1 /* Modified by Broadcom Corp. Portions Copyright (c) Broadcom Corp, 2012. */
3 * linux/mm/page_alloc.c
5 * Manages the free list, the system allocates free pages here.
6 * Note that kmalloc() lives in slab.c
8 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
9 * Swap reorganised 29.12.95, Stephen Tweedie
10 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
11 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
12 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
13 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
14 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
15 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
18 #include <linux/stddef.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/jiffies.h>
24 #include <linux/bootmem.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/oom.h>
34 #include <linux/notifier.h>
35 #include <linux/topology.h>
36 #include <linux/sysctl.h>
37 #include <linux/cpu.h>
38 #include <linux/cpuset.h>
39 #include <linux/memory_hotplug.h>
40 #include <linux/nodemask.h>
41 #include <linux/vmalloc.h>
42 #include <linux/mempolicy.h>
43 #include <linux/stop_machine.h>
44 #include <linux/sort.h>
45 #include <linux/pfn.h>
46 #include <linux/backing-dev.h>
47 #include <linux/fault-inject.h>
48 #include <linux/page-isolation.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/debugobjects.h>
51 #include <linux/kmemleak.h>
52 #include <linux/memory.h>
53 #include <linux/compaction.h>
54 #include <trace/events/kmem.h>
55 #include <linux/ftrace_event.h>
57 #include <asm/tlbflush.h>
58 #include <asm/div64.h>
64 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
65 DEFINE_PER_CPU(int, numa_node
);
66 EXPORT_PER_CPU_SYMBOL(numa_node
);
69 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
71 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
72 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
73 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
74 * defined in <linux/topology.h>.
76 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
77 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
81 * Array of node states.
83 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
84 [N_POSSIBLE
] = NODE_MASK_ALL
,
85 [N_ONLINE
] = { { [0] = 1UL } },
87 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
89 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
91 [N_CPU
] = { { [0] = 1UL } },
94 EXPORT_SYMBOL(node_states
);
96 unsigned long totalram_pages __read_mostly
;
97 unsigned long totalreserve_pages __read_mostly
;
98 int percpu_pagelist_fraction
;
99 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
101 #ifdef CONFIG_PM_SLEEP
103 * The following functions are used by the suspend/hibernate code to temporarily
104 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
105 * while devices are suspended. To avoid races with the suspend/hibernate code,
106 * they should always be called with pm_mutex held (gfp_allowed_mask also should
107 * only be modified with pm_mutex held, unless the suspend/hibernate code is
108 * guaranteed not to run in parallel with that modification).
111 static gfp_t saved_gfp_mask
;
113 void pm_restore_gfp_mask(void)
115 WARN_ON(!mutex_is_locked(&pm_mutex
));
116 if (saved_gfp_mask
) {
117 gfp_allowed_mask
= saved_gfp_mask
;
122 void pm_restrict_gfp_mask(void)
124 WARN_ON(!mutex_is_locked(&pm_mutex
));
125 WARN_ON(saved_gfp_mask
);
126 saved_gfp_mask
= gfp_allowed_mask
;
127 gfp_allowed_mask
&= ~GFP_IOFS
;
129 #endif /* CONFIG_PM_SLEEP */
131 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
132 int pageblock_order __read_mostly
;
135 static void __free_pages_ok(struct page
*page
, unsigned int order
);
138 * results with 256, 32 in the lowmem_reserve sysctl:
139 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
140 * 1G machine -> (16M dma, 784M normal, 224M high)
141 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
142 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
143 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
145 * TBD: should special case ZONE_DMA32 machines here - in those we normally
146 * don't need any ZONE_NORMAL reservation
148 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
149 #ifdef CONFIG_ZONE_DMA
152 #ifdef CONFIG_ZONE_DMA32
155 #ifdef CONFIG_HIGHMEM
161 EXPORT_SYMBOL(totalram_pages
);
163 static char * const zone_names
[MAX_NR_ZONES
] = {
164 #ifdef CONFIG_ZONE_DMA
167 #ifdef CONFIG_ZONE_DMA32
171 #ifdef CONFIG_HIGHMEM
177 int min_free_kbytes
= 1024;
179 static unsigned long __meminitdata nr_kernel_pages
;
180 static unsigned long __meminitdata nr_all_pages
;
181 static unsigned long __meminitdata dma_reserve
;
183 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
185 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
186 * ranges of memory (RAM) that may be registered with add_active_range().
187 * Ranges passed to add_active_range() will be merged if possible
188 * so the number of times add_active_range() can be called is
189 * related to the number of nodes and the number of holes
191 #ifdef CONFIG_MAX_ACTIVE_REGIONS
192 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
193 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
195 #if MAX_NUMNODES >= 32
196 /* If there can be many nodes, allow up to 50 holes per node */
197 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
199 /* By default, allow up to 256 distinct regions */
200 #define MAX_ACTIVE_REGIONS 256
204 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
205 static int __meminitdata nr_nodemap_entries
;
206 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
207 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
208 static unsigned long __initdata required_kernelcore
;
209 static unsigned long __initdata required_movablecore
;
210 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
212 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
214 EXPORT_SYMBOL(movable_zone
);
215 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
218 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
219 int nr_online_nodes __read_mostly
= 1;
220 EXPORT_SYMBOL(nr_node_ids
);
221 EXPORT_SYMBOL(nr_online_nodes
);
224 int page_group_by_mobility_disabled __read_mostly
;
226 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
229 if (unlikely(page_group_by_mobility_disabled
))
230 migratetype
= MIGRATE_UNMOVABLE
;
232 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
233 PB_migrate
, PB_migrate_end
);
236 bool oom_killer_disabled __read_mostly
;
238 #ifdef CONFIG_DEBUG_VM
239 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
243 unsigned long pfn
= page_to_pfn(page
);
246 seq
= zone_span_seqbegin(zone
);
247 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
249 else if (pfn
< zone
->zone_start_pfn
)
251 } while (zone_span_seqretry(zone
, seq
));
256 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
258 if (!pfn_valid_within(page_to_pfn(page
)))
260 if (zone
!= page_zone(page
))
266 * Temporary debugging check for pages not lying within a given zone.
268 static int bad_range(struct zone
*zone
, struct page
*page
)
270 if (page_outside_zone_boundaries(zone
, page
))
272 if (!page_is_consistent(zone
, page
))
278 static inline int bad_range(struct zone
*zone
, struct page
*page
)
284 static void bad_page(struct page
*page
)
286 static unsigned long resume
;
287 static unsigned long nr_shown
;
288 static unsigned long nr_unshown
;
290 /* Don't complain about poisoned pages */
291 if (PageHWPoison(page
)) {
292 __ClearPageBuddy(page
);
297 * Allow a burst of 60 reports, then keep quiet for that minute;
298 * or allow a steady drip of one report per second.
300 if (nr_shown
== 60) {
301 if (time_before(jiffies
, resume
)) {
307 "BUG: Bad page state: %lu messages suppressed\n",
314 resume
= jiffies
+ 60 * HZ
;
316 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
317 current
->comm
, page_to_pfn(page
));
322 /* Leave bad fields for debug, except PageBuddy could make trouble */
323 __ClearPageBuddy(page
);
324 add_taint(TAINT_BAD_PAGE
);
328 * Higher-order pages are called "compound pages". They are structured thusly:
330 * The first PAGE_SIZE page is called the "head page".
332 * The remaining PAGE_SIZE pages are called "tail pages".
334 * All pages have PG_compound set. All pages have their ->private pointing at
335 * the head page (even the head page has this).
337 * The first tail page's ->lru.next holds the address of the compound page's
338 * put_page() function. Its ->lru.prev holds the order of allocation.
339 * This usage means that zero-order pages may not be compound.
342 static void free_compound_page(struct page
*page
)
344 __free_pages_ok(page
, compound_order(page
));
347 void prep_compound_page(struct page
*page
, unsigned long order
)
350 int nr_pages
= 1 << order
;
352 set_compound_page_dtor(page
, free_compound_page
);
353 set_compound_order(page
, order
);
355 for (i
= 1; i
< nr_pages
; i
++) {
356 struct page
*p
= page
+ i
;
359 p
->first_page
= page
;
363 static int destroy_compound_page(struct page
*page
, unsigned long order
)
366 int nr_pages
= 1 << order
;
369 if (unlikely(compound_order(page
) != order
) ||
370 unlikely(!PageHead(page
))) {
375 __ClearPageHead(page
);
377 for (i
= 1; i
< nr_pages
; i
++) {
378 struct page
*p
= page
+ i
;
380 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
390 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
395 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
396 * and __GFP_HIGHMEM from hard or soft interrupt context.
398 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
399 for (i
= 0; i
< (1 << order
); i
++)
400 clear_highpage(page
+ i
);
403 static inline void set_page_order(struct page
*page
, int order
)
405 set_page_private(page
, order
);
406 __SetPageBuddy(page
);
409 static inline void rmv_page_order(struct page
*page
)
411 __ClearPageBuddy(page
);
412 set_page_private(page
, 0);
416 * Locate the struct page for both the matching buddy in our
417 * pair (buddy1) and the combined O(n+1) page they form (page).
419 * 1) Any buddy B1 will have an order O twin B2 which satisfies
420 * the following equation:
422 * For example, if the starting buddy (buddy2) is #8 its order
424 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
426 * 2) Any buddy B will have an order O+1 parent P which
427 * satisfies the following equation:
430 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
432 static inline struct page
*
433 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
435 unsigned long buddy_idx
= page_idx
^ (1 << order
);
437 return page
+ (buddy_idx
- page_idx
);
440 static inline unsigned long
441 __find_combined_index(unsigned long page_idx
, unsigned int order
)
443 return (page_idx
& ~(1 << order
));
447 * This function checks whether a page is free && is the buddy
448 * we can do coalesce a page and its buddy if
449 * (a) the buddy is not in a hole &&
450 * (b) the buddy is in the buddy system &&
451 * (c) a page and its buddy have the same order &&
452 * (d) a page and its buddy are in the same zone.
454 * For recording whether a page is in the buddy system, we use PG_buddy.
455 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
457 * For recording page's order, we use page_private(page).
459 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
462 if (!pfn_valid_within(page_to_pfn(buddy
)))
465 if (page_zone_id(page
) != page_zone_id(buddy
))
468 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
469 VM_BUG_ON(page_count(buddy
) != 0);
476 * Freeing function for a buddy system allocator.
478 * The concept of a buddy system is to maintain direct-mapped table
479 * (containing bit values) for memory blocks of various "orders".
480 * The bottom level table contains the map for the smallest allocatable
481 * units of memory (here, pages), and each level above it describes
482 * pairs of units from the levels below, hence, "buddies".
483 * At a high level, all that happens here is marking the table entry
484 * at the bottom level available, and propagating the changes upward
485 * as necessary, plus some accounting needed to play nicely with other
486 * parts of the VM system.
487 * At each level, we keep a list of pages, which are heads of continuous
488 * free pages of length of (1 << order) and marked with PG_buddy. Page's
489 * order is recorded in page_private(page) field.
490 * So when we are allocating or freeing one, we can derive the state of the
491 * other. That is, if we allocate a small block, and both were
492 * free, the remainder of the region must be split into blocks.
493 * If a block is freed, and its buddy is also free, then this
494 * triggers coalescing into a block of larger size.
499 static inline void __free_one_page(struct page
*page
,
500 struct zone
*zone
, unsigned int order
,
503 unsigned long page_idx
;
504 unsigned long combined_idx
;
507 if (unlikely(PageCompound(page
)))
508 if (unlikely(destroy_compound_page(page
, order
)))
511 VM_BUG_ON(migratetype
== -1);
513 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
515 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
516 VM_BUG_ON(bad_range(zone
, page
));
518 while (order
< MAX_ORDER
-1) {
519 buddy
= __page_find_buddy(page
, page_idx
, order
);
520 if (!page_is_buddy(page
, buddy
, order
))
523 /* Our buddy is free, merge with it and move up one order. */
524 list_del(&buddy
->lru
);
525 zone
->free_area
[order
].nr_free
--;
526 rmv_page_order(buddy
);
527 combined_idx
= __find_combined_index(page_idx
, order
);
528 page
= page
+ (combined_idx
- page_idx
);
529 page_idx
= combined_idx
;
532 set_page_order(page
, order
);
535 * If this is not the largest possible page, check if the buddy
536 * of the next-highest order is free. If it is, it's possible
537 * that pages are being freed that will coalesce soon. In case,
538 * that is happening, add the free page to the tail of the list
539 * so it's less likely to be used soon and more likely to be merged
540 * as a higher order page
542 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
543 struct page
*higher_page
, *higher_buddy
;
544 combined_idx
= __find_combined_index(page_idx
, order
);
545 higher_page
= page
+ combined_idx
- page_idx
;
546 higher_buddy
= __page_find_buddy(higher_page
, combined_idx
, order
+ 1);
547 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
548 list_add_tail(&page
->lru
,
549 &zone
->free_area
[order
].free_list
[migratetype
]);
554 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
556 zone
->free_area
[order
].nr_free
++;
560 * free_page_mlock() -- clean up attempts to free and mlocked() page.
561 * Page should not be on lru, so no need to fix that up.
562 * free_pages_check() will verify...
564 static inline void free_page_mlock(struct page
*page
)
566 __dec_zone_page_state(page
, NR_MLOCK
);
567 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
570 static inline int free_pages_check(struct page
*page
)
572 if (unlikely(page_mapcount(page
) |
573 (page
->mapping
!= NULL
) |
574 (atomic_read(&page
->_count
) != 0) |
575 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
579 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
580 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
585 * Frees a number of pages from the PCP lists
586 * Assumes all pages on list are in same zone, and of same order.
587 * count is the number of pages to free.
589 * If the zone was previously in an "all pages pinned" state then look to
590 * see if this freeing clears that state.
592 * And clear the zone's pages_scanned counter, to hold off the "all pages are
593 * pinned" detection logic.
595 static void free_pcppages_bulk(struct zone
*zone
, int count
,
596 struct per_cpu_pages
*pcp
)
602 spin_lock(&zone
->lock
);
603 zone
->all_unreclaimable
= 0;
604 zone
->pages_scanned
= 0;
608 struct list_head
*list
;
611 * Remove pages from lists in a round-robin fashion. A
612 * batch_free count is maintained that is incremented when an
613 * empty list is encountered. This is so more pages are freed
614 * off fuller lists instead of spinning excessively around empty
619 if (++migratetype
== MIGRATE_PCPTYPES
)
621 list
= &pcp
->lists
[migratetype
];
622 } while (list_empty(list
));
625 page
= list_entry(list
->prev
, struct page
, lru
);
626 /* must delete as __free_one_page list manipulates */
627 list_del(&page
->lru
);
628 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
629 __free_one_page(page
, zone
, 0, page_private(page
));
630 trace_mm_page_pcpu_drain(page
, 0, page_private(page
));
631 } while (--to_free
&& --batch_free
&& !list_empty(list
));
633 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
);
634 spin_unlock(&zone
->lock
);
637 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
640 spin_lock(&zone
->lock
);
641 zone
->all_unreclaimable
= 0;
642 zone
->pages_scanned
= 0;
644 __free_one_page(page
, zone
, order
, migratetype
);
645 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
646 spin_unlock(&zone
->lock
);
649 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
654 trace_mm_page_free_direct(page
, order
);
655 kmemcheck_free_shadow(page
, order
);
657 for (i
= 0; i
< (1 << order
); i
++) {
658 struct page
*pg
= page
+ i
;
662 bad
+= free_pages_check(pg
);
667 if (!PageHighMem(page
)) {
668 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
669 debug_check_no_obj_freed(page_address(page
),
672 arch_free_page(page
, order
);
673 kernel_map_pages(page
, 1 << order
, 0);
678 static void __free_pages_ok(struct page
*page
, unsigned int order
)
681 int wasMlocked
= __TestClearPageMlocked(page
);
683 if (!free_pages_prepare(page
, order
))
686 local_irq_save(flags
);
687 if (unlikely(wasMlocked
))
688 free_page_mlock(page
);
689 __count_vm_events(PGFREE
, 1 << order
);
690 free_one_page(page_zone(page
), page
, order
,
691 get_pageblock_migratetype(page
));
692 local_irq_restore(flags
);
696 * permit the bootmem allocator to evade page validation on high-order frees
698 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
701 __ClearPageReserved(page
);
702 set_page_count(page
, 0);
703 set_page_refcounted(page
);
709 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
710 struct page
*p
= &page
[loop
];
712 if (loop
+ 1 < BITS_PER_LONG
)
714 __ClearPageReserved(p
);
715 set_page_count(p
, 0);
718 set_page_refcounted(page
);
719 __free_pages(page
, order
);
725 * The order of subdivision here is critical for the IO subsystem.
