Add lock_page_killable
[linux-2.6/mini2440.git] / mm / page_alloc.c
blobb5a58d476c1a66a7cc6ce94adaedfc9e2aff0a06
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>
47 #include <asm/tlbflush.h>
48 #include <asm/div64.h>
49 #include "internal.h"
52 * Array of node states.
54 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
55 [N_POSSIBLE] = NODE_MASK_ALL,
56 [N_ONLINE] = { { [0] = 1UL } },
57 #ifndef CONFIG_NUMA
58 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
59 #ifdef CONFIG_HIGHMEM
60 [N_HIGH_MEMORY] = { { [0] = 1UL } },
61 #endif
62 [N_CPU] = { { [0] = 1UL } },
63 #endif /* NUMA */
65 EXPORT_SYMBOL(node_states);
67 unsigned long totalram_pages __read_mostly;
68 unsigned long totalreserve_pages __read_mostly;
69 long nr_swap_pages;
70 int percpu_pagelist_fraction;
72 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
73 int pageblock_order __read_mostly;
74 #endif
76 static void __free_pages_ok(struct page *page, unsigned int order);
79 * results with 256, 32 in the lowmem_reserve sysctl:
80 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
81 * 1G machine -> (16M dma, 784M normal, 224M high)
82 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
83 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
84 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
86 * TBD: should special case ZONE_DMA32 machines here - in those we normally
87 * don't need any ZONE_NORMAL reservation
89 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
90 #ifdef CONFIG_ZONE_DMA
91 256,
92 #endif
93 #ifdef CONFIG_ZONE_DMA32
94 256,
95 #endif
96 #ifdef CONFIG_HIGHMEM
97 32,
98 #endif
99 32,
102 EXPORT_SYMBOL(totalram_pages);
104 static char * const zone_names[MAX_NR_ZONES] = {
105 #ifdef CONFIG_ZONE_DMA
106 "DMA",
107 #endif
108 #ifdef CONFIG_ZONE_DMA32
109 "DMA32",
110 #endif
111 "Normal",
112 #ifdef CONFIG_HIGHMEM
113 "HighMem",
114 #endif
115 "Movable",
118 int min_free_kbytes = 1024;
120 unsigned long __meminitdata nr_kernel_pages;
121 unsigned long __meminitdata nr_all_pages;
122 static unsigned long __meminitdata dma_reserve;
124 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
126 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
127 * ranges of memory (RAM) that may be registered with add_active_range().
128 * Ranges passed to add_active_range() will be merged if possible
129 * so the number of times add_active_range() can be called is
130 * related to the number of nodes and the number of holes
132 #ifdef CONFIG_MAX_ACTIVE_REGIONS
133 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
134 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
135 #else
136 #if MAX_NUMNODES >= 32
137 /* If there can be many nodes, allow up to 50 holes per node */
138 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
139 #else
140 /* By default, allow up to 256 distinct regions */
141 #define MAX_ACTIVE_REGIONS 256
142 #endif
143 #endif
145 static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
146 static int __meminitdata nr_nodemap_entries;
147 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
148 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
149 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
150 static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
151 static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
152 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
153 unsigned long __initdata required_kernelcore;
154 static unsigned long __initdata required_movablecore;
155 unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
157 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
158 int movable_zone;
159 EXPORT_SYMBOL(movable_zone);
160 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
162 #if MAX_NUMNODES > 1
163 int nr_node_ids __read_mostly = MAX_NUMNODES;
164 EXPORT_SYMBOL(nr_node_ids);
165 #endif
167 int page_group_by_mobility_disabled __read_mostly;
169 static void set_pageblock_migratetype(struct page *page, int migratetype)
171 set_pageblock_flags_group(page, (unsigned long)migratetype,
172 PB_migrate, PB_migrate_end);
175 #ifdef CONFIG_DEBUG_VM
176 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
178 int ret = 0;
179 unsigned seq;
180 unsigned long pfn = page_to_pfn(page);
182 do {
183 seq = zone_span_seqbegin(zone);
184 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
185 ret = 1;
186 else if (pfn < zone->zone_start_pfn)
187 ret = 1;
188 } while (zone_span_seqretry(zone, seq));
190 return ret;
193 static int page_is_consistent(struct zone *zone, struct page *page)
195 if (!pfn_valid_within(page_to_pfn(page)))
196 return 0;
197 if (zone != page_zone(page))
198 return 0;
200 return 1;
203 * Temporary debugging check for pages not lying within a given zone.
205 static int bad_range(struct zone *zone, struct page *page)
207 if (page_outside_zone_boundaries(zone, page))
208 return 1;
209 if (!page_is_consistent(zone, page))
210 return 1;
212 return 0;
214 #else
215 static inline int bad_range(struct zone *zone, struct page *page)
217 return 0;
219 #endif
221 static void bad_page(struct page *page)
223 printk(KERN_EMERG "Bad page state in process '%s'\n"
224 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
225 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
226 KERN_EMERG "Backtrace:\n",
227 current->comm, page, (int)(2*sizeof(unsigned long)),
228 (unsigned long)page->flags, page->mapping,
229 page_mapcount(page), page_count(page));
230 dump_stack();
231 page->flags &= ~(1 << PG_lru |
232 1 << PG_private |
233 1 << PG_locked |
234 1 << PG_active |
235 1 << PG_dirty |
236 1 << PG_reclaim |
237 1 << PG_slab |
238 1 << PG_swapcache |
239 1 << PG_writeback |
240 1 << PG_buddy );
241 set_page_count(page, 0);
242 reset_page_mapcount(page);
243 page->mapping = NULL;
244 add_taint(TAINT_BAD_PAGE);
248 * Higher-order pages are called "compound pages". They are structured thusly:
250 * The first PAGE_SIZE page is called the "head page".
252 * The remaining PAGE_SIZE pages are called "tail pages".
254 * All pages have PG_compound set. All pages have their ->private pointing at
255 * the head page (even the head page has this).
257 * The first tail page's ->lru.next holds the address of the compound page's
258 * put_page() function. Its ->lru.prev holds the order of allocation.
259 * This usage means that zero-order pages may not be compound.
262 static void free_compound_page(struct page *page)
264 __free_pages_ok(page, compound_order(page));
267 static void prep_compound_page(struct page *page, unsigned long order)
269 int i;
270 int nr_pages = 1 << order;
272 set_compound_page_dtor(page, free_compound_page);
273 set_compound_order(page, order);
274 __SetPageHead(page);
275 for (i = 1; i < nr_pages; i++) {
276 struct page *p = page + i;
278 __SetPageTail(p);
279 p->first_page = page;
283 static void destroy_compound_page(struct page *page, unsigned long order)
285 int i;
286 int nr_pages = 1 << order;
288 if (unlikely(compound_order(page) != order))
289 bad_page(page);
291 if (unlikely(!PageHead(page)))
292 bad_page(page);
293 __ClearPageHead(page);
294 for (i = 1; i < nr_pages; i++) {
295 struct page *p = page + i;
297 if (unlikely(!PageTail(p) |
298 (p->first_page != page)))
299 bad_page(page);
300 __ClearPageTail(p);
304 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
306 int i;
309 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
310 * and __GFP_HIGHMEM from hard or soft interrupt context.
312 VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
313 for (i = 0; i < (1 << order); i++)
314 clear_highpage(page + i);
317 static inline void set_page_order(struct page *page, int order)
319 set_page_private(page, order);
320 __SetPageBuddy(page);
323 static inline void rmv_page_order(struct page *page)
325 __ClearPageBuddy(page);
326 set_page_private(page, 0);
330 * Locate the struct page for both the matching buddy in our
331 * pair (buddy1) and the combined O(n+1) page they form (page).
333 * 1) Any buddy B1 will have an order O twin B2 which satisfies
334 * the following equation:
335 * B2 = B1 ^ (1 << O)
336 * For example, if the starting buddy (buddy2) is #8 its order
337 * 1 buddy is #10:
338 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
340 * 2) Any buddy B will have an order O+1 parent P which
341 * satisfies the following equation:
342 * P = B & ~(1 << O)
344 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
346 static inline struct page *
347 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
349 unsigned long buddy_idx = page_idx ^ (1 << order);
351 return page + (buddy_idx - page_idx);
354 static inline unsigned long
355 __find_combined_index(unsigned long page_idx, unsigned int order)
357 return (page_idx & ~(1 << order));
361 * This function checks whether a page is free && is the buddy
362 * we can do coalesce a page and its buddy if
363 * (a) the buddy is not in a hole &&
364 * (b) the buddy is in the buddy system &&
365 * (c) a page and its buddy have the same order &&
366 * (d) a page and its buddy are in the same zone.
368 * For recording whether a page is in the buddy system, we use PG_buddy.
369 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
371 * For recording page's order, we use page_private(page).
373 static inline int page_is_buddy(struct page *page, struct page *buddy,
374 int order)
376 if (!pfn_valid_within(page_to_pfn(buddy)))
377 return 0;
379 if (page_zone_id(page) != page_zone_id(buddy))
380 return 0;
382 if (PageBuddy(buddy) && page_order(buddy) == order) {
383 BUG_ON(page_count(buddy) != 0);
384 return 1;
386 return 0;
390 * Freeing function for a buddy system allocator.
392 * The concept of a buddy system is to maintain direct-mapped table
393 * (containing bit values) for memory blocks of various "orders".
394 * The bottom level table contains the map for the smallest allocatable
395 * units of memory (here, pages), and each level above it describes
396 * pairs of units from the levels below, hence, "buddies".
397 * At a high level, all that happens here is marking the table entry
398 * at the bottom level available, and propagating the changes upward
399 * as necessary, plus some accounting needed to play nicely with other
400 * parts of the VM system.
401 * At each level, we keep a list of pages, which are heads of continuous
402 * free pages of length of (1 << order) and marked with PG_buddy. Page's
403 * order is recorded in page_private(page) field.
404 * So when we are allocating or freeing one, we can derive the state of the
405 * other. That is, if we allocate a small block, and both were
406 * free, the remainder of the region must be split into blocks.
407 * If a block is freed, and its buddy is also free, then this
408 * triggers coalescing into a block of larger size.
410 * -- wli
413 static inline void __free_one_page(struct page *page,
414 struct zone *zone, unsigned int order)
416 unsigned long page_idx;
417 int order_size = 1 << order;
418 int migratetype = get_pageblock_migratetype(page);
420 if (unlikely(PageCompound(page)))
421 destroy_compound_page(page, order);
423 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
425 VM_BUG_ON(page_idx & (order_size - 1));
426 VM_BUG_ON(bad_range(zone, page));
428 __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
429 while (order < MAX_ORDER-1) {
430 unsigned long combined_idx;
431 struct page *buddy;
433 buddy = __page_find_buddy(page, page_idx, order);
434 if (!page_is_buddy(page, buddy, order))
435 break; /* Move the buddy up one level. */
437 list_del(&buddy->lru);
438 zone->free_area[order].nr_free--;
439 rmv_page_order(buddy);
440 combined_idx = __find_combined_index(page_idx, order);
441 page = page + (combined_idx - page_idx);
442 page_idx = combined_idx;
443 order++;
445 set_page_order(page, order);
446 list_add(&page->lru,
447 &zone->free_area[order].free_list[migratetype]);
448 zone->free_area[order].nr_free++;
451 static inline int free_pages_check(struct page *page)
453 if (unlikely(page_mapcount(page) |
454 (page->mapping != NULL) |
455 (page_count(page) != 0) |
456 (page->flags & (
457 1 << PG_lru |
458 1 << PG_private |
459 1 << PG_locked |
460 1 << PG_active |
461 1 << PG_slab |
462 1 << PG_swapcache |
463 1 << PG_writeback |
464 1 << PG_reserved |
465 1 << PG_buddy ))))
466 bad_page(page);
467 if (PageDirty(page))
468 __ClearPageDirty(page);
470 * For now, we report if PG_reserved was found set, but do not
471 * clear it, and do not free the page. But we shall soon need
472 * to do more, for when the ZERO_PAGE count wraps negative.
474 return PageReserved(page);
478 * Frees a list of pages.
479 * Assumes all pages on list are in same zone, and of same order.
480 * count is the number of pages to free.
482 * If the zone was previously in an "all pages pinned" state then look to
483 * see if this freeing clears that state.
485 * And clear the zone's pages_scanned counter, to hold off the "all pages are
486 * pinned" detection logic.
488 static void free_pages_bulk(struct zone *zone, int count,
489 struct list_head *list, int order)
491 spin_lock(&zone->lock);
492 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
493 zone->pages_scanned = 0;
494 while (count--) {
495 struct page *page;
497 VM_BUG_ON(list_empty(list));
498 page = list_entry(list->prev, struct page, lru);
499 /* have to delete it as __free_one_page list manipulates */
500 list_del(&page->lru);
501 __free_one_page(page, zone, order);
503 spin_unlock(&zone->lock);
506 static void free_one_page(struct zone *zone, struct page *page, int order)
508 spin_lock(&zone->lock);
509 zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
510 zone->pages_scanned = 0;
511 __free_one_page(page, zone, order);
512 spin_unlock(&zone->lock);
515 static void __free_pages_ok(struct page *page, unsigned int order)
517 unsigned long flags;
518 int i;
519 int reserved = 0;
521 for (i = 0 ; i < (1 << order) ; ++i)
522 reserved += free_pages_check(page + i);
523 if (reserved)
524 return;
526 if (!PageHighMem(page))
527 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
528 arch_free_page(page, order);
529 kernel_map_pages(page, 1 << order, 0);
531 local_irq_save(flags);
532 __count_vm_events(PGFREE, 1 << order);
533 free_one_page(page_zone(page), page, order);
534 local_irq_restore(flags);
538 * permit the bootmem allocator to evade page validation on high-order frees
540 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
542 if (order == 0) {
543 __ClearPageReserved(page);
544 set_page_count(page, 0);
545 set_page_refcounted(page);
546 __free_page(page);
547 } else {
548 int loop;
550 prefetchw(page);
551 for (loop = 0; loop < BITS_PER_LONG; loop++) {
552 struct page *p = &page[loop];
554 if (loop + 1 < BITS_PER_LONG)
555 prefetchw(p + 1);
556 __ClearPageReserved(p);
557 set_page_count(p, 0);
560 set_page_refcounted(page);
561 __free_pages(page, order);
567 * The order of subdivision here is critical for the IO subsystem.
568 * Please do not alter this order without good reasons and regression
569 * testing. Specifically, as large blocks of memory are subdivided,
570 * the order in which smaller blocks are delivered depends on the order
571 * they're subdivided in this function. This is the primary factor
572 * influencing the order in which pages are delivered to the IO
573 * subsystem according to empirical testing, and this is also justified
574 * by considering the behavior of a buddy system containing a single
575 * large block of memory acted on by a series of small allocations.
576 * This behavior is a critical factor in sglist merging's success.
578 * -- wli
580 static inline void expand(struct zone *zone, struct page *page,
581 int low, int high, struct free_area *area,
582 int migratetype)
584 unsigned long size = 1 << high;
586 while (high > low) {
587 area--;
588 high--;
589 size >>= 1;
590 VM_BUG_ON(bad_range(zone, &page[size]));
591 list_add(&page[size].lru, &area->free_list[migratetype]);
592 area->nr_free++;
593 set_page_order(&page[size], high);
598 * This page is about to be returned from the page allocator
600 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
602 if (unlikely(page_mapcount(page) |
603 (page->mapping != NULL) |
604 (page_count(page) != 0) |
605 (page->flags & (
606 1 << PG_lru |
607 1 << PG_private |
608 1 << PG_locked |
609 1 << PG_active |
610 1 << PG_dirty |
611 1 << PG_slab |
612 1 << PG_swapcache |
613 1 << PG_writeback |
614 1 << PG_reserved |
615 1 << PG_buddy ))))
616 bad_page(page);
619 * For now, we report if PG_reserved was found set, but do not
620 * clear it, and do not allocate the page: as a safety net.
