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 <linux/compaction.h>
53 #include <trace/events/kmem.h>
54 #include <linux/ftrace_event.h>
56 #include <asm/tlbflush.h>
57 #include <asm/div64.h>
60 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
61 DEFINE_PER_CPU(int, numa_node
);
62 EXPORT_PER_CPU_SYMBOL(numa_node
);
65 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
67 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
68 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
69 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
70 * defined in <linux/topology.h>.
72 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
73 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
77 * Array of node states.
79 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
80 [N_POSSIBLE
] = NODE_MASK_ALL
,
81 [N_ONLINE
] = { { [0] = 1UL } },
83 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
85 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
87 [N_CPU
] = { { [0] = 1UL } },
90 EXPORT_SYMBOL(node_states
);
92 unsigned long totalram_pages __read_mostly
;
93 unsigned long totalreserve_pages __read_mostly
;
94 int percpu_pagelist_fraction
;
95 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
97 #ifdef CONFIG_PM_SLEEP
99 * The following functions are used by the suspend/hibernate code to temporarily
100 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
101 * while devices are suspended. To avoid races with the suspend/hibernate code,
102 * they should always be called with pm_mutex held (gfp_allowed_mask also should
103 * only be modified with pm_mutex held, unless the suspend/hibernate code is
104 * guaranteed not to run in parallel with that modification).
106 void set_gfp_allowed_mask(gfp_t mask
)
108 WARN_ON(!mutex_is_locked(&pm_mutex
));
109 gfp_allowed_mask
= mask
;
112 gfp_t
clear_gfp_allowed_mask(gfp_t mask
)
114 gfp_t ret
= gfp_allowed_mask
;
116 WARN_ON(!mutex_is_locked(&pm_mutex
));
117 gfp_allowed_mask
&= ~mask
;
120 #endif /* CONFIG_PM_SLEEP */
122 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
123 int pageblock_order __read_mostly
;
126 static void __free_pages_ok(struct page
*page
, unsigned int order
);
129 * results with 256, 32 in the lowmem_reserve sysctl:
130 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
131 * 1G machine -> (16M dma, 784M normal, 224M high)
132 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
133 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
134 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
136 * TBD: should special case ZONE_DMA32 machines here - in those we normally
137 * don't need any ZONE_NORMAL reservation
139 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
140 #ifdef CONFIG_ZONE_DMA
143 #ifdef CONFIG_ZONE_DMA32
146 #ifdef CONFIG_HIGHMEM
152 EXPORT_SYMBOL(totalram_pages
);
154 static char * const zone_names
[MAX_NR_ZONES
] = {
155 #ifdef CONFIG_ZONE_DMA
158 #ifdef CONFIG_ZONE_DMA32
162 #ifdef CONFIG_HIGHMEM
168 int min_free_kbytes
= 1024;
170 static unsigned long __meminitdata nr_kernel_pages
;
171 static unsigned long __meminitdata nr_all_pages
;
172 static unsigned long __meminitdata dma_reserve
;
174 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
176 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
177 * ranges of memory (RAM) that may be registered with add_active_range().
178 * Ranges passed to add_active_range() will be merged if possible
179 * so the number of times add_active_range() can be called is
180 * related to the number of nodes and the number of holes
182 #ifdef CONFIG_MAX_ACTIVE_REGIONS
183 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
184 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
186 #if MAX_NUMNODES >= 32
187 /* If there can be many nodes, allow up to 50 holes per node */
188 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
190 /* By default, allow up to 256 distinct regions */
191 #define MAX_ACTIVE_REGIONS 256
195 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
196 static int __meminitdata nr_nodemap_entries
;
197 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
198 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
199 static unsigned long __initdata required_kernelcore
;
200 static unsigned long __initdata required_movablecore
;
201 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
203 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
205 EXPORT_SYMBOL(movable_zone
);
206 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
209 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
210 int nr_online_nodes __read_mostly
= 1;
211 EXPORT_SYMBOL(nr_node_ids
);
212 EXPORT_SYMBOL(nr_online_nodes
);
215 int page_group_by_mobility_disabled __read_mostly
;
217 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
220 if (unlikely(page_group_by_mobility_disabled
))
221 migratetype
= MIGRATE_UNMOVABLE
;
223 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
224 PB_migrate
, PB_migrate_end
);
227 bool oom_killer_disabled __read_mostly
;
229 #ifdef CONFIG_DEBUG_VM
230 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
234 unsigned long pfn
= page_to_pfn(page
);
237 seq
= zone_span_seqbegin(zone
);
238 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
240 else if (pfn
< zone
->zone_start_pfn
)
242 } while (zone_span_seqretry(zone
, seq
));
247 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
249 if (!pfn_valid_within(page_to_pfn(page
)))
251 if (zone
!= page_zone(page
))
257 * Temporary debugging check for pages not lying within a given zone.
259 static int bad_range(struct zone
*zone
, struct page
*page
)
261 if (page_outside_zone_boundaries(zone
, page
))
263 if (!page_is_consistent(zone
, page
))
269 static inline int bad_range(struct zone
*zone
, struct page
*page
)
275 static void bad_page(struct page
*page
)
277 static unsigned long resume
;
278 static unsigned long nr_shown
;
279 static unsigned long nr_unshown
;
281 /* Don't complain about poisoned pages */
282 if (PageHWPoison(page
)) {
283 __ClearPageBuddy(page
);
288 * Allow a burst of 60 reports, then keep quiet for that minute;
289 * or allow a steady drip of one report per second.
291 if (nr_shown
== 60) {
292 if (time_before(jiffies
, resume
)) {
298 "BUG: Bad page state: %lu messages suppressed\n",
305 resume
= jiffies
+ 60 * HZ
;
307 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
308 current
->comm
, page_to_pfn(page
));
313 /* Leave bad fields for debug, except PageBuddy could make trouble */
314 __ClearPageBuddy(page
);
315 add_taint(TAINT_BAD_PAGE
);
319 * Higher-order pages are called "compound pages". They are structured thusly:
321 * The first PAGE_SIZE page is called the "head page".
323 * The remaining PAGE_SIZE pages are called "tail pages".
325 * All pages have PG_compound set. All pages have their ->private pointing at
326 * the head page (even the head page has this).
328 * The first tail page's ->lru.next holds the address of the compound page's
329 * put_page() function. Its ->lru.prev holds the order of allocation.
330 * This usage means that zero-order pages may not be compound.
333 static void free_compound_page(struct page
*page
)
335 __free_pages_ok(page
, compound_order(page
));
338 void prep_compound_page(struct page
*page
, unsigned long order
)
341 int nr_pages
= 1 << order
;
343 set_compound_page_dtor(page
, free_compound_page
);
344 set_compound_order(page
, order
);
346 for (i
= 1; i
< nr_pages
; i
++) {
347 struct page
*p
= page
+ i
;
350 p
->first_page
= page
;
354 static int destroy_compound_page(struct page
*page
, unsigned long order
)
357 int nr_pages
= 1 << order
;
360 if (unlikely(compound_order(page
) != order
) ||
361 unlikely(!PageHead(page
))) {
366 __ClearPageHead(page
);
368 for (i
= 1; i
< nr_pages
; i
++) {
369 struct page
*p
= page
+ i
;
371 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
381 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
386 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
387 * and __GFP_HIGHMEM from hard or soft interrupt context.
389 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
390 for (i
= 0; i
< (1 << order
); i
++)
391 clear_highpage(page
+ i
);
394 static inline void set_page_order(struct page
*page
, int order
)
396 set_page_private(page
, order
);
397 __SetPageBuddy(page
);
400 static inline void rmv_page_order(struct page
*page
)
402 __ClearPageBuddy(page
);
403 set_page_private(page
, 0);
407 * Locate the struct page for both the matching buddy in our
408 * pair (buddy1) and the combined O(n+1) page they form (page).
410 * 1) Any buddy B1 will have an order O twin B2 which satisfies
411 * the following equation:
413 * For example, if the starting buddy (buddy2) is #8 its order
415 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
417 * 2) Any buddy B will have an order O+1 parent P which
418 * satisfies the following equation:
421 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
423 static inline struct page
*
424 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
426 unsigned long buddy_idx
= page_idx
^ (1 << order
);
428 return page
+ (buddy_idx
- page_idx
);
431 static inline unsigned long
432 __find_combined_index(unsigned long page_idx
, unsigned int order
)
434 return (page_idx
& ~(1 << order
));
438 * This function checks whether a page is free && is the buddy
439 * we can do coalesce a page and its buddy if
440 * (a) the buddy is not in a hole &&
441 * (b) the buddy is in the buddy system &&
442 * (c) a page and its buddy have the same order &&
443 * (d) a page and its buddy are in the same zone.
445 * For recording whether a page is in the buddy system, we use PG_buddy.
446 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
448 * For recording page's order, we use page_private(page).
450 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
453 if (!pfn_valid_within(page_to_pfn(buddy
)))
456 if (page_zone_id(page
) != page_zone_id(buddy
))
459 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
460 VM_BUG_ON(page_count(buddy
) != 0);
467 * Freeing function for a buddy system allocator.
469 * The concept of a buddy system is to maintain direct-mapped table
470 * (containing bit values) for memory blocks of various "orders".
471 * The bottom level table contains the map for the smallest allocatable
472 * units of memory (here, pages), and each level above it describes
473 * pairs of units from the levels below, hence, "buddies".
474 * At a high level, all that happens here is marking the table entry
475 * at the bottom level available, and propagating the changes upward
476 * as necessary, plus some accounting needed to play nicely with other
477 * parts of the VM system.
478 * At each level, we keep a list of pages, which are heads of continuous
479 * free pages of length of (1 << order) and marked with PG_buddy. Page's
480 * order is recorded in page_private(page) field.
481 * So when we are allocating or freeing one, we can derive the state of the
482 * other. That is, if we allocate a small block, and both were
483 * free, the remainder of the region must be split into blocks.
484 * If a block is freed, and its buddy is also free, then this
485 * triggers coalescing into a block of larger size.
490 static inline void __free_one_page(struct page
*page
,
491 struct zone
*zone
, unsigned int order
,
494 unsigned long page_idx
;
495 unsigned long combined_idx
;
498 if (unlikely(PageCompound(page
)))
499 if (unlikely(destroy_compound_page(page
, order
)))
502 VM_BUG_ON(migratetype
== -1);
504 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
506 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
507 VM_BUG_ON(bad_range(zone
, page
));
509 while (order
< MAX_ORDER
-1) {
510 buddy
= __page_find_buddy(page
, page_idx
, order
);
511 if (!page_is_buddy(page
, buddy
, order
))
514 /* Our buddy is free, merge with it and move up one order. */
515 list_del(&buddy
->lru
);
516 zone
->free_area
[order
].nr_free
--;
517 rmv_page_order(buddy
);
518 combined_idx
= __find_combined_index(page_idx
, order
);
519 page
= page
+ (combined_idx
- page_idx
);
520 page_idx
= combined_idx
;
523 set_page_order(page
, order
);
526 * If this is not the largest possible page, check if the buddy
527 * of the next-highest order is free. If it is, it's possible
528 * that pages are being freed that will coalesce soon. In case,
529 * that is happening, add the free page to the tail of the list
530 * so it's less likely to be used soon and more likely to be merged
531 * as a higher order page
533 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
534 struct page
*higher_page
, *higher_buddy
;
535 combined_idx
= __find_combined_index(page_idx
, order
);
536 higher_page
= page
+ combined_idx
- page_idx
;
537 higher_buddy
= __page_find_buddy(higher_page
, combined_idx
, order
+ 1);
538 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
539 list_add_tail(&page
->lru
,
540 &zone
->free_area
[order
].free_list
[migratetype
]);
545 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
547 zone
->free_area
[order
].nr_free
++;
551 * free_page_mlock() -- clean up attempts to free and mlocked() page.
552 * Page should not be on lru, so no need to fix that up.
553 * free_pages_check() will verify...
555 static inline void free_page_mlock(struct page
*page
)
557 __dec_zone_page_state(page
, NR_MLOCK
);
558 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
561 static inline int free_pages_check(struct page
*page
)
563 if (unlikely(page_mapcount(page
) |
564 (page
->mapping
!= NULL
) |
565 (atomic_read(&page
->_count
) != 0) |
566 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
570 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
571 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
576 * Frees a number of pages from the PCP lists
577 * Assumes all pages on list are in same zone, and of same order.
578 * count is the number of pages to free.
580 * If the zone was previously in an "all pages pinned" state then look to
581 * see if this freeing clears that state.
583 * And clear the zone's pages_scanned counter, to hold off the "all pages are
584 * pinned" detection logic.
586 static void free_pcppages_bulk(struct zone
*zone
, int count
,
587 struct per_cpu_pages
*pcp
)
593 spin_lock(&zone
->lock
);
594 zone
->all_unreclaimable
= 0;
595 zone
->pages_scanned
= 0;
599 struct list_head
*list
;
602 * Remove pages from lists in a round-robin fashion. A
603 * batch_free count is maintained that is incremented when an
604 * empty list is encountered. This is so more pages are freed
605 * off fuller lists instead of spinning excessively around empty
610 if (++migratetype
== MIGRATE_PCPTYPES
)
612 list
= &pcp
->lists
[migratetype
];
613 } while (list_empty(list
));
616 page
= list_entry(list
->prev
, struct page
, lru
);
617 /* must delete as __free_one_page list manipulates */
618 list_del(&page
->lru
);
619 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
620 __free_one_page(page
, zone
, 0, page_private(page
));
621 trace_mm_page_pcpu_drain(page
, 0, page_private(page
));
622 } while (--to_free
&& --batch_free
&& !list_empty(list
));
624 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
);
625 spin_unlock(&zone
->lock
);
628 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
631 spin_lock(&zone
->lock
);
632 zone
->all_unreclaimable
= 0;
633 zone
->pages_scanned
= 0;
635 __free_one_page(page
, zone
, order
, migratetype
);
636 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
637 spin_unlock(&zone
->lock
);
640 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
645 trace_mm_page_free_direct(page
, order
);
646 kmemcheck_free_shadow(page
, order
);
648 for (i
= 0; i
< (1 << order
); i
++) {
649 struct page
*pg
= page
+ i
;
653 bad
+= free_pages_check(pg
);
658 if (!PageHighMem(page
)) {
659 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
660 debug_check_no_obj_freed(page_address(page
),
663 arch_free_page(page
, order
);
664 kernel_map_pages(page
, 1 << order
, 0);
669 static void __free_pages_ok(struct page
*page
, unsigned int order
)
672 int wasMlocked
= __TestClearPageMlocked(page
);
674 if (!free_pages_prepare(page
, order
))
677 local_irq_save(flags
);
678 if (unlikely(wasMlocked
))
679 free_page_mlock(page
);
680 __count_vm_events(PGFREE
, 1 << order
);
681 free_one_page(page_zone(page
), page
, order
,
682 get_pageblock_migratetype(page
));
683 local_irq_restore(flags
);
687 * permit the bootmem allocator to evade page validation on high-order frees
689 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
692 __ClearPageReserved(page
);
693 set_page_count(page
, 0);
694 set_page_refcounted(page
);
700 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
701 struct page
*p
= &page
[loop
];
703 if (loop
+ 1 < BITS_PER_LONG
)
705 __ClearPageReserved(p
);
706 set_page_count(p
, 0);
709 set_page_refcounted(page
);
710 __free_pages(page
, order
);
716 * The order of subdivision here is critical for the IO subsystem.
717 * Please do not alter this order without good reasons and regression
718 * testing. Specifically, as large blocks of memory are subdivided,
719 * the order in which smaller blocks are delivered depends on the order
720 * they're subdivided in this function. This is the primary factor
721 * influencing the order in which pages are delivered to the IO
722 * subsystem according to empirical testing, and this is also justified
723 * by considering the behavior of a buddy system containing a single
724 * large block of memory acted on by a series of small allocations.
725 * This behavior is a critical factor in sglist merging's success.
