[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: ITP: Intel e1000 gigabit NIC driver kernel module

On Sun, Oct 20, 2002 at 12:54:58PM +0200, Ard van Breemen wrote:

> On Thu, Oct 17, 2002 at 04:06:36PM -0700, Joseph Carter wrote:

> > I don't know how easy it is to get the e1000 source, but the driver search
> > page Intel's other pages direct you to for the Linux driver is 100%
> > Windows only.  The current version of the e100 driver is GPL, but the
> > version in Debian is under another license and placed in non-free because
> > of patent issues (which have obviously since been resolved..)  I had to
> > find the current driver because I needed ethtool support for ifplugd.
> Ok, ok, I am a little bit slow with the e100 package, but that is
> because if the new package breaks because the source is not ok,
> it will cost the company I work for a lot of money...

Eh, I only now found out this thread is also on debian-devel.
So I will disclose a little bit about my original plans for the
e100. A while back Wichert contacted me and said the e100 driver
looked free, but contained some unknown patented code, which
Intel did not want to disclose: Intel did not want to tell
*which* part of the code is patented, nor did they want to tell
which patents applied.

I then put the package in non-free, and contacted my contacts at
Intel about it. The contact told me that he would try to get some
sense about it, but the latest thing he heard is that it was in
flux. (Actually that was the latest thing I ever heard of them.
At least broadcom keeps supplying me with the latest drivers).

Talking about it with Arjen van der Ven (redhat), he told me the
later drivers were even worse in coding (he had to audit them),
because when something happens that is not resolvable, it would
just panic, or cause other (worse!) harm. An updated eepro100
which should have a better performance than the original eepro100
driver would be in the 2.4.19 series. And if you want my personal
opnions: David Miller once said that the broadcom driver was
horrible (which was probably only about the coding style), but he
has not seen the e100 driver, and I am not talking coding style,
although the driver would be rather clean if you exclude the

So that's when I decided that supporting e100 would not be good
anymore, and that a shift to the eepro100 would be a good thing.

Since then I've been contacted by other people that 1) there is a
GPL'd version (thanks Intel, for keeping me updated), and 2) the
eepro100 does not support pcmcia cards.

At that time I held back phasing out the e100 package, and
uploading a new version. Because the e100 was definitely needed,
but did the GPL cover the patents (which means shifting it to
debian instead of non-free)?

I already have looked at the new drivers, but since they have not
included my (or any) vlan patches that are absolutely necessary
to make it work with 8021.Q, I have not packaged it yet.
Packaging it without vlan support would probably break a lot of
servers. I can try sending my patches again, but including my
patches means, you can have vlans with this driver, without using
Intel's proprietery iANS. Which means you can easily switch NIC
vendors without giving up your infrastructure. This looks a bit
weird, since the e1000 contains HW_VLAN support...
(vlan support with e100 is added by calling
e100_config_long_rx(bdp, true) somewhere in init...)

I will look at the e100 it this week, since I probably have some
time then.

BTW: both e1000 and e100 are in 2.4.20!
But then again: who is going to run 2.4.20 on production without
regression testing?

mail          up   96+03:13,     4 users,  load 0.00, 0.00, 0.00
mistar1       up   28+14:01,    18 users,  load 0.00, 0.16, 0.40
Let your government know you value your freedom: sign the petition:

Reply to: