NVIDIA Still Working On A Generic Allocator - Has Working Open-Source Implementation

[arabek]

I appreciate there is community concern about NVIDIA's lack of upstream open source graphics card support generally, but ironically, NVIDIA *does* have an open source driver stack called nvgpu.

It happens to be a downstream vendor fork and not written against the kernel's DRM interfaces, but it's licensed under an open source license. Based on code commits, their internal team do write the driver stack against the (big) desktop GPU parts, but only provide support for Tegra integrated parts.

Still requires firmware blobs that can't be freely redistributed. Unless nvidia stepps up the game and provides a firmware that allows for power management/reclocking and under linux-firmware package compatbile license… it won't really matter. And i don't mean only for tegra hardware. That's the embedded stuff almost no desktop user cares about.

[emersion]

This article is pretty wrong. The "working open-source implementation" is not about the generic allocator at all.

I've written a wrap-up blog post about this part of XDC, which I hope can help people understand what's this year's talk was about: https://emersion.fr/blog/2019/xdc201...-up/#allocator