726 * Please do not alter this order without good reasons and regression
727 * testing. Specifically, as large blocks of memory are subdivided,
728 * the order in which smaller blocks are delivered depends on the order
729 * they're subdivided in this function. This is the primary factor
730 * influencing the order in which pages are delivered to the IO
731 * subsystem according to empirical testing, and this is also justified
732 * by considering the behavior of a buddy system containing a single
733 * large block of memory acted on by a series of small allocations.
734 * This behavior is a critical factor in sglist merging's success.
738 static inline void expand(struct zone
*zone
, struct page
*page
,
739 int low
, int high
, struct free_area
*area
,
742 unsigned long size
= 1 << high
;
748 VM_BUG_ON(bad_range(zone
, &page
[size
]));
749 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
751 set_page_order(&page
[size
], high
);
756 * This page is about to be returned from the page allocator
758 static inline int check_new_page(struct page
*page
)
760 if (unlikely(page_mapcount(page
) |
761 (page
->mapping
!= NULL
) |
762 (atomic_read(&page
->_count
) != 0) |
763 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
770 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
774 for (i
= 0; i
< (1 << order
); i
++) {
775 struct page
*p
= page
+ i
;
776 if (unlikely(check_new_page(p
)))
780 set_page_private(page
, 0);
781 set_page_refcounted(page
);
783 arch_alloc_page(page
, order
);
784 kernel_map_pages(page
, 1 << order
, 1);
786 if (gfp_flags
& __GFP_ZERO
)
787 prep_zero_page(page
, order
, gfp_flags
);
789 if (order
&& (gfp_flags
& __GFP_COMP
))
790 prep_compound_page(page
, order
);
796 * Go through the free lists for the given migratetype and remove
797 * the smallest available page from the freelists
800 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
803 unsigned int current_order
;
804 struct free_area
* area
;
807 /* Find a page of the appropriate size in the preferred list */
808 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
809 area
= &(zone
->free_area
[current_order
]);
810 if (list_empty(&area
->free_list
[migratetype
]))
813 page
= list_entry(area
->free_list
[migratetype
].next
,
815 list_del(&page
->lru
);
816 rmv_page_order(page
);
818 expand(zone
, page
, order
, current_order
, area
, migratetype
);
827 * This array describes the order lists are fallen back to when
828 * the free lists for the desirable migrate type are depleted
830 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
831 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
832 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
833 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
834 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
838 * Move the free pages in a range to the free lists of the requested type.
839 * Note that start_page and end_pages are not aligned on a pageblock
840 * boundary. If alignment is required, use move_freepages_block()
842 static int move_freepages(struct zone
*zone
,
843 struct page
*start_page
, struct page
*end_page
,
850 #ifndef CONFIG_HOLES_IN_ZONE
852 * page_zone is not safe to call in this context when
853 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
854 * anyway as we check zone boundaries in move_freepages_block().
855 * Remove at a later date when no bug reports exist related to
856 * grouping pages by mobility
858 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
861 for (page
= start_page
; page
<= end_page
;) {
862 /* Make sure we are not inadvertently changing nodes */
863 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
865 if (!pfn_valid_within(page_to_pfn(page
))) {
870 if (!PageBuddy(page
)) {
875 order
= page_order(page
);
876 list_del(&page
->lru
);
878 &zone
->free_area
[order
].free_list
[migratetype
]);
880 pages_moved
+= 1 << order
;
886 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
889 unsigned long start_pfn
, end_pfn
;
890 struct page
*start_page
, *end_page
;
892 start_pfn
= page_to_pfn(page
);
893 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
894 start_page
= pfn_to_page(start_pfn
);
895 end_page
= start_page
+ pageblock_nr_pages
- 1;
896 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
898 /* Do not cross zone boundaries */
899 if (start_pfn
< zone
->zone_start_pfn
)
901 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
904 return move_freepages(zone
, start_page
, end_page
, migratetype
);
907 static void change_pageblock_range(struct page
*pageblock_page
,
908 int start_order
, int migratetype
)
910 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
912 while (nr_pageblocks
--) {
913 set_pageblock_migratetype(pageblock_page
, migratetype
);
914 pageblock_page
+= pageblock_nr_pages
;
918 /* Remove an element from the buddy allocator from the fallback list */
919 static inline struct page
*
920 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
922 struct free_area
* area
;
927 /* Find the largest possible block of pages in the other list */
928 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
930 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
931 migratetype
= fallbacks
[start_migratetype
][i
];
933 /* MIGRATE_RESERVE handled later if necessary */
934 if (migratetype
== MIGRATE_RESERVE
)
937 area
= &(zone
->free_area
[current_order
]);
938 if (list_empty(&area
->free_list
[migratetype
]))
941 page
= list_entry(area
->free_list
[migratetype
].next
,
946 * If breaking a large block of pages, move all free
947 * pages to the preferred allocation list. If falling
948 * back for a reclaimable kernel allocation, be more
949 * agressive about taking ownership of free pages
951 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
952 start_migratetype
== MIGRATE_RECLAIMABLE
||
953 page_group_by_mobility_disabled
) {
955 pages
= move_freepages_block(zone
, page
,
958 /* Claim the whole block if over half of it is free */
959 if (pages
>= (1 << (pageblock_order
-1)) ||
960 page_group_by_mobility_disabled
)
961 set_pageblock_migratetype(page
,
964 migratetype
= start_migratetype
;
967 /* Remove the page from the freelists */
968 list_del(&page
->lru
);
969 rmv_page_order(page
);
971 /* Take ownership for orders >= pageblock_order */
972 if (current_order
>= pageblock_order
)
973 change_pageblock_range(page
, current_order
,
976 expand(zone
, page
, order
, current_order
, area
, migratetype
);
978 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
979 start_migratetype
, migratetype
);
989 * Do the hard work of removing an element from the buddy allocator.
990 * Call me with the zone->lock already held.
992 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
998 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1000 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1001 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1004 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1005 * is used because __rmqueue_smallest is an inline function
1006 * and we want just one call site
1009 migratetype
= MIGRATE_RESERVE
;
1014 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1019 * Obtain a specified number of elements from the buddy allocator, all under
1020 * a single hold of the lock, for efficiency. Add them to the supplied list.
1021 * Returns the number of new pages which were placed at *list.
1023 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1024 unsigned long count
, struct list_head
*list
,
1025 int migratetype
, int cold
)
1029 spin_lock(&zone
->lock
);
1030 for (i
= 0; i
< count
; ++i
) {
1031 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1032 if (unlikely(page
== NULL
))
1036 * Split buddy pages returned by expand() are received here
1037 * in physical page order. The page is added to the callers and
1038 * list and the list head then moves forward. From the callers
1039 * perspective, the linked list is ordered by page number in
1040 * some conditions. This is useful for IO devices that can
1041 * merge IO requests if the physical pages are ordered
1044 if (likely(cold
== 0))
1045 list_add(&page
->lru
, list
);
1047 list_add_tail(&page
->lru
, list
);
1048 set_page_private(page
, migratetype
);
1051 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1052 spin_unlock(&zone
->lock
);
1058 * Called from the vmstat counter updater to drain pagesets of this
1059 * currently executing processor on remote nodes after they have
1062 * Note that this function must be called with the thread pinned to
1063 * a single processor.
1065 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1067 unsigned long flags
;
1070 local_irq_save(flags
);
1071 if (pcp
->count
>= pcp
->batch
)
1072 to_drain
= pcp
->batch
;
1074 to_drain
= pcp
->count
;
1075 free_pcppages_bulk(zone
, to_drain
, pcp
);
1076 pcp
->count
-= to_drain
;
1077 local_irq_restore(flags
);
1082 * Drain pages of the indicated processor.
1084 * The processor must either be the current processor and the
1085 * thread pinned to the current processor or a processor that
1088 static void drain_pages(unsigned int cpu
)
1090 unsigned long flags
;
1093 for_each_populated_zone(zone
) {
1094 struct per_cpu_pageset
*pset
;
1095 struct per_cpu_pages
*pcp
;
1097 local_irq_save(flags
);
1098 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1101 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1103 local_irq_restore(flags
);
1108 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1110 void drain_local_pages(void *arg
)
1112 drain_pages(smp_processor_id());
1116 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1118 void drain_all_pages(void)
1120 on_each_cpu(drain_local_pages
, NULL
, 1);
1123 #ifdef CONFIG_HIBERNATION
1125 void mark_free_pages(struct zone
*zone
)
1127 unsigned long pfn
, max_zone_pfn
;
1128 unsigned long flags
;
1130 struct list_head
*curr
;
1132 if (!zone
->spanned_pages
)
1135 spin_lock_irqsave(&zone
->lock
, flags
);
1137 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1138 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1139 if (pfn_valid(pfn
)) {
1140 struct page
*page
= pfn_to_page(pfn
);
1142 if (!swsusp_page_is_forbidden(page
))
1143 swsusp_unset_page_free(page
);
1146 for_each_migratetype_order(order
, t
) {
1147 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1150 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1151 for (i
= 0; i
< (1UL << order
); i
++)
1152 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1155 spin_unlock_irqrestore(&zone
->lock
, flags
);
1157 #endif /* CONFIG_PM */
1160 * Free a 0-order page
1161 * cold == 1 ? free a cold page : free a hot page
1163 void free_hot_cold_page(struct page
*page
, int cold
)
1165 struct zone
*zone
= page_zone(page
);
1166 struct per_cpu_pages
*pcp
;
1167 unsigned long flags
;
1169 int wasMlocked
= __TestClearPageMlocked(page
);
1171 if (!free_pages_prepare(page
, 0))
1174 migratetype
= get_pageblock_migratetype(page
);
1175 set_page_private(page
, migratetype
);
1176 local_irq_save(flags
);
1177 if (unlikely(wasMlocked
))
1178 free_page_mlock(page
);
1179 __count_vm_event(PGFREE
);
1182 * We only track unmovable, reclaimable and movable on pcp lists.
1183 * Free ISOLATE pages back to the allocator because they are being
1184 * offlined but treat RESERVE as movable pages so we can get those
1185 * areas back if necessary. Otherwise, we may have to free
1186 * excessively into the page allocator
1188 if (migratetype
>= MIGRATE_PCPTYPES
) {
1189 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1190 free_one_page(zone
, page
, 0, migratetype
);
1193 migratetype
= MIGRATE_MOVABLE
;
1196 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1198 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1200 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1202 if (pcp
->count
>= pcp
->high
) {
1203 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1204 pcp
->count
-= pcp
->batch
;
1208 local_irq_restore(flags
);
1212 * split_page takes a non-compound higher-order page, and splits it into
1213 * n (1<<order) sub-pages: page[0..n]
1214 * Each sub-page must be freed individually.
1216 * Note: this is probably too low level an operation for use in drivers.
1217 * Please consult with lkml before using this in your driver.
1219 void split_page(struct page
*page
, unsigned int order
)
1223 VM_BUG_ON(PageCompound(page
));
1224 VM_BUG_ON(!page_count(page
));
1226 #ifdef CONFIG_KMEMCHECK
1228 * Split shadow pages too, because free(page[0]) would
1229 * otherwise free the whole shadow.
1231 if (kmemcheck_page_is_tracked(page
))
1232 split_page(virt_to_page(page
[0].shadow
), order
);
1235 for (i
= 1; i
< (1 << order
); i
++)
1236 set_page_refcounted(page
+ i
);
1240 * Similar to split_page except the page is already free. As this is only
1241 * being used for migration, the migratetype of the block also changes.
1242 * As this is called with interrupts disabled, the caller is responsible
1243 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1246 * Note: this is probably too low level an operation for use in drivers.
1247 * Please consult with lkml before using this in your driver.
1249 int split_free_page(struct page
*page
)
1252 unsigned long watermark
;
1255 BUG_ON(!PageBuddy(page
));
1257 zone
= page_zone(page
);
1258 order
= page_order(page
);
1260 /* Obey watermarks as if the page was being allocated */
1261 watermark
= low_wmark_pages(zone
) + (1 << order
);
1262 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1265 /* Remove page from free list */
1266 list_del(&page
->lru
);
1267 zone
->free_area
[order
].nr_free
--;
1268 rmv_page_order(page
);
1269 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1UL << order
));
1271 /* Split into individual pages */
1272 set_page_refcounted(page
);
1273 split_page(page
, order
);
1275 if (order
>= pageblock_order
- 1) {
1276 struct page
*endpage
= page
+ (1 << order
) - 1;
1277 for (; page
< endpage
; page
+= pageblock_nr_pages
)
1278 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1285 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1286 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1290 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1291 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1294 unsigned long flags
;
1296 int cold
= !!(gfp_flags
& __GFP_COLD
);
1299 if (likely(order
== 0)) {
1300 struct per_cpu_pages
*pcp
;
1301 struct list_head
*list
;
1303 local_irq_save(flags
);
1304 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1305 list
= &pcp
->lists
[migratetype
];
1306 if (list_empty(list
)) {
1307 pcp
->count
+= rmqueue_bulk(zone
, 0,
1310 if (unlikely(list_empty(list
)))
1315 page
= list_entry(list
->prev
, struct page
, lru
);
1317 page
= list_entry(list
->next
, struct page
, lru
);
1319 list_del(&page
->lru
);
1322 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1324 * __GFP_NOFAIL is not to be used in new code.
1326 * All __GFP_NOFAIL callers should be fixed so that they
1327 * properly detect and handle allocation failures.
1329 * We most definitely don't want callers attempting to
1330 * allocate greater than order-1 page units with
1333 WARN_ON_ONCE(order
> 1);
1335 spin_lock_irqsave(&zone
->lock
, flags
);
1336 page
= __rmqueue(zone
, order
, migratetype
);
1337 spin_unlock(&zone
->lock
);
1340 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << order
));
1343 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1344 zone_statistics(preferred_zone
, zone
);
1345 local_irq_restore(flags
);
1347 VM_BUG_ON(bad_range(zone
, page
));
1348 if (prep_new_page(page
, order
, gfp_flags
))
1353 local_irq_restore(flags
);
1357 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1358 #define ALLOC_WMARK_MIN WMARK_MIN
1359 #define ALLOC_WMARK_LOW WMARK_LOW
1360 #define ALLOC_WMARK_HIGH WMARK_HIGH
1361 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1363 /* Mask to get the watermark bits */
1364 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1366 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1367 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1368 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1370 #ifdef CONFIG_FAIL_PAGE_ALLOC
1372 static struct fail_page_alloc_attr
{
1373 struct fault_attr attr
;
1375 u32 ignore_gfp_highmem
;
1376 u32 ignore_gfp_wait
;
1379 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1381 struct dentry
*ignore_gfp_highmem_file
;
1382 struct dentry
*ignore_gfp_wait_file
;
1383 struct dentry
*min_order_file
;
1385 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1387 } fail_page_alloc
= {
1388 .attr
= FAULT_ATTR_INITIALIZER
,
1389 .ignore_gfp_wait
= 1,
1390 .ignore_gfp_highmem
= 1,
1394 static int __init
setup_fail_page_alloc(char *str
)
1396 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1398 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1400 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1402 if (order
< fail_page_alloc
.min_order
)
1404 if (gfp_mask
& __GFP_NOFAIL
)
1406 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1408 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1411 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1414 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1416 static int __init
fail_page_alloc_debugfs(void)
1418 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1422 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1426 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1428 fail_page_alloc
.ignore_gfp_wait_file
=
1429 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1430 &fail_page_alloc
.ignore_gfp_wait
);
1432 fail_page_alloc
.ignore_gfp_highmem_file
=
1433 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1434 &fail_page_alloc
.ignore_gfp_highmem
);
1435 fail_page_alloc
.min_order_file
=
1436 debugfs_create_u32("min-order", mode
, dir
,
1437 &fail_page_alloc
.min_order
);
1439 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1440 !fail_page_alloc
.ignore_gfp_highmem_file
||
1441 !fail_page_alloc
.min_order_file
) {
1443 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1444 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1445 debugfs_remove(fail_page_alloc
.min_order_file
);
1446 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1452 late_initcall(fail_page_alloc_debugfs
);
1454 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1456 #else /* CONFIG_FAIL_PAGE_ALLOC */
1458 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1463 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1466 * Return true if free pages are above 'mark'. This takes into account the order
1467 * of the allocation.
