Re: [PATCH] net: dsa: read mac address from DT for slave device

[Florian Fainelli <f.fainelli@gmail.com>]

On 2/26/2019 6:04 PM, xiaofeis@codeaurora.org wrote:
> On 2019-02-26 15:45, xiaofeis@codeaurora.org wrote:
>> On 2019-02-26 01:27, Florian Fainelli wrote:
>>> On 2/25/19 5:28 AM, xiaofeis@codeaurora.org wrote:
>>>> Hi Florian
>>>>
>>>> We have two slave DSA interfaces, wan0 and lan0, one is for wan port,
>>>> and the other is for lan port. Customer has it's mac address pool, they
>>>> want
>>>> to assign the mac address from the pool on wan0, lan0, and other
>>>> interfaces like
>>>> wifi, bt. Coreboot/uboot will populate it to the DTS node, so the
>>>> driver
>>>> can
>>>> get it from it's node. For DSA slave interface, it already has it's own
>>>> DTS node, it's
>>>> easy to just add one porperty "local-mac-address" there for the
>>>> usage in
>>>> DSA driver.
>>>>
>>>> If not use DSA framework, normally we will use eth0.x and eth0.y for
>>>> wan
>>>> and lan.
>>>> On this case, customer usually also assign the MAC address on these
>>>> logical interface
>>>> from it's pool.
>>>
>>> OK, but this is not necessary per my previous explanation: the CPU <=>
>>> WAN pipe is a separate broadcast domain (otherwise it is a security hole
>>> since you exposing LAN machines to the outside world), and so there is
>>> no need for a separate MAC address. It might be convenient to have one,
>>> especially for the provider, if they run a management software (e.g.:
>>> TR69), but it is not required per-se.
>>>
>>> Let me ask a secondary question here, how many Ethernet MACs connect to
>>> the switch in this configuration? Is there one that is supposed to be
>>> assigned all LAN traffic and one that is supposed to be assigned all WAN
>>> traffic? If so, then what you are doing makes even less
>>>
>>
>> Only one MAC connected to switch cpu port, both lan0 and wan0 are on
>> the top of
>> same interface(eth0).
>>
> Customer doesn't care about the MAC controller's MAC address, just leave
> it as the driver
> randomly generated. They just want to assign the MAC address on wan and
> lan DSA logical
> interface.
> 
> Many customer doesn't use DSA, for example, they use eth0.1/eth0.2 for
> lan/wan with one MAC controller.
> They configure switch wan port in vlan2 group, and lan port in vlan1
> group, they usually assign mac address
> on the logical interface(eth0.1&eth0.2), i think this is similar with
> DSA slave interfaces.

Yes it is a similar use case, and in both cases there is no really a
functional need for a separate MAC address for lan/eth0.1 or wan/eth0.2
since the switch should be configured to perform IVL (Individual VLAN
Learning) and would determine the egress port just fine based on the MAC
DA. Because it is an established practice does not mean we should not
challenge it :).

My issue with your change is that because DSA is meant to be a flexible
framework we do not know the type/nature of DSA master network device
that is going to be used. That DSA master network device may or may not
have it own MAC DA filtering capability. Having to filter its own MAC
address is fine and a well established behavior, having to filter for
more than one unicast address starts to be questionable and eats up
filter space that could be better used for filtering MC addresses
instead. Another possible concern is a station trying to spoof the MAC
address, some switches may support protecting only one UC/management MAC
address, so having more than one could create security attack surfaces.

To give you an example, I work with 3 generations of DSA master network
controllers (bcmgenet and bcmsysport drivers).

- GENET supports 17 perfect filters, but we must include its own MAC
address, the broadcast address and that leaves only 15 filters for MC

- SYSTEMPORT is always attached to a switch but supports filtering the
MAC DA based on its own MAC and then it is in promiscuous mode

So with your scheme, we would leave only 13 filters for MC on GENET and
we would putting the interface in promiscuous mode for SYSTEMPORT.

Until we have a better switch-side filtering framework (and this is
being worked on right now), I would prefer that we defer accepting those
type of features. Andrew and Vivien might feel differently about that
though.
-- 
Florian