The kernel devs wouldn't have to put these defenses if NVIDIA played fair and just open-sourced their driver like every other major GPU vendor.
The kernel devs wouldn't have to put these defenses if NVIDIA played fair and just open-sourced their driver like every other major GPU vendor.
The fear could be that NVIDIAs "hacks" helps with performance and playing by the rules of the kernel developers will cost... say 10% in performance, that would be a pretty big hit for many.
I don't think there's a fear that NVIDIA will stop developing Linux drivers, that doesn't seem realistic, but I also doubt that they are going to open any more than they absolutely have to. NVIDIA isn't a open source friendly company and no Linux kernel hacks are going to change that. They'll change when it's financially beneficial to do so.
You made a bunch of claims based on having no information about the subject that have nothing to do with the post you replied to...
Like... what am I supposed to do with what you wrote?
---
As an aside, from my personal experience, NVidia is a very bureaucratic company with very few people having much of an authority over software it writes. It also has a culture of restricting many freedoms to their employees (as in, mandating OS for their laptops, taking away root privileges on work laptops, requiring a lot of unnecessary reporting when dealing with programming issues etc.)
I would expect that the decision to not open-source the driver was based on some policy established ages ago. Nobody can remember who and for what reason made the decision, but nobody is bold enough to challenge it. Mostly because the organization grew too big and the decision must've been probably made by some title which no longer exists and whoever is in "position of power" now doesn't have any real power.
As for performance... well, I wouldn't expect it to tank. Here's one anecdote that I know about since I'm personally affected by it. On laptops with two graphic adapters (typically Intel and NVidia) where one is used for power-saving mode and another one for better 3D rendering, when such a laptop is connected via HDMI to a big monitor, NVidia's adapter starts overheating and using a lot of unnecessary resources. Turns out someone left out some kernel logging in the very hot loop of the piece of the driver related to HDMI probing.
In other words, I wouldn't be surprised if NVidia's driver would overall benefit from few extra pairs of eyes looking at it. It's hard to imagine that they somehow found a way to utilize system resources so much better than a legitimate driver would. And even if they did, there's a good chance they'd be able to convince Linux kernel developers to let them have it.
I don't disagree that the drivers could benefit from a few extra people looking at them... though I doubt that many outside AMD is looking at the Radeon drivers. There's no harm in NVIDIA opening their drivers, but they won't do it, nor do they like that the kernel developers are saying that users are on their own if they do load their proprietary drivers.
Now what I originally tried to address was people defending NVIDIA, because their work depends heavily on things working as they currently do and what they are looking at now is an ever so slightly more uncertain future.
Edit: Oh I see, you're confused about "if the thing that powers their products suddenly taken from them", is that what you're thinking in terms of authorship? I'm referring to having the drivers taken away, feature, performance, whatever the negative outcome could be in the mind of those defending NVIDIA.
There would be literally zero reduction in performance by relicensing currently-proprietary code under the GPL.
For people who depend on their Nvidia hardware to work under Linux, I understand they are pissed off, since it break their support for a non-technical reason.
It is not unlike DRM when you think about it. In the "Digital Rights Management", it is a technical measure to protect the Linux kernel copyright ("copyleft" in this case). In the "Digital Restrictions Management" sense, it is a feature that is design to break something that works. And it is also about the "Direct Rendering Manager", but at least, no one complains about that DRM.
I just with that Nvidia had more competition. For not, they can do whatever they want because they have the best GPUs, both with regard to open source and price. AMD GPUs are meh right now. Maybe Intel will be competitive one day, from my experience, their Linux / open source support is rather good, but maybe because when it comes to GPUs, they are the underdog.
They need to redirect their anger at the correct culprit -- Nvidia. The non-technical breakage is a result of Nvidia continuing to try to sneak around restrictions that they should have been abiding by. Nvidia could have spent the resources to get their drivers to compliance, but instead they chose to use them to try to circumvent the restrictions.
It says it's dual licensed MIT/GPL2
Also it's not the same codebase as their proprietary driver, it is inferior in performance for desktop/gaming users. So it is not really a complete alternative yet.
also gives them a way to relieve pressure on the AI market. "install the open drivers bro", assuming it works.
