Line data Source code
1 : /* SPDX-License-Identifier: GPL-2.0 */
2 : #ifndef _LINUX_MMZONE_H
3 : #define _LINUX_MMZONE_H
4 :
5 : #ifndef __ASSEMBLY__
6 : #ifndef __GENERATING_BOUNDS_H
7 :
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>
23 : #include <linux/local_lock.h>
24 : #include <asm/page.h>
25 :
26 : /* Free memory management - zoned buddy allocator. */
27 : #ifndef CONFIG_FORCE_MAX_ZONEORDER
28 : #define MAX_ORDER 11
29 : #else
30 : #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER
31 : #endif
32 : #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1))
33 :
34 : /*
35 : * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed
36 : * costly to service. That is between allocation orders which should
37 : * coalesce naturally under reasonable reclaim pressure and those which
38 : * will not.
39 : */
40 : #define PAGE_ALLOC_COSTLY_ORDER 3
41 :
42 : enum migratetype {
43 : MIGRATE_UNMOVABLE,
44 : MIGRATE_MOVABLE,
45 : MIGRATE_RECLAIMABLE,
46 : MIGRATE_PCPTYPES, /* the number of types on the pcp lists */
47 : MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES,
48 : #ifdef CONFIG_CMA
49 : /*
50 : * MIGRATE_CMA migration type is designed to mimic the way
51 : * ZONE_MOVABLE works. Only movable pages can be allocated
52 : * from MIGRATE_CMA pageblocks and page allocator never
53 : * implicitly change migration type of MIGRATE_CMA pageblock.
54 : *
55 : * The way to use it is to change migratetype of a range of
56 : * pageblocks to MIGRATE_CMA which can be done by
57 : * __free_pageblock_cma() function. What is important though
58 : * is that a range of pageblocks must be aligned to
59 : * MAX_ORDER_NR_PAGES should biggest page be bigger than
60 : * a single pageblock.
61 : */
62 : MIGRATE_CMA,
63 : #endif
64 : #ifdef CONFIG_MEMORY_ISOLATION
65 : MIGRATE_ISOLATE, /* can't allocate from here */
66 : #endif
67 : MIGRATE_TYPES
68 : };
69 :
70 : /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */
71 : extern const char * const migratetype_names[MIGRATE_TYPES];
72 :
73 : #ifdef CONFIG_CMA
74 : # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA)
75 : # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA)
76 : #else
77 : # define is_migrate_cma(migratetype) false
78 : # define is_migrate_cma_page(_page) false
79 : #endif
80 :
81 : static inline bool is_migrate_movable(int mt)
82 : {
83 0 : return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE;
84 : }
85 :
86 : /*
87 : * Check whether a migratetype can be merged with another migratetype.
88 : *
89 : * It is only mergeable when it can fall back to other migratetypes for
90 : * allocation. See fallbacks[MIGRATE_TYPES][3] in page_alloc.c.
91 : */
92 : static inline bool migratetype_is_mergeable(int mt)
93 : {
94 : return mt < MIGRATE_PCPTYPES;
95 : }
96 :
97 : #define for_each_migratetype_order(order, type) \
98 : for (order = 0; order < MAX_ORDER; order++) \
99 : for (type = 0; type < MIGRATE_TYPES; type++)
100 :
101 : extern int page_group_by_mobility_disabled;
102 :
103 : #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1)
104 :
105 : #define get_pageblock_migratetype(page) \
106 : get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
107 :
108 : struct free_area {
109 : struct list_head free_list[MIGRATE_TYPES];
110 : unsigned long nr_free;
111 : };
112 :
113 : static inline struct page *get_page_from_free_area(struct free_area *area,
114 : int migratetype)
115 : {
116 1413 : return list_first_entry_or_null(&area->free_list[migratetype],
117 : struct page, lru);
118 : }
119 :
120 : static inline bool free_area_empty(struct free_area *area, int migratetype)
121 : {
122 302 : return list_empty(&area->free_list[migratetype]);
123 : }
124 :
125 : struct pglist_data;
126 :
127 : /*
128 : * Add a wild amount of padding here to ensure data fall into separate
129 : * cachelines. There are very few zone structures in the machine, so space
130 : * consumption is not a concern here.
131 : */
132 : #if defined(CONFIG_SMP)
133 : struct zone_padding {
134 : char x[0];
135 : } ____cacheline_internodealigned_in_smp;
136 : #define ZONE_PADDING(name) struct zone_padding name;
137 : #else
138 : #define ZONE_PADDING(name)
139 : #endif
140 :
141 : #ifdef CONFIG_NUMA
142 : enum numa_stat_item {
143 : NUMA_HIT, /* allocated in intended node */
144 : NUMA_MISS, /* allocated in non intended node */
145 : NUMA_FOREIGN, /* was intended here, hit elsewhere */
146 : NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */
147 : NUMA_LOCAL, /* allocation from local node */
148 : NUMA_OTHER, /* allocation from other node */
149 : NR_VM_NUMA_EVENT_ITEMS
150 : };
151 : #else
152 : #define NR_VM_NUMA_EVENT_ITEMS 0
153 : #endif
154 :
155 : enum zone_stat_item {
156 : /* First 128 byte cacheline (assuming 64 bit words) */
157 : NR_FREE_PAGES,
158 : NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */
159 : NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE,
160 : NR_ZONE_ACTIVE_ANON,
161 : NR_ZONE_INACTIVE_FILE,
162 : NR_ZONE_ACTIVE_FILE,
163 : NR_ZONE_UNEVICTABLE,
164 : NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */
165 : NR_MLOCK, /* mlock()ed pages found and moved off LRU */
166 : /* Second 128 byte cacheline */
167 : NR_BOUNCE,
168 : #if IS_ENABLED(CONFIG_ZSMALLOC)
169 : NR_ZSPAGES, /* allocated in zsmalloc */
170 : #endif
171 : NR_FREE_CMA_PAGES,
172 : NR_VM_ZONE_STAT_ITEMS };
173 :
174 : enum node_stat_item {
175 : NR_LRU_BASE,
176 : NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */
177 : NR_ACTIVE_ANON, /* " " " " " */
178 : NR_INACTIVE_FILE, /* " " " " " */
179 : NR_ACTIVE_FILE, /* " " " " " */
180 : NR_UNEVICTABLE, /* " " " " " */
181 : NR_SLAB_RECLAIMABLE_B,
182 : NR_SLAB_UNRECLAIMABLE_B,
183 : NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */
184 : NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */
185 : WORKINGSET_NODES,
186 : WORKINGSET_REFAULT_BASE,
187 : WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE,
188 : WORKINGSET_REFAULT_FILE,
189 : WORKINGSET_ACTIVATE_BASE,
190 : WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE,
191 : WORKINGSET_ACTIVATE_FILE,
192 : WORKINGSET_RESTORE_BASE,
193 : WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE,
194 : WORKINGSET_RESTORE_FILE,
195 : WORKINGSET_NODERECLAIM,
196 : NR_ANON_MAPPED, /* Mapped anonymous pages */
197 : NR_FILE_MAPPED, /* pagecache pages mapped into pagetables.