729 static inline void expand(struct zone
*zone
, struct page
*page
,
730 int low
, int high
, struct free_area
*area
,
733 unsigned long size
= 1 << high
;
739 VM_BUG_ON(bad_range(zone
, &page
[size
]));
740 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
742 set_page_order(&page
[size
], high
);
747 * This page is about to be returned from the page allocator
749 static inline int check_new_page(struct page
*page
)
751 if (unlikely(page_mapcount(page
) |
752 (page
->mapping
!= NULL
) |
753 (atomic_read(&page
->_count
) != 0) |
754 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
761 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
765 for (i
= 0; i
< (1 << order
); i
++) {
766 struct page
*p
= page
+ i
;
767 if (unlikely(check_new_page(p
)))
771 set_page_private(page
, 0);
772 set_page_refcounted(page
);
774 arch_alloc_page(page
, order
);
775 kernel_map_pages(page
, 1 << order
, 1);
777 if (gfp_flags
& __GFP_ZERO
)
778 prep_zero_page(page
, order
, gfp_flags
);
780 if (order
&& (gfp_flags
& __GFP_COMP
))
781 prep_compound_page(page
, order
);
787 * Go through the free lists for the given migratetype and remove
788 * the smallest available page from the freelists
791 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
794 unsigned int current_order
;
795 struct free_area
* area
;
798 /* Find a page of the appropriate size in the preferred list */
799 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
800 area
= &(zone
->free_area
[current_order
]);
801 if (list_empty(&area
->free_list
[migratetype
]))
804 page
= list_entry(area
->free_list
[migratetype
].next
,
806 list_del(&page
->lru
);
807 rmv_page_order(page
);
809 expand(zone
, page
, order
, current_order
, area
, migratetype
);
818 * This array describes the order lists are fallen back to when
819 * the free lists for the desirable migrate type are depleted
821 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
822 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
823 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
824 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
825 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
829 * Move the free pages in a range to the free lists of the requested type.
830 * Note that start_page and end_pages are not aligned on a pageblock
831 * boundary. If alignment is required, use move_freepages_block()
833 static int move_freepages(struct zone
*zone
,
834 struct page
*start_page
, struct page
*end_page
,
841 #ifndef CONFIG_HOLES_IN_ZONE
843 * page_zone is not safe to call in this context when
844 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
845 * anyway as we check zone boundaries in move_freepages_block().
846 * Remove at a later date when no bug reports exist related to
847 * grouping pages by mobility
849 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
852 for (page
= start_page
; page
<= end_page
;) {
853 /* Make sure we are not inadvertently changing nodes */
854 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
856 if (!pfn_valid_within(page_to_pfn(page
))) {
861 if (!PageBuddy(page
)) {
866 order
= page_order(page
);
867 list_del(&page
->lru
);
869 &zone
->free_area
[order
].free_list
[migratetype
]);
871 pages_moved
+= 1 << order
;
877 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
880 unsigned long start_pfn
, end_pfn
;
881 struct page
*start_page
, *end_page
;
883 start_pfn
= page_to_pfn(page
);
884 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
885 start_page
= pfn_to_page(start_pfn
);
886 end_page
= start_page
+ pageblock_nr_pages
- 1;
887 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
889 /* Do not cross zone boundaries */
890 if (start_pfn
< zone
->zone_start_pfn
)
892 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
895 return move_freepages(zone
, start_page
, end_page
, migratetype
);
898 static void change_pageblock_range(struct page
*pageblock_page
,
899 int start_order
, int migratetype
)
901 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
903 while (nr_pageblocks
--) {
904 set_pageblock_migratetype(pageblock_page
, migratetype
);
905 pageblock_page
+= pageblock_nr_pages
;
909 /* Remove an element from the buddy allocator from the fallback list */
910 static inline struct page
*
911 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
913 struct free_area
* area
;
918 /* Find the largest possible block of pages in the other list */
919 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
921 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
922 migratetype
= fallbacks
[start_migratetype
][i
];
924 /* MIGRATE_RESERVE handled later if necessary */
925 if (migratetype
== MIGRATE_RESERVE
)
928 area
= &(zone
->free_area
[current_order
]);
929 if (list_empty(&area
->free_list
[migratetype
]))
932 page
= list_entry(area
->free_list
[migratetype
].next
,
937 * If breaking a large block of pages, move all free
938 * pages to the preferred allocation list. If falling
939 * back for a reclaimable kernel allocation, be more
940 * agressive about taking ownership of free pages
942 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
943 start_migratetype
== MIGRATE_RECLAIMABLE
||
944 page_group_by_mobility_disabled
) {
946 pages
= move_freepages_block(zone
, page
,
949 /* Claim the whole block if over half of it is free */
950 if (pages
>= (1 << (pageblock_order
-1)) ||
951 page_group_by_mobility_disabled
)
952 set_pageblock_migratetype(page
,
955 migratetype
= start_migratetype
;
958 /* Remove the page from the freelists */
959 list_del(&page
->lru
);
960 rmv_page_order(page
);
962 /* Take ownership for orders >= pageblock_order */
963 if (current_order
>= pageblock_order
)
964 change_pageblock_range(page
, current_order
,
967 expand(zone
, page
, order
, current_order
, area
, migratetype
);
969 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
970 start_migratetype
, migratetype
);
980 * Do the hard work of removing an element from the buddy allocator.
981 * Call me with the zone->lock already held.
983 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
989 page
= __rmqueue_smallest(zone
, order
, migratetype
);
991 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
992 page
= __rmqueue_fallback(zone
, order
, migratetype
);
995 * Use MIGRATE_RESERVE rather than fail an allocation. goto
996 * is used because __rmqueue_smallest is an inline function
997 * and we want just one call site
1000 migratetype
= MIGRATE_RESERVE
;
1005 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1010 * Obtain a specified number of elements from the buddy allocator, all under
1011 * a single hold of the lock, for efficiency. Add them to the supplied list.
1012 * Returns the number of new pages which were placed at *list.
1014 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1015 unsigned long count
, struct list_head
*list
,
1016 int migratetype
, int cold
)
1020 spin_lock(&zone
->lock
);
1021 for (i
= 0; i
< count
; ++i
) {
1022 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1023 if (unlikely(page
== NULL
))
1027 * Split buddy pages returned by expand() are received here
1028 * in physical page order. The page is added to the callers and
1029 * list and the list head then moves forward. From the callers
1030 * perspective, the linked list is ordered by page number in
1031 * some conditions. This is useful for IO devices that can
1032 * merge IO requests if the physical pages are ordered
1035 if (likely(cold
== 0))
1036 list_add(&page
->lru
, list
);
1038 list_add_tail(&page
->lru
, list
);
1039 set_page_private(page
, migratetype
);
1042 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1043 spin_unlock(&zone
->lock
);
1049 * Called from the vmstat counter updater to drain pagesets of this
1050 * currently executing processor on remote nodes after they have
1053 * Note that this function must be called with the thread pinned to
1054 * a single processor.
1056 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1058 unsigned long flags
;
1061 local_irq_save(flags
);
1062 if (pcp
->count
>= pcp
->batch
)
1063 to_drain
= pcp
->batch
;
1065 to_drain
= pcp
->count
;
1066 free_pcppages_bulk(zone
, to_drain
, pcp
);
1067 pcp
->count
-= to_drain
;
1068 local_irq_restore(flags
);
1073 * Drain pages of the indicated processor.
1075 * The processor must either be the current processor and the
1076 * thread pinned to the current processor or a processor that
1079 static void drain_pages(unsigned int cpu
)
1081 unsigned long flags
;
1084 for_each_populated_zone(zone
) {
1085 struct per_cpu_pageset
*pset
;
1086 struct per_cpu_pages
*pcp
;
1088 local_irq_save(flags
);
1089 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1092 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1094 local_irq_restore(flags
);
1099 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1101 void drain_local_pages(void *arg
)
1103 drain_pages(smp_processor_id());
1107 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1109 void drain_all_pages(void)
1111 on_each_cpu(drain_local_pages
, NULL
, 1);
1114 #ifdef CONFIG_HIBERNATION
1116 void mark_free_pages(struct zone
*zone
)
1118 unsigned long pfn
, max_zone_pfn
;
1119 unsigned long flags
;
1121 struct list_head
*curr
;
1123 if (!zone
->spanned_pages
)
1126 spin_lock_irqsave(&zone
->lock
, flags
);
1128 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1129 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1130 if (pfn_valid(pfn
)) {
1131 struct page
*page
= pfn_to_page(pfn
);
1133 if (!swsusp_page_is_forbidden(page
))
1134 swsusp_unset_page_free(page
);
1137 for_each_migratetype_order(order
, t
) {
1138 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1141 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1142 for (i
= 0; i
< (1UL << order
); i
++)
1143 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1146 spin_unlock_irqrestore(&zone
->lock
, flags
);
1148 #endif /* CONFIG_PM */
1151 * Free a 0-order page
1152 * cold == 1 ? free a cold page : free a hot page
1154 void free_hot_cold_page(struct page
*page
, int cold
)
1156 struct zone
*zone
= page_zone(page
);
1157 struct per_cpu_pages
*pcp
;
1158 unsigned long flags
;
1160 int wasMlocked
= __TestClearPageMlocked(page
);
1162 if (!free_pages_prepare(page
, 0))
1165 migratetype
= get_pageblock_migratetype(page
);
1166 set_page_private(page
, migratetype
);
1167 local_irq_save(flags
);
1168 if (unlikely(wasMlocked
))
1169 free_page_mlock(page
);
1170 __count_vm_event(PGFREE
);
1173 * We only track unmovable, reclaimable and movable on pcp lists.
1174 * Free ISOLATE pages back to the allocator because they are being
1175 * offlined but treat RESERVE as movable pages so we can get those
1176 * areas back if necessary. Otherwise, we may have to free
1177 * excessively into the page allocator
1179 if (migratetype
>= MIGRATE_PCPTYPES
) {
1180 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1181 free_one_page(zone
, page
, 0, migratetype
);
1184 migratetype
= MIGRATE_MOVABLE
;
1187 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1189 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1191 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1193 if (pcp
->count
>= pcp
->high
) {
1194 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1195 pcp
->count
-= pcp
->batch
;
1199 local_irq_restore(flags
);
1203 * split_page takes a non-compound higher-order page, and splits it into
1204 * n (1<<order) sub-pages: page[0..n]
1205 * Each sub-page must be freed individually.
1207 * Note: this is probably too low level an operation for use in drivers.
1208 * Please consult with lkml before using this in your driver.
1210 void split_page(struct page
*page
, unsigned int order
)
1214 VM_BUG_ON(PageCompound(page
));
1215 VM_BUG_ON(!page_count(page
));
1217 #ifdef CONFIG_KMEMCHECK
1219 * Split shadow pages too, because free(page[0]) would
1220 * otherwise free the whole shadow.
1222 if (kmemcheck_page_is_tracked(page
))
1223 split_page(virt_to_page(page
[0].shadow
), order
);
1226 for (i
= 1; i
< (1 << order
); i
++)
1227 set_page_refcounted(page
+ i
);
1231 * Similar to split_page except the page is already free. As this is only
1232 * being used for migration, the migratetype of the block also changes.
1233 * As this is called with interrupts disabled, the caller is responsible
1234 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1237 * Note: this is probably too low level an operation for use in drivers.
1238 * Please consult with lkml before using this in your driver.
1240 int split_free_page(struct page
*page
)
1243 unsigned long watermark
;
1246 BUG_ON(!PageBuddy(page
));
1248 zone
= page_zone(page
);
1249 order
= page_order(page
);
1251 /* Obey watermarks as if the page was being allocated */
1252 watermark
= low_wmark_pages(zone
) + (1 << order
);
1253 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1256 /* Remove page from free list */
1257 list_del(&page
->lru
);
1258 zone
->free_area
[order
].nr_free
--;
1259 rmv_page_order(page
);
1260 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1UL << order
));
1262 /* Split into individual pages */
1263 set_page_refcounted(page
);
1264 split_page(page
, order
);
1266 if (order
>= pageblock_order
- 1) {
1267 struct page
*endpage
= page
+ (1 << order
) - 1;
1268 for (; page
< endpage
; page
+= pageblock_nr_pages
)
1269 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1276 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1277 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1281 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1282 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1285 unsigned long flags
;
1287 int cold
= !!(gfp_flags
& __GFP_COLD
);
1290 if (likely(order
== 0)) {
1291 struct per_cpu_pages
*pcp
;
1292 struct list_head
*list
;
1294 local_irq_save(flags
);
1295 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1296 list
= &pcp
->lists
[migratetype
];
1297 if (list_empty(list
)) {
1298 pcp
->count
+= rmqueue_bulk(zone
, 0,
1301 if (unlikely(list_empty(list
)))
1306 page
= list_entry(list
->prev
, struct page
, lru
);
1308 page
= list_entry(list
->next
, struct page
, lru
);
1310 list_del(&page
->lru
);
1313 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1315 * __GFP_NOFAIL is not to be used in new code.
1317 * All __GFP_NOFAIL callers should be fixed so that they
1318 * properly detect and handle allocation failures.
1320 * We most definitely don't want callers attempting to
1321 * allocate greater than order-1 page units with
1324 WARN_ON_ONCE(order
> 1);
1326 spin_lock_irqsave(&zone
->lock
, flags
);
1327 page
= __rmqueue(zone
, order
, migratetype
);
1328 spin_unlock(&zone
->lock
);
1331 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << order
));
1334 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1335 zone_statistics(preferred_zone
, zone
);
1336 local_irq_restore(flags
);
1338 VM_BUG_ON(bad_range(zone
, page
));
1339 if (prep_new_page(page
, order
, gfp_flags
))
1344 local_irq_restore(flags
);
1348 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1349 #define ALLOC_WMARK_MIN WMARK_MIN
1350 #define ALLOC_WMARK_LOW WMARK_LOW
1351 #define ALLOC_WMARK_HIGH WMARK_HIGH
1352 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1354 /* Mask to get the watermark bits */
1355 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1357 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1358 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1359 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1361 #ifdef CONFIG_FAIL_PAGE_ALLOC
1363 static struct fail_page_alloc_attr
{
1364 struct fault_attr attr
;
1366 u32 ignore_gfp_highmem
;
1367 u32 ignore_gfp_wait
;
1370 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1372 struct dentry
*ignore_gfp_highmem_file
;
1373 struct dentry
*ignore_gfp_wait_file
;
1374 struct dentry
*min_order_file
;
1376 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1378 } fail_page_alloc
= {
1379 .attr
= FAULT_ATTR_INITIALIZER
,
1380 .ignore_gfp_wait
= 1,
1381 .ignore_gfp_highmem
= 1,
1385 static int __init
setup_fail_page_alloc(char *str
)
1387 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1389 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1391 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1393 if (order
< fail_page_alloc
.min_order
)
1395 if (gfp_mask
& __GFP_NOFAIL
)
1397 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1399 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1402 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1405 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1407 static int __init
fail_page_alloc_debugfs(void)
1409 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1413 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1417 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1419 fail_page_alloc
.ignore_gfp_wait_file
=
1420 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1421 &fail_page_alloc
.ignore_gfp_wait
);
1423 fail_page_alloc
.ignore_gfp_highmem_file
=
1424 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1425 &fail_page_alloc
.ignore_gfp_highmem
);
1426 fail_page_alloc
.min_order_file
=
1427 debugfs_create_u32("min-order", mode
, dir
,
1428 &fail_page_alloc
.min_order
);
1430 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1431 !fail_page_alloc
.ignore_gfp_highmem_file
||
1432 !fail_page_alloc
.min_order_file
) {
1434 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1435 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1436 debugfs_remove(fail_page_alloc
.min_order_file
);
1437 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1443 late_initcall(fail_page_alloc_debugfs
);
1445 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1447 #else /* CONFIG_FAIL_PAGE_ALLOC */
1449 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1454 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1457 * Return 1 if free pages are above 'mark'. This takes into account the order
1458 * of the allocation.
