NVIDIA 304.37 Linux Driver Brings 41 Official Changes

[ninez]

I haven't followed it closely, has anything happened since this? It was undecided then.

Well, that was 6 months ago and kernel 3.5 was released july 21st - none of the code contains EXPORT_SYMBOLS rather than EXPORT_SYMBOLS_GPL, so I think the decision has already been made... In case you didn't know ~ any blob is going to require EXPORT_SYMBOLS (without _GPL on the end).

Someone else whom is 'in the know', reading this thread hopefully will comment. But l'm pretty sure Nvidia is NOT going to be able to use DMABUF... To be honest, even when reading that original dmabuf/phoronix article that you linked too, i was not very hopeful that the kernel developers were actually going to use EXPORT_SYMBOL / allow nvidia to use dmabuf ~ even if it would've maybe allowed them to support Optimus (which was, from the sound of it, one of their goals along with the tegra bit cited in the posting).

So for all those people bitching how nvidia won't support Optimus, they should be equally bitching how (certain) kernel developers don't actually want Optimus to work seamlessly on linux, either. ..or at the very least, they are making it harder for nvidia to do so. Now, i am sure that is because they are trying to force nvidia's hand - but it's a shame that the end-user is the one who is paying the price, IMO. thus i think RealNC hit the nail on the head;

NVidia requested permission to use DMABUFF in order to implement Optimus on Linux. The kernel devs said no. It's not one-sided.

[johnc]

So for all those people bitching how nvidia won't support Optimus, they should be equally bitching how (certain) kernel developers don't actually want Optimus to work seamlessly on linux, either. ..or at the very least, they are making it harder for nvidia to do so. Now, i am sure that is because they are trying to force nvidia's hand - but it's a shame that the end-user is the one who is paying the price, IMO. thus i think RealNC hit the nail on the head;

That's kind of the irony of Linus responding to somebody's question about Optimus by blaming it on NVIDIA. Despite all of Stallman's constant GNU blathering about how users should be free, the Linux desktop places some of the most ridiculous constraints on users that aren't found in other OSes.

Imagine if they couldn't use dma-buf on the embedded devices.

[ninez]

That's kind of the irony of Linus responding to somebody's question about Optimus by blaming it on NVIDIA. Despite all of Stallman's constant GNU blathering about how users should be free, the Linux desktop places some of the most ridiculous constraints on users that aren't found in other OSes.

yeah, i do see the irony - which is actually, at lleast part of the reason i wrote that. But i don't think i am quite as pessimistic as you. In some ways there are constraints, while in other ways i do have some freedom in terms of flexibility, access to source code, customization, etc. So it's not all bad news

Imagine if they couldn't use dma-buf on the embedded devices.

We don't have to imagine that. If this is in fact the case (that nvidia can't use DMABUF - which i think it is). Nvidia and no other closed-source drivers/blobs can use it ---> embedded devices tend to be using closed-source blobs.

[RealNC]

Not wanting Optimus on Linux could make sense if you consider that many of the more vocal opponents of NVidia using dmabuf, are actually Intel employees. Like Alan Cox, probably the most vocal of them who are against this.

So we are in a situation where the employers of a competitor company decide what you can and can't offer for a platform. Another company known to do these things is Microsoft.

[russofris]

That's kind of the irony of Linus responding to somebody's question about Optimus by blaming it on NVIDIA. Despite all of Stallman's constant GNU blathering about how users should be free, the Linux desktop places some of the most ridiculous constraints on users that aren't found in other OSes.

Imagine if they couldn't use dma-buf on the embedded devices.

irony? Linux isn't putting ridiculous contraints on users... Nvidia is. Puts head on desk and cries for those that don't "get it".

[RealNC]

irony? Linux isn't putting ridiculous contraints on users... Nvidia is. Puts head on desk and cries for those that don't "get it".

Linux actually does put constrains on users. NVidia just follows a tried model to do their support. An OS offers services to drivers, and then the hardware manufacturer writes drivers.

Linux is refusing to offer needed services to the hardware vendor. In other words, Linux says:

"Our licensing model failed; it prevents driver writers to offer good proprietary support. So fuck you NVidia."

Yeah, makes a lot of sense.

[johnc]

We don't have to imagine that. If this is in fact the case (that nvidia can't use DMABUF - which i think it is). Nvidia and no other closed-source drivers/blobs can use it ---> embedded devices tend to be using closed-source blobs.

Ahhh, I must have misinterpreted those comments from the TI guy.

Not wanting Optimus on Linux could make sense if you consider that many of the more vocal opponents of NVidia using dmabuf, are actually Intel employees. Like Alan Cox, probably the most vocal of them who are against this.

So we are in a situation where the employers of a competitor company decide what you can and can't offer for a platform. Another company known to do these things is Microsoft.

This is the political aspect that I was afraid to bring up. I think it's true of a lot of these projects and ideas that get pushed by certain (to be unnamed) entities. There's always the possibility of corporate motives.

[bridgman]

We don't have to imagine that. If this is in fact the case (that nvidia can't use DMABUF - which i think it is). Nvidia and no other closed-source drivers/blobs can use it ---> embedded devices tend to be using closed-source blobs.

AFAIK the most common embedded driver model is a GPL-compatible open source kernel driver plus binary userspace.

[ninez]

AFAIK the most common embedded driver model is a GPL-compatible open source kernel driver plus binary userspace.

oops, my bad/ i always forget that it is the user-space side of embedded driver model that is closed... ~ hence, why you usually see devs trying to implement the 3d stuff that is missing, when they are trying to make an OSS implementation for some ARM device's GFX hardware.

Apparently, i am not paying attention, tonight ~ thank you for the correction, Bridgman.

[pingufunkybeat]

Not wanting Optimus on Linux could make sense if you consider that many of the more vocal opponents of NVidia using dmabuf, are actually Intel employees. Like Alan Cox, probably the most vocal of them who are against this.

So we are in a situation where the employers of a competitor company decide what you can and can't offer for a platform. Another company known to do these things is Microsoft.

They can offer exactly the same thing Intel and AMD offer. They simply refuse to do it.

Oh, and Intel and AMD contribute to the kernel, Nvidia doesn't. But they have special requests.