198 : only modified from process context */
199 : NR_FILE_PAGES,
200 : NR_FILE_DIRTY,
201 : NR_WRITEBACK,
202 : NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
203 : NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */
204 : NR_SHMEM_THPS,
205 : NR_SHMEM_PMDMAPPED,
206 : NR_FILE_THPS,
207 : NR_FILE_PMDMAPPED,
208 : NR_ANON_THPS,
209 : NR_VMSCAN_WRITE,
210 : NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */
211 : NR_DIRTIED, /* page dirtyings since bootup */
212 : NR_WRITTEN, /* page writings since bootup */
213 : NR_THROTTLED_WRITTEN, /* NR_WRITTEN while reclaim throttled */
214 : NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */
215 : NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */
216 : NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */
217 : NR_KERNEL_STACK_KB, /* measured in KiB */
218 : #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK)
219 : NR_KERNEL_SCS_KB, /* measured in KiB */
220 : #endif
221 : NR_PAGETABLE, /* used for pagetables */
222 : #ifdef CONFIG_SWAP
223 : NR_SWAPCACHE,
224 : #endif
225 : #ifdef CONFIG_NUMA_BALANCING
226 : PGPROMOTE_SUCCESS, /* promote successfully */
227 : #endif
228 : NR_VM_NODE_STAT_ITEMS
229 : };
230 :
231 : /*
232 : * Returns true if the item should be printed in THPs (/proc/vmstat
233 : * currently prints number of anon, file and shmem THPs. But the item
234 : * is charged in pages).
235 : */
236 : static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
237 : {
238 : if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
239 : return false;
240 :
241 : return item == NR_ANON_THPS ||
242 : item == NR_FILE_THPS ||
243 : item == NR_SHMEM_THPS ||
244 : item == NR_SHMEM_PMDMAPPED ||
245 : item == NR_FILE_PMDMAPPED;
246 : }
247 :
248 : /*
249 : * Returns true if the value is measured in bytes (most vmstat values are
250 : * measured in pages). This defines the API part, the internal representation
251 : * might be different.
252 : */
253 : static __always_inline bool vmstat_item_in_bytes(int idx)
254 : {
255 : /*
256 : * Global and per-node slab counters track slab pages.
257 : * It's expected that changes are multiples of PAGE_SIZE.
258 : * Internally values are stored in pages.
259 : *
260 : * Per-memcg and per-lruvec counters track memory, consumed
261 : * by individual slab objects. These counters are actually
262 : * byte-precise.
263 : */
264 457 : return (idx == NR_SLAB_RECLAIMABLE_B ||
265 : idx == NR_SLAB_UNRECLAIMABLE_B);
266 : }
267 :
268 : /*
269 : * We do arithmetic on the LRU lists in various places in the code,
270 : * so it is important to keep the active lists LRU_ACTIVE higher in
271 : * the array than the corresponding inactive lists, and to keep
272 : * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists.
273 : *
274 : * This has to be kept in sync with the statistics in zone_stat_item
275 : * above and the descriptions in vmstat_text in mm/vmstat.c
276 : */
277 : #define LRU_BASE 0
278 : #define LRU_ACTIVE 1
279 : #define LRU_FILE 2
280 :
281 : enum lru_list {
282 : LRU_INACTIVE_ANON = LRU_BASE,
283 : LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE,
284 : LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE,
285 : LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE,
286 : LRU_UNEVICTABLE,
287 : NR_LRU_LISTS
288 : };
289 :
290 : enum vmscan_throttle_state {
291 : VMSCAN_THROTTLE_WRITEBACK,
292 : VMSCAN_THROTTLE_ISOLATED,
293 : VMSCAN_THROTTLE_NOPROGRESS,
294 : VMSCAN_THROTTLE_CONGESTED,
295 : NR_VMSCAN_THROTTLE,
296 : };
297 :
298 : #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++)
299 :
300 : #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++)
301 :
302 : static inline bool is_file_lru(enum lru_list lru)
303 : {
304 0 : return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE);
305 : }
306 :
307 : static inline bool is_active_lru(enum lru_list lru)
308 : {
309 0 : return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE);
310 : }
311 :
312 : #define ANON_AND_FILE 2
313 :
314 : enum lruvec_flags {
315 : LRUVEC_CONGESTED, /* lruvec has many dirty pages
316 : * backed by a congested BDI
317 : */
318 : };
319 :
320 : struct lruvec {
321 : struct list_head lists[NR_LRU_LISTS];
322 : /* per lruvec lru_lock for memcg */
323 : spinlock_t lru_lock;
324 : /*
325 : * These track the cost of reclaiming one LRU - file or anon -
326 : * over the other. As the observed cost of reclaiming one LRU
327 : * increases, the reclaim scan balance tips toward the other.
328 : */
329 : unsigned long anon_cost;
330 : unsigned long file_cost;
331 : /* Non-resident age, driven by LRU movement */
332 : atomic_long_t nonresident_age;
333 : /* Refaults at the time of last reclaim cycle */
334 : unsigned long refaults[ANON_AND_FILE];
335 : /* Various lruvec state flags (enum lruvec_flags) */
336 : unsigned long flags;
337 : #ifdef CONFIG_MEMCG
338 : struct pglist_data *pgdat;
339 : #endif
340 : };
341 :
342 : /* Isolate unmapped pages */
343 : #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2)
344 : /* Isolate for asynchronous migration */
345 : #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4)
346 : /* Isolate unevictable pages */
347 : #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8)
348 :
349 : /* LRU Isolation modes. */
350 : typedef unsigned __bitwise isolate_mode_t;
351 :
352 : enum zone_watermarks {
353 : WMARK_MIN,
354 : WMARK_LOW,
355 : WMARK_HIGH,
356 : WMARK_PROMO,
357 : NR_WMARK
358 : };
359 :
360 : /*
361 : * One per migratetype for each PAGE_ALLOC_COSTLY_ORDER plus one additional
362 : * for pageblock size for THP if configured.
363 : */
364 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
365 : #define NR_PCP_THP 1
366 : #else
367 : #define NR_PCP_THP 0
368 : #endif
369 : #define NR_PCP_LISTS (MIGRATE_PCPTYPES * (PAGE_ALLOC_COSTLY_ORDER + 1 + NR_PCP_THP))
370 :
371 : /*
372 : * Shift to encode migratetype and order in the same integer, with order
373 : * in the least significant bits.
374 : */
375 : #define NR_PCP_ORDER_WIDTH 8
376 : #define NR_PCP_ORDER_MASK ((1<<NR_PCP_ORDER_WIDTH) - 1)
377 :
378 : #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost)
379 : #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost)
380 : #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost)
381 : #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost)
382 :
383 : /* Fields and list protected by pagesets local_lock in page_alloc.c */
384 : struct per_cpu_pages {
385 : int count; /* number of pages in the list */
386 : int high; /* high watermark, emptying needed */
387 : int batch; /* chunk size for buddy add/remove */
388 : short free_factor; /* batch scaling factor during free */
389 : #ifdef CONFIG_NUMA
390 : short expire; /* When 0, remote pagesets are drained */
391 : #endif
392 :
393 : /* Lists of pages, one per migrate type stored on the pcp-lists */
394 : struct list_head lists[NR_PCP_LISTS];
395 : };
396 :
397 : struct per_cpu_zonestat {
398 : #ifdef CONFIG_SMP
399 : s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS];
400 : s8 stat_threshold;
401 : #endif
402 : #ifdef CONFIG_NUMA
403 : /*
404 : * Low priority inaccurate counters that are only folded
405 : * on demand. Use a large type to avoid the overhead of
406 : * folding during refresh_cpu_vm_stats.
