[DECNET] ROUTE: remove unecessary alignment
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / mm / page_alloc.c
blob26a54a17dc9f5dfc0191da8f0dd774476ab8eee0
1 /*
2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
18 #include <linux/mm.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/oom.h>
31 #include <linux/notifier.h>
32 #include <linux/topology.h>
33 #include <linux/sysctl.h>
34 #include <linux/cpu.h>
35 #include <linux/cpuset.h>
36 #include <linux/memory_hotplug.h>
37 #include <linux/nodemask.h>
38 #include <linux/vmalloc.h>
39 #include <linux/mempolicy.h>
40 #include <linux/stop_machine.h>
41 #include <linux/sort.h>
42 #include <linux/pfn.h>
43 #include <linux/backing-dev.h>
44 #include <linux/fault-inject.h>
45 #include <linux/page-isolation.h>
46 #include <linux/memcontrol.h>
48 #include <asm/tlbflush.h>
49 #include <asm/div64.h>
50 #include "internal.h"
53 * Array of node states.
55 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
56 [N_POSSIBLE] = NODE_MASK_ALL,
57 [N_ONLINE] = { { [0] = 1UL } },
58 #ifndef CONFIG_NUMA
59 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
60 #ifdef CONFIG_HIGHMEM
61 [N_HIGH_MEMORY] = { { [0] = 1UL } },
62 #endif
63 [N_CPU] = { { [0] = 1UL } },
64 #endif /* NUMA */
66 EXPORT_SYMBOL(node_states);
68 unsigned long totalram_pages __read_mostly;
69 unsigned long totalreserve_pages __read_mostly;
70 long nr_swap_pages;
71 int percpu_pagelist_fraction;
73 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
74 int pageblock_order __read_mostly;
75 #endif
77 static void __free_pages_ok(struct page *page, unsigned int order);
80 * results with 256, 32 in the lowmem_reserve sysctl:
81 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
82 * 1G machine -> (16M dma, 784M normal, 224M high)
83 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
84 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
85 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
87 * TBD: should special case ZONE_DMA32 machines here - in those we normally
88 * don't need any ZONE_NORMAL reservation
90 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
91 #ifdef CONFIG_ZONE_DMA
92 256,
93 #endif
94 #ifdef CONFIG_ZONE_DMA32
95 256,
96 #endif
97 #ifdef CONFIG_HIGHMEM
98 32,
99 #endif
103 EXPORT_SYMBOL(totalram_pages);
105 static char * const zone_names[MAX_NR_ZONES] = {
106 #ifdef CONFIG_ZONE_DMA
107 "DMA",
108 #endif
109 #ifdef CONFIG_ZONE_DMA32
110 "DMA32",
111 #endif
112 "Normal",
113 #ifdef CONFIG_HIGHMEM
114 "HighMem",
115 #endif
116 "Movable",
119 int min_free_kbytes = 1024;
121 unsigned long __meminitdata nr_kernel_pages;
122 unsigned long __meminitdata nr_all_pages;
123 static unsigned long __meminitdata dma_reserve;
125 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
127 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
128 * ranges of memory (RAM) that may be registered with add_active_range().
129 * Ranges passed to add_active_range() will be merged if possible
130 * so the number of times add_active_range() can be called is
131 * related to the number of nodes and the number of holes
133 #ifdef CONFIG_MAX_ACTIVE_REGIONS
134 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
135 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
136 #else
137 #if MAX_NUMNODES >= 32
138 /* If there can be many nodes, allow up to 50 holes per node */
139 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
140 #else
141 /* By default, allow up to 256 distinct regions */
142 #define MAX_ACTIVE_REGIONS 256
143 #endif
144 #endif
146 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
147 static int __meminitdata nr_nodemap_entries;
148 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
149 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
150 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
151 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
152 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
153 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
154 unsigned long __initdata required_kernelcore;
155 static unsigned long __initdata required_movablecore;
156 unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
158 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
159 int movable_zone;
160 EXPORT_SYMBOL(movable_zone);
161 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163 #if MAX_NUMNODES > 1
164 int nr_node_ids __read_mostly = MAX_NUMNODES;
165 EXPORT_SYMBOL(nr_node_ids);
166 #endif
168 int page_group_by_mobility_disabled __read_mostly;
170 static void set_pageblock_migratetype(struct page *page, int migratetype)
172 set_pageblock_flags_group(page, (unsigned long)migratetype,
173 PB_migrate, PB_migrate_end);
176 #ifdef CONFIG_DEBUG_VM
177 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
179 int ret = 0;
180 unsigned seq;
181 unsigned long pfn = page_to_pfn(page);
183 do {
184 seq = zone_span_seqbegin(zone);
185 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
186 ret = 1;
187 else if (pfn < zone->zone_start_pfn)
188 ret = 1;
189 } while (zone_span_seqretry(zone, seq));
191 return ret;
194 static int page_is_consistent(struct zone *zone, struct page *page)
196 if (!pfn_valid_within(page_to_pfn(page)))
197 return 0;
198 if (zone != page_zone(page))
199 return 0;
201 return 1;
204 * Temporary debugging check for pages not lying within a given zone.
206 static int bad_range(struct zone *zone, struct page *page)
208 if (page_outside_zone_boundaries(zone, page))
209 return 1;
210 if (!page_is_consistent(zone, page))
211 return 1;
213 return 0;
215 #else
216 static inline int bad_range(struct zone *zone, struct page *page)
218 return 0;
220 #endif
222 static void bad_page(struct page *page)
224 printk(KERN_EMERG "Bad page state in process '%s'\n"
225 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
226 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
227 KERN_EMERG "Backtrace:\n",
228 current->comm, page, (int)(2*sizeof(unsigned long)),
229 (unsigned long)page->flags, page->mapping,
230 page_mapcount(page), page_count(page));
231 dump_stack();
232 page->flags &= ~(1 << PG_lru |
233 1 << PG_private |
234 1 << PG_locked |
235 1 << PG_active |
236 1 << PG_dirty |
237 1 << PG_reclaim |
238 1 << PG_slab |
239 1 << PG_swapcache |
240 1 << PG_writeback |
241 1 << PG_buddy );
242 set_page_count(page, 0);
243 reset_page_mapcount(page);
244 page->mapping = NULL;
245 add_taint(TAINT_BAD_PAGE);
249 * Higher-order pages are called "compound pages". They are structured thusly:
251 * The first PAGE_SIZE page is called the "head page".
253 * The remaining PAGE_SIZE pages are called "tail pages".
255 * All pages have PG_compound set. All pages have their ->private pointing at
256 * the head page (even the head page has this).
258 * The first tail page's ->lru.next holds the address of the compound page's
259 * put_page() function. Its ->lru.prev holds the order of allocation.
260 * This usage means that zero-order pages may not be compound.
263 static void free_compound_page(struct page *page)
265 __free_pages_ok(page, compound_order(page));
268 static void prep_compound_page(struct page *page, unsigned long order)
270 int i;
271 int nr_pages = 1 << order;
273 set_compound_page_dtor(page, free_compound_page);
274 set_compound_order(page, order);
275 __SetPageHead(page);
276 for (i = 1; i < nr_pages; i++) {
277 struct page *p = page + i;
279 __SetPageTail(p);
280 p->first_page = page;
284 static void destroy_compound_page(struct page *page, unsigned long order)
286 int i;
287 int nr_pages = 1 << order;
289 if (unlikely(compound_order(page) != order))
290 bad_page(page);
292 if (unlikely(!PageHead(page)))
293 bad_page(page);
294 __ClearPageHead(page);
295 for (i = 1; i < nr_pages; i++) {
296 struct page *p = page + i;
298 if (unlikely(!PageTail(p) |
299 (p->first_page != page)))
300 bad_page(page);
301 __ClearPageTail(p);
305 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
307 int i;
310 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
311 * and __GFP_HIGHMEM from hard or soft interrupt context.
313 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
314 for (i = 0; i < (1 << order); i++)
315 clear_highpage(page + i);
318 static inline void set_page_order(struct page *page, int order)
320 set_page_private(page, order);
321 __SetPageBuddy(page);
324 static inline void rmv_page_order(struct page *page)
326 __ClearPageBuddy(page);
327 set_page_private(page, 0);
331 * Locate the struct page for both the matching buddy in our
332 * pair (buddy1) and the combined O(n+1) page they form (page).
334 * 1) Any buddy B1 will have an order O twin B2 which satisfies
335 * the following equation:
336 * B2 = B1 ^ (1 << O)
337 * For example, if the starting buddy (buddy2) is #8 its order
338 * 1 buddy is #10:
339 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
341 * 2) Any buddy B will have an order O+1 parent P which
342 * satisfies the following equation:
343 * P = B & ~(1 << O)
345 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
347 static inline struct page *
348 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
350 unsigned long buddy_idx = page_idx ^ (1 << order);
352 return page + (buddy_idx - page_idx);
355 static inline unsigned long
356 __find_combined_index(unsigned long page_idx, unsigned int order)
358 return (page_idx & ~(1 << order));
362 * This function checks whether a page is free && is the buddy
363 * we can do coalesce a page and its buddy if
364 * (a) the buddy is not in a hole &&
365 * (b) the buddy is in the buddy system &&
366 * (c) a page and its buddy have the same order &&
367 * (d) a page and its buddy are in the same zone.
369 * For recording whether a page is in the buddy system, we use PG_buddy.
370 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
372 * For recording page's order, we use page_private(page).
374 static inline int page_is_buddy(struct page *page, struct page *buddy,
375 int order)
377 if (!pfn_valid_within(page_to_pfn(buddy)))
378 return 0;
380 if (page_zone_id(page) != page_zone_id(buddy))
381 return 0;
383 if (PageBuddy(buddy) && page_order(buddy) == order) {
384 BUG_ON(page_count(buddy) != 0);
385 return 1;
387 return 0;
391 * Freeing function for a buddy system allocator.
393 * The concept of a buddy system is to maintain direct-mapped table
394 * (containing bit values) for memory blocks of various "orders".
395 * The bottom level table contains the map for the smallest allocatable
396 * units of memory (here, pages), and each level above it describes
397 * pairs of units from the levels below, hence, "buddies".
398 * At a high level, all that happens here is marking the table entry
399 * at the bottom level available, and propagating the changes upward
400 * as necessary, plus some accounting needed to play nicely with other
401 * parts of the VM system.
402 * At each level, we keep a list of pages, which are heads of continuous
403 * free pages of length of (1 << order) and marked with PG_buddy. Page's
404 * order is recorded in page_private(page) field.
405 * So when we are allocating or freeing one, we can derive the state of the
406 * other. That is, if we allocate a small block, and both were
407 * free, the remainder of the region must be split into blocks.
408 * If a block is freed, and its buddy is also free, then this
409 * triggers coalescing into a block of larger size.
411 * -- wli
414 static inline void __free_one_page(struct page *page,
415 struct zone *zone, unsigned int order)
417 unsigned long page_idx;
418 int order_size = 1 << order;
419 int migratetype = get_pageblock_migratetype(page);
421 if (unlikely(PageCompound(page)))
422 destroy_compound_page(page, order);
424 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
426 VM_BUG_ON(page_idx & (order_size - 1));
427 VM_BUG_ON(bad_range(zone, page));
429 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
430 while (order < MAX_ORDER-1) {
431 unsigned long combined_idx;
432 struct page *buddy;
434 buddy = __page_find_buddy(page, page_idx, order);
435 if (!page_is_buddy(page, buddy, order))
436 break; /* Move the buddy up one level. */
438 list_del(&buddy->lru);
439 zone->free_area[order].nr_free--;
440 rmv_page_order(buddy);
441 combined_idx = __find_combined_index(page_idx, order);
442 page = page + (combined_idx - page_idx);
443 page_idx = combined_idx;
444 order++;
446 set_page_order(page, order);
447 list_add(&page->lru,
448 &zone->free_area[order].free_list[migratetype]);
449 zone->free_area[order].nr_free++;
452 static inline int free_pages_check(struct page *page)
454 if (unlikely(page_mapcount(page) |
455 (page->mapping != NULL) |
456 (page_count(page) != 0) |
457 (page->flags & (
458 1 << PG_lru |
459 1 << PG_private |
460 1 << PG_locked |
461 1 << PG_active |
462 1 << PG_slab |
463 1 << PG_swapcache |
464 1 << PG_writeback |
465 1 << PG_reserved |
466 1 << PG_buddy ))))
467 bad_page(page);
468 if (PageDirty(page))
469 __ClearPageDirty(page);
471 * For now, we report if PG_reserved was found set, but do not
472 * clear it, and do not free the page. But we shall soon need
473 * to do more, for when the ZERO_PAGE count wraps negative.
475 return PageReserved(page);
479 * Frees a list of pages.
480 * Assumes all pages on list are in same zone, and of same order.
481 * count is the number of pages to free.
483 * If the zone was previously in an "all pages pinned" state then look to
484 * see if this freeing clears that state.
486 * And clear the zone's pages_scanned counter, to hold off the "all pages are
487 * pinned" detection logic.
489 static void free_pages_bulk(struct zone *zone, int count,
490 struct list_head *list, int order)
492 spin_lock(&zone->lock);
493 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
494 zone->pages_scanned = 0;
495 while (count--) {
496 struct page *page;
498 VM_BUG_ON(list_empty(list));
499 page = list_entry(list->prev, struct page, lru);
500 /* have to delete it as __free_one_page list manipulates */
501 list_del(&page->lru);
502 __free_one_page(page, zone, order);
504 spin_unlock(&zone->lock);
507 static void free_one_page(struct zone *zone, struct page *page, int order)
509 spin_lock(&zone->lock);
510 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
511 zone->pages_scanned = 0;
512 __free_one_page(page, zone, order);
513 spin_unlock(&zone->lock);
516 static void __free_pages_ok(struct page *page, unsigned int order)
518 unsigned long flags;
519 int i;
520 int reserved = 0;
522 for (i = 0 ; i < (1 << order) ; ++i)
523 reserved += free_pages_check(page + i);
524 if (reserved)
525 return;
527 if (!PageHighMem(page))
528 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
529 arch_free_page(page, order);
530 kernel_map_pages(page, 1 << order, 0);
532 local_irq_save(flags);
533 __count_vm_events(PGFREE, 1 << order);
534 free_one_page(page_zone(page), page, order);
535 local_irq_restore(flags);
539 * permit the bootmem allocator to evade page validation on high-order frees
541 void __init __free_pages_bootmem(struct page *page, unsigned int order)
543 if (order == 0) {
544 __ClearPageReserved(page);
545 set_page_count(page, 0);
546 set_page_refcounted(page);
547 __free_page(page);
548 } else {
549 int loop;
551 prefetchw(page);
552 for (loop = 0; loop < BITS_PER_LONG; loop++) {
553 struct page *p = &page[loop];
555 if (loop + 1 < BITS_PER_LONG)
556 prefetchw(p + 1);
557 __ClearPageReserved(p);
558 set_page_count(p, 0);
561 set_page_refcounted(page);
562 __free_pages(page, order);
568 * The order of subdivision here is critical for the IO subsystem.
569 * Please do not alter this order without good reasons and regression
570 * testing. Specifically, as large blocks of memory are subdivided,
571 * the order in which smaller blocks are delivered depends on the order
572 * they're subdivided in this function. This is the primary factor
573 * influencing the order in which pages are delivered to the IO
574 * subsystem according to empirical testing, and this is also justified
575 * by considering the behavior of a buddy system containing a single
576 * large block of memory acted on by a series of small allocations.
577 * This behavior is a critical factor in sglist merging's success.
