1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_MMZONE_H
3 #define _LINUX_MMZONE_H
6 #ifndef __GENERATING_BOUNDS_H
8 #include <linux/spinlock.h>
9 #include <linux/list.h>
10 #include <linux/wait.h>
11 #include <linux/bitops.h>
12 #include <linux/cache.h>
13 #include <linux/threads.h>
14 #include <linux/numa.h>
15 #include <linux/init.h>
16 #include <linux/seqlock.h>
17 #include <linux/nodemask.h>
18 #include <linux/pageblock-flags.h>
19 #include <linux/page-flags-layout.h>
20 #include <linux/atomic.h>
21 #include <linux/mm_types.h>
22 #include <linux/page-flags.h>
25 /* Free memory management - zoned buddy allocator. */
26 #ifndef CONFIG_FORCE_MAX_ZONEORDER
29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
35 * costly to service. That is between allocation orders which should
36 * coalesce naturally under reasonable reclaim pressure and those which
39 #define PAGE_ALLOC_COSTLY_ORDER 3
45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
49 * MIGRATE_CMA migration type is designed to mimic the way
50 * ZONE_MOVABLE works. Only movable pages can be allocated
51 * from MIGRATE_CMA pageblocks and page allocator never
52 * implicitly change migration type of MIGRATE_CMA pageblock.
54 * The way to use it is to change migratetype of a range of
55 * pageblocks to MIGRATE_CMA which can be done by
56 * __free_pageblock_cma() function. What is important though
57 * is that a range of pageblocks must be aligned to
58 * MAX_ORDER_NR_PAGES should biggest page be bigger then
63 #ifdef CONFIG_MEMORY_ISOLATION
64 MIGRATE_ISOLATE, /* can't allocate from here */
69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
70 extern const char * const migratetype_names[MIGRATE_TYPES];
73 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
74 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
76 # define is_migrate_cma(migratetype) false
77 # define is_migrate_cma_page(_page) false
80 static inline bool is_migrate_movable(int mt)
82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
85 #define for_each_migratetype_order(order, type) \
86 for (order = 0; order < MAX_ORDER; order++) \
87 for (type = 0; type < MIGRATE_TYPES; type++)
89 extern int page_group_by_mobility_disabled;
91 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
93 #define get_pageblock_migratetype(page) \
94 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
97 struct list_head free_list[MIGRATE_TYPES];
98 unsigned long nr_free;
101 static inline struct page *get_page_from_free_area(struct free_area *area,
104 return list_first_entry_or_null(&area->free_list[migratetype],
108 static inline bool free_area_empty(struct free_area *area, int migratetype)
110 return list_empty(&area->free_list[migratetype]);
116 * zone->lock and the zone lru_lock are two of the hottest locks in the kernel.
117 * So add a wild amount of padding here to ensure that they fall into separate
118 * cachelines. There are very few zone structures in the machine, so space
119 * consumption is not a concern here.
121 #if defined(CONFIG_SMP)
122 struct zone_padding {
124 } ____cacheline_internodealigned_in_smp;
125 #define ZONE_PADDING(name) struct zone_padding name;
127 #define ZONE_PADDING(name)
131 enum numa_stat_item {
132 NUMA_HIT, /* allocated in intended node */
133 NUMA_MISS, /* allocated in non intended node */
134 NUMA_FOREIGN, /* was intended here, hit elsewhere */
135 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
136 NUMA_LOCAL, /* allocation from local node */
137 NUMA_OTHER, /* allocation from other node */
138 NR_VM_NUMA_STAT_ITEMS
141 #define NR_VM_NUMA_STAT_ITEMS 0
144 enum zone_stat_item {
145 /* First 128 byte cacheline (assuming 64 bit words) */
147 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
148 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
150 NR_ZONE_INACTIVE_FILE,
153 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
154 NR_MLOCK, /* mlock()ed pages found and moved off LRU */
155 /* Second 128 byte cacheline */
157 #if IS_ENABLED(CONFIG_ZSMALLOC)
158 NR_ZSPAGES, /* allocated in zsmalloc */
161 NR_VM_ZONE_STAT_ITEMS };
163 enum node_stat_item {
165 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
166 NR_ACTIVE_ANON, /* " " " " " */
167 NR_INACTIVE_FILE, /* " " " " " */
168 NR_ACTIVE_FILE, /* " " " " " */
169 NR_UNEVICTABLE, /* " " " " " */
170 NR_SLAB_RECLAIMABLE_B,
171 NR_SLAB_UNRECLAIMABLE_B,
172 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
173 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
175 WORKINGSET_REFAULT_BASE,
176 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
177 WORKINGSET_REFAULT_FILE,
178 WORKINGSET_ACTIVATE_BASE,
179 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
180 WORKINGSET_ACTIVATE_FILE,
181 WORKINGSET_RESTORE_BASE,
182 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
183 WORKINGSET_RESTORE_FILE,
184 WORKINGSET_NODERECLAIM,
185 NR_ANON_MAPPED, /* Mapped anonymous pages */
186 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
187 only modified from process context */
191 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
192 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
199 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
200 NR_DIRTIED, /* page dirtyings since bootup */
201 NR_WRITTEN, /* page writings since bootup */
202 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
203 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
204 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
205 NR_KERNEL_STACK_KB, /* measured in KiB */
206 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
207 NR_KERNEL_SCS_KB, /* measured in KiB */
209 NR_PAGETABLE, /* used for pagetables */
210 NR_VM_NODE_STAT_ITEMS
214 * Returns true if the value is measured in bytes (most vmstat values are
215 * measured in pages). This defines the API part, the internal representation
216 * might be different.
