Re: [PATCH V4 2/2] gpio: inverter: document the inverter bindings

[Eugeniu Rosca <roscaeugeniu@xxxxxxxxx>]

Hi Geert,

On Fri, Sep 27, 2019 at 11:07:20AM +0200, Geert Uytterhoeven wrote:

[..]

> My standard reply would be: describe the device connected to the GPIO(s)
> in DT.  The GPIO line polarities are specified in the device's "gpios"
> properties.
> 
> BTW, can you give an example of what's actually connected to those
> GPIOs?
> Is it a complex device (the GPIO is only a part of it, it's also hanging
> off e.g. an I2C bus)?
> Is it something simple (e.g. an LED ("gpio-leds"), relay, or actuator)?

Since the targeted user of the new feature is not in immediate vicinity,
we expect some delay in getting this information.

> 
> Next step would be to use the device from Linux.  For that to work, you
> need a dedicated driver (for the complex case), or something generic
> (for the simple case).
> The latter is not unlike e.g. spidev.  Once you have a generic driver,
> you can use "driver_override" in sysfs to bind the generic driver to
> your device.  See e.g. commit 5039563e7c25eccd ("spi: Add
> driver_override SPI device attribute").

We have passed your suggestions along. Many thanks.

> Currently we don't have a "generic" driver for GPIOs. We do have the
> GPIO chardev interface, which exports a full gpio_chip.
> It indeed looks like this "gpio-inverter" could be used as a generic
> driver.  But it is limited to GPIOs that are inverted, which rules out
> some use cases.
> 
> So what about making it more generic, and dropping the "inverter" from
> its name, and the "inverted" from the "inverted-gpios" property? After
> all the inversion can be specified by the polarity of the GPIO cells in
> the "gpios" property, and the GPIO core will take care of it[*]?
> Which boils down to adding a simple DT interface to my gpio-aggregator
> ("[PATCH/RFC v2 0/5] gpio: Add GPIO Aggregator Driver",
>  https://lore.kernel.org/lkml/20190911143858.13024-1-geert+renesas@xxxxxxxxx/).
> And now I have realized[*], we probably no longer need the GPIO
> Forwarder Helper, as there is no need to add inversion on top.

After having a look at the gpio aggregator (and giving it a try on
R-Car3 H3ULCB), here is how I interpret the above comment:

If there is still a compelling reason for having gpio-inverter, then it
probably makes sense to strip it from its "inverter" function (hence,
transforming it into some kind of "repeater") on the basis that the
inverting function is more of a collateral/secondary feature, rather
than its primary one. Just like in the case of gpio aggregator, the
primary function of gpio inverter is to accept a bunch of GPIO lines and
to expose those via a dedicated gpiochip. I hope this is a proper
summary of the first point in your comment. In any case, this is the
understanding I get based on my experiments with both drivers.

What I also infer is that, assuming gpio-inverter will stay (potentially
renamed and stripped of its non-essential inverting function), the gpio
aggregator will need to keep its Forwarder Helper (supposed to act as a
common foundation for both drivers).

The second point which I extract from your comment is that the "gpio
aggregator" could alternatively acquire the role of "gpio-inverter"
(hence superseding it) by adding a "simple DT interface". I actually
tend to like this proposal, since (as said above) both drivers are
essentially doing the same thing, i.e. they cluster a number of gpio
lines and expose this cluster as a new gpiochip (keeping the
reserved/used gpio lines on hold). That looks like a huge overlap in
the functionalities of the two drivers.

The only difference which I see is that "gpio-inverter" is getting its
input from DT and generates the gpiochips at probe time, while
"gpio aggregator" is getting its input from sysfs and generates the
gpiochips at runtime, post-probe.

So, assuming no objections from Harish and other reviewers, I would be
very happy to review and test the DT-based gpio inversion functionality
as part of gpio aggregator. Thanks!

-- 
Best Regards,
Eugeniu