Re: [PATCH 2/2] dt-binding: spi: Document Renesas R-Car RPC controller bindings

[Boris Brezillon <boris.brezillon@xxxxxxxxxxx>]

On Mon, 19 Nov 2018 23:29:00 +0100
Marek Vasut <marek.vasut@xxxxxxxxx> wrote:

> On 11/19/2018 11:25 PM, Boris Brezillon wrote:
> > On Mon, 19 Nov 2018 23:22:45 +0100
> > Marek Vasut <marek.vasut@xxxxxxxxx> wrote:
> >   
> >> On 11/19/2018 11:19 PM, Boris Brezillon wrote:  
> >>> On Mon, 19 Nov 2018 23:11:31 +0100
> >>> Marek Vasut <marek.vasut@xxxxxxxxx> wrote:
> >>>     
> >>>> On 11/19/2018 04:21 PM, Boris Brezillon wrote:    
> >>>>> On Mon, 19 Nov 2018 16:12:41 +0100
> >>>>> Marek Vasut <marek.vasut@xxxxxxxxx> wrote:
> >>>>>       
> >>>>>> On 11/19/2018 03:43 PM, Boris Brezillon wrote:      
> >>>>>>> On Mon, 19 Nov 2018 15:14:07 +0100
> >>>>>>> Marek Vasut <marek.vasut@xxxxxxxxx> wrote:
> >>>>>>>         
> >>>>>>>> On 11/19/2018 03:10 PM, Boris Brezillon wrote:        
> >>>>>>>>> On Mon, 19 Nov 2018 14:49:31 +0100
> >>>>>>>>> Marek Vasut <marek.vasut@xxxxxxxxx> wrote:
> >>>>>>>>>           
> >>>>>>>>>> On 11/19/2018 11:01 AM, Mason Yang wrote:          
> >>>>>>>>>>> Document the bindings used by the Renesas R-Car D3 RPC controller.
> >>>>>>>>>>>
> >>>>>>>>>>> Signed-off-by: Mason Yang <masonccyang@xxxxxxxxxxx>
> >>>>>>>>>>> ---
> >>>>>>>>>>>  .../devicetree/bindings/spi/spi-renesas-rpc.txt    | 33 ++++++++++++++++++++++
> >>>>>>>>>>>  1 file changed, 33 insertions(+)
> >>>>>>>>>>>  create mode 100644 Documentation/devicetree/bindings/spi/spi-renesas-rpc.txt
> >>>>>>>>>>>
> >>>>>>>>>>> diff --git a/Documentation/devicetree/bindings/spi/spi-renesas-rpc.txt b/Documentation/devicetree/bindings/spi/spi-renesas-rpc.txt
> >>>>>>>>>>> new file mode 100644
> >>>>>>>>>>> index 0000000..8286cc8
> >>>>>>>>>>> --- /dev/null
> >>>>>>>>>>> +++ b/Documentation/devicetree/bindings/spi/spi-renesas-rpc.txt
> >>>>>>>>>>> @@ -0,0 +1,33 @@
> >>>>>>>>>>> +Renesas R-Car D3 RPC controller Device Tree Bindings
> >>>>>>>>>>> +----------------------------------------------------
> >>>>>>>>>>> +
> >>>>>>>>>>> +Required properties:
> >>>>>>>>>>> +- compatible: should be "renesas,rpc-r8a77995"
> >>>>>>>>>>> +- #address-cells: should be 1
> >>>>>>>>>>> +- #size-cells: should be 0
> >>>>>>>>>>> +- reg: should contain 2 entries, one for the registers and one for the direct
> >>>>>>>>>>> +       mapping area
> >>>>>>>>>>> +- reg-names: should contain "rpc_regs" and "dirmap"
> >>>>>>>>>>> +- interrupts: interrupt line connected to the RPC SPI controller            
> >>>>>>>>>>
> >>>>>>>>>> Do you also plan to support the RPC HF mode ? And if so, how would that
> >>>>>>>>>> look in the bindings ?          
> >>>>>>>>>
> >>>>>>>>> Not sure this approach is still accepted, but that's how we solved the
> >>>>>>>>> problem for the flexcom block [1].
> >>>>>>>>>
> >>>>>>>>> [1]https://elixir.bootlin.com/linux/v4.20-rc3/source/Documentation/devicetree/bindings/mfd/atmel-flexcom.txt          
> >>>>>>>>
> >>>>>>>> That looks pretty horrible.
> >>>>>>>>
> >>>>>>>> In U-Boot we check whether the device hanging under the controller node
> >>>>>>>> is JEDEC SPI flash or CFI flash and based on that decide what the config
> >>>>>>>> of the controller should be (SPI or HF). Not sure that's much better,but
> >>>>>>>> at least it doesn't need extra nodes which do not really represent any
> >>>>>>>> kind of real hardware.
> >>>>>>>>        
> >>>>>>>
> >>>>>>> The subnodes are not needed, you can just have a property that tells in
> >>>>>>> which mode the controller is supposed to operate, and the MFD would
> >>>>>>> create a sub-device that points to the same device_node.        
> >>>>>>
> >>>>>> Do you even need a dedicated property ? I think you can decide purely on
> >>>>>> what node is hanging under the controller (jedec spi nor or cfi nor).      
> >>>>>
> >>>>> Yes, that could work if they have well-known compatibles. As soon as
> >>>>> people start using flash-specific compats (like some people do for
> >>>>> their SPI NORs) it becomes a maintenance burden.      
> >>>>
> >>>> Which, on this controller, is very likely never gonna happen. Once it
> >>>> does , we can add a custom property.
> >>>>    
> >>>>>>> Or we can have
> >>>>>>> a single driver that decides what to declare (a spi_controller or flash
> >>>>>>> controller), but you'd still have to decide where to place this
> >>>>>>> driver...        
> >>>>>>
> >>>>>> I'd definitely prefer a single driver.
> >>>>>>      
> >>>>>
> >>>>> Where would you put this driver? I really don't like the idea of having
> >>>>> MTD drivers spread over the tree. Don't know what's Mark's opinion on
> >>>>> this matter.      
> >>>>
> >>>> Well, it's both CFI (hyperflash) and SF (well, SPI flash) controller, so
> >>>> where would this go ?
> >>>>    
> >>>
> >>> The spi-mem layer is in drivers/spi/ so it could go in drivers/spi/
> >>> (spi-mem controller) or drivers/mtd/ (CFI controller).    
> >>
> >> drivers/mtd is probably a better option, since it's not a generic SPI
> >> controller.
> >>  
> > 
> > No, spi-mem controller drivers should go in drivers/spi/ even if they
> > don't implement the generic SPI interface (it's allowed to only
> > implement the spi_mem interface).  
> 
> Except this is not only SPI MEM controller, this is also hyperflash
> (that is, CFI) controller. It can drive both types of chips. Thus , I
> think it fits better in drivers/mtd/ .
> 

Okay, then I guess we need an ack from Mark on that.