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 <asm/page.h> 24 25 /* Free memory management - zoned buddy allocator. */ 26 #ifndef CONFIG_FORCE_MAX_ZONEORDER 27 #define MAX_ORDER 11 28 #else 29 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER 30 #endif 31 #define MAX_ORDER_NR_PAGES (1 << (MAX_ORDER - 1)) 32 33 /* 34 * PAGE_ALLOC_COSTLY_ORDER is the order at which allocations are deemed 35 * costly to service. That is between allocation orders which should 36 * coalesce naturally under reasonable reclaim pressure and those which 37 * will not. 38 */ 39 #define PAGE_ALLOC_COSTLY_ORDER 3 40 41 enum migratetype { 42 MIGRATE_UNMOVABLE, 43 MIGRATE_MOVABLE, 44 MIGRATE_RECLAIMABLE, 45 MIGRATE_PCPTYPES, /* the number of types on the pcp lists */ 46 MIGRATE_HIGHATOMIC = MIGRATE_PCPTYPES, 47 #ifdef CONFIG_CMA 48 /* 49 * MIGRATE_CMA migration type is designed to mimic the way 50 * ZONE_MOVABLE works. Only movable pages can be allocated 51 * from MIGRATE_CMA pageblocks and page allocator never 52 * implicitly change migration type of MIGRATE_CMA pageblock. 53 * 54 * The way to use it is to change migratetype of a range of 55 * pageblocks to MIGRATE_CMA which can be done by 56 * __free_pageblock_cma() function. What is important though 57 * is that a range of pageblocks must be aligned to 58 * MAX_ORDER_NR_PAGES should biggest page be bigger then 59 * a single pageblock. 60 */ 61 MIGRATE_CMA, 62 #endif 63 #ifdef CONFIG_MEMORY_ISOLATION 64 MIGRATE_ISOLATE, /* can't allocate from here */ 65 #endif 66 MIGRATE_TYPES 67 }; 68 69 /* In mm/page_alloc.c; keep in sync also with show_migration_types() there */ 70 extern const char * const migratetype_names[MIGRATE_TYPES]; 71 72 #ifdef CONFIG_CMA 73 # define is_migrate_cma(migratetype) unlikely((migratetype) == MIGRATE_CMA) 74 # define is_migrate_cma_page(_page) (get_pageblock_migratetype(_page) == MIGRATE_CMA) 75 #else 76 # define is_migrate_cma(migratetype) false 77 # define is_migrate_cma_page(_page) false 78 #endif 79 80 static inline bool is_migrate_movable(int mt) 81 { 82 return is_migrate_cma(mt) || mt == MIGRATE_MOVABLE; 83 } 84 85 #define for_each_migratetype_order(order, type) \ 86 for (order = 0; order < MAX_ORDER; order++) \ 87 for (type = 0; type < MIGRATE_TYPES; type++) 88 89 extern int page_group_by_mobility_disabled; 90 91 #define MIGRATETYPE_MASK ((1UL << PB_migratetype_bits) - 1) 92 93 #define get_pageblock_migratetype(page) \ 94 get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK) 95 96 struct free_area { 97 struct list_head free_list[MIGRATE_TYPES]; 98 unsigned long nr_free; 99 }; 100 101 static inline struct page *get_page_from_free_area(struct free_area *area, 102 int migratetype) 103 { 104 return list_first_entry_or_null(&area->free_list[migratetype], 105 struct page, lru); 106 } 107 108 static inline bool free_area_empty(struct free_area *area, int migratetype) 109 { 110 return list_empty(&area->free_list[migratetype]); 111 } 112 113 struct pglist_data; 114 115 /* 116 * Add a wild amount of padding here to ensure datas fall into separate 117 * cachelines. There are very few zone structures in the machine, so space 118 * consumption is not a concern here. 119 */ 120 #if defined(CONFIG_SMP) 121 struct zone_padding { 122 char x[0]; 123 } ____cacheline_internodealigned_in_smp; 124 #define ZONE_PADDING(name) struct zone_padding name; 125 #else 126 #define ZONE_PADDING(name) 127 #endif 128 129 #ifdef CONFIG_NUMA 130 enum numa_stat_item { 131 NUMA_HIT, /* allocated in intended node */ 132 NUMA_MISS, /* allocated in non intended node */ 133 NUMA_FOREIGN, /* was intended here, hit elsewhere */ 134 NUMA_INTERLEAVE_HIT, /* interleaver preferred this zone */ 135 NUMA_LOCAL, /* allocation from local node */ 136 NUMA_OTHER, /* allocation from other node */ 137 NR_VM_NUMA_STAT_ITEMS 138 }; 139 #else 140 #define NR_VM_NUMA_STAT_ITEMS 0 141 #endif 142 143 enum zone_stat_item { 144 /* First 128 byte cacheline (assuming 64 bit words) */ 145 NR_FREE_PAGES, 146 NR_ZONE_LRU_BASE, /* Used only for compaction and reclaim retry */ 147 NR_ZONE_INACTIVE_ANON = NR_ZONE_LRU_BASE, 148 NR_ZONE_ACTIVE_ANON, 149 NR_ZONE_INACTIVE_FILE, 150 NR_ZONE_ACTIVE_FILE, 151 NR_ZONE_UNEVICTABLE, 152 NR_ZONE_WRITE_PENDING, /* Count of dirty, writeback and unstable pages */ 153 NR_MLOCK, /* mlock()ed pages found and moved off LRU */ 154 /* Second 128 byte cacheline */ 155 NR_BOUNCE, 156 #if IS_ENABLED(CONFIG_ZSMALLOC) 157 NR_ZSPAGES, /* allocated in zsmalloc */ 158 #endif 159 NR_FREE_CMA_PAGES, 160 NR_VM_ZONE_STAT_ITEMS }; 161 162 enum node_stat_item { 163 NR_LRU_BASE, 164 NR_INACTIVE_ANON = NR_LRU_BASE, /* must match order of LRU_[IN]ACTIVE */ 165 NR_ACTIVE_ANON, /* " " " " " */ 166 NR_INACTIVE_FILE, /* " " " " " */ 167 NR_ACTIVE_FILE, /* " " " " " */ 168 NR_UNEVICTABLE, /* " " " " " */ 169 NR_SLAB_RECLAIMABLE_B, 170 NR_SLAB_UNRECLAIMABLE_B, 171 NR_ISOLATED_ANON, /* Temporary isolated pages from anon lru */ 172 NR_ISOLATED_FILE, /* Temporary isolated pages from file lru */ 173 WORKINGSET_NODES, 174 WORKINGSET_REFAULT_BASE, 175 WORKINGSET_REFAULT_ANON = WORKINGSET_REFAULT_BASE, 176 WORKINGSET_REFAULT_FILE, 177 WORKINGSET_ACTIVATE_BASE, 178 WORKINGSET_ACTIVATE_ANON = WORKINGSET_ACTIVATE_BASE, 179 WORKINGSET_ACTIVATE_FILE, 180 WORKINGSET_RESTORE_BASE, 181 WORKINGSET_RESTORE_ANON = WORKINGSET_RESTORE_BASE, 182 WORKINGSET_RESTORE_FILE, 183 WORKINGSET_NODERECLAIM, 184 NR_ANON_MAPPED, /* Mapped anonymous pages */ 185 NR_FILE_MAPPED, /* pagecache pages mapped into pagetables. 186 only modified from process context */ 187 NR_FILE_PAGES, 188 NR_FILE_DIRTY, 189 NR_WRITEBACK, 190 NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */ 191 NR_SHMEM, /* shmem pages (included tmpfs/GEM pages) */ 192 NR_SHMEM_THPS, 193 NR_SHMEM_PMDMAPPED, 194 NR_FILE_THPS, 195 NR_FILE_PMDMAPPED, 196 NR_ANON_THPS, 197 NR_VMSCAN_WRITE, 198 NR_VMSCAN_IMMEDIATE, /* Prioritise for reclaim when writeback ends */ 199 NR_DIRTIED, /* page dirtyings since bootup */ 200 NR_WRITTEN, /* page writings since bootup */ 201 NR_KERNEL_MISC_RECLAIMABLE, /* reclaimable non-slab kernel pages */ 202 NR_FOLL_PIN_ACQUIRED, /* via: pin_user_page(), gup flag: FOLL_PIN */ 203 NR_FOLL_PIN_RELEASED, /* pages returned via unpin_user_page() */ 204 NR_KERNEL_STACK_KB, /* measured in KiB */ 205 #if IS_ENABLED(CONFIG_SHADOW_CALL_STACK) 206 NR_KERNEL_SCS_KB, /* measured in KiB */ 207 #endif 208 NR_PAGETABLE, /* used for pagetables */ 209 #ifdef CONFIG_SWAP 210 NR_SWAPCACHE, 211 #endif 212 NR_VM_NODE_STAT_ITEMS 213 }; 214 215 /* 216 * Returns true if the item should be printed in THPs (/proc/vmstat 217 * currently prints number of anon, file and shmem THPs. But the item 218 * is charged in pages). 