[U-Boot] [PATCH 1/6] net: gem: Add support for reading MAC from I2C EEPROM

[Joe Hershberger]

Hi Bin,

On Sun, Feb 14, 2016 at 6:00 AM, Bin Meng <bmeng.cn at gmail.com> wrote:
> Hi Michal,
>
> On Sun, Feb 14, 2016 at 6:03 PM, Michal Simek <monstr at monstr.eu> wrote:
>> Hi Bin,
>>
>> 2016-02-14 3:25 GMT+01:00 Bin Meng <bmeng.cn at gmail.com>:
>>>
>>> Hi Michal,
>>>
>>> On Sat, Feb 13, 2016 at 6:39 PM, Michal Simek <monstr at monstr.eu> wrote:
>>> > Add support for reading MAC address from I2C EEPROM.
>>> >
>>>
>>> Is this a feature provided by the GEM MAC IP?
>>>
>>> > Signed-off-by: Michal Simek <monstr at monstr.eu>
>>> > ---
>>> >
>>> >  drivers/net/zynq_gem.c | 16 ++++++++++++++++
>>> >  1 file changed, 16 insertions(+)
>>> >
>>> > diff --git a/drivers/net/zynq_gem.c b/drivers/net/zynq_gem.c
>>> > index b3821c31a91d..ace60c901cb5 100644
>>> > --- a/drivers/net/zynq_gem.c
>>> > +++ b/drivers/net/zynq_gem.c
>>> > @@ -627,6 +627,21 @@ static int zynq_gem_remove(struct udevice *dev)
>>> >         return 0;
>>> >  }
>>> >
>>> > +static int zynq_gem_read_rom_hwaddr(struct udevice *dev)
>>> > +{
>>> > +#if defined(CONFIG_ZYNQ_GEM_EEPROM_ADDR) && \
>>> > +    defined(CONFIG_ZYNQ_GEM_I2C_MAC_OFFSET)
>>> > +       struct eth_pdata *pdata = dev_get_platdata(dev);
>>> > +
>>> > +       if (eeprom_read(CONFIG_ZYNQ_GEM_EEPROM_ADDR,
>>> > +                       CONFIG_ZYNQ_GEM_I2C_MAC_OFFSET,
>>> > +                       pdata->enetaddr, ARRAY_SIZE(pdata->enetaddr)))
>>>
>>> This call to eeprom_read() looks to me a board-specific feature, that
>>> an on-board eeprom is used to store the MAC address for the GEM?
>>>
>>
>> Right. it is board specific feature I can
>> The question is if this should really go to board specific file
>> or to be the part of DT binding. I didn't look at the kernel if someone has
>> some sort of eeprom binding for this case
>> but I expect local mac addresses via DT are used. Or passing via command
>> line does it.
>>
>> Anyway there is mac_read_from_eeprom() but it is ancient one.
>> Do you any preference for the name of function?
>
> I think the intention of the read_rom_hwaddr op is to read the MAC
> address via the ethernet controller defined mechanism (eg: e1000 can
> read its MAC address from either an EEPROM or an SPI flash), or via
> SoC defined mechanism (eg: the ethernet IP is only seen on a specific
> SoC, and reading the MAC address only makes sense on that specific
> SoC). If this is a board-specific mechanism, I don't think we should
> put the codes into driver's read_rom_hwaddr(). Again, if it is
> board-specific, we may not come up with a standard DT binding for such
> stuff, unless we can enumerate all possible ways for storing MAC
> address on a board (EEPROM, SPI flash, eMMC, SD)?

Did you see this: https://patchwork.ozlabs.org/patch/573373/

This is the concept I had in mind for handling this sort of thing.

Thanks,
-Joe