kernel26-alx/patches-2.6.17-r6/0021-2.6.17-mm-kswapd_inherit_prio-1.patch

When kswapd is awoken due to reclaim by a running task, set the priority of
kswapd to that of the calling task thus making memory reclaim cpu activity
affected by nice level.

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

 include/linux/mmzone.h |    2 +-
 mm/page_alloc.c        |    2 +-
 mm/vmscan.c            |   40 ++++++++++++++++++++++++++++++++++++++--
 3 files changed, 40 insertions(+), 4 deletions(-)

Index: linux-ck-dev/include/linux/mmzone.h
===================================================================
--- linux-ck-dev.orig/include/linux/mmzone.h	2006-06-18 15:25:00.000000000 +1000
+++ linux-ck-dev/include/linux/mmzone.h	2006-06-18 15:25:02.000000000 +1000
@@ -330,7 +330,7 @@ void __get_zone_counts(unsigned long *ac
 void get_zone_counts(unsigned long *active, unsigned long *inactive,
 			unsigned long *free);
 void build_all_zonelists(void);
-void wakeup_kswapd(struct zone *zone, int order);
+void wakeup_kswapd(struct zone *zone, int order, struct task_struct *p);
 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
 		int classzone_idx, int alloc_flags);
 
Index: linux-ck-dev/mm/page_alloc.c
===================================================================
--- linux-ck-dev.orig/mm/page_alloc.c	2006-06-18 15:25:00.000000000 +1000
+++ linux-ck-dev/mm/page_alloc.c	2006-06-18 15:25:02.000000000 +1000
@@ -952,7 +952,7 @@ restart:
 
 	do {
 		if (cpuset_zone_allowed(*z, gfp_mask|__GFP_HARDWALL))
-			wakeup_kswapd(*z, order);
+			wakeup_kswapd(*z, order, p);
 	} while (*(++z));
 
 	/*
Index: linux-ck-dev/mm/vmscan.c
===================================================================
--- linux-ck-dev.orig/mm/vmscan.c	2006-06-18 15:25:00.000000000 +1000
+++ linux-ck-dev/mm/vmscan.c	2006-06-18 15:25:02.000000000 +1000
@@ -897,6 +897,38 @@ static unsigned long shrink_zone(int pri
 }
 
 /*
+ * Helper functions to adjust nice level of kswapd, based on the priority of
+ * the task (p) that called it. If it is already higher priority we do not
+ * demote its nice level since it is still working on behalf of a higher
+ * priority task. With kernel threads we leave it at nice 0.
+ *
+ * We don't ever run kswapd real time, so if a real time task calls kswapd we
+ * set it to highest SCHED_NORMAL priority.
+ */
+static int effective_sc_prio(struct task_struct *p)
+{
+	if (likely(p->mm)) {
+		if (rt_task(p))
+			return -20;
+		return task_nice(p);
+	}
+	return 0;
+}
+
+static void set_kswapd_nice(task_t *kswapd, task_t *p, int active)
+{
+	long nice = effective_sc_prio(p);
+
+	if (task_nice(kswapd) > nice || !active)
+		set_user_nice(kswapd, nice);
+}
+
+static int sc_priority(struct task_struct *p)
+{
+	return (DEF_PRIORITY + (DEF_PRIORITY * effective_sc_prio(p) / 40));
+}
+
+/*
  * This is the direct reclaim path, for page-allocating processes.  We only
  * try to reclaim pages from zones which will satisfy the caller's allocation
  * request.
@@ -1266,6 +1298,7 @@ static int kswapd(void *p)
 			 */
 			order = new_order;
 		} else {
+			set_user_nice(tsk, 0);
 			schedule();
 			order = pgdat->kswapd_max_order;
 		}
@@ -1279,9 +1312,10 @@ static int kswapd(void *p)
 /*
  * A zone is low on free memory, so wake its kswapd task to service it.
  */
-void wakeup_kswapd(struct zone *zone, int order)
+void wakeup_kswapd(struct zone *zone, int order, struct task_struct *p)
 {
 	pg_data_t *pgdat;
+	int active;
 
