Dead code CONFIG_ARMADA_39X?

[Stefan Roese]

On 04.03.21 14:58, Pali Rohár wrote:
> On Thursday 04 March 2021 14:50:56 Stefan Roese wrote:
>> On 04.03.21 14:41, Pali Rohár wrote:
>>> On Thursday 04 March 2021 14:29:46 Stefan Roese wrote:
>>>> On 03.03.21 11:21, Pali Rohár wrote:
>>>>> Hello!
>>>>>
>>>>> I see in more U-Boot files check for ifdef CONFIG_ARMADA_39X but I do
>>>>> not see that CONFIG_ARMADA_39X could be defined in some header or board
>>>>> code.
>>>>>
>>>>> So does it mean that all code under ifdef CONFIG_ARMADA_39X is dead? Has
>>>>> U-Boot support for Marvell A39X SoC?
>>>>>
>>>>> If it is really dead code, should not be ifdef CONFIG_ARMADA_39X
>>>>> completely removed?
>>>>
>>>> Frankly, I don't remember the history here. Did you look into the git
>>>> history to see, where this Kconfig option was introduced?
>>>
>>> There is no Kconfig option for ARMADA_39X.
>>>
>>> Only C source files are checking if CONFIG_ARMADA_39X is defined or not.
>>> But there is no code which can define CONFIG_ARMADA_39X, neither header
>>> file nor Kconfig.
>>>
>>> That is suspicious for me.
>>
>> I agree. Thanks for looking into it.
>>
>>>> It could very well be the case, that this was introduced "by accident"
>>>> by including some Marvell code without taking it out. AFAIK, we are not
>>>> supporting any Armada 39x in mainline right now.
>>>
>>> So it looks like that somebody introduced code #ifdef CONFIG_ARMADA_39X
>>> on more places "by accident".
>>>
>>> For example in commit edb470253346f4a882ba9e891c8b102ce388b9cc were
>>> added some these ifdefs and commit was authorized by you. So I thought
>>> that you would know more...
>>
>> Autsch. That was in 2015 - sorry my memory does not cover history
>> that long (anymore). ;)
> 
> Ok :)
> 
>>> So if mainline U-Boot does not support Armada 39x, does it make sense to
>>> remove all code hidden under CONFIG_ARMADA_39X? Following command could
>>> do it:
>>>
>>>     git ls-tree -r --name-only HEAD | xargs unifdef -m -UCONFIG_ARMADA_39X
>>
>> I agree in general. But would it make syncing with Marvell original
>> code  more difficult?
> 
> I was synchronizing only ddr3 training code for a38x. And during
> synchronization I already called unifdef with tons of undef options. So
> adding another one does not make it more difficult.
> 
> And about synchronizing other parts of code I have no idea if it
> happened...

The high-speed serdes code comes to my mind. But anyways, please create
such a removal patch and let's review it on the list then.

Thanks,
Stefan