2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/notifier.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/stop_machine.h>
46 #include <linux/sort.h>
47 #include <linux/pfn.h>
48 #include <linux/backing-dev.h>
49 #include <linux/fault-inject.h>
50 #include <linux/page-isolation.h>
51 #include <linux/page_cgroup.h>
52 #include <linux/debugobjects.h>
53 #include <linux/kmemleak.h>
54 #include <linux/compaction.h>
55 #include <trace/events/kmem.h>
56 #include <linux/ftrace_event.h>
57 #include <linux/memcontrol.h>
58 #include <linux/prefetch.h>
59 #include <linux/migrate.h>
60 #include <linux/page-debug-flags.h>
62 #include <asm/tlbflush.h>
63 #include <asm/div64.h>
66 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
67 DEFINE_PER_CPU(int, numa_node
);
68 EXPORT_PER_CPU_SYMBOL(numa_node
);
71 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
73 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
74 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
75 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
76 * defined in <linux/topology.h>.
78 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
79 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
83 * Array of node states.
85 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
86 [N_POSSIBLE
] = NODE_MASK_ALL
,
87 [N_ONLINE
] = { { [0] = 1UL } },
89 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
91 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
93 [N_CPU
] = { { [0] = 1UL } },
96 EXPORT_SYMBOL(node_states
);
98 unsigned long totalram_pages __read_mostly
;
99 unsigned long totalreserve_pages __read_mostly
;
101 * When calculating the number of globally allowed dirty pages, there
102 * is a certain number of per-zone reserves that should not be
103 * considered dirtyable memory. This is the sum of those reserves
104 * over all existing zones that contribute dirtyable memory.
106 unsigned long dirty_balance_reserve __read_mostly
;
108 int percpu_pagelist_fraction
;
109 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
111 #ifdef CONFIG_PM_SLEEP
113 * The following functions are used by the suspend/hibernate code to temporarily
114 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
115 * while devices are suspended. To avoid races with the suspend/hibernate code,
116 * they should always be called with pm_mutex held (gfp_allowed_mask also should
117 * only be modified with pm_mutex held, unless the suspend/hibernate code is
118 * guaranteed not to run in parallel with that modification).
121 static gfp_t saved_gfp_mask
;
123 void pm_restore_gfp_mask(void)
125 WARN_ON(!mutex_is_locked(&pm_mutex
));
126 if (saved_gfp_mask
) {
127 gfp_allowed_mask
= saved_gfp_mask
;
132 void pm_restrict_gfp_mask(void)
134 WARN_ON(!mutex_is_locked(&pm_mutex
));
135 WARN_ON(saved_gfp_mask
);
136 saved_gfp_mask
= gfp_allowed_mask
;
137 gfp_allowed_mask
&= ~GFP_IOFS
;
140 bool pm_suspended_storage(void)
142 if ((gfp_allowed_mask
& GFP_IOFS
) == GFP_IOFS
)
146 #endif /* CONFIG_PM_SLEEP */
148 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
149 int pageblock_order __read_mostly
;
152 static void __free_pages_ok(struct page
*page
, unsigned int order
);
155 * results with 256, 32 in the lowmem_reserve sysctl:
156 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
157 * 1G machine -> (16M dma, 784M normal, 224M high)
158 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
159 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
160 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
162 * TBD: should special case ZONE_DMA32 machines here - in those we normally
163 * don't need any ZONE_NORMAL reservation
165 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
166 #ifdef CONFIG_ZONE_DMA
169 #ifdef CONFIG_ZONE_DMA32
172 #ifdef CONFIG_HIGHMEM
178 EXPORT_SYMBOL(totalram_pages
);
180 static char * const zone_names
[MAX_NR_ZONES
] = {
181 #ifdef CONFIG_ZONE_DMA
184 #ifdef CONFIG_ZONE_DMA32
188 #ifdef CONFIG_HIGHMEM
194 int min_free_kbytes
= 1024;
196 static unsigned long __meminitdata nr_kernel_pages
;
197 static unsigned long __meminitdata nr_all_pages
;
198 static unsigned long __meminitdata dma_reserve
;
200 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
201 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
202 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
203 static unsigned long __initdata required_kernelcore
;
204 static unsigned long __initdata required_movablecore
;
205 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
207 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
209 EXPORT_SYMBOL(movable_zone
);
210 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
213 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
214 int nr_online_nodes __read_mostly
= 1;
215 EXPORT_SYMBOL(nr_node_ids
);
216 EXPORT_SYMBOL(nr_online_nodes
);
219 int page_group_by_mobility_disabled __read_mostly
;
223 * Don't use set_pageblock_migratetype(page, MIGRATE_ISOLATE) directly.
224 * Instead, use {un}set_pageblock_isolate.
226 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
229 if (unlikely(page_group_by_mobility_disabled
))
230 migratetype
= MIGRATE_UNMOVABLE
;
232 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
233 PB_migrate
, PB_migrate_end
);
236 bool oom_killer_disabled __read_mostly
;
238 #ifdef CONFIG_DEBUG_VM
239 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
243 unsigned long pfn
= page_to_pfn(page
);
246 seq
= zone_span_seqbegin(zone
);
247 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
249 else if (pfn
< zone
->zone_start_pfn
)
251 } while (zone_span_seqretry(zone
, seq
));
256 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
258 if (!pfn_valid_within(page_to_pfn(page
)))
260 if (zone
!= page_zone(page
))
266 * Temporary debugging check for pages not lying within a given zone.
268 static int bad_range(struct zone
*zone
, struct page
*page
)
270 if (page_outside_zone_boundaries(zone
, page
))
272 if (!page_is_consistent(zone
, page
))
278 static inline int bad_range(struct zone
*zone
, struct page
*page
)
284 static void bad_page(struct page
*page
)
286 static unsigned long resume
;
287 static unsigned long nr_shown
;
288 static unsigned long nr_unshown
;
290 /* Don't complain about poisoned pages */
291 if (PageHWPoison(page
)) {
292 reset_page_mapcount(page
); /* remove PageBuddy */
297 * Allow a burst of 60 reports, then keep quiet for that minute;
298 * or allow a steady drip of one report per second.
300 if (nr_shown
== 60) {
301 if (time_before(jiffies
, resume
)) {
307 "BUG: Bad page state: %lu messages suppressed\n",
314 resume
= jiffies
+ 60 * HZ
;
316 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
317 current
->comm
, page_to_pfn(page
));
323 /* Leave bad fields for debug, except PageBuddy could make trouble */
324 reset_page_mapcount(page
); /* remove PageBuddy */
325 add_taint(TAINT_BAD_PAGE
);
329 * Higher-order pages are called "compound pages". They are structured thusly:
331 * The first PAGE_SIZE page is called the "head page".
333 * The remaining PAGE_SIZE pages are called "tail pages".
335 * All pages have PG_compound set. All tail pages have their ->first_page
336 * pointing at the head page.
338 * The first tail page's ->lru.next holds the address of the compound page's
339 * put_page() function. Its ->lru.prev holds the order of allocation.
340 * This usage means that zero-order pages may not be compound.
343 static void free_compound_page(struct page
*page
)
345 __free_pages_ok(page
, compound_order(page
));
348 void prep_compound_page(struct page
*page
, unsigned long order
)
351 int nr_pages
= 1 << order
;
353 set_compound_page_dtor(page
, free_compound_page
);
354 set_compound_order(page
, order
);
356 for (i
= 1; i
< nr_pages
; i
++) {
357 struct page
*p
= page
+ i
;
359 set_page_count(p
, 0);
360 p
->first_page
= page
;
364 /* update __split_huge_page_refcount if you change this function */
365 static int destroy_compound_page(struct page
*page
, unsigned long order
)
368 int nr_pages
= 1 << order
;
371 if (unlikely(compound_order(page
) != order
) ||
372 unlikely(!PageHead(page
))) {
377 __ClearPageHead(page
);
379 for (i
= 1; i
< nr_pages
; i
++) {
380 struct page
*p
= page
+ i
;
382 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
392 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
397 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
398 * and __GFP_HIGHMEM from hard or soft interrupt context.
400 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
401 for (i
= 0; i
< (1 << order
); i
++)
402 clear_highpage(page
+ i
);
405 #ifdef CONFIG_DEBUG_PAGEALLOC
406 unsigned int _debug_guardpage_minorder
;
408 static int __init
debug_guardpage_minorder_setup(char *buf
)
412 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
413 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
416 _debug_guardpage_minorder
= res
;
417 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
420 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
422 static inline void set_page_guard_flag(struct page
*page
)
424 __set_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
427 static inline void clear_page_guard_flag(struct page
*page
)
429 __clear_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
432 static inline void set_page_guard_flag(struct page
*page
) { }
433 static inline void clear_page_guard_flag(struct page
*page
) { }
436 static inline void set_page_order(struct page
*page
, int order
)
438 set_page_private(page
, order
);
439 __SetPageBuddy(page
);
442 static inline void rmv_page_order(struct page
*page
)
444 __ClearPageBuddy(page
);
445 set_page_private(page
, 0);
449 * Locate the struct page for both the matching buddy in our
450 * pair (buddy1) and the combined O(n+1) page they form (page).
452 * 1) Any buddy B1 will have an order O twin B2 which satisfies
453 * the following equation:
455 * For example, if the starting buddy (buddy2) is #8 its order
457 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
459 * 2) Any buddy B will have an order O+1 parent P which
460 * satisfies the following equation:
463 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
465 static inline unsigned long
466 __find_buddy_index(unsigned long page_idx
, unsigned int order
)
468 return page_idx
^ (1 << order
);
472 * This function checks whether a page is free && is the buddy
473 * we can do coalesce a page and its buddy if
474 * (a) the buddy is not in a hole &&
475 * (b) the buddy is in the buddy system &&
476 * (c) a page and its buddy have the same order &&
477 * (d) a page and its buddy are in the same zone.
479 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
480 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
482 * For recording page's order, we use page_private(page).
484 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
487 if (!pfn_valid_within(page_to_pfn(buddy
)))
490 if (page_zone_id(page
) != page_zone_id(buddy
))
493 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
494 VM_BUG_ON(page_count(buddy
) != 0);
498 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
499 VM_BUG_ON(page_count(buddy
) != 0);
506 * Freeing function for a buddy system allocator.
508 * The concept of a buddy system is to maintain direct-mapped table
509 * (containing bit values) for memory blocks of various "orders".
510 * The bottom level table contains the map for the smallest allocatable
511 * units of memory (here, pages), and each level above it describes
512 * pairs of units from the levels below, hence, "buddies".
513 * At a high level, all that happens here is marking the table entry
514 * at the bottom level available, and propagating the changes upward
515 * as necessary, plus some accounting needed to play nicely with other
516 * parts of the VM system.
517 * At each level, we keep a list of pages, which are heads of continuous
518 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
519 * order is recorded in page_private(page) field.
520 * So when we are allocating or freeing one, we can derive the state of the
521 * other. That is, if we allocate a small block, and both were
522 * free, the remainder of the region must be split into blocks.
523 * If a block is freed, and its buddy is also free, then this
524 * triggers coalescing into a block of larger size.
529 static inline void __free_one_page(struct page
*page
,
530 struct zone
*zone
, unsigned int order
,
533 unsigned long page_idx
;
534 unsigned long combined_idx
;
535 unsigned long uninitialized_var(buddy_idx
);
538 if (unlikely(PageCompound(page
)))
539 if (unlikely(destroy_compound_page(page
, order
)))
542 VM_BUG_ON(migratetype
== -1);
544 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
546 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
547 VM_BUG_ON(bad_range(zone
, page
));
549 while (order
< MAX_ORDER
-1) {
550 buddy_idx
= __find_buddy_index(page_idx
, order
);
551 buddy
= page
+ (buddy_idx
- page_idx
);
552 if (!page_is_buddy(page
, buddy
, order
))
555 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
556 * merge with it and move up one order.
558 if (page_is_guard(buddy
)) {
559 clear_page_guard_flag(buddy
);
560 set_page_private(page
, 0);
561 __mod_zone_freepage_state(zone
, 1 << order
,
564 list_del(&buddy
->lru
);
565 zone
->free_area
[order
].nr_free
--;
566 rmv_page_order(buddy
);
568 combined_idx
= buddy_idx
& page_idx
;
569 page
= page
+ (combined_idx
- page_idx
);
570 page_idx
= combined_idx
;
573 set_page_order(page
, order
);
576 * If this is not the largest possible page, check if the buddy
577 * of the next-highest order is free. If it is, it's possible
578 * that pages are being freed that will coalesce soon. In case,
579 * that is happening, add the free page to the tail of the list
580 * so it's less likely to be used soon and more likely to be merged
581 * as a higher order page
583 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
584 struct page
*higher_page
, *higher_buddy
;
585 combined_idx
= buddy_idx
& page_idx
;
586 higher_page
= page
+ (combined_idx
- page_idx
);
587 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
588 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
589 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
590 list_add_tail(&page
->lru
,
591 &zone
->free_area
[order
].free_list
[migratetype
]);
596 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
598 zone
->free_area
[order
].nr_free
++;
601 static inline int free_pages_check(struct page
*page
)
603 if (unlikely(page_mapcount(page
) |
604 (page
->mapping
!= NULL
) |
605 (atomic_read(&page
->_count
) != 0) |
606 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
) |
607 (mem_cgroup_bad_page_check(page
)))) {
611 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
612 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
617 * Frees a number of pages from the PCP lists
618 * Assumes all pages on list are in same zone, and of same order.
619 * count is the number of pages to free.
621 * If the zone was previously in an "all pages pinned" state then look to
622 * see if this freeing clears that state.
624 * And clear the zone's pages_scanned counter, to hold off the "all pages are
625 * pinned" detection logic.
627 static void free_pcppages_bulk(struct zone
*zone
, int count
,
628 struct per_cpu_pages
*pcp
)
634 spin_lock(&zone
->lock
);
635 zone
->all_unreclaimable
= 0;
636 zone
->pages_scanned
= 0;
640 struct list_head
*list
;
643 * Remove pages from lists in a round-robin fashion. A
644 * batch_free count is maintained that is incremented when an
645 * empty list is encountered. This is so more pages are freed
646 * off fuller lists instead of spinning excessively around empty
651 if (++migratetype
== MIGRATE_PCPTYPES
)
653 list
= &pcp
->lists
[migratetype
];
654 } while (list_empty(list
));
656 /* This is the only non-empty list. Free them all. */
657 if (batch_free
== MIGRATE_PCPTYPES
)
658 batch_free
= to_free
;
661 int mt
; /* migratetype of the to-be-freed page */
663 page
= list_entry(list
->prev
, struct page
, lru
);
664 /* must delete as __free_one_page list manipulates */
665 list_del(&page
->lru
);
666 mt
= get_freepage_migratetype(page
);
667 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
668 __free_one_page(page
, zone
, 0, mt
);
669 trace_mm_page_pcpu_drain(page
, 0, mt
);
670 if (likely(get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)) {
671 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1);
672 if (is_migrate_cma(mt
))
673 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
, 1);
675 } while (--to_free
&& --batch_free
&& !list_empty(list
));
677 spin_unlock(&zone
->lock
);
680 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
683 spin_lock(&zone
->lock
);
684 zone
->all_unreclaimable
= 0;
685 zone
->pages_scanned
= 0;
687 __free_one_page(page
, zone
, order
, migratetype
);
688 if (unlikely(migratetype
!= MIGRATE_ISOLATE
))
689 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
690 spin_unlock(&zone
->lock
);
693 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
698 trace_mm_page_free(page
, order
);
699 kmemcheck_free_shadow(page
, order
);
702 page
->mapping
= NULL
;
703 for (i
= 0; i
< (1 << order
); i
++)
704 bad
+= free_pages_check(page
+ i
);
708 if (!PageHighMem(page
)) {
709 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
710 debug_check_no_obj_freed(page_address(page
),
713 arch_free_page(page
, order
);
714 kernel_map_pages(page
, 1 << order
, 0);
719 static void __free_pages_ok(struct page
*page
, unsigned int order
)
724 if (!free_pages_prepare(page
, order
))
727 local_irq_save(flags
);
728 __count_vm_events(PGFREE
, 1 << order
);
729 migratetype
= get_pageblock_migratetype(page
);
730 set_freepage_migratetype(page
, migratetype
);
731 free_one_page(page_zone(page
), page
, order
, migratetype
);
732 local_irq_restore(flags
);
735 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
737 unsigned int nr_pages
= 1 << order
;
741 for (loop
= 0; loop
< nr_pages
; loop
++) {
742 struct page
*p
= &page
[loop
];
744 if (loop
+ 1 < nr_pages
)
746 __ClearPageReserved(p
);
747 set_page_count(p
, 0);
750 set_page_refcounted(page
);
751 __free_pages(page
, order
);
755 /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
756 void __init
init_cma_reserved_pageblock(struct page
*page
)
758 unsigned i
= pageblock_nr_pages
;
759 struct page
*p
= page
;
762 __ClearPageReserved(p
);
763 set_page_count(p
, 0);
766 set_page_refcounted(page
);
767 set_pageblock_migratetype(page
, MIGRATE_CMA
);
768 __free_pages(page
, pageblock_order
);
769 totalram_pages
+= pageblock_nr_pages
;
774 * The order of subdivision here is critical for the IO subsystem.
775 * Please do not alter this order without good reasons and regression
776 * testing. Specifically, as large blocks of memory are subdivided,
777 * the order in which smaller blocks are delivered depends on the order
778 * they're subdivided in this function. This is the primary factor
779 * influencing the order in which pages are delivered to the IO
780 * subsystem according to empirical testing, and this is also justified
781 * by considering the behavior of a buddy system containing a single
782 * large block of memory acted on by a series of small allocations.
783 * This behavior is a critical factor in sglist merging's success.
787 static inline void expand(struct zone
*zone
, struct page
*page
,
788 int low
, int high
, struct free_area
*area
,
791 unsigned long size
= 1 << high
;
797 VM_BUG_ON(bad_range(zone
, &page
[size
]));
799 #ifdef CONFIG_DEBUG_PAGEALLOC
800 if (high
< debug_guardpage_minorder()) {
802 * Mark as guard pages (or page), that will allow to
803 * merge back to allocator when buddy will be freed.
