On Fri 07-06-19 03:57:18, Hugh Dickins wrote: [...] > The addition of "THPeligible" without an "Anon" in its name was > unfortunate. I suppose we're two releases too late to change that. Well, I do not really see any reason why THPeligible should be Anon specific at all. Even if ... > Applying process (PR_SET_THP_DISABLE) and mm (MADV_*HUGEPAGE) > limitations to shared filesystem objects doesn't work all that well. ... this is what we are going with then it is really important to have a single place to query the eligibility IMHO. > I recommend that you continue to treat shmem objects separately from > anon memory, and just make the smaps "THPeligible" more often accurate. Agreed on this. -- Michal Hocko SUSE Labs
