2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/notifier.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/stop_machine.h>
46 #include <linux/sort.h>
47 #include <linux/pfn.h>
48 #include <linux/backing-dev.h>
49 #include <linux/fault-inject.h>
50 #include <linux/page-isolation.h>
51 #include <linux/page_cgroup.h>
52 #include <linux/debugobjects.h>
53 #include <linux/kmemleak.h>
54 #include <linux/compaction.h>
55 #include <trace/events/kmem.h>
56 #include <linux/ftrace_event.h>
57 #include <linux/memcontrol.h>
58 #include <linux/prefetch.h>
59 #include <linux/migrate.h>
60 #include <linux/page-debug-flags.h>
61 #include <linux/hugetlb.h>
62 #include <linux/sched/rt.h>
64 #include <asm/sections.h>
65 #include <asm/tlbflush.h>
66 #include <asm/div64.h>
69 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
70 static DEFINE_MUTEX(pcp_batch_high_lock
);
72 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
73 DEFINE_PER_CPU(int, numa_node
);
74 EXPORT_PER_CPU_SYMBOL(numa_node
);
77 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
79 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
80 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
81 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
82 * defined in <linux/topology.h>.
84 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
85 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
89 * Array of node states.
91 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
92 [N_POSSIBLE
] = NODE_MASK_ALL
,
93 [N_ONLINE
] = { { [0] = 1UL } },
95 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
97 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
99 #ifdef CONFIG_MOVABLE_NODE
100 [N_MEMORY
] = { { [0] = 1UL } },
102 [N_CPU
] = { { [0] = 1UL } },
105 EXPORT_SYMBOL(node_states
);
107 /* Protect totalram_pages and zone->managed_pages */
108 static DEFINE_SPINLOCK(managed_page_count_lock
);
110 unsigned long totalram_pages __read_mostly
;
111 unsigned long totalreserve_pages __read_mostly
;
113 * When calculating the number of globally allowed dirty pages, there
114 * is a certain number of per-zone reserves that should not be
115 * considered dirtyable memory. This is the sum of those reserves
116 * over all existing zones that contribute dirtyable memory.
118 unsigned long dirty_balance_reserve __read_mostly
;
120 int percpu_pagelist_fraction
;
121 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
123 #ifdef CONFIG_PM_SLEEP
125 * The following functions are used by the suspend/hibernate code to temporarily
126 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
127 * while devices are suspended. To avoid races with the suspend/hibernate code,
128 * they should always be called with pm_mutex held (gfp_allowed_mask also should
129 * only be modified with pm_mutex held, unless the suspend/hibernate code is
130 * guaranteed not to run in parallel with that modification).
133 static gfp_t saved_gfp_mask
;
135 void pm_restore_gfp_mask(void)
137 WARN_ON(!mutex_is_locked(&pm_mutex
));
138 if (saved_gfp_mask
) {
139 gfp_allowed_mask
= saved_gfp_mask
;
144 void pm_restrict_gfp_mask(void)
146 WARN_ON(!mutex_is_locked(&pm_mutex
));
147 WARN_ON(saved_gfp_mask
);
148 saved_gfp_mask
= gfp_allowed_mask
;
149 gfp_allowed_mask
&= ~GFP_IOFS
;
152 bool pm_suspended_storage(void)
154 if ((gfp_allowed_mask
& GFP_IOFS
) == GFP_IOFS
)
158 #endif /* CONFIG_PM_SLEEP */
160 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
161 int pageblock_order __read_mostly
;
164 static void __free_pages_ok(struct page
*page
, unsigned int order
);
167 * results with 256, 32 in the lowmem_reserve sysctl:
168 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
169 * 1G machine -> (16M dma, 784M normal, 224M high)
170 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
171 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
172 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
174 * TBD: should special case ZONE_DMA32 machines here - in those we normally
175 * don't need any ZONE_NORMAL reservation
177 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
178 #ifdef CONFIG_ZONE_DMA
181 #ifdef CONFIG_ZONE_DMA32
184 #ifdef CONFIG_HIGHMEM
190 EXPORT_SYMBOL(totalram_pages
);
192 static char * const zone_names
[MAX_NR_ZONES
] = {
193 #ifdef CONFIG_ZONE_DMA
196 #ifdef CONFIG_ZONE_DMA32
200 #ifdef CONFIG_HIGHMEM
206 int min_free_kbytes
= 1024;
207 int user_min_free_kbytes
;
209 static unsigned long __meminitdata nr_kernel_pages
;
210 static unsigned long __meminitdata nr_all_pages
;
211 static unsigned long __meminitdata dma_reserve
;
213 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
214 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
215 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
216 static unsigned long __initdata required_kernelcore
;
217 static unsigned long __initdata required_movablecore
;
218 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
220 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
222 EXPORT_SYMBOL(movable_zone
);
223 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
226 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
227 int nr_online_nodes __read_mostly
= 1;
228 EXPORT_SYMBOL(nr_node_ids
);
229 EXPORT_SYMBOL(nr_online_nodes
);
232 int page_group_by_mobility_disabled __read_mostly
;
234 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
237 if (unlikely(page_group_by_mobility_disabled
))
238 migratetype
= MIGRATE_UNMOVABLE
;
240 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
241 PB_migrate
, PB_migrate_end
);
244 bool oom_killer_disabled __read_mostly
;
246 #ifdef CONFIG_DEBUG_VM
247 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
251 unsigned long pfn
= page_to_pfn(page
);
252 unsigned long sp
, start_pfn
;
255 seq
= zone_span_seqbegin(zone
);
256 start_pfn
= zone
->zone_start_pfn
;
257 sp
= zone
->spanned_pages
;
258 if (!zone_spans_pfn(zone
, pfn
))
260 } while (zone_span_seqretry(zone
, seq
));
263 pr_err("page %lu outside zone [ %lu - %lu ]\n",
264 pfn
, start_pfn
, start_pfn
+ sp
);
269 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
271 if (!pfn_valid_within(page_to_pfn(page
)))
273 if (zone
!= page_zone(page
))
279 * Temporary debugging check for pages not lying within a given zone.
281 static int bad_range(struct zone
*zone
, struct page
*page
)
283 if (page_outside_zone_boundaries(zone
, page
))
285 if (!page_is_consistent(zone
, page
))
291 static inline int bad_range(struct zone
*zone
, struct page
*page
)
297 static void bad_page(struct page
*page
)
299 static unsigned long resume
;
300 static unsigned long nr_shown
;
301 static unsigned long nr_unshown
;
303 /* Don't complain about poisoned pages */
304 if (PageHWPoison(page
)) {
305 page_mapcount_reset(page
); /* remove PageBuddy */
310 * Allow a burst of 60 reports, then keep quiet for that minute;
311 * or allow a steady drip of one report per second.
313 if (nr_shown
== 60) {
314 if (time_before(jiffies
, resume
)) {
320 "BUG: Bad page state: %lu messages suppressed\n",
327 resume
= jiffies
+ 60 * HZ
;
329 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
330 current
->comm
, page_to_pfn(page
));
336 /* Leave bad fields for debug, except PageBuddy could make trouble */
337 page_mapcount_reset(page
); /* remove PageBuddy */
338 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
342 * Higher-order pages are called "compound pages". They are structured thusly:
344 * The first PAGE_SIZE page is called the "head page".
346 * The remaining PAGE_SIZE pages are called "tail pages".
348 * All pages have PG_compound set. All tail pages have their ->first_page
349 * pointing at the head page.
351 * The first tail page's ->lru.next holds the address of the compound page's
352 * put_page() function. Its ->lru.prev holds the order of allocation.
353 * This usage means that zero-order pages may not be compound.
356 static void free_compound_page(struct page
*page
)
358 __free_pages_ok(page
, compound_order(page
));
361 void prep_compound_page(struct page
*page
, unsigned long order
)
364 int nr_pages
= 1 << order
;
366 set_compound_page_dtor(page
, free_compound_page
);
367 set_compound_order(page
, order
);
369 for (i
= 1; i
< nr_pages
; i
++) {
370 struct page
*p
= page
+ i
;
372 set_page_count(p
, 0);
373 p
->first_page
= page
;
377 /* update __split_huge_page_refcount if you change this function */
378 static int destroy_compound_page(struct page
*page
, unsigned long order
)
381 int nr_pages
= 1 << order
;
384 if (unlikely(compound_order(page
) != order
)) {
389 __ClearPageHead(page
);
391 for (i
= 1; i
< nr_pages
; i
++) {
392 struct page
*p
= page
+ i
;
394 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
404 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
409 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
410 * and __GFP_HIGHMEM from hard or soft interrupt context.
412 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
413 for (i
= 0; i
< (1 << order
); i
++)
414 clear_highpage(page
+ i
);
417 #ifdef CONFIG_DEBUG_PAGEALLOC
418 unsigned int _debug_guardpage_minorder
;
420 static int __init
debug_guardpage_minorder_setup(char *buf
)
424 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
425 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
428 _debug_guardpage_minorder
= res
;
429 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
432 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
434 static inline void set_page_guard_flag(struct page
*page
)
436 __set_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
439 static inline void clear_page_guard_flag(struct page
*page
)
441 __clear_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
444 static inline void set_page_guard_flag(struct page
*page
) { }
445 static inline void clear_page_guard_flag(struct page
*page
) { }
448 static inline void set_page_order(struct page
*page
, int order
)
450 set_page_private(page
, order
);
451 __SetPageBuddy(page
);
454 static inline void rmv_page_order(struct page
*page
)
456 __ClearPageBuddy(page
);
457 set_page_private(page
, 0);
461 * Locate the struct page for both the matching buddy in our
462 * pair (buddy1) and the combined O(n+1) page they form (page).
464 * 1) Any buddy B1 will have an order O twin B2 which satisfies
465 * the following equation:
467 * For example, if the starting buddy (buddy2) is #8 its order
469 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
471 * 2) Any buddy B will have an order O+1 parent P which
472 * satisfies the following equation:
475 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
477 static inline unsigned long
478 __find_buddy_index(unsigned long page_idx
, unsigned int order
)
480 return page_idx
^ (1 << order
);
484 * This function checks whether a page is free && is the buddy
485 * we can do coalesce a page and its buddy if
486 * (a) the buddy is not in a hole &&
487 * (b) the buddy is in the buddy system &&
488 * (c) a page and its buddy have the same order &&
489 * (d) a page and its buddy are in the same zone.
491 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
492 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
494 * For recording page's order, we use page_private(page).
496 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
499 if (!pfn_valid_within(page_to_pfn(buddy
)))
502 if (page_zone_id(page
) != page_zone_id(buddy
))
505 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
506 VM_BUG_ON(page_count(buddy
) != 0);
510 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
511 VM_BUG_ON(page_count(buddy
) != 0);
518 * Freeing function for a buddy system allocator.
520 * The concept of a buddy system is to maintain direct-mapped table
521 * (containing bit values) for memory blocks of various "orders".
522 * The bottom level table contains the map for the smallest allocatable
523 * units of memory (here, pages), and each level above it describes
524 * pairs of units from the levels below, hence, "buddies".
525 * At a high level, all that happens here is marking the table entry
526 * at the bottom level available, and propagating the changes upward
527 * as necessary, plus some accounting needed to play nicely with other
528 * parts of the VM system.
529 * At each level, we keep a list of pages, which are heads of continuous
530 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
531 * order is recorded in page_private(page) field.
532 * So when we are allocating or freeing one, we can derive the state of the
533 * other. That is, if we allocate a small block, and both were
534 * free, the remainder of the region must be split into blocks.
535 * If a block is freed, and its buddy is also free, then this
536 * triggers coalescing into a block of larger size.
541 static inline void __free_one_page(struct page
*page
,
542 struct zone
*zone
, unsigned int order
,
545 unsigned long page_idx
;
546 unsigned long combined_idx
;
547 unsigned long uninitialized_var(buddy_idx
);
550 VM_BUG_ON(!zone_is_initialized(zone
));
552 if (unlikely(PageCompound(page
)))
553 if (unlikely(destroy_compound_page(page
, order
)))
556 VM_BUG_ON(migratetype
== -1);
558 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
560 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
561 VM_BUG_ON(bad_range(zone
, page
));
563 while (order
< MAX_ORDER
-1) {
564 buddy_idx
= __find_buddy_index(page_idx
, order
);
565 buddy
= page
+ (buddy_idx
- page_idx
);
566 if (!page_is_buddy(page
, buddy
, order
))
569 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
570 * merge with it and move up one order.
572 if (page_is_guard(buddy
)) {
573 clear_page_guard_flag(buddy
);
574 set_page_private(page
, 0);
575 __mod_zone_freepage_state(zone
, 1 << order
,
578 list_del(&buddy
->lru
);
579 zone
->free_area
[order
].nr_free
--;
580 rmv_page_order(buddy
);
582 combined_idx
= buddy_idx
& page_idx
;
583 page
= page
+ (combined_idx
- page_idx
);
584 page_idx
= combined_idx
;
587 set_page_order(page
, order
);
590 * If this is not the largest possible page, check if the buddy
591 * of the next-highest order is free. If it is, it's possible
592 * that pages are being freed that will coalesce soon. In case,
593 * that is happening, add the free page to the tail of the list
594 * so it's less likely to be used soon and more likely to be merged
595 * as a higher order page
597 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
598 struct page
*higher_page
, *higher_buddy
;
599 combined_idx
= buddy_idx
& page_idx
;
600 higher_page
= page
+ (combined_idx
- page_idx
);
601 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
602 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
603 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
604 list_add_tail(&page
->lru
,
605 &zone
->free_area
[order
].free_list
[migratetype
]);
610 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
612 zone
->free_area
[order
].nr_free
++;
615 static inline int free_pages_check(struct page
*page
)
617 if (unlikely(page_mapcount(page
) |
618 (page
->mapping
!= NULL
) |
619 (atomic_read(&page
->_count
) != 0) |
620 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
) |
621 (mem_cgroup_bad_page_check(page
)))) {
625 page_nid_reset_last(page
);
626 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
627 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
632 * Frees a number of pages from the PCP lists
633 * Assumes all pages on list are in same zone, and of same order.
634 * count is the number of pages to free.
636 * If the zone was previously in an "all pages pinned" state then look to
637 * see if this freeing clears that state.
639 * And clear the zone's pages_scanned counter, to hold off the "all pages are
640 * pinned" detection logic.
642 static void free_pcppages_bulk(struct zone
*zone
, int count
,
643 struct per_cpu_pages
*pcp
)
649 spin_lock(&zone
->lock
);
650 zone
->all_unreclaimable
= 0;
651 zone
->pages_scanned
= 0;
655 struct list_head
*list
;
658 * Remove pages from lists in a round-robin fashion. A
659 * batch_free count is maintained that is incremented when an
660 * empty list is encountered. This is so more pages are freed
661 * off fuller lists instead of spinning excessively around empty
666 if (++migratetype
== MIGRATE_PCPTYPES
)
668 list
= &pcp
->lists
[migratetype
];
669 } while (list_empty(list
));
671 /* This is the only non-empty list. Free them all. */
672 if (batch_free
== MIGRATE_PCPTYPES
)
673 batch_free
= to_free
;
676 int mt
; /* migratetype of the to-be-freed page */
678 page
= list_entry(list
->prev
, struct page
, lru
);
679 /* must delete as __free_one_page list manipulates */
680 list_del(&page
->lru
);
681 mt
= get_freepage_migratetype(page
);
682 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
683 __free_one_page(page
, zone
, 0, mt
);
684 trace_mm_page_pcpu_drain(page
, 0, mt
);
685 if (likely(!is_migrate_isolate_page(page
))) {
686 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1);
687 if (is_migrate_cma(mt
))
688 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
, 1);
690 } while (--to_free
&& --batch_free
&& !list_empty(list
));
692 spin_unlock(&zone
->lock
);
695 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
698 spin_lock(&zone
->lock
);
699 zone
->all_unreclaimable
= 0;
700 zone
->pages_scanned
= 0;
702 __free_one_page(page
, zone
, order
, migratetype
);
703 if (unlikely(!is_migrate_isolate(migratetype
)))
704 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
705 spin_unlock(&zone
->lock
);
708 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
713 trace_mm_page_free(page
, order
);
714 kmemcheck_free_shadow(page
, order
);
717 page
->mapping
= NULL
;
718 for (i
= 0; i
< (1 << order
); i
++)
719 bad
+= free_pages_check(page
+ i
);
723 if (!PageHighMem(page
)) {
724 debug_check_no_locks_freed(page_address(page
),
726 debug_check_no_obj_freed(page_address(page
),
729 arch_free_page(page
, order
);
730 kernel_map_pages(page
, 1 << order
, 0);
735 static void __free_pages_ok(struct page
*page
, unsigned int order
)
740 if (!free_pages_prepare(page
, order
))
743 local_irq_save(flags
);
744 __count_vm_events(PGFREE
, 1 << order
);
745 migratetype
= get_pageblock_migratetype(page
);
746 set_freepage_migratetype(page
, migratetype
);
747 free_one_page(page_zone(page
), page
, order
, migratetype
);
748 local_irq_restore(flags
);
751 void __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
753 unsigned int nr_pages
= 1 << order
;
754 struct page
*p
= page
;
758 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
760 __ClearPageReserved(p
);
761 set_page_count(p
, 0);
763 __ClearPageReserved(p
);
764 set_page_count(p
, 0);
766 page_zone(page
)->managed_pages
+= nr_pages
;
767 set_page_refcounted(page
);
768 __free_pages(page
, order
);
772 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
773 void __init
init_cma_reserved_pageblock(struct page
*page
)
775 unsigned i
= pageblock_nr_pages
;
776 struct page
*p
= page
;
779 __ClearPageReserved(p
);
780 set_page_count(p
, 0);
783 set_page_refcounted(page
);
784 set_pageblock_migratetype(page
, MIGRATE_CMA
);
785 __free_pages(page
, pageblock_order
);
786 adjust_managed_page_count(page
, pageblock_nr_pages
);
791 * The order of subdivision here is critical for the IO subsystem.
792 * Please do not alter this order without good reasons and regression
793 * testing. Specifically, as large blocks of memory are subdivided,
794 * the order in which smaller blocks are delivered depends on the order
795 * they're subdivided in this function. This is the primary factor
796 * influencing the order in which pages are delivered to the IO
797 * subsystem according to empirical testing, and this is also justified
798 * by considering the behavior of a buddy system containing a single
799 * large block of memory acted on by a series of small allocations.
800 * This behavior is a critical factor in sglist merging's success.
804 static inline void expand(struct zone
*zone
, struct page
*page
,
805 int low
, int high
, struct free_area
*area
,
808 unsigned long size
= 1 << high
;
814 VM_BUG_ON(bad_range(zone
, &page
[size
]));
816 #ifdef CONFIG_DEBUG_PAGEALLOC
817 if (high
< debug_guardpage_minorder()) {
819 * Mark as guard pages (or page), that will allow to
820 * merge back to allocator when buddy will be freed.