1460 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1461 int classzone_idx
, int alloc_flags
)
1463 /* free_pages my go negative - that's OK */
1465 long free_pages
= zone_nr_free_pages(z
) - (1 << order
) + 1;
1468 if (alloc_flags
& ALLOC_HIGH
)
1470 if (alloc_flags
& ALLOC_HARDER
)
1473 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1475 for (o
= 0; o
< order
; o
++) {
1476 /* At the next order, this order's pages become unavailable */
1477 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1479 /* Require fewer higher order pages to be free */
1482 if (free_pages
<= min
)
1490 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1491 * skip over zones that are not allowed by the cpuset, or that have
1492 * been recently (in last second) found to be nearly full. See further
1493 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1494 * that have to skip over a lot of full or unallowed zones.
1496 * If the zonelist cache is present in the passed in zonelist, then
1497 * returns a pointer to the allowed node mask (either the current
1498 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1500 * If the zonelist cache is not available for this zonelist, does
1501 * nothing and returns NULL.
1503 * If the fullzones BITMAP in the zonelist cache is stale (more than
1504 * a second since last zap'd) then we zap it out (clear its bits.)
1506 * We hold off even calling zlc_setup, until after we've checked the
1507 * first zone in the zonelist, on the theory that most allocations will
1508 * be satisfied from that first zone, so best to examine that zone as
1509 * quickly as we can.
1511 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1513 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1514 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1516 zlc
= zonelist
->zlcache_ptr
;
1520 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1521 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1522 zlc
->last_full_zap
= jiffies
;
1525 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1526 &cpuset_current_mems_allowed
:
1527 &node_states
[N_HIGH_MEMORY
];
1528 return allowednodes
;
1532 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1533 * if it is worth looking at further for free memory:
1534 * 1) Check that the zone isn't thought to be full (doesn't have its
1535 * bit set in the zonelist_cache fullzones BITMAP).
1536 * 2) Check that the zones node (obtained from the zonelist_cache
1537 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1538 * Return true (non-zero) if zone is worth looking at further, or
1539 * else return false (zero) if it is not.
1541 * This check -ignores- the distinction between various watermarks,
1542 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1543 * found to be full for any variation of these watermarks, it will
1544 * be considered full for up to one second by all requests, unless
1545 * we are so low on memory on all allowed nodes that we are forced
1546 * into the second scan of the zonelist.
1548 * In the second scan we ignore this zonelist cache and exactly
1549 * apply the watermarks to all zones, even it is slower to do so.
1550 * We are low on memory in the second scan, and should leave no stone
1551 * unturned looking for a free page.
1553 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1554 nodemask_t
*allowednodes
)
1556 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1557 int i
; /* index of *z in zonelist zones */
1558 int n
; /* node that zone *z is on */
1560 zlc
= zonelist
->zlcache_ptr
;
1564 i
= z
- zonelist
->_zonerefs
;
1567 /* This zone is worth trying if it is allowed but not full */
1568 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1572 * Given 'z' scanning a zonelist, set the corresponding bit in
1573 * zlc->fullzones, so that subsequent attempts to allocate a page
1574 * from that zone don't waste time re-examining it.
1576 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1578 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1579 int i
; /* index of *z in zonelist zones */
1581 zlc
= zonelist
->zlcache_ptr
;
1585 i
= z
- zonelist
->_zonerefs
;
1587 set_bit(i
, zlc
->fullzones
);
1590 #else /* CONFIG_NUMA */
1592 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1597 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1598 nodemask_t
*allowednodes
)
1603 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1606 #endif /* CONFIG_NUMA */
1609 * get_page_from_freelist goes through the zonelist trying to allocate
1612 static struct page
*
1613 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1614 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1615 struct zone
*preferred_zone
, int migratetype
)
1618 struct page
*page
= NULL
;
1621 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1622 int zlc_active
= 0; /* set if using zonelist_cache */
1623 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1625 classzone_idx
= zone_idx(preferred_zone
);
1628 * Scan zonelist, looking for a zone with enough free.
1629 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1631 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1632 high_zoneidx
, nodemask
) {
1633 if (NUMA_BUILD
&& zlc_active
&&
1634 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1636 if ((alloc_flags
& ALLOC_CPUSET
) &&
1637 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1640 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1641 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1645 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1646 if (zone_watermark_ok(zone
, order
, mark
,
1647 classzone_idx
, alloc_flags
))
1650 if (zone_reclaim_mode
== 0)
1651 goto this_zone_full
;
1653 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1655 case ZONE_RECLAIM_NOSCAN
:
1658 case ZONE_RECLAIM_FULL
:
1659 /* scanned but unreclaimable */
1660 goto this_zone_full
;
1662 /* did we reclaim enough */
1663 if (!zone_watermark_ok(zone
, order
, mark
,
1664 classzone_idx
, alloc_flags
))
1665 goto this_zone_full
;
1670 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1671 gfp_mask
, migratetype
);
1676 zlc_mark_zone_full(zonelist
, z
);
1678 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1680 * we do zlc_setup after the first zone is tried but only
1681 * if there are multiple nodes make it worthwhile
1683 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1689 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1690 /* Disable zlc cache for second zonelist scan */
1698 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1699 unsigned long pages_reclaimed
)
1701 /* Do not loop if specifically requested */
1702 if (gfp_mask
& __GFP_NORETRY
)
1706 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1707 * means __GFP_NOFAIL, but that may not be true in other
1710 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1714 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1715 * specified, then we retry until we no longer reclaim any pages
1716 * (above), or we've reclaimed an order of pages at least as
1717 * large as the allocation's order. In both cases, if the
1718 * allocation still fails, we stop retrying.
1720 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1724 * Don't let big-order allocations loop unless the caller
1725 * explicitly requests that.
1727 if (gfp_mask
& __GFP_NOFAIL
)
1733 static inline struct page
*
1734 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1735 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1736 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1741 /* Acquire the OOM killer lock for the zones in zonelist */
1742 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1743 schedule_timeout_uninterruptible(1);
1748 * Go through the zonelist yet one more time, keep very high watermark
1749 * here, this is only to catch a parallel oom killing, we must fail if
1750 * we're still under heavy pressure.
1752 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1753 order
, zonelist
, high_zoneidx
,
1754 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1755 preferred_zone
, migratetype
);
1759 if (!(gfp_mask
& __GFP_NOFAIL
)) {
1760 /* The OOM killer will not help higher order allocs */
1761 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1764 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1765 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1766 * The caller should handle page allocation failure by itself if
1767 * it specifies __GFP_THISNODE.
1768 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1770 if (gfp_mask
& __GFP_THISNODE
)
1773 /* Exhausted what can be done so it's blamo time */
1774 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
);
1777 clear_zonelist_oom(zonelist
, gfp_mask
);
1781 #ifdef CONFIG_COMPACTION
1782 /* Try memory compaction for high-order allocations before reclaim */
1783 static struct page
*
1784 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1785 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1786 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1787 int migratetype
, unsigned long *did_some_progress
)
1791 if (!order
|| compaction_deferred(preferred_zone
))
1794 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
1796 if (*did_some_progress
!= COMPACT_SKIPPED
) {
1798 /* Page migration frees to the PCP lists but we want merging */
1799 drain_pages(get_cpu());
1802 page
= get_page_from_freelist(gfp_mask
, nodemask
,
1803 order
, zonelist
, high_zoneidx
,
1804 alloc_flags
, preferred_zone
,
1807 preferred_zone
->compact_considered
= 0;
1808 preferred_zone
->compact_defer_shift
= 0;
1809 count_vm_event(COMPACTSUCCESS
);
1814 * It's bad if compaction run occurs and fails.
1815 * The most likely reason is that pages exist,
1816 * but not enough to satisfy watermarks.
1818 count_vm_event(COMPACTFAIL
);
1819 defer_compaction(preferred_zone
);
1827 static inline struct page
*
1828 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1829 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1830 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1831 int migratetype
, unsigned long *did_some_progress
)
1835 #endif /* CONFIG_COMPACTION */
1837 /* The really slow allocator path where we enter direct reclaim */
1838 static inline struct page
*
1839 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1840 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1841 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1842 int migratetype
, unsigned long *did_some_progress
)
1844 struct page
*page
= NULL
;
1845 struct reclaim_state reclaim_state
;
1846 struct task_struct
*p
= current
;
1847 bool drained
= false;
1851 /* We now go into synchronous reclaim */
1852 cpuset_memory_pressure_bump();
1853 p
->flags
|= PF_MEMALLOC
;
1854 lockdep_set_current_reclaim_state(gfp_mask
);
1855 reclaim_state
.reclaimed_slab
= 0;
1856 p
->reclaim_state
= &reclaim_state
;
1858 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1860 p
->reclaim_state
= NULL
;
1861 lockdep_clear_current_reclaim_state();
1862 p
->flags
&= ~PF_MEMALLOC
;
1866 if (unlikely(!(*did_some_progress
)))
1870 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1871 zonelist
, high_zoneidx
,
1872 alloc_flags
, preferred_zone
,
1876 * If an allocation failed after direct reclaim, it could be because
1877 * pages are pinned on the per-cpu lists. Drain them and try again
1879 if (!page
&& !drained
) {
1889 * This is called in the allocator slow-path if the allocation request is of
1890 * sufficient urgency to ignore watermarks and take other desperate measures
1892 static inline struct page
*
1893 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1894 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1895 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1901 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1902 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1903 preferred_zone
, migratetype
);
1905 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1906 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1907 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1913 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1914 enum zone_type high_zoneidx
)
1919 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1920 wakeup_kswapd(zone
, order
);
1924 gfp_to_alloc_flags(gfp_t gfp_mask
)
1926 struct task_struct
*p
= current
;
1927 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
1928 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1930 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1931 BUILD_BUG_ON(__GFP_HIGH
!= ALLOC_HIGH
);
1934 * The caller may dip into page reserves a bit more if the caller
1935 * cannot run direct reclaim, or if the caller has realtime scheduling
1936 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1937 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1939 alloc_flags
|= (gfp_mask
& __GFP_HIGH
);
1942 alloc_flags
|= ALLOC_HARDER
;
1944 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1945 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1947 alloc_flags
&= ~ALLOC_CPUSET
;
1948 } else if (unlikely(rt_task(p
)) && !in_interrupt())
1949 alloc_flags
|= ALLOC_HARDER
;
1951 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
1952 if (!in_interrupt() &&
1953 ((p
->flags
& PF_MEMALLOC
) ||
1954 unlikely(test_thread_flag(TIF_MEMDIE
))))
1955 alloc_flags
|= ALLOC_NO_WATERMARKS
;
1961 static inline struct page
*
1962 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
1963 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1964 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1967 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1968 struct page
*page
= NULL
;
1970 unsigned long pages_reclaimed
= 0;
1971 unsigned long did_some_progress
;
1972 struct task_struct
*p
= current
;
1975 * In the slowpath, we sanity check order to avoid ever trying to
1976 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1977 * be using allocators in order of preference for an area that is
1980 if (order
>= MAX_ORDER
) {
1981 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
1986 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1987 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1988 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1989 * using a larger set of nodes after it has established that the
1990 * allowed per node queues are empty and that nodes are
1993 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1997 wake_all_kswapd(order
, zonelist
, high_zoneidx
);
2000 * OK, we're below the kswapd watermark and have kicked background
2001 * reclaim. Now things get more complex, so set up alloc_flags according
2002 * to how we want to proceed.
2004 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2006 /* This is the last chance, in general, before the goto nopage. */
2007 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2008 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2009 preferred_zone
, migratetype
);
2014 /* Allocate without watermarks if the context allows */
2015 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2016 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2017 zonelist
, high_zoneidx
, nodemask
,
2018 preferred_zone
, migratetype
);
2023 /* Atomic allocations - we can't balance anything */
2027 /* Avoid recursion of direct reclaim */
2028 if (p
->flags
& PF_MEMALLOC
)
2031 /* Avoid allocations with no watermarks from looping endlessly */
2032 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2035 /* Try direct compaction */
2036 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2037 zonelist
, high_zoneidx
,
2039 alloc_flags
, preferred_zone
,
2040 migratetype
, &did_some_progress
);
2044 /* Try direct reclaim and then allocating */
2045 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2046 zonelist
, high_zoneidx
,
2048 alloc_flags
, preferred_zone
,
2049 migratetype
, &did_some_progress
);
2054 * If we failed to make any progress reclaiming, then we are
2055 * running out of options and have to consider going OOM
2057 if (!did_some_progress
) {
2058 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2059 if (oom_killer_disabled
)
2061 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2062 zonelist
, high_zoneidx
,
2063 nodemask
, preferred_zone
,
2069 * The OOM killer does not trigger for high-order
2070 * ~__GFP_NOFAIL allocations so if no progress is being
2071 * made, there are no other options and retrying is
2074 if (order
> PAGE_ALLOC_COSTLY_ORDER
&&
2075 !(gfp_mask
& __GFP_NOFAIL
))
2082 /* Check if we should retry the allocation */
2083 pages_reclaimed
+= did_some_progress
;
2084 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
2085 /* Wait for some write requests to complete then retry */
2086 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
2091 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
2092 printk(KERN_WARNING
"%s: page allocation failure."
2093 " order:%d, mode:0x%x\n",
2094 p
->comm
, order
, gfp_mask
);
2100 if (kmemcheck_enabled
)
2101 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2107 * This is the 'heart' of the zoned buddy allocator.
2110 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2111 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2113 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2114 struct zone
*preferred_zone
;
2116 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2118 gfp_mask
&= gfp_allowed_mask
;
2120 lockdep_trace_alloc(gfp_mask
);
2122 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2124 if (should_fail_alloc_page(gfp_mask
, order
))
2128 * Check the zones suitable for the gfp_mask contain at least one
2129 * valid zone. It's possible to have an empty zonelist as a result
2130 * of GFP_THISNODE and a memoryless node
2132 if (unlikely(!zonelist
->_zonerefs
->zone
))
2136 /* The preferred zone is used for statistics later */
2137 first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
, &preferred_zone
);
2138 if (!preferred_zone
) {
2143 /* First allocation attempt */
2144 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2145 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
2146 preferred_zone
, migratetype
);
2147 if (unlikely(!page
))
2148 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2149 zonelist
, high_zoneidx
, nodemask
,
2150 preferred_zone
, migratetype
);
2153 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2156 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2159 * Common helper functions.
2161 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2166 * __get_free_pages() returns a 32-bit address, which cannot represent
2169 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2171 page
= alloc_pages(gfp_mask
, order
);
2174 return (unsigned long) page_address(page
);
2176 EXPORT_SYMBOL(__get_free_pages
);
2178 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2180 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2182 EXPORT_SYMBOL(get_zeroed_page
);
2184 void __pagevec_free(struct pagevec
*pvec
)
2186 int i
= pagevec_count(pvec
);
2189 trace_mm_pagevec_free(pvec
->pages
[i
], pvec
->cold
);
2190 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
2194 void __free_pages(struct page
*page
, unsigned int order
)
2196 if (put_page_testzero(page
)) {
2198 free_hot_cold_page(page
, 0);
2200 __free_pages_ok(page
, order
);
2204 EXPORT_SYMBOL(__free_pages
);
2206 void free_pages(unsigned long addr
, unsigned int order
)
2209 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2210 __free_pages(virt_to_page((void *)addr
), order
);
2214 EXPORT_SYMBOL(free_pages
);
2217 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2218 * @size: the number of bytes to allocate
2219 * @gfp_mask: GFP flags for the allocation
2221 * This function is similar to alloc_pages(), except that it allocates the
2222 * minimum number of pages to satisfy the request. alloc_pages() can only
2223 * allocate memory in power-of-two pages.
2225 * This function is also limited by MAX_ORDER.
2227 * Memory allocated by this function must be released by free_pages_exact().
2229 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2231 unsigned int order
= get_order(size
);
2234 addr
= __get_free_pages(gfp_mask
, order
);
2236 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2237 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2239 split_page(virt_to_page((void *)addr
), order
);
2240 while (used
< alloc_end
) {
2246 return (void *)addr
;
2248 EXPORT_SYMBOL(alloc_pages_exact
);
2251 * free_pages_exact - release memory allocated via alloc_pages_exact()
2252 * @virt: the value returned by alloc_pages_exact.
