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/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/kmemcheck.h>
27 #include <linux/module.h>
28 #include <linux/suspend.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/slab.h>
32 #include <linux/oom.h>
33 #include <linux/notifier.h>
34 #include <linux/topology.h>
35 #include <linux/sysctl.h>
36 #include <linux/cpu.h>
37 #include <linux/cpuset.h>
38 #include <linux/memory_hotplug.h>
39 #include <linux/nodemask.h>
40 #include <linux/vmalloc.h>
41 #include <linux/mempolicy.h>
42 #include <linux/stop_machine.h>
43 #include <linux/sort.h>
44 #include <linux/pfn.h>
45 #include <linux/backing-dev.h>
46 #include <linux/fault-inject.h>
47 #include <linux/page-isolation.h>
48 #include <linux/page_cgroup.h>
49 #include <linux/debugobjects.h>
50 #include <linux/kmemleak.h>
51 #include <linux/memory.h>
52 #include <trace/events/kmem.h>
53 #include <linux/ftrace_event.h>
55 #include <asm/tlbflush.h>
56 #include <asm/div64.h>
60 * Array of node states.
62 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
63 [N_POSSIBLE
] = NODE_MASK_ALL
,
64 [N_ONLINE
] = { { [0] = 1UL } },
66 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
68 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
70 [N_CPU
] = { { [0] = 1UL } },
73 EXPORT_SYMBOL(node_states
);
75 unsigned long totalram_pages __read_mostly
;
76 unsigned long totalreserve_pages __read_mostly
;
77 int percpu_pagelist_fraction
;
78 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
80 #ifdef CONFIG_PM_SLEEP
82 * The following functions are used by the suspend/hibernate code to temporarily
83 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
84 * while devices are suspended. To avoid races with the suspend/hibernate code,
85 * they should always be called with pm_mutex held (gfp_allowed_mask also should
86 * only be modified with pm_mutex held, unless the suspend/hibernate code is
87 * guaranteed not to run in parallel with that modification).
89 void set_gfp_allowed_mask(gfp_t mask
)
91 WARN_ON(!mutex_is_locked(&pm_mutex
));
92 gfp_allowed_mask
= mask
;
95 gfp_t
clear_gfp_allowed_mask(gfp_t mask
)
97 gfp_t ret
= gfp_allowed_mask
;
99 WARN_ON(!mutex_is_locked(&pm_mutex
));
100 gfp_allowed_mask
&= ~mask
;
103 #endif /* CONFIG_PM_SLEEP */
105 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
106 int pageblock_order __read_mostly
;
109 static void __free_pages_ok(struct page
*page
, unsigned int order
);
112 * results with 256, 32 in the lowmem_reserve sysctl:
113 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
114 * 1G machine -> (16M dma, 784M normal, 224M high)
115 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
116 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
117 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
119 * TBD: should special case ZONE_DMA32 machines here - in those we normally
120 * don't need any ZONE_NORMAL reservation
122 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
123 #ifdef CONFIG_ZONE_DMA
126 #ifdef CONFIG_ZONE_DMA32
129 #ifdef CONFIG_HIGHMEM
135 EXPORT_SYMBOL(totalram_pages
);
137 static char * const zone_names
[MAX_NR_ZONES
] = {
138 #ifdef CONFIG_ZONE_DMA
141 #ifdef CONFIG_ZONE_DMA32
145 #ifdef CONFIG_HIGHMEM
151 int min_free_kbytes
= 1024;
153 static unsigned long __meminitdata nr_kernel_pages
;
154 static unsigned long __meminitdata nr_all_pages
;
155 static unsigned long __meminitdata dma_reserve
;
157 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
159 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
160 * ranges of memory (RAM) that may be registered with add_active_range().
161 * Ranges passed to add_active_range() will be merged if possible
162 * so the number of times add_active_range() can be called is
163 * related to the number of nodes and the number of holes
165 #ifdef CONFIG_MAX_ACTIVE_REGIONS
166 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
167 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
169 #if MAX_NUMNODES >= 32
170 /* If there can be many nodes, allow up to 50 holes per node */
171 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
173 /* By default, allow up to 256 distinct regions */
174 #define MAX_ACTIVE_REGIONS 256
178 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
179 static int __meminitdata nr_nodemap_entries
;
180 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
181 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
182 static unsigned long __initdata required_kernelcore
;
183 static unsigned long __initdata required_movablecore
;
184 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
186 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
188 EXPORT_SYMBOL(movable_zone
);
189 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
192 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
193 int nr_online_nodes __read_mostly
= 1;
194 EXPORT_SYMBOL(nr_node_ids
);
195 EXPORT_SYMBOL(nr_online_nodes
);
198 int page_group_by_mobility_disabled __read_mostly
;
200 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
203 if (unlikely(page_group_by_mobility_disabled
))
204 migratetype
= MIGRATE_UNMOVABLE
;
206 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
207 PB_migrate
, PB_migrate_end
);
210 bool oom_killer_disabled __read_mostly
;
212 #ifdef CONFIG_DEBUG_VM
213 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
217 unsigned long pfn
= page_to_pfn(page
);
220 seq
= zone_span_seqbegin(zone
);
221 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
223 else if (pfn
< zone
->zone_start_pfn
)
225 } while (zone_span_seqretry(zone
, seq
));
230 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
232 if (!pfn_valid_within(page_to_pfn(page
)))
234 if (zone
!= page_zone(page
))
240 * Temporary debugging check for pages not lying within a given zone.
242 static int bad_range(struct zone
*zone
, struct page
*page
)
244 if (page_outside_zone_boundaries(zone
, page
))
246 if (!page_is_consistent(zone
, page
))
252 static inline int bad_range(struct zone
*zone
, struct page
*page
)
258 static void bad_page(struct page
*page
)
260 static unsigned long resume
;
261 static unsigned long nr_shown
;
262 static unsigned long nr_unshown
;
264 /* Don't complain about poisoned pages */
265 if (PageHWPoison(page
)) {
266 __ClearPageBuddy(page
);
271 * Allow a burst of 60 reports, then keep quiet for that minute;
272 * or allow a steady drip of one report per second.
274 if (nr_shown
== 60) {
275 if (time_before(jiffies
, resume
)) {
281 "BUG: Bad page state: %lu messages suppressed\n",
288 resume
= jiffies
+ 60 * HZ
;
290 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
291 current
->comm
, page_to_pfn(page
));
296 /* Leave bad fields for debug, except PageBuddy could make trouble */
297 __ClearPageBuddy(page
);
298 add_taint(TAINT_BAD_PAGE
);
302 * Higher-order pages are called "compound pages". They are structured thusly:
304 * The first PAGE_SIZE page is called the "head page".
306 * The remaining PAGE_SIZE pages are called "tail pages".
308 * All pages have PG_compound set. All pages have their ->private pointing at
309 * the head page (even the head page has this).
311 * The first tail page's ->lru.next holds the address of the compound page's
312 * put_page() function. Its ->lru.prev holds the order of allocation.
313 * This usage means that zero-order pages may not be compound.
316 static void free_compound_page(struct page
*page
)
318 __free_pages_ok(page
, compound_order(page
));
321 void prep_compound_page(struct page
*page
, unsigned long order
)
324 int nr_pages
= 1 << order
;
326 set_compound_page_dtor(page
, free_compound_page
);
327 set_compound_order(page
, order
);
329 for (i
= 1; i
< nr_pages
; i
++) {
330 struct page
*p
= page
+ i
;
333 p
->first_page
= page
;
337 static int destroy_compound_page(struct page
*page
, unsigned long order
)
340 int nr_pages
= 1 << order
;
343 if (unlikely(compound_order(page
) != order
) ||
344 unlikely(!PageHead(page
))) {
349 __ClearPageHead(page
);
351 for (i
= 1; i
< nr_pages
; i
++) {
352 struct page
*p
= page
+ i
;
354 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
364 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
369 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
370 * and __GFP_HIGHMEM from hard or soft interrupt context.
372 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
373 for (i
= 0; i
< (1 << order
); i
++)
374 clear_highpage(page
+ i
);
377 static inline void set_page_order(struct page
*page
, int order
)
379 set_page_private(page
, order
);
380 __SetPageBuddy(page
);
383 static inline void rmv_page_order(struct page
*page
)
385 __ClearPageBuddy(page
);
386 set_page_private(page
, 0);
390 * Locate the struct page for both the matching buddy in our
391 * pair (buddy1) and the combined O(n+1) page they form (page).
393 * 1) Any buddy B1 will have an order O twin B2 which satisfies
394 * the following equation:
396 * For example, if the starting buddy (buddy2) is #8 its order
398 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
400 * 2) Any buddy B will have an order O+1 parent P which
401 * satisfies the following equation:
404 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
406 static inline struct page
*
407 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
409 unsigned long buddy_idx
= page_idx
^ (1 << order
);
411 return page
+ (buddy_idx
- page_idx
);
414 static inline unsigned long
415 __find_combined_index(unsigned long page_idx
, unsigned int order
)
417 return (page_idx
& ~(1 << order
));
421 * This function checks whether a page is free && is the buddy
422 * we can do coalesce a page and its buddy if
423 * (a) the buddy is not in a hole &&
424 * (b) the buddy is in the buddy system &&
425 * (c) a page and its buddy have the same order &&
426 * (d) a page and its buddy are in the same zone.
428 * For recording whether a page is in the buddy system, we use PG_buddy.
429 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
431 * For recording page's order, we use page_private(page).
433 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
436 if (!pfn_valid_within(page_to_pfn(buddy
)))
439 if (page_zone_id(page
) != page_zone_id(buddy
))
442 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
443 VM_BUG_ON(page_count(buddy
) != 0);
450 * Freeing function for a buddy system allocator.
452 * The concept of a buddy system is to maintain direct-mapped table
453 * (containing bit values) for memory blocks of various "orders".
454 * The bottom level table contains the map for the smallest allocatable
455 * units of memory (here, pages), and each level above it describes
456 * pairs of units from the levels below, hence, "buddies".
457 * At a high level, all that happens here is marking the table entry
458 * at the bottom level available, and propagating the changes upward
459 * as necessary, plus some accounting needed to play nicely with other
460 * parts of the VM system.
461 * At each level, we keep a list of pages, which are heads of continuous
462 * free pages of length of (1 << order) and marked with PG_buddy. Page's
463 * order is recorded in page_private(page) field.
464 * So when we are allocating or freeing one, we can derive the state of the
465 * other. That is, if we allocate a small block, and both were
466 * free, the remainder of the region must be split into blocks.
467 * If a block is freed, and its buddy is also free, then this
468 * triggers coalescing into a block of larger size.
473 static inline void __free_one_page(struct page
*page
,
474 struct zone
*zone
, unsigned int order
,
477 unsigned long page_idx
;
478 unsigned long combined_idx
;
481 if (unlikely(PageCompound(page
)))
482 if (unlikely(destroy_compound_page(page
, order
)))
485 VM_BUG_ON(migratetype
== -1);
487 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
489 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
490 VM_BUG_ON(bad_range(zone
, page
));
492 while (order
< MAX_ORDER
-1) {
493 buddy
= __page_find_buddy(page
, page_idx
, order
);
494 if (!page_is_buddy(page
, buddy
, order
))
497 /* Our buddy is free, merge with it and move up one order. */
498 list_del(&buddy
->lru
);
499 zone
->free_area
[order
].nr_free
--;
500 rmv_page_order(buddy
);
501 combined_idx
= __find_combined_index(page_idx
, order
);
502 page
= page
+ (combined_idx
- page_idx
);
503 page_idx
= combined_idx
;
506 set_page_order(page
, order
);
509 * If this is not the largest possible page, check if the buddy
510 * of the next-highest order is free. If it is, it's possible
511 * that pages are being freed that will coalesce soon. In case,
512 * that is happening, add the free page to the tail of the list
513 * so it's less likely to be used soon and more likely to be merged
514 * as a higher order page
516 if ((order
< MAX_ORDER
-1) && pfn_valid_within(page_to_pfn(buddy
))) {
517 struct page
*higher_page
, *higher_buddy
;
518 combined_idx
= __find_combined_index(page_idx
, order
);
519 higher_page
= page
+ combined_idx
- page_idx
;
520 higher_buddy
= __page_find_buddy(higher_page
, combined_idx
, order
+ 1);
521 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
522 list_add_tail(&page
->lru
,
523 &zone
->free_area
[order
].free_list
[migratetype
]);
528 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
530 zone
->free_area
[order
].nr_free
++;
534 * free_page_mlock() -- clean up attempts to free and mlocked() page.
535 * Page should not be on lru, so no need to fix that up.
536 * free_pages_check() will verify...
538 static inline void free_page_mlock(struct page
*page
)
540 __dec_zone_page_state(page
, NR_MLOCK
);
541 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
544 static inline int free_pages_check(struct page
*page
)
546 if (unlikely(page_mapcount(page
) |
547 (page
->mapping
!= NULL
) |
548 (atomic_read(&page
->_count
) != 0) |
549 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
553 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
554 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
559 * Frees a number of pages from the PCP lists
560 * Assumes all pages on list are in same zone, and of same order.
561 * count is the number of pages to free.
563 * If the zone was previously in an "all pages pinned" state then look to
564 * see if this freeing clears that state.
566 * And clear the zone's pages_scanned counter, to hold off the "all pages are
567 * pinned" detection logic.
569 static void free_pcppages_bulk(struct zone
*zone
, int count
,
570 struct per_cpu_pages
*pcp
)
575 spin_lock(&zone
->lock
);
576 zone
->all_unreclaimable
= 0;
577 zone
->pages_scanned
= 0;
579 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
);
582 struct list_head
*list
;
585 * Remove pages from lists in a round-robin fashion. A
586 * batch_free count is maintained that is incremented when an
587 * empty list is encountered. This is so more pages are freed
588 * off fuller lists instead of spinning excessively around empty
593 if (++migratetype
== MIGRATE_PCPTYPES
)
595 list
= &pcp
->lists
[migratetype
];
596 } while (list_empty(list
));
599 page
= list_entry(list
->prev
, struct page
, lru
);
600 /* must delete as __free_one_page list manipulates */
601 list_del(&page
->lru
);
602 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
603 __free_one_page(page
, zone
, 0, page_private(page
));
604 trace_mm_page_pcpu_drain(page
, 0, page_private(page
));
605 } while (--count
&& --batch_free
&& !list_empty(list
));
607 spin_unlock(&zone
->lock
);
610 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
613 spin_lock(&zone
->lock
);
614 zone
->all_unreclaimable
= 0;
615 zone
->pages_scanned
= 0;
617 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
618 __free_one_page(page
, zone
, order
, migratetype
);
619 spin_unlock(&zone
->lock
);
622 static void __free_pages_ok(struct page
*page
, unsigned int order
)
627 int wasMlocked
= __TestClearPageMlocked(page
);
629 trace_mm_page_free_direct(page
, order
);
630 kmemcheck_free_shadow(page
, order
);
632 for (i
= 0 ; i
< (1 << order
) ; ++i
)
633 bad
+= free_pages_check(page
+ i
);
637 if (!PageHighMem(page
)) {
638 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
639 debug_check_no_obj_freed(page_address(page
),
642 arch_free_page(page
, order
);
643 kernel_map_pages(page
, 1 << order
, 0);
645 local_irq_save(flags
);
646 if (unlikely(wasMlocked
))
647 free_page_mlock(page
);
648 __count_vm_events(PGFREE
, 1 << order
);
649 free_one_page(page_zone(page
), page
, order
,
650 get_pageblock_migratetype(page
));
651 local_irq_restore(flags
);
655 * permit the bootmem allocator to evade page validation on high-order frees
657 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
660 __ClearPageReserved(page
);
661 set_page_count(page
, 0);
662 set_page_refcounted(page
);
668 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
669 struct page
*p
= &page
[loop
];
671 if (loop
+ 1 < BITS_PER_LONG
)
673 __ClearPageReserved(p
);
674 set_page_count(p
, 0);
677 set_page_refcounted(page
);
678 __free_pages(page
, order
);
684 * The order of subdivision here is critical for the IO subsystem.
