2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/notifier.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/stop_machine.h>
46 #include <linux/sort.h>
47 #include <linux/pfn.h>
48 #include <linux/backing-dev.h>
49 #include <linux/fault-inject.h>
50 #include <linux/page-isolation.h>
51 #include <linux/page_cgroup.h>
52 #include <linux/debugobjects.h>
53 #include <linux/kmemleak.h>
54 #include <linux/compaction.h>
55 #include <trace/events/kmem.h>
56 #include <linux/ftrace_event.h>
57 #include <linux/memcontrol.h>
58 #include <linux/prefetch.h>
59 #include <linux/migrate.h>
60 #include <linux/page-debug-flags.h>
62 #include <asm/tlbflush.h>
63 #include <asm/div64.h>
66 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
67 DEFINE_PER_CPU(int, numa_node
);
68 EXPORT_PER_CPU_SYMBOL(numa_node
);
71 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
73 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
74 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
75 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
76 * defined in <linux/topology.h>.
78 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
79 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
83 * Array of node states.
85 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
86 [N_POSSIBLE
] = NODE_MASK_ALL
,
87 [N_ONLINE
] = { { [0] = 1UL } },
89 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
91 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
93 #ifdef CONFIG_MOVABLE_NODE
94 [N_MEMORY
] = { { [0] = 1UL } },
96 [N_CPU
] = { { [0] = 1UL } },
99 EXPORT_SYMBOL(node_states
);
101 unsigned long totalram_pages __read_mostly
;
102 unsigned long totalreserve_pages __read_mostly
;
104 * When calculating the number of globally allowed dirty pages, there
105 * is a certain number of per-zone reserves that should not be
106 * considered dirtyable memory. This is the sum of those reserves
107 * over all existing zones that contribute dirtyable memory.
109 unsigned long dirty_balance_reserve __read_mostly
;
111 int percpu_pagelist_fraction
;
112 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
114 #ifdef CONFIG_PM_SLEEP
116 * The following functions are used by the suspend/hibernate code to temporarily
117 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
118 * while devices are suspended. To avoid races with the suspend/hibernate code,
119 * they should always be called with pm_mutex held (gfp_allowed_mask also should
120 * only be modified with pm_mutex held, unless the suspend/hibernate code is
121 * guaranteed not to run in parallel with that modification).
124 static gfp_t saved_gfp_mask
;
126 void pm_restore_gfp_mask(void)
128 WARN_ON(!mutex_is_locked(&pm_mutex
));
129 if (saved_gfp_mask
) {
130 gfp_allowed_mask
= saved_gfp_mask
;
135 void pm_restrict_gfp_mask(void)
137 WARN_ON(!mutex_is_locked(&pm_mutex
));
138 WARN_ON(saved_gfp_mask
);
139 saved_gfp_mask
= gfp_allowed_mask
;
140 gfp_allowed_mask
&= ~GFP_IOFS
;
143 bool pm_suspended_storage(void)
145 if ((gfp_allowed_mask
& GFP_IOFS
) == GFP_IOFS
)
149 #endif /* CONFIG_PM_SLEEP */
151 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
152 int pageblock_order __read_mostly
;
155 static void __free_pages_ok(struct page
*page
, unsigned int order
);
158 * results with 256, 32 in the lowmem_reserve sysctl:
159 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
160 * 1G machine -> (16M dma, 784M normal, 224M high)
161 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
162 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
163 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
165 * TBD: should special case ZONE_DMA32 machines here - in those we normally
166 * don't need any ZONE_NORMAL reservation
168 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
169 #ifdef CONFIG_ZONE_DMA
172 #ifdef CONFIG_ZONE_DMA32
175 #ifdef CONFIG_HIGHMEM
181 EXPORT_SYMBOL(totalram_pages
);
183 static char * const zone_names
[MAX_NR_ZONES
] = {
184 #ifdef CONFIG_ZONE_DMA
187 #ifdef CONFIG_ZONE_DMA32
191 #ifdef CONFIG_HIGHMEM
197 int min_free_kbytes
= 1024;
199 static unsigned long __meminitdata nr_kernel_pages
;
200 static unsigned long __meminitdata nr_all_pages
;
201 static unsigned long __meminitdata dma_reserve
;
203 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
204 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
205 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
206 static unsigned long __initdata required_kernelcore
;
207 static unsigned long __initdata required_movablecore
;
208 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
210 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
212 EXPORT_SYMBOL(movable_zone
);
213 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
216 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
217 int nr_online_nodes __read_mostly
= 1;
218 EXPORT_SYMBOL(nr_node_ids
);
219 EXPORT_SYMBOL(nr_online_nodes
);
222 int page_group_by_mobility_disabled __read_mostly
;
226 * Don't use set_pageblock_migratetype(page, MIGRATE_ISOLATE) directly.
227 * Instead, use {un}set_pageblock_isolate.
229 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
232 if (unlikely(page_group_by_mobility_disabled
))
233 migratetype
= MIGRATE_UNMOVABLE
;
235 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
236 PB_migrate
, PB_migrate_end
);
239 bool oom_killer_disabled __read_mostly
;
241 #ifdef CONFIG_DEBUG_VM
242 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
246 unsigned long pfn
= page_to_pfn(page
);
249 seq
= zone_span_seqbegin(zone
);
250 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
252 else if (pfn
< zone
->zone_start_pfn
)
254 } while (zone_span_seqretry(zone
, seq
));
259 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
261 if (!pfn_valid_within(page_to_pfn(page
)))
263 if (zone
!= page_zone(page
))
269 * Temporary debugging check for pages not lying within a given zone.
271 static int bad_range(struct zone
*zone
, struct page
*page
)
273 if (page_outside_zone_boundaries(zone
, page
))
275 if (!page_is_consistent(zone
, page
))
281 static inline int bad_range(struct zone
*zone
, struct page
*page
)
287 static void bad_page(struct page
*page
)
289 static unsigned long resume
;
290 static unsigned long nr_shown
;
291 static unsigned long nr_unshown
;
293 /* Don't complain about poisoned pages */
294 if (PageHWPoison(page
)) {
295 reset_page_mapcount(page
); /* remove PageBuddy */
300 * Allow a burst of 60 reports, then keep quiet for that minute;
301 * or allow a steady drip of one report per second.
303 if (nr_shown
== 60) {
304 if (time_before(jiffies
, resume
)) {
310 "BUG: Bad page state: %lu messages suppressed\n",
317 resume
= jiffies
+ 60 * HZ
;
319 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
320 current
->comm
, page_to_pfn(page
));
326 /* Leave bad fields for debug, except PageBuddy could make trouble */
327 reset_page_mapcount(page
); /* remove PageBuddy */
328 add_taint(TAINT_BAD_PAGE
);
332 * Higher-order pages are called "compound pages". They are structured thusly:
334 * The first PAGE_SIZE page is called the "head page".
336 * The remaining PAGE_SIZE pages are called "tail pages".
338 * All pages have PG_compound set. All tail pages have their ->first_page
339 * pointing at the head page.
341 * The first tail page's ->lru.next holds the address of the compound page's
342 * put_page() function. Its ->lru.prev holds the order of allocation.
343 * This usage means that zero-order pages may not be compound.
346 static void free_compound_page(struct page
*page
)
348 __free_pages_ok(page
, compound_order(page
));
351 void prep_compound_page(struct page
*page
, unsigned long order
)
354 int nr_pages
= 1 << order
;
356 set_compound_page_dtor(page
, free_compound_page
);
357 set_compound_order(page
, order
);
359 for (i
= 1; i
< nr_pages
; i
++) {
360 struct page
*p
= page
+ i
;
362 set_page_count(p
, 0);
363 p
->first_page
= page
;
367 /* update __split_huge_page_refcount if you change this function */
368 static int destroy_compound_page(struct page
*page
, unsigned long order
)
371 int nr_pages
= 1 << order
;
374 if (unlikely(compound_order(page
) != order
)) {
379 __ClearPageHead(page
);
381 for (i
= 1; i
< nr_pages
; i
++) {
382 struct page
*p
= page
+ i
;
384 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
394 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
399 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
400 * and __GFP_HIGHMEM from hard or soft interrupt context.
402 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
403 for (i
= 0; i
< (1 << order
); i
++)
404 clear_highpage(page
+ i
);
407 #ifdef CONFIG_DEBUG_PAGEALLOC
408 unsigned int _debug_guardpage_minorder
;
410 static int __init
debug_guardpage_minorder_setup(char *buf
)
414 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
415 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
418 _debug_guardpage_minorder
= res
;
419 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
422 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
424 static inline void set_page_guard_flag(struct page
*page
)
426 __set_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
429 static inline void clear_page_guard_flag(struct page
*page
)
431 __clear_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
434 static inline void set_page_guard_flag(struct page
*page
) { }
435 static inline void clear_page_guard_flag(struct page
*page
) { }
438 static inline void set_page_order(struct page
*page
, int order
)
440 set_page_private(page
, order
);
441 __SetPageBuddy(page
);
444 static inline void rmv_page_order(struct page
*page
)
446 __ClearPageBuddy(page
);
447 set_page_private(page
, 0);
451 * Locate the struct page for both the matching buddy in our
452 * pair (buddy1) and the combined O(n+1) page they form (page).
454 * 1) Any buddy B1 will have an order O twin B2 which satisfies
455 * the following equation:
457 * For example, if the starting buddy (buddy2) is #8 its order
459 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
461 * 2) Any buddy B will have an order O+1 parent P which
462 * satisfies the following equation:
465 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
467 static inline unsigned long
468 __find_buddy_index(unsigned long page_idx
, unsigned int order
)
470 return page_idx
^ (1 << order
);
474 * This function checks whether a page is free && is the buddy
475 * we can do coalesce a page and its buddy if
476 * (a) the buddy is not in a hole &&
477 * (b) the buddy is in the buddy system &&
478 * (c) a page and its buddy have the same order &&
479 * (d) a page and its buddy are in the same zone.
481 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
482 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
484 * For recording page's order, we use page_private(page).
486 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
489 if (!pfn_valid_within(page_to_pfn(buddy
)))
492 if (page_zone_id(page
) != page_zone_id(buddy
))
495 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
496 VM_BUG_ON(page_count(buddy
) != 0);
500 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
501 VM_BUG_ON(page_count(buddy
) != 0);
508 * Freeing function for a buddy system allocator.
510 * The concept of a buddy system is to maintain direct-mapped table
511 * (containing bit values) for memory blocks of various "orders".
512 * The bottom level table contains the map for the smallest allocatable
513 * units of memory (here, pages), and each level above it describes
514 * pairs of units from the levels below, hence, "buddies".
515 * At a high level, all that happens here is marking the table entry
516 * at the bottom level available, and propagating the changes upward
517 * as necessary, plus some accounting needed to play nicely with other
518 * parts of the VM system.
519 * At each level, we keep a list of pages, which are heads of continuous
520 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
521 * order is recorded in page_private(page) field.
522 * So when we are allocating or freeing one, we can derive the state of the
523 * other. That is, if we allocate a small block, and both were
524 * free, the remainder of the region must be split into blocks.
525 * If a block is freed, and its buddy is also free, then this
526 * triggers coalescing into a block of larger size.
531 static inline void __free_one_page(struct page
*page
,
532 struct zone
*zone
, unsigned int order
,
535 unsigned long page_idx
;
536 unsigned long combined_idx
;
537 unsigned long uninitialized_var(buddy_idx
);
540 if (unlikely(PageCompound(page
)))
541 if (unlikely(destroy_compound_page(page
, order
)))
544 VM_BUG_ON(migratetype
== -1);
546 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
548 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
549 VM_BUG_ON(bad_range(zone
, page
));
551 while (order
< MAX_ORDER
-1) {
552 buddy_idx
= __find_buddy_index(page_idx
, order
);
553 buddy
= page
+ (buddy_idx
- page_idx
);
554 if (!page_is_buddy(page
, buddy
, order
))
557 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
558 * merge with it and move up one order.
560 if (page_is_guard(buddy
)) {
561 clear_page_guard_flag(buddy
);
562 set_page_private(page
, 0);
563 __mod_zone_freepage_state(zone
, 1 << order
,
566 list_del(&buddy
->lru
);
567 zone
->free_area
[order
].nr_free
--;
568 rmv_page_order(buddy
);
570 combined_idx
= buddy_idx
& page_idx
;
571 page
= page
+ (combined_idx
- page_idx
);
572 page_idx
= combined_idx
;
575 set_page_order(page
, order
);
578 * If this is not the largest possible page, check if the buddy
579 * of the next-highest order is free. If it is, it's possible
580 * that pages are being freed that will coalesce soon. In case,
581 * that is happening, add the free page to the tail of the list
582 * so it's less likely to be used soon and more likely to be merged
583 * as a higher order page
585 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
586 struct page
*higher_page
, *higher_buddy
;
587 combined_idx
= buddy_idx
& page_idx
;
588 higher_page
= page
+ (combined_idx
- page_idx
);
589 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
590 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
591 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
592 list_add_tail(&page
->lru
,
593 &zone
->free_area
[order
].free_list
[migratetype
]);
598 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
600 zone
->free_area
[order
].nr_free
++;
603 static inline int free_pages_check(struct page
*page
)
605 if (unlikely(page_mapcount(page
) |
606 (page
->mapping
!= NULL
) |
607 (atomic_read(&page
->_count
) != 0) |
608 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
) |
609 (mem_cgroup_bad_page_check(page
)))) {
613 reset_page_last_nid(page
);
614 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
615 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
620 * Frees a number of pages from the PCP lists
621 * Assumes all pages on list are in same zone, and of same order.
622 * count is the number of pages to free.
624 * If the zone was previously in an "all pages pinned" state then look to
625 * see if this freeing clears that state.
627 * And clear the zone's pages_scanned counter, to hold off the "all pages are
628 * pinned" detection logic.
630 static void free_pcppages_bulk(struct zone
*zone
, int count
,
631 struct per_cpu_pages
*pcp
)
637 spin_lock(&zone
->lock
);
638 zone
->all_unreclaimable
= 0;
639 zone
->pages_scanned
= 0;
643 struct list_head
*list
;
646 * Remove pages from lists in a round-robin fashion. A
647 * batch_free count is maintained that is incremented when an
648 * empty list is encountered. This is so more pages are freed
649 * off fuller lists instead of spinning excessively around empty
654 if (++migratetype
== MIGRATE_PCPTYPES
)
656 list
= &pcp
->lists
[migratetype
];
657 } while (list_empty(list
));
659 /* This is the only non-empty list. Free them all. */
660 if (batch_free
== MIGRATE_PCPTYPES
)
661 batch_free
= to_free
;
664 int mt
; /* migratetype of the to-be-freed page */
666 page
= list_entry(list
->prev
, struct page
, lru
);
667 /* must delete as __free_one_page list manipulates */
668 list_del(&page
->lru
);
669 mt
= get_freepage_migratetype(page
);
670 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
671 __free_one_page(page
, zone
, 0, mt
);
672 trace_mm_page_pcpu_drain(page
, 0, mt
);
673 if (likely(get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)) {
674 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1);
675 if (is_migrate_cma(mt
))
676 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
, 1);
678 } while (--to_free
&& --batch_free
&& !list_empty(list
));
680 spin_unlock(&zone
->lock
);
683 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
686 spin_lock(&zone
->lock
);
687 zone
->all_unreclaimable
= 0;
688 zone
->pages_scanned
= 0;
690 __free_one_page(page
, zone
, order
, migratetype
);
691 if (unlikely(migratetype
!= MIGRATE_ISOLATE
))
692 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
693 spin_unlock(&zone
->lock
);
696 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
701 trace_mm_page_free(page
, order
);
702 kmemcheck_free_shadow(page
, order
);
705 page
->mapping
= NULL
;
706 for (i
= 0; i
< (1 << order
); i
++)
707 bad
+= free_pages_check(page
+ i
);
711 if (!PageHighMem(page
)) {
712 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
713 debug_check_no_obj_freed(page_address(page
),
716 arch_free_page(page
, order
);
717 kernel_map_pages(page
, 1 << order
, 0);
722 static void __free_pages_ok(struct page
*page
, unsigned int order
)
727 if (!free_pages_prepare(page
, order
))
730 local_irq_save(flags
);
731 __count_vm_events(PGFREE
, 1 << order
);
732 migratetype
= get_pageblock_migratetype(page
);
733 set_freepage_migratetype(page
, migratetype
);
734 free_one_page(page_zone(page
), page
, order
, migratetype
);
735 local_irq_restore(flags
);
739 * Read access to zone->managed_pages is safe because it's unsigned long,
740 * but we still need to serialize writers. Currently all callers of
741 * __free_pages_bootmem() except put_page_bootmem() should only be used
742 * at boot time. So for shorter boot time, we shift the burden to
743 * put_page_bootmem() to serialize writers.
745 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
747 unsigned int nr_pages
= 1 << order
;
751 for (loop
= 0; loop
< nr_pages
; loop
++) {
752 struct page
*p
= &page
[loop
];
754 if (loop
+ 1 < nr_pages
)
756 __ClearPageReserved(p
);
757 set_page_count(p
, 0);
760 page_zone(page
)->managed_pages
+= 1 << order
;
761 set_page_refcounted(page
);
762 __free_pages(page
, order
);
766 /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
767 void __init
init_cma_reserved_pageblock(struct page
*page
)
769 unsigned i
= pageblock_nr_pages
;
770 struct page
*p
= page
;
773 __ClearPageReserved(p
);
774 set_page_count(p
, 0);
777 set_page_refcounted(page
);
778 set_pageblock_migratetype(page
, MIGRATE_CMA
);
779 __free_pages(page
, pageblock_order
);
780 totalram_pages
+= pageblock_nr_pages
;
785 * The order of subdivision here is critical for the IO subsystem.
786 * Please do not alter this order without good reasons and regression
787 * testing. Specifically, as large blocks of memory are subdivided,
788 * the order in which smaller blocks are delivered depends on the order
789 * they're subdivided in this function. This is the primary factor
790 * influencing the order in which pages are delivered to the IO
791 * subsystem according to empirical testing, and this is also justified
792 * by considering the behavior of a buddy system containing a single
793 * large block of memory acted on by a series of small allocations.
