1 #ifndef _LINUX_MMZONE_H 2 #define _LINUX_MMZONE_H 3 4 #ifndef __ASSEMBLY__ 5 #ifndef __GENERATING_BOUNDS_H 6 7 #include <linux/spinlock.h> 8 #include <linux/list.h> 9 #include <linux/wait.h> 10 #include <linux/bitops.h> 11 #include <linux/cache.h> 12 #include <linux/threads.h> 13 #include <linux/numa.h> 14 #include <linux/init.h> 15 #include <linux/seqlock.h> 16 #include <linux/nodemask.h> 17 #include <linux/pageblock-flags.h> 18 #include <linux/page-flags-layout.h> 19 #include <linux/atomic.h> 20 #include <asm/page.h> 21 22 /* Free memory management - zoned buddy allocator. */ 23 #ifndef CONFIG_FORCE_MAX_ZONEORDER 24 #define MAX_ORDER 11 25 #else 26 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER 27 #endif 28 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1)) 29 30 /* 31 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed 32 * costly to service. That is between allocation orders which should 33 * coalesce naturally under reasonable reclaim pressure and those which 34 * will not. 35 */ 36 #define PAGE_ALLOC_COSTLY_ORDER 3 37 38 enum { 39 MIGRATE_UNMOVABLE, 40 MIGRATE_MOVABLE, 41 MIGRATE_RECLAIMABLE, 42 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ 43 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES, 44 #ifdef CONFIG_CMA 45 /* 46 * MIGRATE_CMA migration type is designed to mimic the way 47 * ZONE_MOVABLE works. Only movable pages can be allocated 48 * from MIGRATE_CMA pageblocks and page allocator never 49 * implicitly change migration type of MIGRATE_CMA pageblock. 50 * 51 * The way to use it is to change migratetype of a range of 52 * pageblocks to MIGRATE_CMA which can be done by 53 * __free_pageblock_cma() function. What is important though 54 * is that a range of pageblocks must be aligned to 55 * MAX_ORDER_NR_PAGES should biggest page be bigger then 56 * a single pageblock. 57 */ 58 MIGRATE_CMA, 59 #endif 60 #ifdef CONFIG_MEMORY_ISOLATION 61 MIGRATE_ISOLATE, /* can't allocate from here */ 62 #endif 63 MIGRATE_TYPES 64 }; 65 66 #ifdef CONFIG_CMA 67 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) 68 #else 69 # define is_migrate_cma(migratetype) false 70 #endif 71 72 #define for_each_migratetype_order(order, type) \ 73 for (order = 0; order < MAX_ORDER; order++) \ 74 for (type = 0; type < MIGRATE_TYPES; type++) 75 76 extern int page_group_by_mobility_disabled; 77 78 #define NR_MIGRATETYPE_BITS (PB_migrate_end - PB_migrate + 1) 79 #define MIGRATETYPE_MASK ((1UL << NR_MIGRATETYPE_BITS) - 1) 80 81 #define get_pageblock_migratetype(page) \ 82 get_pfnblock_flags_mask(page, page_to_pfn(page), \ 83 PB_migrate_end, MIGRATETYPE_MASK) 84 85 static inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn) 86 { 87 BUILD_BUG_ON(PB_migrate_end - PB_migrate != 2); 88 return get_pfnblock_flags_mask(page, pfn, PB_migrate_end, 89 MIGRATETYPE_MASK); 90 } 91 92 struct free_area { 93 struct list_head free_list[MIGRATE_TYPES]; 94 unsigned long nr_free; 95 }; 96 97 struct pglist_data; 98 99 /* 100 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel. 101 * So add a wild amount of padding here to ensure that they fall into separate 102 * cachelines. There are very few zone structures in the machine, so space 103 * consumption is not a concern here. 104 */ 105 #if defined(CONFIG_SMP) 106 struct zone_padding { 107 char x[0]; 108 } ____cacheline_internodealigned_in_smp; 109 #define ZONE_PADDING(name) struct zone_padding name; 110 #else 111 #define ZONE_PADDING(name) 112 #endif 113 114 enum zone_stat_item { 115 /* First 128 byte cacheline (assuming 64 bit words) */ 116 NR_FREE_PAGES, 117 NR_ALLOC_BATCH, 118 NR_LRU_BASE, 119 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ 120 NR_ACTIVE_ANON, /* " " " " " */ 121 NR_INACTIVE_FILE, /* " " " " " */ 122 NR_ACTIVE_FILE, /* " " " " " */ 123 NR_UNEVICTABLE, /* " " " " " */ 124 NR_MLOCK, /* mlock()ed pages found and moved off LRU */ 125 NR_ANON_PAGES, /* Mapped anonymous pages */ 126 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. 127 only modified from process context */ 128 NR_FILE_PAGES, 129 NR_FILE_DIRTY, 130 NR_WRITEBACK, 131 NR_SLAB_RECLAIMABLE, 132 NR_SLAB_UNRECLAIMABLE, 133 NR_PAGETABLE, /* used for pagetables */ 134 NR_KERNEL_STACK, 135 /* Second 128 byte cacheline */ 136 NR_UNSTABLE_NFS, /* NFS unstable pages */ 137 NR_BOUNCE, 138 NR_VMSCAN_WRITE, 139 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ 140 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ 141 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ 142 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ 143 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ 144 NR_DIRTIED, /* page dirtyings since bootup */ 145 NR_WRITTEN, /* page writings since bootup */ 146 NR_PAGES_SCANNED, /* pages scanned since last reclaim */ 147 #ifdef CONFIG_NUMA 148 NUMA_HIT, /* allocated in intended node */ 149 NUMA_MISS, /* allocated in non intended node */ 150 NUMA_FOREIGN, /* was intended here, hit elsewhere */ 151 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ 152 NUMA_LOCAL, /* allocation from local node */ 153 NUMA_OTHER, /* allocation from other node */ 154 #endif 155 WORKINGSET_REFAULT, 156 WORKINGSET_ACTIVATE, 157 WORKINGSET_NODERECLAIM, 158 NR_ANON_TRANSPARENT_HUGEPAGES, 159 NR_FREE_CMA_PAGES, 160 NR_VM_ZONE_STAT_ITEMS }; 161 162 /* 163 * We do arithmetic on the LRU lists in various places in the code, 164 * so it is important to keep the active lists LRU_ACTIVE higher in 165 * the array than the corresponding inactive lists, and to keep 166 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. 