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
;
129 * A cached value of the page's pageblock's migratetype, used when the page is
130 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
131 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
132 * Also the migratetype set in the page does not necessarily match the pcplist
133 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
134 * other index - this ensures that it will be put on the correct CMA freelist.
136 static inline int get_pcppage_migratetype(struct page
*page
)
141 static inline void set_pcppage_migratetype(struct page
*page
, int migratetype
)
143 page
->index
= migratetype
;
146 #ifdef CONFIG_PM_SLEEP
148 * The following functions are used by the suspend/hibernate code to temporarily
149 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
150 * while devices are suspended. To avoid races with the suspend/hibernate code,
151 * they should always be called with pm_mutex held (gfp_allowed_mask also should
152 * only be modified with pm_mutex held, unless the suspend/hibernate code is
153 * guaranteed not to run in parallel with that modification).
156 static gfp_t saved_gfp_mask
;
158 void pm_restore_gfp_mask(void)
160 WARN_ON(!mutex_is_locked(&pm_mutex
));
161 if (saved_gfp_mask
) {
162 gfp_allowed_mask
= saved_gfp_mask
;
167 void pm_restrict_gfp_mask(void)
169 WARN_ON(!mutex_is_locked(&pm_mutex
));
170 WARN_ON(saved_gfp_mask
);
171 saved_gfp_mask
= gfp_allowed_mask
;
172 gfp_allowed_mask
&= ~(__GFP_IO
| __GFP_FS
);
175 bool pm_suspended_storage(void)
177 if ((gfp_allowed_mask
& (__GFP_IO
| __GFP_FS
)) == (__GFP_IO
| __GFP_FS
))
181 #endif /* CONFIG_PM_SLEEP */
183 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
184 int pageblock_order __read_mostly
;
187 static void __free_pages_ok(struct page
*page
, unsigned int order
);
190 * results with 256, 32 in the lowmem_reserve sysctl:
191 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
192 * 1G machine -> (16M dma, 784M normal, 224M high)
193 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
194 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
195 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
197 * TBD: should special case ZONE_DMA32 machines here - in those we normally
198 * don't need any ZONE_NORMAL reservation
200 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
201 #ifdef CONFIG_ZONE_DMA
204 #ifdef CONFIG_ZONE_DMA32
207 #ifdef CONFIG_HIGHMEM
213 EXPORT_SYMBOL(totalram_pages
);
215 static char * const zone_names
[MAX_NR_ZONES
] = {
216 #ifdef CONFIG_ZONE_DMA
219 #ifdef CONFIG_ZONE_DMA32
223 #ifdef CONFIG_HIGHMEM
227 #ifdef CONFIG_ZONE_DEVICE
232 int min_free_kbytes
= 1024;
233 int user_min_free_kbytes
= -1;
235 static unsigned long __meminitdata nr_kernel_pages
;
236 static unsigned long __meminitdata nr_all_pages
;
237 static unsigned long __meminitdata dma_reserve
;
239 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
240 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
241 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
242 static unsigned long __initdata required_kernelcore
;
243 static unsigned long __initdata required_movablecore
;
244 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
246 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
248 EXPORT_SYMBOL(movable_zone
);
249 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
252 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
253 int nr_online_nodes __read_mostly
= 1;
254 EXPORT_SYMBOL(nr_node_ids
);
255 EXPORT_SYMBOL(nr_online_nodes
);
258 int page_group_by_mobility_disabled __read_mostly
;
260 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
261 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
263 pgdat
->first_deferred_pfn
= ULONG_MAX
;
266 /* Returns true if the struct page for the pfn is uninitialised */
267 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
269 if (pfn
>= NODE_DATA(early_pfn_to_nid(pfn
))->first_deferred_pfn
)
275 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
277 if (pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
284 * Returns false when the remaining initialisation should be deferred until
285 * later in the boot cycle when it can be parallelised.
287 static inline bool update_defer_init(pg_data_t
*pgdat
,
288 unsigned long pfn
, unsigned long zone_end
,
289 unsigned long *nr_initialised
)
291 /* Always populate low zones for address-contrained allocations */
292 if (zone_end
< pgdat_end_pfn(pgdat
))
295 /* Initialise at least 2G of the highest zone */
297 if (*nr_initialised
> (2UL << (30 - PAGE_SHIFT
)) &&
298 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
299 pgdat
->first_deferred_pfn
= pfn
;
306 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
310 static inline bool early_page_uninitialised(unsigned long pfn
)
315 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
320 static inline bool update_defer_init(pg_data_t
*pgdat
,
321 unsigned long pfn
, unsigned long zone_end
,
322 unsigned long *nr_initialised
)
329 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
331 if (unlikely(page_group_by_mobility_disabled
&&
332 migratetype
< MIGRATE_PCPTYPES
))
333 migratetype
= MIGRATE_UNMOVABLE
;
335 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
336 PB_migrate
, PB_migrate_end
);
339 #ifdef CONFIG_DEBUG_VM
340 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
344 unsigned long pfn
= page_to_pfn(page
);
345 unsigned long sp
, start_pfn
;
348 seq
= zone_span_seqbegin(zone
);
349 start_pfn
= zone
->zone_start_pfn
;
350 sp
= zone
->spanned_pages
;
351 if (!zone_spans_pfn(zone
, pfn
))
353 } while (zone_span_seqretry(zone
, seq
));
356 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
357 pfn
, zone_to_nid(zone
), zone
->name
,
358 start_pfn
, start_pfn
+ sp
);
363 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
365 if (!pfn_valid_within(page_to_pfn(page
)))
367 if (zone
!= page_zone(page
))
373 * Temporary debugging check for pages not lying within a given zone.
375 static int bad_range(struct zone
*zone
, struct page
*page
)
377 if (page_outside_zone_boundaries(zone
, page
))
379 if (!page_is_consistent(zone
, page
))
385 static inline int bad_range(struct zone
*zone
, struct page
*page
)
391 static void bad_page(struct page
*page
, const char *reason
,
392 unsigned long bad_flags
)
394 static unsigned long resume
;
395 static unsigned long nr_shown
;
396 static unsigned long nr_unshown
;
398 /* Don't complain about poisoned pages */
399 if (PageHWPoison(page
)) {
400 page_mapcount_reset(page
); /* remove PageBuddy */
405 * Allow a burst of 60 reports, then keep quiet for that minute;
406 * or allow a steady drip of one report per second.
408 if (nr_shown
== 60) {
409 if (time_before(jiffies
, resume
)) {
415 "BUG: Bad page state: %lu messages suppressed\n",
422 resume
= jiffies
+ 60 * HZ
;
424 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
425 current
->comm
, page_to_pfn(page
));
426 dump_page_badflags(page
, reason
, bad_flags
);
431 /* Leave bad fields for debug, except PageBuddy could make trouble */
432 page_mapcount_reset(page
); /* remove PageBuddy */
433 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
437 * Higher-order pages are called "compound pages". They are structured thusly:
439 * The first PAGE_SIZE page is called the "head page".
441 * The remaining PAGE_SIZE pages are called "tail pages".
443 * All pages have PG_compound set. All tail pages have their ->first_page
444 * pointing at the head page.
446 * The first tail page's ->lru.next holds the address of the compound page's
447 * put_page() function. Its ->lru.prev holds the order of allocation.
448 * This usage means that zero-order pages may not be compound.
451 static void free_compound_page(struct page
*page
)
453 __free_pages_ok(page
, compound_order(page
));
456 void prep_compound_page(struct page
*page
, unsigned long order
)
459 int nr_pages
= 1 << order
;
461 set_compound_page_dtor(page
, free_compound_page
);
462 set_compound_order(page
, order
);
464 for (i
= 1; i
< nr_pages
; i
++) {
465 struct page
*p
= page
+ i
;
466 set_page_count(p
, 0);
467 p
->first_page
= page
;
468 /* Make sure p->first_page is always valid for PageTail() */
474 #ifdef CONFIG_DEBUG_PAGEALLOC
475 unsigned int _debug_guardpage_minorder
;
476 bool _debug_pagealloc_enabled __read_mostly
;
477 bool _debug_guardpage_enabled __read_mostly
;
479 static int __init
early_debug_pagealloc(char *buf
)
484 if (strcmp(buf
, "on") == 0)
485 _debug_pagealloc_enabled
= true;
489 early_param("debug_pagealloc", early_debug_pagealloc
);
491 static bool need_debug_guardpage(void)
493 /* If we don't use debug_pagealloc, we don't need guard page */
494 if (!debug_pagealloc_enabled())
500 static void init_debug_guardpage(void)
502 if (!debug_pagealloc_enabled())
505 _debug_guardpage_enabled
= true;
508 struct page_ext_operations debug_guardpage_ops
= {
509 .need
= need_debug_guardpage
,
510 .init
= init_debug_guardpage
,
513 static int __init
debug_guardpage_minorder_setup(char *buf
)
517 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
518 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
521 _debug_guardpage_minorder
= res
;
522 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
525 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
527 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
528 unsigned int order
, int migratetype
)
530 struct page_ext
*page_ext
;
532 if (!debug_guardpage_enabled())
535 page_ext
= lookup_page_ext(page
);
536 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
538 INIT_LIST_HEAD(&page
->lru
);
539 set_page_private(page
, order
);
540 /* Guard pages are not available for any usage */
541 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
544 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
545 unsigned int order
, int migratetype
)
547 struct page_ext
*page_ext
;
549 if (!debug_guardpage_enabled())
552 page_ext
= lookup_page_ext(page
);
553 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
555 set_page_private(page
, 0);
556 if (!is_migrate_isolate(migratetype
))
557 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
560 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
561 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
562 unsigned int order
, int migratetype
) {}
563 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
564 unsigned int order
, int migratetype
) {}
567 static inline void set_page_order(struct page
*page
, unsigned int order
)
569 set_page_private(page
, order
);
570 __SetPageBuddy(page
);
573 static inline void rmv_page_order(struct page
*page
)
575 __ClearPageBuddy(page
);
576 set_page_private(page
, 0);
580 * This function checks whether a page is free && is the buddy
581 * we can do coalesce a page and its buddy if
582 * (a) the buddy is not in a hole &&
583 * (b) the buddy is in the buddy system &&
584 * (c) a page and its buddy have the same order &&
585 * (d) a page and its buddy are in the same zone.
587 * For recording whether a page is in the buddy system, we set ->_mapcount
588 * PAGE_BUDDY_MAPCOUNT_VALUE.
589 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
590 * serialized by zone->lock.
592 * For recording page's order, we use page_private(page).
594 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
597 if (!pfn_valid_within(page_to_pfn(buddy
)))
600 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
601 if (page_zone_id(page
) != page_zone_id(buddy
))
604 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
609 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
611 * zone check is done late to avoid uselessly
612 * calculating zone/node ids for pages that could
615 if (page_zone_id(page
) != page_zone_id(buddy
))
618 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
626 * Freeing function for a buddy system allocator.
628 * The concept of a buddy system is to maintain direct-mapped table
629 * (containing bit values) for memory blocks of various "orders".
630 * The bottom level table contains the map for the smallest allocatable
631 * units of memory (here, pages), and each level above it describes
632 * pairs of units from the levels below, hence, "buddies".
633 * At a high level, all that happens here is marking the table entry
634 * at the bottom level available, and propagating the changes upward
635 * as necessary, plus some accounting needed to play nicely with other
636 * parts of the VM system.
637 * At each level, we keep a list of pages, which are heads of continuous
638 * free pages of length of (1 << order) and marked with _mapcount
639 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
641 * So when we are allocating or freeing one, we can derive the state of the
642 * other. That is, if we allocate a small block, and both were
643 * free, the remainder of the region must be split into blocks.
644 * If a block is freed, and its buddy is also free, then this
645 * triggers coalescing into a block of larger size.
650 static inline void __free_one_page(struct page
*page
,
652 struct zone
*zone
, unsigned int order
,
655 unsigned long page_idx
;
656 unsigned long combined_idx
;
657 unsigned long uninitialized_var(buddy_idx
);
659 int max_order
= MAX_ORDER
;
661 VM_BUG_ON(!zone_is_initialized(zone
));
662 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
664 VM_BUG_ON(migratetype
== -1);
665 if (is_migrate_isolate(migratetype
)) {
667 * We restrict max order of merging to prevent merge
668 * between freepages on isolate pageblock and normal
669 * pageblock. Without this, pageblock isolation
670 * could cause incorrect freepage accounting.
672 max_order
= min(MAX_ORDER
, pageblock_order
+ 1);
674 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
677 page_idx
= pfn
& ((1 << max_order
) - 1);
679 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
680 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
682 while (order
< max_order
- 1) {
683 buddy_idx
= __find_buddy_index(page_idx
, order
);
684 buddy
= page
+ (buddy_idx
- page_idx
);
685 if (!page_is_buddy(page
, buddy
, order
))
688 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
689 * merge with it and move up one order.
691 if (page_is_guard(buddy
)) {
692 clear_page_guard(zone
, buddy
, order
, migratetype
);
694 list_del(&buddy
->lru
);
695 zone
->free_area
[order
].nr_free
--;
696 rmv_page_order(buddy
);
698 combined_idx
= buddy_idx
& page_idx
;
699 page
= page
+ (combined_idx
- page_idx
);
700 page_idx
= combined_idx
;
703 set_page_order(page
, order
);
706 * If this is not the largest possible page, check if the buddy
707 * of the next-highest order is free. If it is, it's possible
708 * that pages are being freed that will coalesce soon. In case,
709 * that is happening, add the free page to the tail of the list
710 * so it's less likely to be used soon and more likely to be merged
711 * as a higher order page
713 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
714 struct page
*higher_page
, *higher_buddy
;
715 combined_idx
= buddy_idx
& page_idx
;
716 higher_page
= page
+ (combined_idx
- page_idx
);
717 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
718 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
719 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
720 list_add_tail(&page
->lru
,
721 &zone
->free_area
[order
].free_list
[migratetype
]);
726 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
728 zone
->free_area
[order
].nr_free
++;
731 static inline int free_pages_check(struct page
*page
)
733 const char *bad_reason
= NULL
;
734 unsigned long bad_flags
= 0;
736 if (unlikely(page_mapcount(page
)))
737 bad_reason
= "nonzero mapcount";
738 if (unlikely(page
->mapping
!= NULL
))
739 bad_reason
= "non-NULL mapping";
740 if (unlikely(atomic_read(&page
->_count
) != 0))
741 bad_reason
= "nonzero _count";
742 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
743 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
744 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
747 if (unlikely(page
->mem_cgroup
))
748 bad_reason
= "page still charged to cgroup";
750 if (unlikely(bad_reason
)) {
751 bad_page(page
, bad_reason
, bad_flags
);
754 page_cpupid_reset_last(page
);
755 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
756 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
761 * Frees a number of pages from the PCP lists
762 * Assumes all pages on list are in same zone, and of same order.
763 * count is the number of pages to free.
765 * If the zone was previously in an "all pages pinned" state then look to
766 * see if this freeing clears that state.
768 * And clear the zone's pages_scanned counter, to hold off the "all pages are
769 * pinned" detection logic.
771 static void free_pcppages_bulk(struct zone
*zone
, int count
,
772 struct per_cpu_pages
*pcp
)
777 unsigned long nr_scanned
;
779 spin_lock(&zone
->lock
);
780 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
782 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
786 struct list_head
*list
;
789 * Remove pages from lists in a round-robin fashion. A
790 * batch_free count is maintained that is incremented when an
791 * empty list is encountered. This is so more pages are freed
792 * off fuller lists instead of spinning excessively around empty
797 if (++migratetype
== MIGRATE_PCPTYPES
)
799 list
= &pcp
->lists
[migratetype
];
800 } while (list_empty(list
));
802 /* This is the only non-empty list. Free them all. */
803 if (batch_free
== MIGRATE_PCPTYPES
)
804 batch_free
= to_free
;
807 int mt
; /* migratetype of the to-be-freed page */
809 page
= list_entry(list
->prev
, struct page
, lru
);
810 /* must delete as __free_one_page list manipulates */
811 list_del(&page
->lru
);
813 mt
= get_pcppage_migratetype(page
);
814 /* MIGRATE_ISOLATE page should not go to pcplists */
815 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
816 /* Pageblock could have been isolated meanwhile */
817 if (unlikely(has_isolate_pageblock(zone
)))
818 mt
= get_pageblock_migratetype(page
);
820 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
821 trace_mm_page_pcpu_drain(page
, 0, mt
);
822 } while (--to_free
&& --batch_free
&& !list_empty(list
));
824 spin_unlock(&zone
->lock
);
827 static void free_one_page(struct zone
*zone
,
828 struct page
*page
, unsigned long pfn
,
832 unsigned long nr_scanned
;
833 spin_lock(&zone
->lock
);
834 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
836 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
838 if (unlikely(has_isolate_pageblock(zone
) ||
839 is_migrate_isolate(migratetype
))) {
840 migratetype
= get_pfnblock_migratetype(page
, pfn
);
842 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
843 spin_unlock(&zone
->lock
);
846 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
848 if (!IS_ENABLED(CONFIG_DEBUG_VM
))
850 if (unlikely(!PageTail(page
))) {
851 bad_page(page
, "PageTail not set", 0);
854 if (unlikely(page
->first_page
!= head_page
)) {
855 bad_page(page
, "first_page not consistent", 0);
861 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
862 unsigned long zone
, int nid
)
864 set_page_links(page
, zone
, nid
, pfn
);
865 init_page_count(page
);
866 page_mapcount_reset(page
);
867 page_cpupid_reset_last(page
);
869 INIT_LIST_HEAD(&page
->lru
);
870 #ifdef WANT_PAGE_VIRTUAL
871 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
872 if (!is_highmem_idx(zone
))
873 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
877 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
880 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
883 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
884 static void init_reserved_page(unsigned long pfn
)
889 if (!early_page_uninitialised(pfn
))
892 nid
= early_pfn_to_nid(pfn
);
893 pgdat
= NODE_DATA(nid
);
895 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
896 struct zone
*zone
= &pgdat
->node_zones
[zid
];
898 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
901 __init_single_pfn(pfn
, zid
, nid
);
904 static inline void init_reserved_page(unsigned long pfn
)
907 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
910 * Initialised pages do not have PageReserved set. This function is
911 * called for each range allocated by the bootmem allocator and
912 * marks the pages PageReserved. The remaining valid pages are later
913 * sent to the buddy page allocator.