218 static __always_inline bool vmstat_item_in_bytes(int idx)
221 * Global and per-node slab counters track slab pages.
222 * It's expected that changes are multiples of PAGE_SIZE.
223 * Internally values are stored in pages.
225 * Per-memcg and per-lruvec counters track memory, consumed
226 * by individual slab objects. These counters are actually
229 return (idx == NR_SLAB_RECLAIMABLE_B ||
230 idx == NR_SLAB_UNRECLAIMABLE_B);
234 * We do arithmetic on the LRU lists in various places in the code,
235 * so it is important to keep the active lists LRU_ACTIVE higher in
236 * the array than the corresponding inactive lists, and to keep
237 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
239 * This has to be kept in sync with the statistics in zone_stat_item
240 * above and the descriptions in vmstat_text in mm/vmstat.c
247 LRU_INACTIVE_ANON = LRU_BASE,
248 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
249 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
250 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
255 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
257 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
259 static inline bool is_file_lru(enum lru_list lru)
261 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
264 static inline bool is_active_lru(enum lru_list lru)
266 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
269 #define ANON_AND_FILE 2
272 LRUVEC_CONGESTED, /* lruvec has many dirty pages
273 * backed by a congested BDI
278 struct list_head lists[NR_LRU_LISTS];
280 * These track the cost of reclaiming one LRU - file or anon -
281 * over the other. As the observed cost of reclaiming one LRU
282 * increases, the reclaim scan balance tips toward the other.
284 unsigned long anon_cost;
285 unsigned long file_cost;
286 /* Non-resident age, driven by LRU movement */
287 atomic_long_t nonresident_age;
288 /* Refaults at the time of last reclaim cycle */
289 unsigned long refaults[ANON_AND_FILE];
290 /* Various lruvec state flags (enum lruvec_flags) */
293 struct pglist_data *pgdat;
297 /* Isolate unmapped pages */
298 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
299 /* Isolate for asynchronous migration */
300 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
301 /* Isolate unevictable pages */
302 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
304 /* LRU Isolation modes. */
305 typedef unsigned __bitwise isolate_mode_t;
307 enum zone_watermarks {
314 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
315 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
316 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
317 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
319 struct per_cpu_pages {
320 int count; /* number of pages in the list */
321 int high; /* high watermark, emptying needed */
322 int batch; /* chunk size for buddy add/remove */
324 /* Lists of pages, one per migrate type stored on the pcp-lists */
325 struct list_head lists[MIGRATE_PCPTYPES];
328 struct per_cpu_pageset {
329 struct per_cpu_pages pcp;
332 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS];
336 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
340 struct per_cpu_nodestat {
342 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
345 #endif /* !__GENERATING_BOUNDS.H */
349 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
350 * to DMA to all of the addressable memory (ZONE_NORMAL).
351 * On architectures where this area covers the whole 32 bit address
352 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
353 * DMA addressing constraints. This distinction is important as a 32bit
354 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
355 * platforms may need both zones as they support peripherals with
356 * different DMA addressing limitations.
360 * - i386 and x86_64 have a fixed 16M ZONE_DMA and ZONE_DMA32 for the
361 * rest of the lower 4G.
363 * - arm only uses ZONE_DMA, the size, up to 4G, may vary depending on
364 * the specific device.
366 * - arm64 has a fixed 1G ZONE_DMA and ZONE_DMA32 for the rest of the
369 * - powerpc only uses ZONE_DMA, the size, up to 2G, may vary
370 * depending on the specific device.
372 * - s390 uses ZONE_DMA fixed to the lower 2G.
374 * - ia64 and riscv only use ZONE_DMA32.
376 * - parisc uses neither.
378 #ifdef CONFIG_ZONE_DMA
381 #ifdef CONFIG_ZONE_DMA32
385 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
386 * performed on pages in ZONE_NORMAL if the DMA devices support
387 * transfers to all addressable memory.
390 #ifdef CONFIG_HIGHMEM
392 * A memory area that is only addressable by the kernel through
393 * mapping portions into its own address space. This is for example
394 * used by i386 to allow the kernel to address the memory beyond
395 * 900MB. The kernel will set up special mappings (page
396 * table entries on i386) for each page that the kernel needs to
402 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
403 * movable pages with few exceptional cases described below. Main use
404 * cases for ZONE_MOVABLE are to make memory offlining/unplug more
405 * likely to succeed, and to locally limit unmovable allocations - e.g.,
406 * to increase the number of THP/huge pages. Notable special cases are:
408 * 1. Pinned pages: (long-term) pinning of movable pages might
409 * essentially turn such pages unmovable. Memory offlining might
411 * 2. memblock allocations: kernelcore/movablecore setups might create
412 * situations where ZONE_MOVABLE contains unmovable allocations
413 * after boot. Memory offlining and allocations fail early.
414 * 3. Memory holes: kernelcore/movablecore setups might create very rare
415 * situations where ZONE_MOVABLE contains memory holes after boot,
416 * for example, if we have sections that are only partially
417 * populated. Memory offlining and allocations fail early.
418 * 4. PG_hwpoison pages: while poisoned pages can be skipped during
419 * memory offlining, such pages cannot be allocated.
420 * 5. Unmovable PG_offline pages: in paravirtualized environments,
421 * hotplugged memory blocks might only partially be managed by the
422 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
423 * parts not manged by the buddy are unmovable PG_offline pages. In
424 * some cases (virtio-mem), such pages can be skipped during
425 * memory offlining, however, cannot be moved/allocated. These
426 * techniques might use alloc_contig_range() to hide previously
427 * exposed pages from the buddy again (e.g., to implement some sort
428 * of memory unplug in virtio-mem).
430 * In general, no unmovable allocations that degrade memory offlining
431 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
432 * have to expect that migrating pages in ZONE_MOVABLE can fail (even
433 * if has_unmovable_pages() states that there are no unmovable pages,
434 * there can be false negatives).
437 #ifdef CONFIG_ZONE_DEVICE
444 #ifndef __GENERATING_BOUNDS_H
446 #define ASYNC_AND_SYNC 2
449 /* Read-mostly fields */
451 /* zone watermarks, access with *_wmark_pages(zone) macros */
452 unsigned long _watermark[NR_WMARK];
453 unsigned long watermark_boost;
455 unsigned long nr_reserved_highatomic;
458 * We don't know if the memory that we're going to allocate will be
459 * freeable or/and it will be released eventually, so to avoid totally
460 * wasting several GB of ram we must reserve some of the lower zone
461 * memory (otherwise we risk to run OOM on the lower zones despite
462 * there being tons of freeable ram on the higher zones). This array is
463 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
466 long lowmem_reserve[MAX_NR_ZONES];
471 struct pglist_data *zone_pgdat;
472 struct per_cpu_pageset __percpu *pageset;
474 * the high and batch values are copied to individual pagesets for
480 #ifndef CONFIG_SPARSEMEM
482 * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
483 * In SPARSEMEM, this map is stored in struct mem_section
485 unsigned long *pageblock_flags;
486 #endif /* CONFIG_SPARSEMEM */
488 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
489 unsigned long zone_start_pfn;
492 * spanned_pages is the total pages spanned by the zone, including
493 * holes, which is calculated as:
494 * spanned_pages = zone_end_pfn - zone_start_pfn;
496 * present_pages is physical pages existing within the zone, which
498 * present_pages = spanned_pages - absent_pages(pages in holes);
500 * managed_pages is present pages managed by the buddy system, which
501 * is calculated as (reserved_pages includes pages allocated by the
502 * bootmem allocator):
503 * managed_pages = present_pages - reserved_pages;
505 * So present_pages may be used by memory hotplug or memory power
506 * management logic to figure out unmanaged pages by checking
507 * (present_pages - managed_pages). And managed_pages should be used
508 * by page allocator and vm scanner to calculate all kinds of watermarks
513 * zone_start_pfn and spanned_pages are protected by span_seqlock.
