I think the core argument is that distributing a proprietary driver that's obviously meant to be loaded into the Linux kernel to be useful is distributing a derived work -- since the Linux kernel has a syscall exception, but not a linking exception. So the problem isn't that Nvidia may or may not be distributing the kernel itself; it is distributing a derived work of the Linux kernel, and that must be licensed accordingly.

But I may have misunderstood/misremembered the details. It's been years since I cared enough to do a deep dive on this.

> I think the core argument is that distributing a proprietary driver that’s obviously meant to be loaded into the Linux kernel to be useful is distributing a derived work

Whether functional elements needed for interoperability are not covered by copyright (so that a work that seems “derived” through its use of them is not a “derivative work” at all) or whether using them is Fair Use (such that while a work using them might be a derivative work that would, but for Fair Use, require a license, it is nonetheless legal without a license due to Fair Use) seems to be a bit of a mixed bag, but either way using copyright to prevent functional interaction (except when you get into DMCA coverage of technical anti-circumvention measures, which is also part of copyright law, but not really applicable here) seems to be a difficult battle for copyright holders, in US law, at least.

> seems to be a difficult battle for copyright holders, in US law, at least.

heh, I wonder if you could argue that the catalog of functions/methods/symbols is a database and should thus be copyrightable under EU law.

it would obviously be hugely destructive if upheld, but, I think "database right" is fairly incoherent as a concept anyway, other than as an attempt to uphold "sweat of the brow doctrine" (ie "anything I worked to make should be mine by law"), which is already discredited in US law.