915 void __meminit
reserve_bootmem_region(unsigned long start
, unsigned long end
)
917 unsigned long start_pfn
= PFN_DOWN(start
);
918 unsigned long end_pfn
= PFN_UP(end
);
920 for (; start_pfn
< end_pfn
; start_pfn
++) {
921 if (pfn_valid(start_pfn
)) {
922 struct page
*page
= pfn_to_page(start_pfn
);
924 init_reserved_page(start_pfn
);
925 SetPageReserved(page
);
930 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
932 bool compound
= PageCompound(page
);
935 VM_BUG_ON_PAGE(PageTail(page
), page
);
936 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
938 trace_mm_page_free(page
, order
);
939 kmemcheck_free_shadow(page
, order
);
940 kasan_free_pages(page
, order
);
943 page
->mapping
= NULL
;
944 bad
+= free_pages_check(page
);
945 for (i
= 1; i
< (1 << order
); i
++) {
947 bad
+= free_tail_pages_check(page
, page
+ i
);
948 bad
+= free_pages_check(page
+ i
);
953 reset_page_owner(page
, order
);
955 if (!PageHighMem(page
)) {
956 debug_check_no_locks_freed(page_address(page
),
958 debug_check_no_obj_freed(page_address(page
),
961 arch_free_page(page
, order
);
962 kernel_map_pages(page
, 1 << order
, 0);
967 static void __free_pages_ok(struct page
*page
, unsigned int order
)
971 unsigned long pfn
= page_to_pfn(page
);
973 if (!free_pages_prepare(page
, order
))
976 migratetype
= get_pfnblock_migratetype(page
, pfn
);
977 local_irq_save(flags
);
978 __count_vm_events(PGFREE
, 1 << order
);
979 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
980 local_irq_restore(flags
);
983 static void __init
__free_pages_boot_core(struct page
*page
,
984 unsigned long pfn
, unsigned int order
)
986 unsigned int nr_pages
= 1 << order
;
987 struct page
*p
= page
;
991 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
993 __ClearPageReserved(p
);
994 set_page_count(p
, 0);
996 __ClearPageReserved(p
);
997 set_page_count(p
, 0);
999 page_zone(page
)->managed_pages
+= nr_pages
;
1000 set_page_refcounted(page
);
1001 __free_pages(page
, order
);
1004 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1005 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1007 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1009 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1011 static DEFINE_SPINLOCK(early_pfn_lock
);
1014 spin_lock(&early_pfn_lock
);
1015 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1018 spin_unlock(&early_pfn_lock
);
1024 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1025 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1026 struct mminit_pfnnid_cache
*state
)
1030 nid
= __early_pfn_to_nid(pfn
, state
);
1031 if (nid
>= 0 && nid
!= node
)
1036 /* Only safe to use early in boot when initialisation is single-threaded */
1037 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1039 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1044 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1048 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1049 struct mminit_pfnnid_cache
*state
)
1056 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1059 if (early_page_uninitialised(pfn
))
1061 return __free_pages_boot_core(page
, pfn
, order
);
1064 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1065 static void __init
deferred_free_range(struct page
*page
,
1066 unsigned long pfn
, int nr_pages
)
1073 /* Free a large naturally-aligned chunk if possible */
1074 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1075 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1076 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1077 __free_pages_boot_core(page
, pfn
, MAX_ORDER
-1);
1081 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++)
1082 __free_pages_boot_core(page
, pfn
, 0);
1085 /* Completion tracking for deferred_init_memmap() threads */
1086 static atomic_t pgdat_init_n_undone __initdata
;
1087 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1089 static inline void __init
pgdat_init_report_one_done(void)
1091 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1092 complete(&pgdat_init_all_done_comp
);
1095 /* Initialise remaining memory on a node */
1096 static int __init
deferred_init_memmap(void *data
)
1098 pg_data_t
*pgdat
= data
;
1099 int nid
= pgdat
->node_id
;
1100 struct mminit_pfnnid_cache nid_init_state
= { };
1101 unsigned long start
= jiffies
;
1102 unsigned long nr_pages
= 0;
1103 unsigned long walk_start
, walk_end
;
1106 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1107 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1109 if (first_init_pfn
== ULONG_MAX
) {
1110 pgdat_init_report_one_done();
1114 /* Bind memory initialisation thread to a local node if possible */
1115 if (!cpumask_empty(cpumask
))
1116 set_cpus_allowed_ptr(current
, cpumask
);
1118 /* Sanity check boundaries */
1119 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1120 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1121 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1123 /* Only the highest zone is deferred so find it */
1124 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1125 zone
= pgdat
->node_zones
+ zid
;
1126 if (first_init_pfn
< zone_end_pfn(zone
))
1130 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1131 unsigned long pfn
, end_pfn
;
1132 struct page
*page
= NULL
;
1133 struct page
*free_base_page
= NULL
;
1134 unsigned long free_base_pfn
= 0;
1137 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1138 pfn
= first_init_pfn
;
1139 if (pfn
< walk_start
)
1141 if (pfn
< zone
->zone_start_pfn
)
1142 pfn
= zone
->zone_start_pfn
;
1144 for (; pfn
< end_pfn
; pfn
++) {
1145 if (!pfn_valid_within(pfn
))
1149 * Ensure pfn_valid is checked every
1150 * MAX_ORDER_NR_PAGES for memory holes
1152 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1153 if (!pfn_valid(pfn
)) {
1159 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1164 /* Minimise pfn page lookups and scheduler checks */
1165 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1168 nr_pages
+= nr_to_free
;
1169 deferred_free_range(free_base_page
,
1170 free_base_pfn
, nr_to_free
);
1171 free_base_page
= NULL
;
1172 free_base_pfn
= nr_to_free
= 0;
1174 page
= pfn_to_page(pfn
);
1179 VM_BUG_ON(page_zone(page
) != zone
);
1183 __init_single_page(page
, pfn
, zid
, nid
);
1184 if (!free_base_page
) {
1185 free_base_page
= page
;
1186 free_base_pfn
= pfn
;
1191 /* Where possible, batch up pages for a single free */
1194 /* Free the current block of pages to allocator */
1195 nr_pages
+= nr_to_free
;
1196 deferred_free_range(free_base_page
, free_base_pfn
,
1198 free_base_page
= NULL
;
1199 free_base_pfn
= nr_to_free
= 0;
1202 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1205 /* Sanity check that the next zone really is unpopulated */
1206 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1208 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1209 jiffies_to_msecs(jiffies
- start
));
1211 pgdat_init_report_one_done();
1215 void __init
page_alloc_init_late(void)
1219 /* There will be num_node_state(N_MEMORY) threads */
1220 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1221 for_each_node_state(nid
, N_MEMORY
) {
1222 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1225 /* Block until all are initialised */
1226 wait_for_completion(&pgdat_init_all_done_comp
);
1228 /* Reinit limits that are based on free pages after the kernel is up */
1229 files_maxfiles_init();
1231 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1234 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1235 void __init
init_cma_reserved_pageblock(struct page
*page
)
1237 unsigned i
= pageblock_nr_pages
;
1238 struct page
*p
= page
;
1241 __ClearPageReserved(p
);
1242 set_page_count(p
, 0);
1245 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1247 if (pageblock_order
>= MAX_ORDER
) {
1248 i
= pageblock_nr_pages
;
1251 set_page_refcounted(p
);
1252 __free_pages(p
, MAX_ORDER
- 1);
1253 p
+= MAX_ORDER_NR_PAGES
;
1254 } while (i
-= MAX_ORDER_NR_PAGES
);
1256 set_page_refcounted(page
);
1257 __free_pages(page
, pageblock_order
);
1260 adjust_managed_page_count(page
, pageblock_nr_pages
);
1265 * The order of subdivision here is critical for the IO subsystem.
1266 * Please do not alter this order without good reasons and regression
1267 * testing. Specifically, as large blocks of memory are subdivided,
1268 * the order in which smaller blocks are delivered depends on the order
1269 * they're subdivided in this function. This is the primary factor
1270 * influencing the order in which pages are delivered to the IO
1271 * subsystem according to empirical testing, and this is also justified
1272 * by considering the behavior of a buddy system containing a single
1273 * large block of memory acted on by a series of small allocations.
1274 * This behavior is a critical factor in sglist merging's success.
1278 static inline void expand(struct zone
*zone
, struct page
*page
,
1279 int low
, int high
, struct free_area
*area
,
1282 unsigned long size
= 1 << high
;
1284 while (high
> low
) {
1288 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1290 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1291 debug_guardpage_enabled() &&
1292 high
< debug_guardpage_minorder()) {
1294 * Mark as guard pages (or page), that will allow to
1295 * merge back to allocator when buddy will be freed.
1296 * Corresponding page table entries will not be touched,
1297 * pages will stay not present in virtual address space
1299 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1302 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1304 set_page_order(&page
[size
], high
);
1309 * This page is about to be returned from the page allocator
1311 static inline int check_new_page(struct page
*page
)
1313 const char *bad_reason
= NULL
;
1314 unsigned long bad_flags
= 0;
1316 if (unlikely(page_mapcount(page
)))
1317 bad_reason
= "nonzero mapcount";
1318 if (unlikely(page
->mapping
!= NULL
))
1319 bad_reason
= "non-NULL mapping";
1320 if (unlikely(atomic_read(&page
->_count
) != 0))
1321 bad_reason
= "nonzero _count";
1322 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1323 bad_reason
= "HWPoisoned (hardware-corrupted)";
1324 bad_flags
= __PG_HWPOISON
;
1326 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1327 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1328 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1331 if (unlikely(page
->mem_cgroup
))
1332 bad_reason
= "page still charged to cgroup";
1334 if (unlikely(bad_reason
)) {
1335 bad_page(page
, bad_reason
, bad_flags
);
1341 static int prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1346 for (i
= 0; i
< (1 << order
); i
++) {
1347 struct page
*p
= page
+ i
;
1348 if (unlikely(check_new_page(p
)))
1352 set_page_private(page
, 0);
1353 set_page_refcounted(page
);
1355 arch_alloc_page(page
, order
);
1356 kernel_map_pages(page
, 1 << order
, 1);
1357 kasan_alloc_pages(page
, order
);
1359 if (gfp_flags
& __GFP_ZERO
)
1360 for (i
= 0; i
< (1 << order
); i
++)
1361 clear_highpage(page
+ i
);
1363 if (order
&& (gfp_flags
& __GFP_COMP
))
1364 prep_compound_page(page
, order
);
1366 set_page_owner(page
, order
, gfp_flags
);
1369 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1370 * allocate the page. The expectation is that the caller is taking
1371 * steps that will free more memory. The caller should avoid the page
1372 * being used for !PFMEMALLOC purposes.
1374 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1375 set_page_pfmemalloc(page
);
1377 clear_page_pfmemalloc(page
);
1383 * Go through the free lists for the given migratetype and remove
1384 * the smallest available page from the freelists
1387 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1390 unsigned int current_order
;
1391 struct free_area
*area
;
1394 /* Find a page of the appropriate size in the preferred list */
1395 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1396 area
= &(zone
->free_area
[current_order
]);
1397 if (list_empty(&area
->free_list
[migratetype
]))
1400 page
= list_entry(area
->free_list
[migratetype
].next
,
1402 list_del(&page
->lru
);
1403 rmv_page_order(page
);
1405 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1406 set_pcppage_migratetype(page
, migratetype
);
1415 * This array describes the order lists are fallen back to when
1416 * the free lists for the desirable migrate type are depleted
1418 static int fallbacks
[MIGRATE_TYPES
][4] = {
1419 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1420 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1421 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1423 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1425 #ifdef CONFIG_MEMORY_ISOLATION
1426 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1431 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1434 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1437 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1438 unsigned int order
) { return NULL
; }
1442 * Move the free pages in a range to the free lists of the requested type.
1443 * Note that start_page and end_pages are not aligned on a pageblock
1444 * boundary. If alignment is required, use move_freepages_block()
1446 int move_freepages(struct zone
*zone
,
1447 struct page
*start_page
, struct page
*end_page
,
1451 unsigned long order
;
1452 int pages_moved
= 0;
1454 #ifndef CONFIG_HOLES_IN_ZONE
1456 * page_zone is not safe to call in this context when
1457 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1458 * anyway as we check zone boundaries in move_freepages_block().
1459 * Remove at a later date when no bug reports exist related to
1460 * grouping pages by mobility
1462 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1465 for (page
= start_page
; page
<= end_page
;) {
1466 /* Make sure we are not inadvertently changing nodes */
1467 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1469 if (!pfn_valid_within(page_to_pfn(page
))) {
1474 if (!PageBuddy(page
)) {
1479 order
= page_order(page
);
1480 list_move(&page
->lru
,
1481 &zone
->free_area
[order
].free_list
[migratetype
]);
1483 pages_moved
+= 1 << order
;
1489 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1492 unsigned long start_pfn
, end_pfn
;
1493 struct page
*start_page
, *end_page
;
1495 start_pfn
= page_to_pfn(page
);
1496 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1497 start_page
= pfn_to_page(start_pfn
);
1498 end_page
= start_page
+ pageblock_nr_pages
- 1;
1499 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1501 /* Do not cross zone boundaries */
1502 if (!zone_spans_pfn(zone
, start_pfn
))
1504 if (!zone_spans_pfn(zone
, end_pfn
))
1507 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1510 static void change_pageblock_range(struct page
*pageblock_page
,
1511 int start_order
, int migratetype
)
1513 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1515 while (nr_pageblocks
--) {
1516 set_pageblock_migratetype(pageblock_page
, migratetype
);
1517 pageblock_page
+= pageblock_nr_pages
;
1522 * When we are falling back to another migratetype during allocation, try to
1523 * steal extra free pages from the same pageblocks to satisfy further
1524 * allocations, instead of polluting multiple pageblocks.
1526 * If we are stealing a relatively large buddy page, it is likely there will
1527 * be more free pages in the pageblock, so try to steal them all. For
1528 * reclaimable and unmovable allocations, we steal regardless of page size,
1529 * as fragmentation caused by those allocations polluting movable pageblocks
1530 * is worse than movable allocations stealing from unmovable and reclaimable
1533 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1536 * Leaving this order check is intended, although there is
1537 * relaxed order check in next check. The reason is that
1538 * we can actually steal whole pageblock if this condition met,
1539 * but, below check doesn't guarantee it and that is just heuristic
1540 * so could be changed anytime.
1542 if (order
>= pageblock_order
)
1545 if (order
>= pageblock_order
/ 2 ||
1546 start_mt
== MIGRATE_RECLAIMABLE
||
1547 start_mt
== MIGRATE_UNMOVABLE
||
1548 page_group_by_mobility_disabled
)
1555 * This function implements actual steal behaviour. If order is large enough,
1556 * we can steal whole pageblock. If not, we first move freepages in this
1557 * pageblock and check whether half of pages are moved or not. If half of
1558 * pages are moved, we can change migratetype of pageblock and permanently
1559 * use it's pages as requested migratetype in the future.
1561 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1564 int current_order
= page_order(page
);
1567 /* Take ownership for orders >= pageblock_order */
1568 if (current_order
>= pageblock_order
) {
1569 change_pageblock_range(page
, current_order
, start_type
);
1573 pages
= move_freepages_block(zone
, page
, start_type
);
1575 /* Claim the whole block if over half of it is free */
1576 if (pages
>= (1 << (pageblock_order
-1)) ||
1577 page_group_by_mobility_disabled
)
1578 set_pageblock_migratetype(page
, start_type
);
1582 * Check whether there is a suitable fallback freepage with requested order.
1583 * If only_stealable is true, this function returns fallback_mt only if
1584 * we can steal other freepages all together. This would help to reduce
1585 * fragmentation due to mixed migratetype pages in one pageblock.
1587 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1588 int migratetype
, bool only_stealable
, bool *can_steal
)
1593 if (area
->nr_free
== 0)
1598 fallback_mt
= fallbacks
[migratetype
][i
];
1599 if (fallback_mt
== MIGRATE_TYPES
)
1602 if (list_empty(&area
->free_list
[fallback_mt
]))
1605 if (can_steal_fallback(order
, migratetype
))
1608 if (!only_stealable
)
1619 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1620 * there are no empty page blocks that contain a page with a suitable order
1622 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
1623 unsigned int alloc_order
)
1626 unsigned long max_managed
, flags
;
1629 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1630 * Check is race-prone but harmless.
1632 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
1633 if (zone
->nr_reserved_highatomic
>= max_managed
)
1636 spin_lock_irqsave(&zone
->lock
, flags
);
1638 /* Recheck the nr_reserved_highatomic limit under the lock */
1639 if (zone
->nr_reserved_highatomic
>= max_managed
)
1643 mt
= get_pageblock_migratetype(page
);
1644 if (mt
!= MIGRATE_HIGHATOMIC
&&
1645 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
1646 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
1647 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
1648 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
1652 spin_unlock_irqrestore(&zone
->lock
, flags
);
1656 * Used when an allocation is about to fail under memory pressure. This
1657 * potentially hurts the reliability of high-order allocations when under
1658 * intense memory pressure but failed atomic allocations should be easier
1659 * to recover from than an OOM.
1661 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
1663 struct zonelist
*zonelist
= ac
->zonelist
;
1664 unsigned long flags
;
1670 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
1672 /* Preserve at least one pageblock */
1673 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
1676 spin_lock_irqsave(&zone
->lock
, flags
);
1677 for (order
= 0; order
< MAX_ORDER
; order
++) {
1678 struct free_area
*area
= &(zone
->free_area
[order
]);
1680 if (list_empty(&area
->free_list
[MIGRATE_HIGHATOMIC
]))
1683 page
= list_entry(area
->free_list
[MIGRATE_HIGHATOMIC
].next
,
1687 * It should never happen but changes to locking could
1688 * inadvertently allow a per-cpu drain to add pages
1689 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1690 * and watch for underflows.
1692 zone
->nr_reserved_highatomic
-= min(pageblock_nr_pages
,
1693 zone
->nr_reserved_highatomic
);
1696 * Convert to ac->migratetype and avoid the normal
1697 * pageblock stealing heuristics. Minimally, the caller
1698 * is doing the work and needs the pages. More
1699 * importantly, if the block was always converted to
1700 * MIGRATE_UNMOVABLE or another type then the number
1701 * of pageblocks that cannot be completely freed
1704 set_pageblock_migratetype(page
, ac
->migratetype
);
1705 move_freepages_block(zone
, page
, ac
->migratetype
);
1706 spin_unlock_irqrestore(&zone
->lock
, flags
);
1709 spin_unlock_irqrestore(&zone
->lock
, flags
);
1713 /* Remove an element from the buddy allocator from the fallback list */
1714 static inline struct page
*
1715 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
1717 struct free_area
*area
;
1718 unsigned int current_order
;
1723 /* Find the largest possible block of pages in the other list */
1724 for (current_order
= MAX_ORDER
-1;
1725 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
1727 area
= &(zone
->free_area
[current_order
]);
1728 fallback_mt
= find_suitable_fallback(area
, current_order
,
1729 start_migratetype
, false, &can_steal
);
1730 if (fallback_mt
== -1)
1733 page
= list_entry(area
->free_list
[fallback_mt
].next
,
1736 steal_suitable_fallback(zone
, page
, start_migratetype
);
1738 /* Remove the page from the freelists */
1740 list_del(&page
->lru
);
1741 rmv_page_order(page
);
1743 expand(zone
, page
, order
, current_order
, area
,
1746 * The pcppage_migratetype may differ from pageblock's
1747 * migratetype depending on the decisions in
1748 * find_suitable_fallback(). This is OK as long as it does not
1749 * differ for MIGRATE_CMA pageblocks. Those can be used as
1750 * fallback only via special __rmqueue_cma_fallback() function
1752 set_pcppage_migratetype(page
, start_migratetype
);
1754 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1755 start_migratetype
, fallback_mt
);
1764 * Do the hard work of removing an element from the buddy allocator.
1765 * Call me with the zone->lock already held.
1767 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1768 int migratetype
, gfp_t gfp_flags
)
1772 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1773 if (unlikely(!page
)) {
1774 if (migratetype
== MIGRATE_MOVABLE
)
1775 page
= __rmqueue_cma_fallback(zone
, order
);
1778 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1781 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1786 * Obtain a specified number of elements from the buddy allocator, all under
1787 * a single hold of the lock, for efficiency. Add them to the supplied list.
1788 * Returns the number of new pages which were placed at *list.
1790 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1791 unsigned long count
, struct list_head
*list
,
1792 int migratetype
, bool cold
)
1796 spin_lock(&zone
->lock
);
1797 for (i
= 0; i
< count
; ++i
) {
1798 struct page
*page
= __rmqueue(zone
, order
, migratetype
, 0);
1799 if (unlikely(page
== NULL
))
1803 * Split buddy pages returned by expand() are received here
1804 * in physical page order. The page is added to the callers and
1805 * list and the list head then moves forward. From the callers
1806 * perspective, the linked list is ordered by page number in
1807 * some conditions. This is useful for IO devices that can
1808 * merge IO requests if the physical pages are ordered
1812 list_add(&page
->lru
, list
);
1814 list_add_tail(&page
->lru
, list
);
1816 if (is_migrate_cma(get_pcppage_migratetype(page
)))
1817 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1820 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1821 spin_unlock(&zone
->lock
);
1827 * Called from the vmstat counter updater to drain pagesets of this
1828 * currently executing processor on remote nodes after they have
1831 * Note that this function must be called with the thread pinned to
1832 * a single processor.
