Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | #ifndef _LINUX_MMZONE_H |
| 2 | #define _LINUX_MMZONE_H |
| 3 | |
| 4 | #ifdef __KERNEL__ |
| 5 | #ifndef __ASSEMBLY__ |
| 6 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | #include <linux/spinlock.h> |
| 8 | #include <linux/list.h> |
| 9 | #include <linux/wait.h> |
| 10 | #include <linux/cache.h> |
| 11 | #include <linux/threads.h> |
| 12 | #include <linux/numa.h> |
| 13 | #include <linux/init.h> |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 14 | #include <linux/seqlock.h> |
KAMEZAWA Hiroyuki | 8357f86 | 2006-03-27 01:15:57 -0800 | [diff] [blame] | 15 | #include <linux/nodemask.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | #include <asm/atomic.h> |
Ralf Baechle | 93ff66b | 2006-06-04 02:51:29 -0700 | [diff] [blame] | 17 | #include <asm/page.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | |
| 19 | /* Free memory management - zoned buddy allocator. */ |
| 20 | #ifndef CONFIG_FORCE_MAX_ZONEORDER |
| 21 | #define MAX_ORDER 11 |
| 22 | #else |
| 23 | #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER |
| 24 | #endif |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 25 | #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | |
Andy Whitcroft | 5ad333e | 2007-07-17 04:03:16 -0700 | [diff] [blame] | 27 | /* |
| 28 | * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed |
| 29 | * costly to service. That is between allocation orders which should |
| 30 | * coelesce naturally under reasonable reclaim pressure and those which |
| 31 | * will not. |
| 32 | */ |
| 33 | #define PAGE_ALLOC_COSTLY_ORDER 3 |
| 34 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | struct free_area { |
| 36 | struct list_head free_list; |
| 37 | unsigned long nr_free; |
| 38 | }; |
| 39 | |
| 40 | struct pglist_data; |
| 41 | |
| 42 | /* |
| 43 | * zone->lock and zone->lru_lock are two of the hottest locks in the kernel. |
| 44 | * So add a wild amount of padding here to ensure that they fall into separate |
| 45 | * cachelines. There are very few zone structures in the machine, so space |
| 46 | * consumption is not a concern here. |
| 47 | */ |
| 48 | #if defined(CONFIG_SMP) |
| 49 | struct zone_padding { |
| 50 | char x[0]; |
Ravikiran G Thirumalai | 22fc6ec | 2006-01-08 01:01:27 -0800 | [diff] [blame] | 51 | } ____cacheline_internodealigned_in_smp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | #define ZONE_PADDING(name) struct zone_padding name; |
| 53 | #else |
| 54 | #define ZONE_PADDING(name) |
| 55 | #endif |
| 56 | |
Christoph Lameter | 2244b95 | 2006-06-30 01:55:33 -0700 | [diff] [blame] | 57 | enum zone_stat_item { |
Christoph Lameter | 51ed449 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 58 | /* First 128 byte cacheline (assuming 64 bit words) */ |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 59 | NR_FREE_PAGES, |
Christoph Lameter | c878538 | 2007-02-10 01:43:01 -0800 | [diff] [blame] | 60 | NR_INACTIVE, |
| 61 | NR_ACTIVE, |
Christoph Lameter | f3dbd34 | 2006-06-30 01:55:36 -0700 | [diff] [blame] | 62 | NR_ANON_PAGES, /* Mapped anonymous pages */ |
| 63 | NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. |
Christoph Lameter | 65ba55f | 2006-06-30 01:55:34 -0700 | [diff] [blame] | 64 | only modified from process context */ |
Christoph Lameter | 347ce43 | 2006-06-30 01:55:35 -0700 | [diff] [blame] | 65 | NR_FILE_PAGES, |
Christoph Lameter | b1e7a8f | 2006-06-30 01:55:39 -0700 | [diff] [blame] | 66 | NR_FILE_DIRTY, |
Christoph Lameter | ce866b3 | 2006-06-30 01:55:40 -0700 | [diff] [blame] | 67 | NR_WRITEBACK, |
Christoph Lameter | 51ed449 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 68 | /* Second 128 byte cacheline */ |
| 69 | NR_SLAB_RECLAIMABLE, |
| 70 | NR_SLAB_UNRECLAIMABLE, |
| 71 | NR_PAGETABLE, /* used for pagetables */ |
Christoph Lameter | fd39fc8 | 2006-06-30 01:55:40 -0700 | [diff] [blame] | 72 | NR_UNSTABLE_NFS, /* NFS unstable pages */ |
Christoph Lameter | d2c5e30 | 2006-06-30 01:55:41 -0700 | [diff] [blame] | 73 | NR_BOUNCE, |
Andrew Morton | e129b5c | 2006-09-27 01:50:00 -0700 | [diff] [blame] | 74 | NR_VMSCAN_WRITE, |
Christoph Lameter | ca889e6 | 2006-06-30 01:55:44 -0700 | [diff] [blame] | 75 | #ifdef CONFIG_NUMA |
| 76 | NUMA_HIT, /* allocated in intended node */ |
| 77 | NUMA_MISS, /* allocated in non intended node */ |
| 78 | NUMA_FOREIGN, /* was intended here, hit elsewhere */ |
| 79 | NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ |
| 80 | NUMA_LOCAL, /* allocation from local node */ |
| 81 | NUMA_OTHER, /* allocation from other node */ |
| 82 | #endif |
Christoph Lameter | 2244b95 | 2006-06-30 01:55:33 -0700 | [diff] [blame] | 83 | NR_VM_ZONE_STAT_ITEMS }; |
| 84 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 85 | struct per_cpu_pages { |
| 86 | int count; /* number of pages in the list */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 87 | int high; /* high watermark, emptying needed */ |
| 88 | int batch; /* chunk size for buddy add/remove */ |
| 89 | struct list_head list; /* the list of pages */ |
| 90 | }; |
| 91 | |
| 92 | struct per_cpu_pageset { |
| 93 | struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */ |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 94 | #ifdef CONFIG_NUMA |
| 95 | s8 expire; |
| 96 | #endif |
Christoph Lameter | 2244b95 | 2006-06-30 01:55:33 -0700 | [diff] [blame] | 97 | #ifdef CONFIG_SMP |
Christoph Lameter | df9ecab | 2006-08-31 21:27:35 -0700 | [diff] [blame] | 98 | s8 stat_threshold; |
Christoph Lameter | 2244b95 | 2006-06-30 01:55:33 -0700 | [diff] [blame] | 99 | s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; |
| 100 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | } ____cacheline_aligned_in_smp; |
| 102 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 103 | #ifdef CONFIG_NUMA |
| 104 | #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)]) |
