Patch to kill ioremap_mm

[John Rose]

> We discussed that, it's sort of step #2 :) The idea I have is to fold
> normal ioremap into the vmalloc space like other archs, ditch imalloc,
> and for now to keep the top 1TB for the explicit mappings (PHB IO
> space). David or I will do that after we are finished with other things.

<Apologies again for resend.  Removed bad list name from cc and
replaced>

I would assume that we'll still have a path for creating mappings before
mem_init_done?  Ack, I'm already imagining the slot/phb DLPAR problems
that vmalloc may not be able to handle.  For example splitting existing
regions, removing mappings within a given range, etc.

John