2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/notifier.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/stop_machine.h>
46 #include <linux/sort.h>
47 #include <linux/pfn.h>
48 #include <linux/backing-dev.h>
49 #include <linux/fault-inject.h>
50 #include <linux/page-isolation.h>
51 #include <linux/page_cgroup.h>
52 #include <linux/debugobjects.h>
53 #include <linux/kmemleak.h>
54 #include <linux/memory.h>
55 #include <linux/compaction.h>
56 #include <trace/events/kmem.h>
57 #include <linux/ftrace_event.h>
58 #include <linux/memcontrol.h>
59 #include <linux/prefetch.h>
60 #include <linux/migrate.h>
61 #include <linux/page-debug-flags.h>
63 #include <asm/tlbflush.h>
64 #include <asm/div64.h>
67 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
68 DEFINE_PER_CPU(int, numa_node
);
69 EXPORT_PER_CPU_SYMBOL(numa_node
);
72 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
74 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
75 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
76 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
77 * defined in <linux/topology.h>.
79 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
80 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
84 * Array of node states.
86 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
87 [N_POSSIBLE
] = NODE_MASK_ALL
,
88 [N_ONLINE
] = { { [0] = 1UL } },
90 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
92 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
94 [N_CPU
] = { { [0] = 1UL } },
97 EXPORT_SYMBOL(node_states
);
99 unsigned long totalram_pages __read_mostly
;
100 unsigned long totalreserve_pages __read_mostly
;
102 * When calculating the number of globally allowed dirty pages, there
103 * is a certain number of per-zone reserves that should not be
104 * considered dirtyable memory. This is the sum of those reserves
105 * over all existing zones that contribute dirtyable memory.
107 unsigned long dirty_balance_reserve __read_mostly
;
109 int percpu_pagelist_fraction
;
110 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
112 #ifdef CONFIG_PM_SLEEP
114 * The following functions are used by the suspend/hibernate code to temporarily
115 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
116 * while devices are suspended. To avoid races with the suspend/hibernate code,
117 * they should always be called with pm_mutex held (gfp_allowed_mask also should
118 * only be modified with pm_mutex held, unless the suspend/hibernate code is
119 * guaranteed not to run in parallel with that modification).
122 static gfp_t saved_gfp_mask
;
124 void pm_restore_gfp_mask(void)
126 WARN_ON(!mutex_is_locked(&pm_mutex
));
127 if (saved_gfp_mask
) {
128 gfp_allowed_mask
= saved_gfp_mask
;
133 void pm_restrict_gfp_mask(void)
135 WARN_ON(!mutex_is_locked(&pm_mutex
));
136 WARN_ON(saved_gfp_mask
);
137 saved_gfp_mask
= gfp_allowed_mask
;
138 gfp_allowed_mask
&= ~GFP_IOFS
;
141 bool pm_suspended_storage(void)
143 if ((gfp_allowed_mask
& GFP_IOFS
) == GFP_IOFS
)
147 #endif /* CONFIG_PM_SLEEP */
149 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
150 int pageblock_order __read_mostly
;
153 static void __free_pages_ok(struct page
*page
, unsigned int order
);
156 * results with 256, 32 in the lowmem_reserve sysctl:
157 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
158 * 1G machine -> (16M dma, 784M normal, 224M high)
159 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
160 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
161 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
163 * TBD: should special case ZONE_DMA32 machines here - in those we normally
164 * don't need any ZONE_NORMAL reservation
166 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
167 #ifdef CONFIG_ZONE_DMA
170 #ifdef CONFIG_ZONE_DMA32
173 #ifdef CONFIG_HIGHMEM
179 EXPORT_SYMBOL(totalram_pages
);
181 static char * const zone_names
[MAX_NR_ZONES
] = {
182 #ifdef CONFIG_ZONE_DMA
185 #ifdef CONFIG_ZONE_DMA32
189 #ifdef CONFIG_HIGHMEM
195 int min_free_kbytes
= 1024;
197 static unsigned long __meminitdata nr_kernel_pages
;
198 static unsigned long __meminitdata nr_all_pages
;
199 static unsigned long __meminitdata dma_reserve
;
201 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
202 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
203 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
204 static unsigned long __initdata required_kernelcore
;
205 static unsigned long __initdata required_movablecore
;
206 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
208 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
210 EXPORT_SYMBOL(movable_zone
);
211 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
214 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
215 int nr_online_nodes __read_mostly
= 1;
216 EXPORT_SYMBOL(nr_node_ids
);
217 EXPORT_SYMBOL(nr_online_nodes
);
220 int page_group_by_mobility_disabled __read_mostly
;
222 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
225 if (unlikely(page_group_by_mobility_disabled
))
226 migratetype
= MIGRATE_UNMOVABLE
;
228 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
229 PB_migrate
, PB_migrate_end
);
232 bool oom_killer_disabled __read_mostly
;
234 #ifdef CONFIG_DEBUG_VM
235 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
239 unsigned long pfn
= page_to_pfn(page
);
242 seq
= zone_span_seqbegin(zone
);
243 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
245 else if (pfn
< zone
->zone_start_pfn
)
247 } while (zone_span_seqretry(zone
, seq
));
252 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
254 if (!pfn_valid_within(page_to_pfn(page
)))
256 if (zone
!= page_zone(page
))
262 * Temporary debugging check for pages not lying within a given zone.
264 static int bad_range(struct zone
*zone
, struct page
*page
)
266 if (page_outside_zone_boundaries(zone
, page
))
268 if (!page_is_consistent(zone
, page
))
274 static inline int bad_range(struct zone
*zone
, struct page
*page
)
280 static void bad_page(struct page
*page
)
282 static unsigned long resume
;
283 static unsigned long nr_shown
;
284 static unsigned long nr_unshown
;
286 /* Don't complain about poisoned pages */
287 if (PageHWPoison(page
)) {
288 reset_page_mapcount(page
); /* remove PageBuddy */
293 * Allow a burst of 60 reports, then keep quiet for that minute;
294 * or allow a steady drip of one report per second.
296 if (nr_shown
== 60) {
297 if (time_before(jiffies
, resume
)) {
303 "BUG: Bad page state: %lu messages suppressed\n",
310 resume
= jiffies
+ 60 * HZ
;
312 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
313 current
->comm
, page_to_pfn(page
));
319 /* Leave bad fields for debug, except PageBuddy could make trouble */
320 reset_page_mapcount(page
); /* remove PageBuddy */
321 add_taint(TAINT_BAD_PAGE
);
325 * Higher-order pages are called "compound pages". They are structured thusly:
327 * The first PAGE_SIZE page is called the "head page".
329 * The remaining PAGE_SIZE pages are called "tail pages".
331 * All pages have PG_compound set. All tail pages have their ->first_page
332 * pointing at the head page.
334 * The first tail page's ->lru.next holds the address of the compound page's
335 * put_page() function. Its ->lru.prev holds the order of allocation.
336 * This usage means that zero-order pages may not be compound.
339 static void free_compound_page(struct page
*page
)
341 __free_pages_ok(page
, compound_order(page
));
344 void prep_compound_page(struct page
*page
, unsigned long order
)
347 int nr_pages
= 1 << order
;
349 set_compound_page_dtor(page
, free_compound_page
);
350 set_compound_order(page
, order
);
352 for (i
= 1; i
< nr_pages
; i
++) {
353 struct page
*p
= page
+ i
;
355 set_page_count(p
, 0);
356 p
->first_page
= page
;
360 /* update __split_huge_page_refcount if you change this function */
361 static int destroy_compound_page(struct page
*page
, unsigned long order
)
364 int nr_pages
= 1 << order
;
367 if (unlikely(compound_order(page
) != order
) ||
368 unlikely(!PageHead(page
))) {
373 __ClearPageHead(page
);
375 for (i
= 1; i
< nr_pages
; i
++) {
376 struct page
*p
= page
+ i
;
378 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
388 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
393 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
394 * and __GFP_HIGHMEM from hard or soft interrupt context.
396 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
397 for (i
= 0; i
< (1 << order
); i
++)
398 clear_highpage(page
+ i
);
401 #ifdef CONFIG_DEBUG_PAGEALLOC
402 unsigned int _debug_guardpage_minorder
;
404 static int __init
debug_guardpage_minorder_setup(char *buf
)
408 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
409 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
412 _debug_guardpage_minorder
= res
;
413 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
416 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
418 static inline void set_page_guard_flag(struct page
*page
)
420 __set_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
423 static inline void clear_page_guard_flag(struct page
*page
)
425 __clear_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
428 static inline void set_page_guard_flag(struct page
*page
) { }
429 static inline void clear_page_guard_flag(struct page
*page
) { }
432 static inline void set_page_order(struct page
*page
, int order
)
434 set_page_private(page
, order
);
435 __SetPageBuddy(page
);
438 static inline void rmv_page_order(struct page
*page
)
440 __ClearPageBuddy(page
);
441 set_page_private(page
, 0);
445 * Locate the struct page for both the matching buddy in our
446 * pair (buddy1) and the combined O(n+1) page they form (page).
448 * 1) Any buddy B1 will have an order O twin B2 which satisfies
449 * the following equation:
451 * For example, if the starting buddy (buddy2) is #8 its order
453 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
455 * 2) Any buddy B will have an order O+1 parent P which
456 * satisfies the following equation:
459 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
461 static inline unsigned long
462 __find_buddy_index(unsigned long page_idx
, unsigned int order
)
464 return page_idx
^ (1 << order
);
468 * This function checks whether a page is free && is the buddy
469 * we can do coalesce a page and its buddy if
470 * (a) the buddy is not in a hole &&
471 * (b) the buddy is in the buddy system &&
472 * (c) a page and its buddy have the same order &&
473 * (d) a page and its buddy are in the same zone.
475 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
476 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
478 * For recording page's order, we use page_private(page).
480 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
483 if (!pfn_valid_within(page_to_pfn(buddy
)))
486 if (page_zone_id(page
) != page_zone_id(buddy
))
489 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
490 VM_BUG_ON(page_count(buddy
) != 0);
494 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
495 VM_BUG_ON(page_count(buddy
) != 0);
502 * Freeing function for a buddy system allocator.
504 * The concept of a buddy system is to maintain direct-mapped table
505 * (containing bit values) for memory blocks of various "orders".
506 * The bottom level table contains the map for the smallest allocatable
507 * units of memory (here, pages), and each level above it describes
508 * pairs of units from the levels below, hence, "buddies".
509 * At a high level, all that happens here is marking the table entry
510 * at the bottom level available, and propagating the changes upward
511 * as necessary, plus some accounting needed to play nicely with other
512 * parts of the VM system.
513 * At each level, we keep a list of pages, which are heads of continuous
514 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
515 * order is recorded in page_private(page) field.
516 * So when we are allocating or freeing one, we can derive the state of the
517 * other. That is, if we allocate a small block, and both were
518 * free, the remainder of the region must be split into blocks.
519 * If a block is freed, and its buddy is also free, then this
520 * triggers coalescing into a block of larger size.
525 static inline void __free_one_page(struct page
*page
,
526 struct zone
*zone
, unsigned int order
,
529 unsigned long page_idx
;
530 unsigned long combined_idx
;
531 unsigned long uninitialized_var(buddy_idx
);
534 if (unlikely(PageCompound(page
)))
535 if (unlikely(destroy_compound_page(page
, order
)))
538 VM_BUG_ON(migratetype
== -1);
540 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
542 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
543 VM_BUG_ON(bad_range(zone
, page
));
545 while (order
< MAX_ORDER
-1) {
546 buddy_idx
= __find_buddy_index(page_idx
, order
);
547 buddy
= page
+ (buddy_idx
- page_idx
);
548 if (!page_is_buddy(page
, buddy
, order
))
551 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
552 * merge with it and move up one order.
554 if (page_is_guard(buddy
)) {
555 clear_page_guard_flag(buddy
);
556 set_page_private(page
, 0);
557 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
559 list_del(&buddy
->lru
);
560 zone
->free_area
[order
].nr_free
--;
561 rmv_page_order(buddy
);
563 combined_idx
= buddy_idx
& page_idx
;
564 page
= page
+ (combined_idx
- page_idx
);
565 page_idx
= combined_idx
;
568 set_page_order(page
, order
);
571 * If this is not the largest possible page, check if the buddy
572 * of the next-highest order is free. If it is, it's possible
573 * that pages are being freed that will coalesce soon. In case,
574 * that is happening, add the free page to the tail of the list
575 * so it's less likely to be used soon and more likely to be merged
576 * as a higher order page
578 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
579 struct page
*higher_page
, *higher_buddy
;
580 combined_idx
= buddy_idx
& page_idx
;
581 higher_page
= page
+ (combined_idx
- page_idx
);
582 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
583 higher_buddy
= page
+ (buddy_idx
- combined_idx
);
584 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
585 list_add_tail(&page
->lru
,
586 &zone
->free_area
[order
].free_list
[migratetype
]);
591 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
593 zone
->free_area
[order
].nr_free
++;
597 * free_page_mlock() -- clean up attempts to free and mlocked() page.
598 * Page should not be on lru, so no need to fix that up.
599 * free_pages_check() will verify...
601 static inline void free_page_mlock(struct page
*page
)
603 __dec_zone_page_state(page
, NR_MLOCK
);
604 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
607 static inline int free_pages_check(struct page
*page
)
609 if (unlikely(page_mapcount(page
) |
610 (page
->mapping
!= NULL
) |
611 (atomic_read(&page
->_count
) != 0) |
612 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
) |
613 (mem_cgroup_bad_page_check(page
)))) {
617 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
618 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
623 * Frees a number of pages from the PCP lists
624 * Assumes all pages on list are in same zone, and of same order.
625 * count is the number of pages to free.
627 * If the zone was previously in an "all pages pinned" state then look to
628 * see if this freeing clears that state.
630 * And clear the zone's pages_scanned counter, to hold off the "all pages are
631 * pinned" detection logic.
633 static void free_pcppages_bulk(struct zone
*zone
, int count
,
634 struct per_cpu_pages
*pcp
)
640 spin_lock(&zone
->lock
);
641 zone
->all_unreclaimable
= 0;
642 zone
->pages_scanned
= 0;
646 struct list_head
*list
;
649 * Remove pages from lists in a round-robin fashion. A
650 * batch_free count is maintained that is incremented when an
651 * empty list is encountered. This is so more pages are freed
652 * off fuller lists instead of spinning excessively around empty
657 if (++migratetype
== MIGRATE_PCPTYPES
)
659 list
= &pcp
->lists
[migratetype
];
660 } while (list_empty(list
));
662 /* This is the only non-empty list. Free them all. */
663 if (batch_free
== MIGRATE_PCPTYPES
)
664 batch_free
= to_free
;
667 page
= list_entry(list
->prev
, struct page
, lru
);
668 /* must delete as __free_one_page list manipulates */
669 list_del(&page
->lru
);
670 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
671 __free_one_page(page
, zone
, 0, page_private(page
));
672 trace_mm_page_pcpu_drain(page
, 0, page_private(page
));
673 } while (--to_free
&& --batch_free
&& !list_empty(list
));
675 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
);
676 spin_unlock(&zone
->lock
);
679 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
682 spin_lock(&zone
->lock
);
683 zone
->all_unreclaimable
= 0;
684 zone
->pages_scanned
= 0;
686 __free_one_page(page
, zone
, order
, migratetype
);
687 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
688 spin_unlock(&zone
->lock
);
691 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
696 trace_mm_page_free(page
, order
);
697 kmemcheck_free_shadow(page
, order
);
700 page
->mapping
= NULL
;
701 for (i
= 0; i
< (1 << order
); i
++)
702 bad
+= free_pages_check(page
+ i
);
706 if (!PageHighMem(page
)) {
707 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
708 debug_check_no_obj_freed(page_address(page
),
711 arch_free_page(page
, order
);
712 kernel_map_pages(page
, 1 << order
, 0);
717 static void __free_pages_ok(struct page
*page
, unsigned int order
)
720 int wasMlocked
= __TestClearPageMlocked(page
);
722 if (!free_pages_prepare(page
, order
))
725 local_irq_save(flags
);
726 if (unlikely(wasMlocked
))
727 free_page_mlock(page
);
728 __count_vm_events(PGFREE
, 1 << order
);
729 free_one_page(page_zone(page
), page
, order
,
730 get_pageblock_migratetype(page
));
731 local_irq_restore(flags
);
734 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
736 unsigned int nr_pages
= 1 << order
;
740 for (loop
= 0; loop
< nr_pages
; loop
++) {
741 struct page
*p
= &page
[loop
];
743 if (loop
+ 1 < nr_pages
)
745 __ClearPageReserved(p
);
746 set_page_count(p
, 0);
749 set_page_refcounted(page
);
750 __free_pages(page
, order
);
754 /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
755 void __init
init_cma_reserved_pageblock(struct page
*page
)
757 unsigned i
= pageblock_nr_pages
;
758 struct page
*p
= page
;
761 __ClearPageReserved(p
);
762 set_page_count(p
, 0);
765 set_page_refcounted(page
);
766 set_pageblock_migratetype(page
, MIGRATE_CMA
);
767 __free_pages(page
, pageblock_order
);
768 totalram_pages
+= pageblock_nr_pages
;
773 * The order of subdivision here is critical for the IO subsystem.
774 * Please do not alter this order without good reasons and regression
775 * testing. Specifically, as large blocks of memory are subdivided,
776 * the order in which smaller blocks are delivered depends on the order
777 * they're subdivided in this function. This is the primary factor
778 * influencing the order in which pages are delivered to the IO
779 * subsystem according to empirical testing, and this is also justified
780 * by considering the behavior of a buddy system containing a single
781 * large block of memory acted on by a series of small allocations.
