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 compaction_defer_reset(preferred_zone
, order
, true);
2239 count_vm_event(COMPACTSUCCESS
);
2244 * It's bad if compaction run occurs and fails.
2245 * The most likely reason is that pages exist,
2246 * but not enough to satisfy watermarks.
2248 count_vm_event(COMPACTFAIL
);
2251 * As async compaction considers a subset of pageblocks, only
2252 * defer if the failure was a sync compaction failure.
2255 defer_compaction(preferred_zone
, order
);
2263 static inline struct page
*
2264 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2265 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2266 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2267 int migratetype
, bool sync_migration
,
2268 bool *contended_compaction
, bool *deferred_compaction
,
2269 unsigned long *did_some_progress
)
2273 #endif /* CONFIG_COMPACTION */
2275 /* Perform direct synchronous page reclaim */
2277 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
, struct zonelist
*zonelist
,
2278 nodemask_t
*nodemask
)
2280 struct reclaim_state reclaim_state
;
2285 /* We now go into synchronous reclaim */
2286 cpuset_memory_pressure_bump();
2287 current
->flags
|= PF_MEMALLOC
;
2288 lockdep_set_current_reclaim_state(gfp_mask
);
2289 reclaim_state
.reclaimed_slab
= 0;
2290 current
->reclaim_state
= &reclaim_state
;
2292 progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
2294 current
->reclaim_state
= NULL
;
2295 lockdep_clear_current_reclaim_state();
2296 current
->flags
&= ~PF_MEMALLOC
;
2303 /* The really slow allocator path where we enter direct reclaim */
2304 static inline struct page
*
2305 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2306 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2307 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2308 int migratetype
, unsigned long *did_some_progress
)
2310 struct page
*page
= NULL
;
2311 bool drained
= false;
2313 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
2315 if (unlikely(!(*did_some_progress
)))
2318 /* After successful reclaim, reconsider all zones for allocation */
2319 if (IS_ENABLED(CONFIG_NUMA
))
2320 zlc_clear_zones_full(zonelist
);
2323 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2324 zonelist
, high_zoneidx
,
2325 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2326 preferred_zone
, migratetype
);
2329 * If an allocation failed after direct reclaim, it could be because
2330 * pages are pinned on the per-cpu lists. Drain them and try again
2332 if (!page
&& !drained
) {
2342 * This is called in the allocator slow-path if the allocation request is of
2343 * sufficient urgency to ignore watermarks and take other desperate measures
2345 static inline struct page
*
2346 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2347 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2348 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2354 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2355 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
2356 preferred_zone
, migratetype
);
2358 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2359 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2360 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2365 static void prepare_slowpath(gfp_t gfp_mask
, unsigned int order
,
2366 struct zonelist
*zonelist
,
2367 enum zone_type high_zoneidx
,
2368 struct zone
*preferred_zone
)
2373 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
2374 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2375 wakeup_kswapd(zone
, order
, zone_idx(preferred_zone
));
2377 * Only reset the batches of zones that were actually
2378 * considered in the fast path, we don't want to
2379 * thrash fairness information for zones that are not
2380 * actually part of this zonelist's round-robin cycle.
2382 if (!zone_local(preferred_zone
, zone
))
2384 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2385 high_wmark_pages(zone
) -
2386 low_wmark_pages(zone
) -
2387 zone_page_state(zone
, NR_ALLOC_BATCH
));
2392 gfp_to_alloc_flags(gfp_t gfp_mask
)
2394 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2395 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2397 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2398 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2401 * The caller may dip into page reserves a bit more if the caller
2402 * cannot run direct reclaim, or if the caller has realtime scheduling
2403 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2404 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2406 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2410 * Not worth trying to allocate harder for
2411 * __GFP_NOMEMALLOC even if it can't schedule.
2413 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2414 alloc_flags
|= ALLOC_HARDER
;
2416 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2417 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2419 alloc_flags
&= ~ALLOC_CPUSET
;
2420 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2421 alloc_flags
|= ALLOC_HARDER
;
2423 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2424 if (gfp_mask
& __GFP_MEMALLOC
)
2425 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2426 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2427 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2428 else if (!in_interrupt() &&
2429 ((current
->flags
& PF_MEMALLOC
) ||
2430 unlikely(test_thread_flag(TIF_MEMDIE
))))
2431 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2434 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2435 alloc_flags
|= ALLOC_CMA
;
2440 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2442 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2445 static inline struct page
*
2446 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2447 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2448 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2451 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2452 struct page
*page
= NULL
;
2454 unsigned long pages_reclaimed
= 0;
2455 unsigned long did_some_progress
;
2456 bool sync_migration
= false;
2457 bool deferred_compaction
= false;
2458 bool contended_compaction
= false;
2461 * In the slowpath, we sanity check order to avoid ever trying to
2462 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2463 * be using allocators in order of preference for an area that is
2466 if (order
>= MAX_ORDER
) {
2467 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2472 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2473 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2474 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2475 * using a larger set of nodes after it has established that the
2476 * allowed per node queues are empty and that nodes are
2479 if (IS_ENABLED(CONFIG_NUMA
) &&
2480 (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2484 prepare_slowpath(gfp_mask
, order
, zonelist
,
2485 high_zoneidx
, preferred_zone
);
2488 * OK, we're below the kswapd watermark and have kicked background
2489 * reclaim. Now things get more complex, so set up alloc_flags according
2490 * to how we want to proceed.
2492 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2495 * Find the true preferred zone if the allocation is unconstrained by
2498 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2499 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2503 /* This is the last chance, in general, before the goto nopage. */
2504 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2505 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2506 preferred_zone
, migratetype
);
2510 /* Allocate without watermarks if the context allows */
2511 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2513 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2514 * the allocation is high priority and these type of
2515 * allocations are system rather than user orientated
2517 zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
2519 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2520 zonelist
, high_zoneidx
, nodemask
,
2521 preferred_zone
, migratetype
);
2527 /* Atomic allocations - we can't balance anything */
2530 * All existing users of the deprecated __GFP_NOFAIL are
2531 * blockable, so warn of any new users that actually allow this
2532 * type of allocation to fail.
2534 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
2538 /* Avoid recursion of direct reclaim */
2539 if (current
->flags
& PF_MEMALLOC
)
2542 /* Avoid allocations with no watermarks from looping endlessly */
2543 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2547 * Try direct compaction. The first pass is asynchronous. Subsequent
2548 * attempts after direct reclaim are synchronous
2550 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2551 zonelist
, high_zoneidx
,
2553 alloc_flags
, preferred_zone
,
2554 migratetype
, sync_migration
,
2555 &contended_compaction
,
2556 &deferred_compaction
,
2557 &did_some_progress
);
2560 sync_migration
= true;
2563 * If compaction is deferred for high-order allocations, it is because
2564 * sync compaction recently failed. In this is the case and the caller
2565 * requested a movable allocation that does not heavily disrupt the
2566 * system then fail the allocation instead of entering direct reclaim.
2568 if ((deferred_compaction
|| contended_compaction
) &&
2569 (gfp_mask
& __GFP_NO_KSWAPD
))
2572 /* Try direct reclaim and then allocating */
2573 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2574 zonelist
, high_zoneidx
,
2576 alloc_flags
, preferred_zone
,
2577 migratetype
, &did_some_progress
);
2582 * If we failed to make any progress reclaiming, then we are
2583 * running out of options and have to consider going OOM
2585 if (!did_some_progress
) {
2586 if (oom_gfp_allowed(gfp_mask
)) {
2587 if (oom_killer_disabled
)
2589 /* Coredumps can quickly deplete all memory reserves */
2590 if ((current
->flags
& PF_DUMPCORE
) &&
2591 !(gfp_mask
& __GFP_NOFAIL
))
2593 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2594 zonelist
, high_zoneidx
,
2595 nodemask
, preferred_zone
,
2600 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2602 * The oom killer is not called for high-order
2603 * allocations that may fail, so if no progress
2604 * is being made, there are no other options and
2605 * retrying is unlikely to help.
2607 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2610 * The oom killer is not called for lowmem
2611 * allocations to prevent needlessly killing
2614 if (high_zoneidx
< ZONE_NORMAL
)
2622 /* Check if we should retry the allocation */
2623 pages_reclaimed
+= did_some_progress
;
2624 if (should_alloc_retry(gfp_mask
, order
, did_some_progress
,
2626 /* Wait for some write requests to complete then retry */
2627 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2631 * High-order allocations do not necessarily loop after
2632 * direct reclaim and reclaim/compaction depends on compaction
2633 * being called after reclaim so call directly if necessary
2635 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2636 zonelist
, high_zoneidx
,
2638 alloc_flags
, preferred_zone
,
2639 migratetype
, sync_migration
,
2640 &contended_compaction
,
2641 &deferred_compaction
,
2642 &did_some_progress
);
2648 warn_alloc_failed(gfp_mask
, order
, NULL
);
2651 if (kmemcheck_enabled
)
2652 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2658 * This is the 'heart' of the zoned buddy allocator.
2661 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2662 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2664 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2665 struct zone
*preferred_zone
;
2666 struct page
*page
= NULL
;
2667 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2668 unsigned int cpuset_mems_cookie
;
2669 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
;
2670 struct mem_cgroup
*memcg
= NULL
;
2672 gfp_mask
&= gfp_allowed_mask
;
2674 lockdep_trace_alloc(gfp_mask
);
2676 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2678 if (should_fail_alloc_page(gfp_mask
, order
))
2682 * Check the zones suitable for the gfp_mask contain at least one
2683 * valid zone. It's possible to have an empty zonelist as a result
2684 * of GFP_THISNODE and a memoryless node
2686 if (unlikely(!zonelist
->_zonerefs
->zone
))
2690 * Will only have any effect when __GFP_KMEMCG is set. This is
2691 * verified in the (always inline) callee
2693 if (!memcg_kmem_newpage_charge(gfp_mask
, &memcg
, order
))
2697 cpuset_mems_cookie
= get_mems_allowed();
2699 /* The preferred zone is used for statistics later */
2700 first_zones_zonelist(zonelist
, high_zoneidx
,
2701 nodemask
? : &cpuset_current_mems_allowed
,
2703 if (!preferred_zone
)
2707 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2708 alloc_flags
|= ALLOC_CMA
;
2710 /* First allocation attempt */
2711 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2712 zonelist
, high_zoneidx
, alloc_flags
,
2713 preferred_zone
, migratetype
);
2714 if (unlikely(!page
)) {
2716 * Runtime PM, block IO and its error handling path
2717 * can deadlock because I/O on the device might not
2720 gfp_mask
= memalloc_noio_flags(gfp_mask
);
2721 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2722 zonelist
, high_zoneidx
, nodemask
,
2723 preferred_zone
, migratetype
);
2726 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2730 * When updating a task's mems_allowed, it is possible to race with
2731 * parallel threads in such a way that an allocation can fail while
2732 * the mask is being updated. If a page allocation is about to fail,
2733 * check if the cpuset changed during allocation and if so, retry.
2735 if (unlikely(!put_mems_allowed(cpuset_mems_cookie
) && !page
))
2738 memcg_kmem_commit_charge(page
, memcg
, order
);
2742 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2745 * Common helper functions.
2747 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2752 * __get_free_pages() returns a 32-bit address, which cannot represent
2755 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2757 page
= alloc_pages(gfp_mask
, order
);
2760 return (unsigned long) page_address(page
);
2762 EXPORT_SYMBOL(__get_free_pages
);
2764 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2766 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2768 EXPORT_SYMBOL(get_zeroed_page
);
2770 void __free_pages(struct page
*page
, unsigned int order
)
2772 if (put_page_testzero(page
)) {
2774 free_hot_cold_page(page
, 0);
2776 __free_pages_ok(page
, order
);
2780 EXPORT_SYMBOL(__free_pages
);
2782 void free_pages(unsigned long addr
, unsigned int order
)
2785 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2786 __free_pages(virt_to_page((void *)addr
), order
);
2790 EXPORT_SYMBOL(free_pages
);
2793 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2794 * pages allocated with __GFP_KMEMCG.
2796 * Those pages are accounted to a particular memcg, embedded in the
2797 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2798 * for that information only to find out that it is NULL for users who have no
2799 * interest in that whatsoever, we provide these functions.
2801 * The caller knows better which flags it relies on.
2803 void __free_memcg_kmem_pages(struct page
*page
, unsigned int order
)
2805 memcg_kmem_uncharge_pages(page
, order
);
2806 __free_pages(page
, order
);
2809 void free_memcg_kmem_pages(unsigned long addr
, unsigned int order
)
2812 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2813 __free_memcg_kmem_pages(virt_to_page((void *)addr
), order
);
2817 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2820 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2821 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2823 split_page(virt_to_page((void *)addr
), order
);
2824 while (used
< alloc_end
) {
2829 return (void *)addr
;
2833 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2834 * @size: the number of bytes to allocate
2835 * @gfp_mask: GFP flags for the allocation
2837 * This function is similar to alloc_pages(), except that it allocates the
2838 * minimum number of pages to satisfy the request. alloc_pages() can only
2839 * allocate memory in power-of-two pages.
