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/compaction.h>
55 #include <trace/events/kmem.h>
56 #include <linux/ftrace_event.h>
57 #include <linux/memcontrol.h>
58 #include <linux/prefetch.h>
59 #include <linux/migrate.h>
60 #include <linux/page-debug-flags.h>
62 #include <asm/tlbflush.h>
63 #include <asm/div64.h>
66 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
67 DEFINE_PER_CPU(int, numa_node
);
68 EXPORT_PER_CPU_SYMBOL(numa_node
);
71 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
73 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
74 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
75 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
76 * defined in <linux/topology.h>.
78 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
79 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
83 * Array of node states.
85 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
86 [N_POSSIBLE
] = NODE_MASK_ALL
,
87 [N_ONLINE
] = { { [0] = 1UL } },
89 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
91 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
93 [N_CPU
] = { { [0] = 1UL } },
96 EXPORT_SYMBOL(node_states
);
98 unsigned long totalram_pages __read_mostly
;
99 unsigned long totalreserve_pages __read_mostly
;
101 * When calculating the number of globally allowed dirty pages, there
102 * is a certain number of per-zone reserves that should not be
103 * considered dirtyable memory. This is the sum of those reserves
104 * over all existing zones that contribute dirtyable memory.
106 unsigned long dirty_balance_reserve __read_mostly
;
108 int percpu_pagelist_fraction
;
109 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
111 #ifdef CONFIG_PM_SLEEP
113 * The following functions are used by the suspend/hibernate code to temporarily
114 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
115 * while devices are suspended. To avoid races with the suspend/hibernate code,
116 * they should always be called with pm_mutex held (gfp_allowed_mask also should
117 * only be modified with pm_mutex held, unless the suspend/hibernate code is
118 * guaranteed not to run in parallel with that modification).
121 static gfp_t saved_gfp_mask
;
123 void pm_restore_gfp_mask(void)
125 WARN_ON(!mutex_is_locked(&pm_mutex
));
126 if (saved_gfp_mask
) {
127 gfp_allowed_mask
= saved_gfp_mask
;
132 void pm_restrict_gfp_mask(void)
134 WARN_ON(!mutex_is_locked(&pm_mutex
));
135 WARN_ON(saved_gfp_mask
);
136 saved_gfp_mask
= gfp_allowed_mask
;
137 gfp_allowed_mask
&= ~GFP_IOFS
;
140 bool pm_suspended_storage(void)
142 if ((gfp_allowed_mask
& GFP_IOFS
) == GFP_IOFS
)
146 #endif /* CONFIG_PM_SLEEP */
148 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
149 int pageblock_order __read_mostly
;
152 static void __free_pages_ok(struct page
*page
, unsigned int order
);
155 * results with 256, 32 in the lowmem_reserve sysctl:
156 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
157 * 1G machine -> (16M dma, 784M normal, 224M high)
158 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
159 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
160 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
162 * TBD: should special case ZONE_DMA32 machines here - in those we normally
163 * don't need any ZONE_NORMAL reservation
165 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
166 #ifdef CONFIG_ZONE_DMA
169 #ifdef CONFIG_ZONE_DMA32
172 #ifdef CONFIG_HIGHMEM
178 EXPORT_SYMBOL(totalram_pages
);
180 static char * const zone_names
[MAX_NR_ZONES
] = {
181 #ifdef CONFIG_ZONE_DMA
184 #ifdef CONFIG_ZONE_DMA32
188 #ifdef CONFIG_HIGHMEM
194 int min_free_kbytes
= 1024;
196 static unsigned long __meminitdata nr_kernel_pages
;
197 static unsigned long __meminitdata nr_all_pages
;
198 static unsigned long __meminitdata dma_reserve
;
200 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
201 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
202 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
203 static unsigned long __initdata required_kernelcore
;
204 static unsigned long __initdata required_movablecore
;
205 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
207 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
209 EXPORT_SYMBOL(movable_zone
);
210 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
213 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
214 int nr_online_nodes __read_mostly
= 1;
215 EXPORT_SYMBOL(nr_node_ids
);
216 EXPORT_SYMBOL(nr_online_nodes
);
219 int page_group_by_mobility_disabled __read_mostly
;
223 * Don't use set_pageblock_migratetype(page, MIGRATE_ISOLATE) directly.
224 * Instead, use {un}set_pageblock_isolate.
226 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
229 if (unlikely(page_group_by_mobility_disabled
))
230 migratetype
= MIGRATE_UNMOVABLE
;
232 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
233 PB_migrate
, PB_migrate_end
);
236 bool oom_killer_disabled __read_mostly
;
238 #ifdef CONFIG_DEBUG_VM
239 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
243 unsigned long pfn
= page_to_pfn(page
);
246 seq
= zone_span_seqbegin(zone
);
247 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
249 else if (pfn
< zone
->zone_start_pfn
)
251 } while (zone_span_seqretry(zone
, seq
));
256 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
258 if (!pfn_valid_within(page_to_pfn(page
)))
260 if (zone
!= page_zone(page
))
266 * Temporary debugging check for pages not lying within a given zone.
268 static int bad_range(struct zone
*zone
, struct page
*page
)
270 if (page_outside_zone_boundaries(zone
, page
))
272 if (!page_is_consistent(zone
, page
))
278 static inline int bad_range(struct zone
*zone
, struct page
*page
)
284 static void bad_page(struct page
*page
)
286 static unsigned long resume
;
287 static unsigned long nr_shown
;
288 static unsigned long nr_unshown
;
290 /* Don't complain about poisoned pages */
291 if (PageHWPoison(page
)) {
292 reset_page_mapcount(page
); /* remove PageBuddy */
297 * Allow a burst of 60 reports, then keep quiet for that minute;
298 * or allow a steady drip of one report per second.
300 if (nr_shown
== 60) {
301 if (time_before(jiffies
, resume
)) {
307 "BUG: Bad page state: %lu messages suppressed\n",
314 resume
= jiffies
+ 60 * HZ
;
316 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
317 current
->comm
, page_to_pfn(page
));
323 /* Leave bad fields for debug, except PageBuddy could make trouble */
324 reset_page_mapcount(page
); /* remove PageBuddy */
325 add_taint(TAINT_BAD_PAGE
);
329 * Higher-order pages are called "compound pages". They are structured thusly:
331 * The first PAGE_SIZE page is called the "head page".
333 * The remaining PAGE_SIZE pages are called "tail pages".
335 * All pages have PG_compound set. All tail pages have their ->first_page
336 * pointing at the head page.
338 * The first tail page's ->lru.next holds the address of the compound page's
339 * put_page() function. Its ->lru.prev holds the order of allocation.
340 * This usage means that zero-order pages may not be compound.
343 static void free_compound_page(struct page
*page
)
345 __free_pages_ok(page
, compound_order(page
));
348 void prep_compound_page(struct page
*page
, unsigned long order
)
351 int nr_pages
= 1 << order
;
353 set_compound_page_dtor(page
, free_compound_page
);
354 set_compound_order(page
, order
);
356 for (i
= 1; i
< nr_pages
; i
++) {
357 struct page
*p
= page
+ i
;
359 set_page_count(p
, 0);
360 p
->first_page
= page
;
364 /* update __split_huge_page_refcount if you change this function */
365 static int destroy_compound_page(struct page
*page
, unsigned long order
)
368 int nr_pages
= 1 << order
;
371 if (unlikely(compound_order(page
) != order
) ||
372 unlikely(!PageHead(page
))) {
377 __ClearPageHead(page
);
379 for (i
= 1; i
< nr_pages
; i
++) {
380 struct page
*p
= page
+ i
;
382 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
392 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
397 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
398 * and __GFP_HIGHMEM from hard or soft interrupt context.
400 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
401 for (i
= 0; i
< (1 << order
); i
++)
402 clear_highpage(page
+ i
);
405 #ifdef CONFIG_DEBUG_PAGEALLOC
406 unsigned int _debug_guardpage_minorder
;
408 static int __init
debug_guardpage_minorder_setup(char *buf
)
412 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
413 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
416 _debug_guardpage_minorder
= res
;
417 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
420 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
422 static inline void set_page_guard_flag(struct page
*page
)
424 __set_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
427 static inline void clear_page_guard_flag(struct page
*page
)
429 __clear_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
432 static inline void set_page_guard_flag(struct page
*page
) { }
433 static inline void clear_page_guard_flag(struct page
*page
) { }
436 static inline void set_page_order(struct page
*page
, int order
)
438 set_page_private(page
, order
);
439 __SetPageBuddy(page
);
442 static inline void rmv_page_order(struct page
*page
)
444 __ClearPageBuddy(page
);
445 set_page_private(page
, 0);
449 * Locate the struct page for both the matching buddy in our
450 * pair (buddy1) and the combined O(n+1) page they form (page).
452 * 1) Any buddy B1 will have an order O twin B2 which satisfies
453 * the following equation:
455 * For example, if the starting buddy (buddy2) is #8 its order
457 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
459 * 2) Any buddy B will have an order O+1 parent P which
460 * satisfies the following equation:
463 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
465 static inline unsigned long
466 __find_buddy_index(unsigned long page_idx
, unsigned int order
)
468 return page_idx
^ (1 << order
);
472 * This function checks whether a page is free && is the buddy
473 * we can do coalesce a page and its buddy if
474 * (a) the buddy is not in a hole &&
475 * (b) the buddy is in the buddy system &&
476 * (c) a page and its buddy have the same order &&
477 * (d) a page and its buddy are in the same zone.
479 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
480 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
482 * For recording page's order, we use page_private(page).
484 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
487 if (!pfn_valid_within(page_to_pfn(buddy
)))
490 if (page_zone_id(page
) != page_zone_id(buddy
))
493 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
494 VM_BUG_ON(page_count(buddy
) != 0);
498 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
499 VM_BUG_ON(page_count(buddy
) != 0);
506 * Freeing function for a buddy system allocator.
508 * The concept of a buddy system is to maintain direct-mapped table
509 * (containing bit values) for memory blocks of various "orders".
510 * The bottom level table contains the map for the smallest allocatable
511 * units of memory (here, pages), and each level above it describes
512 * pairs of units from the levels below, hence, "buddies".
513 * At a high level, all that happens here is marking the table entry
514 * at the bottom level available, and propagating the changes upward
515 * as necessary, plus some accounting needed to play nicely with other
516 * parts of the VM system.
517 * At each level, we keep a list of pages, which are heads of continuous
518 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
519 * order is recorded in page_private(page) field.
520 * So when we are allocating or freeing one, we can derive the state of the
521 * other. That is, if we allocate a small block, and both were
522 * free, the remainder of the region must be split into blocks.
523 * If a block is freed, and its buddy is also free, then this
524 * triggers coalescing into a block of larger size.
529 static inline void __free_one_page(struct page
*page
,
530 struct zone
*zone
, unsigned int order
,
533 unsigned long page_idx
;
534 unsigned long combined_idx
;
535 unsigned long uninitialized_var(buddy_idx
);
538 if (unlikely(PageCompound(page
)))
539 if (unlikely(destroy_compound_page(page
, order
)))
542 VM_BUG_ON(migratetype
== -1);
544 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
546 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
547 VM_BUG_ON(bad_range(zone
, page
));
549 while (order
< MAX_ORDER
-1) {
550 buddy_idx
= __find_buddy_index(page_idx
, order
);
551 buddy
= page
+ (buddy_idx
- page_idx
);
552 if (!page_is_buddy(page
, buddy
, order
))
555 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
556 * merge with it and move up one order.
558 if (page_is_guard(buddy
)) {
559 clear_page_guard_flag(buddy
);
560 set_page_private(page
, 0);
561 __mod_zone_freepage_state(zone
, 1 << order
,
564 list_del(&buddy
->lru
);
565 zone
->free_area
[order
].nr_free
--;
566 rmv_page_order(buddy
);
568 combined_idx
= buddy_idx
& page_idx
;
569 page
= page
+ (combined_idx
- page_idx
);
570 page_idx
= combined_idx
;
573 set_page_order(page
, order
);
576 * If this is not the largest possible page, check if the buddy
577 * of the next-highest order is free. If it is, it's possible
578 * that pages are being freed that will coalesce soon. In case,
579 * that is happening, add the free page to the tail of the list
580 * so it's less likely to be used soon and more likely to be merged
581 * as a higher order page
583 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
584 struct page
*higher_page
, *higher_buddy
;
585 combined_idx
= buddy_idx
& page_idx
;
586 higher_page
= page
+ (combined_idx
- page_idx
);
587 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
588 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
589 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
590 list_add_tail(&page
->lru
,
591 &zone
->free_area
[order
].free_list
[migratetype
]);
596 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
598 zone
->free_area
[order
].nr_free
++;
602 * free_page_mlock() -- clean up attempts to free and mlocked() page.
603 * Page should not be on lru, so no need to fix that up.
604 * free_pages_check() will verify...
606 static inline void free_page_mlock(struct page
*page
)
608 __dec_zone_page_state(page
, NR_MLOCK
);
609 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
612 static inline int free_pages_check(struct page
*page
)
614 if (unlikely(page_mapcount(page
) |
615 (page
->mapping
!= NULL
) |
616 (atomic_read(&page
->_count
) != 0) |
617 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
) |
618 (mem_cgroup_bad_page_check(page
)))) {
622 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
623 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
628 * Frees a number of pages from the PCP lists
629 * Assumes all pages on list are in same zone, and of same order.
630 * count is the number of pages to free.
632 * If the zone was previously in an "all pages pinned" state then look to
633 * see if this freeing clears that state.
635 * And clear the zone's pages_scanned counter, to hold off the "all pages are
636 * pinned" detection logic.
638 static void free_pcppages_bulk(struct zone
*zone
, int count
,
639 struct per_cpu_pages
*pcp
)
645 spin_lock(&zone
->lock
);
646 zone
->all_unreclaimable
= 0;
647 zone
->pages_scanned
= 0;
651 struct list_head
*list
;
654 * Remove pages from lists in a round-robin fashion. A
655 * batch_free count is maintained that is incremented when an
656 * empty list is encountered. This is so more pages are freed
657 * off fuller lists instead of spinning excessively around empty
662 if (++migratetype
== MIGRATE_PCPTYPES
)
664 list
= &pcp
->lists
[migratetype
];
665 } while (list_empty(list
));
667 /* This is the only non-empty list. Free them all. */
668 if (batch_free
== MIGRATE_PCPTYPES
)
669 batch_free
= to_free
;
672 int mt
; /* migratetype of the to-be-freed page */
674 page
= list_entry(list
->prev
, struct page
, lru
);
675 /* must delete as __free_one_page list manipulates */
676 list_del(&page
->lru
);
677 mt
= page_private(page
);
678 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
679 __free_one_page(page
, zone
, 0, mt
);
680 trace_mm_page_pcpu_drain(page
, 0, mt
);
681 if (is_migrate_cma(mt
))
682 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
, 1);
683 } while (--to_free
&& --batch_free
&& !list_empty(list
));
685 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
);
686 spin_unlock(&zone
->lock
);
689 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
692 spin_lock(&zone
->lock
);
693 zone
->all_unreclaimable
= 0;
694 zone
->pages_scanned
= 0;
696 __free_one_page(page
, zone
, order
, migratetype
);
697 if (unlikely(migratetype
!= MIGRATE_ISOLATE
))
698 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
699 spin_unlock(&zone
->lock
);
702 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
707 trace_mm_page_free(page
, order
);
708 kmemcheck_free_shadow(page
, order
);
711 page
->mapping
= NULL
;
712 for (i
= 0; i
< (1 << order
); i
++)
713 bad
+= free_pages_check(page
+ i
);
717 if (!PageHighMem(page
)) {
718 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
719 debug_check_no_obj_freed(page_address(page
),
722 arch_free_page(page
, order
);
723 kernel_map_pages(page
, 1 << order
, 0);
728 static void __free_pages_ok(struct page
*page
, unsigned int order
)
731 int wasMlocked
= __TestClearPageMlocked(page
);
733 if (!free_pages_prepare(page
, order
))
736 local_irq_save(flags
);
737 if (unlikely(wasMlocked
))
738 free_page_mlock(page
);
739 __count_vm_events(PGFREE
, 1 << order
);
740 free_one_page(page_zone(page
), page
, order
,
741 get_pageblock_migratetype(page
));
742 local_irq_restore(flags
);
745 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
747 unsigned int nr_pages
= 1 << order
;
751 for (loop
= 0; loop
< nr_pages
; loop
++) {
752 struct page
*p
= &page
[loop
];
754 if (loop
+ 1 < nr_pages
)
756 __ClearPageReserved(p
);
757 set_page_count(p
, 0);
760 set_page_refcounted(page
);
761 __free_pages(page
, order
);
765 /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
766 void __init
init_cma_reserved_pageblock(struct page
*page
)
768 unsigned i
= pageblock_nr_pages
;
769 struct page
*p
= page
;
772 __ClearPageReserved(p
);
773 set_page_count(p
, 0);
776 set_page_refcounted(page
);
777 set_pageblock_migratetype(page
, MIGRATE_CMA
);
778 __free_pages(page
, pageblock_order
);
779 totalram_pages
+= pageblock_nr_pages
;
784 * The order of subdivision here is critical for the IO subsystem.
785 * Please do not alter this order without good reasons and regression
786 * testing. Specifically, as large blocks of memory are subdivided,
787 * the order in which smaller blocks are delivered depends on the order
788 * they're subdivided in this function. This is the primary factor
789 * influencing the order in which pages are delivered to the IO
790 * subsystem according to empirical testing, and this is also justified
791 * by considering the behavior of a buddy system containing a single
792 * large block of memory acted on by a series of small allocations.
793 * This behavior is a critical factor in sglist merging's success.
