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/memory.h>
55 #include <linux/compaction.h>
56 #include <trace/events/kmem.h>
57 #include <linux/ftrace_event.h>
58 #include <linux/memcontrol.h>
59 #include <linux/prefetch.h>
61 #include <asm/tlbflush.h>
62 #include <asm/div64.h>
65 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
66 DEFINE_PER_CPU(int, numa_node
);
67 EXPORT_PER_CPU_SYMBOL(numa_node
);
70 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
72 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
73 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
74 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
75 * defined in <linux/topology.h>.
77 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
78 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
82 * Array of node states.
84 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
85 [N_POSSIBLE
] = NODE_MASK_ALL
,
86 [N_ONLINE
] = { { [0] = 1UL } },
88 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
90 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
92 [N_CPU
] = { { [0] = 1UL } },
95 EXPORT_SYMBOL(node_states
);
97 unsigned long totalram_pages __read_mostly
;
98 unsigned long totalreserve_pages __read_mostly
;
99 int percpu_pagelist_fraction
;
100 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
102 #ifdef CONFIG_PM_SLEEP
104 * The following functions are used by the suspend/hibernate code to temporarily
105 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
106 * while devices are suspended. To avoid races with the suspend/hibernate code,
107 * they should always be called with pm_mutex held (gfp_allowed_mask also should
108 * only be modified with pm_mutex held, unless the suspend/hibernate code is
109 * guaranteed not to run in parallel with that modification).
112 static gfp_t saved_gfp_mask
;
114 void pm_restore_gfp_mask(void)
116 WARN_ON(!mutex_is_locked(&pm_mutex
));
117 if (saved_gfp_mask
) {
118 gfp_allowed_mask
= saved_gfp_mask
;
123 void pm_restrict_gfp_mask(void)
125 WARN_ON(!mutex_is_locked(&pm_mutex
));
126 WARN_ON(saved_gfp_mask
);
127 saved_gfp_mask
= gfp_allowed_mask
;
128 gfp_allowed_mask
&= ~GFP_IOFS
;
130 #endif /* CONFIG_PM_SLEEP */
132 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
133 int pageblock_order __read_mostly
;
136 static void __free_pages_ok(struct page
*page
, unsigned int order
);
139 * results with 256, 32 in the lowmem_reserve sysctl:
140 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
141 * 1G machine -> (16M dma, 784M normal, 224M high)
142 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
143 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
144 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
146 * TBD: should special case ZONE_DMA32 machines here - in those we normally
147 * don't need any ZONE_NORMAL reservation
149 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
150 #ifdef CONFIG_ZONE_DMA
153 #ifdef CONFIG_ZONE_DMA32
156 #ifdef CONFIG_HIGHMEM
162 EXPORT_SYMBOL(totalram_pages
);
164 static char * const zone_names
[MAX_NR_ZONES
] = {
165 #ifdef CONFIG_ZONE_DMA
168 #ifdef CONFIG_ZONE_DMA32
172 #ifdef CONFIG_HIGHMEM
178 int min_free_kbytes
= 1024;
180 static unsigned long __meminitdata nr_kernel_pages
;
181 static unsigned long __meminitdata nr_all_pages
;
182 static unsigned long __meminitdata dma_reserve
;
184 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
185 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
186 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
187 static unsigned long __initdata required_kernelcore
;
188 static unsigned long __initdata required_movablecore
;
189 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
191 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
193 EXPORT_SYMBOL(movable_zone
);
194 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
197 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
198 int nr_online_nodes __read_mostly
= 1;
199 EXPORT_SYMBOL(nr_node_ids
);
200 EXPORT_SYMBOL(nr_online_nodes
);
203 int page_group_by_mobility_disabled __read_mostly
;
205 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
208 if (unlikely(page_group_by_mobility_disabled
))
209 migratetype
= MIGRATE_UNMOVABLE
;
211 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
212 PB_migrate
, PB_migrate_end
);
215 bool oom_killer_disabled __read_mostly
;
217 #ifdef CONFIG_DEBUG_VM
218 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
222 unsigned long pfn
= page_to_pfn(page
);
225 seq
= zone_span_seqbegin(zone
);
226 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
228 else if (pfn
< zone
->zone_start_pfn
)
230 } while (zone_span_seqretry(zone
, seq
));
235 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
237 if (!pfn_valid_within(page_to_pfn(page
)))
239 if (zone
!= page_zone(page
))
245 * Temporary debugging check for pages not lying within a given zone.