219 */ 220 static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item) 221 { 222 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) 223 return false; 224 225 return item == NR_ANON_THPS || 226 item == NR_FILE_THPS || 227 item == NR_SHMEM_THPS || 228 item == NR_SHMEM_PMDMAPPED || 229 item == NR_FILE_PMDMAPPED; 230 } 231 232 /* 233 * Returns true if the value is measured in bytes (most vmstat values are 234 * measured in pages). This defines the API part, the internal representation 235 * might be different. 236 */ 237 static __always_inline bool vmstat_item_in_bytes(int idx) 238 { 239 /* 240 * Global and per-node slab counters track slab pages. 241 * It's expected that changes are multiples of PAGE_SIZE. 242 * Internally values are stored in pages. 243 * 244 * Per-memcg and per-lruvec counters track memory, consumed 245 * by individual slab objects. These counters are actually 246 * byte-precise. 247 */ 248 return (idx == NR_SLAB_RECLAIMABLE_B || 249 idx == NR_SLAB_UNRECLAIMABLE_B); 250 } 251 252 /* 253 * We do arithmetic on the LRU lists in various places in the code, 254 * so it is important to keep the active lists LRU_ACTIVE higher in 255 * the array than the corresponding inactive lists, and to keep 256 * the *_FILE lists LRU_FILE higher than the corresponding _ANON lists. 257 * 258 * This has to be kept in sync with the statistics in zone_stat_item 259 * above and the descriptions in vmstat_text in mm/vmstat.c 260 */ 261 #define LRU_BASE 0 262 #define LRU_ACTIVE 1 263 #define LRU_FILE 2 264 265 enum lru_list { 266 LRU_INACTIVE_ANON = LRU_BASE, 267 LRU_ACTIVE_ANON = LRU_BASE + LRU_ACTIVE, 268 LRU_INACTIVE_FILE = LRU_BASE + LRU_FILE, 269 LRU_ACTIVE_FILE = LRU_BASE + LRU_FILE + LRU_ACTIVE, 270 LRU_UNEVICTABLE, 271 NR_LRU_LISTS 272 }; 273 274 #define for_each_lru(lru) for (lru = 0; lru < NR_LRU_LISTS; lru++) 275 276 #define for_each_evictable_lru(lru) for (lru = 0; lru <= LRU_ACTIVE_FILE; lru++) 277 278 static inline bool is_file_lru(enum lru_list lru) 279 { 280 return (lru == LRU_INACTIVE_FILE || lru == LRU_ACTIVE_FILE); 281 } 282 283 static inline bool is_active_lru(enum lru_list lru) 284 { 285 return (lru == LRU_ACTIVE_ANON || lru == LRU_ACTIVE_FILE); 286 } 287 288 #define ANON_AND_FILE 2 289 290 enum lruvec_flags { 291 LRUVEC_CONGESTED, /* lruvec has many dirty pages 292 * backed by a congested BDI 293 */ 294 }; 295 296 struct lruvec { 297 struct list_head lists[NR_LRU_LISTS]; 298 /* per lruvec lru_lock for memcg */ 299 spinlock_t lru_lock; 300 /* 301 * These track the cost of reclaiming one LRU - file or anon - 302 * over the other. As the observed cost of reclaiming one LRU 303 * increases, the reclaim scan balance tips toward the other. 304 */ 305 unsigned long anon_cost; 306 unsigned long file_cost; 307 /* Non-resident age, driven by LRU movement */ 308 atomic_long_t nonresident_age; 309 /* Refaults at the time of last reclaim cycle */ 310 unsigned long refaults[ANON_AND_FILE]; 311 /* Various lruvec state flags (enum lruvec_flags) */ 312 unsigned long flags; 313 #ifdef CONFIG_MEMCG 314 struct pglist_data *pgdat; 315 #endif 316 }; 317 318 /* Isolate unmapped pages */ 319 #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x2) 320 /* Isolate for asynchronous migration */ 321 #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x4) 322 /* Isolate unevictable pages */ 323 #define ISOLATE_UNEVICTABLE ((__force isolate_mode_t)0x8) 324 325 /* LRU Isolation modes. */ 326 typedef unsigned __bitwise isolate_mode_t; 327 328 enum zone_watermarks { 329 WMARK_MIN, 330 WMARK_LOW, 331 WMARK_HIGH, 332 NR_WMARK 333 }; 334 335 #define min_wmark_pages(z) (z->_watermark[WMARK_MIN] + z->watermark_boost) 336 #define low_wmark_pages(z) (z->_watermark[WMARK_LOW] + z->watermark_boost) 337 #define high_wmark_pages(z) (z->_watermark[WMARK_HIGH] + z->watermark_boost) 338 #define wmark_pages(z, i) (z->_watermark[i] + z->watermark_boost) 339 340 struct per_cpu_pages { 341 int count; /* number of pages in the list */ 342 int high; /* high watermark, emptying needed */ 343 int batch; /* chunk size for buddy add/remove */ 344 345 /* Lists of pages, one per migrate type stored on the pcp-lists */ 346 struct list_head lists[MIGRATE_PCPTYPES]; 347 }; 348 349 struct per_cpu_pageset { 350 struct per_cpu_pages pcp; 351 #ifdef CONFIG_NUMA 352 s8 expire; 353 u16 vm_numa_stat_diff[NR_VM_NUMA_STAT_ITEMS]; 354 #endif 355 #ifdef CONFIG_SMP 356 s8 stat_threshold; 357 s8 vm_stat_diff[NR_VM_ZONE_STAT_ITEMS]; 358 #endif 359 }; 360 361 struct per_cpu_nodestat { 362 s8 stat_threshold; 363 s8 vm_node_stat_diff[NR_VM_NODE_STAT_ITEMS]; 364 }; 365 366 #endif /* !__GENERATING_BOUNDS.H */ 367 368 enum zone_type { 369 /* 370 * ZONE_DMA and ZONE_DMA32 are used when there are peripherals not able 371 * to DMA to all of the addressable memory (ZONE_NORMAL). 372 * On architectures where this area covers the whole 32 bit address 373 * space ZONE_DMA32 is used. ZONE_DMA is left for the ones with smaller 374 * DMA addressing constraints. This distinction is important as a 32bit 375 * DMA mask is assumed when ZONE_DMA32 is defined. Some 64-bit 376 * platforms may need both zones as they support peripherals with 377 * different DMA addressing limitations. 378 */ 379 #ifdef CONFIG_ZONE_DMA 380 ZONE_DMA, 381 #endif 382 #ifdef CONFIG_ZONE_DMA32 383 ZONE_DMA32, 384 #endif 385 /* 386 * Normal addressable memory is in ZONE_NORMAL. DMA operations can be 387 * performed on pages in ZONE_NORMAL if the DMA devices support 388 * transfers to all addressable memory. 