On Wed, 4 Mar 1998 20:28:11 +0100 
 Frank van der Linden <frank@wins.uva.nl> wrote:

 > I'm not sure anymore about this case, but: you may also simply be
 > running out of KVM, even if you have enough PDEs. For machines >= 64M,
 > you will always get enough PDEs for all of the possible 256 - 4 Mb
 > of KVM. But the kernel on the i386 simply has a limited amount
 > to work with.

I don't think it's KVM, in this case.  It looks like it might be possible
to lose on a reference to a USER pmap... i.e. for some reason, the PT page
hasn't been allocated by the time pmap_enter() is called, for some reason,
possibly trap() being confused by how the reference is made to the address
(seems like the hueristics used to determine user fault vs. kernel fault
are fragile; this is definitely the case on the m68k, as Chuck Cranor,
Steve Woodford, and I discovered last week while trying to find a panic
condition on the hp300 that occured while using UVM).

Jason R. Thorpe                                       thorpej@nas.nasa.gov
NASA Ames Research Center                            Home: +1 408 866 1912
NAS: M/S 258-5                                       Work: +1 650 604 0935
Moffett Field, CA 94035                             Pager: +1 415 428 6939