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).
107 static gfp_t saved_gfp_mask
;
109 void pm_restore_gfp_mask(void)
111 WARN_ON(!mutex_is_locked(&pm_mutex
));
112 if (saved_gfp_mask
) {
113 gfp_allowed_mask
= saved_gfp_mask
;
118 void pm_restrict_gfp_mask(void)
120 WARN_ON(!mutex_is_locked(&pm_mutex
));
121 WARN_ON(saved_gfp_mask
);
122 saved_gfp_mask
= gfp_allowed_mask
;
123 gfp_allowed_mask
&= ~GFP_IOFS
;
125 #endif /* CONFIG_PM_SLEEP */
127 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
128 int pageblock_order __read_mostly
;
131 static void __free_pages_ok(struct page
*page
, unsigned int order
);
134 * results with 256, 32 in the lowmem_reserve sysctl:
135 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
136 * 1G machine -> (16M dma, 784M normal, 224M high)
137 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
138 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
139 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
141 * TBD: should special case ZONE_DMA32 machines here - in those we normally
142 * don't need any ZONE_NORMAL reservation
144 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
145 #ifdef CONFIG_ZONE_DMA
148 #ifdef CONFIG_ZONE_DMA32
151 #ifdef CONFIG_HIGHMEM
157 EXPORT_SYMBOL(totalram_pages
);
159 static char * const zone_names
[MAX_NR_ZONES
] = {
160 #ifdef CONFIG_ZONE_DMA
163 #ifdef CONFIG_ZONE_DMA32
167 #ifdef CONFIG_HIGHMEM
173 int min_free_kbytes
= 1024;
175 static unsigned long __meminitdata nr_kernel_pages
;
176 static unsigned long __meminitdata nr_all_pages
;
177 static unsigned long __meminitdata dma_reserve
;
179 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
181 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
182 * ranges of memory (RAM) that may be registered with add_active_range().
183 * Ranges passed to add_active_range() will be merged if possible
184 * so the number of times add_active_range() can be called is
185 * related to the number of nodes and the number of holes
187 #ifdef CONFIG_MAX_ACTIVE_REGIONS
188 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
189 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
191 #if MAX_NUMNODES >= 32
192 /* If there can be many nodes, allow up to 50 holes per node */
193 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
195 /* By default, allow up to 256 distinct regions */
196 #define MAX_ACTIVE_REGIONS 256
200 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
201 static int __meminitdata nr_nodemap_entries
;
202 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
203 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
204 static unsigned long __initdata required_kernelcore
;
205 static unsigned long __initdata required_movablecore
;
206 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
208 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
210 EXPORT_SYMBOL(movable_zone
);
211 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
214 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
215 int nr_online_nodes __read_mostly
= 1;
216 EXPORT_SYMBOL(nr_node_ids
);
217 EXPORT_SYMBOL(nr_online_nodes
);
220 int page_group_by_mobility_disabled __read_mostly
;
222 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
225 if (unlikely(page_group_by_mobility_disabled
))
226 migratetype
= MIGRATE_UNMOVABLE
;
228 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
229 PB_migrate
, PB_migrate_end
);
232 bool oom_killer_disabled __read_mostly
;
234 #ifdef CONFIG_DEBUG_VM
235 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
239 unsigned long pfn
= page_to_pfn(page
);
242 seq
= zone_span_seqbegin(zone
);
243 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
245 else if (pfn
< zone
->zone_start_pfn
)
247 } while (zone_span_seqretry(zone
, seq
));
252 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
254 if (!pfn_valid_within(page_to_pfn(page
)))
256 if (zone
!= page_zone(page
))
262 * Temporary debugging check for pages not lying within a given zone.
264 static int bad_range(struct zone
*zone
, struct page
*page
)
266 if (page_outside_zone_boundaries(zone
, page
))
268 if (!page_is_consistent(zone
, page
))
274 static inline int bad_range(struct zone
*zone
, struct page
*page
)
280 static void bad_page(struct page
*page
)
282 static unsigned long resume
;
283 static unsigned long nr_shown
;
284 static unsigned long nr_unshown
;
286 /* Don't complain about poisoned pages */
287 if (PageHWPoison(page
)) {
288 __ClearPageBuddy(page
);
293 * Allow a burst of 60 reports, then keep quiet for that minute;
294 * or allow a steady drip of one report per second.
296 if (nr_shown
== 60) {
297 if (time_before(jiffies
, resume
)) {
303 "BUG: Bad page state: %lu messages suppressed\n",
310 resume
= jiffies
+ 60 * HZ
;
312 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
313 current
->comm
, page_to_pfn(page
));
318 /* Leave bad fields for debug, except PageBuddy could make trouble */
319 __ClearPageBuddy(page
);
320 add_taint(TAINT_BAD_PAGE
);
324 * Higher-order pages are called "compound pages". They are structured thusly:
326 * The first PAGE_SIZE page is called the "head page".
328 * The remaining PAGE_SIZE pages are called "tail pages".
330 * All pages have PG_compound set. All pages have their ->private pointing at
331 * the head page (even the head page has this).
333 * The first tail page's ->lru.next holds the address of the compound page's
334 * put_page() function. Its ->lru.prev holds the order of allocation.
335 * This usage means that zero-order pages may not be compound.
338 static void free_compound_page(struct page
*page
)
340 __free_pages_ok(page
, compound_order(page
));
343 void prep_compound_page(struct page
*page
, unsigned long order
)
346 int nr_pages
= 1 << order
;
348 set_compound_page_dtor(page
, free_compound_page
);
349 set_compound_order(page
, order
);
351 for (i
= 1; i
< nr_pages
; i
++) {
352 struct page
*p
= page
+ i
;
355 p
->first_page
= page
;
359 static int destroy_compound_page(struct page
*page
, unsigned long order
)
362 int nr_pages
= 1 << order
;
365 if (unlikely(compound_order(page
) != order
) ||
366 unlikely(!PageHead(page
))) {
371 __ClearPageHead(page
);
373 for (i
= 1; i
< nr_pages
; i
++) {
374 struct page
*p
= page
+ i
;
376 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
386 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
391 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
392 * and __GFP_HIGHMEM from hard or soft interrupt context.
394 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
395 for (i
= 0; i
< (1 << order
); i
++)
396 clear_highpage(page
+ i
);
399 static inline void set_page_order(struct page
*page
, int order
)
401 set_page_private(page
, order
);
402 __SetPageBuddy(page
);
405 static inline void rmv_page_order(struct page
*page
)
407 __ClearPageBuddy(page
);
408 set_page_private(page
, 0);
412 * Locate the struct page for both the matching buddy in our
413 * pair (buddy1) and the combined O(n+1) page they form (page).
415 * 1) Any buddy B1 will have an order O twin B2 which satisfies
416 * the following equation:
418 * For example, if the starting buddy (buddy2) is #8 its order
420 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
422 * 2) Any buddy B will have an order O+1 parent P which
423 * satisfies the following equation:
426 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
428 static inline struct page
*
429 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
431 unsigned long buddy_idx
= page_idx
^ (1 << order
);
433 return page
+ (buddy_idx
- page_idx
);
436 static inline unsigned long
437 __find_combined_index(unsigned long page_idx
, unsigned int order
)
439 return (page_idx
& ~(1 << order
));
443 * This function checks whether a page is free && is the buddy
444 * we can do coalesce a page and its buddy if
445 * (a) the buddy is not in a hole &&
446 * (b) the buddy is in the buddy system &&
447 * (c) a page and its buddy have the same order &&
448 * (d) a page and its buddy are in the same zone.
450 * For recording whether a page is in the buddy system, we use PG_buddy.
451 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
453 * For recording page's order, we use page_private(page).
455 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
458 if (!pfn_valid_within(page_to_pfn(buddy
)))
461 if (page_zone_id(page
) != page_zone_id(buddy
))
464 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
465 VM_BUG_ON(page_count(buddy
) != 0);
472 * Freeing function for a buddy system allocator.
474 * The concept of a buddy system is to maintain direct-mapped table
475 * (containing bit values) for memory blocks of various "orders".
476 * The bottom level table contains the map for the smallest allocatable
477 * units of memory (here, pages), and each level above it describes
478 * pairs of units from the levels below, hence, "buddies".
479 * At a high level, all that happens here is marking the table entry
480 * at the bottom level available, and propagating the changes upward
481 * as necessary, plus some accounting needed to play nicely with other
482 * parts of the VM system.
483 * At each level, we keep a list of pages, which are heads of continuous
484 * free pages of length of (1 << order) and marked with PG_buddy. Page's
485 * order is recorded in page_private(page) field.
486 * So when we are allocating or freeing one, we can derive the state of the
487 * other. That is, if we allocate a small block, and both were
488 * free, the remainder of the region must be split into blocks.
489 * If a block is freed, and its buddy is also free, then this
490 * triggers coalescing into a block of larger size.
495 static inline void __free_one_page(struct page
*page
,
496 struct zone
*zone
, unsigned int order
,
499 unsigned long page_idx
;
500 unsigned long combined_idx
;
503 if (unlikely(PageCompound(page
)))
504 if (unlikely(destroy_compound_page(page
, order
)))
507 VM_BUG_ON(migratetype
== -1);
509 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
511 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
512 VM_BUG_ON(bad_range(zone
, page
));
514 while (order
< MAX_ORDER
-1) {
515 buddy
= __page_find_buddy(page
, page_idx
, order
);
516 if (!page_is_buddy(page
, buddy
, order
))
519 /* Our buddy is free, merge with it and move up one order. */
520 list_del(&buddy
->lru
);
521 zone
->free_area
[order
].nr_free
--;
522 rmv_page_order(buddy
);
523 combined_idx
= __find_combined_index(page_idx
, order
);
524 page
= page
+ (combined_idx
- page_idx
);
525 page_idx
= combined_idx
;
528 set_page_order(page
, order
);
531 * If this is not the largest possible page, check if the buddy
532 * of the next-highest order is free. If it is, it's possible
533 * that pages are being freed that will coalesce soon. In case,
534 * that is happening, add the free page to the tail of the list
535 * so it's less likely to be used soon and more likely to be merged
536 * as a higher order page
538 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
539 struct page
*higher_page
, *higher_buddy
;
540 combined_idx
= __find_combined_index(page_idx
, order
);
541 higher_page
= page
+ combined_idx
- page_idx
;
542 higher_buddy
= __page_find_buddy(higher_page
, combined_idx
, order
+ 1);
543 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
544 list_add_tail(&page
->lru
,
545 &zone
->free_area
[order
].free_list
[migratetype
]);
550 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
552 zone
->free_area
[order
].nr_free
++;
556 * free_page_mlock() -- clean up attempts to free and mlocked() page.
557 * Page should not be on lru, so no need to fix that up.
558 * free_pages_check() will verify...
560 static inline void free_page_mlock(struct page
*page
)
562 __dec_zone_page_state(page
, NR_MLOCK
);
563 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
566 static inline int free_pages_check(struct page
*page
)
568 if (unlikely(page_mapcount(page
) |
569 (page
->mapping
!= NULL
) |
570 (atomic_read(&page
->_count
) != 0) |
571 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
575 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
576 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
581 * Frees a number of pages from the PCP lists
582 * Assumes all pages on list are in same zone, and of same order.
583 * count is the number of pages to free.
585 * If the zone was previously in an "all pages pinned" state then look to
586 * see if this freeing clears that state.
588 * And clear the zone's pages_scanned counter, to hold off the "all pages are
589 * pinned" detection logic.
591 static void free_pcppages_bulk(struct zone
*zone
, int count
,
592 struct per_cpu_pages
*pcp
)
598 spin_lock(&zone
->lock
);
599 zone
->all_unreclaimable
= 0;
600 zone
->pages_scanned
= 0;
604 struct list_head
*list
;
607 * Remove pages from lists in a round-robin fashion. A
608 * batch_free count is maintained that is incremented when an
609 * empty list is encountered. This is so more pages are freed
610 * off fuller lists instead of spinning excessively around empty
615 if (++migratetype
== MIGRATE_PCPTYPES
)
617 list
= &pcp
->lists
[migratetype
];
618 } while (list_empty(list
));
621 page
= list_entry(list
->prev
, struct page
, lru
);
622 /* must delete as __free_one_page list manipulates */
623 list_del(&page
->lru
);
624 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
625 __free_one_page(page
, zone
, 0, page_private(page
));
626 trace_mm_page_pcpu_drain(page
, 0, page_private(page
));
627 } while (--to_free
&& --batch_free
&& !list_empty(list
));
629 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
);
630 spin_unlock(&zone
->lock
);
633 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
636 spin_lock(&zone
->lock
);
637 zone
->all_unreclaimable
= 0;
638 zone
->pages_scanned
= 0;
640 __free_one_page(page
, zone
, order
, migratetype
);
641 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
642 spin_unlock(&zone
->lock
);
645 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
650 trace_mm_page_free_direct(page
, order
);
651 kmemcheck_free_shadow(page
, order
);
653 for (i
= 0; i
< (1 << order
); i
++) {
654 struct page
*pg
= page
+ i
;
658 bad
+= free_pages_check(pg
);
663 if (!PageHighMem(page
)) {
664 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
665 debug_check_no_obj_freed(page_address(page
),
668 arch_free_page(page
, order
);
669 kernel_map_pages(page
, 1 << order
, 0);
674 static void __free_pages_ok(struct page
*page
, unsigned int order
)
677 int wasMlocked
= __TestClearPageMlocked(page
);
679 if (!free_pages_prepare(page
, order
))
682 local_irq_save(flags
);
683 if (unlikely(wasMlocked
))
684 free_page_mlock(page
);
685 __count_vm_events(PGFREE
, 1 << order
);
686 free_one_page(page_zone(page
), page
, order
,
687 get_pageblock_migratetype(page
));
688 local_irq_restore(flags
);
692 * permit the bootmem allocator to evade page validation on high-order frees
694 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
697 __ClearPageReserved(page
);
698 set_page_count(page
, 0);
699 set_page_refcounted(page
);
705 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
706 struct page
*p
= &page
[loop
];
708 if (loop
+ 1 < BITS_PER_LONG
)
710 __ClearPageReserved(p
);
711 set_page_count(p
, 0);
714 set_page_refcounted(page
);
715 __free_pages(page
, order
);
721 * The order of subdivision here is critical for the IO subsystem.
722 * Please do not alter this order without good reasons and regression
723 * testing. Specifically, as large blocks of memory are subdivided,
724 * the order in which smaller blocks are delivered depends on the order
725 * they're subdivided in this function. This is the primary factor
726 * influencing the order in which pages are delivered to the IO
727 * subsystem according to empirical testing, and this is also justified
728 * by considering the behavior of a buddy system containing a single
729 * large block of memory acted on by a series of small allocations.
730 * This behavior is a critical factor in sglist merging's success.
734 static inline void expand(struct zone
*zone
, struct page
*page
,
735 int low
, int high
, struct free_area
*area
,
738 unsigned long size
= 1 << high
;
744 VM_BUG_ON(bad_range(zone
, &page
[size
]));
745 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
747 set_page_order(&page
[size
], high
);
752 * This page is about to be returned from the page allocator
754 static inline int check_new_page(struct page
*page
)
756 if (unlikely(page_mapcount(page
) |
757 (page
->mapping
!= NULL
) |
758 (atomic_read(&page
->_count
) != 0) |
759 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
766 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
770 for (i
= 0; i
< (1 << order
); i
++) {
771 struct page
*p
= page
+ i
;
772 if (unlikely(check_new_page(p
)))
776 set_page_private(page
, 0);
777 set_page_refcounted(page
);
779 arch_alloc_page(page
, order
);
780 kernel_map_pages(page
, 1 << order
, 1);
782 if (gfp_flags
& __GFP_ZERO
)
783 prep_zero_page(page
, order
, gfp_flags
);
785 if (order
&& (gfp_flags
& __GFP_COMP
))
786 prep_compound_page(page
, order
);
792 * Go through the free lists for the given migratetype and remove
793 * the smallest available page from the freelists
796 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
799 unsigned int current_order
;
800 struct free_area
* area
;
803 /* Find a page of the appropriate size in the preferred list */
804 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
805 area
= &(zone
->free_area
[current_order
]);
806 if (list_empty(&area
->free_list
[migratetype
]))
809 page
= list_entry(area
->free_list
[migratetype
].next
,
811 list_del(&page
->lru
);
812 rmv_page_order(page
);
814 expand(zone
, page
, order
, current_order
, area
, migratetype
);
823 * This array describes the order lists are fallen back to when
824 * the free lists for the desirable migrate type are depleted
826 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
827 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
828 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
829 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
830 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
834 * Move the free pages in a range to the free lists of the requested type.