247 static int bad_range(struct zone
*zone
, struct page
*page
)
249 if (page_outside_zone_boundaries(zone
, page
))
251 if (!page_is_consistent(zone
, page
))
257 static inline int bad_range(struct zone
*zone
, struct page
*page
)
263 static void bad_page(struct page
*page
)
265 static unsigned long resume
;
266 static unsigned long nr_shown
;
267 static unsigned long nr_unshown
;
269 /* Don't complain about poisoned pages */
270 if (PageHWPoison(page
)) {
271 reset_page_mapcount(page
); /* remove PageBuddy */
276 * Allow a burst of 60 reports, then keep quiet for that minute;
277 * or allow a steady drip of one report per second.
279 if (nr_shown
== 60) {
280 if (time_before(jiffies
, resume
)) {
286 "BUG: Bad page state: %lu messages suppressed\n",
293 resume
= jiffies
+ 60 * HZ
;
295 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
296 current
->comm
, page_to_pfn(page
));
302 /* Leave bad fields for debug, except PageBuddy could make trouble */
303 reset_page_mapcount(page
); /* remove PageBuddy */
304 add_taint(TAINT_BAD_PAGE
);
308 * Higher-order pages are called "compound pages". They are structured thusly:
310 * The first PAGE_SIZE page is called the "head page".
312 * The remaining PAGE_SIZE pages are called "tail pages".
314 * All pages have PG_compound set. All tail pages have their ->first_page
315 * pointing at the head page.
317 * The first tail page's ->lru.next holds the address of the compound page's
318 * put_page() function. Its ->lru.prev holds the order of allocation.
319 * This usage means that zero-order pages may not be compound.
322 static void free_compound_page(struct page
*page
)
324 __free_pages_ok(page
, compound_order(page
));
327 void prep_compound_page(struct page
*page
, unsigned long order
)
330 int nr_pages
= 1 << order
;
332 set_compound_page_dtor(page
, free_compound_page
);
333 set_compound_order(page
, order
);
335 for (i
= 1; i
< nr_pages
; i
++) {
336 struct page
*p
= page
+ i
;
338 set_page_count(p
, 0);
339 p
->first_page
= page
;
343 /* update __split_huge_page_refcount if you change this function */
344 static int destroy_compound_page(struct page
*page
, unsigned long order
)
347 int nr_pages
= 1 << order
;
350 if (unlikely(compound_order(page
) != order
) ||
351 unlikely(!PageHead(page
))) {
356 __ClearPageHead(page
);
358 for (i
= 1; i
< nr_pages
; i
++) {
359 struct page
*p
= page
+ i
;
361 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
371 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
376 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
377 * and __GFP_HIGHMEM from hard or soft interrupt context.
379 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
380 for (i
= 0; i
< (1 << order
); i
++)
381 clear_highpage(page
+ i
);
384 static inline void set_page_order(struct page
*page
, int order
)
386 set_page_private(page
, order
);
387 __SetPageBuddy(page
);
390 static inline void rmv_page_order(struct page
*page
)
392 __ClearPageBuddy(page
);
393 set_page_private(page
, 0);
397 * Locate the struct page for both the matching buddy in our
398 * pair (buddy1) and the combined O(n+1) page they form (page).
400 * 1) Any buddy B1 will have an order O twin B2 which satisfies
401 * the following equation:
403 * For example, if the starting buddy (buddy2) is #8 its order
405 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
407 * 2) Any buddy B will have an order O+1 parent P which
408 * satisfies the following equation:
411 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
413 static inline unsigned long
414 __find_buddy_index(unsigned long page_idx
, unsigned int order
)
416 return page_idx
^ (1 << order
);
420 * This function checks whether a page is free && is the buddy
421 * we can do coalesce a page and its buddy if
422 * (a) the buddy is not in a hole &&
423 * (b) the buddy is in the buddy system &&
424 * (c) a page and its buddy have the same order &&
425 * (d) a page and its buddy are in the same zone.
427 * For recording whether a page is in the buddy system, we set ->_mapcount -2.
428 * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock.
430 * For recording page's order, we use page_private(page).
432 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
435 if (!pfn_valid_within(page_to_pfn(buddy
)))
438 if (page_zone_id(page
) != page_zone_id(buddy
))
441 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
442 VM_BUG_ON(page_count(buddy
) != 0);
449 * Freeing function for a buddy system allocator.
451 * The concept of a buddy system is to maintain direct-mapped table
452 * (containing bit values) for memory blocks of various "orders".
453 * The bottom level table contains the map for the smallest allocatable
454 * units of memory (here, pages), and each level above it describes
455 * pairs of units from the levels below, hence, "buddies".
456 * At a high level, all that happens here is marking the table entry
457 * at the bottom level available, and propagating the changes upward
458 * as necessary, plus some accounting needed to play nicely with other
459 * parts of the VM system.
460 * At each level, we keep a list of pages, which are heads of continuous
461 * free pages of length of (1 << order) and marked with _mapcount -2. Page's
462 * order is recorded in page_private(page) field.
