On Sat, Apr 16, 2022 at 11:05:49AM +0800, Miaohe Lin wrote:
> @@ -1797,6 +1797,17 @@ static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
>  		goto out;
>  	}
>  
> +	if (unlikely(!PageUptodate(page))) {
> +		pte_t pteval;
> +
> +		dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
> +		pteval = swp_entry_to_pte(make_swapin_error_entry(page));
> +		set_pte_at(vma->vm_mm, addr, pte, pteval);
> +		swap_free(entry);
> +		ret = 0;
> +		goto out;
> +	}
> +
>  	/* See do_swap_page() */
>  	BUG_ON(!PageAnon(page) && PageMappedToDisk(page));
>  	BUG_ON(PageAnon(page) && PageAnonExclusive(page));

Totally off-topic, but.. today when I was looking at the unuse path I just
found that the swp bits could have got lost for either soft-dirty and
uffd-wp here?  A quick patch attached.

Maybe at some point we should start to have some special helpers for
set_pte_at() when we're converting between present/non-present ptes, so as
to make sure all these will always be taken care of properly.

-- 
Peter Xu