685 * Please do not alter this order without good reasons and regression
686 * testing. Specifically, as large blocks of memory are subdivided,
687 * the order in which smaller blocks are delivered depends on the order
688 * they're subdivided in this function. This is the primary factor
689 * influencing the order in which pages are delivered to the IO
690 * subsystem according to empirical testing, and this is also justified
691 * by considering the behavior of a buddy system containing a single
692 * large block of memory acted on by a series of small allocations.
693 * This behavior is a critical factor in sglist merging's success.
697 static inline void expand(struct zone
*zone
, struct page
*page
,
698 int low
, int high
, struct free_area
*area
,
701 unsigned long size
= 1 << high
;
707 VM_BUG_ON(bad_range(zone
, &page
[size
]));
708 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
710 set_page_order(&page
[size
], high
);
715 * This page is about to be returned from the page allocator
717 static inline int check_new_page(struct page
*page
)
719 if (unlikely(page_mapcount(page
) |
720 (page
->mapping
!= NULL
) |
721 (atomic_read(&page
->_count
) != 0) |
722 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
729 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
733 for (i
= 0; i
< (1 << order
); i
++) {
734 struct page
*p
= page
+ i
;
735 if (unlikely(check_new_page(p
)))
739 set_page_private(page
, 0);
740 set_page_refcounted(page
);
742 arch_alloc_page(page
, order
);
743 kernel_map_pages(page
, 1 << order
, 1);
745 if (gfp_flags
& __GFP_ZERO
)
746 prep_zero_page(page
, order
, gfp_flags
);
748 if (order
&& (gfp_flags
& __GFP_COMP
))
749 prep_compound_page(page
, order
);
755 * Go through the free lists for the given migratetype and remove
756 * the smallest available page from the freelists
759 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
762 unsigned int current_order
;
763 struct free_area
* area
;
766 /* Find a page of the appropriate size in the preferred list */
767 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
768 area
= &(zone
->free_area
[current_order
]);
769 if (list_empty(&area
->free_list
[migratetype
]))
772 page
= list_entry(area
->free_list
[migratetype
].next
,
774 list_del(&page
->lru
);
775 rmv_page_order(page
);
777 expand(zone
, page
, order
, current_order
, area
, migratetype
);
786 * This array describes the order lists are fallen back to when
787 * the free lists for the desirable migrate type are depleted
789 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
790 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
791 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
792 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
793 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
797 * Move the free pages in a range to the free lists of the requested type.
798 * Note that start_page and end_pages are not aligned on a pageblock
799 * boundary. If alignment is required, use move_freepages_block()
801 static int move_freepages(struct zone
*zone
,
802 struct page
*start_page
, struct page
*end_page
,
809 #ifndef CONFIG_HOLES_IN_ZONE
811 * page_zone is not safe to call in this context when
812 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
813 * anyway as we check zone boundaries in move_freepages_block().
814 * Remove at a later date when no bug reports exist related to
815 * grouping pages by mobility
817 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
820 for (page
= start_page
; page
<= end_page
;) {
821 /* Make sure we are not inadvertently changing nodes */
822 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
824 if (!pfn_valid_within(page_to_pfn(page
))) {
829 if (!PageBuddy(page
)) {
834 order
= page_order(page
);
835 list_del(&page
->lru
);
837 &zone
->free_area
[order
].free_list
[migratetype
]);
839 pages_moved
+= 1 << order
;
845 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
848 unsigned long start_pfn
, end_pfn
;
849 struct page
*start_page
, *end_page
;
851 start_pfn
= page_to_pfn(page
);
852 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
853 start_page
= pfn_to_page(start_pfn
);
854 end_page
= start_page
+ pageblock_nr_pages
- 1;
855 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
857 /* Do not cross zone boundaries */
858 if (start_pfn
< zone
->zone_start_pfn
)
860 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
863 return move_freepages(zone
, start_page
, end_page
, migratetype
);
866 static void change_pageblock_range(struct page
*pageblock_page
,
867 int start_order
, int migratetype
)
869 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
871 while (nr_pageblocks
--) {
872 set_pageblock_migratetype(pageblock_page
, migratetype
);
873 pageblock_page
+= pageblock_nr_pages
;
877 /* Remove an element from the buddy allocator from the fallback list */
878 static inline struct page
*
879 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
881 struct free_area
* area
;
886 /* Find the largest possible block of pages in the other list */
887 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
889 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
890 migratetype
= fallbacks
[start_migratetype
][i
];
892 /* MIGRATE_RESERVE handled later if necessary */
893 if (migratetype
== MIGRATE_RESERVE
)
896 area
= &(zone
->free_area
[current_order
]);
897 if (list_empty(&area
->free_list
[migratetype
]))
900 page
= list_entry(area
->free_list
[migratetype
].next
,
905 * If breaking a large block of pages, move all free
906 * pages to the preferred allocation list. If falling
907 * back for a reclaimable kernel allocation, be more
908 * agressive about taking ownership of free pages
910 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
911 start_migratetype
== MIGRATE_RECLAIMABLE
||
912 page_group_by_mobility_disabled
) {
914 pages
= move_freepages_block(zone
, page
,
917 /* Claim the whole block if over half of it is free */
918 if (pages
>= (1 << (pageblock_order
-1)) ||
919 page_group_by_mobility_disabled
)
920 set_pageblock_migratetype(page
,
923 migratetype
= start_migratetype
;
926 /* Remove the page from the freelists */
927 list_del(&page
->lru
);
928 rmv_page_order(page
);
930 /* Take ownership for orders >= pageblock_order */
931 if (current_order
>= pageblock_order
)
932 change_pageblock_range(page
, current_order
,
935 expand(zone
, page
, order
, current_order
, area
, migratetype
);
937 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
938 start_migratetype
, migratetype
);
948 * Do the hard work of removing an element from the buddy allocator.
949 * Call me with the zone->lock already held.
951 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
957 page
= __rmqueue_smallest(zone
, order
, migratetype
);
959 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
960 page
= __rmqueue_fallback(zone
, order
, migratetype
);
963 * Use MIGRATE_RESERVE rather than fail an allocation. goto
964 * is used because __rmqueue_smallest is an inline function
965 * and we want just one call site
968 migratetype
= MIGRATE_RESERVE
;
973 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
978 * Obtain a specified number of elements from the buddy allocator, all under
979 * a single hold of the lock, for efficiency. Add them to the supplied list.
980 * Returns the number of new pages which were placed at *list.
982 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
983 unsigned long count
, struct list_head
*list
,
984 int migratetype
, int cold
)
988 spin_lock(&zone
->lock
);
989 for (i
= 0; i
< count
; ++i
) {
990 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
991 if (unlikely(page
== NULL
))
995 * Split buddy pages returned by expand() are received here
996 * in physical page order. The page is added to the callers and
997 * list and the list head then moves forward. From the callers
998 * perspective, the linked list is ordered by page number in
999 * some conditions. This is useful for IO devices that can
1000 * merge IO requests if the physical pages are ordered
1003 if (likely(cold
== 0))
1004 list_add(&page
->lru
, list
);
1006 list_add_tail(&page
->lru
, list
);
1007 set_page_private(page
, migratetype
);
1010 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1011 spin_unlock(&zone
->lock
);
1017 * Called from the vmstat counter updater to drain pagesets of this
1018 * currently executing processor on remote nodes after they have
1021 * Note that this function must be called with the thread pinned to
1022 * a single processor.
1024 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1026 unsigned long flags
;
1029 local_irq_save(flags
);
1030 if (pcp
->count
>= pcp
->batch
)
1031 to_drain
= pcp
->batch
;
1033 to_drain
= pcp
->count
;
1034 free_pcppages_bulk(zone
, to_drain
, pcp
);
1035 pcp
->count
-= to_drain
;
1036 local_irq_restore(flags
);
1041 * Drain pages of the indicated processor.
1043 * The processor must either be the current processor and the
1044 * thread pinned to the current processor or a processor that
1047 static void drain_pages(unsigned int cpu
)
1049 unsigned long flags
;
1052 for_each_populated_zone(zone
) {
1053 struct per_cpu_pageset
*pset
;
1054 struct per_cpu_pages
*pcp
;
1056 local_irq_save(flags
);
1057 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1060 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1062 local_irq_restore(flags
);
1067 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1069 void drain_local_pages(void *arg
)
1071 drain_pages(smp_processor_id());
1075 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1077 void drain_all_pages(void)
1079 on_each_cpu(drain_local_pages
, NULL
, 1);
1082 #ifdef CONFIG_HIBERNATION
1084 void mark_free_pages(struct zone
*zone
)
1086 unsigned long pfn
, max_zone_pfn
;
1087 unsigned long flags
;
1089 struct list_head
*curr
;
1091 if (!zone
->spanned_pages
)
1094 spin_lock_irqsave(&zone
->lock
, flags
);
1096 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1097 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1098 if (pfn_valid(pfn
)) {
1099 struct page
*page
= pfn_to_page(pfn
);
1101 if (!swsusp_page_is_forbidden(page
))
1102 swsusp_unset_page_free(page
);
1105 for_each_migratetype_order(order
, t
) {
1106 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1109 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1110 for (i
= 0; i
< (1UL << order
); i
++)
1111 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1114 spin_unlock_irqrestore(&zone
->lock
, flags
);
1116 #endif /* CONFIG_PM */
1119 * Free a 0-order page
1120 * cold == 1 ? free a cold page : free a hot page
1122 void free_hot_cold_page(struct page
*page
, int cold
)
1124 struct zone
*zone
= page_zone(page
);
1125 struct per_cpu_pages
*pcp
;
1126 unsigned long flags
;
1128 int wasMlocked
= __TestClearPageMlocked(page
);
1130 trace_mm_page_free_direct(page
, 0);
1131 kmemcheck_free_shadow(page
, 0);
1134 page
->mapping
= NULL
;
1135 if (free_pages_check(page
))
1138 if (!PageHighMem(page
)) {
1139 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1140 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1142 arch_free_page(page
, 0);
1143 kernel_map_pages(page
, 1, 0);
1145 migratetype
= get_pageblock_migratetype(page
);
1146 set_page_private(page
, migratetype
);
1147 local_irq_save(flags
);
1148 if (unlikely(wasMlocked
))
1149 free_page_mlock(page
);
1150 __count_vm_event(PGFREE
);
1153 * We only track unmovable, reclaimable and movable on pcp lists.
1154 * Free ISOLATE pages back to the allocator because they are being
1155 * offlined but treat RESERVE as movable pages so we can get those
1156 * areas back if necessary. Otherwise, we may have to free
1157 * excessively into the page allocator
1159 if (migratetype
>= MIGRATE_PCPTYPES
) {
1160 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1161 free_one_page(zone
, page
, 0, migratetype
);
1164 migratetype
= MIGRATE_MOVABLE
;
1167 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1169 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1171 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1173 if (pcp
->count
>= pcp
->high
) {
1174 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1175 pcp
->count
-= pcp
->batch
;
1179 local_irq_restore(flags
);
1183 * split_page takes a non-compound higher-order page, and splits it into
1184 * n (1<<order) sub-pages: page[0..n]
1185 * Each sub-page must be freed individually.
1187 * Note: this is probably too low level an operation for use in drivers.
1188 * Please consult with lkml before using this in your driver.
1190 void split_page(struct page
*page
, unsigned int order
)
1194 VM_BUG_ON(PageCompound(page
));
1195 VM_BUG_ON(!page_count(page
));
1197 #ifdef CONFIG_KMEMCHECK
1199 * Split shadow pages too, because free(page[0]) would
1200 * otherwise free the whole shadow.
1202 if (kmemcheck_page_is_tracked(page
))
1203 split_page(virt_to_page(page
[0].shadow
), order
);
1206 for (i
= 1; i
< (1 << order
); i
++)
1207 set_page_refcounted(page
+ i
);
1211 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1212 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1216 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1217 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1220 unsigned long flags
;
1222 int cold
= !!(gfp_flags
& __GFP_COLD
);
1225 if (likely(order
== 0)) {
1226 struct per_cpu_pages
*pcp
;
1227 struct list_head
*list
;
1229 local_irq_save(flags
);
1230 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1231 list
= &pcp
->lists
[migratetype
];
1232 if (list_empty(list
)) {
1233 pcp
->count
+= rmqueue_bulk(zone
, 0,
1236 if (unlikely(list_empty(list
)))
1241 page
= list_entry(list
->prev
, struct page
, lru
);
1243 page
= list_entry(list
->next
, struct page
, lru
);
1245 list_del(&page
->lru
);
1248 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1250 * __GFP_NOFAIL is not to be used in new code.
1252 * All __GFP_NOFAIL callers should be fixed so that they
1253 * properly detect and handle allocation failures.
1255 * We most definitely don't want callers attempting to
1256 * allocate greater than order-1 page units with
1259 WARN_ON_ONCE(order
> 1);
1261 spin_lock_irqsave(&zone
->lock
, flags
);
1262 page
= __rmqueue(zone
, order
, migratetype
);
1263 spin_unlock(&zone
->lock
);
1266 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << order
));
1269 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1270 zone_statistics(preferred_zone
, zone
);
1271 local_irq_restore(flags
);
1273 VM_BUG_ON(bad_range(zone
, page
));
1274 if (prep_new_page(page
, order
, gfp_flags
))
1279 local_irq_restore(flags
);
1283 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1284 #define ALLOC_WMARK_MIN WMARK_MIN
1285 #define ALLOC_WMARK_LOW WMARK_LOW
1286 #define ALLOC_WMARK_HIGH WMARK_HIGH
1287 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1289 /* Mask to get the watermark bits */
1290 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1292 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1293 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1294 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1296 #ifdef CONFIG_FAIL_PAGE_ALLOC
1298 static struct fail_page_alloc_attr
{
1299 struct fault_attr attr
;
1301 u32 ignore_gfp_highmem
;
1302 u32 ignore_gfp_wait
;
1305 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1307 struct dentry
*ignore_gfp_highmem_file
;
1308 struct dentry
*ignore_gfp_wait_file
;
1309 struct dentry
*min_order_file
;
1311 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1313 } fail_page_alloc
= {
1314 .attr
= FAULT_ATTR_INITIALIZER
,
1315 .ignore_gfp_wait
= 1,
1316 .ignore_gfp_highmem
= 1,
1320 static int __init
setup_fail_page_alloc(char *str
)
1322 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1324 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1326 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1328 if (order
< fail_page_alloc
.min_order
)
1330 if (gfp_mask
& __GFP_NOFAIL
)
1332 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1334 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1337 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1340 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1342 static int __init
fail_page_alloc_debugfs(void)
1344 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1348 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1352 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1354 fail_page_alloc
.ignore_gfp_wait_file
=
1355 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1356 &fail_page_alloc
.ignore_gfp_wait
);
1358 fail_page_alloc
.ignore_gfp_highmem_file
=
1359 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1360 &fail_page_alloc
.ignore_gfp_highmem
);
1361 fail_page_alloc
.min_order_file
=
1362 debugfs_create_u32("min-order", mode
, dir
,
1363 &fail_page_alloc
.min_order
);
1365 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1366 !fail_page_alloc
.ignore_gfp_highmem_file
||
1367 !fail_page_alloc
.min_order_file
) {
1369 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1370 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1371 debugfs_remove(fail_page_alloc
.min_order_file
);
1372 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1378 late_initcall(fail_page_alloc_debugfs
);
1380 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1382 #else /* CONFIG_FAIL_PAGE_ALLOC */
1384 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1389 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1392 * Return 1 if free pages are above 'mark'. This takes into account the order
1393 * of the allocation.
