Excerpts from netbsd: 16-Feb-98 Re: other fix for the "unwi.. Ignatios
Souvatzis@theor (407*)

> > Would this be too restrictive?
> 
> Hm... thinking about frame buffers, maybe. Thinking about the
> controlling registers for the frame buffer devices, no.

It's actually worse - processes which make /dev/mem to
read kernel data are affected too.
Otoh, the situation is comparable to the coredump case -
device regions are not dumped and nobody has complained
so far.
What should be done Imho: extend the driver mmap() interface
to tell the VM system if the region can be accessed side effect
free and keep that information in a flag in the vm_object.

best regards
Matthias