Linux 6.6 To Better Protect Against The Illicit Behavior Of NVIDIA's Proprietary Driver

[mSparks]

it has always been violating the GPL

Not at all.
e.g.
Someone writes a MIT licenced module using sme_active() - GPL not violated
Someone else takes that module and incorporates it into a propriety module (neither GPL or MIT violated)
sme_active is marked "GPL only" in the 5.9 kernel
the propriety module breaks despite no GPL violation occurring before or after.

This is exactly what happened with pci_set_consistent_dma_mask, which is not "GPL only" but calls functions that became GPL only.

Like there is nothing fundamentally wrong with having "very GPL exports" marked as such, good practice in fact.
But you can't blame other people for the lack of transparency that existed before without appearing like an ignorant, MIT/BSD licence hostile actor that never opened an IDE in their life, or just a jealous moron that is mad better developers are getting paid more than you.

[Grinness]

Not at all.
e.g.
Someone writes a MIT licenced module using sme_active() - GPL not violated
Someone else takes that module and incorporates it into a propriety module (neither GPL or MIT violated)
sme_active is marked "GPL only" in the 5.9 kernel
the propriety module breaks despite no GPL violation occurring before or after.

This is exactly what happened with pci_set_consistent_dma_mask, which is not "GPL only" but calls functions that became GPL only.

Like there is nothing fundamentally wrong with having "very GPL exports" marked as such, good practice in fact.
But you can't blame other people for the lack of transparency that existed before without appearing like an ignorant, MIT/BSD licence hostile actor that never opened an IDE in their life, or just a jealous moron that is mad better developers are getting paid more than you.

Yes it has (specifically the NVIDIA blob) -- otherwise kernel dev would not waste their time on this
It has been written million of times and point out by kernel devs, it is not about 'some one doing this or that'
Possibly other binary modules do the same, the changes have been made in an attempt to take care of the general misbehavior -- period

[Grinness]

or just a jealous moron that is mad better developers are getting paid more than you.

you have no idea how much I am paid, nor what I do, and it is none of your business

[r1348]

The amount of stupid in this forum related to any article about NVIDIA is staggering. Let's look at some market caps.

• NVIDIA: $1.2T USD
• AMD: $171.13B USD
• Intel: $143.69B USD

Guess who is absolutely crushing revenue expectations?

Nvidia is riding the AI hype hard, I hope you don't expect such values to last for long...

However, this is a technical and legal dispute that has nothing to do with their respective market caps. The GPL allows some things and doesn't allow others. It's still infinitely more permissive than the proprietary license Nvidia published their drivers under. They can either play by GPL rules like everyone else, or go play elsewhere.

[mSparks]

Yes it has (specifically the NVIDIA blob) -- otherwise kernel dev would not waste their time on this
It has been written million of times and point out by kernel devs, it is not about 'some one doing this or that'
Possibly other binary modules do the same, the changes have been made in an attempt to take care of the general misbehavior -- period

Well, I think there are very valid questions why the Kernel Devs are creating this chaos, Maybe they are paid by Microsoft and hate Linux?
Or maybe they are just incompentent.
But nvidia doesn't seem to be having anywhere near the same problems working around their BS than everyone else.

https://github.com/openzfs/zfs/discussions/13471

https://forums.ni.com/t5/Instrument-Control-GPIB-Serial/ERROR-GPL-incompatible-module-nikal-ko-uses-GPL-only-symbol-usb/td-p/731730

https://access.redhat.com/solutions/309503

https://linux-kernel.vger.kernel.narkive.com/dNMsfttm/gpl-incompatible-module-error-message

Probably best to just strip out any "GPL_ONLY" symbols from the kernel completely TBH, replace them with their BSD equivalents (that the authors probably copy pasted from anyway), then this wouldn't be a problem.

GPL only mutexes...
Yeah right, that is bad behaviour.

Not using them oblivious to the fact the moron that wrote the export somehow thinks anything with a mutex must be derived from GPL code.

[horizonbrave]

See chinese android tv boxes for a nice primer.

Could you help me out? I'd like to dig more into the Chinese android boxes issues.
Thanks

[horizonbrave]

The amount of stupid in this forum related to any article about NVIDIA is staggering. Let's look at some market caps.

• NVIDIA: $1.2T USD
• AMD: $171.13B USD
• Intel: $143.69B USD

Guess who is absolutely crushing revenue expectations?

