2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/notifier.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/stop_machine.h>
46 #include <linux/sort.h>
47 #include <linux/pfn.h>
48 #include <linux/backing-dev.h>
49 #include <linux/fault-inject.h>
50 #include <linux/page-isolation.h>
51 #include <linux/page_cgroup.h>
52 #include <linux/debugobjects.h>
53 #include <linux/kmemleak.h>
54 #include <linux/compaction.h>
55 #include <trace/events/kmem.h>
56 #include <linux/ftrace_event.h>
57 #include <linux/memcontrol.h>
58 #include <linux/prefetch.h>
59 #include <linux/migrate.h>
60 #include <linux/page-debug-flags.h>
61 #include <linux/hugetlb.h>
62 #include <linux/sched/rt.h>
64 #include <asm/tlbflush.h>
65 #include <asm/div64.h>
68 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
69 DEFINE_PER_CPU(int, numa_node
);
70 EXPORT_PER_CPU_SYMBOL(numa_node
);
73 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
75 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
76 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
77 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
78 * defined in <linux/topology.h>.
80 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
81 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
85 * Array of node states.
87 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
88 [N_POSSIBLE
] = NODE_MASK_ALL
,
89 [N_ONLINE
] = { { [0] = 1UL } },
91 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
93 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
95 #ifdef CONFIG_MOVABLE_NODE
96 [N_MEMORY
] = { { [0] = 1UL } },
98 [N_CPU
] = { { [0] = 1UL } },
101 EXPORT_SYMBOL(node_states
);
103 unsigned long totalram_pages __read_mostly
;
104 unsigned long totalreserve_pages __read_mostly
;
106 * When calculating the number of globally allowed dirty pages, there
107 * is a certain number of per-zone reserves that should not be
108 * considered dirtyable memory. This is the sum of those reserves
109 * over all existing zones that contribute dirtyable memory.
111 unsigned long dirty_balance_reserve __read_mostly
;
113 static unsigned int boot_mode
= 0;
114 static int __init
setup_bootmode(char *str
)
116 printk("%s: boot_mode is %u\n", __func__
, boot_mode
);
117 if (get_option(&str
, &boot_mode
)) {
118 printk("%s: boot_mode is %u\n", __func__
, boot_mode
);
124 early_param("bootmode", setup_bootmode
);
126 int percpu_pagelist_fraction
;
127 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
129 #ifdef CONFIG_PM_SLEEP
131 * The following functions are used by the suspend/hibernate code to temporarily
132 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
133 * while devices are suspended. To avoid races with the suspend/hibernate code,
134 * they should always be called with pm_mutex held (gfp_allowed_mask also should
135 * only be modified with pm_mutex held, unless the suspend/hibernate code is
136 * guaranteed not to run in parallel with that modification).
139 static gfp_t saved_gfp_mask
;
141 void pm_restore_gfp_mask(void)
143 WARN_ON(!mutex_is_locked(&pm_mutex
));
144 if (saved_gfp_mask
) {
145 gfp_allowed_mask
= saved_gfp_mask
;
150 void pm_restrict_gfp_mask(void)
152 WARN_ON(!mutex_is_locked(&pm_mutex
));
153 WARN_ON(saved_gfp_mask
);
154 saved_gfp_mask
= gfp_allowed_mask
;
155 gfp_allowed_mask
&= ~GFP_IOFS
;
158 bool pm_suspended_storage(void)
160 if ((gfp_allowed_mask
& GFP_IOFS
) == GFP_IOFS
)
164 #endif /* CONFIG_PM_SLEEP */
166 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
167 int pageblock_order __read_mostly
;
170 static void __free_pages_ok(struct page
*page
, unsigned int order
);
173 * results with 256, 32 in the lowmem_reserve sysctl:
174 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
175 * 1G machine -> (16M dma, 784M normal, 224M high)
176 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
177 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
178 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
180 * TBD: should special case ZONE_DMA32 machines here - in those we normally
181 * don't need any ZONE_NORMAL reservation
183 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
184 #ifdef CONFIG_ZONE_DMA
187 #ifdef CONFIG_ZONE_DMA32
190 #ifdef CONFIG_HIGHMEM
196 EXPORT_SYMBOL(totalram_pages
);
198 static char * const zone_names
[MAX_NR_ZONES
] = {
199 #ifdef CONFIG_ZONE_DMA
202 #ifdef CONFIG_ZONE_DMA32
206 #ifdef CONFIG_HIGHMEM
213 * Try to keep at least this much lowmem free. Do not allow normal
214 * allocations below this point, only high priority ones. Automatically
215 * tuned according to the amount of memory in the system.
217 int min_free_kbytes
= 1024;
218 int min_free_order_shift
= 1;
221 * Extra memory for the system to try freeing. Used to temporarily
222 * free memory, to make space for new workloads. Anyone can allocate
223 * down to the min watermarks controlled by min_free_kbytes above.
225 int extra_free_kbytes
= 0;
227 static unsigned long __meminitdata nr_kernel_pages
;
228 static unsigned long __meminitdata nr_all_pages
;
229 static unsigned long __meminitdata dma_reserve
;
231 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
232 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
233 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
234 static unsigned long __initdata required_kernelcore
;
235 static unsigned long __initdata required_movablecore
;
236 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
238 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
240 EXPORT_SYMBOL(movable_zone
);
241 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
244 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
245 int nr_online_nodes __read_mostly
= 1;
246 EXPORT_SYMBOL(nr_node_ids
);
247 EXPORT_SYMBOL(nr_online_nodes
);
250 int page_group_by_mobility_disabled __read_mostly
;
252 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
255 if (unlikely(page_group_by_mobility_disabled
))
256 migratetype
= MIGRATE_UNMOVABLE
;
258 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
259 PB_migrate
, PB_migrate_end
);
262 bool oom_killer_disabled __read_mostly
;
264 #ifdef CONFIG_DEBUG_VM
265 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
269 unsigned long pfn
= page_to_pfn(page
);
270 unsigned long sp
, start_pfn
;
273 seq
= zone_span_seqbegin(zone
);
274 start_pfn
= zone
->zone_start_pfn
;
275 sp
= zone
->spanned_pages
;
276 if (!zone_spans_pfn(zone
, pfn
))
278 } while (zone_span_seqretry(zone
, seq
));
281 pr_err("page %lu outside zone [ %lu - %lu ]\n",
282 pfn
, start_pfn
, start_pfn
+ sp
);
287 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
289 if (!pfn_valid_within(page_to_pfn(page
)))
291 if (zone
!= page_zone(page
))
297 * Temporary debugging check for pages not lying within a given zone.
299 static int bad_range(struct zone
*zone
, struct page
*page
)
301 if (page_outside_zone_boundaries(zone
, page
))
303 if (!page_is_consistent(zone
, page
))
309 static inline int bad_range(struct zone
*zone
, struct page
*page
)
315 static void bad_page(struct page
*page
)
317 static unsigned long resume
;
318 static unsigned long nr_shown
;
319 static unsigned long nr_unshown
;
321 /* Don't complain about poisoned pages */
322 if (PageHWPoison(page
)) {
323 page_mapcount_reset(page
); /* remove PageBuddy */
328 * Allow a burst of 60 reports, then keep quiet for that minute;
329 * or allow a steady drip of one report per second.
331 if (nr_shown
== 60) {
332 if (time_before(jiffies
, resume
)) {
338 "BUG: Bad page state: %lu messages suppressed\n",
345 resume
= jiffies
+ 60 * HZ
;
347 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
348 current
->comm
, page_to_pfn(page
));
354 /* Leave bad fields for debug, except PageBuddy could make trouble */
355 page_mapcount_reset(page
); /* remove PageBuddy */
356 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
360 * Higher-order pages are called "compound pages". They are structured thusly:
362 * The first PAGE_SIZE page is called the "head page".
364 * The remaining PAGE_SIZE pages are called "tail pages".
366 * All pages have PG_compound set. All tail pages have their ->first_page
367 * pointing at the head page.
369 * The first tail page's ->lru.next holds the address of the compound page's
370 * put_page() function. Its ->lru.prev holds the order of allocation.
371 * This usage means that zero-order pages may not be compound.
374 static void free_compound_page(struct page
*page
)
376 __free_pages_ok(page
, compound_order(page
));
379 void prep_compound_page(struct page
*page
, unsigned long order
)
382 int nr_pages
= 1 << order
;
384 set_compound_page_dtor(page
, free_compound_page
);
385 set_compound_order(page
, order
);
387 for (i
= 1; i
< nr_pages
; i
++) {
388 struct page
*p
= page
+ i
;
389 set_page_count(p
, 0);
390 p
->first_page
= page
;
391 /* Make sure p->first_page is always valid for PageTail() */
397 /* update __split_huge_page_refcount if you change this function */
398 static int destroy_compound_page(struct page
*page
, unsigned long order
)
401 int nr_pages
= 1 << order
;
404 if (unlikely(compound_order(page
) != order
)) {
409 __ClearPageHead(page
);
411 for (i
= 1; i
< nr_pages
; i
++) {
412 struct page
*p
= page
+ i
;
414 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
424 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
429 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
430 * and __GFP_HIGHMEM from hard or soft interrupt context.
432 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
433 for (i
= 0; i
< (1 << order
); i
++)
434 clear_highpage(page
+ i
);
437 #ifdef CONFIG_DEBUG_PAGEALLOC
438 unsigned int _debug_guardpage_minorder
;
440 static int __init
debug_guardpage_minorder_setup(char *buf
)
444 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
445 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
448 _debug_guardpage_minorder
= res
;
449 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
452 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
454 static inline void set_page_guard_flag(struct page
*page
)
456 __set_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
459 static inline void clear_page_guard_flag(struct page
*page
)
461 __clear_bit(PAGE_DEBUG_FLAG_GUARD
, &page
->debug_flags
);
464 static inline void set_page_guard_flag(struct page
*page
) { }
465 static inline void clear_page_guard_flag(struct page
*page
) { }
468 static inline void set_page_order(struct page
*page
, int order
)
470 set_page_private(page
, order
);
471 __SetPageBuddy(page
);
474 static inline void rmv_page_order(struct page
*page
)
476 __ClearPageBuddy(page
);
477 set_page_private(page
, 0);
481 * Locate the struct page for both the matching buddy in our
482 * pair (buddy1) and the combined O(n+1) page they form (page).
484 * 1) Any buddy B1 will have an order O twin B2 which satisfies
485 * the following equation:
487 * For example, if the starting buddy (buddy2) is #8 its order
489 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
491 * 2) Any buddy B will have an order O+1 parent P which
492 * satisfies the following equation:
495 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
497 static inline unsigned long
498 __find_buddy_index(unsigned long page_idx
, unsigned int order
)
500 return page_idx
^ (1 << order
);
504 * This function checks whether a page is free && is the buddy
505 * we can do coalesce a page and its buddy if
506 * (a) the buddy is not in a hole &&
507 * (b) the buddy is in the buddy system &&
508 * (c) a page and its buddy have the same order &&
509 * (d) a page and its buddy are in the same zone.
511 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
512 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
514 * For recording page's order, we use page_private(page).
516 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
519 if (!pfn_valid_within(page_to_pfn(buddy
)))
522 if (page_zone_id(page
) != page_zone_id(buddy
))
525 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
526 VM_BUG_ON(page_count(buddy
) != 0);
530 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
531 VM_BUG_ON(page_count(buddy
) != 0);
538 * Freeing function for a buddy system allocator.
540 * The concept of a buddy system is to maintain direct-mapped table
541 * (containing bit values) for memory blocks of various "orders".
542 * The bottom level table contains the map for the smallest allocatable
543 * units of memory (here, pages), and each level above it describes
544 * pairs of units from the levels below, hence, "buddies".
545 * At a high level, all that happens here is marking the table entry
546 * at the bottom level available, and propagating the changes upward
547 * as necessary, plus some accounting needed to play nicely with other
548 * parts of the VM system.
549 * At each level, we keep a list of pages, which are heads of continuous
550 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
551 * order is recorded in page_private(page) field.
552 * So when we are allocating or freeing one, we can derive the state of the
553 * other. That is, if we allocate a small block, and both were
554 * free, the remainder of the region must be split into blocks.
555 * If a block is freed, and its buddy is also free, then this
556 * triggers coalescing into a block of larger size.
561 static inline void __free_one_page(struct page
*page
,
562 struct zone
*zone
, unsigned int order
,
565 unsigned long page_idx
;
566 unsigned long combined_idx
;
567 unsigned long uninitialized_var(buddy_idx
);
570 VM_BUG_ON(!zone_is_initialized(zone
));
572 if (unlikely(PageCompound(page
)))
573 if (unlikely(destroy_compound_page(page
, order
)))
576 VM_BUG_ON(migratetype
== -1);
578 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
580 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
581 VM_BUG_ON(bad_range(zone
, page
));
583 while (order
< MAX_ORDER
-1) {
584 buddy_idx
= __find_buddy_index(page_idx
, order
);
585 buddy
= page
+ (buddy_idx
- page_idx
);
586 if (!page_is_buddy(page
, buddy
, order
))
589 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
590 * merge with it and move up one order.
592 if (page_is_guard(buddy
)) {
593 clear_page_guard_flag(buddy
);
594 set_page_private(page
, 0);
595 __mod_zone_freepage_state(zone
, 1 << order
,
598 list_del(&buddy
->lru
);
599 zone
->free_area
[order
].nr_free
--;
600 rmv_page_order(buddy
);
602 combined_idx
= buddy_idx
& page_idx
;
603 page
= page
+ (combined_idx
- page_idx
);
604 page_idx
= combined_idx
;
607 set_page_order(page
, order
);
610 * If this is not the largest possible page, check if the buddy
611 * of the next-highest order is free. If it is, it's possible
612 * that pages are being freed that will coalesce soon. In case,
613 * that is happening, add the free page to the tail of the list
614 * so it's less likely to be used soon and more likely to be merged
615 * as a higher order page
617 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
618 struct page
*higher_page
, *higher_buddy
;
619 combined_idx
= buddy_idx
& page_idx
;
620 higher_page
= page
+ (combined_idx
- page_idx
);
621 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
622 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
623 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
624 list_add_tail(&page
->lru
,
625 &zone
->free_area
[order
].free_list
[migratetype
]);
630 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
632 zone
->free_area
[order
].nr_free
++;
635 static inline int free_pages_check(struct page
*page
)
637 if (unlikely(page_mapcount(page
) |
638 (page
->mapping
!= NULL
) |
639 (atomic_read(&page
->_count
) != 0) |
640 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
) |
641 (mem_cgroup_bad_page_check(page
)))) {
645 page_nid_reset_last(page
);
646 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
647 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
652 * Frees a number of pages from the PCP lists
653 * Assumes all pages on list are in same zone, and of same order.
654 * count is the number of pages to free.
656 * If the zone was previously in an "all pages pinned" state then look to
657 * see if this freeing clears that state.
659 * And clear the zone's pages_scanned counter, to hold off the "all pages are
660 * pinned" detection logic.
662 static void free_pcppages_bulk(struct zone
*zone
, int count
,
663 struct per_cpu_pages
*pcp
)
669 spin_lock(&zone
->lock
);
670 zone
->all_unreclaimable
= 0;
671 zone
->pages_scanned
= 0;
675 struct list_head
*list
;
678 * Remove pages from lists in a round-robin fashion. A
679 * batch_free count is maintained that is incremented when an
680 * empty list is encountered. This is so more pages are freed
681 * off fuller lists instead of spinning excessively around empty
686 if (++migratetype
== MIGRATE_PCPTYPES
)
688 list
= &pcp
->lists
[migratetype
];
689 } while (list_empty(list
));
691 /* This is the only non-empty list. Free them all. */
692 if (batch_free
== MIGRATE_PCPTYPES
)
693 batch_free
= to_free
;
696 int mt
; /* migratetype of the to-be-freed page */
698 page
= list_entry(list
->prev
, struct page
, lru
);
699 /* must delete as __free_one_page list manipulates */
700 list_del(&page
->lru
);
701 mt
= get_freepage_migratetype(page
);
702 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
703 __free_one_page(page
, zone
, 0, mt
);
704 trace_mm_page_pcpu_drain(page
, 0, mt
);
705 if (likely(!is_migrate_isolate_page(page
))) {
706 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1);
707 if (is_migrate_cma(mt
))
708 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
, 1);
710 } while (--to_free
&& --batch_free
&& !list_empty(list
));
712 spin_unlock(&zone
->lock
);
715 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
718 spin_lock(&zone
->lock
);
719 zone
->all_unreclaimable
= 0;
720 zone
->pages_scanned
= 0;
722 __free_one_page(page
, zone
, order
, migratetype
);
723 if (unlikely(!is_migrate_isolate(migratetype
)))
724 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
725 spin_unlock(&zone
->lock
);
728 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
733 trace_mm_page_free(page
, order
);
734 kmemcheck_free_shadow(page
, order
);
737 page
->mapping
= NULL
;
738 for (i
= 0; i
< (1 << order
); i
++)
739 bad
+= free_pages_check(page
+ i
);
743 if (!PageHighMem(page
)) {
744 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
745 debug_check_no_obj_freed(page_address(page
),
748 arch_free_page(page
, order
);
749 kernel_map_pages(page
, 1 << order
, 0);
754 static void __free_pages_ok(struct page
*page
, unsigned int order
)
759 if (!free_pages_prepare(page
, order
))
762 local_irq_save(flags
);
763 __count_vm_events(PGFREE
, 1 << order
);
764 migratetype
= get_pageblock_migratetype(page
);
765 set_freepage_migratetype(page
, migratetype
);
766 free_one_page(page_zone(page
), page
, order
, migratetype
);
767 local_irq_restore(flags
);
771 * Read access to zone->managed_pages is safe because it's unsigned long,
772 * but we still need to serialize writers. Currently all callers of
773 * __free_pages_bootmem() except put_page_bootmem() should only be used
774 * at boot time. So for shorter boot time, we shift the burden to
775 * put_page_bootmem() to serialize writers.
