[RFC} memory unplug patchset prep [0/16]

[RFC} memory unplug patchset prep [0/16]

[KAMEZAWA Hiroyuki]

Hi, this is memory-unplug patch set.......sample.

My purpose is to show how memory-unplug can be implemented on ZONE_MOVABLE.
*Any* comments are welcome. This patch just support ia64, which I can test.
If you want me to add arch support, please mail me.

This patch is a bit old and against 2.6.20-mm2. I'll rebase this and reflect
your comments in the next post (may not soon).
Well booted on ia64 and passed *quick* memory offline test.

I know there are many other works....
- mlocked page off-lru
- anon-witout-swap off-lru
- change zone/node relation ship and create new generic memory management unit
- etc..
So, I think I will have to change amount of part of this patch set.

Anyway, I can't go ahead without some memory encapsulation like ZONE_MOVABLE.

TODO
- add hot-added memory to ZONE_MOVABLE
- show physical memory <-> memory section relation ship.
- node hotplug support. (this may needs some amount of patches.)
- more GFP_MOVABLE checks. Disallow not-movable pages in ZONE_MOVABLE.
- test under heavy work load and more careful race check.
- devide memory unplug core patch into smaller pieces.
- Fix where we should allocate migration target page from.
- more sophisticated boot options. algorythms, controls.
- fix memory encapsulation technique like ZONE_MOVABLE 


Patch series[1]-[16]
== cleanup ==
These patches offerres no functional changes. just changes code.
I like this patch set as clean-up.
[1] zone naming clean up. --- changes #ifdefs around ZONE.
[2] alloc_zeroed_user_highpage clean up --- gathered scatterred defintions.
[3] is_highmem cleaunp --- added is_identity_map() 

== zone movable ==
This ZONE_MOVABLE is based on Mel's ZONE_MOVABLE. some implementation
details are different from his, because of clean-ups[1][2][3].
It seems that pushing ZONE_MOVABLE now is not easy. We(I?)'ll have to
make changes to memory management core for implementing better memory
management structure than zones.
[4] add ZONE_MOVABLE patch
[5] GFP_MOVABLE definition patch
[6] change alloc_zeroed_user_highpage to use ZONE_MOVABLE
[7] change GFP_HIGHUSER callers to use GFP_HIGH_MOVABLE (if they can)
[8] counters patch -- per-zone counter for ZONE_MOVABLE
[9] movable_zone_create patch --- added boot option for creating movable zones.
                                  maybe need more work.
[10] ia64_movable -- creating ZONE_MOVABLE zone on ia64


==page isolation==
page isolation makes page to be *unused* state. Make free pages off-free_area[]
and hide them from system.
[11] page isolation patch ..... basic defintions of page isolation.
[12] drain_all_zone_pages patch ..... drain all cpus' pcp pages.
[13] isolate freed page patch ..... isolate pages in free_area[]

==memory unplug==
offline a section of pages. isolate specified section and migrate contents
of used pages to out of section. (Because free pages in a section is isolated,
it never be returned by alloc_pages())
This patch doesn't care where we should allocate migration new pages from.
[14] memory unplug core patch --- maybe need more work.
[15] interface patch          --- "offline" interface support 

==migration nocontext==
Fix race condition of page migration without process context (not taking mm->sem).
This patch delayes kmem_cache_free() of anon_vma until migration ends.
[16] migration nocontext patch --- support page migration without acquiring mm->sem.
                                   need careful debug...


-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [1/16] zone ids cleanup

[KAMEZAWA Hiroyuki]

OThis patch defines ZONE_DMA,DMA32,HIGHMEM on *any* config.
MAX_NR_ZONES is unchanged and not-configured zones's id is greater than it.
Now, you can check zone is configured or not by (zone_id < MAX_NR_ZONES).

Good bye #ifdefs. Compiler will do enough work, I think.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 include/linux/mmzone.h |   33 ++++++++++++------------
 mm/page_alloc.c        |   65 ++++++++++++++++++++++++++-----------------------
 2 files changed, 51 insertions(+), 47 deletions(-)

Index: devel-tree-2.6.20-mm2/include/linux/mmzone.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/mmzone.h
+++ devel-tree-2.6.20-mm2/include/linux/mmzone.h
@@ -142,9 +142,24 @@ enum zone_type {
 	 */
 	ZONE_HIGHMEM,
 #endif
-	MAX_NR_ZONES
+	MAX_NR_ZONES,
+#ifndef CONFIG_ZONE_DMA
+	ZONE_DMA,
+#endif
+#ifndef CONFIG_ZONE_DMA32
+	ZONE_DMA32,
+#endif
+#ifndef CONFIG_HIGHMEM
+	ZONE_HIGHMEM,
+#endif
+	MAX_POSSIBLE_ZONES
 };
 
+static inline int is_configured_zone(enum zone_type type)
+{
+	return (type < MAX_NR_ZONES);
+}
+
 /*
  * When a memory allocation must conform to specific limitations (such
  * as being suitable for DMA) the caller will pass in hints to the
@@ -500,11 +515,7 @@ static inline int populated_zone(struct 
 
 static inline int is_highmem_idx(enum zone_type idx)
 {
-#ifdef CONFIG_HIGHMEM
 	return (idx == ZONE_HIGHMEM);
-#else
-	return 0;
-#endif
 }
 
 static inline int is_normal_idx(enum zone_type idx)
@@ -520,11 +531,7 @@ static inline int is_normal_idx(enum zon
  */
 static inline int is_highmem(struct zone *zone)
 {
-#ifdef CONFIG_HIGHMEM
 	return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
-#else
-	return 0;
-#endif
 }
 
 static inline int is_normal(struct zone *zone)
@@ -534,20 +541,12 @@ static inline int is_normal(struct zone 
 
 static inline int is_dma32(struct zone *zone)
 {
-#ifdef CONFIG_ZONE_DMA32
 	return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
-#else
-	return 0;
-#endif
 }
 
 static inline int is_dma(struct zone *zone)
 {
-#ifdef CONFIG_ZONE_DMA
 	return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
-#else
-	return 0;
-#endif
 }
 
 /* These two functions are used to setup the per zone pages min values */
Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
+++ devel-tree-2.6.20-mm2/mm/page_alloc.c
@@ -72,32 +72,34 @@ static void __free_pages_ok(struct page 
  * TBD: should special case ZONE_DMA32 machines here - in those we normally
  * don't need any ZONE_NORMAL reservation
  */
-int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
-#ifdef CONFIG_ZONE_DMA
-	 256,
-#endif
-#ifdef CONFIG_ZONE_DMA32
-	 256,
-#endif
-#ifdef CONFIG_HIGHMEM
-	 32
-#endif
-};
+int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
 
 EXPORT_SYMBOL(totalram_pages);
 
-static char * const zone_names[MAX_NR_ZONES] = {
-#ifdef CONFIG_ZONE_DMA
-	 "DMA",
-#endif
-#ifdef CONFIG_ZONE_DMA32
-	 "DMA32",
-#endif
-	 "Normal",
-#ifdef CONFIG_HIGHMEM
-	 "HighMem"
-#endif
-};
+static char *zone_names[MAX_POSSIBLE_ZONES];
+
+static char name_dma[] = "DMA";
+static char name_dma32[] = "DMA32";
+static char name_normal[] = "Normal";
+static char name_highmem[] = "Highmem";
+
+static inline void __meminit zone_variables_init(void)
+{
+	if (zone_names[0] != NULL)
+		return;
+	zone_names[ZONE_DMA] = name_dma;
+	zone_names[ZONE_DMA32] = name_dma32;
+	zone_names[ZONE_NORMAL] = name_normal;
+	zone_names[ZONE_HIGHMEM] = name_highmem;
+
+	/* ZONE below NORAML has ratio 256 */
+	if (is_configured_zone(ZONE_DMA))
+		sysctl_lowmem_reserve_ratio[ZONE_DMA] = 256;
+	if (is_configured_zone(ZONE_DMA32))
+		sysctl_lowmem_reserve_ratio[ZONE_DMA32] = 256;
+	if (is_configured_zone(ZONE_HIGHMEM))
+		sysctl_lowmem_reserve_ratio[ZONE_HIGHMEM] = 32;
+}
 
 int min_free_kbytes = 1024;
 
@@ -1638,14 +1640,16 @@ void si_meminfo_node(struct sysinfo *val
 
 	val->totalram = pgdat->node_present_pages;
 	val->freeram = node_page_state(nid, NR_FREE_PAGES);
-#ifdef CONFIG_HIGHMEM
-	val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
-	val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
+	if (is_configured_zone(ZONE_HIGHMEM)) {
+		val->totalhigh =
+			pgdat->node_zones[ZONE_HIGHMEM].present_pages;
+		val->freehigh =
+			zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
 			NR_FREE_PAGES);
-#else
-	val->totalhigh = 0;
-	val->freehigh = 0;
-#endif
+	} else {
+		val->totalhigh = 0;
+		val->freehigh = 0;
+	}
 	val->mem_unit = PAGE_SIZE;
 }
 #endif
@@ -3048,7 +3052,8 @@ void __init free_area_init_nodes(unsigne
 {
 	unsigned long nid;
 	enum zone_type i;
-
+	/* Parameter Setup */
+	zone_variables_init();
 	/* Sort early_node_map as initialisation assumes it is sorted */
 	sort_node_map();
 
@@ -3119,6 +3124,7 @@ EXPORT_SYMBOL(contig_page_data);
 
 void __init free_area_init(unsigned long *zones_size)
 {
+	zone_variables_init();
 	free_area_init_node(0, NODE_DATA(0), zones_size,
 			__pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
 }

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [2/16] gathering alloc_zeroed_user_highpage()

[KAMEZAWA Hiroyuki]

Definitions of alloc_zeroed_user_highpage() is scattered.
This patch gathers them to linux/highmem.h

To do so, added CONFIG_ARCH_HAS_PREZERO_USERPAGE and
CONFIG_ARCH_HAS_FLUSH_USERNEWZEROPAGE.

If you know better config name, please tell me.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 arch/alpha/Kconfig           |    3 +++
 arch/cris/Kconfig            |    3 +++
 arch/h8300/Kconfig           |    4 ++++
 arch/i386/Kconfig            |    3 +++
 arch/ia64/Kconfig            |    6 ++++++
 arch/m32r/Kconfig            |    3 +++
 arch/m68knommu/Kconfig       |    3 +++
 arch/s390/Kconfig            |    3 +++
 arch/x86_64/Kconfig          |    3 +++
 include/asm-alpha/page.h     |    3 ---
 include/asm-cris/page.h      |    3 ---
 include/asm-h8300/page.h     |    3 ---
 include/asm-i386/page.h      |    3 ---
 include/asm-ia64/page.h      |   10 +---------
 include/asm-m32r/page.h      |    3 ---
 include/asm-m68knommu/page.h |    3 ---
 include/asm-s390/page.h      |    2 --
 include/asm-x86_64/page.h    |    2 --
 include/linux/highmem.h      |   16 +++++++++++++++-
 19 files changed, 47 insertions(+), 32 deletions(-)

Index: devel-tree-2.6.20-mm2/arch/alpha/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/alpha/Kconfig
+++ devel-tree-2.6.20-mm2/arch/alpha/Kconfig
@@ -551,6 +551,9 @@ config ARCH_DISCONTIGMEM_ENABLE
 	  or have huge holes in the physical address space for other reasons.
 	  See <file:Documentation/vm/numa> for more.
 
+config ARCH_HAS_PREZERO_USERPAGE
+	def_bool y
+
 source "mm/Kconfig"
 
 config NUMA
Index: devel-tree-2.6.20-mm2/include/asm-alpha/page.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/asm-alpha/page.h
+++ devel-tree-2.6.20-mm2/include/asm-alpha/page.h
@@ -17,9 +17,6 @@
 extern void clear_page(void *page);
 #define clear_user_page(page, vaddr, pg)	clear_page(page)
 
-#define alloc_zeroed_user_highpage(vma, vaddr) alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vmaddr)
-#define __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
-
 extern void copy_page(void * _to, void * _from);
 #define copy_user_page(to, from, vaddr, pg)	copy_page(to, from)
 
Index: devel-tree-2.6.20-mm2/arch/cris/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/cris/Kconfig
+++ devel-tree-2.6.20-mm2/arch/cris/Kconfig
@@ -97,6 +97,9 @@ config PREEMPT
 	  Say Y here if you are building a kernel for a desktop, embedded
 	  or real-time system.  Say N if you are unsure.
 
+config ARCH_HAS_PREZERO_USERPAGE
+	def_bool y
+
 source mm/Kconfig
 
 endmenu
Index: devel-tree-2.6.20-mm2/include/asm-cris/page.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/asm-cris/page.h
+++ devel-tree-2.6.20-mm2/include/asm-cris/page.h
@@ -20,9 +20,6 @@
 #define clear_user_page(page, vaddr, pg)    clear_page(page)
 #define copy_user_page(to, from, vaddr, pg) copy_page(to, from)
 
-#define alloc_zeroed_user_highpage(vma, vaddr) alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr)
-#define __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
-
 /*
  * These are used to make use of C type-checking..
  */
Index: devel-tree-2.6.20-mm2/arch/h8300/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/h8300/Kconfig
+++ devel-tree-2.6.20-mm2/arch/h8300/Kconfig
@@ -68,6 +68,10 @@ config PCI
 	bool
 	default n
 
+config ARCH_HAS_PREZERO_USERPAGE
+	bool
+	default y
+
 source "init/Kconfig"
 
 source "arch/h8300/Kconfig.cpu"
Index: devel-tree-2.6.20-mm2/include/asm-h8300/page.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/asm-h8300/page.h
+++ devel-tree-2.6.20-mm2/include/asm-h8300/page.h
@@ -22,9 +22,6 @@
 #define clear_user_page(page, vaddr, pg)	clear_page(page)
 #define copy_user_page(to, from, vaddr, pg)	copy_page(to, from)
 
-#define alloc_zeroed_user_highpage(vma, vaddr) alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr)
-#define __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
-
 /*
  * These are used to make use of C type-checking..
  */
Index: devel-tree-2.6.20-mm2/arch/i386/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/i386/Kconfig
+++ devel-tree-2.6.20-mm2/arch/i386/Kconfig
@@ -675,6 +675,9 @@ config ARCH_SELECT_MEMORY_MODEL
 config ARCH_POPULATES_NODE_MAP
 	def_bool y
 
+config ARCH_HAS_PREZERO_USERPAGE
+	def_bool y
+
 source "mm/Kconfig"
 
 config HIGHPTE
Index: devel-tree-2.6.20-mm2/include/asm-i386/page.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/asm-i386/page.h
+++ devel-tree-2.6.20-mm2/include/asm-i386/page.h
@@ -34,9 +34,6 @@
 #define clear_user_page(page, vaddr, pg)	clear_page(page)
 #define copy_user_page(to, from, vaddr, pg)	copy_page(to, from)
 
-#define alloc_zeroed_user_highpage(vma, vaddr) alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr)
-#define __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
-
 /*
  * These are used to make use of C type-checking..
  */
Index: devel-tree-2.6.20-mm2/arch/ia64/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/ia64/Kconfig
+++ devel-tree-2.6.20-mm2/arch/ia64/Kconfig
@@ -329,6 +329,12 @@ config PREEMPT
           Say Y here if you are building a kernel for a desktop, embedded
           or real-time system.  Say N if you are unsure.
 
+config ARCH_HAS_PREZERO_USERPAGE
+	def_bool y
+
+config ARCH_HAS_FLUSH_USERNEWZEROPAGE
+	def_bool y
+
 source "mm/Kconfig"
 
 config ARCH_SELECT_MEMORY_MODEL
Index: devel-tree-2.6.20-mm2/include/asm-ia64/page.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/asm-ia64/page.h
+++ devel-tree-2.6.20-mm2/include/asm-ia64/page.h
@@ -87,15 +87,7 @@ do {						\
 } while (0)
 
 
-#define alloc_zeroed_user_highpage(vma, vaddr) \
-({						\
-	struct page *page = alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr); \
-	if (page)				\
- 		flush_dcache_page(page);	\
-	page;					\
-})
-
-#define __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
+#define flush_user_newzeropage(page)	flush_dcache_page(page)
 
 #define virt_addr_valid(kaddr)	pfn_valid(__pa(kaddr) >> PAGE_SHIFT)
 
Index: devel-tree-2.6.20-mm2/arch/m32r/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/m32r/Kconfig
+++ devel-tree-2.6.20-mm2/arch/m32r/Kconfig
@@ -193,6 +193,9 @@ config ARCH_DISCONTIGMEM_ENABLE
 	depends on CHIP_M32700 || CHIP_M32102 || CHIP_VDEC2 || CHIP_OPSP || CHIP_M32104
 	default y
 
+config ARCH_HAS_PREZERO_USERPAGE
+	def_bool	y
+
 source "mm/Kconfig"
 
 config IRAM_START
Index: devel-tree-2.6.20-mm2/include/asm-m32r/page.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/asm-m32r/page.h
+++ devel-tree-2.6.20-mm2/include/asm-m32r/page.h
@@ -15,9 +15,6 @@ extern void copy_page(void *to, void *fr
 #define clear_user_page(page, vaddr, pg)	clear_page(page)
 #define copy_user_page(to, from, vaddr, pg)	copy_page(to, from)
 
-#define alloc_zeroed_user_highpage(vma, vaddr) alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr)
-#define __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
-
 /*
  * These are used to make use of C type-checking..
  */
Index: devel-tree-2.6.20-mm2/arch/m68knommu/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/m68knommu/Kconfig
+++ devel-tree-2.6.20-mm2/arch/m68knommu/Kconfig
@@ -627,6 +627,9 @@ config ROMKERNEL
 
 endchoice
 
+config ARCH_HAS_PREZERO_USERPAGE
+	def_bool y
+
 source "mm/Kconfig"
 
 endmenu
Index: devel-tree-2.6.20-mm2/include/asm-m68knommu/page.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/asm-m68knommu/page.h
+++ devel-tree-2.6.20-mm2/include/asm-m68knommu/page.h
@@ -22,9 +22,6 @@
 #define clear_user_page(page, vaddr, pg)	clear_page(page)
 #define copy_user_page(to, from, vaddr, pg)	copy_page(to, from)
 
-#define alloc_zeroed_user_highpage(vma, vaddr) alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr)
-#define __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
-
 /*
  * These are used to make use of C type-checking..
  */
Index: devel-tree-2.6.20-mm2/arch/s390/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/s390/Kconfig
+++ devel-tree-2.6.20-mm2/arch/s390/Kconfig
@@ -272,6 +272,9 @@ config WARN_STACK_SIZE
 config ARCH_POPULATES_NODE_MAP
 	def_bool y
 
+config ARCH_HAS_PREZERO_USERPAGE
+	def_bool y
+
 source "mm/Kconfig"
 
 config HOLES_IN_ZONE
Index: devel-tree-2.6.20-mm2/include/asm-s390/page.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/asm-s390/page.h
+++ devel-tree-2.6.20-mm2/include/asm-s390/page.h
@@ -64,8 +64,6 @@ static inline void copy_page(void *to, v
 #define clear_user_page(page, vaddr, pg)	clear_page(page)
 #define copy_user_page(to, from, vaddr, pg)	copy_page(to, from)
 
-#define alloc_zeroed_user_highpage(vma, vaddr) alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr)
-#define __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
 
 /*
  * These are used to make use of C type-checking..
Index: devel-tree-2.6.20-mm2/arch/x86_64/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/x86_64/Kconfig
+++ devel-tree-2.6.20-mm2/arch/x86_64/Kconfig
@@ -400,6 +400,9 @@ config ARCH_FLATMEM_ENABLE
 	def_bool y
 	depends on !NUMA
 
+config ARCH_HAS_PREZERO_USERPAGE
+	def_bool y
+
 source "mm/Kconfig"
 
 config MEMORY_HOTPLUG_RESERVE
Index: devel-tree-2.6.20-mm2/include/asm-x86_64/page.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/asm-x86_64/page.h
+++ devel-tree-2.6.20-mm2/include/asm-x86_64/page.h
@@ -51,8 +51,6 @@ void copy_page(void *, void *);
 #define clear_user_page(page, vaddr, pg)	clear_page(page)
 #define copy_user_page(to, from, vaddr, pg)	copy_page(to, from)
 
-#define alloc_zeroed_user_highpage(vma, vaddr) alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr)
-#define __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
 /*
  * These are used to make use of C type-checking..
  */
Index: devel-tree-2.6.20-mm2/include/linux/highmem.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/highmem.h
+++ devel-tree-2.6.20-mm2/include/linux/highmem.h
@@ -60,8 +60,22 @@ static inline void clear_user_highpage(s
 	/* Make sure this page is cleared on other CPU's too before using it */
 	smp_wmb();
 }
+#ifndef CONFIG_ARCH_HAS_FLUSH_USER_NEWZEROPAGE
+#define flush_user_newzeroapge(page)	do{}while(0);
+#endif
 
-#ifndef __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE
+#ifdef CONFIG_ARCH_HAS_PREZERO_USERPAGE
+static inline struct page *
+alloc_zeroed_user_highpage(struct vm_area_struct *vma, unsigned long vaddr)
+{
+	struct page *page;
+	page = alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr);
+	if (page)
+		flush_user_newzeropage(page);
+	return page;
+}
+
+#else
 static inline struct page *
 alloc_zeroed_user_highpage(struct vm_area_struct *vma, unsigned long vaddr)
 {

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [3/16] define is_identity_mapped

[KAMEZAWA Hiroyuki]

Add is_identity_map() functon and rewrite is_highmem() user to
to use is_identity_map().