1469 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1470 int classzone_idx
, int alloc_flags
, long free_pages
)
1472 /* free_pages my go negative - that's OK */
1476 free_pages
-= (1 << order
) + 1;
1477 if (alloc_flags
& ALLOC_HIGH
)
1479 if (alloc_flags
& ALLOC_HARDER
)
1482 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1484 for (o
= 0; o
< order
; o
++) {
1485 /* At the next order, this order's pages become unavailable */
1486 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1488 /* Require fewer higher order pages to be free */
1491 if (free_pages
<= min
)
1497 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1498 int classzone_idx
, int alloc_flags
)
1500 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1501 zone_page_state(z
, NR_FREE_PAGES
));
1504 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1505 int classzone_idx
, int alloc_flags
)
1507 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1509 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1510 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1512 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1518 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1519 * skip over zones that are not allowed by the cpuset, or that have
1520 * been recently (in last second) found to be nearly full. See further
1521 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1522 * that have to skip over a lot of full or unallowed zones.
1524 * If the zonelist cache is present in the passed in zonelist, then
1525 * returns a pointer to the allowed node mask (either the current
1526 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1528 * If the zonelist cache is not available for this zonelist, does
1529 * nothing and returns NULL.
1531 * If the fullzones BITMAP in the zonelist cache is stale (more than
1532 * a second since last zap'd) then we zap it out (clear its bits.)
1534 * We hold off even calling zlc_setup, until after we've checked the
1535 * first zone in the zonelist, on the theory that most allocations will
1536 * be satisfied from that first zone, so best to examine that zone as
1537 * quickly as we can.
1539 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1541 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1542 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1544 zlc
= zonelist
->zlcache_ptr
;
1548 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1549 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1550 zlc
->last_full_zap
= jiffies
;
1553 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1554 &cpuset_current_mems_allowed
:
1555 &node_states
[N_HIGH_MEMORY
];
1556 return allowednodes
;
1560 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1561 * if it is worth looking at further for free memory:
1562 * 1) Check that the zone isn't thought to be full (doesn't have its
1563 * bit set in the zonelist_cache fullzones BITMAP).
1564 * 2) Check that the zones node (obtained from the zonelist_cache
1565 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1566 * Return true (non-zero) if zone is worth looking at further, or
1567 * else return false (zero) if it is not.
1569 * This check -ignores- the distinction between various watermarks,
1570 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1571 * found to be full for any variation of these watermarks, it will
1572 * be considered full for up to one second by all requests, unless
1573 * we are so low on memory on all allowed nodes that we are forced
1574 * into the second scan of the zonelist.
1576 * In the second scan we ignore this zonelist cache and exactly
1577 * apply the watermarks to all zones, even it is slower to do so.
1578 * We are low on memory in the second scan, and should leave no stone
1579 * unturned looking for a free page.
1581 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1582 nodemask_t
*allowednodes
)
1584 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1585 int i
; /* index of *z in zonelist zones */
1586 int n
; /* node that zone *z is on */
1588 zlc
= zonelist
->zlcache_ptr
;
1592 i
= z
- zonelist
->_zonerefs
;
1595 /* This zone is worth trying if it is allowed but not full */
1596 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1600 * Given 'z' scanning a zonelist, set the corresponding bit in
1601 * zlc->fullzones, so that subsequent attempts to allocate a page
1602 * from that zone don't waste time re-examining it.
1604 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1606 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1607 int i
; /* index of *z in zonelist zones */
1609 zlc
= zonelist
->zlcache_ptr
;
1613 i
= z
- zonelist
->_zonerefs
;
1615 set_bit(i
, zlc
->fullzones
);
1618 #else /* CONFIG_NUMA */
1620 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1625 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1626 nodemask_t
*allowednodes
)
1631 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1634 #endif /* CONFIG_NUMA */
1637 * get_page_from_freelist goes through the zonelist trying to allocate
1640 static struct page
* BCMFASTPATH_HOST
1641 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1642 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1643 struct zone
*preferred_zone
, int migratetype
)
1646 struct page
*page
= NULL
;
1649 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1650 int zlc_active
= 0; /* set if using zonelist_cache */
1651 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1653 classzone_idx
= zone_idx(preferred_zone
);
1656 * Scan zonelist, looking for a zone with enough free.
1657 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1659 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1660 high_zoneidx
, nodemask
) {
1661 if (NUMA_BUILD
&& zlc_active
&&
1662 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1664 if ((alloc_flags
& ALLOC_CPUSET
) &&
1665 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1668 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1669 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1673 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1674 if (zone_watermark_ok(zone
, order
, mark
,
1675 classzone_idx
, alloc_flags
))
1678 if (zone_reclaim_mode
== 0)
1679 goto this_zone_full
;
1681 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1683 case ZONE_RECLAIM_NOSCAN
:
1686 case ZONE_RECLAIM_FULL
:
1687 /* scanned but unreclaimable */
1688 goto this_zone_full
;
1690 /* did we reclaim enough */
1691 if (!zone_watermark_ok(zone
, order
, mark
,
1692 classzone_idx
, alloc_flags
))
1693 goto this_zone_full
;
1698 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1699 gfp_mask
, migratetype
);
1704 zlc_mark_zone_full(zonelist
, z
);
1706 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1708 * we do zlc_setup after the first zone is tried but only
1709 * if there are multiple nodes make it worthwhile
1711 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1717 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1718 /* Disable zlc cache for second zonelist scan */
1726 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1727 unsigned long pages_reclaimed
)
1729 /* Do not loop if specifically requested */
1730 if (gfp_mask
& __GFP_NORETRY
)
1734 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1735 * means __GFP_NOFAIL, but that may not be true in other
1738 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1742 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1743 * specified, then we retry until we no longer reclaim any pages
1744 * (above), or we've reclaimed an order of pages at least as
1745 * large as the allocation's order. In both cases, if the
1746 * allocation still fails, we stop retrying.
1748 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1752 * Don't let big-order allocations loop unless the caller
1753 * explicitly requests that.
1755 if (gfp_mask
& __GFP_NOFAIL
)
1761 static inline struct page
*
1762 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1763 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1764 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1769 /* Acquire the OOM killer lock for the zones in zonelist */
1770 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
1771 schedule_timeout_uninterruptible(1);
1776 * Go through the zonelist yet one more time, keep very high watermark
1777 * here, this is only to catch a parallel oom killing, we must fail if
1778 * we're still under heavy pressure.
1780 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1781 order
, zonelist
, high_zoneidx
,
1782 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1783 preferred_zone
, migratetype
);
1787 if (!(gfp_mask
& __GFP_NOFAIL
)) {
1788 /* The OOM killer will not help higher order allocs */
1789 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1791 /* The OOM killer does not needlessly kill tasks for lowmem */
1792 if (high_zoneidx
< ZONE_NORMAL
)
1795 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1796 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1797 * The caller should handle page allocation failure by itself if
1798 * it specifies __GFP_THISNODE.
1799 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1801 if (gfp_mask
& __GFP_THISNODE
)
1804 /* Exhausted what can be done so it's blamo time */
1805 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
);
1808 clear_zonelist_oom(zonelist
, gfp_mask
);
1812 #ifdef CONFIG_COMPACTION
1813 /* Try memory compaction for high-order allocations before reclaim */
1814 static struct page
*
1815 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1816 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1817 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1818 int migratetype
, unsigned long *did_some_progress
)
1822 if (!order
|| compaction_deferred(preferred_zone
))
1825 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
1827 if (*did_some_progress
!= COMPACT_SKIPPED
) {
1829 /* Page migration frees to the PCP lists but we want merging */
1830 drain_pages(get_cpu());
1833 page
= get_page_from_freelist(gfp_mask
, nodemask
,
1834 order
, zonelist
, high_zoneidx
,
1835 alloc_flags
, preferred_zone
,
1838 preferred_zone
->compact_considered
= 0;
1839 preferred_zone
->compact_defer_shift
= 0;
1840 count_vm_event(COMPACTSUCCESS
);
1845 * It's bad if compaction run occurs and fails.
1846 * The most likely reason is that pages exist,
1847 * but not enough to satisfy watermarks.
1849 count_vm_event(COMPACTFAIL
);
1850 defer_compaction(preferred_zone
);
1858 static inline struct page
*
1859 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1860 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1861 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1862 int migratetype
, unsigned long *did_some_progress
)
1866 #endif /* CONFIG_COMPACTION */
1868 /* The really slow allocator path where we enter direct reclaim */
1869 static inline struct page
*
1870 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1871 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1872 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1873 int migratetype
, unsigned long *did_some_progress
)
1875 struct page
*page
= NULL
;
1876 struct reclaim_state reclaim_state
;
1877 struct task_struct
*p
= current
;
1878 bool drained
= false;
1882 /* We now go into synchronous reclaim */
1883 cpuset_memory_pressure_bump();
1884 p
->flags
|= PF_MEMALLOC
;
1885 lockdep_set_current_reclaim_state(gfp_mask
);
1886 reclaim_state
.reclaimed_slab
= 0;
1887 p
->reclaim_state
= &reclaim_state
;
1889 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1891 p
->reclaim_state
= NULL
;
1892 lockdep_clear_current_reclaim_state();
1893 p
->flags
&= ~PF_MEMALLOC
;
1897 if (unlikely(!(*did_some_progress
)))
1901 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1902 zonelist
, high_zoneidx
,
1903 alloc_flags
, preferred_zone
,
1907 * If an allocation failed after direct reclaim, it could be because
1908 * pages are pinned on the per-cpu lists. Drain them and try again
1910 if (!page
&& !drained
) {
1920 * This is called in the allocator slow-path if the allocation request is of
1921 * sufficient urgency to ignore watermarks and take other desperate measures
1923 static inline struct page
*
1924 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1925 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1926 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1932 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1933 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1934 preferred_zone
, migratetype
);
1936 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1937 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
1938 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1944 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1945 enum zone_type high_zoneidx
)
1950 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1951 wakeup_kswapd(zone
, order
);
1955 gfp_to_alloc_flags(gfp_t gfp_mask
)
1957 struct task_struct
*p
= current
;
1958 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
1959 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1961 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1962 BUILD_BUG_ON(__GFP_HIGH
!= ALLOC_HIGH
);
1965 * The caller may dip into page reserves a bit more if the caller
1966 * cannot run direct reclaim, or if the caller has realtime scheduling
1967 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1968 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1970 alloc_flags
|= (gfp_mask
& __GFP_HIGH
);
1973 alloc_flags
|= ALLOC_HARDER
;
1975 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1976 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1978 alloc_flags
&= ~ALLOC_CPUSET
;
1979 } else if (unlikely(rt_task(p
)) && !in_interrupt())
1980 alloc_flags
|= ALLOC_HARDER
;
1982 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
1983 if (!in_interrupt() &&
1984 ((p
->flags
& PF_MEMALLOC
) ||
1985 unlikely(test_thread_flag(TIF_MEMDIE
))))
1986 alloc_flags
|= ALLOC_NO_WATERMARKS
;
1992 static inline struct page
*
1993 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
1994 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1995 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1998 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1999 struct page
*page
= NULL
;
2001 unsigned long pages_reclaimed
= 0;
2002 unsigned long did_some_progress
;
2003 struct task_struct
*p
= current
;
2006 * In the slowpath, we sanity check order to avoid ever trying to
2007 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2008 * be using allocators in order of preference for an area that is
2011 if (order
>= MAX_ORDER
) {
2012 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2017 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2018 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2019 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2020 * using a larger set of nodes after it has established that the
2021 * allowed per node queues are empty and that nodes are
2024 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2028 wake_all_kswapd(order
, zonelist
, high_zoneidx
);
2031 * OK, we're below the kswapd watermark and have kicked background
2032 * reclaim. Now things get more complex, so set up alloc_flags according
2033 * to how we want to proceed.
2035 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2037 /* This is the last chance, in general, before the goto nopage. */
2038 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2039 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2040 preferred_zone
, migratetype
);
2045 /* Allocate without watermarks if the context allows */
2046 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2047 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2048 zonelist
, high_zoneidx
, nodemask
,
2049 preferred_zone
, migratetype
);
2054 /* Atomic allocations - we can't balance anything */
2058 /* Avoid recursion of direct reclaim */
2059 if (p
->flags
& PF_MEMALLOC
)
2062 /* Avoid allocations with no watermarks from looping endlessly */
2063 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2066 /* Try direct compaction */
2067 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2068 zonelist
, high_zoneidx
,
2070 alloc_flags
, preferred_zone
,
2071 migratetype
, &did_some_progress
);
2075 /* Try direct reclaim and then allocating */
2076 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2077 zonelist
, high_zoneidx
,
2079 alloc_flags
, preferred_zone
,
2080 migratetype
, &did_some_progress
);
2085 * If we failed to make any progress reclaiming, then we are
2086 * running out of options and have to consider going OOM
2088 if (!did_some_progress
) {
2089 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2090 if (oom_killer_disabled
)
2092 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2093 zonelist
, high_zoneidx
,
2094 nodemask
, preferred_zone
,
2099 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2101 * The oom killer is not called for high-order
2102 * allocations that may fail, so if no progress
2103 * is being made, there are no other options and
2104 * retrying is unlikely to help.
2106 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2109 * The oom killer is not called for lowmem
2110 * allocations to prevent needlessly killing
2113 if (high_zoneidx
< ZONE_NORMAL
)
2121 /* Check if we should retry the allocation */
2122 pages_reclaimed
+= did_some_progress
;
2123 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
2124 /* Wait for some write requests to complete then retry */
2125 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
2130 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
2131 printk(KERN_WARNING
"%s: page allocation failure."
2132 " order:%d, mode:0x%x\n",
2133 p
->comm
, order
, gfp_mask
);
2139 if (kmemcheck_enabled
)
2140 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2146 * This is the 'heart' of the zoned buddy allocator.
2149 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2150 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2152 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2153 struct zone
*preferred_zone
;
2155 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2157 gfp_mask
&= gfp_allowed_mask
;
2159 lockdep_trace_alloc(gfp_mask
);
2161 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2163 if (should_fail_alloc_page(gfp_mask
, order
))
2167 * Check the zones suitable for the gfp_mask contain at least one
2168 * valid zone. It's possible to have an empty zonelist as a result
2169 * of GFP_THISNODE and a memoryless node
2171 if (unlikely(!zonelist
->_zonerefs
->zone
))
2175 /* The preferred zone is used for statistics later */
2176 first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
, &preferred_zone
);
2177 if (!preferred_zone
) {
2182 /* First allocation attempt */
2183 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2184 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
2185 preferred_zone
, migratetype
);
2186 if (unlikely(!page
))
2187 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2188 zonelist
, high_zoneidx
, nodemask
,
2189 preferred_zone
, migratetype
);
2192 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2195 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2198 * Common helper functions.