1395 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1396 int classzone_idx
, int alloc_flags
)
1398 /* free_pages my go negative - that's OK */
1400 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1403 if (alloc_flags
& ALLOC_HIGH
)
1405 if (alloc_flags
& ALLOC_HARDER
)
1408 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1410 for (o
= 0; o
< order
; o
++) {
1411 /* At the next order, this order's pages become unavailable */
1412 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1414 /* Require fewer higher order pages to be free */
1417 if (free_pages
<= min
)
1425 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1426 * skip over zones that are not allowed by the cpuset, or that have
1427 * been recently (in last second) found to be nearly full. See further
1428 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1429 * that have to skip over a lot of full or unallowed zones.
1431 * If the zonelist cache is present in the passed in zonelist, then
1432 * returns a pointer to the allowed node mask (either the current
1433 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1435 * If the zonelist cache is not available for this zonelist, does
1436 * nothing and returns NULL.
1438 * If the fullzones BITMAP in the zonelist cache is stale (more than
1439 * a second since last zap'd) then we zap it out (clear its bits.)
1441 * We hold off even calling zlc_setup, until after we've checked the
1442 * first zone in the zonelist, on the theory that most allocations will
1443 * be satisfied from that first zone, so best to examine that zone as
1444 * quickly as we can.
1446 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1448 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1449 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1451 zlc
= zonelist
->zlcache_ptr
;
1455 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1456 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1457 zlc
->last_full_zap
= jiffies
;
1460 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1461 &cpuset_current_mems_allowed
:
1462 &node_states
[N_HIGH_MEMORY
];
1463 return allowednodes
;
1467 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1468 * if it is worth looking at further for free memory:
1469 * 1) Check that the zone isn't thought to be full (doesn't have its
1470 * bit set in the zonelist_cache fullzones BITMAP).
1471 * 2) Check that the zones node (obtained from the zonelist_cache
1472 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1473 * Return true (non-zero) if zone is worth looking at further, or
1474 * else return false (zero) if it is not.
1476 * This check -ignores- the distinction between various watermarks,
1477 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1478 * found to be full for any variation of these watermarks, it will
1479 * be considered full for up to one second by all requests, unless
1480 * we are so low on memory on all allowed nodes that we are forced
1481 * into the second scan of the zonelist.
1483 * In the second scan we ignore this zonelist cache and exactly
1484 * apply the watermarks to all zones, even it is slower to do so.
1485 * We are low on memory in the second scan, and should leave no stone
1486 * unturned looking for a free page.
1488 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1489 nodemask_t
*allowednodes
)
1491 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1492 int i
; /* index of *z in zonelist zones */
1493 int n
; /* node that zone *z is on */
1495 zlc
= zonelist
->zlcache_ptr
;
1499 i
= z
- zonelist
->_zonerefs
;
1502 /* This zone is worth trying if it is allowed but not full */
1503 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1507 * Given 'z' scanning a zonelist, set the corresponding bit in
1508 * zlc->fullzones, so that subsequent attempts to allocate a page
1509 * from that zone don't waste time re-examining it.
1511 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1513 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1514 int i
; /* index of *z in zonelist zones */
1516 zlc
= zonelist
->zlcache_ptr
;
1520 i
= z
- zonelist
->_zonerefs
;
1522 set_bit(i
, zlc
->fullzones
);
1525 #else /* CONFIG_NUMA */
1527 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1532 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1533 nodemask_t
*allowednodes
)
1538 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1541 #endif /* CONFIG_NUMA */
1544 * get_page_from_freelist goes through the zonelist trying to allocate
1547 static struct page
*
1548 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1549 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1550 struct zone
*preferred_zone
, int migratetype
)
1553 struct page
*page
= NULL
;
1556 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1557 int zlc_active
= 0; /* set if using zonelist_cache */
1558 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1560 classzone_idx
= zone_idx(preferred_zone
);
1563 * Scan zonelist, looking for a zone with enough free.
1564 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1566 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1567 high_zoneidx
, nodemask
) {
1568 if (NUMA_BUILD
&& zlc_active
&&
1569 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1571 if ((alloc_flags
& ALLOC_CPUSET
) &&
1572 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1575 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1576 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1580 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1581 if (zone_watermark_ok(zone
, order
, mark
,
1582 classzone_idx
, alloc_flags
))
1585 if (zone_reclaim_mode
== 0)
1586 goto this_zone_full
;
1588 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1590 case ZONE_RECLAIM_NOSCAN
:
1593 case ZONE_RECLAIM_FULL
:
1594 /* scanned but unreclaimable */
1595 goto this_zone_full
;
1597 /* did we reclaim enough */
1598 if (!zone_watermark_ok(zone
, order
, mark
,
1599 classzone_idx
, alloc_flags
))
1600 goto this_zone_full
;
1605 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1606 gfp_mask
, migratetype
);
1611 zlc_mark_zone_full(zonelist
, z
);
1613 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1615 * we do zlc_setup after the first zone is tried but only
1616 * if there are multiple nodes make it worthwhile
1618 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1624 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1625 /* Disable zlc cache for second zonelist scan */
1633 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1634 unsigned long pages_reclaimed
)
1636 /* Do not loop if specifically requested */
1637 if (gfp_mask
& __GFP_NORETRY
)
1641 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1642 * means __GFP_NOFAIL, but that may not be true in other
1645 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1649 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1650 * specified, then we retry until we no longer reclaim any pages
1651 * (above), or we've reclaimed an order of pages at least as
1652 * large as the allocation's order. In both cases, if the
1653 * allocation still fails, we stop retrying.
1655 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1659 * Don't let big-order allocations loop unless the caller
1660 * explicitly requests that.
1662 if (gfp_mask
& __GFP_NOFAIL
)
1668 static inline struct page
*
1669 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1670 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1671 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1676 /* Acquire the OOM killer lock for the zones in zonelist */
1677 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1678 schedule_timeout_uninterruptible(1);
1683 * Go through the zonelist yet one more time, keep very high watermark
1684 * here, this is only to catch a parallel oom killing, we must fail if
1685 * we're still under heavy pressure.
1687 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1688 order
, zonelist
, high_zoneidx
,
1689 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1690 preferred_zone
, migratetype
);
1694 if (!(gfp_mask
& __GFP_NOFAIL
)) {
1695 /* The OOM killer will not help higher order allocs */
1696 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1699 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1700 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1701 * The caller should handle page allocation failure by itself if
1702 * it specifies __GFP_THISNODE.
1703 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1705 if (gfp_mask
& __GFP_THISNODE
)
1708 /* Exhausted what can be done so it's blamo time */
1709 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
);
1712 clear_zonelist_oom(zonelist
, gfp_mask
);
1716 /* The really slow allocator path where we enter direct reclaim */
1717 static inline struct page
*
1718 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1719 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1720 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1721 int migratetype
, unsigned long *did_some_progress
)
1723 struct page
*page
= NULL
;
1724 struct reclaim_state reclaim_state
;
1725 struct task_struct
*p
= current
;
1729 /* We now go into synchronous reclaim */
1730 cpuset_memory_pressure_bump();
1731 p
->flags
|= PF_MEMALLOC
;
1732 lockdep_set_current_reclaim_state(gfp_mask
);
1733 reclaim_state
.reclaimed_slab
= 0;
1734 p
->reclaim_state
= &reclaim_state
;
1736 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1738 p
->reclaim_state
= NULL
;
1739 lockdep_clear_current_reclaim_state();
1740 p
->flags
&= ~PF_MEMALLOC
;
1747 if (likely(*did_some_progress
))
1748 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1749 zonelist
, high_zoneidx
,
1750 alloc_flags
, preferred_zone
,
1756 * This is called in the allocator slow-path if the allocation request is of
1757 * sufficient urgency to ignore watermarks and take other desperate measures
1759 static inline struct page
*
1760 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1761 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1762 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1768 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1769 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1770 preferred_zone
, migratetype
);
1772 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1773 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1774 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1780 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1781 enum zone_type high_zoneidx
)
1786 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1787 wakeup_kswapd(zone
, order
);
1791 gfp_to_alloc_flags(gfp_t gfp_mask
)
1793 struct task_struct
*p
= current
;
1794 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
1795 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1797 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1798 BUILD_BUG_ON(__GFP_HIGH
!= ALLOC_HIGH
);
1801 * The caller may dip into page reserves a bit more if the caller
1802 * cannot run direct reclaim, or if the caller has realtime scheduling
1803 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1804 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1806 alloc_flags
|= (gfp_mask
& __GFP_HIGH
);
1809 alloc_flags
|= ALLOC_HARDER
;
1811 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1812 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1814 alloc_flags
&= ~ALLOC_CPUSET
;
1815 } else if (unlikely(rt_task(p
)) && !in_interrupt())
1816 alloc_flags
|= ALLOC_HARDER
;
1818 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
1819 if (!in_interrupt() &&
1820 ((p
->flags
& PF_MEMALLOC
) ||
1821 unlikely(test_thread_flag(TIF_MEMDIE
))))
1822 alloc_flags
|= ALLOC_NO_WATERMARKS
;
1828 static inline struct page
*
1829 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
1830 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1831 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1834 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1835 struct page
*page
= NULL
;
1837 unsigned long pages_reclaimed
= 0;
1838 unsigned long did_some_progress
;
1839 struct task_struct
*p
= current
;
1842 * In the slowpath, we sanity check order to avoid ever trying to
1843 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1844 * be using allocators in order of preference for an area that is
1847 if (order
>= MAX_ORDER
) {
1848 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
1853 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1854 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1855 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1856 * using a larger set of nodes after it has established that the
1857 * allowed per node queues are empty and that nodes are
1860 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1864 wake_all_kswapd(order
, zonelist
, high_zoneidx
);
1867 * OK, we're below the kswapd watermark and have kicked background
1868 * reclaim. Now things get more complex, so set up alloc_flags according
1869 * to how we want to proceed.
1871 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
1873 /* This is the last chance, in general, before the goto nopage. */
1874 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1875 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
1876 preferred_zone
, migratetype
);
1881 /* Allocate without watermarks if the context allows */
1882 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
1883 page
= __alloc_pages_high_priority(gfp_mask
, order
,
1884 zonelist
, high_zoneidx
, nodemask
,
1885 preferred_zone
, migratetype
);
1890 /* Atomic allocations - we can't balance anything */
1894 /* Avoid recursion of direct reclaim */
1895 if (p
->flags
& PF_MEMALLOC
)
1898 /* Avoid allocations with no watermarks from looping endlessly */
1899 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
1902 /* Try direct reclaim and then allocating */
1903 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
1904 zonelist
, high_zoneidx
,
1906 alloc_flags
, preferred_zone
,
1907 migratetype
, &did_some_progress
);
1912 * If we failed to make any progress reclaiming, then we are
1913 * running out of options and have to consider going OOM
1915 if (!did_some_progress
) {
1916 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1917 if (oom_killer_disabled
)
1919 page
= __alloc_pages_may_oom(gfp_mask
, order
,
1920 zonelist
, high_zoneidx
,
1921 nodemask
, preferred_zone
,
1927 * The OOM killer does not trigger for high-order
1928 * ~__GFP_NOFAIL allocations so if no progress is being
1929 * made, there are no other options and retrying is
1932 if (order
> PAGE_ALLOC_COSTLY_ORDER
&&
1933 !(gfp_mask
& __GFP_NOFAIL
))
1940 /* Check if we should retry the allocation */
1941 pages_reclaimed
+= did_some_progress
;
1942 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
1943 /* Wait for some write requests to complete then retry */
1944 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1949 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1950 printk(KERN_WARNING
"%s: page allocation failure."
1951 " order:%d, mode:0x%x\n",
1952 p
->comm
, order
, gfp_mask
);
1958 if (kmemcheck_enabled
)
1959 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
1965 * This is the 'heart' of the zoned buddy allocator.
1968 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
1969 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1971 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1972 struct zone
*preferred_zone
;
1974 int migratetype
= allocflags_to_migratetype(gfp_mask
);
1976 gfp_mask
&= gfp_allowed_mask
;
1978 lockdep_trace_alloc(gfp_mask
);
1980 might_sleep_if(gfp_mask
& __GFP_WAIT
);
1982 if (should_fail_alloc_page(gfp_mask
, order
))
1986 * Check the zones suitable for the gfp_mask contain at least one
1987 * valid zone. It's possible to have an empty zonelist as a result
1988 * of GFP_THISNODE and a memoryless node
1990 if (unlikely(!zonelist
->_zonerefs
->zone
))
1994 /* The preferred zone is used for statistics later */
1995 first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
, &preferred_zone
);
1996 if (!preferred_zone
) {
2001 /* First allocation attempt */
2002 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2003 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
2004 preferred_zone
, migratetype
);
2005 if (unlikely(!page
))
2006 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2007 zonelist
, high_zoneidx
, nodemask
,
2008 preferred_zone
, migratetype
);
2011 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2014 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2017 * Common helper functions.
2019 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2024 * __get_free_pages() returns a 32-bit address, which cannot represent
2027 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2029 page
= alloc_pages(gfp_mask
, order
);
2032 return (unsigned long) page_address(page
);
2034 EXPORT_SYMBOL(__get_free_pages
);
2036 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2038 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2040 EXPORT_SYMBOL(get_zeroed_page
);
2042 void __pagevec_free(struct pagevec
*pvec
)
2044 int i
= pagevec_count(pvec
);
2047 trace_mm_pagevec_free(pvec
->pages
[i
], pvec
->cold
);
2048 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
2052 void __free_pages(struct page
*page
, unsigned int order
)
2054 if (put_page_testzero(page
)) {
2056 free_hot_cold_page(page
, 0);
2058 __free_pages_ok(page
, order
);
2062 EXPORT_SYMBOL(__free_pages
);
2064 void free_pages(unsigned long addr
, unsigned int order
)
2067 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2068 __free_pages(virt_to_page((void *)addr
), order
);
2072 EXPORT_SYMBOL(free_pages
);
2075 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2076 * @size: the number of bytes to allocate
2077 * @gfp_mask: GFP flags for the allocation
2079 * This function is similar to alloc_pages(), except that it allocates the
2080 * minimum number of pages to satisfy the request. alloc_pages() can only
2081 * allocate memory in power-of-two pages.
2083 * This function is also limited by MAX_ORDER.
2085 * Memory allocated by this function must be released by free_pages_exact().
2087 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2089 unsigned int order
= get_order(size
);
2092 addr
= __get_free_pages(gfp_mask
, order
);
2094 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2095 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2097 split_page(virt_to_page((void *)addr
), order
);
2098 while (used
< alloc_end
) {
2104 return (void *)addr
;
2106 EXPORT_SYMBOL(alloc_pages_exact
);
2109 * free_pages_exact - release memory allocated via alloc_pages_exact()
2110 * @virt: the value returned by alloc_pages_exact.
2111 * @size: size of allocation, same value as passed to alloc_pages_exact().