389 */ 390 ZONE_NORMAL, 391 #ifdef CONFIG_HIGHMEM 392 /* 393 * A memory area that is only addressable by the kernel through 394 * mapping portions into its own address space. This is for example 395 * used by i386 to allow the kernel to address the memory beyond 396 * 900MB. The kernel will set up special mappings (page 397 * table entries on i386) for each page that the kernel needs to 398 * access. 399 */ 400 ZONE_HIGHMEM, 401 #endif 402 /* 403 * ZONE_MOVABLE is similar to ZONE_NORMAL, except that it contains 404 * movable pages with few exceptional cases described below. Main use 405 * cases for ZONE_MOVABLE are to make memory offlining/unplug more 406 * likely to succeed, and to locally limit unmovable allocations - e.g., 407 * to increase the number of THP/huge pages. Notable special cases are: 408 * 409 * 1. Pinned pages: (long-term) pinning of movable pages might 410 * essentially turn such pages unmovable. Memory offlining might 411 * retry a long time. 412 * 2. memblock allocations: kernelcore/movablecore setups might create 413 * situations where ZONE_MOVABLE contains unmovable allocations 414 * after boot. Memory offlining and allocations fail early. 415 * 3. Memory holes: kernelcore/movablecore setups might create very rare 416 * situations where ZONE_MOVABLE contains memory holes after boot, 417 * for example, if we have sections that are only partially 418 * populated. Memory offlining and allocations fail early. 419 * 4. PG_hwpoison pages: while poisoned pages can be skipped during 420 * memory offlining, such pages cannot be allocated. 421 * 5. Unmovable PG_offline pages: in paravirtualized environments, 422 * hotplugged memory blocks might only partially be managed by the 423 * buddy (e.g., via XEN-balloon, Hyper-V balloon, virtio-mem). The 424 * parts not manged by the buddy are unmovable PG_offline pages. In 425 * some cases (virtio-mem), such pages can be skipped during 426 * memory offlining, however, cannot be moved/allocated. These 427 * techniques might use alloc_contig_range() to hide previously 428 * exposed pages from the buddy again (e.g., to implement some sort 429 * of memory unplug in virtio-mem). 430 * 431 * In general, no unmovable allocations that degrade memory offlining 432 * should end up in ZONE_MOVABLE. Allocators (like alloc_contig_range()) 433 * have to expect that migrating pages in ZONE_MOVABLE can fail (even 434 * if has_unmovable_pages() states that there are no unmovable pages, 435 * there can be false negatives). 436 */ 437 ZONE_MOVABLE, 438 #ifdef CONFIG_ZONE_DEVICE 439 ZONE_DEVICE, 440 #endif 441 __MAX_NR_ZONES 442 443 }; 444 445 #ifndef __GENERATING_BOUNDS_H 446 447 #define ASYNC_AND_SYNC 2 448 449 struct zone { 450 /* Read-mostly fields */ 451 452 /* zone watermarks, access with *_wmark_pages(zone) macros */ 453 unsigned long _watermark[NR_WMARK]; 454 unsigned long watermark_boost; 455 456 unsigned long nr_reserved_highatomic; 457 458 /* 459 * We don't know if the memory that we're going to allocate will be 460 * freeable or/and it will be released eventually, so to avoid totally 461 * wasting several GB of ram we must reserve some of the lower zone 462 * memory (otherwise we risk to run OOM on the lower zones despite 463 * there being tons of freeable ram on the higher zones). This array is 464 * recalculated at runtime if the sysctl_lowmem_reserve_ratio sysctl 465 * changes. 466 */ 467 long lowmem_reserve[MAX_NR_ZONES]; 468 469 #ifdef CONFIG_NUMA 470 int node; 471 #endif 472 struct pglist_data *zone_pgdat; 473 struct per_cpu_pageset __percpu *pageset; 474 /* 475 * the high and batch values are copied to individual pagesets for 476 * faster access 477 */ 478 int pageset_high; 479 int pageset_batch; 480 481 #ifndef CONFIG_SPARSEMEM 482 /* 483 * Flags for a pageblock_nr_pages block. See pageblock-flags.h. 484 * In SPARSEMEM, this map is stored in struct mem_section 485 */ 486 unsigned long *pageblock_flags; 487 #endif /* CONFIG_SPARSEMEM */ 488 489 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ 490 unsigned long zone_start_pfn; 491 492 /* 493 * spanned_pages is the total pages spanned by the zone, including 494 * holes, which is calculated as: 495 * spanned_pages = zone_end_pfn - zone_start_pfn; 496 * 497 * present_pages is physical pages existing within the zone, which 498 * is calculated as: 499 * present_pages = spanned_pages - absent_pages(pages in holes); 500 * 501 * managed_pages is present pages managed by the buddy system, which 502 * is calculated as (reserved_pages includes pages allocated by the 503 * bootmem allocator): 504 * managed_pages = present_pages - reserved_pages; 505 * 506 * cma pages is present pages that are assigned for CMA use 507 * (MIGRATE_CMA). 508 * 509 * So present_pages may be used by memory hotplug or memory power 510 * management logic to figure out unmanaged pages by checking 511 * (present_pages - managed_pages). And managed_pages should be used 512 * by page allocator and vm scanner to calculate all kinds of watermarks 513 * and thresholds. 514 * 515 * Locking rules: 516 * 517 * zone_start_pfn and spanned_pages are protected by span_seqlock. 518 * It is a seqlock because it has to be read outside of zone->lock, 519 * and it is done in the main allocator path. But, it is written 520 * quite infrequently. 521 * 522 * The span_seq lock is declared along with zone->lock because it is 523 * frequently read in proximity to zone->lock. It's good to 524 * give them a chance of being in the same cacheline. 525 * 526 * Write access to present_pages at runtime should be protected by 527 * mem_hotplug_begin/end(). Any reader who can't tolerant drift of 528 * present_pages should get_online_mems() to get a stable value. 