On 9/11/25 22:42, David Hildenbrand wrote:
> On 08.09.25 02:04, Balbir Singh wrote:
>> Implement CPU fault handling for zone device THP entries through
>> do_huge_pmd_device_private(), enabling transparent migration of
>> device-private large pages back to system memory on CPU access.
>>
>> When the CPU accesses a zone device THP entry, the fault handler
>> calls the device driver's migrate_to_ram() callback to migrate
>> the entire large page back to system memory.
>>
>> Cc: Andrew Morton <a...@linux-foundation.org>
>> Cc: David Hildenbrand <da...@redhat.com>
>> Cc: Zi Yan <z...@nvidia.com>
>> Cc: Joshua Hahn <joshua.hah...@gmail.com>
>> Cc: Rakie Kim <rakie....@sk.com>
>> Cc: Byungchul Park <byungc...@sk.com>
>> Cc: Gregory Price <gou...@gourry.net>
>> Cc: Ying Huang <ying.hu...@linux.alibaba.com>
>> Cc: Alistair Popple <apop...@nvidia.com>
>> Cc: Oscar Salvador <osalva...@suse.de>
>> Cc: Lorenzo Stoakes <lorenzo.stoa...@oracle.com>
>> Cc: Baolin Wang <baolin.w...@linux.alibaba.com>
>> Cc: "Liam R. Howlett" <liam.howl...@oracle.com>
>> Cc: Nico Pache <npa...@redhat.com>
>> Cc: Ryan Roberts <ryan.robe...@arm.com>
>> Cc: Dev Jain <dev.j...@arm.com>
>> Cc: Barry Song <bao...@kernel.org>
>> Cc: Lyude Paul <ly...@redhat.com>
>> Cc: Danilo Krummrich <d...@kernel.org>
>> Cc: David Airlie <airl...@gmail.com>
>> Cc: Simona Vetter <sim...@ffwll.ch>
>> Cc: Ralph Campbell <rcampb...@nvidia.com>
>> Cc: Mika Penttilä <mpent...@redhat.com>
>> Cc: Matthew Brost <matthew.br...@intel.com>
>> Cc: Francois Dugast <francois.dug...@intel.com>
>>
>> Signed-off-by: Balbir Singh <balb...@nvidia.com>
>> ---
>> include/linux/huge_mm.h | 7 +++++++
>> mm/huge_memory.c | 36 ++++++++++++++++++++++++++++++++++++
>> mm/memory.c | 6 ++++--
>> 3 files changed, 47 insertions(+), 2 deletions(-)
>>
>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>> index 23f124493c47..2c6a0c3c862c 100644
>> --- a/include/linux/huge_mm.h
>> +++ b/include/linux/huge_mm.h
>> @@ -496,6 +496,8 @@ static inline bool folio_test_pmd_mappable(struct folio
>> *folio)
>> vm_fault_t do_huge_pmd_numa_page(struct vm_fault *vmf);
>> +vm_fault_t do_huge_pmd_device_private(struct vm_fault *vmf);
>> +
>> extern struct folio *huge_zero_folio;
>> extern unsigned long huge_zero_pfn;
>> @@ -675,6 +677,11 @@ static inline vm_fault_t do_huge_pmd_numa_page(struct
>> vm_fault *vmf)
>> return 0;
>> }
>> +static inline vm_fault_t do_huge_pmd_device_private(struct vm_fault *vmf)
>> +{
>
> Is this a VM_WARN_ON_ONCE() or similar? (Maybe BUILD_BUG is possible?)
>
>> + return 0;
>> +}
>> +
>> static inline bool is_huge_zero_folio(const struct folio *folio)
>> {
>> return false;
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index b720870c04b2..d634b2157a56 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -1287,6 +1287,42 @@ static vm_fault_t __do_huge_pmd_anonymous_page(struct
>> vm_fault *vmf)
>> }
>> +vm_fault_t do_huge_pmd_device_private(struct vm_fault *vmf)
>> +{
>> + struct vm_area_struct *vma = vmf->vma;
>> + vm_fault_t ret = 0;
>> + spinlock_t *ptl;
>> + swp_entry_t swp_entry;
>> + struct page *page;
>> +
>> + if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
>> + vma_end_read(vma);
>> + return VM_FAULT_RETRY;
>> + }
>> +
>> + ptl = pmd_lock(vma->vm_mm, vmf->pmd);
>> + if (unlikely(!pmd_same(*vmf->pmd, vmf->orig_pmd))) {
>> + spin_unlock(ptl);
>> + return 0;
>> + }
>> +
>> + swp_entry = pmd_to_swp_entry(vmf->orig_pmd);
>> + page = pfn_swap_entry_to_page(swp_entry);
>> + vmf->page = page;
>> + vmf->pte = NULL;
>> + if (trylock_page(vmf->page)) {
>> + get_page(page);
>> + spin_unlock(ptl);
>> + ret = page_pgmap(page)->ops->migrate_to_ram(vmf);
>> + unlock_page(vmf->page);
>> + put_page(page);
>> + } else {
>> + spin_unlock(ptl);
>> + }
>> +
>> + return ret;
>> +}
>> +
>> /*
>> * always: directly stall for all thp allocations
>> * defer: wake kswapd and fail if not immediately available
>> diff --git a/mm/memory.c b/mm/memory.c
>> index d9de6c056179..860665f4b692 100644
>> --- a/mm/memory.c
>> +++ b/mm/memory.c
>> @@ -6298,8 +6298,10 @@ static vm_fault_t __handle_mm_fault(struct
>> vm_area_struct *vma,
>> vmf.orig_pmd = pmdp_get_lockless(vmf.pmd);
>> if (unlikely(is_swap_pmd(vmf.orig_pmd))) {
>> - VM_BUG_ON(thp_migration_supported() &&
>> - !is_pmd_migration_entry(vmf.orig_pmd));
>> + if (is_device_private_entry(
>> + pmd_to_swp_entry(vmf.orig_pmd)))
>
> Single line please.
Ack
>
> But didn't we have a pmd helper for that?
>
This is a single if that handles is_swap_pmd() and then
is_device_private_entry() and is_pmd_migration_entry()
under that
>> + return do_huge_pmd_device_private(&vmf);
>> +
>> if (is_pmd_migration_entry(vmf.orig_pmd))
>> pmd_migration_entry_wait(mm, vmf.pmd);
>> return 0;
>
>
Thanks,
Balbir