1834 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1836 unsigned long flags
;
1837 int to_drain
, batch
;
1839 local_irq_save(flags
);
1840 batch
= READ_ONCE(pcp
->batch
);
1841 to_drain
= min(pcp
->count
, batch
);
1843 free_pcppages_bulk(zone
, to_drain
, pcp
);
1844 pcp
->count
-= to_drain
;
1846 local_irq_restore(flags
);
1851 * Drain pcplists of the indicated processor and zone.
1853 * The processor must either be the current processor and the
1854 * thread pinned to the current processor or a processor that
1857 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
1859 unsigned long flags
;
1860 struct per_cpu_pageset
*pset
;
1861 struct per_cpu_pages
*pcp
;
1863 local_irq_save(flags
);
1864 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1868 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1871 local_irq_restore(flags
);
1875 * Drain pcplists of all zones on the indicated processor.
1877 * The processor must either be the current processor and the
1878 * thread pinned to the current processor or a processor that
1881 static void drain_pages(unsigned int cpu
)
1885 for_each_populated_zone(zone
) {
1886 drain_pages_zone(cpu
, zone
);
1891 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1893 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1894 * the single zone's pages.
1896 void drain_local_pages(struct zone
*zone
)
1898 int cpu
= smp_processor_id();
1901 drain_pages_zone(cpu
, zone
);
1907 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1909 * When zone parameter is non-NULL, spill just the single zone's pages.
1911 * Note that this code is protected against sending an IPI to an offline
1912 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1913 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1914 * nothing keeps CPUs from showing up after we populated the cpumask and
1915 * before the call to on_each_cpu_mask().
1917 void drain_all_pages(struct zone
*zone
)
1922 * Allocate in the BSS so we wont require allocation in
1923 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1925 static cpumask_t cpus_with_pcps
;
1928 * We don't care about racing with CPU hotplug event
1929 * as offline notification will cause the notified
1930 * cpu to drain that CPU pcps and on_each_cpu_mask
1931 * disables preemption as part of its processing
1933 for_each_online_cpu(cpu
) {
1934 struct per_cpu_pageset
*pcp
;
1936 bool has_pcps
= false;
1939 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1943 for_each_populated_zone(z
) {
1944 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
1945 if (pcp
->pcp
.count
) {
1953 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1955 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1957 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
1961 #ifdef CONFIG_HIBERNATION
1963 void mark_free_pages(struct zone
*zone
)
1965 unsigned long pfn
, max_zone_pfn
;
1966 unsigned long flags
;
1967 unsigned int order
, t
;
1968 struct list_head
*curr
;
1970 if (zone_is_empty(zone
))
1973 spin_lock_irqsave(&zone
->lock
, flags
);
1975 max_zone_pfn
= zone_end_pfn(zone
);
1976 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1977 if (pfn_valid(pfn
)) {
1978 struct page
*page
= pfn_to_page(pfn
);
1980 if (!swsusp_page_is_forbidden(page
))
1981 swsusp_unset_page_free(page
);
1984 for_each_migratetype_order(order
, t
) {
1985 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1988 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1989 for (i
= 0; i
< (1UL << order
); i
++)
1990 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1993 spin_unlock_irqrestore(&zone
->lock
, flags
);
1995 #endif /* CONFIG_PM */
1998 * Free a 0-order page
1999 * cold == true ? free a cold page : free a hot page
2001 void free_hot_cold_page(struct page
*page
, bool cold
)
2003 struct zone
*zone
= page_zone(page
);
2004 struct per_cpu_pages
*pcp
;
2005 unsigned long flags
;
2006 unsigned long pfn
= page_to_pfn(page
);
2009 if (!free_pages_prepare(page
, 0))
2012 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2013 set_pcppage_migratetype(page
, migratetype
);
2014 local_irq_save(flags
);
2015 __count_vm_event(PGFREE
);
2018 * We only track unmovable, reclaimable and movable on pcp lists.
2019 * Free ISOLATE pages back to the allocator because they are being
2020 * offlined but treat RESERVE as movable pages so we can get those
2021 * areas back if necessary. Otherwise, we may have to free
2022 * excessively into the page allocator
2024 if (migratetype
>= MIGRATE_PCPTYPES
) {
2025 if (unlikely(is_migrate_isolate(migratetype
))) {
2026 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2029 migratetype
= MIGRATE_MOVABLE
;
2032 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2034 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2036 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2038 if (pcp
->count
>= pcp
->high
) {
2039 unsigned long batch
= READ_ONCE(pcp
->batch
);
2040 free_pcppages_bulk(zone
, batch
, pcp
);
2041 pcp
->count
-= batch
;
2045 local_irq_restore(flags
);
2049 * Free a list of 0-order pages
2051 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2053 struct page
*page
, *next
;
2055 list_for_each_entry_safe(page
, next
, list
, lru
) {
2056 trace_mm_page_free_batched(page
, cold
);
2057 free_hot_cold_page(page
, cold
);
2062 * split_page takes a non-compound higher-order page, and splits it into
2063 * n (1<<order) sub-pages: page[0..n]
2064 * Each sub-page must be freed individually.
2066 * Note: this is probably too low level an operation for use in drivers.
2067 * Please consult with lkml before using this in your driver.
2069 void split_page(struct page
*page
, unsigned int order
)
2074 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2075 VM_BUG_ON_PAGE(!page_count(page
), page
);
2077 #ifdef CONFIG_KMEMCHECK
2079 * Split shadow pages too, because free(page[0]) would
2080 * otherwise free the whole shadow.
2082 if (kmemcheck_page_is_tracked(page
))
2083 split_page(virt_to_page(page
[0].shadow
), order
);
2086 gfp_mask
= get_page_owner_gfp(page
);
2087 set_page_owner(page
, 0, gfp_mask
);
2088 for (i
= 1; i
< (1 << order
); i
++) {
2089 set_page_refcounted(page
+ i
);
2090 set_page_owner(page
+ i
, 0, gfp_mask
);
2093 EXPORT_SYMBOL_GPL(split_page
);
2095 int __isolate_free_page(struct page
*page
, unsigned int order
)
2097 unsigned long watermark
;
2101 BUG_ON(!PageBuddy(page
));
2103 zone
= page_zone(page
);
2104 mt
= get_pageblock_migratetype(page
);
2106 if (!is_migrate_isolate(mt
)) {
2107 /* Obey watermarks as if the page was being allocated */
2108 watermark
= low_wmark_pages(zone
) + (1 << order
);
2109 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2112 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2115 /* Remove page from free list */
2116 list_del(&page
->lru
);
2117 zone
->free_area
[order
].nr_free
--;
2118 rmv_page_order(page
);
2120 set_page_owner(page
, order
, __GFP_MOVABLE
);
2122 /* Set the pageblock if the isolated page is at least a pageblock */
2123 if (order
>= pageblock_order
- 1) {
2124 struct page
*endpage
= page
+ (1 << order
) - 1;
2125 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2126 int mt
= get_pageblock_migratetype(page
);
2127 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2128 set_pageblock_migratetype(page
,
2134 return 1UL << order
;
2138 * Similar to split_page except the page is already free. As this is only
2139 * being used for migration, the migratetype of the block also changes.
2140 * As this is called with interrupts disabled, the caller is responsible
2141 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2144 * Note: this is probably too low level an operation for use in drivers.
2145 * Please consult with lkml before using this in your driver.
2147 int split_free_page(struct page
*page
)
2152 order
= page_order(page
);
2154 nr_pages
= __isolate_free_page(page
, order
);
2158 /* Split into individual pages */
2159 set_page_refcounted(page
);
2160 split_page(page
, order
);
2165 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2168 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2169 struct zone
*zone
, unsigned int order
,
2170 gfp_t gfp_flags
, int alloc_flags
, int migratetype
)
2172 unsigned long flags
;
2174 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2176 if (likely(order
== 0)) {
2177 struct per_cpu_pages
*pcp
;
2178 struct list_head
*list
;
2180 local_irq_save(flags
);
2181 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2182 list
= &pcp
->lists
[migratetype
];
2183 if (list_empty(list
)) {
2184 pcp
->count
+= rmqueue_bulk(zone
, 0,
2187 if (unlikely(list_empty(list
)))
2192 page
= list_entry(list
->prev
, struct page
, lru
);
2194 page
= list_entry(list
->next
, struct page
, lru
);
2196 list_del(&page
->lru
);
2199 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
2201 * __GFP_NOFAIL is not to be used in new code.
2203 * All __GFP_NOFAIL callers should be fixed so that they
2204 * properly detect and handle allocation failures.
2206 * We most definitely don't want callers attempting to
2207 * allocate greater than order-1 page units with
2210 WARN_ON_ONCE(order
> 1);
2212 spin_lock_irqsave(&zone
->lock
, flags
);
2215 if (alloc_flags
& ALLOC_HARDER
) {
2216 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2218 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2221 page
= __rmqueue(zone
, order
, migratetype
, gfp_flags
);
2222 spin_unlock(&zone
->lock
);
2225 __mod_zone_freepage_state(zone
, -(1 << order
),
2226 get_pcppage_migratetype(page
));
2229 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2230 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2231 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2232 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2234 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2235 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2236 local_irq_restore(flags
);
2238 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2242 local_irq_restore(flags
);
2246 #ifdef CONFIG_FAIL_PAGE_ALLOC
2249 struct fault_attr attr
;
2251 bool ignore_gfp_highmem
;
2252 bool ignore_gfp_reclaim
;
2254 } fail_page_alloc
= {
2255 .attr
= FAULT_ATTR_INITIALIZER
,
2256 .ignore_gfp_reclaim
= true,
2257 .ignore_gfp_highmem
= true,
2261 static int __init
setup_fail_page_alloc(char *str
)
2263 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2265 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2267 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2269 if (order
< fail_page_alloc
.min_order
)
2271 if (gfp_mask
& __GFP_NOFAIL
)
2273 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2275 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2276 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2279 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2282 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2284 static int __init
fail_page_alloc_debugfs(void)
2286 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2289 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2290 &fail_page_alloc
.attr
);
2292 return PTR_ERR(dir
);
2294 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2295 &fail_page_alloc
.ignore_gfp_reclaim
))
2297 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2298 &fail_page_alloc
.ignore_gfp_highmem
))
2300 if (!debugfs_create_u32("min-order", mode
, dir
,
2301 &fail_page_alloc
.min_order
))
2306 debugfs_remove_recursive(dir
);
2311 late_initcall(fail_page_alloc_debugfs
);
2313 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2315 #else /* CONFIG_FAIL_PAGE_ALLOC */
2317 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2322 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2325 * Return true if free base pages are above 'mark'. For high-order checks it
2326 * will return true of the order-0 watermark is reached and there is at least
2327 * one free page of a suitable size. Checking now avoids taking the zone lock
2328 * to check in the allocation paths if no pages are free.
2330 static bool __zone_watermark_ok(struct zone
*z
, unsigned int order
,
2331 unsigned long mark
, int classzone_idx
, int alloc_flags
,
2336 const int alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2338 /* free_pages may go negative - that's OK */
2339 free_pages
-= (1 << order
) - 1;
2341 if (alloc_flags
& ALLOC_HIGH
)
2345 * If the caller does not have rights to ALLOC_HARDER then subtract
2346 * the high-atomic reserves. This will over-estimate the size of the
2347 * atomic reserve but it avoids a search.
2349 if (likely(!alloc_harder
))
2350 free_pages
-= z
->nr_reserved_highatomic
;
2355 /* If allocation can't use CMA areas don't use free CMA pages */
2356 if (!(alloc_flags
& ALLOC_CMA
))
2357 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2361 * Check watermarks for an order-0 allocation request. If these
2362 * are not met, then a high-order request also cannot go ahead
2363 * even if a suitable page happened to be free.
2365 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2368 /* If this is an order-0 request then the watermark is fine */
2372 /* For a high-order request, check at least one suitable page is free */
2373 for (o
= order
; o
< MAX_ORDER
; o
++) {
2374 struct free_area
*area
= &z
->free_area
[o
];
2383 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2384 if (!list_empty(&area
->free_list
[mt
]))
2389 if ((alloc_flags
& ALLOC_CMA
) &&
2390 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2398 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2399 int classzone_idx
, int alloc_flags
)
2401 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2402 zone_page_state(z
, NR_FREE_PAGES
));
2405 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2406 unsigned long mark
, int classzone_idx
)
2408 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2410 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2411 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2413 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2418 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2420 return local_zone
->node
== zone
->node
;
2423 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2425 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2428 #else /* CONFIG_NUMA */
2429 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2434 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2438 #endif /* CONFIG_NUMA */
2440 static void reset_alloc_batches(struct zone
*preferred_zone
)
2442 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2445 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2446 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2447 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2448 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2449 } while (zone
++ != preferred_zone
);
2453 * get_page_from_freelist goes through the zonelist trying to allocate
2456 static struct page
*
2457 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2458 const struct alloc_context
*ac
)
2460 struct zonelist
*zonelist
= ac
->zonelist
;
2462 struct page
*page
= NULL
;
2464 int nr_fair_skipped
= 0;
2465 bool zonelist_rescan
;
2468 zonelist_rescan
= false;
2471 * Scan zonelist, looking for a zone with enough free.
2472 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2474 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2478 if (cpusets_enabled() &&
2479 (alloc_flags
& ALLOC_CPUSET
) &&
2480 !cpuset_zone_allowed(zone
, gfp_mask
))
2483 * Distribute pages in proportion to the individual
2484 * zone size to ensure fair page aging. The zone a
2485 * page was allocated in should have no effect on the
2486 * time the page has in memory before being reclaimed.
2488 if (alloc_flags
& ALLOC_FAIR
) {
2489 if (!zone_local(ac
->preferred_zone
, zone
))
2491 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2497 * When allocating a page cache page for writing, we
2498 * want to get it from a zone that is within its dirty
2499 * limit, such that no single zone holds more than its
2500 * proportional share of globally allowed dirty pages.
2501 * The dirty limits take into account the zone's
2502 * lowmem reserves and high watermark so that kswapd
2503 * should be able to balance it without having to
2504 * write pages from its LRU list.
2506 * This may look like it could increase pressure on
2507 * lower zones by failing allocations in higher zones
2508 * before they are full. But the pages that do spill
2509 * over are limited as the lower zones are protected
2510 * by this very same mechanism. It should not become
2511 * a practical burden to them.
2513 * XXX: For now, allow allocations to potentially
2514 * exceed the per-zone dirty limit in the slowpath
2515 * (spread_dirty_pages unset) before going into reclaim,
2516 * which is important when on a NUMA setup the allowed
2517 * zones are together not big enough to reach the
2518 * global limit. The proper fix for these situations
2519 * will require awareness of zones in the
2520 * dirty-throttling and the flusher threads.
2522 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2525 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2526 if (!zone_watermark_ok(zone
, order
, mark
,
2527 ac
->classzone_idx
, alloc_flags
)) {
2530 /* Checked here to keep the fast path fast */
2531 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2532 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2535 if (zone_reclaim_mode
== 0 ||
2536 !zone_allows_reclaim(ac
->preferred_zone
, zone
))
2539 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2541 case ZONE_RECLAIM_NOSCAN
:
2544 case ZONE_RECLAIM_FULL
:
2545 /* scanned but unreclaimable */
2548 /* did we reclaim enough */
2549 if (zone_watermark_ok(zone
, order
, mark
,
2550 ac
->classzone_idx
, alloc_flags
))
2558 page
= buffered_rmqueue(ac
->preferred_zone
, zone
, order
,
2559 gfp_mask
, alloc_flags
, ac
->migratetype
);
2561 if (prep_new_page(page
, order
, gfp_mask
, alloc_flags
))
2565 * If this is a high-order atomic allocation then check
2566 * if the pageblock should be reserved for the future
2568 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
2569 reserve_highatomic_pageblock(page
, zone
, order
);
2576 * The first pass makes sure allocations are spread fairly within the
2577 * local node. However, the local node might have free pages left
2578 * after the fairness batches are exhausted, and remote zones haven't
2579 * even been considered yet. Try once more without fairness, and
2580 * include remote zones now, before entering the slowpath and waking
2581 * kswapd: prefer spilling to a remote zone over swapping locally.
2583 if (alloc_flags
& ALLOC_FAIR
) {
2584 alloc_flags
&= ~ALLOC_FAIR
;
2585 if (nr_fair_skipped
) {
2586 zonelist_rescan
= true;
2587 reset_alloc_batches(ac
->preferred_zone
);
2589 if (nr_online_nodes
> 1)
2590 zonelist_rescan
= true;
2593 if (zonelist_rescan
)
2600 * Large machines with many possible nodes should not always dump per-node
2601 * meminfo in irq context.
2603 static inline bool should_suppress_show_mem(void)
2608 ret
= in_interrupt();
2613 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2614 DEFAULT_RATELIMIT_INTERVAL
,
2615 DEFAULT_RATELIMIT_BURST
);
2617 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
2619 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2621 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2622 debug_guardpage_minorder() > 0)
2626 * This documents exceptions given to allocations in certain
2627 * contexts that are allowed to allocate outside current's set
2630 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2631 if (test_thread_flag(TIF_MEMDIE
) ||
2632 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2633 filter
&= ~SHOW_MEM_FILTER_NODES
;
2634 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
2635 filter
&= ~SHOW_MEM_FILTER_NODES
;
2638 struct va_format vaf
;
2641 va_start(args
, fmt
);
2646 pr_warn("%pV", &vaf
);
2651 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2652 current
->comm
, order
, gfp_mask
);
2655 if (!should_suppress_show_mem())
2659 static inline struct page
*
2660 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2661 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
2663 struct oom_control oc
= {
2664 .zonelist
= ac
->zonelist
,
2665 .nodemask
= ac
->nodemask
,
2666 .gfp_mask
= gfp_mask
,
2671 *did_some_progress
= 0;
2674 * Acquire the oom lock. If that fails, somebody else is
2675 * making progress for us.
2677 if (!mutex_trylock(&oom_lock
)) {
2678 *did_some_progress
= 1;
2679 schedule_timeout_uninterruptible(1);
2684 * Go through the zonelist yet one more time, keep very high watermark
2685 * here, this is only to catch a parallel oom killing, we must fail if
2686 * we're still under heavy pressure.
2688 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
2689 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
2693 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2694 /* Coredumps can quickly deplete all memory reserves */
2695 if (current
->flags
& PF_DUMPCORE
)
2697 /* The OOM killer will not help higher order allocs */
2698 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2700 /* The OOM killer does not needlessly kill tasks for lowmem */
2701 if (ac
->high_zoneidx
< ZONE_NORMAL
)
2703 /* The OOM killer does not compensate for IO-less reclaim */
2704 if (!(gfp_mask
& __GFP_FS
)) {
2706 * XXX: Page reclaim didn't yield anything,
2707 * and the OOM killer can't be invoked, but
2708 * keep looping as per tradition.
