[PATCH] sparsemem memory model

Sparsemem abstracts the use of discontiguous mem_maps[].  This kind of
mem_map[] is needed by discontiguous memory machines (like in the old
CONFIG_DISCONTIGMEM case) as well as memory hotplug systems.  Sparsemem
replaces DISCONTIGMEM when enabled, and it is hoped that it can eventually
become a complete replacement.

A significant advantage over DISCONTIGMEM is that it's completely separated
from CONFIG_NUMA.  When producing this patch, it became apparent in that NUMA
and DISCONTIG are often confused.

Another advantage is that sparse doesn't require each NUMA node's ranges to be
contiguous.  It can handle overlapping ranges between nodes with no problems,
where DISCONTIGMEM currently throws away that memory.

Sparsemem uses an array to provide different pfn_to_page() translations for
each SECTION_SIZE area of physical memory.  This is what allows the mem_map[]
to be chopped up.

In order to do quick pfn_to_page() operations, the section number of the page
is encoded in page->flags.  Part of the sparsemem infrastructure enables
sharing of these bits more dynamically (at compile-time) between the
page_zone() and sparsemem operations.  However, on 32-bit architectures, the
number of bits is quite limited, and may require growing the size of the
page->flags type in certain conditions.  Several things might force this to
occur: a decrease in the SECTION_SIZE (if you want to hotplug smaller areas of
memory), an increase in the physical address space, or an increase in the
number of used page->flags.

One thing to note is that, once sparsemem is present, the NUMA node
information no longer needs to be stored in the page->flags.  It might provide
speed increases on certain platforms and will be stored there if there is
room.  But, if out of room, an alternate (theoretically slower) mechanism is
used.

This patch introduces CONFIG_FLATMEM.  It is used in almost all cases where
there used to be an #ifndef DISCONTIG, because SPARSEMEM and DISCONTIGMEM
often have to compile out the same areas of code.

Signed-off-by: Andy Whitcroft <[email protected]>
Signed-off-by: Dave Hansen <[email protected]>
Signed-off-by: Martin Bligh <[email protected]>
Signed-off-by: Adrian Bunk <[email protected]>
Signed-off-by: Yasunori Goto <[email protected]>
Signed-off-by: Bob Picco <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

arch/i386/Kconfig

@@ -813,6 +813,7 @@ source "mm/Kconfig"
 config HAVE_ARCH_EARLY_PFN_TO_NID
 	bool
 	default y
+	depends on NUMA
 
 config HIGHPTE
 	bool "Allocate 3rd-level pagetables from highmem"

include/linux/mm.h

@@ -397,40 +397,80 @@ static inline void put_page(struct page *page)
  * sets it, so none of the operations on it need to be atomic.
  */
 
-/* Page flags: | NODE | ZONE | ... | FLAGS | */
-#define NODES_PGOFF		((sizeof(page_flags_t)*8) - NODES_SHIFT)
-#define ZONES_PGOFF		(NODES_PGOFF - ZONES_SHIFT)
+
+/*
+ * page->flags layout:
+ *
+ * There are three possibilities for how page->flags get
+ * laid out.  The first is for the normal case, without
+ * sparsemem.  The second is for sparsemem when there is
+ * plenty of space for node and section.  The last is when
+ * we have run out of space and have to fall back to an
+ * alternate (slower) way of determining the node.
+ *
+ *        No sparsemem: |       NODE     | ZONE | ... | FLAGS |
+ * with space for node: | SECTION | NODE | ZONE | ... | FLAGS |
+ *   no space for node: | SECTION |     ZONE    | ... | FLAGS |
+ */
+#ifdef CONFIG_SPARSEMEM
+#define SECTIONS_WIDTH		SECTIONS_SHIFT
+#else
+#define SECTIONS_WIDTH		0
+#endif
+
+#define ZONES_WIDTH		ZONES_SHIFT
+
+#if SECTIONS_WIDTH+ZONES_WIDTH+NODES_SHIFT <= FLAGS_RESERVED
+#define NODES_WIDTH		NODES_SHIFT
+#else
+#define NODES_WIDTH		0
+#endif
+
+/* Page flags: | [SECTION] | [NODE] | ZONE | ... | FLAGS | */
+#define SECTIONS_PGOFF		((sizeof(page_flags_t)*8) - SECTIONS_WIDTH)
+#define NODES_PGOFF		(SECTIONS_PGOFF - NODES_WIDTH)
+#define ZONES_PGOFF		(NODES_PGOFF - ZONES_WIDTH)
+
+/*
+ * We are going to use the flags for the page to node mapping if its in
+ * there.  This includes the case where there is no node, so it is implicit.
+ */
+#define FLAGS_HAS_NODE		(NODES_WIDTH > 0 || NODES_SHIFT == 0)
+
+#ifndef PFN_SECTION_SHIFT
+#define PFN_SECTION_SHIFT 0
+#endif
 
