Linux Developers Still Reject NVIDIA Using DMA-BUF
Phoronix: Linux Developers Still Reject NVIDIA Using DMA-BUF
Going back to the beginning of this year there's been talk of NVIDIA looking at Optimus support for Linux and in August they confirmed they were working on NVIDIA Optimus Linux support. As part of their Optimus Linux implementation they want to use DMA-BUF for the multi-GPU interactions just like the open-source drivers, so that they can all work together. However, kernel developers continue to reject this notion...
Well, what have you expected?
You just DO NOT modyfy somebody else code to just change the license. You just do not.
You ASK for such change.
So why are Nvidia so afraid of opening up their drivers? I mean plenty of worldclass stuff in linux, my own plugins included. They are a hardware seller right? So the drivers that come with are open-source, what harm would it do? And sharing that, does that make anyone more competitive? Think about optimizations from the users aswell. Probably to the point of optimal. And generalized for any driver. So everyone contributes. How can that be negative?
Peace Be With You.
Well, I won't blame the kernel devs for preventing Linux to become stuffed up with more proprietary software.
They don't release code. They don't help with code written by others. They don't even release documentation. But now that they're not having the upper hand, they ask if pretty please Linux itself can weaken its license status in order to enable them to continue doing the f**k that they want, as they have continuously done until today. How brazen-faced can you get?
Anyway, I think that the kernel developers should just give in and allow them to use the API, only because otherwise NVIDIA will probably come up with some proprietary solution, which will be even worse for NVIDIA users. Nothing good is to be expected from NVIDIA when it comes to open source, and binary modules are probably already a GPL violation anyways.
The modification doesn't change the license. The kernel already has this exception when linking userspace interfaces.
Originally Posted by przemoli
Originally Posted by Paradox Uncreated
If I contribute code under the clause that all linked code is also under GPL (or compatible), you can't just change that to GPL + Linking Exception. It's a different license.
Originally Posted by WorBlux
I am happy with the kernel devs' position on this matter. If Nvidia wants to play in GPL land, they should release their source code under GPL. If not, then FUCK YOU NVIDIA!
Not to mention, but in addition to intellectual property license, there is also the very significant security matter to consider in sharing buffers between multiple GPUs, especially when it comes to code that is a closed trap well known backdoor ridden plague like nvidia. Keeping symbols as GPL-only, at least the linked drivers can be audited regarding security concerns. This is obviously not a possibility with closed source crap, so allowing that closed source crap to interface would be introducing even more of a security vulnerability than there would otherwise be in the system.
What is EXPORT_SYMBOL_GPL and why is it so important it is GPL only?