| 105 | #else |
| 106 | #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)]) |
| 107 | #endif |
| 108 | |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 109 | enum zone_type { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 110 | #ifdef CONFIG_ZONE_DMA |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 111 | /* |
| 112 | * ZONE_DMA is used when there are devices that are not able |
| 113 | * to do DMA to all of addressable memory (ZONE_NORMAL). Then we |
| 114 | * carve out the portion of memory that is needed for these devices. |
| 115 | * The range is arch specific. |
| 116 | * |
| 117 | * Some examples |
| 118 | * |
| 119 | * Architecture Limit |
| 120 | * --------------------------- |
| 121 | * parisc, ia64, sparc <4G |
| 122 | * s390 <2G |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 123 | * arm Various |
| 124 | * alpha Unlimited or 0-16MB. |
| 125 | * |
| 126 | * i386, x86_64 and multiple other arches |
| 127 | * <16M. |
| 128 | */ |
| 129 | ZONE_DMA, |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 130 | #endif |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 131 | #ifdef CONFIG_ZONE_DMA32 |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 132 | /* |
| 133 | * x86_64 needs two ZONE_DMAs because it supports devices that are |
| 134 | * only able to do DMA to the lower 16M but also 32 bit devices that |
| 135 | * can only do DMA areas below 4G. |
| 136 | */ |
| 137 | ZONE_DMA32, |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 138 | #endif |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 139 | /* |
| 140 | * Normal addressable memory is in ZONE_NORMAL. DMA operations can be |
| 141 | * performed on pages in ZONE_NORMAL if the DMA devices support |
| 142 | * transfers to all addressable memory. |
| 143 | */ |
| 144 | ZONE_NORMAL, |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 145 | #ifdef CONFIG_HIGHMEM |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 146 | /* |
| 147 | * A memory area that is only addressable by the kernel through |
| 148 | * mapping portions into its own address space. This is for example |
| 149 | * used by i386 to allow the kernel to address the memory beyond |
| 150 | * 900MB. The kernel will set up special mappings (page |
| 151 | * table entries on i386) for each page that the kernel needs to |
| 152 | * access. |
| 153 | */ |
| 154 | ZONE_HIGHMEM, |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 155 | #endif |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 156 | ZONE_MOVABLE, |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 157 | MAX_NR_ZONES |
| 158 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 160 | /* |
| 161 | * When a memory allocation must conform to specific limitations (such |
| 162 | * as being suitable for DMA) the caller will pass in hints to the |
| 163 | * allocator in the gfp_mask, in the zone modifier bits. These bits |
| 164 | * are used to select a priority ordered list of memory zones which |
Christoph Lameter | 19655d3 | 2006-09-25 23:31:19 -0700 | [diff] [blame] | 165 | * match the requested limits. See gfp_zone() in include/linux/gfp.h |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | */ |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 167 | |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 168 | /* |
| 169 | * Count the active zones. Note that the use of defined(X) outside |
| 170 | * #if and family is not necessarily defined so ensure we cannot use |
| 171 | * it later. Use __ZONE_COUNT to work out how many shift bits we need. |
| 172 | */ |
| 173 | #define __ZONE_COUNT ( \ |
| 174 | defined(CONFIG_ZONE_DMA) \ |
| 175 | + defined(CONFIG_ZONE_DMA32) \ |
| 176 | + 1 \ |
| 177 | + defined(CONFIG_HIGHMEM) \ |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 178 | + 1 \ |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 179 | ) |
| 180 | #if __ZONE_COUNT < 2 |
| 181 | #define ZONES_SHIFT 0 |
| 182 | #elif __ZONE_COUNT <= 2 |
Christoph Lameter | 19655d3 | 2006-09-25 23:31:19 -0700 | [diff] [blame] | 183 | #define ZONES_SHIFT 1 |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 184 | #elif __ZONE_COUNT <= 4 |
Christoph Lameter | 19655d3 | 2006-09-25 23:31:19 -0700 | [diff] [blame] | 185 | #define ZONES_SHIFT 2 |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 186 | #else |
| 187 | #error ZONES_SHIFT -- too many zones configured adjust calculation |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 188 | #endif |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 189 | #undef __ZONE_COUNT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | struct zone { |
| 192 | /* Fields commonly accessed by the page allocator */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | unsigned long pages_min, pages_low, pages_high; |
| 194 | /* |
| 195 | * We don't know if the memory that we're going to allocate will be freeable |
| 196 | * or/and it will be released eventually, so to avoid totally wasting several |
| 197 | * GB of ram we must reserve some of the lower zone memory (otherwise we risk |
| 198 | * to run OOM on the lower zones despite there's tons of freeable ram |
| 199 | * on the higher zones). This array is recalculated at runtime if the |
| 200 | * sysctl_lowmem_reserve_ratio sysctl changes. |
| 201 | */ |
| 202 | unsigned long lowmem_reserve[MAX_NR_ZONES]; |
| 203 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 204 | #ifdef CONFIG_NUMA |
Christoph Lameter | d5f541e | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 205 | int node; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 206 | /* |
| 207 | * zone reclaim becomes active if more unmapped pages exist. |
| 208 | */ |
Christoph Lameter | 8417bba | 2006-09-25 23:31:51 -0700 | [diff] [blame] | 209 | unsigned long min_unmapped_pages; |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 210 | unsigned long min_slab_pages; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 211 | struct per_cpu_pageset *pageset[NR_CPUS]; |
