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/kasan.h>
29 #include <linux/module.h>
30 #include <linux/suspend.h>
31 #include <linux/pagevec.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/ratelimit.h>
35 #include <linux/oom.h>
36 #include <linux/notifier.h>
37 #include <linux/topology.h>
38 #include <linux/sysctl.h>
39 #include <linux/cpu.h>
40 #include <linux/cpuset.h>
41 #include <linux/memory_hotplug.h>
42 #include <linux/nodemask.h>
43 #include <linux/vmalloc.h>
44 #include <linux/vmstat.h>
45 #include <linux/mempolicy.h>
46 #include <linux/stop_machine.h>
47 #include <linux/sort.h>
48 #include <linux/pfn.h>
49 #include <linux/backing-dev.h>
50 #include <linux/fault-inject.h>
51 #include <linux/page-isolation.h>
52 #include <linux/page_ext.h>
53 #include <linux/debugobjects.h>
54 #include <linux/kmemleak.h>
55 #include <linux/compaction.h>
56 #include <trace/events/kmem.h>
57 #include <linux/prefetch.h>
58 #include <linux/mm_inline.h>
59 #include <linux/migrate.h>
60 #include <linux/page_ext.h>
61 #include <linux/hugetlb.h>
62 #include <linux/sched/rt.h>
63 #include <linux/page_owner.h>
64 #include <linux/kthread.h>
66 #include <asm/sections.h>
67 #include <asm/tlbflush.h>
68 #include <asm/div64.h>
71 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
72 static DEFINE_MUTEX(pcp_batch_high_lock
);
73 #define MIN_PERCPU_PAGELIST_FRACTION (8)
75 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
76 DEFINE_PER_CPU(int, numa_node
);
77 EXPORT_PER_CPU_SYMBOL(numa_node
);
80 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
82 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
83 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
84 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
85 * defined in <linux/topology.h>.
87 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
88 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
89 int _node_numa_mem_
[MAX_NUMNODES
];
93 * Array of node states.
95 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
96 [N_POSSIBLE
] = NODE_MASK_ALL
,
97 [N_ONLINE
] = { { [0] = 1UL } },
99 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
100 #ifdef CONFIG_HIGHMEM
101 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
103 #ifdef CONFIG_MOVABLE_NODE
104 [N_MEMORY
] = { { [0] = 1UL } },
106 [N_CPU
] = { { [0] = 1UL } },
109 EXPORT_SYMBOL(node_states
);
111 /* Protect totalram_pages and zone->managed_pages */
112 static DEFINE_SPINLOCK(managed_page_count_lock
);
114 unsigned long totalram_pages __read_mostly
;
115 unsigned long totalreserve_pages __read_mostly
;
116 unsigned long totalcma_pages __read_mostly
;
118 * When calculating the number of globally allowed dirty pages, there
119 * is a certain number of per-zone reserves that should not be
120 * considered dirtyable memory. This is the sum of those reserves
121 * over all existing zones that contribute dirtyable memory.
123 unsigned long dirty_balance_reserve __read_mostly
;
125 int percpu_pagelist_fraction
;
126 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
128 #ifdef CONFIG_PM_SLEEP
130 * The following functions are used by the suspend/hibernate code to temporarily
131 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
132 * while devices are suspended. To avoid races with the suspend/hibernate code,
133 * they should always be called with pm_mutex held (gfp_allowed_mask also should
134 * only be modified with pm_mutex held, unless the suspend/hibernate code is
135 * guaranteed not to run in parallel with that modification).
138 static gfp_t saved_gfp_mask
;
140 void pm_restore_gfp_mask(void)
142 WARN_ON(!mutex_is_locked(&pm_mutex
));
143 if (saved_gfp_mask
) {
144 gfp_allowed_mask
= saved_gfp_mask
;
149 void pm_restrict_gfp_mask(void)
151 WARN_ON(!mutex_is_locked(&pm_mutex
));
152 WARN_ON(saved_gfp_mask
);
153 saved_gfp_mask
= gfp_allowed_mask
;
154 gfp_allowed_mask
&= ~(__GFP_IO
| __GFP_FS
);
157 bool pm_suspended_storage(void)
159 if ((gfp_allowed_mask
& (__GFP_IO
| __GFP_FS
)) == (__GFP_IO
| __GFP_FS
))
163 #endif /* CONFIG_PM_SLEEP */
165 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
166 unsigned int pageblock_order __read_mostly
;
169 static void __free_pages_ok(struct page
*page
, unsigned int order
);
172 * results with 256, 32 in the lowmem_reserve sysctl:
173 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
174 * 1G machine -> (16M dma, 784M normal, 224M high)
175 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
176 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
177 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
179 * TBD: should special case ZONE_DMA32 machines here - in those we normally
180 * don't need any ZONE_NORMAL reservation
182 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
183 #ifdef CONFIG_ZONE_DMA
186 #ifdef CONFIG_ZONE_DMA32
189 #ifdef CONFIG_HIGHMEM
195 EXPORT_SYMBOL(totalram_pages
);
197 static char * const zone_names
[MAX_NR_ZONES
] = {
198 #ifdef CONFIG_ZONE_DMA
201 #ifdef CONFIG_ZONE_DMA32
205 #ifdef CONFIG_HIGHMEM
209 #ifdef CONFIG_ZONE_DEVICE
214 static void free_compound_page(struct page
*page
);
215 compound_page_dtor
* const compound_page_dtors
[] = {
218 #ifdef CONFIG_HUGETLB_PAGE
224 * Try to keep at least this much lowmem free. Do not allow normal
225 * allocations below this point, only high priority ones. Automatically
226 * tuned according to the amount of memory in the system.
228 int min_free_kbytes
= 1024;
229 int user_min_free_kbytes
= -1;
232 * Extra memory for the system to try freeing. Used to temporarily
233 * free memory, to make space for new workloads. Anyone can allocate
234 * down to the min watermarks controlled by min_free_kbytes above.
236 int extra_free_kbytes
= 0;
238 static unsigned long __meminitdata nr_kernel_pages
;
239 static unsigned long __meminitdata nr_all_pages
;
240 static unsigned long __meminitdata dma_reserve
;
242 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
243 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
244 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
245 static unsigned long __initdata required_kernelcore
;
246 static unsigned long __initdata required_movablecore
;
247 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
249 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
251 EXPORT_SYMBOL(movable_zone
);
252 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
255 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
256 int nr_online_nodes __read_mostly
= 1;
257 EXPORT_SYMBOL(nr_node_ids
);
258 EXPORT_SYMBOL(nr_online_nodes
);
261 int page_group_by_mobility_disabled __read_mostly
;
263 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
266 * Determine how many pages need to be initialized durig early boot
267 * (non-deferred initialization).
268 * The value of first_deferred_pfn will be set later, once non-deferred pages
269 * are initialized, but for now set it ULONG_MAX.
271 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
273 phys_addr_t start_addr
, end_addr
;
274 unsigned long max_pgcnt
;
275 unsigned long reserved
;
278 * Initialise at least 2G of a node but also take into account that
279 * two large system hashes that can take up 1GB for 0.25TB/node.
281 max_pgcnt
= max(2UL << (30 - PAGE_SHIFT
),
282 (pgdat
->node_spanned_pages
>> 8));
285 * Compensate the all the memblock reservations (e.g. crash kernel)
286 * from the initial estimation to make sure we will initialize enough
289 start_addr
= PFN_PHYS(pgdat
->node_start_pfn
);
290 end_addr
= PFN_PHYS(pgdat
->node_start_pfn
+ max_pgcnt
);
291 reserved
= memblock_reserved_memory_within(start_addr
, end_addr
);
292 max_pgcnt
+= PHYS_PFN(reserved
);
294 pgdat
->static_init_pgcnt
= min(max_pgcnt
, pgdat
->node_spanned_pages
);
295 pgdat
->first_deferred_pfn
= ULONG_MAX
;
298 /* Returns true if the struct page for the pfn is uninitialised */
299 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
301 int nid
= early_pfn_to_nid(pfn
);
303 if (node_online(nid
) && pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
309 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
311 if (pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
318 * Returns false when the remaining initialisation should be deferred until
319 * later in the boot cycle when it can be parallelised.
321 static inline bool update_defer_init(pg_data_t
*pgdat
,
322 unsigned long pfn
, unsigned long zone_end
,
323 unsigned long *nr_initialised
)
325 /* Always populate low zones for address-contrained allocations */
326 if (zone_end
< pgdat_end_pfn(pgdat
))
328 /* Initialise at least 2G of the highest zone */
330 if ((*nr_initialised
> pgdat
->static_init_pgcnt
) &&
331 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
332 pgdat
->first_deferred_pfn
= pfn
;
339 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
343 static inline bool early_page_uninitialised(unsigned long pfn
)
348 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
353 static inline bool update_defer_init(pg_data_t
*pgdat
,
354 unsigned long pfn
, unsigned long zone_end
,
355 unsigned long *nr_initialised
)
362 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
364 if (unlikely(page_group_by_mobility_disabled
&&
365 migratetype
< MIGRATE_PCPTYPES
))
366 migratetype
= MIGRATE_UNMOVABLE
;
368 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
369 PB_migrate
, PB_migrate_end
);
372 #ifdef CONFIG_DEBUG_VM
373 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
377 unsigned long pfn
= page_to_pfn(page
);
378 unsigned long sp
, start_pfn
;
381 seq
= zone_span_seqbegin(zone
);
382 start_pfn
= zone
->zone_start_pfn
;
383 sp
= zone
->spanned_pages
;
384 if (!zone_spans_pfn(zone
, pfn
))
386 } while (zone_span_seqretry(zone
, seq
));
389 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
390 pfn
, zone_to_nid(zone
), zone
->name
,
391 start_pfn
, start_pfn
+ sp
);
396 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
398 if (!pfn_valid_within(page_to_pfn(page
)))
400 if (zone
!= page_zone(page
))
406 * Temporary debugging check for pages not lying within a given zone.
408 static int bad_range(struct zone
*zone
, struct page
*page
)
410 if (page_outside_zone_boundaries(zone
, page
))
412 if (!page_is_consistent(zone
, page
))
418 static inline int bad_range(struct zone
*zone
, struct page
*page
)
424 static void bad_page(struct page
*page
, const char *reason
,
425 unsigned long bad_flags
)
427 static unsigned long resume
;
428 static unsigned long nr_shown
;
429 static unsigned long nr_unshown
;
431 /* Don't complain about poisoned pages */
432 if (PageHWPoison(page
)) {
433 page_mapcount_reset(page
); /* remove PageBuddy */
438 * Allow a burst of 60 reports, then keep quiet for that minute;
439 * or allow a steady drip of one report per second.
441 if (nr_shown
== 60) {
442 if (time_before(jiffies
, resume
)) {
448 "BUG: Bad page state: %lu messages suppressed\n",
455 resume
= jiffies
+ 60 * HZ
;
457 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
458 current
->comm
, page_to_pfn(page
));
459 dump_page_badflags(page
, reason
, bad_flags
);
464 /* Leave bad fields for debug, except PageBuddy could make trouble */
465 page_mapcount_reset(page
); /* remove PageBuddy */
466 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
470 * Higher-order pages are called "compound pages". They are structured thusly:
472 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
474 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
475 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
477 * The first tail page's ->compound_dtor holds the offset in array of compound
478 * page destructors. See compound_page_dtors.
480 * The first tail page's ->compound_order holds the order of allocation.
481 * This usage means that zero-order pages may not be compound.
484 static void free_compound_page(struct page
*page
)
486 __free_pages_ok(page
, compound_order(page
));
489 void prep_compound_page(struct page
*page
, unsigned int order
)
492 int nr_pages
= 1 << order
;
494 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
495 set_compound_order(page
, order
);
497 for (i
= 1; i
< nr_pages
; i
++) {
498 struct page
*p
= page
+ i
;
499 set_page_count(p
, 0);
500 set_compound_head(p
, page
);
504 #ifdef CONFIG_DEBUG_PAGEALLOC
505 unsigned int _debug_guardpage_minorder
;
506 bool _debug_pagealloc_enabled __read_mostly
;
507 bool _debug_guardpage_enabled __read_mostly
;
509 static int __init
early_debug_pagealloc(char *buf
)
514 if (strcmp(buf
, "on") == 0)
515 _debug_pagealloc_enabled
= true;
519 early_param("debug_pagealloc", early_debug_pagealloc
);
521 static bool need_debug_guardpage(void)
523 /* If we don't use debug_pagealloc, we don't need guard page */
524 if (!debug_pagealloc_enabled())
530 static void init_debug_guardpage(void)
532 if (!debug_pagealloc_enabled())
535 _debug_guardpage_enabled
= true;
538 struct page_ext_operations debug_guardpage_ops
= {
539 .need
= need_debug_guardpage
,
540 .init
= init_debug_guardpage
,
543 static int __init
debug_guardpage_minorder_setup(char *buf
)
547 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
548 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
551 _debug_guardpage_minorder
= res
;
552 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
555 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
557 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
558 unsigned int order
, int migratetype
)
560 struct page_ext
*page_ext
;
562 if (!debug_guardpage_enabled())
565 page_ext
= lookup_page_ext(page
);
566 if (unlikely(!page_ext
))
569 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
571 INIT_LIST_HEAD(&page
->lru
);
572 set_page_private(page
, order
);
573 /* Guard pages are not available for any usage */
574 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
577 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
578 unsigned int order
, int migratetype
)
580 struct page_ext
*page_ext
;
582 if (!debug_guardpage_enabled())
585 page_ext
= lookup_page_ext(page
);
586 if (unlikely(!page_ext
))
589 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
591 set_page_private(page
, 0);
592 if (!is_migrate_isolate(migratetype
))
593 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
596 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
597 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
598 unsigned int order
, int migratetype
) {}
599 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
600 unsigned int order
, int migratetype
) {}
603 static inline void set_page_order(struct page
*page
, unsigned int order
)
605 set_page_private(page
, order
);
606 __SetPageBuddy(page
);
609 static inline void rmv_page_order(struct page
*page
)
611 __ClearPageBuddy(page
);
612 set_page_private(page
, 0);
616 * This function checks whether a page is free && is the buddy
617 * we can do coalesce a page and its buddy if
618 * (a) the buddy is not in a hole &&
619 * (b) the buddy is in the buddy system &&
620 * (c) a page and its buddy have the same order &&
621 * (d) a page and its buddy are in the same zone.
623 * For recording whether a page is in the buddy system, we set ->_mapcount
624 * PAGE_BUDDY_MAPCOUNT_VALUE.
625 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
626 * serialized by zone->lock.
628 * For recording page's order, we use page_private(page).
630 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
633 if (!pfn_valid_within(page_to_pfn(buddy
)))
636 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
637 if (page_zone_id(page
) != page_zone_id(buddy
))
640 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
645 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
647 * zone check is done late to avoid uselessly
648 * calculating zone/node ids for pages that could
651 if (page_zone_id(page
) != page_zone_id(buddy
))
654 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
662 * Freeing function for a buddy system allocator.
664 * The concept of a buddy system is to maintain direct-mapped table
665 * (containing bit values) for memory blocks of various "orders".
666 * The bottom level table contains the map for the smallest allocatable
667 * units of memory (here, pages), and each level above it describes
668 * pairs of units from the levels below, hence, "buddies".
669 * At a high level, all that happens here is marking the table entry
670 * at the bottom level available, and propagating the changes upward
671 * as necessary, plus some accounting needed to play nicely with other
672 * parts of the VM system.
673 * At each level, we keep a list of pages, which are heads of continuous
674 * free pages of length of (1 << order) and marked with _mapcount
675 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
677 * So when we are allocating or freeing one, we can derive the state of the
678 * other. That is, if we allocate a small block, and both were
679 * free, the remainder of the region must be split into blocks.
680 * If a block is freed, and its buddy is also free, then this
681 * triggers coalescing into a block of larger size.
686 static inline void __free_one_page(struct page
*page
,
688 struct zone
*zone
, unsigned int order
,
691 unsigned long page_idx
;
692 unsigned long combined_idx
;
693 unsigned long uninitialized_var(buddy_idx
);
695 unsigned int max_order
;
697 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
699 VM_BUG_ON(!zone_is_initialized(zone
));
700 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
702 VM_BUG_ON(migratetype
== -1);
703 if (likely(!is_migrate_isolate(migratetype
)))
704 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
706 page_idx
= pfn
& ((1 << MAX_ORDER
) - 1);
708 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
709 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
712 while (order
< max_order
- 1) {
713 buddy_idx
= __find_buddy_index(page_idx
, order
);
714 buddy
= page
+ (buddy_idx
- page_idx
);
715 if (!page_is_buddy(page
, buddy
, order
))
718 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
719 * merge with it and move up one order.
721 if (page_is_guard(buddy
)) {
722 clear_page_guard(zone
, buddy
, order
, migratetype
);
724 list_del(&buddy
->lru
);
725 zone
->free_area
[order
].nr_free
--;
726 rmv_page_order(buddy
);
728 combined_idx
= buddy_idx
& page_idx
;
729 page
= page
+ (combined_idx
- page_idx
);
730 page_idx
= combined_idx
;
733 if (max_order
< MAX_ORDER
) {
734 /* If we are here, it means order is >= pageblock_order.
735 * We want to prevent merge between freepages on isolate
736 * pageblock and normal pageblock. Without this, pageblock
737 * isolation could cause incorrect freepage or CMA accounting.
739 * We don't want to hit this code for the more frequent
742 if (unlikely(has_isolate_pageblock(zone
))) {
745 buddy_idx
= __find_buddy_index(page_idx
, order
);
746 buddy
= page
+ (buddy_idx
- page_idx
);
747 buddy_mt
= get_pageblock_migratetype(buddy
);
749 if (migratetype
!= buddy_mt
750 && (is_migrate_isolate(migratetype
) ||
751 is_migrate_isolate(buddy_mt
)))
755 goto continue_merging
;
759 set_page_order(page
, order
);
762 * If this is not the largest possible page, check if the buddy
763 * of the next-highest order is free. If it is, it's possible
764 * that pages are being freed that will coalesce soon. In case,
765 * that is happening, add the free page to the tail of the list
766 * so it's less likely to be used soon and more likely to be merged
767 * as a higher order page
769 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
770 struct page
*higher_page
, *higher_buddy
;
771 combined_idx
= buddy_idx
& page_idx
;
772 higher_page
= page
+ (combined_idx
- page_idx
);
773 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
774 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
775 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
776 list_add_tail(&page
->lru
,
777 &zone
->free_area
[order
].free_list
[migratetype
]);
782 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
784 zone
->free_area
[order
].nr_free
++;
787 static inline int free_pages_check(struct page
*page
)
789 const char *bad_reason
= NULL
;
790 unsigned long bad_flags
= 0;
792 if (unlikely(page_mapcount(page
)))
793 bad_reason
= "nonzero mapcount";
794 if (unlikely(page
->mapping
!= NULL
))
795 bad_reason
= "non-NULL mapping";
796 if (unlikely(atomic_read(&page
->_count
) != 0))
797 bad_reason
= "nonzero _count";
798 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
799 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
800 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
803 if (unlikely(page
->mem_cgroup
))
804 bad_reason
= "page still charged to cgroup";
806 if (unlikely(bad_reason
)) {
807 bad_page(page
, bad_reason
, bad_flags
);
810 page_cpupid_reset_last(page
);
811 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
812 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
817 * Frees a number of pages from the PCP lists
818 * Assumes all pages on list are in same zone, and of same order.
819 * count is the number of pages to free.
821 * If the zone was previously in an "all pages pinned" state then look to
822 * see if this freeing clears that state.
824 * And clear the zone's pages_scanned counter, to hold off the "all pages are
825 * pinned" detection logic.