167 * 168 * This has to be kept in sync with the statistics in zone_stat_item 169 * above and the descriptions in vmstat_text in mm/vmstat.c 170 */ 171 #define LRU_BASE 0 172 #define LRU_ACTIVE 1 173 #define LRU_FILE 2 174 175 enum lru_list { 176 LRU_INACTIVE_ANON = LRU_BASE, 177 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, 178 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, 179 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, 180 LRU_UNEVICTABLE, 181 NR_LRU_LISTS 182 }; 183 184 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) 185 186 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) 187 188 static inline int is_file_lru(enum lru_list lru) 189 { 190 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); 191 } 192 193 static inline int is_active_lru(enum lru_list lru) 194 { 195 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); 196 } 197 198 static inline int is_unevictable_lru(enum lru_list lru) 199 { 200 return (lru == LRU_UNEVICTABLE); 201 } 202 203 struct zone_reclaim_stat { 204 /* 205 * The pageout code in vmscan.c keeps track of how many of the 206 * mem/swap backed and file backed pages are referenced. 207 * The higher the rotated/scanned ratio, the more valuable 208 * that cache is. 209 * 210 * The anon LRU stats live in [0], file LRU stats in [1] 211 */ 212 unsigned long recent_rotated[2]; 213 unsigned long recent_scanned[2]; 214 }; 215 216 struct lruvec { 217 struct list_head lists[NR_LRU_LISTS]; 218 struct zone_reclaim_stat reclaim_stat; 219 #ifdef CONFIG_MEMCG 220 struct zone *zone; 221 #endif 222 }; 223 224 /* Mask used at gathering information at once (see memcontrol.c) */ 225 #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) 226 #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) 227 #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) 228 229 /* Isolate clean file */ 230 #define ISOLATE_CLEAN ((__force isolate_mode_t)0x1) 231 /* Isolate unmapped file */ 232 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2) 233 /* Isolate for asynchronous migration */ 234 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) 235 /* Isolate unevictable pages */ 236 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) 237 238 /* LRU Isolation modes. */ 239 typedef unsigned __bitwise__ isolate_mode_t; 240 241 enum zone_watermarks { 242 WMARK_MIN, 243 WMARK_LOW, 244 WMARK_HIGH, 245 NR_WMARK 246 }; 247 248 #define min_wmark_pages(z) (z->watermark[WMARK_MIN]) 249 #define low_wmark_pages(z) (z->watermark[WMARK_LOW]) 250 #define high_wmark_pages(z) (z->watermark[WMARK_HIGH]) 251 252 struct per_cpu_pages { 253 int count; /* number of pages in the list */ 254 int high; /* high watermark, emptying needed */ 255 int batch; /* chunk size for buddy add/remove */ 256 257 /* Lists of pages, one per migrate type stored on the pcp-lists */ 258 struct list_head lists[MIGRATE_PCPTYPES]; 259 }; 260 261 struct per_cpu_pageset { 262 struct per_cpu_pages pcp; 263 #ifdef CONFIG_NUMA 264 s8 expire; 265 #endif 266 #ifdef CONFIG_SMP 267 s8 stat_threshold; 268 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; 269 #endif 270 }; 271 272 #endif /* !__GENERATING_BOUNDS.H */ 273 274 enum zone_type { 275 #ifdef CONFIG_ZONE_DMA 276 /* 277 * ZONE_DMA is used when there are devices that are not able 278 * to do DMA to all of addressable memory (ZONE_NORMAL). Then we 279 * carve out the portion of memory that is needed for these devices. 280 * The range is arch specific. 281 * 282 * Some examples 283 * 284 * Architecture Limit 285 * --------------------------- 286 * parisc, ia64, sparc <4G 287 * s390 <2G 288 * arm Various 289 * alpha Unlimited or 0-16MB. 290 * 291 * i386, x86_64 and multiple other arches 292 * <16M. 293 */ 294 ZONE_DMA, 295 #endif 296 #ifdef CONFIG_ZONE_DMA32 297 /* 298 * x86_64 needs two ZONE_DMAs because it supports devices that are 299 * only able to do DMA to the lower 16M but also 32 bit devices that 300 * can only do DMA areas below 4G. 301 */ 302 ZONE_DMA32, 303 #endif 304 /* 305 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be 306 * performed on pages in ZONE_NORMAL if the DMA devices support 307 * transfers to all addressable memory. 308 */ 309 ZONE_NORMAL, 310 #ifdef CONFIG_HIGHMEM 311 /* 312 * A memory area that is only addressable by the kernel through 313 * mapping portions into its own address space. This is for example 314 * used by i386 to allow the kernel to address the memory beyond 315 * 900MB. The kernel will set up special mappings (page 316 * table entries on i386) for each page that the kernel needs to 317 * access. 