[Patch] New zone ZONE_EASY_RECLAIM take 2[0/5]

[Patch] New zone ZONE_EASY_RECLAIM take 2[0/5]

[Yasunori Goto]

Hello.

I rewrote ZONE_EASY_RECLAIM patches.
It aimed to collect the page which was difficult for removing on a few nodes,
and it made other nodes this zone to be removed easier.


In -mm tree, ZONE_DMA32 is already included. So, I recreate 
ZONE_EASY_RECLAIM as 5th zone against 2.6.14-mm1. It is difference of
previous one.

And, I'll post another patch to solve how to allocate ZONE_EASY_RECLAIM
after these patch. But, it is not direct way for it. So, I would like
to divide from these patche.

Please comment.

-- 
Yasunori Goto 

[Patch] New zone ZONE_EASY_RECLAIM take 2[1/5]

[Yasunori Goto]

This defines __GFP flag for new zone with GFP_DMA32.


Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>

Index: new_zone_mm/include/linux/gfp.h
=================================--- new_zone_mm.orig/include/linux/gfp.h	2005-11-17 16:47:04.000000000 +0900
+++ new_zone_mm/include/linux/gfp.h	2005-11-17 17:29:16.000000000 +0900
@@ -16,10 +16,13 @@ struct vm_area_struct;
 #define __GFP_HIGHMEM	((__force gfp_t)0x02u)
 #ifdef CONFIG_DMA_IS_DMA32
 #define __GFP_DMA32	((__force gfp_t)0x01)	/* ZONE_DMA is ZONE_DMA32 */
+#define __GFP_EASY_RECLAIM ((__force gfp_t)0x04u)
 #elif BITS_PER_LONG < 64
 #define __GFP_DMA32	((__force gfp_t)0x00)	/* ZONE_NORMAL is ZONE_DMA32 */
+#define __GFP_EASY_RECLAIM ((__force gfp_t)0x04u)
 #else
 #define __GFP_DMA32	((__force gfp_t)0x04)	/* Has own ZONE_DMA32 */
+#define __GFP_EASY_RECLAIM ((__force gfp_t)0x08u)
 #endif
 
 /*
@@ -66,7 +69,7 @@ struct vm_area_struct;
 #define GFP_USER	(__GFP_VALID | __GFP_WAIT | __GFP_IO | __GFP_FS | \
 				__GFP_HARDWALL)
 #define GFP_HIGHUSER	(__GFP_VALID | __GFP_WAIT | __GFP_IO | __GFP_FS | \
-				__GFP_HIGHMEM | __GFP_HARDWALL)
+				__GFP_HIGHMEM | __GFP_HARDWALL | __GFP_EASY_RECLAIM)
 
 /* Flag - indicates that the buffer will be suitable for DMA.  Ignored on some
    platforms, used as appropriate on others */
Index: new_zone_mm/include/linux/mmzone.h
=================================--- new_zone_mm.orig/include/linux/mmzone.h	2005-11-17 16:47:04.000000000 +0900
+++ new_zone_mm/include/linux/mmzone.h	2005-11-17 17:29:01.000000000 +0900
@@ -92,7 +92,7 @@ struct per_cpu_pageset {
  * be 8 (2 ** 3) zonelists.  GFP_ZONETYPES defines the number of possible
  * combinations of zone modifiers in "zone modifier space".
  */
-#define GFP_ZONEMASK	0x03
+#define GFP_ZONEMASK	0x0f
 
 /*
  * As an optimisation any zone modifier bits which are only valid when

-- 
Yasunori Goto 

[Patch] New zone ZONE_EASY_RECLAIM take 2[2/5]

[Yasunori Goto]

This defines new zone ZONE_EASY_RECLAIM.
ZONES_SHIFT becomes 3.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>