804 * Corresponding page table entries will not be touched,
805 * pages will stay not present in virtual address space
807 INIT_LIST_HEAD(&page
[size
].lru
);
808 set_page_guard_flag(&page
[size
]);
809 set_page_private(&page
[size
], high
);
810 /* Guard pages are not available for any usage */
811 __mod_zone_freepage_state(zone
, -(1 << high
),
816 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
818 set_page_order(&page
[size
], high
);
823 * This page is about to be returned from the page allocator
825 static inline int check_new_page(struct page
*page
)
827 if (unlikely(page_mapcount(page
) |
828 (page
->mapping
!= NULL
) |
829 (atomic_read(&page
->_count
) != 0) |
830 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
) |
831 (mem_cgroup_bad_page_check(page
)))) {
838 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
842 for (i
= 0; i
< (1 << order
); i
++) {
843 struct page
*p
= page
+ i
;
844 if (unlikely(check_new_page(p
)))
848 set_page_private(page
, 0);
849 set_page_refcounted(page
);
851 arch_alloc_page(page
, order
);
852 kernel_map_pages(page
, 1 << order
, 1);
854 if (gfp_flags
& __GFP_ZERO
)
855 prep_zero_page(page
, order
, gfp_flags
);
857 if (order
&& (gfp_flags
& __GFP_COMP
))
858 prep_compound_page(page
, order
);
864 * Go through the free lists for the given migratetype and remove
865 * the smallest available page from the freelists
868 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
871 unsigned int current_order
;
872 struct free_area
* area
;
875 /* Find a page of the appropriate size in the preferred list */
876 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
877 area
= &(zone
->free_area
[current_order
]);
878 if (list_empty(&area
->free_list
[migratetype
]))
881 page
= list_entry(area
->free_list
[migratetype
].next
,
883 list_del(&page
->lru
);
884 rmv_page_order(page
);
886 expand(zone
, page
, order
, current_order
, area
, migratetype
);
895 * This array describes the order lists are fallen back to when
896 * the free lists for the desirable migrate type are depleted
898 static int fallbacks
[MIGRATE_TYPES
][4] = {
899 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
900 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
902 [MIGRATE_MOVABLE
] = { MIGRATE_CMA
, MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
903 [MIGRATE_CMA
] = { MIGRATE_RESERVE
}, /* Never used */
905 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
907 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
}, /* Never used */
908 [MIGRATE_ISOLATE
] = { MIGRATE_RESERVE
}, /* Never used */
912 * Move the free pages in a range to the free lists of the requested type.
913 * Note that start_page and end_pages are not aligned on a pageblock
914 * boundary. If alignment is required, use move_freepages_block()
916 int move_freepages(struct zone
*zone
,
917 struct page
*start_page
, struct page
*end_page
,
924 #ifndef CONFIG_HOLES_IN_ZONE
926 * page_zone is not safe to call in this context when
927 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
928 * anyway as we check zone boundaries in move_freepages_block().
929 * Remove at a later date when no bug reports exist related to
930 * grouping pages by mobility
932 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
935 for (page
= start_page
; page
<= end_page
;) {
936 /* Make sure we are not inadvertently changing nodes */
937 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
939 if (!pfn_valid_within(page_to_pfn(page
))) {
944 if (!PageBuddy(page
)) {
949 order
= page_order(page
);
950 list_move(&page
->lru
,
951 &zone
->free_area
[order
].free_list
[migratetype
]);
952 set_freepage_migratetype(page
, migratetype
);
954 pages_moved
+= 1 << order
;
960 int move_freepages_block(struct zone
*zone
, struct page
*page
,
963 unsigned long start_pfn
, end_pfn
;
964 struct page
*start_page
, *end_page
;
966 start_pfn
= page_to_pfn(page
);
967 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
968 start_page
= pfn_to_page(start_pfn
);
969 end_page
= start_page
+ pageblock_nr_pages
- 1;
970 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
972 /* Do not cross zone boundaries */
973 if (start_pfn
< zone
->zone_start_pfn
)
975 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
978 return move_freepages(zone
, start_page
, end_page
, migratetype
);
981 static void change_pageblock_range(struct page
*pageblock_page
,
982 int start_order
, int migratetype
)
984 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
986 while (nr_pageblocks
--) {
987 set_pageblock_migratetype(pageblock_page
, migratetype
);
988 pageblock_page
+= pageblock_nr_pages
;
992 /* Remove an element from the buddy allocator from the fallback list */
993 static inline struct page
*
994 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
996 struct free_area
* area
;
1001 /* Find the largest possible block of pages in the other list */
1002 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
1005 migratetype
= fallbacks
[start_migratetype
][i
];
1007 /* MIGRATE_RESERVE handled later if necessary */
1008 if (migratetype
== MIGRATE_RESERVE
)
1011 area
= &(zone
->free_area
[current_order
]);
1012 if (list_empty(&area
->free_list
[migratetype
]))
1015 page
= list_entry(area
->free_list
[migratetype
].next
,
1020 * If breaking a large block of pages, move all free
1021 * pages to the preferred allocation list. If falling
1022 * back for a reclaimable kernel allocation, be more
1023 * aggressive about taking ownership of free pages
1025 * On the other hand, never change migration
1026 * type of MIGRATE_CMA pageblocks nor move CMA
1027 * pages on different free lists. We don't
1028 * want unmovable pages to be allocated from
1029 * MIGRATE_CMA areas.
1031 if (!is_migrate_cma(migratetype
) &&
1032 (unlikely(current_order
>= pageblock_order
/ 2) ||
1033 start_migratetype
== MIGRATE_RECLAIMABLE
||
1034 page_group_by_mobility_disabled
)) {
1036 pages
= move_freepages_block(zone
, page
,
1039 /* Claim the whole block if over half of it is free */
1040 if (pages
>= (1 << (pageblock_order
-1)) ||
1041 page_group_by_mobility_disabled
)
1042 set_pageblock_migratetype(page
,
1045 migratetype
= start_migratetype
;
1048 /* Remove the page from the freelists */
1049 list_del(&page
->lru
);
1050 rmv_page_order(page
);
1052 /* Take ownership for orders >= pageblock_order */
1053 if (current_order
>= pageblock_order
&&
1054 !is_migrate_cma(migratetype
))
1055 change_pageblock_range(page
, current_order
,
1058 expand(zone
, page
, order
, current_order
, area
,
1059 is_migrate_cma(migratetype
)
1060 ? migratetype
: start_migratetype
);
1062 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1063 start_migratetype
, migratetype
);
1073 * Do the hard work of removing an element from the buddy allocator.
1074 * Call me with the zone->lock already held.
1076 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1082 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1084 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1085 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1088 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1089 * is used because __rmqueue_smallest is an inline function
1090 * and we want just one call site
1093 migratetype
= MIGRATE_RESERVE
;
1098 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1103 * Obtain a specified number of elements from the buddy allocator, all under
1104 * a single hold of the lock, for efficiency. Add them to the supplied list.
1105 * Returns the number of new pages which were placed at *list.
1107 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1108 unsigned long count
, struct list_head
*list
,
1109 int migratetype
, int cold
)
1111 int mt
= migratetype
, i
;
1113 spin_lock(&zone
->lock
);
1114 for (i
= 0; i
< count
; ++i
) {
1115 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1116 if (unlikely(page
== NULL
))
1120 * Split buddy pages returned by expand() are received here
1121 * in physical page order. The page is added to the callers and
1122 * list and the list head then moves forward. From the callers
1123 * perspective, the linked list is ordered by page number in
1124 * some conditions. This is useful for IO devices that can
1125 * merge IO requests if the physical pages are ordered
1128 if (likely(cold
== 0))
1129 list_add(&page
->lru
, list
);
1131 list_add_tail(&page
->lru
, list
);
1132 if (IS_ENABLED(CONFIG_CMA
)) {
1133 mt
= get_pageblock_migratetype(page
);
1134 if (!is_migrate_cma(mt
) && mt
!= MIGRATE_ISOLATE
)
1137 set_freepage_migratetype(page
, mt
);
1139 if (is_migrate_cma(mt
))
1140 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1143 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1144 spin_unlock(&zone
->lock
);
1150 * Called from the vmstat counter updater to drain pagesets of this
1151 * currently executing processor on remote nodes after they have
1154 * Note that this function must be called with the thread pinned to
1155 * a single processor.
1157 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1159 unsigned long flags
;
1162 local_irq_save(flags
);
1163 if (pcp
->count
>= pcp
->batch
)
1164 to_drain
= pcp
->batch
;
1166 to_drain
= pcp
->count
;
1168 free_pcppages_bulk(zone
, to_drain
, pcp
);
1169 pcp
->count
-= to_drain
;
1171 local_irq_restore(flags
);
1176 * Drain pages of the indicated processor.
1178 * The processor must either be the current processor and the
1179 * thread pinned to the current processor or a processor that
1182 static void drain_pages(unsigned int cpu
)
1184 unsigned long flags
;
1187 for_each_populated_zone(zone
) {
1188 struct per_cpu_pageset
*pset
;
1189 struct per_cpu_pages
*pcp
;
1191 local_irq_save(flags
);
1192 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1196 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1199 local_irq_restore(flags
);
1204 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1206 void drain_local_pages(void *arg
)
1208 drain_pages(smp_processor_id());
1212 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1214 * Note that this code is protected against sending an IPI to an offline
1215 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1216 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1217 * nothing keeps CPUs from showing up after we populated the cpumask and
1218 * before the call to on_each_cpu_mask().
1220 void drain_all_pages(void)
1223 struct per_cpu_pageset
*pcp
;
1227 * Allocate in the BSS so we wont require allocation in
1228 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1230 static cpumask_t cpus_with_pcps
;
1233 * We don't care about racing with CPU hotplug event
1234 * as offline notification will cause the notified
1235 * cpu to drain that CPU pcps and on_each_cpu_mask
1236 * disables preemption as part of its processing
1238 for_each_online_cpu(cpu
) {
1239 bool has_pcps
= false;
1240 for_each_populated_zone(zone
) {
1241 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1242 if (pcp
->pcp
.count
) {
1248 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1250 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1252 on_each_cpu_mask(&cpus_with_pcps
, drain_local_pages
, NULL
, 1);
1255 #ifdef CONFIG_HIBERNATION
1257 void mark_free_pages(struct zone
*zone
)
1259 unsigned long pfn
, max_zone_pfn
;
1260 unsigned long flags
;
1262 struct list_head
*curr
;
1264 if (!zone
->spanned_pages
)
1267 spin_lock_irqsave(&zone
->lock
, flags
);
1269 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1270 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1271 if (pfn_valid(pfn
)) {
1272 struct page
*page
= pfn_to_page(pfn
);
1274 if (!swsusp_page_is_forbidden(page
))
1275 swsusp_unset_page_free(page
);
1278 for_each_migratetype_order(order
, t
) {
1279 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1282 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1283 for (i
= 0; i
< (1UL << order
); i
++)
1284 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1287 spin_unlock_irqrestore(&zone
->lock
, flags
);
1289 #endif /* CONFIG_PM */
1292 * Free a 0-order page
1293 * cold == 1 ? free a cold page : free a hot page
1295 void free_hot_cold_page(struct page
*page
, int cold
)
1297 struct zone
*zone
= page_zone(page
);
1298 struct per_cpu_pages
*pcp
;
1299 unsigned long flags
;
1302 if (!free_pages_prepare(page
, 0))
1305 migratetype
= get_pageblock_migratetype(page
);
1306 set_freepage_migratetype(page
, migratetype
);
1307 local_irq_save(flags
);
1308 __count_vm_event(PGFREE
);
1311 * We only track unmovable, reclaimable and movable on pcp lists.
1312 * Free ISOLATE pages back to the allocator because they are being
1313 * offlined but treat RESERVE as movable pages so we can get those
1314 * areas back if necessary. Otherwise, we may have to free
1315 * excessively into the page allocator
1317 if (migratetype
>= MIGRATE_PCPTYPES
) {
1318 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1319 free_one_page(zone
, page
, 0, migratetype
);
1322 migratetype
= MIGRATE_MOVABLE
;
1325 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1327 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1329 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1331 if (pcp
->count
>= pcp
->high
) {
1332 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1333 pcp
->count
-= pcp
->batch
;
1337 local_irq_restore(flags
);
1341 * Free a list of 0-order pages
1343 void free_hot_cold_page_list(struct list_head
*list
, int cold
)
1345 struct page
*page
, *next
;
1347 list_for_each_entry_safe(page
, next
, list
, lru
) {
1348 trace_mm_page_free_batched(page
, cold
);
1349 free_hot_cold_page(page
, cold
);
1354 * split_page takes a non-compound higher-order page, and splits it into
1355 * n (1<<order) sub-pages: page[0..n]
1356 * Each sub-page must be freed individually.
1358 * Note: this is probably too low level an operation for use in drivers.
1359 * Please consult with lkml before using this in your driver.
1361 void split_page(struct page
*page
, unsigned int order
)
1365 VM_BUG_ON(PageCompound(page
));
1366 VM_BUG_ON(!page_count(page
));
1368 #ifdef CONFIG_KMEMCHECK
1370 * Split shadow pages too, because free(page[0]) would
1371 * otherwise free the whole shadow.
1373 if (kmemcheck_page_is_tracked(page
))
1374 split_page(virt_to_page(page
[0].shadow
), order
);
1377 for (i
= 1; i
< (1 << order
); i
++)
1378 set_page_refcounted(page
+ i
);
1382 * Similar to the split_page family of functions except that the page
1383 * required at the given order and being isolated now to prevent races
1384 * with parallel allocators
1386 int capture_free_page(struct page
*page
, int alloc_order
, int migratetype
)
1389 unsigned long watermark
;
1393 BUG_ON(!PageBuddy(page
));
1395 zone
= page_zone(page
);
1396 order
= page_order(page
);
1398 /* Obey watermarks as if the page was being allocated */
1399 watermark
= low_wmark_pages(zone
) + (1 << order
);
1400 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1403 /* Remove page from free list */
1404 list_del(&page
->lru
);
1405 zone
->free_area
[order
].nr_free
--;
1406 rmv_page_order(page
);
1408 mt
= get_pageblock_migratetype(page
);
1409 if (unlikely(mt
!= MIGRATE_ISOLATE
))
1410 __mod_zone_freepage_state(zone
, -(1UL << alloc_order
), mt
);
1412 if (alloc_order
!= order
)
1413 expand(zone
, page
, alloc_order
, order
,
1414 &zone
->free_area
[order
], migratetype
);
1416 /* Set the pageblock if the captured page is at least a pageblock */
1417 if (order
>= pageblock_order
- 1) {
1418 struct page
*endpage
= page
+ (1 << order
) - 1;
1419 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
1420 int mt
= get_pageblock_migratetype(page
);
1421 if (mt
!= MIGRATE_ISOLATE
&& !is_migrate_cma(mt
))
1422 set_pageblock_migratetype(page
,
1427 return 1UL << alloc_order
;
1431 * Similar to split_page except the page is already free. As this is only
1432 * being used for migration, the migratetype of the block also changes.
1433 * As this is called with interrupts disabled, the caller is responsible
1434 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1437 * Note: this is probably too low level an operation for use in drivers.
1438 * Please consult with lkml before using this in your driver.
1440 int split_free_page(struct page
*page
)
1445 BUG_ON(!PageBuddy(page
));
1446 order
= page_order(page
);
1448 nr_pages
= capture_free_page(page
, order
, 0);
1452 /* Split into individual pages */
1453 set_page_refcounted(page
);
1454 split_page(page
, order
);
1459 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1460 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1464 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1465 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1468 unsigned long flags
;
1470 int cold
= !!(gfp_flags
& __GFP_COLD
);
1473 if (likely(order
== 0)) {
1474 struct per_cpu_pages
*pcp
;
1475 struct list_head
*list
;
1477 local_irq_save(flags
);
1478 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1479 list
= &pcp
->lists
[migratetype
];
1480 if (list_empty(list
)) {
1481 pcp
->count
+= rmqueue_bulk(zone
, 0,
1484 if (unlikely(list_empty(list
)))
1489 page
= list_entry(list
->prev
, struct page
, lru
);
1491 page
= list_entry(list
->next
, struct page
, lru
);
1493 list_del(&page
->lru
);
1496 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1498 * __GFP_NOFAIL is not to be used in new code.
1500 * All __GFP_NOFAIL callers should be fixed so that they
1501 * properly detect and handle allocation failures.
1503 * We most definitely don't want callers attempting to
1504 * allocate greater than order-1 page units with
1507 WARN_ON_ONCE(order
> 1);
1509 spin_lock_irqsave(&zone
->lock
, flags
);
1510 page
= __rmqueue(zone
, order
, migratetype
);
1511 spin_unlock(&zone
->lock
);
1514 __mod_zone_freepage_state(zone
, -(1 << order
),
1515 get_pageblock_migratetype(page
));
1518 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1519 zone_statistics(preferred_zone
, zone
, gfp_flags
);
1520 local_irq_restore(flags
);
1522 VM_BUG_ON(bad_range(zone
, page
));
1523 if (prep_new_page(page
, order
, gfp_flags
))
1528 local_irq_restore(flags
);
1532 #ifdef CONFIG_FAIL_PAGE_ALLOC
1535 struct fault_attr attr
;
1537 u32 ignore_gfp_highmem
;
1538 u32 ignore_gfp_wait
;
1540 } fail_page_alloc
= {
1541 .attr
= FAULT_ATTR_INITIALIZER
,
1542 .ignore_gfp_wait
= 1,
1543 .ignore_gfp_highmem
= 1,
1547 static int __init
setup_fail_page_alloc(char *str
)
1549 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1551 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1553 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1555 if (order
< fail_page_alloc
.min_order
)
1557 if (gfp_mask
& __GFP_NOFAIL
)
1559 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1561 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1564 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1567 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1569 static int __init
fail_page_alloc_debugfs(void)
1571 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1574 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
1575 &fail_page_alloc
.attr
);
1577 return PTR_ERR(dir
);
1579 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1580 &fail_page_alloc
.ignore_gfp_wait
))
1582 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1583 &fail_page_alloc
.ignore_gfp_highmem
))
1585 if (!debugfs_create_u32("min-order", mode
, dir
,
1586 &fail_page_alloc
.min_order
))
1591 debugfs_remove_recursive(dir
);
1596 late_initcall(fail_page_alloc_debugfs
);
1598 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1600 #else /* CONFIG_FAIL_PAGE_ALLOC */
1602 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1607 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1610 * Return true if free pages are above 'mark'. This takes into account the order
1611 * of the allocation.