835 * Note that start_page and end_pages are not aligned on a pageblock
836 * boundary. If alignment is required, use move_freepages_block()
838 static int move_freepages(struct zone
*zone
,
839 struct page
*start_page
, struct page
*end_page
,
846 #ifndef CONFIG_HOLES_IN_ZONE
848 * page_zone is not safe to call in this context when
849 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
850 * anyway as we check zone boundaries in move_freepages_block().
851 * Remove at a later date when no bug reports exist related to
852 * grouping pages by mobility
854 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
857 for (page
= start_page
; page
<= end_page
;) {
858 /* Make sure we are not inadvertently changing nodes */
859 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
861 if (!pfn_valid_within(page_to_pfn(page
))) {
866 if (!PageBuddy(page
)) {
871 order
= page_order(page
);
872 list_del(&page
->lru
);
874 &zone
->free_area
[order
].free_list
[migratetype
]);
876 pages_moved
+= 1 << order
;
882 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
885 unsigned long start_pfn
, end_pfn
;
886 struct page
*start_page
, *end_page
;
888 start_pfn
= page_to_pfn(page
);
889 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
890 start_page
= pfn_to_page(start_pfn
);
891 end_page
= start_page
+ pageblock_nr_pages
- 1;
892 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
894 /* Do not cross zone boundaries */
895 if (start_pfn
< zone
->zone_start_pfn
)
897 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
900 return move_freepages(zone
, start_page
, end_page
, migratetype
);
903 static void change_pageblock_range(struct page
*pageblock_page
,
904 int start_order
, int migratetype
)
906 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
908 while (nr_pageblocks
--) {
909 set_pageblock_migratetype(pageblock_page
, migratetype
);
910 pageblock_page
+= pageblock_nr_pages
;
914 /* Remove an element from the buddy allocator from the fallback list */
915 static inline struct page
*
916 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
918 struct free_area
* area
;
923 /* Find the largest possible block of pages in the other list */
924 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
926 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
927 migratetype
= fallbacks
[start_migratetype
][i
];
929 /* MIGRATE_RESERVE handled later if necessary */
930 if (migratetype
== MIGRATE_RESERVE
)
933 area
= &(zone
->free_area
[current_order
]);
934 if (list_empty(&area
->free_list
[migratetype
]))
937 page
= list_entry(area
->free_list
[migratetype
].next
,
942 * If breaking a large block of pages, move all free
943 * pages to the preferred allocation list. If falling
944 * back for a reclaimable kernel allocation, be more
945 * agressive about taking ownership of free pages
947 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
948 start_migratetype
== MIGRATE_RECLAIMABLE
||
949 page_group_by_mobility_disabled
) {
951 pages
= move_freepages_block(zone
, page
,
954 /* Claim the whole block if over half of it is free */
955 if (pages
>= (1 << (pageblock_order
-1)) ||
956 page_group_by_mobility_disabled
)
957 set_pageblock_migratetype(page
,
960 migratetype
= start_migratetype
;
963 /* Remove the page from the freelists */
964 list_del(&page
->lru
);
965 rmv_page_order(page
);
967 /* Take ownership for orders >= pageblock_order */
968 if (current_order
>= pageblock_order
)
969 change_pageblock_range(page
, current_order
,
972 expand(zone
, page
, order
, current_order
, area
, migratetype
);
974 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
975 start_migratetype
, migratetype
);
985 * Do the hard work of removing an element from the buddy allocator.
986 * Call me with the zone->lock already held.
988 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
994 page
= __rmqueue_smallest(zone
, order
, migratetype
);
996 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
997 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1000 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1001 * is used because __rmqueue_smallest is an inline function
1002 * and we want just one call site
1005 migratetype
= MIGRATE_RESERVE
;
1010 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1015 * Obtain a specified number of elements from the buddy allocator, all under
1016 * a single hold of the lock, for efficiency. Add them to the supplied list.
1017 * Returns the number of new pages which were placed at *list.
1019 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1020 unsigned long count
, struct list_head
*list
,
1021 int migratetype
, int cold
)
1025 spin_lock(&zone
->lock
);
1026 for (i
= 0; i
< count
; ++i
) {
1027 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1028 if (unlikely(page
== NULL
))
1032 * Split buddy pages returned by expand() are received here
1033 * in physical page order. The page is added to the callers and
1034 * list and the list head then moves forward. From the callers
1035 * perspective, the linked list is ordered by page number in
1036 * some conditions. This is useful for IO devices that can
1037 * merge IO requests if the physical pages are ordered
1040 if (likely(cold
== 0))
1041 list_add(&page
->lru
, list
);
1043 list_add_tail(&page
->lru
, list
);
1044 set_page_private(page
, migratetype
);
1047 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1048 spin_unlock(&zone
->lock
);
1054 * Called from the vmstat counter updater to drain pagesets of this
1055 * currently executing processor on remote nodes after they have
1058 * Note that this function must be called with the thread pinned to
1059 * a single processor.
1061 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1063 unsigned long flags
;
1066 local_irq_save(flags
);
1067 if (pcp
->count
>= pcp
->batch
)
1068 to_drain
= pcp
->batch
;
1070 to_drain
= pcp
->count
;
1071 free_pcppages_bulk(zone
, to_drain
, pcp
);
1072 pcp
->count
-= to_drain
;
1073 local_irq_restore(flags
);
1078 * Drain pages of the indicated processor.
1080 * The processor must either be the current processor and the
1081 * thread pinned to the current processor or a processor that
1084 static void drain_pages(unsigned int cpu
)
1086 unsigned long flags
;
1089 for_each_populated_zone(zone
) {
1090 struct per_cpu_pageset
*pset
;
1091 struct per_cpu_pages
*pcp
;
1093 local_irq_save(flags
);
1094 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1097 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1099 local_irq_restore(flags
);
1104 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1106 void drain_local_pages(void *arg
)
1108 drain_pages(smp_processor_id());
1112 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1114 void drain_all_pages(void)
1116 on_each_cpu(drain_local_pages
, NULL
, 1);
1119 #ifdef CONFIG_HIBERNATION
1121 void mark_free_pages(struct zone
*zone
)
1123 unsigned long pfn
, max_zone_pfn
;
1124 unsigned long flags
;
1126 struct list_head
*curr
;
1128 if (!zone
->spanned_pages
)
1131 spin_lock_irqsave(&zone
->lock
, flags
);
1133 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1134 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1135 if (pfn_valid(pfn
)) {
1136 struct page
*page
= pfn_to_page(pfn
);
1138 if (!swsusp_page_is_forbidden(page
))
1139 swsusp_unset_page_free(page
);
1142 for_each_migratetype_order(order
, t
) {
1143 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1146 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1147 for (i
= 0; i
< (1UL << order
); i
++)
1148 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1151 spin_unlock_irqrestore(&zone
->lock
, flags
);
1153 #endif /* CONFIG_PM */
1156 * Free a 0-order page
1157 * cold == 1 ? free a cold page : free a hot page
1159 void free_hot_cold_page(struct page
*page
, int cold
)
1161 struct zone
*zone
= page_zone(page
);
1162 struct per_cpu_pages
*pcp
;
1163 unsigned long flags
;
1165 int wasMlocked
= __TestClearPageMlocked(page
);
1167 if (!free_pages_prepare(page
, 0))
1170 migratetype
= get_pageblock_migratetype(page
);
1171 set_page_private(page
, migratetype
);
1172 local_irq_save(flags
);
1173 if (unlikely(wasMlocked
))
1174 free_page_mlock(page
);
1175 __count_vm_event(PGFREE
);
1178 * We only track unmovable, reclaimable and movable on pcp lists.
1179 * Free ISOLATE pages back to the allocator because they are being
1180 * offlined but treat RESERVE as movable pages so we can get those
1181 * areas back if necessary. Otherwise, we may have to free
1182 * excessively into the page allocator
1184 if (migratetype
>= MIGRATE_PCPTYPES
) {
1185 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1186 free_one_page(zone
, page
, 0, migratetype
);
1189 migratetype
= MIGRATE_MOVABLE
;
1192 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1194 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1196 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1198 if (pcp
->count
>= pcp
->high
) {
1199 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1200 pcp
->count
-= pcp
->batch
;
1204 local_irq_restore(flags
);
1208 * split_page takes a non-compound higher-order page, and splits it into
1209 * n (1<<order) sub-pages: page[0..n]
1210 * Each sub-page must be freed individually.
1212 * Note: this is probably too low level an operation for use in drivers.
1213 * Please consult with lkml before using this in your driver.
1215 void split_page(struct page
*page
, unsigned int order
)
1219 VM_BUG_ON(PageCompound(page
));
1220 VM_BUG_ON(!page_count(page
));
1222 #ifdef CONFIG_KMEMCHECK
1224 * Split shadow pages too, because free(page[0]) would
1225 * otherwise free the whole shadow.
1227 if (kmemcheck_page_is_tracked(page
))
1228 split_page(virt_to_page(page
[0].shadow
), order
);
1231 for (i
= 1; i
< (1 << order
); i
++)
1232 set_page_refcounted(page
+ i
);
1236 * Similar to split_page except the page is already free. As this is only
1237 * being used for migration, the migratetype of the block also changes.
1238 * As this is called with interrupts disabled, the caller is responsible
1239 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1242 * Note: this is probably too low level an operation for use in drivers.
1243 * Please consult with lkml before using this in your driver.
1245 int split_free_page(struct page
*page
)
1248 unsigned long watermark
;
1251 BUG_ON(!PageBuddy(page
));
1253 zone
= page_zone(page
);
1254 order
= page_order(page
);
1256 /* Obey watermarks as if the page was being allocated */
1257 watermark
= low_wmark_pages(zone
) + (1 << order
);
1258 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1261 /* Remove page from free list */
1262 list_del(&page
->lru
);
1263 zone
->free_area
[order
].nr_free
--;
1264 rmv_page_order(page
);
1265 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1UL << order
));
1267 /* Split into individual pages */
1268 set_page_refcounted(page
);
1269 split_page(page
, order
);
1271 if (order
>= pageblock_order
- 1) {
1272 struct page
*endpage
= page
+ (1 << order
) - 1;
1273 for (; page
< endpage
; page
+= pageblock_nr_pages
)
1274 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1281 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1282 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1286 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1287 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1290 unsigned long flags
;
1292 int cold
= !!(gfp_flags
& __GFP_COLD
);
1295 if (likely(order
== 0)) {
1296 struct per_cpu_pages
*pcp
;
1297 struct list_head
*list
;
1299 local_irq_save(flags
);
1300 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1301 list
= &pcp
->lists
[migratetype
];
1302 if (list_empty(list
)) {
1303 pcp
->count
+= rmqueue_bulk(zone
, 0,
1306 if (unlikely(list_empty(list
)))
1311 page
= list_entry(list
->prev
, struct page
, lru
);
1313 page
= list_entry(list
->next
, struct page
, lru
);
1315 list_del(&page
->lru
);
1318 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1320 * __GFP_NOFAIL is not to be used in new code.
1322 * All __GFP_NOFAIL callers should be fixed so that they
1323 * properly detect and handle allocation failures.
1325 * We most definitely don't want callers attempting to
1326 * allocate greater than order-1 page units with
1329 WARN_ON_ONCE(order
> 1);
1331 spin_lock_irqsave(&zone
->lock
, flags
);
1332 page
= __rmqueue(zone
, order
, migratetype
);
1333 spin_unlock(&zone
->lock
);
1336 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << order
));
1339 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1340 zone_statistics(preferred_zone
, zone
);
1341 local_irq_restore(flags
);
1343 VM_BUG_ON(bad_range(zone
, page
));
1344 if (prep_new_page(page
, order
, gfp_flags
))
1349 local_irq_restore(flags
);
1353 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1354 #define ALLOC_WMARK_MIN WMARK_MIN
1355 #define ALLOC_WMARK_LOW WMARK_LOW
1356 #define ALLOC_WMARK_HIGH WMARK_HIGH
1357 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1359 /* Mask to get the watermark bits */
1360 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1362 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1363 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1364 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1366 #ifdef CONFIG_FAIL_PAGE_ALLOC
1368 static struct fail_page_alloc_attr
{
1369 struct fault_attr attr
;
1371 u32 ignore_gfp_highmem
;
1372 u32 ignore_gfp_wait
;
1375 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1377 struct dentry
*ignore_gfp_highmem_file
;
1378 struct dentry
*ignore_gfp_wait_file
;
1379 struct dentry
*min_order_file
;
1381 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1383 } fail_page_alloc
= {
1384 .attr
= FAULT_ATTR_INITIALIZER
,
1385 .ignore_gfp_wait
= 1,
1386 .ignore_gfp_highmem
= 1,
1390 static int __init
setup_fail_page_alloc(char *str
)
1392 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1394 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1396 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1398 if (order
< fail_page_alloc
.min_order
)
1400 if (gfp_mask
& __GFP_NOFAIL
)
1402 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1404 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1407 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1410 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1412 static int __init
fail_page_alloc_debugfs(void)
1414 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1418 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1422 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1424 fail_page_alloc
.ignore_gfp_wait_file
=
1425 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1426 &fail_page_alloc
.ignore_gfp_wait
);
1428 fail_page_alloc
.ignore_gfp_highmem_file
=
1429 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1430 &fail_page_alloc
.ignore_gfp_highmem
);
1431 fail_page_alloc
.min_order_file
=
1432 debugfs_create_u32("min-order", mode
, dir
,
1433 &fail_page_alloc
.min_order
);
1435 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1436 !fail_page_alloc
.ignore_gfp_highmem_file
||
1437 !fail_page_alloc
.min_order_file
) {
1439 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1440 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1441 debugfs_remove(fail_page_alloc
.min_order_file
);
1442 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1448 late_initcall(fail_page_alloc_debugfs
);
1450 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1452 #else /* CONFIG_FAIL_PAGE_ALLOC */
1454 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1459 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1462 * Return 1 if free pages are above 'mark'. This takes into account the order
1463 * of the allocation.
1465 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1466 int classzone_idx
, int alloc_flags
)
1468 /* free_pages my go negative - that's OK */
1470 long free_pages
= zone_nr_free_pages(z
) - (1 << order
) + 1;
1473 if (alloc_flags
& ALLOC_HIGH
)
1475 if (alloc_flags
& ALLOC_HARDER
)
1478 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1480 for (o
= 0; o
< order
; o
++) {
1481 /* At the next order, this order's pages become unavailable */
1482 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1484 /* Require fewer higher order pages to be free */
1487 if (free_pages
<= min
)
1495 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1496 * skip over zones that are not allowed by the cpuset, or that have
1497 * been recently (in last second) found to be nearly full. See further
1498 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1499 * that have to skip over a lot of full or unallowed zones.
1501 * If the zonelist cache is present in the passed in zonelist, then
1502 * returns a pointer to the allowed node mask (either the current
1503 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1505 * If the zonelist cache is not available for this zonelist, does
1506 * nothing and returns NULL.