2710 *did_some_progress
= 1;
2713 if (pm_suspended_storage())
2715 /* The OOM killer may not free memory on a specific node */
2716 if (gfp_mask
& __GFP_THISNODE
)
2719 /* Exhausted what can be done so it's blamo time */
2720 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
))
2721 *did_some_progress
= 1;
2723 mutex_unlock(&oom_lock
);
2727 #ifdef CONFIG_COMPACTION
2728 /* Try memory compaction for high-order allocations before reclaim */
2729 static struct page
*
2730 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2731 int alloc_flags
, const struct alloc_context
*ac
,
2732 enum migrate_mode mode
, int *contended_compaction
,
2733 bool *deferred_compaction
)
2735 unsigned long compact_result
;
2741 current
->flags
|= PF_MEMALLOC
;
2742 compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
2743 mode
, contended_compaction
);
2744 current
->flags
&= ~PF_MEMALLOC
;
2746 switch (compact_result
) {
2747 case COMPACT_DEFERRED
:
2748 *deferred_compaction
= true;
2750 case COMPACT_SKIPPED
:
2757 * At least in one zone compaction wasn't deferred or skipped, so let's
2758 * count a compaction stall
2760 count_vm_event(COMPACTSTALL
);
2762 page
= get_page_from_freelist(gfp_mask
, order
,
2763 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2766 struct zone
*zone
= page_zone(page
);
2768 zone
->compact_blockskip_flush
= false;
2769 compaction_defer_reset(zone
, order
, true);
2770 count_vm_event(COMPACTSUCCESS
);
2775 * It's bad if compaction run occurs and fails. The most likely reason
2776 * is that pages exist, but not enough to satisfy watermarks.
2778 count_vm_event(COMPACTFAIL
);
2785 static inline struct page
*
2786 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2787 int alloc_flags
, const struct alloc_context
*ac
,
2788 enum migrate_mode mode
, int *contended_compaction
,
2789 bool *deferred_compaction
)
2793 #endif /* CONFIG_COMPACTION */
2795 /* Perform direct synchronous page reclaim */
2797 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
2798 const struct alloc_context
*ac
)
2800 struct reclaim_state reclaim_state
;
2805 /* We now go into synchronous reclaim */
2806 cpuset_memory_pressure_bump();
2807 current
->flags
|= PF_MEMALLOC
;
2808 lockdep_set_current_reclaim_state(gfp_mask
);
2809 reclaim_state
.reclaimed_slab
= 0;
2810 current
->reclaim_state
= &reclaim_state
;
2812 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
2815 current
->reclaim_state
= NULL
;
2816 lockdep_clear_current_reclaim_state();
2817 current
->flags
&= ~PF_MEMALLOC
;
2824 /* The really slow allocator path where we enter direct reclaim */
2825 static inline struct page
*
2826 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2827 int alloc_flags
, const struct alloc_context
*ac
,
2828 unsigned long *did_some_progress
)
2830 struct page
*page
= NULL
;
2831 bool drained
= false;
2833 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
2834 if (unlikely(!(*did_some_progress
)))
2838 page
= get_page_from_freelist(gfp_mask
, order
,
2839 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2842 * If an allocation failed after direct reclaim, it could be because
2843 * pages are pinned on the per-cpu lists or in high alloc reserves.
2844 * Shrink them them and try again
2846 if (!page
&& !drained
) {
2847 unreserve_highatomic_pageblock(ac
);
2848 drain_all_pages(NULL
);
2857 * This is called in the allocator slow-path if the allocation request is of
2858 * sufficient urgency to ignore watermarks and take other desperate measures
2860 static inline struct page
*
2861 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2862 const struct alloc_context
*ac
)
2867 page
= get_page_from_freelist(gfp_mask
, order
,
2868 ALLOC_NO_WATERMARKS
, ac
);
2870 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2871 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
,
2873 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2878 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
2883 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
2884 ac
->high_zoneidx
, ac
->nodemask
)
2885 wakeup_kswapd(zone
, order
, zone_idx(ac
->preferred_zone
));
2889 gfp_to_alloc_flags(gfp_t gfp_mask
)
2891 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2893 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2894 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2897 * The caller may dip into page reserves a bit more if the caller
2898 * cannot run direct reclaim, or if the caller has realtime scheduling
2899 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2900 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
2902 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2904 if (gfp_mask
& __GFP_ATOMIC
) {
2906 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2907 * if it can't schedule.
2909 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2910 alloc_flags
|= ALLOC_HARDER
;
2912 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2913 * comment for __cpuset_node_allowed().
2915 alloc_flags
&= ~ALLOC_CPUSET
;
2916 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2917 alloc_flags
|= ALLOC_HARDER
;
2919 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2920 if (gfp_mask
& __GFP_MEMALLOC
)
2921 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2922 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2923 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2924 else if (!in_interrupt() &&
2925 ((current
->flags
& PF_MEMALLOC
) ||
2926 unlikely(test_thread_flag(TIF_MEMDIE
))))
2927 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2930 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2931 alloc_flags
|= ALLOC_CMA
;
2936 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2938 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2941 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
2943 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
2946 static inline struct page
*
2947 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2948 struct alloc_context
*ac
)
2950 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
2951 struct page
*page
= NULL
;
2953 unsigned long pages_reclaimed
= 0;
2954 unsigned long did_some_progress
;
2955 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
2956 bool deferred_compaction
= false;
2957 int contended_compaction
= COMPACT_CONTENDED_NONE
;
2960 * In the slowpath, we sanity check order to avoid ever trying to
2961 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2962 * be using allocators in order of preference for an area that is
2965 if (order
>= MAX_ORDER
) {
2966 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2971 * We also sanity check to catch abuse of atomic reserves being used by
2972 * callers that are not in atomic context.
2974 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
2975 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
2976 gfp_mask
&= ~__GFP_ATOMIC
;
2979 * If this allocation cannot block and it is for a specific node, then
2980 * fail early. There's no need to wakeup kswapd or retry for a
2981 * speculative node-specific allocation.
2983 if (IS_ENABLED(CONFIG_NUMA
) && (gfp_mask
& __GFP_THISNODE
) && !can_direct_reclaim
)
2987 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
2988 wake_all_kswapds(order
, ac
);
2991 * OK, we're below the kswapd watermark and have kicked background
2992 * reclaim. Now things get more complex, so set up alloc_flags according
2993 * to how we want to proceed.
2995 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2998 * Find the true preferred zone if the allocation is unconstrained by
3001 if (!(alloc_flags
& ALLOC_CPUSET
) && !ac
->nodemask
) {
3002 struct zoneref
*preferred_zoneref
;
3003 preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3004 ac
->high_zoneidx
, NULL
, &ac
->preferred_zone
);
3005 ac
->classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3008 /* This is the last chance, in general, before the goto nopage. */
3009 page
= get_page_from_freelist(gfp_mask
, order
,
3010 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3014 /* Allocate without watermarks if the context allows */
3015 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3017 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3018 * the allocation is high priority and these type of
3019 * allocations are system rather than user orientated
3021 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3023 page
= __alloc_pages_high_priority(gfp_mask
, order
, ac
);
3030 /* Caller is not willing to reclaim, we can't balance anything */
3031 if (!can_direct_reclaim
) {
3033 * All existing users of the deprecated __GFP_NOFAIL are
3034 * blockable, so warn of any new users that actually allow this
3035 * type of allocation to fail.
3037 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3041 /* Avoid recursion of direct reclaim */
3042 if (current
->flags
& PF_MEMALLOC
)
3045 /* Avoid allocations with no watermarks from looping endlessly */
3046 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3050 * Try direct compaction. The first pass is asynchronous. Subsequent
3051 * attempts after direct reclaim are synchronous
3053 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3055 &contended_compaction
,
3056 &deferred_compaction
);
3060 /* Checks for THP-specific high-order allocations */
3061 if (is_thp_gfp_mask(gfp_mask
)) {
3063 * If compaction is deferred for high-order allocations, it is
3064 * because sync compaction recently failed. If this is the case
3065 * and the caller requested a THP allocation, we do not want
3066 * to heavily disrupt the system, so we fail the allocation
3067 * instead of entering direct reclaim.
3069 if (deferred_compaction
)
3073 * In all zones where compaction was attempted (and not
3074 * deferred or skipped), lock contention has been detected.
3075 * For THP allocation we do not want to disrupt the others
3076 * so we fallback to base pages instead.
3078 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3082 * If compaction was aborted due to need_resched(), we do not
3083 * want to further increase allocation latency, unless it is
3084 * khugepaged trying to collapse.
3086 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3087 && !(current
->flags
& PF_KTHREAD
))
3092 * It can become very expensive to allocate transparent hugepages at
3093 * fault, so use asynchronous memory compaction for THP unless it is
3094 * khugepaged trying to collapse.
3096 if (!is_thp_gfp_mask(gfp_mask
) || (current
->flags
& PF_KTHREAD
))
3097 migration_mode
= MIGRATE_SYNC_LIGHT
;
3099 /* Try direct reclaim and then allocating */
3100 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3101 &did_some_progress
);
3105 /* Do not loop if specifically requested */
3106 if (gfp_mask
& __GFP_NORETRY
)
3109 /* Keep reclaiming pages as long as there is reasonable progress */
3110 pages_reclaimed
+= did_some_progress
;
3111 if ((did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
3112 ((gfp_mask
& __GFP_REPEAT
) && pages_reclaimed
< (1 << order
))) {
3113 /* Wait for some write requests to complete then retry */
3114 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
3118 /* Reclaim has failed us, start killing things */
3119 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3123 /* Retry as long as the OOM killer is making progress */
3124 if (did_some_progress
)
3129 * High-order allocations do not necessarily loop after
3130 * direct reclaim and reclaim/compaction depends on compaction
3131 * being called after reclaim so call directly if necessary
3133 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3135 &contended_compaction
,
3136 &deferred_compaction
);
3140 warn_alloc_failed(gfp_mask
, order
, NULL
);
3146 * This is the 'heart' of the zoned buddy allocator.
3149 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3150 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3152 struct zoneref
*preferred_zoneref
;
3153 struct page
*page
= NULL
;
3154 unsigned int cpuset_mems_cookie
;
3155 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
|ALLOC_FAIR
;
3156 gfp_t alloc_mask
; /* The gfp_t that was actually used for allocation */
3157 struct alloc_context ac
= {
3158 .high_zoneidx
= gfp_zone(gfp_mask
),
3159 .nodemask
= nodemask
,
3160 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3163 gfp_mask
&= gfp_allowed_mask
;
3165 lockdep_trace_alloc(gfp_mask
);
3167 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3169 if (should_fail_alloc_page(gfp_mask
, order
))
3173 * Check the zones suitable for the gfp_mask contain at least one
3174 * valid zone. It's possible to have an empty zonelist as a result
3175 * of __GFP_THISNODE and a memoryless node
3177 if (unlikely(!zonelist
->_zonerefs
->zone
))
3180 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3181 alloc_flags
|= ALLOC_CMA
;
3184 cpuset_mems_cookie
= read_mems_allowed_begin();
3186 /* We set it here, as __alloc_pages_slowpath might have changed it */
3187 ac
.zonelist
= zonelist
;
3189 /* Dirty zone balancing only done in the fast path */
3190 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3192 /* The preferred zone is used for statistics later */
3193 preferred_zoneref
= first_zones_zonelist(ac
.zonelist
, ac
.high_zoneidx
,
3194 ac
.nodemask
? : &cpuset_current_mems_allowed
,
3195 &ac
.preferred_zone
);
3196 if (!ac
.preferred_zone
)
3198 ac
.classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3200 /* First allocation attempt */
3201 alloc_mask
= gfp_mask
|__GFP_HARDWALL
;
3202 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3203 if (unlikely(!page
)) {
3205 * Runtime PM, block IO and its error handling path
3206 * can deadlock because I/O on the device might not
3209 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3210 ac
.spread_dirty_pages
= false;
3212 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3215 if (kmemcheck_enabled
&& page
)
3216 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3218 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3222 * When updating a task's mems_allowed, it is possible to race with
3223 * parallel threads in such a way that an allocation can fail while
3224 * the mask is being updated. If a page allocation is about to fail,
3225 * check if the cpuset changed during allocation and if so, retry.
3227 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
)))
3232 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3235 * Common helper functions.
3237 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3242 * __get_free_pages() returns a 32-bit address, which cannot represent
3245 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3247 page
= alloc_pages(gfp_mask
, order
);
3250 return (unsigned long) page_address(page
);
3252 EXPORT_SYMBOL(__get_free_pages
);
3254 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3256 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3258 EXPORT_SYMBOL(get_zeroed_page
);
3260 void __free_pages(struct page
*page
, unsigned int order
)
3262 if (put_page_testzero(page
)) {
3264 free_hot_cold_page(page
, false);
3266 __free_pages_ok(page
, order
);
3270 EXPORT_SYMBOL(__free_pages
);
3272 void free_pages(unsigned long addr
, unsigned int order
)
3275 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3276 __free_pages(virt_to_page((void *)addr
), order
);
3280 EXPORT_SYMBOL(free_pages
);
3284 * An arbitrary-length arbitrary-offset area of memory which resides
3285 * within a 0 or higher order page. Multiple fragments within that page
3286 * are individually refcounted, in the page's reference counter.
3288 * The page_frag functions below provide a simple allocation framework for
3289 * page fragments. This is used by the network stack and network device
3290 * drivers to provide a backing region of memory for use as either an
3291 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3293 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3296 struct page
*page
= NULL
;
3297 gfp_t gfp
= gfp_mask
;
3299 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3300 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3302 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3303 PAGE_FRAG_CACHE_MAX_ORDER
);
3304 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3306 if (unlikely(!page
))
3307 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3309 nc
->va
= page
? page_address(page
) : NULL
;
3314 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3315 unsigned int fragsz
, gfp_t gfp_mask
)
3317 unsigned int size
= PAGE_SIZE
;
3321 if (unlikely(!nc
->va
)) {
3323 page
= __page_frag_refill(nc
, gfp_mask
);
3327 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3328 /* if size can vary use size else just use PAGE_SIZE */
3331 /* Even if we own the page, we do not use atomic_set().
3332 * This would break get_page_unless_zero() users.
3334 atomic_add(size
- 1, &page
->_count
);
3336 /* reset page count bias and offset to start of new frag */
3337 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3338 nc
->pagecnt_bias
= size
;
3342 offset
= nc
->offset
- fragsz
;
3343 if (unlikely(offset
< 0)) {
3344 page
= virt_to_page(nc
->va
);
3346 if (!atomic_sub_and_test(nc
->pagecnt_bias
, &page
->_count
))
3349 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3350 /* if size can vary use size else just use PAGE_SIZE */
3353 /* OK, page count is 0, we can safely set it */
3354 atomic_set(&page
->_count
, size
);
3356 /* reset page count bias and offset to start of new frag */
3357 nc
->pagecnt_bias
= size
;
3358 offset
= size
- fragsz
;
3362 nc
->offset
= offset
;
3364 return nc
->va
+ offset
;
3366 EXPORT_SYMBOL(__alloc_page_frag
);
3369 * Frees a page fragment allocated out of either a compound or order 0 page.
3371 void __free_page_frag(void *addr
)
3373 struct page
*page
= virt_to_head_page(addr
);
3375 if (unlikely(put_page_testzero(page
)))
3376 __free_pages_ok(page
, compound_order(page
));
3378 EXPORT_SYMBOL(__free_page_frag
);
3381 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3382 * of the current memory cgroup.
3384 * It should be used when the caller would like to use kmalloc, but since the
3385 * allocation is large, it has to fall back to the page allocator.
3387 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
3391 page
= alloc_pages(gfp_mask
, order
);
3392 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3393 __free_pages(page
, order
);
3399 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
3403 page
= alloc_pages_node(nid
, gfp_mask
, order
);
3404 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3405 __free_pages(page
, order
);
3412 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3415 void __free_kmem_pages(struct page
*page
, unsigned int order
)
3417 memcg_kmem_uncharge(page
, order
);
3418 __free_pages(page
, order
);
3421 void free_kmem_pages(unsigned long addr
, unsigned int order
)
3424 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3425 __free_kmem_pages(virt_to_page((void *)addr
), order
);
3429 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
3432 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
3433 unsigned long used
= addr
+ PAGE_ALIGN(size
);
3435 split_page(virt_to_page((void *)addr
), order
);
3436 while (used
< alloc_end
) {
3441 return (void *)addr
;
3445 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3446 * @size: the number of bytes to allocate
3447 * @gfp_mask: GFP flags for the allocation
3449 * This function is similar to alloc_pages(), except that it allocates the
3450 * minimum number of pages to satisfy the request. alloc_pages() can only
3451 * allocate memory in power-of-two pages.
3453 * This function is also limited by MAX_ORDER.
3455 * Memory allocated by this function must be released by free_pages_exact().
3457 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
3459 unsigned int order
= get_order(size
);
3462 addr
= __get_free_pages(gfp_mask
, order
);
3463 return make_alloc_exact(addr
, order
, size
);
3465 EXPORT_SYMBOL(alloc_pages_exact
);
3468 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3470 * @nid: the preferred node ID where memory should be allocated
3471 * @size: the number of bytes to allocate
3472 * @gfp_mask: GFP flags for the allocation
3474 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3477 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
3479 unsigned order
= get_order(size
);
3480 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
3483 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
3487 * free_pages_exact - release memory allocated via alloc_pages_exact()
3488 * @virt: the value returned by alloc_pages_exact.
3489 * @size: size of allocation, same value as passed to alloc_pages_exact().
3491 * Release the memory allocated by a previous call to alloc_pages_exact.
3493 void free_pages_exact(void *virt
, size_t size
)
3495 unsigned long addr
= (unsigned long)virt
;
3496 unsigned long end
= addr
+ PAGE_ALIGN(size
);
3498 while (addr
< end
) {
3503 EXPORT_SYMBOL(free_pages_exact
);
3506 * nr_free_zone_pages - count number of pages beyond high watermark
3507 * @offset: The zone index of the highest zone
3509 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3510 * high watermark within all zones at or below a given zone index. For each
3511 * zone, the number of pages is calculated as:
3512 * managed_pages - high_pages
3514 static unsigned long nr_free_zone_pages(int offset
)
3519 /* Just pick one node, since fallback list is circular */
3520 unsigned long sum
= 0;
3522 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
3524 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
3525 unsigned long size
= zone
->managed_pages
;
3526 unsigned long high
= high_wmark_pages(zone
);
3535 * nr_free_buffer_pages - count number of pages beyond high watermark
3537 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3538 * watermark within ZONE_DMA and ZONE_NORMAL.
3540 unsigned long nr_free_buffer_pages(void)
3542 return nr_free_zone_pages(gfp_zone(GFP_USER
));
3544 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
3547 * nr_free_pagecache_pages - count number of pages beyond high watermark
3549 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3550 * high watermark within all zones.
3552 unsigned long nr_free_pagecache_pages(void)
3554 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
3557 static inline void show_node(struct zone
*zone
)
3559 if (IS_ENABLED(CONFIG_NUMA
))
3560 printk("Node %d ", zone_to_nid(zone
));
3563 void si_meminfo(struct sysinfo
*val
)
3565 val
->totalram
= totalram_pages
;
3566 val
->sharedram
= global_page_state(NR_SHMEM
);
3567 val
->freeram
= global_page_state(NR_FREE_PAGES
);
3568 val
->bufferram
= nr_blockdev_pages();
3569 val
->totalhigh
= totalhigh_pages
;
3570 val
->freehigh
= nr_free_highpages();
3571 val
->mem_unit
= PAGE_SIZE
;
3574 EXPORT_SYMBOL(si_meminfo
);
3577 void si_meminfo_node(struct sysinfo
*val
, int nid
)
3579 int zone_type
; /* needs to be signed */
3580 unsigned long managed_pages
= 0;
3581 pg_data_t
*pgdat
= NODE_DATA(nid
);
3583 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
3584 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
3585 val
->totalram
= managed_pages
;
3586 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
3587 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
3588 #ifdef CONFIG_HIGHMEM
3589 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
3590 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
3596 val
->mem_unit
= PAGE_SIZE
;
3601 * Determine whether the node should be displayed or not, depending on whether
3602 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3604 bool skip_free_areas_node(unsigned int flags
, int nid
)
3607 unsigned int cpuset_mems_cookie
;
3609 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3613 cpuset_mems_cookie
= read_mems_allowed_begin();
3614 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3615 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
3620 #define K(x) ((x) << (PAGE_SHIFT-10))
3622 static void show_migration_types(unsigned char type
)
3624 static const char types
[MIGRATE_TYPES
] = {
3625 [MIGRATE_UNMOVABLE
] = 'U',
3626 [MIGRATE_RECLAIMABLE
] = 'E',
3627 [MIGRATE_MOVABLE
] = 'M',
3629 [MIGRATE_CMA
] = 'C',
3631 #ifdef CONFIG_MEMORY_ISOLATION
3632 [MIGRATE_ISOLATE
] = 'I',
3635 char tmp
[MIGRATE_TYPES
+ 1];
3639 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3640 if (type
& (1 << i
))
3645 printk("(%s) ", tmp
);
3649 * Show free area list (used inside shift_scroll-lock stuff)
3650 * We also calculate the percentage fragmentation. We do this by counting the
3651 * memory on each free list with the exception of the first item on the list.