797 static inline void expand(struct zone
*zone
, struct page
*page
,
798 int low
, int high
, struct free_area
*area
,
801 unsigned long size
= 1 << high
;
807 VM_BUG_ON(bad_range(zone
, &page
[size
]));
809 #ifdef CONFIG_DEBUG_PAGEALLOC
810 if (high
< debug_guardpage_minorder()) {
812 * Mark as guard pages (or page), that will allow to
813 * merge back to allocator when buddy will be freed.
814 * Corresponding page table entries will not be touched,
815 * pages will stay not present in virtual address space
817 INIT_LIST_HEAD(&page
[size
].lru
);
818 set_page_guard_flag(&page
[size
]);
819 set_page_private(&page
[size
], high
);
820 /* Guard pages are not available for any usage */
821 __mod_zone_freepage_state(zone
, -(1 << high
),
826 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
828 set_page_order(&page
[size
], high
);
833 * This page is about to be returned from the page allocator
835 static inline int check_new_page(struct page
*page
)
837 if (unlikely(page_mapcount(page
) |
838 (page
->mapping
!= NULL
) |
839 (atomic_read(&page
->_count
) != 0) |
840 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
) |
841 (mem_cgroup_bad_page_check(page
)))) {
848 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
852 for (i
= 0; i
< (1 << order
); i
++) {
853 struct page
*p
= page
+ i
;
854 if (unlikely(check_new_page(p
)))
858 set_page_private(page
, 0);
859 set_page_refcounted(page
);
861 arch_alloc_page(page
, order
);
862 kernel_map_pages(page
, 1 << order
, 1);
864 if (gfp_flags
& __GFP_ZERO
)
865 prep_zero_page(page
, order
, gfp_flags
);
867 if (order
&& (gfp_flags
& __GFP_COMP
))
868 prep_compound_page(page
, order
);
874 * Go through the free lists for the given migratetype and remove
875 * the smallest available page from the freelists
878 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
881 unsigned int current_order
;
882 struct free_area
* area
;
885 /* Find a page of the appropriate size in the preferred list */
886 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
887 area
= &(zone
->free_area
[current_order
]);
888 if (list_empty(&area
->free_list
[migratetype
]))
891 page
= list_entry(area
->free_list
[migratetype
].next
,
893 list_del(&page
->lru
);
894 rmv_page_order(page
);
896 expand(zone
, page
, order
, current_order
, area
, migratetype
);
905 * This array describes the order lists are fallen back to when
906 * the free lists for the desirable migrate type are depleted
908 static int fallbacks
[MIGRATE_TYPES
][4] = {
909 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
910 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
912 [MIGRATE_MOVABLE
] = { MIGRATE_CMA
, MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
913 [MIGRATE_CMA
] = { MIGRATE_RESERVE
}, /* Never used */
915 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
917 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
}, /* Never used */
918 [MIGRATE_ISOLATE
] = { MIGRATE_RESERVE
}, /* Never used */
922 * Move the free pages in a range to the free lists of the requested type.
923 * Note that start_page and end_pages are not aligned on a pageblock
924 * boundary. If alignment is required, use move_freepages_block()
926 static int move_freepages(struct zone
*zone
,
927 struct page
*start_page
, struct page
*end_page
,
934 #ifndef CONFIG_HOLES_IN_ZONE
936 * page_zone is not safe to call in this context when
937 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
938 * anyway as we check zone boundaries in move_freepages_block().
939 * Remove at a later date when no bug reports exist related to
940 * grouping pages by mobility
942 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
945 for (page
= start_page
; page
<= end_page
;) {
946 /* Make sure we are not inadvertently changing nodes */
947 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
949 if (!pfn_valid_within(page_to_pfn(page
))) {
954 if (!PageBuddy(page
)) {
959 order
= page_order(page
);
960 list_move(&page
->lru
,
961 &zone
->free_area
[order
].free_list
[migratetype
]);
963 pages_moved
+= 1 << order
;
969 int move_freepages_block(struct zone
*zone
, struct page
*page
,
972 unsigned long start_pfn
, end_pfn
;
973 struct page
*start_page
, *end_page
;
975 start_pfn
= page_to_pfn(page
);
976 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
977 start_page
= pfn_to_page(start_pfn
);
978 end_page
= start_page
+ pageblock_nr_pages
- 1;
979 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
981 /* Do not cross zone boundaries */
982 if (start_pfn
< zone
->zone_start_pfn
)
984 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
987 return move_freepages(zone
, start_page
, end_page
, migratetype
);
990 static void change_pageblock_range(struct page
*pageblock_page
,
991 int start_order
, int migratetype
)
993 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
995 while (nr_pageblocks
--) {
996 set_pageblock_migratetype(pageblock_page
, migratetype
);
997 pageblock_page
+= pageblock_nr_pages
;
1001 /* Remove an element from the buddy allocator from the fallback list */
1002 static inline struct page
*
1003 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
1005 struct free_area
* area
;
1010 /* Find the largest possible block of pages in the other list */
1011 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
1014 migratetype
= fallbacks
[start_migratetype
][i
];
1016 /* MIGRATE_RESERVE handled later if necessary */
1017 if (migratetype
== MIGRATE_RESERVE
)
1020 area
= &(zone
->free_area
[current_order
]);
1021 if (list_empty(&area
->free_list
[migratetype
]))
1024 page
= list_entry(area
->free_list
[migratetype
].next
,
1029 * If breaking a large block of pages, move all free
1030 * pages to the preferred allocation list. If falling
1031 * back for a reclaimable kernel allocation, be more
1032 * aggressive about taking ownership of free pages
1034 * On the other hand, never change migration
1035 * type of MIGRATE_CMA pageblocks nor move CMA
1036 * pages on different free lists. We don't
1037 * want unmovable pages to be allocated from
1038 * MIGRATE_CMA areas.
1040 if (!is_migrate_cma(migratetype
) &&
1041 (unlikely(current_order
>= pageblock_order
/ 2) ||
1042 start_migratetype
== MIGRATE_RECLAIMABLE
||
1043 page_group_by_mobility_disabled
)) {
1045 pages
= move_freepages_block(zone
, page
,
1048 /* Claim the whole block if over half of it is free */
1049 if (pages
>= (1 << (pageblock_order
-1)) ||
1050 page_group_by_mobility_disabled
)
1051 set_pageblock_migratetype(page
,
1054 migratetype
= start_migratetype
;
1057 /* Remove the page from the freelists */
1058 list_del(&page
->lru
);
1059 rmv_page_order(page
);
1061 /* Take ownership for orders >= pageblock_order */
1062 if (current_order
>= pageblock_order
&&
1063 !is_migrate_cma(migratetype
))
1064 change_pageblock_range(page
, current_order
,
1067 expand(zone
, page
, order
, current_order
, area
,
1068 is_migrate_cma(migratetype
)
1069 ? migratetype
: start_migratetype
);
1071 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1072 start_migratetype
, migratetype
);
1082 * Do the hard work of removing an element from the buddy allocator.
1083 * Call me with the zone->lock already held.
1085 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1091 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1093 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1094 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1097 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1098 * is used because __rmqueue_smallest is an inline function
1099 * and we want just one call site
1102 migratetype
= MIGRATE_RESERVE
;
1107 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1112 * Obtain a specified number of elements from the buddy allocator, all under
1113 * a single hold of the lock, for efficiency. Add them to the supplied list.
1114 * Returns the number of new pages which were placed at *list.
1116 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1117 unsigned long count
, struct list_head
*list
,
1118 int migratetype
, int cold
)
1120 int mt
= migratetype
, i
;
1122 spin_lock(&zone
->lock
);
1123 for (i
= 0; i
< count
; ++i
) {
1124 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1125 if (unlikely(page
== NULL
))
1129 * Split buddy pages returned by expand() are received here
1130 * in physical page order. The page is added to the callers and
1131 * list and the list head then moves forward. From the callers
1132 * perspective, the linked list is ordered by page number in
1133 * some conditions. This is useful for IO devices that can
1134 * merge IO requests if the physical pages are ordered
1137 if (likely(cold
== 0))
1138 list_add(&page
->lru
, list
);
1140 list_add_tail(&page
->lru
, list
);
1141 if (IS_ENABLED(CONFIG_CMA
)) {
1142 mt
= get_pageblock_migratetype(page
);
1143 if (!is_migrate_cma(mt
) && mt
!= MIGRATE_ISOLATE
)
1146 set_page_private(page
, mt
);
1148 if (is_migrate_cma(mt
))
1149 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1152 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1153 spin_unlock(&zone
->lock
);
1159 * Called from the vmstat counter updater to drain pagesets of this
1160 * currently executing processor on remote nodes after they have
1163 * Note that this function must be called with the thread pinned to
1164 * a single processor.
1166 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1168 unsigned long flags
;
1171 local_irq_save(flags
);
1172 if (pcp
->count
>= pcp
->batch
)
1173 to_drain
= pcp
->batch
;
1175 to_drain
= pcp
->count
;
1177 free_pcppages_bulk(zone
, to_drain
, pcp
);
1178 pcp
->count
-= to_drain
;
1180 local_irq_restore(flags
);
1185 * Drain pages of the indicated processor.
1187 * The processor must either be the current processor and the
1188 * thread pinned to the current processor or a processor that
1191 static void drain_pages(unsigned int cpu
)
1193 unsigned long flags
;
1196 for_each_populated_zone(zone
) {
1197 struct per_cpu_pageset
*pset
;
1198 struct per_cpu_pages
*pcp
;
1200 local_irq_save(flags
);
1201 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1205 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1208 local_irq_restore(flags
);
1213 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1215 void drain_local_pages(void *arg
)
1217 drain_pages(smp_processor_id());
1221 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1223 * Note that this code is protected against sending an IPI to an offline
1224 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1225 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1226 * nothing keeps CPUs from showing up after we populated the cpumask and
1227 * before the call to on_each_cpu_mask().
1229 void drain_all_pages(void)
1232 struct per_cpu_pageset
*pcp
;
1236 * Allocate in the BSS so we wont require allocation in
1237 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1239 static cpumask_t cpus_with_pcps
;
1242 * We don't care about racing with CPU hotplug event
1243 * as offline notification will cause the notified
1244 * cpu to drain that CPU pcps and on_each_cpu_mask
1245 * disables preemption as part of its processing
1247 for_each_online_cpu(cpu
) {
1248 bool has_pcps
= false;
1249 for_each_populated_zone(zone
) {
1250 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1251 if (pcp
->pcp
.count
) {
1257 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1259 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1261 on_each_cpu_mask(&cpus_with_pcps
, drain_local_pages
, NULL
, 1);
1264 #ifdef CONFIG_HIBERNATION
1266 void mark_free_pages(struct zone
*zone
)
1268 unsigned long pfn
, max_zone_pfn
;
1269 unsigned long flags
;
1271 struct list_head
*curr
;
1273 if (!zone
->spanned_pages
)
1276 spin_lock_irqsave(&zone
->lock
, flags
);
1278 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1279 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1280 if (pfn_valid(pfn
)) {
1281 struct page
*page
= pfn_to_page(pfn
);
1283 if (!swsusp_page_is_forbidden(page
))
1284 swsusp_unset_page_free(page
);
1287 for_each_migratetype_order(order
, t
) {
1288 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1291 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1292 for (i
= 0; i
< (1UL << order
); i
++)
1293 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1296 spin_unlock_irqrestore(&zone
->lock
, flags
);
1298 #endif /* CONFIG_PM */
1301 * Free a 0-order page
1302 * cold == 1 ? free a cold page : free a hot page
1304 void free_hot_cold_page(struct page
*page
, int cold
)
1306 struct zone
*zone
= page_zone(page
);
1307 struct per_cpu_pages
*pcp
;
1308 unsigned long flags
;
1310 int wasMlocked
= __TestClearPageMlocked(page
);
1312 if (!free_pages_prepare(page
, 0))
1315 migratetype
= get_pageblock_migratetype(page
);
1316 set_page_private(page
, migratetype
);
1317 local_irq_save(flags
);
1318 if (unlikely(wasMlocked
))
1319 free_page_mlock(page
);
1320 __count_vm_event(PGFREE
);
1323 * We only track unmovable, reclaimable and movable on pcp lists.
1324 * Free ISOLATE pages back to the allocator because they are being
1325 * offlined but treat RESERVE as movable pages so we can get those
1326 * areas back if necessary. Otherwise, we may have to free
1327 * excessively into the page allocator
1329 if (migratetype
>= MIGRATE_PCPTYPES
) {
1330 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1331 free_one_page(zone
, page
, 0, migratetype
);
1334 migratetype
= MIGRATE_MOVABLE
;
1337 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1339 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1341 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1343 if (pcp
->count
>= pcp
->high
) {
1344 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1345 pcp
->count
-= pcp
->batch
;
1349 local_irq_restore(flags
);
1353 * Free a list of 0-order pages
1355 void free_hot_cold_page_list(struct list_head
*list
, int cold
)
1357 struct page
*page
, *next
;
1359 list_for_each_entry_safe(page
, next
, list
, lru
) {
1360 trace_mm_page_free_batched(page
, cold
);
1361 free_hot_cold_page(page
, cold
);
1366 * split_page takes a non-compound higher-order page, and splits it into
1367 * n (1<<order) sub-pages: page[0..n]
1368 * Each sub-page must be freed individually.
1370 * Note: this is probably too low level an operation for use in drivers.
1371 * Please consult with lkml before using this in your driver.
1373 void split_page(struct page
*page
, unsigned int order
)
1377 VM_BUG_ON(PageCompound(page
));
1378 VM_BUG_ON(!page_count(page
));
1380 #ifdef CONFIG_KMEMCHECK
1382 * Split shadow pages too, because free(page[0]) would
1383 * otherwise free the whole shadow.
1385 if (kmemcheck_page_is_tracked(page
))
1386 split_page(virt_to_page(page
[0].shadow
), order
);
1389 for (i
= 1; i
< (1 << order
); i
++)
1390 set_page_refcounted(page
+ i
);
1394 * Similar to the split_page family of functions except that the page
1395 * required at the given order and being isolated now to prevent races
1396 * with parallel allocators
1398 int capture_free_page(struct page
*page
, int alloc_order
, int migratetype
)
1401 unsigned long watermark
;
1405 BUG_ON(!PageBuddy(page
));
1407 zone
= page_zone(page
);
1408 order
= page_order(page
);
1410 /* Obey watermarks as if the page was being allocated */
1411 watermark
= low_wmark_pages(zone
) + (1 << order
);
1412 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1415 /* Remove page from free list */
1416 list_del(&page
->lru
);
1417 zone
->free_area
[order
].nr_free
--;
1418 rmv_page_order(page
);
1420 mt
= get_pageblock_migratetype(page
);
1421 if (unlikely(mt
!= MIGRATE_ISOLATE
))
1422 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
1424 if (alloc_order
!= order
)
1425 expand(zone
, page
, alloc_order
, order
,
1426 &zone
->free_area
[order
], migratetype
);
1428 /* Set the pageblock if the captured page is at least a pageblock */
1429 if (order
>= pageblock_order
- 1) {
1430 struct page
*endpage
= page
+ (1 << order
) - 1;
1431 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
1432 int mt
= get_pageblock_migratetype(page
);
1433 if (mt
!= MIGRATE_ISOLATE
&& !is_migrate_cma(mt
))
1434 set_pageblock_migratetype(page
,
1439 return 1UL << order
;
1443 * Similar to split_page except the page is already free. As this is only
1444 * being used for migration, the migratetype of the block also changes.
1445 * As this is called with interrupts disabled, the caller is responsible
1446 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1449 * Note: this is probably too low level an operation for use in drivers.
1450 * Please consult with lkml before using this in your driver.
1452 int split_free_page(struct page
*page
)
1457 BUG_ON(!PageBuddy(page
));
1458 order
= page_order(page
);
1460 nr_pages
= capture_free_page(page
, order
, 0);
1464 /* Split into individual pages */
1465 set_page_refcounted(page
);
1466 split_page(page
, order
);
1471 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1472 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1476 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1477 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1480 unsigned long flags
;
1482 int cold
= !!(gfp_flags
& __GFP_COLD
);
1485 if (likely(order
== 0)) {
1486 struct per_cpu_pages
*pcp
;
1487 struct list_head
*list
;
1489 local_irq_save(flags
);
1490 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1491 list
= &pcp
->lists
[migratetype
];
1492 if (list_empty(list
)) {
1493 pcp
->count
+= rmqueue_bulk(zone
, 0,
1496 if (unlikely(list_empty(list
)))
1501 page
= list_entry(list
->prev
, struct page
, lru
);
1503 page
= list_entry(list
->next
, struct page
, lru
);
1505 list_del(&page
->lru
);
1508 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1510 * __GFP_NOFAIL is not to be used in new code.
1512 * All __GFP_NOFAIL callers should be fixed so that they
1513 * properly detect and handle allocation failures.
1515 * We most definitely don't want callers attempting to
1516 * allocate greater than order-1 page units with
1519 WARN_ON_ONCE(order
> 1);
1521 spin_lock_irqsave(&zone
->lock
, flags
);
1522 page
= __rmqueue(zone
, order
, migratetype
);
1523 spin_unlock(&zone
->lock
);
1526 __mod_zone_freepage_state(zone
, -(1 << order
),
1527 get_pageblock_migratetype(page
));
1530 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1531 zone_statistics(preferred_zone
, zone
, gfp_flags
);
1532 local_irq_restore(flags
);
1534 VM_BUG_ON(bad_range(zone
, page
));
1535 if (prep_new_page(page
, order
, gfp_flags
))
1540 local_irq_restore(flags
);
1544 #ifdef CONFIG_FAIL_PAGE_ALLOC
1547 struct fault_attr attr
;
1549 u32 ignore_gfp_highmem
;
1550 u32 ignore_gfp_wait
;
1552 } fail_page_alloc
= {
1553 .attr
= FAULT_ATTR_INITIALIZER
,
1554 .ignore_gfp_wait
= 1,
1555 .ignore_gfp_highmem
= 1,
1559 static int __init
setup_fail_page_alloc(char *str
)
1561 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1563 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1565 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1567 if (order
< fail_page_alloc
.min_order
)
1569 if (gfp_mask
& __GFP_NOFAIL
)
1571 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1573 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1576 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1579 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1581 static int __init
fail_page_alloc_debugfs(void)
1583 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1586 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
1587 &fail_page_alloc
.attr
);
1589 return PTR_ERR(dir
);
1591 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1592 &fail_page_alloc
.ignore_gfp_wait
))
1594 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1595 &fail_page_alloc
.ignore_gfp_highmem
))
1597 if (!debugfs_create_u32("min-order", mode
, dir
,
1598 &fail_page_alloc
.min_order
))
1603 debugfs_remove_recursive(dir
);
1608 late_initcall(fail_page_alloc_debugfs
);
1610 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1612 #else /* CONFIG_FAIL_PAGE_ALLOC */
1614 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1619 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1622 * Return true if free pages are above 'mark'. This takes into account the order
1623 * of the allocation.
