Subject: Re: Changes to link layer input routines
To: None <firstname.lastname@example.org>
From: Anders Magnusson <email@example.com>
Date: 05/15/1999 20:24:56
> On Sat, 15 May 1999 19:46:49 +0200 (MET DST)
> Anders Magnusson <firstname.lastname@example.org> wrote:
> > >
> > > (3) The VAX Unibus and Q-bus Ethernet drivers confused the h*ll
> > > out of me, and I didn't even try to clean up the way they
> > > use separate Unibus mapping registers for the Ethernet header
> > > and the actual packet. (Wacky, it's not like they even support
> > > trailers in the driver!)
> > >
> > Don't care so much about that; I'm going to change a lot of the uba
> > mapping code. Welcome, bus_dma :-)
> Excellent! Are you going to model it after the Alpha SGMAP code? :-)
Yes, most of which is applicable. I have a Qbus MIPS machine waiting at
home in Sweden when I return, therefore I need to do this :-)
> > About trailers: A fast check shows that someone was cleaning up the
> > code maybe a little bit too much a while ago (Hey, Charles :-)
> > cvs log if_uba.c
> > ...
> > revision 1.5
> > date: 1995/04/11 06:16:38; author: mycroft; state: Exp; lines: +3 -15
> > Sync with the rest of the world. Say goodbye to Mr. Trailer.
> > ----------------------------
> > ...
> Ah! Okay, this is nice. I think we should change ether_ifattach() to
> simply set IFF_NOTRAILERS everywhere, then. I mean, it's not like anyone
> USES trailers anymore :-) And, really, no other Ethernet interfaces
> would support them (that I know of).
Or remove the IFF_NOTRAILERS flag totally, it's use is deprecated anyway.
The only thing is that trailer packets must be received, which doesn't
work quite well right now. That should actually be a compile-time option or
something like TCP_COMPAT_42.