Re: [PATCH v2 1/4] mm: Introduce vm_uffd_ops API

[Mike Rapoport <rppt@kernel.org>]

On Tue, Jul 01, 2025 at 10:04:28AM -0700, Suren Baghdasaryan wrote:
> On Mon, Jun 30, 2025 at 3:16 AM Lorenzo Stoakes
> <lorenzo.stoakes@oracle.com> wrote:
> >
> > It seems like we're assuming a _lot_ of mm understanding in the underlying
> > driver here.
> >
> > I'm not sure it's really normal to be handing around page table state and
> > folios etc. to a driver like this, this is really... worrying to me.
> >
> > This feels like you're trying to put mm functionality outside of mm?
> 
> To second that, two things stick out for me here:
> 1. uffd_copy and uffd_get_folio seem to be at different abstraction
> levels. uffd_copy is almost the entire copy operation for VM_SHARED
> VMAs while uffd_get_folio is a small part of the continue operation.
> 2. shmem_mfill_atomic_pte which becomes uffd_copy for shmem in the
> last patch is quite a complex function which itself calls some IMO
> pretty internal functions like mfill_atomic_install_pte(). Expecting
> modules to implement such functionality seems like a stretch to me but
> maybe this is for some specialized modules which are written by mm
> experts only?

Largely shmem_mfill_atomic_pte() differs from anonymous memory version
(mfill_atomic_pte_copy()) by the way the allocated folio is accounted and
whether it's added to the page cache. So instead of uffd_copy(...) we might
add

	int (*folio_alloc)(struct vm_area_struct *vma, unsigned long dst_addr);
	void (*folio_release)(struct vm_area_struct *vma, struct folio *folio);

and then use them in mfill_atomic_pte_copy():

diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index bc473ad21202..6bad0dd70d3d 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c
@@ -247,8 +247,11 @@ static int mfill_atomic_pte_copy(pmd_t *dst_pmd,
 
 	if (!*foliop) {
 		ret = -ENOMEM;
-		folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, dst_vma,
-					dst_addr);
+		if (uffd_ops(dst_vma) && uffd_ops(dst_vma)->folio_alloc)
+			folio = uffd_ops(dst_vma)->folio_alloc(dst_vma, dst_addr);
+		else
+			folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, dst_vma,
+						dst_addr);
 		if (!folio)
 			goto out;
 
@@ -307,6 +310,8 @@ static int mfill_atomic_pte_copy(pmd_t *dst_pmd,
 	return ret;
 out_release:
 	folio_put(folio);
+	if (uffd_ops(dst_vma) && uffd_ops(dst_vma)->folio_release)
+		uffd_ops(dst_vma)->folio_release(dst_vma, folio);
 	goto out;
 }
 

-- 
Sincerely yours,
Mike.