579 * -- wli
581 static inline void expand(struct zone *zone, struct page *page,
582 int low, int high, struct free_area *area,
583 int migratetype)
585 unsigned long size = 1 << high;
587 while (high > low) {
588 area--;
589 high--;
590 size >>= 1;
591 VM_BUG_ON(bad_range(zone, &page[size]));
592 list_add(&page[size].lru, &area->free_list[migratetype]);
593 area->nr_free++;
594 set_page_order(&page[size], high);
599 * This page is about to be returned from the page allocator
601 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
603 if (unlikely(page_mapcount(page) |
604 (page->mapping != NULL) |
605 (page_count(page) != 0) |
606 (page->flags & (
607 1 << PG_lru |
608 1 << PG_private |
609 1 << PG_locked |
610 1 << PG_active |
611 1 << PG_dirty |
612 1 << PG_slab |
613 1 << PG_swapcache |
614 1 << PG_writeback |
615 1 << PG_reserved |
616 1 << PG_buddy ))))
617 bad_page(page);
620 * For now, we report if PG_reserved was found set, but do not
621 * clear it, and do not allocate the page: as a safety net.
623 if (PageReserved(page))
624 return 1;
626 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
627 1 << PG_referenced | 1 << PG_arch_1 |
628 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
629 set_page_private(page, 0);
630 set_page_refcounted(page);
632 arch_alloc_page(page, order);
633 kernel_map_pages(page, 1 << order, 1);
635 if (gfp_flags & __GFP_ZERO)
636 prep_zero_page(page, order, gfp_flags);
638 if (order && (gfp_flags & __GFP_COMP))
639 prep_compound_page(page, order);
641 return 0;
645 * Go through the free lists for the given migratetype and remove
646 * the smallest available page from the freelists
648 static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
649 int migratetype)
651 unsigned int current_order;
652 struct free_area * area;
653 struct page *page;
655 /* Find a page of the appropriate size in the preferred list */
656 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
657 area = &(zone->free_area[current_order]);
658 if (list_empty(&area->free_list[migratetype]))
659 continue;
661 page = list_entry(area->free_list[migratetype].next,
662 struct page, lru);
663 list_del(&page->lru);
664 rmv_page_order(page);
665 area->nr_free--;
666 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
667 expand(zone, page, order, current_order, area, migratetype);
668 return page;
671 return NULL;
676 * This array describes the order lists are fallen back to when
677 * the free lists for the desirable migrate type are depleted
679 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
680 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
681 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
682 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
683 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
687 * Move the free pages in a range to the free lists of the requested type.
688 * Note that start_page and end_pages are not aligned on a pageblock
689 * boundary. If alignment is required, use move_freepages_block()
691 int move_freepages(struct zone *zone,
692 struct page *start_page, struct page *end_page,
693 int migratetype)
695 struct page *page;
696 unsigned long order;
697 int pages_moved = 0;
699 #ifndef CONFIG_HOLES_IN_ZONE
701 * page_zone is not safe to call in this context when
702 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
703 * anyway as we check zone boundaries in move_freepages_block().
704 * Remove at a later date when no bug reports exist related to
705 * grouping pages by mobility
707 BUG_ON(page_zone(start_page) != page_zone(end_page));
708 #endif
710 for (page = start_page; page <= end_page;) {
711 if (!pfn_valid_within(page_to_pfn(page))) {
712 page++;
713 continue;
716 if (!PageBuddy(page)) {
717 page++;
718 continue;
721 order = page_order(page);
722 list_del(&page->lru);
723 list_add(&page->lru,
724 &zone->free_area[order].free_list[migratetype]);
725 page += 1 << order;
726 pages_moved += 1 << order;
729 return pages_moved;
732 int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
734 unsigned long start_pfn, end_pfn;
735 struct page *start_page, *end_page;
737 start_pfn = page_to_pfn(page);
738 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
739 start_page = pfn_to_page(start_pfn);
740 end_page = start_page + pageblock_nr_pages - 1;
741 end_pfn = start_pfn + pageblock_nr_pages - 1;
743 /* Do not cross zone boundaries */
744 if (start_pfn < zone->zone_start_pfn)
745 start_page = page;
746 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
747 return 0;
749 return move_freepages(zone, start_page, end_page, migratetype);
752 /* Remove an element from the buddy allocator from the fallback list */
753 static struct page *__rmqueue_fallback(struct zone *zone, int order,
754 int start_migratetype)
756 struct free_area * area;
757 int current_order;
758 struct page *page;
759 int migratetype, i;
761 /* Find the largest possible block of pages in the other list */
762 for (current_order = MAX_ORDER-1; current_order >= order;
763 --current_order) {
764 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
765 migratetype = fallbacks[start_migratetype][i];
767 /* MIGRATE_RESERVE handled later if necessary */
768 if (migratetype == MIGRATE_RESERVE)
769 continue;
771 area = &(zone->free_area[current_order]);
772 if (list_empty(&area->free_list[migratetype]))
773 continue;
775 page = list_entry(area->free_list[migratetype].next,
776 struct page, lru);
777 area->nr_free--;
780 * If breaking a large block of pages, move all free
781 * pages to the preferred allocation list. If falling
782 * back for a reclaimable kernel allocation, be more
783 * agressive about taking ownership of free pages
785 if (unlikely(current_order >= (pageblock_order >> 1)) ||
786 start_migratetype == MIGRATE_RECLAIMABLE) {
787 unsigned long pages;
788 pages = move_freepages_block(zone, page,
789 start_migratetype);
791 /* Claim the whole block if over half of it is free */
792 if (pages >= (1 << (pageblock_order-1)))
793 set_pageblock_migratetype(page,
794 start_migratetype);
796 migratetype = start_migratetype;
799 /* Remove the page from the freelists */
800 list_del(&page->lru);
801 rmv_page_order(page);
802 __mod_zone_page_state(zone, NR_FREE_PAGES,
803 -(1UL << order));
805 if (current_order == pageblock_order)
806 set_pageblock_migratetype(page,
807 start_migratetype);
809 expand(zone, page, order, current_order, area, migratetype);
810 return page;
814 /* Use MIGRATE_RESERVE rather than fail an allocation */
815 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
819 * Do the hard work of removing an element from the buddy allocator.
820 * Call me with the zone->lock already held.
822 static struct page *__rmqueue(struct zone *zone, unsigned int order,
823 int migratetype)
825 struct page *page;
827 page = __rmqueue_smallest(zone, order, migratetype);
829 if (unlikely(!page))
830 page = __rmqueue_fallback(zone, order, migratetype);
832 return page;
836 * Obtain a specified number of elements from the buddy allocator, all under
837 * a single hold of the lock, for efficiency. Add them to the supplied list.
838 * Returns the number of new pages which were placed at *list.
840 static int rmqueue_bulk(struct zone *zone, unsigned int order,
841 unsigned long count, struct list_head *list,
842 int migratetype)
844 int i;
846 spin_lock(&zone->lock);
847 for (i = 0; i < count; ++i) {
848 struct page *page = __rmqueue(zone, order, migratetype);
849 if (unlikely(page == NULL))
850 break;
853 * Split buddy pages returned by expand() are received here
854 * in physical page order. The page is added to the callers and
855 * list and the list head then moves forward. From the callers
856 * perspective, the linked list is ordered by page number in
857 * some conditions. This is useful for IO devices that can
858 * merge IO requests if the physical pages are ordered
859 * properly.
861 list_add(&page->lru, list);
862 set_page_private(page, migratetype);
863 list = &page->lru;
865 spin_unlock(&zone->lock);
866 return i;
869 #ifdef CONFIG_NUMA
871 * Called from the vmstat counter updater to drain pagesets of this
872 * currently executing processor on remote nodes after they have
873 * expired.
875 * Note that this function must be called with the thread pinned to
876 * a single processor.
878 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
880 unsigned long flags;
881 int to_drain;
883 local_irq_save(flags);
884 if (pcp->count >= pcp->batch)
885 to_drain = pcp->batch;
886 else
887 to_drain = pcp->count;
888 free_pages_bulk(zone, to_drain, &pcp->list, 0);
889 pcp->count -= to_drain;
890 local_irq_restore(flags);
892 #endif
895 * Drain pages of the indicated processor.
897 * The processor must either be the current processor and the
898 * thread pinned to the current processor or a processor that
899 * is not online.
901 static void drain_pages(unsigned int cpu)
903 unsigned long flags;
904 struct zone *zone;
906 for_each_zone(zone) {
907 struct per_cpu_pageset *pset;
908 struct per_cpu_pages *pcp;
910 if (!populated_zone(zone))
911 continue;
913 pset = zone_pcp(zone, cpu);
915 pcp = &pset->pcp;
916 local_irq_save(flags);
917 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
918 pcp->count = 0;
919 local_irq_restore(flags);
924 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
926 void drain_local_pages(void *arg)
928 drain_pages(smp_processor_id());
932 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
934 void drain_all_pages(void)
936 on_each_cpu(drain_local_pages, NULL, 0, 1);
939 #ifdef CONFIG_HIBERNATION
941 void mark_free_pages(struct zone *zone)
943 unsigned long pfn, max_zone_pfn;
944 unsigned long flags;
945 int order, t;
946 struct list_head *curr;
948 if (!zone->spanned_pages)
949 return;
951 spin_lock_irqsave(&zone->lock, flags);
953 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
954 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
955 if (pfn_valid(pfn)) {
956 struct page *page = pfn_to_page(pfn);
958 if (!swsusp_page_is_forbidden(page))
959 swsusp_unset_page_free(page);
962 for_each_migratetype_order(order, t) {
963 list_for_each(curr, &zone->free_area[order].free_list[t]) {
964 unsigned long i;
966 pfn = page_to_pfn(list_entry(curr, struct page, lru));
967 for (i = 0; i < (1UL << order); i++)
968 swsusp_set_page_free(pfn_to_page(pfn + i));
971 spin_unlock_irqrestore(&zone->lock, flags);
973 #endif /* CONFIG_PM */
976 * Free a 0-order page
978 static void free_hot_cold_page(struct page *page, int cold)
980 struct zone *zone = page_zone(page);
981 struct per_cpu_pages *pcp;
982 unsigned long flags;
984 if (PageAnon(page))
985 page->mapping = NULL;
986 if (free_pages_check(page))
987 return;
989 if (!PageHighMem(page))
990 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
991 VM_BUG_ON(page_get_page_cgroup(page));
992 arch_free_page(page, 0);
993 kernel_map_pages(page, 1, 0);
995 pcp = &zone_pcp(zone, get_cpu())->pcp;
996 local_irq_save(flags);
997 __count_vm_event(PGFREE);
998 if (cold)
999 list_add_tail(&page->lru, &pcp->list);
1000 else
1001 list_add(&page->lru, &pcp->list);
1002 set_page_private(page, get_pageblock_migratetype(page));
1003 pcp->count++;
1004 if (pcp->count >= pcp->high) {
1005 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1006 pcp->count -= pcp->batch;
1008 local_irq_restore(flags);
1009 put_cpu();
1012 void free_hot_page(struct page *page)
1014 free_hot_cold_page(page, 0);
1017 void free_cold_page(struct page *page)
1019 free_hot_cold_page(page, 1);
1023 * split_page takes a non-compound higher-order page, and splits it into
1024 * n (1<<order) sub-pages: page[0..n]
1025 * Each sub-page must be freed individually.
1027 * Note: this is probably too low level an operation for use in drivers.
1028 * Please consult with lkml before using this in your driver.
1030 void split_page(struct page *page, unsigned int order)
1032 int i;
1034 VM_BUG_ON(PageCompound(page));
1035 VM_BUG_ON(!page_count(page));
1036 for (i = 1; i < (1 << order); i++)
1037 set_page_refcounted(page + i);
1041 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1042 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1043 * or two.
1045 static struct page *buffered_rmqueue(struct zonelist *zonelist,
1046 struct zone *zone, int order, gfp_t gfp_flags)
1048 unsigned long flags;
1049 struct page *page;
1050 int cold = !!(gfp_flags & __GFP_COLD);
1051 int cpu;
1052 int migratetype = allocflags_to_migratetype(gfp_flags);
1054 again:
1055 cpu = get_cpu();
1056 if (likely(order == 0)) {
1057 struct per_cpu_pages *pcp;
1059 pcp = &zone_pcp(zone, cpu)->pcp;
1060 local_irq_save(flags);
1061 if (!pcp->count) {
1062 pcp->count = rmqueue_bulk(zone, 0,
1063 pcp->batch, &pcp->list, migratetype);
1064 if (unlikely(!pcp->count))
1065 goto failed;
1068 /* Find a page of the appropriate migrate type */
1069 if (cold) {
1070 list_for_each_entry_reverse(page, &pcp->list, lru)
1071 if (page_private(page) == migratetype)
1072 break;
1073 } else {
1074 list_for_each_entry(page, &pcp->list, lru)
1075 if (page_private(page) == migratetype)
1076 break;
1079 /* Allocate more to the pcp list if necessary */
1080 if (unlikely(&page->lru == &pcp->list)) {
1081 pcp->count += rmqueue_bulk(zone, 0,
1082 pcp->batch, &pcp->list, migratetype);
1083 page = list_entry(pcp->list.next, struct page, lru);
1086 list_del(&page->lru);
1087 pcp->count--;
1088 } else {
1089 spin_lock_irqsave(&zone->lock, flags);
1090 page = __rmqueue(zone, order, migratetype);
1091 spin_unlock(&zone->lock);
1092 if (!page)
1093 goto failed;
1096 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1097 zone_statistics(zonelist, zone);
1098 local_irq_restore(flags);
1099 put_cpu();
1101 VM_BUG_ON(bad_range(zone, page));
1102 if (prep_new_page(page, order, gfp_flags))
1103 goto again;
1104 return page;
1106 failed:
1107 local_irq_restore(flags);
1108 put_cpu();
1109 return NULL;
1112 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1113 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1114 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1115 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1116 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1117 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1118 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1120 #ifdef CONFIG_FAIL_PAGE_ALLOC
1122 static struct fail_page_alloc_attr {
1123 struct fault_attr attr;
1125 u32 ignore_gfp_highmem;
1126 u32 ignore_gfp_wait;
1127 u32 min_order;
1129 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1131 struct dentry *ignore_gfp_highmem_file;
1132 struct dentry *ignore_gfp_wait_file;
1133 struct dentry *min_order_file;
1135 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1137 } fail_page_alloc = {
1138 .attr = FAULT_ATTR_INITIALIZER,
1139 .ignore_gfp_wait = 1,
1140 .ignore_gfp_highmem = 1,
1141 .min_order = 1,
1144 static int __init setup_fail_page_alloc(char *str)
1146 return setup_fault_attr(&fail_page_alloc.attr, str);
1148 __setup("fail_page_alloc=", setup_fail_page_alloc);
1150 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1152 if (order < fail_page_alloc.min_order)
1153 return 0;
1154 if (gfp_mask & __GFP_NOFAIL)
1155 return 0;
1156 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1157 return 0;
1158 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1159 return 0;
1161 return should_fail(&fail_page_alloc.attr, 1 << order);
1164 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1166 static int __init fail_page_alloc_debugfs(void)
1168 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1169 struct dentry *dir;
1170 int err;
1172 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1173 "fail_page_alloc");
1174 if (err)
1175 return err;
1176 dir = fail_page_alloc.attr.dentries.dir;
1178 fail_page_alloc.ignore_gfp_wait_file =
1179 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1180 &fail_page_alloc.ignore_gfp_wait);
1182 fail_page_alloc.ignore_gfp_highmem_file =
1183 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1184 &fail_page_alloc.ignore_gfp_highmem);
1185 fail_page_alloc.min_order_file =
1186 debugfs_create_u32("min-order", mode, dir,
1187 &fail_page_alloc.min_order);
1189 if (!fail_page_alloc.ignore_gfp_wait_file ||
1190 !fail_page_alloc.ignore_gfp_highmem_file ||
1191 !fail_page_alloc.min_order_file) {
1192 err = -ENOMEM;
1193 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1194 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1195 debugfs_remove(fail_page_alloc.min_order_file);
1196 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1199 return err;
1202 late_initcall(fail_page_alloc_debugfs);
1204 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1206 #else /* CONFIG_FAIL_PAGE_ALLOC */
1208 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1210 return 0;
1213 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1216 * Return 1 if free pages are above 'mark'. This takes into account the order
1217 * of the allocation.
