1 #ifndef _LINUX_MMZONE_H
2 #define _LINUX_MMZONE_H
7 #include <linux/config.h>
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/cache.h>
12 #include <linux/threads.h>
13 #include <linux/numa.h>
14 #include <linux/init.h>
15 #include <asm/atomic.h>
17 /* Free memory management - zoned buddy allocator. */
18 #ifndef CONFIG_FORCE_MAX_ZONEORDER
21 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
25 struct list_head free_list;
26 unsigned long nr_free;
32 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
33 * So add a wild amount of padding here to ensure that they fall into separate
34 * cachelines. There are very few zone structures in the machine, so space
35 * consumption is not a concern here.
37 #if defined(CONFIG_SMP)
40 } ____cacheline_maxaligned_in_smp;
41 #define ZONE_PADDING(name) struct zone_padding name;
43 #define ZONE_PADDING(name)
46 struct per_cpu_pages {
47 int count; /* number of pages in the list */
48 int low; /* low watermark, refill needed */
49 int high; /* high watermark, emptying needed */
50 int batch; /* chunk size for buddy add/remove */
51 struct list_head list; /* the list of pages */
54 struct per_cpu_pageset {
55 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
57 unsigned long numa_hit; /* allocated in intended node */
58 unsigned long numa_miss; /* allocated in non intended node */
59 unsigned long numa_foreign; /* was intended here, hit elsewhere */
60 unsigned long interleave_hit; /* interleaver prefered this zone */
61 unsigned long local_node; /* allocation from local node */
62 unsigned long other_node; /* allocation from other node */
64 } ____cacheline_aligned_in_smp;
68 #define ZONE_HIGHMEM 2
70 #define MAX_NR_ZONES 3 /* Sync this with ZONES_SHIFT */
71 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
75 * When a memory allocation must conform to specific limitations (such
76 * as being suitable for DMA) the caller will pass in hints to the
77 * allocator in the gfp_mask, in the zone modifier bits. These bits
78 * are used to select a priority ordered list of memory zones which
79 * match the requested limits. GFP_ZONEMASK defines which bits within
80 * the gfp_mask should be considered as zone modifiers. Each valid
81 * combination of the zone modifier bits has a corresponding list
82 * of zones (in node_zonelists). Thus for two zone modifiers there
83 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
84 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
85 * combinations of zone modifiers in "zone modifier space".
87 #define GFP_ZONEMASK 0x03
89 * As an optimisation any zone modifier bits which are only valid when
90 * no other zone modifier bits are set (loners) should be placed in
91 * the highest order bits of this field. This allows us to reduce the
92 * extent of the zonelists thus saving space. For example in the case
93 * of three zone modifier bits, we could require up to eight zonelists.
94 * If the left most zone modifier is a "loner" then the highest valid
95 * zonelist would be four allowing us to allocate only five zonelists.
96 * Use the first form when the left most bit is not a "loner", otherwise
99 /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */
100 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
103 * On machines where it is needed (eg PCs) we divide physical memory
104 * into multiple physical zones. On a PC we have 3 zones:
106 * ZONE_DMA < 16 MB ISA DMA capable memory
107 * ZONE_NORMAL 16-896 MB direct mapped by the kernel
108 * ZONE_HIGHMEM > 896 MB only page cache and user processes
112 /* Fields commonly accessed by the page allocator */
113 unsigned long free_pages;
114 unsigned long pages_min, pages_low, pages_high;
116 * We don't know if the memory that we're going to allocate will be freeable
117 * or/and it will be released eventually, so to avoid totally wasting several
118 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
119 * to run OOM on the lower zones despite there's tons of freeable ram
120 * on the higher zones). This array is recalculated at runtime if the
121 * sysctl_lowmem_reserve_ratio sysctl changes.
123 unsigned long lowmem_reserve[MAX_NR_ZONES];
125 struct per_cpu_pageset pageset[NR_CPUS];
128 * free areas of different sizes
131 struct free_area free_area[MAX_ORDER];
136 /* Fields commonly accessed by the page reclaim scanner */
138 struct list_head active_list;
139 struct list_head inactive_list;
140 unsigned long nr_scan_active;
141 unsigned long nr_scan_inactive;
142 unsigned long nr_active;
143 unsigned long nr_inactive;
144 unsigned long pages_scanned; /* since last reclaim */
145 int all_unreclaimable; /* All pages pinned */
148 * Does the allocator try to reclaim pages from the zone as soon
149 * as it fails a watermark_ok() in __alloc_pages?
