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