Re: [PATCH] natsemi update 1/4 Use assigned MAC address

Hi Manfred, Gary, Stephen,

"But: I'm not sure that the change is required. What about just
setting the mac to 0, and the actual mac address is set from user
space? It's possible to set the mac address with"

Could I suggest if this is the solution implemented, setting the
first octet of the MAC address to 0x02, as in a Locally Assigned
MAC address ?

If an interface with an all zero's MAC address is added to a
bridge, running the Spanning Tree Protocol (STP), it will always
attempt to take over as the root bridge, unless STP root bridge
priorities are being used. This would disrupt traffic on the
attached LAN.

Ideally, assigning "zero" a MAC address which has almost no
chance of disrupting STP or any other protocol could be useful.
Maybe fe:ff:ff:ff:ff:ff ?

I think this would fit in with the "be conservative in what you
send, liberal in what you receive" philosophy.

Regards,
Mark.

ps, please CC on any replies, I'm not subscribed to the list yet.

On Tue, 8 Jun 2004 07:12:52 +0930
Mark Smith [off-list ref] wrote:

Hi Manfred, Gary, Stephen,

"But: I'm not sure that the change is required. What about just
setting the mac to 0, and the actual mac address is set from user
space? It's possible to set the mac address with"

Could I suggest if this is the solution implemented, setting the
first octet of the MAC address to 0x02, as in a Locally Assigned
MAC address ?

If an interface with an all zero's MAC address is added to a
bridge, running the Spanning Tree Protocol (STP), it will always
attempt to take over as the root bridge, unless STP root bridge
priorities are being used. This would disrupt traffic on the
attached LAN.

Actually, it won't let you add any interface to a bridge without
a valid ether address now (ie non-zero and not multicast).

Ideally, assigning "zero" a MAC address which has almost no
chance of disrupting STP or any other protocol could be useful.
Maybe fe:ff:ff:ff:ff:ff ?

I think this would fit in with the "be conservative in what you
send, liberal in what you receive" philosophy.

Regards,
Mark.

ps, please CC on any replies, I'm not subscribed to the list yet.

On Tue, 8 Jun 2004 07:12:52 +0930
Mark Smith [off-list ref] wrote:

Hi Manfred, Gary, Stephen,

"But: I'm not sure that the change is required. What about just
setting the mac to 0, and the actual mac address is set from user
space? It's possible to set the mac address with"

How about 
	random_ether_addr(dev->dev_addr);

it does the right thing.

On Mon, 7 Jun 2004 14:46:54 -0700
Stephen Hemminger [off-list ref] wrote:

On Tue, 8 Jun 2004 07:12:52 +0930
Mark Smith
[off-list ref]
wrote:

quoted

Hi Manfred, Gary, Stephen,

<snip>

quoted

If an interface with an all zero's MAC address is added to a
bridge, running the Spanning Tree Protocol (STP), it will
always attempt to take over as the root bridge, unless STP
root bridge priorities are being used. This would disrupt
traffic on the attached LAN.

Actually, it won't let you add any interface to a bridge
without a valid ether address now (ie non-zero and not
multicast).

That's good, I first noticed this issue when I happened to add a
dummy interface into a bridge configuration, and found that its
zero MAC address ended up disrupting the spanning tree. 

I also have just noticed that the dummy interfaces now have
random MAC addresses as per the function you mentioned earlier.
Also good.

Regards,
Mark.

quoted

ps, please CC on any replies, I'm not subscribed to the list
yet.