(*) prepare for adding extra zone ZONE_MOVABLE.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 include/linux/mmzone.h     |   10 ++++++++++
 include/linux/page-flags.h |    2 +-
 kernel/power/snapshot.c    |   12 ++++++------
 kernel/power/swsusp.c      |    2 +-
 mm/page_alloc.c            |    8 ++++----
 5 files changed, 22 insertions(+), 12 deletions(-)

Index: devel-tree-2.6.20-mm2/include/linux/mmzone.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/mmzone.h
+++ devel-tree-2.6.20-mm2/include/linux/mmzone.h
@@ -523,6 +523,13 @@ static inline int is_normal_idx(enum zon
 	return (idx == ZONE_NORMAL);
 }
 
+static inline int is_identity_map_idx(enum zone_type idx)
+{
+	if (is_configured_zone(ZONE_HIGHMEM))
+		return (idx < ZONE_HIGHMEM);
+	else
+		return 1;
+}
 /**
  * is_highmem - helper function to quickly check if a struct zone is a 
  *              highmem zone or not.  This is an attempt to keep references
@@ -549,6 +556,14 @@ static inline int is_dma(struct zone *zo
 	return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
 }
 
+static inline int is_identity_map(struct zone *zone)
+{
+	if (is_configured_zone(ZONE_HIGHMEM)
+		return zone_idx(zone) < ZONE_HIGHMEM;
+	else
+		return 1;
+}
+
 /* These two functions are used to setup the per zone pages min values */
 struct ctl_table;
 struct file;
Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
+++ devel-tree-2.6.20-mm2/mm/page_alloc.c
@@ -2090,7 +2090,7 @@ void __meminit memmap_init_zone(unsigned
 		INIT_LIST_HEAD(&page->lru);
 #ifdef WANT_PAGE_VIRTUAL
 		/* The shift won't overflow because ZONE_NORMAL is below 4G. */
-		if (!is_highmem_idx(zone))
+		if (is_identity_map_idx(zone))
 			set_page_address(page, __va(pfn << PAGE_SHIFT));
 #endif
 #ifdef CONFIG_PAGE_OWNER
@@ -2769,7 +2769,7 @@ static void __meminit free_area_init_cor
 					zone_names[0], dma_reserve);
 		}
 
-		if (!is_highmem_idx(j))
+		if (is_identity_map_idx(j))
 			nr_kernel_pages += realsize;
 		nr_all_pages += realsize;
 
@@ -3235,7 +3235,7 @@ void setup_per_zone_pages_min(void)
 
 	/* Calculate total number of !ZONE_HIGHMEM pages */
 	for_each_zone(zone) {
-		if (!is_highmem(zone))
+		if (is_identity_map(zone))
 			lowmem_pages += zone->present_pages;
 	}
 
@@ -3245,7 +3245,7 @@ void setup_per_zone_pages_min(void)
 		spin_lock_irqsave(&zone->lru_lock, flags);
 		tmp = (u64)pages_min * zone->present_pages;
 		do_div(tmp, lowmem_pages);
-		if (is_highmem(zone)) {
+		if (!is_identity_map(zone)) {
 			/*
 			 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
 			 * need highmem pages, so cap pages_min to a small
Index: devel-tree-2.6.20-mm2/include/linux/page-flags.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/page-flags.h
+++ devel-tree-2.6.20-mm2/include/linux/page-flags.h
@@ -162,7 +162,7 @@ static inline void SetPageUptodate(struc
 #define __ClearPageSlab(page)	__clear_bit(PG_slab, &(page)->flags)
 
 #ifdef CONFIG_HIGHMEM
-#define PageHighMem(page)	is_highmem(page_zone(page))
+#define PageHighMem(page)	(!is_identitiy_map(page_zone(page)))
 #else
 #define PageHighMem(page)	0 /* needed to optimize away at compile time */
 #endif
Index: devel-tree-2.6.20-mm2/kernel/power/snapshot.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/kernel/power/snapshot.c
+++ devel-tree-2.6.20-mm2/kernel/power/snapshot.c
@@ -590,7 +590,7 @@ static unsigned int count_free_highmem_p
 	unsigned int cnt = 0;
 
 	for_each_zone(zone)
-		if (populated_zone(zone) && is_highmem(zone))
+		if (populated_zone(zone) && !is_identity_map(zone))
 			cnt += zone_page_state(zone, NR_FREE_PAGES);
 
 	return cnt;
@@ -634,7 +634,7 @@ unsigned int count_highmem_pages(void)
 	for_each_zone(zone) {
 		unsigned long pfn, max_zone_pfn;
 
-		if (!is_highmem(zone))
+		if (is_identity_map(zone))
 			continue;
 
 		mark_free_pages(zone);
@@ -702,7 +702,7 @@ unsigned int count_data_pages(void)
 	unsigned int n = 0;
 
 	for_each_zone(zone) {
-		if (is_highmem(zone))
+		if (!is_identity_map(zone))
 			continue;
 
 		mark_free_pages(zone);
@@ -729,8 +729,8 @@ static inline void do_copy_page(long *ds
 static inline struct page *
 page_is_saveable(struct zone *zone, unsigned long pfn)
 {
-	return is_highmem(zone) ?
-			saveable_highmem_page(pfn) : saveable_page(pfn);
+	return is_identity_map(zone) ?
+			saveable_page(pfn) : savable_highmem_page(pfn);
 }
 
 static inline void
@@ -868,7 +868,7 @@ static int enough_free_mem(unsigned int 
 
 	for_each_zone(zone) {
 		meta += snapshot_additional_pages(zone);
-		if (!is_highmem(zone))
+		if (is_identity_map(zone))
 			free += zone_page_state(zone, NR_FREE_PAGES);
 	}
 
Index: devel-tree-2.6.20-mm2/kernel/power/swsusp.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/kernel/power/swsusp.c
+++ devel-tree-2.6.20-mm2/kernel/power/swsusp.c
@@ -229,7 +229,7 @@ int swsusp_shrink_memory(void)
 		size += highmem_size;
 		for_each_zone (zone)
 			if (populated_zone(zone)) {
-				if (is_highmem(zone)) {
+				if (!is_identity_map(zone)) {
 					highmem_size -=
 					zone_page_state(zone, NR_FREE_PAGES);
 				} else {

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [4/16] ZONE_MOVABLE

[KAMEZAWA Hiroyuki]

Add ZONE_MOVABLE.

This zone is only used for migratable/reclaimable pages.

zone order is
	[ZONE_DMA],
	[ZONE_DMA32],
	ZONE_NORMAL,
	[ZONE_HIGHMEM],
	[ZONE_MOVABLE],
	MAX_NR_ZONES
if highmem is configured, movable zone is not identitiy-mapped.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 include/linux/mmzone.h |   29 +++++++
 mm/Kconfig             |    4 +
 mm/page_alloc.c        |  180 ++++++++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 211 insertions(+), 2 deletions(-)

Index: devel-tree-2.6.20-mm2/include/linux/mmzone.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/mmzone.h
+++ devel-tree-2.6.20-mm2/include/linux/mmzone.h
@@ -142,6 +142,16 @@ enum zone_type {
 	 */
 	ZONE_HIGHMEM,
 #endif
+#ifdef CONFIG_ZONE_MOVABLE
+	/*
+	 * This memory area is used only for migratable pages.
+	 * We have a chance to hot-remove memory in this zone.
+	 * Currently, anonymous memory and usual page cache etc. are included.
+	 * if HIGHMEM is configured, MOVABLE zone is treated as
+         * not-direct-mapped-memory for kernel;.
+	 */
+	ZONE_MOVABLE,
+#endif
 	MAX_NR_ZONES,
 #ifndef CONFIG_ZONE_DMA
 	ZONE_DMA,
@@ -152,6 +162,9 @@ enum zone_type {
 #ifndef CONFIG_HIGHMEM
 	ZONE_HIGHMEM,
 #endif
+#ifndef CONFIG_ZONE_MOVABLE
+	ZONE_MOVABLE,
+#endif
 	MAX_POSSIBLE_ZONES
 };
 
@@ -172,13 +185,18 @@ static inline int is_configured_zone(enu
  * Count the active zones.  Note that the use of defined(X) outside
  * #if and family is not necessarily defined so ensure we cannot use
  * it later.  Use __ZONE_COUNT to work out how many shift bits we need.
+ *
+ * Assumes ZONE_DMA32,ZONE_HIGHMEM, ZONE_MOVABLE can't be configured at
+ * the same time.
  */
 #define __ZONE_COUNT (			\
 	  defined(CONFIG_ZONE_DMA)	\
 	+ defined(CONFIG_ZONE_DMA32)	\
 	+ 1				\
 	+ defined(CONFIG_HIGHMEM)	\
+	+ defined(CONFIG_ZONE_MOVABLE) \
 )
+
 #if __ZONE_COUNT < 2
 #define ZONES_SHIFT 0
 #elif __ZONE_COUNT <= 2
@@ -513,6 +531,11 @@ static inline int populated_zone(struct 
 	return (!!zone->present_pages);
 }
 
+static inline int is_movable_dix(enum zone_type idx)
+{
+	return (idx == ZONE_MOVABLE);
+}
+
 static inline int is_highmem_idx(enum zone_type idx)
 {
 	return (idx == ZONE_HIGHMEM);
@@ -536,6 +559,12 @@ static inline int is_identity_map_idx(en
  *              to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
  * @zone - pointer to struct zone variable
  */
+
+static inline int is_movable(struct zone *zone)
+{
+	return zone == zone->zone_pgdat->node_zones + ZONE_MOVABLE;
+}
+
 static inline int is_highmem(struct zone *zone)
 {
 	return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
+++ devel-tree-2.6.20-mm2/mm/page_alloc.c
@@ -82,6 +82,7 @@ static char name_dma[] = "DMA";
 static char name_dma32[] = "DMA32";
 static char name_normal[] = "Normal";
 static char name_highmem[] = "Highmem";
+static char name_movable[] = "Movable";
 
 static inline void __meminit zone_variables_init(void)
 {
@@ -91,6 +92,7 @@ static inline void __meminit zone_variab
 	zone_names[ZONE_DMA32] = name_dma32;
 	zone_names[ZONE_NORMAL] = name_normal;
 	zone_names[ZONE_HIGHMEM] = name_highmem;
+	zone_names[ZONE_MOVABLE] = name_movable;
 
 	/* ZONE below NORAML has ratio 256 */
 	if (is_configured_zone(ZONE_DMA))
@@ -99,6 +101,8 @@ static inline void __meminit zone_variab
 		sysctl_lowmem_reserve_ratio[ZONE_DMA32] = 256;
 	if (is_configured_zone(ZONE_HIGHMEM))
 		sysctl_lowmem_reserve_ratio[ZONE_HIGHMEM] = 32;
+	if (is_configured_zone(ZONE_MOVABLE))
+		sysctl_lowmem_reserve_ratio[ZONE_MOVABLE] = 32;
 }
 
 int min_free_kbytes = 1024;
@@ -3065,11 +3069,17 @@ void __init free_area_init_nodes(unsigne
 	arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
 	arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
 	for (i = 1; i < MAX_NR_ZONES; i++) {
+		if (i == ZONE_MOVABLE)
+			continue;
 		arch_zone_lowest_possible_pfn[i] =
 			arch_zone_highest_possible_pfn[i-1];
 		arch_zone_highest_possible_pfn[i] =
 			max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
 	}
+	if (is_configured_zone(ZONE_MOVABLE)) {
+		arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
+		arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
+	}
 
 	/* Print out the page size for debugging meminit problems */
 	printk(KERN_DEBUG "sizeof(struct page) = %zd\n", sizeof(struct page));
@@ -3097,6 +3107,7 @@ void __init free_area_init_nodes(unsigne
 				find_min_pfn_for_node(nid), NULL);
 	}
 }
+
 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
 
 /**
Index: devel-tree-2.6.20-mm2/mm/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/Kconfig
+++ devel-tree-2.6.20-mm2/mm/Kconfig
@@ -163,6 +163,10 @@ config ZONE_DMA_FLAG
 	default "0" if !ZONE_DMA
 	default "1"
 
+config ZONE_MOVABLE
+	bool "Create zones for MOVABLE pages"
+	depends on ARCH_POPULATES_NODE_MAP
+	depends on MIGRATION
 #
 # Adaptive file readahead
 #

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [5/16] GFP_MOVABLE

[KAMEZAWA Hiroyuki]

Add GFP_MOVABLE support

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
---
 include/linux/gfp.h |    9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

Index: devel-tree-2.6.20-mm2/include/linux/gfp.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/gfp.h
+++ devel-tree-2.6.20-mm2/include/linux/gfp.h
@@ -46,6 +46,7 @@ struct vm_area_struct;
 #define __GFP_NOMEMALLOC ((__force gfp_t)0x10000u) /* Don't use emergency reserves */
 #define __GFP_HARDWALL   ((__force gfp_t)0x20000u) /* Enforce hardwall cpuset memory allocs */
 #define __GFP_THISNODE	((__force gfp_t)0x40000u)/* No fallback, no policies */
+#define __GFP_MOVABLE	((__force gfp_t)0x80000u) /* Movable page allocation */
 
 #define __GFP_BITS_SHIFT 20	/* Room for 20 __GFP_FOO bits */
 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
@@ -54,7 +55,7 @@ struct vm_area_struct;
 #define GFP_LEVEL_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS| \
 			__GFP_COLD|__GFP_NOWARN|__GFP_REPEAT| \
 			__GFP_NOFAIL|__GFP_NORETRY|__GFP_NO_GROW|__GFP_COMP| \
-			__GFP_NOMEMALLOC|__GFP_HARDWALL|__GFP_THISNODE)
+			__GFP_NOMEMALLOC|__GFP_HARDWALL|__GFP_THISNODE | __GFP_MOVABLE)
 
 /* This equals 0, but use constants in case they ever change */
 #define GFP_NOWAIT	(GFP_ATOMIC & ~__GFP_HIGH)
@@ -66,6 +67,8 @@ struct vm_area_struct;
 #define GFP_USER	(__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
 #define GFP_HIGHUSER	(__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL | \
 			 __GFP_HIGHMEM)
+#define GFP_HIGH_MOVABLE (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL | \
+			 __GFP_HIGHMEM | __GFP_MOVABLE)
 
 #ifdef CONFIG_NUMA
 #define GFP_THISNODE	(__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
@@ -93,6 +96,10 @@ static inline enum zone_type gfp_zone(gf
 	if (flags & __GFP_DMA32)
 		return ZONE_DMA32;
 #endif
+#ifdef CONFIG_ZONE_MOVABLE
+	if (flags & __GFP_MOVABLE) /* we can try to alloc movable pages. */
+		return ZONE_MOVABLE;
+#endif
 #ifdef CONFIG_HIGHMEM
 	if (flags & __GFP_HIGHMEM)
 		return ZONE_HIGHMEM;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [6/16] alloc_zeroed_user_high_movable

[KAMEZAWA Hiroyuki]

Add function alloc_zeroed_user_highmovable() for allocating user pages
from ZONE_MOVABLE.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 include/linux/highmem.h |   25 +++++++++++++++++++++----
 1 file changed, 21 insertions(+), 4 deletions(-)

Index: devel-tree-2.6.20-mm2/include/linux/highmem.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/highmem.h
+++ devel-tree-2.6.20-mm2/include/linux/highmem.h
@@ -66,10 +66,11 @@ static inline void clear_user_highpage(s
 
 #ifdef CONFIG_ARCH_HAS_PREZERO_USERPAGE
 static inline struct page *
-alloc_zeroed_user_highpage(struct vm_area_struct *vma, unsigned long vaddr)
+__alloc_zeroed_user_highpage(gfp_t movable,
+			struct vm_area_struct *vma, unsigned long vaddr)
 {
 	struct page *page;
-	page = alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO, vma, vaddr);
+	page = alloc_page_vma(GFP_HIGHUSER | __GFP_ZERO | movable, vma, vaddr);
 	if (page)
 		flush_user_newzeropage(page);
 	return page;
@@ -77,9 +78,10 @@ alloc_zeroed_user_highpage(struct vm_are
 
 #else
 static inline struct page *
-alloc_zeroed_user_highpage(struct vm_area_struct *vma, unsigned long vaddr)
+__alloc_zeroed_user_highpage(gfp_t movable,
+		struct vm_area_struct *vma, unsigned long vaddr)
 {
-	struct page *page = alloc_page_vma(GFP_HIGHUSER, vma, vaddr);
+	struct page *page = alloc_page_vma(GFP_HIGHUSER | movable, vma, vaddr);
 
 	if (page)
 		clear_user_highpage(page, vaddr);
@@ -88,6 +90,21 @@ alloc_zeroed_user_highpage(struct vm_are
 }
 #endif
 
+
+static inline struct page *
+alloc_zeroed_user_highpage(struct vm_area_struct *vma, unsigned long vaddr)
+{
+	return __alloc_zeroed_user_highpage(0, vma, vaddr);
+}
+
+static inline struct page *
+alloc_zeroed_user_highmovable(struct vm_area_struct *vma, unsigned long vaddr)
+{
+	return __alloc_zeroed_user_highpage(__GFP_MOVABLE, vma, vaddr);
+}
+
+
+
 static inline void clear_highpage(struct page *page)
 {
 	void *kaddr = kmap_atomic(page, KM_USER0);

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [7/16] change caller's gfp_mask

[KAMEZAWA Hiroyuki]

Changes callers of GFP_HIGHUSER to use GFP_HIGH_MOVABLE if it can
some of alloc_zeroed_user_highpage are changed to
alloc_zeroed_user_high_movable.

I think I need more study in this area.