2113 * Release the memory allocated by a previous call to alloc_pages_exact.
2115 void free_pages_exact(void *virt
, size_t size
)
2117 unsigned long addr
= (unsigned long)virt
;
2118 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2120 while (addr
< end
) {
2125 EXPORT_SYMBOL(free_pages_exact
);
2127 static unsigned int nr_free_zone_pages(int offset
)
2132 /* Just pick one node, since fallback list is circular */
2133 unsigned int sum
= 0;
2135 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2137 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2138 unsigned long size
= zone
->present_pages
;
2139 unsigned long high
= high_wmark_pages(zone
);
2148 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2150 unsigned int nr_free_buffer_pages(void)
2152 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2154 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2157 * Amount of free RAM allocatable within all zones
2159 unsigned int nr_free_pagecache_pages(void)
2161 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2164 static inline void show_node(struct zone
*zone
)
2167 printk("Node %d ", zone_to_nid(zone
));
2170 void si_meminfo(struct sysinfo
*val
)
2172 val
->totalram
= totalram_pages
;
2174 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2175 val
->bufferram
= nr_blockdev_pages();
2176 val
->totalhigh
= totalhigh_pages
;
2177 val
->freehigh
= nr_free_highpages();
2178 val
->mem_unit
= PAGE_SIZE
;
2181 EXPORT_SYMBOL(si_meminfo
);
2184 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2186 pg_data_t
*pgdat
= NODE_DATA(nid
);
2188 val
->totalram
= pgdat
->node_present_pages
;
2189 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2190 #ifdef CONFIG_HIGHMEM
2191 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2192 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2198 val
->mem_unit
= PAGE_SIZE
;
2202 #define K(x) ((x) << (PAGE_SHIFT-10))
2205 * Show free area list (used inside shift_scroll-lock stuff)
2206 * We also calculate the percentage fragmentation. We do this by counting the
2207 * memory on each free list with the exception of the first item on the list.
2209 void show_free_areas(void)
2214 for_each_populated_zone(zone
) {
2216 printk("%s per-cpu:\n", zone
->name
);
2218 for_each_online_cpu(cpu
) {
2219 struct per_cpu_pageset
*pageset
;
2221 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2223 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2224 cpu
, pageset
->pcp
.high
,
2225 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2229 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2230 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2232 " dirty:%lu writeback:%lu unstable:%lu\n"
2233 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2234 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2235 global_page_state(NR_ACTIVE_ANON
),
2236 global_page_state(NR_INACTIVE_ANON
),
2237 global_page_state(NR_ISOLATED_ANON
),
2238 global_page_state(NR_ACTIVE_FILE
),
2239 global_page_state(NR_INACTIVE_FILE
),
2240 global_page_state(NR_ISOLATED_FILE
),
2241 global_page_state(NR_UNEVICTABLE
),
2242 global_page_state(NR_FILE_DIRTY
),
2243 global_page_state(NR_WRITEBACK
),
2244 global_page_state(NR_UNSTABLE_NFS
),
2245 global_page_state(NR_FREE_PAGES
),
2246 global_page_state(NR_SLAB_RECLAIMABLE
),
2247 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2248 global_page_state(NR_FILE_MAPPED
),
2249 global_page_state(NR_SHMEM
),
2250 global_page_state(NR_PAGETABLE
),
2251 global_page_state(NR_BOUNCE
));
2253 for_each_populated_zone(zone
) {
2262 " active_anon:%lukB"
2263 " inactive_anon:%lukB"
2264 " active_file:%lukB"
2265 " inactive_file:%lukB"
2266 " unevictable:%lukB"
2267 " isolated(anon):%lukB"
2268 " isolated(file):%lukB"
2275 " slab_reclaimable:%lukB"
2276 " slab_unreclaimable:%lukB"
2277 " kernel_stack:%lukB"
2281 " writeback_tmp:%lukB"
2282 " pages_scanned:%lu"
2283 " all_unreclaimable? %s"
2286 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2287 K(min_wmark_pages(zone
)),
2288 K(low_wmark_pages(zone
)),
2289 K(high_wmark_pages(zone
)),
2290 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2291 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2292 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2293 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2294 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2295 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
2296 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
2297 K(zone
->present_pages
),
2298 K(zone_page_state(zone
, NR_MLOCK
)),
2299 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
2300 K(zone_page_state(zone
, NR_WRITEBACK
)),
2301 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
2302 K(zone_page_state(zone
, NR_SHMEM
)),
2303 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
2304 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
2305 zone_page_state(zone
, NR_KERNEL_STACK
) *
2307 K(zone_page_state(zone
, NR_PAGETABLE
)),
2308 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
2309 K(zone_page_state(zone
, NR_BOUNCE
)),
2310 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
2311 zone
->pages_scanned
,
2312 (zone
->all_unreclaimable
? "yes" : "no")
2314 printk("lowmem_reserve[]:");
2315 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2316 printk(" %lu", zone
->lowmem_reserve
[i
]);
2320 for_each_populated_zone(zone
) {
2321 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2324 printk("%s: ", zone
->name
);
2326 spin_lock_irqsave(&zone
->lock
, flags
);
2327 for (order
= 0; order
< MAX_ORDER
; order
++) {
2328 nr
[order
] = zone
->free_area
[order
].nr_free
;
2329 total
+= nr
[order
] << order
;
2331 spin_unlock_irqrestore(&zone
->lock
, flags
);
2332 for (order
= 0; order
< MAX_ORDER
; order
++)
2333 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2334 printk("= %lukB\n", K(total
));
2337 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2339 show_swap_cache_info();
2342 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2344 zoneref
->zone
= zone
;
2345 zoneref
->zone_idx
= zone_idx(zone
);
2349 * Builds allocation fallback zone lists.
2351 * Add all populated zones of a node to the zonelist.
2353 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2354 int nr_zones
, enum zone_type zone_type
)
2358 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2363 zone
= pgdat
->node_zones
+ zone_type
;
2364 if (populated_zone(zone
)) {
2365 zoneref_set_zone(zone
,
2366 &zonelist
->_zonerefs
[nr_zones
++]);
2367 check_highest_zone(zone_type
);
2370 } while (zone_type
);
2377 * 0 = automatic detection of better ordering.
2378 * 1 = order by ([node] distance, -zonetype)
2379 * 2 = order by (-zonetype, [node] distance)
2381 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2382 * the same zonelist. So only NUMA can configure this param.
2384 #define ZONELIST_ORDER_DEFAULT 0
2385 #define ZONELIST_ORDER_NODE 1
2386 #define ZONELIST_ORDER_ZONE 2
2388 /* zonelist order in the kernel.
2389 * set_zonelist_order() will set this to NODE or ZONE.
2391 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2392 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2396 /* The value user specified ....changed by config */
2397 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2398 /* string for sysctl */
2399 #define NUMA_ZONELIST_ORDER_LEN 16
2400 char numa_zonelist_order
[16] = "default";
2403 * interface for configure zonelist ordering.
2404 * command line option "numa_zonelist_order"
2405 * = "[dD]efault - default, automatic configuration.
2406 * = "[nN]ode - order by node locality, then by zone within node
2407 * = "[zZ]one - order by zone, then by locality within zone
2410 static int __parse_numa_zonelist_order(char *s
)
2412 if (*s
== 'd' || *s
== 'D') {
2413 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2414 } else if (*s
== 'n' || *s
== 'N') {
2415 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2416 } else if (*s
== 'z' || *s
== 'Z') {
2417 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2420 "Ignoring invalid numa_zonelist_order value: "
2427 static __init
int setup_numa_zonelist_order(char *s
)
2430 return __parse_numa_zonelist_order(s
);
2433 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2436 * sysctl handler for numa_zonelist_order
2438 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2439 void __user
*buffer
, size_t *length
,
2442 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2444 static DEFINE_MUTEX(zl_order_mutex
);
2446 mutex_lock(&zl_order_mutex
);
2448 strcpy(saved_string
, (char*)table
->data
);
2449 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
2453 int oldval
= user_zonelist_order
;
2454 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2456 * bogus value. restore saved string
2458 strncpy((char*)table
->data
, saved_string
,
2459 NUMA_ZONELIST_ORDER_LEN
);
2460 user_zonelist_order
= oldval
;
2461 } else if (oldval
!= user_zonelist_order
)
2462 build_all_zonelists();
2465 mutex_unlock(&zl_order_mutex
);
2470 #define MAX_NODE_LOAD (nr_online_nodes)
2471 static int node_load
[MAX_NUMNODES
];
2474 * find_next_best_node - find the next node that should appear in a given node's fallback list
2475 * @node: node whose fallback list we're appending
2476 * @used_node_mask: nodemask_t of already used nodes
2478 * We use a number of factors to determine which is the next node that should
2479 * appear on a given node's fallback list. The node should not have appeared
2480 * already in @node's fallback list, and it should be the next closest node
2481 * according to the distance array (which contains arbitrary distance values
2482 * from each node to each node in the system), and should also prefer nodes
2483 * with no CPUs, since presumably they'll have very little allocation pressure
2484 * on them otherwise.
2485 * It returns -1 if no node is found.
2487 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2490 int min_val
= INT_MAX
;
2492 const struct cpumask
*tmp
= cpumask_of_node(0);
2494 /* Use the local node if we haven't already */
2495 if (!node_isset(node
, *used_node_mask
)) {
2496 node_set(node
, *used_node_mask
);
2500 for_each_node_state(n
, N_HIGH_MEMORY
) {
2502 /* Don't want a node to appear more than once */
2503 if (node_isset(n
, *used_node_mask
))
2506 /* Use the distance array to find the distance */
2507 val
= node_distance(node
, n
);
2509 /* Penalize nodes under us ("prefer the next node") */
2512 /* Give preference to headless and unused nodes */
2513 tmp
= cpumask_of_node(n
);
2514 if (!cpumask_empty(tmp
))
2515 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2517 /* Slight preference for less loaded node */
2518 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2519 val
+= node_load
[n
];
2521 if (val
< min_val
) {
2528 node_set(best_node
, *used_node_mask
);
2535 * Build zonelists ordered by node and zones within node.
2536 * This results in maximum locality--normal zone overflows into local
2537 * DMA zone, if any--but risks exhausting DMA zone.
2539 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2542 struct zonelist
*zonelist
;
2544 zonelist
= &pgdat
->node_zonelists
[0];
2545 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2547 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2549 zonelist
->_zonerefs
[j
].zone
= NULL
;
2550 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2554 * Build gfp_thisnode zonelists
2556 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2559 struct zonelist
*zonelist
;
2561 zonelist
= &pgdat
->node_zonelists
[1];
2562 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2563 zonelist
->_zonerefs
[j
].zone
= NULL
;
2564 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2568 * Build zonelists ordered by zone and nodes within zones.
2569 * This results in conserving DMA zone[s] until all Normal memory is
2570 * exhausted, but results in overflowing to remote node while memory
2571 * may still exist in local DMA zone.
2573 static int node_order
[MAX_NUMNODES
];
2575 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2578 int zone_type
; /* needs to be signed */
2580 struct zonelist
*zonelist
;
2582 zonelist
= &pgdat
->node_zonelists
[0];
2584 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2585 for (j
= 0; j
< nr_nodes
; j
++) {
2586 node
= node_order
[j
];
2587 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2588 if (populated_zone(z
)) {
2590 &zonelist
->_zonerefs
[pos
++]);
2591 check_highest_zone(zone_type
);
2595 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2596 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2599 static int default_zonelist_order(void)
2602 unsigned long low_kmem_size
,total_size
;
2606 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2607 * If they are really small and used heavily, the system can fall
2608 * into OOM very easily.
2609 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2611 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2614 for_each_online_node(nid
) {
2615 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2616 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2617 if (populated_zone(z
)) {
2618 if (zone_type
< ZONE_NORMAL
)
2619 low_kmem_size
+= z
->present_pages
;
2620 total_size
+= z
->present_pages
;
2624 if (!low_kmem_size
|| /* there are no DMA area. */
2625 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2626 return ZONELIST_ORDER_NODE
;
2628 * look into each node's config.
2629 * If there is a node whose DMA/DMA32 memory is very big area on
2630 * local memory, NODE_ORDER may be suitable.
2632 average_size
= total_size
/
2633 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2634 for_each_online_node(nid
) {
2637 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2638 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2639 if (populated_zone(z
)) {
2640 if (zone_type
< ZONE_NORMAL
)
2641 low_kmem_size
+= z
->present_pages
;
2642 total_size
+= z
->present_pages
;
2645 if (low_kmem_size
&&
2646 total_size
> average_size
&& /* ignore small node */
2647 low_kmem_size
> total_size
* 70/100)
2648 return ZONELIST_ORDER_NODE
;
2650 return ZONELIST_ORDER_ZONE
;
2653 static void set_zonelist_order(void)
2655 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2656 current_zonelist_order
= default_zonelist_order();
2658 current_zonelist_order
= user_zonelist_order
;
2661 static void build_zonelists(pg_data_t
*pgdat
)
2665 nodemask_t used_mask
;
2666 int local_node
, prev_node
;
2667 struct zonelist
*zonelist
;
2668 int order
= current_zonelist_order
;
2670 /* initialize zonelists */
2671 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2672 zonelist
= pgdat
->node_zonelists
+ i
;
2673 zonelist
->_zonerefs
[0].zone
= NULL
;
2674 zonelist
->_zonerefs
[0].zone_idx
= 0;
2677 /* NUMA-aware ordering of nodes */
2678 local_node
= pgdat
->node_id
;
2679 load
= nr_online_nodes
;
2680 prev_node
= local_node
;
2681 nodes_clear(used_mask
);
2683 memset(node_order
, 0, sizeof(node_order
));
2686 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2687 int distance
= node_distance(local_node
, node
);
2690 * If another node is sufficiently far away then it is better
2691 * to reclaim pages in a zone before going off node.
2693 if (distance
> RECLAIM_DISTANCE
)
2694 zone_reclaim_mode
= 1;
2697 * We don't want to pressure a particular node.
2698 * So adding penalty to the first node in same
2699 * distance group to make it round-robin.
2701 if (distance
!= node_distance(local_node
, prev_node
))
2702 node_load
[node
] = load
;
2706 if (order
== ZONELIST_ORDER_NODE
)
2707 build_zonelists_in_node_order(pgdat
, node
);
2709 node_order
[j
++] = node
; /* remember order */
2712 if (order
== ZONELIST_ORDER_ZONE
) {
2713 /* calculate node order -- i.e., DMA last! */
2714 build_zonelists_in_zone_order(pgdat
, j
);
2717 build_thisnode_zonelists(pgdat
);
2720 /* Construct the zonelist performance cache - see further mmzone.h */
2721 static void build_zonelist_cache(pg_data_t
*pgdat
)
2723 struct zonelist
*zonelist
;
2724 struct zonelist_cache
*zlc
;
2727 zonelist
= &pgdat
->node_zonelists
[0];
2728 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2729 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2730 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2731 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2735 #else /* CONFIG_NUMA */
2737 static void set_zonelist_order(void)
2739 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2742 static void build_zonelists(pg_data_t
*pgdat
)
2744 int node
, local_node
;
2746 struct zonelist
*zonelist
;
2748 local_node
= pgdat
->node_id
;
2750 zonelist
= &pgdat
->node_zonelists
[0];
2751 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2754 * Now we build the zonelist so that it contains the zones
2755 * of all the other nodes.