821 * Corresponding page table entries will not be touched,
822 * pages will stay not present in virtual address space
824 INIT_LIST_HEAD(&page
[size
].lru
);
825 set_page_guard_flag(&page
[size
]);
826 set_page_private(&page
[size
], high
);
827 /* Guard pages are not available for any usage */
828 __mod_zone_freepage_state(zone
, -(1 << high
),
833 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
835 set_page_order(&page
[size
], high
);
840 * This page is about to be returned from the page allocator
842 static inline int check_new_page(struct page
*page
)
844 if (unlikely(page_mapcount(page
) |
845 (page
->mapping
!= NULL
) |
846 (atomic_read(&page
->_count
) != 0) |
847 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
) |
848 (mem_cgroup_bad_page_check(page
)))) {
855 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
859 for (i
= 0; i
< (1 << order
); i
++) {
860 struct page
*p
= page
+ i
;
861 if (unlikely(check_new_page(p
)))
865 set_page_private(page
, 0);
866 set_page_refcounted(page
);
868 arch_alloc_page(page
, order
);
869 kernel_map_pages(page
, 1 << order
, 1);
871 if (gfp_flags
& __GFP_ZERO
)
872 prep_zero_page(page
, order
, gfp_flags
);
874 if (order
&& (gfp_flags
& __GFP_COMP
))
875 prep_compound_page(page
, order
);
881 * Go through the free lists for the given migratetype and remove
882 * the smallest available page from the freelists
885 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
888 unsigned int current_order
;
889 struct free_area
*area
;
892 /* Find a page of the appropriate size in the preferred list */
893 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
894 area
= &(zone
->free_area
[current_order
]);
895 if (list_empty(&area
->free_list
[migratetype
]))
898 page
= list_entry(area
->free_list
[migratetype
].next
,
900 list_del(&page
->lru
);
901 rmv_page_order(page
);
903 expand(zone
, page
, order
, current_order
, area
, migratetype
);
912 * This array describes the order lists are fallen back to when
913 * the free lists for the desirable migrate type are depleted
915 static int fallbacks
[MIGRATE_TYPES
][4] = {
916 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
917 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
919 [MIGRATE_MOVABLE
] = { MIGRATE_CMA
, MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
920 [MIGRATE_CMA
] = { MIGRATE_RESERVE
}, /* Never used */
922 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
924 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
}, /* Never used */
925 #ifdef CONFIG_MEMORY_ISOLATION
926 [MIGRATE_ISOLATE
] = { MIGRATE_RESERVE
}, /* Never used */
931 * Move the free pages in a range to the free lists of the requested type.
932 * Note that start_page and end_pages are not aligned on a pageblock
933 * boundary. If alignment is required, use move_freepages_block()
935 int move_freepages(struct zone
*zone
,
936 struct page
*start_page
, struct page
*end_page
,
943 #ifndef CONFIG_HOLES_IN_ZONE
945 * page_zone is not safe to call in this context when
946 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
947 * anyway as we check zone boundaries in move_freepages_block().
948 * Remove at a later date when no bug reports exist related to
949 * grouping pages by mobility
951 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
954 for (page
= start_page
; page
<= end_page
;) {
955 /* Make sure we are not inadvertently changing nodes */
956 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
958 if (!pfn_valid_within(page_to_pfn(page
))) {
963 if (!PageBuddy(page
)) {
968 order
= page_order(page
);
969 list_move(&page
->lru
,
970 &zone
->free_area
[order
].free_list
[migratetype
]);
971 set_freepage_migratetype(page
, migratetype
);
973 pages_moved
+= 1 << order
;
979 int move_freepages_block(struct zone
*zone
, struct page
*page
,
982 unsigned long start_pfn
, end_pfn
;
983 struct page
*start_page
, *end_page
;
985 start_pfn
= page_to_pfn(page
);
986 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
987 start_page
= pfn_to_page(start_pfn
);
988 end_page
= start_page
+ pageblock_nr_pages
- 1;
989 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
991 /* Do not cross zone boundaries */
992 if (!zone_spans_pfn(zone
, start_pfn
))
994 if (!zone_spans_pfn(zone
, end_pfn
))
997 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1000 static void change_pageblock_range(struct page
*pageblock_page
,
1001 int start_order
, int migratetype
)
1003 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1005 while (nr_pageblocks
--) {
1006 set_pageblock_migratetype(pageblock_page
, migratetype
);
1007 pageblock_page
+= pageblock_nr_pages
;
1012 * If breaking a large block of pages, move all free pages to the preferred
1013 * allocation list. If falling back for a reclaimable kernel allocation, be
1014 * more aggressive about taking ownership of free pages.
1016 * On the other hand, never change migration type of MIGRATE_CMA pageblocks
1017 * nor move CMA pages to different free lists. We don't want unmovable pages
1018 * to be allocated from MIGRATE_CMA areas.
1020 * Returns the new migratetype of the pageblock (or the same old migratetype
1021 * if it was unchanged).
1023 static int try_to_steal_freepages(struct zone
*zone
, struct page
*page
,
1024 int start_type
, int fallback_type
)
1026 int current_order
= page_order(page
);
1028 if (is_migrate_cma(fallback_type
))
1029 return fallback_type
;
1031 /* Take ownership for orders >= pageblock_order */
1032 if (current_order
>= pageblock_order
) {
1033 change_pageblock_range(page
, current_order
, start_type
);
1037 if (current_order
>= pageblock_order
/ 2 ||
1038 start_type
== MIGRATE_RECLAIMABLE
||
1039 page_group_by_mobility_disabled
) {
1042 pages
= move_freepages_block(zone
, page
, start_type
);
1044 /* Claim the whole block if over half of it is free */
1045 if (pages
>= (1 << (pageblock_order
-1)) ||
1046 page_group_by_mobility_disabled
) {
1048 set_pageblock_migratetype(page
, start_type
);
1054 return fallback_type
;
1057 /* Remove an element from the buddy allocator from the fallback list */
1058 static inline struct page
*
1059 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
1061 struct free_area
*area
;
1064 int migratetype
, new_type
, i
;
1066 /* Find the largest possible block of pages in the other list */
1067 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
1070 migratetype
= fallbacks
[start_migratetype
][i
];
1072 /* MIGRATE_RESERVE handled later if necessary */
1073 if (migratetype
== MIGRATE_RESERVE
)
1076 area
= &(zone
->free_area
[current_order
]);
1077 if (list_empty(&area
->free_list
[migratetype
]))
1080 page
= list_entry(area
->free_list
[migratetype
].next
,
1084 new_type
= try_to_steal_freepages(zone
, page
,
1088 /* Remove the page from the freelists */
1089 list_del(&page
->lru
);
1090 rmv_page_order(page
);
1093 * Borrow the excess buddy pages as well, irrespective
1094 * of whether we stole freepages, or took ownership of
1095 * the pageblock or not.
1097 * Exception: When borrowing from MIGRATE_CMA, release
1098 * the excess buddy pages to CMA itself.
1100 expand(zone
, page
, order
, current_order
, area
,
1101 is_migrate_cma(migratetype
)
1102 ? migratetype
: start_migratetype
);
1104 trace_mm_page_alloc_extfrag(page
, order
,
1105 current_order
, start_migratetype
, migratetype
,
1106 new_type
== start_migratetype
);
1116 * Do the hard work of removing an element from the buddy allocator.
1117 * Call me with the zone->lock already held.
1119 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1125 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1127 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1128 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1131 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1132 * is used because __rmqueue_smallest is an inline function
1133 * and we want just one call site
1136 migratetype
= MIGRATE_RESERVE
;
1141 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1146 * Obtain a specified number of elements from the buddy allocator, all under
1147 * a single hold of the lock, for efficiency. Add them to the supplied list.
1148 * Returns the number of new pages which were placed at *list.
1150 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1151 unsigned long count
, struct list_head
*list
,
1152 int migratetype
, int cold
)
1154 int mt
= migratetype
, i
;
1156 spin_lock(&zone
->lock
);
1157 for (i
= 0; i
< count
; ++i
) {
1158 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1159 if (unlikely(page
== NULL
))
1163 * Split buddy pages returned by expand() are received here
1164 * in physical page order. The page is added to the callers and
1165 * list and the list head then moves forward. From the callers
1166 * perspective, the linked list is ordered by page number in
1167 * some conditions. This is useful for IO devices that can
1168 * merge IO requests if the physical pages are ordered
1171 if (likely(cold
== 0))
1172 list_add(&page
->lru
, list
);
1174 list_add_tail(&page
->lru
, list
);
1175 if (IS_ENABLED(CONFIG_CMA
)) {
1176 mt
= get_pageblock_migratetype(page
);
1177 if (!is_migrate_cma(mt
) && !is_migrate_isolate(mt
))
1180 set_freepage_migratetype(page
, mt
);
1182 if (is_migrate_cma(mt
))
1183 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1186 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1187 spin_unlock(&zone
->lock
);
1193 * Called from the vmstat counter updater to drain pagesets of this
1194 * currently executing processor on remote nodes after they have
1197 * Note that this function must be called with the thread pinned to
1198 * a single processor.
1200 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1202 unsigned long flags
;
1204 unsigned long batch
;
1206 local_irq_save(flags
);
1207 batch
= ACCESS_ONCE(pcp
->batch
);
1208 if (pcp
->count
>= batch
)
1211 to_drain
= pcp
->count
;
1213 free_pcppages_bulk(zone
, to_drain
, pcp
);
1214 pcp
->count
-= to_drain
;
1216 local_irq_restore(flags
);
1221 * Drain pages of the indicated processor.
1223 * The processor must either be the current processor and the
1224 * thread pinned to the current processor or a processor that
1227 static void drain_pages(unsigned int cpu
)
1229 unsigned long flags
;
1232 for_each_populated_zone(zone
) {
1233 struct per_cpu_pageset
*pset
;
1234 struct per_cpu_pages
*pcp
;
1236 local_irq_save(flags
);
1237 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1241 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1244 local_irq_restore(flags
);
1249 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1251 void drain_local_pages(void *arg
)
1253 drain_pages(smp_processor_id());
1257 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1259 * Note that this code is protected against sending an IPI to an offline
1260 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1261 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1262 * nothing keeps CPUs from showing up after we populated the cpumask and
1263 * before the call to on_each_cpu_mask().
1265 void drain_all_pages(void)
1268 struct per_cpu_pageset
*pcp
;
1272 * Allocate in the BSS so we wont require allocation in
1273 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1275 static cpumask_t cpus_with_pcps
;
1278 * We don't care about racing with CPU hotplug event
1279 * as offline notification will cause the notified
1280 * cpu to drain that CPU pcps and on_each_cpu_mask
1281 * disables preemption as part of its processing
1283 for_each_online_cpu(cpu
) {
1284 bool has_pcps
= false;
1285 for_each_populated_zone(zone
) {
1286 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1287 if (pcp
->pcp
.count
) {
1293 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1295 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1297 on_each_cpu_mask(&cpus_with_pcps
, drain_local_pages
, NULL
, 1);
1300 #ifdef CONFIG_HIBERNATION
1302 void mark_free_pages(struct zone
*zone
)
1304 unsigned long pfn
, max_zone_pfn
;
1305 unsigned long flags
;
1307 struct list_head
*curr
;
1309 if (!zone
->spanned_pages
)
1312 spin_lock_irqsave(&zone
->lock
, flags
);
1314 max_zone_pfn
= zone_end_pfn(zone
);
1315 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1316 if (pfn_valid(pfn
)) {
1317 struct page
*page
= pfn_to_page(pfn
);
1319 if (!swsusp_page_is_forbidden(page
))
1320 swsusp_unset_page_free(page
);
1323 for_each_migratetype_order(order
, t
) {
1324 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1327 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1328 for (i
= 0; i
< (1UL << order
); i
++)
1329 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1332 spin_unlock_irqrestore(&zone
->lock
, flags
);
1334 #endif /* CONFIG_PM */
1337 * Free a 0-order page
1338 * cold == 1 ? free a cold page : free a hot page
1340 void free_hot_cold_page(struct page
*page
, int cold
)
1342 struct zone
*zone
= page_zone(page
);
1343 struct per_cpu_pages
*pcp
;
1344 unsigned long flags
;
1347 if (!free_pages_prepare(page
, 0))
1350 migratetype
= get_pageblock_migratetype(page
);
1351 set_freepage_migratetype(page
, migratetype
);
1352 local_irq_save(flags
);
1353 __count_vm_event(PGFREE
);
1356 * We only track unmovable, reclaimable and movable on pcp lists.
1357 * Free ISOLATE pages back to the allocator because they are being
1358 * offlined but treat RESERVE as movable pages so we can get those
1359 * areas back if necessary. Otherwise, we may have to free
1360 * excessively into the page allocator
1362 if (migratetype
>= MIGRATE_PCPTYPES
) {
1363 if (unlikely(is_migrate_isolate(migratetype
))) {
1364 free_one_page(zone
, page
, 0, migratetype
);
1367 migratetype
= MIGRATE_MOVABLE
;
1370 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1372 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1374 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1376 if (pcp
->count
>= pcp
->high
) {
1377 unsigned long batch
= ACCESS_ONCE(pcp
->batch
);
1378 free_pcppages_bulk(zone
, batch
, pcp
);
1379 pcp
->count
-= batch
;
1383 local_irq_restore(flags
);
1387 * Free a list of 0-order pages
1389 void free_hot_cold_page_list(struct list_head
*list
, int cold
)
1391 struct page
*page
, *next
;
1393 list_for_each_entry_safe(page
, next
, list
, lru
) {
1394 trace_mm_page_free_batched(page
, cold
);
1395 free_hot_cold_page(page
, cold
);
1400 * split_page takes a non-compound higher-order page, and splits it into
1401 * n (1<<order) sub-pages: page[0..n]
1402 * Each sub-page must be freed individually.
1404 * Note: this is probably too low level an operation for use in drivers.
1405 * Please consult with lkml before using this in your driver.
1407 void split_page(struct page
*page
, unsigned int order
)
1411 VM_BUG_ON(PageCompound(page
));
1412 VM_BUG_ON(!page_count(page
));
1414 #ifdef CONFIG_KMEMCHECK
1416 * Split shadow pages too, because free(page[0]) would
1417 * otherwise free the whole shadow.
1419 if (kmemcheck_page_is_tracked(page
))
1420 split_page(virt_to_page(page
[0].shadow
), order
);
1423 for (i
= 1; i
< (1 << order
); i
++)
1424 set_page_refcounted(page
+ i
);
1426 EXPORT_SYMBOL_GPL(split_page
);
1428 static int __isolate_free_page(struct page
*page
, unsigned int order
)
1430 unsigned long watermark
;
1434 BUG_ON(!PageBuddy(page
));
1436 zone
= page_zone(page
);
1437 mt
= get_pageblock_migratetype(page
);
1439 if (!is_migrate_isolate(mt
)) {
1440 /* Obey watermarks as if the page was being allocated */
1441 watermark
= low_wmark_pages(zone
) + (1 << order
);
1442 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1445 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
1448 /* Remove page from free list */
1449 list_del(&page
->lru
);
1450 zone
->free_area
[order
].nr_free
--;
1451 rmv_page_order(page
);
1453 /* Set the pageblock if the isolated page is at least a pageblock */
1454 if (order
>= pageblock_order
- 1) {
1455 struct page
*endpage
= page
+ (1 << order
) - 1;
1456 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
1457 int mt
= get_pageblock_migratetype(page
);
1458 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
1459 set_pageblock_migratetype(page
,
1464 return 1UL << order
;
1468 * Similar to split_page except the page is already free. As this is only
1469 * being used for migration, the migratetype of the block also changes.
1470 * As this is called with interrupts disabled, the caller is responsible
1471 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1474 * Note: this is probably too low level an operation for use in drivers.
1475 * Please consult with lkml before using this in your driver.
1477 int split_free_page(struct page
*page
)
1482 order
= page_order(page
);
1484 nr_pages
= __isolate_free_page(page
, order
);
1488 /* Split into individual pages */
1489 set_page_refcounted(page
);
1490 split_page(page
, order
);
1495 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1496 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1500 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1501 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1504 unsigned long flags
;
1506 int cold
= !!(gfp_flags
& __GFP_COLD
);
1509 if (likely(order
== 0)) {
1510 struct per_cpu_pages
*pcp
;
1511 struct list_head
*list
;
1513 local_irq_save(flags
);
1514 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1515 list
= &pcp
->lists
[migratetype
];
1516 if (list_empty(list
)) {
1517 pcp
->count
+= rmqueue_bulk(zone
, 0,
1520 if (unlikely(list_empty(list
)))
1525 page
= list_entry(list
->prev
, struct page
, lru
);
1527 page
= list_entry(list
->next
, struct page
, lru
);
1529 list_del(&page
->lru
);
1532 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1534 * __GFP_NOFAIL is not to be used in new code.
1536 * All __GFP_NOFAIL callers should be fixed so that they
1537 * properly detect and handle allocation failures.
1539 * We most definitely don't want callers attempting to
1540 * allocate greater than order-1 page units with
1543 WARN_ON_ONCE(order
> 1);
1545 spin_lock_irqsave(&zone
->lock
, flags
);
1546 page
= __rmqueue(zone
, order
, migratetype
);
1547 spin_unlock(&zone
->lock
);
1550 __mod_zone_freepage_state(zone
, -(1 << order
),
1551 get_pageblock_migratetype(page
));
1554 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1555 zone_statistics(preferred_zone
, zone
, gfp_flags
);
1556 local_irq_restore(flags
);
1558 VM_BUG_ON(bad_range(zone
, page
));
1559 if (prep_new_page(page
, order
, gfp_flags
))
1564 local_irq_restore(flags
);
1568 #ifdef CONFIG_FAIL_PAGE_ALLOC
1571 struct fault_attr attr
;
1573 u32 ignore_gfp_highmem
;
1574 u32 ignore_gfp_wait
;
1576 } fail_page_alloc
= {
1577 .attr
= FAULT_ATTR_INITIALIZER
,
1578 .ignore_gfp_wait
= 1,
1579 .ignore_gfp_highmem
= 1,
1583 static int __init
setup_fail_page_alloc(char *str
)
1585 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1587 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1589 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1591 if (order
< fail_page_alloc
.min_order
)
1593 if (gfp_mask
& __GFP_NOFAIL
)
1595 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1597 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1600 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1603 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1605 static int __init
fail_page_alloc_debugfs(void)
1607 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1610 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
1611 &fail_page_alloc
.attr
);
1613 return PTR_ERR(dir
);
1615 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1616 &fail_page_alloc
.ignore_gfp_wait
))
1618 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1619 &fail_page_alloc
.ignore_gfp_highmem
))
1621 if (!debugfs_create_u32("min-order", mode
, dir
,
1622 &fail_page_alloc
.min_order
))
1627 debugfs_remove_recursive(dir
);
1632 late_initcall(fail_page_alloc_debugfs
);
1634 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1636 #else /* CONFIG_FAIL_PAGE_ALLOC */
1638 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1643 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1646 * Return true if free pages are above 'mark'. This takes into account the order
1647 * of the allocation.