1625 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1626 int classzone_idx
, int alloc_flags
, long free_pages
)
1628 /* free_pages my go negative - that's OK */
1630 long lowmem_reserve
= z
->lowmem_reserve
[classzone_idx
];
1633 free_pages
-= (1 << order
) - 1;
1634 if (alloc_flags
& ALLOC_HIGH
)
1636 if (alloc_flags
& ALLOC_HARDER
)
1639 /* If allocation can't use CMA areas don't use free CMA pages */
1640 if (!(alloc_flags
& ALLOC_CMA
))
1641 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
1643 if (free_pages
<= min
+ lowmem_reserve
)
1645 for (o
= 0; o
< order
; o
++) {
1646 /* At the next order, this order's pages become unavailable */
1647 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1649 /* Require fewer higher order pages to be free */
1652 if (free_pages
<= min
)
1658 #ifdef CONFIG_MEMORY_ISOLATION
1659 static inline unsigned long nr_zone_isolate_freepages(struct zone
*zone
)
1661 if (unlikely(zone
->nr_pageblock_isolate
))
1662 return zone
->nr_pageblock_isolate
* pageblock_nr_pages
;
1666 static inline unsigned long nr_zone_isolate_freepages(struct zone
*zone
)
1672 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1673 int classzone_idx
, int alloc_flags
)
1675 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1676 zone_page_state(z
, NR_FREE_PAGES
));
1679 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1680 int classzone_idx
, int alloc_flags
)
1682 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1684 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1685 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1688 * If the zone has MIGRATE_ISOLATE type free pages, we should consider
1689 * it. nr_zone_isolate_freepages is never accurate so kswapd might not
1690 * sleep although it could do so. But this is more desirable for memory
1691 * hotplug than sleeping which can cause a livelock in the direct
1694 free_pages
-= nr_zone_isolate_freepages(z
);
1695 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1701 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1702 * skip over zones that are not allowed by the cpuset, or that have
1703 * been recently (in last second) found to be nearly full. See further
1704 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1705 * that have to skip over a lot of full or unallowed zones.
1707 * If the zonelist cache is present in the passed in zonelist, then
1708 * returns a pointer to the allowed node mask (either the current
1709 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1711 * If the zonelist cache is not available for this zonelist, does
1712 * nothing and returns NULL.
1714 * If the fullzones BITMAP in the zonelist cache is stale (more than
1715 * a second since last zap'd) then we zap it out (clear its bits.)
1717 * We hold off even calling zlc_setup, until after we've checked the
1718 * first zone in the zonelist, on the theory that most allocations will
1719 * be satisfied from that first zone, so best to examine that zone as
1720 * quickly as we can.
1722 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1724 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1725 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1727 zlc
= zonelist
->zlcache_ptr
;
1731 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1732 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1733 zlc
->last_full_zap
= jiffies
;
1736 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1737 &cpuset_current_mems_allowed
:
1738 &node_states
[N_HIGH_MEMORY
];
1739 return allowednodes
;
1743 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1744 * if it is worth looking at further for free memory:
1745 * 1) Check that the zone isn't thought to be full (doesn't have its
1746 * bit set in the zonelist_cache fullzones BITMAP).
1747 * 2) Check that the zones node (obtained from the zonelist_cache
1748 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1749 * Return true (non-zero) if zone is worth looking at further, or
1750 * else return false (zero) if it is not.
1752 * This check -ignores- the distinction between various watermarks,
1753 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1754 * found to be full for any variation of these watermarks, it will
1755 * be considered full for up to one second by all requests, unless
1756 * we are so low on memory on all allowed nodes that we are forced
1757 * into the second scan of the zonelist.
1759 * In the second scan we ignore this zonelist cache and exactly
1760 * apply the watermarks to all zones, even it is slower to do so.
1761 * We are low on memory in the second scan, and should leave no stone
1762 * unturned looking for a free page.
1764 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1765 nodemask_t
*allowednodes
)
1767 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1768 int i
; /* index of *z in zonelist zones */
1769 int n
; /* node that zone *z is on */
1771 zlc
= zonelist
->zlcache_ptr
;
1775 i
= z
- zonelist
->_zonerefs
;
1778 /* This zone is worth trying if it is allowed but not full */
1779 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1783 * Given 'z' scanning a zonelist, set the corresponding bit in
1784 * zlc->fullzones, so that subsequent attempts to allocate a page
1785 * from that zone don't waste time re-examining it.
1787 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1789 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1790 int i
; /* index of *z in zonelist zones */
1792 zlc
= zonelist
->zlcache_ptr
;
1796 i
= z
- zonelist
->_zonerefs
;
1798 set_bit(i
, zlc
->fullzones
);
1802 * clear all zones full, called after direct reclaim makes progress so that
1803 * a zone that was recently full is not skipped over for up to a second
1805 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1807 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1809 zlc
= zonelist
->zlcache_ptr
;
1813 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1816 #else /* CONFIG_NUMA */
1818 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1823 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1824 nodemask_t
*allowednodes
)
1829 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1833 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1836 #endif /* CONFIG_NUMA */
1839 * get_page_from_freelist goes through the zonelist trying to allocate
1842 static struct page
*
1843 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1844 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1845 struct zone
*preferred_zone
, int migratetype
)
1848 struct page
*page
= NULL
;
1851 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1852 int zlc_active
= 0; /* set if using zonelist_cache */
1853 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1855 classzone_idx
= zone_idx(preferred_zone
);
1858 * Scan zonelist, looking for a zone with enough free.
1859 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1861 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1862 high_zoneidx
, nodemask
) {
1863 if (NUMA_BUILD
&& zlc_active
&&
1864 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1866 if ((alloc_flags
& ALLOC_CPUSET
) &&
1867 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1870 * When allocating a page cache page for writing, we
1871 * want to get it from a zone that is within its dirty
1872 * limit, such that no single zone holds more than its
1873 * proportional share of globally allowed dirty pages.
1874 * The dirty limits take into account the zone's
1875 * lowmem reserves and high watermark so that kswapd
1876 * should be able to balance it without having to
1877 * write pages from its LRU list.
1879 * This may look like it could increase pressure on
1880 * lower zones by failing allocations in higher zones
1881 * before they are full. But the pages that do spill
1882 * over are limited as the lower zones are protected
1883 * by this very same mechanism. It should not become
1884 * a practical burden to them.
1886 * XXX: For now, allow allocations to potentially
1887 * exceed the per-zone dirty limit in the slowpath
1888 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1889 * which is important when on a NUMA setup the allowed
1890 * zones are together not big enough to reach the
1891 * global limit. The proper fix for these situations
1892 * will require awareness of zones in the
1893 * dirty-throttling and the flusher threads.
1895 if ((alloc_flags
& ALLOC_WMARK_LOW
) &&
1896 (gfp_mask
& __GFP_WRITE
) && !zone_dirty_ok(zone
))
1897 goto this_zone_full
;
1899 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1900 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1904 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1905 if (zone_watermark_ok(zone
, order
, mark
,
1906 classzone_idx
, alloc_flags
))
1909 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1911 * we do zlc_setup if there are multiple nodes
1912 * and before considering the first zone allowed
1915 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1920 if (zone_reclaim_mode
== 0)
1921 goto this_zone_full
;
1924 * As we may have just activated ZLC, check if the first
1925 * eligible zone has failed zone_reclaim recently.
1927 if (NUMA_BUILD
&& zlc_active
&&
1928 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1931 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1933 case ZONE_RECLAIM_NOSCAN
:
1936 case ZONE_RECLAIM_FULL
:
1937 /* scanned but unreclaimable */
1940 /* did we reclaim enough */
1941 if (!zone_watermark_ok(zone
, order
, mark
,
1942 classzone_idx
, alloc_flags
))
1943 goto this_zone_full
;
1948 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1949 gfp_mask
, migratetype
);
1954 zlc_mark_zone_full(zonelist
, z
);
1957 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1958 /* Disable zlc cache for second zonelist scan */
1965 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
1966 * necessary to allocate the page. The expectation is
1967 * that the caller is taking steps that will free more
1968 * memory. The caller should avoid the page being used
1969 * for !PFMEMALLOC purposes.
1971 page
->pfmemalloc
= !!(alloc_flags
& ALLOC_NO_WATERMARKS
);
1977 * Large machines with many possible nodes should not always dump per-node
1978 * meminfo in irq context.
1980 static inline bool should_suppress_show_mem(void)
1985 ret
= in_interrupt();
1990 static DEFINE_RATELIMIT_STATE(nopage_rs
,
1991 DEFAULT_RATELIMIT_INTERVAL
,
1992 DEFAULT_RATELIMIT_BURST
);
1994 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
1996 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
1998 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
1999 debug_guardpage_minorder() > 0)
2003 * This documents exceptions given to allocations in certain
2004 * contexts that are allowed to allocate outside current's set
2007 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2008 if (test_thread_flag(TIF_MEMDIE
) ||
2009 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2010 filter
&= ~SHOW_MEM_FILTER_NODES
;
2011 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
2012 filter
&= ~SHOW_MEM_FILTER_NODES
;
2015 struct va_format vaf
;
2018 va_start(args
, fmt
);
2023 pr_warn("%pV", &vaf
);
2028 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2029 current
->comm
, order
, gfp_mask
);
2032 if (!should_suppress_show_mem())
2037 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
2038 unsigned long did_some_progress
,
2039 unsigned long pages_reclaimed
)
2041 /* Do not loop if specifically requested */
2042 if (gfp_mask
& __GFP_NORETRY
)
2045 /* Always retry if specifically requested */
2046 if (gfp_mask
& __GFP_NOFAIL
)
2050 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2051 * making forward progress without invoking OOM. Suspend also disables
2052 * storage devices so kswapd will not help. Bail if we are suspending.
2054 if (!did_some_progress
&& pm_suspended_storage())
2058 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2059 * means __GFP_NOFAIL, but that may not be true in other
2062 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
2066 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2067 * specified, then we retry until we no longer reclaim any pages
2068 * (above), or we've reclaimed an order of pages at least as
2069 * large as the allocation's order. In both cases, if the
2070 * allocation still fails, we stop retrying.
2072 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
2078 static inline struct page
*
2079 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2080 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2081 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2086 /* Acquire the OOM killer lock for the zones in zonelist */
2087 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
2088 schedule_timeout_uninterruptible(1);
2093 * Go through the zonelist yet one more time, keep very high watermark
2094 * here, this is only to catch a parallel oom killing, we must fail if
2095 * we're still under heavy pressure.
2097 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
2098 order
, zonelist
, high_zoneidx
,
2099 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
2100 preferred_zone
, migratetype
);
2104 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2105 /* The OOM killer will not help higher order allocs */
2106 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2108 /* The OOM killer does not needlessly kill tasks for lowmem */
2109 if (high_zoneidx
< ZONE_NORMAL
)
2112 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2113 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2114 * The caller should handle page allocation failure by itself if
2115 * it specifies __GFP_THISNODE.
2116 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2118 if (gfp_mask
& __GFP_THISNODE
)
2121 /* Exhausted what can be done so it's blamo time */
2122 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
, false);
2125 clear_zonelist_oom(zonelist
, gfp_mask
);
2129 #ifdef CONFIG_COMPACTION
2130 /* Try memory compaction for high-order allocations before reclaim */
2131 static struct page
*
2132 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2133 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2134 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2135 int migratetype
, bool sync_migration
,
2136 bool *contended_compaction
, bool *deferred_compaction
,
2137 unsigned long *did_some_progress
)
2139 struct page
*page
= NULL
;
2144 if (compaction_deferred(preferred_zone
, order
)) {
2145 *deferred_compaction
= true;
2149 current
->flags
|= PF_MEMALLOC
;
2150 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
2151 nodemask
, sync_migration
,
2152 contended_compaction
, &page
);
2153 current
->flags
&= ~PF_MEMALLOC
;
2155 /* If compaction captured a page, prep and use it */
2157 prep_new_page(page
, order
, gfp_mask
);
2161 if (*did_some_progress
!= COMPACT_SKIPPED
) {
2162 /* Page migration frees to the PCP lists but we want merging */
2163 drain_pages(get_cpu());
2166 page
= get_page_from_freelist(gfp_mask
, nodemask
,
2167 order
, zonelist
, high_zoneidx
,
2168 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2169 preferred_zone
, migratetype
);
2172 preferred_zone
->compact_considered
= 0;
2173 preferred_zone
->compact_defer_shift
= 0;
2174 if (order
>= preferred_zone
->compact_order_failed
)
2175 preferred_zone
->compact_order_failed
= order
+ 1;
2176 count_vm_event(COMPACTSUCCESS
);
2181 * It's bad if compaction run occurs and fails.
2182 * The most likely reason is that pages exist,
2183 * but not enough to satisfy watermarks.
2185 count_vm_event(COMPACTFAIL
);
2188 * As async compaction considers a subset of pageblocks, only
2189 * defer if the failure was a sync compaction failure.
2192 defer_compaction(preferred_zone
, order
);
2200 static inline struct page
*
2201 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2202 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2203 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2204 int migratetype
, bool sync_migration
,
2205 bool *contended_compaction
, bool *deferred_compaction
,
2206 unsigned long *did_some_progress
)
2210 #endif /* CONFIG_COMPACTION */
2212 /* Perform direct synchronous page reclaim */
2214 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
, struct zonelist
*zonelist
,
2215 nodemask_t
*nodemask
)
2217 struct reclaim_state reclaim_state
;
2222 /* We now go into synchronous reclaim */
2223 cpuset_memory_pressure_bump();
2224 current
->flags
|= PF_MEMALLOC
;
2225 lockdep_set_current_reclaim_state(gfp_mask
);
2226 reclaim_state
.reclaimed_slab
= 0;
2227 current
->reclaim_state
= &reclaim_state
;
2229 progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
2231 current
->reclaim_state
= NULL
;
2232 lockdep_clear_current_reclaim_state();
2233 current
->flags
&= ~PF_MEMALLOC
;
2240 /* The really slow allocator path where we enter direct reclaim */
2241 static inline struct page
*
2242 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2243 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2244 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2245 int migratetype
, unsigned long *did_some_progress
)
2247 struct page
*page
= NULL
;
2248 bool drained
= false;
2250 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
2252 if (unlikely(!(*did_some_progress
)))
2255 /* After successful reclaim, reconsider all zones for allocation */
2257 zlc_clear_zones_full(zonelist
);
2260 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2261 zonelist
, high_zoneidx
,
2262 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2263 preferred_zone
, migratetype
);
2266 * If an allocation failed after direct reclaim, it could be because
2267 * pages are pinned on the per-cpu lists. Drain them and try again
2269 if (!page
&& !drained
) {
2279 * This is called in the allocator slow-path if the allocation request is of
2280 * sufficient urgency to ignore watermarks and take other desperate measures
2282 static inline struct page
*
2283 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2284 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2285 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2291 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2292 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
2293 preferred_zone
, migratetype
);
2295 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2296 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2297 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2303 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
2304 enum zone_type high_zoneidx
,
2305 enum zone_type classzone_idx
)
2310 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
2311 wakeup_kswapd(zone
, order
, classzone_idx
);
2315 gfp_to_alloc_flags(gfp_t gfp_mask
)
2317 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2318 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2320 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2321 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2324 * The caller may dip into page reserves a bit more if the caller
2325 * cannot run direct reclaim, or if the caller has realtime scheduling
2326 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2327 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2329 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2333 * Not worth trying to allocate harder for
2334 * __GFP_NOMEMALLOC even if it can't schedule.
2336 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2337 alloc_flags
|= ALLOC_HARDER
;
2339 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2340 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2342 alloc_flags
&= ~ALLOC_CPUSET
;
2343 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2344 alloc_flags
|= ALLOC_HARDER
;
2346 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2347 if (gfp_mask
& __GFP_MEMALLOC
)
2348 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2349 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2350 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2351 else if (!in_interrupt() &&
2352 ((current
->flags
& PF_MEMALLOC
) ||
2353 unlikely(test_thread_flag(TIF_MEMDIE
))))
2354 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2357 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2358 alloc_flags
|= ALLOC_CMA
;
2363 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2365 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2368 static inline struct page
*
2369 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2370 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2371 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2374 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2375 struct page
*page
= NULL
;
2377 unsigned long pages_reclaimed
= 0;
2378 unsigned long did_some_progress
;
2379 bool sync_migration
= false;
2380 bool deferred_compaction
= false;
2381 bool contended_compaction
= false;
2384 * In the slowpath, we sanity check order to avoid ever trying to
2385 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2386 * be using allocators in order of preference for an area that is
2389 if (order
>= MAX_ORDER
) {
2390 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2395 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2396 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2397 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2398 * using a larger set of nodes after it has established that the
2399 * allowed per node queues are empty and that nodes are
2402 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2406 wake_all_kswapd(order
, zonelist
, high_zoneidx
,
2407 zone_idx(preferred_zone
));
2410 * OK, we're below the kswapd watermark and have kicked background
2411 * reclaim. Now things get more complex, so set up alloc_flags according
2412 * to how we want to proceed.