514 * It is a seqlock because it has to be read outside of zone->lock,
515 * and it is done in the main allocator path. But, it is written
516 * quite infrequently.
518 * The span_seq lock is declared along with zone->lock because it is
519 * frequently read in proximity to zone->lock. It's good to
520 * give them a chance of being in the same cacheline.
522 * Write access to present_pages at runtime should be protected by
523 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
524 * present_pages should get_online_mems() to get a stable value.
526 atomic_long_t managed_pages;
527 unsigned long spanned_pages;
528 unsigned long present_pages;
532 #ifdef CONFIG_MEMORY_ISOLATION
534 * Number of isolated pageblock. It is used to solve incorrect
535 * freepage counting problem due to racy retrieving migratetype
536 * of pageblock. Protected by zone->lock.
538 unsigned long nr_isolate_pageblock;
541 #ifdef CONFIG_MEMORY_HOTPLUG
542 /* see spanned/present_pages for more description */
543 seqlock_t span_seqlock;
548 /* Write-intensive fields used from the page allocator */
551 /* free areas of different sizes */
552 struct free_area free_area[MAX_ORDER];
554 /* zone flags, see below */
557 /* Primarily protects free_area */
560 /* Write-intensive fields used by compaction and vmstats. */
564 * When free pages are below this point, additional steps are taken
565 * when reading the number of free pages to avoid per-cpu counter
566 * drift allowing watermarks to be breached
568 unsigned long percpu_drift_mark;
570 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
571 /* pfn where compaction free scanner should start */
572 unsigned long compact_cached_free_pfn;
573 /* pfn where compaction migration scanner should start */
574 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
575 unsigned long compact_init_migrate_pfn;
576 unsigned long compact_init_free_pfn;
579 #ifdef CONFIG_COMPACTION
581 * On compaction failure, 1<<compact_defer_shift compactions
582 * are skipped before trying again. The number attempted since
583 * last failure is tracked with compact_considered.
584 * compact_order_failed is the minimum compaction failed order.
586 unsigned int compact_considered;
587 unsigned int compact_defer_shift;
588 int compact_order_failed;
591 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
592 /* Set to true when the PG_migrate_skip bits should be cleared */
593 bool compact_blockskip_flush;
599 /* Zone statistics */
600 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
601 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS];
602 } ____cacheline_internodealigned_in_smp;
605 PGDAT_DIRTY, /* reclaim scanning has recently found
606 * many dirty file pages at the tail
609 PGDAT_WRITEBACK, /* reclaim scanning has recently found
610 * many pages under writeback
612 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
616 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
617 * Cleared when kswapd is woken.
621 static inline unsigned long zone_managed_pages(struct zone *zone)
623 return (unsigned long)atomic_long_read(&zone->managed_pages);
626 static inline unsigned long zone_end_pfn(const struct zone *zone)
628 return zone->zone_start_pfn + zone->spanned_pages;
631 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
633 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
636 static inline bool zone_is_initialized(struct zone *zone)
638 return zone->initialized;
641 static inline bool zone_is_empty(struct zone *zone)
643 return zone->spanned_pages == 0;
647 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
648 * intersection with the given zone
650 static inline bool zone_intersects(struct zone *zone,
651 unsigned long start_pfn, unsigned long nr_pages)
653 if (zone_is_empty(zone))
655 if (start_pfn >= zone_end_pfn(zone) ||
656 start_pfn + nr_pages <= zone->zone_start_pfn)
663 * The "priority" of VM scanning is how much of the queues we will scan in one
664 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
665 * queues ("queue_length >> 12") during an aging round.
667 #define DEF_PRIORITY 12
669 /* Maximum number of zones on a zonelist */
670 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
673 ZONELIST_FALLBACK, /* zonelist with fallback */
676 * The NUMA zonelists are doubled because we need zonelists that
677 * restrict the allocations to a single node for __GFP_THISNODE.
679 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
685 * This struct contains information about a zone in a zonelist. It is stored
686 * here to avoid dereferences into large structures and lookups of tables
689 struct zone *zone; /* Pointer to actual zone */
690 int zone_idx; /* zone_idx(zoneref->zone) */
694 * One allocation request operates on a zonelist. A zonelist
695 * is a list of zones, the first one is the 'goal' of the
696 * allocation, the other zones are fallback zones, in decreasing
699 * To speed the reading of the zonelist, the zonerefs contain the zone index
700 * of the entry being read. Helper functions to access information given
701 * a struct zoneref are
703 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
704 * zonelist_zone_idx() - Return the index of the zone for an entry
705 * zonelist_node_idx() - Return the index of the node for an entry
708 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
711 #ifndef CONFIG_DISCONTIGMEM
712 /* The array of struct pages - for discontigmem use pgdat->lmem_map */
713 extern struct page *mem_map;
716 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
717 struct deferred_split {
718 spinlock_t split_queue_lock;
719 struct list_head split_queue;
720 unsigned long split_queue_len;
725 * On NUMA machines, each NUMA node would have a pg_data_t to describe
726 * it's memory layout. On UMA machines there is a single pglist_data which
727 * describes the whole memory.
