1 #ifndef _LINUX_MMZONE_H 2 #define _LINUX_MMZONE_H 3 4 #ifdef __KERNEL__ 5 #ifndef __ASSEMBLY__ 6 7 #include <linux/config.h> 8 #include <linux/spinlock.h> 9 #include <linux/list.h> 10 #include <linux/wait.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 <asm/atomic.h> 18 19 /* Free memory management - zoned buddy allocator. */ 20 #ifndef CONFIG_FORCE_MAX_ZONEORDER 21 #define MAX_ORDER 11 22 #else 23 #define MAX_ORDER CONFIG_FORCE_MAX_ZONEORDER 24 #endif 25 26 struct free_area { 27 struct list_head free_list; 28 unsigned long nr_free; 29 }; 30 31 struct pglist_data; 32 33 /* 34 * zone->lock and zone->lru_lock are two of the hottest locks in the kernel. 35 * So add a wild amount of padding here to ensure that they fall into separate 36 * cachelines. There are very few zone structures in the machine, so space 37 * consumption is not a concern here. 38 */ 39 #if defined(CONFIG_SMP) 40 struct zone_padding { 41 char x[0]; 42 } ____cacheline_internodealigned_in_smp; 43 #define ZONE_PADDING(name) struct zone_padding name; 44 #else 45 #define ZONE_PADDING(name) 46 #endif 47 48 struct per_cpu_pages { 49 int count; /* number of pages in the list */ 50 int high; /* high watermark, emptying needed */ 51 int batch; /* chunk size for buddy add/remove */ 52 struct list_head list; /* the list of pages */ 53 }; 54 55 struct per_cpu_pageset { 56 struct per_cpu_pages pcp[2]; /* 0: hot. 1: cold */ 57 #ifdef CONFIG_NUMA 58 unsigned long numa_hit; /* allocated in intended node */ 59 unsigned long numa_miss; /* allocated in non intended node */ 60 unsigned long numa_foreign; /* was intended here, hit elsewhere */ 61 unsigned long interleave_hit; /* interleaver prefered this zone */ 62 unsigned long local_node; /* allocation from local node */ 63 unsigned long other_node; /* allocation from other node */ 64 #endif 65 } ____cacheline_aligned_in_smp; 66 67 #ifdef CONFIG_NUMA 68 #define zone_pcp(__z, __cpu) ((__z)->pageset[(__cpu)]) 69 #else 70 #define zone_pcp(__z, __cpu) (&(__z)->pageset[(__cpu)]) 71 #endif 72 73 #define ZONE_DMA 0 74 #define ZONE_DMA32 1 75 #define ZONE_NORMAL 2 76 #define ZONE_HIGHMEM 3 77 78 #define MAX_NR_ZONES 4 /* Sync this with ZONES_SHIFT */ 79 #define ZONES_SHIFT 2 /* ceil(log2(MAX_NR_ZONES)) */ 80 81 82 /* 83 * When a memory allocation must conform to specific limitations (such 84 * as being suitable for DMA) the caller will pass in hints to the 85 * allocator in the gfp_mask, in the zone modifier bits. These bits 86 * are used to select a priority ordered list of memory zones which 87 * match the requested limits. GFP_ZONEMASK defines which bits within 88 * the gfp_mask should be considered as zone modifiers. Each valid 89 * combination of the zone modifier bits has a corresponding list 90 * of zones (in node_zonelists). Thus for two zone modifiers there 91 * will be a maximum of 4 (2 ** 2) zonelists, for 3 modifiers there will 92 * be 8 (2 ** 3) zonelists. GFP_ZONETYPES defines the number of possible 93 * combinations of zone modifiers in "zone modifier space". 94 * 95 * As an optimisation any zone modifier bits which are only valid when 96 * no other zone modifier bits are set (loners) should be placed in 97 * the highest order bits of this field. This allows us to reduce the 98 * extent of the zonelists thus saving space. For example in the case 99 * of three zone modifier bits, we could require up to eight zonelists. 