1613 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1614 int classzone_idx
, int alloc_flags
, long free_pages
)
1616 /* free_pages my go negative - that's OK */
1618 long lowmem_reserve
= z
->lowmem_reserve
[classzone_idx
];
1621 free_pages
-= (1 << order
) - 1;
1622 if (alloc_flags
& ALLOC_HIGH
)
1624 if (alloc_flags
& ALLOC_HARDER
)
1627 /* If allocation can't use CMA areas don't use free CMA pages */
1628 if (!(alloc_flags
& ALLOC_CMA
))
1629 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
1631 if (free_pages
<= min
+ lowmem_reserve
)
1633 for (o
= 0; o
< order
; o
++) {
1634 /* At the next order, this order's pages become unavailable */
1635 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1637 /* Require fewer higher order pages to be free */
1640 if (free_pages
<= min
)
1646 #ifdef CONFIG_MEMORY_ISOLATION
1647 static inline unsigned long nr_zone_isolate_freepages(struct zone
*zone
)
1649 if (unlikely(zone
->nr_pageblock_isolate
))
1650 return zone
->nr_pageblock_isolate
* pageblock_nr_pages
;
1654 static inline unsigned long nr_zone_isolate_freepages(struct zone
*zone
)
1660 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1661 int classzone_idx
, int alloc_flags
)
1663 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1664 zone_page_state(z
, NR_FREE_PAGES
));
1667 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1668 int classzone_idx
, int alloc_flags
)
1670 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1672 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1673 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1676 * If the zone has MIGRATE_ISOLATE type free pages, we should consider
1677 * it. nr_zone_isolate_freepages is never accurate so kswapd might not
1678 * sleep although it could do so. But this is more desirable for memory
1679 * hotplug than sleeping which can cause a livelock in the direct
1682 free_pages
-= nr_zone_isolate_freepages(z
);
1683 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1689 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1690 * skip over zones that are not allowed by the cpuset, or that have
1691 * been recently (in last second) found to be nearly full. See further
1692 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1693 * that have to skip over a lot of full or unallowed zones.
1695 * If the zonelist cache is present in the passed in zonelist, then
1696 * returns a pointer to the allowed node mask (either the current
1697 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1699 * If the zonelist cache is not available for this zonelist, does
1700 * nothing and returns NULL.
1702 * If the fullzones BITMAP in the zonelist cache is stale (more than
1703 * a second since last zap'd) then we zap it out (clear its bits.)
1705 * We hold off even calling zlc_setup, until after we've checked the
1706 * first zone in the zonelist, on the theory that most allocations will
1707 * be satisfied from that first zone, so best to examine that zone as
1708 * quickly as we can.
1710 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1712 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1713 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1715 zlc
= zonelist
->zlcache_ptr
;
1719 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1720 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1721 zlc
->last_full_zap
= jiffies
;
1724 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1725 &cpuset_current_mems_allowed
:
1726 &node_states
[N_HIGH_MEMORY
];
1727 return allowednodes
;
1731 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1732 * if it is worth looking at further for free memory:
1733 * 1) Check that the zone isn't thought to be full (doesn't have its
1734 * bit set in the zonelist_cache fullzones BITMAP).
1735 * 2) Check that the zones node (obtained from the zonelist_cache
1736 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1737 * Return true (non-zero) if zone is worth looking at further, or
1738 * else return false (zero) if it is not.
1740 * This check -ignores- the distinction between various watermarks,
1741 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1742 * found to be full for any variation of these watermarks, it will
1743 * be considered full for up to one second by all requests, unless
1744 * we are so low on memory on all allowed nodes that we are forced
1745 * into the second scan of the zonelist.
1747 * In the second scan we ignore this zonelist cache and exactly
1748 * apply the watermarks to all zones, even it is slower to do so.
1749 * We are low on memory in the second scan, and should leave no stone
1750 * unturned looking for a free page.
1752 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1753 nodemask_t
*allowednodes
)
1755 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1756 int i
; /* index of *z in zonelist zones */
1757 int n
; /* node that zone *z is on */
1759 zlc
= zonelist
->zlcache_ptr
;
1763 i
= z
- zonelist
->_zonerefs
;
1766 /* This zone is worth trying if it is allowed but not full */
1767 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1771 * Given 'z' scanning a zonelist, set the corresponding bit in
1772 * zlc->fullzones, so that subsequent attempts to allocate a page
1773 * from that zone don't waste time re-examining it.
1775 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1777 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1778 int i
; /* index of *z in zonelist zones */
1780 zlc
= zonelist
->zlcache_ptr
;
1784 i
= z
- zonelist
->_zonerefs
;
1786 set_bit(i
, zlc
->fullzones
);
1790 * clear all zones full, called after direct reclaim makes progress so that
1791 * a zone that was recently full is not skipped over for up to a second
1793 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1795 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1797 zlc
= zonelist
->zlcache_ptr
;
1801 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1804 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1806 return node_isset(local_zone
->node
, zone
->zone_pgdat
->reclaim_nodes
);
1809 static void __paginginit
init_zone_allows_reclaim(int nid
)
1813 for_each_online_node(i
)
1814 if (node_distance(nid
, i
) <= RECLAIM_DISTANCE
)
1815 node_set(i
, NODE_DATA(nid
)->reclaim_nodes
);
1817 zone_reclaim_mode
= 1;
1820 #else /* CONFIG_NUMA */
1822 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1827 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1828 nodemask_t
*allowednodes
)
1833 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1837 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1841 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1846 static inline void init_zone_allows_reclaim(int nid
)
1849 #endif /* CONFIG_NUMA */
1852 * get_page_from_freelist goes through the zonelist trying to allocate
1855 static struct page
*
1856 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1857 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1858 struct zone
*preferred_zone
, int migratetype
)
1861 struct page
*page
= NULL
;
1864 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1865 int zlc_active
= 0; /* set if using zonelist_cache */
1866 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1868 classzone_idx
= zone_idx(preferred_zone
);
1871 * Scan zonelist, looking for a zone with enough free.
1872 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1874 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1875 high_zoneidx
, nodemask
) {
1876 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1877 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1879 if ((alloc_flags
& ALLOC_CPUSET
) &&
1880 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1883 * When allocating a page cache page for writing, we
1884 * want to get it from a zone that is within its dirty
1885 * limit, such that no single zone holds more than its
1886 * proportional share of globally allowed dirty pages.
1887 * The dirty limits take into account the zone's
1888 * lowmem reserves and high watermark so that kswapd
1889 * should be able to balance it without having to
1890 * write pages from its LRU list.
1892 * This may look like it could increase pressure on
1893 * lower zones by failing allocations in higher zones
1894 * before they are full. But the pages that do spill
1895 * over are limited as the lower zones are protected
1896 * by this very same mechanism. It should not become
1897 * a practical burden to them.
1899 * XXX: For now, allow allocations to potentially
1900 * exceed the per-zone dirty limit in the slowpath
1901 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1902 * which is important when on a NUMA setup the allowed
1903 * zones are together not big enough to reach the
1904 * global limit. The proper fix for these situations
1905 * will require awareness of zones in the
1906 * dirty-throttling and the flusher threads.
1908 if ((alloc_flags
& ALLOC_WMARK_LOW
) &&
1909 (gfp_mask
& __GFP_WRITE
) && !zone_dirty_ok(zone
))
1910 goto this_zone_full
;
1912 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1913 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1917 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1918 if (zone_watermark_ok(zone
, order
, mark
,
1919 classzone_idx
, alloc_flags
))
1922 if (IS_ENABLED(CONFIG_NUMA
) &&
1923 !did_zlc_setup
&& nr_online_nodes
> 1) {
1925 * we do zlc_setup if there are multiple nodes
1926 * and before considering the first zone allowed
1929 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1934 if (zone_reclaim_mode
== 0 ||
1935 !zone_allows_reclaim(preferred_zone
, zone
))
1936 goto this_zone_full
;
1939 * As we may have just activated ZLC, check if the first
1940 * eligible zone has failed zone_reclaim recently.
1942 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1943 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1946 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1948 case ZONE_RECLAIM_NOSCAN
:
1951 case ZONE_RECLAIM_FULL
:
1952 /* scanned but unreclaimable */
1955 /* did we reclaim enough */
1956 if (!zone_watermark_ok(zone
, order
, mark
,
1957 classzone_idx
, alloc_flags
))
1958 goto this_zone_full
;
1963 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1964 gfp_mask
, migratetype
);
1968 if (IS_ENABLED(CONFIG_NUMA
))
1969 zlc_mark_zone_full(zonelist
, z
);
1972 if (unlikely(IS_ENABLED(CONFIG_NUMA
) && page
== NULL
&& zlc_active
)) {
1973 /* Disable zlc cache for second zonelist scan */
1980 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
1981 * necessary to allocate the page. The expectation is
1982 * that the caller is taking steps that will free more
1983 * memory. The caller should avoid the page being used
1984 * for !PFMEMALLOC purposes.
1986 page
->pfmemalloc
= !!(alloc_flags
& ALLOC_NO_WATERMARKS
);
1992 * Large machines with many possible nodes should not always dump per-node
1993 * meminfo in irq context.
1995 static inline bool should_suppress_show_mem(void)
2000 ret
= in_interrupt();
2005 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2006 DEFAULT_RATELIMIT_INTERVAL
,
2007 DEFAULT_RATELIMIT_BURST
);
2009 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
2011 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2013 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2014 debug_guardpage_minorder() > 0)
2018 * This documents exceptions given to allocations in certain
2019 * contexts that are allowed to allocate outside current's set
2022 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2023 if (test_thread_flag(TIF_MEMDIE
) ||
2024 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2025 filter
&= ~SHOW_MEM_FILTER_NODES
;
2026 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
2027 filter
&= ~SHOW_MEM_FILTER_NODES
;
2030 struct va_format vaf
;
2033 va_start(args
, fmt
);
2038 pr_warn("%pV", &vaf
);
2043 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2044 current
->comm
, order
, gfp_mask
);
2047 if (!should_suppress_show_mem())
2052 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
2053 unsigned long did_some_progress
,
2054 unsigned long pages_reclaimed
)
2056 /* Do not loop if specifically requested */
2057 if (gfp_mask
& __GFP_NORETRY
)
2060 /* Always retry if specifically requested */
2061 if (gfp_mask
& __GFP_NOFAIL
)
2065 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2066 * making forward progress without invoking OOM. Suspend also disables
2067 * storage devices so kswapd will not help. Bail if we are suspending.
2069 if (!did_some_progress
&& pm_suspended_storage())
2073 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2074 * means __GFP_NOFAIL, but that may not be true in other
2077 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
2081 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2082 * specified, then we retry until we no longer reclaim any pages
2083 * (above), or we've reclaimed an order of pages at least as
2084 * large as the allocation's order. In both cases, if the
2085 * allocation still fails, we stop retrying.
2087 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
2093 static inline struct page
*
2094 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2095 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2096 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2101 /* Acquire the OOM killer lock for the zones in zonelist */
2102 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
2103 schedule_timeout_uninterruptible(1);
2108 * Go through the zonelist yet one more time, keep very high watermark
2109 * here, this is only to catch a parallel oom killing, we must fail if
2110 * we're still under heavy pressure.
2112 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
2113 order
, zonelist
, high_zoneidx
,
2114 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
2115 preferred_zone
, migratetype
);
2119 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2120 /* The OOM killer will not help higher order allocs */
2121 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2123 /* The OOM killer does not needlessly kill tasks for lowmem */
2124 if (high_zoneidx
< ZONE_NORMAL
)
2127 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2128 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2129 * The caller should handle page allocation failure by itself if
2130 * it specifies __GFP_THISNODE.
2131 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2133 if (gfp_mask
& __GFP_THISNODE
)
2136 /* Exhausted what can be done so it's blamo time */
2137 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
, false);
2140 clear_zonelist_oom(zonelist
, gfp_mask
);
2144 #ifdef CONFIG_COMPACTION
2145 /* Try memory compaction for high-order allocations before reclaim */
2146 static struct page
*
2147 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2148 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2149 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2150 int migratetype
, bool sync_migration
,
2151 bool *contended_compaction
, bool *deferred_compaction
,
2152 unsigned long *did_some_progress
)
2154 struct page
*page
= NULL
;
2159 if (compaction_deferred(preferred_zone
, order
)) {
2160 *deferred_compaction
= true;
2164 current
->flags
|= PF_MEMALLOC
;
2165 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
2166 nodemask
, sync_migration
,
2167 contended_compaction
, &page
);
2168 current
->flags
&= ~PF_MEMALLOC
;
2170 /* If compaction captured a page, prep and use it */
2172 prep_new_page(page
, order
, gfp_mask
);
2176 if (*did_some_progress
!= COMPACT_SKIPPED
) {
2177 /* Page migration frees to the PCP lists but we want merging */
2178 drain_pages(get_cpu());
2181 page
= get_page_from_freelist(gfp_mask
, nodemask
,
2182 order
, zonelist
, high_zoneidx
,
2183 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2184 preferred_zone
, migratetype
);
2187 preferred_zone
->compact_blockskip_flush
= false;
2188 preferred_zone
->compact_considered
= 0;
2189 preferred_zone
->compact_defer_shift
= 0;
2190 if (order
>= preferred_zone
->compact_order_failed
)
2191 preferred_zone
->compact_order_failed
= order
+ 1;
2192 count_vm_event(COMPACTSUCCESS
);
2197 * It's bad if compaction run occurs and fails.
2198 * The most likely reason is that pages exist,
2199 * but not enough to satisfy watermarks.
2201 count_vm_event(COMPACTFAIL
);
2204 * As async compaction considers a subset of pageblocks, only
2205 * defer if the failure was a sync compaction failure.
2208 defer_compaction(preferred_zone
, order
);
2216 static inline struct page
*
2217 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2218 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2219 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2220 int migratetype
, bool sync_migration
,
2221 bool *contended_compaction
, bool *deferred_compaction
,
2222 unsigned long *did_some_progress
)
2226 #endif /* CONFIG_COMPACTION */
2228 /* Perform direct synchronous page reclaim */
2230 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
, struct zonelist
*zonelist
,
2231 nodemask_t
*nodemask
)
2233 struct reclaim_state reclaim_state
;
2238 /* We now go into synchronous reclaim */
2239 cpuset_memory_pressure_bump();
2240 current
->flags
|= PF_MEMALLOC
;
2241 lockdep_set_current_reclaim_state(gfp_mask
);
2242 reclaim_state
.reclaimed_slab
= 0;
2243 current
->reclaim_state
= &reclaim_state
;
2245 progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
2247 current
->reclaim_state
= NULL
;
2248 lockdep_clear_current_reclaim_state();
2249 current
->flags
&= ~PF_MEMALLOC
;
2256 /* The really slow allocator path where we enter direct reclaim */
2257 static inline struct page
*
2258 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2259 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2260 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2261 int migratetype
, unsigned long *did_some_progress
)
2263 struct page
*page
= NULL
;
2264 bool drained
= false;
2266 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
2268 if (unlikely(!(*did_some_progress
)))
2271 /* After successful reclaim, reconsider all zones for allocation */
2272 if (IS_ENABLED(CONFIG_NUMA
))
2273 zlc_clear_zones_full(zonelist
);
2276 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2277 zonelist
, high_zoneidx
,
2278 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2279 preferred_zone
, migratetype
);
2282 * If an allocation failed after direct reclaim, it could be because
2283 * pages are pinned on the per-cpu lists. Drain them and try again
2285 if (!page
&& !drained
) {
2295 * This is called in the allocator slow-path if the allocation request is of
2296 * sufficient urgency to ignore watermarks and take other desperate measures
2298 static inline struct page
*
2299 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2300 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2301 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2307 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2308 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
2309 preferred_zone
, migratetype
);
2311 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2312 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2313 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2319 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
2320 enum zone_type high_zoneidx
,
2321 enum zone_type classzone_idx
)
2326 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
2327 wakeup_kswapd(zone
, order
, classzone_idx
);
2331 gfp_to_alloc_flags(gfp_t gfp_mask
)
2333 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2334 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2336 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2337 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2340 * The caller may dip into page reserves a bit more if the caller
2341 * cannot run direct reclaim, or if the caller has realtime scheduling
2342 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2343 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2345 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2349 * Not worth trying to allocate harder for
2350 * __GFP_NOMEMALLOC even if it can't schedule.
2352 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2353 alloc_flags
|= ALLOC_HARDER
;
2355 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2356 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2358 alloc_flags
&= ~ALLOC_CPUSET
;
2359 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2360 alloc_flags
|= ALLOC_HARDER
;
2362 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2363 if (gfp_mask
& __GFP_MEMALLOC
)
2364 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2365 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2366 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2367 else if (!in_interrupt() &&
2368 ((current
->flags
& PF_MEMALLOC
) ||
2369 unlikely(test_thread_flag(TIF_MEMDIE
))))
2370 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2373 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2374 alloc_flags
|= ALLOC_CMA
;
2379 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2381 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2384 static inline struct page
*
2385 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2386 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2387 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2390 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2391 struct page
*page
= NULL
;
2393 unsigned long pages_reclaimed
= 0;
2394 unsigned long did_some_progress
;
2395 bool sync_migration
= false;
2396 bool deferred_compaction
= false;
2397 bool contended_compaction
= false;
2400 * In the slowpath, we sanity check order to avoid ever trying to
2401 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2402 * be using allocators in order of preference for an area that is
2405 if (order
>= MAX_ORDER
) {
2406 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2411 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2412 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2413 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2414 * using a larger set of nodes after it has established that the
2415 * allowed per node queues are empty and that nodes are
2418 if (IS_ENABLED(CONFIG_NUMA
) &&
2419 (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2423 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2424 wake_all_kswapd(order
, zonelist
, high_zoneidx
,
2425 zone_idx(preferred_zone
));
2428 * OK, we're below the kswapd watermark and have kicked background
2429 * reclaim. Now things get more complex, so set up alloc_flags according
2430 * to how we want to proceed.
