Linux Developers Still Reject NVIDIA Using DMA-BUF

[droidhacker]

Linux has had these exact policies all this time and yet Android is the best-selling smartphone OS in the world and nexus 7 is bound to have made Android a good dent in the tablet space, so much for being 'encumbered with insanity'. If Google needs to provide such a 'method' for proprietary drivers then yes, they are fully capable of implementing this, it won't be a 'problem'. Your continued trolling of 'oh god, if Linux doesn't make it easy for proprietary drivers it will fail anytime now' is as hollow as it ever was (and yet you keep on repeating the same bs like a broken record, it's almost as if you had an agenda...).

Linux success is obviously tied to it's great out-of-the-box support for hardware in all forms which is exactly what their no-binary-drivers-in-kernel-space policy has brought. Again we have NVidia as the last big holdout, and it's discrete gpu heydays are quickly coming to an end on the desktop. Desktop users in general will look on discrete gpu's the way we nowadays snicker at the voodoo architecture which is what nvidia once replaced, as something arcane and obsolete.

Interesting that you bring up the old voodoo cards... because they were pretty similar to this dual-GPU stuff we're dealing with now. They were a high performance (for the time) 3D graphics accelerator (i.e., discrete GPU) hooked in to the existing 2D graphics card (i.e. IGP) by a special wire. The nice thing about voodoo, though, was that it didn't need special wacky software interfaces to tie the two pieces of hardware together -- they were literally WIRED together to work this way.

Maybe nvidia should consider this as a solution... shouldn't be any kind of licensing issue, since they own what is left of 3dfx....

[GreatEmerald]

This is very important, because the "viral" FUD bullshit is based on such false premises. GPL doesn't force you to do anything with your code. It only governs the redistribution of code based on GPL code. If you want to distribute versions of GPLed code, it has to be under the GPL.

Uhm, but that's a given. Does a license that governs what you can do with anything in private exist to begin with? I don't know any. As far as I know, you can reverse-engineer and decompile anything you wish, as long as it's private and not distributed, there are no problems. And if you don't distribute what you do, it is pretty much as good as not existing. Sure, you can ignore all licenses when making something to run on a private server, but that's pretty much as far as you can go with that. So there are no false premises here, it's already implied as such.

[droidhacker]

Doesn't bode well for Android, unless Google has their own method.

Of course iOS and WinRT won't be encumbered with insanity, so they won't have any problem implementing such a mechanism.

How do you figure?
Binary blobs are free to talk to each other if they wish. No special internal kernel structures are required.
OSS drivers are free to talk to each other with the special internal kernel structure.
Binary blobs and OSS drivers are free to talk to each other through interfaces implemented in the binary blob.

On top of that... how many Android devices are you aware of that have discrete GPUs?

You see, here is the thing about phones, tablets, and things of that nature.... the GPU is integrated into the SoC. That goes for TEGRA (nvidia), SNAPDRAGON (qualcomm), and MALI (arm). Seems that nvidia isn't so worried about this interface on their TEGRA SoC's, you know why? Its because ***THEY*** are the ones with the primary GPU on those units!!!!

This "optimus" nonsense is a temporary fad. Its because Intel is integrating their GPUs with their CPUs, and AMD is integrating their GPUs with THEIR CPUs. NVidia doesn't HAVE a CPU for the desktop/laptop segment of the market, so their product must be some kind of discrete unit in order for them to continue to HAVE a product.

Laptops with dual-AMD (IGP+GPU) only exist to compete with the PERCEPTUAL advantages of dual GPU systems. It sure SOUNDS cool to have a low power + high performance chips in one machine, but when it really comes down to it, it is FAR FAR better to take the scalable graphics approach. Eventually, its going to be GPU components integrated with the CPU in multiple segments that can be scaled and/or powered down independently of each other. That will go for server chips as well as laptop chips, because server chips are going to benefit from the parallel performance benefits of GPGPU and heterogeneous computing. At some point, you won't even be able to distinguish the "GPU" components from the "CPU". The whole concept of "GPU" will become obsolete, and we will be back entirely to SOFTWARE rendering on CPUs that are much better suited for this type of workload than what we've become used to.

[droidhacker]

No no no.

If you modify GPL software, you don't have to redistribute anything. But if you CHOOSE to redistribute, then the resulting code must also be GPL.

You can combine GPL software and proprietary software to your heart's content, as long as you don't distribute the result to others. GPL is a distribution license.


No, GPL is used to ensure that if you receive software, you keep the right to modify and redistribute it under GPL terms. It doesn't say anything about the case where you keep the code for yourself and don't pass it on.

This is very important, because the "viral" FUD bullshit is based on such false premises. GPL doesn't force you to do anything with your code. It only governs the redistribution of code based on GPL code. If you want to distribute versions of GPLed code, it has to be under the GPL.

I think he meant what you said, but was just a bit less than articulate about it.

[droidhacker]

lol... ok. you win teh internets for the day.

And you sir, have lost. You have taken the final recourse of the interwebz troll and are taking on a condescending tone. The reality of the situation is that you have thoroughly dug your hole and failed to provide any valid response to the overwhelming combined intellect you are opposing.

[droidhacker]

Uhm, but that's a given. Does a license that governs what you can do with anything in private exist to begin with? I don't know any. As far as I know, you can reverse-engineer and decompile anything you wish, as long as it's private and not distributed, there are no problems. And if you don't distribute what you do, it is pretty much as good as not existing. Sure, you can ignore all licenses when making something to run on a private server, but that's pretty much as far as you can go with that. So there are no false premises here, it's already implied as such.

Actually, the unfortunate thing is that in many places (such as the USA), you do NOT have those freedoms. Reverse engineering perhaps under certain blackbox scenarios, but certainly not decompilation or use of code that you have obtained unlawfully or under any form of restrictive agreement, open source or not, that instructs you otherwise. Doing so in private does not protect you from police raids, which can be performed under warrant granted for only the SUSPICION of unlawful activities.

[Ramiliez]

Some people here are saying that NVIDIA has right to do whatever they want with their IP but they dont let Linux devs to have the same priviledge

The very same people which are so protective about NVIDIAs IP rights are telling Linux devs that they must do whatever other company tells them to do even if it infringes on their own license

[artivision]

Its actually very simple: Some companies like Nvidia, they understand that Linux will dominate shortly, and they have decided to gain full access to all the goods without giving anything back. Thanks but no thanks. if you want all the GPL yours, then open your driver. That doesn't mean to open everything, just the driver and the things untillOGL3.3, then use OGL4-4.3 like closed extensions for the open driver. Those close extensions they will only have graphics extensions (compilers, programs for the synthesizer, FX), and no driver functionality like memory management, that must be open. Also this open (and if you want unified driver), you can use it for all operating systems.

[kgonzales]

Linux success is obviously tied to it's great out-of-the-box support for hardware in all forms which is exactly what their no-binary-drivers-in-kernel-space policy has brought.

Yeah... right.

Linux success is due to the fact that you can get it for free. Most users could give two shits that its "free as in speech". They only care when it is "free as in beer".

Most Linux users (many corporations included) couldn't give a damn that it has great OOTB support for hardware. They like not paying for software, and further like not wasting their time contributing anything back to the so-called community which continues to give them stuff for free without them spending one bit of time or money.

Sad but true.

[kgonzales]

Its actually very simple: Some companies like Nvidia, they understand that Linux will dominate shortly

If I had a dime for every silly Linux fan boi who said a line like this, I would be wealthy beyond my dreams.

Fortunately, since its not dominating shortly, there will be plenty of time for people to say these lines, and I can work out my deal for dimes.