2841 * This function is also limited by MAX_ORDER.
2843 * Memory allocated by this function must be released by free_pages_exact().
2845 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2847 unsigned int order
= get_order(size
);
2850 addr
= __get_free_pages(gfp_mask
, order
);
2851 return make_alloc_exact(addr
, order
, size
);
2853 EXPORT_SYMBOL(alloc_pages_exact
);
2856 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2858 * @nid: the preferred node ID where memory should be allocated
2859 * @size: the number of bytes to allocate
2860 * @gfp_mask: GFP flags for the allocation
2862 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2864 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2867 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2869 unsigned order
= get_order(size
);
2870 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2873 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2875 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2878 * free_pages_exact - release memory allocated via alloc_pages_exact()
2879 * @virt: the value returned by alloc_pages_exact.
2880 * @size: size of allocation, same value as passed to alloc_pages_exact().
2882 * Release the memory allocated by a previous call to alloc_pages_exact.
2884 void free_pages_exact(void *virt
, size_t size
)
2886 unsigned long addr
= (unsigned long)virt
;
2887 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2889 while (addr
< end
) {
2894 EXPORT_SYMBOL(free_pages_exact
);
2897 * nr_free_zone_pages - count number of pages beyond high watermark
2898 * @offset: The zone index of the highest zone
2900 * nr_free_zone_pages() counts the number of counts pages which are beyond the
2901 * high watermark within all zones at or below a given zone index. For each
2902 * zone, the number of pages is calculated as:
2903 * managed_pages - high_pages
2905 static unsigned long nr_free_zone_pages(int offset
)
2910 /* Just pick one node, since fallback list is circular */
2911 unsigned long sum
= 0;
2913 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2915 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2916 unsigned long size
= zone
->managed_pages
;
2917 unsigned long high
= high_wmark_pages(zone
);
2926 * nr_free_buffer_pages - count number of pages beyond high watermark
2928 * nr_free_buffer_pages() counts the number of pages which are beyond the high
2929 * watermark within ZONE_DMA and ZONE_NORMAL.
2931 unsigned long nr_free_buffer_pages(void)
2933 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2935 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2938 * nr_free_pagecache_pages - count number of pages beyond high watermark
2940 * nr_free_pagecache_pages() counts the number of pages which are beyond the
2941 * high watermark within all zones.
2943 unsigned long nr_free_pagecache_pages(void)
2945 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2948 static inline void show_node(struct zone
*zone
)
2950 if (IS_ENABLED(CONFIG_NUMA
))
2951 printk("Node %d ", zone_to_nid(zone
));
2954 void si_meminfo(struct sysinfo
*val
)
2956 val
->totalram
= totalram_pages
;
2958 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2959 val
->bufferram
= nr_blockdev_pages();
2960 val
->totalhigh
= totalhigh_pages
;
2961 val
->freehigh
= nr_free_highpages();
2962 val
->mem_unit
= PAGE_SIZE
;
2965 EXPORT_SYMBOL(si_meminfo
);
2968 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2970 int zone_type
; /* needs to be signed */
2971 unsigned long managed_pages
= 0;
2972 pg_data_t
*pgdat
= NODE_DATA(nid
);
2974 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
2975 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
2976 val
->totalram
= managed_pages
;
2977 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2978 #ifdef CONFIG_HIGHMEM
2979 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
2980 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2986 val
->mem_unit
= PAGE_SIZE
;
2991 * Determine whether the node should be displayed or not, depending on whether
2992 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2994 bool skip_free_areas_node(unsigned int flags
, int nid
)
2997 unsigned int cpuset_mems_cookie
;
2999 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3003 cpuset_mems_cookie
= get_mems_allowed();
3004 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3005 } while (!put_mems_allowed(cpuset_mems_cookie
));
3010 #define K(x) ((x) << (PAGE_SHIFT-10))
3012 static void show_migration_types(unsigned char type
)
3014 static const char types
[MIGRATE_TYPES
] = {
3015 [MIGRATE_UNMOVABLE
] = 'U',
3016 [MIGRATE_RECLAIMABLE
] = 'E',
3017 [MIGRATE_MOVABLE
] = 'M',
3018 [MIGRATE_RESERVE
] = 'R',
3020 [MIGRATE_CMA
] = 'C',
3022 #ifdef CONFIG_MEMORY_ISOLATION
3023 [MIGRATE_ISOLATE
] = 'I',
3026 char tmp
[MIGRATE_TYPES
+ 1];
3030 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3031 if (type
& (1 << i
))
3036 printk("(%s) ", tmp
);
3040 * Show free area list (used inside shift_scroll-lock stuff)
3041 * We also calculate the percentage fragmentation. We do this by counting the
3042 * memory on each free list with the exception of the first item on the list.
3043 * Suppresses nodes that are not allowed by current's cpuset if
3044 * SHOW_MEM_FILTER_NODES is passed.
3046 void show_free_areas(unsigned int filter
)
3051 for_each_populated_zone(zone
) {
3052 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3055 printk("%s per-cpu:\n", zone
->name
);
3057 for_each_online_cpu(cpu
) {
3058 struct per_cpu_pageset
*pageset
;
3060 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
3062 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3063 cpu
, pageset
->pcp
.high
,
3064 pageset
->pcp
.batch
, pageset
->pcp
.count
);
3068 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3069 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3071 " dirty:%lu writeback:%lu unstable:%lu\n"
3072 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3073 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3075 global_page_state(NR_ACTIVE_ANON
),
3076 global_page_state(NR_INACTIVE_ANON
),
3077 global_page_state(NR_ISOLATED_ANON
),
3078 global_page_state(NR_ACTIVE_FILE
),
3079 global_page_state(NR_INACTIVE_FILE
),
3080 global_page_state(NR_ISOLATED_FILE
),
3081 global_page_state(NR_UNEVICTABLE
),
3082 global_page_state(NR_FILE_DIRTY
),
3083 global_page_state(NR_WRITEBACK
),
3084 global_page_state(NR_UNSTABLE_NFS
),
3085 global_page_state(NR_FREE_PAGES
),
3086 global_page_state(NR_SLAB_RECLAIMABLE
),
3087 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3088 global_page_state(NR_FILE_MAPPED
),
3089 global_page_state(NR_SHMEM
),
3090 global_page_state(NR_PAGETABLE
),
3091 global_page_state(NR_BOUNCE
),
3092 global_page_state(NR_FREE_CMA_PAGES
));
3094 for_each_populated_zone(zone
) {
3097 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3105 " active_anon:%lukB"
3106 " inactive_anon:%lukB"
3107 " active_file:%lukB"
3108 " inactive_file:%lukB"
3109 " unevictable:%lukB"
3110 " isolated(anon):%lukB"
3111 " isolated(file):%lukB"
3119 " slab_reclaimable:%lukB"
3120 " slab_unreclaimable:%lukB"
3121 " kernel_stack:%lukB"
3126 " writeback_tmp:%lukB"
3127 " pages_scanned:%lu"
3128 " all_unreclaimable? %s"
3131 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3132 K(min_wmark_pages(zone
)),
3133 K(low_wmark_pages(zone
)),
3134 K(high_wmark_pages(zone
)),
3135 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3136 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3137 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3138 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3139 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3140 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3141 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3142 K(zone
->present_pages
),
3143 K(zone
->managed_pages
),
3144 K(zone_page_state(zone
, NR_MLOCK
)),
3145 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3146 K(zone_page_state(zone
, NR_WRITEBACK
)),
3147 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3148 K(zone_page_state(zone
, NR_SHMEM
)),
3149 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3150 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3151 zone_page_state(zone
, NR_KERNEL_STACK
) *
3153 K(zone_page_state(zone
, NR_PAGETABLE
)),
3154 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3155 K(zone_page_state(zone
, NR_BOUNCE
)),
3156 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3157 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3158 zone
->pages_scanned
,
3159 (!zone_reclaimable(zone
) ? "yes" : "no")
3161 printk("lowmem_reserve[]:");
3162 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3163 printk(" %lu", zone
->lowmem_reserve
[i
]);
3167 for_each_populated_zone(zone
) {
3168 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
3169 unsigned char types
[MAX_ORDER
];
3171 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3174 printk("%s: ", zone
->name
);
3176 spin_lock_irqsave(&zone
->lock
, flags
);
3177 for (order
= 0; order
< MAX_ORDER
; order
++) {
3178 struct free_area
*area
= &zone
->free_area
[order
];
3181 nr
[order
] = area
->nr_free
;
3182 total
+= nr
[order
] << order
;
3185 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3186 if (!list_empty(&area
->free_list
[type
]))
3187 types
[order
] |= 1 << type
;
3190 spin_unlock_irqrestore(&zone
->lock
, flags
);
3191 for (order
= 0; order
< MAX_ORDER
; order
++) {
3192 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3194 show_migration_types(types
[order
]);
3196 printk("= %lukB\n", K(total
));
3199 hugetlb_show_meminfo();
3201 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3203 show_swap_cache_info();
3206 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3208 zoneref
->zone
= zone
;
3209 zoneref
->zone_idx
= zone_idx(zone
);
3213 * Builds allocation fallback zone lists.
3215 * Add all populated zones of a node to the zonelist.
3217 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3221 enum zone_type zone_type
= MAX_NR_ZONES
;
3225 zone
= pgdat
->node_zones
+ zone_type
;
3226 if (populated_zone(zone
)) {
3227 zoneref_set_zone(zone
,
3228 &zonelist
->_zonerefs
[nr_zones
++]);
3229 check_highest_zone(zone_type
);
3231 } while (zone_type
);
3239 * 0 = automatic detection of better ordering.
3240 * 1 = order by ([node] distance, -zonetype)
3241 * 2 = order by (-zonetype, [node] distance)
3243 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3244 * the same zonelist. So only NUMA can configure this param.
3246 #define ZONELIST_ORDER_DEFAULT 0
3247 #define ZONELIST_ORDER_NODE 1
3248 #define ZONELIST_ORDER_ZONE 2
3250 /* zonelist order in the kernel.
3251 * set_zonelist_order() will set this to NODE or ZONE.
3253 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3254 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3258 /* The value user specified ....changed by config */
3259 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3260 /* string for sysctl */
3261 #define NUMA_ZONELIST_ORDER_LEN 16
3262 char numa_zonelist_order
[16] = "default";
3265 * interface for configure zonelist ordering.
3266 * command line option "numa_zonelist_order"
3267 * = "[dD]efault - default, automatic configuration.
3268 * = "[nN]ode - order by node locality, then by zone within node
3269 * = "[zZ]one - order by zone, then by locality within zone
3272 static int __parse_numa_zonelist_order(char *s
)
3274 if (*s
== 'd' || *s
== 'D') {
3275 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3276 } else if (*s
== 'n' || *s
== 'N') {
3277 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3278 } else if (*s
== 'z' || *s
== 'Z') {
3279 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3282 "Ignoring invalid numa_zonelist_order value: "
3289 static __init
int setup_numa_zonelist_order(char *s
)
3296 ret
= __parse_numa_zonelist_order(s
);
3298 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3302 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3305 * sysctl handler for numa_zonelist_order
3307 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
3308 void __user
*buffer
, size_t *length
,
3311 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3313 static DEFINE_MUTEX(zl_order_mutex
);
3315 mutex_lock(&zl_order_mutex
);
3317 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
3321 strcpy(saved_string
, (char *)table
->data
);
3323 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3327 int oldval
= user_zonelist_order
;
3329 ret
= __parse_numa_zonelist_order((char *)table
->data
);
3332 * bogus value. restore saved string
3334 strncpy((char *)table
->data
, saved_string
,
3335 NUMA_ZONELIST_ORDER_LEN
);
3336 user_zonelist_order
= oldval
;
3337 } else if (oldval
!= user_zonelist_order
) {
3338 mutex_lock(&zonelists_mutex
);
3339 build_all_zonelists(NULL
, NULL
);
3340 mutex_unlock(&zonelists_mutex
);
3344 mutex_unlock(&zl_order_mutex
);
3349 #define MAX_NODE_LOAD (nr_online_nodes)
3350 static int node_load
[MAX_NUMNODES
];
3353 * find_next_best_node - find the next node that should appear in a given node's fallback list
3354 * @node: node whose fallback list we're appending
3355 * @used_node_mask: nodemask_t of already used nodes
3357 * We use a number of factors to determine which is the next node that should
3358 * appear on a given node's fallback list. The node should not have appeared
3359 * already in @node's fallback list, and it should be the next closest node
3360 * according to the distance array (which contains arbitrary distance values
3361 * from each node to each node in the system), and should also prefer nodes
3362 * with no CPUs, since presumably they'll have very little allocation pressure
3363 * on them otherwise.
3364 * It returns -1 if no node is found.
