/panel-hiby-lg35583.c    | 342 ++++++++
> 11 files changed, 1258 insertions(+)
> create mode 100644 
> Documentation/devicetree/bindings/display/ingenic,x1600-dpu.yaml
> create mode 100644 
> Documentation/devicetree/bindings/display/panel/hiby,lg35583.yaml
> create mode 100644 drivers/gpu/drm/ingenic/ingenic-dpu-x1600.c
> create mode 100644 drivers/gpu/drm/panel/panel-hiby-lg35583.c
> 
> 
> base-commit: 8deaade4a5a4ff635df91e246e42139ab2b33f6d
> -- 
> 2.39.5

Finally I could set up a demo with the sharp,lq070y3dg3b panel-simple
which would be a 33 MHz pixel clock 480x800 panel.

Since we have no connector and just test points I could only verify that
there is a pixel clock signal, sync and if I write random data to /dev/fb0
there are varying signals on the RGB pads.

BR and thanks,
Nikolaus

Reply via email to