On Wed 09-02-22 14:59:19, Hugh Dickins wrote: [...] > Ah, but you're worrying about any previously-mapped pages when > VM_LOCKONFAULT is established: those ones get done by mlock_pte_range() > in 07/13, same as when VM_LOCKED is established - I checked again, > VM_LOCKONFAULT is not set without VM_LOCKED. Right you are. mlock_fixup->mlock_vma_pages_range is the missing piece. This is called for both do_mlock and mlockall paths. So all good and thanks for the clarification! I will get back to this sometimes next week. -- Michal Hocko SUSE Labs
