More odd behavior with the new PCI code (and, maybe, ata)

I have noticed a very odd behavior since the new PCI code went into
CVS. I have virtually no real data on the problem and no idea if it's
really serious. It's only happened three time, always in single-user
mode. 

1. boot -s
2. <CR> to start sh
3. fsck -p
4. adjkerntz -i
5. swapon -a
6. mount -a -t ufs
7. cd /usr/src

These are the only things I am sure of as common. After this, an attempt
to run something will simply hang. In mergemaster I had it happen when I
tried to do an edit (e b) in a merge. Another time 'make installworld'
hung in 'mktemp'. Interrupting the operation (^C) and restarting does
not seem to help, but doing some other operation that accesses the disk
seems to clear the problem. (E.g. ls /usr/src).

Once cleared, the problem never seems to recur. I have not seen it in
multi-user mode. I can't imagine what could be causing it, but loss of
interrupts from the ata MIGHT explain it.

Any suggestions on where to look?

/tmp is on real disk, not RAM. (I've seen reports of problems with
md's elsewhere.) 

This is probably not enough information to be useful, but I'll try to
collect more information when/if I see it again. I am hoping to find out
if it's just me or a more common issue.
-- 
R. Kevin Oberman, Network Engineer
Energy Sciences Network (ESnet)
Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab)
E-mail: oberman_at_es.net			Phone: +1 510 486-8634