Re: [RFC PATCH net-next v2 2/3] sh_eth: add support RMIIMODE register

[Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>]

Hello.

On 28-06-2013 5:41, Simon Horman wrote:

>>>>> This change is motivated by the lager board which uses the r8a7790 SoC,
>>>>> an R-Car SoC. For this board setting the RMIIMODE register is necessary.
>>>>> This patch assumes this is valid for all R-Car SoCs.
>>>>     No, RMIIMODE register is not documented on R8A7778/9 SoCs and
>>>> this location is described as reserved. How's it called in R8A7790
>>>> manual, CXR15?

>>> It appears to be undocumented in the R8A7790 I have.

>>      Of the manuals I have now this register is only documented in
>> R-Car H2 user's manual 0.50,

>>> Empirically it seems necessary for the lager/R8A7790.

>>     The manual confirms that.

>>> To be honest I am entirely unsure of its relevance to other SoCs.

>>> I wonder if a better approach would be to add a flag to platform
>>> data rather than struct sh_eth_cpu_data. This would allow the register
>>> to only be accessed on boards where it is known to be useful.

>>     No, this doesn't depend on the board, judging on the H2 manual.
>> You just must write it to 1 on R-Car H2 in order for Ether to operate.

> What is your preferred way to add support for this to the driver?

    Like you did, add a field to 'struct sh_eth_cpu_data' that's 1 if 
RMIIMODE register is present but don't set it in 'r8a777x_data' and 
create separate 'r8a7790_data' instead (where to set this field).

WBR, Sergei