622 if (PageReserved(page))
623 return 1;
625 page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
626 1 << PG_referenced | 1 << PG_arch_1 |
627 1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
628 set_page_private(page, 0);
629 set_page_refcounted(page);
631 arch_alloc_page(page, order);
632 kernel_map_pages(page, 1 << order, 1);
634 if (gfp_flags & __GFP_ZERO)
635 prep_zero_page(page, order, gfp_flags);
637 if (order && (gfp_flags & __GFP_COMP))
638 prep_compound_page(page, order);
640 return 0;
644 * Go through the free lists for the given migratetype and remove
645 * the smallest available page from the freelists
647 static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
648 int migratetype)
650 unsigned int current_order;
651 struct free_area * area;
652 struct page *page;
654 /* Find a page of the appropriate size in the preferred list */
655 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
656 area = &(zone->free_area[current_order]);
657 if (list_empty(&area->free_list[migratetype]))
658 continue;
660 page = list_entry(area->free_list[migratetype].next,
661 struct page, lru);
662 list_del(&page->lru);
663 rmv_page_order(page);
664 area->nr_free--;
665 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
666 expand(zone, page, order, current_order, area, migratetype);
667 return page;
670 return NULL;
675 * This array describes the order lists are fallen back to when
676 * the free lists for the desirable migrate type are depleted
678 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
679 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
680 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE },
681 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
682 [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */
686 * Move the free pages in a range to the free lists of the requested type.
687 * Note that start_page and end_pages are not aligned on a pageblock
688 * boundary. If alignment is required, use move_freepages_block()
690 int move_freepages(struct zone *zone,
691 struct page *start_page, struct page *end_page,
692 int migratetype)
694 struct page *page;
695 unsigned long order;
696 int pages_moved = 0;
698 #ifndef CONFIG_HOLES_IN_ZONE
700 * page_zone is not safe to call in this context when
701 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
702 * anyway as we check zone boundaries in move_freepages_block().
703 * Remove at a later date when no bug reports exist related to
704 * grouping pages by mobility
706 BUG_ON(page_zone(start_page) != page_zone(end_page));
707 #endif
709 for (page = start_page; page <= end_page;) {
710 if (!pfn_valid_within(page_to_pfn(page))) {
711 page++;
712 continue;
715 if (!PageBuddy(page)) {
716 page++;
717 continue;
720 order = page_order(page);
721 list_del(&page->lru);
722 list_add(&page->lru,
723 &zone->free_area[order].free_list[migratetype]);
724 page += 1 << order;
725 pages_moved += 1 << order;
728 return pages_moved;
731 int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
733 unsigned long start_pfn, end_pfn;
734 struct page *start_page, *end_page;
736 start_pfn = page_to_pfn(page);
737 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
738 start_page = pfn_to_page(start_pfn);
739 end_page = start_page + pageblock_nr_pages - 1;
740 end_pfn = start_pfn + pageblock_nr_pages - 1;
742 /* Do not cross zone boundaries */
743 if (start_pfn < zone->zone_start_pfn)
744 start_page = page;
745 if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
746 return 0;
748 return move_freepages(zone, start_page, end_page, migratetype);
751 /* Remove an element from the buddy allocator from the fallback list */
752 static struct page *__rmqueue_fallback(struct zone *zone, int order,
753 int start_migratetype)
755 struct free_area * area;
756 int current_order;
757 struct page *page;
758 int migratetype, i;
760 /* Find the largest possible block of pages in the other list */
761 for (current_order = MAX_ORDER-1; current_order >= order;
762 --current_order) {
763 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
764 migratetype = fallbacks[start_migratetype][i];
766 /* MIGRATE_RESERVE handled later if necessary */
767 if (migratetype == MIGRATE_RESERVE)
768 continue;
770 area = &(zone->free_area[current_order]);
771 if (list_empty(&area->free_list[migratetype]))
772 continue;
774 page = list_entry(area->free_list[migratetype].next,
775 struct page, lru);
776 area->nr_free--;
779 * If breaking a large block of pages, move all free
780 * pages to the preferred allocation list. If falling
781 * back for a reclaimable kernel allocation, be more
782 * agressive about taking ownership of free pages
784 if (unlikely(current_order >= (pageblock_order >> 1)) ||
785 start_migratetype == MIGRATE_RECLAIMABLE) {
786 unsigned long pages;
787 pages = move_freepages_block(zone, page,
788 start_migratetype);
790 /* Claim the whole block if over half of it is free */
791 if (pages >= (1 << (pageblock_order-1)))
792 set_pageblock_migratetype(page,
793 start_migratetype);
795 migratetype = start_migratetype;
798 /* Remove the page from the freelists */
799 list_del(&page->lru);
800 rmv_page_order(page);
801 __mod_zone_page_state(zone, NR_FREE_PAGES,
802 -(1UL << order));
804 if (current_order == pageblock_order)
805 set_pageblock_migratetype(page,
806 start_migratetype);
808 expand(zone, page, order, current_order, area, migratetype);
809 return page;
813 /* Use MIGRATE_RESERVE rather than fail an allocation */
814 return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
818 * Do the hard work of removing an element from the buddy allocator.
819 * Call me with the zone->lock already held.
821 static struct page *__rmqueue(struct zone *zone, unsigned int order,
822 int migratetype)
824 struct page *page;
826 page = __rmqueue_smallest(zone, order, migratetype);
828 if (unlikely(!page))
829 page = __rmqueue_fallback(zone, order, migratetype);
831 return page;
835 * Obtain a specified number of elements from the buddy allocator, all under
836 * a single hold of the lock, for efficiency. Add them to the supplied list.
837 * Returns the number of new pages which were placed at *list.
839 static int rmqueue_bulk(struct zone *zone, unsigned int order,
840 unsigned long count, struct list_head *list,
841 int migratetype)
843 int i;
845 spin_lock(&zone->lock);
846 for (i = 0; i < count; ++i) {
847 struct page *page = __rmqueue(zone, order, migratetype);
848 if (unlikely(page == NULL))
849 break;
850 list_add(&page->lru, list);
851 set_page_private(page, migratetype);
853 spin_unlock(&zone->lock);
854 return i;
857 #ifdef CONFIG_NUMA
859 * Called from the vmstat counter updater to drain pagesets of this
860 * currently executing processor on remote nodes after they have
861 * expired.
863 * Note that this function must be called with the thread pinned to
864 * a single processor.
866 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
868 unsigned long flags;
869 int to_drain;
871 local_irq_save(flags);
872 if (pcp->count >= pcp->batch)
873 to_drain = pcp->batch;
874 else
875 to_drain = pcp->count;
876 free_pages_bulk(zone, to_drain, &pcp->list, 0);
877 pcp->count -= to_drain;
878 local_irq_restore(flags);
880 #endif
882 static void __drain_pages(unsigned int cpu)
884 unsigned long flags;
885 struct zone *zone;
886 int i;
888 for_each_zone(zone) {
889 struct per_cpu_pageset *pset;
891 if (!populated_zone(zone))
892 continue;
894 pset = zone_pcp(zone, cpu);
895 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
896 struct per_cpu_pages *pcp;
898 pcp = &pset->pcp[i];
899 local_irq_save(flags);
900 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
901 pcp->count = 0;
902 local_irq_restore(flags);
907 #ifdef CONFIG_HIBERNATION
909 void mark_free_pages(struct zone *zone)
911 unsigned long pfn, max_zone_pfn;
912 unsigned long flags;
913 int order, t;
914 struct list_head *curr;
916 if (!zone->spanned_pages)
917 return;
919 spin_lock_irqsave(&zone->lock, flags);
921 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
922 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
923 if (pfn_valid(pfn)) {
924 struct page *page = pfn_to_page(pfn);
926 if (!swsusp_page_is_forbidden(page))
927 swsusp_unset_page_free(page);
930 for_each_migratetype_order(order, t) {
931 list_for_each(curr, &zone->free_area[order].free_list[t]) {
932 unsigned long i;
934 pfn = page_to_pfn(list_entry(curr, struct page, lru));
935 for (i = 0; i < (1UL << order); i++)
936 swsusp_set_page_free(pfn_to_page(pfn + i));
939 spin_unlock_irqrestore(&zone->lock, flags);
941 #endif /* CONFIG_PM */
944 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
946 void drain_local_pages(void)
948 unsigned long flags;
950 local_irq_save(flags);
951 __drain_pages(smp_processor_id());
952 local_irq_restore(flags);
955 void smp_drain_local_pages(void *arg)
957 drain_local_pages();
961 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
963 void drain_all_local_pages(void)
965 unsigned long flags;
967 local_irq_save(flags);
968 __drain_pages(smp_processor_id());
969 local_irq_restore(flags);
971 smp_call_function(smp_drain_local_pages, NULL, 0, 1);
975 * Free a 0-order page
977 static void fastcall free_hot_cold_page(struct page *page, int cold)
979 struct zone *zone = page_zone(page);
980 struct per_cpu_pages *pcp;
981 unsigned long flags;
983 if (PageAnon(page))
984 page->mapping = NULL;
985 if (free_pages_check(page))
986 return;
988 if (!PageHighMem(page))
989 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
990 arch_free_page(page, 0);
991 kernel_map_pages(page, 1, 0);
993 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
994 local_irq_save(flags);
995 __count_vm_event(PGFREE);
996 list_add(&page->lru, &pcp->list);
997 set_page_private(page, get_pageblock_migratetype(page));
998 pcp->count++;
999 if (pcp->count >= pcp->high) {
1000 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1001 pcp->count -= pcp->batch;
1003 local_irq_restore(flags);
1004 put_cpu();
1007 void fastcall free_hot_page(struct page *page)
1009 free_hot_cold_page(page, 0);
1012 void fastcall free_cold_page(struct page *page)
1014 free_hot_cold_page(page, 1);
1018 * split_page takes a non-compound higher-order page, and splits it into
1019 * n (1<<order) sub-pages: page[0..n]
1020 * Each sub-page must be freed individually.
1022 * Note: this is probably too low level an operation for use in drivers.
1023 * Please consult with lkml before using this in your driver.
1025 void split_page(struct page *page, unsigned int order)
1027 int i;
1029 VM_BUG_ON(PageCompound(page));
1030 VM_BUG_ON(!page_count(page));
1031 for (i = 1; i < (1 << order); i++)
1032 set_page_refcounted(page + i);
1036 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1037 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1038 * or two.
1040 static struct page *buffered_rmqueue(struct zonelist *zonelist,
1041 struct zone *zone, int order, gfp_t gfp_flags)
1043 unsigned long flags;
1044 struct page *page;
1045 int cold = !!(gfp_flags & __GFP_COLD);
1046 int cpu;
1047 int migratetype = allocflags_to_migratetype(gfp_flags);
1049 again:
1050 cpu = get_cpu();
1051 if (likely(order == 0)) {
1052 struct per_cpu_pages *pcp;
1054 pcp = &zone_pcp(zone, cpu)->pcp[cold];
1055 local_irq_save(flags);
1056 if (!pcp->count) {
1057 pcp->count = rmqueue_bulk(zone, 0,
1058 pcp->batch, &pcp->list, migratetype);
1059 if (unlikely(!pcp->count))
1060 goto failed;
1063 /* Find a page of the appropriate migrate type */
1064 list_for_each_entry(page, &pcp->list, lru)
1065 if (page_private(page) == migratetype)
1066 break;
1068 /* Allocate more to the pcp list if necessary */
1069 if (unlikely(&page->lru == &pcp->list)) {
1070 pcp->count += rmqueue_bulk(zone, 0,
1071 pcp->batch, &pcp->list, migratetype);
1072 page = list_entry(pcp->list.next, struct page, lru);
1075 list_del(&page->lru);
1076 pcp->count--;
1077 } else {
1078 spin_lock_irqsave(&zone->lock, flags);
1079 page = __rmqueue(zone, order, migratetype);
1080 spin_unlock(&zone->lock);
1081 if (!page)
1082 goto failed;
1085 __count_zone_vm_events(PGALLOC, zone, 1 << order);
1086 zone_statistics(zonelist, zone);
1087 local_irq_restore(flags);
1088 put_cpu();
1090 VM_BUG_ON(bad_range(zone, page));
1091 if (prep_new_page(page, order, gfp_flags))
1092 goto again;
1093 return page;
1095 failed:
1096 local_irq_restore(flags);
1097 put_cpu();
1098 return NULL;
1101 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1102 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1103 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1104 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1105 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1106 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1107 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1109 #ifdef CONFIG_FAIL_PAGE_ALLOC
1111 static struct fail_page_alloc_attr {
1112 struct fault_attr attr;
1114 u32 ignore_gfp_highmem;
1115 u32 ignore_gfp_wait;
1116 u32 min_order;
1118 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1120 struct dentry *ignore_gfp_highmem_file;
1121 struct dentry *ignore_gfp_wait_file;
1122 struct dentry *min_order_file;
1124 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1126 } fail_page_alloc = {
1127 .attr = FAULT_ATTR_INITIALIZER,
1128 .ignore_gfp_wait = 1,
1129 .ignore_gfp_highmem = 1,
1130 .min_order = 1,
1133 static int __init setup_fail_page_alloc(char *str)
1135 return setup_fault_attr(&fail_page_alloc.attr, str);
1137 __setup("fail_page_alloc=", setup_fail_page_alloc);
1139 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1141 if (order < fail_page_alloc.min_order)
1142 return 0;
1143 if (gfp_mask & __GFP_NOFAIL)
1144 return 0;
1145 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1146 return 0;
1147 if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1148 return 0;
1150 return should_fail(&fail_page_alloc.attr, 1 << order);
1153 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1155 static int __init fail_page_alloc_debugfs(void)
1157 mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1158 struct dentry *dir;
1159 int err;
1161 err = init_fault_attr_dentries(&fail_page_alloc.attr,
1162 "fail_page_alloc");
1163 if (err)
1164 return err;
1165 dir = fail_page_alloc.attr.dentries.dir;
1167 fail_page_alloc.ignore_gfp_wait_file =
1168 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1169 &fail_page_alloc.ignore_gfp_wait);
1171 fail_page_alloc.ignore_gfp_highmem_file =
1172 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1173 &fail_page_alloc.ignore_gfp_highmem);
1174 fail_page_alloc.min_order_file =
1175 debugfs_create_u32("min-order", mode, dir,
1176 &fail_page_alloc.min_order);
1178 if (!fail_page_alloc.ignore_gfp_wait_file ||
1179 !fail_page_alloc.ignore_gfp_highmem_file ||
1180 !fail_page_alloc.min_order_file) {
1181 err = -ENOMEM;
1182 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1183 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1184 debugfs_remove(fail_page_alloc.min_order_file);
1185 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1188 return err;
1191 late_initcall(fail_page_alloc_debugfs);
1193 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1195 #else /* CONFIG_FAIL_PAGE_ALLOC */
1197 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1199 return 0;
1202 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1205 * Return 1 if free pages are above 'mark'. This takes into account the order
1206 * of the allocation.
