mm/huge_memory.c | 25 ++++++++++++++++++------- 1 file changed, 18 insertions(+), 7 deletions(-)
commit a6ca2ba46390 ("mm: replace pmd_to_swp_entry() with softleaf_from_pmd()")
does not work with device private THP entries. softleaf_is_migration_young()
asserts that the entry be a migration entry, but in the current code, the
entry might already be replaced by a device private entry by the time the
check is made. The issue exists with commit
7385dbdbf841 ("mm/rmap: extend rmap and migration support device-private entries")
Fix this by processing the migration entries prior to conversion to
device private if the folio is device private.
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@kernel.org>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lance Yang <lance.yang@linux.dev>
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
---
mm/huge_memory.c | 25 ++++++++++++++++++-------
1 file changed, 18 insertions(+), 7 deletions(-)
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 942bd8410c54..82b019205216 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -4939,6 +4939,7 @@ void remove_migration_pmd(struct page_vma_mapped_walk *pvmw, struct page *new)
unsigned long haddr = address & HPAGE_PMD_MASK;
pmd_t pmde;
softleaf_t entry;
+ bool old = false, dirty = false, migration_read_entry = false;
if (!(pvmw->pmd && !pvmw->pte))
return;
@@ -4947,6 +4948,19 @@ void remove_migration_pmd(struct page_vma_mapped_walk *pvmw, struct page *new)
folio_get(folio);
pmde = folio_mk_pmd(folio, READ_ONCE(vma->vm_page_prot));
+ if (!softleaf_is_migration_young(entry))
+ old = true;
+
+ /* NOTE: this may contain setting soft-dirty on some archs */
+ if (folio_test_dirty(folio) && softleaf_is_migration_dirty(entry))
+ dirty = true;
+
+ if (softleaf_is_migration_write(entry))
+ pmde = pmd_mkwrite(pmde, vma);
+
+ if (!softleaf_is_migration_read(entry))
+ migration_read_entry = true;
+
if (folio_is_device_private(folio)) {
if (pmd_write(pmde))
entry = make_writable_device_private_entry(
@@ -4959,20 +4973,17 @@ void remove_migration_pmd(struct page_vma_mapped_walk *pvmw, struct page *new)
if (pmd_swp_soft_dirty(*pvmw->pmd))
pmde = pmd_mksoft_dirty(pmde);
- if (softleaf_is_migration_write(entry))
- pmde = pmd_mkwrite(pmde, vma);
+ if (old)
+ pmde = pmd_mkold(pmde);
if (pmd_swp_uffd_wp(*pvmw->pmd))
pmde = pmd_mkuffd_wp(pmde);
- if (!softleaf_is_migration_young(entry))
- pmde = pmd_mkold(pmde);
- /* NOTE: this may contain setting soft-dirty on some archs */
- if (folio_test_dirty(folio) && softleaf_is_migration_dirty(entry))
+ if (dirty)
pmde = pmd_mkdirty(pmde);
if (folio_test_anon(folio)) {
rmap_t rmap_flags = RMAP_NONE;
- if (!softleaf_is_migration_read(entry))
+ if (migration_read_entry)
rmap_flags |= RMAP_EXCLUSIVE;
folio_add_anon_rmap_pmd(folio, new, vma, haddr, rmap_flags);
--
2.51.1On Wed, Nov 12, 2025 at 03:46:34PM +1100, Balbir Singh wrote:
> commit a6ca2ba46390 ("mm: replace pmd_to_swp_entry() with softleaf_from_pmd()")
> does not work with device private THP entries. softleaf_is_migration_young()
> asserts that the entry be a migration entry, but in the current code, the
> entry might already be replaced by a device private entry by the time the
> check is made. The issue exists with commit
> 7385dbdbf841 ("mm/rmap: extend rmap and migration support device-private entries")
OK this is _hugely_ confusing.
Is the bug in my patch or in yours?
Why are you replying to your own series with this patch?
You shouldn't reference non-upstream commit messages in general.
If the bug is in 7385dbdbf841, fix it in your series, then perhaps send a
suggested fix-patch to the appropriate patch in my series to make life easier
for Andrew.
