Re: [PATCH] mm/khugepaged: Do not fail collapse_pte_mapped_thp() on SCAN_PMD_NULL

[Dev Jain <dev.jain@arm.com>]

On 12/09/25 10:28 pm, Kiryl Shutsemau wrote:
> From: Kiryl Shutsemau <kas@kernel.org>
>
> MADV_COLLAPSE on a file mapping behaves inconsistently depending on if
> PMD page table is installed or not.
>
> Consider following example:
>
> 	p = mmap(NULL, 2UL << 20, PROT_READ | PROT_WRITE,
> 		 MAP_SHARED, fd, 0);
> 	err = madvise(p, 2UL << 20, MADV_COLLAPSE);
>
> fd is a populated tmpfs file.
>
> The result depends on the address that the kernel returns on mmap().
> If it is located in an existing PMD table, the madvise() will succeed.
> However, if the table does not exist, it will fail with -EINVAL.
>
> This occurs because find_pmd_or_thp_or_none() returns SCAN_PMD_NULL when
> a page table is missing, which causes collapse_pte_mapped_thp() to fail.
>
> SCAN_PMD_NULL and SCAN_PMD_NONE should be treated the same in
> collapse_pte_mapped_thp(): install the PMD leaf entry and allocate page
> tables as needed.

Thanks.

Besides this patch, the label maybe_install_pmd is misleading -
SCAN_PMD_NONE means that the pmd table exists, just that the pmd
entry is none, so the pmd is already installed. Along with this,
the argument bool install_pmd should likewise be install_huge_pmd.

>
> Signed-off-by: Kiryl Shutsemau <kas@kernel.org>
> ---
>   mm/khugepaged.c | 25 ++++++++++++++++++++++++-
>   1 file changed, 24 insertions(+), 1 deletion(-)
>
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index b486c1d19b2d..9e76a4f46df9 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -1488,6 +1488,28 @@ static int set_huge_pmd(struct vm_area_struct *vma, unsigned long addr,
>   	return SCAN_SUCCEED;
>   }
>   
> +static int install_huge_pmd(struct vm_area_struct *vma, unsigned long haddr,
> +			    pmd_t *pmd, struct folio *folio)
> +{
> +	struct mm_struct *mm = vma->vm_mm;
> +	pgd_t *pgd;
> +	p4d_t *p4d;
> +	pud_t *pud;
> +
> +	pgd = pgd_offset(mm, haddr);
> +	p4d = p4d_alloc(mm, pgd, haddr);
> +	if (!p4d)
> +		return SCAN_FAIL;
> +	pud = pud_alloc(mm, p4d, haddr);
> +	if (!pud)
> +		return SCAN_FAIL;
> +	pmd = pmd_alloc(mm, pud, haddr);
> +	if (!pmd)
> +		return SCAN_FAIL;
> +
> +	return set_huge_pmd(vma, haddr, pmd, folio, &folio->page);
> +}
> +

For the SCAN_PMD_NONE case, we are unconditionally traversing the pagetables
now which is not needed. How about, in set_huge_pmd(), we pass a boolean install_pmd,
and at the start of the function, call install_pmd() which will do the traversal
and the pmd_alloc()? That will also make it crystal clear that in the SCAN_PMD_NULL
case, we are first installing the PMD table and then setting it to huge. Right now
the distinction between the two cases is not clear.
  

>   /**
>    * collapse_pte_mapped_thp - Try to collapse a pte-mapped THP for mm at
>    * address haddr.
> @@ -1556,6 +1578,7 @@ int collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr,
>   	switch (result) {
>   	case SCAN_SUCCEED:
>   		break;
> +	case SCAN_PMD_NULL:
>   	case SCAN_PMD_NONE:
>   		/*
>   		 * All pte entries have been removed and pmd cleared.
> @@ -1700,7 +1723,7 @@ int collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr,
>   maybe_install_pmd:
>   	/* step 5: install pmd entry */
>   	result = install_pmd
> -			? set_huge_pmd(vma, haddr, pmd, folio, &folio->page)
> +			? install_huge_pmd(vma, haddr, pmd, folio)
>   			: SCAN_SUCCEED;
>   	goto drop_folio;
>   abort: