I can see the same behaviour on a Intel Xeon box. At some point I tried mainline build 4.15-rc9 and that behaves exactly the same. So it looks like an issue with the KPTI implementation which still exists upstream. I am planning to experiment with further 4.15-rc's and then try bisecting there.
I can see the same behaviour on a Intel Xeon box. At some point I tried mainline build 4.15-rc9 and that behaves exactly the same. So it looks like an issue with the KPTI implementation which still exists upstream. I am planning to experiment with further 4.15-rc's and then try bisecting there.