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>
61 #include <linux/sched/rt.h>
63 #include <asm/tlbflush.h>
64 #include <asm/div64.h>
67 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
68 DEFINE_PER_CPU(int, numa_node
);
69 EXPORT_PER_CPU_SYMBOL(numa_node
);
72 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
74 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
75 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
76 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
77 * defined in <linux/topology.h>.
79 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
80 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
84 * Array of node states.
86 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
87 [N_POSSIBLE
] = NODE_MASK_ALL
,
88 [N_ONLINE
] = { { [0] = 1UL } },
90 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
92 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
94 #ifdef CONFIG_MOVABLE_NODE
95 [N_MEMORY
] = { { [0] = 1UL } },
97 [N_CPU
] = { { [0] = 1UL } },
100 EXPORT_SYMBOL(node_states
);
102 unsigned long totalram_pages __read_mostly
;
103 unsigned long totalreserve_pages __read_mostly
;
105 * When calculating the number of globally allowed dirty pages, there
106 * is a certain number of per-zone reserves that should not be
107 * considered dirtyable memory. This is the sum of those reserves
108 * over all existing zones that contribute dirtyable memory.
110 unsigned long dirty_balance_reserve __read_mostly
;
112 int percpu_pagelist_fraction
;
113 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
115 #ifdef CONFIG_PM_SLEEP
117 * The following functions are used by the suspend/hibernate code to temporarily
118 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
119 * while devices are suspended. To avoid races with the suspend/hibernate code,
120 * they should always be called with pm_mutex held (gfp_allowed_mask also should
121 * only be modified with pm_mutex held, unless the suspend/hibernate code is
122 * guaranteed not to run in parallel with that modification).
125 static gfp_t saved_gfp_mask
;
127 void pm_restore_gfp_mask(void)
129 WARN_ON(!mutex_is_locked(&pm_mutex
));
130 if (saved_gfp_mask
) {
131 gfp_allowed_mask
= saved_gfp_mask
;
136 void pm_restrict_gfp_mask(void)
138 WARN_ON(!mutex_is_locked(&pm_mutex
));
139 WARN_ON(saved_gfp_mask
);
140 saved_gfp_mask
= gfp_allowed_mask
;
141 gfp_allowed_mask
&= ~GFP_IOFS
;
144 bool pm_suspended_storage(void)
146 if ((gfp_allowed_mask
& GFP_IOFS
) == GFP_IOFS
)
150 #endif /* CONFIG_PM_SLEEP */
152 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
153 int pageblock_order __read_mostly
;
156 static void __free_pages_ok(struct page
*page
, unsigned int order
);
159 * results with 256, 32 in the lowmem_reserve sysctl:
160 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
161 * 1G machine -> (16M dma, 784M normal, 224M high)
162 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
163 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
164 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
166 * TBD: should special case ZONE_DMA32 machines here - in those we normally
167 * don't need any ZONE_NORMAL reservation
169 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
170 #ifdef CONFIG_ZONE_DMA
173 #ifdef CONFIG_ZONE_DMA32
176 #ifdef CONFIG_HIGHMEM
182 EXPORT_SYMBOL(totalram_pages
);
184 static char * const zone_names
[MAX_NR_ZONES
] = {
185 #ifdef CONFIG_ZONE_DMA
188 #ifdef CONFIG_ZONE_DMA32
192 #ifdef CONFIG_HIGHMEM
198 int min_free_kbytes
= 1024;
200 static unsigned long __meminitdata nr_kernel_pages
;
201 static unsigned long __meminitdata nr_all_pages
;
202 static unsigned long __meminitdata dma_reserve
;
204 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
205 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
206 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
207 static unsigned long __initdata required_kernelcore
;
208 static unsigned long __initdata required_movablecore
;
209 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
211 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
213 EXPORT_SYMBOL(movable_zone
);
214 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
217 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
218 int nr_online_nodes __read_mostly
= 1;
219 EXPORT_SYMBOL(nr_node_ids
);
220 EXPORT_SYMBOL(nr_online_nodes
);
223 int page_group_by_mobility_disabled __read_mostly
;
225 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
228 if (unlikely(page_group_by_mobility_disabled
))
229 migratetype
= MIGRATE_UNMOVABLE
;
231 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
232 PB_migrate
, PB_migrate_end
);
235 bool oom_killer_disabled __read_mostly
;
237 #ifdef CONFIG_DEBUG_VM
238 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
242 unsigned long pfn
= page_to_pfn(page
);
245 seq
= zone_span_seqbegin(zone
);
246 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
248 else if (pfn
< zone
->zone_start_pfn
)
250 } while (zone_span_seqretry(zone
, seq
));
255 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
257 if (!pfn_valid_within(page_to_pfn(page
)))
259 if (zone
!= page_zone(page
))
265 * Temporary debugging check for pages not lying within a given zone.
267 static int bad_range(struct zone
*zone
, struct page
*page
)
269 if (page_outside_zone_boundaries(zone
, page
))
271 if (!page_is_consistent(zone
, page
))
277 static inline int bad_range(struct zone
*zone
, struct page
*page
)
283 static void bad_page(struct page
*page
)
285 static unsigned long resume
;
286 static unsigned long nr_shown
;
287 static unsigned long nr_unshown
;
289 /* Don't complain about poisoned pages */
290 if (PageHWPoison(page
)) {
291 reset_page_mapcount(page
); /* remove PageBuddy */
296 * Allow a burst of 60 reports, then keep quiet for that minute;
297 * or allow a steady drip of one report per second.
299 if (nr_shown
== 60) {
300 if (time_before(jiffies
, resume
)) {
306 "BUG: Bad page state: %lu messages suppressed\n",
313 resume
= jiffies
+ 60 * HZ
;
315 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
316 current
->comm
, page_to_pfn(page
));
322 /* Leave bad fields for debug, except PageBuddy could make trouble */
323 reset_page_mapcount(page
); /* remove PageBuddy */
324 add_taint(TAINT_BAD_PAGE
);
328 * Higher-order pages are called "compound pages". They are structured thusly:
330 * The first PAGE_SIZE page is called the "head page".
332 * The remaining PAGE_SIZE pages are called "tail pages".
334 * All pages have PG_compound set. All tail pages have their ->first_page
335 * pointing at the head page.
337 * The first tail page's ->lru.next holds the address of the compound page's
338 * put_page() function. Its ->lru.prev holds the order of allocation.
339 * This usage means that zero-order pages may not be compound.
342 static void free_compound_page(struct page
*page
)
344 __free_pages_ok(page
, compound_order(page
));
347 void prep_compound_page(struct page
*page
, unsigned long order
)
350 int nr_pages
= 1 << order
;
352 set_compound_page_dtor(page
, free_compound_page
);
353 set_compound_order(page
, order
);
355 for (i
= 1; i
< nr_pages
; i
++) {
356 struct page
*p
= page
+ i
;
358 set_page_count(p
, 0);
359 p
->first_page
= page
;
363 /* update __split_huge_page_refcount if you change this function */
364 static int destroy_compound_page(struct page
*page
, unsigned long order
)
367 int nr_pages
= 1 << order
;
370 if (unlikely(compound_order(page
) != order
)) {
375 __ClearPageHead(page
);
377 for (i
= 1; i
< nr_pages
; i
++) {
378 struct page
*p
= page
+ i
;
380 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
390 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
395 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
396 * and __GFP_HIGHMEM from hard or soft interrupt context.
398 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
399 for (i
= 0; i
< (1 << order
); i
++)
400 clear_highpage(page
+ i
);
403 #ifdef CONFIG_DEBUG_PAGEALLOC
404 unsigned int _debug_guardpage_minorder
;
406 static int __init
debug_guardpage_minorder_setup(char *buf
)
410 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
411 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
414 _debug_guardpage_minorder
= res
;
415 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
418 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
420 static inline void set_page_guard_flag(struct page
*page
)
422 __set_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
425 static inline void clear_page_guard_flag(struct page
*page
)
427 __clear_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
430 static inline void set_page_guard_flag(struct page
*page
) { }
431 static inline void clear_page_guard_flag(struct page
*page
) { }
434 static inline void set_page_order(struct page
*page
, int order
)
436 set_page_private(page
, order
);
437 __SetPageBuddy(page
);
440 static inline void rmv_page_order(struct page
*page
)
442 __ClearPageBuddy(page
);
443 set_page_private(page
, 0);
447 * Locate the struct page for both the matching buddy in our
448 * pair (buddy1) and the combined O(n+1) page they form (page).
450 * 1) Any buddy B1 will have an order O twin B2 which satisfies
451 * the following equation:
453 * For example, if the starting buddy (buddy2) is #8 its order
455 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
457 * 2) Any buddy B will have an order O+1 parent P which
458 * satisfies the following equation:
461 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
463 static inline unsigned long
464 __find_buddy_index(unsigned long page_idx
, unsigned int order
)
466 return page_idx
^ (1 << order
);
470 * This function checks whether a page is free && is the buddy
471 * we can do coalesce a page and its buddy if
472 * (a) the buddy is not in a hole &&
473 * (b) the buddy is in the buddy system &&
474 * (c) a page and its buddy have the same order &&
475 * (d) a page and its buddy are in the same zone.
477 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
478 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
480 * For recording page's order, we use page_private(page).
482 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
485 if (!pfn_valid_within(page_to_pfn(buddy
)))
488 if (page_zone_id(page
) != page_zone_id(buddy
))
491 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
492 VM_BUG_ON(page_count(buddy
) != 0);
496 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
497 VM_BUG_ON(page_count(buddy
) != 0);
504 * Freeing function for a buddy system allocator.
506 * The concept of a buddy system is to maintain direct-mapped table
507 * (containing bit values) for memory blocks of various "orders".
508 * The bottom level table contains the map for the smallest allocatable
509 * units of memory (here, pages), and each level above it describes
510 * pairs of units from the levels below, hence, "buddies".
511 * At a high level, all that happens here is marking the table entry
512 * at the bottom level available, and propagating the changes upward
513 * as necessary, plus some accounting needed to play nicely with other
514 * parts of the VM system.
515 * At each level, we keep a list of pages, which are heads of continuous
516 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
517 * order is recorded in page_private(page) field.
518 * So when we are allocating or freeing one, we can derive the state of the
519 * other. That is, if we allocate a small block, and both were
520 * free, the remainder of the region must be split into blocks.
521 * If a block is freed, and its buddy is also free, then this
522 * triggers coalescing into a block of larger size.
527 static inline void __free_one_page(struct page
*page
,
528 struct zone
*zone
, unsigned int order
,
531 unsigned long page_idx
;
532 unsigned long combined_idx
;
533 unsigned long uninitialized_var(buddy_idx
);
536 if (unlikely(PageCompound(page
)))
537 if (unlikely(destroy_compound_page(page
, order
)))
540 VM_BUG_ON(migratetype
== -1);
542 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
544 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
545 VM_BUG_ON(bad_range(zone
, page
));
547 while (order
< MAX_ORDER
-1) {
548 buddy_idx
= __find_buddy_index(page_idx
, order
);
549 buddy
= page
+ (buddy_idx
- page_idx
);
550 if (!page_is_buddy(page
, buddy
, order
))
553 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
554 * merge with it and move up one order.
556 if (page_is_guard(buddy
)) {
557 clear_page_guard_flag(buddy
);
558 set_page_private(page
, 0);
559 __mod_zone_freepage_state(zone
, 1 << order
,
562 list_del(&buddy
->lru
);
563 zone
->free_area
[order
].nr_free
--;
564 rmv_page_order(buddy
);
566 combined_idx
= buddy_idx
& page_idx
;
567 page
= page
+ (combined_idx
- page_idx
);
568 page_idx
= combined_idx
;
571 set_page_order(page
, order
);
574 * If this is not the largest possible page, check if the buddy
575 * of the next-highest order is free. If it is, it's possible
576 * that pages are being freed that will coalesce soon. In case,
577 * that is happening, add the free page to the tail of the list
578 * so it's less likely to be used soon and more likely to be merged
579 * as a higher order page
581 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
582 struct page
*higher_page
, *higher_buddy
;
583 combined_idx
= buddy_idx
& page_idx
;
584 higher_page
= page
+ (combined_idx
- page_idx
);
585 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
586 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
587 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
588 list_add_tail(&page
->lru
,
589 &zone
->free_area
[order
].free_list
[migratetype
]);
594 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
596 zone
->free_area
[order
].nr_free
++;
599 static inline int free_pages_check(struct page
*page
)
601 if (unlikely(page_mapcount(page
) |
602 (page
->mapping
!= NULL
) |
603 (atomic_read(&page
->_count
) != 0) |
604 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
) |
605 (mem_cgroup_bad_page_check(page
)))) {
609 reset_page_last_nid(page
);
610 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
611 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
616 * Frees a number of pages from the PCP lists
617 * Assumes all pages on list are in same zone, and of same order.
618 * count is the number of pages to free.
620 * If the zone was previously in an "all pages pinned" state then look to
621 * see if this freeing clears that state.
623 * And clear the zone's pages_scanned counter, to hold off the "all pages are
624 * pinned" detection logic.
626 static void free_pcppages_bulk(struct zone
*zone
, int count
,
627 struct per_cpu_pages
*pcp
)
633 spin_lock(&zone
->lock
);
634 zone
->all_unreclaimable
= 0;
635 zone
->pages_scanned
= 0;
639 struct list_head
*list
;
642 * Remove pages from lists in a round-robin fashion. A
643 * batch_free count is maintained that is incremented when an
644 * empty list is encountered. This is so more pages are freed
645 * off fuller lists instead of spinning excessively around empty
650 if (++migratetype
== MIGRATE_PCPTYPES
)
652 list
= &pcp
->lists
[migratetype
];
653 } while (list_empty(list
));
655 /* This is the only non-empty list. Free them all. */
656 if (batch_free
== MIGRATE_PCPTYPES
)
657 batch_free
= to_free
;
660 int mt
; /* migratetype of the to-be-freed page */
662 page
= list_entry(list
->prev
, struct page
, lru
);
663 /* must delete as __free_one_page list manipulates */
664 list_del(&page
->lru
);
665 mt
= get_freepage_migratetype(page
);
666 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
667 __free_one_page(page
, zone
, 0, mt
);
668 trace_mm_page_pcpu_drain(page
, 0, mt
);
669 if (likely(get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)) {
670 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1);
671 if (is_migrate_cma(mt
))
672 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
, 1);
674 } while (--to_free
&& --batch_free
&& !list_empty(list
));
676 spin_unlock(&zone
->lock
);
679 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
682 spin_lock(&zone
->lock
);
683 zone
->all_unreclaimable
= 0;
684 zone
->pages_scanned
= 0;
686 __free_one_page(page
, zone
, order
, migratetype
);
687 if (unlikely(migratetype
!= MIGRATE_ISOLATE
))
688 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
689 spin_unlock(&zone
->lock
);
692 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
697 trace_mm_page_free(page
, order
);
698 kmemcheck_free_shadow(page
, order
);
701 page
->mapping
= NULL
;
702 for (i
= 0; i
< (1 << order
); i
++)
703 bad
+= free_pages_check(page
+ i
);
707 if (!PageHighMem(page
)) {
708 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
709 debug_check_no_obj_freed(page_address(page
),
712 arch_free_page(page
, order
);
713 kernel_map_pages(page
, 1 << order
, 0);
718 static void __free_pages_ok(struct page
*page
, unsigned int order
)
723 if (!free_pages_prepare(page
, order
))
726 local_irq_save(flags
);
727 __count_vm_events(PGFREE
, 1 << order
);
728 migratetype
= get_pageblock_migratetype(page
);
729 set_freepage_migratetype(page
, migratetype
);
730 free_one_page(page_zone(page
), page
, order
, migratetype
);
731 local_irq_restore(flags
);
735 * Read access to zone->managed_pages is safe because it's unsigned long,
736 * but we still need to serialize writers. Currently all callers of
737 * __free_pages_bootmem() except put_page_bootmem() should only be used
738 * at boot time. So for shorter boot time, we shift the burden to
739 * put_page_bootmem() to serialize writers.
741 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
743 unsigned int nr_pages
= 1 << order
;
747 for (loop
= 0; loop
< nr_pages
; loop
++) {
748 struct page
*p
= &page
[loop
];
750 if (loop
+ 1 < nr_pages
)
752 __ClearPageReserved(p
);
753 set_page_count(p
, 0);
756 page_zone(page
)->managed_pages
+= 1 << order
;
757 set_page_refcounted(page
);
758 __free_pages(page
, order
);
762 /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
763 void __init
init_cma_reserved_pageblock(struct page
*page
)
765 unsigned i
= pageblock_nr_pages
;
766 struct page
*p
= page
;
769 __ClearPageReserved(p
);
770 set_page_count(p
, 0);
773 set_page_refcounted(page
);
774 set_pageblock_migratetype(page
, MIGRATE_CMA
);
775 __free_pages(page
, pageblock_order
);
776 totalram_pages
+= pageblock_nr_pages
;
781 * The order of subdivision here is critical for the IO subsystem.
782 * Please do not alter this order without good reasons and regression
783 * testing. Specifically, as large blocks of memory are subdivided,
784 * the order in which smaller blocks are delivered depends on the order
785 * they're subdivided in this function. This is the primary factor
786 * influencing the order in which pages are delivered to the IO
787 * subsystem according to empirical testing, and this is also justified
788 * by considering the behavior of a buddy system containing a single
789 * large block of memory acted on by a series of small allocations.
790 * This behavior is a critical factor in sglist merging's success.
794 static inline void expand(struct zone
*zone
, struct page
*page
,
795 int low
, int high
, struct free_area
*area
,
798 unsigned long size
= 1 << high
;
804 VM_BUG_ON(bad_range(zone
, &page
[size
]));
806 #ifdef CONFIG_DEBUG_PAGEALLOC
807 if (high
< debug_guardpage_minorder()) {
809 * Mark as guard pages (or page), that will allow to
810 * merge back to allocator when buddy will be freed.
