Ars Technica article

[Grzegorz Junka]

On 13/04/2020 21:46, Chris wrote:
>
>> Thankfully, Linux code can't enter FreeBSD base because of different 
>> licensing.
>>
>> I hope you don't advocate for re-implementing drivers for FreeBSD? 
>> What about NVidia drivers that aren't even distributed with source 
>> code? Surely, they provide binary packages for FreeBSD, but without 
>> support for Vulkan for example. Without any documentation of the 
>> hardware ports how this could be even achieved?
>>
>> Besides, I am happy when I see a vendor or developer supports Linux, 
>> apart from the usual Windows or Mac, not even mentioning FreeBSD. 
>> Many don't even differentiate beyond Linux, listing support for 
>> FreeBSD under Linux and other distro sections!
>>
>> I think we need to be pragmatic about it. Neither AMD nor NVidia will 
>> provide FreeBSD drivers and FreeBSD can't implement those drivers, 
>> either for objective reasons (man power) or because it's not feasible 
>> (NVidia). If we can reuse existing drivers that the manufacturers are 
>> releasing and updating frequently, and only focus on the 
>> compatibility layer, then why not?
> Agreed. The challenges are many. How? How have we managed so far? Yes.
> We've been (seemingly) forced to import video blobs for some (G|A)PUs. 
> But
> what of the others? What of the French driver project; Nouveau? How have
> they managed? Granted it *too* is Linux based. But they got there 
> *somehow*.
> Would it be a worthy venture to start a project like that that either
> imports && recobbles that project into "native" code? Or? I've been at 
> this
> for over 50yrs, and I've amassed more hardware than you could imagine. 
> Need
> to get FreeBSD on a Mac SE dsdd? NP. a PDP? Oh wait, that was done 
> l-o-n-g
> ago. But I've got several models available -- just in case.
> I'm serious. I'm willing to participate in whatever capacity I can. But
> IMHO importing Linux stuff isn't really "keeping ahead of the curve".
>

Are you trying to encourage developers to start a new project aimed at 
reimplementing major, sometimes closed, graphics drivers, for BSD, or 
even just FreeBSD? A project that would cost them lots of time and no 
pay but possibly a gratitude from happy users? If so, a comment to a 
post in a list group might not be sufficient. The idea would need to go 
out, developers evangelized, people convinced... As they say, the best 
leading is by example :-)

Jokes aside, you might be overestimating what Nouveau was able to 
achieve. First of all, they reverse-engineer. That means a working 
driver might only appear on the market after the hardware has been 
released and reverse-engineered. Sometimes long after, if at all.

Secondly, they only implement the minimum that's required to make them 
usable, which means no CUDA and no Vulkan support. Also look at this 
matrix <https://nouveau.freedesktop.org/wiki/FeatureMatrix/>, especially 
all the mostly and todos for the already quite old chipsets.

Lastly, Nouveau is mostly driven by RedHat's corporation money 
<https://www.phoronix.com/scan.php?page=news_item&px=Nouveau-2020-Early-Status>. 
A project of this size needs a commercial sponsor just to keep up with 
the pace of changes. Or a foundation. FreeBSD already sponsors some work 
in the graphics stack but surely that wouldn't be enough for 
re-implementing drivers, so a new one?

 From my point of view, accelerated graphics I mostly want to use on the 
desktop and I don't really care if it's coming from Linux or elsewhere 
as long as it properly supports all the latest features of my hardware 
so that my heard earned cash I spent on buying it is utilized. On the 
server I don't even use a graphic stack, I have IPMI or SSH, so none of 
the Linux parts affect it.

--
GrzegorzJ