1208 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1209 int classzone_idx, int alloc_flags)
1211 /* free_pages my go negative - that's OK */
1212 long min = mark;
1213 long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1214 int o;
1216 if (alloc_flags & ALLOC_HIGH)
1217 min -= min / 2;
1218 if (alloc_flags & ALLOC_HARDER)
1219 min -= min / 4;
1221 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1222 return 0;
1223 for (o = 0; o < order; o++) {
1224 /* At the next order, this order's pages become unavailable */
1225 free_pages -= z->free_area[o].nr_free << o;
1227 /* Require fewer higher order pages to be free */
1228 min >>= 1;
1230 if (free_pages <= min)
1231 return 0;
1233 return 1;
1236 #ifdef CONFIG_NUMA
1238 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1239 * skip over zones that are not allowed by the cpuset, or that have
1240 * been recently (in last second) found to be nearly full. See further
1241 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1242 * that have to skip over a lot of full or unallowed zones.
1244 * If the zonelist cache is present in the passed in zonelist, then
1245 * returns a pointer to the allowed node mask (either the current
1246 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1248 * If the zonelist cache is not available for this zonelist, does
1249 * nothing and returns NULL.
1251 * If the fullzones BITMAP in the zonelist cache is stale (more than
1252 * a second since last zap'd) then we zap it out (clear its bits.)
1254 * We hold off even calling zlc_setup, until after we've checked the
1255 * first zone in the zonelist, on the theory that most allocations will
1256 * be satisfied from that first zone, so best to examine that zone as
1257 * quickly as we can.
1259 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1261 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1262 nodemask_t *allowednodes; /* zonelist_cache approximation */
1264 zlc = zonelist->zlcache_ptr;
1265 if (!zlc)
1266 return NULL;
1268 if (jiffies - zlc->last_full_zap > 1 * HZ) {
1269 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1270 zlc->last_full_zap = jiffies;
1273 allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1274 &cpuset_current_mems_allowed :
1275 &node_states[N_HIGH_MEMORY];
1276 return allowednodes;
1280 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1281 * if it is worth looking at further for free memory:
1282 * 1) Check that the zone isn't thought to be full (doesn't have its
1283 * bit set in the zonelist_cache fullzones BITMAP).
1284 * 2) Check that the zones node (obtained from the zonelist_cache
1285 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1286 * Return true (non-zero) if zone is worth looking at further, or
1287 * else return false (zero) if it is not.
1289 * This check -ignores- the distinction between various watermarks,
1290 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1291 * found to be full for any variation of these watermarks, it will
1292 * be considered full for up to one second by all requests, unless
1293 * we are so low on memory on all allowed nodes that we are forced
1294 * into the second scan of the zonelist.
1296 * In the second scan we ignore this zonelist cache and exactly
1297 * apply the watermarks to all zones, even it is slower to do so.
1298 * We are low on memory in the second scan, and should leave no stone
1299 * unturned looking for a free page.
1301 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1302 nodemask_t *allowednodes)
1304 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1305 int i; /* index of *z in zonelist zones */
1306 int n; /* node that zone *z is on */
1308 zlc = zonelist->zlcache_ptr;
1309 if (!zlc)
1310 return 1;
1312 i = z - zonelist->zones;
1313 n = zlc->z_to_n[i];
1315 /* This zone is worth trying if it is allowed but not full */
1316 return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1320 * Given 'z' scanning a zonelist, set the corresponding bit in
1321 * zlc->fullzones, so that subsequent attempts to allocate a page
1322 * from that zone don't waste time re-examining it.
1324 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1326 struct zonelist_cache *zlc; /* cached zonelist speedup info */
1327 int i; /* index of *z in zonelist zones */
1329 zlc = zonelist->zlcache_ptr;
1330 if (!zlc)
1331 return;
1333 i = z - zonelist->zones;
1335 set_bit(i, zlc->fullzones);
1338 #else /* CONFIG_NUMA */
1340 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1342 return NULL;
1345 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1346 nodemask_t *allowednodes)
1348 return 1;
1351 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1354 #endif /* CONFIG_NUMA */
1357 * get_page_from_freelist goes through the zonelist trying to allocate
1358 * a page.
1360 static struct page *
1361 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1362 struct zonelist *zonelist, int alloc_flags)
1364 struct zone **z;
1365 struct page *page = NULL;
1366 int classzone_idx = zone_idx(zonelist->zones[0]);
1367 struct zone *zone;
1368 nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1369 int zlc_active = 0; /* set if using zonelist_cache */
1370 int did_zlc_setup = 0; /* just call zlc_setup() one time */
1371 enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
1373 zonelist_scan:
1375 * Scan zonelist, looking for a zone with enough free.
1376 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1378 z = zonelist->zones;
1380 do {
1382 * In NUMA, this could be a policy zonelist which contains
1383 * zones that may not be allowed by the current gfp_mask.
1384 * Check the zone is allowed by the current flags
1386 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1387 if (highest_zoneidx == -1)
1388 highest_zoneidx = gfp_zone(gfp_mask);
1389 if (zone_idx(*z) > highest_zoneidx)
1390 continue;
1393 if (NUMA_BUILD && zlc_active &&
1394 !zlc_zone_worth_trying(zonelist, z, allowednodes))
1395 continue;
1396 zone = *z;
1397 if ((alloc_flags & ALLOC_CPUSET) &&
1398 !cpuset_zone_allowed_softwall(zone, gfp_mask))
1399 goto try_next_zone;
1401 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1402 unsigned long mark;
1403 if (alloc_flags & ALLOC_WMARK_MIN)
1404 mark = zone->pages_min;
1405 else if (alloc_flags & ALLOC_WMARK_LOW)
1406 mark = zone->pages_low;
1407 else
1408 mark = zone->pages_high;
1409 if (!zone_watermark_ok(zone, order, mark,
1410 classzone_idx, alloc_flags)) {
1411 if (!zone_reclaim_mode ||
1412 !zone_reclaim(zone, gfp_mask, order))
1413 goto this_zone_full;
1417 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1418 if (page)
1419 break;
1420 this_zone_full:
1421 if (NUMA_BUILD)
1422 zlc_mark_zone_full(zonelist, z);
1423 try_next_zone:
1424 if (NUMA_BUILD && !did_zlc_setup) {
1425 /* we do zlc_setup after the first zone is tried */
1426 allowednodes = zlc_setup(zonelist, alloc_flags);
1427 zlc_active = 1;
1428 did_zlc_setup = 1;
1430 } while (*(++z) != NULL);
1432 if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1433 /* Disable zlc cache for second zonelist scan */
1434 zlc_active = 0;
1435 goto zonelist_scan;
1437 return page;
1441 * This is the 'heart' of the zoned buddy allocator.
1443 struct page * fastcall
1444 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1445 struct zonelist *zonelist)
1447 const gfp_t wait = gfp_mask & __GFP_WAIT;
1448 struct zone **z;
1449 struct page *page;
1450 struct reclaim_state reclaim_state;
1451 struct task_struct *p = current;
1452 int do_retry;
1453 int alloc_flags;
1454 int did_some_progress;
1456 might_sleep_if(wait);
1458 if (should_fail_alloc_page(gfp_mask, order))
1459 return NULL;
1461 restart:
1462 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
1464 if (unlikely(*z == NULL)) {
1466 * Happens if we have an empty zonelist as a result of
1467 * GFP_THISNODE being used on a memoryless node
1469 return NULL;
1472 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1473 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1474 if (page)
1475 goto got_pg;
1478 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1479 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1480 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1481 * using a larger set of nodes after it has established that the
1482 * allowed per node queues are empty and that nodes are
1483 * over allocated.
1485 if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1486 goto nopage;
1488 for (z = zonelist->zones; *z; z++)
1489 wakeup_kswapd(*z, order);
1492 * OK, we're below the kswapd watermark and have kicked background
1493 * reclaim. Now things get more complex, so set up alloc_flags according
1494 * to how we want to proceed.
1496 * The caller may dip into page reserves a bit more if the caller
1497 * cannot run direct reclaim, or if the caller has realtime scheduling
1498 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1499 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1501 alloc_flags = ALLOC_WMARK_MIN;
1502 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1503 alloc_flags |= ALLOC_HARDER;
1504 if (gfp_mask & __GFP_HIGH)
1505 alloc_flags |= ALLOC_HIGH;
1506 if (wait)
1507 alloc_flags |= ALLOC_CPUSET;
1510 * Go through the zonelist again. Let __GFP_HIGH and allocations
1511 * coming from realtime tasks go deeper into reserves.
1513 * This is the last chance, in general, before the goto nopage.
1514 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1515 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1517 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1518 if (page)
1519 goto got_pg;
1521 /* This allocation should allow future memory freeing. */
1523 rebalance:
1524 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1525 && !in_interrupt()) {
1526 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1527 nofail_alloc:
1528 /* go through the zonelist yet again, ignoring mins */
1529 page = get_page_from_freelist(gfp_mask, order,
1530 zonelist, ALLOC_NO_WATERMARKS);
1531 if (page)
1532 goto got_pg;
1533 if (gfp_mask & __GFP_NOFAIL) {
1534 congestion_wait(WRITE, HZ/50);
1535 goto nofail_alloc;
1538 goto nopage;
1541 /* Atomic allocations - we can't balance anything */
1542 if (!wait)
1543 goto nopage;
1545 cond_resched();
1547 /* We now go into synchronous reclaim */
1548 cpuset_memory_pressure_bump();
1549 p->flags |= PF_MEMALLOC;
1550 reclaim_state.reclaimed_slab = 0;
1551 p->reclaim_state = &reclaim_state;
1553 did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
1555 p->reclaim_state = NULL;
1556 p->flags &= ~PF_MEMALLOC;
1558 cond_resched();
1560 if (order != 0)
1561 drain_all_local_pages();
1563 if (likely(did_some_progress)) {
1564 page = get_page_from_freelist(gfp_mask, order,
1565 zonelist, alloc_flags);
1566 if (page)
1567 goto got_pg;
1568 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1569 if (!try_set_zone_oom(zonelist)) {
1570 schedule_timeout_uninterruptible(1);
1571 goto restart;
1575 * Go through the zonelist yet one more time, keep
1576 * very high watermark here, this is only to catch
1577 * a parallel oom killing, we must fail if we're still
1578 * under heavy pressure.
1580 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1581 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1582 if (page) {
1583 clear_zonelist_oom(zonelist);
1584 goto got_pg;
1587 /* The OOM killer will not help higher order allocs so fail */
1588 if (order > PAGE_ALLOC_COSTLY_ORDER) {
1589 clear_zonelist_oom(zonelist);
1590 goto nopage;
1593 out_of_memory(zonelist, gfp_mask, order);
1594 clear_zonelist_oom(zonelist);
1595 goto restart;
1599 * Don't let big-order allocations loop unless the caller explicitly
1600 * requests that. Wait for some write requests to complete then retry.
1602 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1603 * <= 3, but that may not be true in other implementations.
1605 do_retry = 0;
1606 if (!(gfp_mask & __GFP_NORETRY)) {
1607 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1608 (gfp_mask & __GFP_REPEAT))
1609 do_retry = 1;
1610 if (gfp_mask & __GFP_NOFAIL)
1611 do_retry = 1;
1613 if (do_retry) {
1614 congestion_wait(WRITE, HZ/50);
1615 goto rebalance;
1618 nopage:
1619 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1620 printk(KERN_WARNING "%s: page allocation failure."
1621 " order:%d, mode:0x%x\n",
1622 p->comm, order, gfp_mask);
1623 dump_stack();
1624 show_mem();
1626 got_pg:
1627 return page;
1630 EXPORT_SYMBOL(__alloc_pages);
1633 * Common helper functions.
1635 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1637 struct page * page;
1638 page = alloc_pages(gfp_mask, order);
1639 if (!page)
1640 return 0;
1641 return (unsigned long) page_address(page);
1644 EXPORT_SYMBOL(__get_free_pages);
1646 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1648 struct page * page;
1651 * get_zeroed_page() returns a 32-bit address, which cannot represent
1652 * a highmem page
1654 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1656 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1657 if (page)
1658 return (unsigned long) page_address(page);
1659 return 0;
1662 EXPORT_SYMBOL(get_zeroed_page);
1664 void __pagevec_free(struct pagevec *pvec)
1666 int i = pagevec_count(pvec);
1668 while (--i >= 0)
1669 free_hot_cold_page(pvec->pages[i], pvec->cold);
1672 fastcall void __free_pages(struct page *page, unsigned int order)
1674 if (put_page_testzero(page)) {
1675 if (order == 0)
1676 free_hot_page(page);
1677 else
1678 __free_pages_ok(page, order);
1682 EXPORT_SYMBOL(__free_pages);
1684 fastcall void free_pages(unsigned long addr, unsigned int order)
1686 if (addr != 0) {
1687 VM_BUG_ON(!virt_addr_valid((void *)addr));
1688 __free_pages(virt_to_page((void *)addr), order);
1692 EXPORT_SYMBOL(free_pages);
1694 static unsigned int nr_free_zone_pages(int offset)
1696 /* Just pick one node, since fallback list is circular */
1697 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1698 unsigned int sum = 0;
1700 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1701 struct zone **zonep = zonelist->zones;
1702 struct zone *zone;
1704 for (zone = *zonep++; zone; zone = *zonep++) {
1705 unsigned long size = zone->present_pages;
1706 unsigned long high = zone->pages_high;
1707 if (size > high)
1708 sum += size - high;
1711 return sum;
1715 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1717 unsigned int nr_free_buffer_pages(void)
1719 return nr_free_zone_pages(gfp_zone(GFP_USER));
1721 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1724 * Amount of free RAM allocatable within all zones
1726 unsigned int nr_free_pagecache_pages(void)
1728 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1731 static inline void show_node(struct zone *zone)
1733 if (NUMA_BUILD)
1734 printk("Node %d ", zone_to_nid(zone));
1737 void si_meminfo(struct sysinfo *val)
1739 val->totalram = totalram_pages;
1740 val->sharedram = 0;
1741 val->freeram = global_page_state(NR_FREE_PAGES);
1742 val->bufferram = nr_blockdev_pages();
1743 val->totalhigh = totalhigh_pages;
1744 val->freehigh = nr_free_highpages();
1745 val->mem_unit = PAGE_SIZE;
1748 EXPORT_SYMBOL(si_meminfo);
1750 #ifdef CONFIG_NUMA
1751 void si_meminfo_node(struct sysinfo *val, int nid)
1753 pg_data_t *pgdat = NODE_DATA(nid);
1755 val->totalram = pgdat->node_present_pages;
1756 val->freeram = node_page_state(nid, NR_FREE_PAGES);
1757 #ifdef CONFIG_HIGHMEM
1758 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1759 val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1760 NR_FREE_PAGES);
1761 #else
1762 val->totalhigh = 0;
1763 val->freehigh = 0;
1764 #endif
1765 val->mem_unit = PAGE_SIZE;
1767 #endif
1769 #define K(x) ((x) << (PAGE_SHIFT-10))
1772 * Show free area list (used inside shift_scroll-lock stuff)
1773 * We also calculate the percentage fragmentation. We do this by counting the
1774 * memory on each free list with the exception of the first item on the list.
1776 void show_free_areas(void)
1778 int cpu;
1779 struct zone *zone;
1781 for_each_zone(zone) {
1782 if (!populated_zone(zone))
1783 continue;
1785 show_node(zone);
1786 printk("%s per-cpu:\n", zone->name);
1788 for_each_online_cpu(cpu) {
1789 struct per_cpu_pageset *pageset;
1791 pageset = zone_pcp(zone, cpu);
1793 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1794 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1795 cpu, pageset->pcp[0].high,
1796 pageset->pcp[0].batch, pageset->pcp[0].count,
1797 pageset->pcp[1].high, pageset->pcp[1].batch,
1798 pageset->pcp[1].count);
1802 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1803 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1804 global_page_state(NR_ACTIVE),
1805 global_page_state(NR_INACTIVE),
1806 global_page_state(NR_FILE_DIRTY),
1807 global_page_state(NR_WRITEBACK),
1808 global_page_state(NR_UNSTABLE_NFS),
1809 global_page_state(NR_FREE_PAGES),
1810 global_page_state(NR_SLAB_RECLAIMABLE) +
1811 global_page_state(NR_SLAB_UNRECLAIMABLE),
1812 global_page_state(NR_FILE_MAPPED),
1813 global_page_state(NR_PAGETABLE),
1814 global_page_state(NR_BOUNCE));
1816 for_each_zone(zone) {
1817 int i;
1819 if (!populated_zone(zone))
1820 continue;
1822 show_node(zone);
1823 printk("%s"
1824 " free:%lukB"
1825 " min:%lukB"
1826 " low:%lukB"
1827 " high:%lukB"
1828 " active:%lukB"
1829 " inactive:%lukB"
1830 " present:%lukB"
1831 " pages_scanned:%lu"
1832 " all_unreclaimable? %s"
1833 "\n",
1834 zone->name,
1835 K(zone_page_state(zone, NR_FREE_PAGES)),
1836 K(zone->pages_min),
1837 K(zone->pages_low),
1838 K(zone->pages_high),
1839 K(zone_page_state(zone, NR_ACTIVE)),
1840 K(zone_page_state(zone, NR_INACTIVE)),
1841 K(zone->present_pages),
1842 zone->pages_scanned,
1843 (zone_is_all_unreclaimable(zone) ? "yes" : "no")
1845 printk("lowmem_reserve[]:");
1846 for (i = 0; i < MAX_NR_ZONES; i++)
1847 printk(" %lu", zone->lowmem_reserve[i]);
1848 printk("\n");
1851 for_each_zone(zone) {
1852 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1854 if (!populated_zone(zone))
1855 continue;
1857 show_node(zone);
1858 printk("%s: ", zone->name);
1860 spin_lock_irqsave(&zone->lock, flags);
1861 for (order = 0; order < MAX_ORDER; order++) {
1862 nr[order] = zone->free_area[order].nr_free;
1863 total += nr[order] << order;
1865 spin_unlock_irqrestore(&zone->lock, flags);
1866 for (order = 0; order < MAX_ORDER; order++)
1867 printk("%lu*%lukB ", nr[order], K(1UL) << order);
1868 printk("= %lukB\n", K(total));
1871 show_swap_cache_info();
1875 * Builds allocation fallback zone lists.
1877 * Add all populated zones of a node to the zonelist.
1879 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1880 int nr_zones, enum zone_type zone_type)
1882 struct zone *zone;
1884 BUG_ON(zone_type >= MAX_NR_ZONES);
1885 zone_type++;
1887 do {
1888 zone_type--;
1889 zone = pgdat->node_zones + zone_type;
1890 if (populated_zone(zone)) {
1891 zonelist->zones[nr_zones++] = zone;
1892 check_highest_zone(zone_type);
1895 } while (zone_type);
1896 return nr_zones;
1901 * zonelist_order:
1902 * 0 = automatic detection of better ordering.
1903 * 1 = order by ([node] distance, -zonetype)
1904 * 2 = order by (-zonetype, [node] distance)
1906 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1907 * the same zonelist. So only NUMA can configure this param.
1909 #define ZONELIST_ORDER_DEFAULT 0
1910 #define ZONELIST_ORDER_NODE 1
1911 #define ZONELIST_ORDER_ZONE 2
1913 /* zonelist order in the kernel.
1914 * set_zonelist_order() will set this to NODE or ZONE.
1916 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
1917 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
1920 #ifdef CONFIG_NUMA
1921 /* The value user specified ....changed by config */
1922 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1923 /* string for sysctl */
1924 #define NUMA_ZONELIST_ORDER_LEN 16
1925 char numa_zonelist_order[16] = "default";
1928 * interface for configure zonelist ordering.
1929 * command line option "numa_zonelist_order"
1930 * = "[dD]efault - default, automatic configuration.
1931 * = "[nN]ode - order by node locality, then by zone within node
1932 * = "[zZ]one - order by zone, then by locality within zone
1935 static int __parse_numa_zonelist_order(char *s)
1937 if (*s == 'd' || *s == 'D') {
1938 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
1939 } else if (*s == 'n' || *s == 'N') {
1940 user_zonelist_order = ZONELIST_ORDER_NODE;
1941 } else if (*s == 'z' || *s == 'Z') {
1942 user_zonelist_order = ZONELIST_ORDER_ZONE;
1943 } else {
1944 printk(KERN_WARNING
1945 "Ignoring invalid numa_zonelist_order value: "
1946 "%s\n", s);
1947 return -EINVAL;
1949 return 0;
1952 static __init int setup_numa_zonelist_order(char *s)
1954 if (s)
1955 return __parse_numa_zonelist_order(s);
1956 return 0;
1958 early_param("numa_zonelist_order", setup_numa_zonelist_order);
1961 * sysctl handler for numa_zonelist_order
1963 int numa_zonelist_order_handler(ctl_table *table, int write,
1964 struct file *file, void __user *buffer, size_t *length,
1965 loff_t *ppos)
1967 char saved_string[NUMA_ZONELIST_ORDER_LEN];
1968 int ret;
1970 if (write)
1971 strncpy(saved_string, (char*)table->data,
1972 NUMA_ZONELIST_ORDER_LEN);
1973 ret = proc_dostring(table, write, file, buffer, length, ppos);
1974 if (ret)
1975 return ret;
1976 if (write) {
1977 int oldval = user_zonelist_order;
1978 if (__parse_numa_zonelist_order((char*)table->data)) {
1980 * bogus value. restore saved string
1982 strncpy((char*)table->data, saved_string,
1983 NUMA_ZONELIST_ORDER_LEN);
1984 user_zonelist_order = oldval;
1985 } else if (oldval != user_zonelist_order)
1986 build_all_zonelists();
1988 return 0;
1992 #define MAX_NODE_LOAD (num_online_nodes())
1993 static int node_load[MAX_NUMNODES];
1996 * find_next_best_node - find the next node that should appear in a given node's fallback list
1997 * @node: node whose fallback list we're appending
1998 * @used_node_mask: nodemask_t of already used nodes
2000 * We use a number of factors to determine which is the next node that should
2001 * appear on a given node's fallback list. The node should not have appeared
2002 * already in @node's fallback list, and it should be the next closest node
2003 * according to the distance array (which contains arbitrary distance values
2004 * from each node to each node in the system), and should also prefer nodes
2005 * with no CPUs, since presumably they'll have very little allocation pressure
2006 * on them otherwise.
2007 * It returns -1 if no node is found.
2009 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2011 int n, val;
2012 int min_val = INT_MAX;
2013 int best_node = -1;
2015 /* Use the local node if we haven't already */
2016 if (!node_isset(node, *used_node_mask)) {
2017 node_set(node, *used_node_mask);
2018 return node;
2021 for_each_node_state(n, N_HIGH_MEMORY) {
2022 cpumask_t tmp;
2024 /* Don't want a node to appear more than once */
2025 if (node_isset(n, *used_node_mask))
2026 continue;
2028 /* Use the distance array to find the distance */
2029 val = node_distance(node, n);
2031 /* Penalize nodes under us ("prefer the next node") */
2032 val += (n < node);
2034 /* Give preference to headless and unused nodes */
2035 tmp = node_to_cpumask(n);
2036 if (!cpus_empty(tmp))
2037 val += PENALTY_FOR_NODE_WITH_CPUS;
2039 /* Slight preference for less loaded node */
2040 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2041 val += node_load[n];
2043 if (val < min_val) {
2044 min_val = val;
2045 best_node = n;
2049 if (best_node >= 0)
2050 node_set(best_node, *used_node_mask);
2052 return best_node;
2057 * Build zonelists ordered by node and zones within node.
2058 * This results in maximum locality--normal zone overflows into local
2059 * DMA zone, if any--but risks exhausting DMA zone.
2061 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2063 enum zone_type i;
2064 int j;
2065 struct zonelist *zonelist;
2067 for (i = 0; i < MAX_NR_ZONES; i++) {
2068 zonelist = pgdat->node_zonelists + i;
2069 for (j = 0; zonelist->zones[j] != NULL; j++)
2071 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2072 zonelist->zones[j] = NULL;
2077 * Build gfp_thisnode zonelists
2079 static void build_thisnode_zonelists(pg_data_t *pgdat)
2081 enum zone_type i;
2082 int j;
2083 struct zonelist *zonelist;
2085 for (i = 0; i < MAX_NR_ZONES; i++) {
2086 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2087 j = build_zonelists_node(pgdat, zonelist, 0, i);
2088 zonelist->zones[j] = NULL;
2093 * Build zonelists ordered by zone and nodes within zones.
2094 * This results in conserving DMA zone[s] until all Normal memory is
2095 * exhausted, but results in overflowing to remote node while memory
2096 * may still exist in local DMA zone.
2098 static int node_order[MAX_NUMNODES];
2100 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2102 enum zone_type i;
2103 int pos, j, node;
2104 int zone_type; /* needs to be signed */
2105 struct zone *z;
2106 struct zonelist *zonelist;
2108 for (i = 0; i < MAX_NR_ZONES; i++) {
2109 zonelist = pgdat->node_zonelists + i;
2110 pos = 0;
2111 for (zone_type = i; zone_type >= 0; zone_type--) {
2112 for (j = 0; j < nr_nodes; j++) {
2113 node = node_order[j];
2114 z = &NODE_DATA(node)->node_zones[zone_type];
2115 if (populated_zone(z)) {
2116 zonelist->zones[pos++] = z;
2117 check_highest_zone(zone_type);
2121 zonelist->zones[pos] = NULL;
2125 static int default_zonelist_order(void)
2127 int nid, zone_type;
2128 unsigned long low_kmem_size,total_size;
2129 struct zone *z;
2130 int average_size;
2132 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2133 * If they are really small and used heavily, the system can fall
2134 * into OOM very easily.
2135 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2137 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2138 low_kmem_size = 0;
2139 total_size = 0;
2140 for_each_online_node(nid) {
2141 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2142 z = &NODE_DATA(nid)->node_zones[zone_type];
2143 if (populated_zone(z)) {
2144 if (zone_type < ZONE_NORMAL)
2145 low_kmem_size += z->present_pages;
2146 total_size += z->present_pages;
2150 if (!low_kmem_size || /* there are no DMA area. */
2151 low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2152 return ZONELIST_ORDER_NODE;
2154 * look into each node's config.
2155 * If there is a node whose DMA/DMA32 memory is very big area on
2156 * local memory, NODE_ORDER may be suitable.
2158 average_size = total_size /
2159 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2160 for_each_online_node(nid) {
2161 low_kmem_size = 0;
2162 total_size = 0;
2163 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2164 z = &NODE_DATA(nid)->node_zones[zone_type];
2165 if (populated_zone(z)) {
2166 if (zone_type < ZONE_NORMAL)
2167 low_kmem_size += z->present_pages;
2168 total_size += z->present_pages;
2171 if (low_kmem_size &&
2172 total_size > average_size && /* ignore small node */
2173 low_kmem_size > total_size * 70/100)
2174 return ZONELIST_ORDER_NODE;
2176 return ZONELIST_ORDER_ZONE;
2179 static void set_zonelist_order(void)
2181 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2182 current_zonelist_order = default_zonelist_order();
2183 else
2184 current_zonelist_order = user_zonelist_order;
2187 static void build_zonelists(pg_data_t *pgdat)
2189 int j, node, load;
2190 enum zone_type i;
2191 nodemask_t used_mask;
2192 int local_node, prev_node;
2193 struct zonelist *zonelist;
2194 int order = current_zonelist_order;
2196 /* initialize zonelists */
2197 for (i = 0; i < MAX_ZONELISTS; i++) {
2198 zonelist = pgdat->node_zonelists + i;
2199 zonelist->zones[0] = NULL;
2202 /* NUMA-aware ordering of nodes */
2203 local_node = pgdat->node_id;
2204 load = num_online_nodes();
2205 prev_node = local_node;
2206 nodes_clear(used_mask);
2208 memset(node_load, 0, sizeof(node_load));
2209 memset(node_order, 0, sizeof(node_order));
2210 j = 0;
2212 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2213 int distance = node_distance(local_node, node);
2216 * If another node is sufficiently far away then it is better
2217 * to reclaim pages in a zone before going off node.
2219 if (distance > RECLAIM_DISTANCE)
2220 zone_reclaim_mode = 1;
2223 * We don't want to pressure a particular node.
2224 * So adding penalty to the first node in same
2225 * distance group to make it round-robin.
2227 if (distance != node_distance(local_node, prev_node))
2228 node_load[node] = load;
2230 prev_node = node;
2231 load--;
2232 if (order == ZONELIST_ORDER_NODE)
2233 build_zonelists_in_node_order(pgdat, node);
2234 else
2235 node_order[j++] = node; /* remember order */
2238 if (order == ZONELIST_ORDER_ZONE) {
2239 /* calculate node order -- i.e., DMA last! */
2240 build_zonelists_in_zone_order(pgdat, j);
2243 build_thisnode_zonelists(pgdat);
2246 /* Construct the zonelist performance cache - see further mmzone.h */
2247 static void build_zonelist_cache(pg_data_t *pgdat)
2249 int i;
2251 for (i = 0; i < MAX_NR_ZONES; i++) {
2252 struct zonelist *zonelist;
2253 struct zonelist_cache *zlc;
2254 struct zone **z;
2256 zonelist = pgdat->node_zonelists + i;
2257 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2258 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2259 for (z = zonelist->zones; *z; z++)
2260 zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2265 #else /* CONFIG_NUMA */
2267 static void set_zonelist_order(void)
2269 current_zonelist_order = ZONELIST_ORDER_ZONE;
2272 static void build_zonelists(pg_data_t *pgdat)
2274 int node, local_node;
2275 enum zone_type i,j;
2277 local_node = pgdat->node_id;
2278 for (i = 0; i < MAX_NR_ZONES; i++) {
2279 struct zonelist *zonelist;
2281 zonelist = pgdat->node_zonelists + i;
2283 j = build_zonelists_node(pgdat, zonelist, 0, i);
2285 * Now we build the zonelist so that it contains the zones
2286 * of all the other nodes.
2287 * We don't want to pressure a particular node, so when
2288 * building the zones for node N, we make sure that the
2289 * zones coming right after the local ones are those from
2290 * node N+1 (modulo N)
2292 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2293 if (!node_online(node))
2294 continue;
2295 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2297 for (node = 0; node < local_node; node++) {
2298 if (!node_online(node))
2299 continue;
2300 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2303 zonelist->zones[j] = NULL;
2307 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2308 static void build_zonelist_cache(pg_data_t *pgdat)
2310 int i;
2312 for (i = 0; i < MAX_NR_ZONES; i++)
2313 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2316 #endif /* CONFIG_NUMA */
2318 /* return values int ....just for stop_machine_run() */
2319 static int __build_all_zonelists(void *dummy)
2321 int nid;
2323 for_each_online_node(nid) {
2324 pg_data_t *pgdat = NODE_DATA(nid);
2326 build_zonelists(pgdat);
2327 build_zonelist_cache(pgdat);
2329 return 0;
2332 void build_all_zonelists(void)
2334 set_zonelist_order();
2336 if (system_state == SYSTEM_BOOTING) {
2337 __build_all_zonelists(NULL);
2338 cpuset_init_current_mems_allowed();
2339 } else {
2340 /* we have to stop all cpus to guarantee there is no user
2341 of zonelist */
2342 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2343 /* cpuset refresh routine should be here */
2345 vm_total_pages = nr_free_pagecache_pages();
2347 * Disable grouping by mobility if the number of pages in the
2348 * system is too low to allow the mechanism to work. It would be
2349 * more accurate, but expensive to check per-zone. This check is
2350 * made on memory-hotadd so a system can start with mobility
2351 * disabled and enable it later
2353 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2354 page_group_by_mobility_disabled = 1;
2355 else
2356 page_group_by_mobility_disabled = 0;
2358 printk("Built %i zonelists in %s order, mobility grouping %s. "
2359 "Total pages: %ld\n",
2360 num_online_nodes(),
2361 zonelist_order_name[current_zonelist_order],
2362 page_group_by_mobility_disabled ? "off" : "on",
2363 vm_total_pages);
2364 #ifdef CONFIG_NUMA
2365 printk("Policy zone: %s\n", zone_names[policy_zone]);
2366 #endif
2370 * Helper functions to size the waitqueue hash table.
2371 * Essentially these want to choose hash table sizes sufficiently
2372 * large so that collisions trying to wait on pages are rare.
2373 * But in fact, the number of active page waitqueues on typical
2374 * systems is ridiculously low, less than 200. So this is even
2375 * conservative, even though it seems large.
2377 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2378 * waitqueues, i.e. the size of the waitq table given the number of pages.
2380 #define PAGES_PER_WAITQUEUE 256
2382 #ifndef CONFIG_MEMORY_HOTPLUG
2383 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2385 unsigned long size = 1;
2387 pages /= PAGES_PER_WAITQUEUE;
2389 while (size < pages)
2390 size <<= 1;
2393 * Once we have dozens or even hundreds of threads sleeping
2394 * on IO we've got bigger problems than wait queue collision.
2395 * Limit the size of the wait table to a reasonable size.
2397 size = min(size, 4096UL);
2399 return max(size, 4UL);
2401 #else
2403 * A zone's size might be changed by hot-add, so it is not possible to determine
2404 * a suitable size for its wait_table. So we use the maximum size now.
2406 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2408 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2409 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2410 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2412 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2413 * or more by the traditional way. (See above). It equals:
2415 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2416 * ia64(16K page size) : = ( 8G + 4M)byte.
2417 * powerpc (64K page size) : = (32G +16M)byte.
2419 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2421 return 4096UL;
2423 #endif
2426 * This is an integer logarithm so that shifts can be used later
2427 * to extract the more random high bits from the multiplicative
2428 * hash function before the remainder is taken.
2430 static inline unsigned long wait_table_bits(unsigned long size)
2432 return ffz(~size);
2435 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2438 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2439 * of blocks reserved is based on zone->pages_min. The memory within the
2440 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2441 * higher will lead to a bigger reserve which will get freed as contiguous
2442 * blocks as reclaim kicks in
2444 static void setup_zone_migrate_reserve(struct zone *zone)
2446 unsigned long start_pfn, pfn, end_pfn;
2447 struct page *page;
2448 unsigned long reserve, block_migratetype;
2450 /* Get the start pfn, end pfn and the number of blocks to reserve */
2451 start_pfn = zone->zone_start_pfn;
2452 end_pfn = start_pfn + zone->spanned_pages;
2453 reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2454 pageblock_order;
2456 for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2457 if (!pfn_valid(pfn))
2458 continue;
2459 page = pfn_to_page(pfn);
2461 /* Blocks with reserved pages will never free, skip them. */
2462 if (PageReserved(page))
2463 continue;
2465 block_migratetype = get_pageblock_migratetype(page);
2467 /* If this block is reserved, account for it */
2468 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2469 reserve--;
2470 continue;
2473 /* Suitable for reserving if this block is movable */
2474 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2475 set_pageblock_migratetype(page, MIGRATE_RESERVE);
2476 move_freepages_block(zone, page, MIGRATE_RESERVE);
2477 reserve--;
2478 continue;
2482 * If the reserve is met and this is a previous reserved block,
2483 * take it back
2485 if (block_migratetype == MIGRATE_RESERVE) {
2486 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2487 move_freepages_block(zone, page, MIGRATE_MOVABLE);
2493 * Initially all pages are reserved - free ones are freed
2494 * up by free_all_bootmem() once the early boot process is
2495 * done. Non-atomic initialization, single-pass.
2497 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2498 unsigned long start_pfn, enum memmap_context context)
2500 struct page *page;
2501 unsigned long end_pfn = start_pfn + size;
2502 unsigned long pfn;
2504 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2506 * There can be holes in boot-time mem_map[]s
2507 * handed to this function. They do not
2508 * exist on hotplugged memory.
2510 if (context == MEMMAP_EARLY) {
2511 if (!early_pfn_valid(pfn))
2512 continue;
2513 if (!early_pfn_in_nid(pfn, nid))
2514 continue;
2516 page = pfn_to_page(pfn);
2517 set_page_links(page, zone, nid, pfn);
2518 init_page_count(page);
2519 reset_page_mapcount(page);
2520 SetPageReserved(page);
2523 * Mark the block movable so that blocks are reserved for
2524 * movable at startup. This will force kernel allocations
2525 * to reserve their blocks rather than leaking throughout
2526 * the address space during boot when many long-lived
2527 * kernel allocations are made. Later some blocks near
2528 * the start are marked MIGRATE_RESERVE by
2529 * setup_zone_migrate_reserve()
2531 if ((pfn & (pageblock_nr_pages-1)))
2532 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2534 INIT_LIST_HEAD(&page->lru);
2535 #ifdef WANT_PAGE_VIRTUAL
2536 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2537 if (!is_highmem_idx(zone))
2538 set_page_address(page, __va(pfn << PAGE_SHIFT));
2539 #endif
2543 static void __meminit zone_init_free_lists(struct pglist_data *pgdat,
2544 struct zone *zone, unsigned long size)
2546 int order, t;
2547 for_each_migratetype_order(order, t) {
2548 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2549 zone->free_area[order].nr_free = 0;
2553 #ifndef __HAVE_ARCH_MEMMAP_INIT
2554 #define memmap_init(size, nid, zone, start_pfn) \
2555 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2556 #endif
2558 static int __devinit zone_batchsize(struct zone *zone)
2560 int batch;
2563 * The per-cpu-pages pools are set to around 1000th of the
2564 * size of the zone. But no more than 1/2 of a meg.
2566 * OK, so we don't know how big the cache is. So guess.
2568 batch = zone->present_pages / 1024;
2569 if (batch * PAGE_SIZE > 512 * 1024)
2570 batch = (512 * 1024) / PAGE_SIZE;
2571 batch /= 4; /* We effectively *= 4 below */
2572 if (batch < 1)
2573 batch = 1;
2576 * Clamp the batch to a 2^n - 1 value. Having a power
2577 * of 2 value was found to be more likely to have
2578 * suboptimal cache aliasing properties in some cases.
2580 * For example if 2 tasks are alternately allocating
2581 * batches of pages, one task can end up with a lot
2582 * of pages of one half of the possible page colors
2583 * and the other with pages of the other colors.
2585 batch = (1 << (fls(batch + batch/2)-1)) - 1;
2587 return batch;
2590 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2592 struct per_cpu_pages *pcp;
2594 memset(p, 0, sizeof(*p));
2596 pcp = &p->pcp[0]; /* hot */
2597 pcp->count = 0;
2598 pcp->high = 6 * batch;
2599 pcp->batch = max(1UL, 1 * batch);
2600 INIT_LIST_HEAD(&pcp->list);
2602 pcp = &p->pcp[1]; /* cold*/
2603 pcp->count = 0;
2604 pcp->high = 2 * batch;
2605 pcp->batch = max(1UL, batch/2);
2606 INIT_LIST_HEAD(&pcp->list);
2610 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2611 * to the value high for the pageset p.
2614 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2615 unsigned long high)
2617 struct per_cpu_pages *pcp;
2619 pcp = &p->pcp[0]; /* hot list */
2620 pcp->high = high;
2621 pcp->batch = max(1UL, high/4);
2622 if ((high/4) > (PAGE_SHIFT * 8))
2623 pcp->batch = PAGE_SHIFT * 8;
2627 #ifdef CONFIG_NUMA
2629 * Boot pageset table. One per cpu which is going to be used for all
2630 * zones and all nodes. The parameters will be set in such a way
2631 * that an item put on a list will immediately be handed over to
2632 * the buddy list. This is safe since pageset manipulation is done
2633 * with interrupts disabled.
2635 * Some NUMA counter updates may also be caught by the boot pagesets.
2637 * The boot_pagesets must be kept even after bootup is complete for
2638 * unused processors and/or zones. They do play a role for bootstrapping
2639 * hotplugged processors.
2641 * zoneinfo_show() and maybe other functions do
2642 * not check if the processor is online before following the pageset pointer.
2643 * Other parts of the kernel may not check if the zone is available.
2645 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2648 * Dynamically allocate memory for the
2649 * per cpu pageset array in struct zone.
2651 static int __cpuinit process_zones(int cpu)
2653 struct zone *zone, *dzone;
2654 int node = cpu_to_node(cpu);
2656 node_set_state(node, N_CPU); /* this node has a cpu */
2658 for_each_zone(zone) {
2660 if (!populated_zone(zone))
2661 continue;
2663 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2664 GFP_KERNEL, node);
2665 if (!zone_pcp(zone, cpu))
2666 goto bad;
2668 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2670 if (percpu_pagelist_fraction)
2671 setup_pagelist_highmark(zone_pcp(zone, cpu),
2672 (zone->present_pages / percpu_pagelist_fraction));
2675 return 0;
2676 bad:
2677 for_each_zone(dzone) {
2678 if (!populated_zone(dzone))
2679 continue;
2680 if (dzone == zone)
2681 break;
2682 kfree(zone_pcp(dzone, cpu));
2683 zone_pcp(dzone, cpu) = NULL;
2685 return -ENOMEM;
2688 static inline void free_zone_pagesets(int cpu)
2690 struct zone *zone;
2692 for_each_zone(zone) {
2693 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2695 /* Free per_cpu_pageset if it is slab allocated */
2696 if (pset != &boot_pageset[cpu])
2697 kfree(pset);
2698 zone_pcp(zone, cpu) = NULL;
2702 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2703 unsigned long action,
2704 void *hcpu)
2706 int cpu = (long)hcpu;
2707 int ret = NOTIFY_OK;
2709 switch (action) {
2710 case CPU_UP_PREPARE:
2711 case CPU_UP_PREPARE_FROZEN:
2712 if (process_zones(cpu))
2713 ret = NOTIFY_BAD;
2714 break;
2715 case CPU_UP_CANCELED:
2716 case CPU_UP_CANCELED_FROZEN:
2717 case CPU_DEAD:
2718 case CPU_DEAD_FROZEN:
2719 free_zone_pagesets(cpu);
2720 break;
2721 default:
2722 break;
2724 return ret;
2727 static struct notifier_block __cpuinitdata pageset_notifier =
2728 { &pageset_cpuup_callback, NULL, 0 };
2730 void __init setup_per_cpu_pageset(void)
2732 int err;
2734 /* Initialize per_cpu_pageset for cpu 0.
2735 * A cpuup callback will do this for every cpu
2736 * as it comes online
2738 err = process_zones(smp_processor_id());
2739 BUG_ON(err);
2740 register_cpu_notifier(&pageset_notifier);
2743 #endif
2745 static noinline __init_refok
2746 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2748 int i;
2749 struct pglist_data *pgdat = zone->zone_pgdat;
2750 size_t alloc_size;
2753 * The per-page waitqueue mechanism uses hashed waitqueues
2754 * per zone.
2756 zone->wait_table_hash_nr_entries =
2757 wait_table_hash_nr_entries(zone_size_pages);
2758 zone->wait_table_bits =
2759 wait_table_bits(zone->wait_table_hash_nr_entries);
2760 alloc_size = zone->wait_table_hash_nr_entries
2761 * sizeof(wait_queue_head_t);
2763 if (system_state == SYSTEM_BOOTING) {
2764 zone->wait_table = (wait_queue_head_t *)
2765 alloc_bootmem_node(pgdat, alloc_size);
2766 } else {
2768 * This case means that a zone whose size was 0 gets new memory
2769 * via memory hot-add.
2770 * But it may be the case that a new node was hot-added. In
2771 * this case vmalloc() will not be able to use this new node's
2772 * memory - this wait_table must be initialized to use this new
2773 * node itself as well.
2774 * To use this new node's memory, further consideration will be
2775 * necessary.
2777 zone->wait_table = vmalloc(alloc_size);
2779 if (!zone->wait_table)
2780 return -ENOMEM;
2782 for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2783 init_waitqueue_head(zone->wait_table + i);
2785 return 0;
2788 static __meminit void zone_pcp_init(struct zone *zone)
2790 int cpu;
2791 unsigned long batch = zone_batchsize(zone);
2793 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2794 #ifdef CONFIG_NUMA
2795 /* Early boot. Slab allocator not functional yet */
2796 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2797 setup_pageset(&boot_pageset[cpu],0);
2798 #else
2799 setup_pageset(zone_pcp(zone,cpu), batch);
2800 #endif
2802 if (zone->present_pages)
2803 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2804 zone->name, zone->present_pages, batch);
2807 __meminit int init_currently_empty_zone(struct zone *zone,
2808 unsigned long zone_start_pfn,
2809 unsigned long size,
2810 enum memmap_context context)
2812 struct pglist_data *pgdat = zone->zone_pgdat;
2813 int ret;
2814 ret = zone_wait_table_init(zone, size);
2815 if (ret)
2816 return ret;
2817 pgdat->nr_zones = zone_idx(zone) + 1;
2819 zone->zone_start_pfn = zone_start_pfn;
2821 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2823 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2825 return 0;
2828 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2830 * Basic iterator support. Return the first range of PFNs for a node
2831 * Note: nid == MAX_NUMNODES returns first region regardless of node
2833 static int __meminit first_active_region_index_in_nid(int nid)
2835 int i;
2837 for (i = 0; i < nr_nodemap_entries; i++)
2838 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2839 return i;
2841 return -1;
2845 * Basic iterator support. Return the next active range of PFNs for a node
2846 * Note: nid == MAX_NUMNODES returns next region regardless of node
2848 static int __meminit next_active_region_index_in_nid(int index, int nid)
2850 for (index = index + 1; index < nr_nodemap_entries; index++)
2851 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2852 return index;
2854 return -1;
2857 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2859 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2860 * Architectures may implement their own version but if add_active_range()
2861 * was used and there are no special requirements, this is a convenient
2862 * alternative
2864 int __meminit early_pfn_to_nid(unsigned long pfn)
2866 int i;
2868 for (i = 0; i < nr_nodemap_entries; i++) {
2869 unsigned long start_pfn = early_node_map[i].start_pfn;
2870 unsigned long end_pfn = early_node_map[i].end_pfn;
2872 if (start_pfn <= pfn && pfn < end_pfn)
2873 return early_node_map[i].nid;
2876 return 0;
2878 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2880 /* Basic iterator support to walk early_node_map[] */
2881 #define for_each_active_range_index_in_nid(i, nid) \
2882 for (i = first_active_region_index_in_nid(nid); i != -1; \
2883 i = next_active_region_index_in_nid(i, nid))
2886 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2887 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2888 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2890 * If an architecture guarantees that all ranges registered with
2891 * add_active_ranges() contain no holes and may be freed, this
2892 * this function may be used instead of calling free_bootmem() manually.
