On 12/09/2018 11:54:07+0100, Lee Jones wrote:

On Wed, 12 Sep 2018, Geert Uytterhoeven wrote:

quoted

On Wed, Sep 12, 2018 at 10:41 AM Lee Jones [off-list ref] wrote:

quoted

On Wed, 12 Sep 2018, Alexandre Belloni wrote:

quoted

On 11/09/2018 23:54:40+0100, Lee Jones wrote:

quoted

quoted

quoted

http://ww1.microchip.com/downloads/en/DeviceDoc/Atmel-6438-32-bit-ARM926-Embedded-Microprocessor-SAM9G45_Datasheet.pdf

USART doc starting p572, registers p621.

After looking at the datasheet, I don't see any reason why one of the
two drivers can't be selected using different compatible strings.

Because there is only one IP and we don't use the device tree to selecet
linux specific drivers.

We do it all the time.  There are loads of MFDs (def: same IP, with
different functions) which have separate compatibles for their various
functions.  If you wish this IP to operate as an SPI controller, it
should have an SPI compatible, if you wish it to operate as a U(S)ART,
then it should have a UART compatible.  It's what we do for most of
the other MFDs in the kernel.

There is a big difference: MFD functions are(more or less) independent
functions, which can be used at the same time. It makes perfect sense for a
single IP block that has both SPI and UART interfaces, that can be used at
the same time.

In this case, there is a single piece of hardware that can perform
different functions, but not at the same time. Performing a different
function means configuring the hardware for that function, hence using a
different driver (from a different subsystem).

Yes, I can see that PoV.

But ... we can't have it both ways.  *Either* it's a true MFD, in
which case it can/should have 2 separate compatible strings which can
be specified directly from the DT.  *Or* it's not an MFD.  In the
latter case, which I think we're all agreeing on (else we'd have 2
compatible strings), MFD is not the place to handle this (my original
point).

If that is what bothers you, then let's move it out of mfd.

So ... this is a USART device which can do SPI, right?

My current thinking is that; as this is a USART device first &
foremost, the USART should be probed in the first instance regardless,
then if SPI mode is specified it (the USART driver) registers the SPI
platform driver (as MFD does currently) and exits gracefully, allowing
the SPI driver to take over.

Spanner in the works: is it physically possible to change the mode at
run-time?  :s

Yes it is possible but on Linux that will not happen without probing
the drivers again. I think DT overlays will be the only possible use
case because on SPI, you'd still have to provide nodes for the connected
SPI devices.

-- 
Alexandre Belloni, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com