407 : */
408 : unsigned long vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
409 : #endif
410 : };
411 :
412 : struct per_cpu_nodestat {
413 : s8 stat_threshold;
414 : s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS];
415 : };
416 :
417 : #endif /* !__GENERATING_BOUNDS.H */
418 :
419 : enum zone_type {
420 : /*
421 : * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able
422 : * to DMA to all of the addressable memory (ZONE_NORMAL).
423 : * On architectures where this area covers the whole 32 bit address
424 : * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller
425 : * DMA addressing constraints. This distinction is important as a 32bit
426 : * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit
427 : * platforms may need both zones as they support peripherals with
428 : * different DMA addressing limitations.
429 : */
430 : #ifdef CONFIG_ZONE_DMA
431 : ZONE_DMA,
432 : #endif
433 : #ifdef CONFIG_ZONE_DMA32
434 : ZONE_DMA32,
435 : #endif
436 : /*
437 : * Normal addressable memory is in ZONE_NORMAL. DMA operations can be
438 : * performed on pages in ZONE_NORMAL if the DMA devices support
439 : * transfers to all addressable memory.
440 : */
441 : ZONE_NORMAL,
442 : #ifdef CONFIG_HIGHMEM
443 : /*
444 : * A memory area that is only addressable by the kernel through
445 : * mapping portions into its own address space. This is for example
446 : * used by i386 to allow the kernel to address the memory beyond
447 : * 900MB. The kernel will set up special mappings (page
448 : * table entries on i386) for each page that the kernel needs to
449 : * access.
450 : */
451 : ZONE_HIGHMEM,
452 : #endif
453 : /*
454 : * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains
455 : * movable pages with few exceptional cases described below. Main use
456 : * cases for ZONE_MOVABLE are to make memory offlining/unplug more
457 : * likely to succeed, and to locally limit unmovable allocations - e.g.,
458 : * to increase the number of THP/huge pages. Notable special cases are:
459 : *
460 : * 1. Pinned pages: (long-term) pinning of movable pages might
461 : * essentially turn such pages unmovable. Therefore, we do not allow
462 : * pinning long-term pages in ZONE_MOVABLE. When pages are pinned and
463 : * faulted, they come from the right zone right away. However, it is
464 : * still possible that address space already has pages in
465 : * ZONE_MOVABLE at the time when pages are pinned (i.e. user has
466 : * touches that memory before pinning). In such case we migrate them
467 : * to a different zone. When migration fails - pinning fails.
468 : * 2. memblock allocations: kernelcore/movablecore setups might create
469 : * situations where ZONE_MOVABLE contains unmovable allocations
470 : * after boot. Memory offlining and allocations fail early.
471 : * 3. Memory holes: kernelcore/movablecore setups might create very rare
472 : * situations where ZONE_MOVABLE contains memory holes after boot,
473 : * for example, if we have sections that are only partially
474 : * populated. Memory offlining and allocations fail early.
475 : * 4. PG_hwpoison pages: while poisoned pages can be skipped during
476 : * memory offlining, such pages cannot be allocated.
477 : * 5. Unmovable PG_offline pages: in paravirtualized environments,
478 : * hotplugged memory blocks might only partially be managed by the
479 : * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The
480 : * parts not manged by the buddy are unmovable PG_offline pages. In
481 : * some cases (virtio-mem), such pages can be skipped during
482 : * memory offlining, however, cannot be moved/allocated. These
483 : * techniques might use alloc_contig_range() to hide previously
484 : * exposed pages from the buddy again (e.g., to implement some sort
485 : * of memory unplug in virtio-mem).
486 : * 6. ZERO_PAGE(0), kernelcore/movablecore setups might create
487 : * situations where ZERO_PAGE(0) which is allocated differently
488 : * on different platforms may end up in a movable zone. ZERO_PAGE(0)
489 : * cannot be migrated.
490 : * 7. Memory-hotplug: when using memmap_on_memory and onlining the
491 : * memory to the MOVABLE zone, the vmemmap pages are also placed in
492 : * such zone. Such pages cannot be really moved around as they are
493 : * self-stored in the range, but they are treated as movable when
494 : * the range they describe is about to be offlined.
495 : *
496 : * In general, no unmovable allocations that degrade memory offlining
497 : * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range())
498 : * have to expect that migrating pages in ZONE_MOVABLE can fail (even
499 : * if has_unmovable_pages() states that there are no unmovable pages,
500 : * there can be false negatives).
501 : */
502 : ZONE_MOVABLE,
503 : #ifdef CONFIG_ZONE_DEVICE
504 : ZONE_DEVICE,
505 : #endif
506 : __MAX_NR_ZONES
507 :
508 : };
509 :
510 : #ifndef __GENERATING_BOUNDS_H
511 :
512 : #define ASYNC_AND_SYNC 2
513 :
514 : struct zone {
515 : /* Read-mostly fields */
516 :
517 : /* zone watermarks, access with *_wmark_pages(zone) macros */
518 : unsigned long _watermark[NR_WMARK];
519 : unsigned long watermark_boost;
520 :
521 : unsigned long nr_reserved_highatomic;
522 :
523 : /*
524 : * We don't know if the memory that we're going to allocate will be
525 : * freeable or/and it will be released eventually, so to avoid totally
526 : * wasting several GB of ram we must reserve some of the lower zone
527 : * memory (otherwise we risk to run OOM on the lower zones despite
528 : * there being tons of freeable ram on the higher zones). This array is
529 : * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl
530 : * changes.
531 : */
532 : long lowmem_reserve[MAX_NR_ZONES];
533 :
534 : #ifdef CONFIG_NUMA
535 : int node;
536 : #endif
537 : struct pglist_data *zone_pgdat;
538 : struct per_cpu_pages __percpu *per_cpu_pageset;
539 : struct per_cpu_zonestat __percpu *per_cpu_zonestats;
540 : /*
541 : * the high and batch values are copied to individual pagesets for
542 : * faster access
543 : */
544 : int pageset_high;
545 : int pageset_batch;
546 :
547 : #ifndef CONFIG_SPARSEMEM
548 : /*
549 : * Flags for a pageblock_nr_pages block. See pageblock-flags.h.
550 : * In SPARSEMEM, this map is stored in struct mem_section
551 : */
552 : unsigned long *pageblock_flags;
553 : #endif /* CONFIG_SPARSEMEM */
554 :
555 : /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */
556 : unsigned long zone_start_pfn;
557 :
558 : /*
559 : * spanned_pages is the total pages spanned by the zone, including
560 : * holes, which is calculated as:
561 : * spanned_pages = zone_end_pfn - zone_start_pfn;
562 : *
563 : * present_pages is physical pages existing within the zone, which
564 : * is calculated as:
565 : * present_pages = spanned_pages - absent_pages(pages in holes);
566 : *
567 : * present_early_pages is present pages existing within the zone
568 : * located on memory available since early boot, excluding hotplugged
569 : * memory.
