Re: [PATCH v5 00/17] Basic StarFive JH7110 RISC-V SoC support

[Matthias Brugger <matthias.bgg@gmail.com>]

On 13/04/2023 12:05, yanhong wang wrote:
> 
> 
> On 2023/4/13 17:03, Torsten Duwe wrote:
>> On Thu, 13 Apr 2023 10:05:28 +0800
>> yanhong wang <yanhong.wang@starfivetech.com> wrote:
>>
>>> the definition of DT refers to Linux and is consistent with the definition framework of Linux.
>>
>> This is one of the desired goals, to avoid confusion, usually. But note there are already the
>> -u-boot.dtsi files; in this case it would be vice-versa: U-Boot could be simple, the kernel
>> required a different treatment. As long as the resulting tree matches the hardware!
>>
>>> The difference between 1.2A and 1.3B is the PHY type and phy clock delay configuration,
>>> which are reflected in DT, and the difference in defconfig is the configuration of the DT file.
>>>
>>> Is defconfig defined separately or merged?
>>
>> You are the implementer, this is your decision. You make a proposal, and it will get accepted
>> or not. We only make suggestions, with the intention to improve the code.
>>
> 
> Thanks. A defconfig matches a piece of hardware, which is more developer-friendly and less confusing,
> so defconfig is better defined separately.
> 

My opinion isIn my opinion user-friendlyness is more important then developer 
friendly that from an end-user point of view it's much easier to have one binary 
that works on all different board versions. If not users will have to know which 
board version they have to flash the correct U-Boot.

For the RaspberryPi we even went further and put effort into U-Boot development 
to have one U-Boot binary which can boot RPi3 and RPi4 boards.

In that sense I would advise you to revisit your decision to put a 
developer-friendly approach over an end-user-friendly one.

As Torsten said, it shouldn't be too difficult to update the device-tree at boot 
time to fit the actual board you are running U-Boot on.

Just my thoughts about the issue :)

Best regards,
Matthias

>>> The EEPROM is being prepared and will be submitted as soon as possible. Is it necessary to
>>> incorporate EEPROM into this submission?
>>>
>>> When eeprom is supported, the MAC address will be read from eeprom. The board reversion
>>> can be read from eeprom, but phy clock delay configuration cannot be read from eeprom, only in DT.
>>
>> But the board revision number in EEPROM can be used to differentiate between 1.2 and 1.3, right?
>>
> 
> Yes, board reversion read from eeprom can distinguish between 1.2A and 1.3B.
> 
> 1.2A and 1.3B are two sets of hardware, and the differences between the hardware are defined
> by DT, which is more concise and clear.
> 
>> When I look at the code and my test results, this is my proposal to pull this in, in order to
>> simplify things and avoid duplication. Whether you do so is up to you, see above. Let me recap:
>>
>> * the device tree *must* match the hardware at hand.
>>
>> * the differences are minor and could be patched, Copy&Waste is error prone and causes extra work.
>>
>> It is my firm conviction that this patch set does not introduce hardware variants, and it would be
>> the task of the ethernet driver patch set to split the code (DT+defconfig) OR to provide a patching
>> method. Maybe I find a few cycles to look at the EEPROM.
>>
>> 	Torsten