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>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/oom.h>
34 #include <linux/notifier.h>
35 #include <linux/topology.h>
36 #include <linux/sysctl.h>
37 #include <linux/cpu.h>
38 #include <linux/cpuset.h>
39 #include <linux/memory_hotplug.h>
40 #include <linux/nodemask.h>
41 #include <linux/vmalloc.h>
42 #include <linux/mempolicy.h>
43 #include <linux/stop_machine.h>
44 #include <linux/sort.h>
45 #include <linux/pfn.h>
46 #include <linux/backing-dev.h>
47 #include <linux/fault-inject.h>
48 #include <linux/page-isolation.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/debugobjects.h>
51 #include <linux/kmemleak.h>
52 #include <linux/memory.h>
53 #include <linux/compaction.h>
54 #include <trace/events/kmem.h>
55 #include <linux/ftrace_event.h>
57 #include <asm/tlbflush.h>
58 #include <asm/div64.h>
61 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
62 DEFINE_PER_CPU(int, numa_node
);
63 EXPORT_PER_CPU_SYMBOL(numa_node
);
66 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
68 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
69 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
70 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
71 * defined in <linux/topology.h>.
73 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
74 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
78 * Array of node states.
80 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
81 [N_POSSIBLE
] = NODE_MASK_ALL
,
82 [N_ONLINE
] = { { [0] = 1UL } },
84 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
86 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
88 [N_CPU
] = { { [0] = 1UL } },
91 EXPORT_SYMBOL(node_states
);
93 unsigned long totalram_pages __read_mostly
;
94 unsigned long totalreserve_pages __read_mostly
;
95 int percpu_pagelist_fraction
;
96 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
98 #ifdef CONFIG_PM_SLEEP
100 * The following functions are used by the suspend/hibernate code to temporarily
101 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
102 * while devices are suspended. To avoid races with the suspend/hibernate code,
103 * they should always be called with pm_mutex held (gfp_allowed_mask also should
104 * only be modified with pm_mutex held, unless the suspend/hibernate code is
105 * guaranteed not to run in parallel with that modification).
108 static gfp_t saved_gfp_mask
;
110 void pm_restore_gfp_mask(void)
112 WARN_ON(!mutex_is_locked(&pm_mutex
));
113 if (saved_gfp_mask
) {
114 gfp_allowed_mask
= saved_gfp_mask
;
119 void pm_restrict_gfp_mask(void)
121 WARN_ON(!mutex_is_locked(&pm_mutex
));
122 WARN_ON(saved_gfp_mask
);
123 saved_gfp_mask
= gfp_allowed_mask
;
124 gfp_allowed_mask
&= ~GFP_IOFS
;
126 #endif /* CONFIG_PM_SLEEP */
128 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
129 int pageblock_order __read_mostly
;
132 static void __free_pages_ok(struct page
*page
, unsigned int order
);
135 * results with 256, 32 in the lowmem_reserve sysctl:
136 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
137 * 1G machine -> (16M dma, 784M normal, 224M high)
138 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
139 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
140 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
142 * TBD: should special case ZONE_DMA32 machines here - in those we normally
143 * don't need any ZONE_NORMAL reservation
145 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
146 #ifdef CONFIG_ZONE_DMA
149 #ifdef CONFIG_ZONE_DMA32
152 #ifdef CONFIG_HIGHMEM
158 EXPORT_SYMBOL(totalram_pages
);
160 static char * const zone_names
[MAX_NR_ZONES
] = {
161 #ifdef CONFIG_ZONE_DMA
164 #ifdef CONFIG_ZONE_DMA32
168 #ifdef CONFIG_HIGHMEM
174 int min_free_kbytes
= 1024;
176 static unsigned long __meminitdata nr_kernel_pages
;
177 static unsigned long __meminitdata nr_all_pages
;
178 static unsigned long __meminitdata dma_reserve
;
180 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
182 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
183 * ranges of memory (RAM) that may be registered with add_active_range().
184 * Ranges passed to add_active_range() will be merged if possible
185 * so the number of times add_active_range() can be called is
186 * related to the number of nodes and the number of holes
188 #ifdef CONFIG_MAX_ACTIVE_REGIONS
189 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
190 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
192 #if MAX_NUMNODES >= 32
193 /* If there can be many nodes, allow up to 50 holes per node */
194 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
196 /* By default, allow up to 256 distinct regions */
197 #define MAX_ACTIVE_REGIONS 256
201 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
202 static int __meminitdata nr_nodemap_entries
;
203 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
204 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
205 static unsigned long __initdata required_kernelcore
;
206 static unsigned long __initdata required_movablecore
;
207 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
209 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
211 EXPORT_SYMBOL(movable_zone
);
212 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
215 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
216 int nr_online_nodes __read_mostly
= 1;
217 EXPORT_SYMBOL(nr_node_ids
);
218 EXPORT_SYMBOL(nr_online_nodes
);
221 int page_group_by_mobility_disabled __read_mostly
;
223 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
226 if (unlikely(page_group_by_mobility_disabled
))
227 migratetype
= MIGRATE_UNMOVABLE
;
229 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
230 PB_migrate
, PB_migrate_end
);
233 bool oom_killer_disabled __read_mostly
;
235 #ifdef CONFIG_DEBUG_VM
236 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
240 unsigned long pfn
= page_to_pfn(page
);
243 seq
= zone_span_seqbegin(zone
);
244 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
246 else if (pfn
< zone
->zone_start_pfn
)
248 } while (zone_span_seqretry(zone
, seq
));
253 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
255 if (!pfn_valid_within(page_to_pfn(page
)))
257 if (zone
!= page_zone(page
))
263 * Temporary debugging check for pages not lying within a given zone.
265 static int bad_range(struct zone
*zone
, struct page
*page
)
267 if (page_outside_zone_boundaries(zone
, page
))
269 if (!page_is_consistent(zone
, page
))
275 static inline int bad_range(struct zone
*zone
, struct page
*page
)
281 static void bad_page(struct page
*page
)
283 static unsigned long resume
;
284 static unsigned long nr_shown
;
285 static unsigned long nr_unshown
;
287 /* Don't complain about poisoned pages */
288 if (PageHWPoison(page
)) {
289 __ClearPageBuddy(page
);
294 * Allow a burst of 60 reports, then keep quiet for that minute;
295 * or allow a steady drip of one report per second.
297 if (nr_shown
== 60) {
298 if (time_before(jiffies
, resume
)) {
304 "BUG: Bad page state: %lu messages suppressed\n",
311 resume
= jiffies
+ 60 * HZ
;
313 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
314 current
->comm
, page_to_pfn(page
));
319 /* Leave bad fields for debug, except PageBuddy could make trouble */
320 __ClearPageBuddy(page
);
321 add_taint(TAINT_BAD_PAGE
);
325 * Higher-order pages are called "compound pages". They are structured thusly:
327 * The first PAGE_SIZE page is called the "head page".
329 * The remaining PAGE_SIZE pages are called "tail pages".
331 * All pages have PG_compound set. All pages have their ->private pointing at
332 * the head page (even the head page has this).
334 * The first tail page's ->lru.next holds the address of the compound page's
335 * put_page() function. Its ->lru.prev holds the order of allocation.
336 * This usage means that zero-order pages may not be compound.
339 static void free_compound_page(struct page
*page
)
341 __free_pages_ok(page
, compound_order(page
));
344 void prep_compound_page(struct page
*page
, unsigned long order
)
347 int nr_pages
= 1 << order
;
349 set_compound_page_dtor(page
, free_compound_page
);
350 set_compound_order(page
, order
);
352 for (i
= 1; i
< nr_pages
; i
++) {
353 struct page
*p
= page
+ i
;
356 p
->first_page
= page
;
360 /* update __split_huge_page_refcount if you change this function */
361 static int destroy_compound_page(struct page
*page
, unsigned long order
)
364 int nr_pages
= 1 << order
;
367 if (unlikely(compound_order(page
) != order
) ||
368 unlikely(!PageHead(page
))) {
373 __ClearPageHead(page
);
375 for (i
= 1; i
< nr_pages
; i
++) {
376 struct page
*p
= page
+ i
;
378 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
388 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
393 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
394 * and __GFP_HIGHMEM from hard or soft interrupt context.
396 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
397 for (i
= 0; i
< (1 << order
); i
++)
398 clear_highpage(page
+ i
);
401 static inline void set_page_order(struct page
*page
, int order
)
403 set_page_private(page
, order
);
404 __SetPageBuddy(page
);
407 static inline void rmv_page_order(struct page
*page
)
409 __ClearPageBuddy(page
);
410 set_page_private(page
, 0);
414 * Locate the struct page for both the matching buddy in our
415 * pair (buddy1) and the combined O(n+1) page they form (page).
417 * 1) Any buddy B1 will have an order O twin B2 which satisfies
418 * the following equation:
420 * For example, if the starting buddy (buddy2) is #8 its order
422 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
424 * 2) Any buddy B will have an order O+1 parent P which
425 * satisfies the following equation:
428 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
430 static inline struct page
*
431 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
433 unsigned long buddy_idx
= page_idx
^ (1 << order
);
435 return page
+ (buddy_idx
- page_idx
);
438 static inline unsigned long
439 __find_combined_index(unsigned long page_idx
, unsigned int order
)
441 return (page_idx
& ~(1 << order
));
445 * This function checks whether a page is free && is the buddy
446 * we can do coalesce a page and its buddy if
447 * (a) the buddy is not in a hole &&
448 * (b) the buddy is in the buddy system &&
449 * (c) a page and its buddy have the same order &&
450 * (d) a page and its buddy are in the same zone.
452 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
453 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
455 * For recording page's order, we use page_private(page).
457 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
460 if (!pfn_valid_within(page_to_pfn(buddy
)))
463 if (page_zone_id(page
) != page_zone_id(buddy
))
466 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
467 VM_BUG_ON(page_count(buddy
) != 0);
474 * Freeing function for a buddy system allocator.
476 * The concept of a buddy system is to maintain direct-mapped table
477 * (containing bit values) for memory blocks of various "orders".
478 * The bottom level table contains the map for the smallest allocatable
479 * units of memory (here, pages), and each level above it describes
480 * pairs of units from the levels below, hence, "buddies".
481 * At a high level, all that happens here is marking the table entry
482 * at the bottom level available, and propagating the changes upward
483 * as necessary, plus some accounting needed to play nicely with other
484 * parts of the VM system.
485 * At each level, we keep a list of pages, which are heads of continuous
486 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
487 * order is recorded in page_private(page) field.
488 * So when we are allocating or freeing one, we can derive the state of the
489 * other. That is, if we allocate a small block, and both were
490 * free, the remainder of the region must be split into blocks.
491 * If a block is freed, and its buddy is also free, then this
492 * triggers coalescing into a block of larger size.
497 static inline void __free_one_page(struct page
*page
,
498 struct zone
*zone
, unsigned int order
,
501 unsigned long page_idx
;
502 unsigned long combined_idx
;
505 if (unlikely(PageCompound(page
)))
506 if (unlikely(destroy_compound_page(page
, order
)))
509 VM_BUG_ON(migratetype
== -1);
511 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
513 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
514 VM_BUG_ON(bad_range(zone
, page
));
516 while (order
< MAX_ORDER
-1) {
517 buddy
= __page_find_buddy(page
, page_idx
, order
);
518 if (!page_is_buddy(page
, buddy
, order
))
521 /* Our buddy is free, merge with it and move up one order. */
522 list_del(&buddy
->lru
);
523 zone
->free_area
[order
].nr_free
--;
524 rmv_page_order(buddy
);
525 combined_idx
= __find_combined_index(page_idx
, order
);
526 page
= page
+ (combined_idx
- page_idx
);
527 page_idx
= combined_idx
;
530 set_page_order(page
, order
);
533 * If this is not the largest possible page, check if the buddy
534 * of the next-highest order is free. If it is, it's possible
535 * that pages are being freed that will coalesce soon. In case,
536 * that is happening, add the free page to the tail of the list
537 * so it's less likely to be used soon and more likely to be merged
538 * as a higher order page
540 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
541 struct page
*higher_page
, *higher_buddy
;
542 combined_idx
= __find_combined_index(page_idx
, order
);
543 higher_page
= page
+ combined_idx
- page_idx
;
544 higher_buddy
= __page_find_buddy(higher_page
, combined_idx
, order
+ 1);
545 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
546 list_add_tail(&page
->lru
,
547 &zone
->free_area
[order
].free_list
[migratetype
]);
552 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
554 zone
->free_area
[order
].nr_free
++;
558 * free_page_mlock() -- clean up attempts to free and mlocked() page.
559 * Page should not be on lru, so no need to fix that up.
560 * free_pages_check() will verify...
562 static inline void free_page_mlock(struct page
*page
)
564 __dec_zone_page_state(page
, NR_MLOCK
);
565 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
568 static inline int free_pages_check(struct page
*page
)
570 if (unlikely(page_mapcount(page
) |
571 (page
->mapping
!= NULL
) |
572 (atomic_read(&page
->_count
) != 0) |
573 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
577 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
578 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
583 * Frees a number of pages from the PCP lists
584 * Assumes all pages on list are in same zone, and of same order.
585 * count is the number of pages to free.
587 * If the zone was previously in an "all pages pinned" state then look to
588 * see if this freeing clears that state.
590 * And clear the zone's pages_scanned counter, to hold off the "all pages are
591 * pinned" detection logic.
593 static void free_pcppages_bulk(struct zone
*zone
, int count
,
594 struct per_cpu_pages
*pcp
)
600 spin_lock(&zone
->lock
);
601 zone
->all_unreclaimable
= 0;
602 zone
->pages_scanned
= 0;
606 struct list_head
*list
;
609 * Remove pages from lists in a round-robin fashion. A
610 * batch_free count is maintained that is incremented when an
611 * empty list is encountered. This is so more pages are freed
612 * off fuller lists instead of spinning excessively around empty
617 if (++migratetype
== MIGRATE_PCPTYPES
)
619 list
= &pcp
->lists
[migratetype
];
620 } while (list_empty(list
));
623 page
= list_entry(list
->prev
, struct page
, lru
);
624 /* must delete as __free_one_page list manipulates */
625 list_del(&page
->lru
);
626 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
627 __free_one_page(page
, zone
, 0, page_private(page
));
628 trace_mm_page_pcpu_drain(page
, 0, page_private(page
));
629 } while (--to_free
&& --batch_free
&& !list_empty(list
));
631 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
);
632 spin_unlock(&zone
->lock
);
635 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
638 spin_lock(&zone
->lock
);
639 zone
->all_unreclaimable
= 0;
640 zone
->pages_scanned
= 0;
642 __free_one_page(page
, zone
, order
, migratetype
);
643 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
644 spin_unlock(&zone
->lock
);
647 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
652 trace_mm_page_free_direct(page
, order
);
653 kmemcheck_free_shadow(page
, order
);
656 page
->mapping
= NULL
;
657 for (i
= 0; i
< (1 << order
); i
++)
658 bad
+= free_pages_check(page
+ i
);
662 if (!PageHighMem(page
)) {
663 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
664 debug_check_no_obj_freed(page_address(page
),
667 arch_free_page(page
, order
);
668 kernel_map_pages(page
, 1 << order
, 0);
673 static void __free_pages_ok(struct page
*page
, unsigned int order
)
676 int wasMlocked
= __TestClearPageMlocked(page
);
678 if (!free_pages_prepare(page
, order
))
681 local_irq_save(flags
);
682 if (unlikely(wasMlocked
))
683 free_page_mlock(page
);
684 __count_vm_events(PGFREE
, 1 << order
);
685 free_one_page(page_zone(page
), page
, order
,
686 get_pageblock_migratetype(page
));
687 local_irq_restore(flags
);
691 * permit the bootmem allocator to evade page validation on high-order frees
693 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
696 __ClearPageReserved(page
);
697 set_page_count(page
, 0);
698 set_page_refcounted(page
);
704 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
705 struct page
*p
= &page
[loop
];
707 if (loop
+ 1 < BITS_PER_LONG
)
709 __ClearPageReserved(p
);
710 set_page_count(p
, 0);
713 set_page_refcounted(page
);
714 __free_pages(page
, order
);
720 * The order of subdivision here is critical for the IO subsystem.