As mine I think in this case was purely a mechanical replacement of function
calls I'm guessing it's a bug in yours? So I think this is probably the best
way.
>
> Fix this by processing the migration entries prior to conversion to
> device private if the folio is device private.
>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: David Hildenbrand <david@kernel.org>
> Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
> Cc: Zi Yan <ziy@nvidia.com>
> Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
> Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
> Cc: Nico Pache <npache@redhat.com>
> Cc: Ryan Roberts <ryan.roberts@arm.com>
> Cc: Dev Jain <dev.jain@arm.com>
> Cc: Barry Song <baohua@kernel.org>
> Cc: Lance Yang <lance.yang@linux.dev>
>
> Signed-off-by: Balbir Singh <balbirs@nvidia.com>
> ---
> mm/huge_memory.c | 25 ++++++++++++++++++-------
> 1 file changed, 18 insertions(+), 7 deletions(-)
>
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 942bd8410c54..82b019205216 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -4939,6 +4939,7 @@ void remove_migration_pmd(struct page_vma_mapped_walk *pvmw, struct page *new)
> unsigned long haddr = address & HPAGE_PMD_MASK;
> pmd_t pmde;
> softleaf_t entry;
> + bool old = false, dirty = false, migration_read_entry = false;
>
> if (!(pvmw->pmd && !pvmw->pte))
> return;
> @@ -4947,6 +4948,19 @@ void remove_migration_pmd(struct page_vma_mapped_walk *pvmw, struct page *new)
> folio_get(folio);
> pmde = folio_mk_pmd(folio, READ_ONCE(vma->vm_page_prot));
>
> + if (!softleaf_is_migration_young(entry))
> + old = true;
> +
> + /* NOTE: this may contain setting soft-dirty on some archs */
'This may contain setting soft-dirty' is confusing. 'This may set soft-dirty on some arches' perhaps?
> + if (folio_test_dirty(folio) && softleaf_is_migration_dirty(entry))
> + dirty = true;
> +
> + if (softleaf_is_migration_write(entry))
> + pmde = pmd_mkwrite(pmde, vma);
> +
> + if (!softleaf_is_migration_read(entry))
> + migration_read_entry = true;
> +
> if (folio_is_device_private(folio)) {
> if (pmd_write(pmde))
> entry = make_writable_device_private_entry(
> @@ -4959,20 +4973,17 @@ void remove_migration_pmd(struct page_vma_mapped_walk *pvmw, struct page *new)
>
> if (pmd_swp_soft_dirty(*pvmw->pmd))
> pmde = pmd_mksoft_dirty(pmde);
> - if (softleaf_is_migration_write(entry))
> - pmde = pmd_mkwrite(pmde, vma);
> + if (old)
> + pmde = pmd_mkold(pmde);
> if (pmd_swp_uffd_wp(*pvmw->pmd))
> pmde = pmd_mkuffd_wp(pmde);
> - if (!softleaf_is_migration_young(entry))
> - pmde = pmd_mkold(pmde);
> - /* NOTE: this may contain setting soft-dirty on some archs */
> - if (folio_test_dirty(folio) && softleaf_is_migration_dirty(entry))
> + if (dirty)
> pmde = pmd_mkdirty(pmde);
>
> if (folio_test_anon(folio)) {
> rmap_t rmap_flags = RMAP_NONE;
>
> - if (!softleaf_is_migration_read(entry))
> + if (migration_read_entry)
> rmap_flags |= RMAP_EXCLUSIVE;
>
> folio_add_anon_rmap_pmd(folio, new, vma, haddr, rmap_flags);
> --
> 2.51.1
>
Thanks, Lorenzo
I noticed I did not respond to this <snip> >> + /* NOTE: this may contain setting soft-dirty on some archs */ > > 'This may contain setting soft-dirty' is confusing. 'This may set soft-dirty on some arches' perhaps? > This is the existing comment and it already says some archs. Am I missing something? Balbir
On 11/13/25 00:43, Lorenzo Stoakes wrote:
> On Wed, Nov 12, 2025 at 03:46:34PM +1100, Balbir Singh wrote:
>> commit a6ca2ba46390 ("mm: replace pmd_to_swp_entry() with softleaf_from_pmd()")
>> does not work with device private THP entries. softleaf_is_migration_young()
>> asserts that the entry be a migration entry, but in the current code, the
>> entry might already be replaced by a device private entry by the time the
>> check is made. The issue exists with commit
>> 7385dbdbf841 ("mm/rmap: extend rmap and migration support device-private entries")
>
> OK this is _hugely_ confusing.