2414 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2417 * Find the true preferred zone if the allocation is unconstrained by
2420 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2421 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2425 /* This is the last chance, in general, before the goto nopage. */
2426 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2427 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2428 preferred_zone
, migratetype
);
2432 /* Allocate without watermarks if the context allows */
2433 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2435 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2436 * the allocation is high priority and these type of
2437 * allocations are system rather than user orientated
2439 zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
2441 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2442 zonelist
, high_zoneidx
, nodemask
,
2443 preferred_zone
, migratetype
);
2449 /* Atomic allocations - we can't balance anything */
2453 /* Avoid recursion of direct reclaim */
2454 if (current
->flags
& PF_MEMALLOC
)
2457 /* Avoid allocations with no watermarks from looping endlessly */
2458 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2462 * Try direct compaction. The first pass is asynchronous. Subsequent
2463 * attempts after direct reclaim are synchronous
2465 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2466 zonelist
, high_zoneidx
,
2468 alloc_flags
, preferred_zone
,
2469 migratetype
, sync_migration
,
2470 &contended_compaction
,
2471 &deferred_compaction
,
2472 &did_some_progress
);
2475 sync_migration
= true;
2478 * If compaction is deferred for high-order allocations, it is because
2479 * sync compaction recently failed. In this is the case and the caller
2480 * requested a movable allocation that does not heavily disrupt the
2481 * system then fail the allocation instead of entering direct reclaim.
2483 if ((deferred_compaction
|| contended_compaction
) &&
2484 (gfp_mask
& (__GFP_MOVABLE
|__GFP_REPEAT
)) == __GFP_MOVABLE
)
2487 /* Try direct reclaim and then allocating */
2488 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2489 zonelist
, high_zoneidx
,
2491 alloc_flags
, preferred_zone
,
2492 migratetype
, &did_some_progress
);
2497 * If we failed to make any progress reclaiming, then we are
2498 * running out of options and have to consider going OOM
2500 if (!did_some_progress
) {
2501 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2502 if (oom_killer_disabled
)
2504 /* Coredumps can quickly deplete all memory reserves */
2505 if ((current
->flags
& PF_DUMPCORE
) &&
2506 !(gfp_mask
& __GFP_NOFAIL
))
2508 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2509 zonelist
, high_zoneidx
,
2510 nodemask
, preferred_zone
,
2515 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2517 * The oom killer is not called for high-order
2518 * allocations that may fail, so if no progress
2519 * is being made, there are no other options and
2520 * retrying is unlikely to help.
2522 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2525 * The oom killer is not called for lowmem
2526 * allocations to prevent needlessly killing
2529 if (high_zoneidx
< ZONE_NORMAL
)
2537 /* Check if we should retry the allocation */
2538 pages_reclaimed
+= did_some_progress
;
2539 if (should_alloc_retry(gfp_mask
, order
, did_some_progress
,
2541 /* Wait for some write requests to complete then retry */
2542 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2546 * High-order allocations do not necessarily loop after
2547 * direct reclaim and reclaim/compaction depends on compaction
2548 * being called after reclaim so call directly if necessary
2550 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2551 zonelist
, high_zoneidx
,
2553 alloc_flags
, preferred_zone
,
2554 migratetype
, sync_migration
,
2555 &contended_compaction
,
2556 &deferred_compaction
,
2557 &did_some_progress
);
2563 warn_alloc_failed(gfp_mask
, order
, NULL
);
2566 if (kmemcheck_enabled
)
2567 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2573 * This is the 'heart' of the zoned buddy allocator.
2576 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2577 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2579 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2580 struct zone
*preferred_zone
;
2581 struct page
*page
= NULL
;
2582 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2583 unsigned int cpuset_mems_cookie
;
2584 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
;
2586 gfp_mask
&= gfp_allowed_mask
;
2588 lockdep_trace_alloc(gfp_mask
);
2590 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2592 if (should_fail_alloc_page(gfp_mask
, order
))
2596 * Check the zones suitable for the gfp_mask contain at least one
2597 * valid zone. It's possible to have an empty zonelist as a result
2598 * of GFP_THISNODE and a memoryless node
2600 if (unlikely(!zonelist
->_zonerefs
->zone
))
2604 cpuset_mems_cookie
= get_mems_allowed();
2606 /* The preferred zone is used for statistics later */
2607 first_zones_zonelist(zonelist
, high_zoneidx
,
2608 nodemask
? : &cpuset_current_mems_allowed
,
2610 if (!preferred_zone
)
2614 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2615 alloc_flags
|= ALLOC_CMA
;
2617 /* First allocation attempt */
2618 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2619 zonelist
, high_zoneidx
, alloc_flags
,
2620 preferred_zone
, migratetype
);
2621 if (unlikely(!page
))
2622 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2623 zonelist
, high_zoneidx
, nodemask
,
2624 preferred_zone
, migratetype
);
2626 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2630 * When updating a task's mems_allowed, it is possible to race with
2631 * parallel threads in such a way that an allocation can fail while
2632 * the mask is being updated. If a page allocation is about to fail,
2633 * check if the cpuset changed during allocation and if so, retry.
2635 if (unlikely(!put_mems_allowed(cpuset_mems_cookie
) && !page
))
2640 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2643 * Common helper functions.
2645 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2650 * __get_free_pages() returns a 32-bit address, which cannot represent
2653 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2655 page
= alloc_pages(gfp_mask
, order
);
2658 return (unsigned long) page_address(page
);
2660 EXPORT_SYMBOL(__get_free_pages
);
2662 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2664 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2666 EXPORT_SYMBOL(get_zeroed_page
);
2668 void __free_pages(struct page
*page
, unsigned int order
)
2670 if (put_page_testzero(page
)) {
2672 free_hot_cold_page(page
, 0);
2674 __free_pages_ok(page
, order
);
2678 EXPORT_SYMBOL(__free_pages
);
2680 void free_pages(unsigned long addr
, unsigned int order
)
2683 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2684 __free_pages(virt_to_page((void *)addr
), order
);
2688 EXPORT_SYMBOL(free_pages
);
2690 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2693 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2694 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2696 split_page(virt_to_page((void *)addr
), order
);
2697 while (used
< alloc_end
) {
2702 return (void *)addr
;
2706 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2707 * @size: the number of bytes to allocate
2708 * @gfp_mask: GFP flags for the allocation
2710 * This function is similar to alloc_pages(), except that it allocates the
2711 * minimum number of pages to satisfy the request. alloc_pages() can only
2712 * allocate memory in power-of-two pages.
2714 * This function is also limited by MAX_ORDER.
2716 * Memory allocated by this function must be released by free_pages_exact().
2718 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2720 unsigned int order
= get_order(size
);
2723 addr
= __get_free_pages(gfp_mask
, order
);
2724 return make_alloc_exact(addr
, order
, size
);
2726 EXPORT_SYMBOL(alloc_pages_exact
);
2729 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2731 * @nid: the preferred node ID where memory should be allocated
2732 * @size: the number of bytes to allocate
2733 * @gfp_mask: GFP flags for the allocation
2735 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2737 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2740 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2742 unsigned order
= get_order(size
);
2743 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2746 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2748 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2751 * free_pages_exact - release memory allocated via alloc_pages_exact()
2752 * @virt: the value returned by alloc_pages_exact.
2753 * @size: size of allocation, same value as passed to alloc_pages_exact().
2755 * Release the memory allocated by a previous call to alloc_pages_exact.
2757 void free_pages_exact(void *virt
, size_t size
)
2759 unsigned long addr
= (unsigned long)virt
;
2760 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2762 while (addr
< end
) {
2767 EXPORT_SYMBOL(free_pages_exact
);
2769 static unsigned int nr_free_zone_pages(int offset
)
2774 /* Just pick one node, since fallback list is circular */
2775 unsigned int sum
= 0;
2777 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2779 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2780 unsigned long size
= zone
->present_pages
;
2781 unsigned long high
= high_wmark_pages(zone
);
2790 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2792 unsigned int nr_free_buffer_pages(void)
2794 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2796 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2799 * Amount of free RAM allocatable within all zones
2801 unsigned int nr_free_pagecache_pages(void)
2803 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2806 static inline void show_node(struct zone
*zone
)
2809 printk("Node %d ", zone_to_nid(zone
));
2812 void si_meminfo(struct sysinfo
*val
)
2814 val
->totalram
= totalram_pages
;
2816 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2817 val
->bufferram
= nr_blockdev_pages();
2818 val
->totalhigh
= totalhigh_pages
;
2819 val
->freehigh
= nr_free_highpages();
2820 val
->mem_unit
= PAGE_SIZE
;
2823 EXPORT_SYMBOL(si_meminfo
);
2826 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2828 pg_data_t
*pgdat
= NODE_DATA(nid
);
2830 val
->totalram
= pgdat
->node_present_pages
;
2831 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2832 #ifdef CONFIG_HIGHMEM
2833 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2834 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2840 val
->mem_unit
= PAGE_SIZE
;
2845 * Determine whether the node should be displayed or not, depending on whether
2846 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2848 bool skip_free_areas_node(unsigned int flags
, int nid
)
2851 unsigned int cpuset_mems_cookie
;
2853 if (!(flags
& SHOW_MEM_FILTER_NODES
))
2857 cpuset_mems_cookie
= get_mems_allowed();
2858 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
2859 } while (!put_mems_allowed(cpuset_mems_cookie
));
2864 #define K(x) ((x) << (PAGE_SHIFT-10))
2867 * Show free area list (used inside shift_scroll-lock stuff)
2868 * We also calculate the percentage fragmentation. We do this by counting the
2869 * memory on each free list with the exception of the first item on the list.
2870 * Suppresses nodes that are not allowed by current's cpuset if
2871 * SHOW_MEM_FILTER_NODES is passed.
2873 void show_free_areas(unsigned int filter
)
2878 for_each_populated_zone(zone
) {
2879 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2882 printk("%s per-cpu:\n", zone
->name
);
2884 for_each_online_cpu(cpu
) {
2885 struct per_cpu_pageset
*pageset
;
2887 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2889 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2890 cpu
, pageset
->pcp
.high
,
2891 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2895 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2896 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2898 " dirty:%lu writeback:%lu unstable:%lu\n"
2899 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2900 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
2902 global_page_state(NR_ACTIVE_ANON
),
2903 global_page_state(NR_INACTIVE_ANON
),
2904 global_page_state(NR_ISOLATED_ANON
),
2905 global_page_state(NR_ACTIVE_FILE
),
2906 global_page_state(NR_INACTIVE_FILE
),
2907 global_page_state(NR_ISOLATED_FILE
),
2908 global_page_state(NR_UNEVICTABLE
),
2909 global_page_state(NR_FILE_DIRTY
),
2910 global_page_state(NR_WRITEBACK
),
2911 global_page_state(NR_UNSTABLE_NFS
),
2912 global_page_state(NR_FREE_PAGES
),
2913 global_page_state(NR_SLAB_RECLAIMABLE
),
2914 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2915 global_page_state(NR_FILE_MAPPED
),
2916 global_page_state(NR_SHMEM
),
2917 global_page_state(NR_PAGETABLE
),
2918 global_page_state(NR_BOUNCE
),
2919 global_page_state(NR_FREE_CMA_PAGES
));
2921 for_each_populated_zone(zone
) {
2924 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2932 " active_anon:%lukB"
2933 " inactive_anon:%lukB"
2934 " active_file:%lukB"
2935 " inactive_file:%lukB"
2936 " unevictable:%lukB"
2937 " isolated(anon):%lukB"
2938 " isolated(file):%lukB"
2945 " slab_reclaimable:%lukB"
2946 " slab_unreclaimable:%lukB"
2947 " kernel_stack:%lukB"
2952 " writeback_tmp:%lukB"
2953 " pages_scanned:%lu"
2954 " all_unreclaimable? %s"
2957 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2958 K(min_wmark_pages(zone
)),
2959 K(low_wmark_pages(zone
)),
2960 K(high_wmark_pages(zone
)),
2961 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2962 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2963 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2964 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2965 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2966 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
2967 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
2968 K(zone
->present_pages
),
2969 K(zone_page_state(zone
, NR_MLOCK
)),
2970 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
2971 K(zone_page_state(zone
, NR_WRITEBACK
)),
2972 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
2973 K(zone_page_state(zone
, NR_SHMEM
)),
2974 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
2975 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
2976 zone_page_state(zone
, NR_KERNEL_STACK
) *
2978 K(zone_page_state(zone
, NR_PAGETABLE
)),
2979 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
2980 K(zone_page_state(zone
, NR_BOUNCE
)),
2981 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
2982 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
2983 zone
->pages_scanned
,
2984 (zone
->all_unreclaimable
? "yes" : "no")
2986 printk("lowmem_reserve[]:");
2987 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2988 printk(" %lu", zone
->lowmem_reserve
[i
]);
2992 for_each_populated_zone(zone
) {
2993 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2995 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2998 printk("%s: ", zone
->name
);
3000 spin_lock_irqsave(&zone
->lock
, flags
);
3001 for (order
= 0; order
< MAX_ORDER
; order
++) {
3002 nr
[order
] = zone
->free_area
[order
].nr_free
;
3003 total
+= nr
[order
] << order
;
3005 spin_unlock_irqrestore(&zone
->lock
, flags
);
3006 for (order
= 0; order
< MAX_ORDER
; order
++)
3007 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3008 printk("= %lukB\n", K(total
));
3011 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3013 show_swap_cache_info();
3016 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3018 zoneref
->zone
= zone
;
3019 zoneref
->zone_idx
= zone_idx(zone
);
3023 * Builds allocation fallback zone lists.
3025 * Add all populated zones of a node to the zonelist.
3027 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3028 int nr_zones
, enum zone_type zone_type
)
3032 BUG_ON(zone_type
>= MAX_NR_ZONES
);
3037 zone
= pgdat
->node_zones
+ zone_type
;
3038 if (populated_zone(zone
)) {
3039 zoneref_set_zone(zone
,
3040 &zonelist
->_zonerefs
[nr_zones
++]);
3041 check_highest_zone(zone_type
);
3044 } while (zone_type
);
3051 * 0 = automatic detection of better ordering.
3052 * 1 = order by ([node] distance, -zonetype)
3053 * 2 = order by (-zonetype, [node] distance)
3055 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3056 * the same zonelist. So only NUMA can configure this param.
3058 #define ZONELIST_ORDER_DEFAULT 0
3059 #define ZONELIST_ORDER_NODE 1
3060 #define ZONELIST_ORDER_ZONE 2
3062 /* zonelist order in the kernel.
3063 * set_zonelist_order() will set this to NODE or ZONE.
3065 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3066 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3070 /* The value user specified ....changed by config */
3071 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3072 /* string for sysctl */
3073 #define NUMA_ZONELIST_ORDER_LEN 16
3074 char numa_zonelist_order
[16] = "default";
3077 * interface for configure zonelist ordering.
3078 * command line option "numa_zonelist_order"
3079 * = "[dD]efault - default, automatic configuration.
3080 * = "[nN]ode - order by node locality, then by zone within node
3081 * = "[zZ]one - order by zone, then by locality within zone
3084 static int __parse_numa_zonelist_order(char *s
)
3086 if (*s
== 'd' || *s
== 'D') {
3087 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3088 } else if (*s
== 'n' || *s
== 'N') {
3089 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3090 } else if (*s
== 'z' || *s
== 'Z') {
3091 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3094 "Ignoring invalid numa_zonelist_order value: "
3101 static __init
int setup_numa_zonelist_order(char *s
)
3108 ret
= __parse_numa_zonelist_order(s
);
3110 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3114 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3117 * sysctl handler for numa_zonelist_order
3119 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
3120 void __user
*buffer
, size_t *length
,
3123 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3125 static DEFINE_MUTEX(zl_order_mutex
);
3127 mutex_lock(&zl_order_mutex
);
3129 strcpy(saved_string
, (char*)table
->data
);
3130 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3134 int oldval
= user_zonelist_order
;
3135 if (__parse_numa_zonelist_order((char*)table
->data
)) {
3137 * bogus value. restore saved string
3139 strncpy((char*)table
->data
, saved_string
,
3140 NUMA_ZONELIST_ORDER_LEN
);
3141 user_zonelist_order
= oldval
;
3142 } else if (oldval
!= user_zonelist_order
) {
3143 mutex_lock(&zonelists_mutex
);
3144 build_all_zonelists(NULL
, NULL
);
3145 mutex_unlock(&zonelists_mutex
);
3149 mutex_unlock(&zl_order_mutex
);
3154 #define MAX_NODE_LOAD (nr_online_nodes)
3155 static int node_load
[MAX_NUMNODES
];
3158 * find_next_best_node - find the next node that should appear in a given node's fallback list
3159 * @node: node whose fallback list we're appending
3160 * @used_node_mask: nodemask_t of already used nodes
3162 * We use a number of factors to determine which is the next node that should
3163 * appear on a given node's fallback list. The node should not have appeared
3164 * already in @node's fallback list, and it should be the next closest node
3165 * according to the distance array (which contains arbitrary distance values
3166 * from each node to each node in the system), and should also prefer nodes
3167 * with no CPUs, since presumably they'll have very little allocation pressure
3168 * on them otherwise.
3169 * It returns -1 if no node is found.