2253 * @size: size of allocation, same value as passed to alloc_pages_exact().
2255 * Release the memory allocated by a previous call to alloc_pages_exact.
2257 void free_pages_exact(void *virt
, size_t size
)
2259 unsigned long addr
= (unsigned long)virt
;
2260 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2262 while (addr
< end
) {
2267 EXPORT_SYMBOL(free_pages_exact
);
2269 static unsigned int nr_free_zone_pages(int offset
)
2274 /* Just pick one node, since fallback list is circular */
2275 unsigned int sum
= 0;
2277 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2279 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2280 unsigned long size
= zone
->present_pages
;
2281 unsigned long high
= high_wmark_pages(zone
);
2290 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2292 unsigned int nr_free_buffer_pages(void)
2294 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2296 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2299 * Amount of free RAM allocatable within all zones
2301 unsigned int nr_free_pagecache_pages(void)
2303 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2306 static inline void show_node(struct zone
*zone
)
2309 printk("Node %d ", zone_to_nid(zone
));
2312 void si_meminfo(struct sysinfo
*val
)
2314 val
->totalram
= totalram_pages
;
2316 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2317 val
->bufferram
= nr_blockdev_pages();
2318 val
->totalhigh
= totalhigh_pages
;
2319 val
->freehigh
= nr_free_highpages();
2320 val
->mem_unit
= PAGE_SIZE
;
2323 EXPORT_SYMBOL(si_meminfo
);
2326 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2328 pg_data_t
*pgdat
= NODE_DATA(nid
);
2330 val
->totalram
= pgdat
->node_present_pages
;
2331 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2332 #ifdef CONFIG_HIGHMEM
2333 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2334 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2340 val
->mem_unit
= PAGE_SIZE
;
2344 #define K(x) ((x) << (PAGE_SHIFT-10))
2347 * Show free area list (used inside shift_scroll-lock stuff)
2348 * We also calculate the percentage fragmentation. We do this by counting the
2349 * memory on each free list with the exception of the first item on the list.
2351 void show_free_areas(void)
2356 for_each_populated_zone(zone
) {
2358 printk("%s per-cpu:\n", zone
->name
);
2360 for_each_online_cpu(cpu
) {
2361 struct per_cpu_pageset
*pageset
;
2363 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2365 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2366 cpu
, pageset
->pcp
.high
,
2367 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2371 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2372 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2374 " dirty:%lu writeback:%lu unstable:%lu\n"
2375 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2376 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2377 global_page_state(NR_ACTIVE_ANON
),
2378 global_page_state(NR_INACTIVE_ANON
),
2379 global_page_state(NR_ISOLATED_ANON
),
2380 global_page_state(NR_ACTIVE_FILE
),
2381 global_page_state(NR_INACTIVE_FILE
),
2382 global_page_state(NR_ISOLATED_FILE
),
2383 global_page_state(NR_UNEVICTABLE
),
2384 global_page_state(NR_FILE_DIRTY
),
2385 global_page_state(NR_WRITEBACK
),
2386 global_page_state(NR_UNSTABLE_NFS
),
2387 global_page_state(NR_FREE_PAGES
),
2388 global_page_state(NR_SLAB_RECLAIMABLE
),
2389 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2390 global_page_state(NR_FILE_MAPPED
),
2391 global_page_state(NR_SHMEM
),
2392 global_page_state(NR_PAGETABLE
),
2393 global_page_state(NR_BOUNCE
));
2395 for_each_populated_zone(zone
) {
2404 " active_anon:%lukB"
2405 " inactive_anon:%lukB"
2406 " active_file:%lukB"
2407 " inactive_file:%lukB"
2408 " unevictable:%lukB"
2409 " isolated(anon):%lukB"
2410 " isolated(file):%lukB"
2417 " slab_reclaimable:%lukB"
2418 " slab_unreclaimable:%lukB"
2419 " kernel_stack:%lukB"
2423 " writeback_tmp:%lukB"
2424 " pages_scanned:%lu"
2425 " all_unreclaimable? %s"
2428 K(zone_nr_free_pages(zone
)),
2429 K(min_wmark_pages(zone
)),
2430 K(low_wmark_pages(zone
)),
2431 K(high_wmark_pages(zone
)),
2432 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2433 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2434 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2435 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2436 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2437 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
2438 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
2439 K(zone
->present_pages
),
2440 K(zone_page_state(zone
, NR_MLOCK
)),
2441 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
2442 K(zone_page_state(zone
, NR_WRITEBACK
)),
2443 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
2444 K(zone_page_state(zone
, NR_SHMEM
)),
2445 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
2446 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
2447 zone_page_state(zone
, NR_KERNEL_STACK
) *
2449 K(zone_page_state(zone
, NR_PAGETABLE
)),
2450 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
2451 K(zone_page_state(zone
, NR_BOUNCE
)),
2452 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
2453 zone
->pages_scanned
,
2454 (zone
->all_unreclaimable
? "yes" : "no")
2456 printk("lowmem_reserve[]:");
2457 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2458 printk(" %lu", zone
->lowmem_reserve
[i
]);
2462 for_each_populated_zone(zone
) {
2463 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2466 printk("%s: ", zone
->name
);
2468 spin_lock_irqsave(&zone
->lock
, flags
);
2469 for (order
= 0; order
< MAX_ORDER
; order
++) {
2470 nr
[order
] = zone
->free_area
[order
].nr_free
;
2471 total
+= nr
[order
] << order
;
2473 spin_unlock_irqrestore(&zone
->lock
, flags
);
2474 for (order
= 0; order
< MAX_ORDER
; order
++)
2475 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2476 printk("= %lukB\n", K(total
));
2479 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2481 show_swap_cache_info();
2484 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2486 zoneref
->zone
= zone
;
2487 zoneref
->zone_idx
= zone_idx(zone
);
2491 * Builds allocation fallback zone lists.
2493 * Add all populated zones of a node to the zonelist.
2495 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2496 int nr_zones
, enum zone_type zone_type
)
2500 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2505 zone
= pgdat
->node_zones
+ zone_type
;
2506 if (populated_zone(zone
)) {
2507 zoneref_set_zone(zone
,
2508 &zonelist
->_zonerefs
[nr_zones
++]);
2509 check_highest_zone(zone_type
);
2512 } while (zone_type
);
2519 * 0 = automatic detection of better ordering.
2520 * 1 = order by ([node] distance, -zonetype)
2521 * 2 = order by (-zonetype, [node] distance)
2523 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2524 * the same zonelist. So only NUMA can configure this param.
2526 #define ZONELIST_ORDER_DEFAULT 0
2527 #define ZONELIST_ORDER_NODE 1
2528 #define ZONELIST_ORDER_ZONE 2
2530 /* zonelist order in the kernel.
2531 * set_zonelist_order() will set this to NODE or ZONE.
2533 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2534 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2538 /* The value user specified ....changed by config */
2539 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2540 /* string for sysctl */
2541 #define NUMA_ZONELIST_ORDER_LEN 16
2542 char numa_zonelist_order
[16] = "default";
2545 * interface for configure zonelist ordering.
2546 * command line option "numa_zonelist_order"
2547 * = "[dD]efault - default, automatic configuration.
2548 * = "[nN]ode - order by node locality, then by zone within node
2549 * = "[zZ]one - order by zone, then by locality within zone
2552 static int __parse_numa_zonelist_order(char *s
)
2554 if (*s
== 'd' || *s
== 'D') {
2555 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2556 } else if (*s
== 'n' || *s
== 'N') {
2557 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2558 } else if (*s
== 'z' || *s
== 'Z') {
2559 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2562 "Ignoring invalid numa_zonelist_order value: "
2569 static __init
int setup_numa_zonelist_order(char *s
)
2572 return __parse_numa_zonelist_order(s
);
2575 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2578 * sysctl handler for numa_zonelist_order
2580 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2581 void __user
*buffer
, size_t *length
,
2584 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2586 static DEFINE_MUTEX(zl_order_mutex
);
2588 mutex_lock(&zl_order_mutex
);
2590 strcpy(saved_string
, (char*)table
->data
);
2591 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
2595 int oldval
= user_zonelist_order
;
2596 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2598 * bogus value. restore saved string
2600 strncpy((char*)table
->data
, saved_string
,
2601 NUMA_ZONELIST_ORDER_LEN
);
2602 user_zonelist_order
= oldval
;
2603 } else if (oldval
!= user_zonelist_order
) {
2604 mutex_lock(&zonelists_mutex
);
2605 build_all_zonelists(NULL
);
2606 mutex_unlock(&zonelists_mutex
);
2610 mutex_unlock(&zl_order_mutex
);
2615 #define MAX_NODE_LOAD (nr_online_nodes)
2616 static int node_load
[MAX_NUMNODES
];
2619 * find_next_best_node - find the next node that should appear in a given node's fallback list
2620 * @node: node whose fallback list we're appending
2621 * @used_node_mask: nodemask_t of already used nodes
2623 * We use a number of factors to determine which is the next node that should
2624 * appear on a given node's fallback list. The node should not have appeared
2625 * already in @node's fallback list, and it should be the next closest node
2626 * according to the distance array (which contains arbitrary distance values
2627 * from each node to each node in the system), and should also prefer nodes
2628 * with no CPUs, since presumably they'll have very little allocation pressure
2629 * on them otherwise.
2630 * It returns -1 if no node is found.
2632 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2635 int min_val
= INT_MAX
;
2637 const struct cpumask
*tmp
= cpumask_of_node(0);
2639 /* Use the local node if we haven't already */
2640 if (!node_isset(node
, *used_node_mask
)) {
2641 node_set(node
, *used_node_mask
);
2645 for_each_node_state(n
, N_HIGH_MEMORY
) {
2647 /* Don't want a node to appear more than once */
2648 if (node_isset(n
, *used_node_mask
))
2651 /* Use the distance array to find the distance */
2652 val
= node_distance(node
, n
);
2654 /* Penalize nodes under us ("prefer the next node") */
2657 /* Give preference to headless and unused nodes */
2658 tmp
= cpumask_of_node(n
);
2659 if (!cpumask_empty(tmp
))
2660 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2662 /* Slight preference for less loaded node */
2663 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2664 val
+= node_load
[n
];
2666 if (val
< min_val
) {
2673 node_set(best_node
, *used_node_mask
);
2680 * Build zonelists ordered by node and zones within node.
2681 * This results in maximum locality--normal zone overflows into local
2682 * DMA zone, if any--but risks exhausting DMA zone.
2684 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2687 struct zonelist
*zonelist
;
2689 zonelist
= &pgdat
->node_zonelists
[0];
2690 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2692 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2694 zonelist
->_zonerefs
[j
].zone
= NULL
;
2695 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2699 * Build gfp_thisnode zonelists
2701 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2704 struct zonelist
*zonelist
;
2706 zonelist
= &pgdat
->node_zonelists
[1];
2707 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2708 zonelist
->_zonerefs
[j
].zone
= NULL
;
2709 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2713 * Build zonelists ordered by zone and nodes within zones.
2714 * This results in conserving DMA zone[s] until all Normal memory is
2715 * exhausted, but results in overflowing to remote node while memory
2716 * may still exist in local DMA zone.
2718 static int node_order
[MAX_NUMNODES
];
2720 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2723 int zone_type
; /* needs to be signed */
2725 struct zonelist
*zonelist
;
2727 zonelist
= &pgdat
->node_zonelists
[0];
2729 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2730 for (j
= 0; j
< nr_nodes
; j
++) {
2731 node
= node_order
[j
];
2732 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2733 if (populated_zone(z
)) {
2735 &zonelist
->_zonerefs
[pos
++]);
2736 check_highest_zone(zone_type
);
2740 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2741 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2744 static int default_zonelist_order(void)
2747 unsigned long low_kmem_size
,total_size
;
2751 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2752 * If they are really small and used heavily, the system can fall
2753 * into OOM very easily.
2754 * This function detect ZONE_DMA/DMA32 size and configures zone order.
2756 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2759 for_each_online_node(nid
) {
2760 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2761 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2762 if (populated_zone(z
)) {
2763 if (zone_type
< ZONE_NORMAL
)
2764 low_kmem_size
+= z
->present_pages
;
2765 total_size
+= z
->present_pages
;
2766 } else if (zone_type
== ZONE_NORMAL
) {
2768 * If any node has only lowmem, then node order
2769 * is preferred to allow kernel allocations
2770 * locally; otherwise, they can easily infringe
2771 * on other nodes when there is an abundance of
2772 * lowmem available to allocate from.
2774 return ZONELIST_ORDER_NODE
;
2778 if (!low_kmem_size
|| /* there are no DMA area. */
2779 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2780 return ZONELIST_ORDER_NODE
;
2782 * look into each node's config.
2783 * If there is a node whose DMA/DMA32 memory is very big area on
2784 * local memory, NODE_ORDER may be suitable.
2786 average_size
= total_size
/
2787 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2788 for_each_online_node(nid
) {
2791 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2792 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2793 if (populated_zone(z
)) {
2794 if (zone_type
< ZONE_NORMAL
)
2795 low_kmem_size
+= z
->present_pages
;
2796 total_size
+= z
->present_pages
;
2799 if (low_kmem_size
&&
2800 total_size
> average_size
&& /* ignore small node */
2801 low_kmem_size
> total_size
* 70/100)
2802 return ZONELIST_ORDER_NODE
;
2804 return ZONELIST_ORDER_ZONE
;
2807 static void set_zonelist_order(void)
2809 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2810 current_zonelist_order
= default_zonelist_order();
2812 current_zonelist_order
= user_zonelist_order
;
2815 static void build_zonelists(pg_data_t
*pgdat
)
2819 nodemask_t used_mask
;
2820 int local_node
, prev_node
;
2821 struct zonelist
*zonelist
;
2822 int order
= current_zonelist_order
;
2824 /* initialize zonelists */
2825 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2826 zonelist
= pgdat
->node_zonelists
+ i
;
2827 zonelist
->_zonerefs
[0].zone
= NULL
;
2828 zonelist
->_zonerefs
[0].zone_idx
= 0;
2831 /* NUMA-aware ordering of nodes */
2832 local_node
= pgdat
->node_id
;
2833 load
= nr_online_nodes
;
2834 prev_node
= local_node
;
2835 nodes_clear(used_mask
);
2837 memset(node_order
, 0, sizeof(node_order
));
2840 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2841 int distance
= node_distance(local_node
, node
);
2844 * If another node is sufficiently far away then it is better
2845 * to reclaim pages in a zone before going off node.
2847 if (distance
> RECLAIM_DISTANCE
)
2848 zone_reclaim_mode
= 1;
2851 * We don't want to pressure a particular node.
2852 * So adding penalty to the first node in same
2853 * distance group to make it round-robin.
2855 if (distance
!= node_distance(local_node
, prev_node
))
2856 node_load
[node
] = load
;
2860 if (order
== ZONELIST_ORDER_NODE
)
2861 build_zonelists_in_node_order(pgdat
, node
);
2863 node_order
[j
++] = node
; /* remember order */
2866 if (order
== ZONELIST_ORDER_ZONE
) {
2867 /* calculate node order -- i.e., DMA last! */
2868 build_zonelists_in_zone_order(pgdat
, j
);
2871 build_thisnode_zonelists(pgdat
);
2874 /* Construct the zonelist performance cache - see further mmzone.h */
2875 static void build_zonelist_cache(pg_data_t
*pgdat
)
2877 struct zonelist
*zonelist
;
2878 struct zonelist_cache
*zlc
;
2881 zonelist
= &pgdat
->node_zonelists
[0];
2882 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2883 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2884 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2885 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2888 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
2890 * Return node id of node used for "local" allocations.
2891 * I.e., first node id of first zone in arg node's generic zonelist.