| 212 | #else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | struct per_cpu_pageset pageset[NR_CPUS]; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 214 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | /* |
| 216 | * free areas of different sizes |
| 217 | */ |
| 218 | spinlock_t lock; |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 219 | #ifdef CONFIG_MEMORY_HOTPLUG |
| 220 | /* see spanned/present_pages for more description */ |
| 221 | seqlock_t span_seqlock; |
| 222 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 223 | struct free_area free_area[MAX_ORDER]; |
| 224 | |
| 225 | |
| 226 | ZONE_PADDING(_pad1_) |
| 227 | |
| 228 | /* Fields commonly accessed by the page reclaim scanner */ |
| 229 | spinlock_t lru_lock; |
| 230 | struct list_head active_list; |
| 231 | struct list_head inactive_list; |
| 232 | unsigned long nr_scan_active; |
| 233 | unsigned long nr_scan_inactive; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 234 | unsigned long pages_scanned; /* since last reclaim */ |
| 235 | int all_unreclaimable; /* All pages pinned */ |
| 236 | |
Martin Hicks | 1e7e5a9 | 2005-06-21 17:14:43 -0700 | [diff] [blame] | 237 | /* A count of how many reclaimers are scanning this zone */ |
| 238 | atomic_t reclaim_in_progress; |
Martin Hicks | 753ee72 | 2005-06-21 17:14:41 -0700 | [diff] [blame] | 239 | |
Christoph Lameter | 2244b95 | 2006-06-30 01:55:33 -0700 | [diff] [blame] | 240 | /* Zone statistics */ |
| 241 | atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 242 | |
| 243 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | * prev_priority holds the scanning priority for this zone. It is |
| 245 | * defined as the scanning priority at which we achieved our reclaim |
| 246 | * target at the previous try_to_free_pages() or balance_pgdat() |
| 247 | * invokation. |
| 248 | * |
| 249 | * We use prev_priority as a measure of how much stress page reclaim is |
| 250 | * under - it drives the swappiness decision: whether to unmap mapped |
| 251 | * pages. |
| 252 | * |
Martin Bligh | 3bb1a85 | 2006-10-28 10:38:24 -0700 | [diff] [blame] | 253 | * Access to both this field is quite racy even on uniprocessor. But |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | * it is expected to average out OK. |
| 255 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | int prev_priority; |
| 257 | |
| 258 | |
| 259 | ZONE_PADDING(_pad2_) |
| 260 | /* Rarely used or read-mostly fields */ |
| 261 | |
| 262 | /* |
| 263 | * wait_table -- the array holding the hash table |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 264 | * wait_table_hash_nr_entries -- the size of the hash table array |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | * wait_table_bits -- wait_table_size == (1 << wait_table_bits) |
| 266 | * |
| 267 | * The purpose of all these is to keep track of the people |
| 268 | * waiting for a page to become available and make them |
| 269 | * runnable again when possible. The trouble is that this |
| 270 | * consumes a lot of space, especially when so few things |
| 271 | * wait on pages at a given time. So instead of using |
| 272 | * per-page waitqueues, we use a waitqueue hash table. |
| 273 | * |
| 274 | * The bucket discipline is to sleep on the same queue when |
| 275 | * colliding and wake all in that wait queue when removing. |
| 276 | * When something wakes, it must check to be sure its page is |
| 277 | * truly available, a la thundering herd. The cost of a |
| 278 | * collision is great, but given the expected load of the |
| 279 | * table, they should be so rare as to be outweighed by the |
| 280 | * benefits from the saved space. |
| 281 | * |
| 282 | * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the |
| 283 | * primary users of these fields, and in mm/page_alloc.c |
| 284 | * free_area_init_core() performs the initialization of them. |
| 285 | */ |
| 286 | wait_queue_head_t * wait_table; |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 287 | unsigned long wait_table_hash_nr_entries; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 288 | unsigned long wait_table_bits; |
| 289 | |
| 290 | /* |
| 291 | * Discontig memory support fields. |
| 292 | */ |
| 293 | struct pglist_data *zone_pgdat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ |
| 295 | unsigned long zone_start_pfn; |
| 296 | |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 297 | /* |
| 298 | * zone_start_pfn, spanned_pages and present_pages are all |
| 299 | * protected by span_seqlock. It is a seqlock because it has |
| 300 | * to be read outside of zone->lock, and it is done in the main |
| 301 | * allocator path. But, it is written quite infrequently. |
| 302 | * |
| 303 | * The lock is declared along with zone->lock because it is |
| 304 | * frequently read in proximity to zone->lock. It's good to |
| 305 | * give them a chance of being in the same cacheline. |
| 306 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | unsigned long spanned_pages; /* total size, including holes */ |
| 308 | unsigned long present_pages; /* amount of memory (excluding holes) */ |
| 309 | |
| 310 | /* |
| 311 | * rarely used fields: |
| 312 | */ |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 313 | const char *name; |
Ravikiran G Thirumalai | 22fc6ec | 2006-01-08 01:01:27 -0800 | [diff] [blame] | 314 | } ____cacheline_internodealigned_in_smp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | /* |
| 317 | * The "priority" of VM scanning is how much of the queues we will scan in one |
| 318 | * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the |
| 319 | * queues ("queue_length >> 12") during an aging round. |
| 320 | */ |
| 321 | #define DEF_PRIORITY 12 |
