> > > I don't think uvm_km_free() would actually need to be told which flags were
> > > passed to uvm_km_kmemalloc(), it can probably figure everything out from
> > > the contents of the map.
> > 
> > i don't think keeping more info in the lower layer is a good idea
> > because
> > - it prevents map_entry merge between different allocation types and
> >   thus increases search cost.
> 
> I wouldn't think we should merge pageable and non-pageable entries anyway.
> or would there end up being more than 2 types?

i was talking about kva-only allocations.
because callers of uvm_km_free often knows that there's
no remaining page mapping on them, i'd like to distinguish them
from malloc(9)-type allocations for efficiency.

YAMAMOTO Takashi