kernel26-magellan/patches-2.6.20-r4/0016-2.6.20-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      |    3 ++-
 kernel/sysctl.c             |   16 ++++++++++++----
 mm/vmscan.c                 |   27 ++++++++++++++++-----------
 5 files changed, 55 insertions(+), 17 deletions(-)

Index: linux-2.6.20-ck1/include/linux/swap.h
===================================================================
--- linux-2.6.20-ck1.orig/include/linux/swap.h	2007-02-16 19:01:33.000000000 +1100
+++ linux-2.6.20-ck1/include/linux/swap.h	2007-02-16 19:01:33.000000000 +1100
@@ -189,7 +189,8 @@ 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-2.6.20-ck1/include/linux/sysctl.h
===================================================================
--- linux-2.6.20-ck1.orig/include/linux/sysctl.h	2007-02-16 19:01:33.000000000 +1100
+++ linux-2.6.20-ck1/include/linux/sysctl.h	2007-02-16 19:01:33.000000000 +1100
@@ -185,7 +185,7 @@ 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 */
@@ -203,6 +203,7 @@ enum
 	VM_VDSO_ENABLED=34,	/* map VDSO into new processes? */
 	VM_MIN_SLAB=35,		 /* Percent pages ignored by zone reclaim */
 	VM_SWAP_PREFETCH=36,	/* swap prefetch */
+	VM_HARDMAPLIMIT=37,	/* Make mapped a hard limit */
 };
 
 
Index: linux-2.6.20-ck1/kernel/sysctl.c
===================================================================
--- linux-2.6.20-ck1.orig/kernel/sysctl.c	2007-02-16 19:01:33.000000000 +1100
+++ linux-2.6.20-ck1/kernel/sysctl.c	2007-02-16 19:01:33.000000000 +1100
@@ -900,16 +900,24 @@ static ctl_table vm_table[] = {
 		.proc_handler	= &proc_dointvec,
 	},
 	{
-		.ctl_name	= VM_SWAPPINESS,
-		.procname	= "swappiness",
-		.data		= &vm_swappiness,
-		.maxlen		= sizeof(vm_swappiness),
+		.ctl_name	= VM_MAPPED,
+		.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	= VM_HARDMAPLIMIT,
+		.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-2.6.20-ck1/mm/vmscan.c
===================================================================
--- linux-2.6.20-ck1.orig/mm/vmscan.c	2007-02-16 19:01:33.000000000 +1100
+++ linux-2.6.20-ck1/mm/vmscan.c	2007-02-16 19:01:33.000000000 +1100
@@ -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);
@@ -808,10 +809,14 @@ 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
@@ -1028,7 +1033,7 @@ 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();
@@ -1134,7 +1139,7 @@ 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
@@ -1441,7 +1446,7 @@ 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;
@@ -1476,7 +1481,7 @@ 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--) {
@@ -1624,7 +1629,7 @@ 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-2.6.20-ck1/Documentation/sysctl/vm.txt
===================================================================
--- linux-2.6.20-ck1.orig/Documentation/sysctl/vm.txt	2007-02-16 19:01:33.000000000 +1100
+++ linux-2.6.20-ck1/Documentation/sysctl/vm.txt	2007-02-16 19:01:33.000000000 +1100
@@ -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	132	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 \| 3 ++-
15			kernel/sysctl.c \| 16 ++++++++++++----
16			mm/vmscan.c \| 27 ++++++++++++++++-----------
17			5 files changed, 55 insertions(+), 17 deletions(-)
18
19			Index: linux-2.6.20-ck1/include/linux/swap.h
20			===================================================================