827 static void free_pcppages_bulk(struct zone
*zone
, int count
,
828 struct per_cpu_pages
*pcp
)
833 unsigned long nr_scanned
;
835 spin_lock(&zone
->lock
);
836 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
838 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
842 struct list_head
*list
;
845 * Remove pages from lists in a round-robin fashion. A
846 * batch_free count is maintained that is incremented when an
847 * empty list is encountered. This is so more pages are freed
848 * off fuller lists instead of spinning excessively around empty
853 if (++migratetype
== MIGRATE_PCPTYPES
)
855 list
= &pcp
->lists
[migratetype
];
856 } while (list_empty(list
));
858 /* This is the only non-empty list. Free them all. */
859 if (batch_free
== MIGRATE_PCPTYPES
)
860 batch_free
= to_free
;
863 int mt
; /* migratetype of the to-be-freed page */
865 page
= list_entry(list
->prev
, struct page
, lru
);
866 /* must delete as __free_one_page list manipulates */
867 list_del(&page
->lru
);
869 mt
= get_pcppage_migratetype(page
);
870 /* MIGRATE_ISOLATE page should not go to pcplists */
871 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
872 /* Pageblock could have been isolated meanwhile */
873 if (unlikely(has_isolate_pageblock(zone
)))
874 mt
= get_pageblock_migratetype(page
);
876 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
877 trace_mm_page_pcpu_drain(page
, 0, mt
);
878 } while (--to_free
&& --batch_free
&& !list_empty(list
));
880 spin_unlock(&zone
->lock
);
883 static void free_one_page(struct zone
*zone
,
884 struct page
*page
, unsigned long pfn
,
888 unsigned long nr_scanned
;
889 spin_lock(&zone
->lock
);
890 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
892 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
894 if (unlikely(has_isolate_pageblock(zone
) ||
895 is_migrate_isolate(migratetype
))) {
896 migratetype
= get_pfnblock_migratetype(page
, pfn
);
898 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
899 spin_unlock(&zone
->lock
);
902 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
907 * We rely page->lru.next never has bit 0 set, unless the page
908 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
910 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
912 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
916 if (unlikely(!PageTail(page
))) {
917 bad_page(page
, "PageTail not set", 0);
920 if (unlikely(compound_head(page
) != head_page
)) {
921 bad_page(page
, "compound_head not consistent", 0);
926 clear_compound_head(page
);
930 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
931 unsigned long zone
, int nid
)
933 set_page_links(page
, zone
, nid
, pfn
);
934 init_page_count(page
);
935 page_mapcount_reset(page
);
936 page_cpupid_reset_last(page
);
938 INIT_LIST_HEAD(&page
->lru
);
939 #ifdef WANT_PAGE_VIRTUAL
940 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
941 if (!is_highmem_idx(zone
))
942 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
946 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
949 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
952 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
953 static void init_reserved_page(unsigned long pfn
)
958 if (!early_page_uninitialised(pfn
))
961 nid
= early_pfn_to_nid(pfn
);
962 pgdat
= NODE_DATA(nid
);
964 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
965 struct zone
*zone
= &pgdat
->node_zones
[zid
];
967 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
970 __init_single_pfn(pfn
, zid
, nid
);
973 static inline void init_reserved_page(unsigned long pfn
)
976 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
979 * Initialised pages do not have PageReserved set. This function is
980 * called for each range allocated by the bootmem allocator and
981 * marks the pages PageReserved. The remaining valid pages are later
982 * sent to the buddy page allocator.
984 void __meminit
reserve_bootmem_region(phys_addr_t start
, phys_addr_t end
)
986 unsigned long start_pfn
= PFN_DOWN(start
);
987 unsigned long end_pfn
= PFN_UP(end
);
989 for (; start_pfn
< end_pfn
; start_pfn
++) {
990 if (pfn_valid(start_pfn
)) {
991 struct page
*page
= pfn_to_page(start_pfn
);
993 init_reserved_page(start_pfn
);
995 /* Avoid false-positive PageTail() */
996 INIT_LIST_HEAD(&page
->lru
);
998 SetPageReserved(page
);
1003 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
1005 bool compound
= PageCompound(page
);
1008 VM_BUG_ON_PAGE(PageTail(page
), page
);
1009 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1011 trace_mm_page_free(page
, order
);
1012 kmemcheck_free_shadow(page
, order
);
1013 kasan_free_pages(page
, order
);
1015 if (PageMappingFlags(page
))
1016 page
->mapping
= NULL
;
1017 bad
+= free_pages_check(page
);
1018 for (i
= 1; i
< (1 << order
); i
++) {
1020 bad
+= free_tail_pages_check(page
, page
+ i
);
1021 bad
+= free_pages_check(page
+ i
);
1026 reset_page_owner(page
, order
);
1028 if (!PageHighMem(page
)) {
1029 debug_check_no_locks_freed(page_address(page
),
1030 PAGE_SIZE
<< order
);
1031 debug_check_no_obj_freed(page_address(page
),
1032 PAGE_SIZE
<< order
);
1034 arch_free_page(page
, order
);
1035 kernel_map_pages(page
, 1 << order
, 0);
1040 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1042 unsigned long flags
;
1044 unsigned long pfn
= page_to_pfn(page
);
1046 if (!free_pages_prepare(page
, order
))
1049 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1050 local_irq_save(flags
);
1051 __count_vm_events(PGFREE
, 1 << order
);
1052 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1053 local_irq_restore(flags
);
1056 void put_page_freelist(struct page
*page
)
1058 if (put_page_testzero(page
)) {
1059 mem_cgroup_uncharge(page
);
1060 __free_pages_ok(page
, 0);
1063 EXPORT_SYMBOL(put_page_freelist
);
1065 static void __init
__free_pages_boot_core(struct page
*page
,
1066 unsigned long pfn
, unsigned int order
)
1068 unsigned int nr_pages
= 1 << order
;
1069 struct page
*p
= page
;
1073 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1075 __ClearPageReserved(p
);
1076 set_page_count(p
, 0);
1078 __ClearPageReserved(p
);
1079 set_page_count(p
, 0);
1081 page_zone(page
)->managed_pages
+= nr_pages
;
1082 set_page_refcounted(page
);
1083 __free_pages(page
, order
);
1086 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1087 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1089 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1091 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1093 static DEFINE_SPINLOCK(early_pfn_lock
);
1096 spin_lock(&early_pfn_lock
);
1097 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1099 nid
= first_online_node
;
1100 spin_unlock(&early_pfn_lock
);
1106 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1107 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1108 struct mminit_pfnnid_cache
*state
)
1112 nid
= __early_pfn_to_nid(pfn
, state
);
1113 if (nid
>= 0 && nid
!= node
)
1118 /* Only safe to use early in boot when initialisation is single-threaded */
1119 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1121 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1126 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1130 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1131 struct mminit_pfnnid_cache
*state
)
1138 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1141 if (early_page_uninitialised(pfn
))
1143 return __free_pages_boot_core(page
, pfn
, order
);
1146 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1147 static void __init
deferred_free_range(struct page
*page
,
1148 unsigned long pfn
, int nr_pages
)
1155 /* Free a large naturally-aligned chunk if possible */
1156 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1157 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1158 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1159 __free_pages_boot_core(page
, pfn
, MAX_ORDER
-1);
1163 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++)
1164 __free_pages_boot_core(page
, pfn
, 0);
1167 /* Completion tracking for deferred_init_memmap() threads */
1168 static atomic_t pgdat_init_n_undone __initdata
;
1169 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1171 static inline void __init
pgdat_init_report_one_done(void)
1173 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1174 complete(&pgdat_init_all_done_comp
);
1177 /* Initialise remaining memory on a node */
1178 static int __init
deferred_init_memmap(void *data
)
1180 pg_data_t
*pgdat
= data
;
1181 int nid
= pgdat
->node_id
;
1182 struct mminit_pfnnid_cache nid_init_state
= { };
1183 unsigned long start
= jiffies
;
1184 unsigned long nr_pages
= 0;
1185 unsigned long walk_start
, walk_end
;
1188 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1189 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1191 if (first_init_pfn
== ULONG_MAX
) {
1192 pgdat_init_report_one_done();
1196 /* Bind memory initialisation thread to a local node if possible */
1197 if (!cpumask_empty(cpumask
))
1198 set_cpus_allowed_ptr(current
, cpumask
);
1200 /* Sanity check boundaries */
1201 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1202 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1203 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1205 /* Only the highest zone is deferred so find it */
1206 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1207 zone
= pgdat
->node_zones
+ zid
;
1208 if (first_init_pfn
< zone_end_pfn(zone
))
1212 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1213 unsigned long pfn
, end_pfn
;
1214 struct page
*page
= NULL
;
1215 struct page
*free_base_page
= NULL
;
1216 unsigned long free_base_pfn
= 0;
1219 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1220 pfn
= first_init_pfn
;
1221 if (pfn
< walk_start
)
1223 if (pfn
< zone
->zone_start_pfn
)
1224 pfn
= zone
->zone_start_pfn
;
1226 for (; pfn
< end_pfn
; pfn
++) {
1227 if (!pfn_valid_within(pfn
))
1231 * Ensure pfn_valid is checked every
1232 * MAX_ORDER_NR_PAGES for memory holes
1234 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1235 if (!pfn_valid(pfn
)) {
1241 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1246 /* Minimise pfn page lookups and scheduler checks */
1247 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1250 nr_pages
+= nr_to_free
;
1251 deferred_free_range(free_base_page
,
1252 free_base_pfn
, nr_to_free
);
1253 free_base_page
= NULL
;
1254 free_base_pfn
= nr_to_free
= 0;
1256 page
= pfn_to_page(pfn
);
1261 VM_BUG_ON(page_zone(page
) != zone
);
1265 __init_single_page(page
, pfn
, zid
, nid
);
1266 if (!free_base_page
) {
1267 free_base_page
= page
;
1268 free_base_pfn
= pfn
;
1273 /* Where possible, batch up pages for a single free */
1276 /* Free the current block of pages to allocator */
1277 nr_pages
+= nr_to_free
;
1278 deferred_free_range(free_base_page
, free_base_pfn
,
1280 free_base_page
= NULL
;
1281 free_base_pfn
= nr_to_free
= 0;
1284 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1287 /* Sanity check that the next zone really is unpopulated */
1288 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1290 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1291 jiffies_to_msecs(jiffies
- start
));
1293 pgdat_init_report_one_done();
1297 void __init
page_alloc_init_late(void)
1301 /* There will be num_node_state(N_MEMORY) threads */
1302 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1303 for_each_node_state(nid
, N_MEMORY
) {
1304 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1307 /* Block until all are initialised */
1308 wait_for_completion(&pgdat_init_all_done_comp
);
1310 /* Reinit limits that are based on free pages after the kernel is up */
1311 files_maxfiles_init();
1313 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1316 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1317 void __init
init_cma_reserved_pageblock(struct page
*page
)
1319 unsigned i
= pageblock_nr_pages
;
1320 struct page
*p
= page
;
1323 __ClearPageReserved(p
);
1324 set_page_count(p
, 0);
1327 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1329 if (pageblock_order
>= MAX_ORDER
) {
1330 i
= pageblock_nr_pages
;
1333 set_page_refcounted(p
);
1334 __free_pages(p
, MAX_ORDER
- 1);
1335 p
+= MAX_ORDER_NR_PAGES
;
1336 } while (i
-= MAX_ORDER_NR_PAGES
);
1338 set_page_refcounted(page
);
1339 __free_pages(page
, pageblock_order
);
1342 adjust_managed_page_count(page
, pageblock_nr_pages
);
1347 * The order of subdivision here is critical for the IO subsystem.
1348 * Please do not alter this order without good reasons and regression
1349 * testing. Specifically, as large blocks of memory are subdivided,
1350 * the order in which smaller blocks are delivered depends on the order
1351 * they're subdivided in this function. This is the primary factor
1352 * influencing the order in which pages are delivered to the IO
1353 * subsystem according to empirical testing, and this is also justified
1354 * by considering the behavior of a buddy system containing a single
1355 * large block of memory acted on by a series of small allocations.
1356 * This behavior is a critical factor in sglist merging's success.
1360 static inline void expand(struct zone
*zone
, struct page
*page
,
1361 int low
, int high
, struct free_area
*area
,
1364 unsigned long size
= 1 << high
;
1366 while (high
> low
) {
1370 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1372 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1373 debug_guardpage_enabled() &&
1374 high
< debug_guardpage_minorder()) {
1376 * Mark as guard pages (or page), that will allow to
1377 * merge back to allocator when buddy will be freed.
1378 * Corresponding page table entries will not be touched,
1379 * pages will stay not present in virtual address space
1381 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1384 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1386 set_page_order(&page
[size
], high
);
1391 * This page is about to be returned from the page allocator
1393 static inline int check_new_page(struct page
*page
)
1395 const char *bad_reason
= NULL
;
1396 unsigned long bad_flags
= 0;
1398 if (unlikely(page_mapcount(page
)))
1399 bad_reason
= "nonzero mapcount";
1400 if (unlikely(page
->mapping
!= NULL
))
1401 bad_reason
= "non-NULL mapping";
1402 if (unlikely(atomic_read(&page
->_count
) != 0))
1403 bad_reason
= "nonzero _count";
1404 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1405 bad_reason
= "HWPoisoned (hardware-corrupted)";
1406 bad_flags
= __PG_HWPOISON
;
1408 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1409 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1410 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1413 if (unlikely(page
->mem_cgroup
))
1414 bad_reason
= "page still charged to cgroup";
1416 if (unlikely(bad_reason
)) {
1417 bad_page(page
, bad_reason
, bad_flags
);
1423 static int prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1428 for (i
= 0; i
< (1 << order
); i
++) {
1429 struct page
*p
= page
+ i
;
1430 if (unlikely(check_new_page(p
)))
1434 set_page_private(page
, 0);
1435 set_page_refcounted(page
);
1437 arch_alloc_page(page
, order
);
1438 kernel_map_pages(page
, 1 << order
, 1);
1439 kasan_alloc_pages(page
, order
);
1441 if (gfp_flags
& __GFP_ZERO
)
1442 for (i
= 0; i
< (1 << order
); i
++)
1443 clear_highpage(page
+ i
);
1445 if (order
&& (gfp_flags
& __GFP_COMP
))
1446 prep_compound_page(page
, order
);
1448 set_page_owner(page
, order
, gfp_flags
);
1451 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1452 * allocate the page. The expectation is that the caller is taking
1453 * steps that will free more memory. The caller should avoid the page
1454 * being used for !PFMEMALLOC purposes.
1456 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1457 set_page_pfmemalloc(page
);
1459 clear_page_pfmemalloc(page
);
1465 * Go through the free lists for the given migratetype and remove
1466 * the smallest available page from the freelists
1469 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1472 unsigned int current_order
;
1473 struct free_area
*area
;
1476 /* Find a page of the appropriate size in the preferred list */
1477 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1478 area
= &(zone
->free_area
[current_order
]);
1479 if (list_empty(&area
->free_list
[migratetype
]))
1482 page
= list_entry(area
->free_list
[migratetype
].next
,
1484 list_del(&page
->lru
);
1485 rmv_page_order(page
);
1487 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1488 set_pcppage_migratetype(page
, migratetype
);
1497 * This array describes the order lists are fallen back to when
1498 * the free lists for the desirable migrate type are depleted
1500 static int fallbacks
[MIGRATE_TYPES
][4] = {
1501 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1502 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1503 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1505 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1507 #ifdef CONFIG_MEMORY_ISOLATION
1508 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1513 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1516 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1519 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1520 unsigned int order
) { return NULL
; }
1524 * Move the free pages in a range to the free lists of the requested type.
1525 * Note that start_page and end_pages are not aligned on a pageblock
1526 * boundary. If alignment is required, use move_freepages_block()
1528 int move_freepages(struct zone
*zone
,
1529 struct page
*start_page
, struct page
*end_page
,
1534 int pages_moved
= 0;
1536 #ifndef CONFIG_HOLES_IN_ZONE
1538 * page_zone is not safe to call in this context when
1539 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1540 * anyway as we check zone boundaries in move_freepages_block().
1541 * Remove at a later date when no bug reports exist related to
1542 * grouping pages by mobility
1544 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1547 for (page
= start_page
; page
<= end_page
;) {
1548 if (!pfn_valid_within(page_to_pfn(page
))) {
1553 /* Make sure we are not inadvertently changing nodes */
1554 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1556 if (!PageBuddy(page
)) {
1561 order
= page_order(page
);
1562 list_move(&page
->lru
,
1563 &zone
->free_area
[order
].free_list
[migratetype
]);
1565 pages_moved
+= 1 << order
;
1571 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1574 unsigned long start_pfn
, end_pfn
;
1575 struct page
*start_page
, *end_page
;
1577 start_pfn
= page_to_pfn(page
);
1578 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1579 start_page
= pfn_to_page(start_pfn
);
1580 end_page
= start_page
+ pageblock_nr_pages
- 1;
1581 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1583 /* Do not cross zone boundaries */
1584 if (!zone_spans_pfn(zone
, start_pfn
))
1586 if (!zone_spans_pfn(zone
, end_pfn
))
1589 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1592 static void change_pageblock_range(struct page
*pageblock_page
,
1593 int start_order
, int migratetype
)
1595 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1597 while (nr_pageblocks
--) {
1598 set_pageblock_migratetype(pageblock_page
, migratetype
);
1599 pageblock_page
+= pageblock_nr_pages
;
1604 * When we are falling back to another migratetype during allocation, try to
1605 * steal extra free pages from the same pageblocks to satisfy further
1606 * allocations, instead of polluting multiple pageblocks.
1608 * If we are stealing a relatively large buddy page, it is likely there will
1609 * be more free pages in the pageblock, so try to steal them all. For
1610 * reclaimable and unmovable allocations, we steal regardless of page size,
1611 * as fragmentation caused by those allocations polluting movable pageblocks
1612 * is worse than movable allocations stealing from unmovable and reclaimable
1615 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1618 * Leaving this order check is intended, although there is
1619 * relaxed order check in next check. The reason is that
1620 * we can actually steal whole pageblock if this condition met,
1621 * but, below check doesn't guarantee it and that is just heuristic
1622 * so could be changed anytime.
1624 if (order
>= pageblock_order
)
1627 if (order
>= pageblock_order
/ 2 ||
1628 start_mt
== MIGRATE_RECLAIMABLE
||
1629 start_mt
== MIGRATE_UNMOVABLE
||
1630 start_mt
== MIGRATE_MOVABLE
||
1631 page_group_by_mobility_disabled
)
1638 * This function implements actual steal behaviour. If order is large enough,
1639 * we can steal whole pageblock. If not, we first move freepages in this
1640 * pageblock and check whether half of pages are moved or not. If half of
1641 * pages are moved, we can change migratetype of pageblock and permanently
1642 * use it's pages as requested migratetype in the future.
1644 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1647 unsigned int current_order
= page_order(page
);
1650 /* Take ownership for orders >= pageblock_order */
1651 if (current_order
>= pageblock_order
) {
1652 change_pageblock_range(page
, current_order
, start_type
);
1656 pages
= move_freepages_block(zone
, page
, start_type
);
1658 /* Claim the whole block if over half of it is free */
1659 if (pages
>= (1 << (pageblock_order
-1)) ||
1660 start_type
== MIGRATE_MOVABLE
||
1661 page_group_by_mobility_disabled
)
1662 set_pageblock_migratetype(page
, start_type
);
1666 * Check whether there is a suitable fallback freepage with requested order.
1667 * If only_stealable is true, this function returns fallback_mt only if
1668 * we can steal other freepages all together. This would help to reduce
1669 * fragmentation due to mixed migratetype pages in one pageblock.
1671 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1672 int migratetype
, bool only_stealable
, bool *can_steal
)
1677 if (area
->nr_free
== 0)
1682 fallback_mt
= fallbacks
[migratetype
][i
];
1683 if (fallback_mt
== MIGRATE_TYPES
)
1686 if (list_empty(&area
->free_list
[fallback_mt
]))
1689 if (can_steal_fallback(order
, migratetype
))
1692 if (!only_stealable
)
1703 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1704 * there are no empty page blocks that contain a page with a suitable order
1706 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
1707 unsigned int alloc_order
)
1710 unsigned long max_managed
, flags
;
1713 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1714 * Check is race-prone but harmless.
1716 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
1717 if (zone
->nr_reserved_highatomic
>= max_managed
)
1720 spin_lock_irqsave(&zone
->lock
, flags
);
1722 /* Recheck the nr_reserved_highatomic limit under the lock */
1723 if (zone
->nr_reserved_highatomic
>= max_managed
)
1727 mt
= get_pageblock_migratetype(page
);
1728 if (mt
!= MIGRATE_HIGHATOMIC
&&
1729 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
1730 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
1731 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
1732 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
1736 spin_unlock_irqrestore(&zone
->lock
, flags
);
1740 * Used when an allocation is about to fail under memory pressure. This
1741 * potentially hurts the reliability of high-order allocations when under
1742 * intense memory pressure but failed atomic allocations should be easier
1743 * to recover from than an OOM.
1745 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
1747 struct zonelist
*zonelist
= ac
->zonelist
;
1748 unsigned long flags
;
1754 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
1756 /* Preserve at least one pageblock */
1757 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
1760 spin_lock_irqsave(&zone
->lock
, flags
);
1761 for (order
= 0; order
< MAX_ORDER
; order
++) {
1762 struct free_area
*area
= &(zone
->free_area
[order
]);
1764 if (list_empty(&area
->free_list
[MIGRATE_HIGHATOMIC
]))
1767 page
= list_entry(area
->free_list
[MIGRATE_HIGHATOMIC
].next
,
1771 * In page freeing path, migratetype change is racy so
1772 * we can counter several free pages in a pageblock
1773 * in this loop althoug we changed the pageblock type
1774 * from highatomic to ac->migratetype. So we should
1775 * adjust the count once.