782 * This behavior is a critical factor in sglist merging's success.
786 static inline void expand(struct zone
*zone
, struct page
*page
,
787 int low
, int high
, struct free_area
*area
,
790 unsigned long size
= 1 << high
;
796 VM_BUG_ON(bad_range(zone
, &page
[size
]));
798 #ifdef CONFIG_DEBUG_PAGEALLOC
799 if (high
< debug_guardpage_minorder()) {
801 * Mark as guard pages (or page), that will allow to
802 * merge back to allocator when buddy will be freed.
803 * Corresponding page table entries will not be touched,
804 * pages will stay not present in virtual address space
806 INIT_LIST_HEAD(&page
[size
].lru
);
807 set_page_guard_flag(&page
[size
]);
808 set_page_private(&page
[size
], high
);
809 /* Guard pages are not available for any usage */
810 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << high
));
814 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
816 set_page_order(&page
[size
], high
);
821 * This page is about to be returned from the page allocator
823 static inline int check_new_page(struct page
*page
)
825 if (unlikely(page_mapcount(page
) |
826 (page
->mapping
!= NULL
) |
827 (atomic_read(&page
->_count
) != 0) |
828 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
) |
829 (mem_cgroup_bad_page_check(page
)))) {
836 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
840 for (i
= 0; i
< (1 << order
); i
++) {
841 struct page
*p
= page
+ i
;
842 if (unlikely(check_new_page(p
)))
846 set_page_private(page
, 0);
847 set_page_refcounted(page
);
849 arch_alloc_page(page
, order
);
850 kernel_map_pages(page
, 1 << order
, 1);
852 if (gfp_flags
& __GFP_ZERO
)
853 prep_zero_page(page
, order
, gfp_flags
);
855 if (order
&& (gfp_flags
& __GFP_COMP
))
856 prep_compound_page(page
, order
);
862 * Go through the free lists for the given migratetype and remove
863 * the smallest available page from the freelists
866 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
869 unsigned int current_order
;
870 struct free_area
* area
;
873 /* Find a page of the appropriate size in the preferred list */
874 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
875 area
= &(zone
->free_area
[current_order
]);
876 if (list_empty(&area
->free_list
[migratetype
]))
879 page
= list_entry(area
->free_list
[migratetype
].next
,
881 list_del(&page
->lru
);
882 rmv_page_order(page
);
884 expand(zone
, page
, order
, current_order
, area
, migratetype
);
893 * This array describes the order lists are fallen back to when
894 * the free lists for the desirable migrate type are depleted
896 static int fallbacks
[MIGRATE_TYPES
][4] = {
897 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
898 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
900 [MIGRATE_MOVABLE
] = { MIGRATE_CMA
, MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
901 [MIGRATE_CMA
] = { MIGRATE_RESERVE
}, /* Never used */
903 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
905 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
}, /* Never used */
906 [MIGRATE_ISOLATE
] = { MIGRATE_RESERVE
}, /* Never used */
910 * Move the free pages in a range to the free lists of the requested type.
911 * Note that start_page and end_pages are not aligned on a pageblock
912 * boundary. If alignment is required, use move_freepages_block()
914 static int move_freepages(struct zone
*zone
,
915 struct page
*start_page
, struct page
*end_page
,
922 #ifndef CONFIG_HOLES_IN_ZONE
924 * page_zone is not safe to call in this context when
925 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
926 * anyway as we check zone boundaries in move_freepages_block().
927 * Remove at a later date when no bug reports exist related to
928 * grouping pages by mobility
930 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
933 for (page
= start_page
; page
<= end_page
;) {
934 /* Make sure we are not inadvertently changing nodes */
935 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
937 if (!pfn_valid_within(page_to_pfn(page
))) {
942 if (!PageBuddy(page
)) {
947 order
= page_order(page
);
948 list_move(&page
->lru
,
949 &zone
->free_area
[order
].free_list
[migratetype
]);
951 pages_moved
+= 1 << order
;
957 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
960 unsigned long start_pfn
, end_pfn
;
961 struct page
*start_page
, *end_page
;
963 start_pfn
= page_to_pfn(page
);
964 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
965 start_page
= pfn_to_page(start_pfn
);
966 end_page
= start_page
+ pageblock_nr_pages
- 1;
967 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
969 /* Do not cross zone boundaries */
970 if (start_pfn
< zone
->zone_start_pfn
)
972 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
975 return move_freepages(zone
, start_page
, end_page
, migratetype
);
978 static void change_pageblock_range(struct page
*pageblock_page
,
979 int start_order
, int migratetype
)
981 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
983 while (nr_pageblocks
--) {
984 set_pageblock_migratetype(pageblock_page
, migratetype
);
985 pageblock_page
+= pageblock_nr_pages
;
989 /* Remove an element from the buddy allocator from the fallback list */
990 static inline struct page
*
991 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
993 struct free_area
* area
;
998 /* Find the largest possible block of pages in the other list */
999 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
1002 migratetype
= fallbacks
[start_migratetype
][i
];
1004 /* MIGRATE_RESERVE handled later if necessary */
1005 if (migratetype
== MIGRATE_RESERVE
)
1008 area
= &(zone
->free_area
[current_order
]);
1009 if (list_empty(&area
->free_list
[migratetype
]))
1012 page
= list_entry(area
->free_list
[migratetype
].next
,
1017 * If breaking a large block of pages, move all free
1018 * pages to the preferred allocation list. If falling
1019 * back for a reclaimable kernel allocation, be more
1020 * aggressive about taking ownership of free pages
1022 * On the other hand, never change migration
1023 * type of MIGRATE_CMA pageblocks nor move CMA
1024 * pages on different free lists. We don't
1025 * want unmovable pages to be allocated from
1026 * MIGRATE_CMA areas.
1028 if (!is_migrate_cma(migratetype
) &&
1029 (unlikely(current_order
>= pageblock_order
/ 2) ||
1030 start_migratetype
== MIGRATE_RECLAIMABLE
||
1031 page_group_by_mobility_disabled
)) {
1033 pages
= move_freepages_block(zone
, page
,
1036 /* Claim the whole block if over half of it is free */
1037 if (pages
>= (1 << (pageblock_order
-1)) ||
1038 page_group_by_mobility_disabled
)
1039 set_pageblock_migratetype(page
,
1042 migratetype
= start_migratetype
;
1045 /* Remove the page from the freelists */
1046 list_del(&page
->lru
);
1047 rmv_page_order(page
);
1049 /* Take ownership for orders >= pageblock_order */
1050 if (current_order
>= pageblock_order
&&
1051 !is_migrate_cma(migratetype
))
1052 change_pageblock_range(page
, current_order
,
1055 expand(zone
, page
, order
, current_order
, area
,
1056 is_migrate_cma(migratetype
)
1057 ? migratetype
: start_migratetype
);
1059 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1060 start_migratetype
, migratetype
);
1070 * Do the hard work of removing an element from the buddy allocator.
1071 * Call me with the zone->lock already held.
1073 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1079 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1081 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1082 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1085 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1086 * is used because __rmqueue_smallest is an inline function
1087 * and we want just one call site
1090 migratetype
= MIGRATE_RESERVE
;
1095 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1100 * Obtain a specified number of elements from the buddy allocator, all under
1101 * a single hold of the lock, for efficiency. Add them to the supplied list.
1102 * Returns the number of new pages which were placed at *list.
1104 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1105 unsigned long count
, struct list_head
*list
,
1106 int migratetype
, int cold
)
1108 int mt
= migratetype
, i
;
1110 spin_lock(&zone
->lock
);
1111 for (i
= 0; i
< count
; ++i
) {
1112 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1113 if (unlikely(page
== NULL
))
1117 * Split buddy pages returned by expand() are received here
1118 * in physical page order. The page is added to the callers and
1119 * list and the list head then moves forward. From the callers
1120 * perspective, the linked list is ordered by page number in
1121 * some conditions. This is useful for IO devices that can
1122 * merge IO requests if the physical pages are ordered
1125 if (likely(cold
== 0))
1126 list_add(&page
->lru
, list
);
1128 list_add_tail(&page
->lru
, list
);
1129 if (IS_ENABLED(CONFIG_CMA
)) {
1130 mt
= get_pageblock_migratetype(page
);
1131 if (!is_migrate_cma(mt
) && mt
!= MIGRATE_ISOLATE
)
1134 set_page_private(page
, mt
);
1137 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1138 spin_unlock(&zone
->lock
);
1144 * Called from the vmstat counter updater to drain pagesets of this
1145 * currently executing processor on remote nodes after they have
1148 * Note that this function must be called with the thread pinned to
1149 * a single processor.
1151 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1153 unsigned long flags
;
1156 local_irq_save(flags
);
1157 if (pcp
->count
>= pcp
->batch
)
1158 to_drain
= pcp
->batch
;
1160 to_drain
= pcp
->count
;
1162 free_pcppages_bulk(zone
, to_drain
, pcp
);
1163 pcp
->count
-= to_drain
;
1165 local_irq_restore(flags
);
1170 * Drain pages of the indicated processor.
1172 * The processor must either be the current processor and the
1173 * thread pinned to the current processor or a processor that
1176 static void drain_pages(unsigned int cpu
)
1178 unsigned long flags
;
1181 for_each_populated_zone(zone
) {
1182 struct per_cpu_pageset
*pset
;
1183 struct per_cpu_pages
*pcp
;
1185 local_irq_save(flags
);
1186 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1190 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1193 local_irq_restore(flags
);
1198 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1200 void drain_local_pages(void *arg
)
1202 drain_pages(smp_processor_id());
1206 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1208 * Note that this code is protected against sending an IPI to an offline
1209 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1210 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1211 * nothing keeps CPUs from showing up after we populated the cpumask and
1212 * before the call to on_each_cpu_mask().
1214 void drain_all_pages(void)
1217 struct per_cpu_pageset
*pcp
;
1221 * Allocate in the BSS so we wont require allocation in
1222 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1224 static cpumask_t cpus_with_pcps
;
1227 * We don't care about racing with CPU hotplug event
1228 * as offline notification will cause the notified
1229 * cpu to drain that CPU pcps and on_each_cpu_mask
1230 * disables preemption as part of its processing
1232 for_each_online_cpu(cpu
) {
1233 bool has_pcps
= false;
1234 for_each_populated_zone(zone
) {
1235 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1236 if (pcp
->pcp
.count
) {
1242 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1244 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1246 on_each_cpu_mask(&cpus_with_pcps
, drain_local_pages
, NULL
, 1);
1249 #ifdef CONFIG_HIBERNATION
1251 void mark_free_pages(struct zone
*zone
)
1253 unsigned long pfn
, max_zone_pfn
;
1254 unsigned long flags
;
1256 struct list_head
*curr
;
1258 if (!zone
->spanned_pages
)
1261 spin_lock_irqsave(&zone
->lock
, flags
);
1263 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1264 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1265 if (pfn_valid(pfn
)) {
1266 struct page
*page
= pfn_to_page(pfn
);
1268 if (!swsusp_page_is_forbidden(page
))
1269 swsusp_unset_page_free(page
);
1272 for_each_migratetype_order(order
, t
) {
1273 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1276 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1277 for (i
= 0; i
< (1UL << order
); i
++)
1278 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1281 spin_unlock_irqrestore(&zone
->lock
, flags
);
1283 #endif /* CONFIG_PM */
1286 * Free a 0-order page
1287 * cold == 1 ? free a cold page : free a hot page
1289 void free_hot_cold_page(struct page
*page
, int cold
)
1291 struct zone
*zone
= page_zone(page
);
1292 struct per_cpu_pages
*pcp
;
1293 unsigned long flags
;
1295 int wasMlocked
= __TestClearPageMlocked(page
);
1297 if (!free_pages_prepare(page
, 0))
1300 migratetype
= get_pageblock_migratetype(page
);
1301 set_page_private(page
, migratetype
);
1302 local_irq_save(flags
);
1303 if (unlikely(wasMlocked
))
1304 free_page_mlock(page
);
1305 __count_vm_event(PGFREE
);
1308 * We only track unmovable, reclaimable and movable on pcp lists.
1309 * Free ISOLATE pages back to the allocator because they are being
1310 * offlined but treat RESERVE as movable pages so we can get those
1311 * areas back if necessary. Otherwise, we may have to free
1312 * excessively into the page allocator
1314 if (migratetype
>= MIGRATE_PCPTYPES
) {
1315 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1316 free_one_page(zone
, page
, 0, migratetype
);
1319 migratetype
= MIGRATE_MOVABLE
;
1322 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1324 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1326 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1328 if (pcp
->count
>= pcp
->high
) {
1329 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1330 pcp
->count
-= pcp
->batch
;
1334 local_irq_restore(flags
);
1338 * Free a list of 0-order pages
1340 void free_hot_cold_page_list(struct list_head
*list
, int cold
)
1342 struct page
*page
, *next
;
1344 list_for_each_entry_safe(page
, next
, list
, lru
) {
1345 trace_mm_page_free_batched(page
, cold
);
1346 free_hot_cold_page(page
, cold
);
1351 * split_page takes a non-compound higher-order page, and splits it into
1352 * n (1<<order) sub-pages: page[0..n]
1353 * Each sub-page must be freed individually.
1355 * Note: this is probably too low level an operation for use in drivers.
1356 * Please consult with lkml before using this in your driver.
1358 void split_page(struct page
*page
, unsigned int order
)
1362 VM_BUG_ON(PageCompound(page
));
1363 VM_BUG_ON(!page_count(page
));
1365 #ifdef CONFIG_KMEMCHECK
1367 * Split shadow pages too, because free(page[0]) would
1368 * otherwise free the whole shadow.
1370 if (kmemcheck_page_is_tracked(page
))
1371 split_page(virt_to_page(page
[0].shadow
), order
);
1374 for (i
= 1; i
< (1 << order
); i
++)
1375 set_page_refcounted(page
+ i
);
1379 * Similar to split_page except the page is already free. As this is only
1380 * being used for migration, the migratetype of the block also changes.
1381 * As this is called with interrupts disabled, the caller is responsible
1382 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1385 * Note: this is probably too low level an operation for use in drivers.
1386 * Please consult with lkml before using this in your driver.
1388 int split_free_page(struct page
*page
)
1391 unsigned long watermark
;
1394 BUG_ON(!PageBuddy(page
));
1396 zone
= page_zone(page
);
1397 order
= page_order(page
);
1399 /* Obey watermarks as if the page was being allocated */
1400 watermark
= low_wmark_pages(zone
) + (1 << order
);
1401 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1404 /* Remove page from free list */
1405 list_del(&page
->lru
);
1406 zone
->free_area
[order
].nr_free
--;
1407 rmv_page_order(page
);
1408 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1UL << order
));
1410 /* Split into individual pages */
1411 set_page_refcounted(page
);
1412 split_page(page
, order
);
1414 if (order
>= pageblock_order
- 1) {
1415 struct page
*endpage
= page
+ (1 << order
) - 1;
1416 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
1417 int mt
= get_pageblock_migratetype(page
);
1418 if (mt
!= MIGRATE_ISOLATE
&& !is_migrate_cma(mt
))
1419 set_pageblock_migratetype(page
,
1428 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1429 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1433 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1434 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1437 unsigned long flags
;
1439 int cold
= !!(gfp_flags
& __GFP_COLD
);
1442 if (likely(order
== 0)) {
1443 struct per_cpu_pages
*pcp
;
1444 struct list_head
*list
;
1446 local_irq_save(flags
);
1447 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1448 list
= &pcp
->lists
[migratetype
];
1449 if (list_empty(list
)) {
1450 pcp
->count
+= rmqueue_bulk(zone
, 0,
1453 if (unlikely(list_empty(list
)))
1458 page
= list_entry(list
->prev
, struct page
, lru
);
1460 page
= list_entry(list
->next
, struct page
, lru
);
1462 list_del(&page
->lru
);
1465 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1467 * __GFP_NOFAIL is not to be used in new code.
1469 * All __GFP_NOFAIL callers should be fixed so that they
1470 * properly detect and handle allocation failures.
1472 * We most definitely don't want callers attempting to
1473 * allocate greater than order-1 page units with
1476 WARN_ON_ONCE(order
> 1);
1478 spin_lock_irqsave(&zone
->lock
, flags
);
1479 page
= __rmqueue(zone
, order
, migratetype
);
1480 spin_unlock(&zone
->lock
);
1483 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << order
));
1486 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1487 zone_statistics(preferred_zone
, zone
, gfp_flags
);
1488 local_irq_restore(flags
);
1490 VM_BUG_ON(bad_range(zone
, page
));
1491 if (prep_new_page(page
, order
, gfp_flags
))
1496 local_irq_restore(flags
);
1500 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1501 #define ALLOC_WMARK_MIN WMARK_MIN
1502 #define ALLOC_WMARK_LOW WMARK_LOW
1503 #define ALLOC_WMARK_HIGH WMARK_HIGH
1504 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1506 /* Mask to get the watermark bits */
1507 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1509 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1510 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1511 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1513 #ifdef CONFIG_FAIL_PAGE_ALLOC
1516 struct fault_attr attr
;
1518 u32 ignore_gfp_highmem
;
1519 u32 ignore_gfp_wait
;
1521 } fail_page_alloc
= {
1522 .attr
= FAULT_ATTR_INITIALIZER
,
1523 .ignore_gfp_wait
= 1,
1524 .ignore_gfp_highmem
= 1,
1528 static int __init
setup_fail_page_alloc(char *str
)
1530 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1532 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1534 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1536 if (order
< fail_page_alloc
.min_order
)
1538 if (gfp_mask
& __GFP_NOFAIL
)
1540 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1542 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1545 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1548 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1550 static int __init
fail_page_alloc_debugfs(void)
1552 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1555 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
1556 &fail_page_alloc
.attr
);
1558 return PTR_ERR(dir
);
1560 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1561 &fail_page_alloc
.ignore_gfp_wait
))
1563 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1564 &fail_page_alloc
.ignore_gfp_highmem
))
1566 if (!debugfs_create_u32("min-order", mode
, dir
,
1567 &fail_page_alloc
.min_order
))
1572 debugfs_remove_recursive(dir
);
1577 late_initcall(fail_page_alloc_debugfs
);
1579 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1581 #else /* CONFIG_FAIL_PAGE_ALLOC */
1583 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1588 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1591 * Return true if free pages are above 'mark'. This takes into account the order
1592 * of the allocation.