1219 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1220 int classzone_idx, int alloc_flags)
1222 /* free_pages my go negative - that's OK */
1223 long min = mark;
1224 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1225 int o;
1227 if (alloc_flags & ALLOC_HIGH)
1228 min -= min / 2;
1229 if (alloc_flags & ALLOC_HARDER)
1230 min -= min / 4;
1232 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1233 return 0;
1234 for (o = 0; o < order; o++) {
1235 /* At the next order, this order's pages become unavailable */
1236 free_pages -= z->free_area[o].nr_free << o;
1238 /* Require fewer higher order pages to be free */
1239 min >>= 1;
1241 if (free_pages <= min)
1242 return 0;
1244 return 1;
1247 #ifdef CONFIG_NUMA
1249 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1250 * skip over zones that are not allowed by the cpuset, or that have
1251 * been recently (in last second) found to be nearly full. See further
1252 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1253 * that have to skip over a lot of full or unallowed zones.
1255 * If the zonelist cache is present in the passed in zonelist, then
1256 * returns a pointer to the allowed node mask (either the current
1257 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1259 * If the zonelist cache is not available for this zonelist, does
1260 * nothing and returns NULL.
1262 * If the fullzones BITMAP in the zonelist cache is stale (more than
1263 * a second since last zap'd) then we zap it out (clear its bits.)
1265 * We hold off even calling zlc_setup, until after we've checked the
1266 * first zone in the zonelist, on the theory that most allocations will
1267 * be satisfied from that first zone, so best to examine that zone as
1268 * quickly as we can.
1270 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1272 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1273 nodemask_t *allowednodes; /* zonelist_cache approximation */
1275 zlc = zonelist->zlcache_ptr;
1276 if (!zlc)
1277 return NULL;
1279 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1280 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1281 zlc->last_full_zap = jiffies;
1284 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1285 &cpuset_current_mems_allowed :
1286 &node_states[N_HIGH_MEMORY];
1287 return allowednodes;
1291 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1292 * if it is worth looking at further for free memory:
1293 * 1) Check that the zone isn't thought to be full (doesn't have its
1294 * bit set in the zonelist_cache fullzones BITMAP).
1295 * 2) Check that the zones node (obtained from the zonelist_cache
1296 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1297 * Return true (non-zero) if zone is worth looking at further, or
1298 * else return false (zero) if it is not.
1300 * This check -ignores- the distinction between various watermarks,
1301 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1302 * found to be full for any variation of these watermarks, it will
1303 * be considered full for up to one second by all requests, unless
1304 * we are so low on memory on all allowed nodes that we are forced
1305 * into the second scan of the zonelist.
1307 * In the second scan we ignore this zonelist cache and exactly
1308 * apply the watermarks to all zones, even it is slower to do so.
1309 * We are low on memory in the second scan, and should leave no stone
1310 * unturned looking for a free page.
1312 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1313 nodemask_t *allowednodes)
1315 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1316 int i; /* index of *z in zonelist zones */
1317 int n; /* node that zone *z is on */
1319 zlc = zonelist->zlcache_ptr;
1320 if (!zlc)
1321 return 1;
1323 i = z - zonelist->zones;
1324 n = zlc->z_to_n[i];
1326 /* This zone is worth trying if it is allowed but not full */
1327 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1331 * Given 'z' scanning a zonelist, set the corresponding bit in
1332 * zlc->fullzones, so that subsequent attempts to allocate a page
1333 * from that zone don't waste time re-examining it.
1335 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1337 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1338 int i; /* index of *z in zonelist zones */
1340 zlc = zonelist->zlcache_ptr;
1341 if (!zlc)
1342 return;
1344 i = z - zonelist->zones;
1346 set_bit(i, zlc->fullzones);
1349 #else /* CONFIG_NUMA */
1351 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1353 return NULL;
1356 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1357 nodemask_t *allowednodes)
1359 return 1;
1362 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1365 #endif /* CONFIG_NUMA */
1368 * get_page_from_freelist goes through the zonelist trying to allocate
1369 * a page.
1371 static struct page *
1372 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1373 struct zonelist *zonelist, int alloc_flags)
1375 struct zone **z;
1376 struct page *page = NULL;
1377 int classzone_idx = zone_idx(zonelist->zones[0]);
1378 struct zone *zone;
1379 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1380 int zlc_active = 0; /* set if using zonelist_cache */
1381 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1382 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
1384 zonelist_scan:
1386 * Scan zonelist, looking for a zone with enough free.
1387 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1389 z = zonelist->zones;
1391 do {
1393 * In NUMA, this could be a policy zonelist which contains
1394 * zones that may not be allowed by the current gfp_mask.
1395 * Check the zone is allowed by the current flags
1397 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1398 if (highest_zoneidx == -1)
1399 highest_zoneidx = gfp_zone(gfp_mask);
1400 if (zone_idx(*z) > highest_zoneidx)
1401 continue;
1404 if (NUMA_BUILD && zlc_active &&
1405 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1406 continue;
1407 zone = *z;
1408 if ((alloc_flags & ALLOC_CPUSET) &&
1409 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1410 goto try_next_zone;
1412 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1413 unsigned long mark;
1414 if (alloc_flags & ALLOC_WMARK_MIN)
1415 mark = zone->pages_min;
1416 else if (alloc_flags & ALLOC_WMARK_LOW)
1417 mark = zone->pages_low;
1418 else
1419 mark = zone->pages_high;
1420 if (!zone_watermark_ok(zone, order, mark,
1421 classzone_idx, alloc_flags)) {
1422 if (!zone_reclaim_mode ||
1423 !zone_reclaim(zone, gfp_mask, order))
1424 goto this_zone_full;
1428 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1429 if (page)
1430 break;
1431 this_zone_full:
1432 if (NUMA_BUILD)
1433 zlc_mark_zone_full(zonelist, z);
1434 try_next_zone:
1435 if (NUMA_BUILD && !did_zlc_setup) {
1436 /* we do zlc_setup after the first zone is tried */
1437 allowednodes = zlc_setup(zonelist, alloc_flags);
1438 zlc_active = 1;
1439 did_zlc_setup = 1;
1441 } while (*(++z) != NULL);
1443 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1444 /* Disable zlc cache for second zonelist scan */
1445 zlc_active = 0;
1446 goto zonelist_scan;
1448 return page;
1452 * This is the 'heart' of the zoned buddy allocator.
1454 struct page * fastcall
1455 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1456 struct zonelist *zonelist)
1458 const gfp_t wait = gfp_mask & __GFP_WAIT;
1459 struct zone **z;
1460 struct page *page;
1461 struct reclaim_state reclaim_state;
1462 struct task_struct *p = current;
1463 int do_retry;
1464 int alloc_flags;
1465 int did_some_progress;
1467 might_sleep_if(wait);
1469 if (should_fail_alloc_page(gfp_mask, order))
1470 return NULL;
1472 restart:
1473 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
1475 if (unlikely(*z == NULL)) {
1477 * Happens if we have an empty zonelist as a result of
1478 * GFP_THISNODE being used on a memoryless node
1480 return NULL;
1483 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1484 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1485 if (page)
1486 goto got_pg;
1489 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1490 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1491 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1492 * using a larger set of nodes after it has established that the
1493 * allowed per node queues are empty and that nodes are
1494 * over allocated.
1496 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1497 goto nopage;
1499 for (z = zonelist->zones; *z; z++)
1500 wakeup_kswapd(*z, order);
1503 * OK, we're below the kswapd watermark and have kicked background
1504 * reclaim. Now things get more complex, so set up alloc_flags according
1505 * to how we want to proceed.
1507 * The caller may dip into page reserves a bit more if the caller
1508 * cannot run direct reclaim, or if the caller has realtime scheduling
1509 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1510 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1512 alloc_flags = ALLOC_WMARK_MIN;
1513 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1514 alloc_flags |= ALLOC_HARDER;
1515 if (gfp_mask & __GFP_HIGH)
1516 alloc_flags |= ALLOC_HIGH;
1517 if (wait)
1518 alloc_flags |= ALLOC_CPUSET;
1521 * Go through the zonelist again. Let __GFP_HIGH and allocations
1522 * coming from realtime tasks go deeper into reserves.
1524 * This is the last chance, in general, before the goto nopage.
1525 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1526 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1528 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1529 if (page)
1530 goto got_pg;
1532 /* This allocation should allow future memory freeing. */
1534 rebalance:
1535 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1536 && !in_interrupt()) {
1537 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1538 nofail_alloc:
1539 /* go through the zonelist yet again, ignoring mins */
1540 page = get_page_from_freelist(gfp_mask, order,
1541 zonelist, ALLOC_NO_WATERMARKS);
1542 if (page)
1543 goto got_pg;
1544 if (gfp_mask & __GFP_NOFAIL) {
1545 congestion_wait(WRITE, HZ/50);
1546 goto nofail_alloc;
1549 goto nopage;
1552 /* Atomic allocations - we can't balance anything */
1553 if (!wait)
1554 goto nopage;
1556 cond_resched();
1558 /* We now go into synchronous reclaim */
1559 cpuset_memory_pressure_bump();
1560 p->flags |= PF_MEMALLOC;
1561 reclaim_state.reclaimed_slab = 0;
1562 p->reclaim_state = &reclaim_state;
1564 did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
1566 p->reclaim_state = NULL;
1567 p->flags &= ~PF_MEMALLOC;
1569 cond_resched();
1571 if (order != 0)
1572 drain_all_pages();
1574 if (likely(did_some_progress)) {
1575 page = get_page_from_freelist(gfp_mask, order,
1576 zonelist, alloc_flags);
1577 if (page)
1578 goto got_pg;
1579 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1580 if (!try_set_zone_oom(zonelist)) {
1581 schedule_timeout_uninterruptible(1);
1582 goto restart;
1586 * Go through the zonelist yet one more time, keep
1587 * very high watermark here, this is only to catch
1588 * a parallel oom killing, we must fail if we're still
1589 * under heavy pressure.
1591 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1592 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1593 if (page) {
1594 clear_zonelist_oom(zonelist);
1595 goto got_pg;
1598 /* The OOM killer will not help higher order allocs so fail */
1599 if (order > PAGE_ALLOC_COSTLY_ORDER) {
1600 clear_zonelist_oom(zonelist);
1601 goto nopage;
1604 out_of_memory(zonelist, gfp_mask, order);
1605 clear_zonelist_oom(zonelist);
1606 goto restart;
1610 * Don't let big-order allocations loop unless the caller explicitly
1611 * requests that. Wait for some write requests to complete then retry.
1613 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1614 * <= 3, but that may not be true in other implementations.
1616 do_retry = 0;
1617 if (!(gfp_mask & __GFP_NORETRY)) {
1618 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1619 (gfp_mask & __GFP_REPEAT))
1620 do_retry = 1;
1621 if (gfp_mask & __GFP_NOFAIL)
1622 do_retry = 1;
1624 if (do_retry) {
1625 congestion_wait(WRITE, HZ/50);
1626 goto rebalance;
1629 nopage:
1630 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1631 printk(KERN_WARNING "%s: page allocation failure."
1632 " order:%d, mode:0x%x\n",
1633 p->comm, order, gfp_mask);
1634 dump_stack();
1635 show_mem();
1637 got_pg:
1638 return page;
1641 EXPORT_SYMBOL(__alloc_pages);
1644 * Common helper functions.
1646 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1648 struct page * page;
1649 page = alloc_pages(gfp_mask, order);
1650 if (!page)
1651 return 0;
1652 return (unsigned long) page_address(page);
1655 EXPORT_SYMBOL(__get_free_pages);
1657 unsigned long get_zeroed_page(gfp_t gfp_mask)
1659 struct page * page;
1662 * get_zeroed_page() returns a 32-bit address, which cannot represent
1663 * a highmem page
1665 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1667 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1668 if (page)
1669 return (unsigned long) page_address(page);
1670 return 0;
1673 EXPORT_SYMBOL(get_zeroed_page);
1675 void __pagevec_free(struct pagevec *pvec)
1677 int i = pagevec_count(pvec);
1679 while (--i >= 0)
1680 free_hot_cold_page(pvec->pages[i], pvec->cold);
1683 void __free_pages(struct page *page, unsigned int order)
1685 if (put_page_testzero(page)) {
1686 if (order == 0)
1687 free_hot_page(page);
1688 else
1689 __free_pages_ok(page, order);
1693 EXPORT_SYMBOL(__free_pages);
1695 void free_pages(unsigned long addr, unsigned int order)
1697 if (addr != 0) {
1698 VM_BUG_ON(!virt_addr_valid((void *)addr));
1699 __free_pages(virt_to_page((void *)addr), order);
1703 EXPORT_SYMBOL(free_pages);
1705 static unsigned int nr_free_zone_pages(int offset)
1707 /* Just pick one node, since fallback list is circular */
1708 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1709 unsigned int sum = 0;
1711 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1712 struct zone **zonep = zonelist->zones;
1713 struct zone *zone;
1715 for (zone = *zonep++; zone; zone = *zonep++) {
1716 unsigned long size = zone->present_pages;
1717 unsigned long high = zone->pages_high;
1718 if (size > high)
1719 sum += size - high;
1722 return sum;
1726 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1728 unsigned int nr_free_buffer_pages(void)
1730 return nr_free_zone_pages(gfp_zone(GFP_USER));
1732 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1735 * Amount of free RAM allocatable within all zones
1737 unsigned int nr_free_pagecache_pages(void)
1739 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1742 static inline void show_node(struct zone *zone)
1744 if (NUMA_BUILD)
1745 printk("Node %d ", zone_to_nid(zone));
1748 void si_meminfo(struct sysinfo *val)
1750 val->totalram = totalram_pages;
1751 val->sharedram = 0;
1752 val->freeram = global_page_state(NR_FREE_PAGES);
1753 val->bufferram = nr_blockdev_pages();
1754 val->totalhigh = totalhigh_pages;
1755 val->freehigh = nr_free_highpages();
1756 val->mem_unit = PAGE_SIZE;
1759 EXPORT_SYMBOL(si_meminfo);
1761 #ifdef CONFIG_NUMA
1762 void si_meminfo_node(struct sysinfo *val, int nid)
1764 pg_data_t *pgdat = NODE_DATA(nid);
1766 val->totalram = pgdat->node_present_pages;
1767 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1768 #ifdef CONFIG_HIGHMEM
1769 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1770 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1771 NR_FREE_PAGES);
1772 #else
1773 val->totalhigh = 0;
1774 val->freehigh = 0;
1775 #endif
1776 val->mem_unit = PAGE_SIZE;
1778 #endif
1780 #define K(x) ((x) << (PAGE_SHIFT-10))
1783 * Show free area list (used inside shift_scroll-lock stuff)
1784 * We also calculate the percentage fragmentation. We do this by counting the
1785 * memory on each free list with the exception of the first item on the list.