Things not showing up in courts proves basically nothing, and if you think it does, well, I'd be happy to sell you a lawyer who will charge you $200k to win a case over the course of 3 years netting you a judgement of $25k at the end.
Regardless of the legal aspect though, nvidia's clearly in the moral wrong. The kernel ships with a license that tries to protect user's freedom to modify its source code, and the source code for derived works. nvidia is acting counter to that license, and counter to its user's freedom.
It would lend credibility to your arguments if you understood the difference between criminal and civil actions.
Not every violation needs to be pursued in civil court. It's not even possible -- the courts would melt down and stop operating, or they'd raise their fees to the point of reducing caseload. "No lawsuit" does not imply "no case".
Not everyone can sue; only people who have standing. Traditionally this would be people who own the copyright -- but on the whole, the core Linux developers have been very resistant to the idea of ever suing. I know Greg KH has come out firmly against it, and a quote above has Linus saying something similar. (I think their attitude is as foolish as saying that you should never ever ever use violence to prevent crime or defend your country.)
You'll have to forgive me for not wanted the only functional driver of hardware in I very much rely on in a daily basis for both work and personal use, to lose functionality.
Yes, they shoved all the important bits in the firmware, but this still could make for a much more stable Linux module, if NVIDIA actually respects license/code boundaries.
Either way as you mention, the fact it's only GeForce RTX 20-series and newer is a dealbreaker for people with older cards as well.
Enforcing licences and neutering illegal violation of license is not a witch hunt.
There are many people who want to use on their computers both the Linux kernel and NVIDIA GPUs. NVIDIA maintains their Linux drivers only for the benefit of the NVIDIA GPU owners who desire this, they do not get any direct revenue from this activity.
So at most you might say that Linux is a victim of the GPU owners that want to use Linux, despite the fact that some kernel developers hate the owners of NVIDIA GPUs and try to discourage them to use Linux.
The problem is that Linux does not have a clear license that would enumerate what Linux users are allowed to do, and even if it would have such a list, that would be somewhat ridiculous, because one of the main claimed objectives of the Linux development has always been that its purpose is to remove the restrictions imposed by the commercial operating systems.
If Linux would have a license forbidding the Linux users to use it in weapons of mass destruction, that could be justifiable on moral grounds. On the other hand, a license that would forbid the Linux users to buy NVIDIA GPUs or any other piece of hardware cannot have any justification. While some kernel developers did not attempt to do something so pathetic, they still do not find anything better to do than to find new ways to make the life harder for those Linux users that have good reasons for using something like NVIDIA GPUs or the ZFS file system.
It is normal for the kernel developers to not do anything to contribute to proprietary device drivers, but it is ugly and mean to try to sabotage them by actively preventing them to link to the symbols that must necessarily be used by a device driver in order to accomplish its function.
The linux kernel is respecting it's user's rights to know what code they run, and to be able to modify it.
nvidia is not extending that courtesy to their users, even though the linux kernel's possibly-legally-enforceable copyright insists that nvidia has to extend such a right. nvidia can release their source code as GPL if they want to use the GPL parts of the kernel.
Are you just as mad at nvidia for having a proprietary driver but not letting users apply patches to the driver? I assure you, the license nvidia distributes its drivers under is much more restrictive than linux's GPL license.
Like, if you're mad that the linux developers are making it harder for nvidia to violate linux's license and make a derivative work of the linux kernel, surely you're even madder that nvidia's developers have made it even harder for anyone to make a derivative work of nvidia's proprietary driver blobs, right?
If this theory, which is the only point of disagreement between me and Richard M. Stallman, would be accepted to be true, then what would stop Microsoft to claim that every program that has ever been written to run on Windows is the property of Microsoft?
When there are two hardware vendors, one who offers open-source device drivers and one that offers proprietary drivers, I would always choose the one with open-source device drivers.
Unfortunately, in the case of NVIDIA there are applications for which no other vendor attempts to provide competitive products, so there is no alternative choice. Therefore it does not matter if I would be "mad" at NVIDIA, as that could change nothing.
On the other hand, I am "mad" at this kind of kernel developers, because I do not accept that they have any right to dictate to the Linux users what kind of hardware they should buy. Obviously, they should not support NVIDIA, but inventing methods to sabotage the Linux users that have NVIDIA GPUs is not a justifiable activity.
