Re: [PATCH v2 2/2] mm: Compute first_set_pte to eliminate evaluating redundant ranges

[Barry Song <baohua@kernel.org>]

On Mon, Sep 16, 2024 at 11:08 PM Dev Jain <dev.jain@arm.com> wrote:
>
> For an mTHP allocation, we need to check, for every order, whether for
> that order, we have enough number of contiguous PTEs empty. Instead of
> iterating the while loop for every order, use some information, which
> is the first set PTE found, from the previous iteration to eliminate
> some cases. The key to understanding the correctness of the patch
> is that the ranges we want to examine form a strictly decreasing
> sequence of nested intervals.

Could we include some benchmark data here, as suggested by Ryan in this thread?

https://lore.kernel.org/linux-mm/58f91a56-890a-45d0-8b1f-47c4c70c9600@arm.com/

>
> Suggested-by: Ryan Roberts <ryan.roberts@arm.com>
> Signed-off-by: Dev Jain <dev.jain@arm.com>

> ---
>  mm/memory.c | 20 ++++++++++++++++++--
>  1 file changed, 18 insertions(+), 2 deletions(-)
>
> diff --git a/mm/memory.c b/mm/memory.c
> index 8bb1236de93c..e81c6abe09ce 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -4633,10 +4633,11 @@ static struct folio *alloc_anon_folio(struct vm_fault *vmf)
>  {
>         struct vm_area_struct *vma = vmf->vma;
>  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +       pte_t *first_set_pte = NULL, *align_pte, *pte;
>         unsigned long orders;
>         struct folio *folio;
>         unsigned long addr;
> -       pte_t *pte;
> +       int max_empty;
>         gfp_t gfp;
>         int order;
>
> @@ -4671,8 +4672,23 @@ static struct folio *alloc_anon_folio(struct vm_fault *vmf)
>         order = highest_order(orders);
>         while (orders) {
>                 addr = ALIGN_DOWN(vmf->address, PAGE_SIZE << order);
> -               if (pte_range_none(pte + pte_index(addr), 1 << order) == 1 << order)
> +               align_pte = pte + pte_index(addr);
> +
> +               /* Range to be scanned known to be empty */
> +               if (align_pte + (1 << order) <= first_set_pte)
> +                       break;
> +
> +               /* Range to be scanned contains first_set_pte */
> +               if (align_pte <= first_set_pte)
> +                       goto repeat;
> +
> +               /* align_pte > first_set_pte, so need to check properly */
> +               max_empty = pte_range_none(align_pte, 1 << order);
> +               if (max_empty == 1 << order)
>                         break;
> +
> +               first_set_pte = align_pte + max_empty;
> +repeat:
>                 order = next_order(&orders, order);
>         }
>
> --
> 2.30.2
>

Thanks
barry