2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/notifier.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/stop_machine.h>
46 #include <linux/sort.h>
47 #include <linux/pfn.h>
48 #include <linux/backing-dev.h>
49 #include <linux/fault-inject.h>
50 #include <linux/page-isolation.h>
51 #include <linux/page_cgroup.h>
52 #include <linux/debugobjects.h>
53 #include <linux/kmemleak.h>
54 #include <linux/compaction.h>
55 #include <trace/events/kmem.h>
56 #include <linux/ftrace_event.h>
57 #include <linux/memcontrol.h>
58 #include <linux/prefetch.h>
59 #include <linux/mm_inline.h>
60 #include <linux/migrate.h>
61 #include <linux/page-debug-flags.h>
62 #include <linux/hugetlb.h>
63 #include <linux/sched/rt.h>
65 #include <asm/sections.h>
66 #include <asm/tlbflush.h>
67 #include <asm/div64.h>
70 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
71 static DEFINE_MUTEX(pcp_batch_high_lock
);
73 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
74 DEFINE_PER_CPU(int, numa_node
);
75 EXPORT_PER_CPU_SYMBOL(numa_node
);
78 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
80 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
81 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
82 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
83 * defined in <linux/topology.h>.
85 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
86 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
90 * Array of node states.
92 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
93 [N_POSSIBLE
] = NODE_MASK_ALL
,
94 [N_ONLINE
] = { { [0] = 1UL } },
96 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
98 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
100 #ifdef CONFIG_MOVABLE_NODE
101 [N_MEMORY
] = { { [0] = 1UL } },
103 [N_CPU
] = { { [0] = 1UL } },
106 EXPORT_SYMBOL(node_states
);
108 /* Protect totalram_pages and zone->managed_pages */
109 static DEFINE_SPINLOCK(managed_page_count_lock
);
111 unsigned long totalram_pages __read_mostly
;
112 unsigned long totalreserve_pages __read_mostly
;
114 * When calculating the number of globally allowed dirty pages, there
115 * is a certain number of per-zone reserves that should not be
116 * considered dirtyable memory. This is the sum of those reserves
117 * over all existing zones that contribute dirtyable memory.
119 unsigned long dirty_balance_reserve __read_mostly
;
121 int percpu_pagelist_fraction
;
122 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
124 #ifdef CONFIG_PM_SLEEP
126 * The following functions are used by the suspend/hibernate code to temporarily
127 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
128 * while devices are suspended. To avoid races with the suspend/hibernate code,
129 * they should always be called with pm_mutex held (gfp_allowed_mask also should
130 * only be modified with pm_mutex held, unless the suspend/hibernate code is
131 * guaranteed not to run in parallel with that modification).
134 static gfp_t saved_gfp_mask
;
136 void pm_restore_gfp_mask(void)
138 WARN_ON(!mutex_is_locked(&pm_mutex
));
139 if (saved_gfp_mask
) {
140 gfp_allowed_mask
= saved_gfp_mask
;
145 void pm_restrict_gfp_mask(void)
147 WARN_ON(!mutex_is_locked(&pm_mutex
));
148 WARN_ON(saved_gfp_mask
);
149 saved_gfp_mask
= gfp_allowed_mask
;
150 gfp_allowed_mask
&= ~GFP_IOFS
;
153 bool pm_suspended_storage(void)
155 if ((gfp_allowed_mask
& GFP_IOFS
) == GFP_IOFS
)
159 #endif /* CONFIG_PM_SLEEP */
161 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
162 int pageblock_order __read_mostly
;
165 static void __free_pages_ok(struct page
*page
, unsigned int order
);
168 * results with 256, 32 in the lowmem_reserve sysctl:
169 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
170 * 1G machine -> (16M dma, 784M normal, 224M high)
171 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
172 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
173 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
175 * TBD: should special case ZONE_DMA32 machines here - in those we normally
176 * don't need any ZONE_NORMAL reservation
178 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
179 #ifdef CONFIG_ZONE_DMA
182 #ifdef CONFIG_ZONE_DMA32
185 #ifdef CONFIG_HIGHMEM
191 EXPORT_SYMBOL(totalram_pages
);
193 static char * const zone_names
[MAX_NR_ZONES
] = {
194 #ifdef CONFIG_ZONE_DMA
197 #ifdef CONFIG_ZONE_DMA32
201 #ifdef CONFIG_HIGHMEM
207 int min_free_kbytes
= 1024;
208 int user_min_free_kbytes
;
210 static unsigned long __meminitdata nr_kernel_pages
;
211 static unsigned long __meminitdata nr_all_pages
;
212 static unsigned long __meminitdata dma_reserve
;
214 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
215 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
216 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
217 static unsigned long __initdata required_kernelcore
;
218 static unsigned long __initdata required_movablecore
;
219 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
221 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
223 EXPORT_SYMBOL(movable_zone
);
224 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
227 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
228 int nr_online_nodes __read_mostly
= 1;
229 EXPORT_SYMBOL(nr_node_ids
);
230 EXPORT_SYMBOL(nr_online_nodes
);
233 int page_group_by_mobility_disabled __read_mostly
;
235 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
237 if (unlikely(page_group_by_mobility_disabled
&&
238 migratetype
< MIGRATE_PCPTYPES
))
239 migratetype
= MIGRATE_UNMOVABLE
;
241 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
242 PB_migrate
, PB_migrate_end
);
245 bool oom_killer_disabled __read_mostly
;
247 #ifdef CONFIG_DEBUG_VM
248 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
252 unsigned long pfn
= page_to_pfn(page
);
253 unsigned long sp
, start_pfn
;
256 seq
= zone_span_seqbegin(zone
);
257 start_pfn
= zone
->zone_start_pfn
;
258 sp
= zone
->spanned_pages
;
259 if (!zone_spans_pfn(zone
, pfn
))
261 } while (zone_span_seqretry(zone
, seq
));
264 pr_err("page %lu outside zone [ %lu - %lu ]\n",
265 pfn
, start_pfn
, start_pfn
+ sp
);
270 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
272 if (!pfn_valid_within(page_to_pfn(page
)))
274 if (zone
!= page_zone(page
))
280 * Temporary debugging check for pages not lying within a given zone.
282 static int bad_range(struct zone
*zone
, struct page
*page
)
284 if (page_outside_zone_boundaries(zone
, page
))
286 if (!page_is_consistent(zone
, page
))
292 static inline int bad_range(struct zone
*zone
, struct page
*page
)
298 static void bad_page(struct page
*page
)
300 static unsigned long resume
;
301 static unsigned long nr_shown
;
302 static unsigned long nr_unshown
;
304 /* Don't complain about poisoned pages */
305 if (PageHWPoison(page
)) {
306 page_mapcount_reset(page
); /* remove PageBuddy */
311 * Allow a burst of 60 reports, then keep quiet for that minute;
312 * or allow a steady drip of one report per second.
314 if (nr_shown
== 60) {
315 if (time_before(jiffies
, resume
)) {
321 "BUG: Bad page state: %lu messages suppressed\n",
328 resume
= jiffies
+ 60 * HZ
;
330 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
331 current
->comm
, page_to_pfn(page
));
337 /* Leave bad fields for debug, except PageBuddy could make trouble */
338 page_mapcount_reset(page
); /* remove PageBuddy */
339 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
343 * Higher-order pages are called "compound pages". They are structured thusly:
345 * The first PAGE_SIZE page is called the "head page".
347 * The remaining PAGE_SIZE pages are called "tail pages".
349 * All pages have PG_compound set. All tail pages have their ->first_page
350 * pointing at the head page.
352 * The first tail page's ->lru.next holds the address of the compound page's
353 * put_page() function. Its ->lru.prev holds the order of allocation.
354 * This usage means that zero-order pages may not be compound.
357 static void free_compound_page(struct page
*page
)
359 __free_pages_ok(page
, compound_order(page
));
362 void prep_compound_page(struct page
*page
, unsigned long order
)
365 int nr_pages
= 1 << order
;
367 set_compound_page_dtor(page
, free_compound_page
);
368 set_compound_order(page
, order
);
370 for (i
= 1; i
< nr_pages
; i
++) {
371 struct page
*p
= page
+ i
;
373 set_page_count(p
, 0);
374 p
->first_page
= page
;
378 /* update __split_huge_page_refcount if you change this function */
379 static int destroy_compound_page(struct page
*page
, unsigned long order
)
382 int nr_pages
= 1 << order
;
385 if (unlikely(compound_order(page
) != order
)) {
390 __ClearPageHead(page
);
392 for (i
= 1; i
< nr_pages
; i
++) {
393 struct page
*p
= page
+ i
;
395 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
405 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
410 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
411 * and __GFP_HIGHMEM from hard or soft interrupt context.
413 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
414 for (i
= 0; i
< (1 << order
); i
++)
415 clear_highpage(page
+ i
);
418 #ifdef CONFIG_DEBUG_PAGEALLOC
419 unsigned int _debug_guardpage_minorder
;
421 static int __init
debug_guardpage_minorder_setup(char *buf
)
425 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
426 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
429 _debug_guardpage_minorder
= res
;
430 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
433 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
435 static inline void set_page_guard_flag(struct page
*page
)
437 __set_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
440 static inline void clear_page_guard_flag(struct page
*page
)
442 __clear_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
445 static inline void set_page_guard_flag(struct page
*page
) { }
446 static inline void clear_page_guard_flag(struct page
*page
) { }
449 static inline void set_page_order(struct page
*page
, int order
)
451 set_page_private(page
, order
);
452 __SetPageBuddy(page
);
455 static inline void rmv_page_order(struct page
*page
)
457 __ClearPageBuddy(page
);
458 set_page_private(page
, 0);
462 * Locate the struct page for both the matching buddy in our
463 * pair (buddy1) and the combined O(n+1) page they form (page).
465 * 1) Any buddy B1 will have an order O twin B2 which satisfies
466 * the following equation:
468 * For example, if the starting buddy (buddy2) is #8 its order
470 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
472 * 2) Any buddy B will have an order O+1 parent P which
473 * satisfies the following equation:
476 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
478 static inline unsigned long
479 __find_buddy_index(unsigned long page_idx
, unsigned int order
)
481 return page_idx
^ (1 << order
);
485 * This function checks whether a page is free && is the buddy
486 * we can do coalesce a page and its buddy if
487 * (a) the buddy is not in a hole &&
488 * (b) the buddy is in the buddy system &&
489 * (c) a page and its buddy have the same order &&
490 * (d) a page and its buddy are in the same zone.
492 * For recording whether a page is in the buddy system, we set ->_mapcount
493 * PAGE_BUDDY_MAPCOUNT_VALUE.
494 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
495 * serialized by zone->lock.
497 * For recording page's order, we use page_private(page).
499 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
502 if (!pfn_valid_within(page_to_pfn(buddy
)))
505 if (page_zone_id(page
) != page_zone_id(buddy
))
508 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
509 VM_BUG_ON(page_count(buddy
) != 0);
513 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
514 VM_BUG_ON(page_count(buddy
) != 0);
521 * Freeing function for a buddy system allocator.
523 * The concept of a buddy system is to maintain direct-mapped table
524 * (containing bit values) for memory blocks of various "orders".
525 * The bottom level table contains the map for the smallest allocatable
526 * units of memory (here, pages), and each level above it describes
527 * pairs of units from the levels below, hence, "buddies".
528 * At a high level, all that happens here is marking the table entry
529 * at the bottom level available, and propagating the changes upward
530 * as necessary, plus some accounting needed to play nicely with other
531 * parts of the VM system.
532 * At each level, we keep a list of pages, which are heads of continuous
533 * free pages of length of (1 << order) and marked with _mapcount
534 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
536 * So when we are allocating or freeing one, we can derive the state of the
537 * other. That is, if we allocate a small block, and both were
538 * free, the remainder of the region must be split into blocks.
539 * If a block is freed, and its buddy is also free, then this
540 * triggers coalescing into a block of larger size.
545 static inline void __free_one_page(struct page
*page
,
546 struct zone
*zone
, unsigned int order
,
549 unsigned long page_idx
;
550 unsigned long combined_idx
;
551 unsigned long uninitialized_var(buddy_idx
);
554 VM_BUG_ON(!zone_is_initialized(zone
));
556 if (unlikely(PageCompound(page
)))
557 if (unlikely(destroy_compound_page(page
, order
)))
560 VM_BUG_ON(migratetype
== -1);
562 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
564 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
565 VM_BUG_ON(bad_range(zone
, page
));
567 while (order
< MAX_ORDER
-1) {
568 buddy_idx
= __find_buddy_index(page_idx
, order
);
569 buddy
= page
+ (buddy_idx
- page_idx
);
570 if (!page_is_buddy(page
, buddy
, order
))
573 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
574 * merge with it and move up one order.
576 if (page_is_guard(buddy
)) {
577 clear_page_guard_flag(buddy
);
578 set_page_private(page
, 0);
579 __mod_zone_freepage_state(zone
, 1 << order
,
582 list_del(&buddy
->lru
);
583 zone
->free_area
[order
].nr_free
--;
584 rmv_page_order(buddy
);
586 combined_idx
= buddy_idx
& page_idx
;
587 page
= page
+ (combined_idx
- page_idx
);
588 page_idx
= combined_idx
;
591 set_page_order(page
, order
);
594 * If this is not the largest possible page, check if the buddy
595 * of the next-highest order is free. If it is, it's possible
596 * that pages are being freed that will coalesce soon. In case,
597 * that is happening, add the free page to the tail of the list
598 * so it's less likely to be used soon and more likely to be merged
599 * as a higher order page
601 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
602 struct page
*higher_page
, *higher_buddy
;
603 combined_idx
= buddy_idx
& page_idx
;
604 higher_page
= page
+ (combined_idx
- page_idx
);
605 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
606 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
607 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
608 list_add_tail(&page
->lru
,
609 &zone
->free_area
[order
].free_list
[migratetype
]);
614 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
616 zone
->free_area
[order
].nr_free
++;
619 static inline int free_pages_check(struct page
*page
)
621 if (unlikely(page_mapcount(page
) |
622 (page
->mapping
!= NULL
) |
623 (atomic_read(&page
->_count
) != 0) |
624 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
) |
625 (mem_cgroup_bad_page_check(page
)))) {
629 page_cpupid_reset_last(page
);
630 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
631 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
636 * Frees a number of pages from the PCP lists
637 * Assumes all pages on list are in same zone, and of same order.
638 * count is the number of pages to free.
640 * If the zone was previously in an "all pages pinned" state then look to
641 * see if this freeing clears that state.
643 * And clear the zone's pages_scanned counter, to hold off the "all pages are
644 * pinned" detection logic.
646 static void free_pcppages_bulk(struct zone
*zone
, int count
,
647 struct per_cpu_pages
*pcp
)
653 spin_lock(&zone
->lock
);
654 zone
->pages_scanned
= 0;
658 struct list_head
*list
;
661 * Remove pages from lists in a round-robin fashion. A
662 * batch_free count is maintained that is incremented when an
663 * empty list is encountered. This is so more pages are freed
664 * off fuller lists instead of spinning excessively around empty
669 if (++migratetype
== MIGRATE_PCPTYPES
)
671 list
= &pcp
->lists
[migratetype
];
672 } while (list_empty(list
));
674 /* This is the only non-empty list. Free them all. */
675 if (batch_free
== MIGRATE_PCPTYPES
)
676 batch_free
= to_free
;
679 int mt
; /* migratetype of the to-be-freed page */
681 page
= list_entry(list
->prev
, struct page
, lru
);
682 /* must delete as __free_one_page list manipulates */
683 list_del(&page
->lru
);
684 mt
= get_freepage_migratetype(page
);
685 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
686 __free_one_page(page
, zone
, 0, mt
);
687 trace_mm_page_pcpu_drain(page
, 0, mt
);
688 if (likely(!is_migrate_isolate_page(page
))) {
689 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1);
690 if (is_migrate_cma(mt
))
691 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
, 1);
693 } while (--to_free
&& --batch_free
&& !list_empty(list
));
695 spin_unlock(&zone
->lock
);
698 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
701 spin_lock(&zone
->lock
);
702 zone
->pages_scanned
= 0;
704 __free_one_page(page
, zone
, order
, migratetype
);
705 if (unlikely(!is_migrate_isolate(migratetype
)))
706 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
707 spin_unlock(&zone
->lock
);
710 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
715 trace_mm_page_free(page
, order
);
716 kmemcheck_free_shadow(page
, order
);
719 page
->mapping
= NULL
;
720 for (i
= 0; i
< (1 << order
); i
++)
721 bad
+= free_pages_check(page
+ i
);
725 if (!PageHighMem(page
)) {
726 debug_check_no_locks_freed(page_address(page
),
728 debug_check_no_obj_freed(page_address(page
),
731 arch_free_page(page
, order
);
732 kernel_map_pages(page
, 1 << order
, 0);
737 static void __free_pages_ok(struct page
*page
, unsigned int order
)
742 if (!free_pages_prepare(page
, order
))
745 local_irq_save(flags
);
746 __count_vm_events(PGFREE
, 1 << order
);
747 migratetype
= get_pageblock_migratetype(page
);
748 set_freepage_migratetype(page
, migratetype
);
749 free_one_page(page_zone(page
), page
, order
, migratetype
);
750 local_irq_restore(flags
);
753 void __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
755 unsigned int nr_pages
= 1 << order
;
756 struct page
*p
= page
;
760 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
762 __ClearPageReserved(p
);
763 set_page_count(p
, 0);
765 __ClearPageReserved(p
);
766 set_page_count(p
, 0);
768 page_zone(page
)->managed_pages
+= nr_pages
;
769 set_page_refcounted(page
);
770 __free_pages(page
, order
);
774 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
775 void __init
init_cma_reserved_pageblock(struct page
*page
)
777 unsigned i
= pageblock_nr_pages
;
778 struct page
*p
= page
;
781 __ClearPageReserved(p
);
782 set_page_count(p
, 0);
785 set_page_refcounted(page
);
786 set_pageblock_migratetype(page
, MIGRATE_CMA
);
787 __free_pages(page
, pageblock_order
);
788 adjust_managed_page_count(page
, pageblock_nr_pages
);
793 * The order of subdivision here is critical for the IO subsystem.
794 * Please do not alter this order without good reasons and regression
795 * testing. Specifically, as large blocks of memory are subdivided,
796 * the order in which smaller blocks are delivered depends on the order
797 * they're subdivided in this function. This is the primary factor
798 * influencing the order in which pages are delivered to the IO
799 * subsystem according to empirical testing, and this is also justified
800 * by considering the behavior of a buddy system containing a single
801 * large block of memory acted on by a series of small allocations.
802 * This behavior is a critical factor in sglist merging's success.
806 static inline void expand(struct zone
*zone
, struct page
*page
,
807 int low
, int high
, struct free_area
*area
,
810 unsigned long size
= 1 << high
;
816 VM_BUG_ON(bad_range(zone
, &page
[size
]));
818 #ifdef CONFIG_DEBUG_PAGEALLOC
819 if (high
< debug_guardpage_minorder()) {
821 * Mark as guard pages (or page), that will allow to
822 * merge back to allocator when buddy will be freed.