777 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
779 unsigned int nr_pages
= 1 << order
;
783 for (loop
= 0; loop
< nr_pages
; loop
++) {
784 struct page
*p
= &page
[loop
];
786 if (loop
+ 1 < nr_pages
)
788 __ClearPageReserved(p
);
789 set_page_count(p
, 0);
792 page_zone(page
)->managed_pages
+= 1 << order
;
793 set_page_refcounted(page
);
794 __free_pages(page
, order
);
798 bool is_cma_pageblock(struct page
*page
)
800 return get_pageblock_migratetype(page
) == MIGRATE_CMA
;
803 /* Free whole pageblock and set it's migration type to MIGRATE_CMA. */
804 void __init
init_cma_reserved_pageblock(struct page
*page
)
806 unsigned i
= pageblock_nr_pages
;
807 struct page
*p
= page
;
810 __ClearPageReserved(p
);
811 set_page_count(p
, 0);
814 set_page_refcounted(page
);
815 set_pageblock_migratetype(page
, MIGRATE_CMA
);
816 __free_pages(page
, pageblock_order
);
817 totalram_pages
+= pageblock_nr_pages
;
818 #ifdef CONFIG_HIGHMEM
819 if (PageHighMem(page
))
820 totalhigh_pages
+= pageblock_nr_pages
;
826 * The order of subdivision here is critical for the IO subsystem.
827 * Please do not alter this order without good reasons and regression
828 * testing. Specifically, as large blocks of memory are subdivided,
829 * the order in which smaller blocks are delivered depends on the order
830 * they're subdivided in this function. This is the primary factor
831 * influencing the order in which pages are delivered to the IO
832 * subsystem according to empirical testing, and this is also justified
833 * by considering the behavior of a buddy system containing a single
834 * large block of memory acted on by a series of small allocations.
835 * This behavior is a critical factor in sglist merging's success.
839 static inline void expand(struct zone
*zone
, struct page
*page
,
840 int low
, int high
, struct free_area
*area
,
843 unsigned long size
= 1 << high
;
849 VM_BUG_ON(bad_range(zone
, &page
[size
]));
851 #ifdef CONFIG_DEBUG_PAGEALLOC
852 if (high
< debug_guardpage_minorder()) {
854 * Mark as guard pages (or page), that will allow to
855 * merge back to allocator when buddy will be freed.
856 * Corresponding page table entries will not be touched,
857 * pages will stay not present in virtual address space
859 INIT_LIST_HEAD(&page
[size
].lru
);
860 set_page_guard_flag(&page
[size
]);
861 set_page_private(&page
[size
], high
);
862 /* Guard pages are not available for any usage */
863 __mod_zone_freepage_state(zone
, -(1 << high
),
868 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
870 set_page_order(&page
[size
], high
);
875 * This page is about to be returned from the page allocator
877 static inline int check_new_page(struct page
*page
)
879 if (unlikely(page_mapcount(page
) |
880 (page
->mapping
!= NULL
) |
881 (atomic_read(&page
->_count
) != 0) |
882 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
) |
883 (mem_cgroup_bad_page_check(page
)))) {
890 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
894 for (i
= 0; i
< (1 << order
); i
++) {
895 struct page
*p
= page
+ i
;
896 if (unlikely(check_new_page(p
)))
900 set_page_private(page
, 0);
901 set_page_refcounted(page
);
903 arch_alloc_page(page
, order
);
904 kernel_map_pages(page
, 1 << order
, 1);
906 if (gfp_flags
& __GFP_ZERO
)
907 prep_zero_page(page
, order
, gfp_flags
);
909 if (order
&& (gfp_flags
& __GFP_COMP
))
910 prep_compound_page(page
, order
);
916 * Go through the free lists for the given migratetype and remove
917 * the smallest available page from the freelists
920 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
923 unsigned int current_order
;
924 struct free_area
* area
;
927 /* Find a page of the appropriate size in the preferred list */
928 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
929 area
= &(zone
->free_area
[current_order
]);
930 if (list_empty(&area
->free_list
[migratetype
]))
933 page
= list_entry(area
->free_list
[migratetype
].next
,
935 list_del(&page
->lru
);
936 rmv_page_order(page
);
938 expand(zone
, page
, order
, current_order
, area
, migratetype
);
947 * This array describes the order lists are fallen back to when
948 * the free lists for the desirable migrate type are depleted
950 static int fallbacks
[MIGRATE_TYPES
][4] = {
951 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
952 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
954 [MIGRATE_MOVABLE
] = { MIGRATE_CMA
, MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
955 [MIGRATE_CMA
] = { MIGRATE_RESERVE
}, /* Never used */
957 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
959 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
}, /* Never used */
960 #ifdef CONFIG_MEMORY_ISOLATION
961 [MIGRATE_ISOLATE
] = { MIGRATE_RESERVE
}, /* Never used */
965 int *get_migratetype_fallbacks(int mtype
)
967 return fallbacks
[mtype
];
971 * Move the free pages in a range to the free lists of the requested type.
972 * Note that start_page and end_pages are not aligned on a pageblock
973 * boundary. If alignment is required, use move_freepages_block()
975 int move_freepages(struct zone
*zone
,
976 struct page
*start_page
, struct page
*end_page
,
983 #ifndef CONFIG_HOLES_IN_ZONE
985 * page_zone is not safe to call in this context when
986 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
987 * anyway as we check zone boundaries in move_freepages_block().
988 * Remove at a later date when no bug reports exist related to
989 * grouping pages by mobility
991 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
994 for (page
= start_page
; page
<= end_page
;) {
995 /* Make sure we are not inadvertently changing nodes */
996 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
998 if (!pfn_valid_within(page_to_pfn(page
))) {
1003 if (!PageBuddy(page
)) {
1008 order
= page_order(page
);
1009 list_move(&page
->lru
,
1010 &zone
->free_area
[order
].free_list
[migratetype
]);
1011 set_freepage_migratetype(page
, migratetype
);
1013 pages_moved
+= 1 << order
;
1019 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1022 unsigned long start_pfn
, end_pfn
;
1023 struct page
*start_page
, *end_page
;
1025 start_pfn
= page_to_pfn(page
);
1026 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1027 start_page
= pfn_to_page(start_pfn
);
1028 end_page
= start_page
+ pageblock_nr_pages
- 1;
1029 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1031 /* Do not cross zone boundaries */
1032 if (!zone_spans_pfn(zone
, start_pfn
))
1034 if (!zone_spans_pfn(zone
, end_pfn
))
1037 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1040 static void change_pageblock_range(struct page
*pageblock_page
,
1041 int start_order
, int migratetype
)
1043 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1045 while (nr_pageblocks
--) {
1046 set_pageblock_migratetype(pageblock_page
, migratetype
);
1047 pageblock_page
+= pageblock_nr_pages
;
1051 /* Remove an element from the buddy allocator from the fallback list */
1052 static inline struct page
*
1053 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
1055 struct free_area
* area
;
1060 /* Find the largest possible block of pages in the other list */
1061 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
1064 migratetype
= fallbacks
[start_migratetype
][i
];
1066 /* MIGRATE_RESERVE handled later if necessary */
1067 if (migratetype
== MIGRATE_RESERVE
)
1070 area
= &(zone
->free_area
[current_order
]);
1071 if (list_empty(&area
->free_list
[migratetype
]))
1074 page
= list_entry(area
->free_list
[migratetype
].next
,
1079 * If breaking a large block of pages, move all free
1080 * pages to the preferred allocation list. If falling
1081 * back for a reclaimable kernel allocation, be more
1082 * aggressive about taking ownership of free pages
1084 * On the other hand, never change migration
1085 * type of MIGRATE_CMA pageblocks nor move CMA
1086 * pages on different free lists. We don't
1087 * want unmovable pages to be allocated from
1088 * MIGRATE_CMA areas.
1090 if (!is_migrate_cma(migratetype
) &&
1091 (unlikely(current_order
>= pageblock_order
/ 2) ||
1092 start_migratetype
== MIGRATE_RECLAIMABLE
||
1093 start_migratetype
== MIGRATE_UNMOVABLE
||
1094 start_migratetype
== MIGRATE_MOVABLE
||
1095 page_group_by_mobility_disabled
)) {
1097 pages
= move_freepages_block(zone
, page
,
1100 /* Claim the whole block if over half of it is free */
1101 if (pages
>= (1 << (pageblock_order
-1)) ||
1102 start_migratetype
== MIGRATE_MOVABLE
||
1103 page_group_by_mobility_disabled
)
1104 set_pageblock_migratetype(page
,
1107 migratetype
= start_migratetype
;
1110 /* Remove the page from the freelists */
1111 list_del(&page
->lru
);
1112 rmv_page_order(page
);
1114 /* Take ownership for orders >= pageblock_order */
1115 if (current_order
>= pageblock_order
&&
1116 !is_migrate_cma(migratetype
))
1117 change_pageblock_range(page
, current_order
,
1120 expand(zone
, page
, order
, current_order
, area
,
1121 is_migrate_cma(migratetype
)
1122 ? migratetype
: start_migratetype
);
1124 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1125 start_migratetype
, migratetype
);
1135 * Do the hard work of removing an element from the buddy allocator.
1136 * Call me with the zone->lock already held.
1138 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1144 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1146 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1147 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1150 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1151 * is used because __rmqueue_smallest is an inline function
1152 * and we want just one call site
1155 migratetype
= MIGRATE_RESERVE
;
1160 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1164 static struct page
*__rmqueue_cma(struct zone
*zone
, unsigned int order
,
1167 struct page
*page
= 0;
1169 if (migratetype
== MIGRATE_MOVABLE
&& !zone
->cma_alloc
) {
1170 page
= __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1172 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1176 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1179 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1180 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1183 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1184 * is used because __rmqueue_smallest is an inline function
1185 * and we want just one call site
1188 migratetype
= MIGRATE_RESERVE
;
1193 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1198 * Obtain a specified number of elements from the buddy allocator, all under
1199 * a single hold of the lock, for efficiency. Add them to the supplied list.
1200 * Returns the number of new pages which were placed at *list.
1202 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1203 unsigned long count
, struct list_head
*list
,
1204 int migratetype
, int cold
, int cma
)
1206 int mt
= migratetype
, i
;
1208 spin_lock(&zone
->lock
);
1209 for (i
= 0; i
< count
; ++i
) {
1212 page
= __rmqueue_cma(zone
, order
, migratetype
);
1214 page
= __rmqueue(zone
, order
, migratetype
);
1215 if (unlikely(page
== NULL
))
1219 * Split buddy pages returned by expand() are received here
1220 * in physical page order. The page is added to the callers and
1221 * list and the list head then moves forward. From the callers
1222 * perspective, the linked list is ordered by page number in
1223 * some conditions. This is useful for IO devices that can
1224 * merge IO requests if the physical pages are ordered
1227 if (likely(cold
== 0))
1228 list_add(&page
->lru
, list
);
1230 list_add_tail(&page
->lru
, list
);
1231 if (IS_ENABLED(CONFIG_CMA
)) {
1232 mt
= get_pageblock_migratetype(page
);
1233 if (!is_migrate_cma(mt
) && !is_migrate_isolate(mt
))
1236 set_freepage_migratetype(page
, mt
);
1238 if (is_migrate_cma(mt
))
1239 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1242 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1243 spin_unlock(&zone
->lock
);
1249 * Called from the vmstat counter updater to drain pagesets of this
1250 * currently executing processor on remote nodes after they have
1253 * Note that this function must be called with the thread pinned to
1254 * a single processor.
1256 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1258 unsigned long flags
;
1261 local_irq_save(flags
);
1262 if (pcp
->count
>= pcp
->batch
)
1263 to_drain
= pcp
->batch
;
1265 to_drain
= pcp
->count
;
1267 free_pcppages_bulk(zone
, to_drain
, pcp
);
1268 pcp
->count
-= to_drain
;
1270 local_irq_restore(flags
);
1275 * Drain pages of the indicated processor.
1277 * The processor must either be the current processor and the
1278 * thread pinned to the current processor or a processor that
1281 static void drain_pages(unsigned int cpu
)
1283 unsigned long flags
;
1286 for_each_populated_zone(zone
) {
1287 struct per_cpu_pageset
*pset
;
1288 struct per_cpu_pages
*pcp
;
1290 local_irq_save(flags
);
1291 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1295 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1298 local_irq_restore(flags
);
1303 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1305 void drain_local_pages(void *arg
)
1307 drain_pages(smp_processor_id());
1311 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1313 * Note that this code is protected against sending an IPI to an offline
1314 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1315 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1316 * nothing keeps CPUs from showing up after we populated the cpumask and
1317 * before the call to on_each_cpu_mask().
1319 void drain_all_pages(void)
1322 struct per_cpu_pageset
*pcp
;
1326 * Allocate in the BSS so we wont require allocation in
1327 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1329 static cpumask_t cpus_with_pcps
;
1332 * We don't care about racing with CPU hotplug event
1333 * as offline notification will cause the notified
1334 * cpu to drain that CPU pcps and on_each_cpu_mask
1335 * disables preemption as part of its processing
1337 for_each_online_cpu(cpu
) {
1338 bool has_pcps
= false;
1339 for_each_populated_zone(zone
) {
1340 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1341 if (pcp
->pcp
.count
) {
1347 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1349 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1351 on_each_cpu_mask(&cpus_with_pcps
, drain_local_pages
, NULL
, 1);
1354 #ifdef CONFIG_HIBERNATION
1356 void mark_free_pages(struct zone
*zone
)
1358 unsigned long pfn
, max_zone_pfn
;
1359 unsigned long flags
;
1361 struct list_head
*curr
;
1363 if (!zone
->spanned_pages
)
1366 spin_lock_irqsave(&zone
->lock
, flags
);
1368 max_zone_pfn
= zone_end_pfn(zone
);
1369 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1370 if (pfn_valid(pfn
)) {
1371 struct page
*page
= pfn_to_page(pfn
);
1373 if (!swsusp_page_is_forbidden(page
))
1374 swsusp_unset_page_free(page
);
1377 for_each_migratetype_order(order
, t
) {
1378 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1381 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1382 for (i
= 0; i
< (1UL << order
); i
++)
1383 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1386 spin_unlock_irqrestore(&zone
->lock
, flags
);
1388 #endif /* CONFIG_PM */
1391 * Free a 0-order page
1392 * cold == 1 ? free a cold page : free a hot page
1394 void free_hot_cold_page(struct page
*page
, int cold
)
1396 struct zone
*zone
= page_zone(page
);
1397 struct per_cpu_pages
*pcp
;
1398 unsigned long flags
;
1401 if (!free_pages_prepare(page
, 0))
1404 migratetype
= get_pageblock_migratetype(page
);
1405 set_freepage_migratetype(page
, migratetype
);
1406 local_irq_save(flags
);
1407 __count_vm_event(PGFREE
);
1410 * We only track unmovable, reclaimable and movable on pcp lists.
1411 * Free ISOLATE pages back to the allocator because they are being
1412 * offlined but treat RESERVE as movable pages so we can get those
1413 * areas back if necessary. Otherwise, we may have to free
1414 * excessively into the page allocator
1416 if (migratetype
>= MIGRATE_PCPTYPES
) {
1417 if (unlikely(is_migrate_isolate(migratetype
))) {
1418 free_one_page(zone
, page
, 0, migratetype
);
1421 migratetype
= MIGRATE_MOVABLE
;
1424 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1426 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1428 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1430 if (pcp
->count
>= pcp
->high
) {
1431 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1432 pcp
->count
-= pcp
->batch
;
1436 local_irq_restore(flags
);
1440 * Free a list of 0-order pages
1442 void free_hot_cold_page_list(struct list_head
*list
, int cold
)
1444 struct page
*page
, *next
;
1446 list_for_each_entry_safe(page
, next
, list
, lru
) {
1447 trace_mm_page_free_batched(page
, cold
);
1448 free_hot_cold_page(page
, cold
);
1453 * split_page takes a non-compound higher-order page, and splits it into
1454 * n (1<<order) sub-pages: page[0..n]
1455 * Each sub-page must be freed individually.
1457 * Note: this is probably too low level an operation for use in drivers.
1458 * Please consult with lkml before using this in your driver.
1460 void split_page(struct page
*page
, unsigned int order
)
1464 VM_BUG_ON(PageCompound(page
));
1465 VM_BUG_ON(!page_count(page
));
1467 #ifdef CONFIG_KMEMCHECK
1469 * Split shadow pages too, because free(page[0]) would
1470 * otherwise free the whole shadow.
1472 if (kmemcheck_page_is_tracked(page
))
1473 split_page(virt_to_page(page
[0].shadow
), order
);
1476 for (i
= 1; i
< (1 << order
); i
++)
1477 set_page_refcounted(page
+ i
);
1479 EXPORT_SYMBOL_GPL(split_page
);
1481 static int __isolate_free_page(struct page
*page
, unsigned int order
)
1483 unsigned long watermark
;
1487 BUG_ON(!PageBuddy(page
));
1489 zone
= page_zone(page
);
1490 mt
= get_pageblock_migratetype(page
);
1492 if (!is_migrate_isolate(mt
)) {
1493 /* Obey watermarks as if the page was being allocated */
1494 watermark
= low_wmark_pages(zone
) + (1 << order
);
1495 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1498 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
1501 /* Remove page from free list */
1502 list_del(&page
->lru
);
1503 zone
->free_area
[order
].nr_free
--;
1504 rmv_page_order(page
);
1506 /* Set the pageblock if the isolated page is at least a pageblock */
1507 if (order
>= pageblock_order
- 1) {
1508 struct page
*endpage
= page
+ (1 << order
) - 1;
1509 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
1510 int mt
= get_pageblock_migratetype(page
);
1511 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
1512 set_pageblock_migratetype(page
,
1517 return 1UL << order
;
1521 * Similar to split_page except the page is already free. As this is only
1522 * being used for migration, the migratetype of the block also changes.
1523 * As this is called with interrupts disabled, the caller is responsible
1524 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1527 * Note: this is probably too low level an operation for use in drivers.
1528 * Please consult with lkml before using this in your driver.
