On Mon, Dec 11, 2023 at 03:19:08PM +0200, Andy Shevchenko wrote:

On Fri, Dec 08, 2023 at 09:18:20PM +0100, Noralf Trønnes wrote:

quoted

On 12/8/23 17:00, Andy Shevchenko wrote:

quoted

Included authors and latest (non-white-space) contributors to the drivers
in question along with relevant mailing list and respective (active in the
area) maintainers.

I already had risen the question in times when 4th (sic!) driver for the same
hardware was about to be pulled into upstream that we have to somehow reduce
the code base and unify device properties.

So, the main question here "What is the plan and where are we now?"

I admit that fbtft case is special as it supports, in particular, platform
device (parallel interface) and also well established in the embedded world.
What about the rest?

N.B. Besides the fact that panel drivers are too OF-centric, which is bad
practice for the new kernel code in general and has to be stopped. I.o.w.
seeing of_property_*() or alike in the driver after ca. 2020 should be
immediate NAK unless it's very well justified why it may not be used on
non-OF systems.

Noralf, thanks for your response, my comments below.

TBH I would also like to hear from maintainers, because it seems they got
an additional burden for no benefit.

quoted

Last year drivers/gpu/drm/tiny/panel-mipi-dbi.c was added to support all
MIPI DBI compatible (ili9341) SPI displays.
It loads the initialisation commands from a firmware file. For more info
see https://github.com/notro/panel-mipi-dbi/wiki.

When I started on fbtft in 2013 I didn't know about MIPI DBI so I made
some common bus access functions and one driver per controller and that
driver had an initialisation sequence to match the panel I had. Then I
discovered that displays using the same controller could have different
init sequences so I added a Device Tree <init> property that could
override the driver init.

In 2015 fbtft was added to drivers/staging, but later that year fbdev
was closed for new drivers so it was a dead end.

I started to work on porting fbtft to DRM and almost 2 years later
support for the MI0283QT panel (ILI9341) was added.
I had now learned about MIPI DBI so a library to handle that was added.
I had asked on the Device Tree ML about the <init> property and I was
told that I couldn't have that which meant that I couldn't get away with
having just one driver for the MIPI DBI compatible display panels as I
was first hoping for.

I was aware that there was a challenge going from fbtft to DRM because
in fbtft there is support for all panel setups using the <init>
property, but in DRM every panel needed support in a driver. So I
started to look at adding Device Tree properties to describe the setup
for one controller. This would make it easy to describe a new panel in
Device Tree for a supported controller. Maxime Ripard came up with the
idea to have the controller initialisation commands in a firmware file
which meant that we could get away with having just one driver for all
MIPI DBI SPI panels (which is the vast majority of these SPI pixel
upload panels).

This meant that SPI support could be removed from all the MIPI DBI
compatible controllers in fbtft

I believe it can't. Otherwise we _must_ provide the DT (device property)
parser that uses what is provided for fbtft SPI to be enabled in the other
driver.

We never committed to supporting that binding in the first place, and
that code is in staging, so I don't see why we must do that.

quoted

since there's now a solution for them in DRM. The drivers themselves
must stay since they also have parallel bus support which is lacking
in DRM. My plan was to wait for panel-mipi-dbi to hit an LTS and
then either prepare patches to remove MIPI DBI SPI support from
fbtft or at least send an email to staging about the new driver.

quoted

Unfortunately my health problems got worse and many plans went
out the window.

Oh, sad to hear this, hope you will get better sooner than later!

quoted

ILI9341 DRM drivers

- drivers/gpu/drm/tiny/mi0283qt.c
  This was the first driver added for the MI0283QT panel series

- drivers/gpu/drm/tiny/ili9341.c
  Later ili9341 based panels was decided to be added to a controller
  specific driver.

Why was it appeared in the first place? :-(

It's probably an oversight. The name was far from obvious to begin with.

quoted

- drivers/gpu/drm/tiny/panel-mipi-dbi.c
  Generic MIPI DBI SPI driver that loads init commands from a firmware
  file. It uses of_property_read_string_index() and
  of_get_drm_panel_display_mode(). I don't know if it's possible to make
  device_property_*() versions of those.

Everything like this is possible, just somebody needs to fulfill that.
And as I said, new OF-centric code, has to be NAKed by default.

quoted

- drivers/gpu/drm/panel/panel-ilitek-ili9341.c
  This driver supports the MIPI DPI (RGB) interface on the controller.
  Controller init is done over MIPI DBI SPI. The driver does also for
  some reason support the same panel as the ili9341.c driver does.
  So 2 drivers for the same panel...
  Sidenote: It is possible to make a generic panel-mipi-dpi.c driver for
  panels that use DPI for pixels and DBI for init loaded from a firmware
  file.

I wonder who has enough experience and time to at least point out or do
something about this...

I'm not sure it's worth it tbh. Panel drivers are fairly straightforward
and easy to maintain, and getting a driver that can support all the
variations of the power sequences those driver need would make a very
complicated driver.

And MIPI-DPI display are becoming more and more irrelevant. Small,
cheap, display will keep using MIPI-DBI, and middle and high-end screens
are using MIPI-DSI/eDP.

Maxime