FYI: creating and calling an opencrypto context (either in-kernel or
userspasde) causes a kernel panic on i386, if the context is for AES
plus a hash (sha1, and I think md5).  The problem is AES_specific;
DES/3DES conjoined with a hash work just fine.

Binary search on some old kernels showed this worked around July 2002,
to a kernel I have that claims to be 1.6U.  It's broken in 1.6W.

I'll be looking into this, but I'd also appreciate it if the folks who
helped clean up opencrypto to use thea newer sys/crypto AES could take
a look, too; as it looks like the problem right appeared then.

(I will very shortly be checking in a userspace tool that tests joint
crypto/hash usage.)