Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1435592 > unrolled thread

[PATCH 09/31] mm, vmscan: by default have direct reclaim only shrink once per node

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [PATCH 09/31] mm, vmscan: by default have direct reclaim only shrink once per node Mel Gorman <mgorman@techsingularity.net> - 2016-07-01 22:10 +0200
    Re: [PATCH 09/31] mm, vmscan: by default have direct reclaim only  shrink once per node Joonsoo Kim <iamjoonsoo.kim@lge.com> - 2016-07-07 03:50 +0200
      Re: [PATCH 09/31] mm, vmscan: by default have direct reclaim only  shrink once per node Mel Gorman <mgorman@techsingularity.net> - 2016-07-07 12:30 +0200

#1435592 — [PATCH 09/31] mm, vmscan: by default have direct reclaim only shrink once per node

Direct reclaim iterates over all zones in the zonelist and shrinking them
but this is in conflict with node-based reclaim.  In the default case,
only shrink once per node.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
---
 mm/vmscan.c | 19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index b524d3b72527..34656173a670 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -2552,14 +2552,6 @@ static inline bool compaction_ready(struct zone *zone, int order, int classzone_
  * try to reclaim pages from zones which will satisfy the caller's allocation
  * request.
  *
- * We reclaim from a zone even if that zone is over high_wmark_pages(zone).
- * Because:
- * a) The caller may be trying to free *extra* pages to satisfy a higher-order
- *    allocation or
- * b) The target zone may be at high_wmark_pages(zone) but the lower zones
- *    must go *over* high_wmark_pages(zone) to satisfy the `incremental min'
- *    zone defense algorithm.
- *
  * If a zone is deemed to be full of pinned pages then just give it a light
  * scan then give up on it.
  */
@@ -2571,6 +2563,7 @@ static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
 	unsigned long nr_soft_scanned;
 	gfp_t orig_mask;
 	enum zone_type classzone_idx;
+	pg_data_t *last_pgdat = NULL;
 
 	/*
 	 * If the number of buffer_heads in the machine exceeds the maximum
@@ -2600,6 +2593,16 @@ static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
 			classzone_idx--;
 
 		/*
+		 * Shrink each node in the zonelist once. If the zonelist is
+		 * ordered by zone (not the default) then a node may be
+		 * shrunk multiple times but in that case the user prefers
+		 * lower zones being preserved
+		 */
+		if (zone->zone_pgdat == last_pgdat)
+			continue;
+		last_pgdat = zone->zone_pgdat;
+
+		/*
 		 * Take care memory controller reclaiming has small influence
 		 * to global LRU.
 		 */
-- 
2.6.4

[toc] | [next] | [standalone]

#1438119 — Re: [PATCH 09/31] mm, vmscan: by default have direct reclaim only shrink once per node

On Fri, Jul 01, 2016 at 09:01:17PM +0100, Mel Gorman wrote:
> Direct reclaim iterates over all zones in the zonelist and shrinking them
> but this is in conflict with node-based reclaim.  In the default case,
> only shrink once per node.
> 
> Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
> Acked-by: Johannes Weiner <hannes@cmpxchg.org>
> Acked-by: Vlastimil Babka <vbabka@suse.cz>
> ---
>  mm/vmscan.c | 19 +++++++++++--------
>  1 file changed, 11 insertions(+), 8 deletions(-)
> 
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index b524d3b72527..34656173a670 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -2552,14 +2552,6 @@ static inline bool compaction_ready(struct zone *zone, int order, int classzone_
>   * try to reclaim pages from zones which will satisfy the caller's allocation
>   * request.
>   *
> - * We reclaim from a zone even if that zone is over high_wmark_pages(zone).
> - * Because:
> - * a) The caller may be trying to free *extra* pages to satisfy a higher-order
> - *    allocation or
> - * b) The target zone may be at high_wmark_pages(zone) but the lower zones
> - *    must go *over* high_wmark_pages(zone) to satisfy the `incremental min'
> - *    zone defense algorithm.
> - *
>   * If a zone is deemed to be full of pinned pages then just give it a light
>   * scan then give up on it.
>   */
> @@ -2571,6 +2563,7 @@ static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
>  	unsigned long nr_soft_scanned;
>  	gfp_t orig_mask;
>  	enum zone_type classzone_idx;
> +	pg_data_t *last_pgdat = NULL;
>  
>  	/*
>  	 * If the number of buffer_heads in the machine exceeds the maximum
> @@ -2600,6 +2593,16 @@ static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
>  			classzone_idx--;
>  
>  		/*
> +		 * Shrink each node in the zonelist once. If the zonelist is
> +		 * ordered by zone (not the default) then a node may be
> +		 * shrunk multiple times but in that case the user prefers
> +		 * lower zones being preserved
> +		 */
> +		if (zone->zone_pgdat == last_pgdat)
> +			continue;
> +		last_pgdat = zone->zone_pgdat;
> +
> +		/*

After this change, compaction_ready() which uses zone information
would be called with highest zone in node. So, if some lower zone in
that node is compaction-ready, we cannot stop the reclaim.

Thanks.

[toc] | [prev] | [next] | [standalone]

#1438505 — Re: [PATCH 09/31] mm, vmscan: by default have direct reclaim only shrink once per node

On Thu, Jul 07, 2016 at 10:43:22AM +0900, Joonsoo Kim wrote:
> > @@ -2600,6 +2593,16 @@ static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
> >  			classzone_idx--;
> >  
> >  		/*
> > +		 * Shrink each node in the zonelist once. If the zonelist is
> > +		 * ordered by zone (not the default) then a node may be
> > +		 * shrunk multiple times but in that case the user prefers
> > +		 * lower zones being preserved
> > +		 */
> > +		if (zone->zone_pgdat == last_pgdat)
> > +			continue;
> > +		last_pgdat = zone->zone_pgdat;
> > +
> > +		/*
> 
> After this change, compaction_ready() which uses zone information
> would be called with highest zone in node. So, if some lower zone in
> that node is compaction-ready, we cannot stop the reclaim.
> 

Yes. It only impacts direct reclaim but potentially it's an issue. I'll
fix it.

-- 
Mel Gorman
SUSE Labs

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web