[PATCH 1/2] net: phy: marvell: Support reg config via "marvell, reg-init" DT property

[Stefan Roese]

On 3/30/22 14:16, Marek Behún wrote:
> On Wed, 30 Mar 2022 10:38:25 +0200
> Stefan Roese <sr at denx.de> wrote:
> 
>> This patch adds support for the "marvell,reg-init" DT property, which
>> is used to describe board specific Marvell PHY register configurations
>> in the board dts file. This DT property is supported in the Linux Kernel
>> since a longer time. Adding it to U-Boot now, enables the boards which
>> describe the register settings in their DT files here as well.
>>
>> I've included calling this marvell_of_reg_init() to all foo_config()
>> functions in this patch as well. If CONFIG_DM_ETH is not set, there is
>> no ofnode, or no "marvell,reg-init" property, the PHY initialization is
>> unchanged.
>>
>> The function marvell_of_reg_init() is a port of the Linux version.
>> Please note that I explicitly did not add error checking and handling
>> to the U-Boot version, as this is basically not done for phy_read/write
>> in this Marvell PHY code.
>>
>> This will be used by the upcoming ethernet support on the MIPS
>> Octeon EBB 7304 board.
>>
>> Signed-off-by: Stefan Roese <sr at denx.de>
>> Cc: Ramon Fried <rfried.dev at gmail.com>
>> Cc: Joe Hershberger <joe.hershberger at ni.com>
>> Cc: Aaron Williams <awilliams at marvell.com>
>> Cc: Chandrakala Chavva <cchavva at marvell.com>
> 
> Hi Stefan, I know this property is used also in kernel, but what do you
> want to use it for?

Frankly, I do not know exactly. It's part of the MIPS Octeon ethernet
support in mainline U-Boot, which also uses this DT property on the
EBB7304 board. I've not analyzed the register settings in detail to
check which register is being configured to which value to see, why this
might be the case. Dropping these register settings might have resulted
in problems not directly spotted on the board. So my plan was to
integrate these as well.

For reference, I've added the current DT reg-init setting below [1].
This is for the 88e1240 PHY, and taking a quick look in the manual,
this seems to be LED related (see below).


> Because in kernel the idea is to deprecate it.

This is new to me.

> It is used for example to confiure INT pin, but that should have it's
> own property once someone implements it. It is also used to configure
> LEDs, and that is hopefully gonna be obsoleted by supporting the LEDs
> via the LED subsystem, so afterwards we will write
>     ethernet-phy at 1 {
>        leds {
>            led at 0 {
>                color = ...;
>            };
>        };
>     };

That would be very cool indeed. Even though it's not possible "right
now".

So how to continue? Add this "reg-init" "feature" for now? And perhaps
drop it once this LED configuration is possible via the way you describe
above? Or drop it once it's also dropped in the Kernel? My personal
feeling is, that supporting "reg-init" does not "hurt" and might even be
handy for some testing / debugging of network PHY setups.

Perhaps Ramon and/or Joe also have some comments on this.

Thanks,
Stefan

[1]		marvell,reg-init = <3 0x10 0 0x8665>,
				   <3 0x11 0 0x00aa>,
				   <3 0x12 0 0x4105>,
				   <3 0x13 0 0x8a08>;