Re: [PATCH 1/4] input: Add new sun4i-lradc-keys drivers

[Dmitry Torokhov <dmitry.torokhov-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]

On Fri, Jan 03, 2014 at 06:36:05PM +0100, Maxime Ripard wrote:
> On Thu, Jan 02, 2014 at 11:36:33PM +0100, Hans de Goede wrote:
> > Hi,
> > 
> > On 01/02/2014 09:20 PM, Maxime Ripard wrote:
> > >On Thu, Jan 02, 2014 at 02:45:29PM +0100, Hans de Goede wrote:
> > >>>Also, instead of inventing yet another vendor-specific property, why not re-use
> > >>>a button binding similar to gpio-keys like:
> > >>>
> > >>>        lradc: lradc@01c22800 {
> > >>>                compatible = "allwinner,sun4i-lradc-keys";
> > >>>                reg = <0x01c22800 0x100>;
> > >>>                interrupts = <31>;
> > >>>                allwinner,chan0-step = <200>;
> > >>>
> > >>>		#address-cells = <1>;
> > >>>		#size-cells = <0>;
> > >>>
> > >>>		button@0 {
> > >>>			reg = <0>; /* your channel index from above */
> > >>>			linux,code = <115>; /* already used as dt-property */
> > >>>		};
> > >>>
> > >>>		button@1 {
> > >>>			reg = <1>;
> > >>>			linux,code = <114>;
> > >>>		};
> > >>
> > >>Ugh no. Having a vendor specific property which is KISS certainly
> > >>beats this, both wrt ease of writing dts files as well as wrt the
> > >>dts parsing code in the driver.
> > >
> > >I'd agree with Heiko here. This is pretty much the same construct
> > >that's already in use in other input drivers, like gpio-keys.
> > 
> > In the gpio case there is a 1 on 1 relation between a single hw
> > entity (the gpio-pin) and a single keycode. Here there is 1 hw entity
> > which maps to an array of key-codes, certainly using an array rather
> > then a much more complicated construct is the correct data-structure
> > to represent this.
> 
> You can build an array in your driver out of this very easily, it's 10
> lines in your probe. And you gain from this something that is more
> generic, can be shared by other similar drivers and is consistent with
> what is already in use.

How will it be shared? Surely not code-wise, but basically in spirit
only. It seems to me that the originally proposed binding is simple and
concise and works well for the driver.

> 
> > >This is also something that can really easily be made generic,
> > >since this is something that is rather common.
> > >
> > >Speaking of which. I believe this should actually come in two
> > >different drivers:
> > >   - The ADC driver itself, using IIO
> > >   - A generic button handler driver on top of IIO.
> > >
> > > The fact that on most board this adc is used for buttons doesn't make
> > > any difference, it's actually a hardware designer choice, we should
> > > support that choice, but we should also be able to use it just as an
> > > ADC.
> > 
> > No, this is not a generic adc, as mentioned in the commit msg, this
> > adc is specifically designed to be used this way.
> > 
> > The adc won't start sampling data, and won't generate any interrupts
> > until a button is pressed. That is until the input voltage drops below
> > 2/3 of Vref, this is checked through a built-in analog comparator, which
> > hooks into the control logic.
> > 
> > It has button down and button up interrupts, and can detect long
> > presses (unused) and generate a second type of down interrupt for those.
> > 
> > This really is an input device, which happens to use an adc.
> 
> Hmm, yes, ok.
> 
> > >Carlo Caione already started to work on an IIO driver for the LRADC:
> > >https://github.com/carlocaione/linux/tree/sunxi-lradc
> > >maybe you can take over his work.
> > 
> > That won't work because the adc won't sample if the input gets above
> > 2/3 of Vref. There may be some other mode which does not do that, but
> > that is not clearly documented.
> > 
> > Even if an IIO driver turns out to be doable, I strongly believe that
> > having a separate input driver for this is best, since this device
> > was designed to be used as such. Building input on top of IIO would
> > mean polling the adc, while with my driver it actually generates
> > button down / up interrupts without any polling being involved.
> 
> Not really. iio_channel_read calls the read_raw function (in that
> case) of your driver. If the read_raw function in your driver wants to
> poll the device, fine, but most of the time, it will just block
> waiting for an interrupt to come and return the data to the caller,
> which is obviously the saner behaviour, and you don't actually end up
> polling the device. Which is pretty much the architecture you're using
> already, just with an intermediate layer in between.

What is the benefit of the IIO layer if device can't really be used as
IIO? I am all for moving as many generic devices as we can to IIO but we
should recognize that sometimes the device is not an IIO device.

Thanks.

-- 
Dmitry