2756 * We don't want to pressure a particular node, so when
2757 * building the zones for node N, we make sure that the
2758 * zones coming right after the local ones are those from
2759 * node N+1 (modulo N)
2761 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2762 if (!node_online(node
))
2764 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2767 for (node
= 0; node
< local_node
; node
++) {
2768 if (!node_online(node
))
2770 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2774 zonelist
->_zonerefs
[j
].zone
= NULL
;
2775 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2778 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2779 static void build_zonelist_cache(pg_data_t
*pgdat
)
2781 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2784 #endif /* CONFIG_NUMA */
2787 * Boot pageset table. One per cpu which is going to be used for all
2788 * zones and all nodes. The parameters will be set in such a way
2789 * that an item put on a list will immediately be handed over to
2790 * the buddy list. This is safe since pageset manipulation is done
2791 * with interrupts disabled.
2793 * The boot_pagesets must be kept even after bootup is complete for
2794 * unused processors and/or zones. They do play a role for bootstrapping
2795 * hotplugged processors.
2797 * zoneinfo_show() and maybe other functions do
2798 * not check if the processor is online before following the pageset pointer.
2799 * Other parts of the kernel may not check if the zone is available.
2801 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
2802 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
2804 /* return values int ....just for stop_machine() */
2805 static int __build_all_zonelists(void *dummy
)
2811 memset(node_load
, 0, sizeof(node_load
));
2813 for_each_online_node(nid
) {
2814 pg_data_t
*pgdat
= NODE_DATA(nid
);
2816 build_zonelists(pgdat
);
2817 build_zonelist_cache(pgdat
);
2821 * Initialize the boot_pagesets that are going to be used
2822 * for bootstrapping processors. The real pagesets for
2823 * each zone will be allocated later when the per cpu
2824 * allocator is available.
2826 * boot_pagesets are used also for bootstrapping offline
2827 * cpus if the system is already booted because the pagesets
2828 * are needed to initialize allocators on a specific cpu too.
2829 * F.e. the percpu allocator needs the page allocator which
2830 * needs the percpu allocator in order to allocate its pagesets
2831 * (a chicken-egg dilemma).
2833 for_each_possible_cpu(cpu
)
2834 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
2839 void build_all_zonelists(void)
2841 set_zonelist_order();
2843 if (system_state
== SYSTEM_BOOTING
) {
2844 __build_all_zonelists(NULL
);
2845 mminit_verify_zonelist();
2846 cpuset_init_current_mems_allowed();
2848 /* we have to stop all cpus to guarantee there is no user
2850 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2851 /* cpuset refresh routine should be here */
2853 vm_total_pages
= nr_free_pagecache_pages();
2855 * Disable grouping by mobility if the number of pages in the
2856 * system is too low to allow the mechanism to work. It would be
2857 * more accurate, but expensive to check per-zone. This check is
2858 * made on memory-hotadd so a system can start with mobility
2859 * disabled and enable it later
2861 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2862 page_group_by_mobility_disabled
= 1;
2864 page_group_by_mobility_disabled
= 0;
2866 printk("Built %i zonelists in %s order, mobility grouping %s. "
2867 "Total pages: %ld\n",
2869 zonelist_order_name
[current_zonelist_order
],
2870 page_group_by_mobility_disabled
? "off" : "on",
2873 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2878 * Helper functions to size the waitqueue hash table.
2879 * Essentially these want to choose hash table sizes sufficiently
2880 * large so that collisions trying to wait on pages are rare.
2881 * But in fact, the number of active page waitqueues on typical
2882 * systems is ridiculously low, less than 200. So this is even
2883 * conservative, even though it seems large.
2885 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2886 * waitqueues, i.e. the size of the waitq table given the number of pages.
2888 #define PAGES_PER_WAITQUEUE 256
2890 #ifndef CONFIG_MEMORY_HOTPLUG
2891 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2893 unsigned long size
= 1;
2895 pages
/= PAGES_PER_WAITQUEUE
;
2897 while (size
< pages
)
2901 * Once we have dozens or even hundreds of threads sleeping
2902 * on IO we've got bigger problems than wait queue collision.
2903 * Limit the size of the wait table to a reasonable size.
2905 size
= min(size
, 4096UL);
2907 return max(size
, 4UL);
2911 * A zone's size might be changed by hot-add, so it is not possible to determine
2912 * a suitable size for its wait_table. So we use the maximum size now.
2914 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2916 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2917 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2918 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2920 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2921 * or more by the traditional way. (See above). It equals:
2923 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2924 * ia64(16K page size) : = ( 8G + 4M)byte.
2925 * powerpc (64K page size) : = (32G +16M)byte.
2927 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2934 * This is an integer logarithm so that shifts can be used later
2935 * to extract the more random high bits from the multiplicative
2936 * hash function before the remainder is taken.
2938 static inline unsigned long wait_table_bits(unsigned long size
)
2943 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2946 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2947 * of blocks reserved is based on min_wmark_pages(zone). The memory within
2948 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
2949 * higher will lead to a bigger reserve which will get freed as contiguous
2950 * blocks as reclaim kicks in
2952 static void setup_zone_migrate_reserve(struct zone
*zone
)
2954 unsigned long start_pfn
, pfn
, end_pfn
;
2956 unsigned long block_migratetype
;
2959 /* Get the start pfn, end pfn and the number of blocks to reserve */
2960 start_pfn
= zone
->zone_start_pfn
;
2961 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2962 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
2966 * Reserve blocks are generally in place to help high-order atomic
2967 * allocations that are short-lived. A min_free_kbytes value that
2968 * would result in more than 2 reserve blocks for atomic allocations
2969 * is assumed to be in place to help anti-fragmentation for the
2970 * future allocation of hugepages at runtime.
2972 reserve
= min(2, reserve
);
2974 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2975 if (!pfn_valid(pfn
))
2977 page
= pfn_to_page(pfn
);
2979 /* Watch out for overlapping nodes */
2980 if (page_to_nid(page
) != zone_to_nid(zone
))
2983 /* Blocks with reserved pages will never free, skip them. */
2984 if (PageReserved(page
))
2987 block_migratetype
= get_pageblock_migratetype(page
);
2989 /* If this block is reserved, account for it */
2990 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2995 /* Suitable for reserving if this block is movable */
2996 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2997 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2998 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
3004 * If the reserve is met and this is a previous reserved block,
3007 if (block_migratetype
== MIGRATE_RESERVE
) {
3008 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3009 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3015 * Initially all pages are reserved - free ones are freed
3016 * up by free_all_bootmem() once the early boot process is
3017 * done. Non-atomic initialization, single-pass.
3019 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3020 unsigned long start_pfn
, enum memmap_context context
)
3023 unsigned long end_pfn
= start_pfn
+ size
;
3027 if (highest_memmap_pfn
< end_pfn
- 1)
3028 highest_memmap_pfn
= end_pfn
- 1;
3030 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3031 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3033 * There can be holes in boot-time mem_map[]s
3034 * handed to this function. They do not
3035 * exist on hotplugged memory.
3037 if (context
== MEMMAP_EARLY
) {
3038 if (!early_pfn_valid(pfn
))
3040 if (!early_pfn_in_nid(pfn
, nid
))
3043 page
= pfn_to_page(pfn
);
3044 set_page_links(page
, zone
, nid
, pfn
);
3045 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3046 init_page_count(page
);
3047 reset_page_mapcount(page
);
3048 SetPageReserved(page
);
3050 * Mark the block movable so that blocks are reserved for
3051 * movable at startup. This will force kernel allocations
3052 * to reserve their blocks rather than leaking throughout
3053 * the address space during boot when many long-lived
3054 * kernel allocations are made. Later some blocks near
3055 * the start are marked MIGRATE_RESERVE by
3056 * setup_zone_migrate_reserve()
3058 * bitmap is created for zone's valid pfn range. but memmap
3059 * can be created for invalid pages (for alignment)
3060 * check here not to call set_pageblock_migratetype() against
3063 if ((z
->zone_start_pfn
<= pfn
)
3064 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3065 && !(pfn
& (pageblock_nr_pages
- 1)))
3066 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3068 INIT_LIST_HEAD(&page
->lru
);
3069 #ifdef WANT_PAGE_VIRTUAL
3070 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3071 if (!is_highmem_idx(zone
))
3072 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3077 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3080 for_each_migratetype_order(order
, t
) {
3081 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3082 zone
->free_area
[order
].nr_free
= 0;
3086 #ifndef __HAVE_ARCH_MEMMAP_INIT
3087 #define memmap_init(size, nid, zone, start_pfn) \
3088 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3091 static int zone_batchsize(struct zone
*zone
)
3097 * The per-cpu-pages pools are set to around 1000th of the
3098 * size of the zone. But no more than 1/2 of a meg.
3100 * OK, so we don't know how big the cache is. So guess.
3102 batch
= zone
->present_pages
/ 1024;
3103 if (batch
* PAGE_SIZE
> 512 * 1024)
3104 batch
= (512 * 1024) / PAGE_SIZE
;
3105 batch
/= 4; /* We effectively *= 4 below */
3110 * Clamp the batch to a 2^n - 1 value. Having a power
3111 * of 2 value was found to be more likely to have
3112 * suboptimal cache aliasing properties in some cases.
3114 * For example if 2 tasks are alternately allocating
3115 * batches of pages, one task can end up with a lot
3116 * of pages of one half of the possible page colors
3117 * and the other with pages of the other colors.
3119 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3124 /* The deferral and batching of frees should be suppressed under NOMMU
3127 * The problem is that NOMMU needs to be able to allocate large chunks
3128 * of contiguous memory as there's no hardware page translation to
3129 * assemble apparent contiguous memory from discontiguous pages.
3131 * Queueing large contiguous runs of pages for batching, however,
3132 * causes the pages to actually be freed in smaller chunks. As there
3133 * can be a significant delay between the individual batches being
3134 * recycled, this leads to the once large chunks of space being
3135 * fragmented and becoming unavailable for high-order allocations.
3141 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3143 struct per_cpu_pages
*pcp
;
3146 memset(p
, 0, sizeof(*p
));
3150 pcp
->high
= 6 * batch
;
3151 pcp
->batch
= max(1UL, 1 * batch
);
3152 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3153 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3157 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3158 * to the value high for the pageset p.
3161 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3164 struct per_cpu_pages
*pcp
;
3168 pcp
->batch
= max(1UL, high
/4);
3169 if ((high
/4) > (PAGE_SHIFT
* 8))
3170 pcp
->batch
= PAGE_SHIFT
* 8;
3174 * Allocate per cpu pagesets and initialize them.
3175 * Before this call only boot pagesets were available.
3176 * Boot pagesets will no longer be used by this processorr
3177 * after setup_per_cpu_pageset().
3179 void __init
setup_per_cpu_pageset(void)
3184 for_each_populated_zone(zone
) {
3185 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
3187 for_each_possible_cpu(cpu
) {
3188 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
3190 setup_pageset(pcp
, zone_batchsize(zone
));
3192 if (percpu_pagelist_fraction
)
3193 setup_pagelist_highmark(pcp
,
3194 (zone
->present_pages
/
3195 percpu_pagelist_fraction
));
3200 static noinline __init_refok
3201 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3204 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3208 * The per-page waitqueue mechanism uses hashed waitqueues
3211 zone
->wait_table_hash_nr_entries
=
3212 wait_table_hash_nr_entries(zone_size_pages
);
3213 zone
->wait_table_bits
=
3214 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3215 alloc_size
= zone
->wait_table_hash_nr_entries
3216 * sizeof(wait_queue_head_t
);
3218 if (!slab_is_available()) {
3219 zone
->wait_table
= (wait_queue_head_t
*)
3220 alloc_bootmem_node(pgdat
, alloc_size
);
3223 * This case means that a zone whose size was 0 gets new memory
3224 * via memory hot-add.
3225 * But it may be the case that a new node was hot-added. In
3226 * this case vmalloc() will not be able to use this new node's
3227 * memory - this wait_table must be initialized to use this new
3228 * node itself as well.
3229 * To use this new node's memory, further consideration will be
3232 zone
->wait_table
= vmalloc(alloc_size
);
3234 if (!zone
->wait_table
)
3237 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3238 init_waitqueue_head(zone
->wait_table
+ i
);
3243 static int __zone_pcp_update(void *data
)
3245 struct zone
*zone
= data
;
3247 unsigned long batch
= zone_batchsize(zone
), flags
;
3249 for_each_possible_cpu(cpu
) {
3250 struct per_cpu_pageset
*pset
;
3251 struct per_cpu_pages
*pcp
;
3253 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
3256 local_irq_save(flags
);
3257 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
3258 setup_pageset(pset
, batch
);
3259 local_irq_restore(flags
);
3264 void zone_pcp_update(struct zone
*zone
)
3266 stop_machine(__zone_pcp_update
, zone
, NULL
);
3269 static __meminit
void zone_pcp_init(struct zone
*zone
)
3272 * per cpu subsystem is not up at this point. The following code
3273 * relies on the ability of the linker to provide the
3274 * offset of a (static) per cpu variable into the per cpu area.
3276 zone
->pageset
= &boot_pageset
;
3278 if (zone
->present_pages
)
3279 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
3280 zone
->name
, zone
->present_pages
,
3281 zone_batchsize(zone
));
3284 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3285 unsigned long zone_start_pfn
,
3287 enum memmap_context context
)
3289 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3291 ret
= zone_wait_table_init(zone
, size
);
3294 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3296 zone
->zone_start_pfn
= zone_start_pfn
;
3298 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3299 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3301 (unsigned long)zone_idx(zone
),
3302 zone_start_pfn
, (zone_start_pfn
+ size
));
3304 zone_init_free_lists(zone
);
3309 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3311 * Basic iterator support. Return the first range of PFNs for a node
3312 * Note: nid == MAX_NUMNODES returns first region regardless of node
3314 static int __meminit
first_active_region_index_in_nid(int nid
)
3318 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3319 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3326 * Basic iterator support. Return the next active range of PFNs for a node
3327 * Note: nid == MAX_NUMNODES returns next region regardless of node
3329 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3331 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3332 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3338 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3340 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3341 * Architectures may implement their own version but if add_active_range()
3342 * was used and there are no special requirements, this is a convenient
3345 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3349 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3350 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3351 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3353 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3354 return early_node_map
[i
].nid
;
3356 /* This is a memory hole */
3359 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3361 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3365 nid
= __early_pfn_to_nid(pfn
);
3368 /* just returns 0 */
3372 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3373 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3377 nid
= __early_pfn_to_nid(pfn
);
3378 if (nid
>= 0 && nid
!= node
)
3384 /* Basic iterator support to walk early_node_map[] */
3385 #define for_each_active_range_index_in_nid(i, nid) \
3386 for (i = first_active_region_index_in_nid(nid); i != -1; \
3387 i = next_active_region_index_in_nid(i, nid))
3390 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3391 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3392 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3394 * If an architecture guarantees that all ranges registered with
3395 * add_active_ranges() contain no holes and may be freed, this
3396 * this function may be used instead of calling free_bootmem() manually.