1594 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1595 int classzone_idx
, int alloc_flags
, long free_pages
)
1597 /* free_pages my go negative - that's OK */
1601 free_pages
-= (1 << order
) - 1;
1602 if (alloc_flags
& ALLOC_HIGH
)
1604 if (alloc_flags
& ALLOC_HARDER
)
1607 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1609 for (o
= 0; o
< order
; o
++) {
1610 /* At the next order, this order's pages become unavailable */
1611 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1613 /* Require fewer higher order pages to be free */
1616 if (free_pages
<= min
)
1622 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1623 int classzone_idx
, int alloc_flags
)
1625 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1626 zone_page_state(z
, NR_FREE_PAGES
));
1629 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1630 int classzone_idx
, int alloc_flags
)
1632 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1634 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1635 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1637 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1643 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1644 * skip over zones that are not allowed by the cpuset, or that have
1645 * been recently (in last second) found to be nearly full. See further
1646 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1647 * that have to skip over a lot of full or unallowed zones.
1649 * If the zonelist cache is present in the passed in zonelist, then
1650 * returns a pointer to the allowed node mask (either the current
1651 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1653 * If the zonelist cache is not available for this zonelist, does
1654 * nothing and returns NULL.
1656 * If the fullzones BITMAP in the zonelist cache is stale (more than
1657 * a second since last zap'd) then we zap it out (clear its bits.)
1659 * We hold off even calling zlc_setup, until after we've checked the
1660 * first zone in the zonelist, on the theory that most allocations will
1661 * be satisfied from that first zone, so best to examine that zone as
1662 * quickly as we can.
1664 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1666 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1667 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1669 zlc
= zonelist
->zlcache_ptr
;
1673 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1674 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1675 zlc
->last_full_zap
= jiffies
;
1678 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1679 &cpuset_current_mems_allowed
:
1680 &node_states
[N_HIGH_MEMORY
];
1681 return allowednodes
;
1685 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1686 * if it is worth looking at further for free memory:
1687 * 1) Check that the zone isn't thought to be full (doesn't have its
1688 * bit set in the zonelist_cache fullzones BITMAP).
1689 * 2) Check that the zones node (obtained from the zonelist_cache
1690 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1691 * Return true (non-zero) if zone is worth looking at further, or
1692 * else return false (zero) if it is not.
1694 * This check -ignores- the distinction between various watermarks,
1695 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1696 * found to be full for any variation of these watermarks, it will
1697 * be considered full for up to one second by all requests, unless
1698 * we are so low on memory on all allowed nodes that we are forced
1699 * into the second scan of the zonelist.
1701 * In the second scan we ignore this zonelist cache and exactly
1702 * apply the watermarks to all zones, even it is slower to do so.
1703 * We are low on memory in the second scan, and should leave no stone
1704 * unturned looking for a free page.
1706 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1707 nodemask_t
*allowednodes
)
1709 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1710 int i
; /* index of *z in zonelist zones */
1711 int n
; /* node that zone *z is on */
1713 zlc
= zonelist
->zlcache_ptr
;
1717 i
= z
- zonelist
->_zonerefs
;
1720 /* This zone is worth trying if it is allowed but not full */
1721 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1725 * Given 'z' scanning a zonelist, set the corresponding bit in
1726 * zlc->fullzones, so that subsequent attempts to allocate a page
1727 * from that zone don't waste time re-examining it.
1729 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1731 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1732 int i
; /* index of *z in zonelist zones */
1734 zlc
= zonelist
->zlcache_ptr
;
1738 i
= z
- zonelist
->_zonerefs
;
1740 set_bit(i
, zlc
->fullzones
);
1744 * clear all zones full, called after direct reclaim makes progress so that
1745 * a zone that was recently full is not skipped over for up to a second
1747 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1749 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1751 zlc
= zonelist
->zlcache_ptr
;
1755 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1758 #else /* CONFIG_NUMA */
1760 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1765 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1766 nodemask_t
*allowednodes
)
1771 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1775 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1778 #endif /* CONFIG_NUMA */
1781 * get_page_from_freelist goes through the zonelist trying to allocate
1784 static struct page
*
1785 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1786 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1787 struct zone
*preferred_zone
, int migratetype
)
1790 struct page
*page
= NULL
;
1793 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1794 int zlc_active
= 0; /* set if using zonelist_cache */
1795 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1797 classzone_idx
= zone_idx(preferred_zone
);
1800 * Scan zonelist, looking for a zone with enough free.
1801 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1803 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1804 high_zoneidx
, nodemask
) {
1805 if (NUMA_BUILD
&& zlc_active
&&
1806 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1808 if ((alloc_flags
& ALLOC_CPUSET
) &&
1809 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1812 * When allocating a page cache page for writing, we
1813 * want to get it from a zone that is within its dirty
1814 * limit, such that no single zone holds more than its
1815 * proportional share of globally allowed dirty pages.
1816 * The dirty limits take into account the zone's
1817 * lowmem reserves and high watermark so that kswapd
1818 * should be able to balance it without having to
1819 * write pages from its LRU list.
1821 * This may look like it could increase pressure on
1822 * lower zones by failing allocations in higher zones
1823 * before they are full. But the pages that do spill
1824 * over are limited as the lower zones are protected
1825 * by this very same mechanism. It should not become
1826 * a practical burden to them.
1828 * XXX: For now, allow allocations to potentially
1829 * exceed the per-zone dirty limit in the slowpath
1830 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1831 * which is important when on a NUMA setup the allowed
1832 * zones are together not big enough to reach the
1833 * global limit. The proper fix for these situations
1834 * will require awareness of zones in the
1835 * dirty-throttling and the flusher threads.
1837 if ((alloc_flags
& ALLOC_WMARK_LOW
) &&
1838 (gfp_mask
& __GFP_WRITE
) && !zone_dirty_ok(zone
))
1839 goto this_zone_full
;
1841 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1842 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1846 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1847 if (zone_watermark_ok(zone
, order
, mark
,
1848 classzone_idx
, alloc_flags
))
1851 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1853 * we do zlc_setup if there are multiple nodes
1854 * and before considering the first zone allowed
1857 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1862 if (zone_reclaim_mode
== 0)
1863 goto this_zone_full
;
1866 * As we may have just activated ZLC, check if the first
1867 * eligible zone has failed zone_reclaim recently.
1869 if (NUMA_BUILD
&& zlc_active
&&
1870 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1873 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1875 case ZONE_RECLAIM_NOSCAN
:
1878 case ZONE_RECLAIM_FULL
:
1879 /* scanned but unreclaimable */
1882 /* did we reclaim enough */
1883 if (!zone_watermark_ok(zone
, order
, mark
,
1884 classzone_idx
, alloc_flags
))
1885 goto this_zone_full
;
1890 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1891 gfp_mask
, migratetype
);
1896 zlc_mark_zone_full(zonelist
, z
);
1899 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1900 /* Disable zlc cache for second zonelist scan */
1908 * Large machines with many possible nodes should not always dump per-node
1909 * meminfo in irq context.
1911 static inline bool should_suppress_show_mem(void)
1916 ret
= in_interrupt();
1921 static DEFINE_RATELIMIT_STATE(nopage_rs
,
1922 DEFAULT_RATELIMIT_INTERVAL
,
1923 DEFAULT_RATELIMIT_BURST
);
1925 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
1927 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
1929 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
1930 debug_guardpage_minorder() > 0)
1934 * This documents exceptions given to allocations in certain
1935 * contexts that are allowed to allocate outside current's set
1938 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
1939 if (test_thread_flag(TIF_MEMDIE
) ||
1940 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
1941 filter
&= ~SHOW_MEM_FILTER_NODES
;
1942 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
1943 filter
&= ~SHOW_MEM_FILTER_NODES
;
1946 struct va_format vaf
;
1949 va_start(args
, fmt
);
1954 pr_warn("%pV", &vaf
);
1959 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
1960 current
->comm
, order
, gfp_mask
);
1963 if (!should_suppress_show_mem())
1968 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1969 unsigned long did_some_progress
,
1970 unsigned long pages_reclaimed
)
1972 /* Do not loop if specifically requested */
1973 if (gfp_mask
& __GFP_NORETRY
)
1976 /* Always retry if specifically requested */
1977 if (gfp_mask
& __GFP_NOFAIL
)
1981 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
1982 * making forward progress without invoking OOM. Suspend also disables
1983 * storage devices so kswapd will not help. Bail if we are suspending.
1985 if (!did_some_progress
&& pm_suspended_storage())
1989 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1990 * means __GFP_NOFAIL, but that may not be true in other
1993 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1997 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1998 * specified, then we retry until we no longer reclaim any pages
1999 * (above), or we've reclaimed an order of pages at least as
2000 * large as the allocation's order. In both cases, if the
2001 * allocation still fails, we stop retrying.
2003 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
2009 static inline struct page
*
2010 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2011 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2012 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2017 /* Acquire the OOM killer lock for the zones in zonelist */
2018 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
2019 schedule_timeout_uninterruptible(1);
2024 * Go through the zonelist yet one more time, keep very high watermark
2025 * here, this is only to catch a parallel oom killing, we must fail if
2026 * we're still under heavy pressure.
2028 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
2029 order
, zonelist
, high_zoneidx
,
2030 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
2031 preferred_zone
, migratetype
);
2035 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2036 /* The OOM killer will not help higher order allocs */
2037 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2039 /* The OOM killer does not needlessly kill tasks for lowmem */
2040 if (high_zoneidx
< ZONE_NORMAL
)
2043 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2044 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2045 * The caller should handle page allocation failure by itself if
2046 * it specifies __GFP_THISNODE.
2047 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2049 if (gfp_mask
& __GFP_THISNODE
)
2052 /* Exhausted what can be done so it's blamo time */
2053 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
, false);
2056 clear_zonelist_oom(zonelist
, gfp_mask
);
2060 #ifdef CONFIG_COMPACTION
2061 /* Try memory compaction for high-order allocations before reclaim */
2062 static struct page
*
2063 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2064 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2065 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2066 int migratetype
, bool sync_migration
,
2067 bool *deferred_compaction
,
2068 unsigned long *did_some_progress
)
2075 if (compaction_deferred(preferred_zone
, order
)) {
2076 *deferred_compaction
= true;
2080 current
->flags
|= PF_MEMALLOC
;
2081 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
2082 nodemask
, sync_migration
);
2083 current
->flags
&= ~PF_MEMALLOC
;
2084 if (*did_some_progress
!= COMPACT_SKIPPED
) {
2086 /* Page migration frees to the PCP lists but we want merging */
2087 drain_pages(get_cpu());
2090 page
= get_page_from_freelist(gfp_mask
, nodemask
,
2091 order
, zonelist
, high_zoneidx
,
2092 alloc_flags
, preferred_zone
,
2095 preferred_zone
->compact_considered
= 0;
2096 preferred_zone
->compact_defer_shift
= 0;
2097 if (order
>= preferred_zone
->compact_order_failed
)
2098 preferred_zone
->compact_order_failed
= order
+ 1;
2099 count_vm_event(COMPACTSUCCESS
);
2104 * It's bad if compaction run occurs and fails.
2105 * The most likely reason is that pages exist,
2106 * but not enough to satisfy watermarks.
2108 count_vm_event(COMPACTFAIL
);
2111 * As async compaction considers a subset of pageblocks, only
2112 * defer if the failure was a sync compaction failure.
2115 defer_compaction(preferred_zone
, order
);
2123 static inline struct page
*
2124 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2125 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2126 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2127 int migratetype
, bool sync_migration
,
2128 bool *deferred_compaction
,
2129 unsigned long *did_some_progress
)
2133 #endif /* CONFIG_COMPACTION */
2135 /* Perform direct synchronous page reclaim */
2137 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
, struct zonelist
*zonelist
,
2138 nodemask_t
*nodemask
)
2140 struct reclaim_state reclaim_state
;
2145 /* We now go into synchronous reclaim */
2146 cpuset_memory_pressure_bump();
2147 current
->flags
|= PF_MEMALLOC
;
2148 lockdep_set_current_reclaim_state(gfp_mask
);
2149 reclaim_state
.reclaimed_slab
= 0;
2150 current
->reclaim_state
= &reclaim_state
;
2152 progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
2154 current
->reclaim_state
= NULL
;
2155 lockdep_clear_current_reclaim_state();
2156 current
->flags
&= ~PF_MEMALLOC
;
2163 /* The really slow allocator path where we enter direct reclaim */
2164 static inline struct page
*
2165 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2166 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2167 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2168 int migratetype
, unsigned long *did_some_progress
)
2170 struct page
*page
= NULL
;
2171 bool drained
= false;
2173 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
2175 if (unlikely(!(*did_some_progress
)))
2178 /* After successful reclaim, reconsider all zones for allocation */
2180 zlc_clear_zones_full(zonelist
);
2183 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2184 zonelist
, high_zoneidx
,
2185 alloc_flags
, preferred_zone
,
2189 * If an allocation failed after direct reclaim, it could be because
2190 * pages are pinned on the per-cpu lists. Drain them and try again
2192 if (!page
&& !drained
) {
2202 * This is called in the allocator slow-path if the allocation request is of
2203 * sufficient urgency to ignore watermarks and take other desperate measures
2205 static inline struct page
*
2206 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2207 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2208 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2214 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2215 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
2216 preferred_zone
, migratetype
);
2218 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2219 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2220 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2226 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
2227 enum zone_type high_zoneidx
,
2228 enum zone_type classzone_idx
)
2233 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
2234 wakeup_kswapd(zone
, order
, classzone_idx
);
2238 gfp_to_alloc_flags(gfp_t gfp_mask
)
2240 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2241 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2243 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2244 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2247 * The caller may dip into page reserves a bit more if the caller
2248 * cannot run direct reclaim, or if the caller has realtime scheduling
2249 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2250 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2252 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2256 * Not worth trying to allocate harder for
2257 * __GFP_NOMEMALLOC even if it can't schedule.
2259 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2260 alloc_flags
|= ALLOC_HARDER
;
2262 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2263 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2265 alloc_flags
&= ~ALLOC_CPUSET
;
2266 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2267 alloc_flags
|= ALLOC_HARDER
;
2269 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2270 if (!in_interrupt() &&
2271 ((current
->flags
& PF_MEMALLOC
) ||
2272 unlikely(test_thread_flag(TIF_MEMDIE
))))
2273 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2279 static inline struct page
*
2280 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2281 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2282 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2285 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2286 struct page
*page
= NULL
;
2288 unsigned long pages_reclaimed
= 0;
2289 unsigned long did_some_progress
;
2290 bool sync_migration
= false;
2291 bool deferred_compaction
= false;
2294 * In the slowpath, we sanity check order to avoid ever trying to
2295 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2296 * be using allocators in order of preference for an area that is
2299 if (order
>= MAX_ORDER
) {
2300 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2305 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2306 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2307 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2308 * using a larger set of nodes after it has established that the
2309 * allowed per node queues are empty and that nodes are
2312 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2316 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2317 wake_all_kswapd(order
, zonelist
, high_zoneidx
,
2318 zone_idx(preferred_zone
));
2321 * OK, we're below the kswapd watermark and have kicked background
2322 * reclaim. Now things get more complex, so set up alloc_flags according
2323 * to how we want to proceed.
2325 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2328 * Find the true preferred zone if the allocation is unconstrained by
2331 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2332 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2336 /* This is the last chance, in general, before the goto nopage. */
2337 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2338 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2339 preferred_zone
, migratetype
);
2343 /* Allocate without watermarks if the context allows */
2344 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2345 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2346 zonelist
, high_zoneidx
, nodemask
,
2347 preferred_zone
, migratetype
);
2352 /* Atomic allocations - we can't balance anything */
2356 /* Avoid recursion of direct reclaim */
2357 if (current
->flags
& PF_MEMALLOC
)
2360 /* Avoid allocations with no watermarks from looping endlessly */
2361 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2365 * Try direct compaction. The first pass is asynchronous. Subsequent
2366 * attempts after direct reclaim are synchronous
2368 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2369 zonelist
, high_zoneidx
,
2371 alloc_flags
, preferred_zone
,
2372 migratetype
, sync_migration
,
2373 &deferred_compaction
,
2374 &did_some_progress
);
2377 sync_migration
= true;
2380 * If compaction is deferred for high-order allocations, it is because
2381 * sync compaction recently failed. In this is the case and the caller
2382 * has requested the system not be heavily disrupted, fail the
2383 * allocation now instead of entering direct reclaim
2385 if (deferred_compaction
&& (gfp_mask
& __GFP_NO_KSWAPD
))
2388 /* Try direct reclaim and then allocating */
2389 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2390 zonelist
, high_zoneidx
,
2392 alloc_flags
, preferred_zone
,
2393 migratetype
, &did_some_progress
);
2398 * If we failed to make any progress reclaiming, then we are
2399 * running out of options and have to consider going OOM
2401 if (!did_some_progress
) {
2402 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2403 if (oom_killer_disabled
)
2405 /* Coredumps can quickly deplete all memory reserves */
2406 if ((current
->flags
& PF_DUMPCORE
) &&
2407 !(gfp_mask
& __GFP_NOFAIL
))
2409 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2410 zonelist
, high_zoneidx
,
2411 nodemask
, preferred_zone
,
2416 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2418 * The oom killer is not called for high-order
2419 * allocations that may fail, so if no progress
2420 * is being made, there are no other options and
2421 * retrying is unlikely to help.