570 : *
571 : * managed_pages is present pages managed by the buddy system, which
572 : * is calculated as (reserved_pages includes pages allocated by the
573 : * bootmem allocator):
574 : * managed_pages = present_pages - reserved_pages;
575 : *
576 : * cma pages is present pages that are assigned for CMA use
577 : * (MIGRATE_CMA).
578 : *
579 : * So present_pages may be used by memory hotplug or memory power
580 : * management logic to figure out unmanaged pages by checking
581 : * (present_pages - managed_pages). And managed_pages should be used
582 : * by page allocator and vm scanner to calculate all kinds of watermarks
583 : * and thresholds.
584 : *
585 : * Locking rules:
586 : *
587 : * zone_start_pfn and spanned_pages are protected by span_seqlock.
588 : * It is a seqlock because it has to be read outside of zone->lock,
589 : * and it is done in the main allocator path. But, it is written
590 : * quite infrequently.
591 : *
592 : * The span_seq lock is declared along with zone->lock because it is
593 : * frequently read in proximity to zone->lock. It's good to
594 : * give them a chance of being in the same cacheline.
595 : *
596 : * Write access to present_pages at runtime should be protected by
597 : * mem_hotplug_begin/end(). Any reader who can't tolerant drift of
598 : * present_pages should get_online_mems() to get a stable value.
599 : */
600 : atomic_long_t managed_pages;
601 : unsigned long spanned_pages;
602 : unsigned long present_pages;
603 : #if defined(CONFIG_MEMORY_HOTPLUG)
604 : unsigned long present_early_pages;
605 : #endif
606 : #ifdef CONFIG_CMA
607 : unsigned long cma_pages;
608 : #endif
609 :
610 : const char *name;
611 :
612 : #ifdef CONFIG_MEMORY_ISOLATION
613 : /*
614 : * Number of isolated pageblock. It is used to solve incorrect
615 : * freepage counting problem due to racy retrieving migratetype
616 : * of pageblock. Protected by zone->lock.
617 : */
618 : unsigned long nr_isolate_pageblock;
619 : #endif
620 :
621 : #ifdef CONFIG_MEMORY_HOTPLUG
622 : /* see spanned/present_pages for more description */
623 : seqlock_t span_seqlock;
624 : #endif
625 :
626 : int initialized;
627 :
628 : /* Write-intensive fields used from the page allocator */
629 : ZONE_PADDING(_pad1_)
630 :
631 : /* free areas of different sizes */
632 : struct free_area free_area[MAX_ORDER];
633 :
634 : /* zone flags, see below */
635 : unsigned long flags;
636 :
637 : /* Primarily protects free_area */
638 : spinlock_t lock;
639 :
640 : /* Write-intensive fields used by compaction and vmstats. */
641 : ZONE_PADDING(_pad2_)
642 :
643 : /*
644 : * When free pages are below this point, additional steps are taken
645 : * when reading the number of free pages to avoid per-cpu counter
646 : * drift allowing watermarks to be breached
647 : */
648 : unsigned long percpu_drift_mark;
649 :
650 : #if defined CONFIG_COMPACTION || defined CONFIG_CMA
651 : /* pfn where compaction free scanner should start */
652 : unsigned long compact_cached_free_pfn;
653 : /* pfn where compaction migration scanner should start */
654 : unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC];
655 : unsigned long compact_init_migrate_pfn;
656 : unsigned long compact_init_free_pfn;
657 : #endif
658 :
659 : #ifdef CONFIG_COMPACTION
660 : /*
661 : * On compaction failure, 1<<compact_defer_shift compactions
662 : * are skipped before trying again. The number attempted since
663 : * last failure is tracked with compact_considered.
664 : * compact_order_failed is the minimum compaction failed order.
665 : */
666 : unsigned int compact_considered;
667 : unsigned int compact_defer_shift;
668 : int compact_order_failed;
669 : #endif
670 :
671 : #if defined CONFIG_COMPACTION || defined CONFIG_CMA
672 : /* Set to true when the PG_migrate_skip bits should be cleared */
673 : bool compact_blockskip_flush;
674 : #endif
675 :
676 : bool contiguous;
677 :
678 : ZONE_PADDING(_pad3_)
679 : /* Zone statistics */
680 : atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS];
681 : atomic_long_t vm_numa_event[NR_VM_NUMA_EVENT_ITEMS];
682 : } ____cacheline_internodealigned_in_smp;
683 :
684 : enum pgdat_flags {
685 : PGDAT_DIRTY, /* reclaim scanning has recently found
686 : * many dirty file pages at the tail
687 : * of the LRU.
688 : */
689 : PGDAT_WRITEBACK, /* reclaim scanning has recently found
690 : * many pages under writeback
691 : */
692 : PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */
693 : };
694 :
695 : enum zone_flags {
696 : ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks.
697 : * Cleared when kswapd is woken.
698 : */
699 : ZONE_RECLAIM_ACTIVE, /* kswapd may be scanning the zone. */
700 : };
701 :
702 : static inline unsigned long zone_managed_pages(struct zone *zone)
703 : {
704 42 : return (unsigned long)atomic_long_read(&zone->managed_pages);
705 : }
706 :
707 : static inline unsigned long zone_cma_pages(struct zone *zone)
708 : {
709 : #ifdef CONFIG_CMA
710 : return zone->cma_pages;
711 : #else
712 : return 0;
713 : #endif
714 : }
715 :
716 : static inline unsigned long zone_end_pfn(const struct zone *zone)
717 : {
718 261 : return zone->zone_start_pfn + zone->spanned_pages;
719 : }
720 :
721 : static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn)
722 : {
723 0 : return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone);
724 : }
725 :
726 : static inline bool zone_is_initialized(struct zone *zone)
727 : {
728 : return zone->initialized;
729 : }
730 :
731 : static inline bool zone_is_empty(struct zone *zone)
732 : {
733 : return zone->spanned_pages == 0;
734 : }
735 :
736 : /*
737 : * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty
738 : * intersection with the given zone
739 : */
740 : static inline bool zone_intersects(struct zone *zone,
741 : unsigned long start_pfn, unsigned long nr_pages)
742 : {
743 : if (zone_is_empty(zone))
744 : return false;
745 : if (start_pfn >= zone_end_pfn(zone) ||
746 : start_pfn + nr_pages <= zone->zone_start_pfn)
747 : return false;
748 :
749 : return true;
750 : }
751 :
752 : /*
753 : * The "priority" of VM scanning is how much of the queues we will scan in one
754 : * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the
755 : * queues ("queue_length >> 12") during an aging round.
756 : */
757 : #define DEF_PRIORITY 12
758 :
759 : /* Maximum number of zones on a zonelist */
760 : #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES)
761 :
762 : enum {
763 : ZONELIST_FALLBACK, /* zonelist with fallback */
764 : #ifdef CONFIG_NUMA
765 : /*
766 : * The NUMA zonelists are doubled because we need zonelists that
767 : * restrict the allocations to a single node for __GFP_THISNODE.
768 : */
769 : ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */
770 : #endif
771 : MAX_ZONELISTS
772 : };
773 :
774 : /*
775 : * This struct contains information about a zone in a zonelist. It is stored
776 : * here to avoid dereferences into large structures and lookups of tables
777 : */
778 : struct zoneref {
779 : struct zone *zone; /* Pointer to actual zone */
780 : int zone_idx; /* zone_idx(zoneref->zone) */
781 : };
782 :
783 : /*
784 : * One allocation request operates on a zonelist. A zonelist
785 : * is a list of zones, the first one is the 'goal' of the
786 : * allocation, the other zones are fallback zones, in decreasing
787 : * priority.