2432 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2435 * Find the true preferred zone if the allocation is unconstrained by
2438 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2439 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2443 /* This is the last chance, in general, before the goto nopage. */
2444 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2445 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2446 preferred_zone
, migratetype
);
2450 /* Allocate without watermarks if the context allows */
2451 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2453 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2454 * the allocation is high priority and these type of
2455 * allocations are system rather than user orientated
2457 zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
2459 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2460 zonelist
, high_zoneidx
, nodemask
,
2461 preferred_zone
, migratetype
);
2467 /* Atomic allocations - we can't balance anything */
2471 /* Avoid recursion of direct reclaim */
2472 if (current
->flags
& PF_MEMALLOC
)
2475 /* Avoid allocations with no watermarks from looping endlessly */
2476 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2480 * Try direct compaction. The first pass is asynchronous. Subsequent
2481 * attempts after direct reclaim are synchronous
2483 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2484 zonelist
, high_zoneidx
,
2486 alloc_flags
, preferred_zone
,
2487 migratetype
, sync_migration
,
2488 &contended_compaction
,
2489 &deferred_compaction
,
2490 &did_some_progress
);
2493 sync_migration
= true;
2496 * If compaction is deferred for high-order allocations, it is because
2497 * sync compaction recently failed. In this is the case and the caller
2498 * requested a movable allocation that does not heavily disrupt the
2499 * system then fail the allocation instead of entering direct reclaim.
2501 if ((deferred_compaction
|| contended_compaction
) &&
2502 (gfp_mask
& __GFP_NO_KSWAPD
))
2505 /* Try direct reclaim and then allocating */
2506 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2507 zonelist
, high_zoneidx
,
2509 alloc_flags
, preferred_zone
,
2510 migratetype
, &did_some_progress
);
2515 * If we failed to make any progress reclaiming, then we are
2516 * running out of options and have to consider going OOM
2518 if (!did_some_progress
) {
2519 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2520 if (oom_killer_disabled
)
2522 /* Coredumps can quickly deplete all memory reserves */
2523 if ((current
->flags
& PF_DUMPCORE
) &&
2524 !(gfp_mask
& __GFP_NOFAIL
))
2526 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2527 zonelist
, high_zoneidx
,
2528 nodemask
, preferred_zone
,
2533 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2535 * The oom killer is not called for high-order
2536 * allocations that may fail, so if no progress
2537 * is being made, there are no other options and
2538 * retrying is unlikely to help.
2540 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2543 * The oom killer is not called for lowmem
2544 * allocations to prevent needlessly killing
2547 if (high_zoneidx
< ZONE_NORMAL
)
2555 /* Check if we should retry the allocation */
2556 pages_reclaimed
+= did_some_progress
;
2557 if (should_alloc_retry(gfp_mask
, order
, did_some_progress
,
2559 /* Wait for some write requests to complete then retry */
2560 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2564 * High-order allocations do not necessarily loop after
2565 * direct reclaim and reclaim/compaction depends on compaction
2566 * being called after reclaim so call directly if necessary
2568 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2569 zonelist
, high_zoneidx
,
2571 alloc_flags
, preferred_zone
,
2572 migratetype
, sync_migration
,
2573 &contended_compaction
,
2574 &deferred_compaction
,
2575 &did_some_progress
);
2581 warn_alloc_failed(gfp_mask
, order
, NULL
);
2584 if (kmemcheck_enabled
)
2585 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2591 * This is the 'heart' of the zoned buddy allocator.
2594 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2595 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2597 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2598 struct zone
*preferred_zone
;
2599 struct page
*page
= NULL
;
2600 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2601 unsigned int cpuset_mems_cookie
;
2602 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
;
2604 gfp_mask
&= gfp_allowed_mask
;
2606 lockdep_trace_alloc(gfp_mask
);
2608 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2610 if (should_fail_alloc_page(gfp_mask
, order
))
2614 * Check the zones suitable for the gfp_mask contain at least one
2615 * valid zone. It's possible to have an empty zonelist as a result
2616 * of GFP_THISNODE and a memoryless node
2618 if (unlikely(!zonelist
->_zonerefs
->zone
))
2622 cpuset_mems_cookie
= get_mems_allowed();
2624 /* The preferred zone is used for statistics later */
2625 first_zones_zonelist(zonelist
, high_zoneidx
,
2626 nodemask
? : &cpuset_current_mems_allowed
,
2628 if (!preferred_zone
)
2632 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2633 alloc_flags
|= ALLOC_CMA
;
2635 /* First allocation attempt */
2636 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2637 zonelist
, high_zoneidx
, alloc_flags
,
2638 preferred_zone
, migratetype
);
2639 if (unlikely(!page
))
2640 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2641 zonelist
, high_zoneidx
, nodemask
,
2642 preferred_zone
, migratetype
);
2644 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2648 * When updating a task's mems_allowed, it is possible to race with
2649 * parallel threads in such a way that an allocation can fail while
2650 * the mask is being updated. If a page allocation is about to fail,
2651 * check if the cpuset changed during allocation and if so, retry.
2653 if (unlikely(!put_mems_allowed(cpuset_mems_cookie
) && !page
))
2658 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2661 * Common helper functions.
2663 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2668 * __get_free_pages() returns a 32-bit address, which cannot represent
2671 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2673 page
= alloc_pages(gfp_mask
, order
);
2676 return (unsigned long) page_address(page
);
2678 EXPORT_SYMBOL(__get_free_pages
);
2680 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2682 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2684 EXPORT_SYMBOL(get_zeroed_page
);
2686 void __free_pages(struct page
*page
, unsigned int order
)
2688 if (put_page_testzero(page
)) {
2690 free_hot_cold_page(page
, 0);
2692 __free_pages_ok(page
, order
);
2696 EXPORT_SYMBOL(__free_pages
);
2698 void free_pages(unsigned long addr
, unsigned int order
)
2701 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2702 __free_pages(virt_to_page((void *)addr
), order
);
2706 EXPORT_SYMBOL(free_pages
);
2708 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2711 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2712 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2714 split_page(virt_to_page((void *)addr
), order
);
2715 while (used
< alloc_end
) {
2720 return (void *)addr
;
2724 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2725 * @size: the number of bytes to allocate
2726 * @gfp_mask: GFP flags for the allocation
2728 * This function is similar to alloc_pages(), except that it allocates the
2729 * minimum number of pages to satisfy the request. alloc_pages() can only
2730 * allocate memory in power-of-two pages.
2732 * This function is also limited by MAX_ORDER.
2734 * Memory allocated by this function must be released by free_pages_exact().
2736 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2738 unsigned int order
= get_order(size
);
2741 addr
= __get_free_pages(gfp_mask
, order
);
2742 return make_alloc_exact(addr
, order
, size
);
2744 EXPORT_SYMBOL(alloc_pages_exact
);
2747 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2749 * @nid: the preferred node ID where memory should be allocated
2750 * @size: the number of bytes to allocate
2751 * @gfp_mask: GFP flags for the allocation
2753 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2755 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2758 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2760 unsigned order
= get_order(size
);
2761 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2764 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2766 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2769 * free_pages_exact - release memory allocated via alloc_pages_exact()
2770 * @virt: the value returned by alloc_pages_exact.
2771 * @size: size of allocation, same value as passed to alloc_pages_exact().
2773 * Release the memory allocated by a previous call to alloc_pages_exact.
2775 void free_pages_exact(void *virt
, size_t size
)
2777 unsigned long addr
= (unsigned long)virt
;
2778 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2780 while (addr
< end
) {
2785 EXPORT_SYMBOL(free_pages_exact
);
2787 static unsigned int nr_free_zone_pages(int offset
)
2792 /* Just pick one node, since fallback list is circular */
2793 unsigned int sum
= 0;
2795 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2797 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2798 unsigned long size
= zone
->present_pages
;
2799 unsigned long high
= high_wmark_pages(zone
);
2808 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2810 unsigned int nr_free_buffer_pages(void)
2812 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2814 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2817 * Amount of free RAM allocatable within all zones
2819 unsigned int nr_free_pagecache_pages(void)
2821 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2824 static inline void show_node(struct zone
*zone
)
2826 if (IS_ENABLED(CONFIG_NUMA
))
2827 printk("Node %d ", zone_to_nid(zone
));
2830 void si_meminfo(struct sysinfo
*val
)
2832 val
->totalram
= totalram_pages
;
2834 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2835 val
->bufferram
= nr_blockdev_pages();
2836 val
->totalhigh
= totalhigh_pages
;
2837 val
->freehigh
= nr_free_highpages();
2838 val
->mem_unit
= PAGE_SIZE
;
2841 EXPORT_SYMBOL(si_meminfo
);
2844 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2846 pg_data_t
*pgdat
= NODE_DATA(nid
);
2848 val
->totalram
= pgdat
->node_present_pages
;
2849 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2850 #ifdef CONFIG_HIGHMEM
2851 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2852 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2858 val
->mem_unit
= PAGE_SIZE
;
2863 * Determine whether the node should be displayed or not, depending on whether
2864 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2866 bool skip_free_areas_node(unsigned int flags
, int nid
)
2869 unsigned int cpuset_mems_cookie
;
2871 if (!(flags
& SHOW_MEM_FILTER_NODES
))
2875 cpuset_mems_cookie
= get_mems_allowed();
2876 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
2877 } while (!put_mems_allowed(cpuset_mems_cookie
));
2882 #define K(x) ((x) << (PAGE_SHIFT-10))
2884 static void show_migration_types(unsigned char type
)
2886 static const char types
[MIGRATE_TYPES
] = {
2887 [MIGRATE_UNMOVABLE
] = 'U',
2888 [MIGRATE_RECLAIMABLE
] = 'E',
2889 [MIGRATE_MOVABLE
] = 'M',
2890 [MIGRATE_RESERVE
] = 'R',
2892 [MIGRATE_CMA
] = 'C',
2894 [MIGRATE_ISOLATE
] = 'I',
2896 char tmp
[MIGRATE_TYPES
+ 1];
2900 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
2901 if (type
& (1 << i
))
2906 printk("(%s) ", tmp
);
2910 * Show free area list (used inside shift_scroll-lock stuff)
2911 * We also calculate the percentage fragmentation. We do this by counting the
2912 * memory on each free list with the exception of the first item on the list.
2913 * Suppresses nodes that are not allowed by current's cpuset if
2914 * SHOW_MEM_FILTER_NODES is passed.
2916 void show_free_areas(unsigned int filter
)
2921 for_each_populated_zone(zone
) {
2922 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2925 printk("%s per-cpu:\n", zone
->name
);
2927 for_each_online_cpu(cpu
) {
2928 struct per_cpu_pageset
*pageset
;
2930 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2932 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2933 cpu
, pageset
->pcp
.high
,
2934 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2938 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2939 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2941 " dirty:%lu writeback:%lu unstable:%lu\n"
2942 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2943 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
2945 global_page_state(NR_ACTIVE_ANON
),
2946 global_page_state(NR_INACTIVE_ANON
),
2947 global_page_state(NR_ISOLATED_ANON
),
2948 global_page_state(NR_ACTIVE_FILE
),
2949 global_page_state(NR_INACTIVE_FILE
),
2950 global_page_state(NR_ISOLATED_FILE
),
2951 global_page_state(NR_UNEVICTABLE
),
2952 global_page_state(NR_FILE_DIRTY
),
2953 global_page_state(NR_WRITEBACK
),
2954 global_page_state(NR_UNSTABLE_NFS
),
2955 global_page_state(NR_FREE_PAGES
),
2956 global_page_state(NR_SLAB_RECLAIMABLE
),
2957 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2958 global_page_state(NR_FILE_MAPPED
),
2959 global_page_state(NR_SHMEM
),
2960 global_page_state(NR_PAGETABLE
),
2961 global_page_state(NR_BOUNCE
),
2962 global_page_state(NR_FREE_CMA_PAGES
));
2964 for_each_populated_zone(zone
) {
2967 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2975 " active_anon:%lukB"
2976 " inactive_anon:%lukB"
2977 " active_file:%lukB"
2978 " inactive_file:%lukB"
2979 " unevictable:%lukB"
2980 " isolated(anon):%lukB"
2981 " isolated(file):%lukB"
2988 " slab_reclaimable:%lukB"
2989 " slab_unreclaimable:%lukB"
2990 " kernel_stack:%lukB"
2995 " writeback_tmp:%lukB"
2996 " pages_scanned:%lu"
2997 " all_unreclaimable? %s"
3000 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3001 K(min_wmark_pages(zone
)),
3002 K(low_wmark_pages(zone
)),
3003 K(high_wmark_pages(zone
)),
3004 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3005 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3006 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3007 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3008 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3009 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3010 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3011 K(zone
->present_pages
),
3012 K(zone_page_state(zone
, NR_MLOCK
)),
3013 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3014 K(zone_page_state(zone
, NR_WRITEBACK
)),
3015 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3016 K(zone_page_state(zone
, NR_SHMEM
)),
3017 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3018 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3019 zone_page_state(zone
, NR_KERNEL_STACK
) *
3021 K(zone_page_state(zone
, NR_PAGETABLE
)),
3022 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3023 K(zone_page_state(zone
, NR_BOUNCE
)),
3024 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3025 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3026 zone
->pages_scanned
,
3027 (zone
->all_unreclaimable
? "yes" : "no")
3029 printk("lowmem_reserve[]:");
3030 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3031 printk(" %lu", zone
->lowmem_reserve
[i
]);
3035 for_each_populated_zone(zone
) {
3036 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
3037 unsigned char types
[MAX_ORDER
];
3039 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3042 printk("%s: ", zone
->name
);
3044 spin_lock_irqsave(&zone
->lock
, flags
);
3045 for (order
= 0; order
< MAX_ORDER
; order
++) {
3046 struct free_area
*area
= &zone
->free_area
[order
];
3049 nr
[order
] = area
->nr_free
;
3050 total
+= nr
[order
] << order
;
3053 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3054 if (!list_empty(&area
->free_list
[type
]))
3055 types
[order
] |= 1 << type
;
3058 spin_unlock_irqrestore(&zone
->lock
, flags
);
3059 for (order
= 0; order
< MAX_ORDER
; order
++) {
3060 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3062 show_migration_types(types
[order
]);
3064 printk("= %lukB\n", K(total
));
3067 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3069 show_swap_cache_info();
3072 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3074 zoneref
->zone
= zone
;
3075 zoneref
->zone_idx
= zone_idx(zone
);
3079 * Builds allocation fallback zone lists.
3081 * Add all populated zones of a node to the zonelist.
3083 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3084 int nr_zones
, enum zone_type zone_type
)
3088 BUG_ON(zone_type
>= MAX_NR_ZONES
);
3093 zone
= pgdat
->node_zones
+ zone_type
;
3094 if (populated_zone(zone
)) {
3095 zoneref_set_zone(zone
,
3096 &zonelist
->_zonerefs
[nr_zones
++]);
3097 check_highest_zone(zone_type
);
3100 } while (zone_type
);
3107 * 0 = automatic detection of better ordering.
3108 * 1 = order by ([node] distance, -zonetype)
3109 * 2 = order by (-zonetype, [node] distance)
3111 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3112 * the same zonelist. So only NUMA can configure this param.
3114 #define ZONELIST_ORDER_DEFAULT 0
3115 #define ZONELIST_ORDER_NODE 1
3116 #define ZONELIST_ORDER_ZONE 2
3118 /* zonelist order in the kernel.
3119 * set_zonelist_order() will set this to NODE or ZONE.
3121 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3122 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3126 /* The value user specified ....changed by config */
3127 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3128 /* string for sysctl */
3129 #define NUMA_ZONELIST_ORDER_LEN 16
3130 char numa_zonelist_order
[16] = "default";
3133 * interface for configure zonelist ordering.
3134 * command line option "numa_zonelist_order"
3135 * = "[dD]efault - default, automatic configuration.
3136 * = "[nN]ode - order by node locality, then by zone within node
3137 * = "[zZ]one - order by zone, then by locality within zone
3140 static int __parse_numa_zonelist_order(char *s
)
3142 if (*s
== 'd' || *s
== 'D') {
3143 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3144 } else if (*s
== 'n' || *s
== 'N') {
3145 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3146 } else if (*s
== 'z' || *s
== 'Z') {
3147 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3150 "Ignoring invalid numa_zonelist_order value: "
3157 static __init
int setup_numa_zonelist_order(char *s
)
3164 ret
= __parse_numa_zonelist_order(s
);
3166 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3170 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3173 * sysctl handler for numa_zonelist_order
3175 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
3176 void __user
*buffer
, size_t *length
,
3179 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3181 static DEFINE_MUTEX(zl_order_mutex
);
3183 mutex_lock(&zl_order_mutex
);
3185 strcpy(saved_string
, (char*)table
->data
);
3186 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3190 int oldval
= user_zonelist_order
;
3191 if (__parse_numa_zonelist_order((char*)table
->data
)) {
3193 * bogus value. restore saved string
3195 strncpy((char*)table
->data
, saved_string
,
3196 NUMA_ZONELIST_ORDER_LEN
);
3197 user_zonelist_order
= oldval
;
3198 } else if (oldval
!= user_zonelist_order
) {
3199 mutex_lock(&zonelists_mutex
);
3200 build_all_zonelists(NULL
, NULL
);
3201 mutex_unlock(&zonelists_mutex
);
3205 mutex_unlock(&zl_order_mutex
);
3210 #define MAX_NODE_LOAD (nr_online_nodes)
3211 static int node_load
[MAX_NUMNODES
];
3214 * find_next_best_node - find the next node that should appear in a given node's fallback list
3215 * @node: node whose fallback list we're appending
3216 * @used_node_mask: nodemask_t of already used nodes
3218 * We use a number of factors to determine which is the next node that should
3219 * appear on a given node's fallback list. The node should not have appeared
3220 * already in @node's fallback list, and it should be the next closest node
3221 * according to the distance array (which contains arbitrary distance values
3222 * from each node to each node in the system), and should also prefer nodes
3223 * with no CPUs, since presumably they'll have very little allocation pressure
3224 * on them otherwise.