794 * This behavior is a critical factor in sglist merging's success.
798 static inline void expand(struct zone
*zone
, struct page
*page
,
799 int low
, int high
, struct free_area
*area
,
802 unsigned long size
= 1 << high
;
808 VM_BUG_ON(bad_range(zone
, &page
[size
]));
810 #ifdef CONFIG_DEBUG_PAGEALLOC
811 if (high
< debug_guardpage_minorder()) {
813 * Mark as guard pages (or page), that will allow to
814 * merge back to allocator when buddy will be freed.
815 * Corresponding page table entries will not be touched,
816 * pages will stay not present in virtual address space
818 INIT_LIST_HEAD(&page
[size
].lru
);
819 set_page_guard_flag(&page
[size
]);
820 set_page_private(&page
[size
], high
);
821 /* Guard pages are not available for any usage */
822 __mod_zone_freepage_state(zone
, -(1 << high
),
827 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
829 set_page_order(&page
[size
], high
);
834 * This page is about to be returned from the page allocator
836 static inline int check_new_page(struct page
*page
)
838 if (unlikely(page_mapcount(page
) |
839 (page
->mapping
!= NULL
) |
840 (atomic_read(&page
->_count
) != 0) |
841 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
) |
842 (mem_cgroup_bad_page_check(page
)))) {
849 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
853 for (i
= 0; i
< (1 << order
); i
++) {
854 struct page
*p
= page
+ i
;
855 if (unlikely(check_new_page(p
)))
859 set_page_private(page
, 0);
860 set_page_refcounted(page
);
862 arch_alloc_page(page
, order
);
863 kernel_map_pages(page
, 1 << order
, 1);
865 if (gfp_flags
& __GFP_ZERO
)
866 prep_zero_page(page
, order
, gfp_flags
);
868 if (order
&& (gfp_flags
& __GFP_COMP
))
869 prep_compound_page(page
, order
);
875 * Go through the free lists for the given migratetype and remove
876 * the smallest available page from the freelists
879 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
882 unsigned int current_order
;
883 struct free_area
* area
;
886 /* Find a page of the appropriate size in the preferred list */
887 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
888 area
= &(zone
->free_area
[current_order
]);
889 if (list_empty(&area
->free_list
[migratetype
]))
892 page
= list_entry(area
->free_list
[migratetype
].next
,
894 list_del(&page
->lru
);
895 rmv_page_order(page
);
897 expand(zone
, page
, order
, current_order
, area
, migratetype
);
906 * This array describes the order lists are fallen back to when
907 * the free lists for the desirable migrate type are depleted
909 static int fallbacks
[MIGRATE_TYPES
][4] = {
910 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
911 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
913 [MIGRATE_MOVABLE
] = { MIGRATE_CMA
, MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
914 [MIGRATE_CMA
] = { MIGRATE_RESERVE
}, /* Never used */
916 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
918 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
}, /* Never used */
919 [MIGRATE_ISOLATE
] = { MIGRATE_RESERVE
}, /* Never used */
923 * Move the free pages in a range to the free lists of the requested type.
924 * Note that start_page and end_pages are not aligned on a pageblock
925 * boundary. If alignment is required, use move_freepages_block()
927 int move_freepages(struct zone
*zone
,
928 struct page
*start_page
, struct page
*end_page
,
935 #ifndef CONFIG_HOLES_IN_ZONE
937 * page_zone is not safe to call in this context when
938 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
939 * anyway as we check zone boundaries in move_freepages_block().
940 * Remove at a later date when no bug reports exist related to
941 * grouping pages by mobility
943 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
946 for (page
= start_page
; page
<= end_page
;) {
947 /* Make sure we are not inadvertently changing nodes */
948 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
950 if (!pfn_valid_within(page_to_pfn(page
))) {
955 if (!PageBuddy(page
)) {
960 order
= page_order(page
);
961 list_move(&page
->lru
,
962 &zone
->free_area
[order
].free_list
[migratetype
]);
963 set_freepage_migratetype(page
, migratetype
);
965 pages_moved
+= 1 << order
;
971 int move_freepages_block(struct zone
*zone
, struct page
*page
,
974 unsigned long start_pfn
, end_pfn
;
975 struct page
*start_page
, *end_page
;
977 start_pfn
= page_to_pfn(page
);
978 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
979 start_page
= pfn_to_page(start_pfn
);
980 end_page
= start_page
+ pageblock_nr_pages
- 1;
981 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
983 /* Do not cross zone boundaries */
984 if (start_pfn
< zone
->zone_start_pfn
)
986 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
989 return move_freepages(zone
, start_page
, end_page
, migratetype
);
992 static void change_pageblock_range(struct page
*pageblock_page
,
993 int start_order
, int migratetype
)
995 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
997 while (nr_pageblocks
--) {
998 set_pageblock_migratetype(pageblock_page
, migratetype
);
999 pageblock_page
+= pageblock_nr_pages
;
1003 /* Remove an element from the buddy allocator from the fallback list */
1004 static inline struct page
*
1005 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
1007 struct free_area
* area
;
1012 /* Find the largest possible block of pages in the other list */
1013 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
1016 migratetype
= fallbacks
[start_migratetype
][i
];
1018 /* MIGRATE_RESERVE handled later if necessary */
1019 if (migratetype
== MIGRATE_RESERVE
)
1022 area
= &(zone
->free_area
[current_order
]);
1023 if (list_empty(&area
->free_list
[migratetype
]))
1026 page
= list_entry(area
->free_list
[migratetype
].next
,
1031 * If breaking a large block of pages, move all free
1032 * pages to the preferred allocation list. If falling
1033 * back for a reclaimable kernel allocation, be more
1034 * aggressive about taking ownership of free pages
1036 * On the other hand, never change migration
1037 * type of MIGRATE_CMA pageblocks nor move CMA
1038 * pages on different free lists. We don't
1039 * want unmovable pages to be allocated from
1040 * MIGRATE_CMA areas.
1042 if (!is_migrate_cma(migratetype
) &&
1043 (unlikely(current_order
>= pageblock_order
/ 2) ||
1044 start_migratetype
== MIGRATE_RECLAIMABLE
||
1045 page_group_by_mobility_disabled
)) {
1047 pages
= move_freepages_block(zone
, page
,
1050 /* Claim the whole block if over half of it is free */
1051 if (pages
>= (1 << (pageblock_order
-1)) ||
1052 page_group_by_mobility_disabled
)
1053 set_pageblock_migratetype(page
,
1056 migratetype
= start_migratetype
;
1059 /* Remove the page from the freelists */
1060 list_del(&page
->lru
);
1061 rmv_page_order(page
);
1063 /* Take ownership for orders >= pageblock_order */
1064 if (current_order
>= pageblock_order
&&
1065 !is_migrate_cma(migratetype
))
1066 change_pageblock_range(page
, current_order
,
1069 expand(zone
, page
, order
, current_order
, area
,
1070 is_migrate_cma(migratetype
)
1071 ? migratetype
: start_migratetype
);
1073 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1074 start_migratetype
, migratetype
);
1084 * Do the hard work of removing an element from the buddy allocator.
1085 * Call me with the zone->lock already held.
1087 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1093 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1095 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1096 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1099 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1100 * is used because __rmqueue_smallest is an inline function
1101 * and we want just one call site
1104 migratetype
= MIGRATE_RESERVE
;
1109 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1114 * Obtain a specified number of elements from the buddy allocator, all under
1115 * a single hold of the lock, for efficiency. Add them to the supplied list.
1116 * Returns the number of new pages which were placed at *list.
1118 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1119 unsigned long count
, struct list_head
*list
,
1120 int migratetype
, int cold
)
1122 int mt
= migratetype
, i
;
1124 spin_lock(&zone
->lock
);
1125 for (i
= 0; i
< count
; ++i
) {
1126 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1127 if (unlikely(page
== NULL
))
1131 * Split buddy pages returned by expand() are received here
1132 * in physical page order. The page is added to the callers and
1133 * list and the list head then moves forward. From the callers
1134 * perspective, the linked list is ordered by page number in
1135 * some conditions. This is useful for IO devices that can
1136 * merge IO requests if the physical pages are ordered
1139 if (likely(cold
== 0))
1140 list_add(&page
->lru
, list
);
1142 list_add_tail(&page
->lru
, list
);
1143 if (IS_ENABLED(CONFIG_CMA
)) {
1144 mt
= get_pageblock_migratetype(page
);
1145 if (!is_migrate_cma(mt
) && mt
!= MIGRATE_ISOLATE
)
1148 set_freepage_migratetype(page
, mt
);
1150 if (is_migrate_cma(mt
))
1151 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1154 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1155 spin_unlock(&zone
->lock
);
1161 * Called from the vmstat counter updater to drain pagesets of this
1162 * currently executing processor on remote nodes after they have
1165 * Note that this function must be called with the thread pinned to
1166 * a single processor.
1168 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1170 unsigned long flags
;
1173 local_irq_save(flags
);
1174 if (pcp
->count
>= pcp
->batch
)
1175 to_drain
= pcp
->batch
;
1177 to_drain
= pcp
->count
;
1179 free_pcppages_bulk(zone
, to_drain
, pcp
);
1180 pcp
->count
-= to_drain
;
1182 local_irq_restore(flags
);
1187 * Drain pages of the indicated processor.
1189 * The processor must either be the current processor and the
1190 * thread pinned to the current processor or a processor that
1193 static void drain_pages(unsigned int cpu
)
1195 unsigned long flags
;
1198 for_each_populated_zone(zone
) {
1199 struct per_cpu_pageset
*pset
;
1200 struct per_cpu_pages
*pcp
;
1202 local_irq_save(flags
);
1203 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1207 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1210 local_irq_restore(flags
);
1215 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1217 void drain_local_pages(void *arg
)
1219 drain_pages(smp_processor_id());
1223 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1225 * Note that this code is protected against sending an IPI to an offline
1226 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1227 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1228 * nothing keeps CPUs from showing up after we populated the cpumask and
1229 * before the call to on_each_cpu_mask().
1231 void drain_all_pages(void)
1234 struct per_cpu_pageset
*pcp
;
1238 * Allocate in the BSS so we wont require allocation in
1239 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1241 static cpumask_t cpus_with_pcps
;
1244 * We don't care about racing with CPU hotplug event
1245 * as offline notification will cause the notified
1246 * cpu to drain that CPU pcps and on_each_cpu_mask
1247 * disables preemption as part of its processing
1249 for_each_online_cpu(cpu
) {
1250 bool has_pcps
= false;
1251 for_each_populated_zone(zone
) {
1252 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1253 if (pcp
->pcp
.count
) {
1259 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1261 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1263 on_each_cpu_mask(&cpus_with_pcps
, drain_local_pages
, NULL
, 1);
1266 #ifdef CONFIG_HIBERNATION
1268 void mark_free_pages(struct zone
*zone
)
1270 unsigned long pfn
, max_zone_pfn
;
1271 unsigned long flags
;
1273 struct list_head
*curr
;
1275 if (!zone
->spanned_pages
)
1278 spin_lock_irqsave(&zone
->lock
, flags
);
1280 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1281 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1282 if (pfn_valid(pfn
)) {
1283 struct page
*page
= pfn_to_page(pfn
);
1285 if (!swsusp_page_is_forbidden(page
))
1286 swsusp_unset_page_free(page
);
1289 for_each_migratetype_order(order
, t
) {
1290 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1293 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1294 for (i
= 0; i
< (1UL << order
); i
++)
1295 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1298 spin_unlock_irqrestore(&zone
->lock
, flags
);
1300 #endif /* CONFIG_PM */
1303 * Free a 0-order page
1304 * cold == 1 ? free a cold page : free a hot page
1306 void free_hot_cold_page(struct page
*page
, int cold
)
1308 struct zone
*zone
= page_zone(page
);
1309 struct per_cpu_pages
*pcp
;
1310 unsigned long flags
;
1313 if (!free_pages_prepare(page
, 0))
1316 migratetype
= get_pageblock_migratetype(page
);
1317 set_freepage_migratetype(page
, migratetype
);
1318 local_irq_save(flags
);
1319 __count_vm_event(PGFREE
);
1322 * We only track unmovable, reclaimable and movable on pcp lists.
1323 * Free ISOLATE pages back to the allocator because they are being
1324 * offlined but treat RESERVE as movable pages so we can get those
1325 * areas back if necessary. Otherwise, we may have to free
1326 * excessively into the page allocator
1328 if (migratetype
>= MIGRATE_PCPTYPES
) {
1329 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1330 free_one_page(zone
, page
, 0, migratetype
);
1333 migratetype
= MIGRATE_MOVABLE
;
1336 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1338 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1340 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1342 if (pcp
->count
>= pcp
->high
) {
1343 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1344 pcp
->count
-= pcp
->batch
;
1348 local_irq_restore(flags
);
1352 * Free a list of 0-order pages
1354 void free_hot_cold_page_list(struct list_head
*list
, int cold
)
1356 struct page
*page
, *next
;
1358 list_for_each_entry_safe(page
, next
, list
, lru
) {
1359 trace_mm_page_free_batched(page
, cold
);
1360 free_hot_cold_page(page
, cold
);
1365 * split_page takes a non-compound higher-order page, and splits it into
1366 * n (1<<order) sub-pages: page[0..n]
1367 * Each sub-page must be freed individually.
1369 * Note: this is probably too low level an operation for use in drivers.
1370 * Please consult with lkml before using this in your driver.
1372 void split_page(struct page
*page
, unsigned int order
)
1376 VM_BUG_ON(PageCompound(page
));
1377 VM_BUG_ON(!page_count(page
));
1379 #ifdef CONFIG_KMEMCHECK
1381 * Split shadow pages too, because free(page[0]) would
1382 * otherwise free the whole shadow.
1384 if (kmemcheck_page_is_tracked(page
))
1385 split_page(virt_to_page(page
[0].shadow
), order
);
1388 for (i
= 1; i
< (1 << order
); i
++)
1389 set_page_refcounted(page
+ i
);
1393 * Similar to the split_page family of functions except that the page
1394 * required at the given order and being isolated now to prevent races
1395 * with parallel allocators
1397 int capture_free_page(struct page
*page
, int alloc_order
, int migratetype
)
1400 unsigned long watermark
;
1404 BUG_ON(!PageBuddy(page
));
1406 zone
= page_zone(page
);
1407 order
= page_order(page
);
1408 mt
= get_pageblock_migratetype(page
);
1410 if (mt
!= MIGRATE_ISOLATE
) {
1411 /* Obey watermarks as if the page was being allocated */
1412 watermark
= low_wmark_pages(zone
) + (1 << order
);
1413 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1416 __mod_zone_freepage_state(zone
, -(1UL << alloc_order
), mt
);
1419 /* Remove page from free list */
1420 list_del(&page
->lru
);
1421 zone
->free_area
[order
].nr_free
--;
1422 rmv_page_order(page
);
1424 if (alloc_order
!= order
)
1425 expand(zone
, page
, alloc_order
, order
,
1426 &zone
->free_area
[order
], migratetype
);
1428 /* Set the pageblock if the captured page is at least a pageblock */
1429 if (order
>= pageblock_order
- 1) {
1430 struct page
*endpage
= page
+ (1 << order
) - 1;
1431 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
1432 int mt
= get_pageblock_migratetype(page
);
1433 if (mt
!= MIGRATE_ISOLATE
&& !is_migrate_cma(mt
))
1434 set_pageblock_migratetype(page
,
1439 return 1UL << alloc_order
;
1443 * Similar to split_page except the page is already free. As this is only
1444 * being used for migration, the migratetype of the block also changes.
1445 * As this is called with interrupts disabled, the caller is responsible
1446 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1449 * Note: this is probably too low level an operation for use in drivers.
1450 * Please consult with lkml before using this in your driver.
1452 int split_free_page(struct page
*page
)
1457 BUG_ON(!PageBuddy(page
));
1458 order
= page_order(page
);
1460 nr_pages
= capture_free_page(page
, order
, 0);
1464 /* Split into individual pages */
1465 set_page_refcounted(page
);
1466 split_page(page
, order
);
1471 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1472 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1476 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1477 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1480 unsigned long flags
;
1482 int cold
= !!(gfp_flags
& __GFP_COLD
);
1485 if (likely(order
== 0)) {
1486 struct per_cpu_pages
*pcp
;
1487 struct list_head
*list
;
1489 local_irq_save(flags
);
1490 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1491 list
= &pcp
->lists
[migratetype
];
1492 if (list_empty(list
)) {
1493 pcp
->count
+= rmqueue_bulk(zone
, 0,
1496 if (unlikely(list_empty(list
)))
1501 page
= list_entry(list
->prev
, struct page
, lru
);
1503 page
= list_entry(list
->next
, struct page
, lru
);
1505 list_del(&page
->lru
);
1508 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1510 * __GFP_NOFAIL is not to be used in new code.
1512 * All __GFP_NOFAIL callers should be fixed so that they
1513 * properly detect and handle allocation failures.
1515 * We most definitely don't want callers attempting to
1516 * allocate greater than order-1 page units with
1519 WARN_ON_ONCE(order
> 1);
1521 spin_lock_irqsave(&zone
->lock
, flags
);
1522 page
= __rmqueue(zone
, order
, migratetype
);
1523 spin_unlock(&zone
->lock
);
1526 __mod_zone_freepage_state(zone
, -(1 << order
),
1527 get_pageblock_migratetype(page
));
1530 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1531 zone_statistics(preferred_zone
, zone
, gfp_flags
);
1532 local_irq_restore(flags
);
1534 VM_BUG_ON(bad_range(zone
, page
));
1535 if (prep_new_page(page
, order
, gfp_flags
))
1540 local_irq_restore(flags
);
1544 #ifdef CONFIG_FAIL_PAGE_ALLOC
1547 struct fault_attr attr
;
1549 u32 ignore_gfp_highmem
;
1550 u32 ignore_gfp_wait
;
1552 } fail_page_alloc
= {
1553 .attr
= FAULT_ATTR_INITIALIZER
,
1554 .ignore_gfp_wait
= 1,
1555 .ignore_gfp_highmem
= 1,
1559 static int __init
setup_fail_page_alloc(char *str
)
1561 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1563 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1565 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1567 if (order
< fail_page_alloc
.min_order
)
1569 if (gfp_mask
& __GFP_NOFAIL
)
1571 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1573 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1576 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1579 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1581 static int __init
fail_page_alloc_debugfs(void)
1583 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1586 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
1587 &fail_page_alloc
.attr
);
1589 return PTR_ERR(dir
);
1591 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1592 &fail_page_alloc
.ignore_gfp_wait
))
1594 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1595 &fail_page_alloc
.ignore_gfp_highmem
))
1597 if (!debugfs_create_u32("min-order", mode
, dir
,
1598 &fail_page_alloc
.min_order
))
1603 debugfs_remove_recursive(dir
);
1608 late_initcall(fail_page_alloc_debugfs
);
1610 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1612 #else /* CONFIG_FAIL_PAGE_ALLOC */
1614 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1619 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1622 * Return true if free pages are above 'mark'. This takes into account the order
1623 * of the allocation.