1508 * If the fullzones BITMAP in the zonelist cache is stale (more than
1509 * a second since last zap'd) then we zap it out (clear its bits.)
1511 * We hold off even calling zlc_setup, until after we've checked the
1512 * first zone in the zonelist, on the theory that most allocations will
1513 * be satisfied from that first zone, so best to examine that zone as
1514 * quickly as we can.
1516 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1518 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1519 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1521 zlc
= zonelist
->zlcache_ptr
;
1525 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1526 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1527 zlc
->last_full_zap
= jiffies
;
1530 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1531 &cpuset_current_mems_allowed
:
1532 &node_states
[N_HIGH_MEMORY
];
1533 return allowednodes
;
1537 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1538 * if it is worth looking at further for free memory:
1539 * 1) Check that the zone isn't thought to be full (doesn't have its
1540 * bit set in the zonelist_cache fullzones BITMAP).
1541 * 2) Check that the zones node (obtained from the zonelist_cache
1542 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1543 * Return true (non-zero) if zone is worth looking at further, or
1544 * else return false (zero) if it is not.
1546 * This check -ignores- the distinction between various watermarks,
1547 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1548 * found to be full for any variation of these watermarks, it will
1549 * be considered full for up to one second by all requests, unless
1550 * we are so low on memory on all allowed nodes that we are forced
1551 * into the second scan of the zonelist.
1553 * In the second scan we ignore this zonelist cache and exactly
1554 * apply the watermarks to all zones, even it is slower to do so.
1555 * We are low on memory in the second scan, and should leave no stone
1556 * unturned looking for a free page.
1558 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1559 nodemask_t
*allowednodes
)
1561 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1562 int i
; /* index of *z in zonelist zones */
1563 int n
; /* node that zone *z is on */
1565 zlc
= zonelist
->zlcache_ptr
;
1569 i
= z
- zonelist
->_zonerefs
;
1572 /* This zone is worth trying if it is allowed but not full */
1573 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1577 * Given 'z' scanning a zonelist, set the corresponding bit in
1578 * zlc->fullzones, so that subsequent attempts to allocate a page
1579 * from that zone don't waste time re-examining it.
1581 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1583 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1584 int i
; /* index of *z in zonelist zones */
1586 zlc
= zonelist
->zlcache_ptr
;
1590 i
= z
- zonelist
->_zonerefs
;
1592 set_bit(i
, zlc
->fullzones
);
1595 #else /* CONFIG_NUMA */
1597 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1602 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1603 nodemask_t
*allowednodes
)
1608 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1611 #endif /* CONFIG_NUMA */
1614 * get_page_from_freelist goes through the zonelist trying to allocate
1617 static struct page
*
1618 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1619 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1620 struct zone
*preferred_zone
, int migratetype
)
1623 struct page
*page
= NULL
;
1626 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1627 int zlc_active
= 0; /* set if using zonelist_cache */
1628 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1630 classzone_idx
= zone_idx(preferred_zone
);
1633 * Scan zonelist, looking for a zone with enough free.
1634 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1636 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1637 high_zoneidx
, nodemask
) {
1638 if (NUMA_BUILD
&& zlc_active
&&
1639 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1641 if ((alloc_flags
& ALLOC_CPUSET
) &&
1642 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1645 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1646 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1650 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1651 if (zone_watermark_ok(zone
, order
, mark
,
1652 classzone_idx
, alloc_flags
))
1655 if (zone_reclaim_mode
== 0)
1656 goto this_zone_full
;
1658 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1660 case ZONE_RECLAIM_NOSCAN
:
1663 case ZONE_RECLAIM_FULL
:
1664 /* scanned but unreclaimable */
1665 goto this_zone_full
;
1667 /* did we reclaim enough */
1668 if (!zone_watermark_ok(zone
, order
, mark
,
1669 classzone_idx
, alloc_flags
))
1670 goto this_zone_full
;
1675 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1676 gfp_mask
, migratetype
);
1681 zlc_mark_zone_full(zonelist
, z
);
1683 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1685 * we do zlc_setup after the first zone is tried but only
1686 * if there are multiple nodes make it worthwhile
1688 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1694 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1695 /* Disable zlc cache for second zonelist scan */
1703 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1704 unsigned long pages_reclaimed
)
1706 /* Do not loop if specifically requested */
1707 if (gfp_mask
& __GFP_NORETRY
)
1711 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1712 * means __GFP_NOFAIL, but that may not be true in other
1715 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1719 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1720 * specified, then we retry until we no longer reclaim any pages
1721 * (above), or we've reclaimed an order of pages at least as
1722 * large as the allocation's order. In both cases, if the
1723 * allocation still fails, we stop retrying.
1725 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1729 * Don't let big-order allocations loop unless the caller
1730 * explicitly requests that.
1732 if (gfp_mask
& __GFP_NOFAIL
)
1738 static inline struct page
*
1739 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1740 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1741 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1746 /* Acquire the OOM killer lock for the zones in zonelist */
1747 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
1748 schedule_timeout_uninterruptible(1);
1753 * Go through the zonelist yet one more time, keep very high watermark
1754 * here, this is only to catch a parallel oom killing, we must fail if
1755 * we're still under heavy pressure.
1757 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1758 order
, zonelist
, high_zoneidx
,
1759 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1760 preferred_zone
, migratetype
);
1764 if (!(gfp_mask
& __GFP_NOFAIL
)) {
1765 /* The OOM killer will not help higher order allocs */
1766 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1768 /* The OOM killer does not needlessly kill tasks for lowmem */
1769 if (high_zoneidx
< ZONE_NORMAL
)
1772 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1773 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1774 * The caller should handle page allocation failure by itself if
1775 * it specifies __GFP_THISNODE.
1776 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1778 if (gfp_mask
& __GFP_THISNODE
)
1781 /* Exhausted what can be done so it's blamo time */
1782 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
);
1785 clear_zonelist_oom(zonelist
, gfp_mask
);
1789 #ifdef CONFIG_COMPACTION
1790 /* Try memory compaction for high-order allocations before reclaim */
1791 static struct page
*
1792 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1793 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1794 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1795 int migratetype
, unsigned long *did_some_progress
)
1799 if (!order
|| compaction_deferred(preferred_zone
))
1802 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
1804 if (*did_some_progress
!= COMPACT_SKIPPED
) {
1806 /* Page migration frees to the PCP lists but we want merging */
1807 drain_pages(get_cpu());
1810 page
= get_page_from_freelist(gfp_mask
, nodemask
,
1811 order
, zonelist
, high_zoneidx
,
1812 alloc_flags
, preferred_zone
,
1815 preferred_zone
->compact_considered
= 0;
1816 preferred_zone
->compact_defer_shift
= 0;
1817 count_vm_event(COMPACTSUCCESS
);
1822 * It's bad if compaction run occurs and fails.
1823 * The most likely reason is that pages exist,
1824 * but not enough to satisfy watermarks.
1826 count_vm_event(COMPACTFAIL
);
1827 defer_compaction(preferred_zone
);
1835 static inline struct page
*
1836 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1837 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1838 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1839 int migratetype
, unsigned long *did_some_progress
)
1843 #endif /* CONFIG_COMPACTION */
1845 /* The really slow allocator path where we enter direct reclaim */
1846 static inline struct page
*
1847 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1848 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1849 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1850 int migratetype
, unsigned long *did_some_progress
)
1852 struct page
*page
= NULL
;
1853 struct reclaim_state reclaim_state
;
1854 struct task_struct
*p
= current
;
1855 bool drained
= false;
1859 /* We now go into synchronous reclaim */
1860 cpuset_memory_pressure_bump();
1861 p
->flags
|= PF_MEMALLOC
;
1862 lockdep_set_current_reclaim_state(gfp_mask
);
1863 reclaim_state
.reclaimed_slab
= 0;
1864 p
->reclaim_state
= &reclaim_state
;
1866 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1868 p
->reclaim_state
= NULL
;
1869 lockdep_clear_current_reclaim_state();
1870 p
->flags
&= ~PF_MEMALLOC
;
1874 if (unlikely(!(*did_some_progress
)))
1878 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1879 zonelist
, high_zoneidx
,
1880 alloc_flags
, preferred_zone
,
1884 * If an allocation failed after direct reclaim, it could be because
1885 * pages are pinned on the per-cpu lists. Drain them and try again
1887 if (!page
&& !drained
) {
1897 * This is called in the allocator slow-path if the allocation request is of
1898 * sufficient urgency to ignore watermarks and take other desperate measures
1900 static inline struct page
*
1901 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1902 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1903 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1909 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1910 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1911 preferred_zone
, migratetype
);
1913 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1914 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1915 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1921 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1922 enum zone_type high_zoneidx
)
1927 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1928 wakeup_kswapd(zone
, order
);
1932 gfp_to_alloc_flags(gfp_t gfp_mask
)
1934 struct task_struct
*p
= current
;
1935 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
1936 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1938 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1939 BUILD_BUG_ON(__GFP_HIGH
!= ALLOC_HIGH
);
1942 * The caller may dip into page reserves a bit more if the caller
1943 * cannot run direct reclaim, or if the caller has realtime scheduling
1944 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1945 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1947 alloc_flags
|= (gfp_mask
& __GFP_HIGH
);
1950 alloc_flags
|= ALLOC_HARDER
;
1952 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1953 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1955 alloc_flags
&= ~ALLOC_CPUSET
;
1956 } else if (unlikely(rt_task(p
)) && !in_interrupt())
1957 alloc_flags
|= ALLOC_HARDER
;
1959 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
1960 if (!in_interrupt() &&
1961 ((p
->flags
& PF_MEMALLOC
) ||
1962 unlikely(test_thread_flag(TIF_MEMDIE
))))
1963 alloc_flags
|= ALLOC_NO_WATERMARKS
;
1969 static inline struct page
*
1970 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
1971 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1972 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1975 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1976 struct page
*page
= NULL
;
1978 unsigned long pages_reclaimed
= 0;
1979 unsigned long did_some_progress
;
1980 struct task_struct
*p
= current
;
1983 * In the slowpath, we sanity check order to avoid ever trying to
1984 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
1985 * be using allocators in order of preference for an area that is
1988 if (order
>= MAX_ORDER
) {
1989 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
1994 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1995 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1996 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1997 * using a larger set of nodes after it has established that the
1998 * allowed per node queues are empty and that nodes are
2001 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2005 wake_all_kswapd(order
, zonelist
, high_zoneidx
);
2008 * OK, we're below the kswapd watermark and have kicked background
2009 * reclaim. Now things get more complex, so set up alloc_flags according
2010 * to how we want to proceed.
2012 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2014 /* This is the last chance, in general, before the goto nopage. */
2015 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2016 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2017 preferred_zone
, migratetype
);
2022 /* Allocate without watermarks if the context allows */
2023 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2024 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2025 zonelist
, high_zoneidx
, nodemask
,
2026 preferred_zone
, migratetype
);
2031 /* Atomic allocations - we can't balance anything */
2035 /* Avoid recursion of direct reclaim */
2036 if (p
->flags
& PF_MEMALLOC
)
2039 /* Avoid allocations with no watermarks from looping endlessly */
2040 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2043 /* Try direct compaction */
2044 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2045 zonelist
, high_zoneidx
,
2047 alloc_flags
, preferred_zone
,
2048 migratetype
, &did_some_progress
);
2052 /* Try direct reclaim and then allocating */
2053 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2054 zonelist
, high_zoneidx
,
2056 alloc_flags
, preferred_zone
,
2057 migratetype
, &did_some_progress
);
2062 * If we failed to make any progress reclaiming, then we are
2063 * running out of options and have to consider going OOM
2065 if (!did_some_progress
) {
2066 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2067 if (oom_killer_disabled
)
2069 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2070 zonelist
, high_zoneidx
,
2071 nodemask
, preferred_zone
,
2076 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2078 * The oom killer is not called for high-order
2079 * allocations that may fail, so if no progress
2080 * is being made, there are no other options and
2081 * retrying is unlikely to help.
2083 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2086 * The oom killer is not called for lowmem
2087 * allocations to prevent needlessly killing
2090 if (high_zoneidx
< ZONE_NORMAL
)
2098 /* Check if we should retry the allocation */
2099 pages_reclaimed
+= did_some_progress
;
2100 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
2101 /* Wait for some write requests to complete then retry */
2102 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
2107 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
2108 printk(KERN_WARNING
"%s: page allocation failure."
2109 " order:%d, mode:0x%x\n",
2110 p
->comm
, order
, gfp_mask
);
2116 if (kmemcheck_enabled
)
2117 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2123 * This is the 'heart' of the zoned buddy allocator.
2126 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2127 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2129 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2130 struct zone
*preferred_zone
;
2132 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2134 gfp_mask
&= gfp_allowed_mask
;
2136 lockdep_trace_alloc(gfp_mask
);
2138 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2140 if (should_fail_alloc_page(gfp_mask
, order
))
2144 * Check the zones suitable for the gfp_mask contain at least one
2145 * valid zone. It's possible to have an empty zonelist as a result
2146 * of GFP_THISNODE and a memoryless node
2148 if (unlikely(!zonelist
->_zonerefs
->zone
))
2152 /* The preferred zone is used for statistics later */
2153 first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
, &preferred_zone
);
2154 if (!preferred_zone
) {
2159 /* First allocation attempt */
2160 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2161 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
2162 preferred_zone
, migratetype
);
2163 if (unlikely(!page
))
2164 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2165 zonelist
, high_zoneidx
, nodemask
,
2166 preferred_zone
, migratetype
);
2169 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2172 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2175 * Common helper functions.
2177 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2182 * __get_free_pages() returns a 32-bit address, which cannot represent
2185 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2187 page
= alloc_pages(gfp_mask
, order
);
2190 return (unsigned long) page_address(page
);
2192 EXPORT_SYMBOL(__get_free_pages
);
2194 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2196 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2198 EXPORT_SYMBOL(get_zeroed_page
);
2200 void __pagevec_free(struct pagevec
*pvec
)
2202 int i
= pagevec_count(pvec
);
2205 trace_mm_pagevec_free(pvec
->pages
[i
], pvec
->cold
);
2206 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
2210 void __free_pages(struct page
*page
, unsigned int order
)
2212 if (put_page_testzero(page
)) {
2214 free_hot_cold_page(page
, 0);
2216 __free_pages_ok(page
, order
);
2220 EXPORT_SYMBOL(__free_pages
);
2222 void free_pages(unsigned long addr
, unsigned int order
)
2225 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2226 __free_pages(virt_to_page((void *)addr
), order
);
2230 EXPORT_SYMBOL(free_pages
);
2233 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2234 * @size: the number of bytes to allocate
2235 * @gfp_mask: GFP flags for the allocation
2237 * This function is similar to alloc_pages(), except that it allocates the
2238 * minimum number of pages to satisfy the request. alloc_pages() can only
2239 * allocate memory in power-of-two pages.
2241 * This function is also limited by MAX_ORDER.
2243 * Memory allocated by this function must be released by free_pages_exact().
2245 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2247 unsigned int order
= get_order(size
);
2250 addr
= __get_free_pages(gfp_mask
, order
);
2252 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2253 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2255 split_page(virt_to_page((void *)addr
), order
);
2256 while (used
< alloc_end
) {
2262 return (void *)addr
;
2264 EXPORT_SYMBOL(alloc_pages_exact
);
2267 * free_pages_exact - release memory allocated via alloc_pages_exact()
2268 * @virt: the value returned by alloc_pages_exact.
2269 * @size: size of allocation, same value as passed to alloc_pages_exact().
2271 * Release the memory allocated by a previous call to alloc_pages_exact.