1530 int split_free_page(struct page
*page
)
1535 order
= page_order(page
);
1537 nr_pages
= __isolate_free_page(page
, order
);
1541 /* Split into individual pages */
1542 set_page_refcounted(page
);
1543 split_page(page
, order
);
1548 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1549 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1553 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1554 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1557 unsigned long flags
;
1559 int cold
= !!(gfp_flags
& __GFP_COLD
);
1562 if (likely(order
== 0)) {
1563 struct per_cpu_pages
*pcp
;
1564 struct list_head
*list
;
1566 local_irq_save(flags
);
1567 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1568 list
= &pcp
->lists
[migratetype
];
1569 if (list_empty(list
)) {
1570 pcp
->count
+= rmqueue_bulk(zone
, 0,
1573 gfp_flags
& __GFP_CMA
);
1574 if (unlikely(list_empty(list
)))
1579 page
= list_entry(list
->prev
, struct page
, lru
);
1581 page
= list_entry(list
->next
, struct page
, lru
);
1583 list_del(&page
->lru
);
1586 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1588 * __GFP_NOFAIL is not to be used in new code.
1590 * All __GFP_NOFAIL callers should be fixed so that they
1591 * properly detect and handle allocation failures.
1593 * We most definitely don't want callers attempting to
1594 * allocate greater than order-1 page units with
1597 WARN_ON_ONCE(order
> 1);
1599 spin_lock_irqsave(&zone
->lock
, flags
);
1600 if (gfp_flags
& __GFP_CMA
)
1601 page
= __rmqueue_cma(zone
, order
, migratetype
);
1603 page
= __rmqueue(zone
, order
, migratetype
);
1604 spin_unlock(&zone
->lock
);
1607 __mod_zone_freepage_state(zone
, -(1 << order
),
1608 get_pageblock_migratetype(page
));
1611 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1612 zone_statistics(preferred_zone
, zone
, gfp_flags
);
1613 local_irq_restore(flags
);
1615 VM_BUG_ON(bad_range(zone
, page
));
1616 if (prep_new_page(page
, order
, gfp_flags
))
1621 local_irq_restore(flags
);
1625 #ifdef CONFIG_FAIL_PAGE_ALLOC
1628 struct fault_attr attr
;
1630 u32 ignore_gfp_highmem
;
1631 u32 ignore_gfp_wait
;
1633 } fail_page_alloc
= {
1634 .attr
= FAULT_ATTR_INITIALIZER
,
1635 .ignore_gfp_wait
= 1,
1636 .ignore_gfp_highmem
= 1,
1640 static int __init
setup_fail_page_alloc(char *str
)
1642 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1644 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1646 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1648 if (order
< fail_page_alloc
.min_order
)
1650 if (gfp_mask
& __GFP_NOFAIL
)
1652 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1654 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1657 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1660 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1662 static int __init
fail_page_alloc_debugfs(void)
1664 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1667 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
1668 &fail_page_alloc
.attr
);
1670 return PTR_ERR(dir
);
1672 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1673 &fail_page_alloc
.ignore_gfp_wait
))
1675 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1676 &fail_page_alloc
.ignore_gfp_highmem
))
1678 if (!debugfs_create_u32("min-order", mode
, dir
,
1679 &fail_page_alloc
.min_order
))
1684 debugfs_remove_recursive(dir
);
1689 late_initcall(fail_page_alloc_debugfs
);
1691 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1693 #else /* CONFIG_FAIL_PAGE_ALLOC */
1695 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1700 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1703 * Return true if free pages are above 'mark'. This takes into account the order
1704 * of the allocation.
1706 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1707 int classzone_idx
, int alloc_flags
, long free_pages
)
1709 /* free_pages my go negative - that's OK */
1711 long lowmem_reserve
= z
->lowmem_reserve
[classzone_idx
];
1715 free_pages
-= (1 << order
) - 1;
1716 if (alloc_flags
& ALLOC_HIGH
)
1718 if (alloc_flags
& ALLOC_HARDER
)
1721 /* If allocation can't use CMA areas don't use free CMA pages */
1722 if (!(alloc_flags
& ALLOC_CMA
))
1723 free_cma
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
1726 if (free_pages
- free_cma
<= min
+ lowmem_reserve
)
1728 for (o
= 0; o
< order
; o
++) {
1729 /* At the next order, this order's pages become unavailable */
1730 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1732 /* Require fewer higher order pages to be free */
1733 min
>>= min_free_order_shift
;
1735 if (free_pages
<= min
)
1741 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1742 int classzone_idx
, int alloc_flags
)
1744 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1745 zone_page_state(z
, NR_FREE_PAGES
));
1748 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1749 int classzone_idx
, int alloc_flags
)
1751 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1753 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1754 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1756 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1762 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1763 * skip over zones that are not allowed by the cpuset, or that have
1764 * been recently (in last second) found to be nearly full. See further
1765 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1766 * that have to skip over a lot of full or unallowed zones.
1768 * If the zonelist cache is present in the passed in zonelist, then
1769 * returns a pointer to the allowed node mask (either the current
1770 * tasks mems_allowed, or node_states[N_MEMORY].)
1772 * If the zonelist cache is not available for this zonelist, does
1773 * nothing and returns NULL.
1775 * If the fullzones BITMAP in the zonelist cache is stale (more than
1776 * a second since last zap'd) then we zap it out (clear its bits.)
1778 * We hold off even calling zlc_setup, until after we've checked the
1779 * first zone in the zonelist, on the theory that most allocations will
1780 * be satisfied from that first zone, so best to examine that zone as
1781 * quickly as we can.
1783 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1785 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1786 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1788 zlc
= zonelist
->zlcache_ptr
;
1792 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1793 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1794 zlc
->last_full_zap
= jiffies
;
1797 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1798 &cpuset_current_mems_allowed
:
1799 &node_states
[N_MEMORY
];
1800 return allowednodes
;
1804 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1805 * if it is worth looking at further for free memory:
1806 * 1) Check that the zone isn't thought to be full (doesn't have its
1807 * bit set in the zonelist_cache fullzones BITMAP).
1808 * 2) Check that the zones node (obtained from the zonelist_cache
1809 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1810 * Return true (non-zero) if zone is worth looking at further, or
1811 * else return false (zero) if it is not.
1813 * This check -ignores- the distinction between various watermarks,
1814 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1815 * found to be full for any variation of these watermarks, it will
1816 * be considered full for up to one second by all requests, unless
1817 * we are so low on memory on all allowed nodes that we are forced
1818 * into the second scan of the zonelist.
1820 * In the second scan we ignore this zonelist cache and exactly
1821 * apply the watermarks to all zones, even it is slower to do so.
1822 * We are low on memory in the second scan, and should leave no stone
1823 * unturned looking for a free page.
1825 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1826 nodemask_t
*allowednodes
)
1828 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1829 int i
; /* index of *z in zonelist zones */
1830 int n
; /* node that zone *z is on */
1832 zlc
= zonelist
->zlcache_ptr
;
1836 i
= z
- zonelist
->_zonerefs
;
1839 /* This zone is worth trying if it is allowed but not full */
1840 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1844 * Given 'z' scanning a zonelist, set the corresponding bit in
1845 * zlc->fullzones, so that subsequent attempts to allocate a page
1846 * from that zone don't waste time re-examining it.
1848 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1850 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1851 int i
; /* index of *z in zonelist zones */
1853 zlc
= zonelist
->zlcache_ptr
;
1857 i
= z
- zonelist
->_zonerefs
;
1859 set_bit(i
, zlc
->fullzones
);
1863 * clear all zones full, called after direct reclaim makes progress so that
1864 * a zone that was recently full is not skipped over for up to a second
1866 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1868 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1870 zlc
= zonelist
->zlcache_ptr
;
1874 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1877 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1879 return node_isset(local_zone
->node
, zone
->zone_pgdat
->reclaim_nodes
);
1882 static void __paginginit
init_zone_allows_reclaim(int nid
)
1886 for_each_online_node(i
)
1887 if (node_distance(nid
, i
) <= RECLAIM_DISTANCE
)
1888 node_set(i
, NODE_DATA(nid
)->reclaim_nodes
);
1890 zone_reclaim_mode
= 1;
1893 #else /* CONFIG_NUMA */
1895 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1900 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1901 nodemask_t
*allowednodes
)
1906 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1910 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1914 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
1919 static inline void init_zone_allows_reclaim(int nid
)
1922 #endif /* CONFIG_NUMA */
1925 * get_page_from_freelist goes through the zonelist trying to allocate
1928 static struct page
*
1929 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1930 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1931 struct zone
*preferred_zone
, int migratetype
)
1934 struct page
*page
= NULL
;
1937 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1938 int zlc_active
= 0; /* set if using zonelist_cache */
1939 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1941 classzone_idx
= zone_idx(preferred_zone
);
1944 * Scan zonelist, looking for a zone with enough free.
1945 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1947 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1948 high_zoneidx
, nodemask
) {
1949 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
1950 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1952 if ((alloc_flags
& ALLOC_CPUSET
) &&
1953 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1956 * When allocating a page cache page for writing, we
1957 * want to get it from a zone that is within its dirty
1958 * limit, such that no single zone holds more than its
1959 * proportional share of globally allowed dirty pages.
1960 * The dirty limits take into account the zone's
1961 * lowmem reserves and high watermark so that kswapd
1962 * should be able to balance it without having to
1963 * write pages from its LRU list.
1965 * This may look like it could increase pressure on
1966 * lower zones by failing allocations in higher zones
1967 * before they are full. But the pages that do spill
1968 * over are limited as the lower zones are protected
1969 * by this very same mechanism. It should not become
1970 * a practical burden to them.
1972 * XXX: For now, allow allocations to potentially
1973 * exceed the per-zone dirty limit in the slowpath
1974 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1975 * which is important when on a NUMA setup the allowed
1976 * zones are together not big enough to reach the
1977 * global limit. The proper fix for these situations
1978 * will require awareness of zones in the
1979 * dirty-throttling and the flusher threads.
1981 if ((alloc_flags
& ALLOC_WMARK_LOW
) &&
1982 (gfp_mask
& __GFP_WRITE
) && !zone_dirty_ok(zone
))
1983 goto this_zone_full
;
1985 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1986 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1990 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1991 if (zone_watermark_ok(zone
, order
, mark
,
1992 classzone_idx
, alloc_flags
))
1995 if (IS_ENABLED(CONFIG_NUMA
) &&
1996 !did_zlc_setup
&& nr_online_nodes
> 1) {
1998 * we do zlc_setup if there are multiple nodes
1999 * and before considering the first zone allowed
2002 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
2007 if (zone_reclaim_mode
== 0 ||
2008 !zone_allows_reclaim(preferred_zone
, zone
))
2009 goto this_zone_full
;
2012 * As we may have just activated ZLC, check if the first
2013 * eligible zone has failed zone_reclaim recently.
2015 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
2016 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
2019 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2021 case ZONE_RECLAIM_NOSCAN
:
2024 case ZONE_RECLAIM_FULL
:
2025 /* scanned but unreclaimable */
2028 /* did we reclaim enough */
2029 if (zone_watermark_ok(zone
, order
, mark
,
2030 classzone_idx
, alloc_flags
))
2034 * Failed to reclaim enough to meet watermark.
2035 * Only mark the zone full if checking the min
2036 * watermark or if we failed to reclaim just
2037 * 1<<order pages or else the page allocator
2038 * fastpath will prematurely mark zones full
2039 * when the watermark is between the low and
2042 if (((alloc_flags
& ALLOC_WMARK_MASK
) == ALLOC_WMARK_MIN
) ||
2043 ret
== ZONE_RECLAIM_SOME
)
2044 goto this_zone_full
;
2051 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
2052 gfp_mask
, migratetype
);
2056 if (IS_ENABLED(CONFIG_NUMA
))
2057 zlc_mark_zone_full(zonelist
, z
);
2060 if (unlikely(IS_ENABLED(CONFIG_NUMA
) && page
== NULL
&& zlc_active
)) {
2061 /* Disable zlc cache for second zonelist scan */
2068 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2069 * necessary to allocate the page. The expectation is
2070 * that the caller is taking steps that will free more
2071 * memory. The caller should avoid the page being used
2072 * for !PFMEMALLOC purposes.
2074 page
->pfmemalloc
= !!(alloc_flags
& ALLOC_NO_WATERMARKS
);
2080 * Large machines with many possible nodes should not always dump per-node
2081 * meminfo in irq context.
2083 static inline bool should_suppress_show_mem(void)
2088 ret
= in_interrupt();
2093 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2094 DEFAULT_RATELIMIT_INTERVAL
,
2095 DEFAULT_RATELIMIT_BURST
);
2097 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
2099 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2101 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2102 debug_guardpage_minorder() > 0)
2106 * Walking all memory to count page types is very expensive and should
2107 * be inhibited in non-blockable contexts.
2109 if (!(gfp_mask
& __GFP_WAIT
))
2110 filter
|= SHOW_MEM_FILTER_PAGE_COUNT
;
2113 * This documents exceptions given to allocations in certain
2114 * contexts that are allowed to allocate outside current's set
2117 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2118 if (test_thread_flag(TIF_MEMDIE
) ||
2119 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2120 filter
&= ~SHOW_MEM_FILTER_NODES
;
2121 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
2122 filter
&= ~SHOW_MEM_FILTER_NODES
;
2125 struct va_format vaf
;
2128 va_start(args
, fmt
);
2133 pr_warn("%pV", &vaf
);
2138 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2139 current
->comm
, order
, gfp_mask
);
2142 if (!should_suppress_show_mem())
2147 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
2148 unsigned long did_some_progress
,
2149 unsigned long pages_reclaimed
)
2151 /* Do not loop if specifically requested */
2152 if (gfp_mask
& __GFP_NORETRY
)
2155 /* Always retry if specifically requested */
2156 if (gfp_mask
& __GFP_NOFAIL
)
2160 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2161 * making forward progress without invoking OOM. Suspend also disables
2162 * storage devices so kswapd will not help. Bail if we are suspending.
2164 if (!did_some_progress
&& pm_suspended_storage())
2168 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2169 * means __GFP_NOFAIL, but that may not be true in other
2172 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
2176 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2177 * specified, then we retry until we no longer reclaim any pages
2178 * (above), or we've reclaimed an order of pages at least as
2179 * large as the allocation's order. In both cases, if the
2180 * allocation still fails, we stop retrying.
2182 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
2188 static inline struct page
*
2189 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2190 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2191 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2196 /* Acquire the OOM killer lock for the zones in zonelist */
2197 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
2198 schedule_timeout_uninterruptible(1);
2203 * PM-freezer should be notified that there might be an OOM killer on
2204 * its way to kill and wake somebody up. This is too early and we might
2205 * end up not killing anything but false positives are acceptable.
2206 * See freeze_processes.
2211 * Go through the zonelist yet one more time, keep very high watermark
2212 * here, this is only to catch a parallel oom killing, we must fail if
2213 * we're still under heavy pressure.
2215 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
2216 order
, zonelist
, high_zoneidx
,
2217 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
2218 preferred_zone
, migratetype
);
2222 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2223 /* The OOM killer will not help higher order allocs */
2224 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2226 /* The OOM killer does not needlessly kill tasks for lowmem */
2227 if (high_zoneidx
< ZONE_NORMAL
)
2230 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2231 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2232 * The caller should handle page allocation failure by itself if
2233 * it specifies __GFP_THISNODE.
2234 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2236 if (gfp_mask
& __GFP_THISNODE
)
2239 /* Exhausted what can be done so it's blamo time */
2240 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
, false);
2243 clear_zonelist_oom(zonelist
, gfp_mask
);
2247 #ifdef CONFIG_COMPACTION
2248 /* Try memory compaction for high-order allocations before reclaim */
2249 static struct page
*
2250 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2251 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2252 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2253 int migratetype
, bool sync_migration
,
2254 bool *contended_compaction
, bool *deferred_compaction
,
2255 unsigned long *did_some_progress
)
2260 if (compaction_deferred(preferred_zone
, order
)) {
2261 *deferred_compaction
= true;
2265 current
->flags
|= PF_MEMALLOC
;
2266 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
2267 nodemask
, sync_migration
,
2268 contended_compaction
);
2269 current
->flags
&= ~PF_MEMALLOC
;
2271 if (*did_some_progress
!= COMPACT_SKIPPED
) {
2274 /* Page migration frees to the PCP lists but we want merging */
2275 drain_pages(get_cpu());
2278 page
= get_page_from_freelist(gfp_mask
, nodemask
,
2279 order
, zonelist
, high_zoneidx
,
2280 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2281 preferred_zone
, migratetype
);
2283 preferred_zone
->compact_blockskip_flush
= false;
2284 preferred_zone
->compact_considered
= 0;
2285 preferred_zone
->compact_defer_shift
= 0;
2286 if (order
>= preferred_zone
->compact_order_failed
)
2287 preferred_zone
->compact_order_failed
= order
+ 1;
2288 count_vm_event(COMPACTSUCCESS
);
2293 * It's bad if compaction run occurs and fails.
2294 * The most likely reason is that pages exist,
2295 * but not enough to satisfy watermarks.
2297 count_vm_event(COMPACTFAIL
);
2300 * As async compaction considers a subset of pageblocks, only
2301 * defer if the failure was a sync compaction failure.