3366 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3369 int min_val
= INT_MAX
;
3370 int best_node
= NUMA_NO_NODE
;
3371 const struct cpumask
*tmp
= cpumask_of_node(0);
3373 /* Use the local node if we haven't already */
3374 if (!node_isset(node
, *used_node_mask
)) {
3375 node_set(node
, *used_node_mask
);
3379 for_each_node_state(n
, N_MEMORY
) {
3381 /* Don't want a node to appear more than once */
3382 if (node_isset(n
, *used_node_mask
))
3385 /* Use the distance array to find the distance */
3386 val
= node_distance(node
, n
);
3388 /* Penalize nodes under us ("prefer the next node") */
3391 /* Give preference to headless and unused nodes */
3392 tmp
= cpumask_of_node(n
);
3393 if (!cpumask_empty(tmp
))
3394 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
3396 /* Slight preference for less loaded node */
3397 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
3398 val
+= node_load
[n
];
3400 if (val
< min_val
) {
3407 node_set(best_node
, *used_node_mask
);
3414 * Build zonelists ordered by node and zones within node.
3415 * This results in maximum locality--normal zone overflows into local
3416 * DMA zone, if any--but risks exhausting DMA zone.
3418 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
3421 struct zonelist
*zonelist
;
3423 zonelist
= &pgdat
->node_zonelists
[0];
3424 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
3426 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
3427 zonelist
->_zonerefs
[j
].zone
= NULL
;
3428 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3432 * Build gfp_thisnode zonelists
3434 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
3437 struct zonelist
*zonelist
;
3439 zonelist
= &pgdat
->node_zonelists
[1];
3440 j
= build_zonelists_node(pgdat
, zonelist
, 0);
3441 zonelist
->_zonerefs
[j
].zone
= NULL
;
3442 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3446 * Build zonelists ordered by zone and nodes within zones.
3447 * This results in conserving DMA zone[s] until all Normal memory is
3448 * exhausted, but results in overflowing to remote node while memory
3449 * may still exist in local DMA zone.
3451 static int node_order
[MAX_NUMNODES
];
3453 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
3456 int zone_type
; /* needs to be signed */
3458 struct zonelist
*zonelist
;
3460 zonelist
= &pgdat
->node_zonelists
[0];
3462 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
3463 for (j
= 0; j
< nr_nodes
; j
++) {
3464 node
= node_order
[j
];
3465 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
3466 if (populated_zone(z
)) {
3468 &zonelist
->_zonerefs
[pos
++]);
3469 check_highest_zone(zone_type
);
3473 zonelist
->_zonerefs
[pos
].zone
= NULL
;
3474 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
3477 static int default_zonelist_order(void)
3480 unsigned long low_kmem_size
, total_size
;
3484 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3485 * If they are really small and used heavily, the system can fall
3486 * into OOM very easily.
3487 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3489 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3492 for_each_online_node(nid
) {
3493 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3494 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3495 if (populated_zone(z
)) {
3496 if (zone_type
< ZONE_NORMAL
)
3497 low_kmem_size
+= z
->managed_pages
;
3498 total_size
+= z
->managed_pages
;
3499 } else if (zone_type
== ZONE_NORMAL
) {
3501 * If any node has only lowmem, then node order
3502 * is preferred to allow kernel allocations
3503 * locally; otherwise, they can easily infringe
3504 * on other nodes when there is an abundance of
3505 * lowmem available to allocate from.
3507 return ZONELIST_ORDER_NODE
;
3511 if (!low_kmem_size
|| /* there are no DMA area. */
3512 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
3513 return ZONELIST_ORDER_NODE
;
3515 * look into each node's config.
3516 * If there is a node whose DMA/DMA32 memory is very big area on
3517 * local memory, NODE_ORDER may be suitable.
3519 average_size
= total_size
/
3520 (nodes_weight(node_states
[N_MEMORY
]) + 1);
3521 for_each_online_node(nid
) {
3524 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3525 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3526 if (populated_zone(z
)) {
3527 if (zone_type
< ZONE_NORMAL
)
3528 low_kmem_size
+= z
->present_pages
;
3529 total_size
+= z
->present_pages
;
3532 if (low_kmem_size
&&
3533 total_size
> average_size
&& /* ignore small node */
3534 low_kmem_size
> total_size
* 70/100)
3535 return ZONELIST_ORDER_NODE
;
3537 return ZONELIST_ORDER_ZONE
;
3540 static void set_zonelist_order(void)
3542 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
3543 current_zonelist_order
= default_zonelist_order();
3545 current_zonelist_order
= user_zonelist_order
;
3548 static void build_zonelists(pg_data_t
*pgdat
)
3552 nodemask_t used_mask
;
3553 int local_node
, prev_node
;
3554 struct zonelist
*zonelist
;
3555 int order
= current_zonelist_order
;
3557 /* initialize zonelists */
3558 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3559 zonelist
= pgdat
->node_zonelists
+ i
;
3560 zonelist
->_zonerefs
[0].zone
= NULL
;
3561 zonelist
->_zonerefs
[0].zone_idx
= 0;
3564 /* NUMA-aware ordering of nodes */
3565 local_node
= pgdat
->node_id
;
3566 load
= nr_online_nodes
;
3567 prev_node
= local_node
;
3568 nodes_clear(used_mask
);
3570 memset(node_order
, 0, sizeof(node_order
));
3573 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3575 * We don't want to pressure a particular node.
3576 * So adding penalty to the first node in same
3577 * distance group to make it round-robin.
3579 if (node_distance(local_node
, node
) !=
3580 node_distance(local_node
, prev_node
))
3581 node_load
[node
] = load
;
3585 if (order
== ZONELIST_ORDER_NODE
)
3586 build_zonelists_in_node_order(pgdat
, node
);
3588 node_order
[j
++] = node
; /* remember order */
3591 if (order
== ZONELIST_ORDER_ZONE
) {
3592 /* calculate node order -- i.e., DMA last! */
3593 build_zonelists_in_zone_order(pgdat
, j
);
3596 build_thisnode_zonelists(pgdat
);
3599 /* Construct the zonelist performance cache - see further mmzone.h */
3600 static void build_zonelist_cache(pg_data_t
*pgdat
)
3602 struct zonelist
*zonelist
;
3603 struct zonelist_cache
*zlc
;
3606 zonelist
= &pgdat
->node_zonelists
[0];
3607 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3608 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3609 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3610 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3613 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3615 * Return node id of node used for "local" allocations.
3616 * I.e., first node id of first zone in arg node's generic zonelist.
3617 * Used for initializing percpu 'numa_mem', which is used primarily
3618 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3620 int local_memory_node(int node
)
3624 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3625 gfp_zone(GFP_KERNEL
),
3632 #else /* CONFIG_NUMA */
3634 static void set_zonelist_order(void)
3636 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3639 static void build_zonelists(pg_data_t
*pgdat
)
3641 int node
, local_node
;
3643 struct zonelist
*zonelist
;
3645 local_node
= pgdat
->node_id
;
3647 zonelist
= &pgdat
->node_zonelists
[0];
3648 j
= build_zonelists_node(pgdat
, zonelist
, 0);
3651 * Now we build the zonelist so that it contains the zones
3652 * of all the other nodes.
3653 * We don't want to pressure a particular node, so when
3654 * building the zones for node N, we make sure that the
3655 * zones coming right after the local ones are those from
3656 * node N+1 (modulo N)
3658 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3659 if (!node_online(node
))
3661 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
3663 for (node
= 0; node
< local_node
; node
++) {
3664 if (!node_online(node
))
3666 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
3669 zonelist
->_zonerefs
[j
].zone
= NULL
;
3670 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3673 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3674 static void build_zonelist_cache(pg_data_t
*pgdat
)
3676 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3679 #endif /* CONFIG_NUMA */
3682 * Boot pageset table. One per cpu which is going to be used for all
3683 * zones and all nodes. The parameters will be set in such a way
3684 * that an item put on a list will immediately be handed over to
3685 * the buddy list. This is safe since pageset manipulation is done
3686 * with interrupts disabled.
3688 * The boot_pagesets must be kept even after bootup is complete for
3689 * unused processors and/or zones. They do play a role for bootstrapping
3690 * hotplugged processors.
3692 * zoneinfo_show() and maybe other functions do
3693 * not check if the processor is online before following the pageset pointer.
3694 * Other parts of the kernel may not check if the zone is available.
3696 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3697 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3698 static void setup_zone_pageset(struct zone
*zone
);
3701 * Global mutex to protect against size modification of zonelists
3702 * as well as to serialize pageset setup for the new populated zone.
3704 DEFINE_MUTEX(zonelists_mutex
);
3706 /* return values int ....just for stop_machine() */
3707 static int __build_all_zonelists(void *data
)
3711 pg_data_t
*self
= data
;
3714 memset(node_load
, 0, sizeof(node_load
));
3717 if (self
&& !node_online(self
->node_id
)) {
3718 build_zonelists(self
);
3719 build_zonelist_cache(self
);
3722 for_each_online_node(nid
) {
3723 pg_data_t
*pgdat
= NODE_DATA(nid
);
3725 build_zonelists(pgdat
);
3726 build_zonelist_cache(pgdat
);
3730 * Initialize the boot_pagesets that are going to be used
3731 * for bootstrapping processors. The real pagesets for
3732 * each zone will be allocated later when the per cpu
3733 * allocator is available.
3735 * boot_pagesets are used also for bootstrapping offline
3736 * cpus if the system is already booted because the pagesets
3737 * are needed to initialize allocators on a specific cpu too.
3738 * F.e. the percpu allocator needs the page allocator which
3739 * needs the percpu allocator in order to allocate its pagesets
3740 * (a chicken-egg dilemma).
3742 for_each_possible_cpu(cpu
) {
3743 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3745 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3747 * We now know the "local memory node" for each node--
3748 * i.e., the node of the first zone in the generic zonelist.
3749 * Set up numa_mem percpu variable for on-line cpus. During
3750 * boot, only the boot cpu should be on-line; we'll init the
3751 * secondary cpus' numa_mem as they come on-line. During
3752 * node/memory hotplug, we'll fixup all on-line cpus.
3754 if (cpu_online(cpu
))
3755 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3763 * Called with zonelists_mutex held always
3764 * unless system_state == SYSTEM_BOOTING.
3766 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
3768 set_zonelist_order();
3770 if (system_state
== SYSTEM_BOOTING
) {
3771 __build_all_zonelists(NULL
);
3772 mminit_verify_zonelist();
3773 cpuset_init_current_mems_allowed();
3775 #ifdef CONFIG_MEMORY_HOTPLUG
3777 setup_zone_pageset(zone
);
3779 /* we have to stop all cpus to guarantee there is no user
3781 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
3782 /* cpuset refresh routine should be here */
3784 vm_total_pages
= nr_free_pagecache_pages();
3786 * Disable grouping by mobility if the number of pages in the
3787 * system is too low to allow the mechanism to work. It would be
3788 * more accurate, but expensive to check per-zone. This check is
3789 * made on memory-hotadd so a system can start with mobility
3790 * disabled and enable it later
3792 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3793 page_group_by_mobility_disabled
= 1;
3795 page_group_by_mobility_disabled
= 0;
3797 printk("Built %i zonelists in %s order, mobility grouping %s. "
3798 "Total pages: %ld\n",
3800 zonelist_order_name
[current_zonelist_order
],
3801 page_group_by_mobility_disabled
? "off" : "on",
3804 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3809 * Helper functions to size the waitqueue hash table.
3810 * Essentially these want to choose hash table sizes sufficiently
3811 * large so that collisions trying to wait on pages are rare.
3812 * But in fact, the number of active page waitqueues on typical
3813 * systems is ridiculously low, less than 200. So this is even
3814 * conservative, even though it seems large.
3816 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3817 * waitqueues, i.e. the size of the waitq table given the number of pages.
3819 #define PAGES_PER_WAITQUEUE 256
3821 #ifndef CONFIG_MEMORY_HOTPLUG
3822 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3824 unsigned long size
= 1;
3826 pages
/= PAGES_PER_WAITQUEUE
;
3828 while (size
< pages
)
3832 * Once we have dozens or even hundreds of threads sleeping
3833 * on IO we've got bigger problems than wait queue collision.
3834 * Limit the size of the wait table to a reasonable size.
3836 size
= min(size
, 4096UL);
3838 return max(size
, 4UL);
3842 * A zone's size might be changed by hot-add, so it is not possible to determine
3843 * a suitable size for its wait_table. So we use the maximum size now.
3845 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3847 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3848 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3849 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3851 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3852 * or more by the traditional way. (See above). It equals:
3854 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3855 * ia64(16K page size) : = ( 8G + 4M)byte.
3856 * powerpc (64K page size) : = (32G +16M)byte.
3858 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3865 * This is an integer logarithm so that shifts can be used later
3866 * to extract the more random high bits from the multiplicative
3867 * hash function before the remainder is taken.