721 * Please do not alter this order without good reasons and regression
722 * testing. Specifically, as large blocks of memory are subdivided,
723 * the order in which smaller blocks are delivered depends on the order
724 * they're subdivided in this function. This is the primary factor
725 * influencing the order in which pages are delivered to the IO
726 * subsystem according to empirical testing, and this is also justified
727 * by considering the behavior of a buddy system containing a single
728 * large block of memory acted on by a series of small allocations.
729 * This behavior is a critical factor in sglist merging's success.
733 static inline void expand(struct zone
*zone
, struct page
*page
,
734 int low
, int high
, struct free_area
*area
,
737 unsigned long size
= 1 << high
;
743 VM_BUG_ON(bad_range(zone
, &page
[size
]));
744 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
746 set_page_order(&page
[size
], high
);
751 * This page is about to be returned from the page allocator
753 static inline int check_new_page(struct page
*page
)
755 if (unlikely(page_mapcount(page
) |
756 (page
->mapping
!= NULL
) |
757 (atomic_read(&page
->_count
) != 0) |
758 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
765 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
769 for (i
= 0; i
< (1 << order
); i
++) {
770 struct page
*p
= page
+ i
;
771 if (unlikely(check_new_page(p
)))
775 set_page_private(page
, 0);
776 set_page_refcounted(page
);
778 arch_alloc_page(page
, order
);
779 kernel_map_pages(page
, 1 << order
, 1);
781 if (gfp_flags
& __GFP_ZERO
)
782 prep_zero_page(page
, order
, gfp_flags
);
784 if (order
&& (gfp_flags
& __GFP_COMP
))
785 prep_compound_page(page
, order
);
791 * Go through the free lists for the given migratetype and remove
792 * the smallest available page from the freelists
795 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
798 unsigned int current_order
;
799 struct free_area
* area
;
802 /* Find a page of the appropriate size in the preferred list */
803 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
804 area
= &(zone
->free_area
[current_order
]);
805 if (list_empty(&area
->free_list
[migratetype
]))
808 page
= list_entry(area
->free_list
[migratetype
].next
,
810 list_del(&page
->lru
);
811 rmv_page_order(page
);
813 expand(zone
, page
, order
, current_order
, area
, migratetype
);
822 * This array describes the order lists are fallen back to when
823 * the free lists for the desirable migrate type are depleted
825 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
826 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
827 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
828 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
829 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
833 * Move the free pages in a range to the free lists of the requested type.
834 * Note that start_page and end_pages are not aligned on a pageblock
835 * boundary. If alignment is required, use move_freepages_block()
837 static int move_freepages(struct zone
*zone
,
838 struct page
*start_page
, struct page
*end_page
,
845 #ifndef CONFIG_HOLES_IN_ZONE
847 * page_zone is not safe to call in this context when
848 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
849 * anyway as we check zone boundaries in move_freepages_block().
850 * Remove at a later date when no bug reports exist related to
851 * grouping pages by mobility
853 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
856 for (page
= start_page
; page
<= end_page
;) {
857 /* Make sure we are not inadvertently changing nodes */
858 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
860 if (!pfn_valid_within(page_to_pfn(page
))) {
865 if (!PageBuddy(page
)) {
870 order
= page_order(page
);
871 list_del(&page
->lru
);
873 &zone
->free_area
[order
].free_list
[migratetype
]);
875 pages_moved
+= 1 << order
;
881 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
884 unsigned long start_pfn
, end_pfn
;
885 struct page
*start_page
, *end_page
;
887 start_pfn
= page_to_pfn(page
);
888 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
889 start_page
= pfn_to_page(start_pfn
);
890 end_page
= start_page
+ pageblock_nr_pages
- 1;
891 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
893 /* Do not cross zone boundaries */
894 if (start_pfn
< zone
->zone_start_pfn
)
896 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
899 return move_freepages(zone
, start_page
, end_page
, migratetype
);
902 static void change_pageblock_range(struct page
*pageblock_page
,
903 int start_order
, int migratetype
)
905 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
907 while (nr_pageblocks
--) {
908 set_pageblock_migratetype(pageblock_page
, migratetype
);
909 pageblock_page
+= pageblock_nr_pages
;
913 /* Remove an element from the buddy allocator from the fallback list */
914 static inline struct page
*
915 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
917 struct free_area
* area
;
922 /* Find the largest possible block of pages in the other list */
923 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
925 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
926 migratetype
= fallbacks
[start_migratetype
][i
];
928 /* MIGRATE_RESERVE handled later if necessary */
929 if (migratetype
== MIGRATE_RESERVE
)
932 area
= &(zone
->free_area
[current_order
]);
933 if (list_empty(&area
->free_list
[migratetype
]))
936 page
= list_entry(area
->free_list
[migratetype
].next
,
941 * If breaking a large block of pages, move all free
942 * pages to the preferred allocation list. If falling
943 * back for a reclaimable kernel allocation, be more
944 * agressive about taking ownership of free pages
946 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
947 start_migratetype
== MIGRATE_RECLAIMABLE
||
948 page_group_by_mobility_disabled
) {
950 pages
= move_freepages_block(zone
, page
,
953 /* Claim the whole block if over half of it is free */
954 if (pages
>= (1 << (pageblock_order
-1)) ||
955 page_group_by_mobility_disabled
)
956 set_pageblock_migratetype(page
,
959 migratetype
= start_migratetype
;
962 /* Remove the page from the freelists */
963 list_del(&page
->lru
);
964 rmv_page_order(page
);
966 /* Take ownership for orders >= pageblock_order */
967 if (current_order
>= pageblock_order
)
968 change_pageblock_range(page
, current_order
,
971 expand(zone
, page
, order
, current_order
, area
, migratetype
);
973 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
974 start_migratetype
, migratetype
);
984 * Do the hard work of removing an element from the buddy allocator.
985 * Call me with the zone->lock already held.
987 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
993 page
= __rmqueue_smallest(zone
, order
, migratetype
);
995 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
996 page
= __rmqueue_fallback(zone
, order
, migratetype
);
999 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1000 * is used because __rmqueue_smallest is an inline function
1001 * and we want just one call site
1004 migratetype
= MIGRATE_RESERVE
;
1009 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1014 * Obtain a specified number of elements from the buddy allocator, all under
1015 * a single hold of the lock, for efficiency. Add them to the supplied list.
1016 * Returns the number of new pages which were placed at *list.
1018 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1019 unsigned long count
, struct list_head
*list
,
1020 int migratetype
, int cold
)
1024 spin_lock(&zone
->lock
);
1025 for (i
= 0; i
< count
; ++i
) {
1026 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1027 if (unlikely(page
== NULL
))
1031 * Split buddy pages returned by expand() are received here
1032 * in physical page order. The page is added to the callers and
1033 * list and the list head then moves forward. From the callers
1034 * perspective, the linked list is ordered by page number in
1035 * some conditions. This is useful for IO devices that can
1036 * merge IO requests if the physical pages are ordered
1039 if (likely(cold
== 0))
1040 list_add(&page
->lru
, list
);
1042 list_add_tail(&page
->lru
, list
);
1043 set_page_private(page
, migratetype
);
1046 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1047 spin_unlock(&zone
->lock
);
1053 * Called from the vmstat counter updater to drain pagesets of this
1054 * currently executing processor on remote nodes after they have
1057 * Note that this function must be called with the thread pinned to
1058 * a single processor.
1060 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1062 unsigned long flags
;
1065 local_irq_save(flags
);
1066 if (pcp
->count
>= pcp
->batch
)
1067 to_drain
= pcp
->batch
;
1069 to_drain
= pcp
->count
;
1070 free_pcppages_bulk(zone
, to_drain
, pcp
);
1071 pcp
->count
-= to_drain
;
1072 local_irq_restore(flags
);
1077 * Drain pages of the indicated processor.
1079 * The processor must either be the current processor and the
1080 * thread pinned to the current processor or a processor that
1083 static void drain_pages(unsigned int cpu
)
1085 unsigned long flags
;
1088 for_each_populated_zone(zone
) {
1089 struct per_cpu_pageset
*pset
;
1090 struct per_cpu_pages
*pcp
;
1092 local_irq_save(flags
);
1093 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1096 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1098 local_irq_restore(flags
);
1103 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1105 void drain_local_pages(void *arg
)
1107 drain_pages(smp_processor_id());
1111 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1113 void drain_all_pages(void)
1115 on_each_cpu(drain_local_pages
, NULL
, 1);
1118 #ifdef CONFIG_HIBERNATION
1120 void mark_free_pages(struct zone
*zone
)
1122 unsigned long pfn
, max_zone_pfn
;
1123 unsigned long flags
;
1125 struct list_head
*curr
;
1127 if (!zone
->spanned_pages
)
1130 spin_lock_irqsave(&zone
->lock
, flags
);
1132 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1133 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1134 if (pfn_valid(pfn
)) {
1135 struct page
*page
= pfn_to_page(pfn
);
1137 if (!swsusp_page_is_forbidden(page
))
1138 swsusp_unset_page_free(page
);
1141 for_each_migratetype_order(order
, t
) {
1142 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1145 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1146 for (i
= 0; i
< (1UL << order
); i
++)
1147 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1150 spin_unlock_irqrestore(&zone
->lock
, flags
);
1152 #endif /* CONFIG_PM */
1155 * Free a 0-order page
1156 * cold == 1 ? free a cold page : free a hot page
1158 void free_hot_cold_page(struct page
*page
, int cold
)
1160 struct zone
*zone
= page_zone(page
);
1161 struct per_cpu_pages
*pcp
;
1162 unsigned long flags
;
1164 int wasMlocked
= __TestClearPageMlocked(page
);
1166 if (!free_pages_prepare(page
, 0))
1169 migratetype
= get_pageblock_migratetype(page
);
1170 set_page_private(page
, migratetype
);
1171 local_irq_save(flags
);
1172 if (unlikely(wasMlocked
))
1173 free_page_mlock(page
);
1174 __count_vm_event(PGFREE
);
1177 * We only track unmovable, reclaimable and movable on pcp lists.
1178 * Free ISOLATE pages back to the allocator because they are being
1179 * offlined but treat RESERVE as movable pages so we can get those
1180 * areas back if necessary. Otherwise, we may have to free
1181 * excessively into the page allocator
1183 if (migratetype
>= MIGRATE_PCPTYPES
) {
1184 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1185 free_one_page(zone
, page
, 0, migratetype
);
1188 migratetype
= MIGRATE_MOVABLE
;
1191 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1193 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1195 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1197 if (pcp
->count
>= pcp
->high
) {
1198 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1199 pcp
->count
-= pcp
->batch
;
1203 local_irq_restore(flags
);
1207 * split_page takes a non-compound higher-order page, and splits it into
1208 * n (1<<order) sub-pages: page[0..n]
1209 * Each sub-page must be freed individually.
1211 * Note: this is probably too low level an operation for use in drivers.
1212 * Please consult with lkml before using this in your driver.
1214 void split_page(struct page
*page
, unsigned int order
)
1218 VM_BUG_ON(PageCompound(page
));
1219 VM_BUG_ON(!page_count(page
));
1221 #ifdef CONFIG_KMEMCHECK
1223 * Split shadow pages too, because free(page[0]) would
1224 * otherwise free the whole shadow.
1226 if (kmemcheck_page_is_tracked(page
))
1227 split_page(virt_to_page(page
[0].shadow
), order
);
1230 for (i
= 1; i
< (1 << order
); i
++)
1231 set_page_refcounted(page
+ i
);
1235 * Similar to split_page except the page is already free. As this is only
1236 * being used for migration, the migratetype of the block also changes.
1237 * As this is called with interrupts disabled, the caller is responsible
1238 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1241 * Note: this is probably too low level an operation for use in drivers.
1242 * Please consult with lkml before using this in your driver.
1244 int split_free_page(struct page
*page
)
1247 unsigned long watermark
;
1250 BUG_ON(!PageBuddy(page
));
1252 zone
= page_zone(page
);
1253 order
= page_order(page
);
1255 /* Obey watermarks as if the page was being allocated */
1256 watermark
= low_wmark_pages(zone
) + (1 << order
);
1257 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1260 /* Remove page from free list */
1261 list_del(&page
->lru
);
1262 zone
->free_area
[order
].nr_free
--;
1263 rmv_page_order(page
);
1264 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1UL << order
));
1266 /* Split into individual pages */
1267 set_page_refcounted(page
);
1268 split_page(page
, order
);
1270 if (order
>= pageblock_order
- 1) {
1271 struct page
*endpage
= page
+ (1 << order
) - 1;
1272 for (; page
< endpage
; page
+= pageblock_nr_pages
)
1273 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1280 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1281 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1285 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1286 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1289 unsigned long flags
;
1291 int cold
= !!(gfp_flags
& __GFP_COLD
);
1294 if (likely(order
== 0)) {
1295 struct per_cpu_pages
*pcp
;
1296 struct list_head
*list
;
1298 local_irq_save(flags
);
1299 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1300 list
= &pcp
->lists
[migratetype
];
1301 if (list_empty(list
)) {
1302 pcp
->count
+= rmqueue_bulk(zone
, 0,
1305 if (unlikely(list_empty(list
)))
1310 page
= list_entry(list
->prev
, struct page
, lru
);
1312 page
= list_entry(list
->next
, struct page
, lru
);
1314 list_del(&page
->lru
);
1317 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1319 * __GFP_NOFAIL is not to be used in new code.
1321 * All __GFP_NOFAIL callers should be fixed so that they
1322 * properly detect and handle allocation failures.
1324 * We most definitely don't want callers attempting to
1325 * allocate greater than order-1 page units with
1328 WARN_ON_ONCE(order
> 1);
1330 spin_lock_irqsave(&zone
->lock
, flags
);
1331 page
= __rmqueue(zone
, order
, migratetype
);
1332 spin_unlock(&zone
->lock
);
1335 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << order
));
1338 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1339 zone_statistics(preferred_zone
, zone
);
1340 local_irq_restore(flags
);
1342 VM_BUG_ON(bad_range(zone
, page
));
1343 if (prep_new_page(page
, order
, gfp_flags
))
1348 local_irq_restore(flags
);
1352 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1353 #define ALLOC_WMARK_MIN WMARK_MIN
1354 #define ALLOC_WMARK_LOW WMARK_LOW
1355 #define ALLOC_WMARK_HIGH WMARK_HIGH
1356 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1358 /* Mask to get the watermark bits */
1359 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1361 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1362 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1363 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1365 #ifdef CONFIG_FAIL_PAGE_ALLOC
1367 static struct fail_page_alloc_attr
{
1368 struct fault_attr attr
;
1370 u32 ignore_gfp_highmem
;
1371 u32 ignore_gfp_wait
;
1374 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1376 struct dentry
*ignore_gfp_highmem_file
;
1377 struct dentry
*ignore_gfp_wait_file
;
1378 struct dentry
*min_order_file
;
1380 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1382 } fail_page_alloc
= {
1383 .attr
= FAULT_ATTR_INITIALIZER
,
1384 .ignore_gfp_wait
= 1,
1385 .ignore_gfp_highmem
= 1,
1389 static int __init
setup_fail_page_alloc(char *str
)
1391 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1393 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1395 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1397 if (order
< fail_page_alloc
.min_order
)
1399 if (gfp_mask
& __GFP_NOFAIL
)
1401 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1403 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1406 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1409 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1411 static int __init
fail_page_alloc_debugfs(void)
1413 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1417 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1421 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1423 fail_page_alloc
.ignore_gfp_wait_file
=
1424 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1425 &fail_page_alloc
.ignore_gfp_wait
);
1427 fail_page_alloc
.ignore_gfp_highmem_file
=
1428 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1429 &fail_page_alloc
.ignore_gfp_highmem
);
1430 fail_page_alloc
.min_order_file
=
1431 debugfs_create_u32("min-order", mode
, dir
,
1432 &fail_page_alloc
.min_order
);
1434 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1435 !fail_page_alloc
.ignore_gfp_highmem_file
||
1436 !fail_page_alloc
.min_order_file
) {
1438 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1439 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1440 debugfs_remove(fail_page_alloc
.min_order_file
);
1441 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1447 late_initcall(fail_page_alloc_debugfs
);
1449 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1451 #else /* CONFIG_FAIL_PAGE_ALLOC */
1453 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1458 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1461 * Return true if free pages are above 'mark'. This takes into account the order
1462 * of the allocation.