463 * So when we are allocating or freeing one, we can derive the state of the
464 * other. That is, if we allocate a small block, and both were
465 * free, the remainder of the region must be split into blocks.
466 * If a block is freed, and its buddy is also free, then this
467 * triggers coalescing into a block of larger size.
472 static inline void __free_one_page(struct page
*page
,
473 struct zone
*zone
, unsigned int order
,
476 unsigned long page_idx
;
477 unsigned long combined_idx
;
478 unsigned long uninitialized_var(buddy_idx
);
481 if (unlikely(PageCompound(page
)))
482 if (unlikely(destroy_compound_page(page
, order
)))
485 VM_BUG_ON(migratetype
== -1);
487 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
489 VM_BUG_ON(page_idx
& ((1 << order
) - 1));
490 VM_BUG_ON(bad_range(zone
, page
));
492 while (order
< MAX_ORDER
-1) {
493 buddy_idx
= __find_buddy_index(page_idx
, order
);
494 buddy
= page
+ (buddy_idx
- page_idx
);
495 if (!page_is_buddy(page
, buddy
, order
))
498 /* Our buddy is free, merge with it and move up one order. */
499 list_del(&buddy
->lru
);
500 zone
->free_area
[order
].nr_free
--;
501 rmv_page_order(buddy
);
502 combined_idx
= buddy_idx
& page_idx
;
503 page
= page
+ (combined_idx
- page_idx
);
504 page_idx
= combined_idx
;
507 set_page_order(page
, order
);
510 * If this is not the largest possible page, check if the buddy
511 * of the next-highest order is free. If it is, it's possible
512 * that pages are being freed that will coalesce soon. In case,
513 * that is happening, add the free page to the tail of the list
514 * so it's less likely to be used soon and more likely to be merged
515 * as a higher order page
517 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
518 struct page
*higher_page
, *higher_buddy
;
519 combined_idx
= buddy_idx
& page_idx
;
520 higher_page
= page
+ (combined_idx
- page_idx
);
521 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
522 higher_buddy
= page
+ (buddy_idx
- combined_idx
);
523 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
524 list_add_tail(&page
->lru
,
525 &zone
->free_area
[order
].free_list
[migratetype
]);
530 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
532 zone
->free_area
[order
].nr_free
++;
536 * free_page_mlock() -- clean up attempts to free and mlocked() page.
537 * Page should not be on lru, so no need to fix that up.
538 * free_pages_check() will verify...
540 static inline void free_page_mlock(struct page
*page
)
542 __dec_zone_page_state(page
, NR_MLOCK
);
543 __count_vm_event(UNEVICTABLE_MLOCKFREED
);
546 static inline int free_pages_check(struct page
*page
)
548 if (unlikely(page_mapcount(page
) |
549 (page
->mapping
!= NULL
) |
550 (atomic_read(&page
->_count
) != 0) |
551 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
) |
552 (mem_cgroup_bad_page_check(page
)))) {
556 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
557 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
562 * Frees a number of pages from the PCP lists
563 * Assumes all pages on list are in same zone, and of same order.
564 * count is the number of pages to free.
566 * If the zone was previously in an "all pages pinned" state then look to
567 * see if this freeing clears that state.
569 * And clear the zone's pages_scanned counter, to hold off the "all pages are
570 * pinned" detection logic.