2894 void __init free_bootmem_with_active_regions(int nid,
2895 unsigned long max_low_pfn)
2897 int i;
2899 for_each_active_range_index_in_nid(i, nid) {
2900 unsigned long size_pages = 0;
2901 unsigned long end_pfn = early_node_map[i].end_pfn;
2903 if (early_node_map[i].start_pfn >= max_low_pfn)
2904 continue;
2906 if (end_pfn > max_low_pfn)
2907 end_pfn = max_low_pfn;
2909 size_pages = end_pfn - early_node_map[i].start_pfn;
2910 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2911 PFN_PHYS(early_node_map[i].start_pfn),
2912 size_pages << PAGE_SHIFT);
2917 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2918 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2920 * If an architecture guarantees that all ranges registered with
2921 * add_active_ranges() contain no holes and may be freed, this
2922 * function may be used instead of calling memory_present() manually.
2924 void __init sparse_memory_present_with_active_regions(int nid)
2926 int i;
2928 for_each_active_range_index_in_nid(i, nid)
2929 memory_present(early_node_map[i].nid,
2930 early_node_map[i].start_pfn,
2931 early_node_map[i].end_pfn);
2935 * push_node_boundaries - Push node boundaries to at least the requested boundary
2936 * @nid: The nid of the node to push the boundary for
2937 * @start_pfn: The start pfn of the node
2938 * @end_pfn: The end pfn of the node
2940 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2941 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2942 * be hotplugged even though no physical memory exists. This function allows
2943 * an arch to push out the node boundaries so mem_map is allocated that can
2944 * be used later.
2946 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2947 void __init push_node_boundaries(unsigned int nid,
2948 unsigned long start_pfn, unsigned long end_pfn)
2950 printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2951 nid, start_pfn, end_pfn);
2953 /* Initialise the boundary for this node if necessary */
2954 if (node_boundary_end_pfn[nid] == 0)
2955 node_boundary_start_pfn[nid] = -1UL;
2957 /* Update the boundaries */
2958 if (node_boundary_start_pfn[nid] > start_pfn)
2959 node_boundary_start_pfn[nid] = start_pfn;
2960 if (node_boundary_end_pfn[nid] < end_pfn)
2961 node_boundary_end_pfn[nid] = end_pfn;
2964 /* If necessary, push the node boundary out for reserve hotadd */
2965 static void __meminit account_node_boundary(unsigned int nid,
2966 unsigned long *start_pfn, unsigned long *end_pfn)
2968 printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2969 nid, *start_pfn, *end_pfn);
2971 /* Return if boundary information has not been provided */
2972 if (node_boundary_end_pfn[nid] == 0)
2973 return;
2975 /* Check the boundaries and update if necessary */
2976 if (node_boundary_start_pfn[nid] < *start_pfn)
2977 *start_pfn = node_boundary_start_pfn[nid];
2978 if (node_boundary_end_pfn[nid] > *end_pfn)
2979 *end_pfn = node_boundary_end_pfn[nid];
2981 #else
2982 void __init push_node_boundaries(unsigned int nid,
2983 unsigned long start_pfn, unsigned long end_pfn) {}
2985 static void __meminit account_node_boundary(unsigned int nid,
2986 unsigned long *start_pfn, unsigned long *end_pfn) {}
2987 #endif
2991 * get_pfn_range_for_nid - Return the start and end page frames for a node
2992 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2993 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2994 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2996 * It returns the start and end page frame of a node based on information
2997 * provided by an arch calling add_active_range(). If called for a node
2998 * with no available memory, a warning is printed and the start and end
2999 * PFNs will be 0.
3001 void __meminit get_pfn_range_for_nid(unsigned int nid,
3002 unsigned long *start_pfn, unsigned long *end_pfn)
3004 int i;
3005 *start_pfn = -1UL;
3006 *end_pfn = 0;
3008 for_each_active_range_index_in_nid(i, nid) {
3009 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3010 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3013 if (*start_pfn == -1UL)
3014 *start_pfn = 0;
3016 /* Push the node boundaries out if requested */
3017 account_node_boundary(nid, start_pfn, end_pfn);
3021 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3022 * assumption is made that zones within a node are ordered in monotonic
3023 * increasing memory addresses so that the "highest" populated zone is used
3025 void __init find_usable_zone_for_movable(void)
3027 int zone_index;
3028 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3029 if (zone_index == ZONE_MOVABLE)
3030 continue;
3032 if (arch_zone_highest_possible_pfn[zone_index] >
3033 arch_zone_lowest_possible_pfn[zone_index])
3034 break;
3037 VM_BUG_ON(zone_index == -1);
3038 movable_zone = zone_index;
3042 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3043 * because it is sized independant of architecture. Unlike the other zones,
3044 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3045 * in each node depending on the size of each node and how evenly kernelcore
3046 * is distributed. This helper function adjusts the zone ranges
3047 * provided by the architecture for a given node by using the end of the
3048 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3049 * zones within a node are in order of monotonic increases memory addresses
3051 void __meminit adjust_zone_range_for_zone_movable(int nid,
3052 unsigned long zone_type,
3053 unsigned long node_start_pfn,
3054 unsigned long node_end_pfn,
3055 unsigned long *zone_start_pfn,
3056 unsigned long *zone_end_pfn)
3058 /* Only adjust if ZONE_MOVABLE is on this node */
3059 if (zone_movable_pfn[nid]) {
3060 /* Size ZONE_MOVABLE */
3061 if (zone_type == ZONE_MOVABLE) {
3062 *zone_start_pfn = zone_movable_pfn[nid];
3063 *zone_end_pfn = min(node_end_pfn,
3064 arch_zone_highest_possible_pfn[movable_zone]);
3066 /* Adjust for ZONE_MOVABLE starting within this range */
3067 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3068 *zone_end_pfn > zone_movable_pfn[nid]) {
3069 *zone_end_pfn = zone_movable_pfn[nid];
3071 /* Check if this whole range is within ZONE_MOVABLE */
3072 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3073 *zone_start_pfn = *zone_end_pfn;
3078 * Return the number of pages a zone spans in a node, including holes
3079 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3081 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3082 unsigned long zone_type,
3083 unsigned long *ignored)
3085 unsigned long node_start_pfn, node_end_pfn;
3086 unsigned long zone_start_pfn, zone_end_pfn;
3088 /* Get the start and end of the node and zone */
3089 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3090 zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3091 zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3092 adjust_zone_range_for_zone_movable(nid, zone_type,
3093 node_start_pfn, node_end_pfn,
3094 &zone_start_pfn, &zone_end_pfn);
3096 /* Check that this node has pages within the zone's required range */
3097 if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3098 return 0;
3100 /* Move the zone boundaries inside the node if necessary */
3101 zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3102 zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3104 /* Return the spanned pages */
3105 return zone_end_pfn - zone_start_pfn;
3109 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3110 * then all holes in the requested range will be accounted for.
3112 unsigned long __meminit __absent_pages_in_range(int nid,
3113 unsigned long range_start_pfn,
3114 unsigned long range_end_pfn)
3116 int i = 0;
3117 unsigned long prev_end_pfn = 0, hole_pages = 0;
3118 unsigned long start_pfn;
3120 /* Find the end_pfn of the first active range of pfns in the node */
3121 i = first_active_region_index_in_nid(nid);
3122 if (i == -1)
3123 return 0;
3125 prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3127 /* Account for ranges before physical memory on this node */
3128 if (early_node_map[i].start_pfn > range_start_pfn)
3129 hole_pages = prev_end_pfn - range_start_pfn;
3131 /* Find all holes for the zone within the node */
3132 for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3134 /* No need to continue if prev_end_pfn is outside the zone */
3135 if (prev_end_pfn >= range_end_pfn)
3136 break;
3138 /* Make sure the end of the zone is not within the hole */
3139 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3140 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3142 /* Update the hole size cound and move on */
3143 if (start_pfn > range_start_pfn) {
3144 BUG_ON(prev_end_pfn > start_pfn);
3145 hole_pages += start_pfn - prev_end_pfn;
3147 prev_end_pfn = early_node_map[i].end_pfn;
3150 /* Account for ranges past physical memory on this node */
3151 if (range_end_pfn > prev_end_pfn)
3152 hole_pages += range_end_pfn -
3153 max(range_start_pfn, prev_end_pfn);
3155 return hole_pages;
3159 * absent_pages_in_range - Return number of page frames in holes within a range
3160 * @start_pfn: The start PFN to start searching for holes
3161 * @end_pfn: The end PFN to stop searching for holes
3163 * It returns the number of pages frames in memory holes within a range.
3165 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3166 unsigned long end_pfn)
3168 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3171 /* Return the number of page frames in holes in a zone on a node */
3172 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3173 unsigned long zone_type,
3174 unsigned long *ignored)
3176 unsigned long node_start_pfn, node_end_pfn;
3177 unsigned long zone_start_pfn, zone_end_pfn;
3179 get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3180 zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3181 node_start_pfn);
3182 zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3183 node_end_pfn);
3185 adjust_zone_range_for_zone_movable(nid, zone_type,
3186 node_start_pfn, node_end_pfn,
3187 &zone_start_pfn, &zone_end_pfn);
3188 return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3191 #else
3192 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3193 unsigned long zone_type,
3194 unsigned long *zones_size)
3196 return zones_size[zone_type];
3199 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3200 unsigned long zone_type,
3201 unsigned long *zholes_size)
3203 if (!zholes_size)
3204 return 0;
3206 return zholes_size[zone_type];
3209 #endif
3211 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3212 unsigned long *zones_size, unsigned long *zholes_size)
3214 unsigned long realtotalpages, totalpages = 0;
3215 enum zone_type i;
3217 for (i = 0; i < MAX_NR_ZONES; i++)
3218 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3219 zones_size);
3220 pgdat->node_spanned_pages = totalpages;
3222 realtotalpages = totalpages;
3223 for (i = 0; i < MAX_NR_ZONES; i++)
3224 realtotalpages -=
3225 zone_absent_pages_in_node(pgdat->node_id, i,
3226 zholes_size);
3227 pgdat->node_present_pages = realtotalpages;
3228 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3229 realtotalpages);
3232 #ifndef CONFIG_SPARSEMEM
3234 * Calculate the size of the zone->blockflags rounded to an unsigned long
3235 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3236 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3237 * round what is now in bits to nearest long in bits, then return it in
3238 * bytes.
3240 static unsigned long __init usemap_size(unsigned long zonesize)
3242 unsigned long usemapsize;
3244 usemapsize = roundup(zonesize, pageblock_nr_pages);
3245 usemapsize = usemapsize >> pageblock_order;
3246 usemapsize *= NR_PAGEBLOCK_BITS;
3247 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3249 return usemapsize / 8;
3252 static void __init setup_usemap(struct pglist_data *pgdat,
3253 struct zone *zone, unsigned long zonesize)
3255 unsigned long usemapsize = usemap_size(zonesize);
3256 zone->pageblock_flags = NULL;
3257 if (usemapsize) {
3258 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3259 memset(zone->pageblock_flags, 0, usemapsize);
3262 #else
3263 static void inline setup_usemap(struct pglist_data *pgdat,
3264 struct zone *zone, unsigned long zonesize) {}
3265 #endif /* CONFIG_SPARSEMEM */
3267 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3269 /* Return a sensible default order for the pageblock size. */
3270 static inline int pageblock_default_order(void)
3272 if (HPAGE_SHIFT > PAGE_SHIFT)
3273 return HUGETLB_PAGE_ORDER;
3275 return MAX_ORDER-1;
3278 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3279 static inline void __init set_pageblock_order(unsigned int order)
3281 /* Check that pageblock_nr_pages has not already been setup */
3282 if (pageblock_order)
3283 return;
3286 * Assume the largest contiguous order of interest is a huge page.
3287 * This value may be variable depending on boot parameters on IA64
3289 pageblock_order = order;
3291 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3294 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3295 * and pageblock_default_order() are unused as pageblock_order is set
3296 * at compile-time. See include/linux/pageblock-flags.h for the values of
3297 * pageblock_order based on the kernel config
3299 static inline int pageblock_default_order(unsigned int order)
3301 return MAX_ORDER-1;
3303 #define set_pageblock_order(x) do {} while (0)
3305 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3308 * Set up the zone data structures:
3309 * - mark all pages reserved
3310 * - mark all memory queues empty
3311 * - clear the memory bitmaps
3313 static void __meminit free_area_init_core(struct pglist_data *pgdat,
3314 unsigned long *zones_size, unsigned long *zholes_size)
3316 enum zone_type j;
3317 int nid = pgdat->node_id;
3318 unsigned long zone_start_pfn = pgdat->node_start_pfn;
3319 int ret;
3321 pgdat_resize_init(pgdat);
3322 pgdat->nr_zones = 0;
3323 init_waitqueue_head(&pgdat->kswapd_wait);
3324 pgdat->kswapd_max_order = 0;
3326 for (j = 0; j < MAX_NR_ZONES; j++) {
3327 struct zone *zone = pgdat->node_zones + j;
3328 unsigned long size, realsize, memmap_pages;
3330 size = zone_spanned_pages_in_node(nid, j, zones_size);
3331 realsize = size - zone_absent_pages_in_node(nid, j,
3332 zholes_size);
3335 * Adjust realsize so that it accounts for how much memory
3336 * is used by this zone for memmap. This affects the watermark
3337 * and per-cpu initialisations
3339 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3340 if (realsize >= memmap_pages) {
3341 realsize -= memmap_pages;
3342 printk(KERN_DEBUG
3343 " %s zone: %lu pages used for memmap\n",
3344 zone_names[j], memmap_pages);
3345 } else
3346 printk(KERN_WARNING
3347 " %s zone: %lu pages exceeds realsize %lu\n",
3348 zone_names[j], memmap_pages, realsize);
3350 /* Account for reserved pages */
3351 if (j == 0 && realsize > dma_reserve) {
3352 realsize -= dma_reserve;
3353 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
3354 zone_names[0], dma_reserve);
3357 if (!is_highmem_idx(j))
3358 nr_kernel_pages += realsize;
3359 nr_all_pages += realsize;
3361 zone->spanned_pages = size;
3362 zone->present_pages = realsize;
3363 #ifdef CONFIG_NUMA
3364 zone->node = nid;
3365 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3366 / 100;
3367 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3368 #endif
3369 zone->name = zone_names[j];
3370 spin_lock_init(&zone->lock);
3371 spin_lock_init(&zone->lru_lock);
3372 zone_seqlock_init(zone);
3373 zone->zone_pgdat = pgdat;
3375 zone->prev_priority = DEF_PRIORITY;
3377 zone_pcp_init(zone);
3378 INIT_LIST_HEAD(&zone->active_list);
3379 INIT_LIST_HEAD(&zone->inactive_list);
3380 zone->nr_scan_active = 0;
3381 zone->nr_scan_inactive = 0;
3382 zap_zone_vm_stats(zone);
3383 zone->flags = 0;
3384 if (!size)
3385 continue;
3387 set_pageblock_order(pageblock_default_order());
3388 setup_usemap(pgdat, zone, size);
3389 ret = init_currently_empty_zone(zone, zone_start_pfn,
3390 size, MEMMAP_EARLY);
3391 BUG_ON(ret);
3392 zone_start_pfn += size;
3396 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3398 /* Skip empty nodes */
3399 if (!pgdat->node_spanned_pages)
3400 return;
3402 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3403 /* ia64 gets its own node_mem_map, before this, without bootmem */
3404 if (!pgdat->node_mem_map) {
3405 unsigned long size, start, end;
3406 struct page *map;
3409 * The zone's endpoints aren't required to be MAX_ORDER
3410 * aligned but the node_mem_map endpoints must be in order
3411 * for the buddy allocator to function correctly.