529 */ 530 atomic_long_t managed_pages; 531 unsigned long spanned_pages; 532 unsigned long present_pages; 533 #ifdef CONFIG_CMA 534 unsigned long cma_pages; 535 #endif 536 537 const char *name; 538 539 #ifdef CONFIG_MEMORY_ISOLATION 540 /* 541 * Number of isolated pageblock. It is used to solve incorrect 542 * freepage counting problem due to racy retrieving migratetype 543 * of pageblock. Protected by zone->lock. 544 */ 545 unsigned long nr_isolate_pageblock; 546 #endif 547 548 #ifdef CONFIG_MEMORY_HOTPLUG 549 /* see spanned/present_pages for more description */ 550 seqlock_t span_seqlock; 551 #endif 552 553 int initialized; 554 555 /* Write-intensive fields used from the page allocator */ 556 ZONE_PADDING(_pad1_) 557 558 /* free areas of different sizes */ 559 struct free_area free_area[MAX_ORDER]; 560 561 /* zone flags, see below */ 562 unsigned long flags; 563 564 /* Primarily protects free_area */ 565 spinlock_t lock; 566 567 /* Write-intensive fields used by compaction and vmstats. */ 568 ZONE_PADDING(_pad2_) 569 570 /* 571 * When free pages are below this point, additional steps are taken 572 * when reading the number of free pages to avoid per-cpu counter 573 * drift allowing watermarks to be breached 574 */ 575 unsigned long percpu_drift_mark; 576 577 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 578 /* pfn where compaction free scanner should start */ 579 unsigned long compact_cached_free_pfn; 580 /* pfn where compaction migration scanner should start */ 581 unsigned long compact_cached_migrate_pfn[ASYNC_AND_SYNC]; 582 unsigned long compact_init_migrate_pfn; 583 unsigned long compact_init_free_pfn; 584 #endif 585 586 #ifdef CONFIG_COMPACTION 587 /* 588 * On compaction failure, 1<<compact_defer_shift compactions 589 * are skipped before trying again. The number attempted since 590 * last failure is tracked with compact_considered. 591 * compact_order_failed is the minimum compaction failed order. 592 */ 593 unsigned int compact_considered; 594 unsigned int compact_defer_shift; 595 int compact_order_failed; 596 #endif 597 598 #if defined CONFIG_COMPACTION || defined CONFIG_CMA 599 /* Set to true when the PG_migrate_skip bits should be cleared */ 600 bool compact_blockskip_flush; 601 #endif 602 603 bool contiguous; 604 605 ZONE_PADDING(_pad3_) 606 /* Zone statistics */ 607 atomic_long_t vm_stat[NR_VM_ZONE_STAT_ITEMS]; 608 atomic_long_t vm_numa_stat[NR_VM_NUMA_STAT_ITEMS]; 609 } ____cacheline_internodealigned_in_smp; 610 611 enum pgdat_flags { 612 PGDAT_DIRTY, /* reclaim scanning has recently found 613 * many dirty file pages at the tail 614 * of the LRU. 615 */ 616 PGDAT_WRITEBACK, /* reclaim scanning has recently found 617 * many pages under writeback 618 */ 619 PGDAT_RECLAIM_LOCKED, /* prevents concurrent reclaim */ 620 }; 621 622 enum zone_flags { 623 ZONE_BOOSTED_WATERMARK, /* zone recently boosted watermarks. 624 * Cleared when kswapd is woken. 625 */ 626 }; 627 628 static inline unsigned long zone_managed_pages(struct zone *zone) 629 { 630 return (unsigned long)atomic_long_read(&zone->managed_pages); 631 } 632 633 static inline unsigned long zone_cma_pages(struct zone *zone) 634 { 635 #ifdef CONFIG_CMA 636 return zone->cma_pages; 637 #else 638 return 0; 639 #endif 640 } 641 642 static inline unsigned long zone_end_pfn(const struct zone *zone) 643 { 644 return zone->zone_start_pfn + zone->spanned_pages; 645 } 646 647 static inline bool zone_spans_pfn(const struct zone *zone, unsigned long pfn) 648 { 649 return zone->zone_start_pfn <= pfn && pfn < zone_end_pfn(zone); 650 } 651 652 static inline bool zone_is_initialized(struct zone *zone) 653 { 654 return zone->initialized; 655 } 656 657 static inline bool zone_is_empty(struct zone *zone) 658 { 659 return zone->spanned_pages == 0; 660 } 661 662 /* 663 * Return true if [start_pfn, start_pfn + nr_pages) range has a non-empty 664 * intersection with the given zone 665 */ 666 static inline bool zone_intersects(struct zone *zone, 667 unsigned long start_pfn, unsigned long nr_pages) 668 { 669 if (zone_is_empty(zone)) 670 return false; 671 if (start_pfn >= zone_end_pfn(zone) || 672 start_pfn + nr_pages <= zone->zone_start_pfn) 673 return false; 674 675 return true; 676 } 677 678 /* 679 * The "priority" of VM scanning is how much of the queues we will scan in one 680 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the 681 * queues ("queue_length >> 12") during an aging round. 682 */ 683 #define DEF_PRIORITY 12 684 685 /* Maximum number of zones on a zonelist */ 686 #define MAX_ZONES_PER_ZONELIST (MAX_NUMNODES * MAX_NR_ZONES) 687 688 enum { 689 ZONELIST_FALLBACK, /* zonelist with fallback */ 690 #ifdef CONFIG_NUMA 691 /* 692 * The NUMA zonelists are doubled because we need zonelists that 693 * restrict the allocations to a single node for __GFP_THISNODE. 694 */ 695 ZONELIST_NOFALLBACK, /* zonelist without fallback (__GFP_THISNODE) */ 696 #endif 697 MAX_ZONELISTS 698 }; 699 700 /* 701 * This struct contains information about a zone in a zonelist. It is stored 702 * here to avoid dereferences into large structures and lookups of tables 703 */ 704 struct zoneref { 705 struct zone *zone; /* Pointer to actual zone */ 706 int zone_idx; /* zone_idx(zoneref->zone) */ 707 }; 708 709 /* 710 * One allocation request operates on a zonelist. A zonelist 711 * is a list of zones, the first one is the 'goal' of the 712 * allocation, the other zones are fallback zones, in decreasing 713 * priority. 714 * 715 * To speed the reading of the zonelist, the zonerefs contain the zone index 716 * of the entry being read. Helper functions to access information given 717 * a struct zoneref are 718 * 719 * zonelist_zone() - Return the struct zone * for an entry in _zonerefs 720 * zonelist_zone_idx() - Return the index of the zone for an entry 721 * zonelist_node_idx() - Return the index of the node for an entry 722 */ 723 struct zonelist { 724 struct zoneref _zonerefs[MAX_ZONES_PER_ZONELIST + 1]; 725 }; 726 727 #ifndef CONFIG_DISCONTIGMEM 728 /* The array of struct pages - for discontigmem use pgdat->lmem_map */ 729 extern struct page *mem_map; 730 #endif 731 732 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 733 struct deferred_split { 734 spinlock_t split_queue_lock; 735 struct list_head split_queue; 736 unsigned long split_queue_len; 737 }; 738 #endif 739 740 /* 741 * On NUMA machines, each NUMA node would have a pg_data_t to describe 742 * it's memory layout. On UMA machines there is a single pglist_data which 743 * describes the whole memory. 744 * 745 * Memory statistics and page replacement data structures are maintained on a 746 * per-zone basis. 