2200 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2205 * __get_free_pages() returns a 32-bit address, which cannot represent
2208 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2210 page
= alloc_pages(gfp_mask
, order
);
2213 return (unsigned long) page_address(page
);
2215 EXPORT_SYMBOL(__get_free_pages
);
2217 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2219 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2221 EXPORT_SYMBOL(get_zeroed_page
);
2223 void __pagevec_free(struct pagevec
*pvec
)
2225 int i
= pagevec_count(pvec
);
2228 trace_mm_pagevec_free(pvec
->pages
[i
], pvec
->cold
);
2229 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
2233 void __free_pages(struct page
*page
, unsigned int order
)
2235 if (put_page_testzero(page
)) {
2237 free_hot_cold_page(page
, 0);
2239 __free_pages_ok(page
, order
);
2243 EXPORT_SYMBOL(__free_pages
);
2245 void free_pages(unsigned long addr
, unsigned int order
)
2248 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2249 __free_pages(virt_to_page((void *)addr
), order
);
2253 EXPORT_SYMBOL(free_pages
);
2256 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2257 * @size: the number of bytes to allocate
2258 * @gfp_mask: GFP flags for the allocation
2260 * This function is similar to alloc_pages(), except that it allocates the
2261 * minimum number of pages to satisfy the request. alloc_pages() can only
2262 * allocate memory in power-of-two pages.
2264 * This function is also limited by MAX_ORDER.
2266 * Memory allocated by this function must be released by free_pages_exact().
2268 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2270 unsigned int order
= get_order(size
);
2273 addr
= __get_free_pages(gfp_mask
, order
);
2275 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2276 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2278 split_page(virt_to_page((void *)addr
), order
);
2279 while (used
< alloc_end
) {
2285 return (void *)addr
;
2287 EXPORT_SYMBOL(alloc_pages_exact
);
2290 * free_pages_exact - release memory allocated via alloc_pages_exact()
2291 * @virt: the value returned by alloc_pages_exact.
2292 * @size: size of allocation, same value as passed to alloc_pages_exact().
2294 * Release the memory allocated by a previous call to alloc_pages_exact.
2296 void free_pages_exact(void *virt
, size_t size
)
2298 unsigned long addr
= (unsigned long)virt
;
2299 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2301 while (addr
< end
) {
2306 EXPORT_SYMBOL(free_pages_exact
);
2308 static unsigned int nr_free_zone_pages(int offset
)
2313 /* Just pick one node, since fallback list is circular */
2314 unsigned int sum
= 0;
2316 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2318 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2319 unsigned long size
= zone
->present_pages
;
2320 unsigned long high
= high_wmark_pages(zone
);
2329 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2331 unsigned int nr_free_buffer_pages(void)
2333 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2335 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2338 * Amount of free RAM allocatable within all zones
2340 unsigned int nr_free_pagecache_pages(void)
2342 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2345 static inline void show_node(struct zone
*zone
)
2348 printk("Node %d ", zone_to_nid(zone
));
2351 void si_meminfo(struct sysinfo
*val
)
2353 val
->totalram
= totalram_pages
;
2355 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2356 val
->bufferram
= nr_blockdev_pages();
2357 val
->totalhigh
= totalhigh_pages
;
2358 val
->freehigh
= nr_free_highpages();
2359 val
->mem_unit
= PAGE_SIZE
;
2362 EXPORT_SYMBOL(si_meminfo
);
2365 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2367 pg_data_t
*pgdat
= NODE_DATA(nid
);
2369 val
->totalram
= pgdat
->node_present_pages
;
2370 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2371 #ifdef CONFIG_HIGHMEM
2372 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2373 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2379 val
->mem_unit
= PAGE_SIZE
;
2383 #define K(x) ((x) << (PAGE_SHIFT-10))
2386 * Show free area list (used inside shift_scroll-lock stuff)
2387 * We also calculate the percentage fragmentation. We do this by counting the
2388 * memory on each free list with the exception of the first item on the list.
2390 void show_free_areas(void)
2395 for_each_populated_zone(zone
) {
2397 printk("%s per-cpu:\n", zone
->name
);
2399 for_each_online_cpu(cpu
) {
2400 struct per_cpu_pageset
*pageset
;
2402 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2404 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2405 cpu
, pageset
->pcp
.high
,
2406 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2410 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2411 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2413 " dirty:%lu writeback:%lu unstable:%lu\n"
2414 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2415 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2416 global_page_state(NR_ACTIVE_ANON
),
2417 global_page_state(NR_INACTIVE_ANON
),
2418 global_page_state(NR_ISOLATED_ANON
),
2419 global_page_state(NR_ACTIVE_FILE
),
2420 global_page_state(NR_INACTIVE_FILE
),
2421 global_page_state(NR_ISOLATED_FILE
),
2422 global_page_state(NR_UNEVICTABLE
),
2423 global_page_state(NR_FILE_DIRTY
),
2424 global_page_state(NR_WRITEBACK
),
2425 global_page_state(NR_UNSTABLE_NFS
),
2426 global_page_state(NR_FREE_PAGES
),
2427 global_page_state(NR_SLAB_RECLAIMABLE
),
2428 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2429 global_page_state(NR_FILE_MAPPED
),
2430 global_page_state(NR_SHMEM
),
2431 global_page_state(NR_PAGETABLE
),
2432 global_page_state(NR_BOUNCE
));
2434 for_each_populated_zone(zone
) {
2443 " active_anon:%lukB"
2444 " inactive_anon:%lukB"
2445 " active_file:%lukB"
2446 " inactive_file:%lukB"
2447 " unevictable:%lukB"
2448 " isolated(anon):%lukB"
2449 " isolated(file):%lukB"
2456 " slab_reclaimable:%lukB"
2457 " slab_unreclaimable:%lukB"
2458 " kernel_stack:%lukB"
2462 " writeback_tmp:%lukB"
2463 " pages_scanned:%lu"
2464 " all_unreclaimable? %s"
2467 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2468 K(min_wmark_pages(zone
)),
2469 K(low_wmark_pages(zone
)),
2470 K(high_wmark_pages(zone
)),
2471 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2472 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2473 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2474 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2475 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2476 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
2477 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
2478 K(zone
->present_pages
),
2479 K(zone_page_state(zone
, NR_MLOCK
)),
2480 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
2481 K(zone_page_state(zone
, NR_WRITEBACK
)),
2482 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
2483 K(zone_page_state(zone
, NR_SHMEM
)),
2484 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
2485 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
2486 zone_page_state(zone
, NR_KERNEL_STACK
) *
2488 K(zone_page_state(zone
, NR_PAGETABLE
)),
2489 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
2490 K(zone_page_state(zone
, NR_BOUNCE
)),
2491 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
2492 zone
->pages_scanned
,
2493 (zone
->all_unreclaimable
? "yes" : "no")
2495 printk("lowmem_reserve[]:");
2496 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2497 printk(" %lu", zone
->lowmem_reserve
[i
]);
2501 for_each_populated_zone(zone
) {
2502 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2505 printk("%s: ", zone
->name
);
2507 spin_lock_irqsave(&zone
->lock
, flags
);
2508 for (order
= 0; order
< MAX_ORDER
; order
++) {
2509 nr
[order
] = zone
->free_area
[order
].nr_free
;
2510 total
+= nr
[order
] << order
;
2512 spin_unlock_irqrestore(&zone
->lock
, flags
);
2513 for (order
= 0; order
< MAX_ORDER
; order
++)
2514 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2515 printk("= %lukB\n", K(total
));
2518 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2520 show_swap_cache_info();
2523 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2525 zoneref
->zone
= zone
;
2526 zoneref
->zone_idx
= zone_idx(zone
);
2530 * Builds allocation fallback zone lists.
2532 * Add all populated zones of a node to the zonelist.
2534 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2535 int nr_zones
, enum zone_type zone_type
)
2539 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2544 zone
= pgdat
->node_zones
+ zone_type
;
2545 if (populated_zone(zone
)) {
2546 zoneref_set_zone(zone
,
2547 &zonelist
->_zonerefs
[nr_zones
++]);
2548 check_highest_zone(zone_type
);
2551 } while (zone_type
);
2558 * 0 = automatic detection of better ordering.
2559 * 1 = order by ([node] distance, -zonetype)
2560 * 2 = order by (-zonetype, [node] distance)
2562 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2563 * the same zonelist. So only NUMA can configure this param.
2565 #define ZONELIST_ORDER_DEFAULT 0
2566 #define ZONELIST_ORDER_NODE 1
2567 #define ZONELIST_ORDER_ZONE 2
2569 /* zonelist order in the kernel.
2570 * set_zonelist_order() will set this to NODE or ZONE.
2572 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2573 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2577 /* The value user specified ....changed by config */
2578 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2579 /* string for sysctl */
2580 #define NUMA_ZONELIST_ORDER_LEN 16
2581 char numa_zonelist_order
[16] = "default";
2584 * interface for configure zonelist ordering.
2585 * command line option "numa_zonelist_order"
2586 * = "[dD]efault - default, automatic configuration.
2587 * = "[nN]ode - order by node locality, then by zone within node
2588 * = "[zZ]one - order by zone, then by locality within zone
2591 static int __parse_numa_zonelist_order(char *s
)
2593 if (*s
== 'd' || *s
== 'D') {
2594 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2595 } else if (*s
== 'n' || *s
== 'N') {
2596 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2597 } else if (*s
== 'z' || *s
== 'Z') {
2598 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2601 "Ignoring invalid numa_zonelist_order value: "
2608 static __init
int setup_numa_zonelist_order(char *s
)
2611 return __parse_numa_zonelist_order(s
);
2614 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2617 * sysctl handler for numa_zonelist_order
2619 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2620 void __user
*buffer
, size_t *length
,
2623 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2625 static DEFINE_MUTEX(zl_order_mutex
);
2627 mutex_lock(&zl_order_mutex
);
2629 strcpy(saved_string
, (char*)table
->data
);
2630 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
2634 int oldval
= user_zonelist_order
;
2635 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2637 * bogus value. restore saved string
2639 strncpy((char*)table
->data
, saved_string
,
2640 NUMA_ZONELIST_ORDER_LEN
);
2641 user_zonelist_order
= oldval
;
2642 } else if (oldval
!= user_zonelist_order
) {
2643 mutex_lock(&zonelists_mutex
);
2644 build_all_zonelists(NULL
);
2645 mutex_unlock(&zonelists_mutex
);
2649 mutex_unlock(&zl_order_mutex
);
2654 #define MAX_NODE_LOAD (nr_online_nodes)
2655 static int node_load
[MAX_NUMNODES
];
2658 * find_next_best_node - find the next node that should appear in a given node's fallback list
2659 * @node: node whose fallback list we're appending
2660 * @used_node_mask: nodemask_t of already used nodes
2662 * We use a number of factors to determine which is the next node that should
2663 * appear on a given node's fallback list. The node should not have appeared
2664 * already in @node's fallback list, and it should be the next closest node
2665 * according to the distance array (which contains arbitrary distance values
2666 * from each node to each node in the system), and should also prefer nodes
2667 * with no CPUs, since presumably they'll have very little allocation pressure
2668 * on them otherwise.
2669 * It returns -1 if no node is found.
2671 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2674 int min_val
= INT_MAX
;
2676 const struct cpumask
*tmp
= cpumask_of_node(0);
2678 /* Use the local node if we haven't already */
2679 if (!node_isset(node
, *used_node_mask
)) {
2680 node_set(node
, *used_node_mask
);
2684 for_each_node_state(n
, N_HIGH_MEMORY
) {
2686 /* Don't want a node to appear more than once */
2687 if (node_isset(n
, *used_node_mask
))
2690 /* Use the distance array to find the distance */
2691 val
= node_distance(node
, n
);
2693 /* Penalize nodes under us ("prefer the next node") */
2696 /* Give preference to headless and unused nodes */
2697 tmp
= cpumask_of_node(n
);
2698 if (!cpumask_empty(tmp
))
2699 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2701 /* Slight preference for less loaded node */
2702 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2703 val
+= node_load
[n
];
2705 if (val
< min_val
) {
2712 node_set(best_node
, *used_node_mask
);
2719 * Build zonelists ordered by node and zones within node.
2720 * This results in maximum locality--normal zone overflows into local
2721 * DMA zone, if any--but risks exhausting DMA zone.
2723 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2726 struct zonelist
*zonelist
;
2728 zonelist
= &pgdat
->node_zonelists
[0];
2729 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2731 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2733 zonelist
->_zonerefs
[j
].zone
= NULL
;
2734 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2738 * Build gfp_thisnode zonelists
2740 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2743 struct zonelist
*zonelist
;
2745 zonelist
= &pgdat
->node_zonelists
[1];
2746 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2747 zonelist
->_zonerefs
[j
].zone
= NULL
;
2748 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2752 * Build zonelists ordered by zone and nodes within zones.
2753 * This results in conserving DMA zone[s] until all Normal memory is
2754 * exhausted, but results in overflowing to remote node while memory
2755 * may still exist in local DMA zone.
2757 static int node_order
[MAX_NUMNODES
];
2759 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2762 int zone_type
; /* needs to be signed */
2764 struct zonelist
*zonelist
;
2766 zonelist
= &pgdat
->node_zonelists
[0];
2768 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2769 for (j
= 0; j
< nr_nodes
; j
++) {
2770 node
= node_order
[j
];
2771 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2772 if (populated_zone(z
)) {
2774 &zonelist
->_zonerefs
[pos
++]);
2775 check_highest_zone(zone_type
);
2779 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2780 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2783 static int default_zonelist_order(void)
2786 unsigned long low_kmem_size
,total_size
;
2790 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2791 * If they are really small and used heavily, the system can fall
2792 * into OOM very easily.
2793 * This function detect ZONE_DMA/DMA32 size and configures zone order.
2795 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2798 for_each_online_node(nid
) {
2799 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2800 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2801 if (populated_zone(z
)) {
2802 if (zone_type
< ZONE_NORMAL
)
2803 low_kmem_size
+= z
->present_pages
;
2804 total_size
+= z
->present_pages
;
2805 } else if (zone_type
== ZONE_NORMAL
) {
2807 * If any node has only lowmem, then node order
2808 * is preferred to allow kernel allocations
2809 * locally; otherwise, they can easily infringe
2810 * on other nodes when there is an abundance of
2811 * lowmem available to allocate from.
2813 return ZONELIST_ORDER_NODE
;
2817 if (!low_kmem_size
|| /* there are no DMA area. */
2818 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2819 return ZONELIST_ORDER_NODE
;
2821 * look into each node's config.
2822 * If there is a node whose DMA/DMA32 memory is very big area on
2823 * local memory, NODE_ORDER may be suitable.
2825 average_size
= total_size
/
2826 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2827 for_each_online_node(nid
) {
2830 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2831 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2832 if (populated_zone(z
)) {
2833 if (zone_type
< ZONE_NORMAL
)
2834 low_kmem_size
+= z
->present_pages
;
2835 total_size
+= z
->present_pages
;
2838 if (low_kmem_size
&&
2839 total_size
> average_size
&& /* ignore small node */
2840 low_kmem_size
> total_size
* 70/100)
2841 return ZONELIST_ORDER_NODE
;
2843 return ZONELIST_ORDER_ZONE
;
2846 static void set_zonelist_order(void)
2848 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2849 current_zonelist_order
= default_zonelist_order();
2851 current_zonelist_order
= user_zonelist_order
;
2854 static void build_zonelists(pg_data_t
*pgdat
)
2858 nodemask_t used_mask
;
2859 int local_node
, prev_node
;
2860 struct zonelist
*zonelist
;
2861 int order
= current_zonelist_order
;
2863 /* initialize zonelists */
2864 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2865 zonelist
= pgdat
->node_zonelists
+ i
;
2866 zonelist
->_zonerefs
[0].zone
= NULL
;
2867 zonelist
->_zonerefs
[0].zone_idx
= 0;
2870 /* NUMA-aware ordering of nodes */
2871 local_node
= pgdat
->node_id
;
2872 load
= nr_online_nodes
;
2873 prev_node
= local_node
;
2874 nodes_clear(used_mask
);
2876 memset(node_order
, 0, sizeof(node_order
));
2879 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2880 int distance
= node_distance(local_node
, node
);
2883 * If another node is sufficiently far away then it is better
2884 * to reclaim pages in a zone before going off node.
2886 if (distance
> RECLAIM_DISTANCE
)
2887 zone_reclaim_mode
= 1;
2890 * We don't want to pressure a particular node.
2891 * So adding penalty to the first node in same
2892 * distance group to make it round-robin.