3398 void __init
free_bootmem_with_active_regions(int nid
,
3399 unsigned long max_low_pfn
)
3403 for_each_active_range_index_in_nid(i
, nid
) {
3404 unsigned long size_pages
= 0;
3405 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3407 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3410 if (end_pfn
> max_low_pfn
)
3411 end_pfn
= max_low_pfn
;
3413 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3414 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3415 PFN_PHYS(early_node_map
[i
].start_pfn
),
3416 size_pages
<< PAGE_SHIFT
);
3420 int __init
add_from_early_node_map(struct range
*range
, int az
,
3421 int nr_range
, int nid
)
3426 /* need to go over early_node_map to find out good range for node */
3427 for_each_active_range_index_in_nid(i
, nid
) {
3428 start
= early_node_map
[i
].start_pfn
;
3429 end
= early_node_map
[i
].end_pfn
;
3430 nr_range
= add_range(range
, az
, nr_range
, start
, end
);
3435 #ifdef CONFIG_NO_BOOTMEM
3436 void * __init
__alloc_memory_core_early(int nid
, u64 size
, u64 align
,
3437 u64 goal
, u64 limit
)
3442 /* need to go over early_node_map to find out good range for node */
3443 for_each_active_range_index_in_nid(i
, nid
) {
3445 u64 ei_start
, ei_last
;
3447 ei_last
= early_node_map
[i
].end_pfn
;
3448 ei_last
<<= PAGE_SHIFT
;
3449 ei_start
= early_node_map
[i
].start_pfn
;
3450 ei_start
<<= PAGE_SHIFT
;
3451 addr
= find_early_area(ei_start
, ei_last
,
3452 goal
, limit
, size
, align
);
3458 printk(KERN_DEBUG
"alloc (nid=%d %llx - %llx) (%llx - %llx) %llx %llx => %llx\n",
3460 ei_start
, ei_last
, goal
, limit
, size
,
3464 ptr
= phys_to_virt(addr
);
3465 memset(ptr
, 0, size
);
3466 reserve_early_without_check(addr
, addr
+ size
, "BOOTMEM");
3475 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3480 for_each_active_range_index_in_nid(i
, nid
) {
3481 ret
= work_fn(early_node_map
[i
].start_pfn
,
3482 early_node_map
[i
].end_pfn
, data
);
3488 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3489 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3491 * If an architecture guarantees that all ranges registered with
3492 * add_active_ranges() contain no holes and may be freed, this
3493 * function may be used instead of calling memory_present() manually.
3495 void __init
sparse_memory_present_with_active_regions(int nid
)
3499 for_each_active_range_index_in_nid(i
, nid
)
3500 memory_present(early_node_map
[i
].nid
,
3501 early_node_map
[i
].start_pfn
,
3502 early_node_map
[i
].end_pfn
);
3506 * get_pfn_range_for_nid - Return the start and end page frames for a node
3507 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3508 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3509 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3511 * It returns the start and end page frame of a node based on information
3512 * provided by an arch calling add_active_range(). If called for a node
3513 * with no available memory, a warning is printed and the start and end
3516 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3517 unsigned long *start_pfn
, unsigned long *end_pfn
)
3523 for_each_active_range_index_in_nid(i
, nid
) {
3524 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3525 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3528 if (*start_pfn
== -1UL)
3533 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3534 * assumption is made that zones within a node are ordered in monotonic
3535 * increasing memory addresses so that the "highest" populated zone is used
3537 static void __init
find_usable_zone_for_movable(void)
3540 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3541 if (zone_index
== ZONE_MOVABLE
)
3544 if (arch_zone_highest_possible_pfn
[zone_index
] >
3545 arch_zone_lowest_possible_pfn
[zone_index
])
3549 VM_BUG_ON(zone_index
== -1);
3550 movable_zone
= zone_index
;
3554 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3555 * because it is sized independant of architecture. Unlike the other zones,
3556 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3557 * in each node depending on the size of each node and how evenly kernelcore
3558 * is distributed. This helper function adjusts the zone ranges
3559 * provided by the architecture for a given node by using the end of the
3560 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3561 * zones within a node are in order of monotonic increases memory addresses
3563 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3564 unsigned long zone_type
,
3565 unsigned long node_start_pfn
,
3566 unsigned long node_end_pfn
,
3567 unsigned long *zone_start_pfn
,
3568 unsigned long *zone_end_pfn
)
3570 /* Only adjust if ZONE_MOVABLE is on this node */
3571 if (zone_movable_pfn
[nid
]) {
3572 /* Size ZONE_MOVABLE */
3573 if (zone_type
== ZONE_MOVABLE
) {
3574 *zone_start_pfn
= zone_movable_pfn
[nid
];
3575 *zone_end_pfn
= min(node_end_pfn
,
3576 arch_zone_highest_possible_pfn
[movable_zone
]);
3578 /* Adjust for ZONE_MOVABLE starting within this range */
3579 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3580 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3581 *zone_end_pfn
= zone_movable_pfn
[nid
];
3583 /* Check if this whole range is within ZONE_MOVABLE */
3584 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3585 *zone_start_pfn
= *zone_end_pfn
;
3590 * Return the number of pages a zone spans in a node, including holes
3591 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3593 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3594 unsigned long zone_type
,
3595 unsigned long *ignored
)
3597 unsigned long node_start_pfn
, node_end_pfn
;
3598 unsigned long zone_start_pfn
, zone_end_pfn
;
3600 /* Get the start and end of the node and zone */
3601 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3602 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3603 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3604 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3605 node_start_pfn
, node_end_pfn
,
3606 &zone_start_pfn
, &zone_end_pfn
);
3608 /* Check that this node has pages within the zone's required range */
3609 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3612 /* Move the zone boundaries inside the node if necessary */
3613 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3614 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3616 /* Return the spanned pages */
3617 return zone_end_pfn
- zone_start_pfn
;
3621 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3622 * then all holes in the requested range will be accounted for.
3624 unsigned long __meminit
__absent_pages_in_range(int nid
,
3625 unsigned long range_start_pfn
,
3626 unsigned long range_end_pfn
)
3629 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3630 unsigned long start_pfn
;
3632 /* Find the end_pfn of the first active range of pfns in the node */
3633 i
= first_active_region_index_in_nid(nid
);
3637 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3639 /* Account for ranges before physical memory on this node */
3640 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3641 hole_pages
= prev_end_pfn
- range_start_pfn
;
3643 /* Find all holes for the zone within the node */
3644 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3646 /* No need to continue if prev_end_pfn is outside the zone */
3647 if (prev_end_pfn
>= range_end_pfn
)
3650 /* Make sure the end of the zone is not within the hole */
3651 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3652 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3654 /* Update the hole size cound and move on */
3655 if (start_pfn
> range_start_pfn
) {
3656 BUG_ON(prev_end_pfn
> start_pfn
);
3657 hole_pages
+= start_pfn
- prev_end_pfn
;
3659 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3662 /* Account for ranges past physical memory on this node */
3663 if (range_end_pfn
> prev_end_pfn
)
3664 hole_pages
+= range_end_pfn
-
3665 max(range_start_pfn
, prev_end_pfn
);
3671 * absent_pages_in_range - Return number of page frames in holes within a range
3672 * @start_pfn: The start PFN to start searching for holes
3673 * @end_pfn: The end PFN to stop searching for holes
3675 * It returns the number of pages frames in memory holes within a range.
3677 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3678 unsigned long end_pfn
)
3680 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3683 /* Return the number of page frames in holes in a zone on a node */
3684 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3685 unsigned long zone_type
,
3686 unsigned long *ignored
)
3688 unsigned long node_start_pfn
, node_end_pfn
;
3689 unsigned long zone_start_pfn
, zone_end_pfn
;
3691 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3692 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3694 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3697 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3698 node_start_pfn
, node_end_pfn
,
3699 &zone_start_pfn
, &zone_end_pfn
);
3700 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3704 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3705 unsigned long zone_type
,
3706 unsigned long *zones_size
)
3708 return zones_size
[zone_type
];
3711 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3712 unsigned long zone_type
,
3713 unsigned long *zholes_size
)
3718 return zholes_size
[zone_type
];
3723 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3724 unsigned long *zones_size
, unsigned long *zholes_size
)
3726 unsigned long realtotalpages
, totalpages
= 0;
3729 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3730 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3732 pgdat
->node_spanned_pages
= totalpages
;
3734 realtotalpages
= totalpages
;
3735 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3737 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3739 pgdat
->node_present_pages
= realtotalpages
;
3740 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3744 #ifndef CONFIG_SPARSEMEM
3746 * Calculate the size of the zone->blockflags rounded to an unsigned long
3747 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3748 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3749 * round what is now in bits to nearest long in bits, then return it in
3752 static unsigned long __init
usemap_size(unsigned long zonesize
)
3754 unsigned long usemapsize
;
3756 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3757 usemapsize
= usemapsize
>> pageblock_order
;
3758 usemapsize
*= NR_PAGEBLOCK_BITS
;
3759 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3761 return usemapsize
/ 8;
3764 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3765 struct zone
*zone
, unsigned long zonesize
)
3767 unsigned long usemapsize
= usemap_size(zonesize
);
3768 zone
->pageblock_flags
= NULL
;
3770 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3773 static void inline setup_usemap(struct pglist_data
*pgdat
,
3774 struct zone
*zone
, unsigned long zonesize
) {}
3775 #endif /* CONFIG_SPARSEMEM */
3777 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3779 /* Return a sensible default order for the pageblock size. */
3780 static inline int pageblock_default_order(void)
3782 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3783 return HUGETLB_PAGE_ORDER
;
3788 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3789 static inline void __init
set_pageblock_order(unsigned int order
)
3791 /* Check that pageblock_nr_pages has not already been setup */
3792 if (pageblock_order
)
3796 * Assume the largest contiguous order of interest is a huge page.
3797 * This value may be variable depending on boot parameters on IA64
3799 pageblock_order
= order
;
3801 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3804 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3805 * and pageblock_default_order() are unused as pageblock_order is set
3806 * at compile-time. See include/linux/pageblock-flags.h for the values of
3807 * pageblock_order based on the kernel config
3809 static inline int pageblock_default_order(unsigned int order
)
3813 #define set_pageblock_order(x) do {} while (0)
3815 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3818 * Set up the zone data structures:
3819 * - mark all pages reserved
3820 * - mark all memory queues empty
3821 * - clear the memory bitmaps
3823 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3824 unsigned long *zones_size
, unsigned long *zholes_size
)
3827 int nid
= pgdat
->node_id
;
3828 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3831 pgdat_resize_init(pgdat
);
3832 pgdat
->nr_zones
= 0;
3833 init_waitqueue_head(&pgdat
->kswapd_wait
);
3834 pgdat
->kswapd_max_order
= 0;
3835 pgdat_page_cgroup_init(pgdat
);
3837 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3838 struct zone
*zone
= pgdat
->node_zones
+ j
;
3839 unsigned long size
, realsize
, memmap_pages
;
3842 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3843 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3847 * Adjust realsize so that it accounts for how much memory
3848 * is used by this zone for memmap. This affects the watermark
3849 * and per-cpu initialisations
3852 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3853 if (realsize
>= memmap_pages
) {
3854 realsize
-= memmap_pages
;
3857 " %s zone: %lu pages used for memmap\n",
3858 zone_names
[j
], memmap_pages
);
3861 " %s zone: %lu pages exceeds realsize %lu\n",
3862 zone_names
[j
], memmap_pages
, realsize
);
3864 /* Account for reserved pages */
3865 if (j
== 0 && realsize
> dma_reserve
) {
3866 realsize
-= dma_reserve
;
3867 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3868 zone_names
[0], dma_reserve
);
3871 if (!is_highmem_idx(j
))
3872 nr_kernel_pages
+= realsize
;
3873 nr_all_pages
+= realsize
;
3875 zone
->spanned_pages
= size
;
3876 zone
->present_pages
= realsize
;
3879 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3881 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3883 zone
->name
= zone_names
[j
];
3884 spin_lock_init(&zone
->lock
);
3885 spin_lock_init(&zone
->lru_lock
);
3886 zone_seqlock_init(zone
);
3887 zone
->zone_pgdat
= pgdat
;
3889 zone
->prev_priority
= DEF_PRIORITY
;
3891 zone_pcp_init(zone
);
3893 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3894 zone
->reclaim_stat
.nr_saved_scan
[l
] = 0;
3896 zone
->reclaim_stat
.recent_rotated
[0] = 0;
3897 zone
->reclaim_stat
.recent_rotated
[1] = 0;
3898 zone
->reclaim_stat
.recent_scanned
[0] = 0;
3899 zone
->reclaim_stat
.recent_scanned
[1] = 0;
3900 zap_zone_vm_stats(zone
);
3905 set_pageblock_order(pageblock_default_order());
3906 setup_usemap(pgdat
, zone
, size
);
3907 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3908 size
, MEMMAP_EARLY
);
3910 memmap_init(size
, nid
, j
, zone_start_pfn
);
3911 zone_start_pfn
+= size
;
3915 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3917 /* Skip empty nodes */
3918 if (!pgdat
->node_spanned_pages
)
3921 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3922 /* ia64 gets its own node_mem_map, before this, without bootmem */
3923 if (!pgdat
->node_mem_map
) {
3924 unsigned long size
, start
, end
;
3928 * The zone's endpoints aren't required to be MAX_ORDER
3929 * aligned but the node_mem_map endpoints must be in order
3930 * for the buddy allocator to function correctly.
3932 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3933 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3934 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3935 size
= (end
- start
) * sizeof(struct page
);
3936 map
= alloc_remap(pgdat
->node_id
, size
);
3938 map
= alloc_bootmem_node(pgdat
, size
);
3939 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3941 #ifndef CONFIG_NEED_MULTIPLE_NODES
3943 * With no DISCONTIG, the global mem_map is just set as node 0's
3945 if (pgdat
== NODE_DATA(0)) {
3946 mem_map
= NODE_DATA(0)->node_mem_map
;
3947 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3948 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3949 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3950 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3953 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3956 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3957 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3959 pg_data_t
*pgdat
= NODE_DATA(nid
);
3961 pgdat
->node_id
= nid
;
3962 pgdat
->node_start_pfn
= node_start_pfn
;
3963 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3965 alloc_node_mem_map(pgdat
);
3966 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3967 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3968 nid
, (unsigned long)pgdat
,
3969 (unsigned long)pgdat
->node_mem_map
);
3972 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3975 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3977 #if MAX_NUMNODES > 1
3979 * Figure out the number of possible node ids.
3981 static void __init
setup_nr_node_ids(void)
3984 unsigned int highest
= 0;
3986 for_each_node_mask(node
, node_possible_map
)
3988 nr_node_ids
= highest
+ 1;
3991 static inline void setup_nr_node_ids(void)
3997 * add_active_range - Register a range of PFNs backed by physical memory
3998 * @nid: The node ID the range resides on
3999 * @start_pfn: The start PFN of the available physical memory
4000 * @end_pfn: The end PFN of the available physical memory
4002 * These ranges are stored in an early_node_map[] and later used by
4003 * free_area_init_nodes() to calculate zone sizes and holes. If the
4004 * range spans a memory hole, it is up to the architecture to ensure
4005 * the memory is not freed by the bootmem allocator. If possible
4006 * the range being registered will be merged with existing ranges.