729 * Memory statistics and page replacement data structures are maintained on a
732 typedef struct pglist_data {
734 * node_zones contains just the zones for THIS node. Not all of the
735 * zones may be populated, but it is the full list. It is referenced by
736 * this node's node_zonelists as well as other node's node_zonelists.
738 struct zone node_zones[MAX_NR_ZONES];
741 * node_zonelists contains references to all zones in all nodes.
742 * Generally the first zones will be references to this node's
745 struct zonelist node_zonelists[MAX_ZONELISTS];
747 int nr_zones; /* number of populated zones in this node */
748 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */
749 struct page *node_mem_map;
750 #ifdef CONFIG_PAGE_EXTENSION
751 struct page_ext *node_page_ext;
754 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
756 * Must be held any time you expect node_start_pfn,
757 * node_present_pages, node_spanned_pages or nr_zones to stay constant.
758 * Also synchronizes pgdat->first_deferred_pfn during deferred page
761 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
762 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
763 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
765 * Nests above zone->lock and zone->span_seqlock
767 spinlock_t node_size_lock;
769 unsigned long node_start_pfn;
770 unsigned long node_present_pages; /* total number of physical pages */
771 unsigned long node_spanned_pages; /* total size of physical page
772 range, including holes */
774 wait_queue_head_t kswapd_wait;
775 wait_queue_head_t pfmemalloc_wait;
776 struct task_struct *kswapd; /* Protected by
777 mem_hotplug_begin/end() */
779 enum zone_type kswapd_highest_zoneidx;
781 int kswapd_failures; /* Number of 'reclaimed == 0' runs */
783 #ifdef CONFIG_COMPACTION
784 int kcompactd_max_order;
785 enum zone_type kcompactd_highest_zoneidx;
786 wait_queue_head_t kcompactd_wait;
787 struct task_struct *kcompactd;
790 * This is a per-node reserve of pages that are not available
791 * to userspace allocations.
793 unsigned long totalreserve_pages;
797 * node reclaim becomes active if more unmapped pages exist.
799 unsigned long min_unmapped_pages;
800 unsigned long min_slab_pages;
801 #endif /* CONFIG_NUMA */
803 /* Write-intensive fields used by page reclaim */
807 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
809 * If memory initialisation on large machines is deferred then this
810 * is the first PFN that needs to be initialised.
812 unsigned long first_deferred_pfn;
813 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
815 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
816 struct deferred_split deferred_split_queue;
819 /* Fields commonly accessed by the page reclaim scanner */
822 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
824 * Use mem_cgroup_lruvec() to look up lruvecs.
826 struct lruvec __lruvec;
832 /* Per-node vmstats */
833 struct per_cpu_nodestat __percpu *per_cpu_nodestats;
834 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
837 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
838 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
839 #ifdef CONFIG_FLAT_NODE_MEM_MAP
840 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr))
842 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr))
844 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr))
846 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
847 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
849 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
851 return pgdat->node_start_pfn + pgdat->node_spanned_pages;
854 static inline bool pgdat_is_empty(pg_data_t *pgdat)
856 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
859 #include <linux/memory_hotplug.h>
861 void build_all_zonelists(pg_data_t *pgdat);
862 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
863 enum zone_type highest_zoneidx);
864 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
865 int highest_zoneidx, unsigned int alloc_flags,
867 bool zone_watermark_ok(struct zone *z, unsigned int order,
868 unsigned long mark, int highest_zoneidx,
869 unsigned int alloc_flags);
870 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
871 unsigned long mark, int highest_zoneidx);
873 * Memory initialization context, use to differentiate memory added by
874 * the platform statically or via memory hotplug interface.
876 enum meminit_context {
881 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
884 extern void lruvec_init(struct lruvec *lruvec);
886 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
889 return lruvec->pgdat;
891 return container_of(lruvec, struct pglist_data, __lruvec);
895 extern unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru, int zone_idx);
897 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
898 int local_memory_node(int node_id);
900 static inline int local_memory_node(int node_id) { return node_id; };
904 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
906 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
909 * Returns true if a zone has pages managed by the buddy allocator.