3171 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3174 int min_val
= INT_MAX
;
3176 const struct cpumask
*tmp
= cpumask_of_node(0);
3178 /* Use the local node if we haven't already */
3179 if (!node_isset(node
, *used_node_mask
)) {
3180 node_set(node
, *used_node_mask
);
3184 for_each_node_state(n
, N_HIGH_MEMORY
) {
3186 /* Don't want a node to appear more than once */
3187 if (node_isset(n
, *used_node_mask
))
3190 /* Use the distance array to find the distance */
3191 val
= node_distance(node
, n
);
3193 /* Penalize nodes under us ("prefer the next node") */
3196 /* Give preference to headless and unused nodes */
3197 tmp
= cpumask_of_node(n
);
3198 if (!cpumask_empty(tmp
))
3199 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
3201 /* Slight preference for less loaded node */
3202 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
3203 val
+= node_load
[n
];
3205 if (val
< min_val
) {
3212 node_set(best_node
, *used_node_mask
);
3219 * Build zonelists ordered by node and zones within node.
3220 * This results in maximum locality--normal zone overflows into local
3221 * DMA zone, if any--but risks exhausting DMA zone.
3223 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
3226 struct zonelist
*zonelist
;
3228 zonelist
= &pgdat
->node_zonelists
[0];
3229 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
3231 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3233 zonelist
->_zonerefs
[j
].zone
= NULL
;
3234 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3238 * Build gfp_thisnode zonelists
3240 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
3243 struct zonelist
*zonelist
;
3245 zonelist
= &pgdat
->node_zonelists
[1];
3246 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3247 zonelist
->_zonerefs
[j
].zone
= NULL
;
3248 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3252 * Build zonelists ordered by zone and nodes within zones.
3253 * This results in conserving DMA zone[s] until all Normal memory is
3254 * exhausted, but results in overflowing to remote node while memory
3255 * may still exist in local DMA zone.
3257 static int node_order
[MAX_NUMNODES
];
3259 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
3262 int zone_type
; /* needs to be signed */
3264 struct zonelist
*zonelist
;
3266 zonelist
= &pgdat
->node_zonelists
[0];
3268 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
3269 for (j
= 0; j
< nr_nodes
; j
++) {
3270 node
= node_order
[j
];
3271 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
3272 if (populated_zone(z
)) {
3274 &zonelist
->_zonerefs
[pos
++]);
3275 check_highest_zone(zone_type
);
3279 zonelist
->_zonerefs
[pos
].zone
= NULL
;
3280 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
3283 static int default_zonelist_order(void)
3286 unsigned long low_kmem_size
,total_size
;
3290 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3291 * If they are really small and used heavily, the system can fall
3292 * into OOM very easily.
3293 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3295 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3298 for_each_online_node(nid
) {
3299 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3300 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3301 if (populated_zone(z
)) {
3302 if (zone_type
< ZONE_NORMAL
)
3303 low_kmem_size
+= z
->present_pages
;
3304 total_size
+= z
->present_pages
;
3305 } else if (zone_type
== ZONE_NORMAL
) {
3307 * If any node has only lowmem, then node order
3308 * is preferred to allow kernel allocations
3309 * locally; otherwise, they can easily infringe
3310 * on other nodes when there is an abundance of
3311 * lowmem available to allocate from.
3313 return ZONELIST_ORDER_NODE
;
3317 if (!low_kmem_size
|| /* there are no DMA area. */
3318 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
3319 return ZONELIST_ORDER_NODE
;
3321 * look into each node's config.
3322 * If there is a node whose DMA/DMA32 memory is very big area on
3323 * local memory, NODE_ORDER may be suitable.
3325 average_size
= total_size
/
3326 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
3327 for_each_online_node(nid
) {
3330 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3331 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3332 if (populated_zone(z
)) {
3333 if (zone_type
< ZONE_NORMAL
)
3334 low_kmem_size
+= z
->present_pages
;
3335 total_size
+= z
->present_pages
;
3338 if (low_kmem_size
&&
3339 total_size
> average_size
&& /* ignore small node */
3340 low_kmem_size
> total_size
* 70/100)
3341 return ZONELIST_ORDER_NODE
;
3343 return ZONELIST_ORDER_ZONE
;
3346 static void set_zonelist_order(void)
3348 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
3349 current_zonelist_order
= default_zonelist_order();
3351 current_zonelist_order
= user_zonelist_order
;
3354 static void build_zonelists(pg_data_t
*pgdat
)
3358 nodemask_t used_mask
;
3359 int local_node
, prev_node
;
3360 struct zonelist
*zonelist
;
3361 int order
= current_zonelist_order
;
3363 /* initialize zonelists */
3364 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3365 zonelist
= pgdat
->node_zonelists
+ i
;
3366 zonelist
->_zonerefs
[0].zone
= NULL
;
3367 zonelist
->_zonerefs
[0].zone_idx
= 0;
3370 /* NUMA-aware ordering of nodes */
3371 local_node
= pgdat
->node_id
;
3372 load
= nr_online_nodes
;
3373 prev_node
= local_node
;
3374 nodes_clear(used_mask
);
3376 memset(node_order
, 0, sizeof(node_order
));
3379 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3380 int distance
= node_distance(local_node
, node
);
3383 * If another node is sufficiently far away then it is better
3384 * to reclaim pages in a zone before going off node.
3386 if (distance
> RECLAIM_DISTANCE
)
3387 zone_reclaim_mode
= 1;
3390 * We don't want to pressure a particular node.
3391 * So adding penalty to the first node in same
3392 * distance group to make it round-robin.
3394 if (distance
!= node_distance(local_node
, prev_node
))
3395 node_load
[node
] = load
;
3399 if (order
== ZONELIST_ORDER_NODE
)
3400 build_zonelists_in_node_order(pgdat
, node
);
3402 node_order
[j
++] = node
; /* remember order */
3405 if (order
== ZONELIST_ORDER_ZONE
) {
3406 /* calculate node order -- i.e., DMA last! */
3407 build_zonelists_in_zone_order(pgdat
, j
);
3410 build_thisnode_zonelists(pgdat
);
3413 /* Construct the zonelist performance cache - see further mmzone.h */
3414 static void build_zonelist_cache(pg_data_t
*pgdat
)
3416 struct zonelist
*zonelist
;
3417 struct zonelist_cache
*zlc
;
3420 zonelist
= &pgdat
->node_zonelists
[0];
3421 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3422 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3423 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3424 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3427 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3429 * Return node id of node used for "local" allocations.
3430 * I.e., first node id of first zone in arg node's generic zonelist.
3431 * Used for initializing percpu 'numa_mem', which is used primarily
3432 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3434 int local_memory_node(int node
)
3438 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3439 gfp_zone(GFP_KERNEL
),
3446 #else /* CONFIG_NUMA */
3448 static void set_zonelist_order(void)
3450 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3453 static void build_zonelists(pg_data_t
*pgdat
)
3455 int node
, local_node
;
3457 struct zonelist
*zonelist
;
3459 local_node
= pgdat
->node_id
;
3461 zonelist
= &pgdat
->node_zonelists
[0];
3462 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3465 * Now we build the zonelist so that it contains the zones
3466 * of all the other nodes.
3467 * We don't want to pressure a particular node, so when
3468 * building the zones for node N, we make sure that the
3469 * zones coming right after the local ones are those from
3470 * node N+1 (modulo N)
3472 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3473 if (!node_online(node
))
3475 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3478 for (node
= 0; node
< local_node
; node
++) {
3479 if (!node_online(node
))
3481 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3485 zonelist
->_zonerefs
[j
].zone
= NULL
;
3486 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3489 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3490 static void build_zonelist_cache(pg_data_t
*pgdat
)
3492 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3495 #endif /* CONFIG_NUMA */
3498 * Boot pageset table. One per cpu which is going to be used for all
3499 * zones and all nodes. The parameters will be set in such a way
3500 * that an item put on a list will immediately be handed over to
3501 * the buddy list. This is safe since pageset manipulation is done
3502 * with interrupts disabled.
3504 * The boot_pagesets must be kept even after bootup is complete for
3505 * unused processors and/or zones. They do play a role for bootstrapping
3506 * hotplugged processors.
3508 * zoneinfo_show() and maybe other functions do
3509 * not check if the processor is online before following the pageset pointer.
3510 * Other parts of the kernel may not check if the zone is available.
3512 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3513 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3514 static void setup_zone_pageset(struct zone
*zone
);
3517 * Global mutex to protect against size modification of zonelists
3518 * as well as to serialize pageset setup for the new populated zone.
3520 DEFINE_MUTEX(zonelists_mutex
);
3522 /* return values int ....just for stop_machine() */
3523 static int __build_all_zonelists(void *data
)
3527 pg_data_t
*self
= data
;
3530 memset(node_load
, 0, sizeof(node_load
));
3533 if (self
&& !node_online(self
->node_id
)) {
3534 build_zonelists(self
);
3535 build_zonelist_cache(self
);
3538 for_each_online_node(nid
) {
3539 pg_data_t
*pgdat
= NODE_DATA(nid
);
3541 build_zonelists(pgdat
);
3542 build_zonelist_cache(pgdat
);
3546 * Initialize the boot_pagesets that are going to be used
3547 * for bootstrapping processors. The real pagesets for
3548 * each zone will be allocated later when the per cpu
3549 * allocator is available.
3551 * boot_pagesets are used also for bootstrapping offline
3552 * cpus if the system is already booted because the pagesets
3553 * are needed to initialize allocators on a specific cpu too.
3554 * F.e. the percpu allocator needs the page allocator which
3555 * needs the percpu allocator in order to allocate its pagesets
3556 * (a chicken-egg dilemma).
3558 for_each_possible_cpu(cpu
) {
3559 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3561 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3563 * We now know the "local memory node" for each node--
3564 * i.e., the node of the first zone in the generic zonelist.
3565 * Set up numa_mem percpu variable for on-line cpus. During
3566 * boot, only the boot cpu should be on-line; we'll init the
3567 * secondary cpus' numa_mem as they come on-line. During
3568 * node/memory hotplug, we'll fixup all on-line cpus.
3570 if (cpu_online(cpu
))
3571 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3579 * Called with zonelists_mutex held always
3580 * unless system_state == SYSTEM_BOOTING.
3582 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
3584 set_zonelist_order();
3586 if (system_state
== SYSTEM_BOOTING
) {
3587 __build_all_zonelists(NULL
);
3588 mminit_verify_zonelist();
3589 cpuset_init_current_mems_allowed();
3591 /* we have to stop all cpus to guarantee there is no user
3593 #ifdef CONFIG_MEMORY_HOTPLUG
3595 setup_zone_pageset(zone
);
3597 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
3598 /* cpuset refresh routine should be here */
3600 vm_total_pages
= nr_free_pagecache_pages();
3602 * Disable grouping by mobility if the number of pages in the
3603 * system is too low to allow the mechanism to work. It would be
3604 * more accurate, but expensive to check per-zone. This check is
3605 * made on memory-hotadd so a system can start with mobility
3606 * disabled and enable it later
3608 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3609 page_group_by_mobility_disabled
= 1;
3611 page_group_by_mobility_disabled
= 0;
3613 printk("Built %i zonelists in %s order, mobility grouping %s. "
3614 "Total pages: %ld\n",
3616 zonelist_order_name
[current_zonelist_order
],
3617 page_group_by_mobility_disabled
? "off" : "on",
3620 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3625 * Helper functions to size the waitqueue hash table.
3626 * Essentially these want to choose hash table sizes sufficiently
3627 * large so that collisions trying to wait on pages are rare.
3628 * But in fact, the number of active page waitqueues on typical
3629 * systems is ridiculously low, less than 200. So this is even
3630 * conservative, even though it seems large.
3632 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3633 * waitqueues, i.e. the size of the waitq table given the number of pages.
3635 #define PAGES_PER_WAITQUEUE 256
3637 #ifndef CONFIG_MEMORY_HOTPLUG
3638 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3640 unsigned long size
= 1;
3642 pages
/= PAGES_PER_WAITQUEUE
;
3644 while (size
< pages
)
3648 * Once we have dozens or even hundreds of threads sleeping
3649 * on IO we've got bigger problems than wait queue collision.
3650 * Limit the size of the wait table to a reasonable size.
3652 size
= min(size
, 4096UL);
3654 return max(size
, 4UL);
3658 * A zone's size might be changed by hot-add, so it is not possible to determine
3659 * a suitable size for its wait_table. So we use the maximum size now.
3661 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3663 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3664 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3665 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3667 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3668 * or more by the traditional way. (See above). It equals:
3670 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3671 * ia64(16K page size) : = ( 8G + 4M)byte.
3672 * powerpc (64K page size) : = (32G +16M)byte.
3674 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3681 * This is an integer logarithm so that shifts can be used later
3682 * to extract the more random high bits from the multiplicative
3683 * hash function before the remainder is taken.
3685 static inline unsigned long wait_table_bits(unsigned long size
)
3690 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3693 * Check if a pageblock contains reserved pages
3695 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3699 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3700 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3707 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3708 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3709 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3710 * higher will lead to a bigger reserve which will get freed as contiguous
3711 * blocks as reclaim kicks in
3713 static void setup_zone_migrate_reserve(struct zone
*zone
)
3715 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3717 unsigned long block_migratetype
;
3721 * Get the start pfn, end pfn and the number of blocks to reserve
3722 * We have to be careful to be aligned to pageblock_nr_pages to
3723 * make sure that we always check pfn_valid for the first page in
3726 start_pfn
= zone
->zone_start_pfn
;
3727 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3728 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3729 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3733 * Reserve blocks are generally in place to help high-order atomic
3734 * allocations that are short-lived. A min_free_kbytes value that
3735 * would result in more than 2 reserve blocks for atomic allocations
3736 * is assumed to be in place to help anti-fragmentation for the
3737 * future allocation of hugepages at runtime.
3739 reserve
= min(2, reserve
);
3741 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3742 if (!pfn_valid(pfn
))
3744 page
= pfn_to_page(pfn
);
3746 /* Watch out for overlapping nodes */
3747 if (page_to_nid(page
) != zone_to_nid(zone
))
3750 block_migratetype
= get_pageblock_migratetype(page
);
3752 /* Only test what is necessary when the reserves are not met */
3755 * Blocks with reserved pages will never free, skip
3758 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3759 if (pageblock_is_reserved(pfn
, block_end_pfn
))
3762 /* If this block is reserved, account for it */
3763 if (block_migratetype
== MIGRATE_RESERVE
) {
3768 /* Suitable for reserving if this block is movable */
3769 if (block_migratetype
== MIGRATE_MOVABLE
) {
3770 set_pageblock_migratetype(page
,
3772 move_freepages_block(zone
, page
,
3780 * If the reserve is met and this is a previous reserved block,
3783 if (block_migratetype
== MIGRATE_RESERVE
) {
3784 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3785 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3791 * Initially all pages are reserved - free ones are freed
3792 * up by free_all_bootmem() once the early boot process is
3793 * done. Non-atomic initialization, single-pass.
3795 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3796 unsigned long start_pfn
, enum memmap_context context
)
3799 unsigned long end_pfn
= start_pfn
+ size
;
3803 if (highest_memmap_pfn
< end_pfn
- 1)
3804 highest_memmap_pfn
= end_pfn
- 1;
3806 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3807 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3809 * There can be holes in boot-time mem_map[]s
3810 * handed to this function. They do not
3811 * exist on hotplugged memory.
3813 if (context
== MEMMAP_EARLY
) {
3814 if (!early_pfn_valid(pfn
))
3816 if (!early_pfn_in_nid(pfn
, nid
))
3819 page
= pfn_to_page(pfn
);
3820 set_page_links(page
, zone
, nid
, pfn
);
3821 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3822 init_page_count(page
);
3823 reset_page_mapcount(page
);
3824 SetPageReserved(page
);
3826 * Mark the block movable so that blocks are reserved for
3827 * movable at startup. This will force kernel allocations
3828 * to reserve their blocks rather than leaking throughout
3829 * the address space during boot when many long-lived
3830 * kernel allocations are made. Later some blocks near
3831 * the start are marked MIGRATE_RESERVE by
3832 * setup_zone_migrate_reserve()
3834 * bitmap is created for zone's valid pfn range. but memmap
3835 * can be created for invalid pages (for alignment)
3836 * check here not to call set_pageblock_migratetype() against
3839 if ((z
->zone_start_pfn
<= pfn
)
3840 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3841 && !(pfn
& (pageblock_nr_pages
- 1)))
3842 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3844 INIT_LIST_HEAD(&page
->lru
);
3845 #ifdef WANT_PAGE_VIRTUAL
3846 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3847 if (!is_highmem_idx(zone
))
3848 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3853 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3856 for_each_migratetype_order(order
, t
) {
3857 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3858 zone
->free_area
[order
].nr_free
= 0;
3862 #ifndef __HAVE_ARCH_MEMMAP_INIT
3863 #define memmap_init(size, nid, zone, start_pfn) \
3864 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3867 static int __meminit
zone_batchsize(struct zone
*zone
)
3873 * The per-cpu-pages pools are set to around 1000th of the
3874 * size of the zone. But no more than 1/2 of a meg.
3876 * OK, so we don't know how big the cache is. So guess.
3878 batch
= zone
->present_pages
/ 1024;
3879 if (batch
* PAGE_SIZE
> 512 * 1024)
3880 batch
= (512 * 1024) / PAGE_SIZE
;
3881 batch
/= 4; /* We effectively *= 4 below */
3886 * Clamp the batch to a 2^n - 1 value. Having a power
3887 * of 2 value was found to be more likely to have
3888 * suboptimal cache aliasing properties in some cases.
3890 * For example if 2 tasks are alternately allocating
3891 * batches of pages, one task can end up with a lot
3892 * of pages of one half of the possible page colors
3893 * and the other with pages of the other colors.
3895 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3900 /* The deferral and batching of frees should be suppressed under NOMMU
3903 * The problem is that NOMMU needs to be able to allocate large chunks
3904 * of contiguous memory as there's no hardware page translation to
3905 * assemble apparent contiguous memory from discontiguous pages.
3907 * Queueing large contiguous runs of pages for batching, however,
3908 * causes the pages to actually be freed in smaller chunks. As there
3909 * can be a significant delay between the individual batches being
3910 * recycled, this leads to the once large chunks of space being
3911 * fragmented and becoming unavailable for high-order allocations.
