With the changes I checked in today, and the ones from a few days agou,
most or all of the VM-related problems on the ARM port should be
fixed.  In particular, I have not been able to reproduce any problems
with programs SEGVing under load, MFS blowing curly pink chunks, or
copy-on-write problems with grep(1).

There is still a problem with GDB -- not losing breakpoints precisely,
but not being able to add/remove them, because uvm_io() can fail under
heavy paging load.  This isn't really `my department', but I may look
at it.

If there are still problems after tomorrow morning's SUP, let me know.