linux-kernel - Re: [PATCH] ethernet: aquantia: Try MAC address from device tree

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <80083aba-c906-f076-c887-910d3fe186f6@marcan.st>
Date:   Thu, 2 Dec 2021 14:10:49 +0900
From:   Hector Martin <marcan@...can.st>
To:     Andrew Lunn <andrew@...n.ch>
Cc:     Tianhao Chai <cth451@...il.com>,
        Igor Russkikh <irusskikh@...vell.com>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>, netdev@...r.kernel.org,
        linux-kernel@...r.kernel.org, Sven Peter <sven@...npeter.dev>,
        Alyssa Rosenzweig <alyssa@...enzweig.io>
Subject: Re: [PATCH] ethernet: aquantia: Try MAC address from device tree

On 30/11/2021 11.32, Andrew Lunn wrote:
> On Mon, Nov 29, 2021 at 02:08:28AM +0900, Hector Martin wrote:
>> On 29/11/2021 01.33, Andrew Lunn wrote:
>>> On Sat, Nov 27, 2021 at 08:37:33PM -0600, Tianhao Chai wrote:
>>>> Apple M1 Mac minis (2020) with 10GE NICs do not have MAC address in the
>>>> card, but instead need to obtain MAC addresses from the device tree. In
>>>> this case the hardware will report an invalid MAC.
>>>>
>>>> Currently atlantic driver does not query the DT for MAC address and will
>>>> randomly assign a MAC if the NIC doesn't have a permanent MAC burnt in.
>>>> This patch causes the driver to perfer a valid MAC address from OF (if
>>>> present) over HW self-reported MAC and only fall back to a random MAC
>>>> address when neither of them is valid.
>>>
>>> This is a change in behaviour, and could cause regressions. It would
>>> be better to keep with the current flow. Call
>>> aq_fw_ops->get_mac_permanent() first. If that does not give a valid
>>> MAC address, then try DT, and lastly use a random MAC address.
>>
>> On DT platforms, it is expected that the device tree MAC will override
>> whatever the device thinks is its MAC address.
> 
> Can you point to any documentation of that expectation?

I don't think this is explicitly clarified anywhere, but the DT binding 
says:

 > Specifies the MAC address that was assigned to the network device.

It certainly doesn't say this should be a fallback only to be used if 
the device doesn't have some idea of its MAC. Usually you'd expect 
firmware information to override whatever the device's built-in defaults 
are.

>> I would not expect any other existing platform to have a MAC assigned to the
>> device in this way using these cards; if any platforms do, chances are they
>> intended it for it to be used and this patch will fix a current bug. If some
>> platforms out there really have bogus MACs assigned in this way, that's a
>> firmware bug, and we'd have to find out and add explicit, targeted
>> workaround code. Are you aware of any such platforms? :)
> 
> I'm not aware of any, because i try to avoid making behaviour changes.
> 
> Anyway, lets go with this, and if stuff breaks we can always change
> the order to what i suggested in order to unbreak stuff. I'm assuming
> for Apple M1 Mac minis the order does not actually matter?

Correct, on these machines the burned-in MAC is invalid so it doesn't 
matter.


-- 
Hector Martin (marcan@...can.st)
Public Key: https://mrcn.st/pub