3869 static inline unsigned long wait_table_bits(unsigned long size
)
3875 * Check if a pageblock contains reserved pages
3877 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3881 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3882 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3889 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3890 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3891 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3892 * higher will lead to a bigger reserve which will get freed as contiguous
3893 * blocks as reclaim kicks in
3895 static void setup_zone_migrate_reserve(struct zone
*zone
)
3897 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3899 unsigned long block_migratetype
;
3904 * Get the start pfn, end pfn and the number of blocks to reserve
3905 * We have to be careful to be aligned to pageblock_nr_pages to
3906 * make sure that we always check pfn_valid for the first page in
3909 start_pfn
= zone
->zone_start_pfn
;
3910 end_pfn
= zone_end_pfn(zone
);
3911 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3912 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3916 * Reserve blocks are generally in place to help high-order atomic
3917 * allocations that are short-lived. A min_free_kbytes value that
3918 * would result in more than 2 reserve blocks for atomic allocations
3919 * is assumed to be in place to help anti-fragmentation for the
3920 * future allocation of hugepages at runtime.
3922 reserve
= min(2, reserve
);
3923 old_reserve
= zone
->nr_migrate_reserve_block
;
3925 /* When memory hot-add, we almost always need to do nothing */
3926 if (reserve
== old_reserve
)
3928 zone
->nr_migrate_reserve_block
= reserve
;
3930 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3931 if (!pfn_valid(pfn
))
3933 page
= pfn_to_page(pfn
);
3935 /* Watch out for overlapping nodes */
3936 if (page_to_nid(page
) != zone_to_nid(zone
))
3939 block_migratetype
= get_pageblock_migratetype(page
);
3941 /* Only test what is necessary when the reserves are not met */
3944 * Blocks with reserved pages will never free, skip
3947 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3948 if (pageblock_is_reserved(pfn
, block_end_pfn
))
3951 /* If this block is reserved, account for it */
3952 if (block_migratetype
== MIGRATE_RESERVE
) {
3957 /* Suitable for reserving if this block is movable */
3958 if (block_migratetype
== MIGRATE_MOVABLE
) {
3959 set_pageblock_migratetype(page
,
3961 move_freepages_block(zone
, page
,
3966 } else if (!old_reserve
) {
3968 * At boot time we don't need to scan the whole zone
3969 * for turning off MIGRATE_RESERVE.
3975 * If the reserve is met and this is a previous reserved block,
3978 if (block_migratetype
== MIGRATE_RESERVE
) {
3979 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3980 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3986 * Initially all pages are reserved - free ones are freed
3987 * up by free_all_bootmem() once the early boot process is
3988 * done. Non-atomic initialization, single-pass.
3990 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3991 unsigned long start_pfn
, enum memmap_context context
)
3994 unsigned long end_pfn
= start_pfn
+ size
;
3998 if (highest_memmap_pfn
< end_pfn
- 1)
3999 highest_memmap_pfn
= end_pfn
- 1;
4001 z
= &NODE_DATA(nid
)->node_zones
[zone
];
4002 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4004 * There can be holes in boot-time mem_map[]s
4005 * handed to this function. They do not
4006 * exist on hotplugged memory.
4008 if (context
== MEMMAP_EARLY
) {
4009 if (!early_pfn_valid(pfn
))
4011 if (!early_pfn_in_nid(pfn
, nid
))
4014 page
= pfn_to_page(pfn
);
4015 set_page_links(page
, zone
, nid
, pfn
);
4016 mminit_verify_page_links(page
, zone
, nid
, pfn
);
4017 init_page_count(page
);
4018 page_mapcount_reset(page
);
4019 page_cpupid_reset_last(page
);
4020 SetPageReserved(page
);
4022 * Mark the block movable so that blocks are reserved for
4023 * movable at startup. This will force kernel allocations
4024 * to reserve their blocks rather than leaking throughout
4025 * the address space during boot when many long-lived
4026 * kernel allocations are made. Later some blocks near
4027 * the start are marked MIGRATE_RESERVE by
4028 * setup_zone_migrate_reserve()
4030 * bitmap is created for zone's valid pfn range. but memmap
4031 * can be created for invalid pages (for alignment)
4032 * check here not to call set_pageblock_migratetype() against
4035 if ((z
->zone_start_pfn
<= pfn
)
4036 && (pfn
< zone_end_pfn(z
))
4037 && !(pfn
& (pageblock_nr_pages
- 1)))
4038 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4040 INIT_LIST_HEAD(&page
->lru
);
4041 #ifdef WANT_PAGE_VIRTUAL
4042 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4043 if (!is_highmem_idx(zone
))
4044 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
4049 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4052 for_each_migratetype_order(order
, t
) {
4053 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4054 zone
->free_area
[order
].nr_free
= 0;
4058 #ifndef __HAVE_ARCH_MEMMAP_INIT
4059 #define memmap_init(size, nid, zone, start_pfn) \
4060 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4063 static int __meminit
zone_batchsize(struct zone
*zone
)
4069 * The per-cpu-pages pools are set to around 1000th of the
4070 * size of the zone. But no more than 1/2 of a meg.
4072 * OK, so we don't know how big the cache is. So guess.
4074 batch
= zone
->managed_pages
/ 1024;
4075 if (batch
* PAGE_SIZE
> 512 * 1024)
4076 batch
= (512 * 1024) / PAGE_SIZE
;
4077 batch
/= 4; /* We effectively *= 4 below */
4082 * Clamp the batch to a 2^n - 1 value. Having a power
4083 * of 2 value was found to be more likely to have
4084 * suboptimal cache aliasing properties in some cases.
4086 * For example if 2 tasks are alternately allocating
4087 * batches of pages, one task can end up with a lot
4088 * of pages of one half of the possible page colors
4089 * and the other with pages of the other colors.
4091 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4096 /* The deferral and batching of frees should be suppressed under NOMMU
4099 * The problem is that NOMMU needs to be able to allocate large chunks
4100 * of contiguous memory as there's no hardware page translation to
4101 * assemble apparent contiguous memory from discontiguous pages.
4103 * Queueing large contiguous runs of pages for batching, however,
4104 * causes the pages to actually be freed in smaller chunks. As there
4105 * can be a significant delay between the individual batches being
4106 * recycled, this leads to the once large chunks of space being
4107 * fragmented and becoming unavailable for high-order allocations.
4114 * pcp->high and pcp->batch values are related and dependent on one another:
4115 * ->batch must never be higher then ->high.
4116 * The following function updates them in a safe manner without read side
4119 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4120 * those fields changing asynchronously (acording the the above rule).
4122 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4123 * outside of boot time (or some other assurance that no concurrent updaters
4126 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4127 unsigned long batch
)
4129 /* start with a fail safe value for batch */
4133 /* Update high, then batch, in order */
4140 /* a companion to pageset_set_high() */
4141 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4143 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4146 static void pageset_init(struct per_cpu_pageset
*p
)
4148 struct per_cpu_pages
*pcp
;
4151 memset(p
, 0, sizeof(*p
));
4155 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4156 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4159 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4162 pageset_set_batch(p
, batch
);
4166 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4167 * to the value high for the pageset p.
4169 static void pageset_set_high(struct per_cpu_pageset
*p
,
4172 unsigned long batch
= max(1UL, high
/ 4);
4173 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4174 batch
= PAGE_SHIFT
* 8;
4176 pageset_update(&p
->pcp
, high
, batch
);
4179 static void __meminit
pageset_set_high_and_batch(struct zone
*zone
,
4180 struct per_cpu_pageset
*pcp
)
4182 if (percpu_pagelist_fraction
)
4183 pageset_set_high(pcp
,
4184 (zone
->managed_pages
/
4185 percpu_pagelist_fraction
));
4187 pageset_set_batch(pcp
, zone_batchsize(zone
));
4190 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4192 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4195 pageset_set_high_and_batch(zone
, pcp
);
4198 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4201 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4202 for_each_possible_cpu(cpu
)
4203 zone_pageset_init(zone
, cpu
);
4207 * Allocate per cpu pagesets and initialize them.
4208 * Before this call only boot pagesets were available.
4210 void __init
setup_per_cpu_pageset(void)
4214 for_each_populated_zone(zone
)
4215 setup_zone_pageset(zone
);
4218 static noinline __init_refok
4219 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4225 * The per-page waitqueue mechanism uses hashed waitqueues
4228 zone
->wait_table_hash_nr_entries
=
4229 wait_table_hash_nr_entries(zone_size_pages
);
4230 zone
->wait_table_bits
=
4231 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4232 alloc_size
= zone
->wait_table_hash_nr_entries
4233 * sizeof(wait_queue_head_t
);
4235 if (!slab_is_available()) {
4236 zone
->wait_table
= (wait_queue_head_t
*)
4237 memblock_virt_alloc_node_nopanic(
4238 alloc_size
, zone
->zone_pgdat
->node_id
);
4241 * This case means that a zone whose size was 0 gets new memory
4242 * via memory hot-add.
4243 * But it may be the case that a new node was hot-added. In
4244 * this case vmalloc() will not be able to use this new node's
4245 * memory - this wait_table must be initialized to use this new
4246 * node itself as well.
4247 * To use this new node's memory, further consideration will be
4250 zone
->wait_table
= vmalloc(alloc_size
);
4252 if (!zone
->wait_table
)
4255 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4256 init_waitqueue_head(zone
->wait_table
+ i
);
4261 static __meminit
void zone_pcp_init(struct zone
*zone
)
4264 * per cpu subsystem is not up at this point. The following code
4265 * relies on the ability of the linker to provide the
4266 * offset of a (static) per cpu variable into the per cpu area.
4268 zone
->pageset
= &boot_pageset
;
4270 if (populated_zone(zone
))
4271 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4272 zone
->name
, zone
->present_pages
,
4273 zone_batchsize(zone
));
4276 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4277 unsigned long zone_start_pfn
,
4279 enum memmap_context context
)
4281 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4283 ret
= zone_wait_table_init(zone
, size
);
4286 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4288 zone
->zone_start_pfn
= zone_start_pfn
;
4290 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4291 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4293 (unsigned long)zone_idx(zone
),
4294 zone_start_pfn
, (zone_start_pfn
+ size
));
4296 zone_init_free_lists(zone
);
4301 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4302 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4304 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4305 * Architectures may implement their own version but if add_active_range()
4306 * was used and there are no special requirements, this is a convenient
4309 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
4311 unsigned long start_pfn
, end_pfn
;
4314 * NOTE: The following SMP-unsafe globals are only used early in boot
4315 * when the kernel is running single-threaded.
4317 static unsigned long __meminitdata last_start_pfn
, last_end_pfn
;
4318 static int __meminitdata last_nid
;
4320 if (last_start_pfn
<= pfn
&& pfn
< last_end_pfn
)
4323 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4325 last_start_pfn
= start_pfn
;
4326 last_end_pfn
= end_pfn
;
4332 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4334 int __meminit
early_pfn_to_nid(unsigned long pfn
)
4338 nid
= __early_pfn_to_nid(pfn
);
4341 /* just returns 0 */
4345 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4346 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
4350 nid
= __early_pfn_to_nid(pfn
);
4351 if (nid
>= 0 && nid
!= node
)
4358 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4359 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4360 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4362 * If an architecture guarantees that all ranges registered with
4363 * add_active_ranges() contain no holes and may be freed, this
4364 * this function may be used instead of calling memblock_free_early_nid()
4367 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4369 unsigned long start_pfn
, end_pfn
;
4372 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4373 start_pfn
= min(start_pfn
, max_low_pfn
);
4374 end_pfn
= min(end_pfn
, max_low_pfn
);
4376 if (start_pfn
< end_pfn
)
4377 memblock_free_early_nid(PFN_PHYS(start_pfn
),
4378 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
4384 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4385 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4387 * If an architecture guarantees that all ranges registered with
4388 * add_active_ranges() contain no holes and may be freed, this
4389 * function may be used instead of calling memory_present() manually.