788 : *
789 : * To speed the reading of the zonelist, the zonerefs contain the zone index
790 : * of the entry being read. Helper functions to access information given
791 : * a struct zoneref are
792 : *
793 : * zonelist_zone() - Return the struct zone * for an entry in _zonerefs
794 : * zonelist_zone_idx() - Return the index of the zone for an entry
795 : * zonelist_node_idx() - Return the index of the node for an entry
796 : */
797 : struct zonelist {
798 : struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1];
799 : };
800 :
801 : /*
802 : * The array of struct pages for flatmem.
803 : * It must be declared for SPARSEMEM as well because there are configurations
804 : * that rely on that.
805 : */
806 : extern struct page *mem_map;
807 :
808 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
809 : struct deferred_split {
810 : spinlock_t split_queue_lock;
811 : struct list_head split_queue;
812 : unsigned long split_queue_len;
813 : };
814 : #endif
815 :
816 : /*
817 : * On NUMA machines, each NUMA node would have a pg_data_t to describe
818 : * it's memory layout. On UMA machines there is a single pglist_data which
819 : * describes the whole memory.
820 : *
821 : * Memory statistics and page replacement data structures are maintained on a
822 : * per-zone basis.
823 : */
824 : typedef struct pglist_data {
825 : /*
826 : * node_zones contains just the zones for THIS node. Not all of the
827 : * zones may be populated, but it is the full list. It is referenced by
828 : * this node's node_zonelists as well as other node's node_zonelists.
829 : */
830 : struct zone node_zones[MAX_NR_ZONES];
831 :
832 : /*
833 : * node_zonelists contains references to all zones in all nodes.
834 : * Generally the first zones will be references to this node's
835 : * node_zones.
836 : */
837 : struct zonelist node_zonelists[MAX_ZONELISTS];
838 :
839 : int nr_zones; /* number of populated zones in this node */
840 : #ifdef CONFIG_FLATMEM /* means !SPARSEMEM */
841 : struct page *node_mem_map;
842 : #ifdef CONFIG_PAGE_EXTENSION
843 : struct page_ext *node_page_ext;
844 : #endif
845 : #endif
846 : #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
847 : /*
848 : * Must be held any time you expect node_start_pfn,
849 : * node_present_pages, node_spanned_pages or nr_zones to stay constant.
850 : * Also synchronizes pgdat->first_deferred_pfn during deferred page
851 : * init.
852 : *
853 : * pgdat_resize_lock() and pgdat_resize_unlock() are provided to
854 : * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG
855 : * or CONFIG_DEFERRED_STRUCT_PAGE_INIT.
856 : *
857 : * Nests above zone->lock and zone->span_seqlock
858 : */
859 : spinlock_t node_size_lock;
860 : #endif
861 : unsigned long node_start_pfn;
862 : unsigned long node_present_pages; /* total number of physical pages */
863 : unsigned long node_spanned_pages; /* total size of physical page
864 : range, including holes */
865 : int node_id;
866 : wait_queue_head_t kswapd_wait;
867 : wait_queue_head_t pfmemalloc_wait;
868 :
869 : /* workqueues for throttling reclaim for different reasons. */
870 : wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
871 :
872 : atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
873 : unsigned long nr_reclaim_start; /* nr pages written while throttled
874 : * when throttling started. */
875 : struct task_struct *kswapd; /* Protected by
876 : mem_hotplug_begin/end() */
877 : int kswapd_order;
878 : enum zone_type kswapd_highest_zoneidx;
879 :
880 : int kswapd_failures; /* Number of 'reclaimed == 0' runs */
881 :
882 : #ifdef CONFIG_COMPACTION
883 : int kcompactd_max_order;
884 : enum zone_type kcompactd_highest_zoneidx;
885 : wait_queue_head_t kcompactd_wait;
886 : struct task_struct *kcompactd;
887 : bool proactive_compact_trigger;
888 : #endif
889 : /*
890 : * This is a per-node reserve of pages that are not available
891 : * to userspace allocations.
892 : */
893 : unsigned long totalreserve_pages;
894 :
895 : #ifdef CONFIG_NUMA
896 : /*
897 : * node reclaim becomes active if more unmapped pages exist.
898 : */
899 : unsigned long min_unmapped_pages;
900 : unsigned long min_slab_pages;
901 : #endif /* CONFIG_NUMA */
902 :
903 : /* Write-intensive fields used by page reclaim */
904 : ZONE_PADDING(_pad1_)
905 :
906 : #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
907 : /*
908 : * If memory initialisation on large machines is deferred then this
909 : * is the first PFN that needs to be initialised.
910 : */
911 : unsigned long first_deferred_pfn;
912 : #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
913 :
914 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
915 : struct deferred_split deferred_split_queue;
916 : #endif
917 :
918 : /* Fields commonly accessed by the page reclaim scanner */
919 :
920 : /*
921 : * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED.
922 : *
923 : * Use mem_cgroup_lruvec() to look up lruvecs.
924 : */
925 : struct lruvec __lruvec;
926 :
927 : unsigned long flags;
928 :
929 : ZONE_PADDING(_pad2_)
930 :
931 : /* Per-node vmstats */
932 : struct per_cpu_nodestat __percpu *per_cpu_nodestats;
933 : atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS];
934 : } pg_data_t;
935 :
936 : #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages)
937 : #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages)
938 :
939 : #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn)
940 : #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid))
941 :
942 : static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat)
943 : {
944 1 : return pgdat->node_start_pfn + pgdat->node_spanned_pages;
945 : }
946 :
947 : static inline bool pgdat_is_empty(pg_data_t *pgdat)
948 : {
949 : return !pgdat->node_start_pfn && !pgdat->node_spanned_pages;
950 : }
951 :
952 : #include <linux/memory_hotplug.h>
953 :
954 : void build_all_zonelists(pg_data_t *pgdat);
955 : void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
956 : enum zone_type highest_zoneidx);
957 : bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
958 : int highest_zoneidx, unsigned int alloc_flags,
959 : long free_pages);
960 : bool zone_watermark_ok(struct zone *z, unsigned int order,
961 : unsigned long mark, int highest_zoneidx,
962 : unsigned int alloc_flags);
963 : bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
964 : unsigned long mark, int highest_zoneidx);
965 : /*
966 : * Memory initialization context, use to differentiate memory added by
967 : * the platform statically or via memory hotplug interface.
968 : */
969 : enum meminit_context {
970 : MEMINIT_EARLY,
971 : MEMINIT_HOTPLUG,
972 : };
973 :
974 : extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn,
975 : unsigned long size);
976 :
977 : extern void lruvec_init(struct lruvec *lruvec);
978 :
979 : static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec)
980 : {
981 : #ifdef CONFIG_MEMCG
982 : return lruvec->pgdat;
983 : #else
984 0 : return container_of(lruvec, struct pglist_data, __lruvec);
985 : #endif
986 : }
987 :
988 : #ifdef CONFIG_HAVE_MEMORYLESS_NODES
989 : int local_memory_node(int node_id);
990 : #else
991 : static inline int local_memory_node(int node_id) { return node_id; };
992 : #endif
993 :
994 : /*
995 : * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc.