823 * Corresponding page table entries will not be touched,
824 * pages will stay not present in virtual address space
826 INIT_LIST_HEAD(&page
[size
].lru
);
827 set_page_guard_flag(&page
[size
]);
828 set_page_private(&page
[size
], high
);
829 /* Guard pages are not available for any usage */
830 __mod_zone_freepage_state(zone
, -(1 << high
),
835 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
837 set_page_order(&page
[size
], high
);
842 * This page is about to be returned from the page allocator
844 static inline int check_new_page(struct page
*page
)
846 if (unlikely(page_mapcount(page
) |
847 (page
->mapping
!= NULL
) |
848 (atomic_read(&page
->_count
) != 0) |
849 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
) |
850 (mem_cgroup_bad_page_check(page
)))) {
857 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
861 for (i
= 0; i
< (1 << order
); i
++) {
862 struct page
*p
= page
+ i
;
863 if (unlikely(check_new_page(p
)))
867 set_page_private(page
, 0);
868 set_page_refcounted(page
);
870 arch_alloc_page(page
, order
);
871 kernel_map_pages(page
, 1 << order
, 1);
873 if (gfp_flags
& __GFP_ZERO
)
874 prep_zero_page(page
, order
, gfp_flags
);
876 if (order
&& (gfp_flags
& __GFP_COMP
))
877 prep_compound_page(page
, order
);
883 * Go through the free lists for the given migratetype and remove
884 * the smallest available page from the freelists
887 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
890 unsigned int current_order
;
891 struct free_area
*area
;
894 /* Find a page of the appropriate size in the preferred list */
895 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
896 area
= &(zone
->free_area
[current_order
]);
897 if (list_empty(&area
->free_list
[migratetype
]))
900 page
= list_entry(area
->free_list
[migratetype
].next
,
902 list_del(&page
->lru
);
903 rmv_page_order(page
);
905 expand(zone
, page
, order
, current_order
, area
, migratetype
);
914 * This array describes the order lists are fallen back to when
915 * the free lists for the desirable migrate type are depleted
917 static int fallbacks
[MIGRATE_TYPES
][4] = {
918 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
919 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
921 [MIGRATE_MOVABLE
] = { MIGRATE_CMA
, MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
922 [MIGRATE_CMA
] = { MIGRATE_RESERVE
}, /* Never used */
924 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
926 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
}, /* Never used */
927 #ifdef CONFIG_MEMORY_ISOLATION
928 [MIGRATE_ISOLATE
] = { MIGRATE_RESERVE
}, /* Never used */
933 * Move the free pages in a range to the free lists of the requested type.
934 * Note that start_page and end_pages are not aligned on a pageblock
935 * boundary. If alignment is required, use move_freepages_block()
937 int move_freepages(struct zone
*zone
,
938 struct page
*start_page
, struct page
*end_page
,
945 #ifndef CONFIG_HOLES_IN_ZONE
947 * page_zone is not safe to call in this context when
948 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
949 * anyway as we check zone boundaries in move_freepages_block().
950 * Remove at a later date when no bug reports exist related to
951 * grouping pages by mobility
953 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
956 for (page
= start_page
; page
<= end_page
;) {
957 /* Make sure we are not inadvertently changing nodes */
958 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
960 if (!pfn_valid_within(page_to_pfn(page
))) {
965 if (!PageBuddy(page
)) {
970 order
= page_order(page
);
971 list_move(&page
->lru
,
972 &zone
->free_area
[order
].free_list
[migratetype
]);
973 set_freepage_migratetype(page
, migratetype
);
975 pages_moved
+= 1 << order
;
981 int move_freepages_block(struct zone
*zone
, struct page
*page
,
984 unsigned long start_pfn
, end_pfn
;
985 struct page
*start_page
, *end_page
;
987 start_pfn
= page_to_pfn(page
);
988 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
989 start_page
= pfn_to_page(start_pfn
);
990 end_page
= start_page
+ pageblock_nr_pages
- 1;
991 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
993 /* Do not cross zone boundaries */
994 if (!zone_spans_pfn(zone
, start_pfn
))
996 if (!zone_spans_pfn(zone
, end_pfn
))
999 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1002 static void change_pageblock_range(struct page
*pageblock_page
,
1003 int start_order
, int migratetype
)
1005 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1007 while (nr_pageblocks
--) {
1008 set_pageblock_migratetype(pageblock_page
, migratetype
);
1009 pageblock_page
+= pageblock_nr_pages
;
1014 * If breaking a large block of pages, move all free pages to the preferred
1015 * allocation list. If falling back for a reclaimable kernel allocation, be
1016 * more aggressive about taking ownership of free pages.
1018 * On the other hand, never change migration type of MIGRATE_CMA pageblocks
1019 * nor move CMA pages to different free lists. We don't want unmovable pages
1020 * to be allocated from MIGRATE_CMA areas.
1022 * Returns the new migratetype of the pageblock (or the same old migratetype
1023 * if it was unchanged).
1025 static int try_to_steal_freepages(struct zone
*zone
, struct page
*page
,
1026 int start_type
, int fallback_type
)
1028 int current_order
= page_order(page
);
1031 * When borrowing from MIGRATE_CMA, we need to release the excess
1032 * buddy pages to CMA itself.
1034 if (is_migrate_cma(fallback_type
))
1035 return fallback_type
;
1037 /* Take ownership for orders >= pageblock_order */
1038 if (current_order
>= pageblock_order
) {
1039 change_pageblock_range(page
, current_order
, start_type
);
1043 if (current_order
>= pageblock_order
/ 2 ||
1044 start_type
== MIGRATE_RECLAIMABLE
||
1045 page_group_by_mobility_disabled
) {
1048 pages
= move_freepages_block(zone
, page
, start_type
);
1050 /* Claim the whole block if over half of it is free */
1051 if (pages
>= (1 << (pageblock_order
-1)) ||
1052 page_group_by_mobility_disabled
) {
1054 set_pageblock_migratetype(page
, start_type
);
1060 return fallback_type
;
1063 /* Remove an element from the buddy allocator from the fallback list */
1064 static inline struct page
*
1065 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
1067 struct free_area
*area
;
1070 int migratetype
, new_type
, i
;
1072 /* Find the largest possible block of pages in the other list */
1073 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
1076 migratetype
= fallbacks
[start_migratetype
][i
];
1078 /* MIGRATE_RESERVE handled later if necessary */
1079 if (migratetype
== MIGRATE_RESERVE
)
1082 area
= &(zone
->free_area
[current_order
]);
1083 if (list_empty(&area
->free_list
[migratetype
]))
1086 page
= list_entry(area
->free_list
[migratetype
].next
,
1090 new_type
= try_to_steal_freepages(zone
, page
,
1094 /* Remove the page from the freelists */
1095 list_del(&page
->lru
);
1096 rmv_page_order(page
);
1098 expand(zone
, page
, order
, current_order
, area
,
1101 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1102 start_migratetype
, migratetype
, new_type
);
1112 * Do the hard work of removing an element from the buddy allocator.
1113 * Call me with the zone->lock already held.
1115 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1121 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1123 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1124 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1127 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1128 * is used because __rmqueue_smallest is an inline function
1129 * and we want just one call site
1132 migratetype
= MIGRATE_RESERVE
;
1137 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1142 * Obtain a specified number of elements from the buddy allocator, all under
1143 * a single hold of the lock, for efficiency. Add them to the supplied list.
1144 * Returns the number of new pages which were placed at *list.
1146 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1147 unsigned long count
, struct list_head
*list
,
1148 int migratetype
, int cold
)
1150 int mt
= migratetype
, i
;
1152 spin_lock(&zone
->lock
);
1153 for (i
= 0; i
< count
; ++i
) {
1154 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1155 if (unlikely(page
== NULL
))
1159 * Split buddy pages returned by expand() are received here
1160 * in physical page order. The page is added to the callers and
1161 * list and the list head then moves forward. From the callers
1162 * perspective, the linked list is ordered by page number in
1163 * some conditions. This is useful for IO devices that can
1164 * merge IO requests if the physical pages are ordered
1167 if (likely(cold
== 0))
1168 list_add(&page
->lru
, list
);
1170 list_add_tail(&page
->lru
, list
);
1171 if (IS_ENABLED(CONFIG_CMA
)) {
1172 mt
= get_pageblock_migratetype(page
);
1173 if (!is_migrate_cma(mt
) && !is_migrate_isolate(mt
))
1176 set_freepage_migratetype(page
, mt
);
1178 if (is_migrate_cma(mt
))
1179 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1182 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1183 spin_unlock(&zone
->lock
);
1189 * Called from the vmstat counter updater to drain pagesets of this
1190 * currently executing processor on remote nodes after they have
1193 * Note that this function must be called with the thread pinned to
1194 * a single processor.
1196 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1198 unsigned long flags
;
1200 unsigned long batch
;
1202 local_irq_save(flags
);
1203 batch
= ACCESS_ONCE(pcp
->batch
);
1204 if (pcp
->count
>= batch
)
1207 to_drain
= pcp
->count
;
1209 free_pcppages_bulk(zone
, to_drain
, pcp
);
1210 pcp
->count
-= to_drain
;
1212 local_irq_restore(flags
);
1217 * Drain pages of the indicated processor.
1219 * The processor must either be the current processor and the
1220 * thread pinned to the current processor or a processor that
1223 static void drain_pages(unsigned int cpu
)
1225 unsigned long flags
;
1228 for_each_populated_zone(zone
) {
1229 struct per_cpu_pageset
*pset
;
1230 struct per_cpu_pages
*pcp
;
1232 local_irq_save(flags
);
1233 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1237 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1240 local_irq_restore(flags
);
1245 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1247 void drain_local_pages(void *arg
)
1249 drain_pages(smp_processor_id());
1253 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1255 * Note that this code is protected against sending an IPI to an offline
1256 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1257 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1258 * nothing keeps CPUs from showing up after we populated the cpumask and
1259 * before the call to on_each_cpu_mask().
1261 void drain_all_pages(void)
1264 struct per_cpu_pageset
*pcp
;
1268 * Allocate in the BSS so we wont require allocation in
1269 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1271 static cpumask_t cpus_with_pcps
;
1274 * We don't care about racing with CPU hotplug event
1275 * as offline notification will cause the notified
1276 * cpu to drain that CPU pcps and on_each_cpu_mask
1277 * disables preemption as part of its processing
1279 for_each_online_cpu(cpu
) {
1280 bool has_pcps
= false;
1281 for_each_populated_zone(zone
) {
1282 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1283 if (pcp
->pcp
.count
) {
1289 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1291 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1293 on_each_cpu_mask(&cpus_with_pcps
, drain_local_pages
, NULL
, 1);
1296 #ifdef CONFIG_HIBERNATION
1298 void mark_free_pages(struct zone
*zone
)
1300 unsigned long pfn
, max_zone_pfn
;
1301 unsigned long flags
;
1303 struct list_head
*curr
;
1305 if (zone_is_empty(zone
))
1308 spin_lock_irqsave(&zone
->lock
, flags
);
1310 max_zone_pfn
= zone_end_pfn(zone
);
1311 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1312 if (pfn_valid(pfn
)) {
1313 struct page
*page
= pfn_to_page(pfn
);
1315 if (!swsusp_page_is_forbidden(page
))
1316 swsusp_unset_page_free(page
);
1319 for_each_migratetype_order(order
, t
) {
1320 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1323 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1324 for (i
= 0; i
< (1UL << order
); i
++)
1325 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1328 spin_unlock_irqrestore(&zone
->lock
, flags
);
1330 #endif /* CONFIG_PM */
1333 * Free a 0-order page
1334 * cold == 1 ? free a cold page : free a hot page
1336 void free_hot_cold_page(struct page
*page
, int cold
)
1338 struct zone
*zone
= page_zone(page
);
1339 struct per_cpu_pages
*pcp
;
1340 unsigned long flags
;
1343 if (!free_pages_prepare(page
, 0))
1346 migratetype
= get_pageblock_migratetype(page
);
1347 set_freepage_migratetype(page
, migratetype
);
1348 local_irq_save(flags
);
1349 __count_vm_event(PGFREE
);
1352 * We only track unmovable, reclaimable and movable on pcp lists.
1353 * Free ISOLATE pages back to the allocator because they are being
1354 * offlined but treat RESERVE as movable pages so we can get those
1355 * areas back if necessary. Otherwise, we may have to free
1356 * excessively into the page allocator
1358 if (migratetype
>= MIGRATE_PCPTYPES
) {
1359 if (unlikely(is_migrate_isolate(migratetype
))) {
1360 free_one_page(zone
, page
, 0, migratetype
);
1363 migratetype
= MIGRATE_MOVABLE
;
1366 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1368 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1370 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1372 if (pcp
->count
>= pcp
->high
) {
1373 unsigned long batch
= ACCESS_ONCE(pcp
->batch
);
1374 free_pcppages_bulk(zone
, batch
, pcp
);
1375 pcp
->count
-= batch
;
1379 local_irq_restore(flags
);
1383 * Free a list of 0-order pages
1385 void free_hot_cold_page_list(struct list_head
*list
, int cold
)
1387 struct page
*page
, *next
;
1389 list_for_each_entry_safe(page
, next
, list
, lru
) {
1390 trace_mm_page_free_batched(page
, cold
);
1391 free_hot_cold_page(page
, cold
);
1396 * split_page takes a non-compound higher-order page, and splits it into
1397 * n (1<<order) sub-pages: page[0..n]
1398 * Each sub-page must be freed individually.
1400 * Note: this is probably too low level an operation for use in drivers.
1401 * Please consult with lkml before using this in your driver.
1403 void split_page(struct page
*page
, unsigned int order
)
1407 VM_BUG_ON(PageCompound(page
));
1408 VM_BUG_ON(!page_count(page
));
1410 #ifdef CONFIG_KMEMCHECK
1412 * Split shadow pages too, because free(page[0]) would
1413 * otherwise free the whole shadow.
1415 if (kmemcheck_page_is_tracked(page
))
1416 split_page(virt_to_page(page
[0].shadow
), order
);
1419 for (i
= 1; i
< (1 << order
); i
++)
1420 set_page_refcounted(page
+ i
);
1422 EXPORT_SYMBOL_GPL(split_page
);
1424 static int __isolate_free_page(struct page
*page
, unsigned int order
)
1426 unsigned long watermark
;
1430 BUG_ON(!PageBuddy(page
));
1432 zone
= page_zone(page
);
1433 mt
= get_pageblock_migratetype(page
);
1435 if (!is_migrate_isolate(mt
)) {
1436 /* Obey watermarks as if the page was being allocated */
1437 watermark
= low_wmark_pages(zone
) + (1 << order
);
1438 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1441 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
1444 /* Remove page from free list */
1445 list_del(&page
->lru
);
1446 zone
->free_area
[order
].nr_free
--;
1447 rmv_page_order(page
);
1449 /* Set the pageblock if the isolated page is at least a pageblock */
1450 if (order
>= pageblock_order
- 1) {
1451 struct page
*endpage
= page
+ (1 << order
) - 1;
1452 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
1453 int mt
= get_pageblock_migratetype(page
);
1454 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
1455 set_pageblock_migratetype(page
,
1460 return 1UL << order
;
1464 * Similar to split_page except the page is already free. As this is only
1465 * being used for migration, the migratetype of the block also changes.
1466 * As this is called with interrupts disabled, the caller is responsible
1467 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1470 * Note: this is probably too low level an operation for use in drivers.
1471 * Please consult with lkml before using this in your driver.
1473 int split_free_page(struct page
*page
)
1478 order
= page_order(page
);
1480 nr_pages
= __isolate_free_page(page
, order
);
1484 /* Split into individual pages */
1485 set_page_refcounted(page
);
1486 split_page(page
, order
);
1491 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1492 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1496 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1497 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1500 unsigned long flags
;
1502 int cold
= !!(gfp_flags
& __GFP_COLD
);
1505 if (likely(order
== 0)) {
1506 struct per_cpu_pages
*pcp
;
1507 struct list_head
*list
;
1509 local_irq_save(flags
);
1510 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1511 list
= &pcp
->lists
[migratetype
];
1512 if (list_empty(list
)) {
1513 pcp
->count
+= rmqueue_bulk(zone
, 0,
1516 if (unlikely(list_empty(list
)))
1521 page
= list_entry(list
->prev
, struct page
, lru
);
1523 page
= list_entry(list
->next
, struct page
, lru
);
1525 list_del(&page
->lru
);
1528 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1530 * __GFP_NOFAIL is not to be used in new code.
1532 * All __GFP_NOFAIL callers should be fixed so that they
1533 * properly detect and handle allocation failures.
1535 * We most definitely don't want callers attempting to
1536 * allocate greater than order-1 page units with
1539 WARN_ON_ONCE(order
> 1);
1541 spin_lock_irqsave(&zone
->lock
, flags
);
1542 page
= __rmqueue(zone
, order
, migratetype
);
1543 spin_unlock(&zone
->lock
);
1546 __mod_zone_freepage_state(zone
, -(1 << order
),
1547 get_pageblock_migratetype(page
));
1550 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
1551 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1552 zone_statistics(preferred_zone
, zone
, gfp_flags
);
1553 local_irq_restore(flags
);
1555 VM_BUG_ON(bad_range(zone
, page
));
1556 if (prep_new_page(page
, order
, gfp_flags
))
1561 local_irq_restore(flags
);
1565 #ifdef CONFIG_FAIL_PAGE_ALLOC
1568 struct fault_attr attr
;
1570 u32 ignore_gfp_highmem
;
1571 u32 ignore_gfp_wait
;
1573 } fail_page_alloc
= {
1574 .attr
= FAULT_ATTR_INITIALIZER
,
1575 .ignore_gfp_wait
= 1,
1576 .ignore_gfp_highmem
= 1,
1580 static int __init
setup_fail_page_alloc(char *str
)
1582 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1584 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1586 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1588 if (order
< fail_page_alloc
.min_order
)
1590 if (gfp_mask
& __GFP_NOFAIL
)
1592 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1594 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1597 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1600 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1602 static int __init
fail_page_alloc_debugfs(void)
1604 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1607 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
1608 &fail_page_alloc
.attr
);
1610 return PTR_ERR(dir
);
1612 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1613 &fail_page_alloc
.ignore_gfp_wait
))
1615 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1616 &fail_page_alloc
.ignore_gfp_highmem
))
1618 if (!debugfs_create_u32("min-order", mode
, dir
,
1619 &fail_page_alloc
.min_order
))
1624 debugfs_remove_recursive(dir
);
1629 late_initcall(fail_page_alloc_debugfs
);
1631 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1633 #else /* CONFIG_FAIL_PAGE_ALLOC */
1635 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1640 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1643 * Return true if free pages are above 'mark'. This takes into account the order
1644 * of the allocation.
1646 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1647 int classzone_idx
, int alloc_flags
, long free_pages
)
1649 /* free_pages my go negative - that's OK */
1651 long lowmem_reserve
= z
->lowmem_reserve
[classzone_idx
];
1655 free_pages
-= (1 << order
) - 1;
1656 if (alloc_flags
& ALLOC_HIGH
)
1658 if (alloc_flags
& ALLOC_HARDER
)
1661 /* If allocation can't use CMA areas don't use free CMA pages */
1662 if (!(alloc_flags
& ALLOC_CMA
))
1663 free_cma
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
1666 if (free_pages
- free_cma
<= min
+ lowmem_reserve
)
1668 for (o
= 0; o
< order
; o
++) {
1669 /* At the next order, this order's pages become unavailable */
1670 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1672 /* Require fewer higher order pages to be free */
1675 if (free_pages
<= min
)
1681 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1682 int classzone_idx
, int alloc_flags
)
1684 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1685 zone_page_state(z
, NR_FREE_PAGES
));
1688 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1689 int classzone_idx
, int alloc_flags
)
1691 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1693 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1694 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1696 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1702 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1703 * skip over zones that are not allowed by the cpuset, or that have
1704 * been recently (in last second) found to be nearly full. See further
1705 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1706 * that have to skip over a lot of full or unallowed zones.
