Re: [PATCH v1 2/5] pardata: new bus for parallel data access

From: Sam Ravnborg
Date: Wed Aug 08 2018 - 04:24:38 EST


Hi Noralf.

On Tue, Aug 07, 2018 at 06:40:29PM +0200, Noralf Trønnes wrote:
> Hi Sam,
>
> Den 02.08.2018 21.45, skrev Sam Ravnborg:
> >The pardata supports implement a simple bus for devices
> >that are connected using a parallel bus driven by GPIOs.
> >The is often used in combination with simple displays
> >that is often seen in older embedded designs.
> >There is a demand for this support also in the linux
> >kernel for HW designs that uses these kind of displays.
> >
> >The pardata bus uses a platfrom_driver that when probed
> >creates devices for all child nodes in the DT,
> >which are then supposed to be handled by pardata_drivers.
> >
> >Signed-off-by: Sam Ravnborg <sam@xxxxxxxxxxxx>
> >---
>
> From a quick look at this I have these comments:
> 1. There can only be one implementation of this bus, the gpio one.
>    There are SOC's with parallel bus hardware so you need to allow for
>    more implementations.
> 2. The client shouldn't do the bus signaling. This should be hidden
>    behind read and write functions in pardata.
> 3. I would also suggest you add an address bus instead of the RS pin
> 4. I don't think reset belongs in the bus. It's a device thing.

Excellent feedback - I will address this in v2.

> 5. You can use gpiod_set_array_value() in the gpio implementation.
>    Some gpio drivers can set all gpios at once.

Yes, it is already used. But only for DB0 to DB7 as there
are timing constraings for the others.

>
> I made an attempt at implementing a bus like this a while back:
> https://github.com/notro/fbdbi/tree/master/i80

Thanks, very helpfull.

v2 will take a while as I plan to have something that actually works
before posting next version.

One open question. Miguel Ojeda mentioned that there is already a
limited fbdev driver made on top of parport (part of auxdisplay).
Is it the correct design to come up with a new bus or should this
try to build on top of parport?

I did not check in details if using parport is possible, but from a
quick look it is doable.
But then we use parport for something that it originally was not
designed for and we drag with us a lot of extra functionality.
So I like the slimmer pardata bus.

Sam