Signed-Off-By: KAMEZAWA Hiruyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 fs/inode.c         |    6 +++++-
 fs/namei.c         |    1 +
 fs/ramfs/inode.c   |    1 +
 mm/filemap.c       |    2 +-
 mm/memory.c        |    8 ++++----
 mm/mempolicy.c     |    4 ++--
 mm/migrate.c       |    2 +-
 mm/shmem.c         |    5 ++++-
 mm/swap_prefetch.c |    2 +-
 mm/swap_state.c    |    2 +-
 10 files changed, 21 insertions(+), 12 deletions(-)

Index: devel-tree-2.6.20-mm2/fs/inode.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/fs/inode.c
+++ devel-tree-2.6.20-mm2/fs/inode.c
@@ -145,7 +145,7 @@ static struct inode *alloc_inode(struct 
 		mapping->a_ops = &empty_aops;
  		mapping->host = inode;
 		mapping->flags = 0;
-		mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
+		mapping_set_gfp_mask(mapping, GFP_HIGH_MOVABLE);
 		mapping->assoc_mapping = NULL;
 		mapping->backing_dev_info = &default_backing_dev_info;
 
@@ -522,6 +522,10 @@ repeat:
  *	@sb: superblock
  *
  *	Allocates a new inode for given superblock.
+ *	Newly allocated inode's gfp_flag is set to GFP_HIGH_MOVABLE(default).
+ *	If fs doesn't support page migration, is hould be overriden
+ *	by GFP_HIGHUSER.
+ *	mapping_set_gfp_mask() can be used for this purpose.
  */
 struct inode *new_inode(struct super_block *sb)
 {
Index: devel-tree-2.6.20-mm2/fs/ramfs/inode.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/fs/ramfs/inode.c
+++ devel-tree-2.6.20-mm2/fs/ramfs/inode.c
@@ -61,6 +61,7 @@ struct inode *ramfs_get_inode(struct sup
 		inode->i_blocks = 0;
 		inode->i_mapping->a_ops = &ramfs_aops;
 		inode->i_mapping->backing_dev_info = &ramfs_backing_dev_info;
+		mapping_set_gfp_mask(inode->i_mapping, GFP_HIGHUSER);
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 		switch (mode & S_IFMT) {
 		default:
Index: devel-tree-2.6.20-mm2/mm/memory.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/memory.c
+++ devel-tree-2.6.20-mm2/mm/memory.c
@@ -1761,11 +1761,11 @@ gotten:
 	if (unlikely(anon_vma_prepare(vma)))
 		goto oom;
 	if (old_page == ZERO_PAGE(address)) {
-		new_page = alloc_zeroed_user_highpage(vma, address);
+		new_page = alloc_zeroed_user_highmovable(vma, address);
 		if (!new_page)
 			goto oom;
 	} else {
-		new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
+		new_page = alloc_page_vma(GFP_HIGH_MOVABLE, vma, address);
 		if (!new_page)
 			goto oom;
 		cow_user_page(new_page, old_page, address, vma);
@@ -2283,7 +2283,7 @@ static int do_anonymous_page(struct mm_s
 
 		if (unlikely(anon_vma_prepare(vma)))
 			goto oom;
-		page = alloc_zeroed_user_highpage(vma, address);
+		page = alloc_zeroed_user_highmovable(vma, address);
 		if (!page)
 			goto oom;
 
@@ -2384,7 +2384,7 @@ retry:
 
 			if (unlikely(anon_vma_prepare(vma)))
 				goto oom;
-			page = alloc_page_vma(GFP_HIGHUSER, vma, address);
+			page = alloc_page_vma(GFP_HIGH_MOVABLE, vma, address);
 			if (!page)
 				goto oom;
 			copy_user_highpage(page, new_page, address, vma);
Index: devel-tree-2.6.20-mm2/mm/mempolicy.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/mempolicy.c
+++ devel-tree-2.6.20-mm2/mm/mempolicy.c
@@ -603,7 +603,7 @@ static void migrate_page_add(struct page
 
 static struct page *new_node_page(struct page *page, unsigned long node, int **x)
 {
-	return alloc_pages_node(node, GFP_HIGHUSER, 0);
+	return alloc_pages_node(node, GFP_HIGH_MOVABLE, 0);
 }
 
 /*
@@ -719,7 +719,7 @@ static struct page *new_vma_page(struct 
 {
 	struct vm_area_struct *vma = (struct vm_area_struct *)private;
 
-	return alloc_page_vma(GFP_HIGHUSER, vma, page_address_in_vma(page, vma));
+	return alloc_page_vma(GFP_HIGH_MOVABLE, vma, page_address_in_vma(page, vma));
 }
 #else
 
Index: devel-tree-2.6.20-mm2/mm/migrate.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/migrate.c
+++ devel-tree-2.6.20-mm2/mm/migrate.c
@@ -755,7 +755,7 @@ static struct page *new_page_node(struct
 
 	*result = &pm->status;
 
-	return alloc_pages_node(pm->node, GFP_HIGHUSER | GFP_THISNODE, 0);
+	return alloc_pages_node(pm->node, GFP_HIGH_MOVABLE | GFP_THISNODE, 0);
 }
 
 /*
Index: devel-tree-2.6.20-mm2/mm/shmem.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/shmem.c
+++ devel-tree-2.6.20-mm2/mm/shmem.c
@@ -93,8 +93,11 @@ static inline struct page *shmem_dir_all
 	 * The above definition of ENTRIES_PER_PAGE, and the use of
 	 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
 	 * might be reconsidered if it ever diverges from PAGE_SIZE.
+	 *
+	 * shmem's dir is not movable page.
 	 */
-	return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
+	return alloc_pages(gfp_mask & ~__GFP_MOVABLE,
+				PAGE_CACHE_SHIFT-PAGE_SHIFT);
 }
 
 static inline void shmem_dir_free(struct page *page)
Index: devel-tree-2.6.20-mm2/mm/swap_prefetch.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/swap_prefetch.c
+++ devel-tree-2.6.20-mm2/mm/swap_prefetch.c
@@ -204,7 +204,7 @@ static enum trickle_return trickle_swap_
 	 * Get a new page to read from swap. We have already checked the
 	 * watermarks so __alloc_pages will not call on reclaim.
 	 */
-	page = alloc_pages_node(node, GFP_HIGHUSER & ~__GFP_WAIT, 0);
+	page = alloc_pages_node(node, GFP_HIGH_MOVABLE & ~__GFP_WAIT, 0);
 	if (unlikely(!page)) {
 		ret = TRICKLE_DELAY;
 		goto out;
Index: devel-tree-2.6.20-mm2/mm/swap_state.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/swap_state.c
+++ devel-tree-2.6.20-mm2/mm/swap_state.c
@@ -340,7 +340,7 @@ struct page *read_swap_cache_async(swp_e
 		 * Get a new page to read into from swap.
 		 */
 		if (!new_page) {
-			new_page = alloc_page_vma(GFP_HIGHUSER, vma, addr);
+			new_page = alloc_page_vma(GFP_HIGH_MOVABLE, vma, addr);
 			if (!new_page)
 				break;		/* Out of memory */
 		}
Index: devel-tree-2.6.20-mm2/fs/namei.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/fs/namei.c
+++ devel-tree-2.6.20-mm2/fs/namei.c
@@ -2691,6 +2691,7 @@ int __page_symlink(struct inode *inode, 
 	int err;
 	char *kaddr;
 
+	gfp_mask &= ~(__GFP_MOVABLE);
 retry:
 	err = -ENOMEM;
 	page = find_or_create_page(mapping, 0, gfp_mask);
Index: devel-tree-2.6.20-mm2/mm/filemap.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/filemap.c
+++ devel-tree-2.6.20-mm2/mm/filemap.c
@@ -423,7 +423,7 @@ 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_MOVABLE));
 
 	if (error == 0) {
 		write_lock_irq(&mapping->tree_lock);

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [8/16] counter for ZONE_MOVABLE

[KAMEZAWA Hiroyuki]

Show #of Movable pages and vmstat.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 fs/proc/proc_misc.c    |    8 ++++++++
 include/linux/kernel.h |    2 ++
 include/linux/vmstat.h |    8 +++++++-
 mm/page_alloc.c        |   28 +++++++++++++++++++++++++++-
 mm/vmstat.c            |    8 +++++++-
 5 files changed, 51 insertions(+), 3 deletions(-)

Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
+++ devel-tree-2.6.20-mm2/mm/page_alloc.c
@@ -58,6 +58,7 @@ unsigned long totalram_pages __read_most
 unsigned long totalreserve_pages __read_mostly;
 long nr_swap_pages;
 int percpu_pagelist_fraction;
+unsigned long total_movable_pages __read_mostly;
 
 static void __free_pages_ok(struct page *page, unsigned int order);
 
@@ -1571,6 +1572,20 @@ static unsigned int nr_free_zone_pages(i
 	return sum;
 }
 
+unsigned int nr_free_movable_pages(void)
+{
+	unsigned long nr_pages = 0;
+	struct zone *zone;
+	int nid;
+	if (is_configured_zone(ZONE_MOVABLE)) {
+		/* we want to count *only* pages in movable zone */
+		for_each_online_node(nid) {
+			zone = &(NODE_DATA(nid)->node_zones[ZONE_MOVABLE]);
+			nr_pages += zone_page_state(zone, NR_FREE_PAGES);
+		}
+	}
+	return nr_pages;
+}
 /*
  * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
  */
@@ -1584,7 +1599,7 @@ unsigned int nr_free_buffer_pages(void)
  */
 unsigned int nr_free_pagecache_pages(void)
 {
-	return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
+	return nr_free_zone_pages(gfp_zone(GFP_HIGH_MOVABLE));
 }
 
 /*
@@ -1633,6 +1648,8 @@ void si_meminfo(struct sysinfo *val)
 	val->totalhigh = totalhigh_pages;
 	val->freehigh = nr_free_highpages();
 	val->mem_unit = PAGE_SIZE;
+	val->movable = total_movable_pages;
+	val->free_movable = nr_free_movable_pages();
 }
 
 EXPORT_SYMBOL(si_meminfo);
@@ -1654,6 +1671,13 @@ void si_meminfo_node(struct sysinfo *val
 		val->totalhigh = 0;
 		val->freehigh = 0;
 	}
+	if (is_configured_zone(ZONE_MOVABLE)) {
+		val->movable +=
+			pgdat->node_zones[ZONE_MOVABLE].present_pages;
+		val->free_movable +=
+			zone_page_state(&pgdat->node_zones[ZONE_MOVABLE],
+				NR_FREE_PAGES);
+	}
 	val->mem_unit = PAGE_SIZE;
 }
 #endif
@@ -2779,6 +2803,8 @@ static void __meminit free_area_init_cor
 
 		zone->spanned_pages = size;
 		zone->present_pages = realsize;
+		if (j == ZONE_MOVABLE)
+			total_movable_pages += realsize;
 #ifdef CONFIG_NUMA
 		zone->node = nid;
 		zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
Index: devel-tree-2.6.20-mm2/include/linux/kernel.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/kernel.h
+++ devel-tree-2.6.20-mm2/include/linux/kernel.h
@@ -329,6 +329,8 @@ struct sysinfo {
 	unsigned short pad;		/* explicit padding for m68k */
 	unsigned long totalhigh;	/* Total high memory size */
 	unsigned long freehigh;		/* Available high memory size */
+	unsigned long movable;		/* pages used only for data */
+	unsigned long free_movable;	/* Avaiable pages in movable */
 	unsigned int mem_unit;		/* Memory unit size in bytes */
 	char _f[20-2*sizeof(long)-sizeof(int)];	/* Padding: libc5 uses this.. */
 };
Index: devel-tree-2.6.20-mm2/fs/proc/proc_misc.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/fs/proc/proc_misc.c
+++ devel-tree-2.6.20-mm2/fs/proc/proc_misc.c
@@ -160,6 +160,10 @@ static int meminfo_read_proc(char *page,
 		"LowTotal:     %8lu kB\n"
 		"LowFree:      %8lu kB\n"
 #endif
+#ifdef CONFIG_ZONE_MOVABLE
+		"MovableTotal: %8lu kB\n"
+		"MovableFree:  %8lu kB\n"
+#endif
 		"SwapTotal:    %8lu kB\n"
 		"SwapFree:     %8lu kB\n"
 		"Dirty:        %8lu kB\n"
@@ -191,6 +195,10 @@ static int meminfo_read_proc(char *page,
 		K(i.totalram-i.totalhigh),
 		K(i.freeram-i.freehigh),
 #endif
+#ifdef CONFIG_ZONE_MOVABLE
+		K(i.movable),
+		K(i.free_movable),
+#endif
 		K(i.totalswap),
 		K(i.freeswap),
 		K(global_page_state(NR_FILE_DIRTY)),
Index: devel-tree-2.6.20-mm2/include/linux/vmstat.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/vmstat.h
+++ devel-tree-2.6.20-mm2/include/linux/vmstat.h
@@ -25,7 +25,13 @@
 #define HIGHMEM_ZONE(xx)
 #endif
 
-#define FOR_ALL_ZONES(xx) DMA_ZONE(xx) DMA32_ZONE(xx) xx##_NORMAL HIGHMEM_ZONE(xx)
+#ifdef CONFIG_ZONE_MOVABLE
+#define MOVABLE_ZONE(xx) , xx##_MOVABLE
+#else
+#define MOVABLE_ZONE(xx)
+#endif
+
+#define FOR_ALL_ZONES(xx) DMA_ZONE(xx) DMA32_ZONE(xx) xx##_NORMAL HIGHMEM_ZONE(xx) MOVABLE_ZONE(xx)
 
 enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
 		FOR_ALL_ZONES(PGALLOC),
Index: devel-tree-2.6.20-mm2/mm/vmstat.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/vmstat.c
+++ devel-tree-2.6.20-mm2/mm/vmstat.c
@@ -426,8 +426,14 @@ const struct seq_operations fragmentatio
 #define TEXT_FOR_HIGHMEM(xx)
 #endif
 
+#ifdef CONFIG_ZONE_MOVABLE
+#define TEXT_FOR_MOVABLE(xx) xx "_movable",
+#else
+#define TXT_FOR_MOVABLE(xx)
+#endif
+
 #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
-					TEXT_FOR_HIGHMEM(xx)
+					TEXT_FOR_HIGHMEM(xx) TEXT_FOR_MOVABLE(xx)
 
 static const char * const vmstat_text[] = {
 	/* Zoned VM counters */

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [9/16] create movable zone at boot

[KAMEZAWA Hiroyuki]

This patch adds codes for creating movable zones.

Add 2 kernel paramers.
- kernel_core_pages=XXX[KMG]
- kernel_core_ratio=xx

When kernel_core_pages is specified, create zone(s) for not-movable pages
from lower address and make the amount of it to specified size.
Maybe good for non-NUMA environment and node-hot-remove.

When kernel_core_ratio is specified, create zone(s) for not-movable pages
on each node. The amount of not-movable-zone is calucated as

 memory_on_node * kernel_core_ratio/100.

Maybe good for NUMA environment and just want to use MOVABLE zone.

Note:
Changes to zone_spanned_pages_in_node()/absent_pages_in_node() looks ugly...
And, this boot option is just a sample. I'll change this when I find a better
way to go.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 Documentation/kernel-parameters.txt |   11 ++
 include/linux/mmzone.h              |    3 
 mm/page_alloc.c                     |  198 +++++++++++++++++++++++++++++++++---
 3 files changed, 199 insertions(+), 13 deletions(-)

Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
+++ devel-tree-2.6.20-mm2/mm/page_alloc.c
@@ -137,12 +137,16 @@ static unsigned long __initdata dma_rese
   int __initdata nr_nodemap_entries;
   unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
   unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
+  unsigned long __initdata lowest_movable_pfn[MAX_NUMNODES];
+  unsigned long kernel_core_ratio;
+  unsigned long kernel_core_pages;
 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
   unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
   unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
 
+
 #ifdef CONFIG_DEBUG_VM
 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
 {
@@ -2604,6 +2608,8 @@ void __init get_pfn_range_for_nid(unsign
  */
 unsigned long __init zone_spanned_pages_in_node(int nid,
 					unsigned long zone_type,
+					unsigned long *start_pfn,
+					unsigned long *end_pfn,
 					unsigned long *ignored)
 {
 	unsigned long node_start_pfn, node_end_pfn;
@@ -2611,8 +2617,30 @@ unsigned long __init zone_spanned_pages_
 
 	/* Get the start and end of the node and zone */
 	get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
-	zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
-	zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
+	if (start_pfn)
+		*start_pfn = 0;
+	if (end_pfn)
+		*end_pfn = 0;
+	if (!is_configured_zone(ZONE_MOVABLE) ||
+		   lowest_movable_pfn[nid] == 0) {
+		/* we don't use ZONE_MOVABLE */
+		zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
+		zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
+	} else if (zone_type == ZONE_MOVABLE) {
+		zone_start_pfn = lowest_movable_pfn[nid];
+		zone_end_pfn = node_end_pfn;
+	} else {
+		/* adjust range to lowest_movable_pfn[] */
+		zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
+		zone_start_pfn = max(zone_start_pfn, node_start_pfn);
+
+		if (zone_start_pfn >= lowest_movable_pfn[nid])
+			return 0;
+		zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
+		zone_end_pfn = min(zone_end_pfn, node_end_pfn);
+		if (zone_end_pfn > lowest_movable_pfn[nid])
+			zone_end_pfn = lowest_movable_pfn[nid];
+	}
 
 	/* Check that this node has pages within the zone's required range */
 	if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
@@ -2621,8 +2649,11 @@ unsigned long __init zone_spanned_pages_
 	/* Move the zone boundaries inside the node if necessary */
 	zone_end_pfn = min(zone_end_pfn, node_end_pfn);
 	zone_start_pfn = max(zone_start_pfn, node_start_pfn);
-
 	/* Return the spanned pages */
+	if (start_pfn)
+		*start_pfn = zone_start_pfn;
+	if (end_pfn)
+		*end_pfn = zone_end_pfn;
 	return zone_end_pfn - zone_start_pfn;
 }
 
@@ -2692,16 +2723,24 @@ unsigned long __init absent_pages_in_ran
 /* Return the number of page frames in holes in a zone on a node */
 unsigned long __init zone_absent_pages_in_node(int nid,
 					unsigned long zone_type,
+					unsigned long start,
+					unsigned long end,
 					unsigned long *ignored)
 {
 	unsigned long node_start_pfn, node_end_pfn;
 	unsigned long zone_start_pfn, zone_end_pfn;
 
 	get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
-	zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
-							node_start_pfn);
-	zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
-							node_end_pfn);
+	if (start == 0 && end == 0) {
+		zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
+								node_start_pfn);
+		zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
+								node_end_pfn);
+	} else {
+		/* ZONE_MOVABLE always use passed params */
+		zone_start_pfn = start;
+		zone_end_pfn = end;
+	}
 
 	return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
 }
@@ -2709,13 +2748,22 @@ unsigned long __init zone_absent_pages_i
 #else
 static inline unsigned long zone_spanned_pages_in_node(int nid,
 					unsigned long zone_type,
+					unsigned long *start_pfn,
+					unsigned long *end_pfn,
 					unsigned long *zones_size)
 {
+	/* this will not be used by caller*/
+	if (start_pfn)
+		*start_pfn = 0;
+	if (end_pfn)
+		*end_pfn = 0;
 	return zones_size[zone_type];
 }
 
 static inline unsigned long zone_absent_pages_in_node(int nid,
 						unsigned long zone_type,
+						unsigned long start,
+						unsigned long end,
 						unsigned long *zholes_size)
 {
 	if (!zholes_size)
@@ -2733,20 +2781,115 @@ static void __init calculate_node_totalp
 	enum zone_type i;
 
 	for (i = 0; i < MAX_NR_ZONES; i++)
-		totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
+		totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, NULL, NULL,
 								zones_size);
 	pgdat->node_spanned_pages = totalpages;
 
 	realtotalpages = totalpages;
 	for (i = 0; i < MAX_NR_ZONES; i++)
 		realtotalpages -=
-			zone_absent_pages_in_node(pgdat->node_id, i,
+			zone_absent_pages_in_node(pgdat->node_id, i, 0, 0,
 								zholes_size);
 	pgdat->node_present_pages = realtotalpages;
 	printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
 							realtotalpages);
 }
 