100 * If the left most zone modifier is a "loner" then the highest valid 101 * zonelist would be four allowing us to allocate only five zonelists. 102 * Use the first form for GFP_ZONETYPES when the left most bit is not 103 * a "loner", otherwise use the second. 104 * 105 * NOTE! Make sure this matches the zones in <linux/gfp.h> 106 */ 107 #define GFP_ZONEMASK 0x07 108 /* #define GFP_ZONETYPES (GFP_ZONEMASK + 1) */ /* Non-loner */ 109 #define GFP_ZONETYPES ((GFP_ZONEMASK + 1) / 2 + 1) /* Loner */ 110 111 /* 112 * On machines where it is needed (eg PCs) we divide physical memory 113 * into multiple physical zones. On a 32bit PC we have 4 zones: 114 * 115 * ZONE_DMA < 16 MB ISA DMA capable memory 116 * ZONE_DMA32 0 MB Empty 117 * ZONE_NORMAL 16-896 MB direct mapped by the kernel 118 * ZONE_HIGHMEM > 896 MB only page cache and user processes 119 */ 120 121 struct zone { 122 /* Fields commonly accessed by the page allocator */ 123 unsigned long free_pages; 124 unsigned long pages_min, pages_low, pages_high; 125 /* 126 * We don't know if the memory that we're going to allocate will be freeable 127 * or/and it will be released eventually, so to avoid totally wasting several 128 * GB of ram we must reserve some of the lower zone memory (otherwise we risk 129 * to run OOM on the lower zones despite there's tons of freeable ram 130 * on the higher zones). This array is recalculated at runtime if the 131 * sysctl_lowmem_reserve_ratio sysctl changes. 132 */ 133 unsigned long lowmem_reserve[MAX_NR_ZONES]; 134 135 #ifdef CONFIG_NUMA 136 struct per_cpu_pageset *pageset[NR_CPUS]; 137 #else 138 struct per_cpu_pageset pageset[NR_CPUS]; 139 #endif 140 /* 141 * free areas of different sizes 142 */ 143 spinlock_t lock; 144 #ifdef CONFIG_MEMORY_HOTPLUG 145 /* see spanned/present_pages for more description */ 146 seqlock_t span_seqlock; 147 #endif 148 struct free_area free_area[MAX_ORDER]; 149 150 151 ZONE_PADDING(_pad1_) 152 153 /* Fields commonly accessed by the page reclaim scanner */ 154 spinlock_t lru_lock; 155 struct list_head active_list; 156 struct list_head inactive_list; 157 unsigned long nr_scan_active; 158 unsigned long nr_scan_inactive; 159 unsigned long nr_active; 160 unsigned long nr_inactive; 161 unsigned long pages_scanned; /* since last reclaim */ 162 int all_unreclaimable; /* All pages pinned */ 163 164 /* A count of how many reclaimers are scanning this zone */ 165 atomic_t reclaim_in_progress; 166 167 /* 168 * timestamp (in jiffies) of the last zone reclaim that did not 169 * result in freeing of pages. This is used to avoid repeated scans 170 * if all memory in the zone is in use. 171 */ 172 unsigned long last_unsuccessful_zone_reclaim; 173 174 /* 175 * prev_priority holds the scanning priority for this zone. It is 176 * defined as the scanning priority at which we achieved our reclaim 177 * target at the previous try_to_free_pages() or balance_pgdat() 178 * invokation. 179 * 180 * We use prev_priority as a measure of how much stress page reclaim is 181 * under - it drives the swappiness decision: whether to unmap mapped 182 * pages. 183 * 184 * temp_priority is used to remember the scanning priority at which 185 * this zone was successfully refilled to free_pages == pages_high. 186 * 187 * Access to both these fields is quite racy even on uniprocessor. But 188 * it is expected to average out OK. 189 */ 190 int temp_priority; 191 int prev_priority; 192 193 194 ZONE_PADDING(_pad2_) 195 /* Rarely used or read-mostly fields */ 196 197 /* 198 * wait_table -- the array holding the hash table 199 * wait_table_size -- the size of the hash table array 200 * wait_table_bits -- wait_table_size == (1 << wait_table_bits) 201 * 202 * The purpose of all these is to keep track of the people 203 * waiting for a page to become available and make them 204 * runnable again when possible. The trouble is that this 205 * consumes a lot of space, especially when so few things 206 * wait on pages at a given time. So instead of using 207 * per-page waitqueues, we use a waitqueue hash table. 208 * 209 * The bucket discipline is to sleep on the same queue when 210 * colliding and wake all in that wait queue when removing. 211 * When something wakes, it must check to be sure its page is 212 * truly available, a la thundering herd. The cost of a 213 * collision is great, but given the expected load of the 214 * table, they should be so rare as to be outweighed by the 215 * benefits from the saved space. 216 * 217 * __wait_on_page_locked() and unlock_page() in mm/filemap.c, are the 218 * primary users of these fields, and in mm/page_alloc.c 219 * free_area_init_core() performs the initialization of them. 220 */ 221 wait_queue_head_t * wait_table; 222 unsigned long wait_table_size; 223 unsigned long wait_table_bits; 224 225 /* 226 * Discontig memory support fields. 227 */ 228 struct pglist_data *zone_pgdat; 229 /* zone_start_pfn == zone_start_paddr >> PAGE_SHIFT */ 230 unsigned long zone_start_pfn; 231 232 /* 233 * zone_start_pfn, spanned_pages and present_pages are all 234 * protected by span_seqlock. It is a seqlock because it has 235 * to be read outside of zone->lock, and it is done in the main 236 * allocator path. But, it is written quite infrequently. 237 * 238 * The lock is declared along with zone->lock because it is 239 * frequently read in proximity to zone->lock. It's good to 240 * give them a chance of being in the same cacheline. 241 */ 242 unsigned long spanned_pages; /* total size, including holes */ 243 unsigned long present_pages; /* amount of memory (excluding holes) */ 244 245 /* 246 * rarely used fields: 247 */ 248 char *name; 249 } ____cacheline_internodealigned_in_smp; 250 251 252 /* 253 * The "priority" of VM scanning is how much of the queues we will scan in one 254 * go. A value of 12 for DEF_PRIORITY implies that we will scan 1/4096th of the 255 * queues ("queue_length >> 12") during an aging round. 256 */ 257 #define DEF_PRIORITY 12 258 259 /* 260 * One allocation request operates on a zonelist. A zonelist 261 * is a list of zones, the first one is the 'goal' of the 262 * allocation, the other zones are fallback zones, in decreasing 263 * priority. 264 * 265 * Right now a zonelist takes up less than a cacheline. We never 266 * modify it apart from boot-up, and only a few indices are used, 267 * so despite the zonelist table being relatively big, the cache 268 * footprint of this construct is very small. 269 */ 270 struct zonelist { 271 struct zone *zones[MAX_NUMNODES * MAX_NR_ZONES + 1]; // NULL delimited 272 }; 273 274 275 /* 276 * The pg_data_t structure is used in machines with CONFIG_DISCONTIGMEM 277 * (mostly NUMA machines?) to denote a higher-level memory zone than the 278 * zone denotes. 279 * 280 * On NUMA machines, each NUMA node would have a pg_data_t to describe 281 * it's memory layout. 282 * 283 * Memory statistics and page replacement data structures are maintained on a 284 * per-zone basis. 