NVIDIA Provides Open-Source Tegra K1 Support In Nouveau

[robclark]

I think reason doing open source tegra driver for nvidia, is getting licensing kepler/maxwell gpus more attractive for other arm soc manufacturers.

well, it probably has more to do with the fact that you can't ask end users to download and install the nv binary driver first thing when they get their shiny new tablet/phone, especially when there is a chicken/egg problem (ie. need GPU to get UI to come up at all). In case you were ever wondering why closed src kernel drivers is one problem we don't have with android.

[robclark]

PVR isn't closed.

That was a joke, right? Not that any of the mobile GPU vendors have opened up docs yet (or code.. but that is pretty unlikely). But at least nv tegra folks seem to be working well with upstream kernel so there is hope for both open and closed userspace on same kernel. (Which if nothing else, will make life easier for folks doing r/e and working on open src driver... having to reboot back to android to get cmdstream dumps from blob is a pita.) I'm hopefully that we can get to the same point with qcom (drm/msm).

I may be a bit jaded when it comes to pvr, but it is hard to find anyone who has worked with that driver who isn't.. I'm pretty sure at least two of IMGtech's big customers (from back when pvr actually mattered) have brought up the topic of open src with IMG. And I expect IMG was equally dismissive of open src community in both cases.

[mrugiero]

well, it probably has more to do with the fact that you can't ask end users to download and install the nv binary driver first thing when they get their shiny new tablet/phone, especially when there is a chicken/egg problem (ie. need GPU to get UI to come up at all). In case you were ever wondering why closed src kernel drivers is one problem we don't have with android.

Yes, but you don't tell them to install an OS either, so the OEM probably takes care of installing the drivers and locking the device. Also, closed source kernel drivers, sure, but we have closed source user space drivers in most of them, AFAIK.

[robclark]

Yes, but you don't tell them to install an OS either, so the OEM probably takes care of installing the drivers and locking the device. Also, closed source kernel drivers, sure, but we have closed source user space drivers in most of them, AFAIK.

oh, right, there are plenty of closed src userspace blobs on a modern phone/tablet. I was just pointing out that everyone is afraid of closed src kernel parts, because you can't rely on the normal desktop blob end-around of GPL (ie. it is end user combining the GPL kernel with closed src component).

[smitty3268]

AMD also had huge disclaimers about DPM possibly never happening. But it happened.

Yeah, as essentially the very last hardware feature exposed. I'll have hope nvidia might release those docs after they release 90% of the rest of their specs, and I don't see that ever happening, either. They're still working on the first 1% for now.

[chithanh]

I may be a bit jaded when it comes to pvr, but it is hard to find anyone who has worked with that driver who isn't.. I'm pretty sure at least two of IMGtech's big customers (from back when pvr actually mattered) have brought up the topic of open src with IMG. And I expect IMG was equally dismissive of open src community in both cases.

I don't think they were always or hard-line dismissive. From back when Jim Page and Kristof Beets worked on the PowerVR Kyro Linux drivers, IMGtec at least entertained the idea of having open source drivers (though ultimately the Kyro drivers were released as blobs).

It will be interesting to see whether NVidia is going to help the nouveau project in supporting Tegra K1 past the modesetting part.

[robclark]

I don't think they were always or hard-line dismissive. From back when Jim Page and Kristof Beets worked on the PowerVR Kyro Linux drivers, IMGtec at least entertained the idea of having open source drivers (though ultimately the Kyro drivers were released as blobs).

It will be interesting to see whether NVidia is going to help the nouveau project in supporting Tegra K1 past the modesetting part.

(just fwiw)

tegra k1 will still be using the tegra display controller, so I think they'll use nouveau in "render node" mode with prime / dmabuf+fence for synchronization between display and gpu.. so these patches are really all about the gpu side of things.

It could be that the modesetting help from nv for desktop parts is unrelated (or possibly just to have a good mood with the community before landing the tegra K1 stuff).. the patches do contain some initialization that is normally done by bios, which perhaps is interesting for the understanding of the desktop parts. (Otoh, it is all pretty hard-coded, ie. like 'nv_mask(priv, 0x000200, 0x20, 0x20);', so maybe doesn't help much.) But either way, I expect nv will continue the good upstream tegra work.