3654 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3657 void show_free_areas(unsigned int filter
)
3659 unsigned long free_pcp
= 0;
3663 for_each_populated_zone(zone
) {
3664 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3667 for_each_online_cpu(cpu
)
3668 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3671 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3672 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3673 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3674 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3675 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3676 " free:%lu free_pcp:%lu free_cma:%lu\n",
3677 global_page_state(NR_ACTIVE_ANON
),
3678 global_page_state(NR_INACTIVE_ANON
),
3679 global_page_state(NR_ISOLATED_ANON
),
3680 global_page_state(NR_ACTIVE_FILE
),
3681 global_page_state(NR_INACTIVE_FILE
),
3682 global_page_state(NR_ISOLATED_FILE
),
3683 global_page_state(NR_UNEVICTABLE
),
3684 global_page_state(NR_FILE_DIRTY
),
3685 global_page_state(NR_WRITEBACK
),
3686 global_page_state(NR_UNSTABLE_NFS
),
3687 global_page_state(NR_SLAB_RECLAIMABLE
),
3688 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3689 global_page_state(NR_FILE_MAPPED
),
3690 global_page_state(NR_SHMEM
),
3691 global_page_state(NR_PAGETABLE
),
3692 global_page_state(NR_BOUNCE
),
3693 global_page_state(NR_FREE_PAGES
),
3695 global_page_state(NR_FREE_CMA_PAGES
));
3697 for_each_populated_zone(zone
) {
3700 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3704 for_each_online_cpu(cpu
)
3705 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3713 " active_anon:%lukB"
3714 " inactive_anon:%lukB"
3715 " active_file:%lukB"
3716 " inactive_file:%lukB"
3717 " unevictable:%lukB"
3718 " isolated(anon):%lukB"
3719 " isolated(file):%lukB"
3727 " slab_reclaimable:%lukB"
3728 " slab_unreclaimable:%lukB"
3729 " kernel_stack:%lukB"
3736 " writeback_tmp:%lukB"
3737 " pages_scanned:%lu"
3738 " all_unreclaimable? %s"
3741 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3742 K(min_wmark_pages(zone
)),
3743 K(low_wmark_pages(zone
)),
3744 K(high_wmark_pages(zone
)),
3745 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3746 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3747 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3748 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3749 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3750 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3751 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3752 K(zone
->present_pages
),
3753 K(zone
->managed_pages
),
3754 K(zone_page_state(zone
, NR_MLOCK
)),
3755 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3756 K(zone_page_state(zone
, NR_WRITEBACK
)),
3757 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3758 K(zone_page_state(zone
, NR_SHMEM
)),
3759 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3760 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3761 zone_page_state(zone
, NR_KERNEL_STACK
) *
3763 K(zone_page_state(zone
, NR_PAGETABLE
)),
3764 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3765 K(zone_page_state(zone
, NR_BOUNCE
)),
3767 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
3768 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3769 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3770 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
3771 (!zone_reclaimable(zone
) ? "yes" : "no")
3773 printk("lowmem_reserve[]:");
3774 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3775 printk(" %ld", zone
->lowmem_reserve
[i
]);
3779 for_each_populated_zone(zone
) {
3780 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
3781 unsigned char types
[MAX_ORDER
];
3783 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3786 printk("%s: ", zone
->name
);
3788 spin_lock_irqsave(&zone
->lock
, flags
);
3789 for (order
= 0; order
< MAX_ORDER
; order
++) {
3790 struct free_area
*area
= &zone
->free_area
[order
];
3793 nr
[order
] = area
->nr_free
;
3794 total
+= nr
[order
] << order
;
3797 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3798 if (!list_empty(&area
->free_list
[type
]))
3799 types
[order
] |= 1 << type
;
3802 spin_unlock_irqrestore(&zone
->lock
, flags
);
3803 for (order
= 0; order
< MAX_ORDER
; order
++) {
3804 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3806 show_migration_types(types
[order
]);
3808 printk("= %lukB\n", K(total
));
3811 hugetlb_show_meminfo();
3813 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3815 show_swap_cache_info();
3818 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3820 zoneref
->zone
= zone
;
3821 zoneref
->zone_idx
= zone_idx(zone
);
3825 * Builds allocation fallback zone lists.
3827 * Add all populated zones of a node to the zonelist.
3829 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3833 enum zone_type zone_type
= MAX_NR_ZONES
;
3837 zone
= pgdat
->node_zones
+ zone_type
;
3838 if (populated_zone(zone
)) {
3839 zoneref_set_zone(zone
,
3840 &zonelist
->_zonerefs
[nr_zones
++]);
3841 check_highest_zone(zone_type
);
3843 } while (zone_type
);
3851 * 0 = automatic detection of better ordering.
3852 * 1 = order by ([node] distance, -zonetype)
3853 * 2 = order by (-zonetype, [node] distance)
3855 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3856 * the same zonelist. So only NUMA can configure this param.
3858 #define ZONELIST_ORDER_DEFAULT 0
3859 #define ZONELIST_ORDER_NODE 1
3860 #define ZONELIST_ORDER_ZONE 2
3862 /* zonelist order in the kernel.
3863 * set_zonelist_order() will set this to NODE or ZONE.
3865 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3866 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3870 /* The value user specified ....changed by config */
3871 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3872 /* string for sysctl */
3873 #define NUMA_ZONELIST_ORDER_LEN 16
3874 char numa_zonelist_order
[16] = "default";
3877 * interface for configure zonelist ordering.
3878 * command line option "numa_zonelist_order"
3879 * = "[dD]efault - default, automatic configuration.
3880 * = "[nN]ode - order by node locality, then by zone within node
3881 * = "[zZ]one - order by zone, then by locality within zone
3884 static int __parse_numa_zonelist_order(char *s
)
3886 if (*s
== 'd' || *s
== 'D') {
3887 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3888 } else if (*s
== 'n' || *s
== 'N') {
3889 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3890 } else if (*s
== 'z' || *s
== 'Z') {
3891 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3894 "Ignoring invalid numa_zonelist_order value: "
3901 static __init
int setup_numa_zonelist_order(char *s
)
3908 ret
= __parse_numa_zonelist_order(s
);
3910 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3914 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3917 * sysctl handler for numa_zonelist_order
3919 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
3920 void __user
*buffer
, size_t *length
,
3923 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3925 static DEFINE_MUTEX(zl_order_mutex
);
3927 mutex_lock(&zl_order_mutex
);
3929 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
3933 strcpy(saved_string
, (char *)table
->data
);
3935 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3939 int oldval
= user_zonelist_order
;
3941 ret
= __parse_numa_zonelist_order((char *)table
->data
);
3944 * bogus value. restore saved string
3946 strncpy((char *)table
->data
, saved_string
,
3947 NUMA_ZONELIST_ORDER_LEN
);
3948 user_zonelist_order
= oldval
;
3949 } else if (oldval
!= user_zonelist_order
) {
3950 mutex_lock(&zonelists_mutex
);
3951 build_all_zonelists(NULL
, NULL
);
3952 mutex_unlock(&zonelists_mutex
);
3956 mutex_unlock(&zl_order_mutex
);
3961 #define MAX_NODE_LOAD (nr_online_nodes)
3962 static int node_load
[MAX_NUMNODES
];
3965 * find_next_best_node - find the next node that should appear in a given node's fallback list
3966 * @node: node whose fallback list we're appending
3967 * @used_node_mask: nodemask_t of already used nodes
3969 * We use a number of factors to determine which is the next node that should
3970 * appear on a given node's fallback list. The node should not have appeared
3971 * already in @node's fallback list, and it should be the next closest node
3972 * according to the distance array (which contains arbitrary distance values
3973 * from each node to each node in the system), and should also prefer nodes
3974 * with no CPUs, since presumably they'll have very little allocation pressure
3975 * on them otherwise.
3976 * It returns -1 if no node is found.
3978 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3981 int min_val
= INT_MAX
;
3982 int best_node
= NUMA_NO_NODE
;
3983 const struct cpumask
*tmp
= cpumask_of_node(0);
3985 /* Use the local node if we haven't already */
3986 if (!node_isset(node
, *used_node_mask
)) {
3987 node_set(node
, *used_node_mask
);
3991 for_each_node_state(n
, N_MEMORY
) {
3993 /* Don't want a node to appear more than once */
3994 if (node_isset(n
, *used_node_mask
))
3997 /* Use the distance array to find the distance */
3998 val
= node_distance(node
, n
);
4000 /* Penalize nodes under us ("prefer the next node") */
4003 /* Give preference to headless and unused nodes */
4004 tmp
= cpumask_of_node(n
);
4005 if (!cpumask_empty(tmp
))
4006 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4008 /* Slight preference for less loaded node */
4009 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4010 val
+= node_load
[n
];
4012 if (val
< min_val
) {
4019 node_set(best_node
, *used_node_mask
);
4026 * Build zonelists ordered by node and zones within node.
4027 * This results in maximum locality--normal zone overflows into local
4028 * DMA zone, if any--but risks exhausting DMA zone.
4030 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4033 struct zonelist
*zonelist
;
4035 zonelist
= &pgdat
->node_zonelists
[0];
4036 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4038 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4039 zonelist
->_zonerefs
[j
].zone
= NULL
;
4040 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4044 * Build gfp_thisnode zonelists
4046 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4049 struct zonelist
*zonelist
;
4051 zonelist
= &pgdat
->node_zonelists
[1];
4052 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4053 zonelist
->_zonerefs
[j
].zone
= NULL
;
4054 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4058 * Build zonelists ordered by zone and nodes within zones.
4059 * This results in conserving DMA zone[s] until all Normal memory is
4060 * exhausted, but results in overflowing to remote node while memory
4061 * may still exist in local DMA zone.
4063 static int node_order
[MAX_NUMNODES
];
4065 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4068 int zone_type
; /* needs to be signed */
4070 struct zonelist
*zonelist
;
4072 zonelist
= &pgdat
->node_zonelists
[0];
4074 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4075 for (j
= 0; j
< nr_nodes
; j
++) {
4076 node
= node_order
[j
];
4077 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4078 if (populated_zone(z
)) {
4080 &zonelist
->_zonerefs
[pos
++]);
4081 check_highest_zone(zone_type
);
4085 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4086 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4089 #if defined(CONFIG_64BIT)
4091 * Devices that require DMA32/DMA are relatively rare and do not justify a
4092 * penalty to every machine in case the specialised case applies. Default
4093 * to Node-ordering on 64-bit NUMA machines
4095 static int default_zonelist_order(void)
4097 return ZONELIST_ORDER_NODE
;
4101 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4102 * by the kernel. If processes running on node 0 deplete the low memory zone
4103 * then reclaim will occur more frequency increasing stalls and potentially
4104 * be easier to OOM if a large percentage of the zone is under writeback or
4105 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4106 * Hence, default to zone ordering on 32-bit.
4108 static int default_zonelist_order(void)
4110 return ZONELIST_ORDER_ZONE
;
4112 #endif /* CONFIG_64BIT */
4114 static void set_zonelist_order(void)
4116 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4117 current_zonelist_order
= default_zonelist_order();
4119 current_zonelist_order
= user_zonelist_order
;
4122 static void build_zonelists(pg_data_t
*pgdat
)
4126 nodemask_t used_mask
;
4127 int local_node
, prev_node
;
4128 struct zonelist
*zonelist
;
4129 int order
= current_zonelist_order
;
4131 /* initialize zonelists */
4132 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4133 zonelist
= pgdat
->node_zonelists
+ i
;
4134 zonelist
->_zonerefs
[0].zone
= NULL
;
4135 zonelist
->_zonerefs
[0].zone_idx
= 0;
4138 /* NUMA-aware ordering of nodes */
4139 local_node
= pgdat
->node_id
;
4140 load
= nr_online_nodes
;
4141 prev_node
= local_node
;
4142 nodes_clear(used_mask
);
4144 memset(node_order
, 0, sizeof(node_order
));
4147 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4149 * We don't want to pressure a particular node.
4150 * So adding penalty to the first node in same
4151 * distance group to make it round-robin.
4153 if (node_distance(local_node
, node
) !=
4154 node_distance(local_node
, prev_node
))
4155 node_load
[node
] = load
;
4159 if (order
== ZONELIST_ORDER_NODE
)
4160 build_zonelists_in_node_order(pgdat
, node
);
4162 node_order
[j
++] = node
; /* remember order */
4165 if (order
== ZONELIST_ORDER_ZONE
) {
4166 /* calculate node order -- i.e., DMA last! */
4167 build_zonelists_in_zone_order(pgdat
, j
);
4170 build_thisnode_zonelists(pgdat
);
4173 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4175 * Return node id of node used for "local" allocations.
4176 * I.e., first node id of first zone in arg node's generic zonelist.
4177 * Used for initializing percpu 'numa_mem', which is used primarily
4178 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4180 int local_memory_node(int node
)
4184 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4185 gfp_zone(GFP_KERNEL
),
4192 #else /* CONFIG_NUMA */
4194 static void set_zonelist_order(void)
4196 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4199 static void build_zonelists(pg_data_t
*pgdat
)
4201 int node
, local_node
;
4203 struct zonelist
*zonelist
;
4205 local_node
= pgdat
->node_id
;
4207 zonelist
= &pgdat
->node_zonelists
[0];
4208 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4211 * Now we build the zonelist so that it contains the zones
4212 * of all the other nodes.
4213 * We don't want to pressure a particular node, so when
4214 * building the zones for node N, we make sure that the
4215 * zones coming right after the local ones are those from
4216 * node N+1 (modulo N)
4218 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4219 if (!node_online(node
))
4221 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4223 for (node
= 0; node
< local_node
; node
++) {
4224 if (!node_online(node
))
4226 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4229 zonelist
->_zonerefs
[j
].zone
= NULL
;
4230 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4233 #endif /* CONFIG_NUMA */
4236 * Boot pageset table. One per cpu which is going to be used for all
4237 * zones and all nodes. The parameters will be set in such a way
4238 * that an item put on a list will immediately be handed over to
4239 * the buddy list. This is safe since pageset manipulation is done
4240 * with interrupts disabled.
4242 * The boot_pagesets must be kept even after bootup is complete for
4243 * unused processors and/or zones. They do play a role for bootstrapping
4244 * hotplugged processors.
4246 * zoneinfo_show() and maybe other functions do
4247 * not check if the processor is online before following the pageset pointer.
4248 * Other parts of the kernel may not check if the zone is available.
4250 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4251 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4252 static void setup_zone_pageset(struct zone
*zone
);
4255 * Global mutex to protect against size modification of zonelists
4256 * as well as to serialize pageset setup for the new populated zone.
4258 DEFINE_MUTEX(zonelists_mutex
);
4260 /* return values int ....just for stop_machine() */
4261 static int __build_all_zonelists(void *data
)
4265 pg_data_t
*self
= data
;
4268 memset(node_load
, 0, sizeof(node_load
));
4271 if (self
&& !node_online(self
->node_id
)) {
4272 build_zonelists(self
);
4275 for_each_online_node(nid
) {
4276 pg_data_t
*pgdat
= NODE_DATA(nid
);
4278 build_zonelists(pgdat
);
4282 * Initialize the boot_pagesets that are going to be used
4283 * for bootstrapping processors. The real pagesets for
4284 * each zone will be allocated later when the per cpu
4285 * allocator is available.
4287 * boot_pagesets are used also for bootstrapping offline
4288 * cpus if the system is already booted because the pagesets
4289 * are needed to initialize allocators on a specific cpu too.
4290 * F.e. the percpu allocator needs the page allocator which
4291 * needs the percpu allocator in order to allocate its pagesets
4292 * (a chicken-egg dilemma).
4294 for_each_possible_cpu(cpu
) {
4295 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4297 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4299 * We now know the "local memory node" for each node--
4300 * i.e., the node of the first zone in the generic zonelist.
4301 * Set up numa_mem percpu variable for on-line cpus. During
4302 * boot, only the boot cpu should be on-line; we'll init the
4303 * secondary cpus' numa_mem as they come on-line. During
4304 * node/memory hotplug, we'll fixup all on-line cpus.
4306 if (cpu_online(cpu
))
4307 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4314 static noinline
void __init
4315 build_all_zonelists_init(void)
4317 __build_all_zonelists(NULL
);
4318 mminit_verify_zonelist();
4319 cpuset_init_current_mems_allowed();
4323 * Called with zonelists_mutex held always
4324 * unless system_state == SYSTEM_BOOTING.
4326 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4327 * [we're only called with non-NULL zone through __meminit paths] and
4328 * (2) call of __init annotated helper build_all_zonelists_init
4329 * [protected by SYSTEM_BOOTING].
4331 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
4333 set_zonelist_order();
4335 if (system_state
== SYSTEM_BOOTING
) {
4336 build_all_zonelists_init();
4338 #ifdef CONFIG_MEMORY_HOTPLUG
4340 setup_zone_pageset(zone
);
4342 /* we have to stop all cpus to guarantee there is no user
4344 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
4345 /* cpuset refresh routine should be here */
4347 vm_total_pages
= nr_free_pagecache_pages();
4349 * Disable grouping by mobility if the number of pages in the
4350 * system is too low to allow the mechanism to work. It would be
4351 * more accurate, but expensive to check per-zone. This check is
4352 * made on memory-hotadd so a system can start with mobility
4353 * disabled and enable it later
4355 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
4356 page_group_by_mobility_disabled
= 1;
4358 page_group_by_mobility_disabled
= 0;
4360 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
4361 "Total pages: %ld\n",
4363 zonelist_order_name
[current_zonelist_order
],
4364 page_group_by_mobility_disabled
? "off" : "on",
4367 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
4372 * Helper functions to size the waitqueue hash table.
4373 * Essentially these want to choose hash table sizes sufficiently
4374 * large so that collisions trying to wait on pages are rare.
4375 * But in fact, the number of active page waitqueues on typical
4376 * systems is ridiculously low, less than 200. So this is even
4377 * conservative, even though it seems large.
4379 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4380 * waitqueues, i.e. the size of the waitq table given the number of pages.
4382 #define PAGES_PER_WAITQUEUE 256
4384 #ifndef CONFIG_MEMORY_HOTPLUG
4385 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4387 unsigned long size
= 1;
4389 pages
/= PAGES_PER_WAITQUEUE
;
4391 while (size
< pages
)
4395 * Once we have dozens or even hundreds of threads sleeping
4396 * on IO we've got bigger problems than wait queue collision.
4397 * Limit the size of the wait table to a reasonable size.
4399 size
= min(size
, 4096UL);
4401 return max(size
, 4UL);
4405 * A zone's size might be changed by hot-add, so it is not possible to determine
4406 * a suitable size for its wait_table. So we use the maximum size now.
