kernel26-alx/patches-2.6.21-r14/0014-2.6.21-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            |   36 ++++++++++++++++++++++++++++++++++--
 3 files changed, 36 insertions(+), 4 deletions(-)

Index: linux-2.6.21-ck2/include/linux/mmzone.h
===================================================================
--- linux-2.6.21-ck2.orig/include/linux/mmzone.h	2007-05-14 19:49:55.000000000 +1000
+++ linux-2.6.21-ck2/include/linux/mmzone.h	2007-05-14 19:49:55.000000000 +1000
@@ -465,7 +465,7 @@ typedef struct pglist_data {
 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);
 enum memmap_context {
Index: linux-2.6.21-ck2/mm/page_alloc.c
===================================================================
--- linux-2.6.21-ck2.orig/mm/page_alloc.c	2007-05-14 19:49:55.000000000 +1000
+++ linux-2.6.21-ck2/mm/page_alloc.c	2007-05-14 19:49:55.000000000 +1000
@@ -1277,7 +1277,7 @@ restart:
 		goto nopage;
 
 	for (z = zonelist->zones; *z; z++)
-		wakeup_kswapd(*z, order);
+		wakeup_kswapd(*z, order, p);
 
 	/*
 	 * OK, we're below the kswapd watermark and have kicked background
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
@@ -966,6 +966,34 @@ 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(struct task_struct *kswapd, struct task_struct *p,
+			    int active)
+{
+	long nice = effective_sc_prio(p);
+
+	if (task_nice(kswapd) > nice || !active)
+		set_user_nice(kswapd, nice);
+}
+
+/*
  * 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.
@@ -1355,6 +1383,7 @@ static int kswapd(void *p)
 			 */
 			order = new_order;
 		} else {
+			set_user_nice(tsk, 0);
 			schedule();
 			order = pgdat->kswapd_max_order;
 		}
@@ -1368,9 +1397,10 @@ 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;
@@ -1382,7 +1412,9 @@ void wakeup_kswapd(struct zone *zone, in
 		pgdat->kswapd_max_order = order;
 	if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
 		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	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 \| 36 ++++++++++++++++++++++++++++++++++--
10	3 files changed, 36 insertions(+), 4 deletions(-)
11
12	Index: linux-2.6.21-ck2/include/linux/mmzone.h
13	===================================================================
14	--- linux-2.6.21-ck2.orig/include/linux/mmzone.h 2007-05-14 19:49:55.000000000 +1000
15	+++ linux-2.6.21-ck2/include/linux/mmzone.h 2007-05-14 19:49:55.000000000 +1000
16	@@ -465,7 +465,7 @@ typedef struct pglist_data {
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	enum memmap_context {
25	Index: linux-2.6.21-ck2/mm/page_alloc.c
26	===================================================================
27	--- linux-2.6.21-ck2.orig/mm/page_alloc.c 2007-05-14 19:49:55.000000000 +1000
28	+++ linux-2.6.21-ck2/mm/page_alloc.c 2007-05-14 19:49:55.000000000 +1000
29	@@ -1277,7 +1277,7 @@ restart:
30	goto nopage;
31
32	for (z = zonelist->zones; *z; z++)
33	- wakeup_kswapd(*z, order);
34	+ wakeup_kswapd(*z, order, p);
35
36	/*
37	* OK, we're below the kswapd watermark and have kicked background
38	Index: linux-2.6.21-ck2/mm/vmscan.c
39	===================================================================
40	--- linux-2.6.21-ck2.orig/mm/vmscan.c 2007-05-14 19:49:55.000000000 +1000
41	+++ linux-2.6.21-ck2/mm/vmscan.c 2007-05-14 19:49:55.000000000 +1000
42	@@ -966,6 +966,34 @@ 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(struct task_struct kswapd, struct task_struct p,
65	+ int active)
66	+{
67	+ long nice = effective_sc_prio(p);
68	+
69	+ if (task_nice(kswapd) > nice \|\| !active)
70	+ set_user_nice(kswapd, nice);
71	+}
72	+
73	+/*
74	* This is the direct reclaim path, for page-allocating processes. We only
75	* try to reclaim pages from zones which will satisfy the caller's allocation
76	* request.
77	@@ -1355,6 +1383,7 @@ static int kswapd(void *p)
78	*/
79	order = new_order;
80	} else {
81	+ set_user_nice(tsk, 0);
82	schedule();
83	order = pgdat->kswapd_max_order;
84	}
85	@@ -1368,9 +1397,10 @@ static int kswapd(void *p)
86	/*
87	* A zone is low on free memory, so wake its kswapd task to service it.
88	*/
89	-void wakeup_kswapd(struct zone *zone, int order)
90	+void wakeup_kswapd(struct zone zone, int order, struct task_struct p)
91	{
92	pg_data_t *pgdat;
93	+ int active;
94
95	if (!populated_zone(zone))
96	return;
97	@@ -1382,7 +1412,9 @@ void wakeup_kswapd(struct zone *zone, in
98	pgdat->kswapd_max_order = order;
99	if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL))
100	return;
101	- if (!waitqueue_active(&pgdat->kswapd_wait))
102	+ active = waitqueue_active(&pgdat->kswapd_wait);
103	+ set_kswapd_nice(pgdat->kswapd, p, active);
104	+ if (!active)
105	return;
106	wake_up_interruptible(&pgdat->kswapd_wait);
107	}