[PATCH] PM: hibernate: swap: defer linking the next map page

[PATCH] PM: hibernate: swap: defer linking the next map page

[Haesung Kim]

Delay allocating and linking the next swap_map_page until another image
page actually needs to be recorded.

The previous code linked and wrote a new swap map page as soon as the
current one became full. When the image size was an exact multiple of
MAP_PAGE_ENTRIES, that left an empty final map page that existed only to
terminate the on-disk chain.

Instead, keep a full map page in memory and only allocate the next map page
when the next image page arrives. This preserves the resume chain while
avoiding an unnecessary swap slot allocation and write for the empty
swap map page.

Signed-off-by: Haesung Kim <mattkim513@gmail.com>
---
 kernel/power/swap.c | 30 ++++++++++++++++++++----------
 1 file changed, 20 insertions(+), 10 deletions(-)

diff --git a/kernel/power/swap.c b/kernel/power/swap.c
index c626e9dc3c1c..e5cffb399e3a 100644
--- a/kernel/power/swap.c
+++ b/kernel/power/swap.c
@@ -430,19 +430,22 @@ static int swap_write_page(struct swap_map_handle *handle, void *buf,
 
 	if (!handle->cur)
 		return -EINVAL;
-	offset = alloc_swapdev_block(root_swap);
-	error = write_page(buf, offset, hb);
-	if (error)
-		return error;
-	handle->cur->entries[handle->k++] = offset;
+
+	/*
+	 * If the current map page is full, allocate and link next one first.
+	 * Delaying this until here avoids writing an empty swap map page when
+	 * the image size is an exact MAP_PAGE_ENTRIES multiple.
+	 */
 	if (handle->k >= MAP_PAGE_ENTRIES) {
 		offset = alloc_swapdev_block(root_swap);
 		if (!offset)
 			return -ENOSPC;
+
 		handle->cur->next_swap = offset;
 		error = write_page(handle->cur, handle->cur_swap, hb);
 		if (error)
-			goto out;
+			return error;
+
 		clear_page(handle->cur);
 		handle->cur_swap = offset;
 		handle->k = 0;
@@ -450,7 +453,7 @@ static int swap_write_page(struct swap_map_handle *handle, void *buf,
 		if (hb && low_free_pages() <= handle->reqd_free_pages) {
 			error = hib_wait_io(hb);
 			if (error)
-				goto out;
+				return error;
 			/*
 			 * Recalculate the number of required free pages, to
 			 * make sure we never take more than half.
@@ -458,14 +461,21 @@ static int swap_write_page(struct swap_map_handle *handle, void *buf,
 			handle->reqd_free_pages = reqd_free_pages();
 		}
 	}
- out:
-	return error;
+
+	offset = alloc_swapdev_block(root_swap);
+	error = write_page(buf, offset, hb);
+	if (error)
+		return error;
+	handle->cur->entries[handle->k++] = offset;
+	return 0;
 }
 
 static int flush_swap_writer(struct swap_map_handle *handle)
 {
-	if (handle->cur && handle->cur_swap)
+	if (handle->cur && handle->cur_swap && handle->k)
 		return write_page(handle->cur, handle->cur_swap, NULL);
+	else if (handle->cur && handle->cur_swap)
+		return 0;
 	else
 		return -EINVAL;
 }
-- 
2.34.1

Re: [PATCH] PM: hibernate: swap: defer linking the next map page

[Rafael J. Wysocki (Intel)]

On Tue, Jul 14, 2026 at 9:26 AM Haesung Kim <mattkim513@gmail.com> wrote:
>
> Delay allocating and linking the next swap_map_page until another image
> page actually needs to be recorded.
>
> The previous code linked and wrote a new swap map page as soon as the
> current one became full. When the image size was an exact multiple of
> MAP_PAGE_ENTRIES, that left an empty final map page that existed only to
> terminate the on-disk chain.
>
> Instead, keep a full map page in memory and only allocate the next map page
> when the next image page arrives. This preserves the resume chain while
> avoiding an unnecessary swap slot allocation and write for the empty
> swap map page.
>
> Signed-off-by: Haesung Kim <mattkim513@gmail.com>
> ---
>  kernel/power/swap.c | 30 ++++++++++++++++++++----------
>  1 file changed, 20 insertions(+), 10 deletions(-)
>
> diff --git a/kernel/power/swap.c b/kernel/power/swap.c
> index c626e9dc3c1c..e5cffb399e3a 100644
> --- a/kernel/power/swap.c
> +++ b/kernel/power/swap.c
> @@ -430,19 +430,22 @@ static int swap_write_page(struct swap_map_handle *handle, void *buf,
>
>         if (!handle->cur)
>                 return -EINVAL;
> -       offset = alloc_swapdev_block(root_swap);
> -       error = write_page(buf, offset, hb);
> -       if (error)
> -               return error;
> -       handle->cur->entries[handle->k++] = offset;
> +
> +       /*
> +        * If the current map page is full, allocate and link next one first.
> +        * Delaying this until here avoids writing an empty swap map page when
> +        * the image size is an exact MAP_PAGE_ENTRIES multiple.
> +        */
>         if (handle->k >= MAP_PAGE_ENTRIES) {
>                 offset = alloc_swapdev_block(root_swap);
>                 if (!offset)
>                         return -ENOSPC;
> +
>                 handle->cur->next_swap = offset;
>                 error = write_page(handle->cur, handle->cur_swap, hb);
>                 if (error)
> -                       goto out;
> +                       return error;
> +
>                 clear_page(handle->cur);
>                 handle->cur_swap = offset;
>                 handle->k = 0;
> @@ -450,7 +453,7 @@ static int swap_write_page(struct swap_map_handle *handle, void *buf,
>                 if (hb && low_free_pages() <= handle->reqd_free_pages) {
>                         error = hib_wait_io(hb);
>                         if (error)
> -                               goto out;
> +                               return error;
>                         /*
>                          * Recalculate the number of required free pages, to
>                          * make sure we never take more than half.
> @@ -458,14 +461,21 @@ static int swap_write_page(struct swap_map_handle *handle, void *buf,
>                         handle->reqd_free_pages = reqd_free_pages();
>                 }
>         }
> - out:
> -       return error;
> +
> +       offset = alloc_swapdev_block(root_swap);
> +       error = write_page(buf, offset, hb);
> +       if (error)
> +               return error;
> +       handle->cur->entries[handle->k++] = offset;
> +       return 0;
>  }
>
>  static int flush_swap_writer(struct swap_map_handle *handle)
>  {
> -       if (handle->cur && handle->cur_swap)
> +       if (handle->cur && handle->cur_swap && handle->k)
>                 return write_page(handle->cur, handle->cur_swap, NULL);
> +       else if (handle->cur && handle->cur_swap)
> +               return 0;
>         else
>                 return -EINVAL;
>  }
> --

Applied as 7.3 material, thanks!