Re: [PATCH v2 01/14] mm: Convert pXd_devmap checks to vma_is_dax

On Tue, Jun 17, 2025 at 11:19:34AM +0200, David Hildenbrand wrote:

On 16.06.25 13:58, Alistair Popple wrote:

quoted

Currently dax is the only user of pmd and pud mapped ZONE_DEVICE
pages. Therefore page walkers that want to exclude DAX pages can check
pmd_devmap or pud_devmap. However soon dax will no longer set PFN_DEV,
meaning dax pages are mapped as normal pages.

Ensure page walkers that currently use pXd_devmap to skip DAX pages
continue to do so by adding explicit checks of the VMA instead.

quoted

Signed-off-by: Alistair Popple <apopple@nvidia.com>

Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Dan Williams <redacted>

---

Changes from v1:

  - Remove vma_is_dax() check from mm/userfaultfd.c as
    validate_move_areas() will already skip DAX VMA's on account of them
    not being anonymous.

This should be documented in the patch description above.

Ok.

quoted

---
  fs/userfaultfd.c | 2 +-
  mm/hmm.c         | 2 +-
  mm/userfaultfd.c | 6 ------
  3 files changed, 2 insertions(+), 8 deletions(-)

diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index ef054b3..a886750 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c

@@ -304,7 +304,7 @@ static inline bool userfaultfd_must_wait(struct userfaultfd_ctx *ctx,
  		goto out;
  	ret = false;
-	if (!pmd_present(_pmd) || pmd_devmap(_pmd))
+	if (!pmd_present(_pmd) || vma_is_dax(vmf->vma))
  		goto out;

VMA checks should be done before doing any page table walk.

Actually upon review I think this check was always redundant as well -
vma_can_userfault() checks limit userfaultfd to anon/hugetlb/shmem vma's. Boy we
sure have a lot of these "normal vma" checks around the place ... at least for
certain definitions of normal.

Anyway will remove this and add a note to the commit (and apologies for not
catching this last time as I think you may have ready mentioned this or at least
the general concept).

quoted

  	if (pmd_trans_huge(_pmd)) {

diff --git a/mm/hmm.c b/mm/hmm.c
index feac861..5311753 100644
--- a/mm/hmm.c
+++ b/mm/hmm.c

@@ -441,7 +441,7 @@ static int hmm_vma_walk_pud(pud_t *pudp, unsigned long start, unsigned long end,
  		return hmm_vma_walk_hole(start, end, -1, walk);
  	}
-	if (pud_leaf(pud) && pud_devmap(pud)) {
+	if (pud_leaf(pud) && vma_is_dax(walk->vma)) {
  		unsigned long i, npages, pfn;
  		unsigned int required_fault;
  		unsigned long *hmm_pfns;

Dito.

Actually I see little reason to restrict this only to DAX for HMM ... we don't
end up doing that for the equivalent PMD path so will drop this as well. Thanks!

quoted

diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index 58b3ad6..8395db2 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c

@@ -1818,12 +1818,6 @@ ssize_t move_pages(struct userfaultfd_ctx *ctx, unsigned long dst_start,
  		ptl = pmd_trans_huge_lock(src_pmd, src_vma);
  		if (ptl) {
-			if (pmd_devmap(*src_pmd)) {
-				spin_unlock(ptl);
-				err = -ENOENT;
-				break;
-			}
-
  			/* Check if we can move the pmd without splitting it. */
  			if (move_splits_huge_pmd(dst_addr, src_addr, src_start + len) ||
  			    !pmd_none(dst_pmdval)) {

-- 
Cheers,

David / dhildenb