1708 * If the zonelist cache is present in the passed zonelist, then
1709 * returns a pointer to the allowed node mask (either the current
1710 * tasks mems_allowed, or node_states[N_MEMORY].)
1712 * If the zonelist cache is not available for this zonelist, does
1713 * nothing and returns NULL.
1715 * If the fullzones BITMAP in the zonelist cache is stale (more than
1716 * a second since last zap'd) then we zap it out (clear its bits.)
1718 * We hold off even calling zlc_setup, until after we've checked the
1719 * first zone in the zonelist, on the theory that most allocations will
1720 * be satisfied from that first zone, so best to examine that zone as
1721 * quickly as we can.
1723 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1725 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1726 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1728 zlc
= zonelist
->zlcache_ptr
;
1732 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1733 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1734 zlc
->last_full_zap
= jiffies
;
1737 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1738 &cpuset_current_mems_allowed
:
1739 &node_states
[N_MEMORY
];
1740 return allowednodes
;
1744 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1745 * if it is worth looking at further for free memory:
1746 * 1) Check that the zone isn't thought to be full (doesn't have its
1747 * bit set in the zonelist_cache fullzones BITMAP).
1748 * 2) Check that the zones node (obtained from the zonelist_cache
1749 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1750 * Return true (non-zero) if zone is worth looking at further, or
1751 * else return false (zero) if it is not.
1753 * This check -ignores- the distinction between various watermarks,
1754 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1755 * found to be full for any variation of these watermarks, it will
1756 * be considered full for up to one second by all requests, unless
1757 * we are so low on memory on all allowed nodes that we are forced
1758 * into the second scan of the zonelist.
1760 * In the second scan we ignore this zonelist cache and exactly
1761 * apply the watermarks to all zones, even it is slower to do so.
1762 * We are low on memory in the second scan, and should leave no stone
1763 * unturned looking for a free page.
1765 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1766 nodemask_t
*allowednodes
)
1768 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1769 int i
; /* index of *z in zonelist zones */
1770 int n
; /* node that zone *z is on */
1772 zlc
= zonelist
->zlcache_ptr
;
1776 i
= z
- zonelist
->_zonerefs
;
1779 /* This zone is worth trying if it is allowed but not full */
1780 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1784 * Given 'z' scanning a zonelist, set the corresponding bit in
1785 * zlc->fullzones, so that subsequent attempts to allocate a page
1786 * from that zone don't waste time re-examining it.
1788 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1790 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1791 int i
; /* index of *z in zonelist zones */
1793 zlc
= zonelist
->zlcache_ptr
;
1797 i
= z
- zonelist
->_zonerefs
;
1799 set_bit(i
, zlc
->fullzones
);
1803 * clear all zones full, called after direct reclaim makes progress so that
1804 * a zone that was recently full is not skipped over for up to a second
1806 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1808 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1810 zlc
= zonelist
->zlcache_ptr
;
1814 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1817 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
1819 return local_zone
->node
== zone
->node
;
1822 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1824 return node_isset(local_zone
->node
, zone
->zone_pgdat
->reclaim_nodes
);
1827 static void __paginginit
init_zone_allows_reclaim(int nid
)
1831 for_each_online_node(i
)
1832 if (node_distance(nid
, i
) <= RECLAIM_DISTANCE
)
1833 node_set(i
, NODE_DATA(nid
)->reclaim_nodes
);
1835 zone_reclaim_mode
= 1;
1838 #else /* CONFIG_NUMA */
1840 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1845 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1846 nodemask_t
*allowednodes
)
1851 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1855 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1859 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
1864 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1869 static inline void init_zone_allows_reclaim(int nid
)
1872 #endif /* CONFIG_NUMA */
1875 * get_page_from_freelist goes through the zonelist trying to allocate
1878 static struct page
*
1879 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1880 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1881 struct zone
*preferred_zone
, int migratetype
)
1884 struct page
*page
= NULL
;
1887 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1888 int zlc_active
= 0; /* set if using zonelist_cache */
1889 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1891 classzone_idx
= zone_idx(preferred_zone
);
1894 * Scan zonelist, looking for a zone with enough free.
1895 * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c.
1897 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1898 high_zoneidx
, nodemask
) {
1901 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1902 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1904 if ((alloc_flags
& ALLOC_CPUSET
) &&
1905 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1907 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1908 if (unlikely(alloc_flags
& ALLOC_NO_WATERMARKS
))
1911 * Distribute pages in proportion to the individual
1912 * zone size to ensure fair page aging. The zone a
1913 * page was allocated in should have no effect on the
1914 * time the page has in memory before being reclaimed.
1916 * Try to stay in local zones in the fastpath. If
1917 * that fails, the slowpath is entered, which will do
1918 * another pass starting with the local zones, but
1919 * ultimately fall back to remote zones that do not
1920 * partake in the fairness round-robin cycle of this
1923 if (alloc_flags
& ALLOC_WMARK_LOW
) {
1924 if (zone_page_state(zone
, NR_ALLOC_BATCH
) <= 0)
1926 if (!zone_local(preferred_zone
, zone
))
1930 * When allocating a page cache page for writing, we
1931 * want to get it from a zone that is within its dirty
1932 * limit, such that no single zone holds more than its
1933 * proportional share of globally allowed dirty pages.
1934 * The dirty limits take into account the zone's
1935 * lowmem reserves and high watermark so that kswapd
1936 * should be able to balance it without having to
1937 * write pages from its LRU list.
1939 * This may look like it could increase pressure on
1940 * lower zones by failing allocations in higher zones
1941 * before they are full. But the pages that do spill
1942 * over are limited as the lower zones are protected
1943 * by this very same mechanism. It should not become
1944 * a practical burden to them.
1946 * XXX: For now, allow allocations to potentially
1947 * exceed the per-zone dirty limit in the slowpath
1948 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1949 * which is important when on a NUMA setup the allowed
1950 * zones are together not big enough to reach the
1951 * global limit. The proper fix for these situations
1952 * will require awareness of zones in the
1953 * dirty-throttling and the flusher threads.
1955 if ((alloc_flags
& ALLOC_WMARK_LOW
) &&
1956 (gfp_mask
& __GFP_WRITE
) && !zone_dirty_ok(zone
))
1957 goto this_zone_full
;
1959 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1960 if (!zone_watermark_ok(zone
, order
, mark
,
1961 classzone_idx
, alloc_flags
)) {
1964 if (IS_ENABLED(CONFIG_NUMA
) &&
1965 !did_zlc_setup
&& nr_online_nodes
> 1) {
1967 * we do zlc_setup if there are multiple nodes
1968 * and before considering the first zone allowed
1971 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1976 if (zone_reclaim_mode
== 0 ||
1977 !zone_allows_reclaim(preferred_zone
, zone
))
1978 goto this_zone_full
;
1981 * As we may have just activated ZLC, check if the first
1982 * eligible zone has failed zone_reclaim recently.
1984 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1985 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1988 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1990 case ZONE_RECLAIM_NOSCAN
:
1993 case ZONE_RECLAIM_FULL
:
1994 /* scanned but unreclaimable */
1997 /* did we reclaim enough */
1998 if (zone_watermark_ok(zone
, order
, mark
,
1999 classzone_idx
, alloc_flags
))
2003 * Failed to reclaim enough to meet watermark.
2004 * Only mark the zone full if checking the min
2005 * watermark or if we failed to reclaim just
2006 * 1<<order pages or else the page allocator
2007 * fastpath will prematurely mark zones full
2008 * when the watermark is between the low and
2011 if (((alloc_flags
& ALLOC_WMARK_MASK
) == ALLOC_WMARK_MIN
) ||
2012 ret
== ZONE_RECLAIM_SOME
)
2013 goto this_zone_full
;
2020 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
2021 gfp_mask
, migratetype
);
2025 if (IS_ENABLED(CONFIG_NUMA
))
2026 zlc_mark_zone_full(zonelist
, z
);
2029 if (unlikely(IS_ENABLED(CONFIG_NUMA
) && page
== NULL
&& zlc_active
)) {
2030 /* Disable zlc cache for second zonelist scan */
2037 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2038 * necessary to allocate the page. The expectation is
2039 * that the caller is taking steps that will free more
2040 * memory. The caller should avoid the page being used
2041 * for !PFMEMALLOC purposes.
2043 page
->pfmemalloc
= !!(alloc_flags
& ALLOC_NO_WATERMARKS
);
2049 * Large machines with many possible nodes should not always dump per-node
2050 * meminfo in irq context.
2052 static inline bool should_suppress_show_mem(void)
2057 ret
= in_interrupt();
2062 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2063 DEFAULT_RATELIMIT_INTERVAL
,
2064 DEFAULT_RATELIMIT_BURST
);
2066 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
2068 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2070 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2071 debug_guardpage_minorder() > 0)
2075 * This documents exceptions given to allocations in certain
2076 * contexts that are allowed to allocate outside current's set
2079 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2080 if (test_thread_flag(TIF_MEMDIE
) ||
2081 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2082 filter
&= ~SHOW_MEM_FILTER_NODES
;
2083 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
2084 filter
&= ~SHOW_MEM_FILTER_NODES
;
2087 struct va_format vaf
;
2090 va_start(args
, fmt
);
2095 pr_warn("%pV", &vaf
);
2100 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2101 current
->comm
, order
, gfp_mask
);
2104 if (!should_suppress_show_mem())
2109 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
2110 unsigned long did_some_progress
,
2111 unsigned long pages_reclaimed
)
2113 /* Do not loop if specifically requested */
2114 if (gfp_mask
& __GFP_NORETRY
)
2117 /* Always retry if specifically requested */
2118 if (gfp_mask
& __GFP_NOFAIL
)
2122 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2123 * making forward progress without invoking OOM. Suspend also disables
2124 * storage devices so kswapd will not help. Bail if we are suspending.
2126 if (!did_some_progress
&& pm_suspended_storage())
2130 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2131 * means __GFP_NOFAIL, but that may not be true in other
2134 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
2138 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2139 * specified, then we retry until we no longer reclaim any pages
2140 * (above), or we've reclaimed an order of pages at least as
2141 * large as the allocation's order. In both cases, if the
2142 * allocation still fails, we stop retrying.
2144 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
2150 static inline struct page
*
2151 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2152 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2153 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2158 /* Acquire the OOM killer lock for the zones in zonelist */
2159 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
2160 schedule_timeout_uninterruptible(1);
2165 * Go through the zonelist yet one more time, keep very high watermark
2166 * here, this is only to catch a parallel oom killing, we must fail if
2167 * we're still under heavy pressure.
2169 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
2170 order
, zonelist
, high_zoneidx
,
2171 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
2172 preferred_zone
, migratetype
);
2176 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2177 /* The OOM killer will not help higher order allocs */
2178 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2180 /* The OOM killer does not needlessly kill tasks for lowmem */
2181 if (high_zoneidx
< ZONE_NORMAL
)
2184 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2185 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2186 * The caller should handle page allocation failure by itself if
2187 * it specifies __GFP_THISNODE.
2188 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2190 if (gfp_mask
& __GFP_THISNODE
)
2193 /* Exhausted what can be done so it's blamo time */
2194 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
, false);
2197 clear_zonelist_oom(zonelist
, gfp_mask
);
2201 #ifdef CONFIG_COMPACTION
2202 /* Try memory compaction for high-order allocations before reclaim */
2203 static struct page
*
2204 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2205 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2206 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2207 int migratetype
, bool sync_migration
,
2208 bool *contended_compaction
, bool *deferred_compaction
,
2209 unsigned long *did_some_progress
)
2214 if (compaction_deferred(preferred_zone
, order
)) {
2215 *deferred_compaction
= true;
2219 current
->flags
|= PF_MEMALLOC
;
2220 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
2221 nodemask
, sync_migration
,
2222 contended_compaction
);
2223 current
->flags
&= ~PF_MEMALLOC
;
2225 if (*did_some_progress
!= COMPACT_SKIPPED
) {
2228 /* Page migration frees to the PCP lists but we want merging */
2229 drain_pages(get_cpu());
2232 page
= get_page_from_freelist(gfp_mask
, nodemask
,
2233 order
, zonelist
, high_zoneidx
,
2234 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2235 preferred_zone
, migratetype
);
2237 preferred_zone
->compact_blockskip_flush
= false;
2238 preferred_zone
->compact_considered
= 0;
2239 preferred_zone
->compact_defer_shift
= 0;
2240 if (order
>= preferred_zone
->compact_order_failed
)
2241 preferred_zone
->compact_order_failed
= order
+ 1;
2242 count_vm_event(COMPACTSUCCESS
);
2247 * It's bad if compaction run occurs and fails.
2248 * The most likely reason is that pages exist,
2249 * but not enough to satisfy watermarks.
2251 count_vm_event(COMPACTFAIL
);
2254 * As async compaction considers a subset of pageblocks, only
2255 * defer if the failure was a sync compaction failure.
2258 defer_compaction(preferred_zone
, order
);
2266 static inline struct page
*
2267 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2268 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2269 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2270 int migratetype
, bool sync_migration
,
2271 bool *contended_compaction
, bool *deferred_compaction
,
2272 unsigned long *did_some_progress
)
2276 #endif /* CONFIG_COMPACTION */
2278 /* Perform direct synchronous page reclaim */
2280 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
, struct zonelist
*zonelist
,
2281 nodemask_t
*nodemask
)
2283 struct reclaim_state reclaim_state
;
2288 /* We now go into synchronous reclaim */
2289 cpuset_memory_pressure_bump();
2290 current
->flags
|= PF_MEMALLOC
;
2291 lockdep_set_current_reclaim_state(gfp_mask
);
2292 reclaim_state
.reclaimed_slab
= 0;
2293 current
->reclaim_state
= &reclaim_state
;
2295 progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
2297 current
->reclaim_state
= NULL
;
2298 lockdep_clear_current_reclaim_state();
2299 current
->flags
&= ~PF_MEMALLOC
;
2306 /* The really slow allocator path where we enter direct reclaim */
2307 static inline struct page
*
2308 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2309 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2310 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2311 int migratetype
, unsigned long *did_some_progress
)
2313 struct page
*page
= NULL
;
2314 bool drained
= false;
2316 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
2318 if (unlikely(!(*did_some_progress
)))
2321 /* After successful reclaim, reconsider all zones for allocation */
2322 if (IS_ENABLED(CONFIG_NUMA
))
2323 zlc_clear_zones_full(zonelist
);
2326 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2327 zonelist
, high_zoneidx
,
2328 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2329 preferred_zone
, migratetype
);
2332 * If an allocation failed after direct reclaim, it could be because
2333 * pages are pinned on the per-cpu lists. Drain them and try again
2335 if (!page
&& !drained
) {
2345 * This is called in the allocator slow-path if the allocation request is of
2346 * sufficient urgency to ignore watermarks and take other desperate measures
2348 static inline struct page
*
2349 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2350 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2351 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2357 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2358 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
2359 preferred_zone
, migratetype
);
2361 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2362 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2363 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2368 static void prepare_slowpath(gfp_t gfp_mask
, unsigned int order
,
2369 struct zonelist
*zonelist
,
2370 enum zone_type high_zoneidx
,
2371 struct zone
*preferred_zone
)
2376 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
2377 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2378 wakeup_kswapd(zone
, order
, zone_idx(preferred_zone
));
2380 * Only reset the batches of zones that were actually
2381 * considered in the fast path, we don't want to
2382 * thrash fairness information for zones that are not
2383 * actually part of this zonelist's round-robin cycle.
2385 if (!zone_local(preferred_zone
, zone
))
2387 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2388 high_wmark_pages(zone
) -
2389 low_wmark_pages(zone
) -
2390 zone_page_state(zone
, NR_ALLOC_BATCH
));
2395 gfp_to_alloc_flags(gfp_t gfp_mask
)
2397 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2398 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2400 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2401 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2404 * The caller may dip into page reserves a bit more if the caller
2405 * cannot run direct reclaim, or if the caller has realtime scheduling
2406 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2407 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2409 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2413 * Not worth trying to allocate harder for
2414 * __GFP_NOMEMALLOC even if it can't schedule.
2416 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2417 alloc_flags
|= ALLOC_HARDER
;
2419 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2420 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2422 alloc_flags
&= ~ALLOC_CPUSET
;
2423 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2424 alloc_flags
|= ALLOC_HARDER
;
2426 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2427 if (gfp_mask
& __GFP_MEMALLOC
)
2428 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2429 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2430 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2431 else if (!in_interrupt() &&
2432 ((current
->flags
& PF_MEMALLOC
) ||
2433 unlikely(test_thread_flag(TIF_MEMDIE
))))
2434 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2437 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2438 alloc_flags
|= ALLOC_CMA
;
2443 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2445 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2448 static inline struct page
*
2449 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2450 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2451 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2454 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2455 struct page
*page
= NULL
;
2457 unsigned long pages_reclaimed
= 0;
2458 unsigned long did_some_progress
;
2459 bool sync_migration
= false;
2460 bool deferred_compaction
= false;
2461 bool contended_compaction
= false;
2464 * In the slowpath, we sanity check order to avoid ever trying to
2465 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2466 * be using allocators in order of preference for an area that is
2469 if (order
>= MAX_ORDER
) {
2470 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2475 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2476 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2477 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2478 * using a larger set of nodes after it has established that the
2479 * allowed per node queues are empty and that nodes are
2482 if (IS_ENABLED(CONFIG_NUMA
) &&
2483 (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2487 prepare_slowpath(gfp_mask
, order
, zonelist
,
2488 high_zoneidx
, preferred_zone
);
2491 * OK, we're below the kswapd watermark and have kicked background
2492 * reclaim. Now things get more complex, so set up alloc_flags according
2493 * to how we want to proceed.
2495 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2498 * Find the true preferred zone if the allocation is unconstrained by
2501 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2502 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2506 /* This is the last chance, in general, before the goto nopage. */
2507 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2508 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2509 preferred_zone
, migratetype
);
2513 /* Allocate without watermarks if the context allows */
2514 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2516 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2517 * the allocation is high priority and these type of
2518 * allocations are system rather than user orientated
2520 zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
2522 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2523 zonelist
, high_zoneidx
, nodemask
,
2524 preferred_zone
, migratetype
);
2530 /* Atomic allocations - we can't balance anything */
2534 /* Avoid recursion of direct reclaim */
2535 if (current
->flags
& PF_MEMALLOC
)
2538 /* Avoid allocations with no watermarks from looping endlessly */
2539 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2543 * Try direct compaction. The first pass is asynchronous. Subsequent
2544 * attempts after direct reclaim are synchronous
2546 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2547 zonelist
, high_zoneidx
,
2549 alloc_flags
, preferred_zone
,
2550 migratetype
, sync_migration
,
2551 &contended_compaction
,
2552 &deferred_compaction
,
2553 &did_some_progress
);
2556 sync_migration
= true;
2559 * If compaction is deferred for high-order allocations, it is because
2560 * sync compaction recently failed. In this is the case and the caller
2561 * requested a movable allocation that does not heavily disrupt the
2562 * system then fail the allocation instead of entering direct reclaim.