3225 * It returns -1 if no node is found.
3227 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3230 int min_val
= INT_MAX
;
3232 const struct cpumask
*tmp
= cpumask_of_node(0);
3234 /* Use the local node if we haven't already */
3235 if (!node_isset(node
, *used_node_mask
)) {
3236 node_set(node
, *used_node_mask
);
3240 for_each_node_state(n
, N_HIGH_MEMORY
) {
3242 /* Don't want a node to appear more than once */
3243 if (node_isset(n
, *used_node_mask
))
3246 /* Use the distance array to find the distance */
3247 val
= node_distance(node
, n
);
3249 /* Penalize nodes under us ("prefer the next node") */
3252 /* Give preference to headless and unused nodes */
3253 tmp
= cpumask_of_node(n
);
3254 if (!cpumask_empty(tmp
))
3255 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
3257 /* Slight preference for less loaded node */
3258 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
3259 val
+= node_load
[n
];
3261 if (val
< min_val
) {
3268 node_set(best_node
, *used_node_mask
);
3275 * Build zonelists ordered by node and zones within node.
3276 * This results in maximum locality--normal zone overflows into local
3277 * DMA zone, if any--but risks exhausting DMA zone.
3279 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
3282 struct zonelist
*zonelist
;
3284 zonelist
= &pgdat
->node_zonelists
[0];
3285 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
3287 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3289 zonelist
->_zonerefs
[j
].zone
= NULL
;
3290 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3294 * Build gfp_thisnode zonelists
3296 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
3299 struct zonelist
*zonelist
;
3301 zonelist
= &pgdat
->node_zonelists
[1];
3302 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3303 zonelist
->_zonerefs
[j
].zone
= NULL
;
3304 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3308 * Build zonelists ordered by zone and nodes within zones.
3309 * This results in conserving DMA zone[s] until all Normal memory is
3310 * exhausted, but results in overflowing to remote node while memory
3311 * may still exist in local DMA zone.
3313 static int node_order
[MAX_NUMNODES
];
3315 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
3318 int zone_type
; /* needs to be signed */
3320 struct zonelist
*zonelist
;
3322 zonelist
= &pgdat
->node_zonelists
[0];
3324 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
3325 for (j
= 0; j
< nr_nodes
; j
++) {
3326 node
= node_order
[j
];
3327 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
3328 if (populated_zone(z
)) {
3330 &zonelist
->_zonerefs
[pos
++]);
3331 check_highest_zone(zone_type
);
3335 zonelist
->_zonerefs
[pos
].zone
= NULL
;
3336 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
3339 static int default_zonelist_order(void)
3342 unsigned long low_kmem_size
,total_size
;
3346 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3347 * If they are really small and used heavily, the system can fall
3348 * into OOM very easily.
3349 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3351 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3354 for_each_online_node(nid
) {
3355 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3356 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3357 if (populated_zone(z
)) {
3358 if (zone_type
< ZONE_NORMAL
)
3359 low_kmem_size
+= z
->present_pages
;
3360 total_size
+= z
->present_pages
;
3361 } else if (zone_type
== ZONE_NORMAL
) {
3363 * If any node has only lowmem, then node order
3364 * is preferred to allow kernel allocations
3365 * locally; otherwise, they can easily infringe
3366 * on other nodes when there is an abundance of
3367 * lowmem available to allocate from.
3369 return ZONELIST_ORDER_NODE
;
3373 if (!low_kmem_size
|| /* there are no DMA area. */
3374 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
3375 return ZONELIST_ORDER_NODE
;
3377 * look into each node's config.
3378 * If there is a node whose DMA/DMA32 memory is very big area on
3379 * local memory, NODE_ORDER may be suitable.
3381 average_size
= total_size
/
3382 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
3383 for_each_online_node(nid
) {
3386 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3387 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3388 if (populated_zone(z
)) {
3389 if (zone_type
< ZONE_NORMAL
)
3390 low_kmem_size
+= z
->present_pages
;
3391 total_size
+= z
->present_pages
;
3394 if (low_kmem_size
&&
3395 total_size
> average_size
&& /* ignore small node */
3396 low_kmem_size
> total_size
* 70/100)
3397 return ZONELIST_ORDER_NODE
;
3399 return ZONELIST_ORDER_ZONE
;
3402 static void set_zonelist_order(void)
3404 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
3405 current_zonelist_order
= default_zonelist_order();
3407 current_zonelist_order
= user_zonelist_order
;
3410 static void build_zonelists(pg_data_t
*pgdat
)
3414 nodemask_t used_mask
;
3415 int local_node
, prev_node
;
3416 struct zonelist
*zonelist
;
3417 int order
= current_zonelist_order
;
3419 /* initialize zonelists */
3420 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3421 zonelist
= pgdat
->node_zonelists
+ i
;
3422 zonelist
->_zonerefs
[0].zone
= NULL
;
3423 zonelist
->_zonerefs
[0].zone_idx
= 0;
3426 /* NUMA-aware ordering of nodes */
3427 local_node
= pgdat
->node_id
;
3428 load
= nr_online_nodes
;
3429 prev_node
= local_node
;
3430 nodes_clear(used_mask
);
3432 memset(node_order
, 0, sizeof(node_order
));
3435 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3437 * We don't want to pressure a particular node.
3438 * So adding penalty to the first node in same
3439 * distance group to make it round-robin.
3441 if (node_distance(local_node
, node
) !=
3442 node_distance(local_node
, prev_node
))
3443 node_load
[node
] = load
;
3447 if (order
== ZONELIST_ORDER_NODE
)
3448 build_zonelists_in_node_order(pgdat
, node
);
3450 node_order
[j
++] = node
; /* remember order */
3453 if (order
== ZONELIST_ORDER_ZONE
) {
3454 /* calculate node order -- i.e., DMA last! */
3455 build_zonelists_in_zone_order(pgdat
, j
);
3458 build_thisnode_zonelists(pgdat
);
3461 /* Construct the zonelist performance cache - see further mmzone.h */
3462 static void build_zonelist_cache(pg_data_t
*pgdat
)
3464 struct zonelist
*zonelist
;
3465 struct zonelist_cache
*zlc
;
3468 zonelist
= &pgdat
->node_zonelists
[0];
3469 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3470 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3471 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3472 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3475 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3477 * Return node id of node used for "local" allocations.
3478 * I.e., first node id of first zone in arg node's generic zonelist.
3479 * Used for initializing percpu 'numa_mem', which is used primarily
3480 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3482 int local_memory_node(int node
)
3486 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3487 gfp_zone(GFP_KERNEL
),
3494 #else /* CONFIG_NUMA */
3496 static void set_zonelist_order(void)
3498 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3501 static void build_zonelists(pg_data_t
*pgdat
)
3503 int node
, local_node
;
3505 struct zonelist
*zonelist
;
3507 local_node
= pgdat
->node_id
;
3509 zonelist
= &pgdat
->node_zonelists
[0];
3510 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3513 * Now we build the zonelist so that it contains the zones
3514 * of all the other nodes.
3515 * We don't want to pressure a particular node, so when
3516 * building the zones for node N, we make sure that the
3517 * zones coming right after the local ones are those from
3518 * node N+1 (modulo N)
3520 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3521 if (!node_online(node
))
3523 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3526 for (node
= 0; node
< local_node
; node
++) {
3527 if (!node_online(node
))
3529 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3533 zonelist
->_zonerefs
[j
].zone
= NULL
;
3534 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3537 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3538 static void build_zonelist_cache(pg_data_t
*pgdat
)
3540 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3543 #endif /* CONFIG_NUMA */
3546 * Boot pageset table. One per cpu which is going to be used for all
3547 * zones and all nodes. The parameters will be set in such a way
3548 * that an item put on a list will immediately be handed over to
3549 * the buddy list. This is safe since pageset manipulation is done
3550 * with interrupts disabled.
3552 * The boot_pagesets must be kept even after bootup is complete for
3553 * unused processors and/or zones. They do play a role for bootstrapping
3554 * hotplugged processors.
3556 * zoneinfo_show() and maybe other functions do
3557 * not check if the processor is online before following the pageset pointer.
3558 * Other parts of the kernel may not check if the zone is available.
3560 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3561 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3562 static void setup_zone_pageset(struct zone
*zone
);
3565 * Global mutex to protect against size modification of zonelists
3566 * as well as to serialize pageset setup for the new populated zone.
3568 DEFINE_MUTEX(zonelists_mutex
);
3570 /* return values int ....just for stop_machine() */
3571 static int __build_all_zonelists(void *data
)
3575 pg_data_t
*self
= data
;
3578 memset(node_load
, 0, sizeof(node_load
));
3581 if (self
&& !node_online(self
->node_id
)) {
3582 build_zonelists(self
);
3583 build_zonelist_cache(self
);
3586 for_each_online_node(nid
) {
3587 pg_data_t
*pgdat
= NODE_DATA(nid
);
3589 build_zonelists(pgdat
);
3590 build_zonelist_cache(pgdat
);
3594 * Initialize the boot_pagesets that are going to be used
3595 * for bootstrapping processors. The real pagesets for
3596 * each zone will be allocated later when the per cpu
3597 * allocator is available.
3599 * boot_pagesets are used also for bootstrapping offline
3600 * cpus if the system is already booted because the pagesets
3601 * are needed to initialize allocators on a specific cpu too.
3602 * F.e. the percpu allocator needs the page allocator which
3603 * needs the percpu allocator in order to allocate its pagesets
3604 * (a chicken-egg dilemma).
3606 for_each_possible_cpu(cpu
) {
3607 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3609 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3611 * We now know the "local memory node" for each node--
3612 * i.e., the node of the first zone in the generic zonelist.
3613 * Set up numa_mem percpu variable for on-line cpus. During
3614 * boot, only the boot cpu should be on-line; we'll init the
3615 * secondary cpus' numa_mem as they come on-line. During
3616 * node/memory hotplug, we'll fixup all on-line cpus.
3618 if (cpu_online(cpu
))
3619 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3627 * Called with zonelists_mutex held always
3628 * unless system_state == SYSTEM_BOOTING.
3630 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
3632 set_zonelist_order();
3634 if (system_state
== SYSTEM_BOOTING
) {
3635 __build_all_zonelists(NULL
);
3636 mminit_verify_zonelist();
3637 cpuset_init_current_mems_allowed();
3639 /* we have to stop all cpus to guarantee there is no user
3641 #ifdef CONFIG_MEMORY_HOTPLUG
3643 setup_zone_pageset(zone
);
3645 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
3646 /* cpuset refresh routine should be here */
3648 vm_total_pages
= nr_free_pagecache_pages();
3650 * Disable grouping by mobility if the number of pages in the
3651 * system is too low to allow the mechanism to work. It would be
3652 * more accurate, but expensive to check per-zone. This check is
3653 * made on memory-hotadd so a system can start with mobility
3654 * disabled and enable it later
3656 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3657 page_group_by_mobility_disabled
= 1;
3659 page_group_by_mobility_disabled
= 0;
3661 printk("Built %i zonelists in %s order, mobility grouping %s. "
3662 "Total pages: %ld\n",
3664 zonelist_order_name
[current_zonelist_order
],
3665 page_group_by_mobility_disabled
? "off" : "on",
3668 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3673 * Helper functions to size the waitqueue hash table.
3674 * Essentially these want to choose hash table sizes sufficiently
3675 * large so that collisions trying to wait on pages are rare.
3676 * But in fact, the number of active page waitqueues on typical
3677 * systems is ridiculously low, less than 200. So this is even
3678 * conservative, even though it seems large.
3680 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3681 * waitqueues, i.e. the size of the waitq table given the number of pages.
3683 #define PAGES_PER_WAITQUEUE 256
3685 #ifndef CONFIG_MEMORY_HOTPLUG
3686 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3688 unsigned long size
= 1;
3690 pages
/= PAGES_PER_WAITQUEUE
;
3692 while (size
< pages
)
3696 * Once we have dozens or even hundreds of threads sleeping
3697 * on IO we've got bigger problems than wait queue collision.
3698 * Limit the size of the wait table to a reasonable size.
3700 size
= min(size
, 4096UL);
3702 return max(size
, 4UL);
3706 * A zone's size might be changed by hot-add, so it is not possible to determine
3707 * a suitable size for its wait_table. So we use the maximum size now.
3709 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3711 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3712 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3713 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3715 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3716 * or more by the traditional way. (See above). It equals:
3718 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3719 * ia64(16K page size) : = ( 8G + 4M)byte.
3720 * powerpc (64K page size) : = (32G +16M)byte.
3722 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3729 * This is an integer logarithm so that shifts can be used later
3730 * to extract the more random high bits from the multiplicative
3731 * hash function before the remainder is taken.
3733 static inline unsigned long wait_table_bits(unsigned long size
)
3738 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3741 * Check if a pageblock contains reserved pages
3743 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3747 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3748 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3755 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3756 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3757 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3758 * higher will lead to a bigger reserve which will get freed as contiguous
3759 * blocks as reclaim kicks in
3761 static void setup_zone_migrate_reserve(struct zone
*zone
)
3763 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3765 unsigned long block_migratetype
;
3769 * Get the start pfn, end pfn and the number of blocks to reserve
3770 * We have to be careful to be aligned to pageblock_nr_pages to
3771 * make sure that we always check pfn_valid for the first page in
3774 start_pfn
= zone
->zone_start_pfn
;
3775 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3776 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3777 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3781 * Reserve blocks are generally in place to help high-order atomic
3782 * allocations that are short-lived. A min_free_kbytes value that
3783 * would result in more than 2 reserve blocks for atomic allocations
3784 * is assumed to be in place to help anti-fragmentation for the
3785 * future allocation of hugepages at runtime.
3787 reserve
= min(2, reserve
);
3789 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3790 if (!pfn_valid(pfn
))
3792 page
= pfn_to_page(pfn
);
3794 /* Watch out for overlapping nodes */
3795 if (page_to_nid(page
) != zone_to_nid(zone
))
3798 block_migratetype
= get_pageblock_migratetype(page
);
3800 /* Only test what is necessary when the reserves are not met */
3803 * Blocks with reserved pages will never free, skip
3806 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3807 if (pageblock_is_reserved(pfn
, block_end_pfn
))
3810 /* If this block is reserved, account for it */
3811 if (block_migratetype
== MIGRATE_RESERVE
) {
3816 /* Suitable for reserving if this block is movable */
3817 if (block_migratetype
== MIGRATE_MOVABLE
) {
3818 set_pageblock_migratetype(page
,
3820 move_freepages_block(zone
, page
,
3828 * If the reserve is met and this is a previous reserved block,
3831 if (block_migratetype
== MIGRATE_RESERVE
) {
3832 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3833 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3839 * Initially all pages are reserved - free ones are freed
3840 * up by free_all_bootmem() once the early boot process is
3841 * done. Non-atomic initialization, single-pass.
3843 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3844 unsigned long start_pfn
, enum memmap_context context
)
3847 unsigned long end_pfn
= start_pfn
+ size
;
3851 if (highest_memmap_pfn
< end_pfn
- 1)
3852 highest_memmap_pfn
= end_pfn
- 1;
3854 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3855 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3857 * There can be holes in boot-time mem_map[]s
3858 * handed to this function. They do not
3859 * exist on hotplugged memory.
3861 if (context
== MEMMAP_EARLY
) {
3862 if (!early_pfn_valid(pfn
))
3864 if (!early_pfn_in_nid(pfn
, nid
))
3867 page
= pfn_to_page(pfn
);
3868 set_page_links(page
, zone
, nid
, pfn
);
3869 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3870 init_page_count(page
);
3871 reset_page_mapcount(page
);
3872 SetPageReserved(page
);
3874 * Mark the block movable so that blocks are reserved for
3875 * movable at startup. This will force kernel allocations
3876 * to reserve their blocks rather than leaking throughout
3877 * the address space during boot when many long-lived
3878 * kernel allocations are made. Later some blocks near
3879 * the start are marked MIGRATE_RESERVE by
3880 * setup_zone_migrate_reserve()
3882 * bitmap is created for zone's valid pfn range. but memmap
3883 * can be created for invalid pages (for alignment)
3884 * check here not to call set_pageblock_migratetype() against
3887 if ((z
->zone_start_pfn
<= pfn
)
3888 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3889 && !(pfn
& (pageblock_nr_pages
- 1)))
3890 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3892 INIT_LIST_HEAD(&page
->lru
);
3893 #ifdef WANT_PAGE_VIRTUAL
3894 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3895 if (!is_highmem_idx(zone
))
3896 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3901 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3904 for_each_migratetype_order(order
, t
) {
3905 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3906 zone
->free_area
[order
].nr_free
= 0;
3910 #ifndef __HAVE_ARCH_MEMMAP_INIT
3911 #define memmap_init(size, nid, zone, start_pfn) \
3912 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3915 static int __meminit
zone_batchsize(struct zone
*zone
)
3921 * The per-cpu-pages pools are set to around 1000th of the
3922 * size of the zone. But no more than 1/2 of a meg.
3924 * OK, so we don't know how big the cache is. So guess.
3926 batch
= zone
->present_pages
/ 1024;
3927 if (batch
* PAGE_SIZE
> 512 * 1024)
3928 batch
= (512 * 1024) / PAGE_SIZE
;
3929 batch
/= 4; /* We effectively *= 4 below */
3934 * Clamp the batch to a 2^n - 1 value. Having a power
3935 * of 2 value was found to be more likely to have
3936 * suboptimal cache aliasing properties in some cases.