996 : */
997 : #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones)
998 :
999 : #ifdef CONFIG_ZONE_DEVICE
1000 : static inline bool zone_is_zone_device(struct zone *zone)
1001 : {
1002 : return zone_idx(zone) == ZONE_DEVICE;
1003 : }
1004 : #else
1005 : static inline bool zone_is_zone_device(struct zone *zone)
1006 : {
1007 : return false;
1008 : }
1009 : #endif
1010 :
1011 : /*
1012 : * Returns true if a zone has pages managed by the buddy allocator.
1013 : * All the reclaim decisions have to use this function rather than
1014 : * populated_zone(). If the whole zone is reserved then we can easily
1015 : * end up with populated_zone() && !managed_zone().
1016 : */
1017 : static inline bool managed_zone(struct zone *zone)
1018 : {
1019 1 : return zone_managed_pages(zone);
1020 : }
1021 :
1022 : /* Returns true if a zone has memory */
1023 : static inline bool populated_zone(struct zone *zone)
1024 : {
1025 : return zone->present_pages;
1026 : }
1027 :
1028 : #ifdef CONFIG_NUMA
1029 : static inline int zone_to_nid(struct zone *zone)
1030 : {
1031 : return zone->node;
1032 : }
1033 :
1034 : static inline void zone_set_nid(struct zone *zone, int nid)
1035 : {
1036 : zone->node = nid;
1037 : }
1038 : #else
1039 : static inline int zone_to_nid(struct zone *zone)
1040 : {
1041 : return 0;
1042 : }
1043 :
1044 : static inline void zone_set_nid(struct zone *zone, int nid) {}
1045 : #endif
1046 :
1047 : extern int movable_zone;
1048 :
1049 : static inline int is_highmem_idx(enum zone_type idx)
1050 : {
1051 : #ifdef CONFIG_HIGHMEM
1052 : return (idx == ZONE_HIGHMEM ||
1053 : (idx == ZONE_MOVABLE && movable_zone == ZONE_HIGHMEM));
1054 : #else
1055 : return 0;
1056 : #endif
1057 : }
1058 :
1059 : #ifdef CONFIG_ZONE_DMA
1060 : bool has_managed_dma(void);
1061 : #else
1062 : static inline bool has_managed_dma(void)
1063 : {
1064 : return false;
1065 : }
1066 : #endif
1067 :
1068 : /**
1069 : * is_highmem - helper function to quickly check if a struct zone is a
1070 : * highmem zone or not. This is an attempt to keep references
1071 : * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum.
1072 : * @zone: pointer to struct zone variable
1073 : * Return: 1 for a highmem zone, 0 otherwise
1074 : */
1075 : static inline int is_highmem(struct zone *zone)
1076 : {
1077 : #ifdef CONFIG_HIGHMEM
1078 : return is_highmem_idx(zone_idx(zone));
1079 : #else
1080 : return 0;
1081 : #endif
1082 : }
1083 :
1084 : /* These two functions are used to setup the per zone pages min values */
1085 : struct ctl_table;
1086 :
1087 : int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *,
1088 : loff_t *);
1089 : int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *,
1090 : size_t *, loff_t *);
1091 : extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES];
1092 : int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *,
1093 : size_t *, loff_t *);
1094 : int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *, int,
1095 : void *, size_t *, loff_t *);
1096 : int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int,
1097 : void *, size_t *, loff_t *);
1098 : int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int,
1099 : void *, size_t *, loff_t *);
1100 : int numa_zonelist_order_handler(struct ctl_table *, int,
1101 : void *, size_t *, loff_t *);
1102 : extern int percpu_pagelist_high_fraction;
1103 : extern char numa_zonelist_order[];
1104 : #define NUMA_ZONELIST_ORDER_LEN 16
1105 :
1106 : #ifndef CONFIG_NUMA
1107 :
1108 : extern struct pglist_data contig_page_data;
1109 : static inline struct pglist_data *NODE_DATA(int nid)
1110 : {
1111 : return &contig_page_data;
1112 : }
1113 :
1114 : #else /* CONFIG_NUMA */
1115 :
1116 : #include <asm/mmzone.h>
1117 :
1118 : #endif /* !CONFIG_NUMA */
1119 :
1120 : extern struct pglist_data *first_online_pgdat(void);
1121 : extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat);
1122 : extern struct zone *next_zone(struct zone *zone);
1123 :
1124 : /**
1125 : * for_each_online_pgdat - helper macro to iterate over all online nodes
1126 : * @pgdat: pointer to a pg_data_t variable
1127 : */
1128 : #define for_each_online_pgdat(pgdat) \
1129 : for (pgdat = first_online_pgdat(); \
1130 : pgdat; \
1131 : pgdat = next_online_pgdat(pgdat))
1132 : /**
1133 : * for_each_zone - helper macro to iterate over all memory zones
1134 : * @zone: pointer to struct zone variable
1135 : *
1136 : * The user only needs to declare the zone variable, for_each_zone
1137 : * fills it in.
1138 : */
1139 : #define for_each_zone(zone) \
1140 : for (zone = (first_online_pgdat())->node_zones; \
1141 : zone; \
1142 : zone = next_zone(zone))
1143 :
1144 : #define for_each_populated_zone(zone) \
1145 : for (zone = (first_online_pgdat())->node_zones; \
1146 : zone; \
1147 : zone = next_zone(zone)) \
1148 : if (!populated_zone(zone)) \
1149 : ; /* do nothing */ \
1150 : else
1151 :
1152 : static inline struct zone *zonelist_zone(struct zoneref *zoneref)
1153 : {
1154 : return zoneref->zone;
1155 : }
1156 :
1157 : static inline int zonelist_zone_idx(struct zoneref *zoneref)
1158 : {
1159 : return zoneref->zone_idx;
1160 : }
1161 :
1162 : static inline int zonelist_node_idx(struct zoneref *zoneref)
1163 : {
1164 : return zone_to_nid(zoneref->zone);
1165 : }
1166 :
1167 : struct zoneref *__next_zones_zonelist(struct zoneref *z,
1168 : enum zone_type highest_zoneidx,
1169 : nodemask_t *nodes);
1170 :
1171 : /**
1172 : * 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
1173 : * @z: The cursor used as a starting point for the search
1174 : * @highest_zoneidx: The zone index of the highest zone to return
1175 : * @nodes: An optional nodemask to filter the zonelist with
1176 : *
1177 : * This function returns the next zone at or below a given zone index that is
1178 : * within the allowed nodemask using a cursor as the starting point for the
1179 : * search. The zoneref returned is a cursor that represents the current zone
1180 : * being examined. It should be advanced by one before calling
1181 : * next_zones_zonelist again.