3413 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3414 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3415 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3416 size = (end - start) * sizeof(struct page);
3417 map = alloc_remap(pgdat->node_id, size);
3418 if (!map)
3419 map = alloc_bootmem_node(pgdat, size);
3420 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3422 #ifndef CONFIG_NEED_MULTIPLE_NODES
3424 * With no DISCONTIG, the global mem_map is just set as node 0's
3426 if (pgdat == NODE_DATA(0)) {
3427 mem_map = NODE_DATA(0)->node_mem_map;
3428 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3429 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3430 mem_map -= pgdat->node_start_pfn;
3431 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3433 #endif
3434 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3437 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
3438 unsigned long *zones_size, unsigned long node_start_pfn,
3439 unsigned long *zholes_size)
3441 pgdat->node_id = nid;
3442 pgdat->node_start_pfn = node_start_pfn;
3443 calculate_node_totalpages(pgdat, zones_size, zholes_size);
3445 alloc_node_mem_map(pgdat);
3447 free_area_init_core(pgdat, zones_size, zholes_size);
3450 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3452 #if MAX_NUMNODES > 1
3454 * Figure out the number of possible node ids.
3456 static void __init setup_nr_node_ids(void)
3458 unsigned int node;
3459 unsigned int highest = 0;
3461 for_each_node_mask(node, node_possible_map)
3462 highest = node;
3463 nr_node_ids = highest + 1;
3465 #else
3466 static inline void setup_nr_node_ids(void)
3469 #endif
3472 * add_active_range - Register a range of PFNs backed by physical memory
3473 * @nid: The node ID the range resides on
3474 * @start_pfn: The start PFN of the available physical memory
3475 * @end_pfn: The end PFN of the available physical memory
3477 * These ranges are stored in an early_node_map[] and later used by
3478 * free_area_init_nodes() to calculate zone sizes and holes. If the
3479 * range spans a memory hole, it is up to the architecture to ensure
3480 * the memory is not freed by the bootmem allocator. If possible
3481 * the range being registered will be merged with existing ranges.
3483 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3484 unsigned long end_pfn)
3486 int i;
3488 printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3489 "%d entries of %d used\n",
3490 nid, start_pfn, end_pfn,
3491 nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3493 /* Merge with existing active regions if possible */
3494 for (i = 0; i < nr_nodemap_entries; i++) {
3495 if (early_node_map[i].nid != nid)
3496 continue;
3498 /* Skip if an existing region covers this new one */
3499 if (start_pfn >= early_node_map[i].start_pfn &&
3500 end_pfn <= early_node_map[i].end_pfn)
3501 return;
3503 /* Merge forward if suitable */
3504 if (start_pfn <= early_node_map[i].end_pfn &&
3505 end_pfn > early_node_map[i].end_pfn) {
3506 early_node_map[i].end_pfn = end_pfn;
3507 return;
3510 /* Merge backward if suitable */
3511 if (start_pfn < early_node_map[i].end_pfn &&
3512 end_pfn >= early_node_map[i].start_pfn) {
3513 early_node_map[i].start_pfn = start_pfn;
3514 return;
3518 /* Check that early_node_map is large enough */
3519 if (i >= MAX_ACTIVE_REGIONS) {
3520 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3521 MAX_ACTIVE_REGIONS);
3522 return;
3525 early_node_map[i].nid = nid;
3526 early_node_map[i].start_pfn = start_pfn;
3527 early_node_map[i].end_pfn = end_pfn;
3528 nr_nodemap_entries = i + 1;
3532 * shrink_active_range - Shrink an existing registered range of PFNs
3533 * @nid: The node id the range is on that should be shrunk
3534 * @old_end_pfn: The old end PFN of the range
3535 * @new_end_pfn: The new PFN of the range
3537 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3538 * The map is kept at the end physical page range that has already been
3539 * registered with add_active_range(). This function allows an arch to shrink
3540 * an existing registered range.
3542 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3543 unsigned long new_end_pfn)
3545 int i;
3547 /* Find the old active region end and shrink */
3548 for_each_active_range_index_in_nid(i, nid)
3549 if (early_node_map[i].end_pfn == old_end_pfn) {
3550 early_node_map[i].end_pfn = new_end_pfn;
3551 break;
3556 * remove_all_active_ranges - Remove all currently registered regions
3558 * During discovery, it may be found that a table like SRAT is invalid
3559 * and an alternative discovery method must be used. This function removes
3560 * all currently registered regions.
3562 void __init remove_all_active_ranges(void)
3564 memset(early_node_map, 0, sizeof(early_node_map));
3565 nr_nodemap_entries = 0;
3566 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3567 memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3568 memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3569 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3572 /* Compare two active node_active_regions */
3573 static int __init cmp_node_active_region(const void *a, const void *b)
3575 struct node_active_region *arange = (struct node_active_region *)a;
3576 struct node_active_region *brange = (struct node_active_region *)b;
3578 /* Done this way to avoid overflows */
3579 if (arange->start_pfn > brange->start_pfn)
3580 return 1;
3581 if (arange->start_pfn < brange->start_pfn)
3582 return -1;
3584 return 0;
3587 /* sort the node_map by start_pfn */
3588 static void __init sort_node_map(void)
3590 sort(early_node_map, (size_t)nr_nodemap_entries,
3591 sizeof(struct node_active_region),
3592 cmp_node_active_region, NULL);
3595 /* Find the lowest pfn for a node */
3596 unsigned long __init find_min_pfn_for_node(unsigned long nid)
3598 int i;
3599 unsigned long min_pfn = ULONG_MAX;
3601 /* Assuming a sorted map, the first range found has the starting pfn */
3602 for_each_active_range_index_in_nid(i, nid)
3603 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3605 if (min_pfn == ULONG_MAX) {
3606 printk(KERN_WARNING
3607 "Could not find start_pfn for node %lu\n", nid);
3608 return 0;
3611 return min_pfn;
3615 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3617 * It returns the minimum PFN based on information provided via
3618 * add_active_range().
3620 unsigned long __init find_min_pfn_with_active_regions(void)
3622 return find_min_pfn_for_node(MAX_NUMNODES);
3626 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3628 * It returns the maximum PFN based on information provided via
3629 * add_active_range().
3631 unsigned long __init find_max_pfn_with_active_regions(void)
3633 int i;
3634 unsigned long max_pfn = 0;
3636 for (i = 0; i < nr_nodemap_entries; i++)
3637 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3639 return max_pfn;
3643 * early_calculate_totalpages()
3644 * Sum pages in active regions for movable zone.
3645 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3647 static unsigned long __init early_calculate_totalpages(void)
3649 int i;
3650 unsigned long totalpages = 0;
3652 for (i = 0; i < nr_nodemap_entries; i++) {
3653 unsigned long pages = early_node_map[i].end_pfn -
3654 early_node_map[i].start_pfn;
3655 totalpages += pages;
3656 if (pages)
3657 node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3659 return totalpages;
3663 * Find the PFN the Movable zone begins in each node. Kernel memory
3664 * is spread evenly between nodes as long as the nodes have enough
3665 * memory. When they don't, some nodes will have more kernelcore than
3666 * others
3668 void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3670 int i, nid;
3671 unsigned long usable_startpfn;
3672 unsigned long kernelcore_node, kernelcore_remaining;
3673 unsigned long totalpages = early_calculate_totalpages();
3674 int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3677 * If movablecore was specified, calculate what size of
3678 * kernelcore that corresponds so that memory usable for
3679 * any allocation type is evenly spread. If both kernelcore
3680 * and movablecore are specified, then the value of kernelcore
3681 * will be used for required_kernelcore if it's greater than
3682 * what movablecore would have allowed.
3684 if (required_movablecore) {
3685 unsigned long corepages;
3688 * Round-up so that ZONE_MOVABLE is at least as large as what
3689 * was requested by the user
3691 required_movablecore =
3692 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3693 corepages = totalpages - required_movablecore;
3695 required_kernelcore = max(required_kernelcore, corepages);
3698 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3699 if (!required_kernelcore)
3700 return;
3702 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3703 find_usable_zone_for_movable();
3704 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3706 restart:
3707 /* Spread kernelcore memory as evenly as possible throughout nodes */
3708 kernelcore_node = required_kernelcore / usable_nodes;
3709 for_each_node_state(nid, N_HIGH_MEMORY) {
3711 * Recalculate kernelcore_node if the division per node
3712 * now exceeds what is necessary to satisfy the requested
3713 * amount of memory for the kernel
3715 if (required_kernelcore < kernelcore_node)
3716 kernelcore_node = required_kernelcore / usable_nodes;
3719 * As the map is walked, we track how much memory is usable
3720 * by the kernel using kernelcore_remaining. When it is
3721 * 0, the rest of the node is usable by ZONE_MOVABLE
3723 kernelcore_remaining = kernelcore_node;
3725 /* Go through each range of PFNs within this node */
3726 for_each_active_range_index_in_nid(i, nid) {
3727 unsigned long start_pfn, end_pfn;
3728 unsigned long size_pages;
3730 start_pfn = max(early_node_map[i].start_pfn,
3731 zone_movable_pfn[nid]);
3732 end_pfn = early_node_map[i].end_pfn;
3733 if (start_pfn >= end_pfn)
3734 continue;
3736 /* Account for what is only usable for kernelcore */
3737 if (start_pfn < usable_startpfn) {
3738 unsigned long kernel_pages;
3739 kernel_pages = min(end_pfn, usable_startpfn)
3740 - start_pfn;
3742 kernelcore_remaining -= min(kernel_pages,
3743 kernelcore_remaining);
3744 required_kernelcore -= min(kernel_pages,
3745 required_kernelcore);
3747 /* Continue if range is now fully accounted */
3748 if (end_pfn <= usable_startpfn) {
3751 * Push zone_movable_pfn to the end so
3752 * that if we have to rebalance
3753 * kernelcore across nodes, we will
3754 * not double account here
3756 zone_movable_pfn[nid] = end_pfn;
3757 continue;
3759 start_pfn = usable_startpfn;
3763 * The usable PFN range for ZONE_MOVABLE is from
3764 * start_pfn->end_pfn. Calculate size_pages as the
3765 * number of pages used as kernelcore
3767 size_pages = end_pfn - start_pfn;
3768 if (size_pages > kernelcore_remaining)
3769 size_pages = kernelcore_remaining;
3770 zone_movable_pfn[nid] = start_pfn + size_pages;
3773 * Some kernelcore has been met, update counts and
3774 * break if the kernelcore for this node has been
3775 * satisified
3777 required_kernelcore -= min(required_kernelcore,
3778 size_pages);
3779 kernelcore_remaining -= size_pages;
3780 if (!kernelcore_remaining)
3781 break;
3786 * If there is still required_kernelcore, we do another pass with one
3787 * less node in the count. This will push zone_movable_pfn[nid] further
3788 * along on the nodes that still have memory until kernelcore is
3789 * satisified
3791 usable_nodes--;
3792 if (usable_nodes && required_kernelcore > usable_nodes)
3793 goto restart;
3795 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3796 for (nid = 0; nid < MAX_NUMNODES; nid++)
3797 zone_movable_pfn[nid] =
3798 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3801 /* Any regular memory on that node ? */
3802 static void check_for_regular_memory(pg_data_t *pgdat)
3804 #ifdef CONFIG_HIGHMEM
3805 enum zone_type zone_type;
3807 for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3808 struct zone *zone = &pgdat->node_zones[zone_type];
3809 if (zone->present_pages)
3810 node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3812 #endif
3816 * free_area_init_nodes - Initialise all pg_data_t and zone data
3817 * @max_zone_pfn: an array of max PFNs for each zone
3819 * This will call free_area_init_node() for each active node in the system.
3820 * Using the page ranges provided by add_active_range(), the size of each
3821 * zone in each node and their holes is calculated. If the maximum PFN
3822 * between two adjacent zones match, it is assumed that the zone is empty.
3823 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3824 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3825 * starts where the previous one ended. For example, ZONE_DMA32 starts
3826 * at arch_max_dma_pfn.
3828 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3830 unsigned long nid;
3831 enum zone_type i;
3833 /* Sort early_node_map as initialisation assumes it is sorted */
3834 sort_node_map();
3836 /* Record where the zone boundaries are */
3837 memset(arch_zone_lowest_possible_pfn, 0,
3838 sizeof(arch_zone_lowest_possible_pfn));
3839 memset(arch_zone_highest_possible_pfn, 0,
3840 sizeof(arch_zone_highest_possible_pfn));
3841 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3842 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3843 for (i = 1; i < MAX_NR_ZONES; i++) {
3844 if (i == ZONE_MOVABLE)
3845 continue;
3846 arch_zone_lowest_possible_pfn[i] =
3847 arch_zone_highest_possible_pfn[i-1];
3848 arch_zone_highest_possible_pfn[i] =
3849 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3851 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3852 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3854 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3855 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3856 find_zone_movable_pfns_for_nodes(zone_movable_pfn);
3858 /* Print out the zone ranges */
3859 printk("Zone PFN ranges:\n");
3860 for (i = 0; i < MAX_NR_ZONES; i++) {
3861 if (i == ZONE_MOVABLE)
3862 continue;
3863 printk(" %-8s %8lu -> %8lu\n",
3864 zone_names[i],
3865 arch_zone_lowest_possible_pfn[i],
3866 arch_zone_highest_possible_pfn[i]);
3869 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3870 printk("Movable zone start PFN for each node\n");
3871 for (i = 0; i < MAX_NUMNODES; i++) {
3872 if (zone_movable_pfn[i])
3873 printk(" Node %d: %lu\n", i, zone_movable_pfn[i]);
3876 /* Print out the early_node_map[] */
3877 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3878 for (i = 0; i < nr_nodemap_entries; i++)
3879 printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3880 early_node_map[i].start_pfn,
3881 early_node_map[i].end_pfn);
3883 /* Initialise every node */
3884 setup_nr_node_ids();
3885 for_each_online_node(nid) {
3886 pg_data_t *pgdat = NODE_DATA(nid);
3887 free_area_init_node(nid, pgdat, NULL,
3888 find_min_pfn_for_node(nid), NULL);
3890 /* Any memory on that node */
3891 if (pgdat->node_present_pages)
3892 node_set_state(nid, N_HIGH_MEMORY);
3893 check_for_regular_memory(pgdat);
3897 static int __init cmdline_parse_core(char *p, unsigned long *core)
3899 unsigned long long coremem;
3900 if (!p)
3901 return -EINVAL;
3903 coremem = memparse(p, &p);
3904 *core = coremem >> PAGE_SHIFT;
3906 /* Paranoid check that UL is enough for the coremem value */
3907 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
3909 return 0;
3913 * kernelcore=size sets the amount of memory for use for allocations that
3914 * cannot be reclaimed or migrated.
3916 static int __init cmdline_parse_kernelcore(char *p)
3918 return cmdline_parse_core(p, &required_kernelcore);
3922 * movablecore=size sets the amount of memory for use for allocations that
3923 * can be reclaimed or migrated.
3925 static int __init cmdline_parse_movablecore(char *p)
3927 return cmdline_parse_core(p, &required_movablecore);
3930 early_param("kernelcore", cmdline_parse_kernelcore);
3931 early_param("movablecore", cmdline_parse_movablecore);
3933 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3936 * set_dma_reserve - set the specified number of pages reserved in the first zone
3937 * @new_dma_reserve: The number of pages to mark reserved
3939 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3940 * In the DMA zone, a significant percentage may be consumed by kernel image
3941 * and other unfreeable allocations which can skew the watermarks badly. This
3942 * function may optionally be used to account for unfreeable pages in the
3943 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3944 * smaller per-cpu batchsize.
