Re: [PATCH 3/6] dt-bindings: net: remove outdated hisilicon-femac

[Krzysztof Kozlowski]

On 16/02/2024 09:03, Yang Xiwen wrote:
> On 2/16/2024 3:21 PM, Krzysztof Kozlowski wrote:
>> On 16/02/2024 00:48, Yang Xiwen via B4 Relay wrote:
>>> From: Yang Xiwen <forbidden405@xxxxxxxxxxx>
>>>
>>> The user documented(Hi3516) is not found in current kernel anymore.
>>> Remove this binding entirely due to recent driver changes.
>> Hardware does not change because you decided to re-implement driver.
> 
> The only hardware i have is the hi3798mv200. According to downstream 
> driver name, this is supposed to be a hisi-femac-v3 actually. I don't 
> know much about Hi3516, but it confuses me a lot. According to the 
> device tree node example in the text binding file, the MDIO bus is 
> supposed to be inside the femac core (femac core is at 0x0-0x1000 & 
> 0x1100-0x1300 and mdio bus is at 0x1100-0x1120). So i think it's highly 
> possible they are the same hardware. But according to the TRM and my 
> tests, there are 3 clocks in total for femac core in hi3798mv200, one 
> for mac ctrl, one for ahb bus(I'm sure this "bus" clock is not MDIO bus 
> clock), and one for phy, which is very similar to the hisi-gmac driver.
> 
> Which complicates things a lot is the complex clock enabling timing 
> requirements here. at least for hi3798mv200(and all SoCs with 
> hisi-femac-v3 core i think according to the downstream kernel source), 
> It must strictly follow the sequence in hisi_femac_phy_reset() (disable 
> MAC clk and BUS clk first before asserting PHY reset), or the PHY would 
> fail to work. So as said in previous reply, the simplest way is to do 
> all resets and clocks management in the MAC driver, or else it'll be 
> very hard to implement. I can't find an easy way to "tell" a driver to 
> kindly disable its clocks remotely.
> 

None of these explain why support for these devices should be dropped...

Best regards,
Krzysztof