2564 if ((deferred_compaction
|| contended_compaction
) &&
2565 (gfp_mask
& __GFP_NO_KSWAPD
))
2568 /* Try direct reclaim and then allocating */
2569 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2570 zonelist
, high_zoneidx
,
2572 alloc_flags
, preferred_zone
,
2573 migratetype
, &did_some_progress
);
2578 * If we failed to make any progress reclaiming, then we are
2579 * running out of options and have to consider going OOM
2581 if (!did_some_progress
) {
2582 if (oom_gfp_allowed(gfp_mask
)) {
2583 if (oom_killer_disabled
)
2585 /* Coredumps can quickly deplete all memory reserves */
2586 if ((current
->flags
& PF_DUMPCORE
) &&
2587 !(gfp_mask
& __GFP_NOFAIL
))
2589 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2590 zonelist
, high_zoneidx
,
2591 nodemask
, preferred_zone
,
2596 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2598 * The oom killer is not called for high-order
2599 * allocations that may fail, so if no progress
2600 * is being made, there are no other options and
2601 * retrying is unlikely to help.
2603 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2606 * The oom killer is not called for lowmem
2607 * allocations to prevent needlessly killing
2610 if (high_zoneidx
< ZONE_NORMAL
)
2618 /* Check if we should retry the allocation */
2619 pages_reclaimed
+= did_some_progress
;
2620 if (should_alloc_retry(gfp_mask
, order
, did_some_progress
,
2622 /* Wait for some write requests to complete then retry */
2623 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2627 * High-order allocations do not necessarily loop after
2628 * direct reclaim and reclaim/compaction depends on compaction
2629 * being called after reclaim so call directly if necessary
2631 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2632 zonelist
, high_zoneidx
,
2634 alloc_flags
, preferred_zone
,
2635 migratetype
, sync_migration
,
2636 &contended_compaction
,
2637 &deferred_compaction
,
2638 &did_some_progress
);
2644 warn_alloc_failed(gfp_mask
, order
, NULL
);
2647 if (kmemcheck_enabled
)
2648 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2654 * This is the 'heart' of the zoned buddy allocator.
2657 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2658 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2660 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2661 struct zone
*preferred_zone
;
2662 struct page
*page
= NULL
;
2663 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2664 unsigned int cpuset_mems_cookie
;
2665 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
;
2666 struct mem_cgroup
*memcg
= NULL
;
2668 gfp_mask
&= gfp_allowed_mask
;
2670 lockdep_trace_alloc(gfp_mask
);
2672 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2674 if (should_fail_alloc_page(gfp_mask
, order
))
2678 * Check the zones suitable for the gfp_mask contain at least one
2679 * valid zone. It's possible to have an empty zonelist as a result
2680 * of GFP_THISNODE and a memoryless node
2682 if (unlikely(!zonelist
->_zonerefs
->zone
))
2686 * Will only have any effect when __GFP_KMEMCG is set. This is
2687 * verified in the (always inline) callee
2689 if (!memcg_kmem_newpage_charge(gfp_mask
, &memcg
, order
))
2693 cpuset_mems_cookie
= get_mems_allowed();
2695 /* The preferred zone is used for statistics later */
2696 first_zones_zonelist(zonelist
, high_zoneidx
,
2697 nodemask
? : &cpuset_current_mems_allowed
,
2699 if (!preferred_zone
)
2703 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2704 alloc_flags
|= ALLOC_CMA
;
2706 /* First allocation attempt */
2707 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2708 zonelist
, high_zoneidx
, alloc_flags
,
2709 preferred_zone
, migratetype
);
2710 if (unlikely(!page
)) {
2712 * Runtime PM, block IO and its error handling path
2713 * can deadlock because I/O on the device might not
2716 gfp_mask
= memalloc_noio_flags(gfp_mask
);
2717 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2718 zonelist
, high_zoneidx
, nodemask
,
2719 preferred_zone
, migratetype
);
2722 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2726 * When updating a task's mems_allowed, it is possible to race with
2727 * parallel threads in such a way that an allocation can fail while
2728 * the mask is being updated. If a page allocation is about to fail,
2729 * check if the cpuset changed during allocation and if so, retry.
2731 if (unlikely(!put_mems_allowed(cpuset_mems_cookie
) && !page
))
2734 memcg_kmem_commit_charge(page
, memcg
, order
);
2738 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2741 * Common helper functions.
2743 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2748 * __get_free_pages() returns a 32-bit address, which cannot represent
2751 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2753 page
= alloc_pages(gfp_mask
, order
);
2756 return (unsigned long) page_address(page
);
2758 EXPORT_SYMBOL(__get_free_pages
);
2760 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2762 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2764 EXPORT_SYMBOL(get_zeroed_page
);
2766 void __free_pages(struct page
*page
, unsigned int order
)
2768 if (put_page_testzero(page
)) {
2770 free_hot_cold_page(page
, 0);
2772 __free_pages_ok(page
, order
);
2776 EXPORT_SYMBOL(__free_pages
);
2778 void free_pages(unsigned long addr
, unsigned int order
)
2781 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2782 __free_pages(virt_to_page((void *)addr
), order
);
2786 EXPORT_SYMBOL(free_pages
);
2789 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2790 * pages allocated with __GFP_KMEMCG.
2792 * Those pages are accounted to a particular memcg, embedded in the
2793 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2794 * for that information only to find out that it is NULL for users who have no
2795 * interest in that whatsoever, we provide these functions.
2797 * The caller knows better which flags it relies on.
2799 void __free_memcg_kmem_pages(struct page
*page
, unsigned int order
)
2801 memcg_kmem_uncharge_pages(page
, order
);
2802 __free_pages(page
, order
);
2805 void free_memcg_kmem_pages(unsigned long addr
, unsigned int order
)
2808 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2809 __free_memcg_kmem_pages(virt_to_page((void *)addr
), order
);
2813 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2816 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2817 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2819 split_page(virt_to_page((void *)addr
), order
);
2820 while (used
< alloc_end
) {
2825 return (void *)addr
;
2829 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2830 * @size: the number of bytes to allocate
2831 * @gfp_mask: GFP flags for the allocation
2833 * This function is similar to alloc_pages(), except that it allocates the
2834 * minimum number of pages to satisfy the request. alloc_pages() can only
2835 * allocate memory in power-of-two pages.
2837 * This function is also limited by MAX_ORDER.
2839 * Memory allocated by this function must be released by free_pages_exact().
2841 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2843 unsigned int order
= get_order(size
);
2846 addr
= __get_free_pages(gfp_mask
, order
);
2847 return make_alloc_exact(addr
, order
, size
);
2849 EXPORT_SYMBOL(alloc_pages_exact
);
2852 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2854 * @nid: the preferred node ID where memory should be allocated
2855 * @size: the number of bytes to allocate
2856 * @gfp_mask: GFP flags for the allocation
2858 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2860 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2863 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2865 unsigned order
= get_order(size
);
2866 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2869 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2871 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2874 * free_pages_exact - release memory allocated via alloc_pages_exact()
2875 * @virt: the value returned by alloc_pages_exact.
2876 * @size: size of allocation, same value as passed to alloc_pages_exact().
2878 * Release the memory allocated by a previous call to alloc_pages_exact.
2880 void free_pages_exact(void *virt
, size_t size
)
2882 unsigned long addr
= (unsigned long)virt
;
2883 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2885 while (addr
< end
) {
2890 EXPORT_SYMBOL(free_pages_exact
);
2893 * nr_free_zone_pages - count number of pages beyond high watermark
2894 * @offset: The zone index of the highest zone
2896 * nr_free_zone_pages() counts the number of counts pages which are beyond the
2897 * high watermark within all zones at or below a given zone index. For each
2898 * zone, the number of pages is calculated as:
2899 * managed_pages - high_pages
2901 static unsigned long nr_free_zone_pages(int offset
)
2906 /* Just pick one node, since fallback list is circular */
2907 unsigned long sum
= 0;
2909 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2911 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2912 unsigned long size
= zone
->managed_pages
;
2913 unsigned long high
= high_wmark_pages(zone
);
2922 * nr_free_buffer_pages - count number of pages beyond high watermark
2924 * nr_free_buffer_pages() counts the number of pages which are beyond the high
2925 * watermark within ZONE_DMA and ZONE_NORMAL.
2927 unsigned long nr_free_buffer_pages(void)
2929 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2931 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2934 * nr_free_pagecache_pages - count number of pages beyond high watermark
2936 * nr_free_pagecache_pages() counts the number of pages which are beyond the
2937 * high watermark within all zones.
2939 unsigned long nr_free_pagecache_pages(void)
2941 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2944 static inline void show_node(struct zone
*zone
)
2946 if (IS_ENABLED(CONFIG_NUMA
))
2947 printk("Node %d ", zone_to_nid(zone
));
2950 void si_meminfo(struct sysinfo
*val
)
2952 val
->totalram
= totalram_pages
;
2954 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2955 val
->bufferram
= nr_blockdev_pages();
2956 val
->totalhigh
= totalhigh_pages
;
2957 val
->freehigh
= nr_free_highpages();
2958 val
->mem_unit
= PAGE_SIZE
;
2961 EXPORT_SYMBOL(si_meminfo
);
2964 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2966 int zone_type
; /* needs to be signed */
2967 unsigned long managed_pages
= 0;
2968 pg_data_t
*pgdat
= NODE_DATA(nid
);
2970 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
2971 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
2972 val
->totalram
= managed_pages
;
2973 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2974 #ifdef CONFIG_HIGHMEM
2975 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
2976 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2982 val
->mem_unit
= PAGE_SIZE
;
2987 * Determine whether the node should be displayed or not, depending on whether
2988 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2990 bool skip_free_areas_node(unsigned int flags
, int nid
)
2993 unsigned int cpuset_mems_cookie
;
2995 if (!(flags
& SHOW_MEM_FILTER_NODES
))
2999 cpuset_mems_cookie
= get_mems_allowed();
3000 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3001 } while (!put_mems_allowed(cpuset_mems_cookie
));
3006 #define K(x) ((x) << (PAGE_SHIFT-10))
3008 static void show_migration_types(unsigned char type
)
3010 static const char types
[MIGRATE_TYPES
] = {
3011 [MIGRATE_UNMOVABLE
] = 'U',
3012 [MIGRATE_RECLAIMABLE
] = 'E',
3013 [MIGRATE_MOVABLE
] = 'M',
3014 [MIGRATE_RESERVE
] = 'R',
3016 [MIGRATE_CMA
] = 'C',
3018 #ifdef CONFIG_MEMORY_ISOLATION
3019 [MIGRATE_ISOLATE
] = 'I',
3022 char tmp
[MIGRATE_TYPES
+ 1];
3026 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3027 if (type
& (1 << i
))
3032 printk("(%s) ", tmp
);
3036 * Show free area list (used inside shift_scroll-lock stuff)
3037 * We also calculate the percentage fragmentation. We do this by counting the
3038 * memory on each free list with the exception of the first item on the list.
3039 * Suppresses nodes that are not allowed by current's cpuset if
3040 * SHOW_MEM_FILTER_NODES is passed.
3042 void show_free_areas(unsigned int filter
)
3047 for_each_populated_zone(zone
) {
3048 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3051 printk("%s per-cpu:\n", zone
->name
);
3053 for_each_online_cpu(cpu
) {
3054 struct per_cpu_pageset
*pageset
;
3056 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
3058 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3059 cpu
, pageset
->pcp
.high
,
3060 pageset
->pcp
.batch
, pageset
->pcp
.count
);
3064 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3065 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3067 " dirty:%lu writeback:%lu unstable:%lu\n"
3068 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3069 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3071 global_page_state(NR_ACTIVE_ANON
),
3072 global_page_state(NR_INACTIVE_ANON
),
3073 global_page_state(NR_ISOLATED_ANON
),
3074 global_page_state(NR_ACTIVE_FILE
),
3075 global_page_state(NR_INACTIVE_FILE
),
3076 global_page_state(NR_ISOLATED_FILE
),
3077 global_page_state(NR_UNEVICTABLE
),
3078 global_page_state(NR_FILE_DIRTY
),
3079 global_page_state(NR_WRITEBACK
),
3080 global_page_state(NR_UNSTABLE_NFS
),
3081 global_page_state(NR_FREE_PAGES
),
3082 global_page_state(NR_SLAB_RECLAIMABLE
),
3083 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3084 global_page_state(NR_FILE_MAPPED
),
3085 global_page_state(NR_SHMEM
),
3086 global_page_state(NR_PAGETABLE
),
3087 global_page_state(NR_BOUNCE
),
3088 global_page_state(NR_FREE_CMA_PAGES
));
3090 for_each_populated_zone(zone
) {
3093 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3101 " active_anon:%lukB"
3102 " inactive_anon:%lukB"
3103 " active_file:%lukB"
3104 " inactive_file:%lukB"
3105 " unevictable:%lukB"
3106 " isolated(anon):%lukB"
3107 " isolated(file):%lukB"
3115 " slab_reclaimable:%lukB"
3116 " slab_unreclaimable:%lukB"
3117 " kernel_stack:%lukB"
3122 " writeback_tmp:%lukB"
3123 " pages_scanned:%lu"
3124 " all_unreclaimable? %s"
3127 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3128 K(min_wmark_pages(zone
)),
3129 K(low_wmark_pages(zone
)),
3130 K(high_wmark_pages(zone
)),
3131 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3132 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3133 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3134 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3135 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3136 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3137 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3138 K(zone
->present_pages
),
3139 K(zone
->managed_pages
),
3140 K(zone_page_state(zone
, NR_MLOCK
)),
3141 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3142 K(zone_page_state(zone
, NR_WRITEBACK
)),
3143 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3144 K(zone_page_state(zone
, NR_SHMEM
)),
3145 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3146 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3147 zone_page_state(zone
, NR_KERNEL_STACK
) *
3149 K(zone_page_state(zone
, NR_PAGETABLE
)),
3150 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3151 K(zone_page_state(zone
, NR_BOUNCE
)),
3152 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3153 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3154 zone
->pages_scanned
,
3155 (!zone_reclaimable(zone
) ? "yes" : "no")
3157 printk("lowmem_reserve[]:");
3158 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3159 printk(" %lu", zone
->lowmem_reserve
[i
]);
3163 for_each_populated_zone(zone
) {
3164 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
3165 unsigned char types
[MAX_ORDER
];
3167 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3170 printk("%s: ", zone
->name
);
3172 spin_lock_irqsave(&zone
->lock
, flags
);
3173 for (order
= 0; order
< MAX_ORDER
; order
++) {
3174 struct free_area
*area
= &zone
->free_area
[order
];
3177 nr
[order
] = area
->nr_free
;
3178 total
+= nr
[order
] << order
;
3181 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3182 if (!list_empty(&area
->free_list
[type
]))
3183 types
[order
] |= 1 << type
;
3186 spin_unlock_irqrestore(&zone
->lock
, flags
);
3187 for (order
= 0; order
< MAX_ORDER
; order
++) {
3188 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3190 show_migration_types(types
[order
]);
3192 printk("= %lukB\n", K(total
));
3195 hugetlb_show_meminfo();
3197 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3199 show_swap_cache_info();
3202 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3204 zoneref
->zone
= zone
;
3205 zoneref
->zone_idx
= zone_idx(zone
);
3209 * Builds allocation fallback zone lists.
3211 * Add all populated zones of a node to the zonelist.
3213 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3217 enum zone_type zone_type
= MAX_NR_ZONES
;
3221 zone
= pgdat
->node_zones
+ zone_type
;
3222 if (populated_zone(zone
)) {
3223 zoneref_set_zone(zone
,
3224 &zonelist
->_zonerefs
[nr_zones
++]);
3225 check_highest_zone(zone_type
);
3227 } while (zone_type
);
3235 * 0 = automatic detection of better ordering.
3236 * 1 = order by ([node] distance, -zonetype)
3237 * 2 = order by (-zonetype, [node] distance)
3239 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3240 * the same zonelist. So only NUMA can configure this param.
3242 #define ZONELIST_ORDER_DEFAULT 0
3243 #define ZONELIST_ORDER_NODE 1
3244 #define ZONELIST_ORDER_ZONE 2
3246 /* zonelist order in the kernel.
3247 * set_zonelist_order() will set this to NODE or ZONE.
3249 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3250 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3254 /* The value user specified ....changed by config */
3255 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3256 /* string for sysctl */
3257 #define NUMA_ZONELIST_ORDER_LEN 16
3258 char numa_zonelist_order
[16] = "default";
3261 * interface for configure zonelist ordering.
3262 * command line option "numa_zonelist_order"
3263 * = "[dD]efault - default, automatic configuration.
3264 * = "[nN]ode - order by node locality, then by zone within node
3265 * = "[zZ]one - order by zone, then by locality within zone
3268 static int __parse_numa_zonelist_order(char *s
)
3270 if (*s
== 'd' || *s
== 'D') {
3271 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3272 } else if (*s
== 'n' || *s
== 'N') {
3273 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3274 } else if (*s
== 'z' || *s
== 'Z') {
3275 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3278 "Ignoring invalid numa_zonelist_order value: "
3285 static __init
int setup_numa_zonelist_order(char *s
)
3292 ret
= __parse_numa_zonelist_order(s
);
3294 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3298 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3301 * sysctl handler for numa_zonelist_order
3303 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
3304 void __user
*buffer
, size_t *length
,
3307 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3309 static DEFINE_MUTEX(zl_order_mutex
);
3311 mutex_lock(&zl_order_mutex
);
3313 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
3317 strcpy(saved_string
, (char *)table
->data
);
3319 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3323 int oldval
= user_zonelist_order
;
3325 ret
= __parse_numa_zonelist_order((char *)table
->data
);
3328 * bogus value. restore saved string
3330 strncpy((char *)table
->data
, saved_string
,
3331 NUMA_ZONELIST_ORDER_LEN
);
3332 user_zonelist_order
= oldval
;
3333 } else if (oldval
!= user_zonelist_order
) {
3334 mutex_lock(&zonelists_mutex
);
3335 build_all_zonelists(NULL
, NULL
);
3336 mutex_unlock(&zonelists_mutex
);
3340 mutex_unlock(&zl_order_mutex
);
3345 #define MAX_NODE_LOAD (nr_online_nodes)
3346 static int node_load
[MAX_NUMNODES
];
3349 * find_next_best_node - find the next node that should appear in a given node's fallback list
3350 * @node: node whose fallback list we're appending
3351 * @used_node_mask: nodemask_t of already used nodes
3353 * We use a number of factors to determine which is the next node that should
3354 * appear on a given node's fallback list. The node should not have appeared
3355 * already in @node's fallback list, and it should be the next closest node
3356 * according to the distance array (which contains arbitrary distance values
3357 * from each node to each node in the system), and should also prefer nodes
3358 * with no CPUs, since presumably they'll have very little allocation pressure
3359 * on them otherwise.
3360 * It returns -1 if no node is found.