1787 void show_free_areas(void)
1789 int cpu;
1790 struct zone *zone;
1792 for_each_zone(zone) {
1793 if (!populated_zone(zone))
1794 continue;
1796 show_node(zone);
1797 printk("%s per-cpu:\n", zone->name);
1799 for_each_online_cpu(cpu) {
1800 struct per_cpu_pageset *pageset;
1802 pageset = zone_pcp(zone, cpu);
1804 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1805 cpu, pageset->pcp.high,
1806 pageset->pcp.batch, pageset->pcp.count);
1810 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1811 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1812 global_page_state(NR_ACTIVE),
1813 global_page_state(NR_INACTIVE),
1814 global_page_state(NR_FILE_DIRTY),
1815 global_page_state(NR_WRITEBACK),
1816 global_page_state(NR_UNSTABLE_NFS),
1817 global_page_state(NR_FREE_PAGES),
1818 global_page_state(NR_SLAB_RECLAIMABLE) +
1819 global_page_state(NR_SLAB_UNRECLAIMABLE),
1820 global_page_state(NR_FILE_MAPPED),
1821 global_page_state(NR_PAGETABLE),
1822 global_page_state(NR_BOUNCE));
1824 for_each_zone(zone) {
1825 int i;
1827 if (!populated_zone(zone))
1828 continue;
1830 show_node(zone);
1831 printk("%s"
1832 " free:%lukB"
1833 " min:%lukB"
1834 " low:%lukB"
1835 " high:%lukB"
1836 " active:%lukB"
1837 " inactive:%lukB"
1838 " present:%lukB"
1839 " pages_scanned:%lu"
1840 " all_unreclaimable? %s"
1841 "\n",
1842 zone->name,
1843 K(zone_page_state(zone, NR_FREE_PAGES)),
1844 K(zone->pages_min),
1845 K(zone->pages_low),
1846 K(zone->pages_high),
1847 K(zone_page_state(zone, NR_ACTIVE)),
1848 K(zone_page_state(zone, NR_INACTIVE)),
1849 K(zone->present_pages),
1850 zone->pages_scanned,
1851 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
1853 printk("lowmem_reserve[]:");
1854 for (i = 0; i < MAX_NR_ZONES; i++)
1855 printk(" %lu", zone->lowmem_reserve[i]);
1856 printk("\n");
1859 for_each_zone(zone) {
1860 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1862 if (!populated_zone(zone))
1863 continue;
1865 show_node(zone);
1866 printk("%s: ", zone->name);
1868 spin_lock_irqsave(&zone->lock, flags);
1869 for (order = 0; order < MAX_ORDER; order++) {
1870 nr[order] = zone->free_area[order].nr_free;
1871 total += nr[order] << order;
1873 spin_unlock_irqrestore(&zone->lock, flags);
1874 for (order = 0; order < MAX_ORDER; order++)
1875 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1876 printk("= %lukB\n", K(total));
1879 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
1881 show_swap_cache_info();
1885 * Builds allocation fallback zone lists.
1887 * Add all populated zones of a node to the zonelist.
1889 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1890 int nr_zones, enum zone_type zone_type)
1892 struct zone *zone;
1894 BUG_ON(zone_type >= MAX_NR_ZONES);
1895 zone_type++;
1897 do {
1898 zone_type--;
1899 zone = pgdat->node_zones + zone_type;
1900 if (populated_zone(zone)) {
1901 zonelist->zones[nr_zones++] = zone;
1902 check_highest_zone(zone_type);
1905 } while (zone_type);
1906 return nr_zones;
1911 * zonelist_order:
1912 * 0 = automatic detection of better ordering.
1913 * 1 = order by ([node] distance, -zonetype)
1914 * 2 = order by (-zonetype, [node] distance)
1916 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1917 * the same zonelist. So only NUMA can configure this param.
1919 #define ZONELIST_ORDER_DEFAULT 0
1920 #define ZONELIST_ORDER_NODE 1
1921 #define ZONELIST_ORDER_ZONE 2
1923 /* zonelist order in the kernel.
1924 * set_zonelist_order() will set this to NODE or ZONE.
1926 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1927 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1930 #ifdef CONFIG_NUMA
1931 /* The value user specified ....changed by config */
1932 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1933 /* string for sysctl */
1934 #define NUMA_ZONELIST_ORDER_LEN 16
1935 char numa_zonelist_order[16] = "default";
1938 * interface for configure zonelist ordering.
1939 * command line option "numa_zonelist_order"
1940 * = "[dD]efault - default, automatic configuration.
1941 * = "[nN]ode - order by node locality, then by zone within node
1942 * = "[zZ]one - order by zone, then by locality within zone
1945 static int __parse_numa_zonelist_order(char *s)
1947 if (*s == 'd' || *s == 'D') {
1948 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1949 } else if (*s == 'n' || *s == 'N') {
1950 user_zonelist_order = ZONELIST_ORDER_NODE;
1951 } else if (*s == 'z' || *s == 'Z') {
1952 user_zonelist_order = ZONELIST_ORDER_ZONE;
1953 } else {
1954 printk(KERN_WARNING
1955 "Ignoring invalid numa_zonelist_order value: "
1956 "%s\n", s);
1957 return -EINVAL;
1959 return 0;
1962 static __init int setup_numa_zonelist_order(char *s)
1964 if (s)
1965 return __parse_numa_zonelist_order(s);
1966 return 0;
1968 early_param("numa_zonelist_order", setup_numa_zonelist_order);
1971 * sysctl handler for numa_zonelist_order
1973 int numa_zonelist_order_handler(ctl_table *table, int write,
1974 struct file *file, void __user *buffer, size_t *length,
1975 loff_t *ppos)
1977 char saved_string[NUMA_ZONELIST_ORDER_LEN];
1978 int ret;
1980 if (write)
1981 strncpy(saved_string, (char*)table->data,
1982 NUMA_ZONELIST_ORDER_LEN);
1983 ret = proc_dostring(table, write, file, buffer, length, ppos);
1984 if (ret)
1985 return ret;
1986 if (write) {
1987 int oldval = user_zonelist_order;
1988 if (__parse_numa_zonelist_order((char*)table->data)) {
1990 * bogus value. restore saved string
1992 strncpy((char*)table->data, saved_string,
1993 NUMA_ZONELIST_ORDER_LEN);
1994 user_zonelist_order = oldval;
1995 } else if (oldval != user_zonelist_order)
1996 build_all_zonelists();
1998 return 0;
2002 #define MAX_NODE_LOAD (num_online_nodes())
2003 static int node_load[MAX_NUMNODES];
2006 * find_next_best_node - find the next node that should appear in a given node's fallback list
2007 * @node: node whose fallback list we're appending
2008 * @used_node_mask: nodemask_t of already used nodes
2010 * We use a number of factors to determine which is the next node that should
2011 * appear on a given node's fallback list. The node should not have appeared
2012 * already in @node's fallback list, and it should be the next closest node
2013 * according to the distance array (which contains arbitrary distance values
2014 * from each node to each node in the system), and should also prefer nodes
2015 * with no CPUs, since presumably they'll have very little allocation pressure
2016 * on them otherwise.
2017 * It returns -1 if no node is found.
2019 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2021 int n, val;
2022 int min_val = INT_MAX;
2023 int best_node = -1;
2025 /* Use the local node if we haven't already */
2026 if (!node_isset(node, *used_node_mask)) {
2027 node_set(node, *used_node_mask);
2028 return node;
2031 for_each_node_state(n, N_HIGH_MEMORY) {
2032 cpumask_t tmp;
2034 /* Don't want a node to appear more than once */
2035 if (node_isset(n, *used_node_mask))
2036 continue;
2038 /* Use the distance array to find the distance */
2039 val = node_distance(node, n);
2041 /* Penalize nodes under us ("prefer the next node") */
2042 val += (n < node);
2044 /* Give preference to headless and unused nodes */
2045 tmp = node_to_cpumask(n);
2046 if (!cpus_empty(tmp))
2047 val += PENALTY_FOR_NODE_WITH_CPUS;
2049 /* Slight preference for less loaded node */
2050 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2051 val += node_load[n];
2053 if (val < min_val) {
2054 min_val = val;
2055 best_node = n;
2059 if (best_node >= 0)
2060 node_set(best_node, *used_node_mask);
2062 return best_node;
2067 * Build zonelists ordered by node and zones within node.
2068 * This results in maximum locality--normal zone overflows into local
2069 * DMA zone, if any--but risks exhausting DMA zone.
2071 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2073 enum zone_type i;
2074 int j;
2075 struct zonelist *zonelist;
2077 for (i = 0; i < MAX_NR_ZONES; i++) {
2078 zonelist = pgdat->node_zonelists + i;
2079 for (j = 0; zonelist->zones[j] != NULL; j++)
2081 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2082 zonelist->zones[j] = NULL;
2087 * Build gfp_thisnode zonelists
2089 static void build_thisnode_zonelists(pg_data_t *pgdat)
2091 enum zone_type i;
2092 int j;
2093 struct zonelist *zonelist;
2095 for (i = 0; i < MAX_NR_ZONES; i++) {
2096 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2097 j = build_zonelists_node(pgdat, zonelist, 0, i);
2098 zonelist->zones[j] = NULL;
2103 * Build zonelists ordered by zone and nodes within zones.
2104 * This results in conserving DMA zone[s] until all Normal memory is
2105 * exhausted, but results in overflowing to remote node while memory
2106 * may still exist in local DMA zone.
2108 static int node_order[MAX_NUMNODES];
2110 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2112 enum zone_type i;
2113 int pos, j, node;
2114 int zone_type; /* needs to be signed */
2115 struct zone *z;
2116 struct zonelist *zonelist;
2118 for (i = 0; i < MAX_NR_ZONES; i++) {
2119 zonelist = pgdat->node_zonelists + i;
2120 pos = 0;
2121 for (zone_type = i; zone_type >= 0; zone_type--) {
2122 for (j = 0; j < nr_nodes; j++) {
2123 node = node_order[j];
2124 z = &NODE_DATA(node)->node_zones[zone_type];
2125 if (populated_zone(z)) {
2126 zonelist->zones[pos++] = z;
2127 check_highest_zone(zone_type);
2131 zonelist->zones[pos] = NULL;
2135 static int default_zonelist_order(void)
2137 int nid, zone_type;
2138 unsigned long low_kmem_size,total_size;
2139 struct zone *z;
2140 int average_size;
2142 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2143 * If they are really small and used heavily, the system can fall
2144 * into OOM very easily.
2145 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2147 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2148 low_kmem_size = 0;
2149 total_size = 0;
2150 for_each_online_node(nid) {
2151 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2152 z = &NODE_DATA(nid)->node_zones[zone_type];
2153 if (populated_zone(z)) {
2154 if (zone_type < ZONE_NORMAL)
2155 low_kmem_size += z->present_pages;
2156 total_size += z->present_pages;
2160 if (!low_kmem_size || /* there are no DMA area. */
2161 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2162 return ZONELIST_ORDER_NODE;
2164 * look into each node's config.
2165 * If there is a node whose DMA/DMA32 memory is very big area on
2166 * local memory, NODE_ORDER may be suitable.
2168 average_size = total_size /
2169 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2170 for_each_online_node(nid) {
2171 low_kmem_size = 0;
2172 total_size = 0;
2173 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2174 z = &NODE_DATA(nid)->node_zones[zone_type];
2175 if (populated_zone(z)) {
2176 if (zone_type < ZONE_NORMAL)
2177 low_kmem_size += z->present_pages;
2178 total_size += z->present_pages;
2181 if (low_kmem_size &&
2182 total_size > average_size && /* ignore small node */
2183 low_kmem_size > total_size * 70/100)
2184 return ZONELIST_ORDER_NODE;
2186 return ZONELIST_ORDER_ZONE;
2189 static void set_zonelist_order(void)
2191 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2192 current_zonelist_order = default_zonelist_order();
2193 else
2194 current_zonelist_order = user_zonelist_order;
2197 static void build_zonelists(pg_data_t *pgdat)
2199 int j, node, load;
2200 enum zone_type i;
2201 nodemask_t used_mask;
2202 int local_node, prev_node;
2203 struct zonelist *zonelist;
2204 int order = current_zonelist_order;
2206 /* initialize zonelists */
2207 for (i = 0; i < MAX_ZONELISTS; i++) {
2208 zonelist = pgdat->node_zonelists + i;
2209 zonelist->zones[0] = NULL;
2212 /* NUMA-aware ordering of nodes */
2213 local_node = pgdat->node_id;
2214 load = num_online_nodes();
2215 prev_node = local_node;
2216 nodes_clear(used_mask);
2218 memset(node_load, 0, sizeof(node_load));
2219 memset(node_order, 0, sizeof(node_order));
2220 j = 0;
2222 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2223 int distance = node_distance(local_node, node);
2226 * If another node is sufficiently far away then it is better
2227 * to reclaim pages in a zone before going off node.
2229 if (distance > RECLAIM_DISTANCE)
2230 zone_reclaim_mode = 1;
2233 * We don't want to pressure a particular node.
2234 * So adding penalty to the first node in same
2235 * distance group to make it round-robin.
2237 if (distance != node_distance(local_node, prev_node))
2238 node_load[node] = load;
2240 prev_node = node;
2241 load--;
2242 if (order == ZONELIST_ORDER_NODE)
2243 build_zonelists_in_node_order(pgdat, node);
2244 else
2245 node_order[j++] = node; /* remember order */
2248 if (order == ZONELIST_ORDER_ZONE) {
2249 /* calculate node order -- i.e., DMA last! */
2250 build_zonelists_in_zone_order(pgdat, j);
2253 build_thisnode_zonelists(pgdat);
2256 /* Construct the zonelist performance cache - see further mmzone.h */
2257 static void build_zonelist_cache(pg_data_t *pgdat)
2259 int i;
2261 for (i = 0; i < MAX_NR_ZONES; i++) {
2262 struct zonelist *zonelist;
2263 struct zonelist_cache *zlc;
2264 struct zone **z;
2266 zonelist = pgdat->node_zonelists + i;
2267 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2268 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2269 for (z = zonelist->zones; *z; z++)
2270 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2275 #else /* CONFIG_NUMA */
2277 static void set_zonelist_order(void)
2279 current_zonelist_order = ZONELIST_ORDER_ZONE;
2282 static void build_zonelists(pg_data_t *pgdat)
2284 int node, local_node;
2285 enum zone_type i,j;
2287 local_node = pgdat->node_id;
2288 for (i = 0; i < MAX_NR_ZONES; i++) {
2289 struct zonelist *zonelist;
2291 zonelist = pgdat->node_zonelists + i;
2293 j = build_zonelists_node(pgdat, zonelist, 0, i);
2295 * Now we build the zonelist so that it contains the zones
2296 * of all the other nodes.