2273 void free_pages_exact(void *virt
, size_t size
)
2275 unsigned long addr
= (unsigned long)virt
;
2276 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2278 while (addr
< end
) {
2283 EXPORT_SYMBOL(free_pages_exact
);
2285 static unsigned int nr_free_zone_pages(int offset
)
2290 /* Just pick one node, since fallback list is circular */
2291 unsigned int sum
= 0;
2293 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2295 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2296 unsigned long size
= zone
->present_pages
;
2297 unsigned long high
= high_wmark_pages(zone
);
2306 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2308 unsigned int nr_free_buffer_pages(void)
2310 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2312 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2315 * Amount of free RAM allocatable within all zones
2317 unsigned int nr_free_pagecache_pages(void)
2319 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2322 static inline void show_node(struct zone
*zone
)
2325 printk("Node %d ", zone_to_nid(zone
));
2328 void si_meminfo(struct sysinfo
*val
)
2330 val
->totalram
= totalram_pages
;
2332 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2333 val
->bufferram
= nr_blockdev_pages();
2334 val
->totalhigh
= totalhigh_pages
;
2335 val
->freehigh
= nr_free_highpages();
2336 val
->mem_unit
= PAGE_SIZE
;
2339 EXPORT_SYMBOL(si_meminfo
);
2342 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2344 pg_data_t
*pgdat
= NODE_DATA(nid
);
2346 val
->totalram
= pgdat
->node_present_pages
;
2347 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2348 #ifdef CONFIG_HIGHMEM
2349 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2350 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2356 val
->mem_unit
= PAGE_SIZE
;
2360 #define K(x) ((x) << (PAGE_SHIFT-10))
2363 * Show free area list (used inside shift_scroll-lock stuff)
2364 * We also calculate the percentage fragmentation. We do this by counting the
2365 * memory on each free list with the exception of the first item on the list.
2367 void show_free_areas(void)
2372 for_each_populated_zone(zone
) {
2374 printk("%s per-cpu:\n", zone
->name
);
2376 for_each_online_cpu(cpu
) {
2377 struct per_cpu_pageset
*pageset
;
2379 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2381 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2382 cpu
, pageset
->pcp
.high
,
2383 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2387 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2388 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2390 " dirty:%lu writeback:%lu unstable:%lu\n"
2391 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2392 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2393 global_page_state(NR_ACTIVE_ANON
),
2394 global_page_state(NR_INACTIVE_ANON
),
2395 global_page_state(NR_ISOLATED_ANON
),
2396 global_page_state(NR_ACTIVE_FILE
),
2397 global_page_state(NR_INACTIVE_FILE
),
2398 global_page_state(NR_ISOLATED_FILE
),
2399 global_page_state(NR_UNEVICTABLE
),
2400 global_page_state(NR_FILE_DIRTY
),
2401 global_page_state(NR_WRITEBACK
),
2402 global_page_state(NR_UNSTABLE_NFS
),
2403 global_page_state(NR_FREE_PAGES
),
2404 global_page_state(NR_SLAB_RECLAIMABLE
),
2405 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2406 global_page_state(NR_FILE_MAPPED
),
2407 global_page_state(NR_SHMEM
),
2408 global_page_state(NR_PAGETABLE
),
2409 global_page_state(NR_BOUNCE
));
2411 for_each_populated_zone(zone
) {
2420 " active_anon:%lukB"
2421 " inactive_anon:%lukB"
2422 " active_file:%lukB"
2423 " inactive_file:%lukB"
2424 " unevictable:%lukB"
2425 " isolated(anon):%lukB"
2426 " isolated(file):%lukB"
2433 " slab_reclaimable:%lukB"
2434 " slab_unreclaimable:%lukB"
2435 " kernel_stack:%lukB"
2439 " writeback_tmp:%lukB"
2440 " pages_scanned:%lu"
2441 " all_unreclaimable? %s"
2444 K(zone_nr_free_pages(zone
)),
2445 K(min_wmark_pages(zone
)),
2446 K(low_wmark_pages(zone
)),
2447 K(high_wmark_pages(zone
)),
2448 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2449 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2450 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2451 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2452 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2453 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
2454 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
2455 K(zone
->present_pages
),
2456 K(zone_page_state(zone
, NR_MLOCK
)),
2457 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
2458 K(zone_page_state(zone
, NR_WRITEBACK
)),
2459 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
2460 K(zone_page_state(zone
, NR_SHMEM
)),
2461 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
2462 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
2463 zone_page_state(zone
, NR_KERNEL_STACK
) *
2465 K(zone_page_state(zone
, NR_PAGETABLE
)),
2466 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
2467 K(zone_page_state(zone
, NR_BOUNCE
)),
2468 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
2469 zone
->pages_scanned
,
2470 (zone
->all_unreclaimable
? "yes" : "no")
2472 printk("lowmem_reserve[]:");
2473 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2474 printk(" %lu", zone
->lowmem_reserve
[i
]);
2478 for_each_populated_zone(zone
) {
2479 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2482 printk("%s: ", zone
->name
);
2484 spin_lock_irqsave(&zone
->lock
, flags
);
2485 for (order
= 0; order
< MAX_ORDER
; order
++) {
2486 nr
[order
] = zone
->free_area
[order
].nr_free
;
2487 total
+= nr
[order
] << order
;
2489 spin_unlock_irqrestore(&zone
->lock
, flags
);
2490 for (order
= 0; order
< MAX_ORDER
; order
++)
2491 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2492 printk("= %lukB\n", K(total
));
2495 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2497 show_swap_cache_info();
2500 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2502 zoneref
->zone
= zone
;
2503 zoneref
->zone_idx
= zone_idx(zone
);
2507 * Builds allocation fallback zone lists.
2509 * Add all populated zones of a node to the zonelist.
2511 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2512 int nr_zones
, enum zone_type zone_type
)
2516 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2521 zone
= pgdat
->node_zones
+ zone_type
;
2522 if (populated_zone(zone
)) {
2523 zoneref_set_zone(zone
,
2524 &zonelist
->_zonerefs
[nr_zones
++]);
2525 check_highest_zone(zone_type
);
2528 } while (zone_type
);
2535 * 0 = automatic detection of better ordering.
2536 * 1 = order by ([node] distance, -zonetype)
2537 * 2 = order by (-zonetype, [node] distance)
2539 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2540 * the same zonelist. So only NUMA can configure this param.
2542 #define ZONELIST_ORDER_DEFAULT 0
2543 #define ZONELIST_ORDER_NODE 1
2544 #define ZONELIST_ORDER_ZONE 2
2546 /* zonelist order in the kernel.
2547 * set_zonelist_order() will set this to NODE or ZONE.
2549 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2550 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2554 /* The value user specified ....changed by config */
2555 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2556 /* string for sysctl */
2557 #define NUMA_ZONELIST_ORDER_LEN 16
2558 char numa_zonelist_order
[16] = "default";
2561 * interface for configure zonelist ordering.
2562 * command line option "numa_zonelist_order"
2563 * = "[dD]efault - default, automatic configuration.
2564 * = "[nN]ode - order by node locality, then by zone within node
2565 * = "[zZ]one - order by zone, then by locality within zone
2568 static int __parse_numa_zonelist_order(char *s
)
2570 if (*s
== 'd' || *s
== 'D') {
2571 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2572 } else if (*s
== 'n' || *s
== 'N') {
2573 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2574 } else if (*s
== 'z' || *s
== 'Z') {
2575 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2578 "Ignoring invalid numa_zonelist_order value: "
2585 static __init
int setup_numa_zonelist_order(char *s
)
2588 return __parse_numa_zonelist_order(s
);
2591 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2594 * sysctl handler for numa_zonelist_order
2596 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2597 void __user
*buffer
, size_t *length
,
2600 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2602 static DEFINE_MUTEX(zl_order_mutex
);
2604 mutex_lock(&zl_order_mutex
);
2606 strcpy(saved_string
, (char*)table
->data
);
2607 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
2611 int oldval
= user_zonelist_order
;
2612 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2614 * bogus value. restore saved string
2616 strncpy((char*)table
->data
, saved_string
,
2617 NUMA_ZONELIST_ORDER_LEN
);
2618 user_zonelist_order
= oldval
;
2619 } else if (oldval
!= user_zonelist_order
) {
2620 mutex_lock(&zonelists_mutex
);
2621 build_all_zonelists(NULL
);
2622 mutex_unlock(&zonelists_mutex
);
2626 mutex_unlock(&zl_order_mutex
);
2631 #define MAX_NODE_LOAD (nr_online_nodes)
2632 static int node_load
[MAX_NUMNODES
];
2635 * find_next_best_node - find the next node that should appear in a given node's fallback list
2636 * @node: node whose fallback list we're appending
2637 * @used_node_mask: nodemask_t of already used nodes
2639 * We use a number of factors to determine which is the next node that should
2640 * appear on a given node's fallback list. The node should not have appeared
2641 * already in @node's fallback list, and it should be the next closest node
2642 * according to the distance array (which contains arbitrary distance values
2643 * from each node to each node in the system), and should also prefer nodes
2644 * with no CPUs, since presumably they'll have very little allocation pressure
2645 * on them otherwise.
2646 * It returns -1 if no node is found.
2648 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2651 int min_val
= INT_MAX
;
2653 const struct cpumask
*tmp
= cpumask_of_node(0);
2655 /* Use the local node if we haven't already */
2656 if (!node_isset(node
, *used_node_mask
)) {
2657 node_set(node
, *used_node_mask
);
2661 for_each_node_state(n
, N_HIGH_MEMORY
) {
2663 /* Don't want a node to appear more than once */
2664 if (node_isset(n
, *used_node_mask
))
2667 /* Use the distance array to find the distance */
2668 val
= node_distance(node
, n
);
2670 /* Penalize nodes under us ("prefer the next node") */
2673 /* Give preference to headless and unused nodes */
2674 tmp
= cpumask_of_node(n
);
2675 if (!cpumask_empty(tmp
))
2676 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2678 /* Slight preference for less loaded node */
2679 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2680 val
+= node_load
[n
];
2682 if (val
< min_val
) {
2689 node_set(best_node
, *used_node_mask
);
2696 * Build zonelists ordered by node and zones within node.
2697 * This results in maximum locality--normal zone overflows into local
2698 * DMA zone, if any--but risks exhausting DMA zone.
2700 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2703 struct zonelist
*zonelist
;
2705 zonelist
= &pgdat
->node_zonelists
[0];
2706 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2708 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2710 zonelist
->_zonerefs
[j
].zone
= NULL
;
2711 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2715 * Build gfp_thisnode zonelists
2717 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2720 struct zonelist
*zonelist
;
2722 zonelist
= &pgdat
->node_zonelists
[1];
2723 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2724 zonelist
->_zonerefs
[j
].zone
= NULL
;
2725 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2729 * Build zonelists ordered by zone and nodes within zones.
2730 * This results in conserving DMA zone[s] until all Normal memory is
2731 * exhausted, but results in overflowing to remote node while memory
2732 * may still exist in local DMA zone.
2734 static int node_order
[MAX_NUMNODES
];
2736 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2739 int zone_type
; /* needs to be signed */
2741 struct zonelist
*zonelist
;
2743 zonelist
= &pgdat
->node_zonelists
[0];
2745 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2746 for (j
= 0; j
< nr_nodes
; j
++) {
2747 node
= node_order
[j
];
2748 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2749 if (populated_zone(z
)) {
2751 &zonelist
->_zonerefs
[pos
++]);
2752 check_highest_zone(zone_type
);
2756 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2757 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2760 static int default_zonelist_order(void)
2763 unsigned long low_kmem_size
,total_size
;
2767 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2768 * If they are really small and used heavily, the system can fall
2769 * into OOM very easily.
2770 * This function detect ZONE_DMA/DMA32 size and configures zone order.
2772 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2775 for_each_online_node(nid
) {
2776 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2777 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2778 if (populated_zone(z
)) {
2779 if (zone_type
< ZONE_NORMAL
)
2780 low_kmem_size
+= z
->present_pages
;
2781 total_size
+= z
->present_pages
;
2782 } else if (zone_type
== ZONE_NORMAL
) {
2784 * If any node has only lowmem, then node order
2785 * is preferred to allow kernel allocations
2786 * locally; otherwise, they can easily infringe
2787 * on other nodes when there is an abundance of
2788 * lowmem available to allocate from.
2790 return ZONELIST_ORDER_NODE
;
2794 if (!low_kmem_size
|| /* there are no DMA area. */
2795 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2796 return ZONELIST_ORDER_NODE
;
2798 * look into each node's config.
2799 * If there is a node whose DMA/DMA32 memory is very big area on
2800 * local memory, NODE_ORDER may be suitable.
2802 average_size
= total_size
/
2803 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2804 for_each_online_node(nid
) {
2807 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2808 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2809 if (populated_zone(z
)) {
2810 if (zone_type
< ZONE_NORMAL
)
2811 low_kmem_size
+= z
->present_pages
;
2812 total_size
+= z
->present_pages
;
2815 if (low_kmem_size
&&
2816 total_size
> average_size
&& /* ignore small node */
2817 low_kmem_size
> total_size
* 70/100)
2818 return ZONELIST_ORDER_NODE
;
2820 return ZONELIST_ORDER_ZONE
;
2823 static void set_zonelist_order(void)
2825 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2826 current_zonelist_order
= default_zonelist_order();
2828 current_zonelist_order
= user_zonelist_order
;
2831 static void build_zonelists(pg_data_t
*pgdat
)
2835 nodemask_t used_mask
;
2836 int local_node
, prev_node
;
2837 struct zonelist
*zonelist
;
2838 int order
= current_zonelist_order
;
2840 /* initialize zonelists */
2841 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2842 zonelist
= pgdat
->node_zonelists
+ i
;
2843 zonelist
->_zonerefs
[0].zone
= NULL
;
2844 zonelist
->_zonerefs
[0].zone_idx
= 0;
2847 /* NUMA-aware ordering of nodes */
2848 local_node
= pgdat
->node_id
;
2849 load
= nr_online_nodes
;
2850 prev_node
= local_node
;
2851 nodes_clear(used_mask
);
2853 memset(node_order
, 0, sizeof(node_order
));
2856 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2857 int distance
= node_distance(local_node
, node
);
2860 * If another node is sufficiently far away then it is better
2861 * to reclaim pages in a zone before going off node.
2863 if (distance
> RECLAIM_DISTANCE
)
2864 zone_reclaim_mode
= 1;
2867 * We don't want to pressure a particular node.
2868 * So adding penalty to the first node in same
2869 * distance group to make it round-robin.
2871 if (distance
!= node_distance(local_node
, prev_node
))
2872 node_load
[node
] = load
;
2876 if (order
== ZONELIST_ORDER_NODE
)
2877 build_zonelists_in_node_order(pgdat
, node
);
2879 node_order
[j
++] = node
; /* remember order */
2882 if (order
== ZONELIST_ORDER_ZONE
) {
2883 /* calculate node order -- i.e., DMA last! */
2884 build_zonelists_in_zone_order(pgdat
, j
);
2887 build_thisnode_zonelists(pgdat
);
2890 /* Construct the zonelist performance cache - see further mmzone.h */
2891 static void build_zonelist_cache(pg_data_t
*pgdat
)
2893 struct zonelist
*zonelist
;
2894 struct zonelist_cache
*zlc
;
2897 zonelist
= &pgdat
->node_zonelists
[0];
2898 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2899 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2900 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2901 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2904 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
2906 * Return node id of node used for "local" allocations.
2907 * I.e., first node id of first zone in arg node's generic zonelist.
2908 * Used for initializing percpu 'numa_mem', which is used primarily
2909 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
2911 int local_memory_node(int node
)
2915 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
2916 gfp_zone(GFP_KERNEL
),
2923 #else /* CONFIG_NUMA */
2925 static void set_zonelist_order(void)
2927 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2930 static void build_zonelists(pg_data_t
*pgdat
)
2932 int node
, local_node
;
2934 struct zonelist
*zonelist
;
2936 local_node
= pgdat
->node_id
;
2938 zonelist
= &pgdat
->node_zonelists
[0];
2939 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2942 * Now we build the zonelist so that it contains the zones
2943 * of all the other nodes.