>
> Is the bug in my patch or in yours?
>
The bug exists in my series (as pointed out in the the issue exists with),
but it is exposed by your changes with the VM_WARN_ON in your changes.
> Why are you replying to your own series with this patch?
>
> You shouldn't reference non-upstream commit messages in general.
>
> If the bug is in 7385dbdbf841, fix it in your series, then perhaps send a
> suggested fix-patch to the appropriate patch in my series to make life easier
> for Andrew.
>
OK, let me split it up then
> As mine I think in this case was purely a mechanical replacement of function
> calls I'm guessing it's a bug in yours? So I think this is probably the best
> way.
>
[...]
Balbir
On 12.11.25 05:46, Balbir Singh wrote:
> commit a6ca2ba46390 ("mm: replace pmd_to_swp_entry() with softleaf_from_pmd()")
So should this be squashed into Lorenzo patch, or incorporated in his
series in case he has to resend?
> does not work with device private THP entries. softleaf_is_migration_young()
> asserts that the entry be a migration entry, but in the current code, the
> entry might already be replaced by a device private entry by the time the
> check is made. The issue exists with commit
> 7385dbdbf841 ("mm/rmap: extend rmap and migration support device-private entries")
>
Because this confuses me. If it's already a problem in the
commit-to-go-upstream-first, it should be fixed in that commit?
--
Cheers
David
On 11/12/25 22:37, David Hildenbrand (Red Hat) wrote:
> On 12.11.25 05:46, Balbir Singh wrote:
>> commit a6ca2ba46390 ("mm: replace pmd_to_swp_entry() with softleaf_from_pmd()")
>
> So should this be squashed into Lorenzo patch, or incorporated in his series in case he has to resend?
>
>> does not work with device private THP entries. softleaf_is_migration_young()
>> asserts that the entry be a migration entry, but in the current code, the
>> entry might already be replaced by a device private entry by the time the
>> check is made. The issue exists with commit
>> 7385dbdbf841 ("mm/rmap: extend rmap and migration support device-private entries")
>>
>
> Because this confuses me. If it's already a problem in the commit-to-go-upstream-first, it should be fixed in that commit?
>
Not sure how to handle this, because that would break rebase of mm/mm-new
or I'd have to send a replacement patch for the original patch from Lorenzo
(which does not seem right).
I'll post a simpler patch, but it needs to be on top of the series
Balbir
On 13.11.25 06:03, Balbir Singh wrote:
> On 11/12/25 22:37, David Hildenbrand (Red Hat) wrote:
>> On 12.11.25 05:46, Balbir Singh wrote:
>>> commit a6ca2ba46390 ("mm: replace pmd_to_swp_entry() with softleaf_from_pmd()")
>>
>> So should this be squashed into Lorenzo patch, or incorporated in his series in case he has to resend?
>>
>>> does not work with device private THP entries. softleaf_is_migration_young()
>>> asserts that the entry be a migration entry, but in the current code, the
>>> entry might already be replaced by a device private entry by the time the
>>> check is made. The issue exists with commit
>>> 7385dbdbf841 ("mm/rmap: extend rmap and migration support device-private entries")
>>>
>>
>> Because this confuses me. If it's already a problem in the commit-to-go-upstream-first, it should be fixed in that commit?
>>
>
> Not sure how to handle this, because that would break rebase of mm/mm-new
> or I'd have to send a replacement patch for the original patch from Lorenzo
> (which does not seem right).
Yes, to be expected. Maybe Andrew can figure out how do address the
rebase, or we can give him a helping hand :)
>
> I'll post a simpler patch, but it needs to be on top of the series
Agreed, thanks.
--
Cheers
David
© 2016 - 2026 Red Hat, Inc.