1649 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1650 int classzone_idx
, int alloc_flags
, long free_pages
)
1652 /* free_pages my go negative - that's OK */
1654 long lowmem_reserve
= z
->lowmem_reserve
[classzone_idx
];
1658 free_pages
-= (1 << order
) - 1;
1659 if (alloc_flags
& ALLOC_HIGH
)
1661 if (alloc_flags
& ALLOC_HARDER
)
1664 /* If allocation can't use CMA areas don't use free CMA pages */
1665 if (!(alloc_flags
& ALLOC_CMA
))
1666 free_cma
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
1669 if (free_pages
- free_cma
<= min
+ lowmem_reserve
)
1671 for (o
= 0; o
< order
; o
++) {
1672 /* At the next order, this order's pages become unavailable */
1673 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1675 /* Require fewer higher order pages to be free */
1678 if (free_pages
<= min
)
1684 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1685 int classzone_idx
, int alloc_flags
)
1687 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1688 zone_page_state(z
, NR_FREE_PAGES
));
1691 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1692 int classzone_idx
, int alloc_flags
)
1694 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1696 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1697 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1699 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1705 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1706 * skip over zones that are not allowed by the cpuset, or that have
1707 * been recently (in last second) found to be nearly full. See further
1708 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1709 * that have to skip over a lot of full or unallowed zones.
1711 * If the zonelist cache is present in the passed in zonelist, then
1712 * returns a pointer to the allowed node mask (either the current
1713 * tasks mems_allowed, or node_states[N_MEMORY].)
1715 * If the zonelist cache is not available for this zonelist, does
1716 * nothing and returns NULL.
1718 * If the fullzones BITMAP in the zonelist cache is stale (more than
1719 * a second since last zap'd) then we zap it out (clear its bits.)
1721 * We hold off even calling zlc_setup, until after we've checked the
1722 * first zone in the zonelist, on the theory that most allocations will
1723 * be satisfied from that first zone, so best to examine that zone as
1724 * quickly as we can.
1726 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1728 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1729 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1731 zlc
= zonelist
->zlcache_ptr
;
1735 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1736 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1737 zlc
->last_full_zap
= jiffies
;
1740 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1741 &cpuset_current_mems_allowed
:
1742 &node_states
[N_MEMORY
];
1743 return allowednodes
;
1747 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1748 * if it is worth looking at further for free memory:
1749 * 1) Check that the zone isn't thought to be full (doesn't have its
1750 * bit set in the zonelist_cache fullzones BITMAP).
1751 * 2) Check that the zones node (obtained from the zonelist_cache
1752 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1753 * Return true (non-zero) if zone is worth looking at further, or
1754 * else return false (zero) if it is not.
1756 * This check -ignores- the distinction between various watermarks,
1757 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1758 * found to be full for any variation of these watermarks, it will
1759 * be considered full for up to one second by all requests, unless
1760 * we are so low on memory on all allowed nodes that we are forced
1761 * into the second scan of the zonelist.
1763 * In the second scan we ignore this zonelist cache and exactly
1764 * apply the watermarks to all zones, even it is slower to do so.
1765 * We are low on memory in the second scan, and should leave no stone
1766 * unturned looking for a free page.
1768 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1769 nodemask_t
*allowednodes
)
1771 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1772 int i
; /* index of *z in zonelist zones */
1773 int n
; /* node that zone *z is on */
1775 zlc
= zonelist
->zlcache_ptr
;
1779 i
= z
- zonelist
->_zonerefs
;
1782 /* This zone is worth trying if it is allowed but not full */
1783 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1787 * Given 'z' scanning a zonelist, set the corresponding bit in
1788 * zlc->fullzones, so that subsequent attempts to allocate a page
1789 * from that zone don't waste time re-examining it.
1791 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1793 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1794 int i
; /* index of *z in zonelist zones */
1796 zlc
= zonelist
->zlcache_ptr
;
1800 i
= z
- zonelist
->_zonerefs
;
1802 set_bit(i
, zlc
->fullzones
);
1806 * clear all zones full, called after direct reclaim makes progress so that
1807 * a zone that was recently full is not skipped over for up to a second
1809 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1811 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1813 zlc
= zonelist
->zlcache_ptr
;
1817 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1820 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1822 return node_isset(local_zone
->node
, zone
->zone_pgdat
->reclaim_nodes
);
1825 static void __paginginit
init_zone_allows_reclaim(int nid
)
1829 for_each_online_node(i
)
1830 if (node_distance(nid
, i
) <= RECLAIM_DISTANCE
)
1831 node_set(i
, NODE_DATA(nid
)->reclaim_nodes
);
1833 zone_reclaim_mode
= 1;
1836 #else /* CONFIG_NUMA */
1838 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1843 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1844 nodemask_t
*allowednodes
)
1849 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1853 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1857 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1862 static inline void init_zone_allows_reclaim(int nid
)
1865 #endif /* CONFIG_NUMA */
1868 * get_page_from_freelist goes through the zonelist trying to allocate
1871 static struct page
*
1872 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1873 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1874 struct zone
*preferred_zone
, int migratetype
)
1877 struct page
*page
= NULL
;
1880 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1881 int zlc_active
= 0; /* set if using zonelist_cache */
1882 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1884 classzone_idx
= zone_idx(preferred_zone
);
1887 * Scan zonelist, looking for a zone with enough free.
1888 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1890 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1891 high_zoneidx
, nodemask
) {
1894 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1895 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1897 if ((alloc_flags
& ALLOC_CPUSET
) &&
1898 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1900 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1901 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1904 * When allocating a page cache page for writing, we
1905 * want to get it from a zone that is within its dirty
1906 * limit, such that no single zone holds more than its
1907 * proportional share of globally allowed dirty pages.
1908 * The dirty limits take into account the zone's
1909 * lowmem reserves and high watermark so that kswapd
1910 * should be able to balance it without having to
1911 * write pages from its LRU list.
1913 * This may look like it could increase pressure on
1914 * lower zones by failing allocations in higher zones
1915 * before they are full. But the pages that do spill
1916 * over are limited as the lower zones are protected
1917 * by this very same mechanism. It should not become
1918 * a practical burden to them.
1920 * XXX: For now, allow allocations to potentially
1921 * exceed the per-zone dirty limit in the slowpath
1922 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1923 * which is important when on a NUMA setup the allowed
1924 * zones are together not big enough to reach the
1925 * global limit. The proper fix for these situations
1926 * will require awareness of zones in the
1927 * dirty-throttling and the flusher threads.
1929 if ((alloc_flags
& ALLOC_WMARK_LOW
) &&
1930 (gfp_mask
& __GFP_WRITE
) && !zone_dirty_ok(zone
))
1931 goto this_zone_full
;
1933 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1934 if (!zone_watermark_ok(zone
, order
, mark
,
1935 classzone_idx
, alloc_flags
)) {
1938 if (IS_ENABLED(CONFIG_NUMA
) &&
1939 !did_zlc_setup
&& nr_online_nodes
> 1) {
1941 * we do zlc_setup if there are multiple nodes
1942 * and before considering the first zone allowed
1945 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1950 if (zone_reclaim_mode
== 0 ||
1951 !zone_allows_reclaim(preferred_zone
, zone
))
1952 goto this_zone_full
;
1955 * As we may have just activated ZLC, check if the first
1956 * eligible zone has failed zone_reclaim recently.
1958 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1959 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1962 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1964 case ZONE_RECLAIM_NOSCAN
:
1967 case ZONE_RECLAIM_FULL
:
1968 /* scanned but unreclaimable */
1971 /* did we reclaim enough */
1972 if (zone_watermark_ok(zone
, order
, mark
,
1973 classzone_idx
, alloc_flags
))
1977 * Failed to reclaim enough to meet watermark.
1978 * Only mark the zone full if checking the min
1979 * watermark or if we failed to reclaim just
1980 * 1<<order pages or else the page allocator
1981 * fastpath will prematurely mark zones full
1982 * when the watermark is between the low and
1985 if (((alloc_flags
& ALLOC_WMARK_MASK
) == ALLOC_WMARK_MIN
) ||
1986 ret
== ZONE_RECLAIM_SOME
)
1987 goto this_zone_full
;
1994 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1995 gfp_mask
, migratetype
);
1999 if (IS_ENABLED(CONFIG_NUMA
))
2000 zlc_mark_zone_full(zonelist
, z
);
2003 if (unlikely(IS_ENABLED(CONFIG_NUMA
) && page
== NULL
&& zlc_active
)) {
2004 /* Disable zlc cache for second zonelist scan */
2011 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2012 * necessary to allocate the page. The expectation is
2013 * that the caller is taking steps that will free more
2014 * memory. The caller should avoid the page being used
2015 * for !PFMEMALLOC purposes.
2017 page
->pfmemalloc
= !!(alloc_flags
& ALLOC_NO_WATERMARKS
);
2023 * Large machines with many possible nodes should not always dump per-node
2024 * meminfo in irq context.
2026 static inline bool should_suppress_show_mem(void)
2031 ret
= in_interrupt();
2036 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2037 DEFAULT_RATELIMIT_INTERVAL
,
2038 DEFAULT_RATELIMIT_BURST
);
2040 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
2042 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2044 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2045 debug_guardpage_minorder() > 0)
2049 * Walking all memory to count page types is very expensive and should
2050 * be inhibited in non-blockable contexts.
2052 if (!(gfp_mask
& __GFP_WAIT
))
2053 filter
|= SHOW_MEM_FILTER_PAGE_COUNT
;
2056 * This documents exceptions given to allocations in certain
2057 * contexts that are allowed to allocate outside current's set
2060 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2061 if (test_thread_flag(TIF_MEMDIE
) ||
2062 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2063 filter
&= ~SHOW_MEM_FILTER_NODES
;
2064 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
2065 filter
&= ~SHOW_MEM_FILTER_NODES
;
2068 struct va_format vaf
;
2071 va_start(args
, fmt
);
2076 pr_warn("%pV", &vaf
);
2081 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2082 current
->comm
, order
, gfp_mask
);
2085 if (!should_suppress_show_mem())
2090 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
2091 unsigned long did_some_progress
,
2092 unsigned long pages_reclaimed
)
2094 /* Do not loop if specifically requested */
2095 if (gfp_mask
& __GFP_NORETRY
)
2098 /* Always retry if specifically requested */
2099 if (gfp_mask
& __GFP_NOFAIL
)
2103 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2104 * making forward progress without invoking OOM. Suspend also disables
2105 * storage devices so kswapd will not help. Bail if we are suspending.
2107 if (!did_some_progress
&& pm_suspended_storage())
2111 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2112 * means __GFP_NOFAIL, but that may not be true in other
2115 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
2119 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2120 * specified, then we retry until we no longer reclaim any pages
2121 * (above), or we've reclaimed an order of pages at least as
2122 * large as the allocation's order. In both cases, if the
2123 * allocation still fails, we stop retrying.
2125 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
2131 static inline struct page
*
2132 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2133 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2134 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2139 /* Acquire the OOM killer lock for the zones in zonelist */
2140 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
2141 schedule_timeout_uninterruptible(1);
2146 * Go through the zonelist yet one more time, keep very high watermark
2147 * here, this is only to catch a parallel oom killing, we must fail if
2148 * we're still under heavy pressure.
2150 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
2151 order
, zonelist
, high_zoneidx
,
2152 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
2153 preferred_zone
, migratetype
);
2157 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2158 /* The OOM killer will not help higher order allocs */
2159 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2161 /* The OOM killer does not needlessly kill tasks for lowmem */
2162 if (high_zoneidx
< ZONE_NORMAL
)
2165 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2166 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2167 * The caller should handle page allocation failure by itself if
2168 * it specifies __GFP_THISNODE.
2169 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2171 if (gfp_mask
& __GFP_THISNODE
)
2174 /* Exhausted what can be done so it's blamo time */
2175 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
, false);
2178 clear_zonelist_oom(zonelist
, gfp_mask
);
2182 #ifdef CONFIG_COMPACTION
2183 /* Try memory compaction for high-order allocations before reclaim */
2184 static struct page
*
2185 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2186 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2187 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2188 int migratetype
, bool sync_migration
,
2189 bool *contended_compaction
, bool *deferred_compaction
,
2190 unsigned long *did_some_progress
)
2195 if (compaction_deferred(preferred_zone
, order
)) {
2196 *deferred_compaction
= true;
2200 current
->flags
|= PF_MEMALLOC
;
2201 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
2202 nodemask
, sync_migration
,
2203 contended_compaction
);
2204 current
->flags
&= ~PF_MEMALLOC
;
2206 if (*did_some_progress
!= COMPACT_SKIPPED
) {
2209 /* Page migration frees to the PCP lists but we want merging */
2210 drain_pages(get_cpu());
2213 page
= get_page_from_freelist(gfp_mask
, nodemask
,
2214 order
, zonelist
, high_zoneidx
,
2215 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2216 preferred_zone
, migratetype
);
2218 preferred_zone
->compact_blockskip_flush
= false;
2219 preferred_zone
->compact_considered
= 0;
2220 preferred_zone
->compact_defer_shift
= 0;
2221 if (order
>= preferred_zone
->compact_order_failed
)
2222 preferred_zone
->compact_order_failed
= order
+ 1;
2223 count_vm_event(COMPACTSUCCESS
);
2228 * It's bad if compaction run occurs and fails.
2229 * The most likely reason is that pages exist,
2230 * but not enough to satisfy watermarks.
2232 count_vm_event(COMPACTFAIL
);
2235 * As async compaction considers a subset of pageblocks, only
2236 * defer if the failure was a sync compaction failure.
2239 defer_compaction(preferred_zone
, order
);
2247 static inline struct page
*
2248 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2249 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2250 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2251 int migratetype
, bool sync_migration
,
2252 bool *contended_compaction
, bool *deferred_compaction
,
2253 unsigned long *did_some_progress
)
2257 #endif /* CONFIG_COMPACTION */
2259 /* Perform direct synchronous page reclaim */
2261 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
, struct zonelist
*zonelist
,
2262 nodemask_t
*nodemask
)
2264 struct reclaim_state reclaim_state
;
2269 /* We now go into synchronous reclaim */
2270 cpuset_memory_pressure_bump();
2271 current
->flags
|= PF_MEMALLOC
;
2272 lockdep_set_current_reclaim_state(gfp_mask
);
2273 reclaim_state
.reclaimed_slab
= 0;
2274 current
->reclaim_state
= &reclaim_state
;
2276 progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
2278 current
->reclaim_state
= NULL
;
2279 lockdep_clear_current_reclaim_state();
2280 current
->flags
&= ~PF_MEMALLOC
;
2287 /* The really slow allocator path where we enter direct reclaim */
2288 static inline struct page
*
2289 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2290 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2291 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2292 int migratetype
, unsigned long *did_some_progress
)
2294 struct page
*page
= NULL
;
2295 bool drained
= false;
2297 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
2299 if (unlikely(!(*did_some_progress
)))
2302 /* After successful reclaim, reconsider all zones for allocation */
2303 if (IS_ENABLED(CONFIG_NUMA
))
2304 zlc_clear_zones_full(zonelist
);
2307 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2308 zonelist
, high_zoneidx
,
2309 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2310 preferred_zone
, migratetype
);
2313 * If an allocation failed after direct reclaim, it could be because
2314 * pages are pinned on the per-cpu lists. Drain them and try again
2316 if (!page
&& !drained
) {
2326 * This is called in the allocator slow-path if the allocation request is of
2327 * sufficient urgency to ignore watermarks and take other desperate measures
2329 static inline struct page
*
2330 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2331 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2332 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2338 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2339 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
2340 preferred_zone
, migratetype
);
2342 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2343 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2344 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2350 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
2351 enum zone_type high_zoneidx
,
2352 enum zone_type classzone_idx
)
2357 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
2358 wakeup_kswapd(zone
, order
, classzone_idx
);
2362 gfp_to_alloc_flags(gfp_t gfp_mask
)
2364 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2365 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2367 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2368 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2371 * The caller may dip into page reserves a bit more if the caller
2372 * cannot run direct reclaim, or if the caller has realtime scheduling
2373 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2374 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2376 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2380 * Not worth trying to allocate harder for
2381 * __GFP_NOMEMALLOC even if it can't schedule.
2383 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2384 alloc_flags
|= ALLOC_HARDER
;
2386 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2387 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2389 alloc_flags
&= ~ALLOC_CPUSET
;
2390 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2391 alloc_flags
|= ALLOC_HARDER
;
2393 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2394 if (gfp_mask
& __GFP_MEMALLOC
)
2395 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2396 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2397 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2398 else if (!in_interrupt() &&
2399 ((current
->flags
& PF_MEMALLOC
) ||
2400 unlikely(test_thread_flag(TIF_MEMDIE
))))
2401 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2404 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2405 alloc_flags
|= ALLOC_CMA
;
2410 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2412 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2415 static inline struct page
*
2416 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2417 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2418 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2421 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2422 struct page
*page
= NULL
;
2424 unsigned long pages_reclaimed
= 0;
2425 unsigned long did_some_progress
;
2426 bool sync_migration
= false;
2427 bool deferred_compaction
= false;
2428 bool contended_compaction
= false;
2431 * In the slowpath, we sanity check order to avoid ever trying to
2432 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2433 * be using allocators in order of preference for an area that is
2436 if (order
>= MAX_ORDER
) {
2437 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2442 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2443 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2444 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2445 * using a larger set of nodes after it has established that the
2446 * allowed per node queues are empty and that nodes are
2449 if (IS_ENABLED(CONFIG_NUMA
) &&
2450 (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2454 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2455 wake_all_kswapd(order
, zonelist
, high_zoneidx
,
2456 zone_idx(preferred_zone
));
2459 * OK, we're below the kswapd watermark and have kicked background
2460 * reclaim. Now things get more complex, so set up alloc_flags according
2461 * to how we want to proceed.
2463 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2466 * Find the true preferred zone if the allocation is unconstrained by
2469 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2470 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2474 /* This is the last chance, in general, before the goto nopage. */
2475 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2476 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2477 preferred_zone
, migratetype
);
2481 /* Allocate without watermarks if the context allows */
2482 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2484 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2485 * the allocation is high priority and these type of
2486 * allocations are system rather than user orientated
2488 zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
2490 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2491 zonelist
, high_zoneidx
, nodemask
,
2492 preferred_zone
, migratetype
);
2498 /* Atomic allocations - we can't balance anything */
2502 /* Avoid recursion of direct reclaim */
2503 if (current
->flags
& PF_MEMALLOC
)
2506 /* Avoid allocations with no watermarks from looping endlessly */
2507 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2511 * Try direct compaction. The first pass is asynchronous. Subsequent
2512 * attempts after direct reclaim are synchronous
2514 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2515 zonelist
, high_zoneidx
,
2517 alloc_flags
, preferred_zone
,
2518 migratetype
, sync_migration
,
2519 &contended_compaction
,
2520 &deferred_compaction
,
2521 &did_some_progress
);
2524 sync_migration
= true;
2527 * If compaction is deferred for high-order allocations, it is because
2528 * sync compaction recently failed. In this is the case and the caller
2529 * requested a movable allocation that does not heavily disrupt the
2530 * system then fail the allocation instead of entering direct reclaim.