2944 * We don't want to pressure a particular node, so when
2945 * building the zones for node N, we make sure that the
2946 * zones coming right after the local ones are those from
2947 * node N+1 (modulo N)
2949 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2950 if (!node_online(node
))
2952 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2955 for (node
= 0; node
< local_node
; node
++) {
2956 if (!node_online(node
))
2958 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2962 zonelist
->_zonerefs
[j
].zone
= NULL
;
2963 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2966 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2967 static void build_zonelist_cache(pg_data_t
*pgdat
)
2969 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2972 #endif /* CONFIG_NUMA */
2975 * Boot pageset table. One per cpu which is going to be used for all
2976 * zones and all nodes. The parameters will be set in such a way
2977 * that an item put on a list will immediately be handed over to
2978 * the buddy list. This is safe since pageset manipulation is done
2979 * with interrupts disabled.
2981 * The boot_pagesets must be kept even after bootup is complete for
2982 * unused processors and/or zones. They do play a role for bootstrapping
2983 * hotplugged processors.
2985 * zoneinfo_show() and maybe other functions do
2986 * not check if the processor is online before following the pageset pointer.
2987 * Other parts of the kernel may not check if the zone is available.
2989 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
2990 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
2991 static void setup_zone_pageset(struct zone
*zone
);
2994 * Global mutex to protect against size modification of zonelists
2995 * as well as to serialize pageset setup for the new populated zone.
2997 DEFINE_MUTEX(zonelists_mutex
);
2999 /* return values int ....just for stop_machine() */
3000 static __init_refok
int __build_all_zonelists(void *data
)
3006 memset(node_load
, 0, sizeof(node_load
));
3008 for_each_online_node(nid
) {
3009 pg_data_t
*pgdat
= NODE_DATA(nid
);
3011 build_zonelists(pgdat
);
3012 build_zonelist_cache(pgdat
);
3015 #ifdef CONFIG_MEMORY_HOTPLUG
3016 /* Setup real pagesets for the new zone */
3018 struct zone
*zone
= data
;
3019 setup_zone_pageset(zone
);
3024 * Initialize the boot_pagesets that are going to be used
3025 * for bootstrapping processors. The real pagesets for
3026 * each zone will be allocated later when the per cpu
3027 * allocator is available.
3029 * boot_pagesets are used also for bootstrapping offline
3030 * cpus if the system is already booted because the pagesets
3031 * are needed to initialize allocators on a specific cpu too.
3032 * F.e. the percpu allocator needs the page allocator which
3033 * needs the percpu allocator in order to allocate its pagesets
3034 * (a chicken-egg dilemma).
3036 for_each_possible_cpu(cpu
) {
3037 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3039 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3041 * We now know the "local memory node" for each node--
3042 * i.e., the node of the first zone in the generic zonelist.
3043 * Set up numa_mem percpu variable for on-line cpus. During
3044 * boot, only the boot cpu should be on-line; we'll init the
3045 * secondary cpus' numa_mem as they come on-line. During
3046 * node/memory hotplug, we'll fixup all on-line cpus.
3048 if (cpu_online(cpu
))
3049 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3057 * Called with zonelists_mutex held always
3058 * unless system_state == SYSTEM_BOOTING.
3060 void build_all_zonelists(void *data
)
3062 set_zonelist_order();
3064 if (system_state
== SYSTEM_BOOTING
) {
3065 __build_all_zonelists(NULL
);
3066 mminit_verify_zonelist();
3067 cpuset_init_current_mems_allowed();
3069 /* we have to stop all cpus to guarantee there is no user
3071 stop_machine(__build_all_zonelists
, data
, NULL
);
3072 /* cpuset refresh routine should be here */
3074 vm_total_pages
= nr_free_pagecache_pages();
3076 * Disable grouping by mobility if the number of pages in the
3077 * system is too low to allow the mechanism to work. It would be
3078 * more accurate, but expensive to check per-zone. This check is
3079 * made on memory-hotadd so a system can start with mobility
3080 * disabled and enable it later
3082 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3083 page_group_by_mobility_disabled
= 1;
3085 page_group_by_mobility_disabled
= 0;
3087 printk("Built %i zonelists in %s order, mobility grouping %s. "
3088 "Total pages: %ld\n",
3090 zonelist_order_name
[current_zonelist_order
],
3091 page_group_by_mobility_disabled
? "off" : "on",
3094 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3099 * Helper functions to size the waitqueue hash table.
3100 * Essentially these want to choose hash table sizes sufficiently
3101 * large so that collisions trying to wait on pages are rare.
3102 * But in fact, the number of active page waitqueues on typical
3103 * systems is ridiculously low, less than 200. So this is even
3104 * conservative, even though it seems large.
3106 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3107 * waitqueues, i.e. the size of the waitq table given the number of pages.
3109 #define PAGES_PER_WAITQUEUE 256
3111 #ifndef CONFIG_MEMORY_HOTPLUG
3112 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3114 unsigned long size
= 1;
3116 pages
/= PAGES_PER_WAITQUEUE
;
3118 while (size
< pages
)
3122 * Once we have dozens or even hundreds of threads sleeping
3123 * on IO we've got bigger problems than wait queue collision.
3124 * Limit the size of the wait table to a reasonable size.
3126 size
= min(size
, 4096UL);
3128 return max(size
, 4UL);
3132 * A zone's size might be changed by hot-add, so it is not possible to determine
3133 * a suitable size for its wait_table. So we use the maximum size now.
3135 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3137 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3138 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3139 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3141 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3142 * or more by the traditional way. (See above). It equals:
3144 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3145 * ia64(16K page size) : = ( 8G + 4M)byte.
3146 * powerpc (64K page size) : = (32G +16M)byte.
3148 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3155 * This is an integer logarithm so that shifts can be used later
3156 * to extract the more random high bits from the multiplicative
3157 * hash function before the remainder is taken.
3159 static inline unsigned long wait_table_bits(unsigned long size
)
3164 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3167 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3168 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3169 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3170 * higher will lead to a bigger reserve which will get freed as contiguous
3171 * blocks as reclaim kicks in
3173 static void setup_zone_migrate_reserve(struct zone
*zone
)
3175 unsigned long start_pfn
, pfn
, end_pfn
;
3177 unsigned long block_migratetype
;
3180 /* Get the start pfn, end pfn and the number of blocks to reserve */
3181 start_pfn
= zone
->zone_start_pfn
;
3182 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3183 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3187 * Reserve blocks are generally in place to help high-order atomic
3188 * allocations that are short-lived. A min_free_kbytes value that
3189 * would result in more than 2 reserve blocks for atomic allocations
3190 * is assumed to be in place to help anti-fragmentation for the
3191 * future allocation of hugepages at runtime.
3193 reserve
= min(2, reserve
);
3195 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3196 if (!pfn_valid(pfn
))
3198 page
= pfn_to_page(pfn
);
3200 /* Watch out for overlapping nodes */
3201 if (page_to_nid(page
) != zone_to_nid(zone
))
3204 /* Blocks with reserved pages will never free, skip them. */
3205 if (PageReserved(page
))
3208 block_migratetype
= get_pageblock_migratetype(page
);
3210 /* If this block is reserved, account for it */
3211 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
3216 /* Suitable for reserving if this block is movable */
3217 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
3218 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
3219 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
3225 * If the reserve is met and this is a previous reserved block,
3228 if (block_migratetype
== MIGRATE_RESERVE
) {
3229 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3230 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3236 * Initially all pages are reserved - free ones are freed
3237 * up by free_all_bootmem() once the early boot process is
3238 * done. Non-atomic initialization, single-pass.
3240 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3241 unsigned long start_pfn
, enum memmap_context context
)
3244 unsigned long end_pfn
= start_pfn
+ size
;
3248 if (highest_memmap_pfn
< end_pfn
- 1)
3249 highest_memmap_pfn
= end_pfn
- 1;
3251 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3252 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3254 * There can be holes in boot-time mem_map[]s
3255 * handed to this function. They do not
3256 * exist on hotplugged memory.
3258 if (context
== MEMMAP_EARLY
) {
3259 if (!early_pfn_valid(pfn
))
3261 if (!early_pfn_in_nid(pfn
, nid
))
3264 page
= pfn_to_page(pfn
);
3265 set_page_links(page
, zone
, nid
, pfn
);
3266 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3267 init_page_count(page
);
3268 reset_page_mapcount(page
);
3269 SetPageReserved(page
);
3271 * Mark the block movable so that blocks are reserved for
3272 * movable at startup. This will force kernel allocations
3273 * to reserve their blocks rather than leaking throughout
3274 * the address space during boot when many long-lived
3275 * kernel allocations are made. Later some blocks near
3276 * the start are marked MIGRATE_RESERVE by
3277 * setup_zone_migrate_reserve()
3279 * bitmap is created for zone's valid pfn range. but memmap
3280 * can be created for invalid pages (for alignment)
3281 * check here not to call set_pageblock_migratetype() against
3284 if ((z
->zone_start_pfn
<= pfn
)
3285 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3286 && !(pfn
& (pageblock_nr_pages
- 1)))
3287 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3289 INIT_LIST_HEAD(&page
->lru
);
3290 #ifdef WANT_PAGE_VIRTUAL
3291 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3292 if (!is_highmem_idx(zone
))
3293 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3298 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3301 for_each_migratetype_order(order
, t
) {
3302 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3303 zone
->free_area
[order
].nr_free
= 0;
3307 #ifndef __HAVE_ARCH_MEMMAP_INIT
3308 #define memmap_init(size, nid, zone, start_pfn) \
3309 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3312 static int zone_batchsize(struct zone
*zone
)
3318 * The per-cpu-pages pools are set to around 1000th of the
3319 * size of the zone. But no more than 1/2 of a meg.
3321 * OK, so we don't know how big the cache is. So guess.
3323 batch
= zone
->present_pages
/ 1024;
3324 if (batch
* PAGE_SIZE
> 512 * 1024)
3325 batch
= (512 * 1024) / PAGE_SIZE
;
3326 batch
/= 4; /* We effectively *= 4 below */
3331 * Clamp the batch to a 2^n - 1 value. Having a power
3332 * of 2 value was found to be more likely to have
3333 * suboptimal cache aliasing properties in some cases.
3335 * For example if 2 tasks are alternately allocating
3336 * batches of pages, one task can end up with a lot
3337 * of pages of one half of the possible page colors
3338 * and the other with pages of the other colors.
3340 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3345 /* The deferral and batching of frees should be suppressed under NOMMU
3348 * The problem is that NOMMU needs to be able to allocate large chunks
3349 * of contiguous memory as there's no hardware page translation to
3350 * assemble apparent contiguous memory from discontiguous pages.
3352 * Queueing large contiguous runs of pages for batching, however,
3353 * causes the pages to actually be freed in smaller chunks. As there
3354 * can be a significant delay between the individual batches being
3355 * recycled, this leads to the once large chunks of space being
3356 * fragmented and becoming unavailable for high-order allocations.
3362 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3364 struct per_cpu_pages
*pcp
;
3367 memset(p
, 0, sizeof(*p
));
3371 pcp
->high
= 6 * batch
;
3372 pcp
->batch
= max(1UL, 1 * batch
);
3373 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3374 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3378 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3379 * to the value high for the pageset p.
3382 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3385 struct per_cpu_pages
*pcp
;
3389 pcp
->batch
= max(1UL, high
/4);
3390 if ((high
/4) > (PAGE_SHIFT
* 8))
3391 pcp
->batch
= PAGE_SHIFT
* 8;
3394 static __meminit
void setup_zone_pageset(struct zone
*zone
)
3398 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
3400 for_each_possible_cpu(cpu
) {
3401 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
3403 setup_pageset(pcp
, zone_batchsize(zone
));
3405 if (percpu_pagelist_fraction
)
3406 setup_pagelist_highmark(pcp
,
3407 (zone
->present_pages
/
3408 percpu_pagelist_fraction
));
3413 * Allocate per cpu pagesets and initialize them.
3414 * Before this call only boot pagesets were available.
3416 void __init
setup_per_cpu_pageset(void)
3420 for_each_populated_zone(zone
)
3421 setup_zone_pageset(zone
);
3424 static noinline __init_refok
3425 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3428 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3432 * The per-page waitqueue mechanism uses hashed waitqueues
3435 zone
->wait_table_hash_nr_entries
=
3436 wait_table_hash_nr_entries(zone_size_pages
);
3437 zone
->wait_table_bits
=
3438 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3439 alloc_size
= zone
->wait_table_hash_nr_entries
3440 * sizeof(wait_queue_head_t
);
3442 if (!slab_is_available()) {
3443 zone
->wait_table
= (wait_queue_head_t
*)
3444 alloc_bootmem_node(pgdat
, alloc_size
);
3447 * This case means that a zone whose size was 0 gets new memory
3448 * via memory hot-add.
3449 * But it may be the case that a new node was hot-added. In
3450 * this case vmalloc() will not be able to use this new node's
3451 * memory - this wait_table must be initialized to use this new
3452 * node itself as well.
3453 * To use this new node's memory, further consideration will be
3456 zone
->wait_table
= vmalloc(alloc_size
);
3458 if (!zone
->wait_table
)
3461 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3462 init_waitqueue_head(zone
->wait_table
+ i
);
3467 static int __zone_pcp_update(void *data
)
3469 struct zone
*zone
= data
;
3471 unsigned long batch
= zone_batchsize(zone
), flags
;
3473 for_each_possible_cpu(cpu
) {
3474 struct per_cpu_pageset
*pset
;
3475 struct per_cpu_pages
*pcp
;
3477 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
3480 local_irq_save(flags
);
3481 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
3482 setup_pageset(pset
, batch
);
3483 local_irq_restore(flags
);
3488 void zone_pcp_update(struct zone
*zone
)
3490 stop_machine(__zone_pcp_update
, zone
, NULL
);
3493 static __meminit
void zone_pcp_init(struct zone
*zone
)
3496 * per cpu subsystem is not up at this point. The following code
3497 * relies on the ability of the linker to provide the
3498 * offset of a (static) per cpu variable into the per cpu area.
3500 zone
->pageset
= &boot_pageset
;
3502 if (zone
->present_pages
)
3503 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
3504 zone
->name
, zone
->present_pages
,
3505 zone_batchsize(zone
));
3508 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3509 unsigned long zone_start_pfn
,
3511 enum memmap_context context
)
3513 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3515 ret
= zone_wait_table_init(zone
, size
);
3518 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3520 zone
->zone_start_pfn
= zone_start_pfn
;
3522 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3523 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3525 (unsigned long)zone_idx(zone
),
3526 zone_start_pfn
, (zone_start_pfn
+ size
));
3528 zone_init_free_lists(zone
);
3533 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3535 * Basic iterator support. Return the first range of PFNs for a node
3536 * Note: nid == MAX_NUMNODES returns first region regardless of node
3538 static int __meminit
first_active_region_index_in_nid(int nid
)
3542 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3543 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3550 * Basic iterator support. Return the next active range of PFNs for a node
3551 * Note: nid == MAX_NUMNODES returns next region regardless of node
3553 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3555 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3556 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3562 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3564 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3565 * Architectures may implement their own version but if add_active_range()
3566 * was used and there are no special requirements, this is a convenient
3569 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3573 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3574 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3575 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3577 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3578 return early_node_map
[i
].nid
;
3580 /* This is a memory hole */
3583 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3585 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3589 nid
= __early_pfn_to_nid(pfn
);
3592 /* just returns 0 */
3596 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3597 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3601 nid
= __early_pfn_to_nid(pfn
);
3602 if (nid
>= 0 && nid
!= node
)
3608 /* Basic iterator support to walk early_node_map[] */
3609 #define for_each_active_range_index_in_nid(i, nid) \
3610 for (i = first_active_region_index_in_nid(nid); i != -1; \
3611 i = next_active_region_index_in_nid(i, nid))
3614 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3615 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3616 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3618 * If an architecture guarantees that all ranges registered with
3619 * add_active_ranges() contain no holes and may be freed, this
3620 * this function may be used instead of calling free_bootmem() manually.