2304 defer_compaction(preferred_zone
, order
);
2312 static inline struct page
*
2313 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2314 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2315 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2316 int migratetype
, bool sync_migration
,
2317 bool *contended_compaction
, bool *deferred_compaction
,
2318 unsigned long *did_some_progress
)
2322 #endif /* CONFIG_COMPACTION */
2324 /* Perform direct synchronous page reclaim */
2326 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
, struct zonelist
*zonelist
,
2327 nodemask_t
*nodemask
)
2329 struct reclaim_state reclaim_state
;
2334 /* We now go into synchronous reclaim */
2335 cpuset_memory_pressure_bump();
2336 current
->flags
|= PF_MEMALLOC
;
2337 lockdep_set_current_reclaim_state(gfp_mask
);
2338 reclaim_state
.reclaimed_slab
= 0;
2339 current
->reclaim_state
= &reclaim_state
;
2341 progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
2343 current
->reclaim_state
= NULL
;
2344 lockdep_clear_current_reclaim_state();
2345 current
->flags
&= ~PF_MEMALLOC
;
2352 /* The really slow allocator path where we enter direct reclaim */
2353 static inline struct page
*
2354 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2355 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2356 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
2357 int migratetype
, unsigned long *did_some_progress
)
2359 struct page
*page
= NULL
;
2360 bool drained
= false;
2362 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, zonelist
,
2364 if (unlikely(!(*did_some_progress
)))
2367 /* After successful reclaim, reconsider all zones for allocation */
2368 if (IS_ENABLED(CONFIG_NUMA
))
2369 zlc_clear_zones_full(zonelist
);
2372 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2373 zonelist
, high_zoneidx
,
2374 alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2375 preferred_zone
, migratetype
);
2378 * If an allocation failed after direct reclaim, it could be because
2379 * pages are pinned on the per-cpu lists. Drain them and try again
2381 if (!page
&& !drained
) {
2391 * This is called in the allocator slow-path if the allocation request is of
2392 * sufficient urgency to ignore watermarks and take other desperate measures
2394 static inline struct page
*
2395 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2396 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2397 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2403 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
2404 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
2405 preferred_zone
, migratetype
);
2407 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2408 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2409 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2415 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
2416 enum zone_type high_zoneidx
,
2417 enum zone_type classzone_idx
)
2422 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
2423 wakeup_kswapd(zone
, order
, classzone_idx
);
2427 gfp_to_alloc_flags(gfp_t gfp_mask
)
2429 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2430 const bool atomic
= !(gfp_mask
& (__GFP_WAIT
| __GFP_NO_KSWAPD
));
2432 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2433 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2436 * The caller may dip into page reserves a bit more if the caller
2437 * cannot run direct reclaim, or if the caller has realtime scheduling
2438 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2439 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
2441 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2445 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2446 * if it can't schedule.
2448 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2449 alloc_flags
|= ALLOC_HARDER
;
2451 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2452 * comment for __cpuset_node_allowed_softwall().
2454 alloc_flags
&= ~ALLOC_CPUSET
;
2455 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2456 alloc_flags
|= ALLOC_HARDER
;
2458 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2459 if (gfp_mask
& __GFP_MEMALLOC
)
2460 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2461 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2462 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2463 else if (!in_interrupt() &&
2464 ((current
->flags
& PF_MEMALLOC
) ||
2465 unlikely(test_thread_flag(TIF_MEMDIE
))))
2466 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2469 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2470 alloc_flags
|= ALLOC_CMA
;
2475 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2477 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2480 static inline struct page
*
2481 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2482 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2483 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2486 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2487 struct page
*page
= NULL
;
2489 unsigned long pages_reclaimed
= 0;
2490 unsigned long did_some_progress
;
2491 bool sync_migration
= false;
2492 bool deferred_compaction
= false;
2493 bool contended_compaction
= false;
2494 #ifdef CONFIG_SEC_OOM_KILLER
2495 unsigned long oom_invoke_timeout
= jiffies
+ HZ
/4;
2499 * In the slowpath, we sanity check order to avoid ever trying to
2500 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2501 * be using allocators in order of preference for an area that is
2504 if (order
>= MAX_ORDER
) {
2505 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2509 #if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
2510 set_tsk_thread_flag(current
, TIF_MEMALLOC
);
2514 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2515 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2516 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2517 * using a larger set of nodes after it has established that the
2518 * allowed per node queues are empty and that nodes are
2521 if (IS_ENABLED(CONFIG_NUMA
) &&
2522 (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2526 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2527 wake_all_kswapd(order
, zonelist
, high_zoneidx
,
2528 zone_idx(preferred_zone
));
2531 * OK, we're below the kswapd watermark and have kicked background
2532 * reclaim. Now things get more complex, so set up alloc_flags according
2533 * to how we want to proceed.
2535 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2538 * Find the true preferred zone if the allocation is unconstrained by
2541 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2542 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2546 /* This is the last chance, in general, before the goto nopage. */
2547 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2548 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2549 preferred_zone
, migratetype
);
2553 /* Allocate without watermarks if the context allows */
2554 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2556 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2557 * the allocation is high priority and these type of
2558 * allocations are system rather than user orientated
2560 zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
2562 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2563 zonelist
, high_zoneidx
, nodemask
,
2564 preferred_zone
, migratetype
);
2570 /* Atomic allocations - we can't balance anything */
2574 /* Avoid recursion of direct reclaim */
2575 if (current
->flags
& PF_MEMALLOC
)
2578 /* Avoid allocations with no watermarks from looping endlessly */
2579 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2583 * Try direct compaction. The first pass is asynchronous. Subsequent
2584 * attempts after direct reclaim are synchronous
2586 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2587 zonelist
, high_zoneidx
,
2589 alloc_flags
, preferred_zone
,
2590 migratetype
, sync_migration
,
2591 &contended_compaction
,
2592 &deferred_compaction
,
2593 &did_some_progress
);
2596 sync_migration
= true;
2599 * If compaction is deferred for high-order allocations, it is because
2600 * sync compaction recently failed. In this is the case and the caller
2601 * requested a movable allocation that does not heavily disrupt the
2602 * system then fail the allocation instead of entering direct reclaim.
2604 if ((deferred_compaction
|| contended_compaction
) &&
2605 (gfp_mask
& __GFP_NO_KSWAPD
))
2608 /* Try direct reclaim and then allocating */
2609 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2610 zonelist
, high_zoneidx
,
2612 alloc_flags
, preferred_zone
,
2613 migratetype
, &did_some_progress
);
2618 * If we failed to make any progress reclaiming, then we are
2619 * running out of options and have to consider going OOM
2620 * If we are looping more than 250 ms, go to OOM
2623 #ifdef CONFIG_SEC_OOM_KILLER
2624 #define SHOULD_CONSIDER_OOM (!did_some_progress || time_after(jiffies, oom_invoke_timeout)) && (boot_mode != 2)
2626 #define SHOULD_CONSIDER_OOM !did_some_progress && (boot_mode != 2)
2629 if (SHOULD_CONSIDER_OOM
) {
2630 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2631 if (oom_killer_disabled
)
2633 /* Coredumps can quickly deplete all memory reserves */
2634 if ((current
->flags
& PF_DUMPCORE
) &&
2635 !(gfp_mask
& __GFP_NOFAIL
))
2638 #ifdef CONFIG_SEC_OOM_KILLER
2639 if (did_some_progress
)
2640 pr_info("time's up : calling "
2641 "__alloc_pages_may_oom(o:%u, gfp:0x%x)\n",
2644 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2645 zonelist
, high_zoneidx
,
2646 nodemask
, preferred_zone
,
2651 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2653 * The oom killer is not called for high-order
2654 * allocations that may fail, so if no progress
2655 * is being made, there are no other options and
2656 * retrying is unlikely to help.
2658 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2661 * The oom killer is not called for lowmem
2662 * allocations to prevent needlessly killing
2665 if (high_zoneidx
< ZONE_NORMAL
)
2669 #ifdef CONFIG_SEC_OOM_KILLER
2670 oom_invoke_timeout
= jiffies
+ HZ
/4;
2676 /* Check if we should retry the allocation */
2677 pages_reclaimed
+= did_some_progress
;
2678 if (should_alloc_retry(gfp_mask
, order
, did_some_progress
,
2680 /* Wait for some write requests to complete then retry */
2681 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2685 * High-order allocations do not necessarily loop after
2686 * direct reclaim and reclaim/compaction depends on compaction
2687 * being called after reclaim so call directly if necessary
2689 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2690 zonelist
, high_zoneidx
,
2692 alloc_flags
, preferred_zone
,
2693 migratetype
, sync_migration
,
2694 &contended_compaction
,
2695 &deferred_compaction
,
2696 &did_some_progress
);
2702 #if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
2703 clear_tsk_thread_flag(current
, TIF_MEMALLOC
);
2705 warn_alloc_failed(gfp_mask
, order
, NULL
);
2708 #if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
2709 clear_tsk_thread_flag(current
, TIF_MEMALLOC
);
2711 if (kmemcheck_enabled
)
2712 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2718 * This is the 'heart' of the zoned buddy allocator.
2721 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2722 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2724 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2725 struct zone
*preferred_zone
;
2726 struct page
*page
= NULL
;
2727 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2728 unsigned int cpuset_mems_cookie
;
2729 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
;
2730 struct mem_cgroup
*memcg
= NULL
;
2732 gfp_mask
&= gfp_allowed_mask
;
2734 lockdep_trace_alloc(gfp_mask
);
2736 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2738 if (should_fail_alloc_page(gfp_mask
, order
))
2742 * Check the zones suitable for the gfp_mask contain at least one
2743 * valid zone. It's possible to have an empty zonelist as a result
2744 * of GFP_THISNODE and a memoryless node
2746 if (unlikely(!zonelist
->_zonerefs
->zone
))
2750 * Will only have any effect when __GFP_KMEMCG is set. This is
2751 * verified in the (always inline) callee
2753 if (!memcg_kmem_newpage_charge(gfp_mask
, &memcg
, order
))
2757 cpuset_mems_cookie
= get_mems_allowed();
2759 /* The preferred zone is used for statistics later */
2760 first_zones_zonelist(zonelist
, high_zoneidx
,
2761 nodemask
? : &cpuset_current_mems_allowed
,
2763 if (!preferred_zone
)
2767 if (allocflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2768 alloc_flags
|= ALLOC_CMA
;
2770 /* First allocation attempt */
2771 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2772 zonelist
, high_zoneidx
, alloc_flags
,
2773 preferred_zone
, migratetype
);
2774 if (unlikely(!page
)) {
2776 * Runtime PM, block IO and its error handling path
2777 * can deadlock because I/O on the device might not
2780 gfp_mask
= memalloc_noio_flags(gfp_mask
);
2781 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2782 zonelist
, high_zoneidx
, nodemask
,
2783 preferred_zone
, migratetype
);
2786 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2790 * When updating a task's mems_allowed, it is possible to race with
2791 * parallel threads in such a way that an allocation can fail while
2792 * the mask is being updated. If a page allocation is about to fail,
2793 * check if the cpuset changed during allocation and if so, retry.
2795 if (unlikely(!put_mems_allowed(cpuset_mems_cookie
) && !page
))
2798 memcg_kmem_commit_charge(page
, memcg
, order
);
2802 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2805 * Common helper functions.
2807 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2812 * __get_free_pages() returns a 32-bit address, which cannot represent
2815 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2817 page
= alloc_pages(gfp_mask
, order
);
2820 return (unsigned long) page_address(page
);
2822 EXPORT_SYMBOL(__get_free_pages
);
2824 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2826 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2828 EXPORT_SYMBOL(get_zeroed_page
);
2830 void __free_pages(struct page
*page
, unsigned int order
)
2832 if (put_page_testzero(page
)) {
2834 free_hot_cold_page(page
, 0);
2836 __free_pages_ok(page
, order
);
2840 EXPORT_SYMBOL(__free_pages
);
2842 void free_pages(unsigned long addr
, unsigned int order
)
2845 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2846 __free_pages(virt_to_page((void *)addr
), order
);
2850 EXPORT_SYMBOL(free_pages
);
2853 * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2854 * pages allocated with __GFP_KMEMCG.
2856 * Those pages are accounted to a particular memcg, embedded in the
2857 * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2858 * for that information only to find out that it is NULL for users who have no
2859 * interest in that whatsoever, we provide these functions.
2861 * The caller knows better which flags it relies on.
2863 void __free_memcg_kmem_pages(struct page
*page
, unsigned int order
)
2865 memcg_kmem_uncharge_pages(page
, order
);
2866 __free_pages(page
, order
);
2869 void free_memcg_kmem_pages(unsigned long addr
, unsigned int order
)
2872 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2873 __free_memcg_kmem_pages(virt_to_page((void *)addr
), order
);
2877 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2880 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2881 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2883 split_page(virt_to_page((void *)addr
), order
);
2884 while (used
< alloc_end
) {
2889 return (void *)addr
;
2893 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2894 * @size: the number of bytes to allocate
2895 * @gfp_mask: GFP flags for the allocation
2897 * This function is similar to alloc_pages(), except that it allocates the
2898 * minimum number of pages to satisfy the request. alloc_pages() can only
2899 * allocate memory in power-of-two pages.
2901 * This function is also limited by MAX_ORDER.
2903 * Memory allocated by this function must be released by free_pages_exact().
2905 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2907 unsigned int order
= get_order(size
);
2910 addr
= __get_free_pages(gfp_mask
, order
);
2911 return make_alloc_exact(addr
, order
, size
);
2913 EXPORT_SYMBOL(alloc_pages_exact
);
2916 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2918 * @nid: the preferred node ID where memory should be allocated
2919 * @size: the number of bytes to allocate
2920 * @gfp_mask: GFP flags for the allocation
2922 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2924 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2927 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2929 unsigned order
= get_order(size
);
2930 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2933 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2935 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2938 * free_pages_exact - release memory allocated via alloc_pages_exact()
2939 * @virt: the value returned by alloc_pages_exact.
2940 * @size: size of allocation, same value as passed to alloc_pages_exact().
2942 * Release the memory allocated by a previous call to alloc_pages_exact.
2944 void free_pages_exact(void *virt
, size_t size
)
2946 unsigned long addr
= (unsigned long)virt
;
2947 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2949 while (addr
< end
) {
2954 EXPORT_SYMBOL(free_pages_exact
);
2957 * nr_free_zone_pages - count number of pages beyond high watermark
2958 * @offset: The zone index of the highest zone
2960 * nr_free_zone_pages() counts the number of counts pages which are beyond the
2961 * high watermark within all zones at or below a given zone index. For each
2962 * zone, the number of pages is calculated as:
2963 * present_pages - high_pages
2965 static unsigned long nr_free_zone_pages(int offset
)
2970 /* Just pick one node, since fallback list is circular */
2971 unsigned long sum
= 0;
2973 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2975 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2976 unsigned long size
= zone
->managed_pages
;
2977 unsigned long high
= high_wmark_pages(zone
);
2986 * nr_free_buffer_pages - count number of pages beyond high watermark
2988 * nr_free_buffer_pages() counts the number of pages which are beyond the high
2989 * watermark within ZONE_DMA and ZONE_NORMAL.
2991 unsigned long nr_free_buffer_pages(void)
2993 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2995 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2998 * nr_free_pagecache_pages - count number of pages beyond high watermark
3000 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3001 * high watermark within all zones.
3003 unsigned long nr_free_pagecache_pages(void)
3005 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
3008 static inline void show_node(struct zone
*zone
)
3010 if (IS_ENABLED(CONFIG_NUMA
))
3011 printk("Node %d ", zone_to_nid(zone
));
3014 void si_meminfo(struct sysinfo
*val
)
3016 val
->totalram
= totalram_pages
;
3018 val
->freeram
= global_page_state(NR_FREE_PAGES
);
3019 val
->bufferram
= nr_blockdev_pages();
3020 val
->totalhigh
= totalhigh_pages
;
3021 val
->freehigh
= nr_free_highpages();
3022 val
->mem_unit
= PAGE_SIZE
;
3025 EXPORT_SYMBOL(si_meminfo
);
3028 void si_meminfo_node(struct sysinfo
*val
, int nid
)
3030 pg_data_t
*pgdat
= NODE_DATA(nid
);
3032 val
->totalram
= pgdat
->node_present_pages
;
3033 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
3034 #ifdef CONFIG_HIGHMEM
3035 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
3036 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
3042 val
->mem_unit
= PAGE_SIZE
;
3047 * Determine whether the node should be displayed or not, depending on whether
3048 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3050 bool skip_free_areas_node(unsigned int flags
, int nid
)
3053 unsigned int cpuset_mems_cookie
;
3055 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3059 cpuset_mems_cookie
= get_mems_allowed();
3060 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3061 } while (!put_mems_allowed(cpuset_mems_cookie
));
3066 #define K(x) ((x) << (PAGE_SHIFT-10))
3068 static void show_migration_types(unsigned char type
)
3070 static const char types
[MIGRATE_TYPES
] = {
3071 [MIGRATE_UNMOVABLE
] = 'U',
3072 [MIGRATE_RECLAIMABLE
] = 'E',
3073 [MIGRATE_MOVABLE
] = 'M',
3074 [MIGRATE_RESERVE
] = 'R',
3076 [MIGRATE_CMA
] = 'C',
3078 #ifdef CONFIG_MEMORY_ISOLATION
3079 [MIGRATE_ISOLATE
] = 'I',
3082 char tmp
[MIGRATE_TYPES
+ 1];
3086 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3087 if (type
& (1 << i
))
3092 printk("(%s) ", tmp
);
3096 * Show free area list (used inside shift_scroll-lock stuff)
3097 * We also calculate the percentage fragmentation. We do this by counting the
3098 * memory on each free list with the exception of the first item on the list.
3099 * Suppresses nodes that are not allowed by current's cpuset if
3100 * SHOW_MEM_FILTER_NODES is passed.
