Nouveau Developers Remain Blocked By NVIDIA From Advancing Open-Source Driver

[chithanh]

I'd be curious to see how many of you would really care about open-source from the second it would threaten IP your economic well-being depends from. Then you would perhaps see the open-source die-hards as a pack of vultures trying to get for free what you worked your ass off to achieve.. All things are relative, like Einstein proved..

If anything, events in recent years where poopy proprietary software affected millions of users reinforced my view that people who demand FOSS and reject blobs are right.

• Proprietary software in cars was used to hide their actual environmental impact from regulators and owners.
• Proprietary software used in elections could be used to manipulate or at least undermine public trust in election results.
• Proprietary software running on some auxiliary processor in your computer allows the vendor, or anyone who coerces or sweet-talks them, and sometimes, anybody to secretly take complete control over your computer. Ironically, Intel's latest ME firmware derived from the open source MINIX operating system, after Intel met with resistance messing around with Linux to fit in such tiny platforms.

I'm (unpleasantly) surprised they started developing a separate driver when they already used nouveau for Tegra/Android. :/
https://www.phoronix.com/scan.php?pa...ixel-C-Default

I think Pixel C is a somewhat special situation. Reports indicate that the device was originally intended to run Chrome OS, and only later in development they switched to Android.

[RSpliet]

As you can see - it's all about taking and giving nothing in return. I am kind of fascinated that you guys still spend your (spare) time to improve the drivers for a company that show you this clearly how valuable you are for them...

I don't know about others, but I do this to build my knowledge and engineering skills while contributing to the ideal of "good open source drivers for everything". I do this for myself and for the users, not for NVIDIA.

[lepetit]

my future gpu, will be amd

[artivision]

Sometimes I miss the early days of the open source effort when everyone was saying that "thousands of community developers would write drivers that were much better than AMD's in a few months". These days it seems that even building and publishing packages from published open source driver code has become an "oh we can't do that only AMD can do it" thing.

Maybe that *is* Linux in the 21st century... I sure hope not though

That is because if one free developer design a good patch that you didn't think yourselves you will probably hire him (like Marek). So technically the community writes better drivers vs you. This includes people from other companies to that you cannot hire (like Valve or other game porting companies).

[aht0]

my future gpu, will be amd

Fine, until you would stick to Linux and only plan on gaming.
And are willing to give up OpenCL & etc.

Forget FreeBSD, Solaris support - never-ever seen FreeBSD official driver from AMD. Nvidia has always provided one

[karolherbst]

Most distributions aren't even bundling any recent Nouveau components. You can't even boot a Manjaro or Fedora install media with open source drivers on a Pascal GPU, they die with "Nouveau: unknown chipset" error. Nvidia probably isn't in a hurry, as distro maintainers don't seem to be interested in shipping even mode setting support, except for Ubuntu. Maybe when Debian decides to ship it sometime in 2037, Nvidia might think it's time to release firmwares, and let things progress.

well for modesetting you don't need signed firmwares, but we can't just turn it on for every not yet released hardware. There are usually changed in the display part of the GPU we jave to adjust. And we should also rather fail loading Nouveau than to run on hardware Nouveau was never tested on.

[karolherbst]

As you can see - it's all about taking and giving nothing in return. I am kind of fascinated that you guys still spend your (spare) time to improve the drivers for a company that show you this clearly how valuable you are for them...

well as RSpliet already said: we mainly do it for ourselves and the community. And it's fun, I wouldn't work on it if it wouldn't be fun.

[pcxmac]

it might be easier to move amd code over to things like free bsd once X11 starts declining steeply enough and there is enough abstraction overhead/well written interface code for the (a) kernel.

[misp]

I don't know about others, but I do this to build my knowledge and engineering skills while contributing to the ideal of "good open source drivers for everything". I do this for myself and for the users, not for NVIDIA.

Honest question here: why not do it for something where it makes more sense? E.g. Android universe doesn't have proper open drivers for graphics. There are some projects, but most of it is still using proprietary blobs afaik

Or someone mentioned in another thread that amd drivers are also reverse-engineered. At least these guys cooperate with the community, RE some part would bring their stack to a much better level then nouveau.

[imirkin]

Honest question here: why not do it for something where it makes more sense? E.g. Android universe doesn't have proper open drivers for graphics. There are some projects, but most of it is still using proprietary blobs afaik

Or someone mentioned in another thread that amd drivers are also reverse-engineered. At least these guys cooperate with the community, RE some part would bring their stack to a much better level then nouveau.

Can't speak for everyone, but I do tend to contribute to other drivers as well -- mostly freedreno (all Adreno generations), and a bit to swr. Maybe the odd i965/anv contribution. Not to mention all kinds of core improvements.

Of course eventually you build up an investment in a certain driver -- e.g. I have a dozen NVIDIA GPUs of various generations that I've bought off eBay (never new). And I know a *ton* about how NVIDIA stuff works across all generations (and it's still probably like 10% of what Ben knows, which in turn is 10% of the full knowledge one might get with docs), whereas it'd be a considerable investment to learn how a different GPU architecture worked.

Also it'd be no use contributing to something like the AMD RadeonSI driver - it's backed by full-time developers who are already familiar with the hardware, driver stack, etc., and have the hardware on hand. Not a lot of low-hanging fruit left there for a volunteer part-time contributor. (Not to mention that it uses LLVM for its compiler backend, which is an instant turn-off.)