If a Windows program is not a derivative of the Win32 libraries owned by Microsoft, what use does it serve once separated from the Microsoft libraries?
If an Android application is not a derivative of the Android OS owned by Google, what use does it then serve once separated from the Android base system?
If any C program that has ever been written (except those written for freestanding environments) is not a derivative of the C Standard Library, what use does it serve once separated from libc?
If your argument were right, with the exception of the very small number of programmers who write programs for bare-metal embedded systems, all the other programmers have written all their lives only derivative programs, so none of them (or of their employers) has any right to claim copyright ownership on what they have written.
And the GPL has always noted exceptions for system-provided libraries.
Kernel modules run inside kernel-space. They are completely unlike user space applications.
Why?
I cannot see any difference from the POV of copyright. There is no difference between the loading of a kernel module that must discover the addresses of the kernel symbols that it must use and the loading of a user-mode executable that must discover the addresses of the symbols that it needs from the libc shared library.
Also the GPL does not allow the bundling of non-GPL kernel modules with the Linux kernel but it cannot restrict in any way the right of any Linux users to use the kernel by loading any kernel modules they want.
These attempts made periodically by some kernel developers to cripple the performance of the non-GPL kernel modules by hiding the useful kernel symbols when such modules are loaded by the end users have nothing to do with the GPL.
They also do not hurt NVDIA or other hardware vendors, but only the Linux end users, whose systems become broken after kernel updates.
2. Yes, a kernel module will have to discover the address of kernel symbols in the kernel. And a user space application will have to discover the address of symbols in the libraries it dynamically links against. But neither the kernel module nor the application cross the OS/user-space boundary. The kernel module is derived from the kernel, and the user space application is derived from the libraries that it uses. But the user space application is not derived from the kernel.
3. Yes, users can do whatever they want. So sure, an enterprising coder could write their own NVidia driver module, have it use GPL symbols from the kernel, and nobody could stop them. What they cannot do is distribute that, and that is what NVidia does.
First, the "System Libraries" exemption also names explicitly the kernel.
More importantly, this exemption has nothing to do with determining whether some work is covered by GPL or not.
The "System Libraries" exemption allows a GPL program to be distributed together with "System Libraries", without providing the source code for those "System Libraries".
The GPL only claims to cover a work that is distributed to others ("conveyed") and which includes a GPL program that is "combined with it such as to form a larger program".
The GPL is not applicable to programs that use interfaces provided by a GPL program, but which are distributed independently of it to the end users, for instance kernel modules or shared libraries.
You're going to get nowhere trying to claim that a user space application and a kernel module have the same relationship to the kernel. Actually, less than nowhere.
So the answer is it’s still a library of functions that provides a substantial portion of a display-driver, even separated from the kernel. a windows driver module/library backend for such is at least as useful as any other code!
NVIDIA deliberately violates GPL license.
> The problem is that Linux does not have a clear license
are you sure about that?
The GPL is not that kind of license.
While GPL includes some use restrictions, they are logically contradictory.
GPL says:
"This License explicitly affirms your unlimited permission to run the unmodified Program."
So anyone can run the Linux kernel without any restrictions imposed by GPL.
The GPL then has a convoluted specification of how it should forbid a GPL program to be combined with other programs, so that copies of the combination would be transferred to others.
However those restrictions do not apply to kernel modules like those distributed by NVIDIA, which are not "combined" with the GPL Linux kernel until the end user loads the NVIDIA kernel module.
Because the GPL cannot be applied to such kernel modules, the anti-NVIDIA kernel developers take advantage of the fact that when a kernel module is loaded dynamically, in order to work, it must discover the entry addresses for the kernel symbols, exactly like a user-mode executable must discover the entry addresses for the libc shared library.
So they attempt to make difficult for the NVIDIA module to obtain the addresses for the kernel symbols that it needs. This is exactly like if in user space the libc shared library would attempt to not provide e.g. the entry address of "printf", if it would not like the name of the executable file from which "printf" is invoked.
If anyone breaks the GPL, these kernel developers break it, because instead of providing "unlimited permission to run", which means "unlimited permission to" invoke any of the interfaces provided by the kernel, they attempt to hide many of the interfaces when they discover that the Linux users own NVIDIA GPUs, with the purpose of crippling the performance of such GPUs.