[CRIU] [PATCH 2/2] restore: Restore as many pages at once as possible

[Christopher Covington]

Hi Pavel,

On 11/09/2015 07:46 AM, Pavel Emelyanov wrote:
> When the VMA being restored is not COW-ed we read pages from images
> one-by-one which results in suboptimal pages.img access. Fix this
> by reading as many pages from iamge at once as possible withing the
> active pagemap and VMA.
>
> Signed-off-by: Pavel Emelyanov <xemul at parallels.com>
> ---
>  cr-restore.c | 28 ++++++++++++++++++++++++----
>  1 file changed, 24 insertions(+), 4 deletions(-)
> 
> diff --git a/cr-restore.c b/cr-restore.c
> index 8770f57..e06570d 100644
> --- a/cr-restore.c
> +++ b/cr-restore.c
> @@ -468,8 +468,8 @@ static int restore_priv_vma_content(void)
>  				ret = pr.read_pages(&pr, va, 1, buf);
>  				if (ret < 0)
>  					goto err_read;
> -				va += PAGE_SIZE;
>  
> +				va += PAGE_SIZE;
>  				nr_compared++;
>  
>  				if (memcmp(p, buf, PAGE_SIZE) == 0) {
> @@ -477,15 +477,35 @@ static int restore_priv_vma_content(void)
>  					continue;
>  				}
>  
> +				nr_restored++;
>  				memcpy(p, buf, PAGE_SIZE);
>  			} else {
> -				ret = pr.read_pages(&pr, va, 1, p);
> +				int nr, j;
> +
> +				/*
> +				 * Try to read as many pages as possible at once.
> +				 *
> +				 * Within the current pagemap we still have
> +				 * nr_pages - i pages (not all, as we might have
> +				 * switched VMA above), within the current VMA
> +				 * we have at most (vma->end - current_addr) bytes.
> +				 */
> +
> +				nr = min(nr_pages - i, (vma->e->end - va) / PAGE_SIZE);

Using page_size() instead of PAGE_SIZE would be nice for AArch64 (and PowerPC
if they add 4K page support to CRIU). On AArch32 and x86, page_size() is
#define'd to PAGE_SIZE so you shouldn't see any slowdown there.

Thanks,
Christopher Covington

-- 
Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project