[PATCH v3] staging: android: lowmemorykiller: neglect swap cached pages in other_file

[PATCH v3] staging: android: lowmemorykiller: neglect swap cached pages in other_file

[vinayakm.list]

From: Vinayak Menon <vinayakm.list@gmail.com>

With ZRAM enabled it is observed that lowmemory killer
doesn't trigger properly. swap cached pages are
accounted in NR_FILE, and lowmemorykiller considers
this as reclaimable and adds to other_file. But these
pages can't be reclaimed unless lowmemorykiller triggers.
So subtract swap pages from other_file.

Signed-off-by: Vinayak Menon <vinayakm.list@gmail.com>
---
 drivers/staging/android/lowmemorykiller.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/staging/android/lowmemorykiller.c b/drivers/staging/android/lowmemorykiller.c
index 6f094b3..9e931b2 100644
--- a/drivers/staging/android/lowmemorykiller.c
+++ b/drivers/staging/android/lowmemorykiller.c
@@ -88,7 +88,8 @@ static unsigned long lowmem_scan(struct shrinker *s, struct shrink_control *sc)
 	int array_size = ARRAY_SIZE(lowmem_adj);
 	int other_free = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
 	int other_file = global_page_state(NR_FILE_PAGES) -
-						global_page_state(NR_SHMEM);
+						global_page_state(NR_SHMEM) -
+						total_swapcache_pages();
 
 	if (lowmem_adj_size < array_size)
 		array_size = lowmem_adj_size;
-- 
1.7.10.4

Re: [PATCH v3] staging: android: lowmemorykiller: neglect swap cached pages in other_file

[Dave Hansen]

On 02/26/2014 11:06 AM, vinayakm.list@gmail.com wrote:
> With ZRAM enabled it is observed that lowmemory killer
> doesn't trigger properly. swap cached pages are
> accounted in NR_FILE, and lowmemorykiller considers
> this as reclaimable and adds to other_file. But these
> pages can't be reclaimed unless lowmemorykiller triggers.
> So subtract swap pages from other_file.
...
> --- a/drivers/staging/android/lowmemorykiller.c
> +++ b/drivers/staging/android/lowmemorykiller.c
> @@ -88,7 +88,8 @@ static unsigned long lowmem_scan(struct shrinker *s, struct shrink_control *sc)
>  	int array_size = ARRAY_SIZE(lowmem_adj);
>  	int other_free = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
>  	int other_file = global_page_state(NR_FILE_PAGES) -
> -						global_page_state(NR_SHMEM);
> +						global_page_state(NR_SHMEM) -
> +						total_swapcache_pages();

Sorry to dredge up an ancient thread.  But, I'm not sure this is a good
commit

The swap cache is there for pages which are still in RAM but are also in
some stage of being written to disk.  In this case, the "disk" in
question is ZRAM.

So, pages in the swap cache are probably "present" simultaneously in the
swap cache (the original copy) *and* in ZRAM (the compressed copy).
Swap cache pages *do* become reclaimable once their write to the backing
store is complete (just like any other dirty data).

I assume that this patch did *some* good, but perhaps it was via another
mechanism that what was described in the commit message.  Or, perhaps
the fact that ZRAM consumes RAM has thrown the lowmemorykiller off
somehow, and this helps to compensate for that effect.

Vinayak, how have you seen this patch help?