I have brought up this issue on LKML, asking for the kernel community's feedback on whether this is best solved by, in some way, allowing spurious wake-ups: https://lkml.org/lkml/2014/11/27/472