Jerome Glisse was talking at this week's XDC2013 conference about using the progress address space by GPU drivers. It's a rather technical presentation targeting Linux driver developers, but within the presentation he shared that a partnership has occurred between Red Hat -- where Glisse is employed -- and NVIDIA to work on a generic device agnostic API dealing with using the CPU address space on the GPU. They have a working prototype right now and hope to publish it soon for upstream inclusion after receiving comments from other upstream developers.
This new API will allow for graphics drivers to support texture uploading without involving a memory copy, next sparse texture extension loading from the disk, OpenGL using process addresses for seamless compute shader support.
Jerome's presentation PDF slides aren't the most detailed and no video recording is yet available, but hopefully the prototype code will be published soon so we can get a better look at what's going on.