318 */ 319 ZONE_HIGHMEM, 320 #endif 321 ZONE_MOVABLE, 322 #ifdef CONFIG_ZONE_DEVICE 323 ZONE_DEVICE, 324 #endif 325 __MAX_NR_ZONES 326 327 }; 328 329 #ifndef __GENERATING_BOUNDS_H 330 331 struct zone { 332 /* Read-mostly fields */ 333 334 /* zone watermarks, access with *_wmark_pages(zone) macros */ 335 unsigned long watermark[NR_WMARK]; 336 337 unsigned long nr_reserved_highatomic; 338 339 /* 340 * We don't know if the memory that we're going to allocate will be 341 * freeable or/and it will be released eventually, so to avoid totally 342 * wasting several GB of ram we must reserve some of the lower zone 343 * memory (otherwise we risk to run OOM on the lower zones despite 344 * there being tons of freeable ram on the higher zones). This array is 345 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl 346 * changes. 347 */ 348 long lowmem_reserve[MAX_NR_ZONES]; 349 350 #ifdef CONFIG_NUMA 351 int node; 352 #endif 353 354 /* 355 * The target ratio of ACTIVE_ANON to INACTIVE_ANON pages on 356 * this zone's LRU. Maintained by the pageout code. 357 */ 358 unsigned int inactive_ratio; 359 360 struct pglist_data *zone_pgdat; 361 struct per_cpu_pageset __percpu *pageset; 362 363 /* 364 * This is a per-zone reserve of pages that should not be 365 * considered dirtyable memory. 366 */ 367 unsigned long dirty_balance_reserve; 368 369 #ifndef CONFIG_SPARSEMEM 370 /* 371 * Flags for a pageblock_nr_pages block. See pageblock-flags.h. 372 * In SPARSEMEM, this map is stored in struct mem_section 373 */ 374 unsigned long *pageblock_flags; 375 #endif /* CONFIG_SPARSEMEM */ 376 377 #ifdef CONFIG_NUMA 378 /* 379 * zone reclaim becomes active if more unmapped pages exist. 380 */ 381 unsigned long min_unmapped_pages; 382 unsigned long min_slab_pages; 383 #endif /* CONFIG_NUMA */ 384 385 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ 386 unsigned long zone_start_pfn; 387 388 /* 389 * spanned_pages is the total pages spanned by the zone, including 390 * holes, which is calculated as: 391 * spanned_pages = zone_end_pfn - zone_start_pfn; 392 * 393 * present_pages is physical pages existing within the zone, which 394 * is calculated as: 395 * present_pages = spanned_pages - absent_pages(pages in holes); 396 * 397 * managed_pages is present pages managed by the buddy system, which 398 * is calculated as (reserved_pages includes pages allocated by the 399 * bootmem allocator): 400 * managed_pages = present_pages - reserved_pages; 401 * 402 * So present_pages may be used by memory hotplug or memory power 403 * management logic to figure out unmanaged pages by checking 404 * (present_pages - managed_pages). And managed_pages should be used 405 * by page allocator and vm scanner to calculate all kinds of watermarks 406 * and thresholds. 407 * 408 * Locking rules: 409 * 410 * zone_start_pfn and spanned_pages are protected by span_seqlock. 411 * It is a seqlock because it has to be read outside of zone->lock, 412 * and it is done in the main allocator path. But, it is written 413 * quite infrequently. 414 * 415 * The span_seq lock is declared along with zone->lock because it is 416 * frequently read in proximity to zone->lock. It's good to 417 * give them a chance of being in the same cacheline. 418 * 419 * Write access to present_pages at runtime should be protected by 420 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of 421 * present_pages should get_online_mems() to get a stable value. 422 * 423 * Read access to managed_pages should be safe because it's unsigned 424 * long. Write access to zone->managed_pages and totalram_pages are 425 * protected by managed_page_count_lock at runtime. Idealy only 426 * adjust_managed_page_count() should be used instead of directly 427 * touching zone->managed_pages and totalram_pages. 428 */ 429 unsigned long managed_pages; 430 unsigned long spanned_pages; 431 unsigned long present_pages; 432 433 const char *name; 434 435 #ifdef CONFIG_MEMORY_ISOLATION 436 /* 437 * Number of isolated pageblock. It is used to solve incorrect 438 * freepage counting problem due to racy retrieving migratetype 439 * of pageblock. Protected by zone->lock. 440 */ 441 unsigned long nr_isolate_pageblock; 442 #endif 443 444 #ifdef CONFIG_MEMORY_HOTPLUG 445 /* see spanned/present_pages for more description */ 446 seqlock_t span_seqlock; 447 #endif 448 449 /* 450 * wait_table -- the array holding the hash table 451 * wait_table_hash_nr_entries -- the size of the hash table array 452 * wait_table_bits -- wait_table_size == (1 << wait_table_bits) 453 * 454 * The purpose of all these is to keep track of the people 455 * waiting for a page to become available and make them 456 * runnable again when possible. The trouble is that this 457 * consumes a lot of space, especially when so few things 458 * wait on pages at a given time. So instead of using 459 * per-page waitqueues, we use a waitqueue hash table. 460 * 461 * The bucket discipline is to sleep on the same queue when 462 * colliding and wake all in that wait queue when removing. 