747 */ 748 typedef struct pglist_data { 749 /* 750 * node_zones contains just the zones for THIS node. Not all of the 751 * zones may be populated, but it is the full list. It is referenced by 752 * this node's node_zonelists as well as other node's node_zonelists. 753 */ 754 struct zone node_zones[MAX_NR_ZONES]; 755 756 /* 757 * node_zonelists contains references to all zones in all nodes. 758 * Generally the first zones will be references to this node's 759 * node_zones. 760 */ 761 struct zonelist node_zonelists[MAX_ZONELISTS]; 762 763 int nr_zones; /* number of populated zones in this node */ 764 #ifdef CONFIG_FLAT_NODE_MEM_MAP /* means !SPARSEMEM */ 765 struct page *node_mem_map; 766 #ifdef CONFIG_PAGE_EXTENSION 767 struct page_ext *node_page_ext; 768 #endif 769 #endif 770 #if defined(CONFIG_MEMORY_HOTPLUG) || defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT) 771 /* 772 * Must be held any time you expect node_start_pfn, 773 * node_present_pages, node_spanned_pages or nr_zones to stay constant. 774 * Also synchronizes pgdat->first_deferred_pfn during deferred page 775 * init. 776 * 777 * pgdat_resize_lock() and pgdat_resize_unlock() are provided to 778 * manipulate node_size_lock without checking for CONFIG_MEMORY_HOTPLUG 779 * or CONFIG_DEFERRED_STRUCT_PAGE_INIT. 780 * 781 * Nests above zone->lock and zone->span_seqlock 782 */ 783 spinlock_t node_size_lock; 784 #endif 785 unsigned long node_start_pfn; 786 unsigned long node_present_pages; /* total number of physical pages */ 787 unsigned long node_spanned_pages; /* total size of physical page 788 range, including holes */ 789 int node_id; 790 wait_queue_head_t kswapd_wait; 791 wait_queue_head_t pfmemalloc_wait; 792 struct task_struct *kswapd; /* Protected by 793 mem_hotplug_begin/end() */ 794 int kswapd_order; 795 enum zone_type kswapd_highest_zoneidx; 796 797 int kswapd_failures; /* Number of 'reclaimed == 0' runs */ 798 799 #ifdef CONFIG_COMPACTION 800 int kcompactd_max_order; 801 enum zone_type kcompactd_highest_zoneidx; 802 wait_queue_head_t kcompactd_wait; 803 struct task_struct *kcompactd; 804 #endif 805 /* 806 * This is a per-node reserve of pages that are not available 807 * to userspace allocations. 808 */ 809 unsigned long totalreserve_pages; 810 811 #ifdef CONFIG_NUMA 812 /* 813 * node reclaim becomes active if more unmapped pages exist. 814 */ 815 unsigned long min_unmapped_pages; 816 unsigned long min_slab_pages; 817 #endif /* CONFIG_NUMA */ 818 819 /* Write-intensive fields used by page reclaim */ 820 ZONE_PADDING(_pad1_) 821 822 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT 823 /* 824 * If memory initialisation on large machines is deferred then this 825 * is the first PFN that needs to be initialised. 826 */ 827 unsigned long first_deferred_pfn; 828 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ 829 830 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 831 struct deferred_split deferred_split_queue; 832 #endif 833 834 /* Fields commonly accessed by the page reclaim scanner */ 835 836 /* 837 * NOTE: THIS IS UNUSED IF MEMCG IS ENABLED. 838 * 839 * Use mem_cgroup_lruvec() to look up lruvecs. 840 */ 841 struct lruvec __lruvec; 842 843 unsigned long flags; 844 845 ZONE_PADDING(_pad2_) 846 847 /* Per-node vmstats */ 848 struct per_cpu_nodestat __percpu *per_cpu_nodestats; 849 atomic_long_t vm_stat[NR_VM_NODE_STAT_ITEMS]; 850 } pg_data_t; 851 852 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) 853 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) 854 #ifdef CONFIG_FLAT_NODE_MEM_MAP 855 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) 856 #else 857 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) 858 #endif 859 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) 860 861 #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) 862 #define node_end_pfn(nid) pgdat_end_pfn(NODE_DATA(nid)) 863 864 static inline unsigned long pgdat_end_pfn(pg_data_t *pgdat) 865 { 866 return pgdat->node_start_pfn + pgdat->node_spanned_pages; 867 } 868 869 static inline bool pgdat_is_empty(pg_data_t *pgdat) 870 { 871 return !pgdat->node_start_pfn && !pgdat->node_spanned_pages; 872 } 873 874 #include <linux/memory_hotplug.h> 875 876 void build_all_zonelists(pg_data_t *pgdat); 877 void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order, 878 enum zone_type highest_zoneidx); 879 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, 880 int highest_zoneidx, unsigned int alloc_flags, 881 long free_pages); 882 bool zone_watermark_ok(struct zone *z, unsigned int order, 883 unsigned long mark, int highest_zoneidx, 884 unsigned int alloc_flags); 885 bool zone_watermark_ok_safe(struct zone *z, unsigned int order, 886 unsigned long mark, int highest_zoneidx); 887 /* 888 * Memory initialization context, use to differentiate memory added by 889 * the platform statically or via memory hotplug interface. 890 */ 891 enum meminit_context { 892 MEMINIT_EARLY, 893 MEMINIT_HOTPLUG, 894 }; 895 896 extern void init_currently_empty_zone(struct zone *zone, unsigned long start_pfn, 897 unsigned long size); 898 899 extern void lruvec_init(struct lruvec *lruvec); 900 901 static inline struct pglist_data *lruvec_pgdat(struct lruvec *lruvec) 902 { 903 #ifdef CONFIG_MEMCG 904 return lruvec->pgdat; 905 #else 906 return container_of(lruvec, struct pglist_data, __lruvec); 907 #endif 908 } 909 910 #ifdef CONFIG_HAVE_MEMORYLESS_NODES 911 int local_memory_node(int node_id); 912 #else 913 static inline int local_memory_node(int node_id) { return node_id; }; 914 #endif 915 916 /* 917 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. 918 */ 919 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) 920 921 #ifdef CONFIG_ZONE_DEVICE 922 static inline bool zone_is_zone_device(struct zone *zone) 923 { 924 return zone_idx(zone) == ZONE_DEVICE; 925 } 926 #else 927 static inline bool zone_is_zone_device(struct zone *zone) 928 { 929 return false; 930 } 931 #endif 932 933 /* 934 * Returns true if a zone has pages managed by the buddy allocator. 