kernel26-alx/patches-2.6.17-r5/0020-2.6.17-mm-lots_watermark.diff

The vm currently performs scanning when allocating ram once the watermarks
are below the pages_low value and tries to restore them to the pages_high
watermark. The disadvantage of this is that we are scanning most aggresssively
at the same time we are allocating ram regardless of the stress the vm is
under. Add a pages_lots watermark and allow the watermark to be relaxed
according to the stress the vm is at the time (according to the priority
value). Thus we have more in reserve next time we are allocating ram and end
up scanning less aggresssively. Note the actual pages_lots isn't used directly
in this code.

Signed-off-by: Con Kolivas <kernel@kolivas.org>

 include/linux/mmzone.h |    2 +-
 mm/page_alloc.c        |    5 +++++
 mm/vmscan.c            |   18 +++++++++++++++---
 3 files changed, 21 insertions(+), 4 deletions(-)

Index: linux-ck-dev/include/linux/mmzone.h
===================================================================
--- linux-ck-dev.orig/include/linux/mmzone.h	2006-06-18 15:20:12.000000000 +1000
+++ linux-ck-dev/include/linux/mmzone.h	2006-06-18 15:25:00.000000000 +1000
@@ -123,7 +123,7 @@ struct per_cpu_pageset {
 struct zone {
 	/* Fields commonly accessed by the page allocator */
 	unsigned long		free_pages;
-	unsigned long		pages_min, pages_low, pages_high;
+	unsigned long		pages_min, pages_low, pages_high, pages_lots;
 	/*
 	 * We don't know if the memory that we're going to allocate will be freeable
 	 * or/and it will be released eventually, so to avoid totally wasting several
Index: linux-ck-dev/mm/page_alloc.c
===================================================================
--- linux-ck-dev.orig/mm/page_alloc.c	2006-06-18 15:20:12.000000000 +1000
+++ linux-ck-dev/mm/page_alloc.c	2006-06-18 15:25:00.000000000 +1000
@@ -1461,6 +1461,7 @@ void show_free_areas(void)
 			" min:%lukB"
 			" low:%lukB"
 			" high:%lukB"
+			" lots:%lukB"
 			" active:%lukB"
 			" inactive:%lukB"
 			" present:%lukB"
@@ -1472,6 +1473,7 @@ void show_free_areas(void)
 			K(zone->pages_min),
 			K(zone->pages_low),
 			K(zone->pages_high),
+			K(zone->pages_lots),
 			K(zone->nr_active),
 			K(zone->nr_inactive),
 			K(zone->present_pages),
@@ -2261,6 +2263,7 @@ static int zoneinfo_show(struct seq_file
 			   "\n        min      %lu"
 			   "\n        low      %lu"
 			   "\n        high     %lu"
+			   "\n        lots     %lu"
 			   "\n        active   %lu"
 			   "\n        inactive %lu"
 			   "\n        scanned  %lu (a: %lu i: %lu)"
@@ -2270,6 +2273,7 @@ static int zoneinfo_show(struct seq_file
 			   zone->pages_min,
 			   zone->pages_low,
 			   zone->pages_high,
+			   zone->pages_lots,
 			   zone->nr_active,
 			   zone->nr_inactive,
 			   zone->pages_scanned,
@@ -2609,6 +2613,7 @@ void setup_per_zone_pages_min(void)
 
 		zone->pages_low   = zone->pages_min + (tmp >> 2);
 		zone->pages_high  = zone->pages_min + (tmp >> 1);
+		zone->pages_lots  = zone->pages_min + tmp;
 		spin_unlock_irqrestore(&zone->lru_lock, flags);
 	}
 
Index: linux-ck-dev/mm/vmscan.c
===================================================================
--- linux-ck-dev.orig/mm/vmscan.c	2006-06-18 15:24:58.000000000 +1000
+++ linux-ck-dev/mm/vmscan.c	2006-06-18 15:25:00.000000000 +1000
@@ -1095,6 +1095,7 @@ loop_again:
 		 */
 		for (i = pgdat->nr_zones - 1; i >= 0; i--) {
 			struct zone *zone = pgdat->node_zones + i;
+			unsigned long watermark;
 
 			if (!populated_zone(zone))
 				continue;
@@ -1102,11 +1103,18 @@ loop_again:
 			if (zone->all_unreclaimable && priority != DEF_PRIORITY)
 				continue;
 
-			if (!zone_watermark_ok(zone, order, zone->pages_high,
-					       0, 0)) {
+			/*
+			 * The watermark is relaxed depending on the
+			 * level of "priority" till it drops to
+			 * pages_high.
+			 */
+			watermark = zone->pages_high + (zone->pages_high *
+				    priority / DEF_PRIORITY);
+			if (!zone_watermark_ok(zone, order, watermark, 0, 0)) {
 				end_zone = i;
 				goto scan;
 			}
+
 		}
 		goto out;
 scan:
@@ -1128,6 +1136,7 @@ scan:
 		for (i = 0; i <= end_zone; i++) {
 			struct zone *zone = pgdat->node_zones + i;
 			int nr_slab;
+			unsigned long watermark;
 
 			if (!populated_zone(zone))
 				continue;
@@ -1135,7 +1144,10 @@ scan:
 			if (zone->all_unreclaimable && priority != DEF_PRIORITY)
 				continue;
 