21			--- linux-2.6.20-ck1.orig/include/linux/swap.h 2007-02-16 19:01:33.000000000 +1100
22			+++ linux-2.6.20-ck1/include/linux/swap.h 2007-02-16 19:01:33.000000000 +1100
23			@@ -189,7 +189,8 @@ 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-2.6.20-ck1/include/linux/sysctl.h
34			===================================================================
35			--- linux-2.6.20-ck1.orig/include/linux/sysctl.h 2007-02-16 19:01:33.000000000 +1100
36			+++ linux-2.6.20-ck1/include/linux/sysctl.h 2007-02-16 19:01:33.000000000 +1100
37			@@ -185,7 +185,7 @@ 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			@@ -203,6 +203,7 @@ enum
47			VM_VDSO_ENABLED=34, /* map VDSO into new processes? */
48			VM_MIN_SLAB=35, /* Percent pages ignored by zone reclaim */
49			VM_SWAP_PREFETCH=36, /* swap prefetch */
50			+ VM_HARDMAPLIMIT=37, /* Make mapped a hard limit */
51			};
52
53
54			Index: linux-2.6.20-ck1/kernel/sysctl.c
55			===================================================================
56			--- linux-2.6.20-ck1.orig/kernel/sysctl.c 2007-02-16 19:01:33.000000000 +1100
57			+++ linux-2.6.20-ck1/kernel/sysctl.c 2007-02-16 19:01:33.000000000 +1100
58			@@ -900,16 +900,24 @@ static ctl_table vm_table[] = {
59			.proc_handler = &proc_dointvec,
60			},
61			{
62			- .ctl_name = VM_SWAPPINESS,
63			- .procname = "swappiness",
64			- .data = &vm_swappiness,
65			- .maxlen = sizeof(vm_swappiness),
66			+ .ctl_name = VM_MAPPED,
67			+ .procname = "mapped",
68			+ .data = &vm_mapped,
69			+ .maxlen = sizeof(vm_mapped),
70			.mode = 0644,
71			.proc_handler = &proc_dointvec_minmax,
72			.strategy = &sysctl_intvec,
73			.extra1 = &zero,
74			.extra2 = &one_hundred,
75			},
76			+ {
77			+ .ctl_name = VM_HARDMAPLIMIT,
78			+ .procname = "hardmaplimit",
79			+ .data = &vm_hardmaplimit,
80			+ .maxlen = sizeof(int),
81			+ .mode = 0644,
82			+ .proc_handler = &proc_dointvec,
83			+ },
84			#ifdef CONFIG_HUGETLB_PAGE
85			{
86			.ctl_name = VM_HUGETLB_PAGES,
87			Index: linux-2.6.20-ck1/mm/vmscan.c
88			===================================================================
89			--- linux-2.6.20-ck1.orig/mm/vmscan.c 2007-02-16 19:01:33.000000000 +1100
90			+++ linux-2.6.20-ck1/mm/vmscan.c 2007-02-16 19:01:33.000000000 +1100
91			@@ -64,7 +64,7 @@ struct scan_control {
92			* whole list at once. */
93			int swap_cluster_max;
94
95			- int swappiness;
96			+ int mapped;
97
98			int all_unreclaimable;
99			};
100			@@ -111,9 +111,10 @@ struct shrinker {
101			#endif
102
103			/*
104			- * From 0 .. 100. Higher means more swappy.
105			+ * From 0 .. 100. Lower means more swappy.
106			*/
107			-int vm_swappiness = 60;
108			+int vm_mapped __read_mostly = 66;
109			+int vm_hardmaplimit __read_mostly = 1;
110			long vm_total_pages; /* The total number of pages which the VM controls */
111
112			static LIST_HEAD(shrinker_list);
113			@@ -808,10 +809,14 @@ static void shrink_active_list(unsigned
114			* The distress ratio is important - we don't want to start
115			* going oom.
116			*
117			- * A 100% value of vm_swappiness overrides this algorithm
118			- * altogether.
119			+ * This distress value is ignored if we apply a hardmaplimit except
120			+ * in extreme distress.
121			+ *
122			+ * A 0% value of vm_mapped overrides this algorithm altogether.