3362 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3365 int min_val
= INT_MAX
;
3366 int best_node
= NUMA_NO_NODE
;
3367 const struct cpumask
*tmp
= cpumask_of_node(0);
3369 /* Use the local node if we haven't already */
3370 if (!node_isset(node
, *used_node_mask
)) {
3371 node_set(node
, *used_node_mask
);
3375 for_each_node_state(n
, N_MEMORY
) {
3377 /* Don't want a node to appear more than once */
3378 if (node_isset(n
, *used_node_mask
))
3381 /* Use the distance array to find the distance */
3382 val
= node_distance(node
, n
);
3384 /* Penalize nodes under us ("prefer the next node") */
3387 /* Give preference to headless and unused nodes */
3388 tmp
= cpumask_of_node(n
);
3389 if (!cpumask_empty(tmp
))
3390 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
3392 /* Slight preference for less loaded node */
3393 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
3394 val
+= node_load
[n
];
3396 if (val
< min_val
) {
3403 node_set(best_node
, *used_node_mask
);
3410 * Build zonelists ordered by node and zones within node.
3411 * This results in maximum locality--normal zone overflows into local
3412 * DMA zone, if any--but risks exhausting DMA zone.
3414 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
3417 struct zonelist
*zonelist
;
3419 zonelist
= &pgdat
->node_zonelists
[0];
3420 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
3422 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
3423 zonelist
->_zonerefs
[j
].zone
= NULL
;
3424 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3428 * Build gfp_thisnode zonelists
3430 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
3433 struct zonelist
*zonelist
;
3435 zonelist
= &pgdat
->node_zonelists
[1];
3436 j
= build_zonelists_node(pgdat
, zonelist
, 0);
3437 zonelist
->_zonerefs
[j
].zone
= NULL
;
3438 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3442 * Build zonelists ordered by zone and nodes within zones.
3443 * This results in conserving DMA zone[s] until all Normal memory is
3444 * exhausted, but results in overflowing to remote node while memory
3445 * may still exist in local DMA zone.
3447 static int node_order
[MAX_NUMNODES
];
3449 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
3452 int zone_type
; /* needs to be signed */
3454 struct zonelist
*zonelist
;
3456 zonelist
= &pgdat
->node_zonelists
[0];
3458 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
3459 for (j
= 0; j
< nr_nodes
; j
++) {
3460 node
= node_order
[j
];
3461 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
3462 if (populated_zone(z
)) {
3464 &zonelist
->_zonerefs
[pos
++]);
3465 check_highest_zone(zone_type
);
3469 zonelist
->_zonerefs
[pos
].zone
= NULL
;
3470 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
3473 static int default_zonelist_order(void)
3476 unsigned long low_kmem_size
, total_size
;
3480 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3481 * If they are really small and used heavily, the system can fall
3482 * into OOM very easily.
3483 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3485 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3488 for_each_online_node(nid
) {
3489 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3490 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3491 if (populated_zone(z
)) {
3492 if (zone_type
< ZONE_NORMAL
)
3493 low_kmem_size
+= z
->managed_pages
;
3494 total_size
+= z
->managed_pages
;
3495 } else if (zone_type
== ZONE_NORMAL
) {
3497 * If any node has only lowmem, then node order
3498 * is preferred to allow kernel allocations
3499 * locally; otherwise, they can easily infringe
3500 * on other nodes when there is an abundance of
3501 * lowmem available to allocate from.
3503 return ZONELIST_ORDER_NODE
;
3507 if (!low_kmem_size
|| /* there are no DMA area. */
3508 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
3509 return ZONELIST_ORDER_NODE
;
3511 * look into each node's config.
3512 * If there is a node whose DMA/DMA32 memory is very big area on
3513 * local memory, NODE_ORDER may be suitable.
3515 average_size
= total_size
/
3516 (nodes_weight(node_states
[N_MEMORY
]) + 1);
3517 for_each_online_node(nid
) {
3520 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3521 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3522 if (populated_zone(z
)) {
3523 if (zone_type
< ZONE_NORMAL
)
3524 low_kmem_size
+= z
->present_pages
;
3525 total_size
+= z
->present_pages
;
3528 if (low_kmem_size
&&
3529 total_size
> average_size
&& /* ignore small node */
3530 low_kmem_size
> total_size
* 70/100)
3531 return ZONELIST_ORDER_NODE
;
3533 return ZONELIST_ORDER_ZONE
;
3536 static void set_zonelist_order(void)
3538 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
3539 current_zonelist_order
= default_zonelist_order();
3541 current_zonelist_order
= user_zonelist_order
;
3544 static void build_zonelists(pg_data_t
*pgdat
)
3548 nodemask_t used_mask
;
3549 int local_node
, prev_node
;
3550 struct zonelist
*zonelist
;
3551 int order
= current_zonelist_order
;
3553 /* initialize zonelists */
3554 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3555 zonelist
= pgdat
->node_zonelists
+ i
;
3556 zonelist
->_zonerefs
[0].zone
= NULL
;
3557 zonelist
->_zonerefs
[0].zone_idx
= 0;
3560 /* NUMA-aware ordering of nodes */
3561 local_node
= pgdat
->node_id
;
3562 load
= nr_online_nodes
;
3563 prev_node
= local_node
;
3564 nodes_clear(used_mask
);
3566 memset(node_order
, 0, sizeof(node_order
));
3569 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3571 * We don't want to pressure a particular node.
3572 * So adding penalty to the first node in same
3573 * distance group to make it round-robin.
3575 if (node_distance(local_node
, node
) !=
3576 node_distance(local_node
, prev_node
))
3577 node_load
[node
] = load
;
3581 if (order
== ZONELIST_ORDER_NODE
)
3582 build_zonelists_in_node_order(pgdat
, node
);
3584 node_order
[j
++] = node
; /* remember order */
3587 if (order
== ZONELIST_ORDER_ZONE
) {
3588 /* calculate node order -- i.e., DMA last! */
3589 build_zonelists_in_zone_order(pgdat
, j
);
3592 build_thisnode_zonelists(pgdat
);
3595 /* Construct the zonelist performance cache - see further mmzone.h */
3596 static void build_zonelist_cache(pg_data_t
*pgdat
)
3598 struct zonelist
*zonelist
;
3599 struct zonelist_cache
*zlc
;
3602 zonelist
= &pgdat
->node_zonelists
[0];
3603 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3604 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3605 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3606 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3609 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3611 * Return node id of node used for "local" allocations.
3612 * I.e., first node id of first zone in arg node's generic zonelist.
3613 * Used for initializing percpu 'numa_mem', which is used primarily
3614 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3616 int local_memory_node(int node
)
3620 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3621 gfp_zone(GFP_KERNEL
),
3628 #else /* CONFIG_NUMA */
3630 static void set_zonelist_order(void)
3632 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3635 static void build_zonelists(pg_data_t
*pgdat
)
3637 int node
, local_node
;
3639 struct zonelist
*zonelist
;
3641 local_node
= pgdat
->node_id
;
3643 zonelist
= &pgdat
->node_zonelists
[0];
3644 j
= build_zonelists_node(pgdat
, zonelist
, 0);
3647 * Now we build the zonelist so that it contains the zones
3648 * of all the other nodes.
3649 * We don't want to pressure a particular node, so when
3650 * building the zones for node N, we make sure that the
3651 * zones coming right after the local ones are those from
3652 * node N+1 (modulo N)
3654 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3655 if (!node_online(node
))
3657 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
3659 for (node
= 0; node
< local_node
; node
++) {
3660 if (!node_online(node
))
3662 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
3665 zonelist
->_zonerefs
[j
].zone
= NULL
;
3666 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3669 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3670 static void build_zonelist_cache(pg_data_t
*pgdat
)
3672 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3675 #endif /* CONFIG_NUMA */
3678 * Boot pageset table. One per cpu which is going to be used for all
3679 * zones and all nodes. The parameters will be set in such a way
3680 * that an item put on a list will immediately be handed over to
3681 * the buddy list. This is safe since pageset manipulation is done
3682 * with interrupts disabled.
3684 * The boot_pagesets must be kept even after bootup is complete for
3685 * unused processors and/or zones. They do play a role for bootstrapping
3686 * hotplugged processors.
3688 * zoneinfo_show() and maybe other functions do
3689 * not check if the processor is online before following the pageset pointer.
3690 * Other parts of the kernel may not check if the zone is available.
3692 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3693 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3694 static void setup_zone_pageset(struct zone
*zone
);
3697 * Global mutex to protect against size modification of zonelists
3698 * as well as to serialize pageset setup for the new populated zone.
3700 DEFINE_MUTEX(zonelists_mutex
);
3702 /* return values int ....just for stop_machine() */
3703 static int __build_all_zonelists(void *data
)
3707 pg_data_t
*self
= data
;
3710 memset(node_load
, 0, sizeof(node_load
));
3713 if (self
&& !node_online(self
->node_id
)) {
3714 build_zonelists(self
);
3715 build_zonelist_cache(self
);
3718 for_each_online_node(nid
) {
3719 pg_data_t
*pgdat
= NODE_DATA(nid
);
3721 build_zonelists(pgdat
);
3722 build_zonelist_cache(pgdat
);
3726 * Initialize the boot_pagesets that are going to be used
3727 * for bootstrapping processors. The real pagesets for
3728 * each zone will be allocated later when the per cpu
3729 * allocator is available.
3731 * boot_pagesets are used also for bootstrapping offline
3732 * cpus if the system is already booted because the pagesets
3733 * are needed to initialize allocators on a specific cpu too.
3734 * F.e. the percpu allocator needs the page allocator which
3735 * needs the percpu allocator in order to allocate its pagesets
3736 * (a chicken-egg dilemma).
3738 for_each_possible_cpu(cpu
) {
3739 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3741 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3743 * We now know the "local memory node" for each node--
3744 * i.e., the node of the first zone in the generic zonelist.
3745 * Set up numa_mem percpu variable for on-line cpus. During
3746 * boot, only the boot cpu should be on-line; we'll init the
3747 * secondary cpus' numa_mem as they come on-line. During
3748 * node/memory hotplug, we'll fixup all on-line cpus.
3750 if (cpu_online(cpu
))
3751 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3759 * Called with zonelists_mutex held always
3760 * unless system_state == SYSTEM_BOOTING.
3762 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
3764 set_zonelist_order();
3766 if (system_state
== SYSTEM_BOOTING
) {
3767 __build_all_zonelists(NULL
);
3768 mminit_verify_zonelist();
3769 cpuset_init_current_mems_allowed();
3771 #ifdef CONFIG_MEMORY_HOTPLUG
3773 setup_zone_pageset(zone
);
3775 /* we have to stop all cpus to guarantee there is no user
3777 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
3778 /* cpuset refresh routine should be here */
3780 vm_total_pages
= nr_free_pagecache_pages();
3782 * Disable grouping by mobility if the number of pages in the
3783 * system is too low to allow the mechanism to work. It would be
3784 * more accurate, but expensive to check per-zone. This check is
3785 * made on memory-hotadd so a system can start with mobility
3786 * disabled and enable it later
3788 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3789 page_group_by_mobility_disabled
= 1;
3791 page_group_by_mobility_disabled
= 0;
3793 printk("Built %i zonelists in %s order, mobility grouping %s. "
3794 "Total pages: %ld\n",
3796 zonelist_order_name
[current_zonelist_order
],
3797 page_group_by_mobility_disabled
? "off" : "on",
3800 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3805 * Helper functions to size the waitqueue hash table.
3806 * Essentially these want to choose hash table sizes sufficiently
3807 * large so that collisions trying to wait on pages are rare.
3808 * But in fact, the number of active page waitqueues on typical
3809 * systems is ridiculously low, less than 200. So this is even
3810 * conservative, even though it seems large.
3812 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3813 * waitqueues, i.e. the size of the waitq table given the number of pages.
3815 #define PAGES_PER_WAITQUEUE 256
3817 #ifndef CONFIG_MEMORY_HOTPLUG
3818 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3820 unsigned long size
= 1;
3822 pages
/= PAGES_PER_WAITQUEUE
;
3824 while (size
< pages
)
3828 * Once we have dozens or even hundreds of threads sleeping
3829 * on IO we've got bigger problems than wait queue collision.
3830 * Limit the size of the wait table to a reasonable size.
3832 size
= min(size
, 4096UL);
3834 return max(size
, 4UL);
3838 * A zone's size might be changed by hot-add, so it is not possible to determine
3839 * a suitable size for its wait_table. So we use the maximum size now.
3841 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3843 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3844 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3845 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3847 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3848 * or more by the traditional way. (See above). It equals:
3850 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3851 * ia64(16K page size) : = ( 8G + 4M)byte.
3852 * powerpc (64K page size) : = (32G +16M)byte.
3854 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3861 * This is an integer logarithm so that shifts can be used later
3862 * to extract the more random high bits from the multiplicative
3863 * hash function before the remainder is taken.
3865 static inline unsigned long wait_table_bits(unsigned long size
)
3871 * Check if a pageblock contains reserved pages
3873 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3877 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3878 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3885 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3886 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3887 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3888 * higher will lead to a bigger reserve which will get freed as contiguous
3889 * blocks as reclaim kicks in
3891 static void setup_zone_migrate_reserve(struct zone
*zone
)
3893 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3895 unsigned long block_migratetype
;
3900 * Get the start pfn, end pfn and the number of blocks to reserve
3901 * We have to be careful to be aligned to pageblock_nr_pages to
3902 * make sure that we always check pfn_valid for the first page in
3905 start_pfn
= zone
->zone_start_pfn
;
3906 end_pfn
= zone_end_pfn(zone
);
3907 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3908 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3912 * Reserve blocks are generally in place to help high-order atomic
3913 * allocations that are short-lived. A min_free_kbytes value that
3914 * would result in more than 2 reserve blocks for atomic allocations
3915 * is assumed to be in place to help anti-fragmentation for the
3916 * future allocation of hugepages at runtime.
3918 reserve
= min(2, reserve
);
3919 old_reserve
= zone
->nr_migrate_reserve_block
;
3921 /* When memory hot-add, we almost always need to do nothing */
3922 if (reserve
== old_reserve
)
3924 zone
->nr_migrate_reserve_block
= reserve
;
3926 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3927 if (!pfn_valid(pfn
))
3929 page
= pfn_to_page(pfn
);
3931 /* Watch out for overlapping nodes */
3932 if (page_to_nid(page
) != zone_to_nid(zone
))
3935 block_migratetype
= get_pageblock_migratetype(page
);
3937 /* Only test what is necessary when the reserves are not met */
3940 * Blocks with reserved pages will never free, skip
3943 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3944 if (pageblock_is_reserved(pfn
, block_end_pfn
))
3947 /* If this block is reserved, account for it */
3948 if (block_migratetype
== MIGRATE_RESERVE
) {
3953 /* Suitable for reserving if this block is movable */
3954 if (block_migratetype
== MIGRATE_MOVABLE
) {
3955 set_pageblock_migratetype(page
,
3957 move_freepages_block(zone
, page
,
3962 } else if (!old_reserve
) {
3964 * At boot time we don't need to scan the whole zone
3965 * for turning off MIGRATE_RESERVE.
3971 * If the reserve is met and this is a previous reserved block,
3974 if (block_migratetype
== MIGRATE_RESERVE
) {
3975 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3976 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3982 * Initially all pages are reserved - free ones are freed
3983 * up by free_all_bootmem() once the early boot process is
3984 * done. Non-atomic initialization, single-pass.
3986 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3987 unsigned long start_pfn
, enum memmap_context context
)
3990 unsigned long end_pfn
= start_pfn
+ size
;
3994 if (highest_memmap_pfn
< end_pfn
- 1)
3995 highest_memmap_pfn
= end_pfn
- 1;
3997 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3998 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4000 * There can be holes in boot-time mem_map[]s
4001 * handed to this function. They do not
4002 * exist on hotplugged memory.
4004 if (context
== MEMMAP_EARLY
) {
4005 if (!early_pfn_valid(pfn
))
4007 if (!early_pfn_in_nid(pfn
, nid
))
4010 page
= pfn_to_page(pfn
);
4011 set_page_links(page
, zone
, nid
, pfn
);
4012 mminit_verify_page_links(page
, zone
, nid
, pfn
);
4013 init_page_count(page
);
4014 page_mapcount_reset(page
);
4015 page_cpupid_reset_last(page
);
4016 SetPageReserved(page
);
4018 * Mark the block movable so that blocks are reserved for
4019 * movable at startup. This will force kernel allocations
4020 * to reserve their blocks rather than leaking throughout
4021 * the address space during boot when many long-lived
4022 * kernel allocations are made. Later some blocks near
4023 * the start are marked MIGRATE_RESERVE by
4024 * setup_zone_migrate_reserve()
4026 * bitmap is created for zone's valid pfn range. but memmap
4027 * can be created for invalid pages (for alignment)
4028 * check here not to call set_pageblock_migratetype() against
4031 if ((z
->zone_start_pfn
<= pfn
)
4032 && (pfn
< zone_end_pfn(z
))
4033 && !(pfn
& (pageblock_nr_pages
- 1)))
4034 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4036 INIT_LIST_HEAD(&page
->lru
);
4037 #ifdef WANT_PAGE_VIRTUAL
4038 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4039 if (!is_highmem_idx(zone
))
4040 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
4045 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4048 for_each_migratetype_order(order
, t
) {
4049 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4050 zone
->free_area
[order
].nr_free
= 0;
4054 #ifndef __HAVE_ARCH_MEMMAP_INIT
4055 #define memmap_init(size, nid, zone, start_pfn) \
4056 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4059 static int __meminit
zone_batchsize(struct zone
*zone
)
4065 * The per-cpu-pages pools are set to around 1000th of the
4066 * size of the zone. But no more than 1/2 of a meg.
4068 * OK, so we don't know how big the cache is. So guess.
4070 batch
= zone
->managed_pages
/ 1024;
4071 if (batch
* PAGE_SIZE
> 512 * 1024)
4072 batch
= (512 * 1024) / PAGE_SIZE
;
4073 batch
/= 4; /* We effectively *= 4 below */
4078 * Clamp the batch to a 2^n - 1 value. Having a power
4079 * of 2 value was found to be more likely to have
4080 * suboptimal cache aliasing properties in some cases.
4082 * For example if 2 tasks are alternately allocating
4083 * batches of pages, one task can end up with a lot
4084 * of pages of one half of the possible page colors
4085 * and the other with pages of the other colors.
4087 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4092 /* The deferral and batching of frees should be suppressed under NOMMU
4095 * The problem is that NOMMU needs to be able to allocate large chunks
4096 * of contiguous memory as there's no hardware page translation to
4097 * assemble apparent contiguous memory from discontiguous pages.
4099 * Queueing large contiguous runs of pages for batching, however,
4100 * causes the pages to actually be freed in smaller chunks. As there
4101 * can be a significant delay between the individual batches being
4102 * recycled, this leads to the once large chunks of space being
4103 * fragmented and becoming unavailable for high-order allocations.
4110 * pcp->high and pcp->batch values are related and dependent on one another:
4111 * ->batch must never be higher then ->high.
4112 * The following function updates them in a safe manner without read side
4115 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4116 * those fields changing asynchronously (acording the the above rule).