811 * Corresponding page table entries will not be touched,
812 * pages will stay not present in virtual address space
814 INIT_LIST_HEAD(&page
[size
].lru
);
815 set_page_guard_flag(&page
[size
]);
816 set_page_private(&page
[size
], high
);
817 /* Guard pages are not available for any usage */
818 __mod_zone_freepage_state(zone
, -(1 << high
),
823 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
825 set_page_order(&page
[size
], high
);
830 * This page is about to be returned from the page allocator
832 static inline int check_new_page(struct page
*page
)
834 if (unlikely(page_mapcount(page
) |
835 (page
->mapping
!= NULL
) |
836 (atomic_read(&page
->_count
) != 0) |
837 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
) |
838 (mem_cgroup_bad_page_check(page
)))) {
845 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
849 for (i
= 0; i
< (1 << order
); i
++) {
850 struct page
*p
= page
+ i
;
851 if (unlikely(check_new_page(p
)))
855 set_page_private(page
, 0);
856 set_page_refcounted(page
);
858 arch_alloc_page(page
, order
);
859 kernel_map_pages(page
, 1 << order
, 1);
861 if (gfp_flags
& __GFP_ZERO
)
862 prep_zero_page(page
, order
, gfp_flags
);
864 if (order
&& (gfp_flags
& __GFP_COMP
))
865 prep_compound_page(page
, order
);
871 * Go through the free lists for the given migratetype and remove
872 * the smallest available page from the freelists
875 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
878 unsigned int current_order
;
879 struct free_area
* area
;
882 /* Find a page of the appropriate size in the preferred list */
883 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
884 area
= &(zone
->free_area
[current_order
]);
885 if (list_empty(&area
->free_list
[migratetype
]))
888 page
= list_entry(area
->free_list
[migratetype
].next
,
890 list_del(&page
->lru
);
891 rmv_page_order(page
);
893 expand(zone
, page
, order
, current_order
, area
, migratetype
);
902 * This array describes the order lists are fallen back to when
903 * the free lists for the desirable migrate type are depleted
905 static int fallbacks
[MIGRATE_TYPES
][4] = {
906 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
907 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
909 [MIGRATE_MOVABLE
] = { MIGRATE_CMA
, MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
910 [MIGRATE_CMA
] = { MIGRATE_RESERVE
}, /* Never used */
912 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
914 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
}, /* Never used */
915 [MIGRATE_ISOLATE
] = { MIGRATE_RESERVE
}, /* Never used */
919 * Move the free pages in a range to the free lists of the requested type.
920 * Note that start_page and end_pages are not aligned on a pageblock
921 * boundary. If alignment is required, use move_freepages_block()
923 int move_freepages(struct zone
*zone
,
924 struct page
*start_page
, struct page
*end_page
,
931 #ifndef CONFIG_HOLES_IN_ZONE
933 * page_zone is not safe to call in this context when
934 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
935 * anyway as we check zone boundaries in move_freepages_block().
936 * Remove at a later date when no bug reports exist related to
937 * grouping pages by mobility
939 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
942 for (page
= start_page
; page
<= end_page
;) {
943 /* Make sure we are not inadvertently changing nodes */
944 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
946 if (!pfn_valid_within(page_to_pfn(page
))) {
951 if (!PageBuddy(page
)) {
956 order
= page_order(page
);
957 list_move(&page
->lru
,
958 &zone
->free_area
[order
].free_list
[migratetype
]);
959 set_freepage_migratetype(page
, migratetype
);
961 pages_moved
+= 1 << order
;
967 int move_freepages_block(struct zone
*zone
, struct page
*page
,
970 unsigned long start_pfn
, end_pfn
;
971 struct page
*start_page
, *end_page
;
973 start_pfn
= page_to_pfn(page
);
974 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
975 start_page
= pfn_to_page(start_pfn
);
976 end_page
= start_page
+ pageblock_nr_pages
- 1;
977 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
979 /* Do not cross zone boundaries */
980 if (start_pfn
< zone
->zone_start_pfn
)
982 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
985 return move_freepages(zone
, start_page
, end_page
, migratetype
);
988 static void change_pageblock_range(struct page
*pageblock_page
,
989 int start_order
, int migratetype
)
991 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
993 while (nr_pageblocks
--) {
994 set_pageblock_migratetype(pageblock_page
, migratetype
);
995 pageblock_page
+= pageblock_nr_pages
;
999 /* Remove an element from the buddy allocator from the fallback list */
1000 static inline struct page
*
1001 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
1003 struct free_area
* area
;
1008 /* Find the largest possible block of pages in the other list */
1009 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
1012 migratetype
= fallbacks
[start_migratetype
][i
];
1014 /* MIGRATE_RESERVE handled later if necessary */
1015 if (migratetype
== MIGRATE_RESERVE
)
1018 area
= &(zone
->free_area
[current_order
]);
1019 if (list_empty(&area
->free_list
[migratetype
]))
1022 page
= list_entry(area
->free_list
[migratetype
].next
,
1027 * If breaking a large block of pages, move all free
1028 * pages to the preferred allocation list. If falling
1029 * back for a reclaimable kernel allocation, be more
1030 * aggressive about taking ownership of free pages
1032 * On the other hand, never change migration
1033 * type of MIGRATE_CMA pageblocks nor move CMA
1034 * pages on different free lists. We don't
1035 * want unmovable pages to be allocated from
1036 * MIGRATE_CMA areas.
1038 if (!is_migrate_cma(migratetype
) &&
1039 (unlikely(current_order
>= pageblock_order
/ 2) ||
1040 start_migratetype
== MIGRATE_RECLAIMABLE
||
1041 page_group_by_mobility_disabled
)) {
1043 pages
= move_freepages_block(zone
, page
,
1046 /* Claim the whole block if over half of it is free */
1047 if (pages
>= (1 << (pageblock_order
-1)) ||
1048 page_group_by_mobility_disabled
)
1049 set_pageblock_migratetype(page
,
1052 migratetype
= start_migratetype
;
1055 /* Remove the page from the freelists */
1056 list_del(&page
->lru
);
1057 rmv_page_order(page
);
1059 /* Take ownership for orders >= pageblock_order */
1060 if (current_order
>= pageblock_order
&&
1061 !is_migrate_cma(migratetype
))
1062 change_pageblock_range(page
, current_order
,
1065 expand(zone
, page
, order
, current_order
, area
,
1066 is_migrate_cma(migratetype
)
1067 ? migratetype
: start_migratetype
);
1069 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1070 start_migratetype
, migratetype
);
1080 * Do the hard work of removing an element from the buddy allocator.
1081 * Call me with the zone->lock already held.
1083 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1089 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1091 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1092 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1095 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1096 * is used because __rmqueue_smallest is an inline function
1097 * and we want just one call site
1100 migratetype
= MIGRATE_RESERVE
;
1105 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1110 * Obtain a specified number of elements from the buddy allocator, all under
1111 * a single hold of the lock, for efficiency. Add them to the supplied list.
1112 * Returns the number of new pages which were placed at *list.
1114 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1115 unsigned long count
, struct list_head
*list
,
1116 int migratetype
, int cold
)
1118 int mt
= migratetype
, i
;
1120 spin_lock(&zone
->lock
);
1121 for (i
= 0; i
< count
; ++i
) {
1122 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1123 if (unlikely(page
== NULL
))
1127 * Split buddy pages returned by expand() are received here
1128 * in physical page order. The page is added to the callers and
1129 * list and the list head then moves forward. From the callers
1130 * perspective, the linked list is ordered by page number in
1131 * some conditions. This is useful for IO devices that can
1132 * merge IO requests if the physical pages are ordered
1135 if (likely(cold
== 0))
1136 list_add(&page
->lru
, list
);
1138 list_add_tail(&page
->lru
, list
);
1139 if (IS_ENABLED(CONFIG_CMA
)) {
1140 mt
= get_pageblock_migratetype(page
);
1141 if (!is_migrate_cma(mt
) && mt
!= MIGRATE_ISOLATE
)
1144 set_freepage_migratetype(page
, mt
);
1146 if (is_migrate_cma(mt
))
1147 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1150 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1151 spin_unlock(&zone
->lock
);
1157 * Called from the vmstat counter updater to drain pagesets of this
1158 * currently executing processor on remote nodes after they have
1161 * Note that this function must be called with the thread pinned to
1162 * a single processor.
1164 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1166 unsigned long flags
;
1169 local_irq_save(flags
);
1170 if (pcp
->count
>= pcp
->batch
)
1171 to_drain
= pcp
->batch
;
1173 to_drain
= pcp
->count
;
1175 free_pcppages_bulk(zone
, to_drain
, pcp
);
1176 pcp
->count
-= to_drain
;
1178 local_irq_restore(flags
);
1183 * Drain pages of the indicated processor.
1185 * The processor must either be the current processor and the
1186 * thread pinned to the current processor or a processor that
1189 static void drain_pages(unsigned int cpu
)
1191 unsigned long flags
;
1194 for_each_populated_zone(zone
) {
1195 struct per_cpu_pageset
*pset
;
1196 struct per_cpu_pages
*pcp
;
1198 local_irq_save(flags
);
1199 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1203 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1206 local_irq_restore(flags
);
1211 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1213 void drain_local_pages(void *arg
)
1215 drain_pages(smp_processor_id());
1219 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1221 * Note that this code is protected against sending an IPI to an offline
1222 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1223 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1224 * nothing keeps CPUs from showing up after we populated the cpumask and
1225 * before the call to on_each_cpu_mask().
1227 void drain_all_pages(void)
1230 struct per_cpu_pageset
*pcp
;
1234 * Allocate in the BSS so we wont require allocation in
1235 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1237 static cpumask_t cpus_with_pcps
;
1240 * We don't care about racing with CPU hotplug event
1241 * as offline notification will cause the notified
1242 * cpu to drain that CPU pcps and on_each_cpu_mask
1243 * disables preemption as part of its processing
1245 for_each_online_cpu(cpu
) {
1246 bool has_pcps
= false;
1247 for_each_populated_zone(zone
) {
1248 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1249 if (pcp
->pcp
.count
) {
1255 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1257 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1259 on_each_cpu_mask(&cpus_with_pcps
, drain_local_pages
, NULL
, 1);
1262 #ifdef CONFIG_HIBERNATION
1264 void mark_free_pages(struct zone
*zone
)
1266 unsigned long pfn
, max_zone_pfn
;
1267 unsigned long flags
;
1269 struct list_head
*curr
;
1271 if (!zone
->spanned_pages
)
1274 spin_lock_irqsave(&zone
->lock
, flags
);
1276 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1277 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1278 if (pfn_valid(pfn
)) {
1279 struct page
*page
= pfn_to_page(pfn
);
1281 if (!swsusp_page_is_forbidden(page
))
1282 swsusp_unset_page_free(page
);
1285 for_each_migratetype_order(order
, t
) {
1286 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1289 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1290 for (i
= 0; i
< (1UL << order
); i
++)
1291 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1294 spin_unlock_irqrestore(&zone
->lock
, flags
);
1296 #endif /* CONFIG_PM */
1299 * Free a 0-order page
1300 * cold == 1 ? free a cold page : free a hot page
1302 void free_hot_cold_page(struct page
*page
, int cold
)
1304 struct zone
*zone
= page_zone(page
);
1305 struct per_cpu_pages
*pcp
;
1306 unsigned long flags
;
1309 if (!free_pages_prepare(page
, 0))
1312 migratetype
= get_pageblock_migratetype(page
);
1313 set_freepage_migratetype(page
, migratetype
);
1314 local_irq_save(flags
);
1315 __count_vm_event(PGFREE
);
1318 * We only track unmovable, reclaimable and movable on pcp lists.
1319 * Free ISOLATE pages back to the allocator because they are being
1320 * offlined but treat RESERVE as movable pages so we can get those
1321 * areas back if necessary. Otherwise, we may have to free
1322 * excessively into the page allocator
1324 if (migratetype
>= MIGRATE_PCPTYPES
) {
1325 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1326 free_one_page(zone
, page
, 0, migratetype
);
1329 migratetype
= MIGRATE_MOVABLE
;
1332 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1334 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1336 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1338 if (pcp
->count
>= pcp
->high
) {
1339 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1340 pcp
->count
-= pcp
->batch
;
1344 local_irq_restore(flags
);
1348 * Free a list of 0-order pages
1350 void free_hot_cold_page_list(struct list_head
*list
, int cold
)
1352 struct page
*page
, *next
;
1354 list_for_each_entry_safe(page
, next
, list
, lru
) {
1355 trace_mm_page_free_batched(page
, cold
);
1356 free_hot_cold_page(page
, cold
);
1361 * split_page takes a non-compound higher-order page, and splits it into
1362 * n (1<<order) sub-pages: page[0..n]
1363 * Each sub-page must be freed individually.
1365 * Note: this is probably too low level an operation for use in drivers.
1366 * Please consult with lkml before using this in your driver.
1368 void split_page(struct page
*page
, unsigned int order
)
1372 VM_BUG_ON(PageCompound(page
));
1373 VM_BUG_ON(!page_count(page
));
1375 #ifdef CONFIG_KMEMCHECK
1377 * Split shadow pages too, because free(page[0]) would
1378 * otherwise free the whole shadow.
1380 if (kmemcheck_page_is_tracked(page
))
1381 split_page(virt_to_page(page
[0].shadow
), order
);
1384 for (i
= 1; i
< (1 << order
); i
++)
1385 set_page_refcounted(page
+ i
);
1388 static int __isolate_free_page(struct page
*page
, unsigned int order
)
1390 unsigned long watermark
;
1394 BUG_ON(!PageBuddy(page
));
1396 zone
= page_zone(page
);
1397 mt
= get_pageblock_migratetype(page
);
1399 if (mt
!= MIGRATE_ISOLATE
) {
1400 /* Obey watermarks as if the page was being allocated */
1401 watermark
= low_wmark_pages(zone
) + (1 << order
);
1402 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1405 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
1408 /* Remove page from free list */
1409 list_del(&page
->lru
);
1410 zone
->free_area
[order
].nr_free
--;
1411 rmv_page_order(page
);
1413 /* Set the pageblock if the isolated page is at least a pageblock */
1414 if (order
>= pageblock_order
- 1) {
1415 struct page
*endpage
= page
+ (1 << order
) - 1;
1416 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
1417 int mt
= get_pageblock_migratetype(page
);
1418 if (mt
!= MIGRATE_ISOLATE
&& !is_migrate_cma(mt
))
1419 set_pageblock_migratetype(page
,
1424 return 1UL << order
;
1428 * Similar to split_page except the page is already free. As this is only
1429 * being used for migration, the migratetype of the block also changes.
1430 * As this is called with interrupts disabled, the caller is responsible
1431 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1434 * Note: this is probably too low level an operation for use in drivers.
1435 * Please consult with lkml before using this in your driver.
1437 int split_free_page(struct page
*page
)
1442 order
= page_order(page
);
1444 nr_pages
= __isolate_free_page(page
, order
);
1448 /* Split into individual pages */
1449 set_page_refcounted(page
);
1450 split_page(page
, order
);
1455 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1456 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1460 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1461 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1464 unsigned long flags
;
1466 int cold
= !!(gfp_flags
& __GFP_COLD
);
1469 if (likely(order
== 0)) {
1470 struct per_cpu_pages
*pcp
;
1471 struct list_head
*list
;
1473 local_irq_save(flags
);
1474 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1475 list
= &pcp
->lists
[migratetype
];
1476 if (list_empty(list
)) {
1477 pcp
->count
+= rmqueue_bulk(zone
, 0,
1480 if (unlikely(list_empty(list
)))
1485 page
= list_entry(list
->prev
, struct page
, lru
);
1487 page
= list_entry(list
->next
, struct page
, lru
);
1489 list_del(&page
->lru
);
1492 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1494 * __GFP_NOFAIL is not to be used in new code.
1496 * All __GFP_NOFAIL callers should be fixed so that they
1497 * properly detect and handle allocation failures.
1499 * We most definitely don't want callers attempting to
1500 * allocate greater than order-1 page units with
1503 WARN_ON_ONCE(order
> 1);
1505 spin_lock_irqsave(&zone
->lock
, flags
);
1506 page
= __rmqueue(zone
, order
, migratetype
);
1507 spin_unlock(&zone
->lock
);
1510 __mod_zone_freepage_state(zone
, -(1 << order
),
1511 get_pageblock_migratetype(page
));
1514 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1515 zone_statistics(preferred_zone
, zone
, gfp_flags
);
1516 local_irq_restore(flags
);
1518 VM_BUG_ON(bad_range(zone
, page
));
1519 if (prep_new_page(page
, order
, gfp_flags
))
1524 local_irq_restore(flags
);
1528 #ifdef CONFIG_FAIL_PAGE_ALLOC
1531 struct fault_attr attr
;
1533 u32 ignore_gfp_highmem
;
1534 u32 ignore_gfp_wait
;
1536 } fail_page_alloc
= {
1537 .attr
= FAULT_ATTR_INITIALIZER
,
1538 .ignore_gfp_wait
= 1,
1539 .ignore_gfp_highmem
= 1,
1543 static int __init
setup_fail_page_alloc(char *str
)
1545 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1547 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1549 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1551 if (order
< fail_page_alloc
.min_order
)
1553 if (gfp_mask
& __GFP_NOFAIL
)
1555 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1557 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1560 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1563 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1565 static int __init
fail_page_alloc_debugfs(void)
1567 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1570 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
1571 &fail_page_alloc
.attr
);
1573 return PTR_ERR(dir
);
1575 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1576 &fail_page_alloc
.ignore_gfp_wait
))
1578 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1579 &fail_page_alloc
.ignore_gfp_highmem
))
1581 if (!debugfs_create_u32("min-order", mode
, dir
,
1582 &fail_page_alloc
.min_order
))
1587 debugfs_remove_recursive(dir
);
1592 late_initcall(fail_page_alloc_debugfs
);
1594 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1596 #else /* CONFIG_FAIL_PAGE_ALLOC */
1598 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1603 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1606 * Return true if free pages are above 'mark'. This takes into account the order
1607 * of the allocation.