1182 : *
1183 : * Return: the next zone at or below highest_zoneidx within the allowed
1184 : * nodemask using a cursor within a zonelist as a starting point
1185 : */
1186 : static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z,
1187 : enum zone_type highest_zoneidx,
1188 : nodemask_t *nodes)
1189 : {
1190 504 : if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx))
1191 : return z;
1192 0 : return __next_zones_zonelist(z, highest_zoneidx, nodes);
1193 : }
1194 :
1195 : /**
1196 : * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist
1197 : * @zonelist: The zonelist to search for a suitable zone
1198 : * @highest_zoneidx: The zone index of the highest zone to return
1199 : * @nodes: An optional nodemask to filter the zonelist with
1200 : *
1201 : * This function returns the first zone at or below a given zone index that is
1202 : * within the allowed nodemask. The zoneref returned is a cursor that can be
1203 : * used to iterate the zonelist with next_zones_zonelist by advancing it by
1204 : * one before calling.
1205 : *
1206 : * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is
1207 : * never NULL). This may happen either genuinely, or due to concurrent nodemask
1208 : * update due to cpuset modification.
1209 : *
1210 : * Return: Zoneref pointer for the first suitable zone found
1211 : */
1212 : static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist,
1213 : enum zone_type highest_zoneidx,
1214 : nodemask_t *nodes)
1215 : {
1216 1002 : return next_zones_zonelist(zonelist->_zonerefs,
1217 : highest_zoneidx, nodes);
1218 : }
1219 :
1220 : /**
1221 : * 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
1222 : * @zone: The current zone in the iterator
1223 : * @z: The current pointer within zonelist->_zonerefs being iterated
1224 : * @zlist: The zonelist being iterated
1225 : * @highidx: The zone index of the highest zone to return
1226 : * @nodemask: Nodemask allowed by the allocator
1227 : *
1228 : * This iterator iterates though all zones at or below a given zone index and
1229 : * within a given nodemask
1230 : */
1231 : #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \
1232 : for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
1233 : zone; \
1234 : z = next_zones_zonelist(++z, highidx, nodemask), \
1235 : zone = zonelist_zone(z))
1236 :
1237 : #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \
1238 : for (zone = z->zone; \
1239 : zone; \
1240 : z = next_zones_zonelist(++z, highidx, nodemask), \
1241 : zone = zonelist_zone(z))
1242 :
1243 :
1244 : /**
1245 : * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index
1246 : * @zone: The current zone in the iterator
1247 : * @z: The current pointer within zonelist->zones being iterated
1248 : * @zlist: The zonelist being iterated
1249 : * @highidx: The zone index of the highest zone to return
1250 : *
1251 : * This iterator iterates though all zones at or below a given zone index.
1252 : */
1253 : #define for_each_zone_zonelist(zone, z, zlist, highidx) \
1254 : for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
1255 :
1256 : /* Whether the 'nodes' are all movable nodes */
1257 : static inline bool movable_only_nodes(nodemask_t *nodes)
1258 : {
1259 : struct zonelist *zonelist;
1260 : struct zoneref *z;
1261 : int nid;
1262 :
1263 : if (nodes_empty(*nodes))
1264 : return false;
1265 :
1266 : /*
1267 : * We can chose arbitrary node from the nodemask to get a
1268 : * zonelist as they are interlinked. We just need to find
1269 : * at least one zone that can satisfy kernel allocations.
1270 : */
1271 : nid = first_node(*nodes);
1272 : zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
1273 : z = first_zones_zonelist(zonelist, ZONE_NORMAL, nodes);
1274 : return (!z->zone) ? true : false;
1275 : }
1276 :
1277 :
1278 : #ifdef CONFIG_SPARSEMEM
1279 : #include <asm/sparsemem.h>
1280 : #endif
1281 :
1282 : #ifdef CONFIG_FLATMEM
1283 : #define pfn_to_nid(pfn) (0)
1284 : #endif
1285 :
1286 : #ifdef CONFIG_SPARSEMEM
1287 :
1288 : /*
1289 : * PA_SECTION_SHIFT physical address to/from section number
1290 : * PFN_SECTION_SHIFT pfn to/from section number
1291 : */
1292 : #define PA_SECTION_SHIFT (SECTION_SIZE_BITS)
1293 : #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT)
1294 :
1295 : #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT)
1296 :
1297 : #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT)
1298 : #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1))
1299 :
1300 : #define SECTION_BLOCKFLAGS_BITS \
1301 : ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS)
1302 :
1303 : #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS
1304 : #error Allocator MAX_ORDER exceeds SECTION_SIZE
1305 : #endif
1306 :
1307 : static inline unsigned long pfn_to_section_nr(unsigned long pfn)
1308 : {
1309 : return pfn >> PFN_SECTION_SHIFT;
1310 : }
1311 : static inline unsigned long section_nr_to_pfn(unsigned long sec)
1312 : {
1313 : return sec << PFN_SECTION_SHIFT;
1314 : }
1315 :
1316 : #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
1317 : #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
1318 :
1319 : #define SUBSECTION_SHIFT 21
1320 : #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT)
1321 :
1322 : #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT)
1323 : #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT)
1324 : #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1))
1325 :
1326 : #if SUBSECTION_SHIFT > SECTION_SIZE_BITS
1327 : #error Subsection size exceeds section size
1328 : #else
1329 : #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT))
1330 : #endif
1331 :
1332 : #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION)
1333 : #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK)
1334 :
1335 : struct mem_section_usage {
1336 : #ifdef CONFIG_SPARSEMEM_VMEMMAP
1337 : DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION);
1338 : #endif
1339 : /* See declaration of similar field in struct zone */
1340 : unsigned long pageblock_flags[0];
1341 : };
1342 :
1343 : void subsection_map_init(unsigned long pfn, unsigned long nr_pages);
1344 :
1345 : struct page;
1346 : struct page_ext;
1347 : struct mem_section {
1348 : /*
1349 : * This is, logically, a pointer to an array of struct
1350 : * pages. However, it is stored with some other magic.
1351 : * (see sparse.c::sparse_init_one_section())
1352 : *
1353 : * Additionally during early boot we encode node id of
1354 : * the location of the section here to guide allocation.
1355 : * (see sparse.c::memory_present())
1356 : *
1357 : * Making it a UL at least makes someone do a cast
1358 : * before using it wrong.
1359 : */
1360 : unsigned long section_mem_map;
1361 :
1362 : struct mem_section_usage *usage;
1363 : #ifdef CONFIG_PAGE_EXTENSION
1364 : /*
1365 : * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use
1366 : * section. (see page_ext.h about this.)
1367 : */
1368 : struct page_ext *page_ext;
1369 : unsigned long pad;
1370 : #endif
1371 : /*
1372 : * WARNING: mem_section must be a power-of-2 in size for the
1373 : * calculation and use of SECTION_ROOT_MASK to make sense.
1374 : */
1375 : };
1376 :
1377 : #ifdef CONFIG_SPARSEMEM_EXTREME
1378 : #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section))
1379 : #else
1380 : #define SECTIONS_PER_ROOT 1
1381 : #endif
1382 :
1383 : #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT)
1384 : #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT)
1385 : #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1)
1386 :
1387 : #ifdef CONFIG_SPARSEMEM_EXTREME
1388 : extern struct mem_section **mem_section;
1389 : #else
1390 : extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
1391 : #endif
1392 :
1393 : static inline unsigned long *section_to_usemap(struct mem_section *ms)
1394 : {
1395 : return ms->usage->pageblock_flags;
1396 : }
1397 :
1398 : static inline struct mem_section *__nr_to_section(unsigned long nr)
1399 : {
1400 : unsigned long root = SECTION_NR_TO_ROOT(nr);
1401 :
1402 : if (unlikely(root >= NR_SECTION_ROOTS))
1403 : return NULL;
1404 :
1405 : #ifdef CONFIG_SPARSEMEM_EXTREME
1406 : if (!mem_section || !mem_section[root])
1407 : return NULL;
1408 : #endif
1409 : return &mem_section[root][nr & SECTION_ROOT_MASK];
1410 : }
1411 : extern size_t mem_section_usage_size(void);
1412 :
1413 : /*
1414 : * We use the lower bits of the mem_map pointer to store
1415 : * a little bit of information. The pointer is calculated
1416 : * as mem_map - section_nr_to_pfn(pnum). The result is
1417 : * aligned to the minimum alignment of the two values:
1418 : * 1. All mem_map arrays are page-aligned.
