Re: [PATCH 1/2] vmalloc: Free pages as a batch

From: Michal Hocko
Date: Wed Sep 23 2020 - 04:35:43 EST


On Mon 21-09-20 23:46:27, Matthew Wilcox wrote:
> Use release_pages() to free the pages allocated by vmalloc(). This is
> slightly more efficient in terms of disabling and enabling IRQs once
> per batch instead of once per page.

Hmm, does this really lead to runtime improvements? Batching IRQ is
certainly nice but release_pages is much more heavy weight and all
additional checks are simply always false for vmalloc pages so all those
checks are pointless.

Maybe storing those pages into the linked list and use
free_unref_page_list instead would achieve what you want?

> Signed-off-by: Matthew Wilcox (Oracle) <willy@xxxxxxxxxxxxx>
> ---
> mm/vmalloc.c | 10 +---------
> 1 file changed, 1 insertion(+), 9 deletions(-)
>
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index be4724b916b3..3893fc8915c4 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -2263,16 +2263,8 @@ static void __vunmap(const void *addr, int deallocate_pages)
> vm_remove_mappings(area, deallocate_pages);
>
> if (deallocate_pages) {
> - int i;
> -
> - for (i = 0; i < area->nr_pages; i++) {
> - struct page *page = area->pages[i];
> -
> - BUG_ON(!page);
> - __free_pages(page, 0);
> - }
> + release_pages(area->pages, area->nr_pages);
> atomic_long_sub(area->nr_pages, &nr_vmalloc_pages);
> -
> kvfree(area->pages);
> }
>
> --
> 2.28.0

--
Michal Hocko
SUSE Labs