What a troll

[citral]

Could you help me out? I'd like to dig more into the Chinese android boxes issues.
Thanks

https://github.com/DesktopECHO/T95-H616-Malware

[qarium]

The promise (whether real or imagined) of ROCm support on the consumer 7000 series sparked interest, but never materialised.

tell me why is the consumer 7000 cards even interesting ? in AI workloads you always hit the memory wall...

the AMD PRO W7900 has more shaders than the 7900XTX 6100vs6440 and also 24GB vram vs 48GB vram.

of course the price 4000€ you can sometimes find it at 3960€ but if you compare it to nvidia prices this is like nothing.

[piotrj3]

Forking doesn't allow NVIDIA to re-license the relevant parts of the code... and altering the export symbols is pretty much a slam dunk way to show willful infringement.

And yes it's about patents, there are a whole heap of royalty free licenses granted to the kernel provided the gpl if followed.

And under what license line Export_SYMBOL (non GPL?) is?

That is right it is under GPL license. Same as EXPORT_SYMBOL_GPL.

Now why 2nd exist? The reason is very simple, GPL does not set a thick line in stone where GPL violation does happen. Router maker deploys router made on Linux. They do develop certain utility let's say something to control antenna. They release 3 versions of router :

1st. User mode utility, doesn't exist in kernel mode.
2nd. non-GPL kernel module not using GPL symbols.
3rd. non_GPL kernel module using GPL symbols.

So what GPL license thinks about it according to their FAQ. https://www.gnu.org/licenses/gpl-faq...ereAggregation

By contrast, pipes, sockets and command-line arguments are communication mechanisms normally used between two separate programs. So when they are used for communication, the modules normally are separate programs. But if the semantics of the communication are intimate enough, exchanging complex internal data structures, that too could be a basis to consider the two parts as combined into a larger program.

So in nutshell GPL doesn't define hard barriers. What matters is intention and intentions are up to interpretation. When it would be hard to prove intention of that user-mode utility was made in mind of creating derivitive of work - probably that utility could work on other systems POSIX compliant and there are non-GPL POSIX systems.

2nd. case is more interesting one. Could be made a case that they do only use symbols meant for general use, could be made a case that it is distributed as one program to control router. Anyone's game.

3rd. case - when you use symbol exported as GPL-only it means you fully know intention and you are clearly aware of what you are doing. Proving intention in court would be easy and I think Linus Torvalds wrote something in this style and said he discussed it with lawyers and they liked that idea.

This is real reason why EXPORT_SYMBOL_GPL exist - because if true GPL violation happens (including distribution) and you truly use GPL symbol, court will not have mercy at you, you will be striked down because your intention of breaching GPL license is clear.

Now can you rename EXPORT_SYMBOL_GPL to EXPORT_SYMBOL?

Technically yes you can. But you still use symbol in reality GPL based (relicensing didn't happen) and it is clear you renamed it, in mind of intention mentioned above. So in nutshell if you would rename EXPORT_SYMBOL_GPL to non-GPL and after ship ZFS or Nvidia driver together with renamed kernel - you probably still have bad time because your intention didn't change.

Thing is neither ZFS neither Nvidia ship linux kernel with their drivers. So nvidia doesn't breach GPL license - they give you that code and you are free to do whatever you want as long as nvidia license permits it. You take that code (you yourself) install it into linux system. Now you have bound GPL code with non-GPL code. Now you did it yourself and distributed (compiled stuff) stuff to yourself so at worst you can sue yourself (lol). Nvidia corporation or Oracle with ZFS is not responsible for your actions. Now you probably don't want to sue yourself. Now imagine you (specifically you) distribute such system to diffrent user. Now that user can say hey i have that linux with GPL code but there is some module i cannot find source for, please send me source code of it. And this is moment when GPL breach happens and it is your breach, not Nvidia's breach.

So just saying it is not possible to sue Nvidia for breaking GPL license as they don't distribute GPL code themselves.

DMCA breaking (that some people are claiming) is in my opinion very stupid claim as DRM is not compatible with GPL. You would have to mark that code or feature explicitly as DMCA protection measure.... but that wouldn't be compatible with GPL license itself that says explicitly you can do whatever you want with code as long as you meet certain distribution criteria. In fact GPLv3 has explicit mention that you

When you convey a covered work, you waive any legal power to forbid circumvention of technological measures to the extent such circumvention is effected by exercising rights under this License with respect to the covered work, and you disclaim any intention to limit operation or modification of the work as a means of enforcing, against the work's users, your or third parties' legal rights to forbid circumvention of technological measures.

GPLv2 happened before such thing as DMCA exist, but I believe it would end up in similar manner. In nutshell yes you circumvented DRM. But you weaved the right to execute it as your legal right.