910 * All the reclaim decisions have to use this function rather than
911 * populated_zone(). If the whole zone is reserved then we can easily
912 * end up with populated_zone() && !managed_zone().
914 static inline bool managed_zone(struct zone *zone)
916 return zone_managed_pages(zone);
919 /* Returns true if a zone has memory */
920 static inline bool populated_zone(struct zone *zone)
922 return zone->present_pages;
926 static inline int zone_to_nid(struct zone *zone)
931 static inline void zone_set_nid(struct zone *zone, int nid)
936 static inline int zone_to_nid(struct zone *zone)
941 static inline void zone_set_nid(struct zone *zone, int nid) {}
944 extern int movable_zone;
946 #ifdef CONFIG_HIGHMEM
947 static inline int zone_movable_is_highmem(void)
949 #ifdef CONFIG_NEED_MULTIPLE_NODES
950 return movable_zone == ZONE_HIGHMEM;
952 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM;
957 static inline int is_highmem_idx(enum zone_type idx)
959 #ifdef CONFIG_HIGHMEM
960 return (idx == ZONE_HIGHMEM ||
961 (idx == ZONE_MOVABLE && zone_movable_is_highmem()));
968 * is_highmem - helper function to quickly check if a struct zone is a
969 * highmem zone or not. This is an attempt to keep references
970 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
971 * @zone - pointer to struct zone variable
973 static inline int is_highmem(struct zone *zone)
975 #ifdef CONFIG_HIGHMEM
976 return is_highmem_idx(zone_idx(zone));
982 /* These two functions are used to setup the per zone pages min values */
985 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
987 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
989 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
990 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
992 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int,
993 void *, size_t *, loff_t *);
994 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
995 void *, size_t *, loff_t *);
996 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
997 void *, size_t *, loff_t *);
998 int numa_zonelist_order_handler(struct ctl_table *, int,
999 void *, size_t *, loff_t *);
1000 extern int percpu_pagelist_fraction;
1001 extern char numa_zonelist_order[];
1002 #define NUMA_ZONELIST_ORDER_LEN 16
1004 #ifndef CONFIG_NEED_MULTIPLE_NODES
1006 extern struct pglist_data contig_page_data;
1007 #define NODE_DATA(nid) (&contig_page_data)
1008 #define NODE_MEM_MAP(nid) mem_map
1010 #else /* CONFIG_NEED_MULTIPLE_NODES */
1012 #include <asm/mmzone.h>
1014 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1016 extern struct pglist_data *first_online_pgdat(void);
1017 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1018 extern struct zone *next_zone(struct zone *zone);
1021 * for_each_online_pgdat - helper macro to iterate over all online nodes
1022 * @pgdat - pointer to a pg_data_t variable
1024 #define for_each_online_pgdat(pgdat) \
1025 for (pgdat = first_online_pgdat(); \
1027 pgdat = next_online_pgdat(pgdat))
1029 * for_each_zone - helper macro to iterate over all memory zones
1030 * @zone - pointer to struct zone variable
1032 * The user only needs to declare the zone variable, for_each_zone
1035 #define for_each_zone(zone) \
1036 for (zone = (first_online_pgdat())->node_zones; \
1038 zone = next_zone(zone))
1040 #define for_each_populated_zone(zone) \
1041 for (zone = (first_online_pgdat())->node_zones; \
1043 zone = next_zone(zone)) \
1044 if (!populated_zone(zone)) \
1045 ; /* do nothing */ \
1048 static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1050 return zoneref->zone;
1053 static inline int zonelist_zone_idx(struct zoneref *zoneref)
1055 return zoneref->zone_idx;
1058 static inline int zonelist_node_idx(struct zoneref *zoneref)
1060 return zone_to_nid(zoneref->zone);
1063 struct zoneref *__next_zones_zonelist(struct zoneref *z,
1064 enum zone_type highest_zoneidx,
1068 * next_zones_zonelist - Returns the next zone at or below highest_zoneidx within the allowed nodemask using a cursor within a zonelist as a starting point
1069 * @z - The cursor used as a starting point for the search
1070 * @highest_zoneidx - The zone index of the highest zone to return
1071 * @nodes - An optional nodemask to filter the zonelist with
1073 * This function returns the next zone at or below a given zone index that is
1074 * within the allowed nodemask using a cursor as the starting point for the
1075 * search. The zoneref returned is a cursor that represents the current zone
1076 * being examined. It should be advanced by one before calling
1077 * next_zones_zonelist again.