Index: new_zone_mm/include/linux/mmzone.h
=================================--- new_zone_mm.orig/include/linux/mmzone.h	2005-11-17 16:57:15.000000000 +0900
+++ new_zone_mm/include/linux/mmzone.h	2005-11-17 17:07:30.000000000 +0900
@@ -74,9 +74,10 @@ struct per_cpu_pageset {
 #define ZONE_DMA32		1
 #define ZONE_NORMAL		2
 #define ZONE_HIGHMEM		3
+#define ZONE_EASY_RECLAIM	4
 
-#define MAX_NR_ZONES		4	/* Sync this with ZONES_SHIFT */
-#define ZONES_SHIFT		2	/* ceil(log2(MAX_NR_ZONES)) */
+#define MAX_NR_ZONES		5	/* Sync this with ZONES_SHIFT */
+#define ZONES_SHIFT		3	/* ceil(log2(MAX_NR_ZONES)) */
 
 
 /*
Index: new_zone_mm/mm/page_alloc.c
=================================--- new_zone_mm.orig/mm/page_alloc.c	2005-11-17 17:05:12.000000000 +0900
+++ new_zone_mm/mm/page_alloc.c	2005-11-17 17:08:17.000000000 +0900
@@ -75,7 +75,7 @@ EXPORT_SYMBOL(totalram_pages);
 struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
 EXPORT_SYMBOL(zone_table);
 
-static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
+static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem", "Easy Reclaim"};
 int min_free_kbytes = 1024;
 
 unsigned long __initdata nr_kernel_pages;

-- 
Yasunori Goto 

[Patch] New zone ZONE_EASY_RECLAIM take 2[3/5]

[Yasunori Goto]

This is changing build_zonelists for new zone.

__GFP_xxxs are flag for requires of page allocation which zone
is prefered. But, it is used as an index number for zonelists[] too.
But after my patch, __GFP_xxx might be set at same time. So,
last set bit number of __GFP is recognized for zonelists' index
by this patch.


Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>

Index: new_zone_mm/mm/page_alloc.c
=================================--- new_zone_mm.orig/mm/page_alloc.c	2005-11-28 16:12:53.000000000 +0900
+++ new_zone_mm/mm/page_alloc.c	2005-11-28 16:15:55.000000000 +0900
@@ -1498,6 +1498,10 @@ static int __init build_zonelists_node(p
 		struct zone *zone;
 	default:
 		BUG();
+	case ZONE_EASY_RECLAIM:
+		zone = pgdat->node_zones + ZONE_EASY_RECLAIM;
+		if (zone->present_pages)
+			zonelist->zones[j++] = zone;
 	case ZONE_HIGHMEM:
 		zone = pgdat->node_zones + ZONE_HIGHMEM;
 		if (zone->present_pages) {
@@ -1526,11 +1530,14 @@ static int __init build_zonelists_node(p
 static inline int highest_zone(int zone_bits)
 {
 	int res = ZONE_NORMAL;
-	if (zone_bits & (__force int)__GFP_HIGHMEM)
+
+	if (zone_bits = fls((__force int)__GFP_EASY_RECLAIM))
+		res = ZONE_EASY_RECLAIM;
+	else if (zone_bits = fls((__force int)__GFP_HIGHMEM))
 		res = ZONE_HIGHMEM;
-	if (zone_bits & (__force int)__GFP_DMA32)
+	else if (zone_bits = fls((__force int)__GFP_DMA32))
 		res = ZONE_DMA32;
-	if (zone_bits & (__force int)__GFP_DMA)
+	else if (zone_bits = fls((__force int)__GFP_DMA))
 		res = ZONE_DMA;
 	return res;
 }
Index: new_zone_mm/include/linux/gfp.h
=================================--- new_zone_mm.orig/include/linux/gfp.h	2005-11-28 16:15:48.000000000 +0900
+++ new_zone_mm/include/linux/gfp.h	2005-11-28 16:15:55.000000000 +0900
@@ -79,8 +79,10 @@ struct vm_area_struct;
 /* 4GB DMA on some platforms */
 #define GFP_DMA32	__GFP_DMA32
 