-			if (!zone_watermark_ok(zone, order, zone->pages_high,
+			watermark = zone->pages_high + (zone->pages_high *
+				    priority / DEF_PRIORITY);
+
+			if (!zone_watermark_ok(zone, order, watermark,
 					       end_zone, 0))
 				all_zones_ok = 0;
 			zone->temp_priority = priority;
1	niro	199	The vm currently performs scanning when allocating ram once the watermarks
2			are below the pages_low value and tries to restore them to the pages_high
3			watermark. The disadvantage of this is that we are scanning most aggresssively
4			at the same time we are allocating ram regardless of the stress the vm is
5			under. Add a pages_lots watermark and allow the watermark to be relaxed
6			according to the stress the vm is at the time (according to the priority
7			value). Thus we have more in reserve next time we are allocating ram and end
8			up scanning less aggresssively. Note the actual pages_lots isn't used directly
9			in this code.
10
11			Signed-off-by: Con Kolivas <kernel@kolivas.org>
12
13			include/linux/mmzone.h \| 2 +-
14			mm/page_alloc.c \| 5 +++++
15			mm/vmscan.c \| 18 +++++++++++++++---
16			3 files changed, 21 insertions(+), 4 deletions(-)
17
18			Index: linux-ck-dev/include/linux/mmzone.h
19			===================================================================
20			--- linux-ck-dev.orig/include/linux/mmzone.h 2006-06-18 15:20:12.000000000 +1000
21			+++ linux-ck-dev/include/linux/mmzone.h 2006-06-18 15:25:00.000000000 +1000
22			@@ -123,7 +123,7 @@ struct per_cpu_pageset {
23			struct zone {
24			/* Fields commonly accessed by the page allocator */
25			unsigned long free_pages;
26			- unsigned long pages_min, pages_low, pages_high;
27			+ unsigned long pages_min, pages_low, pages_high, pages_lots;
28			/*
29			* We don't know if the memory that we're going to allocate will be freeable
30			* or/and it will be released eventually, so to avoid totally wasting several
31			Index: linux-ck-dev/mm/page_alloc.c
32			===================================================================
33			--- linux-ck-dev.orig/mm/page_alloc.c 2006-06-18 15:20:12.000000000 +1000
34			+++ linux-ck-dev/mm/page_alloc.c 2006-06-18 15:25:00.000000000 +1000
35			@@ -1461,6 +1461,7 @@ void show_free_areas(void)
36			" min:%lukB"
37			" low:%lukB"
38			" high:%lukB"
39			+ " lots:%lukB"
40			" active:%lukB"
41			" inactive:%lukB"
42			" present:%lukB"
43			@@ -1472,6 +1473,7 @@ void show_free_areas(void)
44			K(zone->pages_min),
45			K(zone->pages_low),
46			K(zone->pages_high),
47			+ K(zone->pages_lots),
48			K(zone->nr_active),
49			K(zone->nr_inactive),
50			K(zone->present_pages),
51			@@ -2261,6 +2263,7 @@ static int zoneinfo_show(struct seq_file
52			"\n min %lu"
53			"\n low %lu"
54			"\n high %lu"
55			+ "\n lots %lu"
56			"\n active %lu"
57			"\n inactive %lu"
58			"\n scanned %lu (a: %lu i: %lu)"
59			@@ -2270,6 +2273,7 @@ static int zoneinfo_show(struct seq_file
60			zone->pages_min,
61			zone->pages_low,
62			zone->pages_high,
63			+ zone->pages_lots,
64			zone->nr_active,
65			zone->nr_inactive,
66			zone->pages_scanned,
67			@@ -2609,6 +2613,7 @@ void setup_per_zone_pages_min(void)
68
69			zone->pages_low = zone->pages_min + (tmp >> 2);
70			zone->pages_high = zone->pages_min + (tmp >> 1);
71			+ zone->pages_lots = zone->pages_min + tmp;
72			spin_unlock_irqrestore(&zone->lru_lock, flags);
73			}
74
75			Index: linux-ck-dev/mm/vmscan.c
76			===================================================================
77			--- linux-ck-dev.orig/mm/vmscan.c 2006-06-18 15:24:58.000000000 +1000
78			+++ linux-ck-dev/mm/vmscan.c 2006-06-18 15:25:00.000000000 +1000
79			@@ -1095,6 +1095,7 @@ loop_again:
80			*/
81			for (i = pgdat->nr_zones - 1; i >= 0; i--) {
82			struct zone *zone = pgdat->node_zones + i;
83			+ unsigned long watermark;
84
85			if (!populated_zone(zone))
86			continue;
87			@@ -1102,11 +1103,18 @@ loop_again:
88			if (zone->all_unreclaimable && priority != DEF_PRIORITY)
89			continue;
90
91			- if (!zone_watermark_ok(zone, order, zone->pages_high,
92			- 0, 0)) {
93			+ /*
94			+ * The watermark is relaxed depending on the
95			+ * level of "priority" till it drops to
96			+ * pages_high.
97			+ */
98			+ watermark = zone->pages_high + (zone->pages_high *
99			+ priority / DEF_PRIORITY);
100			+ if (!zone_watermark_ok(zone, order, watermark, 0, 0)) {
101			end_zone = i;
102			goto scan;
103			}
104			+
105			}
106			goto out;
107			scan:
108			@@ -1128,6 +1136,7 @@ scan:
109			for (i = 0; i <= end_zone; i++) {
110			struct zone *zone = pgdat->node_zones + i;
111			int nr_slab;
112			+ unsigned long watermark;
113
114			if (!populated_zone(zone))
115			continue;
116			@@ -1135,7 +1144,10 @@ scan:
117			if (zone->all_unreclaimable && priority != DEF_PRIORITY)
118			continue;
119
120			- if (!zone_watermark_ok(zone, order, zone->pages_high,
121			+ watermark = zone->pages_high + (zone->pages_high *
122			+ priority / DEF_PRIORITY);
123			+
124			+ if (!zone_watermark_ok(zone, order, watermark,
125			end_zone, 0))
126			all_zones_ok = 0;
127			zone->temp_priority = priority;