None of these are regressions afaict, so considering how late we are in the 4.15 process, I can see reasons to not take any of these. All of them are backporting candidates though, imo.
1: correct off-by-1 in number-of-IOMMUs check 2: leave FECTL write to vtd_resume() 3: re-order register restoring in vtd_resume() 4: restore flush hooks when disabling qinval Jan