[PATCH RFC] mm: do not start node_reclaim for page order > MAX_ORDER

[PATCH RFC] mm: do not start node_reclaim for page order > MAX_ORDER

[Konstantin Khlebnikov]

Page allocator has check in __alloc_pages_slowpath() but nowdays
there is earlier entry point into reclimer without such check:
get_page_from_freelist() -> node_reclaim().

Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
---
 mm/vmscan.c |    6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 62ac0c488624..52f672420f0b 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -4117,6 +4117,12 @@ int node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned int order)
 {
 	int ret;
 
+	/*
+	 * Do not scan if allocation will never succeed.
+	 */
+	if (order >= MAX_ORDER)
+		return NODE_RECLAIM_NOSCAN;
+
 	/*
 	 * Node reclaim reclaims unmapped file backed pages and
 	 * slab pages if we are over the defined limits.

Re: [PATCH RFC] mm: do not start node_reclaim for page order > MAX_ORDER

[Michal Hocko]

On Thu 01-11-18 20:37:52, Konstantin Khlebnikov wrote:
> Page allocator has check in __alloc_pages_slowpath() but nowdays
> there is earlier entry point into reclimer without such check:
> get_page_from_freelist() -> node_reclaim().

Is the order check so expensive that it would be visible in the fast
path? Spreading these MAX_ORDER checks sounds quite fragile to me.

> Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
> ---
>  mm/vmscan.c |    6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 62ac0c488624..52f672420f0b 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -4117,6 +4117,12 @@ int node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned int order)
>  {
>  	int ret;
>  
> +	/*
> +	 * Do not scan if allocation will never succeed.
> +	 */
> +	if (order >= MAX_ORDER)
> +		return NODE_RECLAIM_NOSCAN;
> +
>  	/*
>  	 * Node reclaim reclaims unmapped file backed pages and
>  	 * slab pages if we are over the defined limits.
> 

-- 
Michal Hocko
SUSE Labs