Re: [PATCH v2] mm/damon/vaddr: attempt per-vma lock during page table walk

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] `damos_va_migrate_pmd_entry` and `damos_va_stat_pmd_entry` incorrectly advance iteration by the full folio size, skipping unrelated PTEs for mTHP.
--

commit 25c04a646a951de8c86c5c718af6cfc459b7b5e0
Author: Kefeng Wang <wangkefeng.wang@huawei.com>

mm/damon/vaddr: attempt per-vma lock during page table walk

Currently, DAMON virtual address operations use mmap_read_lock
during page table walks, which can cause unnecessary contention
under high concurrency.

> diff --git a/mm/damon/vaddr.c b/mm/damon/vaddr.c
> index 1b0ebe3b6951e..d271476035641 100644
> --- a/mm/damon/vaddr.c
> +++ b/mm/damon/vaddr.c
[ ... ]
> @@ -750,7 +773,6 @@ static unsigned long damos_va_migrate(struct damon_target *target,
>  	struct mm_walk_ops walk_ops = {
>  		.pmd_entry = damos_va_migrate_pmd_entry,
>  		.pte_entry = NULL,
> -		.walk_lock = PGWALK_RDLOCK,
>  	};

This isn't a regression from this patch, but there appears to be a pre-existing
issue in damos_va_migrate_pmd_entry() and damos_va_stat_pmd_entry() when
handling large folios.

In the PTE iteration loops, when a large folio is found, the code sets
nr = folio_nr_pages(folio) and updates pte += nr and addr += nr * PAGE_SIZE.

Could this skip unrelated PTEs if the iteration started on a tail page of this
folio (e.g., halfway into an mTHP)?

Advancing by the full folio size could overshoot the end of the folio and skip
unrelated PTEs mapped immediately after it within the same PMD.

Would it be better to subtract the offset within the folio to safely skip only
the remaining pages of the current folio?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260512151523.2092638-1-wangkefeng.wang@huawei.com?part=1