1464 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1465 int classzone_idx
, int alloc_flags
, long free_pages
)
1467 /* free_pages my go negative - that's OK */
1471 free_pages
-= (1 << order
) + 1;
1472 if (alloc_flags
& ALLOC_HIGH
)
1474 if (alloc_flags
& ALLOC_HARDER
)
1477 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1479 for (o
= 0; o
< order
; o
++) {
1480 /* At the next order, this order's pages become unavailable */
1481 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1483 /* Require fewer higher order pages to be free */
1486 if (free_pages
<= min
)
1492 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1493 int classzone_idx
, int alloc_flags
)
1495 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1496 zone_page_state(z
, NR_FREE_PAGES
));
1499 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1500 int classzone_idx
, int alloc_flags
)
1502 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1504 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1505 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1507 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1513 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1514 * skip over zones that are not allowed by the cpuset, or that have
1515 * been recently (in last second) found to be nearly full. See further
1516 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1517 * that have to skip over a lot of full or unallowed zones.
1519 * If the zonelist cache is present in the passed in zonelist, then
1520 * returns a pointer to the allowed node mask (either the current
1521 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1523 * If the zonelist cache is not available for this zonelist, does
1524 * nothing and returns NULL.
1526 * If the fullzones BITMAP in the zonelist cache is stale (more than
1527 * a second since last zap'd) then we zap it out (clear its bits.)
1529 * We hold off even calling zlc_setup, until after we've checked the
1530 * first zone in the zonelist, on the theory that most allocations will
1531 * be satisfied from that first zone, so best to examine that zone as
1532 * quickly as we can.
1534 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1536 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1537 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1539 zlc
= zonelist
->zlcache_ptr
;
1543 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1544 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1545 zlc
->last_full_zap
= jiffies
;
1548 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1549 &cpuset_current_mems_allowed
:
1550 &node_states
[N_HIGH_MEMORY
];
1551 return allowednodes
;
1555 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1556 * if it is worth looking at further for free memory:
1557 * 1) Check that the zone isn't thought to be full (doesn't have its
1558 * bit set in the zonelist_cache fullzones BITMAP).
1559 * 2) Check that the zones node (obtained from the zonelist_cache
1560 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1561 * Return true (non-zero) if zone is worth looking at further, or
1562 * else return false (zero) if it is not.
1564 * This check -ignores- the distinction between various watermarks,
1565 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1566 * found to be full for any variation of these watermarks, it will
1567 * be considered full for up to one second by all requests, unless
1568 * we are so low on memory on all allowed nodes that we are forced
1569 * into the second scan of the zonelist.
1571 * In the second scan we ignore this zonelist cache and exactly
1572 * apply the watermarks to all zones, even it is slower to do so.
1573 * We are low on memory in the second scan, and should leave no stone
1574 * unturned looking for a free page.
1576 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1577 nodemask_t
*allowednodes
)
1579 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1580 int i
; /* index of *z in zonelist zones */
1581 int n
; /* node that zone *z is on */
1583 zlc
= zonelist
->zlcache_ptr
;
1587 i
= z
- zonelist
->_zonerefs
;
1590 /* This zone is worth trying if it is allowed but not full */
1591 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1595 * Given 'z' scanning a zonelist, set the corresponding bit in
1596 * zlc->fullzones, so that subsequent attempts to allocate a page
1597 * from that zone don't waste time re-examining it.
1599 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1601 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1602 int i
; /* index of *z in zonelist zones */
1604 zlc
= zonelist
->zlcache_ptr
;
1608 i
= z
- zonelist
->_zonerefs
;
1610 set_bit(i
, zlc
->fullzones
);
1613 #else /* CONFIG_NUMA */
1615 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1620 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1621 nodemask_t
*allowednodes
)
1626 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1629 #endif /* CONFIG_NUMA */
1632 * get_page_from_freelist goes through the zonelist trying to allocate
1635 static struct page
*
1636 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1637 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1638 struct zone
*preferred_zone
, int migratetype
)
1641 struct page
*page
= NULL
;
1644 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1645 int zlc_active
= 0; /* set if using zonelist_cache */
1646 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1648 classzone_idx
= zone_idx(preferred_zone
);
1651 * Scan zonelist, looking for a zone with enough free.
1652 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1654 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1655 high_zoneidx
, nodemask
) {
1656 if (NUMA_BUILD
&& zlc_active
&&
1657 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1659 if ((alloc_flags
& ALLOC_CPUSET
) &&
1660 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1663 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1664 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1668 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1669 if (zone_watermark_ok(zone
, order
, mark
,
1670 classzone_idx
, alloc_flags
))
1673 if (zone_reclaim_mode
== 0)
1674 goto this_zone_full
;
1676 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1678 case ZONE_RECLAIM_NOSCAN
:
1681 case ZONE_RECLAIM_FULL
:
1682 /* scanned but unreclaimable */
1683 goto this_zone_full
;
1685 /* did we reclaim enough */
1686 if (!zone_watermark_ok(zone
, order
, mark
,
1687 classzone_idx
, alloc_flags
))
1688 goto this_zone_full
;
1693 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1694 gfp_mask
, migratetype
);
1699 zlc_mark_zone_full(zonelist
, z
);
1701 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1703 * we do zlc_setup after the first zone is tried but only
1704 * if there are multiple nodes make it worthwhile
1706 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1712 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1713 /* Disable zlc cache for second zonelist scan */
1721 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1722 unsigned long pages_reclaimed
)
1724 /* Do not loop if specifically requested */
1725 if (gfp_mask
& __GFP_NORETRY
)
1729 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1730 * means __GFP_NOFAIL, but that may not be true in other
1733 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1737 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1738 * specified, then we retry until we no longer reclaim any pages
1739 * (above), or we've reclaimed an order of pages at least as
1740 * large as the allocation's order. In both cases, if the
1741 * allocation still fails, we stop retrying.
1743 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1747 * Don't let big-order allocations loop unless the caller
1748 * explicitly requests that.
1750 if (gfp_mask
& __GFP_NOFAIL
)
1756 static inline struct page
*
1757 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1758 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1759 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1764 /* Acquire the OOM killer lock for the zones in zonelist */
1765 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
1766 schedule_timeout_uninterruptible(1);
1771 * Go through the zonelist yet one more time, keep very high watermark
1772 * here, this is only to catch a parallel oom killing, we must fail if
1773 * we're still under heavy pressure.
1775 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1776 order
, zonelist
, high_zoneidx
,
1777 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1778 preferred_zone
, migratetype
);
1782 if (!(gfp_mask
& __GFP_NOFAIL
)) {
1783 /* The OOM killer will not help higher order allocs */
1784 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1786 /* The OOM killer does not needlessly kill tasks for lowmem */
1787 if (high_zoneidx
< ZONE_NORMAL
)
1790 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1791 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1792 * The caller should handle page allocation failure by itself if
1793 * it specifies __GFP_THISNODE.
1794 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1796 if (gfp_mask
& __GFP_THISNODE
)
1799 /* Exhausted what can be done so it's blamo time */
1800 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
);
1803 clear_zonelist_oom(zonelist
, gfp_mask
);
1807 #ifdef CONFIG_COMPACTION
1808 /* Try memory compaction for high-order allocations before reclaim */
1809 static struct page
*
1810 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1811 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1812 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1813 int migratetype
, unsigned long *did_some_progress
,
1814 bool sync_migration
)
1817 struct task_struct
*tsk
= current
;
1819 if (!order
|| compaction_deferred(preferred_zone
))
1822 tsk
->flags
|= PF_MEMALLOC
;
1823 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
1824 nodemask
, sync_migration
);
1825 tsk
->flags
&= ~PF_MEMALLOC
;
1826 if (*did_some_progress
!= COMPACT_SKIPPED
) {
1828 /* Page migration frees to the PCP lists but we want merging */
1829 drain_pages(get_cpu());
1832 page
= get_page_from_freelist(gfp_mask
, nodemask
,
1833 order
, zonelist
, high_zoneidx
,
1834 alloc_flags
, preferred_zone
,
1837 preferred_zone
->compact_considered
= 0;
1838 preferred_zone
->compact_defer_shift
= 0;
1839 count_vm_event(COMPACTSUCCESS
);
1844 * It's bad if compaction run occurs and fails.
1845 * The most likely reason is that pages exist,
1846 * but not enough to satisfy watermarks.
1848 count_vm_event(COMPACTFAIL
);
1849 defer_compaction(preferred_zone
);
1857 static inline struct page
*
1858 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1859 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1860 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1861 int migratetype
, unsigned long *did_some_progress
,
1862 bool sync_migration
)
1866 #endif /* CONFIG_COMPACTION */
1868 /* The really slow allocator path where we enter direct reclaim */
1869 static inline struct page
*
1870 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1871 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1872 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1873 int migratetype
, unsigned long *did_some_progress
)
1875 struct page
*page
= NULL
;
1876 struct reclaim_state reclaim_state
;
1877 struct task_struct
*p
= current
;
1878 bool drained
= false;
1882 /* We now go into synchronous reclaim */
1883 cpuset_memory_pressure_bump();
1884 p
->flags
|= PF_MEMALLOC
;
1885 lockdep_set_current_reclaim_state(gfp_mask
);
1886 reclaim_state
.reclaimed_slab
= 0;
1887 p
->reclaim_state
= &reclaim_state
;
1889 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1891 p
->reclaim_state
= NULL
;
1892 lockdep_clear_current_reclaim_state();
1893 p
->flags
&= ~PF_MEMALLOC
;
1897 if (unlikely(!(*did_some_progress
)))
1901 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1902 zonelist
, high_zoneidx
,
1903 alloc_flags
, preferred_zone
,
1907 * If an allocation failed after direct reclaim, it could be because
1908 * pages are pinned on the per-cpu lists. Drain them and try again
1910 if (!page
&& !drained
) {
1920 * This is called in the allocator slow-path if the allocation request is of
1921 * sufficient urgency to ignore watermarks and take other desperate measures
1923 static inline struct page
*
1924 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1925 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1926 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1932 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1933 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1934 preferred_zone
, migratetype
);
1936 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1937 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
1938 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1944 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1945 enum zone_type high_zoneidx
,
1946 enum zone_type classzone_idx
)
1951 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1952 wakeup_kswapd(zone
, order
, classzone_idx
);
1956 gfp_to_alloc_flags(gfp_t gfp_mask
)
1958 struct task_struct
*p
= current
;
1959 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
1960 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1962 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1963 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
1966 * The caller may dip into page reserves a bit more if the caller
1967 * cannot run direct reclaim, or if the caller has realtime scheduling
1968 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1969 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1971 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
1975 * Not worth trying to allocate harder for
1976 * __GFP_NOMEMALLOC even if it can't schedule.
1978 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
1979 alloc_flags
|= ALLOC_HARDER
;
1981 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1982 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1984 alloc_flags
&= ~ALLOC_CPUSET
;
1985 } else if (unlikely(rt_task(p
)) && !in_interrupt())
1986 alloc_flags
|= ALLOC_HARDER
;
1988 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
1989 if (!in_interrupt() &&
1990 ((p
->flags
& PF_MEMALLOC
) ||
1991 unlikely(test_thread_flag(TIF_MEMDIE
))))
1992 alloc_flags
|= ALLOC_NO_WATERMARKS
;
1998 static inline struct page
*
1999 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2000 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2001 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2004 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2005 struct page
*page
= NULL
;
2007 unsigned long pages_reclaimed
= 0;
2008 unsigned long did_some_progress
;
2009 struct task_struct
*p
= current
;
2010 bool sync_migration
= false;
2013 * In the slowpath, we sanity check order to avoid ever trying to
2014 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2015 * be using allocators in order of preference for an area that is
2018 if (order
>= MAX_ORDER
) {
2019 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2024 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2025 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2026 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2027 * using a larger set of nodes after it has established that the
2028 * allowed per node queues are empty and that nodes are
2031 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2035 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2036 wake_all_kswapd(order
, zonelist
, high_zoneidx
,
2037 zone_idx(preferred_zone
));
2040 * OK, we're below the kswapd watermark and have kicked background
2041 * reclaim. Now things get more complex, so set up alloc_flags according
2042 * to how we want to proceed.
2044 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2046 /* This is the last chance, in general, before the goto nopage. */
2047 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2048 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2049 preferred_zone
, migratetype
);
2054 /* Allocate without watermarks if the context allows */
2055 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2056 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2057 zonelist
, high_zoneidx
, nodemask
,
2058 preferred_zone
, migratetype
);
2063 /* Atomic allocations - we can't balance anything */
2067 /* Avoid recursion of direct reclaim */
2068 if (p
->flags
& PF_MEMALLOC
)
2071 /* Avoid allocations with no watermarks from looping endlessly */
2072 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2076 * Try direct compaction. The first pass is asynchronous. Subsequent
2077 * attempts after direct reclaim are synchronous
2079 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2080 zonelist
, high_zoneidx
,
2082 alloc_flags
, preferred_zone
,
2083 migratetype
, &did_some_progress
,
2087 sync_migration
= true;
2089 /* Try direct reclaim and then allocating */
2090 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2091 zonelist
, high_zoneidx
,
2093 alloc_flags
, preferred_zone
,
2094 migratetype
, &did_some_progress
);
2099 * If we failed to make any progress reclaiming, then we are
2100 * running out of options and have to consider going OOM
2102 if (!did_some_progress
) {
2103 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2104 if (oom_killer_disabled
)
2106 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2107 zonelist
, high_zoneidx
,
2108 nodemask
, preferred_zone
,
2113 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2115 * The oom killer is not called for high-order
2116 * allocations that may fail, so if no progress
2117 * is being made, there are no other options and
2118 * retrying is unlikely to help.
2120 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2123 * The oom killer is not called for lowmem
2124 * allocations to prevent needlessly killing
2127 if (high_zoneidx
< ZONE_NORMAL
)
2135 /* Check if we should retry the allocation */
2136 pages_reclaimed
+= did_some_progress
;
2137 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
2138 /* Wait for some write requests to complete then retry */
2139 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2143 * High-order allocations do not necessarily loop after
2144 * direct reclaim and reclaim/compaction depends on compaction
2145 * being called after reclaim so call directly if necessary
2147 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2148 zonelist
, high_zoneidx
,
2150 alloc_flags
, preferred_zone
,
2151 migratetype
, &did_some_progress
,
2158 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
2159 printk(KERN_WARNING
"%s: page allocation failure."
2160 " order:%d, mode:0x%x\n",
2161 p
->comm
, order
, gfp_mask
);
2167 if (kmemcheck_enabled
)
2168 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2174 * This is the 'heart' of the zoned buddy allocator.
2177 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2178 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2180 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2181 struct zone
*preferred_zone
;
2183 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2185 gfp_mask
&= gfp_allowed_mask
;
2187 lockdep_trace_alloc(gfp_mask
);
2189 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2191 if (should_fail_alloc_page(gfp_mask
, order
))
2195 * Check the zones suitable for the gfp_mask contain at least one
2196 * valid zone. It's possible to have an empty zonelist as a result
2197 * of GFP_THISNODE and a memoryless node
2199 if (unlikely(!zonelist
->_zonerefs
->zone
))
2203 /* The preferred zone is used for statistics later */
2204 first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
, &preferred_zone
);
2205 if (!preferred_zone
) {
2210 /* First allocation attempt */
2211 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2212 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
2213 preferred_zone
, migratetype
);
2214 if (unlikely(!page
))
2215 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2216 zonelist
, high_zoneidx
, nodemask
,
2217 preferred_zone
, migratetype
);
2220 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2223 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2226 * Common helper functions.
2228 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2233 * __get_free_pages() returns a 32-bit address, which cannot represent
2236 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2238 page
= alloc_pages(gfp_mask
, order
);
2241 return (unsigned long) page_address(page
);
2243 EXPORT_SYMBOL(__get_free_pages
);
2245 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2247 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2249 EXPORT_SYMBOL(get_zeroed_page
);
2251 void __pagevec_free(struct pagevec
*pvec
)
2253 int i
= pagevec_count(pvec
);
2256 trace_mm_pagevec_free(pvec
->pages
[i
], pvec
->cold
);
2257 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
2261 void __free_pages(struct page
*page
, unsigned int order
)
2263 if (put_page_testzero(page
)) {
2265 free_hot_cold_page(page
, 0);
2267 __free_pages_ok(page
, order
);
2271 EXPORT_SYMBOL(__free_pages
);
2273 void free_pages(unsigned long addr
, unsigned int order
)
2276 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2277 __free_pages(virt_to_page((void *)addr
), order
);
2281 EXPORT_SYMBOL(free_pages
);
2284 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2285 * @size: the number of bytes to allocate
2286 * @gfp_mask: GFP flags for the allocation
2288 * This function is similar to alloc_pages(), except that it allocates the
2289 * minimum number of pages to satisfy the request. alloc_pages() can only
2290 * allocate memory in power-of-two pages.