2892 * Used for initializing percpu 'numa_mem', which is used primarily
2893 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
2895 int local_memory_node(int node
)
2899 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
2900 gfp_zone(GFP_KERNEL
),
2907 #else /* CONFIG_NUMA */
2909 static void set_zonelist_order(void)
2911 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2914 static void build_zonelists(pg_data_t
*pgdat
)
2916 int node
, local_node
;
2918 struct zonelist
*zonelist
;
2920 local_node
= pgdat
->node_id
;
2922 zonelist
= &pgdat
->node_zonelists
[0];
2923 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2926 * Now we build the zonelist so that it contains the zones
2927 * of all the other nodes.
2928 * We don't want to pressure a particular node, so when
2929 * building the zones for node N, we make sure that the
2930 * zones coming right after the local ones are those from
2931 * node N+1 (modulo N)
2933 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2934 if (!node_online(node
))
2936 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2939 for (node
= 0; node
< local_node
; node
++) {
2940 if (!node_online(node
))
2942 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2946 zonelist
->_zonerefs
[j
].zone
= NULL
;
2947 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2950 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2951 static void build_zonelist_cache(pg_data_t
*pgdat
)
2953 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2956 #endif /* CONFIG_NUMA */
2959 * Boot pageset table. One per cpu which is going to be used for all
2960 * zones and all nodes. The parameters will be set in such a way
2961 * that an item put on a list will immediately be handed over to
2962 * the buddy list. This is safe since pageset manipulation is done
2963 * with interrupts disabled.
2965 * The boot_pagesets must be kept even after bootup is complete for
2966 * unused processors and/or zones. They do play a role for bootstrapping
2967 * hotplugged processors.
2969 * zoneinfo_show() and maybe other functions do
2970 * not check if the processor is online before following the pageset pointer.
2971 * Other parts of the kernel may not check if the zone is available.
2973 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
2974 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
2975 static void setup_zone_pageset(struct zone
*zone
);
2978 * Global mutex to protect against size modification of zonelists
2979 * as well as to serialize pageset setup for the new populated zone.
2981 DEFINE_MUTEX(zonelists_mutex
);
2983 /* return values int ....just for stop_machine() */
2984 static __init_refok
int __build_all_zonelists(void *data
)
2990 memset(node_load
, 0, sizeof(node_load
));
2992 for_each_online_node(nid
) {
2993 pg_data_t
*pgdat
= NODE_DATA(nid
);
2995 build_zonelists(pgdat
);
2996 build_zonelist_cache(pgdat
);
2999 #ifdef CONFIG_MEMORY_HOTPLUG
3000 /* Setup real pagesets for the new zone */
3002 struct zone
*zone
= data
;
3003 setup_zone_pageset(zone
);
3008 * Initialize the boot_pagesets that are going to be used
3009 * for bootstrapping processors. The real pagesets for
3010 * each zone will be allocated later when the per cpu
3011 * allocator is available.
3013 * boot_pagesets are used also for bootstrapping offline
3014 * cpus if the system is already booted because the pagesets
3015 * are needed to initialize allocators on a specific cpu too.
3016 * F.e. the percpu allocator needs the page allocator which
3017 * needs the percpu allocator in order to allocate its pagesets
3018 * (a chicken-egg dilemma).
3020 for_each_possible_cpu(cpu
) {
3021 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3023 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3025 * We now know the "local memory node" for each node--
3026 * i.e., the node of the first zone in the generic zonelist.
3027 * Set up numa_mem percpu variable for on-line cpus. During
3028 * boot, only the boot cpu should be on-line; we'll init the
3029 * secondary cpus' numa_mem as they come on-line. During
3030 * node/memory hotplug, we'll fixup all on-line cpus.
3032 if (cpu_online(cpu
))
3033 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3041 * Called with zonelists_mutex held always
3042 * unless system_state == SYSTEM_BOOTING.
3044 void build_all_zonelists(void *data
)
3046 set_zonelist_order();
3048 if (system_state
== SYSTEM_BOOTING
) {
3049 __build_all_zonelists(NULL
);
3050 mminit_verify_zonelist();
3051 cpuset_init_current_mems_allowed();
3053 /* we have to stop all cpus to guarantee there is no user
3055 stop_machine(__build_all_zonelists
, data
, NULL
);
3056 /* cpuset refresh routine should be here */
3058 vm_total_pages
= nr_free_pagecache_pages();
3060 * Disable grouping by mobility if the number of pages in the
3061 * system is too low to allow the mechanism to work. It would be
3062 * more accurate, but expensive to check per-zone. This check is
3063 * made on memory-hotadd so a system can start with mobility
3064 * disabled and enable it later
3066 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3067 page_group_by_mobility_disabled
= 1;
3069 page_group_by_mobility_disabled
= 0;
3071 printk("Built %i zonelists in %s order, mobility grouping %s. "
3072 "Total pages: %ld\n",
3074 zonelist_order_name
[current_zonelist_order
],
3075 page_group_by_mobility_disabled
? "off" : "on",
3078 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3083 * Helper functions to size the waitqueue hash table.
3084 * Essentially these want to choose hash table sizes sufficiently
3085 * large so that collisions trying to wait on pages are rare.
3086 * But in fact, the number of active page waitqueues on typical
3087 * systems is ridiculously low, less than 200. So this is even
3088 * conservative, even though it seems large.
3090 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3091 * waitqueues, i.e. the size of the waitq table given the number of pages.
3093 #define PAGES_PER_WAITQUEUE 256
3095 #ifndef CONFIG_MEMORY_HOTPLUG
3096 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3098 unsigned long size
= 1;
3100 pages
/= PAGES_PER_WAITQUEUE
;
3102 while (size
< pages
)
3106 * Once we have dozens or even hundreds of threads sleeping
3107 * on IO we've got bigger problems than wait queue collision.
3108 * Limit the size of the wait table to a reasonable size.
3110 size
= min(size
, 4096UL);
3112 return max(size
, 4UL);
3116 * A zone's size might be changed by hot-add, so it is not possible to determine
3117 * a suitable size for its wait_table. So we use the maximum size now.
3119 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3121 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3122 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3123 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3125 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3126 * or more by the traditional way. (See above). It equals:
3128 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3129 * ia64(16K page size) : = ( 8G + 4M)byte.
3130 * powerpc (64K page size) : = (32G +16M)byte.
3132 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3139 * This is an integer logarithm so that shifts can be used later
3140 * to extract the more random high bits from the multiplicative
3141 * hash function before the remainder is taken.
3143 static inline unsigned long wait_table_bits(unsigned long size
)
3148 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3151 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3152 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3153 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3154 * higher will lead to a bigger reserve which will get freed as contiguous
3155 * blocks as reclaim kicks in
3157 static void setup_zone_migrate_reserve(struct zone
*zone
)
3159 unsigned long start_pfn
, pfn
, end_pfn
;
3161 unsigned long block_migratetype
;
3164 /* Get the start pfn, end pfn and the number of blocks to reserve */
3165 start_pfn
= zone
->zone_start_pfn
;
3166 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3167 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3171 * Reserve blocks are generally in place to help high-order atomic
3172 * allocations that are short-lived. A min_free_kbytes value that
3173 * would result in more than 2 reserve blocks for atomic allocations
3174 * is assumed to be in place to help anti-fragmentation for the
3175 * future allocation of hugepages at runtime.
3177 reserve
= min(2, reserve
);
3179 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3180 if (!pfn_valid(pfn
))
3182 page
= pfn_to_page(pfn
);
3184 /* Watch out for overlapping nodes */
3185 if (page_to_nid(page
) != zone_to_nid(zone
))
3188 /* Blocks with reserved pages will never free, skip them. */
3189 if (PageReserved(page
))
3192 block_migratetype
= get_pageblock_migratetype(page
);
3194 /* If this block is reserved, account for it */
3195 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
3200 /* Suitable for reserving if this block is movable */
3201 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
3202 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
3203 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
3209 * If the reserve is met and this is a previous reserved block,
3212 if (block_migratetype
== MIGRATE_RESERVE
) {
3213 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3214 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3220 * Initially all pages are reserved - free ones are freed
3221 * up by free_all_bootmem() once the early boot process is
3222 * done. Non-atomic initialization, single-pass.
3224 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3225 unsigned long start_pfn
, enum memmap_context context
)
3228 unsigned long end_pfn
= start_pfn
+ size
;
3232 if (highest_memmap_pfn
< end_pfn
- 1)
3233 highest_memmap_pfn
= end_pfn
- 1;
3235 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3236 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3238 * There can be holes in boot-time mem_map[]s
3239 * handed to this function. They do not
3240 * exist on hotplugged memory.
3242 if (context
== MEMMAP_EARLY
) {
3243 if (!early_pfn_valid(pfn
))
3245 if (!early_pfn_in_nid(pfn
, nid
))
3248 page
= pfn_to_page(pfn
);
3249 set_page_links(page
, zone
, nid
, pfn
);
3250 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3251 init_page_count(page
);
3252 reset_page_mapcount(page
);
3253 SetPageReserved(page
);
3255 * Mark the block movable so that blocks are reserved for
3256 * movable at startup. This will force kernel allocations
3257 * to reserve their blocks rather than leaking throughout
3258 * the address space during boot when many long-lived
3259 * kernel allocations are made. Later some blocks near
3260 * the start are marked MIGRATE_RESERVE by
3261 * setup_zone_migrate_reserve()
3263 * bitmap is created for zone's valid pfn range. but memmap
3264 * can be created for invalid pages (for alignment)
3265 * check here not to call set_pageblock_migratetype() against
3268 if ((z
->zone_start_pfn
<= pfn
)
3269 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3270 && !(pfn
& (pageblock_nr_pages
- 1)))
3271 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3273 INIT_LIST_HEAD(&page
->lru
);
3274 #ifdef WANT_PAGE_VIRTUAL
3275 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3276 if (!is_highmem_idx(zone
))
3277 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3282 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3285 for_each_migratetype_order(order
, t
) {
3286 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3287 zone
->free_area
[order
].nr_free
= 0;
3291 #ifndef __HAVE_ARCH_MEMMAP_INIT
3292 #define memmap_init(size, nid, zone, start_pfn) \
3293 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3296 static int zone_batchsize(struct zone
*zone
)
3302 * The per-cpu-pages pools are set to around 1000th of the
3303 * size of the zone. But no more than 1/2 of a meg.
3305 * OK, so we don't know how big the cache is. So guess.
3307 batch
= zone
->present_pages
/ 1024;
3308 if (batch
* PAGE_SIZE
> 512 * 1024)
3309 batch
= (512 * 1024) / PAGE_SIZE
;
3310 batch
/= 4; /* We effectively *= 4 below */
3315 * Clamp the batch to a 2^n - 1 value. Having a power
3316 * of 2 value was found to be more likely to have
3317 * suboptimal cache aliasing properties in some cases.
3319 * For example if 2 tasks are alternately allocating
3320 * batches of pages, one task can end up with a lot
3321 * of pages of one half of the possible page colors
3322 * and the other with pages of the other colors.
3324 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3329 /* The deferral and batching of frees should be suppressed under NOMMU
3332 * The problem is that NOMMU needs to be able to allocate large chunks
3333 * of contiguous memory as there's no hardware page translation to
3334 * assemble apparent contiguous memory from discontiguous pages.
3336 * Queueing large contiguous runs of pages for batching, however,
3337 * causes the pages to actually be freed in smaller chunks. As there
3338 * can be a significant delay between the individual batches being
3339 * recycled, this leads to the once large chunks of space being
3340 * fragmented and becoming unavailable for high-order allocations.
3346 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3348 struct per_cpu_pages
*pcp
;
3351 memset(p
, 0, sizeof(*p
));
3355 pcp
->high
= 6 * batch
;
3356 pcp
->batch
= max(1UL, 1 * batch
);
3357 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3358 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3362 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3363 * to the value high for the pageset p.
3366 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3369 struct per_cpu_pages
*pcp
;
3373 pcp
->batch
= max(1UL, high
/4);
3374 if ((high
/4) > (PAGE_SHIFT
* 8))
3375 pcp
->batch
= PAGE_SHIFT
* 8;
3378 static __meminit
void setup_zone_pageset(struct zone
*zone
)
3382 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
3384 for_each_possible_cpu(cpu
) {
3385 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
3387 setup_pageset(pcp
, zone_batchsize(zone
));
3389 if (percpu_pagelist_fraction
)
3390 setup_pagelist_highmark(pcp
,
3391 (zone
->present_pages
/
3392 percpu_pagelist_fraction
));
3397 * Allocate per cpu pagesets and initialize them.
3398 * Before this call only boot pagesets were available.
3400 void __init
setup_per_cpu_pageset(void)
3404 for_each_populated_zone(zone
)
3405 setup_zone_pageset(zone
);
3408 static noinline __init_refok
3409 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3412 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3416 * The per-page waitqueue mechanism uses hashed waitqueues
3419 zone
->wait_table_hash_nr_entries
=
3420 wait_table_hash_nr_entries(zone_size_pages
);
3421 zone
->wait_table_bits
=
3422 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3423 alloc_size
= zone
->wait_table_hash_nr_entries
3424 * sizeof(wait_queue_head_t
);
3426 if (!slab_is_available()) {
3427 zone
->wait_table
= (wait_queue_head_t
*)
3428 alloc_bootmem_node(pgdat
, alloc_size
);
3431 * This case means that a zone whose size was 0 gets new memory
3432 * via memory hot-add.
3433 * But it may be the case that a new node was hot-added. In
3434 * this case vmalloc() will not be able to use this new node's
3435 * memory - this wait_table must be initialized to use this new
3436 * node itself as well.
3437 * To use this new node's memory, further consideration will be
3440 zone
->wait_table
= vmalloc(alloc_size
);
3442 if (!zone
->wait_table
)
3445 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3446 init_waitqueue_head(zone
->wait_table
+ i
);
3451 static int __zone_pcp_update(void *data
)
3453 struct zone
*zone
= data
;
3455 unsigned long batch
= zone_batchsize(zone
), flags
;
3457 for_each_possible_cpu(cpu
) {
3458 struct per_cpu_pageset
*pset
;
3459 struct per_cpu_pages
*pcp
;
3461 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
3464 local_irq_save(flags
);
3465 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
3466 setup_pageset(pset
, batch
);
3467 local_irq_restore(flags
);
3472 void zone_pcp_update(struct zone
*zone
)
3474 stop_machine(__zone_pcp_update
, zone
, NULL
);
3477 static __meminit
void zone_pcp_init(struct zone
*zone
)
3480 * per cpu subsystem is not up at this point. The following code
3481 * relies on the ability of the linker to provide the
3482 * offset of a (static) per cpu variable into the per cpu area.
3484 zone
->pageset
= &boot_pageset
;
3486 if (zone
->present_pages
)
3487 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
3488 zone
->name
, zone
->present_pages
,
3489 zone_batchsize(zone
));
3492 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3493 unsigned long zone_start_pfn
,
3495 enum memmap_context context
)
3497 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3499 ret
= zone_wait_table_init(zone
, size
);
3502 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3504 zone
->zone_start_pfn
= zone_start_pfn
;
3506 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3507 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3509 (unsigned long)zone_idx(zone
),
3510 zone_start_pfn
, (zone_start_pfn
+ size
));
3512 zone_init_free_lists(zone
);
3517 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3519 * Basic iterator support. Return the first range of PFNs for a node
3520 * Note: nid == MAX_NUMNODES returns first region regardless of node
3522 static int __meminit
first_active_region_index_in_nid(int nid
)
3526 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3527 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3534 * Basic iterator support. Return the next active range of PFNs for a node
3535 * Note: nid == MAX_NUMNODES returns next region regardless of node
3537 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3539 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3540 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3546 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3548 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3549 * Architectures may implement their own version but if add_active_range()
3550 * was used and there are no special requirements, this is a convenient
3553 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3557 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3558 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3559 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3561 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3562 return early_node_map
[i
].nid
;
3564 /* This is a memory hole */
3567 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3569 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3573 nid
= __early_pfn_to_nid(pfn
);
3576 /* just returns 0 */
3580 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3581 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3585 nid
= __early_pfn_to_nid(pfn
);
3586 if (nid
>= 0 && nid
!= node
)
3592 /* Basic iterator support to walk early_node_map[] */
3593 #define for_each_active_range_index_in_nid(i, nid) \
3594 for (i = first_active_region_index_in_nid(nid); i != -1; \
3595 i = next_active_region_index_in_nid(i, nid))
3598 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3599 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3600 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3602 * If an architecture guarantees that all ranges registered with
3603 * add_active_ranges() contain no holes and may be freed, this
3604 * this function may be used instead of calling free_bootmem() manually.
