> where it calls
> kthread_create1() to make the new threads. At that point, we are
> actually running at spl0(), just like we should. But the kernel threads
> are being created at splhigh(), regardless.

New process always starts splhigh(), duh.  It made little matter
because any forked process dives into user mode immediately right
after the first CPU tick is assigned for it.  The introduction of
kthread uncoverted the potential issue.

It's a bug should be fixed, I think.

Tohru Nishimura