1609 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1610 int classzone_idx
, int alloc_flags
, long free_pages
)
1612 /* free_pages my go negative - that's OK */
1614 long lowmem_reserve
= z
->lowmem_reserve
[classzone_idx
];
1617 free_pages
-= (1 << order
) - 1;
1618 if (alloc_flags
& ALLOC_HIGH
)
1620 if (alloc_flags
& ALLOC_HARDER
)
1623 /* If allocation can't use CMA areas don't use free CMA pages */
1624 if (!(alloc_flags
& ALLOC_CMA
))
1625 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
1627 if (free_pages
<= min
+ lowmem_reserve
)
1629 for (o
= 0; o
< order
; o
++) {
1630 /* At the next order, this order's pages become unavailable */
1631 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1633 /* Require fewer higher order pages to be free */
1636 if (free_pages
<= min
)
1642 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1643 int classzone_idx
, int alloc_flags
)
1645 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1646 zone_page_state(z
, NR_FREE_PAGES
));
1649 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1650 int classzone_idx
, int alloc_flags
)
1652 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1654 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1655 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1657 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1663 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1664 * skip over zones that are not allowed by the cpuset, or that have
1665 * been recently (in last second) found to be nearly full. See further
1666 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1667 * that have to skip over a lot of full or unallowed zones.
1669 * If the zonelist cache is present in the passed in zonelist, then
1670 * returns a pointer to the allowed node mask (either the current
1671 * tasks mems_allowed, or node_states[N_MEMORY].)
1673 * If the zonelist cache is not available for this zonelist, does
1674 * nothing and returns NULL.
1676 * If the fullzones BITMAP in the zonelist cache is stale (more than
1677 * a second since last zap'd) then we zap it out (clear its bits.)
1679 * We hold off even calling zlc_setup, until after we've checked the
1680 * first zone in the zonelist, on the theory that most allocations will
1681 * be satisfied from that first zone, so best to examine that zone as
1682 * quickly as we can.
1684 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1686 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1687 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1689 zlc
= zonelist
->zlcache_ptr
;
1693 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1694 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1695 zlc
->last_full_zap
= jiffies
;
1698 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1699 &cpuset_current_mems_allowed
:
1700 &node_states
[N_MEMORY
];
1701 return allowednodes
;
1705 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1706 * if it is worth looking at further for free memory:
1707 * 1) Check that the zone isn't thought to be full (doesn't have its
1708 * bit set in the zonelist_cache fullzones BITMAP).
1709 * 2) Check that the zones node (obtained from the zonelist_cache
1710 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1711 * Return true (non-zero) if zone is worth looking at further, or
1712 * else return false (zero) if it is not.
1714 * This check -ignores- the distinction between various watermarks,
1715 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1716 * found to be full for any variation of these watermarks, it will
1717 * be considered full for up to one second by all requests, unless
1718 * we are so low on memory on all allowed nodes that we are forced
1719 * into the second scan of the zonelist.
1721 * In the second scan we ignore this zonelist cache and exactly
1722 * apply the watermarks to all zones, even it is slower to do so.
1723 * We are low on memory in the second scan, and should leave no stone
1724 * unturned looking for a free page.
1726 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1727 nodemask_t
*allowednodes
)
1729 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1730 int i
; /* index of *z in zonelist zones */
1731 int n
; /* node that zone *z is on */
1733 zlc
= zonelist
->zlcache_ptr
;
1737 i
= z
- zonelist
->_zonerefs
;
1740 /* This zone is worth trying if it is allowed but not full */
1741 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1745 * Given 'z' scanning a zonelist, set the corresponding bit in
1746 * zlc->fullzones, so that subsequent attempts to allocate a page
1747 * from that zone don't waste time re-examining it.
1749 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1751 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1752 int i
; /* index of *z in zonelist zones */
1754 zlc
= zonelist
->zlcache_ptr
;
1758 i
= z
- zonelist
->_zonerefs
;
1760 set_bit(i
, zlc
->fullzones
);
1764 * clear all zones full, called after direct reclaim makes progress so that
1765 * a zone that was recently full is not skipped over for up to a second
1767 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1769 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1771 zlc
= zonelist
->zlcache_ptr
;
1775 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1778 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1780 return node_isset(local_zone
->node
, zone
->zone_pgdat
->reclaim_nodes
);
1783 static void __paginginit
init_zone_allows_reclaim(int nid
)
1787 for_each_online_node(i
)
1788 if (node_distance(nid
, i
) <= RECLAIM_DISTANCE
)
1789 node_set(i
, NODE_DATA(nid
)->reclaim_nodes
);
1791 zone_reclaim_mode
= 1;
1794 #else /* CONFIG_NUMA */
1796 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1801 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1802 nodemask_t
*allowednodes
)
1807 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1811 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1815 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1820 static inline void init_zone_allows_reclaim(int nid
)
1823 #endif /* CONFIG_NUMA */
1826 * get_page_from_freelist goes through the zonelist trying to allocate
1829 static struct page
*
1830 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1831 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1832 struct zone
*preferred_zone
, int migratetype
)
1835 struct page
*page
= NULL
;
1838 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1839 int zlc_active
= 0; /* set if using zonelist_cache */
1840 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1842 classzone_idx
= zone_idx(preferred_zone
);
1845 * Scan zonelist, looking for a zone with enough free.
1846 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1848 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1849 high_zoneidx
, nodemask
) {
1850 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1851 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1853 if ((alloc_flags
& ALLOC_CPUSET
) &&
1854 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1857 * When allocating a page cache page for writing, we
1858 * want to get it from a zone that is within its dirty
1859 * limit, such that no single zone holds more than its
1860 * proportional share of globally allowed dirty pages.
1861 * The dirty limits take into account the zone's
1862 * lowmem reserves and high watermark so that kswapd
1863 * should be able to balance it without having to
1864 * write pages from its LRU list.
1866 * This may look like it could increase pressure on
1867 * lower zones by failing allocations in higher zones
1868 * before they are full. But the pages that do spill
1869 * over are limited as the lower zones are protected
1870 * by this very same mechanism. It should not become
1871 * a practical burden to them.
1873 * XXX: For now, allow allocations to potentially
1874 * exceed the per-zone dirty limit in the slowpath
1875 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1876 * which is important when on a NUMA setup the allowed
1877 * zones are together not big enough to reach the
1878 * global limit. The proper fix for these situations
1879 * will require awareness of zones in the
1880 * dirty-throttling and the flusher threads.
1882 if ((alloc_flags
& ALLOC_WMARK_LOW
) &&
1883 (gfp_mask
& __GFP_WRITE
) && !zone_dirty_ok(zone
))
1884 goto this_zone_full
;
1886 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1887 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1891 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1892 if (zone_watermark_ok(zone
, order
, mark
,
1893 classzone_idx
, alloc_flags
))
1896 if (IS_ENABLED(CONFIG_NUMA
) &&
1897 !did_zlc_setup
&& nr_online_nodes
> 1) {
1899 * we do zlc_setup if there are multiple nodes
1900 * and before considering the first zone allowed
1903 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1908 if (zone_reclaim_mode
== 0 ||
1909 !zone_allows_reclaim(preferred_zone
, zone
))
1910 goto this_zone_full
;
1913 * As we may have just activated ZLC, check if the first
1914 * eligible zone has failed zone_reclaim recently.
1916 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1917 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1920 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1922 case ZONE_RECLAIM_NOSCAN
:
1925 case ZONE_RECLAIM_FULL
:
1926 /* scanned but unreclaimable */
1929 /* did we reclaim enough */
1930 if (!zone_watermark_ok(zone
, order
, mark
,
1931 classzone_idx
, alloc_flags
))
1932 goto this_zone_full
;
1937 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1938 gfp_mask
, migratetype
);
1942 if (IS_ENABLED(CONFIG_NUMA
))
1943 zlc_mark_zone_full(zonelist
, z
);
1946 if (unlikely(IS_ENABLED(CONFIG_NUMA
) && page
== NULL
&& zlc_active
)) {
1947 /* Disable zlc cache for second zonelist scan */
1954 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
1955 * necessary to allocate the page. The expectation is
1956 * that the caller is taking steps that will free more
1957 * memory. The caller should avoid the page being used
1958 * for !PFMEMALLOC purposes.
1960 page
->pfmemalloc
= !!(alloc_flags
& ALLOC_NO_WATERMARKS
);
1966 * Large machines with many possible nodes should not always dump per-node
1967 * meminfo in irq context.
1969 static inline bool should_suppress_show_mem(void)
1974 ret
= in_interrupt();
1979 static DEFINE_RATELIMIT_STATE(nopage_rs
,
1980 DEFAULT_RATELIMIT_INTERVAL
,
1981 DEFAULT_RATELIMIT_BURST
);
1983 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
1985 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
1987 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
1988 debug_guardpage_minorder() > 0)
1992 * This documents exceptions given to allocations in certain
1993 * contexts that are allowed to allocate outside current's set
1996 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
1997 if (test_thread_flag(TIF_MEMDIE
) ||
1998 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
1999 filter
&= ~SHOW_MEM_FILTER_NODES
;
2000 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
2001 filter
&= ~SHOW_MEM_FILTER_NODES
;
2004 struct va_format vaf
;
2007 va_start(args
, fmt
);
2012 pr_warn("%pV", &vaf
);
2017 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2018 current
->comm
, order
, gfp_mask
);
2021 if (!should_suppress_show_mem())
2026 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
2027 unsigned long did_some_progress
,
2028 unsigned long pages_reclaimed
)
2030 /* Do not loop if specifically requested */
2031 if (gfp_mask
& __GFP_NORETRY
)
2034 /* Always retry if specifically requested */
2035 if (gfp_mask
& __GFP_NOFAIL
)
2039 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2040 * making forward progress without invoking OOM. Suspend also disables
2041 * storage devices so kswapd will not help. Bail if we are suspending.
2043 if (!did_some_progress
&& pm_suspended_storage())
2047 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2048 * means __GFP_NOFAIL, but that may not be true in other
2051 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
2055 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2056 * specified, then we retry until we no longer reclaim any pages
2057 * (above), or we've reclaimed an order of pages at least as
2058 * large as the allocation's order. In both cases, if the
2059 * allocation still fails, we stop retrying.
2061 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
2067 static inline struct page
*
2068 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2069 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2070 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2075 /* Acquire the OOM killer lock for the zones in zonelist */
2076 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
2077 schedule_timeout_uninterruptible(1);
2082 * Go through the zonelist yet one more time, keep very high watermark
2083 * here, this is only to catch a parallel oom killing, we must fail if
2084 * we're still under heavy pressure.
2086 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
2087 order
, zonelist
, high_zoneidx
,
2088 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
2089 preferred_zone
, migratetype
);
2093 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2094 /* The OOM killer will not help higher order allocs */
2095 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2097 /* The OOM killer does not needlessly kill tasks for lowmem */
2098 if (high_zoneidx
< ZONE_NORMAL
)
2101 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2102 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2103 * The caller should handle page allocation failure by itself if
2104 * it specifies __GFP_THISNODE.
2105 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2107 if (gfp_mask
& __GFP_THISNODE
)
2110 /* Exhausted what can be done so it's blamo time */
2111 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
, false);
2114 clear_zonelist_oom(zonelist
, gfp_mask
);
2118 #ifdef CONFIG_COMPACTION
2119 /* Try memory compaction for high-order allocations before reclaim */
2120 static struct page
*
2121 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2122 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2123 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2124 int migratetype
, bool sync_migration
,
2125 bool *contended_compaction
, bool *deferred_compaction
,
2126 unsigned long *did_some_progress
)
2131 if (compaction_deferred(preferred_zone
, order
)) {
2132 *deferred_compaction
= true;
2136 current
->flags
|= PF_MEMALLOC
;
2137 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
2138 nodemask
, sync_migration
,
2139 contended_compaction
);
2140 current
->flags
&= ~PF_MEMALLOC
;
2142 if (*did_some_progress
!= COMPACT_SKIPPED
) {
2145 /* Page migration frees to the PCP lists but we want merging */
2146 drain_pages(get_cpu());
2149 page
= get_page_from_freelist(gfp_mask
, nodemask
,
2150 order
, zonelist
, high_zoneidx
,
2151 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2152 preferred_zone
, migratetype
);
2154 preferred_zone
->compact_blockskip_flush
= false;
2155 preferred_zone
->compact_considered
= 0;
2156 preferred_zone
->compact_defer_shift
= 0;
2157 if (order
>= preferred_zone
->compact_order_failed
)
2158 preferred_zone
->compact_order_failed
= order
+ 1;
2159 count_vm_event(COMPACTSUCCESS
);
2164 * It's bad if compaction run occurs and fails.
2165 * The most likely reason is that pages exist,
2166 * but not enough to satisfy watermarks.
2168 count_vm_event(COMPACTFAIL
);
2171 * As async compaction considers a subset of pageblocks, only
2172 * defer if the failure was a sync compaction failure.
2175 defer_compaction(preferred_zone
, order
);
2183 static inline struct page
*
2184 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2185 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2186 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2187 int migratetype
, bool sync_migration
,
2188 bool *contended_compaction
, bool *deferred_compaction
,
2189 unsigned long *did_some_progress
)
2193 #endif /* CONFIG_COMPACTION */
2195 /* Perform direct synchronous page reclaim */
2197 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
, struct zonelist
*zonelist
,
2198 nodemask_t
*nodemask
)
2200 struct reclaim_state reclaim_state
;
2205 /* We now go into synchronous reclaim */
2206 cpuset_memory_pressure_bump();
2207 current
->flags
|= PF_MEMALLOC
;
2208 lockdep_set_current_reclaim_state(gfp_mask
);
2209 reclaim_state
.reclaimed_slab
= 0;
2210 current
->reclaim_state
= &reclaim_state
;
2212 progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
2214 current
->reclaim_state
= NULL
;
2215 lockdep_clear_current_reclaim_state();
2216 current
->flags
&= ~PF_MEMALLOC
;
2223 /* The really slow allocator path where we enter direct reclaim */
2224 static inline struct page
*
2225 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2226 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2227 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2228 int migratetype
, unsigned long *did_some_progress
)
2230 struct page
*page
= NULL
;
2231 bool drained
= false;
2233 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
2235 if (unlikely(!(*did_some_progress
)))
2238 /* After successful reclaim, reconsider all zones for allocation */
2239 if (IS_ENABLED(CONFIG_NUMA
))
2240 zlc_clear_zones_full(zonelist
);
2243 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2244 zonelist
, high_zoneidx
,
2245 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2246 preferred_zone
, migratetype
);
2249 * If an allocation failed after direct reclaim, it could be because
2250 * pages are pinned on the per-cpu lists. Drain them and try again
2252 if (!page
&& !drained
) {
2262 * This is called in the allocator slow-path if the allocation request is of
2263 * sufficient urgency to ignore watermarks and take other desperate measures
2265 static inline struct page
*
2266 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2267 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2268 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2274 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2275 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
2276 preferred_zone
, migratetype
);
2278 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2279 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2280 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2286 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
2287 enum zone_type high_zoneidx
,
2288 enum zone_type classzone_idx
)
2293 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
2294 wakeup_kswapd(zone
, order
, classzone_idx
);
2298 gfp_to_alloc_flags(gfp_t gfp_mask
)
2300 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2301 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2303 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2304 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2307 * The caller may dip into page reserves a bit more if the caller
2308 * cannot run direct reclaim, or if the caller has realtime scheduling
2309 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2310 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2312 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2316 * Not worth trying to allocate harder for
2317 * __GFP_NOMEMALLOC even if it can't schedule.
2319 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2320 alloc_flags
|= ALLOC_HARDER
;
2322 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2323 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2325 alloc_flags
&= ~ALLOC_CPUSET
;
2326 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2327 alloc_flags
|= ALLOC_HARDER
;
2329 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2330 if (gfp_mask
& __GFP_MEMALLOC
)
2331 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2332 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2333 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2334 else if (!in_interrupt() &&
2335 ((current
->flags
& PF_MEMALLOC
) ||
2336 unlikely(test_thread_flag(TIF_MEMDIE
))))
2337 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2340 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2341 alloc_flags
|= ALLOC_CMA
;
2346 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2348 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2351 static inline struct page
*
2352 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2353 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2354 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2357 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2358 struct page
*page
= NULL
;
2360 unsigned long pages_reclaimed
= 0;
2361 unsigned long did_some_progress
;
2362 bool sync_migration
= false;
2363 bool deferred_compaction
= false;
2364 bool contended_compaction
= false;
2367 * In the slowpath, we sanity check order to avoid ever trying to
2368 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2369 * be using allocators in order of preference for an area that is
2372 if (order
>= MAX_ORDER
) {
2373 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2378 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2379 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2380 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2381 * using a larger set of nodes after it has established that the
2382 * allowed per node queues are empty and that nodes are
2385 if (IS_ENABLED(CONFIG_NUMA
) &&
2386 (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2390 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2391 wake_all_kswapd(order
, zonelist
, high_zoneidx
,
2392 zone_idx(preferred_zone
));
2395 * OK, we're below the kswapd watermark and have kicked background
2396 * reclaim. Now things get more complex, so set up alloc_flags according
2397 * to how we want to proceed.