 /*
  * Define the bit shifts to access each section.  For non-existant
  * sections we define the shift as 0; that plus a 0 mask ensures
  * the compiler will optimise away reference to them.
  */
-#define NODES_PGSHIFT		(NODES_PGOFF * (NODES_SHIFT != 0))
-#define ZONES_PGSHIFT		(ZONES_PGOFF * (ZONES_SHIFT != 0))
+#define SECTIONS_PGSHIFT	(SECTIONS_PGOFF * (SECTIONS_WIDTH != 0))
+#define NODES_PGSHIFT		(NODES_PGOFF * (NODES_WIDTH != 0))
+#define ZONES_PGSHIFT		(ZONES_PGOFF * (ZONES_WIDTH != 0))
 
-/* NODE:ZONE is used to lookup the zone from a page. */
+/* NODE:ZONE or SECTION:ZONE is used to lookup the zone from a page. */
+#if FLAGS_HAS_NODE
 #define ZONETABLE_SHIFT		(NODES_SHIFT + ZONES_SHIFT)
+#else
+#define ZONETABLE_SHIFT		(SECTIONS_SHIFT + ZONES_SHIFT)
+#endif
 #define ZONETABLE_PGSHIFT	ZONES_PGSHIFT
 
-#if NODES_SHIFT+ZONES_SHIFT > FLAGS_RESERVED
-#error NODES_SHIFT+ZONES_SHIFT > FLAGS_RESERVED
+#if SECTIONS_WIDTH+NODES_WIDTH+ZONES_WIDTH > FLAGS_RESERVED
+#error SECTIONS_WIDTH+NODES_WIDTH+ZONES_WIDTH > FLAGS_RESERVED
 #endif
 
-#define NODEZONE(node, zone)	((node << ZONES_SHIFT) | zone)
-
-#define ZONES_MASK		((1UL << ZONES_SHIFT) - 1)
-#define NODES_MASK		((1UL << NODES_SHIFT) - 1)
+#define ZONES_MASK		((1UL << ZONES_WIDTH) - 1)
+#define NODES_MASK		((1UL << NODES_WIDTH) - 1)
+#define SECTIONS_MASK		((1UL << SECTIONS_WIDTH) - 1)
 #define ZONETABLE_MASK		((1UL << ZONETABLE_SHIFT) - 1)
 
 static inline unsigned long page_zonenum(struct page *page)
 {
 	return (page->flags >> ZONES_PGSHIFT) & ZONES_MASK;
 }
-static inline unsigned long page_to_nid(struct page *page)
-{
-	return (page->flags >> NODES_PGSHIFT) & NODES_MASK;
-}
 
 struct zone;
 extern struct zone *zone_table[];
@@ -441,6 +481,18 @@ static inline struct zone *page_zone(struct page *page)
 			ZONETABLE_MASK];
 }
 
+static inline unsigned long page_to_nid(struct page *page)
+{
+	if (FLAGS_HAS_NODE)
+		return (page->flags >> NODES_PGSHIFT) & NODES_MASK;
+	else
+		return page_zone(page)->zone_pgdat->node_id;
+}
+static inline unsigned long page_to_section(struct page *page)
+{
+	return (page->flags >> SECTIONS_PGSHIFT) & SECTIONS_MASK;
+}
+
 static inline void set_page_zone(struct page *page, unsigned long zone)
 {
 	page->flags &= ~(ZONES_MASK << ZONES_PGSHIFT);
@@ -451,12 +503,18 @@ static inline void set_page_node(struct page *page, unsigned long node)
 	page->flags &= ~(NODES_MASK << NODES_PGSHIFT);
 	page->flags |= (node & NODES_MASK) << NODES_PGSHIFT;
 }
+static inline void set_page_section(struct page *page, unsigned long section)
+{
+	page->flags &= ~(SECTIONS_MASK << SECTIONS_PGSHIFT);
+	page->flags |= (section & SECTIONS_MASK) << SECTIONS_PGSHIFT;
+}
 
 static inline void set_page_links(struct page *page, unsigned long zone,
-	unsigned long node)
+	unsigned long node, unsigned long pfn)
 {
 	set_page_zone(page, zone);
 	set_page_node(page, node);
+	set_page_section(page, pfn_to_section_nr(pfn));
 }
 