4391 void __init
sparse_memory_present_with_active_regions(int nid
)
4393 unsigned long start_pfn
, end_pfn
;
4396 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4397 memory_present(this_nid
, start_pfn
, end_pfn
);
4401 * get_pfn_range_for_nid - Return the start and end page frames for a node
4402 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4403 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4404 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4406 * It returns the start and end page frame of a node based on information
4407 * provided by an arch calling add_active_range(). If called for a node
4408 * with no available memory, a warning is printed and the start and end
4411 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4412 unsigned long *start_pfn
, unsigned long *end_pfn
)
4414 unsigned long this_start_pfn
, this_end_pfn
;
4420 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4421 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4422 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4425 if (*start_pfn
== -1UL)
4430 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4431 * assumption is made that zones within a node are ordered in monotonic
4432 * increasing memory addresses so that the "highest" populated zone is used
4434 static void __init
find_usable_zone_for_movable(void)
4437 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4438 if (zone_index
== ZONE_MOVABLE
)
4441 if (arch_zone_highest_possible_pfn
[zone_index
] >
4442 arch_zone_lowest_possible_pfn
[zone_index
])
4446 VM_BUG_ON(zone_index
== -1);
4447 movable_zone
= zone_index
;
4451 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4452 * because it is sized independent of architecture. Unlike the other zones,
4453 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4454 * in each node depending on the size of each node and how evenly kernelcore
4455 * is distributed. This helper function adjusts the zone ranges
4456 * provided by the architecture for a given node by using the end of the
4457 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4458 * zones within a node are in order of monotonic increases memory addresses
4460 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4461 unsigned long zone_type
,
4462 unsigned long node_start_pfn
,
4463 unsigned long node_end_pfn
,
4464 unsigned long *zone_start_pfn
,
4465 unsigned long *zone_end_pfn
)
4467 /* Only adjust if ZONE_MOVABLE is on this node */
4468 if (zone_movable_pfn
[nid
]) {
4469 /* Size ZONE_MOVABLE */
4470 if (zone_type
== ZONE_MOVABLE
) {
4471 *zone_start_pfn
= zone_movable_pfn
[nid
];
4472 *zone_end_pfn
= min(node_end_pfn
,
4473 arch_zone_highest_possible_pfn
[movable_zone
]);
4475 /* Adjust for ZONE_MOVABLE starting within this range */
4476 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4477 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4478 *zone_end_pfn
= zone_movable_pfn
[nid
];
4480 /* Check if this whole range is within ZONE_MOVABLE */
4481 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4482 *zone_start_pfn
= *zone_end_pfn
;
4487 * Return the number of pages a zone spans in a node, including holes
4488 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4490 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4491 unsigned long zone_type
,
4492 unsigned long node_start_pfn
,
4493 unsigned long node_end_pfn
,
4494 unsigned long *ignored
)
4496 unsigned long zone_start_pfn
, zone_end_pfn
;
4498 /* Get the start and end of the zone */
4499 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4500 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4501 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4502 node_start_pfn
, node_end_pfn
,
4503 &zone_start_pfn
, &zone_end_pfn
);
4505 /* Check that this node has pages within the zone's required range */
4506 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4509 /* Move the zone boundaries inside the node if necessary */
4510 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4511 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4513 /* Return the spanned pages */
4514 return zone_end_pfn
- zone_start_pfn
;
4518 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4519 * then all holes in the requested range will be accounted for.
4521 unsigned long __meminit
__absent_pages_in_range(int nid
,
4522 unsigned long range_start_pfn
,
4523 unsigned long range_end_pfn
)
4525 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4526 unsigned long start_pfn
, end_pfn
;
4529 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4530 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4531 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4532 nr_absent
-= end_pfn
- start_pfn
;
4538 * absent_pages_in_range - Return number of page frames in holes within a range
4539 * @start_pfn: The start PFN to start searching for holes
4540 * @end_pfn: The end PFN to stop searching for holes
4542 * It returns the number of pages frames in memory holes within a range.
4544 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4545 unsigned long end_pfn
)
4547 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4550 /* Return the number of page frames in holes in a zone on a node */
4551 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4552 unsigned long zone_type
,
4553 unsigned long node_start_pfn
,
4554 unsigned long node_end_pfn
,
4555 unsigned long *ignored
)
4557 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4558 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4559 unsigned long zone_start_pfn
, zone_end_pfn
;
4561 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4562 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4564 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4565 node_start_pfn
, node_end_pfn
,
4566 &zone_start_pfn
, &zone_end_pfn
);
4567 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4570 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4571 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4572 unsigned long zone_type
,
4573 unsigned long node_start_pfn
,
4574 unsigned long node_end_pfn
,
4575 unsigned long *zones_size
)
4577 return zones_size
[zone_type
];
4580 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4581 unsigned long zone_type
,
4582 unsigned long node_start_pfn
,
4583 unsigned long node_end_pfn
,
4584 unsigned long *zholes_size
)
4589 return zholes_size
[zone_type
];
4592 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4594 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4595 unsigned long node_start_pfn
,
4596 unsigned long node_end_pfn
,
4597 unsigned long *zones_size
,
4598 unsigned long *zholes_size
)
4600 unsigned long realtotalpages
, totalpages
= 0;
4603 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4604 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4608 pgdat
->node_spanned_pages
= totalpages
;
4610 realtotalpages
= totalpages
;
4611 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4613 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4614 node_start_pfn
, node_end_pfn
,
4616 pgdat
->node_present_pages
= realtotalpages
;
4617 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4621 #ifndef CONFIG_SPARSEMEM
4623 * Calculate the size of the zone->blockflags rounded to an unsigned long
4624 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4625 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4626 * round what is now in bits to nearest long in bits, then return it in
4629 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
4631 unsigned long usemapsize
;
4633 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
4634 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4635 usemapsize
= usemapsize
>> pageblock_order
;
4636 usemapsize
*= NR_PAGEBLOCK_BITS
;
4637 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4639 return usemapsize
/ 8;
4642 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4644 unsigned long zone_start_pfn
,
4645 unsigned long zonesize
)
4647 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
4648 zone
->pageblock_flags
= NULL
;
4650 zone
->pageblock_flags
=
4651 memblock_virt_alloc_node_nopanic(usemapsize
,
4655 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
4656 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
4657 #endif /* CONFIG_SPARSEMEM */
4659 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4661 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4662 void __paginginit
set_pageblock_order(void)
4666 /* Check that pageblock_nr_pages has not already been setup */
4667 if (pageblock_order
)
4670 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4671 order
= HUGETLB_PAGE_ORDER
;
4673 order
= MAX_ORDER
- 1;
4676 * Assume the largest contiguous order of interest is a huge page.
4677 * This value may be variable depending on boot parameters on IA64 and
4680 pageblock_order
= order
;
4682 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4685 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4686 * is unused as pageblock_order is set at compile-time. See
4687 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4690 void __paginginit
set_pageblock_order(void)
4694 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4696 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
4697 unsigned long present_pages
)
4699 unsigned long pages
= spanned_pages
;
4702 * Provide a more accurate estimation if there are holes within
4703 * the zone and SPARSEMEM is in use. If there are holes within the
4704 * zone, each populated memory region may cost us one or two extra
4705 * memmap pages due to alignment because memmap pages for each
4706 * populated regions may not naturally algined on page boundary.
4707 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4709 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
4710 IS_ENABLED(CONFIG_SPARSEMEM
))
4711 pages
= present_pages
;
4713 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
4717 * Set up the zone data structures:
4718 * - mark all pages reserved
4719 * - mark all memory queues empty
4720 * - clear the memory bitmaps
4722 * NOTE: pgdat should get zeroed by caller.
4724 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4725 unsigned long node_start_pfn
, unsigned long node_end_pfn
,
4726 unsigned long *zones_size
, unsigned long *zholes_size
)
4729 int nid
= pgdat
->node_id
;
4730 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4733 pgdat_resize_init(pgdat
);
4734 #ifdef CONFIG_NUMA_BALANCING
4735 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
4736 pgdat
->numabalancing_migrate_nr_pages
= 0;
4737 pgdat
->numabalancing_migrate_next_window
= jiffies
;
4739 init_waitqueue_head(&pgdat
->kswapd_wait
);
4740 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
4741 pgdat_page_cgroup_init(pgdat
);
4743 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4744 struct zone
*zone
= pgdat
->node_zones
+ j
;
4745 unsigned long size
, realsize
, freesize
, memmap_pages
;
4747 size
= zone_spanned_pages_in_node(nid
, j
, node_start_pfn
,
4748 node_end_pfn
, zones_size
);
4749 realsize
= freesize
= size
- zone_absent_pages_in_node(nid
, j
,
4755 * Adjust freesize so that it accounts for how much memory
4756 * is used by this zone for memmap. This affects the watermark
4757 * and per-cpu initialisations
4759 memmap_pages
= calc_memmap_size(size
, realsize
);
4760 if (freesize
>= memmap_pages
) {
4761 freesize
-= memmap_pages
;
4764 " %s zone: %lu pages used for memmap\n",
4765 zone_names
[j
], memmap_pages
);
4768 " %s zone: %lu pages exceeds freesize %lu\n",
4769 zone_names
[j
], memmap_pages
, freesize
);
4771 /* Account for reserved pages */
4772 if (j
== 0 && freesize
> dma_reserve
) {
4773 freesize
-= dma_reserve
;
4774 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4775 zone_names
[0], dma_reserve
);
4778 if (!is_highmem_idx(j
))
4779 nr_kernel_pages
+= freesize
;
4780 /* Charge for highmem memmap if there are enough kernel pages */
4781 else if (nr_kernel_pages
> memmap_pages
* 2)
4782 nr_kernel_pages
-= memmap_pages
;
4783 nr_all_pages
+= freesize
;
4785 zone
->spanned_pages
= size
;
4786 zone
->present_pages
= realsize
;
4788 * Set an approximate value for lowmem here, it will be adjusted
4789 * when the bootmem allocator frees pages into the buddy system.
4790 * And all highmem pages will be managed by the buddy system.
4792 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
4795 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
4797 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
4799 zone
->name
= zone_names
[j
];
4800 spin_lock_init(&zone
->lock
);
4801 spin_lock_init(&zone
->lru_lock
);
4802 zone_seqlock_init(zone
);
4803 zone
->zone_pgdat
= pgdat
;
4804 zone_pcp_init(zone
);
4806 /* For bootup, initialized properly in watermark setup */
4807 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
4809 lruvec_init(&zone
->lruvec
);
4813 set_pageblock_order();
4814 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
4815 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4816 size
, MEMMAP_EARLY
);
4818 memmap_init(size
, nid
, j
, zone_start_pfn
);
4819 zone_start_pfn
+= size
;
4823 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4825 /* Skip empty nodes */
4826 if (!pgdat
->node_spanned_pages
)
4829 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4830 /* ia64 gets its own node_mem_map, before this, without bootmem */
4831 if (!pgdat
->node_mem_map
) {
4832 unsigned long size
, start
, end
;
4836 * The zone's endpoints aren't required to be MAX_ORDER
4837 * aligned but the node_mem_map endpoints must be in order
4838 * for the buddy allocator to function correctly.
4840 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4841 end
= pgdat_end_pfn(pgdat
);
4842 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4843 size
= (end
- start
) * sizeof(struct page
);
4844 map
= alloc_remap(pgdat
->node_id
, size
);
4846 map
= memblock_virt_alloc_node_nopanic(size
,
4848 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4850 #ifndef CONFIG_NEED_MULTIPLE_NODES
4852 * With no DISCONTIG, the global mem_map is just set as node 0's
4854 if (pgdat
== NODE_DATA(0)) {
4855 mem_map
= NODE_DATA(0)->node_mem_map
;
4856 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4857 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4858 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4859 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4862 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4865 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4866 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4868 pg_data_t
*pgdat
= NODE_DATA(nid
);
4869 unsigned long start_pfn
= 0;
4870 unsigned long end_pfn
= 0;
4872 /* pg_data_t should be reset to zero when it's allocated */
4873 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
4875 pgdat
->node_id
= nid
;
4876 pgdat
->node_start_pfn
= node_start_pfn
;
4877 init_zone_allows_reclaim(nid
);
4878 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4879 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
4881 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
4882 zones_size
, zholes_size
);
4884 alloc_node_mem_map(pgdat
);
4885 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4886 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4887 nid
, (unsigned long)pgdat
,
4888 (unsigned long)pgdat
->node_mem_map
);
4891 free_area_init_core(pgdat
, start_pfn
, end_pfn
,
4892 zones_size
, zholes_size
);
4895 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4897 #if MAX_NUMNODES > 1
4899 * Figure out the number of possible node ids.
4901 void __init
setup_nr_node_ids(void)
4904 unsigned int highest
= 0;
4906 for_each_node_mask(node
, node_possible_map
)
4908 nr_node_ids
= highest
+ 1;
4913 * node_map_pfn_alignment - determine the maximum internode alignment
4915 * This function should be called after node map is populated and sorted.
4916 * It calculates the maximum power of two alignment which can distinguish
4919 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4920 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4921 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4922 * shifted, 1GiB is enough and this function will indicate so.
4924 * This is used to test whether pfn -> nid mapping of the chosen memory
4925 * model has fine enough granularity to avoid incorrect mapping for the
4926 * populated node map.
4928 * Returns the determined alignment in pfn's. 0 if there is no alignment
4929 * requirement (single node).
4931 unsigned long __init
node_map_pfn_alignment(void)
4933 unsigned long accl_mask
= 0, last_end
= 0;
4934 unsigned long start
, end
, mask
;
4938 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4939 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4946 * Start with a mask granular enough to pin-point to the
4947 * start pfn and tick off bits one-by-one until it becomes
4948 * too coarse to separate the current node from the last.