123			*/
124			- swap_tendency = mapped_ratio / 2 + distress + sc->swappiness;
125			+ swap_tendency = mapped_ratio * 100 / (sc->mapped + 1);
126			+ if (!vm_hardmaplimit \|\| distress == 100)
127			+ swap_tendency += distress;
128
129			/*
130			* Now use this metric to decide whether to start moving mapped
131			@@ -1028,7 +1033,7 @@ unsigned long try_to_free_pages(struct z
132			.may_writepage = !laptop_mode,
133			.swap_cluster_max = SWAP_CLUSTER_MAX,
134			.may_swap = 1,
135			- .swappiness = vm_swappiness,
136			+ .mapped = vm_mapped,
137			};
138
139			delay_swap_prefetch();
140			@@ -1134,7 +1139,7 @@ static unsigned long balance_pgdat(pg_da
141			.gfp_mask = GFP_KERNEL,
142			.may_swap = 1,
143			.swap_cluster_max = SWAP_CLUSTER_MAX,
144			- .swappiness = vm_swappiness,
145			+ .mapped = vm_mapped,
146			};
147			/*
148			* temp_priority is used to remember the scanning priority at which
149			@@ -1441,7 +1446,7 @@ unsigned long shrink_all_memory(unsigned
150			.may_swap = 0,
151			.swap_cluster_max = nr_pages,
152			.may_writepage = 1,
153			- .swappiness = vm_swappiness,
154			+ .mapped = vm_mapped,
155			};
156
157			current->reclaim_state = &reclaim_state;
158			@@ -1476,7 +1481,7 @@ unsigned long shrink_all_memory(unsigned
159			/* Force reclaiming mapped pages in the passes #3 and #4 */
160			if (pass > 2) {
161			sc.may_swap = 1;
162			- sc.swappiness = 100;
163			+ sc.mapped = 0;
164			}
165
166			for (prio = DEF_PRIORITY; prio >= 0; prio--) {
167			@@ -1624,7 +1629,7 @@ static int __zone_reclaim(struct zone *z
168			.swap_cluster_max = max_t(unsigned long, nr_pages,
169			SWAP_CLUSTER_MAX),
170			.gfp_mask = gfp_mask,
171			- .swappiness = vm_swappiness,
172			+ .mapped = vm_mapped,
173			};
174			unsigned long slab_reclaimable;
175
176			Index: linux-2.6.20-ck1/Documentation/sysctl/vm.txt
177			===================================================================
178			--- linux-2.6.20-ck1.orig/Documentation/sysctl/vm.txt 2007-02-16 19:01:33.000000000 +1100
179			+++ linux-2.6.20-ck1/Documentation/sysctl/vm.txt 2007-02-16 19:01:33.000000000 +1100
180			@@ -22,6 +22,8 @@ Currently, these files are in /proc/sys/
181			- dirty_background_ratio
182			- dirty_expire_centisecs
183			- dirty_writeback_centisecs
184			+- hardmaplimit
185			+- mapped
186			- max_map_count
187			- min_free_kbytes
188			- laptop_mode
189			@@ -87,6 +89,27 @@ for swap because we only cluster swap da
190
191			==============================================================
192
193			+hardmaplimit:
194			+
195			+This flag makes the vm adhere to the mapped value as closely as possible
196			+except in the most extreme vm stress where doing so would provoke an out
197			+of memory condition (see mapped below).
198			+
199			+Enabled by default.
200			+
201			+==============================================================
202			+
203			+mapped:
204			+
205			+This is the percentage ram that is filled with mapped pages (applications)
206			+before the vm will start reclaiming mapped pages by moving them to swap.
207			+It is altered by the relative stress of the vm at the time so is not
208			+strictly adhered to to prevent provoking out of memory kills.
209			+
210			+Set to 66 by default.
211			+
212			+==============================================================
213			+
214			max_map_count:
215
216			This file contains the maximum number of memory map areas a process