 #ifndef CONFIG_DISCONTIGMEM

include/linux/mmzone.h

@@ -269,7 +269,9 @@ typedef struct pglist_data {
 	struct zone node_zones[MAX_NR_ZONES];
 	struct zonelist node_zonelists[GFP_ZONETYPES];
 	int nr_zones;
+#ifdef CONFIG_FLAT_NODE_MEM_MAP
 	struct page *node_mem_map;
+#endif
 	struct bootmem_data *bdata;
 	unsigned long node_start_pfn;
 	unsigned long node_present_pages; /* total number of physical pages */
@@ -284,7 +286,11 @@ typedef struct pglist_data {
 
 #define node_present_pages(nid)	(NODE_DATA(nid)->node_present_pages)
 #define node_spanned_pages(nid)	(NODE_DATA(nid)->node_spanned_pages)
+#ifdef CONFIG_FLAT_NODE_MEM_MAP
 #define pgdat_page_nr(pgdat, pagenr)	((pgdat)->node_mem_map + (pagenr))
+#else
+#define pgdat_page_nr(pgdat, pagenr)	pfn_to_page((pgdat)->node_start_pfn + (pagenr))
+#endif
 #define nid_page_nr(nid, pagenr) 	pgdat_page_nr(NODE_DATA(nid),(pagenr))
 
 extern struct pglist_data *pgdat_list;
@@ -416,6 +422,10 @@ extern struct pglist_data contig_page_data;
 
 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 
+#ifdef CONFIG_SPARSEMEM
+#include <asm/sparsemem.h>
+#endif
+
 #if BITS_PER_LONG == 32 || defined(ARCH_HAS_ATOMIC_UNSIGNED)
 /*
  * with 32 bit page->flags field, we reserve 8 bits for node/zone info.
@@ -439,6 +449,92 @@ extern struct pglist_data contig_page_data;
 #define early_pfn_to_nid(nid)  (0UL)
 #endif
 
+#define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT)
+#define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT)
+
+#ifdef CONFIG_SPARSEMEM
+
+/*
+ * SECTION_SHIFT    		#bits space required to store a section #
+ *
+ * PA_SECTION_SHIFT		physical address to/from section number
+ * PFN_SECTION_SHIFT		pfn to/from section number
+ */
+#define SECTIONS_SHIFT		(MAX_PHYSMEM_BITS - SECTION_SIZE_BITS)
+
+#define PA_SECTION_SHIFT	(SECTION_SIZE_BITS)
+#define PFN_SECTION_SHIFT	(SECTION_SIZE_BITS - PAGE_SHIFT)
+
+#define NR_MEM_SECTIONS		(1UL << SECTIONS_SHIFT)
+
+#define PAGES_PER_SECTION       (1UL << PFN_SECTION_SHIFT)
+#define PAGE_SECTION_MASK	(~(PAGES_PER_SECTION-1))
+
+#if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
+#error Allocator MAX_ORDER exceeds SECTION_SIZE
+#endif
+
+struct page;
+struct mem_section {
+	struct page *section_mem_map;
+};
+
+extern struct mem_section mem_section[NR_MEM_SECTIONS];
+
+/*
+ * Given a kernel address, find the home node of the underlying memory.
+ */
+#define kvaddr_to_nid(kaddr)	pfn_to_nid(__pa(kaddr) >> PAGE_SHIFT)
+
+static inline struct mem_section *__pfn_to_section(unsigned long pfn)
+{
+	return &mem_section[pfn_to_section_nr(pfn)];
+}
+
+#define pfn_to_page(pfn) 						\
+({ 									\
+	unsigned long __pfn = (pfn);					\
+	__pfn_to_section(__pfn)->section_mem_map + __pfn;		\
+})
+#define page_to_pfn(page)						\
+({									\
+	page - mem_section[page_to_section(page)].section_mem_map;	\
+})
+
+static inline int pfn_valid(unsigned long pfn)
+{
+	if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
+		return 0;
+	return mem_section[pfn_to_section_nr(pfn)].section_mem_map != 0;
+}
+
+/*
+ * These are _only_ used during initialisation, therefore they
+ * can use __initdata ...  They could have names to indicate
+ * this restriction.
+ */
+#ifdef CONFIG_NUMA
+#define pfn_to_nid		early_pfn_to_nid
+#endif
+
+#define pfn_to_pgdat(pfn)						\
+({									\
+	NODE_DATA(pfn_to_nid(pfn));					\
+})
+
+#define early_pfn_valid(pfn)	pfn_valid(pfn)
+void sparse_init(void);
+#else
+#define sparse_init()	do {} while (0)
+#endif /* CONFIG_SPARSEMEM */
+
+#ifndef early_pfn_valid
+#define early_pfn_valid(pfn)	(1)
+#endif
+
+void memory_present(int nid, unsigned long start, unsigned long end);
+unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
+
 #endif /* !__ASSEMBLY__ */
 #endif /* __KERNEL__ */
 #endif /* _LINUX_MMZONE_H */