3102 void show_free_areas(unsigned int filter
)
3107 for_each_populated_zone(zone
) {
3108 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3111 printk("%s per-cpu:\n", zone
->name
);
3113 for_each_online_cpu(cpu
) {
3114 struct per_cpu_pageset
*pageset
;
3116 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
3118 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3119 cpu
, pageset
->pcp
.high
,
3120 pageset
->pcp
.batch
, pageset
->pcp
.count
);
3124 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3125 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3127 " dirty:%lu writeback:%lu unstable:%lu\n"
3128 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3129 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3131 global_page_state(NR_ACTIVE_ANON
),
3132 global_page_state(NR_INACTIVE_ANON
),
3133 global_page_state(NR_ISOLATED_ANON
),
3134 global_page_state(NR_ACTIVE_FILE
),
3135 global_page_state(NR_INACTIVE_FILE
),
3136 global_page_state(NR_ISOLATED_FILE
),
3137 global_page_state(NR_UNEVICTABLE
),
3138 global_page_state(NR_FILE_DIRTY
),
3139 global_page_state(NR_WRITEBACK
),
3140 global_page_state(NR_UNSTABLE_NFS
),
3141 global_page_state(NR_FREE_PAGES
),
3142 global_page_state(NR_SLAB_RECLAIMABLE
),
3143 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3144 global_page_state(NR_FILE_MAPPED
),
3145 global_page_state(NR_SHMEM
),
3146 global_page_state(NR_PAGETABLE
),
3147 global_page_state(NR_BOUNCE
),
3148 global_page_state(NR_FREE_CMA_PAGES
));
3150 for_each_populated_zone(zone
) {
3153 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3161 " active_anon:%lukB"
3162 " inactive_anon:%lukB"
3163 " active_file:%lukB"
3164 " inactive_file:%lukB"
3165 " unevictable:%lukB"
3166 " isolated(anon):%lukB"
3167 " isolated(file):%lukB"
3175 " slab_reclaimable:%lukB"
3176 " slab_unreclaimable:%lukB"
3177 " kernel_stack:%lukB"
3182 " writeback_tmp:%lukB"
3183 " pages_scanned:%lu"
3184 " all_unreclaimable? %s"
3187 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3188 K(min_wmark_pages(zone
)),
3189 K(low_wmark_pages(zone
)),
3190 K(high_wmark_pages(zone
)),
3191 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3192 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3193 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3194 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3195 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3196 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3197 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3198 K(zone
->present_pages
),
3199 K(zone
->managed_pages
),
3200 K(zone_page_state(zone
, NR_MLOCK
)),
3201 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3202 K(zone_page_state(zone
, NR_WRITEBACK
)),
3203 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3204 K(zone_page_state(zone
, NR_SHMEM
)),
3205 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3206 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3207 zone_page_state(zone
, NR_KERNEL_STACK
) *
3209 K(zone_page_state(zone
, NR_PAGETABLE
)),
3210 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3211 K(zone_page_state(zone
, NR_BOUNCE
)),
3212 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3213 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3214 zone
->pages_scanned
,
3215 (zone
->all_unreclaimable
? "yes" : "no")
3217 printk("lowmem_reserve[]:");
3218 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3219 printk(" %lu", zone
->lowmem_reserve
[i
]);
3223 for_each_populated_zone(zone
) {
3224 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
3225 unsigned char types
[MAX_ORDER
];
3227 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3230 printk("%s: ", zone
->name
);
3232 spin_lock_irqsave(&zone
->lock
, flags
);
3233 for (order
= 0; order
< MAX_ORDER
; order
++) {
3234 struct free_area
*area
= &zone
->free_area
[order
];
3237 nr
[order
] = area
->nr_free
;
3238 total
+= nr
[order
] << order
;
3241 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3242 if (!list_empty(&area
->free_list
[type
]))
3243 types
[order
] |= 1 << type
;
3246 spin_unlock_irqrestore(&zone
->lock
, flags
);
3247 for (order
= 0; order
< MAX_ORDER
; order
++) {
3248 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3250 show_migration_types(types
[order
]);
3252 printk("= %lukB\n", K(total
));
3255 hugetlb_show_meminfo();
3257 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3259 show_swap_cache_info();
3262 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3264 zoneref
->zone
= zone
;
3265 zoneref
->zone_idx
= zone_idx(zone
);
3269 * Builds allocation fallback zone lists.
3271 * Add all populated zones of a node to the zonelist.
3273 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3274 int nr_zones
, enum zone_type zone_type
)
3278 BUG_ON(zone_type
>= MAX_NR_ZONES
);
3283 zone
= pgdat
->node_zones
+ zone_type
;
3284 if (populated_zone(zone
)) {
3285 zoneref_set_zone(zone
,
3286 &zonelist
->_zonerefs
[nr_zones
++]);
3287 check_highest_zone(zone_type
);
3290 } while (zone_type
);
3297 * 0 = automatic detection of better ordering.
3298 * 1 = order by ([node] distance, -zonetype)
3299 * 2 = order by (-zonetype, [node] distance)
3301 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3302 * the same zonelist. So only NUMA can configure this param.
3304 #define ZONELIST_ORDER_DEFAULT 0
3305 #define ZONELIST_ORDER_NODE 1
3306 #define ZONELIST_ORDER_ZONE 2
3308 /* zonelist order in the kernel.
3309 * set_zonelist_order() will set this to NODE or ZONE.
3311 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3312 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3316 /* The value user specified ....changed by config */
3317 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3318 /* string for sysctl */
3319 #define NUMA_ZONELIST_ORDER_LEN 16
3320 char numa_zonelist_order
[16] = "default";
3323 * interface for configure zonelist ordering.
3324 * command line option "numa_zonelist_order"
3325 * = "[dD]efault - default, automatic configuration.
3326 * = "[nN]ode - order by node locality, then by zone within node
3327 * = "[zZ]one - order by zone, then by locality within zone
3330 static int __parse_numa_zonelist_order(char *s
)
3332 if (*s
== 'd' || *s
== 'D') {
3333 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3334 } else if (*s
== 'n' || *s
== 'N') {
3335 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3336 } else if (*s
== 'z' || *s
== 'Z') {
3337 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3340 "Ignoring invalid numa_zonelist_order value: "
3347 static __init
int setup_numa_zonelist_order(char *s
)
3354 ret
= __parse_numa_zonelist_order(s
);
3356 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3360 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3363 * sysctl handler for numa_zonelist_order
3365 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
3366 void __user
*buffer
, size_t *length
,
3369 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3371 static DEFINE_MUTEX(zl_order_mutex
);
3373 mutex_lock(&zl_order_mutex
);
3375 strcpy(saved_string
, (char*)table
->data
);
3376 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3380 int oldval
= user_zonelist_order
;
3381 if (__parse_numa_zonelist_order((char*)table
->data
)) {
3383 * bogus value. restore saved string
3385 strncpy((char*)table
->data
, saved_string
,
3386 NUMA_ZONELIST_ORDER_LEN
);
3387 user_zonelist_order
= oldval
;
3388 } else if (oldval
!= user_zonelist_order
) {
3389 mutex_lock(&zonelists_mutex
);
3390 build_all_zonelists(NULL
, NULL
);
3391 mutex_unlock(&zonelists_mutex
);
3395 mutex_unlock(&zl_order_mutex
);
3400 #define MAX_NODE_LOAD (nr_online_nodes)
3401 static int node_load
[MAX_NUMNODES
];
3404 * find_next_best_node - find the next node that should appear in a given node's fallback list
3405 * @node: node whose fallback list we're appending
3406 * @used_node_mask: nodemask_t of already used nodes
3408 * We use a number of factors to determine which is the next node that should
3409 * appear on a given node's fallback list. The node should not have appeared
3410 * already in @node's fallback list, and it should be the next closest node
3411 * according to the distance array (which contains arbitrary distance values
3412 * from each node to each node in the system), and should also prefer nodes
3413 * with no CPUs, since presumably they'll have very little allocation pressure
3414 * on them otherwise.
3415 * It returns -1 if no node is found.
3417 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3420 int min_val
= INT_MAX
;
3421 int best_node
= NUMA_NO_NODE
;
3422 const struct cpumask
*tmp
= cpumask_of_node(0);
3424 /* Use the local node if we haven't already */
3425 if (!node_isset(node
, *used_node_mask
)) {
3426 node_set(node
, *used_node_mask
);
3430 for_each_node_state(n
, N_MEMORY
) {
3432 /* Don't want a node to appear more than once */
3433 if (node_isset(n
, *used_node_mask
))
3436 /* Use the distance array to find the distance */
3437 val
= node_distance(node
, n
);
3439 /* Penalize nodes under us ("prefer the next node") */
3442 /* Give preference to headless and unused nodes */
3443 tmp
= cpumask_of_node(n
);
3444 if (!cpumask_empty(tmp
))
3445 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
3447 /* Slight preference for less loaded node */
3448 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
3449 val
+= node_load
[n
];
3451 if (val
< min_val
) {
3458 node_set(best_node
, *used_node_mask
);
3465 * Build zonelists ordered by node and zones within node.
3466 * This results in maximum locality--normal zone overflows into local
3467 * DMA zone, if any--but risks exhausting DMA zone.
3469 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
3472 struct zonelist
*zonelist
;
3474 zonelist
= &pgdat
->node_zonelists
[0];
3475 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
3477 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3479 zonelist
->_zonerefs
[j
].zone
= NULL
;
3480 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3484 * Build gfp_thisnode zonelists
3486 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
3489 struct zonelist
*zonelist
;
3491 zonelist
= &pgdat
->node_zonelists
[1];
3492 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3493 zonelist
->_zonerefs
[j
].zone
= NULL
;
3494 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3498 * Build zonelists ordered by zone and nodes within zones.
3499 * This results in conserving DMA zone[s] until all Normal memory is
3500 * exhausted, but results in overflowing to remote node while memory
3501 * may still exist in local DMA zone.
3503 static int node_order
[MAX_NUMNODES
];
3505 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
3508 int zone_type
; /* needs to be signed */
3510 struct zonelist
*zonelist
;
3512 zonelist
= &pgdat
->node_zonelists
[0];
3514 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
3515 for (j
= 0; j
< nr_nodes
; j
++) {
3516 node
= node_order
[j
];
3517 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
3518 if (populated_zone(z
)) {
3520 &zonelist
->_zonerefs
[pos
++]);
3521 check_highest_zone(zone_type
);
3525 zonelist
->_zonerefs
[pos
].zone
= NULL
;
3526 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
3529 static int default_zonelist_order(void)
3532 unsigned long low_kmem_size
,total_size
;
3536 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3537 * If they are really small and used heavily, the system can fall
3538 * into OOM very easily.
3539 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3541 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3544 for_each_online_node(nid
) {
3545 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3546 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3547 if (populated_zone(z
)) {
3548 if (zone_type
< ZONE_NORMAL
)
3549 low_kmem_size
+= z
->present_pages
;
3550 total_size
+= z
->present_pages
;
3551 } else if (zone_type
== ZONE_NORMAL
) {
3553 * If any node has only lowmem, then node order
3554 * is preferred to allow kernel allocations
3555 * locally; otherwise, they can easily infringe
3556 * on other nodes when there is an abundance of
3557 * lowmem available to allocate from.
3559 return ZONELIST_ORDER_NODE
;
3563 if (!low_kmem_size
|| /* there are no DMA area. */
3564 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
3565 return ZONELIST_ORDER_NODE
;
3567 * look into each node's config.
3568 * If there is a node whose DMA/DMA32 memory is very big area on
3569 * local memory, NODE_ORDER may be suitable.
3571 average_size
= total_size
/
3572 (nodes_weight(node_states
[N_MEMORY
]) + 1);
3573 for_each_online_node(nid
) {
3576 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
3577 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
3578 if (populated_zone(z
)) {
3579 if (zone_type
< ZONE_NORMAL
)
3580 low_kmem_size
+= z
->present_pages
;
3581 total_size
+= z
->present_pages
;
3584 if (low_kmem_size
&&
3585 total_size
> average_size
&& /* ignore small node */
3586 low_kmem_size
> total_size
* 70/100)
3587 return ZONELIST_ORDER_NODE
;
3589 return ZONELIST_ORDER_ZONE
;
3592 static void set_zonelist_order(void)
3594 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
3595 current_zonelist_order
= default_zonelist_order();
3597 current_zonelist_order
= user_zonelist_order
;
3600 static void build_zonelists(pg_data_t
*pgdat
)
3604 nodemask_t used_mask
;
3605 int local_node
, prev_node
;
3606 struct zonelist
*zonelist
;
3607 int order
= current_zonelist_order
;
3609 /* initialize zonelists */
3610 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3611 zonelist
= pgdat
->node_zonelists
+ i
;
3612 zonelist
->_zonerefs
[0].zone
= NULL
;
3613 zonelist
->_zonerefs
[0].zone_idx
= 0;
3616 /* NUMA-aware ordering of nodes */
3617 local_node
= pgdat
->node_id
;
3618 load
= nr_online_nodes
;
3619 prev_node
= local_node
;
3620 nodes_clear(used_mask
);
3622 memset(node_order
, 0, sizeof(node_order
));
3625 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3627 * We don't want to pressure a particular node.
3628 * So adding penalty to the first node in same
3629 * distance group to make it round-robin.
3631 if (node_distance(local_node
, node
) !=
3632 node_distance(local_node
, prev_node
))
3633 node_load
[node
] = load
;
3637 if (order
== ZONELIST_ORDER_NODE
)
3638 build_zonelists_in_node_order(pgdat
, node
);
3640 node_order
[j
++] = node
; /* remember order */
3643 if (order
== ZONELIST_ORDER_ZONE
) {
3644 /* calculate node order -- i.e., DMA last! */
3645 build_zonelists_in_zone_order(pgdat
, j
);
3648 build_thisnode_zonelists(pgdat
);
3651 /* Construct the zonelist performance cache - see further mmzone.h */
3652 static void build_zonelist_cache(pg_data_t
*pgdat
)
3654 struct zonelist
*zonelist
;
3655 struct zonelist_cache
*zlc
;
3658 zonelist
= &pgdat
->node_zonelists
[0];
3659 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3660 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3661 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3662 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3665 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3667 * Return node id of node used for "local" allocations.
3668 * I.e., first node id of first zone in arg node's generic zonelist.
3669 * Used for initializing percpu 'numa_mem', which is used primarily
3670 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3672 int local_memory_node(int node
)
3676 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3677 gfp_zone(GFP_KERNEL
),
3684 #else /* CONFIG_NUMA */
3686 static void set_zonelist_order(void)
3688 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3691 static void build_zonelists(pg_data_t
*pgdat
)
3693 int node
, local_node
;
3695 struct zonelist
*zonelist
;
3697 local_node
= pgdat
->node_id
;
3699 zonelist
= &pgdat
->node_zonelists
[0];
3700 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3703 * Now we build the zonelist so that it contains the zones
3704 * of all the other nodes.
3705 * We don't want to pressure a particular node, so when
3706 * building the zones for node N, we make sure that the
3707 * zones coming right after the local ones are those from
3708 * node N+1 (modulo N)
3710 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3711 if (!node_online(node
))
3713 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3716 for (node
= 0; node
< local_node
; node
++) {
3717 if (!node_online(node
))
3719 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3723 zonelist
->_zonerefs
[j
].zone
= NULL
;
3724 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3727 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3728 static void build_zonelist_cache(pg_data_t
*pgdat
)
3730 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3733 #endif /* CONFIG_NUMA */
3736 * Boot pageset table. One per cpu which is going to be used for all
3737 * zones and all nodes. The parameters will be set in such a way
3738 * that an item put on a list will immediately be handed over to
3739 * the buddy list. This is safe since pageset manipulation is done
3740 * with interrupts disabled.
3742 * The boot_pagesets must be kept even after bootup is complete for
3743 * unused processors and/or zones. They do play a role for bootstrapping
3744 * hotplugged processors.
3746 * zoneinfo_show() and maybe other functions do
3747 * not check if the processor is online before following the pageset pointer.
3748 * Other parts of the kernel may not check if the zone is available.
3750 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3751 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3752 static void setup_zone_pageset(struct zone
*zone
);
3755 * Global mutex to protect against size modification of zonelists
3756 * as well as to serialize pageset setup for the new populated zone.
3758 DEFINE_MUTEX(zonelists_mutex
);
3760 /* return values int ....just for stop_machine() */
3761 static int __build_all_zonelists(void *data
)
3765 pg_data_t
*self
= data
;
3768 memset(node_load
, 0, sizeof(node_load
));
3771 if (self
&& !node_online(self
->node_id
)) {
3772 build_zonelists(self
);
3773 build_zonelist_cache(self
);
3776 for_each_online_node(nid
) {
3777 pg_data_t
*pgdat
= NODE_DATA(nid
);
3779 build_zonelists(pgdat
);
3780 build_zonelist_cache(pgdat
);
3784 * Initialize the boot_pagesets that are going to be used
3785 * for bootstrapping processors. The real pagesets for
3786 * each zone will be allocated later when the per cpu
3787 * allocator is available.
3789 * boot_pagesets are used also for bootstrapping offline
3790 * cpus if the system is already booted because the pagesets
3791 * are needed to initialize allocators on a specific cpu too.
3792 * F.e. the percpu allocator needs the page allocator which
3793 * needs the percpu allocator in order to allocate its pagesets
3794 * (a chicken-egg dilemma).
3796 for_each_possible_cpu(cpu
) {
3797 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3799 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3801 * We now know the "local memory node" for each node--
3802 * i.e., the node of the first zone in the generic zonelist.
3803 * Set up numa_mem percpu variable for on-line cpus. During
3804 * boot, only the boot cpu should be on-line; we'll init the
3805 * secondary cpus' numa_mem as they come on-line. During
3806 * node/memory hotplug, we'll fixup all on-line cpus.
3808 if (cpu_online(cpu
))
3809 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3817 * Called with zonelists_mutex held always
3818 * unless system_state == SYSTEM_BOOTING.
3820 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
3822 set_zonelist_order();
3824 if (system_state
== SYSTEM_BOOTING
) {
3825 __build_all_zonelists(NULL
);
3826 mminit_verify_zonelist();
3827 cpuset_init_current_mems_allowed();
3829 /* we have to stop all cpus to guarantee there is no user
3831 #ifdef CONFIG_MEMORY_HOTPLUG
3833 setup_zone_pageset(zone
);
3835 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
3836 /* cpuset refresh routine should be here */
3838 vm_total_pages
= nr_free_pagecache_pages();
3840 * Disable grouping by mobility if the number of pages in the
3841 * system is too low to allow the mechanism to work. It would be
3842 * more accurate, but expensive to check per-zone. This check is
3843 * made on memory-hotadd so a system can start with mobility
3844 * disabled and enable it later
3846 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3847 page_group_by_mobility_disabled
= 1;
3849 page_group_by_mobility_disabled
= 0;
3851 printk("Built %i zonelists in %s order, mobility grouping %s. "
3852 "Total pages: %ld\n",
3854 zonelist_order_name
[current_zonelist_order
],
3855 page_group_by_mobility_disabled
? "off" : "on",
3858 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3863 * Helper functions to size the waitqueue hash table.
3864 * Essentially these want to choose hash table sizes sufficiently
3865 * large so that collisions trying to wait on pages are rare.
3866 * But in fact, the number of active page waitqueues on typical
3867 * systems is ridiculously low, less than 200. So this is even
3868 * conservative, even though it seems large.
3870 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3871 * waitqueues, i.e. the size of the waitq table given the number of pages.