4408 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4410 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4411 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4412 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4414 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4415 * or more by the traditional way. (See above). It equals:
4417 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4418 * ia64(16K page size) : = ( 8G + 4M)byte.
4419 * powerpc (64K page size) : = (32G +16M)byte.
4421 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4428 * This is an integer logarithm so that shifts can be used later
4429 * to extract the more random high bits from the multiplicative
4430 * hash function before the remainder is taken.
4432 static inline unsigned long wait_table_bits(unsigned long size
)
4438 * Initially all pages are reserved - free ones are freed
4439 * up by free_all_bootmem() once the early boot process is
4440 * done. Non-atomic initialization, single-pass.
4442 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
4443 unsigned long start_pfn
, enum memmap_context context
)
4445 pg_data_t
*pgdat
= NODE_DATA(nid
);
4446 unsigned long end_pfn
= start_pfn
+ size
;
4449 unsigned long nr_initialised
= 0;
4451 if (highest_memmap_pfn
< end_pfn
- 1)
4452 highest_memmap_pfn
= end_pfn
- 1;
4454 z
= &pgdat
->node_zones
[zone
];
4455 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4457 * There can be holes in boot-time mem_map[]s
4458 * handed to this function. They do not
4459 * exist on hotplugged memory.
4461 if (context
== MEMMAP_EARLY
) {
4462 if (!early_pfn_valid(pfn
))
4464 if (!early_pfn_in_nid(pfn
, nid
))
4466 if (!update_defer_init(pgdat
, pfn
, end_pfn
,
4472 * Mark the block movable so that blocks are reserved for
4473 * movable at startup. This will force kernel allocations
4474 * to reserve their blocks rather than leaking throughout
4475 * the address space during boot when many long-lived
4476 * kernel allocations are made.
4478 * bitmap is created for zone's valid pfn range. but memmap
4479 * can be created for invalid pages (for alignment)
4480 * check here not to call set_pageblock_migratetype() against
4483 if (!(pfn
& (pageblock_nr_pages
- 1))) {
4484 struct page
*page
= pfn_to_page(pfn
);
4486 __init_single_page(page
, pfn
, zone
, nid
);
4487 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4489 __init_single_pfn(pfn
, zone
, nid
);
4494 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4496 unsigned int order
, t
;
4497 for_each_migratetype_order(order
, t
) {
4498 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4499 zone
->free_area
[order
].nr_free
= 0;
4503 #ifndef __HAVE_ARCH_MEMMAP_INIT
4504 #define memmap_init(size, nid, zone, start_pfn) \
4505 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4508 static int zone_batchsize(struct zone
*zone
)
4514 * The per-cpu-pages pools are set to around 1000th of the
4515 * size of the zone. But no more than 1/2 of a meg.
4517 * OK, so we don't know how big the cache is. So guess.
4519 batch
= zone
->managed_pages
/ 1024;
4520 if (batch
* PAGE_SIZE
> 512 * 1024)
4521 batch
= (512 * 1024) / PAGE_SIZE
;
4522 batch
/= 4; /* We effectively *= 4 below */
4527 * Clamp the batch to a 2^n - 1 value. Having a power
4528 * of 2 value was found to be more likely to have
4529 * suboptimal cache aliasing properties in some cases.
4531 * For example if 2 tasks are alternately allocating
4532 * batches of pages, one task can end up with a lot
4533 * of pages of one half of the possible page colors
4534 * and the other with pages of the other colors.
4536 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4541 /* The deferral and batching of frees should be suppressed under NOMMU
4544 * The problem is that NOMMU needs to be able to allocate large chunks
4545 * of contiguous memory as there's no hardware page translation to
4546 * assemble apparent contiguous memory from discontiguous pages.
4548 * Queueing large contiguous runs of pages for batching, however,
4549 * causes the pages to actually be freed in smaller chunks. As there
4550 * can be a significant delay between the individual batches being
4551 * recycled, this leads to the once large chunks of space being
4552 * fragmented and becoming unavailable for high-order allocations.
4559 * pcp->high and pcp->batch values are related and dependent on one another:
4560 * ->batch must never be higher then ->high.
4561 * The following function updates them in a safe manner without read side
4564 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4565 * those fields changing asynchronously (acording the the above rule).
4567 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4568 * outside of boot time (or some other assurance that no concurrent updaters
4571 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4572 unsigned long batch
)
4574 /* start with a fail safe value for batch */
4578 /* Update high, then batch, in order */
4585 /* a companion to pageset_set_high() */
4586 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4588 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4591 static void pageset_init(struct per_cpu_pageset
*p
)
4593 struct per_cpu_pages
*pcp
;
4596 memset(p
, 0, sizeof(*p
));
4600 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4601 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4604 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4607 pageset_set_batch(p
, batch
);
4611 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4612 * to the value high for the pageset p.
4614 static void pageset_set_high(struct per_cpu_pageset
*p
,
4617 unsigned long batch
= max(1UL, high
/ 4);
4618 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4619 batch
= PAGE_SHIFT
* 8;
4621 pageset_update(&p
->pcp
, high
, batch
);
4624 static void pageset_set_high_and_batch(struct zone
*zone
,
4625 struct per_cpu_pageset
*pcp
)
4627 if (percpu_pagelist_fraction
)
4628 pageset_set_high(pcp
,
4629 (zone
->managed_pages
/
4630 percpu_pagelist_fraction
));
4632 pageset_set_batch(pcp
, zone_batchsize(zone
));
4635 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4637 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4640 pageset_set_high_and_batch(zone
, pcp
);
4643 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4646 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4647 for_each_possible_cpu(cpu
)
4648 zone_pageset_init(zone
, cpu
);
4652 * Allocate per cpu pagesets and initialize them.
4653 * Before this call only boot pagesets were available.
4655 void __init
setup_per_cpu_pageset(void)
4659 for_each_populated_zone(zone
)
4660 setup_zone_pageset(zone
);
4663 static noinline __init_refok
4664 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4670 * The per-page waitqueue mechanism uses hashed waitqueues
4673 zone
->wait_table_hash_nr_entries
=
4674 wait_table_hash_nr_entries(zone_size_pages
);
4675 zone
->wait_table_bits
=
4676 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4677 alloc_size
= zone
->wait_table_hash_nr_entries
4678 * sizeof(wait_queue_head_t
);
4680 if (!slab_is_available()) {
4681 zone
->wait_table
= (wait_queue_head_t
*)
4682 memblock_virt_alloc_node_nopanic(
4683 alloc_size
, zone
->zone_pgdat
->node_id
);
4686 * This case means that a zone whose size was 0 gets new memory
4687 * via memory hot-add.
4688 * But it may be the case that a new node was hot-added. In
4689 * this case vmalloc() will not be able to use this new node's
4690 * memory - this wait_table must be initialized to use this new
4691 * node itself as well.
4692 * To use this new node's memory, further consideration will be
4695 zone
->wait_table
= vmalloc(alloc_size
);
4697 if (!zone
->wait_table
)
4700 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4701 init_waitqueue_head(zone
->wait_table
+ i
);
4706 static __meminit
void zone_pcp_init(struct zone
*zone
)
4709 * per cpu subsystem is not up at this point. The following code
4710 * relies on the ability of the linker to provide the
4711 * offset of a (static) per cpu variable into the per cpu area.
4713 zone
->pageset
= &boot_pageset
;
4715 if (populated_zone(zone
))
4716 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4717 zone
->name
, zone
->present_pages
,
4718 zone_batchsize(zone
));
4721 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4722 unsigned long zone_start_pfn
,
4725 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4727 ret
= zone_wait_table_init(zone
, size
);
4730 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4732 zone
->zone_start_pfn
= zone_start_pfn
;
4734 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4735 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4737 (unsigned long)zone_idx(zone
),
4738 zone_start_pfn
, (zone_start_pfn
+ size
));
4740 zone_init_free_lists(zone
);
4745 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4746 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4749 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4751 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
4752 struct mminit_pfnnid_cache
*state
)
4754 unsigned long start_pfn
, end_pfn
;
4757 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
4758 return state
->last_nid
;
4760 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4762 state
->last_start
= start_pfn
;
4763 state
->last_end
= end_pfn
;
4764 state
->last_nid
= nid
;
4769 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4772 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4773 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4774 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4776 * If an architecture guarantees that all ranges registered contain no holes
4777 * and may be freed, this this function may be used instead of calling
4778 * memblock_free_early_nid() manually.
4780 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4782 unsigned long start_pfn
, end_pfn
;
4785 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4786 start_pfn
= min(start_pfn
, max_low_pfn
);
4787 end_pfn
= min(end_pfn
, max_low_pfn
);
4789 if (start_pfn
< end_pfn
)
4790 memblock_free_early_nid(PFN_PHYS(start_pfn
),
4791 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
4797 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4798 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4800 * If an architecture guarantees that all ranges registered contain no holes and may
4801 * be freed, this function may be used instead of calling memory_present() manually.
4803 void __init
sparse_memory_present_with_active_regions(int nid
)
4805 unsigned long start_pfn
, end_pfn
;
4808 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4809 memory_present(this_nid
, start_pfn
, end_pfn
);
4813 * get_pfn_range_for_nid - Return the start and end page frames for a node
4814 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4815 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4816 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4818 * It returns the start and end page frame of a node based on information
4819 * provided by memblock_set_node(). If called for a node
4820 * with no available memory, a warning is printed and the start and end
4823 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4824 unsigned long *start_pfn
, unsigned long *end_pfn
)
4826 unsigned long this_start_pfn
, this_end_pfn
;
4832 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4833 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4834 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4837 if (*start_pfn
== -1UL)
4842 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4843 * assumption is made that zones within a node are ordered in monotonic
4844 * increasing memory addresses so that the "highest" populated zone is used
4846 static void __init
find_usable_zone_for_movable(void)
4849 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4850 if (zone_index
== ZONE_MOVABLE
)
4853 if (arch_zone_highest_possible_pfn
[zone_index
] >
4854 arch_zone_lowest_possible_pfn
[zone_index
])
4858 VM_BUG_ON(zone_index
== -1);
4859 movable_zone
= zone_index
;
4863 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4864 * because it is sized independent of architecture. Unlike the other zones,
4865 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4866 * in each node depending on the size of each node and how evenly kernelcore
4867 * is distributed. This helper function adjusts the zone ranges
4868 * provided by the architecture for a given node by using the end of the
4869 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4870 * zones within a node are in order of monotonic increases memory addresses
4872 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4873 unsigned long zone_type
,
4874 unsigned long node_start_pfn
,
4875 unsigned long node_end_pfn
,
4876 unsigned long *zone_start_pfn
,
4877 unsigned long *zone_end_pfn
)
4879 /* Only adjust if ZONE_MOVABLE is on this node */
4880 if (zone_movable_pfn
[nid
]) {
4881 /* Size ZONE_MOVABLE */
4882 if (zone_type
== ZONE_MOVABLE
) {
4883 *zone_start_pfn
= zone_movable_pfn
[nid
];
4884 *zone_end_pfn
= min(node_end_pfn
,
4885 arch_zone_highest_possible_pfn
[movable_zone
]);
4887 /* Adjust for ZONE_MOVABLE starting within this range */
4888 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4889 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4890 *zone_end_pfn
= zone_movable_pfn
[nid
];
4892 /* Check if this whole range is within ZONE_MOVABLE */
4893 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4894 *zone_start_pfn
= *zone_end_pfn
;
4899 * Return the number of pages a zone spans in a node, including holes
4900 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4902 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4903 unsigned long zone_type
,
4904 unsigned long node_start_pfn
,
4905 unsigned long node_end_pfn
,
4906 unsigned long *ignored
)
4908 unsigned long zone_start_pfn
, zone_end_pfn
;
4910 /* When hotadd a new node from cpu_up(), the node should be empty */
4911 if (!node_start_pfn
&& !node_end_pfn
)
4914 /* Get the start and end of the zone */
4915 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4916 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4917 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4918 node_start_pfn
, node_end_pfn
,
4919 &zone_start_pfn
, &zone_end_pfn
);
4921 /* Check that this node has pages within the zone's required range */
4922 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4925 /* Move the zone boundaries inside the node if necessary */
4926 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4927 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4929 /* Return the spanned pages */
4930 return zone_end_pfn
- zone_start_pfn
;
4934 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4935 * then all holes in the requested range will be accounted for.
4937 unsigned long __meminit
__absent_pages_in_range(int nid
,
4938 unsigned long range_start_pfn
,
4939 unsigned long range_end_pfn
)
4941 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4942 unsigned long start_pfn
, end_pfn
;
4945 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4946 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4947 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4948 nr_absent
-= end_pfn
- start_pfn
;
4954 * absent_pages_in_range - Return number of page frames in holes within a range
4955 * @start_pfn: The start PFN to start searching for holes
4956 * @end_pfn: The end PFN to stop searching for holes
4958 * It returns the number of pages frames in memory holes within a range.
4960 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4961 unsigned long end_pfn
)
4963 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4966 /* Return the number of page frames in holes in a zone on a node */
4967 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4968 unsigned long zone_type
,
4969 unsigned long node_start_pfn
,
4970 unsigned long node_end_pfn
,
4971 unsigned long *ignored
)
4973 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4974 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4975 unsigned long zone_start_pfn
, zone_end_pfn
;
4977 /* When hotadd a new node from cpu_up(), the node should be empty */
4978 if (!node_start_pfn
&& !node_end_pfn
)
4981 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4982 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4984 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4985 node_start_pfn
, node_end_pfn
,
4986 &zone_start_pfn
, &zone_end_pfn
);
4987 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4990 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4991 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4992 unsigned long zone_type
,
4993 unsigned long node_start_pfn
,
4994 unsigned long node_end_pfn
,
4995 unsigned long *zones_size
)
4997 return zones_size
[zone_type
];
5000 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5001 unsigned long zone_type
,
5002 unsigned long node_start_pfn
,
5003 unsigned long node_end_pfn
,
5004 unsigned long *zholes_size
)
5009 return zholes_size
[zone_type
];
5012 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5014 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5015 unsigned long node_start_pfn
,
5016 unsigned long node_end_pfn
,
5017 unsigned long *zones_size
,
5018 unsigned long *zholes_size
)
5020 unsigned long realtotalpages
= 0, totalpages
= 0;
5023 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5024 struct zone
*zone
= pgdat
->node_zones
+ i
;
5025 unsigned long size
, real_size
;
5027 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5031 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5032 node_start_pfn
, node_end_pfn
,
5034 zone
->spanned_pages
= size
;
5035 zone
->present_pages
= real_size
;
5038 realtotalpages
+= real_size
;
5041 pgdat
->node_spanned_pages
= totalpages
;
5042 pgdat
->node_present_pages
= realtotalpages
;
5043 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5047 #ifndef CONFIG_SPARSEMEM
5049 * Calculate the size of the zone->blockflags rounded to an unsigned long
5050 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5051 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5052 * round what is now in bits to nearest long in bits, then return it in
5055 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5057 unsigned long usemapsize
;
5059 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5060 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5061 usemapsize
= usemapsize
>> pageblock_order
;
5062 usemapsize
*= NR_PAGEBLOCK_BITS
;
5063 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5065 return usemapsize
/ 8;
5068 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5070 unsigned long zone_start_pfn
,
5071 unsigned long zonesize
)
5073 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5074 zone
->pageblock_flags
= NULL
;
5076 zone
->pageblock_flags
=
5077 memblock_virt_alloc_node_nopanic(usemapsize
,
5081 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5082 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5083 #endif /* CONFIG_SPARSEMEM */
5085 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5087 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5088 void __paginginit
set_pageblock_order(void)
5092 /* Check that pageblock_nr_pages has not already been setup */
5093 if (pageblock_order
)
5096 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5097 order
= HUGETLB_PAGE_ORDER
;
5099 order
= MAX_ORDER
- 1;
5102 * Assume the largest contiguous order of interest is a huge page.
5103 * This value may be variable depending on boot parameters on IA64 and
5106 pageblock_order
= order
;
5108 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5111 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5112 * is unused as pageblock_order is set at compile-time. See
5113 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5116 void __paginginit
set_pageblock_order(void)
5120 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5122 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5123 unsigned long present_pages
)
5125 unsigned long pages
= spanned_pages
;
5128 * Provide a more accurate estimation if there are holes within
5129 * the zone and SPARSEMEM is in use. If there are holes within the
5130 * zone, each populated memory region may cost us one or two extra
5131 * memmap pages due to alignment because memmap pages for each
5132 * populated regions may not naturally algined on page boundary.
5133 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5135 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5136 IS_ENABLED(CONFIG_SPARSEMEM
))
5137 pages
= present_pages
;
5139 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5143 * Set up the zone data structures:
5144 * - mark all pages reserved
5145 * - mark all memory queues empty
5146 * - clear the memory bitmaps
5148 * NOTE: pgdat should get zeroed by caller.
5150 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5153 int nid
= pgdat
->node_id
;
5154 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
5157 pgdat_resize_init(pgdat
);
5158 #ifdef CONFIG_NUMA_BALANCING
5159 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5160 pgdat
->numabalancing_migrate_nr_pages
= 0;
5161 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5163 init_waitqueue_head(&pgdat
->kswapd_wait
);
5164 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5165 pgdat_page_ext_init(pgdat
);
5167 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5168 struct zone
*zone
= pgdat
->node_zones
+ j
;
5169 unsigned long size
, realsize
, freesize
, memmap_pages
;
5171 size
= zone
->spanned_pages
;
5172 realsize
= freesize
= zone
->present_pages
;
5175 * Adjust freesize so that it accounts for how much memory
5176 * is used by this zone for memmap. This affects the watermark
5177 * and per-cpu initialisations
5179 memmap_pages
= calc_memmap_size(size
, realsize
);
5180 if (!is_highmem_idx(j
)) {
5181 if (freesize
>= memmap_pages
) {
5182 freesize
-= memmap_pages
;
5185 " %s zone: %lu pages used for memmap\n",
5186 zone_names
[j
], memmap_pages
);
5189 " %s zone: %lu pages exceeds freesize %lu\n",
5190 zone_names
[j
], memmap_pages
, freesize
);
5193 /* Account for reserved pages */
5194 if (j
== 0 && freesize
> dma_reserve
) {
5195 freesize
-= dma_reserve
;
5196 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5197 zone_names
[0], dma_reserve
);
5200 if (!is_highmem_idx(j
))
5201 nr_kernel_pages
+= freesize
;
5202 /* Charge for highmem memmap if there are enough kernel pages */
5203 else if (nr_kernel_pages
> memmap_pages
* 2)
5204 nr_kernel_pages
-= memmap_pages
;
5205 nr_all_pages
+= freesize
;
5208 * Set an approximate value for lowmem here, it will be adjusted
5209 * when the bootmem allocator frees pages into the buddy system.
5210 * And all highmem pages will be managed by the buddy system.
5212 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5215 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5217 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5219 zone
->name
= zone_names
[j
];
5220 spin_lock_init(&zone
->lock
);
5221 spin_lock_init(&zone
->lru_lock
);
5222 zone_seqlock_init(zone
);
5223 zone
->zone_pgdat
= pgdat
;
5224 zone_pcp_init(zone
);
5226 /* For bootup, initialized properly in watermark setup */
5227 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5229 lruvec_init(&zone
->lruvec
);
5233 set_pageblock_order();
5234 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
5235 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
5237 memmap_init(size
, nid
, j
, zone_start_pfn
);
5238 zone_start_pfn
+= size
;
5242 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
5244 unsigned long __maybe_unused offset
= 0;
5246 /* Skip empty nodes */
5247 if (!pgdat
->node_spanned_pages
)
5250 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5251 /* ia64 gets its own node_mem_map, before this, without bootmem */
5252 if (!pgdat
->node_mem_map
) {
5253 unsigned long size
, start
, end
;
5257 * The zone's endpoints aren't required to be MAX_ORDER
5258 * aligned but the node_mem_map endpoints must be in order
5259 * for the buddy allocator to function correctly.
