Hi Michel,
On Tue, Apr 17, 2018 at 9:56 AM, Michel Pollet
<michel.pollet@xxxxxxxxxxxxxx> wrote:
> On 13 April 2018 19:06, Rob Herring:
>> On Tue, Apr 10, 2018 at 09:30:03AM +0100, Michel Pollet wrote:
>> > The Renesas RZ/N1 Family (Part #R9A06G0xx) has a multi-function system
>> > controller. This documents the node used to encapsulate it's sub
>> > drivers.
>> >
>> > Signed-off-by: Michel Pollet <michel.pollet@xxxxxxxxxxxxxx>
>> > ---
>> > .../bindings/mfd/renesas,rzn1-sysctrl.txt | 23
>> ++++++++++++++++++++++
>> > 1 file changed, 23 insertions(+)
>> > create mode 100644
>> > Documentation/devicetree/bindings/mfd/renesas,rzn1-sysctrl.txt
>> >
>> > diff --git
>> > a/Documentation/devicetree/bindings/mfd/renesas,rzn1-sysctrl.txt
>> > b/Documentation/devicetree/bindings/mfd/renesas,rzn1-sysctrl.txt
>> > new file mode 100644
>> > index 0000000..9897f8f
>> > --- /dev/null
>> > +++ b/Documentation/devicetree/bindings/mfd/renesas,rzn1-sysctrl.txt
>> > @@ -0,0 +1,23 @@
>> > +DT bindings for the Renesas RZ/N1 System Controller
>> > +
>> > +== System Controller Node ==
>> > +
>> > +The system controller node currently only hosts a single sub-node to
>> > +handle the rebooting of the CPU. Eventually it will host the clock
>> > +driver, SMP start handler, watchdog etc.
>>
>> Please submit a complete binding for the h/w block.
>>
>> Again, if the only reason you have sub nodes is to define compatible strings
>> and in turn enumerate drivers, then you don't need the nodes in DT. DT is
>> not the only way to instantiate drivers.
>
> I can't document it before I have the code. There is 0.000% chance of my clock
> driver for example to be upstreamed the way I would imagine making it -- in
> fact pretty much any other driver will have to be reworked to fit, so documenting
> bindings first is impossible.
>
> So, if I understand correctly, you are telling me to make a 'sysctrl' driver and use
> platform_device to instantiate my sub-drivers? Isn't that what machine files used
> to do? And they are now banned?
>
> Geert, any guidance here?
It depends on how many and which subnodes you want to add to the sysctrl
node. Without a complete binding for the block, we cannot know.
If the major part will be the clock driver, I would make that the main
driver for the sysctrl node. The clock driver can easily register
e.g. a simple reset handler.
Cfr. the renesas-cpg-mssr driver, which also handles (module) resets.
There are plenty of other examples of drivers providing multiple
functionalities (e.g. pinctrl drivers also registering GPIO controllers).
If a monolithic driver becomes too large, it can still be split using the
MFD framework.
E.g. the BD9571 PMIC driver is split in an MFD core driver, and 2 drivers
for different functionalities:
drivers/gpio/gpio-bd9571mwv.c
drivers/mfd/bd9571mwv.c
drivers/regulator/bd9571mwv-regulator.c
include/linux/mfd/bd9571mwv.h
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