2399 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2402 * Find the true preferred zone if the allocation is unconstrained by
2405 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2406 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2410 /* This is the last chance, in general, before the goto nopage. */
2411 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2412 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2413 preferred_zone
, migratetype
);
2417 /* Allocate without watermarks if the context allows */
2418 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2420 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2421 * the allocation is high priority and these type of
2422 * allocations are system rather than user orientated
2424 zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
2426 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2427 zonelist
, high_zoneidx
, nodemask
,
2428 preferred_zone
, migratetype
);
2434 /* Atomic allocations - we can't balance anything */
2438 /* Avoid recursion of direct reclaim */
2439 if (current
->flags
& PF_MEMALLOC
)
2442 /* Avoid allocations with no watermarks from looping endlessly */
2443 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2447 * Try direct compaction. The first pass is asynchronous. Subsequent
2448 * attempts after direct reclaim are synchronous
2450 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2451 zonelist
, high_zoneidx
,
2453 alloc_flags
, preferred_zone
,
2454 migratetype
, sync_migration
,
2455 &contended_compaction
,
2456 &deferred_compaction
,
2457 &did_some_progress
);
2460 sync_migration
= true;
2463 * If compaction is deferred for high-order allocations, it is because
2464 * sync compaction recently failed. In this is the case and the caller
2465 * requested a movable allocation that does not heavily disrupt the
2466 * system then fail the allocation instead of entering direct reclaim.
2468 if ((deferred_compaction
|| contended_compaction
) &&
2469 (gfp_mask
& __GFP_NO_KSWAPD
))
2472 /* Try direct reclaim and then allocating */
2473 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2474 zonelist
, high_zoneidx
,
2476 alloc_flags
, preferred_zone
,
2477 migratetype
, &did_some_progress
);
2482 * If we failed to make any progress reclaiming, then we are
2483 * running out of options and have to consider going OOM
2485 if (!did_some_progress
) {
2486 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2487 if (oom_killer_disabled
)
2489 /* Coredumps can quickly deplete all memory reserves */
2490 if ((current
->flags
& PF_DUMPCORE
) &&
2491 !(gfp_mask
& __GFP_NOFAIL
))
2493 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2494 zonelist
, high_zoneidx
,
2495 nodemask
, preferred_zone
,
2500 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2502 * The oom killer is not called for high-order
2503 * allocations that may fail, so if no progress
2504 * is being made, there are no other options and
2505 * retrying is unlikely to help.
2507 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2510 * The oom killer is not called for lowmem
2511 * allocations to prevent needlessly killing
2514 if (high_zoneidx
< ZONE_NORMAL
)
2522 /* Check if we should retry the allocation */
2523 pages_reclaimed
+= did_some_progress
;
2524 if (should_alloc_retry(gfp_mask
, order
, did_some_progress
,
2526 /* Wait for some write requests to complete then retry */
2527 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2531 * High-order allocations do not necessarily loop after
2532 * direct reclaim and reclaim/compaction depends on compaction
2533 * being called after reclaim so call directly if necessary
2535 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2536 zonelist
, high_zoneidx
,
2538 alloc_flags
, preferred_zone
,
2539 migratetype
, sync_migration
,
2540 &contended_compaction
,
2541 &deferred_compaction
,
2542 &did_some_progress
);
2548 warn_alloc_failed(gfp_mask
, order
, NULL
);
2551 if (kmemcheck_enabled
)
2552 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2558 * This is the 'heart' of the zoned buddy allocator.
2561 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2562 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2564 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2565 struct zone
*preferred_zone
;
2566 struct page
*page
= NULL
;
2567 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2568 unsigned int cpuset_mems_cookie
;
2569 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
;
2570 struct mem_cgroup
*memcg
= NULL
;
2572 gfp_mask
&= gfp_allowed_mask
;
2574 lockdep_trace_alloc(gfp_mask
);
2576 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2578 if (should_fail_alloc_page(gfp_mask
, order
))
2582 * Check the zones suitable for the gfp_mask contain at least one
2583 * valid zone. It's possible to have an empty zonelist as a result
2584 * of GFP_THISNODE and a memoryless node
2586 if (unlikely(!zonelist
->_zonerefs
->zone
))
2590 * Will only have any effect when __GFP_KMEMCG is set. This is
2591 * verified in the (always inline) callee
2593 if (!memcg_kmem_newpage_charge(gfp_mask
, &memcg
, order
))
2597 cpuset_mems_cookie
= get_mems_allowed();
2599 /* The preferred zone is used for statistics later */
2600 first_zones_zonelist(zonelist
, high_zoneidx
,
2601 nodemask
? : &cpuset_current_mems_allowed
,
2603 if (!preferred_zone
)
2607 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2608 alloc_flags
|= ALLOC_CMA
;
2610 /* First allocation attempt */
2611 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2612 zonelist
, high_zoneidx
, alloc_flags
,
2613 preferred_zone
, migratetype
);
2614 if (unlikely(!page
))
2615 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2616 zonelist
, high_zoneidx
, nodemask
,
2617 preferred_zone
, migratetype
);
2619 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2623 * When updating a task's mems_allowed, it is possible to race with
2624 * parallel threads in such a way that an allocation can fail while
2625 * the mask is being updated. If a page allocation is about to fail,
2626 * check if the cpuset changed during allocation and if so, retry.
2628 if (unlikely(!put_mems_allowed(cpuset_mems_cookie
) && !page
))
2631 memcg_kmem_commit_charge(page
, memcg
, order
);
2635 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2638 * Common helper functions.
2640 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2645 * __get_free_pages() returns a 32-bit address, which cannot represent
2648 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2650 page
= alloc_pages(gfp_mask
, order
);
2653 return (unsigned long) page_address(page
);
2655 EXPORT_SYMBOL(__get_free_pages
);
2657 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2659 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2661 EXPORT_SYMBOL(get_zeroed_page
);
2663 void __free_pages(struct page
*page
, unsigned int order
)
2665 if (put_page_testzero(page
)) {
2667 free_hot_cold_page(page
, 0);
2669 __free_pages_ok(page
, order
);
2673 EXPORT_SYMBOL(__free_pages
);
2675 void free_pages(unsigned long addr
, unsigned int order
)
2678 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2679 __free_pages(virt_to_page((void *)addr
), order
);
2683 EXPORT_SYMBOL(free_pages
);
2686 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2687 * pages allocated with __GFP_KMEMCG.
2689 * Those pages are accounted to a particular memcg, embedded in the
2690 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2691 * for that information only to find out that it is NULL for users who have no
2692 * interest in that whatsoever, we provide these functions.
2694 * The caller knows better which flags it relies on.
2696 void __free_memcg_kmem_pages(struct page
*page
, unsigned int order
)
2698 memcg_kmem_uncharge_pages(page
, order
);
2699 __free_pages(page
, order
);
2702 void free_memcg_kmem_pages(unsigned long addr
, unsigned int order
)
2705 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2706 __free_memcg_kmem_pages(virt_to_page((void *)addr
), order
);
2710 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2713 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2714 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2716 split_page(virt_to_page((void *)addr
), order
);
2717 while (used
< alloc_end
) {
2722 return (void *)addr
;
2726 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2727 * @size: the number of bytes to allocate
2728 * @gfp_mask: GFP flags for the allocation
2730 * This function is similar to alloc_pages(), except that it allocates the
2731 * minimum number of pages to satisfy the request. alloc_pages() can only
2732 * allocate memory in power-of-two pages.
2734 * This function is also limited by MAX_ORDER.
2736 * Memory allocated by this function must be released by free_pages_exact().
2738 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2740 unsigned int order
= get_order(size
);
2743 addr
= __get_free_pages(gfp_mask
, order
);
2744 return make_alloc_exact(addr
, order
, size
);
2746 EXPORT_SYMBOL(alloc_pages_exact
);
2749 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2751 * @nid: the preferred node ID where memory should be allocated
2752 * @size: the number of bytes to allocate
2753 * @gfp_mask: GFP flags for the allocation
2755 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2757 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2760 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2762 unsigned order
= get_order(size
);
2763 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2766 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2768 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2771 * free_pages_exact - release memory allocated via alloc_pages_exact()
2772 * @virt: the value returned by alloc_pages_exact.
2773 * @size: size of allocation, same value as passed to alloc_pages_exact().
2775 * Release the memory allocated by a previous call to alloc_pages_exact.
2777 void free_pages_exact(void *virt
, size_t size
)
2779 unsigned long addr
= (unsigned long)virt
;
2780 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2782 while (addr
< end
) {
2787 EXPORT_SYMBOL(free_pages_exact
);
2789 static unsigned int nr_free_zone_pages(int offset
)
2794 /* Just pick one node, since fallback list is circular */
2795 unsigned int sum
= 0;
2797 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2799 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2800 unsigned long size
= zone
->present_pages
;
2801 unsigned long high
= high_wmark_pages(zone
);
2810 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2812 unsigned int nr_free_buffer_pages(void)
2814 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2816 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2819 * Amount of free RAM allocatable within all zones
2821 unsigned int nr_free_pagecache_pages(void)
2823 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2826 static inline void show_node(struct zone
*zone
)
2828 if (IS_ENABLED(CONFIG_NUMA
))
2829 printk("Node %d ", zone_to_nid(zone
));
2832 void si_meminfo(struct sysinfo
*val
)
2834 val
->totalram
= totalram_pages
;
2836 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2837 val
->bufferram
= nr_blockdev_pages();
2838 val
->totalhigh
= totalhigh_pages
;
2839 val
->freehigh
= nr_free_highpages();
2840 val
->mem_unit
= PAGE_SIZE
;
2843 EXPORT_SYMBOL(si_meminfo
);
2846 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2848 pg_data_t
*pgdat
= NODE_DATA(nid
);
2850 val
->totalram
= pgdat
->node_present_pages
;
2851 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2852 #ifdef CONFIG_HIGHMEM
2853 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2854 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2860 val
->mem_unit
= PAGE_SIZE
;
2865 * Determine whether the node should be displayed or not, depending on whether
2866 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2868 bool skip_free_areas_node(unsigned int flags
, int nid
)
2871 unsigned int cpuset_mems_cookie
;
2873 if (!(flags
& SHOW_MEM_FILTER_NODES
))
2877 cpuset_mems_cookie
= get_mems_allowed();
2878 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
2879 } while (!put_mems_allowed(cpuset_mems_cookie
));
2884 #define K(x) ((x) << (PAGE_SHIFT-10))
2886 static void show_migration_types(unsigned char type
)
2888 static const char types
[MIGRATE_TYPES
] = {
2889 [MIGRATE_UNMOVABLE
] = 'U',
2890 [MIGRATE_RECLAIMABLE
] = 'E',
2891 [MIGRATE_MOVABLE
] = 'M',
2892 [MIGRATE_RESERVE
] = 'R',
2894 [MIGRATE_CMA
] = 'C',
2896 [MIGRATE_ISOLATE
] = 'I',
2898 char tmp
[MIGRATE_TYPES
+ 1];
2902 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
2903 if (type
& (1 << i
))
2908 printk("(%s) ", tmp
);
2912 * Show free area list (used inside shift_scroll-lock stuff)
2913 * We also calculate the percentage fragmentation. We do this by counting the
2914 * memory on each free list with the exception of the first item on the list.
2915 * Suppresses nodes that are not allowed by current's cpuset if
2916 * SHOW_MEM_FILTER_NODES is passed.
2918 void show_free_areas(unsigned int filter
)
2923 for_each_populated_zone(zone
) {
2924 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2927 printk("%s per-cpu:\n", zone
->name
);
2929 for_each_online_cpu(cpu
) {
2930 struct per_cpu_pageset
*pageset
;
2932 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2934 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2935 cpu
, pageset
->pcp
.high
,
2936 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2940 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2941 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2943 " dirty:%lu writeback:%lu unstable:%lu\n"
2944 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2945 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
2947 global_page_state(NR_ACTIVE_ANON
),
2948 global_page_state(NR_INACTIVE_ANON
),
2949 global_page_state(NR_ISOLATED_ANON
),
2950 global_page_state(NR_ACTIVE_FILE
),
2951 global_page_state(NR_INACTIVE_FILE
),
2952 global_page_state(NR_ISOLATED_FILE
),
2953 global_page_state(NR_UNEVICTABLE
),
2954 global_page_state(NR_FILE_DIRTY
),
2955 global_page_state(NR_WRITEBACK
),
2956 global_page_state(NR_UNSTABLE_NFS
),
2957 global_page_state(NR_FREE_PAGES
),
2958 global_page_state(NR_SLAB_RECLAIMABLE
),
2959 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2960 global_page_state(NR_FILE_MAPPED
),
2961 global_page_state(NR_SHMEM
),
2962 global_page_state(NR_PAGETABLE
),
2963 global_page_state(NR_BOUNCE
),
2964 global_page_state(NR_FREE_CMA_PAGES
));
2966 for_each_populated_zone(zone
) {
2969 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2977 " active_anon:%lukB"
2978 " inactive_anon:%lukB"
2979 " active_file:%lukB"
2980 " inactive_file:%lukB"
2981 " unevictable:%lukB"
2982 " isolated(anon):%lukB"
2983 " isolated(file):%lukB"
2991 " slab_reclaimable:%lukB"
2992 " slab_unreclaimable:%lukB"
2993 " kernel_stack:%lukB"
2998 " writeback_tmp:%lukB"
2999 " pages_scanned:%lu"
3000 " all_unreclaimable? %s"
3003 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3004 K(min_wmark_pages(zone
)),
3005 K(low_wmark_pages(zone
)),
3006 K(high_wmark_pages(zone
)),
3007 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3008 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3009 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3010 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3011 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3012 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3013 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3014 K(zone
->present_pages
),
3015 K(zone
->managed_pages
),
3016 K(zone_page_state(zone
, NR_MLOCK
)),
3017 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3018 K(zone_page_state(zone
, NR_WRITEBACK
)),
3019 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3020 K(zone_page_state(zone
, NR_SHMEM
)),
3021 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3022 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3023 zone_page_state(zone
, NR_KERNEL_STACK
) *
3025 K(zone_page_state(zone
, NR_PAGETABLE
)),
3026 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3027 K(zone_page_state(zone
, NR_BOUNCE
)),
3028 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3029 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3030 zone
->pages_scanned
,
3031 (zone
->all_unreclaimable
? "yes" : "no")
3033 printk("lowmem_reserve[]:");
3034 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3035 printk(" %lu", zone
->lowmem_reserve
[i
]);
3039 for_each_populated_zone(zone
) {
3040 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
3041 unsigned char types
[MAX_ORDER
];
3043 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3046 printk("%s: ", zone
->name
);
3048 spin_lock_irqsave(&zone
->lock
, flags
);
3049 for (order
= 0; order
< MAX_ORDER
; order
++) {
3050 struct free_area
*area
= &zone
->free_area
[order
];
3053 nr
[order
] = area
->nr_free
;
3054 total
+= nr
[order
] << order
;
3057 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3058 if (!list_empty(&area
->free_list
[type
]))
3059 types
[order
] |= 1 << type
;
3062 spin_unlock_irqrestore(&zone
->lock
, flags
);
3063 for (order
= 0; order
< MAX_ORDER
; order
++) {
3064 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3066 show_migration_types(types
[order
]);
3068 printk("= %lukB\n", K(total
));
3071 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3073 show_swap_cache_info();
3076 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3078 zoneref
->zone
= zone
;
3079 zoneref
->zone_idx
= zone_idx(zone
);
3083 * Builds allocation fallback zone lists.
3085 * Add all populated zones of a node to the zonelist.
3087 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3088 int nr_zones
, enum zone_type zone_type
)
3092 BUG_ON(zone_type
>= MAX_NR_ZONES
);
3097 zone
= pgdat
->node_zones
+ zone_type
;
3098 if (populated_zone(zone
)) {
3099 zoneref_set_zone(zone
,
3100 &zonelist
->_zonerefs
[nr_zones
++]);
3101 check_highest_zone(zone_type
);
3104 } while (zone_type
);
3111 * 0 = automatic detection of better ordering.
3112 * 1 = order by ([node] distance, -zonetype)
3113 * 2 = order by (-zonetype, [node] distance)
3115 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3116 * the same zonelist. So only NUMA can configure this param.
3118 #define ZONELIST_ORDER_DEFAULT 0
3119 #define ZONELIST_ORDER_NODE 1
3120 #define ZONELIST_ORDER_ZONE 2
3122 /* zonelist order in the kernel.
3123 * set_zonelist_order() will set this to NODE or ZONE.
3125 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3126 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3130 /* The value user specified ....changed by config */
3131 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3132 /* string for sysctl */
3133 #define NUMA_ZONELIST_ORDER_LEN 16
3134 char numa_zonelist_order
[16] = "default";
3137 * interface for configure zonelist ordering.
3138 * command line option "numa_zonelist_order"
3139 * = "[dD]efault - default, automatic configuration.
3140 * = "[nN]ode - order by node locality, then by zone within node
3141 * = "[zZ]one - order by zone, then by locality within zone
3144 static int __parse_numa_zonelist_order(char *s
)
3146 if (*s
== 'd' || *s
== 'D') {
3147 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3148 } else if (*s
== 'n' || *s
== 'N') {
3149 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3150 } else if (*s
== 'z' || *s
== 'Z') {
3151 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3154 "Ignoring invalid numa_zonelist_order value: "
3161 static __init
int setup_numa_zonelist_order(char *s
)
3168 ret
= __parse_numa_zonelist_order(s
);
3170 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3174 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3177 * sysctl handler for numa_zonelist_order
3179 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
3180 void __user
*buffer
, size_t *length
,
3183 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3185 static DEFINE_MUTEX(zl_order_mutex
);
3187 mutex_lock(&zl_order_mutex
);
3189 strcpy(saved_string
, (char*)table
->data
);
3190 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3194 int oldval
= user_zonelist_order
;
3195 if (__parse_numa_zonelist_order((char*)table
->data
)) {
3197 * bogus value. restore saved string
3199 strncpy((char*)table
->data
, saved_string
,
3200 NUMA_ZONELIST_ORDER_LEN
);
3201 user_zonelist_order
= oldval
;
3202 } else if (oldval
!= user_zonelist_order
) {
3203 mutex_lock(&zonelists_mutex
);
3204 build_all_zonelists(NULL
, NULL
);
3205 mutex_unlock(&zonelists_mutex
);
3209 mutex_unlock(&zl_order_mutex
);
3214 #define MAX_NODE_LOAD (nr_online_nodes)
3215 static int node_load
[MAX_NUMNODES
];
3218 * find_next_best_node - find the next node that should appear in a given node's fallback list
3219 * @node: node whose fallback list we're appending
3220 * @used_node_mask: nodemask_t of already used nodes
3222 * We use a number of factors to determine which is the next node that should
3223 * appear on a given node's fallback list. The node should not have appeared
3224 * already in @node's fallback list, and it should be the next closest node
3225 * according to the distance array (which contains arbitrary distance values
3226 * from each node to each node in the system), and should also prefer nodes
3227 * with no CPUs, since presumably they'll have very little allocation pressure
3228 * on them otherwise.
