Defence of the GPL realm (The H)

Posted Dec 18, 2012 9:52 UTC (Tue) by farnz (guest, #17727) [Link]

While I'm not a lawyer, I've obtained advice on this subject before. I'll try and summarize it here. Note that I'm deliberately not identifying my jurisdiction here - it's not legal advice, and may not apply where you are, it's just a summary of a discussion I've had with someone who's qualified to give such advice if you pay them.

The legal situation in my jurisdiction is likely to be that the module as component parts (i.e. the blob .o, and the shim source) is legally acceptable - you are being given a set of components written to an API that the kernel implements, and as long as the two are not combined, only the copyright holder of the binary blob components you've been given can control redistribution (so, in the case of the NVIDIA blob, you can distribute the binary blob plus the shim source freely, as that's what NVIDIA says is OK).

My lawyer suspected this theory would continue to hold if you distributed the compiled blob for a specific kernel version, as long as you distributed it independently of the kernel (e.g. if NVIDIA provided precompiled binaries for RHEL 5's kernel), but was less confident of this pronouncement. The legal reasoning would be that the module was not a derived work - it was written to the Linux functional API only, which is not protected by copyright law.

Where you hit problems is when you try to distribute both the kernel and the binary module together. At this point, my lawyer thinks you have to comply with the licenses on both parts at once - as the kernel license requires source under GPL, and the binary license says you can't distribute source, you're stuck breaking one set of obligations or the other, and thus hit the GPL's catch-22 - either you comply with the GPL (which means supplying NVIDIA's secret source in breach of your license from NVIDIA), or you comply with NVIDIA's license, which means that you cannot be licensing the kernel under the GPL, and thus cannot point to any license that permits you to distribute the kernel.

Note that my lawyer believed that distributing the two bits separately for self-assembly by your end customer is acceptable, as in that case it's possible that the customer will not assemble them, but instead will use the binary blob with a differently licensed kernel (e.g. if someone wrote a BSD-licensed shim layer for NetBSD that could make the NVIDIA blob work). It's distributing the combination together that's a problem; if the end-user has to go to enough extra effort to combine them, it's no longer a problem.

This affects (for example) embedded systems; if I build an arcade machine around a PC with NVIDIA graphics, I have to get the final owner of the machine to combine the NVIDIA module with the kernel before it works; I am not allowed to do that for them, as the resulting machine is no longer legally redistributable thanks to copyright law. In turn, the only plausible attack on NVIDIA is via this embedded route - find a customer who NVIDIA can't afford to lose who's distributing the binary blob together with the matching kernel binary, and go for them, in the knowledge that NVIDIA will step in to protect them.