2297 * We don't want to pressure a particular node, so when
2298 * building the zones for node N, we make sure that the
2299 * zones coming right after the local ones are those from
2300 * node N+1 (modulo N)
2302 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2303 if (!node_online(node))
2304 continue;
2305 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2307 for (node = 0; node < local_node; node++) {
2308 if (!node_online(node))
2309 continue;
2310 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2313 zonelist->zones[j] = NULL;
2317 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2318 static void build_zonelist_cache(pg_data_t *pgdat)
2320 int i;
2322 for (i = 0; i < MAX_NR_ZONES; i++)
2323 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2326 #endif /* CONFIG_NUMA */
2328 /* return values int ....just for stop_machine_run() */
2329 static int __build_all_zonelists(void *dummy)
2331 int nid;
2333 for_each_online_node(nid) {
2334 pg_data_t *pgdat = NODE_DATA(nid);
2336 build_zonelists(pgdat);
2337 build_zonelist_cache(pgdat);
2339 return 0;
2342 void build_all_zonelists(void)
2344 set_zonelist_order();
2346 if (system_state == SYSTEM_BOOTING) {
2347 __build_all_zonelists(NULL);
2348 cpuset_init_current_mems_allowed();
2349 } else {
2350 /* we have to stop all cpus to guarantee there is no user
2351 of zonelist */
2352 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2353 /* cpuset refresh routine should be here */
2355 vm_total_pages = nr_free_pagecache_pages();
2357 * Disable grouping by mobility if the number of pages in the
2358 * system is too low to allow the mechanism to work. It would be
2359 * more accurate, but expensive to check per-zone. This check is
2360 * made on memory-hotadd so a system can start with mobility
2361 * disabled and enable it later
2363 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2364 page_group_by_mobility_disabled = 1;
2365 else
2366 page_group_by_mobility_disabled = 0;
2368 printk("Built %i zonelists in %s order, mobility grouping %s. "
2369 "Total pages: %ld\n",
2370 num_online_nodes(),
2371 zonelist_order_name[current_zonelist_order],
2372 page_group_by_mobility_disabled ? "off" : "on",
2373 vm_total_pages);
2374 #ifdef CONFIG_NUMA
2375 printk("Policy zone: %s\n", zone_names[policy_zone]);
2376 #endif
2380 * Helper functions to size the waitqueue hash table.
2381 * Essentially these want to choose hash table sizes sufficiently
2382 * large so that collisions trying to wait on pages are rare.
2383 * But in fact, the number of active page waitqueues on typical
2384 * systems is ridiculously low, less than 200. So this is even
2385 * conservative, even though it seems large.
2387 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2388 * waitqueues, i.e. the size of the waitq table given the number of pages.
2390 #define PAGES_PER_WAITQUEUE 256
2392 #ifndef CONFIG_MEMORY_HOTPLUG
2393 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2395 unsigned long size = 1;
2397 pages /= PAGES_PER_WAITQUEUE;
2399 while (size < pages)
2400 size <<= 1;
2403 * Once we have dozens or even hundreds of threads sleeping
2404 * on IO we've got bigger problems than wait queue collision.
2405 * Limit the size of the wait table to a reasonable size.
2407 size = min(size, 4096UL);
2409 return max(size, 4UL);
2411 #else
2413 * A zone's size might be changed by hot-add, so it is not possible to determine
2414 * a suitable size for its wait_table. So we use the maximum size now.
2416 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2418 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2419 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2420 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2422 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2423 * or more by the traditional way. (See above). It equals:
2425 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2426 * ia64(16K page size) : = ( 8G + 4M)byte.
2427 * powerpc (64K page size) : = (32G +16M)byte.
2429 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2431 return 4096UL;
2433 #endif
2436 * This is an integer logarithm so that shifts can be used later
2437 * to extract the more random high bits from the multiplicative
2438 * hash function before the remainder is taken.
2440 static inline unsigned long wait_table_bits(unsigned long size)
2442 return ffz(~size);
2445 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2448 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2449 * of blocks reserved is based on zone->pages_min. The memory within the
2450 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2451 * higher will lead to a bigger reserve which will get freed as contiguous
2452 * blocks as reclaim kicks in
2454 static void setup_zone_migrate_reserve(struct zone *zone)
2456 unsigned long start_pfn, pfn, end_pfn;
2457 struct page *page;
2458 unsigned long reserve, block_migratetype;
2460 /* Get the start pfn, end pfn and the number of blocks to reserve */
2461 start_pfn = zone->zone_start_pfn;
2462 end_pfn = start_pfn + zone->spanned_pages;
2463 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2464 pageblock_order;
2466 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2467 if (!pfn_valid(pfn))
2468 continue;
2469 page = pfn_to_page(pfn);
2471 /* Blocks with reserved pages will never free, skip them. */
2472 if (PageReserved(page))
2473 continue;
2475 block_migratetype = get_pageblock_migratetype(page);
2477 /* If this block is reserved, account for it */
2478 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2479 reserve--;
2480 continue;
2483 /* Suitable for reserving if this block is movable */
2484 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2485 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2486 move_freepages_block(zone, page, MIGRATE_RESERVE);
2487 reserve--;
2488 continue;
2492 * If the reserve is met and this is a previous reserved block,
2493 * take it back
2495 if (block_migratetype == MIGRATE_RESERVE) {
2496 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2497 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2503 * Initially all pages are reserved - free ones are freed
2504 * up by free_all_bootmem() once the early boot process is
2505 * done. Non-atomic initialization, single-pass.
2507 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2508 unsigned long start_pfn, enum memmap_context context)
2510 struct page *page;
2511 unsigned long end_pfn = start_pfn + size;
2512 unsigned long pfn;
2514 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2516 * There can be holes in boot-time mem_map[]s
2517 * handed to this function. They do not
2518 * exist on hotplugged memory.
2520 if (context == MEMMAP_EARLY) {
2521 if (!early_pfn_valid(pfn))
2522 continue;
2523 if (!early_pfn_in_nid(pfn, nid))
2524 continue;
2526 page = pfn_to_page(pfn);
2527 set_page_links(page, zone, nid, pfn);
2528 init_page_count(page);
2529 reset_page_mapcount(page);
2530 page_assign_page_cgroup(page, NULL);
2531 SetPageReserved(page);
2534 * Mark the block movable so that blocks are reserved for
2535 * movable at startup. This will force kernel allocations
2536 * to reserve their blocks rather than leaking throughout
2537 * the address space during boot when many long-lived
2538 * kernel allocations are made. Later some blocks near
2539 * the start are marked MIGRATE_RESERVE by
2540 * setup_zone_migrate_reserve()
2542 if ((pfn & (pageblock_nr_pages-1)))
2543 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2545 INIT_LIST_HEAD(&page->lru);
2546 #ifdef WANT_PAGE_VIRTUAL
2547 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2548 if (!is_highmem_idx(zone))
2549 set_page_address(page, __va(pfn << PAGE_SHIFT));
2550 #endif
2554 static void __meminit zone_init_free_lists(struct zone *zone)
2556 int order, t;
2557 for_each_migratetype_order(order, t) {
2558 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2559 zone->free_area[order].nr_free = 0;
2563 #ifndef __HAVE_ARCH_MEMMAP_INIT
2564 #define memmap_init(size, nid, zone, start_pfn) \
2565 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2566 #endif
2568 static int zone_batchsize(struct zone *zone)
2570 int batch;
2573 * The per-cpu-pages pools are set to around 1000th of the
2574 * size of the zone. But no more than 1/2 of a meg.
2576 * OK, so we don't know how big the cache is. So guess.
2578 batch = zone->present_pages / 1024;
2579 if (batch * PAGE_SIZE > 512 * 1024)
2580 batch = (512 * 1024) / PAGE_SIZE;
2581 batch /= 4; /* We effectively *= 4 below */
2582 if (batch < 1)
2583 batch = 1;
2586 * Clamp the batch to a 2^n - 1 value. Having a power
2587 * of 2 value was found to be more likely to have
2588 * suboptimal cache aliasing properties in some cases.
2590 * For example if 2 tasks are alternately allocating
2591 * batches of pages, one task can end up with a lot
2592 * of pages of one half of the possible page colors
2593 * and the other with pages of the other colors.
2595 batch = (1 << (fls(batch + batch/2)-1)) - 1;
2597 return batch;
2600 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2602 struct per_cpu_pages *pcp;
2604 memset(p, 0, sizeof(*p));
2606 pcp = &p->pcp;
2607 pcp->count = 0;
2608 pcp->high = 6 * batch;
2609 pcp->batch = max(1UL, 1 * batch);
2610 INIT_LIST_HEAD(&pcp->list);
2614 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2615 * to the value high for the pageset p.
2618 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2619 unsigned long high)
2621 struct per_cpu_pages *pcp;
2623 pcp = &p->pcp;
2624 pcp->high = high;
2625 pcp->batch = max(1UL, high/4);
2626 if ((high/4) > (PAGE_SHIFT * 8))
2627 pcp->batch = PAGE_SHIFT * 8;
2631 #ifdef CONFIG_NUMA
2633 * Boot pageset table. One per cpu which is going to be used for all
2634 * zones and all nodes. The parameters will be set in such a way
2635 * that an item put on a list will immediately be handed over to
2636 * the buddy list. This is safe since pageset manipulation is done
2637 * with interrupts disabled.
2639 * Some NUMA counter updates may also be caught by the boot pagesets.
2641 * The boot_pagesets must be kept even after bootup is complete for
2642 * unused processors and/or zones. They do play a role for bootstrapping
2643 * hotplugged processors.
2645 * zoneinfo_show() and maybe other functions do
2646 * not check if the processor is online before following the pageset pointer.
2647 * Other parts of the kernel may not check if the zone is available.
2649 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2652 * Dynamically allocate memory for the
2653 * per cpu pageset array in struct zone.
2655 static int __cpuinit process_zones(int cpu)
2657 struct zone *zone, *dzone;
2658 int node = cpu_to_node(cpu);
2660 node_set_state(node, N_CPU); /* this node has a cpu */
2662 for_each_zone(zone) {
2664 if (!populated_zone(zone))
2665 continue;
2667 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2668 GFP_KERNEL, node);
2669 if (!zone_pcp(zone, cpu))
2670 goto bad;
2672 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2674 if (percpu_pagelist_fraction)
2675 setup_pagelist_highmark(zone_pcp(zone, cpu),
2676 (zone->present_pages / percpu_pagelist_fraction));
2679 return 0;
2680 bad:
2681 for_each_zone(dzone) {
2682 if (!populated_zone(dzone))
2683 continue;
2684 if (dzone == zone)
2685 break;
2686 kfree(zone_pcp(dzone, cpu));
2687 zone_pcp(dzone, cpu) = NULL;
2689 return -ENOMEM;
2692 static inline void free_zone_pagesets(int cpu)
2694 struct zone *zone;
2696 for_each_zone(zone) {
2697 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2699 /* Free per_cpu_pageset if it is slab allocated */
2700 if (pset != &boot_pageset[cpu])
2701 kfree(pset);
2702 zone_pcp(zone, cpu) = NULL;
2706 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2707 unsigned long action,
2708 void *hcpu)
2710 int cpu = (long)hcpu;
2711 int ret = NOTIFY_OK;
2713 switch (action) {
2714 case CPU_UP_PREPARE:
2715 case CPU_UP_PREPARE_FROZEN:
2716 if (process_zones(cpu))
2717 ret = NOTIFY_BAD;
2718 break;
2719 case CPU_UP_CANCELED:
2720 case CPU_UP_CANCELED_FROZEN:
2721 case CPU_DEAD:
2722 case CPU_DEAD_FROZEN:
2723 free_zone_pagesets(cpu);
2724 break;
2725 default:
2726 break;
2728 return ret;
2731 static struct notifier_block __cpuinitdata pageset_notifier =
2732 { &pageset_cpuup_callback, NULL, 0 };
2734 void __init setup_per_cpu_pageset(void)
2736 int err;
2738 /* Initialize per_cpu_pageset for cpu 0.
2739 * A cpuup callback will do this for every cpu
2740 * as it comes online
2742 err = process_zones(smp_processor_id());
2743 BUG_ON(err);
2744 register_cpu_notifier(&pageset_notifier);
2747 #endif
2749 static noinline __init_refok
2750 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2752 int i;
2753 struct pglist_data *pgdat = zone->zone_pgdat;
2754 size_t alloc_size;
2757 * The per-page waitqueue mechanism uses hashed waitqueues
2758 * per zone.
2760 zone->wait_table_hash_nr_entries =
2761 wait_table_hash_nr_entries(zone_size_pages);
2762 zone->wait_table_bits =
2763 wait_table_bits(zone->wait_table_hash_nr_entries);
2764 alloc_size = zone->wait_table_hash_nr_entries
2765 * sizeof(wait_queue_head_t);
2767 if (system_state == SYSTEM_BOOTING) {
2768 zone->wait_table = (wait_queue_head_t *)
2769 alloc_bootmem_node(pgdat, alloc_size);
2770 } else {
2772 * This case means that a zone whose size was 0 gets new memory
2773 * via memory hot-add.
2774 * But it may be the case that a new node was hot-added. In
2775 * this case vmalloc() will not be able to use this new node's
2776 * memory - this wait_table must be initialized to use this new
2777 * node itself as well.
2778 * To use this new node's memory, further consideration will be
2779 * necessary.
2781 zone->wait_table = vmalloc(alloc_size);
2783 if (!zone->wait_table)
2784 return -ENOMEM;
2786 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2787 init_waitqueue_head(zone->wait_table + i);
2789 return 0;
2792 static __meminit void zone_pcp_init(struct zone *zone)
2794 int cpu;
2795 unsigned long batch = zone_batchsize(zone);
2797 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2798 #ifdef CONFIG_NUMA
2799 /* Early boot. Slab allocator not functional yet */
2800 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2801 setup_pageset(&boot_pageset[cpu],0);
2802 #else
2803 setup_pageset(zone_pcp(zone,cpu), batch);
2804 #endif
2806 if (zone->present_pages)
2807 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2808 zone->name, zone->present_pages, batch);
2811 __meminit int init_currently_empty_zone(struct zone *zone,
2812 unsigned long zone_start_pfn,
2813 unsigned long size,
2814 enum memmap_context context)
2816 struct pglist_data *pgdat = zone->zone_pgdat;
2817 int ret;
2818 ret = zone_wait_table_init(zone, size);
2819 if (ret)
2820 return ret;
2821 pgdat->nr_zones = zone_idx(zone) + 1;
2823 zone->zone_start_pfn = zone_start_pfn;
2825 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2827 zone_init_free_lists(zone);
2829 return 0;
2832 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2834 * Basic iterator support. Return the first range of PFNs for a node
2835 * Note: nid == MAX_NUMNODES returns first region regardless of node
2837 static int __meminit first_active_region_index_in_nid(int nid)
2839 int i;
2841 for (i = 0; i < nr_nodemap_entries; i++)
2842 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2843 return i;
2845 return -1;
2849 * Basic iterator support. Return the next active range of PFNs for a node
2850 * Note: nid == MAX_NUMNODES returns next region regardless of node
2852 static int __meminit next_active_region_index_in_nid(int index, int nid)
2854 for (index = index + 1; index < nr_nodemap_entries; index++)
2855 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2856 return index;
2858 return -1;
2861 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2863 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2864 * Architectures may implement their own version but if add_active_range()
2865 * was used and there are no special requirements, this is a convenient
2866 * alternative
2868 int __meminit early_pfn_to_nid(unsigned long pfn)
2870 int i;
2872 for (i = 0; i < nr_nodemap_entries; i++) {
2873 unsigned long start_pfn = early_node_map[i].start_pfn;
2874 unsigned long end_pfn = early_node_map[i].end_pfn;
2876 if (start_pfn <= pfn && pfn < end_pfn)
2877 return early_node_map[i].nid;
2880 return 0;
2882 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2884 /* Basic iterator support to walk early_node_map[] */
2885 #define for_each_active_range_index_in_nid(i, nid) \
2886 for (i = first_active_region_index_in_nid(nid); i != -1; \
2887 i = next_active_region_index_in_nid(i, nid))
2890 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2891 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2892 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2894 * If an architecture guarantees that all ranges registered with
2895 * add_active_ranges() contain no holes and may be freed, this
2896 * this function may be used instead of calling free_bootmem() manually.
