Re: [PATCH 3/4] arm64: add support for i.MX8M EVK board

[Dong Aisheng <aisheng.dong-sgV2jX0FEOL9JmXXK+q4OQ@public.gmane.org>]

On 2018-01-25 19:09, Lucas Stach wrote:
> Am Donnerstag, den 25.01.2018, 18:49 +0800 schrieb Dong Aisheng:
>> On 2018-01-25 18:31, Lucas Stach wrote:
>> > Am Donnerstag, den 25.01.2018, 18:10 +0800 schrieb 
>> > aisheng.dong-sgV2jX0FEOL9JmXXK+q4OQ@public.gmane.org:
> [...]
>> AFAIK we switched to generic pinconfig since MX7ULP as maintainer 
>> > > won't
>> > > access old binding pinctrl drivers.
>> >
>> > I'm not convinced that the generic pinconf is good fit. For pingroups
>> > with different configs for some of the pins, like the example above, we
>> > would need to split things into multiple DT nodes. This really hurts
>> > readability, so I'm not going to switch to the generic stuff without
>> > some really convincing arguments.
>> >
>> 
>> Per my understanding, based on the last discussion with Linus W, we 
>> actually did this in order to increase the readability that 1) user 
>> does
>> not need to see the 'ugly' unreadable raw data and refer to reference
>> manual 2) unified generic binding format which already exist in kernel
>> and used by many platforms.
>> 
>> Actually MXS platform already used it for many years in a similar way.
>> So IMHO a little hurt to add another node for different pad setting 
>> in 
>> the same group won't be enough reason to stop switching to generic 
>> config.
>> 
>> Does it make sense?
> 
> I know that Linus W is pushing for this common pinconf thing in the
> name of readability. It's just that I don't think it's such a clear
> win.
> 
> After all you still need to look into the reference manual or binding
> to see which values in the common binding correspond to a specific
> drive/pull strength, etc.
> 

User don't need to look into reference manual and they don't need to
compose the 'ulgy' raw data which is the most tough thing.

With generic binding, it probably can saving ~80% pad setting effort
by refer to the defined generic config properties.
And things can be even better when the reference code is already there
as user becomes know which property supported.

> On the other hand it really bloats the DT description of the pin
> configuration. If you want to look at an (IMHO) bad example, go look at
> the Tegra DTs. The Tegra pincontrol implements the "separate properties
> for each pinconf option" that is pushed by Linus W. This bloated the DT
> description to the point that no-one is able/willing to write those
> descriptions anymore and the only viable way to get them is to auto-
> generate them from some spreadsheets. Not really what I would call an
> readable...
> 

I wonder the worst case you're worrying whether exist in reality.
Take imx6qdl-sabresd as an example, about half of pingroups having the 
same
pad setting while others have two different settings at most.
That means it may not bloat the device tree too much.

> Maybe I'm a little stubborn when it comes to this topic, but at
> Pengutronix we see a lot of customer designs where we need to come up
> with the board DT. Bloating each one of those and making the work of
> the developers harder in the name of a readability win that I just
> don't see doesn't sound like something I want to support. :)
> 

Hmm.. In contrast, what i feel currently is that it may ease the using 
of
pad setting, not make it harder. Not sure if i overlooked something.

Let's listen to Shawn and Linus W if they have some comments.

Regards
Dong Aisheng

> Regards,
> Lucas
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html