> another thing to be concerned about is fragmentation of kernel_map.
> I wouldn't be surprised if this change increased the number of kernel_map
> entries by an order of magnitude typically.  this may not be as much
> of a problem now that we're using rbtrees for vm_maps, but it's worth
> looking at.
> 
Would it be a bad idea to use a separate submap for it? So that
it won't have to compete with other stuff using kernel_map, and 
so avoid problem on machines with small kerbel virtual memory.

-- Ragge