2423 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2426 * The oom killer is not called for lowmem
2427 * allocations to prevent needlessly killing
2430 if (high_zoneidx
< ZONE_NORMAL
)
2438 /* Check if we should retry the allocation */
2439 pages_reclaimed
+= did_some_progress
;
2440 if (should_alloc_retry(gfp_mask
, order
, did_some_progress
,
2442 /* Wait for some write requests to complete then retry */
2443 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2447 * High-order allocations do not necessarily loop after
2448 * direct reclaim and reclaim/compaction depends on compaction
2449 * being called after reclaim so call directly if necessary
2451 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2452 zonelist
, high_zoneidx
,
2454 alloc_flags
, preferred_zone
,
2455 migratetype
, sync_migration
,
2456 &deferred_compaction
,
2457 &did_some_progress
);
2463 warn_alloc_failed(gfp_mask
, order
, NULL
);
2466 if (kmemcheck_enabled
)
2467 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2473 * This is the 'heart' of the zoned buddy allocator.
2476 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2477 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2479 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2480 struct zone
*preferred_zone
;
2481 struct page
*page
= NULL
;
2482 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2483 unsigned int cpuset_mems_cookie
;
2485 gfp_mask
&= gfp_allowed_mask
;
2487 lockdep_trace_alloc(gfp_mask
);
2489 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2491 if (should_fail_alloc_page(gfp_mask
, order
))
2495 * Check the zones suitable for the gfp_mask contain at least one
2496 * valid zone. It's possible to have an empty zonelist as a result
2497 * of GFP_THISNODE and a memoryless node
2499 if (unlikely(!zonelist
->_zonerefs
->zone
))
2503 cpuset_mems_cookie
= get_mems_allowed();
2505 /* The preferred zone is used for statistics later */
2506 first_zones_zonelist(zonelist
, high_zoneidx
,
2507 nodemask
? : &cpuset_current_mems_allowed
,
2509 if (!preferred_zone
)
2512 /* First allocation attempt */
2513 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2514 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
2515 preferred_zone
, migratetype
);
2516 if (unlikely(!page
))
2517 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2518 zonelist
, high_zoneidx
, nodemask
,
2519 preferred_zone
, migratetype
);
2521 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2525 * When updating a task's mems_allowed, it is possible to race with
2526 * parallel threads in such a way that an allocation can fail while
2527 * the mask is being updated. If a page allocation is about to fail,
2528 * check if the cpuset changed during allocation and if so, retry.
2530 if (unlikely(!put_mems_allowed(cpuset_mems_cookie
) && !page
))
2535 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2538 * Common helper functions.
2540 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2545 * __get_free_pages() returns a 32-bit address, which cannot represent
2548 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2550 page
= alloc_pages(gfp_mask
, order
);
2553 return (unsigned long) page_address(page
);
2555 EXPORT_SYMBOL(__get_free_pages
);
2557 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2559 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2561 EXPORT_SYMBOL(get_zeroed_page
);
2563 void __free_pages(struct page
*page
, unsigned int order
)
2565 if (put_page_testzero(page
)) {
2567 free_hot_cold_page(page
, 0);
2569 __free_pages_ok(page
, order
);
2573 EXPORT_SYMBOL(__free_pages
);
2575 void free_pages(unsigned long addr
, unsigned int order
)
2578 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2579 __free_pages(virt_to_page((void *)addr
), order
);
2583 EXPORT_SYMBOL(free_pages
);
2585 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2588 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2589 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2591 split_page(virt_to_page((void *)addr
), order
);
2592 while (used
< alloc_end
) {
2597 return (void *)addr
;
2601 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2602 * @size: the number of bytes to allocate
2603 * @gfp_mask: GFP flags for the allocation
2605 * This function is similar to alloc_pages(), except that it allocates the
2606 * minimum number of pages to satisfy the request. alloc_pages() can only
2607 * allocate memory in power-of-two pages.
2609 * This function is also limited by MAX_ORDER.
2611 * Memory allocated by this function must be released by free_pages_exact().
2613 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2615 unsigned int order
= get_order(size
);
2618 addr
= __get_free_pages(gfp_mask
, order
);
2619 return make_alloc_exact(addr
, order
, size
);
2621 EXPORT_SYMBOL(alloc_pages_exact
);
2624 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2626 * @nid: the preferred node ID where memory should be allocated
2627 * @size: the number of bytes to allocate
2628 * @gfp_mask: GFP flags for the allocation
2630 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2632 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2635 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2637 unsigned order
= get_order(size
);
2638 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2641 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2643 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2646 * free_pages_exact - release memory allocated via alloc_pages_exact()
2647 * @virt: the value returned by alloc_pages_exact.
2648 * @size: size of allocation, same value as passed to alloc_pages_exact().
2650 * Release the memory allocated by a previous call to alloc_pages_exact.
2652 void free_pages_exact(void *virt
, size_t size
)
2654 unsigned long addr
= (unsigned long)virt
;
2655 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2657 while (addr
< end
) {
2662 EXPORT_SYMBOL(free_pages_exact
);
2664 static unsigned int nr_free_zone_pages(int offset
)
2669 /* Just pick one node, since fallback list is circular */
2670 unsigned int sum
= 0;
2672 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2674 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2675 unsigned long size
= zone
->present_pages
;
2676 unsigned long high
= high_wmark_pages(zone
);
2685 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2687 unsigned int nr_free_buffer_pages(void)
2689 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2691 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2694 * Amount of free RAM allocatable within all zones
2696 unsigned int nr_free_pagecache_pages(void)
2698 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2701 static inline void show_node(struct zone
*zone
)
2704 printk("Node %d ", zone_to_nid(zone
));
2707 void si_meminfo(struct sysinfo
*val
)
2709 val
->totalram
= totalram_pages
;
2711 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2712 val
->bufferram
= nr_blockdev_pages();
2713 val
->totalhigh
= totalhigh_pages
;
2714 val
->freehigh
= nr_free_highpages();
2715 val
->mem_unit
= PAGE_SIZE
;
2718 EXPORT_SYMBOL(si_meminfo
);
2721 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2723 pg_data_t
*pgdat
= NODE_DATA(nid
);
2725 val
->totalram
= pgdat
->node_present_pages
;
2726 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2727 #ifdef CONFIG_HIGHMEM
2728 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2729 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2735 val
->mem_unit
= PAGE_SIZE
;
2740 * Determine whether the node should be displayed or not, depending on whether
2741 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2743 bool skip_free_areas_node(unsigned int flags
, int nid
)
2746 unsigned int cpuset_mems_cookie
;
2748 if (!(flags
& SHOW_MEM_FILTER_NODES
))
2752 cpuset_mems_cookie
= get_mems_allowed();
2753 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
2754 } while (!put_mems_allowed(cpuset_mems_cookie
));
2759 #define K(x) ((x) << (PAGE_SHIFT-10))
2762 * Show free area list (used inside shift_scroll-lock stuff)
2763 * We also calculate the percentage fragmentation. We do this by counting the
2764 * memory on each free list with the exception of the first item on the list.
2765 * Suppresses nodes that are not allowed by current's cpuset if
2766 * SHOW_MEM_FILTER_NODES is passed.
2768 void show_free_areas(unsigned int filter
)
2773 for_each_populated_zone(zone
) {
2774 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2777 printk("%s per-cpu:\n", zone
->name
);
2779 for_each_online_cpu(cpu
) {
2780 struct per_cpu_pageset
*pageset
;
2782 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2784 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2785 cpu
, pageset
->pcp
.high
,
2786 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2790 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2791 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2793 " dirty:%lu writeback:%lu unstable:%lu\n"
2794 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2795 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2796 global_page_state(NR_ACTIVE_ANON
),
2797 global_page_state(NR_INACTIVE_ANON
),
2798 global_page_state(NR_ISOLATED_ANON
),
2799 global_page_state(NR_ACTIVE_FILE
),
2800 global_page_state(NR_INACTIVE_FILE
),
2801 global_page_state(NR_ISOLATED_FILE
),
2802 global_page_state(NR_UNEVICTABLE
),
2803 global_page_state(NR_FILE_DIRTY
),
2804 global_page_state(NR_WRITEBACK
),
2805 global_page_state(NR_UNSTABLE_NFS
),
2806 global_page_state(NR_FREE_PAGES
),
2807 global_page_state(NR_SLAB_RECLAIMABLE
),
2808 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2809 global_page_state(NR_FILE_MAPPED
),
2810 global_page_state(NR_SHMEM
),
2811 global_page_state(NR_PAGETABLE
),
2812 global_page_state(NR_BOUNCE
));
2814 for_each_populated_zone(zone
) {
2817 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2825 " active_anon:%lukB"
2826 " inactive_anon:%lukB"
2827 " active_file:%lukB"
2828 " inactive_file:%lukB"
2829 " unevictable:%lukB"
2830 " isolated(anon):%lukB"
2831 " isolated(file):%lukB"
2838 " slab_reclaimable:%lukB"
2839 " slab_unreclaimable:%lukB"
2840 " kernel_stack:%lukB"
2844 " writeback_tmp:%lukB"
2845 " pages_scanned:%lu"
2846 " all_unreclaimable? %s"
2849 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2850 K(min_wmark_pages(zone
)),
2851 K(low_wmark_pages(zone
)),
2852 K(high_wmark_pages(zone
)),
2853 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2854 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2855 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2856 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2857 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2858 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
2859 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
2860 K(zone
->present_pages
),
2861 K(zone_page_state(zone
, NR_MLOCK
)),
2862 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
2863 K(zone_page_state(zone
, NR_WRITEBACK
)),
2864 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
2865 K(zone_page_state(zone
, NR_SHMEM
)),
2866 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
2867 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
2868 zone_page_state(zone
, NR_KERNEL_STACK
) *
2870 K(zone_page_state(zone
, NR_PAGETABLE
)),
2871 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
2872 K(zone_page_state(zone
, NR_BOUNCE
)),
2873 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
2874 zone
->pages_scanned
,
2875 (zone
->all_unreclaimable
? "yes" : "no")
2877 printk("lowmem_reserve[]:");
2878 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2879 printk(" %lu", zone
->lowmem_reserve
[i
]);
2883 for_each_populated_zone(zone
) {
2884 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2886 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2889 printk("%s: ", zone
->name
);
2891 spin_lock_irqsave(&zone
->lock
, flags
);
2892 for (order
= 0; order
< MAX_ORDER
; order
++) {
2893 nr
[order
] = zone
->free_area
[order
].nr_free
;
2894 total
+= nr
[order
] << order
;
2896 spin_unlock_irqrestore(&zone
->lock
, flags
);
2897 for (order
= 0; order
< MAX_ORDER
; order
++)
2898 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2899 printk("= %lukB\n", K(total
));
2902 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2904 show_swap_cache_info();
2907 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2909 zoneref
->zone
= zone
;
2910 zoneref
->zone_idx
= zone_idx(zone
);
2914 * Builds allocation fallback zone lists.
2916 * Add all populated zones of a node to the zonelist.
2918 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2919 int nr_zones
, enum zone_type zone_type
)
2923 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2928 zone
= pgdat
->node_zones
+ zone_type
;
2929 if (populated_zone(zone
)) {
2930 zoneref_set_zone(zone
,
2931 &zonelist
->_zonerefs
[nr_zones
++]);
2932 check_highest_zone(zone_type
);
2935 } while (zone_type
);
2942 * 0 = automatic detection of better ordering.
2943 * 1 = order by ([node] distance, -zonetype)
2944 * 2 = order by (-zonetype, [node] distance)
2946 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2947 * the same zonelist. So only NUMA can configure this param.
2949 #define ZONELIST_ORDER_DEFAULT 0
2950 #define ZONELIST_ORDER_NODE 1
2951 #define ZONELIST_ORDER_ZONE 2
2953 /* zonelist order in the kernel.
2954 * set_zonelist_order() will set this to NODE or ZONE.
2956 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2957 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2961 /* The value user specified ....changed by config */
2962 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2963 /* string for sysctl */
2964 #define NUMA_ZONELIST_ORDER_LEN 16
2965 char numa_zonelist_order
[16] = "default";
2968 * interface for configure zonelist ordering.
2969 * command line option "numa_zonelist_order"
2970 * = "[dD]efault - default, automatic configuration.
2971 * = "[nN]ode - order by node locality, then by zone within node
2972 * = "[zZ]one - order by zone, then by locality within zone
2975 static int __parse_numa_zonelist_order(char *s
)
2977 if (*s
== 'd' || *s
== 'D') {
2978 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2979 } else if (*s
== 'n' || *s
== 'N') {
2980 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2981 } else if (*s
== 'z' || *s
== 'Z') {
2982 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2985 "Ignoring invalid numa_zonelist_order value: "
2992 static __init
int setup_numa_zonelist_order(char *s
)
2999 ret
= __parse_numa_zonelist_order(s
);
3001 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3005 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3008 * sysctl handler for numa_zonelist_order
3010 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
3011 void __user
*buffer
, size_t *length
,
3014 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3016 static DEFINE_MUTEX(zl_order_mutex
);
3018 mutex_lock(&zl_order_mutex
);
3020 strcpy(saved_string
, (char*)table
->data
);
3021 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3025 int oldval
= user_zonelist_order
;
3026 if (__parse_numa_zonelist_order((char*)table
->data
)) {
3028 * bogus value. restore saved string
3030 strncpy((char*)table
->data
, saved_string
,
3031 NUMA_ZONELIST_ORDER_LEN
);
3032 user_zonelist_order
= oldval
;
3033 } else if (oldval
!= user_zonelist_order
) {
3034 mutex_lock(&zonelists_mutex
);
3035 build_all_zonelists(NULL
);
3036 mutex_unlock(&zonelists_mutex
);
3040 mutex_unlock(&zl_order_mutex
);
3045 #define MAX_NODE_LOAD (nr_online_nodes)
3046 static int node_load
[MAX_NUMNODES
];
3049 * find_next_best_node - find the next node that should appear in a given node's fallback list
3050 * @node: node whose fallback list we're appending
3051 * @used_node_mask: nodemask_t of already used nodes
3053 * We use a number of factors to determine which is the next node that should
3054 * appear on a given node's fallback list. The node should not have appeared
3055 * already in @node's fallback list, and it should be the next closest node
3056 * according to the distance array (which contains arbitrary distance values
3057 * from each node to each node in the system), and should also prefer nodes
3058 * with no CPUs, since presumably they'll have very little allocation pressure
3059 * on them otherwise.
3060 * It returns -1 if no node is found.
3062 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3065 int min_val
= INT_MAX
;
3067 const struct cpumask
*tmp
= cpumask_of_node(0);
3069 /* Use the local node if we haven't already */
3070 if (!node_isset(node
, *used_node_mask
)) {
3071 node_set(node
, *used_node_mask
);
3075 for_each_node_state(n
, N_HIGH_MEMORY
) {
3077 /* Don't want a node to appear more than once */
3078 if (node_isset(n
, *used_node_mask
))
3081 /* Use the distance array to find the distance */
3082 val
= node_distance(node
, n
);
3084 /* Penalize nodes under us ("prefer the next node") */
3087 /* Give preference to headless and unused nodes */
3088 tmp
= cpumask_of_node(n
);
3089 if (!cpumask_empty(tmp
))
3090 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
3092 /* Slight preference for less loaded node */
3093 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
3094 val
+= node_load
[n
];
3096 if (val
< min_val
) {
3103 node_set(best_node
, *used_node_mask
);
3110 * Build zonelists ordered by node and zones within node.
3111 * This results in maximum locality--normal zone overflows into local
3112 * DMA zone, if any--but risks exhausting DMA zone.
3114 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
3117 struct zonelist
*zonelist
;
3119 zonelist
= &pgdat
->node_zonelists
[0];
3120 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
3122 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3124 zonelist
->_zonerefs
[j
].zone
= NULL
;
3125 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3129 * Build gfp_thisnode zonelists
3131 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
3134 struct zonelist
*zonelist
;
3136 zonelist
= &pgdat
->node_zonelists
[1];
3137 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3138 zonelist
->_zonerefs
[j
].zone
= NULL
;
3139 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3143 * Build zonelists ordered by zone and nodes within zones.
3144 * This results in conserving DMA zone[s] until all Normal memory is
3145 * exhausted, but results in overflowing to remote node while memory
3146 * may still exist in local DMA zone.
3148 static int node_order
[MAX_NUMNODES
];
3150 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
3153 int zone_type
; /* needs to be signed */
3155 struct zonelist
*zonelist
;
3157 zonelist
= &pgdat
->node_zonelists
[0];
3159 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
3160 for (j
= 0; j
< nr_nodes
; j
++) {
3161 node
= node_order
[j
];
3162 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
3163 if (populated_zone(z
)) {
3165 &zonelist
->_zonerefs
[pos
++]);
3166 check_highest_zone(zone_type
);
3170 zonelist
->_zonerefs
[pos
].zone
= NULL
;
3171 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
3174 static int default_zonelist_order(void)
3177 unsigned long low_kmem_size
,total_size
;
3181 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3182 * If they are really small and used heavily, the system can fall
3183 * into OOM very easily.