3917 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3919 struct per_cpu_pages
*pcp
;
3922 memset(p
, 0, sizeof(*p
));
3926 pcp
->high
= 6 * batch
;
3927 pcp
->batch
= max(1UL, 1 * batch
);
3928 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3929 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3933 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3934 * to the value high for the pageset p.
3937 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3940 struct per_cpu_pages
*pcp
;
3944 pcp
->batch
= max(1UL, high
/4);
3945 if ((high
/4) > (PAGE_SHIFT
* 8))
3946 pcp
->batch
= PAGE_SHIFT
* 8;
3949 static void __meminit
setup_zone_pageset(struct zone
*zone
)
3953 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
3955 for_each_possible_cpu(cpu
) {
3956 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
3958 setup_pageset(pcp
, zone_batchsize(zone
));
3960 if (percpu_pagelist_fraction
)
3961 setup_pagelist_highmark(pcp
,
3962 (zone
->present_pages
/
3963 percpu_pagelist_fraction
));
3968 * Allocate per cpu pagesets and initialize them.
3969 * Before this call only boot pagesets were available.
3971 void __init
setup_per_cpu_pageset(void)
3975 for_each_populated_zone(zone
)
3976 setup_zone_pageset(zone
);
3979 static noinline __init_refok
3980 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3983 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3987 * The per-page waitqueue mechanism uses hashed waitqueues
3990 zone
->wait_table_hash_nr_entries
=
3991 wait_table_hash_nr_entries(zone_size_pages
);
3992 zone
->wait_table_bits
=
3993 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3994 alloc_size
= zone
->wait_table_hash_nr_entries
3995 * sizeof(wait_queue_head_t
);
3997 if (!slab_is_available()) {
3998 zone
->wait_table
= (wait_queue_head_t
*)
3999 alloc_bootmem_node_nopanic(pgdat
, alloc_size
);
4002 * This case means that a zone whose size was 0 gets new memory
4003 * via memory hot-add.
4004 * But it may be the case that a new node was hot-added. In
4005 * this case vmalloc() will not be able to use this new node's
4006 * memory - this wait_table must be initialized to use this new
4007 * node itself as well.
4008 * To use this new node's memory, further consideration will be
4011 zone
->wait_table
= vmalloc(alloc_size
);
4013 if (!zone
->wait_table
)
4016 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4017 init_waitqueue_head(zone
->wait_table
+ i
);
4022 static __meminit
void zone_pcp_init(struct zone
*zone
)
4025 * per cpu subsystem is not up at this point. The following code
4026 * relies on the ability of the linker to provide the
4027 * offset of a (static) per cpu variable into the per cpu area.
4029 zone
->pageset
= &boot_pageset
;
4031 if (zone
->present_pages
)
4032 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4033 zone
->name
, zone
->present_pages
,
4034 zone_batchsize(zone
));
4037 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4038 unsigned long zone_start_pfn
,
4040 enum memmap_context context
)
4042 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4044 ret
= zone_wait_table_init(zone
, size
);
4047 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4049 zone
->zone_start_pfn
= zone_start_pfn
;
4051 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4052 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4054 (unsigned long)zone_idx(zone
),
4055 zone_start_pfn
, (zone_start_pfn
+ size
));
4057 zone_init_free_lists(zone
);
4062 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4063 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4065 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4066 * Architectures may implement their own version but if add_active_range()
4067 * was used and there are no special requirements, this is a convenient
4070 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
4072 unsigned long start_pfn
, end_pfn
;
4075 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4076 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
4078 /* This is a memory hole */
4081 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4083 int __meminit
early_pfn_to_nid(unsigned long pfn
)
4087 nid
= __early_pfn_to_nid(pfn
);
4090 /* just returns 0 */
4094 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4095 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
4099 nid
= __early_pfn_to_nid(pfn
);
4100 if (nid
>= 0 && nid
!= node
)
4107 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
4108 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4109 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
4111 * If an architecture guarantees that all ranges registered with
4112 * add_active_ranges() contain no holes and may be freed, this
4113 * this function may be used instead of calling free_bootmem() manually.
4115 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4117 unsigned long start_pfn
, end_pfn
;
4120 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4121 start_pfn
= min(start_pfn
, max_low_pfn
);
4122 end_pfn
= min(end_pfn
, max_low_pfn
);
4124 if (start_pfn
< end_pfn
)
4125 free_bootmem_node(NODE_DATA(this_nid
),
4126 PFN_PHYS(start_pfn
),
4127 (end_pfn
- start_pfn
) << PAGE_SHIFT
);
4132 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4133 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4135 * If an architecture guarantees that all ranges registered with
4136 * add_active_ranges() contain no holes and may be freed, this
4137 * function may be used instead of calling memory_present() manually.
4139 void __init
sparse_memory_present_with_active_regions(int nid
)
4141 unsigned long start_pfn
, end_pfn
;
4144 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4145 memory_present(this_nid
, start_pfn
, end_pfn
);
4149 * get_pfn_range_for_nid - Return the start and end page frames for a node
4150 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4151 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4152 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4154 * It returns the start and end page frame of a node based on information
4155 * provided by an arch calling add_active_range(). If called for a node
4156 * with no available memory, a warning is printed and the start and end
4159 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4160 unsigned long *start_pfn
, unsigned long *end_pfn
)
4162 unsigned long this_start_pfn
, this_end_pfn
;
4168 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4169 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4170 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4173 if (*start_pfn
== -1UL)
4178 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4179 * assumption is made that zones within a node are ordered in monotonic
4180 * increasing memory addresses so that the "highest" populated zone is used
4182 static void __init
find_usable_zone_for_movable(void)
4185 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4186 if (zone_index
== ZONE_MOVABLE
)
4189 if (arch_zone_highest_possible_pfn
[zone_index
] >
4190 arch_zone_lowest_possible_pfn
[zone_index
])
4194 VM_BUG_ON(zone_index
== -1);
4195 movable_zone
= zone_index
;
4199 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4200 * because it is sized independent of architecture. Unlike the other zones,
4201 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4202 * in each node depending on the size of each node and how evenly kernelcore
4203 * is distributed. This helper function adjusts the zone ranges
4204 * provided by the architecture for a given node by using the end of the
4205 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4206 * zones within a node are in order of monotonic increases memory addresses
4208 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4209 unsigned long zone_type
,
4210 unsigned long node_start_pfn
,
4211 unsigned long node_end_pfn
,
4212 unsigned long *zone_start_pfn
,
4213 unsigned long *zone_end_pfn
)
4215 /* Only adjust if ZONE_MOVABLE is on this node */
4216 if (zone_movable_pfn
[nid
]) {
4217 /* Size ZONE_MOVABLE */
4218 if (zone_type
== ZONE_MOVABLE
) {
4219 *zone_start_pfn
= zone_movable_pfn
[nid
];
4220 *zone_end_pfn
= min(node_end_pfn
,
4221 arch_zone_highest_possible_pfn
[movable_zone
]);
4223 /* Adjust for ZONE_MOVABLE starting within this range */
4224 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4225 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4226 *zone_end_pfn
= zone_movable_pfn
[nid
];
4228 /* Check if this whole range is within ZONE_MOVABLE */
4229 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4230 *zone_start_pfn
= *zone_end_pfn
;
4235 * Return the number of pages a zone spans in a node, including holes
4236 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4238 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4239 unsigned long zone_type
,
4240 unsigned long *ignored
)
4242 unsigned long node_start_pfn
, node_end_pfn
;
4243 unsigned long zone_start_pfn
, zone_end_pfn
;
4245 /* Get the start and end of the node and zone */
4246 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4247 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4248 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4249 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4250 node_start_pfn
, node_end_pfn
,
4251 &zone_start_pfn
, &zone_end_pfn
);
4253 /* Check that this node has pages within the zone's required range */
4254 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4257 /* Move the zone boundaries inside the node if necessary */
4258 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4259 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4261 /* Return the spanned pages */
4262 return zone_end_pfn
- zone_start_pfn
;
4266 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4267 * then all holes in the requested range will be accounted for.
4269 unsigned long __meminit
__absent_pages_in_range(int nid
,
4270 unsigned long range_start_pfn
,
4271 unsigned long range_end_pfn
)
4273 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4274 unsigned long start_pfn
, end_pfn
;
4277 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4278 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4279 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4280 nr_absent
-= end_pfn
- start_pfn
;
4286 * absent_pages_in_range - Return number of page frames in holes within a range
4287 * @start_pfn: The start PFN to start searching for holes
4288 * @end_pfn: The end PFN to stop searching for holes
4290 * It returns the number of pages frames in memory holes within a range.
4292 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4293 unsigned long end_pfn
)
4295 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4298 /* Return the number of page frames in holes in a zone on a node */
4299 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4300 unsigned long zone_type
,
4301 unsigned long *ignored
)
4303 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4304 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4305 unsigned long node_start_pfn
, node_end_pfn
;
4306 unsigned long zone_start_pfn
, zone_end_pfn
;
4308 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4309 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4310 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4312 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4313 node_start_pfn
, node_end_pfn
,
4314 &zone_start_pfn
, &zone_end_pfn
);
4315 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4318 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4319 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4320 unsigned long zone_type
,
4321 unsigned long *zones_size
)
4323 return zones_size
[zone_type
];
4326 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4327 unsigned long zone_type
,
4328 unsigned long *zholes_size
)
4333 return zholes_size
[zone_type
];
4336 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4338 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4339 unsigned long *zones_size
, unsigned long *zholes_size
)
4341 unsigned long realtotalpages
, totalpages
= 0;
4344 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4345 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4347 pgdat
->node_spanned_pages
= totalpages
;
4349 realtotalpages
= totalpages
;
4350 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4352 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4354 pgdat
->node_present_pages
= realtotalpages
;
4355 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4359 #ifndef CONFIG_SPARSEMEM
4361 * Calculate the size of the zone->blockflags rounded to an unsigned long
4362 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4363 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4364 * round what is now in bits to nearest long in bits, then return it in
4367 static unsigned long __init
usemap_size(unsigned long zonesize
)
4369 unsigned long usemapsize
;
4371 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4372 usemapsize
= usemapsize
>> pageblock_order
;
4373 usemapsize
*= NR_PAGEBLOCK_BITS
;
4374 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4376 return usemapsize
/ 8;
4379 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4380 struct zone
*zone
, unsigned long zonesize
)
4382 unsigned long usemapsize
= usemap_size(zonesize
);
4383 zone
->pageblock_flags
= NULL
;
4385 zone
->pageblock_flags
= alloc_bootmem_node_nopanic(pgdat
,
4389 static inline void setup_usemap(struct pglist_data
*pgdat
,
4390 struct zone
*zone
, unsigned long zonesize
) {}
4391 #endif /* CONFIG_SPARSEMEM */
4393 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4395 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4396 void __init
set_pageblock_order(void)
4400 /* Check that pageblock_nr_pages has not already been setup */
4401 if (pageblock_order
)
4404 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4405 order
= HUGETLB_PAGE_ORDER
;
4407 order
= MAX_ORDER
- 1;
4410 * Assume the largest contiguous order of interest is a huge page.
4411 * This value may be variable depending on boot parameters on IA64 and
4414 pageblock_order
= order
;
4416 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4419 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4420 * is unused as pageblock_order is set at compile-time. See
4421 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4424 void __init
set_pageblock_order(void)
4428 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4431 * Set up the zone data structures:
4432 * - mark all pages reserved
4433 * - mark all memory queues empty
4434 * - clear the memory bitmaps
4436 * NOTE: pgdat should get zeroed by caller.
4438 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4439 unsigned long *zones_size
, unsigned long *zholes_size
)
4442 int nid
= pgdat
->node_id
;
4443 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4446 pgdat_resize_init(pgdat
);
4447 init_waitqueue_head(&pgdat
->kswapd_wait
);
4448 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
4449 pgdat_page_cgroup_init(pgdat
);
4451 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4452 struct zone
*zone
= pgdat
->node_zones
+ j
;
4453 unsigned long size
, realsize
, memmap_pages
;
4455 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4456 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
4460 * Adjust realsize so that it accounts for how much memory
4461 * is used by this zone for memmap. This affects the watermark
4462 * and per-cpu initialisations
4465 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
4466 if (realsize
>= memmap_pages
) {
4467 realsize
-= memmap_pages
;
4470 " %s zone: %lu pages used for memmap\n",
4471 zone_names
[j
], memmap_pages
);
4474 " %s zone: %lu pages exceeds realsize %lu\n",
4475 zone_names
[j
], memmap_pages
, realsize
);
4477 /* Account for reserved pages */
4478 if (j
== 0 && realsize
> dma_reserve
) {
4479 realsize
-= dma_reserve
;
4480 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4481 zone_names
[0], dma_reserve
);
4484 if (!is_highmem_idx(j
))
4485 nr_kernel_pages
+= realsize
;
4486 nr_all_pages
+= realsize
;
4488 zone
->spanned_pages
= size
;
4489 zone
->present_pages
= realsize
;
4490 #if defined CONFIG_COMPACTION || defined CONFIG_CMA
4491 zone
->compact_cached_free_pfn
= zone
->zone_start_pfn
+
4492 zone
->spanned_pages
;
4493 zone
->compact_cached_free_pfn
&= ~(pageblock_nr_pages
-1);
4497 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
4499 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
4501 zone
->name
= zone_names
[j
];
4502 spin_lock_init(&zone
->lock
);
4503 spin_lock_init(&zone
->lru_lock
);
4504 zone_seqlock_init(zone
);
4505 zone
->zone_pgdat
= pgdat
;
4507 zone_pcp_init(zone
);
4508 lruvec_init(&zone
->lruvec
, zone
);
4512 set_pageblock_order();
4513 setup_usemap(pgdat
, zone
, size
);
4514 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4515 size
, MEMMAP_EARLY
);
4517 memmap_init(size
, nid
, j
, zone_start_pfn
);
4518 zone_start_pfn
+= size
;
4522 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4524 /* Skip empty nodes */
4525 if (!pgdat
->node_spanned_pages
)
4528 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4529 /* ia64 gets its own node_mem_map, before this, without bootmem */
4530 if (!pgdat
->node_mem_map
) {
4531 unsigned long size
, start
, end
;
4535 * The zone's endpoints aren't required to be MAX_ORDER
4536 * aligned but the node_mem_map endpoints must be in order
4537 * for the buddy allocator to function correctly.
4539 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4540 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4541 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4542 size
= (end
- start
) * sizeof(struct page
);
4543 map
= alloc_remap(pgdat
->node_id
, size
);
4545 map
= alloc_bootmem_node_nopanic(pgdat
, size
);
4546 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4548 #ifndef CONFIG_NEED_MULTIPLE_NODES
4550 * With no DISCONTIG, the global mem_map is just set as node 0's
4552 if (pgdat
== NODE_DATA(0)) {
4553 mem_map
= NODE_DATA(0)->node_mem_map
;
4554 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4555 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4556 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4557 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4560 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4563 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4564 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4566 pg_data_t
*pgdat
= NODE_DATA(nid
);
4568 /* pg_data_t should be reset to zero when it's allocated */
4569 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
4571 pgdat
->node_id
= nid
;
4572 pgdat
->node_start_pfn
= node_start_pfn
;
4573 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4575 alloc_node_mem_map(pgdat
);
4576 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4577 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4578 nid
, (unsigned long)pgdat
,
4579 (unsigned long)pgdat
->node_mem_map
);
4582 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4585 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4587 #if MAX_NUMNODES > 1
4589 * Figure out the number of possible node ids.
4591 static void __init
setup_nr_node_ids(void)
4594 unsigned int highest
= 0;
4596 for_each_node_mask(node
, node_possible_map
)
4598 nr_node_ids
= highest
+ 1;
4601 static inline void setup_nr_node_ids(void)
4607 * node_map_pfn_alignment - determine the maximum internode alignment
4609 * This function should be called after node map is populated and sorted.
4610 * It calculates the maximum power of two alignment which can distinguish
4613 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4614 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4615 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4616 * shifted, 1GiB is enough and this function will indicate so.
4618 * This is used to test whether pfn -> nid mapping of the chosen memory
4619 * model has fine enough granularity to avoid incorrect mapping for the
4620 * populated node map.
4622 * Returns the determined alignment in pfn's. 0 if there is no alignment
4623 * requirement (single node).
4625 unsigned long __init
node_map_pfn_alignment(void)
4627 unsigned long accl_mask
= 0, last_end
= 0;
4628 unsigned long start
, end
, mask
;
4632 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4633 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4640 * Start with a mask granular enough to pin-point to the
4641 * start pfn and tick off bits one-by-one until it becomes
4642 * too coarse to separate the current node from the last.
4644 mask
= ~((1 << __ffs(start
)) - 1);
4645 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4648 /* accumulate all internode masks */
4652 /* convert mask to number of pages */
4653 return ~accl_mask
+ 1;
4656 /* Find the lowest pfn for a node */
4657 static unsigned long __init
find_min_pfn_for_node(int nid
)
4659 unsigned long min_pfn
= ULONG_MAX
;
4660 unsigned long start_pfn
;
4663 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4664 min_pfn
= min(min_pfn
, start_pfn
);
4666 if (min_pfn
== ULONG_MAX
) {
4668 "Could not find start_pfn for node %d\n", nid
);
4676 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4678 * It returns the minimum PFN based on information provided via
4679 * add_active_range().
4681 unsigned long __init
find_min_pfn_with_active_regions(void)
4683 return find_min_pfn_for_node(MAX_NUMNODES
);
4687 * early_calculate_totalpages()
4688 * Sum pages in active regions for movable zone.