3622 void __init
free_bootmem_with_active_regions(int nid
,
3623 unsigned long max_low_pfn
)
3627 for_each_active_range_index_in_nid(i
, nid
) {
3628 unsigned long size_pages
= 0;
3629 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3631 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3634 if (end_pfn
> max_low_pfn
)
3635 end_pfn
= max_low_pfn
;
3637 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3638 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3639 PFN_PHYS(early_node_map
[i
].start_pfn
),
3640 size_pages
<< PAGE_SHIFT
);
3644 int __init
add_from_early_node_map(struct range
*range
, int az
,
3645 int nr_range
, int nid
)
3650 /* need to go over early_node_map to find out good range for node */
3651 for_each_active_range_index_in_nid(i
, nid
) {
3652 start
= early_node_map
[i
].start_pfn
;
3653 end
= early_node_map
[i
].end_pfn
;
3654 nr_range
= add_range(range
, az
, nr_range
, start
, end
);
3659 #ifdef CONFIG_NO_BOOTMEM
3660 void * __init
__alloc_memory_core_early(int nid
, u64 size
, u64 align
,
3661 u64 goal
, u64 limit
)
3666 if (limit
> get_max_mapped())
3667 limit
= get_max_mapped();
3669 /* need to go over early_node_map to find out good range for node */
3670 for_each_active_range_index_in_nid(i
, nid
) {
3672 u64 ei_start
, ei_last
;
3674 ei_last
= early_node_map
[i
].end_pfn
;
3675 ei_last
<<= PAGE_SHIFT
;
3676 ei_start
= early_node_map
[i
].start_pfn
;
3677 ei_start
<<= PAGE_SHIFT
;
3678 addr
= find_early_area(ei_start
, ei_last
,
3679 goal
, limit
, size
, align
);
3685 printk(KERN_DEBUG
"alloc (nid=%d %llx - %llx) (%llx - %llx) %llx %llx => %llx\n",
3687 ei_start
, ei_last
, goal
, limit
, size
,
3691 ptr
= phys_to_virt(addr
);
3692 memset(ptr
, 0, size
);
3693 reserve_early_without_check(addr
, addr
+ size
, "BOOTMEM");
3695 * The min_count is set to 0 so that bootmem allocated blocks
3696 * are never reported as leaks.
3698 kmemleak_alloc(ptr
, size
, 0, 0);
3707 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3712 for_each_active_range_index_in_nid(i
, nid
) {
3713 ret
= work_fn(early_node_map
[i
].start_pfn
,
3714 early_node_map
[i
].end_pfn
, data
);
3720 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3721 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3723 * If an architecture guarantees that all ranges registered with
3724 * add_active_ranges() contain no holes and may be freed, this
3725 * function may be used instead of calling memory_present() manually.
3727 void __init
sparse_memory_present_with_active_regions(int nid
)
3731 for_each_active_range_index_in_nid(i
, nid
)
3732 memory_present(early_node_map
[i
].nid
,
3733 early_node_map
[i
].start_pfn
,
3734 early_node_map
[i
].end_pfn
);
3738 * get_pfn_range_for_nid - Return the start and end page frames for a node
3739 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3740 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3741 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3743 * It returns the start and end page frame of a node based on information
3744 * provided by an arch calling add_active_range(). If called for a node
3745 * with no available memory, a warning is printed and the start and end
3748 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3749 unsigned long *start_pfn
, unsigned long *end_pfn
)
3755 for_each_active_range_index_in_nid(i
, nid
) {
3756 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3757 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3760 if (*start_pfn
== -1UL)
3765 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3766 * assumption is made that zones within a node are ordered in monotonic
3767 * increasing memory addresses so that the "highest" populated zone is used
3769 static void __init
find_usable_zone_for_movable(void)
3772 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3773 if (zone_index
== ZONE_MOVABLE
)
3776 if (arch_zone_highest_possible_pfn
[zone_index
] >
3777 arch_zone_lowest_possible_pfn
[zone_index
])
3781 VM_BUG_ON(zone_index
== -1);
3782 movable_zone
= zone_index
;
3786 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3787 * because it is sized independant of architecture. Unlike the other zones,
3788 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3789 * in each node depending on the size of each node and how evenly kernelcore
3790 * is distributed. This helper function adjusts the zone ranges
3791 * provided by the architecture for a given node by using the end of the
3792 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3793 * zones within a node are in order of monotonic increases memory addresses
3795 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3796 unsigned long zone_type
,
3797 unsigned long node_start_pfn
,
3798 unsigned long node_end_pfn
,
3799 unsigned long *zone_start_pfn
,
3800 unsigned long *zone_end_pfn
)
3802 /* Only adjust if ZONE_MOVABLE is on this node */
3803 if (zone_movable_pfn
[nid
]) {
3804 /* Size ZONE_MOVABLE */
3805 if (zone_type
== ZONE_MOVABLE
) {
3806 *zone_start_pfn
= zone_movable_pfn
[nid
];
3807 *zone_end_pfn
= min(node_end_pfn
,
3808 arch_zone_highest_possible_pfn
[movable_zone
]);
3810 /* Adjust for ZONE_MOVABLE starting within this range */
3811 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3812 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3813 *zone_end_pfn
= zone_movable_pfn
[nid
];
3815 /* Check if this whole range is within ZONE_MOVABLE */
3816 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3817 *zone_start_pfn
= *zone_end_pfn
;
3822 * Return the number of pages a zone spans in a node, including holes
3823 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3825 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3826 unsigned long zone_type
,
3827 unsigned long *ignored
)
3829 unsigned long node_start_pfn
, node_end_pfn
;
3830 unsigned long zone_start_pfn
, zone_end_pfn
;
3832 /* Get the start and end of the node and zone */
3833 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3834 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3835 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3836 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3837 node_start_pfn
, node_end_pfn
,
3838 &zone_start_pfn
, &zone_end_pfn
);
3840 /* Check that this node has pages within the zone's required range */
3841 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3844 /* Move the zone boundaries inside the node if necessary */
3845 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3846 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3848 /* Return the spanned pages */
3849 return zone_end_pfn
- zone_start_pfn
;
3853 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3854 * then all holes in the requested range will be accounted for.
3856 unsigned long __meminit
__absent_pages_in_range(int nid
,
3857 unsigned long range_start_pfn
,
3858 unsigned long range_end_pfn
)
3861 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3862 unsigned long start_pfn
;
3864 /* Find the end_pfn of the first active range of pfns in the node */
3865 i
= first_active_region_index_in_nid(nid
);
3869 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3871 /* Account for ranges before physical memory on this node */
3872 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3873 hole_pages
= prev_end_pfn
- range_start_pfn
;
3875 /* Find all holes for the zone within the node */
3876 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3878 /* No need to continue if prev_end_pfn is outside the zone */
3879 if (prev_end_pfn
>= range_end_pfn
)
3882 /* Make sure the end of the zone is not within the hole */
3883 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3884 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3886 /* Update the hole size cound and move on */
3887 if (start_pfn
> range_start_pfn
) {
3888 BUG_ON(prev_end_pfn
> start_pfn
);
3889 hole_pages
+= start_pfn
- prev_end_pfn
;
3891 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3894 /* Account for ranges past physical memory on this node */
3895 if (range_end_pfn
> prev_end_pfn
)
3896 hole_pages
+= range_end_pfn
-
3897 max(range_start_pfn
, prev_end_pfn
);
3903 * absent_pages_in_range - Return number of page frames in holes within a range
3904 * @start_pfn: The start PFN to start searching for holes
3905 * @end_pfn: The end PFN to stop searching for holes
3907 * It returns the number of pages frames in memory holes within a range.
3909 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3910 unsigned long end_pfn
)
3912 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3915 /* Return the number of page frames in holes in a zone on a node */
3916 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3917 unsigned long zone_type
,
3918 unsigned long *ignored
)
3920 unsigned long node_start_pfn
, node_end_pfn
;
3921 unsigned long zone_start_pfn
, zone_end_pfn
;
3923 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3924 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3926 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3929 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3930 node_start_pfn
, node_end_pfn
,
3931 &zone_start_pfn
, &zone_end_pfn
);
3932 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3936 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3937 unsigned long zone_type
,
3938 unsigned long *zones_size
)
3940 return zones_size
[zone_type
];
3943 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3944 unsigned long zone_type
,
3945 unsigned long *zholes_size
)
3950 return zholes_size
[zone_type
];
3955 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3956 unsigned long *zones_size
, unsigned long *zholes_size
)
3958 unsigned long realtotalpages
, totalpages
= 0;
3961 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3962 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3964 pgdat
->node_spanned_pages
= totalpages
;
3966 realtotalpages
= totalpages
;
3967 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3969 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3971 pgdat
->node_present_pages
= realtotalpages
;
3972 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3976 #ifndef CONFIG_SPARSEMEM
3978 * Calculate the size of the zone->blockflags rounded to an unsigned long
3979 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3980 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3981 * round what is now in bits to nearest long in bits, then return it in
3984 static unsigned long __init
usemap_size(unsigned long zonesize
)
3986 unsigned long usemapsize
;
3988 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3989 usemapsize
= usemapsize
>> pageblock_order
;
3990 usemapsize
*= NR_PAGEBLOCK_BITS
;
3991 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3993 return usemapsize
/ 8;
3996 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3997 struct zone
*zone
, unsigned long zonesize
)
3999 unsigned long usemapsize
= usemap_size(zonesize
);
4000 zone
->pageblock_flags
= NULL
;
4002 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
4005 static void inline setup_usemap(struct pglist_data
*pgdat
,
4006 struct zone
*zone
, unsigned long zonesize
) {}
4007 #endif /* CONFIG_SPARSEMEM */
4009 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4011 /* Return a sensible default order for the pageblock size. */
4012 static inline int pageblock_default_order(void)
4014 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4015 return HUGETLB_PAGE_ORDER
;
4020 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4021 static inline void __init
set_pageblock_order(unsigned int order
)
4023 /* Check that pageblock_nr_pages has not already been setup */
4024 if (pageblock_order
)
4028 * Assume the largest contiguous order of interest is a huge page.
4029 * This value may be variable depending on boot parameters on IA64
4031 pageblock_order
= order
;
4033 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4036 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4037 * and pageblock_default_order() are unused as pageblock_order is set
4038 * at compile-time. See include/linux/pageblock-flags.h for the values of
4039 * pageblock_order based on the kernel config
4041 static inline int pageblock_default_order(unsigned int order
)
4045 #define set_pageblock_order(x) do {} while (0)
4047 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4050 * Set up the zone data structures:
4051 * - mark all pages reserved
4052 * - mark all memory queues empty
4053 * - clear the memory bitmaps
4055 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4056 unsigned long *zones_size
, unsigned long *zholes_size
)
4059 int nid
= pgdat
->node_id
;
4060 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4063 pgdat_resize_init(pgdat
);
4064 pgdat
->nr_zones
= 0;
4065 init_waitqueue_head(&pgdat
->kswapd_wait
);
4066 pgdat
->kswapd_max_order
= 0;
4067 pgdat_page_cgroup_init(pgdat
);
4069 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4070 struct zone
*zone
= pgdat
->node_zones
+ j
;
4071 unsigned long size
, realsize
, memmap_pages
;
4074 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4075 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
4079 * Adjust realsize so that it accounts for how much memory
4080 * is used by this zone for memmap. This affects the watermark
4081 * and per-cpu initialisations
4084 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
4085 if (realsize
>= memmap_pages
) {
4086 realsize
-= memmap_pages
;
4089 " %s zone: %lu pages used for memmap\n",
4090 zone_names
[j
], memmap_pages
);
4093 " %s zone: %lu pages exceeds realsize %lu\n",
4094 zone_names
[j
], memmap_pages
, realsize
);
4096 /* Account for reserved pages */
4097 if (j
== 0 && realsize
> dma_reserve
) {
4098 realsize
-= dma_reserve
;
4099 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4100 zone_names
[0], dma_reserve
);
4103 if (!is_highmem_idx(j
))
4104 nr_kernel_pages
+= realsize
;
4105 nr_all_pages
+= realsize
;
4107 zone
->spanned_pages
= size
;
4108 zone
->present_pages
= realsize
;
4111 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
4113 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
4115 zone
->name
= zone_names
[j
];
4116 spin_lock_init(&zone
->lock
);
4117 spin_lock_init(&zone
->lru_lock
);
4118 zone_seqlock_init(zone
);
4119 zone
->zone_pgdat
= pgdat
;
4121 zone_pcp_init(zone
);
4123 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
4124 zone
->reclaim_stat
.nr_saved_scan
[l
] = 0;
4126 zone
->reclaim_stat
.recent_rotated
[0] = 0;
4127 zone
->reclaim_stat
.recent_rotated
[1] = 0;
4128 zone
->reclaim_stat
.recent_scanned
[0] = 0;
4129 zone
->reclaim_stat
.recent_scanned
[1] = 0;
4130 zap_zone_vm_stats(zone
);
4135 set_pageblock_order(pageblock_default_order());
4136 setup_usemap(pgdat
, zone
, size
);
4137 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4138 size
, MEMMAP_EARLY
);
4140 memmap_init(size
, nid
, j
, zone_start_pfn
);
4141 zone_start_pfn
+= size
;
4145 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4147 /* Skip empty nodes */
4148 if (!pgdat
->node_spanned_pages
)
4151 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4152 /* ia64 gets its own node_mem_map, before this, without bootmem */
4153 if (!pgdat
->node_mem_map
) {
4154 unsigned long size
, start
, end
;
4158 * The zone's endpoints aren't required to be MAX_ORDER
4159 * aligned but the node_mem_map endpoints must be in order
4160 * for the buddy allocator to function correctly.
4162 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4163 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4164 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4165 size
= (end
- start
) * sizeof(struct page
);
4166 map
= alloc_remap(pgdat
->node_id
, size
);
4168 map
= alloc_bootmem_node(pgdat
, size
);
4169 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4171 #ifndef CONFIG_NEED_MULTIPLE_NODES
4173 * With no DISCONTIG, the global mem_map is just set as node 0's
4175 if (pgdat
== NODE_DATA(0)) {
4176 mem_map
= NODE_DATA(0)->node_mem_map
;
4177 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4178 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4179 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4180 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4183 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4186 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4187 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4189 pg_data_t
*pgdat
= NODE_DATA(nid
);
4191 pgdat
->node_id
= nid
;
4192 pgdat
->node_start_pfn
= node_start_pfn
;
4193 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4195 alloc_node_mem_map(pgdat
);
4196 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4197 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4198 nid
, (unsigned long)pgdat
,
4199 (unsigned long)pgdat
->node_mem_map
);
4202 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4205 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4207 #if MAX_NUMNODES > 1
4209 * Figure out the number of possible node ids.
4211 static void __init
setup_nr_node_ids(void)
4214 unsigned int highest
= 0;
4216 for_each_node_mask(node
, node_possible_map
)
4218 nr_node_ids
= highest
+ 1;
4221 static inline void setup_nr_node_ids(void)
4227 * add_active_range - Register a range of PFNs backed by physical memory
4228 * @nid: The node ID the range resides on
4229 * @start_pfn: The start PFN of the available physical memory
4230 * @end_pfn: The end PFN of the available physical memory
4232 * These ranges are stored in an early_node_map[] and later used by
4233 * free_area_init_nodes() to calculate zone sizes and holes. If the
4234 * range spans a memory hole, it is up to the architecture to ensure
4235 * the memory is not freed by the bootmem allocator. If possible
4236 * the range being registered will be merged with existing ranges.