| 322 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 323 | /* Maximum number of zones on a zonelist */ |
| 324 | #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) |
| 325 | |
| 326 | #ifdef CONFIG_NUMA |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame^] | 327 | |
| 328 | /* |
| 329 | * The NUMA zonelists are doubled becausse we need zonelists that restrict the |
| 330 | * allocations to a single node for GFP_THISNODE. |
| 331 | * |
| 332 | * [0 .. MAX_NR_ZONES -1] : Zonelists with fallback |
| 333 | * [MAZ_NR_ZONES ... MAZ_ZONELISTS -1] : No fallback (GFP_THISNODE) |
| 334 | */ |
| 335 | #define MAX_ZONELISTS (2 * MAX_NR_ZONES) |
| 336 | |
| 337 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 338 | /* |
| 339 | * We cache key information from each zonelist for smaller cache |
| 340 | * footprint when scanning for free pages in get_page_from_freelist(). |
| 341 | * |
| 342 | * 1) The BITMAP fullzones tracks which zones in a zonelist have come |
| 343 | * up short of free memory since the last time (last_fullzone_zap) |
| 344 | * we zero'd fullzones. |
| 345 | * 2) The array z_to_n[] maps each zone in the zonelist to its node |
| 346 | * id, so that we can efficiently evaluate whether that node is |
| 347 | * set in the current tasks mems_allowed. |
| 348 | * |
| 349 | * Both fullzones and z_to_n[] are one-to-one with the zonelist, |
| 350 | * indexed by a zones offset in the zonelist zones[] array. |
| 351 | * |
| 352 | * The get_page_from_freelist() routine does two scans. During the |
| 353 | * first scan, we skip zones whose corresponding bit in 'fullzones' |
| 354 | * is set or whose corresponding node in current->mems_allowed (which |
| 355 | * comes from cpusets) is not set. During the second scan, we bypass |
| 356 | * this zonelist_cache, to ensure we look methodically at each zone. |
| 357 | * |
| 358 | * Once per second, we zero out (zap) fullzones, forcing us to |
| 359 | * reconsider nodes that might have regained more free memory. |
| 360 | * The field last_full_zap is the time we last zapped fullzones. |
| 361 | * |
| 362 | * This mechanism reduces the amount of time we waste repeatedly |
| 363 | * reexaming zones for free memory when they just came up low on |
| 364 | * memory momentarilly ago. |
| 365 | * |
| 366 | * The zonelist_cache struct members logically belong in struct |
| 367 | * zonelist. However, the mempolicy zonelists constructed for |
| 368 | * MPOL_BIND are intentionally variable length (and usually much |
| 369 | * shorter). A general purpose mechanism for handling structs with |
| 370 | * multiple variable length members is more mechanism than we want |
| 371 | * here. We resort to some special case hackery instead. |
| 372 | * |
| 373 | * The MPOL_BIND zonelists don't need this zonelist_cache (in good |
| 374 | * part because they are shorter), so we put the fixed length stuff |
| 375 | * at the front of the zonelist struct, ending in a variable length |
| 376 | * zones[], as is needed by MPOL_BIND. |
| 377 | * |
| 378 | * Then we put the optional zonelist cache on the end of the zonelist |
| 379 | * struct. This optional stuff is found by a 'zlcache_ptr' pointer in |
| 380 | * the fixed length portion at the front of the struct. This pointer |
| 381 | * both enables us to find the zonelist cache, and in the case of |
| 382 | * MPOL_BIND zonelists, (which will just set the zlcache_ptr to NULL) |
| 383 | * to know that the zonelist cache is not there. |
| 384 | * |
| 385 | * The end result is that struct zonelists come in two flavors: |
| 386 | * 1) The full, fixed length version, shown below, and |
| 387 | * 2) The custom zonelists for MPOL_BIND. |
| 388 | * The custom MPOL_BIND zonelists have a NULL zlcache_ptr and no zlcache. |
| 389 | * |
| 390 | * Even though there may be multiple CPU cores on a node modifying |
| 391 | * fullzones or last_full_zap in the same zonelist_cache at the same |
| 392 | * time, we don't lock it. This is just hint data - if it is wrong now |
| 393 | * and then, the allocator will still function, perhaps a bit slower. |
| 394 | */ |
| 395 | |
| 396 | |
| 397 | struct zonelist_cache { |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 398 | unsigned short z_to_n[MAX_ZONES_PER_ZONELIST]; /* zone->nid */ |
Paul Jackson | 7253f4e | 2006-12-06 20:31:49 -0800 | [diff] [blame] | 399 | DECLARE_BITMAP(fullzones, MAX_ZONES_PER_ZONELIST); /* zone full? */ |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 400 | unsigned long last_full_zap; /* when last zap'd (jiffies) */ |
| 401 | }; |
| 402 | #else |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame^] | 403 | #define MAX_ZONELISTS MAX_NR_ZONES |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 404 | struct zonelist_cache; |
| 405 | #endif |
| 406 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | /* |
| 408 | * One allocation request operates on a zonelist. A zonelist |
| 409 | * is a list of zones, the first one is the 'goal' of the |
| 410 | * allocation, the other zones are fallback zones, in decreasing |
| 411 | * priority. |
| 412 | * |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 413 | * If zlcache_ptr is not NULL, then it is just the address of zlcache, |
| 414 | * as explained above. If zlcache_ptr is NULL, there is no zlcache. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | */ |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 416 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 417 | struct zonelist { |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 418 | struct zonelist_cache *zlcache_ptr; // NULL or &zlcache |
| 419 | struct zone *zones[MAX_ZONES_PER_ZONELIST + 1]; // NULL delimited |
| 420 | #ifdef CONFIG_NUMA |
| 421 | struct zonelist_cache zlcache; // optional ... |
| 422 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | }; |
| 424 | |