285 */ 286 struct bootmem_data; 287 typedef struct pglist_data { 288 struct zone node_zones[MAX_NR_ZONES]; 289 struct zonelist node_zonelists[GFP_ZONETYPES]; 290 int nr_zones; 291 #ifdef CONFIG_FLAT_NODE_MEM_MAP 292 struct page *node_mem_map; 293 #endif 294 struct bootmem_data *bdata; 295 #ifdef CONFIG_MEMORY_HOTPLUG 296 /* 297 * Must be held any time you expect node_start_pfn, node_present_pages 298 * or node_spanned_pages stay constant. Holding this will also 299 * guarantee that any pfn_valid() stays that way. 300 * 301 * Nests above zone->lock and zone->size_seqlock. 302 */ 303 spinlock_t node_size_lock; 304 #endif 305 unsigned long node_start_pfn; 306 unsigned long node_present_pages; /* total number of physical pages */ 307 unsigned long node_spanned_pages; /* total size of physical page 308 range, including holes */ 309 int node_id; 310 wait_queue_head_t kswapd_wait; 311 struct task_struct *kswapd; 312 int kswapd_max_order; 313 } pg_data_t; 314 315 #define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) 316 #define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) 317 #ifdef CONFIG_FLAT_NODE_MEM_MAP 318 #define pgdat_page_nr(pgdat, pagenr) ((pgdat)->node_mem_map + (pagenr)) 319 #else 320 #define pgdat_page_nr(pgdat, pagenr) pfn_to_page((pgdat)->node_start_pfn + (pagenr)) 321 #endif 322 #define nid_page_nr(nid, pagenr) pgdat_page_nr(NODE_DATA(nid),(pagenr)) 323 324 #include <linux/memory_hotplug.h> 325 326 void __get_zone_counts(unsigned long *active, unsigned long *inactive, 327 unsigned long *free, struct pglist_data *pgdat); 328 void get_zone_counts(unsigned long *active, unsigned long *inactive, 329 unsigned long *free); 330 void build_all_zonelists(void); 331 void wakeup_kswapd(struct zone *zone, int order); 332 int zone_watermark_ok(struct zone *z, int order, unsigned long mark, 333 int classzone_idx, int alloc_flags); 334 335 #ifdef CONFIG_HAVE_MEMORY_PRESENT 336 void memory_present(int nid, unsigned long start, unsigned long end); 337 #else 338 static inline void memory_present(int nid, unsigned long start, unsigned long end) {} 339 #endif 340 341 #ifdef CONFIG_NEED_NODE_MEMMAP_SIZE 342 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); 343 #endif 344 345 /* 346 * zone_idx() returns 0 for the ZONE_DMA zone, 1 for the ZONE_NORMAL zone, etc. 347 */ 348 #define zone_idx(zone) ((zone) - (zone)->zone_pgdat->node_zones) 349 350 static inline int populated_zone(struct zone *zone) 351 { 352 return (!!zone->present_pages); 353 } 354 355 static inline int is_highmem_idx(int idx) 356 { 357 return (idx == ZONE_HIGHMEM); 358 } 359 360 static inline int is_normal_idx(int idx) 361 { 362 return (idx == ZONE_NORMAL); 363 } 364 365 /** 366 * is_highmem - helper function to quickly check if a struct zone is a 367 * highmem zone or not. This is an attempt to keep references 368 * to ZONE_{DMA/NORMAL/HIGHMEM/etc} in general code to a minimum. 369 * @zone - pointer to struct zone variable 370 */ 371 static inline int is_highmem(struct zone *zone) 372 { 373 return zone == zone->zone_pgdat->node_zones + ZONE_HIGHMEM; 374 } 375 376 static inline int is_normal(struct zone *zone) 377 { 378 return zone == zone->zone_pgdat->node_zones + ZONE_NORMAL; 379 } 380 381 static inline int is_dma32(struct zone *zone) 382 { 383 return zone == zone->zone_pgdat->node_zones + ZONE_DMA32; 384 } 385 386 static inline int is_dma(struct zone *zone) 387 { 388 return zone == zone->zone_pgdat->node_zones + ZONE_DMA; 389 } 390 391 /* These two functions are used to setup the per zone pages min values */ 392 struct ctl_table; 393 struct file; 394 int min_free_kbytes_sysctl_handler(struct ctl_table *, int, struct file *, 395 void __user *, size_t *, loff_t *); 396 extern int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1]; 397 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *, int, struct file *, 398 void __user *, size_t *, loff_t *); 399 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *, int, struct file *, 400 void __user *, size_t *, loff_t *); 401 402 #include <linux/topology.h> 403 /* Returns the number of the current Node. */ 404 #ifndef numa_node_id 405 #define numa_node_id() (cpu_to_node(raw_smp_processor_id())) 406 #endif 407 408 #ifndef CONFIG_NEED_MULTIPLE_NODES 409 410 extern struct pglist_data contig_page_data; 411 #define NODE_DATA(nid) (&contig_page_data) 412 #define NODE_MEM_MAP(nid) mem_map 413 #define MAX_NODES_SHIFT 1 414 415 #else /* CONFIG_NEED_MULTIPLE_NODES */ 416 417 #include <asm/mmzone.h> 418 419 #endif /* !CONFIG_NEED_MULTIPLE_NODES */ 420 421 extern struct pglist_data *first_online_pgdat(void); 422 extern struct pglist_data *next_online_pgdat(struct pglist_data *pgdat); 423 extern struct zone *next_zone(struct zone *zone); 424 425 /** 426 * for_each_pgdat - helper macro to iterate over all nodes 427 * @pgdat - pointer to a pg_data_t variable 428 */ 429 #define for_each_online_pgdat(pgdat) \ 430 for (pgdat = first_online_pgdat(); \ 431 pgdat; \ 432 pgdat = next_online_pgdat(pgdat)) 433 /** 434 * for_each_zone - helper macro to iterate over all memory zones 435 * @zone - pointer to struct zone variable 436 * 437 * The user only needs to declare the zone variable, for_each_zone 438 * fills it in. 439 */ 440 #define for_each_zone(zone) \ 441 for (zone = (first_online_pgdat())->node_zones; \ 442 zone; \ 443 zone = next_zone(zone)) 444 445 #ifdef CONFIG_SPARSEMEM 446 #include <asm/sparsemem.h> 447 #endif 448 449 #if BITS_PER_LONG == 32 450 /* 451 * with 32 bit page->flags field, we reserve 9 bits for node/zone info. 452 * there are 4 zones (3 bits) and this leaves 9-3=6 bits for nodes. 453 */ 454 #define FLAGS_RESERVED 9 455 456 #elif BITS_PER_LONG == 64 457 /* 458 * with 64 bit flags field, there's plenty of room. 459 */ 460 #define FLAGS_RESERVED 32 461 462 #else 463 464 #error BITS_PER_LONG not defined 465 466 #endif 467 468 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID 469 #define early_pfn_to_nid(nid) (0UL) 470 #endif 471 472 #ifdef CONFIG_FLATMEM 473 #define pfn_to_nid(pfn) (0) 474 #endif 475 476 #define pfn_to_section_nr(pfn) ((pfn) >> PFN_SECTION_SHIFT) 477 #define section_nr_to_pfn(sec) ((sec) << PFN_SECTION_SHIFT) 478 479 #ifdef CONFIG_SPARSEMEM 480 481 /* 482 * SECTION_SHIFT #bits space required to store a section # 483 * 484 * PA_SECTION_SHIFT physical address to/from section number 485 * PFN_SECTION_SHIFT pfn to/from section number 486 */ 487 #define SECTIONS_SHIFT (MAX_PHYSMEM_BITS - SECTION_SIZE_BITS) 488 489 #define PA_SECTION_SHIFT (SECTION_SIZE_BITS) 490 #define PFN_SECTION_SHIFT (SECTION_SIZE_BITS - PAGE_SHIFT) 491 492 #define NR_MEM_SECTIONS (1UL << SECTIONS_SHIFT) 493 494 #define PAGES_PER_SECTION (1UL << PFN_SECTION_SHIFT) 495 #define PAGE_SECTION_MASK (~(PAGES_PER_SECTION-1)) 496 497 #if (MAX_ORDER - 1 + PAGE_SHIFT) > SECTION_SIZE_BITS 498 #error Allocator MAX_ORDER exceeds SECTION_SIZE 499 #endif 500 501 struct page; 502 struct mem_section { 503 /* 504 * This is, logically, a pointer to an array of struct 505 * pages. However, it is stored with some other magic. 