4238 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
4239 unsigned long end_pfn
)
4243 mminit_dprintk(MMINIT_TRACE
, "memory_register",
4244 "Entering add_active_range(%d, %#lx, %#lx) "
4245 "%d entries of %d used\n",
4246 nid
, start_pfn
, end_pfn
,
4247 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
4249 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
4251 /* Merge with existing active regions if possible */
4252 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4253 if (early_node_map
[i
].nid
!= nid
)
4256 /* Skip if an existing region covers this new one */
4257 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
4258 end_pfn
<= early_node_map
[i
].end_pfn
)
4261 /* Merge forward if suitable */
4262 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
4263 end_pfn
> early_node_map
[i
].end_pfn
) {
4264 early_node_map
[i
].end_pfn
= end_pfn
;
4268 /* Merge backward if suitable */
4269 if (start_pfn
< early_node_map
[i
].start_pfn
&&
4270 end_pfn
>= early_node_map
[i
].start_pfn
) {
4271 early_node_map
[i
].start_pfn
= start_pfn
;
4276 /* Check that early_node_map is large enough */
4277 if (i
>= MAX_ACTIVE_REGIONS
) {
4278 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
4279 MAX_ACTIVE_REGIONS
);
4283 early_node_map
[i
].nid
= nid
;
4284 early_node_map
[i
].start_pfn
= start_pfn
;
4285 early_node_map
[i
].end_pfn
= end_pfn
;
4286 nr_nodemap_entries
= i
+ 1;
4290 * remove_active_range - Shrink an existing registered range of PFNs
4291 * @nid: The node id the range is on that should be shrunk
4292 * @start_pfn: The new PFN of the range
4293 * @end_pfn: The new PFN of the range
4295 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
4296 * The map is kept near the end physical page range that has already been
4297 * registered. This function allows an arch to shrink an existing registered
4300 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
4301 unsigned long end_pfn
)
4306 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
4307 nid
, start_pfn
, end_pfn
);
4309 /* Find the old active region end and shrink */
4310 for_each_active_range_index_in_nid(i
, nid
) {
4311 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4312 early_node_map
[i
].end_pfn
<= end_pfn
) {
4314 early_node_map
[i
].start_pfn
= 0;
4315 early_node_map
[i
].end_pfn
= 0;
4319 if (early_node_map
[i
].start_pfn
< start_pfn
&&
4320 early_node_map
[i
].end_pfn
> start_pfn
) {
4321 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
4322 early_node_map
[i
].end_pfn
= start_pfn
;
4323 if (temp_end_pfn
> end_pfn
)
4324 add_active_range(nid
, end_pfn
, temp_end_pfn
);
4327 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4328 early_node_map
[i
].end_pfn
> end_pfn
&&
4329 early_node_map
[i
].start_pfn
< end_pfn
) {
4330 early_node_map
[i
].start_pfn
= end_pfn
;
4338 /* remove the blank ones */
4339 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
4340 if (early_node_map
[i
].nid
!= nid
)
4342 if (early_node_map
[i
].end_pfn
)
4344 /* we found it, get rid of it */
4345 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
4346 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
4347 sizeof(early_node_map
[j
]));
4348 j
= nr_nodemap_entries
- 1;
4349 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
4350 nr_nodemap_entries
--;
4355 * remove_all_active_ranges - Remove all currently registered regions
4357 * During discovery, it may be found that a table like SRAT is invalid
4358 * and an alternative discovery method must be used. This function removes
4359 * all currently registered regions.
4361 void __init
remove_all_active_ranges(void)
4363 memset(early_node_map
, 0, sizeof(early_node_map
));
4364 nr_nodemap_entries
= 0;
4367 /* Compare two active node_active_regions */
4368 static int __init
cmp_node_active_region(const void *a
, const void *b
)
4370 struct node_active_region
*arange
= (struct node_active_region
*)a
;
4371 struct node_active_region
*brange
= (struct node_active_region
*)b
;
4373 /* Done this way to avoid overflows */
4374 if (arange
->start_pfn
> brange
->start_pfn
)
4376 if (arange
->start_pfn
< brange
->start_pfn
)
4382 /* sort the node_map by start_pfn */
4383 void __init
sort_node_map(void)
4385 sort(early_node_map
, (size_t)nr_nodemap_entries
,
4386 sizeof(struct node_active_region
),
4387 cmp_node_active_region
, NULL
);
4390 /* Find the lowest pfn for a node */
4391 static unsigned long __init
find_min_pfn_for_node(int nid
)
4394 unsigned long min_pfn
= ULONG_MAX
;
4396 /* Assuming a sorted map, the first range found has the starting pfn */
4397 for_each_active_range_index_in_nid(i
, nid
)
4398 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
4400 if (min_pfn
== ULONG_MAX
) {
4402 "Could not find start_pfn for node %d\n", nid
);
4410 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4412 * It returns the minimum PFN based on information provided via
4413 * add_active_range().
4415 unsigned long __init
find_min_pfn_with_active_regions(void)
4417 return find_min_pfn_for_node(MAX_NUMNODES
);
4421 * early_calculate_totalpages()
4422 * Sum pages in active regions for movable zone.
4423 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4425 static unsigned long __init
early_calculate_totalpages(void)
4428 unsigned long totalpages
= 0;
4430 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4431 unsigned long pages
= early_node_map
[i
].end_pfn
-
4432 early_node_map
[i
].start_pfn
;
4433 totalpages
+= pages
;
4435 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
4441 * Find the PFN the Movable zone begins in each node. Kernel memory
4442 * is spread evenly between nodes as long as the nodes have enough
4443 * memory. When they don't, some nodes will have more kernelcore than
4446 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
4449 unsigned long usable_startpfn
;
4450 unsigned long kernelcore_node
, kernelcore_remaining
;
4451 /* save the state before borrow the nodemask */
4452 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4453 unsigned long totalpages
= early_calculate_totalpages();
4454 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4457 * If movablecore was specified, calculate what size of
4458 * kernelcore that corresponds so that memory usable for
4459 * any allocation type is evenly spread. If both kernelcore
4460 * and movablecore are specified, then the value of kernelcore
4461 * will be used for required_kernelcore if it's greater than
4462 * what movablecore would have allowed.
4464 if (required_movablecore
) {
4465 unsigned long corepages
;
4468 * Round-up so that ZONE_MOVABLE is at least as large as what
4469 * was requested by the user
4471 required_movablecore
=
4472 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4473 corepages
= totalpages
- required_movablecore
;
4475 required_kernelcore
= max(required_kernelcore
, corepages
);
4478 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4479 if (!required_kernelcore
)
4482 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4483 find_usable_zone_for_movable();
4484 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4487 /* Spread kernelcore memory as evenly as possible throughout nodes */
4488 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4489 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4491 * Recalculate kernelcore_node if the division per node
4492 * now exceeds what is necessary to satisfy the requested
4493 * amount of memory for the kernel
4495 if (required_kernelcore
< kernelcore_node
)
4496 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4499 * As the map is walked, we track how much memory is usable
4500 * by the kernel using kernelcore_remaining. When it is
4501 * 0, the rest of the node is usable by ZONE_MOVABLE
4503 kernelcore_remaining
= kernelcore_node
;
4505 /* Go through each range of PFNs within this node */
4506 for_each_active_range_index_in_nid(i
, nid
) {
4507 unsigned long start_pfn
, end_pfn
;
4508 unsigned long size_pages
;
4510 start_pfn
= max(early_node_map
[i
].start_pfn
,
4511 zone_movable_pfn
[nid
]);
4512 end_pfn
= early_node_map
[i
].end_pfn
;
4513 if (start_pfn
>= end_pfn
)
4516 /* Account for what is only usable for kernelcore */
4517 if (start_pfn
< usable_startpfn
) {
4518 unsigned long kernel_pages
;
4519 kernel_pages
= min(end_pfn
, usable_startpfn
)
4522 kernelcore_remaining
-= min(kernel_pages
,
4523 kernelcore_remaining
);
4524 required_kernelcore
-= min(kernel_pages
,
4525 required_kernelcore
);
4527 /* Continue if range is now fully accounted */
4528 if (end_pfn
<= usable_startpfn
) {
4531 * Push zone_movable_pfn to the end so
4532 * that if we have to rebalance
4533 * kernelcore across nodes, we will
4534 * not double account here
4536 zone_movable_pfn
[nid
] = end_pfn
;
4539 start_pfn
= usable_startpfn
;
4543 * The usable PFN range for ZONE_MOVABLE is from
4544 * start_pfn->end_pfn. Calculate size_pages as the
4545 * number of pages used as kernelcore
4547 size_pages
= end_pfn
- start_pfn
;
4548 if (size_pages
> kernelcore_remaining
)
4549 size_pages
= kernelcore_remaining
;
4550 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4553 * Some kernelcore has been met, update counts and
4554 * break if the kernelcore for this node has been
4557 required_kernelcore
-= min(required_kernelcore
,
4559 kernelcore_remaining
-= size_pages
;
4560 if (!kernelcore_remaining
)
4566 * If there is still required_kernelcore, we do another pass with one
4567 * less node in the count. This will push zone_movable_pfn[nid] further
4568 * along on the nodes that still have memory until kernelcore is
4572 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4575 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4576 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4577 zone_movable_pfn
[nid
] =
4578 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4581 /* restore the node_state */
4582 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4585 /* Any regular memory on that node ? */
4586 static void check_for_regular_memory(pg_data_t
*pgdat
)
4588 #ifdef CONFIG_HIGHMEM
4589 enum zone_type zone_type
;
4591 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4592 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4593 if (zone
->present_pages
)
4594 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4600 * free_area_init_nodes - Initialise all pg_data_t and zone data
4601 * @max_zone_pfn: an array of max PFNs for each zone
4603 * This will call free_area_init_node() for each active node in the system.
4604 * Using the page ranges provided by add_active_range(), the size of each
4605 * zone in each node and their holes is calculated. If the maximum PFN
4606 * between two adjacent zones match, it is assumed that the zone is empty.
4607 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4608 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4609 * starts where the previous one ended. For example, ZONE_DMA32 starts
4610 * at arch_max_dma_pfn.
4612 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4617 /* Sort early_node_map as initialisation assumes it is sorted */
4620 /* Record where the zone boundaries are */
4621 memset(arch_zone_lowest_possible_pfn
, 0,
4622 sizeof(arch_zone_lowest_possible_pfn
));
4623 memset(arch_zone_highest_possible_pfn
, 0,
4624 sizeof(arch_zone_highest_possible_pfn
));
4625 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4626 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4627 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4628 if (i
== ZONE_MOVABLE
)
4630 arch_zone_lowest_possible_pfn
[i
] =
4631 arch_zone_highest_possible_pfn
[i
-1];
4632 arch_zone_highest_possible_pfn
[i
] =
4633 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4635 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4636 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4638 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4639 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4640 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4642 /* Print out the zone ranges */
4643 printk("Zone PFN ranges:\n");
4644 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4645 if (i
== ZONE_MOVABLE
)
4647 printk(" %-8s ", zone_names
[i
]);
4648 if (arch_zone_lowest_possible_pfn
[i
] ==
4649 arch_zone_highest_possible_pfn
[i
])
4652 printk("%0#10lx -> %0#10lx\n",
4653 arch_zone_lowest_possible_pfn
[i
],
4654 arch_zone_highest_possible_pfn
[i
]);
4657 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4658 printk("Movable zone start PFN for each node\n");
4659 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4660 if (zone_movable_pfn
[i
])
4661 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4664 /* Print out the early_node_map[] */
4665 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4666 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4667 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4668 early_node_map
[i
].start_pfn
,
4669 early_node_map
[i
].end_pfn
);
4671 /* Initialise every node */
4672 mminit_verify_pageflags_layout();
4673 setup_nr_node_ids();
4674 for_each_online_node(nid
) {
4675 pg_data_t
*pgdat
= NODE_DATA(nid
);
4676 free_area_init_node(nid
, NULL
,
4677 find_min_pfn_for_node(nid
), NULL
);
4679 /* Any memory on that node */
4680 if (pgdat
->node_present_pages
)
4681 node_set_state(nid
, N_HIGH_MEMORY
);
4682 check_for_regular_memory(pgdat
);
4686 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4688 unsigned long long coremem
;
4692 coremem
= memparse(p
, &p
);
4693 *core
= coremem
>> PAGE_SHIFT
;
4695 /* Paranoid check that UL is enough for the coremem value */
4696 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4702 * kernelcore=size sets the amount of memory for use for allocations that
4703 * cannot be reclaimed or migrated.
4705 static int __init
cmdline_parse_kernelcore(char *p
)
4707 return cmdline_parse_core(p
, &required_kernelcore
);
4711 * movablecore=size sets the amount of memory for use for allocations that
4712 * can be reclaimed or migrated.
4714 static int __init
cmdline_parse_movablecore(char *p
)
4716 return cmdline_parse_core(p
, &required_movablecore
);
4719 early_param("kernelcore", cmdline_parse_kernelcore
);
4720 early_param("movablecore", cmdline_parse_movablecore
);
4722 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4725 * set_dma_reserve - set the specified number of pages reserved in the first zone
4726 * @new_dma_reserve: The number of pages to mark reserved
4728 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4729 * In the DMA zone, a significant percentage may be consumed by kernel image
4730 * and other unfreeable allocations which can skew the watermarks badly. This
4731 * function may optionally be used to account for unfreeable pages in the
4732 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4733 * smaller per-cpu batchsize.
4735 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4737 dma_reserve
= new_dma_reserve
;
4740 #ifndef CONFIG_NEED_MULTIPLE_NODES
4741 struct pglist_data __refdata contig_page_data
= {
4742 #ifndef CONFIG_NO_BOOTMEM
4743 .bdata
= &bootmem_node_data
[0]
4746 EXPORT_SYMBOL(contig_page_data
);
4749 void __init
free_area_init(unsigned long *zones_size
)
4751 free_area_init_node(0, zones_size
,
4752 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4755 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4756 unsigned long action
, void *hcpu
)
4758 int cpu
= (unsigned long)hcpu
;
4760 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4764 * Spill the event counters of the dead processor
4765 * into the current processors event counters.
4766 * This artificially elevates the count of the current
4769 vm_events_fold_cpu(cpu
);
4772 * Zero the differential counters of the dead processor
4773 * so that the vm statistics are consistent.
4775 * This is only okay since the processor is dead and cannot
4776 * race with what we are doing.
4778 refresh_cpu_vm_stats(cpu
);
4783 void __init
page_alloc_init(void)
4785 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4789 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4790 * or min_free_kbytes changes.
4792 static void calculate_totalreserve_pages(void)
4794 struct pglist_data
*pgdat
;
4795 unsigned long reserve_pages
= 0;
4796 enum zone_type i
, j
;
4798 for_each_online_pgdat(pgdat
) {
4799 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4800 struct zone
*zone
= pgdat
->node_zones
+ i
;
4801 unsigned long max
= 0;
4803 /* Find valid and maximum lowmem_reserve in the zone */
4804 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4805 if (zone
->lowmem_reserve
[j
] > max
)
4806 max
= zone
->lowmem_reserve
[j
];
4809 /* we treat the high watermark as reserved pages. */
4810 max
+= high_wmark_pages(zone
);
4812 if (max
> zone
->present_pages
)
4813 max
= zone
->present_pages
;
4814 reserve_pages
+= max
;
4817 totalreserve_pages
= reserve_pages
;
4821 * setup_per_zone_lowmem_reserve - called whenever
4822 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4823 * has a correct pages reserved value, so an adequate number of
4824 * pages are left in the zone after a successful __alloc_pages().
