Peter Memishian wrote:
> > /*
> > * Compare two Ethernet addresses - assumes that the two given
> > * pointers can be referenced as shorts. On architectures
> > * where this is not the case, use bcmp instead. Note that like
> > * bcmp, we return zero if they are the SAME.
> > */
> >
> > #if defined(__sparc) || defined(__i386) || defined(__amd64)
> > #define ether_cmp(a, b) (((short *)b)[2] != ((short *)a)[2] || \
> > ((short *)b)[1] != ((short *)a)[1] || \
> > ((short *)b)[0] != ((short *)a)[0])
> > #else
> > #define ether_cmp(a, b) (bcmp((caddr_t)a, (caddr_t)b, 6))
> > #endif
>
> As the comment you included states, the code is assuming that the
> passed-in Ethernet address is at least 2-byte aligned -- which makes some
> sense since that's going to tbe usual case with an Ethernet header.
That's not the way I interpret that comment. The first sentence is
describing the first implementation (before the #else) which assumes
that the *architecture* allows accessing arbitrary addresses as shorts.
As originally written (both code and comment), ether_cmp was meant to be
usable on arbitrarily aligned memory, and when defined(__sparc) was
added, it was broken (or the interface was redefined with a new
restriction).
-=] Mike [=-
_______________________________________________
networking-discuss mailing list
[email protected]