2532 if ((deferred_compaction
|| contended_compaction
) &&
2533 (gfp_mask
& __GFP_NO_KSWAPD
))
2536 /* Try direct reclaim and then allocating */
2537 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2538 zonelist
, high_zoneidx
,
2540 alloc_flags
, preferred_zone
,
2541 migratetype
, &did_some_progress
);
2546 * If we failed to make any progress reclaiming, then we are
2547 * running out of options and have to consider going OOM
2549 if (!did_some_progress
) {
2550 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2551 if (oom_killer_disabled
)
2553 /* Coredumps can quickly deplete all memory reserves */
2554 if ((current
->flags
& PF_DUMPCORE
) &&
2555 !(gfp_mask
& __GFP_NOFAIL
))
2557 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2558 zonelist
, high_zoneidx
,
2559 nodemask
, preferred_zone
,
2564 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2566 * The oom killer is not called for high-order
2567 * allocations that may fail, so if no progress
2568 * is being made, there are no other options and
2569 * retrying is unlikely to help.
2571 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2574 * The oom killer is not called for lowmem
2575 * allocations to prevent needlessly killing
2578 if (high_zoneidx
< ZONE_NORMAL
)
2586 /* Check if we should retry the allocation */
2587 pages_reclaimed
+= did_some_progress
;
2588 if (should_alloc_retry(gfp_mask
, order
, did_some_progress
,
2590 /* Wait for some write requests to complete then retry */
2591 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2595 * High-order allocations do not necessarily loop after
2596 * direct reclaim and reclaim/compaction depends on compaction
2597 * being called after reclaim so call directly if necessary
2599 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2600 zonelist
, high_zoneidx
,
2602 alloc_flags
, preferred_zone
,
2603 migratetype
, sync_migration
,
2604 &contended_compaction
,
2605 &deferred_compaction
,
2606 &did_some_progress
);
2612 warn_alloc_failed(gfp_mask
, order
, NULL
);
2615 if (kmemcheck_enabled
)
2616 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2622 * This is the 'heart' of the zoned buddy allocator.
2625 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2626 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2628 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2629 struct zone
*preferred_zone
;
2630 struct page
*page
= NULL
;
2631 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2632 unsigned int cpuset_mems_cookie
;
2633 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
;
2634 struct mem_cgroup
*memcg
= NULL
;
2636 gfp_mask
&= gfp_allowed_mask
;
2638 lockdep_trace_alloc(gfp_mask
);
2640 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2642 if (should_fail_alloc_page(gfp_mask
, order
))
2646 * Check the zones suitable for the gfp_mask contain at least one
2647 * valid zone. It's possible to have an empty zonelist as a result
2648 * of GFP_THISNODE and a memoryless node
2650 if (unlikely(!zonelist
->_zonerefs
->zone
))
2654 * Will only have any effect when __GFP_KMEMCG is set. This is
2655 * verified in the (always inline) callee
2657 if (!memcg_kmem_newpage_charge(gfp_mask
, &memcg
, order
))
2661 cpuset_mems_cookie
= get_mems_allowed();
2663 /* The preferred zone is used for statistics later */
2664 first_zones_zonelist(zonelist
, high_zoneidx
,
2665 nodemask
? : &cpuset_current_mems_allowed
,
2667 if (!preferred_zone
)
2671 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2672 alloc_flags
|= ALLOC_CMA
;
2674 /* First allocation attempt */
2675 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2676 zonelist
, high_zoneidx
, alloc_flags
,
2677 preferred_zone
, migratetype
);
2678 if (unlikely(!page
)) {
2680 * Runtime PM, block IO and its error handling path
2681 * can deadlock because I/O on the device might not
2684 gfp_mask
= memalloc_noio_flags(gfp_mask
);
2685 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2686 zonelist
, high_zoneidx
, nodemask
,
2687 preferred_zone
, migratetype
);
2690 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2694 * When updating a task's mems_allowed, it is possible to race with
2695 * parallel threads in such a way that an allocation can fail while
2696 * the mask is being updated. If a page allocation is about to fail,
2697 * check if the cpuset changed during allocation and if so, retry.
2699 if (unlikely(!put_mems_allowed(cpuset_mems_cookie
) && !page
))
2702 memcg_kmem_commit_charge(page
, memcg
, order
);
2706 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2709 * Common helper functions.
2711 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2716 * __get_free_pages() returns a 32-bit address, which cannot represent
2719 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2721 page
= alloc_pages(gfp_mask
, order
);
2724 return (unsigned long) page_address(page
);
2726 EXPORT_SYMBOL(__get_free_pages
);
2728 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2730 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2732 EXPORT_SYMBOL(get_zeroed_page
);
2734 void __free_pages(struct page
*page
, unsigned int order
)
2736 if (put_page_testzero(page
)) {
2738 free_hot_cold_page(page
, 0);
2740 __free_pages_ok(page
, order
);
2744 EXPORT_SYMBOL(__free_pages
);
2746 void free_pages(unsigned long addr
, unsigned int order
)
2749 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2750 __free_pages(virt_to_page((void *)addr
), order
);
2754 EXPORT_SYMBOL(free_pages
);
2757 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2758 * pages allocated with __GFP_KMEMCG.
2760 * Those pages are accounted to a particular memcg, embedded in the
2761 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2762 * for that information only to find out that it is NULL for users who have no
2763 * interest in that whatsoever, we provide these functions.
2765 * The caller knows better which flags it relies on.
2767 void __free_memcg_kmem_pages(struct page
*page
, unsigned int order
)
2769 memcg_kmem_uncharge_pages(page
, order
);
2770 __free_pages(page
, order
);
2773 void free_memcg_kmem_pages(unsigned long addr
, unsigned int order
)
2776 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2777 __free_memcg_kmem_pages(virt_to_page((void *)addr
), order
);
2781 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2784 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2785 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2787 split_page(virt_to_page((void *)addr
), order
);
2788 while (used
< alloc_end
) {
2793 return (void *)addr
;
2797 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2798 * @size: the number of bytes to allocate
2799 * @gfp_mask: GFP flags for the allocation
2801 * This function is similar to alloc_pages(), except that it allocates the
2802 * minimum number of pages to satisfy the request. alloc_pages() can only
2803 * allocate memory in power-of-two pages.
2805 * This function is also limited by MAX_ORDER.
2807 * Memory allocated by this function must be released by free_pages_exact().
2809 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2811 unsigned int order
= get_order(size
);
2814 addr
= __get_free_pages(gfp_mask
, order
);
2815 return make_alloc_exact(addr
, order
, size
);
2817 EXPORT_SYMBOL(alloc_pages_exact
);
2820 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2822 * @nid: the preferred node ID where memory should be allocated
2823 * @size: the number of bytes to allocate
2824 * @gfp_mask: GFP flags for the allocation
2826 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2828 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2831 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2833 unsigned order
= get_order(size
);
2834 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2837 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2839 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2842 * free_pages_exact - release memory allocated via alloc_pages_exact()
2843 * @virt: the value returned by alloc_pages_exact.
2844 * @size: size of allocation, same value as passed to alloc_pages_exact().
2846 * Release the memory allocated by a previous call to alloc_pages_exact.
2848 void free_pages_exact(void *virt
, size_t size
)
2850 unsigned long addr
= (unsigned long)virt
;
2851 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2853 while (addr
< end
) {
2858 EXPORT_SYMBOL(free_pages_exact
);
2861 * nr_free_zone_pages - count number of pages beyond high watermark
2862 * @offset: The zone index of the highest zone
2864 * nr_free_zone_pages() counts the number of counts pages which are beyond the
2865 * high watermark within all zones at or below a given zone index. For each
2866 * zone, the number of pages is calculated as:
2867 * managed_pages - high_pages
2869 static unsigned long nr_free_zone_pages(int offset
)
2874 /* Just pick one node, since fallback list is circular */
2875 unsigned long sum
= 0;
2877 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2879 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2880 unsigned long size
= zone
->managed_pages
;
2881 unsigned long high
= high_wmark_pages(zone
);
2890 * nr_free_buffer_pages - count number of pages beyond high watermark
2892 * nr_free_buffer_pages() counts the number of pages which are beyond the high
2893 * watermark within ZONE_DMA and ZONE_NORMAL.
2895 unsigned long nr_free_buffer_pages(void)
2897 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2899 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2902 * nr_free_pagecache_pages - count number of pages beyond high watermark
2904 * nr_free_pagecache_pages() counts the number of pages which are beyond the
2905 * high watermark within all zones.
2907 unsigned long nr_free_pagecache_pages(void)
2909 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2912 static inline void show_node(struct zone
*zone
)
2914 if (IS_ENABLED(CONFIG_NUMA
))
2915 printk("Node %d ", zone_to_nid(zone
));
2918 void si_meminfo(struct sysinfo
*val
)
2920 val
->totalram
= totalram_pages
;
2922 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2923 val
->bufferram
= nr_blockdev_pages();
2924 val
->totalhigh
= totalhigh_pages
;
2925 val
->freehigh
= nr_free_highpages();
2926 val
->mem_unit
= PAGE_SIZE
;
2929 EXPORT_SYMBOL(si_meminfo
);
2932 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2934 int zone_type
; /* needs to be signed */
2935 unsigned long managed_pages
= 0;
2936 pg_data_t
*pgdat
= NODE_DATA(nid
);
2938 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
2939 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
2940 val
->totalram
= managed_pages
;
2941 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2942 #ifdef CONFIG_HIGHMEM
2943 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
2944 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2950 val
->mem_unit
= PAGE_SIZE
;
2955 * Determine whether the node should be displayed or not, depending on whether
2956 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2958 bool skip_free_areas_node(unsigned int flags
, int nid
)
2961 unsigned int cpuset_mems_cookie
;
2963 if (!(flags
& SHOW_MEM_FILTER_NODES
))
2967 cpuset_mems_cookie
= get_mems_allowed();
2968 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
2969 } while (!put_mems_allowed(cpuset_mems_cookie
));
2974 #define K(x) ((x) << (PAGE_SHIFT-10))
2976 static void show_migration_types(unsigned char type
)
2978 static const char types
[MIGRATE_TYPES
] = {
2979 [MIGRATE_UNMOVABLE
] = 'U',
2980 [MIGRATE_RECLAIMABLE
] = 'E',
2981 [MIGRATE_MOVABLE
] = 'M',
2982 [MIGRATE_RESERVE
] = 'R',
2984 [MIGRATE_CMA
] = 'C',
2986 #ifdef CONFIG_MEMORY_ISOLATION
2987 [MIGRATE_ISOLATE
] = 'I',
2990 char tmp
[MIGRATE_TYPES
+ 1];
2994 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
2995 if (type
& (1 << i
))
3000 printk("(%s) ", tmp
);
3004 * Show free area list (used inside shift_scroll-lock stuff)
3005 * We also calculate the percentage fragmentation. We do this by counting the
3006 * memory on each free list with the exception of the first item on the list.
3007 * Suppresses nodes that are not allowed by current's cpuset if
3008 * SHOW_MEM_FILTER_NODES is passed.
3010 void show_free_areas(unsigned int filter
)
3015 for_each_populated_zone(zone
) {
3016 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3019 printk("%s per-cpu:\n", zone
->name
);
3021 for_each_online_cpu(cpu
) {
3022 struct per_cpu_pageset
*pageset
;
3024 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
3026 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3027 cpu
, pageset
->pcp
.high
,
3028 pageset
->pcp
.batch
, pageset
->pcp
.count
);
3032 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3033 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3035 " dirty:%lu writeback:%lu unstable:%lu\n"
3036 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3037 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3039 global_page_state(NR_ACTIVE_ANON
),
3040 global_page_state(NR_INACTIVE_ANON
),
3041 global_page_state(NR_ISOLATED_ANON
),
3042 global_page_state(NR_ACTIVE_FILE
),
3043 global_page_state(NR_INACTIVE_FILE
),
3044 global_page_state(NR_ISOLATED_FILE
),
3045 global_page_state(NR_UNEVICTABLE
),
3046 global_page_state(NR_FILE_DIRTY
),
3047 global_page_state(NR_WRITEBACK
),
3048 global_page_state(NR_UNSTABLE_NFS
),
3049 global_page_state(NR_FREE_PAGES
),
3050 global_page_state(NR_SLAB_RECLAIMABLE
),
3051 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3052 global_page_state(NR_FILE_MAPPED
),
3053 global_page_state(NR_SHMEM
),
3054 global_page_state(NR_PAGETABLE
),
3055 global_page_state(NR_BOUNCE
),
3056 global_page_state(NR_FREE_CMA_PAGES
));
3058 for_each_populated_zone(zone
) {
3061 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3069 " active_anon:%lukB"
3070 " inactive_anon:%lukB"
3071 " active_file:%lukB"
3072 " inactive_file:%lukB"
3073 " unevictable:%lukB"
3074 " isolated(anon):%lukB"
3075 " isolated(file):%lukB"
3083 " slab_reclaimable:%lukB"
3084 " slab_unreclaimable:%lukB"
3085 " kernel_stack:%lukB"
3090 " writeback_tmp:%lukB"
3091 " pages_scanned:%lu"
3092 " all_unreclaimable? %s"
3095 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3096 K(min_wmark_pages(zone
)),
3097 K(low_wmark_pages(zone
)),
3098 K(high_wmark_pages(zone
)),
3099 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3100 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3101 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3102 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3103 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3104 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3105 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3106 K(zone
->present_pages
),
3107 K(zone
->managed_pages
),
3108 K(zone_page_state(zone
, NR_MLOCK
)),
3109 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3110 K(zone_page_state(zone
, NR_WRITEBACK
)),
3111 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3112 K(zone_page_state(zone
, NR_SHMEM
)),
3113 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3114 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3115 zone_page_state(zone
, NR_KERNEL_STACK
) *
3117 K(zone_page_state(zone
, NR_PAGETABLE
)),
3118 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3119 K(zone_page_state(zone
, NR_BOUNCE
)),
3120 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3121 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3122 zone
->pages_scanned
,
3123 (zone
->all_unreclaimable
? "yes" : "no")
3125 printk("lowmem_reserve[]:");
3126 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3127 printk(" %lu", zone
->lowmem_reserve
[i
]);
3131 for_each_populated_zone(zone
) {
3132 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
3133 unsigned char types
[MAX_ORDER
];
3135 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3138 printk("%s: ", zone
->name
);
3140 spin_lock_irqsave(&zone
->lock
, flags
);
3141 for (order
= 0; order
< MAX_ORDER
; order
++) {
3142 struct free_area
*area
= &zone
->free_area
[order
];
3145 nr
[order
] = area
->nr_free
;
3146 total
+= nr
[order
] << order
;
3149 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3150 if (!list_empty(&area
->free_list
[type
]))
3151 types
[order
] |= 1 << type
;
3154 spin_unlock_irqrestore(&zone
->lock
, flags
);
3155 for (order
= 0; order
< MAX_ORDER
; order
++) {
3156 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3158 show_migration_types(types
[order
]);
3160 printk("= %lukB\n", K(total
));
3163 hugetlb_show_meminfo();
3165 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3167 show_swap_cache_info();
3170 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3172 zoneref
->zone
= zone
;
3173 zoneref
->zone_idx
= zone_idx(zone
);
3177 * Builds allocation fallback zone lists.
3179 * Add all populated zones of a node to the zonelist.
3181 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3185 enum zone_type zone_type
= MAX_NR_ZONES
;
3189 zone
= pgdat
->node_zones
+ zone_type
;
3190 if (populated_zone(zone
)) {
3191 zoneref_set_zone(zone
,
3192 &zonelist
->_zonerefs
[nr_zones
++]);
3193 check_highest_zone(zone_type
);
3195 } while (zone_type
);
3203 * 0 = automatic detection of better ordering.
3204 * 1 = order by ([node] distance, -zonetype)
3205 * 2 = order by (-zonetype, [node] distance)
3207 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3208 * the same zonelist. So only NUMA can configure this param.
3210 #define ZONELIST_ORDER_DEFAULT 0
3211 #define ZONELIST_ORDER_NODE 1
3212 #define ZONELIST_ORDER_ZONE 2
3214 /* zonelist order in the kernel.
3215 * set_zonelist_order() will set this to NODE or ZONE.
3217 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3218 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3222 /* The value user specified ....changed by config */
3223 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3224 /* string for sysctl */
3225 #define NUMA_ZONELIST_ORDER_LEN 16
3226 char numa_zonelist_order
[16] = "default";
3229 * interface for configure zonelist ordering.
3230 * command line option "numa_zonelist_order"
3231 * = "[dD]efault - default, automatic configuration.
3232 * = "[nN]ode - order by node locality, then by zone within node
3233 * = "[zZ]one - order by zone, then by locality within zone
3236 static int __parse_numa_zonelist_order(char *s
)
3238 if (*s
== 'd' || *s
== 'D') {
3239 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3240 } else if (*s
== 'n' || *s
== 'N') {
3241 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3242 } else if (*s
== 'z' || *s
== 'Z') {
3243 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3246 "Ignoring invalid numa_zonelist_order value: "
3253 static __init
int setup_numa_zonelist_order(char *s
)
3260 ret
= __parse_numa_zonelist_order(s
);
3262 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3266 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3269 * sysctl handler for numa_zonelist_order
3271 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
3272 void __user
*buffer
, size_t *length
,
3275 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3277 static DEFINE_MUTEX(zl_order_mutex
);
3279 mutex_lock(&zl_order_mutex
);
3281 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
3285 strcpy(saved_string
, (char *)table
->data
);
3287 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3291 int oldval
= user_zonelist_order
;
3293 ret
= __parse_numa_zonelist_order((char *)table
->data
);
3296 * bogus value. restore saved string
3298 strncpy((char *)table
->data
, saved_string
,
3299 NUMA_ZONELIST_ORDER_LEN
);
3300 user_zonelist_order
= oldval
;
3301 } else if (oldval
!= user_zonelist_order
) {
3302 mutex_lock(&zonelists_mutex
);
3303 build_all_zonelists(NULL
, NULL
);
3304 mutex_unlock(&zonelists_mutex
);
3308 mutex_unlock(&zl_order_mutex
);
3313 #define MAX_NODE_LOAD (nr_online_nodes)
3314 static int node_load
[MAX_NUMNODES
];
3317 * find_next_best_node - find the next node that should appear in a given node's fallback list
3318 * @node: node whose fallback list we're appending
3319 * @used_node_mask: nodemask_t of already used nodes
3321 * We use a number of factors to determine which is the next node that should
3322 * appear on a given node's fallback list. The node should not have appeared
3323 * already in @node's fallback list, and it should be the next closest node
3324 * according to the distance array (which contains arbitrary distance values
3325 * from each node to each node in the system), and should also prefer nodes
3326 * with no CPUs, since presumably they'll have very little allocation pressure
3327 * on them otherwise.
