[PATCHv2 1/2] zram: clear IDLE flag after recompression

Sergey Senozhatsky posted 2 patches 3 weeks, 6 days ago
[PATCHv2 1/2] zram: clear IDLE flag after recompression
Posted by Sergey Senozhatsky 3 weeks, 6 days ago
Recompression should clear ZRAM_IDLE flag on the entries
it has accessed, because otherwise some entries, specifically
those for which recompression has failed, become immediate
candidate entries for another post-processing (e.g. writeback).

Consider the following case:
- recompression marks entries IDLE every 4 hours and attempts
  to recompress them
- some entries are incompressible, so we keep them intact and
  hence preserve IDLE flag
- writeback marks entries IDLE every 8 hours and writebacks
  IDLE entries, however we have IDLE entries left from
  recompression, so writeback prematurely writebacks those
  entries.

The bug was reported by Shin Kawamura.

Fixes: 84b33bf78889 ("zram: introduce recompress sysfs knob")
Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
---
 drivers/block/zram/zram_drv.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index e6d12e81241d..a16dbffcdca3 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c
@@ -1864,6 +1864,13 @@ static int recompress_slot(struct zram *zram, u32 index, struct page *page,
 	if (ret)
 		return ret;
 
+	/*
+	 * We touched this entry so mark it as non-IDLE. This makes sure that
+	 * we don't preserve IDLE flag and don't incorrectly pick this entry
+	 * for different post-processing type (e.g. writeback).
+	 */
+	zram_clear_flag(zram, index, ZRAM_IDLE);
+
 	class_index_old = zs_lookup_class_index(zram->mem_pool, comp_len_old);
 	/*
 	 * Iterate the secondary comp algorithms list (in order of priority)
-- 
2.47.0.163.g1226f6d8fa-goog
Re: [PATCHv2 1/2] zram: clear IDLE flag after recompression
Posted by Brian Geffon 3 weeks, 6 days ago
On Tue, Oct 29, 2024 at 12:36:14AM +0900, Sergey Senozhatsky wrote:
> Recompression should clear ZRAM_IDLE flag on the entries
> it has accessed, because otherwise some entries, specifically
> those for which recompression has failed, become immediate
> candidate entries for another post-processing (e.g. writeback).
> 
> Consider the following case:
> - recompression marks entries IDLE every 4 hours and attempts
>   to recompress them
> - some entries are incompressible, so we keep them intact and
>   hence preserve IDLE flag
> - writeback marks entries IDLE every 8 hours and writebacks
>   IDLE entries, however we have IDLE entries left from
>   recompression, so writeback prematurely writebacks those
>   entries.
> 
> The bug was reported by Shin Kawamura.
Reported-by: Shin Kawamura <kawasin@google.com>

> 
> Fixes: 84b33bf78889 ("zram: introduce recompress sysfs knob")
> Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Acked-by: Brian Geffon <bgeffon@google.com>

> ---
>  drivers/block/zram/zram_drv.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
> index e6d12e81241d..a16dbffcdca3 100644
> --- a/drivers/block/zram/zram_drv.c
> +++ b/drivers/block/zram/zram_drv.c
> @@ -1864,6 +1864,13 @@ static int recompress_slot(struct zram *zram, u32 index, struct page *page,
>  	if (ret)
>  		return ret;
>  
> +	/*
> +	 * We touched this entry so mark it as non-IDLE. This makes sure that
> +	 * we don't preserve IDLE flag and don't incorrectly pick this entry
> +	 * for different post-processing type (e.g. writeback).
> +	 */
> +	zram_clear_flag(zram, index, ZRAM_IDLE);
> +
>  	class_index_old = zs_lookup_class_index(zram->mem_pool, comp_len_old);
>  	/*
>  	 * Iterate the secondary comp algorithms list (in order of priority)
> -- 
> 2.47.0.163.g1226f6d8fa-goog
>