[RFC] arm64: dts: ti: k3-j721e: Use nvmem to model the efuses

  The nvmem subsystem is used to model efuse areas. Currently our efuse
areas are described using raw syscon nodes and offsets into that. The
use case for this efuse area is to store the SoC Ethernet MAC address.
This is already such a commonly supported use case that both the nvmem
and networking subsystem have built-in support for it. Use that here
to get the following:

1. More standard, complete, and accurate description of this hardware
   efuse area.

2. Remove the need for this custom 'ti,syscon-efuse' property (which
   I'm surprised was accepted into the binding in the first place..)

3. Allow the parent node to not have to be a syscon which fixes a DT
   check warning.

RFC for now as we don't have a simple generic compatible for memory
mapped nvmem areas. For now we use the 'uniphier-efuse' compatible
as it is essentially just a generic mmio nvmem area, which means any
generic area is technically "compatible".

If this all is acceptable I'll split this patch into three and do
the same for all the other K3 devices still using this odd
'ti,syscon-efuse' pattern for v1.

Signed-off-by: Andrew Davis <afd@ti.com>
---
 .../boot/dts/ti/k3-j721e-mcu-wakeup.dtsi      | 24 ++++++++++++++++---
 1 file changed, 21 insertions(+), 3 deletions(-)
  

+ Siddharth for CSPW driver, + Vibhore of CPUFreq

On 25/10/23 21:13, Andrew Davis wrote:
> The nvmem subsystem is used to model efuse areas. Currently our efuse
> areas are described using raw syscon nodes and offsets into that. The
> use case for this efuse area is to store the SoC Ethernet MAC address.
> This is already such a commonly supported use case that both the nvmem
> and networking subsystem have built-in support for it. Use that here
> to get the following:
> 
> 1. More standard, complete, and accurate description of this hardware
>    efuse area.
> 
> 2. Remove the need for this custom 'ti,syscon-efuse' property (which
>    I'm surprised was accepted into the binding in the first place..)
> 
> 3. Allow the parent node to not have to be a syscon which fixes a DT
>    check warning.
> 

Sounds like a good plan to me.

> RFC for now as we don't have a simple generic compatible for memory
> mapped nvmem areas. For now we use the 'uniphier-efuse' compatible
> as it is essentially just a generic mmio nvmem area, which means any
> generic area is technically "compatible".
> 

Looking at drivers/nvmem/ its seems like each SoC arch seems to have
their own individual efuse driver?

For K3 SoCs (at least AM62), there are other properties in addition to
MAC address in the eFUSE which are useful for different drivers such as
Speed grades. Can this be extended for the same?

> If this all is acceptable I'll split this patch into three and do
> the same for all the other K3 devices still using this odd
> 'ti,syscon-efuse' pattern for v1.
> 
> Signed-off-by: Andrew Davis <afd@ti.com>
> ---
>  .../boot/dts/ti/k3-j721e-mcu-wakeup.dtsi      | 24 ++++++++++++++++---
>  1 file changed, 21 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/arm64/boot/dts/ti/k3-j721e-mcu-wakeup.dtsi b/arch/arm64/boot/dts/ti/k3-j721e-mcu-wakeup.dtsi
> index 05d6ef127ba78..8f488d43d49df 100644
> --- a/arch/arm64/boot/dts/ti/k3-j721e-mcu-wakeup.dtsi
> +++ b/arch/arm64/boot/dts/ti/k3-j721e-mcu-wakeup.dtsi
> @@ -34,13 +34,30 @@ k3_reset: reset-controller {
>  		};
>  	};
>  
> -	mcu_conf: syscon@40f00000 {
> -		compatible = "syscon", "simple-mfd";
> +	mcu_conf: bus@40f00000 {
> +		compatible = "simple-bus";
>  		reg = <0x0 0x40f00000 0x0 0x20000>;
>  		#address-cells = <1>;
>  		#size-cells = <1>;
>  		ranges = <0x0 0x0 0x40f00000 0x20000>;
>  
> +		efuse@200 {
> +			compatible = "socionext,uniphier-efuse";
> +			#address-cells = <1>;
> +			#size-cells = <1>;
> +			reg = <0x200 0x40>;
> +
> +			nvmem-layout {
> +				compatible = "fixed-layout";
> +				#address-cells = <1>;
> +				#size-cells = <1>;
> +
> +				mac_address: mac@0 {
> +					reg = <0x0 0x6>;
> +				};
> +			};
> +		};
> +
>  		phy_gmii_sel: phy@4040 {
>  			compatible = "ti,am654-phy-gmii-sel";
>  			reg = <0x4040 0x4>;
> @@ -533,7 +550,8 @@ cpsw_port1: port@1 {
>  				reg = <1>;
>  				ti,mac-only;
>  				label = "port1";
> -				ti,syscon-efuse = <&mcu_conf 0x200>;
> +				nvmem-cells = <&mac_address>;
> +				nvmem-cell-names = "mac-address";
>  				phys = <&phy_gmii_sel 1>;
>  			};
>  		};
  

