On Mon, Aug 13, 2018 at 08:18:08PM +0200, Sam Ravnborg wrote:
Hi Nicholas.
On Mon, Aug 13, 2018 at 05:54:48PM +0200, Nicolas Ferre wrote:
On 12/08/2018 at 20:41, Sam Ravnborg wrote:
New DRM based driver for at91sam9 SOC's that uses the Atmel LCDC IP core.
I'm delighted to see this work: Thanks a lot Sam!
Glad to hear. I was a bit worried that the response would be "this is waste of time as we have a working driver already".
This is first version of a patch set that adds drivers for the Atmel LCDC IP core. Posted for review as the basics works now.
The LCDC IP core contains two devices:
- a PWM often used for backlight
- a LCD display controller
Both devices are supported today by the atmel_lcdfb driver. For this new set of drivers the compatible strings was selected to avoid clash with the existing compatible strings used for the atmel_lcdfb driver to allow them to co-exist.
Would be good to have a plan to phase-out the old atmel_lcdfb fbdev driver when this one addresses some TODO items that make sense.
Agree on this. One approach could be to say that when all in-kernel users of atmel_lcdfb are ported, then the old driver could be dropped after a kernel release.
The mfd suffix seems strange to me. What about "atmel,at91sam9263-lcdc-mfd" => "atmel,at91sam9263-lcd" (or "microchip,at91sam9263-lcdc").
The "-mfd" suffix was added to avoid clashing with the current compatible string used by the atmel_lcdfb driver.
I don't know that 2 registers for a backlight PWM constitute an MFD. A single node can be both an LCD controller and a PWM.
The fact that the OS has 2 drivers is irrelevant to the binding and DT is not a way to select drivers. Your choice with Linux is either kconfig or manual module loading.
I susggest we do the following:
- use the microchip prefix, as this is now owned by microchip
- and add the driver to a drm/microchip/ directory
(Then we can only hope that microchip do not change name or are purchased by someone else).
I would not do that. Then you have a dts with a mixture based on when you got around to writing each binding. i.MX has continued using 'fsl' even on new chips today. Probably a good choice had the QCom acq had gone thru.
Rob