Bickering Continues About NVIDIA Using DMA-BUF

[pingufunkybeat]

Question: Whats the point of DMA-BUF if the people who need to use it, can't?

Most drivers can use it just fine.

[energyman]

why attacking Alan Cox?

He wrote the code. His choice of licence. End of story.

Nvidia are the ones moving around in a legally dark grey area doing morally despictable stuff. If they really care they could always open up their drivers. Or support linux driver development like AMD or Intel.

Nvidia wants to play dirty - and moronix should not help them.

[boast]

All this issue would be non existent if only nVidia went and fixed their own broken hardware (by writing their own DMA-BUF) when they released it.

But they never released it for Linux. They're just trying to do this based on user demand.


Nvidia should just create their own alternative dma-buf. Thats the point of the kernel, to be modified to your needs.

[Rigaldo]

Linux Kernel is GPL2 period. Don't like open source? Go somewhere else.

What part of what I said implies I don't like opensource? What part of preventing NVIDIA from using DMA-BUF promotes opensource?
Not allowing the use of DMA-BUF means they either make their own CLOSED stuff or no Optimus. So please explain me how that would help opensource?

No, people will probably not start jumping to nouveau. And if they do, many will have serious issues and won't care much about whether it's open or not when they just out of it, or from Linux in general. This won't help opensource in anyway, maybe hurt it though.
And NVIDIA will NOT opensource their driver. Heck, even if their management wanted to, they couldn't.

So even if I mostly said that as a joke, I stand by it, since I'd prefer the best experience for Linux users and the best fate for Linux. Nvidia delving more in open things will spread more openness, not the opposite. "Cox blocking" them will likely do the opposite.
Btw, I don't have anything Optimus related(not even an NVIDIA card), it's like It will affect me directly in any way.

I like opensource, but thinking realistically is cool too.

In the end, making an exception for NVIDIA would likely help opensource more than hurt it, if it happened.

Btw, I won't do anywhere because of some "OpenSource Nazis".(Not necessarily referring to Alan and co)

[tomato]

But they never released it for Linux. They're just trying to do this based on user demand.


Nvidia should just create their own alternative dma-buf. Thats the point of the kernel, to be modified to your needs.

it's not like the mobile chipsets are blacklisted in the binary driver, they knew about the issue and made a official release to their users on the line of "sucks to be you, we won't be fixing our hardware".

[mememe]

Can't NVidia just release their own implementation of the DMA-BUF API? After all one of the recent Dalvik/Java suits made it clear that an API itself cannot be copyrighted and therefore subject to a license agreement.

Edit:

I generally wonder why dynamic linking A to B makes A a derivative work of B. It's not like every Windows program, which links dynamically to Windows components such as kernel32.dll or user32.dll is automatically a derivative work of Windows.

[LLStarks]

Alan Cox is an asshole.

I have an Optimus laptop and as things stand, I will never get the expected performance out of my GT 555M on Linux.

Nouveau's use of dma-buf is not a solution. It works, but it's as slow as using the Intel GPU. Plus, the power management puts my card in an unusable state when I want to reboot into Windows.

I SHOULD BE ABLE TO USE THE NVIDIA BLOB AND DMA-BUF AS I SEE FIT FOR MY OWN PERSONAL NEEDS.

[GreatEmerald]

Guys, once again:

Kernel devs did not pick the kernel licence yesterday. They (Linus) picked it more than 20 years ago. They can't change it.

Binary kernel module using internal stuff like DMA-BUF violates the GPL, and nobody would be allowed to distribute that without risking a devastating lawsuit. That's the way it is. Renaming symbols just clarifies this, but does not in any way change the GPL, which prohibits it.

What Nvidia wants is some kind of tacit approval that this violation is "OK" so in the case of a lawsuit they can claim that they were acting in good faith and were misled. What Alan wants is for them to take the responsibility for the violation. If they think that it's legal for them to call kernel code from their binary module, they should simply ignore the symbol names. After all, there's nothing to worry about, right?

This. We have a few simple facts here:

1. The kernel is GPL, and the license states that everything that communicates to it through anything but system calls is considered a derivative work.
2. EXPORT_SYMBOL_GPL is part of the kernel.
3. EXPORT_SYMBOL is part of the kernel.

So they are both under the GPL, just the former explicitly notes that. Changing it makes zero difference, and the fact that it was changed itself can be considered an acknowledgement that it is under the GPL (if they didn't know that, they wouldn't have changed it).

The only ways to make DMA-BUF available for use with proprietary software is to either make it no longer a part of the kernel, or make it somehow accessible via system calls. Otherwise the functionality must be reimplemented in userland.

It's really that simple and clear-cut, when you look at it.

[Xilanaz]

What part of what I said implies I don't like opensource? What part of preventing NVIDIA from using DMA-BUF promotes opensource?
Not allowing the use of DMA-BUF means they either make their own CLOSED stuff or no Optimus. So please explain me how that would help opensource?

No, people will probably not start jumping to nouveau. And if they do, many will have serious issues and won't care much about whether it's open or not when they just out of it, or from Linux in general. This won't help opensource in anyway, maybe hurt it though.

+1

.....

[GreatEmerald]

I SHOULD BE ABLE TO USE THE NVIDIA BLOB AND DMA-BUF AS I SEE FIT FOR MY OWN PERSONAL NEEDS.

And you are. There is nothing saying that you can't do that. You just need to get the source code of the blob and then compile it to link and work with DMA-BUF. For your personal needs only, of course, not distribution.