4950 mask
= ~((1 << __ffs(start
)) - 1);
4951 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4954 /* accumulate all internode masks */
4958 /* convert mask to number of pages */
4959 return ~accl_mask
+ 1;
4962 /* Find the lowest pfn for a node */
4963 static unsigned long __init
find_min_pfn_for_node(int nid
)
4965 unsigned long min_pfn
= ULONG_MAX
;
4966 unsigned long start_pfn
;
4969 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4970 min_pfn
= min(min_pfn
, start_pfn
);
4972 if (min_pfn
== ULONG_MAX
) {
4974 "Could not find start_pfn for node %d\n", nid
);
4982 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4984 * It returns the minimum PFN based on information provided via
4985 * add_active_range().
4987 unsigned long __init
find_min_pfn_with_active_regions(void)
4989 return find_min_pfn_for_node(MAX_NUMNODES
);
4993 * early_calculate_totalpages()
4994 * Sum pages in active regions for movable zone.
4995 * Populate N_MEMORY for calculating usable_nodes.
4997 static unsigned long __init
early_calculate_totalpages(void)
4999 unsigned long totalpages
= 0;
5000 unsigned long start_pfn
, end_pfn
;
5003 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5004 unsigned long pages
= end_pfn
- start_pfn
;
5006 totalpages
+= pages
;
5008 node_set_state(nid
, N_MEMORY
);
5014 * Find the PFN the Movable zone begins in each node. Kernel memory
5015 * is spread evenly between nodes as long as the nodes have enough
5016 * memory. When they don't, some nodes will have more kernelcore than
5019 static void __init
find_zone_movable_pfns_for_nodes(void)
5022 unsigned long usable_startpfn
;
5023 unsigned long kernelcore_node
, kernelcore_remaining
;
5024 /* save the state before borrow the nodemask */
5025 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5026 unsigned long totalpages
= early_calculate_totalpages();
5027 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5028 struct memblock_type
*type
= &memblock
.memory
;
5030 /* Need to find movable_zone earlier when movable_node is specified. */
5031 find_usable_zone_for_movable();
5034 * If movable_node is specified, ignore kernelcore and movablecore
5037 if (movable_node_is_enabled()) {
5038 for (i
= 0; i
< type
->cnt
; i
++) {
5039 if (!memblock_is_hotpluggable(&type
->regions
[i
]))
5042 nid
= type
->regions
[i
].nid
;
5044 usable_startpfn
= PFN_DOWN(type
->regions
[i
].base
);
5045 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5046 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5054 * If movablecore=nn[KMG] was specified, calculate what size of
5055 * kernelcore that corresponds so that memory usable for
5056 * any allocation type is evenly spread. If both kernelcore
5057 * and movablecore are specified, then the value of kernelcore
5058 * will be used for required_kernelcore if it's greater than
5059 * what movablecore would have allowed.
5061 if (required_movablecore
) {
5062 unsigned long corepages
;
5065 * Round-up so that ZONE_MOVABLE is at least as large as what
5066 * was requested by the user
5068 required_movablecore
=
5069 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5070 corepages
= totalpages
- required_movablecore
;
5072 required_kernelcore
= max(required_kernelcore
, corepages
);
5075 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5076 if (!required_kernelcore
)
5079 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5080 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5083 /* Spread kernelcore memory as evenly as possible throughout nodes */
5084 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5085 for_each_node_state(nid
, N_MEMORY
) {
5086 unsigned long start_pfn
, end_pfn
;
5089 * Recalculate kernelcore_node if the division per node
5090 * now exceeds what is necessary to satisfy the requested
5091 * amount of memory for the kernel
5093 if (required_kernelcore
< kernelcore_node
)
5094 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5097 * As the map is walked, we track how much memory is usable
5098 * by the kernel using kernelcore_remaining. When it is
5099 * 0, the rest of the node is usable by ZONE_MOVABLE
5101 kernelcore_remaining
= kernelcore_node
;
5103 /* Go through each range of PFNs within this node */
5104 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5105 unsigned long size_pages
;
5107 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5108 if (start_pfn
>= end_pfn
)
5111 /* Account for what is only usable for kernelcore */
5112 if (start_pfn
< usable_startpfn
) {
5113 unsigned long kernel_pages
;
5114 kernel_pages
= min(end_pfn
, usable_startpfn
)
5117 kernelcore_remaining
-= min(kernel_pages
,
5118 kernelcore_remaining
);
5119 required_kernelcore
-= min(kernel_pages
,
5120 required_kernelcore
);
5122 /* Continue if range is now fully accounted */
5123 if (end_pfn
<= usable_startpfn
) {
5126 * Push zone_movable_pfn to the end so
5127 * that if we have to rebalance
5128 * kernelcore across nodes, we will
5129 * not double account here
5131 zone_movable_pfn
[nid
] = end_pfn
;
5134 start_pfn
= usable_startpfn
;
5138 * The usable PFN range for ZONE_MOVABLE is from
5139 * start_pfn->end_pfn. Calculate size_pages as the
5140 * number of pages used as kernelcore
5142 size_pages
= end_pfn
- start_pfn
;
5143 if (size_pages
> kernelcore_remaining
)
5144 size_pages
= kernelcore_remaining
;
5145 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5148 * Some kernelcore has been met, update counts and
5149 * break if the kernelcore for this node has been
5152 required_kernelcore
-= min(required_kernelcore
,
5154 kernelcore_remaining
-= size_pages
;
5155 if (!kernelcore_remaining
)
5161 * If there is still required_kernelcore, we do another pass with one
5162 * less node in the count. This will push zone_movable_pfn[nid] further
5163 * along on the nodes that still have memory until kernelcore is
5167 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5171 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5172 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5173 zone_movable_pfn
[nid
] =
5174 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5177 /* restore the node_state */
5178 node_states
[N_MEMORY
] = saved_node_state
;
5181 /* Any regular or high memory on that node ? */
5182 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5184 enum zone_type zone_type
;
5186 if (N_MEMORY
== N_NORMAL_MEMORY
)
5189 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5190 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5191 if (populated_zone(zone
)) {
5192 node_set_state(nid
, N_HIGH_MEMORY
);
5193 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5194 zone_type
<= ZONE_NORMAL
)
5195 node_set_state(nid
, N_NORMAL_MEMORY
);
5202 * free_area_init_nodes - Initialise all pg_data_t and zone data
5203 * @max_zone_pfn: an array of max PFNs for each zone
5205 * This will call free_area_init_node() for each active node in the system.
5206 * Using the page ranges provided by add_active_range(), the size of each
5207 * zone in each node and their holes is calculated. If the maximum PFN
5208 * between two adjacent zones match, it is assumed that the zone is empty.
5209 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5210 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5211 * starts where the previous one ended. For example, ZONE_DMA32 starts
5212 * at arch_max_dma_pfn.
5214 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5216 unsigned long start_pfn
, end_pfn
;
5219 /* Record where the zone boundaries are */
5220 memset(arch_zone_lowest_possible_pfn
, 0,
5221 sizeof(arch_zone_lowest_possible_pfn
));
5222 memset(arch_zone_highest_possible_pfn
, 0,
5223 sizeof(arch_zone_highest_possible_pfn
));
5224 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5225 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5226 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5227 if (i
== ZONE_MOVABLE
)
5229 arch_zone_lowest_possible_pfn
[i
] =
5230 arch_zone_highest_possible_pfn
[i
-1];
5231 arch_zone_highest_possible_pfn
[i
] =
5232 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5234 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5235 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5237 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5238 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5239 find_zone_movable_pfns_for_nodes();
5241 /* Print out the zone ranges */
5242 printk("Zone ranges:\n");
5243 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5244 if (i
== ZONE_MOVABLE
)
5246 printk(KERN_CONT
" %-8s ", zone_names
[i
]);
5247 if (arch_zone_lowest_possible_pfn
[i
] ==
5248 arch_zone_highest_possible_pfn
[i
])
5249 printk(KERN_CONT
"empty\n");
5251 printk(KERN_CONT
"[mem %0#10lx-%0#10lx]\n",
5252 arch_zone_lowest_possible_pfn
[i
] << PAGE_SHIFT
,
5253 (arch_zone_highest_possible_pfn
[i
]
5254 << PAGE_SHIFT
) - 1);
5257 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5258 printk("Movable zone start for each node\n");
5259 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5260 if (zone_movable_pfn
[i
])
5261 printk(" Node %d: %#010lx\n", i
,
5262 zone_movable_pfn
[i
] << PAGE_SHIFT
);
5265 /* Print out the early node map */
5266 printk("Early memory node ranges\n");
5267 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5268 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid
,
5269 start_pfn
<< PAGE_SHIFT
, (end_pfn
<< PAGE_SHIFT
) - 1);
5271 /* Initialise every node */
5272 mminit_verify_pageflags_layout();
5273 setup_nr_node_ids();
5274 for_each_online_node(nid
) {
5275 pg_data_t
*pgdat
= NODE_DATA(nid
);
5276 free_area_init_node(nid
, NULL
,
5277 find_min_pfn_for_node(nid
), NULL
);
5279 /* Any memory on that node */
5280 if (pgdat
->node_present_pages
)
5281 node_set_state(nid
, N_MEMORY
);
5282 check_for_memory(pgdat
, nid
);
5286 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5288 unsigned long long coremem
;
5292 coremem
= memparse(p
, &p
);
5293 *core
= coremem
>> PAGE_SHIFT
;
5295 /* Paranoid check that UL is enough for the coremem value */
5296 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5302 * kernelcore=size sets the amount of memory for use for allocations that
5303 * cannot be reclaimed or migrated.
5305 static int __init
cmdline_parse_kernelcore(char *p
)
5307 return cmdline_parse_core(p
, &required_kernelcore
);
5311 * movablecore=size sets the amount of memory for use for allocations that
5312 * can be reclaimed or migrated.
5314 static int __init
cmdline_parse_movablecore(char *p
)
5316 return cmdline_parse_core(p
, &required_movablecore
);
5319 early_param("kernelcore", cmdline_parse_kernelcore
);
5320 early_param("movablecore", cmdline_parse_movablecore
);
5322 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5324 void adjust_managed_page_count(struct page
*page
, long count
)
5326 spin_lock(&managed_page_count_lock
);
5327 page_zone(page
)->managed_pages
+= count
;
5328 totalram_pages
+= count
;
5329 #ifdef CONFIG_HIGHMEM
5330 if (PageHighMem(page
))
5331 totalhigh_pages
+= count
;
5333 spin_unlock(&managed_page_count_lock
);
5335 EXPORT_SYMBOL(adjust_managed_page_count
);
5337 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
5340 unsigned long pages
= 0;
5342 start
= (void *)PAGE_ALIGN((unsigned long)start
);
5343 end
= (void *)((unsigned long)end
& PAGE_MASK
);
5344 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5345 if ((unsigned int)poison
<= 0xFF)
5346 memset(pos
, poison
, PAGE_SIZE
);
5347 free_reserved_page(virt_to_page(pos
));
5351 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5352 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
5356 EXPORT_SYMBOL(free_reserved_area
);
5358 #ifdef CONFIG_HIGHMEM
5359 void free_highmem_page(struct page
*page
)
5361 __free_reserved_page(page
);
5363 page_zone(page
)->managed_pages
++;
5369 void __init
mem_init_print_info(const char *str
)
5371 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
5372 unsigned long init_code_size
, init_data_size
;
5374 physpages
= get_num_physpages();
5375 codesize
= _etext
- _stext
;
5376 datasize
= _edata
- _sdata
;
5377 rosize
= __end_rodata
- __start_rodata
;
5378 bss_size
= __bss_stop
- __bss_start
;
5379 init_data_size
= __init_end
- __init_begin
;
5380 init_code_size
= _einittext
- _sinittext
;
5383 * Detect special cases and adjust section sizes accordingly:
5384 * 1) .init.* may be embedded into .data sections
5385 * 2) .init.text.* may be out of [__init_begin, __init_end],
5386 * please refer to arch/tile/kernel/vmlinux.lds.S.
5387 * 3) .rodata.* may be embedded into .text or .data sections.
5389 #define adj_init_size(start, end, size, pos, adj) \
5391 if (start <= pos && pos < end && size > adj) \
5395 adj_init_size(__init_begin
, __init_end
, init_data_size
,
5396 _sinittext
, init_code_size
);
5397 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
5398 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
5399 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
5400 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
5402 #undef adj_init_size
5404 printk("Memory: %luK/%luK available "
5405 "(%luK kernel code, %luK rwdata, %luK rodata, "
5406 "%luK init, %luK bss, %luK reserved"
5407 #ifdef CONFIG_HIGHMEM
5411 nr_free_pages() << (PAGE_SHIFT
-10), physpages
<< (PAGE_SHIFT
-10),
5412 codesize
>> 10, datasize
>> 10, rosize
>> 10,
5413 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
5414 (physpages
- totalram_pages
) << (PAGE_SHIFT
-10),
5415 #ifdef CONFIG_HIGHMEM
5416 totalhigh_pages
<< (PAGE_SHIFT
-10),
5418 str
? ", " : "", str
? str
: "");
5422 * set_dma_reserve - set the specified number of pages reserved in the first zone
5423 * @new_dma_reserve: The number of pages to mark reserved
5425 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5426 * In the DMA zone, a significant percentage may be consumed by kernel image
5427 * and other unfreeable allocations which can skew the watermarks badly. This
5428 * function may optionally be used to account for unfreeable pages in the
5429 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5430 * smaller per-cpu batchsize.