935 * All the reclaim decisions have to use this function rather than 936 * populated_zone(). If the whole zone is reserved then we can easily 937 * end up with populated_zone() && !managed_zone(). 938 */ 939 static inline bool managed_zone(struct zone *zone) 940 { 941 return zone_managed_pages(zone); 942 } 943 944 /* Returns true if a zone has memory */ 945 static inline bool populated_zone(struct zone *zone) 946 { 947 return zone->present_pages; 948 } 949 950 #ifdef CONFIG_NUMA 951 static inline int zone_to_nid(struct zone *zone) 952 { 953 return zone->node; 954 } 955 956 static inline void zone_set_nid(struct zone *zone, int nid) 957 { 958 zone->node = nid; 959 } 960 #else 961 static inline int zone_to_nid(struct zone *zone) 962 { 963 return 0; 964 } 965 966 static inline void zone_set_nid(struct zone *zone, int nid) {} 967 #endif 968 969 extern int movable_zone; 970 971 #ifdef CONFIG_HIGHMEM 972 static inline int zone_movable_is_highmem(void) 973 { 974 #ifdef CONFIG_NEED_MULTIPLE_NODES 975 return movable_zone == ZONE_HIGHMEM; 976 #else 977 return (ZONE_MOVABLE - 1) == ZONE_HIGHMEM; 978 #endif 979 } 980 #endif 981 982 static inline int is_highmem_idx(enum zone_type idx) 983 { 984 #ifdef CONFIG_HIGHMEM 985 return (idx == ZONE_HIGHMEM || 986 (idx == ZONE_MOVABLE && zone_movable_is_highmem())); 987 #else 988 return 0; 989 #endif 990 } 991 992 /** 993 * is_highmem - helper function to quickly check if a struct zone is a 994 * highmem zone or not. This is an attempt to keep references 995 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. 996 * @zone: pointer to struct zone variable 997 * Return: 1 for a highmem zone, 0 otherwise 998 */ 999 static inline int is_highmem(struct zone *zone) 1000 { 1001 #ifdef CONFIG_HIGHMEM 1002 return is_highmem_idx(zone_idx(zone)); 1003 #else 1004 return 0; 1005 #endif 1006 } 1007 1008 /* These two functions are used to setup the per zone pages min values */ 1009 struct ctl_table; 1010 1011 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, void *, size_t *, 1012 loff_t *); 1013 int watermark_scale_factor_sysctl_handler(struct ctl_table *, int, void *, 1014 size_t *, loff_t *); 1015 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES]; 1016 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, void *, 1017 size_t *, loff_t *); 1018 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, 1019 void *, size_t *, loff_t *); 1020 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *, int, 1021 void *, size_t *, loff_t *); 1022 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *, int, 1023 void *, size_t *, loff_t *); 1024 int numa_zonelist_order_handler(struct ctl_table *, int, 1025 void *, size_t *, loff_t *); 1026 extern int percpu_pagelist_fraction; 1027 extern char numa_zonelist_order[]; 1028 #define NUMA_ZONELIST_ORDER_LEN 16 1029 1030 #ifndef CONFIG_NEED_MULTIPLE_NODES 1031 1032 extern struct pglist_data contig_page_data; 1033 #define NODE_DATA(nid) (&contig_page_data) 1034 #define NODE_MEM_MAP(nid) mem_map 1035 1036 #else /* CONFIG_NEED_MULTIPLE_NODES */ 1037 1038 #include <asm/mmzone.h> 1039 1040 #endif /* !CONFIG_NEED_MULTIPLE_NODES */ 1041 1042 extern struct pglist_data *first_online_pgdat(void); 1043 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); 1044 extern struct zone *next_zone(struct zone *zone); 1045 1046 /** 1047 * for_each_online_pgdat - helper macro to iterate over all online nodes 1048 * @pgdat: pointer to a pg_data_t variable 1049 */ 1050 #define for_each_online_pgdat(pgdat) \ 1051 for (pgdat = first_online_pgdat(); \ 1052 pgdat; \ 1053 pgdat = next_online_pgdat(pgdat)) 1054 /** 1055 * for_each_zone - helper macro to iterate over all memory zones 1056 * @zone: pointer to struct zone variable 1057 * 1058 * The user only needs to declare the zone variable, for_each_zone 1059 * fills it in. 1060 */ 1061 #define for_each_zone(zone) \ 1062 for (zone = (first_online_pgdat())->node_zones; \ 1063 zone; \ 1064 zone = next_zone(zone)) 1065 1066 #define for_each_populated_zone(zone) \ 1067 for (zone = (first_online_pgdat())->node_zones; \ 1068 zone; \ 1069 zone = next_zone(zone)) \ 1070 if (!populated_zone(zone)) \ 1071 ; /* do nothing */ \ 1072 else 1073 1074 static inline struct zone *zonelist_zone(struct zoneref *zoneref) 1075 { 1076 return zoneref->zone; 1077 } 1078 1079 static inline int zonelist_zone_idx(struct zoneref *zoneref) 1080 { 1081 return zoneref->zone_idx; 1082 } 1083 1084 static inline int zonelist_node_idx(struct zoneref *zoneref) 1085 { 1086 return zone_to_nid(zoneref->zone); 1087 } 1088 1089 struct zoneref *__next_zones_zonelist(struct zoneref *z, 1090 enum zone_type highest_zoneidx, 1091 nodemask_t *nodes); 1092 1093 /** 1094 * 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 1095 * @z: The cursor used as a starting point for the search 1096 * @highest_zoneidx: The zone index of the highest zone to return 1097 * @nodes: An optional nodemask to filter the zonelist with 1098 * 1099 * This function returns the next zone at or below a given zone index that is 1100 * within the allowed nodemask using a cursor as the starting point for the 1101 * search. The zoneref returned is a cursor that represents the current zone 1102 * being examined. It should be advanced by one before calling 1103 * next_zones_zonelist again. 1104 * 1105 * Return: the next zone at or below highest_zoneidx within the allowed 1106 * nodemask using a cursor within a zonelist as a starting point 1107 */ 1108 static __always_inline struct zoneref *next_zones_zonelist(struct zoneref *z, 1109 enum zone_type highest_zoneidx, 1110 nodemask_t *nodes) 1111 { 1112 if (likely(!nodes && zonelist_zone_idx(z) <= highest_zoneidx)) 1113 return z; 1114 return __next_zones_zonelist(z, highest_zoneidx, nodes); 1115 } 1116 1117 /** 1118 * first_zones_zonelist - Returns the first zone at or below highest_zoneidx within the allowed nodemask in a zonelist 1119 * @zonelist: The zonelist to search for a suitable zone 1120 * @highest_zoneidx: The zone index of the highest zone to return 1121 * @nodes: An optional nodemask to filter the zonelist with 1122 * 1123 * This function returns the first zone at or below a given zone index that is 1124 * within the allowed nodemask. The zoneref returned is a cursor that can be 1125 * used to iterate the zonelist with next_zones_zonelist by advancing it by 1126 * one before calling. 