3938 * For example if 2 tasks are alternately allocating
3939 * batches of pages, one task can end up with a lot
3940 * of pages of one half of the possible page colors
3941 * and the other with pages of the other colors.
3943 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3948 /* The deferral and batching of frees should be suppressed under NOMMU
3951 * The problem is that NOMMU needs to be able to allocate large chunks
3952 * of contiguous memory as there's no hardware page translation to
3953 * assemble apparent contiguous memory from discontiguous pages.
3955 * Queueing large contiguous runs of pages for batching, however,
3956 * causes the pages to actually be freed in smaller chunks. As there
3957 * can be a significant delay between the individual batches being
3958 * recycled, this leads to the once large chunks of space being
3959 * fragmented and becoming unavailable for high-order allocations.
3965 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3967 struct per_cpu_pages
*pcp
;
3970 memset(p
, 0, sizeof(*p
));
3974 pcp
->high
= 6 * batch
;
3975 pcp
->batch
= max(1UL, 1 * batch
);
3976 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3977 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3981 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3982 * to the value high for the pageset p.
3985 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3988 struct per_cpu_pages
*pcp
;
3992 pcp
->batch
= max(1UL, high
/4);
3993 if ((high
/4) > (PAGE_SHIFT
* 8))
3994 pcp
->batch
= PAGE_SHIFT
* 8;
3997 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4001 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4003 for_each_possible_cpu(cpu
) {
4004 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4006 setup_pageset(pcp
, zone_batchsize(zone
));
4008 if (percpu_pagelist_fraction
)
4009 setup_pagelist_highmark(pcp
,
4010 (zone
->present_pages
/
4011 percpu_pagelist_fraction
));
4016 * Allocate per cpu pagesets and initialize them.
4017 * Before this call only boot pagesets were available.
4019 void __init
setup_per_cpu_pageset(void)
4023 for_each_populated_zone(zone
)
4024 setup_zone_pageset(zone
);
4027 static noinline __init_refok
4028 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4031 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4035 * The per-page waitqueue mechanism uses hashed waitqueues
4038 zone
->wait_table_hash_nr_entries
=
4039 wait_table_hash_nr_entries(zone_size_pages
);
4040 zone
->wait_table_bits
=
4041 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4042 alloc_size
= zone
->wait_table_hash_nr_entries
4043 * sizeof(wait_queue_head_t
);
4045 if (!slab_is_available()) {
4046 zone
->wait_table
= (wait_queue_head_t
*)
4047 alloc_bootmem_node_nopanic(pgdat
, alloc_size
);
4050 * This case means that a zone whose size was 0 gets new memory
4051 * via memory hot-add.
4052 * But it may be the case that a new node was hot-added. In
4053 * this case vmalloc() will not be able to use this new node's
4054 * memory - this wait_table must be initialized to use this new
4055 * node itself as well.
4056 * To use this new node's memory, further consideration will be
4059 zone
->wait_table
= vmalloc(alloc_size
);
4061 if (!zone
->wait_table
)
4064 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4065 init_waitqueue_head(zone
->wait_table
+ i
);
4070 static __meminit
void zone_pcp_init(struct zone
*zone
)
4073 * per cpu subsystem is not up at this point. The following code
4074 * relies on the ability of the linker to provide the
4075 * offset of a (static) per cpu variable into the per cpu area.
4077 zone
->pageset
= &boot_pageset
;
4079 if (zone
->present_pages
)
4080 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4081 zone
->name
, zone
->present_pages
,
4082 zone_batchsize(zone
));
4085 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4086 unsigned long zone_start_pfn
,
4088 enum memmap_context context
)
4090 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4092 ret
= zone_wait_table_init(zone
, size
);
4095 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4097 zone
->zone_start_pfn
= zone_start_pfn
;
4099 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4100 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4102 (unsigned long)zone_idx(zone
),
4103 zone_start_pfn
, (zone_start_pfn
+ size
));
4105 zone_init_free_lists(zone
);
4110 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4111 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4113 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4114 * Architectures may implement their own version but if add_active_range()
4115 * was used and there are no special requirements, this is a convenient
4118 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
4120 unsigned long start_pfn
, end_pfn
;
4123 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4124 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
4126 /* This is a memory hole */
4129 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4131 int __meminit
early_pfn_to_nid(unsigned long pfn
)
4135 nid
= __early_pfn_to_nid(pfn
);
4138 /* just returns 0 */
4142 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4143 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
4147 nid
= __early_pfn_to_nid(pfn
);
4148 if (nid
>= 0 && nid
!= node
)
4155 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
4156 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4157 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
4159 * If an architecture guarantees that all ranges registered with
4160 * add_active_ranges() contain no holes and may be freed, this
4161 * this function may be used instead of calling free_bootmem() manually.
4163 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4165 unsigned long start_pfn
, end_pfn
;
4168 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4169 start_pfn
= min(start_pfn
, max_low_pfn
);
4170 end_pfn
= min(end_pfn
, max_low_pfn
);
4172 if (start_pfn
< end_pfn
)
4173 free_bootmem_node(NODE_DATA(this_nid
),
4174 PFN_PHYS(start_pfn
),
4175 (end_pfn
- start_pfn
) << PAGE_SHIFT
);
4180 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4181 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4183 * If an architecture guarantees that all ranges registered with
4184 * add_active_ranges() contain no holes and may be freed, this
4185 * function may be used instead of calling memory_present() manually.
4187 void __init
sparse_memory_present_with_active_regions(int nid
)
4189 unsigned long start_pfn
, end_pfn
;
4192 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4193 memory_present(this_nid
, start_pfn
, end_pfn
);
4197 * get_pfn_range_for_nid - Return the start and end page frames for a node
4198 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4199 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4200 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4202 * It returns the start and end page frame of a node based on information
4203 * provided by an arch calling add_active_range(). If called for a node
4204 * with no available memory, a warning is printed and the start and end
4207 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4208 unsigned long *start_pfn
, unsigned long *end_pfn
)
4210 unsigned long this_start_pfn
, this_end_pfn
;
4216 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4217 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4218 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4221 if (*start_pfn
== -1UL)
4226 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4227 * assumption is made that zones within a node are ordered in monotonic
4228 * increasing memory addresses so that the "highest" populated zone is used
4230 static void __init
find_usable_zone_for_movable(void)
4233 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4234 if (zone_index
== ZONE_MOVABLE
)
4237 if (arch_zone_highest_possible_pfn
[zone_index
] >
4238 arch_zone_lowest_possible_pfn
[zone_index
])
4242 VM_BUG_ON(zone_index
== -1);
4243 movable_zone
= zone_index
;
4247 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4248 * because it is sized independent of architecture. Unlike the other zones,
4249 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4250 * in each node depending on the size of each node and how evenly kernelcore
4251 * is distributed. This helper function adjusts the zone ranges
4252 * provided by the architecture for a given node by using the end of the
4253 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4254 * zones within a node are in order of monotonic increases memory addresses
4256 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4257 unsigned long zone_type
,
4258 unsigned long node_start_pfn
,
4259 unsigned long node_end_pfn
,
4260 unsigned long *zone_start_pfn
,
4261 unsigned long *zone_end_pfn
)
4263 /* Only adjust if ZONE_MOVABLE is on this node */
4264 if (zone_movable_pfn
[nid
]) {
4265 /* Size ZONE_MOVABLE */
4266 if (zone_type
== ZONE_MOVABLE
) {
4267 *zone_start_pfn
= zone_movable_pfn
[nid
];
4268 *zone_end_pfn
= min(node_end_pfn
,
4269 arch_zone_highest_possible_pfn
[movable_zone
]);
4271 /* Adjust for ZONE_MOVABLE starting within this range */
4272 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4273 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4274 *zone_end_pfn
= zone_movable_pfn
[nid
];
4276 /* Check if this whole range is within ZONE_MOVABLE */
4277 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4278 *zone_start_pfn
= *zone_end_pfn
;
4283 * Return the number of pages a zone spans in a node, including holes
4284 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4286 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4287 unsigned long zone_type
,
4288 unsigned long *ignored
)
4290 unsigned long node_start_pfn
, node_end_pfn
;
4291 unsigned long zone_start_pfn
, zone_end_pfn
;
4293 /* Get the start and end of the node and zone */
4294 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4295 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4296 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4297 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4298 node_start_pfn
, node_end_pfn
,
4299 &zone_start_pfn
, &zone_end_pfn
);
4301 /* Check that this node has pages within the zone's required range */
4302 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4305 /* Move the zone boundaries inside the node if necessary */
4306 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4307 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4309 /* Return the spanned pages */
4310 return zone_end_pfn
- zone_start_pfn
;
4314 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4315 * then all holes in the requested range will be accounted for.
4317 unsigned long __meminit
__absent_pages_in_range(int nid
,
4318 unsigned long range_start_pfn
,
4319 unsigned long range_end_pfn
)
4321 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4322 unsigned long start_pfn
, end_pfn
;
4325 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4326 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4327 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4328 nr_absent
-= end_pfn
- start_pfn
;
4334 * absent_pages_in_range - Return number of page frames in holes within a range
4335 * @start_pfn: The start PFN to start searching for holes
4336 * @end_pfn: The end PFN to stop searching for holes
4338 * It returns the number of pages frames in memory holes within a range.
4340 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4341 unsigned long end_pfn
)
4343 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4346 /* Return the number of page frames in holes in a zone on a node */
4347 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4348 unsigned long zone_type
,
4349 unsigned long *ignored
)
4351 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4352 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4353 unsigned long node_start_pfn
, node_end_pfn
;
4354 unsigned long zone_start_pfn
, zone_end_pfn
;
4356 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4357 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4358 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4360 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4361 node_start_pfn
, node_end_pfn
,
4362 &zone_start_pfn
, &zone_end_pfn
);
4363 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4366 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4367 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4368 unsigned long zone_type
,
4369 unsigned long *zones_size
)
4371 return zones_size
[zone_type
];
4374 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4375 unsigned long zone_type
,
4376 unsigned long *zholes_size
)
4381 return zholes_size
[zone_type
];
4384 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4386 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4387 unsigned long *zones_size
, unsigned long *zholes_size
)
4389 unsigned long realtotalpages
, totalpages
= 0;
4392 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4393 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4395 pgdat
->node_spanned_pages
= totalpages
;
4397 realtotalpages
= totalpages
;
4398 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4400 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4402 pgdat
->node_present_pages
= realtotalpages
;
4403 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4407 #ifndef CONFIG_SPARSEMEM
4409 * Calculate the size of the zone->blockflags rounded to an unsigned long
4410 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4411 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4412 * round what is now in bits to nearest long in bits, then return it in
4415 static unsigned long __init
usemap_size(unsigned long zonesize
)
4417 unsigned long usemapsize
;
4419 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4420 usemapsize
= usemapsize
>> pageblock_order
;
4421 usemapsize
*= NR_PAGEBLOCK_BITS
;
4422 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4424 return usemapsize
/ 8;
4427 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4428 struct zone
*zone
, unsigned long zonesize
)
4430 unsigned long usemapsize
= usemap_size(zonesize
);
4431 zone
->pageblock_flags
= NULL
;
4433 zone
->pageblock_flags
= alloc_bootmem_node_nopanic(pgdat
,
4437 static inline void setup_usemap(struct pglist_data
*pgdat
,
4438 struct zone
*zone
, unsigned long zonesize
) {}
4439 #endif /* CONFIG_SPARSEMEM */
4441 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4443 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4444 void __init
set_pageblock_order(void)
4448 /* Check that pageblock_nr_pages has not already been setup */
4449 if (pageblock_order
)
4452 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4453 order
= HUGETLB_PAGE_ORDER
;
4455 order
= MAX_ORDER
- 1;
4458 * Assume the largest contiguous order of interest is a huge page.
4459 * This value may be variable depending on boot parameters on IA64 and
4462 pageblock_order
= order
;
4464 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4467 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4468 * is unused as pageblock_order is set at compile-time. See
4469 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4472 void __init
set_pageblock_order(void)
4476 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4479 * Set up the zone data structures:
4480 * - mark all pages reserved
4481 * - mark all memory queues empty
4482 * - clear the memory bitmaps
4484 * NOTE: pgdat should get zeroed by caller.
4486 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4487 unsigned long *zones_size
, unsigned long *zholes_size
)
4490 int nid
= pgdat
->node_id
;
4491 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4494 pgdat_resize_init(pgdat
);
4495 init_waitqueue_head(&pgdat
->kswapd_wait
);
4496 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
4497 pgdat_page_cgroup_init(pgdat
);
4499 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4500 struct zone
*zone
= pgdat
->node_zones
+ j
;
4501 unsigned long size
, realsize
, memmap_pages
;
4503 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4504 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
4508 * Adjust realsize so that it accounts for how much memory
4509 * is used by this zone for memmap. This affects the watermark
4510 * and per-cpu initialisations
4513 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
4514 if (realsize
>= memmap_pages
) {
4515 realsize
-= memmap_pages
;
4518 " %s zone: %lu pages used for memmap\n",
4519 zone_names
[j
], memmap_pages
);
4522 " %s zone: %lu pages exceeds realsize %lu\n",
4523 zone_names
[j
], memmap_pages
, realsize
);
4525 /* Account for reserved pages */
4526 if (j
== 0 && realsize
> dma_reserve
) {
4527 realsize
-= dma_reserve
;
4528 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4529 zone_names
[0], dma_reserve
);
4532 if (!is_highmem_idx(j
))
4533 nr_kernel_pages
+= realsize
;
4534 nr_all_pages
+= realsize
;
4536 zone
->spanned_pages
= size
;
4537 zone
->present_pages
= realsize
;
4540 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
4542 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
4544 zone
->name
= zone_names
[j
];
4545 spin_lock_init(&zone
->lock
);
4546 spin_lock_init(&zone
->lru_lock
);
4547 zone_seqlock_init(zone
);
4548 zone
->zone_pgdat
= pgdat
;
4550 zone_pcp_init(zone
);
4551 lruvec_init(&zone
->lruvec
);
4555 set_pageblock_order();
4556 setup_usemap(pgdat
, zone
, size
);
4557 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4558 size
, MEMMAP_EARLY
);
4560 memmap_init(size
, nid
, j
, zone_start_pfn
);
4561 zone_start_pfn
+= size
;
4565 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4567 /* Skip empty nodes */
4568 if (!pgdat
->node_spanned_pages
)
4571 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4572 /* ia64 gets its own node_mem_map, before this, without bootmem */
4573 if (!pgdat
->node_mem_map
) {
4574 unsigned long size
, start
, end
;
4578 * The zone's endpoints aren't required to be MAX_ORDER
4579 * aligned but the node_mem_map endpoints must be in order
4580 * for the buddy allocator to function correctly.
4582 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4583 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4584 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4585 size
= (end
- start
) * sizeof(struct page
);
4586 map
= alloc_remap(pgdat
->node_id
, size
);
4588 map
= alloc_bootmem_node_nopanic(pgdat
, size
);
4589 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4591 #ifndef CONFIG_NEED_MULTIPLE_NODES
4593 * With no DISCONTIG, the global mem_map is just set as node 0's
4595 if (pgdat
== NODE_DATA(0)) {
4596 mem_map
= NODE_DATA(0)->node_mem_map
;
4597 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4598 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4599 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4600 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4603 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4606 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4607 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4609 pg_data_t
*pgdat
= NODE_DATA(nid
);
4611 /* pg_data_t should be reset to zero when it's allocated */
4612 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
4614 pgdat
->node_id
= nid
;
4615 pgdat
->node_start_pfn
= node_start_pfn
;
4616 init_zone_allows_reclaim(nid
);
4617 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4619 alloc_node_mem_map(pgdat
);
4620 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4621 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4622 nid
, (unsigned long)pgdat
,
4623 (unsigned long)pgdat
->node_mem_map
);
4626 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4629 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4631 #if MAX_NUMNODES > 1
4633 * Figure out the number of possible node ids.
4635 static void __init
setup_nr_node_ids(void)
4638 unsigned int highest
= 0;
4640 for_each_node_mask(node
, node_possible_map
)
4642 nr_node_ids
= highest
+ 1;
4645 static inline void setup_nr_node_ids(void)
4651 * node_map_pfn_alignment - determine the maximum internode alignment
4653 * This function should be called after node map is populated and sorted.
4654 * It calculates the maximum power of two alignment which can distinguish
4657 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4658 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4659 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4660 * shifted, 1GiB is enough and this function will indicate so.
4662 * This is used to test whether pfn -> nid mapping of the chosen memory
4663 * model has fine enough granularity to avoid incorrect mapping for the
4664 * populated node map.
4666 * Returns the determined alignment in pfn's. 0 if there is no alignment
4667 * requirement (single node).
4669 unsigned long __init
node_map_pfn_alignment(void)
4671 unsigned long accl_mask
= 0, last_end
= 0;
4672 unsigned long start
, end
, mask
;
4676 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4677 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4684 * Start with a mask granular enough to pin-point to the
4685 * start pfn and tick off bits one-by-one until it becomes
4686 * too coarse to separate the current node from the last.
4688 mask
= ~((1 << __ffs(start
)) - 1);
4689 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4692 /* accumulate all internode masks */
4696 /* convert mask to number of pages */
4697 return ~accl_mask
+ 1;
4700 /* Find the lowest pfn for a node */
4701 static unsigned long __init
find_min_pfn_for_node(int nid
)
4703 unsigned long min_pfn
= ULONG_MAX
;
4704 unsigned long start_pfn
;
4707 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4708 min_pfn
= min(min_pfn
, start_pfn
);
4710 if (min_pfn
== ULONG_MAX
) {
4712 "Could not find start_pfn for node %d\n", nid
);
4720 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4722 * It returns the minimum PFN based on information provided via
4723 * add_active_range().
4725 unsigned long __init
find_min_pfn_with_active_regions(void)
4727 return find_min_pfn_for_node(MAX_NUMNODES
);
4731 * early_calculate_totalpages()
4732 * Sum pages in active regions for movable zone.
