Re: [PATCH v2 3/3] ARM64: dts: meson-gx: Add MALI nodes for GXBB and GXL

From: Andreas FÃrber
Date: Thu Mar 02 2017 - 13:01:01 EST


Hi,

Am 02.03.2017 um 13:47 schrieb Neil Armstrong:
> On 03/02/2017 01:31 PM, Andreas Färber wrote:
>> Am 01.03.2017 um 11:46 schrieb Neil Armstrong:
>>> For GXL, since a lot is shared with the GXM that has a MALI-T820 IP, this
>>> patch adds a new meson-gxl-mali.dtsi and is included in the SoC specific
>>> dtsi files.
>>
>> This part is slightly confusing though.
>>
>> What exactly is the GXL vs. GXM difference that this can't be handled by
>> overriding node properties compatible/interrupts/clocks? I am missing a
>> GXM patch in this series as rationale for doing it this way.
>>
>> In particular I am wondering whether the whole GXM-inherits-from-GXL
>> concept is flawed and should be adjusted if this leads to secondary
>> .dtsi files like this: My proposal would be to instead create a
>> meson-gxl-gxm.dtsi, that meson-gxl.dtsi and meson-gxm.dtsi can inherit
>> the current common parts from, then the Mali bits can simply go into
>> meson-gxl.dtsi without extra #includes needed in S905X and S905D. While
>> it's slightly more work to split once again, I think it would be cleaner.
>
> The GXL and GXM differences are very small :
> - They share the same clock tree
> - They share the same pinctrl and even the same pinout (S905D and S912 are pin-to-pin compatible)
> - They share all the peripherals
>
> The only changes are :
> - Enhanced video encoding and decoding support, this will need a family-specific compatible when pushed
> - Slightly differences in the Video Processing Unit, this is why I introduced family-specific compatibles
> - A secondary Cortex-A53 cluster
> - A secondary SCPI cpufreq clock entry
> - A different Mali core, but with the same interrupts (less but they share the same lower interrupts), clocks and memory space
>
> This is why it was decided to have a sub-dtsi, having a secondary dtsi will simply copy 99% of the GXL dtsi,
> but surely we could also have an intermediate dtsi but for boards I'm ok with it, but less for a SoC dtsi,
> since it could lead to some confusion.
>
> Finally, yes I could have added the mali node to the GXL dtsi, but the midgard Mali dt-bindings are not upstream
> and the family is too big and recent enough to consider having stable bindings for now.

OK, my question really was specific to Mali differences. :)

> Nevertheless, nothing is final, this gxl-mali.dtsi could be merged into the GXL dtsi in the future when we
> have proper dt-bindings and a real support of the T820 Mali on the S912.

What about a /delete-node/ &mali; in meson-gxm.dtsi?
That would avoid having any new .dtsi.

Regards,
Andreas

--
SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)