+#ifdef CONFIG_ZONE_MOVABLE
+
+unsigned long calc_zone_alignment(unsigned long pfn)
+{
+#ifdef CONFIG_SPARSEMEM
+	return (pfn + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK;
+#else
+	return (pfn + MAX_ORDER_NR_PAGES - 1) & ~(MAX_ORDER_NR_PAGES - 1)
+#endif
+}
+
+
+static void alloc_core_pages_from_low(void)
+{
+	unsigned long nr_pages, start_pfn, end_pfn, pfn;
+	int i, nid;
+	long kcore_pages = kernel_core_pages;
+	for_each_online_node(nid) {
+		for_each_active_range_index_in_nid(i, nid) {
+			start_pfn = early_node_map[i].start_pfn;
+			end_pfn = early_node_map[i].end_pfn;
+			nr_pages = end_pfn - start_pfn;
+			if (nr_pages > kcore_pages) {
+				pfn = start_pfn + kcore_pages;
+				pfn = calc_zone_alignment(pfn);
+				if (pfn < end_pfn) {
+					lowest_movable_pfn[nid] = pfn;
+					kcore_pages = 0;
+					break;
+				} else {
+					kcore_pages = 0;
+				}
+			} else {
+				kcore_pages -= nr_pages;
+			}
+		}
+	}
+	return;
+}
+
+static void split_movable_pages(void)
+{
+	int i, nid;
+	unsigned long total_pages, nr_pages, start_pfn, end_pfn, pfn;
+	long core;
+	for_each_online_node(nid) {
+		lowest_movable_pfn[nid] = 0;
+		pfn = 0;
+		total_pages = 0;
+		for_each_active_range_index_in_nid(i, nid) {
+			start_pfn = early_node_map[i].start_pfn;
+			end_pfn = early_node_map[i].end_pfn;
+			total_pages += end_pfn - start_pfn;
+		}
+		core = total_pages * kernel_core_ratio/100;
+		for_each_active_range_index_in_nid(i, nid) {
+			start_pfn = early_node_map[i].start_pfn;
+			end_pfn = early_node_map[i].end_pfn;
+			nr_pages = end_pfn - start_pfn;
+			if (nr_pages > core) {
+				pfn = start_pfn + core;
+				pfn = calc_zone_alignment(pfn);
+				if (pfn < end_pfn) {
+					lowest_movable_pfn[nid] = pfn;
+					break;
+				} else {
+					core -= nr_pages;
+					if (core < 0)
+						core = 0;
+				}
+			} else {
+				core -= nr_pages;
+			}
+		}
+	}
+	return;
+}
+
+
+static void reserve_movable_pages(void)
+{
+	memset(lowest_movable_pfn, 0, MAX_NUMNODES);
+	if (kernel_core_pages) {
+		alloc_core_pages_from_low();
+	} else if (kernel_core_ratio) {
+		split_movable_pages();
+	}
+	return;
+}
+#else
+static void reserve_movable_pages(void)
+{
+	return;
+}
+#endif
 /*
  * Set up the zone data structures:
  *   - mark all pages reserved
@@ -2768,10 +2911,10 @@ static void __meminit free_area_init_cor
 	
 	for (j = 0; j < MAX_NR_ZONES; j++) {
 		struct zone *zone = pgdat->node_zones + j;
-		unsigned long size, realsize, memmap_pages;
+		unsigned long size, realsize, memmap_pages, start, end;
 
-		size = zone_spanned_pages_in_node(nid, j, zones_size);
-		realsize = size - zone_absent_pages_in_node(nid, j,
+		size = zone_spanned_pages_in_node(nid, j, &start, &end, zones_size);
+		realsize = size - zone_absent_pages_in_node(nid, j, start, end,
 								zholes_size);
 
 		/*
@@ -3065,6 +3208,7 @@ unsigned long __init find_max_pfn_with_a
 	return max_pfn;
 }
 
+
 /**
  * free_area_init_nodes - Initialise all pg_data_t and zone data
  * @max_zone_pfn: an array of max PFNs for each zone
@@ -3127,6 +3271,8 @@ void __init free_area_init_nodes(unsigne
 
 	/* Initialise every node */
 	setup_nr_node_ids();
+	/* setup movable pages */
+	reserve_movable_pages();
 	for_each_online_node(nid) {
 		pg_data_t *pgdat = NODE_DATA(nid);
 		free_area_init_node(nid, pgdat, NULL,
@@ -3542,6 +3688,33 @@ void *__init alloc_large_system_hash(con
 	return table;
 }
 
+#ifdef CONFIG_ZONE_MOVABLE
+
+char * __init parse_kernel_core_pages(char *p)
+{
+	unsigned long long coremem;
+	if (!p)
+		return NULL;
+	coremem = memparse(p, &p);
+	kernel_core_pages = coremem >> PAGE_SHIFT;
+	return p;
+}
+
+char * __init parse_kernel_core_ratio(char *p)
+{
+	int ratio[1];
+	ratio[0] = 0;
+	if (!p)
+		return NULL;
+	p = get_options(p, 1, ratio);
+	if (ratio[0])
+		kernel_core_ratio = ratio[0];
+	if (kernel_core_ratio > 100)
+		kernel_core_ratio = 0; /* ll memory is not movable */
+	return p;
+}
+#endif /* CONFIG_ZONE_MOVABLE */
+
 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
 struct page *pfn_to_page(unsigned long pfn)
 {
@@ -3555,4 +3728,3 @@ EXPORT_SYMBOL(pfn_to_page);
 EXPORT_SYMBOL(page_to_pfn);
 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
 
-
Index: devel-tree-2.6.20-mm2/Documentation/kernel-parameters.txt
===================================================================
--- devel-tree-2.6.20-mm2.orig/Documentation/kernel-parameters.txt
+++ devel-tree-2.6.20-mm2/Documentation/kernel-parameters.txt
@@ -764,6 +764,17 @@ and is between 256 and 4096 characters. 
 
 	keepinitrd	[HW,ARM]
 
+	kernel_core_pages=nn[KMG] [KNL, BOOT] divide the whole memory into
+			not-movable and movable. movable memory can be
+			used only for page cache and user data. This option
+			specifies the amount of not-movable pages, called core
+			pages. core pages are allocated from the lower address.
+
+	kernel_core_ratio=nn [KND, BOOT] specifies the amount of the core
+			pages(see kernel_core_pages) by the ratio against
+			total memory. If NUMA, core pages are allocated for
+			each node by this ratio. "0" is not allowed.
+
 	kstack=N	[IA-32,X86-64] Print N words from the kernel stack
 			in oops dumps.
 
Index: devel-tree-2.6.20-mm2/include/linux/mmzone.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/mmzone.h
+++ devel-tree-2.6.20-mm2/include/linux/mmzone.h
@@ -608,6 +608,9 @@ int sysctl_min_unmapped_ratio_sysctl_han
 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
 			struct file *, void __user *, size_t *, loff_t *);
 
+extern char* parse_kernel_core_pages(char *cp);
+extern char*  parse_kernel_core_ratio(char *cp);
+
 #include <linux/topology.h>
 /* Returns the number of the current Node. */
 #ifndef numa_node_id

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [10/16] ia64 support

[KAMEZAWA Hiroyuki]

Add ia64 support for kernel_core_pages/kernel_core_ratio.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 arch/ia64/kernel/efi.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

Index: devel-tree-2.6.20-mm2/arch/ia64/kernel/efi.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/ia64/kernel/efi.c
+++ devel-tree-2.6.20-mm2/arch/ia64/kernel/efi.c
@@ -424,7 +424,11 @@ efi_init (void)
 			max_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
 		} else if (memcmp(cp, "min_addr=", 9) == 0) {
 			min_addr = GRANULEROUNDDOWN(memparse(cp + 9, &cp));
-		} else {
+		} else if (memcmp(cp, "kernel_core_pages=",18) == 0) {
+			cp = parse_kernel_core_pages(cp + 18);
+		} else if (memcmp(cp, "kernel_core_ratio=", 18) == 0) {
+			cp = parse_kernel_core_ratio(cp + 18);
+		} else  {
 			while (*cp != ' ' && *cp)
 				++cp;
 			while (*cp == ' ')

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [11/16] page isolation core

[KAMEZAWA Hiroyuki]

This patch is for supporting making page unused.

Isolate pages by capturing freed pages before inserting free_area[],
buddy allocator.
If you have an idea for avoiding spin_lock(), please advise me.

Isolating pages in free_area[] is implemented in other patch.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 include/linux/mmzone.h         |    8 +
 include/linux/page_isolation.h |   52 +++++++++++
 mm/Kconfig                     |    7 +
 mm/page_alloc.c                |  184 +++++++++++++++++++++++++++++++++++++++++
 4 files changed, 251 insertions(+)

Index: devel-tree-2.6.20-mm2/include/linux/mmzone.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/mmzone.h
+++ devel-tree-2.6.20-mm2/include/linux/mmzone.h
@@ -315,6 +315,14 @@ struct zone {
 	/* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
 	unsigned long		zone_start_pfn;
 
+#ifdef CONFIG_PAGE_ISOLATION
+	/*
+	 *  For pages which are not used but not free.
+	 *  See include/linux/page_isolation.h
+	 */
+	spinlock_t		isolation_lock;
+	struct list_head	isolation_list;
+#endif
 	/*
 	 * zone_start_pfn, spanned_pages and present_pages are all
 	 * protected by span_seqlock.  It is a seqlock because it has
Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
+++ devel-tree-2.6.20-mm2/mm/page_alloc.c
@@ -41,6 +41,7 @@
 #include <linux/pfn.h>
 #include <linux/backing-dev.h>
 #include <linux/fault-inject.h>
+#include <linux/page_isolation.h>
 
 #include <asm/tlbflush.h>
 #include <asm/div64.h>
@@ -421,6 +422,9 @@ static inline void __free_one_page(struc
 	if (unlikely(PageCompound(page)))
 		destroy_compound_page(page, order);
 
+	if (page_under_isolation(zone, page, order))
+		return;
+
 	page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
 
 	VM_BUG_ON(page_idx & (order_size - 1));
@@ -2969,6 +2973,10 @@ static void __meminit free_area_init_cor
 		zone->nr_scan_inactive = 0;
 		zap_zone_vm_stats(zone);
 		atomic_set(&zone->reclaim_in_progress, 0);
+#ifdef CONFIG_PAGE_ISOLATION
+		spin_lock_init(&zone->isolation_lock);
+		INIT_LIST_HEAD(&zone->isolation_list);
+#endif
 		if (!size)
 			continue;
 
@@ -3728,3 +3736,179 @@ EXPORT_SYMBOL(pfn_to_page);
 EXPORT_SYMBOL(page_to_pfn);
 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
 
+#ifdef CONFIG_PAGE_ISOLATION
+/*
+ * Page Isolation.
+ *
+ * If a page is removed from usual free_list and will never be used,
+ * It is linked to "struct isolation_info" and set Reserved, Private
+ * bit. page->mapping points to isolation_info in it.
+ * and page_count(page) is 0.
+ *
+ * This can be used for creating a chunk of contiguous *unused* memory.
+ *
+ * current user is Memory-Hot-Remove.
+ * maybe move to some other file is better.
+ */
+static void
+isolate_page_nolock(struct isolation_info *info, struct page *page, int order)
+{
+	int pagenum;
+	pagenum = 1 << order;
+	while (pagenum > 0) {
+		SetPageReserved(page);
+		SetPagePrivate(page);
+		page->private = (unsigned long)info;
+		list_add(&page->lru, &info->pages);
+		page++;
+		pagenum--;
+	}
+}
+
+/*
+ * This function is called from page_under_isolation()l
+ */
+
+int __page_under_isolation(struct zone *zone, struct page *page, int order)
+{
+	struct isolation_info *info;
+	unsigned long pfn = page_to_pfn(page);
+	unsigned long flags;
+	int found = 0;
+
+	spin_lock_irqsave(&zone->isolation_lock,flags);
+	list_for_each_entry(info, &zone->isolation_list, list) {
+		if (info->start_pfn <= pfn && pfn < info->end_pfn) {
+			found = 1;
+			break;
+		}
+	}
+	if (found) {
+		isolate_page_nolock(info, page, order);
+	}
+	spin_unlock_irqrestore(&zone->isolation_lock, flags);
+	return found;
+}
+
+/*
+ * start and end must be in the same zone.
+ *
+ */
+struct isolation_info  *
+register_isolation(unsigned long start, unsigned long end)
+{
+	struct zone *zone;
+	struct isolation_info *info = NULL, *tmp;
+	unsigned long flags;
+	unsigned long last_pfn = end - 1;
+
+	if (!pfn_valid(start) || !pfn_valid(last_pfn) || (start >= end))
+		return ERR_PTR(-EINVAL);
+	/* check start and end is in the same zone */
+	zone = page_zone(pfn_to_page(start));
+
+	if (zone != page_zone(pfn_to_page(last_pfn)))
+		return ERR_PTR(-EINVAL);
+	/* target range has to match MAX_ORDER alignmet */
+	if ((start & (MAX_ORDER_NR_PAGES - 1)) ||
+		(end & (MAX_ORDER_NR_PAGES - 1)))
+		return ERR_PTR(-EINVAL);
+	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	if (!info)
+		return ERR_PTR(-ENOMEM);
+	spin_lock_irqsave(&zone->isolation_lock, flags);
+	/* we don't allow overlap among isolation areas */
+	if (!list_empty(&zone->isolation_list)) {
+		list_for_each_entry(tmp, &zone->isolation_list, list) {
+			if (start < tmp->end_pfn && end > tmp->start_pfn) {
+				goto out_free;
+			}
+		}
+	}
+	info->start_pfn = start;
+	info->end_pfn = end;
+	info->zone = zone;
+	INIT_LIST_HEAD(&info->list);
+	INIT_LIST_HEAD(&info->pages);
+	list_add(&info->list, &zone->isolation_list);
+out_unlock:
+	spin_unlock_irqrestore(&zone->isolation_lock, flags);
+	return info;
+out_free:
+	kfree(info);
+	info = ERR_PTR(-EBUSY);
+	goto out_unlock;
+}
+/*
+ * Remove IsolationInfo from zone.
+ * After this, we can unuse memory in info or
+ * free back to freelist.
+ */
+
+void
+detach_isolation_info_zone(struct isolation_info *info)
+{
+	unsigned long flags;
+	struct zone *zone = info->zone;
+	spin_lock_irqsave(&zone->isolation_lock,flags);
+	list_del(&info->list);
+	info->zone = NULL;
+	spin_unlock_irqrestore(&zone->isolation_lock,flags);
+}
+
+/*
+ * All pages in info->pages should be remvoed before calling this.
+ * And info should be detached from zone.
+ */
+void
+free_isolation_info(struct isolation_info *info)
+{
+	BUG_ON(!list_empty(&info->pages));
+	BUG_ON(info->zone);
+	kfree(info);
+	return;
+}
+
+/*
+ * Mark All pages in the isolation_info to be Reserved.
+ * When onlining these pages again, a user must check
+ * which page is usable by IORESOURCE_RAM
+ * please see memory_hotplug.c/online_pages() if unclear.
+ *
+ * info should be detached from zone before calling this.
+ */
+void
+unuse_all_isolated_pages(struct isolation_info *info)
+{
+	struct page *page, *n;
+	BUG_ON(info->zone);
+	list_for_each_entry_safe(page, n, &info->pages, lru) {
+		SetPageReserved(page);
+		page->private = 0;
+		ClearPagePrivate(page);
+		list_del(&page->lru);
+	}
+}
+
+/*
+ * Free all pages connected in isolation list.
+ * pages are moved back to free_list.
+ */
+void
+free_all_isolated_pages(struct isolation_info *info)
+{
+	struct page *page, *n;
+	BUG_ON(info->zone);
+	list_for_each_entry_safe(page, n ,&info->pages, lru) {
+		ClearPagePrivate(page);
+		ClearPageReserved(page);
+		page->private = 0;
+		list_del(&page->lru);
+		set_page_count(page, 0);
+		set_page_refcounted(page);
+		/* This is sage because info is detached from zone */
+		__free_page(page);
+	}
+}
+
+#endif
Index: devel-tree-2.6.20-mm2/mm/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/Kconfig
+++ devel-tree-2.6.20-mm2/mm/Kconfig
@@ -224,3 +224,10 @@ config DEBUG_READAHEAD
 
 	  Say N for production servers.
 
+config PAGE_ISOLATION
+	bool	"Page Isolation Framework"
+	help
+	  This option adds page isolation framework to mm.
+	  This is used for isolate amount of contiguous pages from linux
+	  memory management.
+	  Say N if unsure.
Index: devel-tree-2.6.20-mm2/include/linux/page_isolation.h
===================================================================
--- /dev/null
+++ devel-tree-2.6.20-mm2/include/linux/page_isolation.h
@@ -0,0 +1,52 @@
+#ifndef __LINIX_PAGE_ISOLATION_H
+#define __LINUX_PAGE_ISOLATION_H
+
+#ifdef CONFIG_PAGE_ISOLATION
+
+struct isolation_info {
+	struct list_head	list;
+	unsigned long	start_pfn;
+	unsigned long	end_pfn;
+	struct zone		*zone;
+	struct list_head	pages;
+};
+
+extern int
+__page_under_isolation(struct zone *zone, struct page *page, int order);
+
+static inline int
+page_under_isolation(struct zone *zone, struct page *page, int order)
+{
+	if (likely(list_empty(&zone->isolation_list)))
+		return 0;
+	return __page_under_isolation(zone, page, order);
+}
+
+static inline int
+is_page_isolated(struct isolation_info *info, struct page *page)
+{
+	if (PageReserved(page) && PagePrivate(page) &&
+	    page_count(page) == 0 &&
+	    page->private == (unsigned long)info)
+		return 1;
+	return 0;
+}
+
+extern struct isolation_info *
+register_isolation(unsigned long start, unsigned long end);
+
+extern void detach_isolation_info_zone(struct isolation_info *info);
+extern void free_isolation_info(struct isolation_info *info);
+extern void unuse_all_isolated_pages(struct isolation_info *info);
+extern void free_all_isolated_pages(struct isolation_info *info);
+
+#else
+
+static inline int
+page_under_isolation(struct zone *zone, struct page *page, int order)
+{
+	return 0;
+}
+
+#endif
+#endif

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [12/16] drain all pages

[KAMEZAWA Hiroyuki]

This patch add function drain_all_pages(void) to drain all 
pages on per-cpu-freelist.
Page isolation will catch them in free_one_page.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>


---
 include/linux/page_isolation.h |    1 +
 mm/page_alloc.c                |   17 ++++++++++++++++-
 2 files changed, 17 insertions(+), 1 deletion(-)

Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
+++ devel-tree-2.6.20-mm2/mm/page_alloc.c
@@ -822,6 +822,9 @@ void mark_free_pages(struct zone *zone)
 
 	spin_unlock_irqrestore(&zone->lock, flags);
 }
+#endif /* CONFIG_PM */
+
+#if defined(CONFIG_PM) || defined(CONFIG_PAGE_ISOLATION)
 
 /*
  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
@@ -834,8 +837,20 @@ void drain_local_pages(void)
 	__drain_pages(smp_processor_id());
 	local_irq_restore(flags);	
 }
-#endif /* CONFIG_PM */
+#endif
 
+#ifdef CONFIG_PAGE_ISOLATION
+static void drain_local_zone_pages(struct work_struct *work)
+{
+	drain_local_pages();
+}
+
+void drain_all_pages(void)
+{
+	schedule_on_each_cpu(drain_local_zone_pages);
+}
+
+#endif
 /*
  * Free a 0-order page
  */
Index: devel-tree-2.6.20-mm2/include/linux/page_isolation.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/page_isolation.h
+++ devel-tree-2.6.20-mm2/include/linux/page_isolation.h
@@ -39,6 +39,7 @@ extern void detach_isolation_info_zone(s
 extern void free_isolation_info(struct isolation_info *info);
 extern void unuse_all_isolated_pages(struct isolation_info *info);
 extern void free_all_isolated_pages(struct isolation_info *info);
+extern void drain_all_pages(void);
 
 #else
 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [13/16] isolate freed pages.

