kernel26-alx/patches-2.6.21-r14/0012-2.6.21-mm-convert_swappiness_to_mapped.patch

Turn the "swappiness" knob into one with well defined semantics. Rename it
"mapped" to correspond directly with the percentage of mapped ram or
"applications" as users think of it. Currently the swappiness algorithm can
easily lead to swapping situations on simple file copies due to the distress
algorithm which too easily overrides the swappiness value. Add a
"hardmaplimit" tunable, on by default, which only allows the vm to override
the "mapped" tunable when distress is at its greatest to prevent false
out-of-memory situations.

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

 Documentation/sysctl/vm.txt |   23 +++++++++++++++++++++++
 include/linux/swap.h        |    3 ++-
 include/linux/sysctl.h      |    2 +-
 kernel/sysctl.c             |   16 ++++++++++++----
 mm/vmscan.c                 |   27 ++++++++++++++++-----------
 5 files changed, 54 insertions(+), 17 deletions(-)

Index: linux-2.6.21-ck2/include/linux/swap.h
===================================================================
--- linux-2.6.21-ck2.orig/include/linux/swap.h	2007-05-14 19:49:55.000000000 +1000
+++ linux-2.6.21-ck2/include/linux/swap.h	2007-05-14 19:49:55.000000000 +1000
@@ -191,7 +191,8 @@ extern void swap_setup(void);
 /* linux/mm/vmscan.c */
 extern unsigned long try_to_free_pages(struct zone **, gfp_t);
 extern unsigned long shrink_all_memory(unsigned long nr_pages);
-extern int vm_swappiness;
+extern int vm_mapped;
+extern int vm_hardmaplimit;
 extern int remove_mapping(struct address_space *mapping, struct page *page);
 extern long vm_total_pages;
 
Index: linux-2.6.21-ck2/include/linux/sysctl.h
===================================================================
--- linux-2.6.21-ck2.orig/include/linux/sysctl.h	2007-05-14 19:49:19.000000000 +1000
+++ linux-2.6.21-ck2/include/linux/sysctl.h	2007-05-14 19:49:55.000000000 +1000
@@ -190,7 +190,7 @@ enum
 	VM_OVERCOMMIT_RATIO=16, /* percent of RAM to allow overcommit in */
 	VM_PAGEBUF=17,		/* struct: Control pagebuf parameters */
 	VM_HUGETLB_PAGES=18,	/* int: Number of available Huge Pages */
-	VM_SWAPPINESS=19,	/* Tendency to steal mapped memory */
+	VM_MAPPED=19,		/* percent mapped min while evicting cache */
 	VM_LOWMEM_RESERVE_RATIO=20,/* reservation ratio for lower memory zones */
 	VM_MIN_FREE_KBYTES=21,	/* Minimum free kilobytes to maintain */
 	VM_MAX_MAP_COUNT=22,	/* int: Maximum number of mmaps/address-space */
Index: linux-2.6.21-ck2/kernel/sysctl.c
===================================================================
--- linux-2.6.21-ck2.orig/kernel/sysctl.c	2007-05-14 19:49:55.000000000 +1000
+++ linux-2.6.21-ck2/kernel/sysctl.c	2007-05-14 19:49:55.000000000 +1000
@@ -741,16 +741,24 @@ static ctl_table vm_table[] = {
 		.proc_handler	= &proc_dointvec,
 	},
 	{
-		.ctl_name	= VM_SWAPPINESS,
-		.procname	= "swappiness",
-		.data		= &vm_swappiness,
-		.maxlen		= sizeof(vm_swappiness),
+		.ctl_name	= CTL_UNNUMBERED,
+		.procname	= "mapped",
+		.data		= &vm_mapped,
+		.maxlen		= sizeof(vm_mapped),
 		.mode		= 0644,
 		.proc_handler	= &proc_dointvec_minmax,
 		.strategy	= &sysctl_intvec,
 		.extra1		= &zero,
 		.extra2		= &one_hundred,
 	},
+	{
+		.ctl_name	= CTL_UNNUMBERED,
+		.procname	= "hardmaplimit",
+		.data		= &vm_hardmaplimit,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
 #ifdef CONFIG_HUGETLB_PAGE
 	 {
 		.ctl_name	= VM_HUGETLB_PAGES,
Index: linux-2.6.21-ck2/mm/vmscan.c
===================================================================
--- linux-2.6.21-ck2.orig/mm/vmscan.c	2007-05-14 19:49:55.000000000 +1000
+++ linux-2.6.21-ck2/mm/vmscan.c	2007-05-14 19:49:55.000000000 +1000
@@ -64,7 +64,7 @@ struct scan_control {
 	 * whole list at once. */
 	int swap_cluster_max;
 