1777 if (get_pageblock_migratetype(page
) ==
1778 MIGRATE_HIGHATOMIC
) {
1780 * It should never happen but changes to
1781 * locking could inadvertently allow a per-cpu
1782 * drain to add pages to MIGRATE_HIGHATOMIC
1783 * while unreserving so be safe and watch for
1786 zone
->nr_reserved_highatomic
-= min(
1788 zone
->nr_reserved_highatomic
);
1792 * Convert to ac->migratetype and avoid the normal
1793 * pageblock stealing heuristics. Minimally, the caller
1794 * is doing the work and needs the pages. More
1795 * importantly, if the block was always converted to
1796 * MIGRATE_UNMOVABLE or another type then the number
1797 * of pageblocks that cannot be completely freed
1800 set_pageblock_migratetype(page
, ac
->migratetype
);
1801 move_freepages_block(zone
, page
, ac
->migratetype
);
1802 spin_unlock_irqrestore(&zone
->lock
, flags
);
1805 spin_unlock_irqrestore(&zone
->lock
, flags
);
1809 /* Remove an element from the buddy allocator from the fallback list */
1810 static inline struct page
*
1811 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
1813 struct free_area
*area
;
1814 unsigned int current_order
;
1819 /* Find the largest possible block of pages in the other list */
1820 for (current_order
= MAX_ORDER
-1;
1821 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
1823 area
= &(zone
->free_area
[current_order
]);
1824 fallback_mt
= find_suitable_fallback(area
, current_order
,
1825 start_migratetype
, false, &can_steal
);
1826 if (fallback_mt
== -1)
1829 page
= list_entry(area
->free_list
[fallback_mt
].next
,
1832 get_pageblock_migratetype(page
) != MIGRATE_HIGHATOMIC
)
1833 steal_suitable_fallback(zone
, page
, start_migratetype
);
1835 /* Remove the page from the freelists */
1837 list_del(&page
->lru
);
1838 rmv_page_order(page
);
1840 expand(zone
, page
, order
, current_order
, area
,
1843 * The pcppage_migratetype may differ from pageblock's
1844 * migratetype depending on the decisions in
1845 * find_suitable_fallback(). This is OK as long as it does not
1846 * differ for MIGRATE_CMA pageblocks. Those can be used as
1847 * fallback only via special __rmqueue_cma_fallback() function
1849 set_pcppage_migratetype(page
, start_migratetype
);
1851 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1852 start_migratetype
, fallback_mt
);
1861 * Do the hard work of removing an element from the buddy allocator.
1862 * Call me with the zone->lock already held.
1864 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1865 int migratetype
, gfp_t gfp_flags
)
1867 struct page
*page
= NULL
;
1869 if ((migratetype
== MIGRATE_MOVABLE
) && (gfp_flags
& __GFP_CMA
))
1870 page
= __rmqueue_cma_fallback(zone
, order
);
1873 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1875 if (unlikely(!page
))
1876 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1878 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1883 * Obtain a specified number of elements from the buddy allocator, all under
1884 * a single hold of the lock, for efficiency. Add them to the supplied list.
1885 * Returns the number of new pages which were placed at *list.
1887 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1888 unsigned long count
, struct list_head
*list
,
1889 int migratetype
, gfp_t gfp_flags
)
1893 spin_lock(&zone
->lock
);
1894 for (i
= 0; i
< count
; ++i
) {
1895 struct page
*page
= __rmqueue(zone
, order
, migratetype
, gfp_flags
);
1896 if (unlikely(page
== NULL
))
1900 * Split buddy pages returned by expand() are received here
1901 * in physical page order. The page is added to the callers and
1902 * list and the list head then moves forward. From the callers
1903 * perspective, the linked list is ordered by page number in
1904 * some conditions. This is useful for IO devices that can
1905 * merge IO requests if the physical pages are ordered
1908 if (likely(!(gfp_flags
& __GFP_COLD
)))
1909 list_add(&page
->lru
, list
);
1911 list_add_tail(&page
->lru
, list
);
1913 if (is_migrate_cma(get_pcppage_migratetype(page
)))
1914 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1917 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1918 spin_unlock(&zone
->lock
);
1924 * Called from the vmstat counter updater to drain pagesets of this
1925 * currently executing processor on remote nodes after they have
1928 * Note that this function must be called with the thread pinned to
1929 * a single processor.
1931 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1933 unsigned long flags
;
1934 int to_drain
, batch
;
1936 local_irq_save(flags
);
1937 batch
= READ_ONCE(pcp
->batch
);
1938 to_drain
= min(pcp
->count
, batch
);
1940 free_pcppages_bulk(zone
, to_drain
, pcp
);
1941 pcp
->count
-= to_drain
;
1943 local_irq_restore(flags
);
1948 * Drain pcplists of the indicated processor and zone.
1950 * The processor must either be the current processor and the
1951 * thread pinned to the current processor or a processor that
1954 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
1956 unsigned long flags
;
1957 struct per_cpu_pageset
*pset
;
1958 struct per_cpu_pages
*pcp
;
1960 local_irq_save(flags
);
1961 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1965 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1968 local_irq_restore(flags
);
1972 * Drain pcplists of all zones on the indicated processor.
1974 * The processor must either be the current processor and the
1975 * thread pinned to the current processor or a processor that
1978 static void drain_pages(unsigned int cpu
)
1982 for_each_populated_zone(zone
) {
1983 drain_pages_zone(cpu
, zone
);
1988 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1990 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1991 * the single zone's pages.
1993 void drain_local_pages(struct zone
*zone
)
1995 int cpu
= smp_processor_id();
1998 drain_pages_zone(cpu
, zone
);
2004 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2006 * When zone parameter is non-NULL, spill just the single zone's pages.
2008 * Note that this code is protected against sending an IPI to an offline
2009 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2010 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2011 * nothing keeps CPUs from showing up after we populated the cpumask and
2012 * before the call to on_each_cpu_mask().
2014 void drain_all_pages(struct zone
*zone
)
2019 * Allocate in the BSS so we wont require allocation in
2020 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2022 static cpumask_t cpus_with_pcps
;
2025 * We don't care about racing with CPU hotplug event
2026 * as offline notification will cause the notified
2027 * cpu to drain that CPU pcps and on_each_cpu_mask
2028 * disables preemption as part of its processing
2030 for_each_online_cpu(cpu
) {
2031 struct per_cpu_pageset
*pcp
;
2033 bool has_pcps
= false;
2036 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2040 for_each_populated_zone(z
) {
2041 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2042 if (pcp
->pcp
.count
) {
2050 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2052 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2054 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
2058 #ifdef CONFIG_HIBERNATION
2060 void mark_free_pages(struct zone
*zone
)
2062 unsigned long pfn
, max_zone_pfn
;
2063 unsigned long flags
;
2064 unsigned int order
, t
;
2065 struct list_head
*curr
;
2067 if (zone_is_empty(zone
))
2070 spin_lock_irqsave(&zone
->lock
, flags
);
2072 max_zone_pfn
= zone_end_pfn(zone
);
2073 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2074 if (pfn_valid(pfn
)) {
2075 struct page
*page
= pfn_to_page(pfn
);
2077 if (!swsusp_page_is_forbidden(page
))
2078 swsusp_unset_page_free(page
);
2081 for_each_migratetype_order(order
, t
) {
2082 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
2085 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
2086 for (i
= 0; i
< (1UL << order
); i
++)
2087 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2090 spin_unlock_irqrestore(&zone
->lock
, flags
);
2092 #endif /* CONFIG_PM */
2095 * Free a 0-order page
2096 * cold == true ? free a cold page : free a hot page
2098 void free_hot_cold_page(struct page
*page
, bool cold
)
2100 struct zone
*zone
= page_zone(page
);
2101 struct per_cpu_pages
*pcp
;
2102 unsigned long flags
;
2103 unsigned long pfn
= page_to_pfn(page
);
2106 if (!free_pages_prepare(page
, 0))
2109 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2110 set_pcppage_migratetype(page
, migratetype
);
2111 local_irq_save(flags
);
2112 __count_vm_event(PGFREE
);
2115 * We only track unmovable, reclaimable and movable on pcp lists.
2116 * Free ISOLATE pages back to the allocator because they are being
2117 * offlined but treat RESERVE as movable pages so we can get those
2118 * areas back if necessary. Otherwise, we may have to free
2119 * excessively into the page allocator
2121 if (migratetype
>= MIGRATE_PCPTYPES
) {
2122 if (unlikely(is_migrate_isolate(migratetype
))) {
2123 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2126 migratetype
= MIGRATE_MOVABLE
;
2129 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2131 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2133 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2135 if (pcp
->count
>= pcp
->high
) {
2136 unsigned long batch
= READ_ONCE(pcp
->batch
);
2137 free_pcppages_bulk(zone
, batch
, pcp
);
2138 pcp
->count
-= batch
;
2142 local_irq_restore(flags
);
2146 * Free a list of 0-order pages
2148 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2150 struct page
*page
, *next
;
2152 list_for_each_entry_safe(page
, next
, list
, lru
) {
2153 trace_mm_page_free_batched(page
, cold
);
2154 free_hot_cold_page(page
, cold
);
2159 * split_page takes a non-compound higher-order page, and splits it into
2160 * n (1<<order) sub-pages: page[0..n]
2161 * Each sub-page must be freed individually.
2163 * Note: this is probably too low level an operation for use in drivers.
2164 * Please consult with lkml before using this in your driver.
2166 void split_page(struct page
*page
, unsigned int order
)
2171 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2172 VM_BUG_ON_PAGE(!page_count(page
), page
);
2174 #ifdef CONFIG_KMEMCHECK
2176 * Split shadow pages too, because free(page[0]) would
2177 * otherwise free the whole shadow.
2179 if (kmemcheck_page_is_tracked(page
))
2180 split_page(virt_to_page(page
[0].shadow
), order
);
2183 gfp_mask
= get_page_owner_gfp(page
);
2184 set_page_owner(page
, 0, gfp_mask
);
2185 for (i
= 1; i
< (1 << order
); i
++) {
2186 set_page_refcounted(page
+ i
);
2187 set_page_owner(page
+ i
, 0, gfp_mask
);
2190 EXPORT_SYMBOL_GPL(split_page
);
2192 int __isolate_free_page(struct page
*page
, unsigned int order
)
2194 unsigned long watermark
;
2198 BUG_ON(!PageBuddy(page
));
2200 zone
= page_zone(page
);
2201 mt
= get_pageblock_migratetype(page
);
2203 if (!is_migrate_isolate(mt
)) {
2204 /* Obey watermarks as if the page was being allocated */
2205 watermark
= low_wmark_pages(zone
) + (1 << order
);
2206 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2209 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2212 /* Remove page from free list */
2213 list_del(&page
->lru
);
2214 zone
->free_area
[order
].nr_free
--;
2215 rmv_page_order(page
);
2217 set_page_owner(page
, order
, __GFP_MOVABLE
);
2219 /* Set the pageblock if the isolated page is at least a pageblock */
2220 if (order
>= pageblock_order
- 1) {
2221 struct page
*endpage
= page
+ (1 << order
) - 1;
2222 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2223 int mt
= get_pageblock_migratetype(page
);
2224 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
)
2225 && mt
!= MIGRATE_HIGHATOMIC
)
2226 set_pageblock_migratetype(page
,
2232 return 1UL << order
;
2236 * Similar to split_page except the page is already free. As this is only
2237 * being used for migration, the migratetype of the block also changes.
2238 * As this is called with interrupts disabled, the caller is responsible
2239 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2242 * Note: this is probably too low level an operation for use in drivers.
2243 * Please consult with lkml before using this in your driver.
2245 int split_free_page(struct page
*page
)
2250 order
= page_order(page
);
2252 nr_pages
= __isolate_free_page(page
, order
);
2256 /* Split into individual pages */
2257 set_page_refcounted(page
);
2258 split_page(page
, order
);
2263 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2266 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2267 struct zone
*zone
, unsigned int order
,
2268 gfp_t gfp_flags
, int alloc_flags
, int migratetype
)
2270 unsigned long flags
;
2271 struct page
*page
= NULL
;
2273 if (likely(order
== 0)) {
2274 struct per_cpu_pages
*pcp
;
2275 struct list_head
*list
;
2277 local_irq_save(flags
);
2278 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2279 list
= &pcp
->lists
[migratetype
];
2280 if (list_empty(list
)) {
2281 pcp
->count
+= rmqueue_bulk(zone
, 0,
2283 migratetype
, gfp_flags
);
2284 if (unlikely(list_empty(list
)))
2288 if ((gfp_flags
& __GFP_COLD
) != 0)
2289 page
= list_entry(list
->prev
, struct page
, lru
);
2291 page
= list_entry(list
->next
, struct page
, lru
);
2294 * If the head or the tail page in the pcp list is CMA page and
2295 * the gfp flags does not have __GFP_CMA, try allocation from
2296 * free list. If a page in the pcp list is CMA page, all pages
2297 * in the pcp list is probabily CMA pages because rmqueue_bulk()
2298 * fills the list from the free list of the same migratetype.
2300 if (!(gfp_flags
& __GFP_CMA
) &&
2301 is_migrate_cma(get_pcppage_migratetype(page
))) {
2303 local_irq_restore(flags
);
2305 list_del(&page
->lru
);
2311 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
2313 * __GFP_NOFAIL is not to be used in new code.
2315 * All __GFP_NOFAIL callers should be fixed so that they
2316 * properly detect and handle allocation failures.
2318 * We most definitely don't want callers attempting to
2319 * allocate greater than order-1 page units with
2322 WARN_ON_ONCE(order
> 1);
2324 spin_lock_irqsave(&zone
->lock
, flags
);
2326 if (alloc_flags
& ALLOC_HARDER
) {
2327 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2329 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2332 page
= __rmqueue(zone
, order
, migratetype
, gfp_flags
);
2333 spin_unlock(&zone
->lock
);
2336 __mod_zone_freepage_state(zone
, -(1 << order
),
2337 get_pcppage_migratetype(page
));
2340 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2341 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2342 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2343 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2345 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2346 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2347 local_irq_restore(flags
);
2349 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2353 local_irq_restore(flags
);
2357 #ifdef CONFIG_FAIL_PAGE_ALLOC
2360 struct fault_attr attr
;
2362 bool ignore_gfp_highmem
;
2363 bool ignore_gfp_reclaim
;
2365 } fail_page_alloc
= {
2366 .attr
= FAULT_ATTR_INITIALIZER
,
2367 .ignore_gfp_reclaim
= true,
2368 .ignore_gfp_highmem
= true,
2372 static int __init
setup_fail_page_alloc(char *str
)
2374 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2376 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2378 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2380 if (order
< fail_page_alloc
.min_order
)
2382 if (gfp_mask
& __GFP_NOFAIL
)
2384 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2386 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2387 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2390 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2393 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2395 static int __init
fail_page_alloc_debugfs(void)
2397 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2400 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2401 &fail_page_alloc
.attr
);
2403 return PTR_ERR(dir
);
2405 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2406 &fail_page_alloc
.ignore_gfp_reclaim
))
2408 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2409 &fail_page_alloc
.ignore_gfp_highmem
))
2411 if (!debugfs_create_u32("min-order", mode
, dir
,
2412 &fail_page_alloc
.min_order
))
2417 debugfs_remove_recursive(dir
);
2422 late_initcall(fail_page_alloc_debugfs
);
2424 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2426 #else /* CONFIG_FAIL_PAGE_ALLOC */
2428 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2433 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2436 * Return true if free base pages are above 'mark'. For high-order checks it
2437 * will return true of the order-0 watermark is reached and there is at least
2438 * one free page of a suitable size. Checking now avoids taking the zone lock
2439 * to check in the allocation paths if no pages are free.
2441 static bool __zone_watermark_ok(struct zone
*z
, unsigned int order
,
2442 unsigned long mark
, int classzone_idx
, int alloc_flags
,
2447 const int alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2449 /* free_pages may go negative - that's OK */
2450 free_pages
-= (1 << order
) - 1;
2452 if (alloc_flags
& ALLOC_HIGH
)
2456 * If the caller does not have rights to ALLOC_HARDER then subtract
2457 * the high-atomic reserves. This will over-estimate the size of the
2458 * atomic reserve but it avoids a search.
2460 if (likely(!alloc_harder
))
2461 free_pages
-= z
->nr_reserved_highatomic
;
2466 /* If allocation can't use CMA areas don't use free CMA pages */
2467 if (!(alloc_flags
& ALLOC_CMA
))
2468 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2472 * Check watermarks for an order-0 allocation request. If these
2473 * are not met, then a high-order request also cannot go ahead
2474 * even if a suitable page happened to be free.
2476 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2479 /* If this is an order-0 request then the watermark is fine */
2483 /* For a high-order request, check at least one suitable page is free */
2484 for (o
= order
; o
< MAX_ORDER
; o
++) {
2485 struct free_area
*area
= &z
->free_area
[o
];
2494 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2495 if (!list_empty(&area
->free_list
[mt
]))
2500 if ((alloc_flags
& ALLOC_CMA
) &&
2501 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2509 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2510 int classzone_idx
, int alloc_flags
)
2512 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2513 zone_page_state(z
, NR_FREE_PAGES
));
2516 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2517 unsigned long mark
, int classzone_idx
)
2519 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2521 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2522 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2524 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2529 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2531 return local_zone
->node
== zone
->node
;
2534 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2536 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <=
2539 #else /* CONFIG_NUMA */
2540 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2545 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2549 #endif /* CONFIG_NUMA */
2551 static void reset_alloc_batches(struct zone
*preferred_zone
)
2553 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2556 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2557 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2558 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2559 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2560 } while (zone
++ != preferred_zone
);
2564 * get_page_from_freelist goes through the zonelist trying to allocate
2567 static struct page
*
2568 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2569 const struct alloc_context
*ac
)
2571 struct zonelist
*zonelist
= ac
->zonelist
;
2573 struct page
*page
= NULL
;
2575 int nr_fair_skipped
= 0;
2576 bool zonelist_rescan
;
2579 zonelist_rescan
= false;
2582 * Scan zonelist, looking for a zone with enough free.
2583 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2585 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2589 if (cpusets_enabled() &&
2590 (alloc_flags
& ALLOC_CPUSET
) &&
2591 !cpuset_zone_allowed(zone
, gfp_mask
))
2594 * Distribute pages in proportion to the individual
2595 * zone size to ensure fair page aging. The zone a
2596 * page was allocated in should have no effect on the
2597 * time the page has in memory before being reclaimed.
2599 if (alloc_flags
& ALLOC_FAIR
) {
2600 if (!zone_local(ac
->preferred_zone
, zone
))
2602 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2608 * When allocating a page cache page for writing, we
2609 * want to get it from a zone that is within its dirty
2610 * limit, such that no single zone holds more than its
2611 * proportional share of globally allowed dirty pages.
2612 * The dirty limits take into account the zone's
2613 * lowmem reserves and high watermark so that kswapd
2614 * should be able to balance it without having to
2615 * write pages from its LRU list.
2617 * This may look like it could increase pressure on
2618 * lower zones by failing allocations in higher zones
2619 * before they are full. But the pages that do spill
2620 * over are limited as the lower zones are protected
2621 * by this very same mechanism. It should not become
2622 * a practical burden to them.
2624 * XXX: For now, allow allocations to potentially
2625 * exceed the per-zone dirty limit in the slowpath
2626 * (spread_dirty_pages unset) before going into reclaim,
2627 * which is important when on a NUMA setup the allowed
2628 * zones are together not big enough to reach the
2629 * global limit. The proper fix for these situations
2630 * will require awareness of zones in the
2631 * dirty-throttling and the flusher threads.
2633 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2636 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2637 if (!zone_watermark_ok(zone
, order
, mark
,
2638 ac
->classzone_idx
, alloc_flags
)) {
2641 /* Checked here to keep the fast path fast */
2642 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2643 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2646 if (zone_reclaim_mode
== 0 ||
2647 !zone_allows_reclaim(ac
->preferred_zone
, zone
))
2650 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2652 case ZONE_RECLAIM_NOSCAN
:
2655 case ZONE_RECLAIM_FULL
:
2656 /* scanned but unreclaimable */
2659 /* did we reclaim enough */
2660 if (zone_watermark_ok(zone
, order
, mark
,
2661 ac
->classzone_idx
, alloc_flags
))
2669 page
= buffered_rmqueue(ac
->preferred_zone
, zone
, order
,
2670 gfp_mask
, alloc_flags
, ac
->migratetype
);
2672 if (prep_new_page(page
, order
, gfp_mask
, alloc_flags
))
2676 * If this is a high-order atomic allocation then check
2677 * if the pageblock should be reserved for the future
2679 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
2680 reserve_highatomic_pageblock(page
, zone
, order
);
2687 * The first pass makes sure allocations are spread fairly within the
2688 * local node. However, the local node might have free pages left
2689 * after the fairness batches are exhausted, and remote zones haven't
2690 * even been considered yet. Try once more without fairness, and
2691 * include remote zones now, before entering the slowpath and waking
2692 * kswapd: prefer spilling to a remote zone over swapping locally.