2292 * This function is also limited by MAX_ORDER.
2294 * Memory allocated by this function must be released by free_pages_exact().
2296 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2298 unsigned int order
= get_order(size
);
2301 addr
= __get_free_pages(gfp_mask
, order
);
2303 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2304 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2306 split_page(virt_to_page((void *)addr
), order
);
2307 while (used
< alloc_end
) {
2313 return (void *)addr
;
2315 EXPORT_SYMBOL(alloc_pages_exact
);
2318 * free_pages_exact - release memory allocated via alloc_pages_exact()
2319 * @virt: the value returned by alloc_pages_exact.
2320 * @size: size of allocation, same value as passed to alloc_pages_exact().
2322 * Release the memory allocated by a previous call to alloc_pages_exact.
2324 void free_pages_exact(void *virt
, size_t size
)
2326 unsigned long addr
= (unsigned long)virt
;
2327 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2329 while (addr
< end
) {
2334 EXPORT_SYMBOL(free_pages_exact
);
2336 static unsigned int nr_free_zone_pages(int offset
)
2341 /* Just pick one node, since fallback list is circular */
2342 unsigned int sum
= 0;
2344 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2346 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2347 unsigned long size
= zone
->present_pages
;
2348 unsigned long high
= high_wmark_pages(zone
);
2357 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2359 unsigned int nr_free_buffer_pages(void)
2361 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2363 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2366 * Amount of free RAM allocatable within all zones
2368 unsigned int nr_free_pagecache_pages(void)
2370 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2373 static inline void show_node(struct zone
*zone
)
2376 printk("Node %d ", zone_to_nid(zone
));
2379 void si_meminfo(struct sysinfo
*val
)
2381 val
->totalram
= totalram_pages
;
2383 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2384 val
->bufferram
= nr_blockdev_pages();
2385 val
->totalhigh
= totalhigh_pages
;
2386 val
->freehigh
= nr_free_highpages();
2387 val
->mem_unit
= PAGE_SIZE
;
2390 EXPORT_SYMBOL(si_meminfo
);
2393 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2395 pg_data_t
*pgdat
= NODE_DATA(nid
);
2397 val
->totalram
= pgdat
->node_present_pages
;
2398 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2399 #ifdef CONFIG_HIGHMEM
2400 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2401 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2407 val
->mem_unit
= PAGE_SIZE
;
2411 #define K(x) ((x) << (PAGE_SHIFT-10))
2414 * Show free area list (used inside shift_scroll-lock stuff)
2415 * We also calculate the percentage fragmentation. We do this by counting the
2416 * memory on each free list with the exception of the first item on the list.
2418 void show_free_areas(void)
2423 for_each_populated_zone(zone
) {
2425 printk("%s per-cpu:\n", zone
->name
);
2427 for_each_online_cpu(cpu
) {
2428 struct per_cpu_pageset
*pageset
;
2430 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2432 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2433 cpu
, pageset
->pcp
.high
,
2434 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2438 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2439 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2441 " dirty:%lu writeback:%lu unstable:%lu\n"
2442 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2443 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2444 global_page_state(NR_ACTIVE_ANON
),
2445 global_page_state(NR_INACTIVE_ANON
),
2446 global_page_state(NR_ISOLATED_ANON
),
2447 global_page_state(NR_ACTIVE_FILE
),
2448 global_page_state(NR_INACTIVE_FILE
),
2449 global_page_state(NR_ISOLATED_FILE
),
2450 global_page_state(NR_UNEVICTABLE
),
2451 global_page_state(NR_FILE_DIRTY
),
2452 global_page_state(NR_WRITEBACK
),
2453 global_page_state(NR_UNSTABLE_NFS
),
2454 global_page_state(NR_FREE_PAGES
),
2455 global_page_state(NR_SLAB_RECLAIMABLE
),
2456 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2457 global_page_state(NR_FILE_MAPPED
),
2458 global_page_state(NR_SHMEM
),
2459 global_page_state(NR_PAGETABLE
),
2460 global_page_state(NR_BOUNCE
));
2462 for_each_populated_zone(zone
) {
2471 " active_anon:%lukB"
2472 " inactive_anon:%lukB"
2473 " active_file:%lukB"
2474 " inactive_file:%lukB"
2475 " unevictable:%lukB"
2476 " isolated(anon):%lukB"
2477 " isolated(file):%lukB"
2484 " slab_reclaimable:%lukB"
2485 " slab_unreclaimable:%lukB"
2486 " kernel_stack:%lukB"
2490 " writeback_tmp:%lukB"
2491 " pages_scanned:%lu"
2492 " all_unreclaimable? %s"
2495 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2496 K(min_wmark_pages(zone
)),
2497 K(low_wmark_pages(zone
)),
2498 K(high_wmark_pages(zone
)),
2499 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2500 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2501 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2502 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2503 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2504 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
2505 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
2506 K(zone
->present_pages
),
2507 K(zone_page_state(zone
, NR_MLOCK
)),
2508 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
2509 K(zone_page_state(zone
, NR_WRITEBACK
)),
2510 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
2511 K(zone_page_state(zone
, NR_SHMEM
)),
2512 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
2513 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
2514 zone_page_state(zone
, NR_KERNEL_STACK
) *
2516 K(zone_page_state(zone
, NR_PAGETABLE
)),
2517 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
2518 K(zone_page_state(zone
, NR_BOUNCE
)),
2519 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
2520 zone
->pages_scanned
,
2521 (zone
->all_unreclaimable
? "yes" : "no")
2523 printk("lowmem_reserve[]:");
2524 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2525 printk(" %lu", zone
->lowmem_reserve
[i
]);
2529 for_each_populated_zone(zone
) {
2530 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2533 printk("%s: ", zone
->name
);
2535 spin_lock_irqsave(&zone
->lock
, flags
);
2536 for (order
= 0; order
< MAX_ORDER
; order
++) {
2537 nr
[order
] = zone
->free_area
[order
].nr_free
;
2538 total
+= nr
[order
] << order
;
2540 spin_unlock_irqrestore(&zone
->lock
, flags
);
2541 for (order
= 0; order
< MAX_ORDER
; order
++)
2542 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2543 printk("= %lukB\n", K(total
));
2546 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2548 show_swap_cache_info();
2551 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2553 zoneref
->zone
= zone
;
2554 zoneref
->zone_idx
= zone_idx(zone
);
2558 * Builds allocation fallback zone lists.
2560 * Add all populated zones of a node to the zonelist.
2562 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2563 int nr_zones
, enum zone_type zone_type
)
2567 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2572 zone
= pgdat
->node_zones
+ zone_type
;
2573 if (populated_zone(zone
)) {
2574 zoneref_set_zone(zone
,
2575 &zonelist
->_zonerefs
[nr_zones
++]);
2576 check_highest_zone(zone_type
);
2579 } while (zone_type
);
2586 * 0 = automatic detection of better ordering.
2587 * 1 = order by ([node] distance, -zonetype)
2588 * 2 = order by (-zonetype, [node] distance)
2590 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2591 * the same zonelist. So only NUMA can configure this param.
2593 #define ZONELIST_ORDER_DEFAULT 0
2594 #define ZONELIST_ORDER_NODE 1
2595 #define ZONELIST_ORDER_ZONE 2
2597 /* zonelist order in the kernel.
2598 * set_zonelist_order() will set this to NODE or ZONE.
2600 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2601 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2605 /* The value user specified ....changed by config */
2606 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2607 /* string for sysctl */
2608 #define NUMA_ZONELIST_ORDER_LEN 16
2609 char numa_zonelist_order
[16] = "default";
2612 * interface for configure zonelist ordering.
2613 * command line option "numa_zonelist_order"
2614 * = "[dD]efault - default, automatic configuration.
2615 * = "[nN]ode - order by node locality, then by zone within node
2616 * = "[zZ]one - order by zone, then by locality within zone
2619 static int __parse_numa_zonelist_order(char *s
)
2621 if (*s
== 'd' || *s
== 'D') {
2622 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2623 } else if (*s
== 'n' || *s
== 'N') {
2624 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2625 } else if (*s
== 'z' || *s
== 'Z') {
2626 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2629 "Ignoring invalid numa_zonelist_order value: "
2636 static __init
int setup_numa_zonelist_order(char *s
)
2643 ret
= __parse_numa_zonelist_order(s
);
2645 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
2649 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2652 * sysctl handler for numa_zonelist_order
2654 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2655 void __user
*buffer
, size_t *length
,
2658 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2660 static DEFINE_MUTEX(zl_order_mutex
);
2662 mutex_lock(&zl_order_mutex
);
2664 strcpy(saved_string
, (char*)table
->data
);
2665 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
2669 int oldval
= user_zonelist_order
;
2670 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2672 * bogus value. restore saved string
2674 strncpy((char*)table
->data
, saved_string
,
2675 NUMA_ZONELIST_ORDER_LEN
);
2676 user_zonelist_order
= oldval
;
2677 } else if (oldval
!= user_zonelist_order
) {
2678 mutex_lock(&zonelists_mutex
);
2679 build_all_zonelists(NULL
);
2680 mutex_unlock(&zonelists_mutex
);
2684 mutex_unlock(&zl_order_mutex
);
2689 #define MAX_NODE_LOAD (nr_online_nodes)
2690 static int node_load
[MAX_NUMNODES
];
2693 * find_next_best_node - find the next node that should appear in a given node's fallback list
2694 * @node: node whose fallback list we're appending
2695 * @used_node_mask: nodemask_t of already used nodes
2697 * We use a number of factors to determine which is the next node that should
2698 * appear on a given node's fallback list. The node should not have appeared
2699 * already in @node's fallback list, and it should be the next closest node
2700 * according to the distance array (which contains arbitrary distance values
2701 * from each node to each node in the system), and should also prefer nodes
2702 * with no CPUs, since presumably they'll have very little allocation pressure
2703 * on them otherwise.
2704 * It returns -1 if no node is found.
2706 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2709 int min_val
= INT_MAX
;
2711 const struct cpumask
*tmp
= cpumask_of_node(0);
2713 /* Use the local node if we haven't already */
2714 if (!node_isset(node
, *used_node_mask
)) {
2715 node_set(node
, *used_node_mask
);
2719 for_each_node_state(n
, N_HIGH_MEMORY
) {
2721 /* Don't want a node to appear more than once */
2722 if (node_isset(n
, *used_node_mask
))
2725 /* Use the distance array to find the distance */
2726 val
= node_distance(node
, n
);
2728 /* Penalize nodes under us ("prefer the next node") */
2731 /* Give preference to headless and unused nodes */
2732 tmp
= cpumask_of_node(n
);
2733 if (!cpumask_empty(tmp
))
2734 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2736 /* Slight preference for less loaded node */
2737 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2738 val
+= node_load
[n
];
2740 if (val
< min_val
) {
2747 node_set(best_node
, *used_node_mask
);
2754 * Build zonelists ordered by node and zones within node.
2755 * This results in maximum locality--normal zone overflows into local
2756 * DMA zone, if any--but risks exhausting DMA zone.
2758 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2761 struct zonelist
*zonelist
;
2763 zonelist
= &pgdat
->node_zonelists
[0];
2764 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2766 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2768 zonelist
->_zonerefs
[j
].zone
= NULL
;
2769 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2773 * Build gfp_thisnode zonelists
2775 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2778 struct zonelist
*zonelist
;
2780 zonelist
= &pgdat
->node_zonelists
[1];
2781 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2782 zonelist
->_zonerefs
[j
].zone
= NULL
;
2783 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2787 * Build zonelists ordered by zone and nodes within zones.
2788 * This results in conserving DMA zone[s] until all Normal memory is
2789 * exhausted, but results in overflowing to remote node while memory
2790 * may still exist in local DMA zone.
2792 static int node_order
[MAX_NUMNODES
];
2794 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2797 int zone_type
; /* needs to be signed */
2799 struct zonelist
*zonelist
;
2801 zonelist
= &pgdat
->node_zonelists
[0];
2803 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2804 for (j
= 0; j
< nr_nodes
; j
++) {
2805 node
= node_order
[j
];
2806 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2807 if (populated_zone(z
)) {
2809 &zonelist
->_zonerefs
[pos
++]);
2810 check_highest_zone(zone_type
);
2814 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2815 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2818 static int default_zonelist_order(void)
2821 unsigned long low_kmem_size
,total_size
;
2825 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2826 * If they are really small and used heavily, the system can fall
2827 * into OOM very easily.
2828 * This function detect ZONE_DMA/DMA32 size and configures zone order.
2830 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2833 for_each_online_node(nid
) {
2834 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2835 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2836 if (populated_zone(z
)) {
2837 if (zone_type
< ZONE_NORMAL
)
2838 low_kmem_size
+= z
->present_pages
;
2839 total_size
+= z
->present_pages
;
2840 } else if (zone_type
== ZONE_NORMAL
) {
2842 * If any node has only lowmem, then node order
2843 * is preferred to allow kernel allocations
2844 * locally; otherwise, they can easily infringe
2845 * on other nodes when there is an abundance of
2846 * lowmem available to allocate from.
2848 return ZONELIST_ORDER_NODE
;
2852 if (!low_kmem_size
|| /* there are no DMA area. */
2853 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2854 return ZONELIST_ORDER_NODE
;
2856 * look into each node's config.
2857 * If there is a node whose DMA/DMA32 memory is very big area on
2858 * local memory, NODE_ORDER may be suitable.
2860 average_size
= total_size
/
2861 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2862 for_each_online_node(nid
) {
2865 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2866 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2867 if (populated_zone(z
)) {
2868 if (zone_type
< ZONE_NORMAL
)
2869 low_kmem_size
+= z
->present_pages
;
2870 total_size
+= z
->present_pages
;
2873 if (low_kmem_size
&&
2874 total_size
> average_size
&& /* ignore small node */
2875 low_kmem_size
> total_size
* 70/100)
2876 return ZONELIST_ORDER_NODE
;
2878 return ZONELIST_ORDER_ZONE
;
2881 static void set_zonelist_order(void)
2883 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2884 current_zonelist_order
= default_zonelist_order();
2886 current_zonelist_order
= user_zonelist_order
;
2889 static void build_zonelists(pg_data_t
*pgdat
)
2893 nodemask_t used_mask
;
2894 int local_node
, prev_node
;
2895 struct zonelist
*zonelist
;
2896 int order
= current_zonelist_order
;
2898 /* initialize zonelists */
2899 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2900 zonelist
= pgdat
->node_zonelists
+ i
;
2901 zonelist
->_zonerefs
[0].zone
= NULL
;
2902 zonelist
->_zonerefs
[0].zone_idx
= 0;
2905 /* NUMA-aware ordering of nodes */
2906 local_node
= pgdat
->node_id
;
2907 load
= nr_online_nodes
;
2908 prev_node
= local_node
;
2909 nodes_clear(used_mask
);
2911 memset(node_order
, 0, sizeof(node_order
));
2914 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2915 int distance
= node_distance(local_node
, node
);
2918 * If another node is sufficiently far away then it is better
2919 * to reclaim pages in a zone before going off node.
2921 if (distance
> RECLAIM_DISTANCE
)
2922 zone_reclaim_mode
= 1;
2925 * We don't want to pressure a particular node.
2926 * So adding penalty to the first node in same
2927 * distance group to make it round-robin.
2929 if (distance
!= node_distance(local_node
, prev_node
))
2930 node_load
[node
] = load
;
2934 if (order
== ZONELIST_ORDER_NODE
)
2935 build_zonelists_in_node_order(pgdat
, node
);
2937 node_order
[j
++] = node
; /* remember order */
2940 if (order
== ZONELIST_ORDER_ZONE
) {
2941 /* calculate node order -- i.e., DMA last! */
2942 build_zonelists_in_zone_order(pgdat
, j
);
2945 build_thisnode_zonelists(pgdat
);
2948 /* Construct the zonelist performance cache - see further mmzone.h */
2949 static void build_zonelist_cache(pg_data_t
*pgdat
)
2951 struct zonelist
*zonelist
;
2952 struct zonelist_cache
*zlc
;
2955 zonelist
= &pgdat
->node_zonelists
[0];
2956 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2957 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2958 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2959 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2962 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
2964 * Return node id of node used for "local" allocations.