-	int swappiness;
+	int mapped;
 
 	int all_unreclaimable;
 };
@@ -111,9 +111,10 @@ struct shrinker {
 #endif
 
 /*
- * From 0 .. 100.  Higher means more swappy.
+ * From 0 .. 100.  Lower means more swappy.
  */
-int vm_swappiness = 60;
+int vm_mapped __read_mostly = 66;
+int vm_hardmaplimit __read_mostly = 1;
 long vm_total_pages;	/* The total number of pages which the VM controls */
 
 static LIST_HEAD(shrinker_list);
@@ -809,10 +810,14 @@ static void shrink_active_list(unsigned 
 		 * The distress ratio is important - we don't want to start
 		 * going oom.
 		 *
-		 * A 100% value of vm_swappiness overrides this algorithm
-		 * altogether.
+		 * This distress value is ignored if we apply a hardmaplimit except
+		 * in extreme distress.
+		 *
+		 * A 0% value of vm_mapped overrides this algorithm altogether.
 		 */
-		swap_tendency = mapped_ratio / 2 + distress + sc->swappiness;
+		swap_tendency = mapped_ratio * 100 / (sc->mapped + 1);
+		if (!vm_hardmaplimit || distress == 100)
+			swap_tendency += distress;
 
 		/*
 		 * Now use this metric to decide whether to start moving mapped
@@ -1031,7 +1036,7 @@ unsigned long try_to_free_pages(struct z
 		.may_writepage = !laptop_mode,
 		.swap_cluster_max = SWAP_CLUSTER_MAX,
 		.may_swap = 1,
-		.swappiness = vm_swappiness,
+		.mapped = vm_mapped,
 	};
 
 	delay_swap_prefetch();
@@ -1138,7 +1143,7 @@ static unsigned long balance_pgdat(pg_da
 		.gfp_mask = GFP_KERNEL,
 		.may_swap = 1,
 		.swap_cluster_max = SWAP_CLUSTER_MAX,
-		.swappiness = vm_swappiness,
+		.mapped = vm_mapped,
 	};
 	/*
 	 * temp_priority is used to remember the scanning priority at which
@@ -1446,7 +1451,7 @@ unsigned long shrink_all_memory(unsigned
 		.may_swap = 0,
 		.swap_cluster_max = nr_pages,
 		.may_writepage = 1,
-		.swappiness = vm_swappiness,
+		.mapped = vm_mapped,
 	};
 
 	current->reclaim_state = &reclaim_state;
@@ -1481,7 +1486,7 @@ unsigned long shrink_all_memory(unsigned
 		/* Force reclaiming mapped pages in the passes #3 and #4 */
 		if (pass > 2) {
 			sc.may_swap = 1;
-			sc.swappiness = 100;
+			sc.mapped = 0;
 		}
 
 		for (prio = DEF_PRIORITY; prio >= 0; prio--) {
@@ -1629,7 +1634,7 @@ static int __zone_reclaim(struct zone *z
 		.swap_cluster_max = max_t(unsigned long, nr_pages,
 					SWAP_CLUSTER_MAX),
 		.gfp_mask = gfp_mask,
-		.swappiness = vm_swappiness,
+		.mapped = vm_mapped,
 	};
 	unsigned long slab_reclaimable;
 