1079 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1080 enum zone_type highest_zoneidx,
1083 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1085 return __next_zones_zonelist(z, highest_zoneidx, nodes);
1089 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1090 * @zonelist - The zonelist to search for a suitable zone
1091 * @highest_zoneidx - The zone index of the highest zone to return
1092 * @nodes - An optional nodemask to filter the zonelist with
1093 * @return - Zoneref pointer for the first suitable zone found (see below)
1095 * This function returns the first zone at or below a given zone index that is
1096 * within the allowed nodemask. The zoneref returned is a cursor that can be
1097 * used to iterate the zonelist with next_zones_zonelist by advancing it by
1098 * one before calling.
1100 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1101 * never NULL). This may happen either genuinely, or due to concurrent nodemask
1102 * update due to cpuset modification.
1104 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1105 enum zone_type highest_zoneidx,
1108 return next_zones_zonelist(zonelist->_zonerefs,
1109 highest_zoneidx, nodes);
1113 * for_each_zone_zonelist_nodemask - helper macro to iterate over valid zones in a zonelist at or below a given zone index and within a nodemask
1114 * @zone - The current zone in the iterator
1115 * @z - The current pointer within zonelist->_zonerefs being iterated
1116 * @zlist - The zonelist being iterated
1117 * @highidx - The zone index of the highest zone to return
1118 * @nodemask - Nodemask allowed by the allocator
1120 * This iterator iterates though all zones at or below a given zone index and
1121 * within a given nodemask
1123 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1124 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1126 z = next_zones_zonelist(++z, highidx, nodemask), \
1127 zone = zonelist_zone(z))
1129 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1130 for (zone = z->zone; \
1132 z = next_zones_zonelist(++z, highidx, nodemask), \
1133 zone = zonelist_zone(z))
1137 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1138 * @zone - The current zone in the iterator
1139 * @z - The current pointer within zonelist->zones being iterated
1140 * @zlist - The zonelist being iterated
1141 * @highidx - The zone index of the highest zone to return
1143 * This iterator iterates though all zones at or below a given zone index.
1145 #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1146 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1148 #ifdef CONFIG_SPARSEMEM
1149 #include <asm/sparsemem.h>
1152 #ifdef CONFIG_FLATMEM
1153 #define pfn_to_nid(pfn) (0)
1156 #ifdef CONFIG_SPARSEMEM
1159 * SECTION_SHIFT #bits space required to store a section #
1161 * PA_SECTION_SHIFT physical address to/from section number
1162 * PFN_SECTION_SHIFT pfn to/from section number
1164 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1165 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1167 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1169 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1170 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1172 #define SECTION_BLOCKFLAGS_BITS \
1173 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1175 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1176 #error Allocator MAX_ORDER exceeds SECTION_SIZE
1179 static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1181 return pfn >> PFN_SECTION_SHIFT;
1183 static inline unsigned long section_nr_to_pfn(unsigned long sec)
1185 return sec << PFN_SECTION_SHIFT;
1188 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1189 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1191 #define SUBSECTION_SHIFT 21
1192 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1194 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1195 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1196 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1198 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1199 #error Subsection size exceeds section size
1201 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1204 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1205 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1207 struct mem_section_usage {
1208 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1209 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1211 /* See declaration of similar field in struct zone */
1212 unsigned long pageblock_flags[0];
1215 void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1219 struct mem_section {
1221 * This is, logically, a pointer to an array of struct
1222 * pages. However, it is stored with some other magic.
1223 * (see sparse.c::sparse_init_one_section())
1225 * Additionally during early boot we encode node id of
1226 * the location of the section here to guide allocation.
1227 * (see sparse.c::memory_present())
1229 * Making it a UL at least makes someone do a cast
1230 * before using it wrong.
1232 unsigned long section_mem_map;
1234 struct mem_section_usage *usage;
1235 #ifdef CONFIG_PAGE_EXTENSION
1237 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1238 * section. (see page_ext.h about this.)
1240 struct page_ext *page_ext;
1244 * WARNING: mem_section must be a power-of-2 in size for the
1245 * calculation and use of SECTION_ROOT_MASK to make sense.
