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 <asm/atomic.h>
16 /* Free memory management - zoned buddy allocator. */
17 #ifndef CONFIG_FORCE_MAX_ZONEORDER
20 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
24 struct list_head free_list;
25 unsigned long nr_free;
31 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel.
32 * So add a wild amount of padding here to ensure that they fall into separate
33 * cachelines. There are very few zone structures in the machine, so space
34 * consumption is not a concern here.
36 #if defined(CONFIG_SMP)
39 } ____cacheline_maxaligned_in_smp;
40 #define ZONE_PADDING(name) struct zone_padding name;
42 #define ZONE_PADDING(name)
45 struct per_cpu_pages {
46 int count; /* number of pages in the list */
47 int low; /* low watermark, refill needed */
48 int high; /* high watermark, emptying needed */
49 int batch; /* chunk size for buddy add/remove */
50 struct list_head list; /* the list of pages */
53 struct per_cpu_pageset {
54 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */
56 unsigned long numa_hit; /* allocated in intended node */
57 unsigned long numa_miss; /* allocated in non intended node */
58 unsigned long numa_foreign; /* was intended here, hit elsewhere */
59 unsigned long interleave_hit; /* interleaver prefered this zone */
60 unsigned long local_node; /* allocation from local node */
61 unsigned long other_node; /* allocation from other node */
63 } ____cacheline_aligned_in_smp;
67 #define ZONE_HIGHMEM 2
69 #define MAX_NR_ZONES 3 /* Sync this with ZONES_SHIFT */
70 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */
74 * When a memory allocation must conform to specific limitations (such
75 * as being suitable for DMA) the caller will pass in hints to the
76 * allocator in the gfp_mask, in the zone modifier bits. These bits
77 * are used to select a priority ordered list of memory zones which
78 * match the requested limits. GFP_ZONEMASK defines which bits within
79 * the gfp_mask should be considered as zone modifiers. Each valid
80 * combination of the zone modifier bits has a corresponding list
81 * of zones (in node_zonelists). Thus for two zone modifiers there
82 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will
83 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible
84 * combinations of zone modifiers in "zone modifier space".
86 #define GFP_ZONEMASK 0x03
88 * As an optimisation any zone modifier bits which are only valid when
89 * no other zone modifier bits are set (loners) should be placed in
90 * the highest order bits of this field. This allows us to reduce the
91 * extent of the zonelists thus saving space. For example in the case
92 * of three zone modifier bits, we could require up to eight zonelists.
93 * If the left most zone modifier is a "loner" then the highest valid
94 * zonelist would be four allowing us to allocate only five zonelists.
95 * Use the first form when the left most bit is not a "loner", otherwise
98 /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */
99 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */
102 * On machines where it is needed (eg PCs) we divide physical memory
103 * into multiple physical zones. On a PC we have 3 zones:
105 * ZONE_DMA < 16 MB ISA DMA capable memory
106 * ZONE_NORMAL 16-896 MB direct mapped by the kernel
107 * ZONE_HIGHMEM > 896 MB only page cache and user processes
111 /* Fields commonly accessed by the page allocator */
112 unsigned long free_pages;
113 unsigned long pages_min, pages_low, pages_high;
115 * We don't know if the memory that we're going to allocate will be freeable
116 * or/and it will be released eventually, so to avoid totally wasting several
117 * GB of ram we must reserve some of the lower zone memory (otherwise we risk
118 * to run OOM on the lower zones despite there's tons of freeable ram
119 * on the higher zones). This array is recalculated at runtime if the
120 * sysctl_lowmem_reserve_ratio sysctl changes.