Mel Gorman | b377fd3 | 2007-08-22 14:02:05 -0700 | [diff] [blame] | 425 | #ifdef CONFIG_NUMA |
| 426 | /* |
| 427 | * Only custom zonelists like MPOL_BIND need to be filtered as part of |
| 428 | * policies. As described in the comment for struct zonelist_cache, these |
| 429 | * zonelists will not have a zlcache so zlcache_ptr will not be set. Use |
| 430 | * that to determine if the zonelists needs to be filtered or not. |
| 431 | */ |
| 432 | static inline int alloc_should_filter_zonelist(struct zonelist *zonelist) |
| 433 | { |
| 434 | return !zonelist->zlcache_ptr; |
| 435 | } |
| 436 | #else |
| 437 | static inline int alloc_should_filter_zonelist(struct zonelist *zonelist) |
| 438 | { |
| 439 | return 0; |
| 440 | } |
| 441 | #endif /* CONFIG_NUMA */ |
| 442 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 443 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
| 444 | struct node_active_region { |
| 445 | unsigned long start_pfn; |
| 446 | unsigned long end_pfn; |
| 447 | int nid; |
| 448 | }; |
| 449 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 450 | |
Heiko Carstens | 5b99cd0 | 2006-09-27 01:50:01 -0700 | [diff] [blame] | 451 | #ifndef CONFIG_DISCONTIGMEM |
| 452 | /* The array of struct pages - for discontigmem use pgdat->lmem_map */ |
| 453 | extern struct page *mem_map; |
| 454 | #endif |
| 455 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | /* |
| 457 | * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM |
| 458 | * (mostly NUMA machines?) to denote a higher-level memory zone than the |
| 459 | * zone denotes. |
| 460 | * |
| 461 | * On NUMA machines, each NUMA node would have a pg_data_t to describe |
| 462 | * it's memory layout. |
| 463 | * |
| 464 | * Memory statistics and page replacement data structures are maintained on a |
| 465 | * per-zone basis. |
| 466 | */ |
| 467 | struct bootmem_data; |
| 468 | typedef struct pglist_data { |
| 469 | struct zone node_zones[MAX_NR_ZONES]; |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame^] | 470 | struct zonelist node_zonelists[MAX_ZONELISTS]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 471 | int nr_zones; |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 472 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | struct page *node_mem_map; |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 474 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 475 | struct bootmem_data *bdata; |
Dave Hansen | 208d54e | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 476 | #ifdef CONFIG_MEMORY_HOTPLUG |
| 477 | /* |
| 478 | * Must be held any time you expect node_start_pfn, node_present_pages |
| 479 | * or node_spanned_pages stay constant. Holding this will also |
| 480 | * guarantee that any pfn_valid() stays that way. |
| 481 | * |
| 482 | * Nests above zone->lock and zone->size_seqlock. |
| 483 | */ |
| 484 | spinlock_t node_size_lock; |
| 485 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | unsigned long node_start_pfn; |
| 487 | unsigned long node_present_pages; /* total number of physical pages */ |
| 488 | unsigned long node_spanned_pages; /* total size of physical page |
| 489 | range, including holes */ |
| 490 | int node_id; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 491 | wait_queue_head_t kswapd_wait; |
| 492 | struct task_struct *kswapd; |
| 493 | int kswapd_max_order; |
| 494 | } pg_data_t; |
| 495 | |
| 496 | #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) |
| 497 | #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 498 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
Dave Hansen | 408fde8 | 2005-06-23 00:07:37 -0700 | [diff] [blame] | 499 | #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 500 | #else |
| 501 | #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) |
| 502 | #endif |
Dave Hansen | 408fde8 | 2005-06-23 00:07:37 -0700 | [diff] [blame] | 503 | #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 504 | |
Dave Hansen | 208d54e | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 505 | #include <linux/memory_hotplug.h> |
| 506 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | void get_zone_counts(unsigned long *active, unsigned long *inactive, |
| 508 | unsigned long *free); |
| 509 | void build_all_zonelists(void); |
| 510 | void wakeup_kswapd(struct zone *zone, int order); |
| 511 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 512 | int classzone_idx, int alloc_flags); |
Dave Hansen | a2f3aa0 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 513 | enum memmap_context { |
| 514 | MEMMAP_EARLY, |
| 515 | MEMMAP_HOTPLUG, |
| 516 | }; |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 517 | extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, |
Dave Hansen | a2f3aa0 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 518 | unsigned long size, |
| 519 | enum memmap_context context); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 520 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | #ifdef CONFIG_HAVE_MEMORY_PRESENT |
| 522 | void memory_present(int nid, unsigned long start, unsigned long end); |
| 523 | #else |
| 524 | static inline void memory_present(int nid, unsigned long start, unsigned long end) {} |
| 525 | #endif |
| 526 | |
| 527 | #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE |
| 528 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); |
| 529 | #endif |
| 530 | |
| 531 | /* |
| 532 | * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. |
| 533 | */ |
| 534 | #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) |
| 535 | |
Con Kolivas | f3fe651 | 2006-01-06 00:11:15 -0800 | [diff] [blame] | 536 | static inline int populated_zone(struct zone *zone) |
| 537 | { |
| 538 | return (!!zone->present_pages); |
| 539 | } |
| 540 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 541 | extern int movable_zone; |
| 542 | |