2898 void __init free_bootmem_with_active_regions(int nid,
2899 unsigned long max_low_pfn)
2901 int i;
2903 for_each_active_range_index_in_nid(i, nid) {
2904 unsigned long size_pages = 0;
2905 unsigned long end_pfn = early_node_map[i].end_pfn;
2907 if (early_node_map[i].start_pfn >= max_low_pfn)
2908 continue;
2910 if (end_pfn > max_low_pfn)
2911 end_pfn = max_low_pfn;
2913 size_pages = end_pfn - early_node_map[i].start_pfn;
2914 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2915 PFN_PHYS(early_node_map[i].start_pfn),
2916 size_pages << PAGE_SHIFT);
2921 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2922 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2924 * If an architecture guarantees that all ranges registered with
2925 * add_active_ranges() contain no holes and may be freed, this
2926 * function may be used instead of calling memory_present() manually.
2928 void __init sparse_memory_present_with_active_regions(int nid)
2930 int i;
2932 for_each_active_range_index_in_nid(i, nid)
2933 memory_present(early_node_map[i].nid,
2934 early_node_map[i].start_pfn,
2935 early_node_map[i].end_pfn);
2939 * push_node_boundaries - Push node boundaries to at least the requested boundary
2940 * @nid: The nid of the node to push the boundary for
2941 * @start_pfn: The start pfn of the node
2942 * @end_pfn: The end pfn of the node
2944 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2945 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2946 * be hotplugged even though no physical memory exists. This function allows
2947 * an arch to push out the node boundaries so mem_map is allocated that can
2948 * be used later.
2950 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2951 void __init push_node_boundaries(unsigned int nid,
2952 unsigned long start_pfn, unsigned long end_pfn)
2954 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2955 nid, start_pfn, end_pfn);
2957 /* Initialise the boundary for this node if necessary */
2958 if (node_boundary_end_pfn[nid] == 0)
2959 node_boundary_start_pfn[nid] = -1UL;
2961 /* Update the boundaries */
2962 if (node_boundary_start_pfn[nid] > start_pfn)
2963 node_boundary_start_pfn[nid] = start_pfn;
2964 if (node_boundary_end_pfn[nid] < end_pfn)
2965 node_boundary_end_pfn[nid] = end_pfn;
2968 /* If necessary, push the node boundary out for reserve hotadd */
2969 static void __meminit account_node_boundary(unsigned int nid,
2970 unsigned long *start_pfn, unsigned long *end_pfn)
2972 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2973 nid, *start_pfn, *end_pfn);
2975 /* Return if boundary information has not been provided */
2976 if (node_boundary_end_pfn[nid] == 0)
2977 return;
2979 /* Check the boundaries and update if necessary */
2980 if (node_boundary_start_pfn[nid] < *start_pfn)
2981 *start_pfn = node_boundary_start_pfn[nid];
2982 if (node_boundary_end_pfn[nid] > *end_pfn)
2983 *end_pfn = node_boundary_end_pfn[nid];
2985 #else
2986 void __init push_node_boundaries(unsigned int nid,
2987 unsigned long start_pfn, unsigned long end_pfn) {}
2989 static void __meminit account_node_boundary(unsigned int nid,
2990 unsigned long *start_pfn, unsigned long *end_pfn) {}
2991 #endif
2995 * get_pfn_range_for_nid - Return the start and end page frames for a node
2996 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2997 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2998 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3000 * It returns the start and end page frame of a node based on information
3001 * provided by an arch calling add_active_range(). If called for a node
3002 * with no available memory, a warning is printed and the start and end
3003 * PFNs will be 0.
3005 void __meminit get_pfn_range_for_nid(unsigned int nid,
3006 unsigned long *start_pfn, unsigned long *end_pfn)
3008 int i;
3009 *start_pfn = -1UL;
3010 *end_pfn = 0;
3012 for_each_active_range_index_in_nid(i, nid) {
3013 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3014 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3017 if (*start_pfn == -1UL)
3018 *start_pfn = 0;
3020 /* Push the node boundaries out if requested */
3021 account_node_boundary(nid, start_pfn, end_pfn);
3025 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3026 * assumption is made that zones within a node are ordered in monotonic
3027 * increasing memory addresses so that the "highest" populated zone is used
3029 void __init find_usable_zone_for_movable(void)
3031 int zone_index;
3032 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3033 if (zone_index == ZONE_MOVABLE)
3034 continue;
3036 if (arch_zone_highest_possible_pfn[zone_index] >
3037 arch_zone_lowest_possible_pfn[zone_index])
3038 break;
3041 VM_BUG_ON(zone_index == -1);
3042 movable_zone = zone_index;
3046 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3047 * because it is sized independant of architecture. Unlike the other zones,
3048 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3049 * in each node depending on the size of each node and how evenly kernelcore
3050 * is distributed. This helper function adjusts the zone ranges
3051 * provided by the architecture for a given node by using the end of the
3052 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3053 * zones within a node are in order of monotonic increases memory addresses
3055 void __meminit adjust_zone_range_for_zone_movable(int nid,
3056 unsigned long zone_type,
3057 unsigned long node_start_pfn,
3058 unsigned long node_end_pfn,
3059 unsigned long *zone_start_pfn,
3060 unsigned long *zone_end_pfn)
3062 /* Only adjust if ZONE_MOVABLE is on this node */
3063 if (zone_movable_pfn[nid]) {
3064 /* Size ZONE_MOVABLE */
3065 if (zone_type == ZONE_MOVABLE) {
3066 *zone_start_pfn = zone_movable_pfn[nid];
3067 *zone_end_pfn = min(node_end_pfn,
3068 arch_zone_highest_possible_pfn[movable_zone]);
3070 /* Adjust for ZONE_MOVABLE starting within this range */
3071 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3072 *zone_end_pfn > zone_movable_pfn[nid]) {
3073 *zone_end_pfn = zone_movable_pfn[nid];
3075 /* Check if this whole range is within ZONE_MOVABLE */
3076 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3077 *zone_start_pfn = *zone_end_pfn;
3082 * Return the number of pages a zone spans in a node, including holes
3083 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3085 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3086 unsigned long zone_type,
3087 unsigned long *ignored)
3089 unsigned long node_start_pfn, node_end_pfn;
3090 unsigned long zone_start_pfn, zone_end_pfn;
3092 /* Get the start and end of the node and zone */
3093 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3094 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3095 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3096 adjust_zone_range_for_zone_movable(nid, zone_type,
3097 node_start_pfn, node_end_pfn,
3098 &zone_start_pfn, &zone_end_pfn);
3100 /* Check that this node has pages within the zone's required range */
3101 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3102 return 0;
3104 /* Move the zone boundaries inside the node if necessary */
3105 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3106 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3108 /* Return the spanned pages */
3109 return zone_end_pfn - zone_start_pfn;
3113 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3114 * then all holes in the requested range will be accounted for.
3116 unsigned long __meminit __absent_pages_in_range(int nid,
3117 unsigned long range_start_pfn,
3118 unsigned long range_end_pfn)
3120 int i = 0;
3121 unsigned long prev_end_pfn = 0, hole_pages = 0;
3122 unsigned long start_pfn;
3124 /* Find the end_pfn of the first active range of pfns in the node */
3125 i = first_active_region_index_in_nid(nid);
3126 if (i == -1)
3127 return 0;
3129 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3131 /* Account for ranges before physical memory on this node */
3132 if (early_node_map[i].start_pfn > range_start_pfn)
3133 hole_pages = prev_end_pfn - range_start_pfn;
3135 /* Find all holes for the zone within the node */
3136 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3138 /* No need to continue if prev_end_pfn is outside the zone */
3139 if (prev_end_pfn >= range_end_pfn)
3140 break;
3142 /* Make sure the end of the zone is not within the hole */
3143 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3144 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3146 /* Update the hole size cound and move on */
3147 if (start_pfn > range_start_pfn) {
3148 BUG_ON(prev_end_pfn > start_pfn);
3149 hole_pages += start_pfn - prev_end_pfn;
3151 prev_end_pfn = early_node_map[i].end_pfn;
3154 /* Account for ranges past physical memory on this node */
3155 if (range_end_pfn > prev_end_pfn)
3156 hole_pages += range_end_pfn -
3157 max(range_start_pfn, prev_end_pfn);
3159 return hole_pages;
3163 * absent_pages_in_range - Return number of page frames in holes within a range
3164 * @start_pfn: The start PFN to start searching for holes
3165 * @end_pfn: The end PFN to stop searching for holes
3167 * It returns the number of pages frames in memory holes within a range.
3169 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3170 unsigned long end_pfn)
3172 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3175 /* Return the number of page frames in holes in a zone on a node */
3176 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3177 unsigned long zone_type,
3178 unsigned long *ignored)
3180 unsigned long node_start_pfn, node_end_pfn;
3181 unsigned long zone_start_pfn, zone_end_pfn;
3183 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3184 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3185 node_start_pfn);
3186 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3187 node_end_pfn);
3189 adjust_zone_range_for_zone_movable(nid, zone_type,
3190 node_start_pfn, node_end_pfn,
3191 &zone_start_pfn, &zone_end_pfn);
3192 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3195 #else
3196 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3197 unsigned long zone_type,
3198 unsigned long *zones_size)
3200 return zones_size[zone_type];
3203 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3204 unsigned long zone_type,
3205 unsigned long *zholes_size)
3207 if (!zholes_size)
3208 return 0;
3210 return zholes_size[zone_type];
3213 #endif
3215 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3216 unsigned long *zones_size, unsigned long *zholes_size)
3218 unsigned long realtotalpages, totalpages = 0;
3219 enum zone_type i;
3221 for (i = 0; i < MAX_NR_ZONES; i++)
3222 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3223 zones_size);
3224 pgdat->node_spanned_pages = totalpages;
3226 realtotalpages = totalpages;
3227 for (i = 0; i < MAX_NR_ZONES; i++)
3228 realtotalpages -=
3229 zone_absent_pages_in_node(pgdat->node_id, i,
3230 zholes_size);
3231 pgdat->node_present_pages = realtotalpages;
3232 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3233 realtotalpages);
3236 #ifndef CONFIG_SPARSEMEM
3238 * Calculate the size of the zone->blockflags rounded to an unsigned long
3239 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3240 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3241 * round what is now in bits to nearest long in bits, then return it in
3242 * bytes.
3244 static unsigned long __init usemap_size(unsigned long zonesize)
3246 unsigned long usemapsize;
3248 usemapsize = roundup(zonesize, pageblock_nr_pages);
3249 usemapsize = usemapsize >> pageblock_order;
3250 usemapsize *= NR_PAGEBLOCK_BITS;
3251 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3253 return usemapsize / 8;
3256 static void __init setup_usemap(struct pglist_data *pgdat,
3257 struct zone *zone, unsigned long zonesize)
3259 unsigned long usemapsize = usemap_size(zonesize);
3260 zone->pageblock_flags = NULL;
3261 if (usemapsize) {
3262 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3263 memset(zone->pageblock_flags, 0, usemapsize);
3266 #else
3267 static void inline setup_usemap(struct pglist_data *pgdat,
3268 struct zone *zone, unsigned long zonesize) {}
3269 #endif /* CONFIG_SPARSEMEM */
3271 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3273 /* Return a sensible default order for the pageblock size. */
3274 static inline int pageblock_default_order(void)
3276 if (HPAGE_SHIFT > PAGE_SHIFT)
3277 return HUGETLB_PAGE_ORDER;
3279 return MAX_ORDER-1;
3282 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3283 static inline void __init set_pageblock_order(unsigned int order)
3285 /* Check that pageblock_nr_pages has not already been setup */
3286 if (pageblock_order)
3287 return;
3290 * Assume the largest contiguous order of interest is a huge page.
3291 * This value may be variable depending on boot parameters on IA64
3293 pageblock_order = order;
3295 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3298 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3299 * and pageblock_default_order() are unused as pageblock_order is set
3300 * at compile-time. See include/linux/pageblock-flags.h for the values of
3301 * pageblock_order based on the kernel config
3303 static inline int pageblock_default_order(unsigned int order)
3305 return MAX_ORDER-1;
3307 #define set_pageblock_order(x) do {} while (0)
3309 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3312 * Set up the zone data structures:
3313 * - mark all pages reserved
3314 * - mark all memory queues empty
3315 * - clear the memory bitmaps
3317 static void __meminit free_area_init_core(struct pglist_data *pgdat,
3318 unsigned long *zones_size, unsigned long *zholes_size)
3320 enum zone_type j;
3321 int nid = pgdat->node_id;
3322 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3323 int ret;
3325 pgdat_resize_init(pgdat);
3326 pgdat->nr_zones = 0;
3327 init_waitqueue_head(&pgdat->kswapd_wait);
3328 pgdat->kswapd_max_order = 0;
3330 for (j = 0; j < MAX_NR_ZONES; j++) {
3331 struct zone *zone = pgdat->node_zones + j;
3332 unsigned long size, realsize, memmap_pages;
3334 size = zone_spanned_pages_in_node(nid, j, zones_size);
3335 realsize = size - zone_absent_pages_in_node(nid, j,
3336 zholes_size);
3339 * Adjust realsize so that it accounts for how much memory
3340 * is used by this zone for memmap. This affects the watermark
3341 * and per-cpu initialisations
3343 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3344 if (realsize >= memmap_pages) {
3345 realsize -= memmap_pages;
3346 printk(KERN_DEBUG
3347 " %s zone: %lu pages used for memmap\n",
3348 zone_names[j], memmap_pages);
3349 } else
3350 printk(KERN_WARNING
3351 " %s zone: %lu pages exceeds realsize %lu\n",
3352 zone_names[j], memmap_pages, realsize);
3354 /* Account for reserved pages */
3355 if (j == 0 && realsize > dma_reserve) {
3356 realsize -= dma_reserve;
3357 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3358 zone_names[0], dma_reserve);
3361 if (!is_highmem_idx(j))
3362 nr_kernel_pages += realsize;
3363 nr_all_pages += realsize;
3365 zone->spanned_pages = size;
3366 zone->present_pages = realsize;
3367 #ifdef CONFIG_NUMA
3368 zone->node = nid;
3369 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3370 / 100;
3371 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3372 #endif
3373 zone->name = zone_names[j];
3374 spin_lock_init(&zone->lock);
3375 spin_lock_init(&zone->lru_lock);
3376 zone_seqlock_init(zone);
3377 zone->zone_pgdat = pgdat;
3379 zone->prev_priority = DEF_PRIORITY;
3381 zone_pcp_init(zone);
3382 INIT_LIST_HEAD(&zone->active_list);
3383 INIT_LIST_HEAD(&zone->inactive_list);
3384 zone->nr_scan_active = 0;
3385 zone->nr_scan_inactive = 0;
3386 zap_zone_vm_stats(zone);
3387 zone->flags = 0;
3388 if (!size)
3389 continue;
3391 set_pageblock_order(pageblock_default_order());
3392 setup_usemap(pgdat, zone, size);
3393 ret = init_currently_empty_zone(zone, zone_start_pfn,
3394 size, MEMMAP_EARLY);
3395 BUG_ON(ret);
3396 zone_start_pfn += size;
3400 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3402 /* Skip empty nodes */
3403 if (!pgdat->node_spanned_pages)
3404 return;
3406 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3407 /* ia64 gets its own node_mem_map, before this, without bootmem */
3408 if (!pgdat->node_mem_map) {
3409 unsigned long size, start, end;
3410 struct page *map;
3413 * The zone's endpoints aren't required to be MAX_ORDER
3414 * aligned but the node_mem_map endpoints must be in order
3415 * for the buddy allocator to function correctly.
