Yeah...if that is the case, wouldn't nvidia graphic driver's kernel-shim-to-closed-source-binary not be legit, since the shim defines an API in the kernel and then the closed source blob uses that GPL'd API?
Nvidia are already moving to a world where the Linux kernel driver is GPLed and in Linux mainline and only the firmware and userspace parts would be proprietary. So that potential GPL violation is in the process of being resolved. Eventually the signed proprietary firmware and the upstream Linux kernel driver would be shared between the proprietary nvidia driver and the open source nouveau driver. Of course they shoved parts of the Linux kernel driver into the proprietary firmware as part of the process, so....
The bulk of a modern video driver on linux (and most operating systems) is not in the kernel. The kernel module just facilitates communication between the user space drivers in mesa and the hardware.
Precisely. Though perhaps NVidia would argue they're giving themselves a license exemption?
If you understand drivers as plugins (which makes sense, really, drivers are kernel plugins), and if you understand plugins as derivative, it would be even worse. It would mean every driver on Windows's copyright is Microsoft's by default. And that the original NVidia closed source is already GPL.
Not just NVIDIA, ZFS too. From the GPL perspective, any non GPL compatible license is considered the same way.
That said, that part of the GPL on library linking was never attempted to be enforced in court.
For additional entertainment: redistributing a linked GPL program to a non-GPL compatible non-OS bundled (those have an exception) library is supposed to be forbidden too, but is very often done. For example, for the Visual C++ runtimes.
It’s supposed to not be okay according to the FSF, which does quite some overreach in their FAQs. A reminder that the FAQs themselves are _not_ part of the license.
There is a way that Oracle could solve the ZFS situation; release a CDDLv2 that is compatible with the GPL, then the CDDLv1 auto-upgrade clause would mean that all CDDLv1 software that hasn't opted out of the upgrade clause would now be GPL compatible. Some parts of OpenZFS would need to be relicensed or reimplemented, but that job would be much smaller than reimplementing ZFS for GPL compatibility. More discussion here:
Drivers aren't plugins. They link to and incorporate GPL code. You can't grant special terms in that scenario. The best they can do is dual license their headers. The hazy area is where no linking happens and it's just a pure API without code sharing.
Do DKMS drivers have to include headers as redistributed? I was under the impression they didn't have to, but I have very little expertise with DKMS and could definitely be wrong.