3184 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3186 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3189 for_each_online_node(nid
) {
3190 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3191 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3192 if (populated_zone(z
)) {
3193 if (zone_type
< ZONE_NORMAL
)
3194 low_kmem_size
+= z
->present_pages
;
3195 total_size
+= z
->present_pages
;
3196 } else if (zone_type
== ZONE_NORMAL
) {
3198 * If any node has only lowmem, then node order
3199 * is preferred to allow kernel allocations
3200 * locally; otherwise, they can easily infringe
3201 * on other nodes when there is an abundance of
3202 * lowmem available to allocate from.
3204 return ZONELIST_ORDER_NODE
;
3208 if (!low_kmem_size
|| /* there are no DMA area. */
3209 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
3210 return ZONELIST_ORDER_NODE
;
3212 * look into each node's config.
3213 * If there is a node whose DMA/DMA32 memory is very big area on
3214 * local memory, NODE_ORDER may be suitable.
3216 average_size
= total_size
/
3217 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
3218 for_each_online_node(nid
) {
3221 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3222 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3223 if (populated_zone(z
)) {
3224 if (zone_type
< ZONE_NORMAL
)
3225 low_kmem_size
+= z
->present_pages
;
3226 total_size
+= z
->present_pages
;
3229 if (low_kmem_size
&&
3230 total_size
> average_size
&& /* ignore small node */
3231 low_kmem_size
> total_size
* 70/100)
3232 return ZONELIST_ORDER_NODE
;
3234 return ZONELIST_ORDER_ZONE
;
3237 static void set_zonelist_order(void)
3239 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
3240 current_zonelist_order
= default_zonelist_order();
3242 current_zonelist_order
= user_zonelist_order
;
3245 static void build_zonelists(pg_data_t
*pgdat
)
3249 nodemask_t used_mask
;
3250 int local_node
, prev_node
;
3251 struct zonelist
*zonelist
;
3252 int order
= current_zonelist_order
;
3254 /* initialize zonelists */
3255 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3256 zonelist
= pgdat
->node_zonelists
+ i
;
3257 zonelist
->_zonerefs
[0].zone
= NULL
;
3258 zonelist
->_zonerefs
[0].zone_idx
= 0;
3261 /* NUMA-aware ordering of nodes */
3262 local_node
= pgdat
->node_id
;
3263 load
= nr_online_nodes
;
3264 prev_node
= local_node
;
3265 nodes_clear(used_mask
);
3267 memset(node_order
, 0, sizeof(node_order
));
3270 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3271 int distance
= node_distance(local_node
, node
);
3274 * If another node is sufficiently far away then it is better
3275 * to reclaim pages in a zone before going off node.
3277 if (distance
> RECLAIM_DISTANCE
)
3278 zone_reclaim_mode
= 1;
3281 * We don't want to pressure a particular node.
3282 * So adding penalty to the first node in same
3283 * distance group to make it round-robin.
3285 if (distance
!= node_distance(local_node
, prev_node
))
3286 node_load
[node
] = load
;
3290 if (order
== ZONELIST_ORDER_NODE
)
3291 build_zonelists_in_node_order(pgdat
, node
);
3293 node_order
[j
++] = node
; /* remember order */
3296 if (order
== ZONELIST_ORDER_ZONE
) {
3297 /* calculate node order -- i.e., DMA last! */
3298 build_zonelists_in_zone_order(pgdat
, j
);
3301 build_thisnode_zonelists(pgdat
);
3304 /* Construct the zonelist performance cache - see further mmzone.h */
3305 static void build_zonelist_cache(pg_data_t
*pgdat
)
3307 struct zonelist
*zonelist
;
3308 struct zonelist_cache
*zlc
;
3311 zonelist
= &pgdat
->node_zonelists
[0];
3312 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3313 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3314 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3315 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3318 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3320 * Return node id of node used for "local" allocations.
3321 * I.e., first node id of first zone in arg node's generic zonelist.
3322 * Used for initializing percpu 'numa_mem', which is used primarily
3323 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3325 int local_memory_node(int node
)
3329 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3330 gfp_zone(GFP_KERNEL
),
3337 #else /* CONFIG_NUMA */
3339 static void set_zonelist_order(void)
3341 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3344 static void build_zonelists(pg_data_t
*pgdat
)
3346 int node
, local_node
;
3348 struct zonelist
*zonelist
;
3350 local_node
= pgdat
->node_id
;
3352 zonelist
= &pgdat
->node_zonelists
[0];
3353 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3356 * Now we build the zonelist so that it contains the zones
3357 * of all the other nodes.
3358 * We don't want to pressure a particular node, so when
3359 * building the zones for node N, we make sure that the
3360 * zones coming right after the local ones are those from
3361 * node N+1 (modulo N)
3363 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3364 if (!node_online(node
))
3366 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3369 for (node
= 0; node
< local_node
; node
++) {
3370 if (!node_online(node
))
3372 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3376 zonelist
->_zonerefs
[j
].zone
= NULL
;
3377 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3380 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3381 static void build_zonelist_cache(pg_data_t
*pgdat
)
3383 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3386 #endif /* CONFIG_NUMA */
3389 * Boot pageset table. One per cpu which is going to be used for all
3390 * zones and all nodes. The parameters will be set in such a way
3391 * that an item put on a list will immediately be handed over to
3392 * the buddy list. This is safe since pageset manipulation is done
3393 * with interrupts disabled.
3395 * The boot_pagesets must be kept even after bootup is complete for
3396 * unused processors and/or zones. They do play a role for bootstrapping
3397 * hotplugged processors.
3399 * zoneinfo_show() and maybe other functions do
3400 * not check if the processor is online before following the pageset pointer.
3401 * Other parts of the kernel may not check if the zone is available.
3403 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3404 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3405 static void setup_zone_pageset(struct zone
*zone
);
3408 * Global mutex to protect against size modification of zonelists
3409 * as well as to serialize pageset setup for the new populated zone.
3411 DEFINE_MUTEX(zonelists_mutex
);
3413 /* return values int ....just for stop_machine() */
3414 static __init_refok
int __build_all_zonelists(void *data
)
3420 memset(node_load
, 0, sizeof(node_load
));
3422 for_each_online_node(nid
) {
3423 pg_data_t
*pgdat
= NODE_DATA(nid
);
3425 build_zonelists(pgdat
);
3426 build_zonelist_cache(pgdat
);
3430 * Initialize the boot_pagesets that are going to be used
3431 * for bootstrapping processors. The real pagesets for
3432 * each zone will be allocated later when the per cpu
3433 * allocator is available.
3435 * boot_pagesets are used also for bootstrapping offline
3436 * cpus if the system is already booted because the pagesets
3437 * are needed to initialize allocators on a specific cpu too.
3438 * F.e. the percpu allocator needs the page allocator which
3439 * needs the percpu allocator in order to allocate its pagesets
3440 * (a chicken-egg dilemma).
3442 for_each_possible_cpu(cpu
) {
3443 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3445 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3447 * We now know the "local memory node" for each node--
3448 * i.e., the node of the first zone in the generic zonelist.
3449 * Set up numa_mem percpu variable for on-line cpus. During
3450 * boot, only the boot cpu should be on-line; we'll init the
3451 * secondary cpus' numa_mem as they come on-line. During
3452 * node/memory hotplug, we'll fixup all on-line cpus.
3454 if (cpu_online(cpu
))
3455 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3463 * Called with zonelists_mutex held always
3464 * unless system_state == SYSTEM_BOOTING.
3466 void __ref
build_all_zonelists(void *data
)
3468 set_zonelist_order();
3470 if (system_state
== SYSTEM_BOOTING
) {
3471 __build_all_zonelists(NULL
);
3472 mminit_verify_zonelist();
3473 cpuset_init_current_mems_allowed();
3475 /* we have to stop all cpus to guarantee there is no user
3477 #ifdef CONFIG_MEMORY_HOTPLUG
3479 setup_zone_pageset((struct zone
*)data
);
3481 stop_machine(__build_all_zonelists
, NULL
, NULL
);
3482 /* cpuset refresh routine should be here */
3484 vm_total_pages
= nr_free_pagecache_pages();
3486 * Disable grouping by mobility if the number of pages in the
3487 * system is too low to allow the mechanism to work. It would be
3488 * more accurate, but expensive to check per-zone. This check is
3489 * made on memory-hotadd so a system can start with mobility
3490 * disabled and enable it later
3492 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3493 page_group_by_mobility_disabled
= 1;
3495 page_group_by_mobility_disabled
= 0;
3497 printk("Built %i zonelists in %s order, mobility grouping %s. "
3498 "Total pages: %ld\n",
3500 zonelist_order_name
[current_zonelist_order
],
3501 page_group_by_mobility_disabled
? "off" : "on",
3504 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3509 * Helper functions to size the waitqueue hash table.
3510 * Essentially these want to choose hash table sizes sufficiently
3511 * large so that collisions trying to wait on pages are rare.
3512 * But in fact, the number of active page waitqueues on typical
3513 * systems is ridiculously low, less than 200. So this is even
3514 * conservative, even though it seems large.
3516 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3517 * waitqueues, i.e. the size of the waitq table given the number of pages.
3519 #define PAGES_PER_WAITQUEUE 256
3521 #ifndef CONFIG_MEMORY_HOTPLUG
3522 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3524 unsigned long size
= 1;
3526 pages
/= PAGES_PER_WAITQUEUE
;
3528 while (size
< pages
)
3532 * Once we have dozens or even hundreds of threads sleeping
3533 * on IO we've got bigger problems than wait queue collision.
3534 * Limit the size of the wait table to a reasonable size.
3536 size
= min(size
, 4096UL);
3538 return max(size
, 4UL);
3542 * A zone's size might be changed by hot-add, so it is not possible to determine
3543 * a suitable size for its wait_table. So we use the maximum size now.
3545 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3547 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3548 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3549 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3551 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3552 * or more by the traditional way. (See above). It equals:
3554 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3555 * ia64(16K page size) : = ( 8G + 4M)byte.
3556 * powerpc (64K page size) : = (32G +16M)byte.
3558 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3565 * This is an integer logarithm so that shifts can be used later
3566 * to extract the more random high bits from the multiplicative
3567 * hash function before the remainder is taken.
3569 static inline unsigned long wait_table_bits(unsigned long size
)
3574 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3577 * Check if a pageblock contains reserved pages
3579 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3583 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3584 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3591 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3592 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3593 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3594 * higher will lead to a bigger reserve which will get freed as contiguous
3595 * blocks as reclaim kicks in
3597 static void setup_zone_migrate_reserve(struct zone
*zone
)
3599 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3601 unsigned long block_migratetype
;
3605 * Get the start pfn, end pfn and the number of blocks to reserve
3606 * We have to be careful to be aligned to pageblock_nr_pages to
3607 * make sure that we always check pfn_valid for the first page in
3610 start_pfn
= zone
->zone_start_pfn
;
3611 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3612 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3613 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3617 * Reserve blocks are generally in place to help high-order atomic
3618 * allocations that are short-lived. A min_free_kbytes value that
3619 * would result in more than 2 reserve blocks for atomic allocations
3620 * is assumed to be in place to help anti-fragmentation for the
3621 * future allocation of hugepages at runtime.
3623 reserve
= min(2, reserve
);
3625 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3626 if (!pfn_valid(pfn
))
3628 page
= pfn_to_page(pfn
);
3630 /* Watch out for overlapping nodes */
3631 if (page_to_nid(page
) != zone_to_nid(zone
))
3634 block_migratetype
= get_pageblock_migratetype(page
);
3636 /* Only test what is necessary when the reserves are not met */
3639 * Blocks with reserved pages will never free, skip
3642 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3643 if (pageblock_is_reserved(pfn
, block_end_pfn
))
3646 /* If this block is reserved, account for it */
3647 if (block_migratetype
== MIGRATE_RESERVE
) {
3652 /* Suitable for reserving if this block is movable */
3653 if (block_migratetype
== MIGRATE_MOVABLE
) {
3654 set_pageblock_migratetype(page
,
3656 move_freepages_block(zone
, page
,
3664 * If the reserve is met and this is a previous reserved block,
3667 if (block_migratetype
== MIGRATE_RESERVE
) {
3668 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3669 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3675 * Initially all pages are reserved - free ones are freed
3676 * up by free_all_bootmem() once the early boot process is
3677 * done. Non-atomic initialization, single-pass.
3679 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3680 unsigned long start_pfn
, enum memmap_context context
)
3683 unsigned long end_pfn
= start_pfn
+ size
;
3687 if (highest_memmap_pfn
< end_pfn
- 1)
3688 highest_memmap_pfn
= end_pfn
- 1;
3690 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3691 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3693 * There can be holes in boot-time mem_map[]s
3694 * handed to this function. They do not
3695 * exist on hotplugged memory.
3697 if (context
== MEMMAP_EARLY
) {
3698 if (!early_pfn_valid(pfn
))
3700 if (!early_pfn_in_nid(pfn
, nid
))
3703 page
= pfn_to_page(pfn
);
3704 set_page_links(page
, zone
, nid
, pfn
);
3705 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3706 init_page_count(page
);
3707 reset_page_mapcount(page
);
3708 SetPageReserved(page
);
3710 * Mark the block movable so that blocks are reserved for
3711 * movable at startup. This will force kernel allocations
3712 * to reserve their blocks rather than leaking throughout
3713 * the address space during boot when many long-lived
3714 * kernel allocations are made. Later some blocks near
3715 * the start are marked MIGRATE_RESERVE by
3716 * setup_zone_migrate_reserve()
3718 * bitmap is created for zone's valid pfn range. but memmap
3719 * can be created for invalid pages (for alignment)
3720 * check here not to call set_pageblock_migratetype() against
3723 if ((z
->zone_start_pfn
<= pfn
)
3724 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3725 && !(pfn
& (pageblock_nr_pages
- 1)))
3726 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3728 INIT_LIST_HEAD(&page
->lru
);
3729 #ifdef WANT_PAGE_VIRTUAL
3730 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3731 if (!is_highmem_idx(zone
))
3732 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3737 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3740 for_each_migratetype_order(order
, t
) {
3741 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3742 zone
->free_area
[order
].nr_free
= 0;
3746 #ifndef __HAVE_ARCH_MEMMAP_INIT
3747 #define memmap_init(size, nid, zone, start_pfn) \
3748 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3751 static int zone_batchsize(struct zone
*zone
)
3757 * The per-cpu-pages pools are set to around 1000th of the
3758 * size of the zone. But no more than 1/2 of a meg.
3760 * OK, so we don't know how big the cache is. So guess.
3762 batch
= zone
->present_pages
/ 1024;
3763 if (batch
* PAGE_SIZE
> 512 * 1024)
3764 batch
= (512 * 1024) / PAGE_SIZE
;
3765 batch
/= 4; /* We effectively *= 4 below */
3770 * Clamp the batch to a 2^n - 1 value. Having a power
3771 * of 2 value was found to be more likely to have
3772 * suboptimal cache aliasing properties in some cases.
3774 * For example if 2 tasks are alternately allocating
3775 * batches of pages, one task can end up with a lot
3776 * of pages of one half of the possible page colors
3777 * and the other with pages of the other colors.
3779 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3784 /* The deferral and batching of frees should be suppressed under NOMMU
3787 * The problem is that NOMMU needs to be able to allocate large chunks
3788 * of contiguous memory as there's no hardware page translation to
3789 * assemble apparent contiguous memory from discontiguous pages.
3791 * Queueing large contiguous runs of pages for batching, however,
3792 * causes the pages to actually be freed in smaller chunks. As there
3793 * can be a significant delay between the individual batches being
3794 * recycled, this leads to the once large chunks of space being
3795 * fragmented and becoming unavailable for high-order allocations.
3801 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3803 struct per_cpu_pages
*pcp
;
3806 memset(p
, 0, sizeof(*p
));
3810 pcp
->high
= 6 * batch
;
3811 pcp
->batch
= max(1UL, 1 * batch
);
3812 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3813 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3817 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3818 * to the value high for the pageset p.
3821 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3824 struct per_cpu_pages
*pcp
;
3828 pcp
->batch
= max(1UL, high
/4);
3829 if ((high
/4) > (PAGE_SHIFT
* 8))
3830 pcp
->batch
= PAGE_SHIFT
* 8;
3833 static void setup_zone_pageset(struct zone
*zone
)
3837 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
3839 for_each_possible_cpu(cpu
) {
3840 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
3842 setup_pageset(pcp
, zone_batchsize(zone
));
3844 if (percpu_pagelist_fraction
)
3845 setup_pagelist_highmark(pcp
,
3846 (zone
->present_pages
/
3847 percpu_pagelist_fraction
));
3852 * Allocate per cpu pagesets and initialize them.
3853 * Before this call only boot pagesets were available.
3855 void __init
setup_per_cpu_pageset(void)
3859 for_each_populated_zone(zone
)
3860 setup_zone_pageset(zone
);
3863 static noinline __init_refok
3864 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3867 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3871 * The per-page waitqueue mechanism uses hashed waitqueues
3874 zone
->wait_table_hash_nr_entries
=
3875 wait_table_hash_nr_entries(zone_size_pages
);
3876 zone
->wait_table_bits
=
3877 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3878 alloc_size
= zone
->wait_table_hash_nr_entries
3879 * sizeof(wait_queue_head_t
);
3881 if (!slab_is_available()) {
3882 zone
->wait_table
= (wait_queue_head_t
*)
3883 alloc_bootmem_node_nopanic(pgdat
, alloc_size
);
3886 * This case means that a zone whose size was 0 gets new memory
3887 * via memory hot-add.