2894 if (distance
!= node_distance(local_node
, prev_node
))
2895 node_load
[node
] = load
;
2899 if (order
== ZONELIST_ORDER_NODE
)
2900 build_zonelists_in_node_order(pgdat
, node
);
2902 node_order
[j
++] = node
; /* remember order */
2905 if (order
== ZONELIST_ORDER_ZONE
) {
2906 /* calculate node order -- i.e., DMA last! */
2907 build_zonelists_in_zone_order(pgdat
, j
);
2910 build_thisnode_zonelists(pgdat
);
2913 /* Construct the zonelist performance cache - see further mmzone.h */
2914 static void build_zonelist_cache(pg_data_t
*pgdat
)
2916 struct zonelist
*zonelist
;
2917 struct zonelist_cache
*zlc
;
2920 zonelist
= &pgdat
->node_zonelists
[0];
2921 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2922 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2923 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2924 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2927 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
2929 * Return node id of node used for "local" allocations.
2930 * I.e., first node id of first zone in arg node's generic zonelist.
2931 * Used for initializing percpu 'numa_mem', which is used primarily
2932 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
2934 int local_memory_node(int node
)
2938 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
2939 gfp_zone(GFP_KERNEL
),
2946 #else /* CONFIG_NUMA */
2948 static void set_zonelist_order(void)
2950 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2953 static void build_zonelists(pg_data_t
*pgdat
)
2955 int node
, local_node
;
2957 struct zonelist
*zonelist
;
2959 local_node
= pgdat
->node_id
;
2961 zonelist
= &pgdat
->node_zonelists
[0];
2962 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2965 * Now we build the zonelist so that it contains the zones
2966 * of all the other nodes.
2967 * We don't want to pressure a particular node, so when
2968 * building the zones for node N, we make sure that the
2969 * zones coming right after the local ones are those from
2970 * node N+1 (modulo N)
2972 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2973 if (!node_online(node
))
2975 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2978 for (node
= 0; node
< local_node
; node
++) {
2979 if (!node_online(node
))
2981 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2985 zonelist
->_zonerefs
[j
].zone
= NULL
;
2986 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2989 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2990 static void build_zonelist_cache(pg_data_t
*pgdat
)
2992 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2995 #endif /* CONFIG_NUMA */
2998 * Boot pageset table. One per cpu which is going to be used for all
2999 * zones and all nodes. The parameters will be set in such a way
3000 * that an item put on a list will immediately be handed over to
3001 * the buddy list. This is safe since pageset manipulation is done
3002 * with interrupts disabled.
3004 * The boot_pagesets must be kept even after bootup is complete for
3005 * unused processors and/or zones. They do play a role for bootstrapping
3006 * hotplugged processors.
3008 * zoneinfo_show() and maybe other functions do
3009 * not check if the processor is online before following the pageset pointer.
3010 * Other parts of the kernel may not check if the zone is available.
3012 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3013 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3014 static void setup_zone_pageset(struct zone
*zone
);
3017 * Global mutex to protect against size modification of zonelists
3018 * as well as to serialize pageset setup for the new populated zone.
3020 DEFINE_MUTEX(zonelists_mutex
);
3022 /* return values int ....just for stop_machine() */
3023 static __init_refok
int __build_all_zonelists(void *data
)
3029 memset(node_load
, 0, sizeof(node_load
));
3031 for_each_online_node(nid
) {
3032 pg_data_t
*pgdat
= NODE_DATA(nid
);
3034 build_zonelists(pgdat
);
3035 build_zonelist_cache(pgdat
);
3038 #ifdef CONFIG_MEMORY_HOTPLUG
3039 /* Setup real pagesets for the new zone */
3041 struct zone
*zone
= data
;
3042 setup_zone_pageset(zone
);
3047 * Initialize the boot_pagesets that are going to be used
3048 * for bootstrapping processors. The real pagesets for
3049 * each zone will be allocated later when the per cpu
3050 * allocator is available.
3052 * boot_pagesets are used also for bootstrapping offline
3053 * cpus if the system is already booted because the pagesets
3054 * are needed to initialize allocators on a specific cpu too.
3055 * F.e. the percpu allocator needs the page allocator which
3056 * needs the percpu allocator in order to allocate its pagesets
3057 * (a chicken-egg dilemma).
3059 for_each_possible_cpu(cpu
) {
3060 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3062 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3064 * We now know the "local memory node" for each node--
3065 * i.e., the node of the first zone in the generic zonelist.
3066 * Set up numa_mem percpu variable for on-line cpus. During
3067 * boot, only the boot cpu should be on-line; we'll init the
3068 * secondary cpus' numa_mem as they come on-line. During
3069 * node/memory hotplug, we'll fixup all on-line cpus.
3071 if (cpu_online(cpu
))
3072 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3080 * Called with zonelists_mutex held always
3081 * unless system_state == SYSTEM_BOOTING.
3083 void build_all_zonelists(void *data
)
3085 set_zonelist_order();
3087 if (system_state
== SYSTEM_BOOTING
) {
3088 __build_all_zonelists(NULL
);
3089 mminit_verify_zonelist();
3090 cpuset_init_current_mems_allowed();
3092 /* we have to stop all cpus to guarantee there is no user
3094 stop_machine(__build_all_zonelists
, data
, NULL
);
3095 /* cpuset refresh routine should be here */
3097 vm_total_pages
= nr_free_pagecache_pages();
3099 * Disable grouping by mobility if the number of pages in the
3100 * system is too low to allow the mechanism to work. It would be
3101 * more accurate, but expensive to check per-zone. This check is
3102 * made on memory-hotadd so a system can start with mobility
3103 * disabled and enable it later
3105 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3106 page_group_by_mobility_disabled
= 1;
3108 page_group_by_mobility_disabled
= 0;
3110 printk("Built %i zonelists in %s order, mobility grouping %s. "
3111 "Total pages: %ld\n",
3113 zonelist_order_name
[current_zonelist_order
],
3114 page_group_by_mobility_disabled
? "off" : "on",
3117 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3122 * Helper functions to size the waitqueue hash table.
3123 * Essentially these want to choose hash table sizes sufficiently
3124 * large so that collisions trying to wait on pages are rare.
3125 * But in fact, the number of active page waitqueues on typical
3126 * systems is ridiculously low, less than 200. So this is even
3127 * conservative, even though it seems large.
3129 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3130 * waitqueues, i.e. the size of the waitq table given the number of pages.
3132 #define PAGES_PER_WAITQUEUE 256
3134 #ifndef CONFIG_MEMORY_HOTPLUG
3135 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3137 unsigned long size
= 1;
3139 pages
/= PAGES_PER_WAITQUEUE
;
3141 while (size
< pages
)
3145 * Once we have dozens or even hundreds of threads sleeping
3146 * on IO we've got bigger problems than wait queue collision.
3147 * Limit the size of the wait table to a reasonable size.
3149 size
= min(size
, 4096UL);
3151 return max(size
, 4UL);
3155 * A zone's size might be changed by hot-add, so it is not possible to determine
3156 * a suitable size for its wait_table. So we use the maximum size now.
3158 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3160 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3161 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3162 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3164 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3165 * or more by the traditional way. (See above). It equals:
3167 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3168 * ia64(16K page size) : = ( 8G + 4M)byte.
3169 * powerpc (64K page size) : = (32G +16M)byte.
3171 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3178 * This is an integer logarithm so that shifts can be used later
3179 * to extract the more random high bits from the multiplicative
3180 * hash function before the remainder is taken.
3182 static inline unsigned long wait_table_bits(unsigned long size
)
3187 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3190 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3191 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3192 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3193 * higher will lead to a bigger reserve which will get freed as contiguous
3194 * blocks as reclaim kicks in
3196 static void setup_zone_migrate_reserve(struct zone
*zone
)
3198 unsigned long start_pfn
, pfn
, end_pfn
;
3200 unsigned long block_migratetype
;
3203 /* Get the start pfn, end pfn and the number of blocks to reserve */
3204 start_pfn
= zone
->zone_start_pfn
;
3205 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3206 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3210 * Reserve blocks are generally in place to help high-order atomic
3211 * allocations that are short-lived. A min_free_kbytes value that
3212 * would result in more than 2 reserve blocks for atomic allocations
3213 * is assumed to be in place to help anti-fragmentation for the
3214 * future allocation of hugepages at runtime.
3216 reserve
= min(2, reserve
);
3218 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3219 if (!pfn_valid(pfn
))
3221 page
= pfn_to_page(pfn
);
3223 /* Watch out for overlapping nodes */
3224 if (page_to_nid(page
) != zone_to_nid(zone
))
3227 /* Blocks with reserved pages will never free, skip them. */
3228 if (PageReserved(page
))
3231 block_migratetype
= get_pageblock_migratetype(page
);
3233 /* If this block is reserved, account for it */
3234 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
3239 /* Suitable for reserving if this block is movable */
3240 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
3241 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
3242 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
3248 * If the reserve is met and this is a previous reserved block,
3251 if (block_migratetype
== MIGRATE_RESERVE
) {
3252 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3253 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3259 * Initially all pages are reserved - free ones are freed
3260 * up by free_all_bootmem() once the early boot process is
3261 * done. Non-atomic initialization, single-pass.
3263 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3264 unsigned long start_pfn
, enum memmap_context context
)
3267 unsigned long end_pfn
= start_pfn
+ size
;
3271 if (highest_memmap_pfn
< end_pfn
- 1)
3272 highest_memmap_pfn
= end_pfn
- 1;
3274 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3275 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3277 * There can be holes in boot-time mem_map[]s
3278 * handed to this function. They do not
3279 * exist on hotplugged memory.
3281 if (context
== MEMMAP_EARLY
) {
3282 if (!early_pfn_valid(pfn
))
3284 if (!early_pfn_in_nid(pfn
, nid
))
3287 page
= pfn_to_page(pfn
);
3288 set_page_links(page
, zone
, nid
, pfn
);
3289 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3290 init_page_count(page
);
3291 reset_page_mapcount(page
);
3292 SetPageReserved(page
);
3294 * Mark the block movable so that blocks are reserved for
3295 * movable at startup. This will force kernel allocations
3296 * to reserve their blocks rather than leaking throughout
3297 * the address space during boot when many long-lived
3298 * kernel allocations are made. Later some blocks near
3299 * the start are marked MIGRATE_RESERVE by
3300 * setup_zone_migrate_reserve()
3302 * bitmap is created for zone's valid pfn range. but memmap
3303 * can be created for invalid pages (for alignment)
3304 * check here not to call set_pageblock_migratetype() against
3307 if ((z
->zone_start_pfn
<= pfn
)
3308 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3309 && !(pfn
& (pageblock_nr_pages
- 1)))
3310 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3312 INIT_LIST_HEAD(&page
->lru
);
3313 #ifdef WANT_PAGE_VIRTUAL
3314 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3315 if (!is_highmem_idx(zone
))
3316 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3321 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3324 for_each_migratetype_order(order
, t
) {
3325 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3326 zone
->free_area
[order
].nr_free
= 0;
3330 #ifndef __HAVE_ARCH_MEMMAP_INIT
3331 #define memmap_init(size, nid, zone, start_pfn) \
3332 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3335 static int zone_batchsize(struct zone
*zone
)
3341 * The per-cpu-pages pools are set to around 1000th of the
3342 * size of the zone. But no more than 1/2 of a meg.
3344 * OK, so we don't know how big the cache is. So guess.
3346 batch
= zone
->present_pages
/ 1024;
3347 if (batch
* PAGE_SIZE
> 512 * 1024)
3348 batch
= (512 * 1024) / PAGE_SIZE
;
3349 batch
/= 4; /* We effectively *= 4 below */
3354 * Clamp the batch to a 2^n - 1 value. Having a power
3355 * of 2 value was found to be more likely to have
3356 * suboptimal cache aliasing properties in some cases.
3358 * For example if 2 tasks are alternately allocating
3359 * batches of pages, one task can end up with a lot
3360 * of pages of one half of the possible page colors
3361 * and the other with pages of the other colors.
3363 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3368 /* The deferral and batching of frees should be suppressed under NOMMU
3371 * The problem is that NOMMU needs to be able to allocate large chunks
3372 * of contiguous memory as there's no hardware page translation to
3373 * assemble apparent contiguous memory from discontiguous pages.
3375 * Queueing large contiguous runs of pages for batching, however,
3376 * causes the pages to actually be freed in smaller chunks. As there
3377 * can be a significant delay between the individual batches being
3378 * recycled, this leads to the once large chunks of space being
3379 * fragmented and becoming unavailable for high-order allocations.
3385 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3387 struct per_cpu_pages
*pcp
;
3390 memset(p
, 0, sizeof(*p
));
3394 pcp
->high
= 6 * batch
;
3395 pcp
->batch
= max(1UL, 1 * batch
);
3396 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3397 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3401 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3402 * to the value high for the pageset p.
3405 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3408 struct per_cpu_pages
*pcp
;
3412 pcp
->batch
= max(1UL, high
/4);
3413 if ((high
/4) > (PAGE_SHIFT
* 8))
3414 pcp
->batch
= PAGE_SHIFT
* 8;
3417 static __meminit
void setup_zone_pageset(struct zone
*zone
)
3421 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
3423 for_each_possible_cpu(cpu
) {
3424 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
3426 setup_pageset(pcp
, zone_batchsize(zone
));
3428 if (percpu_pagelist_fraction
)
3429 setup_pagelist_highmark(pcp
,
3430 (zone
->present_pages
/
3431 percpu_pagelist_fraction
));
3436 * Allocate per cpu pagesets and initialize them.
3437 * Before this call only boot pagesets were available.
3439 void __init
setup_per_cpu_pageset(void)
3443 for_each_populated_zone(zone
)
3444 setup_zone_pageset(zone
);
3447 static noinline __init_refok
3448 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3451 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3455 * The per-page waitqueue mechanism uses hashed waitqueues
3458 zone
->wait_table_hash_nr_entries
=
3459 wait_table_hash_nr_entries(zone_size_pages
);
3460 zone
->wait_table_bits
=
3461 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3462 alloc_size
= zone
->wait_table_hash_nr_entries
3463 * sizeof(wait_queue_head_t
);
3465 if (!slab_is_available()) {
3466 zone
->wait_table
= (wait_queue_head_t
*)
3467 alloc_bootmem_node(pgdat
, alloc_size
);
3470 * This case means that a zone whose size was 0 gets new memory
3471 * via memory hot-add.
3472 * But it may be the case that a new node was hot-added. In
3473 * this case vmalloc() will not be able to use this new node's
3474 * memory - this wait_table must be initialized to use this new
3475 * node itself as well.
3476 * To use this new node's memory, further consideration will be
3479 zone
->wait_table
= vmalloc(alloc_size
);
3481 if (!zone
->wait_table
)
3484 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3485 init_waitqueue_head(zone
->wait_table
+ i
);
3490 static int __zone_pcp_update(void *data
)
3492 struct zone
*zone
= data
;
3494 unsigned long batch
= zone_batchsize(zone
), flags
;
3496 for_each_possible_cpu(cpu
) {
3497 struct per_cpu_pageset
*pset
;
3498 struct per_cpu_pages
*pcp
;
3500 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
3503 local_irq_save(flags
);
3504 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
3505 setup_pageset(pset
, batch
);
3506 local_irq_restore(flags
);
3511 void zone_pcp_update(struct zone
*zone
)
3513 stop_machine(__zone_pcp_update
, zone
, NULL
);
3516 static __meminit
void zone_pcp_init(struct zone
*zone
)
3519 * per cpu subsystem is not up at this point. The following code
3520 * relies on the ability of the linker to provide the
3521 * offset of a (static) per cpu variable into the per cpu area.
