[PATCH] net: uclass: Save ethernet MAC address when generated

[Michal Simek]

On 11/4/21 12:37, Tom Rini wrote:
> On Thu, Nov 04, 2021 at 12:18:46PM +0100, Michal Simek wrote:
>>
>>
>> On 11/3/21 17:57, Tom Rini wrote:
>>> On Tue, Nov 02, 2021 at 11:27:22AM +0100, Michal Simek wrote:
>>>>
>>>>
>>>> On 11/2/21 10:00, Michael Walle wrote:
>>>>>> On Fri, Oct 29, 2021 at 2:14 PM Michal Simek <michal.simek at xilinx.com> wrote:
>>>>>>>
>>>>>>> When MAC address is randomly generated it should be also saved to
>>>>>>> variables. This step is there when MAC address is passed via pdata but not
>>>>>>> when it is randomly generated.
>>>>>>>
>>>>>>> Signed-off-by: Michal Simek <michal.simek at xilinx.com>
>>>>>>> ---
>>>>>>>
>>>>>>>     net/eth-uclass.c | 2 ++
>>>>>>>     1 file changed, 2 insertions(+)
>>>>>>>
>>>>>>> diff --git a/net/eth-uclass.c b/net/eth-uclass.c
>>>>>>> index 0da0e85be031..58c308f33276 100644
>>>>>>> --- a/net/eth-uclass.c
>>>>>>> +++ b/net/eth-uclass.c
>>>>>>> @@ -583,6 +583,8 @@ static int eth_post_probe(struct udevice *dev)
>>>>>>>                    net_random_ethaddr(pdata->enetaddr);
>>>>>>>                    printf("\nWarning: %s (eth%d) using random MAC address - %pM\n",
>>>>>>>                           dev->name, dev_seq(dev), pdata->enetaddr);
>>>>>>> +               eth_env_set_enetaddr_by_index("eth", dev_seq(dev),
>>>>>>> +                                             pdata->enetaddr);
>>>>>>>     #else
>>>>>>>                    printf("\nError: %s address not set.\n",
>>>>>>>                           dev->name);
>>>>>>> --
>>>>>>> 2.33.1
>>>>>>>
>>>>>> Reviewed-by: Ramon Fried <rfried.dev at gmail.com>
>>>>>
>>>>> Please note, that this will change behavior. Before this commit, the
>>>>> random mac address was local to u-boot (at least for most network drivers).
>>>>> After this commit, it will also be communicated to linux.
>>>>>
>>>>> I'm not sure what to think of this. At the very least, this should be
>>>>> documented in the commit message and in the Kconfig help text.
>>>>
>>>> Thanks for bringing this up. I have no issue that this address is being
>>>> propagated to Linux but others can feel this as an issue.
>>>> I can definitely extend commit message to say it.
>>>>
>>>> I found this via net list command where you can see controllers but you
>>>> can't see their mac addresses which is IMHO wrong.
>>>
>>> So, this causes dhcp to simply timeout for me on a pine64_plus board
>>> (where yes, I believe we do end up using a random MAC) when running the
>>> pytest suite.  I wonder if this particular change messes up the state
>>> machine?  Checking the dnsmasq logs it looks like U-Boot keeps asking
>>> and getting a reply, over and over.
>>
>> It is just saving that values to variables. If you want to propagete that
>> values in the stack that's up to your system configuration.
>>
>> What state machine are you talking about?
> 
> I was just making a guess about state machine, given what I had recalled
> of the network stack before.  To be clearer, I bisected down my failure
> of 3 tests that run "setenv autoload no ; dhcp" and then dhcp timing
> out, to this commit (in the network PR).

First of all I didn't know that it was the part of networking pull 
request because didn't see apply message only reviewed from Ramon.

I would be very surprised this patch to caused your issue. It should be 
covering something else but that should be debugged.

Thanks,
Michal