On Fri, May 29, 2026 at 06:26:37PM +0100, Kiryl Shutsemau (Meta) wrote: > Three mm paths outside the fault handler gate on the uffd PTE bit > today: khugepaged (skip collapse on ranges carrying markers), rmap > (cap unmap batching), and GUP (force a fault through > gup_can_follow_protnone). Extend each to treat VM_UFFD_RWP the same > as VM_UFFD_WP; otherwise per-PTE RWP state is silently destroyed or > bypassed. > > khugepaged: try_collapse_pte_mapped_thp() and > file_backed_vma_is_retractable() already refuse to collapse or > retract page tables on ranges carrying the uffd PTE bit. Broaden the > VMA predicate from userfaultfd_wp() to userfaultfd_protected() so > VM_UFFD_RWP ranges get the same protection. hpage_collapse_scan_pmd() > needs no change — its existing pte_uffd() check already catches an > RWP PTE because it carries the uffd bit. > > rmap: folio_unmap_pte_batch() caps batching at 1 for VM_UFFD_RWP so > the restore path handles each PTE with its own marker. > > GUP: gup_can_follow_protnone() forces a fault on VM_UFFD_RWP VMAs > regardless of FOLL_HONOR_NUMA_FAULT. RWP uses protnone as an > access-tracking marker, not for NUMA hinting, so any GUP — read or > write — must go through the userfaultfd fault path. > > Signed-off-by: Kiryl Shutsemau <[email protected]> > Assisted-by: Claude:claude-opus-4-6 > Acked-by: Mike Rapoport (Microsoft) <[email protected]>
Nit below but LGTM, so: Reviewed-by: Lorenzo Stoakes <[email protected]> > --- > include/linux/mm.h | 16 +++++++++++++++- > mm/khugepaged.c | 18 +++++++++++------- > mm/rmap.c | 2 +- > 3 files changed, 27 insertions(+), 9 deletions(-) > > diff --git a/include/linux/mm.h b/include/linux/mm.h > index 3d4d5f9a6f1b..2b04f690b516 100644 > --- a/include/linux/mm.h > +++ b/include/linux/mm.h > @@ -4644,11 +4644,25 @@ static inline int vm_fault_to_errno(vm_fault_t > vm_fault, int foll_flags) > > /* > * Indicates whether GUP can follow a PROT_NONE mapped page, or whether > - * a (NUMA hinting) fault is required. > + * a (NUMA hinting or userfaultfd RWP) fault is required. > */ > static inline bool gup_can_follow_protnone(const struct vm_area_struct *vma, > unsigned int flags) > { > + /* > + * VM_UFFD_RWP uses protnone as an access-tracking marker, not for > + * NUMA hinting. GUP must always take a fault so the access is > + * delivered to userfaultfd, regardless of FOLL_HONOR_NUMA_FAULT. > + * > + * Only do so while the VMA is accessible. If it has been made > + * inaccessible (e.g. mprotect(PROT_NONE)), fall through to the guard > + * below: forcing a fault there would loop, as handle_mm_fault() makes > + * no progress on protnone in an inaccessible VMA, and the access is > + * denied regardless of RWP anyway. > + */ > + if ((vma->vm_flags & VM_UFFD_RWP) && vma_is_accessible(vma)) > + return false; Can be: if (vma_test_single_mask(vma, VMA_UFFD_RWP) && vma_is_accessible(vma)) return false; > + > /* > * If callers don't want to honor NUMA hinting faults, no need to > * determine if we would actually have to trigger a NUMA hinting fault. > diff --git a/mm/khugepaged.c b/mm/khugepaged.c > index afa218be15de..4f3fedcd75cf 100644 > --- a/mm/khugepaged.c > +++ b/mm/khugepaged.c > @@ -1895,8 +1895,11 @@ static enum scan_result > try_collapse_pte_mapped_thp(struct mm_struct *mm, unsign > if (!thp_vma_allowable_order(vma, vma->vm_flags, TVA_FORCED_COLLAPSE, > PMD_ORDER)) > return SCAN_VMA_CHECK; > > - /* Keep pmd pgtable for uffd-wp; see comment in retract_page_tables() */ > - if (userfaultfd_wp(vma)) > + /* > + * Keep pmd pgtable while the uffd bit is in use; see comment in > + * retract_page_tables(). > + */ > + if (userfaultfd_protected(vma)) > return SCAN_PTE_UFFD; > > folio = filemap_lock_folio(vma->vm_file->f_mapping, > @@ -2109,13 +2112,14 @@ static bool file_backed_vma_is_retractable(struct > vm_area_struct *vma) > return false; > > /* > - * When a vma is registered with uffd-wp, we cannot recycle > + * When a vma is registered with uffd-wp or RWP, we cannot recycle > * the page table because there may be pte markers installed. > - * Other vmas can still have the same file mapped hugely, but > - * skip this one: it will always be mapped in small page size > - * for uffd-wp registered ranges. > + * VM_UFFD_RWP ranges similarly rely on per-PTE uffd state > + * and cannot be recycled to a shared PMD. Other vmas can still > + * have the same file mapped hugely, but skip this one: it will > + * always be mapped in small page size for these registrations. > */ > - if (userfaultfd_wp(vma)) > + if (userfaultfd_protected(vma)) > return false; > > /* > diff --git a/mm/rmap.c b/mm/rmap.c > index 546bc1cf9391..9fb733489898 100644 > --- a/mm/rmap.c > +++ b/mm/rmap.c > @@ -1965,7 +1965,7 @@ static inline unsigned int folio_unmap_pte_batch(struct > folio *folio, > if (pte_unused(pte)) > return 1; > > - if (userfaultfd_wp(vma)) > + if (userfaultfd_protected(vma)) > return 1; > > /* > -- > 2.54.0 >