3523 zone
->pageset
= &boot_pageset
;
3525 if (zone
->present_pages
)
3526 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
3527 zone
->name
, zone
->present_pages
,
3528 zone_batchsize(zone
));
3531 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3532 unsigned long zone_start_pfn
,
3534 enum memmap_context context
)
3536 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3538 ret
= zone_wait_table_init(zone
, size
);
3541 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3543 zone
->zone_start_pfn
= zone_start_pfn
;
3545 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3546 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3548 (unsigned long)zone_idx(zone
),
3549 zone_start_pfn
, (zone_start_pfn
+ size
));
3551 zone_init_free_lists(zone
);
3556 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3558 * Basic iterator support. Return the first range of PFNs for a node
3559 * Note: nid == MAX_NUMNODES returns first region regardless of node
3561 static int __meminit
first_active_region_index_in_nid(int nid
)
3565 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3566 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3573 * Basic iterator support. Return the next active range of PFNs for a node
3574 * Note: nid == MAX_NUMNODES returns next region regardless of node
3576 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3578 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3579 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3585 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3587 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3588 * Architectures may implement their own version but if add_active_range()
3589 * was used and there are no special requirements, this is a convenient
3592 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3596 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3597 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3598 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3600 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3601 return early_node_map
[i
].nid
;
3603 /* This is a memory hole */
3606 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3608 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3612 nid
= __early_pfn_to_nid(pfn
);
3615 /* just returns 0 */
3619 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3620 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3624 nid
= __early_pfn_to_nid(pfn
);
3625 if (nid
>= 0 && nid
!= node
)
3631 /* Basic iterator support to walk early_node_map[] */
3632 #define for_each_active_range_index_in_nid(i, nid) \
3633 for (i = first_active_region_index_in_nid(nid); i != -1; \
3634 i = next_active_region_index_in_nid(i, nid))
3637 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3638 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3639 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3641 * If an architecture guarantees that all ranges registered with
3642 * add_active_ranges() contain no holes and may be freed, this
3643 * this function may be used instead of calling free_bootmem() manually.
3645 void __init
free_bootmem_with_active_regions(int nid
,
3646 unsigned long max_low_pfn
)
3650 for_each_active_range_index_in_nid(i
, nid
) {
3651 unsigned long size_pages
= 0;
3652 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3654 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3657 if (end_pfn
> max_low_pfn
)
3658 end_pfn
= max_low_pfn
;
3660 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3661 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3662 PFN_PHYS(early_node_map
[i
].start_pfn
),
3663 size_pages
<< PAGE_SHIFT
);
3667 int __init
add_from_early_node_map(struct range
*range
, int az
,
3668 int nr_range
, int nid
)
3673 /* need to go over early_node_map to find out good range for node */
3674 for_each_active_range_index_in_nid(i
, nid
) {
3675 start
= early_node_map
[i
].start_pfn
;
3676 end
= early_node_map
[i
].end_pfn
;
3677 nr_range
= add_range(range
, az
, nr_range
, start
, end
);
3682 #ifdef CONFIG_NO_BOOTMEM
3683 void * __init
__alloc_memory_core_early(int nid
, u64 size
, u64 align
,
3684 u64 goal
, u64 limit
)
3689 if (limit
> get_max_mapped())
3690 limit
= get_max_mapped();
3692 /* need to go over early_node_map to find out good range for node */
3693 for_each_active_range_index_in_nid(i
, nid
) {
3695 u64 ei_start
, ei_last
;
3697 ei_last
= early_node_map
[i
].end_pfn
;
3698 ei_last
<<= PAGE_SHIFT
;
3699 ei_start
= early_node_map
[i
].start_pfn
;
3700 ei_start
<<= PAGE_SHIFT
;
3701 addr
= find_early_area(ei_start
, ei_last
,
3702 goal
, limit
, size
, align
);
3708 ptr
= phys_to_virt(addr
);
3709 memset(ptr
, 0, size
);
3710 reserve_early_without_check(addr
, addr
+ size
, "BOOTMEM");
3712 * The min_count is set to 0 so that bootmem allocated blocks
3713 * are never reported as leaks.
3715 kmemleak_alloc(ptr
, size
, 0, 0);
3724 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3729 for_each_active_range_index_in_nid(i
, nid
) {
3730 ret
= work_fn(early_node_map
[i
].start_pfn
,
3731 early_node_map
[i
].end_pfn
, data
);
3737 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3738 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3740 * If an architecture guarantees that all ranges registered with
3741 * add_active_ranges() contain no holes and may be freed, this
3742 * function may be used instead of calling memory_present() manually.
3744 void __init
sparse_memory_present_with_active_regions(int nid
)
3748 for_each_active_range_index_in_nid(i
, nid
)
3749 memory_present(early_node_map
[i
].nid
,
3750 early_node_map
[i
].start_pfn
,
3751 early_node_map
[i
].end_pfn
);
3755 * get_pfn_range_for_nid - Return the start and end page frames for a node
3756 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3757 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3758 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3760 * It returns the start and end page frame of a node based on information
3761 * provided by an arch calling add_active_range(). If called for a node
3762 * with no available memory, a warning is printed and the start and end
3765 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3766 unsigned long *start_pfn
, unsigned long *end_pfn
)
3772 for_each_active_range_index_in_nid(i
, nid
) {
3773 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3774 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3777 if (*start_pfn
== -1UL)
3782 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3783 * assumption is made that zones within a node are ordered in monotonic
3784 * increasing memory addresses so that the "highest" populated zone is used
3786 static void __init
find_usable_zone_for_movable(void)
3789 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3790 if (zone_index
== ZONE_MOVABLE
)
3793 if (arch_zone_highest_possible_pfn
[zone_index
] >
3794 arch_zone_lowest_possible_pfn
[zone_index
])
3798 VM_BUG_ON(zone_index
== -1);
3799 movable_zone
= zone_index
;
3803 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3804 * because it is sized independant of architecture. Unlike the other zones,
3805 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3806 * in each node depending on the size of each node and how evenly kernelcore
3807 * is distributed. This helper function adjusts the zone ranges
3808 * provided by the architecture for a given node by using the end of the
3809 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3810 * zones within a node are in order of monotonic increases memory addresses
3812 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3813 unsigned long zone_type
,
3814 unsigned long node_start_pfn
,
3815 unsigned long node_end_pfn
,
3816 unsigned long *zone_start_pfn
,
3817 unsigned long *zone_end_pfn
)
3819 /* Only adjust if ZONE_MOVABLE is on this node */
3820 if (zone_movable_pfn
[nid
]) {
3821 /* Size ZONE_MOVABLE */
3822 if (zone_type
== ZONE_MOVABLE
) {
3823 *zone_start_pfn
= zone_movable_pfn
[nid
];
3824 *zone_end_pfn
= min(node_end_pfn
,
3825 arch_zone_highest_possible_pfn
[movable_zone
]);
3827 /* Adjust for ZONE_MOVABLE starting within this range */
3828 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3829 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3830 *zone_end_pfn
= zone_movable_pfn
[nid
];
3832 /* Check if this whole range is within ZONE_MOVABLE */
3833 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3834 *zone_start_pfn
= *zone_end_pfn
;
3839 * Return the number of pages a zone spans in a node, including holes
3840 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3842 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3843 unsigned long zone_type
,
3844 unsigned long *ignored
)
3846 unsigned long node_start_pfn
, node_end_pfn
;
3847 unsigned long zone_start_pfn
, zone_end_pfn
;
3849 /* Get the start and end of the node and zone */
3850 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3851 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3852 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3853 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3854 node_start_pfn
, node_end_pfn
,
3855 &zone_start_pfn
, &zone_end_pfn
);
3857 /* Check that this node has pages within the zone's required range */
3858 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3861 /* Move the zone boundaries inside the node if necessary */
3862 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3863 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3865 /* Return the spanned pages */
3866 return zone_end_pfn
- zone_start_pfn
;
3870 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3871 * then all holes in the requested range will be accounted for.
3873 unsigned long __meminit
__absent_pages_in_range(int nid
,
3874 unsigned long range_start_pfn
,
3875 unsigned long range_end_pfn
)
3878 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3879 unsigned long start_pfn
;
3881 /* Find the end_pfn of the first active range of pfns in the node */
3882 i
= first_active_region_index_in_nid(nid
);
3886 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3888 /* Account for ranges before physical memory on this node */
3889 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3890 hole_pages
= prev_end_pfn
- range_start_pfn
;
3892 /* Find all holes for the zone within the node */
3893 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3895 /* No need to continue if prev_end_pfn is outside the zone */
3896 if (prev_end_pfn
>= range_end_pfn
)
3899 /* Make sure the end of the zone is not within the hole */
3900 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3901 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3903 /* Update the hole size cound and move on */
3904 if (start_pfn
> range_start_pfn
) {
3905 BUG_ON(prev_end_pfn
> start_pfn
);
3906 hole_pages
+= start_pfn
- prev_end_pfn
;
3908 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3911 /* Account for ranges past physical memory on this node */
3912 if (range_end_pfn
> prev_end_pfn
)
3913 hole_pages
+= range_end_pfn
-
3914 max(range_start_pfn
, prev_end_pfn
);
3920 * absent_pages_in_range - Return number of page frames in holes within a range
3921 * @start_pfn: The start PFN to start searching for holes
3922 * @end_pfn: The end PFN to stop searching for holes
3924 * It returns the number of pages frames in memory holes within a range.
3926 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3927 unsigned long end_pfn
)
3929 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3932 /* Return the number of page frames in holes in a zone on a node */
3933 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3934 unsigned long zone_type
,
3935 unsigned long *ignored
)
3937 unsigned long node_start_pfn
, node_end_pfn
;
3938 unsigned long zone_start_pfn
, zone_end_pfn
;
3940 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3941 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3943 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3946 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3947 node_start_pfn
, node_end_pfn
,
3948 &zone_start_pfn
, &zone_end_pfn
);
3949 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3953 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3954 unsigned long zone_type
,
3955 unsigned long *zones_size
)
3957 return zones_size
[zone_type
];
3960 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3961 unsigned long zone_type
,
3962 unsigned long *zholes_size
)
3967 return zholes_size
[zone_type
];
3972 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3973 unsigned long *zones_size
, unsigned long *zholes_size
)
3975 unsigned long realtotalpages
, totalpages
= 0;
3978 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3979 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3981 pgdat
->node_spanned_pages
= totalpages
;
3983 realtotalpages
= totalpages
;
3984 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3986 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3988 pgdat
->node_present_pages
= realtotalpages
;
3989 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3993 #ifndef CONFIG_SPARSEMEM
3995 * Calculate the size of the zone->blockflags rounded to an unsigned long
3996 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3997 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3998 * round what is now in bits to nearest long in bits, then return it in
4001 static unsigned long __init
usemap_size(unsigned long zonesize
)
4003 unsigned long usemapsize
;
4005 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4006 usemapsize
= usemapsize
>> pageblock_order
;
4007 usemapsize
*= NR_PAGEBLOCK_BITS
;
4008 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4010 return usemapsize
/ 8;
4013 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4014 struct zone
*zone
, unsigned long zonesize
)
4016 unsigned long usemapsize
= usemap_size(zonesize
);
4017 zone
->pageblock_flags
= NULL
;
4019 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
4022 static void inline setup_usemap(struct pglist_data
*pgdat
,
4023 struct zone
*zone
, unsigned long zonesize
) {}
4024 #endif /* CONFIG_SPARSEMEM */
4026 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4028 /* Return a sensible default order for the pageblock size. */
4029 static inline int pageblock_default_order(void)
4031 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4032 return HUGETLB_PAGE_ORDER
;
4037 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4038 static inline void __init
set_pageblock_order(unsigned int order
)
4040 /* Check that pageblock_nr_pages has not already been setup */
4041 if (pageblock_order
)
4045 * Assume the largest contiguous order of interest is a huge page.
4046 * This value may be variable depending on boot parameters on IA64
4048 pageblock_order
= order
;
4050 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4053 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4054 * and pageblock_default_order() are unused as pageblock_order is set
4055 * at compile-time. See include/linux/pageblock-flags.h for the values of
4056 * pageblock_order based on the kernel config
4058 static inline int pageblock_default_order(unsigned int order
)
4062 #define set_pageblock_order(x) do {} while (0)
4064 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4067 * Set up the zone data structures:
4068 * - mark all pages reserved
4069 * - mark all memory queues empty
4070 * - clear the memory bitmaps
4072 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4073 unsigned long *zones_size
, unsigned long *zholes_size
)
4076 int nid
= pgdat
->node_id
;
4077 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4080 pgdat_resize_init(pgdat
);
4081 pgdat
->nr_zones
= 0;
4082 init_waitqueue_head(&pgdat
->kswapd_wait
);
4083 pgdat
->kswapd_max_order
= 0;
4084 pgdat_page_cgroup_init(pgdat
);
4086 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4087 struct zone
*zone
= pgdat
->node_zones
+ j
;
4088 unsigned long size
, realsize
, memmap_pages
;
4091 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4092 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
4096 * Adjust realsize so that it accounts for how much memory
4097 * is used by this zone for memmap. This affects the watermark
4098 * and per-cpu initialisations
4101 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
4102 if (realsize
>= memmap_pages
) {
4103 realsize
-= memmap_pages
;
4106 " %s zone: %lu pages used for memmap\n",
4107 zone_names
[j
], memmap_pages
);
4110 " %s zone: %lu pages exceeds realsize %lu\n",
4111 zone_names
[j
], memmap_pages
, realsize
);
4113 /* Account for reserved pages */
4114 if (j
== 0 && realsize
> dma_reserve
) {
4115 realsize
-= dma_reserve
;
4116 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4117 zone_names
[0], dma_reserve
);
4120 if (!is_highmem_idx(j
))
4121 nr_kernel_pages
+= realsize
;
4122 nr_all_pages
+= realsize
;
4124 zone
->spanned_pages
= size
;
4125 zone
->present_pages
= realsize
;
4128 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
4130 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
4132 zone
->name
= zone_names
[j
];
4133 spin_lock_init(&zone
->lock
);
4134 spin_lock_init(&zone
->lru_lock
);
4135 zone_seqlock_init(zone
);
4136 zone
->zone_pgdat
= pgdat
;
4138 zone_pcp_init(zone
);
4140 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
4141 zone
->reclaim_stat
.nr_saved_scan
[l
] = 0;
4143 zone
->reclaim_stat
.recent_rotated
[0] = 0;
4144 zone
->reclaim_stat
.recent_rotated
[1] = 0;
4145 zone
->reclaim_stat
.recent_scanned
[0] = 0;
4146 zone
->reclaim_stat
.recent_scanned
[1] = 0;
4147 zap_zone_vm_stats(zone
);
4152 set_pageblock_order(pageblock_default_order());
4153 setup_usemap(pgdat
, zone
, size
);
4154 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4155 size
, MEMMAP_EARLY
);
4157 memmap_init(size
, nid
, j
, zone_start_pfn
);
4158 zone_start_pfn
+= size
;
4162 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4164 /* Skip empty nodes */
4165 if (!pgdat
->node_spanned_pages
)
4168 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4169 /* ia64 gets its own node_mem_map, before this, without bootmem */
4170 if (!pgdat
->node_mem_map
) {
4171 unsigned long size
, start
, end
;
4175 * The zone's endpoints aren't required to be MAX_ORDER
4176 * aligned but the node_mem_map endpoints must be in order
4177 * for the buddy allocator to function correctly.
4179 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4180 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4181 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4182 size
= (end
- start
) * sizeof(struct page
);
4183 map
= alloc_remap(pgdat
->node_id
, size
);
4185 map
= alloc_bootmem_node(pgdat
, size
);
4186 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4188 #ifndef CONFIG_NEED_MULTIPLE_NODES
4190 * With no DISCONTIG, the global mem_map is just set as node 0's
4192 if (pgdat
== NODE_DATA(0)) {
4193 mem_map
= NODE_DATA(0)->node_mem_map
;
4194 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4195 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4196 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4197 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4200 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4203 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4204 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4206 pg_data_t
*pgdat
= NODE_DATA(nid
);
4208 pgdat
->node_id
= nid
;
4209 pgdat
->node_start_pfn
= node_start_pfn
;
4210 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4212 alloc_node_mem_map(pgdat
);
4213 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4214 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4215 nid
, (unsigned long)pgdat
,
4216 (unsigned long)pgdat
->node_mem_map
);
4219 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4222 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
4224 #if MAX_NUMNODES > 1
4226 * Figure out the number of possible node ids.
