The code in question always executes in the "top half" of the kernel in
process context.  Since there is never more than one process executing
in the kernel at once in NetBSD (at least until SMP stuff gets worked on :)
there is no race here because neither process yields the processor (or
calls something which might do so) with the flags in an inconsistent
state.

When SMP arrives, it'll have to put locks around manipulations such as
this and/or recast them in a more MP-friendly paradigm.

==John