On 05/24/2018 09:13 AM, Linus Walleij wrote:

On Fri, May 18, 2018 at 9:29 AM, Amelie DELAUNAY [off-list ref] wrote:

quoted

On 05/17/2018 08:36 AM, Lee Jones wrote:

quoted

On Wed, 16 May 2018, Amelie DELAUNAY wrote:

quoted

On 05/16/2018 04:20 PM, Linus Walleij wrote:

quoted

On Wed, May 9, 2018 at 9:56 AM, Amelie DELAUNAY [off-list ref] wrote:

quoted

Indeed, stmfx has other functions than GPIO. But, after comments done
here: [1] and there: [2], it has been decided to move MFD parent/GPIO
child drivers into a single PINCTRL/GPIO driver because of the following
reasons:
- Other stmfx functions (IDD measurement and TouchScreen controller) are
not used on any of the boards using an stmfx and supported by Linux, so
no way to test these functions, and no need to maintain them while they
are not being used.
- But, in the case a new board will use more than GPIO function on
stmfx, the actual implementation allow to easily extract common init
part of stmfx and put it in an MFD driver.

So I could remove gpio sub-node and put its contents in stmfx node and
keep single PINCTRL/GPIO driver for the time being.
Please advise,

I would normally advice to use the right modeling from the start, create
the MFD driver and spawn the devices from there. It is confusing
if the layout of the driver(s) doesn't really match the layout of the
hardware.

I understand that it is a pain to write new MFD drivers to get your
things going and it would be "nice to get this working really quick
now" but in my experience it is better to do it right from the start.

Hi Linus,

Thanks for your advice. I understand the point.
So, the right modeling would be to:
- create an MFD driver with the common init part of stmfx
- remove all common init part of stmfx-pinctrl driver and keep only all
gpio/pinctrl functions.

I will not develop the other stmfx functions (IDD measurement driver and
TouchScreen controller driver) because, as explained ealier, they are
not used on any of the boards using an stmfx and supported by Linux, so
no way to test these functions, and no need to maintain them while they
are not being used.

Lee, are you OK with that ?

I missed a lot of this conversation I think, but from what I've read,
it sounds fine.

I summarize the situation:
- I still don't have an official datasheet for STMFX device which could
justify the use of an MFD driver;
- the MFD driver will contain the STMFX chip initialization stuff such
as regmap initialization (regmap structure will be shared with the
child), chip initialization, global interrupt management;
- there will be only one child (GPIO/PINCTRL node) for the time being.

But there will be more devices in it. And they will invariably be put
to use later, and there will be new versions of the chip as well, and
then you will be happy about doing the MFD core, which makes it
easy to add new variants with different subdevices.

quoted

So, is "MFD driver + GPIO/PINCTRL driver" the right modeling, and does
it still sound fine after this summary ? :)

No I think it should use an MFD core.

Mainly because of device tree concerns.

The main reason is that the device tree bindings will be different if
you add an MFD core later, the GPIO and pinctrl driver will
move to a child node, making old device trees incompatible.

We could have a single driver in GPIO+pin control if it is a child
of an MFD node in the device tree, but it doesn't make much
sense as the I2C device need to be probing to the MFD core.

I agree with you Linus, and that's why all STMFX chip initialization 
stuff was decorrelated in pinctrl-stmfx. This shows that this stuff 
needs to be in an MFD core.

But as there is only one child for now (due to the reasons mentioned 
earlier), it can suggest that it is not a Multi-Function Device.

I'm not able to target when IDD or TS functions will be required on a 
Linux product, but it still makes sense to consider that these functions 
will be used on a Linux product.

So, I think MFD core + GPIO/pinctrl driver is the right modeling, but I 
wanted to be sure that this is okay for everyone. I don't want to spend 
time on something that will not be accepted due to its modeling.

Regards,
Amelie