1419 : * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT
1420 : * lowest bits. PFN_SECTION_SHIFT is arch-specific
1421 : * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the
1422 : * worst combination is powerpc with 256k pages,
1423 : * which results in PFN_SECTION_SHIFT equal 6.
1424 : * To sum it up, at least 6 bits are available.
1425 : */
1426 : #define SECTION_MARKED_PRESENT (1UL<<0)
1427 : #define SECTION_HAS_MEM_MAP (1UL<<1)
1428 : #define SECTION_IS_ONLINE (1UL<<2)
1429 : #define SECTION_IS_EARLY (1UL<<3)
1430 : #define SECTION_TAINT_ZONE_DEVICE (1UL<<4)
1431 : #define SECTION_MAP_LAST_BIT (1UL<<5)
1432 : #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1))
1433 : #define SECTION_NID_SHIFT 6
1434 :
1435 : static inline struct page *__section_mem_map_addr(struct mem_section *section)
1436 : {
1437 : unsigned long map = section->section_mem_map;
1438 : map &= SECTION_MAP_MASK;
1439 : return (struct page *)map;
1440 : }
1441 :
1442 : static inline int present_section(struct mem_section *section)
1443 : {
1444 : return (section && (section->section_mem_map & SECTION_MARKED_PRESENT));
1445 : }
1446 :
1447 : static inline int present_section_nr(unsigned long nr)
1448 : {
1449 : return present_section(__nr_to_section(nr));
1450 : }
1451 :
1452 : static inline int valid_section(struct mem_section *section)
1453 : {
1454 : return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP));
1455 : }
1456 :
1457 : static inline int early_section(struct mem_section *section)
1458 : {
1459 : return (section && (section->section_mem_map & SECTION_IS_EARLY));
1460 : }
1461 :
1462 : static inline int valid_section_nr(unsigned long nr)
1463 : {
1464 : return valid_section(__nr_to_section(nr));
1465 : }
1466 :
1467 : static inline int online_section(struct mem_section *section)
1468 : {
1469 : return (section && (section->section_mem_map & SECTION_IS_ONLINE));
1470 : }
1471 :
1472 : static inline int online_device_section(struct mem_section *section)
1473 : {
1474 : unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE;
1475 :
1476 : return section && ((section->section_mem_map & flags) == flags);
1477 : }
1478 :
1479 : static inline int online_section_nr(unsigned long nr)
1480 : {
1481 : return online_section(__nr_to_section(nr));
1482 : }
1483 :
1484 : #ifdef CONFIG_MEMORY_HOTPLUG
1485 : void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1486 : void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn);
1487 : #endif
1488 :
1489 : static inline struct mem_section *__pfn_to_section(unsigned long pfn)
1490 : {
1491 : return __nr_to_section(pfn_to_section_nr(pfn));
1492 : }
1493 :
1494 : extern unsigned long __highest_present_section_nr;
1495 :
1496 : static inline int subsection_map_index(unsigned long pfn)
1497 : {
1498 : return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION;
1499 : }
1500 :
1501 : #ifdef CONFIG_SPARSEMEM_VMEMMAP
1502 : static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1503 : {
1504 : int idx = subsection_map_index(pfn);
1505 :
1506 : return test_bit(idx, ms->usage->subsection_map);
1507 : }
1508 : #else
1509 : static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn)
1510 : {
1511 : return 1;
1512 : }
1513 : #endif
1514 :
1515 : #ifndef CONFIG_HAVE_ARCH_PFN_VALID
1516 : /**
1517 : * pfn_valid - check if there is a valid memory map entry for a PFN
1518 : * @pfn: the page frame number to check
1519 : *
1520 : * Check if there is a valid memory map entry aka struct page for the @pfn.
1521 : * Note, that availability of the memory map entry does not imply that
1522 : * there is actual usable memory at that @pfn. The struct page may
1523 : * represent a hole or an unusable page frame.
1524 : *
1525 : * Return: 1 for PFNs that have memory map entries and 0 otherwise
1526 : */
1527 : static inline int pfn_valid(unsigned long pfn)
1528 : {
1529 : struct mem_section *ms;
1530 :
1531 : /*
1532 : * Ensure the upper PAGE_SHIFT bits are clear in the
1533 : * pfn. Else it might lead to false positives when
1534 : * some of the upper bits are set, but the lower bits
1535 : * match a valid pfn.
1536 : */
1537 : if (PHYS_PFN(PFN_PHYS(pfn)) != pfn)
1538 : return 0;
1539 :
1540 : if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1541 : return 0;
1542 : ms = __pfn_to_section(pfn);
1543 : if (!valid_section(ms))
1544 : return 0;
1545 : /*
1546 : * Traditionally early sections always returned pfn_valid() for
1547 : * the entire section-sized span.
1548 : */
1549 : return early_section(ms) || pfn_section_valid(ms, pfn);
1550 : }
1551 : #endif
1552 :
1553 : static inline int pfn_in_present_section(unsigned long pfn)
1554 : {
1555 : if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
1556 : return 0;
1557 : return present_section(__pfn_to_section(pfn));
1558 : }
1559 :
1560 : static inline unsigned long next_present_section_nr(unsigned long section_nr)
1561 : {
1562 : while (++section_nr <= __highest_present_section_nr) {
1563 : if (present_section_nr(section_nr))
1564 : return section_nr;
1565 : }
1566 :
1567 : return -1;
1568 : }
1569 :
1570 : /*
1571 : * These are _only_ used during initialisation, therefore they
1572 : * can use __initdata ... They could have names to indicate
1573 : * this restriction.
1574 : */
1575 : #ifdef CONFIG_NUMA
1576 : #define pfn_to_nid(pfn) \
1577 : ({ \
1578 : unsigned long __pfn_to_nid_pfn = (pfn); \
1579 : page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \
1580 : })
1581 : #else
1582 : #define pfn_to_nid(pfn) (0)
1583 : #endif
1584 :
1585 : void sparse_init(void);
1586 : #else
1587 : #define sparse_init() do {} while (0)
1588 : #define sparse_index_init(_sec, _nid) do {} while (0)
1589 : #define pfn_in_present_section pfn_valid
1590 : #define subsection_map_init(_pfn, _nr_pages) do {} while (0)
1591 : #endif /* CONFIG_SPARSEMEM */
1592 :
1593 : #endif /* !__GENERATING_BOUNDS.H */
1594 : #endif /* !__ASSEMBLY__ */
1595 : #endif /* _LINUX_MMZONE_H */
|
