Yeah- I hadn't spotted the EXEC_32 changes- I think I just assumed it was
   the random scroggling of pmap and other internals that's been going on in
   alpha space in the last month- admittedly a bias on my part. Does anyone
   know for sure which one of the four different places are generating this
   error?


i reversed the logic in eeh's patch and it no longer fails on my
alpha.  i am awaiting my 32bit (shark) kernel to build so i can
test that it works there as well before commiting (ah, it is linking,
this should be soon).