Re: [PATCH v2 0/2] Exynos850 GPU support

["Alexey Klimov" <alexey.klimov@linaro.org>]

On Fri Aug 28, 2026 at 10:35 AM BST, Peter Griffin wrote:
> Hi Thomas, Alexey, Sam
>
> On Wed, 26 Aug 2026 at 20:44, Thomas Turner <tturner@lineageos.org> wrote:
>>
>> On 26/08/2026 20:02, Alexey Klimov wrote:
>> > On Tue Aug 25, 2026 at 10:05 PM BST, Thomas Turner wrote:
>> >> Hi Sam and Alexey,
>> >>
>> >> On 19/08/2026 20:51, Sam Protsenko wrote:
>> >>> Can you please tell us a bit about how you tested this patch, and on
>> >>> which board? As DPU is not currently enabled on Exynos850 SoC, I'm not
>> >>> sure how exactly GPU can be useful. If you have a use case or a
>> >>> testing procedure we can run on the E850-96 board, please share.
>> >>>
>> >>
>> >> Sam, Alexey could you please test GPU on the E850-96 board by following
>> >> these instructions:
>> >>
>> >> 1) Apply this patch series locally
>> >> 2) Change status="disabled"; to status="okay";
>> >> 3) Make sure these configs are enabled:
>> >> CONFIG_DRM_SIMPLEDRM=y
>> >> CONFIG_DRM_PANFROST=y
>> >> 4) In terms of regulators we set up a basic mali regulator via our
>> >> secondary bootloader [1] and configured with this dts setup [2], NOTE
>> >> you may be able to get away with a dummy regulator so try that first.
>> >> If not I am sure you can replicate something simple with u-boot PMIC.
>> >> 5) Boot a GUI, I saw Debian was used for the bring-up, that should work
>> >> 6) Run the following command:
>> >> $ dmesg | grep "panfrost"
>> >> 7) Compare that to the output at [3]
>> >
>> > Your email subject still seems to be null.
>> >
>> > [   15.206193] panfrost 11500000.gpu: clock rate = 377000000
>> > [   15.206322] panfrost 11500000.gpu: error -ENODEV: _opp_set_regulators: no regulator (mali) found
>> > [   15.206513] panfrost 11500000.gpu: supply mali not found, using dummy regulator
>> > [   15.206840] panfrost 11500000.gpu: mali-g52 id 0x7212 major 0x0 minor 0x0 status 0x0
>> > [   15.213137] panfrost 11500000.gpu: features: 00000000,00000df7, issues: 00000000,00000400
>> > [   15.221271] panfrost 11500000.gpu: Features: L2:0x07100206 Shader:0x00000000 Tiler:0x00000809 Mem:0x1 MMU:0x00002830 AS:0xff JS:0x7
>> > [   15.233080] panfrost 11500000.gpu: shader_present=0x1 l2_present=0x1
>> > [   15.240488] panfrost 11500000.gpu: [drm] Using Transparent Hugepage
>> > [   15.250497] [drm] Initialized panfrost 1.6.0 for 11500000.gpu on minor 0
>>
>> You seemed to have missed out step (4) of my instructions, if you set-up
>> regulators correctly this error will disappear.
>>
>> > [root@e850usb ~]# uname -a
>> > Linux e850usb 7.2.0-next-20260825-00009-g647b48e80572-dirty #8 SMP PREEMPT Wed Aug 26 17:05:45 BST 2026 aarch64 GNU/Linux
>> >
>> > I have different DRM_SIMPLEDRM options and related  enabled but nothing
>> > pops up via HDMI (at least I tried simple xorg-server things).
>> > Is it possible that your bootloader initilises some framebuffer and inserts
>> > it into DTB that passed to kernel? Or you have some other pieces in
>> > kernel/dt that implement missing functionality?
>>
>> Our bootloader does intialise a framebuffer, and we have a node in our device
>> tree which makes it usable. Not sure how exactly the e850-96 device sets
>> things up but you can take a look at our fork if you want, [1].
>>
>> > Here it seems there are no outputs:
>> > [root@e850usb ~]# ls -al /dev/fb*
>> > ls: cannot access '/dev/fb*': No such file or directory
>> >
>> > and there no simle framebuffer or related logs. "DPU is not currently
>> > enabled on Exynos850 SoC" looks like the reason.
>>
>> hm, I am pretty sure our bootloader configures DPU for us so that might be
>> why it works for us and not for you.

Yes, exactly, and see below too please.

> @Alexey - I think the LK bootloader leaves DPU framebuffer updates off
> to ensure a seamless handover from the bootsplash to the full DRM
> driver. We have the following DPU poke in gs101 (which if I recall
> correctly we got the inspiration for this from uniloader)
> https://gitlab.com/LinaroLtd/googlelt/meta-bsp-google-gs/-/commit/4057117e4bc6e08fbb6ff281cbfef8782a68ec0a
>
> So for the E850 board, you likely need to do something similar, either
> in LK or like the above poke. If UniLoader is already doing this for
> the e850 SoC then hopefully you can check there to see the address of
> the register etc. That should then able you to have simple-framebuffer
> enabled.

Am not an expert in grapics but from all observations I see, that even
if bootloader (say LK) sets up DECON/DPU and ley them stream DSI,
then we are still missing support for Lonthium MIPI/DSI-to-HDMI bridge
to output anything via HDMI on Winlink E850-96 board.
There is high-speed connector though.

For smartphones with simple panels with DSI interface all that actually
could work if bootloader takes care about initial set up (exactly what is
done in referenced repos of uniLoader AFAICS).

BR,
Alexey