include/linux/numa.h

@@ -3,7 +3,7 @@
 
 #include <linux/config.h>
 
-#ifdef CONFIG_DISCONTIGMEM
+#ifndef CONFIG_FLATMEM
 #include <asm/numnodes.h>
 #endif
 

mm/Kconfig

@@ -6,6 +6,7 @@ choice
 	prompt "Memory model"
 	depends on SELECT_MEMORY_MODEL
 	default DISCONTIGMEM_MANUAL if ARCH_DISCONTIGMEM_DEFAULT
+	default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT
 	default FLATMEM_MANUAL
 
 config FLATMEM_MANUAL
@@ -17,7 +18,15 @@ config FLATMEM_MANUAL
 	  only have one option here: FLATMEM.  This is normal
 	  and a correct option.
 
-	  If unsure, choose this option over any other.
+	  Some users of more advanced features like NUMA and
+	  memory hotplug may have different options here.
+	  DISCONTIGMEM is an more mature, better tested system,
+	  but is incompatible with memory hotplug and may suffer
+	  decreased performance over SPARSEMEM.  If unsure between
+	  "Sparse Memory" and "Discontiguous Memory", choose
+	  "Discontiguous Memory".
+
+	  If unsure, choose this option (Flat Memory) over any other.
 
 config DISCONTIGMEM_MANUAL
 	bool "Discontigious Memory"
@@ -35,15 +44,38 @@ config DISCONTIGMEM_MANUAL
 
 	  If unsure, choose "Flat Memory" over this option.
 
+config SPARSEMEM_MANUAL
+	bool "Sparse Memory"
+	depends on ARCH_SPARSEMEM_ENABLE
+	help
+	  This will be the only option for some systems, including
+	  memory hotplug systems.  This is normal.
+
+	  For many other systems, this will be an alternative to
+	  "Discontigious Memory".  This option provides some potential
+	  performance benefits, along with decreased code complexity,
+	  but it is newer, and more experimental.
+
+	  If unsure, choose "Discontiguous Memory" or "Flat Memory"
+	  over this option.
+
 endchoice
 
 config DISCONTIGMEM
 	def_bool y
 	depends on (!SELECT_MEMORY_MODEL && ARCH_DISCONTIGMEM_ENABLE) || DISCONTIGMEM_MANUAL
 
+config SPARSEMEM
+	def_bool y
+	depends on SPARSEMEM_MANUAL
+
 config FLATMEM
 	def_bool y
-	depends on !DISCONTIGMEM || FLATMEM_MANUAL
+	depends on (!DISCONTIGMEM && !SPARSEMEM) || FLATMEM_MANUAL
+
+config FLAT_NODE_MEM_MAP
+	def_bool y
+	depends on !SPARSEMEM
 
 #
 # Both the NUMA code and DISCONTIGMEM use arrays of pg_data_t's
@@ -56,4 +88,4 @@ config NEED_MULTIPLE_NODES
 
 config HAVE_MEMORY_PRESENT
 	def_bool y
-	depends on ARCH_HAVE_MEMORY_PRESENT
+	depends on ARCH_HAVE_MEMORY_PRESENT || SPARSEMEM

mm/Makefile

@@ -15,6 +15,7 @@ obj-y			:= bootmem.o filemap.o mempool.o oom_kill.o fadvise.o \
 obj-$(CONFIG_SWAP)	+= page_io.o swap_state.o swapfile.o thrash.o
 obj-$(CONFIG_HUGETLBFS)	+= hugetlb.o
 obj-$(CONFIG_NUMA) 	+= mempolicy.o
+obj-$(CONFIG_SPARSEMEM)	+= sparse.o
 obj-$(CONFIG_SHMEM) += shmem.o
 obj-$(CONFIG_TINY_SHMEM) += tiny-shmem.o