| 543 | static inline int zone_movable_is_highmem(void) |
| 544 | { |
| 545 | #if defined(CONFIG_HIGHMEM) && defined(CONFIG_ARCH_POPULATES_NODE_MAP) |
| 546 | return movable_zone == ZONE_HIGHMEM; |
| 547 | #else |
| 548 | return 0; |
| 549 | #endif |
| 550 | } |
| 551 | |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 552 | static inline int is_highmem_idx(enum zone_type idx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | { |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 554 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 555 | return (idx == ZONE_HIGHMEM || |
| 556 | (idx == ZONE_MOVABLE && zone_movable_is_highmem())); |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 557 | #else |
| 558 | return 0; |
| 559 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | } |
| 561 | |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 562 | static inline int is_normal_idx(enum zone_type idx) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 563 | { |
| 564 | return (idx == ZONE_NORMAL); |
| 565 | } |
Nick Piggin | 9328b8f | 2006-01-06 00:11:10 -0800 | [diff] [blame] | 566 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 567 | /** |
| 568 | * is_highmem - helper function to quickly check if a struct zone is a |
| 569 | * highmem zone or not. This is an attempt to keep references |
| 570 | * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. |
| 571 | * @zone - pointer to struct zone variable |
| 572 | */ |
| 573 | static inline int is_highmem(struct zone *zone) |
| 574 | { |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 575 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 576 | int zone_idx = zone - zone->zone_pgdat->node_zones; |
| 577 | return zone_idx == ZONE_HIGHMEM || |
| 578 | (zone_idx == ZONE_MOVABLE && zone_movable_is_highmem()); |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 579 | #else |
| 580 | return 0; |
| 581 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 582 | } |
| 583 | |
| 584 | static inline int is_normal(struct zone *zone) |
| 585 | { |
| 586 | return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL; |
| 587 | } |
| 588 | |
Nick Piggin | 9328b8f | 2006-01-06 00:11:10 -0800 | [diff] [blame] | 589 | static inline int is_dma32(struct zone *zone) |
| 590 | { |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 591 | #ifdef CONFIG_ZONE_DMA32 |
Nick Piggin | 9328b8f | 2006-01-06 00:11:10 -0800 | [diff] [blame] | 592 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA32; |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 593 | #else |
| 594 | return 0; |
| 595 | #endif |
Nick Piggin | 9328b8f | 2006-01-06 00:11:10 -0800 | [diff] [blame] | 596 | } |
| 597 | |
| 598 | static inline int is_dma(struct zone *zone) |
| 599 | { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 600 | #ifdef CONFIG_ZONE_DMA |
Nick Piggin | 9328b8f | 2006-01-06 00:11:10 -0800 | [diff] [blame] | 601 | return zone == zone->zone_pgdat->node_zones + ZONE_DMA; |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 602 | #else |
| 603 | return 0; |
| 604 | #endif |
Nick Piggin | 9328b8f | 2006-01-06 00:11:10 -0800 | [diff] [blame] | 605 | } |
| 606 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | /* These two functions are used to setup the per zone pages min values */ |
| 608 | struct ctl_table; |
| 609 | struct file; |
| 610 | int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *, |
| 611 | void __user *, size_t *, loff_t *); |
| 612 | extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; |
| 613 | int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *, |
| 614 | void __user *, size_t *, loff_t *); |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 615 | int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *, |
| 616 | void __user *, size_t *, loff_t *); |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 617 | int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, |
| 618 | struct file *, void __user *, size_t *, loff_t *); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 619 | int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, |
| 620 | struct file *, void __user *, size_t *, loff_t *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 621 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 622 | extern int numa_zonelist_order_handler(struct ctl_table *, int, |
| 623 | struct file *, void __user *, size_t *, loff_t *); |
| 624 | extern char numa_zonelist_order[]; |
| 625 | #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */ |
| 626 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | #include <linux/topology.h> |
| 628 | /* Returns the number of the current Node. */ |
Andi Kleen | 69d81fc | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 629 | #ifndef numa_node_id |
Ingo Molnar | 39c715b | 2005-06-21 17:14:34 -0700 | [diff] [blame] | 630 | #define numa_node_id() (cpu_to_node(raw_smp_processor_id())) |
Andi Kleen | 69d81fc | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 631 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | |
Dave Hansen | 93b7504 | 2005-06-23 00:07:47 -0700 | [diff] [blame] | 633 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 634 | |
| 635 | extern struct pglist_data contig_page_data; |
| 636 | #define NODE_DATA(nid) (&contig_page_data) |
| 637 | #define NODE_MEM_MAP(nid) mem_map |
| 638 | #define MAX_NODES_SHIFT 1 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | |
Dave Hansen | 93b7504 | 2005-06-23 00:07:47 -0700 | [diff] [blame] | 640 | #else /* CONFIG_NEED_MULTIPLE_NODES */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | |
| 642 | #include <asm/mmzone.h> |
| 643 | |
Dave Hansen | 93b7504 | 2005-06-23 00:07:47 -0700 | [diff] [blame] | 644 | #endif /* !CONFIG_NEED_MULTIPLE_NODES */ |
Dave Hansen | 348f8b6 | 2005-06-23 00:07:40 -0700 | [diff] [blame] | 645 | |