2965 * I.e., first node id of first zone in arg node's generic zonelist.
2966 * Used for initializing percpu 'numa_mem', which is used primarily
2967 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
2969 int local_memory_node(int node
)
2973 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
2974 gfp_zone(GFP_KERNEL
),
2981 #else /* CONFIG_NUMA */
2983 static void set_zonelist_order(void)
2985 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2988 static void build_zonelists(pg_data_t
*pgdat
)
2990 int node
, local_node
;
2992 struct zonelist
*zonelist
;
2994 local_node
= pgdat
->node_id
;
2996 zonelist
= &pgdat
->node_zonelists
[0];
2997 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3000 * Now we build the zonelist so that it contains the zones
3001 * of all the other nodes.
3002 * We don't want to pressure a particular node, so when
3003 * building the zones for node N, we make sure that the
3004 * zones coming right after the local ones are those from
3005 * node N+1 (modulo N)
3007 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3008 if (!node_online(node
))
3010 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3013 for (node
= 0; node
< local_node
; node
++) {
3014 if (!node_online(node
))
3016 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3020 zonelist
->_zonerefs
[j
].zone
= NULL
;
3021 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3024 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3025 static void build_zonelist_cache(pg_data_t
*pgdat
)
3027 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3030 #endif /* CONFIG_NUMA */
3033 * Boot pageset table. One per cpu which is going to be used for all
3034 * zones and all nodes. The parameters will be set in such a way
3035 * that an item put on a list will immediately be handed over to
3036 * the buddy list. This is safe since pageset manipulation is done
3037 * with interrupts disabled.
3039 * The boot_pagesets must be kept even after bootup is complete for
3040 * unused processors and/or zones. They do play a role for bootstrapping
3041 * hotplugged processors.
3043 * zoneinfo_show() and maybe other functions do
3044 * not check if the processor is online before following the pageset pointer.
3045 * Other parts of the kernel may not check if the zone is available.
3047 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3048 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3049 static void setup_zone_pageset(struct zone
*zone
);
3052 * Global mutex to protect against size modification of zonelists
3053 * as well as to serialize pageset setup for the new populated zone.
3055 DEFINE_MUTEX(zonelists_mutex
);
3057 /* return values int ....just for stop_machine() */
3058 static __init_refok
int __build_all_zonelists(void *data
)
3064 memset(node_load
, 0, sizeof(node_load
));
3066 for_each_online_node(nid
) {
3067 pg_data_t
*pgdat
= NODE_DATA(nid
);
3069 build_zonelists(pgdat
);
3070 build_zonelist_cache(pgdat
);
3074 * Initialize the boot_pagesets that are going to be used
3075 * for bootstrapping processors. The real pagesets for
3076 * each zone will be allocated later when the per cpu
3077 * allocator is available.
3079 * boot_pagesets are used also for bootstrapping offline
3080 * cpus if the system is already booted because the pagesets
3081 * are needed to initialize allocators on a specific cpu too.
3082 * F.e. the percpu allocator needs the page allocator which
3083 * needs the percpu allocator in order to allocate its pagesets
3084 * (a chicken-egg dilemma).
3086 for_each_possible_cpu(cpu
) {
3087 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3089 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3091 * We now know the "local memory node" for each node--
3092 * i.e., the node of the first zone in the generic zonelist.
3093 * Set up numa_mem percpu variable for on-line cpus. During
3094 * boot, only the boot cpu should be on-line; we'll init the
3095 * secondary cpus' numa_mem as they come on-line. During
3096 * node/memory hotplug, we'll fixup all on-line cpus.
3098 if (cpu_online(cpu
))
3099 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3107 * Called with zonelists_mutex held always
3108 * unless system_state == SYSTEM_BOOTING.
3110 void build_all_zonelists(void *data
)
3112 set_zonelist_order();
3114 if (system_state
== SYSTEM_BOOTING
) {
3115 __build_all_zonelists(NULL
);
3116 mminit_verify_zonelist();
3117 cpuset_init_current_mems_allowed();
3119 /* we have to stop all cpus to guarantee there is no user
3121 #ifdef CONFIG_MEMORY_HOTPLUG
3123 setup_zone_pageset((struct zone
*)data
);
3125 stop_machine(__build_all_zonelists
, NULL
, NULL
);
3126 /* cpuset refresh routine should be here */
3128 vm_total_pages
= nr_free_pagecache_pages();
3130 * Disable grouping by mobility if the number of pages in the
3131 * system is too low to allow the mechanism to work. It would be
3132 * more accurate, but expensive to check per-zone. This check is
3133 * made on memory-hotadd so a system can start with mobility
3134 * disabled and enable it later
3136 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3137 page_group_by_mobility_disabled
= 1;
3139 page_group_by_mobility_disabled
= 0;
3141 printk("Built %i zonelists in %s order, mobility grouping %s. "
3142 "Total pages: %ld\n",
3144 zonelist_order_name
[current_zonelist_order
],
3145 page_group_by_mobility_disabled
? "off" : "on",
3148 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3153 * Helper functions to size the waitqueue hash table.
3154 * Essentially these want to choose hash table sizes sufficiently
3155 * large so that collisions trying to wait on pages are rare.
3156 * But in fact, the number of active page waitqueues on typical
3157 * systems is ridiculously low, less than 200. So this is even
3158 * conservative, even though it seems large.
3160 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3161 * waitqueues, i.e. the size of the waitq table given the number of pages.
3163 #define PAGES_PER_WAITQUEUE 256
3165 #ifndef CONFIG_MEMORY_HOTPLUG
3166 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3168 unsigned long size
= 1;
3170 pages
/= PAGES_PER_WAITQUEUE
;
3172 while (size
< pages
)
3176 * Once we have dozens or even hundreds of threads sleeping
3177 * on IO we've got bigger problems than wait queue collision.
3178 * Limit the size of the wait table to a reasonable size.
3180 size
= min(size
, 4096UL);
3182 return max(size
, 4UL);
3186 * A zone's size might be changed by hot-add, so it is not possible to determine
3187 * a suitable size for its wait_table. So we use the maximum size now.
3189 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3191 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3192 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3193 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3195 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3196 * or more by the traditional way. (See above). It equals:
3198 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3199 * ia64(16K page size) : = ( 8G + 4M)byte.
3200 * powerpc (64K page size) : = (32G +16M)byte.
3202 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3209 * This is an integer logarithm so that shifts can be used later
3210 * to extract the more random high bits from the multiplicative
3211 * hash function before the remainder is taken.
3213 static inline unsigned long wait_table_bits(unsigned long size
)
3218 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3221 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3222 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3223 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3224 * higher will lead to a bigger reserve which will get freed as contiguous
3225 * blocks as reclaim kicks in
3227 static void setup_zone_migrate_reserve(struct zone
*zone
)
3229 unsigned long start_pfn
, pfn
, end_pfn
;
3231 unsigned long block_migratetype
;
3234 /* Get the start pfn, end pfn and the number of blocks to reserve */
3235 start_pfn
= zone
->zone_start_pfn
;
3236 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3237 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3241 * Reserve blocks are generally in place to help high-order atomic
3242 * allocations that are short-lived. A min_free_kbytes value that
3243 * would result in more than 2 reserve blocks for atomic allocations
3244 * is assumed to be in place to help anti-fragmentation for the
3245 * future allocation of hugepages at runtime.
3247 reserve
= min(2, reserve
);
3249 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3250 if (!pfn_valid(pfn
))
3252 page
= pfn_to_page(pfn
);
3254 /* Watch out for overlapping nodes */
3255 if (page_to_nid(page
) != zone_to_nid(zone
))
3258 /* Blocks with reserved pages will never free, skip them. */
3259 if (PageReserved(page
))
3262 block_migratetype
= get_pageblock_migratetype(page
);
3264 /* If this block is reserved, account for it */
3265 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
3270 /* Suitable for reserving if this block is movable */
3271 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
3272 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
3273 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
3279 * If the reserve is met and this is a previous reserved block,
3282 if (block_migratetype
== MIGRATE_RESERVE
) {
3283 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3284 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3290 * Initially all pages are reserved - free ones are freed
3291 * up by free_all_bootmem() once the early boot process is
3292 * done. Non-atomic initialization, single-pass.
3294 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3295 unsigned long start_pfn
, enum memmap_context context
)
3298 unsigned long end_pfn
= start_pfn
+ size
;
3302 if (highest_memmap_pfn
< end_pfn
- 1)
3303 highest_memmap_pfn
= end_pfn
- 1;
3305 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3306 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3308 * There can be holes in boot-time mem_map[]s
3309 * handed to this function. They do not
3310 * exist on hotplugged memory.
3312 if (context
== MEMMAP_EARLY
) {
3313 if (!early_pfn_valid(pfn
))
3315 if (!early_pfn_in_nid(pfn
, nid
))
3318 page
= pfn_to_page(pfn
);
3319 set_page_links(page
, zone
, nid
, pfn
);
3320 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3321 init_page_count(page
);
3322 reset_page_mapcount(page
);
3323 SetPageReserved(page
);
3325 * Mark the block movable so that blocks are reserved for
3326 * movable at startup. This will force kernel allocations
3327 * to reserve their blocks rather than leaking throughout
3328 * the address space during boot when many long-lived
3329 * kernel allocations are made. Later some blocks near
3330 * the start are marked MIGRATE_RESERVE by
3331 * setup_zone_migrate_reserve()
3333 * bitmap is created for zone's valid pfn range. but memmap
3334 * can be created for invalid pages (for alignment)
3335 * check here not to call set_pageblock_migratetype() against
3338 if ((z
->zone_start_pfn
<= pfn
)
3339 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3340 && !(pfn
& (pageblock_nr_pages
- 1)))
3341 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3343 INIT_LIST_HEAD(&page
->lru
);
3344 #ifdef WANT_PAGE_VIRTUAL
3345 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3346 if (!is_highmem_idx(zone
))
3347 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3352 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3355 for_each_migratetype_order(order
, t
) {
3356 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3357 zone
->free_area
[order
].nr_free
= 0;
3361 #ifndef __HAVE_ARCH_MEMMAP_INIT
3362 #define memmap_init(size, nid, zone, start_pfn) \
3363 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3366 static int zone_batchsize(struct zone
*zone
)
3372 * The per-cpu-pages pools are set to around 1000th of the
3373 * size of the zone. But no more than 1/2 of a meg.
3375 * OK, so we don't know how big the cache is. So guess.
3377 batch
= zone
->present_pages
/ 1024;
3378 if (batch
* PAGE_SIZE
> 512 * 1024)
3379 batch
= (512 * 1024) / PAGE_SIZE
;
3380 batch
/= 4; /* We effectively *= 4 below */
3385 * Clamp the batch to a 2^n - 1 value. Having a power
3386 * of 2 value was found to be more likely to have
3387 * suboptimal cache aliasing properties in some cases.
3389 * For example if 2 tasks are alternately allocating
3390 * batches of pages, one task can end up with a lot
3391 * of pages of one half of the possible page colors
3392 * and the other with pages of the other colors.
3394 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3399 /* The deferral and batching of frees should be suppressed under NOMMU
3402 * The problem is that NOMMU needs to be able to allocate large chunks
3403 * of contiguous memory as there's no hardware page translation to
3404 * assemble apparent contiguous memory from discontiguous pages.
3406 * Queueing large contiguous runs of pages for batching, however,
3407 * causes the pages to actually be freed in smaller chunks. As there
3408 * can be a significant delay between the individual batches being
3409 * recycled, this leads to the once large chunks of space being
3410 * fragmented and becoming unavailable for high-order allocations.
3416 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3418 struct per_cpu_pages
*pcp
;
3421 memset(p
, 0, sizeof(*p
));
3425 pcp
->high
= 6 * batch
;
3426 pcp
->batch
= max(1UL, 1 * batch
);
3427 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3428 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3432 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3433 * to the value high for the pageset p.
3436 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3439 struct per_cpu_pages
*pcp
;
3443 pcp
->batch
= max(1UL, high
/4);
3444 if ((high
/4) > (PAGE_SHIFT
* 8))
3445 pcp
->batch
= PAGE_SHIFT
* 8;
3448 static __meminit
void setup_zone_pageset(struct zone
*zone
)
3452 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
3454 for_each_possible_cpu(cpu
) {
3455 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
3457 setup_pageset(pcp
, zone_batchsize(zone
));
3459 if (percpu_pagelist_fraction
)
3460 setup_pagelist_highmark(pcp
,
3461 (zone
->present_pages
/
3462 percpu_pagelist_fraction
));
3467 * Allocate per cpu pagesets and initialize them.
3468 * Before this call only boot pagesets were available.
3470 void __init
setup_per_cpu_pageset(void)
3474 for_each_populated_zone(zone
)
3475 setup_zone_pageset(zone
);
3478 static noinline __init_refok
3479 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3482 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3486 * The per-page waitqueue mechanism uses hashed waitqueues
3489 zone
->wait_table_hash_nr_entries
=
3490 wait_table_hash_nr_entries(zone_size_pages
);
3491 zone
->wait_table_bits
=
3492 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3493 alloc_size
= zone
->wait_table_hash_nr_entries
3494 * sizeof(wait_queue_head_t
);
3496 if (!slab_is_available()) {
3497 zone
->wait_table
= (wait_queue_head_t
*)
3498 alloc_bootmem_node(pgdat
, alloc_size
);
3501 * This case means that a zone whose size was 0 gets new memory
3502 * via memory hot-add.
3503 * But it may be the case that a new node was hot-added. In
3504 * this case vmalloc() will not be able to use this new node's
3505 * memory - this wait_table must be initialized to use this new
3506 * node itself as well.
3507 * To use this new node's memory, further consideration will be
3510 zone
->wait_table
= vmalloc(alloc_size
);
3512 if (!zone
->wait_table
)
3515 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3516 init_waitqueue_head(zone
->wait_table
+ i
);
3521 static int __zone_pcp_update(void *data
)
3523 struct zone
*zone
= data
;
3525 unsigned long batch
= zone_batchsize(zone
), flags
;
3527 for_each_possible_cpu(cpu
) {
3528 struct per_cpu_pageset
*pset
;
3529 struct per_cpu_pages
*pcp
;
3531 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
3534 local_irq_save(flags
);
3535 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
3536 setup_pageset(pset
, batch
);
3537 local_irq_restore(flags
);
3542 void zone_pcp_update(struct zone
*zone
)
3544 stop_machine(__zone_pcp_update
, zone
, NULL
);
3547 static __meminit
void zone_pcp_init(struct zone
*zone
)
3550 * per cpu subsystem is not up at this point. The following code
3551 * relies on the ability of the linker to provide the
3552 * offset of a (static) per cpu variable into the per cpu area.
3554 zone
->pageset
= &boot_pageset
;
3556 if (zone
->present_pages
)
3557 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
3558 zone
->name
, zone
->present_pages
,
3559 zone_batchsize(zone
));
3562 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3563 unsigned long zone_start_pfn
,
3565 enum memmap_context context
)
3567 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3569 ret
= zone_wait_table_init(zone
, size
);
3572 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3574 zone
->zone_start_pfn
= zone_start_pfn
;
3576 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3577 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3579 (unsigned long)zone_idx(zone
),
3580 zone_start_pfn
, (zone_start_pfn
+ size
));
3582 zone_init_free_lists(zone
);
3587 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3589 * Basic iterator support. Return the first range of PFNs for a node
3590 * Note: nid == MAX_NUMNODES returns first region regardless of node
3592 static int __meminit
first_active_region_index_in_nid(int nid
)
3596 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3597 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3604 * Basic iterator support. Return the next active range of PFNs for a node
3605 * Note: nid == MAX_NUMNODES returns next region regardless of node
3607 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3609 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3610 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3616 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3618 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3619 * Architectures may implement their own version but if add_active_range()
3620 * was used and there are no special requirements, this is a convenient
3623 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3627 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3628 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3629 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3631 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3632 return early_node_map
[i
].nid
;
3634 /* This is a memory hole */
3637 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3639 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3643 nid
= __early_pfn_to_nid(pfn
);
3646 /* just returns 0 */
3650 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3651 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3655 nid
= __early_pfn_to_nid(pfn
);
3656 if (nid
>= 0 && nid
!= node
)
3662 /* Basic iterator support to walk early_node_map[] */
3663 #define for_each_active_range_index_in_nid(i, nid) \
3664 for (i = first_active_region_index_in_nid(nid); i != -1; \
3665 i = next_active_region_index_in_nid(i, nid))
3668 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3669 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3670 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3672 * If an architecture guarantees that all ranges registered with
3673 * add_active_ranges() contain no holes and may be freed, this
3674 * this function may be used instead of calling free_bootmem() manually.