463 * When something wakes, it must check to be sure its page is 464 * truly available, a la thundering herd. The cost of a 465 * collision is great, but given the expected load of the 466 * table, they should be so rare as to be outweighed by the 467 * benefits from the saved space. 468 * 469 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the 470 * primary users of these fields, and in mm/page_alloc.c 471 * free_area_init_core() performs the initialization of them. 472 */ 473 wait_queue_head_t *wait_table; 474 unsigned long wait_table_hash_nr_entries; 475 unsigned long wait_table_bits; 476 477 ZONE_PADDING(_pad1_) 478 /* free areas of different sizes */ 479 struct free_area free_area[MAX_ORDER]; 480 481 /* zone flags, see below */ 482 unsigned long flags; 483 484 /* Write-intensive fields used from the page allocator */ 485 spinlock_t lock; 486 487 ZONE_PADDING(_pad2_) 488 489 /* Write-intensive fields used by page reclaim */ 490 491 /* Fields commonly accessed by the page reclaim scanner */ 492 spinlock_t lru_lock; 493 struct lruvec lruvec; 494 495 /* Evictions & activations on the inactive file list */ 496 atomic_long_t inactive_age; 497 498 /* 499 * When free pages are below this point, additional steps are taken 500 * when reading the number of free pages to avoid per-cpu counter 501 * drift allowing watermarks to be breached 502 */ 503 unsigned long percpu_drift_mark; 504 505 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 506 /* pfn where compaction free scanner should start */ 507 unsigned long compact_cached_free_pfn; 508 /* pfn where async and sync compaction migration scanner should start */ 509 unsigned long compact_cached_migrate_pfn[2]; 510 #endif 511 512 #ifdef CONFIG_COMPACTION 513 /* 514 * On compaction failure, 1<<compact_defer_shift compactions 515 * are skipped before trying again. The number attempted since 516 * last failure is tracked with compact_considered. 517 */ 518 unsigned int compact_considered; 519 unsigned int compact_defer_shift; 520 int compact_order_failed; 521 #endif 522 523 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 524 /* Set to true when the PG_migrate_skip bits should be cleared */ 525 bool compact_blockskip_flush; 526 #endif 527 528 ZONE_PADDING(_pad3_) 529 /* Zone statistics */ 530 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; 531 } ____cacheline_internodealigned_in_smp; 532 533 enum zone_flags { 534 ZONE_RECLAIM_LOCKED, /* prevents concurrent reclaim */ 535 ZONE_OOM_LOCKED, /* zone is in OOM killer zonelist */ 536 ZONE_CONGESTED, /* zone has many dirty pages backed by 537 * a congested BDI 538 */ 539 ZONE_DIRTY, /* reclaim scanning has recently found 540 * many dirty file pages at the tail 541 * of the LRU. 542 */ 543 ZONE_WRITEBACK, /* reclaim scanning has recently found 544 * many pages under writeback 545 */ 546 ZONE_FAIR_DEPLETED, /* fair zone policy batch depleted */ 547 }; 548 549 static inline unsigned long zone_end_pfn(const struct zone *zone) 550 { 551 return zone->zone_start_pfn + zone->spanned_pages; 552 } 553 554 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) 555 { 556 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); 557 } 558 559 static inline bool zone_is_initialized(struct zone *zone) 560 { 561 return !!zone->wait_table; 562 } 563 564 static inline bool zone_is_empty(struct zone *zone) 565 { 566 return zone->spanned_pages == 0; 567 } 568 569 /* 570 * The "priority" of VM scanning is how much of the queues we will scan in one 571 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the 572 * queues ("queue_length >> 12") during an aging round. 573 */ 574 #define DEF_PRIORITY 12 575 576 /* Maximum number of zones on a zonelist */ 577 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) 578 579 #ifdef CONFIG_NUMA 580 581 /* 582 * The NUMA zonelists are doubled because we need zonelists that restrict the 583 * allocations to a single node for __GFP_THISNODE. 584 * 585 * [0] : Zonelist with fallback 586 * [1] : No fallback (__GFP_THISNODE) 587 */ 588 #define MAX_ZONELISTS 2 589 #else 590 #define MAX_ZONELISTS 1 591 #endif 592 593 /* 594 * This struct contains information about a zone in a zonelist. It is stored 595 * here to avoid dereferences into large structures and lookups of tables 596 */ 597 struct zoneref { 598 struct zone *zone; /* Pointer to actual zone */ 599 int zone_idx; /* zone_idx(zoneref->zone) */ 600 }; 601 602 /* 603 * One allocation request operates on a zonelist. A zonelist 604 * is a list of zones, the first one is the 'goal' of the 605 * allocation, the other zones are fallback zones, in decreasing 606 * priority. 607 * 608 * To speed the reading of the zonelist, the zonerefs contain the zone index 609 * of the entry being read. Helper functions to access information given 610 * a struct zoneref are 611 * 612 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs 613 * zonelist_zone_idx() - Return the index of the zone for an entry 614 * zonelist_node_idx() - Return the index of the node for an entry 615 */ 616 struct zonelist { 617 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; 618 }; 619 620 #ifndef CONFIG_DISCONTIGMEM 621 /* The array of struct pages - for discontigmem use pgdat->lmem_map */ 622 extern struct page *mem_map; 623 #endif 624 625 /* 626 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM 627 * (mostly NUMA machines?) to denote a higher-level memory zone than the 628 * zone denotes. 