4228 static void __init
setup_nr_node_ids(void)
4231 unsigned int highest
= 0;
4233 for_each_node_mask(node
, node_possible_map
)
4235 nr_node_ids
= highest
+ 1;
4238 static inline void setup_nr_node_ids(void)
4244 * add_active_range - Register a range of PFNs backed by physical memory
4245 * @nid: The node ID the range resides on
4246 * @start_pfn: The start PFN of the available physical memory
4247 * @end_pfn: The end PFN of the available physical memory
4249 * These ranges are stored in an early_node_map[] and later used by
4250 * free_area_init_nodes() to calculate zone sizes and holes. If the
4251 * range spans a memory hole, it is up to the architecture to ensure
4252 * the memory is not freed by the bootmem allocator. If possible
4253 * the range being registered will be merged with existing ranges.
4255 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
4256 unsigned long end_pfn
)
4260 mminit_dprintk(MMINIT_TRACE
, "memory_register",
4261 "Entering add_active_range(%d, %#lx, %#lx) "
4262 "%d entries of %d used\n",
4263 nid
, start_pfn
, end_pfn
,
4264 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
4266 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
4268 /* Merge with existing active regions if possible */
4269 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4270 if (early_node_map
[i
].nid
!= nid
)
4273 /* Skip if an existing region covers this new one */
4274 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
4275 end_pfn
<= early_node_map
[i
].end_pfn
)
4278 /* Merge forward if suitable */
4279 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
4280 end_pfn
> early_node_map
[i
].end_pfn
) {
4281 early_node_map
[i
].end_pfn
= end_pfn
;
4285 /* Merge backward if suitable */
4286 if (start_pfn
< early_node_map
[i
].start_pfn
&&
4287 end_pfn
>= early_node_map
[i
].start_pfn
) {
4288 early_node_map
[i
].start_pfn
= start_pfn
;
4293 /* Check that early_node_map is large enough */
4294 if (i
>= MAX_ACTIVE_REGIONS
) {
4295 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
4296 MAX_ACTIVE_REGIONS
);
4300 early_node_map
[i
].nid
= nid
;
4301 early_node_map
[i
].start_pfn
= start_pfn
;
4302 early_node_map
[i
].end_pfn
= end_pfn
;
4303 nr_nodemap_entries
= i
+ 1;
4307 * remove_active_range - Shrink an existing registered range of PFNs
4308 * @nid: The node id the range is on that should be shrunk
4309 * @start_pfn: The new PFN of the range
4310 * @end_pfn: The new PFN of the range
4312 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
4313 * The map is kept near the end physical page range that has already been
4314 * registered. This function allows an arch to shrink an existing registered
4317 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
4318 unsigned long end_pfn
)
4323 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
4324 nid
, start_pfn
, end_pfn
);
4326 /* Find the old active region end and shrink */
4327 for_each_active_range_index_in_nid(i
, nid
) {
4328 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4329 early_node_map
[i
].end_pfn
<= end_pfn
) {
4331 early_node_map
[i
].start_pfn
= 0;
4332 early_node_map
[i
].end_pfn
= 0;
4336 if (early_node_map
[i
].start_pfn
< start_pfn
&&
4337 early_node_map
[i
].end_pfn
> start_pfn
) {
4338 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
4339 early_node_map
[i
].end_pfn
= start_pfn
;
4340 if (temp_end_pfn
> end_pfn
)
4341 add_active_range(nid
, end_pfn
, temp_end_pfn
);
4344 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
4345 early_node_map
[i
].end_pfn
> end_pfn
&&
4346 early_node_map
[i
].start_pfn
< end_pfn
) {
4347 early_node_map
[i
].start_pfn
= end_pfn
;
4355 /* remove the blank ones */
4356 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
4357 if (early_node_map
[i
].nid
!= nid
)
4359 if (early_node_map
[i
].end_pfn
)
4361 /* we found it, get rid of it */
4362 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
4363 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
4364 sizeof(early_node_map
[j
]));
4365 j
= nr_nodemap_entries
- 1;
4366 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
4367 nr_nodemap_entries
--;
4372 * remove_all_active_ranges - Remove all currently registered regions
4374 * During discovery, it may be found that a table like SRAT is invalid
4375 * and an alternative discovery method must be used. This function removes
4376 * all currently registered regions.
4378 void __init
remove_all_active_ranges(void)
4380 memset(early_node_map
, 0, sizeof(early_node_map
));
4381 nr_nodemap_entries
= 0;
4384 /* Compare two active node_active_regions */
4385 static int __init
cmp_node_active_region(const void *a
, const void *b
)
4387 struct node_active_region
*arange
= (struct node_active_region
*)a
;
4388 struct node_active_region
*brange
= (struct node_active_region
*)b
;
4390 /* Done this way to avoid overflows */
4391 if (arange
->start_pfn
> brange
->start_pfn
)
4393 if (arange
->start_pfn
< brange
->start_pfn
)
4399 /* sort the node_map by start_pfn */
4400 void __init
sort_node_map(void)
4402 sort(early_node_map
, (size_t)nr_nodemap_entries
,
4403 sizeof(struct node_active_region
),
4404 cmp_node_active_region
, NULL
);
4407 /* Find the lowest pfn for a node */
4408 static unsigned long __init
find_min_pfn_for_node(int nid
)
4411 unsigned long min_pfn
= ULONG_MAX
;
4413 /* Assuming a sorted map, the first range found has the starting pfn */
4414 for_each_active_range_index_in_nid(i
, nid
)
4415 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
4417 if (min_pfn
== ULONG_MAX
) {
4419 "Could not find start_pfn for node %d\n", nid
);
4427 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4429 * It returns the minimum PFN based on information provided via
4430 * add_active_range().
4432 unsigned long __init
find_min_pfn_with_active_regions(void)
4434 return find_min_pfn_for_node(MAX_NUMNODES
);
4438 * early_calculate_totalpages()
4439 * Sum pages in active regions for movable zone.
4440 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4442 static unsigned long __init
early_calculate_totalpages(void)
4445 unsigned long totalpages
= 0;
4447 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
4448 unsigned long pages
= early_node_map
[i
].end_pfn
-
4449 early_node_map
[i
].start_pfn
;
4450 totalpages
+= pages
;
4452 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
4458 * Find the PFN the Movable zone begins in each node. Kernel memory
4459 * is spread evenly between nodes as long as the nodes have enough
4460 * memory. When they don't, some nodes will have more kernelcore than
4463 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
4466 unsigned long usable_startpfn
;
4467 unsigned long kernelcore_node
, kernelcore_remaining
;
4468 /* save the state before borrow the nodemask */
4469 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4470 unsigned long totalpages
= early_calculate_totalpages();
4471 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4474 * If movablecore was specified, calculate what size of
4475 * kernelcore that corresponds so that memory usable for
4476 * any allocation type is evenly spread. If both kernelcore
4477 * and movablecore are specified, then the value of kernelcore
4478 * will be used for required_kernelcore if it's greater than
4479 * what movablecore would have allowed.
4481 if (required_movablecore
) {
4482 unsigned long corepages
;
4485 * Round-up so that ZONE_MOVABLE is at least as large as what
4486 * was requested by the user
4488 required_movablecore
=
4489 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4490 corepages
= totalpages
- required_movablecore
;
4492 required_kernelcore
= max(required_kernelcore
, corepages
);
4495 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4496 if (!required_kernelcore
)
4499 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4500 find_usable_zone_for_movable();
4501 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4504 /* Spread kernelcore memory as evenly as possible throughout nodes */
4505 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4506 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4508 * Recalculate kernelcore_node if the division per node
4509 * now exceeds what is necessary to satisfy the requested
4510 * amount of memory for the kernel
4512 if (required_kernelcore
< kernelcore_node
)
4513 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4516 * As the map is walked, we track how much memory is usable
4517 * by the kernel using kernelcore_remaining. When it is
4518 * 0, the rest of the node is usable by ZONE_MOVABLE
4520 kernelcore_remaining
= kernelcore_node
;
4522 /* Go through each range of PFNs within this node */
4523 for_each_active_range_index_in_nid(i
, nid
) {
4524 unsigned long start_pfn
, end_pfn
;
4525 unsigned long size_pages
;
4527 start_pfn
= max(early_node_map
[i
].start_pfn
,
4528 zone_movable_pfn
[nid
]);
4529 end_pfn
= early_node_map
[i
].end_pfn
;
4530 if (start_pfn
>= end_pfn
)
4533 /* Account for what is only usable for kernelcore */
4534 if (start_pfn
< usable_startpfn
) {
4535 unsigned long kernel_pages
;
4536 kernel_pages
= min(end_pfn
, usable_startpfn
)
4539 kernelcore_remaining
-= min(kernel_pages
,
4540 kernelcore_remaining
);
4541 required_kernelcore
-= min(kernel_pages
,
4542 required_kernelcore
);
4544 /* Continue if range is now fully accounted */
4545 if (end_pfn
<= usable_startpfn
) {
4548 * Push zone_movable_pfn to the end so
4549 * that if we have to rebalance
4550 * kernelcore across nodes, we will
4551 * not double account here
4553 zone_movable_pfn
[nid
] = end_pfn
;
4556 start_pfn
= usable_startpfn
;
4560 * The usable PFN range for ZONE_MOVABLE is from
4561 * start_pfn->end_pfn. Calculate size_pages as the
4562 * number of pages used as kernelcore
4564 size_pages
= end_pfn
- start_pfn
;
4565 if (size_pages
> kernelcore_remaining
)
4566 size_pages
= kernelcore_remaining
;
4567 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4570 * Some kernelcore has been met, update counts and
4571 * break if the kernelcore for this node has been
4574 required_kernelcore
-= min(required_kernelcore
,
4576 kernelcore_remaining
-= size_pages
;
4577 if (!kernelcore_remaining
)
4583 * If there is still required_kernelcore, we do another pass with one
4584 * less node in the count. This will push zone_movable_pfn[nid] further
4585 * along on the nodes that still have memory until kernelcore is
4589 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4592 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4593 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4594 zone_movable_pfn
[nid
] =
4595 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4598 /* restore the node_state */
4599 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4602 /* Any regular memory on that node ? */
4603 static void check_for_regular_memory(pg_data_t
*pgdat
)
4605 #ifdef CONFIG_HIGHMEM
4606 enum zone_type zone_type
;
4608 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4609 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4610 if (zone
->present_pages
)
4611 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4617 * free_area_init_nodes - Initialise all pg_data_t and zone data
4618 * @max_zone_pfn: an array of max PFNs for each zone
4620 * This will call free_area_init_node() for each active node in the system.
4621 * Using the page ranges provided by add_active_range(), the size of each
4622 * zone in each node and their holes is calculated. If the maximum PFN
4623 * between two adjacent zones match, it is assumed that the zone is empty.
4624 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4625 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4626 * starts where the previous one ended. For example, ZONE_DMA32 starts
4627 * at arch_max_dma_pfn.
4629 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4634 /* Sort early_node_map as initialisation assumes it is sorted */
4637 /* Record where the zone boundaries are */
4638 memset(arch_zone_lowest_possible_pfn
, 0,
4639 sizeof(arch_zone_lowest_possible_pfn
));
4640 memset(arch_zone_highest_possible_pfn
, 0,
4641 sizeof(arch_zone_highest_possible_pfn
));
4642 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4643 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4644 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4645 if (i
== ZONE_MOVABLE
)
4647 arch_zone_lowest_possible_pfn
[i
] =
4648 arch_zone_highest_possible_pfn
[i
-1];
4649 arch_zone_highest_possible_pfn
[i
] =
4650 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4652 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4653 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4655 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4656 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4657 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4659 /* Print out the zone ranges */
4660 printk("Zone PFN ranges:\n");
4661 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4662 if (i
== ZONE_MOVABLE
)
4664 printk(" %-8s ", zone_names
[i
]);
4665 if (arch_zone_lowest_possible_pfn
[i
] ==
4666 arch_zone_highest_possible_pfn
[i
])
4669 printk("%0#10lx -> %0#10lx\n",
4670 arch_zone_lowest_possible_pfn
[i
],
4671 arch_zone_highest_possible_pfn
[i
]);
4674 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4675 printk("Movable zone start PFN for each node\n");
4676 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4677 if (zone_movable_pfn
[i
])
4678 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4681 /* Print out the early_node_map[] */
4682 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4683 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4684 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4685 early_node_map
[i
].start_pfn
,
4686 early_node_map
[i
].end_pfn
);
4688 /* Initialise every node */
4689 mminit_verify_pageflags_layout();
4690 setup_nr_node_ids();
4691 for_each_online_node(nid
) {
4692 pg_data_t
*pgdat
= NODE_DATA(nid
);
4693 free_area_init_node(nid
, NULL
,
4694 find_min_pfn_for_node(nid
), NULL
);
4696 /* Any memory on that node */
4697 if (pgdat
->node_present_pages
)
4698 node_set_state(nid
, N_HIGH_MEMORY
);
4699 check_for_regular_memory(pgdat
);
4703 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4705 unsigned long long coremem
;
4709 coremem
= memparse(p
, &p
);
4710 *core
= coremem
>> PAGE_SHIFT
;
4712 /* Paranoid check that UL is enough for the coremem value */
4713 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4719 * kernelcore=size sets the amount of memory for use for allocations that
4720 * cannot be reclaimed or migrated.
4722 static int __init
cmdline_parse_kernelcore(char *p
)
4724 return cmdline_parse_core(p
, &required_kernelcore
);
4728 * movablecore=size sets the amount of memory for use for allocations that
4729 * can be reclaimed or migrated.
4731 static int __init
cmdline_parse_movablecore(char *p
)
4733 return cmdline_parse_core(p
, &required_movablecore
);
4736 early_param("kernelcore", cmdline_parse_kernelcore
);
4737 early_param("movablecore", cmdline_parse_movablecore
);
4739 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4742 * set_dma_reserve - set the specified number of pages reserved in the first zone
4743 * @new_dma_reserve: The number of pages to mark reserved
4745 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4746 * In the DMA zone, a significant percentage may be consumed by kernel image
4747 * and other unfreeable allocations which can skew the watermarks badly. This
4748 * function may optionally be used to account for unfreeable pages in the
4749 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4750 * smaller per-cpu batchsize.
4752 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4754 dma_reserve
= new_dma_reserve
;
4757 #ifndef CONFIG_NEED_MULTIPLE_NODES
4758 struct pglist_data __refdata contig_page_data
= {
4759 #ifndef CONFIG_NO_BOOTMEM
4760 .bdata
= &bootmem_node_data
[0]
4763 EXPORT_SYMBOL(contig_page_data
);
4766 void __init
free_area_init(unsigned long *zones_size
)
4768 free_area_init_node(0, zones_size
,
4769 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4772 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4773 unsigned long action
, void *hcpu
)
4775 int cpu
= (unsigned long)hcpu
;
4777 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4781 * Spill the event counters of the dead processor
4782 * into the current processors event counters.
4783 * This artificially elevates the count of the current
4786 vm_events_fold_cpu(cpu
);
4789 * Zero the differential counters of the dead processor
4790 * so that the vm statistics are consistent.
4792 * This is only okay since the processor is dead and cannot
4793 * race with what we are doing.
4795 refresh_cpu_vm_stats(cpu
);
4800 void __init
page_alloc_init(void)
4802 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4806 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4807 * or min_free_kbytes changes.
4809 static void calculate_totalreserve_pages(void)
4811 struct pglist_data
*pgdat
;
4812 unsigned long reserve_pages
= 0;
4813 enum zone_type i
, j
;
4815 for_each_online_pgdat(pgdat
) {
4816 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4817 struct zone
*zone
= pgdat
->node_zones
+ i
;
4818 unsigned long max
= 0;
4820 /* Find valid and maximum lowmem_reserve in the zone */
4821 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4822 if (zone
->lowmem_reserve
[j
] > max
)
4823 max
= zone
->lowmem_reserve
[j
];
4826 /* we treat the high watermark as reserved pages. */
4827 max
+= high_wmark_pages(zone
);
4829 if (max
> zone
->present_pages
)
4830 max
= zone
->present_pages
;
4831 reserve_pages
+= max
;
4834 totalreserve_pages
= reserve_pages
;
4838 * setup_per_zone_lowmem_reserve - called whenever
4839 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4840 * has a correct pages reserved value, so an adequate number of
4841 * pages are left in the zone after a successful __alloc_pages().