572 static void free_pcppages_bulk(struct zone
*zone
, int count
,
573 struct per_cpu_pages
*pcp
)
579 spin_lock(&zone
->lock
);
580 zone
->all_unreclaimable
= 0;
581 zone
->pages_scanned
= 0;
585 struct list_head
*list
;
588 * Remove pages from lists in a round-robin fashion. A
589 * batch_free count is maintained that is incremented when an
590 * empty list is encountered. This is so more pages are freed
591 * off fuller lists instead of spinning excessively around empty
596 if (++migratetype
== MIGRATE_PCPTYPES
)
598 list
= &pcp
->lists
[migratetype
];
599 } while (list_empty(list
));
601 /* This is the only non-empty list. Free them all. */
602 if (batch_free
== MIGRATE_PCPTYPES
)
603 batch_free
= to_free
;
606 page
= list_entry(list
->prev
, struct page
, lru
);
607 /* must delete as __free_one_page list manipulates */
608 list_del(&page
->lru
);
609 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
610 __free_one_page(page
, zone
, 0, page_private(page
));
611 trace_mm_page_pcpu_drain(page
, 0, page_private(page
));
612 } while (--to_free
&& --batch_free
&& !list_empty(list
));
614 __mod_zone_page_state(zone
, NR_FREE_PAGES
, count
);
615 spin_unlock(&zone
->lock
);
618 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
,
621 spin_lock(&zone
->lock
);
622 zone
->all_unreclaimable
= 0;
623 zone
->pages_scanned
= 0;
625 __free_one_page(page
, zone
, order
, migratetype
);
626 __mod_zone_page_state(zone
, NR_FREE_PAGES
, 1 << order
);
627 spin_unlock(&zone
->lock
);
630 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
635 trace_mm_page_free_direct(page
, order
);
636 kmemcheck_free_shadow(page
, order
);
639 page
->mapping
= NULL
;
640 for (i
= 0; i
< (1 << order
); i
++)
641 bad
+= free_pages_check(page
+ i
);
645 if (!PageHighMem(page
)) {
646 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
647 debug_check_no_obj_freed(page_address(page
),
650 arch_free_page(page
, order
);
651 kernel_map_pages(page
, 1 << order
, 0);
656 static void __free_pages_ok(struct page
*page
, unsigned int order
)
659 int wasMlocked
= __TestClearPageMlocked(page
);
661 if (!free_pages_prepare(page
, order
))
664 local_irq_save(flags
);
665 if (unlikely(wasMlocked
))
666 free_page_mlock(page
);
667 __count_vm_events(PGFREE
, 1 << order
);
668 free_one_page(page_zone(page
), page
, order
,
669 get_pageblock_migratetype(page
));
670 local_irq_restore(flags
);
674 * permit the bootmem allocator to evade page validation on high-order frees
676 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
679 __ClearPageReserved(page
);
680 set_page_count(page
, 0);
681 set_page_refcounted(page
);
687 for (loop
= 0; loop
< (1 << order
); loop
++) {
688 struct page
*p
= &page
[loop
];
690 if (loop
+ 1 < (1 << order
))
692 __ClearPageReserved(p
);
693 set_page_count(p
, 0);
696 set_page_refcounted(page
);
697 __free_pages(page
, order
);
703 * The order of subdivision here is critical for the IO subsystem.
704 * Please do not alter this order without good reasons and regression
705 * testing. Specifically, as large blocks of memory are subdivided,
706 * the order in which smaller blocks are delivered depends on the order
707 * they're subdivided in this function. This is the primary factor
708 * influencing the order in which pages are delivered to the IO
709 * subsystem according to empirical testing, and this is also justified
710 * by considering the behavior of a buddy system containing a single
711 * large block of memory acted on by a series of small allocations.
712 * This behavior is a critical factor in sglist merging's success.
716 static inline void expand(struct zone
*zone
, struct page
*page
,
717 int low
, int high
, struct free_area
*area
,
720 unsigned long size
= 1 << high
;
726 VM_BUG_ON(bad_range(zone
, &page
[size
]));
727 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
729 set_page_order(&page
[size
], high
);
734 * This page is about to be returned from the page allocator
736 static inline int check_new_page(struct page
*page
)
738 if (unlikely(page_mapcount(page
) |
739 (page
->mapping
!= NULL
) |
740 (atomic_read(&page
->_count
) != 0) |
741 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
) |
742 (mem_cgroup_bad_page_check(page
)))) {
749 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
753 for (i
= 0; i
< (1 << order
); i
++) {
754 struct page
*p
= page
+ i
;
755 if (unlikely(check_new_page(p
)))
759 set_page_private(page
, 0);
760 set_page_refcounted(page
);
762 arch_alloc_page(page
, order
);
763 kernel_map_pages(page
, 1 << order
, 1);
765 if (gfp_flags
& __GFP_ZERO
)
766 prep_zero_page(page
, order
, gfp_flags
);
768 if (order
&& (gfp_flags
& __GFP_COMP
))
769 prep_compound_page(page
, order
);
775 * Go through the free lists for the given migratetype and remove
776 * the smallest available page from the freelists
779 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
782 unsigned int current_order
;
783 struct free_area
* area
;
786 /* Find a page of the appropriate size in the preferred list */
787 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
788 area
= &(zone
->free_area
[current_order
]);
789 if (list_empty(&area
->free_list
[migratetype
]))
792 page
= list_entry(area
->free_list
[migratetype
].next
,
794 list_del(&page
->lru
);
795 rmv_page_order(page
);
797 expand(zone
, page
, order
, current_order
, area
, migratetype
);
806 * This array describes the order lists are fallen back to when
807 * the free lists for the desirable migrate type are depleted
809 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
810 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
811 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
812 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
813 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
817 * Move the free pages in a range to the free lists of the requested type.
818 * Note that start_page and end_pages are not aligned on a pageblock
819 * boundary. If alignment is required, use move_freepages_block()
821 static int move_freepages(struct zone
*zone
,
822 struct page
*start_page
, struct page
*end_page
,
829 #ifndef CONFIG_HOLES_IN_ZONE
831 * page_zone is not safe to call in this context when
832 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
833 * anyway as we check zone boundaries in move_freepages_block().