3676 void __init
free_bootmem_with_active_regions(int nid
,
3677 unsigned long max_low_pfn
)
3681 for_each_active_range_index_in_nid(i
, nid
) {
3682 unsigned long size_pages
= 0;
3683 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3685 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3688 if (end_pfn
> max_low_pfn
)
3689 end_pfn
= max_low_pfn
;
3691 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3692 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3693 PFN_PHYS(early_node_map
[i
].start_pfn
),
3694 size_pages
<< PAGE_SHIFT
);
3698 #ifdef CONFIG_HAVE_MEMBLOCK
3699 u64 __init
find_memory_core_early(int nid
, u64 size
, u64 align
,
3700 u64 goal
, u64 limit
)
3704 /* Need to go over early_node_map to find out good range for node */
3705 for_each_active_range_index_in_nid(i
, nid
) {
3707 u64 ei_start
, ei_last
;
3708 u64 final_start
, final_end
;
3710 ei_last
= early_node_map
[i
].end_pfn
;
3711 ei_last
<<= PAGE_SHIFT
;
3712 ei_start
= early_node_map
[i
].start_pfn
;
3713 ei_start
<<= PAGE_SHIFT
;
3715 final_start
= max(ei_start
, goal
);
3716 final_end
= min(ei_last
, limit
);
3718 if (final_start
>= final_end
)
3721 addr
= memblock_find_in_range(final_start
, final_end
, size
, align
);
3723 if (addr
== MEMBLOCK_ERROR
)
3729 return MEMBLOCK_ERROR
;
3733 int __init
add_from_early_node_map(struct range
*range
, int az
,
3734 int nr_range
, int nid
)
3739 /* need to go over early_node_map to find out good range for node */
3740 for_each_active_range_index_in_nid(i
, nid
) {
3741 start
= early_node_map
[i
].start_pfn
;
3742 end
= early_node_map
[i
].end_pfn
;
3743 nr_range
= add_range(range
, az
, nr_range
, start
, end
);
3748 #ifdef CONFIG_NO_BOOTMEM
3749 void * __init
__alloc_memory_core_early(int nid
, u64 size
, u64 align
,
3750 u64 goal
, u64 limit
)
3755 if (limit
> memblock
.current_limit
)
3756 limit
= memblock
.current_limit
;
3758 addr
= find_memory_core_early(nid
, size
, align
, goal
, limit
);
3760 if (addr
== MEMBLOCK_ERROR
)
3763 ptr
= phys_to_virt(addr
);
3764 memset(ptr
, 0, size
);
3765 memblock_x86_reserve_range(addr
, addr
+ size
, "BOOTMEM");
3767 * The min_count is set to 0 so that bootmem allocated blocks
3768 * are never reported as leaks.
3770 kmemleak_alloc(ptr
, size
, 0, 0);
3776 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3781 for_each_active_range_index_in_nid(i
, nid
) {
3782 ret
= work_fn(early_node_map
[i
].start_pfn
,
3783 early_node_map
[i
].end_pfn
, data
);
3789 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3790 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3792 * If an architecture guarantees that all ranges registered with
3793 * add_active_ranges() contain no holes and may be freed, this
3794 * function may be used instead of calling memory_present() manually.
3796 void __init
sparse_memory_present_with_active_regions(int nid
)
3800 for_each_active_range_index_in_nid(i
, nid
)
3801 memory_present(early_node_map
[i
].nid
,
3802 early_node_map
[i
].start_pfn
,
3803 early_node_map
[i
].end_pfn
);
3807 * get_pfn_range_for_nid - Return the start and end page frames for a node
3808 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3809 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3810 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3812 * It returns the start and end page frame of a node based on information
3813 * provided by an arch calling add_active_range(). If called for a node
3814 * with no available memory, a warning is printed and the start and end
3817 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3818 unsigned long *start_pfn
, unsigned long *end_pfn
)
3824 for_each_active_range_index_in_nid(i
, nid
) {
3825 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3826 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3829 if (*start_pfn
== -1UL)
3834 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3835 * assumption is made that zones within a node are ordered in monotonic
3836 * increasing memory addresses so that the "highest" populated zone is used
3838 static void __init
find_usable_zone_for_movable(void)
3841 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3842 if (zone_index
== ZONE_MOVABLE
)
3845 if (arch_zone_highest_possible_pfn
[zone_index
] >
3846 arch_zone_lowest_possible_pfn
[zone_index
])
3850 VM_BUG_ON(zone_index
== -1);
3851 movable_zone
= zone_index
;
3855 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3856 * because it is sized independant of architecture. Unlike the other zones,
3857 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3858 * in each node depending on the size of each node and how evenly kernelcore
3859 * is distributed. This helper function adjusts the zone ranges
3860 * provided by the architecture for a given node by using the end of the
3861 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3862 * zones within a node are in order of monotonic increases memory addresses
3864 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3865 unsigned long zone_type
,
3866 unsigned long node_start_pfn
,
3867 unsigned long node_end_pfn
,
3868 unsigned long *zone_start_pfn
,
3869 unsigned long *zone_end_pfn
)
3871 /* Only adjust if ZONE_MOVABLE is on this node */
3872 if (zone_movable_pfn
[nid
]) {
3873 /* Size ZONE_MOVABLE */
3874 if (zone_type
== ZONE_MOVABLE
) {
3875 *zone_start_pfn
= zone_movable_pfn
[nid
];
3876 *zone_end_pfn
= min(node_end_pfn
,
3877 arch_zone_highest_possible_pfn
[movable_zone
]);
3879 /* Adjust for ZONE_MOVABLE starting within this range */
3880 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3881 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3882 *zone_end_pfn
= zone_movable_pfn
[nid
];
3884 /* Check if this whole range is within ZONE_MOVABLE */
3885 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3886 *zone_start_pfn
= *zone_end_pfn
;
3891 * Return the number of pages a zone spans in a node, including holes
3892 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3894 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3895 unsigned long zone_type
,
3896 unsigned long *ignored
)
3898 unsigned long node_start_pfn
, node_end_pfn
;
3899 unsigned long zone_start_pfn
, zone_end_pfn
;
3901 /* Get the start and end of the node and zone */
3902 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3903 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3904 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3905 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3906 node_start_pfn
, node_end_pfn
,
3907 &zone_start_pfn
, &zone_end_pfn
);
3909 /* Check that this node has pages within the zone's required range */
3910 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3913 /* Move the zone boundaries inside the node if necessary */
3914 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3915 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3917 /* Return the spanned pages */
3918 return zone_end_pfn
- zone_start_pfn
;
3922 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3923 * then all holes in the requested range will be accounted for.
3925 unsigned long __meminit
__absent_pages_in_range(int nid
,
3926 unsigned long range_start_pfn
,
3927 unsigned long range_end_pfn
)
3930 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3931 unsigned long start_pfn
;
3933 /* Find the end_pfn of the first active range of pfns in the node */
3934 i
= first_active_region_index_in_nid(nid
);
3938 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3940 /* Account for ranges before physical memory on this node */
3941 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3942 hole_pages
= prev_end_pfn
- range_start_pfn
;
3944 /* Find all holes for the zone within the node */
3945 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3947 /* No need to continue if prev_end_pfn is outside the zone */
3948 if (prev_end_pfn
>= range_end_pfn
)
3951 /* Make sure the end of the zone is not within the hole */
3952 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3953 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3955 /* Update the hole size cound and move on */
3956 if (start_pfn
> range_start_pfn
) {
3957 BUG_ON(prev_end_pfn
> start_pfn
);
3958 hole_pages
+= start_pfn
- prev_end_pfn
;
3960 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3963 /* Account for ranges past physical memory on this node */
3964 if (range_end_pfn
> prev_end_pfn
)
3965 hole_pages
+= range_end_pfn
-
3966 max(range_start_pfn
, prev_end_pfn
);
3972 * absent_pages_in_range - Return number of page frames in holes within a range
3973 * @start_pfn: The start PFN to start searching for holes
3974 * @end_pfn: The end PFN to stop searching for holes
3976 * It returns the number of pages frames in memory holes within a range.
3978 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3979 unsigned long end_pfn
)
3981 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3984 /* Return the number of page frames in holes in a zone on a node */
3985 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3986 unsigned long zone_type
,
3987 unsigned long *ignored
)
3989 unsigned long node_start_pfn
, node_end_pfn
;
3990 unsigned long zone_start_pfn
, zone_end_pfn
;
3992 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3993 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3995 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3998 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3999 node_start_pfn
, node_end_pfn
,
4000 &zone_start_pfn
, &zone_end_pfn
);
4001 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4005 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4006 unsigned long zone_type
,
4007 unsigned long *zones_size
)
4009 return zones_size
[zone_type
];
4012 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4013 unsigned long zone_type
,
4014 unsigned long *zholes_size
)
4019 return zholes_size
[zone_type
];
4024 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4025 unsigned long *zones_size
, unsigned long *zholes_size
)
4027 unsigned long realtotalpages
, totalpages
= 0;
4030 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4031 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4033 pgdat
->node_spanned_pages
= totalpages
;
4035 realtotalpages
= totalpages
;
4036 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4038 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4040 pgdat
->node_present_pages
= realtotalpages
;
4041 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4045 #ifndef CONFIG_SPARSEMEM
4047 * Calculate the size of the zone->blockflags rounded to an unsigned long
4048 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4049 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4050 * round what is now in bits to nearest long in bits, then return it in
4053 static unsigned long __init
usemap_size(unsigned long zonesize
)
4055 unsigned long usemapsize
;
4057 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4058 usemapsize
= usemapsize
>> pageblock_order
;
4059 usemapsize
*= NR_PAGEBLOCK_BITS
;
4060 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4062 return usemapsize
/ 8;
4065 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4066 struct zone
*zone
, unsigned long zonesize
)
4068 unsigned long usemapsize
= usemap_size(zonesize
);
4069 zone
->pageblock_flags
= NULL
;
4071 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
4074 static inline void setup_usemap(struct pglist_data
*pgdat
,
4075 struct zone
*zone
, unsigned long zonesize
) {}
4076 #endif /* CONFIG_SPARSEMEM */
4078 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4080 /* Return a sensible default order for the pageblock size. */
4081 static inline int pageblock_default_order(void)
4083 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4084 return HUGETLB_PAGE_ORDER
;
4089 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4090 static inline void __init
set_pageblock_order(unsigned int order
)
4092 /* Check that pageblock_nr_pages has not already been setup */
4093 if (pageblock_order
)
4097 * Assume the largest contiguous order of interest is a huge page.
4098 * This value may be variable depending on boot parameters on IA64
4100 pageblock_order
= order
;
4102 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4105 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4106 * and pageblock_default_order() are unused as pageblock_order is set
4107 * at compile-time. See include/linux/pageblock-flags.h for the values of
4108 * pageblock_order based on the kernel config
4110 static inline int pageblock_default_order(unsigned int order
)
4114 #define set_pageblock_order(x) do {} while (0)
4116 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4119 * Set up the zone data structures:
4120 * - mark all pages reserved
4121 * - mark all memory queues empty
4122 * - clear the memory bitmaps
4124 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4125 unsigned long *zones_size
, unsigned long *zholes_size
)
4128 int nid
= pgdat
->node_id
;
4129 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4132 pgdat_resize_init(pgdat
);
4133 pgdat
->nr_zones
= 0;
4134 init_waitqueue_head(&pgdat
->kswapd_wait
);
4135 pgdat
->kswapd_max_order
= 0;
4136 pgdat_page_cgroup_init(pgdat
);
4138 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4139 struct zone
*zone
= pgdat
->node_zones
+ j
;
4140 unsigned long size
, realsize
, memmap_pages
;
4143 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4144 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
4148 * Adjust realsize so that it accounts for how much memory
4149 * is used by this zone for memmap. This affects the watermark
4150 * and per-cpu initialisations
4153 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
4154 if (realsize
>= memmap_pages
) {
4155 realsize
-= memmap_pages
;
4158 " %s zone: %lu pages used for memmap\n",
4159 zone_names
[j
], memmap_pages
);
4162 " %s zone: %lu pages exceeds realsize %lu\n",
4163 zone_names
[j
], memmap_pages
, realsize
);
4165 /* Account for reserved pages */
4166 if (j
== 0 && realsize
> dma_reserve
) {
4167 realsize
-= dma_reserve
;
4168 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4169 zone_names
[0], dma_reserve
);
4172 if (!is_highmem_idx(j
))
4173 nr_kernel_pages
+= realsize
;
4174 nr_all_pages
+= realsize
;
4176 zone
->spanned_pages
= size
;
4177 zone
->present_pages
= realsize
;
4180 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
4182 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
4184 zone
->name
= zone_names
[j
];
4185 spin_lock_init(&zone
->lock
);
4186 spin_lock_init(&zone
->lru_lock
);
4187 zone_seqlock_init(zone
);
4188 zone
->zone_pgdat
= pgdat
;
4190 zone_pcp_init(zone
);
4192 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
4193 zone
->reclaim_stat
.nr_saved_scan
[l
] = 0;
4195 zone
->reclaim_stat
.recent_rotated
[0] = 0;
4196 zone
->reclaim_stat
.recent_rotated
[1] = 0;
4197 zone
->reclaim_stat
.recent_scanned
[0] = 0;
4198 zone
->reclaim_stat
.recent_scanned
[1] = 0;
4199 zap_zone_vm_stats(zone
);
4204 set_pageblock_order(pageblock_default_order());
4205 setup_usemap(pgdat
, zone
, size
);
4206 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4207 size
, MEMMAP_EARLY
);
4209 memmap_init(size
, nid
, j
, zone_start_pfn
);
4210 zone_start_pfn
+= size
;
4214 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4216 /* Skip empty nodes */
4217 if (!pgdat
->node_spanned_pages
)
4220 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4221 /* ia64 gets its own node_mem_map, before this, without bootmem */
4222 if (!pgdat
->node_mem_map
) {
4223 unsigned long size
, start
, end
;
4227 * The zone's endpoints aren't required to be MAX_ORDER
4228 * aligned but the node_mem_map endpoints must be in order
4229 * for the buddy allocator to function correctly.
4231 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4232 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4233 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4234 size
= (end
- start
) * sizeof(struct page
);
4235 map
= alloc_remap(pgdat
->node_id
, size
);
4237 map
= alloc_bootmem_node(pgdat
, size
);
4238 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4240 #ifndef CONFIG_NEED_MULTIPLE_NODES
4242 * With no DISCONTIG, the global mem_map is just set as node 0's
4244 if (pgdat
== NODE_DATA(0)) {
4245 mem_map
= NODE_DATA(0)->node_mem_map
;
4246 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4247 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4248 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4249 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4252 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4255 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4256 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4258 pg_data_t
*pgdat
= NODE_DATA(nid
);
4260 pgdat
->node_id
= nid
;
4261 pgdat
->node_start_pfn
= node_start_pfn
;
4262 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4264 alloc_node_mem_map(pgdat
);
4265 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4266 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4267 nid
, (unsigned long)pgdat
,
4268 (unsigned long)pgdat
->node_mem_map
);
4271 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4274 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4276 #if MAX_NUMNODES > 1
4278 * Figure out the number of possible node ids.
4280 static void __init
setup_nr_node_ids(void)
4283 unsigned int highest
= 0;
4285 for_each_node_mask(node
, node_possible_map
)
4287 nr_node_ids
= highest
+ 1;
4290 static inline void setup_nr_node_ids(void)
4296 * add_active_range - Register a range of PFNs backed by physical memory
4297 * @nid: The node ID the range resides on
4298 * @start_pfn: The start PFN of the available physical memory
4299 * @end_pfn: The end PFN of the available physical memory
4301 * These ranges are stored in an early_node_map[] and later used by
4302 * free_area_init_nodes() to calculate zone sizes and holes. If the
4303 * range spans a memory hole, it is up to the architecture to ensure
4304 * the memory is not freed by the bootmem allocator. If possible
4305 * the range being registered will be merged with existing ranges.
