> Well, after todays SUP, a new kernel died while building libc:
>
> May 29 17:45:46 sockdev /netbsd: panic: utlbmod: invalid pte
> May 29 17:45:47 sockdev /netbsd: syncing disks... 29 29 29 29 29 29 29 29 29 29 29 29 29 29 29 29 29 29 29 29 giving up
>
>
> After rebooting and completing the build of libc, the system is now
> working on the rest of userland and is still alive.

It's very interesting.  Panic condition could not be reproduced.
Further more, dump is now broken (in a different way than mine...) I
don't touch SCSI code (yet), so the reason might sit behind reworked
spl (but code was just a replacement to do exactly same things). I go
back to code to what's the possible cause.  Thank you for your
valuable feedback. 

Tohru Nishimura