1625 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1626 int classzone_idx
, int alloc_flags
, long free_pages
)
1628 /* free_pages my go negative - that's OK */
1630 long lowmem_reserve
= z
->lowmem_reserve
[classzone_idx
];
1633 free_pages
-= (1 << order
) - 1;
1634 if (alloc_flags
& ALLOC_HIGH
)
1636 if (alloc_flags
& ALLOC_HARDER
)
1639 /* If allocation can't use CMA areas don't use free CMA pages */
1640 if (!(alloc_flags
& ALLOC_CMA
))
1641 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
1643 if (free_pages
<= min
+ lowmem_reserve
)
1645 for (o
= 0; o
< order
; o
++) {
1646 /* At the next order, this order's pages become unavailable */
1647 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1649 /* Require fewer higher order pages to be free */
1652 if (free_pages
<= min
)
1658 #ifdef CONFIG_MEMORY_ISOLATION
1659 static inline unsigned long nr_zone_isolate_freepages(struct zone
*zone
)
1661 if (unlikely(zone
->nr_pageblock_isolate
))
1662 return zone
->nr_pageblock_isolate
* pageblock_nr_pages
;
1666 static inline unsigned long nr_zone_isolate_freepages(struct zone
*zone
)
1672 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1673 int classzone_idx
, int alloc_flags
)
1675 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1676 zone_page_state(z
, NR_FREE_PAGES
));
1679 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1680 int classzone_idx
, int alloc_flags
)
1682 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1684 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1685 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1688 * If the zone has MIGRATE_ISOLATE type free pages, we should consider
1689 * it. nr_zone_isolate_freepages is never accurate so kswapd might not
1690 * sleep although it could do so. But this is more desirable for memory
1691 * hotplug than sleeping which can cause a livelock in the direct
1694 free_pages
-= nr_zone_isolate_freepages(z
);
1695 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1701 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1702 * skip over zones that are not allowed by the cpuset, or that have
1703 * been recently (in last second) found to be nearly full. See further
1704 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1705 * that have to skip over a lot of full or unallowed zones.
1707 * If the zonelist cache is present in the passed in zonelist, then
1708 * returns a pointer to the allowed node mask (either the current
1709 * tasks mems_allowed, or node_states[N_MEMORY].)
1711 * If the zonelist cache is not available for this zonelist, does
1712 * nothing and returns NULL.
1714 * If the fullzones BITMAP in the zonelist cache is stale (more than
1715 * a second since last zap'd) then we zap it out (clear its bits.)
1717 * We hold off even calling zlc_setup, until after we've checked the
1718 * first zone in the zonelist, on the theory that most allocations will
1719 * be satisfied from that first zone, so best to examine that zone as
1720 * quickly as we can.
1722 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1724 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1725 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1727 zlc
= zonelist
->zlcache_ptr
;
1731 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1732 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1733 zlc
->last_full_zap
= jiffies
;
1736 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1737 &cpuset_current_mems_allowed
:
1738 &node_states
[N_MEMORY
];
1739 return allowednodes
;
1743 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1744 * if it is worth looking at further for free memory:
1745 * 1) Check that the zone isn't thought to be full (doesn't have its
1746 * bit set in the zonelist_cache fullzones BITMAP).
1747 * 2) Check that the zones node (obtained from the zonelist_cache
1748 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1749 * Return true (non-zero) if zone is worth looking at further, or
1750 * else return false (zero) if it is not.
1752 * This check -ignores- the distinction between various watermarks,
1753 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1754 * found to be full for any variation of these watermarks, it will
1755 * be considered full for up to one second by all requests, unless
1756 * we are so low on memory on all allowed nodes that we are forced
1757 * into the second scan of the zonelist.
1759 * In the second scan we ignore this zonelist cache and exactly
1760 * apply the watermarks to all zones, even it is slower to do so.
1761 * We are low on memory in the second scan, and should leave no stone
1762 * unturned looking for a free page.
1764 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1765 nodemask_t
*allowednodes
)
1767 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1768 int i
; /* index of *z in zonelist zones */
1769 int n
; /* node that zone *z is on */
1771 zlc
= zonelist
->zlcache_ptr
;
1775 i
= z
- zonelist
->_zonerefs
;
1778 /* This zone is worth trying if it is allowed but not full */
1779 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1783 * Given 'z' scanning a zonelist, set the corresponding bit in
1784 * zlc->fullzones, so that subsequent attempts to allocate a page
1785 * from that zone don't waste time re-examining it.
1787 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1789 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1790 int i
; /* index of *z in zonelist zones */
1792 zlc
= zonelist
->zlcache_ptr
;
1796 i
= z
- zonelist
->_zonerefs
;
1798 set_bit(i
, zlc
->fullzones
);
1802 * clear all zones full, called after direct reclaim makes progress so that
1803 * a zone that was recently full is not skipped over for up to a second
1805 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1807 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1809 zlc
= zonelist
->zlcache_ptr
;
1813 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1816 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1818 return node_isset(local_zone
->node
, zone
->zone_pgdat
->reclaim_nodes
);
1821 static void __paginginit
init_zone_allows_reclaim(int nid
)
1825 for_each_online_node(i
)
1826 if (node_distance(nid
, i
) <= RECLAIM_DISTANCE
)
1827 node_set(i
, NODE_DATA(nid
)->reclaim_nodes
);
1829 zone_reclaim_mode
= 1;
1832 #else /* CONFIG_NUMA */
1834 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1839 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1840 nodemask_t
*allowednodes
)
1845 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1849 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1853 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1858 static inline void init_zone_allows_reclaim(int nid
)
1861 #endif /* CONFIG_NUMA */
1864 * get_page_from_freelist goes through the zonelist trying to allocate
1867 static struct page
*
1868 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1869 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1870 struct zone
*preferred_zone
, int migratetype
)
1873 struct page
*page
= NULL
;
1876 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1877 int zlc_active
= 0; /* set if using zonelist_cache */
1878 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1880 classzone_idx
= zone_idx(preferred_zone
);
1883 * Scan zonelist, looking for a zone with enough free.
1884 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1886 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1887 high_zoneidx
, nodemask
) {
1888 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1889 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1891 if ((alloc_flags
& ALLOC_CPUSET
) &&
1892 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1895 * When allocating a page cache page for writing, we
1896 * want to get it from a zone that is within its dirty
1897 * limit, such that no single zone holds more than its
1898 * proportional share of globally allowed dirty pages.
1899 * The dirty limits take into account the zone's
1900 * lowmem reserves and high watermark so that kswapd
1901 * should be able to balance it without having to
1902 * write pages from its LRU list.
1904 * This may look like it could increase pressure on
1905 * lower zones by failing allocations in higher zones
1906 * before they are full. But the pages that do spill
1907 * over are limited as the lower zones are protected
1908 * by this very same mechanism. It should not become
1909 * a practical burden to them.
1911 * XXX: For now, allow allocations to potentially
1912 * exceed the per-zone dirty limit in the slowpath
1913 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1914 * which is important when on a NUMA setup the allowed
1915 * zones are together not big enough to reach the
1916 * global limit. The proper fix for these situations
1917 * will require awareness of zones in the
1918 * dirty-throttling and the flusher threads.
1920 if ((alloc_flags
& ALLOC_WMARK_LOW
) &&
1921 (gfp_mask
& __GFP_WRITE
) && !zone_dirty_ok(zone
))
1922 goto this_zone_full
;
1924 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1925 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1929 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1930 if (zone_watermark_ok(zone
, order
, mark
,
1931 classzone_idx
, alloc_flags
))
1934 if (IS_ENABLED(CONFIG_NUMA
) &&
1935 !did_zlc_setup
&& nr_online_nodes
> 1) {
1937 * we do zlc_setup if there are multiple nodes
1938 * and before considering the first zone allowed
1941 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1946 if (zone_reclaim_mode
== 0 ||
1947 !zone_allows_reclaim(preferred_zone
, zone
))
1948 goto this_zone_full
;
1951 * As we may have just activated ZLC, check if the first
1952 * eligible zone has failed zone_reclaim recently.
1954 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1955 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1958 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1960 case ZONE_RECLAIM_NOSCAN
:
1963 case ZONE_RECLAIM_FULL
:
1964 /* scanned but unreclaimable */
1967 /* did we reclaim enough */
1968 if (!zone_watermark_ok(zone
, order
, mark
,
1969 classzone_idx
, alloc_flags
))
1970 goto this_zone_full
;
1975 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1976 gfp_mask
, migratetype
);
1980 if (IS_ENABLED(CONFIG_NUMA
))
1981 zlc_mark_zone_full(zonelist
, z
);
1984 if (unlikely(IS_ENABLED(CONFIG_NUMA
) && page
== NULL
&& zlc_active
)) {
1985 /* Disable zlc cache for second zonelist scan */
1992 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
1993 * necessary to allocate the page. The expectation is
1994 * that the caller is taking steps that will free more
1995 * memory. The caller should avoid the page being used
1996 * for !PFMEMALLOC purposes.
1998 page
->pfmemalloc
= !!(alloc_flags
& ALLOC_NO_WATERMARKS
);
2004 * Large machines with many possible nodes should not always dump per-node
2005 * meminfo in irq context.
2007 static inline bool should_suppress_show_mem(void)
2012 ret
= in_interrupt();
2017 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2018 DEFAULT_RATELIMIT_INTERVAL
,
2019 DEFAULT_RATELIMIT_BURST
);
2021 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
2023 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2025 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2026 debug_guardpage_minorder() > 0)
2030 * This documents exceptions given to allocations in certain
2031 * contexts that are allowed to allocate outside current's set
2034 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2035 if (test_thread_flag(TIF_MEMDIE
) ||
2036 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2037 filter
&= ~SHOW_MEM_FILTER_NODES
;
2038 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
2039 filter
&= ~SHOW_MEM_FILTER_NODES
;
2042 struct va_format vaf
;
2045 va_start(args
, fmt
);
2050 pr_warn("%pV", &vaf
);
2055 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2056 current
->comm
, order
, gfp_mask
);
2059 if (!should_suppress_show_mem())
2064 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
2065 unsigned long did_some_progress
,
2066 unsigned long pages_reclaimed
)
2068 /* Do not loop if specifically requested */
2069 if (gfp_mask
& __GFP_NORETRY
)
2072 /* Always retry if specifically requested */
2073 if (gfp_mask
& __GFP_NOFAIL
)
2077 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2078 * making forward progress without invoking OOM. Suspend also disables
2079 * storage devices so kswapd will not help. Bail if we are suspending.
2081 if (!did_some_progress
&& pm_suspended_storage())
2085 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2086 * means __GFP_NOFAIL, but that may not be true in other
2089 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
2093 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2094 * specified, then we retry until we no longer reclaim any pages
2095 * (above), or we've reclaimed an order of pages at least as
2096 * large as the allocation's order. In both cases, if the
2097 * allocation still fails, we stop retrying.
2099 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
2105 static inline struct page
*
2106 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2107 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2108 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2113 /* Acquire the OOM killer lock for the zones in zonelist */
2114 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
2115 schedule_timeout_uninterruptible(1);
2120 * Go through the zonelist yet one more time, keep very high watermark
2121 * here, this is only to catch a parallel oom killing, we must fail if
2122 * we're still under heavy pressure.
2124 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
2125 order
, zonelist
, high_zoneidx
,
2126 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
2127 preferred_zone
, migratetype
);
2131 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2132 /* The OOM killer will not help higher order allocs */
2133 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2135 /* The OOM killer does not needlessly kill tasks for lowmem */
2136 if (high_zoneidx
< ZONE_NORMAL
)
2139 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2140 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2141 * The caller should handle page allocation failure by itself if
2142 * it specifies __GFP_THISNODE.
2143 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2145 if (gfp_mask
& __GFP_THISNODE
)
2148 /* Exhausted what can be done so it's blamo time */
2149 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
, false);
2152 clear_zonelist_oom(zonelist
, gfp_mask
);
2156 #ifdef CONFIG_COMPACTION
2157 /* Try memory compaction for high-order allocations before reclaim */
2158 static struct page
*
2159 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2160 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2161 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2162 int migratetype
, bool sync_migration
,
2163 bool *contended_compaction
, bool *deferred_compaction
,
2164 unsigned long *did_some_progress
)
2166 struct page
*page
= NULL
;
2171 if (compaction_deferred(preferred_zone
, order
)) {
2172 *deferred_compaction
= true;
2176 current
->flags
|= PF_MEMALLOC
;
2177 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
2178 nodemask
, sync_migration
,
2179 contended_compaction
, &page
);
2180 current
->flags
&= ~PF_MEMALLOC
;
2182 /* If compaction captured a page, prep and use it */
2184 prep_new_page(page
, order
, gfp_mask
);
2188 if (*did_some_progress
!= COMPACT_SKIPPED
) {
2189 /* Page migration frees to the PCP lists but we want merging */
2190 drain_pages(get_cpu());
2193 page
= get_page_from_freelist(gfp_mask
, nodemask
,
2194 order
, zonelist
, high_zoneidx
,
2195 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2196 preferred_zone
, migratetype
);
2199 preferred_zone
->compact_blockskip_flush
= false;
2200 preferred_zone
->compact_considered
= 0;
2201 preferred_zone
->compact_defer_shift
= 0;
2202 if (order
>= preferred_zone
->compact_order_failed
)
2203 preferred_zone
->compact_order_failed
= order
+ 1;
2204 count_vm_event(COMPACTSUCCESS
);
2209 * It's bad if compaction run occurs and fails.
2210 * The most likely reason is that pages exist,
2211 * but not enough to satisfy watermarks.
2213 count_vm_event(COMPACTFAIL
);
2216 * As async compaction considers a subset of pageblocks, only
2217 * defer if the failure was a sync compaction failure.
2220 defer_compaction(preferred_zone
, order
);
2228 static inline struct page
*
2229 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2230 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2231 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2232 int migratetype
, bool sync_migration
,
2233 bool *contended_compaction
, bool *deferred_compaction
,
2234 unsigned long *did_some_progress
)
2238 #endif /* CONFIG_COMPACTION */
2240 /* Perform direct synchronous page reclaim */
2242 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
, struct zonelist
*zonelist
,
2243 nodemask_t
*nodemask
)
2245 struct reclaim_state reclaim_state
;
2250 /* We now go into synchronous reclaim */
2251 cpuset_memory_pressure_bump();
2252 current
->flags
|= PF_MEMALLOC
;
2253 lockdep_set_current_reclaim_state(gfp_mask
);
2254 reclaim_state
.reclaimed_slab
= 0;
2255 current
->reclaim_state
= &reclaim_state
;
2257 progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
2259 current
->reclaim_state
= NULL
;
2260 lockdep_clear_current_reclaim_state();
2261 current
->flags
&= ~PF_MEMALLOC
;
2268 /* The really slow allocator path where we enter direct reclaim */
2269 static inline struct page
*
2270 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2271 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2272 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2273 int migratetype
, unsigned long *did_some_progress
)
2275 struct page
*page
= NULL
;
2276 bool drained
= false;
2278 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
2280 if (unlikely(!(*did_some_progress
)))
2283 /* After successful reclaim, reconsider all zones for allocation */
2284 if (IS_ENABLED(CONFIG_NUMA
))
2285 zlc_clear_zones_full(zonelist
);
2288 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2289 zonelist
, high_zoneidx
,
2290 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2291 preferred_zone
, migratetype
);
2294 * If an allocation failed after direct reclaim, it could be because
2295 * pages are pinned on the per-cpu lists. Drain them and try again
2297 if (!page
&& !drained
) {
2307 * This is called in the allocator slow-path if the allocation request is of
2308 * sufficient urgency to ignore watermarks and take other desperate measures
2310 static inline struct page
*
2311 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2312 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2313 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2319 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2320 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
2321 preferred_zone
, migratetype
);
2323 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2324 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2325 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2331 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
2332 enum zone_type high_zoneidx
,
2333 enum zone_type classzone_idx
)
2338 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
2339 wakeup_kswapd(zone
, order
, classzone_idx
);
2343 gfp_to_alloc_flags(gfp_t gfp_mask
)
2345 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2346 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2348 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2349 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2352 * The caller may dip into page reserves a bit more if the caller
2353 * cannot run direct reclaim, or if the caller has realtime scheduling
2354 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2355 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2357 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2361 * Not worth trying to allocate harder for
2362 * __GFP_NOMEMALLOC even if it can't schedule.
2364 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2365 alloc_flags
|= ALLOC_HARDER
;
2367 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2368 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2370 alloc_flags
&= ~ALLOC_CPUSET
;
2371 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2372 alloc_flags
|= ALLOC_HARDER
;
2374 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2375 if (gfp_mask
& __GFP_MEMALLOC
)
2376 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2377 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2378 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2379 else if (!in_interrupt() &&
2380 ((current
->flags
& PF_MEMALLOC
) ||
2381 unlikely(test_thread_flag(TIF_MEMDIE
))))
2382 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2385 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2386 alloc_flags
|= ALLOC_CMA
;
2391 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2393 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2396 static inline struct page
*
2397 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2398 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2399 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2402 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2403 struct page
*page
= NULL
;
2405 unsigned long pages_reclaimed
= 0;
2406 unsigned long did_some_progress
;
2407 bool sync_migration
= false;
2408 bool deferred_compaction
= false;
2409 bool contended_compaction
= false;
2412 * In the slowpath, we sanity check order to avoid ever trying to
2413 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2414 * be using allocators in order of preference for an area that is
2417 if (order
>= MAX_ORDER
) {
2418 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2423 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2424 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2425 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2426 * using a larger set of nodes after it has established that the
2427 * allowed per node queues are empty and that nodes are
2430 if (IS_ENABLED(CONFIG_NUMA
) &&
2431 (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2435 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2436 wake_all_kswapd(order
, zonelist
, high_zoneidx
,
2437 zone_idx(preferred_zone
));
2440 * OK, we're below the kswapd watermark and have kicked background
2441 * reclaim. Now things get more complex, so set up alloc_flags according
2442 * to how we want to proceed.