834 * Remove at a later date when no bug reports exist related to
835 * grouping pages by mobility
837 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
840 for (page
= start_page
; page
<= end_page
;) {
841 /* Make sure we are not inadvertently changing nodes */
842 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
844 if (!pfn_valid_within(page_to_pfn(page
))) {
849 if (!PageBuddy(page
)) {
854 order
= page_order(page
);
855 list_move(&page
->lru
,
856 &zone
->free_area
[order
].free_list
[migratetype
]);
858 pages_moved
+= 1 << order
;
864 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
867 unsigned long start_pfn
, end_pfn
;
868 struct page
*start_page
, *end_page
;
870 start_pfn
= page_to_pfn(page
);
871 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
872 start_page
= pfn_to_page(start_pfn
);
873 end_page
= start_page
+ pageblock_nr_pages
- 1;
874 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
876 /* Do not cross zone boundaries */
877 if (start_pfn
< zone
->zone_start_pfn
)
879 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
882 return move_freepages(zone
, start_page
, end_page
, migratetype
);
885 static void change_pageblock_range(struct page
*pageblock_page
,
886 int start_order
, int migratetype
)
888 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
890 while (nr_pageblocks
--) {
891 set_pageblock_migratetype(pageblock_page
, migratetype
);
892 pageblock_page
+= pageblock_nr_pages
;
896 /* Remove an element from the buddy allocator from the fallback list */
897 static inline struct page
*
898 __rmqueue_fallback(struct zone
*zone
, int order
, int start_migratetype
)
900 struct free_area
* area
;
905 /* Find the largest possible block of pages in the other list */
906 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
908 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
909 migratetype
= fallbacks
[start_migratetype
][i
];
911 /* MIGRATE_RESERVE handled later if necessary */
912 if (migratetype
== MIGRATE_RESERVE
)
915 area
= &(zone
->free_area
[current_order
]);
916 if (list_empty(&area
->free_list
[migratetype
]))
919 page
= list_entry(area
->free_list
[migratetype
].next
,
924 * If breaking a large block of pages, move all free
925 * pages to the preferred allocation list. If falling
926 * back for a reclaimable kernel allocation, be more
927 * aggressive about taking ownership of free pages
929 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
930 start_migratetype
== MIGRATE_RECLAIMABLE
||
931 page_group_by_mobility_disabled
) {
933 pages
= move_freepages_block(zone
, page
,
936 /* Claim the whole block if over half of it is free */
937 if (pages
>= (1 << (pageblock_order
-1)) ||
938 page_group_by_mobility_disabled
)
939 set_pageblock_migratetype(page
,
942 migratetype
= start_migratetype
;
945 /* Remove the page from the freelists */
946 list_del(&page
->lru
);
947 rmv_page_order(page
);
949 /* Take ownership for orders >= pageblock_order */
950 if (current_order
>= pageblock_order
)
951 change_pageblock_range(page
, current_order
,
954 expand(zone
, page
, order
, current_order
, area
, migratetype
);
956 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
957 start_migratetype
, migratetype
);
967 * Do the hard work of removing an element from the buddy allocator.
968 * Call me with the zone->lock already held.
970 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
976 page
= __rmqueue_smallest(zone
, order
, migratetype
);
978 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
979 page
= __rmqueue_fallback(zone
, order
, migratetype
);
982 * Use MIGRATE_RESERVE rather than fail an allocation. goto
983 * is used because __rmqueue_smallest is an inline function
984 * and we want just one call site
987 migratetype
= MIGRATE_RESERVE
;
992 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
997 * Obtain a specified number of elements from the buddy allocator, all under
998 * a single hold of the lock, for efficiency. Add them to the supplied list.
999 * Returns the number of new pages which were placed at *list.
1001 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1002 unsigned long count
, struct list_head
*list
,
1003 int migratetype
, int cold
)
1007 spin_lock(&zone
->lock
);
1008 for (i
= 0; i
< count
; ++i
) {
1009 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1010 if (unlikely(page
== NULL
))
1014 * Split buddy pages returned by expand() are received here
1015 * in physical page order. The page is added to the callers and
1016 * list and the list head then moves forward. From the callers
1017 * perspective, the linked list is ordered by page number in
1018 * some conditions. This is useful for IO devices that can
1019 * merge IO requests if the physical pages are ordered
1022 if (likely(cold
== 0))
1023 list_add(&page
->lru
, list
);
1025 list_add_tail(&page
->lru
, list
);
1026 set_page_private(page
, migratetype
);
1029 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1030 spin_unlock(&zone
->lock
);
1036 * Called from the vmstat counter updater to drain pagesets of this
1037 * currently executing processor on remote nodes after they have
1040 * Note that this function must be called with the thread pinned to
1041 * a single processor.