2694 if (alloc_flags
& ALLOC_FAIR
) {
2695 alloc_flags
&= ~ALLOC_FAIR
;
2696 if (nr_fair_skipped
) {
2697 zonelist_rescan
= true;
2698 reset_alloc_batches(ac
->preferred_zone
);
2700 if (nr_online_nodes
> 1)
2701 zonelist_rescan
= true;
2704 if (zonelist_rescan
)
2711 * Large machines with many possible nodes should not always dump per-node
2712 * meminfo in irq context.
2714 static inline bool should_suppress_show_mem(void)
2719 ret
= in_interrupt();
2724 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2725 DEFAULT_RATELIMIT_INTERVAL
,
2726 DEFAULT_RATELIMIT_BURST
);
2728 void warn_alloc_failed(gfp_t gfp_mask
, unsigned int order
, const char *fmt
, ...)
2730 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2732 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2733 debug_guardpage_minorder() > 0)
2737 * This documents exceptions given to allocations in certain
2738 * contexts that are allowed to allocate outside current's set
2741 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2742 if (test_thread_flag(TIF_MEMDIE
) ||
2743 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2744 filter
&= ~SHOW_MEM_FILTER_NODES
;
2745 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
2746 filter
&= ~SHOW_MEM_FILTER_NODES
;
2749 struct va_format vaf
;
2752 va_start(args
, fmt
);
2757 pr_warn("%pV", &vaf
);
2762 pr_warn("%s: page allocation failure: order:%u, mode:0x%x\n",
2763 current
->comm
, order
, gfp_mask
);
2766 if (!should_suppress_show_mem())
2770 static inline struct page
*
2771 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2772 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
2774 struct oom_control oc
= {
2775 .zonelist
= ac
->zonelist
,
2776 .nodemask
= ac
->nodemask
,
2777 .gfp_mask
= gfp_mask
,
2782 *did_some_progress
= 0;
2785 * Acquire the oom lock. If that fails, somebody else is
2786 * making progress for us.
2788 if (!mutex_trylock(&oom_lock
)) {
2789 *did_some_progress
= 1;
2790 schedule_timeout_uninterruptible(1);
2795 * Go through the zonelist yet one more time, keep very high watermark
2796 * here, this is only to catch a parallel oom killing, we must fail if
2797 * we're still under heavy pressure.
2799 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
2800 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
2804 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2805 /* Coredumps can quickly deplete all memory reserves */
2806 if (current
->flags
& PF_DUMPCORE
)
2808 /* The OOM killer will not help higher order allocs */
2809 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2811 /* The OOM killer does not needlessly kill tasks for lowmem */
2812 if (ac
->high_zoneidx
< ZONE_NORMAL
)
2814 /* The OOM killer does not compensate for IO-less reclaim */
2815 if (!(gfp_mask
& __GFP_FS
)) {
2817 * XXX: Page reclaim didn't yield anything,
2818 * and the OOM killer can't be invoked, but
2819 * keep looping as per tradition.
2821 *did_some_progress
= 1;
2824 if (pm_suspended_storage())
2826 /* The OOM killer may not free memory on a specific node */
2827 if (gfp_mask
& __GFP_THISNODE
)
2830 /* Exhausted what can be done so it's blamo time */
2831 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
))
2832 *did_some_progress
= 1;
2834 mutex_unlock(&oom_lock
);
2838 #ifdef CONFIG_COMPACTION
2839 /* Try memory compaction for high-order allocations before reclaim */
2840 static struct page
*
2841 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2842 int alloc_flags
, const struct alloc_context
*ac
,
2843 enum migrate_mode mode
, int *contended_compaction
,
2844 bool *deferred_compaction
)
2846 unsigned long compact_result
;
2852 current
->flags
|= PF_MEMALLOC
;
2853 compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
2854 mode
, contended_compaction
);
2855 current
->flags
&= ~PF_MEMALLOC
;
2857 switch (compact_result
) {
2858 case COMPACT_DEFERRED
:
2859 *deferred_compaction
= true;
2861 case COMPACT_SKIPPED
:
2868 * At least in one zone compaction wasn't deferred or skipped, so let's
2869 * count a compaction stall
2871 count_vm_event(COMPACTSTALL
);
2873 page
= get_page_from_freelist(gfp_mask
, order
,
2874 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2877 struct zone
*zone
= page_zone(page
);
2879 zone
->compact_blockskip_flush
= false;
2880 compaction_defer_reset(zone
, order
, true);
2881 count_vm_event(COMPACTSUCCESS
);
2886 * It's bad if compaction run occurs and fails. The most likely reason
2887 * is that pages exist, but not enough to satisfy watermarks.
2889 count_vm_event(COMPACTFAIL
);
2896 static inline struct page
*
2897 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2898 int alloc_flags
, const struct alloc_context
*ac
,
2899 enum migrate_mode mode
, int *contended_compaction
,
2900 bool *deferred_compaction
)
2904 #endif /* CONFIG_COMPACTION */
2906 /* Perform direct synchronous page reclaim */
2908 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
2909 const struct alloc_context
*ac
)
2911 struct reclaim_state reclaim_state
;
2916 /* We now go into synchronous reclaim */
2917 cpuset_memory_pressure_bump();
2918 current
->flags
|= PF_MEMALLOC
;
2919 lockdep_set_current_reclaim_state(gfp_mask
);
2920 reclaim_state
.reclaimed_slab
= 0;
2921 current
->reclaim_state
= &reclaim_state
;
2923 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
2926 current
->reclaim_state
= NULL
;
2927 lockdep_clear_current_reclaim_state();
2928 current
->flags
&= ~PF_MEMALLOC
;
2935 /* The really slow allocator path where we enter direct reclaim */
2936 static inline struct page
*
2937 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2938 int alloc_flags
, const struct alloc_context
*ac
,
2939 unsigned long *did_some_progress
)
2941 struct page
*page
= NULL
;
2942 bool drained
= false;
2944 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
2945 if (unlikely(!(*did_some_progress
)))
2949 page
= get_page_from_freelist(gfp_mask
, order
,
2950 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2953 * If an allocation failed after direct reclaim, it could be because
2954 * pages are pinned on the per-cpu lists or in high alloc reserves.
2955 * Shrink them them and try again
2957 if (!page
&& !drained
) {
2958 unreserve_highatomic_pageblock(ac
);
2959 drain_all_pages(NULL
);
2968 * This is called in the allocator slow-path if the allocation request is of
2969 * sufficient urgency to ignore watermarks and take other desperate measures
2971 static inline struct page
*
2972 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2973 const struct alloc_context
*ac
)
2978 page
= get_page_from_freelist(gfp_mask
, order
,
2979 ALLOC_NO_WATERMARKS
, ac
);
2981 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2982 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
,
2984 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2989 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
2994 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
2995 ac
->high_zoneidx
, ac
->nodemask
)
2996 wakeup_kswapd(zone
, order
, zone_idx(ac
->preferred_zone
));
3000 gfp_to_alloc_flags(gfp_t gfp_mask
)
3002 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
3004 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3005 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
3008 * The caller may dip into page reserves a bit more if the caller
3009 * cannot run direct reclaim, or if the caller has realtime scheduling
3010 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3011 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3013 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
3015 if (gfp_mask
& __GFP_ATOMIC
) {
3017 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3018 * if it can't schedule.
3020 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3021 alloc_flags
|= ALLOC_HARDER
;
3023 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3024 * comment for __cpuset_node_allowed().
3026 alloc_flags
&= ~ALLOC_CPUSET
;
3027 } else if (unlikely(rt_task(current
)) && !in_interrupt())
3028 alloc_flags
|= ALLOC_HARDER
;
3030 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
3031 if (gfp_mask
& __GFP_MEMALLOC
)
3032 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3033 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
3034 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3035 else if (!in_interrupt() &&
3036 ((current
->flags
& PF_MEMALLOC
) ||
3037 unlikely(test_thread_flag(TIF_MEMDIE
))))
3038 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3041 if ((gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
3042 && !!(gfp_mask
& __GFP_CMA
))
3043 alloc_flags
|= ALLOC_CMA
;
3048 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
3050 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
3053 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
3055 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
3058 static inline struct page
*
3059 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
3060 struct alloc_context
*ac
)
3062 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3063 struct page
*page
= NULL
;
3065 unsigned long pages_reclaimed
= 0;
3066 unsigned long did_some_progress
;
3067 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
3068 bool deferred_compaction
= false;
3069 int contended_compaction
= COMPACT_CONTENDED_NONE
;
3072 * In the slowpath, we sanity check order to avoid ever trying to
3073 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3074 * be using allocators in order of preference for an area that is
3077 if (order
>= MAX_ORDER
) {
3078 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3082 #if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
3083 set_tsk_thread_flag(current
, TIF_MEMALLOC
);
3087 * We also sanity check to catch abuse of atomic reserves being used by
3088 * callers that are not in atomic context.
3090 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3091 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3092 gfp_mask
&= ~__GFP_ATOMIC
;
3095 * If this allocation cannot block and it is for a specific node, then
3096 * fail early. There's no need to wakeup kswapd or retry for a
3097 * speculative node-specific allocation.
3099 if (IS_ENABLED(CONFIG_NUMA
) && (gfp_mask
& __GFP_THISNODE
) && !can_direct_reclaim
)
3103 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3104 wake_all_kswapds(order
, ac
);
3107 * OK, we're below the kswapd watermark and have kicked background
3108 * reclaim. Now things get more complex, so set up alloc_flags according
3109 * to how we want to proceed.
3111 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3114 * Find the true preferred zone if the allocation is unconstrained by
3117 if (!(alloc_flags
& ALLOC_CPUSET
) && !ac
->nodemask
) {
3118 struct zoneref
*preferred_zoneref
;
3119 preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3120 ac
->high_zoneidx
, NULL
, &ac
->preferred_zone
);
3121 ac
->classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3124 /* This is the last chance, in general, before the goto nopage. */
3125 page
= get_page_from_freelist(gfp_mask
, order
,
3126 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3130 /* Allocate without watermarks if the context allows */
3131 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3133 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3134 * the allocation is high priority and these type of
3135 * allocations are system rather than user orientated
3137 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3139 page
= __alloc_pages_high_priority(gfp_mask
, order
, ac
);
3146 /* Caller is not willing to reclaim, we can't balance anything */
3147 if (!can_direct_reclaim
) {
3149 * All existing users of the deprecated __GFP_NOFAIL are
3150 * blockable, so warn of any new users that actually allow this
3151 * type of allocation to fail.
3153 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3157 /* Avoid recursion of direct reclaim */
3158 if (current
->flags
& PF_MEMALLOC
)
3161 /* Avoid allocations with no watermarks from looping endlessly */
3162 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3166 * Try direct compaction. The first pass is asynchronous. Subsequent
3167 * attempts after direct reclaim are synchronous
3169 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3171 &contended_compaction
,
3172 &deferred_compaction
);
3176 /* Checks for THP-specific high-order allocations */
3177 if (is_thp_gfp_mask(gfp_mask
)) {
3179 * If compaction is deferred for high-order allocations, it is
3180 * because sync compaction recently failed. If this is the case
3181 * and the caller requested a THP allocation, we do not want
3182 * to heavily disrupt the system, so we fail the allocation
3183 * instead of entering direct reclaim.
3185 if (deferred_compaction
)
3189 * In all zones where compaction was attempted (and not
3190 * deferred or skipped), lock contention has been detected.
3191 * For THP allocation we do not want to disrupt the others
3192 * so we fallback to base pages instead.
3194 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3198 * If compaction was aborted due to need_resched(), we do not
3199 * want to further increase allocation latency, unless it is
3200 * khugepaged trying to collapse.
3202 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3203 && !(current
->flags
& PF_KTHREAD
))
3208 * It can become very expensive to allocate transparent hugepages at
3209 * fault, so use asynchronous memory compaction for THP unless it is
3210 * khugepaged trying to collapse.
3212 if (!is_thp_gfp_mask(gfp_mask
) || (current
->flags
& PF_KTHREAD
))
3213 migration_mode
= MIGRATE_SYNC_LIGHT
;
3215 /* Try direct reclaim and then allocating */
3216 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3217 &did_some_progress
);
3221 /* Do not loop if specifically requested */
3222 if (gfp_mask
& __GFP_NORETRY
)
3225 /* Keep reclaiming pages as long as there is reasonable progress */
3226 pages_reclaimed
+= did_some_progress
;
3227 if ((did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
3228 ((gfp_mask
& __GFP_REPEAT
) && pages_reclaimed
< (1 << order
))) {
3229 /* Wait for some write requests to complete then retry */
3230 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
3234 /* Reclaim has failed us, start killing things */
3235 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3239 /* Retry as long as the OOM killer is making progress */
3240 if (did_some_progress
)
3245 * High-order allocations do not necessarily loop after
3246 * direct reclaim and reclaim/compaction depends on compaction
3247 * being called after reclaim so call directly if necessary
3249 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3251 &contended_compaction
,
3252 &deferred_compaction
);
3256 #if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
3257 clear_tsk_thread_flag(current
, TIF_MEMALLOC
);
3259 warn_alloc_failed(gfp_mask
, order
, NULL
);
3261 #if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
3262 clear_tsk_thread_flag(current
, TIF_MEMALLOC
);
3268 * This is the 'heart' of the zoned buddy allocator.
3271 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3272 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3274 struct zoneref
*preferred_zoneref
;
3275 struct page
*page
= NULL
;
3276 unsigned int cpuset_mems_cookie
;
3277 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
|ALLOC_FAIR
;
3278 gfp_t alloc_mask
; /* The gfp_t that was actually used for allocation */
3279 struct alloc_context ac
= {
3280 .high_zoneidx
= gfp_zone(gfp_mask
),
3281 .nodemask
= nodemask
,
3282 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3285 gfp_mask
&= gfp_allowed_mask
;
3287 lockdep_trace_alloc(gfp_mask
);
3289 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3291 if (should_fail_alloc_page(gfp_mask
, order
))
3295 * Check the zones suitable for the gfp_mask contain at least one
3296 * valid zone. It's possible to have an empty zonelist as a result
3297 * of __GFP_THISNODE and a memoryless node
3299 if (unlikely(!zonelist
->_zonerefs
->zone
))
3302 if (IS_ENABLED(CONFIG_CMA
) && (ac
.migratetype
== MIGRATE_MOVABLE
)
3303 && !!(gfp_mask
& __GFP_CMA
))
3304 alloc_flags
|= ALLOC_CMA
;
3307 cpuset_mems_cookie
= read_mems_allowed_begin();
3309 /* We set it here, as __alloc_pages_slowpath might have changed it */
3310 ac
.zonelist
= zonelist
;
3312 /* Dirty zone balancing only done in the fast path */
3313 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3315 /* The preferred zone is used for statistics later */
3316 preferred_zoneref
= first_zones_zonelist(ac
.zonelist
, ac
.high_zoneidx
,
3317 ac
.nodemask
? : &cpuset_current_mems_allowed
,
3318 &ac
.preferred_zone
);
3319 if (!ac
.preferred_zone
)
3321 ac
.classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3323 /* First allocation attempt */
3324 alloc_mask
= gfp_mask
|__GFP_HARDWALL
;
3325 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3326 if (unlikely(!page
)) {
3328 * Runtime PM, block IO and its error handling path
3329 * can deadlock because I/O on the device might not
3332 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3333 ac
.spread_dirty_pages
= false;
3335 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3338 if (kmemcheck_enabled
&& page
)
3339 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3341 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3345 * When updating a task's mems_allowed, it is possible to race with
3346 * parallel threads in such a way that an allocation can fail while
3347 * the mask is being updated. If a page allocation is about to fail,
3348 * check if the cpuset changed during allocation and if so, retry.
3350 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
)))
3355 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3358 * Common helper functions.
3360 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3365 * __get_free_pages() returns a 32-bit address, which cannot represent
3368 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3370 page
= alloc_pages(gfp_mask
, order
);
3373 return (unsigned long) page_address(page
);
3375 EXPORT_SYMBOL(__get_free_pages
);
3377 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3379 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3381 EXPORT_SYMBOL(get_zeroed_page
);
3383 void __free_pages(struct page
*page
, unsigned int order
)
3385 if (put_page_testzero(page
)) {
3387 free_hot_cold_page(page
, false);
3389 __free_pages_ok(page
, order
);
3393 EXPORT_SYMBOL(__free_pages
);
3395 void free_pages(unsigned long addr
, unsigned int order
)
3398 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3399 __free_pages(virt_to_page((void *)addr
), order
);
3403 EXPORT_SYMBOL(free_pages
);
3407 * An arbitrary-length arbitrary-offset area of memory which resides
3408 * within a 0 or higher order page. Multiple fragments within that page
3409 * are individually refcounted, in the page's reference counter.
3411 * The page_frag functions below provide a simple allocation framework for
3412 * page fragments. This is used by the network stack and network device
3413 * drivers to provide a backing region of memory for use as either an
3414 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3416 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3419 struct page
*page
= NULL
;
3420 gfp_t gfp
= gfp_mask
;
3422 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3423 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3425 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3426 PAGE_FRAG_CACHE_MAX_ORDER
);
3427 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3429 if (unlikely(!page
))
3430 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3432 nc
->va
= page
? page_address(page
) : NULL
;
3437 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3438 unsigned int fragsz
, gfp_t gfp_mask
)
3440 unsigned int size
= PAGE_SIZE
;
3444 if (unlikely(!nc
->va
)) {
3446 page
= __page_frag_refill(nc
, gfp_mask
);
3450 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3451 /* if size can vary use size else just use PAGE_SIZE */
3454 /* Even if we own the page, we do not use atomic_set().
3455 * This would break get_page_unless_zero() users.
3457 atomic_add(size
- 1, &page
->_count
);
3459 /* reset page count bias and offset to start of new frag */
3460 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3461 nc
->pagecnt_bias
= size
;
3465 offset
= nc
->offset
- fragsz
;
3466 if (unlikely(offset
< 0)) {
3467 page
= virt_to_page(nc
->va
);
3469 if (!atomic_sub_and_test(nc
->pagecnt_bias
, &page
->_count
))
3472 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3473 /* if size can vary use size else just use PAGE_SIZE */
3476 /* OK, page count is 0, we can safely set it */
3477 atomic_set(&page
->_count
, size
);
3479 /* reset page count bias and offset to start of new frag */
3480 nc
->pagecnt_bias
= size
;
3481 offset
= size
- fragsz
;
3485 nc
->offset
= offset
;
3487 return nc
->va
+ offset
;
3489 EXPORT_SYMBOL(__alloc_page_frag
);
3492 * Frees a page fragment allocated out of either a compound or order 0 page.
3494 void __free_page_frag(void *addr
)
3496 struct page
*page
= virt_to_head_page(addr
);
3498 if (unlikely(put_page_testzero(page
)))
3499 __free_pages_ok(page
, compound_order(page
));
3501 EXPORT_SYMBOL(__free_page_frag
);
3504 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3505 * of the current memory cgroup.
3507 * It should be used when the caller would like to use kmalloc, but since the
3508 * allocation is large, it has to fall back to the page allocator.
3510 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
3514 page
= alloc_pages(gfp_mask
, order
);
3515 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3516 __free_pages(page
, order
);
3522 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
3526 page
= alloc_pages_node(nid
, gfp_mask
, order
);
3527 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3528 __free_pages(page
, order
);
3535 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3538 void __free_kmem_pages(struct page
*page
, unsigned int order
)
3540 memcg_kmem_uncharge(page
, order
);
3541 __free_pages(page
, order
);
3544 void free_kmem_pages(unsigned long addr
, unsigned int order
)
3547 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3548 __free_kmem_pages(virt_to_page((void *)addr
), order
);
3552 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
3556 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
3557 unsigned long used
= addr
+ PAGE_ALIGN(size
);
3559 split_page(virt_to_page((void *)addr
), order
);
3560 while (used
< alloc_end
) {
3565 return (void *)addr
;
3569 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3570 * @size: the number of bytes to allocate
3571 * @gfp_mask: GFP flags for the allocation
3573 * This function is similar to alloc_pages(), except that it allocates the
3574 * minimum number of pages to satisfy the request. alloc_pages() can only
3575 * allocate memory in power-of-two pages.
3577 * This function is also limited by MAX_ORDER.
3579 * Memory allocated by this function must be released by free_pages_exact().
3581 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
3583 unsigned int order
= get_order(size
);
3586 addr
= __get_free_pages(gfp_mask
, order
);
3587 return make_alloc_exact(addr
, order
, size
);
3589 EXPORT_SYMBOL(alloc_pages_exact
);
3592 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3594 * @nid: the preferred node ID where memory should be allocated
3595 * @size: the number of bytes to allocate
3596 * @gfp_mask: GFP flags for the allocation
3598 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3601 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
3603 unsigned int order
= get_order(size
);
3604 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
3607 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
3611 * free_pages_exact - release memory allocated via alloc_pages_exact()
3612 * @virt: the value returned by alloc_pages_exact.
