Performance problems
Just over a year ago I implemented an optimization to the SPI core code in Linux that avoids some needless context switches to a worker thread in the main data path that most clients use. This was really nice, it was simple to do but saved a bunch of work