3946 void __init set_dma_reserve(unsigned long new_dma_reserve)
3948 dma_reserve = new_dma_reserve;
3951 #ifndef CONFIG_NEED_MULTIPLE_NODES
3952 static bootmem_data_t contig_bootmem_data;
3953 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
3955 EXPORT_SYMBOL(contig_page_data);
3956 #endif
3958 void __init free_area_init(unsigned long *zones_size)
3960 free_area_init_node(0, NODE_DATA(0), zones_size,
3961 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
3964 static int page_alloc_cpu_notify(struct notifier_block *self,
3965 unsigned long action, void *hcpu)
3967 int cpu = (unsigned long)hcpu;
3969 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
3970 local_irq_disable();
3971 __drain_pages(cpu);
3972 vm_events_fold_cpu(cpu);
3973 local_irq_enable();
3974 refresh_cpu_vm_stats(cpu);
3976 return NOTIFY_OK;
3979 void __init page_alloc_init(void)
3981 hotcpu_notifier(page_alloc_cpu_notify, 0);
3985 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3986 * or min_free_kbytes changes.
3988 static void calculate_totalreserve_pages(void)
3990 struct pglist_data *pgdat;
3991 unsigned long reserve_pages = 0;
3992 enum zone_type i, j;
3994 for_each_online_pgdat(pgdat) {
3995 for (i = 0; i < MAX_NR_ZONES; i++) {
3996 struct zone *zone = pgdat->node_zones + i;
3997 unsigned long max = 0;
3999 /* Find valid and maximum lowmem_reserve in the zone */
4000 for (j = i; j < MAX_NR_ZONES; j++) {
4001 if (zone->lowmem_reserve[j] > max)
4002 max = zone->lowmem_reserve[j];
4005 /* we treat pages_high as reserved pages. */
4006 max += zone->pages_high;
4008 if (max > zone->present_pages)
4009 max = zone->present_pages;
4010 reserve_pages += max;
4013 totalreserve_pages = reserve_pages;
4017 * setup_per_zone_lowmem_reserve - called whenever
4018 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4019 * has a correct pages reserved value, so an adequate number of
4020 * pages are left in the zone after a successful __alloc_pages().
4022 static void setup_per_zone_lowmem_reserve(void)
4024 struct pglist_data *pgdat;
4025 enum zone_type j, idx;
4027 for_each_online_pgdat(pgdat) {
4028 for (j = 0; j < MAX_NR_ZONES; j++) {
4029 struct zone *zone = pgdat->node_zones + j;
4030 unsigned long present_pages = zone->present_pages;
4032 zone->lowmem_reserve[j] = 0;
4034 idx = j;
4035 while (idx) {
4036 struct zone *lower_zone;
4038 idx--;
4040 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4041 sysctl_lowmem_reserve_ratio[idx] = 1;
4043 lower_zone = pgdat->node_zones + idx;
4044 lower_zone->lowmem_reserve[j] = present_pages /
4045 sysctl_lowmem_reserve_ratio[idx];
4046 present_pages += lower_zone->present_pages;
4051 /* update totalreserve_pages */
4052 calculate_totalreserve_pages();
4056 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4058 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4059 * with respect to min_free_kbytes.
4061 void setup_per_zone_pages_min(void)
4063 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4064 unsigned long lowmem_pages = 0;
4065 struct zone *zone;
4066 unsigned long flags;
4068 /* Calculate total number of !ZONE_HIGHMEM pages */
4069 for_each_zone(zone) {
4070 if (!is_highmem(zone))
4071 lowmem_pages += zone->present_pages;
4074 for_each_zone(zone) {
4075 u64 tmp;
4077 spin_lock_irqsave(&zone->lru_lock, flags);
4078 tmp = (u64)pages_min * zone->present_pages;
4079 do_div(tmp, lowmem_pages);
4080 if (is_highmem(zone)) {
4082 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4083 * need highmem pages, so cap pages_min to a small
4084 * value here.
4086 * The (pages_high-pages_low) and (pages_low-pages_min)
4087 * deltas controls asynch page reclaim, and so should
4088 * not be capped for highmem.
4090 int min_pages;
4092 min_pages = zone->present_pages / 1024;
4093 if (min_pages < SWAP_CLUSTER_MAX)
4094 min_pages = SWAP_CLUSTER_MAX;
4095 if (min_pages > 128)
4096 min_pages = 128;
4097 zone->pages_min = min_pages;
4098 } else {
4100 * If it's a lowmem zone, reserve a number of pages
4101 * proportionate to the zone's size.
4103 zone->pages_min = tmp;
4106 zone->pages_low = zone->pages_min + (tmp >> 2);
4107 zone->pages_high = zone->pages_min + (tmp >> 1);
4108 setup_zone_migrate_reserve(zone);
4109 spin_unlock_irqrestore(&zone->lru_lock, flags);
4112 /* update totalreserve_pages */
4113 calculate_totalreserve_pages();
4117 * Initialise min_free_kbytes.
4119 * For small machines we want it small (128k min). For large machines
4120 * we want it large (64MB max). But it is not linear, because network
4121 * bandwidth does not increase linearly with machine size. We use
4123 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4124 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4126 * which yields
4128 * 16MB: 512k
4129 * 32MB: 724k
4130 * 64MB: 1024k
4131 * 128MB: 1448k
4132 * 256MB: 2048k
4133 * 512MB: 2896k
4134 * 1024MB: 4096k
4135 * 2048MB: 5792k
4136 * 4096MB: 8192k
4137 * 8192MB: 11584k
4138 * 16384MB: 16384k
4140 static int __init init_per_zone_pages_min(void)
4142 unsigned long lowmem_kbytes;
4144 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4146 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4147 if (min_free_kbytes < 128)
4148 min_free_kbytes = 128;
4149 if (min_free_kbytes > 65536)
4150 min_free_kbytes = 65536;
4151 setup_per_zone_pages_min();
4152 setup_per_zone_lowmem_reserve();
4153 return 0;
4155 module_init(init_per_zone_pages_min)
4158 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4159 * that we can call two helper functions whenever min_free_kbytes
4160 * changes.
4162 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
4163 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4165 proc_dointvec(table, write, file, buffer, length, ppos);
4166 if (write)
4167 setup_per_zone_pages_min();
4168 return 0;
4171 #ifdef CONFIG_NUMA
4172 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4173 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4175 struct zone *zone;
4176 int rc;
4178 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4179 if (rc)
4180 return rc;
4182 for_each_zone(zone)
4183 zone->min_unmapped_pages = (zone->present_pages *
4184 sysctl_min_unmapped_ratio) / 100;
4185 return 0;
4188 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4189 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4191 struct zone *zone;
4192 int rc;
4194 rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4195 if (rc)
4196 return rc;
4198 for_each_zone(zone)
4199 zone->min_slab_pages = (zone->present_pages *
4200 sysctl_min_slab_ratio) / 100;
4201 return 0;
4203 #endif
4206 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4207 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4208 * whenever sysctl_lowmem_reserve_ratio changes.
4210 * The reserve ratio obviously has absolutely no relation with the
4211 * pages_min watermarks. The lowmem reserve ratio can only make sense
4212 * if in function of the boot time zone sizes.
4214 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4215 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4217 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4218 setup_per_zone_lowmem_reserve();
4219 return 0;
4223 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4224 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4225 * can have before it gets flushed back to buddy allocator.
4228 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4229 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4231 struct zone *zone;
4232 unsigned int cpu;
4233 int ret;
4235 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4236 if (!write || (ret == -EINVAL))
4237 return ret;
4238 for_each_zone(zone) {
4239 for_each_online_cpu(cpu) {
4240 unsigned long high;
4241 high = zone->present_pages / percpu_pagelist_fraction;
4242 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4245 return 0;
4248 int hashdist = HASHDIST_DEFAULT;
4250 #ifdef CONFIG_NUMA
4251 static int __init set_hashdist(char *str)
4253 if (!str)
4254 return 0;
4255 hashdist = simple_strtoul(str, &str, 0);
4256 return 1;
4258 __setup("hashdist=", set_hashdist);
4259 #endif
4262 * allocate a large system hash table from bootmem
4263 * - it is assumed that the hash table must contain an exact power-of-2
4264 * quantity of entries
4265 * - limit is the number of hash buckets, not the total allocation size
4267 void *__init alloc_large_system_hash(const char *tablename,
4268 unsigned long bucketsize,
4269 unsigned long numentries,
4270 int scale,
4271 int flags,
4272 unsigned int *_hash_shift,
4273 unsigned int *_hash_mask,
4274 unsigned long limit)
4276 unsigned long long max = limit;
4277 unsigned long log2qty, size;
4278 void *table = NULL;
4280 /* allow the kernel cmdline to have a say */
4281 if (!numentries) {
4282 /* round applicable memory size up to nearest megabyte */
4283 numentries = nr_kernel_pages;
4284 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4285 numentries >>= 20 - PAGE_SHIFT;
4286 numentries <<= 20 - PAGE_SHIFT;
4288 /* limit to 1 bucket per 2^scale bytes of low memory */
4289 if (scale > PAGE_SHIFT)
4290 numentries >>= (scale - PAGE_SHIFT);
4291 else
4292 numentries <<= (PAGE_SHIFT - scale);
4294 /* Make sure we've got at least a 0-order allocation.. */
4295 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4296 numentries = PAGE_SIZE / bucketsize;
4298 numentries = roundup_pow_of_two(numentries);
4300 /* limit allocation size to 1/16 total memory by default */
4301 if (max == 0) {
4302 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4303 do_div(max, bucketsize);
4306 if (numentries > max)
4307 numentries = max;
4309 log2qty = ilog2(numentries);
4311 do {
4312 size = bucketsize << log2qty;
4313 if (flags & HASH_EARLY)
4314 table = alloc_bootmem(size);
4315 else if (hashdist)
4316 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4317 else {
4318 unsigned long order;
4319 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4321 table = (void*) __get_free_pages(GFP_ATOMIC, order);
4323 * If bucketsize is not a power-of-two, we may free
4324 * some pages at the end of hash table.
4326 if (table) {
4327 unsigned long alloc_end = (unsigned long)table +
4328 (PAGE_SIZE << order);
4329 unsigned long used = (unsigned long)table +
4330 PAGE_ALIGN(size);
4331 split_page(virt_to_page(table), order);
4332 while (used < alloc_end) {
4333 free_page(used);
4334 used += PAGE_SIZE;
4338 } while (!table && size > PAGE_SIZE && --log2qty);
4340 if (!table)
4341 panic("Failed to allocate %s hash table\n", tablename);
4343 printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4344 tablename,
4345 (1U << log2qty),
4346 ilog2(size) - PAGE_SHIFT,
4347 size);
4349 if (_hash_shift)
4350 *_hash_shift = log2qty;
4351 if (_hash_mask)
4352 *_hash_mask = (1 << log2qty) - 1;
4354 return table;
4357 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4358 struct page *pfn_to_page(unsigned long pfn)
4360 return __pfn_to_page(pfn);
4362 unsigned long page_to_pfn(struct page *page)
4364 return __page_to_pfn(page);
4366 EXPORT_SYMBOL(pfn_to_page);
4367 EXPORT_SYMBOL(page_to_pfn);
4368 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4370 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4371 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4372 unsigned long pfn)
4374 #ifdef CONFIG_SPARSEMEM
4375 return __pfn_to_section(pfn)->pageblock_flags;
4376 #else
4377 return zone->pageblock_flags;
4378 #endif /* CONFIG_SPARSEMEM */
4381 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4383 #ifdef CONFIG_SPARSEMEM
4384 pfn &= (PAGES_PER_SECTION-1);
4385 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4386 #else
4387 pfn = pfn - zone->zone_start_pfn;
4388 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4389 #endif /* CONFIG_SPARSEMEM */
4393 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4394 * @page: The page within the block of interest
4395 * @start_bitidx: The first bit of interest to retrieve
4396 * @end_bitidx: The last bit of interest
4397 * returns pageblock_bits flags
4399 unsigned long get_pageblock_flags_group(struct page *page,
4400 int start_bitidx, int end_bitidx)
4402 struct zone *zone;
4403 unsigned long *bitmap;
4404 unsigned long pfn, bitidx;
4405 unsigned long flags = 0;
4406 unsigned long value = 1;
4408 zone = page_zone(page);
4409 pfn = page_to_pfn(page);
4410 bitmap = get_pageblock_bitmap(zone, pfn);
4411 bitidx = pfn_to_bitidx(zone, pfn);
4413 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4414 if (test_bit(bitidx + start_bitidx, bitmap))
4415 flags |= value;
4417 return flags;
4421 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4422 * @page: The page within the block of interest
4423 * @start_bitidx: The first bit of interest
4424 * @end_bitidx: The last bit of interest
4425 * @flags: The flags to set
4427 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4428 int start_bitidx, int end_bitidx)
4430 struct zone *zone;
4431 unsigned long *bitmap;
4432 unsigned long pfn, bitidx;
4433 unsigned long value = 1;
4435 zone = page_zone(page);
4436 pfn = page_to_pfn(page);
4437 bitmap = get_pageblock_bitmap(zone, pfn);
4438 bitidx = pfn_to_bitidx(zone, pfn);
4440 for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4441 if (flags & value)
4442 __set_bit(bitidx + start_bitidx, bitmap);
4443 else
4444 __clear_bit(bitidx + start_bitidx, bitmap);
4448 * This is designed as sub function...plz see page_isolation.c also.
4449 * set/clear page block's type to be ISOLATE.
4450 * page allocater never alloc memory from ISOLATE block.
4453 int set_migratetype_isolate(struct page *page)
4455 struct zone *zone;
4456 unsigned long flags;
4457 int ret = -EBUSY;
4459 zone = page_zone(page);
4460 spin_lock_irqsave(&zone->lock, flags);
4462 * In future, more migrate types will be able to be isolation target.
4464 if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4465 goto out;
4466 set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4467 move_freepages_block(zone, page, MIGRATE_ISOLATE);
4468 ret = 0;
4469 out:
4470 spin_unlock_irqrestore(&zone->lock, flags);
4471 if (!ret)
4472 drain_all_local_pages();
4473 return ret;
4476 void unset_migratetype_isolate(struct page *page)
4478 struct zone *zone;
4479 unsigned long flags;
4480 zone = page_zone(page);
4481 spin_lock_irqsave(&zone->lock, flags);
4482 if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4483 goto out;
4484 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4485 move_freepages_block(zone, page, MIGRATE_MOVABLE);
4486 out:
4487 spin_unlock_irqrestore(&zone->lock, flags);
4490 #ifdef CONFIG_MEMORY_HOTREMOVE
4492 * All pages in the range must be isolated before calling this.
4494 void
4495 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4497 struct page *page;
4498 struct zone *zone;
4499 int order, i;
4500 unsigned long pfn;
4501 unsigned long flags;
4502 /* find the first valid pfn */
4503 for (pfn = start_pfn; pfn < end_pfn; pfn++)
4504 if (pfn_valid(pfn))
4505 break;
4506 if (pfn == end_pfn)
4507 return;
4508 zone = page_zone(pfn_to_page(pfn));
4509 spin_lock_irqsave(&zone->lock, flags);
4510 pfn = start_pfn;
4511 while (pfn < end_pfn) {
4512 if (!pfn_valid(pfn)) {
4513 pfn++;
4514 continue;
4516 page = pfn_to_page(pfn);
4517 BUG_ON(page_count(page));
4518 BUG_ON(!PageBuddy(page));
4519 order = page_order(page);
4520 #ifdef CONFIG_DEBUG_VM
4521 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4522 pfn, 1 << order, end_pfn);
4523 #endif
4524 list_del(&page->lru);
4525 rmv_page_order(page);
4526 zone->free_area[order].nr_free--;
4527 __mod_zone_page_state(zone, NR_FREE_PAGES,
4528 - (1UL << order));
4529 for (i = 0; i < (1 << order); i++)
4530 SetPageReserved((page+i));
4531 pfn += (1 << order);
4533 spin_unlock_irqrestore(&zone->lock, flags);
4535 #endif