4826 static void setup_per_zone_lowmem_reserve(void)
4828 struct pglist_data
*pgdat
;
4829 enum zone_type j
, idx
;
4831 for_each_online_pgdat(pgdat
) {
4832 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4833 struct zone
*zone
= pgdat
->node_zones
+ j
;
4834 unsigned long present_pages
= zone
->present_pages
;
4836 zone
->lowmem_reserve
[j
] = 0;
4840 struct zone
*lower_zone
;
4844 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4845 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4847 lower_zone
= pgdat
->node_zones
+ idx
;
4848 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4849 sysctl_lowmem_reserve_ratio
[idx
];
4850 present_pages
+= lower_zone
->present_pages
;
4855 /* update totalreserve_pages */
4856 calculate_totalreserve_pages();
4860 * setup_per_zone_wmarks - called when min_free_kbytes changes
4861 * or when memory is hot-{added|removed}
4863 * Ensures that the watermark[min,low,high] values for each zone are set
4864 * correctly with respect to min_free_kbytes.
4866 void setup_per_zone_wmarks(void)
4868 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4869 unsigned long lowmem_pages
= 0;
4871 unsigned long flags
;
4873 /* Calculate total number of !ZONE_HIGHMEM pages */
4874 for_each_zone(zone
) {
4875 if (!is_highmem(zone
))
4876 lowmem_pages
+= zone
->present_pages
;
4879 for_each_zone(zone
) {
4882 spin_lock_irqsave(&zone
->lock
, flags
);
4883 tmp
= (u64
)pages_min
* zone
->present_pages
;
4884 do_div(tmp
, lowmem_pages
);
4885 if (is_highmem(zone
)) {
4887 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4888 * need highmem pages, so cap pages_min to a small
4891 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4892 * deltas controls asynch page reclaim, and so should
4893 * not be capped for highmem.
4897 min_pages
= zone
->present_pages
/ 1024;
4898 if (min_pages
< SWAP_CLUSTER_MAX
)
4899 min_pages
= SWAP_CLUSTER_MAX
;
4900 if (min_pages
> 128)
4902 zone
->watermark
[WMARK_MIN
] = min_pages
;
4905 * If it's a lowmem zone, reserve a number of pages
4906 * proportionate to the zone's size.
4908 zone
->watermark
[WMARK_MIN
] = tmp
;
4911 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
4912 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
4913 setup_zone_migrate_reserve(zone
);
4914 spin_unlock_irqrestore(&zone
->lock
, flags
);
4917 /* update totalreserve_pages */
4918 calculate_totalreserve_pages();
4922 * The inactive anon list should be small enough that the VM never has to
4923 * do too much work, but large enough that each inactive page has a chance
4924 * to be referenced again before it is swapped out.
4926 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4927 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4928 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4929 * the anonymous pages are kept on the inactive list.
4932 * memory ratio inactive anon
4933 * -------------------------------------
4942 void calculate_zone_inactive_ratio(struct zone
*zone
)
4944 unsigned int gb
, ratio
;
4946 /* Zone size in gigabytes */
4947 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4949 ratio
= int_sqrt(10 * gb
);
4953 zone
->inactive_ratio
= ratio
;
4956 static void __init
setup_per_zone_inactive_ratio(void)
4961 calculate_zone_inactive_ratio(zone
);
4965 * Initialise min_free_kbytes.
4967 * For small machines we want it small (128k min). For large machines
4968 * we want it large (64MB max). But it is not linear, because network
4969 * bandwidth does not increase linearly with machine size. We use
4971 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4972 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4988 static int __init
init_per_zone_wmark_min(void)
4990 unsigned long lowmem_kbytes
;
4992 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4994 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4995 if (min_free_kbytes
< 128)
4996 min_free_kbytes
= 128;
4997 if (min_free_kbytes
> 65536)
4998 min_free_kbytes
= 65536;
4999 setup_per_zone_wmarks();
5000 setup_per_zone_lowmem_reserve();
5001 setup_per_zone_inactive_ratio();
5004 module_init(init_per_zone_wmark_min
)
5007 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5008 * that we can call two helper functions whenever min_free_kbytes
5011 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5012 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5014 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5016 setup_per_zone_wmarks();
5021 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5022 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5027 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5032 zone
->min_unmapped_pages
= (zone
->present_pages
*
5033 sysctl_min_unmapped_ratio
) / 100;
5037 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5038 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5043 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5048 zone
->min_slab_pages
= (zone
->present_pages
*
5049 sysctl_min_slab_ratio
) / 100;
5055 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5056 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5057 * whenever sysctl_lowmem_reserve_ratio changes.
5059 * The reserve ratio obviously has absolutely no relation with the
5060 * minimum watermarks. The lowmem reserve ratio can only make sense
5061 * if in function of the boot time zone sizes.
5063 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5064 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5066 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5067 setup_per_zone_lowmem_reserve();
5072 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5073 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5074 * can have before it gets flushed back to buddy allocator.
5077 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5078 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5084 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5085 if (!write
|| (ret
== -EINVAL
))
5087 for_each_populated_zone(zone
) {
5088 for_each_possible_cpu(cpu
) {
5090 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5091 setup_pagelist_highmark(
5092 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5098 int hashdist
= HASHDIST_DEFAULT
;
5101 static int __init
set_hashdist(char *str
)
5105 hashdist
= simple_strtoul(str
, &str
, 0);
5108 __setup("hashdist=", set_hashdist
);
5112 * allocate a large system hash table from bootmem
5113 * - it is assumed that the hash table must contain an exact power-of-2
5114 * quantity of entries
5115 * - limit is the number of hash buckets, not the total allocation size
5117 void *__init
alloc_large_system_hash(const char *tablename
,
5118 unsigned long bucketsize
,
5119 unsigned long numentries
,
5122 unsigned int *_hash_shift
,
5123 unsigned int *_hash_mask
,
5124 unsigned long limit
)
5126 unsigned long long max
= limit
;
5127 unsigned long log2qty
, size
;
5130 /* allow the kernel cmdline to have a say */
5132 /* round applicable memory size up to nearest megabyte */
5133 numentries
= nr_kernel_pages
;
5134 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5135 numentries
>>= 20 - PAGE_SHIFT
;
5136 numentries
<<= 20 - PAGE_SHIFT
;
5138 /* limit to 1 bucket per 2^scale bytes of low memory */
5139 if (scale
> PAGE_SHIFT
)
5140 numentries
>>= (scale
- PAGE_SHIFT
);
5142 numentries
<<= (PAGE_SHIFT
- scale
);
5144 /* Make sure we've got at least a 0-order allocation.. */
5145 if (unlikely(flags
& HASH_SMALL
)) {
5146 /* Makes no sense without HASH_EARLY */
5147 WARN_ON(!(flags
& HASH_EARLY
));
5148 if (!(numentries
>> *_hash_shift
)) {
5149 numentries
= 1UL << *_hash_shift
;
5150 BUG_ON(!numentries
);
5152 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5153 numentries
= PAGE_SIZE
/ bucketsize
;
5155 numentries
= roundup_pow_of_two(numentries
);
5157 /* limit allocation size to 1/16 total memory by default */
5159 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5160 do_div(max
, bucketsize
);
5163 if (numentries
> max
)
5166 log2qty
= ilog2(numentries
);
5169 size
= bucketsize
<< log2qty
;
5170 if (flags
& HASH_EARLY
)
5171 table
= alloc_bootmem_nopanic(size
);
5173 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5176 * If bucketsize is not a power-of-two, we may free
5177 * some pages at the end of hash table which
5178 * alloc_pages_exact() automatically does
5180 if (get_order(size
) < MAX_ORDER
) {
5181 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5182 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5185 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5188 panic("Failed to allocate %s hash table\n", tablename
);
5190 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5193 ilog2(size
) - PAGE_SHIFT
,
5197 *_hash_shift
= log2qty
;
5199 *_hash_mask
= (1 << log2qty
) - 1;
5204 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5205 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5208 #ifdef CONFIG_SPARSEMEM
5209 return __pfn_to_section(pfn
)->pageblock_flags
;
5211 return zone
->pageblock_flags
;
5212 #endif /* CONFIG_SPARSEMEM */
5215 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5217 #ifdef CONFIG_SPARSEMEM
5218 pfn
&= (PAGES_PER_SECTION
-1);
5219 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5221 pfn
= pfn
- zone
->zone_start_pfn
;
5222 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5223 #endif /* CONFIG_SPARSEMEM */
5227 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5228 * @page: The page within the block of interest
5229 * @start_bitidx: The first bit of interest to retrieve
5230 * @end_bitidx: The last bit of interest
5231 * returns pageblock_bits flags
5233 unsigned long get_pageblock_flags_group(struct page
*page
,
5234 int start_bitidx
, int end_bitidx
)
5237 unsigned long *bitmap
;
5238 unsigned long pfn
, bitidx
;
5239 unsigned long flags
= 0;
5240 unsigned long value
= 1;
5242 zone
= page_zone(page
);
5243 pfn
= page_to_pfn(page
);
5244 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5245 bitidx
= pfn_to_bitidx(zone
, pfn
);
5247 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5248 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5255 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5256 * @page: The page within the block of interest
5257 * @start_bitidx: The first bit of interest
5258 * @end_bitidx: The last bit of interest
5259 * @flags: The flags to set
5261 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5262 int start_bitidx
, int end_bitidx
)
5265 unsigned long *bitmap
;
5266 unsigned long pfn
, bitidx
;
5267 unsigned long value
= 1;
5269 zone
= page_zone(page
);
5270 pfn
= page_to_pfn(page
);
5271 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5272 bitidx
= pfn_to_bitidx(zone
, pfn
);
5273 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5274 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5276 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5278 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5280 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5284 * This is designed as sub function...plz see page_isolation.c also.
5285 * set/clear page block's type to be ISOLATE.
5286 * page allocater never alloc memory from ISOLATE block.
5289 int set_migratetype_isolate(struct page
*page
)
5292 struct page
*curr_page
;
5293 unsigned long flags
, pfn
, iter
;
5294 unsigned long immobile
= 0;
5295 struct memory_isolate_notify arg
;
5300 zone
= page_zone(page
);
5301 zone_idx
= zone_idx(zone
);
5303 spin_lock_irqsave(&zone
->lock
, flags
);
5304 if (get_pageblock_migratetype(page
) == MIGRATE_MOVABLE
||
5305 zone_idx
== ZONE_MOVABLE
) {
5310 pfn
= page_to_pfn(page
);
5311 arg
.start_pfn
= pfn
;
5312 arg
.nr_pages
= pageblock_nr_pages
;
5313 arg
.pages_found
= 0;
5316 * It may be possible to isolate a pageblock even if the
5317 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5318 * notifier chain is used by balloon drivers to return the
5319 * number of pages in a range that are held by the balloon
5320 * driver to shrink memory. If all the pages are accounted for
5321 * by balloons, are free, or on the LRU, isolation can continue.
5322 * Later, for example, when memory hotplug notifier runs, these
5323 * pages reported as "can be isolated" should be isolated(freed)
5324 * by the balloon driver through the memory notifier chain.
5326 notifier_ret
= memory_isolate_notify(MEM_ISOLATE_COUNT
, &arg
);
5327 notifier_ret
= notifier_to_errno(notifier_ret
);
5328 if (notifier_ret
|| !arg
.pages_found
)
5331 for (iter
= pfn
; iter
< (pfn
+ pageblock_nr_pages
); iter
++) {
5332 if (!pfn_valid_within(pfn
))
5335 curr_page
= pfn_to_page(iter
);
5336 if (!page_count(curr_page
) || PageLRU(curr_page
))
5342 if (arg
.pages_found
== immobile
)
5347 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
5348 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
5351 spin_unlock_irqrestore(&zone
->lock
, flags
);
5357 void unset_migratetype_isolate(struct page
*page
)
5360 unsigned long flags
;
5361 zone
= page_zone(page
);
5362 spin_lock_irqsave(&zone
->lock
, flags
);
5363 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
5365 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5366 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
5368 spin_unlock_irqrestore(&zone
->lock
, flags
);
5371 #ifdef CONFIG_MEMORY_HOTREMOVE
5373 * All pages in the range must be isolated before calling this.
5376 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
5382 unsigned long flags
;
5383 /* find the first valid pfn */
5384 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
5389 zone
= page_zone(pfn_to_page(pfn
));
5390 spin_lock_irqsave(&zone
->lock
, flags
);
5392 while (pfn
< end_pfn
) {
5393 if (!pfn_valid(pfn
)) {
5397 page
= pfn_to_page(pfn
);
5398 BUG_ON(page_count(page
));
5399 BUG_ON(!PageBuddy(page
));
5400 order
= page_order(page
);
5401 #ifdef CONFIG_DEBUG_VM
5402 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
5403 pfn
, 1 << order
, end_pfn
);
5405 list_del(&page
->lru
);
5406 rmv_page_order(page
);
5407 zone
->free_area
[order
].nr_free
--;
5408 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
5410 for (i
= 0; i
< (1 << order
); i
++)
5411 SetPageReserved((page
+i
));
5412 pfn
+= (1 << order
);
5414 spin_unlock_irqrestore(&zone
->lock
, flags
);
5418 #ifdef CONFIG_MEMORY_FAILURE
5419 bool is_free_buddy_page(struct page
*page
)
5421 struct zone
*zone
= page_zone(page
);
5422 unsigned long pfn
= page_to_pfn(page
);
5423 unsigned long flags
;
5426 spin_lock_irqsave(&zone
->lock
, flags
);
5427 for (order
= 0; order
< MAX_ORDER
; order
++) {
5428 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
5430 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
5433 spin_unlock_irqrestore(&zone
->lock
, flags
);
5435 return order
< MAX_ORDER
;
5439 static struct trace_print_flags pageflag_names
[] = {
5440 {1UL << PG_locked
, "locked" },
5441 {1UL << PG_error
, "error" },
5442 {1UL << PG_referenced
, "referenced" },
5443 {1UL << PG_uptodate
, "uptodate" },
5444 {1UL << PG_dirty
, "dirty" },
5445 {1UL << PG_lru
, "lru" },
5446 {1UL << PG_active
, "active" },
5447 {1UL << PG_slab
, "slab" },
5448 {1UL << PG_owner_priv_1
, "owner_priv_1" },
5449 {1UL << PG_arch_1
, "arch_1" },
5450 {1UL << PG_reserved
, "reserved" },
5451 {1UL << PG_private
, "private" },
5452 {1UL << PG_private_2
, "private_2" },
5453 {1UL << PG_writeback
, "writeback" },
5454 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5455 {1UL << PG_head
, "head" },
5456 {1UL << PG_tail
, "tail" },
5458 {1UL << PG_compound
, "compound" },
5460 {1UL << PG_swapcache
, "swapcache" },
5461 {1UL << PG_mappedtodisk
, "mappedtodisk" },
5462 {1UL << PG_reclaim
, "reclaim" },
5463 {1UL << PG_buddy
, "buddy" },
5464 {1UL << PG_swapbacked
, "swapbacked" },
5465 {1UL << PG_unevictable
, "unevictable" },
5467 {1UL << PG_mlocked
, "mlocked" },
5469 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5470 {1UL << PG_uncached
, "uncached" },
5472 #ifdef CONFIG_MEMORY_FAILURE
5473 {1UL << PG_hwpoison
, "hwpoison" },
5478 static void dump_page_flags(unsigned long flags
)
5480 const char *delim
= "";
5484 printk(KERN_ALERT
"page flags: %#lx(", flags
);
5486 /* remove zone id */
5487 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
5489 for (i
= 0; pageflag_names
[i
].name
&& flags
; i
++) {
5491 mask
= pageflag_names
[i
].mask
;
5492 if ((flags
& mask
) != mask
)
5496 printk("%s%s", delim
, pageflag_names
[i
].name
);
5500 /* check for left over flags */
5502 printk("%s%#lx", delim
, flags
);
5507 void dump_page(struct page
*page
)
5510 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5511 page
, page_count(page
), page_mapcount(page
),
5512 page
->mapping
, page
->index
);
5513 dump_page_flags(page
->flags
);