3613 * @size: size of allocation, same value as passed to alloc_pages_exact().
3615 * Release the memory allocated by a previous call to alloc_pages_exact.
3617 void free_pages_exact(void *virt
, size_t size
)
3619 unsigned long addr
= (unsigned long)virt
;
3620 unsigned long end
= addr
+ PAGE_ALIGN(size
);
3622 while (addr
< end
) {
3627 EXPORT_SYMBOL(free_pages_exact
);
3630 * nr_free_zone_pages - count number of pages beyond high watermark
3631 * @offset: The zone index of the highest zone
3633 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3634 * high watermark within all zones at or below a given zone index. For each
3635 * zone, the number of pages is calculated as:
3636 * managed_pages - high_pages
3638 static unsigned long nr_free_zone_pages(int offset
)
3643 /* Just pick one node, since fallback list is circular */
3644 unsigned long sum
= 0;
3646 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
3648 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
3649 unsigned long size
= zone
->managed_pages
;
3650 unsigned long high
= high_wmark_pages(zone
);
3659 * nr_free_buffer_pages - count number of pages beyond high watermark
3661 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3662 * watermark within ZONE_DMA and ZONE_NORMAL.
3664 unsigned long nr_free_buffer_pages(void)
3666 return nr_free_zone_pages(gfp_zone(GFP_USER
));
3668 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
3671 * nr_free_pagecache_pages - count number of pages beyond high watermark
3673 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3674 * high watermark within all zones.
3676 unsigned long nr_free_pagecache_pages(void)
3678 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
3681 static inline void show_node(struct zone
*zone
)
3683 if (IS_ENABLED(CONFIG_NUMA
))
3684 printk("Node %d ", zone_to_nid(zone
));
3687 void si_meminfo(struct sysinfo
*val
)
3689 val
->totalram
= totalram_pages
;
3690 val
->sharedram
= global_page_state(NR_SHMEM
);
3691 val
->freeram
= global_page_state(NR_FREE_PAGES
);
3692 val
->bufferram
= nr_blockdev_pages();
3693 val
->totalhigh
= totalhigh_pages
;
3694 val
->freehigh
= nr_free_highpages();
3695 val
->mem_unit
= PAGE_SIZE
;
3698 EXPORT_SYMBOL(si_meminfo
);
3701 void si_meminfo_node(struct sysinfo
*val
, int nid
)
3703 int zone_type
; /* needs to be signed */
3704 unsigned long managed_pages
= 0;
3705 pg_data_t
*pgdat
= NODE_DATA(nid
);
3707 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
3708 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
3709 val
->totalram
= managed_pages
;
3710 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
3711 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
3712 #ifdef CONFIG_HIGHMEM
3713 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
3714 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
3720 val
->mem_unit
= PAGE_SIZE
;
3725 * Determine whether the node should be displayed or not, depending on whether
3726 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3728 bool skip_free_areas_node(unsigned int flags
, int nid
)
3731 unsigned int cpuset_mems_cookie
;
3733 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3737 cpuset_mems_cookie
= read_mems_allowed_begin();
3738 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3739 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
3744 #define K(x) ((x) << (PAGE_SHIFT-10))
3746 static void show_migration_types(unsigned char type
)
3748 static const char types
[MIGRATE_TYPES
] = {
3749 [MIGRATE_UNMOVABLE
] = 'U',
3750 [MIGRATE_MOVABLE
] = 'M',
3751 [MIGRATE_RECLAIMABLE
] = 'E',
3752 [MIGRATE_HIGHATOMIC
] = 'H',
3754 [MIGRATE_CMA
] = 'C',
3756 #ifdef CONFIG_MEMORY_ISOLATION
3757 [MIGRATE_ISOLATE
] = 'I',
3760 char tmp
[MIGRATE_TYPES
+ 1];
3764 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3765 if (type
& (1 << i
))
3770 printk("(%s) ", tmp
);
3774 * Show free area list (used inside shift_scroll-lock stuff)
3775 * We also calculate the percentage fragmentation. We do this by counting the
3776 * memory on each free list with the exception of the first item on the list.
3779 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3782 void show_free_areas(unsigned int filter
)
3784 unsigned long free_pcp
= 0;
3788 for_each_populated_zone(zone
) {
3789 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3792 for_each_online_cpu(cpu
)
3793 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3796 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3797 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3798 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3799 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3800 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3801 " free:%lu free_pcp:%lu free_cma:%lu\n",
3802 global_page_state(NR_ACTIVE_ANON
),
3803 global_page_state(NR_INACTIVE_ANON
),
3804 global_page_state(NR_ISOLATED_ANON
),
3805 global_page_state(NR_ACTIVE_FILE
),
3806 global_page_state(NR_INACTIVE_FILE
),
3807 global_page_state(NR_ISOLATED_FILE
),
3808 global_page_state(NR_UNEVICTABLE
),
3809 global_page_state(NR_FILE_DIRTY
),
3810 global_page_state(NR_WRITEBACK
),
3811 global_page_state(NR_UNSTABLE_NFS
),
3812 global_page_state(NR_SLAB_RECLAIMABLE
),
3813 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3814 global_page_state(NR_FILE_MAPPED
),
3815 global_page_state(NR_SHMEM
),
3816 global_page_state(NR_PAGETABLE
),
3817 global_page_state(NR_BOUNCE
),
3818 global_page_state(NR_FREE_PAGES
),
3820 global_page_state(NR_FREE_CMA_PAGES
));
3822 for_each_populated_zone(zone
) {
3825 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3829 for_each_online_cpu(cpu
)
3830 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3838 " active_anon:%lukB"
3839 " inactive_anon:%lukB"
3840 " active_file:%lukB"
3841 " inactive_file:%lukB"
3842 " unevictable:%lukB"
3843 " isolated(anon):%lukB"
3844 " isolated(file):%lukB"
3852 " slab_reclaimable:%lukB"
3853 " slab_unreclaimable:%lukB"
3854 " kernel_stack:%lukB"
3861 " writeback_tmp:%lukB"
3862 " pages_scanned:%lu"
3863 " all_unreclaimable? %s"
3866 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3867 K(min_wmark_pages(zone
)),
3868 K(low_wmark_pages(zone
)),
3869 K(high_wmark_pages(zone
)),
3870 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3871 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3872 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3873 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3874 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3875 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3876 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3877 K(zone
->present_pages
),
3878 K(zone
->managed_pages
),
3879 K(zone_page_state(zone
, NR_MLOCK
)),
3880 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3881 K(zone_page_state(zone
, NR_WRITEBACK
)),
3882 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3883 K(zone_page_state(zone
, NR_SHMEM
)),
3884 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3885 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3886 zone_page_state(zone
, NR_KERNEL_STACK
) *
3888 K(zone_page_state(zone
, NR_PAGETABLE
)),
3889 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3890 K(zone_page_state(zone
, NR_BOUNCE
)),
3892 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
3893 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3894 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3895 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
3896 (!zone_reclaimable(zone
) ? "yes" : "no")
3898 printk("lowmem_reserve[]:");
3899 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3900 printk(" %ld", zone
->lowmem_reserve
[i
]);
3904 for_each_populated_zone(zone
) {
3906 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
3907 unsigned char types
[MAX_ORDER
];
3909 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3912 printk("%s: ", zone
->name
);
3914 spin_lock_irqsave(&zone
->lock
, flags
);
3915 for (order
= 0; order
< MAX_ORDER
; order
++) {
3916 struct free_area
*area
= &zone
->free_area
[order
];
3919 nr
[order
] = area
->nr_free
;
3920 total
+= nr
[order
] << order
;
3923 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3924 if (!list_empty(&area
->free_list
[type
]))
3925 types
[order
] |= 1 << type
;
3928 spin_unlock_irqrestore(&zone
->lock
, flags
);
3929 for (order
= 0; order
< MAX_ORDER
; order
++) {
3930 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3932 show_migration_types(types
[order
]);
3934 printk("= %lukB\n", K(total
));
3937 hugetlb_show_meminfo();
3939 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3941 show_swap_cache_info();
3944 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3946 zoneref
->zone
= zone
;
3947 zoneref
->zone_idx
= zone_idx(zone
);
3951 * Builds allocation fallback zone lists.
3953 * Add all populated zones of a node to the zonelist.
3955 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3959 enum zone_type zone_type
= MAX_NR_ZONES
;
3963 zone
= pgdat
->node_zones
+ zone_type
;
3964 if (populated_zone(zone
)) {
3965 zoneref_set_zone(zone
,
3966 &zonelist
->_zonerefs
[nr_zones
++]);
3967 check_highest_zone(zone_type
);
3969 } while (zone_type
);
3977 * 0 = automatic detection of better ordering.
3978 * 1 = order by ([node] distance, -zonetype)
3979 * 2 = order by (-zonetype, [node] distance)
3981 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3982 * the same zonelist. So only NUMA can configure this param.
3984 #define ZONELIST_ORDER_DEFAULT 0
3985 #define ZONELIST_ORDER_NODE 1
3986 #define ZONELIST_ORDER_ZONE 2
3988 /* zonelist order in the kernel.
3989 * set_zonelist_order() will set this to NODE or ZONE.
3991 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3992 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3996 /* The value user specified ....changed by config */
3997 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3998 /* string for sysctl */
3999 #define NUMA_ZONELIST_ORDER_LEN 16
4000 char numa_zonelist_order
[16] = "default";
4003 * interface for configure zonelist ordering.
4004 * command line option "numa_zonelist_order"
4005 * = "[dD]efault - default, automatic configuration.
4006 * = "[nN]ode - order by node locality, then by zone within node
4007 * = "[zZ]one - order by zone, then by locality within zone
4010 static int __parse_numa_zonelist_order(char *s
)
4012 if (*s
== 'd' || *s
== 'D') {
4013 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4014 } else if (*s
== 'n' || *s
== 'N') {
4015 user_zonelist_order
= ZONELIST_ORDER_NODE
;
4016 } else if (*s
== 'z' || *s
== 'Z') {
4017 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
4020 "Ignoring invalid numa_zonelist_order value: %s\n", s
);
4026 static __init
int setup_numa_zonelist_order(char *s
)
4033 ret
= __parse_numa_zonelist_order(s
);
4035 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
4039 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
4042 * sysctl handler for numa_zonelist_order
4044 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
4045 void __user
*buffer
, size_t *length
,
4048 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
4050 static DEFINE_MUTEX(zl_order_mutex
);
4052 mutex_lock(&zl_order_mutex
);
4054 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
4058 strcpy(saved_string
, (char *)table
->data
);
4060 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4064 int oldval
= user_zonelist_order
;
4066 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4069 * bogus value. restore saved string
4071 strncpy((char *)table
->data
, saved_string
,
4072 NUMA_ZONELIST_ORDER_LEN
);
4073 user_zonelist_order
= oldval
;
4074 } else if (oldval
!= user_zonelist_order
) {
4075 mutex_lock(&zonelists_mutex
);
4076 build_all_zonelists(NULL
, NULL
);
4077 mutex_unlock(&zonelists_mutex
);
4081 mutex_unlock(&zl_order_mutex
);
4086 #define MAX_NODE_LOAD (nr_online_nodes)
4087 static int node_load
[MAX_NUMNODES
];
4090 * find_next_best_node - find the next node that should appear in a given node's fallback list
4091 * @node: node whose fallback list we're appending
4092 * @used_node_mask: nodemask_t of already used nodes
4094 * We use a number of factors to determine which is the next node that should
4095 * appear on a given node's fallback list. The node should not have appeared
4096 * already in @node's fallback list, and it should be the next closest node
4097 * according to the distance array (which contains arbitrary distance values
4098 * from each node to each node in the system), and should also prefer nodes
4099 * with no CPUs, since presumably they'll have very little allocation pressure
4100 * on them otherwise.
4101 * It returns -1 if no node is found.
4103 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4106 int min_val
= INT_MAX
;
4107 int best_node
= NUMA_NO_NODE
;
4108 const struct cpumask
*tmp
= cpumask_of_node(0);
4110 /* Use the local node if we haven't already */
4111 if (!node_isset(node
, *used_node_mask
)) {
4112 node_set(node
, *used_node_mask
);
4116 for_each_node_state(n
, N_MEMORY
) {
4118 /* Don't want a node to appear more than once */
4119 if (node_isset(n
, *used_node_mask
))
4122 /* Use the distance array to find the distance */
4123 val
= node_distance(node
, n
);
4125 /* Penalize nodes under us ("prefer the next node") */
4128 /* Give preference to headless and unused nodes */
4129 tmp
= cpumask_of_node(n
);
4130 if (!cpumask_empty(tmp
))
4131 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4133 /* Slight preference for less loaded node */
4134 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4135 val
+= node_load
[n
];
4137 if (val
< min_val
) {
4144 node_set(best_node
, *used_node_mask
);
4151 * Build zonelists ordered by node and zones within node.
4152 * This results in maximum locality--normal zone overflows into local
4153 * DMA zone, if any--but risks exhausting DMA zone.
4155 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4158 struct zonelist
*zonelist
;
4160 zonelist
= &pgdat
->node_zonelists
[0];
4161 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4163 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4164 zonelist
->_zonerefs
[j
].zone
= NULL
;
4165 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4169 * Build gfp_thisnode zonelists
4171 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4174 struct zonelist
*zonelist
;
4176 zonelist
= &pgdat
->node_zonelists
[1];
4177 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4178 zonelist
->_zonerefs
[j
].zone
= NULL
;
4179 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4183 * Build zonelists ordered by zone and nodes within zones.
4184 * This results in conserving DMA zone[s] until all Normal memory is
4185 * exhausted, but results in overflowing to remote node while memory
4186 * may still exist in local DMA zone.
4188 static int node_order
[MAX_NUMNODES
];
4190 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4193 int zone_type
; /* needs to be signed */
4195 struct zonelist
*zonelist
;
4197 zonelist
= &pgdat
->node_zonelists
[0];
4199 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4200 for (j
= 0; j
< nr_nodes
; j
++) {
4201 node
= node_order
[j
];
4202 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4203 if (populated_zone(z
)) {
4205 &zonelist
->_zonerefs
[pos
++]);
4206 check_highest_zone(zone_type
);
4210 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4211 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4214 #if defined(CONFIG_64BIT)
4216 * Devices that require DMA32/DMA are relatively rare and do not justify a
4217 * penalty to every machine in case the specialised case applies. Default
4218 * to Node-ordering on 64-bit NUMA machines
4220 static int default_zonelist_order(void)
4222 return ZONELIST_ORDER_NODE
;
4226 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4227 * by the kernel. If processes running on node 0 deplete the low memory zone
4228 * then reclaim will occur more frequency increasing stalls and potentially
4229 * be easier to OOM if a large percentage of the zone is under writeback or
4230 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4231 * Hence, default to zone ordering on 32-bit.
4233 static int default_zonelist_order(void)
4235 return ZONELIST_ORDER_ZONE
;
4237 #endif /* CONFIG_64BIT */
4239 static void set_zonelist_order(void)
4241 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4242 current_zonelist_order
= default_zonelist_order();
4244 current_zonelist_order
= user_zonelist_order
;
4247 static void build_zonelists(pg_data_t
*pgdat
)
4251 nodemask_t used_mask
;
4252 int local_node
, prev_node
;
4253 struct zonelist
*zonelist
;
4254 unsigned int order
= current_zonelist_order
;
4256 /* initialize zonelists */
4257 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4258 zonelist
= pgdat
->node_zonelists
+ i
;
4259 zonelist
->_zonerefs
[0].zone
= NULL
;
4260 zonelist
->_zonerefs
[0].zone_idx
= 0;
4263 /* NUMA-aware ordering of nodes */
4264 local_node
= pgdat
->node_id
;
4265 load
= nr_online_nodes
;
4266 prev_node
= local_node
;
4267 nodes_clear(used_mask
);
4269 memset(node_order
, 0, sizeof(node_order
));
4272 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4274 * We don't want to pressure a particular node.
4275 * So adding penalty to the first node in same
4276 * distance group to make it round-robin.
4278 if (node_distance(local_node
, node
) !=
4279 node_distance(local_node
, prev_node
))
4280 node_load
[node
] = load
;
4284 if (order
== ZONELIST_ORDER_NODE
)
4285 build_zonelists_in_node_order(pgdat
, node
);
4287 node_order
[j
++] = node
; /* remember order */
4290 if (order
== ZONELIST_ORDER_ZONE
) {
4291 /* calculate node order -- i.e., DMA last! */
4292 build_zonelists_in_zone_order(pgdat
, j
);
4295 build_thisnode_zonelists(pgdat
);
4298 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4300 * Return node id of node used for "local" allocations.
4301 * I.e., first node id of first zone in arg node's generic zonelist.
4302 * Used for initializing percpu 'numa_mem', which is used primarily
4303 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4305 int local_memory_node(int node
)
4309 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4310 gfp_zone(GFP_KERNEL
),
4317 #else /* CONFIG_NUMA */
4319 static void set_zonelist_order(void)
4321 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4324 static void build_zonelists(pg_data_t
*pgdat
)
4326 int node
, local_node
;
4328 struct zonelist
*zonelist
;
4330 local_node
= pgdat
->node_id
;
4332 zonelist
= &pgdat
->node_zonelists
[0];
4333 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4336 * Now we build the zonelist so that it contains the zones
4337 * of all the other nodes.
4338 * We don't want to pressure a particular node, so when
4339 * building the zones for node N, we make sure that the
4340 * zones coming right after the local ones are those from
4341 * node N+1 (modulo N)
4343 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4344 if (!node_online(node
))
4346 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4348 for (node
= 0; node
< local_node
; node
++) {
4349 if (!node_online(node
))
4351 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4354 zonelist
->_zonerefs
[j
].zone
= NULL
;
4355 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4358 #endif /* CONFIG_NUMA */
4361 * Boot pageset table. One per cpu which is going to be used for all
4362 * zones and all nodes. The parameters will be set in such a way
4363 * that an item put on a list will immediately be handed over to
4364 * the buddy list. This is safe since pageset manipulation is done
4365 * with interrupts disabled.
4367 * The boot_pagesets must be kept even after bootup is complete for
4368 * unused processors and/or zones. They do play a role for bootstrapping
4369 * hotplugged processors.
4371 * zoneinfo_show() and maybe other functions do
4372 * not check if the processor is online before following the pageset pointer.
4373 * Other parts of the kernel may not check if the zone is available.
4375 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4376 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4377 static void setup_zone_pageset(struct zone
*zone
);
4380 * Global mutex to protect against size modification of zonelists
4381 * as well as to serialize pageset setup for the new populated zone.
4383 DEFINE_MUTEX(zonelists_mutex
);
4385 /* return values int ....just for stop_machine() */
4386 static int __build_all_zonelists(void *data
)
4390 pg_data_t
*self
= data
;
4393 memset(node_load
, 0, sizeof(node_load
));
4396 if (self
&& !node_online(self
->node_id
)) {
4397 build_zonelists(self
);
4400 for_each_online_node(nid
) {
4401 pg_data_t
*pgdat
= NODE_DATA(nid
);
4403 build_zonelists(pgdat
);
4407 * Initialize the boot_pagesets that are going to be used
4408 * for bootstrapping processors. The real pagesets for
4409 * each zone will be allocated later when the per cpu
4410 * allocator is available.
4412 * boot_pagesets are used also for bootstrapping offline
4413 * cpus if the system is already booted because the pagesets
4414 * are needed to initialize allocators on a specific cpu too.
4415 * F.e. the percpu allocator needs the page allocator which
4416 * needs the percpu allocator in order to allocate its pagesets
4417 * (a chicken-egg dilemma).
4419 for_each_possible_cpu(cpu
) {
4420 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4422 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4424 * We now know the "local memory node" for each node--
4425 * i.e., the node of the first zone in the generic zonelist.
4426 * Set up numa_mem percpu variable for on-line cpus. During
4427 * boot, only the boot cpu should be on-line; we'll init the
4428 * secondary cpus' numa_mem as they come on-line. During
4429 * node/memory hotplug, we'll fixup all on-line cpus.
4431 if (cpu_online(cpu
))
4432 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4439 static noinline
void __init
4440 build_all_zonelists_init(void)
4442 __build_all_zonelists(NULL
);
4443 mminit_verify_zonelist();
4444 cpuset_init_current_mems_allowed();
4448 * Called with zonelists_mutex held always
4449 * unless system_state == SYSTEM_BOOTING.
4451 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4452 * [we're only called with non-NULL zone through __meminit paths] and
4453 * (2) call of __init annotated helper build_all_zonelists_init
4454 * [protected by SYSTEM_BOOTING].