3328 * It returns -1 if no node is found.
3330 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3333 int min_val
= INT_MAX
;
3334 int best_node
= NUMA_NO_NODE
;
3335 const struct cpumask
*tmp
= cpumask_of_node(0);
3337 /* Use the local node if we haven't already */
3338 if (!node_isset(node
, *used_node_mask
)) {
3339 node_set(node
, *used_node_mask
);
3343 for_each_node_state(n
, N_MEMORY
) {
3345 /* Don't want a node to appear more than once */
3346 if (node_isset(n
, *used_node_mask
))
3349 /* Use the distance array to find the distance */
3350 val
= node_distance(node
, n
);
3352 /* Penalize nodes under us ("prefer the next node") */
3355 /* Give preference to headless and unused nodes */
3356 tmp
= cpumask_of_node(n
);
3357 if (!cpumask_empty(tmp
))
3358 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
3360 /* Slight preference for less loaded node */
3361 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
3362 val
+= node_load
[n
];
3364 if (val
< min_val
) {
3371 node_set(best_node
, *used_node_mask
);
3378 * Build zonelists ordered by node and zones within node.
3379 * This results in maximum locality--normal zone overflows into local
3380 * DMA zone, if any--but risks exhausting DMA zone.
3382 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
3385 struct zonelist
*zonelist
;
3387 zonelist
= &pgdat
->node_zonelists
[0];
3388 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
3390 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
3391 zonelist
->_zonerefs
[j
].zone
= NULL
;
3392 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3396 * Build gfp_thisnode zonelists
3398 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
3401 struct zonelist
*zonelist
;
3403 zonelist
= &pgdat
->node_zonelists
[1];
3404 j
= build_zonelists_node(pgdat
, zonelist
, 0);
3405 zonelist
->_zonerefs
[j
].zone
= NULL
;
3406 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3410 * Build zonelists ordered by zone and nodes within zones.
3411 * This results in conserving DMA zone[s] until all Normal memory is
3412 * exhausted, but results in overflowing to remote node while memory
3413 * may still exist in local DMA zone.
3415 static int node_order
[MAX_NUMNODES
];
3417 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
3420 int zone_type
; /* needs to be signed */
3422 struct zonelist
*zonelist
;
3424 zonelist
= &pgdat
->node_zonelists
[0];
3426 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
3427 for (j
= 0; j
< nr_nodes
; j
++) {
3428 node
= node_order
[j
];
3429 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
3430 if (populated_zone(z
)) {
3432 &zonelist
->_zonerefs
[pos
++]);
3433 check_highest_zone(zone_type
);
3437 zonelist
->_zonerefs
[pos
].zone
= NULL
;
3438 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
3441 static int default_zonelist_order(void)
3444 unsigned long low_kmem_size
, total_size
;
3448 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3449 * If they are really small and used heavily, the system can fall
3450 * into OOM very easily.
3451 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3453 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3456 for_each_online_node(nid
) {
3457 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3458 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3459 if (populated_zone(z
)) {
3460 if (zone_type
< ZONE_NORMAL
)
3461 low_kmem_size
+= z
->managed_pages
;
3462 total_size
+= z
->managed_pages
;
3463 } else if (zone_type
== ZONE_NORMAL
) {
3465 * If any node has only lowmem, then node order
3466 * is preferred to allow kernel allocations
3467 * locally; otherwise, they can easily infringe
3468 * on other nodes when there is an abundance of
3469 * lowmem available to allocate from.
3471 return ZONELIST_ORDER_NODE
;
3475 if (!low_kmem_size
|| /* there are no DMA area. */
3476 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
3477 return ZONELIST_ORDER_NODE
;
3479 * look into each node's config.
3480 * If there is a node whose DMA/DMA32 memory is very big area on
3481 * local memory, NODE_ORDER may be suitable.
3483 average_size
= total_size
/
3484 (nodes_weight(node_states
[N_MEMORY
]) + 1);
3485 for_each_online_node(nid
) {
3488 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3489 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3490 if (populated_zone(z
)) {
3491 if (zone_type
< ZONE_NORMAL
)
3492 low_kmem_size
+= z
->present_pages
;
3493 total_size
+= z
->present_pages
;
3496 if (low_kmem_size
&&
3497 total_size
> average_size
&& /* ignore small node */
3498 low_kmem_size
> total_size
* 70/100)
3499 return ZONELIST_ORDER_NODE
;
3501 return ZONELIST_ORDER_ZONE
;
3504 static void set_zonelist_order(void)
3506 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
3507 current_zonelist_order
= default_zonelist_order();
3509 current_zonelist_order
= user_zonelist_order
;
3512 static void build_zonelists(pg_data_t
*pgdat
)
3516 nodemask_t used_mask
;
3517 int local_node
, prev_node
;
3518 struct zonelist
*zonelist
;
3519 int order
= current_zonelist_order
;
3521 /* initialize zonelists */
3522 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3523 zonelist
= pgdat
->node_zonelists
+ i
;
3524 zonelist
->_zonerefs
[0].zone
= NULL
;
3525 zonelist
->_zonerefs
[0].zone_idx
= 0;
3528 /* NUMA-aware ordering of nodes */
3529 local_node
= pgdat
->node_id
;
3530 load
= nr_online_nodes
;
3531 prev_node
= local_node
;
3532 nodes_clear(used_mask
);
3534 memset(node_order
, 0, sizeof(node_order
));
3537 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3539 * We don't want to pressure a particular node.
3540 * So adding penalty to the first node in same
3541 * distance group to make it round-robin.
3543 if (node_distance(local_node
, node
) !=
3544 node_distance(local_node
, prev_node
))
3545 node_load
[node
] = load
;
3549 if (order
== ZONELIST_ORDER_NODE
)
3550 build_zonelists_in_node_order(pgdat
, node
);
3552 node_order
[j
++] = node
; /* remember order */
3555 if (order
== ZONELIST_ORDER_ZONE
) {
3556 /* calculate node order -- i.e., DMA last! */
3557 build_zonelists_in_zone_order(pgdat
, j
);
3560 build_thisnode_zonelists(pgdat
);
3563 /* Construct the zonelist performance cache - see further mmzone.h */
3564 static void build_zonelist_cache(pg_data_t
*pgdat
)
3566 struct zonelist
*zonelist
;
3567 struct zonelist_cache
*zlc
;
3570 zonelist
= &pgdat
->node_zonelists
[0];
3571 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3572 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3573 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3574 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3577 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3579 * Return node id of node used for "local" allocations.
3580 * I.e., first node id of first zone in arg node's generic zonelist.
3581 * Used for initializing percpu 'numa_mem', which is used primarily
3582 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3584 int local_memory_node(int node
)
3588 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3589 gfp_zone(GFP_KERNEL
),
3596 #else /* CONFIG_NUMA */
3598 static void set_zonelist_order(void)
3600 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3603 static void build_zonelists(pg_data_t
*pgdat
)
3605 int node
, local_node
;
3607 struct zonelist
*zonelist
;
3609 local_node
= pgdat
->node_id
;
3611 zonelist
= &pgdat
->node_zonelists
[0];
3612 j
= build_zonelists_node(pgdat
, zonelist
, 0);
3615 * Now we build the zonelist so that it contains the zones
3616 * of all the other nodes.
3617 * We don't want to pressure a particular node, so when
3618 * building the zones for node N, we make sure that the
3619 * zones coming right after the local ones are those from
3620 * node N+1 (modulo N)
3622 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3623 if (!node_online(node
))
3625 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
3627 for (node
= 0; node
< local_node
; node
++) {
3628 if (!node_online(node
))
3630 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
3633 zonelist
->_zonerefs
[j
].zone
= NULL
;
3634 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3637 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3638 static void build_zonelist_cache(pg_data_t
*pgdat
)
3640 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3643 #endif /* CONFIG_NUMA */
3646 * Boot pageset table. One per cpu which is going to be used for all
3647 * zones and all nodes. The parameters will be set in such a way
3648 * that an item put on a list will immediately be handed over to
3649 * the buddy list. This is safe since pageset manipulation is done
3650 * with interrupts disabled.
3652 * The boot_pagesets must be kept even after bootup is complete for
3653 * unused processors and/or zones. They do play a role for bootstrapping
3654 * hotplugged processors.
3656 * zoneinfo_show() and maybe other functions do
3657 * not check if the processor is online before following the pageset pointer.
3658 * Other parts of the kernel may not check if the zone is available.
3660 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3661 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3662 static void setup_zone_pageset(struct zone
*zone
);
3665 * Global mutex to protect against size modification of zonelists
3666 * as well as to serialize pageset setup for the new populated zone.
3668 DEFINE_MUTEX(zonelists_mutex
);
3670 /* return values int ....just for stop_machine() */
3671 static int __build_all_zonelists(void *data
)
3675 pg_data_t
*self
= data
;
3678 memset(node_load
, 0, sizeof(node_load
));
3681 if (self
&& !node_online(self
->node_id
)) {
3682 build_zonelists(self
);
3683 build_zonelist_cache(self
);
3686 for_each_online_node(nid
) {
3687 pg_data_t
*pgdat
= NODE_DATA(nid
);
3689 build_zonelists(pgdat
);
3690 build_zonelist_cache(pgdat
);
3694 * Initialize the boot_pagesets that are going to be used
3695 * for bootstrapping processors. The real pagesets for
3696 * each zone will be allocated later when the per cpu
3697 * allocator is available.
3699 * boot_pagesets are used also for bootstrapping offline
3700 * cpus if the system is already booted because the pagesets
3701 * are needed to initialize allocators on a specific cpu too.
3702 * F.e. the percpu allocator needs the page allocator which
3703 * needs the percpu allocator in order to allocate its pagesets
3704 * (a chicken-egg dilemma).
3706 for_each_possible_cpu(cpu
) {
3707 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3709 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3711 * We now know the "local memory node" for each node--
3712 * i.e., the node of the first zone in the generic zonelist.
3713 * Set up numa_mem percpu variable for on-line cpus. During
3714 * boot, only the boot cpu should be on-line; we'll init the
3715 * secondary cpus' numa_mem as they come on-line. During
3716 * node/memory hotplug, we'll fixup all on-line cpus.
3718 if (cpu_online(cpu
))
3719 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3727 * Called with zonelists_mutex held always
3728 * unless system_state == SYSTEM_BOOTING.
3730 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
3732 set_zonelist_order();
3734 if (system_state
== SYSTEM_BOOTING
) {
3735 __build_all_zonelists(NULL
);
3736 mminit_verify_zonelist();
3737 cpuset_init_current_mems_allowed();
3739 #ifdef CONFIG_MEMORY_HOTPLUG
3741 setup_zone_pageset(zone
);
3743 /* we have to stop all cpus to guarantee there is no user
3745 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
3746 /* cpuset refresh routine should be here */
3748 vm_total_pages
= nr_free_pagecache_pages();
3750 * Disable grouping by mobility if the number of pages in the
3751 * system is too low to allow the mechanism to work. It would be
3752 * more accurate, but expensive to check per-zone. This check is
3753 * made on memory-hotadd so a system can start with mobility
3754 * disabled and enable it later
3756 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3757 page_group_by_mobility_disabled
= 1;
3759 page_group_by_mobility_disabled
= 0;
3761 printk("Built %i zonelists in %s order, mobility grouping %s. "
3762 "Total pages: %ld\n",
3764 zonelist_order_name
[current_zonelist_order
],
3765 page_group_by_mobility_disabled
? "off" : "on",
3768 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3773 * Helper functions to size the waitqueue hash table.
3774 * Essentially these want to choose hash table sizes sufficiently
3775 * large so that collisions trying to wait on pages are rare.
3776 * But in fact, the number of active page waitqueues on typical
3777 * systems is ridiculously low, less than 200. So this is even
3778 * conservative, even though it seems large.
3780 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3781 * waitqueues, i.e. the size of the waitq table given the number of pages.
3783 #define PAGES_PER_WAITQUEUE 256
3785 #ifndef CONFIG_MEMORY_HOTPLUG
3786 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3788 unsigned long size
= 1;
3790 pages
/= PAGES_PER_WAITQUEUE
;
3792 while (size
< pages
)
3796 * Once we have dozens or even hundreds of threads sleeping
3797 * on IO we've got bigger problems than wait queue collision.
3798 * Limit the size of the wait table to a reasonable size.
3800 size
= min(size
, 4096UL);
3802 return max(size
, 4UL);
3806 * A zone's size might be changed by hot-add, so it is not possible to determine
3807 * a suitable size for its wait_table. So we use the maximum size now.
3809 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3811 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3812 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3813 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3815 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3816 * or more by the traditional way. (See above). It equals:
3818 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3819 * ia64(16K page size) : = ( 8G + 4M)byte.
3820 * powerpc (64K page size) : = (32G +16M)byte.
3822 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3829 * This is an integer logarithm so that shifts can be used later
3830 * to extract the more random high bits from the multiplicative
3831 * hash function before the remainder is taken.
3833 static inline unsigned long wait_table_bits(unsigned long size
)
3838 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3841 * Check if a pageblock contains reserved pages
3843 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3847 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3848 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3855 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3856 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3857 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3858 * higher will lead to a bigger reserve which will get freed as contiguous
3859 * blocks as reclaim kicks in
3861 static void setup_zone_migrate_reserve(struct zone
*zone
)
3863 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3865 unsigned long block_migratetype
;
3869 * Get the start pfn, end pfn and the number of blocks to reserve
3870 * We have to be careful to be aligned to pageblock_nr_pages to
3871 * make sure that we always check pfn_valid for the first page in
3874 start_pfn
= zone
->zone_start_pfn
;
3875 end_pfn
= zone_end_pfn(zone
);
3876 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3877 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3881 * Reserve blocks are generally in place to help high-order atomic
3882 * allocations that are short-lived. A min_free_kbytes value that
3883 * would result in more than 2 reserve blocks for atomic allocations
3884 * is assumed to be in place to help anti-fragmentation for the
3885 * future allocation of hugepages at runtime.
3887 reserve
= min(2, reserve
);
3889 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3890 if (!pfn_valid(pfn
))
3892 page
= pfn_to_page(pfn
);
3894 /* Watch out for overlapping nodes */
3895 if (page_to_nid(page
) != zone_to_nid(zone
))
3898 block_migratetype
= get_pageblock_migratetype(page
);
3900 /* Only test what is necessary when the reserves are not met */
3903 * Blocks with reserved pages will never free, skip
3906 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3907 if (pageblock_is_reserved(pfn
, block_end_pfn
))
3910 /* If this block is reserved, account for it */
3911 if (block_migratetype
== MIGRATE_RESERVE
) {
3916 /* Suitable for reserving if this block is movable */
3917 if (block_migratetype
== MIGRATE_MOVABLE
) {
3918 set_pageblock_migratetype(page
,
3920 move_freepages_block(zone
, page
,
3928 * If the reserve is met and this is a previous reserved block,
3931 if (block_migratetype
== MIGRATE_RESERVE
) {
3932 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3933 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3939 * Initially all pages are reserved - free ones are freed
3940 * up by free_all_bootmem() once the early boot process is
3941 * done. Non-atomic initialization, single-pass.
3943 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3944 unsigned long start_pfn
, enum memmap_context context
)
3947 unsigned long end_pfn
= start_pfn
+ size
;
3951 if (highest_memmap_pfn
< end_pfn
- 1)
3952 highest_memmap_pfn
= end_pfn
- 1;
3954 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3955 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3957 * There can be holes in boot-time mem_map[]s
3958 * handed to this function. They do not
3959 * exist on hotplugged memory.
3961 if (context
== MEMMAP_EARLY
) {
3962 if (!early_pfn_valid(pfn
))
3964 if (!early_pfn_in_nid(pfn
, nid
))
3967 page
= pfn_to_page(pfn
);
3968 set_page_links(page
, zone
, nid
, pfn
);
3969 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3970 init_page_count(page
);
3971 page_mapcount_reset(page
);
3972 page_nid_reset_last(page
);
3973 SetPageReserved(page
);
3975 * Mark the block movable so that blocks are reserved for
3976 * movable at startup. This will force kernel allocations
3977 * to reserve their blocks rather than leaking throughout
3978 * the address space during boot when many long-lived
3979 * kernel allocations are made. Later some blocks near
3980 * the start are marked MIGRATE_RESERVE by
3981 * setup_zone_migrate_reserve()
3983 * bitmap is created for zone's valid pfn range. but memmap
3984 * can be created for invalid pages (for alignment)
3985 * check here not to call set_pageblock_migratetype() against
3988 if ((z
->zone_start_pfn
<= pfn
)
3989 && (pfn
< zone_end_pfn(z
))
3990 && !(pfn
& (pageblock_nr_pages
- 1)))
3991 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3993 INIT_LIST_HEAD(&page
->lru
);
3994 #ifdef WANT_PAGE_VIRTUAL
3995 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3996 if (!is_highmem_idx(zone
))
3997 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
4002 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4005 for_each_migratetype_order(order
, t
) {
4006 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4007 zone
->free_area
[order
].nr_free
= 0;
4011 #ifndef __HAVE_ARCH_MEMMAP_INIT
4012 #define memmap_init(size, nid, zone, start_pfn) \
4013 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4016 static int __meminit
zone_batchsize(struct zone
*zone
)
4022 * The per-cpu-pages pools are set to around 1000th of the
4023 * size of the zone. But no more than 1/2 of a meg.
4025 * OK, so we don't know how big the cache is. So guess.
4027 batch
= zone
->managed_pages
/ 1024;
4028 if (batch
* PAGE_SIZE
> 512 * 1024)
4029 batch
= (512 * 1024) / PAGE_SIZE
;
4030 batch
/= 4; /* We effectively *= 4 below */
4035 * Clamp the batch to a 2^n - 1 value. Having a power
4036 * of 2 value was found to be more likely to have
4037 * suboptimal cache aliasing properties in some cases.
4039 * For example if 2 tasks are alternately allocating
4040 * batches of pages, one task can end up with a lot
4041 * of pages of one half of the possible page colors
4042 * and the other with pages of the other colors.
4044 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4049 /* The deferral and batching of frees should be suppressed under NOMMU
4052 * The problem is that NOMMU needs to be able to allocate large chunks
4053 * of contiguous memory as there's no hardware page translation to
4054 * assemble apparent contiguous memory from discontiguous pages.
4056 * Queueing large contiguous runs of pages for batching, however,
4057 * causes the pages to actually be freed in smaller chunks. As there
4058 * can be a significant delay between the individual batches being
4059 * recycled, this leads to the once large chunks of space being
4060 * fragmented and becoming unavailable for high-order allocations.
4067 * pcp->high and pcp->batch values are related and dependent on one another:
4068 * ->batch must never be higher then ->high.
4069 * The following function updates them in a safe manner without read side
4072 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4073 * those fields changing asynchronously (acording the the above rule).