3606 void __init
free_bootmem_with_active_regions(int nid
,
3607 unsigned long max_low_pfn
)
3611 for_each_active_range_index_in_nid(i
, nid
) {
3612 unsigned long size_pages
= 0;
3613 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3615 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3618 if (end_pfn
> max_low_pfn
)
3619 end_pfn
= max_low_pfn
;
3621 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3622 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3623 PFN_PHYS(early_node_map
[i
].start_pfn
),
3624 size_pages
<< PAGE_SHIFT
);
3628 int __init
add_from_early_node_map(struct range
*range
, int az
,
3629 int nr_range
, int nid
)
3634 /* need to go over early_node_map to find out good range for node */
3635 for_each_active_range_index_in_nid(i
, nid
) {
3636 start
= early_node_map
[i
].start_pfn
;
3637 end
= early_node_map
[i
].end_pfn
;
3638 nr_range
= add_range(range
, az
, nr_range
, start
, end
);
3643 #ifdef CONFIG_NO_BOOTMEM
3644 void * __init
__alloc_memory_core_early(int nid
, u64 size
, u64 align
,
3645 u64 goal
, u64 limit
)
3650 if (limit
> get_max_mapped())
3651 limit
= get_max_mapped();
3653 /* need to go over early_node_map to find out good range for node */
3654 for_each_active_range_index_in_nid(i
, nid
) {
3656 u64 ei_start
, ei_last
;
3658 ei_last
= early_node_map
[i
].end_pfn
;
3659 ei_last
<<= PAGE_SHIFT
;
3660 ei_start
= early_node_map
[i
].start_pfn
;
3661 ei_start
<<= PAGE_SHIFT
;
3662 addr
= find_early_area(ei_start
, ei_last
,
3663 goal
, limit
, size
, align
);
3669 printk(KERN_DEBUG
"alloc (nid=%d %llx - %llx) (%llx - %llx) %llx %llx => %llx\n",
3671 ei_start
, ei_last
, goal
, limit
, size
,
3675 ptr
= phys_to_virt(addr
);
3676 memset(ptr
, 0, size
);
3677 reserve_early_without_check(addr
, addr
+ size
, "BOOTMEM");
3679 * The min_count is set to 0 so that bootmem allocated blocks
3680 * are never reported as leaks.
3682 kmemleak_alloc(ptr
, size
, 0, 0);
3691 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3696 for_each_active_range_index_in_nid(i
, nid
) {
3697 ret
= work_fn(early_node_map
[i
].start_pfn
,
3698 early_node_map
[i
].end_pfn
, data
);
3704 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3705 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3707 * If an architecture guarantees that all ranges registered with
3708 * add_active_ranges() contain no holes and may be freed, this
3709 * function may be used instead of calling memory_present() manually.
3711 void __init
sparse_memory_present_with_active_regions(int nid
)
3715 for_each_active_range_index_in_nid(i
, nid
)
3716 memory_present(early_node_map
[i
].nid
,
3717 early_node_map
[i
].start_pfn
,
3718 early_node_map
[i
].end_pfn
);
3722 * get_pfn_range_for_nid - Return the start and end page frames for a node
3723 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3724 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3725 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3727 * It returns the start and end page frame of a node based on information
3728 * provided by an arch calling add_active_range(). If called for a node
3729 * with no available memory, a warning is printed and the start and end
3732 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3733 unsigned long *start_pfn
, unsigned long *end_pfn
)
3739 for_each_active_range_index_in_nid(i
, nid
) {
3740 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3741 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3744 if (*start_pfn
== -1UL)
3749 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3750 * assumption is made that zones within a node are ordered in monotonic
3751 * increasing memory addresses so that the "highest" populated zone is used
3753 static void __init
find_usable_zone_for_movable(void)
3756 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3757 if (zone_index
== ZONE_MOVABLE
)
3760 if (arch_zone_highest_possible_pfn
[zone_index
] >
3761 arch_zone_lowest_possible_pfn
[zone_index
])
3765 VM_BUG_ON(zone_index
== -1);
3766 movable_zone
= zone_index
;
3770 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3771 * because it is sized independant of architecture. Unlike the other zones,
3772 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3773 * in each node depending on the size of each node and how evenly kernelcore
3774 * is distributed. This helper function adjusts the zone ranges
3775 * provided by the architecture for a given node by using the end of the
3776 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3777 * zones within a node are in order of monotonic increases memory addresses
3779 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3780 unsigned long zone_type
,
3781 unsigned long node_start_pfn
,
3782 unsigned long node_end_pfn
,
3783 unsigned long *zone_start_pfn
,
3784 unsigned long *zone_end_pfn
)
3786 /* Only adjust if ZONE_MOVABLE is on this node */
3787 if (zone_movable_pfn
[nid
]) {
3788 /* Size ZONE_MOVABLE */
3789 if (zone_type
== ZONE_MOVABLE
) {
3790 *zone_start_pfn
= zone_movable_pfn
[nid
];
3791 *zone_end_pfn
= min(node_end_pfn
,
3792 arch_zone_highest_possible_pfn
[movable_zone
]);
3794 /* Adjust for ZONE_MOVABLE starting within this range */
3795 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3796 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3797 *zone_end_pfn
= zone_movable_pfn
[nid
];
3799 /* Check if this whole range is within ZONE_MOVABLE */
3800 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3801 *zone_start_pfn
= *zone_end_pfn
;
3806 * Return the number of pages a zone spans in a node, including holes
3807 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3809 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3810 unsigned long zone_type
,
3811 unsigned long *ignored
)
3813 unsigned long node_start_pfn
, node_end_pfn
;
3814 unsigned long zone_start_pfn
, zone_end_pfn
;
3816 /* Get the start and end of the node and zone */
3817 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3818 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3819 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3820 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3821 node_start_pfn
, node_end_pfn
,
3822 &zone_start_pfn
, &zone_end_pfn
);
3824 /* Check that this node has pages within the zone's required range */
3825 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3828 /* Move the zone boundaries inside the node if necessary */
3829 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3830 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3832 /* Return the spanned pages */
3833 return zone_end_pfn
- zone_start_pfn
;
3837 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3838 * then all holes in the requested range will be accounted for.
3840 unsigned long __meminit
__absent_pages_in_range(int nid
,
3841 unsigned long range_start_pfn
,
3842 unsigned long range_end_pfn
)
3845 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3846 unsigned long start_pfn
;
3848 /* Find the end_pfn of the first active range of pfns in the node */
3849 i
= first_active_region_index_in_nid(nid
);
3853 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3855 /* Account for ranges before physical memory on this node */
3856 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3857 hole_pages
= prev_end_pfn
- range_start_pfn
;
3859 /* Find all holes for the zone within the node */
3860 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3862 /* No need to continue if prev_end_pfn is outside the zone */
3863 if (prev_end_pfn
>= range_end_pfn
)
3866 /* Make sure the end of the zone is not within the hole */
3867 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3868 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3870 /* Update the hole size cound and move on */
3871 if (start_pfn
> range_start_pfn
) {
3872 BUG_ON(prev_end_pfn
> start_pfn
);
3873 hole_pages
+= start_pfn
- prev_end_pfn
;
3875 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3878 /* Account for ranges past physical memory on this node */
3879 if (range_end_pfn
> prev_end_pfn
)
3880 hole_pages
+= range_end_pfn
-
3881 max(range_start_pfn
, prev_end_pfn
);
3887 * absent_pages_in_range - Return number of page frames in holes within a range
3888 * @start_pfn: The start PFN to start searching for holes
3889 * @end_pfn: The end PFN to stop searching for holes
3891 * It returns the number of pages frames in memory holes within a range.
3893 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3894 unsigned long end_pfn
)
3896 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3899 /* Return the number of page frames in holes in a zone on a node */
3900 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3901 unsigned long zone_type
,
3902 unsigned long *ignored
)
3904 unsigned long node_start_pfn
, node_end_pfn
;
3905 unsigned long zone_start_pfn
, zone_end_pfn
;
3907 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3908 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3910 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3913 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3914 node_start_pfn
, node_end_pfn
,
3915 &zone_start_pfn
, &zone_end_pfn
);
3916 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3920 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3921 unsigned long zone_type
,
3922 unsigned long *zones_size
)
3924 return zones_size
[zone_type
];
3927 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3928 unsigned long zone_type
,
3929 unsigned long *zholes_size
)
3934 return zholes_size
[zone_type
];
3939 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3940 unsigned long *zones_size
, unsigned long *zholes_size
)
3942 unsigned long realtotalpages
, totalpages
= 0;
3945 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3946 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3948 pgdat
->node_spanned_pages
= totalpages
;
3950 realtotalpages
= totalpages
;
3951 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3953 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3955 pgdat
->node_present_pages
= realtotalpages
;
3956 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3960 #ifndef CONFIG_SPARSEMEM
3962 * Calculate the size of the zone->blockflags rounded to an unsigned long
3963 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3964 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3965 * round what is now in bits to nearest long in bits, then return it in
3968 static unsigned long __init
usemap_size(unsigned long zonesize
)
3970 unsigned long usemapsize
;
3972 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3973 usemapsize
= usemapsize
>> pageblock_order
;
3974 usemapsize
*= NR_PAGEBLOCK_BITS
;
3975 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3977 return usemapsize
/ 8;
3980 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3981 struct zone
*zone
, unsigned long zonesize
)
3983 unsigned long usemapsize
= usemap_size(zonesize
);
3984 zone
->pageblock_flags
= NULL
;
3986 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3989 static void inline setup_usemap(struct pglist_data
*pgdat
,
3990 struct zone
*zone
, unsigned long zonesize
) {}
3991 #endif /* CONFIG_SPARSEMEM */
3993 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3995 /* Return a sensible default order for the pageblock size. */
3996 static inline int pageblock_default_order(void)
3998 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3999 return HUGETLB_PAGE_ORDER
;
4004 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4005 static inline void __init
set_pageblock_order(unsigned int order
)
4007 /* Check that pageblock_nr_pages has not already been setup */
4008 if (pageblock_order
)
4012 * Assume the largest contiguous order of interest is a huge page.
4013 * This value may be variable depending on boot parameters on IA64
4015 pageblock_order
= order
;
4017 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4020 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4021 * and pageblock_default_order() are unused as pageblock_order is set
4022 * at compile-time. See include/linux/pageblock-flags.h for the values of
4023 * pageblock_order based on the kernel config
4025 static inline int pageblock_default_order(unsigned int order
)
4029 #define set_pageblock_order(x) do {} while (0)
4031 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4034 * Set up the zone data structures:
4035 * - mark all pages reserved
4036 * - mark all memory queues empty
4037 * - clear the memory bitmaps
4039 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4040 unsigned long *zones_size
, unsigned long *zholes_size
)
4043 int nid
= pgdat
->node_id
;
4044 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4047 pgdat_resize_init(pgdat
);
4048 pgdat
->nr_zones
= 0;
4049 init_waitqueue_head(&pgdat
->kswapd_wait
);
4050 pgdat
->kswapd_max_order
= 0;
4051 pgdat_page_cgroup_init(pgdat
);
4053 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4054 struct zone
*zone
= pgdat
->node_zones
+ j
;
4055 unsigned long size
, realsize
, memmap_pages
;
4058 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4059 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
4063 * Adjust realsize so that it accounts for how much memory
4064 * is used by this zone for memmap. This affects the watermark
4065 * and per-cpu initialisations
4068 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
4069 if (realsize
>= memmap_pages
) {
4070 realsize
-= memmap_pages
;
4073 " %s zone: %lu pages used for memmap\n",
4074 zone_names
[j
], memmap_pages
);
4077 " %s zone: %lu pages exceeds realsize %lu\n",
4078 zone_names
[j
], memmap_pages
, realsize
);
4080 /* Account for reserved pages */
4081 if (j
== 0 && realsize
> dma_reserve
) {
4082 realsize
-= dma_reserve
;
4083 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4084 zone_names
[0], dma_reserve
);
4087 if (!is_highmem_idx(j
))
4088 nr_kernel_pages
+= realsize
;
4089 nr_all_pages
+= realsize
;
4091 zone
->spanned_pages
= size
;
4092 zone
->present_pages
= realsize
;
4095 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
4097 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
4099 zone
->name
= zone_names
[j
];
4100 spin_lock_init(&zone
->lock
);
4101 spin_lock_init(&zone
->lru_lock
);
4102 zone_seqlock_init(zone
);
4103 zone
->zone_pgdat
= pgdat
;
4105 zone
->prev_priority
= DEF_PRIORITY
;
4107 zone_pcp_init(zone
);
4109 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
4110 zone
->reclaim_stat
.nr_saved_scan
[l
] = 0;
4112 zone
->reclaim_stat
.recent_rotated
[0] = 0;
4113 zone
->reclaim_stat
.recent_rotated
[1] = 0;
4114 zone
->reclaim_stat
.recent_scanned
[0] = 0;
4115 zone
->reclaim_stat
.recent_scanned
[1] = 0;
4116 zap_zone_vm_stats(zone
);
4121 set_pageblock_order(pageblock_default_order());
4122 setup_usemap(pgdat
, zone
, size
);
4123 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4124 size
, MEMMAP_EARLY
);
4126 memmap_init(size
, nid
, j
, zone_start_pfn
);
4127 zone_start_pfn
+= size
;
4131 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4133 /* Skip empty nodes */
4134 if (!pgdat
->node_spanned_pages
)
4137 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4138 /* ia64 gets its own node_mem_map, before this, without bootmem */
4139 if (!pgdat
->node_mem_map
) {
4140 unsigned long size
, start
, end
;
4144 * The zone's endpoints aren't required to be MAX_ORDER
4145 * aligned but the node_mem_map endpoints must be in order
4146 * for the buddy allocator to function correctly.
4148 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4149 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4150 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4151 size
= (end
- start
) * sizeof(struct page
);
4152 map
= alloc_remap(pgdat
->node_id
, size
);
4154 map
= alloc_bootmem_node(pgdat
, size
);
4155 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4157 #ifndef CONFIG_NEED_MULTIPLE_NODES
4159 * With no DISCONTIG, the global mem_map is just set as node 0's
4161 if (pgdat
== NODE_DATA(0)) {
4162 mem_map
= NODE_DATA(0)->node_mem_map
;
4163 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4164 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4165 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4166 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4169 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4172 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4173 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4175 pg_data_t
*pgdat
= NODE_DATA(nid
);
4177 pgdat
->node_id
= nid
;
4178 pgdat
->node_start_pfn
= node_start_pfn
;
4179 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4181 alloc_node_mem_map(pgdat
);
4182 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4183 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4184 nid
, (unsigned long)pgdat
,
4185 (unsigned long)pgdat
->node_mem_map
);
4188 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4191 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4193 #if MAX_NUMNODES > 1
4195 * Figure out the number of possible node ids.
4197 static void __init
setup_nr_node_ids(void)
4200 unsigned int highest
= 0;
4202 for_each_node_mask(node
, node_possible_map
)
4204 nr_node_ids
= highest
+ 1;
4207 static inline void setup_nr_node_ids(void)
4213 * add_active_range - Register a range of PFNs backed by physical memory
4214 * @nid: The node ID the range resides on
4215 * @start_pfn: The start PFN of the available physical memory
4216 * @end_pfn: The end PFN of the available physical memory
4218 * These ranges are stored in an early_node_map[] and later used by
4219 * free_area_init_nodes() to calculate zone sizes and holes. If the
4220 * range spans a memory hole, it is up to the architecture to ensure
4221 * the memory is not freed by the bootmem allocator. If possible
4222 * the range being registered will be merged with existing ranges.