3888 * But it may be the case that a new node was hot-added. In
3889 * this case vmalloc() will not be able to use this new node's
3890 * memory - this wait_table must be initialized to use this new
3891 * node itself as well.
3892 * To use this new node's memory, further consideration will be
3895 zone
->wait_table
= vmalloc(alloc_size
);
3897 if (!zone
->wait_table
)
3900 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3901 init_waitqueue_head(zone
->wait_table
+ i
);
3906 static int __zone_pcp_update(void *data
)
3908 struct zone
*zone
= data
;
3910 unsigned long batch
= zone_batchsize(zone
), flags
;
3912 for_each_possible_cpu(cpu
) {
3913 struct per_cpu_pageset
*pset
;
3914 struct per_cpu_pages
*pcp
;
3916 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
3919 local_irq_save(flags
);
3921 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
3922 setup_pageset(pset
, batch
);
3923 local_irq_restore(flags
);
3928 void zone_pcp_update(struct zone
*zone
)
3930 stop_machine(__zone_pcp_update
, zone
, NULL
);
3933 static __meminit
void zone_pcp_init(struct zone
*zone
)
3936 * per cpu subsystem is not up at this point. The following code
3937 * relies on the ability of the linker to provide the
3938 * offset of a (static) per cpu variable into the per cpu area.
3940 zone
->pageset
= &boot_pageset
;
3942 if (zone
->present_pages
)
3943 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
3944 zone
->name
, zone
->present_pages
,
3945 zone_batchsize(zone
));
3948 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3949 unsigned long zone_start_pfn
,
3951 enum memmap_context context
)
3953 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3955 ret
= zone_wait_table_init(zone
, size
);
3958 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3960 zone
->zone_start_pfn
= zone_start_pfn
;
3962 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3963 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3965 (unsigned long)zone_idx(zone
),
3966 zone_start_pfn
, (zone_start_pfn
+ size
));
3968 zone_init_free_lists(zone
);
3973 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
3974 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3976 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3977 * Architectures may implement their own version but if add_active_range()
3978 * was used and there are no special requirements, this is a convenient
3981 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3983 unsigned long start_pfn
, end_pfn
;
3986 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
3987 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3989 /* This is a memory hole */
3992 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3994 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3998 nid
= __early_pfn_to_nid(pfn
);
4001 /* just returns 0 */
4005 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4006 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
4010 nid
= __early_pfn_to_nid(pfn
);
4011 if (nid
>= 0 && nid
!= node
)
4018 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
4019 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4020 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
4022 * If an architecture guarantees that all ranges registered with
4023 * add_active_ranges() contain no holes and may be freed, this
4024 * this function may be used instead of calling free_bootmem() manually.
4026 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4028 unsigned long start_pfn
, end_pfn
;
4031 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4032 start_pfn
= min(start_pfn
, max_low_pfn
);
4033 end_pfn
= min(end_pfn
, max_low_pfn
);
4035 if (start_pfn
< end_pfn
)
4036 free_bootmem_node(NODE_DATA(this_nid
),
4037 PFN_PHYS(start_pfn
),
4038 (end_pfn
- start_pfn
) << PAGE_SHIFT
);
4043 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4044 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4046 * If an architecture guarantees that all ranges registered with
4047 * add_active_ranges() contain no holes and may be freed, this
4048 * function may be used instead of calling memory_present() manually.
4050 void __init
sparse_memory_present_with_active_regions(int nid
)
4052 unsigned long start_pfn
, end_pfn
;
4055 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4056 memory_present(this_nid
, start_pfn
, end_pfn
);
4060 * get_pfn_range_for_nid - Return the start and end page frames for a node
4061 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4062 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4063 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4065 * It returns the start and end page frame of a node based on information
4066 * provided by an arch calling add_active_range(). If called for a node
4067 * with no available memory, a warning is printed and the start and end
4070 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4071 unsigned long *start_pfn
, unsigned long *end_pfn
)
4073 unsigned long this_start_pfn
, this_end_pfn
;
4079 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4080 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4081 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4084 if (*start_pfn
== -1UL)
4089 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4090 * assumption is made that zones within a node are ordered in monotonic
4091 * increasing memory addresses so that the "highest" populated zone is used
4093 static void __init
find_usable_zone_for_movable(void)
4096 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4097 if (zone_index
== ZONE_MOVABLE
)
4100 if (arch_zone_highest_possible_pfn
[zone_index
] >
4101 arch_zone_lowest_possible_pfn
[zone_index
])
4105 VM_BUG_ON(zone_index
== -1);
4106 movable_zone
= zone_index
;
4110 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4111 * because it is sized independent of architecture. Unlike the other zones,
4112 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4113 * in each node depending on the size of each node and how evenly kernelcore
4114 * is distributed. This helper function adjusts the zone ranges
4115 * provided by the architecture for a given node by using the end of the
4116 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4117 * zones within a node are in order of monotonic increases memory addresses
4119 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4120 unsigned long zone_type
,
4121 unsigned long node_start_pfn
,
4122 unsigned long node_end_pfn
,
4123 unsigned long *zone_start_pfn
,
4124 unsigned long *zone_end_pfn
)
4126 /* Only adjust if ZONE_MOVABLE is on this node */
4127 if (zone_movable_pfn
[nid
]) {
4128 /* Size ZONE_MOVABLE */
4129 if (zone_type
== ZONE_MOVABLE
) {
4130 *zone_start_pfn
= zone_movable_pfn
[nid
];
4131 *zone_end_pfn
= min(node_end_pfn
,
4132 arch_zone_highest_possible_pfn
[movable_zone
]);
4134 /* Adjust for ZONE_MOVABLE starting within this range */
4135 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4136 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4137 *zone_end_pfn
= zone_movable_pfn
[nid
];
4139 /* Check if this whole range is within ZONE_MOVABLE */
4140 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4141 *zone_start_pfn
= *zone_end_pfn
;
4146 * Return the number of pages a zone spans in a node, including holes
4147 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4149 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4150 unsigned long zone_type
,
4151 unsigned long *ignored
)
4153 unsigned long node_start_pfn
, node_end_pfn
;
4154 unsigned long zone_start_pfn
, zone_end_pfn
;
4156 /* Get the start and end of the node and zone */
4157 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4158 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4159 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4160 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4161 node_start_pfn
, node_end_pfn
,
4162 &zone_start_pfn
, &zone_end_pfn
);
4164 /* Check that this node has pages within the zone's required range */
4165 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4168 /* Move the zone boundaries inside the node if necessary */
4169 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4170 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4172 /* Return the spanned pages */
4173 return zone_end_pfn
- zone_start_pfn
;
4177 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4178 * then all holes in the requested range will be accounted for.
4180 unsigned long __meminit
__absent_pages_in_range(int nid
,
4181 unsigned long range_start_pfn
,
4182 unsigned long range_end_pfn
)
4184 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4185 unsigned long start_pfn
, end_pfn
;
4188 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4189 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4190 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4191 nr_absent
-= end_pfn
- start_pfn
;
4197 * absent_pages_in_range - Return number of page frames in holes within a range
4198 * @start_pfn: The start PFN to start searching for holes
4199 * @end_pfn: The end PFN to stop searching for holes
4201 * It returns the number of pages frames in memory holes within a range.
4203 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4204 unsigned long end_pfn
)
4206 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4209 /* Return the number of page frames in holes in a zone on a node */
4210 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4211 unsigned long zone_type
,
4212 unsigned long *ignored
)
4214 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4215 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4216 unsigned long node_start_pfn
, node_end_pfn
;
4217 unsigned long zone_start_pfn
, zone_end_pfn
;
4219 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4220 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4221 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4223 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4224 node_start_pfn
, node_end_pfn
,
4225 &zone_start_pfn
, &zone_end_pfn
);
4226 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4229 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4230 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4231 unsigned long zone_type
,
4232 unsigned long *zones_size
)
4234 return zones_size
[zone_type
];
4237 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4238 unsigned long zone_type
,
4239 unsigned long *zholes_size
)
4244 return zholes_size
[zone_type
];
4247 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4249 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4250 unsigned long *zones_size
, unsigned long *zholes_size
)
4252 unsigned long realtotalpages
, totalpages
= 0;
4255 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4256 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4258 pgdat
->node_spanned_pages
= totalpages
;
4260 realtotalpages
= totalpages
;
4261 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4263 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4265 pgdat
->node_present_pages
= realtotalpages
;
4266 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4270 #ifndef CONFIG_SPARSEMEM
4272 * Calculate the size of the zone->blockflags rounded to an unsigned long
4273 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4274 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4275 * round what is now in bits to nearest long in bits, then return it in
4278 static unsigned long __init
usemap_size(unsigned long zonesize
)
4280 unsigned long usemapsize
;
4282 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4283 usemapsize
= usemapsize
>> pageblock_order
;
4284 usemapsize
*= NR_PAGEBLOCK_BITS
;
4285 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4287 return usemapsize
/ 8;
4290 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4291 struct zone
*zone
, unsigned long zonesize
)
4293 unsigned long usemapsize
= usemap_size(zonesize
);
4294 zone
->pageblock_flags
= NULL
;
4296 zone
->pageblock_flags
= alloc_bootmem_node_nopanic(pgdat
,
4300 static inline void setup_usemap(struct pglist_data
*pgdat
,
4301 struct zone
*zone
, unsigned long zonesize
) {}
4302 #endif /* CONFIG_SPARSEMEM */
4304 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4306 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4307 static inline void __init
set_pageblock_order(void)
4311 /* Check that pageblock_nr_pages has not already been setup */
4312 if (pageblock_order
)
4315 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4316 order
= HUGETLB_PAGE_ORDER
;
4318 order
= MAX_ORDER
- 1;
4321 * Assume the largest contiguous order of interest is a huge page.
4322 * This value may be variable depending on boot parameters on IA64 and
4325 pageblock_order
= order
;
4327 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4330 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4331 * is unused as pageblock_order is set at compile-time. See
4332 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4335 static inline void set_pageblock_order(void)
4339 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4342 * Set up the zone data structures:
4343 * - mark all pages reserved
4344 * - mark all memory queues empty
4345 * - clear the memory bitmaps
4347 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4348 unsigned long *zones_size
, unsigned long *zholes_size
)
4351 int nid
= pgdat
->node_id
;
4352 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4355 pgdat_resize_init(pgdat
);
4356 pgdat
->nr_zones
= 0;
4357 init_waitqueue_head(&pgdat
->kswapd_wait
);
4358 pgdat
->kswapd_max_order
= 0;
4359 pgdat_page_cgroup_init(pgdat
);
4361 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4362 struct zone
*zone
= pgdat
->node_zones
+ j
;
4363 unsigned long size
, realsize
, memmap_pages
;
4365 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4366 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
4370 * Adjust realsize so that it accounts for how much memory
4371 * is used by this zone for memmap. This affects the watermark
4372 * and per-cpu initialisations
4375 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
4376 if (realsize
>= memmap_pages
) {
4377 realsize
-= memmap_pages
;
4380 " %s zone: %lu pages used for memmap\n",
4381 zone_names
[j
], memmap_pages
);
4384 " %s zone: %lu pages exceeds realsize %lu\n",
4385 zone_names
[j
], memmap_pages
, realsize
);
4387 /* Account for reserved pages */
4388 if (j
== 0 && realsize
> dma_reserve
) {
4389 realsize
-= dma_reserve
;
4390 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4391 zone_names
[0], dma_reserve
);
4394 if (!is_highmem_idx(j
))
4395 nr_kernel_pages
+= realsize
;
4396 nr_all_pages
+= realsize
;
4398 zone
->spanned_pages
= size
;
4399 zone
->present_pages
= realsize
;
4402 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
4404 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
4406 zone
->name
= zone_names
[j
];
4407 spin_lock_init(&zone
->lock
);
4408 spin_lock_init(&zone
->lru_lock
);
4409 zone_seqlock_init(zone
);
4410 zone
->zone_pgdat
= pgdat
;
4412 zone_pcp_init(zone
);
4413 lruvec_init(&zone
->lruvec
, zone
);
4414 zap_zone_vm_stats(zone
);
4419 set_pageblock_order();
4420 setup_usemap(pgdat
, zone
, size
);
4421 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4422 size
, MEMMAP_EARLY
);
4424 memmap_init(size
, nid
, j
, zone_start_pfn
);
4425 zone_start_pfn
+= size
;
4429 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4431 /* Skip empty nodes */
4432 if (!pgdat
->node_spanned_pages
)
4435 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4436 /* ia64 gets its own node_mem_map, before this, without bootmem */
4437 if (!pgdat
->node_mem_map
) {
4438 unsigned long size
, start
, end
;
4442 * The zone's endpoints aren't required to be MAX_ORDER
4443 * aligned but the node_mem_map endpoints must be in order
4444 * for the buddy allocator to function correctly.
4446 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4447 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4448 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4449 size
= (end
- start
) * sizeof(struct page
);
4450 map
= alloc_remap(pgdat
->node_id
, size
);
4452 map
= alloc_bootmem_node_nopanic(pgdat
, size
);
4453 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4455 #ifndef CONFIG_NEED_MULTIPLE_NODES
4457 * With no DISCONTIG, the global mem_map is just set as node 0's
4459 if (pgdat
== NODE_DATA(0)) {
4460 mem_map
= NODE_DATA(0)->node_mem_map
;
4461 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4462 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4463 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4464 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4467 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4470 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4471 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4473 pg_data_t
*pgdat
= NODE_DATA(nid
);
4475 pgdat
->node_id
= nid
;
4476 pgdat
->node_start_pfn
= node_start_pfn
;
4477 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4479 alloc_node_mem_map(pgdat
);
4480 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4481 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4482 nid
, (unsigned long)pgdat
,
4483 (unsigned long)pgdat
->node_mem_map
);
4486 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4489 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4491 #if MAX_NUMNODES > 1
4493 * Figure out the number of possible node ids.
4495 static void __init
setup_nr_node_ids(void)
4498 unsigned int highest
= 0;
4500 for_each_node_mask(node
, node_possible_map
)
4502 nr_node_ids
= highest
+ 1;
4505 static inline void setup_nr_node_ids(void)
4511 * node_map_pfn_alignment - determine the maximum internode alignment
4513 * This function should be called after node map is populated and sorted.
4514 * It calculates the maximum power of two alignment which can distinguish
4517 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4518 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4519 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4520 * shifted, 1GiB is enough and this function will indicate so.
4522 * This is used to test whether pfn -> nid mapping of the chosen memory
4523 * model has fine enough granularity to avoid incorrect mapping for the
4524 * populated node map.
4526 * Returns the determined alignment in pfn's. 0 if there is no alignment
4527 * requirement (single node).
4529 unsigned long __init
node_map_pfn_alignment(void)
4531 unsigned long accl_mask
= 0, last_end
= 0;
4532 unsigned long start
, end
, mask
;
4536 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4537 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4544 * Start with a mask granular enough to pin-point to the
4545 * start pfn and tick off bits one-by-one until it becomes
4546 * too coarse to separate the current node from the last.
4548 mask
= ~((1 << __ffs(start
)) - 1);
4549 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4552 /* accumulate all internode masks */
4556 /* convert mask to number of pages */
4557 return ~accl_mask
+ 1;
4560 /* Find the lowest pfn for a node */
4561 static unsigned long __init
find_min_pfn_for_node(int nid
)
4563 unsigned long min_pfn
= ULONG_MAX
;
4564 unsigned long start_pfn
;
4567 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4568 min_pfn
= min(min_pfn
, start_pfn
);
4570 if (min_pfn
== ULONG_MAX
) {
4572 "Could not find start_pfn for node %d\n", nid
);
4580 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4582 * It returns the minimum PFN based on information provided via
4583 * add_active_range().
4585 unsigned long __init
find_min_pfn_with_active_regions(void)
4587 return find_min_pfn_for_node(MAX_NUMNODES
);
4591 * early_calculate_totalpages()
4592 * Sum pages in active regions for movable zone.
4593 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4595 static unsigned long __init
early_calculate_totalpages(void)
4597 unsigned long totalpages
= 0;
4598 unsigned long start_pfn
, end_pfn
;
4601 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
4602 unsigned long pages
= end_pfn
- start_pfn
;
4604 totalpages
+= pages
;
4606 node_set_state(nid
, N_HIGH_MEMORY
);
4612 * Find the PFN the Movable zone begins in each node. Kernel memory
4613 * is spread evenly between nodes as long as the nodes have enough
4614 * memory. When they don't, some nodes will have more kernelcore than
4617 static void __init
find_zone_movable_pfns_for_nodes(void)
4620 unsigned long usable_startpfn
;
4621 unsigned long kernelcore_node
, kernelcore_remaining
;
4622 /* save the state before borrow the nodemask */
4623 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4624 unsigned long totalpages
= early_calculate_totalpages();
4625 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4628 * If movablecore was specified, calculate what size of
4629 * kernelcore that corresponds so that memory usable for
4630 * any allocation type is evenly spread. If both kernelcore
4631 * and movablecore are specified, then the value of kernelcore
4632 * will be used for required_kernelcore if it's greater than
4633 * what movablecore would have allowed.