5261 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
5262 offset
= pgdat
->node_start_pfn
- start
;
5263 end
= pgdat_end_pfn(pgdat
);
5264 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
5265 size
= (end
- start
) * sizeof(struct page
);
5266 map
= alloc_remap(pgdat
->node_id
, size
);
5268 map
= memblock_virt_alloc_node_nopanic(size
,
5270 pgdat
->node_mem_map
= map
+ offset
;
5272 #ifndef CONFIG_NEED_MULTIPLE_NODES
5274 * With no DISCONTIG, the global mem_map is just set as node 0's
5276 if (pgdat
== NODE_DATA(0)) {
5277 mem_map
= NODE_DATA(0)->node_mem_map
;
5278 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5279 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
5281 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5284 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5287 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
5288 unsigned long node_start_pfn
, unsigned long *zholes_size
)
5290 pg_data_t
*pgdat
= NODE_DATA(nid
);
5291 unsigned long start_pfn
= 0;
5292 unsigned long end_pfn
= 0;
5294 /* pg_data_t should be reset to zero when it's allocated */
5295 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
5297 reset_deferred_meminit(pgdat
);
5298 pgdat
->node_id
= nid
;
5299 pgdat
->node_start_pfn
= node_start_pfn
;
5300 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5301 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
5302 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
5303 (u64
)start_pfn
<< PAGE_SHIFT
,
5304 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
5306 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
5307 zones_size
, zholes_size
);
5309 alloc_node_mem_map(pgdat
);
5310 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5311 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5312 nid
, (unsigned long)pgdat
,
5313 (unsigned long)pgdat
->node_mem_map
);
5316 free_area_init_core(pgdat
);
5319 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5321 #if MAX_NUMNODES > 1
5323 * Figure out the number of possible node ids.
5325 void __init
setup_nr_node_ids(void)
5327 unsigned int highest
;
5329 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
5330 nr_node_ids
= highest
+ 1;
5335 * node_map_pfn_alignment - determine the maximum internode alignment
5337 * This function should be called after node map is populated and sorted.
5338 * It calculates the maximum power of two alignment which can distinguish
5341 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5342 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5343 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5344 * shifted, 1GiB is enough and this function will indicate so.
5346 * This is used to test whether pfn -> nid mapping of the chosen memory
5347 * model has fine enough granularity to avoid incorrect mapping for the
5348 * populated node map.
5350 * Returns the determined alignment in pfn's. 0 if there is no alignment
5351 * requirement (single node).
5353 unsigned long __init
node_map_pfn_alignment(void)
5355 unsigned long accl_mask
= 0, last_end
= 0;
5356 unsigned long start
, end
, mask
;
5360 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
5361 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
5368 * Start with a mask granular enough to pin-point to the
5369 * start pfn and tick off bits one-by-one until it becomes
5370 * too coarse to separate the current node from the last.
5372 mask
= ~((1 << __ffs(start
)) - 1);
5373 while (mask
&& last_end
<= (start
& (mask
<< 1)))
5376 /* accumulate all internode masks */
5380 /* convert mask to number of pages */
5381 return ~accl_mask
+ 1;
5384 /* Find the lowest pfn for a node */
5385 static unsigned long __init
find_min_pfn_for_node(int nid
)
5387 unsigned long min_pfn
= ULONG_MAX
;
5388 unsigned long start_pfn
;
5391 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
5392 min_pfn
= min(min_pfn
, start_pfn
);
5394 if (min_pfn
== ULONG_MAX
) {
5396 "Could not find start_pfn for node %d\n", nid
);
5404 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5406 * It returns the minimum PFN based on information provided via
5407 * memblock_set_node().
5409 unsigned long __init
find_min_pfn_with_active_regions(void)
5411 return find_min_pfn_for_node(MAX_NUMNODES
);
5415 * early_calculate_totalpages()
5416 * Sum pages in active regions for movable zone.
5417 * Populate N_MEMORY for calculating usable_nodes.
5419 static unsigned long __init
early_calculate_totalpages(void)
5421 unsigned long totalpages
= 0;
5422 unsigned long start_pfn
, end_pfn
;
5425 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5426 unsigned long pages
= end_pfn
- start_pfn
;
5428 totalpages
+= pages
;
5430 node_set_state(nid
, N_MEMORY
);
5436 * Find the PFN the Movable zone begins in each node. Kernel memory
5437 * is spread evenly between nodes as long as the nodes have enough
5438 * memory. When they don't, some nodes will have more kernelcore than
5441 static void __init
find_zone_movable_pfns_for_nodes(void)
5444 unsigned long usable_startpfn
;
5445 unsigned long kernelcore_node
, kernelcore_remaining
;
5446 /* save the state before borrow the nodemask */
5447 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5448 unsigned long totalpages
= early_calculate_totalpages();
5449 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5450 struct memblock_region
*r
;
5452 /* Need to find movable_zone earlier when movable_node is specified. */
5453 find_usable_zone_for_movable();
5456 * If movable_node is specified, ignore kernelcore and movablecore
5459 if (movable_node_is_enabled()) {
5460 for_each_memblock(memory
, r
) {
5461 if (!memblock_is_hotpluggable(r
))
5466 usable_startpfn
= PFN_DOWN(r
->base
);
5467 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5468 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5476 * If movablecore=nn[KMG] was specified, calculate what size of
5477 * kernelcore that corresponds so that memory usable for
5478 * any allocation type is evenly spread. If both kernelcore
5479 * and movablecore are specified, then the value of kernelcore
5480 * will be used for required_kernelcore if it's greater than
5481 * what movablecore would have allowed.
5483 if (required_movablecore
) {
5484 unsigned long corepages
;
5487 * Round-up so that ZONE_MOVABLE is at least as large as what
5488 * was requested by the user
5490 required_movablecore
=
5491 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5492 required_movablecore
= min(totalpages
, required_movablecore
);
5493 corepages
= totalpages
- required_movablecore
;
5495 required_kernelcore
= max(required_kernelcore
, corepages
);
5499 * If kernelcore was not specified or kernelcore size is larger
5500 * than totalpages, there is no ZONE_MOVABLE.
5502 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
5505 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5506 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5509 /* Spread kernelcore memory as evenly as possible throughout nodes */
5510 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5511 for_each_node_state(nid
, N_MEMORY
) {
5512 unsigned long start_pfn
, end_pfn
;
5515 * Recalculate kernelcore_node if the division per node
5516 * now exceeds what is necessary to satisfy the requested
5517 * amount of memory for the kernel
5519 if (required_kernelcore
< kernelcore_node
)
5520 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5523 * As the map is walked, we track how much memory is usable
5524 * by the kernel using kernelcore_remaining. When it is
5525 * 0, the rest of the node is usable by ZONE_MOVABLE
5527 kernelcore_remaining
= kernelcore_node
;
5529 /* Go through each range of PFNs within this node */
5530 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5531 unsigned long size_pages
;
5533 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5534 if (start_pfn
>= end_pfn
)
5537 /* Account for what is only usable for kernelcore */
5538 if (start_pfn
< usable_startpfn
) {
5539 unsigned long kernel_pages
;
5540 kernel_pages
= min(end_pfn
, usable_startpfn
)
5543 kernelcore_remaining
-= min(kernel_pages
,
5544 kernelcore_remaining
);
5545 required_kernelcore
-= min(kernel_pages
,
5546 required_kernelcore
);
5548 /* Continue if range is now fully accounted */
5549 if (end_pfn
<= usable_startpfn
) {
5552 * Push zone_movable_pfn to the end so
5553 * that if we have to rebalance
5554 * kernelcore across nodes, we will
5555 * not double account here
5557 zone_movable_pfn
[nid
] = end_pfn
;
5560 start_pfn
= usable_startpfn
;
5564 * The usable PFN range for ZONE_MOVABLE is from
5565 * start_pfn->end_pfn. Calculate size_pages as the
5566 * number of pages used as kernelcore
5568 size_pages
= end_pfn
- start_pfn
;
5569 if (size_pages
> kernelcore_remaining
)
5570 size_pages
= kernelcore_remaining
;
5571 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5574 * Some kernelcore has been met, update counts and
5575 * break if the kernelcore for this node has been
5578 required_kernelcore
-= min(required_kernelcore
,
5580 kernelcore_remaining
-= size_pages
;
5581 if (!kernelcore_remaining
)
5587 * If there is still required_kernelcore, we do another pass with one
5588 * less node in the count. This will push zone_movable_pfn[nid] further
5589 * along on the nodes that still have memory until kernelcore is
5593 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5597 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5598 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5599 zone_movable_pfn
[nid
] =
5600 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5603 /* restore the node_state */
5604 node_states
[N_MEMORY
] = saved_node_state
;
5607 /* Any regular or high memory on that node ? */
5608 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5610 enum zone_type zone_type
;
5612 if (N_MEMORY
== N_NORMAL_MEMORY
)
5615 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5616 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5617 if (populated_zone(zone
)) {
5618 node_set_state(nid
, N_HIGH_MEMORY
);
5619 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5620 zone_type
<= ZONE_NORMAL
)
5621 node_set_state(nid
, N_NORMAL_MEMORY
);
5628 * free_area_init_nodes - Initialise all pg_data_t and zone data
5629 * @max_zone_pfn: an array of max PFNs for each zone
5631 * This will call free_area_init_node() for each active node in the system.
5632 * Using the page ranges provided by memblock_set_node(), the size of each
5633 * zone in each node and their holes is calculated. If the maximum PFN
5634 * between two adjacent zones match, it is assumed that the zone is empty.
5635 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5636 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5637 * starts where the previous one ended. For example, ZONE_DMA32 starts
5638 * at arch_max_dma_pfn.
5640 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5642 unsigned long start_pfn
, end_pfn
;
5645 /* Record where the zone boundaries are */
5646 memset(arch_zone_lowest_possible_pfn
, 0,
5647 sizeof(arch_zone_lowest_possible_pfn
));
5648 memset(arch_zone_highest_possible_pfn
, 0,
5649 sizeof(arch_zone_highest_possible_pfn
));
5650 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5651 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5652 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5653 if (i
== ZONE_MOVABLE
)
5655 arch_zone_lowest_possible_pfn
[i
] =
5656 arch_zone_highest_possible_pfn
[i
-1];
5657 arch_zone_highest_possible_pfn
[i
] =
5658 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5660 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5661 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5663 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5664 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5665 find_zone_movable_pfns_for_nodes();
5667 /* Print out the zone ranges */
5668 pr_info("Zone ranges:\n");
5669 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5670 if (i
== ZONE_MOVABLE
)
5672 pr_info(" %-8s ", zone_names
[i
]);
5673 if (arch_zone_lowest_possible_pfn
[i
] ==
5674 arch_zone_highest_possible_pfn
[i
])
5677 pr_cont("[mem %#018Lx-%#018Lx]\n",
5678 (u64
)arch_zone_lowest_possible_pfn
[i
]
5680 ((u64
)arch_zone_highest_possible_pfn
[i
]
5681 << PAGE_SHIFT
) - 1);
5684 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5685 pr_info("Movable zone start for each node\n");
5686 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5687 if (zone_movable_pfn
[i
])
5688 pr_info(" Node %d: %#018Lx\n", i
,
5689 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
5692 /* Print out the early node map */
5693 pr_info("Early memory node ranges\n");
5694 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5695 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
5696 (u64
)start_pfn
<< PAGE_SHIFT
,
5697 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
5699 /* Initialise every node */
5700 mminit_verify_pageflags_layout();
5701 setup_nr_node_ids();
5702 for_each_online_node(nid
) {
5703 pg_data_t
*pgdat
= NODE_DATA(nid
);
5704 free_area_init_node(nid
, NULL
,
5705 find_min_pfn_for_node(nid
), NULL
);
5707 /* Any memory on that node */
5708 if (pgdat
->node_present_pages
)
5709 node_set_state(nid
, N_MEMORY
);
5710 check_for_memory(pgdat
, nid
);
5714 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5716 unsigned long long coremem
;
5720 coremem
= memparse(p
, &p
);
5721 *core
= coremem
>> PAGE_SHIFT
;
5723 /* Paranoid check that UL is enough for the coremem value */
5724 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5730 * kernelcore=size sets the amount of memory for use for allocations that
5731 * cannot be reclaimed or migrated.
5733 static int __init
cmdline_parse_kernelcore(char *p
)
5735 return cmdline_parse_core(p
, &required_kernelcore
);
5739 * movablecore=size sets the amount of memory for use for allocations that
5740 * can be reclaimed or migrated.
5742 static int __init
cmdline_parse_movablecore(char *p
)
5744 return cmdline_parse_core(p
, &required_movablecore
);
5747 early_param("kernelcore", cmdline_parse_kernelcore
);
5748 early_param("movablecore", cmdline_parse_movablecore
);
5750 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5752 void adjust_managed_page_count(struct page
*page
, long count
)
5754 spin_lock(&managed_page_count_lock
);
5755 page_zone(page
)->managed_pages
+= count
;
5756 totalram_pages
+= count
;
5757 #ifdef CONFIG_HIGHMEM
5758 if (PageHighMem(page
))
5759 totalhigh_pages
+= count
;
5761 spin_unlock(&managed_page_count_lock
);
5763 EXPORT_SYMBOL(adjust_managed_page_count
);
5765 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
5768 unsigned long pages
= 0;
5770 start
= (void *)PAGE_ALIGN((unsigned long)start
);
5771 end
= (void *)((unsigned long)end
& PAGE_MASK
);
5772 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5773 if ((unsigned int)poison
<= 0xFF)
5774 memset(pos
, poison
, PAGE_SIZE
);
5775 free_reserved_page(virt_to_page(pos
));
5779 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5780 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
5784 EXPORT_SYMBOL(free_reserved_area
);
5786 #ifdef CONFIG_HIGHMEM
5787 void free_highmem_page(struct page
*page
)
5789 __free_reserved_page(page
);
5791 page_zone(page
)->managed_pages
++;
5797 void __init
mem_init_print_info(const char *str
)
5799 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
5800 unsigned long init_code_size
, init_data_size
;
5802 physpages
= get_num_physpages();
5803 codesize
= _etext
- _stext
;
5804 datasize
= _edata
- _sdata
;
5805 rosize
= __end_rodata
- __start_rodata
;
5806 bss_size
= __bss_stop
- __bss_start
;
5807 init_data_size
= __init_end
- __init_begin
;
5808 init_code_size
= _einittext
- _sinittext
;
5811 * Detect special cases and adjust section sizes accordingly:
5812 * 1) .init.* may be embedded into .data sections
5813 * 2) .init.text.* may be out of [__init_begin, __init_end],
5814 * please refer to arch/tile/kernel/vmlinux.lds.S.
5815 * 3) .rodata.* may be embedded into .text or .data sections.
5817 #define adj_init_size(start, end, size, pos, adj) \
5819 if (start <= pos && pos < end && size > adj) \
5823 adj_init_size(__init_begin
, __init_end
, init_data_size
,
5824 _sinittext
, init_code_size
);
5825 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
5826 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
5827 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
5828 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
5830 #undef adj_init_size
5832 pr_info("Memory: %luK/%luK available "
5833 "(%luK kernel code, %luK rwdata, %luK rodata, "
5834 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
5835 #ifdef CONFIG_HIGHMEM
5839 nr_free_pages() << (PAGE_SHIFT
-10), physpages
<< (PAGE_SHIFT
-10),
5840 codesize
>> 10, datasize
>> 10, rosize
>> 10,
5841 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
5842 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
-10),
5843 totalcma_pages
<< (PAGE_SHIFT
-10),
5844 #ifdef CONFIG_HIGHMEM
5845 totalhigh_pages
<< (PAGE_SHIFT
-10),
5847 str
? ", " : "", str
? str
: "");
5851 * set_dma_reserve - set the specified number of pages reserved in the first zone
5852 * @new_dma_reserve: The number of pages to mark reserved
5854 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
5855 * In the DMA zone, a significant percentage may be consumed by kernel image
5856 * and other unfreeable allocations which can skew the watermarks badly. This
5857 * function may optionally be used to account for unfreeable pages in the
5858 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5859 * smaller per-cpu batchsize.
5861 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5863 dma_reserve
= new_dma_reserve
;
5866 void __init
free_area_init(unsigned long *zones_size
)
5868 free_area_init_node(0, zones_size
,
5869 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5872 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5873 unsigned long action
, void *hcpu
)
5875 int cpu
= (unsigned long)hcpu
;
5877 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5878 lru_add_drain_cpu(cpu
);
5882 * Spill the event counters of the dead processor
5883 * into the current processors event counters.
5884 * This artificially elevates the count of the current
5887 vm_events_fold_cpu(cpu
);
5890 * Zero the differential counters of the dead processor
5891 * so that the vm statistics are consistent.
5893 * This is only okay since the processor is dead and cannot
5894 * race with what we are doing.
5896 cpu_vm_stats_fold(cpu
);
5901 void __init
page_alloc_init(void)
5903 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5907 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
5908 * or min_free_kbytes changes.
5910 static void calculate_totalreserve_pages(void)
5912 struct pglist_data
*pgdat
;
5913 unsigned long reserve_pages
= 0;
5914 enum zone_type i
, j
;
5916 for_each_online_pgdat(pgdat
) {
5917 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5918 struct zone
*zone
= pgdat
->node_zones
+ i
;
5921 /* Find valid and maximum lowmem_reserve in the zone */
5922 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5923 if (zone
->lowmem_reserve
[j
] > max
)
5924 max
= zone
->lowmem_reserve
[j
];
5927 /* we treat the high watermark as reserved pages. */
5928 max
+= high_wmark_pages(zone
);
5930 if (max
> zone
->managed_pages
)
5931 max
= zone
->managed_pages
;
5932 reserve_pages
+= max
;
5934 * Lowmem reserves are not available to
5935 * GFP_HIGHUSER page cache allocations and
5936 * kswapd tries to balance zones to their high
5937 * watermark. As a result, neither should be
5938 * regarded as dirtyable memory, to prevent a
5939 * situation where reclaim has to clean pages
5940 * in order to balance the zones.
5942 zone
->dirty_balance_reserve
= max
;
5945 dirty_balance_reserve
= reserve_pages
;
5946 totalreserve_pages
= reserve_pages
;
5950 * setup_per_zone_lowmem_reserve - called whenever
5951 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
5952 * has a correct pages reserved value, so an adequate number of
5953 * pages are left in the zone after a successful __alloc_pages().
5955 static void setup_per_zone_lowmem_reserve(void)
5957 struct pglist_data
*pgdat
;
5958 enum zone_type j
, idx
;
5960 for_each_online_pgdat(pgdat
) {
5961 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5962 struct zone
*zone
= pgdat
->node_zones
+ j
;
5963 unsigned long managed_pages
= zone
->managed_pages
;
5965 zone
->lowmem_reserve
[j
] = 0;
5969 struct zone
*lower_zone
;
5973 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5974 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5976 lower_zone
= pgdat
->node_zones
+ idx
;
5977 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
5978 sysctl_lowmem_reserve_ratio
[idx
];
5979 managed_pages
+= lower_zone
->managed_pages
;
5984 /* update totalreserve_pages */
5985 calculate_totalreserve_pages();
5988 static void __setup_per_zone_wmarks(void)
5990 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5991 unsigned long lowmem_pages
= 0;
5993 unsigned long flags
;
5995 /* Calculate total number of !ZONE_HIGHMEM pages */
5996 for_each_zone(zone
) {
5997 if (!is_highmem(zone
))
5998 lowmem_pages
+= zone
->managed_pages
;
6001 for_each_zone(zone
) {
6004 spin_lock_irqsave(&zone
->lock
, flags
);
6005 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6006 do_div(tmp
, lowmem_pages
);
6007 if (is_highmem(zone
)) {
6009 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6010 * need highmem pages, so cap pages_min to a small
6013 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6014 * deltas control asynch page reclaim, and so should
6015 * not be capped for highmem.