4118 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4119 * outside of boot time (or some other assurance that no concurrent updaters
4122 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4123 unsigned long batch
)
4125 /* start with a fail safe value for batch */
4129 /* Update high, then batch, in order */
4136 /* a companion to pageset_set_high() */
4137 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4139 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4142 static void pageset_init(struct per_cpu_pageset
*p
)
4144 struct per_cpu_pages
*pcp
;
4147 memset(p
, 0, sizeof(*p
));
4151 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4152 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4155 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4158 pageset_set_batch(p
, batch
);
4162 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4163 * to the value high for the pageset p.
4165 static void pageset_set_high(struct per_cpu_pageset
*p
,
4168 unsigned long batch
= max(1UL, high
/ 4);
4169 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4170 batch
= PAGE_SHIFT
* 8;
4172 pageset_update(&p
->pcp
, high
, batch
);
4175 static void __meminit
pageset_set_high_and_batch(struct zone
*zone
,
4176 struct per_cpu_pageset
*pcp
)
4178 if (percpu_pagelist_fraction
)
4179 pageset_set_high(pcp
,
4180 (zone
->managed_pages
/
4181 percpu_pagelist_fraction
));
4183 pageset_set_batch(pcp
, zone_batchsize(zone
));
4186 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4188 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4191 pageset_set_high_and_batch(zone
, pcp
);
4194 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4197 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4198 for_each_possible_cpu(cpu
)
4199 zone_pageset_init(zone
, cpu
);
4203 * Allocate per cpu pagesets and initialize them.
4204 * Before this call only boot pagesets were available.
4206 void __init
setup_per_cpu_pageset(void)
4210 for_each_populated_zone(zone
)
4211 setup_zone_pageset(zone
);
4214 static noinline __init_refok
4215 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4221 * The per-page waitqueue mechanism uses hashed waitqueues
4224 zone
->wait_table_hash_nr_entries
=
4225 wait_table_hash_nr_entries(zone_size_pages
);
4226 zone
->wait_table_bits
=
4227 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4228 alloc_size
= zone
->wait_table_hash_nr_entries
4229 * sizeof(wait_queue_head_t
);
4231 if (!slab_is_available()) {
4232 zone
->wait_table
= (wait_queue_head_t
*)
4233 memblock_virt_alloc_node_nopanic(
4234 alloc_size
, zone
->zone_pgdat
->node_id
);
4237 * This case means that a zone whose size was 0 gets new memory
4238 * via memory hot-add.
4239 * But it may be the case that a new node was hot-added. In
4240 * this case vmalloc() will not be able to use this new node's
4241 * memory - this wait_table must be initialized to use this new
4242 * node itself as well.
4243 * To use this new node's memory, further consideration will be
4246 zone
->wait_table
= vmalloc(alloc_size
);
4248 if (!zone
->wait_table
)
4251 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4252 init_waitqueue_head(zone
->wait_table
+ i
);
4257 static __meminit
void zone_pcp_init(struct zone
*zone
)
4260 * per cpu subsystem is not up at this point. The following code
4261 * relies on the ability of the linker to provide the
4262 * offset of a (static) per cpu variable into the per cpu area.
4264 zone
->pageset
= &boot_pageset
;
4266 if (populated_zone(zone
))
4267 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4268 zone
->name
, zone
->present_pages
,
4269 zone_batchsize(zone
));
4272 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4273 unsigned long zone_start_pfn
,
4275 enum memmap_context context
)
4277 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4279 ret
= zone_wait_table_init(zone
, size
);
4282 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4284 zone
->zone_start_pfn
= zone_start_pfn
;
4286 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4287 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4289 (unsigned long)zone_idx(zone
),
4290 zone_start_pfn
, (zone_start_pfn
+ size
));
4292 zone_init_free_lists(zone
);
4297 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4298 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4300 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4301 * Architectures may implement their own version but if add_active_range()
4302 * was used and there are no special requirements, this is a convenient
4305 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
4307 unsigned long start_pfn
, end_pfn
;
4310 * NOTE: The following SMP-unsafe globals are only used early in boot
4311 * when the kernel is running single-threaded.
4313 static unsigned long __meminitdata last_start_pfn
, last_end_pfn
;
4314 static int __meminitdata last_nid
;
4316 if (last_start_pfn
<= pfn
&& pfn
< last_end_pfn
)
4319 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4321 last_start_pfn
= start_pfn
;
4322 last_end_pfn
= end_pfn
;
4328 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4330 int __meminit
early_pfn_to_nid(unsigned long pfn
)
4334 nid
= __early_pfn_to_nid(pfn
);
4337 /* just returns 0 */
4341 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4342 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
4346 nid
= __early_pfn_to_nid(pfn
);
4347 if (nid
>= 0 && nid
!= node
)
4354 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4355 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4356 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4358 * If an architecture guarantees that all ranges registered with
4359 * add_active_ranges() contain no holes and may be freed, this
4360 * this function may be used instead of calling memblock_free_early_nid()
4363 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4365 unsigned long start_pfn
, end_pfn
;
4368 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4369 start_pfn
= min(start_pfn
, max_low_pfn
);
4370 end_pfn
= min(end_pfn
, max_low_pfn
);
4372 if (start_pfn
< end_pfn
)
4373 memblock_free_early_nid(PFN_PHYS(start_pfn
),
4374 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
4380 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4381 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4383 * If an architecture guarantees that all ranges registered with
4384 * add_active_ranges() contain no holes and may be freed, this
4385 * function may be used instead of calling memory_present() manually.
4387 void __init
sparse_memory_present_with_active_regions(int nid
)
4389 unsigned long start_pfn
, end_pfn
;
4392 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4393 memory_present(this_nid
, start_pfn
, end_pfn
);
4397 * get_pfn_range_for_nid - Return the start and end page frames for a node
4398 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4399 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4400 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4402 * It returns the start and end page frame of a node based on information
4403 * provided by an arch calling add_active_range(). If called for a node
4404 * with no available memory, a warning is printed and the start and end
4407 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4408 unsigned long *start_pfn
, unsigned long *end_pfn
)
4410 unsigned long this_start_pfn
, this_end_pfn
;
4416 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4417 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4418 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4421 if (*start_pfn
== -1UL)
4426 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4427 * assumption is made that zones within a node are ordered in monotonic
4428 * increasing memory addresses so that the "highest" populated zone is used
4430 static void __init
find_usable_zone_for_movable(void)
4433 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4434 if (zone_index
== ZONE_MOVABLE
)
4437 if (arch_zone_highest_possible_pfn
[zone_index
] >
4438 arch_zone_lowest_possible_pfn
[zone_index
])
4442 VM_BUG_ON(zone_index
== -1);
4443 movable_zone
= zone_index
;
4447 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4448 * because it is sized independent of architecture. Unlike the other zones,
4449 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4450 * in each node depending on the size of each node and how evenly kernelcore
4451 * is distributed. This helper function adjusts the zone ranges
4452 * provided by the architecture for a given node by using the end of the
4453 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4454 * zones within a node are in order of monotonic increases memory addresses
4456 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4457 unsigned long zone_type
,
4458 unsigned long node_start_pfn
,
4459 unsigned long node_end_pfn
,
4460 unsigned long *zone_start_pfn
,
4461 unsigned long *zone_end_pfn
)
4463 /* Only adjust if ZONE_MOVABLE is on this node */
4464 if (zone_movable_pfn
[nid
]) {
4465 /* Size ZONE_MOVABLE */
4466 if (zone_type
== ZONE_MOVABLE
) {
4467 *zone_start_pfn
= zone_movable_pfn
[nid
];
4468 *zone_end_pfn
= min(node_end_pfn
,
4469 arch_zone_highest_possible_pfn
[movable_zone
]);
4471 /* Adjust for ZONE_MOVABLE starting within this range */
4472 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4473 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4474 *zone_end_pfn
= zone_movable_pfn
[nid
];
4476 /* Check if this whole range is within ZONE_MOVABLE */
4477 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4478 *zone_start_pfn
= *zone_end_pfn
;
4483 * Return the number of pages a zone spans in a node, including holes
4484 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4486 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4487 unsigned long zone_type
,
4488 unsigned long node_start_pfn
,
4489 unsigned long node_end_pfn
,
4490 unsigned long *ignored
)
4492 unsigned long zone_start_pfn
, zone_end_pfn
;
4494 /* Get the start and end of the zone */
4495 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4496 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4497 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4498 node_start_pfn
, node_end_pfn
,
4499 &zone_start_pfn
, &zone_end_pfn
);
4501 /* Check that this node has pages within the zone's required range */
4502 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4505 /* Move the zone boundaries inside the node if necessary */
4506 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4507 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4509 /* Return the spanned pages */
4510 return zone_end_pfn
- zone_start_pfn
;
4514 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4515 * then all holes in the requested range will be accounted for.
4517 unsigned long __meminit
__absent_pages_in_range(int nid
,
4518 unsigned long range_start_pfn
,
4519 unsigned long range_end_pfn
)
4521 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4522 unsigned long start_pfn
, end_pfn
;
4525 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4526 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4527 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4528 nr_absent
-= end_pfn
- start_pfn
;
4534 * absent_pages_in_range - Return number of page frames in holes within a range
4535 * @start_pfn: The start PFN to start searching for holes
4536 * @end_pfn: The end PFN to stop searching for holes
4538 * It returns the number of pages frames in memory holes within a range.
4540 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4541 unsigned long end_pfn
)
4543 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4546 /* Return the number of page frames in holes in a zone on a node */
4547 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4548 unsigned long zone_type
,
4549 unsigned long node_start_pfn
,
4550 unsigned long node_end_pfn
,
4551 unsigned long *ignored
)
4553 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4554 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4555 unsigned long zone_start_pfn
, zone_end_pfn
;
4557 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4558 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4560 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4561 node_start_pfn
, node_end_pfn
,
4562 &zone_start_pfn
, &zone_end_pfn
);
4563 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4566 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4567 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4568 unsigned long zone_type
,
4569 unsigned long node_start_pfn
,
4570 unsigned long node_end_pfn
,
4571 unsigned long *zones_size
)
4573 return zones_size
[zone_type
];
4576 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4577 unsigned long zone_type
,
4578 unsigned long node_start_pfn
,
4579 unsigned long node_end_pfn
,
4580 unsigned long *zholes_size
)
4585 return zholes_size
[zone_type
];
4588 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4590 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4591 unsigned long node_start_pfn
,
4592 unsigned long node_end_pfn
,
4593 unsigned long *zones_size
,
4594 unsigned long *zholes_size
)
4596 unsigned long realtotalpages
, totalpages
= 0;
4599 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4600 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4604 pgdat
->node_spanned_pages
= totalpages
;
4606 realtotalpages
= totalpages
;
4607 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4609 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4610 node_start_pfn
, node_end_pfn
,
4612 pgdat
->node_present_pages
= realtotalpages
;
4613 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4617 #ifndef CONFIG_SPARSEMEM
4619 * Calculate the size of the zone->blockflags rounded to an unsigned long
4620 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4621 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4622 * round what is now in bits to nearest long in bits, then return it in
4625 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
4627 unsigned long usemapsize
;
4629 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
4630 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4631 usemapsize
= usemapsize
>> pageblock_order
;
4632 usemapsize
*= NR_PAGEBLOCK_BITS
;
4633 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4635 return usemapsize
/ 8;
4638 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4640 unsigned long zone_start_pfn
,
4641 unsigned long zonesize
)
4643 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
4644 zone
->pageblock_flags
= NULL
;
4646 zone
->pageblock_flags
=
4647 memblock_virt_alloc_node_nopanic(usemapsize
,
4651 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
4652 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
4653 #endif /* CONFIG_SPARSEMEM */
4655 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4657 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4658 void __paginginit
set_pageblock_order(void)
4662 /* Check that pageblock_nr_pages has not already been setup */
4663 if (pageblock_order
)
4666 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4667 order
= HUGETLB_PAGE_ORDER
;
4669 order
= MAX_ORDER
- 1;
4672 * Assume the largest contiguous order of interest is a huge page.
4673 * This value may be variable depending on boot parameters on IA64 and
4676 pageblock_order
= order
;
4678 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4681 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4682 * is unused as pageblock_order is set at compile-time. See
4683 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4686 void __paginginit
set_pageblock_order(void)
4690 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4692 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
4693 unsigned long present_pages
)
4695 unsigned long pages
= spanned_pages
;
4698 * Provide a more accurate estimation if there are holes within
4699 * the zone and SPARSEMEM is in use. If there are holes within the
4700 * zone, each populated memory region may cost us one or two extra
4701 * memmap pages due to alignment because memmap pages for each
4702 * populated regions may not naturally algined on page boundary.
4703 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4705 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
4706 IS_ENABLED(CONFIG_SPARSEMEM
))
4707 pages
= present_pages
;
4709 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
4713 * Set up the zone data structures:
4714 * - mark all pages reserved
4715 * - mark all memory queues empty
4716 * - clear the memory bitmaps
4718 * NOTE: pgdat should get zeroed by caller.
4720 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4721 unsigned long node_start_pfn
, unsigned long node_end_pfn
,
4722 unsigned long *zones_size
, unsigned long *zholes_size
)
4725 int nid
= pgdat
->node_id
;
4726 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4729 pgdat_resize_init(pgdat
);
4730 #ifdef CONFIG_NUMA_BALANCING
4731 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
4732 pgdat
->numabalancing_migrate_nr_pages
= 0;
4733 pgdat
->numabalancing_migrate_next_window
= jiffies
;
4735 init_waitqueue_head(&pgdat
->kswapd_wait
);
4736 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
4737 pgdat_page_cgroup_init(pgdat
);
4739 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4740 struct zone
*zone
= pgdat
->node_zones
+ j
;
4741 unsigned long size
, realsize
, freesize
, memmap_pages
;
4743 size
= zone_spanned_pages_in_node(nid
, j
, node_start_pfn
,
4744 node_end_pfn
, zones_size
);
4745 realsize
= freesize
= size
- zone_absent_pages_in_node(nid
, j
,
4751 * Adjust freesize so that it accounts for how much memory
4752 * is used by this zone for memmap. This affects the watermark
4753 * and per-cpu initialisations
4755 memmap_pages
= calc_memmap_size(size
, realsize
);
4756 if (freesize
>= memmap_pages
) {
4757 freesize
-= memmap_pages
;
4760 " %s zone: %lu pages used for memmap\n",
4761 zone_names
[j
], memmap_pages
);
4764 " %s zone: %lu pages exceeds freesize %lu\n",
4765 zone_names
[j
], memmap_pages
, freesize
);
4767 /* Account for reserved pages */
4768 if (j
== 0 && freesize
> dma_reserve
) {
4769 freesize
-= dma_reserve
;
4770 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4771 zone_names
[0], dma_reserve
);
4774 if (!is_highmem_idx(j
))
4775 nr_kernel_pages
+= freesize
;
4776 /* Charge for highmem memmap if there are enough kernel pages */
4777 else if (nr_kernel_pages
> memmap_pages
* 2)
4778 nr_kernel_pages
-= memmap_pages
;
4779 nr_all_pages
+= freesize
;
4781 zone
->spanned_pages
= size
;
4782 zone
->present_pages
= realsize
;
4784 * Set an approximate value for lowmem here, it will be adjusted
4785 * when the bootmem allocator frees pages into the buddy system.
4786 * And all highmem pages will be managed by the buddy system.
4788 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
4791 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
4793 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
4795 zone
->name
= zone_names
[j
];
4796 spin_lock_init(&zone
->lock
);
4797 spin_lock_init(&zone
->lru_lock
);
4798 zone_seqlock_init(zone
);
4799 zone
->zone_pgdat
= pgdat
;
4800 zone_pcp_init(zone
);
4802 /* For bootup, initialized properly in watermark setup */
4803 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
4805 lruvec_init(&zone
->lruvec
);
4809 set_pageblock_order();
4810 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
4811 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4812 size
, MEMMAP_EARLY
);
4814 memmap_init(size
, nid
, j
, zone_start_pfn
);
4815 zone_start_pfn
+= size
;
4819 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4821 /* Skip empty nodes */
4822 if (!pgdat
->node_spanned_pages
)
4825 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4826 /* ia64 gets its own node_mem_map, before this, without bootmem */
4827 if (!pgdat
->node_mem_map
) {
4828 unsigned long size
, start
, end
;
4832 * The zone's endpoints aren't required to be MAX_ORDER
4833 * aligned but the node_mem_map endpoints must be in order
4834 * for the buddy allocator to function correctly.
4836 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4837 end
= pgdat_end_pfn(pgdat
);
4838 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4839 size
= (end
- start
) * sizeof(struct page
);
4840 map
= alloc_remap(pgdat
->node_id
, size
);
4842 map
= memblock_virt_alloc_node_nopanic(size
,
4844 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4846 #ifndef CONFIG_NEED_MULTIPLE_NODES
4848 * With no DISCONTIG, the global mem_map is just set as node 0's
4850 if (pgdat
== NODE_DATA(0)) {
4851 mem_map
= NODE_DATA(0)->node_mem_map
;
4852 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4853 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4854 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4855 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4858 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4861 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4862 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4864 pg_data_t
*pgdat
= NODE_DATA(nid
);
4865 unsigned long start_pfn
= 0;
4866 unsigned long end_pfn
= 0;
4868 /* pg_data_t should be reset to zero when it's allocated */
4869 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
4871 pgdat
->node_id
= nid
;
4872 pgdat
->node_start_pfn
= node_start_pfn
;
4873 init_zone_allows_reclaim(nid
);
4874 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4875 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
4877 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
4878 zones_size
, zholes_size
);
4880 alloc_node_mem_map(pgdat
);
4881 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4882 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4883 nid
, (unsigned long)pgdat
,
4884 (unsigned long)pgdat
->node_mem_map
);
4887 free_area_init_core(pgdat
, start_pfn
, end_pfn
,
4888 zones_size
, zholes_size
);
4891 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4893 #if MAX_NUMNODES > 1
4895 * Figure out the number of possible node ids.
4897 void __init
setup_nr_node_ids(void)
4900 unsigned int highest
= 0;
4902 for_each_node_mask(node
, node_possible_map
)
4904 nr_node_ids
= highest
+ 1;
4909 * node_map_pfn_alignment - determine the maximum internode alignment
4911 * This function should be called after node map is populated and sorted.
4912 * It calculates the maximum power of two alignment which can distinguish
4915 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4916 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4917 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4918 * shifted, 1GiB is enough and this function will indicate so.
4920 * This is used to test whether pfn -> nid mapping of the chosen memory
4921 * model has fine enough granularity to avoid incorrect mapping for the
4922 * populated node map.
4924 * Returns the determined alignment in pfn's. 0 if there is no alignment
4925 * requirement (single node).
4927 unsigned long __init
node_map_pfn_alignment(void)
4929 unsigned long accl_mask
= 0, last_end
= 0;
4930 unsigned long start
, end
, mask
;
4934 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4935 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4942 * Start with a mask granular enough to pin-point to the
4943 * start pfn and tick off bits one-by-one until it becomes
4944 * too coarse to separate the current node from the last.