4843 static void setup_per_zone_lowmem_reserve(void)
4845 struct pglist_data
*pgdat
;
4846 enum zone_type j
, idx
;
4848 for_each_online_pgdat(pgdat
) {
4849 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4850 struct zone
*zone
= pgdat
->node_zones
+ j
;
4851 unsigned long present_pages
= zone
->present_pages
;
4853 zone
->lowmem_reserve
[j
] = 0;
4857 struct zone
*lower_zone
;
4861 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4862 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4864 lower_zone
= pgdat
->node_zones
+ idx
;
4865 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4866 sysctl_lowmem_reserve_ratio
[idx
];
4867 present_pages
+= lower_zone
->present_pages
;
4872 /* update totalreserve_pages */
4873 calculate_totalreserve_pages();
4877 * setup_per_zone_wmarks - called when min_free_kbytes changes
4878 * or when memory is hot-{added|removed}
4880 * Ensures that the watermark[min,low,high] values for each zone are set
4881 * correctly with respect to min_free_kbytes.
4883 void setup_per_zone_wmarks(void)
4885 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4886 unsigned long lowmem_pages
= 0;
4888 unsigned long flags
;
4890 /* Calculate total number of !ZONE_HIGHMEM pages */
4891 for_each_zone(zone
) {
4892 if (!is_highmem(zone
))
4893 lowmem_pages
+= zone
->present_pages
;
4896 for_each_zone(zone
) {
4899 spin_lock_irqsave(&zone
->lock
, flags
);
4900 tmp
= (u64
)pages_min
* zone
->present_pages
;
4901 do_div(tmp
, lowmem_pages
);
4902 if (is_highmem(zone
)) {
4904 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4905 * need highmem pages, so cap pages_min to a small
4908 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4909 * deltas controls asynch page reclaim, and so should
4910 * not be capped for highmem.
4914 min_pages
= zone
->present_pages
/ 1024;
4915 if (min_pages
< SWAP_CLUSTER_MAX
)
4916 min_pages
= SWAP_CLUSTER_MAX
;
4917 if (min_pages
> 128)
4919 zone
->watermark
[WMARK_MIN
] = min_pages
;
4922 * If it's a lowmem zone, reserve a number of pages
4923 * proportionate to the zone's size.
4925 zone
->watermark
[WMARK_MIN
] = tmp
;
4928 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
4929 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
4930 setup_zone_migrate_reserve(zone
);
4931 spin_unlock_irqrestore(&zone
->lock
, flags
);
4934 /* update totalreserve_pages */
4935 calculate_totalreserve_pages();
4939 * The inactive anon list should be small enough that the VM never has to
4940 * do too much work, but large enough that each inactive page has a chance
4941 * to be referenced again before it is swapped out.
4943 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4944 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4945 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4946 * the anonymous pages are kept on the inactive list.
4949 * memory ratio inactive anon
4950 * -------------------------------------
4959 void calculate_zone_inactive_ratio(struct zone
*zone
)
4961 unsigned int gb
, ratio
;
4963 /* Zone size in gigabytes */
4964 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4966 ratio
= int_sqrt(10 * gb
);
4970 zone
->inactive_ratio
= ratio
;
4973 static void __init
setup_per_zone_inactive_ratio(void)
4978 calculate_zone_inactive_ratio(zone
);
4982 * Initialise min_free_kbytes.
4984 * For small machines we want it small (128k min). For large machines
4985 * we want it large (64MB max). But it is not linear, because network
4986 * bandwidth does not increase linearly with machine size. We use
4988 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4989 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5005 static int __init
init_per_zone_wmark_min(void)
5007 unsigned long lowmem_kbytes
;
5009 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5011 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5012 if (min_free_kbytes
< 128)
5013 min_free_kbytes
= 128;
5014 if (min_free_kbytes
> 65536)
5015 min_free_kbytes
= 65536;
5016 setup_per_zone_wmarks();
5017 setup_per_zone_lowmem_reserve();
5018 setup_per_zone_inactive_ratio();
5021 module_init(init_per_zone_wmark_min
)
5024 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5025 * that we can call two helper functions whenever min_free_kbytes
5028 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5029 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5031 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5033 setup_per_zone_wmarks();
5038 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5039 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5044 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5049 zone
->min_unmapped_pages
= (zone
->present_pages
*
5050 sysctl_min_unmapped_ratio
) / 100;
5054 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5055 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5060 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5065 zone
->min_slab_pages
= (zone
->present_pages
*
5066 sysctl_min_slab_ratio
) / 100;
5072 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5073 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5074 * whenever sysctl_lowmem_reserve_ratio changes.
5076 * The reserve ratio obviously has absolutely no relation with the
5077 * minimum watermarks. The lowmem reserve ratio can only make sense
5078 * if in function of the boot time zone sizes.
5080 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5081 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5083 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5084 setup_per_zone_lowmem_reserve();
5089 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5090 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5091 * can have before it gets flushed back to buddy allocator.
5094 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5095 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5101 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5102 if (!write
|| (ret
== -EINVAL
))
5104 for_each_populated_zone(zone
) {
5105 for_each_possible_cpu(cpu
) {
5107 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5108 setup_pagelist_highmark(
5109 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5115 int hashdist
= HASHDIST_DEFAULT
;
5118 static int __init
set_hashdist(char *str
)
5122 hashdist
= simple_strtoul(str
, &str
, 0);
5125 __setup("hashdist=", set_hashdist
);
5129 * allocate a large system hash table from bootmem
5130 * - it is assumed that the hash table must contain an exact power-of-2
5131 * quantity of entries
5132 * - limit is the number of hash buckets, not the total allocation size
5134 void *__init
alloc_large_system_hash(const char *tablename
,
5135 unsigned long bucketsize
,
5136 unsigned long numentries
,
5139 unsigned int *_hash_shift
,
5140 unsigned int *_hash_mask
,
5141 unsigned long limit
)
5143 unsigned long long max
= limit
;
5144 unsigned long log2qty
, size
;
5147 /* allow the kernel cmdline to have a say */
5149 /* round applicable memory size up to nearest megabyte */
5150 numentries
= nr_kernel_pages
;
5151 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5152 numentries
>>= 20 - PAGE_SHIFT
;
5153 numentries
<<= 20 - PAGE_SHIFT
;
5155 /* limit to 1 bucket per 2^scale bytes of low memory */
5156 if (scale
> PAGE_SHIFT
)
5157 numentries
>>= (scale
- PAGE_SHIFT
);
5159 numentries
<<= (PAGE_SHIFT
- scale
);
5161 /* Make sure we've got at least a 0-order allocation.. */
5162 if (unlikely(flags
& HASH_SMALL
)) {
5163 /* Makes no sense without HASH_EARLY */
5164 WARN_ON(!(flags
& HASH_EARLY
));
5165 if (!(numentries
>> *_hash_shift
)) {
5166 numentries
= 1UL << *_hash_shift
;
5167 BUG_ON(!numentries
);
5169 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5170 numentries
= PAGE_SIZE
/ bucketsize
;
5172 numentries
= roundup_pow_of_two(numentries
);
5174 /* limit allocation size to 1/16 total memory by default */
5176 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5177 do_div(max
, bucketsize
);
5180 if (numentries
> max
)
5183 log2qty
= ilog2(numentries
);
5186 size
= bucketsize
<< log2qty
;
5187 if (flags
& HASH_EARLY
)
5188 table
= alloc_bootmem_nopanic(size
);
5190 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5193 * If bucketsize is not a power-of-two, we may free
5194 * some pages at the end of hash table which
5195 * alloc_pages_exact() automatically does
5197 if (get_order(size
) < MAX_ORDER
) {
5198 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5199 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5202 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5205 panic("Failed to allocate %s hash table\n", tablename
);
5207 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5210 ilog2(size
) - PAGE_SHIFT
,
5214 *_hash_shift
= log2qty
;
5216 *_hash_mask
= (1 << log2qty
) - 1;
5221 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5222 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5225 #ifdef CONFIG_SPARSEMEM
5226 return __pfn_to_section(pfn
)->pageblock_flags
;
5228 return zone
->pageblock_flags
;
5229 #endif /* CONFIG_SPARSEMEM */
5232 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5234 #ifdef CONFIG_SPARSEMEM
5235 pfn
&= (PAGES_PER_SECTION
-1);
5236 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5238 pfn
= pfn
- zone
->zone_start_pfn
;
5239 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5240 #endif /* CONFIG_SPARSEMEM */
5244 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5245 * @page: The page within the block of interest
5246 * @start_bitidx: The first bit of interest to retrieve
5247 * @end_bitidx: The last bit of interest
5248 * returns pageblock_bits flags
5250 unsigned long get_pageblock_flags_group(struct page
*page
,
5251 int start_bitidx
, int end_bitidx
)
5254 unsigned long *bitmap
;
5255 unsigned long pfn
, bitidx
;
5256 unsigned long flags
= 0;
5257 unsigned long value
= 1;
5259 zone
= page_zone(page
);
5260 pfn
= page_to_pfn(page
);
5261 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5262 bitidx
= pfn_to_bitidx(zone
, pfn
);
5264 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5265 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5272 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5273 * @page: The page within the block of interest
5274 * @start_bitidx: The first bit of interest
5275 * @end_bitidx: The last bit of interest
5276 * @flags: The flags to set
5278 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5279 int start_bitidx
, int end_bitidx
)
5282 unsigned long *bitmap
;
5283 unsigned long pfn
, bitidx
;
5284 unsigned long value
= 1;
5286 zone
= page_zone(page
);
5287 pfn
= page_to_pfn(page
);
5288 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5289 bitidx
= pfn_to_bitidx(zone
, pfn
);
5290 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5291 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5293 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5295 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5297 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5301 * This is designed as sub function...plz see page_isolation.c also.
5302 * set/clear page block's type to be ISOLATE.
5303 * page allocater never alloc memory from ISOLATE block.
5306 int set_migratetype_isolate(struct page
*page
)
5309 struct page
*curr_page
;
5310 unsigned long flags
, pfn
, iter
;
5311 unsigned long immobile
= 0;
5312 struct memory_isolate_notify arg
;
5317 zone
= page_zone(page
);
5318 zone_idx
= zone_idx(zone
);
5320 spin_lock_irqsave(&zone
->lock
, flags
);
5321 if (get_pageblock_migratetype(page
) == MIGRATE_MOVABLE
||
5322 zone_idx
== ZONE_MOVABLE
) {
5327 pfn
= page_to_pfn(page
);
5328 arg
.start_pfn
= pfn
;
5329 arg
.nr_pages
= pageblock_nr_pages
;
5330 arg
.pages_found
= 0;
5333 * It may be possible to isolate a pageblock even if the
5334 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5335 * notifier chain is used by balloon drivers to return the
5336 * number of pages in a range that are held by the balloon
5337 * driver to shrink memory. If all the pages are accounted for
5338 * by balloons, are free, or on the LRU, isolation can continue.
5339 * Later, for example, when memory hotplug notifier runs, these
5340 * pages reported as "can be isolated" should be isolated(freed)
5341 * by the balloon driver through the memory notifier chain.
5343 notifier_ret
= memory_isolate_notify(MEM_ISOLATE_COUNT
, &arg
);
5344 notifier_ret
= notifier_to_errno(notifier_ret
);
5345 if (notifier_ret
|| !arg
.pages_found
)
5348 for (iter
= pfn
; iter
< (pfn
+ pageblock_nr_pages
); iter
++) {
5349 if (!pfn_valid_within(pfn
))
5352 curr_page
= pfn_to_page(iter
);
5353 if (!page_count(curr_page
) || PageLRU(curr_page
))
5359 if (arg
.pages_found
== immobile
)
5364 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
5365 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
5368 spin_unlock_irqrestore(&zone
->lock
, flags
);
5374 void unset_migratetype_isolate(struct page
*page
)
5377 unsigned long flags
;
5378 zone
= page_zone(page
);
5379 spin_lock_irqsave(&zone
->lock
, flags
);
5380 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
5382 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5383 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
5385 spin_unlock_irqrestore(&zone
->lock
, flags
);
5388 #ifdef CONFIG_MEMORY_HOTREMOVE
5390 * All pages in the range must be isolated before calling this.
5393 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
5399 unsigned long flags
;
5400 /* find the first valid pfn */
5401 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
5406 zone
= page_zone(pfn_to_page(pfn
));
5407 spin_lock_irqsave(&zone
->lock
, flags
);
5409 while (pfn
< end_pfn
) {
5410 if (!pfn_valid(pfn
)) {
5414 page
= pfn_to_page(pfn
);
5415 BUG_ON(page_count(page
));
5416 BUG_ON(!PageBuddy(page
));
5417 order
= page_order(page
);
5418 #ifdef CONFIG_DEBUG_VM
5419 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
5420 pfn
, 1 << order
, end_pfn
);
5422 list_del(&page
->lru
);
5423 rmv_page_order(page
);
5424 zone
->free_area
[order
].nr_free
--;
5425 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
5427 for (i
= 0; i
< (1 << order
); i
++)
5428 SetPageReserved((page
+i
));
5429 pfn
+= (1 << order
);
5431 spin_unlock_irqrestore(&zone
->lock
, flags
);
5435 #ifdef CONFIG_MEMORY_FAILURE
5436 bool is_free_buddy_page(struct page
*page
)
5438 struct zone
*zone
= page_zone(page
);
5439 unsigned long pfn
= page_to_pfn(page
);
5440 unsigned long flags
;
5443 spin_lock_irqsave(&zone
->lock
, flags
);
5444 for (order
= 0; order
< MAX_ORDER
; order
++) {
5445 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
5447 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
5450 spin_unlock_irqrestore(&zone
->lock
, flags
);
5452 return order
< MAX_ORDER
;
5456 static struct trace_print_flags pageflag_names
[] = {
5457 {1UL << PG_locked
, "locked" },
5458 {1UL << PG_error
, "error" },
5459 {1UL << PG_referenced
, "referenced" },
5460 {1UL << PG_uptodate
, "uptodate" },
5461 {1UL << PG_dirty
, "dirty" },
5462 {1UL << PG_lru
, "lru" },
5463 {1UL << PG_active
, "active" },
5464 {1UL << PG_slab
, "slab" },
5465 {1UL << PG_owner_priv_1
, "owner_priv_1" },
5466 {1UL << PG_arch_1
, "arch_1" },
5467 {1UL << PG_reserved
, "reserved" },
5468 {1UL << PG_private
, "private" },
5469 {1UL << PG_private_2
, "private_2" },
5470 {1UL << PG_writeback
, "writeback" },
5471 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5472 {1UL << PG_head
, "head" },
5473 {1UL << PG_tail
, "tail" },
5475 {1UL << PG_compound
, "compound" },
5477 {1UL << PG_swapcache
, "swapcache" },
5478 {1UL << PG_mappedtodisk
, "mappedtodisk" },
5479 {1UL << PG_reclaim
, "reclaim" },
5480 {1UL << PG_buddy
, "buddy" },
5481 {1UL << PG_swapbacked
, "swapbacked" },
5482 {1UL << PG_unevictable
, "unevictable" },
5484 {1UL << PG_mlocked
, "mlocked" },
5486 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5487 {1UL << PG_uncached
, "uncached" },
5489 #ifdef CONFIG_MEMORY_FAILURE
5490 {1UL << PG_hwpoison
, "hwpoison" },
5495 static void dump_page_flags(unsigned long flags
)
5497 const char *delim
= "";
5501 printk(KERN_ALERT
"page flags: %#lx(", flags
);
5503 /* remove zone id */
5504 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
5506 for (i
= 0; pageflag_names
[i
].name
&& flags
; i
++) {
5508 mask
= pageflag_names
[i
].mask
;
5509 if ((flags
& mask
) != mask
)
5513 printk("%s%s", delim
, pageflag_names
[i
].name
);
5517 /* check for left over flags */
5519 printk("%s%#lx", delim
, flags
);
5524 void dump_page(struct page
*page
)
5527 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5528 page
, page_count(page
), page_mapcount(page
),
5529 page
->mapping
, page
->index
);
5530 dump_page_flags(page
->flags
);