Re: [PATCH] mm/huge_memory: fix override of entry in remove_migration_pmd

[Balbir Singh]

On 11/14/25 20:23, David Hildenbrand (Red Hat) wrote:
> On 14.11.25 02:21, Balbir Singh wrote:
>> Recent changes exposed a BUG in remove_migration_pmd() where the
>> migration entry was being overridden when the folio is device private.
>>
>> Use scope local entry for creating the device private pmde. Make the
>> pmde writable if the migration entry is writable by moving the check
>> is_migration_write() prior to creating the device private entry.
>>
>> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>> Cc: David Hildenbrand <david@xxxxxxxxxx>
>> Cc: Zi Yan <ziy@xxxxxxxxxx>
>> Cc: Joshua Hahn <joshua.hahnjy@xxxxxxxxx>
>> Cc: Rakie Kim <rakie.kim@xxxxxx>
>> Cc: Byungchul Park <byungchul@xxxxxx>
>> Cc: Gregory Price <gourry@xxxxxxxxxx>
>> Cc: Ying Huang <ying.huang@xxxxxxxxxxxxxxxxx>
>> Cc: Alistair Popple <apopple@xxxxxxxxxx>
>> Cc: Oscar Salvador <osalvador@xxxxxxx>
>> Cc: Lorenzo Stoakes <lorenzo.stoakes@xxxxxxxxxx>
>> Cc: Baolin Wang <baolin.wang@xxxxxxxxxxxxxxxxx>
>> Cc: "Liam R. Howlett" <Liam.Howlett@xxxxxxxxxx>
>> Cc: Nico Pache <npache@xxxxxxxxxx>
>> Cc: Ryan Roberts <ryan.roberts@xxxxxxx>
>> Cc: Dev Jain <dev.jain@xxxxxxx>
>> Cc: Barry Song <baohua@xxxxxxxxxx>
>> Cc: Lyude Paul <lyude@xxxxxxxxxx>
>> Cc: Danilo Krummrich <dakr@xxxxxxxxxx>
>> Cc: David Airlie <airlied@xxxxxxxxx>
>> Cc: Simona Vetter <simona@xxxxxxxx>
>> Cc: Ralph Campbell <rcampbell@xxxxxxxxxx>
>> Cc: Mika Penttilä <mpenttil@xxxxxxxxxx>
>> Cc: Matthew Brost <matthew.brost@xxxxxxxxx>
>> Cc: Francois Dugast <francois.dugast@xxxxxxxxx>
>>
>> Signed-off-by: Balbir Singh <balbirs@xxxxxxxxxx>
>> Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@xxxxxxxxxx>
>> ---
>>   mm/huge_memory.c | 7 +++++--
>>   1 file changed, 5 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index 9dda8c48daca..df93768a6e15 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -4698,7 +4698,12 @@ 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 (is_writable_migration_entry(entry))
>> +        pmde = pmd_mkwrite(pmde, vma);
>> +
>>       if (folio_is_device_private(folio)) {
>> +        swp_entry_t entry;
>> +
>>           if (pmd_write(pmde))
>>               entry = make_writable_device_private_entry(
>>                               page_to_pfn(new));
>> @@ -4710,8 +4715,6 @@ 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 (is_writable_migration_entry(entry))
>> -        pmde = pmd_mkwrite(pmde, vma);
>>       if (pmd_swp_uffd_wp(*pvmw->pmd))
>>           pmde = pmd_mkuffd_wp(pmde);
>>       if (!is_migration_entry_young(entry))
> 
> There are more problems here: you cannot call pmd_mksoft_dirty() etc on something that is not a present pmd! We have pmd_swp_mksoft_dirty() and friends for that.
> 
> So you'll have to completely split both paths.
> 

I'll follow the pattern of remove_migration_pte(), where all the device private handling
and that of pte_swp_soft_dirty and pte_swp_mkuffd_wp is done at the every end after
migration processing.

Thanks for the review! That might also take up any softleaf fix-up's required, I'll check

Balbir

Balbir