Index: linux-2.6.21-ck2/Documentation/sysctl/vm.txt
===================================================================
--- linux-2.6.21-ck2.orig/Documentation/sysctl/vm.txt	2007-05-14 19:49:55.000000000 +1000
+++ linux-2.6.21-ck2/Documentation/sysctl/vm.txt	2007-05-14 19:49:55.000000000 +1000
@@ -22,6 +22,8 @@ Currently, these files are in /proc/sys/
 - dirty_background_ratio
 - dirty_expire_centisecs
 - dirty_writeback_centisecs
+- hardmaplimit
+- mapped
 - max_map_count
 - min_free_kbytes
 - laptop_mode
@@ -87,6 +89,27 @@ for swap because we only cluster swap da
 
 ==============================================================
 
+hardmaplimit:
+
+This flag makes the vm adhere to the mapped value as closely as possible
+except in the most extreme vm stress where doing so would provoke an out
+of memory condition (see mapped below).
+
+Enabled by default.
+
+==============================================================
+
+mapped:
+
+This is the percentage ram that is filled with mapped pages (applications)
+before the vm will start reclaiming mapped pages by moving them to swap.
+It is altered by the relative stress of the vm at the time so is not
+strictly adhered to to prevent provoking out of memory kills.
+
+Set to 66 by default.
+
+==============================================================
+
 max_map_count:
 