3417 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3418 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3419 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3420 size = (end - start) * sizeof(struct page);
3421 map = alloc_remap(pgdat->node_id, size);
3422 if (!map)
3423 map = alloc_bootmem_node(pgdat, size);
3424 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3426 #ifndef CONFIG_NEED_MULTIPLE_NODES
3428 * With no DISCONTIG, the global mem_map is just set as node 0's
3430 if (pgdat == NODE_DATA(0)) {
3431 mem_map = NODE_DATA(0)->node_mem_map;
3432 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3433 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3434 mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3435 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3437 #endif
3438 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3441 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
3442 unsigned long *zones_size, unsigned long node_start_pfn,
3443 unsigned long *zholes_size)
3445 pgdat->node_id = nid;
3446 pgdat->node_start_pfn = node_start_pfn;
3447 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3449 alloc_node_mem_map(pgdat);
3451 free_area_init_core(pgdat, zones_size, zholes_size);
3454 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3456 #if MAX_NUMNODES > 1
3458 * Figure out the number of possible node ids.
3460 static void __init setup_nr_node_ids(void)
3462 unsigned int node;
3463 unsigned int highest = 0;
3465 for_each_node_mask(node, node_possible_map)
3466 highest = node;
3467 nr_node_ids = highest + 1;
3469 #else
3470 static inline void setup_nr_node_ids(void)
3473 #endif
3476 * add_active_range - Register a range of PFNs backed by physical memory
3477 * @nid: The node ID the range resides on
3478 * @start_pfn: The start PFN of the available physical memory
3479 * @end_pfn: The end PFN of the available physical memory
3481 * These ranges are stored in an early_node_map[] and later used by
3482 * free_area_init_nodes() to calculate zone sizes and holes. If the
3483 * range spans a memory hole, it is up to the architecture to ensure
3484 * the memory is not freed by the bootmem allocator. If possible
3485 * the range being registered will be merged with existing ranges.
3487 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3488 unsigned long end_pfn)
3490 int i;
3492 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3493 "%d entries of %d used\n",
3494 nid, start_pfn, end_pfn,
3495 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3497 /* Merge with existing active regions if possible */
3498 for (i = 0; i < nr_nodemap_entries; i++) {
3499 if (early_node_map[i].nid != nid)
3500 continue;
3502 /* Skip if an existing region covers this new one */
3503 if (start_pfn >= early_node_map[i].start_pfn &&
3504 end_pfn <= early_node_map[i].end_pfn)
3505 return;
3507 /* Merge forward if suitable */
3508 if (start_pfn <= early_node_map[i].end_pfn &&
3509 end_pfn > early_node_map[i].end_pfn) {
3510 early_node_map[i].end_pfn = end_pfn;
3511 return;
3514 /* Merge backward if suitable */
3515 if (start_pfn < early_node_map[i].end_pfn &&
3516 end_pfn >= early_node_map[i].start_pfn) {
3517 early_node_map[i].start_pfn = start_pfn;
3518 return;
3522 /* Check that early_node_map is large enough */
3523 if (i >= MAX_ACTIVE_REGIONS) {
3524 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3525 MAX_ACTIVE_REGIONS);
3526 return;
3529 early_node_map[i].nid = nid;
3530 early_node_map[i].start_pfn = start_pfn;
3531 early_node_map[i].end_pfn = end_pfn;
3532 nr_nodemap_entries = i + 1;
3536 * shrink_active_range - Shrink an existing registered range of PFNs
3537 * @nid: The node id the range is on that should be shrunk
3538 * @old_end_pfn: The old end PFN of the range
3539 * @new_end_pfn: The new PFN of the range
3541 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3542 * The map is kept at the end physical page range that has already been
3543 * registered with add_active_range(). This function allows an arch to shrink
3544 * an existing registered range.
3546 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3547 unsigned long new_end_pfn)
3549 int i;
3551 /* Find the old active region end and shrink */
3552 for_each_active_range_index_in_nid(i, nid)
3553 if (early_node_map[i].end_pfn == old_end_pfn) {
3554 early_node_map[i].end_pfn = new_end_pfn;
3555 break;
3560 * remove_all_active_ranges - Remove all currently registered regions
3562 * During discovery, it may be found that a table like SRAT is invalid
3563 * and an alternative discovery method must be used. This function removes
3564 * all currently registered regions.
3566 void __init remove_all_active_ranges(void)
3568 memset(early_node_map, 0, sizeof(early_node_map));
3569 nr_nodemap_entries = 0;
3570 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3571 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3572 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3573 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3576 /* Compare two active node_active_regions */
3577 static int __init cmp_node_active_region(const void *a, const void *b)
3579 struct node_active_region *arange = (struct node_active_region *)a;
3580 struct node_active_region *brange = (struct node_active_region *)b;
3582 /* Done this way to avoid overflows */
3583 if (arange->start_pfn > brange->start_pfn)
3584 return 1;
3585 if (arange->start_pfn < brange->start_pfn)
3586 return -1;
3588 return 0;
3591 /* sort the node_map by start_pfn */
3592 static void __init sort_node_map(void)
3594 sort(early_node_map, (size_t)nr_nodemap_entries,
3595 sizeof(struct node_active_region),
3596 cmp_node_active_region, NULL);
3599 /* Find the lowest pfn for a node */
3600 unsigned long __init find_min_pfn_for_node(unsigned long nid)
3602 int i;
3603 unsigned long min_pfn = ULONG_MAX;
3605 /* Assuming a sorted map, the first range found has the starting pfn */
3606 for_each_active_range_index_in_nid(i, nid)
3607 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3609 if (min_pfn == ULONG_MAX) {
3610 printk(KERN_WARNING
3611 "Could not find start_pfn for node %lu\n", nid);
3612 return 0;
3615 return min_pfn;
3619 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3621 * It returns the minimum PFN based on information provided via
3622 * add_active_range().
3624 unsigned long __init find_min_pfn_with_active_regions(void)
3626 return find_min_pfn_for_node(MAX_NUMNODES);
3630 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3632 * It returns the maximum PFN based on information provided via
3633 * add_active_range().
3635 unsigned long __init find_max_pfn_with_active_regions(void)
3637 int i;
3638 unsigned long max_pfn = 0;
3640 for (i = 0; i < nr_nodemap_entries; i++)
3641 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3643 return max_pfn;
3647 * early_calculate_totalpages()
3648 * Sum pages in active regions for movable zone.
3649 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3651 static unsigned long __init early_calculate_totalpages(void)
3653 int i;
3654 unsigned long totalpages = 0;
3656 for (i = 0; i < nr_nodemap_entries; i++) {
3657 unsigned long pages = early_node_map[i].end_pfn -
3658 early_node_map[i].start_pfn;
3659 totalpages += pages;
3660 if (pages)
3661 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3663 return totalpages;
3667 * Find the PFN the Movable zone begins in each node. Kernel memory
3668 * is spread evenly between nodes as long as the nodes have enough
3669 * memory. When they don't, some nodes will have more kernelcore than
3670 * others
3672 void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3674 int i, nid;
3675 unsigned long usable_startpfn;
3676 unsigned long kernelcore_node, kernelcore_remaining;
3677 unsigned long totalpages = early_calculate_totalpages();
3678 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3681 * If movablecore was specified, calculate what size of
3682 * kernelcore that corresponds so that memory usable for
3683 * any allocation type is evenly spread. If both kernelcore
3684 * and movablecore are specified, then the value of kernelcore
3685 * will be used for required_kernelcore if it's greater than
3686 * what movablecore would have allowed.
3688 if (required_movablecore) {
3689 unsigned long corepages;
3692 * Round-up so that ZONE_MOVABLE is at least as large as what
3693 * was requested by the user
3695 required_movablecore =
3696 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3697 corepages = totalpages - required_movablecore;
3699 required_kernelcore = max(required_kernelcore, corepages);
3702 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3703 if (!required_kernelcore)
3704 return;
3706 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3707 find_usable_zone_for_movable();
3708 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3710 restart:
3711 /* Spread kernelcore memory as evenly as possible throughout nodes */
3712 kernelcore_node = required_kernelcore / usable_nodes;
3713 for_each_node_state(nid, N_HIGH_MEMORY) {
3715 * Recalculate kernelcore_node if the division per node
3716 * now exceeds what is necessary to satisfy the requested
3717 * amount of memory for the kernel
3719 if (required_kernelcore < kernelcore_node)
3720 kernelcore_node = required_kernelcore / usable_nodes;
3723 * As the map is walked, we track how much memory is usable
3724 * by the kernel using kernelcore_remaining. When it is
3725 * 0, the rest of the node is usable by ZONE_MOVABLE
3727 kernelcore_remaining = kernelcore_node;
3729 /* Go through each range of PFNs within this node */
3730 for_each_active_range_index_in_nid(i, nid) {
3731 unsigned long start_pfn, end_pfn;
3732 unsigned long size_pages;
3734 start_pfn = max(early_node_map[i].start_pfn,
3735 zone_movable_pfn[nid]);
3736 end_pfn = early_node_map[i].end_pfn;
3737 if (start_pfn >= end_pfn)
3738 continue;
3740 /* Account for what is only usable for kernelcore */
3741 if (start_pfn < usable_startpfn) {
3742 unsigned long kernel_pages;
3743 kernel_pages = min(end_pfn, usable_startpfn)
3744 - start_pfn;
3746 kernelcore_remaining -= min(kernel_pages,
3747 kernelcore_remaining);
3748 required_kernelcore -= min(kernel_pages,
3749 required_kernelcore);
3751 /* Continue if range is now fully accounted */
3752 if (end_pfn <= usable_startpfn) {
3755 * Push zone_movable_pfn to the end so
3756 * that if we have to rebalance
3757 * kernelcore across nodes, we will
3758 * not double account here
3760 zone_movable_pfn[nid] = end_pfn;
3761 continue;
3763 start_pfn = usable_startpfn;
3767 * The usable PFN range for ZONE_MOVABLE is from
3768 * start_pfn->end_pfn. Calculate size_pages as the
3769 * number of pages used as kernelcore
3771 size_pages = end_pfn - start_pfn;
3772 if (size_pages > kernelcore_remaining)
3773 size_pages = kernelcore_remaining;
3774 zone_movable_pfn[nid] = start_pfn + size_pages;
3777 * Some kernelcore has been met, update counts and
3778 * break if the kernelcore for this node has been
3779 * satisified
3781 required_kernelcore -= min(required_kernelcore,
3782 size_pages);
3783 kernelcore_remaining -= size_pages;
3784 if (!kernelcore_remaining)
3785 break;
3790 * If there is still required_kernelcore, we do another pass with one
3791 * less node in the count. This will push zone_movable_pfn[nid] further
3792 * along on the nodes that still have memory until kernelcore is
3793 * satisified
3795 usable_nodes--;
3796 if (usable_nodes && required_kernelcore > usable_nodes)
3797 goto restart;
3799 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3800 for (nid = 0; nid < MAX_NUMNODES; nid++)
3801 zone_movable_pfn[nid] =
3802 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3805 /* Any regular memory on that node ? */
3806 static void check_for_regular_memory(pg_data_t *pgdat)
3808 #ifdef CONFIG_HIGHMEM
3809 enum zone_type zone_type;
3811 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3812 struct zone *zone = &pgdat->node_zones[zone_type];
3813 if (zone->present_pages)
3814 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3816 #endif
3820 * free_area_init_nodes - Initialise all pg_data_t and zone data
3821 * @max_zone_pfn: an array of max PFNs for each zone
3823 * This will call free_area_init_node() for each active node in the system.
3824 * Using the page ranges provided by add_active_range(), the size of each
3825 * zone in each node and their holes is calculated. If the maximum PFN
3826 * between two adjacent zones match, it is assumed that the zone is empty.
3827 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3828 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3829 * starts where the previous one ended. For example, ZONE_DMA32 starts
3830 * at arch_max_dma_pfn.
3832 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3834 unsigned long nid;
3835 enum zone_type i;
3837 /* Sort early_node_map as initialisation assumes it is sorted */
3838 sort_node_map();
3840 /* Record where the zone boundaries are */
3841 memset(arch_zone_lowest_possible_pfn, 0,
3842 sizeof(arch_zone_lowest_possible_pfn));
3843 memset(arch_zone_highest_possible_pfn, 0,
3844 sizeof(arch_zone_highest_possible_pfn));
3845 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3846 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3847 for (i = 1; i < MAX_NR_ZONES; i++) {
3848 if (i == ZONE_MOVABLE)
3849 continue;
3850 arch_zone_lowest_possible_pfn[i] =
3851 arch_zone_highest_possible_pfn[i-1];
3852 arch_zone_highest_possible_pfn[i] =
3853 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3855 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3856 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3858 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3859 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3860 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
3862 /* Print out the zone ranges */
3863 printk("Zone PFN ranges:\n");
3864 for (i = 0; i < MAX_NR_ZONES; i++) {
3865 if (i == ZONE_MOVABLE)
3866 continue;
3867 printk(" %-8s %8lu -> %8lu\n",
3868 zone_names[i],
3869 arch_zone_lowest_possible_pfn[i],
3870 arch_zone_highest_possible_pfn[i]);
3873 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3874 printk("Movable zone start PFN for each node\n");
3875 for (i = 0; i < MAX_NUMNODES; i++) {
3876 if (zone_movable_pfn[i])
3877 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3880 /* Print out the early_node_map[] */
3881 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3882 for (i = 0; i < nr_nodemap_entries; i++)
3883 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3884 early_node_map[i].start_pfn,
3885 early_node_map[i].end_pfn);
3887 /* Initialise every node */
3888 setup_nr_node_ids();
3889 for_each_online_node(nid) {
3890 pg_data_t *pgdat = NODE_DATA(nid);
3891 free_area_init_node(nid, pgdat, NULL,
3892 find_min_pfn_for_node(nid), NULL);
3894 /* Any memory on that node */
3895 if (pgdat->node_present_pages)
3896 node_set_state(nid, N_HIGH_MEMORY);
3897 check_for_regular_memory(pgdat);
3901 static int __init cmdline_parse_core(char *p, unsigned long *core)
3903 unsigned long long coremem;
3904 if (!p)
3905 return -EINVAL;
3907 coremem = memparse(p, &p);
3908 *core = coremem >> PAGE_SHIFT;
3910 /* Paranoid check that UL is enough for the coremem value */
3911 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3913 return 0;
3917 * kernelcore=size sets the amount of memory for use for allocations that
3918 * cannot be reclaimed or migrated.
3920 static int __init cmdline_parse_kernelcore(char *p)
3922 return cmdline_parse_core(p, &required_kernelcore);
3926 * movablecore=size sets the amount of memory for use for allocations that
3927 * can be reclaimed or migrated.
3929 static int __init cmdline_parse_movablecore(char *p)
3931 return cmdline_parse_core(p, &required_movablecore);
3934 early_param("kernelcore", cmdline_parse_kernelcore);
3935 early_param("movablecore", cmdline_parse_movablecore);
3937 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3940 * set_dma_reserve - set the specified number of pages reserved in the first zone
3941 * @new_dma_reserve: The number of pages to mark reserved
3943 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3944 * In the DMA zone, a significant percentage may be consumed by kernel image
3945 * and other unfreeable allocations which can skew the watermarks badly. This
3946 * function may optionally be used to account for unfreeable pages in the
3947 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3948 * smaller per-cpu batchsize.
3950 void __init set_dma_reserve(unsigned long new_dma_reserve)
3952 dma_reserve = new_dma_reserve;
3955 #ifndef CONFIG_NEED_MULTIPLE_NODES
3956 static bootmem_data_t contig_bootmem_data;
3957 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3959 EXPORT_SYMBOL(contig_page_data);
3960 #endif
3962 void __init free_area_init(unsigned long *zones_size)
3964 free_area_init_node(0, NODE_DATA(0), zones_size,
3965 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3968 static int page_alloc_cpu_notify(struct notifier_block *self,
3969 unsigned long action, void *hcpu)
3971 int cpu = (unsigned long)hcpu;
3973 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
3974 drain_pages(cpu);
3977 * Spill the event counters of the dead processor
3978 * into the current processors event counters.