4456 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
4458 set_zonelist_order();
4460 if (system_state
== SYSTEM_BOOTING
) {
4461 build_all_zonelists_init();
4463 #ifdef CONFIG_MEMORY_HOTPLUG
4465 setup_zone_pageset(zone
);
4467 /* we have to stop all cpus to guarantee there is no user
4469 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
4470 /* cpuset refresh routine should be here */
4472 vm_total_pages
= nr_free_pagecache_pages();
4474 * Disable grouping by mobility if the number of pages in the
4475 * system is too low to allow the mechanism to work. It would be
4476 * more accurate, but expensive to check per-zone. This check is
4477 * made on memory-hotadd so a system can start with mobility
4478 * disabled and enable it later
4480 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
4481 page_group_by_mobility_disabled
= 1;
4483 page_group_by_mobility_disabled
= 0;
4485 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
4487 zonelist_order_name
[current_zonelist_order
],
4488 page_group_by_mobility_disabled
? "off" : "on",
4491 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
4496 * Helper functions to size the waitqueue hash table.
4497 * Essentially these want to choose hash table sizes sufficiently
4498 * large so that collisions trying to wait on pages are rare.
4499 * But in fact, the number of active page waitqueues on typical
4500 * systems is ridiculously low, less than 200. So this is even
4501 * conservative, even though it seems large.
4503 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4504 * waitqueues, i.e. the size of the waitq table given the number of pages.
4506 #define PAGES_PER_WAITQUEUE 256
4508 #ifndef CONFIG_MEMORY_HOTPLUG
4509 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4511 unsigned long size
= 1;
4513 pages
/= PAGES_PER_WAITQUEUE
;
4515 while (size
< pages
)
4519 * Once we have dozens or even hundreds of threads sleeping
4520 * on IO we've got bigger problems than wait queue collision.
4521 * Limit the size of the wait table to a reasonable size.
4523 size
= min(size
, 4096UL);
4525 return max(size
, 4UL);
4529 * A zone's size might be changed by hot-add, so it is not possible to determine
4530 * a suitable size for its wait_table. So we use the maximum size now.
4532 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4534 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4535 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4536 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4538 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4539 * or more by the traditional way. (See above). It equals:
4541 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4542 * ia64(16K page size) : = ( 8G + 4M)byte.
4543 * powerpc (64K page size) : = (32G +16M)byte.
4545 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4552 * This is an integer logarithm so that shifts can be used later
4553 * to extract the more random high bits from the multiplicative
4554 * hash function before the remainder is taken.
4556 static inline unsigned long wait_table_bits(unsigned long size
)
4562 * Initially all pages are reserved - free ones are freed
4563 * up by free_all_bootmem() once the early boot process is
4564 * done. Non-atomic initialization, single-pass.
4566 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
4567 unsigned long start_pfn
, enum memmap_context context
)
4569 pg_data_t
*pgdat
= NODE_DATA(nid
);
4570 unsigned long end_pfn
= start_pfn
+ size
;
4573 unsigned long nr_initialised
= 0;
4575 if (highest_memmap_pfn
< end_pfn
- 1)
4576 highest_memmap_pfn
= end_pfn
- 1;
4578 z
= &pgdat
->node_zones
[zone
];
4579 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4581 * There can be holes in boot-time mem_map[]s
4582 * handed to this function. They do not
4583 * exist on hotplugged memory.
4585 if (context
== MEMMAP_EARLY
) {
4586 if (!early_pfn_valid(pfn
))
4588 if (!early_pfn_in_nid(pfn
, nid
))
4590 if (!update_defer_init(pgdat
, pfn
, end_pfn
,
4596 * Mark the block movable so that blocks are reserved for
4597 * movable at startup. This will force kernel allocations
4598 * to reserve their blocks rather than leaking throughout
4599 * the address space during boot when many long-lived
4600 * kernel allocations are made.
4602 * bitmap is created for zone's valid pfn range. but memmap
4603 * can be created for invalid pages (for alignment)
4604 * check here not to call set_pageblock_migratetype() against
4607 if (!(pfn
& (pageblock_nr_pages
- 1))) {
4608 struct page
*page
= pfn_to_page(pfn
);
4610 __init_single_page(page
, pfn
, zone
, nid
);
4611 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4613 __init_single_pfn(pfn
, zone
, nid
);
4618 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4620 unsigned int order
, t
;
4621 for_each_migratetype_order(order
, t
) {
4622 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4623 zone
->free_area
[order
].nr_free
= 0;
4627 #ifndef __HAVE_ARCH_MEMMAP_INIT
4628 #define memmap_init(size, nid, zone, start_pfn) \
4629 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4632 static int zone_batchsize(struct zone
*zone
)
4638 * The per-cpu-pages pools are set to around 1000th of the
4639 * size of the zone. But no more than 1/2 of a meg.
4641 * OK, so we don't know how big the cache is. So guess.
4643 batch
= zone
->managed_pages
/ 1024;
4644 if (batch
* PAGE_SIZE
> 512 * 1024)
4645 batch
= (512 * 1024) / PAGE_SIZE
;
4646 batch
/= 4; /* We effectively *= 4 below */
4651 * Clamp the batch to a 2^n - 1 value. Having a power
4652 * of 2 value was found to be more likely to have
4653 * suboptimal cache aliasing properties in some cases.
4655 * For example if 2 tasks are alternately allocating
4656 * batches of pages, one task can end up with a lot
4657 * of pages of one half of the possible page colors
4658 * and the other with pages of the other colors.
4660 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4665 /* The deferral and batching of frees should be suppressed under NOMMU
4668 * The problem is that NOMMU needs to be able to allocate large chunks
4669 * of contiguous memory as there's no hardware page translation to
4670 * assemble apparent contiguous memory from discontiguous pages.
4672 * Queueing large contiguous runs of pages for batching, however,
4673 * causes the pages to actually be freed in smaller chunks. As there
4674 * can be a significant delay between the individual batches being
4675 * recycled, this leads to the once large chunks of space being
4676 * fragmented and becoming unavailable for high-order allocations.
4683 * pcp->high and pcp->batch values are related and dependent on one another:
4684 * ->batch must never be higher then ->high.
4685 * The following function updates them in a safe manner without read side
4688 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4689 * those fields changing asynchronously (acording the the above rule).
4691 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4692 * outside of boot time (or some other assurance that no concurrent updaters
4695 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4696 unsigned long batch
)
4698 /* start with a fail safe value for batch */
4702 /* Update high, then batch, in order */
4709 /* a companion to pageset_set_high() */
4710 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4712 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4715 static void pageset_init(struct per_cpu_pageset
*p
)
4717 struct per_cpu_pages
*pcp
;
4720 memset(p
, 0, sizeof(*p
));
4724 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4725 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4728 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4731 pageset_set_batch(p
, batch
);
4735 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4736 * to the value high for the pageset p.
4738 static void pageset_set_high(struct per_cpu_pageset
*p
,
4741 unsigned long batch
= max(1UL, high
/ 4);
4742 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4743 batch
= PAGE_SHIFT
* 8;
4745 pageset_update(&p
->pcp
, high
, batch
);
4748 static void pageset_set_high_and_batch(struct zone
*zone
,
4749 struct per_cpu_pageset
*pcp
)
4751 if (percpu_pagelist_fraction
)
4752 pageset_set_high(pcp
,
4753 (zone
->managed_pages
/
4754 percpu_pagelist_fraction
));
4756 pageset_set_batch(pcp
, zone_batchsize(zone
));
4759 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4761 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4764 pageset_set_high_and_batch(zone
, pcp
);
4767 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4770 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4771 for_each_possible_cpu(cpu
)
4772 zone_pageset_init(zone
, cpu
);
4776 * Allocate per cpu pagesets and initialize them.
4777 * Before this call only boot pagesets were available.
4779 void __init
setup_per_cpu_pageset(void)
4783 for_each_populated_zone(zone
)
4784 setup_zone_pageset(zone
);
4787 static noinline __init_refok
4788 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4794 * The per-page waitqueue mechanism uses hashed waitqueues
4797 zone
->wait_table_hash_nr_entries
=
4798 wait_table_hash_nr_entries(zone_size_pages
);
4799 zone
->wait_table_bits
=
4800 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4801 alloc_size
= zone
->wait_table_hash_nr_entries
4802 * sizeof(wait_queue_head_t
);
4804 if (!slab_is_available()) {
4805 zone
->wait_table
= (wait_queue_head_t
*)
4806 memblock_virt_alloc_node_nopanic(
4807 alloc_size
, zone
->zone_pgdat
->node_id
);
4810 * This case means that a zone whose size was 0 gets new memory
4811 * via memory hot-add.
4812 * But it may be the case that a new node was hot-added. In
4813 * this case vmalloc() will not be able to use this new node's
4814 * memory - this wait_table must be initialized to use this new
4815 * node itself as well.
4816 * To use this new node's memory, further consideration will be
4819 zone
->wait_table
= vmalloc(alloc_size
);
4821 if (!zone
->wait_table
)
4824 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4825 init_waitqueue_head(zone
->wait_table
+ i
);
4830 static __meminit
void zone_pcp_init(struct zone
*zone
)
4833 * per cpu subsystem is not up at this point. The following code
4834 * relies on the ability of the linker to provide the
4835 * offset of a (static) per cpu variable into the per cpu area.
4837 zone
->pageset
= &boot_pageset
;
4839 if (populated_zone(zone
))
4840 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4841 zone
->name
, zone
->present_pages
,
4842 zone_batchsize(zone
));
4845 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4846 unsigned long zone_start_pfn
,
4849 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4851 ret
= zone_wait_table_init(zone
, size
);
4854 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4856 zone
->zone_start_pfn
= zone_start_pfn
;
4858 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4859 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4861 (unsigned long)zone_idx(zone
),
4862 zone_start_pfn
, (zone_start_pfn
+ size
));
4864 zone_init_free_lists(zone
);
4869 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4870 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4873 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4875 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
4876 struct mminit_pfnnid_cache
*state
)
4878 unsigned long start_pfn
, end_pfn
;
4881 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
4882 return state
->last_nid
;
4884 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4886 state
->last_start
= start_pfn
;
4887 state
->last_end
= end_pfn
;
4888 state
->last_nid
= nid
;
4893 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4896 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4897 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4898 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4900 * If an architecture guarantees that all ranges registered contain no holes
4901 * and may be freed, this this function may be used instead of calling
4902 * memblock_free_early_nid() manually.
4904 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4906 unsigned long start_pfn
, end_pfn
;
4909 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4910 start_pfn
= min(start_pfn
, max_low_pfn
);
4911 end_pfn
= min(end_pfn
, max_low_pfn
);
4913 if (start_pfn
< end_pfn
)
4914 memblock_free_early_nid(PFN_PHYS(start_pfn
),
4915 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
4921 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4922 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4924 * If an architecture guarantees that all ranges registered contain no holes and may
4925 * be freed, this function may be used instead of calling memory_present() manually.
4927 void __init
sparse_memory_present_with_active_regions(int nid
)
4929 unsigned long start_pfn
, end_pfn
;
4932 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4933 memory_present(this_nid
, start_pfn
, end_pfn
);
4937 * get_pfn_range_for_nid - Return the start and end page frames for a node
4938 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4939 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4940 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4942 * It returns the start and end page frame of a node based on information
4943 * provided by memblock_set_node(). If called for a node
4944 * with no available memory, a warning is printed and the start and end
4947 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4948 unsigned long *start_pfn
, unsigned long *end_pfn
)
4950 unsigned long this_start_pfn
, this_end_pfn
;
4956 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4957 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4958 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4961 if (*start_pfn
== -1UL)
4966 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4967 * assumption is made that zones within a node are ordered in monotonic
4968 * increasing memory addresses so that the "highest" populated zone is used
4970 static void __init
find_usable_zone_for_movable(void)
4973 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4974 if (zone_index
== ZONE_MOVABLE
)
4977 if (arch_zone_highest_possible_pfn
[zone_index
] >
4978 arch_zone_lowest_possible_pfn
[zone_index
])
4982 VM_BUG_ON(zone_index
== -1);
4983 movable_zone
= zone_index
;
4987 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4988 * because it is sized independent of architecture. Unlike the other zones,
4989 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4990 * in each node depending on the size of each node and how evenly kernelcore
4991 * is distributed. This helper function adjusts the zone ranges
4992 * provided by the architecture for a given node by using the end of the
4993 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4994 * zones within a node are in order of monotonic increases memory addresses
4996 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4997 unsigned long zone_type
,
4998 unsigned long node_start_pfn
,
4999 unsigned long node_end_pfn
,
5000 unsigned long *zone_start_pfn
,
5001 unsigned long *zone_end_pfn
)
5003 /* Only adjust if ZONE_MOVABLE is on this node */
5004 if (zone_movable_pfn
[nid
]) {
5005 /* Size ZONE_MOVABLE */
5006 if (zone_type
== ZONE_MOVABLE
) {
5007 *zone_start_pfn
= zone_movable_pfn
[nid
];
5008 *zone_end_pfn
= min(node_end_pfn
,
5009 arch_zone_highest_possible_pfn
[movable_zone
]);
5011 /* Adjust for ZONE_MOVABLE starting within this range */
5012 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
5013 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
5014 *zone_end_pfn
= zone_movable_pfn
[nid
];
5016 /* Check if this whole range is within ZONE_MOVABLE */
5017 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5018 *zone_start_pfn
= *zone_end_pfn
;
5023 * Return the number of pages a zone spans in a node, including holes
5024 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5026 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5027 unsigned long zone_type
,
5028 unsigned long node_start_pfn
,
5029 unsigned long node_end_pfn
,
5030 unsigned long *ignored
)
5032 unsigned long zone_start_pfn
, zone_end_pfn
;
5034 /* When hotadd a new node from cpu_up(), the node should be empty */
5035 if (!node_start_pfn
&& !node_end_pfn
)
5038 /* Get the start and end of the zone */
5039 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5040 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5041 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5042 node_start_pfn
, node_end_pfn
,
5043 &zone_start_pfn
, &zone_end_pfn
);
5045 /* Check that this node has pages within the zone's required range */
5046 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
5049 /* Move the zone boundaries inside the node if necessary */
5050 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
5051 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
5053 /* Return the spanned pages */
5054 return zone_end_pfn
- zone_start_pfn
;
5058 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5059 * then all holes in the requested range will be accounted for.
5061 unsigned long __meminit
__absent_pages_in_range(int nid
,
5062 unsigned long range_start_pfn
,
5063 unsigned long range_end_pfn
)
5065 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5066 unsigned long start_pfn
, end_pfn
;
5069 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5070 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5071 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5072 nr_absent
-= end_pfn
- start_pfn
;
5078 * absent_pages_in_range - Return number of page frames in holes within a range
5079 * @start_pfn: The start PFN to start searching for holes
5080 * @end_pfn: The end PFN to stop searching for holes
5082 * It returns the number of pages frames in memory holes within a range.
5084 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5085 unsigned long end_pfn
)
5087 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5090 /* Return the number of page frames in holes in a zone on a node */
5091 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5092 unsigned long zone_type
,
5093 unsigned long node_start_pfn
,
5094 unsigned long node_end_pfn
,
5095 unsigned long *ignored
)
5097 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5098 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5099 unsigned long zone_start_pfn
, zone_end_pfn
;
5101 /* When hotadd a new node from cpu_up(), the node should be empty */
5102 if (!node_start_pfn
&& !node_end_pfn
)
5105 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5106 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5108 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5109 node_start_pfn
, node_end_pfn
,
5110 &zone_start_pfn
, &zone_end_pfn
);
5111 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5114 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5115 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5116 unsigned long zone_type
,
5117 unsigned long node_start_pfn
,
5118 unsigned long node_end_pfn
,
5119 unsigned long *zones_size
)
5121 return zones_size
[zone_type
];
5124 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5125 unsigned long zone_type
,
5126 unsigned long node_start_pfn
,
5127 unsigned long node_end_pfn
,
5128 unsigned long *zholes_size
)
5133 return zholes_size
[zone_type
];
5136 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5138 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5139 unsigned long node_start_pfn
,
5140 unsigned long node_end_pfn
,
5141 unsigned long *zones_size
,
5142 unsigned long *zholes_size
)
5144 unsigned long realtotalpages
= 0, totalpages
= 0;
5147 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5148 struct zone
*zone
= pgdat
->node_zones
+ i
;
5149 unsigned long size
, real_size
;
5151 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5155 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5156 node_start_pfn
, node_end_pfn
,
5158 zone
->spanned_pages
= size
;
5159 zone
->present_pages
= real_size
;
5162 realtotalpages
+= real_size
;
5165 pgdat
->node_spanned_pages
= totalpages
;
5166 pgdat
->node_present_pages
= realtotalpages
;
5167 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5171 #ifndef CONFIG_SPARSEMEM
5173 * Calculate the size of the zone->blockflags rounded to an unsigned long
5174 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5175 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5176 * round what is now in bits to nearest long in bits, then return it in
5179 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5181 unsigned long usemapsize
;
5183 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5184 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5185 usemapsize
= usemapsize
>> pageblock_order
;
5186 usemapsize
*= NR_PAGEBLOCK_BITS
;
5187 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5189 return usemapsize
/ 8;
5192 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5194 unsigned long zone_start_pfn
,
5195 unsigned long zonesize
)
5197 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5198 zone
->pageblock_flags
= NULL
;
5200 zone
->pageblock_flags
=
5201 memblock_virt_alloc_node_nopanic(usemapsize
,
5205 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5206 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5207 #endif /* CONFIG_SPARSEMEM */
5209 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5211 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5212 void __paginginit
set_pageblock_order(void)
5216 /* Check that pageblock_nr_pages has not already been setup */
5217 if (pageblock_order
)
5220 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5221 order
= HUGETLB_PAGE_ORDER
;
5223 order
= MAX_ORDER
- 1;
5226 * Assume the largest contiguous order of interest is a huge page.
5227 * This value may be variable depending on boot parameters on IA64 and
5230 pageblock_order
= order
;
5232 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5235 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5236 * is unused as pageblock_order is set at compile-time. See
5237 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5240 void __paginginit
set_pageblock_order(void)
5244 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5246 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5247 unsigned long present_pages
)
5249 unsigned long pages
= spanned_pages
;
5252 * Provide a more accurate estimation if there are holes within
5253 * the zone and SPARSEMEM is in use. If there are holes within the
5254 * zone, each populated memory region may cost us one or two extra
5255 * memmap pages due to alignment because memmap pages for each
5256 * populated regions may not naturally algined on page boundary.
5257 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5259 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5260 IS_ENABLED(CONFIG_SPARSEMEM
))
5261 pages
= present_pages
;
5263 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5267 * Set up the zone data structures:
5268 * - mark all pages reserved
5269 * - mark all memory queues empty
5270 * - clear the memory bitmaps
5272 * NOTE: pgdat should get zeroed by caller.
5274 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5277 int nid
= pgdat
->node_id
;
5278 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
5281 pgdat_resize_init(pgdat
);
5282 #ifdef CONFIG_NUMA_BALANCING
5283 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5284 pgdat
->numabalancing_migrate_nr_pages
= 0;
5285 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5287 init_waitqueue_head(&pgdat
->kswapd_wait
);
5288 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5289 #ifdef CONFIG_COMPACTION
5290 init_waitqueue_head(&pgdat
->kcompactd_wait
);
5292 pgdat_page_ext_init(pgdat
);
5294 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5295 struct zone
*zone
= pgdat
->node_zones
+ j
;
5296 unsigned long size
, realsize
, freesize
, memmap_pages
;
5298 size
= zone
->spanned_pages
;
5299 realsize
= freesize
= zone
->present_pages
;
5302 * Adjust freesize so that it accounts for how much memory
5303 * is used by this zone for memmap. This affects the watermark
5304 * and per-cpu initialisations
5306 memmap_pages
= calc_memmap_size(size
, realsize
);
5307 if (!is_highmem_idx(j
)) {
5308 if (freesize
>= memmap_pages
) {
5309 freesize
-= memmap_pages
;
5312 " %s zone: %lu pages used for memmap\n",
5313 zone_names
[j
], memmap_pages
);
5316 " %s zone: %lu pages exceeds freesize %lu\n",
5317 zone_names
[j
], memmap_pages
, freesize
);
5320 /* Account for reserved pages */
5321 if (j
== 0 && freesize
> dma_reserve
) {
5322 freesize
-= dma_reserve
;
5323 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5324 zone_names
[0], dma_reserve
);
5327 if (!is_highmem_idx(j
))
5328 nr_kernel_pages
+= freesize
;
5329 /* Charge for highmem memmap if there are enough kernel pages */
5330 else if (nr_kernel_pages
> memmap_pages
* 2)
5331 nr_kernel_pages
-= memmap_pages
;
5332 nr_all_pages
+= freesize
;
5335 * Set an approximate value for lowmem here, it will be adjusted
5336 * when the bootmem allocator frees pages into the buddy system.