506 * (see sparse.c::sparse_init_one_section()) 507 * 508 * Making it a UL at least makes someone do a cast 509 * before using it wrong. 510 */ 511 unsigned long section_mem_map; 512 }; 513 514 #ifdef CONFIG_SPARSEMEM_EXTREME 515 #define SECTIONS_PER_ROOT (PAGE_SIZE / sizeof (struct mem_section)) 516 #else 517 #define SECTIONS_PER_ROOT 1 518 #endif 519 520 #define SECTION_NR_TO_ROOT(sec) ((sec) / SECTIONS_PER_ROOT) 521 #define NR_SECTION_ROOTS (NR_MEM_SECTIONS / SECTIONS_PER_ROOT) 522 #define SECTION_ROOT_MASK (SECTIONS_PER_ROOT - 1) 523 524 #ifdef CONFIG_SPARSEMEM_EXTREME 525 extern struct mem_section *mem_section[NR_SECTION_ROOTS]; 526 #else 527 extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]; 528 #endif 529 530 static inline struct mem_section *__nr_to_section(unsigned long nr) 531 { 532 if (!mem_section[SECTION_NR_TO_ROOT(nr)]) 533 return NULL; 534 return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; 535 } 536 extern int __section_nr(struct mem_section* ms); 537 538 /* 539 * We use the lower bits of the mem_map pointer to store 540 * a little bit of information. There should be at least 541 * 3 bits here due to 32-bit alignment. 542 */ 543 #define SECTION_MARKED_PRESENT (1UL<<0) 544 #define SECTION_HAS_MEM_MAP (1UL<<1) 545 #define SECTION_MAP_LAST_BIT (1UL<<2) 546 #define SECTION_MAP_MASK (~(SECTION_MAP_LAST_BIT-1)) 547 548 static inline struct page *__section_mem_map_addr(struct mem_section *section) 549 { 550 unsigned long map = section->section_mem_map; 551 map &= SECTION_MAP_MASK; 552 return (struct page *)map; 553 } 554 555 static inline int valid_section(struct mem_section *section) 556 { 557 return (section && (section->section_mem_map & SECTION_MARKED_PRESENT)); 558 } 559 560 static inline int section_has_mem_map(struct mem_section *section) 561 { 562 return (section && (section->section_mem_map & SECTION_HAS_MEM_MAP)); 563 } 564 565 static inline int valid_section_nr(unsigned long nr) 566 { 567 return valid_section(__nr_to_section(nr)); 568 } 569 570 static inline struct mem_section *__pfn_to_section(unsigned long pfn) 571 { 572 return __nr_to_section(pfn_to_section_nr(pfn)); 573 } 574 575 static inline int pfn_valid(unsigned long pfn) 576 { 577 if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS) 578 return 0; 579 return valid_section(__nr_to_section(pfn_to_section_nr(pfn))); 580 } 581 582 /* 583 * These are _only_ used during initialisation, therefore they 584 * can use __initdata ... They could have names to indicate 585 * this restriction. 586 */ 587 #ifdef CONFIG_NUMA 588 #define pfn_to_nid(pfn) \ 589 ({ \ 590 unsigned long __pfn_to_nid_pfn = (pfn); \ 591 page_to_nid(pfn_to_page(__pfn_to_nid_pfn)); \ 592 }) 593 #else 594 #define pfn_to_nid(pfn) (0) 595 #endif 596 597 #define early_pfn_valid(pfn) pfn_valid(pfn) 598 void sparse_init(void); 599 #else 600 #define sparse_init() do {} while (0) 601 #define sparse_index_init(_sec, _nid) do {} while (0) 602 #endif /* CONFIG_SPARSEMEM */ 603 604 #ifndef early_pfn_valid 605 #define early_pfn_valid(pfn) (1) 606 #endif 607 608 void memory_present(int nid, unsigned long start, unsigned long end); 609 unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long); 610 611 #endif /* !__ASSEMBLY__ */ 612 #endif /* __KERNEL__ */ 613 #endif /* _LINUX_MMZONE_H */ 614