 This file contains the maximum number of memory map areas a process
1	niro	447	Turn the "swappiness" knob into one with well defined semantics. Rename it
2			"mapped" to correspond directly with the percentage of mapped ram or
3			"applications" as users think of it. Currently the swappiness algorithm can
4			easily lead to swapping situations on simple file copies due to the distress
5			algorithm which too easily overrides the swappiness value. Add a
6			"hardmaplimit" tunable, on by default, which only allows the vm to override
7			the "mapped" tunable when distress is at its greatest to prevent false
8			out-of-memory situations.
9
10			Signed-off-by: Con Kolivas <kernel@kolivas.org>
11
12			Documentation/sysctl/vm.txt \| 23 +++++++++++++++++++++++
13			include/linux/swap.h \| 3 ++-
14			include/linux/sysctl.h \| 2 +-
15			kernel/sysctl.c \| 16 ++++++++++++----
16			mm/vmscan.c \| 27 ++++++++++++++++-----------
17			5 files changed, 54 insertions(+), 17 deletions(-)
18
19			Index: linux-2.6.21-ck2/include/linux/swap.h
20			===================================================================
21			--- linux-2.6.21-ck2.orig/include/linux/swap.h 2007-05-14 19:49:55.000000000 +1000
22			+++ linux-2.6.21-ck2/include/linux/swap.h 2007-05-14 19:49:55.000000000 +1000
23			@@ -191,7 +191,8 @@ extern void swap_setup(void);
24			/* linux/mm/vmscan.c */
25			extern unsigned long try_to_free_pages(struct zone **, gfp_t);
26			extern unsigned long shrink_all_memory(unsigned long nr_pages);
27			-extern int vm_swappiness;
28			+extern int vm_mapped;
29			+extern int vm_hardmaplimit;
30			extern int remove_mapping(struct address_space mapping, struct page page);
31			extern long vm_total_pages;
32
33			Index: linux-2.6.21-ck2/include/linux/sysctl.h
34			===================================================================
35			--- linux-2.6.21-ck2.orig/include/linux/sysctl.h 2007-05-14 19:49:19.000000000 +1000
36			+++ linux-2.6.21-ck2/include/linux/sysctl.h 2007-05-14 19:49:55.000000000 +1000
37			@@ -190,7 +190,7 @@ enum
38			VM_OVERCOMMIT_RATIO=16, /* percent of RAM to allow overcommit in */
39			VM_PAGEBUF=17, /* struct: Control pagebuf parameters */
40			VM_HUGETLB_PAGES=18, /* int: Number of available Huge Pages */
41			- VM_SWAPPINESS=19, /* Tendency to steal mapped memory */
42			+ VM_MAPPED=19, /* percent mapped min while evicting cache */
43			VM_LOWMEM_RESERVE_RATIO=20,/* reservation ratio for lower memory zones */
44			VM_MIN_FREE_KBYTES=21, /* Minimum free kilobytes to maintain */
45			VM_MAX_MAP_COUNT=22, /* int: Maximum number of mmaps/address-space */
46			Index: linux-2.6.21-ck2/kernel/sysctl.c
47			===================================================================
48			--- linux-2.6.21-ck2.orig/kernel/sysctl.c 2007-05-14 19:49:55.000000000 +1000
49			+++ linux-2.6.21-ck2/kernel/sysctl.c 2007-05-14 19:49:55.000000000 +1000
50			@@ -741,16 +741,24 @@ static ctl_table vm_table[] = {
51			.proc_handler = &proc_dointvec,
52			},
53			{
54			- .ctl_name = VM_SWAPPINESS,
55			- .procname = "swappiness",
56			- .data = &vm_swappiness,
57			- .maxlen = sizeof(vm_swappiness),
58			+ .ctl_name = CTL_UNNUMBERED,
59			+ .procname = "mapped",
60			+ .data = &vm_mapped,
61			+ .maxlen = sizeof(vm_mapped),
62			.mode = 0644,
63			.proc_handler = &proc_dointvec_minmax,
64			.strategy = &sysctl_intvec,
65			.extra1 = &zero,
66			.extra2 = &one_hundred,
67			},
68			+ {
69			+ .ctl_name = CTL_UNNUMBERED,
70			+ .procname = "hardmaplimit",
71			+ .data = &vm_hardmaplimit,
72			+ .maxlen = sizeof(int),
73			+ .mode = 0644,
74			+ .proc_handler = &proc_dointvec,
75			+ },
76			#ifdef CONFIG_HUGETLB_PAGE
77			{
78			.ctl_name = VM_HUGETLB_PAGES,
79			Index: linux-2.6.21-ck2/mm/vmscan.c
80			===================================================================
81			--- linux-2.6.21-ck2.orig/mm/vmscan.c 2007-05-14 19:49:55.000000000 +1000
82			+++ linux-2.6.21-ck2/mm/vmscan.c 2007-05-14 19:49:55.000000000 +1000
83			@@ -64,7 +64,7 @@ struct scan_control {
84			* whole list at once. */
85			int swap_cluster_max;
86
87			- int swappiness;
88			+ int mapped;
89
90			int all_unreclaimable;
91			};
92			@@ -111,9 +111,10 @@ struct shrinker {
93			#endif
94