4635 if (required_movablecore
) {
4636 unsigned long corepages
;
4639 * Round-up so that ZONE_MOVABLE is at least as large as what
4640 * was requested by the user
4642 required_movablecore
=
4643 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4644 corepages
= totalpages
- required_movablecore
;
4646 required_kernelcore
= max(required_kernelcore
, corepages
);
4649 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4650 if (!required_kernelcore
)
4653 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4654 find_usable_zone_for_movable();
4655 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4658 /* Spread kernelcore memory as evenly as possible throughout nodes */
4659 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4660 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4661 unsigned long start_pfn
, end_pfn
;
4664 * Recalculate kernelcore_node if the division per node
4665 * now exceeds what is necessary to satisfy the requested
4666 * amount of memory for the kernel
4668 if (required_kernelcore
< kernelcore_node
)
4669 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4672 * As the map is walked, we track how much memory is usable
4673 * by the kernel using kernelcore_remaining. When it is
4674 * 0, the rest of the node is usable by ZONE_MOVABLE
4676 kernelcore_remaining
= kernelcore_node
;
4678 /* Go through each range of PFNs within this node */
4679 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4680 unsigned long size_pages
;
4682 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
4683 if (start_pfn
>= end_pfn
)
4686 /* Account for what is only usable for kernelcore */
4687 if (start_pfn
< usable_startpfn
) {
4688 unsigned long kernel_pages
;
4689 kernel_pages
= min(end_pfn
, usable_startpfn
)
4692 kernelcore_remaining
-= min(kernel_pages
,
4693 kernelcore_remaining
);
4694 required_kernelcore
-= min(kernel_pages
,
4695 required_kernelcore
);
4697 /* Continue if range is now fully accounted */
4698 if (end_pfn
<= usable_startpfn
) {
4701 * Push zone_movable_pfn to the end so
4702 * that if we have to rebalance
4703 * kernelcore across nodes, we will
4704 * not double account here
4706 zone_movable_pfn
[nid
] = end_pfn
;
4709 start_pfn
= usable_startpfn
;
4713 * The usable PFN range for ZONE_MOVABLE is from
4714 * start_pfn->end_pfn. Calculate size_pages as the
4715 * number of pages used as kernelcore
4717 size_pages
= end_pfn
- start_pfn
;
4718 if (size_pages
> kernelcore_remaining
)
4719 size_pages
= kernelcore_remaining
;
4720 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4723 * Some kernelcore has been met, update counts and
4724 * break if the kernelcore for this node has been
4727 required_kernelcore
-= min(required_kernelcore
,
4729 kernelcore_remaining
-= size_pages
;
4730 if (!kernelcore_remaining
)
4736 * If there is still required_kernelcore, we do another pass with one
4737 * less node in the count. This will push zone_movable_pfn[nid] further
4738 * along on the nodes that still have memory until kernelcore is
4742 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4745 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4746 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4747 zone_movable_pfn
[nid
] =
4748 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4751 /* restore the node_state */
4752 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4755 /* Any regular memory on that node ? */
4756 static void check_for_regular_memory(pg_data_t
*pgdat
)
4758 #ifdef CONFIG_HIGHMEM
4759 enum zone_type zone_type
;
4761 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4762 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4763 if (zone
->present_pages
) {
4764 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4772 * free_area_init_nodes - Initialise all pg_data_t and zone data
4773 * @max_zone_pfn: an array of max PFNs for each zone
4775 * This will call free_area_init_node() for each active node in the system.
4776 * Using the page ranges provided by add_active_range(), the size of each
4777 * zone in each node and their holes is calculated. If the maximum PFN
4778 * between two adjacent zones match, it is assumed that the zone is empty.
4779 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4780 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4781 * starts where the previous one ended. For example, ZONE_DMA32 starts
4782 * at arch_max_dma_pfn.
4784 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4786 unsigned long start_pfn
, end_pfn
;
4789 /* Record where the zone boundaries are */
4790 memset(arch_zone_lowest_possible_pfn
, 0,
4791 sizeof(arch_zone_lowest_possible_pfn
));
4792 memset(arch_zone_highest_possible_pfn
, 0,
4793 sizeof(arch_zone_highest_possible_pfn
));
4794 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4795 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4796 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4797 if (i
== ZONE_MOVABLE
)
4799 arch_zone_lowest_possible_pfn
[i
] =
4800 arch_zone_highest_possible_pfn
[i
-1];
4801 arch_zone_highest_possible_pfn
[i
] =
4802 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4804 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4805 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4807 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4808 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4809 find_zone_movable_pfns_for_nodes();
4811 /* Print out the zone ranges */
4812 printk("Zone ranges:\n");
4813 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4814 if (i
== ZONE_MOVABLE
)
4816 printk(KERN_CONT
" %-8s ", zone_names
[i
]);
4817 if (arch_zone_lowest_possible_pfn
[i
] ==
4818 arch_zone_highest_possible_pfn
[i
])
4819 printk(KERN_CONT
"empty\n");
4821 printk(KERN_CONT
"[mem %0#10lx-%0#10lx]\n",
4822 arch_zone_lowest_possible_pfn
[i
] << PAGE_SHIFT
,
4823 (arch_zone_highest_possible_pfn
[i
]
4824 << PAGE_SHIFT
) - 1);
4827 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4828 printk("Movable zone start for each node\n");
4829 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4830 if (zone_movable_pfn
[i
])
4831 printk(" Node %d: %#010lx\n", i
,
4832 zone_movable_pfn
[i
] << PAGE_SHIFT
);
4835 /* Print out the early_node_map[] */
4836 printk("Early memory node ranges\n");
4837 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4838 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid
,
4839 start_pfn
<< PAGE_SHIFT
, (end_pfn
<< PAGE_SHIFT
) - 1);
4841 /* Initialise every node */
4842 mminit_verify_pageflags_layout();
4843 setup_nr_node_ids();
4844 for_each_online_node(nid
) {
4845 pg_data_t
*pgdat
= NODE_DATA(nid
);
4846 free_area_init_node(nid
, NULL
,
4847 find_min_pfn_for_node(nid
), NULL
);
4849 /* Any memory on that node */
4850 if (pgdat
->node_present_pages
)
4851 node_set_state(nid
, N_HIGH_MEMORY
);
4852 check_for_regular_memory(pgdat
);
4856 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4858 unsigned long long coremem
;
4862 coremem
= memparse(p
, &p
);
4863 *core
= coremem
>> PAGE_SHIFT
;
4865 /* Paranoid check that UL is enough for the coremem value */
4866 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4872 * kernelcore=size sets the amount of memory for use for allocations that
4873 * cannot be reclaimed or migrated.
4875 static int __init
cmdline_parse_kernelcore(char *p
)
4877 return cmdline_parse_core(p
, &required_kernelcore
);
4881 * movablecore=size sets the amount of memory for use for allocations that
4882 * can be reclaimed or migrated.
4884 static int __init
cmdline_parse_movablecore(char *p
)
4886 return cmdline_parse_core(p
, &required_movablecore
);
4889 early_param("kernelcore", cmdline_parse_kernelcore
);
4890 early_param("movablecore", cmdline_parse_movablecore
);
4892 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4895 * set_dma_reserve - set the specified number of pages reserved in the first zone
4896 * @new_dma_reserve: The number of pages to mark reserved
4898 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4899 * In the DMA zone, a significant percentage may be consumed by kernel image
4900 * and other unfreeable allocations which can skew the watermarks badly. This
4901 * function may optionally be used to account for unfreeable pages in the
4902 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4903 * smaller per-cpu batchsize.
4905 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4907 dma_reserve
= new_dma_reserve
;
4910 void __init
free_area_init(unsigned long *zones_size
)
4912 free_area_init_node(0, zones_size
,
4913 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4916 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4917 unsigned long action
, void *hcpu
)
4919 int cpu
= (unsigned long)hcpu
;
4921 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4922 lru_add_drain_cpu(cpu
);
4926 * Spill the event counters of the dead processor
4927 * into the current processors event counters.
4928 * This artificially elevates the count of the current
4931 vm_events_fold_cpu(cpu
);
4934 * Zero the differential counters of the dead processor
4935 * so that the vm statistics are consistent.
4937 * This is only okay since the processor is dead and cannot
4938 * race with what we are doing.
4940 refresh_cpu_vm_stats(cpu
);
4945 void __init
page_alloc_init(void)
4947 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4951 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4952 * or min_free_kbytes changes.
4954 static void calculate_totalreserve_pages(void)
4956 struct pglist_data
*pgdat
;
4957 unsigned long reserve_pages
= 0;
4958 enum zone_type i
, j
;
4960 for_each_online_pgdat(pgdat
) {
4961 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4962 struct zone
*zone
= pgdat
->node_zones
+ i
;
4963 unsigned long max
= 0;
4965 /* Find valid and maximum lowmem_reserve in the zone */
4966 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4967 if (zone
->lowmem_reserve
[j
] > max
)
4968 max
= zone
->lowmem_reserve
[j
];
4971 /* we treat the high watermark as reserved pages. */
4972 max
+= high_wmark_pages(zone
);
4974 if (max
> zone
->present_pages
)
4975 max
= zone
->present_pages
;
4976 reserve_pages
+= max
;
4978 * Lowmem reserves are not available to
4979 * GFP_HIGHUSER page cache allocations and
4980 * kswapd tries to balance zones to their high
4981 * watermark. As a result, neither should be
4982 * regarded as dirtyable memory, to prevent a
4983 * situation where reclaim has to clean pages
4984 * in order to balance the zones.
4986 zone
->dirty_balance_reserve
= max
;
4989 dirty_balance_reserve
= reserve_pages
;
4990 totalreserve_pages
= reserve_pages
;
4994 * setup_per_zone_lowmem_reserve - called whenever
4995 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4996 * has a correct pages reserved value, so an adequate number of
4997 * pages are left in the zone after a successful __alloc_pages().
4999 static void setup_per_zone_lowmem_reserve(void)
5001 struct pglist_data
*pgdat
;
5002 enum zone_type j
, idx
;
5004 for_each_online_pgdat(pgdat
) {
5005 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5006 struct zone
*zone
= pgdat
->node_zones
+ j
;
5007 unsigned long present_pages
= zone
->present_pages
;
5009 zone
->lowmem_reserve
[j
] = 0;
5013 struct zone
*lower_zone
;
5017 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5018 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5020 lower_zone
= pgdat
->node_zones
+ idx
;
5021 lower_zone
->lowmem_reserve
[j
] = present_pages
/
5022 sysctl_lowmem_reserve_ratio
[idx
];
5023 present_pages
+= lower_zone
->present_pages
;
5028 /* update totalreserve_pages */
5029 calculate_totalreserve_pages();
5032 static void __setup_per_zone_wmarks(void)
5034 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5035 unsigned long lowmem_pages
= 0;
5037 unsigned long flags
;
5039 /* Calculate total number of !ZONE_HIGHMEM pages */
5040 for_each_zone(zone
) {
5041 if (!is_highmem(zone
))
5042 lowmem_pages
+= zone
->present_pages
;
5045 for_each_zone(zone
) {
5048 spin_lock_irqsave(&zone
->lock
, flags
);
5049 tmp
= (u64
)pages_min
* zone
->present_pages
;
5050 do_div(tmp
, lowmem_pages
);
5051 if (is_highmem(zone
)) {
5053 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5054 * need highmem pages, so cap pages_min to a small
5057 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5058 * deltas controls asynch page reclaim, and so should
5059 * not be capped for highmem.
5063 min_pages
= zone
->present_pages
/ 1024;
5064 if (min_pages
< SWAP_CLUSTER_MAX
)
5065 min_pages
= SWAP_CLUSTER_MAX
;
5066 if (min_pages
> 128)
5068 zone
->watermark
[WMARK_MIN
] = min_pages
;
5071 * If it's a lowmem zone, reserve a number of pages
5072 * proportionate to the zone's size.
5074 zone
->watermark
[WMARK_MIN
] = tmp
;
5077 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
5078 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
5080 zone
->watermark
[WMARK_MIN
] += cma_wmark_pages(zone
);
5081 zone
->watermark
[WMARK_LOW
] += cma_wmark_pages(zone
);
5082 zone
->watermark
[WMARK_HIGH
] += cma_wmark_pages(zone
);
5084 setup_zone_migrate_reserve(zone
);
5085 spin_unlock_irqrestore(&zone
->lock
, flags
);
5088 /* update totalreserve_pages */
5089 calculate_totalreserve_pages();
5093 * setup_per_zone_wmarks - called when min_free_kbytes changes
5094 * or when memory is hot-{added|removed}
5096 * Ensures that the watermark[min,low,high] values for each zone are set
5097 * correctly with respect to min_free_kbytes.
5099 void setup_per_zone_wmarks(void)
5101 mutex_lock(&zonelists_mutex
);
5102 __setup_per_zone_wmarks();
5103 mutex_unlock(&zonelists_mutex
);
5107 * The inactive anon list should be small enough that the VM never has to
5108 * do too much work, but large enough that each inactive page has a chance
5109 * to be referenced again before it is swapped out.
5111 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5112 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5113 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5114 * the anonymous pages are kept on the inactive list.
5117 * memory ratio inactive anon
5118 * -------------------------------------
5127 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
5129 unsigned int gb
, ratio
;
5131 /* Zone size in gigabytes */
5132 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
5134 ratio
= int_sqrt(10 * gb
);
5138 zone
->inactive_ratio
= ratio
;
5141 static void __meminit
setup_per_zone_inactive_ratio(void)
5146 calculate_zone_inactive_ratio(zone
);
5150 * Initialise min_free_kbytes.
5152 * For small machines we want it small (128k min). For large machines
5153 * we want it large (64MB max). But it is not linear, because network
5154 * bandwidth does not increase linearly with machine size. We use
5156 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5157 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5173 int __meminit
init_per_zone_wmark_min(void)
5175 unsigned long lowmem_kbytes
;
5177 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5179 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5180 if (min_free_kbytes
< 128)
5181 min_free_kbytes
= 128;
5182 if (min_free_kbytes
> 65536)
5183 min_free_kbytes
= 65536;
5184 setup_per_zone_wmarks();
5185 refresh_zone_stat_thresholds();
5186 setup_per_zone_lowmem_reserve();
5187 setup_per_zone_inactive_ratio();
5190 module_init(init_per_zone_wmark_min
)
5193 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5194 * that we can call two helper functions whenever min_free_kbytes
5197 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5198 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5200 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5202 setup_per_zone_wmarks();
5207 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5208 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5213 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5218 zone
->min_unmapped_pages
= (zone
->present_pages
*
5219 sysctl_min_unmapped_ratio
) / 100;
5223 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5224 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5229 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5234 zone
->min_slab_pages
= (zone
->present_pages
*
5235 sysctl_min_slab_ratio
) / 100;
5241 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5242 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5243 * whenever sysctl_lowmem_reserve_ratio changes.
5245 * The reserve ratio obviously has absolutely no relation with the
5246 * minimum watermarks. The lowmem reserve ratio can only make sense
5247 * if in function of the boot time zone sizes.
5249 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5250 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5252 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5253 setup_per_zone_lowmem_reserve();
5258 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5259 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5260 * can have before it gets flushed back to buddy allocator.
5263 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5264 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5270 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5271 if (!write
|| (ret
< 0))
5273 for_each_populated_zone(zone
) {
5274 for_each_possible_cpu(cpu
) {
5276 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5277 setup_pagelist_highmark(
5278 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5284 int hashdist
= HASHDIST_DEFAULT
;
5287 static int __init
set_hashdist(char *str
)
5291 hashdist
= simple_strtoul(str
, &str
, 0);
5294 __setup("hashdist=", set_hashdist
);
5298 * allocate a large system hash table from bootmem
5299 * - it is assumed that the hash table must contain an exact power-of-2
5300 * quantity of entries
5301 * - limit is the number of hash buckets, not the total allocation size
5303 void *__init
alloc_large_system_hash(const char *tablename
,
5304 unsigned long bucketsize
,
5305 unsigned long numentries
,
5308 unsigned int *_hash_shift
,
5309 unsigned int *_hash_mask
,
5310 unsigned long low_limit
,
5311 unsigned long high_limit
)
5313 unsigned long long max
= high_limit
;
5314 unsigned long log2qty
, size
;
5317 /* allow the kernel cmdline to have a say */
5319 /* round applicable memory size up to nearest megabyte */
5320 numentries
= nr_kernel_pages
;
5321 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5322 numentries
>>= 20 - PAGE_SHIFT
;
5323 numentries
<<= 20 - PAGE_SHIFT
;
5325 /* limit to 1 bucket per 2^scale bytes of low memory */
5326 if (scale
> PAGE_SHIFT
)
5327 numentries
>>= (scale
- PAGE_SHIFT
);
5329 numentries
<<= (PAGE_SHIFT
- scale
);
5331 /* Make sure we've got at least a 0-order allocation.. */
5332 if (unlikely(flags
& HASH_SMALL
)) {
5333 /* Makes no sense without HASH_EARLY */
5334 WARN_ON(!(flags
& HASH_EARLY
));
5335 if (!(numentries
>> *_hash_shift
)) {
5336 numentries
= 1UL << *_hash_shift
;
5337 BUG_ON(!numentries
);
5339 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5340 numentries
= PAGE_SIZE
/ bucketsize
;
5342 numentries
= roundup_pow_of_two(numentries
);
5344 /* limit allocation size to 1/16 total memory by default */
5346 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5347 do_div(max
, bucketsize
);
5349 max
= min(max
, 0x80000000ULL
);
5351 if (numentries
< low_limit
)
5352 numentries
= low_limit
;
5353 if (numentries
> max
)
5356 log2qty
= ilog2(numentries
);
5359 size
= bucketsize
<< log2qty
;
5360 if (flags
& HASH_EARLY
)
5361 table
= alloc_bootmem_nopanic(size
);
5363 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5366 * If bucketsize is not a power-of-two, we may free
5367 * some pages at the end of hash table which
5368 * alloc_pages_exact() automatically does
5370 if (get_order(size
) < MAX_ORDER
) {
5371 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5372 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5375 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5378 panic("Failed to allocate %s hash table\n", tablename
);
5380 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5383 ilog2(size
) - PAGE_SHIFT
,
5387 *_hash_shift
= log2qty
;
5389 *_hash_mask
= (1 << log2qty
) - 1;
5394 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5395 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5398 #ifdef CONFIG_SPARSEMEM
5399 return __pfn_to_section(pfn
)->pageblock_flags
;
5401 return zone
->pageblock_flags
;
5402 #endif /* CONFIG_SPARSEMEM */
5405 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5407 #ifdef CONFIG_SPARSEMEM
5408 pfn
&= (PAGES_PER_SECTION
-1);
5409 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5411 pfn
= pfn
- zone
->zone_start_pfn
;
5412 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5413 #endif /* CONFIG_SPARSEMEM */
5417 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5418 * @page: The page within the block of interest
5419 * @start_bitidx: The first bit of interest to retrieve
5420 * @end_bitidx: The last bit of interest
5421 * returns pageblock_bits flags
5423 unsigned long get_pageblock_flags_group(struct page
*page
,
5424 int start_bitidx
, int end_bitidx
)
5427 unsigned long *bitmap
;
5428 unsigned long pfn
, bitidx
;
5429 unsigned long flags
= 0;
5430 unsigned long value
= 1;
5432 zone
= page_zone(page
);
5433 pfn
= page_to_pfn(page
);
5434 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5435 bitidx
= pfn_to_bitidx(zone
, pfn
);
5437 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5438 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5445 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5446 * @page: The page within the block of interest
5447 * @start_bitidx: The first bit of interest
5448 * @end_bitidx: The last bit of interest
5449 * @flags: The flags to set
5451 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5452 int start_bitidx
, int end_bitidx
)
5455 unsigned long *bitmap
;
5456 unsigned long pfn
, bitidx
;
5457 unsigned long value
= 1;
5459 zone
= page_zone(page
);
5460 pfn
= page_to_pfn(page
);
5461 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5462 bitidx
= pfn_to_bitidx(zone
, pfn
);
5463 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5464 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5466 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5468 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5470 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5474 * This is designed as sub function...plz see page_isolation.c also.