152 /* A count of how many reclaimers are scanning this zone */
153 atomic_t reclaim_in_progress;
156 * prev_priority holds the scanning priority for this zone. It is
157 * defined as the scanning priority at which we achieved our reclaim
158 * target at the previous try_to_free_pages() or balance_pgdat()
161 * We use prev_priority as a measure of how much stress page reclaim is
162 * under - it drives the swappiness decision: whether to unmap mapped
165 * temp_priority is used to remember the scanning priority at which
166 * this zone was successfully refilled to free_pages == pages_high.
168 * Access to both these fields is quite racy even on uniprocessor. But
169 * it is expected to average out OK.
176 /* Rarely used or read-mostly fields */
179 * wait_table -- the array holding the hash table
180 * wait_table_size -- the size of the hash table array
181 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
183 * The purpose of all these is to keep track of the people
184 * waiting for a page to become available and make them
185 * runnable again when possible. The trouble is that this
186 * consumes a lot of space, especially when so few things
187 * wait on pages at a given time. So instead of using
188 * per-page waitqueues, we use a waitqueue hash table.
190 * The bucket discipline is to sleep on the same queue when
191 * colliding and wake all in that wait queue when removing.
192 * When something wakes, it must check to be sure its page is
193 * truly available, a la thundering herd. The cost of a
194 * collision is great, but given the expected load of the
195 * table, they should be so rare as to be outweighed by the
196 * benefits from the saved space.
198 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
199 * primary users of these fields, and in mm/page_alloc.c
200 * free_area_init_core() performs the initialization of them.
202 wait_queue_head_t * wait_table;
203 unsigned long wait_table_size;
204 unsigned long wait_table_bits;
207 * Discontig memory support fields.
209 struct pglist_data *zone_pgdat;
210 struct page *zone_mem_map;
211 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
212 unsigned long zone_start_pfn;
214 unsigned long spanned_pages; /* total size, including holes */
215 unsigned long present_pages; /* amount of memory (excluding holes) */
218 * rarely used fields:
221 } ____cacheline_maxaligned_in_smp;
225 * The "priority" of VM scanning is how much of the queues we will scan in one
226 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
227 * queues ("queue_length >> 12") during an aging round.
229 #define DEF_PRIORITY 12
232 * One allocation request operates on a zonelist. A zonelist
233 * is a list of zones, the first one is the 'goal' of the
234 * allocation, the other zones are fallback zones, in decreasing
237 * Right now a zonelist takes up less than a cacheline. We never
238 * modify it apart from boot-up, and only a few indices are used,
239 * so despite the zonelist table being relatively big, the cache
240 * footprint of this construct is very small.
243 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
248 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
249 * (mostly NUMA machines?) to denote a higher-level memory zone than the
252 * On NUMA machines, each NUMA node would have a pg_data_t to describe
253 * it's memory layout.
255 * Memory statistics and page replacement data structures are maintained on a
259 typedef struct pglist_data {
260 struct zone node_zones[MAX_NR_ZONES];
261 struct zonelist node_zonelists[GFP_ZONETYPES];
263 struct page *node_mem_map;
264 struct bootmem_data *bdata;
265 unsigned long node_start_pfn;
266 unsigned long node_present_pages; /* total number of physical pages */
267 unsigned long node_spanned_pages; /* total size of physical page
268 range, including holes */
270 struct pglist_data *pgdat_next;
271 wait_queue_head_t kswapd_wait;
272 struct task_struct *kswapd;
273 int kswapd_max_order;
276 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
277 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
279 extern struct pglist_data *pgdat_list;
281 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
282 unsigned long *free, struct pglist_data *pgdat);
283 void get_zone_counts(unsigned long *active, unsigned long *inactive,
284 unsigned long *free);
285 void build_all_zonelists(void);
286 void wakeup_kswapd(struct zone *zone, int order);
287 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
288 int alloc_type, int can_try_harder, int gfp_high);
290 #ifdef CONFIG_HAVE_MEMORY_PRESENT
291 void memory_present(int nid, unsigned long start, unsigned long end);
293 static inline void memory_present(int nid, unsigned long start, unsigned long end) {}
296 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE
297 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
301 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
303 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
306 * for_each_pgdat - helper macro to iterate over all nodes
307 * @pgdat - pointer to a pg_data_t variable
309 * Meant to help with common loops of the form
310 * pgdat = pgdat_list;
313 * pgdat = pgdat->pgdat_next;
316 #define for_each_pgdat(pgdat) \
317 for (pgdat = pgdat_list; pgdat; pgdat = pgdat->pgdat_next)
320 * next_zone - helper magic for for_each_zone()
321 * Thanks to William Lee Irwin III for this piece of ingenuity.
