On Mon, Oct 12, 2015 at 10:54:16AM +0900, Minchan Kim wrote:
> Use is_zero_pfn on pteval only after pte_present check on pteval
> (It might be better idea to introduce is_zero_pte where checks
> pte_present first). Otherwise, it could work with swap or
> migration entry and if pte_pfn's result is equal to zero_pfn
> by chance, we lose user's data in __collapse_huge_page_copy.
> So if you're luck, the application is segfaulted and finally you
> could see below message when the application is exit.
>
> BUG: Bad rss-counter state mm:ffff88007f099300 idx:2 val:3
Did you acctually steped on the bug?
If yes it's subject for stable@, I think.
> Signed-off-by: Minchan Kim <minc...@kernel.org>
> ---
>
> I found this bug with MADV_FREE hard test. Sometime, I saw
> "Bad rss-counter" message with MM_SWAPENTS but it's really
> rare, once a day if I was luck or once in five days if I was
> unlucky so I am doing test still and just pass a few days but
> I hope it will fix the issue.
>
> mm/huge_memory.c | 12 +++++++++++-
> 1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 4b06b8db9df2..349590aa4533 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -2665,15 +2665,25 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
> for (_address = address, _pte = pte; _pte < pte+HPAGE_PMD_NR;
> _pte++, _address += PAGE_SIZE) {
> pte_t pteval = *_pte;
> - if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
> + if (pte_none(pteval)) {
In -mm tree we have is_swap_pte() check before this point in
khugepaged_scan_pmd()
Also, what about similar pattern in __collapse_huge_page_isolate() and
__collapse_huge_page_copy()? Shouldn't they be fixed as well?
> if (!userfaultfd_armed(vma) &&
> ++none_or_zero <= khugepaged_max_ptes_none)
> continue;
> else
> goto out_unmap;
> }
> +
> if (!pte_present(pteval))
> goto out_unmap;
> +
> + if (is_zero_pfn(pte_pfn(pteval))) {
> + if (!userfaultfd_armed(vma) &&
> + ++none_or_zero <= khugepaged_max_ptes_none)
> + continue;
> + else
> + goto out_unmap;
> + }
> +
> if (pte_write(pteval))
> writable = true;
>
> --
> 1.9.1
>
--
Kirill A. Shutemov
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
