Re: [PATCH] gpio: zynq: Setup chip->base based on alias ID

[Linus Walleij]

On Thu, Apr 26, 2018 at 3:35 PM, Michal Simek <michal.simek@xxxxxxxxxx> wrote:

>> The only use case which I can think about is userspace sysfs
>> and then I would really like to know why these userspace
>> users cannot use the character device that is nowadays
>> supported by libgpiod and there is even patches for some
>> IoT libraries to use it. The character device makes the
>> GPIO Linux "base" irrelevant for userspace.
>>
>> GPIO sysfs is deprecated and moved to the obsolete ABI.
>>
>> If there are legacy applications that use this I would have
>> to consider it, but since this has been -1 since the driver
>> was merged I find that unlikely.
>
> Yes, it is about legacy application which I have seen recently and there
> is no source code for application calls it because board vendor doesn't
> provide it.
>
> You are right that -1 was used from the beginning in mainline but
> unfortunately this driver was in vendor tree for a while and it uses 0
> there.
>
> In upstreaming this was changed to -1 but customers have a lot of code
> which developed against vendor tree and they want to use
> latest&greatest. And without this they are not able to run that
> applications.
>
> I found that this logic is already in 5 drivers in mainline that's why I
> send this patch to be +1.

I see.

Sadly comaptibility with out-of-tree driver code is none of our
(community) business.

We do pay a lot of effort not to break the ABI to userspace, but
it needs to be an ABI coming from the mainline kernel, not from
a vendor tree.

So to the mainline kernel this is no regression.

Yours,
Linus Walleij