[KAMEZAWA Hiroyuki]

Isolate all freed pages (means in buddy_list) in the range.
See page_buddy() and free_one_page() function if unsure.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 include/linux/page_isolation.h |    2 +
 mm/page_alloc.c                |   48 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+)

Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
+++ devel-tree-2.6.20-mm2/mm/page_alloc.c
@@ -3927,10 +3929,59 @@ free_all_isolated_pages(struct isolation
 		list_del(&page->lru);
 		set_page_count(page, 0);
 		set_page_refcounted(page);
 		/* This is sage because info is detached from zone */
 		__free_page(page);
 	}
 }
 
+
+/*
+ * Isolate already freed pages.
+ */
+int
+capture_isolate_freed_pages(struct isolation_info *info)
+{
+	struct zone *zone;
+	unsigned long pfn;
+	struct page *page;
+	int order, order_size;
+	int nr_pages = 0;
+	unsigned long last_pfn = info->end_pfn - 1;
+	pfn = info->start_pfn;
+	if (!pfn_valid(pfn))
+		return -EINVAL;
+	zone = info->zone;
+	if ((zone != page_zone(pfn_to_page(pfn))) ||
+	    (zone != page_zone(pfn_to_page(last_pfn))))
+		return -EINVAL;
+	drain_all_pages();
+	spin_lock(&zone->lock);
+	while (pfn < info->end_pfn) {
+		if (!pfn_valid(pfn)) {
+			pfn++;
+			continue;
+		}
+		page = pfn_to_page(pfn);
+		/* See page_is_buddy()  */
+		if (page_count(page) == 0 && PageBuddy(page)) {
+			order = page_order(page);
+			order_size = 1 << order;
+			zone->free_area[order].nr_free--;
+			__mod_zone_page_state(zone, NR_FREE_PAGES, -order_size);
+			list_del(&page->lru);
+			rmv_page_order(page);
+			isolate_page_nolock(info, page, order);
+			nr_pages += order_size;
+			pfn += order_size;
+		} else {
+			pfn++;
+		}
+	}
+	spin_unlock(&zone->lock);
+	return nr_pages;
+}
+
+
 #endif

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [14/16] memory unplug core

[KAMEZAWA Hiroyuki]

Add MEMORY_HOTREMOVE config and implements basic algorythm.

This config selects ZONE_MOVABLE and PAGE_ISOLATION

how work:
1. register Isololation area of specified section
2. search mem_map and migrate pages.
3. detach isolation and make pages unused.

This works on my easy test, but I think I need more work on loop algorythm 
and policy.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>


---
 include/linux/memory_hotplug.h |    1 
 mm/Kconfig                     |    9 +
 mm/memory_hotplug.c            |  219 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 228 insertions(+), 1 deletion(-)

Index: devel-tree-2.6.20-mm2/mm/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/Kconfig
+++ devel-tree-2.6.20-mm2/mm/Kconfig
@@ -126,6 +126,13 @@ config MEMORY_HOTPLUG_SPARSE
 	def_bool y
 	depends on SPARSEMEM && MEMORY_HOTPLUG
 
+config MEMORY_HOTREMOVE
+	bool "Allow for memory hot-remvoe"
+	depends on MEMORY_HOTPLUG_SPARSE
+	select	ZONE_MOVABLE
+	select	MIGRATION
+	select  PAGE_ISOLATION
+
 # Heavily threaded applications may benefit from splitting the mm-wide
 # page_table_lock, so that faults on different parts of the user address
 # space can be handled with less contention: split it at this NR_CPUS.
@@ -145,7 +152,7 @@ config SPLIT_PTLOCK_CPUS
 config MIGRATION
 	bool "Page migration"
 	def_bool y
-	depends on NUMA
+	depends on NUMA || MEMORY_HOTREMOVE
 	help
 	  Allows the migration of the physical location of pages of processes
 	  while the virtual addresses are not changed. This is useful for
Index: devel-tree-2.6.20-mm2/mm/memory_hotplug.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/memory_hotplug.c
+++ devel-tree-2.6.20-mm2/mm/memory_hotplug.c
@@ -23,6 +23,9 @@
 #include <linux/vmalloc.h>
 #include <linux/ioport.h>
 #include <linux/cpuset.h>
+#include <linux/page_isolation.h>
+#include <linux/delay.h>
+#include <linux/migrate.h>
 
 #include <asm/tlbflush.h>
 
@@ -308,3 +311,221 @@ error:
 	return ret;
 }
 EXPORT_SYMBOL_GPL(add_memory);
+
+
+
+#ifdef CONFIG_MEMORY_HOTREMOVE
+
+/*
+ * Just an easy implementation.
+ */
+static struct page *
+hotremove_migrate_alloc(struct page *page,
+			unsigned long private,
+			int **x)
+{
+	return alloc_page(GFP_HIGH_MOVABLE);
+}
+
+/* scans # of pages per itelation */
+#define HOTREMOVE_UNIT	(1024)
+
+static int do_migrate_and_isolate_pages(struct isolation_info *info,
+				unsigned long start_pfn,
+				unsigned long end_pfn)
+{
+	int move_pages = HOTREMOVE_UNIT;
+	int ret, managed, not_managed;
+	unsigned long pfn;
+	struct page *page;
+	LIST_HEAD(source);
+
+	not_managed = 0;
+	for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
+		if (!pfn_valid(pfn))  /* never happens in sparsemem */
+			continue;
+		page = pfn_to_page(pfn);
+		if (is_page_isolated(info,page))
+			continue;
+		ret = isolate_lru_page(page, &source);
+
+		if (ret == 1) {
+			move_pages--;
+			managed++;
+		} else {
+			if (page_count(page))
+				not_managed++; /* someone uses this */
+		}
+	}
+	ret = -EBUSY;
+	if (not_managed) {
+		if (!list_empty(&source))
+			putback_lru_pages(&source);
+		goto out;
+	}
+	ret = 0;
+	if (list_empty(&source))
+		goto out;
+	/* this function returns # of failed pages */
+	ret = migrate_pages(&source, hotremove_migrate_alloc,
+			   (unsigned long)info);
+out:
+	return ret;
+}
+
+
+/*
+ * Check All pages registered as IORESOURCE_RAM are isolated or not.
+ */
+static int check_removal_success(struct isolation_info *info)
+{
+	struct resource res;
+	unsigned long section_end;
+	unsigned long start_pfn, i, nr_pages;
+	struct page *page;
+	int removed = 0;
+	res.start = info->start_pfn << PAGE_SHIFT;
+	res.end = (info->end_pfn - 1) << PAGE_SHIFT;
+	res.flags = IORESOURCE_MEM;
+	section_end = res.end;
+	while ((res.start < res.end) && (find_next_system_ram(&res) >= 0)) {
+		start_pfn =(res.start >> PAGE_SHIFT);
+		nr_pages = (res.end + 1UL - res.start) >> PAGE_SHIFT;
+		for (i = 0; i < nr_pages; i++) {
+			page = pfn_to_page(start_pfn + i);
+			if (!is_page_isolated(info,page))
+				return -EBUSY;
+			removed++;
+		}
+		res.start = res.end + 1;
+		res.end = section_end;
+	}
+	return removed;
+}
+/*
+ * start_pfn and end_pfn myst be aligned to SECTION_SIZE.
+ * start_pfn and end_pfn must be in the same zone.
+ * target page range must be in ZONE_MOVABLE.
+ *
+ * Under this, [start_pfn, end_pfn) pages are isolated.
+ * All freed pages in the range is captured info isolation_info.
+ *
+ * If all pages in the range are isolated, offline_pages() returns 0.
+ *
+ * Note: memory holes in section are marked as Reserved Memory.
+ *       So we igonre Reserved pages in the first check.
+ *       But bootmem is aslo makred as Reserved.
+ *	 We check memory resouce information and confirm we freed
+ *	 All necessary pages.
+ */
+
+int offline_pages(unsigned long start_pfn,
+		  unsigned long end_pfn,
+		  unsigned long timeout)
+{
+	struct isolation_info *info;
+	struct page *page;
+	LIST_HEAD(pagelist);
+	int ret, nr_pages;
+	unsigned long expire = jiffies + timeout;
+	struct zone *zone;
+	unsigned long pfn, offlined_pages;
+
+	if (start_pfn & (PAGES_PER_SECTION - 1))
+		return -EINVAL;
+	if (end_pfn & (PAGES_PER_SECTION - 1))
+		return -EINVAL;
+
+	zone = page_zone(pfn_to_page(start_pfn));
+
+	if (!is_movable(zone))
+		return -EBUSY;
+
+	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
+		if (!pfn_valid(pfn))
+			continue;
+		page = pfn_to_page(pfn);
+		if (PageSlab(page) ||
+			PageUncached(page) ||
+			PageCompound(page))
+			break;
+	}
+	if (pfn < end_pfn)
+		return -EBUSY;
+
+	info = register_isolation(start_pfn, end_pfn);
+	if (IS_ERR(info))
+		return PTR_ERR(info);
+	/* start memory hot removal */
+
+	ret = capture_isolate_freed_pages(info);
+	if(ret < 0)
+		goto failed_removal;
+
+	nr_pages = end_pfn - start_pfn;
+	pfn = start_pfn;
+repeat:
+	ret = -EAGAIN;
+	if (time_after(jiffies, expire))
+		goto failed_removal;
+	ret = -EINTR;
+	if (signal_pending(current))
+		goto failed_removal;
+
+	lru_add_drain_all();
+
+	for(;pfn < end_pfn;pfn++) {
+		if (!pfn_valid(pfn))
+			continue;
+		page = pfn_to_page(pfn);
+		if (PageReserved(page)) /* ignore Resrved page for now */
+			continue;
+		if (!is_page_isolated(info,page))
+			break;
+	}
+
+	if (pfn != end_pfn) {
+		ret = do_migrate_and_isolate_pages(info, pfn, end_pfn);
+		if (!ret) {
+			cond_resched();
+			goto repeat;
+		} else if (ret < 0) {
+			ret = -EBUSY;
+			goto failed_removal;
+		} else if (ret > 0) {
+			/* some congestion found. sleep a bit */
+			msleep(10);
+			goto repeat;
+		}
+	}
+	/* check memory holes and bootmem */
+	ret = check_removal_success(info);
+	if (ret < 0) {
+		goto failed_removal;
+	}
+	offlined_pages = ret;
+	/* all pages are isolated */
+	detach_isolation_info_zone(info);
+	unuse_all_isolated_pages(info);
+	free_isolation_info(info);
+	zone->present_pages -= offlined_pages;
+	zone->zone_pgdat->node_present_pages -= offlined_pages;
+	totalram_pages -= offlined_pages;
+	num_physpages -= offlined_pages;
+	vm_total_pages = nr_free_pagecache_pages();
+	writeback_set_ratelimit();
+	return 0;
+
+failed_removal:
+	if (ret == -EBUSY) {
+		printk("some unremovable pages are included in %lx to %lx\n",
+			info->start_pfn, info->end_pfn);
+	}
+	/* push back to free_list */
+	detach_isolation_info_zone(info);
+	free_all_isolated_pages(info);
+	free_isolation_info(info);
+	return ret;
+}
+
+#endif
Index: devel-tree-2.6.20-mm2/include/linux/memory_hotplug.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/memory_hotplug.h
+++ devel-tree-2.6.20-mm2/include/linux/memory_hotplug.h
@@ -59,6 +59,7 @@ extern int add_one_highpage(struct page 
 extern void online_page(struct page *page);
 /* VM interface that may be used by firmware interface */
 extern int online_pages(unsigned long, unsigned long);
+extern int offline_pages(unsigned long, unsigned long, unsigned long);
 
 /* reasonably generic interface to expand the physical pages in a zone  */
 extern int __add_pages(struct zone *zone, unsigned long start_pfn,

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [15/16] hot-unplug interface for ia64

[KAMEZAWA Hiroyuki]

Call offline pages from remove_memory().

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
---
 arch/ia64/mm/init.c |   13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

Index: devel-tree-2.6.20-mm2/arch/ia64/mm/init.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/arch/ia64/mm/init.c
+++ devel-tree-2.6.20-mm2/arch/ia64/mm/init.c
@@ -759,7 +759,18 @@ int arch_add_memory(int nid, u64 start, 
 
 int remove_memory(u64 start, u64 size)
 {
-	return -EINVAL;
+	unsigned long start_pfn, end_pfn;
+	unsigned long timeout = 120 * HZ;
+	int ret;
+	start_pfn = start >> PAGE_SHIFT;
+	end_pfn = start_pfn + (size >> PAGE_SHIFT);
+	ret = offline_pages(start_pfn, end_pfn, timeout);
+	if (ret)
+		goto out;
+	/* we can free mem_map at this point */
+out:
+	return ret;
 }
+
 EXPORT_SYMBOL_GPL(remove_memory);
 #endif

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC} memory unplug patchset prep [16/16] migration nocontext

[KAMEZAWA Hiroyuki]

Delaying freeing anon_vma until migration finishes.

We cannot trust page->mapping (of ANON) when page_mapcount(page) ==0.

page migration puts page_mocount(page) to be 0. So we have to
guarantee anon_vma pointed by page->mapping is valid by some hook.

Usual page migration guarantees this by mm->sem. but we can't do it.
So, just delaying freeing anon_vma.

Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

---
 include/linux/migrate.h |    2 ++
 include/linux/rmap.h    |   21 +++++++++++++++++++++
 mm/Kconfig              |   12 ++++++++++++
 mm/memory_hotplug.c     |    4 ++--
 mm/migrate.c            |   35 +++++++++++++++++++++++++++++------
 mm/rmap.c               |   36 +++++++++++++++++++++++++++++++++++-
 6 files changed, 101 insertions(+), 9 deletions(-)

Index: devel-tree-2.6.20-mm2/mm/migrate.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/migrate.c
+++ devel-tree-2.6.20-mm2/mm/migrate.c
@@ -601,7 +601,7 @@ static int move_to_new_page(struct page 
  * to the newly allocated page in newpage.
  */
 static int unmap_and_move(new_page_t get_new_page, unsigned long private,
-			struct page *page, int force)
+			struct page *page, int force, int nocontext)
 {
 	int rc = 0;
 	int *result = NULL;
@@ -626,7 +626,10 @@ static int unmap_and_move(new_page_t get
 			goto unlock;
 		wait_on_page_writeback(page);
 	}
-
+	if (PageAnon(page) && nocontext) {
+		/* hold this anon_vma until remove_migration_ptes() finishes */
+		anon_vma_hold(page);
+	}
 	/*
 	 * Establish migration ptes or remove ptes
 	 */
@@ -634,8 +637,14 @@ static int unmap_and_move(new_page_t get
 	if (!page_mapped(page))
 		rc = move_to_new_page(newpage, page);
 
-	if (rc)
+	if (rc) {
 		remove_migration_ptes(page, page);
+		if (PageAnon(page) && nocontext)
+			anon_vma_release(page);
+	} else {
+		if (PageAnon(newpage) && nocontext)
+			anon_vma_release(page);
+	}
 
 unlock:
 	unlock_page(page);
@@ -680,8 +689,8 @@ move_newpage:
  *
  * Return: Number of pages not migrated or error code.
  */
-int migrate_pages(struct list_head *from,
-		new_page_t get_new_page, unsigned long private)
+static int __migrate_pages(struct list_head *from,
+		new_page_t get_new_page, unsigned long private, int nocontext)
 {
 	int retry = 1;
 	int nr_failed = 0;
@@ -701,7 +710,7 @@ int migrate_pages(struct list_head *from
 			cond_resched();
 
 			rc = unmap_and_move(get_new_page, private,
-						page, pass > 2);
+						page, pass > 2, nocontext);
 
 			switch(rc) {
 			case -ENOMEM:
@@ -731,6 +740,20 @@ out:
 	return nr_failed + retry;
 }
 
+int migrate_pages(struct list_head *from,
+	new_page_t get_new_page, unsigned long private)
+{
+	return __migrate_pages(from, get_new_page, private, 0);
+}
+
+#ifdef CONFIG_MIGRATION_NOCONTEXT
+int migrate_pages_nocontext(struct list_head *from,
+	new_page_t get_new_page, unsigned long private)
+{
+	return __migrate_pages(from, get_new_page, private, 1);
+}
+#endif
+
 #ifdef CONFIG_NUMA
 /*
  * Move a list of individual pages
Index: devel-tree-2.6.20-mm2/include/linux/rmap.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/rmap.h
+++ devel-tree-2.6.20-mm2/include/linux/rmap.h
@@ -26,6 +26,9 @@
 struct anon_vma {
 	spinlock_t lock;	/* Serialize access to vma list */
 	struct list_head head;	/* List of private "related" vmas */
+#ifdef CONFIG_MIGRATION_NOCONTEXT
+	atomic_t	hold;	/* == 0 if we can free this immediately */
+#endif
 };
 
 #ifdef CONFIG_MMU
@@ -37,10 +40,14 @@ static inline struct anon_vma *anon_vma_
 	return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
 }
 
+#ifndef CONFIG_MIGRATION_NOCONTEXT
 static inline void anon_vma_free(struct anon_vma *anon_vma)
 {
 	kmem_cache_free(anon_vma_cachep, anon_vma);
 }
+#else
+extern void anon_vma_free(struct anon_vma *anon_vma);
+#endif
 
 static inline void anon_vma_lock(struct vm_area_struct *vma)
 {
@@ -74,6 +81,20 @@ void page_add_new_anon_rmap(struct page 
 void page_add_file_rmap(struct page *);
 void page_remove_rmap(struct page *, struct vm_area_struct *);
 
+#ifdef CONFIG_MIGRATION_NOCONTEXT
+/*
+ * While Page migration without any process context, we doesn't have
+ * mm->sem. Because page->mapcount goes down to 0 while migration,
+ * we cannot trust page->mapping value.
+ * THese two functions prevents anon_vma from being freed  while
+ * migration.
+ */
+void anon_vma_hold(struct page *page);
+void anon_vma_release(struct page *page);
+#else
+#define anon_vma_hold(page)	do{}while(0)
+#define anon_vma_release(page)	do{}while(0)
+#endif
 /**
  * page_dup_rmap - duplicate pte mapping to a page
  * @page:	the page to add the mapping to
Index: devel-tree-2.6.20-mm2/mm/rmap.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/rmap.c
+++ devel-tree-2.6.20-mm2/mm/rmap.c
@@ -155,8 +155,9 @@ void anon_vma_unlink(struct vm_area_stru
 	empty = list_empty(&anon_vma->head);
 	spin_unlock(&anon_vma->lock);
 
-	if (empty)
+	if (empty) {
 		anon_vma_free(anon_vma);
+	}
 }
 
 static void anon_vma_ctor(void *data, struct kmem_cache *cachep,
@@ -939,3 +940,36 @@ int try_to_unmap(struct page *page, int 
 	return ret;
 }
 
+#ifdef CONFIG_MIGRATION_NOCONTEXT
+
+void anon_vma_free(struct anon_vma *anon)
+{
+	if (atomic_read(&anon->hold) == 0) {
+		kmem_cache_free(anon_vma_cachep, anon);
+	}
+}
+
+void anon_vma_hold(struct page *page)
+{
+	struct anon_vma *anon_vma;
+	anon_vma = page_lock_anon_vma(page);
+	if (!anon_vma)
+		return;
+	atomic_set(&anon_vma->hold, 1);
+	spin_unlock(&anon_vma->lock);
+}
+
+void anon_vma_release(struct page *page)
+{
+	struct anon_vma *anon_vma;
+	int empty;
+	anon_vma = page_lock_anon_vma(page);
+	if (!anon_vma)
+		return;
+	atomic_set(&anon_vma->hold, 0);
+	empty = list_empty(&anon_vma->head);
+	spin_unlock(&anon_vma->lock);
+	if (empty)
+		anon_vma_free(anon_vma);
+}
+#endif
Index: devel-tree-2.6.20-mm2/mm/Kconfig
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/Kconfig
+++ devel-tree-2.6.20-mm2/mm/Kconfig
@@ -132,6 +132,7 @@ config MEMORY_HOTREMOVE
 	select	ZONE_MOVABLE
 	select	MIGRATION
 	select  PAGE_ISOLATION
+	select  MIGRATION_NOCONTEXT
 
 # Heavily threaded applications may benefit from splitting the mm-wide
 # page_table_lock, so that faults on different parts of the user address
@@ -159,6 +160,17 @@ config MIGRATION
 	  example on NUMA systems to put pages nearer to the processors accessing
 	  the page.
 