 	if (!populated_zone(zone))
 		return;
@@ -1293,7 +1327,9 @@ void wakeup_kswapd(struct zone *zone, in
 		pgdat->kswapd_max_order = order;
 	if (!cpuset_zone_allowed(zone, __GFP_HARDWALL))
 		return;
-	if (!waitqueue_active(&pgdat->kswapd_wait))
+	active = waitqueue_active(&pgdat->kswapd_wait);
+	set_kswapd_nice(pgdat->kswapd, p, active);
+	if (!active)
 		return;
 	wake_up_interruptible(&pgdat->kswapd_wait);
 }
1	niro	199	When kswapd is awoken due to reclaim by a running task, set the priority of
2			kswapd to that of the calling task thus making memory reclaim cpu activity
3			affected by nice level.
4
5			Signed-off-by: Con Kolivas <kernel@kolivas.org>
6
7			include/linux/mmzone.h \| 2 +-
8			mm/page_alloc.c \| 2 +-
9			mm/vmscan.c \| 40 ++++++++++++++++++++++++++++++++++++++--
10			3 files changed, 40 insertions(+), 4 deletions(-)
11
12			Index: linux-ck-dev/include/linux/mmzone.h
13			===================================================================
14			--- linux-ck-dev.orig/include/linux/mmzone.h 2006-06-18 15:25:00.000000000 +1000
15			+++ linux-ck-dev/include/linux/mmzone.h 2006-06-18 15:25:02.000000000 +1000
16			@@ -330,7 +330,7 @@ void __get_zone_counts(unsigned long *ac
17			void get_zone_counts(unsigned long active, unsigned long inactive,
18			unsigned long *free);
19			void build_all_zonelists(void);
20			-void wakeup_kswapd(struct zone *zone, int order);
21			+void wakeup_kswapd(struct zone zone, int order, struct task_struct p);
22			int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
23			int classzone_idx, int alloc_flags);
24
25			Index: linux-ck-dev/mm/page_alloc.c
26			===================================================================
27			--- linux-ck-dev.orig/mm/page_alloc.c 2006-06-18 15:25:00.000000000 +1000
28			+++ linux-ck-dev/mm/page_alloc.c 2006-06-18 15:25:02.000000000 +1000
29			@@ -952,7 +952,7 @@ restart:
30
31			do {
32			if (cpuset_zone_allowed(*z, gfp_mask\|__GFP_HARDWALL))
33			- wakeup_kswapd(*z, order);
34			+ wakeup_kswapd(*z, order, p);
35			} while (*(++z));
36
37			/*
38			Index: linux-ck-dev/mm/vmscan.c
39			===================================================================
40			--- linux-ck-dev.orig/mm/vmscan.c 2006-06-18 15:25:00.000000000 +1000
41			+++ linux-ck-dev/mm/vmscan.c 2006-06-18 15:25:02.000000000 +1000
42			@@ -897,6 +897,38 @@ static unsigned long shrink_zone(int pri
43			}
44
45			/*
46			+ * Helper functions to adjust nice level of kswapd, based on the priority of
47			+ * the task (p) that called it. If it is already higher priority we do not
48			+ * demote its nice level since it is still working on behalf of a higher
49			+ * priority task. With kernel threads we leave it at nice 0.
50			+ *
51			+ * We don't ever run kswapd real time, so if a real time task calls kswapd we
52			+ * set it to highest SCHED_NORMAL priority.
53			+ */
54			+static int effective_sc_prio(struct task_struct *p)
55			+{
56			+ if (likely(p->mm)) {
57			+ if (rt_task(p))
58			+ return -20;
59			+ return task_nice(p);
60			+ }
61			+ return 0;
62			+}
63			+
64			+static void set_kswapd_nice(task_t kswapd, task_t p, int active)
65			+{
66			+ long nice = effective_sc_prio(p);
67			+
68			+ if (task_nice(kswapd) > nice \|\| !active)
69			+ set_user_nice(kswapd, nice);
70			+}
71			+
72			+static int sc_priority(struct task_struct *p)
73			+{
74			+ return (DEF_PRIORITY + (DEF_PRIORITY * effective_sc_prio(p) / 40));
75			+}
76			+
77			+/*
78			* This is the direct reclaim path, for page-allocating processes. We only
79			* try to reclaim pages from zones which will satisfy the caller's allocation
80			* request.
81			@@ -1266,6 +1298,7 @@ static int kswapd(void *p)
82			*/
83			order = new_order;
84			} else {
85			+ set_user_nice(tsk, 0);
86			schedule();
87			order = pgdat->kswapd_max_order;
88			}
89			@@ -1279,9 +1312,10 @@ static int kswapd(void *p)
90			/*
91			* A zone is low on free memory, so wake its kswapd task to service it.
92			*/
93			-void wakeup_kswapd(struct zone *zone, int order)
94			+void wakeup_kswapd(struct zone zone, int order, struct task_struct p)
95			{
96			pg_data_t *pgdat;
97			+ int active;
98
99			if (!populated_zone(zone))
100			return;
101			@@ -1293,7 +1327,9 @@ void wakeup_kswapd(struct zone *zone, in
102			pgdat->kswapd_max_order = order;
103			if (!cpuset_zone_allowed(zone, __GFP_HARDWALL))
104			return;
105			- if (!waitqueue_active(&pgdat->kswapd_wait))
106			+ active = waitqueue_active(&pgdat->kswapd_wait);
107			+ set_kswapd_nice(pgdat->kswapd, p, active);
108			+ if (!active)
109			return;
110			wake_up_interruptible(&pgdat->kswapd_wait);
111			}