On 10/29/23 11:25 PM, Vignesh Raghavendra wrote:
> + Siddharth for CSPW driver, + Vibhore of CPUFreq
> 
> On 25/10/23 21:13, Andrew Davis wrote:
>> The nvmem subsystem is used to model efuse areas. Currently our efuse
>> areas are described using raw syscon nodes and offsets into that. The
>> use case for this efuse area is to store the SoC Ethernet MAC address.
>> This is already such a commonly supported use case that both the nvmem
>> and networking subsystem have built-in support for it. Use that here
>> to get the following:
>>
>> 1. More standard, complete, and accurate description of this hardware
>>     efuse area.
>>
>> 2. Remove the need for this custom 'ti,syscon-efuse' property (which
>>     I'm surprised was accepted into the binding in the first place..)
>>
>> 3. Allow the parent node to not have to be a syscon which fixes a DT
>>     check warning.
>>
> 
> Sounds like a good plan to me.
> 
>> RFC for now as we don't have a simple generic compatible for memory
>> mapped nvmem areas. For now we use the 'uniphier-efuse' compatible
>> as it is essentially just a generic mmio nvmem area, which means any
>> generic area is technically "compatible".
>>
> 
> Looking at drivers/nvmem/ its seems like each SoC arch seems to have
> their own individual efuse driver?
> 

Yes, but I'm not sure I see the need for a whole new driver/compatible
for every simple MMIO-based efuse ranges. Would be nice to have a
generic one.

> For K3 SoCs (at least AM62), there are other properties in addition to
> MAC address in the eFUSE which are useful for different drivers such as
> Speed grades. Can this be extended for the same?
> 

Yes, with the "nvmem-layout" being worked on right now by the nvmem
folks I see a clear path to extend this over the rest of our efuse
area and replace the remaining "syscon" hacks we have been using.

Andrew

>> If this all is acceptable I'll split this patch into three and do
>> the same for all the other K3 devices still using this odd
>> 'ti,syscon-efuse' pattern for v1.
>>
>> Signed-off-by: Andrew Davis <afd@ti.com>
>> ---
>>   .../boot/dts/ti/k3-j721e-mcu-wakeup.dtsi      | 24 ++++++++++++++++---
>>   1 file changed, 21 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/arm64/boot/dts/ti/k3-j721e-mcu-wakeup.dtsi b/arch/arm64/boot/dts/ti/k3-j721e-mcu-wakeup.dtsi
>> index 05d6ef127ba78..8f488d43d49df 100644
>> --- a/arch/arm64/boot/dts/ti/k3-j721e-mcu-wakeup.dtsi
>> +++ b/arch/arm64/boot/dts/ti/k3-j721e-mcu-wakeup.dtsi
>> @@ -34,13 +34,30 @@ k3_reset: reset-controller {
>>   		};
>>   	};
>>   
>> -	mcu_conf: syscon@40f00000 {
>> -		compatible = "syscon", "simple-mfd";
>> +	mcu_conf: bus@40f00000 {
>> +		compatible = "simple-bus";
>>   		reg = <0x0 0x40f00000 0x0 0x20000>;
>>   		#address-cells = <1>;
>>   		#size-cells = <1>;
>>   		ranges = <0x0 0x0 0x40f00000 0x20000>;
>>   
>> +		efuse@200 {
>> +			compatible = "socionext,uniphier-efuse";
>> +			#address-cells = <1>;
>> +			#size-cells = <1>;
>> +			reg = <0x200 0x40>;
>> +
>> +			nvmem-layout {
>> +				compatible = "fixed-layout";
>> +				#address-cells = <1>;
>> +				#size-cells = <1>;
>> +
>> +				mac_address: mac@0 {
>> +					reg = <0x0 0x6>;
>> +				};
>> +			};
>> +		};
>> +
>>   		phy_gmii_sel: phy@4040 {
>>   			compatible = "ti,am654-phy-gmii-sel";
>>   			reg = <0x4040 0x4>;
>> @@ -533,7 +550,8 @@ cpsw_port1: port@1 {
>>   				reg = <1>;
>>   				ti,mac-only;
>>   				label = "port1";
>> -				ti,syscon-efuse = <&mcu_conf 0x200>;
>> +				nvmem-cells = <&mac_address>;
>> +				nvmem-cell-names = "mac-address";
>>   				phys = <&phy_gmii_sel 1>;
>>   			};
>>   		};
>