[stefan_]

The whole "map the PCIE device into userspace process memory" thing is called DPDK (https://www.dpdk.org/)

[ay]

And you can combine the two:

https://fd.io/docs/vpp/master/whatisvpp/hoststack.html

And there is a sister project using this tech to get noticeable speed-ups:

https://wiki.fd.io/view/VSAP

Disclaimer: I am involved with the VPP project.