Re: [PATCH mm-new v2 1/1] mm/khugepaged: guard is_zero_pfn() calls with pte_present()

[Lorenzo Stoakes <lorenzo.stoakes@oracle.com>]

On Fri, Oct 17, 2025 at 05:38:47PM +0800, Lance Yang wrote:
> From: Lance Yang <lance.yang@linux.dev>
>
> A non-present entry, like a swap PTE, contains completely different data
> (swap type and offset). pte_pfn() doesn't know this, so if we feed it a
> non-present entry, it will spit out a junk PFN.

It feels like this somewhat contradicts points I've made on the original series
re the is_swap_pte() stuff. Sigh.

I guess that's _such a mess_ it's hard to avoid though.

And I guess it's reasonable that !pte_present() means we can't expect a valid
PFN though.

>
> What if that junk PFN happens to match the zeropage's PFN by sheer
> chance? While really unlikely, this would be really bad if it did.
>
> So, let's fix this potential bug by ensuring all calls to is_zero_pfn()
> in khugepaged.c are properly guarded by a pte_present() check.
>
> Suggested-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>

Not sure I really suggested something that strictly contradicts points I
made... but I guess I did suggest guarding this stuff more carefully.

> Reviewed-by: Dev Jain <dev.jain@arm.com>
> Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
> Signed-off-by: Lance Yang <lance.yang@linux.dev>
> ---
> Applies against commit 0f22abd9096e in mm-new.
>
> v1 -> v2:
>  - Collect Reviewed-by from Dev, Wei and Baolin - thanks!
>  - Reduce a level of indentation (per Dev)
>  - https://lore.kernel.org/linux-mm/20251016033643.10848-1-lance.yang@linux.dev/
>
>  mm/khugepaged.c | 29 ++++++++++++++++-------------
>  1 file changed, 16 insertions(+), 13 deletions(-)
>
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index d635d821f611..648d9335de00 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -516,7 +516,7 @@ static void release_pte_pages(pte_t *pte, pte_t *_pte,
>  		pte_t pteval = ptep_get(_pte);
>  		unsigned long pfn;
>
> -		if (pte_none(pteval))
> +		if (!pte_present(pteval))
>  			continue;
>  		pfn = pte_pfn(pteval);
>  		if (is_zero_pfn(pfn))
> @@ -690,17 +690,18 @@ static void __collapse_huge_page_copy_succeeded(pte_t *pte,
>  	     address += nr_ptes * PAGE_SIZE) {
>  		nr_ptes = 1;
>  		pteval = ptep_get(_pte);
> -		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
> +		if (pte_none(pteval) ||
> +		    (pte_present(pteval) && is_zero_pfn(pte_pfn(pteval)))) {
>  			add_mm_counter(vma->vm_mm, MM_ANONPAGES, 1);
> -			if (is_zero_pfn(pte_pfn(pteval))) {
> -				/*
> -				 * ptl mostly unnecessary.
> -				 */
> -				spin_lock(ptl);
> -				ptep_clear(vma->vm_mm, address, _pte);
> -				spin_unlock(ptl);
> -				ksm_might_unmap_zero_page(vma->vm_mm, pteval);
> -			}
> +			if (pte_none(pteval))
> +				continue;

Yeah I'm not sure I really love this refactoring.

Can be:

		if (!is_swap_pte(pteval)) {
			add_mm_counter(vma->vm_mm, MM_ANONPAGES, 1);
			if (!is_zero_pfn(pte_pfn(pteval)))
				continue;

			...
		}

Doing pte_pfn() on a pte_none() PTE is fine.

Obviously as theree's a lot of hate for is_swap_pte() you could also do:

		if (pte_none(pteval) || pte_present(pteval)) {
			...
		}

Which literally open-codes !is_swap_pte().

At the same time, this makes very clear that PTE none is OK.

> +			/*
> +			 * ptl mostly unnecessary.
> +			 */
> +			spin_lock(ptl);
> +			ptep_clear(vma->vm_mm, address, _pte);
> +			spin_unlock(ptl);
> +			ksm_might_unmap_zero_page(vma->vm_mm, pteval);
>  		} else {
>  			struct page *src_page = pte_page(pteval);
>
> @@ -794,7 +795,8 @@ static int __collapse_huge_page_copy(pte_t *pte, struct folio *folio,
>  		unsigned long src_addr = address + i * PAGE_SIZE;
>  		struct page *src_page;
>
> -		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
> +		if (pte_none(pteval) ||
> +		    (pte_present(pteval) && is_zero_pfn(pte_pfn(pteval)))) {
>  			clear_user_highpage(page, src_addr);
>  			continue;
>  		}
> @@ -1294,7 +1296,8 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
>  				goto out_unmap;
>  			}
>  		}
> -		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
> +		if (pte_none(pteval) ||
> +		    (pte_present(pteval) && is_zero_pfn(pte_pfn(pteval)))) {
>  			++none_or_zero;
>  			if (!userfaultfd_armed(vma) &&
>  			    (!cc->is_khugepaged ||
> --
> 2.49.0
>

I mean all of this seems super gross anyway. We're constantly open-coding the
same check over and over again.

static inline bool pte_is_none_or_zero(pte_t pteval)
{
	if (is_swap_pte(pteval))
		return false;

	return is_zero_pfn(pte_pfn(pteval));
}

Put somewhere in a relevant header file.

Or again, if there's distaste at is_swap_pte(), and here maybe it's more valid
not to use it (given name of function).

static inline bool pte_is_none_or_zero(pte_t pteval)
{
	/* Non-present entries do not have a PFN to check. */
	if (!pte_present(pteval))
		return false;

	if (pte_none(pteval))
		return true;

	return is_zero_pfn(pte_pfn(pteval));
}

I think I'm going to do a series to addres the is_swap_pte() mess actually, as
this whole thing is very frustrating.

Thanks, Lorenzo