1043 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1045 unsigned long flags
;
1048 local_irq_save(flags
);
1049 if (pcp
->count
>= pcp
->batch
)
1050 to_drain
= pcp
->batch
;
1052 to_drain
= pcp
->count
;
1053 free_pcppages_bulk(zone
, to_drain
, pcp
);
1054 pcp
->count
-= to_drain
;
1055 local_irq_restore(flags
);
1060 * Drain pages of the indicated processor.
1062 * The processor must either be the current processor and the
1063 * thread pinned to the current processor or a processor that
1066 static void drain_pages(unsigned int cpu
)
1068 unsigned long flags
;
1071 for_each_populated_zone(zone
) {
1072 struct per_cpu_pageset
*pset
;
1073 struct per_cpu_pages
*pcp
;
1075 local_irq_save(flags
);
1076 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1080 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1083 local_irq_restore(flags
);
1088 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1090 void drain_local_pages(void *arg
)
1092 drain_pages(smp_processor_id());
1096 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
1098 void drain_all_pages(void)
1100 on_each_cpu(drain_local_pages
, NULL
, 1);
1103 #ifdef CONFIG_HIBERNATION
1105 void mark_free_pages(struct zone
*zone
)
1107 unsigned long pfn
, max_zone_pfn
;
1108 unsigned long flags
;
1110 struct list_head
*curr
;
1112 if (!zone
->spanned_pages
)
1115 spin_lock_irqsave(&zone
->lock
, flags
);
1117 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
1118 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1119 if (pfn_valid(pfn
)) {
1120 struct page
*page
= pfn_to_page(pfn
);
1122 if (!swsusp_page_is_forbidden(page
))
1123 swsusp_unset_page_free(page
);
1126 for_each_migratetype_order(order
, t
) {
1127 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1130 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1131 for (i
= 0; i
< (1UL << order
); i
++)
1132 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1135 spin_unlock_irqrestore(&zone
->lock
, flags
);
1137 #endif /* CONFIG_PM */
1140 * Free a 0-order page
1141 * cold == 1 ? free a cold page : free a hot page
1143 void free_hot_cold_page(struct page
*page
, int cold
)
1145 struct zone
*zone
= page_zone(page
);
1146 struct per_cpu_pages
*pcp
;
1147 unsigned long flags
;
1149 int wasMlocked
= __TestClearPageMlocked(page
);
1151 if (!free_pages_prepare(page
, 0))
1154 migratetype
= get_pageblock_migratetype(page
);
1155 set_page_private(page
, migratetype
);
1156 local_irq_save(flags
);
1157 if (unlikely(wasMlocked
))
1158 free_page_mlock(page
);
1159 __count_vm_event(PGFREE
);
1162 * We only track unmovable, reclaimable and movable on pcp lists.
1163 * Free ISOLATE pages back to the allocator because they are being
1164 * offlined but treat RESERVE as movable pages so we can get those
1165 * areas back if necessary. Otherwise, we may have to free
1166 * excessively into the page allocator
1168 if (migratetype
>= MIGRATE_PCPTYPES
) {
1169 if (unlikely(migratetype
== MIGRATE_ISOLATE
)) {
1170 free_one_page(zone
, page
, 0, migratetype
);
1173 migratetype
= MIGRATE_MOVABLE
;
1176 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1178 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1180 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1182 if (pcp
->count
>= pcp
->high
) {
1183 free_pcppages_bulk(zone
, pcp
->batch
, pcp
);
1184 pcp
->count
-= pcp
->batch
;
1188 local_irq_restore(flags
);
1192 * split_page takes a non-compound higher-order page, and splits it into
1193 * n (1<<order) sub-pages: page[0..n]
1194 * Each sub-page must be freed individually.
1196 * Note: this is probably too low level an operation for use in drivers.
1197 * Please consult with lkml before using this in your driver.
1199 void split_page(struct page
*page
, unsigned int order
)
1203 VM_BUG_ON(PageCompound(page
));
1204 VM_BUG_ON(!page_count(page
));
1206 #ifdef CONFIG_KMEMCHECK
1208 * Split shadow pages too, because free(page[0]) would
1209 * otherwise free the whole shadow.
1211 if (kmemcheck_page_is_tracked(page
))
1212 split_page(virt_to_page(page
[0].shadow
), order
);
1215 for (i
= 1; i
< (1 << order
); i
++)
1216 set_page_refcounted(page
+ i
);
1220 * Similar to split_page except the page is already free. As this is only
1221 * being used for migration, the migratetype of the block also changes.