4075 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4076 * outside of boot time (or some other assurance that no concurrent updaters
4079 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4080 unsigned long batch
)
4082 /* start with a fail safe value for batch */
4086 /* Update high, then batch, in order */
4093 /* a companion to pageset_set_high() */
4094 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4096 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4099 static void pageset_init(struct per_cpu_pageset
*p
)
4101 struct per_cpu_pages
*pcp
;
4104 memset(p
, 0, sizeof(*p
));
4108 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4109 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4112 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4115 pageset_set_batch(p
, batch
);
4119 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4120 * to the value high for the pageset p.
4122 static void pageset_set_high(struct per_cpu_pageset
*p
,
4125 unsigned long batch
= max(1UL, high
/ 4);
4126 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4127 batch
= PAGE_SHIFT
* 8;
4129 pageset_update(&p
->pcp
, high
, batch
);
4132 static void __meminit
pageset_set_high_and_batch(struct zone
*zone
,
4133 struct per_cpu_pageset
*pcp
)
4135 if (percpu_pagelist_fraction
)
4136 pageset_set_high(pcp
,
4137 (zone
->managed_pages
/
4138 percpu_pagelist_fraction
));
4140 pageset_set_batch(pcp
, zone_batchsize(zone
));
4143 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4145 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4148 pageset_set_high_and_batch(zone
, pcp
);
4151 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4154 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4155 for_each_possible_cpu(cpu
)
4156 zone_pageset_init(zone
, cpu
);
4160 * Allocate per cpu pagesets and initialize them.
4161 * Before this call only boot pagesets were available.
4163 void __init
setup_per_cpu_pageset(void)
4167 for_each_populated_zone(zone
)
4168 setup_zone_pageset(zone
);
4171 static noinline __init_refok
4172 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4175 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4179 * The per-page waitqueue mechanism uses hashed waitqueues
4182 zone
->wait_table_hash_nr_entries
=
4183 wait_table_hash_nr_entries(zone_size_pages
);
4184 zone
->wait_table_bits
=
4185 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4186 alloc_size
= zone
->wait_table_hash_nr_entries
4187 * sizeof(wait_queue_head_t
);
4189 if (!slab_is_available()) {
4190 zone
->wait_table
= (wait_queue_head_t
*)
4191 alloc_bootmem_node_nopanic(pgdat
, alloc_size
);
4194 * This case means that a zone whose size was 0 gets new memory
4195 * via memory hot-add.
4196 * But it may be the case that a new node was hot-added. In
4197 * this case vmalloc() will not be able to use this new node's
4198 * memory - this wait_table must be initialized to use this new
4199 * node itself as well.
4200 * To use this new node's memory, further consideration will be
4203 zone
->wait_table
= vmalloc(alloc_size
);
4205 if (!zone
->wait_table
)
4208 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4209 init_waitqueue_head(zone
->wait_table
+ i
);
4214 static __meminit
void zone_pcp_init(struct zone
*zone
)
4217 * per cpu subsystem is not up at this point. The following code
4218 * relies on the ability of the linker to provide the
4219 * offset of a (static) per cpu variable into the per cpu area.
4221 zone
->pageset
= &boot_pageset
;
4223 if (zone
->present_pages
)
4224 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4225 zone
->name
, zone
->present_pages
,
4226 zone_batchsize(zone
));
4229 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4230 unsigned long zone_start_pfn
,
4232 enum memmap_context context
)
4234 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4236 ret
= zone_wait_table_init(zone
, size
);
4239 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4241 zone
->zone_start_pfn
= zone_start_pfn
;
4243 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4244 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4246 (unsigned long)zone_idx(zone
),
4247 zone_start_pfn
, (zone_start_pfn
+ size
));
4249 zone_init_free_lists(zone
);
4254 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4255 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4257 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4258 * Architectures may implement their own version but if add_active_range()
4259 * was used and there are no special requirements, this is a convenient
4262 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
4264 unsigned long start_pfn
, end_pfn
;
4267 * NOTE: The following SMP-unsafe globals are only used early in boot
4268 * when the kernel is running single-threaded.
4270 static unsigned long __meminitdata last_start_pfn
, last_end_pfn
;
4271 static int __meminitdata last_nid
;
4273 if (last_start_pfn
<= pfn
&& pfn
< last_end_pfn
)
4276 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4277 if (start_pfn
<= pfn
&& pfn
< end_pfn
) {
4278 last_start_pfn
= start_pfn
;
4279 last_end_pfn
= end_pfn
;
4283 /* This is a memory hole */
4286 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4288 int __meminit
early_pfn_to_nid(unsigned long pfn
)
4292 nid
= __early_pfn_to_nid(pfn
);
4295 /* just returns 0 */
4299 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4300 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
4304 nid
= __early_pfn_to_nid(pfn
);
4305 if (nid
>= 0 && nid
!= node
)
4312 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
4313 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4314 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
4316 * If an architecture guarantees that all ranges registered with
4317 * add_active_ranges() contain no holes and may be freed, this
4318 * this function may be used instead of calling free_bootmem() manually.
4320 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4322 unsigned long start_pfn
, end_pfn
;
4325 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4326 start_pfn
= min(start_pfn
, max_low_pfn
);
4327 end_pfn
= min(end_pfn
, max_low_pfn
);
4329 if (start_pfn
< end_pfn
)
4330 free_bootmem_node(NODE_DATA(this_nid
),
4331 PFN_PHYS(start_pfn
),
4332 (end_pfn
- start_pfn
) << PAGE_SHIFT
);
4337 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4338 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4340 * If an architecture guarantees that all ranges registered with
4341 * add_active_ranges() contain no holes and may be freed, this
4342 * function may be used instead of calling memory_present() manually.
4344 void __init
sparse_memory_present_with_active_regions(int nid
)
4346 unsigned long start_pfn
, end_pfn
;
4349 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4350 memory_present(this_nid
, start_pfn
, end_pfn
);
4354 * get_pfn_range_for_nid - Return the start and end page frames for a node
4355 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4356 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4357 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4359 * It returns the start and end page frame of a node based on information
4360 * provided by an arch calling add_active_range(). If called for a node
4361 * with no available memory, a warning is printed and the start and end
4364 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4365 unsigned long *start_pfn
, unsigned long *end_pfn
)
4367 unsigned long this_start_pfn
, this_end_pfn
;
4373 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4374 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4375 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4378 if (*start_pfn
== -1UL)
4383 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4384 * assumption is made that zones within a node are ordered in monotonic
4385 * increasing memory addresses so that the "highest" populated zone is used
4387 static void __init
find_usable_zone_for_movable(void)
4390 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4391 if (zone_index
== ZONE_MOVABLE
)
4394 if (arch_zone_highest_possible_pfn
[zone_index
] >
4395 arch_zone_lowest_possible_pfn
[zone_index
])
4399 VM_BUG_ON(zone_index
== -1);
4400 movable_zone
= zone_index
;
4404 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4405 * because it is sized independent of architecture. Unlike the other zones,
4406 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4407 * in each node depending on the size of each node and how evenly kernelcore
4408 * is distributed. This helper function adjusts the zone ranges
4409 * provided by the architecture for a given node by using the end of the
4410 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4411 * zones within a node are in order of monotonic increases memory addresses
4413 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4414 unsigned long zone_type
,
4415 unsigned long node_start_pfn
,
4416 unsigned long node_end_pfn
,
4417 unsigned long *zone_start_pfn
,
4418 unsigned long *zone_end_pfn
)
4420 /* Only adjust if ZONE_MOVABLE is on this node */
4421 if (zone_movable_pfn
[nid
]) {
4422 /* Size ZONE_MOVABLE */
4423 if (zone_type
== ZONE_MOVABLE
) {
4424 *zone_start_pfn
= zone_movable_pfn
[nid
];
4425 *zone_end_pfn
= min(node_end_pfn
,
4426 arch_zone_highest_possible_pfn
[movable_zone
]);
4428 /* Adjust for ZONE_MOVABLE starting within this range */
4429 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4430 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4431 *zone_end_pfn
= zone_movable_pfn
[nid
];
4433 /* Check if this whole range is within ZONE_MOVABLE */
4434 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4435 *zone_start_pfn
= *zone_end_pfn
;
4440 * Return the number of pages a zone spans in a node, including holes
4441 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4443 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4444 unsigned long zone_type
,
4445 unsigned long node_start_pfn
,
4446 unsigned long node_end_pfn
,
4447 unsigned long *ignored
)
4449 unsigned long zone_start_pfn
, zone_end_pfn
;
4451 /* Get the start and end of the zone */
4452 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4453 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4454 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4455 node_start_pfn
, node_end_pfn
,
4456 &zone_start_pfn
, &zone_end_pfn
);
4458 /* Check that this node has pages within the zone's required range */
4459 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4462 /* Move the zone boundaries inside the node if necessary */
4463 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4464 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4466 /* Return the spanned pages */
4467 return zone_end_pfn
- zone_start_pfn
;
4471 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4472 * then all holes in the requested range will be accounted for.
4474 unsigned long __meminit
__absent_pages_in_range(int nid
,
4475 unsigned long range_start_pfn
,
4476 unsigned long range_end_pfn
)
4478 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4479 unsigned long start_pfn
, end_pfn
;
4482 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4483 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4484 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4485 nr_absent
-= end_pfn
- start_pfn
;
4491 * absent_pages_in_range - Return number of page frames in holes within a range
4492 * @start_pfn: The start PFN to start searching for holes
4493 * @end_pfn: The end PFN to stop searching for holes
4495 * It returns the number of pages frames in memory holes within a range.
4497 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4498 unsigned long end_pfn
)
4500 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4503 /* Return the number of page frames in holes in a zone on a node */
4504 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4505 unsigned long zone_type
,
4506 unsigned long node_start_pfn
,
4507 unsigned long node_end_pfn
,
4508 unsigned long *ignored
)
4510 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4511 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4512 unsigned long zone_start_pfn
, zone_end_pfn
;
4514 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4515 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4517 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4518 node_start_pfn
, node_end_pfn
,
4519 &zone_start_pfn
, &zone_end_pfn
);
4520 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4523 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4524 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4525 unsigned long zone_type
,
4526 unsigned long node_start_pfn
,
4527 unsigned long node_end_pfn
,
4528 unsigned long *zones_size
)
4530 return zones_size
[zone_type
];
4533 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4534 unsigned long zone_type
,
4535 unsigned long node_start_pfn
,
4536 unsigned long node_end_pfn
,
4537 unsigned long *zholes_size
)
4542 return zholes_size
[zone_type
];
4545 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4547 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4548 unsigned long node_start_pfn
,
4549 unsigned long node_end_pfn
,
4550 unsigned long *zones_size
,
4551 unsigned long *zholes_size
)
4553 unsigned long realtotalpages
, totalpages
= 0;
4556 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4557 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4561 pgdat
->node_spanned_pages
= totalpages
;
4563 realtotalpages
= totalpages
;
4564 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4566 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4567 node_start_pfn
, node_end_pfn
,
4569 pgdat
->node_present_pages
= realtotalpages
;
4570 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4574 #ifndef CONFIG_SPARSEMEM
4576 * Calculate the size of the zone->blockflags rounded to an unsigned long
4577 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4578 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4579 * round what is now in bits to nearest long in bits, then return it in
4582 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
4584 unsigned long usemapsize
;
4586 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
4587 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4588 usemapsize
= usemapsize
>> pageblock_order
;
4589 usemapsize
*= NR_PAGEBLOCK_BITS
;
4590 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4592 return usemapsize
/ 8;
4595 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4597 unsigned long zone_start_pfn
,
4598 unsigned long zonesize
)
4600 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
4601 zone
->pageblock_flags
= NULL
;
4603 zone
->pageblock_flags
= alloc_bootmem_node_nopanic(pgdat
,
4607 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
4608 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
4609 #endif /* CONFIG_SPARSEMEM */
4611 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4613 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4614 void __paginginit
set_pageblock_order(void)
4618 /* Check that pageblock_nr_pages has not already been setup */
4619 if (pageblock_order
)
4622 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4623 order
= HUGETLB_PAGE_ORDER
;
4625 order
= MAX_ORDER
- 1;
4628 * Assume the largest contiguous order of interest is a huge page.
4629 * This value may be variable depending on boot parameters on IA64 and
4632 pageblock_order
= order
;
4634 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4637 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4638 * is unused as pageblock_order is set at compile-time. See
4639 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4642 void __paginginit
set_pageblock_order(void)
4646 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4648 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
4649 unsigned long present_pages
)
4651 unsigned long pages
= spanned_pages
;
4654 * Provide a more accurate estimation if there are holes within
4655 * the zone and SPARSEMEM is in use. If there are holes within the
4656 * zone, each populated memory region may cost us one or two extra
4657 * memmap pages due to alignment because memmap pages for each
4658 * populated regions may not naturally algined on page boundary.
4659 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4661 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
4662 IS_ENABLED(CONFIG_SPARSEMEM
))
4663 pages
= present_pages
;
4665 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
4669 * Set up the zone data structures:
4670 * - mark all pages reserved
4671 * - mark all memory queues empty
4672 * - clear the memory bitmaps
4674 * NOTE: pgdat should get zeroed by caller.
4676 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4677 unsigned long node_start_pfn
, unsigned long node_end_pfn
,
4678 unsigned long *zones_size
, unsigned long *zholes_size
)
4681 int nid
= pgdat
->node_id
;
4682 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4685 pgdat_resize_init(pgdat
);
4686 #ifdef CONFIG_NUMA_BALANCING
4687 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
4688 pgdat
->numabalancing_migrate_nr_pages
= 0;
4689 pgdat
->numabalancing_migrate_next_window
= jiffies
;
4691 init_waitqueue_head(&pgdat
->kswapd_wait
);
4692 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
4693 pgdat_page_cgroup_init(pgdat
);
4695 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4696 struct zone
*zone
= pgdat
->node_zones
+ j
;
4697 unsigned long size
, realsize
, freesize
, memmap_pages
;
4699 size
= zone_spanned_pages_in_node(nid
, j
, node_start_pfn
,
4700 node_end_pfn
, zones_size
);
4701 realsize
= freesize
= size
- zone_absent_pages_in_node(nid
, j
,
4707 * Adjust freesize so that it accounts for how much memory
4708 * is used by this zone for memmap. This affects the watermark
4709 * and per-cpu initialisations
4711 memmap_pages
= calc_memmap_size(size
, realsize
);
4712 if (freesize
>= memmap_pages
) {
4713 freesize
-= memmap_pages
;
4716 " %s zone: %lu pages used for memmap\n",
4717 zone_names
[j
], memmap_pages
);
4720 " %s zone: %lu pages exceeds freesize %lu\n",
4721 zone_names
[j
], memmap_pages
, freesize
);
4723 /* Account for reserved pages */
4724 if (j
== 0 && freesize
> dma_reserve
) {
4725 freesize
-= dma_reserve
;
4726 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4727 zone_names
[0], dma_reserve
);
4730 if (!is_highmem_idx(j
))
4731 nr_kernel_pages
+= freesize
;
4732 /* Charge for highmem memmap if there are enough kernel pages */
4733 else if (nr_kernel_pages
> memmap_pages
* 2)
4734 nr_kernel_pages
-= memmap_pages
;
4735 nr_all_pages
+= freesize
;
4737 zone
->spanned_pages
= size
;
4738 zone
->present_pages
= realsize
;
4740 * Set an approximate value for lowmem here, it will be adjusted
4741 * when the bootmem allocator frees pages into the buddy system.
4742 * And all highmem pages will be managed by the buddy system.
4744 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
4747 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
4749 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
4751 zone
->name
= zone_names
[j
];
4752 spin_lock_init(&zone
->lock
);
4753 spin_lock_init(&zone
->lru_lock
);
4754 zone_seqlock_init(zone
);
4755 zone
->zone_pgdat
= pgdat
;
4757 zone_pcp_init(zone
);
4758 lruvec_init(&zone
->lruvec
);
4762 set_pageblock_order();
4763 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
4764 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4765 size
, MEMMAP_EARLY
);
4767 memmap_init(size
, nid
, j
, zone_start_pfn
);
4768 zone_start_pfn
+= size
;
4772 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4774 /* Skip empty nodes */
4775 if (!pgdat
->node_spanned_pages
)
4778 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4779 /* ia64 gets its own node_mem_map, before this, without bootmem */
4780 if (!pgdat
->node_mem_map
) {
4781 unsigned long size
, start
, end
;
4785 * The zone's endpoints aren't required to be MAX_ORDER
4786 * aligned but the node_mem_map endpoints must be in order
4787 * for the buddy allocator to function correctly.
4789 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4790 end
= pgdat_end_pfn(pgdat
);
4791 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4792 size
= (end
- start
) * sizeof(struct page
);
4793 map
= alloc_remap(pgdat
->node_id
, size
);
4795 map
= alloc_bootmem_node_nopanic(pgdat
, size
);
4796 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4798 #ifndef CONFIG_NEED_MULTIPLE_NODES
4800 * With no DISCONTIG, the global mem_map is just set as node 0's
4802 if (pgdat
== NODE_DATA(0)) {
4803 mem_map
= NODE_DATA(0)->node_mem_map
;
4804 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4805 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4806 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4807 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4810 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4813 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4814 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4816 pg_data_t
*pgdat
= NODE_DATA(nid
);
4817 unsigned long start_pfn
= 0;
4818 unsigned long end_pfn
= 0;
4820 /* pg_data_t should be reset to zero when it's allocated */
4821 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
4823 pgdat
->node_id
= nid
;
4824 pgdat
->node_start_pfn
= node_start_pfn
;
4825 init_zone_allows_reclaim(nid
);
4826 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4827 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
4829 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
4830 zones_size
, zholes_size
);
4832 alloc_node_mem_map(pgdat
);
4833 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4834 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4835 nid
, (unsigned long)pgdat
,
4836 (unsigned long)pgdat
->node_mem_map
);
4839 free_area_init_core(pgdat
, start_pfn
, end_pfn
,
4840 zones_size
, zholes_size
);
4843 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4845 #if MAX_NUMNODES > 1
4847 * Figure out the number of possible node ids.
4849 void __init
setup_nr_node_ids(void)
4852 unsigned int highest
= 0;
4854 for_each_node_mask(node
, node_possible_map
)
4856 nr_node_ids
= highest
+ 1;
4861 * node_map_pfn_alignment - determine the maximum internode alignment
4863 * This function should be called after node map is populated and sorted.
4864 * It calculates the maximum power of two alignment which can distinguish
4867 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4868 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4869 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4870 * shifted, 1GiB is enough and this function will indicate so.
4872 * This is used to test whether pfn -> nid mapping of the chosen memory
4873 * model has fine enough granularity to avoid incorrect mapping for the
4874 * populated node map.
4876 * Returns the determined alignment in pfn's. 0 if there is no alignment
4877 * requirement (single node).
4879 unsigned long __init
node_map_pfn_alignment(void)
4881 unsigned long accl_mask
= 0, last_end
= 0;
4882 unsigned long start
, end
, mask
;
4886 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4887 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4894 * Start with a mask granular enough to pin-point to the
4895 * start pfn and tick off bits one-by-one until it becomes
4896 * too coarse to separate the current node from the last.