5432 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5434 dma_reserve
= new_dma_reserve
;
5437 void __init
free_area_init(unsigned long *zones_size
)
5439 free_area_init_node(0, zones_size
,
5440 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5443 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5444 unsigned long action
, void *hcpu
)
5446 int cpu
= (unsigned long)hcpu
;
5448 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5449 lru_add_drain_cpu(cpu
);
5453 * Spill the event counters of the dead processor
5454 * into the current processors event counters.
5455 * This artificially elevates the count of the current
5458 vm_events_fold_cpu(cpu
);
5461 * Zero the differential counters of the dead processor
5462 * so that the vm statistics are consistent.
5464 * This is only okay since the processor is dead and cannot
5465 * race with what we are doing.
5467 cpu_vm_stats_fold(cpu
);
5472 void __init
page_alloc_init(void)
5474 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5478 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5479 * or min_free_kbytes changes.
5481 static void calculate_totalreserve_pages(void)
5483 struct pglist_data
*pgdat
;
5484 unsigned long reserve_pages
= 0;
5485 enum zone_type i
, j
;
5487 for_each_online_pgdat(pgdat
) {
5488 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5489 struct zone
*zone
= pgdat
->node_zones
+ i
;
5490 unsigned long max
= 0;
5492 /* Find valid and maximum lowmem_reserve in the zone */
5493 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5494 if (zone
->lowmem_reserve
[j
] > max
)
5495 max
= zone
->lowmem_reserve
[j
];
5498 /* we treat the high watermark as reserved pages. */
5499 max
+= high_wmark_pages(zone
);
5501 if (max
> zone
->managed_pages
)
5502 max
= zone
->managed_pages
;
5503 reserve_pages
+= max
;
5505 * Lowmem reserves are not available to
5506 * GFP_HIGHUSER page cache allocations and
5507 * kswapd tries to balance zones to their high
5508 * watermark. As a result, neither should be
5509 * regarded as dirtyable memory, to prevent a
5510 * situation where reclaim has to clean pages
5511 * in order to balance the zones.
5513 zone
->dirty_balance_reserve
= max
;
5516 dirty_balance_reserve
= reserve_pages
;
5517 totalreserve_pages
= reserve_pages
;
5521 * setup_per_zone_lowmem_reserve - called whenever
5522 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5523 * has a correct pages reserved value, so an adequate number of
5524 * pages are left in the zone after a successful __alloc_pages().
5526 static void setup_per_zone_lowmem_reserve(void)
5528 struct pglist_data
*pgdat
;
5529 enum zone_type j
, idx
;
5531 for_each_online_pgdat(pgdat
) {
5532 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5533 struct zone
*zone
= pgdat
->node_zones
+ j
;
5534 unsigned long managed_pages
= zone
->managed_pages
;
5536 zone
->lowmem_reserve
[j
] = 0;
5540 struct zone
*lower_zone
;
5544 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5545 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5547 lower_zone
= pgdat
->node_zones
+ idx
;
5548 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
5549 sysctl_lowmem_reserve_ratio
[idx
];
5550 managed_pages
+= lower_zone
->managed_pages
;
5555 /* update totalreserve_pages */
5556 calculate_totalreserve_pages();
5559 static void __setup_per_zone_wmarks(void)
5561 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5562 unsigned long lowmem_pages
= 0;
5564 unsigned long flags
;
5566 /* Calculate total number of !ZONE_HIGHMEM pages */
5567 for_each_zone(zone
) {
5568 if (!is_highmem(zone
))
5569 lowmem_pages
+= zone
->managed_pages
;
5572 for_each_zone(zone
) {
5575 spin_lock_irqsave(&zone
->lock
, flags
);
5576 tmp
= (u64
)pages_min
* zone
->managed_pages
;
5577 do_div(tmp
, lowmem_pages
);
5578 if (is_highmem(zone
)) {
5580 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5581 * need highmem pages, so cap pages_min to a small
5584 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5585 * deltas controls asynch page reclaim, and so should
5586 * not be capped for highmem.
5588 unsigned long min_pages
;
5590 min_pages
= zone
->managed_pages
/ 1024;
5591 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
5592 zone
->watermark
[WMARK_MIN
] = min_pages
;
5595 * If it's a lowmem zone, reserve a number of pages
5596 * proportionate to the zone's size.
5598 zone
->watermark
[WMARK_MIN
] = tmp
;
5601 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
5602 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
5604 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
5605 high_wmark_pages(zone
) -
5606 low_wmark_pages(zone
) -
5607 zone_page_state(zone
, NR_ALLOC_BATCH
));
5609 setup_zone_migrate_reserve(zone
);
5610 spin_unlock_irqrestore(&zone
->lock
, flags
);
5613 /* update totalreserve_pages */
5614 calculate_totalreserve_pages();
5618 * setup_per_zone_wmarks - called when min_free_kbytes changes
5619 * or when memory is hot-{added|removed}
5621 * Ensures that the watermark[min,low,high] values for each zone are set
5622 * correctly with respect to min_free_kbytes.
5624 void setup_per_zone_wmarks(void)
5626 mutex_lock(&zonelists_mutex
);
5627 __setup_per_zone_wmarks();
5628 mutex_unlock(&zonelists_mutex
);
5632 * The inactive anon list should be small enough that the VM never has to
5633 * do too much work, but large enough that each inactive page has a chance
5634 * to be referenced again before it is swapped out.
5636 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5637 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5638 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5639 * the anonymous pages are kept on the inactive list.
5642 * memory ratio inactive anon
5643 * -------------------------------------
5652 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
5654 unsigned int gb
, ratio
;
5656 /* Zone size in gigabytes */
5657 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
5659 ratio
= int_sqrt(10 * gb
);
5663 zone
->inactive_ratio
= ratio
;
5666 static void __meminit
setup_per_zone_inactive_ratio(void)
5671 calculate_zone_inactive_ratio(zone
);
5675 * Initialise min_free_kbytes.
5677 * For small machines we want it small (128k min). For large machines
5678 * we want it large (64MB max). But it is not linear, because network
5679 * bandwidth does not increase linearly with machine size. We use
5681 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5682 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5698 int __meminit
init_per_zone_wmark_min(void)
5700 unsigned long lowmem_kbytes
;
5701 int new_min_free_kbytes
;
5703 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5704 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5706 if (new_min_free_kbytes
> user_min_free_kbytes
) {
5707 min_free_kbytes
= new_min_free_kbytes
;
5708 if (min_free_kbytes
< 128)
5709 min_free_kbytes
= 128;
5710 if (min_free_kbytes
> 65536)
5711 min_free_kbytes
= 65536;
5713 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5714 new_min_free_kbytes
, user_min_free_kbytes
);
5716 setup_per_zone_wmarks();
5717 refresh_zone_stat_thresholds();
5718 setup_per_zone_lowmem_reserve();
5719 setup_per_zone_inactive_ratio();
5722 module_init(init_per_zone_wmark_min
)
5725 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5726 * that we can call two helper functions whenever min_free_kbytes
5729 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5730 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5732 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5734 user_min_free_kbytes
= min_free_kbytes
;
5735 setup_per_zone_wmarks();
5741 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5742 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5747 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5752 zone
->min_unmapped_pages
= (zone
->managed_pages
*
5753 sysctl_min_unmapped_ratio
) / 100;
5757 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5758 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5763 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5768 zone
->min_slab_pages
= (zone
->managed_pages
*
5769 sysctl_min_slab_ratio
) / 100;
5775 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5776 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5777 * whenever sysctl_lowmem_reserve_ratio changes.
5779 * The reserve ratio obviously has absolutely no relation with the
5780 * minimum watermarks. The lowmem reserve ratio can only make sense
5781 * if in function of the boot time zone sizes.
5783 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5784 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5786 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5787 setup_per_zone_lowmem_reserve();
5792 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5793 * cpu. It is the fraction of total pages in each zone that a hot per cpu
5794 * pagelist can have before it gets flushed back to buddy allocator.
5796 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5797 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5803 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5804 if (!write
|| (ret
< 0))
5807 mutex_lock(&pcp_batch_high_lock
);
5808 for_each_populated_zone(zone
) {
5810 high
= zone
->managed_pages
/ percpu_pagelist_fraction
;
5811 for_each_possible_cpu(cpu
)
5812 pageset_set_high(per_cpu_ptr(zone
->pageset
, cpu
),
5815 mutex_unlock(&pcp_batch_high_lock
);
5819 int hashdist
= HASHDIST_DEFAULT
;
5822 static int __init
set_hashdist(char *str
)
5826 hashdist
= simple_strtoul(str
, &str
, 0);
5829 __setup("hashdist=", set_hashdist
);
5833 * allocate a large system hash table from bootmem
5834 * - it is assumed that the hash table must contain an exact power-of-2
5835 * quantity of entries
5836 * - limit is the number of hash buckets, not the total allocation size
5838 void *__init
alloc_large_system_hash(const char *tablename
,
5839 unsigned long bucketsize
,
5840 unsigned long numentries
,
5843 unsigned int *_hash_shift
,
5844 unsigned int *_hash_mask
,
5845 unsigned long low_limit
,
5846 unsigned long high_limit
)
5848 unsigned long long max
= high_limit
;
5849 unsigned long log2qty
, size
;
5852 /* allow the kernel cmdline to have a say */
5854 /* round applicable memory size up to nearest megabyte */
5855 numentries
= nr_kernel_pages
;
5857 /* It isn't necessary when PAGE_SIZE >= 1MB */
5858 if (PAGE_SHIFT
< 20)
5859 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
5861 /* limit to 1 bucket per 2^scale bytes of low memory */
5862 if (scale
> PAGE_SHIFT
)
5863 numentries
>>= (scale
- PAGE_SHIFT
);
5865 numentries
<<= (PAGE_SHIFT
- scale
);
5867 /* Make sure we've got at least a 0-order allocation.. */
5868 if (unlikely(flags
& HASH_SMALL
)) {
5869 /* Makes no sense without HASH_EARLY */
5870 WARN_ON(!(flags
& HASH_EARLY
));
5871 if (!(numentries
>> *_hash_shift
)) {
5872 numentries
= 1UL << *_hash_shift
;
5873 BUG_ON(!numentries
);
5875 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5876 numentries
= PAGE_SIZE
/ bucketsize
;
5878 numentries
= roundup_pow_of_two(numentries
);
5880 /* limit allocation size to 1/16 total memory by default */
5882 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5883 do_div(max
, bucketsize
);
5885 max
= min(max
, 0x80000000ULL
);
5887 if (numentries
< low_limit
)
5888 numentries
= low_limit
;
5889 if (numentries
> max
)
5892 log2qty
= ilog2(numentries
);
5895 size
= bucketsize
<< log2qty
;
5896 if (flags
& HASH_EARLY
)
5897 table
= memblock_virt_alloc_nopanic(size
, 0);
5899 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5902 * If bucketsize is not a power-of-two, we may free
5903 * some pages at the end of hash table which
5904 * alloc_pages_exact() automatically does
5906 if (get_order(size
) < MAX_ORDER
) {
5907 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5908 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5911 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5914 panic("Failed to allocate %s hash table\n", tablename
);
5916 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5919 ilog2(size
) - PAGE_SHIFT
,
5923 *_hash_shift
= log2qty
;
5925 *_hash_mask
= (1 << log2qty
) - 1;
5930 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5931 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5934 #ifdef CONFIG_SPARSEMEM
5935 return __pfn_to_section(pfn
)->pageblock_flags
;
5937 return zone
->pageblock_flags
;
5938 #endif /* CONFIG_SPARSEMEM */
5941 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5943 #ifdef CONFIG_SPARSEMEM
5944 pfn
&= (PAGES_PER_SECTION
-1);
5945 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5947 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
5948 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5949 #endif /* CONFIG_SPARSEMEM */
5953 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5954 * @page: The page within the block of interest
5955 * @start_bitidx: The first bit of interest to retrieve
5956 * @end_bitidx: The last bit of interest
5957 * returns pageblock_bits flags
5959 unsigned long get_pageblock_flags_group(struct page
*page
,
5960 int start_bitidx
, int end_bitidx
)
5963 unsigned long *bitmap
;
5964 unsigned long pfn
, bitidx
;
5965 unsigned long flags
= 0;
5966 unsigned long value
= 1;
5968 zone
= page_zone(page
);
5969 pfn
= page_to_pfn(page
);
5970 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5971 bitidx
= pfn_to_bitidx(zone
, pfn
);
5973 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5974 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5981 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5982 * @page: The page within the block of interest
5983 * @start_bitidx: The first bit of interest
5984 * @end_bitidx: The last bit of interest
5985 * @flags: The flags to set
5987 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5988 int start_bitidx
, int end_bitidx
)
5991 unsigned long *bitmap
;
5992 unsigned long pfn
, bitidx
;
5993 unsigned long value
= 1;
5995 zone
= page_zone(page
);
5996 pfn
= page_to_pfn(page
);
5997 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5998 bitidx
= pfn_to_bitidx(zone
, pfn
);
5999 VM_BUG_ON(!zone_spans_pfn(zone
, pfn
));
6001 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
6003 __set_bit(bitidx
+ start_bitidx
, bitmap
);
6005 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
6009 * This function checks whether pageblock includes unmovable pages or not.