4008 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
4009 unsigned long end_pfn
)
4013 mminit_dprintk(MMINIT_TRACE
, "memory_register",
4014 "Entering add_active_range(%d, %#lx, %#lx) "
4015 "%d entries of %d used\n",
4016 nid
, start_pfn
, end_pfn
,
4017 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
4019 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
4021 /* Merge with existing active regions if possible */
4022 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4023 if (early_node_map
[i
].nid
!= nid
)
4026 /* Skip if an existing region covers this new one */
4027 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
4028 end_pfn
<= early_node_map
[i
].end_pfn
)
4031 /* Merge forward if suitable */
4032 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
4033 end_pfn
> early_node_map
[i
].end_pfn
) {
4034 early_node_map
[i
].end_pfn
= end_pfn
;
4038 /* Merge backward if suitable */
4039 if (start_pfn
< early_node_map
[i
].start_pfn
&&
4040 end_pfn
>= early_node_map
[i
].start_pfn
) {
4041 early_node_map
[i
].start_pfn
= start_pfn
;
4046 /* Check that early_node_map is large enough */
4047 if (i
>= MAX_ACTIVE_REGIONS
) {
4048 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
4049 MAX_ACTIVE_REGIONS
);
4053 early_node_map
[i
].nid
= nid
;
4054 early_node_map
[i
].start_pfn
= start_pfn
;
4055 early_node_map
[i
].end_pfn
= end_pfn
;
4056 nr_nodemap_entries
= i
+ 1;
4060 * remove_active_range - Shrink an existing registered range of PFNs
4061 * @nid: The node id the range is on that should be shrunk
4062 * @start_pfn: The new PFN of the range
4063 * @end_pfn: The new PFN of the range
4065 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
4066 * The map is kept near the end physical page range that has already been
4067 * registered. This function allows an arch to shrink an existing registered
4070 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
4071 unsigned long end_pfn
)
4076 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
4077 nid
, start_pfn
, end_pfn
);
4079 /* Find the old active region end and shrink */
4080 for_each_active_range_index_in_nid(i
, nid
) {
4081 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4082 early_node_map
[i
].end_pfn
<= end_pfn
) {
4084 early_node_map
[i
].start_pfn
= 0;
4085 early_node_map
[i
].end_pfn
= 0;
4089 if (early_node_map
[i
].start_pfn
< start_pfn
&&
4090 early_node_map
[i
].end_pfn
> start_pfn
) {
4091 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
4092 early_node_map
[i
].end_pfn
= start_pfn
;
4093 if (temp_end_pfn
> end_pfn
)
4094 add_active_range(nid
, end_pfn
, temp_end_pfn
);
4097 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4098 early_node_map
[i
].end_pfn
> end_pfn
&&
4099 early_node_map
[i
].start_pfn
< end_pfn
) {
4100 early_node_map
[i
].start_pfn
= end_pfn
;
4108 /* remove the blank ones */
4109 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
4110 if (early_node_map
[i
].nid
!= nid
)
4112 if (early_node_map
[i
].end_pfn
)
4114 /* we found it, get rid of it */
4115 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
4116 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
4117 sizeof(early_node_map
[j
]));
4118 j
= nr_nodemap_entries
- 1;
4119 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
4120 nr_nodemap_entries
--;
4125 * remove_all_active_ranges - Remove all currently registered regions
4127 * During discovery, it may be found that a table like SRAT is invalid
4128 * and an alternative discovery method must be used. This function removes
4129 * all currently registered regions.
4131 void __init
remove_all_active_ranges(void)
4133 memset(early_node_map
, 0, sizeof(early_node_map
));
4134 nr_nodemap_entries
= 0;
4137 /* Compare two active node_active_regions */
4138 static int __init
cmp_node_active_region(const void *a
, const void *b
)
4140 struct node_active_region
*arange
= (struct node_active_region
*)a
;
4141 struct node_active_region
*brange
= (struct node_active_region
*)b
;
4143 /* Done this way to avoid overflows */
4144 if (arange
->start_pfn
> brange
->start_pfn
)
4146 if (arange
->start_pfn
< brange
->start_pfn
)
4152 /* sort the node_map by start_pfn */
4153 void __init
sort_node_map(void)
4155 sort(early_node_map
, (size_t)nr_nodemap_entries
,
4156 sizeof(struct node_active_region
),
4157 cmp_node_active_region
, NULL
);
4160 /* Find the lowest pfn for a node */
4161 static unsigned long __init
find_min_pfn_for_node(int nid
)
4164 unsigned long min_pfn
= ULONG_MAX
;
4166 /* Assuming a sorted map, the first range found has the starting pfn */
4167 for_each_active_range_index_in_nid(i
, nid
)
4168 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
4170 if (min_pfn
== ULONG_MAX
) {
4172 "Could not find start_pfn for node %d\n", nid
);
4180 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4182 * It returns the minimum PFN based on information provided via
4183 * add_active_range().
4185 unsigned long __init
find_min_pfn_with_active_regions(void)
4187 return find_min_pfn_for_node(MAX_NUMNODES
);
4191 * early_calculate_totalpages()
4192 * Sum pages in active regions for movable zone.
4193 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4195 static unsigned long __init
early_calculate_totalpages(void)
4198 unsigned long totalpages
= 0;
4200 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4201 unsigned long pages
= early_node_map
[i
].end_pfn
-
4202 early_node_map
[i
].start_pfn
;
4203 totalpages
+= pages
;
4205 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
4211 * Find the PFN the Movable zone begins in each node. Kernel memory
4212 * is spread evenly between nodes as long as the nodes have enough
4213 * memory. When they don't, some nodes will have more kernelcore than
4216 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
4219 unsigned long usable_startpfn
;
4220 unsigned long kernelcore_node
, kernelcore_remaining
;
4221 /* save the state before borrow the nodemask */
4222 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4223 unsigned long totalpages
= early_calculate_totalpages();
4224 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4227 * If movablecore was specified, calculate what size of
4228 * kernelcore that corresponds so that memory usable for
4229 * any allocation type is evenly spread. If both kernelcore
4230 * and movablecore are specified, then the value of kernelcore
4231 * will be used for required_kernelcore if it's greater than
4232 * what movablecore would have allowed.
4234 if (required_movablecore
) {
4235 unsigned long corepages
;
4238 * Round-up so that ZONE_MOVABLE is at least as large as what
4239 * was requested by the user
4241 required_movablecore
=
4242 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4243 corepages
= totalpages
- required_movablecore
;
4245 required_kernelcore
= max(required_kernelcore
, corepages
);
4248 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4249 if (!required_kernelcore
)
4252 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4253 find_usable_zone_for_movable();
4254 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4257 /* Spread kernelcore memory as evenly as possible throughout nodes */
4258 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4259 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4261 * Recalculate kernelcore_node if the division per node
4262 * now exceeds what is necessary to satisfy the requested
4263 * amount of memory for the kernel
4265 if (required_kernelcore
< kernelcore_node
)
4266 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4269 * As the map is walked, we track how much memory is usable
4270 * by the kernel using kernelcore_remaining. When it is
4271 * 0, the rest of the node is usable by ZONE_MOVABLE
4273 kernelcore_remaining
= kernelcore_node
;
4275 /* Go through each range of PFNs within this node */
4276 for_each_active_range_index_in_nid(i
, nid
) {
4277 unsigned long start_pfn
, end_pfn
;
4278 unsigned long size_pages
;
4280 start_pfn
= max(early_node_map
[i
].start_pfn
,
4281 zone_movable_pfn
[nid
]);
4282 end_pfn
= early_node_map
[i
].end_pfn
;
4283 if (start_pfn
>= end_pfn
)
4286 /* Account for what is only usable for kernelcore */
4287 if (start_pfn
< usable_startpfn
) {
4288 unsigned long kernel_pages
;
4289 kernel_pages
= min(end_pfn
, usable_startpfn
)
4292 kernelcore_remaining
-= min(kernel_pages
,
4293 kernelcore_remaining
);
4294 required_kernelcore
-= min(kernel_pages
,
4295 required_kernelcore
);
4297 /* Continue if range is now fully accounted */
4298 if (end_pfn
<= usable_startpfn
) {
4301 * Push zone_movable_pfn to the end so
4302 * that if we have to rebalance
4303 * kernelcore across nodes, we will
4304 * not double account here
4306 zone_movable_pfn
[nid
] = end_pfn
;
4309 start_pfn
= usable_startpfn
;
4313 * The usable PFN range for ZONE_MOVABLE is from
4314 * start_pfn->end_pfn. Calculate size_pages as the
4315 * number of pages used as kernelcore
4317 size_pages
= end_pfn
- start_pfn
;
4318 if (size_pages
> kernelcore_remaining
)
4319 size_pages
= kernelcore_remaining
;
4320 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4323 * Some kernelcore has been met, update counts and
4324 * break if the kernelcore for this node has been
4327 required_kernelcore
-= min(required_kernelcore
,
4329 kernelcore_remaining
-= size_pages
;
4330 if (!kernelcore_remaining
)
4336 * If there is still required_kernelcore, we do another pass with one
4337 * less node in the count. This will push zone_movable_pfn[nid] further
4338 * along on the nodes that still have memory until kernelcore is
4342 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4345 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4346 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4347 zone_movable_pfn
[nid
] =
4348 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4351 /* restore the node_state */
4352 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4355 /* Any regular memory on that node ? */
4356 static void check_for_regular_memory(pg_data_t
*pgdat
)
4358 #ifdef CONFIG_HIGHMEM
4359 enum zone_type zone_type
;
4361 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4362 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4363 if (zone
->present_pages
)
4364 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4370 * free_area_init_nodes - Initialise all pg_data_t and zone data
4371 * @max_zone_pfn: an array of max PFNs for each zone
4373 * This will call free_area_init_node() for each active node in the system.
4374 * Using the page ranges provided by add_active_range(), the size of each
4375 * zone in each node and their holes is calculated. If the maximum PFN
4376 * between two adjacent zones match, it is assumed that the zone is empty.
4377 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4378 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4379 * starts where the previous one ended. For example, ZONE_DMA32 starts
4380 * at arch_max_dma_pfn.
4382 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4387 /* Sort early_node_map as initialisation assumes it is sorted */
4390 /* Record where the zone boundaries are */
4391 memset(arch_zone_lowest_possible_pfn
, 0,
4392 sizeof(arch_zone_lowest_possible_pfn
));
4393 memset(arch_zone_highest_possible_pfn
, 0,
4394 sizeof(arch_zone_highest_possible_pfn
));
4395 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4396 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4397 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4398 if (i
== ZONE_MOVABLE
)
4400 arch_zone_lowest_possible_pfn
[i
] =
4401 arch_zone_highest_possible_pfn
[i
-1];
4402 arch_zone_highest_possible_pfn
[i
] =
4403 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4405 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4406 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4408 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4409 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4410 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4412 /* Print out the zone ranges */
4413 printk("Zone PFN ranges:\n");
4414 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4415 if (i
== ZONE_MOVABLE
)
4417 printk(" %-8s ", zone_names
[i
]);
4418 if (arch_zone_lowest_possible_pfn
[i
] ==
4419 arch_zone_highest_possible_pfn
[i
])
4422 printk("%0#10lx -> %0#10lx\n",
4423 arch_zone_lowest_possible_pfn
[i
],
4424 arch_zone_highest_possible_pfn
[i
]);
4427 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4428 printk("Movable zone start PFN for each node\n");
4429 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4430 if (zone_movable_pfn
[i
])
4431 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4434 /* Print out the early_node_map[] */
4435 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4436 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4437 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4438 early_node_map
[i
].start_pfn
,
4439 early_node_map
[i
].end_pfn
);
4441 /* Initialise every node */
4442 mminit_verify_pageflags_layout();
4443 setup_nr_node_ids();
4444 for_each_online_node(nid
) {
4445 pg_data_t
*pgdat
= NODE_DATA(nid
);
4446 free_area_init_node(nid
, NULL
,
4447 find_min_pfn_for_node(nid
), NULL
);
4449 /* Any memory on that node */
4450 if (pgdat
->node_present_pages
)
4451 node_set_state(nid
, N_HIGH_MEMORY
);
4452 check_for_regular_memory(pgdat
);
4456 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4458 unsigned long long coremem
;
4462 coremem
= memparse(p
, &p
);
4463 *core
= coremem
>> PAGE_SHIFT
;
4465 /* Paranoid check that UL is enough for the coremem value */
4466 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4472 * kernelcore=size sets the amount of memory for use for allocations that
4473 * cannot be reclaimed or migrated.
4475 static int __init
cmdline_parse_kernelcore(char *p
)
4477 return cmdline_parse_core(p
, &required_kernelcore
);
4481 * movablecore=size sets the amount of memory for use for allocations that
4482 * can be reclaimed or migrated.
4484 static int __init
cmdline_parse_movablecore(char *p
)
4486 return cmdline_parse_core(p
, &required_movablecore
);
4489 early_param("kernelcore", cmdline_parse_kernelcore
);
4490 early_param("movablecore", cmdline_parse_movablecore
);
4492 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4495 * set_dma_reserve - set the specified number of pages reserved in the first zone
4496 * @new_dma_reserve: The number of pages to mark reserved
4498 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4499 * In the DMA zone, a significant percentage may be consumed by kernel image
4500 * and other unfreeable allocations which can skew the watermarks badly. This
4501 * function may optionally be used to account for unfreeable pages in the
4502 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4503 * smaller per-cpu batchsize.
4505 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4507 dma_reserve
= new_dma_reserve
;
4510 #ifndef CONFIG_NEED_MULTIPLE_NODES
4511 struct pglist_data __refdata contig_page_data
= {
4512 #ifndef CONFIG_NO_BOOTMEM
4513 .bdata
= &bootmem_node_data
[0]
4516 EXPORT_SYMBOL(contig_page_data
);
4519 void __init
free_area_init(unsigned long *zones_size
)
4521 free_area_init_node(0, zones_size
,
4522 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4525 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4526 unsigned long action
, void *hcpu
)
4528 int cpu
= (unsigned long)hcpu
;
4530 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4534 * Spill the event counters of the dead processor
4535 * into the current processors event counters.
4536 * This artificially elevates the count of the current
4539 vm_events_fold_cpu(cpu
);
4542 * Zero the differential counters of the dead processor
4543 * so that the vm statistics are consistent.
4545 * This is only okay since the processor is dead and cannot
4546 * race with what we are doing.
4548 refresh_cpu_vm_stats(cpu
);
4553 void __init
page_alloc_init(void)
4555 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4559 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4560 * or min_free_kbytes changes.
4562 static void calculate_totalreserve_pages(void)
4564 struct pglist_data
*pgdat
;
4565 unsigned long reserve_pages
= 0;
4566 enum zone_type i
, j
;
4568 for_each_online_pgdat(pgdat
) {
4569 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4570 struct zone
*zone
= pgdat
->node_zones
+ i
;
4571 unsigned long max
= 0;
4573 /* Find valid and maximum lowmem_reserve in the zone */
4574 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4575 if (zone
->lowmem_reserve
[j
] > max
)
4576 max
= zone
->lowmem_reserve
[j
];
4579 /* we treat the high watermark as reserved pages. */
4580 max
+= high_wmark_pages(zone
);
4582 if (max
> zone
->present_pages
)
4583 max
= zone
->present_pages
;
4584 reserve_pages
+= max
;
4587 totalreserve_pages
= reserve_pages
;
4591 * setup_per_zone_lowmem_reserve - called whenever
4592 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4593 * has a correct pages reserved value, so an adequate number of
4594 * pages are left in the zone after a successful __alloc_pages().
4596 static void setup_per_zone_lowmem_reserve(void)
4598 struct pglist_data
*pgdat
;
4599 enum zone_type j
, idx
;
4601 for_each_online_pgdat(pgdat
) {
4602 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4603 struct zone
*zone
= pgdat
->node_zones
+ j
;
4604 unsigned long present_pages
= zone
->present_pages
;
4606 zone
->lowmem_reserve
[j
] = 0;
4610 struct zone
*lower_zone
;
4614 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4615 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4617 lower_zone
= pgdat
->node_zones
+ idx
;
4618 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4619 sysctl_lowmem_reserve_ratio
[idx
];
4620 present_pages
+= lower_zone
->present_pages
;
4625 /* update totalreserve_pages */
4626 calculate_totalreserve_pages();
4630 * setup_per_zone_wmarks - called when min_free_kbytes changes
4631 * or when memory is hot-{added|removed}
4633 * Ensures that the watermark[min,low,high] values for each zone are set
4634 * correctly with respect to min_free_kbytes.