+config MIGRATION_NOCONTEXT
+	bool "Page migration without process context"
+	def_bool y
+	depends on MEMORY_HOTREMOVE
+	help
+	  When Memory-Hotremove is executed, page migraion runs.
+	  But a process which does page migraion doesn't have context of
+	  migration target pages. This has a small race condition.
+	  If this config is selected, some workaround for fix them is enabled.
+	  This may be add slight performance influence.
+
 config RESOURCES_64BIT
 	bool "64 bit Memory and IO resources (EXPERIMENTAL)" if (!64BIT && EXPERIMENTAL)
 	default 64BIT
Index: devel-tree-2.6.20-mm2/include/linux/migrate.h
===================================================================
--- devel-tree-2.6.20-mm2.orig/include/linux/migrate.h
+++ devel-tree-2.6.20-mm2/include/linux/migrate.h
@@ -11,6 +11,8 @@ extern int putback_lru_pages(struct list
 extern int migrate_page(struct address_space *,
 			struct page *, struct page *);
 extern int migrate_pages(struct list_head *l, new_page_t x, unsigned long);
+extern int migrate_pages_nocontext(struct list_head *l,
+					new_page_t x, unsigned long);
 
 extern int fail_migrate_page(struct address_space *,
 			struct page *, struct page *);
Index: devel-tree-2.6.20-mm2/mm/memory_hotplug.c
===================================================================
--- devel-tree-2.6.20-mm2.orig/mm/memory_hotplug.c
+++ devel-tree-2.6.20-mm2/mm/memory_hotplug.c
@@ -345,7 +345,7 @@ static int do_migrate_and_isolate_pages(
 		if (!pfn_valid(pfn))  /* never happens in sparsemem */
 			continue;
 		page = pfn_to_page(pfn);
-		if (is_page_isolated(info,page))
+		if (PageReserved(page))
 			continue;
 		ret = isolate_lru_page(page, &source);
 
@@ -367,7 +367,7 @@ static int do_migrate_and_isolate_pages(
 	if (list_empty(&source))
 		goto out;
 	/* this function returns # of failed pages */
-	ret = migrate_pages(&source, hotremove_migrate_alloc,
+	ret = migrate_pages_nocontext(&source, hotremove_migrate_alloc,
 			   (unsigned long)info);
 out:
 	return ret;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [0/16]

[David Rientjes]

On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:

> My purpose is to show how memory-unplug can be implemented on ZONE_MOVABLE.
> *Any* comments are welcome. This patch just support ia64, which I can test.
> If you want me to add arch support, please mail me.
> 

It's great to see progress being made in the memory hot-unplug direction.  
The implementation seems to be following the plan you had in our 
coversations from the end of last year.

> This patch is a bit old and against 2.6.20-mm2. I'll rebase this and reflect
> your comments in the next post (may not soon).
> Well booted on ia64 and passed *quick* memory offline test.
> 

When it's rebased, it might be better to apply it to the latest -mm with 
Mel Gorman's patch series merged.  They're in 2.6.21-rc2-mm2.

It appears as though you're using a subset of the ZONE_MOVABLE patches as 
posted from March 1.  What about the additional capabilities of 
ZONE_MOVABLE that aren't included in your patchset, such as allowing 
hugetlb pages from being allocated under GFP_HIGH_MOVABLE, are going to 
need to be changed to support memory hot-unplug?  Since your patchset 
wasn't based on the entire ZONE_MOVABLE set, it leads me to believe that 
some of what it does diverges with the memory hot-unplug use case.

Are you aiming to target both ia64 and x86_64 with this patchset or are 
you focusing on ia64 exclusively at the moment?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [1/16] zone ids cleanup

[David Rientjes]

On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:

> OThis patch defines ZONE_DMA,DMA32,HIGHMEM on *any* config.
> MAX_NR_ZONES is unchanged and not-configured zones's id is greater than it.
> Now, you can check zone is configured or not by (zone_id < MAX_NR_ZONES).
> 
> Good bye #ifdefs. Compiler will do enough work, I think.
> 

Eliminating the abundance of #ifdef's certainly seems like a worthwhile 
goal.

Few comments below.

> Index: devel-tree-2.6.20-mm2/include/linux/mmzone.h
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/include/linux/mmzone.h
> +++ devel-tree-2.6.20-mm2/include/linux/mmzone.h
> @@ -142,9 +142,24 @@ enum zone_type {
>  	 */
>  	ZONE_HIGHMEM,
>  #endif
> -	MAX_NR_ZONES
> +	MAX_NR_ZONES,
> +#ifndef CONFIG_ZONE_DMA
> +	ZONE_DMA,
> +#endif
> +#ifndef CONFIG_ZONE_DMA32
> +	ZONE_DMA32,
> +#endif
> +#ifndef CONFIG_HIGHMEM
> +	ZONE_HIGHMEM,
> +#endif
> +	MAX_POSSIBLE_ZONES
>  };
>  
> +static inline int is_configured_zone(enum zone_type type)
> +{
> +	return (type < MAX_NR_ZONES);
> +}
> +
>  /*
>   * When a memory allocation must conform to specific limitations (such
>   * as being suitable for DMA) the caller will pass in hints to the
> @@ -500,11 +515,7 @@ static inline int populated_zone(struct 
>  
>  static inline int is_highmem_idx(enum zone_type idx)
>  {
> -#ifdef CONFIG_HIGHMEM
>  	return (idx == ZONE_HIGHMEM);
> -#else
> -	return 0;
> -#endif
>  }
>  

Doesn't this need a check for is_configured_zone(idx) as well since this 
will return 1 if we pass in idx == ZONE_HIGHMEM even though it's above 
MAX_NR_ZONES?

>  static inline int is_normal_idx(enum zone_type idx)
> @@ -520,11 +531,7 @@ static inline int is_normal_idx(enum zon
>   */
>  static inline int is_highmem(struct zone *zone)
>  {
> -#ifdef CONFIG_HIGHMEM
>  	return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
> -#else
> -	return 0;
> -#endif
>  }
>  

The only call site for this after your patchset is applied is in i386 code 
which you can probably remove with the identity idx.

>  static inline int is_normal(struct zone *zone)
> @@ -534,20 +541,12 @@ static inline int is_normal(struct zone 
>  
>  static inline int is_dma32(struct zone *zone)
>  {
> -#ifdef CONFIG_ZONE_DMA32
>  	return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
> -#else
> -	return 0;
> -#endif
>  }
>  
>  static inline int is_dma(struct zone *zone)
>  {
> -#ifdef CONFIG_ZONE_DMA
>  	return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
> -#else
> -	return 0;
> -#endif
>  }
>  

Neither is_dma32() nor is_dma() are even used anymore.

>  /* These two functions are used to setup the per zone pages min values */
> Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
> +++ devel-tree-2.6.20-mm2/mm/page_alloc.c
> @@ -72,32 +72,34 @@ static void __free_pages_ok(struct page 
>   * TBD: should special case ZONE_DMA32 machines here - in those we normally
>   * don't need any ZONE_NORMAL reservation
>   */
> -int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
> -#ifdef CONFIG_ZONE_DMA
> -	 256,
> -#endif
> -#ifdef CONFIG_ZONE_DMA32
> -	 256,
> -#endif
> -#ifdef CONFIG_HIGHMEM
> -	 32
> -#endif
> -};
> +int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
>  

Probably an easier way to initialize these instead of 
zone_variables_init() is like this:

	int sysctl_lowmem_reserve_ratio[MAX_POSSIBLE_ZONES-1] = {
		[ZONE_DMA]	= 256,
		[ZONE_DMA32]	= 256,
		[ZONE_HIGHMEM]	= 32 };

>  EXPORT_SYMBOL(totalram_pages);
>  
> -static char * const zone_names[MAX_NR_ZONES] = {
> -#ifdef CONFIG_ZONE_DMA
> -	 "DMA",
> -#endif
> -#ifdef CONFIG_ZONE_DMA32
> -	 "DMA32",
> -#endif
> -	 "Normal",
> -#ifdef CONFIG_HIGHMEM
> -	 "HighMem"
> -#endif
> -};
> +static char *zone_names[MAX_POSSIBLE_ZONES];
> +

Likewise:

	static const char *zone_names[MAX_POSSIBLE_ZONES-1] = {
		[ZONE_DMA]	= "DMA",
		[ZONE_DMA32]	= "DMA32",
		[ZONE_NORMAL]	= "Normal",
		[ZONE_HIGHMEM]	= "HighMem" };

> +static char name_dma[] = "DMA";
> +static char name_dma32[] = "DMA32";
> +static char name_normal[] = "Normal";
> +static char name_highmem[] = "Highmem";
> +
> +static inline void __meminit zone_variables_init(void)
> +{
> +	if (zone_names[0] != NULL)
> +		return;
> +	zone_names[ZONE_DMA] = name_dma;
> +	zone_names[ZONE_DMA32] = name_dma32;
> +	zone_names[ZONE_NORMAL] = name_normal;
> +	zone_names[ZONE_HIGHMEM] = name_highmem;
> +
> +	/* ZONE below NORAML has ratio 256 */
> +	if (is_configured_zone(ZONE_DMA))
> +		sysctl_lowmem_reserve_ratio[ZONE_DMA] = 256;
> +	if (is_configured_zone(ZONE_DMA32))
> +		sysctl_lowmem_reserve_ratio[ZONE_DMA32] = 256;
> +	if (is_configured_zone(ZONE_HIGHMEM))
> +		sysctl_lowmem_reserve_ratio[ZONE_HIGHMEM] = 32;
> +}
>  

Then you can avoid this.

		David

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [2/16] gathering alloc_zeroed_user_highpage()

[David Rientjes]

On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:

> Definitions of alloc_zeroed_user_highpage() is scattered.
> This patch gathers them to linux/highmem.h
> 
> To do so, added CONFIG_ARCH_HAS_PREZERO_USERPAGE and
> CONFIG_ARCH_HAS_FLUSH_USERNEWZEROPAGE.
> 

Previous to this patch, __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE was never 
configurable by the user and was totally dependant on the architecture, 
which seems appropriate.  Are there cases when a user would actually 
prefer to disable the new CONFIG_ARCH_HAS_PREZERO_USERPAGE to avoid 
__GFP_ZERO allocations?

> Index: devel-tree-2.6.20-mm2/include/linux/highmem.h
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/include/linux/highmem.h
> +++ devel-tree-2.6.20-mm2/include/linux/highmem.h
> @@ -60,8 +60,22 @@ static inline void clear_user_highpage(s
>  	/* Make sure this page is cleared on other CPU's too before using it */
>  	smp_wmb();
>  }
> +#ifndef CONFIG_ARCH_HAS_FLUSH_USER_NEWZEROPAGE
> +#define flush_user_newzeroapge(page)	do{}while(0);
> +#endif
>  

Well, I guess this supports my point.  It doesn't appear as it was ever 
tested in disabling __GFP_ZERO allocations because 
flush_user_newzeropage() is misspelled above so it wouldn't even compile.

		David

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [3/16] define is_identity_mapped

[David Rientjes]

On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:

> Index: devel-tree-2.6.20-mm2/include/linux/mmzone.h
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/include/linux/mmzone.h
> +++ devel-tree-2.6.20-mm2/include/linux/mmzone.h
> @@ -523,6 +523,13 @@ static inline int is_normal_idx(enum zon
>  	return (idx == ZONE_NORMAL);
>  }
>  
> +static inline int is_identity_map_idx(enum zone_type idx)
> +{
> +	if (is_configured_zone(ZONE_HIGHMEM))
> +		return (idx < ZONE_HIGHMEM);
> +	else
> +		return 1;
> +}
>  /**
>   * is_highmem - helper function to quickly check if a struct zone is a 
>   *              highmem zone or not.  This is an attempt to keep references
> @@ -549,6 +556,14 @@ static inline int is_dma(struct zone *zo
>  	return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
>  }
>  
> +static inline int is_identity_map(struct zone *zone)
> +{
> +	if (is_configured_zone(ZONE_HIGHMEM)
> +		return zone_idx(zone) < ZONE_HIGHMEM;
> +	else
> +		return 1;
> +}
> +

is_identity_map() isn't specific to any particular architecture nor is it 
dependent on a configuration option.  Since there's a missing ) in its 
conditional, I'm wondering how this entire patch was ever tested.

> Index: devel-tree-2.6.20-mm2/include/linux/page-flags.h
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/include/linux/page-flags.h
> +++ devel-tree-2.6.20-mm2/include/linux/page-flags.h
> @@ -162,7 +162,7 @@ static inline void SetPageUptodate(struc
>  #define __ClearPageSlab(page)	__clear_bit(PG_slab, &(page)->flags)
>  
>  #ifdef CONFIG_HIGHMEM
> -#define PageHighMem(page)	is_highmem(page_zone(page))
> +#define PageHighMem(page)	(!is_identitiy_map(page_zone(page)))
>  #else
>  #define PageHighMem(page)	0 /* needed to optimize away at compile time */
>  #endif

I assume this should be defined to !is_identity_map(page_zone(page)).

		David

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [4/16] ZONE_MOVABLE

[David Rientjes]

On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:

> Index: devel-tree-2.6.20-mm2/include/linux/mmzone.h
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/include/linux/mmzone.h
> +++ devel-tree-2.6.20-mm2/include/linux/mmzone.h
> @@ -142,6 +142,16 @@ enum zone_type {
>  	 */
>  	ZONE_HIGHMEM,
>  #endif
> +#ifdef CONFIG_ZONE_MOVABLE
> +	/*
> +	 * This memory area is used only for migratable pages.
> +	 * We have a chance to hot-remove memory in this zone.
> +	 * Currently, anonymous memory and usual page cache etc. are included.
> +	 * if HIGHMEM is configured, MOVABLE zone is treated as
> +         * not-direct-mapped-memory for kernel;.
> +	 */
> +	ZONE_MOVABLE,
> +#endif
>  	MAX_NR_ZONES,
>  #ifndef CONFIG_ZONE_DMA
>  	ZONE_DMA,
> @@ -152,6 +162,9 @@ enum zone_type {
>  #ifndef CONFIG_HIGHMEM
>  	ZONE_HIGHMEM,
>  #endif
> +#ifndef CONFIG_ZONE_MOVABLE
> +	ZONE_MOVABLE,
> +#endif
>  	MAX_POSSIBLE_ZONES
>  };
>  
> @@ -172,13 +185,18 @@ static inline int is_configured_zone(enu
>   * Count the active zones.  Note that the use of defined(X) outside
>   * #if and family is not necessarily defined so ensure we cannot use
>   * it later.  Use __ZONE_COUNT to work out how many shift bits we need.
> + *
> + * Assumes ZONE_DMA32,ZONE_HIGHMEM, ZONE_MOVABLE can't be configured at
> + * the same time.
>   */
>  #define __ZONE_COUNT (			\
>  	  defined(CONFIG_ZONE_DMA)	\
>  	+ defined(CONFIG_ZONE_DMA32)	\
>  	+ 1				\
>  	+ defined(CONFIG_HIGHMEM)	\
> +	+ defined(CONFIG_ZONE_MOVABLE) \
>  )
> +
>  #if __ZONE_COUNT < 2
>  #define ZONES_SHIFT 0
>  #elif __ZONE_COUNT <= 2
> @@ -513,6 +531,11 @@ static inline int populated_zone(struct 
>  	return (!!zone->present_pages);
>  }
>  
> +static inline int is_movable_dix(enum zone_type idx)
> +{
> +	return (idx == ZONE_MOVABLE);
> +}
> +

Should be is_movable_idx() maybe?  I assume this function is here for 
completeness since it's never referenced in the patchset.

> Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
> +++ devel-tree-2.6.20-mm2/mm/page_alloc.c
> @@ -82,6 +82,7 @@ static char name_dma[] = "DMA";
>  static char name_dma32[] = "DMA32";
>  static char name_normal[] = "Normal";
>  static char name_highmem[] = "Highmem";
> +static char name_movable[] = "Movable";
>  
>  static inline void __meminit zone_variables_init(void)
>  {
> @@ -91,6 +92,7 @@ static inline void __meminit zone_variab
>  	zone_names[ZONE_DMA32] = name_dma32;
>  	zone_names[ZONE_NORMAL] = name_normal;
>  	zone_names[ZONE_HIGHMEM] = name_highmem;
> +	zone_names[ZONE_MOVABLE] = name_movable;
>  
>  	/* ZONE below NORAML has ratio 256 */
>  	if (is_configured_zone(ZONE_DMA))
> @@ -99,6 +101,8 @@ static inline void __meminit zone_variab
>  		sysctl_lowmem_reserve_ratio[ZONE_DMA32] = 256;
>  	if (is_configured_zone(ZONE_HIGHMEM))
>  		sysctl_lowmem_reserve_ratio[ZONE_HIGHMEM] = 32;
> +	if (is_configured_zone(ZONE_MOVABLE))
> +		sysctl_lowmem_reserve_ratio[ZONE_MOVABLE] = 32;
>  }
>  
>  int min_free_kbytes = 1024;
> @@ -3065,11 +3069,17 @@ void __init free_area_init_nodes(unsigne
>  	arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
>  	arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
>  	for (i = 1; i < MAX_NR_ZONES; i++) {
> +		if (i == ZONE_MOVABLE)
> +			continue;
>  		arch_zone_lowest_possible_pfn[i] =
>  			arch_zone_highest_possible_pfn[i-1];
>  		arch_zone_highest_possible_pfn[i] =
>  			max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
>  	}
> +	if (is_configured_zone(ZONE_MOVABLE)) {
> +		arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
> +		arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
> +	}
>  
>  	/* Print out the page size for debugging meminit problems */
>  	printk(KERN_DEBUG "sizeof(struct page) = %zd\n", sizeof(struct page));

Aren't the arch_zone_{lowest|highest}_possible_pfn's for ZONE_MOVABLE 
already at 0?  If not, it should definitely be memset early on to avoid 
any possible assignment mistakes amongst all these conditionals.

