From the project page: "The idea behind KGPU is to treat the GPU as a computing co-processor for the operating system, enabling data-parallel computation inside the Linux kernel. This allows us to use SIMD (or SIMT in CUDA) style code to accelerate Linux kernel functionality, and to bring new functionality formerly considered too compute intensive into the kernel. Simply put, KGPU enables vector computing for the kernel."
Additionally, "it makes the Linux kernel really parallelized: it is not only processing multiple requests concurrently, but can also partition a single large requested computation into tiles and spread them across the large number of cores on a GPU."
While it sounds like a novel concept, this is more of a research project at this point and there are a few factors that basically rule out this "KGPU" project from making much mainstream adoption for the foreseeable future. The big problem is that none of the open-source graphics drivers living within the Linux kernel DRM (Direct Rendering Manager) sub-system are yet capable of GPGPU support. There is support planned for an OpenCL Gallium3D state tracker with Clover, but that's far from being ready.
The other problem is that this current work just targets the Linux kernel on calling the GPU using CUDA. This work is sponsored by NVIDIA so the university went with using the Compute Unified Device Architecture, which is only supported on NVIDIA hardware, and when using their latest proprietary driver. A better choice would have been OpenCL, which can run on both AMD and NVIDIA GPUs and is an open industry standard.
At this point in the development of KGPU, they have a GPU-accelerated AES cipher for use by the eCryptfs encrypted file-system, which is showing off promising results considering how well modern GPUs can deal with cryptography.
Those interested in learning more about KGPU to augment the Linux kernel with the GPU, visit the Google Code page. Code is also available on GitHub. It would be interesting to see the Linux kernel begin to take advantage of the processing capabilities offered by modern GPUs, but first the open-source kernel drivers must be improved and be able to handle OpenCL and/or other GPGPU interfaces.