4733 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4735 static unsigned long __init
early_calculate_totalpages(void)
4737 unsigned long totalpages
= 0;
4738 unsigned long start_pfn
, end_pfn
;
4741 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
4742 unsigned long pages
= end_pfn
- start_pfn
;
4744 totalpages
+= pages
;
4746 node_set_state(nid
, N_HIGH_MEMORY
);
4752 * Find the PFN the Movable zone begins in each node. Kernel memory
4753 * is spread evenly between nodes as long as the nodes have enough
4754 * memory. When they don't, some nodes will have more kernelcore than
4757 static void __init
find_zone_movable_pfns_for_nodes(void)
4760 unsigned long usable_startpfn
;
4761 unsigned long kernelcore_node
, kernelcore_remaining
;
4762 /* save the state before borrow the nodemask */
4763 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4764 unsigned long totalpages
= early_calculate_totalpages();
4765 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4768 * If movablecore was specified, calculate what size of
4769 * kernelcore that corresponds so that memory usable for
4770 * any allocation type is evenly spread. If both kernelcore
4771 * and movablecore are specified, then the value of kernelcore
4772 * will be used for required_kernelcore if it's greater than
4773 * what movablecore would have allowed.
4775 if (required_movablecore
) {
4776 unsigned long corepages
;
4779 * Round-up so that ZONE_MOVABLE is at least as large as what
4780 * was requested by the user
4782 required_movablecore
=
4783 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4784 corepages
= totalpages
- required_movablecore
;
4786 required_kernelcore
= max(required_kernelcore
, corepages
);
4789 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4790 if (!required_kernelcore
)
4793 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4794 find_usable_zone_for_movable();
4795 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4798 /* Spread kernelcore memory as evenly as possible throughout nodes */
4799 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4800 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4801 unsigned long start_pfn
, end_pfn
;
4804 * Recalculate kernelcore_node if the division per node
4805 * now exceeds what is necessary to satisfy the requested
4806 * amount of memory for the kernel
4808 if (required_kernelcore
< kernelcore_node
)
4809 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4812 * As the map is walked, we track how much memory is usable
4813 * by the kernel using kernelcore_remaining. When it is
4814 * 0, the rest of the node is usable by ZONE_MOVABLE
4816 kernelcore_remaining
= kernelcore_node
;
4818 /* Go through each range of PFNs within this node */
4819 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4820 unsigned long size_pages
;
4822 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
4823 if (start_pfn
>= end_pfn
)
4826 /* Account for what is only usable for kernelcore */
4827 if (start_pfn
< usable_startpfn
) {
4828 unsigned long kernel_pages
;
4829 kernel_pages
= min(end_pfn
, usable_startpfn
)
4832 kernelcore_remaining
-= min(kernel_pages
,
4833 kernelcore_remaining
);
4834 required_kernelcore
-= min(kernel_pages
,
4835 required_kernelcore
);
4837 /* Continue if range is now fully accounted */
4838 if (end_pfn
<= usable_startpfn
) {
4841 * Push zone_movable_pfn to the end so
4842 * that if we have to rebalance
4843 * kernelcore across nodes, we will
4844 * not double account here
4846 zone_movable_pfn
[nid
] = end_pfn
;
4849 start_pfn
= usable_startpfn
;
4853 * The usable PFN range for ZONE_MOVABLE is from
4854 * start_pfn->end_pfn. Calculate size_pages as the
4855 * number of pages used as kernelcore
4857 size_pages
= end_pfn
- start_pfn
;
4858 if (size_pages
> kernelcore_remaining
)
4859 size_pages
= kernelcore_remaining
;
4860 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4863 * Some kernelcore has been met, update counts and
4864 * break if the kernelcore for this node has been
4867 required_kernelcore
-= min(required_kernelcore
,
4869 kernelcore_remaining
-= size_pages
;
4870 if (!kernelcore_remaining
)
4876 * If there is still required_kernelcore, we do another pass with one
4877 * less node in the count. This will push zone_movable_pfn[nid] further
4878 * along on the nodes that still have memory until kernelcore is
4882 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4885 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4886 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4887 zone_movable_pfn
[nid
] =
4888 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4891 /* restore the node_state */
4892 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4895 /* Any regular memory on that node ? */
4896 static void __init
check_for_regular_memory(pg_data_t
*pgdat
)
4898 #ifdef CONFIG_HIGHMEM
4899 enum zone_type zone_type
;
4901 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4902 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4903 if (zone
->present_pages
) {
4904 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4912 * free_area_init_nodes - Initialise all pg_data_t and zone data
4913 * @max_zone_pfn: an array of max PFNs for each zone
4915 * This will call free_area_init_node() for each active node in the system.
4916 * Using the page ranges provided by add_active_range(), the size of each
4917 * zone in each node and their holes is calculated. If the maximum PFN
4918 * between two adjacent zones match, it is assumed that the zone is empty.
4919 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4920 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4921 * starts where the previous one ended. For example, ZONE_DMA32 starts
4922 * at arch_max_dma_pfn.
4924 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4926 unsigned long start_pfn
, end_pfn
;
4929 /* Record where the zone boundaries are */
4930 memset(arch_zone_lowest_possible_pfn
, 0,
4931 sizeof(arch_zone_lowest_possible_pfn
));
4932 memset(arch_zone_highest_possible_pfn
, 0,
4933 sizeof(arch_zone_highest_possible_pfn
));
4934 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4935 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4936 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4937 if (i
== ZONE_MOVABLE
)
4939 arch_zone_lowest_possible_pfn
[i
] =
4940 arch_zone_highest_possible_pfn
[i
-1];
4941 arch_zone_highest_possible_pfn
[i
] =
4942 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4944 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4945 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4947 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4948 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4949 find_zone_movable_pfns_for_nodes();
4951 /* Print out the zone ranges */
4952 printk("Zone ranges:\n");
4953 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4954 if (i
== ZONE_MOVABLE
)
4956 printk(KERN_CONT
" %-8s ", zone_names
[i
]);
4957 if (arch_zone_lowest_possible_pfn
[i
] ==
4958 arch_zone_highest_possible_pfn
[i
])
4959 printk(KERN_CONT
"empty\n");
4961 printk(KERN_CONT
"[mem %0#10lx-%0#10lx]\n",
4962 arch_zone_lowest_possible_pfn
[i
] << PAGE_SHIFT
,
4963 (arch_zone_highest_possible_pfn
[i
]
4964 << PAGE_SHIFT
) - 1);
4967 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4968 printk("Movable zone start for each node\n");
4969 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4970 if (zone_movable_pfn
[i
])
4971 printk(" Node %d: %#010lx\n", i
,
4972 zone_movable_pfn
[i
] << PAGE_SHIFT
);
4975 /* Print out the early node map */
4976 printk("Early memory node ranges\n");
4977 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4978 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid
,
4979 start_pfn
<< PAGE_SHIFT
, (end_pfn
<< PAGE_SHIFT
) - 1);
4981 /* Initialise every node */
4982 mminit_verify_pageflags_layout();
4983 setup_nr_node_ids();
4984 for_each_online_node(nid
) {
4985 pg_data_t
*pgdat
= NODE_DATA(nid
);
4986 free_area_init_node(nid
, NULL
,
4987 find_min_pfn_for_node(nid
), NULL
);
4989 /* Any memory on that node */
4990 if (pgdat
->node_present_pages
)
4991 node_set_state(nid
, N_HIGH_MEMORY
);
4992 check_for_regular_memory(pgdat
);
4996 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4998 unsigned long long coremem
;
5002 coremem
= memparse(p
, &p
);
5003 *core
= coremem
>> PAGE_SHIFT
;
5005 /* Paranoid check that UL is enough for the coremem value */
5006 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5012 * kernelcore=size sets the amount of memory for use for allocations that
5013 * cannot be reclaimed or migrated.
5015 static int __init
cmdline_parse_kernelcore(char *p
)
5017 return cmdline_parse_core(p
, &required_kernelcore
);
5021 * movablecore=size sets the amount of memory for use for allocations that
5022 * can be reclaimed or migrated.
5024 static int __init
cmdline_parse_movablecore(char *p
)
5026 return cmdline_parse_core(p
, &required_movablecore
);
5029 early_param("kernelcore", cmdline_parse_kernelcore
);
5030 early_param("movablecore", cmdline_parse_movablecore
);
5032 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5035 * set_dma_reserve - set the specified number of pages reserved in the first zone
5036 * @new_dma_reserve: The number of pages to mark reserved
5038 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5039 * In the DMA zone, a significant percentage may be consumed by kernel image
5040 * and other unfreeable allocations which can skew the watermarks badly. This
5041 * function may optionally be used to account for unfreeable pages in the
5042 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5043 * smaller per-cpu batchsize.
5045 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5047 dma_reserve
= new_dma_reserve
;
5050 void __init
free_area_init(unsigned long *zones_size
)
5052 free_area_init_node(0, zones_size
,
5053 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5056 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5057 unsigned long action
, void *hcpu
)
5059 int cpu
= (unsigned long)hcpu
;
5061 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5062 lru_add_drain_cpu(cpu
);
5066 * Spill the event counters of the dead processor
5067 * into the current processors event counters.
5068 * This artificially elevates the count of the current
5071 vm_events_fold_cpu(cpu
);
5074 * Zero the differential counters of the dead processor
5075 * so that the vm statistics are consistent.
5077 * This is only okay since the processor is dead and cannot
5078 * race with what we are doing.
5080 refresh_cpu_vm_stats(cpu
);
5085 void __init
page_alloc_init(void)
5087 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5091 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5092 * or min_free_kbytes changes.
5094 static void calculate_totalreserve_pages(void)
5096 struct pglist_data
*pgdat
;
5097 unsigned long reserve_pages
= 0;
5098 enum zone_type i
, j
;
5100 for_each_online_pgdat(pgdat
) {
5101 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5102 struct zone
*zone
= pgdat
->node_zones
+ i
;
5103 unsigned long max
= 0;
5105 /* Find valid and maximum lowmem_reserve in the zone */
5106 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5107 if (zone
->lowmem_reserve
[j
] > max
)
5108 max
= zone
->lowmem_reserve
[j
];
5111 /* we treat the high watermark as reserved pages. */
5112 max
+= high_wmark_pages(zone
);
5114 if (max
> zone
->present_pages
)
5115 max
= zone
->present_pages
;
5116 reserve_pages
+= max
;
5118 * Lowmem reserves are not available to
5119 * GFP_HIGHUSER page cache allocations and
5120 * kswapd tries to balance zones to their high
5121 * watermark. As a result, neither should be
5122 * regarded as dirtyable memory, to prevent a
5123 * situation where reclaim has to clean pages
5124 * in order to balance the zones.
5126 zone
->dirty_balance_reserve
= max
;
5129 dirty_balance_reserve
= reserve_pages
;
5130 totalreserve_pages
= reserve_pages
;
5134 * setup_per_zone_lowmem_reserve - called whenever
5135 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5136 * has a correct pages reserved value, so an adequate number of
5137 * pages are left in the zone after a successful __alloc_pages().
5139 static void setup_per_zone_lowmem_reserve(void)
5141 struct pglist_data
*pgdat
;
5142 enum zone_type j
, idx
;
5144 for_each_online_pgdat(pgdat
) {
5145 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5146 struct zone
*zone
= pgdat
->node_zones
+ j
;
5147 unsigned long present_pages
= zone
->present_pages
;
5149 zone
->lowmem_reserve
[j
] = 0;
5153 struct zone
*lower_zone
;
5157 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5158 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5160 lower_zone
= pgdat
->node_zones
+ idx
;
5161 lower_zone
->lowmem_reserve
[j
] = present_pages
/
5162 sysctl_lowmem_reserve_ratio
[idx
];
5163 present_pages
+= lower_zone
->present_pages
;
5168 /* update totalreserve_pages */
5169 calculate_totalreserve_pages();
5172 static void __setup_per_zone_wmarks(void)
5174 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5175 unsigned long lowmem_pages
= 0;
5177 unsigned long flags
;
5179 /* Calculate total number of !ZONE_HIGHMEM pages */
5180 for_each_zone(zone
) {
5181 if (!is_highmem(zone
))
5182 lowmem_pages
+= zone
->present_pages
;
5185 for_each_zone(zone
) {
5188 spin_lock_irqsave(&zone
->lock
, flags
);
5189 tmp
= (u64
)pages_min
* zone
->present_pages
;
5190 do_div(tmp
, lowmem_pages
);
5191 if (is_highmem(zone
)) {
5193 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5194 * need highmem pages, so cap pages_min to a small
5197 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5198 * deltas controls asynch page reclaim, and so should
5199 * not be capped for highmem.
5203 min_pages
= zone
->present_pages
/ 1024;
5204 if (min_pages
< SWAP_CLUSTER_MAX
)
5205 min_pages
= SWAP_CLUSTER_MAX
;
5206 if (min_pages
> 128)
5208 zone
->watermark
[WMARK_MIN
] = min_pages
;
5211 * If it's a lowmem zone, reserve a number of pages
5212 * proportionate to the zone's size.
5214 zone
->watermark
[WMARK_MIN
] = tmp
;
5217 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
5218 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
5220 zone
->watermark
[WMARK_MIN
] += cma_wmark_pages(zone
);
5221 zone
->watermark
[WMARK_LOW
] += cma_wmark_pages(zone
);
5222 zone
->watermark
[WMARK_HIGH
] += cma_wmark_pages(zone
);
5224 setup_zone_migrate_reserve(zone
);
5225 spin_unlock_irqrestore(&zone
->lock
, flags
);
5228 /* update totalreserve_pages */
5229 calculate_totalreserve_pages();
5233 * setup_per_zone_wmarks - called when min_free_kbytes changes
5234 * or when memory is hot-{added|removed}
5236 * Ensures that the watermark[min,low,high] values for each zone are set
5237 * correctly with respect to min_free_kbytes.
5239 void setup_per_zone_wmarks(void)
5241 mutex_lock(&zonelists_mutex
);
5242 __setup_per_zone_wmarks();
5243 mutex_unlock(&zonelists_mutex
);
5247 * The inactive anon list should be small enough that the VM never has to
5248 * do too much work, but large enough that each inactive page has a chance
5249 * to be referenced again before it is swapped out.
5251 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5252 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5253 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5254 * the anonymous pages are kept on the inactive list.
5257 * memory ratio inactive anon
5258 * -------------------------------------
5267 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
5269 unsigned int gb
, ratio
;
5271 /* Zone size in gigabytes */
5272 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
5274 ratio
= int_sqrt(10 * gb
);
5278 zone
->inactive_ratio
= ratio
;
5281 static void __meminit
setup_per_zone_inactive_ratio(void)
5286 calculate_zone_inactive_ratio(zone
);
5290 * Initialise min_free_kbytes.
5292 * For small machines we want it small (128k min). For large machines
5293 * we want it large (64MB max). But it is not linear, because network
5294 * bandwidth does not increase linearly with machine size. We use
5296 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5297 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5313 int __meminit
init_per_zone_wmark_min(void)
5315 unsigned long lowmem_kbytes
;
5317 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5319 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5320 if (min_free_kbytes
< 128)
5321 min_free_kbytes
= 128;
5322 if (min_free_kbytes
> 65536)
5323 min_free_kbytes
= 65536;
5324 setup_per_zone_wmarks();
5325 refresh_zone_stat_thresholds();
5326 setup_per_zone_lowmem_reserve();
5327 setup_per_zone_inactive_ratio();
5330 module_init(init_per_zone_wmark_min
)
5333 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5334 * that we can call two helper functions whenever min_free_kbytes
5337 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5338 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5340 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5342 setup_per_zone_wmarks();
5347 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5348 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5353 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5358 zone
->min_unmapped_pages
= (zone
->present_pages
*
5359 sysctl_min_unmapped_ratio
) / 100;
5363 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5364 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5369 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5374 zone
->min_slab_pages
= (zone
->present_pages
*
5375 sysctl_min_slab_ratio
) / 100;
5381 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5382 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5383 * whenever sysctl_lowmem_reserve_ratio changes.
5385 * The reserve ratio obviously has absolutely no relation with the
5386 * minimum watermarks. The lowmem reserve ratio can only make sense
5387 * if in function of the boot time zone sizes.
5389 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5390 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5392 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5393 setup_per_zone_lowmem_reserve();
5398 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5399 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5400 * can have before it gets flushed back to buddy allocator.