3229 * It returns -1 if no node is found.
3231 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3234 int min_val
= INT_MAX
;
3236 const struct cpumask
*tmp
= cpumask_of_node(0);
3238 /* Use the local node if we haven't already */
3239 if (!node_isset(node
, *used_node_mask
)) {
3240 node_set(node
, *used_node_mask
);
3244 for_each_node_state(n
, N_MEMORY
) {
3246 /* Don't want a node to appear more than once */
3247 if (node_isset(n
, *used_node_mask
))
3250 /* Use the distance array to find the distance */
3251 val
= node_distance(node
, n
);
3253 /* Penalize nodes under us ("prefer the next node") */
3256 /* Give preference to headless and unused nodes */
3257 tmp
= cpumask_of_node(n
);
3258 if (!cpumask_empty(tmp
))
3259 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
3261 /* Slight preference for less loaded node */
3262 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
3263 val
+= node_load
[n
];
3265 if (val
< min_val
) {
3272 node_set(best_node
, *used_node_mask
);
3279 * Build zonelists ordered by node and zones within node.
3280 * This results in maximum locality--normal zone overflows into local
3281 * DMA zone, if any--but risks exhausting DMA zone.
3283 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
3286 struct zonelist
*zonelist
;
3288 zonelist
= &pgdat
->node_zonelists
[0];
3289 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
3291 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3293 zonelist
->_zonerefs
[j
].zone
= NULL
;
3294 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3298 * Build gfp_thisnode zonelists
3300 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
3303 struct zonelist
*zonelist
;
3305 zonelist
= &pgdat
->node_zonelists
[1];
3306 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3307 zonelist
->_zonerefs
[j
].zone
= NULL
;
3308 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3312 * Build zonelists ordered by zone and nodes within zones.
3313 * This results in conserving DMA zone[s] until all Normal memory is
3314 * exhausted, but results in overflowing to remote node while memory
3315 * may still exist in local DMA zone.
3317 static int node_order
[MAX_NUMNODES
];
3319 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
3322 int zone_type
; /* needs to be signed */
3324 struct zonelist
*zonelist
;
3326 zonelist
= &pgdat
->node_zonelists
[0];
3328 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
3329 for (j
= 0; j
< nr_nodes
; j
++) {
3330 node
= node_order
[j
];
3331 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
3332 if (populated_zone(z
)) {
3334 &zonelist
->_zonerefs
[pos
++]);
3335 check_highest_zone(zone_type
);
3339 zonelist
->_zonerefs
[pos
].zone
= NULL
;
3340 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
3343 static int default_zonelist_order(void)
3346 unsigned long low_kmem_size
,total_size
;
3350 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3351 * If they are really small and used heavily, the system can fall
3352 * into OOM very easily.
3353 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3355 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3358 for_each_online_node(nid
) {
3359 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3360 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3361 if (populated_zone(z
)) {
3362 if (zone_type
< ZONE_NORMAL
)
3363 low_kmem_size
+= z
->present_pages
;
3364 total_size
+= z
->present_pages
;
3365 } else if (zone_type
== ZONE_NORMAL
) {
3367 * If any node has only lowmem, then node order
3368 * is preferred to allow kernel allocations
3369 * locally; otherwise, they can easily infringe
3370 * on other nodes when there is an abundance of
3371 * lowmem available to allocate from.
3373 return ZONELIST_ORDER_NODE
;
3377 if (!low_kmem_size
|| /* there are no DMA area. */
3378 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
3379 return ZONELIST_ORDER_NODE
;
3381 * look into each node's config.
3382 * If there is a node whose DMA/DMA32 memory is very big area on
3383 * local memory, NODE_ORDER may be suitable.
3385 average_size
= total_size
/
3386 (nodes_weight(node_states
[N_MEMORY
]) + 1);
3387 for_each_online_node(nid
) {
3390 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3391 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3392 if (populated_zone(z
)) {
3393 if (zone_type
< ZONE_NORMAL
)
3394 low_kmem_size
+= z
->present_pages
;
3395 total_size
+= z
->present_pages
;
3398 if (low_kmem_size
&&
3399 total_size
> average_size
&& /* ignore small node */
3400 low_kmem_size
> total_size
* 70/100)
3401 return ZONELIST_ORDER_NODE
;
3403 return ZONELIST_ORDER_ZONE
;
3406 static void set_zonelist_order(void)
3408 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
3409 current_zonelist_order
= default_zonelist_order();
3411 current_zonelist_order
= user_zonelist_order
;
3414 static void build_zonelists(pg_data_t
*pgdat
)
3418 nodemask_t used_mask
;
3419 int local_node
, prev_node
;
3420 struct zonelist
*zonelist
;
3421 int order
= current_zonelist_order
;
3423 /* initialize zonelists */
3424 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3425 zonelist
= pgdat
->node_zonelists
+ i
;
3426 zonelist
->_zonerefs
[0].zone
= NULL
;
3427 zonelist
->_zonerefs
[0].zone_idx
= 0;
3430 /* NUMA-aware ordering of nodes */
3431 local_node
= pgdat
->node_id
;
3432 load
= nr_online_nodes
;
3433 prev_node
= local_node
;
3434 nodes_clear(used_mask
);
3436 memset(node_order
, 0, sizeof(node_order
));
3439 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3441 * We don't want to pressure a particular node.
3442 * So adding penalty to the first node in same
3443 * distance group to make it round-robin.
3445 if (node_distance(local_node
, node
) !=
3446 node_distance(local_node
, prev_node
))
3447 node_load
[node
] = load
;
3451 if (order
== ZONELIST_ORDER_NODE
)
3452 build_zonelists_in_node_order(pgdat
, node
);
3454 node_order
[j
++] = node
; /* remember order */
3457 if (order
== ZONELIST_ORDER_ZONE
) {
3458 /* calculate node order -- i.e., DMA last! */
3459 build_zonelists_in_zone_order(pgdat
, j
);
3462 build_thisnode_zonelists(pgdat
);
3465 /* Construct the zonelist performance cache - see further mmzone.h */
3466 static void build_zonelist_cache(pg_data_t
*pgdat
)
3468 struct zonelist
*zonelist
;
3469 struct zonelist_cache
*zlc
;
3472 zonelist
= &pgdat
->node_zonelists
[0];
3473 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3474 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3475 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3476 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3479 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3481 * Return node id of node used for "local" allocations.
3482 * I.e., first node id of first zone in arg node's generic zonelist.
3483 * Used for initializing percpu 'numa_mem', which is used primarily
3484 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3486 int local_memory_node(int node
)
3490 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3491 gfp_zone(GFP_KERNEL
),
3498 #else /* CONFIG_NUMA */
3500 static void set_zonelist_order(void)
3502 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3505 static void build_zonelists(pg_data_t
*pgdat
)
3507 int node
, local_node
;
3509 struct zonelist
*zonelist
;
3511 local_node
= pgdat
->node_id
;
3513 zonelist
= &pgdat
->node_zonelists
[0];
3514 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3517 * Now we build the zonelist so that it contains the zones
3518 * of all the other nodes.
3519 * We don't want to pressure a particular node, so when
3520 * building the zones for node N, we make sure that the
3521 * zones coming right after the local ones are those from
3522 * node N+1 (modulo N)
3524 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3525 if (!node_online(node
))
3527 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3530 for (node
= 0; node
< local_node
; node
++) {
3531 if (!node_online(node
))
3533 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3537 zonelist
->_zonerefs
[j
].zone
= NULL
;
3538 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3541 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3542 static void build_zonelist_cache(pg_data_t
*pgdat
)
3544 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3547 #endif /* CONFIG_NUMA */
3550 * Boot pageset table. One per cpu which is going to be used for all
3551 * zones and all nodes. The parameters will be set in such a way
3552 * that an item put on a list will immediately be handed over to
3553 * the buddy list. This is safe since pageset manipulation is done
3554 * with interrupts disabled.
3556 * The boot_pagesets must be kept even after bootup is complete for
3557 * unused processors and/or zones. They do play a role for bootstrapping
3558 * hotplugged processors.
3560 * zoneinfo_show() and maybe other functions do
3561 * not check if the processor is online before following the pageset pointer.
3562 * Other parts of the kernel may not check if the zone is available.
3564 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3565 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3566 static void setup_zone_pageset(struct zone
*zone
);
3569 * Global mutex to protect against size modification of zonelists
3570 * as well as to serialize pageset setup for the new populated zone.
3572 DEFINE_MUTEX(zonelists_mutex
);
3574 /* return values int ....just for stop_machine() */
3575 static int __build_all_zonelists(void *data
)
3579 pg_data_t
*self
= data
;
3582 memset(node_load
, 0, sizeof(node_load
));
3585 if (self
&& !node_online(self
->node_id
)) {
3586 build_zonelists(self
);
3587 build_zonelist_cache(self
);
3590 for_each_online_node(nid
) {
3591 pg_data_t
*pgdat
= NODE_DATA(nid
);
3593 build_zonelists(pgdat
);
3594 build_zonelist_cache(pgdat
);
3598 * Initialize the boot_pagesets that are going to be used
3599 * for bootstrapping processors. The real pagesets for
3600 * each zone will be allocated later when the per cpu
3601 * allocator is available.
3603 * boot_pagesets are used also for bootstrapping offline
3604 * cpus if the system is already booted because the pagesets
3605 * are needed to initialize allocators on a specific cpu too.
3606 * F.e. the percpu allocator needs the page allocator which
3607 * needs the percpu allocator in order to allocate its pagesets
3608 * (a chicken-egg dilemma).
3610 for_each_possible_cpu(cpu
) {
3611 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3613 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3615 * We now know the "local memory node" for each node--
3616 * i.e., the node of the first zone in the generic zonelist.
3617 * Set up numa_mem percpu variable for on-line cpus. During
3618 * boot, only the boot cpu should be on-line; we'll init the
3619 * secondary cpus' numa_mem as they come on-line. During
3620 * node/memory hotplug, we'll fixup all on-line cpus.
3622 if (cpu_online(cpu
))
3623 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3631 * Called with zonelists_mutex held always
3632 * unless system_state == SYSTEM_BOOTING.
3634 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
3636 set_zonelist_order();
3638 if (system_state
== SYSTEM_BOOTING
) {
3639 __build_all_zonelists(NULL
);
3640 mminit_verify_zonelist();
3641 cpuset_init_current_mems_allowed();
3643 /* we have to stop all cpus to guarantee there is no user
3645 #ifdef CONFIG_MEMORY_HOTPLUG
3647 setup_zone_pageset(zone
);
3649 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
3650 /* cpuset refresh routine should be here */
3652 vm_total_pages
= nr_free_pagecache_pages();
3654 * Disable grouping by mobility if the number of pages in the
3655 * system is too low to allow the mechanism to work. It would be
3656 * more accurate, but expensive to check per-zone. This check is
3657 * made on memory-hotadd so a system can start with mobility
3658 * disabled and enable it later
3660 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3661 page_group_by_mobility_disabled
= 1;
3663 page_group_by_mobility_disabled
= 0;
3665 printk("Built %i zonelists in %s order, mobility grouping %s. "
3666 "Total pages: %ld\n",
3668 zonelist_order_name
[current_zonelist_order
],
3669 page_group_by_mobility_disabled
? "off" : "on",
3672 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3677 * Helper functions to size the waitqueue hash table.
3678 * Essentially these want to choose hash table sizes sufficiently
3679 * large so that collisions trying to wait on pages are rare.
3680 * But in fact, the number of active page waitqueues on typical
3681 * systems is ridiculously low, less than 200. So this is even
3682 * conservative, even though it seems large.
3684 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3685 * waitqueues, i.e. the size of the waitq table given the number of pages.
3687 #define PAGES_PER_WAITQUEUE 256
3689 #ifndef CONFIG_MEMORY_HOTPLUG
3690 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3692 unsigned long size
= 1;
3694 pages
/= PAGES_PER_WAITQUEUE
;
3696 while (size
< pages
)
3700 * Once we have dozens or even hundreds of threads sleeping
3701 * on IO we've got bigger problems than wait queue collision.
3702 * Limit the size of the wait table to a reasonable size.
3704 size
= min(size
, 4096UL);
3706 return max(size
, 4UL);
3710 * A zone's size might be changed by hot-add, so it is not possible to determine
3711 * a suitable size for its wait_table. So we use the maximum size now.
3713 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3715 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3716 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3717 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3719 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3720 * or more by the traditional way. (See above). It equals:
3722 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3723 * ia64(16K page size) : = ( 8G + 4M)byte.
3724 * powerpc (64K page size) : = (32G +16M)byte.
3726 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3733 * This is an integer logarithm so that shifts can be used later
3734 * to extract the more random high bits from the multiplicative
3735 * hash function before the remainder is taken.
3737 static inline unsigned long wait_table_bits(unsigned long size
)
3742 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3745 * Check if a pageblock contains reserved pages
3747 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3751 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3752 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3759 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3760 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3761 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3762 * higher will lead to a bigger reserve which will get freed as contiguous
3763 * blocks as reclaim kicks in
3765 static void setup_zone_migrate_reserve(struct zone
*zone
)
3767 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3769 unsigned long block_migratetype
;
3773 * Get the start pfn, end pfn and the number of blocks to reserve
3774 * We have to be careful to be aligned to pageblock_nr_pages to
3775 * make sure that we always check pfn_valid for the first page in
3778 start_pfn
= zone
->zone_start_pfn
;
3779 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3780 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3781 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3785 * Reserve blocks are generally in place to help high-order atomic
3786 * allocations that are short-lived. A min_free_kbytes value that
3787 * would result in more than 2 reserve blocks for atomic allocations
3788 * is assumed to be in place to help anti-fragmentation for the
3789 * future allocation of hugepages at runtime.
3791 reserve
= min(2, reserve
);
3793 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3794 if (!pfn_valid(pfn
))
3796 page
= pfn_to_page(pfn
);
3798 /* Watch out for overlapping nodes */
3799 if (page_to_nid(page
) != zone_to_nid(zone
))
3802 block_migratetype
= get_pageblock_migratetype(page
);
3804 /* Only test what is necessary when the reserves are not met */
3807 * Blocks with reserved pages will never free, skip
3810 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3811 if (pageblock_is_reserved(pfn
, block_end_pfn
))
3814 /* If this block is reserved, account for it */
3815 if (block_migratetype
== MIGRATE_RESERVE
) {
3820 /* Suitable for reserving if this block is movable */
3821 if (block_migratetype
== MIGRATE_MOVABLE
) {
3822 set_pageblock_migratetype(page
,
3824 move_freepages_block(zone
, page
,
3832 * If the reserve is met and this is a previous reserved block,
3835 if (block_migratetype
== MIGRATE_RESERVE
) {
3836 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3837 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3843 * Initially all pages are reserved - free ones are freed
3844 * up by free_all_bootmem() once the early boot process is
3845 * done. Non-atomic initialization, single-pass.
3847 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3848 unsigned long start_pfn
, enum memmap_context context
)
3851 unsigned long end_pfn
= start_pfn
+ size
;
3855 if (highest_memmap_pfn
< end_pfn
- 1)
3856 highest_memmap_pfn
= end_pfn
- 1;
3858 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3859 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3861 * There can be holes in boot-time mem_map[]s
3862 * handed to this function. They do not
3863 * exist on hotplugged memory.
3865 if (context
== MEMMAP_EARLY
) {
3866 if (!early_pfn_valid(pfn
))
3868 if (!early_pfn_in_nid(pfn
, nid
))
3871 page
= pfn_to_page(pfn
);
3872 set_page_links(page
, zone
, nid
, pfn
);
3873 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3874 init_page_count(page
);
3875 reset_page_mapcount(page
);
3876 reset_page_last_nid(page
);
3877 SetPageReserved(page
);
3879 * Mark the block movable so that blocks are reserved for
3880 * movable at startup. This will force kernel allocations
3881 * to reserve their blocks rather than leaking throughout
3882 * the address space during boot when many long-lived
3883 * kernel allocations are made. Later some blocks near
3884 * the start are marked MIGRATE_RESERVE by
3885 * setup_zone_migrate_reserve()
3887 * bitmap is created for zone's valid pfn range. but memmap
3888 * can be created for invalid pages (for alignment)
3889 * check here not to call set_pageblock_migratetype() against
3892 if ((z
->zone_start_pfn
<= pfn
)
3893 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3894 && !(pfn
& (pageblock_nr_pages
- 1)))
3895 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3897 INIT_LIST_HEAD(&page
->lru
);
3898 #ifdef WANT_PAGE_VIRTUAL
3899 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3900 if (!is_highmem_idx(zone
))
3901 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3906 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3909 for_each_migratetype_order(order
, t
) {
3910 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3911 zone
->free_area
[order
].nr_free
= 0;
3915 #ifndef __HAVE_ARCH_MEMMAP_INIT
3916 #define memmap_init(size, nid, zone, start_pfn) \
3917 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3920 static int __meminit
zone_batchsize(struct zone
*zone
)
3926 * The per-cpu-pages pools are set to around 1000th of the
3927 * size of the zone. But no more than 1/2 of a meg.
3929 * OK, so we don't know how big the cache is. So guess.