1127 * 1128 * When no eligible zone is found, zoneref->zone is NULL (zoneref itself is 1129 * never NULL). This may happen either genuinely, or due to concurrent nodemask 1130 * update due to cpuset modification. 1131 * 1132 * Return: Zoneref pointer for the first suitable zone found 1133 */ 1134 static inline struct zoneref *first_zones_zonelist(struct zonelist *zonelist, 1135 enum zone_type highest_zoneidx, 1136 nodemask_t *nodes) 1137 { 1138 return next_zones_zonelist(zonelist->_zonerefs, 1139 highest_zoneidx, nodes); 1140 } 1141 1142 /** 1143 * 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 1144 * @zone: The current zone in the iterator 1145 * @z: The current pointer within zonelist->_zonerefs being iterated 1146 * @zlist: The zonelist being iterated 1147 * @highidx: The zone index of the highest zone to return 1148 * @nodemask: Nodemask allowed by the allocator 1149 * 1150 * This iterator iterates though all zones at or below a given zone index and 1151 * within a given nodemask 1152 */ 1153 #define for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, nodemask) \ 1154 for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \ 1155 zone; \ 1156 z = next_zones_zonelist(++z, highidx, nodemask), \ 1157 zone = zonelist_zone(z)) 1158 1159 #define for_next_zone_zonelist_nodemask(zone, z, highidx, nodemask) \ 1160 for (zone = z->zone; \ 1161 zone; \ 1162 z = next_zones_zonelist(++z, highidx, nodemask), \ 1163 zone = zonelist_zone(z)) 1164 1165 1166 /** 1167 * for_each_zone_zonelist - helper macro to iterate over valid zones in a zonelist at or below a given zone index 1168 * @zone: The current zone in the iterator 1169 * @z: The current pointer within zonelist->zones being iterated 1170 * @zlist: The zonelist being iterated 1171 * @highidx: The zone index of the highest zone to return 1172 * 1173 * This iterator iterates though all zones at or below a given zone index. 1174 */ 1175 #define for_each_zone_zonelist(zone, z, zlist, highidx) \ 1176 for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL) 1177 1178 #ifdef CONFIG_SPARSEMEM 1179 #include <asm/sparsemem.h> 1180 #endif 1181 1182 #ifdef CONFIG_FLATMEM 1183 #define pfn_to_nid(pfn) (0) 1184 #endif 1185 1186 #ifdef CONFIG_SPARSEMEM 1187 1188 /* 1189 * SECTION_SHIFT #bits space required to store a section # 1190 * 1191 * PA_SECTION_SHIFT physical address to/from section number 1192 * PFN_SECTION_SHIFT pfn to/from section number 1193 */ 1194 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) 1195 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) 1196 1197 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) 1198 1199 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) 1200 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) 1201 1202 #define SECTION_BLOCKFLAGS_BITS \ 1203 ((1UL << (PFN_SECTION_SHIFT - pageblock_order)) * NR_PAGEBLOCK_BITS) 1204 1205 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS 1206 #error Allocator MAX_ORDER exceeds SECTION_SIZE 1207 #endif 1208 1209 static inline unsigned long pfn_to_section_nr(unsigned long pfn) 1210 { 1211 return pfn >> PFN_SECTION_SHIFT; 1212 } 1213 static inline unsigned long section_nr_to_pfn(unsigned long sec) 1214 { 1215 return sec << PFN_SECTION_SHIFT; 1216 } 1217 1218 #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK) 1219 #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK) 1220 1221 #define SUBSECTION_SHIFT 21 1222 #define SUBSECTION_SIZE (1UL << SUBSECTION_SHIFT) 1223 1224 #define PFN_SUBSECTION_SHIFT (SUBSECTION_SHIFT - PAGE_SHIFT) 1225 #define PAGES_PER_SUBSECTION (1UL << PFN_SUBSECTION_SHIFT) 1226 #define PAGE_SUBSECTION_MASK (~(PAGES_PER_SUBSECTION-1)) 1227 1228 #if SUBSECTION_SHIFT > SECTION_SIZE_BITS 1229 #error Subsection size exceeds section size 1230 #else 1231 #define SUBSECTIONS_PER_SECTION (1UL << (SECTION_SIZE_BITS - SUBSECTION_SHIFT)) 1232 #endif 1233 1234 #define SUBSECTION_ALIGN_UP(pfn) ALIGN((pfn), PAGES_PER_SUBSECTION) 1235 #define SUBSECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SUBSECTION_MASK) 1236 1237 struct mem_section_usage { 1238 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1239 DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION); 1240 #endif 1241 /* See declaration of similar field in struct zone */ 1242 unsigned long pageblock_flags[0]; 1243 }; 1244 1245 void subsection_map_init(unsigned long pfn, unsigned long nr_pages); 1246 1247 struct page; 1248 struct page_ext; 1249 struct mem_section { 1250 /* 1251 * This is, logically, a pointer to an array of struct 1252 * pages. However, it is stored with some other magic. 1253 * (see sparse.c::sparse_init_one_section()) 1254 * 1255 * Additionally during early boot we encode node id of 1256 * the location of the section here to guide allocation. 1257 * (see sparse.c::memory_present()) 1258 * 1259 * Making it a UL at least makes someone do a cast 1260 * before using it wrong. 1261 */ 1262 unsigned long section_mem_map; 1263 1264 struct mem_section_usage *usage; 1265 #ifdef CONFIG_PAGE_EXTENSION 1266 /* 1267 * If SPARSEMEM, pgdat doesn't have page_ext pointer. We use 1268 * section. (see page_ext.h about this.) 1269 */ 1270 struct page_ext *page_ext; 1271 unsigned long pad; 1272 #endif 1273 /* 1274 * WARNING: mem_section must be a power-of-2 in size for the 1275 * calculation and use of SECTION_ROOT_MASK to make sense. 