5337 * And all highmem pages will be managed by the buddy system.
5339 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5342 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5344 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5346 zone
->name
= zone_names
[j
];
5347 spin_lock_init(&zone
->lock
);
5348 spin_lock_init(&zone
->lru_lock
);
5349 zone_seqlock_init(zone
);
5350 zone
->zone_pgdat
= pgdat
;
5351 zone_pcp_init(zone
);
5353 /* For bootup, initialized properly in watermark setup */
5354 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5356 lruvec_init(&zone
->lruvec
);
5360 set_pageblock_order();
5361 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
5362 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
5364 memmap_init(size
, nid
, j
, zone_start_pfn
);
5365 zone_start_pfn
+= size
;
5369 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
5371 unsigned long __maybe_unused start
= 0;
5372 unsigned long __maybe_unused offset
= 0;
5374 /* Skip empty nodes */
5375 if (!pgdat
->node_spanned_pages
)
5378 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5379 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
5380 offset
= pgdat
->node_start_pfn
- start
;
5381 /* ia64 gets its own node_mem_map, before this, without bootmem */
5382 if (!pgdat
->node_mem_map
) {
5383 unsigned long size
, end
;
5387 * The zone's endpoints aren't required to be MAX_ORDER
5388 * aligned but the node_mem_map endpoints must be in order
5389 * for the buddy allocator to function correctly.
5391 end
= pgdat_end_pfn(pgdat
);
5392 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
5393 size
= (end
- start
) * sizeof(struct page
);
5394 map
= alloc_remap(pgdat
->node_id
, size
);
5396 map
= memblock_virt_alloc_node_nopanic(size
,
5398 pgdat
->node_mem_map
= map
+ offset
;
5400 #ifndef CONFIG_NEED_MULTIPLE_NODES
5402 * With no DISCONTIG, the global mem_map is just set as node 0's
5404 if (pgdat
== NODE_DATA(0)) {
5405 mem_map
= NODE_DATA(0)->node_mem_map
;
5406 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5407 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
5409 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5412 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5415 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
5416 unsigned long node_start_pfn
, unsigned long *zholes_size
)
5418 pg_data_t
*pgdat
= NODE_DATA(nid
);
5419 unsigned long start_pfn
= 0;
5420 unsigned long end_pfn
= 0;
5422 /* pg_data_t should be reset to zero when it's allocated */
5423 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
5425 pgdat
->node_id
= nid
;
5426 pgdat
->node_start_pfn
= node_start_pfn
;
5427 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5428 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
5429 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
5430 (u64
)start_pfn
<< PAGE_SHIFT
,
5431 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
5433 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
5434 zones_size
, zholes_size
);
5436 alloc_node_mem_map(pgdat
);
5437 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5438 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5439 nid
, (unsigned long)pgdat
,
5440 (unsigned long)pgdat
->node_mem_map
);
5443 reset_deferred_meminit(pgdat
);
5444 free_area_init_core(pgdat
);
5447 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5449 #if MAX_NUMNODES > 1
5451 * Figure out the number of possible node ids.
5453 void __init
setup_nr_node_ids(void)
5455 unsigned int highest
;
5457 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
5458 nr_node_ids
= highest
+ 1;
5463 * node_map_pfn_alignment - determine the maximum internode alignment
5465 * This function should be called after node map is populated and sorted.
5466 * It calculates the maximum power of two alignment which can distinguish
5469 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5470 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5471 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5472 * shifted, 1GiB is enough and this function will indicate so.
5474 * This is used to test whether pfn -> nid mapping of the chosen memory
5475 * model has fine enough granularity to avoid incorrect mapping for the
5476 * populated node map.
5478 * Returns the determined alignment in pfn's. 0 if there is no alignment
5479 * requirement (single node).
5481 unsigned long __init
node_map_pfn_alignment(void)
5483 unsigned long accl_mask
= 0, last_end
= 0;
5484 unsigned long start
, end
, mask
;
5488 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
5489 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
5496 * Start with a mask granular enough to pin-point to the
5497 * start pfn and tick off bits one-by-one until it becomes
5498 * too coarse to separate the current node from the last.
5500 mask
= ~((1 << __ffs(start
)) - 1);
5501 while (mask
&& last_end
<= (start
& (mask
<< 1)))
5504 /* accumulate all internode masks */
5508 /* convert mask to number of pages */
5509 return ~accl_mask
+ 1;
5512 /* Find the lowest pfn for a node */
5513 static unsigned long __init
find_min_pfn_for_node(int nid
)
5515 unsigned long min_pfn
= ULONG_MAX
;
5516 unsigned long start_pfn
;
5519 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
5520 min_pfn
= min(min_pfn
, start_pfn
);
5522 if (min_pfn
== ULONG_MAX
) {
5524 "Could not find start_pfn for node %d\n", nid
);
5532 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5534 * It returns the minimum PFN based on information provided via
5535 * memblock_set_node().
5537 unsigned long __init
find_min_pfn_with_active_regions(void)
5539 return find_min_pfn_for_node(MAX_NUMNODES
);
5543 * early_calculate_totalpages()
5544 * Sum pages in active regions for movable zone.
5545 * Populate N_MEMORY for calculating usable_nodes.
5547 static unsigned long __init
early_calculate_totalpages(void)
5549 unsigned long totalpages
= 0;
5550 unsigned long start_pfn
, end_pfn
;
5553 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5554 unsigned long pages
= end_pfn
- start_pfn
;
5556 totalpages
+= pages
;
5558 node_set_state(nid
, N_MEMORY
);
5564 * Find the PFN the Movable zone begins in each node. Kernel memory
5565 * is spread evenly between nodes as long as the nodes have enough
5566 * memory. When they don't, some nodes will have more kernelcore than
5569 static void __init
find_zone_movable_pfns_for_nodes(void)
5572 unsigned long usable_startpfn
;
5573 unsigned long kernelcore_node
, kernelcore_remaining
;
5574 /* save the state before borrow the nodemask */
5575 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5576 unsigned long totalpages
= early_calculate_totalpages();
5577 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5578 struct memblock_region
*r
;
5580 /* Need to find movable_zone earlier when movable_node is specified. */
5581 find_usable_zone_for_movable();
5584 * If movable_node is specified, ignore kernelcore and movablecore
5587 if (movable_node_is_enabled()) {
5588 for_each_memblock(memory
, r
) {
5589 if (!memblock_is_hotpluggable(r
))
5594 usable_startpfn
= PFN_DOWN(r
->base
);
5595 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5596 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5604 * If movablecore=nn[KMG] was specified, calculate what size of
5605 * kernelcore that corresponds so that memory usable for
5606 * any allocation type is evenly spread. If both kernelcore
5607 * and movablecore are specified, then the value of kernelcore
5608 * will be used for required_kernelcore if it's greater than
5609 * what movablecore would have allowed.
5611 if (required_movablecore
) {
5612 unsigned long corepages
;
5615 * Round-up so that ZONE_MOVABLE is at least as large as what
5616 * was requested by the user
5618 required_movablecore
=
5619 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5620 required_movablecore
= min(totalpages
, required_movablecore
);
5621 corepages
= totalpages
- required_movablecore
;
5623 required_kernelcore
= max(required_kernelcore
, corepages
);
5627 * If kernelcore was not specified or kernelcore size is larger
5628 * than totalpages, there is no ZONE_MOVABLE.
5630 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
5633 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5634 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5637 /* Spread kernelcore memory as evenly as possible throughout nodes */
5638 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5639 for_each_node_state(nid
, N_MEMORY
) {
5640 unsigned long start_pfn
, end_pfn
;
5643 * Recalculate kernelcore_node if the division per node
5644 * now exceeds what is necessary to satisfy the requested
5645 * amount of memory for the kernel
5647 if (required_kernelcore
< kernelcore_node
)
5648 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5651 * As the map is walked, we track how much memory is usable
5652 * by the kernel using kernelcore_remaining. When it is
5653 * 0, the rest of the node is usable by ZONE_MOVABLE
5655 kernelcore_remaining
= kernelcore_node
;
5657 /* Go through each range of PFNs within this node */
5658 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5659 unsigned long size_pages
;
5661 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5662 if (start_pfn
>= end_pfn
)
5665 /* Account for what is only usable for kernelcore */
5666 if (start_pfn
< usable_startpfn
) {
5667 unsigned long kernel_pages
;
5668 kernel_pages
= min(end_pfn
, usable_startpfn
)
5671 kernelcore_remaining
-= min(kernel_pages
,
5672 kernelcore_remaining
);
5673 required_kernelcore
-= min(kernel_pages
,
5674 required_kernelcore
);
5676 /* Continue if range is now fully accounted */
5677 if (end_pfn
<= usable_startpfn
) {
5680 * Push zone_movable_pfn to the end so
5681 * that if we have to rebalance
5682 * kernelcore across nodes, we will
5683 * not double account here
5685 zone_movable_pfn
[nid
] = end_pfn
;
5688 start_pfn
= usable_startpfn
;
5692 * The usable PFN range for ZONE_MOVABLE is from
5693 * start_pfn->end_pfn. Calculate size_pages as the
5694 * number of pages used as kernelcore
5696 size_pages
= end_pfn
- start_pfn
;
5697 if (size_pages
> kernelcore_remaining
)
5698 size_pages
= kernelcore_remaining
;
5699 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5702 * Some kernelcore has been met, update counts and
5703 * break if the kernelcore for this node has been
5706 required_kernelcore
-= min(required_kernelcore
,
5708 kernelcore_remaining
-= size_pages
;
5709 if (!kernelcore_remaining
)
5715 * If there is still required_kernelcore, we do another pass with one
5716 * less node in the count. This will push zone_movable_pfn[nid] further
5717 * along on the nodes that still have memory until kernelcore is
5721 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5725 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5726 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5727 zone_movable_pfn
[nid
] =
5728 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5731 /* restore the node_state */
5732 node_states
[N_MEMORY
] = saved_node_state
;
5735 /* Any regular or high memory on that node ? */
5736 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5738 enum zone_type zone_type
;
5740 if (N_MEMORY
== N_NORMAL_MEMORY
)
5743 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5744 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5745 if (populated_zone(zone
)) {
5746 node_set_state(nid
, N_HIGH_MEMORY
);
5747 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5748 zone_type
<= ZONE_NORMAL
)
5749 node_set_state(nid
, N_NORMAL_MEMORY
);
5756 * free_area_init_nodes - Initialise all pg_data_t and zone data
5757 * @max_zone_pfn: an array of max PFNs for each zone
5759 * This will call free_area_init_node() for each active node in the system.
5760 * Using the page ranges provided by memblock_set_node(), the size of each
5761 * zone in each node and their holes is calculated. If the maximum PFN
5762 * between two adjacent zones match, it is assumed that the zone is empty.
5763 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5764 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5765 * starts where the previous one ended. For example, ZONE_DMA32 starts
5766 * at arch_max_dma_pfn.
5768 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5770 unsigned long start_pfn
, end_pfn
;
5773 /* Record where the zone boundaries are */
5774 memset(arch_zone_lowest_possible_pfn
, 0,
5775 sizeof(arch_zone_lowest_possible_pfn
));
5776 memset(arch_zone_highest_possible_pfn
, 0,
5777 sizeof(arch_zone_highest_possible_pfn
));
5779 start_pfn
= find_min_pfn_with_active_regions();
5781 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5782 if (i
== ZONE_MOVABLE
)
5785 end_pfn
= max(max_zone_pfn
[i
], start_pfn
);
5786 arch_zone_lowest_possible_pfn
[i
] = start_pfn
;
5787 arch_zone_highest_possible_pfn
[i
] = end_pfn
;
5789 start_pfn
= end_pfn
;
5791 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5792 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5794 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5795 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5796 find_zone_movable_pfns_for_nodes();
5798 /* Print out the zone ranges */
5799 pr_info("Zone ranges:\n");
5800 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5801 if (i
== ZONE_MOVABLE
)
5803 pr_info(" %-8s ", zone_names
[i
]);
5804 if (arch_zone_lowest_possible_pfn
[i
] ==
5805 arch_zone_highest_possible_pfn
[i
])
5808 pr_cont("[mem %#018Lx-%#018Lx]\n",
5809 (u64
)arch_zone_lowest_possible_pfn
[i
]
5811 ((u64
)arch_zone_highest_possible_pfn
[i
]
5812 << PAGE_SHIFT
) - 1);
5815 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5816 pr_info("Movable zone start for each node\n");
5817 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5818 if (zone_movable_pfn
[i
])
5819 pr_info(" Node %d: %#018Lx\n", i
,
5820 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
5823 /* Print out the early node map */
5824 pr_info("Early memory node ranges\n");
5825 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5826 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
5827 (u64
)start_pfn
<< PAGE_SHIFT
,
5828 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
5830 /* Initialise every node */
5831 mminit_verify_pageflags_layout();
5832 setup_nr_node_ids();
5833 for_each_online_node(nid
) {
5834 pg_data_t
*pgdat
= NODE_DATA(nid
);
5835 free_area_init_node(nid
, NULL
,
5836 find_min_pfn_for_node(nid
), NULL
);
5838 /* Any memory on that node */
5839 if (pgdat
->node_present_pages
)
5840 node_set_state(nid
, N_MEMORY
);
5841 check_for_memory(pgdat
, nid
);
5845 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5847 unsigned long long coremem
;
5851 coremem
= memparse(p
, &p
);
5852 *core
= coremem
>> PAGE_SHIFT
;
5854 /* Paranoid check that UL is enough for the coremem value */
5855 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5861 * kernelcore=size sets the amount of memory for use for allocations that
5862 * cannot be reclaimed or migrated.
5864 static int __init
cmdline_parse_kernelcore(char *p
)
5866 return cmdline_parse_core(p
, &required_kernelcore
);
5870 * movablecore=size sets the amount of memory for use for allocations that
5871 * can be reclaimed or migrated.
5873 static int __init
cmdline_parse_movablecore(char *p
)
5875 return cmdline_parse_core(p
, &required_movablecore
);
5878 early_param("kernelcore", cmdline_parse_kernelcore
);
5879 early_param("movablecore", cmdline_parse_movablecore
);
5881 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5883 void adjust_managed_page_count(struct page
*page
, long count
)
5885 spin_lock(&managed_page_count_lock
);
5886 page_zone(page
)->managed_pages
+= count
;
5887 totalram_pages
+= count
;
5888 #ifdef CONFIG_HIGHMEM
5889 if (PageHighMem(page
))
5890 totalhigh_pages
+= count
;
5892 spin_unlock(&managed_page_count_lock
);
5894 EXPORT_SYMBOL(adjust_managed_page_count
);
5896 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
5899 unsigned long pages
= 0;
5901 start
= (void *)PAGE_ALIGN((unsigned long)start
);
5902 end
= (void *)((unsigned long)end
& PAGE_MASK
);
5903 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5904 if ((unsigned int)poison
<= 0xFF)
5905 memset(pos
, poison
, PAGE_SIZE
);
5906 free_reserved_page(virt_to_page(pos
));
5910 pr_info("Freeing %s memory: %ldK\n",
5911 s
, pages
<< (PAGE_SHIFT
- 10));
5915 EXPORT_SYMBOL(free_reserved_area
);
5917 #ifdef CONFIG_HIGHMEM
5918 void free_highmem_page(struct page
*page
)
5920 __free_reserved_page(page
);
5922 page_zone(page
)->managed_pages
++;
5928 void __init
mem_init_print_info(const char *str
)
5930 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
5931 unsigned long init_code_size
, init_data_size
;
5933 physpages
= get_num_physpages();
5934 codesize
= _etext
- _stext
;
5935 datasize
= _edata
- _sdata
;
5936 rosize
= __end_rodata
- __start_rodata
;
5937 bss_size
= __bss_stop
- __bss_start
;
5938 init_data_size
= __init_end
- __init_begin
;
5939 init_code_size
= _einittext
- _sinittext
;
5942 * Detect special cases and adjust section sizes accordingly:
5943 * 1) .init.* may be embedded into .data sections
5944 * 2) .init.text.* may be out of [__init_begin, __init_end],
5945 * please refer to arch/tile/kernel/vmlinux.lds.S.
5946 * 3) .rodata.* may be embedded into .text or .data sections.
5948 #define adj_init_size(start, end, size, pos, adj) \
5950 if (start <= pos && pos < end && size > adj) \
5954 adj_init_size(__init_begin
, __init_end
, init_data_size
,
5955 _sinittext
, init_code_size
);
5956 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
5957 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
5958 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
5959 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
5961 #undef adj_init_size
5963 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
5964 #ifdef CONFIG_HIGHMEM
5968 nr_free_pages() << (PAGE_SHIFT
- 10),
5969 physpages
<< (PAGE_SHIFT
- 10),
5970 codesize
>> 10, datasize
>> 10, rosize
>> 10,
5971 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
5972 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
- 10),
5973 totalcma_pages
<< (PAGE_SHIFT
- 10),
5974 #ifdef CONFIG_HIGHMEM
5975 totalhigh_pages
<< (PAGE_SHIFT
- 10),
5977 str
? ", " : "", str
? str
: "");
5981 * set_dma_reserve - set the specified number of pages reserved in the first zone
5982 * @new_dma_reserve: The number of pages to mark reserved
5984 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
5985 * In the DMA zone, a significant percentage may be consumed by kernel image
5986 * and other unfreeable allocations which can skew the watermarks badly. This
5987 * function may optionally be used to account for unfreeable pages in the
5988 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5989 * smaller per-cpu batchsize.
5991 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5993 dma_reserve
= new_dma_reserve
;
5996 void __init
free_area_init(unsigned long *zones_size
)
5998 free_area_init_node(0, zones_size
,
5999 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6002 static int page_alloc_cpu_notify(struct notifier_block
*self
,
6003 unsigned long action
, void *hcpu
)
6005 int cpu
= (unsigned long)hcpu
;
6007 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
6008 lru_add_drain_cpu(cpu
);
6012 * Spill the event counters of the dead processor
6013 * into the current processors event counters.
6014 * This artificially elevates the count of the current
6017 vm_events_fold_cpu(cpu
);
6020 * Zero the differential counters of the dead processor
6021 * so that the vm statistics are consistent.
6023 * This is only okay since the processor is dead and cannot
6024 * race with what we are doing.
6026 cpu_vm_stats_fold(cpu
);
6031 void __init
page_alloc_init(void)
6033 hotcpu_notifier(page_alloc_cpu_notify
, 0);
6037 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6038 * or min_free_kbytes changes.
6040 static void calculate_totalreserve_pages(void)
6042 struct pglist_data
*pgdat
;
6043 unsigned long reserve_pages
= 0;
6044 enum zone_type i
, j
;
6046 for_each_online_pgdat(pgdat
) {
6047 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6048 struct zone
*zone
= pgdat
->node_zones
+ i
;
6051 /* Find valid and maximum lowmem_reserve in the zone */
6052 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6053 if (zone
->lowmem_reserve
[j
] > max
)
6054 max
= zone
->lowmem_reserve
[j
];
6057 /* we treat the high watermark as reserved pages. */
6058 max
+= high_wmark_pages(zone
);
6060 if (max
> zone
->managed_pages
)
6061 max
= zone
->managed_pages
;
6062 reserve_pages
+= max
;
6064 * Lowmem reserves are not available to
6065 * GFP_HIGHUSER page cache allocations and
6066 * kswapd tries to balance zones to their high
6067 * watermark. As a result, neither should be
6068 * regarded as dirtyable memory, to prevent a
6069 * situation where reclaim has to clean pages
6070 * in order to balance the zones.
6072 zone
->dirty_balance_reserve
= max
;
6075 dirty_balance_reserve
= reserve_pages
;
6076 totalreserve_pages
= reserve_pages
;
6080 * setup_per_zone_lowmem_reserve - called whenever
6081 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6082 * has a correct pages reserved value, so an adequate number of
6083 * pages are left in the zone after a successful __alloc_pages().
6085 static void setup_per_zone_lowmem_reserve(void)
6087 struct pglist_data
*pgdat
;
6088 enum zone_type j
, idx
;
6090 for_each_online_pgdat(pgdat
) {
6091 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6092 struct zone
*zone
= pgdat
->node_zones
+ j
;
6093 unsigned long managed_pages
= zone
->managed_pages
;
6095 zone
->lowmem_reserve
[j
] = 0;
6099 struct zone
*lower_zone
;
6103 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6104 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6106 lower_zone
= pgdat
->node_zones
+ idx
;
6107 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6108 sysctl_lowmem_reserve_ratio
[idx
];
6109 managed_pages
+= lower_zone
->managed_pages
;
6114 /* update totalreserve_pages */
6115 calculate_totalreserve_pages();
6118 static void __setup_per_zone_wmarks(void)
6120 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6121 unsigned long pages_low
= extra_free_kbytes
>> (PAGE_SHIFT
- 10);
6122 unsigned long lowmem_pages
= 0;
6124 unsigned long flags
;
6126 /* Calculate total number of !ZONE_HIGHMEM pages */
6127 for_each_zone(zone
) {
6128 if (!is_highmem(zone
))
6129 lowmem_pages
+= zone
->managed_pages
;
6132 for_each_zone(zone
) {
6135 spin_lock_irqsave(&zone
->lock
, flags
);
6136 min
= (u64
)pages_min
* zone
->managed_pages
;
6137 do_div(min
, lowmem_pages
);
6138 low
= (u64
)pages_low
* zone
->managed_pages
;
6139 do_div(low
, vm_total_pages
);
6141 if (is_highmem(zone
)) {
6143 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6144 * need highmem pages, so cap pages_min to a small
6147 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6148 * deltas control asynch page reclaim, and so should
6149 * not be capped for highmem.