4946 mask
= ~((1 << __ffs(start
)) - 1);
4947 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4950 /* accumulate all internode masks */
4954 /* convert mask to number of pages */
4955 return ~accl_mask
+ 1;
4958 /* Find the lowest pfn for a node */
4959 static unsigned long __init
find_min_pfn_for_node(int nid
)
4961 unsigned long min_pfn
= ULONG_MAX
;
4962 unsigned long start_pfn
;
4965 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4966 min_pfn
= min(min_pfn
, start_pfn
);
4968 if (min_pfn
== ULONG_MAX
) {
4970 "Could not find start_pfn for node %d\n", nid
);
4978 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4980 * It returns the minimum PFN based on information provided via
4981 * add_active_range().
4983 unsigned long __init
find_min_pfn_with_active_regions(void)
4985 return find_min_pfn_for_node(MAX_NUMNODES
);
4989 * early_calculate_totalpages()
4990 * Sum pages in active regions for movable zone.
4991 * Populate N_MEMORY for calculating usable_nodes.
4993 static unsigned long __init
early_calculate_totalpages(void)
4995 unsigned long totalpages
= 0;
4996 unsigned long start_pfn
, end_pfn
;
4999 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5000 unsigned long pages
= end_pfn
- start_pfn
;
5002 totalpages
+= pages
;
5004 node_set_state(nid
, N_MEMORY
);
5010 * Find the PFN the Movable zone begins in each node. Kernel memory
5011 * is spread evenly between nodes as long as the nodes have enough
5012 * memory. When they don't, some nodes will have more kernelcore than
5015 static void __init
find_zone_movable_pfns_for_nodes(void)
5018 unsigned long usable_startpfn
;
5019 unsigned long kernelcore_node
, kernelcore_remaining
;
5020 /* save the state before borrow the nodemask */
5021 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5022 unsigned long totalpages
= early_calculate_totalpages();
5023 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5024 struct memblock_type
*type
= &memblock
.memory
;
5026 /* Need to find movable_zone earlier when movable_node is specified. */
5027 find_usable_zone_for_movable();
5030 * If movable_node is specified, ignore kernelcore and movablecore
5033 if (movable_node_is_enabled()) {
5034 for (i
= 0; i
< type
->cnt
; i
++) {
5035 if (!memblock_is_hotpluggable(&type
->regions
[i
]))
5038 nid
= type
->regions
[i
].nid
;
5040 usable_startpfn
= PFN_DOWN(type
->regions
[i
].base
);
5041 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5042 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5050 * If movablecore=nn[KMG] was specified, calculate what size of
5051 * kernelcore that corresponds so that memory usable for
5052 * any allocation type is evenly spread. If both kernelcore
5053 * and movablecore are specified, then the value of kernelcore
5054 * will be used for required_kernelcore if it's greater than
5055 * what movablecore would have allowed.
5057 if (required_movablecore
) {
5058 unsigned long corepages
;
5061 * Round-up so that ZONE_MOVABLE is at least as large as what
5062 * was requested by the user
5064 required_movablecore
=
5065 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5066 corepages
= totalpages
- required_movablecore
;
5068 required_kernelcore
= max(required_kernelcore
, corepages
);
5071 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5072 if (!required_kernelcore
)
5075 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5076 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5079 /* Spread kernelcore memory as evenly as possible throughout nodes */
5080 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5081 for_each_node_state(nid
, N_MEMORY
) {
5082 unsigned long start_pfn
, end_pfn
;
5085 * Recalculate kernelcore_node if the division per node
5086 * now exceeds what is necessary to satisfy the requested
5087 * amount of memory for the kernel
5089 if (required_kernelcore
< kernelcore_node
)
5090 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5093 * As the map is walked, we track how much memory is usable
5094 * by the kernel using kernelcore_remaining. When it is
5095 * 0, the rest of the node is usable by ZONE_MOVABLE
5097 kernelcore_remaining
= kernelcore_node
;
5099 /* Go through each range of PFNs within this node */
5100 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5101 unsigned long size_pages
;
5103 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5104 if (start_pfn
>= end_pfn
)
5107 /* Account for what is only usable for kernelcore */
5108 if (start_pfn
< usable_startpfn
) {
5109 unsigned long kernel_pages
;
5110 kernel_pages
= min(end_pfn
, usable_startpfn
)
5113 kernelcore_remaining
-= min(kernel_pages
,
5114 kernelcore_remaining
);
5115 required_kernelcore
-= min(kernel_pages
,
5116 required_kernelcore
);
5118 /* Continue if range is now fully accounted */
5119 if (end_pfn
<= usable_startpfn
) {
5122 * Push zone_movable_pfn to the end so
5123 * that if we have to rebalance
5124 * kernelcore across nodes, we will
5125 * not double account here
5127 zone_movable_pfn
[nid
] = end_pfn
;
5130 start_pfn
= usable_startpfn
;
5134 * The usable PFN range for ZONE_MOVABLE is from
5135 * start_pfn->end_pfn. Calculate size_pages as the
5136 * number of pages used as kernelcore
5138 size_pages
= end_pfn
- start_pfn
;
5139 if (size_pages
> kernelcore_remaining
)
5140 size_pages
= kernelcore_remaining
;
5141 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5144 * Some kernelcore has been met, update counts and
5145 * break if the kernelcore for this node has been
5148 required_kernelcore
-= min(required_kernelcore
,
5150 kernelcore_remaining
-= size_pages
;
5151 if (!kernelcore_remaining
)
5157 * If there is still required_kernelcore, we do another pass with one
5158 * less node in the count. This will push zone_movable_pfn[nid] further
5159 * along on the nodes that still have memory until kernelcore is
5163 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5167 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5168 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5169 zone_movable_pfn
[nid
] =
5170 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5173 /* restore the node_state */
5174 node_states
[N_MEMORY
] = saved_node_state
;
5177 /* Any regular or high memory on that node ? */
5178 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5180 enum zone_type zone_type
;
5182 if (N_MEMORY
== N_NORMAL_MEMORY
)
5185 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5186 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5187 if (populated_zone(zone
)) {
5188 node_set_state(nid
, N_HIGH_MEMORY
);
5189 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5190 zone_type
<= ZONE_NORMAL
)
5191 node_set_state(nid
, N_NORMAL_MEMORY
);
5198 * free_area_init_nodes - Initialise all pg_data_t and zone data
5199 * @max_zone_pfn: an array of max PFNs for each zone
5201 * This will call free_area_init_node() for each active node in the system.
5202 * Using the page ranges provided by add_active_range(), the size of each
5203 * zone in each node and their holes is calculated. If the maximum PFN
5204 * between two adjacent zones match, it is assumed that the zone is empty.
5205 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5206 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5207 * starts where the previous one ended. For example, ZONE_DMA32 starts
5208 * at arch_max_dma_pfn.
5210 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5212 unsigned long start_pfn
, end_pfn
;
5215 /* Record where the zone boundaries are */
5216 memset(arch_zone_lowest_possible_pfn
, 0,
5217 sizeof(arch_zone_lowest_possible_pfn
));
5218 memset(arch_zone_highest_possible_pfn
, 0,
5219 sizeof(arch_zone_highest_possible_pfn
));
5220 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5221 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5222 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5223 if (i
== ZONE_MOVABLE
)
5225 arch_zone_lowest_possible_pfn
[i
] =
5226 arch_zone_highest_possible_pfn
[i
-1];
5227 arch_zone_highest_possible_pfn
[i
] =
5228 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5230 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5231 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5233 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5234 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5235 find_zone_movable_pfns_for_nodes();
5237 /* Print out the zone ranges */
5238 printk("Zone ranges:\n");
5239 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5240 if (i
== ZONE_MOVABLE
)
5242 printk(KERN_CONT
" %-8s ", zone_names
[i
]);
5243 if (arch_zone_lowest_possible_pfn
[i
] ==
5244 arch_zone_highest_possible_pfn
[i
])
5245 printk(KERN_CONT
"empty\n");
5247 printk(KERN_CONT
"[mem %0#10lx-%0#10lx]\n",
5248 arch_zone_lowest_possible_pfn
[i
] << PAGE_SHIFT
,
5249 (arch_zone_highest_possible_pfn
[i
]
5250 << PAGE_SHIFT
) - 1);
5253 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5254 printk("Movable zone start for each node\n");
5255 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5256 if (zone_movable_pfn
[i
])
5257 printk(" Node %d: %#010lx\n", i
,
5258 zone_movable_pfn
[i
] << PAGE_SHIFT
);
5261 /* Print out the early node map */
5262 printk("Early memory node ranges\n");
5263 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5264 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid
,
5265 start_pfn
<< PAGE_SHIFT
, (end_pfn
<< PAGE_SHIFT
) - 1);
5267 /* Initialise every node */
5268 mminit_verify_pageflags_layout();
5269 setup_nr_node_ids();
5270 for_each_online_node(nid
) {
5271 pg_data_t
*pgdat
= NODE_DATA(nid
);
5272 free_area_init_node(nid
, NULL
,
5273 find_min_pfn_for_node(nid
), NULL
);
5275 /* Any memory on that node */
5276 if (pgdat
->node_present_pages
)
5277 node_set_state(nid
, N_MEMORY
);
5278 check_for_memory(pgdat
, nid
);
5282 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5284 unsigned long long coremem
;
5288 coremem
= memparse(p
, &p
);
5289 *core
= coremem
>> PAGE_SHIFT
;
5291 /* Paranoid check that UL is enough for the coremem value */
5292 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5298 * kernelcore=size sets the amount of memory for use for allocations that
5299 * cannot be reclaimed or migrated.
5301 static int __init
cmdline_parse_kernelcore(char *p
)
5303 return cmdline_parse_core(p
, &required_kernelcore
);
5307 * movablecore=size sets the amount of memory for use for allocations that
5308 * can be reclaimed or migrated.
5310 static int __init
cmdline_parse_movablecore(char *p
)
5312 return cmdline_parse_core(p
, &required_movablecore
);
5315 early_param("kernelcore", cmdline_parse_kernelcore
);
5316 early_param("movablecore", cmdline_parse_movablecore
);
5318 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5320 void adjust_managed_page_count(struct page
*page
, long count
)
5322 spin_lock(&managed_page_count_lock
);
5323 page_zone(page
)->managed_pages
+= count
;
5324 totalram_pages
+= count
;
5325 #ifdef CONFIG_HIGHMEM
5326 if (PageHighMem(page
))
5327 totalhigh_pages
+= count
;
5329 spin_unlock(&managed_page_count_lock
);
5331 EXPORT_SYMBOL(adjust_managed_page_count
);
5333 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
5336 unsigned long pages
= 0;
5338 start
= (void *)PAGE_ALIGN((unsigned long)start
);
5339 end
= (void *)((unsigned long)end
& PAGE_MASK
);
5340 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5341 if ((unsigned int)poison
<= 0xFF)
5342 memset(pos
, poison
, PAGE_SIZE
);
5343 free_reserved_page(virt_to_page(pos
));
5347 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5348 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
5352 EXPORT_SYMBOL(free_reserved_area
);
5354 #ifdef CONFIG_HIGHMEM
5355 void free_highmem_page(struct page
*page
)
5357 __free_reserved_page(page
);
5359 page_zone(page
)->managed_pages
++;
5365 void __init
mem_init_print_info(const char *str
)
5367 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
5368 unsigned long init_code_size
, init_data_size
;
5370 physpages
= get_num_physpages();
5371 codesize
= _etext
- _stext
;
5372 datasize
= _edata
- _sdata
;
5373 rosize
= __end_rodata
- __start_rodata
;
5374 bss_size
= __bss_stop
- __bss_start
;
5375 init_data_size
= __init_end
- __init_begin
;
5376 init_code_size
= _einittext
- _sinittext
;
5379 * Detect special cases and adjust section sizes accordingly:
5380 * 1) .init.* may be embedded into .data sections
5381 * 2) .init.text.* may be out of [__init_begin, __init_end],
5382 * please refer to arch/tile/kernel/vmlinux.lds.S.
5383 * 3) .rodata.* may be embedded into .text or .data sections.
5385 #define adj_init_size(start, end, size, pos, adj) \
5387 if (start <= pos && pos < end && size > adj) \
5391 adj_init_size(__init_begin
, __init_end
, init_data_size
,
5392 _sinittext
, init_code_size
);
5393 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
5394 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
5395 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
5396 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
5398 #undef adj_init_size
5400 printk("Memory: %luK/%luK available "
5401 "(%luK kernel code, %luK rwdata, %luK rodata, "
5402 "%luK init, %luK bss, %luK reserved"
5403 #ifdef CONFIG_HIGHMEM
5407 nr_free_pages() << (PAGE_SHIFT
-10), physpages
<< (PAGE_SHIFT
-10),
5408 codesize
>> 10, datasize
>> 10, rosize
>> 10,
5409 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
5410 (physpages
- totalram_pages
) << (PAGE_SHIFT
-10),
5411 #ifdef CONFIG_HIGHMEM
5412 totalhigh_pages
<< (PAGE_SHIFT
-10),
5414 str
? ", " : "", str
? str
: "");
5418 * set_dma_reserve - set the specified number of pages reserved in the first zone
5419 * @new_dma_reserve: The number of pages to mark reserved
5421 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5422 * In the DMA zone, a significant percentage may be consumed by kernel image
5423 * and other unfreeable allocations which can skew the watermarks badly. This
5424 * function may optionally be used to account for unfreeable pages in the
5425 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5426 * smaller per-cpu batchsize.
5428 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5430 dma_reserve
= new_dma_reserve
;
5433 void __init
free_area_init(unsigned long *zones_size
)
5435 free_area_init_node(0, zones_size
,
5436 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5439 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5440 unsigned long action
, void *hcpu
)
5442 int cpu
= (unsigned long)hcpu
;
5444 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5445 lru_add_drain_cpu(cpu
);
5449 * Spill the event counters of the dead processor
5450 * into the current processors event counters.
5451 * This artificially elevates the count of the current
5454 vm_events_fold_cpu(cpu
);
5457 * Zero the differential counters of the dead processor
5458 * so that the vm statistics are consistent.
5460 * This is only okay since the processor is dead and cannot
5461 * race with what we are doing.
5463 cpu_vm_stats_fold(cpu
);
5468 void __init
page_alloc_init(void)
5470 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5474 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5475 * or min_free_kbytes changes.
5477 static void calculate_totalreserve_pages(void)
5479 struct pglist_data
*pgdat
;
5480 unsigned long reserve_pages
= 0;
5481 enum zone_type i
, j
;
5483 for_each_online_pgdat(pgdat
) {
5484 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5485 struct zone
*zone
= pgdat
->node_zones
+ i
;
5486 unsigned long max
= 0;
5488 /* Find valid and maximum lowmem_reserve in the zone */
5489 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5490 if (zone
->lowmem_reserve
[j
] > max
)
5491 max
= zone
->lowmem_reserve
[j
];
5494 /* we treat the high watermark as reserved pages. */
5495 max
+= high_wmark_pages(zone
);
5497 if (max
> zone
->managed_pages
)
5498 max
= zone
->managed_pages
;
5499 reserve_pages
+= max
;
5501 * Lowmem reserves are not available to
5502 * GFP_HIGHUSER page cache allocations and
5503 * kswapd tries to balance zones to their high
5504 * watermark. As a result, neither should be
5505 * regarded as dirtyable memory, to prevent a
5506 * situation where reclaim has to clean pages
5507 * in order to balance the zones.
5509 zone
->dirty_balance_reserve
= max
;
5512 dirty_balance_reserve
= reserve_pages
;
5513 totalreserve_pages
= reserve_pages
;
5517 * setup_per_zone_lowmem_reserve - called whenever
5518 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5519 * has a correct pages reserved value, so an adequate number of
5520 * pages are left in the zone after a successful __alloc_pages().
5522 static void setup_per_zone_lowmem_reserve(void)
5524 struct pglist_data
*pgdat
;
5525 enum zone_type j
, idx
;
5527 for_each_online_pgdat(pgdat
) {
5528 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5529 struct zone
*zone
= pgdat
->node_zones
+ j
;
5530 unsigned long managed_pages
= zone
->managed_pages
;
5532 zone
->lowmem_reserve
[j
] = 0;
5536 struct zone
*lower_zone
;
5540 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5541 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5543 lower_zone
= pgdat
->node_zones
+ idx
;
5544 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
5545 sysctl_lowmem_reserve_ratio
[idx
];
5546 managed_pages
+= lower_zone
->managed_pages
;
5551 /* update totalreserve_pages */
5552 calculate_totalreserve_pages();
5555 static void __setup_per_zone_wmarks(void)
5557 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5558 unsigned long lowmem_pages
= 0;
5560 unsigned long flags
;
5562 /* Calculate total number of !ZONE_HIGHMEM pages */
5563 for_each_zone(zone
) {
5564 if (!is_highmem(zone
))
5565 lowmem_pages
+= zone
->managed_pages
;
5568 for_each_zone(zone
) {
5571 spin_lock_irqsave(&zone
->lock
, flags
);
5572 tmp
= (u64
)pages_min
* zone
->managed_pages
;
5573 do_div(tmp
, lowmem_pages
);
5574 if (is_highmem(zone
)) {
5576 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5577 * need highmem pages, so cap pages_min to a small
5580 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5581 * deltas controls asynch page reclaim, and so should
5582 * not be capped for highmem.
5584 unsigned long min_pages
;
5586 min_pages
= zone
->managed_pages
/ 1024;
5587 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
5588 zone
->watermark
[WMARK_MIN
] = min_pages
;
5591 * If it's a lowmem zone, reserve a number of pages
5592 * proportionate to the zone's size.
5594 zone
->watermark
[WMARK_MIN
] = tmp
;
5597 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
5598 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
5600 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
5601 high_wmark_pages(zone
) -
5602 low_wmark_pages(zone
) -
5603 zone_page_state(zone
, NR_ALLOC_BATCH
));
5605 setup_zone_migrate_reserve(zone
);
5606 spin_unlock_irqrestore(&zone
->lock
, flags
);
5609 /* update totalreserve_pages */
5610 calculate_totalreserve_pages();
5614 * setup_per_zone_wmarks - called when min_free_kbytes changes
5615 * or when memory is hot-{added|removed}
5617 * Ensures that the watermark[min,low,high] values for each zone are set
5618 * correctly with respect to min_free_kbytes.
5620 void setup_per_zone_wmarks(void)
5622 mutex_lock(&zonelists_mutex
);
5623 __setup_per_zone_wmarks();
5624 mutex_unlock(&zonelists_mutex
);
5628 * The inactive anon list should be small enough that the VM never has to
5629 * do too much work, but large enough that each inactive page has a chance
5630 * to be referenced again before it is swapped out.
5632 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5633 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5634 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5635 * the anonymous pages are kept on the inactive list.
5638 * memory ratio inactive anon
5639 * -------------------------------------
5648 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
5650 unsigned int gb
, ratio
;
5652 /* Zone size in gigabytes */
5653 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
5655 ratio
= int_sqrt(10 * gb
);
5659 zone
->inactive_ratio
= ratio
;
5662 static void __meminit
setup_per_zone_inactive_ratio(void)
5667 calculate_zone_inactive_ratio(zone
);
5671 * Initialise min_free_kbytes.