3931 batch
= zone
->present_pages
/ 1024;
3932 if (batch
* PAGE_SIZE
> 512 * 1024)
3933 batch
= (512 * 1024) / PAGE_SIZE
;
3934 batch
/= 4; /* We effectively *= 4 below */
3939 * Clamp the batch to a 2^n - 1 value. Having a power
3940 * of 2 value was found to be more likely to have
3941 * suboptimal cache aliasing properties in some cases.
3943 * For example if 2 tasks are alternately allocating
3944 * batches of pages, one task can end up with a lot
3945 * of pages of one half of the possible page colors
3946 * and the other with pages of the other colors.
3948 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3953 /* The deferral and batching of frees should be suppressed under NOMMU
3956 * The problem is that NOMMU needs to be able to allocate large chunks
3957 * of contiguous memory as there's no hardware page translation to
3958 * assemble apparent contiguous memory from discontiguous pages.
3960 * Queueing large contiguous runs of pages for batching, however,
3961 * causes the pages to actually be freed in smaller chunks. As there
3962 * can be a significant delay between the individual batches being
3963 * recycled, this leads to the once large chunks of space being
3964 * fragmented and becoming unavailable for high-order allocations.
3970 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3972 struct per_cpu_pages
*pcp
;
3975 memset(p
, 0, sizeof(*p
));
3979 pcp
->high
= 6 * batch
;
3980 pcp
->batch
= max(1UL, 1 * batch
);
3981 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3982 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3986 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3987 * to the value high for the pageset p.
3990 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3993 struct per_cpu_pages
*pcp
;
3997 pcp
->batch
= max(1UL, high
/4);
3998 if ((high
/4) > (PAGE_SHIFT
* 8))
3999 pcp
->batch
= PAGE_SHIFT
* 8;
4002 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4006 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4008 for_each_possible_cpu(cpu
) {
4009 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4011 setup_pageset(pcp
, zone_batchsize(zone
));
4013 if (percpu_pagelist_fraction
)
4014 setup_pagelist_highmark(pcp
,
4015 (zone
->present_pages
/
4016 percpu_pagelist_fraction
));
4021 * Allocate per cpu pagesets and initialize them.
4022 * Before this call only boot pagesets were available.
4024 void __init
setup_per_cpu_pageset(void)
4028 for_each_populated_zone(zone
)
4029 setup_zone_pageset(zone
);
4032 static noinline __init_refok
4033 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4036 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4040 * The per-page waitqueue mechanism uses hashed waitqueues
4043 zone
->wait_table_hash_nr_entries
=
4044 wait_table_hash_nr_entries(zone_size_pages
);
4045 zone
->wait_table_bits
=
4046 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4047 alloc_size
= zone
->wait_table_hash_nr_entries
4048 * sizeof(wait_queue_head_t
);
4050 if (!slab_is_available()) {
4051 zone
->wait_table
= (wait_queue_head_t
*)
4052 alloc_bootmem_node_nopanic(pgdat
, alloc_size
);
4055 * This case means that a zone whose size was 0 gets new memory
4056 * via memory hot-add.
4057 * But it may be the case that a new node was hot-added. In
4058 * this case vmalloc() will not be able to use this new node's
4059 * memory - this wait_table must be initialized to use this new
4060 * node itself as well.
4061 * To use this new node's memory, further consideration will be
4064 zone
->wait_table
= vmalloc(alloc_size
);
4066 if (!zone
->wait_table
)
4069 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4070 init_waitqueue_head(zone
->wait_table
+ i
);
4075 static __meminit
void zone_pcp_init(struct zone
*zone
)
4078 * per cpu subsystem is not up at this point. The following code
4079 * relies on the ability of the linker to provide the
4080 * offset of a (static) per cpu variable into the per cpu area.
4082 zone
->pageset
= &boot_pageset
;
4084 if (zone
->present_pages
)
4085 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4086 zone
->name
, zone
->present_pages
,
4087 zone_batchsize(zone
));
4090 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4091 unsigned long zone_start_pfn
,
4093 enum memmap_context context
)
4095 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4097 ret
= zone_wait_table_init(zone
, size
);
4100 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4102 zone
->zone_start_pfn
= zone_start_pfn
;
4104 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4105 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4107 (unsigned long)zone_idx(zone
),
4108 zone_start_pfn
, (zone_start_pfn
+ size
));
4110 zone_init_free_lists(zone
);
4115 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4116 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4118 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4119 * Architectures may implement their own version but if add_active_range()
4120 * was used and there are no special requirements, this is a convenient
4123 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
4125 unsigned long start_pfn
, end_pfn
;
4128 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4129 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
4131 /* This is a memory hole */
4134 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4136 int __meminit
early_pfn_to_nid(unsigned long pfn
)
4140 nid
= __early_pfn_to_nid(pfn
);
4143 /* just returns 0 */
4147 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4148 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
4152 nid
= __early_pfn_to_nid(pfn
);
4153 if (nid
>= 0 && nid
!= node
)
4160 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
4161 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4162 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
4164 * If an architecture guarantees that all ranges registered with
4165 * add_active_ranges() contain no holes and may be freed, this
4166 * this function may be used instead of calling free_bootmem() manually.
4168 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4170 unsigned long start_pfn
, end_pfn
;
4173 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4174 start_pfn
= min(start_pfn
, max_low_pfn
);
4175 end_pfn
= min(end_pfn
, max_low_pfn
);
4177 if (start_pfn
< end_pfn
)
4178 free_bootmem_node(NODE_DATA(this_nid
),
4179 PFN_PHYS(start_pfn
),
4180 (end_pfn
- start_pfn
) << PAGE_SHIFT
);
4185 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4186 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4188 * If an architecture guarantees that all ranges registered with
4189 * add_active_ranges() contain no holes and may be freed, this
4190 * function may be used instead of calling memory_present() manually.
4192 void __init
sparse_memory_present_with_active_regions(int nid
)
4194 unsigned long start_pfn
, end_pfn
;
4197 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4198 memory_present(this_nid
, start_pfn
, end_pfn
);
4202 * get_pfn_range_for_nid - Return the start and end page frames for a node
4203 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4204 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4205 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4207 * It returns the start and end page frame of a node based on information
4208 * provided by an arch calling add_active_range(). If called for a node
4209 * with no available memory, a warning is printed and the start and end
4212 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4213 unsigned long *start_pfn
, unsigned long *end_pfn
)
4215 unsigned long this_start_pfn
, this_end_pfn
;
4221 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4222 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4223 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4226 if (*start_pfn
== -1UL)
4231 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4232 * assumption is made that zones within a node are ordered in monotonic
4233 * increasing memory addresses so that the "highest" populated zone is used
4235 static void __init
find_usable_zone_for_movable(void)
4238 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4239 if (zone_index
== ZONE_MOVABLE
)
4242 if (arch_zone_highest_possible_pfn
[zone_index
] >
4243 arch_zone_lowest_possible_pfn
[zone_index
])
4247 VM_BUG_ON(zone_index
== -1);
4248 movable_zone
= zone_index
;
4252 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4253 * because it is sized independent of architecture. Unlike the other zones,
4254 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4255 * in each node depending on the size of each node and how evenly kernelcore
4256 * is distributed. This helper function adjusts the zone ranges
4257 * provided by the architecture for a given node by using the end of the
4258 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4259 * zones within a node are in order of monotonic increases memory addresses
4261 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4262 unsigned long zone_type
,
4263 unsigned long node_start_pfn
,
4264 unsigned long node_end_pfn
,
4265 unsigned long *zone_start_pfn
,
4266 unsigned long *zone_end_pfn
)
4268 /* Only adjust if ZONE_MOVABLE is on this node */
4269 if (zone_movable_pfn
[nid
]) {
4270 /* Size ZONE_MOVABLE */
4271 if (zone_type
== ZONE_MOVABLE
) {
4272 *zone_start_pfn
= zone_movable_pfn
[nid
];
4273 *zone_end_pfn
= min(node_end_pfn
,
4274 arch_zone_highest_possible_pfn
[movable_zone
]);
4276 /* Adjust for ZONE_MOVABLE starting within this range */
4277 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4278 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4279 *zone_end_pfn
= zone_movable_pfn
[nid
];
4281 /* Check if this whole range is within ZONE_MOVABLE */
4282 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4283 *zone_start_pfn
= *zone_end_pfn
;
4288 * Return the number of pages a zone spans in a node, including holes
4289 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4291 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4292 unsigned long zone_type
,
4293 unsigned long *ignored
)
4295 unsigned long node_start_pfn
, node_end_pfn
;
4296 unsigned long zone_start_pfn
, zone_end_pfn
;
4298 /* Get the start and end of the node and zone */
4299 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4300 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4301 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4302 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4303 node_start_pfn
, node_end_pfn
,
4304 &zone_start_pfn
, &zone_end_pfn
);
4306 /* Check that this node has pages within the zone's required range */
4307 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4310 /* Move the zone boundaries inside the node if necessary */
4311 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4312 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4314 /* Return the spanned pages */
4315 return zone_end_pfn
- zone_start_pfn
;
4319 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4320 * then all holes in the requested range will be accounted for.
4322 unsigned long __meminit
__absent_pages_in_range(int nid
,
4323 unsigned long range_start_pfn
,
4324 unsigned long range_end_pfn
)
4326 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4327 unsigned long start_pfn
, end_pfn
;
4330 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4331 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4332 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4333 nr_absent
-= end_pfn
- start_pfn
;
4339 * absent_pages_in_range - Return number of page frames in holes within a range
4340 * @start_pfn: The start PFN to start searching for holes
4341 * @end_pfn: The end PFN to stop searching for holes
4343 * It returns the number of pages frames in memory holes within a range.
4345 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4346 unsigned long end_pfn
)
4348 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4351 /* Return the number of page frames in holes in a zone on a node */
4352 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4353 unsigned long zone_type
,
4354 unsigned long *ignored
)
4356 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4357 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4358 unsigned long node_start_pfn
, node_end_pfn
;
4359 unsigned long zone_start_pfn
, zone_end_pfn
;
4361 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4362 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4363 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4365 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4366 node_start_pfn
, node_end_pfn
,
4367 &zone_start_pfn
, &zone_end_pfn
);
4368 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4371 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4372 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4373 unsigned long zone_type
,
4374 unsigned long *zones_size
)
4376 return zones_size
[zone_type
];
4379 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4380 unsigned long zone_type
,
4381 unsigned long *zholes_size
)
4386 return zholes_size
[zone_type
];
4389 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4391 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4392 unsigned long *zones_size
, unsigned long *zholes_size
)
4394 unsigned long realtotalpages
, totalpages
= 0;
4397 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4398 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4400 pgdat
->node_spanned_pages
= totalpages
;
4402 realtotalpages
= totalpages
;
4403 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4405 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4407 pgdat
->node_present_pages
= realtotalpages
;
4408 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4412 #ifndef CONFIG_SPARSEMEM
4414 * Calculate the size of the zone->blockflags rounded to an unsigned long
4415 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4416 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4417 * round what is now in bits to nearest long in bits, then return it in
4420 static unsigned long __init
usemap_size(unsigned long zonesize
)
4422 unsigned long usemapsize
;
4424 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4425 usemapsize
= usemapsize
>> pageblock_order
;
4426 usemapsize
*= NR_PAGEBLOCK_BITS
;
4427 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4429 return usemapsize
/ 8;
4432 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4433 struct zone
*zone
, unsigned long zonesize
)
4435 unsigned long usemapsize
= usemap_size(zonesize
);
4436 zone
->pageblock_flags
= NULL
;
4438 zone
->pageblock_flags
= alloc_bootmem_node_nopanic(pgdat
,
4442 static inline void setup_usemap(struct pglist_data
*pgdat
,
4443 struct zone
*zone
, unsigned long zonesize
) {}
4444 #endif /* CONFIG_SPARSEMEM */
4446 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4448 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4449 void __init
set_pageblock_order(void)
4453 /* Check that pageblock_nr_pages has not already been setup */
4454 if (pageblock_order
)
4457 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4458 order
= HUGETLB_PAGE_ORDER
;
4460 order
= MAX_ORDER
- 1;
4463 * Assume the largest contiguous order of interest is a huge page.
4464 * This value may be variable depending on boot parameters on IA64 and
4467 pageblock_order
= order
;
4469 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4472 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4473 * is unused as pageblock_order is set at compile-time. See
4474 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4477 void __init
set_pageblock_order(void)
4481 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4483 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
4484 unsigned long present_pages
)
4486 unsigned long pages
= spanned_pages
;
4489 * Provide a more accurate estimation if there are holes within
4490 * the zone and SPARSEMEM is in use. If there are holes within the
4491 * zone, each populated memory region may cost us one or two extra
4492 * memmap pages due to alignment because memmap pages for each
4493 * populated regions may not naturally algined on page boundary.
4494 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4496 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
4497 IS_ENABLED(CONFIG_SPARSEMEM
))
4498 pages
= present_pages
;
4500 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
4504 * Set up the zone data structures:
4505 * - mark all pages reserved
4506 * - mark all memory queues empty
4507 * - clear the memory bitmaps
4509 * NOTE: pgdat should get zeroed by caller.
4511 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4512 unsigned long *zones_size
, unsigned long *zholes_size
)
4515 int nid
= pgdat
->node_id
;
4516 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4519 pgdat_resize_init(pgdat
);
4520 #ifdef CONFIG_NUMA_BALANCING
4521 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
4522 pgdat
->numabalancing_migrate_nr_pages
= 0;
4523 pgdat
->numabalancing_migrate_next_window
= jiffies
;
4525 init_waitqueue_head(&pgdat
->kswapd_wait
);
4526 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
4527 pgdat_page_cgroup_init(pgdat
);
4529 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4530 struct zone
*zone
= pgdat
->node_zones
+ j
;
4531 unsigned long size
, realsize
, freesize
, memmap_pages
;
4533 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4534 realsize
= freesize
= size
- zone_absent_pages_in_node(nid
, j
,
4538 * Adjust freesize so that it accounts for how much memory
4539 * is used by this zone for memmap. This affects the watermark
4540 * and per-cpu initialisations
4542 memmap_pages
= calc_memmap_size(size
, realsize
);
4543 if (freesize
>= memmap_pages
) {
4544 freesize
-= memmap_pages
;
4547 " %s zone: %lu pages used for memmap\n",
4548 zone_names
[j
], memmap_pages
);
4551 " %s zone: %lu pages exceeds freesize %lu\n",
4552 zone_names
[j
], memmap_pages
, freesize
);
4554 /* Account for reserved pages */
4555 if (j
== 0 && freesize
> dma_reserve
) {
4556 freesize
-= dma_reserve
;
4557 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4558 zone_names
[0], dma_reserve
);
4561 if (!is_highmem_idx(j
))
4562 nr_kernel_pages
+= freesize
;
4563 /* Charge for highmem memmap if there are enough kernel pages */
4564 else if (nr_kernel_pages
> memmap_pages
* 2)
4565 nr_kernel_pages
-= memmap_pages
;
4566 nr_all_pages
+= freesize
;
4568 zone
->spanned_pages
= size
;
4569 zone
->present_pages
= freesize
;
4571 * Set an approximate value for lowmem here, it will be adjusted
4572 * when the bootmem allocator frees pages into the buddy system.
4573 * And all highmem pages will be managed by the buddy system.
4575 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
4578 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
4580 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
4582 zone
->name
= zone_names
[j
];
4583 spin_lock_init(&zone
->lock
);
4584 spin_lock_init(&zone
->lru_lock
);
4585 zone_seqlock_init(zone
);
4586 zone
->zone_pgdat
= pgdat
;
4588 zone_pcp_init(zone
);
4589 lruvec_init(&zone
->lruvec
);
4593 set_pageblock_order();
4594 setup_usemap(pgdat
, zone
, size
);
4595 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4596 size
, MEMMAP_EARLY
);
4598 memmap_init(size
, nid
, j
, zone_start_pfn
);
4599 zone_start_pfn
+= size
;
4603 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4605 /* Skip empty nodes */
4606 if (!pgdat
->node_spanned_pages
)
4609 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4610 /* ia64 gets its own node_mem_map, before this, without bootmem */
4611 if (!pgdat
->node_mem_map
) {
4612 unsigned long size
, start
, end
;
4616 * The zone's endpoints aren't required to be MAX_ORDER
4617 * aligned but the node_mem_map endpoints must be in order
4618 * for the buddy allocator to function correctly.
4620 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4621 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4622 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4623 size
= (end
- start
) * sizeof(struct page
);
4624 map
= alloc_remap(pgdat
->node_id
, size
);
4626 map
= alloc_bootmem_node_nopanic(pgdat
, size
);
4627 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4629 #ifndef CONFIG_NEED_MULTIPLE_NODES
4631 * With no DISCONTIG, the global mem_map is just set as node 0's
4633 if (pgdat
== NODE_DATA(0)) {
4634 mem_map
= NODE_DATA(0)->node_mem_map
;
4635 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4636 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4637 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4638 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4641 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4644 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4645 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4647 pg_data_t
*pgdat
= NODE_DATA(nid
);
4649 /* pg_data_t should be reset to zero when it's allocated */
4650 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
4652 pgdat
->node_id
= nid
;
4653 pgdat
->node_start_pfn
= node_start_pfn
;
4654 init_zone_allows_reclaim(nid
);
4655 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4657 alloc_node_mem_map(pgdat
);
4658 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4659 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4660 nid
, (unsigned long)pgdat
,
4661 (unsigned long)pgdat
->node_mem_map
);
4664 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4667 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4669 #if MAX_NUMNODES > 1
4671 * Figure out the number of possible node ids.
4673 static void __init
setup_nr_node_ids(void)
4676 unsigned int highest
= 0;
4678 for_each_node_mask(node
, node_possible_map
)
4680 nr_node_ids
= highest
+ 1;
4683 static inline void setup_nr_node_ids(void)
4689 * node_map_pfn_alignment - determine the maximum internode alignment
4691 * This function should be called after node map is populated and sorted.
4692 * It calculates the maximum power of two alignment which can distinguish
4695 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4696 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4697 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4698 * shifted, 1GiB is enough and this function will indicate so.
