Well right now everyone who cares about latency (including me) is spending time integrating some kind of kernel bypass library into their application that's cus

Well right now everyone who cares about latency (including me) is spending time integrating some kind of kernel bypass library into their application that's custom and/or proprietary to a single card vendor and often pretty finicky to configure and integrate correctly. It's a massive waste of resources. If the kernel could figure out DRI for graphics then they can figure out universal passthrough networking. All the bypass libraries seem to use the same buffer pool + spinlock slot architecture already. And by your logic, how does anything ever get into the kernel? Nobody should ever contribute anything because "upstream".