3873 #define PAGES_PER_WAITQUEUE 256
3875 #ifndef CONFIG_MEMORY_HOTPLUG
3876 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3878 unsigned long size
= 1;
3880 pages
/= PAGES_PER_WAITQUEUE
;
3882 while (size
< pages
)
3886 * Once we have dozens or even hundreds of threads sleeping
3887 * on IO we've got bigger problems than wait queue collision.
3888 * Limit the size of the wait table to a reasonable size.
3890 size
= min(size
, 4096UL);
3892 return max(size
, 4UL);
3896 * A zone's size might be changed by hot-add, so it is not possible to determine
3897 * a suitable size for its wait_table. So we use the maximum size now.
3899 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3901 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3902 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3903 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3905 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3906 * or more by the traditional way. (See above). It equals:
3908 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3909 * ia64(16K page size) : = ( 8G + 4M)byte.
3910 * powerpc (64K page size) : = (32G +16M)byte.
3912 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3919 * This is an integer logarithm so that shifts can be used later
3920 * to extract the more random high bits from the multiplicative
3921 * hash function before the remainder is taken.
3923 static inline unsigned long wait_table_bits(unsigned long size
)
3928 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3931 * Check if a pageblock contains reserved pages
3933 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3937 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3938 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3945 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3946 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3947 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3948 * higher will lead to a bigger reserve which will get freed as contiguous
3949 * blocks as reclaim kicks in
3951 static void setup_zone_migrate_reserve(struct zone
*zone
)
3953 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3955 unsigned long block_migratetype
;
3959 * Get the start pfn, end pfn and the number of blocks to reserve
3960 * We have to be careful to be aligned to pageblock_nr_pages to
3961 * make sure that we always check pfn_valid for the first page in
3964 start_pfn
= zone
->zone_start_pfn
;
3965 end_pfn
= zone_end_pfn(zone
);
3966 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3967 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3971 * Reserve blocks are generally in place to help high-order atomic
3972 * allocations that are short-lived. A min_free_kbytes value that
3973 * would result in more than 2 reserve blocks for atomic allocations
3974 * is assumed to be in place to help anti-fragmentation for the
3975 * future allocation of hugepages at runtime.
3977 reserve
= min(2, reserve
);
3979 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3980 if (!pfn_valid(pfn
))
3982 page
= pfn_to_page(pfn
);
3984 /* Watch out for overlapping nodes */
3985 if (page_to_nid(page
) != zone_to_nid(zone
))
3988 block_migratetype
= get_pageblock_migratetype(page
);
3990 /* Only test what is necessary when the reserves are not met */
3993 * Blocks with reserved pages will never free, skip
3996 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3997 if (pageblock_is_reserved(pfn
, block_end_pfn
))
4000 /* If this block is reserved, account for it */
4001 if (block_migratetype
== MIGRATE_RESERVE
) {
4006 /* Suitable for reserving if this block is movable */
4007 if (block_migratetype
== MIGRATE_MOVABLE
) {
4008 set_pageblock_migratetype(page
,
4010 move_freepages_block(zone
, page
,
4018 * If the reserve is met and this is a previous reserved block,
4021 if (block_migratetype
== MIGRATE_RESERVE
) {
4022 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4023 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4029 * Initially all pages are reserved - free ones are freed
4030 * up by free_all_bootmem() once the early boot process is
4031 * done. Non-atomic initialization, single-pass.
4033 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
4034 unsigned long start_pfn
, enum memmap_context context
)
4037 unsigned long end_pfn
= start_pfn
+ size
;
4041 if (highest_memmap_pfn
< end_pfn
- 1)
4042 highest_memmap_pfn
= end_pfn
- 1;
4044 z
= &NODE_DATA(nid
)->node_zones
[zone
];
4045 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4047 * There can be holes in boot-time mem_map[]s
4048 * handed to this function. They do not
4049 * exist on hotplugged memory.
4051 if (context
== MEMMAP_EARLY
) {
4052 if (!early_pfn_valid(pfn
))
4054 if (!early_pfn_in_nid(pfn
, nid
))
4057 page
= pfn_to_page(pfn
);
4058 set_page_links(page
, zone
, nid
, pfn
);
4059 mminit_verify_page_links(page
, zone
, nid
, pfn
);
4060 init_page_count(page
);
4061 page_mapcount_reset(page
);
4062 page_nid_reset_last(page
);
4063 SetPageReserved(page
);
4065 * Mark the block movable so that blocks are reserved for
4066 * movable at startup. This will force kernel allocations
4067 * to reserve their blocks rather than leaking throughout
4068 * the address space during boot when many long-lived
4069 * kernel allocations are made. Later some blocks near
4070 * the start are marked MIGRATE_RESERVE by
4071 * setup_zone_migrate_reserve()
4073 * bitmap is created for zone's valid pfn range. but memmap
4074 * can be created for invalid pages (for alignment)
4075 * check here not to call set_pageblock_migratetype() against
4078 if ((z
->zone_start_pfn
<= pfn
)
4079 && (pfn
< zone_end_pfn(z
))
4080 && !(pfn
& (pageblock_nr_pages
- 1)))
4081 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4083 INIT_LIST_HEAD(&page
->lru
);
4084 #ifdef WANT_PAGE_VIRTUAL
4085 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4086 if (!is_highmem_idx(zone
))
4087 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
4092 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4095 for_each_migratetype_order(order
, t
) {
4096 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4097 zone
->free_area
[order
].nr_free
= 0;
4101 #ifndef __HAVE_ARCH_MEMMAP_INIT
4102 #define memmap_init(size, nid, zone, start_pfn) \
4103 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4106 static int __meminit
zone_batchsize(struct zone
*zone
)
4112 * The per-cpu-pages pools are set to around 1000th of the
4113 * size of the zone. But no more than 1/2 of a meg.
4115 * OK, so we don't know how big the cache is. So guess.
4117 batch
= zone
->managed_pages
/ 1024;
4118 if (batch
* PAGE_SIZE
> 512 * 1024)
4119 batch
= (512 * 1024) / PAGE_SIZE
;
4120 batch
/= 4; /* We effectively *= 4 below */
4125 * Clamp the batch to a 2^n - 1 value. Having a power
4126 * of 2 value was found to be more likely to have
4127 * suboptimal cache aliasing properties in some cases.
4129 * For example if 2 tasks are alternately allocating
4130 * batches of pages, one task can end up with a lot
4131 * of pages of one half of the possible page colors
4132 * and the other with pages of the other colors.
4134 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4139 /* The deferral and batching of frees should be suppressed under NOMMU
4142 * The problem is that NOMMU needs to be able to allocate large chunks
4143 * of contiguous memory as there's no hardware page translation to
4144 * assemble apparent contiguous memory from discontiguous pages.
4146 * Queueing large contiguous runs of pages for batching, however,
4147 * causes the pages to actually be freed in smaller chunks. As there
4148 * can be a significant delay between the individual batches being
4149 * recycled, this leads to the once large chunks of space being
4150 * fragmented and becoming unavailable for high-order allocations.
4156 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4158 struct per_cpu_pages
*pcp
;
4161 memset(p
, 0, sizeof(*p
));
4165 pcp
->high
= 6 * batch
;
4166 pcp
->batch
= max(1UL, 1 * batch
);
4167 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4168 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4172 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
4173 * to the value high for the pageset p.
4176 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
4179 struct per_cpu_pages
*pcp
;
4183 pcp
->batch
= max(1UL, high
/4);
4184 if ((high
/4) > (PAGE_SHIFT
* 8))
4185 pcp
->batch
= PAGE_SHIFT
* 8;
4188 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4192 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4194 for_each_possible_cpu(cpu
) {
4195 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4197 setup_pageset(pcp
, zone_batchsize(zone
));
4199 if (percpu_pagelist_fraction
)
4200 setup_pagelist_highmark(pcp
,
4201 (zone
->managed_pages
/
4202 percpu_pagelist_fraction
));
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
)
4222 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4226 * The per-page waitqueue mechanism uses hashed waitqueues
4229 zone
->wait_table_hash_nr_entries
=
4230 wait_table_hash_nr_entries(zone_size_pages
);
4231 zone
->wait_table_bits
=
4232 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4233 alloc_size
= zone
->wait_table_hash_nr_entries
4234 * sizeof(wait_queue_head_t
);
4236 if (!slab_is_available()) {
4237 zone
->wait_table
= (wait_queue_head_t
*)
4238 alloc_bootmem_node_nopanic(pgdat
, alloc_size
);
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 (zone
->present_pages
)
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 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4324 if (start_pfn
<= pfn
&& pfn
< end_pfn
) {
4325 last_start_pfn
= start_pfn
;
4326 last_end_pfn
= end_pfn
;
4330 /* This is a memory hole */
4333 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4335 int __meminit
early_pfn_to_nid(unsigned long pfn
)
4339 nid
= __early_pfn_to_nid(pfn
);
4342 /* just returns 0 */
4346 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4347 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
4351 nid
= __early_pfn_to_nid(pfn
);
4352 if (nid
>= 0 && nid
!= node
)
4359 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
4360 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4361 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
4363 * If an architecture guarantees that all ranges registered with
4364 * add_active_ranges() contain no holes and may be freed, this
4365 * this function may be used instead of calling free_bootmem() manually.
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 free_bootmem_node(NODE_DATA(this_nid
),
4378 PFN_PHYS(start_pfn
),
4379 (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 *ignored
)
4494 unsigned long node_start_pfn
, node_end_pfn
;
4495 unsigned long zone_start_pfn
, zone_end_pfn
;
4497 /* Get the start and end of the node and zone */
4498 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
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 *ignored
)
4555 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4556 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4557 unsigned long node_start_pfn
, node_end_pfn
;
4558 unsigned long zone_start_pfn
, zone_end_pfn
;
4560 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_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 *zones_size
)
4575 return zones_size
[zone_type
];
4578 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4579 unsigned long zone_type
,
4580 unsigned long *zholes_size
)
4585 return zholes_size
[zone_type
];
4588 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4590 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4591 unsigned long *zones_size
, unsigned long *zholes_size
)
4593 unsigned long realtotalpages
, totalpages
= 0;
4596 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4597 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4599 pgdat
->node_spanned_pages
= totalpages
;
4601 realtotalpages
= totalpages
;
4602 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4604 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4606 pgdat
->node_present_pages
= realtotalpages
;
4607 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4611 #ifndef CONFIG_SPARSEMEM
4613 * Calculate the size of the zone->blockflags rounded to an unsigned long
4614 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4615 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4616 * round what is now in bits to nearest long in bits, then return it in
4619 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
4621 unsigned long usemapsize
;
4623 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
4624 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4625 usemapsize
= usemapsize
>> pageblock_order
;
4626 usemapsize
*= NR_PAGEBLOCK_BITS
;
4627 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4629 return usemapsize
/ 8;
4632 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4634 unsigned long zone_start_pfn
,
4635 unsigned long zonesize
)
4637 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
4638 zone
->pageblock_flags
= NULL
;
4640 zone
->pageblock_flags
= alloc_bootmem_node_nopanic(pgdat
,
4644 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
4645 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
4646 #endif /* CONFIG_SPARSEMEM */
4648 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4650 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4651 void __paginginit
set_pageblock_order(void)
4655 /* Check that pageblock_nr_pages has not already been setup */
4656 if (pageblock_order
)
4659 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4660 order
= HUGETLB_PAGE_ORDER
;
4662 order
= MAX_ORDER
- 1;
4665 * Assume the largest contiguous order of interest is a huge page.
4666 * This value may be variable depending on boot parameters on IA64 and
4669 pageblock_order
= order
;
4671 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4674 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4675 * is unused as pageblock_order is set at compile-time. See
4676 * include/linux/pageblock-flags.h for the values of pageblock_order based on
4679 void __paginginit
set_pageblock_order(void)
4683 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4685 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
4686 unsigned long present_pages
)
4688 unsigned long pages
= spanned_pages
;
4691 * Provide a more accurate estimation if there are holes within
4692 * the zone and SPARSEMEM is in use. If there are holes within the
4693 * zone, each populated memory region may cost us one or two extra
4694 * memmap pages due to alignment because memmap pages for each
4695 * populated regions may not naturally algined on page boundary.
4696 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4698 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
4699 IS_ENABLED(CONFIG_SPARSEMEM
))
4700 pages
= present_pages
;
4702 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
4706 * Set up the zone data structures:
4707 * - mark all pages reserved
4708 * - mark all memory queues empty
4709 * - clear the memory bitmaps
4711 * NOTE: pgdat should get zeroed by caller.
4713 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4714 unsigned long *zones_size
, unsigned long *zholes_size
)
4717 int nid
= pgdat
->node_id
;
4718 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4721 pgdat_resize_init(pgdat
);
4722 #ifdef CONFIG_NUMA_BALANCING
4723 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
4724 pgdat
->numabalancing_migrate_nr_pages
= 0;
4725 pgdat
->numabalancing_migrate_next_window
= jiffies
;
4727 init_waitqueue_head(&pgdat
->kswapd_wait
);
4728 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
4729 pgdat_page_cgroup_init(pgdat
);
4731 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4732 struct zone
*zone
= pgdat
->node_zones
+ j
;
4733 unsigned long size
, realsize
, freesize
, memmap_pages
;
4735 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4736 realsize
= freesize
= size
- zone_absent_pages_in_node(nid
, j
,
4740 * Adjust freesize so that it accounts for how much memory
4741 * is used by this zone for memmap. This affects the watermark
4742 * and per-cpu initialisations
4744 memmap_pages
= calc_memmap_size(size
, realsize
);
4745 if (freesize
>= memmap_pages
) {
4746 freesize
-= memmap_pages
;
4749 " %s zone: %lu pages used for memmap\n",
4750 zone_names
[j
], memmap_pages
);
4753 " %s zone: %lu pages exceeds freesize %lu\n",
4754 zone_names
[j
], memmap_pages
, freesize
);
4756 /* Account for reserved pages */
4757 if (j
== 0 && freesize
> dma_reserve
) {
4758 freesize
-= dma_reserve
;
4759 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4760 zone_names
[0], dma_reserve
);
4763 if (!is_highmem_idx(j
))
4764 nr_kernel_pages
+= freesize
;
4765 /* Charge for highmem memmap if there are enough kernel pages */
4766 else if (nr_kernel_pages
> memmap_pages
* 2)
4767 nr_kernel_pages
-= memmap_pages
;
4768 nr_all_pages
+= freesize
;
4770 zone
->spanned_pages
= size
;
4771 zone
->present_pages
= realsize
;
4773 * Set an approximate value for lowmem here, it will be adjusted
4774 * when the bootmem allocator frees pages into the buddy system.
4775 * And all highmem pages will be managed by the buddy system.
4777 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
4780 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
4782 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
4784 zone
->name
= zone_names
[j
];
4785 spin_lock_init(&zone
->lock
);
4786 spin_lock_init(&zone
->lru_lock
);
4787 zone_seqlock_init(zone
);
4788 zone
->zone_pgdat
= pgdat
;
4790 zone_pcp_init(zone
);
4791 lruvec_init(&zone
->lruvec
);
4795 set_pageblock_order();
4796 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
4797 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4798 size
, MEMMAP_EARLY
);
4800 memmap_init(size
, nid
, j
, zone_start_pfn
);
4801 zone_start_pfn
+= size
;
4805 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4807 /* Skip empty nodes */
4808 if (!pgdat
->node_spanned_pages
)
4811 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4812 /* ia64 gets its own node_mem_map, before this, without bootmem */
4813 if (!pgdat
->node_mem_map
) {
4814 unsigned long size
, start
, end
;
4818 * The zone's endpoints aren't required to be MAX_ORDER
4819 * aligned but the node_mem_map endpoints must be in order
4820 * for the buddy allocator to function correctly.
4822 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4823 end
= pgdat_end_pfn(pgdat
);
4824 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4825 size
= (end
- start
) * sizeof(struct page
);
4826 map
= alloc_remap(pgdat
->node_id
, size
);
4828 map
= alloc_bootmem_node_nopanic(pgdat
, size
);
4829 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4831 #ifndef CONFIG_NEED_MULTIPLE_NODES
4833 * With no DISCONTIG, the global mem_map is just set as node 0's
4835 if (pgdat
== NODE_DATA(0)) {
4836 mem_map
= NODE_DATA(0)->node_mem_map
;
4837 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4838 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4839 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4840 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4843 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4846 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4847 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4849 pg_data_t
*pgdat
= NODE_DATA(nid
);
4851 /* pg_data_t should be reset to zero when it's allocated */
4852 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
4854 pgdat
->node_id
= nid
;
4855 pgdat
->node_start_pfn
= node_start_pfn
;
4856 init_zone_allows_reclaim(nid
);
4857 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4859 alloc_node_mem_map(pgdat
);
4860 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4861 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4862 nid
, (unsigned long)pgdat
,
4863 (unsigned long)pgdat
->node_mem_map
);
4866 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4869 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4871 #if MAX_NUMNODES > 1
4873 * Figure out the number of possible node ids.
4875 void __init
setup_nr_node_ids(void)
4878 unsigned int highest
= 0;
4880 for_each_node_mask(node
, node_possible_map
)
4882 nr_node_ids
= highest
+ 1;
4887 * node_map_pfn_alignment - determine the maximum internode alignment
4889 * This function should be called after node map is populated and sorted.
4890 * It calculates the maximum power of two alignment which can distinguish
4893 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4894 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4895 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4896 * shifted, 1GiB is enough and this function will indicate so.
4898 * This is used to test whether pfn -> nid mapping of the chosen memory
4899 * model has fine enough granularity to avoid incorrect mapping for the
4900 * populated node map.
4902 * Returns the determined alignment in pfn's. 0 if there is no alignment
4903 * requirement (single node).
4905 unsigned long __init
node_map_pfn_alignment(void)
4907 unsigned long accl_mask
= 0, last_end
= 0;
4908 unsigned long start
, end
, mask
;
4912 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4913 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4920 * Start with a mask granular enough to pin-point to the
4921 * start pfn and tick off bits one-by-one until it becomes
4922 * too coarse to separate the current node from the last.