2444 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2447 * Find the true preferred zone if the allocation is unconstrained by
2450 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2451 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2455 /* This is the last chance, in general, before the goto nopage. */
2456 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2457 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2458 preferred_zone
, migratetype
);
2462 /* Allocate without watermarks if the context allows */
2463 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2465 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2466 * the allocation is high priority and these type of
2467 * allocations are system rather than user orientated
2469 zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
2471 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2472 zonelist
, high_zoneidx
, nodemask
,
2473 preferred_zone
, migratetype
);
2479 /* Atomic allocations - we can't balance anything */
2483 /* Avoid recursion of direct reclaim */
2484 if (current
->flags
& PF_MEMALLOC
)
2487 /* Avoid allocations with no watermarks from looping endlessly */
2488 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2492 * Try direct compaction. The first pass is asynchronous. Subsequent
2493 * attempts after direct reclaim are synchronous
2495 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2496 zonelist
, high_zoneidx
,
2498 alloc_flags
, preferred_zone
,
2499 migratetype
, sync_migration
,
2500 &contended_compaction
,
2501 &deferred_compaction
,
2502 &did_some_progress
);
2505 sync_migration
= true;
2508 * If compaction is deferred for high-order allocations, it is because
2509 * sync compaction recently failed. In this is the case and the caller
2510 * requested a movable allocation that does not heavily disrupt the
2511 * system then fail the allocation instead of entering direct reclaim.
2513 if ((deferred_compaction
|| contended_compaction
) &&
2514 (gfp_mask
& __GFP_NO_KSWAPD
))
2517 /* Try direct reclaim and then allocating */
2518 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2519 zonelist
, high_zoneidx
,
2521 alloc_flags
, preferred_zone
,
2522 migratetype
, &did_some_progress
);
2527 * If we failed to make any progress reclaiming, then we are
2528 * running out of options and have to consider going OOM
2530 if (!did_some_progress
) {
2531 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2532 if (oom_killer_disabled
)
2534 /* Coredumps can quickly deplete all memory reserves */
2535 if ((current
->flags
& PF_DUMPCORE
) &&
2536 !(gfp_mask
& __GFP_NOFAIL
))
2538 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2539 zonelist
, high_zoneidx
,
2540 nodemask
, preferred_zone
,
2545 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2547 * The oom killer is not called for high-order
2548 * allocations that may fail, so if no progress
2549 * is being made, there are no other options and
2550 * retrying is unlikely to help.
2552 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2555 * The oom killer is not called for lowmem
2556 * allocations to prevent needlessly killing
2559 if (high_zoneidx
< ZONE_NORMAL
)
2567 /* Check if we should retry the allocation */
2568 pages_reclaimed
+= did_some_progress
;
2569 if (should_alloc_retry(gfp_mask
, order
, did_some_progress
,
2571 /* Wait for some write requests to complete then retry */
2572 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2576 * High-order allocations do not necessarily loop after
2577 * direct reclaim and reclaim/compaction depends on compaction
2578 * being called after reclaim so call directly if necessary
2580 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2581 zonelist
, high_zoneidx
,
2583 alloc_flags
, preferred_zone
,
2584 migratetype
, sync_migration
,
2585 &contended_compaction
,
2586 &deferred_compaction
,
2587 &did_some_progress
);
2593 warn_alloc_failed(gfp_mask
, order
, NULL
);
2596 if (kmemcheck_enabled
)
2597 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2603 * This is the 'heart' of the zoned buddy allocator.
2606 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2607 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2609 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2610 struct zone
*preferred_zone
;
2611 struct page
*page
= NULL
;
2612 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2613 unsigned int cpuset_mems_cookie
;
2614 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
;
2615 struct mem_cgroup
*memcg
= NULL
;
2617 gfp_mask
&= gfp_allowed_mask
;
2619 lockdep_trace_alloc(gfp_mask
);
2621 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2623 if (should_fail_alloc_page(gfp_mask
, order
))
2627 * Check the zones suitable for the gfp_mask contain at least one
2628 * valid zone. It's possible to have an empty zonelist as a result
2629 * of GFP_THISNODE and a memoryless node
2631 if (unlikely(!zonelist
->_zonerefs
->zone
))
2635 * Will only have any effect when __GFP_KMEMCG is set. This is
2636 * verified in the (always inline) callee
2638 if (!memcg_kmem_newpage_charge(gfp_mask
, &memcg
, order
))
2642 cpuset_mems_cookie
= get_mems_allowed();
2644 /* The preferred zone is used for statistics later */
2645 first_zones_zonelist(zonelist
, high_zoneidx
,
2646 nodemask
? : &cpuset_current_mems_allowed
,
2648 if (!preferred_zone
)
2652 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2653 alloc_flags
|= ALLOC_CMA
;
2655 /* First allocation attempt */
2656 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2657 zonelist
, high_zoneidx
, alloc_flags
,
2658 preferred_zone
, migratetype
);
2659 if (unlikely(!page
))
2660 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2661 zonelist
, high_zoneidx
, nodemask
,
2662 preferred_zone
, migratetype
);
2664 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2668 * When updating a task's mems_allowed, it is possible to race with
2669 * parallel threads in such a way that an allocation can fail while
2670 * the mask is being updated. If a page allocation is about to fail,
2671 * check if the cpuset changed during allocation and if so, retry.
2673 if (unlikely(!put_mems_allowed(cpuset_mems_cookie
) && !page
))
2676 memcg_kmem_commit_charge(page
, memcg
, order
);
2680 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2683 * Common helper functions.
2685 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2690 * __get_free_pages() returns a 32-bit address, which cannot represent
2693 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2695 page
= alloc_pages(gfp_mask
, order
);
2698 return (unsigned long) page_address(page
);
2700 EXPORT_SYMBOL(__get_free_pages
);
2702 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2704 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2706 EXPORT_SYMBOL(get_zeroed_page
);
2708 void __free_pages(struct page
*page
, unsigned int order
)
2710 if (put_page_testzero(page
)) {
2712 free_hot_cold_page(page
, 0);
2714 __free_pages_ok(page
, order
);
2718 EXPORT_SYMBOL(__free_pages
);
2720 void free_pages(unsigned long addr
, unsigned int order
)
2723 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2724 __free_pages(virt_to_page((void *)addr
), order
);
2728 EXPORT_SYMBOL(free_pages
);
2731 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2732 * pages allocated with __GFP_KMEMCG.
2734 * Those pages are accounted to a particular memcg, embedded in the
2735 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2736 * for that information only to find out that it is NULL for users who have no
2737 * interest in that whatsoever, we provide these functions.
2739 * The caller knows better which flags it relies on.
2741 void __free_memcg_kmem_pages(struct page
*page
, unsigned int order
)
2743 memcg_kmem_uncharge_pages(page
, order
);
2744 __free_pages(page
, order
);
2747 void free_memcg_kmem_pages(unsigned long addr
, unsigned int order
)
2750 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2751 __free_memcg_kmem_pages(virt_to_page((void *)addr
), order
);
2755 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2758 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2759 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2761 split_page(virt_to_page((void *)addr
), order
);
2762 while (used
< alloc_end
) {
2767 return (void *)addr
;
2771 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2772 * @size: the number of bytes to allocate
2773 * @gfp_mask: GFP flags for the allocation
2775 * This function is similar to alloc_pages(), except that it allocates the
2776 * minimum number of pages to satisfy the request. alloc_pages() can only
2777 * allocate memory in power-of-two pages.
2779 * This function is also limited by MAX_ORDER.
2781 * Memory allocated by this function must be released by free_pages_exact().
2783 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2785 unsigned int order
= get_order(size
);
2788 addr
= __get_free_pages(gfp_mask
, order
);
2789 return make_alloc_exact(addr
, order
, size
);
2791 EXPORT_SYMBOL(alloc_pages_exact
);
2794 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2796 * @nid: the preferred node ID where memory should be allocated
2797 * @size: the number of bytes to allocate
2798 * @gfp_mask: GFP flags for the allocation
2800 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2802 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2805 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2807 unsigned order
= get_order(size
);
2808 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2811 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2813 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2816 * free_pages_exact - release memory allocated via alloc_pages_exact()
2817 * @virt: the value returned by alloc_pages_exact.
2818 * @size: size of allocation, same value as passed to alloc_pages_exact().
2820 * Release the memory allocated by a previous call to alloc_pages_exact.
2822 void free_pages_exact(void *virt
, size_t size
)
2824 unsigned long addr
= (unsigned long)virt
;
2825 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2827 while (addr
< end
) {
2832 EXPORT_SYMBOL(free_pages_exact
);
2834 static unsigned int nr_free_zone_pages(int offset
)
2839 /* Just pick one node, since fallback list is circular */
2840 unsigned int sum
= 0;
2842 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2844 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2845 unsigned long size
= zone
->present_pages
;
2846 unsigned long high
= high_wmark_pages(zone
);
2855 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2857 unsigned int nr_free_buffer_pages(void)
2859 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2861 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2864 * Amount of free RAM allocatable within all zones
2866 unsigned int nr_free_pagecache_pages(void)
2868 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2871 static inline void show_node(struct zone
*zone
)
2873 if (IS_ENABLED(CONFIG_NUMA
))
2874 printk("Node %d ", zone_to_nid(zone
));
2877 void si_meminfo(struct sysinfo
*val
)
2879 val
->totalram
= totalram_pages
;
2881 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2882 val
->bufferram
= nr_blockdev_pages();
2883 val
->totalhigh
= totalhigh_pages
;
2884 val
->freehigh
= nr_free_highpages();
2885 val
->mem_unit
= PAGE_SIZE
;
2888 EXPORT_SYMBOL(si_meminfo
);
2891 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2893 pg_data_t
*pgdat
= NODE_DATA(nid
);
2895 val
->totalram
= pgdat
->node_present_pages
;
2896 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2897 #ifdef CONFIG_HIGHMEM
2898 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2899 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2905 val
->mem_unit
= PAGE_SIZE
;
2910 * Determine whether the node should be displayed or not, depending on whether
2911 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2913 bool skip_free_areas_node(unsigned int flags
, int nid
)
2916 unsigned int cpuset_mems_cookie
;
2918 if (!(flags
& SHOW_MEM_FILTER_NODES
))
2922 cpuset_mems_cookie
= get_mems_allowed();
2923 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
2924 } while (!put_mems_allowed(cpuset_mems_cookie
));
2929 #define K(x) ((x) << (PAGE_SHIFT-10))
2931 static void show_migration_types(unsigned char type
)
2933 static const char types
[MIGRATE_TYPES
] = {
2934 [MIGRATE_UNMOVABLE
] = 'U',
2935 [MIGRATE_RECLAIMABLE
] = 'E',
2936 [MIGRATE_MOVABLE
] = 'M',
2937 [MIGRATE_RESERVE
] = 'R',
2939 [MIGRATE_CMA
] = 'C',
2941 [MIGRATE_ISOLATE
] = 'I',
2943 char tmp
[MIGRATE_TYPES
+ 1];
2947 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
2948 if (type
& (1 << i
))
2953 printk("(%s) ", tmp
);
2957 * Show free area list (used inside shift_scroll-lock stuff)
2958 * We also calculate the percentage fragmentation. We do this by counting the
2959 * memory on each free list with the exception of the first item on the list.
2960 * Suppresses nodes that are not allowed by current's cpuset if
2961 * SHOW_MEM_FILTER_NODES is passed.
2963 void show_free_areas(unsigned int filter
)
2968 for_each_populated_zone(zone
) {
2969 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2972 printk("%s per-cpu:\n", zone
->name
);
2974 for_each_online_cpu(cpu
) {
2975 struct per_cpu_pageset
*pageset
;
2977 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2979 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2980 cpu
, pageset
->pcp
.high
,
2981 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2985 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2986 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2988 " dirty:%lu writeback:%lu unstable:%lu\n"
2989 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2990 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
2992 global_page_state(NR_ACTIVE_ANON
),
2993 global_page_state(NR_INACTIVE_ANON
),
2994 global_page_state(NR_ISOLATED_ANON
),
2995 global_page_state(NR_ACTIVE_FILE
),
2996 global_page_state(NR_INACTIVE_FILE
),
2997 global_page_state(NR_ISOLATED_FILE
),
2998 global_page_state(NR_UNEVICTABLE
),
2999 global_page_state(NR_FILE_DIRTY
),
3000 global_page_state(NR_WRITEBACK
),
3001 global_page_state(NR_UNSTABLE_NFS
),
3002 global_page_state(NR_FREE_PAGES
),
3003 global_page_state(NR_SLAB_RECLAIMABLE
),
3004 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3005 global_page_state(NR_FILE_MAPPED
),
3006 global_page_state(NR_SHMEM
),
3007 global_page_state(NR_PAGETABLE
),
3008 global_page_state(NR_BOUNCE
),
3009 global_page_state(NR_FREE_CMA_PAGES
));
3011 for_each_populated_zone(zone
) {
3014 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3022 " active_anon:%lukB"
3023 " inactive_anon:%lukB"
3024 " active_file:%lukB"
3025 " inactive_file:%lukB"
3026 " unevictable:%lukB"
3027 " isolated(anon):%lukB"
3028 " isolated(file):%lukB"
3036 " slab_reclaimable:%lukB"
3037 " slab_unreclaimable:%lukB"
3038 " kernel_stack:%lukB"
3043 " writeback_tmp:%lukB"
3044 " pages_scanned:%lu"
3045 " all_unreclaimable? %s"
3048 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3049 K(min_wmark_pages(zone
)),
3050 K(low_wmark_pages(zone
)),
3051 K(high_wmark_pages(zone
)),
3052 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3053 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3054 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3055 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3056 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3057 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3058 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3059 K(zone
->present_pages
),
3060 K(zone
->managed_pages
),
3061 K(zone_page_state(zone
, NR_MLOCK
)),
3062 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3063 K(zone_page_state(zone
, NR_WRITEBACK
)),
3064 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3065 K(zone_page_state(zone
, NR_SHMEM
)),
3066 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3067 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3068 zone_page_state(zone
, NR_KERNEL_STACK
) *
3070 K(zone_page_state(zone
, NR_PAGETABLE
)),
3071 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3072 K(zone_page_state(zone
, NR_BOUNCE
)),
3073 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3074 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3075 zone
->pages_scanned
,
3076 (zone
->all_unreclaimable
? "yes" : "no")
3078 printk("lowmem_reserve[]:");
3079 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3080 printk(" %lu", zone
->lowmem_reserve
[i
]);
3084 for_each_populated_zone(zone
) {
3085 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
3086 unsigned char types
[MAX_ORDER
];
3088 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3091 printk("%s: ", zone
->name
);
3093 spin_lock_irqsave(&zone
->lock
, flags
);
3094 for (order
= 0; order
< MAX_ORDER
; order
++) {
3095 struct free_area
*area
= &zone
->free_area
[order
];
3098 nr
[order
] = area
->nr_free
;
3099 total
+= nr
[order
] << order
;
3102 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3103 if (!list_empty(&area
->free_list
[type
]))
3104 types
[order
] |= 1 << type
;
3107 spin_unlock_irqrestore(&zone
->lock
, flags
);
3108 for (order
= 0; order
< MAX_ORDER
; order
++) {
3109 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3111 show_migration_types(types
[order
]);
3113 printk("= %lukB\n", K(total
));
3116 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3118 show_swap_cache_info();
3121 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3123 zoneref
->zone
= zone
;
3124 zoneref
->zone_idx
= zone_idx(zone
);
3128 * Builds allocation fallback zone lists.
3130 * Add all populated zones of a node to the zonelist.
3132 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3133 int nr_zones
, enum zone_type zone_type
)
3137 BUG_ON(zone_type
>= MAX_NR_ZONES
);
3142 zone
= pgdat
->node_zones
+ zone_type
;
3143 if (populated_zone(zone
)) {
3144 zoneref_set_zone(zone
,
3145 &zonelist
->_zonerefs
[nr_zones
++]);
3146 check_highest_zone(zone_type
);
3149 } while (zone_type
);
3156 * 0 = automatic detection of better ordering.
3157 * 1 = order by ([node] distance, -zonetype)
3158 * 2 = order by (-zonetype, [node] distance)
3160 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3161 * the same zonelist. So only NUMA can configure this param.
3163 #define ZONELIST_ORDER_DEFAULT 0
3164 #define ZONELIST_ORDER_NODE 1
3165 #define ZONELIST_ORDER_ZONE 2
3167 /* zonelist order in the kernel.
3168 * set_zonelist_order() will set this to NODE or ZONE.
3170 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3171 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3175 /* The value user specified ....changed by config */
3176 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3177 /* string for sysctl */
3178 #define NUMA_ZONELIST_ORDER_LEN 16
3179 char numa_zonelist_order
[16] = "default";
3182 * interface for configure zonelist ordering.
3183 * command line option "numa_zonelist_order"
3184 * = "[dD]efault - default, automatic configuration.
