Revamping optimised in_cksum/in4_cksum/in6_cksum support

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

Hi all,
there are currently three very similiar checksum algorithms in the
IPv4/IPv6 stack that platforms should provide assembler versions for.

in_cksum is the core and implemented on all platforms.

in4_cksum and in6_cksum are variants of in_cksum that skip parts of the
mbuf chain and instead use a separately computed initial checksum over
the header. in6_cksum is currently only implement for i386.

I would like to simplify this and implement in4_cksum and in6_cksum in C
and have a single MD backend function (md_in_cksum).

md_in_cksum takes the mbuf, len, initial cksum (as uint32_t) and offset
to skip. The required modifications for the existing in_cksum
implementation is loading the initial checksum instead of zeroing it and
a small loop to compensate for the offset. I expect this to two loads
and a branch for in_cksum, which should be cheap enough to not bother
with a single entry point.

Opinions?

Joerg