4924 mask
= ~((1 << __ffs(start
)) - 1);
4925 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4928 /* accumulate all internode masks */
4932 /* convert mask to number of pages */
4933 return ~accl_mask
+ 1;
4936 /* Find the lowest pfn for a node */
4937 static unsigned long __init
find_min_pfn_for_node(int nid
)
4939 unsigned long min_pfn
= ULONG_MAX
;
4940 unsigned long start_pfn
;
4943 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4944 min_pfn
= min(min_pfn
, start_pfn
);
4946 if (min_pfn
== ULONG_MAX
) {
4948 "Could not find start_pfn for node %d\n", nid
);
4956 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4958 * It returns the minimum PFN based on information provided via
4959 * add_active_range().
4961 unsigned long __init
find_min_pfn_with_active_regions(void)
4963 return find_min_pfn_for_node(MAX_NUMNODES
);
4967 * early_calculate_totalpages()
4968 * Sum pages in active regions for movable zone.
4969 * Populate N_MEMORY for calculating usable_nodes.
4971 static unsigned long __init
early_calculate_totalpages(void)
4973 unsigned long totalpages
= 0;
4974 unsigned long start_pfn
, end_pfn
;
4977 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
4978 unsigned long pages
= end_pfn
- start_pfn
;
4980 totalpages
+= pages
;
4982 node_set_state(nid
, N_MEMORY
);
4988 * Find the PFN the Movable zone begins in each node. Kernel memory
4989 * is spread evenly between nodes as long as the nodes have enough
4990 * memory. When they don't, some nodes will have more kernelcore than
4993 static void __init
find_zone_movable_pfns_for_nodes(void)
4996 unsigned long usable_startpfn
;
4997 unsigned long kernelcore_node
, kernelcore_remaining
;
4998 /* save the state before borrow the nodemask */
4999 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5000 unsigned long totalpages
= early_calculate_totalpages();
5001 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5004 * If movablecore was specified, calculate what size of
5005 * kernelcore that corresponds so that memory usable for
5006 * any allocation type is evenly spread. If both kernelcore
5007 * and movablecore are specified, then the value of kernelcore
5008 * will be used for required_kernelcore if it's greater than
5009 * what movablecore would have allowed.
5011 if (required_movablecore
) {
5012 unsigned long corepages
;
5015 * Round-up so that ZONE_MOVABLE is at least as large as what
5016 * was requested by the user
5018 required_movablecore
=
5019 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5020 corepages
= totalpages
- required_movablecore
;
5022 required_kernelcore
= max(required_kernelcore
, corepages
);
5025 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5026 if (!required_kernelcore
)
5029 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5030 find_usable_zone_for_movable();
5031 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5034 /* Spread kernelcore memory as evenly as possible throughout nodes */
5035 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5036 for_each_node_state(nid
, N_MEMORY
) {
5037 unsigned long start_pfn
, end_pfn
;
5040 * Recalculate kernelcore_node if the division per node
5041 * now exceeds what is necessary to satisfy the requested
5042 * amount of memory for the kernel
5044 if (required_kernelcore
< kernelcore_node
)
5045 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5048 * As the map is walked, we track how much memory is usable
5049 * by the kernel using kernelcore_remaining. When it is
5050 * 0, the rest of the node is usable by ZONE_MOVABLE
5052 kernelcore_remaining
= kernelcore_node
;
5054 /* Go through each range of PFNs within this node */
5055 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5056 unsigned long size_pages
;
5058 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5059 if (start_pfn
>= end_pfn
)
5062 /* Account for what is only usable for kernelcore */
5063 if (start_pfn
< usable_startpfn
) {
5064 unsigned long kernel_pages
;
5065 kernel_pages
= min(end_pfn
, usable_startpfn
)
5068 kernelcore_remaining
-= min(kernel_pages
,
5069 kernelcore_remaining
);
5070 required_kernelcore
-= min(kernel_pages
,
5071 required_kernelcore
);
5073 /* Continue if range is now fully accounted */
5074 if (end_pfn
<= usable_startpfn
) {
5077 * Push zone_movable_pfn to the end so
5078 * that if we have to rebalance
5079 * kernelcore across nodes, we will
5080 * not double account here
5082 zone_movable_pfn
[nid
] = end_pfn
;
5085 start_pfn
= usable_startpfn
;
5089 * The usable PFN range for ZONE_MOVABLE is from
5090 * start_pfn->end_pfn. Calculate size_pages as the
5091 * number of pages used as kernelcore
5093 size_pages
= end_pfn
- start_pfn
;
5094 if (size_pages
> kernelcore_remaining
)
5095 size_pages
= kernelcore_remaining
;
5096 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5099 * Some kernelcore has been met, update counts and
5100 * break if the kernelcore for this node has been
5103 required_kernelcore
-= min(required_kernelcore
,
5105 kernelcore_remaining
-= size_pages
;
5106 if (!kernelcore_remaining
)
5112 * If there is still required_kernelcore, we do another pass with one
5113 * less node in the count. This will push zone_movable_pfn[nid] further
5114 * along on the nodes that still have memory until kernelcore is
5118 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5121 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5122 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5123 zone_movable_pfn
[nid
] =
5124 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5127 /* restore the node_state */
5128 node_states
[N_MEMORY
] = saved_node_state
;
5131 /* Any regular or high memory on that node ? */
5132 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5134 enum zone_type zone_type
;
5136 if (N_MEMORY
== N_NORMAL_MEMORY
)
5139 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5140 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5141 if (zone
->present_pages
) {
5142 node_set_state(nid
, N_HIGH_MEMORY
);
5143 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5144 zone_type
<= ZONE_NORMAL
)
5145 node_set_state(nid
, N_NORMAL_MEMORY
);
5152 * free_area_init_nodes - Initialise all pg_data_t and zone data
5153 * @max_zone_pfn: an array of max PFNs for each zone
5155 * This will call free_area_init_node() for each active node in the system.
5156 * Using the page ranges provided by add_active_range(), the size of each
5157 * zone in each node and their holes is calculated. If the maximum PFN
5158 * between two adjacent zones match, it is assumed that the zone is empty.
5159 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5160 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5161 * starts where the previous one ended. For example, ZONE_DMA32 starts
5162 * at arch_max_dma_pfn.
5164 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5166 unsigned long start_pfn
, end_pfn
;
5169 /* Record where the zone boundaries are */
5170 memset(arch_zone_lowest_possible_pfn
, 0,
5171 sizeof(arch_zone_lowest_possible_pfn
));
5172 memset(arch_zone_highest_possible_pfn
, 0,
5173 sizeof(arch_zone_highest_possible_pfn
));
5175 start_pfn
= find_min_pfn_with_active_regions();
5177 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5178 if (i
== ZONE_MOVABLE
)
5181 end_pfn
= max(max_zone_pfn
[i
], start_pfn
);
5182 arch_zone_lowest_possible_pfn
[i
] = start_pfn
;
5183 arch_zone_highest_possible_pfn
[i
] = end_pfn
;
5185 start_pfn
= end_pfn
;
5187 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5188 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5190 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5191 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5192 find_zone_movable_pfns_for_nodes();
5194 /* Print out the zone ranges */
5195 printk("Zone ranges:\n");
5196 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5197 if (i
== ZONE_MOVABLE
)
5199 printk(KERN_CONT
" %-8s ", zone_names
[i
]);
5200 if (arch_zone_lowest_possible_pfn
[i
] ==
5201 arch_zone_highest_possible_pfn
[i
])
5202 printk(KERN_CONT
"empty\n");
5204 printk(KERN_CONT
"[mem %0#10lx-%0#10lx]\n",
5205 arch_zone_lowest_possible_pfn
[i
] << PAGE_SHIFT
,
5206 (arch_zone_highest_possible_pfn
[i
]
5207 << PAGE_SHIFT
) - 1);
5210 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5211 printk("Movable zone start for each node\n");
5212 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5213 if (zone_movable_pfn
[i
])
5214 printk(" Node %d: %#010lx\n", i
,
5215 zone_movable_pfn
[i
] << PAGE_SHIFT
);
5218 /* Print out the early node map */
5219 printk("Early memory node ranges\n");
5220 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5221 printk(" node %3d: [mem %#010lx-%#010lx]\n", nid
,
5222 start_pfn
<< PAGE_SHIFT
, (end_pfn
<< PAGE_SHIFT
) - 1);
5224 /* Initialise every node */
5225 mminit_verify_pageflags_layout();
5226 setup_nr_node_ids();
5227 for_each_online_node(nid
) {
5228 pg_data_t
*pgdat
= NODE_DATA(nid
);
5229 free_area_init_node(nid
, NULL
,
5230 find_min_pfn_for_node(nid
), NULL
);
5232 /* Any memory on that node */
5233 if (pgdat
->node_present_pages
)
5234 node_set_state(nid
, N_MEMORY
);
5235 check_for_memory(pgdat
, nid
);
5239 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5241 unsigned long long coremem
;
5245 coremem
= memparse(p
, &p
);
5246 *core
= coremem
>> PAGE_SHIFT
;
5248 /* Paranoid check that UL is enough for the coremem value */
5249 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5255 * kernelcore=size sets the amount of memory for use for allocations that
5256 * cannot be reclaimed or migrated.
5258 static int __init
cmdline_parse_kernelcore(char *p
)
5260 return cmdline_parse_core(p
, &required_kernelcore
);
5264 * movablecore=size sets the amount of memory for use for allocations that
5265 * can be reclaimed or migrated.
5267 static int __init
cmdline_parse_movablecore(char *p
)
5269 return cmdline_parse_core(p
, &required_movablecore
);
5272 early_param("kernelcore", cmdline_parse_kernelcore
);
5273 early_param("movablecore", cmdline_parse_movablecore
);
5275 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5277 unsigned long free_reserved_area(unsigned long start
, unsigned long end
,
5278 int poison
, char *s
)
5280 unsigned long pages
, pos
;
5282 pos
= start
= PAGE_ALIGN(start
);
5284 for (pages
= 0; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5286 memset((void *)pos
, poison
, PAGE_SIZE
);
5287 free_reserved_page(virt_to_page((void *)pos
));
5291 pr_info("Freeing %s memory: %ldK\n",
5292 s
, pages
<< (PAGE_SHIFT
- 10));
5297 #ifdef CONFIG_HIGHMEM
5298 void free_highmem_page(struct page
*page
)
5300 __free_reserved_page(page
);
5307 * set_dma_reserve - set the specified number of pages reserved in the first zone
5308 * @new_dma_reserve: The number of pages to mark reserved
5310 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5311 * In the DMA zone, a significant percentage may be consumed by kernel image
5312 * and other unfreeable allocations which can skew the watermarks badly. This
5313 * function may optionally be used to account for unfreeable pages in the
5314 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5315 * smaller per-cpu batchsize.
5317 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5319 dma_reserve
= new_dma_reserve
;
5322 void __init
free_area_init(unsigned long *zones_size
)
5324 free_area_init_node(0, zones_size
,
5325 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5328 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5329 unsigned long action
, void *hcpu
)
5331 int cpu
= (unsigned long)hcpu
;
5333 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5334 lru_add_drain_cpu(cpu
);
5338 * Spill the event counters of the dead processor
5339 * into the current processors event counters.
5340 * This artificially elevates the count of the current
5343 vm_events_fold_cpu(cpu
);
5346 * Zero the differential counters of the dead processor
5347 * so that the vm statistics are consistent.
5349 * This is only okay since the processor is dead and cannot
5350 * race with what we are doing.
5352 refresh_cpu_vm_stats(cpu
);
5357 void __init
page_alloc_init(void)
5359 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5363 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5364 * or min_free_kbytes changes.
5366 static void calculate_totalreserve_pages(void)
5368 struct pglist_data
*pgdat
;
5369 unsigned long reserve_pages
= 0;
5370 enum zone_type i
, j
;
5372 for_each_online_pgdat(pgdat
) {
5373 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5374 struct zone
*zone
= pgdat
->node_zones
+ i
;
5375 unsigned long max
= 0;
5377 /* Find valid and maximum lowmem_reserve in the zone */
5378 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5379 if (zone
->lowmem_reserve
[j
] > max
)
5380 max
= zone
->lowmem_reserve
[j
];
5383 /* we treat the high watermark as reserved pages. */
5384 max
+= high_wmark_pages(zone
);
5386 if (max
> zone
->managed_pages
)
5387 max
= zone
->managed_pages
;
5388 reserve_pages
+= max
;
5390 * Lowmem reserves are not available to
5391 * GFP_HIGHUSER page cache allocations and
5392 * kswapd tries to balance zones to their high
5393 * watermark. As a result, neither should be
5394 * regarded as dirtyable memory, to prevent a
5395 * situation where reclaim has to clean pages
5396 * in order to balance the zones.
5398 zone
->dirty_balance_reserve
= max
;
5401 dirty_balance_reserve
= reserve_pages
;
5402 totalreserve_pages
= reserve_pages
;
5406 * setup_per_zone_lowmem_reserve - called whenever
5407 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
5408 * has a correct pages reserved value, so an adequate number of
5409 * pages are left in the zone after a successful __alloc_pages().
5411 static void setup_per_zone_lowmem_reserve(void)
5413 struct pglist_data
*pgdat
;
5414 enum zone_type j
, idx
;
5416 for_each_online_pgdat(pgdat
) {
5417 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5418 struct zone
*zone
= pgdat
->node_zones
+ j
;
5419 unsigned long managed_pages
= zone
->managed_pages
;
5421 zone
->lowmem_reserve
[j
] = 0;
5425 struct zone
*lower_zone
;
5429 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5430 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5432 lower_zone
= pgdat
->node_zones
+ idx
;
5433 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
5434 sysctl_lowmem_reserve_ratio
[idx
];
5435 managed_pages
+= lower_zone
->managed_pages
;
5440 /* update totalreserve_pages */
5441 calculate_totalreserve_pages();
5444 static void __setup_per_zone_wmarks(void)
5446 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5447 unsigned long pages_low
= extra_free_kbytes
>> (PAGE_SHIFT
- 10);
5448 unsigned long lowmem_pages
= 0;
5450 unsigned long flags
;
5452 /* Calculate total number of !ZONE_HIGHMEM pages */
5453 for_each_zone(zone
) {
5454 if (!is_highmem(zone
))
5455 lowmem_pages
+= zone
->managed_pages
;
5458 for_each_zone(zone
) {
5461 spin_lock_irqsave(&zone
->lock
, flags
);
5462 min
= (u64
)pages_min
* zone
->managed_pages
;
5463 do_div(min
, lowmem_pages
);
5464 low
= (u64
)pages_low
* zone
->managed_pages
;
5465 do_div(low
, vm_total_pages
);
5467 if (is_highmem(zone
)) {
5469 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5470 * need highmem pages, so cap pages_min to a small
5473 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5474 * deltas controls asynch page reclaim, and so should
5475 * not be capped for highmem.
5477 unsigned long min_pages
;
5479 min_pages
= zone
->managed_pages
/ 1024;
5480 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
5481 zone
->watermark
[WMARK_MIN
] = min_pages
;
5484 * If it's a lowmem zone, reserve a number of pages
5485 * proportionate to the zone's size.
5487 zone
->watermark
[WMARK_MIN
] = min
;
5490 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) +
5492 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) +
5495 setup_zone_migrate_reserve(zone
);
5496 spin_unlock_irqrestore(&zone
->lock
, flags
);
5499 /* update totalreserve_pages */
5500 calculate_totalreserve_pages();
5504 * setup_per_zone_wmarks - called when min_free_kbytes changes
5505 * or when memory is hot-{added|removed}
5507 * Ensures that the watermark[min,low,high] values for each zone are set
5508 * correctly with respect to min_free_kbytes.
5510 void setup_per_zone_wmarks(void)
5512 mutex_lock(&zonelists_mutex
);
5513 __setup_per_zone_wmarks();
5514 mutex_unlock(&zonelists_mutex
);
5518 * The inactive anon list should be small enough that the VM never has to
5519 * do too much work, but large enough that each inactive page has a chance
5520 * to be referenced again before it is swapped out.
5522 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5523 * INACTIVE_ANON pages on this zone's LRU, maintained by the
5524 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5525 * the anonymous pages are kept on the inactive list.
5528 * memory ratio inactive anon
5529 * -------------------------------------
5538 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
5540 #ifdef CONFIG_FIX_INACTIVE_RATIO
5541 zone
->inactive_ratio
= 1;
5543 unsigned int gb
, ratio
;
5545 /* Zone size in gigabytes */
5546 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
5548 ratio
= int_sqrt(10 * gb
);
5552 zone
->inactive_ratio
= ratio
;
5556 static void __meminit
setup_per_zone_inactive_ratio(void)
5561 calculate_zone_inactive_ratio(zone
);
5565 * Initialise min_free_kbytes.
5567 * For small machines we want it small (128k min). For large machines
5568 * we want it large (64MB max). But it is not linear, because network
5569 * bandwidth does not increase linearly with machine size. We use
5571 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5572 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
5588 int __meminit
init_per_zone_wmark_min(void)
5590 unsigned long lowmem_kbytes
;
5592 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
5594 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
5595 if (min_free_kbytes
< 128)
5596 min_free_kbytes
= 128;
5597 if (min_free_kbytes
> 65536)
5598 min_free_kbytes
= 65536;
5599 setup_per_zone_wmarks();
5600 refresh_zone_stat_thresholds();
5601 setup_per_zone_lowmem_reserve();
5602 setup_per_zone_inactive_ratio();
5605 module_init(init_per_zone_wmark_min
)
5608 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5609 * that we can call two helper functions whenever min_free_kbytes
5610 * or extra_free_kbytes changes.
5612 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
5613 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5615 proc_dointvec(table
, write
, buffer
, length
, ppos
);
5617 setup_per_zone_wmarks();
5622 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
5623 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5628 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5633 zone
->min_unmapped_pages
= (zone
->managed_pages
*
5634 sysctl_min_unmapped_ratio
) / 100;
5638 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
5639 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5644 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5649 zone
->min_slab_pages
= (zone
->managed_pages
*
5650 sysctl_min_slab_ratio
) / 100;
5656 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5657 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5658 * whenever sysctl_lowmem_reserve_ratio changes.