3185 * = "[nN]ode - order by node locality, then by zone within node
3186 * = "[zZ]one - order by zone, then by locality within zone
3189 static int __parse_numa_zonelist_order(char *s
)
3191 if (*s
== 'd' || *s
== 'D') {
3192 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3193 } else if (*s
== 'n' || *s
== 'N') {
3194 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3195 } else if (*s
== 'z' || *s
== 'Z') {
3196 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3199 "Ignoring invalid numa_zonelist_order value: "
3206 static __init
int setup_numa_zonelist_order(char *s
)
3213 ret
= __parse_numa_zonelist_order(s
);
3215 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3219 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3222 * sysctl handler for numa_zonelist_order
3224 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
3225 void __user
*buffer
, size_t *length
,
3228 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3230 static DEFINE_MUTEX(zl_order_mutex
);
3232 mutex_lock(&zl_order_mutex
);
3234 strcpy(saved_string
, (char*)table
->data
);
3235 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3239 int oldval
= user_zonelist_order
;
3240 if (__parse_numa_zonelist_order((char*)table
->data
)) {
3242 * bogus value. restore saved string
3244 strncpy((char*)table
->data
, saved_string
,
3245 NUMA_ZONELIST_ORDER_LEN
);
3246 user_zonelist_order
= oldval
;
3247 } else if (oldval
!= user_zonelist_order
) {
3248 mutex_lock(&zonelists_mutex
);
3249 build_all_zonelists(NULL
, NULL
);
3250 mutex_unlock(&zonelists_mutex
);
3254 mutex_unlock(&zl_order_mutex
);
3259 #define MAX_NODE_LOAD (nr_online_nodes)
3260 static int node_load
[MAX_NUMNODES
];
3263 * find_next_best_node - find the next node that should appear in a given node's fallback list
3264 * @node: node whose fallback list we're appending
3265 * @used_node_mask: nodemask_t of already used nodes
3267 * We use a number of factors to determine which is the next node that should
3268 * appear on a given node's fallback list. The node should not have appeared
3269 * already in @node's fallback list, and it should be the next closest node
3270 * according to the distance array (which contains arbitrary distance values
3271 * from each node to each node in the system), and should also prefer nodes
3272 * with no CPUs, since presumably they'll have very little allocation pressure
3273 * on them otherwise.
3274 * It returns -1 if no node is found.
3276 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3279 int min_val
= INT_MAX
;
3281 const struct cpumask
*tmp
= cpumask_of_node(0);
3283 /* Use the local node if we haven't already */
3284 if (!node_isset(node
, *used_node_mask
)) {
3285 node_set(node
, *used_node_mask
);
3289 for_each_node_state(n
, N_MEMORY
) {
3291 /* Don't want a node to appear more than once */
3292 if (node_isset(n
, *used_node_mask
))
3295 /* Use the distance array to find the distance */
3296 val
= node_distance(node
, n
);
3298 /* Penalize nodes under us ("prefer the next node") */
3301 /* Give preference to headless and unused nodes */
3302 tmp
= cpumask_of_node(n
);
3303 if (!cpumask_empty(tmp
))
3304 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
3306 /* Slight preference for less loaded node */
3307 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
3308 val
+= node_load
[n
];
3310 if (val
< min_val
) {
3317 node_set(best_node
, *used_node_mask
);
3324 * Build zonelists ordered by node and zones within node.
3325 * This results in maximum locality--normal zone overflows into local
3326 * DMA zone, if any--but risks exhausting DMA zone.
3328 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
3331 struct zonelist
*zonelist
;
3333 zonelist
= &pgdat
->node_zonelists
[0];
3334 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
3336 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3338 zonelist
->_zonerefs
[j
].zone
= NULL
;
3339 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3343 * Build gfp_thisnode zonelists
3345 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
3348 struct zonelist
*zonelist
;
3350 zonelist
= &pgdat
->node_zonelists
[1];
3351 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3352 zonelist
->_zonerefs
[j
].zone
= NULL
;
3353 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3357 * Build zonelists ordered by zone and nodes within zones.
3358 * This results in conserving DMA zone[s] until all Normal memory is
3359 * exhausted, but results in overflowing to remote node while memory
3360 * may still exist in local DMA zone.
3362 static int node_order
[MAX_NUMNODES
];
3364 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
3367 int zone_type
; /* needs to be signed */
3369 struct zonelist
*zonelist
;
3371 zonelist
= &pgdat
->node_zonelists
[0];
3373 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
3374 for (j
= 0; j
< nr_nodes
; j
++) {
3375 node
= node_order
[j
];
3376 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
3377 if (populated_zone(z
)) {
3379 &zonelist
->_zonerefs
[pos
++]);
3380 check_highest_zone(zone_type
);
3384 zonelist
->_zonerefs
[pos
].zone
= NULL
;
3385 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
3388 static int default_zonelist_order(void)
3391 unsigned long low_kmem_size
,total_size
;
3395 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3396 * If they are really small and used heavily, the system can fall
3397 * into OOM very easily.
3398 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3400 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3403 for_each_online_node(nid
) {
3404 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3405 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3406 if (populated_zone(z
)) {
3407 if (zone_type
< ZONE_NORMAL
)
3408 low_kmem_size
+= z
->present_pages
;
3409 total_size
+= z
->present_pages
;
3410 } else if (zone_type
== ZONE_NORMAL
) {
3412 * If any node has only lowmem, then node order
3413 * is preferred to allow kernel allocations
3414 * locally; otherwise, they can easily infringe
3415 * on other nodes when there is an abundance of
3416 * lowmem available to allocate from.
3418 return ZONELIST_ORDER_NODE
;
3422 if (!low_kmem_size
|| /* there are no DMA area. */
3423 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
3424 return ZONELIST_ORDER_NODE
;
3426 * look into each node's config.
3427 * If there is a node whose DMA/DMA32 memory is very big area on
3428 * local memory, NODE_ORDER may be suitable.
3430 average_size
= total_size
/
3431 (nodes_weight(node_states
[N_MEMORY
]) + 1);
3432 for_each_online_node(nid
) {
3435 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3436 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3437 if (populated_zone(z
)) {
3438 if (zone_type
< ZONE_NORMAL
)
3439 low_kmem_size
+= z
->present_pages
;
3440 total_size
+= z
->present_pages
;
3443 if (low_kmem_size
&&
3444 total_size
> average_size
&& /* ignore small node */
3445 low_kmem_size
> total_size
* 70/100)
3446 return ZONELIST_ORDER_NODE
;
3448 return ZONELIST_ORDER_ZONE
;
3451 static void set_zonelist_order(void)
3453 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
3454 current_zonelist_order
= default_zonelist_order();
3456 current_zonelist_order
= user_zonelist_order
;
3459 static void build_zonelists(pg_data_t
*pgdat
)
3463 nodemask_t used_mask
;
3464 int local_node
, prev_node
;
3465 struct zonelist
*zonelist
;
3466 int order
= current_zonelist_order
;
3468 /* initialize zonelists */
3469 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3470 zonelist
= pgdat
->node_zonelists
+ i
;
3471 zonelist
->_zonerefs
[0].zone
= NULL
;
3472 zonelist
->_zonerefs
[0].zone_idx
= 0;
3475 /* NUMA-aware ordering of nodes */
3476 local_node
= pgdat
->node_id
;
3477 load
= nr_online_nodes
;
3478 prev_node
= local_node
;
3479 nodes_clear(used_mask
);
3481 memset(node_order
, 0, sizeof(node_order
));
3484 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3486 * We don't want to pressure a particular node.
3487 * So adding penalty to the first node in same
3488 * distance group to make it round-robin.
3490 if (node_distance(local_node
, node
) !=
3491 node_distance(local_node
, prev_node
))
3492 node_load
[node
] = load
;
3496 if (order
== ZONELIST_ORDER_NODE
)
3497 build_zonelists_in_node_order(pgdat
, node
);
3499 node_order
[j
++] = node
; /* remember order */
3502 if (order
== ZONELIST_ORDER_ZONE
) {
3503 /* calculate node order -- i.e., DMA last! */
3504 build_zonelists_in_zone_order(pgdat
, j
);
3507 build_thisnode_zonelists(pgdat
);
3510 /* Construct the zonelist performance cache - see further mmzone.h */
3511 static void build_zonelist_cache(pg_data_t
*pgdat
)
3513 struct zonelist
*zonelist
;
3514 struct zonelist_cache
*zlc
;
3517 zonelist
= &pgdat
->node_zonelists
[0];
3518 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3519 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3520 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3521 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3524 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3526 * Return node id of node used for "local" allocations.
3527 * I.e., first node id of first zone in arg node's generic zonelist.
3528 * Used for initializing percpu 'numa_mem', which is used primarily
3529 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3531 int local_memory_node(int node
)
3535 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3536 gfp_zone(GFP_KERNEL
),
3543 #else /* CONFIG_NUMA */
3545 static void set_zonelist_order(void)
3547 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3550 static void build_zonelists(pg_data_t
*pgdat
)
3552 int node
, local_node
;
3554 struct zonelist
*zonelist
;
3556 local_node
= pgdat
->node_id
;
3558 zonelist
= &pgdat
->node_zonelists
[0];
3559 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3562 * Now we build the zonelist so that it contains the zones
3563 * of all the other nodes.
3564 * We don't want to pressure a particular node, so when
3565 * building the zones for node N, we make sure that the
3566 * zones coming right after the local ones are those from
3567 * node N+1 (modulo N)
3569 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3570 if (!node_online(node
))
3572 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3575 for (node
= 0; node
< local_node
; node
++) {
3576 if (!node_online(node
))
3578 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3582 zonelist
->_zonerefs
[j
].zone
= NULL
;
3583 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3586 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3587 static void build_zonelist_cache(pg_data_t
*pgdat
)
3589 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3592 #endif /* CONFIG_NUMA */
3595 * Boot pageset table. One per cpu which is going to be used for all
3596 * zones and all nodes. The parameters will be set in such a way
3597 * that an item put on a list will immediately be handed over to
3598 * the buddy list. This is safe since pageset manipulation is done
3599 * with interrupts disabled.
3601 * The boot_pagesets must be kept even after bootup is complete for
3602 * unused processors and/or zones. They do play a role for bootstrapping
3603 * hotplugged processors.
3605 * zoneinfo_show() and maybe other functions do
3606 * not check if the processor is online before following the pageset pointer.
3607 * Other parts of the kernel may not check if the zone is available.
3609 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3610 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3611 static void setup_zone_pageset(struct zone
*zone
);
3614 * Global mutex to protect against size modification of zonelists
3615 * as well as to serialize pageset setup for the new populated zone.
3617 DEFINE_MUTEX(zonelists_mutex
);
3619 /* return values int ....just for stop_machine() */
3620 static int __build_all_zonelists(void *data
)
3624 pg_data_t
*self
= data
;
3627 memset(node_load
, 0, sizeof(node_load
));
3630 if (self
&& !node_online(self
->node_id
)) {
3631 build_zonelists(self
);
3632 build_zonelist_cache(self
);
3635 for_each_online_node(nid
) {
3636 pg_data_t
*pgdat
= NODE_DATA(nid
);
3638 build_zonelists(pgdat
);
3639 build_zonelist_cache(pgdat
);
3643 * Initialize the boot_pagesets that are going to be used
3644 * for bootstrapping processors. The real pagesets for
3645 * each zone will be allocated later when the per cpu
3646 * allocator is available.
3648 * boot_pagesets are used also for bootstrapping offline
3649 * cpus if the system is already booted because the pagesets
3650 * are needed to initialize allocators on a specific cpu too.
3651 * F.e. the percpu allocator needs the page allocator which
3652 * needs the percpu allocator in order to allocate its pagesets
3653 * (a chicken-egg dilemma).
3655 for_each_possible_cpu(cpu
) {
3656 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3658 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3660 * We now know the "local memory node" for each node--
3661 * i.e., the node of the first zone in the generic zonelist.
3662 * Set up numa_mem percpu variable for on-line cpus. During
3663 * boot, only the boot cpu should be on-line; we'll init the
3664 * secondary cpus' numa_mem as they come on-line. During
3665 * node/memory hotplug, we'll fixup all on-line cpus.
3667 if (cpu_online(cpu
))
3668 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3676 * Called with zonelists_mutex held always
3677 * unless system_state == SYSTEM_BOOTING.
3679 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
3681 set_zonelist_order();
3683 if (system_state
== SYSTEM_BOOTING
) {
3684 __build_all_zonelists(NULL
);
3685 mminit_verify_zonelist();
3686 cpuset_init_current_mems_allowed();
3688 /* we have to stop all cpus to guarantee there is no user
3690 #ifdef CONFIG_MEMORY_HOTPLUG
3692 setup_zone_pageset(zone
);
3694 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
3695 /* cpuset refresh routine should be here */
3697 vm_total_pages
= nr_free_pagecache_pages();
3699 * Disable grouping by mobility if the number of pages in the
3700 * system is too low to allow the mechanism to work. It would be
3701 * more accurate, but expensive to check per-zone. This check is
3702 * made on memory-hotadd so a system can start with mobility
3703 * disabled and enable it later
3705 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3706 page_group_by_mobility_disabled
= 1;
3708 page_group_by_mobility_disabled
= 0;
3710 printk("Built %i zonelists in %s order, mobility grouping %s. "
3711 "Total pages: %ld\n",
3713 zonelist_order_name
[current_zonelist_order
],
3714 page_group_by_mobility_disabled
? "off" : "on",
3717 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3722 * Helper functions to size the waitqueue hash table.
3723 * Essentially these want to choose hash table sizes sufficiently
3724 * large so that collisions trying to wait on pages are rare.
3725 * But in fact, the number of active page waitqueues on typical
3726 * systems is ridiculously low, less than 200. So this is even
3727 * conservative, even though it seems large.
3729 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3730 * waitqueues, i.e. the size of the waitq table given the number of pages.
3732 #define PAGES_PER_WAITQUEUE 256
3734 #ifndef CONFIG_MEMORY_HOTPLUG
3735 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3737 unsigned long size
= 1;
3739 pages
/= PAGES_PER_WAITQUEUE
;
3741 while (size
< pages
)
3745 * Once we have dozens or even hundreds of threads sleeping
3746 * on IO we've got bigger problems than wait queue collision.
3747 * Limit the size of the wait table to a reasonable size.
3749 size
= min(size
, 4096UL);
3751 return max(size
, 4UL);
3755 * A zone's size might be changed by hot-add, so it is not possible to determine
3756 * a suitable size for its wait_table. So we use the maximum size now.
3758 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3760 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3761 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3762 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3764 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3765 * or more by the traditional way. (See above). It equals:
3767 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3768 * ia64(16K page size) : = ( 8G + 4M)byte.
3769 * powerpc (64K page size) : = (32G +16M)byte.
3771 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3778 * This is an integer logarithm so that shifts can be used later
3779 * to extract the more random high bits from the multiplicative
3780 * hash function before the remainder is taken.
3782 static inline unsigned long wait_table_bits(unsigned long size
)
3787 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3790 * Check if a pageblock contains reserved pages
3792 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3796 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3797 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3804 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3805 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3806 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3807 * higher will lead to a bigger reserve which will get freed as contiguous
3808 * blocks as reclaim kicks in
3810 static void setup_zone_migrate_reserve(struct zone
*zone
)
3812 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3814 unsigned long block_migratetype
;
3818 * Get the start pfn, end pfn and the number of blocks to reserve
3819 * We have to be careful to be aligned to pageblock_nr_pages to
3820 * make sure that we always check pfn_valid for the first page in
3823 start_pfn
= zone
->zone_start_pfn
;
3824 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3825 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3826 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3830 * Reserve blocks are generally in place to help high-order atomic
3831 * allocations that are short-lived. A min_free_kbytes value that
3832 * would result in more than 2 reserve blocks for atomic allocations
3833 * is assumed to be in place to help anti-fragmentation for the
3834 * future allocation of hugepages at runtime.
3836 reserve
= min(2, reserve
);
3838 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3839 if (!pfn_valid(pfn
))
3841 page
= pfn_to_page(pfn
);
3843 /* Watch out for overlapping nodes */
3844 if (page_to_nid(page
) != zone_to_nid(zone
))
3847 block_migratetype
= get_pageblock_migratetype(page
);
3849 /* Only test what is necessary when the reserves are not met */
3852 * Blocks with reserved pages will never free, skip
3855 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3856 if (pageblock_is_reserved(pfn
, block_end_pfn
))
3859 /* If this block is reserved, account for it */
3860 if (block_migratetype
== MIGRATE_RESERVE
) {
3865 /* Suitable for reserving if this block is movable */
3866 if (block_migratetype
== MIGRATE_MOVABLE
) {
3867 set_pageblock_migratetype(page
,
3869 move_freepages_block(zone
, page
,
3877 * If the reserve is met and this is a previous reserved block,
3880 if (block_migratetype
== MIGRATE_RESERVE
) {
3881 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3882 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3888 * Initially all pages are reserved - free ones are freed
3889 * up by free_all_bootmem() once the early boot process is
3890 * done. Non-atomic initialization, single-pass.
3892 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3893 unsigned long start_pfn
, enum memmap_context context
)
3896 unsigned long end_pfn
= start_pfn
+ size
;
3900 if (highest_memmap_pfn
< end_pfn
- 1)
3901 highest_memmap_pfn
= end_pfn
- 1;
3903 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3904 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3906 * There can be holes in boot-time mem_map[]s
3907 * handed to this function. They do not
3908 * exist on hotplugged memory.
3910 if (context
== MEMMAP_EARLY
) {
3911 if (!early_pfn_valid(pfn
))
3913 if (!early_pfn_in_nid(pfn
, nid
))
3916 page
= pfn_to_page(pfn
);
3917 set_page_links(page
, zone
, nid
, pfn
);
3918 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3919 init_page_count(page
);
3920 reset_page_mapcount(page
);
3921 reset_page_last_nid(page
);
3922 SetPageReserved(page
);
3924 * Mark the block movable so that blocks are reserved for
3925 * movable at startup. This will force kernel allocations
3926 * to reserve their blocks rather than leaking throughout
3927 * the address space during boot when many long-lived
3928 * kernel allocations are made. Later some blocks near
3929 * the start are marked MIGRATE_RESERVE by
3930 * setup_zone_migrate_reserve()
3932 * bitmap is created for zone's valid pfn range. but memmap
3933 * can be created for invalid pages (for alignment)
3934 * check here not to call set_pageblock_migratetype() against
3937 if ((z
->zone_start_pfn
<= pfn
)
3938 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3939 && !(pfn
& (pageblock_nr_pages
- 1)))
3940 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3942 INIT_LIST_HEAD(&page
->lru
);
3943 #ifdef WANT_PAGE_VIRTUAL
3944 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3945 if (!is_highmem_idx(zone
))
3946 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3951 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3954 for_each_migratetype_order(order
, t
) {
3955 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3956 zone
->free_area
[order
].nr_free
= 0;
3960 #ifndef __HAVE_ARCH_MEMMAP_INIT
3961 #define memmap_init(size, nid, zone, start_pfn) \
3962 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3965 static int __meminit
zone_batchsize(struct zone
*zone
)
3971 * The per-cpu-pages pools are set to around 1000th of the
3972 * size of the zone. But no more than 1/2 of a meg.