> Index: devel-tree-2.6.20-mm2/mm/Kconfig
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/mm/Kconfig
> +++ devel-tree-2.6.20-mm2/mm/Kconfig
> @@ -163,6 +163,10 @@ config ZONE_DMA_FLAG
>  	default "0" if !ZONE_DMA
>  	default "1"
>  
> +config ZONE_MOVABLE
> +	bool "Create zones for MOVABLE pages"
> +	depends on ARCH_POPULATES_NODE_MAP
> +	depends on MIGRATION
>  #
>  # Adaptive file readahead
>  #
> 

This patchset is heavily dependent on Mel Gorman's work with ZONE_MOVABLE 
so perhaps it would be better to base it off of the latest -mm with his 
patchset applied?  And if CONFIG_ZONE_MOVABLE wasn't documented in Kconfig 
prior to this, it might be a good opportunity to do so if you're going to 
get community adoption.

		David

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [9/16] create movable zone at boot

[David Rientjes]

On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:

> This patch adds codes for creating movable zones.
> 
> Add 2 kernel paramers.
> - kernel_core_pages=XXX[KMG]
> - kernel_core_ratio=xx
> 

These would never be specified together, right?

> Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
> +++ devel-tree-2.6.20-mm2/mm/page_alloc.c
> @@ -137,12 +137,16 @@ static unsigned long __initdata dma_rese
>    int __initdata nr_nodemap_entries;
>    unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
>    unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
> +  unsigned long __initdata lowest_movable_pfn[MAX_NUMNODES];
> +  unsigned long kernel_core_ratio;
> +  unsigned long kernel_core_pages;
>  #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
>    unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
>    unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
>  #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
>  #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
>  
> +
>  #ifdef CONFIG_DEBUG_VM
>  static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
>  {

You could probably get away with:

	union {
		unsigned long kernel_core_ratio;
		unsigned long kernel_core_pages;
	};

> @@ -2604,6 +2608,8 @@ void __init get_pfn_range_for_nid(unsign
>   */
>  unsigned long __init zone_spanned_pages_in_node(int nid,
>  					unsigned long zone_type,
> +					unsigned long *start_pfn,
> +					unsigned long *end_pfn,
>  					unsigned long *ignored)
>  {
>  	unsigned long node_start_pfn, node_end_pfn;
> @@ -2611,8 +2617,30 @@ unsigned long __init zone_spanned_pages_
>  
>  	/* Get the start and end of the node and zone */
>  	get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
> -	zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
> -	zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
> +	if (start_pfn)
> +		*start_pfn = 0;
> +	if (end_pfn)
> +		*end_pfn = 0;
> +	if (!is_configured_zone(ZONE_MOVABLE) ||
> +		   lowest_movable_pfn[nid] == 0) {
> +		/* we don't use ZONE_MOVABLE */
> +		zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
> +		zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
> +	} else if (zone_type == ZONE_MOVABLE) {
> +		zone_start_pfn = lowest_movable_pfn[nid];
> +		zone_end_pfn = node_end_pfn;
> +	} else {
> +		/* adjust range to lowest_movable_pfn[] */
> +		zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
> +		zone_start_pfn = max(zone_start_pfn, node_start_pfn);
> +
> +		if (zone_start_pfn >= lowest_movable_pfn[nid])
> +			return 0;
> +		zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
> +		zone_end_pfn = min(zone_end_pfn, node_end_pfn);
> +		if (zone_end_pfn > lowest_movable_pfn[nid])
> +			zone_end_pfn = lowest_movable_pfn[nid];
> +	}
>  
>  	/* Check that this node has pages within the zone's required range */
>  	if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)

These hacks of returning start_pfn and end_pfn depending on where it was 
called from and testing for things like start_pfn == end_pfn doesn't make 
much sense.  It'd probably be better to separate this logic out into a 
helper function and then call it from zone_absent_pages_in_node() and 
zone_spanned_pages_in_node(), respectively.

> @@ -2733,20 +2781,115 @@ static void __init calculate_node_totalp
>  	enum zone_type i;
>  
>  	for (i = 0; i < MAX_NR_ZONES; i++)
> -		totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
> +		totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, NULL, NULL,
>  								zones_size);
>  	pgdat->node_spanned_pages = totalpages;
>  
>  	realtotalpages = totalpages;
>  	for (i = 0; i < MAX_NR_ZONES; i++)
>  		realtotalpages -=
> -			zone_absent_pages_in_node(pgdat->node_id, i,
> +			zone_absent_pages_in_node(pgdat->node_id, i, 0, 0,
>  								zholes_size);
>  	pgdat->node_present_pages = realtotalpages;
>  	printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
>  							realtotalpages);
>  }
>  
> +#ifdef CONFIG_ZONE_MOVABLE
> +
> +unsigned long calc_zone_alignment(unsigned long pfn)
> +{
> +#ifdef CONFIG_SPARSEMEM
> +	return (pfn + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK;
> +#else
> +	return (pfn + MAX_ORDER_NR_PAGES - 1) & ~(MAX_ORDER_NR_PAGES - 1)
> +#endif
> +}
> +

Another missing semicolon.

> +static void split_movable_pages(void)
> +{
> +	int i, nid;
> +	unsigned long total_pages, nr_pages, start_pfn, end_pfn, pfn;
> +	long core;
> +	for_each_online_node(nid) {
> +		lowest_movable_pfn[nid] = 0;
> +		pfn = 0;
> +		total_pages = 0;
> +		for_each_active_range_index_in_nid(i, nid) {
> +			start_pfn = early_node_map[i].start_pfn;
> +			end_pfn = early_node_map[i].end_pfn;
> +			total_pages += end_pfn - start_pfn;
> +		}
> +		core = total_pages * kernel_core_ratio/100;
> +		for_each_active_range_index_in_nid(i, nid) {
> +			start_pfn = early_node_map[i].start_pfn;
> +			end_pfn = early_node_map[i].end_pfn;
> +			nr_pages = end_pfn - start_pfn;
> +			if (nr_pages > core) {
> +				pfn = start_pfn + core;
> +				pfn = calc_zone_alignment(pfn);
> +				if (pfn < end_pfn) {
> +					lowest_movable_pfn[nid] = pfn;
> +					break;
> +				} else {
> +					core -= nr_pages;
> +					if (core < 0)
> +						core = 0;
> +				}
> +			} else {
> +				core -= nr_pages;
> +			}
> +		}
> +	}
> +	return;
> +}
> +
> +
> +static void reserve_movable_pages(void)
> +{
> +	memset(lowest_movable_pfn, 0, MAX_NUMNODES);
> +	if (kernel_core_pages) {
> +		alloc_core_pages_from_low();
> +	} else if (kernel_core_ratio) {
> +		split_movable_pages();
> +	}
> +	return;
> +}
> +#else
> +static void reserve_movable_pages(void)
> +{
> +	return;
> +}
> +#endif
>  /*
>   * Set up the zone data structures:
>   *   - mark all pages reserved

reserve_movable_pages() and it's two helper functions, 
alloc_core_pages_from_low() and split_movable_pages(), can be __init?

If so, then both kernel_core_pages and kernel_core_ratio should be 
__initdata.

> Index: devel-tree-2.6.20-mm2/Documentation/kernel-parameters.txt
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/Documentation/kernel-parameters.txt
> +++ devel-tree-2.6.20-mm2/Documentation/kernel-parameters.txt
> @@ -764,6 +764,17 @@ and is between 256 and 4096 characters. 
>  
>  	keepinitrd	[HW,ARM]
>  
> +	kernel_core_pages=nn[KMG] [KNL, BOOT] divide the whole memory into
> +			not-movable and movable. movable memory can be
> +			used only for page cache and user data. This option
> +			specifies the amount of not-movable pages, called core
> +			pages. core pages are allocated from the lower address.
> +
> +	kernel_core_ratio=nn [KND, BOOT] specifies the amount of the core
> +			pages(see kernel_core_pages) by the ratio against
> +			total memory. If NUMA, core pages are allocated for
> +			each node by this ratio. "0" is not allowed.
> +
>  	kstack=N	[IA-32,X86-64] Print N words from the kernel stack
>  			in oops dumps.
>  

This documentation doesn't specify that we can't use both parameters 
together even though we can't.

		David

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [8/16] counter for ZONE_MOVABLE

[David Rientjes]

On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:

> Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
> +++ devel-tree-2.6.20-mm2/mm/page_alloc.c
> @@ -58,6 +58,7 @@ unsigned long totalram_pages __read_most
>  unsigned long totalreserve_pages __read_mostly;
>  long nr_swap_pages;
>  int percpu_pagelist_fraction;
> +unsigned long total_movable_pages __read_mostly;
>  
>  static void __free_pages_ok(struct page *page, unsigned int order);
>  
> @@ -1571,6 +1572,20 @@ static unsigned int nr_free_zone_pages(i
>  	return sum;
>  }
>  
> +unsigned int nr_free_movable_pages(void)
> +{
> +	unsigned long nr_pages = 0;
> +	struct zone *zone;
> +	int nid;
> +	if (is_configured_zone(ZONE_MOVABLE)) {
> +		/* we want to count *only* pages in movable zone */
> +		for_each_online_node(nid) {
> +			zone = &(NODE_DATA(nid)->node_zones[ZONE_MOVABLE]);
> +			nr_pages += zone_page_state(zone, NR_FREE_PAGES);
> +		}
> +	}
> +	return nr_pages;
> +}
>  /*
>   * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
>   */

On each online node, zone should be

	zone = NODE_DATA(nid)->node_sizes + ZONE_MOVABLE;

Also, you should probably only declare this function on #ifdef 
CONFIG_ZONE_MOVABLE and #define it to "do {} while(0)" otherwise.

> @@ -1584,7 +1599,7 @@ unsigned int nr_free_buffer_pages(void)
>   */
>  unsigned int nr_free_pagecache_pages(void)
>  {
> -	return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
> +	return nr_free_zone_pages(gfp_zone(GFP_HIGH_MOVABLE));
>  }
>  
>  /*
> @@ -1633,6 +1648,8 @@ void si_meminfo(struct sysinfo *val)
>  	val->totalhigh = totalhigh_pages;
>  	val->freehigh = nr_free_highpages();
>  	val->mem_unit = PAGE_SIZE;
> +	val->movable = total_movable_pages;
> +	val->free_movable = nr_free_movable_pages();
>  }
>  
>  EXPORT_SYMBOL(si_meminfo);
> @@ -1654,6 +1671,13 @@ void si_meminfo_node(struct sysinfo *val
>  		val->totalhigh = 0;
>  		val->freehigh = 0;
>  	}
> +	if (is_configured_zone(ZONE_MOVABLE)) {
> +		val->movable +=
> +			pgdat->node_zones[ZONE_MOVABLE].present_pages;
> +		val->free_movable +=
> +			zone_page_state(&pgdat->node_zones[ZONE_MOVABLE],
> +				NR_FREE_PAGES);
> +	}
>  	val->mem_unit = PAGE_SIZE;
>  }
>  #endif

Don't you want assignments here instead of accumulations?  val->movable 
and val->free_movable probably shouldn't be the only members in 
si_meminfo_node() that accumulate.

Your first patch in this patchset actually sets val->totalhigh and 
val->freehigh both to 0 in the !is_configured_zone(ZONE_HIGHMEM) case.  Do 
these need the same assignments for movable and free_movable in the 
!is_configured_zone(ZONE_MOVABLE) case?

> Index: devel-tree-2.6.20-mm2/include/linux/kernel.h
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/include/linux/kernel.h
> +++ devel-tree-2.6.20-mm2/include/linux/kernel.h
> @@ -329,6 +329,8 @@ struct sysinfo {
>  	unsigned short pad;		/* explicit padding for m68k */
>  	unsigned long totalhigh;	/* Total high memory size */
>  	unsigned long freehigh;		/* Available high memory size */
> +	unsigned long movable;		/* pages used only for data */
> +	unsigned long free_movable;	/* Avaiable pages in movable */
>  	unsigned int mem_unit;		/* Memory unit size in bytes */
>  	char _f[20-2*sizeof(long)-sizeof(int)];	/* Padding: libc5 uses this.. */
>  };

Please add #ifdef's to CONFIG_ZONE_MOVABLE around these members in struct 
sysinfo so we incur no penalty if we choose not to enable this option.

> Index: devel-tree-2.6.20-mm2/mm/vmstat.c
> ===================================================================
> --- devel-tree-2.6.20-mm2.orig/mm/vmstat.c
> +++ devel-tree-2.6.20-mm2/mm/vmstat.c
> @@ -426,8 +426,14 @@ const struct seq_operations fragmentatio
>  #define TEXT_FOR_HIGHMEM(xx)
>  #endif
>  
> +#ifdef CONFIG_ZONE_MOVABLE
> +#define TEXT_FOR_MOVABLE(xx) xx "_movable",
> +#else
> +#define TXT_FOR_MOVABLE(xx)
> +#endif
> +
>  #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
> -					TEXT_FOR_HIGHMEM(xx)
> +					TEXT_FOR_HIGHMEM(xx) TEXT_FOR_MOVABLE(xx)
>  
>  static const char * const vmstat_text[] = {
>  	/* Zoned VM counters */
> 

This broke my build because TEXT_FOR_MOVABLE() is misspelled on 
!CONFIG_ZONE_MOVABLE.

		David

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [1/16] zone ids cleanup

[KAMEZAWA Hiroyuki]

On Tue, 6 Mar 2007 07:36:30 -0800 (PST)
David Rientjes <rientjes@google.com> wrote:

> >  static inline int is_highmem_idx(enum zone_type idx)
> >  {
> > -#ifdef CONFIG_HIGHMEM
> >  	return (idx == ZONE_HIGHMEM);
> > -#else
> > -	return 0;
> > -#endif
> >  }
> >  
> 
> Doesn't this need a check for is_configured_zone(idx) as well since this 
> will return 1 if we pass in idx == ZONE_HIGHMEM even though it's above 
> MAX_NR_ZONES?
Hmm, I'll add
==
BUG_ON(idx >= MAX_NR_ZONES)
==
here.

> 
> >  static inline int is_normal_idx(enum zone_type idx)
> > @@ -520,11 +531,7 @@ static inline int is_normal_idx(enum zon
> >   */
> >  static inline int is_highmem(struct zone *zone)
> >  {
> > -#ifdef CONFIG_HIGHMEM
> >  	return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
> > -#else
> > -	return 0;
> > -#endif
> >  }
> >  
> 
> The only call site for this after your patchset is applied is in i386 code 
> which you can probably remove with the identity idx.
Ok, look into.


> 
> >  static inline int is_normal(struct zone *zone)
> > @@ -534,20 +541,12 @@ static inline int is_normal(struct zone 
> >  
> >  static inline int is_dma32(struct zone *zone)
> >  {
> > -#ifdef CONFIG_ZONE_DMA32
> >  	return zone == zone->zone_pgdat->node_zones + ZONE_DMA32;
> > -#else
> > -	return 0;
> > -#endif
> >  }
> >  
> >  static inline int is_dma(struct zone *zone)
> >  {
> > -#ifdef CONFIG_ZONE_DMA
> >  	return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
> > -#else
> > -	return 0;
> > -#endif
> >  }
> >  
> 
> Neither is_dma32() nor is_dma() are even used anymore.

I see. maybe removing entire call patch should be applied before this.

> 
> >  /* These two functions are used to setup the per zone pages min values */
> > Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
> > +++ devel-tree-2.6.20-mm2/mm/page_alloc.c
> > @@ -72,32 +72,34 @@ static void __free_pages_ok(struct page 
> >   * TBD: should special case ZONE_DMA32 machines here - in those we normally
> >   * don't need any ZONE_NORMAL reservation
> >   */
> > -int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
> > -#ifdef CONFIG_ZONE_DMA
> > -	 256,
> > -#endif
> > -#ifdef CONFIG_ZONE_DMA32
> > -	 256,
> > -#endif
> > -#ifdef CONFIG_HIGHMEM
> > -	 32
> > -#endif
> > -};
> > +int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
> >  
> 
> Probably an easier way to initialize these instead of 
> zone_variables_init() is like this:
> 
> 	int sysctl_lowmem_reserve_ratio[MAX_POSSIBLE_ZONES-1] = {
> 		[ZONE_DMA]	= 256,
> 		[ZONE_DMA32]	= 256,
> 		[ZONE_HIGHMEM]	= 32 };
> 

AH, I didn't know this initialization method. thanks. will try.

-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [2/16] gathering alloc_zeroed_user_highpage()

[KAMEZAWA Hiroyuki]

On Tue, 6 Mar 2007 07:54:29 -0800 (PST)
David Rientjes <rientjes@google.com> wrote:

> On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:
> 
> > Definitions of alloc_zeroed_user_highpage() is scattered.
> > This patch gathers them to linux/highmem.h
> > 
> > To do so, added CONFIG_ARCH_HAS_PREZERO_USERPAGE and
> > CONFIG_ARCH_HAS_FLUSH_USERNEWZEROPAGE.
> > 
> 
> Previous to this patch, __HAVE_ARCH_ALLOC_ZEROED_USER_HIGHPAGE was never 
> configurable by the user and was totally dependant on the architecture, 
> which seems appropriate.  Are there cases when a user would actually 
> prefer to disable the new CONFIG_ARCH_HAS_PREZERO_USERPAGE to avoid 
> __GFP_ZERO allocations?
> 
no case. I like CONFIG_ARCH_xx rather than #define in header file.



> > Index: devel-tree-2.6.20-mm2/include/linux/highmem.h
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/include/linux/highmem.h
> > +++ devel-tree-2.6.20-mm2/include/linux/highmem.h
> > @@ -60,8 +60,22 @@ static inline void clear_user_highpage(s
> >  	/* Make sure this page is cleared on other CPU's too before using it */
> >  	smp_wmb();
> >  }
> > +#ifndef CONFIG_ARCH_HAS_FLUSH_USER_NEWZEROPAGE
> > +#define flush_user_newzeroapge(page)	do{}while(0);
> > +#endif
> >  
> 
> Well, I guess this supports my point.  It doesn't appear as it was ever 
> tested in disabling __GFP_ZERO allocations because 
> flush_user_newzeropage() is misspelled above so it wouldn't even compile.
> 
Ah, okay. I'll add i386 to my testset, at least. 

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [3/16] define is_identity_mapped

[KAMEZAWA Hiroyuki]

On Tue, 6 Mar 2007 07:55:54 -0800 (PST)
David Rientjes <rientjes@google.com> wrote:

> On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:
> 
> > Index: devel-tree-2.6.20-mm2/include/linux/mmzone.h
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/include/linux/mmzone.h
> > +++ devel-tree-2.6.20-mm2/include/linux/mmzone.h
> > @@ -523,6 +523,13 @@ static inline int is_normal_idx(enum zon
> >  	return (idx == ZONE_NORMAL);
> >  }
> >  
> > +static inline int is_identity_map_idx(enum zone_type idx)
> > +{
> > +	if (is_configured_zone(ZONE_HIGHMEM))
> > +		return (idx < ZONE_HIGHMEM);
> > +	else
> > +		return 1;
> > +}
> >  /**
> >   * is_highmem - helper function to quickly check if a struct zone is a 
> >   *              highmem zone or not.  This is an attempt to keep references
> > @@ -549,6 +556,14 @@ static inline int is_dma(struct zone *zo
> >  	return zone == zone->zone_pgdat->node_zones + ZONE_DMA;
> >  }
> >  
> > +static inline int is_identity_map(struct zone *zone)
> > +{
> > +	if (is_configured_zone(ZONE_HIGHMEM)
> > +		return zone_idx(zone) < ZONE_HIGHMEM;
> > +	else
> > +		return 1;
> > +}
> > +
> 
> is_identity_map() isn't specific to any particular architecture nor is it 
> dependent on a configuration option.  Since there's a missing ) in its 
> conditional, I'm wondering how this entire patch was ever tested.
> 
I tested and my tree looks fine...maybe this is patch refresh miss...sorry.