323 static inline struct zone *next_zone(struct zone *zone)
325 pg_data_t *pgdat = zone->zone_pgdat;
327 if (zone < pgdat->node_zones + MAX_NR_ZONES - 1)
329 else if (pgdat->pgdat_next) {
330 pgdat = pgdat->pgdat_next;
331 zone = pgdat->node_zones;
339 * for_each_zone - helper macro to iterate over all memory zones
340 * @zone - pointer to struct zone variable
342 * The user only needs to declare the zone variable, for_each_zone
343 * fills it in. This basically means for_each_zone() is an
344 * easier to read version of this piece of code:
346 * for (pgdat = pgdat_list; pgdat; pgdat = pgdat->node_next)
347 * for (i = 0; i < MAX_NR_ZONES; ++i) {
348 * struct zone * z = pgdat->node_zones + i;
353 #define for_each_zone(zone) \
354 for (zone = pgdat_list->node_zones; zone; zone = next_zone(zone))
356 static inline int is_highmem_idx(int idx)
358 return (idx == ZONE_HIGHMEM);
361 static inline int is_normal_idx(int idx)
363 return (idx == ZONE_NORMAL);
366 * is_highmem - helper function to quickly check if a struct zone is a
367 * highmem zone or not. This is an attempt to keep references
368 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
369 * @zone - pointer to struct zone variable
371 static inline int is_highmem(struct zone *zone)
373 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
376 static inline int is_normal(struct zone *zone)
378 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
381 /* These two functions are used to setup the per zone pages min values */
384 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
385 void __user *, size_t *, loff_t *);
386 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
387 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
388 void __user *, size_t *, loff_t *);
390 #include <linux/topology.h>
391 /* Returns the number of the current Node. */
392 #define numa_node_id() (cpu_to_node(raw_smp_processor_id()))
394 #ifndef CONFIG_DISCONTIGMEM
396 extern struct pglist_data contig_page_data;
397 #define NODE_DATA(nid) (&contig_page_data)
398 #define NODE_MEM_MAP(nid) mem_map
399 #define MAX_NODES_SHIFT 1
400 #define pfn_to_nid(pfn) (0)
402 #else /* CONFIG_DISCONTIGMEM */
404 #include <asm/mmzone.h>
406 #if BITS_PER_LONG == 32 || defined(ARCH_HAS_ATOMIC_UNSIGNED)
408 * with 32 bit page->flags field, we reserve 8 bits for node/zone info.
409 * there are 3 zones (2 bits) and this leaves 8-2=6 bits for nodes.
411 #define MAX_NODES_SHIFT 6
412 #elif BITS_PER_LONG == 64
414 * with 64 bit flags field, there's plenty of room.
416 #define MAX_NODES_SHIFT 10
419 #endif /* !CONFIG_DISCONTIGMEM */
421 #if NODES_SHIFT > MAX_NODES_SHIFT
422 #error NODES_SHIFT > MAX_NODES_SHIFT
425 /* There are currently 3 zones: DMA, Normal & Highmem, thus we need 2 bits */
426 #define MAX_ZONES_SHIFT 2
428 #if ZONES_SHIFT > MAX_ZONES_SHIFT
429 #error ZONES_SHIFT > MAX_ZONES_SHIFT
432 #endif /* !__ASSEMBLY__ */
433 #endif /* __KERNEL__ */
434 #endif /* _LINUX_MMZONE_H */