3974 * OK, so we don't know how big the cache is. So guess.
3976 batch
= zone
->present_pages
/ 1024;
3977 if (batch
* PAGE_SIZE
> 512 * 1024)
3978 batch
= (512 * 1024) / PAGE_SIZE
;
3979 batch
/= 4; /* We effectively *= 4 below */
3984 * Clamp the batch to a 2^n - 1 value. Having a power
3985 * of 2 value was found to be more likely to have
3986 * suboptimal cache aliasing properties in some cases.
3988 * For example if 2 tasks are alternately allocating
3989 * batches of pages, one task can end up with a lot
3990 * of pages of one half of the possible page colors
3991 * and the other with pages of the other colors.
3993 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3998 /* The deferral and batching of frees should be suppressed under NOMMU
4001 * The problem is that NOMMU needs to be able to allocate large chunks
4002 * of contiguous memory as there's no hardware page translation to
4003 * assemble apparent contiguous memory from discontiguous pages.
4005 * Queueing large contiguous runs of pages for batching, however,
4006 * causes the pages to actually be freed in smaller chunks. As there
4007 * can be a significant delay between the individual batches being
4008 * recycled, this leads to the once large chunks of space being
4009 * fragmented and becoming unavailable for high-order allocations.
4015 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4017 struct per_cpu_pages
*pcp
;
4020 memset(p
, 0, sizeof(*p
));
4024 pcp
->high
= 6 * batch
;
4025 pcp
->batch
= max(1UL, 1 * batch
);
4026 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4027 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4031 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
4032 * to the value high for the pageset p.
4035 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
4038 struct per_cpu_pages
*pcp
;
4042 pcp
->batch
= max(1UL, high
/4);
4043 if ((high
/4) > (PAGE_SHIFT
* 8))
4044 pcp
->batch
= PAGE_SHIFT
* 8;
4047 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4051 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4053 for_each_possible_cpu(cpu
) {
4054 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4056 setup_pageset(pcp
, zone_batchsize(zone
));
4058 if (percpu_pagelist_fraction
)
4059 setup_pagelist_highmark(pcp
,
4060 (zone
->present_pages
/
4061 percpu_pagelist_fraction
));
4066 * Allocate per cpu pagesets and initialize them.
4067 * Before this call only boot pagesets were available.
4069 void __init
setup_per_cpu_pageset(void)
4073 for_each_populated_zone(zone
)
4074 setup_zone_pageset(zone
);
4077 static noinline __init_refok
4078 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4081 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4085 * The per-page waitqueue mechanism uses hashed waitqueues
4088 zone
->wait_table_hash_nr_entries
=
4089 wait_table_hash_nr_entries(zone_size_pages
);
4090 zone
->wait_table_bits
=
4091 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4092 alloc_size
= zone
->wait_table_hash_nr_entries
4093 * sizeof(wait_queue_head_t
);
4095 if (!slab_is_available()) {
4096 zone
->wait_table
= (wait_queue_head_t
*)
4097 alloc_bootmem_node_nopanic(pgdat
, alloc_size
);
4100 * This case means that a zone whose size was 0 gets new memory
4101 * via memory hot-add.
4102 * But it may be the case that a new node was hot-added. In
4103 * this case vmalloc() will not be able to use this new node's
4104 * memory - this wait_table must be initialized to use this new
4105 * node itself as well.
4106 * To use this new node's memory, further consideration will be
4109 zone
->wait_table
= vmalloc(alloc_size
);
4111 if (!zone
->wait_table
)
4114 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4115 init_waitqueue_head(zone
->wait_table
+ i
);
4120 static __meminit
void zone_pcp_init(struct zone
*zone
)
4123 * per cpu subsystem is not up at this point. The following code
4124 * relies on the ability of the linker to provide the
4125 * offset of a (static) per cpu variable into the per cpu area.
4127 zone
->pageset
= &boot_pageset
;
4129 if (zone
->present_pages
)
4130 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4131 zone
->name
, zone
->present_pages
,
4132 zone_batchsize(zone
));
4135 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4136 unsigned long zone_start_pfn
,
4138 enum memmap_context context
)
4140 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4142 ret
= zone_wait_table_init(zone
, size
);
4145 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4147 zone
->zone_start_pfn
= zone_start_pfn
;
4149 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4150 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4152 (unsigned long)zone_idx(zone
),
4153 zone_start_pfn
, (zone_start_pfn
+ size
));
4155 zone_init_free_lists(zone
);
4160 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4161 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4163 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4164 * Architectures may implement their own version but if add_active_range()
4165 * was used and there are no special requirements, this is a convenient
4168 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
4170 unsigned long start_pfn
, end_pfn
;
4173 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4174 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
4176 /* This is a memory hole */
4179 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4181 int __meminit
early_pfn_to_nid(unsigned long pfn
)
4185 nid
= __early_pfn_to_nid(pfn
);
4188 /* just returns 0 */
4192 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4193 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
4197 nid
= __early_pfn_to_nid(pfn
);
4198 if (nid
>= 0 && nid
!= node
)
4205 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
4206 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4207 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
4209 * If an architecture guarantees that all ranges registered with
4210 * add_active_ranges() contain no holes and may be freed, this
4211 * this function may be used instead of calling free_bootmem() manually.
4213 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4215 unsigned long start_pfn
, end_pfn
;
4218 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4219 start_pfn
= min(start_pfn
, max_low_pfn
);
4220 end_pfn
= min(end_pfn
, max_low_pfn
);
4222 if (start_pfn
< end_pfn
)
4223 free_bootmem_node(NODE_DATA(this_nid
),
4224 PFN_PHYS(start_pfn
),
4225 (end_pfn
- start_pfn
) << PAGE_SHIFT
);
4230 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4231 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4233 * If an architecture guarantees that all ranges registered with
4234 * add_active_ranges() contain no holes and may be freed, this
4235 * function may be used instead of calling memory_present() manually.
4237 void __init
sparse_memory_present_with_active_regions(int nid
)
4239 unsigned long start_pfn
, end_pfn
;
4242 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4243 memory_present(this_nid
, start_pfn
, end_pfn
);
4247 * get_pfn_range_for_nid - Return the start and end page frames for a node
4248 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4249 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4250 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4252 * It returns the start and end page frame of a node based on information
4253 * provided by an arch calling add_active_range(). If called for a node
4254 * with no available memory, a warning is printed and the start and end
4257 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4258 unsigned long *start_pfn
, unsigned long *end_pfn
)
4260 unsigned long this_start_pfn
, this_end_pfn
;
4266 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4267 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4268 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4271 if (*start_pfn
== -1UL)
4276 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4277 * assumption is made that zones within a node are ordered in monotonic
4278 * increasing memory addresses so that the "highest" populated zone is used
4280 static void __init
find_usable_zone_for_movable(void)
4283 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4284 if (zone_index
== ZONE_MOVABLE
)
4287 if (arch_zone_highest_possible_pfn
[zone_index
] >
4288 arch_zone_lowest_possible_pfn
[zone_index
])
4292 VM_BUG_ON(zone_index
== -1);
4293 movable_zone
= zone_index
;
4297 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4298 * because it is sized independent of architecture. Unlike the other zones,
4299 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4300 * in each node depending on the size of each node and how evenly kernelcore
4301 * is distributed. This helper function adjusts the zone ranges
4302 * provided by the architecture for a given node by using the end of the
4303 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4304 * zones within a node are in order of monotonic increases memory addresses
4306 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4307 unsigned long zone_type
,
4308 unsigned long node_start_pfn
,
4309 unsigned long node_end_pfn
,
4310 unsigned long *zone_start_pfn
,
4311 unsigned long *zone_end_pfn
)
4313 /* Only adjust if ZONE_MOVABLE is on this node */
4314 if (zone_movable_pfn
[nid
]) {
4315 /* Size ZONE_MOVABLE */
4316 if (zone_type
== ZONE_MOVABLE
) {
4317 *zone_start_pfn
= zone_movable_pfn
[nid
];
4318 *zone_end_pfn
= min(node_end_pfn
,
4319 arch_zone_highest_possible_pfn
[movable_zone
]);
4321 /* Adjust for ZONE_MOVABLE starting within this range */
4322 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4323 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4324 *zone_end_pfn
= zone_movable_pfn
[nid
];
4326 /* Check if this whole range is within ZONE_MOVABLE */
4327 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4328 *zone_start_pfn
= *zone_end_pfn
;
4333 * Return the number of pages a zone spans in a node, including holes
4334 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4336 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4337 unsigned long zone_type
,
4338 unsigned long *ignored
)
4340 unsigned long node_start_pfn
, node_end_pfn
;
4341 unsigned long zone_start_pfn
, zone_end_pfn
;
4343 /* Get the start and end of the node and zone */
4344 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4345 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4346 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4347 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4348 node_start_pfn
, node_end_pfn
,
4349 &zone_start_pfn
, &zone_end_pfn
);
4351 /* Check that this node has pages within the zone's required range */
4352 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4355 /* Move the zone boundaries inside the node if necessary */
4356 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4357 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4359 /* Return the spanned pages */
4360 return zone_end_pfn
- zone_start_pfn
;
4364 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4365 * then all holes in the requested range will be accounted for.
4367 unsigned long __meminit
__absent_pages_in_range(int nid
,
4368 unsigned long range_start_pfn
,
4369 unsigned long range_end_pfn
)
4371 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4372 unsigned long start_pfn
, end_pfn
;
4375 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4376 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4377 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4378 nr_absent
-= end_pfn
- start_pfn
;
4384 * absent_pages_in_range - Return number of page frames in holes within a range
4385 * @start_pfn: The start PFN to start searching for holes
4386 * @end_pfn: The end PFN to stop searching for holes
4388 * It returns the number of pages frames in memory holes within a range.
4390 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4391 unsigned long end_pfn
)
4393 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4396 /* Return the number of page frames in holes in a zone on a node */
4397 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4398 unsigned long zone_type
,
4399 unsigned long *ignored
)
4401 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4402 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4403 unsigned long node_start_pfn
, node_end_pfn
;
4404 unsigned long zone_start_pfn
, zone_end_pfn
;
4406 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
4407 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4408 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4410 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4411 node_start_pfn
, node_end_pfn
,
4412 &zone_start_pfn
, &zone_end_pfn
);
4413 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4416 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4417 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4418 unsigned long zone_type
,
4419 unsigned long *zones_size
)
4421 return zones_size
[zone_type
];
4424 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4425 unsigned long zone_type
,
4426 unsigned long *zholes_size
)
4431 return zholes_size
[zone_type
];
4434 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4436 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4437 unsigned long *zones_size
, unsigned long *zholes_size
)
4439 unsigned long realtotalpages
, totalpages
= 0;
4442 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4443 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4445 pgdat
->node_spanned_pages
= totalpages
;
4447 realtotalpages
= totalpages
;
4448 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4450 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4452 pgdat
->node_present_pages
= realtotalpages
;
4453 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4457 #ifndef CONFIG_SPARSEMEM
4459 * Calculate the size of the zone->blockflags rounded to an unsigned long
4460 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4461 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4462 * round what is now in bits to nearest long in bits, then return it in
4465 static unsigned long __init
usemap_size(unsigned long zonesize
)
4467 unsigned long usemapsize
;
4469 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4470 usemapsize
= usemapsize
>> pageblock_order
;
4471 usemapsize
*= NR_PAGEBLOCK_BITS
;
4472 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4474 return usemapsize
/ 8;
4477 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4478 struct zone
*zone
, unsigned long zonesize
)
4480 unsigned long usemapsize
= usemap_size(zonesize
);
4481 zone
->pageblock_flags
= NULL
;
4483 zone
->pageblock_flags
= alloc_bootmem_node_nopanic(pgdat
,
4487 static inline void setup_usemap(struct pglist_data
*pgdat
,
4488 struct zone
*zone
, unsigned long zonesize
) {}
4489 #endif /* CONFIG_SPARSEMEM */
4491 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4493 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4494 void __init
set_pageblock_order(void)
4498 /* Check that pageblock_nr_pages has not already been setup */
4499 if (pageblock_order
)
4502 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4503 order
= HUGETLB_PAGE_ORDER
;
4505 order
= MAX_ORDER
- 1;
4508 * Assume the largest contiguous order of interest is a huge page.
4509 * This value may be variable depending on boot parameters on IA64 and
4512 pageblock_order
= order
;
4514 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4517 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4518 * is unused as pageblock_order is set at compile-time. See
4519 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4522 void __init
set_pageblock_order(void)
4526 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4528 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
4529 unsigned long present_pages
)
4531 unsigned long pages
= spanned_pages
;
4534 * Provide a more accurate estimation if there are holes within
4535 * the zone and SPARSEMEM is in use. If there are holes within the
4536 * zone, each populated memory region may cost us one or two extra
4537 * memmap pages due to alignment because memmap pages for each
4538 * populated regions may not naturally algined on page boundary.
4539 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4541 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
4542 IS_ENABLED(CONFIG_SPARSEMEM
))
4543 pages
= present_pages
;
4545 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
4549 * Set up the zone data structures:
4550 * - mark all pages reserved
4551 * - mark all memory queues empty
4552 * - clear the memory bitmaps
4554 * NOTE: pgdat should get zeroed by caller.
4556 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4557 unsigned long *zones_size
, unsigned long *zholes_size
)
4560 int nid
= pgdat
->node_id
;
4561 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4564 pgdat_resize_init(pgdat
);
4565 #ifdef CONFIG_NUMA_BALANCING
4566 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
4567 pgdat
->numabalancing_migrate_nr_pages
= 0;
4568 pgdat
->numabalancing_migrate_next_window
= jiffies
;
4570 init_waitqueue_head(&pgdat
->kswapd_wait
);
4571 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
4572 pgdat_page_cgroup_init(pgdat
);
4574 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4575 struct zone
*zone
= pgdat
->node_zones
+ j
;
4576 unsigned long size
, realsize
, freesize
, memmap_pages
;
4578 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4579 realsize
= freesize
= size
- zone_absent_pages_in_node(nid
, j
,
4583 * Adjust freesize so that it accounts for how much memory
4584 * is used by this zone for memmap. This affects the watermark
4585 * and per-cpu initialisations
4587 memmap_pages
= calc_memmap_size(size
, realsize
);
4588 if (freesize
>= memmap_pages
) {
4589 freesize
-= memmap_pages
;
4592 " %s zone: %lu pages used for memmap\n",
4593 zone_names
[j
], memmap_pages
);
4596 " %s zone: %lu pages exceeds freesize %lu\n",
4597 zone_names
[j
], memmap_pages
, freesize
);
4599 /* Account for reserved pages */
4600 if (j
== 0 && freesize
> dma_reserve
) {
4601 freesize
-= dma_reserve
;
4602 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4603 zone_names
[0], dma_reserve
);
4606 if (!is_highmem_idx(j
))
4607 nr_kernel_pages
+= freesize
;
4608 /* Charge for highmem memmap if there are enough kernel pages */
4609 else if (nr_kernel_pages
> memmap_pages
* 2)
4610 nr_kernel_pages
-= memmap_pages
;
4611 nr_all_pages
+= freesize
;
4613 zone
->spanned_pages
= size
;
4614 zone
->present_pages
= freesize
;
4616 * Set an approximate value for lowmem here, it will be adjusted
4617 * when the bootmem allocator frees pages into the buddy system.
4618 * And all highmem pages will be managed by the buddy system.
4620 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
4623 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
4625 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
4627 zone
->name
= zone_names
[j
];
4628 spin_lock_init(&zone
->lock
);
4629 spin_lock_init(&zone
->lru_lock
);
4630 zone_seqlock_init(zone
);
4631 zone
->zone_pgdat
= pgdat
;
4633 zone_pcp_init(zone
);
4634 lruvec_init(&zone
->lruvec
);
4638 set_pageblock_order();
4639 setup_usemap(pgdat
, zone
, size
);
4640 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4641 size
, MEMMAP_EARLY
);
4643 memmap_init(size
, nid
, j
, zone_start_pfn
);
4644 zone_start_pfn
+= size
;
4648 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4650 /* Skip empty nodes */
4651 if (!pgdat
->node_spanned_pages
)
4654 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4655 /* ia64 gets its own node_mem_map, before this, without bootmem */
4656 if (!pgdat
->node_mem_map
) {
4657 unsigned long size
, start
, end
;
4661 * The zone's endpoints aren't required to be MAX_ORDER
4662 * aligned but the node_mem_map endpoints must be in order
4663 * for the buddy allocator to function correctly.
4665 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4666 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4667 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4668 size
= (end
- start
) * sizeof(struct page
);
4669 map
= alloc_remap(pgdat
->node_id
, size
);
4671 map
= alloc_bootmem_node_nopanic(pgdat
, size
);
4672 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4674 #ifndef CONFIG_NEED_MULTIPLE_NODES
4676 * With no DISCONTIG, the global mem_map is just set as node 0's
4678 if (pgdat
== NODE_DATA(0)) {
4679 mem_map
= NODE_DATA(0)->node_mem_map
;
4680 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4681 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4682 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4683 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4686 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4689 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4690 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4692 pg_data_t
*pgdat
= NODE_DATA(nid
);
4694 /* pg_data_t should be reset to zero when it's allocated */
4695 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
4697 pgdat
->node_id
= nid
;
4698 pgdat
->node_start_pfn
= node_start_pfn
;
4699 init_zone_allows_reclaim(nid
);
4700 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4702 alloc_node_mem_map(pgdat
);
4703 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4704 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4705 nid
, (unsigned long)pgdat
,
4706 (unsigned long)pgdat
->node_mem_map
);
4709 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4712 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4714 #if MAX_NUMNODES > 1
4716 * Figure out the number of possible node ids.
