Re: [PATCH v2 9/9] x86/clear_huge_page: make clear_contig_region() preemptible

[Peter Zijlstra]

On Wed, Aug 30, 2023 at 11:49:58AM -0700, Ankur Arora wrote:
> clear_contig_region() can be used for clearing  regions as large as a
> gigantic page. Allow preemption in irqentry_exit to make sure we don't
> hold on to the CPU for an arbitrarily long period.

It would be very nice to quantify the latency improvement here. At the
very least perhaps mention how long a 2M or 1G clear takes (on you
favourite machine) to illustrate why this is needed.

> Signed-off-by: Ankur Arora <ankur.a.arora@xxxxxxxxxx>
> ---
>  arch/x86/mm/hugetlbpage.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/arch/x86/mm/hugetlbpage.c b/arch/x86/mm/hugetlbpage.c
> index 0b9f7a6dad93..55d1d15ea618 100644
> --- a/arch/x86/mm/hugetlbpage.c
> +++ b/arch/x86/mm/hugetlbpage.c
> @@ -152,7 +152,12 @@ hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
>  #ifndef CONFIG_HIGHMEM
>  static void clear_contig_region(struct page *page, unsigned int npages)
>  {
> +	/*
> +	 * We might be clearing a large region. Allow rescheduling.
> +	 */
> +	allow_resched();
>  	clear_pages(page_address(page), npages);
> +	disallow_resched();
>  }

Why is this in clear_contig_region() and not clear_pages() ? Any other
clear_pages() user will have the same problems no?