> > Index: devel-tree-2.6.20-mm2/include/linux/page-flags.h
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/include/linux/page-flags.h
> > +++ devel-tree-2.6.20-mm2/include/linux/page-flags.h
> > @@ -162,7 +162,7 @@ static inline void SetPageUptodate(struc
> >  #define __ClearPageSlab(page)	__clear_bit(PG_slab, &(page)->flags)
> >  
> >  #ifdef CONFIG_HIGHMEM
> > -#define PageHighMem(page)	is_highmem(page_zone(page))
> > +#define PageHighMem(page)	(!is_identitiy_map(page_zone(page)))
> >  #else
> >  #define PageHighMem(page)	0 /* needed to optimize away at compile time */
> >  #endif
> 
> I assume this should be defined to !is_identity_map(page_zone(page)).
> 
ok.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [4/16] ZONE_MOVABLE

[KAMEZAWA Hiroyuki]

On Tue, 6 Mar 2007 08:06:33 -0800 (PST)
David Rientjes <rientjes@google.com> wrote:

> > +static inline int is_movable_dix(enum zone_type idx)
> > +{
> > +	return (idx == ZONE_MOVABLE);
> > +}
> > +
> 
> Should be is_movable_idx() maybe?  I assume this function is here for 
> completeness since it's never referenced in the patchset.
> 
yes.. this is never called. I may drop this function.

> > Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
> > +++ devel-tree-2.6.20-mm2/mm/page_alloc.c
> > @@ -82,6 +82,7 @@ static char name_dma[] = "DMA";
> >  static char name_dma32[] = "DMA32";
> >  static char name_normal[] = "Normal";
> >  static char name_highmem[] = "Highmem";
> > +static char name_movable[] = "Movable";
> >  
> >  static inline void __meminit zone_variables_init(void)
> >  {
> > @@ -91,6 +92,7 @@ static inline void __meminit zone_variab
> >  	zone_names[ZONE_DMA32] = name_dma32;
> >  	zone_names[ZONE_NORMAL] = name_normal;
> >  	zone_names[ZONE_HIGHMEM] = name_highmem;
> > +	zone_names[ZONE_MOVABLE] = name_movable;
> >  
> >  	/* ZONE below NORAML has ratio 256 */
> >  	if (is_configured_zone(ZONE_DMA))
> > @@ -99,6 +101,8 @@ static inline void __meminit zone_variab
> >  		sysctl_lowmem_reserve_ratio[ZONE_DMA32] = 256;
> >  	if (is_configured_zone(ZONE_HIGHMEM))
> >  		sysctl_lowmem_reserve_ratio[ZONE_HIGHMEM] = 32;
> > +	if (is_configured_zone(ZONE_MOVABLE))
> > +		sysctl_lowmem_reserve_ratio[ZONE_MOVABLE] = 32;
> >  }
> >  
> >  int min_free_kbytes = 1024;
> > @@ -3065,11 +3069,17 @@ void __init free_area_init_nodes(unsigne
> >  	arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
> >  	arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
> >  	for (i = 1; i < MAX_NR_ZONES; i++) {
> > +		if (i == ZONE_MOVABLE)
> > +			continue;
> >  		arch_zone_lowest_possible_pfn[i] =
> >  			arch_zone_highest_possible_pfn[i-1];
> >  		arch_zone_highest_possible_pfn[i] =
> >  			max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
> >  	}
> > +	if (is_configured_zone(ZONE_MOVABLE)) {
> > +		arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
> > +		arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
> > +	}
> >  
> >  	/* Print out the page size for debugging meminit problems */
> >  	printk(KERN_DEBUG "sizeof(struct page) = %zd\n", sizeof(struct page));
> 
> Aren't the arch_zone_{lowest|highest}_possible_pfn's for ZONE_MOVABLE 
> already at 0?  If not, it should definitely be memset early on to avoid 
> any possible assignment mistakes amongst all these conditionals.
> 
ok.


> > Index: devel-tree-2.6.20-mm2/mm/Kconfig
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/mm/Kconfig
> > +++ devel-tree-2.6.20-mm2/mm/Kconfig
> > @@ -163,6 +163,10 @@ config ZONE_DMA_FLAG
> >  	default "0" if !ZONE_DMA
> >  	default "1"
> >  
> > +config ZONE_MOVABLE
> > +	bool "Create zones for MOVABLE pages"
> > +	depends on ARCH_POPULATES_NODE_MAP
> > +	depends on MIGRATION
> >  #
> >  # Adaptive file readahead
> >  #
> > 
> 
> This patchset is heavily dependent on Mel Gorman's work with ZONE_MOVABLE 
> so perhaps it would be better to base it off of the latest -mm with his 
> patchset applied?  And if CONFIG_ZONE_MOVABLE wasn't documented in Kconfig 
> prior to this, it might be a good opportunity to do so if you're going to 
> get community adoption.
> 
I'm very glad to see the latest -mm including Mel's.
And yes. I'll use his patch set. 

-Kame


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [8/16] counter for ZONE_MOVABLE

[KAMEZAWA Hiroyuki]

On Tue, 6 Mar 2007 08:11:22 -0800 (PST)
David Rientjes <rientjes@google.com> wrote:

> On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:
> 
> > Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
> > +++ devel-tree-2.6.20-mm2/mm/page_alloc.c
> > @@ -58,6 +58,7 @@ unsigned long totalram_pages __read_most
> >  unsigned long totalreserve_pages __read_mostly;
> >  long nr_swap_pages;
> >  int percpu_pagelist_fraction;
> > +unsigned long total_movable_pages __read_mostly;
> >  
> >  static void __free_pages_ok(struct page *page, unsigned int order);
> >  
> > @@ -1571,6 +1572,20 @@ static unsigned int nr_free_zone_pages(i
> >  	return sum;
> >  }
> >  
> > +unsigned int nr_free_movable_pages(void)
> > +{
> > +	unsigned long nr_pages = 0;
> > +	struct zone *zone;
> > +	int nid;
> > +	if (is_configured_zone(ZONE_MOVABLE)) {
> > +		/* we want to count *only* pages in movable zone */
> > +		for_each_online_node(nid) {
> > +			zone = &(NODE_DATA(nid)->node_zones[ZONE_MOVABLE]);
> > +			nr_pages += zone_page_state(zone, NR_FREE_PAGES);
> > +		}
> > +	}
> > +	return nr_pages;
> > +}
> >  /*
> >   * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
> >   */
> 
> On each online node, zone should be
> 
> 	zone = NODE_DATA(nid)->node_sizes + ZONE_MOVABLE;
> 
> Also, you should probably only declare this function on #ifdef 
> CONFIG_ZONE_MOVABLE and #define it to "do {} while(0)" otherwise.
> 
is_configure_zone() does enough work. (But I'll move to the latest -mm.)



> > @@ -1584,7 +1599,7 @@ unsigned int nr_free_buffer_pages(void)
> >   */
> >  unsigned int nr_free_pagecache_pages(void)
> >  {
> > -	return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
> > +	return nr_free_zone_pages(gfp_zone(GFP_HIGH_MOVABLE));
> >  }
> >  
> >  /*
> > @@ -1633,6 +1648,8 @@ void si_meminfo(struct sysinfo *val)
> >  	val->totalhigh = totalhigh_pages;
> >  	val->freehigh = nr_free_highpages();
> >  	val->mem_unit = PAGE_SIZE;
> > +	val->movable = total_movable_pages;
> > +	val->free_movable = nr_free_movable_pages();
> >  }
> >  
> >  EXPORT_SYMBOL(si_meminfo);
> > @@ -1654,6 +1671,13 @@ void si_meminfo_node(struct sysinfo *val
> >  		val->totalhigh = 0;
> >  		val->freehigh = 0;
> >  	}
> > +	if (is_configured_zone(ZONE_MOVABLE)) {
> > +		val->movable +=
> > +			pgdat->node_zones[ZONE_MOVABLE].present_pages;
> > +		val->free_movable +=
> > +			zone_page_state(&pgdat->node_zones[ZONE_MOVABLE],
> > +				NR_FREE_PAGES);
> > +	}
> >  	val->mem_unit = PAGE_SIZE;
> >  }
> >  #endif
> 
> Don't you want assignments here instead of accumulations?  val->movable 
> and val->free_movable probably shouldn't be the only members in 
> si_meminfo_node() that accumulate.
> 
> Your first patch in this patchset actually sets val->totalhigh and 
> val->freehigh both to 0 in the !is_configured_zone(ZONE_HIGHMEM) case.  Do 
> these need the same assignments for movable and free_movable in the 
> !is_configured_zone(ZONE_MOVABLE) case?
> 
> > Index: devel-tree-2.6.20-mm2/include/linux/kernel.h
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/include/linux/kernel.h
> > +++ devel-tree-2.6.20-mm2/include/linux/kernel.h
> > @@ -329,6 +329,8 @@ struct sysinfo {
> >  	unsigned short pad;		/* explicit padding for m68k */
> >  	unsigned long totalhigh;	/* Total high memory size */
> >  	unsigned long freehigh;		/* Available high memory size */
> > +	unsigned long movable;		/* pages used only for data */
> > +	unsigned long free_movable;	/* Avaiable pages in movable */
> >  	unsigned int mem_unit;		/* Memory unit size in bytes */
> >  	char _f[20-2*sizeof(long)-sizeof(int)];	/* Padding: libc5 uses this.. */
> >  };
> 
> Please add #ifdef's to CONFIG_ZONE_MOVABLE around these members in struct 
> sysinfo so we incur no penalty if we choose not to enable this option.
> 
I just do this becasue highmem is not covered by CONFIG_HIGHMEM.

> > Index: devel-tree-2.6.20-mm2/mm/vmstat.c
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/mm/vmstat.c
> > +++ devel-tree-2.6.20-mm2/mm/vmstat.c
> > @@ -426,8 +426,14 @@ const struct seq_operations fragmentatio
> >  #define TEXT_FOR_HIGHMEM(xx)
> >  #endif
> >  
> > +#ifdef CONFIG_ZONE_MOVABLE
> > +#define TEXT_FOR_MOVABLE(xx) xx "_movable",
> > +#else
> > +#define TXT_FOR_MOVABLE(xx)
> > +#endif
> > +
> >  #define TEXTS_FOR_ZONES(xx) TEXT_FOR_DMA(xx) TEXT_FOR_DMA32(xx) xx "_normal", \
> > -					TEXT_FOR_HIGHMEM(xx)
> > +					TEXT_FOR_HIGHMEM(xx) TEXT_FOR_MOVABLE(xx)
> >  
> >  static const char * const vmstat_text[] = {
> >  	/* Zoned VM counters */
> > 
> 
> This broke my build because TEXT_FOR_MOVABLE() is misspelled on 
> !CONFIG_ZONE_MOVABLE.
> 
Ah.. sigh. ok. I'll do better test in the next time.

-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [9/16] create movable zone at boot

[KAMEZAWA Hiroyuki]

On Tue, 6 Mar 2007 08:06:48 -0800 (PST)
David Rientjes <rientjes@google.com> wrote:

> On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:
> 
> > This patch adds codes for creating movable zones.
> > 
> > Add 2 kernel paramers.
> > - kernel_core_pages=XXX[KMG]
> > - kernel_core_ratio=xx
> > 
> 
> These would never be specified together, right?
> 
No. never be specified together.

> > Index: devel-tree-2.6.20-mm2/mm/page_alloc.c
> > ===================================================================
> > --- devel-tree-2.6.20-mm2.orig/mm/page_alloc.c
> > +++ devel-tree-2.6.20-mm2/mm/page_alloc.c
> > @@ -137,12 +137,16 @@ static unsigned long __initdata dma_rese
> >    int __initdata nr_nodemap_entries;
> >    unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
> >    unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
> > +  unsigned long __initdata lowest_movable_pfn[MAX_NUMNODES];
> > +  unsigned long kernel_core_ratio;
> > +  unsigned long kernel_core_pages;
> >  #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
> >    unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
> >    unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
> >  #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
> >  #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
> >  
> > +
> >  #ifdef CONFIG_DEBUG_VM
> >  static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
> >  {
> 
> You could probably get away with:
> 
> 	union {
> 		unsigned long kernel_core_ratio;
> 		unsigned long kernel_core_pages;
> 	};
> 
> > @@ -2604,6 +2608,8 @@ void __init get_pfn_range_for_nid(unsign
> >   */
> >  unsigned long __init zone_spanned_pages_in_node(int nid,
> >  					unsigned long zone_type,
> > +					unsigned long *start_pfn,
> > +					unsigned long *end_pfn,
> >  					unsigned long *ignored)
> >  {
> >  	unsigned long node_start_pfn, node_end_pfn;
> > @@ -2611,8 +2617,30 @@ unsigned long __init zone_spanned_pages_
> >  
> >  	/* Get the start and end of the node and zone */
> >  	get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
> > -	zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
> > -	zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
> > +	if (start_pfn)
> > +		*start_pfn = 0;
> > +	if (end_pfn)
> > +		*end_pfn = 0;
> > +	if (!is_configured_zone(ZONE_MOVABLE) ||
> > +		   lowest_movable_pfn[nid] == 0) {
> > +		/* we don't use ZONE_MOVABLE */
> > +		zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
> > +		zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
> > +	} else if (zone_type == ZONE_MOVABLE) {
> > +		zone_start_pfn = lowest_movable_pfn[nid];
> > +		zone_end_pfn = node_end_pfn;
> > +	} else {
> > +		/* adjust range to lowest_movable_pfn[] */
> > +		zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
> > +		zone_start_pfn = max(zone_start_pfn, node_start_pfn);
> > +
> > +		if (zone_start_pfn >= lowest_movable_pfn[nid])
> > +			return 0;
> > +		zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
> > +		zone_end_pfn = min(zone_end_pfn, node_end_pfn);
> > +		if (zone_end_pfn > lowest_movable_pfn[nid])
> > +			zone_end_pfn = lowest_movable_pfn[nid];
> > +	}
> >  
> >  	/* Check that this node has pages within the zone's required range */
> >  	if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
> 
> These hacks of returning start_pfn and end_pfn depending on where it was 
> called from and testing for things like start_pfn == end_pfn doesn't make 
> much sense.  It'd probably be better to separate this logic out into a 
> helper function and then call it from zone_absent_pages_in_node() and 
> zone_spanned_pages_in_node(), respectively.
> 
Hmm. This whole logic is different from Mel's. 
I'll look into his and reconsider again.

-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [0/16]

[KAMEZAWA Hiroyuki]

On Tue, 6 Mar 2007 07:24:52 -0800 (PST)
David Rientjes <rientjes@google.com> wrote:

> On Tue, 6 Mar 2007, KAMEZAWA Hiroyuki wrote:
> 
> > My purpose is to show how memory-unplug can be implemented on ZONE_MOVABLE.
> > *Any* comments are welcome. This patch just support ia64, which I can test.
> > If you want me to add arch support, please mail me.
> > 
> 
> It's great to see progress being made in the memory hot-unplug direction.  
> The implementation seems to be following the plan you had in our 
> coversations from the end of last year.
> 
> > This patch is a bit old and against 2.6.20-mm2. I'll rebase this and reflect
> > your comments in the next post (may not soon).
> > Well booted on ia64 and passed *quick* memory offline test.
> > 
> 
> When it's rebased, it might be better to apply it to the latest -mm with 
> Mel Gorman's patch series merged.  They're in 2.6.21-rc2-mm2.
> 
I will do.

> It appears as though you're using a subset of the ZONE_MOVABLE patches as 
> posted from March 1.  What about the additional capabilities of 
> ZONE_MOVABLE that aren't included in your patchset, such as allowing 
> hugetlb pages from being allocated under GFP_HIGH_MOVABLE, are going to 
> need to be changed to support memory hot-unplug?

We need extra patches for "migrating HUGEPAGE". It's in my plan but not
scheduled yet.

>  Since your patchset 
> wasn't based on the entire ZONE_MOVABLE set, it leads me to believe that 
> some of what it does diverges with the memory hot-unplug use case.
> 
> Are you aiming to target both ia64 and x86_64 with this patchset or are 
> you focusing on ia64 exclusively at the moment?
> 
Just because a machine, which I can use as much as I want, is ia64.
I don't have x86_64 now. I'll add i386 in the next post.
I think all arch which support MEMORY_HOTPLUG will support unplug at last.

-Kame



--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [0/16]

[David Rientjes]

On Wed, 7 Mar 2007, KAMEZAWA Hiroyuki wrote:

> > Are you aiming to target both ia64 and x86_64 with this patchset or are 
> > you focusing on ia64 exclusively at the moment?
> > 
> Just because a machine, which I can use as much as I want, is ia64.
> I don't have x86_64 now. I'll add i386 in the next post.
> I think all arch which support MEMORY_HOTPLUG will support unplug at last.
> 

Ok, sounds good.  I can offer quite extensive x86_64 testing coverage.  I 
think it's going to be much better to base this patchset on 2.6.21-rc2-mm2 
so we don't have a couple different GFP_MOVABLE implementations floating 
around.

I'll await your next patchset and then I'll play around with it for 
x86_64.  I'd like to eventually combine your memory unplug work with Mark 
Gross's PM-memory enabling node flags (cc'd).  We can wire it up through a 
sysfs interface for userspace manipulation and see it working in action.

Looking forward to the next series.

		David

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [0/16]

[KAMEZAWA Hiroyuki]

On Tue, 6 Mar 2007 18:31:35 -0800 (PST)
David Rientjes <rientjes@google.com> wrote:

> On Wed, 7 Mar 2007, KAMEZAWA Hiroyuki wrote:
> 
> > > Are you aiming to target both ia64 and x86_64 with this patchset or are 
> > > you focusing on ia64 exclusively at the moment?
> > > 
> > Just because a machine, which I can use as much as I want, is ia64.
> > I don't have x86_64 now. I'll add i386 in the next post.
> > I think all arch which support MEMORY_HOTPLUG will support unplug at last.
> > 
> 
> Ok, sounds good.  I can offer quite extensive x86_64 testing coverage.  I 
> think it's going to be much better to base this patchset on 2.6.21-rc2-mm2 
> so we don't have a couple different GFP_MOVABLE implementations floating 
> around.
> 
Thank you and I'll rebase to -mm and write next patch set. 
(But I sometimes stops by other works. please wait.)


> I'll await your next patchset and then I'll play around with it for 
> x86_64.  I'd like to eventually combine your memory unplug work with Mark 
> Gross's PM-memory enabling node flags (cc'd).  We can wire it up through a 
> sysfs interface for userspace manipulation and see it working in action.
> 
looks interesiting :)

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC} memory unplug patchset prep [0/16]

[Mark Gross]

On Tue, Mar 06, 2007 at 06:31:35PM -0800, David Rientjes wrote:
> On Wed, 7 Mar 2007, KAMEZAWA Hiroyuki wrote:
> 
> > > Are you aiming to target both ia64 and x86_64 with this patchset or are 
> > > you focusing on ia64 exclusively at the moment?
> > > 
> > Just because a machine, which I can use as much as I want, is ia64.
> > I don't have x86_64 now. I'll add i386 in the next post.
> > I think all arch which support MEMORY_HOTPLUG will support unplug at last.
> > 
> 
> Ok, sounds good.  I can offer quite extensive x86_64 testing coverage.  I 
> think it's going to be much better to base this patchset on 2.6.21-rc2-mm2 
> so we don't have a couple different GFP_MOVABLE implementations floating 
> around.
> 
> I'll await your next patchset and then I'll play around with it for 
> x86_64.  I'd like to eventually combine your memory unplug work with Mark 
> Gross's PM-memory enabling node flags (cc'd).  We can wire it up through a 
> sysfs interface for userspace manipulation and see it working in action.
> 
> Looking forward to the next series.

Me too!

--mgross

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>