4689 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4691 static unsigned long __init
early_calculate_totalpages(void)
4693 unsigned long totalpages
= 0;
4694 unsigned long start_pfn
, end_pfn
;
4697 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
4698 unsigned long pages
= end_pfn
- start_pfn
;
4700 totalpages
+= pages
;
4702 node_set_state(nid
, N_HIGH_MEMORY
);
4708 * Find the PFN the Movable zone begins in each node. Kernel memory
4709 * is spread evenly between nodes as long as the nodes have enough
4710 * memory. When they don't, some nodes will have more kernelcore than
4713 static void __init
find_zone_movable_pfns_for_nodes(void)
4716 unsigned long usable_startpfn
;
4717 unsigned long kernelcore_node
, kernelcore_remaining
;
4718 /* save the state before borrow the nodemask */
4719 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4720 unsigned long totalpages
= early_calculate_totalpages();
4721 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4724 * If movablecore was specified, calculate what size of
4725 * kernelcore that corresponds so that memory usable for
4726 * any allocation type is evenly spread. If both kernelcore
4727 * and movablecore are specified, then the value of kernelcore
4728 * will be used for required_kernelcore if it's greater than
4729 * what movablecore would have allowed.
4731 if (required_movablecore
) {
4732 unsigned long corepages
;
4735 * Round-up so that ZONE_MOVABLE is at least as large as what
4736 * was requested by the user
4738 required_movablecore
=
4739 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4740 corepages
= totalpages
- required_movablecore
;
4742 required_kernelcore
= max(required_kernelcore
, corepages
);
4745 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4746 if (!required_kernelcore
)
4749 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4750 find_usable_zone_for_movable();
4751 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4754 /* Spread kernelcore memory as evenly as possible throughout nodes */
4755 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4756 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4757 unsigned long start_pfn
, end_pfn
;
4760 * Recalculate kernelcore_node if the division per node
4761 * now exceeds what is necessary to satisfy the requested
4762 * amount of memory for the kernel
4764 if (required_kernelcore
< kernelcore_node
)
4765 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4768 * As the map is walked, we track how much memory is usable
4769 * by the kernel using kernelcore_remaining. When it is
4770 * 0, the rest of the node is usable by ZONE_MOVABLE
4772 kernelcore_remaining
= kernelcore_node
;
4774 /* Go through each range of PFNs within this node */
4775 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4776 unsigned long size_pages
;
4778 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
4779 if (start_pfn
>= end_pfn
)
4782 /* Account for what is only usable for kernelcore */
4783 if (start_pfn
< usable_startpfn
) {
4784 unsigned long kernel_pages
;
4785 kernel_pages
= min(end_pfn
, usable_startpfn
)
4788 kernelcore_remaining
-= min(kernel_pages
,
4789 kernelcore_remaining
);
4790 required_kernelcore
-= min(kernel_pages
,
4791 required_kernelcore
);
4793 /* Continue if range is now fully accounted */
4794 if (end_pfn
<= usable_startpfn
) {
4797 * Push zone_movable_pfn to the end so
4798 * that if we have to rebalance
4799 * kernelcore across nodes, we will
4800 * not double account here
4802 zone_movable_pfn
[nid
] = end_pfn
;
4805 start_pfn
= usable_startpfn
;
4809 * The usable PFN range for ZONE_MOVABLE is from
4810 * start_pfn->end_pfn. Calculate size_pages as the
4811 * number of pages used as kernelcore
4813 size_pages
= end_pfn
- start_pfn
;
4814 if (size_pages
> kernelcore_remaining
)
4815 size_pages
= kernelcore_remaining
;
4816 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4819 * Some kernelcore has been met, update counts and
4820 * break if the kernelcore for this node has been
4823 required_kernelcore
-= min(required_kernelcore
,
4825 kernelcore_remaining
-= size_pages
;
4826 if (!kernelcore_remaining
)
4832 * If there is still required_kernelcore, we do another pass with one
4833 * less node in the count. This will push zone_movable_pfn[nid] further
4834 * along on the nodes that still have memory until kernelcore is
4838 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4841 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4842 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4843 zone_movable_pfn
[nid
] =
4844 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4847 /* restore the node_state */
4848 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4851 /* Any regular memory on that node ? */
4852 static void __init
check_for_regular_memory(pg_data_t
*pgdat
)
4854 #ifdef CONFIG_HIGHMEM
4855 enum zone_type zone_type
;
4857 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4858 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4859 if (zone
->present_pages
) {
4860 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4868 * free_area_init_nodes - Initialise all pg_data_t and zone data
4869 * @max_zone_pfn: an array of max PFNs for each zone
4871 * This will call free_area_init_node() for each active node in the system.
4872 * Using the page ranges provided by add_active_range(), the size of each
4873 * zone in each node and their holes is calculated. If the maximum PFN
4874 * between two adjacent zones match, it is assumed that the zone is empty.
4875 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4876 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4877 * starts where the previous one ended. For example, ZONE_DMA32 starts
4878 * at arch_max_dma_pfn.
4880 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4882 unsigned long start_pfn
, end_pfn
;
4885 /* Record where the zone boundaries are */
4886 memset(arch_zone_lowest_possible_pfn
, 0,
4887 sizeof(arch_zone_lowest_possible_pfn
));
4888 memset(arch_zone_highest_possible_pfn
, 0,
4889 sizeof(arch_zone_highest_possible_pfn
));
4890 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4891 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4892 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4893 if (i
== ZONE_MOVABLE
)
4895 arch_zone_lowest_possible_pfn
[i
] =
4896 arch_zone_highest_possible_pfn
[i
-1];
4897 arch_zone_highest_possible_pfn
[i
] =
4898 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4900 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4901 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4903 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4904 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4905 find_zone_movable_pfns_for_nodes();
4907 /* Print out the zone ranges */
4908 printk("Zone ranges:\n");
4909 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4910 if (i
== ZONE_MOVABLE
)
4912 printk(KERN_CONT
" %-8s ", zone_names
[i
]);
4913 if (arch_zone_lowest_possible_pfn
[i
] ==
4914 arch_zone_highest_possible_pfn
[i
])
4915 printk(KERN_CONT
"empty\n");
4917 printk(KERN_CONT
"[mem %0#10lx-%0#10lx]\n",
4918 arch_zone_lowest_possible_pfn
[i
] << PAGE_SHIFT
,
4919 (arch_zone_highest_possible_pfn
[i
]
4920 << PAGE_SHIFT
) - 1);
4923 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4924 printk("Movable zone start for each node\n");
4925 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4926 if (zone_movable_pfn
[i
])
4927 printk(" Node %d: %#010lx\n", i
,
4928 zone_movable_pfn
[i
] << PAGE_SHIFT
);
4931 /* Print out the early_node_map[] */
4932 printk("Early memory node ranges\n");
4933 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4934 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid
,
4935 start_pfn
<< PAGE_SHIFT
, (end_pfn
<< PAGE_SHIFT
) - 1);
4937 /* Initialise every node */
4938 mminit_verify_pageflags_layout();
4939 setup_nr_node_ids();
4940 for_each_online_node(nid
) {
4941 pg_data_t
*pgdat
= NODE_DATA(nid
);
4942 free_area_init_node(nid
, NULL
,
4943 find_min_pfn_for_node(nid
), NULL
);
4945 /* Any memory on that node */
4946 if (pgdat
->node_present_pages
)
4947 node_set_state(nid
, N_HIGH_MEMORY
);
4948 check_for_regular_memory(pgdat
);
4952 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4954 unsigned long long coremem
;
4958 coremem
= memparse(p
, &p
);
4959 *core
= coremem
>> PAGE_SHIFT
;
4961 /* Paranoid check that UL is enough for the coremem value */
4962 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4968 * kernelcore=size sets the amount of memory for use for allocations that
4969 * cannot be reclaimed or migrated.
4971 static int __init
cmdline_parse_kernelcore(char *p
)
4973 return cmdline_parse_core(p
, &required_kernelcore
);
4977 * movablecore=size sets the amount of memory for use for allocations that
4978 * can be reclaimed or migrated.
4980 static int __init
cmdline_parse_movablecore(char *p
)
4982 return cmdline_parse_core(p
, &required_movablecore
);
4985 early_param("kernelcore", cmdline_parse_kernelcore
);
4986 early_param("movablecore", cmdline_parse_movablecore
);
4988 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4991 * set_dma_reserve - set the specified number of pages reserved in the first zone
4992 * @new_dma_reserve: The number of pages to mark reserved
4994 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4995 * In the DMA zone, a significant percentage may be consumed by kernel image
4996 * and other unfreeable allocations which can skew the watermarks badly. This
4997 * function may optionally be used to account for unfreeable pages in the
4998 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4999 * smaller per-cpu batchsize.
5001 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5003 dma_reserve
= new_dma_reserve
;
5006 void __init
free_area_init(unsigned long *zones_size
)
5008 free_area_init_node(0, zones_size
,
5009 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5012 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5013 unsigned long action
, void *hcpu
)
5015 int cpu
= (unsigned long)hcpu
;
5017 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5018 lru_add_drain_cpu(cpu
);
5022 * Spill the event counters of the dead processor
5023 * into the current processors event counters.
5024 * This artificially elevates the count of the current
5027 vm_events_fold_cpu(cpu
);
5030 * Zero the differential counters of the dead processor
5031 * so that the vm statistics are consistent.
5033 * This is only okay since the processor is dead and cannot
5034 * race with what we are doing.
5036 refresh_cpu_vm_stats(cpu
);
5041 void __init
page_alloc_init(void)
5043 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5047 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5048 * or min_free_kbytes changes.
5050 static void calculate_totalreserve_pages(void)
5052 struct pglist_data
*pgdat
;
5053 unsigned long reserve_pages
= 0;
5054 enum zone_type i
, j
;
5056 for_each_online_pgdat(pgdat
) {
5057 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5058 struct zone
*zone
= pgdat
->node_zones
+ i
;
5059 unsigned long max
= 0;
5061 /* Find valid and maximum lowmem_reserve in the zone */
5062 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5063 if (zone
->lowmem_reserve
[j
] > max
)
5064 max
= zone
->lowmem_reserve
[j
];
5067 /* we treat the high watermark as reserved pages. */
5068 max
+= high_wmark_pages(zone
);
5070 if (max
> zone
->present_pages
)
5071 max
= zone
->present_pages
;
5072 reserve_pages
+= max
;
5074 * Lowmem reserves are not available to
5075 * GFP_HIGHUSER page cache allocations and
5076 * kswapd tries to balance zones to their high
5077 * watermark. As a result, neither should be
5078 * regarded as dirtyable memory, to prevent a
5079 * situation where reclaim has to clean pages
5080 * in order to balance the zones.
5082 zone
->dirty_balance_reserve
= max
;
5085 dirty_balance_reserve
= reserve_pages
;
5086 totalreserve_pages
= reserve_pages
;
5090 * setup_per_zone_lowmem_reserve - called whenever
5091 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5092 * has a correct pages reserved value, so an adequate number of
5093 * pages are left in the zone after a successful __alloc_pages().
5095 static void setup_per_zone_lowmem_reserve(void)
5097 struct pglist_data
*pgdat
;
5098 enum zone_type j
, idx
;
5100 for_each_online_pgdat(pgdat
) {
5101 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5102 struct zone
*zone
= pgdat
->node_zones
+ j
;
5103 unsigned long present_pages
= zone
->present_pages
;
5105 zone
->lowmem_reserve
[j
] = 0;
5109 struct zone
*lower_zone
;
5113 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5114 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5116 lower_zone
= pgdat
->node_zones
+ idx
;
5117 lower_zone
->lowmem_reserve
[j
] = present_pages
/
5118 sysctl_lowmem_reserve_ratio
[idx
];
5119 present_pages
+= lower_zone
->present_pages
;
5124 /* update totalreserve_pages */
5125 calculate_totalreserve_pages();
5128 static void __setup_per_zone_wmarks(void)
5130 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5131 unsigned long lowmem_pages
= 0;
5133 unsigned long flags
;
5135 /* Calculate total number of !ZONE_HIGHMEM pages */
5136 for_each_zone(zone
) {
5137 if (!is_highmem(zone
))
5138 lowmem_pages
+= zone
->present_pages
;
5141 for_each_zone(zone
) {
5144 spin_lock_irqsave(&zone
->lock
, flags
);
5145 tmp
= (u64
)pages_min
* zone
->present_pages
;
5146 do_div(tmp
, lowmem_pages
);
5147 if (is_highmem(zone
)) {
5149 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5150 * need highmem pages, so cap pages_min to a small
5153 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5154 * deltas controls asynch page reclaim, and so should
5155 * not be capped for highmem.
5159 min_pages
= zone
->present_pages
/ 1024;
5160 if (min_pages
< SWAP_CLUSTER_MAX
)
5161 min_pages
= SWAP_CLUSTER_MAX
;
5162 if (min_pages
> 128)
5164 zone
->watermark
[WMARK_MIN
] = min_pages
;
5167 * If it's a lowmem zone, reserve a number of pages
5168 * proportionate to the zone's size.
5170 zone
->watermark
[WMARK_MIN
] = tmp
;
5173 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
5174 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
5176 zone
->watermark
[WMARK_MIN
] += cma_wmark_pages(zone
);
5177 zone
->watermark
[WMARK_LOW
] += cma_wmark_pages(zone
);
5178 zone
->watermark
[WMARK_HIGH
] += cma_wmark_pages(zone
);
5180 setup_zone_migrate_reserve(zone
);
5181 spin_unlock_irqrestore(&zone
->lock
, flags
);
5184 /* update totalreserve_pages */
5185 calculate_totalreserve_pages();
5189 * setup_per_zone_wmarks - called when min_free_kbytes changes
5190 * or when memory is hot-{added|removed}
5192 * Ensures that the watermark[min,low,high] values for each zone are set
5193 * correctly with respect to min_free_kbytes.
5195 void setup_per_zone_wmarks(void)
5197 mutex_lock(&zonelists_mutex
);
5198 __setup_per_zone_wmarks();
5199 mutex_unlock(&zonelists_mutex
);
5203 * The inactive anon list should be small enough that the VM never has to
5204 * do too much work, but large enough that each inactive page has a chance
5205 * to be referenced again before it is swapped out.
5207 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5208 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5209 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5210 * the anonymous pages are kept on the inactive list.
5213 * memory ratio inactive anon
5214 * -------------------------------------
5223 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
5225 unsigned int gb
, ratio
;
5227 /* Zone size in gigabytes */
5228 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
5230 ratio
= int_sqrt(10 * gb
);
5234 zone
->inactive_ratio
= ratio
;
5237 static void __meminit
setup_per_zone_inactive_ratio(void)
5242 calculate_zone_inactive_ratio(zone
);
5246 * Initialise min_free_kbytes.
5248 * For small machines we want it small (128k min). For large machines
5249 * we want it large (64MB max). But it is not linear, because network
5250 * bandwidth does not increase linearly with machine size. We use
5252 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5253 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5269 int __meminit
init_per_zone_wmark_min(void)
5271 unsigned long lowmem_kbytes
;
5273 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5275 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5276 if (min_free_kbytes
< 128)
5277 min_free_kbytes
= 128;
5278 if (min_free_kbytes
> 65536)
5279 min_free_kbytes
= 65536;
5280 setup_per_zone_wmarks();
5281 refresh_zone_stat_thresholds();
5282 setup_per_zone_lowmem_reserve();
5283 setup_per_zone_inactive_ratio();
5286 module_init(init_per_zone_wmark_min
)
5289 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5290 * that we can call two helper functions whenever min_free_kbytes
5293 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5294 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5296 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5298 setup_per_zone_wmarks();
5303 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5304 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5309 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5314 zone
->min_unmapped_pages
= (zone
->present_pages
*
5315 sysctl_min_unmapped_ratio
) / 100;
5319 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5320 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5325 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5330 zone
->min_slab_pages
= (zone
->present_pages
*
5331 sysctl_min_slab_ratio
) / 100;
5337 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5338 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5339 * whenever sysctl_lowmem_reserve_ratio changes.
5341 * The reserve ratio obviously has absolutely no relation with the
5342 * minimum watermarks. The lowmem reserve ratio can only make sense
5343 * if in function of the boot time zone sizes.
5345 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5346 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5348 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5349 setup_per_zone_lowmem_reserve();
5354 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5355 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5356 * can have before it gets flushed back to buddy allocator.
