Both Nvidia and AMD bypass legal issues because they don't distribute mixed GPL and proprietary code. Basically their proprietary kernel modules is being linked with Linux kernel by end-user and this totally fine for GPL as long as result is not distributed.
So blob drivers that clearly wasn't originally designed for Linux and work via GPL-ed shim (which isn't the case with VMWare) may be not derivative work. Anyway I think it's wrong to say that kernel developers (at least Dave Airlie as we talk about DRM/GPU drivers) don't do anything because they do have anti-leech rules that actually working:
- Some features only exposed as GPL-only as their usage clearly prove that driver is derivative work of kernel. See DMA-BUF.
- Open source kernel drivers that only used by proprietary code aren't going mainline. This is benefit whole ecosystem more than GPL-only kernel with blobs in user-space.
So it's become more expensive to maintain out-of-tree drivers and not all features are available for them. This actually working because Intel does have open source kernel and user-space driver even if their Android user-space is proprietary. AMD also switching to hybrid model.
Personally I find these rules more reasonable than attempts to enforce copyleft that are extremely hard from legal standpoint as there is no single copyright holder. For example FSF require to sign CLA to contribute to their projects in order to relicense code to newer GPL version or enforce it.
> Some features only exposed as GPL-only as their usage clearly prove that driver is derivative work of kernel.
The marking or not marking of things as GPL-only has no affect on whether or not using those things makes the using thing a derivative work. Whether or not X is a derivate work of Y is determined by how much of Y's copyrighted expression is incorporated into X.
Thanks for clarification. Sadly I'm not an expert in copyright law and post of "jnky" about GPL_ONLY probably explain better what I attempted to say.
What I wanted to say is that GPL_ONLY exist and does what it's intend to do: make it clear to driver maintainers (Nvidia as GPU vendor) that certain functionality shouldn't be used within proprietary code.
Of course this isn't actual legal limitation, but Nvidia also want to have open source drivers for Tegra and certainly don't want to get a finger on their next pull request.
The license-writer's published opinion on whether an activity is a copyright-protected activity that needs a license is of no special status (if the license-writer is also the copyright holder, it might have special status as itself being an implicit license in those circumstances where the opinion is that a license is not necessary.)
What is controlling on this question is copyright law, not the FSF's opinion. And, clearly, the FSF isn't a neutral authority here, they have a positive interest in portraying the need for a license as being as broadly as they can to encourage use of the license and adherence to its other conditions.
User-does-the-link does not change whether the blob is derivative work or not.
No, but things are significantly more straightforward if the combination Linux-kernel-plus-nVidia-module is distributed; this combination is clearly a derivative of both the Linux kernel and the nVidia module that make it up. There is far more room for argument if the module is distributed on its own.