6010 * If @count is not zero, it is okay to include less @count unmovable pages
6012 * PageLRU check without isolation or lru_lock could race so that
6013 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6014 * expect this function should be exact.
6016 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6017 bool skip_hwpoisoned_pages
)
6019 unsigned long pfn
, iter
, found
;
6023 * For avoiding noise data, lru_add_drain_all() should be called
6024 * If ZONE_MOVABLE, the zone never contains unmovable pages
6026 if (zone_idx(zone
) == ZONE_MOVABLE
)
6028 mt
= get_pageblock_migratetype(page
);
6029 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6032 pfn
= page_to_pfn(page
);
6033 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6034 unsigned long check
= pfn
+ iter
;
6036 if (!pfn_valid_within(check
))
6039 page
= pfn_to_page(check
);
6042 * Hugepages are not in LRU lists, but they're movable.
6043 * We need not scan over tail pages bacause we don't
6044 * handle each tail page individually in migration.
6046 if (PageHuge(page
)) {
6047 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6052 * We can't use page_count without pin a page
6053 * because another CPU can free compound page.
6054 * This check already skips compound tails of THP
6055 * because their page->_count is zero at all time.
6057 if (!atomic_read(&page
->_count
)) {
6058 if (PageBuddy(page
))
6059 iter
+= (1 << page_order(page
)) - 1;
6064 * The HWPoisoned page may be not in buddy system, and
6065 * page_count() is not 0.
6067 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6073 * If there are RECLAIMABLE pages, we need to check it.
6074 * But now, memory offline itself doesn't call shrink_slab()
6075 * and it still to be fixed.
6078 * If the page is not RAM, page_count()should be 0.
6079 * we don't need more check. This is an _used_ not-movable page.
6081 * The problematic thing here is PG_reserved pages. PG_reserved
6082 * is set to both of a memory hole page and a _used_ kernel
6091 bool is_pageblock_removable_nolock(struct page
*page
)
6097 * We have to be careful here because we are iterating over memory
6098 * sections which are not zone aware so we might end up outside of
6099 * the zone but still within the section.
6100 * We have to take care about the node as well. If the node is offline
6101 * its NODE_DATA will be NULL - see page_zone.
6103 if (!node_online(page_to_nid(page
)))
6106 zone
= page_zone(page
);
6107 pfn
= page_to_pfn(page
);
6108 if (!zone_spans_pfn(zone
, pfn
))
6111 return !has_unmovable_pages(zone
, page
, 0, true);
6116 static unsigned long pfn_max_align_down(unsigned long pfn
)
6118 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6119 pageblock_nr_pages
) - 1);
6122 static unsigned long pfn_max_align_up(unsigned long pfn
)
6124 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6125 pageblock_nr_pages
));
6128 /* [start, end) must belong to a single zone. */
6129 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6130 unsigned long start
, unsigned long end
)
6132 /* This function is based on compact_zone() from compaction.c. */
6133 unsigned long nr_reclaimed
;
6134 unsigned long pfn
= start
;
6135 unsigned int tries
= 0;
6140 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6141 if (fatal_signal_pending(current
)) {
6146 if (list_empty(&cc
->migratepages
)) {
6147 cc
->nr_migratepages
= 0;
6148 pfn
= isolate_migratepages_range(cc
->zone
, cc
,
6155 } else if (++tries
== 5) {
6156 ret
= ret
< 0 ? ret
: -EBUSY
;
6160 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6162 cc
->nr_migratepages
-= nr_reclaimed
;
6164 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6165 0, MIGRATE_SYNC
, MR_CMA
);
6168 putback_movable_pages(&cc
->migratepages
);
6175 * alloc_contig_range() -- tries to allocate given range of pages
6176 * @start: start PFN to allocate
6177 * @end: one-past-the-last PFN to allocate
6178 * @migratetype: migratetype of the underlaying pageblocks (either
6179 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6180 * in range must have the same migratetype and it must
6181 * be either of the two.
6183 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6184 * aligned, however it's the caller's responsibility to guarantee that
6185 * we are the only thread that changes migrate type of pageblocks the
6188 * The PFN range must belong to a single zone.
6190 * Returns zero on success or negative error code. On success all
6191 * pages which PFN is in [start, end) are allocated for the caller and
6192 * need to be freed with free_contig_range().
6194 int alloc_contig_range(unsigned long start
, unsigned long end
,
6195 unsigned migratetype
)
6197 unsigned long outer_start
, outer_end
;
6200 struct compact_control cc
= {
6201 .nr_migratepages
= 0,
6203 .zone
= page_zone(pfn_to_page(start
)),
6205 .ignore_skip_hint
= true,
6207 INIT_LIST_HEAD(&cc
.migratepages
);
6210 * What we do here is we mark all pageblocks in range as
6211 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6212 * have different sizes, and due to the way page allocator
6213 * work, we align the range to biggest of the two pages so
6214 * that page allocator won't try to merge buddies from
6215 * different pageblocks and change MIGRATE_ISOLATE to some
6216 * other migration type.
6218 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6219 * migrate the pages from an unaligned range (ie. pages that
6220 * we are interested in). This will put all the pages in
6221 * range back to page allocator as MIGRATE_ISOLATE.
6223 * When this is done, we take the pages in range from page
6224 * allocator removing them from the buddy system. This way
6225 * page allocator will never consider using them.
6227 * This lets us mark the pageblocks back as
6228 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6229 * aligned range but not in the unaligned, original range are
6230 * put back to page allocator so that buddy can use them.
6233 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6234 pfn_max_align_up(end
), migratetype
,
6239 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6244 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6245 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6246 * more, all pages in [start, end) are free in page allocator.
6247 * What we are going to do is to allocate all pages from
6248 * [start, end) (that is remove them from page allocator).
6250 * The only problem is that pages at the beginning and at the
6251 * end of interesting range may be not aligned with pages that
6252 * page allocator holds, ie. they can be part of higher order
6253 * pages. Because of this, we reserve the bigger range and
6254 * once this is done free the pages we are not interested in.
6256 * We don't have to hold zone->lock here because the pages are
6257 * isolated thus they won't get removed from buddy.
6260 lru_add_drain_all();
6264 outer_start
= start
;
6265 while (!PageBuddy(pfn_to_page(outer_start
))) {
6266 if (++order
>= MAX_ORDER
) {
6270 outer_start
&= ~0UL << order
;
6273 /* Make sure the range is really isolated. */
6274 if (test_pages_isolated(outer_start
, end
, false)) {
6275 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6282 /* Grab isolated pages from freelists. */
6283 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6289 /* Free head and tail (if any) */
6290 if (start
!= outer_start
)
6291 free_contig_range(outer_start
, start
- outer_start
);
6292 if (end
!= outer_end
)
6293 free_contig_range(end
, outer_end
- end
);
6296 undo_isolate_page_range(pfn_max_align_down(start
),
6297 pfn_max_align_up(end
), migratetype
);
6301 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6303 unsigned int count
= 0;
6305 for (; nr_pages
--; pfn
++) {
6306 struct page
*page
= pfn_to_page(pfn
);
6308 count
+= page_count(page
) != 1;
6311 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6315 #ifdef CONFIG_MEMORY_HOTPLUG
6317 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6318 * page high values need to be recalulated.
6320 void __meminit
zone_pcp_update(struct zone
*zone
)
6323 mutex_lock(&pcp_batch_high_lock
);
6324 for_each_possible_cpu(cpu
)
6325 pageset_set_high_and_batch(zone
,
6326 per_cpu_ptr(zone
->pageset
, cpu
));
6327 mutex_unlock(&pcp_batch_high_lock
);
6331 void zone_pcp_reset(struct zone
*zone
)
6333 unsigned long flags
;
6335 struct per_cpu_pageset
*pset
;
6337 /* avoid races with drain_pages() */
6338 local_irq_save(flags
);
6339 if (zone
->pageset
!= &boot_pageset
) {
6340 for_each_online_cpu(cpu
) {
6341 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6342 drain_zonestat(zone
, pset
);
6344 free_percpu(zone
->pageset
);
6345 zone
->pageset
= &boot_pageset
;
6347 local_irq_restore(flags
);
6350 #ifdef CONFIG_MEMORY_HOTREMOVE
6352 * All pages in the range must be isolated before calling this.
6355 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6361 unsigned long flags
;
6362 /* find the first valid pfn */
6363 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6368 zone
= page_zone(pfn_to_page(pfn
));
6369 spin_lock_irqsave(&zone
->lock
, flags
);
6371 while (pfn
< end_pfn
) {
6372 if (!pfn_valid(pfn
)) {
6376 page
= pfn_to_page(pfn
);
6378 * The HWPoisoned page may be not in buddy system, and
6379 * page_count() is not 0.
6381 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6383 SetPageReserved(page
);
6387 BUG_ON(page_count(page
));
6388 BUG_ON(!PageBuddy(page
));
6389 order
= page_order(page
);
6390 #ifdef CONFIG_DEBUG_VM
6391 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6392 pfn
, 1 << order
, end_pfn
);
6394 list_del(&page
->lru
);
6395 rmv_page_order(page
);
6396 zone
->free_area
[order
].nr_free
--;
6397 for (i
= 0; i
< (1 << order
); i
++)
6398 SetPageReserved((page
+i
));
6399 pfn
+= (1 << order
);
6401 spin_unlock_irqrestore(&zone
->lock
, flags
);
6405 #ifdef CONFIG_MEMORY_FAILURE
6406 bool is_free_buddy_page(struct page
*page
)
6408 struct zone
*zone
= page_zone(page
);
6409 unsigned long pfn
= page_to_pfn(page
);
6410 unsigned long flags
;
6413 spin_lock_irqsave(&zone
->lock
, flags
);
6414 for (order
= 0; order
< MAX_ORDER
; order
++) {
6415 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6417 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6420 spin_unlock_irqrestore(&zone
->lock
, flags
);
6422 return order
< MAX_ORDER
;
6426 static const struct trace_print_flags pageflag_names
[] = {
6427 {1UL << PG_locked
, "locked" },
6428 {1UL << PG_error
, "error" },
6429 {1UL << PG_referenced
, "referenced" },
6430 {1UL << PG_uptodate
, "uptodate" },
6431 {1UL << PG_dirty
, "dirty" },
6432 {1UL << PG_lru
, "lru" },
6433 {1UL << PG_active
, "active" },
6434 {1UL << PG_slab
, "slab" },
6435 {1UL << PG_owner_priv_1
, "owner_priv_1" },
6436 {1UL << PG_arch_1
, "arch_1" },
6437 {1UL << PG_reserved
, "reserved" },
6438 {1UL << PG_private
, "private" },
6439 {1UL << PG_private_2
, "private_2" },
6440 {1UL << PG_writeback
, "writeback" },
6441 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6442 {1UL << PG_head
, "head" },
6443 {1UL << PG_tail
, "tail" },
6445 {1UL << PG_compound
, "compound" },
6447 {1UL << PG_swapcache
, "swapcache" },
6448 {1UL << PG_mappedtodisk
, "mappedtodisk" },
6449 {1UL << PG_reclaim
, "reclaim" },
6450 {1UL << PG_swapbacked
, "swapbacked" },
6451 {1UL << PG_unevictable
, "unevictable" },
6453 {1UL << PG_mlocked
, "mlocked" },
6455 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6456 {1UL << PG_uncached
, "uncached" },
6458 #ifdef CONFIG_MEMORY_FAILURE
6459 {1UL << PG_hwpoison
, "hwpoison" },
6461 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6462 {1UL << PG_compound_lock
, "compound_lock" },
6466 static void dump_page_flags(unsigned long flags
)
6468 const char *delim
= "";
6472 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names
) != __NR_PAGEFLAGS
);
6474 printk(KERN_ALERT
"page flags: %#lx(", flags
);
6476 /* remove zone id */
6477 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
6479 for (i
= 0; i
< ARRAY_SIZE(pageflag_names
) && flags
; i
++) {
6481 mask
= pageflag_names
[i
].mask
;
6482 if ((flags
& mask
) != mask
)
6486 printk("%s%s", delim
, pageflag_names
[i
].name
);
6490 /* check for left over flags */
6492 printk("%s%#lx", delim
, flags
);
6497 void dump_page(struct page
*page
)
6500 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6501 page
, atomic_read(&page
->_count
), page_mapcount(page
),
6502 page
->mapping
, page
->index
);
6503 dump_page_flags(page
->flags
);
6504 mem_cgroup_print_bad_page(page
);