4636 void setup_per_zone_wmarks(void)
4638 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4639 unsigned long lowmem_pages
= 0;
4641 unsigned long flags
;
4643 /* Calculate total number of !ZONE_HIGHMEM pages */
4644 for_each_zone(zone
) {
4645 if (!is_highmem(zone
))
4646 lowmem_pages
+= zone
->present_pages
;
4649 for_each_zone(zone
) {
4652 spin_lock_irqsave(&zone
->lock
, flags
);
4653 tmp
= (u64
)pages_min
* zone
->present_pages
;
4654 do_div(tmp
, lowmem_pages
);
4655 if (is_highmem(zone
)) {
4657 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4658 * need highmem pages, so cap pages_min to a small
4661 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4662 * deltas controls asynch page reclaim, and so should
4663 * not be capped for highmem.
4667 min_pages
= zone
->present_pages
/ 1024;
4668 if (min_pages
< SWAP_CLUSTER_MAX
)
4669 min_pages
= SWAP_CLUSTER_MAX
;
4670 if (min_pages
> 128)
4672 zone
->watermark
[WMARK_MIN
] = min_pages
;
4675 * If it's a lowmem zone, reserve a number of pages
4676 * proportionate to the zone's size.
4678 zone
->watermark
[WMARK_MIN
] = tmp
;
4681 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
4682 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
4683 setup_zone_migrate_reserve(zone
);
4684 spin_unlock_irqrestore(&zone
->lock
, flags
);
4687 /* update totalreserve_pages */
4688 calculate_totalreserve_pages();
4692 * The inactive anon list should be small enough that the VM never has to
4693 * do too much work, but large enough that each inactive page has a chance
4694 * to be referenced again before it is swapped out.
4696 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4697 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4698 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4699 * the anonymous pages are kept on the inactive list.
4702 * memory ratio inactive anon
4703 * -------------------------------------
4712 void calculate_zone_inactive_ratio(struct zone
*zone
)
4714 unsigned int gb
, ratio
;
4716 /* Zone size in gigabytes */
4717 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4719 ratio
= int_sqrt(10 * gb
);
4723 zone
->inactive_ratio
= ratio
;
4726 static void __init
setup_per_zone_inactive_ratio(void)
4731 calculate_zone_inactive_ratio(zone
);
4735 * Initialise min_free_kbytes.
4737 * For small machines we want it small (128k min). For large machines
4738 * we want it large (64MB max). But it is not linear, because network
4739 * bandwidth does not increase linearly with machine size. We use
4741 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4742 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4758 static int __init
init_per_zone_wmark_min(void)
4760 unsigned long lowmem_kbytes
;
4762 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4764 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4765 if (min_free_kbytes
< 128)
4766 min_free_kbytes
= 128;
4767 if (min_free_kbytes
> 65536)
4768 min_free_kbytes
= 65536;
4769 setup_per_zone_wmarks();
4770 setup_per_zone_lowmem_reserve();
4771 setup_per_zone_inactive_ratio();
4774 module_init(init_per_zone_wmark_min
)
4777 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4778 * that we can call two helper functions whenever min_free_kbytes
4781 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4782 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4784 proc_dointvec(table
, write
, buffer
, length
, ppos
);
4786 setup_per_zone_wmarks();
4791 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4792 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4797 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
4802 zone
->min_unmapped_pages
= (zone
->present_pages
*
4803 sysctl_min_unmapped_ratio
) / 100;
4807 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4808 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4813 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
4818 zone
->min_slab_pages
= (zone
->present_pages
*
4819 sysctl_min_slab_ratio
) / 100;
4825 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4826 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4827 * whenever sysctl_lowmem_reserve_ratio changes.
4829 * The reserve ratio obviously has absolutely no relation with the
4830 * minimum watermarks. The lowmem reserve ratio can only make sense
4831 * if in function of the boot time zone sizes.
4833 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4834 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4836 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
4837 setup_per_zone_lowmem_reserve();
4842 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4843 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4844 * can have before it gets flushed back to buddy allocator.
4847 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4848 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4854 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
4855 if (!write
|| (ret
== -EINVAL
))
4857 for_each_populated_zone(zone
) {
4858 for_each_possible_cpu(cpu
) {
4860 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4861 setup_pagelist_highmark(
4862 per_cpu_ptr(zone
->pageset
, cpu
), high
);
4868 int hashdist
= HASHDIST_DEFAULT
;
4871 static int __init
set_hashdist(char *str
)
4875 hashdist
= simple_strtoul(str
, &str
, 0);
4878 __setup("hashdist=", set_hashdist
);
4882 * allocate a large system hash table from bootmem
4883 * - it is assumed that the hash table must contain an exact power-of-2
4884 * quantity of entries
4885 * - limit is the number of hash buckets, not the total allocation size
4887 void *__init
alloc_large_system_hash(const char *tablename
,
4888 unsigned long bucketsize
,
4889 unsigned long numentries
,
4892 unsigned int *_hash_shift
,
4893 unsigned int *_hash_mask
,
4894 unsigned long limit
)
4896 unsigned long long max
= limit
;
4897 unsigned long log2qty
, size
;
4900 /* allow the kernel cmdline to have a say */
4902 /* round applicable memory size up to nearest megabyte */
4903 numentries
= nr_kernel_pages
;
4904 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4905 numentries
>>= 20 - PAGE_SHIFT
;
4906 numentries
<<= 20 - PAGE_SHIFT
;
4908 /* limit to 1 bucket per 2^scale bytes of low memory */
4909 if (scale
> PAGE_SHIFT
)
4910 numentries
>>= (scale
- PAGE_SHIFT
);
4912 numentries
<<= (PAGE_SHIFT
- scale
);
4914 /* Make sure we've got at least a 0-order allocation.. */
4915 if (unlikely(flags
& HASH_SMALL
)) {
4916 /* Makes no sense without HASH_EARLY */
4917 WARN_ON(!(flags
& HASH_EARLY
));
4918 if (!(numentries
>> *_hash_shift
)) {
4919 numentries
= 1UL << *_hash_shift
;
4920 BUG_ON(!numentries
);
4922 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4923 numentries
= PAGE_SIZE
/ bucketsize
;
4925 numentries
= roundup_pow_of_two(numentries
);
4927 /* limit allocation size to 1/16 total memory by default */
4929 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4930 do_div(max
, bucketsize
);
4933 if (numentries
> max
)
4936 log2qty
= ilog2(numentries
);
4939 size
= bucketsize
<< log2qty
;
4940 if (flags
& HASH_EARLY
)
4941 table
= alloc_bootmem_nopanic(size
);
4943 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4946 * If bucketsize is not a power-of-two, we may free
4947 * some pages at the end of hash table which
4948 * alloc_pages_exact() automatically does
4950 if (get_order(size
) < MAX_ORDER
) {
4951 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
4952 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
4955 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4958 panic("Failed to allocate %s hash table\n", tablename
);
4960 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4963 ilog2(size
) - PAGE_SHIFT
,
4967 *_hash_shift
= log2qty
;
4969 *_hash_mask
= (1 << log2qty
) - 1;
4974 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4975 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4978 #ifdef CONFIG_SPARSEMEM
4979 return __pfn_to_section(pfn
)->pageblock_flags
;
4981 return zone
->pageblock_flags
;
4982 #endif /* CONFIG_SPARSEMEM */
4985 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4987 #ifdef CONFIG_SPARSEMEM
4988 pfn
&= (PAGES_PER_SECTION
-1);
4989 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4991 pfn
= pfn
- zone
->zone_start_pfn
;
4992 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4993 #endif /* CONFIG_SPARSEMEM */
4997 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4998 * @page: The page within the block of interest
4999 * @start_bitidx: The first bit of interest to retrieve
5000 * @end_bitidx: The last bit of interest
5001 * returns pageblock_bits flags
5003 unsigned long get_pageblock_flags_group(struct page
*page
,
5004 int start_bitidx
, int end_bitidx
)
5007 unsigned long *bitmap
;
5008 unsigned long pfn
, bitidx
;
5009 unsigned long flags
= 0;
5010 unsigned long value
= 1;
5012 zone
= page_zone(page
);
5013 pfn
= page_to_pfn(page
);
5014 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5015 bitidx
= pfn_to_bitidx(zone
, pfn
);
5017 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5018 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5025 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5026 * @page: The page within the block of interest
5027 * @start_bitidx: The first bit of interest
5028 * @end_bitidx: The last bit of interest
5029 * @flags: The flags to set
5031 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5032 int start_bitidx
, int end_bitidx
)
5035 unsigned long *bitmap
;
5036 unsigned long pfn
, bitidx
;
5037 unsigned long value
= 1;
5039 zone
= page_zone(page
);
5040 pfn
= page_to_pfn(page
);
5041 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5042 bitidx
= pfn_to_bitidx(zone
, pfn
);
5043 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5044 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5046 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5048 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5050 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5054 * This is designed as sub function...plz see page_isolation.c also.
5055 * set/clear page block's type to be ISOLATE.
5056 * page allocater never alloc memory from ISOLATE block.
5059 int set_migratetype_isolate(struct page
*page
)
5062 struct page
*curr_page
;
5063 unsigned long flags
, pfn
, iter
;
5064 unsigned long immobile
= 0;
5065 struct memory_isolate_notify arg
;
5070 zone
= page_zone(page
);
5071 zone_idx
= zone_idx(zone
);
5073 spin_lock_irqsave(&zone
->lock
, flags
);
5074 if (get_pageblock_migratetype(page
) == MIGRATE_MOVABLE
||
5075 zone_idx
== ZONE_MOVABLE
) {
5080 pfn
= page_to_pfn(page
);
5081 arg
.start_pfn
= pfn
;
5082 arg
.nr_pages
= pageblock_nr_pages
;
5083 arg
.pages_found
= 0;
5086 * It may be possible to isolate a pageblock even if the
5087 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5088 * notifier chain is used by balloon drivers to return the
5089 * number of pages in a range that are held by the balloon
5090 * driver to shrink memory. If all the pages are accounted for
5091 * by balloons, are free, or on the LRU, isolation can continue.
5092 * Later, for example, when memory hotplug notifier runs, these
5093 * pages reported as "can be isolated" should be isolated(freed)
5094 * by the balloon driver through the memory notifier chain.
5096 notifier_ret
= memory_isolate_notify(MEM_ISOLATE_COUNT
, &arg
);
5097 notifier_ret
= notifier_to_errno(notifier_ret
);
5098 if (notifier_ret
|| !arg
.pages_found
)
5101 for (iter
= pfn
; iter
< (pfn
+ pageblock_nr_pages
); iter
++) {
5102 if (!pfn_valid_within(pfn
))
5105 curr_page
= pfn_to_page(iter
);
5106 if (!page_count(curr_page
) || PageLRU(curr_page
))
5112 if (arg
.pages_found
== immobile
)
5117 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
5118 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
5121 spin_unlock_irqrestore(&zone
->lock
, flags
);
5127 void unset_migratetype_isolate(struct page
*page
)
5130 unsigned long flags
;
5131 zone
= page_zone(page
);
5132 spin_lock_irqsave(&zone
->lock
, flags
);
5133 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
5135 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5136 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
5138 spin_unlock_irqrestore(&zone
->lock
, flags
);
5141 #ifdef CONFIG_MEMORY_HOTREMOVE
5143 * All pages in the range must be isolated before calling this.
5146 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
5152 unsigned long flags
;
5153 /* find the first valid pfn */
5154 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
5159 zone
= page_zone(pfn_to_page(pfn
));
5160 spin_lock_irqsave(&zone
->lock
, flags
);
5162 while (pfn
< end_pfn
) {
5163 if (!pfn_valid(pfn
)) {
5167 page
= pfn_to_page(pfn
);
5168 BUG_ON(page_count(page
));
5169 BUG_ON(!PageBuddy(page
));
5170 order
= page_order(page
);
5171 #ifdef CONFIG_DEBUG_VM
5172 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
5173 pfn
, 1 << order
, end_pfn
);
5175 list_del(&page
->lru
);
5176 rmv_page_order(page
);
5177 zone
->free_area
[order
].nr_free
--;
5178 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
5180 for (i
= 0; i
< (1 << order
); i
++)
5181 SetPageReserved((page
+i
));
5182 pfn
+= (1 << order
);
5184 spin_unlock_irqrestore(&zone
->lock
, flags
);
5188 #ifdef CONFIG_MEMORY_FAILURE
5189 bool is_free_buddy_page(struct page
*page
)
5191 struct zone
*zone
= page_zone(page
);
5192 unsigned long pfn
= page_to_pfn(page
);
5193 unsigned long flags
;
5196 spin_lock_irqsave(&zone
->lock
, flags
);
5197 for (order
= 0; order
< MAX_ORDER
; order
++) {
5198 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
5200 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
5203 spin_unlock_irqrestore(&zone
->lock
, flags
);
5205 return order
< MAX_ORDER
;
5209 static struct trace_print_flags pageflag_names
[] = {
5210 {1UL << PG_locked
, "locked" },
5211 {1UL << PG_error
, "error" },
5212 {1UL << PG_referenced
, "referenced" },
5213 {1UL << PG_uptodate
, "uptodate" },
5214 {1UL << PG_dirty
, "dirty" },
5215 {1UL << PG_lru
, "lru" },
5216 {1UL << PG_active
, "active" },
5217 {1UL << PG_slab
, "slab" },
5218 {1UL << PG_owner_priv_1
, "owner_priv_1" },
5219 {1UL << PG_arch_1
, "arch_1" },
5220 {1UL << PG_reserved
, "reserved" },
5221 {1UL << PG_private
, "private" },
5222 {1UL << PG_private_2
, "private_2" },
5223 {1UL << PG_writeback
, "writeback" },
5224 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5225 {1UL << PG_head
, "head" },
5226 {1UL << PG_tail
, "tail" },
5228 {1UL << PG_compound
, "compound" },
5230 {1UL << PG_swapcache
, "swapcache" },
5231 {1UL << PG_mappedtodisk
, "mappedtodisk" },
5232 {1UL << PG_reclaim
, "reclaim" },
5233 {1UL << PG_buddy
, "buddy" },
5234 {1UL << PG_swapbacked
, "swapbacked" },
5235 {1UL << PG_unevictable
, "unevictable" },
5237 {1UL << PG_mlocked
, "mlocked" },
5239 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5240 {1UL << PG_uncached
, "uncached" },
5242 #ifdef CONFIG_MEMORY_FAILURE
5243 {1UL << PG_hwpoison
, "hwpoison" },
5248 static void dump_page_flags(unsigned long flags
)
5250 const char *delim
= "";
5254 printk(KERN_ALERT
"page flags: %#lx(", flags
);
5256 /* remove zone id */
5257 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
5259 for (i
= 0; pageflag_names
[i
].name
&& flags
; i
++) {
5261 mask
= pageflag_names
[i
].mask
;
5262 if ((flags
& mask
) != mask
)
5266 printk("%s%s", delim
, pageflag_names
[i
].name
);
5270 /* check for left over flags */
5272 printk("%s%#lx", delim
, flags
);
5277 void dump_page(struct page
*page
)
5280 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5281 page
, page_count(page
), page_mapcount(page
),
5282 page
->mapping
, page
->index
);
5283 dump_page_flags(page
->flags
);