629 * 630 * On NUMA machines, each NUMA node would have a pg_data_t to describe 631 * it's memory layout. 632 * 633 * Memory statistics and page replacement data structures are maintained on a 634 * per-zone basis. 635 */ 636 struct bootmem_data; 637 typedef struct pglist_data { 638 struct zone node_zones[MAX_NR_ZONES]; 639 struct zonelist node_zonelists[MAX_ZONELISTS]; 640 int nr_zones; 641 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */ 642 struct page *node_mem_map; 643 #ifdef CONFIG_PAGE_EXTENSION 644 struct page_ext *node_page_ext; 645 #endif 646 #endif 647 #ifndef CONFIG_NO_BOOTMEM 648 struct bootmem_data *bdata; 649 #endif 650 #ifdef CONFIG_MEMORY_HOTPLUG 651 /* 652 * Must be held any time you expect node_start_pfn, node_present_pages 653 * or node_spanned_pages stay constant. Holding this will also 654 * guarantee that any pfn_valid() stays that way. 655 * 656 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to 657 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG. 658 * 659 * Nests above zone->lock and zone->span_seqlock 660 */ 661 spinlock_t node_size_lock; 662 #endif 663 unsigned long node_start_pfn; 664 unsigned long node_present_pages; /* total number of physical pages */ 665 unsigned long node_spanned_pages; /* total size of physical page 666 range, including holes */ 667 int node_id; 668 wait_queue_head_t kswapd_wait; 669 wait_queue_head_t pfmemalloc_wait; 670 struct task_struct *kswapd; /* Protected by 671 mem_hotplug_begin/end() */ 672 int kswapd_max_order; 673 enum zone_type classzone_idx; 674 #ifdef CONFIG_NUMA_BALANCING 675 /* Lock serializing the migrate rate limiting window */ 676 spinlock_t numabalancing_migrate_lock; 677 678 /* Rate limiting time interval */ 679 unsigned long numabalancing_migrate_next_window; 680 681 /* Number of pages migrated during the rate limiting time interval */ 682 unsigned long numabalancing_migrate_nr_pages; 683 #endif 684 685 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 686 /* 687 * If memory initialisation on large machines is deferred then this 688 * is the first PFN that needs to be initialised. 689 */ 690 unsigned long first_deferred_pfn; 691 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 692 } pg_data_t; 693 694 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) 695 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) 696 #ifdef CONFIG_FLAT_NODE_MEM_MAP 697 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) 698 #else 699 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) 700 #endif 701 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) 702 703 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) 704 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) 705 706 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) 707 { 708 return pgdat->node_start_pfn + pgdat->node_spanned_pages; 709 } 710 711 static inline bool pgdat_is_empty(pg_data_t *pgdat) 712 { 713 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages; 714 } 715 716 static inline int zone_id(const struct zone *zone) 717 { 718 struct pglist_data *pgdat = zone->zone_pgdat; 719 720 return zone - pgdat->node_zones; 721 } 722 723 #ifdef CONFIG_ZONE_DEVICE 724 static inline bool is_dev_zone(const struct zone *zone) 725 { 726 return zone_id(zone) == ZONE_DEVICE; 727 } 728 #else 729 static inline bool is_dev_zone(const struct zone *zone) 730 { 731 return false; 732 } 733 #endif 734 735 #include <linux/memory_hotplug.h> 736 737 extern struct mutex zonelists_mutex; 738 void build_all_zonelists(pg_data_t *pgdat, struct zone *zone); 739 void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx); 740 bool zone_watermark_ok(struct zone *z, unsigned int order, 741 unsigned long mark, int classzone_idx, int alloc_flags); 742 bool zone_watermark_ok_safe(struct zone *z, unsigned int order, 743 unsigned long mark, int classzone_idx); 744 enum memmap_context { 745 MEMMAP_EARLY, 746 MEMMAP_HOTPLUG, 747 }; 748 extern int init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, 749 unsigned long size); 750 751 extern void lruvec_init(struct lruvec *lruvec); 752 753 static inline struct zone *lruvec_zone(struct lruvec *lruvec) 754 { 755 #ifdef CONFIG_MEMCG 756 return lruvec->zone; 757 #else 758 return container_of(lruvec, struct zone, lruvec); 759 #endif 760 } 761 762 #ifdef CONFIG_HAVE_MEMORY_PRESENT 763 void memory_present(int nid, unsigned long start, unsigned long end); 764 #else 765 static inline void memory_present(int nid, unsigned long start, unsigned long end) {} 766 #endif 767 768 #ifdef CONFIG_HAVE_MEMORYLESS_NODES 769 int local_memory_node(int node_id); 770 #else 771 static inline int local_memory_node(int node_id) { return node_id; }; 772 #endif 773 774 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE 775 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); 776 #endif 777 778 /* 779 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. 