Re: [PATCH] RFC: clear 1G pages with streaming stores on x86

From: Andrew Morton
Date: Tue Jul 24 2018 - 16:53:54 EST


On Tue, 24 Jul 2018 13:46:39 -0700 Cannon Matthews <cannonmatthews@xxxxxxxxxx> wrote:

> Reimplement clear_gigantic_page() to clear gigabytes pages using the
> non-temporal streaming store instructions that bypass the cache
> (movnti), since an entire 1GiB region will not fit in the cache anyway.
>
> ...
>
> Tested:
> Time to `mlock()` a 512GiB region on broadwell CPU
> AVG time (s) % imp. ms/page
> clear_page_erms 133.584 - 261
> clear_page_nt 34.154 74.43% 67

A gigantic improvement!

> --- a/arch/x86/include/asm/page_64.h
> +++ b/arch/x86/include/asm/page_64.h
> @@ -56,6 +56,9 @@ static inline void clear_page(void *page)
>
> void copy_page(void *to, void *from);
>
> +#define __HAVE_ARCH_CLEAR_GIGANTIC_PAGE
> +void __clear_page_nt(void *page, u64 page_size);

Nit: the modern way is

#ifndef __clear_page_nt
void __clear_page_nt(void *page, u64 page_size);
#define __clear_page_nt __clear_page_nt
#endif

Not sure why, really. I guess it avoids adding two symbols and
having to remember and maintain the relationship between them.

> --- /dev/null
> +++ b/arch/x86/lib/clear_gigantic_page.c
> @@ -0,0 +1,30 @@
> +#include <asm/page.h>
> +
> +#include <linux/kernel.h>
> +#include <linux/mm.h>
> +#include <linux/sched.h>
> +
> +#if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
> +#define PAGES_BETWEEN_RESCHED 64
> +void clear_gigantic_page(struct page *page,
> + unsigned long addr,
> + unsigned int pages_per_huge_page)
> +{
> + int i;
> + void *dest = page_to_virt(page);
> + int resched_count = 0;
> +
> + BUG_ON(pages_per_huge_page % PAGES_BETWEEN_RESCHED != 0);
> + BUG_ON(!dest);
> +
> + might_sleep();

cond_resched() already does might_sleep() - it doesn't seem needed here.

> + for (i = 0; i < pages_per_huge_page; i += PAGES_BETWEEN_RESCHED) {
> + __clear_page_nt(dest + (i * PAGE_SIZE),
> + PAGES_BETWEEN_RESCHED * PAGE_SIZE);
> + resched_count += cond_resched();
> + }
> + /* __clear_page_nt requrires and `sfence` barrier. */
> + wmb();
> + pr_debug("clear_gigantic_page: rescheduled %d times\n", resched_count);
> +}
> +#endif