5359 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5360 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5366 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5367 if (!write
|| (ret
< 0))
5369 for_each_populated_zone(zone
) {
5370 for_each_possible_cpu(cpu
) {
5372 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5373 setup_pagelist_highmark(
5374 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5380 int hashdist
= HASHDIST_DEFAULT
;
5383 static int __init
set_hashdist(char *str
)
5387 hashdist
= simple_strtoul(str
, &str
, 0);
5390 __setup("hashdist=", set_hashdist
);
5394 * allocate a large system hash table from bootmem
5395 * - it is assumed that the hash table must contain an exact power-of-2
5396 * quantity of entries
5397 * - limit is the number of hash buckets, not the total allocation size
5399 void *__init
alloc_large_system_hash(const char *tablename
,
5400 unsigned long bucketsize
,
5401 unsigned long numentries
,
5404 unsigned int *_hash_shift
,
5405 unsigned int *_hash_mask
,
5406 unsigned long low_limit
,
5407 unsigned long high_limit
)
5409 unsigned long long max
= high_limit
;
5410 unsigned long log2qty
, size
;
5413 /* allow the kernel cmdline to have a say */
5415 /* round applicable memory size up to nearest megabyte */
5416 numentries
= nr_kernel_pages
;
5417 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5418 numentries
>>= 20 - PAGE_SHIFT
;
5419 numentries
<<= 20 - PAGE_SHIFT
;
5421 /* limit to 1 bucket per 2^scale bytes of low memory */
5422 if (scale
> PAGE_SHIFT
)
5423 numentries
>>= (scale
- PAGE_SHIFT
);
5425 numentries
<<= (PAGE_SHIFT
- scale
);
5427 /* Make sure we've got at least a 0-order allocation.. */
5428 if (unlikely(flags
& HASH_SMALL
)) {
5429 /* Makes no sense without HASH_EARLY */
5430 WARN_ON(!(flags
& HASH_EARLY
));
5431 if (!(numentries
>> *_hash_shift
)) {
5432 numentries
= 1UL << *_hash_shift
;
5433 BUG_ON(!numentries
);
5435 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5436 numentries
= PAGE_SIZE
/ bucketsize
;
5438 numentries
= roundup_pow_of_two(numentries
);
5440 /* limit allocation size to 1/16 total memory by default */
5442 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5443 do_div(max
, bucketsize
);
5445 max
= min(max
, 0x80000000ULL
);
5447 if (numentries
< low_limit
)
5448 numentries
= low_limit
;
5449 if (numentries
> max
)
5452 log2qty
= ilog2(numentries
);
5455 size
= bucketsize
<< log2qty
;
5456 if (flags
& HASH_EARLY
)
5457 table
= alloc_bootmem_nopanic(size
);
5459 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5462 * If bucketsize is not a power-of-two, we may free
5463 * some pages at the end of hash table which
5464 * alloc_pages_exact() automatically does
5466 if (get_order(size
) < MAX_ORDER
) {
5467 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5468 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5471 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5474 panic("Failed to allocate %s hash table\n", tablename
);
5476 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5479 ilog2(size
) - PAGE_SHIFT
,
5483 *_hash_shift
= log2qty
;
5485 *_hash_mask
= (1 << log2qty
) - 1;
5490 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5491 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5494 #ifdef CONFIG_SPARSEMEM
5495 return __pfn_to_section(pfn
)->pageblock_flags
;
5497 return zone
->pageblock_flags
;
5498 #endif /* CONFIG_SPARSEMEM */
5501 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5503 #ifdef CONFIG_SPARSEMEM
5504 pfn
&= (PAGES_PER_SECTION
-1);
5505 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5507 pfn
= pfn
- zone
->zone_start_pfn
;
5508 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5509 #endif /* CONFIG_SPARSEMEM */
5513 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5514 * @page: The page within the block of interest
5515 * @start_bitidx: The first bit of interest to retrieve
5516 * @end_bitidx: The last bit of interest
5517 * returns pageblock_bits flags
5519 unsigned long get_pageblock_flags_group(struct page
*page
,
5520 int start_bitidx
, int end_bitidx
)
5523 unsigned long *bitmap
;
5524 unsigned long pfn
, bitidx
;
5525 unsigned long flags
= 0;
5526 unsigned long value
= 1;
5528 zone
= page_zone(page
);
5529 pfn
= page_to_pfn(page
);
5530 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5531 bitidx
= pfn_to_bitidx(zone
, pfn
);
5533 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5534 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5541 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5542 * @page: The page within the block of interest
5543 * @start_bitidx: The first bit of interest
5544 * @end_bitidx: The last bit of interest
5545 * @flags: The flags to set
5547 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5548 int start_bitidx
, int end_bitidx
)
5551 unsigned long *bitmap
;
5552 unsigned long pfn
, bitidx
;
5553 unsigned long value
= 1;
5555 zone
= page_zone(page
);
5556 pfn
= page_to_pfn(page
);
5557 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5558 bitidx
= pfn_to_bitidx(zone
, pfn
);
5559 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5560 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5562 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5564 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5566 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5570 * This function checks whether pageblock includes unmovable pages or not.
5571 * If @count is not zero, it is okay to include less @count unmovable pages
5573 * PageLRU check wihtout isolation or lru_lock could race so that
5574 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5575 * expect this function should be exact.
5577 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
)
5579 unsigned long pfn
, iter
, found
;
5583 * For avoiding noise data, lru_add_drain_all() should be called
5584 * If ZONE_MOVABLE, the zone never contains unmovable pages
5586 if (zone_idx(zone
) == ZONE_MOVABLE
)
5588 mt
= get_pageblock_migratetype(page
);
5589 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
5592 pfn
= page_to_pfn(page
);
5593 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
5594 unsigned long check
= pfn
+ iter
;
5596 if (!pfn_valid_within(check
))
5599 page
= pfn_to_page(check
);
5601 * We can't use page_count without pin a page
5602 * because another CPU can free compound page.
5603 * This check already skips compound tails of THP
5604 * because their page->_count is zero at all time.
5606 if (!atomic_read(&page
->_count
)) {
5607 if (PageBuddy(page
))
5608 iter
+= (1 << page_order(page
)) - 1;
5615 * If there are RECLAIMABLE pages, we need to check it.
5616 * But now, memory offline itself doesn't call shrink_slab()
5617 * and it still to be fixed.
5620 * If the page is not RAM, page_count()should be 0.
5621 * we don't need more check. This is an _used_ not-movable page.
5623 * The problematic thing here is PG_reserved pages. PG_reserved
5624 * is set to both of a memory hole page and a _used_ kernel
5633 bool is_pageblock_removable_nolock(struct page
*page
)
5639 * We have to be careful here because we are iterating over memory
5640 * sections which are not zone aware so we might end up outside of
5641 * the zone but still within the section.
5642 * We have to take care about the node as well. If the node is offline
5643 * its NODE_DATA will be NULL - see page_zone.
5645 if (!node_online(page_to_nid(page
)))
5648 zone
= page_zone(page
);
5649 pfn
= page_to_pfn(page
);
5650 if (zone
->zone_start_pfn
> pfn
||
5651 zone
->zone_start_pfn
+ zone
->spanned_pages
<= pfn
)
5654 return !has_unmovable_pages(zone
, page
, 0);
5659 static unsigned long pfn_max_align_down(unsigned long pfn
)
5661 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5662 pageblock_nr_pages
) - 1);
5665 static unsigned long pfn_max_align_up(unsigned long pfn
)
5667 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5668 pageblock_nr_pages
));
5671 static struct page
*
5672 __alloc_contig_migrate_alloc(struct page
*page
, unsigned long private,
5675 gfp_t gfp_mask
= GFP_USER
| __GFP_MOVABLE
;
5677 if (PageHighMem(page
))
5678 gfp_mask
|= __GFP_HIGHMEM
;
5680 return alloc_page(gfp_mask
);
5683 /* [start, end) must belong to a single zone. */
5684 static int __alloc_contig_migrate_range(unsigned long start
, unsigned long end
)
5686 /* This function is based on compact_zone() from compaction.c. */
5688 unsigned long pfn
= start
;
5689 unsigned int tries
= 0;
5692 struct compact_control cc
= {
5693 .nr_migratepages
= 0,
5695 .zone
= page_zone(pfn_to_page(start
)),
5698 INIT_LIST_HEAD(&cc
.migratepages
);
5700 migrate_prep_local();
5702 while (pfn
< end
|| !list_empty(&cc
.migratepages
)) {
5703 if (fatal_signal_pending(current
)) {
5708 if (list_empty(&cc
.migratepages
)) {
5709 cc
.nr_migratepages
= 0;
5710 pfn
= isolate_migratepages_range(cc
.zone
, &cc
,
5717 } else if (++tries
== 5) {
5718 ret
= ret
< 0 ? ret
: -EBUSY
;
5722 reclaim_clean_pages_from_list(cc
.zone
, &cc
.migratepages
);
5724 ret
= migrate_pages(&cc
.migratepages
,
5725 __alloc_contig_migrate_alloc
,
5726 0, false, MIGRATE_SYNC
);
5729 putback_lru_pages(&cc
.migratepages
);
5730 return ret
> 0 ? 0 : ret
;
5734 * Update zone's cma pages counter used for watermark level calculation.
5736 static inline void __update_cma_watermarks(struct zone
*zone
, int count
)
5738 unsigned long flags
;
5739 spin_lock_irqsave(&zone
->lock
, flags
);
5740 zone
->min_cma_pages
+= count
;
5741 spin_unlock_irqrestore(&zone
->lock
, flags
);
5742 setup_per_zone_wmarks();
5746 * Trigger memory pressure bump to reclaim some pages in order to be able to
5747 * allocate 'count' pages in single page units. Does similar work as
5748 *__alloc_pages_slowpath() function.
5750 static int __reclaim_pages(struct zone
*zone
, gfp_t gfp_mask
, int count
)
5752 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
5753 struct zonelist
*zonelist
= node_zonelist(0, gfp_mask
);
5754 int did_some_progress
= 0;
5758 * Increase level of watermarks to force kswapd do his job
5759 * to stabilise at new watermark level.
5761 __update_cma_watermarks(zone
, count
);
5763 /* Obey watermarks as if the page was being allocated */
5764 while (!zone_watermark_ok(zone
, 0, low_wmark_pages(zone
), 0, 0)) {
5765 wake_all_kswapd(order
, zonelist
, high_zoneidx
, zone_idx(zone
));
5767 did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
5769 if (!did_some_progress
) {
5770 /* Exhausted what can be done so it's blamo time */
5771 out_of_memory(zonelist
, gfp_mask
, order
, NULL
, false);
5775 /* Restore original watermark levels. */
5776 __update_cma_watermarks(zone
, -count
);
5782 * alloc_contig_range() -- tries to allocate given range of pages
5783 * @start: start PFN to allocate
5784 * @end: one-past-the-last PFN to allocate
5785 * @migratetype: migratetype of the underlaying pageblocks (either
5786 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
5787 * in range must have the same migratetype and it must
5788 * be either of the two.
5790 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
5791 * aligned, however it's the caller's responsibility to guarantee that
5792 * we are the only thread that changes migrate type of pageblocks the
5795 * The PFN range must belong to a single zone.
5797 * Returns zero on success or negative error code. On success all
5798 * pages which PFN is in [start, end) are allocated for the caller and
5799 * need to be freed with free_contig_range().
5801 int alloc_contig_range(unsigned long start
, unsigned long end
,
5802 unsigned migratetype
)
5804 struct zone
*zone
= page_zone(pfn_to_page(start
));
5805 unsigned long outer_start
, outer_end
;
5809 * What we do here is we mark all pageblocks in range as
5810 * MIGRATE_ISOLATE. Because pageblock and max order pages may
5811 * have different sizes, and due to the way page allocator
5812 * work, we align the range to biggest of the two pages so
5813 * that page allocator won't try to merge buddies from
5814 * different pageblocks and change MIGRATE_ISOLATE to some
5815 * other migration type.
5817 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
5818 * migrate the pages from an unaligned range (ie. pages that
5819 * we are interested in). This will put all the pages in
5820 * range back to page allocator as MIGRATE_ISOLATE.
5822 * When this is done, we take the pages in range from page
5823 * allocator removing them from the buddy system. This way
5824 * page allocator will never consider using them.
5826 * This lets us mark the pageblocks back as
5827 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
5828 * aligned range but not in the unaligned, original range are
5829 * put back to page allocator so that buddy can use them.
5832 ret
= start_isolate_page_range(pfn_max_align_down(start
),
5833 pfn_max_align_up(end
), migratetype
);
5837 ret
= __alloc_contig_migrate_range(start
, end
);
5842 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
5843 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
5844 * more, all pages in [start, end) are free in page allocator.
5845 * What we are going to do is to allocate all pages from
5846 * [start, end) (that is remove them from page allocator).
5848 * The only problem is that pages at the beginning and at the
5849 * end of interesting range may be not aligned with pages that
5850 * page allocator holds, ie. they can be part of higher order
5851 * pages. Because of this, we reserve the bigger range and
5852 * once this is done free the pages we are not interested in.
5854 * We don't have to hold zone->lock here because the pages are
5855 * isolated thus they won't get removed from buddy.
5858 lru_add_drain_all();
5862 outer_start
= start
;
5863 while (!PageBuddy(pfn_to_page(outer_start
))) {
5864 if (++order
>= MAX_ORDER
) {
5868 outer_start
&= ~0UL << order
;
5871 /* Make sure the range is really isolated. */
5872 if (test_pages_isolated(outer_start
, end
)) {
5873 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
5880 * Reclaim enough pages to make sure that contiguous allocation
5881 * will not starve the system.
5883 __reclaim_pages(zone
, GFP_HIGHUSER_MOVABLE
, end
-start
);
5885 /* Grab isolated pages from freelists. */
5886 outer_end
= isolate_freepages_range(outer_start
, end
);
5892 /* Free head and tail (if any) */
5893 if (start
!= outer_start
)
5894 free_contig_range(outer_start
, start
- outer_start
);
5895 if (end
!= outer_end
)
5896 free_contig_range(end
, outer_end
- end
);
5899 undo_isolate_page_range(pfn_max_align_down(start
),
5900 pfn_max_align_up(end
), migratetype
);
5904 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
5906 for (; nr_pages
--; ++pfn
)
5907 __free_page(pfn_to_page(pfn
));
5911 #ifdef CONFIG_MEMORY_HOTPLUG
5912 static int __meminit
__zone_pcp_update(void *data
)
5914 struct zone
*zone
= data
;
5916 unsigned long batch
= zone_batchsize(zone
), flags
;
5918 for_each_possible_cpu(cpu
) {
5919 struct per_cpu_pageset
*pset
;
5920 struct per_cpu_pages
*pcp
;
5922 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
5925 local_irq_save(flags
);
5927 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
5928 setup_pageset(pset
, batch
);
5929 local_irq_restore(flags
);
5934 void __meminit
zone_pcp_update(struct zone
*zone
)
5936 stop_machine(__zone_pcp_update
, zone
, NULL
);
5940 #ifdef CONFIG_MEMORY_HOTREMOVE
5941 void zone_pcp_reset(struct zone
*zone
)
5943 unsigned long flags
;
5945 /* avoid races with drain_pages() */
5946 local_irq_save(flags
);
5947 if (zone
->pageset
!= &boot_pageset
) {
5948 free_percpu(zone
->pageset
);
5949 zone
->pageset
= &boot_pageset
;
5951 local_irq_restore(flags
);
5955 * All pages in the range must be isolated before calling this.
5958 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
5964 unsigned long flags
;
5965 /* find the first valid pfn */
5966 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
5971 zone
= page_zone(pfn_to_page(pfn
));
5972 spin_lock_irqsave(&zone
->lock
, flags
);
5974 while (pfn
< end_pfn
) {
5975 if (!pfn_valid(pfn
)) {
5979 page
= pfn_to_page(pfn
);
5980 BUG_ON(page_count(page
));
5981 BUG_ON(!PageBuddy(page
));
5982 order
= page_order(page
);
5983 #ifdef CONFIG_DEBUG_VM
5984 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
5985 pfn
, 1 << order
, end_pfn
);
5987 list_del(&page
->lru
);
5988 rmv_page_order(page
);
5989 zone
->free_area
[order
].nr_free
--;
5990 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
5992 for (i
= 0; i
< (1 << order
); i
++)
5993 SetPageReserved((page
+i
));
5994 pfn
+= (1 << order
);
5996 spin_unlock_irqrestore(&zone
->lock
, flags
);
6000 #ifdef CONFIG_MEMORY_FAILURE
6001 bool is_free_buddy_page(struct page
*page
)
6003 struct zone
*zone
= page_zone(page
);
6004 unsigned long pfn
= page_to_pfn(page
);
6005 unsigned long flags
;
6008 spin_lock_irqsave(&zone
->lock
, flags
);
6009 for (order
= 0; order
< MAX_ORDER
; order
++) {
6010 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6012 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6015 spin_unlock_irqrestore(&zone
->lock
, flags
);
6017 return order
< MAX_ORDER
;
6021 static const struct trace_print_flags pageflag_names
[] = {
6022 {1UL << PG_locked
, "locked" },
6023 {1UL << PG_error
, "error" },
6024 {1UL << PG_referenced
, "referenced" },
6025 {1UL << PG_uptodate
, "uptodate" },
6026 {1UL << PG_dirty
, "dirty" },
6027 {1UL << PG_lru
, "lru" },
6028 {1UL << PG_active
, "active" },
6029 {1UL << PG_slab
, "slab" },
6030 {1UL << PG_owner_priv_1
, "owner_priv_1" },
6031 {1UL << PG_arch_1
, "arch_1" },
6032 {1UL << PG_reserved
, "reserved" },
6033 {1UL << PG_private
, "private" },
6034 {1UL << PG_private_2
, "private_2" },
6035 {1UL << PG_writeback
, "writeback" },
6036 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6037 {1UL << PG_head
, "head" },
6038 {1UL << PG_tail
, "tail" },
6040 {1UL << PG_compound
, "compound" },
6042 {1UL << PG_swapcache
, "swapcache" },
6043 {1UL << PG_mappedtodisk
, "mappedtodisk" },
6044 {1UL << PG_reclaim
, "reclaim" },
6045 {1UL << PG_swapbacked
, "swapbacked" },
6046 {1UL << PG_unevictable
, "unevictable" },
6048 {1UL << PG_mlocked
, "mlocked" },
6050 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6051 {1UL << PG_uncached
, "uncached" },
6053 #ifdef CONFIG_MEMORY_FAILURE
6054 {1UL << PG_hwpoison
, "hwpoison" },
6056 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6057 {1UL << PG_compound_lock
, "compound_lock" },
6061 static void dump_page_flags(unsigned long flags
)
6063 const char *delim
= "";
6067 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names
) != __NR_PAGEFLAGS
);
6069 printk(KERN_ALERT
"page flags: %#lx(", flags
);
6071 /* remove zone id */
6072 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
6074 for (i
= 0; i
< ARRAY_SIZE(pageflag_names
) && flags
; i
++) {
6076 mask
= pageflag_names
[i
].mask
;
6077 if ((flags
& mask
) != mask
)
6081 printk("%s%s", delim
, pageflag_names
[i
].name
);
6085 /* check for left over flags */
6087 printk("%s%#lx", delim
, flags
);
6092 void dump_page(struct page
*page
)
6095 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6096 page
, atomic_read(&page
->_count
), page_mapcount(page
),
6097 page
->mapping
, page
->index
);
6098 dump_page_flags(page
->flags
);
6099 mem_cgroup_print_bad_page(page
);