KAMEZAWA Hiroyuki | 95144c7 | 2006-03-27 01:16:02 -0800 | [diff] [blame] | 646 | extern struct pglist_data *first_online_pgdat(void); |
| 647 | extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); |
| 648 | extern struct zone *next_zone(struct zone *zone); |
KAMEZAWA Hiroyuki | 8357f86 | 2006-03-27 01:15:57 -0800 | [diff] [blame] | 649 | |
| 650 | /** |
| 651 | * for_each_pgdat - helper macro to iterate over all nodes |
| 652 | * @pgdat - pointer to a pg_data_t variable |
| 653 | */ |
| 654 | #define for_each_online_pgdat(pgdat) \ |
| 655 | for (pgdat = first_online_pgdat(); \ |
| 656 | pgdat; \ |
| 657 | pgdat = next_online_pgdat(pgdat)) |
KAMEZAWA Hiroyuki | 8357f86 | 2006-03-27 01:15:57 -0800 | [diff] [blame] | 658 | /** |
| 659 | * for_each_zone - helper macro to iterate over all memory zones |
| 660 | * @zone - pointer to struct zone variable |
| 661 | * |
| 662 | * The user only needs to declare the zone variable, for_each_zone |
| 663 | * fills it in. |
| 664 | */ |
| 665 | #define for_each_zone(zone) \ |
| 666 | for (zone = (first_online_pgdat())->node_zones; \ |
| 667 | zone; \ |
| 668 | zone = next_zone(zone)) |
| 669 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 670 | #ifdef CONFIG_SPARSEMEM |
| 671 | #include <asm/sparsemem.h> |
| 672 | #endif |
| 673 | |
Andi Kleen | 07808b7 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 674 | #if BITS_PER_LONG == 32 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | /* |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 676 | * with 32 bit page->flags field, we reserve 9 bits for node/zone info. |
| 677 | * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 678 | */ |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 679 | #define FLAGS_RESERVED 9 |
Dave Hansen | 348f8b6 | 2005-06-23 00:07:40 -0700 | [diff] [blame] | 680 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 681 | #elif BITS_PER_LONG == 64 |
| 682 | /* |
| 683 | * with 64 bit flags field, there's plenty of room. |
| 684 | */ |
Dave Hansen | 348f8b6 | 2005-06-23 00:07:40 -0700 | [diff] [blame] | 685 | #define FLAGS_RESERVED 32 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | |
Dave Hansen | 348f8b6 | 2005-06-23 00:07:40 -0700 | [diff] [blame] | 687 | #else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | |
Dave Hansen | 348f8b6 | 2005-06-23 00:07:40 -0700 | [diff] [blame] | 689 | #error BITS_PER_LONG not defined |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 691 | #endif |
| 692 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 693 | #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \ |
| 694 | !defined(CONFIG_ARCH_POPULATES_NODE_MAP) |
Andy Whitcroft | b159d43 | 2005-06-23 00:07:52 -0700 | [diff] [blame] | 695 | #define early_pfn_to_nid(nid) (0UL) |
| 696 | #endif |
| 697 | |
Andy Whitcroft | 2bdaf11 | 2006-01-06 00:10:53 -0800 | [diff] [blame] | 698 | #ifdef CONFIG_FLATMEM |
| 699 | #define pfn_to_nid(pfn) (0) |
| 700 | #endif |
| 701 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 702 | #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) |
| 703 | #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) |
| 704 | |
| 705 | #ifdef CONFIG_SPARSEMEM |
| 706 | |
| 707 | /* |
| 708 | * SECTION_SHIFT #bits space required to store a section # |
| 709 | * |
| 710 | * PA_SECTION_SHIFT physical address to/from section number |
| 711 | * PFN_SECTION_SHIFT pfn to/from section number |
| 712 | */ |
| 713 | #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS) |
| 714 | |
| 715 | #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) |
| 716 | #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) |
| 717 | |
| 718 | #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) |
| 719 | |
| 720 | #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) |
| 721 | #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) |
| 722 | |
| 723 | #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS |
| 724 | #error Allocator MAX_ORDER exceeds SECTION_SIZE |
| 725 | #endif |
| 726 | |
| 727 | struct page; |
| 728 | struct mem_section { |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 729 | /* |
| 730 | * This is, logically, a pointer to an array of struct |
| 731 | * pages. However, it is stored with some other magic. |
| 732 | * (see sparse.c::sparse_init_one_section()) |
| 733 | * |
Andy Whitcroft | 30c253e | 2006-06-23 02:03:41 -0700 | [diff] [blame] | 734 | * Additionally during early boot we encode node id of |
| 735 | * the location of the section here to guide allocation. |
| 736 | * (see sparse.c::memory_present()) |
| 737 | * |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 738 | * Making it a UL at least makes someone do a cast |
| 739 | * before using it wrong. |
| 740 | */ |
| 741 | unsigned long section_mem_map; |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 742 | }; |
| 743 | |
Bob Picco | 3e34726 | 2005-09-03 15:54:28 -0700 | [diff] [blame] | 744 | #ifdef CONFIG_SPARSEMEM_EXTREME |
| 745 | #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) |
Bob Picco | 802f192 | 2005-09-03 15:54:26 -0700 | [diff] [blame] | 746 | #else |
Bob Picco | 3e34726 | 2005-09-03 15:54:28 -0700 | [diff] [blame] | 747 | #define SECTIONS_PER_ROOT 1 |
| 748 | #endif |
Bob Picco | 802f192 | 2005-09-03 15:54:26 -0700 | [diff] [blame] | 749 | |
Bob Picco | 3e34726 | 2005-09-03 15:54:28 -0700 | [diff] [blame] | 750 | #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) |
| 751 | #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT) |
| 752 | #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) |
| 753 | |
| 754 | #ifdef CONFIG_SPARSEMEM_EXTREME |
| 755 | extern struct mem_section *mem_section[NR_SECTION_ROOTS]; |
| 756 | #else |
| 757 | extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; |
| 758 | #endif |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 759 | |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 760 | static inline struct mem_section *__nr_to_section(unsigned long nr) |