1276 */ 1277 }; 1278 1279 #ifdef CONFIG_SPARSEMEM_EXTREME 1280 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) 1281 #else 1282 #define SECTIONS_PER_ROOT 1 1283 #endif 1284 1285 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) 1286 #define NR_SECTION_ROOTS DIV_ROUND_UP(NR_MEM_SECTIONS, SECTIONS_PER_ROOT) 1287 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) 1288 1289 #ifdef CONFIG_SPARSEMEM_EXTREME 1290 extern struct mem_section **mem_section; 1291 #else 1292 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; 1293 #endif 1294 1295 static inline unsigned long *section_to_usemap(struct mem_section *ms) 1296 { 1297 return ms->usage->pageblock_flags; 1298 } 1299 1300 static inline struct mem_section *__nr_to_section(unsigned long nr) 1301 { 1302 #ifdef CONFIG_SPARSEMEM_EXTREME 1303 if (!mem_section) 1304 return NULL; 1305 #endif 1306 if (!mem_section[SECTION_NR_TO_ROOT(nr)]) 1307 return NULL; 1308 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; 1309 } 1310 extern unsigned long __section_nr(struct mem_section *ms); 1311 extern size_t mem_section_usage_size(void); 1312 1313 /* 1314 * We use the lower bits of the mem_map pointer to store 1315 * a little bit of information. The pointer is calculated 1316 * as mem_map - section_nr_to_pfn(pnum). The result is 1317 * aligned to the minimum alignment of the two values: 1318 * 1. All mem_map arrays are page-aligned. 1319 * 2. section_nr_to_pfn() always clears PFN_SECTION_SHIFT 1320 * lowest bits. PFN_SECTION_SHIFT is arch-specific 1321 * (equal SECTION_SIZE_BITS - PAGE_SHIFT), and the 1322 * worst combination is powerpc with 256k pages, 1323 * which results in PFN_SECTION_SHIFT equal 6. 1324 * To sum it up, at least 6 bits are available. 1325 */ 1326 #define SECTION_MARKED_PRESENT (1UL<<0) 1327 #define SECTION_HAS_MEM_MAP (1UL<<1) 1328 #define SECTION_IS_ONLINE (1UL<<2) 1329 #define SECTION_IS_EARLY (1UL<<3) 1330 #define SECTION_TAINT_ZONE_DEVICE (1UL<<4) 1331 #define SECTION_MAP_LAST_BIT (1UL<<5) 1332 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) 1333 #define SECTION_NID_SHIFT 3 1334 1335 static inline struct page *__section_mem_map_addr(struct mem_section *section) 1336 { 1337 unsigned long map = section->section_mem_map; 1338 map &= SECTION_MAP_MASK; 1339 return (struct page *)map; 1340 } 1341 1342 static inline int present_section(struct mem_section *section) 1343 { 1344 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); 1345 } 1346 1347 static inline int present_section_nr(unsigned long nr) 1348 { 1349 return present_section(__nr_to_section(nr)); 1350 } 1351 1352 static inline int valid_section(struct mem_section *section) 1353 { 1354 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); 1355 } 1356 1357 static inline int early_section(struct mem_section *section) 1358 { 1359 return (section && (section->section_mem_map & SECTION_IS_EARLY)); 1360 } 1361 1362 static inline int valid_section_nr(unsigned long nr) 1363 { 1364 return valid_section(__nr_to_section(nr)); 1365 } 1366 1367 static inline int online_section(struct mem_section *section) 1368 { 1369 return (section && (section->section_mem_map & SECTION_IS_ONLINE)); 1370 } 1371 1372 static inline int online_device_section(struct mem_section *section) 1373 { 1374 unsigned long flags = SECTION_IS_ONLINE | SECTION_TAINT_ZONE_DEVICE; 1375 1376 return section && ((section->section_mem_map & flags) == flags); 1377 } 1378 1379 static inline int online_section_nr(unsigned long nr) 1380 { 1381 return online_section(__nr_to_section(nr)); 1382 } 1383 1384 #ifdef CONFIG_MEMORY_HOTPLUG 1385 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1386 #ifdef CONFIG_MEMORY_HOTREMOVE 1387 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn); 1388 #endif 1389 #endif 1390 1391 static inline struct mem_section *__pfn_to_section(unsigned long pfn) 1392 { 1393 return __nr_to_section(pfn_to_section_nr(pfn)); 1394 } 1395 1396 extern unsigned long __highest_present_section_nr; 1397 1398 static inline int subsection_map_index(unsigned long pfn) 1399 { 1400 return (pfn & ~(PAGE_SECTION_MASK)) / PAGES_PER_SUBSECTION; 1401 } 1402 1403 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1404 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1405 { 1406 int idx = subsection_map_index(pfn); 1407 1408 return test_bit(idx, ms->usage->subsection_map); 1409 } 1410 #else 1411 static inline int pfn_section_valid(struct mem_section *ms, unsigned long pfn) 1412 { 1413 return 1; 1414 } 1415 #endif 1416 1417 #ifndef CONFIG_HAVE_ARCH_PFN_VALID 1418 static inline int pfn_valid(unsigned long pfn) 1419 { 1420 struct mem_section *ms; 1421 1422 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 1423 return 0; 1424 ms = __nr_to_section(pfn_to_section_nr(pfn)); 1425 if (!valid_section(ms)) 1426 return 0; 1427 /* 1428 * Traditionally early sections always returned pfn_valid() for 1429 * the entire section-sized span. 1430 */ 1431 return early_section(ms) || pfn_section_valid(ms, pfn); 1432 } 1433 #endif 1434 1435 static inline int pfn_in_present_section(unsigned long pfn) 1436 { 1437 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 1438 return 0; 1439 return present_section(__nr_to_section(pfn_to_section_nr(pfn))); 1440 } 1441 1442 static inline unsigned long next_present_section_nr(unsigned long section_nr) 1443 { 1444 while (++section_nr <= __highest_present_section_nr) { 1445 if (present_section_nr(section_nr)) 1446 return section_nr; 1447 } 1448 1449 return -1; 1450 } 1451 1452 /* 1453 * These are _only_ used during initialisation, therefore they 1454 * can use __initdata ... They could have names to indicate 1455 * this restriction. 1456 */ 1457 #ifdef CONFIG_NUMA 1458 #define pfn_to_nid(pfn) \ 1459 ({ \ 1460 unsigned long __pfn_to_nid_pfn = (pfn); \ 1461 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ 1462 }) 1463 #else 1464 #define pfn_to_nid(pfn) (0) 1465 #endif 1466 1467 void sparse_init(void); 1468 #else 1469 #define sparse_init() do {} while (0) 1470 #define sparse_index_init(_sec, _nid) do {} while (0) 1471 #define pfn_in_present_section pfn_valid 1472 #define subsection_map_init(_pfn, _nr_pages) do {} while (0) 1473 #endif /* CONFIG_SPARSEMEM */ 1474 1475 /* 1476 * If it is possible to have holes within a MAX_ORDER_NR_PAGES, then we 1477 * need to check pfn validity within that MAX_ORDER_NR_PAGES block. 1478 * pfn_valid_within() should be used in this case; we optimise this away 1479 * when we have no holes within a MAX_ORDER_NR_PAGES block. 1480 */ 1481 #ifdef CONFIG_HOLES_IN_ZONE 1482 #define pfn_valid_within(pfn) pfn_valid(pfn) 1483 #else 1484 #define pfn_valid_within(pfn) (1) 1485 #endif 1486 1487 #endif /* !__GENERATING_BOUNDS.H */ 1488 #endif /* !__ASSEMBLY__ */ 1489 #endif /* _LINUX_MMZONE_H */ 1490