Re: [PATCH v4 1/6] dt-bindings: spi: spi-rockchip: add description for rv1126 and rk3568

From: Kever Yang
Date: Tue Jun 08 2021 - 22:48:05 EST

Hi Heiko, Johan,

On 2021/6/7 下午5:04, Heiko Stübner wrote:
Your comment in [PATCH v3 3/8]:
Adding "rockchip,rv1126-spi" to rockchip_spi_dt_match[] is strictly not
needed when using "rockchip,rk3066-spi" as fall back string.
Could a maintainer advise?

Maybe this bug of mine should revert too?? Or is it legacy?
spi: rockchip: add compatible string for px30 rk3308 rk3328
I agree with you. If the maintainer doesn't have any comments, I will use
"rockchip,spi" as compatible names for the subsequent rk platform.
Compatibility strings are supposed to be SoC orientated.
So generic ones like in the manufacturer tree can't be used here.
Johan ist right :-) .

rockchip,spi won't work at all, especially as these controllers always change
over time. [0]

Best example is the iommu. We started with "rockchip,iommu" thinking this
won't change over time, but with the rk3568 we get a new slightly different

Rockchip SPI and SFC controller can use a generic compatible string, because there is a version

register inside the IP, and all the feature update will have a new IP version, so the driver is

used for the SPI/SFC IP  in all SoCs, we don't need to care which SoC is using this driver.

If we have to use the compatible string "rockchip,rk3066-spi" and each for a new soc, then we

have to update the driver compatible id list and document for each soc which is totally not need

and not correct  to do it.

The example "iommu" is different, because there is no version register inside the IP and the IP

can not identify itself, which need a software define "-vX".


- Kever

The vendor-kernel then introduces somewhat random "-vX" additions to
distinguish them, but often they do seem to be very software-centric.

Meaning, hardware-designers moved stuff around and software-developers
then invented the versioning to differentiate between versions.

The devicetree is supposed to describe the hardware though, so going with
the relevant soc-specific compatible gives us the necessary hardware-centric

Also this allows to catch later issues with specific soc implementations ;-)
Like 6 monts down the road we discover some special behaviour on the
rk3568 and devicetree is supposed to be stable.

So having the relevant compatibles in place allows us to just add driver
fixes and have those apply on the rk3568 if that is need at some point.


Linux-rockchip mailing list