780 */ 781 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) 782 783 static inline int populated_zone(struct zone *zone) 784 { 785 return (!!zone->present_pages); 786 } 787 788 extern int movable_zone; 789 790 #ifdef CONFIG_HIGHMEM 791 static inline int zone_movable_is_highmem(void) 792 { 793 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP 794 return movable_zone == ZONE_HIGHMEM; 795 #else 796 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM; 797 #endif 798 } 799 #endif 800 801 static inline int is_highmem_idx(enum zone_type idx) 802 { 803 #ifdef CONFIG_HIGHMEM 804 return (idx == ZONE_HIGHMEM || 805 (idx == ZONE_MOVABLE && zone_movable_is_highmem())); 806 #else 807 return 0; 808 #endif 809 } 810 811 /** 812 * is_highmem - helper function to quickly check if a struct zone is a 813 * highmem zone or not. This is an attempt to keep references 814 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. 815 * @zone - pointer to struct zone variable 816 */ 817 static inline int is_highmem(struct zone *zone) 818 { 819 #ifdef CONFIG_HIGHMEM 820 int zone_off = (char *)zone - (char *)zone->zone_pgdat->node_zones; 821 return zone_off == ZONE_HIGHMEM * sizeof(*zone) || 822 (zone_off == ZONE_MOVABLE * sizeof(*zone) && 823 zone_movable_is_highmem()); 824 #else 825 return 0; 826 #endif 827 } 828 829 /* These two functions are used to setup the per zone pages min values */ 830 struct ctl_table; 831 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, 832 void __user *, size_t *, loff_t *); 833 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; 834 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, 835 void __user *, size_t *, loff_t *); 836 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, 837 void __user *, size_t *, loff_t *); 838 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, 839 void __user *, size_t *, loff_t *); 840 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, 841 void __user *, size_t *, loff_t *); 842 843 extern int numa_zonelist_order_handler(struct ctl_table *, int, 844 void __user *, size_t *, loff_t *); 845 extern char numa_zonelist_order[]; 846 #define NUMA_ZONELIST_ORDER_LEN 16 /* string buffer size */ 847 848 #ifndef CONFIG_NEED_MULTIPLE_NODES 849 850 extern struct pglist_data contig_page_data; 851 #define NODE_DATA(nid) (&contig_page_data) 852 #define NODE_MEM_MAP(nid) mem_map 853 854 #else /* CONFIG_NEED_MULTIPLE_NODES */ 855 856 #include <asm/mmzone.h> 857 858 #endif /* !CONFIG_NEED_MULTIPLE_NODES */ 859 860 extern struct pglist_data *first_online_pgdat(void); 861 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); 862 extern struct zone *next_zone(struct zone *zone); 863 864 /** 865 * for_each_online_pgdat - helper macro to iterate over all online nodes 866 * @pgdat - pointer to a pg_data_t variable 867 */ 868 #define for_each_online_pgdat(pgdat) \ 869 for (pgdat = first_online_pgdat(); \ 870 pgdat; \ 871 pgdat = next_online_pgdat(pgdat)) 872 /** 873 * for_each_zone - helper macro to iterate over all memory zones 874 * @zone - pointer to struct zone variable 875 * 876 * The user only needs to declare the zone variable, for_each_zone 877 * fills it in. 878 */ 879 #define for_each_zone(zone) \ 880 for (zone = (first_online_pgdat())->node_zones; \ 881 zone; \ 882 zone = next_zone(zone)) 883 884 #define for_each_populated_zone(zone) \ 885 for (zone = (first_online_pgdat())->node_zones; \ 886 zone; \ 887 zone = next_zone(zone)) \ 888 if (!populated_zone(zone)) \ 889 ; /* do nothing */ \ 890 else 891 892 static inline struct zone *zonelist_zone(struct zoneref *zoneref) 893 { 894 return zoneref->zone; 895 } 896 897 static inline int zonelist_zone_idx(struct zoneref *zoneref) 898 { 899 return zoneref->zone_idx; 900 } 901 902 static inline int zonelist_node_idx(struct zoneref *zoneref) 903 { 904 #ifdef CONFIG_NUMA 905 /* zone_to_nid not available in this context */ 906 return zoneref->zone->node; 907 #else 908 return 0; 909 #endif /* CONFIG_NUMA */ 910 } 911 912 /** 913 * 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 914 * @z - The cursor used as a starting point for the search 915 * @highest_zoneidx - The zone index of the highest zone to return 916 * @nodes - An optional nodemask to filter the zonelist with 917 * 918 * This function returns the next zone at or below a given zone index that is 919 * within the allowed nodemask using a cursor as the starting point for the 920 * search. The zoneref returned is a cursor that represents the current zone 921 * being examined. It should be advanced by one before calling 922 * next_zones_zonelist again. 