This needs thorough review... I don't know what I'm doing here. Probably not yet ready to be committed because of that.
"The AT> keyword takes the name of a memory region as an argument"
I still need to debug this, since it's different with ld.bfd.