[3660] in linux-net channel archive
Re: new lance patch
daemon@ATHENA.MIT.EDU (Donald Becker)
Sat Jul 13 03:37:30 1996
Date: Fri, 12 Jul 1996 10:18:44 -0400 (EDT)
From: Donald Becker <becker@cesdis1.gsfc.nasa.gov>
To: Paul Gortmaker <gpg109@rsphy1.anu.edu.au>
Cc: Tekno Soft Snc <MC3641@mclink.it>, linux-net@vger.rutgers.edu,
tsbogend@bigbug.franken.de
In-Reply-To: <9607120155.AA26830@rsphy7.anu.edu.au>
On Fri, 12 Jul 1996, Paul Gortmaker wrote:
> > this patch reorganize the lance driver when is used with the lance32
> > option enabled. It free about 6/8Kb.
>
> However, I am not clear on why lance32 exists at all. Perhaps I am
> missing some crucial point that Thomas B. can clarify for me. When I
There is a major reorganization of the bit field sizes, which means
that although the source code is very similar the object file is quite
different.
I had earlier written a PCnet32 32 bit mode driver and benchmarked it
against the 24 bit mode. Even with bounce buffers active there was no
noticable performance difference. The chips are smart enough to use
aligned 32 bit data transfers independent of the 24/32 bit address
width mode.
There are two justifications for having a 32 bit mode driver:
Once in 32 bit mode, the 24 bit drivers will not recognize the card.
There is a slight performance advantage from avoiding bounce-buffers.
(Also it's probably slightly easier to convert a 32 bit driver to the
Rx-in-place driver structure, but only by 3 or 4 lines of code.)
Overall, I didn't think the win was big enough to justify releasing
another driver. But if someone else is maintaining it...
[[ I understand that it does mean a noticable amount of documentation
work for Paul, and some background bug-report-forwarding for the rest
of us. And some extra kernel bloat. ]]
Donald Becker becker@cesdis.gsfc.nasa.gov
USRA-CESDIS, Center of Excellence in Space Data and Information Sciences.
Code 930.5, Goddard Space Flight Center, Greenbelt, MD. 20771
301-286-0882 http://cesdis.gsfc.nasa.gov/pub/people/becker/whoiam.html