5673 * For small machines we want it small (128k min). For large machines
5674 * we want it large (64MB max). But it is not linear, because network
5675 * bandwidth does not increase linearly with machine size. We use
5677 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5678 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5694 int __meminit
init_per_zone_wmark_min(void)
5696 unsigned long lowmem_kbytes
;
5697 int new_min_free_kbytes
;
5699 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5700 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5702 if (new_min_free_kbytes
> user_min_free_kbytes
) {
5703 min_free_kbytes
= new_min_free_kbytes
;
5704 if (min_free_kbytes
< 128)
5705 min_free_kbytes
= 128;
5706 if (min_free_kbytes
> 65536)
5707 min_free_kbytes
= 65536;
5709 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5710 new_min_free_kbytes
, user_min_free_kbytes
);
5712 setup_per_zone_wmarks();
5713 refresh_zone_stat_thresholds();
5714 setup_per_zone_lowmem_reserve();
5715 setup_per_zone_inactive_ratio();
5718 module_init(init_per_zone_wmark_min
)
5721 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5722 * that we can call two helper functions whenever min_free_kbytes
5725 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5726 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5728 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5730 user_min_free_kbytes
= min_free_kbytes
;
5731 setup_per_zone_wmarks();
5737 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5738 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5743 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5748 zone
->min_unmapped_pages
= (zone
->managed_pages
*
5749 sysctl_min_unmapped_ratio
) / 100;
5753 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5754 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5759 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5764 zone
->min_slab_pages
= (zone
->managed_pages
*
5765 sysctl_min_slab_ratio
) / 100;
5771 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5772 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5773 * whenever sysctl_lowmem_reserve_ratio changes.
5775 * The reserve ratio obviously has absolutely no relation with the
5776 * minimum watermarks. The lowmem reserve ratio can only make sense
5777 * if in function of the boot time zone sizes.
5779 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5780 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5782 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5783 setup_per_zone_lowmem_reserve();
5788 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5789 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5790 * pagelist can have before it gets flushed back to buddy allocator.
5792 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5793 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5799 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5800 if (!write
|| (ret
< 0))
5803 mutex_lock(&pcp_batch_high_lock
);
5804 for_each_populated_zone(zone
) {
5806 high
= zone
->managed_pages
/ percpu_pagelist_fraction
;
5807 for_each_possible_cpu(cpu
)
5808 pageset_set_high(per_cpu_ptr(zone
->pageset
, cpu
),
5811 mutex_unlock(&pcp_batch_high_lock
);
5815 int hashdist
= HASHDIST_DEFAULT
;
5818 static int __init
set_hashdist(char *str
)
5822 hashdist
= simple_strtoul(str
, &str
, 0);
5825 __setup("hashdist=", set_hashdist
);
5829 * allocate a large system hash table from bootmem
5830 * - it is assumed that the hash table must contain an exact power-of-2
5831 * quantity of entries
5832 * - limit is the number of hash buckets, not the total allocation size
5834 void *__init
alloc_large_system_hash(const char *tablename
,
5835 unsigned long bucketsize
,
5836 unsigned long numentries
,
5839 unsigned int *_hash_shift
,
5840 unsigned int *_hash_mask
,
5841 unsigned long low_limit
,
5842 unsigned long high_limit
)
5844 unsigned long long max
= high_limit
;
5845 unsigned long log2qty
, size
;
5848 /* allow the kernel cmdline to have a say */
5850 /* round applicable memory size up to nearest megabyte */
5851 numentries
= nr_kernel_pages
;
5853 /* It isn't necessary when PAGE_SIZE >= 1MB */
5854 if (PAGE_SHIFT
< 20)
5855 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
5857 /* limit to 1 bucket per 2^scale bytes of low memory */
5858 if (scale
> PAGE_SHIFT
)
5859 numentries
>>= (scale
- PAGE_SHIFT
);
5861 numentries
<<= (PAGE_SHIFT
- scale
);
5863 /* Make sure we've got at least a 0-order allocation.. */
5864 if (unlikely(flags
& HASH_SMALL
)) {
5865 /* Makes no sense without HASH_EARLY */
5866 WARN_ON(!(flags
& HASH_EARLY
));
5867 if (!(numentries
>> *_hash_shift
)) {
5868 numentries
= 1UL << *_hash_shift
;
5869 BUG_ON(!numentries
);
5871 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5872 numentries
= PAGE_SIZE
/ bucketsize
;
5874 numentries
= roundup_pow_of_two(numentries
);
5876 /* limit allocation size to 1/16 total memory by default */
5878 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5879 do_div(max
, bucketsize
);
5881 max
= min(max
, 0x80000000ULL
);
5883 if (numentries
< low_limit
)
5884 numentries
= low_limit
;
5885 if (numentries
> max
)
5888 log2qty
= ilog2(numentries
);
5891 size
= bucketsize
<< log2qty
;
5892 if (flags
& HASH_EARLY
)
5893 table
= memblock_virt_alloc_nopanic(size
, 0);
5895 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5898 * If bucketsize is not a power-of-two, we may free
5899 * some pages at the end of hash table which
5900 * alloc_pages_exact() automatically does
5902 if (get_order(size
) < MAX_ORDER
) {
5903 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5904 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5907 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5910 panic("Failed to allocate %s hash table\n", tablename
);
5912 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5915 ilog2(size
) - PAGE_SHIFT
,
5919 *_hash_shift
= log2qty
;
5921 *_hash_mask
= (1 << log2qty
) - 1;
5926 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5927 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5930 #ifdef CONFIG_SPARSEMEM
5931 return __pfn_to_section(pfn
)->pageblock_flags
;
5933 return zone
->pageblock_flags
;
5934 #endif /* CONFIG_SPARSEMEM */
5937 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5939 #ifdef CONFIG_SPARSEMEM
5940 pfn
&= (PAGES_PER_SECTION
-1);
5941 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5943 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
5944 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5945 #endif /* CONFIG_SPARSEMEM */
5949 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5950 * @page: The page within the block of interest
5951 * @start_bitidx: The first bit of interest to retrieve
5952 * @end_bitidx: The last bit of interest
5953 * returns pageblock_bits flags
5955 unsigned long get_pageblock_flags_group(struct page
*page
,
5956 int start_bitidx
, int end_bitidx
)
5959 unsigned long *bitmap
;
5960 unsigned long pfn
, bitidx
;
5961 unsigned long flags
= 0;
5962 unsigned long value
= 1;
5964 zone
= page_zone(page
);
5965 pfn
= page_to_pfn(page
);
5966 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5967 bitidx
= pfn_to_bitidx(zone
, pfn
);
5969 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5970 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5977 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5978 * @page: The page within the block of interest
5979 * @start_bitidx: The first bit of interest
5980 * @end_bitidx: The last bit of interest
5981 * @flags: The flags to set
5983 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5984 int start_bitidx
, int end_bitidx
)
5987 unsigned long *bitmap
;
5988 unsigned long pfn
, bitidx
;
5989 unsigned long value
= 1;
5991 zone
= page_zone(page
);
5992 pfn
= page_to_pfn(page
);
5993 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5994 bitidx
= pfn_to_bitidx(zone
, pfn
);
5995 VM_BUG_ON(!zone_spans_pfn(zone
, pfn
));
5997 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5999 __set_bit(bitidx
+ start_bitidx
, bitmap
);
6001 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
6005 * This function checks whether pageblock includes unmovable pages or not.
6006 * If @count is not zero, it is okay to include less @count unmovable pages
6008 * PageLRU check without isolation or lru_lock could race so that
6009 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6010 * expect this function should be exact.
6012 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6013 bool skip_hwpoisoned_pages
)
6015 unsigned long pfn
, iter
, found
;
6019 * For avoiding noise data, lru_add_drain_all() should be called
6020 * If ZONE_MOVABLE, the zone never contains unmovable pages
6022 if (zone_idx(zone
) == ZONE_MOVABLE
)
6024 mt
= get_pageblock_migratetype(page
);
6025 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6028 pfn
= page_to_pfn(page
);
6029 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6030 unsigned long check
= pfn
+ iter
;
6032 if (!pfn_valid_within(check
))
6035 page
= pfn_to_page(check
);
6038 * Hugepages are not in LRU lists, but they're movable.
6039 * We need not scan over tail pages bacause we don't
6040 * handle each tail page individually in migration.
6042 if (PageHuge(page
)) {
6043 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6048 * We can't use page_count without pin a page
6049 * because another CPU can free compound page.
6050 * This check already skips compound tails of THP
6051 * because their page->_count is zero at all time.
6053 if (!atomic_read(&page
->_count
)) {
6054 if (PageBuddy(page
))
6055 iter
+= (1 << page_order(page
)) - 1;
6060 * The HWPoisoned page may be not in buddy system, and
6061 * page_count() is not 0.
6063 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6069 * If there are RECLAIMABLE pages, we need to check it.
6070 * But now, memory offline itself doesn't call shrink_slab()
6071 * and it still to be fixed.
6074 * If the page is not RAM, page_count()should be 0.
6075 * we don't need more check. This is an _used_ not-movable page.
6077 * The problematic thing here is PG_reserved pages. PG_reserved
6078 * is set to both of a memory hole page and a _used_ kernel
6087 bool is_pageblock_removable_nolock(struct page
*page
)
6093 * We have to be careful here because we are iterating over memory
6094 * sections which are not zone aware so we might end up outside of
6095 * the zone but still within the section.
6096 * We have to take care about the node as well. If the node is offline
6097 * its NODE_DATA will be NULL - see page_zone.
6099 if (!node_online(page_to_nid(page
)))
6102 zone
= page_zone(page
);
6103 pfn
= page_to_pfn(page
);
6104 if (!zone_spans_pfn(zone
, pfn
))
6107 return !has_unmovable_pages(zone
, page
, 0, true);
6112 static unsigned long pfn_max_align_down(unsigned long pfn
)
6114 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6115 pageblock_nr_pages
) - 1);
6118 static unsigned long pfn_max_align_up(unsigned long pfn
)
6120 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6121 pageblock_nr_pages
));
6124 /* [start, end) must belong to a single zone. */
6125 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6126 unsigned long start
, unsigned long end
)
6128 /* This function is based on compact_zone() from compaction.c. */
6129 unsigned long nr_reclaimed
;
6130 unsigned long pfn
= start
;
6131 unsigned int tries
= 0;
6136 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6137 if (fatal_signal_pending(current
)) {
6142 if (list_empty(&cc
->migratepages
)) {
6143 cc
->nr_migratepages
= 0;
6144 pfn
= isolate_migratepages_range(cc
->zone
, cc
,
6151 } else if (++tries
== 5) {
6152 ret
= ret
< 0 ? ret
: -EBUSY
;
6156 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6158 cc
->nr_migratepages
-= nr_reclaimed
;
6160 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6161 0, MIGRATE_SYNC
, MR_CMA
);
6164 putback_movable_pages(&cc
->migratepages
);
6171 * alloc_contig_range() -- tries to allocate given range of pages
6172 * @start: start PFN to allocate
6173 * @end: one-past-the-last PFN to allocate
6174 * @migratetype: migratetype of the underlaying pageblocks (either
6175 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6176 * in range must have the same migratetype and it must
6177 * be either of the two.
6179 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6180 * aligned, however it's the caller's responsibility to guarantee that
6181 * we are the only thread that changes migrate type of pageblocks the
6184 * The PFN range must belong to a single zone.
6186 * Returns zero on success or negative error code. On success all
6187 * pages which PFN is in [start, end) are allocated for the caller and
6188 * need to be freed with free_contig_range().
6190 int alloc_contig_range(unsigned long start
, unsigned long end
,
6191 unsigned migratetype
)
6193 unsigned long outer_start
, outer_end
;
6196 struct compact_control cc
= {
6197 .nr_migratepages
= 0,
6199 .zone
= page_zone(pfn_to_page(start
)),
6201 .ignore_skip_hint
= true,
6203 INIT_LIST_HEAD(&cc
.migratepages
);
6206 * What we do here is we mark all pageblocks in range as
6207 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6208 * have different sizes, and due to the way page allocator
6209 * work, we align the range to biggest of the two pages so
6210 * that page allocator won't try to merge buddies from
6211 * different pageblocks and change MIGRATE_ISOLATE to some
6212 * other migration type.
6214 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6215 * migrate the pages from an unaligned range (ie. pages that
6216 * we are interested in). This will put all the pages in
6217 * range back to page allocator as MIGRATE_ISOLATE.
6219 * When this is done, we take the pages in range from page
6220 * allocator removing them from the buddy system. This way
6221 * page allocator will never consider using them.
6223 * This lets us mark the pageblocks back as
6224 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6225 * aligned range but not in the unaligned, original range are
6226 * put back to page allocator so that buddy can use them.
6229 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6230 pfn_max_align_up(end
), migratetype
,
6235 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6240 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6241 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6242 * more, all pages in [start, end) are free in page allocator.
6243 * What we are going to do is to allocate all pages from
6244 * [start, end) (that is remove them from page allocator).
6246 * The only problem is that pages at the beginning and at the
6247 * end of interesting range may be not aligned with pages that
6248 * page allocator holds, ie. they can be part of higher order
6249 * pages. Because of this, we reserve the bigger range and
6250 * once this is done free the pages we are not interested in.
6252 * We don't have to hold zone->lock here because the pages are
6253 * isolated thus they won't get removed from buddy.
6256 lru_add_drain_all();
6260 outer_start
= start
;
6261 while (!PageBuddy(pfn_to_page(outer_start
))) {
6262 if (++order
>= MAX_ORDER
) {
6266 outer_start
&= ~0UL << order
;
6269 /* Make sure the range is really isolated. */
6270 if (test_pages_isolated(outer_start
, end
, false)) {
6271 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6278 /* Grab isolated pages from freelists. */
6279 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6285 /* Free head and tail (if any) */
6286 if (start
!= outer_start
)
6287 free_contig_range(outer_start
, start
- outer_start
);
6288 if (end
!= outer_end
)
6289 free_contig_range(end
, outer_end
- end
);
6292 undo_isolate_page_range(pfn_max_align_down(start
),
6293 pfn_max_align_up(end
), migratetype
);
6297 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6299 unsigned int count
= 0;
6301 for (; nr_pages
--; pfn
++) {
6302 struct page
*page
= pfn_to_page(pfn
);
6304 count
+= page_count(page
) != 1;
6307 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6311 #ifdef CONFIG_MEMORY_HOTPLUG
6313 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6314 * page high values need to be recalulated.
6316 void __meminit
zone_pcp_update(struct zone
*zone
)
6319 mutex_lock(&pcp_batch_high_lock
);
6320 for_each_possible_cpu(cpu
)
6321 pageset_set_high_and_batch(zone
,
6322 per_cpu_ptr(zone
->pageset
, cpu
));
6323 mutex_unlock(&pcp_batch_high_lock
);
6327 void zone_pcp_reset(struct zone
*zone
)
6329 unsigned long flags
;
6331 struct per_cpu_pageset
*pset
;
6333 /* avoid races with drain_pages() */
6334 local_irq_save(flags
);
6335 if (zone
->pageset
!= &boot_pageset
) {
6336 for_each_online_cpu(cpu
) {
6337 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6338 drain_zonestat(zone
, pset
);
6340 free_percpu(zone
->pageset
);
6341 zone
->pageset
= &boot_pageset
;
6343 local_irq_restore(flags
);
6346 #ifdef CONFIG_MEMORY_HOTREMOVE
6348 * All pages in the range must be isolated before calling this.
6351 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6357 unsigned long flags
;
6358 /* find the first valid pfn */
6359 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6364 zone
= page_zone(pfn_to_page(pfn
));
6365 spin_lock_irqsave(&zone
->lock
, flags
);
6367 while (pfn
< end_pfn
) {
6368 if (!pfn_valid(pfn
)) {
6372 page
= pfn_to_page(pfn
);
6374 * The HWPoisoned page may be not in buddy system, and
6375 * page_count() is not 0.
6377 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6379 SetPageReserved(page
);
6383 BUG_ON(page_count(page
));
6384 BUG_ON(!PageBuddy(page
));
6385 order
= page_order(page
);
6386 #ifdef CONFIG_DEBUG_VM
6387 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6388 pfn
, 1 << order
, end_pfn
);
6390 list_del(&page
->lru
);
6391 rmv_page_order(page
);
6392 zone
->free_area
[order
].nr_free
--;
6393 for (i
= 0; i
< (1 << order
); i
++)
6394 SetPageReserved((page
+i
));
6395 pfn
+= (1 << order
);
6397 spin_unlock_irqrestore(&zone
->lock
, flags
);
6401 #ifdef CONFIG_MEMORY_FAILURE
6402 bool is_free_buddy_page(struct page
*page
)
6404 struct zone
*zone
= page_zone(page
);
6405 unsigned long pfn
= page_to_pfn(page
);
6406 unsigned long flags
;
6409 spin_lock_irqsave(&zone
->lock
, flags
);
6410 for (order
= 0; order
< MAX_ORDER
; order
++) {
6411 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6413 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6416 spin_unlock_irqrestore(&zone
->lock
, flags
);
6418 return order
< MAX_ORDER
;
6422 static const struct trace_print_flags pageflag_names
[] = {
6423 {1UL << PG_locked
, "locked" },
6424 {1UL << PG_error
, "error" },
6425 {1UL << PG_referenced
, "referenced" },
6426 {1UL << PG_uptodate
, "uptodate" },
6427 {1UL << PG_dirty
, "dirty" },
6428 {1UL << PG_lru
, "lru" },
6429 {1UL << PG_active
, "active" },
6430 {1UL << PG_slab
, "slab" },
6431 {1UL << PG_owner_priv_1
, "owner_priv_1" },
6432 {1UL << PG_arch_1
, "arch_1" },
6433 {1UL << PG_reserved
, "reserved" },
6434 {1UL << PG_private
, "private" },
6435 {1UL << PG_private_2
, "private_2" },
6436 {1UL << PG_writeback
, "writeback" },
6437 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6438 {1UL << PG_head
, "head" },
6439 {1UL << PG_tail
, "tail" },
6441 {1UL << PG_compound
, "compound" },
6443 {1UL << PG_swapcache
, "swapcache" },
6444 {1UL << PG_mappedtodisk
, "mappedtodisk" },
6445 {1UL << PG_reclaim
, "reclaim" },
6446 {1UL << PG_swapbacked
, "swapbacked" },
6447 {1UL << PG_unevictable
, "unevictable" },
6449 {1UL << PG_mlocked
, "mlocked" },
6451 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6452 {1UL << PG_uncached
, "uncached" },
6454 #ifdef CONFIG_MEMORY_FAILURE
6455 {1UL << PG_hwpoison
, "hwpoison" },
6457 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6458 {1UL << PG_compound_lock
, "compound_lock" },
6462 static void dump_page_flags(unsigned long flags
)
6464 const char *delim
= "";
6468 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names
) != __NR_PAGEFLAGS
);
6470 printk(KERN_ALERT
"page flags: %#lx(", flags
);
6472 /* remove zone id */
6473 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
6475 for (i
= 0; i
< ARRAY_SIZE(pageflag_names
) && flags
; i
++) {
6477 mask
= pageflag_names
[i
].mask
;
6478 if ((flags
& mask
) != mask
)
6482 printk("%s%s", delim
, pageflag_names
[i
].name
);
6486 /* check for left over flags */
6488 printk("%s%#lx", delim
, flags
);
6493 void dump_page(struct page
*page
)
6496 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6497 page
, atomic_read(&page
->_count
), page_mapcount(page
),
6498 page
->mapping
, page
->index
);
6499 dump_page_flags(page
->flags
);
6500 mem_cgroup_print_bad_page(page
);