4898 mask
= ~((1 << __ffs(start
)) - 1);
4899 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4902 /* accumulate all internode masks */
4906 /* convert mask to number of pages */
4907 return ~accl_mask
+ 1;
4910 /* Find the lowest pfn for a node */
4911 static unsigned long __init
find_min_pfn_for_node(int nid
)
4913 unsigned long min_pfn
= ULONG_MAX
;
4914 unsigned long start_pfn
;
4917 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4918 min_pfn
= min(min_pfn
, start_pfn
);
4920 if (min_pfn
== ULONG_MAX
) {
4922 "Could not find start_pfn for node %d\n", nid
);
4930 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4932 * It returns the minimum PFN based on information provided via
4933 * add_active_range().
4935 unsigned long __init
find_min_pfn_with_active_regions(void)
4937 return find_min_pfn_for_node(MAX_NUMNODES
);
4941 * early_calculate_totalpages()
4942 * Sum pages in active regions for movable zone.
4943 * Populate N_MEMORY for calculating usable_nodes.
4945 static unsigned long __init
early_calculate_totalpages(void)
4947 unsigned long totalpages
= 0;
4948 unsigned long start_pfn
, end_pfn
;
4951 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
4952 unsigned long pages
= end_pfn
- start_pfn
;
4954 totalpages
+= pages
;
4956 node_set_state(nid
, N_MEMORY
);
4962 * Find the PFN the Movable zone begins in each node. Kernel memory
4963 * is spread evenly between nodes as long as the nodes have enough
4964 * memory. When they don't, some nodes will have more kernelcore than
4967 static void __init
find_zone_movable_pfns_for_nodes(void)
4970 unsigned long usable_startpfn
;
4971 unsigned long kernelcore_node
, kernelcore_remaining
;
4972 /* save the state before borrow the nodemask */
4973 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
4974 unsigned long totalpages
= early_calculate_totalpages();
4975 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
4978 * If movablecore was specified, calculate what size of
4979 * kernelcore that corresponds so that memory usable for
4980 * any allocation type is evenly spread. If both kernelcore
4981 * and movablecore are specified, then the value of kernelcore
4982 * will be used for required_kernelcore if it's greater than
4983 * what movablecore would have allowed.
4985 if (required_movablecore
) {
4986 unsigned long corepages
;
4989 * Round-up so that ZONE_MOVABLE is at least as large as what
4990 * was requested by the user
4992 required_movablecore
=
4993 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4994 corepages
= totalpages
- required_movablecore
;
4996 required_kernelcore
= max(required_kernelcore
, corepages
);
4999 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5000 if (!required_kernelcore
)
5003 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5004 find_usable_zone_for_movable();
5005 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5008 /* Spread kernelcore memory as evenly as possible throughout nodes */
5009 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5010 for_each_node_state(nid
, N_MEMORY
) {
5011 unsigned long start_pfn
, end_pfn
;
5014 * Recalculate kernelcore_node if the division per node
5015 * now exceeds what is necessary to satisfy the requested
5016 * amount of memory for the kernel
5018 if (required_kernelcore
< kernelcore_node
)
5019 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5022 * As the map is walked, we track how much memory is usable
5023 * by the kernel using kernelcore_remaining. When it is
5024 * 0, the rest of the node is usable by ZONE_MOVABLE
5026 kernelcore_remaining
= kernelcore_node
;
5028 /* Go through each range of PFNs within this node */
5029 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5030 unsigned long size_pages
;
5032 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5033 if (start_pfn
>= end_pfn
)
5036 /* Account for what is only usable for kernelcore */
5037 if (start_pfn
< usable_startpfn
) {
5038 unsigned long kernel_pages
;
5039 kernel_pages
= min(end_pfn
, usable_startpfn
)
5042 kernelcore_remaining
-= min(kernel_pages
,
5043 kernelcore_remaining
);
5044 required_kernelcore
-= min(kernel_pages
,
5045 required_kernelcore
);
5047 /* Continue if range is now fully accounted */
5048 if (end_pfn
<= usable_startpfn
) {
5051 * Push zone_movable_pfn to the end so
5052 * that if we have to rebalance
5053 * kernelcore across nodes, we will
5054 * not double account here
5056 zone_movable_pfn
[nid
] = end_pfn
;
5059 start_pfn
= usable_startpfn
;
5063 * The usable PFN range for ZONE_MOVABLE is from
5064 * start_pfn->end_pfn. Calculate size_pages as the
5065 * number of pages used as kernelcore
5067 size_pages
= end_pfn
- start_pfn
;
5068 if (size_pages
> kernelcore_remaining
)
5069 size_pages
= kernelcore_remaining
;
5070 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5073 * Some kernelcore has been met, update counts and
5074 * break if the kernelcore for this node has been
5077 required_kernelcore
-= min(required_kernelcore
,
5079 kernelcore_remaining
-= size_pages
;
5080 if (!kernelcore_remaining
)
5086 * If there is still required_kernelcore, we do another pass with one
5087 * less node in the count. This will push zone_movable_pfn[nid] further
5088 * along on the nodes that still have memory until kernelcore is
5092 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5095 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5096 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5097 zone_movable_pfn
[nid
] =
5098 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5101 /* restore the node_state */
5102 node_states
[N_MEMORY
] = saved_node_state
;
5105 /* Any regular or high memory on that node ? */
5106 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5108 enum zone_type zone_type
;
5110 if (N_MEMORY
== N_NORMAL_MEMORY
)
5113 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5114 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5115 if (zone
->present_pages
) {
5116 node_set_state(nid
, N_HIGH_MEMORY
);
5117 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5118 zone_type
<= ZONE_NORMAL
)
5119 node_set_state(nid
, N_NORMAL_MEMORY
);
5126 * free_area_init_nodes - Initialise all pg_data_t and zone data
5127 * @max_zone_pfn: an array of max PFNs for each zone
5129 * This will call free_area_init_node() for each active node in the system.
5130 * Using the page ranges provided by add_active_range(), the size of each
5131 * zone in each node and their holes is calculated. If the maximum PFN
5132 * between two adjacent zones match, it is assumed that the zone is empty.
5133 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5134 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5135 * starts where the previous one ended. For example, ZONE_DMA32 starts
5136 * at arch_max_dma_pfn.
5138 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5140 unsigned long start_pfn
, end_pfn
;
5143 /* Record where the zone boundaries are */
5144 memset(arch_zone_lowest_possible_pfn
, 0,
5145 sizeof(arch_zone_lowest_possible_pfn
));
5146 memset(arch_zone_highest_possible_pfn
, 0,
5147 sizeof(arch_zone_highest_possible_pfn
));
5148 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5149 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5150 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5151 if (i
== ZONE_MOVABLE
)
5153 arch_zone_lowest_possible_pfn
[i
] =
5154 arch_zone_highest_possible_pfn
[i
-1];
5155 arch_zone_highest_possible_pfn
[i
] =
5156 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5158 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5159 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5161 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5162 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5163 find_zone_movable_pfns_for_nodes();
5165 /* Print out the zone ranges */
5166 printk("Zone ranges:\n");
5167 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5168 if (i
== ZONE_MOVABLE
)
5170 printk(KERN_CONT
" %-8s ", zone_names
[i
]);
5171 if (arch_zone_lowest_possible_pfn
[i
] ==
5172 arch_zone_highest_possible_pfn
[i
])
5173 printk(KERN_CONT
"empty\n");
5175 printk(KERN_CONT
"[mem %0#10lx-%0#10lx]\n",
5176 arch_zone_lowest_possible_pfn
[i
] << PAGE_SHIFT
,
5177 (arch_zone_highest_possible_pfn
[i
]
5178 << PAGE_SHIFT
) - 1);
5181 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5182 printk("Movable zone start for each node\n");
5183 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5184 if (zone_movable_pfn
[i
])
5185 printk(" Node %d: %#010lx\n", i
,
5186 zone_movable_pfn
[i
] << PAGE_SHIFT
);
5189 /* Print out the early node map */
5190 printk("Early memory node ranges\n");
5191 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5192 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid
,
5193 start_pfn
<< PAGE_SHIFT
, (end_pfn
<< PAGE_SHIFT
) - 1);
5195 /* Initialise every node */
5196 mminit_verify_pageflags_layout();
5197 setup_nr_node_ids();
5198 for_each_online_node(nid
) {
5199 pg_data_t
*pgdat
= NODE_DATA(nid
);
5200 free_area_init_node(nid
, NULL
,
5201 find_min_pfn_for_node(nid
), NULL
);
5203 /* Any memory on that node */
5204 if (pgdat
->node_present_pages
)
5205 node_set_state(nid
, N_MEMORY
);
5206 check_for_memory(pgdat
, nid
);
5210 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5212 unsigned long long coremem
;
5216 coremem
= memparse(p
, &p
);
5217 *core
= coremem
>> PAGE_SHIFT
;
5219 /* Paranoid check that UL is enough for the coremem value */
5220 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5226 * kernelcore=size sets the amount of memory for use for allocations that
5227 * cannot be reclaimed or migrated.
5229 static int __init
cmdline_parse_kernelcore(char *p
)
5231 return cmdline_parse_core(p
, &required_kernelcore
);
5235 * movablecore=size sets the amount of memory for use for allocations that
5236 * can be reclaimed or migrated.
5238 static int __init
cmdline_parse_movablecore(char *p
)
5240 return cmdline_parse_core(p
, &required_movablecore
);
5243 early_param("kernelcore", cmdline_parse_kernelcore
);
5244 early_param("movablecore", cmdline_parse_movablecore
);
5246 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5248 void adjust_managed_page_count(struct page
*page
, long count
)
5250 spin_lock(&managed_page_count_lock
);
5251 page_zone(page
)->managed_pages
+= count
;
5252 totalram_pages
+= count
;
5253 #ifdef CONFIG_HIGHMEM
5254 if (PageHighMem(page
))
5255 totalhigh_pages
+= count
;
5257 spin_unlock(&managed_page_count_lock
);
5259 EXPORT_SYMBOL(adjust_managed_page_count
);
5261 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
5264 unsigned long pages
= 0;
5266 start
= (void *)PAGE_ALIGN((unsigned long)start
);
5267 end
= (void *)((unsigned long)end
& PAGE_MASK
);
5268 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5269 if ((unsigned int)poison
<= 0xFF)
5270 memset(pos
, poison
, PAGE_SIZE
);
5271 free_reserved_page(virt_to_page(pos
));
5275 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5276 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
5280 EXPORT_SYMBOL(free_reserved_area
);
5282 #ifdef CONFIG_HIGHMEM
5283 void free_highmem_page(struct page
*page
)
5285 __free_reserved_page(page
);
5287 page_zone(page
)->managed_pages
++;
5293 void __init
mem_init_print_info(const char *str
)
5295 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
5296 unsigned long init_code_size
, init_data_size
;
5298 physpages
= get_num_physpages();
5299 codesize
= _etext
- _stext
;
5300 datasize
= _edata
- _sdata
;
5301 rosize
= __end_rodata
- __start_rodata
;
5302 bss_size
= __bss_stop
- __bss_start
;
5303 init_data_size
= __init_end
- __init_begin
;
5304 init_code_size
= _einittext
- _sinittext
;
5307 * Detect special cases and adjust section sizes accordingly:
5308 * 1) .init.* may be embedded into .data sections
5309 * 2) .init.text.* may be out of [__init_begin, __init_end],
5310 * please refer to arch/tile/kernel/vmlinux.lds.S.
5311 * 3) .rodata.* may be embedded into .text or .data sections.
5313 #define adj_init_size(start, end, size, pos, adj) \
5315 if (start <= pos && pos < end && size > adj) \
5319 adj_init_size(__init_begin
, __init_end
, init_data_size
,
5320 _sinittext
, init_code_size
);
5321 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
5322 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
5323 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
5324 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
5326 #undef adj_init_size
5328 printk("Memory: %luK/%luK available "
5329 "(%luK kernel code, %luK rwdata, %luK rodata, "
5330 "%luK init, %luK bss, %luK reserved"
5331 #ifdef CONFIG_HIGHMEM
5335 nr_free_pages() << (PAGE_SHIFT
-10), physpages
<< (PAGE_SHIFT
-10),
5336 codesize
>> 10, datasize
>> 10, rosize
>> 10,
5337 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
5338 (physpages
- totalram_pages
) << (PAGE_SHIFT
-10),
5339 #ifdef CONFIG_HIGHMEM
5340 totalhigh_pages
<< (PAGE_SHIFT
-10),
5342 str
? ", " : "", str
? str
: "");
5346 * set_dma_reserve - set the specified number of pages reserved in the first zone
5347 * @new_dma_reserve: The number of pages to mark reserved
5349 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5350 * In the DMA zone, a significant percentage may be consumed by kernel image
5351 * and other unfreeable allocations which can skew the watermarks badly. This
5352 * function may optionally be used to account for unfreeable pages in the
5353 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5354 * smaller per-cpu batchsize.
5356 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5358 dma_reserve
= new_dma_reserve
;
5361 void __init
free_area_init(unsigned long *zones_size
)
5363 free_area_init_node(0, zones_size
,
5364 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5367 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5368 unsigned long action
, void *hcpu
)
5370 int cpu
= (unsigned long)hcpu
;
5372 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5373 lru_add_drain_cpu(cpu
);
5377 * Spill the event counters of the dead processor
5378 * into the current processors event counters.
5379 * This artificially elevates the count of the current
5382 vm_events_fold_cpu(cpu
);
5385 * Zero the differential counters of the dead processor
5386 * so that the vm statistics are consistent.
5388 * This is only okay since the processor is dead and cannot
5389 * race with what we are doing.
5391 refresh_cpu_vm_stats(cpu
);
5396 void __init
page_alloc_init(void)
5398 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5402 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5403 * or min_free_kbytes changes.
5405 static void calculate_totalreserve_pages(void)
5407 struct pglist_data
*pgdat
;
5408 unsigned long reserve_pages
= 0;
5409 enum zone_type i
, j
;
5411 for_each_online_pgdat(pgdat
) {
5412 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5413 struct zone
*zone
= pgdat
->node_zones
+ i
;
5414 unsigned long max
= 0;
5416 /* Find valid and maximum lowmem_reserve in the zone */
5417 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5418 if (zone
->lowmem_reserve
[j
] > max
)
5419 max
= zone
->lowmem_reserve
[j
];
5422 /* we treat the high watermark as reserved pages. */
5423 max
+= high_wmark_pages(zone
);
5425 if (max
> zone
->managed_pages
)
5426 max
= zone
->managed_pages
;
5427 reserve_pages
+= max
;
5429 * Lowmem reserves are not available to
5430 * GFP_HIGHUSER page cache allocations and
5431 * kswapd tries to balance zones to their high
5432 * watermark. As a result, neither should be
5433 * regarded as dirtyable memory, to prevent a
5434 * situation where reclaim has to clean pages
5435 * in order to balance the zones.
5437 zone
->dirty_balance_reserve
= max
;
5440 dirty_balance_reserve
= reserve_pages
;
5441 totalreserve_pages
= reserve_pages
;
5445 * setup_per_zone_lowmem_reserve - called whenever
5446 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5447 * has a correct pages reserved value, so an adequate number of
5448 * pages are left in the zone after a successful __alloc_pages().
5450 static void setup_per_zone_lowmem_reserve(void)
5452 struct pglist_data
*pgdat
;
5453 enum zone_type j
, idx
;
5455 for_each_online_pgdat(pgdat
) {
5456 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5457 struct zone
*zone
= pgdat
->node_zones
+ j
;
5458 unsigned long managed_pages
= zone
->managed_pages
;
5460 zone
->lowmem_reserve
[j
] = 0;
5464 struct zone
*lower_zone
;
5468 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5469 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5471 lower_zone
= pgdat
->node_zones
+ idx
;
5472 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
5473 sysctl_lowmem_reserve_ratio
[idx
];
5474 managed_pages
+= lower_zone
->managed_pages
;
5479 /* update totalreserve_pages */
5480 calculate_totalreserve_pages();
5483 static void __setup_per_zone_wmarks(void)
5485 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5486 unsigned long lowmem_pages
= 0;
5488 unsigned long flags
;
5490 /* Calculate total number of !ZONE_HIGHMEM pages */
5491 for_each_zone(zone
) {
5492 if (!is_highmem(zone
))
5493 lowmem_pages
+= zone
->managed_pages
;
5496 for_each_zone(zone
) {
5499 spin_lock_irqsave(&zone
->lock
, flags
);
5500 tmp
= (u64
)pages_min
* zone
->managed_pages
;
5501 do_div(tmp
, lowmem_pages
);
5502 if (is_highmem(zone
)) {
5504 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5505 * need highmem pages, so cap pages_min to a small
5508 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5509 * deltas controls asynch page reclaim, and so should
5510 * not be capped for highmem.
5512 unsigned long min_pages
;
5514 min_pages
= zone
->managed_pages
/ 1024;
5515 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
5516 zone
->watermark
[WMARK_MIN
] = min_pages
;
5519 * If it's a lowmem zone, reserve a number of pages
5520 * proportionate to the zone's size.
5522 zone
->watermark
[WMARK_MIN
] = tmp
;
5525 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
5526 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
5528 setup_zone_migrate_reserve(zone
);
5529 spin_unlock_irqrestore(&zone
->lock
, flags
);
5532 /* update totalreserve_pages */
5533 calculate_totalreserve_pages();
5537 * setup_per_zone_wmarks - called when min_free_kbytes changes
5538 * or when memory is hot-{added|removed}
5540 * Ensures that the watermark[min,low,high] values for each zone are set
5541 * correctly with respect to min_free_kbytes.
5543 void setup_per_zone_wmarks(void)
5545 mutex_lock(&zonelists_mutex
);
5546 __setup_per_zone_wmarks();
5547 mutex_unlock(&zonelists_mutex
);
5551 * The inactive anon list should be small enough that the VM never has to
5552 * do too much work, but large enough that each inactive page has a chance
5553 * to be referenced again before it is swapped out.
5555 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5556 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5557 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5558 * the anonymous pages are kept on the inactive list.
5561 * memory ratio inactive anon
5562 * -------------------------------------
5571 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
5573 unsigned int gb
, ratio
;
5575 /* Zone size in gigabytes */
5576 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
5578 ratio
= int_sqrt(10 * gb
);
5582 zone
->inactive_ratio
= ratio
;
5585 static void __meminit
setup_per_zone_inactive_ratio(void)
5590 calculate_zone_inactive_ratio(zone
);
5594 * Initialise min_free_kbytes.