4307 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
4308 unsigned long end_pfn
)
4312 mminit_dprintk(MMINIT_TRACE
, "memory_register",
4313 "Entering add_active_range(%d, %#lx, %#lx) "
4314 "%d entries of %d used\n",
4315 nid
, start_pfn
, end_pfn
,
4316 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
4318 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
4320 /* Merge with existing active regions if possible */
4321 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4322 if (early_node_map
[i
].nid
!= nid
)
4325 /* Skip if an existing region covers this new one */
4326 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
4327 end_pfn
<= early_node_map
[i
].end_pfn
)
4330 /* Merge forward if suitable */
4331 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
4332 end_pfn
> early_node_map
[i
].end_pfn
) {
4333 early_node_map
[i
].end_pfn
= end_pfn
;
4337 /* Merge backward if suitable */
4338 if (start_pfn
< early_node_map
[i
].start_pfn
&&
4339 end_pfn
>= early_node_map
[i
].start_pfn
) {
4340 early_node_map
[i
].start_pfn
= start_pfn
;
4345 /* Check that early_node_map is large enough */
4346 if (i
>= MAX_ACTIVE_REGIONS
) {
4347 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
4348 MAX_ACTIVE_REGIONS
);
4352 early_node_map
[i
].nid
= nid
;
4353 early_node_map
[i
].start_pfn
= start_pfn
;
4354 early_node_map
[i
].end_pfn
= end_pfn
;
4355 nr_nodemap_entries
= i
+ 1;
4359 * remove_active_range - Shrink an existing registered range of PFNs
4360 * @nid: The node id the range is on that should be shrunk
4361 * @start_pfn: The new PFN of the range
4362 * @end_pfn: The new PFN of the range
4364 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
4365 * The map is kept near the end physical page range that has already been
4366 * registered. This function allows an arch to shrink an existing registered
4369 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
4370 unsigned long end_pfn
)
4375 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
4376 nid
, start_pfn
, end_pfn
);
4378 /* Find the old active region end and shrink */
4379 for_each_active_range_index_in_nid(i
, nid
) {
4380 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4381 early_node_map
[i
].end_pfn
<= end_pfn
) {
4383 early_node_map
[i
].start_pfn
= 0;
4384 early_node_map
[i
].end_pfn
= 0;
4388 if (early_node_map
[i
].start_pfn
< start_pfn
&&
4389 early_node_map
[i
].end_pfn
> start_pfn
) {
4390 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
4391 early_node_map
[i
].end_pfn
= start_pfn
;
4392 if (temp_end_pfn
> end_pfn
)
4393 add_active_range(nid
, end_pfn
, temp_end_pfn
);
4396 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4397 early_node_map
[i
].end_pfn
> end_pfn
&&
4398 early_node_map
[i
].start_pfn
< end_pfn
) {
4399 early_node_map
[i
].start_pfn
= end_pfn
;
4407 /* remove the blank ones */
4408 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
4409 if (early_node_map
[i
].nid
!= nid
)
4411 if (early_node_map
[i
].end_pfn
)
4413 /* we found it, get rid of it */
4414 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
4415 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
4416 sizeof(early_node_map
[j
]));
4417 j
= nr_nodemap_entries
- 1;
4418 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
4419 nr_nodemap_entries
--;
4424 * remove_all_active_ranges - Remove all currently registered regions
4426 * During discovery, it may be found that a table like SRAT is invalid
4427 * and an alternative discovery method must be used. This function removes
4428 * all currently registered regions.
4430 void __init
remove_all_active_ranges(void)
4432 memset(early_node_map
, 0, sizeof(early_node_map
));
4433 nr_nodemap_entries
= 0;
4436 /* Compare two active node_active_regions */
4437 static int __init
cmp_node_active_region(const void *a
, const void *b
)
4439 struct node_active_region
*arange
= (struct node_active_region
*)a
;
4440 struct node_active_region
*brange
= (struct node_active_region
*)b
;
4442 /* Done this way to avoid overflows */
4443 if (arange
->start_pfn
> brange
->start_pfn
)
4445 if (arange
->start_pfn
< brange
->start_pfn
)
4451 /* sort the node_map by start_pfn */
4452 void __init
sort_node_map(void)
4454 sort(early_node_map
, (size_t)nr_nodemap_entries
,
4455 sizeof(struct node_active_region
),
4456 cmp_node_active_region
, NULL
);
4459 /* Find the lowest pfn for a node */
4460 static unsigned long __init
find_min_pfn_for_node(int nid
)
4463 unsigned long min_pfn
= ULONG_MAX
;
4465 /* Assuming a sorted map, the first range found has the starting pfn */
4466 for_each_active_range_index_in_nid(i
, nid
)
4467 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
4469 if (min_pfn
== ULONG_MAX
) {
4471 "Could not find start_pfn for node %d\n", nid
);
4479 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4481 * It returns the minimum PFN based on information provided via
4482 * add_active_range().
4484 unsigned long __init
find_min_pfn_with_active_regions(void)
4486 return find_min_pfn_for_node(MAX_NUMNODES
);
4490 * early_calculate_totalpages()
4491 * Sum pages in active regions for movable zone.
4492 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4494 static unsigned long __init
early_calculate_totalpages(void)
4497 unsigned long totalpages
= 0;
4499 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4500 unsigned long pages
= early_node_map
[i
].end_pfn
-
4501 early_node_map
[i
].start_pfn
;
4502 totalpages
+= pages
;
4504 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
4510 * Find the PFN the Movable zone begins in each node. Kernel memory
4511 * is spread evenly between nodes as long as the nodes have enough
4512 * memory. When they don't, some nodes will have more kernelcore than
4515 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
4518 unsigned long usable_startpfn
;
4519 unsigned long kernelcore_node
, kernelcore_remaining
;
4520 /* save the state before borrow the nodemask */
4521 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4522 unsigned long totalpages
= early_calculate_totalpages();
4523 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4526 * If movablecore was specified, calculate what size of
4527 * kernelcore that corresponds so that memory usable for
4528 * any allocation type is evenly spread. If both kernelcore
4529 * and movablecore are specified, then the value of kernelcore
4530 * will be used for required_kernelcore if it's greater than
4531 * what movablecore would have allowed.
4533 if (required_movablecore
) {
4534 unsigned long corepages
;
4537 * Round-up so that ZONE_MOVABLE is at least as large as what
4538 * was requested by the user
4540 required_movablecore
=
4541 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4542 corepages
= totalpages
- required_movablecore
;
4544 required_kernelcore
= max(required_kernelcore
, corepages
);
4547 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4548 if (!required_kernelcore
)
4551 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4552 find_usable_zone_for_movable();
4553 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4556 /* Spread kernelcore memory as evenly as possible throughout nodes */
4557 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4558 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4560 * Recalculate kernelcore_node if the division per node
4561 * now exceeds what is necessary to satisfy the requested
4562 * amount of memory for the kernel
4564 if (required_kernelcore
< kernelcore_node
)
4565 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4568 * As the map is walked, we track how much memory is usable
4569 * by the kernel using kernelcore_remaining. When it is
4570 * 0, the rest of the node is usable by ZONE_MOVABLE
4572 kernelcore_remaining
= kernelcore_node
;
4574 /* Go through each range of PFNs within this node */
4575 for_each_active_range_index_in_nid(i
, nid
) {
4576 unsigned long start_pfn
, end_pfn
;
4577 unsigned long size_pages
;
4579 start_pfn
= max(early_node_map
[i
].start_pfn
,
4580 zone_movable_pfn
[nid
]);
4581 end_pfn
= early_node_map
[i
].end_pfn
;
4582 if (start_pfn
>= end_pfn
)
4585 /* Account for what is only usable for kernelcore */
4586 if (start_pfn
< usable_startpfn
) {
4587 unsigned long kernel_pages
;
4588 kernel_pages
= min(end_pfn
, usable_startpfn
)
4591 kernelcore_remaining
-= min(kernel_pages
,
4592 kernelcore_remaining
);
4593 required_kernelcore
-= min(kernel_pages
,
4594 required_kernelcore
);
4596 /* Continue if range is now fully accounted */
4597 if (end_pfn
<= usable_startpfn
) {
4600 * Push zone_movable_pfn to the end so
4601 * that if we have to rebalance
4602 * kernelcore across nodes, we will
4603 * not double account here
4605 zone_movable_pfn
[nid
] = end_pfn
;
4608 start_pfn
= usable_startpfn
;
4612 * The usable PFN range for ZONE_MOVABLE is from
4613 * start_pfn->end_pfn. Calculate size_pages as the
4614 * number of pages used as kernelcore
4616 size_pages
= end_pfn
- start_pfn
;
4617 if (size_pages
> kernelcore_remaining
)
4618 size_pages
= kernelcore_remaining
;
4619 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4622 * Some kernelcore has been met, update counts and
4623 * break if the kernelcore for this node has been
4626 required_kernelcore
-= min(required_kernelcore
,
4628 kernelcore_remaining
-= size_pages
;
4629 if (!kernelcore_remaining
)
4635 * If there is still required_kernelcore, we do another pass with one
4636 * less node in the count. This will push zone_movable_pfn[nid] further
4637 * along on the nodes that still have memory until kernelcore is
4641 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4644 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4645 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4646 zone_movable_pfn
[nid
] =
4647 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4650 /* restore the node_state */
4651 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4654 /* Any regular memory on that node ? */
4655 static void check_for_regular_memory(pg_data_t
*pgdat
)
4657 #ifdef CONFIG_HIGHMEM
4658 enum zone_type zone_type
;
4660 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4661 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4662 if (zone
->present_pages
)
4663 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4669 * free_area_init_nodes - Initialise all pg_data_t and zone data
4670 * @max_zone_pfn: an array of max PFNs for each zone
4672 * This will call free_area_init_node() for each active node in the system.
4673 * Using the page ranges provided by add_active_range(), the size of each
4674 * zone in each node and their holes is calculated. If the maximum PFN
4675 * between two adjacent zones match, it is assumed that the zone is empty.
4676 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4677 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4678 * starts where the previous one ended. For example, ZONE_DMA32 starts
4679 * at arch_max_dma_pfn.
4681 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4686 /* Sort early_node_map as initialisation assumes it is sorted */
4689 /* Record where the zone boundaries are */
4690 memset(arch_zone_lowest_possible_pfn
, 0,
4691 sizeof(arch_zone_lowest_possible_pfn
));
4692 memset(arch_zone_highest_possible_pfn
, 0,
4693 sizeof(arch_zone_highest_possible_pfn
));
4694 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4695 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4696 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4697 if (i
== ZONE_MOVABLE
)
4699 arch_zone_lowest_possible_pfn
[i
] =
4700 arch_zone_highest_possible_pfn
[i
-1];
4701 arch_zone_highest_possible_pfn
[i
] =
4702 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4704 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4705 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4707 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4708 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4709 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4711 /* Print out the zone ranges */
4712 printk("Zone PFN ranges:\n");
4713 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4714 if (i
== ZONE_MOVABLE
)
4716 printk(" %-8s ", zone_names
[i
]);
4717 if (arch_zone_lowest_possible_pfn
[i
] ==
4718 arch_zone_highest_possible_pfn
[i
])
4721 printk("%0#10lx -> %0#10lx\n",
4722 arch_zone_lowest_possible_pfn
[i
],
4723 arch_zone_highest_possible_pfn
[i
]);
4726 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4727 printk("Movable zone start PFN for each node\n");
4728 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4729 if (zone_movable_pfn
[i
])
4730 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4733 /* Print out the early_node_map[] */
4734 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4735 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4736 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4737 early_node_map
[i
].start_pfn
,
4738 early_node_map
[i
].end_pfn
);
4740 /* Initialise every node */
4741 mminit_verify_pageflags_layout();
4742 setup_nr_node_ids();
4743 for_each_online_node(nid
) {
4744 pg_data_t
*pgdat
= NODE_DATA(nid
);
4745 free_area_init_node(nid
, NULL
,
4746 find_min_pfn_for_node(nid
), NULL
);
4748 /* Any memory on that node */
4749 if (pgdat
->node_present_pages
)
4750 node_set_state(nid
, N_HIGH_MEMORY
);
4751 check_for_regular_memory(pgdat
);
4755 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4757 unsigned long long coremem
;
4761 coremem
= memparse(p
, &p
);
4762 *core
= coremem
>> PAGE_SHIFT
;
4764 /* Paranoid check that UL is enough for the coremem value */
4765 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4771 * kernelcore=size sets the amount of memory for use for allocations that
4772 * cannot be reclaimed or migrated.
4774 static int __init
cmdline_parse_kernelcore(char *p
)
4776 return cmdline_parse_core(p
, &required_kernelcore
);
4780 * movablecore=size sets the amount of memory for use for allocations that
4781 * can be reclaimed or migrated.
4783 static int __init
cmdline_parse_movablecore(char *p
)
4785 return cmdline_parse_core(p
, &required_movablecore
);
4788 early_param("kernelcore", cmdline_parse_kernelcore
);
4789 early_param("movablecore", cmdline_parse_movablecore
);
4791 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4794 * set_dma_reserve - set the specified number of pages reserved in the first zone
4795 * @new_dma_reserve: The number of pages to mark reserved
4797 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4798 * In the DMA zone, a significant percentage may be consumed by kernel image
4799 * and other unfreeable allocations which can skew the watermarks badly. This
4800 * function may optionally be used to account for unfreeable pages in the
4801 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4802 * smaller per-cpu batchsize.
4804 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4806 dma_reserve
= new_dma_reserve
;
4809 #ifndef CONFIG_NEED_MULTIPLE_NODES
4810 struct pglist_data __refdata contig_page_data
= {
4811 #ifndef CONFIG_NO_BOOTMEM
4812 .bdata
= &bootmem_node_data
[0]
4815 EXPORT_SYMBOL(contig_page_data
);
4818 void __init
free_area_init(unsigned long *zones_size
)
4820 free_area_init_node(0, zones_size
,
4821 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4824 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4825 unsigned long action
, void *hcpu
)
4827 int cpu
= (unsigned long)hcpu
;
4829 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4833 * Spill the event counters of the dead processor
4834 * into the current processors event counters.
4835 * This artificially elevates the count of the current
4838 vm_events_fold_cpu(cpu
);
4841 * Zero the differential counters of the dead processor
4842 * so that the vm statistics are consistent.
4844 * This is only okay since the processor is dead and cannot
4845 * race with what we are doing.
4847 refresh_cpu_vm_stats(cpu
);
4852 void __init
page_alloc_init(void)
4854 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4858 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4859 * or min_free_kbytes changes.
4861 static void calculate_totalreserve_pages(void)
4863 struct pglist_data
*pgdat
;
4864 unsigned long reserve_pages
= 0;
4865 enum zone_type i
, j
;
4867 for_each_online_pgdat(pgdat
) {
4868 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4869 struct zone
*zone
= pgdat
->node_zones
+ i
;
4870 unsigned long max
= 0;
4872 /* Find valid and maximum lowmem_reserve in the zone */
4873 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4874 if (zone
->lowmem_reserve
[j
] > max
)
4875 max
= zone
->lowmem_reserve
[j
];
4878 /* we treat the high watermark as reserved pages. */
4879 max
+= high_wmark_pages(zone
);
4881 if (max
> zone
->present_pages
)
4882 max
= zone
->present_pages
;
4883 reserve_pages
+= max
;
4886 totalreserve_pages
= reserve_pages
;
4890 * setup_per_zone_lowmem_reserve - called whenever
4891 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4892 * has a correct pages reserved value, so an adequate number of
4893 * pages are left in the zone after a successful __alloc_pages().
4895 static void setup_per_zone_lowmem_reserve(void)
4897 struct pglist_data
*pgdat
;
4898 enum zone_type j
, idx
;
4900 for_each_online_pgdat(pgdat
) {
4901 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4902 struct zone
*zone
= pgdat
->node_zones
+ j
;
4903 unsigned long present_pages
= zone
->present_pages
;
4905 zone
->lowmem_reserve
[j
] = 0;
4909 struct zone
*lower_zone
;
4913 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4914 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4916 lower_zone
= pgdat
->node_zones
+ idx
;
4917 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4918 sysctl_lowmem_reserve_ratio
[idx
];
4919 present_pages
+= lower_zone
->present_pages
;
4924 /* update totalreserve_pages */
4925 calculate_totalreserve_pages();
4929 * setup_per_zone_wmarks - called when min_free_kbytes changes
4930 * or when memory is hot-{added|removed}
4932 * Ensures that the watermark[min,low,high] values for each zone are set
4933 * correctly with respect to min_free_kbytes.