95			/*
96			- * From 0 .. 100. Higher means more swappy.
97			+ * From 0 .. 100. Lower means more swappy.
98			*/
99			-int vm_swappiness = 60;
100			+int vm_mapped __read_mostly = 66;
101			+int vm_hardmaplimit __read_mostly = 1;
102			long vm_total_pages; /* The total number of pages which the VM controls */
103
104			static LIST_HEAD(shrinker_list);
105			@@ -809,10 +810,14 @@ static void shrink_active_list(unsigned
106			* The distress ratio is important - we don't want to start
107			* going oom.
108			*
109			- * A 100% value of vm_swappiness overrides this algorithm
110			- * altogether.
111			+ * This distress value is ignored if we apply a hardmaplimit except
112			+ * in extreme distress.
113			+ *
114			+ * A 0% value of vm_mapped overrides this algorithm altogether.
115			*/
116			- swap_tendency = mapped_ratio / 2 + distress + sc->swappiness;
117			+ swap_tendency = mapped_ratio * 100 / (sc->mapped + 1);
118			+ if (!vm_hardmaplimit \|\| distress == 100)
119			+ swap_tendency += distress;
120
121			/*
122			* Now use this metric to decide whether to start moving mapped
123			@@ -1031,7 +1036,7 @@ unsigned long try_to_free_pages(struct z
124			.may_writepage = !laptop_mode,
125			.swap_cluster_max = SWAP_CLUSTER_MAX,
126			.may_swap = 1,
127			- .swappiness = vm_swappiness,
128			+ .mapped = vm_mapped,
129			};
130
131			delay_swap_prefetch();
132			@@ -1138,7 +1143,7 @@ static unsigned long balance_pgdat(pg_da
133			.gfp_mask = GFP_KERNEL,
134			.may_swap = 1,
135			.swap_cluster_max = SWAP_CLUSTER_MAX,
136			- .swappiness = vm_swappiness,
137			+ .mapped = vm_mapped,
138			};
139			/*
140			* temp_priority is used to remember the scanning priority at which
141			@@ -1446,7 +1451,7 @@ unsigned long shrink_all_memory(unsigned
142			.may_swap = 0,
143			.swap_cluster_max = nr_pages,
144			.may_writepage = 1,
145			- .swappiness = vm_swappiness,
146			+ .mapped = vm_mapped,
147			};
148
149			current->reclaim_state = &reclaim_state;
150			@@ -1481,7 +1486,7 @@ unsigned long shrink_all_memory(unsigned
151			/* Force reclaiming mapped pages in the passes #3 and #4 */
152			if (pass > 2) {
153			sc.may_swap = 1;
154			- sc.swappiness = 100;
155			+ sc.mapped = 0;
156			}
157
158			for (prio = DEF_PRIORITY; prio >= 0; prio--) {
159			@@ -1629,7 +1634,7 @@ static int __zone_reclaim(struct zone *z
160			.swap_cluster_max = max_t(unsigned long, nr_pages,
161			SWAP_CLUSTER_MAX),
162			.gfp_mask = gfp_mask,
163			- .swappiness = vm_swappiness,
164			+ .mapped = vm_mapped,
165			};
166			unsigned long slab_reclaimable;
167
168			Index: linux-2.6.21-ck2/Documentation/sysctl/vm.txt
169			===================================================================
170			--- linux-2.6.21-ck2.orig/Documentation/sysctl/vm.txt 2007-05-14 19:49:55.000000000 +1000
171			+++ linux-2.6.21-ck2/Documentation/sysctl/vm.txt 2007-05-14 19:49:55.000000000 +1000
172			@@ -22,6 +22,8 @@ Currently, these files are in /proc/sys/
173			- dirty_background_ratio
174			- dirty_expire_centisecs
175			- dirty_writeback_centisecs
176			+- hardmaplimit
177			+- mapped
178			- max_map_count
179			- min_free_kbytes
180			- laptop_mode
181			@@ -87,6 +89,27 @@ for swap because we only cluster swap da
182
183			==============================================================
184
185			+hardmaplimit:
186			+
187			+This flag makes the vm adhere to the mapped value as closely as possible
188			+except in the most extreme vm stress where doing so would provoke an out
189			+of memory condition (see mapped below).
190			+
191			+Enabled by default.
192			+
193			+==============================================================
194			+
195			+mapped:
196			+
197			+This is the percentage ram that is filled with mapped pages (applications)
198			+before the vm will start reclaiming mapped pages by moving them to swap.
199			+It is altered by the relative stress of the vm at the time so is not
200			+strictly adhered to to prevent provoking out of memory kills.
201			+
202			+Set to 66 by default.
203			+
204			+==============================================================
205			+
206			max_map_count:
207
208			This file contains the maximum number of memory map areas a process