I just learned that you can peer IPv6 BGP across link-local addresses, even on a tunnel interface. This came as a surprise to me. I had always peered across site local addresses and I had assumed that you could not peer across link-local addresses because they are not unique.
(OTOH, I should not have been surprised. After all, IPv6 OSPF can form adjacencies across link-local addresses, so why not BGP).
So, if you are going to peer to, say, fe80::5, how does the router know which interface to connect through? Wel, in the example in this lab, we specify the update source too, so I am guessing that it uses that to determine which interface to connect through.
This has some strange implications. Suppose you have two link-neighbors, but they both have the address fe80::2. You can only peer with one of them. For the other one, you would have to use a unique address.
Futhermore, if your neighbor is a link-local address, then the update source must be mandatory.
The wierd and wonderful world of IP6!
