from what I can see the kernel developers don't like the particular means used in the patch, there's no fundamental objection to the idea of enabling device-to-device DMA as a performance optimisation.