[c-nsp] AToM between 2 PEs, originating on and accessible by the PEs

[Justin Shore]

I have a situation where I need run AToM across 2 links between 2 pairs 
of routers (7600s and 6524s):

7600-1 <-----> 6524-1

7600-2 <-----> 6524-2

I'm doing this because of what's between these two pairs of routers. 
It's a cobbled up solution across a dozen broadband aggregation routers. 
  We're riding a L2 VLAN across these devices between each pair of 
routers.  Unfortunately these damn things are stripping out IS-IS.  No, 
I can't eliminate these devices; I'm stuck building a solution around them.

I've tried GRE but that jacked the MTU.  I looked into L2TPv3 but that 
isn't supported on the ME6524s.  I'm now looking at AToM.  Fortunately 
the devices in the middle appear to be ignoring MPLS packets.

We meet the broadband routers on each side with a trunk port per router. 
  The native VLAN is L3 for the broadband subnets.  The tagged VLANs are 
carried across the network to our device on the other end.

I enabled LDP and MPLS on the VLAN for the pair of routers I'm working 
with.  I forced the LDP ID on both ends to Lo0.  I set up xconnect on 
the SVI on the 7600 and it didn't complain.  When I did the same thing 
on the 6524's SVI it threw this at me:

6524-2(config-if)#xconnect aa.bb.cc.dd 4020 encapsulation mpls
Incompatible with ip address command on Vl4020 - command rejected.
Error creating new xconnect config object

On each side I've set the SVI MTU to 9216, addressed it out of a /31, 
turned on ISIS, set the IS-IS circuit-type and metric, turned on MPLS 
and turned on LDP.  What am I missing on the 6524?  Eliminating the IP 
and reissuing the xconnect command brings AToM up immediately.

I've been searching through cisco.com and reading through MPLS 
Fundamentals and Layer-2 VPN Architectures.  Can AToM do what I'm trying 
to do?  I know that this isn't exactly the normal model for Ethernet via 
AToM but is this possible?

Any other suggestions on how to get between these 2 sites over this 
equipment would be helpful.  The fixed MTU on GRE tunnels is what hosed 
our earlier attempt to work around the IS-IS stripping problem.  I'm 
nearing my whits ends...

Thanks
  Justin