1249 #ifdef CONFIG_SPARSEMEM_EXTREME
1250 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1252 #define SECTIONS_PER_ROOT 1
1255 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1256 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1257 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1259 #ifdef CONFIG_SPARSEMEM_EXTREME
1260 extern struct mem_section **mem_section;
1262 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1265 static inline unsigned long *section_to_usemap(struct mem_section *ms)
1267 return ms->usage->pageblock_flags;
1270 static inline struct mem_section *__nr_to_section(unsigned long nr)
1272 #ifdef CONFIG_SPARSEMEM_EXTREME
1276 if (!mem_section[SECTION_NR_TO_ROOT(nr)])
1278 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
1280 extern unsigned long __section_nr(struct mem_section *ms);
1281 extern size_t mem_section_usage_size(void);
1284 * We use the lower bits of the mem_map pointer to store
1285 * a little bit of information. The pointer is calculated
1286 * as mem_map - section_nr_to_pfn(pnum). The result is
1287 * aligned to the minimum alignment of the two values:
1288 * 1. All mem_map arrays are page-aligned.
1289 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1290 * lowest bits. PFN_SECTION_SHIFT is arch-specific
1291 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1292 * worst combination is powerpc with 256k pages,
1293 * which results in PFN_SECTION_SHIFT equal 6.
1294 * To sum it up, at least 6 bits are available.
1296 #define SECTION_MARKED_PRESENT (1UL<<0)
1297 #define SECTION_HAS_MEM_MAP (1UL<<1)
1298 #define SECTION_IS_ONLINE (1UL<<2)
1299 #define SECTION_IS_EARLY (1UL<<3)
1300 #define SECTION_MAP_LAST_BIT (1UL<<4)
1301 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1302 #define SECTION_NID_SHIFT 3
1304 static inline struct page *__section_mem_map_addr(struct mem_section *section)
1306 unsigned long map = section->section_mem_map;
1307 map &= SECTION_MAP_MASK;
1308 return (struct page *)map;
1311 static inline int present_section(struct mem_section *section)
1313 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1316 static inline int present_section_nr(unsigned long nr)
1318 return present_section(__nr_to_section(nr));
1321 static inline int valid_section(struct mem_section *section)
1323 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1326 static inline int early_section(struct mem_section *section)
1328 return (section && (section->section_mem_map & SECTION_IS_EARLY));
1331 static inline int valid_section_nr(unsigned long nr)
1333 return valid_section(__nr_to_section(nr));
1336 static inline int online_section(struct mem_section *section)
1338 return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1341 static inline int online_section_nr(unsigned long nr)
1343 return online_section(__nr_to_section(nr));
1346 #ifdef CONFIG_MEMORY_HOTPLUG
1347 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1348 #ifdef CONFIG_MEMORY_HOTREMOVE
1349 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1353 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1355 return __nr_to_section(pfn_to_section_nr(pfn));
1358 extern unsigned long __highest_present_section_nr;
1360 static inline int subsection_map_index(unsigned long pfn)
1362 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1365 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1366 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1368 int idx = subsection_map_index(pfn);
1370 return test_bit(idx, ms->usage->subsection_map);
1373 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1379 #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1380 static inline int pfn_valid(unsigned long pfn)
1382 struct mem_section *ms;
1384 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1386 ms = __nr_to_section(pfn_to_section_nr(pfn));
1387 if (!valid_section(ms))
1390 * Traditionally early sections always returned pfn_valid() for
1391 * the entire section-sized span.
1393 return early_section(ms) || pfn_section_valid(ms, pfn);
1397 static inline int pfn_in_present_section(unsigned long pfn)
1399 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1401 return present_section(__nr_to_section(pfn_to_section_nr(pfn)));
1404 static inline unsigned long next_present_section_nr(unsigned long section_nr)
1406 while (++section_nr <= __highest_present_section_nr) {
1407 if (present_section_nr(section_nr))
1415 * These are _only_ used during initialisation, therefore they
1416 * can use __initdata ... They could have names to indicate
1420 #define pfn_to_nid(pfn) \
1422 unsigned long __pfn_to_nid_pfn = (pfn); \
1423 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1426 #define pfn_to_nid(pfn) (0)
1429 void sparse_init(void);
1431 #define sparse_init() do {} while (0)
1432 #define sparse_index_init(_sec, _nid) do {} while (0)
1433 #define pfn_in_present_section pfn_valid
1434 #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1435 #endif /* CONFIG_SPARSEMEM */
1438 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we
1439 * need to check pfn validity within that MAX_ORDER_NR_PAGES block.
1440 * pfn_valid_within() should be used in this case; we optimise this away
1441 * when we have no holes within a MAX_ORDER_NR_PAGES block.
1443 #ifdef CONFIG_HOLES_IN_ZONE
1444 #define pfn_valid_within(pfn) pfn_valid(pfn)
1446 #define pfn_valid_within(pfn) (1)
1449 #endif /* !__GENERATING_BOUNDS.H */
1450 #endif /* !__ASSEMBLY__ */
1451 #endif /* _LINUX_MMZONE_H */