Re: [PATCH v5 00/21] drm: starfive: jh7110: Enable display subsystem

[Michal Wilczynski]

Hi,

On 10/12/26 19:59, OOHehir wrote:
> Tested on a VisionFive 2 v1.3B with patches 1-17 & 19-21 of v5 on
> Linux 7.2.8, plus the clk patch that was 14/20 in v4. Patch 18 was not
> applied: it conflicts on 7.2 & there is no Rockchip hardware here. The
> same kernel also carries the out-of-tree PowerVR BXE-4-32 enablement &
> the SiFive XPbmtUC errata patches, so the GPU could render to the
> display.
> 
> - The dc8200, the HDMI controller & PHY & the vout clock controller
>   all probe, with nothing left in deferred probe; HDMI-A-1 reports
>   connected.
> - kmscube (GLES via zink on the PowerVR Vulkan driver) runs at the
>   monitor's preferred mode: 1683 frames in 28.1 s, 59.93 fps with
>   vsync, picture confirmed on the monitor, no GPU firmware resets.
> 
> Only that one mode was tested; the modetest sweep in the cover letter
> was not repeated.
> 
> One error at boot:
> 
>   i2c i2c-7: of_i2c: modalias failure on /soc/display-subsystem@29400000/hdmi@29590000/controller/ports
>   i2c i2c-7: Failed to create I2C device for /soc/display-subsystem@29400000/hdmi@29590000/controller/ports
> 
> inno_hdmi_i2c_adapter() sets adap->dev.of_node to the controller's
> node, so the I2C core registers the controller's children as I2C
> clients, & on the JH7110 that node has a 'ports' child. The display
> still comes up. Should the adapter get no of_node, or a dedicated
> child node?
> 
> For patches 1-17 & 19-21:
> 
> Tested-by: OOHehir <oo.hehir@xxxxxxxxx>
> 

Hi,

Thanks for testing. I'll carry your Tested-by into v6.

The modalias failure reproduces for me as well. It doesn't come from
this series: inno_hdmi_i2c_adapter() has been assigning the adapter's
of_node to the controller's node ever since the driver landed, and the
RK3036 and RK3128 carry the same ports child beneath their hdmi node, so
they should print the same message.

Nothing on the DDC bus is described in DT so the adapter shouldn't
receive an of_node at all - dw-hdmi doesn't set one either. Removing
that line clears the error here with EDID and every mode still
working. I don't think the fix belongs in this series though.

Best regards,
-- 
Michal Wilczynski <m.wilczynski@xxxxxxxxxxx>