[v6,20/26] mm/mlock: Skip ZONE_DEVICE PMDs during mlock

Commit Message

Alistair Popple Jan. 10, 2025, 6 a.m. UTC

At present mlock skips ptes mapping ZONE_DEVICE pages. A future change
to remove pmd_devmap will allow pmd_trans_huge_lock() to return
ZONE_DEVICE folios so make sure we continue to skip those.

Signed-off-by: Alistair Popple <[e-mail]>
Acked-by: David Hildenbrand <[e-mail]>
---
 mm/mlock.c | 2 ++
 1 file changed, 2 insertions(+)

Comments

Dan Williams Jan. 14, 2025, 2:42 a.m. UTC | #1

Alistair Popple wrote:
> At present mlock skips ptes mapping ZONE_DEVICE pages. A future change
> to remove pmd_devmap will allow pmd_trans_huge_lock() to return
> ZONE_DEVICE folios so make sure we continue to skip those.
> 
> Signed-off-by: Alistair Popple <[e-mail]>
> Acked-by: David Hildenbrand <[e-mail]>

This looks like a fix in that mlock_pte_range() *does* call mlock_folio() 
when pmd_trans_huge_lock() returns a non-NULL @ptl.

So it is not in preparation for a future change it is making the pte and
pmd cases behave the same to drop mlock requests.

The code change looks good, but do add a Fixes tag and reword the
changelog a bit before adding:

Reviewed-by: Dan Williams <[e-mail]>

Alistair Popple Jan. 17, 2025, 1:54 a.m. UTC | #2

On Mon, Jan 13, 2025 at 06:42:46PM -0800, Dan Williams wrote:
> Alistair Popple wrote:
> > At present mlock skips ptes mapping ZONE_DEVICE pages. A future change
> > to remove pmd_devmap will allow pmd_trans_huge_lock() to return
> > ZONE_DEVICE folios so make sure we continue to skip those.
> > 
> > Signed-off-by: Alistair Popple <[e-mail]>
> > Acked-by: David Hildenbrand <[e-mail]>
> 
> This looks like a fix in that mlock_pte_range() *does* call mlock_folio() 
> when pmd_trans_huge_lock() returns a non-NULL @ptl.
> 
> So it is not in preparation for a future change it is making the pte and
> pmd cases behave the same to drop mlock requests.
> 
> The code change looks good, but do add a Fixes tag and reword the
> changelog a bit before adding:

Yeah, that changelog is a bit whacked. In fact it's not a fix - because
mlock_fixup() (the only caller) already filters dax VMAs. So this is really
about fixing a possible future bug when we start having PMDs for other types of
ZONE_DEVICE pages (ie. private, coherent, etc).

So probably I should just roll this into "mm: Allow compound zone device pages".

> Reviewed-by: Dan Williams <[e-mail]>

Patch

diff --git a/mm/mlock.c b/mm/mlock.c
index cde076f..3cb72b5 100644
--- a/mm/mlock.c
+++ b/mm/mlock.c
@@ -368,6 +368,8 @@  static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
 		if (is_huge_zero_pmd(*pmd))
 			goto out;
 		folio = pmd_folio(*pmd);
+		if (folio_is_zone_device(folio))
+			goto out;
 		if (vma->vm_flags & VM_LOCKED)
 			mlock_folio(folio);
 		else