Mark Kettenis <mark.kettenis@xs4all.nl> wrote:

> > It probably cannot use the guts of uvm_map_teardown(), because I think
> > on shared-U+K architectures it can damage the kernel mappings.
> 
> I don't see how.  The userland vmspace should conatain no mappings
> with kernel addresses.

This is just a worry about some MD/MI mistake that might exist in
actual code.  Same as I worry about the selectors.

Obviously on the pmap side, there are multiple probable issues.

claudio did something clever by moving the pmap calls, and calling it
twice.  That moves much of the sleeping-sensitive de-allocation code
to the process itself.  The pmap parts remain in the wrong place, because
we need to tread more carefully with that.