923 */ 924 struct zoneref *next_zones_zonelist(struct zoneref *z, 925 enum zone_type highest_zoneidx, 926 nodemask_t *nodes); 927 928 /** 929 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist 930 * @zonelist - The zonelist to search for a suitable zone 931 * @highest_zoneidx - The zone index of the highest zone to return 932 * @nodes - An optional nodemask to filter the zonelist with 933 * @zone - The first suitable zone found is returned via this parameter 934 * 935 * This function returns the first zone at or below a given zone index that is 936 * within the allowed nodemask. The zoneref returned is a cursor that can be 937 * used to iterate the zonelist with next_zones_zonelist by advancing it by 938 * one before calling. 939 */ 940 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, 941 enum zone_type highest_zoneidx, 942 nodemask_t *nodes, 943 struct zone **zone) 944 { 945 struct zoneref *z = next_zones_zonelist(zonelist->_zonerefs, 946 highest_zoneidx, nodes); 947 *zone = zonelist_zone(z); 948 return z; 949 } 950 951 /** 952 * 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 953 * @zone - The current zone in the iterator 954 * @z - The current pointer within zonelist->zones being iterated 955 * @zlist - The zonelist being iterated 956 * @highidx - The zone index of the highest zone to return 957 * @nodemask - Nodemask allowed by the allocator 958 * 959 * This iterator iterates though all zones at or below a given zone index and 960 * within a given nodemask 961 */ 962 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ 963 for (z = first_zones_zonelist(zlist, highidx, nodemask, &zone); \ 964 zone; \ 965 z = next_zones_zonelist(++z, highidx, nodemask), \ 966 zone = zonelist_zone(z)) \ 967 968 /** 969 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index 970 * @zone - The current zone in the iterator 971 * @z - The current pointer within zonelist->zones being iterated 972 * @zlist - The zonelist being iterated 973 * @highidx - The zone index of the highest zone to return 974 * 975 * This iterator iterates though all zones at or below a given zone index. 976 */ 977 #define for_each_zone_zonelist(zone, z, zlist, highidx) \ 978 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) 979 980 #ifdef CONFIG_SPARSEMEM 981 #include <asm/sparsemem.h> 982 #endif 983 984 #if !defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) && \ 985 !defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) 986 static inline unsigned long early_pfn_to_nid(unsigned long pfn) 987 { 988 return 0; 989 } 990 #endif 991 992 #ifdef CONFIG_FLATMEM 993 #define pfn_to_nid(pfn) (0) 994 #endif 995 996 #ifdef CONFIG_SPARSEMEM 997 998 /* 999 * SECTION_SHIFT #bits space required to store a section # 1000 * 1001 * PA_SECTION_SHIFT physical address to/from section number 1002 * PFN_SECTION_SHIFT pfn to/from section number 1003 */ 1004 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) 1005 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) 1006 1007 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) 1008 1009 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) 1010 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) 1011 1012 #define SECTION_BLOCKFLAGS_BITS \ 1013 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) 1014 1015 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS 1016 #error Allocator MAX_ORDER exceeds SECTION_SIZE 1017 #endif 1018 1019 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) 1020 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) 1021 1022 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) 1023 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) 1024 1025 struct page; 1026 struct page_ext; 1027 struct mem_section { 1028 /* 1029 * This is, logically, a pointer to an array of struct 1030 * pages. However, it is stored with some other magic. 1031 * (see sparse.c::sparse_init_one_section()) 1032 * 1033 * Additionally during early boot we encode node id of 1034 * the location of the section here to guide allocation. 1035 * (see sparse.c::memory_present()) 1036 * 1037 * Making it a UL at least makes someone do a cast 1038 * before using it wrong. 1039 */ 1040 unsigned long section_mem_map; 1041 1042 /* See declaration of similar field in struct zone */ 1043 unsigned long *pageblock_flags; 1044 #ifdef CONFIG_PAGE_EXTENSION 1045 /* 1046 * If !SPARSEMEM, pgdat doesn't have page_ext pointer. We use 1047 * section. (see page_ext.h about this.) 1048 */ 1049 struct page_ext *page_ext; 1050 unsigned long pad; 1051 #endif 1052 /* 1053 * WARNING: mem_section must be a power-of-2 in size for the 1054 * calculation and use of SECTION_ROOT_MASK to make sense. 1055 */ 1056 }; 1057 1058 #ifdef CONFIG_SPARSEMEM_EXTREME 1059 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) 1060 #else 1061 #define SECTIONS_PER_ROOT 1 1062 #endif 1063 1064 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) 1065 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) 1066 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) 1067 1068 #ifdef CONFIG_SPARSEMEM_EXTREME 1069 extern struct mem_section *mem_section[NR_SECTION_ROOTS]; 1070 #else 1071 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; 1072 #endif 1073 1074 static inline struct mem_section *__nr_to_section(unsigned long nr) 1075 { 1076 if (!mem_section[SECTION_NR_TO_ROOT(nr)]) 1077 return NULL; 1078 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; 1079 } 1080 extern int __section_nr(struct mem_section* ms); 1081 extern unsigned long usemap_size(void); 1082 1083 /* 1084 * We use the lower bits of the mem_map pointer to store 1085 * a little bit of information. There should be at least 1086 * 3 bits here due to 32-bit alignment. 