-
-#define gfp_zone(mask) ((__force int)((mask) & (__force gfp_t)GFP_ZONEMASK))
+static inline unsigned int gfp_zone(unsigned int mask)
+{
+	return fls(mask & GFP_ZONEMASK);
+}
 
 /*
  * There is only one page-allocator function, and two main namespaces to

-- 
Yasunori Goto 

[Patch] New zone ZONE_EASY_RECLAIM take 2[4/5]

[Yasunori Goto]

This is for calculation of the watermark zone->pages_min/low/high.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>

Index: new_zone_mm/include/linux/mmzone.h
=================================--- new_zone_mm.orig/include/linux/mmzone.h	2005-11-17 17:07:30.000000000 +0900
+++ new_zone_mm/include/linux/mmzone.h	2005-11-17 17:17:51.000000000 +0900
@@ -401,6 +401,11 @@ static inline struct zone *next_zone(str
 #define for_each_zone(zone) \
 	for (zone = pgdat_list->node_zones; zone; zone = next_zone(zone))
 
+static inline int is_easy_reclaim_idx(int idx)
+{
+	return (idx = ZONE_EASY_RECLAIM);
+}
+
 static inline int is_highmem_idx(int idx)
 {
 	return (idx = ZONE_HIGHMEM);
@@ -416,6 +421,11 @@ static inline int is_normal_idx(int idx)
  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
  * @zone - pointer to struct zone variable
  */
+static inline int is_easy_reclaim(struct zone *zone)
+{
+	return zone = zone->zone_pgdat->node_zones + ZONE_EASY_RECLAIM;
+}
+
 static inline int is_highmem(struct zone *zone)
 {
 	return zone = zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
Index: new_zone_mm/mm/page_alloc.c
=================================--- new_zone_mm.orig/mm/page_alloc.c	2005-11-17 17:09:05.000000000 +0900
+++ new_zone_mm/mm/page_alloc.c	2005-11-17 17:17:51.000000000 +0900
@@ -2495,13 +2495,13 @@ void setup_per_zone_pages_min(void)
 
 	/* Calculate total number of !ZONE_HIGHMEM pages */
 	for_each_zone(zone) {
-		if (!is_highmem(zone))
+		if (!is_highmem(zone) && !is_easy_reclaim(zone))
 			lowmem_pages += zone->present_pages;
 	}
 
 	for_each_zone(zone) {
 		spin_lock_irqsave(&zone->lru_lock, flags);
-		if (is_highmem(zone)) {
+		if (is_highmem(zone) || is_easy_reclaim(zone)) {
 			/*
 			 * Often, highmem doesn't need to reserve any pages.
 			 * But the pages_min/low/high values are also used for

-- 
Yasunori Goto 

[Patch] New zone ZONE_EASY_RECLAIM take 2[5/5]

[Yasunori Goto]

This is to disable __GFP_EASY_RECLAIM flag at add_to_page_cache().
If this patch is not applied, cache_grow() checks and call BUG(),
at here. 

	if (flags & ~(SLAB_DMA|SLAB_LEVEL_MASK|SLAB_NO_GROW))
		BUG();

This patch is to solve it.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>

Index: new_zone_mm/mm/filemap.c
=================================--- new_zone_mm.orig/mm/filemap.c	2005-11-22 15:21:23.000000000 +0900
+++ new_zone_mm/mm/filemap.c	2005-11-22 15:21:27.000000000 +0900
@@ -381,7 +381,7 @@ int filemap_write_and_wait_range(struct 
 int add_to_page_cache(struct page *page, struct address_space *mapping,
 		pgoff_t offset, gfp_t gfp_mask)
 {
-	int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
+	int error = radix_tree_preload(gfp_mask & ~(__GFP_HIGHMEM | __GFP_EASY_RECLAIM));
 
 	if (error = 0) {
 		write_lock_irq(&mapping->tree_lock);

-- 
Yasunori Goto