4935 void setup_per_zone_wmarks(void)
4937 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4938 unsigned long lowmem_pages
= 0;
4940 unsigned long flags
;
4942 /* Calculate total number of !ZONE_HIGHMEM pages */
4943 for_each_zone(zone
) {
4944 if (!is_highmem(zone
))
4945 lowmem_pages
+= zone
->present_pages
;
4948 for_each_zone(zone
) {
4951 spin_lock_irqsave(&zone
->lock
, flags
);
4952 tmp
= (u64
)pages_min
* zone
->present_pages
;
4953 do_div(tmp
, lowmem_pages
);
4954 if (is_highmem(zone
)) {
4956 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4957 * need highmem pages, so cap pages_min to a small
4960 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4961 * deltas controls asynch page reclaim, and so should
4962 * not be capped for highmem.
4966 min_pages
= zone
->present_pages
/ 1024;
4967 if (min_pages
< SWAP_CLUSTER_MAX
)
4968 min_pages
= SWAP_CLUSTER_MAX
;
4969 if (min_pages
> 128)
4971 zone
->watermark
[WMARK_MIN
] = min_pages
;
4974 * If it's a lowmem zone, reserve a number of pages
4975 * proportionate to the zone's size.
4977 zone
->watermark
[WMARK_MIN
] = tmp
;
4980 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
4981 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
4982 setup_zone_migrate_reserve(zone
);
4983 spin_unlock_irqrestore(&zone
->lock
, flags
);
4986 /* update totalreserve_pages */
4987 calculate_totalreserve_pages();
4991 * The inactive anon list should be small enough that the VM never has to
4992 * do too much work, but large enough that each inactive page has a chance
4993 * to be referenced again before it is swapped out.
4995 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4996 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4997 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4998 * the anonymous pages are kept on the inactive list.
5001 * memory ratio inactive anon
5002 * -------------------------------------
5011 void calculate_zone_inactive_ratio(struct zone
*zone
)
5013 unsigned int gb
, ratio
;
5015 /* Zone size in gigabytes */
5016 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
5018 ratio
= int_sqrt(10 * gb
);
5022 zone
->inactive_ratio
= ratio
;
5025 static void __init
setup_per_zone_inactive_ratio(void)
5030 calculate_zone_inactive_ratio(zone
);
5034 * Initialise min_free_kbytes.
5036 * For small machines we want it small (128k min). For large machines
5037 * we want it large (64MB max). But it is not linear, because network
5038 * bandwidth does not increase linearly with machine size. We use
5040 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5041 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5057 static int __init
init_per_zone_wmark_min(void)
5059 unsigned long lowmem_kbytes
;
5061 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5063 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5064 if (min_free_kbytes
< 128)
5065 min_free_kbytes
= 128;
5066 if (min_free_kbytes
> 65536)
5067 min_free_kbytes
= 65536;
5068 setup_per_zone_wmarks();
5069 setup_per_zone_lowmem_reserve();
5070 setup_per_zone_inactive_ratio();
5073 module_init(init_per_zone_wmark_min
)
5076 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5077 * that we can call two helper functions whenever min_free_kbytes
5080 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5081 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5083 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5085 setup_per_zone_wmarks();
5090 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5091 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5096 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5101 zone
->min_unmapped_pages
= (zone
->present_pages
*
5102 sysctl_min_unmapped_ratio
) / 100;
5106 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5107 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5112 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5117 zone
->min_slab_pages
= (zone
->present_pages
*
5118 sysctl_min_slab_ratio
) / 100;
5124 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5125 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5126 * whenever sysctl_lowmem_reserve_ratio changes.
5128 * The reserve ratio obviously has absolutely no relation with the
5129 * minimum watermarks. The lowmem reserve ratio can only make sense
5130 * if in function of the boot time zone sizes.
5132 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5133 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5135 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5136 setup_per_zone_lowmem_reserve();
5141 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5142 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5143 * can have before it gets flushed back to buddy allocator.
5146 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5147 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5153 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5154 if (!write
|| (ret
== -EINVAL
))
5156 for_each_populated_zone(zone
) {
5157 for_each_possible_cpu(cpu
) {
5159 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5160 setup_pagelist_highmark(
5161 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5167 int hashdist
= HASHDIST_DEFAULT
;
5170 static int __init
set_hashdist(char *str
)
5174 hashdist
= simple_strtoul(str
, &str
, 0);
5177 __setup("hashdist=", set_hashdist
);
5181 * allocate a large system hash table from bootmem
5182 * - it is assumed that the hash table must contain an exact power-of-2
5183 * quantity of entries
5184 * - limit is the number of hash buckets, not the total allocation size
5186 void *__init
alloc_large_system_hash(const char *tablename
,
5187 unsigned long bucketsize
,
5188 unsigned long numentries
,
5191 unsigned int *_hash_shift
,
5192 unsigned int *_hash_mask
,
5193 unsigned long limit
)
5195 unsigned long long max
= limit
;
5196 unsigned long log2qty
, size
;
5199 /* allow the kernel cmdline to have a say */
5201 /* round applicable memory size up to nearest megabyte */
5202 numentries
= nr_kernel_pages
;
5203 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5204 numentries
>>= 20 - PAGE_SHIFT
;
5205 numentries
<<= 20 - PAGE_SHIFT
;
5207 /* limit to 1 bucket per 2^scale bytes of low memory */
5208 if (scale
> PAGE_SHIFT
)
5209 numentries
>>= (scale
- PAGE_SHIFT
);
5211 numentries
<<= (PAGE_SHIFT
- scale
);
5213 /* Make sure we've got at least a 0-order allocation.. */
5214 if (unlikely(flags
& HASH_SMALL
)) {
5215 /* Makes no sense without HASH_EARLY */
5216 WARN_ON(!(flags
& HASH_EARLY
));
5217 if (!(numentries
>> *_hash_shift
)) {
5218 numentries
= 1UL << *_hash_shift
;
5219 BUG_ON(!numentries
);
5221 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5222 numentries
= PAGE_SIZE
/ bucketsize
;
5224 numentries
= roundup_pow_of_two(numentries
);
5226 /* limit allocation size to 1/16 total memory by default */
5228 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5229 do_div(max
, bucketsize
);
5232 if (numentries
> max
)
5235 log2qty
= ilog2(numentries
);
5238 size
= bucketsize
<< log2qty
;
5239 if (flags
& HASH_EARLY
)
5240 table
= alloc_bootmem_nopanic(size
);
5242 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5245 * If bucketsize is not a power-of-two, we may free
5246 * some pages at the end of hash table which
5247 * alloc_pages_exact() automatically does
5249 if (get_order(size
) < MAX_ORDER
) {
5250 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5251 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5254 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5257 panic("Failed to allocate %s hash table\n", tablename
);
5259 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5262 ilog2(size
) - PAGE_SHIFT
,
5266 *_hash_shift
= log2qty
;
5268 *_hash_mask
= (1 << log2qty
) - 1;
5273 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5274 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5277 #ifdef CONFIG_SPARSEMEM
5278 return __pfn_to_section(pfn
)->pageblock_flags
;
5280 return zone
->pageblock_flags
;
5281 #endif /* CONFIG_SPARSEMEM */
5284 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5286 #ifdef CONFIG_SPARSEMEM
5287 pfn
&= (PAGES_PER_SECTION
-1);
5288 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5290 pfn
= pfn
- zone
->zone_start_pfn
;
5291 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5292 #endif /* CONFIG_SPARSEMEM */
5296 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5297 * @page: The page within the block of interest
5298 * @start_bitidx: The first bit of interest to retrieve
5299 * @end_bitidx: The last bit of interest
5300 * returns pageblock_bits flags
5302 unsigned long get_pageblock_flags_group(struct page
*page
,
5303 int start_bitidx
, int end_bitidx
)
5306 unsigned long *bitmap
;
5307 unsigned long pfn
, bitidx
;
5308 unsigned long flags
= 0;
5309 unsigned long value
= 1;
5311 zone
= page_zone(page
);
5312 pfn
= page_to_pfn(page
);
5313 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5314 bitidx
= pfn_to_bitidx(zone
, pfn
);
5316 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5317 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5324 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5325 * @page: The page within the block of interest
5326 * @start_bitidx: The first bit of interest
5327 * @end_bitidx: The last bit of interest
5328 * @flags: The flags to set
5330 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5331 int start_bitidx
, int end_bitidx
)
5334 unsigned long *bitmap
;
5335 unsigned long pfn
, bitidx
;
5336 unsigned long value
= 1;
5338 zone
= page_zone(page
);
5339 pfn
= page_to_pfn(page
);
5340 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5341 bitidx
= pfn_to_bitidx(zone
, pfn
);
5342 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5343 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5345 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5347 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5349 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5353 * This is designed as sub function...plz see page_isolation.c also.
5354 * set/clear page block's type to be ISOLATE.
5355 * page allocater never alloc memory from ISOLATE block.
5359 __count_immobile_pages(struct zone
*zone
, struct page
*page
, int count
)
5361 unsigned long pfn
, iter
, found
;
5363 * For avoiding noise data, lru_add_drain_all() should be called
5364 * If ZONE_MOVABLE, the zone never contains immobile pages
5366 if (zone_idx(zone
) == ZONE_MOVABLE
)
5369 if (get_pageblock_migratetype(page
) == MIGRATE_MOVABLE
)
5372 pfn
= page_to_pfn(page
);
5373 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
5374 unsigned long check
= pfn
+ iter
;
5376 if (!pfn_valid_within(check
)) {
5380 page
= pfn_to_page(check
);
5381 if (!page_count(page
)) {
5382 if (PageBuddy(page
))
5383 iter
+= (1 << page_order(page
)) - 1;
5389 * If there are RECLAIMABLE pages, we need to check it.
5390 * But now, memory offline itself doesn't call shrink_slab()
5391 * and it still to be fixed.
5394 * If the page is not RAM, page_count()should be 0.
5395 * we don't need more check. This is an _used_ not-movable page.
5397 * The problematic thing here is PG_reserved pages. PG_reserved
5398 * is set to both of a memory hole page and a _used_ kernel
5407 bool is_pageblock_removable_nolock(struct page
*page
)
5409 struct zone
*zone
= page_zone(page
);
5410 return __count_immobile_pages(zone
, page
, 0);
5413 int set_migratetype_isolate(struct page
*page
)
5416 unsigned long flags
, pfn
;
5417 struct memory_isolate_notify arg
;
5422 zone
= page_zone(page
);
5423 zone_idx
= zone_idx(zone
);
5425 spin_lock_irqsave(&zone
->lock
, flags
);
5427 pfn
= page_to_pfn(page
);
5428 arg
.start_pfn
= pfn
;
5429 arg
.nr_pages
= pageblock_nr_pages
;
5430 arg
.pages_found
= 0;
5433 * It may be possible to isolate a pageblock even if the
5434 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5435 * notifier chain is used by balloon drivers to return the
5436 * number of pages in a range that are held by the balloon
5437 * driver to shrink memory. If all the pages are accounted for
5438 * by balloons, are free, or on the LRU, isolation can continue.
5439 * Later, for example, when memory hotplug notifier runs, these
5440 * pages reported as "can be isolated" should be isolated(freed)
5441 * by the balloon driver through the memory notifier chain.
5443 notifier_ret
= memory_isolate_notify(MEM_ISOLATE_COUNT
, &arg
);
5444 notifier_ret
= notifier_to_errno(notifier_ret
);
5448 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5449 * We just check MOVABLE pages.
5451 if (__count_immobile_pages(zone
, page
, arg
.pages_found
))
5455 * immobile means "not-on-lru" paes. If immobile is larger than
5456 * removable-by-driver pages reported by notifier, we'll fail.
5461 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
5462 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
5465 spin_unlock_irqrestore(&zone
->lock
, flags
);
5471 void unset_migratetype_isolate(struct page
*page
)
5474 unsigned long flags
;
5475 zone
= page_zone(page
);
5476 spin_lock_irqsave(&zone
->lock
, flags
);
5477 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
5479 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5480 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
5482 spin_unlock_irqrestore(&zone
->lock
, flags
);
5485 #ifdef CONFIG_MEMORY_HOTREMOVE
5487 * All pages in the range must be isolated before calling this.
5490 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
5496 unsigned long flags
;
5497 /* find the first valid pfn */
5498 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
5503 zone
= page_zone(pfn_to_page(pfn
));
5504 spin_lock_irqsave(&zone
->lock
, flags
);
5506 while (pfn
< end_pfn
) {
5507 if (!pfn_valid(pfn
)) {
5511 page
= pfn_to_page(pfn
);
5512 BUG_ON(page_count(page
));
5513 BUG_ON(!PageBuddy(page
));
5514 order
= page_order(page
);
5515 #ifdef CONFIG_DEBUG_VM
5516 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
5517 pfn
, 1 << order
, end_pfn
);
5519 list_del(&page
->lru
);
5520 rmv_page_order(page
);
5521 zone
->free_area
[order
].nr_free
--;
5522 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
5524 for (i
= 0; i
< (1 << order
); i
++)
5525 SetPageReserved((page
+i
));
5526 pfn
+= (1 << order
);
5528 spin_unlock_irqrestore(&zone
->lock
, flags
);
5532 #ifdef CONFIG_MEMORY_FAILURE
5533 bool is_free_buddy_page(struct page
*page
)
5535 struct zone
*zone
= page_zone(page
);
5536 unsigned long pfn
= page_to_pfn(page
);
5537 unsigned long flags
;
5540 spin_lock_irqsave(&zone
->lock
, flags
);
5541 for (order
= 0; order
< MAX_ORDER
; order
++) {
5542 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
5544 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
5547 spin_unlock_irqrestore(&zone
->lock
, flags
);
5549 return order
< MAX_ORDER
;
5553 static struct trace_print_flags pageflag_names
[] = {
5554 {1UL << PG_locked
, "locked" },
5555 {1UL << PG_error
, "error" },
5556 {1UL << PG_referenced
, "referenced" },
5557 {1UL << PG_uptodate
, "uptodate" },
5558 {1UL << PG_dirty
, "dirty" },
5559 {1UL << PG_lru
, "lru" },
5560 {1UL << PG_active
, "active" },
5561 {1UL << PG_slab
, "slab" },
5562 {1UL << PG_owner_priv_1
, "owner_priv_1" },
5563 {1UL << PG_arch_1
, "arch_1" },
5564 {1UL << PG_reserved
, "reserved" },
5565 {1UL << PG_private
, "private" },
5566 {1UL << PG_private_2
, "private_2" },
5567 {1UL << PG_writeback
, "writeback" },
5568 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5569 {1UL << PG_head
, "head" },
5570 {1UL << PG_tail
, "tail" },
5572 {1UL << PG_compound
, "compound" },
5574 {1UL << PG_swapcache
, "swapcache" },
5575 {1UL << PG_mappedtodisk
, "mappedtodisk" },
5576 {1UL << PG_reclaim
, "reclaim" },
5577 {1UL << PG_swapbacked
, "swapbacked" },
5578 {1UL << PG_unevictable
, "unevictable" },
5580 {1UL << PG_mlocked
, "mlocked" },
5582 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5583 {1UL << PG_uncached
, "uncached" },
5585 #ifdef CONFIG_MEMORY_FAILURE
5586 {1UL << PG_hwpoison
, "hwpoison" },
5591 static void dump_page_flags(unsigned long flags
)
5593 const char *delim
= "";
5597 printk(KERN_ALERT
"page flags: %#lx(", flags
);
5599 /* remove zone id */
5600 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
5602 for (i
= 0; pageflag_names
[i
].name
&& flags
; i
++) {
5604 mask
= pageflag_names
[i
].mask
;
5605 if ((flags
& mask
) != mask
)
5609 printk("%s%s", delim
, pageflag_names
[i
].name
);
5613 /* check for left over flags */
5615 printk("%s%#lx", delim
, flags
);
5620 void dump_page(struct page
*page
)
5623 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5624 page
, atomic_read(&page
->_count
), page_mapcount(page
),
5625 page
->mapping
, page
->index
);
5626 dump_page_flags(page
->flags
);