1087 */ 1088 #define SECTION_MARKED_PRESENT (1UL<<0) 1089 #define SECTION_HAS_MEM_MAP (1UL<<1) 1090 #define SECTION_MAP_LAST_BIT (1UL<<2) 1091 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) 1092 #define SECTION_NID_SHIFT 2 1093 1094 static inline struct page *__section_mem_map_addr(struct mem_section *section) 1095 { 1096 unsigned long map = section->section_mem_map; 1097 map &= SECTION_MAP_MASK; 1098 return (struct page *)map; 1099 } 1100 1101 static inline int present_section(struct mem_section *section) 1102 { 1103 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); 1104 } 1105 1106 static inline int present_section_nr(unsigned long nr) 1107 { 1108 return present_section(__nr_to_section(nr)); 1109 } 1110 1111 static inline int valid_section(struct mem_section *section) 1112 { 1113 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); 1114 } 1115 1116 static inline int valid_section_nr(unsigned long nr) 1117 { 1118 return valid_section(__nr_to_section(nr)); 1119 } 1120 1121 static inline struct mem_section *__pfn_to_section(unsigned long pfn) 1122 { 1123 return __nr_to_section(pfn_to_section_nr(pfn)); 1124 } 1125 1126 #ifndef CONFIG_HAVE_ARCH_PFN_VALID 1127 static inline int pfn_valid(unsigned long pfn) 1128 { 1129 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 1130 return 0; 1131 return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); 1132 } 1133 #endif 1134 1135 static inline int pfn_present(unsigned long pfn) 1136 { 1137 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 1138 return 0; 1139 return present_section(__nr_to_section(pfn_to_section_nr(pfn))); 1140 } 1141 1142 /* 1143 * These are _only_ used during initialisation, therefore they 1144 * can use __initdata ... They could have names to indicate 1145 * this restriction. 1146 */ 1147 #ifdef CONFIG_NUMA 1148 #define pfn_to_nid(pfn) \ 1149 ({ \ 1150 unsigned long __pfn_to_nid_pfn = (pfn); \ 1151 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ 1152 }) 1153 #else 1154 #define pfn_to_nid(pfn) (0) 1155 #endif 1156 1157 #define early_pfn_valid(pfn) pfn_valid(pfn) 1158 void sparse_init(void); 1159 #else 1160 #define sparse_init() do {} while (0) 1161 #define sparse_index_init(_sec, _nid) do {} while (0) 1162 #endif /* CONFIG_SPARSEMEM */ 1163 1164 /* 1165 * During memory init memblocks map pfns to nids. The search is expensive and 1166 * this caches recent lookups. The implementation of __early_pfn_to_nid 1167 * may treat start/end as pfns or sections. 1168 */ 1169 struct mminit_pfnnid_cache { 1170 unsigned long last_start; 1171 unsigned long last_end; 1172 int last_nid; 1173 }; 1174 1175 #ifndef early_pfn_valid 1176 #define early_pfn_valid(pfn) (1) 1177 #endif 1178 1179 void memory_present(int nid, unsigned long start, unsigned long end); 1180 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); 1181 1182 /* 1183 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we 1184 * need to check pfn validility within that MAX_ORDER_NR_PAGES block. 1185 * pfn_valid_within() should be used in this case; we optimise this away 1186 * when we have no holes within a MAX_ORDER_NR_PAGES block. 1187 */ 1188 #ifdef CONFIG_HOLES_IN_ZONE 1189 #define pfn_valid_within(pfn) pfn_valid(pfn) 1190 #else 1191 #define pfn_valid_within(pfn) (1) 1192 #endif 1193 1194 #ifdef CONFIG_ARCH_HAS_HOLES_MEMORYMODEL 1195 /* 1196 * pfn_valid() is meant to be able to tell if a given PFN has valid memmap 1197 * associated with it or not. In FLATMEM, it is expected that holes always 1198 * have valid memmap as long as there is valid PFNs either side of the hole. 1199 * In SPARSEMEM, it is assumed that a valid section has a memmap for the 1200 * entire section. 1201 * 1202 * However, an ARM, and maybe other embedded architectures in the future 1203 * free memmap backing holes to save memory on the assumption the memmap is 1204 * never used. The page_zone linkages are then broken even though pfn_valid() 1205 * returns true. A walker of the full memmap must then do this additional 1206 * check to ensure the memmap they are looking at is sane by making sure 1207 * the zone and PFN linkages are still valid. This is expensive, but walkers 1208 * of the full memmap are extremely rare. 1209 */ 1210 int memmap_valid_within(unsigned long pfn, 1211 struct page *page, struct zone *zone); 1212 #else 1213 static inline int memmap_valid_within(unsigned long pfn, 1214 struct page *page, struct zone *zone) 1215 { 1216 return 1; 1217 } 1218 #endif /* CONFIG_ARCH_HAS_HOLES_MEMORYMODEL */ 1219 1220 #endif /* !__GENERATING_BOUNDS.H */ 1221 #endif /* !__ASSEMBLY__ */ 1222 #endif /* _LINUX_MMZONE_H */ 1223