4224 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
4225 unsigned long end_pfn
)
4229 mminit_dprintk(MMINIT_TRACE
, "memory_register",
4230 "Entering add_active_range(%d, %#lx, %#lx) "
4231 "%d entries of %d used\n",
4232 nid
, start_pfn
, end_pfn
,
4233 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
4235 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
4237 /* Merge with existing active regions if possible */
4238 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4239 if (early_node_map
[i
].nid
!= nid
)
4242 /* Skip if an existing region covers this new one */
4243 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
4244 end_pfn
<= early_node_map
[i
].end_pfn
)
4247 /* Merge forward if suitable */
4248 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
4249 end_pfn
> early_node_map
[i
].end_pfn
) {
4250 early_node_map
[i
].end_pfn
= end_pfn
;
4254 /* Merge backward if suitable */
4255 if (start_pfn
< early_node_map
[i
].start_pfn
&&
4256 end_pfn
>= early_node_map
[i
].start_pfn
) {
4257 early_node_map
[i
].start_pfn
= start_pfn
;
4262 /* Check that early_node_map is large enough */
4263 if (i
>= MAX_ACTIVE_REGIONS
) {
4264 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
4265 MAX_ACTIVE_REGIONS
);
4269 early_node_map
[i
].nid
= nid
;
4270 early_node_map
[i
].start_pfn
= start_pfn
;
4271 early_node_map
[i
].end_pfn
= end_pfn
;
4272 nr_nodemap_entries
= i
+ 1;
4276 * remove_active_range - Shrink an existing registered range of PFNs
4277 * @nid: The node id the range is on that should be shrunk
4278 * @start_pfn: The new PFN of the range
4279 * @end_pfn: The new PFN of the range
4281 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
4282 * The map is kept near the end physical page range that has already been
4283 * registered. This function allows an arch to shrink an existing registered
4286 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
4287 unsigned long end_pfn
)
4292 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
4293 nid
, start_pfn
, end_pfn
);
4295 /* Find the old active region end and shrink */
4296 for_each_active_range_index_in_nid(i
, nid
) {
4297 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4298 early_node_map
[i
].end_pfn
<= end_pfn
) {
4300 early_node_map
[i
].start_pfn
= 0;
4301 early_node_map
[i
].end_pfn
= 0;
4305 if (early_node_map
[i
].start_pfn
< start_pfn
&&
4306 early_node_map
[i
].end_pfn
> start_pfn
) {
4307 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
4308 early_node_map
[i
].end_pfn
= start_pfn
;
4309 if (temp_end_pfn
> end_pfn
)
4310 add_active_range(nid
, end_pfn
, temp_end_pfn
);
4313 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4314 early_node_map
[i
].end_pfn
> end_pfn
&&
4315 early_node_map
[i
].start_pfn
< end_pfn
) {
4316 early_node_map
[i
].start_pfn
= end_pfn
;
4324 /* remove the blank ones */
4325 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
4326 if (early_node_map
[i
].nid
!= nid
)
4328 if (early_node_map
[i
].end_pfn
)
4330 /* we found it, get rid of it */
4331 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
4332 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
4333 sizeof(early_node_map
[j
]));
4334 j
= nr_nodemap_entries
- 1;
4335 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
4336 nr_nodemap_entries
--;
4341 * remove_all_active_ranges - Remove all currently registered regions
4343 * During discovery, it may be found that a table like SRAT is invalid
4344 * and an alternative discovery method must be used. This function removes
4345 * all currently registered regions.
4347 void __init
remove_all_active_ranges(void)
4349 memset(early_node_map
, 0, sizeof(early_node_map
));
4350 nr_nodemap_entries
= 0;
4353 /* Compare two active node_active_regions */
4354 static int __init
cmp_node_active_region(const void *a
, const void *b
)
4356 struct node_active_region
*arange
= (struct node_active_region
*)a
;
4357 struct node_active_region
*brange
= (struct node_active_region
*)b
;
4359 /* Done this way to avoid overflows */
4360 if (arange
->start_pfn
> brange
->start_pfn
)
4362 if (arange
->start_pfn
< brange
->start_pfn
)
4368 /* sort the node_map by start_pfn */
4369 void __init
sort_node_map(void)
4371 sort(early_node_map
, (size_t)nr_nodemap_entries
,
4372 sizeof(struct node_active_region
),
4373 cmp_node_active_region
, NULL
);
4376 /* Find the lowest pfn for a node */
4377 static unsigned long __init
find_min_pfn_for_node(int nid
)
4380 unsigned long min_pfn
= ULONG_MAX
;
4382 /* Assuming a sorted map, the first range found has the starting pfn */
4383 for_each_active_range_index_in_nid(i
, nid
)
4384 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
4386 if (min_pfn
== ULONG_MAX
) {
4388 "Could not find start_pfn for node %d\n", nid
);
4396 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4398 * It returns the minimum PFN based on information provided via
4399 * add_active_range().
4401 unsigned long __init
find_min_pfn_with_active_regions(void)
4403 return find_min_pfn_for_node(MAX_NUMNODES
);
4407 * early_calculate_totalpages()
4408 * Sum pages in active regions for movable zone.
4409 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4411 static unsigned long __init
early_calculate_totalpages(void)
4414 unsigned long totalpages
= 0;
4416 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4417 unsigned long pages
= early_node_map
[i
].end_pfn
-
4418 early_node_map
[i
].start_pfn
;
4419 totalpages
+= pages
;
4421 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
4427 * Find the PFN the Movable zone begins in each node. Kernel memory
4428 * is spread evenly between nodes as long as the nodes have enough
4429 * memory. When they don't, some nodes will have more kernelcore than
4432 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
4435 unsigned long usable_startpfn
;
4436 unsigned long kernelcore_node
, kernelcore_remaining
;
4437 /* save the state before borrow the nodemask */
4438 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4439 unsigned long totalpages
= early_calculate_totalpages();
4440 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4443 * If movablecore was specified, calculate what size of
4444 * kernelcore that corresponds so that memory usable for
4445 * any allocation type is evenly spread. If both kernelcore
4446 * and movablecore are specified, then the value of kernelcore
4447 * will be used for required_kernelcore if it's greater than
4448 * what movablecore would have allowed.
4450 if (required_movablecore
) {
4451 unsigned long corepages
;
4454 * Round-up so that ZONE_MOVABLE is at least as large as what
4455 * was requested by the user
4457 required_movablecore
=
4458 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4459 corepages
= totalpages
- required_movablecore
;
4461 required_kernelcore
= max(required_kernelcore
, corepages
);
4464 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4465 if (!required_kernelcore
)
4468 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4469 find_usable_zone_for_movable();
4470 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4473 /* Spread kernelcore memory as evenly as possible throughout nodes */
4474 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4475 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4477 * Recalculate kernelcore_node if the division per node
4478 * now exceeds what is necessary to satisfy the requested
4479 * amount of memory for the kernel
4481 if (required_kernelcore
< kernelcore_node
)
4482 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4485 * As the map is walked, we track how much memory is usable
4486 * by the kernel using kernelcore_remaining. When it is
4487 * 0, the rest of the node is usable by ZONE_MOVABLE
4489 kernelcore_remaining
= kernelcore_node
;
4491 /* Go through each range of PFNs within this node */
4492 for_each_active_range_index_in_nid(i
, nid
) {
4493 unsigned long start_pfn
, end_pfn
;
4494 unsigned long size_pages
;
4496 start_pfn
= max(early_node_map
[i
].start_pfn
,
4497 zone_movable_pfn
[nid
]);
4498 end_pfn
= early_node_map
[i
].end_pfn
;
4499 if (start_pfn
>= end_pfn
)
4502 /* Account for what is only usable for kernelcore */
4503 if (start_pfn
< usable_startpfn
) {
4504 unsigned long kernel_pages
;
4505 kernel_pages
= min(end_pfn
, usable_startpfn
)
4508 kernelcore_remaining
-= min(kernel_pages
,
4509 kernelcore_remaining
);
4510 required_kernelcore
-= min(kernel_pages
,
4511 required_kernelcore
);
4513 /* Continue if range is now fully accounted */
4514 if (end_pfn
<= usable_startpfn
) {
4517 * Push zone_movable_pfn to the end so
4518 * that if we have to rebalance
4519 * kernelcore across nodes, we will
4520 * not double account here
4522 zone_movable_pfn
[nid
] = end_pfn
;
4525 start_pfn
= usable_startpfn
;
4529 * The usable PFN range for ZONE_MOVABLE is from
4530 * start_pfn->end_pfn. Calculate size_pages as the
4531 * number of pages used as kernelcore
4533 size_pages
= end_pfn
- start_pfn
;
4534 if (size_pages
> kernelcore_remaining
)
4535 size_pages
= kernelcore_remaining
;
4536 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4539 * Some kernelcore has been met, update counts and
4540 * break if the kernelcore for this node has been
4543 required_kernelcore
-= min(required_kernelcore
,
4545 kernelcore_remaining
-= size_pages
;
4546 if (!kernelcore_remaining
)
4552 * If there is still required_kernelcore, we do another pass with one
4553 * less node in the count. This will push zone_movable_pfn[nid] further
4554 * along on the nodes that still have memory until kernelcore is
4558 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4561 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4562 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4563 zone_movable_pfn
[nid
] =
4564 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4567 /* restore the node_state */
4568 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4571 /* Any regular memory on that node ? */
4572 static void check_for_regular_memory(pg_data_t
*pgdat
)
4574 #ifdef CONFIG_HIGHMEM
4575 enum zone_type zone_type
;
4577 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4578 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4579 if (zone
->present_pages
)
4580 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4586 * free_area_init_nodes - Initialise all pg_data_t and zone data
4587 * @max_zone_pfn: an array of max PFNs for each zone
4589 * This will call free_area_init_node() for each active node in the system.
4590 * Using the page ranges provided by add_active_range(), the size of each
4591 * zone in each node and their holes is calculated. If the maximum PFN
4592 * between two adjacent zones match, it is assumed that the zone is empty.
4593 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4594 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4595 * starts where the previous one ended. For example, ZONE_DMA32 starts
4596 * at arch_max_dma_pfn.
4598 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4603 /* Sort early_node_map as initialisation assumes it is sorted */
4606 /* Record where the zone boundaries are */
4607 memset(arch_zone_lowest_possible_pfn
, 0,
4608 sizeof(arch_zone_lowest_possible_pfn
));
4609 memset(arch_zone_highest_possible_pfn
, 0,
4610 sizeof(arch_zone_highest_possible_pfn
));
4611 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4612 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4613 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4614 if (i
== ZONE_MOVABLE
)
4616 arch_zone_lowest_possible_pfn
[i
] =
4617 arch_zone_highest_possible_pfn
[i
-1];
4618 arch_zone_highest_possible_pfn
[i
] =
4619 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4621 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4622 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4624 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4625 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4626 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4628 /* Print out the zone ranges */
4629 printk("Zone PFN ranges:\n");
4630 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4631 if (i
== ZONE_MOVABLE
)
4633 printk(" %-8s ", zone_names
[i
]);
4634 if (arch_zone_lowest_possible_pfn
[i
] ==
4635 arch_zone_highest_possible_pfn
[i
])
4638 printk("%0#10lx -> %0#10lx\n",
4639 arch_zone_lowest_possible_pfn
[i
],
4640 arch_zone_highest_possible_pfn
[i
]);
4643 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4644 printk("Movable zone start PFN for each node\n");
4645 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4646 if (zone_movable_pfn
[i
])
4647 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4650 /* Print out the early_node_map[] */
4651 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4652 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4653 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4654 early_node_map
[i
].start_pfn
,
4655 early_node_map
[i
].end_pfn
);
4657 /* Initialise every node */
4658 mminit_verify_pageflags_layout();
4659 setup_nr_node_ids();
4660 for_each_online_node(nid
) {
4661 pg_data_t
*pgdat
= NODE_DATA(nid
);
4662 free_area_init_node(nid
, NULL
,
4663 find_min_pfn_for_node(nid
), NULL
);
4665 /* Any memory on that node */
4666 if (pgdat
->node_present_pages
)
4667 node_set_state(nid
, N_HIGH_MEMORY
);
4668 check_for_regular_memory(pgdat
);
4672 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4674 unsigned long long coremem
;
4678 coremem
= memparse(p
, &p
);
4679 *core
= coremem
>> PAGE_SHIFT
;
4681 /* Paranoid check that UL is enough for the coremem value */
4682 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4688 * kernelcore=size sets the amount of memory for use for allocations that
4689 * cannot be reclaimed or migrated.
4691 static int __init
cmdline_parse_kernelcore(char *p
)
4693 return cmdline_parse_core(p
, &required_kernelcore
);
4697 * movablecore=size sets the amount of memory for use for allocations that
4698 * can be reclaimed or migrated.
4700 static int __init
cmdline_parse_movablecore(char *p
)
4702 return cmdline_parse_core(p
, &required_movablecore
);
4705 early_param("kernelcore", cmdline_parse_kernelcore
);
4706 early_param("movablecore", cmdline_parse_movablecore
);
4708 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4711 * set_dma_reserve - set the specified number of pages reserved in the first zone
4712 * @new_dma_reserve: The number of pages to mark reserved
4714 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4715 * In the DMA zone, a significant percentage may be consumed by kernel image
4716 * and other unfreeable allocations which can skew the watermarks badly. This
4717 * function may optionally be used to account for unfreeable pages in the
4718 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4719 * smaller per-cpu batchsize.
4721 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4723 dma_reserve
= new_dma_reserve
;
4726 #ifndef CONFIG_NEED_MULTIPLE_NODES
4727 struct pglist_data __refdata contig_page_data
= {
4728 #ifndef CONFIG_NO_BOOTMEM
4729 .bdata
= &bootmem_node_data
[0]
4732 EXPORT_SYMBOL(contig_page_data
);
4735 void __init
free_area_init(unsigned long *zones_size
)
4737 free_area_init_node(0, zones_size
,
4738 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4741 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4742 unsigned long action
, void *hcpu
)
4744 int cpu
= (unsigned long)hcpu
;
4746 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4750 * Spill the event counters of the dead processor
4751 * into the current processors event counters.
4752 * This artificially elevates the count of the current
4755 vm_events_fold_cpu(cpu
);
4758 * Zero the differential counters of the dead processor
4759 * so that the vm statistics are consistent.
4761 * This is only okay since the processor is dead and cannot
4762 * race with what we are doing.
4764 refresh_cpu_vm_stats(cpu
);
4769 void __init
page_alloc_init(void)
4771 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4775 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4776 * or min_free_kbytes changes.
4778 static void calculate_totalreserve_pages(void)
4780 struct pglist_data
*pgdat
;
4781 unsigned long reserve_pages
= 0;
4782 enum zone_type i
, j
;
4784 for_each_online_pgdat(pgdat
) {
4785 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4786 struct zone
*zone
= pgdat
->node_zones
+ i
;
4787 unsigned long max
= 0;
4789 /* Find valid and maximum lowmem_reserve in the zone */
4790 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4791 if (zone
->lowmem_reserve
[j
] > max
)
4792 max
= zone
->lowmem_reserve
[j
];
4795 /* we treat the high watermark as reserved pages. */
4796 max
+= high_wmark_pages(zone
);
4798 if (max
> zone
->present_pages
)
4799 max
= zone
->present_pages
;
4800 reserve_pages
+= max
;
4803 totalreserve_pages
= reserve_pages
;
4807 * setup_per_zone_lowmem_reserve - called whenever
4808 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4809 * has a correct pages reserved value, so an adequate number of
4810 * pages are left in the zone after a successful __alloc_pages().