5475 * set/clear page block's type to be ISOLATE.
5476 * page allocater never alloc memory from ISOLATE block.
5480 __count_immobile_pages(struct zone
*zone
, struct page
*page
, int count
)
5482 unsigned long pfn
, iter
, found
;
5486 * For avoiding noise data, lru_add_drain_all() should be called
5487 * If ZONE_MOVABLE, the zone never contains immobile pages
5489 if (zone_idx(zone
) == ZONE_MOVABLE
)
5491 mt
= get_pageblock_migratetype(page
);
5492 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
5495 pfn
= page_to_pfn(page
);
5496 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
5497 unsigned long check
= pfn
+ iter
;
5499 if (!pfn_valid_within(check
))
5502 page
= pfn_to_page(check
);
5503 if (!page_count(page
)) {
5504 if (PageBuddy(page
))
5505 iter
+= (1 << page_order(page
)) - 1;
5511 * If there are RECLAIMABLE pages, we need to check it.
5512 * But now, memory offline itself doesn't call shrink_slab()
5513 * and it still to be fixed.
5516 * If the page is not RAM, page_count()should be 0.
5517 * we don't need more check. This is an _used_ not-movable page.
5519 * The problematic thing here is PG_reserved pages. PG_reserved
5520 * is set to both of a memory hole page and a _used_ kernel
5529 bool is_pageblock_removable_nolock(struct page
*page
)
5535 * We have to be careful here because we are iterating over memory
5536 * sections which are not zone aware so we might end up outside of
5537 * the zone but still within the section.
5538 * We have to take care about the node as well. If the node is offline
5539 * its NODE_DATA will be NULL - see page_zone.
5541 if (!node_online(page_to_nid(page
)))
5544 zone
= page_zone(page
);
5545 pfn
= page_to_pfn(page
);
5546 if (zone
->zone_start_pfn
> pfn
||
5547 zone
->zone_start_pfn
+ zone
->spanned_pages
<= pfn
)
5550 return __count_immobile_pages(zone
, page
, 0);
5553 int set_migratetype_isolate(struct page
*page
)
5556 unsigned long flags
, pfn
;
5557 struct memory_isolate_notify arg
;
5561 zone
= page_zone(page
);
5563 spin_lock_irqsave(&zone
->lock
, flags
);
5565 pfn
= page_to_pfn(page
);
5566 arg
.start_pfn
= pfn
;
5567 arg
.nr_pages
= pageblock_nr_pages
;
5568 arg
.pages_found
= 0;
5571 * It may be possible to isolate a pageblock even if the
5572 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5573 * notifier chain is used by balloon drivers to return the
5574 * number of pages in a range that are held by the balloon
5575 * driver to shrink memory. If all the pages are accounted for
5576 * by balloons, are free, or on the LRU, isolation can continue.
5577 * Later, for example, when memory hotplug notifier runs, these
5578 * pages reported as "can be isolated" should be isolated(freed)
5579 * by the balloon driver through the memory notifier chain.
5581 notifier_ret
= memory_isolate_notify(MEM_ISOLATE_COUNT
, &arg
);
5582 notifier_ret
= notifier_to_errno(notifier_ret
);
5586 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5587 * We just check MOVABLE pages.
5589 if (__count_immobile_pages(zone
, page
, arg
.pages_found
))
5593 * immobile means "not-on-lru" paes. If immobile is larger than
5594 * removable-by-driver pages reported by notifier, we'll fail.
5599 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
5600 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
5603 spin_unlock_irqrestore(&zone
->lock
, flags
);
5609 void unset_migratetype_isolate(struct page
*page
, unsigned migratetype
)
5612 unsigned long flags
;
5613 zone
= page_zone(page
);
5614 spin_lock_irqsave(&zone
->lock
, flags
);
5615 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
5617 set_pageblock_migratetype(page
, migratetype
);
5618 move_freepages_block(zone
, page
, migratetype
);
5620 spin_unlock_irqrestore(&zone
->lock
, flags
);
5625 static unsigned long pfn_max_align_down(unsigned long pfn
)
5627 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5628 pageblock_nr_pages
) - 1);
5631 static unsigned long pfn_max_align_up(unsigned long pfn
)
5633 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5634 pageblock_nr_pages
));
5637 static struct page
*
5638 __alloc_contig_migrate_alloc(struct page
*page
, unsigned long private,
5641 gfp_t gfp_mask
= GFP_USER
| __GFP_MOVABLE
;
5643 if (PageHighMem(page
))
5644 gfp_mask
|= __GFP_HIGHMEM
;
5646 return alloc_page(gfp_mask
);
5649 /* [start, end) must belong to a single zone. */
5650 static int __alloc_contig_migrate_range(unsigned long start
, unsigned long end
)
5652 /* This function is based on compact_zone() from compaction.c. */
5654 unsigned long pfn
= start
;
5655 unsigned int tries
= 0;
5658 struct compact_control cc
= {
5659 .nr_migratepages
= 0,
5661 .zone
= page_zone(pfn_to_page(start
)),
5664 INIT_LIST_HEAD(&cc
.migratepages
);
5666 migrate_prep_local();
5668 while (pfn
< end
|| !list_empty(&cc
.migratepages
)) {
5669 if (fatal_signal_pending(current
)) {
5674 if (list_empty(&cc
.migratepages
)) {
5675 cc
.nr_migratepages
= 0;
5676 pfn
= isolate_migratepages_range(cc
.zone
, &cc
,
5683 } else if (++tries
== 5) {
5684 ret
= ret
< 0 ? ret
: -EBUSY
;
5688 ret
= migrate_pages(&cc
.migratepages
,
5689 __alloc_contig_migrate_alloc
,
5690 0, false, MIGRATE_SYNC
);
5693 putback_lru_pages(&cc
.migratepages
);
5694 return ret
> 0 ? 0 : ret
;
5698 * Update zone's cma pages counter used for watermark level calculation.
5700 static inline void __update_cma_watermarks(struct zone
*zone
, int count
)
5702 unsigned long flags
;
5703 spin_lock_irqsave(&zone
->lock
, flags
);
5704 zone
->min_cma_pages
+= count
;
5705 spin_unlock_irqrestore(&zone
->lock
, flags
);
5706 setup_per_zone_wmarks();
5710 * Trigger memory pressure bump to reclaim some pages in order to be able to
5711 * allocate 'count' pages in single page units. Does similar work as
5712 *__alloc_pages_slowpath() function.
5714 static int __reclaim_pages(struct zone
*zone
, gfp_t gfp_mask
, int count
)
5716 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
5717 struct zonelist
*zonelist
= node_zonelist(0, gfp_mask
);
5718 int did_some_progress
= 0;
5722 * Increase level of watermarks to force kswapd do his job
5723 * to stabilise at new watermark level.
5725 __update_cma_watermarks(zone
, count
);
5727 /* Obey watermarks as if the page was being allocated */
5728 while (!zone_watermark_ok(zone
, 0, low_wmark_pages(zone
), 0, 0)) {
5729 wake_all_kswapd(order
, zonelist
, high_zoneidx
, zone_idx(zone
));
5731 did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
5733 if (!did_some_progress
) {
5734 /* Exhausted what can be done so it's blamo time */
5735 out_of_memory(zonelist
, gfp_mask
, order
, NULL
, false);
5739 /* Restore original watermark levels. */
5740 __update_cma_watermarks(zone
, -count
);
5746 * alloc_contig_range() -- tries to allocate given range of pages
5747 * @start: start PFN to allocate
5748 * @end: one-past-the-last PFN to allocate
5749 * @migratetype: migratetype of the underlaying pageblocks (either
5750 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
5751 * in range must have the same migratetype and it must
5752 * be either of the two.
5754 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
5755 * aligned, however it's the caller's responsibility to guarantee that
5756 * we are the only thread that changes migrate type of pageblocks the
5759 * The PFN range must belong to a single zone.
5761 * Returns zero on success or negative error code. On success all
5762 * pages which PFN is in [start, end) are allocated for the caller and
5763 * need to be freed with free_contig_range().
5765 int alloc_contig_range(unsigned long start
, unsigned long end
,
5766 unsigned migratetype
)
5768 struct zone
*zone
= page_zone(pfn_to_page(start
));
5769 unsigned long outer_start
, outer_end
;
5773 * What we do here is we mark all pageblocks in range as
5774 * MIGRATE_ISOLATE. Because pageblock and max order pages may
5775 * have different sizes, and due to the way page allocator
5776 * work, we align the range to biggest of the two pages so
5777 * that page allocator won't try to merge buddies from
5778 * different pageblocks and change MIGRATE_ISOLATE to some
5779 * other migration type.
5781 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
5782 * migrate the pages from an unaligned range (ie. pages that
5783 * we are interested in). This will put all the pages in
5784 * range back to page allocator as MIGRATE_ISOLATE.
5786 * When this is done, we take the pages in range from page
5787 * allocator removing them from the buddy system. This way
5788 * page allocator will never consider using them.
5790 * This lets us mark the pageblocks back as
5791 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
5792 * aligned range but not in the unaligned, original range are
5793 * put back to page allocator so that buddy can use them.
5796 ret
= start_isolate_page_range(pfn_max_align_down(start
),
5797 pfn_max_align_up(end
), migratetype
);
5801 ret
= __alloc_contig_migrate_range(start
, end
);
5806 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
5807 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
5808 * more, all pages in [start, end) are free in page allocator.
5809 * What we are going to do is to allocate all pages from
5810 * [start, end) (that is remove them from page allocator).
5812 * The only problem is that pages at the beginning and at the
5813 * end of interesting range may be not aligned with pages that
5814 * page allocator holds, ie. they can be part of higher order
5815 * pages. Because of this, we reserve the bigger range and
5816 * once this is done free the pages we are not interested in.
5818 * We don't have to hold zone->lock here because the pages are
5819 * isolated thus they won't get removed from buddy.
5822 lru_add_drain_all();
5826 outer_start
= start
;
5827 while (!PageBuddy(pfn_to_page(outer_start
))) {
5828 if (++order
>= MAX_ORDER
) {
5832 outer_start
&= ~0UL << order
;
5835 /* Make sure the range is really isolated. */
5836 if (test_pages_isolated(outer_start
, end
)) {
5837 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
5844 * Reclaim enough pages to make sure that contiguous allocation
5845 * will not starve the system.
5847 __reclaim_pages(zone
, GFP_HIGHUSER_MOVABLE
, end
-start
);
5849 /* Grab isolated pages from freelists. */
5850 outer_end
= isolate_freepages_range(outer_start
, end
);
5856 /* Free head and tail (if any) */
5857 if (start
!= outer_start
)
5858 free_contig_range(outer_start
, start
- outer_start
);
5859 if (end
!= outer_end
)
5860 free_contig_range(end
, outer_end
- end
);
5863 undo_isolate_page_range(pfn_max_align_down(start
),
5864 pfn_max_align_up(end
), migratetype
);
5868 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
5870 for (; nr_pages
--; ++pfn
)
5871 __free_page(pfn_to_page(pfn
));
5875 #ifdef CONFIG_MEMORY_HOTREMOVE
5877 * All pages in the range must be isolated before calling this.
5880 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
5886 unsigned long flags
;
5887 /* find the first valid pfn */
5888 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
5893 zone
= page_zone(pfn_to_page(pfn
));
5894 spin_lock_irqsave(&zone
->lock
, flags
);
5896 while (pfn
< end_pfn
) {
5897 if (!pfn_valid(pfn
)) {
5901 page
= pfn_to_page(pfn
);
5902 BUG_ON(page_count(page
));
5903 BUG_ON(!PageBuddy(page
));
5904 order
= page_order(page
);
5905 #ifdef CONFIG_DEBUG_VM
5906 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
5907 pfn
, 1 << order
, end_pfn
);
5909 list_del(&page
->lru
);
5910 rmv_page_order(page
);
5911 zone
->free_area
[order
].nr_free
--;
5912 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
5914 for (i
= 0; i
< (1 << order
); i
++)
5915 SetPageReserved((page
+i
));
5916 pfn
+= (1 << order
);
5918 spin_unlock_irqrestore(&zone
->lock
, flags
);
5922 #ifdef CONFIG_MEMORY_FAILURE
5923 bool is_free_buddy_page(struct page
*page
)
5925 struct zone
*zone
= page_zone(page
);
5926 unsigned long pfn
= page_to_pfn(page
);
5927 unsigned long flags
;
5930 spin_lock_irqsave(&zone
->lock
, flags
);
5931 for (order
= 0; order
< MAX_ORDER
; order
++) {
5932 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
5934 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
5937 spin_unlock_irqrestore(&zone
->lock
, flags
);
5939 return order
< MAX_ORDER
;
5943 static const struct trace_print_flags pageflag_names
[] = {
5944 {1UL << PG_locked
, "locked" },
5945 {1UL << PG_error
, "error" },
5946 {1UL << PG_referenced
, "referenced" },
5947 {1UL << PG_uptodate
, "uptodate" },
5948 {1UL << PG_dirty
, "dirty" },
5949 {1UL << PG_lru
, "lru" },
5950 {1UL << PG_active
, "active" },
5951 {1UL << PG_slab
, "slab" },
5952 {1UL << PG_owner_priv_1
, "owner_priv_1" },
5953 {1UL << PG_arch_1
, "arch_1" },
5954 {1UL << PG_reserved
, "reserved" },
5955 {1UL << PG_private
, "private" },
5956 {1UL << PG_private_2
, "private_2" },
5957 {1UL << PG_writeback
, "writeback" },
5958 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5959 {1UL << PG_head
, "head" },
5960 {1UL << PG_tail
, "tail" },
5962 {1UL << PG_compound
, "compound" },
5964 {1UL << PG_swapcache
, "swapcache" },
5965 {1UL << PG_mappedtodisk
, "mappedtodisk" },
5966 {1UL << PG_reclaim
, "reclaim" },
5967 {1UL << PG_swapbacked
, "swapbacked" },
5968 {1UL << PG_unevictable
, "unevictable" },
5970 {1UL << PG_mlocked
, "mlocked" },
5972 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5973 {1UL << PG_uncached
, "uncached" },
5975 #ifdef CONFIG_MEMORY_FAILURE
5976 {1UL << PG_hwpoison
, "hwpoison" },
5978 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5979 {1UL << PG_compound_lock
, "compound_lock" },
5983 static void dump_page_flags(unsigned long flags
)
5985 const char *delim
= "";
5989 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names
) != __NR_PAGEFLAGS
);
5991 printk(KERN_ALERT
"page flags: %#lx(", flags
);
5993 /* remove zone id */
5994 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
5996 for (i
= 0; i
< ARRAY_SIZE(pageflag_names
) && flags
; i
++) {
5998 mask
= pageflag_names
[i
].mask
;
5999 if ((flags
& mask
) != mask
)
6003 printk("%s%s", delim
, pageflag_names
[i
].name
);
6007 /* check for left over flags */
6009 printk("%s%#lx", delim
, flags
);
6014 void dump_page(struct page
*page
)
6017 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6018 page
, atomic_read(&page
->_count
), page_mapcount(page
),
6019 page
->mapping
, page
->index
);
6020 dump_page_flags(page
->flags
);
6021 mem_cgroup_print_bad_page(page
);