6017 unsigned long min_pages
;
6019 min_pages
= zone
->managed_pages
/ 1024;
6020 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6021 zone
->watermark
[WMARK_MIN
] = min_pages
;
6024 * If it's a lowmem zone, reserve a number of pages
6025 * proportionate to the zone's size.
6027 zone
->watermark
[WMARK_MIN
] = tmp
;
6030 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
6031 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
6033 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6034 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6035 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6037 spin_unlock_irqrestore(&zone
->lock
, flags
);
6040 /* update totalreserve_pages */
6041 calculate_totalreserve_pages();
6045 * setup_per_zone_wmarks - called when min_free_kbytes changes
6046 * or when memory is hot-{added|removed}
6048 * Ensures that the watermark[min,low,high] values for each zone are set
6049 * correctly with respect to min_free_kbytes.
6051 void setup_per_zone_wmarks(void)
6053 mutex_lock(&zonelists_mutex
);
6054 __setup_per_zone_wmarks();
6055 mutex_unlock(&zonelists_mutex
);
6059 * The inactive anon list should be small enough that the VM never has to
6060 * do too much work, but large enough that each inactive page has a chance
6061 * to be referenced again before it is swapped out.
6063 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6064 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6065 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6066 * the anonymous pages are kept on the inactive list.
6069 * memory ratio inactive anon
6070 * -------------------------------------
6079 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
6081 unsigned int gb
, ratio
;
6083 /* Zone size in gigabytes */
6084 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
6086 ratio
= int_sqrt(10 * gb
);
6090 zone
->inactive_ratio
= ratio
;
6093 static void __meminit
setup_per_zone_inactive_ratio(void)
6098 calculate_zone_inactive_ratio(zone
);
6102 * Initialise min_free_kbytes.
6104 * For small machines we want it small (128k min). For large machines
6105 * we want it large (64MB max). But it is not linear, because network
6106 * bandwidth does not increase linearly with machine size. We use
6108 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6109 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6125 int __meminit
init_per_zone_wmark_min(void)
6127 unsigned long lowmem_kbytes
;
6128 int new_min_free_kbytes
;
6130 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6131 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6133 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6134 min_free_kbytes
= new_min_free_kbytes
;
6135 if (min_free_kbytes
< 128)
6136 min_free_kbytes
= 128;
6137 if (min_free_kbytes
> 65536)
6138 min_free_kbytes
= 65536;
6140 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6141 new_min_free_kbytes
, user_min_free_kbytes
);
6143 setup_per_zone_wmarks();
6144 refresh_zone_stat_thresholds();
6145 setup_per_zone_lowmem_reserve();
6146 setup_per_zone_inactive_ratio();
6149 module_init(init_per_zone_wmark_min
)
6152 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6153 * that we can call two helper functions whenever min_free_kbytes
6156 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6157 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6161 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6166 user_min_free_kbytes
= min_free_kbytes
;
6167 setup_per_zone_wmarks();
6173 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6174 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6179 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6184 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6185 sysctl_min_unmapped_ratio
) / 100;
6189 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6190 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6195 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6200 zone
->min_slab_pages
= (zone
->managed_pages
*
6201 sysctl_min_slab_ratio
) / 100;
6207 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6208 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6209 * whenever sysctl_lowmem_reserve_ratio changes.
6211 * The reserve ratio obviously has absolutely no relation with the
6212 * minimum watermarks. The lowmem reserve ratio can only make sense
6213 * if in function of the boot time zone sizes.
6215 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6216 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6218 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6219 setup_per_zone_lowmem_reserve();
6224 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6225 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6226 * pagelist can have before it gets flushed back to buddy allocator.
6228 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
6229 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6232 int old_percpu_pagelist_fraction
;
6235 mutex_lock(&pcp_batch_high_lock
);
6236 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
6238 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6239 if (!write
|| ret
< 0)
6242 /* Sanity checking to avoid pcp imbalance */
6243 if (percpu_pagelist_fraction
&&
6244 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
6245 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
6251 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
6254 for_each_populated_zone(zone
) {
6257 for_each_possible_cpu(cpu
)
6258 pageset_set_high_and_batch(zone
,
6259 per_cpu_ptr(zone
->pageset
, cpu
));
6262 mutex_unlock(&pcp_batch_high_lock
);
6267 int hashdist
= HASHDIST_DEFAULT
;
6269 static int __init
set_hashdist(char *str
)
6273 hashdist
= simple_strtoul(str
, &str
, 0);
6276 __setup("hashdist=", set_hashdist
);
6280 * allocate a large system hash table from bootmem
6281 * - it is assumed that the hash table must contain an exact power-of-2
6282 * quantity of entries
6283 * - limit is the number of hash buckets, not the total allocation size
6285 void *__init
alloc_large_system_hash(const char *tablename
,
6286 unsigned long bucketsize
,
6287 unsigned long numentries
,
6290 unsigned int *_hash_shift
,
6291 unsigned int *_hash_mask
,
6292 unsigned long low_limit
,
6293 unsigned long high_limit
)
6295 unsigned long long max
= high_limit
;
6296 unsigned long log2qty
, size
;
6299 /* allow the kernel cmdline to have a say */
6301 /* round applicable memory size up to nearest megabyte */
6302 numentries
= nr_kernel_pages
;
6304 /* It isn't necessary when PAGE_SIZE >= 1MB */
6305 if (PAGE_SHIFT
< 20)
6306 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
6308 /* limit to 1 bucket per 2^scale bytes of low memory */
6309 if (scale
> PAGE_SHIFT
)
6310 numentries
>>= (scale
- PAGE_SHIFT
);
6312 numentries
<<= (PAGE_SHIFT
- scale
);
6314 /* Make sure we've got at least a 0-order allocation.. */
6315 if (unlikely(flags
& HASH_SMALL
)) {
6316 /* Makes no sense without HASH_EARLY */
6317 WARN_ON(!(flags
& HASH_EARLY
));
6318 if (!(numentries
>> *_hash_shift
)) {
6319 numentries
= 1UL << *_hash_shift
;
6320 BUG_ON(!numentries
);
6322 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
6323 numentries
= PAGE_SIZE
/ bucketsize
;
6325 numentries
= roundup_pow_of_two(numentries
);
6327 /* limit allocation size to 1/16 total memory by default */
6329 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
6330 do_div(max
, bucketsize
);
6332 max
= min(max
, 0x80000000ULL
);
6334 if (numentries
< low_limit
)
6335 numentries
= low_limit
;
6336 if (numentries
> max
)
6339 log2qty
= ilog2(numentries
);
6342 size
= bucketsize
<< log2qty
;
6343 if (flags
& HASH_EARLY
)
6344 table
= memblock_virt_alloc_nopanic(size
, 0);
6346 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
6349 * If bucketsize is not a power-of-two, we may free
6350 * some pages at the end of hash table which
6351 * alloc_pages_exact() automatically does
6353 if (get_order(size
) < MAX_ORDER
) {
6354 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
6355 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
6358 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
6361 panic("Failed to allocate %s hash table\n", tablename
);
6363 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
6366 ilog2(size
) - PAGE_SHIFT
,
6370 *_hash_shift
= log2qty
;
6372 *_hash_mask
= (1 << log2qty
) - 1;
6377 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6378 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
6381 #ifdef CONFIG_SPARSEMEM
6382 return __pfn_to_section(pfn
)->pageblock_flags
;
6384 return zone
->pageblock_flags
;
6385 #endif /* CONFIG_SPARSEMEM */
6388 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
6390 #ifdef CONFIG_SPARSEMEM
6391 pfn
&= (PAGES_PER_SECTION
-1);
6392 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6394 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
6395 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6396 #endif /* CONFIG_SPARSEMEM */
6400 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6401 * @page: The page within the block of interest
6402 * @pfn: The target page frame number
6403 * @end_bitidx: The last bit of interest to retrieve
6404 * @mask: mask of bits that the caller is interested in
6406 * Return: pageblock_bits flags
6408 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
6409 unsigned long end_bitidx
,
6413 unsigned long *bitmap
;
6414 unsigned long bitidx
, word_bitidx
;
6417 zone
= page_zone(page
);
6418 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6419 bitidx
= pfn_to_bitidx(zone
, pfn
);
6420 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6421 bitidx
&= (BITS_PER_LONG
-1);
6423 word
= bitmap
[word_bitidx
];
6424 bitidx
+= end_bitidx
;
6425 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
6429 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6430 * @page: The page within the block of interest
6431 * @flags: The flags to set
6432 * @pfn: The target page frame number
6433 * @end_bitidx: The last bit of interest
6434 * @mask: mask of bits that the caller is interested in
6436 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
6438 unsigned long end_bitidx
,
6442 unsigned long *bitmap
;
6443 unsigned long bitidx
, word_bitidx
;
6444 unsigned long old_word
, word
;
6446 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
6448 zone
= page_zone(page
);
6449 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6450 bitidx
= pfn_to_bitidx(zone
, pfn
);
6451 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6452 bitidx
&= (BITS_PER_LONG
-1);
6454 VM_BUG_ON_PAGE(!zone_spans_pfn(zone
, pfn
), page
);
6456 bitidx
+= end_bitidx
;
6457 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
6458 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
6460 word
= READ_ONCE(bitmap
[word_bitidx
]);
6462 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
6463 if (word
== old_word
)
6470 * This function checks whether pageblock includes unmovable pages or not.
6471 * If @count is not zero, it is okay to include less @count unmovable pages
6473 * PageLRU check without isolation or lru_lock could race so that
6474 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6475 * expect this function should be exact.
6477 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6478 bool skip_hwpoisoned_pages
)
6480 unsigned long pfn
, iter
, found
;
6484 * For avoiding noise data, lru_add_drain_all() should be called
6485 * If ZONE_MOVABLE, the zone never contains unmovable pages
6487 if (zone_idx(zone
) == ZONE_MOVABLE
)
6489 mt
= get_pageblock_migratetype(page
);
6490 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6493 pfn
= page_to_pfn(page
);
6494 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6495 unsigned long check
= pfn
+ iter
;
6497 if (!pfn_valid_within(check
))
6500 page
= pfn_to_page(check
);
6503 * Hugepages are not in LRU lists, but they're movable.
6504 * We need not scan over tail pages bacause we don't
6505 * handle each tail page individually in migration.
6507 if (PageHuge(page
)) {
6508 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6513 * We can't use page_count without pin a page
6514 * because another CPU can free compound page.
6515 * This check already skips compound tails of THP
6516 * because their page->_count is zero at all time.
6518 if (!atomic_read(&page
->_count
)) {
6519 if (PageBuddy(page
))
6520 iter
+= (1 << page_order(page
)) - 1;
6525 * The HWPoisoned page may be not in buddy system, and
6526 * page_count() is not 0.
6528 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6534 * If there are RECLAIMABLE pages, we need to check
6535 * it. But now, memory offline itself doesn't call
6536 * shrink_node_slabs() and it still to be fixed.
6539 * If the page is not RAM, page_count()should be 0.
6540 * we don't need more check. This is an _used_ not-movable page.
6542 * The problematic thing here is PG_reserved pages. PG_reserved
6543 * is set to both of a memory hole page and a _used_ kernel
6552 bool is_pageblock_removable_nolock(struct page
*page
)
6558 * We have to be careful here because we are iterating over memory
6559 * sections which are not zone aware so we might end up outside of
6560 * the zone but still within the section.
6561 * We have to take care about the node as well. If the node is offline
6562 * its NODE_DATA will be NULL - see page_zone.
6564 if (!node_online(page_to_nid(page
)))
6567 zone
= page_zone(page
);
6568 pfn
= page_to_pfn(page
);
6569 if (!zone_spans_pfn(zone
, pfn
))
6572 return !has_unmovable_pages(zone
, page
, 0, true);
6577 static unsigned long pfn_max_align_down(unsigned long pfn
)
6579 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6580 pageblock_nr_pages
) - 1);
6583 static unsigned long pfn_max_align_up(unsigned long pfn
)
6585 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6586 pageblock_nr_pages
));
6589 /* [start, end) must belong to a single zone. */
6590 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6591 unsigned long start
, unsigned long end
)
6593 /* This function is based on compact_zone() from compaction.c. */
6594 unsigned long nr_reclaimed
;
6595 unsigned long pfn
= start
;
6596 unsigned int tries
= 0;
6601 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6602 if (fatal_signal_pending(current
)) {
6607 if (list_empty(&cc
->migratepages
)) {
6608 cc
->nr_migratepages
= 0;
6609 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
6615 } else if (++tries
== 5) {
6616 ret
= ret
< 0 ? ret
: -EBUSY
;
6620 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6622 cc
->nr_migratepages
-= nr_reclaimed
;
6624 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6625 NULL
, 0, cc
->mode
, MR_CMA
);
6628 putback_movable_pages(&cc
->migratepages
);
6635 * alloc_contig_range() -- tries to allocate given range of pages
6636 * @start: start PFN to allocate
6637 * @end: one-past-the-last PFN to allocate
6638 * @migratetype: migratetype of the underlaying pageblocks (either
6639 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6640 * in range must have the same migratetype and it must
6641 * be either of the two.
6643 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6644 * aligned, however it's the caller's responsibility to guarantee that
6645 * we are the only thread that changes migrate type of pageblocks the
6648 * The PFN range must belong to a single zone.
6650 * Returns zero on success or negative error code. On success all
6651 * pages which PFN is in [start, end) are allocated for the caller and
6652 * need to be freed with free_contig_range().
6654 int alloc_contig_range(unsigned long start
, unsigned long end
,
6655 unsigned migratetype
)
6657 unsigned long outer_start
, outer_end
;
6660 struct compact_control cc
= {
6661 .nr_migratepages
= 0,
6663 .zone
= page_zone(pfn_to_page(start
)),
6664 .mode
= MIGRATE_SYNC
,
6665 .ignore_skip_hint
= true,
6667 INIT_LIST_HEAD(&cc
.migratepages
);
6670 * What we do here is we mark all pageblocks in range as
6671 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6672 * have different sizes, and due to the way page allocator
6673 * work, we align the range to biggest of the two pages so
6674 * that page allocator won't try to merge buddies from
6675 * different pageblocks and change MIGRATE_ISOLATE to some
6676 * other migration type.
6678 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6679 * migrate the pages from an unaligned range (ie. pages that
6680 * we are interested in). This will put all the pages in
6681 * range back to page allocator as MIGRATE_ISOLATE.
6683 * When this is done, we take the pages in range from page
6684 * allocator removing them from the buddy system. This way
6685 * page allocator will never consider using them.
6687 * This lets us mark the pageblocks back as
6688 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6689 * aligned range but not in the unaligned, original range are
6690 * put back to page allocator so that buddy can use them.
6693 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6694 pfn_max_align_up(end
), migratetype
,
6699 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6704 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6705 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6706 * more, all pages in [start, end) are free in page allocator.
6707 * What we are going to do is to allocate all pages from
6708 * [start, end) (that is remove them from page allocator).
6710 * The only problem is that pages at the beginning and at the
6711 * end of interesting range may be not aligned with pages that
6712 * page allocator holds, ie. they can be part of higher order
6713 * pages. Because of this, we reserve the bigger range and
6714 * once this is done free the pages we are not interested in.
6716 * We don't have to hold zone->lock here because the pages are
6717 * isolated thus they won't get removed from buddy.
6720 lru_add_drain_all();
6721 drain_all_pages(cc
.zone
);
6724 outer_start
= start
;
6725 while (!PageBuddy(pfn_to_page(outer_start
))) {
6726 if (++order
>= MAX_ORDER
) {
6730 outer_start
&= ~0UL << order
;
6733 /* Make sure the range is really isolated. */
6734 if (test_pages_isolated(outer_start
, end
, false)) {
6735 pr_info("%s: [%lx, %lx) PFNs busy\n",
6736 __func__
, outer_start
, end
);
6741 /* Grab isolated pages from freelists. */
6742 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6748 /* Free head and tail (if any) */
6749 if (start
!= outer_start
)
6750 free_contig_range(outer_start
, start
- outer_start
);
6751 if (end
!= outer_end
)
6752 free_contig_range(end
, outer_end
- end
);
6755 undo_isolate_page_range(pfn_max_align_down(start
),
6756 pfn_max_align_up(end
), migratetype
);
6760 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6762 unsigned int count
= 0;
6764 for (; nr_pages
--; pfn
++) {
6765 struct page
*page
= pfn_to_page(pfn
);
6767 count
+= page_count(page
) != 1;
6770 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6774 #ifdef CONFIG_MEMORY_HOTPLUG
6776 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6777 * page high values need to be recalulated.
6779 void __meminit
zone_pcp_update(struct zone
*zone
)
6782 mutex_lock(&pcp_batch_high_lock
);
6783 for_each_possible_cpu(cpu
)
6784 pageset_set_high_and_batch(zone
,
6785 per_cpu_ptr(zone
->pageset
, cpu
));
6786 mutex_unlock(&pcp_batch_high_lock
);
6790 void zone_pcp_reset(struct zone
*zone
)
6792 unsigned long flags
;
6794 struct per_cpu_pageset
*pset
;
6796 /* avoid races with drain_pages() */
6797 local_irq_save(flags
);
6798 if (zone
->pageset
!= &boot_pageset
) {
6799 for_each_online_cpu(cpu
) {
6800 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6801 drain_zonestat(zone
, pset
);
6803 free_percpu(zone
->pageset
);
6804 zone
->pageset
= &boot_pageset
;
6806 local_irq_restore(flags
);
6809 #ifdef CONFIG_MEMORY_HOTREMOVE
6811 * All pages in the range must be isolated before calling this.
6814 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6818 unsigned int order
, i
;
6820 unsigned long flags
;
6821 /* find the first valid pfn */
6822 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6827 zone
= page_zone(pfn_to_page(pfn
));
6828 spin_lock_irqsave(&zone
->lock
, flags
);
6830 while (pfn
< end_pfn
) {
6831 if (!pfn_valid(pfn
)) {
6835 page
= pfn_to_page(pfn
);
6837 * The HWPoisoned page may be not in buddy system, and
6838 * page_count() is not 0.
6840 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6842 SetPageReserved(page
);
6846 BUG_ON(page_count(page
));
6847 BUG_ON(!PageBuddy(page
));
6848 order
= page_order(page
);
6849 #ifdef CONFIG_DEBUG_VM
6850 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6851 pfn
, 1 << order
, end_pfn
);
6853 list_del(&page
->lru
);
6854 rmv_page_order(page
);
6855 zone
->free_area
[order
].nr_free
--;
6856 for (i
= 0; i
< (1 << order
); i
++)
6857 SetPageReserved((page
+i
));
6858 pfn
+= (1 << order
);
6860 spin_unlock_irqrestore(&zone
->lock
, flags
);
6864 #ifdef CONFIG_MEMORY_FAILURE
6865 bool is_free_buddy_page(struct page
*page
)
6867 struct zone
*zone
= page_zone(page
);
6868 unsigned long pfn
= page_to_pfn(page
);
6869 unsigned long flags
;
6872 spin_lock_irqsave(&zone
->lock
, flags
);
6873 for (order
= 0; order
< MAX_ORDER
; order
++) {
6874 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6876 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6879 spin_unlock_irqrestore(&zone
->lock
, flags
);
6881 return order
< MAX_ORDER
;