4700 * This is used to test whether pfn -> nid mapping of the chosen memory
4701 * model has fine enough granularity to avoid incorrect mapping for the
4702 * populated node map.
4704 * Returns the determined alignment in pfn's. 0 if there is no alignment
4705 * requirement (single node).
4707 unsigned long __init
node_map_pfn_alignment(void)
4709 unsigned long accl_mask
= 0, last_end
= 0;
4710 unsigned long start
, end
, mask
;
4714 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4715 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4722 * Start with a mask granular enough to pin-point to the
4723 * start pfn and tick off bits one-by-one until it becomes
4724 * too coarse to separate the current node from the last.
4726 mask
= ~((1 << __ffs(start
)) - 1);
4727 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4730 /* accumulate all internode masks */
4734 /* convert mask to number of pages */
4735 return ~accl_mask
+ 1;
4738 /* Find the lowest pfn for a node */
4739 static unsigned long __init
find_min_pfn_for_node(int nid
)
4741 unsigned long min_pfn
= ULONG_MAX
;
4742 unsigned long start_pfn
;
4745 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4746 min_pfn
= min(min_pfn
, start_pfn
);
4748 if (min_pfn
== ULONG_MAX
) {
4750 "Could not find start_pfn for node %d\n", nid
);
4758 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4760 * It returns the minimum PFN based on information provided via
4761 * add_active_range().
4763 unsigned long __init
find_min_pfn_with_active_regions(void)
4765 return find_min_pfn_for_node(MAX_NUMNODES
);
4769 * early_calculate_totalpages()
4770 * Sum pages in active regions for movable zone.
4771 * Populate N_MEMORY for calculating usable_nodes.
4773 static unsigned long __init
early_calculate_totalpages(void)
4775 unsigned long totalpages
= 0;
4776 unsigned long start_pfn
, end_pfn
;
4779 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
4780 unsigned long pages
= end_pfn
- start_pfn
;
4782 totalpages
+= pages
;
4784 node_set_state(nid
, N_MEMORY
);
4790 * Find the PFN the Movable zone begins in each node. Kernel memory
4791 * is spread evenly between nodes as long as the nodes have enough
4792 * memory. When they don't, some nodes will have more kernelcore than
4795 static void __init
find_zone_movable_pfns_for_nodes(void)
4798 unsigned long usable_startpfn
;
4799 unsigned long kernelcore_node
, kernelcore_remaining
;
4800 /* save the state before borrow the nodemask */
4801 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
4802 unsigned long totalpages
= early_calculate_totalpages();
4803 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
4806 * If movablecore was specified, calculate what size of
4807 * kernelcore that corresponds so that memory usable for
4808 * any allocation type is evenly spread. If both kernelcore
4809 * and movablecore are specified, then the value of kernelcore
4810 * will be used for required_kernelcore if it's greater than
4811 * what movablecore would have allowed.
4813 if (required_movablecore
) {
4814 unsigned long corepages
;
4817 * Round-up so that ZONE_MOVABLE is at least as large as what
4818 * was requested by the user
4820 required_movablecore
=
4821 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4822 corepages
= totalpages
- required_movablecore
;
4824 required_kernelcore
= max(required_kernelcore
, corepages
);
4827 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4828 if (!required_kernelcore
)
4831 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4832 find_usable_zone_for_movable();
4833 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4836 /* Spread kernelcore memory as evenly as possible throughout nodes */
4837 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4838 for_each_node_state(nid
, N_MEMORY
) {
4839 unsigned long start_pfn
, end_pfn
;
4842 * Recalculate kernelcore_node if the division per node
4843 * now exceeds what is necessary to satisfy the requested
4844 * amount of memory for the kernel
4846 if (required_kernelcore
< kernelcore_node
)
4847 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4850 * As the map is walked, we track how much memory is usable
4851 * by the kernel using kernelcore_remaining. When it is
4852 * 0, the rest of the node is usable by ZONE_MOVABLE
4854 kernelcore_remaining
= kernelcore_node
;
4856 /* Go through each range of PFNs within this node */
4857 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4858 unsigned long size_pages
;
4860 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
4861 if (start_pfn
>= end_pfn
)
4864 /* Account for what is only usable for kernelcore */
4865 if (start_pfn
< usable_startpfn
) {
4866 unsigned long kernel_pages
;
4867 kernel_pages
= min(end_pfn
, usable_startpfn
)
4870 kernelcore_remaining
-= min(kernel_pages
,
4871 kernelcore_remaining
);
4872 required_kernelcore
-= min(kernel_pages
,
4873 required_kernelcore
);
4875 /* Continue if range is now fully accounted */
4876 if (end_pfn
<= usable_startpfn
) {
4879 * Push zone_movable_pfn to the end so
4880 * that if we have to rebalance
4881 * kernelcore across nodes, we will
4882 * not double account here
4884 zone_movable_pfn
[nid
] = end_pfn
;
4887 start_pfn
= usable_startpfn
;
4891 * The usable PFN range for ZONE_MOVABLE is from
4892 * start_pfn->end_pfn. Calculate size_pages as the
4893 * number of pages used as kernelcore
4895 size_pages
= end_pfn
- start_pfn
;
4896 if (size_pages
> kernelcore_remaining
)
4897 size_pages
= kernelcore_remaining
;
4898 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4901 * Some kernelcore has been met, update counts and
4902 * break if the kernelcore for this node has been
4905 required_kernelcore
-= min(required_kernelcore
,
4907 kernelcore_remaining
-= size_pages
;
4908 if (!kernelcore_remaining
)
4914 * If there is still required_kernelcore, we do another pass with one
4915 * less node in the count. This will push zone_movable_pfn[nid] further
4916 * along on the nodes that still have memory until kernelcore is
4920 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4923 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4924 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4925 zone_movable_pfn
[nid
] =
4926 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4929 /* restore the node_state */
4930 node_states
[N_MEMORY
] = saved_node_state
;
4933 /* Any regular or high memory on that node ? */
4934 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
4936 enum zone_type zone_type
;
4938 if (N_MEMORY
== N_NORMAL_MEMORY
)
4941 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
4942 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4943 if (zone
->present_pages
) {
4944 node_set_state(nid
, N_HIGH_MEMORY
);
4945 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
4946 zone_type
<= ZONE_NORMAL
)
4947 node_set_state(nid
, N_NORMAL_MEMORY
);
4954 * free_area_init_nodes - Initialise all pg_data_t and zone data
4955 * @max_zone_pfn: an array of max PFNs for each zone
4957 * This will call free_area_init_node() for each active node in the system.
4958 * Using the page ranges provided by add_active_range(), the size of each
4959 * zone in each node and their holes is calculated. If the maximum PFN
4960 * between two adjacent zones match, it is assumed that the zone is empty.
4961 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4962 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4963 * starts where the previous one ended. For example, ZONE_DMA32 starts
4964 * at arch_max_dma_pfn.
4966 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4968 unsigned long start_pfn
, end_pfn
;
4971 /* Record where the zone boundaries are */
4972 memset(arch_zone_lowest_possible_pfn
, 0,
4973 sizeof(arch_zone_lowest_possible_pfn
));
4974 memset(arch_zone_highest_possible_pfn
, 0,
4975 sizeof(arch_zone_highest_possible_pfn
));
4976 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4977 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4978 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4979 if (i
== ZONE_MOVABLE
)
4981 arch_zone_lowest_possible_pfn
[i
] =
4982 arch_zone_highest_possible_pfn
[i
-1];
4983 arch_zone_highest_possible_pfn
[i
] =
4984 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4986 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4987 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4989 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4990 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4991 find_zone_movable_pfns_for_nodes();
4993 /* Print out the zone ranges */
4994 printk("Zone ranges:\n");
4995 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4996 if (i
== ZONE_MOVABLE
)
4998 printk(KERN_CONT
" %-8s ", zone_names
[i
]);
4999 if (arch_zone_lowest_possible_pfn
[i
] ==
5000 arch_zone_highest_possible_pfn
[i
])
5001 printk(KERN_CONT
"empty\n");
5003 printk(KERN_CONT
"[mem %0#10lx-%0#10lx]\n",
5004 arch_zone_lowest_possible_pfn
[i
] << PAGE_SHIFT
,
5005 (arch_zone_highest_possible_pfn
[i
]
5006 << PAGE_SHIFT
) - 1);
5009 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5010 printk("Movable zone start for each node\n");
5011 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5012 if (zone_movable_pfn
[i
])
5013 printk(" Node %d: %#010lx\n", i
,
5014 zone_movable_pfn
[i
] << PAGE_SHIFT
);
5017 /* Print out the early node map */
5018 printk("Early memory node ranges\n");
5019 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5020 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid
,
5021 start_pfn
<< PAGE_SHIFT
, (end_pfn
<< PAGE_SHIFT
) - 1);
5023 /* Initialise every node */
5024 mminit_verify_pageflags_layout();
5025 setup_nr_node_ids();
5026 for_each_online_node(nid
) {
5027 pg_data_t
*pgdat
= NODE_DATA(nid
);
5028 free_area_init_node(nid
, NULL
,
5029 find_min_pfn_for_node(nid
), NULL
);
5031 /* Any memory on that node */
5032 if (pgdat
->node_present_pages
)
5033 node_set_state(nid
, N_MEMORY
);
5034 check_for_memory(pgdat
, nid
);
5038 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5040 unsigned long long coremem
;
5044 coremem
= memparse(p
, &p
);
5045 *core
= coremem
>> PAGE_SHIFT
;
5047 /* Paranoid check that UL is enough for the coremem value */
5048 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5054 * kernelcore=size sets the amount of memory for use for allocations that
5055 * cannot be reclaimed or migrated.
5057 static int __init
cmdline_parse_kernelcore(char *p
)
5059 return cmdline_parse_core(p
, &required_kernelcore
);
5063 * movablecore=size sets the amount of memory for use for allocations that
5064 * can be reclaimed or migrated.
5066 static int __init
cmdline_parse_movablecore(char *p
)
5068 return cmdline_parse_core(p
, &required_movablecore
);
5071 early_param("kernelcore", cmdline_parse_kernelcore
);
5072 early_param("movablecore", cmdline_parse_movablecore
);
5074 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5077 * set_dma_reserve - set the specified number of pages reserved in the first zone
5078 * @new_dma_reserve: The number of pages to mark reserved
5080 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5081 * In the DMA zone, a significant percentage may be consumed by kernel image
5082 * and other unfreeable allocations which can skew the watermarks badly. This
5083 * function may optionally be used to account for unfreeable pages in the
5084 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5085 * smaller per-cpu batchsize.
5087 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5089 dma_reserve
= new_dma_reserve
;
5092 void __init
free_area_init(unsigned long *zones_size
)
5094 free_area_init_node(0, zones_size
,
5095 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5098 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5099 unsigned long action
, void *hcpu
)
5101 int cpu
= (unsigned long)hcpu
;
5103 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5104 lru_add_drain_cpu(cpu
);
5108 * Spill the event counters of the dead processor
5109 * into the current processors event counters.
5110 * This artificially elevates the count of the current
5113 vm_events_fold_cpu(cpu
);
5116 * Zero the differential counters of the dead processor
5117 * so that the vm statistics are consistent.
5119 * This is only okay since the processor is dead and cannot
5120 * race with what we are doing.
5122 refresh_cpu_vm_stats(cpu
);
5127 void __init
page_alloc_init(void)
5129 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5133 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5134 * or min_free_kbytes changes.
5136 static void calculate_totalreserve_pages(void)
5138 struct pglist_data
*pgdat
;
5139 unsigned long reserve_pages
= 0;
5140 enum zone_type i
, j
;
5142 for_each_online_pgdat(pgdat
) {
5143 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5144 struct zone
*zone
= pgdat
->node_zones
+ i
;
5145 unsigned long max
= 0;
5147 /* Find valid and maximum lowmem_reserve in the zone */
5148 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5149 if (zone
->lowmem_reserve
[j
] > max
)
5150 max
= zone
->lowmem_reserve
[j
];
5153 /* we treat the high watermark as reserved pages. */
5154 max
+= high_wmark_pages(zone
);
5156 if (max
> zone
->present_pages
)
5157 max
= zone
->present_pages
;
5158 reserve_pages
+= max
;
5160 * Lowmem reserves are not available to
5161 * GFP_HIGHUSER page cache allocations and
5162 * kswapd tries to balance zones to their high
5163 * watermark. As a result, neither should be
5164 * regarded as dirtyable memory, to prevent a
5165 * situation where reclaim has to clean pages
5166 * in order to balance the zones.
5168 zone
->dirty_balance_reserve
= max
;
5171 dirty_balance_reserve
= reserve_pages
;
5172 totalreserve_pages
= reserve_pages
;
5176 * setup_per_zone_lowmem_reserve - called whenever
5177 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5178 * has a correct pages reserved value, so an adequate number of
5179 * pages are left in the zone after a successful __alloc_pages().
5181 static void setup_per_zone_lowmem_reserve(void)
5183 struct pglist_data
*pgdat
;
5184 enum zone_type j
, idx
;
5186 for_each_online_pgdat(pgdat
) {
5187 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5188 struct zone
*zone
= pgdat
->node_zones
+ j
;
5189 unsigned long present_pages
= zone
->present_pages
;
5191 zone
->lowmem_reserve
[j
] = 0;
5195 struct zone
*lower_zone
;
5199 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5200 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5202 lower_zone
= pgdat
->node_zones
+ idx
;
5203 lower_zone
->lowmem_reserve
[j
] = present_pages
/
5204 sysctl_lowmem_reserve_ratio
[idx
];
5205 present_pages
+= lower_zone
->present_pages
;
5210 /* update totalreserve_pages */
5211 calculate_totalreserve_pages();
5214 static void __setup_per_zone_wmarks(void)
5216 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5217 unsigned long lowmem_pages
= 0;
5219 unsigned long flags
;
5221 /* Calculate total number of !ZONE_HIGHMEM pages */
5222 for_each_zone(zone
) {
5223 if (!is_highmem(zone
))
5224 lowmem_pages
+= zone
->present_pages
;
5227 for_each_zone(zone
) {
5230 spin_lock_irqsave(&zone
->lock
, flags
);
5231 tmp
= (u64
)pages_min
* zone
->present_pages
;
5232 do_div(tmp
, lowmem_pages
);
5233 if (is_highmem(zone
)) {
5235 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5236 * need highmem pages, so cap pages_min to a small
5239 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5240 * deltas controls asynch page reclaim, and so should
5241 * not be capped for highmem.
5245 min_pages
= zone
->present_pages
/ 1024;
5246 if (min_pages
< SWAP_CLUSTER_MAX
)
5247 min_pages
= SWAP_CLUSTER_MAX
;
5248 if (min_pages
> 128)
5250 zone
->watermark
[WMARK_MIN
] = min_pages
;
5253 * If it's a lowmem zone, reserve a number of pages
5254 * proportionate to the zone's size.
5256 zone
->watermark
[WMARK_MIN
] = tmp
;
5259 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
5260 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
5262 setup_zone_migrate_reserve(zone
);
5263 spin_unlock_irqrestore(&zone
->lock
, flags
);
5266 /* update totalreserve_pages */
5267 calculate_totalreserve_pages();
5271 * setup_per_zone_wmarks - called when min_free_kbytes changes
5272 * or when memory is hot-{added|removed}
5274 * Ensures that the watermark[min,low,high] values for each zone are set
5275 * correctly with respect to min_free_kbytes.
5277 void setup_per_zone_wmarks(void)
5279 mutex_lock(&zonelists_mutex
);
5280 __setup_per_zone_wmarks();
5281 mutex_unlock(&zonelists_mutex
);
5285 * The inactive anon list should be small enough that the VM never has to
5286 * do too much work, but large enough that each inactive page has a chance
5287 * to be referenced again before it is swapped out.
5289 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5290 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5291 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5292 * the anonymous pages are kept on the inactive list.
5295 * memory ratio inactive anon
5296 * -------------------------------------
5305 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
5307 unsigned int gb
, ratio
;
5309 /* Zone size in gigabytes */
5310 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
5312 ratio
= int_sqrt(10 * gb
);
5316 zone
->inactive_ratio
= ratio
;
5319 static void __meminit
setup_per_zone_inactive_ratio(void)
5324 calculate_zone_inactive_ratio(zone
);
5328 * Initialise min_free_kbytes.
5330 * For small machines we want it small (128k min). For large machines
5331 * we want it large (64MB max). But it is not linear, because network
5332 * bandwidth does not increase linearly with machine size. We use
5334 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5335 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5351 int __meminit
init_per_zone_wmark_min(void)
5353 unsigned long lowmem_kbytes
;
5355 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5357 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5358 if (min_free_kbytes
< 128)
5359 min_free_kbytes
= 128;
5360 if (min_free_kbytes
> 65536)
5361 min_free_kbytes
= 65536;
5362 setup_per_zone_wmarks();
5363 refresh_zone_stat_thresholds();
5364 setup_per_zone_lowmem_reserve();
5365 setup_per_zone_inactive_ratio();
5368 module_init(init_per_zone_wmark_min
)
5371 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5372 * that we can call two helper functions whenever min_free_kbytes
5375 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5376 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5378 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5380 setup_per_zone_wmarks();
5385 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5386 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5391 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5396 zone
->min_unmapped_pages
= (zone
->present_pages
*
5397 sysctl_min_unmapped_ratio
) / 100;
5401 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5402 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5407 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5412 zone
->min_slab_pages
= (zone
->present_pages
*
5413 sysctl_min_slab_ratio
) / 100;
5419 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5420 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5421 * whenever sysctl_lowmem_reserve_ratio changes.
5423 * The reserve ratio obviously has absolutely no relation with the
5424 * minimum watermarks. The lowmem reserve ratio can only make sense
5425 * if in function of the boot time zone sizes.
5427 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5428 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5430 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5431 setup_per_zone_lowmem_reserve();
5436 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5437 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5438 * can have before it gets flushed back to buddy allocator.