5660 * The reserve ratio obviously has absolutely no relation with the
5661 * minimum watermarks. The lowmem reserve ratio can only make sense
5662 * if in function of the boot time zone sizes.
5664 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
5665 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5667 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5668 setup_per_zone_lowmem_reserve();
5673 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5674 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
5675 * can have before it gets flushed back to buddy allocator.
5678 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5679 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5685 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5686 if (!write
|| (ret
< 0))
5688 for_each_populated_zone(zone
) {
5689 for_each_possible_cpu(cpu
) {
5691 high
= zone
->managed_pages
/ percpu_pagelist_fraction
;
5692 setup_pagelist_highmark(
5693 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5699 int hashdist
= HASHDIST_DEFAULT
;
5702 static int __init
set_hashdist(char *str
)
5706 hashdist
= simple_strtoul(str
, &str
, 0);
5709 __setup("hashdist=", set_hashdist
);
5713 * allocate a large system hash table from bootmem
5714 * - it is assumed that the hash table must contain an exact power-of-2
5715 * quantity of entries
5716 * - limit is the number of hash buckets, not the total allocation size
5718 void *__init
alloc_large_system_hash(const char *tablename
,
5719 unsigned long bucketsize
,
5720 unsigned long numentries
,
5723 unsigned int *_hash_shift
,
5724 unsigned int *_hash_mask
,
5725 unsigned long low_limit
,
5726 unsigned long high_limit
)
5728 unsigned long long max
= high_limit
;
5729 unsigned long log2qty
, size
;
5732 /* allow the kernel cmdline to have a say */
5734 /* round applicable memory size up to nearest megabyte */
5735 numentries
= nr_kernel_pages
;
5736 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5737 numentries
>>= 20 - PAGE_SHIFT
;
5738 numentries
<<= 20 - PAGE_SHIFT
;
5740 /* limit to 1 bucket per 2^scale bytes of low memory */
5741 if (scale
> PAGE_SHIFT
)
5742 numentries
>>= (scale
- PAGE_SHIFT
);
5744 numentries
<<= (PAGE_SHIFT
- scale
);
5746 /* Make sure we've got at least a 0-order allocation.. */
5747 if (unlikely(flags
& HASH_SMALL
)) {
5748 /* Makes no sense without HASH_EARLY */
5749 WARN_ON(!(flags
& HASH_EARLY
));
5750 if (!(numentries
>> *_hash_shift
)) {
5751 numentries
= 1UL << *_hash_shift
;
5752 BUG_ON(!numentries
);
5754 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5755 numentries
= PAGE_SIZE
/ bucketsize
;
5757 numentries
= roundup_pow_of_two(numentries
);
5759 /* limit allocation size to 1/16 total memory by default */
5761 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5762 do_div(max
, bucketsize
);
5764 max
= min(max
, 0x80000000ULL
);
5766 if (numentries
< low_limit
)
5767 numentries
= low_limit
;
5768 if (numentries
> max
)
5771 log2qty
= ilog2(numentries
);
5774 size
= bucketsize
<< log2qty
;
5775 if (flags
& HASH_EARLY
)
5776 table
= alloc_bootmem_nopanic(size
);
5778 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5781 * If bucketsize is not a power-of-two, we may free
5782 * some pages at the end of hash table which
5783 * alloc_pages_exact() automatically does
5785 if (get_order(size
) < MAX_ORDER
) {
5786 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5787 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5790 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5793 panic("Failed to allocate %s hash table\n", tablename
);
5795 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5798 ilog2(size
) - PAGE_SHIFT
,
5802 *_hash_shift
= log2qty
;
5804 *_hash_mask
= (1 << log2qty
) - 1;
5809 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5810 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5813 #ifdef CONFIG_SPARSEMEM
5814 return __pfn_to_section(pfn
)->pageblock_flags
;
5816 return zone
->pageblock_flags
;
5817 #endif /* CONFIG_SPARSEMEM */
5820 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5822 #ifdef CONFIG_SPARSEMEM
5823 pfn
&= (PAGES_PER_SECTION
-1);
5824 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5826 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
5827 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5828 #endif /* CONFIG_SPARSEMEM */
5832 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5833 * @page: The page within the block of interest
5834 * @start_bitidx: The first bit of interest to retrieve
5835 * @end_bitidx: The last bit of interest
5836 * returns pageblock_bits flags
5838 unsigned long get_pageblock_flags_group(struct page
*page
,
5839 int start_bitidx
, int end_bitidx
)
5842 unsigned long *bitmap
;
5843 unsigned long pfn
, bitidx
;
5844 unsigned long flags
= 0;
5845 unsigned long value
= 1;
5847 zone
= page_zone(page
);
5848 pfn
= page_to_pfn(page
);
5849 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5850 bitidx
= pfn_to_bitidx(zone
, pfn
);
5852 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5853 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5860 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5861 * @page: The page within the block of interest
5862 * @start_bitidx: The first bit of interest
5863 * @end_bitidx: The last bit of interest
5864 * @flags: The flags to set
5866 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5867 int start_bitidx
, int end_bitidx
)
5870 unsigned long *bitmap
;
5871 unsigned long pfn
, bitidx
;
5872 unsigned long value
= 1;
5874 zone
= page_zone(page
);
5875 pfn
= page_to_pfn(page
);
5876 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5877 bitidx
= pfn_to_bitidx(zone
, pfn
);
5878 VM_BUG_ON(!zone_spans_pfn(zone
, pfn
));
5880 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5882 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5884 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5888 * This function checks whether pageblock includes unmovable pages or not.
5889 * If @count is not zero, it is okay to include less @count unmovable pages
5891 * PageLRU check wihtout isolation or lru_lock could race so that
5892 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
5893 * expect this function should be exact.
5895 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
5896 bool skip_hwpoisoned_pages
)
5898 unsigned long pfn
, iter
, found
;
5902 * For avoiding noise data, lru_add_drain_all() should be called
5903 * If ZONE_MOVABLE, the zone never contains unmovable pages
5905 if (zone_idx(zone
) == ZONE_MOVABLE
)
5907 mt
= get_pageblock_migratetype(page
);
5908 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
5911 pfn
= page_to_pfn(page
);
5912 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
5913 unsigned long check
= pfn
+ iter
;
5915 if (!pfn_valid_within(check
))
5918 page
= pfn_to_page(check
);
5920 * We can't use page_count without pin a page
5921 * because another CPU can free compound page.
5922 * This check already skips compound tails of THP
5923 * because their page->_count is zero at all time.
5925 if (!atomic_read(&page
->_count
)) {
5926 if (PageBuddy(page
))
5927 iter
+= (1 << page_order(page
)) - 1;
5932 * The HWPoisoned page may be not in buddy system, and
5933 * page_count() is not 0.
5935 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
5941 * If there are RECLAIMABLE pages, we need to check it.
5942 * But now, memory offline itself doesn't call shrink_slab()
5943 * and it still to be fixed.
5946 * If the page is not RAM, page_count()should be 0.
5947 * we don't need more check. This is an _used_ not-movable page.
5949 * The problematic thing here is PG_reserved pages. PG_reserved
5950 * is set to both of a memory hole page and a _used_ kernel
5959 bool is_pageblock_removable_nolock(struct page
*page
)
5965 * We have to be careful here because we are iterating over memory
5966 * sections which are not zone aware so we might end up outside of
5967 * the zone but still within the section.
5968 * We have to take care about the node as well. If the node is offline
5969 * its NODE_DATA will be NULL - see page_zone.
5971 if (!node_online(page_to_nid(page
)))
5974 zone
= page_zone(page
);
5975 pfn
= page_to_pfn(page
);
5976 if (!zone_spans_pfn(zone
, pfn
))
5979 return !has_unmovable_pages(zone
, page
, 0, true);
5984 static unsigned long pfn_max_align_down(unsigned long pfn
)
5986 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5987 pageblock_nr_pages
) - 1);
5990 static unsigned long pfn_max_align_up(unsigned long pfn
)
5992 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
5993 pageblock_nr_pages
));
5996 /* [start, end) must belong to a single zone. */
5997 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
5998 unsigned long start
, unsigned long end
)
6000 /* This function is based on compact_zone() from compaction.c. */
6001 unsigned long nr_reclaimed
;
6002 unsigned long pfn
= start
;
6003 unsigned int tries
= 0;
6008 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6009 if (fatal_signal_pending(current
)) {
6014 if (list_empty(&cc
->migratepages
)) {
6015 cc
->nr_migratepages
= 0;
6016 pfn
= isolate_migratepages_range(cc
->zone
, cc
,
6023 } else if (++tries
== 5) {
6024 ret
= ret
< 0 ? ret
: -EBUSY
;
6028 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6030 cc
->nr_migratepages
-= nr_reclaimed
;
6032 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6033 0, MIGRATE_SYNC
, MR_CMA
);
6036 putback_movable_pages(&cc
->migratepages
);
6043 * alloc_contig_range() -- tries to allocate given range of pages
6044 * @start: start PFN to allocate
6045 * @end: one-past-the-last PFN to allocate
6046 * @migratetype: migratetype of the underlaying pageblocks (either
6047 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6048 * in range must have the same migratetype and it must
6049 * be either of the two.
6051 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6052 * aligned, however it's the caller's responsibility to guarantee that
6053 * we are the only thread that changes migrate type of pageblocks the
6056 * The PFN range must belong to a single zone.
6058 * Returns zero on success or negative error code. On success all
6059 * pages which PFN is in [start, end) are allocated for the caller and
6060 * need to be freed with free_contig_range().
6062 int alloc_contig_range(unsigned long start
, unsigned long end
,
6063 unsigned migratetype
)
6065 unsigned long outer_start
, outer_end
;
6067 struct zone
*zone
= page_zone(pfn_to_page(start
));
6069 struct compact_control cc
= {
6070 .nr_migratepages
= 0,
6074 .ignore_skip_hint
= true,
6076 INIT_LIST_HEAD(&cc
.migratepages
);
6079 * What we do here is we mark all pageblocks in range as
6080 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6081 * have different sizes, and due to the way page allocator
6082 * work, we align the range to biggest of the two pages so
6083 * that page allocator won't try to merge buddies from
6084 * different pageblocks and change MIGRATE_ISOLATE to some
6085 * other migration type.
6087 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6088 * migrate the pages from an unaligned range (ie. pages that
6089 * we are interested in). This will put all the pages in
6090 * range back to page allocator as MIGRATE_ISOLATE.
6092 * When this is done, we take the pages in range from page
6093 * allocator removing them from the buddy system. This way
6094 * page allocator will never consider using them.
6096 * This lets us mark the pageblocks back as
6097 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6098 * aligned range but not in the unaligned, original range are
6099 * put back to page allocator so that buddy can use them.
6102 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6103 pfn_max_align_up(end
), migratetype
,
6108 zone
->cma_alloc
= 1;
6110 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6115 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6116 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6117 * more, all pages in [start, end) are free in page allocator.
6118 * What we are going to do is to allocate all pages from
6119 * [start, end) (that is remove them from page allocator).
6121 * The only problem is that pages at the beginning and at the
6122 * end of interesting range may be not aligned with pages that
6123 * page allocator holds, ie. they can be part of higher order
6124 * pages. Because of this, we reserve the bigger range and
6125 * once this is done free the pages we are not interested in.
6127 * We don't have to hold zone->lock here because the pages are
6128 * isolated thus they won't get removed from buddy.
6131 lru_add_drain_all();
6135 outer_start
= start
;
6136 while (!PageBuddy(pfn_to_page(outer_start
))) {
6137 if (++order
>= MAX_ORDER
) {
6141 outer_start
&= ~0UL << order
;
6144 /* Make sure the range is really isolated. */
6145 if (test_pages_isolated(outer_start
, end
, false)) {
6146 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6153 /* Grab isolated pages from freelists. */
6154 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6160 /* Free head and tail (if any) */
6161 if (start
!= outer_start
)
6162 free_contig_range(outer_start
, start
- outer_start
);
6163 if (end
!= outer_end
)
6164 free_contig_range(end
, outer_end
- end
);
6167 undo_isolate_page_range(pfn_max_align_down(start
),
6168 pfn_max_align_up(end
), migratetype
);
6169 zone
->cma_alloc
= 0;
6173 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6175 unsigned int count
= 0;
6177 for (; nr_pages
--; pfn
++) {
6178 struct page
*page
= pfn_to_page(pfn
);
6180 count
+= page_count(page
) != 1;
6183 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6187 #ifdef CONFIG_MEMORY_HOTPLUG
6188 static int __meminit
__zone_pcp_update(void *data
)
6190 struct zone
*zone
= data
;
6192 unsigned long batch
= zone_batchsize(zone
), flags
;
6194 for_each_possible_cpu(cpu
) {
6195 struct per_cpu_pageset
*pset
;
6196 struct per_cpu_pages
*pcp
;
6198 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6201 local_irq_save(flags
);
6203 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
6204 drain_zonestat(zone
, pset
);
6205 setup_pageset(pset
, batch
);
6206 local_irq_restore(flags
);
6211 void __meminit
zone_pcp_update(struct zone
*zone
)
6213 stop_machine(__zone_pcp_update
, zone
, NULL
);
6217 void zone_pcp_reset(struct zone
*zone
)
6219 unsigned long flags
;
6221 struct per_cpu_pageset
*pset
;
6223 /* avoid races with drain_pages() */
6224 local_irq_save(flags
);
6225 if (zone
->pageset
!= &boot_pageset
) {
6226 for_each_online_cpu(cpu
) {
6227 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6228 drain_zonestat(zone
, pset
);
6230 free_percpu(zone
->pageset
);
6231 zone
->pageset
= &boot_pageset
;
6233 local_irq_restore(flags
);
6236 #ifdef CONFIG_MEMORY_HOTREMOVE
6238 * All pages in the range must be isolated before calling this.
6241 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6247 unsigned long flags
;
6248 /* find the first valid pfn */
6249 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6254 zone
= page_zone(pfn_to_page(pfn
));
6255 spin_lock_irqsave(&zone
->lock
, flags
);
6257 while (pfn
< end_pfn
) {
6258 if (!pfn_valid(pfn
)) {
6262 page
= pfn_to_page(pfn
);
6264 * The HWPoisoned page may be not in buddy system, and
6265 * page_count() is not 0.
6267 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6269 SetPageReserved(page
);
6273 BUG_ON(page_count(page
));
6274 BUG_ON(!PageBuddy(page
));
6275 order
= page_order(page
);
6276 #ifdef CONFIG_DEBUG_VM
6277 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6278 pfn
, 1 << order
, end_pfn
);
6280 list_del(&page
->lru
);
6281 rmv_page_order(page
);
6282 zone
->free_area
[order
].nr_free
--;
6283 #ifdef CONFIG_HIGHMEM
6284 if (PageHighMem(page
))
6285 totalhigh_pages
-= 1 << order
;
6287 for (i
= 0; i
< (1 << order
); i
++)
6288 SetPageReserved((page
+i
));
6289 pfn
+= (1 << order
);
6291 spin_unlock_irqrestore(&zone
->lock
, flags
);
6295 #ifdef CONFIG_MEMORY_FAILURE
6296 bool is_free_buddy_page(struct page
*page
)
6298 struct zone
*zone
= page_zone(page
);
6299 unsigned long pfn
= page_to_pfn(page
);
6300 unsigned long flags
;
6303 spin_lock_irqsave(&zone
->lock
, flags
);
6304 for (order
= 0; order
< MAX_ORDER
; order
++) {
6305 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6307 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6310 spin_unlock_irqrestore(&zone
->lock
, flags
);
6312 return order
< MAX_ORDER
;
6316 static const struct trace_print_flags pageflag_names
[] = {
6317 {1UL << PG_locked
, "locked" },
6318 {1UL << PG_error
, "error" },
6319 {1UL << PG_referenced
, "referenced" },
6320 {1UL << PG_uptodate
, "uptodate" },
6321 {1UL << PG_dirty
, "dirty" },
6322 {1UL << PG_lru
, "lru" },
6323 {1UL << PG_active
, "active" },
6324 {1UL << PG_slab
, "slab" },
6325 {1UL << PG_owner_priv_1
, "owner_priv_1" },
6326 {1UL << PG_arch_1
, "arch_1" },
6327 {1UL << PG_reserved
, "reserved" },
6328 {1UL << PG_private
, "private" },
6329 {1UL << PG_private_2
, "private_2" },
6330 {1UL << PG_writeback
, "writeback" },
6331 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6332 {1UL << PG_head
, "head" },
6333 {1UL << PG_tail
, "tail" },
6335 {1UL << PG_compound
, "compound" },
6337 {1UL << PG_swapcache
, "swapcache" },
6338 {1UL << PG_mappedtodisk
, "mappedtodisk" },
6339 {1UL << PG_reclaim
, "reclaim" },
6340 {1UL << PG_swapbacked
, "swapbacked" },
6341 {1UL << PG_unevictable
, "unevictable" },
6343 {1UL << PG_mlocked
, "mlocked" },
6345 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6346 {1UL << PG_uncached
, "uncached" },
6348 #ifdef CONFIG_MEMORY_FAILURE
6349 {1UL << PG_hwpoison
, "hwpoison" },
6351 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6352 {1UL << PG_compound_lock
, "compound_lock" },
6356 static void dump_page_flags(unsigned long flags
)
6358 const char *delim
= "";
6362 BUILD_BUG_ON(ARRAY_SIZE(pageflag_names
) != __NR_PAGEFLAGS
);
6364 printk(KERN_ALERT
"page flags: %#lx(", flags
);
6366 /* remove zone id */
6367 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
6369 for (i
= 0; i
< ARRAY_SIZE(pageflag_names
) && flags
; i
++) {
6371 mask
= pageflag_names
[i
].mask
;
6372 if ((flags
& mask
) != mask
)
6376 printk("%s%s", delim
, pageflag_names
[i
].name
);
6380 /* check for left over flags */
6382 printk("%s%#lx", delim
, flags
);
6387 void dump_page(struct page
*page
)
6390 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6391 page
, atomic_read(&page
->_count
), page_mapcount(page
),
6392 page
->mapping
, page
->index
);
6393 dump_page_flags(page
->flags
);
6394 mem_cgroup_print_bad_page(page
);