IP address conflict (transparent bridge)

[Val P]

I have a bridge configured as follows:

Bge0	- management interface, ip a.b.c.d
Bge1	- down (not used)
Bge2 	- "inside" bridge leg, no ip, layer-2 only
Bge3	- "outside" bridge leg, no ip, layer-2 only

Now, I am seeing some weird behavior. If I try to do any type of  net access
over bge0, I get console messages such as (paraphrasing, since I don't have
it in front of me):

/kernel: arp: 1.2.3.4.5.6 is using my ip address a.b.c.d!
/kernel: arp: 7.8.9.10.11.12 is using my ip address a.b.c.d!

1.2.3.4.5.6 is the mac of the bge0 interface
7.8.9.10.11.12 is the mac of the bge3 interface

So in other words, why am I getting a conflict between the management
interface and a leg of the bridge? Is there some weird loop going on? There
is no bridging loop happening, this is the only bridge on the subnet and
connects two datacenter switches (it's a firewall). The router is upstream
on the bge3 side of the bridge.

All these interfaces appear to be working fine, the  machine is bridging and
I can use the management interface with no apparent problems. If I ping
continuously, I get those two messages with every ping. Same for telnet,
etc. These messages only appeared for the past week, after the machine was
brought down and back up. 

This is freebsd 4.8. stock ipf is in use, ipfw is not in use. Disabling ipf
makes no difference.

Any input would be appreciated.