1222 * As this is called with interrupts disabled, the caller is responsible
1223 * for calling arch_alloc_page() and kernel_map_page() after interrupts
1226 * Note: this is probably too low level an operation for use in drivers.
1227 * Please consult with lkml before using this in your driver.
1229 int split_free_page(struct page
*page
)
1232 unsigned long watermark
;
1235 BUG_ON(!PageBuddy(page
));
1237 zone
= page_zone(page
);
1238 order
= page_order(page
);
1240 /* Obey watermarks as if the page was being allocated */
1241 watermark
= low_wmark_pages(zone
) + (1 << order
);
1242 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1245 /* Remove page from free list */
1246 list_del(&page
->lru
);
1247 zone
->free_area
[order
].nr_free
--;
1248 rmv_page_order(page
);
1249 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1UL << order
));
1251 /* Split into individual pages */
1252 set_page_refcounted(page
);
1253 split_page(page
, order
);
1255 if (order
>= pageblock_order
- 1) {
1256 struct page
*endpage
= page
+ (1 << order
) - 1;
1257 for (; page
< endpage
; page
+= pageblock_nr_pages
)
1258 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1265 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1266 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1270 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1271 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1274 unsigned long flags
;
1276 int cold
= !!(gfp_flags
& __GFP_COLD
);
1279 if (likely(order
== 0)) {
1280 struct per_cpu_pages
*pcp
;
1281 struct list_head
*list
;
1283 local_irq_save(flags
);
1284 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1285 list
= &pcp
->lists
[migratetype
];
1286 if (list_empty(list
)) {
1287 pcp
->count
+= rmqueue_bulk(zone
, 0,
1290 if (unlikely(list_empty(list
)))
1295 page
= list_entry(list
->prev
, struct page
, lru
);
1297 page
= list_entry(list
->next
, struct page
, lru
);
1299 list_del(&page
->lru
);
1302 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
1304 * __GFP_NOFAIL is not to be used in new code.
1306 * All __GFP_NOFAIL callers should be fixed so that they
1307 * properly detect and handle allocation failures.
1309 * We most definitely don't want callers attempting to
1310 * allocate greater than order-1 page units with
1313 WARN_ON_ONCE(order
> 1);
1315 spin_lock_irqsave(&zone
->lock
, flags
);
1316 page
= __rmqueue(zone
, order
, migratetype
);
1317 spin_unlock(&zone
->lock
);
1320 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(1 << order
));
1323 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1324 zone_statistics(preferred_zone
, zone
, gfp_flags
);
1325 local_irq_restore(flags
);
1327 VM_BUG_ON(bad_range(zone
, page
));
1328 if (prep_new_page(page
, order
, gfp_flags
))
1333 local_irq_restore(flags
);
1337 /* The ALLOC_WMARK bits are used as an index to zone->watermark */
1338 #define ALLOC_WMARK_MIN WMARK_MIN
1339 #define ALLOC_WMARK_LOW WMARK_LOW
1340 #define ALLOC_WMARK_HIGH WMARK_HIGH
1341 #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */
1343 /* Mask to get the watermark bits */
1344 #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1)
1346 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1347 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1348 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1350 #ifdef CONFIG_FAIL_PAGE_ALLOC
1353 struct fault_attr attr
;
1355 u32 ignore_gfp_highmem
;
1356 u32 ignore_gfp_wait
;
1358 } fail_page_alloc
= {
1359 .attr
= FAULT_ATTR_INITIALIZER
,
1360 .ignore_gfp_wait
= 1,
1361 .ignore_gfp_highmem
= 1,
1365 static int __init
setup_fail_page_alloc(char *str
)
1367 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1369 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1371 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1373 if (order
< fail_page_alloc
.min_order
)
1375 if (gfp_mask
& __GFP_NOFAIL
)
1377 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1379 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1382 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1385 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1387 static int __init
fail_page_alloc_debugfs(void)
1389 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1392 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
1393 &fail_page_alloc
.attr
);
1395 return PTR_ERR(dir
);
1397 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1398 &fail_page_alloc
.ignore_gfp_wait
))
1400 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1401 &fail_page_alloc
.ignore_gfp_highmem
))
1403 if (!debugfs_create_u32("min-order", mode
, dir
,
1404 &fail_page_alloc
.min_order
))
1409 debugfs_remove_recursive(dir
);
1414 late_initcall(fail_page_alloc_debugfs
);
1416 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1418 #else /* CONFIG_FAIL_PAGE_ALLOC */
1420 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1425 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1428 * Return true if free pages are above 'mark'. This takes into account the order
1429 * of the allocation.
