Re: Compatibility of Nvidia NVNET driver license with GPL

[Jeff Garzik <jgarzik@pobox.com>]

Karel Kulhavý wrote:
> On Wed, Dec 31, 2003 at 01:36:58AM -0500, Jeff Garzik wrote:
> 
>>Karel Kulhavý wrote:
>>
>>>Hello
>>>
>>>I faintly remember reading some article on the Net from Linus Torvalds stating
>>>something like if a piece of code is written specifically for Linux kernel, it
>>>must be under GPL.
>>>
>>>I have got an nVidia NForce2 board and downloaded their Ethernet driver (nvnet)
>>>and they say in README: "the network driver provided by NVIDIA is subject to
>>>the NVIDIA software license". How is with compatibility of such a behaviour
>>>with GPL of the kernel sources?
>>
>>
>>Since I am not a lawyer, my engineering suggestion would be to sidestep 
>>legal issues by using "forcedeth" driver, to drive your nForce NIC. 
>>It's fully GPL'd, and full open source.
> 
> 
> Suppose we would like to overcome these perpetual problems with misbehaving
> manufacturers by designing a PCI network card from scratch in the soul of
> free source hardware.
> 
> What are the requirements of the kernel for such a card to be cool instead of
> piece of shit? How should such a card behave from PCI point of view, should it
> support scatter, gather, how should interrupts be handled, what should be
> programmable and what not? How should be busmastering implemented?
> 
> I am seriously thinking about designing something like that (tailored specially
> for Linux) because of developping Ronja - if I included native Ronja support in
> "my" network card, I could remove the superfluous circuits that implement
> "bureaucracy" linke autonegotiation, TP link integrity etc. and concentrate on
> raw performance. Also multiple ports could be on one card (say 4) which would
> make the whole thing more nifty.
> 
> I have made to work the whole design chain from schematic design to production
> of raw files for PCB manufacturers. Also seen a design of video capture
> board for IDE connector so I judge implementing something on a PCI should'n
> be a pain in the ass higher than moderate.

If you are serious about this, we have tons of good ideas, and tons of 
suggestions on how to avoid bad ideas :)

OpenCores (http://www.opencores.org/) might be a good place to start, as 
they already have a 10/100 ethernet MAC which is working, and has been 
silicon'd:  http://www.opencores.org/projects/ethmac/  Full "source" for 
the MAC is available, in VHDL I think.  OpenCores also has PCI VHDL and 
other glue you may need.

You'll definitely want to implement autonegotiation.  It's a showstopper 
without that.  And if it's not gigabit ethernet, it's already outdated. 
  So it's a tough challenge.

Once you have basic gigabit ethernet working with 10/100/1000 
autonegotiation, let us know, and we'll pelt you with good suggestions :)


> Is it possible to obtain some PCI specs without NDA's and such bullshit?
> Is here anyone who has taken the PCI specs and rewritten them in their
> own words?

You don't need an NDA, but you do need to pay US$50 or so for the specs. 
  Or a nice person might just send them to you :)

	Jeff