4812 static void setup_per_zone_lowmem_reserve(void)
4814 struct pglist_data
*pgdat
;
4815 enum zone_type j
, idx
;
4817 for_each_online_pgdat(pgdat
) {
4818 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4819 struct zone
*zone
= pgdat
->node_zones
+ j
;
4820 unsigned long present_pages
= zone
->present_pages
;
4822 zone
->lowmem_reserve
[j
] = 0;
4826 struct zone
*lower_zone
;
4830 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4831 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4833 lower_zone
= pgdat
->node_zones
+ idx
;
4834 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4835 sysctl_lowmem_reserve_ratio
[idx
];
4836 present_pages
+= lower_zone
->present_pages
;
4841 /* update totalreserve_pages */
4842 calculate_totalreserve_pages();
4846 * setup_per_zone_wmarks - called when min_free_kbytes changes
4847 * or when memory is hot-{added|removed}
4849 * Ensures that the watermark[min,low,high] values for each zone are set
4850 * correctly with respect to min_free_kbytes.
4852 void setup_per_zone_wmarks(void)
4854 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4855 unsigned long lowmem_pages
= 0;
4857 unsigned long flags
;
4859 /* Calculate total number of !ZONE_HIGHMEM pages */
4860 for_each_zone(zone
) {
4861 if (!is_highmem(zone
))
4862 lowmem_pages
+= zone
->present_pages
;
4865 for_each_zone(zone
) {
4868 spin_lock_irqsave(&zone
->lock
, flags
);
4869 tmp
= (u64
)pages_min
* zone
->present_pages
;
4870 do_div(tmp
, lowmem_pages
);
4871 if (is_highmem(zone
)) {
4873 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4874 * need highmem pages, so cap pages_min to a small
4877 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4878 * deltas controls asynch page reclaim, and so should
4879 * not be capped for highmem.
4883 min_pages
= zone
->present_pages
/ 1024;
4884 if (min_pages
< SWAP_CLUSTER_MAX
)
4885 min_pages
= SWAP_CLUSTER_MAX
;
4886 if (min_pages
> 128)
4888 zone
->watermark
[WMARK_MIN
] = min_pages
;
4891 * If it's a lowmem zone, reserve a number of pages
4892 * proportionate to the zone's size.
4894 zone
->watermark
[WMARK_MIN
] = tmp
;
4897 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
4898 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
4899 setup_zone_migrate_reserve(zone
);
4900 spin_unlock_irqrestore(&zone
->lock
, flags
);
4903 /* update totalreserve_pages */
4904 calculate_totalreserve_pages();
4908 * The inactive anon list should be small enough that the VM never has to
4909 * do too much work, but large enough that each inactive page has a chance
4910 * to be referenced again before it is swapped out.
4912 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4913 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4914 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4915 * the anonymous pages are kept on the inactive list.
4918 * memory ratio inactive anon
4919 * -------------------------------------
4928 void calculate_zone_inactive_ratio(struct zone
*zone
)
4930 unsigned int gb
, ratio
;
4932 /* Zone size in gigabytes */
4933 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4935 ratio
= int_sqrt(10 * gb
);
4939 zone
->inactive_ratio
= ratio
;
4942 static void __init
setup_per_zone_inactive_ratio(void)
4947 calculate_zone_inactive_ratio(zone
);
4951 * Initialise min_free_kbytes.
4953 * For small machines we want it small (128k min). For large machines
4954 * we want it large (64MB max). But it is not linear, because network
4955 * bandwidth does not increase linearly with machine size. We use
4957 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4958 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4974 static int __init
init_per_zone_wmark_min(void)
4976 unsigned long lowmem_kbytes
;
4978 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4980 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4981 if (min_free_kbytes
< 128)
4982 min_free_kbytes
= 128;
4983 if (min_free_kbytes
> 65536)
4984 min_free_kbytes
= 65536;
4985 setup_per_zone_wmarks();
4986 setup_per_zone_lowmem_reserve();
4987 setup_per_zone_inactive_ratio();
4990 module_init(init_per_zone_wmark_min
)
4993 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4994 * that we can call two helper functions whenever min_free_kbytes
4997 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4998 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5000 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5002 setup_per_zone_wmarks();
5007 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5008 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5013 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5018 zone
->min_unmapped_pages
= (zone
->present_pages
*
5019 sysctl_min_unmapped_ratio
) / 100;
5023 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5024 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5029 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5034 zone
->min_slab_pages
= (zone
->present_pages
*
5035 sysctl_min_slab_ratio
) / 100;
5041 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5042 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5043 * whenever sysctl_lowmem_reserve_ratio changes.
5045 * The reserve ratio obviously has absolutely no relation with the
5046 * minimum watermarks. The lowmem reserve ratio can only make sense
5047 * if in function of the boot time zone sizes.
5049 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5050 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5052 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5053 setup_per_zone_lowmem_reserve();
5058 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5059 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5060 * can have before it gets flushed back to buddy allocator.
5063 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5064 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5070 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5071 if (!write
|| (ret
== -EINVAL
))
5073 for_each_populated_zone(zone
) {
5074 for_each_possible_cpu(cpu
) {
5076 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5077 setup_pagelist_highmark(
5078 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5084 int hashdist
= HASHDIST_DEFAULT
;
5087 static int __init
set_hashdist(char *str
)
5091 hashdist
= simple_strtoul(str
, &str
, 0);
5094 __setup("hashdist=", set_hashdist
);
5098 * allocate a large system hash table from bootmem
5099 * - it is assumed that the hash table must contain an exact power-of-2
5100 * quantity of entries
5101 * - limit is the number of hash buckets, not the total allocation size
5103 void *__init
alloc_large_system_hash(const char *tablename
,
5104 unsigned long bucketsize
,
5105 unsigned long numentries
,
5108 unsigned int *_hash_shift
,
5109 unsigned int *_hash_mask
,
5110 unsigned long limit
)
5112 unsigned long long max
= limit
;
5113 unsigned long log2qty
, size
;
5116 /* allow the kernel cmdline to have a say */
5118 /* round applicable memory size up to nearest megabyte */
5119 numentries
= nr_kernel_pages
;
5120 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5121 numentries
>>= 20 - PAGE_SHIFT
;
5122 numentries
<<= 20 - PAGE_SHIFT
;
5124 /* limit to 1 bucket per 2^scale bytes of low memory */
5125 if (scale
> PAGE_SHIFT
)
5126 numentries
>>= (scale
- PAGE_SHIFT
);
5128 numentries
<<= (PAGE_SHIFT
- scale
);
5130 /* Make sure we've got at least a 0-order allocation.. */
5131 if (unlikely(flags
& HASH_SMALL
)) {
5132 /* Makes no sense without HASH_EARLY */
5133 WARN_ON(!(flags
& HASH_EARLY
));
5134 if (!(numentries
>> *_hash_shift
)) {
5135 numentries
= 1UL << *_hash_shift
;
5136 BUG_ON(!numentries
);
5138 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5139 numentries
= PAGE_SIZE
/ bucketsize
;
5141 numentries
= roundup_pow_of_two(numentries
);
5143 /* limit allocation size to 1/16 total memory by default */
5145 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5146 do_div(max
, bucketsize
);
5149 if (numentries
> max
)
5152 log2qty
= ilog2(numentries
);
5155 size
= bucketsize
<< log2qty
;
5156 if (flags
& HASH_EARLY
)
5157 table
= alloc_bootmem_nopanic(size
);
5159 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5162 * If bucketsize is not a power-of-two, we may free
5163 * some pages at the end of hash table which
5164 * alloc_pages_exact() automatically does
5166 if (get_order(size
) < MAX_ORDER
) {
5167 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5168 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5171 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5174 panic("Failed to allocate %s hash table\n", tablename
);
5176 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
5179 ilog2(size
) - PAGE_SHIFT
,
5183 *_hash_shift
= log2qty
;
5185 *_hash_mask
= (1 << log2qty
) - 1;
5190 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5191 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5194 #ifdef CONFIG_SPARSEMEM
5195 return __pfn_to_section(pfn
)->pageblock_flags
;
5197 return zone
->pageblock_flags
;
5198 #endif /* CONFIG_SPARSEMEM */
5201 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5203 #ifdef CONFIG_SPARSEMEM
5204 pfn
&= (PAGES_PER_SECTION
-1);
5205 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5207 pfn
= pfn
- zone
->zone_start_pfn
;
5208 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5209 #endif /* CONFIG_SPARSEMEM */
5213 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5214 * @page: The page within the block of interest
5215 * @start_bitidx: The first bit of interest to retrieve
5216 * @end_bitidx: The last bit of interest
5217 * returns pageblock_bits flags
5219 unsigned long get_pageblock_flags_group(struct page
*page
,
5220 int start_bitidx
, int end_bitidx
)
5223 unsigned long *bitmap
;
5224 unsigned long pfn
, bitidx
;
5225 unsigned long flags
= 0;
5226 unsigned long value
= 1;
5228 zone
= page_zone(page
);
5229 pfn
= page_to_pfn(page
);
5230 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5231 bitidx
= pfn_to_bitidx(zone
, pfn
);
5233 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5234 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5241 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5242 * @page: The page within the block of interest
5243 * @start_bitidx: The first bit of interest
5244 * @end_bitidx: The last bit of interest
5245 * @flags: The flags to set
5247 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5248 int start_bitidx
, int end_bitidx
)
5251 unsigned long *bitmap
;
5252 unsigned long pfn
, bitidx
;
5253 unsigned long value
= 1;
5255 zone
= page_zone(page
);
5256 pfn
= page_to_pfn(page
);
5257 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5258 bitidx
= pfn_to_bitidx(zone
, pfn
);
5259 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5260 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5262 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5264 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5266 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5270 * This is designed as sub function...plz see page_isolation.c also.
5271 * set/clear page block's type to be ISOLATE.
5272 * page allocater never alloc memory from ISOLATE block.
5275 int set_migratetype_isolate(struct page
*page
)
5278 struct page
*curr_page
;
5279 unsigned long flags
, pfn
, iter
;
5280 unsigned long immobile
= 0;
5281 struct memory_isolate_notify arg
;
5286 zone
= page_zone(page
);
5287 zone_idx
= zone_idx(zone
);
5289 spin_lock_irqsave(&zone
->lock
, flags
);
5290 if (get_pageblock_migratetype(page
) == MIGRATE_MOVABLE
||
5291 zone_idx
== ZONE_MOVABLE
) {
5296 pfn
= page_to_pfn(page
);
5297 arg
.start_pfn
= pfn
;
5298 arg
.nr_pages
= pageblock_nr_pages
;
5299 arg
.pages_found
= 0;
5302 * It may be possible to isolate a pageblock even if the
5303 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5304 * notifier chain is used by balloon drivers to return the
5305 * number of pages in a range that are held by the balloon
5306 * driver to shrink memory. If all the pages are accounted for
5307 * by balloons, are free, or on the LRU, isolation can continue.
5308 * Later, for example, when memory hotplug notifier runs, these
5309 * pages reported as "can be isolated" should be isolated(freed)
5310 * by the balloon driver through the memory notifier chain.
5312 notifier_ret
= memory_isolate_notify(MEM_ISOLATE_COUNT
, &arg
);
5313 notifier_ret
= notifier_to_errno(notifier_ret
);
5314 if (notifier_ret
|| !arg
.pages_found
)
5317 for (iter
= pfn
; iter
< (pfn
+ pageblock_nr_pages
); iter
++) {
5318 if (!pfn_valid_within(pfn
))
5321 curr_page
= pfn_to_page(iter
);
5322 if (!page_count(curr_page
) || PageLRU(curr_page
))
5328 if (arg
.pages_found
== immobile
)
5333 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
5334 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
5337 spin_unlock_irqrestore(&zone
->lock
, flags
);
5343 void unset_migratetype_isolate(struct page
*page
)
5346 unsigned long flags
;
5347 zone
= page_zone(page
);
5348 spin_lock_irqsave(&zone
->lock
, flags
);
5349 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
5351 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5352 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
5354 spin_unlock_irqrestore(&zone
->lock
, flags
);
5357 #ifdef CONFIG_MEMORY_HOTREMOVE
5359 * All pages in the range must be isolated before calling this.
5362 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
5368 unsigned long flags
;
5369 /* find the first valid pfn */
5370 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
5375 zone
= page_zone(pfn_to_page(pfn
));
5376 spin_lock_irqsave(&zone
->lock
, flags
);
5378 while (pfn
< end_pfn
) {
5379 if (!pfn_valid(pfn
)) {
5383 page
= pfn_to_page(pfn
);
5384 BUG_ON(page_count(page
));
5385 BUG_ON(!PageBuddy(page
));
5386 order
= page_order(page
);
5387 #ifdef CONFIG_DEBUG_VM
5388 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
5389 pfn
, 1 << order
, end_pfn
);
5391 list_del(&page
->lru
);
5392 rmv_page_order(page
);
5393 zone
->free_area
[order
].nr_free
--;
5394 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
5396 for (i
= 0; i
< (1 << order
); i
++)
5397 SetPageReserved((page
+i
));
5398 pfn
+= (1 << order
);
5400 spin_unlock_irqrestore(&zone
->lock
, flags
);
5404 #ifdef CONFIG_MEMORY_FAILURE
5405 bool is_free_buddy_page(struct page
*page
)
5407 struct zone
*zone
= page_zone(page
);
5408 unsigned long pfn
= page_to_pfn(page
);
5409 unsigned long flags
;
5412 spin_lock_irqsave(&zone
->lock
, flags
);
5413 for (order
= 0; order
< MAX_ORDER
; order
++) {
5414 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
5416 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
5419 spin_unlock_irqrestore(&zone
->lock
, flags
);
5421 return order
< MAX_ORDER
;
5425 static struct trace_print_flags pageflag_names
[] = {
5426 {1UL << PG_locked
, "locked" },
5427 {1UL << PG_error
, "error" },
5428 {1UL << PG_referenced
, "referenced" },
5429 {1UL << PG_uptodate
, "uptodate" },
5430 {1UL << PG_dirty
, "dirty" },
5431 {1UL << PG_lru
, "lru" },
5432 {1UL << PG_active
, "active" },
5433 {1UL << PG_slab
, "slab" },
5434 {1UL << PG_owner_priv_1
, "owner_priv_1" },
5435 {1UL << PG_arch_1
, "arch_1" },
5436 {1UL << PG_reserved
, "reserved" },
5437 {1UL << PG_private
, "private" },
5438 {1UL << PG_private_2
, "private_2" },
5439 {1UL << PG_writeback
, "writeback" },
5440 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5441 {1UL << PG_head
, "head" },
5442 {1UL << PG_tail
, "tail" },
5444 {1UL << PG_compound
, "compound" },
5446 {1UL << PG_swapcache
, "swapcache" },
5447 {1UL << PG_mappedtodisk
, "mappedtodisk" },
5448 {1UL << PG_reclaim
, "reclaim" },
5449 {1UL << PG_buddy
, "buddy" },
5450 {1UL << PG_swapbacked
, "swapbacked" },
5451 {1UL << PG_unevictable
, "unevictable" },
5453 {1UL << PG_mlocked
, "mlocked" },
5455 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5456 {1UL << PG_uncached
, "uncached" },
5458 #ifdef CONFIG_MEMORY_FAILURE
5459 {1UL << PG_hwpoison
, "hwpoison" },
5464 static void dump_page_flags(unsigned long flags
)
5466 const char *delim
= "";
5470 printk(KERN_ALERT
"page flags: %#lx(", flags
);
5472 /* remove zone id */
5473 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
5475 for (i
= 0; pageflag_names
[i
].name
&& flags
; i
++) {
5477 mask
= pageflag_names
[i
].mask
;
5478 if ((flags
& mask
) != mask
)
5482 printk("%s%s", delim
, pageflag_names
[i
].name
);
5486 /* check for left over flags */
5488 printk("%s%#lx", delim
, flags
);
5493 void dump_page(struct page
*page
)
5496 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5497 page
, page_count(page
), page_mapcount(page
),
5498 page
->mapping
, page
->index
);
5499 dump_page_flags(page
->flags
);