4718 static void __init
setup_nr_node_ids(void)
4721 unsigned int highest
= 0;
4723 for_each_node_mask(node
, node_possible_map
)
4725 nr_node_ids
= highest
+ 1;
4728 static inline void setup_nr_node_ids(void)
4734 * node_map_pfn_alignment - determine the maximum internode alignment
4736 * This function should be called after node map is populated and sorted.
4737 * It calculates the maximum power of two alignment which can distinguish
4740 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4741 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4742 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4743 * shifted, 1GiB is enough and this function will indicate so.
4745 * This is used to test whether pfn -> nid mapping of the chosen memory
4746 * model has fine enough granularity to avoid incorrect mapping for the
4747 * populated node map.
4749 * Returns the determined alignment in pfn's. 0 if there is no alignment
4750 * requirement (single node).
4752 unsigned long __init
node_map_pfn_alignment(void)
4754 unsigned long accl_mask
= 0, last_end
= 0;
4755 unsigned long start
, end
, mask
;
4759 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4760 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4767 * Start with a mask granular enough to pin-point to the
4768 * start pfn and tick off bits one-by-one until it becomes
4769 * too coarse to separate the current node from the last.
4771 mask
= ~((1 << __ffs(start
)) - 1);
4772 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4775 /* accumulate all internode masks */
4779 /* convert mask to number of pages */
4780 return ~accl_mask
+ 1;
4783 /* Find the lowest pfn for a node */
4784 static unsigned long __init
find_min_pfn_for_node(int nid
)
4786 unsigned long min_pfn
= ULONG_MAX
;
4787 unsigned long start_pfn
;
4790 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4791 min_pfn
= min(min_pfn
, start_pfn
);
4793 if (min_pfn
== ULONG_MAX
) {
4795 "Could not find start_pfn for node %d\n", nid
);
4803 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4805 * It returns the minimum PFN based on information provided via
4806 * add_active_range().
4808 unsigned long __init
find_min_pfn_with_active_regions(void)
4810 return find_min_pfn_for_node(MAX_NUMNODES
);
4814 * early_calculate_totalpages()
4815 * Sum pages in active regions for movable zone.
4816 * Populate N_MEMORY for calculating usable_nodes.
4818 static unsigned long __init
early_calculate_totalpages(void)
4820 unsigned long totalpages
= 0;
4821 unsigned long start_pfn
, end_pfn
;
4824 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
4825 unsigned long pages
= end_pfn
- start_pfn
;
4827 totalpages
+= pages
;
4829 node_set_state(nid
, N_MEMORY
);
4835 * Find the PFN the Movable zone begins in each node. Kernel memory
4836 * is spread evenly between nodes as long as the nodes have enough
4837 * memory. When they don't, some nodes will have more kernelcore than
4840 static void __init
find_zone_movable_pfns_for_nodes(void)
4843 unsigned long usable_startpfn
;
4844 unsigned long kernelcore_node
, kernelcore_remaining
;
4845 /* save the state before borrow the nodemask */
4846 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
4847 unsigned long totalpages
= early_calculate_totalpages();
4848 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
4851 * If movablecore was specified, calculate what size of
4852 * kernelcore that corresponds so that memory usable for
4853 * any allocation type is evenly spread. If both kernelcore
4854 * and movablecore are specified, then the value of kernelcore
4855 * will be used for required_kernelcore if it's greater than
4856 * what movablecore would have allowed.
4858 if (required_movablecore
) {
4859 unsigned long corepages
;
4862 * Round-up so that ZONE_MOVABLE is at least as large as what
4863 * was requested by the user
4865 required_movablecore
=
4866 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4867 corepages
= totalpages
- required_movablecore
;
4869 required_kernelcore
= max(required_kernelcore
, corepages
);
4872 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4873 if (!required_kernelcore
)
4876 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4877 find_usable_zone_for_movable();
4878 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4881 /* Spread kernelcore memory as evenly as possible throughout nodes */
4882 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4883 for_each_node_state(nid
, N_MEMORY
) {
4884 unsigned long start_pfn
, end_pfn
;
4887 * Recalculate kernelcore_node if the division per node
4888 * now exceeds what is necessary to satisfy the requested
4889 * amount of memory for the kernel
4891 if (required_kernelcore
< kernelcore_node
)
4892 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4895 * As the map is walked, we track how much memory is usable
4896 * by the kernel using kernelcore_remaining. When it is
4897 * 0, the rest of the node is usable by ZONE_MOVABLE
4899 kernelcore_remaining
= kernelcore_node
;
4901 /* Go through each range of PFNs within this node */
4902 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4903 unsigned long size_pages
;
4905 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
4906 if (start_pfn
>= end_pfn
)
4909 /* Account for what is only usable for kernelcore */
4910 if (start_pfn
< usable_startpfn
) {
4911 unsigned long kernel_pages
;
4912 kernel_pages
= min(end_pfn
, usable_startpfn
)
4915 kernelcore_remaining
-= min(kernel_pages
,
4916 kernelcore_remaining
);
4917 required_kernelcore
-= min(kernel_pages
,
4918 required_kernelcore
);
4920 /* Continue if range is now fully accounted */
4921 if (end_pfn
<= usable_startpfn
) {
4924 * Push zone_movable_pfn to the end so
4925 * that if we have to rebalance
4926 * kernelcore across nodes, we will
4927 * not double account here
4929 zone_movable_pfn
[nid
] = end_pfn
;
4932 start_pfn
= usable_startpfn
;
4936 * The usable PFN range for ZONE_MOVABLE is from
4937 * start_pfn->end_pfn. Calculate size_pages as the
4938 * number of pages used as kernelcore
4940 size_pages
= end_pfn
- start_pfn
;
4941 if (size_pages
> kernelcore_remaining
)
4942 size_pages
= kernelcore_remaining
;
4943 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4946 * Some kernelcore has been met, update counts and
4947 * break if the kernelcore for this node has been
4950 required_kernelcore
-= min(required_kernelcore
,
4952 kernelcore_remaining
-= size_pages
;
4953 if (!kernelcore_remaining
)
4959 * If there is still required_kernelcore, we do another pass with one
4960 * less node in the count. This will push zone_movable_pfn[nid] further
4961 * along on the nodes that still have memory until kernelcore is
4965 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4968 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4969 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4970 zone_movable_pfn
[nid
] =
4971 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4974 /* restore the node_state */
4975 node_states
[N_MEMORY
] = saved_node_state
;
4978 /* Any regular or high memory on that node ? */
4979 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
4981 enum zone_type zone_type
;
4983 if (N_MEMORY
== N_NORMAL_MEMORY
)
4986 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
4987 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4988 if (zone
->present_pages
) {
4989 node_set_state(nid
, N_HIGH_MEMORY
);
4990 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
4991 zone_type
<= ZONE_NORMAL
)
4992 node_set_state(nid
, N_NORMAL_MEMORY
);
4999 * free_area_init_nodes - Initialise all pg_data_t and zone data
5000 * @max_zone_pfn: an array of max PFNs for each zone
5002 * This will call free_area_init_node() for each active node in the system.
5003 * Using the page ranges provided by add_active_range(), the size of each
5004 * zone in each node and their holes is calculated. If the maximum PFN
5005 * between two adjacent zones match, it is assumed that the zone is empty.
5006 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5007 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5008 * starts where the previous one ended. For example, ZONE_DMA32 starts
5009 * at arch_max_dma_pfn.
5011 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5013 unsigned long start_pfn
, end_pfn
;
5016 /* Record where the zone boundaries are */
5017 memset(arch_zone_lowest_possible_pfn
, 0,
5018 sizeof(arch_zone_lowest_possible_pfn
));
5019 memset(arch_zone_highest_possible_pfn
, 0,
5020 sizeof(arch_zone_highest_possible_pfn
));
5021 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5022 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5023 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5024 if (i
== ZONE_MOVABLE
)
5026 arch_zone_lowest_possible_pfn
[i
] =
5027 arch_zone_highest_possible_pfn
[i
-1];
5028 arch_zone_highest_possible_pfn
[i
] =
5029 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5031 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5032 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5034 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5035 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5036 find_zone_movable_pfns_for_nodes();
5038 /* Print out the zone ranges */
5039 printk("Zone ranges:\n");
5040 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5041 if (i
== ZONE_MOVABLE
)
5043 printk(KERN_CONT
" %-8s ", zone_names
[i
]);
5044 if (arch_zone_lowest_possible_pfn
[i
] ==
5045 arch_zone_highest_possible_pfn
[i
])
5046 printk(KERN_CONT
"empty\n");
5048 printk(KERN_CONT
"[mem %0#10lx-%0#10lx]\n",
5049 arch_zone_lowest_possible_pfn
[i
] << PAGE_SHIFT
,
5050 (arch_zone_highest_possible_pfn
[i
]
5051 << PAGE_SHIFT
) - 1);
5054 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5055 printk("Movable zone start for each node\n");
5056 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5057 if (zone_movable_pfn
[i
])
5058 printk(" Node %d: %#010lx\n", i
,
5059 zone_movable_pfn
[i
] << PAGE_SHIFT
);
5062 /* Print out the early node map */
5063 printk("Early memory node ranges\n");
5064 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5065 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid
,
5066 start_pfn
<< PAGE_SHIFT
, (end_pfn
<< PAGE_SHIFT
) - 1);
5068 /* Initialise every node */
5069 mminit_verify_pageflags_layout();
5070 setup_nr_node_ids();
5071 for_each_online_node(nid
) {
5072 pg_data_t
*pgdat
= NODE_DATA(nid
);
5073 free_area_init_node(nid
, NULL
,
5074 find_min_pfn_for_node(nid
), NULL
);
5076 /* Any memory on that node */
5077 if (pgdat
->node_present_pages
)
5078 node_set_state(nid
, N_MEMORY
);
5079 check_for_memory(pgdat
, nid
);
5083 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5085 unsigned long long coremem
;
5089 coremem
= memparse(p
, &p
);
5090 *core
= coremem
>> PAGE_SHIFT
;
5092 /* Paranoid check that UL is enough for the coremem value */
5093 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5099 * kernelcore=size sets the amount of memory for use for allocations that
5100 * cannot be reclaimed or migrated.
5102 static int __init
cmdline_parse_kernelcore(char *p
)
5104 return cmdline_parse_core(p
, &required_kernelcore
);
5108 * movablecore=size sets the amount of memory for use for allocations that
5109 * can be reclaimed or migrated.
5111 static int __init
cmdline_parse_movablecore(char *p
)
5113 return cmdline_parse_core(p
, &required_movablecore
);
5116 early_param("kernelcore", cmdline_parse_kernelcore
);
5117 early_param("movablecore", cmdline_parse_movablecore
);
5119 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5122 * set_dma_reserve - set the specified number of pages reserved in the first zone
5123 * @new_dma_reserve: The number of pages to mark reserved
5125 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5126 * In the DMA zone, a significant percentage may be consumed by kernel image
5127 * and other unfreeable allocations which can skew the watermarks badly. This
5128 * function may optionally be used to account for unfreeable pages in the
5129 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5130 * smaller per-cpu batchsize.
5132 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5134 dma_reserve
= new_dma_reserve
;
5137 void __init
free_area_init(unsigned long *zones_size
)
5139 free_area_init_node(0, zones_size
,
5140 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5143 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5144 unsigned long action
, void *hcpu
)
5146 int cpu
= (unsigned long)hcpu
;
5148 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5149 lru_add_drain_cpu(cpu
);
5153 * Spill the event counters of the dead processor
5154 * into the current processors event counters.
5155 * This artificially elevates the count of the current
5158 vm_events_fold_cpu(cpu
);
5161 * Zero the differential counters of the dead processor
5162 * so that the vm statistics are consistent.
5164 * This is only okay since the processor is dead and cannot
5165 * race with what we are doing.
5167 refresh_cpu_vm_stats(cpu
);
5172 void __init
page_alloc_init(void)
5174 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5178 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5179 * or min_free_kbytes changes.
5181 static void calculate_totalreserve_pages(void)
5183 struct pglist_data
*pgdat
;
5184 unsigned long reserve_pages
= 0;
5185 enum zone_type i
, j
;
5187 for_each_online_pgdat(pgdat
) {
5188 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5189 struct zone
*zone
= pgdat
->node_zones
+ i
;
5190 unsigned long max
= 0;
5192 /* Find valid and maximum lowmem_reserve in the zone */
5193 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5194 if (zone
->lowmem_reserve
[j
] > max
)
5195 max
= zone
->lowmem_reserve
[j
];
5198 /* we treat the high watermark as reserved pages. */
5199 max
+= high_wmark_pages(zone
);
5201 if (max
> zone
->present_pages
)
5202 max
= zone
->present_pages
;
5203 reserve_pages
+= max
;
5205 * Lowmem reserves are not available to
5206 * GFP_HIGHUSER page cache allocations and
5207 * kswapd tries to balance zones to their high
5208 * watermark. As a result, neither should be
5209 * regarded as dirtyable memory, to prevent a
5210 * situation where reclaim has to clean pages
5211 * in order to balance the zones.
5213 zone
->dirty_balance_reserve
= max
;
5216 dirty_balance_reserve
= reserve_pages
;
5217 totalreserve_pages
= reserve_pages
;
5221 * setup_per_zone_lowmem_reserve - called whenever
5222 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5223 * has a correct pages reserved value, so an adequate number of
5224 * pages are left in the zone after a successful __alloc_pages().
5226 static void setup_per_zone_lowmem_reserve(void)
5228 struct pglist_data
*pgdat
;
5229 enum zone_type j
, idx
;
5231 for_each_online_pgdat(pgdat
) {
5232 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5233 struct zone
*zone
= pgdat
->node_zones
+ j
;
5234 unsigned long present_pages
= zone
->present_pages
;
5236 zone
->lowmem_reserve
[j
] = 0;
5240 struct zone
*lower_zone
;
5244 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5245 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5247 lower_zone
= pgdat
->node_zones
+ idx
;
5248 lower_zone
->lowmem_reserve
[j
] = present_pages
/
5249 sysctl_lowmem_reserve_ratio
[idx
];
5250 present_pages
+= lower_zone
->present_pages
;
5255 /* update totalreserve_pages */
5256 calculate_totalreserve_pages();
5259 static void __setup_per_zone_wmarks(void)
5261 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5262 unsigned long lowmem_pages
= 0;
5264 unsigned long flags
;
5266 /* Calculate total number of !ZONE_HIGHMEM pages */
5267 for_each_zone(zone
) {
5268 if (!is_highmem(zone
))
5269 lowmem_pages
+= zone
->present_pages
;
5272 for_each_zone(zone
) {
5275 spin_lock_irqsave(&zone
->lock
, flags
);
5276 tmp
= (u64
)pages_min
* zone
->present_pages
;
5277 do_div(tmp
, lowmem_pages
);
5278 if (is_highmem(zone
)) {
5280 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5281 * need highmem pages, so cap pages_min to a small
5284 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5285 * deltas controls asynch page reclaim, and so should
5286 * not be capped for highmem.
5290 min_pages
= zone
->present_pages
/ 1024;
5291 if (min_pages
< SWAP_CLUSTER_MAX
)
5292 min_pages
= SWAP_CLUSTER_MAX
;
5293 if (min_pages
> 128)
5295 zone
->watermark
[WMARK_MIN
] = min_pages
;
5298 * If it's a lowmem zone, reserve a number of pages
5299 * proportionate to the zone's size.
5301 zone
->watermark
[WMARK_MIN
] = tmp
;
5304 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
5305 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
5307 setup_zone_migrate_reserve(zone
);
5308 spin_unlock_irqrestore(&zone
->lock
, flags
);
5311 /* update totalreserve_pages */
5312 calculate_totalreserve_pages();
5316 * setup_per_zone_wmarks - called when min_free_kbytes changes
5317 * or when memory is hot-{added|removed}
5319 * Ensures that the watermark[min,low,high] values for each zone are set
5320 * correctly with respect to min_free_kbytes.
5322 void setup_per_zone_wmarks(void)
5324 mutex_lock(&zonelists_mutex
);
5325 __setup_per_zone_wmarks();
5326 mutex_unlock(&zonelists_mutex
);
5330 * The inactive anon list should be small enough that the VM never has to
5331 * do too much work, but large enough that each inactive page has a chance
5332 * to be referenced again before it is swapped out.
5334 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5335 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5336 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5337 * the anonymous pages are kept on the inactive list.
5340 * memory ratio inactive anon
5341 * -------------------------------------
5350 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
5352 unsigned int gb
, ratio
;
5354 /* Zone size in gigabytes */
5355 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
5357 ratio
= int_sqrt(10 * gb
);
5361 zone
->inactive_ratio
= ratio
;
5364 static void __meminit
setup_per_zone_inactive_ratio(void)
5369 calculate_zone_inactive_ratio(zone
);
5373 * Initialise min_free_kbytes.
5375 * For small machines we want it small (128k min). For large machines
5376 * we want it large (64MB max). But it is not linear, because network
5377 * bandwidth does not increase linearly with machine size. We use
5379 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5380 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5396 int __meminit
init_per_zone_wmark_min(void)
5398 unsigned long lowmem_kbytes
;
5400 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5402 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5403 if (min_free_kbytes
< 128)
5404 min_free_kbytes
= 128;
5405 if (min_free_kbytes
> 65536)
5406 min_free_kbytes
= 65536;
5407 setup_per_zone_wmarks();
5408 refresh_zone_stat_thresholds();
5409 setup_per_zone_lowmem_reserve();
5410 setup_per_zone_inactive_ratio();
5413 module_init(init_per_zone_wmark_min
)
5416 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5417 * that we can call two helper functions whenever min_free_kbytes
5420 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5421 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5423 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5425 setup_per_zone_wmarks();
5430 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5431 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5436 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5441 zone
->min_unmapped_pages
= (zone
->present_pages
*
5442 sysctl_min_unmapped_ratio
) / 100;
5446 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5447 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5452 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5457 zone
->min_slab_pages
= (zone
->present_pages
*
5458 sysctl_min_slab_ratio
) / 100;
5464 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5465 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5466 * whenever sysctl_lowmem_reserve_ratio changes.