5596 * For small machines we want it small (128k min). For large machines
5597 * we want it large (64MB max). But it is not linear, because network
5598 * bandwidth does not increase linearly with machine size. We use
5600 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5601 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5617 int __meminit
init_per_zone_wmark_min(void)
5619 unsigned long lowmem_kbytes
;
5620 int new_min_free_kbytes
;
5622 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5623 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5625 if (new_min_free_kbytes
> user_min_free_kbytes
) {
5626 min_free_kbytes
= new_min_free_kbytes
;
5627 if (min_free_kbytes
< 128)
5628 min_free_kbytes
= 128;
5629 if (min_free_kbytes
> 65536)
5630 min_free_kbytes
= 65536;
5632 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5633 new_min_free_kbytes
, user_min_free_kbytes
);
5635 setup_per_zone_wmarks();
5636 refresh_zone_stat_thresholds();
5637 setup_per_zone_lowmem_reserve();
5638 setup_per_zone_inactive_ratio();
5641 module_init(init_per_zone_wmark_min
)
5644 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5645 * that we can call two helper functions whenever min_free_kbytes
5648 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5649 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5651 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5653 user_min_free_kbytes
= min_free_kbytes
;
5654 setup_per_zone_wmarks();
5660 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5661 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5666 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5671 zone
->min_unmapped_pages
= (zone
->managed_pages
*
5672 sysctl_min_unmapped_ratio
) / 100;
5676 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5677 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5682 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5687 zone
->min_slab_pages
= (zone
->managed_pages
*
5688 sysctl_min_slab_ratio
) / 100;
5694 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5695 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5696 * whenever sysctl_lowmem_reserve_ratio changes.
5698 * The reserve ratio obviously has absolutely no relation with the
5699 * minimum watermarks. The lowmem reserve ratio can only make sense
5700 * if in function of the boot time zone sizes.
5702 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5703 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5705 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5706 setup_per_zone_lowmem_reserve();
5711 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5712 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5713 * pagelist can have before it gets flushed back to buddy allocator.
5715 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5716 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5722 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5723 if (!write
|| (ret
< 0))
5726 mutex_lock(&pcp_batch_high_lock
);
5727 for_each_populated_zone(zone
) {
5729 high
= zone
->managed_pages
/ percpu_pagelist_fraction
;
5730 for_each_possible_cpu(cpu
)
5731 pageset_set_high(per_cpu_ptr(zone
->pageset
, cpu
),
5734 mutex_unlock(&pcp_batch_high_lock
);
5738 int hashdist
= HASHDIST_DEFAULT
;
5741 static int __init
set_hashdist(char *str
)
5745 hashdist
= simple_strtoul(str
, &str
, 0);
5748 __setup("hashdist=", set_hashdist
);
5752 * allocate a large system hash table from bootmem
5753 * - it is assumed that the hash table must contain an exact power-of-2
5754 * quantity of entries
5755 * - limit is the number of hash buckets, not the total allocation size
5757 void *__init
alloc_large_system_hash(const char *tablename
,
5758 unsigned long bucketsize
,
5759 unsigned long numentries
,
5762 unsigned int *_hash_shift
,
5763 unsigned int *_hash_mask
,
5764 unsigned long low_limit
,
5765 unsigned long high_limit
)
5767 unsigned long long max
= high_limit
;
5768 unsigned long log2qty
, size
;
5771 /* allow the kernel cmdline to have a say */
5773 /* round applicable memory size up to nearest megabyte */
5774 numentries
= nr_kernel_pages
;
5776 /* It isn't necessary when PAGE_SIZE >= 1MB */
5777 if (PAGE_SHIFT
< 20)
5778 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
5780 /* limit to 1 bucket per 2^scale bytes of low memory */
5781 if (scale
> PAGE_SHIFT
)
5782 numentries
>>= (scale
- PAGE_SHIFT
);
5784 numentries
<<= (PAGE_SHIFT
- scale
);
5786 /* Make sure we've got at least a 0-order allocation.. */
5787 if (unlikely(flags
& HASH_SMALL
)) {
5788 /* Makes no sense without HASH_EARLY */
5789 WARN_ON(!(flags
& HASH_EARLY
));
5790 if (!(numentries
>> *_hash_shift
)) {
5791 numentries
= 1UL << *_hash_shift
;
5792 BUG_ON(!numentries
);
5794 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5795 numentries
= PAGE_SIZE
/ bucketsize
;
5797 numentries
= roundup_pow_of_two(numentries
);
5799 /* limit allocation size to 1/16 total memory by default */
5801 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5802 do_div(max
, bucketsize
);
5804 max
= min(max
, 0x80000000ULL
);
5806 if (numentries
< low_limit
)
5807 numentries
= low_limit
;
5808 if (numentries
> max
)
5811 log2qty
= ilog2(numentries
);
5814 size
= bucketsize
<< log2qty
;
5815 if (flags
& HASH_EARLY
)
5816 table
= alloc_bootmem_nopanic(size
);
5818 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5821 * If bucketsize is not a power-of-two, we may free
5822 * some pages at the end of hash table which
5823 * alloc_pages_exact() automatically does
5825 if (get_order(size
) < MAX_ORDER
) {
5826 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5827 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5830 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5833 panic("Failed to allocate %s hash table\n", tablename
);
5835 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5838 ilog2(size
) - PAGE_SHIFT
,
5842 *_hash_shift
= log2qty
;
5844 *_hash_mask
= (1 << log2qty
) - 1;
5849 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5850 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5853 #ifdef CONFIG_SPARSEMEM
5854 return __pfn_to_section(pfn
)->pageblock_flags
;
5856 return zone
->pageblock_flags
;
5857 #endif /* CONFIG_SPARSEMEM */
5860 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5862 #ifdef CONFIG_SPARSEMEM
5863 pfn
&= (PAGES_PER_SECTION
-1);
5864 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5866 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
5867 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5868 #endif /* CONFIG_SPARSEMEM */
5872 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5873 * @page: The page within the block of interest
5874 * @start_bitidx: The first bit of interest to retrieve
5875 * @end_bitidx: The last bit of interest
5876 * returns pageblock_bits flags
5878 unsigned long get_pageblock_flags_group(struct page
*page
,
5879 int start_bitidx
, int end_bitidx
)
5882 unsigned long *bitmap
;
5883 unsigned long pfn
, bitidx
;
5884 unsigned long flags
= 0;
5885 unsigned long value
= 1;
5887 zone
= page_zone(page
);
5888 pfn
= page_to_pfn(page
);
5889 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5890 bitidx
= pfn_to_bitidx(zone
, pfn
);
5892 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5893 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5900 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5901 * @page: The page within the block of interest
5902 * @start_bitidx: The first bit of interest
5903 * @end_bitidx: The last bit of interest
5904 * @flags: The flags to set
5906 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5907 int start_bitidx
, int end_bitidx
)
5910 unsigned long *bitmap
;
5911 unsigned long pfn
, bitidx
;
5912 unsigned long value
= 1;
5914 zone
= page_zone(page
);
5915 pfn
= page_to_pfn(page
);
5916 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5917 bitidx
= pfn_to_bitidx(zone
, pfn
);
5918 VM_BUG_ON(!zone_spans_pfn(zone
, pfn
));
5920 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5922 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5924 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5928 * This function checks whether pageblock includes unmovable pages or not.
5929 * If @count is not zero, it is okay to include less @count unmovable pages
5931 * PageLRU check without isolation or lru_lock could race so that
5932 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5933 * expect this function should be exact.
5935 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
5936 bool skip_hwpoisoned_pages
)
5938 unsigned long pfn
, iter
, found
;
5942 * For avoiding noise data, lru_add_drain_all() should be called
5943 * If ZONE_MOVABLE, the zone never contains unmovable pages
5945 if (zone_idx(zone
) == ZONE_MOVABLE
)
5947 mt
= get_pageblock_migratetype(page
);
5948 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
5951 pfn
= page_to_pfn(page
);
5952 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
5953 unsigned long check
= pfn
+ iter
;
5955 if (!pfn_valid_within(check
))
5958 page
= pfn_to_page(check
);
5960 * We can't use page_count without pin a page
5961 * because another CPU can free compound page.
5962 * This check already skips compound tails of THP
5963 * because their page->_count is zero at all time.
5965 if (!atomic_read(&page
->_count
)) {
5966 if (PageBuddy(page
))
5967 iter
+= (1 << page_order(page
)) - 1;
5972 * The HWPoisoned page may be not in buddy system, and
5973 * page_count() is not 0.
5975 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
5981 * If there are RECLAIMABLE pages, we need to check it.
5982 * But now, memory offline itself doesn't call shrink_slab()
5983 * and it still to be fixed.
5986 * If the page is not RAM, page_count()should be 0.
5987 * we don't need more check. This is an _used_ not-movable page.
5989 * The problematic thing here is PG_reserved pages. PG_reserved
5990 * is set to both of a memory hole page and a _used_ kernel
5999 bool is_pageblock_removable_nolock(struct page
*page
)
6005 * We have to be careful here because we are iterating over memory
6006 * sections which are not zone aware so we might end up outside of
6007 * the zone but still within the section.
6008 * We have to take care about the node as well. If the node is offline
6009 * its NODE_DATA will be NULL - see page_zone.
6011 if (!node_online(page_to_nid(page
)))
6014 zone
= page_zone(page
);
6015 pfn
= page_to_pfn(page
);
6016 if (!zone_spans_pfn(zone
, pfn
))
6019 return !has_unmovable_pages(zone
, page
, 0, true);
6024 static unsigned long pfn_max_align_down(unsigned long pfn
)
6026 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6027 pageblock_nr_pages
) - 1);
6030 static unsigned long pfn_max_align_up(unsigned long pfn
)
6032 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6033 pageblock_nr_pages
));
6036 /* [start, end) must belong to a single zone. */
6037 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6038 unsigned long start
, unsigned long end
)
6040 /* This function is based on compact_zone() from compaction.c. */
6041 unsigned long nr_reclaimed
;
6042 unsigned long pfn
= start
;
6043 unsigned int tries
= 0;
6048 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6049 if (fatal_signal_pending(current
)) {
6054 if (list_empty(&cc
->migratepages
)) {
6055 cc
->nr_migratepages
= 0;
6056 pfn
= isolate_migratepages_range(cc
->zone
, cc
,
6063 } else if (++tries
== 5) {
6064 ret
= ret
< 0 ? ret
: -EBUSY
;
6068 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6070 cc
->nr_migratepages
-= nr_reclaimed
;
6072 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6073 0, MIGRATE_SYNC
, MR_CMA
);
6076 putback_movable_pages(&cc
->migratepages
);
6083 * alloc_contig_range() -- tries to allocate given range of pages
6084 * @start: start PFN to allocate
6085 * @end: one-past-the-last PFN to allocate
6086 * @migratetype: migratetype of the underlaying pageblocks (either
6087 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6088 * in range must have the same migratetype and it must
6089 * be either of the two.
6091 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6092 * aligned, however it's the caller's responsibility to guarantee that
6093 * we are the only thread that changes migrate type of pageblocks the
6096 * The PFN range must belong to a single zone.
6098 * Returns zero on success or negative error code. On success all
6099 * pages which PFN is in [start, end) are allocated for the caller and
6100 * need to be freed with free_contig_range().
6102 int alloc_contig_range(unsigned long start
, unsigned long end
,
6103 unsigned migratetype
)
6105 unsigned long outer_start
, outer_end
;
6108 struct compact_control cc
= {
6109 .nr_migratepages
= 0,
6111 .zone
= page_zone(pfn_to_page(start
)),
6113 .ignore_skip_hint
= true,
6115 INIT_LIST_HEAD(&cc
.migratepages
);
6118 * What we do here is we mark all pageblocks in range as
6119 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6120 * have different sizes, and due to the way page allocator
6121 * work, we align the range to biggest of the two pages so
6122 * that page allocator won't try to merge buddies from
6123 * different pageblocks and change MIGRATE_ISOLATE to some
6124 * other migration type.
6126 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6127 * migrate the pages from an unaligned range (ie. pages that
6128 * we are interested in). This will put all the pages in
6129 * range back to page allocator as MIGRATE_ISOLATE.
6131 * When this is done, we take the pages in range from page
6132 * allocator removing them from the buddy system. This way
6133 * page allocator will never consider using them.
6135 * This lets us mark the pageblocks back as
6136 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6137 * aligned range but not in the unaligned, original range are
6138 * put back to page allocator so that buddy can use them.
6141 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6142 pfn_max_align_up(end
), migratetype
,
6147 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6152 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6153 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6154 * more, all pages in [start, end) are free in page allocator.
6155 * What we are going to do is to allocate all pages from
6156 * [start, end) (that is remove them from page allocator).
6158 * The only problem is that pages at the beginning and at the
6159 * end of interesting range may be not aligned with pages that
6160 * page allocator holds, ie. they can be part of higher order
6161 * pages. Because of this, we reserve the bigger range and
6162 * once this is done free the pages we are not interested in.
6164 * We don't have to hold zone->lock here because the pages are
6165 * isolated thus they won't get removed from buddy.
6168 lru_add_drain_all();
6172 outer_start
= start
;
6173 while (!PageBuddy(pfn_to_page(outer_start
))) {
6174 if (++order
>= MAX_ORDER
) {
6178 outer_start
&= ~0UL << order
;
6181 /* Make sure the range is really isolated. */
6182 if (test_pages_isolated(outer_start
, end
, false)) {
6183 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6190 /* Grab isolated pages from freelists. */
6191 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6197 /* Free head and tail (if any) */
6198 if (start
!= outer_start
)
6199 free_contig_range(outer_start
, start
- outer_start
);
6200 if (end
!= outer_end
)
6201 free_contig_range(end
, outer_end
- end
);
6204 undo_isolate_page_range(pfn_max_align_down(start
),
6205 pfn_max_align_up(end
), migratetype
);
6209 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6211 unsigned int count
= 0;
6213 for (; nr_pages
--; pfn
++) {
6214 struct page
*page
= pfn_to_page(pfn
);
6216 count
+= page_count(page
) != 1;
6219 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6223 #ifdef CONFIG_MEMORY_HOTPLUG
6225 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6226 * page high values need to be recalulated.
6228 void __meminit
zone_pcp_update(struct zone
*zone
)
6231 mutex_lock(&pcp_batch_high_lock
);
6232 for_each_possible_cpu(cpu
)
6233 pageset_set_high_and_batch(zone
,
6234 per_cpu_ptr(zone
->pageset
, cpu
));
6235 mutex_unlock(&pcp_batch_high_lock
);
6239 void zone_pcp_reset(struct zone
*zone
)
6241 unsigned long flags
;
6243 struct per_cpu_pageset
*pset
;
6245 /* avoid races with drain_pages() */
6246 local_irq_save(flags
);
6247 if (zone
->pageset
!= &boot_pageset
) {
6248 for_each_online_cpu(cpu
) {
6249 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6250 drain_zonestat(zone
, pset
);
6252 free_percpu(zone
->pageset
);
6253 zone
->pageset
= &boot_pageset
;
6255 local_irq_restore(flags
);
6258 #ifdef CONFIG_MEMORY_HOTREMOVE
6260 * All pages in the range must be isolated before calling this.
6263 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6269 unsigned long flags
;
6270 /* find the first valid pfn */
6271 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6276 zone
= page_zone(pfn_to_page(pfn
));
6277 spin_lock_irqsave(&zone
->lock
, flags
);
6279 while (pfn
< end_pfn
) {
6280 if (!pfn_valid(pfn
)) {
6284 page
= pfn_to_page(pfn
);
6286 * The HWPoisoned page may be not in buddy system, and
6287 * page_count() is not 0.
6289 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6291 SetPageReserved(page
);
6295 BUG_ON(page_count(page
));
6296 BUG_ON(!PageBuddy(page
));
6297 order
= page_order(page
);
6298 #ifdef CONFIG_DEBUG_VM
6299 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6300 pfn
, 1 << order
, end_pfn
);
6302 list_del(&page
->lru
);
6303 rmv_page_order(page
);
6304 zone
->free_area
[order
].nr_free
--;
6305 #ifdef CONFIG_HIGHMEM
6306 if (PageHighMem(page
))
6307 totalhigh_pages
-= 1 << order
;
6309 for (i
= 0; i
< (1 << order
); i
++)
6310 SetPageReserved((page
+i
));
6311 pfn
+= (1 << order
);
6313 spin_unlock_irqrestore(&zone
->lock
, flags
);
6317 #ifdef CONFIG_MEMORY_FAILURE
6318 bool is_free_buddy_page(struct page
*page
)
6320 struct zone
*zone
= page_zone(page
);
6321 unsigned long pfn
= page_to_pfn(page
);
6322 unsigned long flags
;
6325 spin_lock_irqsave(&zone
->lock
, flags
);
6326 for (order
= 0; order
< MAX_ORDER
; order
++) {
6327 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6329 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6332 spin_unlock_irqrestore(&zone
->lock
, flags
);
6334 return order
< MAX_ORDER
;
6338 static const struct trace_print_flags pageflag_names
[] = {
6339 {1UL << PG_locked
, "locked" },
6340 {1UL << PG_error
, "error" },
6341 {1UL << PG_referenced
, "referenced" },
6342 {1UL << PG_uptodate
, "uptodate" },
6343 {1UL << PG_dirty
, "dirty" },
6344 {1UL << PG_lru
, "lru" },
6345 {1UL << PG_active
, "active" },
6346 {1UL << PG_slab
, "slab" },
6347 {1UL << PG_owner_priv_1
, "owner_priv_1" },
6348 {1UL << PG_arch_1
, "arch_1" },
6349 {1UL << PG_reserved
, "reserved" },
6350 {1UL << PG_private
, "private" },
6351 {1UL << PG_private_2
, "private_2" },
6352 {1UL << PG_writeback
, "writeback" },
6353 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6354 {1UL << PG_head
, "head" },
6355 {1UL << PG_tail
, "tail" },
6357 {1UL << PG_compound
, "compound" },
6359 {1UL << PG_swapcache
, "swapcache" },
6360 {1UL << PG_mappedtodisk
, "mappedtodisk" },
6361 {1UL << PG_reclaim
, "reclaim" },
6362 {1UL << PG_swapbacked
, "swapbacked" },
6363 {1UL << PG_unevictable
, "unevictable" },
6365 {1UL << PG_mlocked
, "mlocked" },
6367 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6368 {1UL << PG_uncached
, "uncached" },
6370 #ifdef CONFIG_MEMORY_FAILURE
6371 {1UL << PG_hwpoison
, "hwpoison" },
6373 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6374 {1UL << PG_compound_lock
, "compound_lock" },
6378 static void dump_page_flags(unsigned long flags
)
6380 const char *delim
= "";
6384 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names
) != __NR_PAGEFLAGS
);
6386 printk(KERN_ALERT
"page flags: %#lx(", flags
);
6388 /* remove zone id */
6389 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
6391 for (i
= 0; i
< ARRAY_SIZE(pageflag_names
) && flags
; i
++) {
6393 mask
= pageflag_names
[i
].mask
;
6394 if ((flags
& mask
) != mask
)
6398 printk("%s%s", delim
, pageflag_names
[i
].name
);
6402 /* check for left over flags */
6404 printk("%s%#lx", delim
, flags
);
6409 void dump_page(struct page
*page
)
6412 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6413 page
, atomic_read(&page
->_count
), page_mapcount(page
),
6414 page
->mapping
, page
->index
);
6415 dump_page_flags(page
->flags
);
6416 mem_cgroup_print_bad_page(page
);