5403 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5404 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5410 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5411 if (!write
|| (ret
< 0))
5413 for_each_populated_zone(zone
) {
5414 for_each_possible_cpu(cpu
) {
5416 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5417 setup_pagelist_highmark(
5418 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5424 int hashdist
= HASHDIST_DEFAULT
;
5427 static int __init
set_hashdist(char *str
)
5431 hashdist
= simple_strtoul(str
, &str
, 0);
5434 __setup("hashdist=", set_hashdist
);
5438 * allocate a large system hash table from bootmem
5439 * - it is assumed that the hash table must contain an exact power-of-2
5440 * quantity of entries
5441 * - limit is the number of hash buckets, not the total allocation size
5443 void *__init
alloc_large_system_hash(const char *tablename
,
5444 unsigned long bucketsize
,
5445 unsigned long numentries
,
5448 unsigned int *_hash_shift
,
5449 unsigned int *_hash_mask
,
5450 unsigned long low_limit
,
5451 unsigned long high_limit
)
5453 unsigned long long max
= high_limit
;
5454 unsigned long log2qty
, size
;
5457 /* allow the kernel cmdline to have a say */
5459 /* round applicable memory size up to nearest megabyte */
5460 numentries
= nr_kernel_pages
;
5461 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5462 numentries
>>= 20 - PAGE_SHIFT
;
5463 numentries
<<= 20 - PAGE_SHIFT
;
5465 /* limit to 1 bucket per 2^scale bytes of low memory */
5466 if (scale
> PAGE_SHIFT
)
5467 numentries
>>= (scale
- PAGE_SHIFT
);
5469 numentries
<<= (PAGE_SHIFT
- scale
);
5471 /* Make sure we've got at least a 0-order allocation.. */
5472 if (unlikely(flags
& HASH_SMALL
)) {
5473 /* Makes no sense without HASH_EARLY */
5474 WARN_ON(!(flags
& HASH_EARLY
));
5475 if (!(numentries
>> *_hash_shift
)) {
5476 numentries
= 1UL << *_hash_shift
;
5477 BUG_ON(!numentries
);
5479 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5480 numentries
= PAGE_SIZE
/ bucketsize
;
5482 numentries
= roundup_pow_of_two(numentries
);
5484 /* limit allocation size to 1/16 total memory by default */
5486 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5487 do_div(max
, bucketsize
);
5489 max
= min(max
, 0x80000000ULL
);
5491 if (numentries
< low_limit
)
5492 numentries
= low_limit
;
5493 if (numentries
> max
)
5496 log2qty
= ilog2(numentries
);
5499 size
= bucketsize
<< log2qty
;
5500 if (flags
& HASH_EARLY
)
5501 table
= alloc_bootmem_nopanic(size
);
5503 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5506 * If bucketsize is not a power-of-two, we may free
5507 * some pages at the end of hash table which
5508 * alloc_pages_exact() automatically does
5510 if (get_order(size
) < MAX_ORDER
) {
5511 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5512 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5515 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5518 panic("Failed to allocate %s hash table\n", tablename
);
5520 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5523 ilog2(size
) - PAGE_SHIFT
,
5527 *_hash_shift
= log2qty
;
5529 *_hash_mask
= (1 << log2qty
) - 1;
5534 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5535 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5538 #ifdef CONFIG_SPARSEMEM
5539 return __pfn_to_section(pfn
)->pageblock_flags
;
5541 return zone
->pageblock_flags
;
5542 #endif /* CONFIG_SPARSEMEM */
5545 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5547 #ifdef CONFIG_SPARSEMEM
5548 pfn
&= (PAGES_PER_SECTION
-1);
5549 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5551 pfn
= pfn
- zone
->zone_start_pfn
;
5552 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5553 #endif /* CONFIG_SPARSEMEM */
5557 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5558 * @page: The page within the block of interest
5559 * @start_bitidx: The first bit of interest to retrieve
5560 * @end_bitidx: The last bit of interest
5561 * returns pageblock_bits flags
5563 unsigned long get_pageblock_flags_group(struct page
*page
,
5564 int start_bitidx
, int end_bitidx
)
5567 unsigned long *bitmap
;
5568 unsigned long pfn
, bitidx
;
5569 unsigned long flags
= 0;
5570 unsigned long value
= 1;
5572 zone
= page_zone(page
);
5573 pfn
= page_to_pfn(page
);
5574 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5575 bitidx
= pfn_to_bitidx(zone
, pfn
);
5577 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5578 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5585 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5586 * @page: The page within the block of interest
5587 * @start_bitidx: The first bit of interest
5588 * @end_bitidx: The last bit of interest
5589 * @flags: The flags to set
5591 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5592 int start_bitidx
, int end_bitidx
)
5595 unsigned long *bitmap
;
5596 unsigned long pfn
, bitidx
;
5597 unsigned long value
= 1;
5599 zone
= page_zone(page
);
5600 pfn
= page_to_pfn(page
);
5601 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5602 bitidx
= pfn_to_bitidx(zone
, pfn
);
5603 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5604 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5606 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5608 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5610 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5614 * This function checks whether pageblock includes unmovable pages or not.
5615 * If @count is not zero, it is okay to include less @count unmovable pages
5617 * PageLRU check wihtout isolation or lru_lock could race so that
5618 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5619 * expect this function should be exact.
5621 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
5622 bool skip_hwpoisoned_pages
)
5624 unsigned long pfn
, iter
, found
;
5628 * For avoiding noise data, lru_add_drain_all() should be called
5629 * If ZONE_MOVABLE, the zone never contains unmovable pages
5631 if (zone_idx(zone
) == ZONE_MOVABLE
)
5633 mt
= get_pageblock_migratetype(page
);
5634 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
5637 pfn
= page_to_pfn(page
);
5638 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
5639 unsigned long check
= pfn
+ iter
;
5641 if (!pfn_valid_within(check
))
5644 page
= pfn_to_page(check
);
5646 * We can't use page_count without pin a page
5647 * because another CPU can free compound page.
5648 * This check already skips compound tails of THP
5649 * because their page->_count is zero at all time.
5651 if (!atomic_read(&page
->_count
)) {
5652 if (PageBuddy(page
))
5653 iter
+= (1 << page_order(page
)) - 1;
5658 * The HWPoisoned page may be not in buddy system, and
5659 * page_count() is not 0.
5661 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
5667 * If there are RECLAIMABLE pages, we need to check it.
5668 * But now, memory offline itself doesn't call shrink_slab()
5669 * and it still to be fixed.
5672 * If the page is not RAM, page_count()should be 0.
5673 * we don't need more check. This is an _used_ not-movable page.
5675 * The problematic thing here is PG_reserved pages. PG_reserved
5676 * is set to both of a memory hole page and a _used_ kernel
5685 bool is_pageblock_removable_nolock(struct page
*page
)
5691 * We have to be careful here because we are iterating over memory
5692 * sections which are not zone aware so we might end up outside of
5693 * the zone but still within the section.
5694 * We have to take care about the node as well. If the node is offline
5695 * its NODE_DATA will be NULL - see page_zone.
5697 if (!node_online(page_to_nid(page
)))
5700 zone
= page_zone(page
);
5701 pfn
= page_to_pfn(page
);
5702 if (zone
->zone_start_pfn
> pfn
||
5703 zone
->zone_start_pfn
+ zone
->spanned_pages
<= pfn
)
5706 return !has_unmovable_pages(zone
, page
, 0, true);
5711 static unsigned long pfn_max_align_down(unsigned long pfn
)
5713 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5714 pageblock_nr_pages
) - 1);
5717 static unsigned long pfn_max_align_up(unsigned long pfn
)
5719 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5720 pageblock_nr_pages
));
5723 /* [start, end) must belong to a single zone. */
5724 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
5725 unsigned long start
, unsigned long end
)
5727 /* This function is based on compact_zone() from compaction.c. */
5728 unsigned long nr_reclaimed
;
5729 unsigned long pfn
= start
;
5730 unsigned int tries
= 0;
5733 migrate_prep_local();
5735 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
5736 if (fatal_signal_pending(current
)) {
5741 if (list_empty(&cc
->migratepages
)) {
5742 cc
->nr_migratepages
= 0;
5743 pfn
= isolate_migratepages_range(cc
->zone
, cc
,
5750 } else if (++tries
== 5) {
5751 ret
= ret
< 0 ? ret
: -EBUSY
;
5755 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
5757 cc
->nr_migratepages
-= nr_reclaimed
;
5759 ret
= migrate_pages(&cc
->migratepages
,
5760 alloc_migrate_target
,
5761 0, false, MIGRATE_SYNC
);
5764 putback_lru_pages(&cc
->migratepages
);
5765 return ret
> 0 ? 0 : ret
;
5769 * Update zone's cma pages counter used for watermark level calculation.
5771 static inline void __update_cma_watermarks(struct zone
*zone
, int count
)
5773 unsigned long flags
;
5774 spin_lock_irqsave(&zone
->lock
, flags
);
5775 zone
->min_cma_pages
+= count
;
5776 spin_unlock_irqrestore(&zone
->lock
, flags
);
5777 setup_per_zone_wmarks();
5781 * Trigger memory pressure bump to reclaim some pages in order to be able to
5782 * allocate 'count' pages in single page units. Does similar work as
5783 *__alloc_pages_slowpath() function.
5785 static int __reclaim_pages(struct zone
*zone
, gfp_t gfp_mask
, int count
)
5787 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
5788 struct zonelist
*zonelist
= node_zonelist(0, gfp_mask
);
5789 int did_some_progress
= 0;
5793 * Increase level of watermarks to force kswapd do his job
5794 * to stabilise at new watermark level.
5796 __update_cma_watermarks(zone
, count
);
5798 /* Obey watermarks as if the page was being allocated */
5799 while (!zone_watermark_ok(zone
, 0, low_wmark_pages(zone
), 0, 0)) {
5800 wake_all_kswapd(order
, zonelist
, high_zoneidx
, zone_idx(zone
));
5802 did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
5804 if (!did_some_progress
) {
5805 /* Exhausted what can be done so it's blamo time */
5806 out_of_memory(zonelist
, gfp_mask
, order
, NULL
, false);
5810 /* Restore original watermark levels. */
5811 __update_cma_watermarks(zone
, -count
);
5817 * alloc_contig_range() -- tries to allocate given range of pages
5818 * @start: start PFN to allocate
5819 * @end: one-past-the-last PFN to allocate
5820 * @migratetype: migratetype of the underlaying pageblocks (either
5821 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
5822 * in range must have the same migratetype and it must
5823 * be either of the two.
5825 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
5826 * aligned, however it's the caller's responsibility to guarantee that
5827 * we are the only thread that changes migrate type of pageblocks the
5830 * The PFN range must belong to a single zone.
5832 * Returns zero on success or negative error code. On success all
5833 * pages which PFN is in [start, end) are allocated for the caller and
5834 * need to be freed with free_contig_range().
5836 int alloc_contig_range(unsigned long start
, unsigned long end
,
5837 unsigned migratetype
)
5839 struct zone
*zone
= page_zone(pfn_to_page(start
));
5840 unsigned long outer_start
, outer_end
;
5843 struct compact_control cc
= {
5844 .nr_migratepages
= 0,
5846 .zone
= page_zone(pfn_to_page(start
)),
5848 .ignore_skip_hint
= true,
5850 INIT_LIST_HEAD(&cc
.migratepages
);
5853 * What we do here is we mark all pageblocks in range as
5854 * MIGRATE_ISOLATE. Because pageblock and max order pages may
5855 * have different sizes, and due to the way page allocator
5856 * work, we align the range to biggest of the two pages so
5857 * that page allocator won't try to merge buddies from
5858 * different pageblocks and change MIGRATE_ISOLATE to some
5859 * other migration type.
5861 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
5862 * migrate the pages from an unaligned range (ie. pages that
5863 * we are interested in). This will put all the pages in
5864 * range back to page allocator as MIGRATE_ISOLATE.
5866 * When this is done, we take the pages in range from page
5867 * allocator removing them from the buddy system. This way
5868 * page allocator will never consider using them.
5870 * This lets us mark the pageblocks back as
5871 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
5872 * aligned range but not in the unaligned, original range are
5873 * put back to page allocator so that buddy can use them.
5876 ret
= start_isolate_page_range(pfn_max_align_down(start
),
5877 pfn_max_align_up(end
), migratetype
,
5882 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
5887 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
5888 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
5889 * more, all pages in [start, end) are free in page allocator.
5890 * What we are going to do is to allocate all pages from
5891 * [start, end) (that is remove them from page allocator).
5893 * The only problem is that pages at the beginning and at the
5894 * end of interesting range may be not aligned with pages that
5895 * page allocator holds, ie. they can be part of higher order
5896 * pages. Because of this, we reserve the bigger range and
5897 * once this is done free the pages we are not interested in.
5899 * We don't have to hold zone->lock here because the pages are
5900 * isolated thus they won't get removed from buddy.
5903 lru_add_drain_all();
5907 outer_start
= start
;
5908 while (!PageBuddy(pfn_to_page(outer_start
))) {
5909 if (++order
>= MAX_ORDER
) {
5913 outer_start
&= ~0UL << order
;
5916 /* Make sure the range is really isolated. */
5917 if (test_pages_isolated(outer_start
, end
, false)) {
5918 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
5925 * Reclaim enough pages to make sure that contiguous allocation
5926 * will not starve the system.
5928 __reclaim_pages(zone
, GFP_HIGHUSER_MOVABLE
, end
-start
);
5930 /* Grab isolated pages from freelists. */
5931 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
5937 /* Free head and tail (if any) */
5938 if (start
!= outer_start
)
5939 free_contig_range(outer_start
, start
- outer_start
);
5940 if (end
!= outer_end
)
5941 free_contig_range(end
, outer_end
- end
);
5944 undo_isolate_page_range(pfn_max_align_down(start
),
5945 pfn_max_align_up(end
), migratetype
);
5949 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
5951 for (; nr_pages
--; ++pfn
)
5952 __free_page(pfn_to_page(pfn
));
5956 #ifdef CONFIG_MEMORY_HOTPLUG
5957 static int __meminit
__zone_pcp_update(void *data
)
5959 struct zone
*zone
= data
;
5961 unsigned long batch
= zone_batchsize(zone
), flags
;
5963 for_each_possible_cpu(cpu
) {
5964 struct per_cpu_pageset
*pset
;
5965 struct per_cpu_pages
*pcp
;
5967 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
5970 local_irq_save(flags
);
5972 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
5973 drain_zonestat(zone
, pset
);
5974 setup_pageset(pset
, batch
);
5975 local_irq_restore(flags
);
5980 void __meminit
zone_pcp_update(struct zone
*zone
)
5982 stop_machine(__zone_pcp_update
, zone
, NULL
);
5986 void zone_pcp_reset(struct zone
*zone
)
5988 unsigned long flags
;
5990 struct per_cpu_pageset
*pset
;
5992 /* avoid races with drain_pages() */
5993 local_irq_save(flags
);
5994 if (zone
->pageset
!= &boot_pageset
) {
5995 for_each_online_cpu(cpu
) {
5996 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
5997 drain_zonestat(zone
, pset
);
5999 free_percpu(zone
->pageset
);
6000 zone
->pageset
= &boot_pageset
;
6002 local_irq_restore(flags
);
6005 #ifdef CONFIG_MEMORY_HOTREMOVE
6007 * All pages in the range must be isolated before calling this.
6010 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6016 unsigned long flags
;
6017 /* find the first valid pfn */
6018 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6023 zone
= page_zone(pfn_to_page(pfn
));
6024 spin_lock_irqsave(&zone
->lock
, flags
);
6026 while (pfn
< end_pfn
) {
6027 if (!pfn_valid(pfn
)) {
6031 page
= pfn_to_page(pfn
);
6033 * The HWPoisoned page may be not in buddy system, and
6034 * page_count() is not 0.
6036 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6038 SetPageReserved(page
);
6042 BUG_ON(page_count(page
));
6043 BUG_ON(!PageBuddy(page
));
6044 order
= page_order(page
);
6045 #ifdef CONFIG_DEBUG_VM
6046 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6047 pfn
, 1 << order
, end_pfn
);
6049 list_del(&page
->lru
);
6050 rmv_page_order(page
);
6051 zone
->free_area
[order
].nr_free
--;
6052 for (i
= 0; i
< (1 << order
); i
++)
6053 SetPageReserved((page
+i
));
6054 pfn
+= (1 << order
);
6056 spin_unlock_irqrestore(&zone
->lock
, flags
);
6060 #ifdef CONFIG_MEMORY_FAILURE
6061 bool is_free_buddy_page(struct page
*page
)
6063 struct zone
*zone
= page_zone(page
);
6064 unsigned long pfn
= page_to_pfn(page
);
6065 unsigned long flags
;
6068 spin_lock_irqsave(&zone
->lock
, flags
);
6069 for (order
= 0; order
< MAX_ORDER
; order
++) {
6070 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6072 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6075 spin_unlock_irqrestore(&zone
->lock
, flags
);
6077 return order
< MAX_ORDER
;
6081 static const struct trace_print_flags pageflag_names
[] = {
6082 {1UL << PG_locked
, "locked" },
6083 {1UL << PG_error
, "error" },
6084 {1UL << PG_referenced
, "referenced" },
6085 {1UL << PG_uptodate
, "uptodate" },
6086 {1UL << PG_dirty
, "dirty" },
6087 {1UL << PG_lru
, "lru" },
6088 {1UL << PG_active
, "active" },
6089 {1UL << PG_slab
, "slab" },
6090 {1UL << PG_owner_priv_1
, "owner_priv_1" },
6091 {1UL << PG_arch_1
, "arch_1" },
6092 {1UL << PG_reserved
, "reserved" },
6093 {1UL << PG_private
, "private" },
6094 {1UL << PG_private_2
, "private_2" },
6095 {1UL << PG_writeback
, "writeback" },
6096 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6097 {1UL << PG_head
, "head" },
6098 {1UL << PG_tail
, "tail" },
6100 {1UL << PG_compound
, "compound" },
6102 {1UL << PG_swapcache
, "swapcache" },
6103 {1UL << PG_mappedtodisk
, "mappedtodisk" },
6104 {1UL << PG_reclaim
, "reclaim" },
6105 {1UL << PG_swapbacked
, "swapbacked" },
6106 {1UL << PG_unevictable
, "unevictable" },
6108 {1UL << PG_mlocked
, "mlocked" },
6110 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6111 {1UL << PG_uncached
, "uncached" },
6113 #ifdef CONFIG_MEMORY_FAILURE
6114 {1UL << PG_hwpoison
, "hwpoison" },
6116 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6117 {1UL << PG_compound_lock
, "compound_lock" },
6121 static void dump_page_flags(unsigned long flags
)
6123 const char *delim
= "";
6127 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names
) != __NR_PAGEFLAGS
);
6129 printk(KERN_ALERT
"page flags: %#lx(", flags
);
6131 /* remove zone id */
6132 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
6134 for (i
= 0; i
< ARRAY_SIZE(pageflag_names
) && flags
; i
++) {
6136 mask
= pageflag_names
[i
].mask
;
6137 if ((flags
& mask
) != mask
)
6141 printk("%s%s", delim
, pageflag_names
[i
].name
);
6145 /* check for left over flags */
6147 printk("%s%#lx", delim
, flags
);
6152 void dump_page(struct page
*page
)
6155 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6156 page
, atomic_read(&page
->_count
), page_mapcount(page
),
6157 page
->mapping
, page
->index
);
6158 dump_page_flags(page
->flags
);
6159 mem_cgroup_print_bad_page(page
);