5468 * The reserve ratio obviously has absolutely no relation with the
5469 * minimum watermarks. The lowmem reserve ratio can only make sense
5470 * if in function of the boot time zone sizes.
5472 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5473 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5475 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5476 setup_per_zone_lowmem_reserve();
5481 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5482 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5483 * can have before it gets flushed back to buddy allocator.
5486 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5487 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5493 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5494 if (!write
|| (ret
< 0))
5496 for_each_populated_zone(zone
) {
5497 for_each_possible_cpu(cpu
) {
5499 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5500 setup_pagelist_highmark(
5501 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5507 int hashdist
= HASHDIST_DEFAULT
;
5510 static int __init
set_hashdist(char *str
)
5514 hashdist
= simple_strtoul(str
, &str
, 0);
5517 __setup("hashdist=", set_hashdist
);
5521 * allocate a large system hash table from bootmem
5522 * - it is assumed that the hash table must contain an exact power-of-2
5523 * quantity of entries
5524 * - limit is the number of hash buckets, not the total allocation size
5526 void *__init
alloc_large_system_hash(const char *tablename
,
5527 unsigned long bucketsize
,
5528 unsigned long numentries
,
5531 unsigned int *_hash_shift
,
5532 unsigned int *_hash_mask
,
5533 unsigned long low_limit
,
5534 unsigned long high_limit
)
5536 unsigned long long max
= high_limit
;
5537 unsigned long log2qty
, size
;
5540 /* allow the kernel cmdline to have a say */
5542 /* round applicable memory size up to nearest megabyte */
5543 numentries
= nr_kernel_pages
;
5544 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5545 numentries
>>= 20 - PAGE_SHIFT
;
5546 numentries
<<= 20 - PAGE_SHIFT
;
5548 /* limit to 1 bucket per 2^scale bytes of low memory */
5549 if (scale
> PAGE_SHIFT
)
5550 numentries
>>= (scale
- PAGE_SHIFT
);
5552 numentries
<<= (PAGE_SHIFT
- scale
);
5554 /* Make sure we've got at least a 0-order allocation.. */
5555 if (unlikely(flags
& HASH_SMALL
)) {
5556 /* Makes no sense without HASH_EARLY */
5557 WARN_ON(!(flags
& HASH_EARLY
));
5558 if (!(numentries
>> *_hash_shift
)) {
5559 numentries
= 1UL << *_hash_shift
;
5560 BUG_ON(!numentries
);
5562 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5563 numentries
= PAGE_SIZE
/ bucketsize
;
5565 numentries
= roundup_pow_of_two(numentries
);
5567 /* limit allocation size to 1/16 total memory by default */
5569 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5570 do_div(max
, bucketsize
);
5572 max
= min(max
, 0x80000000ULL
);
5574 if (numentries
< low_limit
)
5575 numentries
= low_limit
;
5576 if (numentries
> max
)
5579 log2qty
= ilog2(numentries
);
5582 size
= bucketsize
<< log2qty
;
5583 if (flags
& HASH_EARLY
)
5584 table
= alloc_bootmem_nopanic(size
);
5586 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5589 * If bucketsize is not a power-of-two, we may free
5590 * some pages at the end of hash table which
5591 * alloc_pages_exact() automatically does
5593 if (get_order(size
) < MAX_ORDER
) {
5594 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5595 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5598 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5601 panic("Failed to allocate %s hash table\n", tablename
);
5603 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5606 ilog2(size
) - PAGE_SHIFT
,
5610 *_hash_shift
= log2qty
;
5612 *_hash_mask
= (1 << log2qty
) - 1;
5617 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5618 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5621 #ifdef CONFIG_SPARSEMEM
5622 return __pfn_to_section(pfn
)->pageblock_flags
;
5624 return zone
->pageblock_flags
;
5625 #endif /* CONFIG_SPARSEMEM */
5628 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5630 #ifdef CONFIG_SPARSEMEM
5631 pfn
&= (PAGES_PER_SECTION
-1);
5632 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5634 pfn
= pfn
- zone
->zone_start_pfn
;
5635 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5636 #endif /* CONFIG_SPARSEMEM */
5640 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5641 * @page: The page within the block of interest
5642 * @start_bitidx: The first bit of interest to retrieve
5643 * @end_bitidx: The last bit of interest
5644 * returns pageblock_bits flags
5646 unsigned long get_pageblock_flags_group(struct page
*page
,
5647 int start_bitidx
, int end_bitidx
)
5650 unsigned long *bitmap
;
5651 unsigned long pfn
, bitidx
;
5652 unsigned long flags
= 0;
5653 unsigned long value
= 1;
5655 zone
= page_zone(page
);
5656 pfn
= page_to_pfn(page
);
5657 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5658 bitidx
= pfn_to_bitidx(zone
, pfn
);
5660 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5661 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5668 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5669 * @page: The page within the block of interest
5670 * @start_bitidx: The first bit of interest
5671 * @end_bitidx: The last bit of interest
5672 * @flags: The flags to set
5674 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5675 int start_bitidx
, int end_bitidx
)
5678 unsigned long *bitmap
;
5679 unsigned long pfn
, bitidx
;
5680 unsigned long value
= 1;
5682 zone
= page_zone(page
);
5683 pfn
= page_to_pfn(page
);
5684 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5685 bitidx
= pfn_to_bitidx(zone
, pfn
);
5686 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5687 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5689 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5691 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5693 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5697 * This function checks whether pageblock includes unmovable pages or not.
5698 * If @count is not zero, it is okay to include less @count unmovable pages
5700 * PageLRU check wihtout isolation or lru_lock could race so that
5701 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5702 * expect this function should be exact.
5704 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
5705 bool skip_hwpoisoned_pages
)
5707 unsigned long pfn
, iter
, found
;
5711 * For avoiding noise data, lru_add_drain_all() should be called
5712 * If ZONE_MOVABLE, the zone never contains unmovable pages
5714 if (zone_idx(zone
) == ZONE_MOVABLE
)
5716 mt
= get_pageblock_migratetype(page
);
5717 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
5720 pfn
= page_to_pfn(page
);
5721 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
5722 unsigned long check
= pfn
+ iter
;
5724 if (!pfn_valid_within(check
))
5727 page
= pfn_to_page(check
);
5729 * We can't use page_count without pin a page
5730 * because another CPU can free compound page.
5731 * This check already skips compound tails of THP
5732 * because their page->_count is zero at all time.
5734 if (!atomic_read(&page
->_count
)) {
5735 if (PageBuddy(page
))
5736 iter
+= (1 << page_order(page
)) - 1;
5741 * The HWPoisoned page may be not in buddy system, and
5742 * page_count() is not 0.
5744 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
5750 * If there are RECLAIMABLE pages, we need to check it.
5751 * But now, memory offline itself doesn't call shrink_slab()
5752 * and it still to be fixed.
5755 * If the page is not RAM, page_count()should be 0.
5756 * we don't need more check. This is an _used_ not-movable page.
5758 * The problematic thing here is PG_reserved pages. PG_reserved
5759 * is set to both of a memory hole page and a _used_ kernel
5768 bool is_pageblock_removable_nolock(struct page
*page
)
5774 * We have to be careful here because we are iterating over memory
5775 * sections which are not zone aware so we might end up outside of
5776 * the zone but still within the section.
5777 * We have to take care about the node as well. If the node is offline
5778 * its NODE_DATA will be NULL - see page_zone.
5780 if (!node_online(page_to_nid(page
)))
5783 zone
= page_zone(page
);
5784 pfn
= page_to_pfn(page
);
5785 if (zone
->zone_start_pfn
> pfn
||
5786 zone
->zone_start_pfn
+ zone
->spanned_pages
<= pfn
)
5789 return !has_unmovable_pages(zone
, page
, 0, true);
5794 static unsigned long pfn_max_align_down(unsigned long pfn
)
5796 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5797 pageblock_nr_pages
) - 1);
5800 static unsigned long pfn_max_align_up(unsigned long pfn
)
5802 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5803 pageblock_nr_pages
));
5806 /* [start, end) must belong to a single zone. */
5807 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
5808 unsigned long start
, unsigned long end
)
5810 /* This function is based on compact_zone() from compaction.c. */
5811 unsigned long nr_reclaimed
;
5812 unsigned long pfn
= start
;
5813 unsigned int tries
= 0;
5818 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
5819 if (fatal_signal_pending(current
)) {
5824 if (list_empty(&cc
->migratepages
)) {
5825 cc
->nr_migratepages
= 0;
5826 pfn
= isolate_migratepages_range(cc
->zone
, cc
,
5833 } else if (++tries
== 5) {
5834 ret
= ret
< 0 ? ret
: -EBUSY
;
5838 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
5840 cc
->nr_migratepages
-= nr_reclaimed
;
5842 ret
= migrate_pages(&cc
->migratepages
,
5843 alloc_migrate_target
,
5844 0, false, MIGRATE_SYNC
,
5848 putback_movable_pages(&cc
->migratepages
);
5849 return ret
> 0 ? 0 : ret
;
5853 * alloc_contig_range() -- tries to allocate given range of pages
5854 * @start: start PFN to allocate
5855 * @end: one-past-the-last PFN to allocate
5856 * @migratetype: migratetype of the underlaying pageblocks (either
5857 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
5858 * in range must have the same migratetype and it must
5859 * be either of the two.
5861 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
5862 * aligned, however it's the caller's responsibility to guarantee that
5863 * we are the only thread that changes migrate type of pageblocks the
5866 * The PFN range must belong to a single zone.
5868 * Returns zero on success or negative error code. On success all
5869 * pages which PFN is in [start, end) are allocated for the caller and
5870 * need to be freed with free_contig_range().
5872 int alloc_contig_range(unsigned long start
, unsigned long end
,
5873 unsigned migratetype
)
5875 unsigned long outer_start
, outer_end
;
5878 struct compact_control cc
= {
5879 .nr_migratepages
= 0,
5881 .zone
= page_zone(pfn_to_page(start
)),
5883 .ignore_skip_hint
= true,
5885 INIT_LIST_HEAD(&cc
.migratepages
);
5888 * What we do here is we mark all pageblocks in range as
5889 * MIGRATE_ISOLATE. Because pageblock and max order pages may
5890 * have different sizes, and due to the way page allocator
5891 * work, we align the range to biggest of the two pages so
5892 * that page allocator won't try to merge buddies from
5893 * different pageblocks and change MIGRATE_ISOLATE to some
5894 * other migration type.
5896 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
5897 * migrate the pages from an unaligned range (ie. pages that
5898 * we are interested in). This will put all the pages in
5899 * range back to page allocator as MIGRATE_ISOLATE.
5901 * When this is done, we take the pages in range from page
5902 * allocator removing them from the buddy system. This way
5903 * page allocator will never consider using them.
5905 * This lets us mark the pageblocks back as
5906 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
5907 * aligned range but not in the unaligned, original range are
5908 * put back to page allocator so that buddy can use them.
5911 ret
= start_isolate_page_range(pfn_max_align_down(start
),
5912 pfn_max_align_up(end
), migratetype
,
5917 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
5922 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
5923 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
5924 * more, all pages in [start, end) are free in page allocator.
5925 * What we are going to do is to allocate all pages from
5926 * [start, end) (that is remove them from page allocator).
5928 * The only problem is that pages at the beginning and at the
5929 * end of interesting range may be not aligned with pages that
5930 * page allocator holds, ie. they can be part of higher order
5931 * pages. Because of this, we reserve the bigger range and
5932 * once this is done free the pages we are not interested in.
5934 * We don't have to hold zone->lock here because the pages are
5935 * isolated thus they won't get removed from buddy.
5938 lru_add_drain_all();
5942 outer_start
= start
;
5943 while (!PageBuddy(pfn_to_page(outer_start
))) {
5944 if (++order
>= MAX_ORDER
) {
5948 outer_start
&= ~0UL << order
;
5951 /* Make sure the range is really isolated. */
5952 if (test_pages_isolated(outer_start
, end
, false)) {
5953 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
5960 /* Grab isolated pages from freelists. */
5961 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
5967 /* Free head and tail (if any) */
5968 if (start
!= outer_start
)
5969 free_contig_range(outer_start
, start
- outer_start
);
5970 if (end
!= outer_end
)
5971 free_contig_range(end
, outer_end
- end
);
5974 undo_isolate_page_range(pfn_max_align_down(start
),
5975 pfn_max_align_up(end
), migratetype
);
5979 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
5981 unsigned int count
= 0;
5983 for (; nr_pages
--; pfn
++) {
5984 struct page
*page
= pfn_to_page(pfn
);
5986 count
+= page_count(page
) != 1;
5989 WARN(count
!= 0, "%d pages are still in use!\n", count
);
5993 #ifdef CONFIG_MEMORY_HOTPLUG
5994 static int __meminit
__zone_pcp_update(void *data
)
5996 struct zone
*zone
= data
;
5998 unsigned long batch
= zone_batchsize(zone
), flags
;
6000 for_each_possible_cpu(cpu
) {
6001 struct per_cpu_pageset
*pset
;
6002 struct per_cpu_pages
*pcp
;
6004 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6007 local_irq_save(flags
);
6009 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
6010 drain_zonestat(zone
, pset
);
6011 setup_pageset(pset
, batch
);
6012 local_irq_restore(flags
);
6017 void __meminit
zone_pcp_update(struct zone
*zone
)
6019 stop_machine(__zone_pcp_update
, zone
, NULL
);
6023 void zone_pcp_reset(struct zone
*zone
)
6025 unsigned long flags
;
6027 struct per_cpu_pageset
*pset
;
6029 /* avoid races with drain_pages() */
6030 local_irq_save(flags
);
6031 if (zone
->pageset
!= &boot_pageset
) {
6032 for_each_online_cpu(cpu
) {
6033 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6034 drain_zonestat(zone
, pset
);
6036 free_percpu(zone
->pageset
);
6037 zone
->pageset
= &boot_pageset
;
6039 local_irq_restore(flags
);
6042 #ifdef CONFIG_MEMORY_HOTREMOVE
6044 * All pages in the range must be isolated before calling this.
6047 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6053 unsigned long flags
;
6054 /* find the first valid pfn */
6055 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6060 zone
= page_zone(pfn_to_page(pfn
));
6061 spin_lock_irqsave(&zone
->lock
, flags
);
6063 while (pfn
< end_pfn
) {
6064 if (!pfn_valid(pfn
)) {
6068 page
= pfn_to_page(pfn
);
6070 * The HWPoisoned page may be not in buddy system, and
6071 * page_count() is not 0.
6073 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6075 SetPageReserved(page
);
6079 BUG_ON(page_count(page
));
6080 BUG_ON(!PageBuddy(page
));
6081 order
= page_order(page
);
6082 #ifdef CONFIG_DEBUG_VM
6083 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6084 pfn
, 1 << order
, end_pfn
);
6086 list_del(&page
->lru
);
6087 rmv_page_order(page
);
6088 zone
->free_area
[order
].nr_free
--;
6089 for (i
= 0; i
< (1 << order
); i
++)
6090 SetPageReserved((page
+i
));
6091 pfn
+= (1 << order
);
6093 spin_unlock_irqrestore(&zone
->lock
, flags
);
6097 #ifdef CONFIG_MEMORY_FAILURE
6098 bool is_free_buddy_page(struct page
*page
)
6100 struct zone
*zone
= page_zone(page
);
6101 unsigned long pfn
= page_to_pfn(page
);
6102 unsigned long flags
;
6105 spin_lock_irqsave(&zone
->lock
, flags
);
6106 for (order
= 0; order
< MAX_ORDER
; order
++) {
6107 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6109 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6112 spin_unlock_irqrestore(&zone
->lock
, flags
);
6114 return order
< MAX_ORDER
;
6118 static const struct trace_print_flags pageflag_names
[] = {
6119 {1UL << PG_locked
, "locked" },
6120 {1UL << PG_error
, "error" },
6121 {1UL << PG_referenced
, "referenced" },
6122 {1UL << PG_uptodate
, "uptodate" },
6123 {1UL << PG_dirty
, "dirty" },
6124 {1UL << PG_lru
, "lru" },
6125 {1UL << PG_active
, "active" },
6126 {1UL << PG_slab
, "slab" },
6127 {1UL << PG_owner_priv_1
, "owner_priv_1" },
6128 {1UL << PG_arch_1
, "arch_1" },
6129 {1UL << PG_reserved
, "reserved" },
6130 {1UL << PG_private
, "private" },
6131 {1UL << PG_private_2
, "private_2" },
6132 {1UL << PG_writeback
, "writeback" },
6133 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6134 {1UL << PG_head
, "head" },
6135 {1UL << PG_tail
, "tail" },
6137 {1UL << PG_compound
, "compound" },
6139 {1UL << PG_swapcache
, "swapcache" },
6140 {1UL << PG_mappedtodisk
, "mappedtodisk" },
6141 {1UL << PG_reclaim
, "reclaim" },
6142 {1UL << PG_swapbacked
, "swapbacked" },
6143 {1UL << PG_unevictable
, "unevictable" },
6145 {1UL << PG_mlocked
, "mlocked" },
6147 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6148 {1UL << PG_uncached
, "uncached" },
6150 #ifdef CONFIG_MEMORY_FAILURE
6151 {1UL << PG_hwpoison
, "hwpoison" },
6153 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6154 {1UL << PG_compound_lock
, "compound_lock" },
6158 static void dump_page_flags(unsigned long flags
)
6160 const char *delim
= "";
6164 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names
) != __NR_PAGEFLAGS
);
6166 printk(KERN_ALERT
"page flags: %#lx(", flags
);
6168 /* remove zone id */
6169 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
6171 for (i
= 0; i
< ARRAY_SIZE(pageflag_names
) && flags
; i
++) {
6173 mask
= pageflag_names
[i
].mask
;
6174 if ((flags
& mask
) != mask
)
6178 printk("%s%s", delim
, pageflag_names
[i
].name
);
6182 /* check for left over flags */
6184 printk("%s%#lx", delim
, flags
);
6189 void dump_page(struct page
*page
)
6192 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6193 page
, atomic_read(&page
->_count
), page_mapcount(page
),
6194 page
->mapping
, page
->index
);
6195 dump_page_flags(page
->flags
);
6196 mem_cgroup_print_bad_page(page
);