[U-Boot] [PATCH 12/14] fdt: eth_fixup: Add hook for board to override MAC

[Joe Hershberger]

On Wed, Nov 30, 2016 at 4:54 AM, Hans de Goede <hdegoede at redhat.com> wrote:
> Hi,
>
>
> On 30-11-16 11:50, Olliver Schinagl wrote:
>>
>> Hey maime,
>>
>> Sorry for constantly getting your e-mail address wrong! Sorry!
>>
>> On 30-11-16 10:12, maxime.ripard at free-electrons.com wrote:
>>>
>>> On Wed, Nov 30, 2016 at 09:00:51AM +0000, Marcel Ziswiler wrote:
>>>>
>>>> Hi Olliver
>>>>
>>>> On Fri, 2016-11-25 at 16:30 +0100, Olliver Schinagl wrote:
>>>>>
>>>>> This patch adds a method for the board to set the MAC address if the
>>>>> environment is not yet set. The environment based MAC addresses are
>>>>> not
>>>>> touched, but if the fdt has an alias set, it is parsed and put into
>>>>> the
>>>>> environment.
>>>>>
>>>>> E.g. The environment contains ethaddr and eth1addr, and the fdt
>>>>> contains
>>>>> an ethernet1 nothing happens. If the fdt contains ethernet2 however,
>>>>> it
>>>>> is parsed and inserted into the environment as eth2addr.
>>>>
>>>> My humble understanding of device tree fixup is that it works the other
>>>> way around (e.g. it is the device tree that usually gets fixed up). So
>>>> the least I would advice for this patch is to change its naming but
>>>> most possibly such code also does not belong into the common
>>>> fdt_support implementation.
>>
>> First, yes I noticed this as well, the nameing is the wrong way around. It
>> took me a few reading times as well. I guess I did not full understand
>> Hans's suggestion comment. So there's some work needed here.
>>>
>>> I don't really have the context of this patch, but in the DT at least,
>>> you can specify the mac address using the local-mac-address
>>> property. I guess we should honor that too. But I don't really know
>>> how that's related to an alias. If the device is probed and the
>>> property is there, use it, otherwise don't.
>>
>> This came from the sdio_realtek module on some sunxi boards, where the
>> device tree has configured it obviously for linux, but u-boot has no notion
>> of it. But u-boot IS responsible to generate (the same) MAC address for the
>> device. So yes, u-boot inserts a mac address into the device-tree for a
>> found alias.
>>
>> E.g. ethernet2 is an alias without a MAC address configured for it. Thus
>> u-boot is used to generate the MAC address for this node and inserts it into
>> the dtb. What I haven't double checked (just blindly assumed, which is the
>> mother) if this code also inserts the mac into the environment, but the
>> kernel only gets this information via the dtb anyway, right? Either via the
>> dtb, or via the MAC register bytes where it is stored in some drivers.
>
>
> About the setting of the MAC in the environment for devices u-boot
> knows nothing about, but which have an alias in the dt, this is
> not necessary, it was done in the old code as a way to make the
> u-boot fdt code pick up the MAC and inject it into the dtb
> passed to the kernel. If this can now happen without setting
> it in the env that step can be skipped.

I agree that the env should not be updated.

-Joe