6151 unsigned long min_pages
;
6153 min_pages
= zone
->managed_pages
/ 1024;
6154 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6155 zone
->watermark
[WMARK_MIN
] = min_pages
;
6158 * If it's a lowmem zone, reserve a number of pages
6159 * proportionate to the zone's size.
6161 zone
->watermark
[WMARK_MIN
] = min
;
6164 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) +
6166 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) +
6169 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6170 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6171 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6173 spin_unlock_irqrestore(&zone
->lock
, flags
);
6176 /* update totalreserve_pages */
6177 calculate_totalreserve_pages();
6181 * setup_per_zone_wmarks - called when min_free_kbytes changes
6182 * or when memory is hot-{added|removed}
6184 * Ensures that the watermark[min,low,high] values for each zone are set
6185 * correctly with respect to min_free_kbytes.
6187 void setup_per_zone_wmarks(void)
6189 mutex_lock(&zonelists_mutex
);
6190 __setup_per_zone_wmarks();
6191 mutex_unlock(&zonelists_mutex
);
6195 * The inactive anon list should be small enough that the VM never has to
6196 * do too much work, but large enough that each inactive page has a chance
6197 * to be referenced again before it is swapped out.
6199 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6200 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6201 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6202 * the anonymous pages are kept on the inactive list.
6205 * memory ratio inactive anon
6206 * -------------------------------------
6215 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
6217 #ifdef CONFIG_FIX_INACTIVE_RATIO
6218 zone
->inactive_ratio
= 1;
6220 unsigned int gb
, ratio
;
6222 /* Zone size in gigabytes */
6223 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
6225 ratio
= int_sqrt(10 * gb
);
6229 zone
->inactive_ratio
= ratio
;
6233 static void __meminit
setup_per_zone_inactive_ratio(void)
6238 calculate_zone_inactive_ratio(zone
);
6242 * Initialise min_free_kbytes.
6244 * For small machines we want it small (128k min). For large machines
6245 * we want it large (64MB max). But it is not linear, because network
6246 * bandwidth does not increase linearly with machine size. We use
6248 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6249 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6265 int __meminit
init_per_zone_wmark_min(void)
6267 unsigned long lowmem_kbytes
;
6268 int new_min_free_kbytes
;
6270 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6271 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6273 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6274 min_free_kbytes
= new_min_free_kbytes
;
6275 if (min_free_kbytes
< 128)
6276 min_free_kbytes
= 128;
6277 if (min_free_kbytes
> 65536)
6278 min_free_kbytes
= 65536;
6280 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6281 new_min_free_kbytes
, user_min_free_kbytes
);
6283 setup_per_zone_wmarks();
6284 refresh_zone_stat_thresholds();
6285 setup_per_zone_lowmem_reserve();
6286 setup_per_zone_inactive_ratio();
6289 core_initcall(init_per_zone_wmark_min
)
6292 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6293 * that we can call two helper functions whenever min_free_kbytes
6294 * or extra_free_kbytes changes.
6296 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6297 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6301 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6306 user_min_free_kbytes
= min_free_kbytes
;
6307 setup_per_zone_wmarks();
6313 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6314 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6319 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6324 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6325 sysctl_min_unmapped_ratio
) / 100;
6329 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6330 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6335 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6340 zone
->min_slab_pages
= (zone
->managed_pages
*
6341 sysctl_min_slab_ratio
) / 100;
6347 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6348 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6349 * whenever sysctl_lowmem_reserve_ratio changes.
6351 * The reserve ratio obviously has absolutely no relation with the
6352 * minimum watermarks. The lowmem reserve ratio can only make sense
6353 * if in function of the boot time zone sizes.
6355 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6356 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6358 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6359 setup_per_zone_lowmem_reserve();
6364 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6365 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6366 * pagelist can have before it gets flushed back to buddy allocator.
6368 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
6369 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6372 int old_percpu_pagelist_fraction
;
6375 mutex_lock(&pcp_batch_high_lock
);
6376 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
6378 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6379 if (!write
|| ret
< 0)
6382 /* Sanity checking to avoid pcp imbalance */
6383 if (percpu_pagelist_fraction
&&
6384 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
6385 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
6391 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
6394 for_each_populated_zone(zone
) {
6397 for_each_possible_cpu(cpu
)
6398 pageset_set_high_and_batch(zone
,
6399 per_cpu_ptr(zone
->pageset
, cpu
));
6402 mutex_unlock(&pcp_batch_high_lock
);
6407 int hashdist
= HASHDIST_DEFAULT
;
6409 static int __init
set_hashdist(char *str
)
6413 hashdist
= simple_strtoul(str
, &str
, 0);
6416 __setup("hashdist=", set_hashdist
);
6420 * allocate a large system hash table from bootmem
6421 * - it is assumed that the hash table must contain an exact power-of-2
6422 * quantity of entries
6423 * - limit is the number of hash buckets, not the total allocation size
6425 void *__init
alloc_large_system_hash(const char *tablename
,
6426 unsigned long bucketsize
,
6427 unsigned long numentries
,
6430 unsigned int *_hash_shift
,
6431 unsigned int *_hash_mask
,
6432 unsigned long low_limit
,
6433 unsigned long high_limit
)
6435 unsigned long long max
= high_limit
;
6436 unsigned long log2qty
, size
;
6439 /* allow the kernel cmdline to have a say */
6441 /* round applicable memory size up to nearest megabyte */
6442 numentries
= nr_kernel_pages
;
6444 /* It isn't necessary when PAGE_SIZE >= 1MB */
6445 if (PAGE_SHIFT
< 20)
6446 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
6448 /* limit to 1 bucket per 2^scale bytes of low memory */
6449 if (scale
> PAGE_SHIFT
)
6450 numentries
>>= (scale
- PAGE_SHIFT
);
6452 numentries
<<= (PAGE_SHIFT
- scale
);
6454 /* Make sure we've got at least a 0-order allocation.. */
6455 if (unlikely(flags
& HASH_SMALL
)) {
6456 /* Makes no sense without HASH_EARLY */
6457 WARN_ON(!(flags
& HASH_EARLY
));
6458 if (!(numentries
>> *_hash_shift
)) {
6459 numentries
= 1UL << *_hash_shift
;
6460 BUG_ON(!numentries
);
6462 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
6463 numentries
= PAGE_SIZE
/ bucketsize
;
6465 numentries
= roundup_pow_of_two(numentries
);
6467 /* limit allocation size to 1/16 total memory by default */
6469 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
6470 do_div(max
, bucketsize
);
6472 max
= min(max
, 0x80000000ULL
);
6474 if (numentries
< low_limit
)
6475 numentries
= low_limit
;
6476 if (numentries
> max
)
6479 log2qty
= ilog2(numentries
);
6482 size
= bucketsize
<< log2qty
;
6483 if (flags
& HASH_EARLY
)
6484 table
= memblock_virt_alloc_nopanic(size
, 0);
6486 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
6489 * If bucketsize is not a power-of-two, we may free
6490 * some pages at the end of hash table which
6491 * alloc_pages_exact() automatically does
6493 if (get_order(size
) < MAX_ORDER
) {
6494 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
6495 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
6498 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
6501 panic("Failed to allocate %s hash table\n", tablename
);
6503 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
6506 ilog2(size
) - PAGE_SHIFT
,
6510 *_hash_shift
= log2qty
;
6512 *_hash_mask
= (1 << log2qty
) - 1;
6517 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6518 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
6521 #ifdef CONFIG_SPARSEMEM
6522 return __pfn_to_section(pfn
)->pageblock_flags
;
6524 return zone
->pageblock_flags
;
6525 #endif /* CONFIG_SPARSEMEM */
6528 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
6530 #ifdef CONFIG_SPARSEMEM
6531 pfn
&= (PAGES_PER_SECTION
-1);
6532 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6534 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
6535 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6536 #endif /* CONFIG_SPARSEMEM */
6540 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6541 * @page: The page within the block of interest
6542 * @pfn: The target page frame number
6543 * @end_bitidx: The last bit of interest to retrieve
6544 * @mask: mask of bits that the caller is interested in
6546 * Return: pageblock_bits flags
6548 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
6549 unsigned long end_bitidx
,
6553 unsigned long *bitmap
;
6554 unsigned long bitidx
, word_bitidx
;
6557 zone
= page_zone(page
);
6558 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6559 bitidx
= pfn_to_bitidx(zone
, pfn
);
6560 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6561 bitidx
&= (BITS_PER_LONG
-1);
6563 word
= bitmap
[word_bitidx
];
6564 bitidx
+= end_bitidx
;
6565 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
6569 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6570 * @page: The page within the block of interest
6571 * @flags: The flags to set
6572 * @pfn: The target page frame number
6573 * @end_bitidx: The last bit of interest
6574 * @mask: mask of bits that the caller is interested in
6576 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
6578 unsigned long end_bitidx
,
6582 unsigned long *bitmap
;
6583 unsigned long bitidx
, word_bitidx
;
6584 unsigned long old_word
, word
;
6586 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
6588 zone
= page_zone(page
);
6589 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6590 bitidx
= pfn_to_bitidx(zone
, pfn
);
6591 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6592 bitidx
&= (BITS_PER_LONG
-1);
6594 VM_BUG_ON_PAGE(!zone_spans_pfn(zone
, pfn
), page
);
6596 bitidx
+= end_bitidx
;
6597 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
6598 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
6600 word
= READ_ONCE(bitmap
[word_bitidx
]);
6602 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
6603 if (word
== old_word
)
6610 * This function checks whether pageblock includes unmovable pages or not.
6611 * If @count is not zero, it is okay to include less @count unmovable pages
6613 * PageLRU check without isolation or lru_lock could race so that
6614 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6615 * expect this function should be exact.
6617 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6618 bool skip_hwpoisoned_pages
)
6620 unsigned long pfn
, iter
, found
;
6624 * For avoiding noise data, lru_add_drain_all() should be called
6625 * If ZONE_MOVABLE, the zone never contains unmovable pages
6627 if (zone_idx(zone
) == ZONE_MOVABLE
)
6629 mt
= get_pageblock_migratetype(page
);
6630 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6633 pfn
= page_to_pfn(page
);
6634 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6635 unsigned long check
= pfn
+ iter
;
6637 if (!pfn_valid_within(check
))
6640 page
= pfn_to_page(check
);
6643 * Hugepages are not in LRU lists, but they're movable.
6644 * We need not scan over tail pages bacause we don't
6645 * handle each tail page individually in migration.
6647 if (PageHuge(page
)) {
6648 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6653 * We can't use page_count without pin a page
6654 * because another CPU can free compound page.
6655 * This check already skips compound tails of THP
6656 * because their page->_count is zero at all time.
6658 if (!atomic_read(&page
->_count
)) {
6659 if (PageBuddy(page
))
6660 iter
+= (1 << page_order(page
)) - 1;
6665 * The HWPoisoned page may be not in buddy system, and
6666 * page_count() is not 0.
6668 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6674 * If there are RECLAIMABLE pages, we need to check
6675 * it. But now, memory offline itself doesn't call
6676 * shrink_node_slabs() and it still to be fixed.
6679 * If the page is not RAM, page_count()should be 0.
6680 * we don't need more check. This is an _used_ not-movable page.
6682 * The problematic thing here is PG_reserved pages. PG_reserved
6683 * is set to both of a memory hole page and a _used_ kernel
6692 bool is_pageblock_removable_nolock(struct page
*page
)
6698 * We have to be careful here because we are iterating over memory
6699 * sections which are not zone aware so we might end up outside of
6700 * the zone but still within the section.
6701 * We have to take care about the node as well. If the node is offline
6702 * its NODE_DATA will be NULL - see page_zone.
6704 if (!node_online(page_to_nid(page
)))
6707 zone
= page_zone(page
);
6708 pfn
= page_to_pfn(page
);
6709 if (!zone_spans_pfn(zone
, pfn
))
6712 return !has_unmovable_pages(zone
, page
, 0, true);
6717 static unsigned long pfn_max_align_down(unsigned long pfn
)
6719 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6720 pageblock_nr_pages
) - 1);
6723 static unsigned long pfn_max_align_up(unsigned long pfn
)
6725 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6726 pageblock_nr_pages
));
6729 /* [start, end) must belong to a single zone. */
6730 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6731 unsigned long start
, unsigned long end
,
6734 /* This function is based on compact_zone() from compaction.c. */
6735 unsigned long nr_reclaimed
;
6736 unsigned long pfn
= start
;
6737 unsigned int tries
= 0;
6743 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6744 if (fatal_signal_pending(current
)) {
6749 if (list_empty(&cc
->migratepages
)) {
6750 cc
->nr_migratepages
= 0;
6751 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
6757 } else if (++tries
== 5) {
6758 ret
= ret
< 0 ? ret
: -EBUSY
;
6762 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6764 cc
->nr_migratepages
-= nr_reclaimed
;
6766 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6767 NULL
, 0, cc
->mode
, cma
? MR_CMA
: MR_HPA
);
6770 putback_movable_pages(&cc
->migratepages
);
6777 * __alloc_contig_range() -- tries to allocate given range of pages
6778 * @start: start PFN to allocate
6779 * @end: one-past-the-last PFN to allocate
6780 * @migratetype: migratetype of the underlaying pageblocks (either
6781 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6782 * in range must have the same migratetype and it must
6783 * be either of the two.
6784 * @cma: true if cma allocation
6786 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6787 * aligned, however it's the caller's responsibility to guarantee that
6788 * we are the only thread that changes migrate type of pageblocks the
6791 * The PFN range must belong to a single zone.
6793 * If the 'cma' is not true, caller allows and assumes that this allocation
6796 * Returns zero on success or negative error code. On success all
6797 * pages which PFN is in [start, end) are allocated for the caller and
6798 * need to be freed with free_contig_range().
6800 int __alloc_contig_range(unsigned long start
, unsigned long end
,
6801 unsigned migratetype
, bool cma
)
6803 unsigned long outer_start
, outer_end
;
6807 struct compact_control cc
= {
6808 .nr_migratepages
= 0,
6810 .zone
= page_zone(pfn_to_page(start
)),
6811 .mode
= MIGRATE_SYNC
,
6812 .ignore_skip_hint
= true,
6814 INIT_LIST_HEAD(&cc
.migratepages
);
6817 * What we do here is we mark all pageblocks in range as
6818 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6819 * have different sizes, and due to the way page allocator
6820 * work, we align the range to biggest of the two pages so
6821 * that page allocator won't try to merge buddies from
6822 * different pageblocks and change MIGRATE_ISOLATE to some
6823 * other migration type.
6825 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6826 * migrate the pages from an unaligned range (ie. pages that
6827 * we are interested in). This will put all the pages in
6828 * range back to page allocator as MIGRATE_ISOLATE.
6830 * When this is done, we take the pages in range from page
6831 * allocator removing them from the buddy system. This way
6832 * page allocator will never consider using them.
6834 * This lets us mark the pageblocks back as
6835 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6836 * aligned range but not in the unaligned, original range are
6837 * put back to page allocator so that buddy can use them.
6840 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6841 pfn_max_align_up(end
), migratetype
,
6846 ret
= __alloc_contig_migrate_range(&cc
, start
, end
, cma
);
6851 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6852 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6853 * more, all pages in [start, end) are free in page allocator.
6854 * What we are going to do is to allocate all pages from
6855 * [start, end) (that is remove them from page allocator).
6857 * The only problem is that pages at the beginning and at the
6858 * end of interesting range may be not aligned with pages that
6859 * page allocator holds, ie. they can be part of higher order
6860 * pages. Because of this, we reserve the bigger range and
6861 * once this is done free the pages we are not interested in.
6863 * We don't have to hold zone->lock here because the pages are
6864 * isolated thus they won't get removed from buddy.
6868 outer_start
= start
;
6871 lru_add_drain_all();
6872 drain_all_pages(cc
.zone
);
6874 while (!PageBuddy(pfn_to_page(outer_start
))) {
6875 if (++order
>= MAX_ORDER
) {
6879 outer_start
&= ~0UL << order
;
6882 /* Make sure the range is really isolated. */
6883 if (test_pages_isolated(outer_start
, end
, false)) {
6884 pr_info("%s: [%lx, %lx) PFNs busy\n",
6885 __func__
, outer_start
, end
);
6891 /* Grab isolated pages from freelists. */
6892 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6898 /* Free head and tail (if any) */
6899 if (start
!= outer_start
)
6900 free_contig_range(outer_start
, start
- outer_start
);
6901 if (end
!= outer_end
)
6902 free_contig_range(end
, outer_end
- end
);
6905 undo_isolate_page_range(pfn_max_align_down(start
),
6906 pfn_max_align_up(end
), migratetype
);
6910 int alloc_contig_range(unsigned long start
, unsigned long end
,
6911 unsigned migratetype
)
6913 return __alloc_contig_range(start
, end
, migratetype
, true);
6916 int alloc_contig_range_fast(unsigned long start
, unsigned long end
,
6917 unsigned migratetype
)
6919 return __alloc_contig_range(start
, end
, migratetype
, false);
6922 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6924 unsigned int count
= 0;
6926 for (; nr_pages
--; pfn
++) {
6927 struct page
*page
= pfn_to_page(pfn
);
6929 count
+= page_count(page
) != 1;
6932 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6936 #ifdef CONFIG_MEMORY_HOTPLUG
6938 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6939 * page high values need to be recalulated.
6941 void __meminit
zone_pcp_update(struct zone
*zone
)
6944 mutex_lock(&pcp_batch_high_lock
);
6945 for_each_possible_cpu(cpu
)
6946 pageset_set_high_and_batch(zone
,
6947 per_cpu_ptr(zone
->pageset
, cpu
));
6948 mutex_unlock(&pcp_batch_high_lock
);
6952 void zone_pcp_reset(struct zone
*zone
)
6954 unsigned long flags
;
6956 struct per_cpu_pageset
*pset
;
6958 /* avoid races with drain_pages() */
6959 local_irq_save(flags
);
6960 if (zone
->pageset
!= &boot_pageset
) {
6961 for_each_online_cpu(cpu
) {
6962 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6963 drain_zonestat(zone
, pset
);
6965 free_percpu(zone
->pageset
);
6966 zone
->pageset
= &boot_pageset
;
6968 local_irq_restore(flags
);
6971 #ifdef CONFIG_MEMORY_HOTREMOVE
6973 * All pages in the range must be isolated before calling this.
6976 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6980 unsigned int order
, i
;
6982 unsigned long flags
;
6983 /* find the first valid pfn */
6984 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6989 zone
= page_zone(pfn_to_page(pfn
));
6990 spin_lock_irqsave(&zone
->lock
, flags
);
6992 while (pfn
< end_pfn
) {
6993 if (!pfn_valid(pfn
)) {
6997 page
= pfn_to_page(pfn
);
6999 * The HWPoisoned page may be not in buddy system, and
7000 * page_count() is not 0.
7002 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7004 SetPageReserved(page
);
7008 BUG_ON(page_count(page
));
7009 BUG_ON(!PageBuddy(page
));
7010 order
= page_order(page
);
7011 #ifdef CONFIG_DEBUG_VM
7012 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
7013 pfn
, 1 << order
, end_pfn
);
7015 list_del(&page
->lru
);
7016 rmv_page_order(page
);
7017 zone
->free_area
[order
].nr_free
--;
7018 for (i
= 0; i
< (1 << order
); i
++)
7019 SetPageReserved((page
+i
));
7020 pfn
+= (1 << order
);
7022 spin_unlock_irqrestore(&zone
->lock
, flags
);
7026 #ifdef CONFIG_MEMORY_FAILURE
7027 bool is_free_buddy_page(struct page
*page
)
7029 struct zone
*zone
= page_zone(page
);
7030 unsigned long pfn
= page_to_pfn(page
);
7031 unsigned long flags
;
7034 spin_lock_irqsave(&zone
->lock
, flags
);
7035 for (order
= 0; order
< MAX_ORDER
; order
++) {
7036 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7038 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7041 spin_unlock_irqrestore(&zone
->lock
, flags
);
7043 return order
< MAX_ORDER
;