5441 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5442 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5448 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5449 if (!write
|| (ret
< 0))
5451 for_each_populated_zone(zone
) {
5452 for_each_possible_cpu(cpu
) {
5454 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5455 setup_pagelist_highmark(
5456 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5462 int hashdist
= HASHDIST_DEFAULT
;
5465 static int __init
set_hashdist(char *str
)
5469 hashdist
= simple_strtoul(str
, &str
, 0);
5472 __setup("hashdist=", set_hashdist
);
5476 * allocate a large system hash table from bootmem
5477 * - it is assumed that the hash table must contain an exact power-of-2
5478 * quantity of entries
5479 * - limit is the number of hash buckets, not the total allocation size
5481 void *__init
alloc_large_system_hash(const char *tablename
,
5482 unsigned long bucketsize
,
5483 unsigned long numentries
,
5486 unsigned int *_hash_shift
,
5487 unsigned int *_hash_mask
,
5488 unsigned long low_limit
,
5489 unsigned long high_limit
)
5491 unsigned long long max
= high_limit
;
5492 unsigned long log2qty
, size
;
5495 /* allow the kernel cmdline to have a say */
5497 /* round applicable memory size up to nearest megabyte */
5498 numentries
= nr_kernel_pages
;
5499 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5500 numentries
>>= 20 - PAGE_SHIFT
;
5501 numentries
<<= 20 - PAGE_SHIFT
;
5503 /* limit to 1 bucket per 2^scale bytes of low memory */
5504 if (scale
> PAGE_SHIFT
)
5505 numentries
>>= (scale
- PAGE_SHIFT
);
5507 numentries
<<= (PAGE_SHIFT
- scale
);
5509 /* Make sure we've got at least a 0-order allocation.. */
5510 if (unlikely(flags
& HASH_SMALL
)) {
5511 /* Makes no sense without HASH_EARLY */
5512 WARN_ON(!(flags
& HASH_EARLY
));
5513 if (!(numentries
>> *_hash_shift
)) {
5514 numentries
= 1UL << *_hash_shift
;
5515 BUG_ON(!numentries
);
5517 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5518 numentries
= PAGE_SIZE
/ bucketsize
;
5520 numentries
= roundup_pow_of_two(numentries
);
5522 /* limit allocation size to 1/16 total memory by default */
5524 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5525 do_div(max
, bucketsize
);
5527 max
= min(max
, 0x80000000ULL
);
5529 if (numentries
< low_limit
)
5530 numentries
= low_limit
;
5531 if (numentries
> max
)
5534 log2qty
= ilog2(numentries
);
5537 size
= bucketsize
<< log2qty
;
5538 if (flags
& HASH_EARLY
)
5539 table
= alloc_bootmem_nopanic(size
);
5541 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5544 * If bucketsize is not a power-of-two, we may free
5545 * some pages at the end of hash table which
5546 * alloc_pages_exact() automatically does
5548 if (get_order(size
) < MAX_ORDER
) {
5549 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5550 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5553 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5556 panic("Failed to allocate %s hash table\n", tablename
);
5558 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5561 ilog2(size
) - PAGE_SHIFT
,
5565 *_hash_shift
= log2qty
;
5567 *_hash_mask
= (1 << log2qty
) - 1;
5572 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5573 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5576 #ifdef CONFIG_SPARSEMEM
5577 return __pfn_to_section(pfn
)->pageblock_flags
;
5579 return zone
->pageblock_flags
;
5580 #endif /* CONFIG_SPARSEMEM */
5583 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5585 #ifdef CONFIG_SPARSEMEM
5586 pfn
&= (PAGES_PER_SECTION
-1);
5587 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5589 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
5590 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5591 #endif /* CONFIG_SPARSEMEM */
5595 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5596 * @page: The page within the block of interest
5597 * @start_bitidx: The first bit of interest to retrieve
5598 * @end_bitidx: The last bit of interest
5599 * returns pageblock_bits flags
5601 unsigned long get_pageblock_flags_group(struct page
*page
,
5602 int start_bitidx
, int end_bitidx
)
5605 unsigned long *bitmap
;
5606 unsigned long pfn
, bitidx
;
5607 unsigned long flags
= 0;
5608 unsigned long value
= 1;
5610 zone
= page_zone(page
);
5611 pfn
= page_to_pfn(page
);
5612 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5613 bitidx
= pfn_to_bitidx(zone
, pfn
);
5615 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5616 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5623 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5624 * @page: The page within the block of interest
5625 * @start_bitidx: The first bit of interest
5626 * @end_bitidx: The last bit of interest
5627 * @flags: The flags to set
5629 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5630 int start_bitidx
, int end_bitidx
)
5633 unsigned long *bitmap
;
5634 unsigned long pfn
, bitidx
;
5635 unsigned long value
= 1;
5637 zone
= page_zone(page
);
5638 pfn
= page_to_pfn(page
);
5639 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5640 bitidx
= pfn_to_bitidx(zone
, pfn
);
5641 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5642 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5644 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5646 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5648 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5652 * This function checks whether pageblock includes unmovable pages or not.
5653 * If @count is not zero, it is okay to include less @count unmovable pages
5655 * PageLRU check wihtout isolation or lru_lock could race so that
5656 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5657 * expect this function should be exact.
5659 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
5660 bool skip_hwpoisoned_pages
)
5662 unsigned long pfn
, iter
, found
;
5666 * For avoiding noise data, lru_add_drain_all() should be called
5667 * If ZONE_MOVABLE, the zone never contains unmovable pages
5669 if (zone_idx(zone
) == ZONE_MOVABLE
)
5671 mt
= get_pageblock_migratetype(page
);
5672 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
5675 pfn
= page_to_pfn(page
);
5676 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
5677 unsigned long check
= pfn
+ iter
;
5679 if (!pfn_valid_within(check
))
5682 page
= pfn_to_page(check
);
5684 * We can't use page_count without pin a page
5685 * because another CPU can free compound page.
5686 * This check already skips compound tails of THP
5687 * because their page->_count is zero at all time.
5689 if (!atomic_read(&page
->_count
)) {
5690 if (PageBuddy(page
))
5691 iter
+= (1 << page_order(page
)) - 1;
5696 * The HWPoisoned page may be not in buddy system, and
5697 * page_count() is not 0.
5699 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
5705 * If there are RECLAIMABLE pages, we need to check it.
5706 * But now, memory offline itself doesn't call shrink_slab()
5707 * and it still to be fixed.
5710 * If the page is not RAM, page_count()should be 0.
5711 * we don't need more check. This is an _used_ not-movable page.
5713 * The problematic thing here is PG_reserved pages. PG_reserved
5714 * is set to both of a memory hole page and a _used_ kernel
5723 bool is_pageblock_removable_nolock(struct page
*page
)
5729 * We have to be careful here because we are iterating over memory
5730 * sections which are not zone aware so we might end up outside of
5731 * the zone but still within the section.
5732 * We have to take care about the node as well. If the node is offline
5733 * its NODE_DATA will be NULL - see page_zone.
5735 if (!node_online(page_to_nid(page
)))
5738 zone
= page_zone(page
);
5739 pfn
= page_to_pfn(page
);
5740 if (zone
->zone_start_pfn
> pfn
||
5741 zone
->zone_start_pfn
+ zone
->spanned_pages
<= pfn
)
5744 return !has_unmovable_pages(zone
, page
, 0, true);
5749 static unsigned long pfn_max_align_down(unsigned long pfn
)
5751 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5752 pageblock_nr_pages
) - 1);
5755 static unsigned long pfn_max_align_up(unsigned long pfn
)
5757 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5758 pageblock_nr_pages
));
5761 /* [start, end) must belong to a single zone. */
5762 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
5763 unsigned long start
, unsigned long end
)
5765 /* This function is based on compact_zone() from compaction.c. */
5766 unsigned long nr_reclaimed
;
5767 unsigned long pfn
= start
;
5768 unsigned int tries
= 0;
5773 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
5774 if (fatal_signal_pending(current
)) {
5779 if (list_empty(&cc
->migratepages
)) {
5780 cc
->nr_migratepages
= 0;
5781 pfn
= isolate_migratepages_range(cc
->zone
, cc
,
5788 } else if (++tries
== 5) {
5789 ret
= ret
< 0 ? ret
: -EBUSY
;
5793 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
5795 cc
->nr_migratepages
-= nr_reclaimed
;
5797 ret
= migrate_pages(&cc
->migratepages
,
5798 alloc_migrate_target
,
5799 0, false, MIGRATE_SYNC
,
5803 putback_movable_pages(&cc
->migratepages
);
5804 return ret
> 0 ? 0 : ret
;
5808 * alloc_contig_range() -- tries to allocate given range of pages
5809 * @start: start PFN to allocate
5810 * @end: one-past-the-last PFN to allocate
5811 * @migratetype: migratetype of the underlaying pageblocks (either
5812 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
5813 * in range must have the same migratetype and it must
5814 * be either of the two.
5816 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
5817 * aligned, however it's the caller's responsibility to guarantee that
5818 * we are the only thread that changes migrate type of pageblocks the
5821 * The PFN range must belong to a single zone.
5823 * Returns zero on success or negative error code. On success all
5824 * pages which PFN is in [start, end) are allocated for the caller and
5825 * need to be freed with free_contig_range().
5827 int alloc_contig_range(unsigned long start
, unsigned long end
,
5828 unsigned migratetype
)
5830 unsigned long outer_start
, outer_end
;
5833 struct compact_control cc
= {
5834 .nr_migratepages
= 0,
5836 .zone
= page_zone(pfn_to_page(start
)),
5838 .ignore_skip_hint
= true,
5840 INIT_LIST_HEAD(&cc
.migratepages
);
5843 * What we do here is we mark all pageblocks in range as
5844 * MIGRATE_ISOLATE. Because pageblock and max order pages may
5845 * have different sizes, and due to the way page allocator
5846 * work, we align the range to biggest of the two pages so
5847 * that page allocator won't try to merge buddies from
5848 * different pageblocks and change MIGRATE_ISOLATE to some
5849 * other migration type.
5851 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
5852 * migrate the pages from an unaligned range (ie. pages that
5853 * we are interested in). This will put all the pages in
5854 * range back to page allocator as MIGRATE_ISOLATE.
5856 * When this is done, we take the pages in range from page
5857 * allocator removing them from the buddy system. This way
5858 * page allocator will never consider using them.
5860 * This lets us mark the pageblocks back as
5861 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
5862 * aligned range but not in the unaligned, original range are
5863 * put back to page allocator so that buddy can use them.
5866 ret
= start_isolate_page_range(pfn_max_align_down(start
),
5867 pfn_max_align_up(end
), migratetype
,
5872 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
5877 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
5878 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
5879 * more, all pages in [start, end) are free in page allocator.
5880 * What we are going to do is to allocate all pages from
5881 * [start, end) (that is remove them from page allocator).
5883 * The only problem is that pages at the beginning and at the
5884 * end of interesting range may be not aligned with pages that
5885 * page allocator holds, ie. they can be part of higher order
5886 * pages. Because of this, we reserve the bigger range and
5887 * once this is done free the pages we are not interested in.
5889 * We don't have to hold zone->lock here because the pages are
5890 * isolated thus they won't get removed from buddy.
5893 lru_add_drain_all();
5897 outer_start
= start
;
5898 while (!PageBuddy(pfn_to_page(outer_start
))) {
5899 if (++order
>= MAX_ORDER
) {
5903 outer_start
&= ~0UL << order
;
5906 /* Make sure the range is really isolated. */
5907 if (test_pages_isolated(outer_start
, end
, false)) {
5908 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
5915 /* Grab isolated pages from freelists. */
5916 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
5922 /* Free head and tail (if any) */
5923 if (start
!= outer_start
)
5924 free_contig_range(outer_start
, start
- outer_start
);
5925 if (end
!= outer_end
)
5926 free_contig_range(end
, outer_end
- end
);
5929 undo_isolate_page_range(pfn_max_align_down(start
),
5930 pfn_max_align_up(end
), migratetype
);
5934 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
5936 unsigned int count
= 0;
5938 for (; nr_pages
--; pfn
++) {
5939 struct page
*page
= pfn_to_page(pfn
);
5941 count
+= page_count(page
) != 1;
5944 WARN(count
!= 0, "%d pages are still in use!\n", count
);
5948 #ifdef CONFIG_MEMORY_HOTPLUG
5949 static int __meminit
__zone_pcp_update(void *data
)
5951 struct zone
*zone
= data
;
5953 unsigned long batch
= zone_batchsize(zone
), flags
;
5955 for_each_possible_cpu(cpu
) {
5956 struct per_cpu_pageset
*pset
;
5957 struct per_cpu_pages
*pcp
;
5959 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
5962 local_irq_save(flags
);
5964 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
5965 drain_zonestat(zone
, pset
);
5966 setup_pageset(pset
, batch
);
5967 local_irq_restore(flags
);
5972 void __meminit
zone_pcp_update(struct zone
*zone
)
5974 stop_machine(__zone_pcp_update
, zone
, NULL
);
5978 void zone_pcp_reset(struct zone
*zone
)
5980 unsigned long flags
;
5982 struct per_cpu_pageset
*pset
;
5984 /* avoid races with drain_pages() */
5985 local_irq_save(flags
);
5986 if (zone
->pageset
!= &boot_pageset
) {
5987 for_each_online_cpu(cpu
) {
5988 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
5989 drain_zonestat(zone
, pset
);
5991 free_percpu(zone
->pageset
);
5992 zone
->pageset
= &boot_pageset
;
5994 local_irq_restore(flags
);
5997 #ifdef CONFIG_MEMORY_HOTREMOVE
5999 * All pages in the range must be isolated before calling this.
6002 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6008 unsigned long flags
;
6009 /* find the first valid pfn */
6010 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6015 zone
= page_zone(pfn_to_page(pfn
));
6016 spin_lock_irqsave(&zone
->lock
, flags
);
6018 while (pfn
< end_pfn
) {
6019 if (!pfn_valid(pfn
)) {
6023 page
= pfn_to_page(pfn
);
6025 * The HWPoisoned page may be not in buddy system, and
6026 * page_count() is not 0.
6028 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6030 SetPageReserved(page
);
6034 BUG_ON(page_count(page
));
6035 BUG_ON(!PageBuddy(page
));
6036 order
= page_order(page
);
6037 #ifdef CONFIG_DEBUG_VM
6038 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6039 pfn
, 1 << order
, end_pfn
);
6041 list_del(&page
->lru
);
6042 rmv_page_order(page
);
6043 zone
->free_area
[order
].nr_free
--;
6044 for (i
= 0; i
< (1 << order
); i
++)
6045 SetPageReserved((page
+i
));
6046 pfn
+= (1 << order
);
6048 spin_unlock_irqrestore(&zone
->lock
, flags
);
6052 #ifdef CONFIG_MEMORY_FAILURE
6053 bool is_free_buddy_page(struct page
*page
)
6055 struct zone
*zone
= page_zone(page
);
6056 unsigned long pfn
= page_to_pfn(page
);
6057 unsigned long flags
;
6060 spin_lock_irqsave(&zone
->lock
, flags
);
6061 for (order
= 0; order
< MAX_ORDER
; order
++) {
6062 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6064 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6067 spin_unlock_irqrestore(&zone
->lock
, flags
);
6069 return order
< MAX_ORDER
;
6073 static const struct trace_print_flags pageflag_names
[] = {
6074 {1UL << PG_locked
, "locked" },
6075 {1UL << PG_error
, "error" },
6076 {1UL << PG_referenced
, "referenced" },
6077 {1UL << PG_uptodate
, "uptodate" },
6078 {1UL << PG_dirty
, "dirty" },
6079 {1UL << PG_lru
, "lru" },
6080 {1UL << PG_active
, "active" },
6081 {1UL << PG_slab
, "slab" },
6082 {1UL << PG_owner_priv_1
, "owner_priv_1" },
6083 {1UL << PG_arch_1
, "arch_1" },
6084 {1UL << PG_reserved
, "reserved" },
6085 {1UL << PG_private
, "private" },
6086 {1UL << PG_private_2
, "private_2" },
6087 {1UL << PG_writeback
, "writeback" },
6088 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6089 {1UL << PG_head
, "head" },
6090 {1UL << PG_tail
, "tail" },
6092 {1UL << PG_compound
, "compound" },
6094 {1UL << PG_swapcache
, "swapcache" },
6095 {1UL << PG_mappedtodisk
, "mappedtodisk" },
6096 {1UL << PG_reclaim
, "reclaim" },
6097 {1UL << PG_swapbacked
, "swapbacked" },
6098 {1UL << PG_unevictable
, "unevictable" },
6100 {1UL << PG_mlocked
, "mlocked" },
6102 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6103 {1UL << PG_uncached
, "uncached" },
6105 #ifdef CONFIG_MEMORY_FAILURE
6106 {1UL << PG_hwpoison
, "hwpoison" },
6108 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6109 {1UL << PG_compound_lock
, "compound_lock" },
6113 static void dump_page_flags(unsigned long flags
)
6115 const char *delim
= "";
6119 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names
) != __NR_PAGEFLAGS
);
6121 printk(KERN_ALERT
"page flags: %#lx(", flags
);
6123 /* remove zone id */
6124 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
6126 for (i
= 0; i
< ARRAY_SIZE(pageflag_names
) && flags
; i
++) {
6128 mask
= pageflag_names
[i
].mask
;
6129 if ((flags
& mask
) != mask
)
6133 printk("%s%s", delim
, pageflag_names
[i
].name
);
6137 /* check for left over flags */
6139 printk("%s%#lx", delim
, flags
);
6144 void dump_page(struct page
*page
)
6147 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6148 page
, atomic_read(&page
->_count
), page_mapcount(page
),
6149 page
->mapping
, page
->index
);
6150 dump_page_flags(page
->flags
);
6151 mem_cgroup_print_bad_page(page
);