| 761 | { |
Bob Picco | 3e34726 | 2005-09-03 15:54:28 -0700 | [diff] [blame] | 762 | if (!mem_section[SECTION_NR_TO_ROOT(nr)]) |
| 763 | return NULL; |
| 764 | return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 765 | } |
Dave Hansen | 4ca644d | 2005-10-29 18:16:51 -0700 | [diff] [blame] | 766 | extern int __section_nr(struct mem_section* ms); |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 767 | |
| 768 | /* |
| 769 | * We use the lower bits of the mem_map pointer to store |
| 770 | * a little bit of information. There should be at least |
| 771 | * 3 bits here due to 32-bit alignment. |
| 772 | */ |
| 773 | #define SECTION_MARKED_PRESENT (1UL<<0) |
| 774 | #define SECTION_HAS_MEM_MAP (1UL<<1) |
| 775 | #define SECTION_MAP_LAST_BIT (1UL<<2) |
| 776 | #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) |
Andy Whitcroft | 30c253e | 2006-06-23 02:03:41 -0700 | [diff] [blame] | 777 | #define SECTION_NID_SHIFT 2 |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 778 | |
| 779 | static inline struct page *__section_mem_map_addr(struct mem_section *section) |
| 780 | { |
| 781 | unsigned long map = section->section_mem_map; |
| 782 | map &= SECTION_MAP_MASK; |
| 783 | return (struct page *)map; |
| 784 | } |
| 785 | |
Andy Whitcroft | 540557b | 2007-10-16 01:24:11 -0700 | [diff] [blame] | 786 | static inline int present_section(struct mem_section *section) |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 787 | { |
Bob Picco | 802f192 | 2005-09-03 15:54:26 -0700 | [diff] [blame] | 788 | return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 789 | } |
| 790 | |
Andy Whitcroft | 540557b | 2007-10-16 01:24:11 -0700 | [diff] [blame] | 791 | static inline int present_section_nr(unsigned long nr) |
| 792 | { |
| 793 | return present_section(__nr_to_section(nr)); |
| 794 | } |
| 795 | |
| 796 | static inline int valid_section(struct mem_section *section) |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 797 | { |
Bob Picco | 802f192 | 2005-09-03 15:54:26 -0700 | [diff] [blame] | 798 | return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 799 | } |
| 800 | |
| 801 | static inline int valid_section_nr(unsigned long nr) |
| 802 | { |
| 803 | return valid_section(__nr_to_section(nr)); |
| 804 | } |
| 805 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 806 | static inline struct mem_section *__pfn_to_section(unsigned long pfn) |
| 807 | { |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 808 | return __nr_to_section(pfn_to_section_nr(pfn)); |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 809 | } |
| 810 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 811 | static inline int pfn_valid(unsigned long pfn) |
| 812 | { |
| 813 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) |
| 814 | return 0; |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 815 | return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 816 | } |
| 817 | |
Andy Whitcroft | 540557b | 2007-10-16 01:24:11 -0700 | [diff] [blame] | 818 | static inline int pfn_present(unsigned long pfn) |
| 819 | { |
| 820 | if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) |
| 821 | return 0; |
| 822 | return present_section(__nr_to_section(pfn_to_section_nr(pfn))); |
| 823 | } |
| 824 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 825 | /* |
| 826 | * These are _only_ used during initialisation, therefore they |
| 827 | * can use __initdata ... They could have names to indicate |
| 828 | * this restriction. |
| 829 | */ |
| 830 | #ifdef CONFIG_NUMA |
Andy Whitcroft | 161599f | 2006-01-06 00:10:54 -0800 | [diff] [blame] | 831 | #define pfn_to_nid(pfn) \ |
| 832 | ({ \ |
| 833 | unsigned long __pfn_to_nid_pfn = (pfn); \ |
| 834 | page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ |
| 835 | }) |
Andy Whitcroft | 2bdaf11 | 2006-01-06 00:10:53 -0800 | [diff] [blame] | 836 | #else |
| 837 | #define pfn_to_nid(pfn) (0) |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 838 | #endif |
| 839 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 840 | #define early_pfn_valid(pfn) pfn_valid(pfn) |
| 841 | void sparse_init(void); |
| 842 | #else |
| 843 | #define sparse_init() do {} while (0) |
Dave Hansen | 28ae55c | 2005-09-03 15:54:29 -0700 | [diff] [blame] | 844 | #define sparse_index_init(_sec, _nid) do {} while (0) |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 845 | #endif /* CONFIG_SPARSEMEM */ |
| 846 | |
Andy Whitcroft | 7516795 | 2006-10-21 10:24:14 -0700 | [diff] [blame] | 847 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
| 848 | #define early_pfn_in_nid(pfn, nid) (early_pfn_to_nid(pfn) == (nid)) |
| 849 | #else |
| 850 | #define early_pfn_in_nid(pfn, nid) (1) |
| 851 | #endif |
| 852 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 853 | #ifndef early_pfn_valid |
| 854 | #define early_pfn_valid(pfn) (1) |
| 855 | #endif |
| 856 | |
| 857 | void memory_present(int nid, unsigned long start, unsigned long end); |
| 858 | unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); |
| 859 | |
Andy Whitcroft | 14e0729 | 2007-05-06 14:49:14 -0700 | [diff] [blame] | 860 | /* |
| 861 | * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we |
| 862 | * need to check pfn validility within that MAX_ORDER_NR_PAGES block. |
| 863 | * pfn_valid_within() should be used in this case; we optimise this away |
| 864 | * when we have no holes within a MAX_ORDER_NR_PAGES block. |
| 865 | */ |
| 866 | #ifdef CONFIG_HOLES_IN_ZONE |
| 867 | #define pfn_valid_within(pfn) pfn_valid(pfn) |
| 868 | #else |
| 869 | #define pfn_valid_within(pfn) (1) |
| 870 | #endif |
| 871 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 872 | #endif /* !__ASSEMBLY__ */ |
| 873 | #endif /* __KERNEL__ */ |
| 874 | #endif /* _LINUX_MMZONE_H */ |