122 unsigned long lowmem_reserve[MAX_NR_ZONES];
124 struct per_cpu_pageset pageset[NR_CPUS];
127 * free areas of different sizes
130 struct free_area free_area[MAX_ORDER];
135 /* Fields commonly accessed by the page reclaim scanner */
137 struct list_head active_list;
138 struct list_head inactive_list;
139 unsigned long nr_scan_active;
140 unsigned long nr_scan_inactive;
141 unsigned long nr_active;
142 unsigned long nr_inactive;
143 unsigned long pages_scanned; /* since last reclaim */
144 int all_unreclaimable; /* All pages pinned */
147 * prev_priority holds the scanning priority for this zone. It is
148 * defined as the scanning priority at which we achieved our reclaim
149 * target at the previous try_to_free_pages() or balance_pgdat()
152 * We use prev_priority as a measure of how much stress page reclaim is
153 * under - it drives the swappiness decision: whether to unmap mapped
156 * temp_priority is used to remember the scanning priority at which
157 * this zone was successfully refilled to free_pages == pages_high.
159 * Access to both these fields is quite racy even on uniprocessor. But
160 * it is expected to average out OK.
167 /* Rarely used or read-mostly fields */
170 * wait_table -- the array holding the hash table
171 * wait_table_size -- the size of the hash table array
172 * wait_table_bits -- wait_table_size == (1 << wait_table_bits)
174 * The purpose of all these is to keep track of the people
175 * waiting for a page to become available and make them
176 * runnable again when possible. The trouble is that this
177 * consumes a lot of space, especially when so few things
178 * wait on pages at a given time. So instead of using
179 * per-page waitqueues, we use a waitqueue hash table.
181 * The bucket discipline is to sleep on the same queue when
182 * colliding and wake all in that wait queue when removing.
183 * When something wakes, it must check to be sure its page is
184 * truly available, a la thundering herd. The cost of a
185 * collision is great, but given the expected load of the
186 * table, they should be so rare as to be outweighed by the
187 * benefits from the saved space.
189 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the
190 * primary users of these fields, and in mm/page_alloc.c
191 * free_area_init_core() performs the initialization of them.
193 wait_queue_head_t * wait_table;
194 unsigned long wait_table_size;
195 unsigned long wait_table_bits;
198 * Discontig memory support fields.
200 struct pglist_data *zone_pgdat;
201 struct page *zone_mem_map;
202 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
203 unsigned long zone_start_pfn;
205 unsigned long spanned_pages; /* total size, including holes */
206 unsigned long present_pages; /* amount of memory (excluding holes) */
209 * rarely used fields:
212 } ____cacheline_maxaligned_in_smp;
216 * The "priority" of VM scanning is how much of the queues we will scan in one
217 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
218 * queues ("queue_length >> 12") during an aging round.
220 #define DEF_PRIORITY 12
223 * One allocation request operates on a zonelist. A zonelist
224 * is a list of zones, the first one is the 'goal' of the
225 * allocation, the other zones are fallback zones, in decreasing
228 * Right now a zonelist takes up less than a cacheline. We never
229 * modify it apart from boot-up, and only a few indices are used,
230 * so despite the zonelist table being relatively big, the cache
231 * footprint of this construct is very small.
234 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited
239 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM
240 * (mostly NUMA machines?) to denote a higher-level memory zone than the
243 * On NUMA machines, each NUMA node would have a pg_data_t to describe
244 * it's memory layout.
246 * Memory statistics and page replacement data structures are maintained on a
250 typedef struct pglist_data {
251 struct zone node_zones[MAX_NR_ZONES];
252 struct zonelist node_zonelists[GFP_ZONETYPES];
254 struct page *node_mem_map;
255 struct bootmem_data *bdata;
256 unsigned long node_start_pfn;
257 unsigned long node_present_pages; /* total number of physical pages */
258 unsigned long node_spanned_pages; /* total size of physical page
259 range, including holes */
261 struct pglist_data *pgdat_next;
262 wait_queue_head_t kswapd_wait;
263 struct task_struct *kswapd;
264 int kswapd_max_order;
267 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
268 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
270 extern struct pglist_data *pgdat_list;
272 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
273 unsigned long *free, struct pglist_data *pgdat);
274 void get_zone_counts(unsigned long *active, unsigned long *inactive,
275 unsigned long *free);
276 void build_all_zonelists(void);
277 void wakeup_kswapd(struct zone *zone, int order);
278 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
279 int alloc_type, int can_try_harder, int gfp_high);
282 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
284 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
287 * for_each_pgdat - helper macro to iterate over all nodes
288 * @pgdat - pointer to a pg_data_t variable
290 * Meant to help with common loops of the form
291 * pgdat = pgdat_list;
294 * pgdat = pgdat->pgdat_next;
297 #define for_each_pgdat(pgdat) \
298 for (pgdat = pgdat_list; pgdat; pgdat = pgdat->pgdat_next)
301 * next_zone - helper magic for for_each_zone()
302 * Thanks to William Lee Irwin III for this piece of ingenuity.
