There's Hope For DMA-BUF With Non-GPL Drivers

There's some resurrected hope for the kernel symbols of the DMA-BUF buffer sharing mechanism to be not restricted to only GPL drivers, which started off as a request by NVIDIA. This could lead to better NVIDIA Optimus support under Linux, among other benefits.

DMA-BUF is the buffer sharing mechanism for the Linux kernel that was introduced in the 3.3 cycle for zero-copy sharing of buffers between kernel drivers. This work originally came out of the Linaro project since there's a need for such a mechanism by various ARM SoC driver developers. However, there's also benefits to this work when it comes to GPU hot-plugging, OpenCL / GPGPU computing, and multi-GPU configurations like NVIDIA Optimus.

To learn more about DMA-BUF for Linux, watch this DMA-BUF video from FOSDEM 2012 when Daniel Vetter of Intel was speaking about this infrastructure.

Back in January there was a request by NVIDIA that the DMA-BUF kernel symbols be not exported GPL-only, which would prevent them from taking advantage of this buffer sharing mechanism in their proprietary driver. This would inhibit them from being able to easily/cleanly share buffers between their binary driver and say the Intel open-source driver in Optimus cases where there's both Intel and NVIDIA graphics. Or for buffer-sharing between an open-source NVIDIA Tegra kernel driver and the binary driver with a GeForce GPU, a case that one of the requesting NVIDIA engineer mentioned.

The upstream open-source Linux kernel developers weren't really in favor of this change to allow non-GPL drivers access to DMA-BUF. However, one of the developers involved with DMA-BUF, Rob Clark of Texas Instruments, has more positive news to now share. Below is the message he posted to the kernel mailing list on Sunday concerning this DMA-BUF licensing.

We discussed this topic at the kernel-gfx mini-summit at ELC. Following the discussion, I agree that dma-buf infrastructure is intended as an interface between driver subsystems. And because (for now) all the other arm SoC gl(es) stacks unfortunately involve a closed src userspace, and since I consider userspace and kernel as tightly coupled in the realm of graphics stacks, I don't think we can really claim the moral high-ground here. So I can't object to use of EXPORT_SYMBOL() instead of EXPORT_SYMBOL_GPL().

That said, I expect the dma-buf infrastructure to still be evolving and it is the responsibility for out-of-tree users of the API (GPL or otherwise) to adapt as the dma-buf API changes. And there isn't much the in-tree drivers can do when it comes to debugging of buffer sharing issues with out-of-tree drivers (binary or otherwise), leaving the debugging responsibility to the owners of different out-of-tree drivers.

BR,
-R

This would be marked as a win for NVIDIA (and potentially AMD, among others) as well as Linux desktop users just looking for the best supported hardware experience, assuming this export symbol change goes through.