AFAIU, Nvidia's legal argument is that their driver can't be 'derived from' the Linux kernel even though it's linked into the kernel since it's older than their Linux port and only is linked against their GPLed (and if push comes to shove, dual licensed GPL and proprietary) kernel abstraction layer. Judges don't care about linkers and the GPL doesn't actually say anything about linking.
This is not the derivation argument (that would be forcing GPL on Nvidia, which they want to counter, of course).
This is about GPL infrigment; you can combine GPL licensed code with differently licensed code, but only in your privacy, without distributing the result to others. Once you would distribute the result, you would be breaking GPL, thus having no right to distribute.
The GPLv2 literally defines itself in relation to trees of derived works:
> The "Program", below, refers to any such program or work, and a "work based on the Program" means either the Program or any derivative work under copyright law: that is to say, a work containing the Program or a portion of it, either verbatim or with modifications and/or translated into another language.
The argument that I've is that even when linked together, the Nvidia binary driver isn't a 'derivative under copyright law' because it doesn't have Linux specific entry points, and is older than their Linux port, and therefore isn't part of "The 'Program'" as listed in the GPL.
Your argument is, why some piece of software should (or should not) be relicensed to GPL.
My argument is, that since that piece of software is currently not GPL licensed; avoids the relicensing topics and explores how to solve the current situation wrt distribution of the combined work. Use by the end-user is non-issue, due to freedom 0.
This isn’t correct. WiFi router ships with GPL Linux and proprietary router software. The company / foundation creating the distribution just needs to have a license for the proprietary bits.
No, that's not entirely true either. Depending on where the proprietary bits are, it's the other side that is the problem. It is a violation of the GPL. So the usual workaround is to shift the responsibility to the user so that incompliant stuff is done by the user by the way of a script that runs on the user's behalf.
If the wifi routers are shipping with binary kernel modules, then yes, it they are breaking GPL. This problem is actually common with embedded devices.
First, if you're talking about the lawsuits over wireless routers, that wasn't Conservancy, that was the Software Freedom Law Center (SFLC) working with the FSF. (And some feel those lawsuits were poorly timed, preempting active negotiations in progress and being largely responsible for Cisco ceasing large Linux-based projects in progress.)
And second, to answer your question, partly because Linux is sufficiently good and has wide enough hardware support that it's still easier to use Linux legally under the GPL than to use another operating system, and partly because if you're going to use another operating system with narrower hardware support you might as well use a tinier embedded one and ship cheaper hardware (which is what Linksys/Cisco did with the WRT54G, shipping VxWorks and less RAM, and then selling the more-Linux-capable WRT54GL as a higher-end product).
The combining of proprietary and GPL code has to be made by the user and the result is un-distributable. That's why distributions cannot ship it.
What they can do is to make the above step easy (which Ubuntu does).