304 static inline struct zone *next_zone(struct zone *zone)
306 pg_data_t *pgdat = zone->zone_pgdat;
308 if (zone < pgdat->node_zones + MAX_NR_ZONES - 1)
310 else if (pgdat->pgdat_next) {
311 pgdat = pgdat->pgdat_next;
312 zone = pgdat->node_zones;
320 * for_each_zone - helper macro to iterate over all memory zones
321 * @zone - pointer to struct zone variable
323 * The user only needs to declare the zone variable, for_each_zone
324 * fills it in. This basically means for_each_zone() is an
325 * easier to read version of this piece of code:
327 * for (pgdat = pgdat_list; pgdat; pgdat = pgdat->node_next)
328 * for (i = 0; i < MAX_NR_ZONES; ++i) {
329 * struct zone * z = pgdat->node_zones + i;
334 #define for_each_zone(zone) \
335 for (zone = pgdat_list->node_zones; zone; zone = next_zone(zone))
337 static inline int is_highmem_idx(int idx)
339 return (idx == ZONE_HIGHMEM);
342 static inline int is_normal_idx(int idx)
344 return (idx == ZONE_NORMAL);
347 * is_highmem - helper function to quickly check if a struct zone is a
348 * highmem zone or not. This is an attempt to keep references
349 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
350 * @zone - pointer to struct zone variable
352 static inline int is_highmem(struct zone *zone)
354 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM;
357 static inline int is_normal(struct zone *zone)
359 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL;
362 /* These two functions are used to setup the per zone pages min values */
365 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *,
366 void __user *, size_t *, loff_t *);
367 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1];
368 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *,
369 void __user *, size_t *, loff_t *);
371 #include <linux/topology.h>
372 /* Returns the number of the current Node. */
373 #define numa_node_id() (cpu_to_node(_smp_processor_id()))
375 #ifndef CONFIG_DISCONTIGMEM
377 extern struct pglist_data contig_page_data;
378 #define NODE_DATA(nid) (&contig_page_data)
379 #define NODE_MEM_MAP(nid) mem_map
380 #define MAX_NODES_SHIFT 1
381 #define pfn_to_nid(pfn) (0)
383 #else /* CONFIG_DISCONTIGMEM */
385 #include <asm/mmzone.h>
387 #if BITS_PER_LONG == 32 || defined(ARCH_HAS_ATOMIC_UNSIGNED)
389 * with 32 bit page->flags field, we reserve 8 bits for node/zone info.
390 * there are 3 zones (2 bits) and this leaves 8-2=6 bits for nodes.
392 #define MAX_NODES_SHIFT 6
393 #elif BITS_PER_LONG == 64
395 * with 64 bit flags field, there's plenty of room.
397 #define MAX_NODES_SHIFT 10
400 #endif /* !CONFIG_DISCONTIGMEM */
402 #if NODES_SHIFT > MAX_NODES_SHIFT
403 #error NODES_SHIFT > MAX_NODES_SHIFT
406 /* There are currently 3 zones: DMA, Normal & Highmem, thus we need 2 bits */
407 #define MAX_ZONES_SHIFT 2
409 #if ZONES_SHIFT > MAX_ZONES_SHIFT
410 #error ZONES_SHIFT > MAX_ZONES_SHIFT
413 #endif /* !__ASSEMBLY__ */
414 #endif /* __KERNEL__ */
415 #endif /* _LINUX_MMZONE_H */