3979 * This artificially elevates the count of the current
3980 * processor.
3982 vm_events_fold_cpu(cpu);
3985 * Zero the differential counters of the dead processor
3986 * so that the vm statistics are consistent.
3988 * This is only okay since the processor is dead and cannot
3989 * race with what we are doing.
3991 refresh_cpu_vm_stats(cpu);
3993 return NOTIFY_OK;
3996 void __init page_alloc_init(void)
3998 hotcpu_notifier(page_alloc_cpu_notify, 0);
4002 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4003 * or min_free_kbytes changes.
4005 static void calculate_totalreserve_pages(void)
4007 struct pglist_data *pgdat;
4008 unsigned long reserve_pages = 0;
4009 enum zone_type i, j;
4011 for_each_online_pgdat(pgdat) {
4012 for (i = 0; i < MAX_NR_ZONES; i++) {
4013 struct zone *zone = pgdat->node_zones + i;
4014 unsigned long max = 0;
4016 /* Find valid and maximum lowmem_reserve in the zone */
4017 for (j = i; j < MAX_NR_ZONES; j++) {
4018 if (zone->lowmem_reserve[j] > max)
4019 max = zone->lowmem_reserve[j];
4022 /* we treat pages_high as reserved pages. */
4023 max += zone->pages_high;
4025 if (max > zone->present_pages)
4026 max = zone->present_pages;
4027 reserve_pages += max;
4030 totalreserve_pages = reserve_pages;
4034 * setup_per_zone_lowmem_reserve - called whenever
4035 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4036 * has a correct pages reserved value, so an adequate number of
4037 * pages are left in the zone after a successful __alloc_pages().
4039 static void setup_per_zone_lowmem_reserve(void)
4041 struct pglist_data *pgdat;
4042 enum zone_type j, idx;
4044 for_each_online_pgdat(pgdat) {
4045 for (j = 0; j < MAX_NR_ZONES; j++) {
4046 struct zone *zone = pgdat->node_zones + j;
4047 unsigned long present_pages = zone->present_pages;
4049 zone->lowmem_reserve[j] = 0;
4051 idx = j;
4052 while (idx) {
4053 struct zone *lower_zone;
4055 idx--;
4057 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4058 sysctl_lowmem_reserve_ratio[idx] = 1;
4060 lower_zone = pgdat->node_zones + idx;
4061 lower_zone->lowmem_reserve[j] = present_pages /
4062 sysctl_lowmem_reserve_ratio[idx];
4063 present_pages += lower_zone->present_pages;
4068 /* update totalreserve_pages */
4069 calculate_totalreserve_pages();
4073 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4075 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4076 * with respect to min_free_kbytes.
4078 void setup_per_zone_pages_min(void)
4080 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4081 unsigned long lowmem_pages = 0;
4082 struct zone *zone;
4083 unsigned long flags;
4085 /* Calculate total number of !ZONE_HIGHMEM pages */
4086 for_each_zone(zone) {
4087 if (!is_highmem(zone))
4088 lowmem_pages += zone->present_pages;
4091 for_each_zone(zone) {
4092 u64 tmp;
4094 spin_lock_irqsave(&zone->lru_lock, flags);
4095 tmp = (u64)pages_min * zone->present_pages;
4096 do_div(tmp, lowmem_pages);
4097 if (is_highmem(zone)) {
4099 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4100 * need highmem pages, so cap pages_min to a small
4101 * value here.
4103 * The (pages_high-pages_low) and (pages_low-pages_min)
4104 * deltas controls asynch page reclaim, and so should
4105 * not be capped for highmem.
4107 int min_pages;
4109 min_pages = zone->present_pages / 1024;
4110 if (min_pages < SWAP_CLUSTER_MAX)
4111 min_pages = SWAP_CLUSTER_MAX;
4112 if (min_pages > 128)
4113 min_pages = 128;
4114 zone->pages_min = min_pages;
4115 } else {
4117 * If it's a lowmem zone, reserve a number of pages
4118 * proportionate to the zone's size.
4120 zone->pages_min = tmp;
4123 zone->pages_low = zone->pages_min + (tmp >> 2);
4124 zone->pages_high = zone->pages_min + (tmp >> 1);
4125 setup_zone_migrate_reserve(zone);
4126 spin_unlock_irqrestore(&zone->lru_lock, flags);
4129 /* update totalreserve_pages */
4130 calculate_totalreserve_pages();
4134 * Initialise min_free_kbytes.
4136 * For small machines we want it small (128k min). For large machines
4137 * we want it large (64MB max). But it is not linear, because network
4138 * bandwidth does not increase linearly with machine size. We use
4140 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4141 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4143 * which yields
4145 * 16MB: 512k
4146 * 32MB: 724k
4147 * 64MB: 1024k
4148 * 128MB: 1448k
4149 * 256MB: 2048k
4150 * 512MB: 2896k
4151 * 1024MB: 4096k
4152 * 2048MB: 5792k
4153 * 4096MB: 8192k
4154 * 8192MB: 11584k
4155 * 16384MB: 16384k
4157 static int __init init_per_zone_pages_min(void)
4159 unsigned long lowmem_kbytes;
4161 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4163 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4164 if (min_free_kbytes < 128)
4165 min_free_kbytes = 128;
4166 if (min_free_kbytes > 65536)
4167 min_free_kbytes = 65536;
4168 setup_per_zone_pages_min();
4169 setup_per_zone_lowmem_reserve();
4170 return 0;
4172 module_init(init_per_zone_pages_min)
4175 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4176 * that we can call two helper functions whenever min_free_kbytes
4177 * changes.
4179 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4180 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4182 proc_dointvec(table, write, file, buffer, length, ppos);
4183 if (write)
4184 setup_per_zone_pages_min();
4185 return 0;
4188 #ifdef CONFIG_NUMA
4189 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4190 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4192 struct zone *zone;
4193 int rc;
4195 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4196 if (rc)
4197 return rc;
4199 for_each_zone(zone)
4200 zone->min_unmapped_pages = (zone->present_pages *
4201 sysctl_min_unmapped_ratio) / 100;
4202 return 0;
4205 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4206 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4208 struct zone *zone;
4209 int rc;
4211 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4212 if (rc)
4213 return rc;
4215 for_each_zone(zone)
4216 zone->min_slab_pages = (zone->present_pages *
4217 sysctl_min_slab_ratio) / 100;
4218 return 0;
4220 #endif
4223 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4224 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4225 * whenever sysctl_lowmem_reserve_ratio changes.
4227 * The reserve ratio obviously has absolutely no relation with the
4228 * pages_min watermarks. The lowmem reserve ratio can only make sense
4229 * if in function of the boot time zone sizes.
4231 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4232 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4234 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4235 setup_per_zone_lowmem_reserve();
4236 return 0;
4240 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4241 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4242 * can have before it gets flushed back to buddy allocator.
4245 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4246 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4248 struct zone *zone;
4249 unsigned int cpu;
4250 int ret;
4252 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4253 if (!write || (ret == -EINVAL))
4254 return ret;
4255 for_each_zone(zone) {
4256 for_each_online_cpu(cpu) {
4257 unsigned long high;
4258 high = zone->present_pages / percpu_pagelist_fraction;
4259 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4262 return 0;
4265 int hashdist = HASHDIST_DEFAULT;
4267 #ifdef CONFIG_NUMA
4268 static int __init set_hashdist(char *str)
4270 if (!str)
4271 return 0;
4272 hashdist = simple_strtoul(str, &str, 0);
4273 return 1;
4275 __setup("hashdist=", set_hashdist);
4276 #endif
4279 * allocate a large system hash table from bootmem
4280 * - it is assumed that the hash table must contain an exact power-of-2
4281 * quantity of entries
4282 * - limit is the number of hash buckets, not the total allocation size
4284 void *__init alloc_large_system_hash(const char *tablename,
4285 unsigned long bucketsize,
4286 unsigned long numentries,
4287 int scale,
4288 int flags,
4289 unsigned int *_hash_shift,
4290 unsigned int *_hash_mask,
4291 unsigned long limit)
4293 unsigned long long max = limit;
4294 unsigned long log2qty, size;
4295 void *table = NULL;
4297 /* allow the kernel cmdline to have a say */
4298 if (!numentries) {
4299 /* round applicable memory size up to nearest megabyte */
4300 numentries = nr_kernel_pages;
4301 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4302 numentries >>= 20 - PAGE_SHIFT;
4303 numentries <<= 20 - PAGE_SHIFT;
4305 /* limit to 1 bucket per 2^scale bytes of low memory */
4306 if (scale > PAGE_SHIFT)
4307 numentries >>= (scale - PAGE_SHIFT);
4308 else
4309 numentries <<= (PAGE_SHIFT - scale);
4311 /* Make sure we've got at least a 0-order allocation.. */
4312 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4313 numentries = PAGE_SIZE / bucketsize;
4315 numentries = roundup_pow_of_two(numentries);
4317 /* limit allocation size to 1/16 total memory by default */
4318 if (max == 0) {
4319 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4320 do_div(max, bucketsize);
4323 if (numentries > max)
4324 numentries = max;
4326 log2qty = ilog2(numentries);
4328 do {
4329 size = bucketsize << log2qty;
4330 if (flags & HASH_EARLY)
4331 table = alloc_bootmem(size);
4332 else if (hashdist)
4333 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4334 else {
4335 unsigned long order;
4336 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4338 table = (void*) __get_free_pages(GFP_ATOMIC, order);
4340 * If bucketsize is not a power-of-two, we may free
4341 * some pages at the end of hash table.
4343 if (table) {
4344 unsigned long alloc_end = (unsigned long)table +
4345 (PAGE_SIZE << order);
4346 unsigned long used = (unsigned long)table +
4347 PAGE_ALIGN(size);
4348 split_page(virt_to_page(table), order);
4349 while (used < alloc_end) {
4350 free_page(used);
4351 used += PAGE_SIZE;
4355 } while (!table && size > PAGE_SIZE && --log2qty);
4357 if (!table)
4358 panic("Failed to allocate %s hash table\n", tablename);
4360 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4361 tablename,
4362 (1U << log2qty),
4363 ilog2(size) - PAGE_SHIFT,
4364 size);
4366 if (_hash_shift)
4367 *_hash_shift = log2qty;
4368 if (_hash_mask)
4369 *_hash_mask = (1 << log2qty) - 1;
4371 return table;
4374 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4375 struct page *pfn_to_page(unsigned long pfn)
4377 return __pfn_to_page(pfn);
4379 unsigned long page_to_pfn(struct page *page)
4381 return __page_to_pfn(page);
4383 EXPORT_SYMBOL(pfn_to_page);
4384 EXPORT_SYMBOL(page_to_pfn);
4385 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4387 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4388 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4389 unsigned long pfn)
4391 #ifdef CONFIG_SPARSEMEM
4392 return __pfn_to_section(pfn)->pageblock_flags;
4393 #else
4394 return zone->pageblock_flags;
4395 #endif /* CONFIG_SPARSEMEM */
4398 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4400 #ifdef CONFIG_SPARSEMEM
4401 pfn &= (PAGES_PER_SECTION-1);
4402 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4403 #else
4404 pfn = pfn - zone->zone_start_pfn;
4405 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4406 #endif /* CONFIG_SPARSEMEM */
4410 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4411 * @page: The page within the block of interest
4412 * @start_bitidx: The first bit of interest to retrieve
4413 * @end_bitidx: The last bit of interest
4414 * returns pageblock_bits flags
4416 unsigned long get_pageblock_flags_group(struct page *page,
4417 int start_bitidx, int end_bitidx)
4419 struct zone *zone;
4420 unsigned long *bitmap;
4421 unsigned long pfn, bitidx;
4422 unsigned long flags = 0;
4423 unsigned long value = 1;
4425 zone = page_zone(page);
4426 pfn = page_to_pfn(page);
4427 bitmap = get_pageblock_bitmap(zone, pfn);
4428 bitidx = pfn_to_bitidx(zone, pfn);
4430 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4431 if (test_bit(bitidx + start_bitidx, bitmap))
4432 flags |= value;
4434 return flags;
4438 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4439 * @page: The page within the block of interest
4440 * @start_bitidx: The first bit of interest
4441 * @end_bitidx: The last bit of interest
4442 * @flags: The flags to set
4444 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4445 int start_bitidx, int end_bitidx)
4447 struct zone *zone;
4448 unsigned long *bitmap;
4449 unsigned long pfn, bitidx;
4450 unsigned long value = 1;
4452 zone = page_zone(page);
4453 pfn = page_to_pfn(page);
4454 bitmap = get_pageblock_bitmap(zone, pfn);
4455 bitidx = pfn_to_bitidx(zone, pfn);
4457 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4458 if (flags & value)
4459 __set_bit(bitidx + start_bitidx, bitmap);
4460 else
4461 __clear_bit(bitidx + start_bitidx, bitmap);
4465 * This is designed as sub function...plz see page_isolation.c also.
4466 * set/clear page block's type to be ISOLATE.
4467 * page allocater never alloc memory from ISOLATE block.
4470 int set_migratetype_isolate(struct page *page)
4472 struct zone *zone;
4473 unsigned long flags;
4474 int ret = -EBUSY;
4476 zone = page_zone(page);
4477 spin_lock_irqsave(&zone->lock, flags);
4479 * In future, more migrate types will be able to be isolation target.
4481 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4482 goto out;
4483 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4484 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4485 ret = 0;
4486 out:
4487 spin_unlock_irqrestore(&zone->lock, flags);
4488 if (!ret)
4489 drain_all_pages();
4490 return ret;
4493 void unset_migratetype_isolate(struct page *page)
4495 struct zone *zone;
4496 unsigned long flags;
4497 zone = page_zone(page);
4498 spin_lock_irqsave(&zone->lock, flags);
4499 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4500 goto out;
4501 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4502 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4503 out:
4504 spin_unlock_irqrestore(&zone->lock, flags);
4507 #ifdef CONFIG_MEMORY_HOTREMOVE
4509 * All pages in the range must be isolated before calling this.
4511 void
4512 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4514 struct page *page;
4515 struct zone *zone;
4516 int order, i;
4517 unsigned long pfn;
4518 unsigned long flags;
4519 /* find the first valid pfn */
4520 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4521 if (pfn_valid(pfn))
4522 break;
4523 if (pfn == end_pfn)
4524 return;
4525 zone = page_zone(pfn_to_page(pfn));
4526 spin_lock_irqsave(&zone->lock, flags);
4527 pfn = start_pfn;
4528 while (pfn < end_pfn) {
4529 if (!pfn_valid(pfn)) {
4530 pfn++;
4531 continue;
4533 page = pfn_to_page(pfn);
4534 BUG_ON(page_count(page));
4535 BUG_ON(!PageBuddy(page));
4536 order = page_order(page);
4537 #ifdef CONFIG_DEBUG_VM
4538 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4539 pfn, 1 << order, end_pfn);
4540 #endif
4541 list_del(&page->lru);
4542 rmv_page_order(page);
4543 zone->free_area[order].nr_free--;
4544 __mod_zone_page_state(zone, NR_FREE_PAGES,
4545 - (1UL << order));
4546 for (i = 0; i < (1 << order); i++)
4547 SetPageReserved((page+i));
4548 pfn += (1 << order);
4550 spin_unlock_irqrestore(&zone->lock, flags);
4552 #endif