Re: [PATCH v6 02/12] mm: add PMD swap entry detection support

["David Hildenbrand (Arm)" <david@kernel.org>]

On 8/18/26 15:09, Usama Arif wrote:
> Currently when a PMD-mapped THP is swapped out, the PMD is always
> split into HPAGE_PMD_NR PTE-level swap entries. To preserve huge
> page information across swap cycles, later patches will install a
> single PMD-level swap entry instead. Add the infrastructure to detect
> those entries.
> 
> Teach the softleaf layer to recognise PMD swap entries:
> pmd_is_swap_entry() detects them and softleaf_is_valid_pmd_entry()
> accepts them as a valid non-present type. Because swap entries do not
> encode a PFN, make pmd_softleaf_to_folio() warn and return NULL for them
> instead of passing the swap offset to softleaf_to_folio(). Clear the
> exclusive overlay bit in softleaf_from_pmd() before decoding, matching
> how soft_dirty and uffd bits are already stripped.
> 
> Add pmd_swp_mkexclusive(), pmd_swp_exclusive(), and
> pmd_swp_clear_exclusive() helpers to each architecture that supports
> PMD softleaf entries (x86, arm64, s390, riscv, loongarch, powerpc),
> mirroring the existing PTE swap exclusive helpers in each arch's
> pgtable.h.

Ah, for migration entries we still use a dedicated migratetype. I actually have
on my todo list to move to PTE bits as well.

(likely the _swp_ part should then be renamed to indicate that this is for
softdirty entries, not just swap entries)

> Provide generic no-op PMD swap exclusive fallbacks for
> architectures without PMD softleaf support, matching the generic PMD
> swap soft-dirty fallbacks.

No softleaf implies to migration and no swap, so this would work.

You should extend mm/debug_vm_pgtable.c in a separate commit to test what
pte_swap_exclusive_tests() tests for PMDs.

> 
> Signed-off-by: Usama Arif <usama.arif@linux.dev>
> ---
>  arch/arm64/include/asm/pgtable.h             |  6 +++++
>  arch/loongarch/include/asm/pgtable.h         | 19 ++++++++++++++
>  arch/powerpc/include/asm/book3s/64/pgtable.h | 17 +++++++++++++
>  arch/riscv/include/asm/pgtable.h             | 15 +++++++++++
>  arch/s390/include/asm/pgtable.h              | 17 +++++++++++++
>  arch/x86/include/asm/pgtable.h               | 17 +++++++++++++

I'm sorry for asking you to compete with Kiryll's series by creating one patch
for each architecture that directly jumps at arch maintainers :)

[...]

>  #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
>  static inline pte_t pte_swp_mksoft_dirty(pte_t pte)
>  {
> diff --git a/include/linux/leafops.h b/include/linux/leafops.h
> index 7c13c58a5e218..4a6c52974b305 100644
> --- a/include/linux/leafops.h
> +++ b/include/linux/leafops.h
> @@ -102,6 +102,8 @@ static inline softleaf_t softleaf_from_pmd(pmd_t pmd)
>  		pmd = pmd_swp_clear_soft_dirty(pmd);
>  	if (pmd_swp_uffd(pmd))
>  		pmd = pmd_swp_clear_uffd(pmd);
> +	if (pmd_swp_exclusive(pmd))
> +		pmd = pmd_swp_clear_exclusive(pmd);

Can't we just unconditionally clear these flags?

pmd = pmd_swp_clear_uffd(pmd);
pmd = pmd_swp_clear_exclusive(pmd);

Avoids these rather unnecessary conditionals unless I am missing something.


>  	arch_entry = __pmd_to_swp_entry(pmd);
>  
>  	/* Temporary until swp_entry_t eliminated. */
> @@ -634,18 +636,30 @@ static inline bool pmd_is_migration_entry(pmd_t pmd)
>   */
>  static inline bool softleaf_is_valid_pmd_entry(softleaf_t entry)
>  {
> -	/* Only device private, migration entries valid for PMD. */
> +	/* Device private, migration, and swap entries valid for PMD. */

Can we just drop that comment? I mean, it's as clear as it gets in the code
immediately below :)

>  	return softleaf_is_device_private(entry) ||
> -		softleaf_is_migration(entry);
> +		softleaf_is_migration(entry) ||
> +		softleaf_is_swap(entry);
> +}
> +

-- 
Cheers,

David