1431 static bool __zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1432 int classzone_idx
, int alloc_flags
, long free_pages
)
1434 /* free_pages my go negative - that's OK */
1438 free_pages
-= (1 << order
) + 1;
1439 if (alloc_flags
& ALLOC_HIGH
)
1441 if (alloc_flags
& ALLOC_HARDER
)
1444 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1446 for (o
= 0; o
< order
; o
++) {
1447 /* At the next order, this order's pages become unavailable */
1448 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1450 /* Require fewer higher order pages to be free */
1453 if (free_pages
<= min
)
1459 bool zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1460 int classzone_idx
, int alloc_flags
)
1462 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1463 zone_page_state(z
, NR_FREE_PAGES
));
1466 bool zone_watermark_ok_safe(struct zone
*z
, int order
, unsigned long mark
,
1467 int classzone_idx
, int alloc_flags
)
1469 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
1471 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
1472 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
1474 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
1480 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1481 * skip over zones that are not allowed by the cpuset, or that have
1482 * been recently (in last second) found to be nearly full. See further
1483 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1484 * that have to skip over a lot of full or unallowed zones.
1486 * If the zonelist cache is present in the passed in zonelist, then
1487 * returns a pointer to the allowed node mask (either the current
1488 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1490 * If the zonelist cache is not available for this zonelist, does
1491 * nothing and returns NULL.
1493 * If the fullzones BITMAP in the zonelist cache is stale (more than
1494 * a second since last zap'd) then we zap it out (clear its bits.)
1496 * We hold off even calling zlc_setup, until after we've checked the
1497 * first zone in the zonelist, on the theory that most allocations will
1498 * be satisfied from that first zone, so best to examine that zone as
1499 * quickly as we can.
1501 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1503 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1504 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1506 zlc
= zonelist
->zlcache_ptr
;
1510 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1511 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1512 zlc
->last_full_zap
= jiffies
;
1515 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1516 &cpuset_current_mems_allowed
:
1517 &node_states
[N_HIGH_MEMORY
];
1518 return allowednodes
;
1522 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1523 * if it is worth looking at further for free memory:
1524 * 1) Check that the zone isn't thought to be full (doesn't have its
1525 * bit set in the zonelist_cache fullzones BITMAP).
1526 * 2) Check that the zones node (obtained from the zonelist_cache
1527 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1528 * Return true (non-zero) if zone is worth looking at further, or
1529 * else return false (zero) if it is not.
1531 * This check -ignores- the distinction between various watermarks,
1532 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1533 * found to be full for any variation of these watermarks, it will
1534 * be considered full for up to one second by all requests, unless
1535 * we are so low on memory on all allowed nodes that we are forced
1536 * into the second scan of the zonelist.
1538 * In the second scan we ignore this zonelist cache and exactly
1539 * apply the watermarks to all zones, even it is slower to do so.
1540 * We are low on memory in the second scan, and should leave no stone
1541 * unturned looking for a free page.
1543 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1544 nodemask_t
*allowednodes
)
1546 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1547 int i
; /* index of *z in zonelist zones */
1548 int n
; /* node that zone *z is on */
1550 zlc
= zonelist
->zlcache_ptr
;
1554 i
= z
- zonelist
->_zonerefs
;
1557 /* This zone is worth trying if it is allowed but not full */
1558 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1562 * Given 'z' scanning a zonelist, set the corresponding bit in
1563 * zlc->fullzones, so that subsequent attempts to allocate a page
1564 * from that zone don't waste time re-examining it.
1566 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1568 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1569 int i
; /* index of *z in zonelist zones */
1571 zlc
= zonelist
->zlcache_ptr
;
1575 i
= z
- zonelist
->_zonerefs
;
1577 set_bit(i
, zlc
->fullzones
);
1581 * clear all zones full, called after direct reclaim makes progress so that
1582 * a zone that was recently full is not skipped over for up to a second
1584 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1586 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1588 zlc
= zonelist
->zlcache_ptr
;
1592 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1595 #else /* CONFIG_NUMA */
1597 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1602 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1603 nodemask_t
*allowednodes
)
1608 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1612 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
1615 #endif /* CONFIG_NUMA */
1618 * get_page_from_freelist goes through the zonelist trying to allocate
1621 static struct page
*
1622 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1623 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1624 struct zone
*preferred_zone
, int migratetype
)
1627 struct page
*page
= NULL
;
1630 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1631 int zlc_active
= 0; /* set if using zonelist_cache */
1632 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1634 classzone_idx
= zone_idx(preferred_zone
);
1637 * Scan zonelist, looking for a zone with enough free.
1638 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1640 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1641 high_zoneidx
, nodemask
) {
1642 if (NUMA_BUILD
&& zlc_active
&&
1643 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1645 if ((alloc_flags
& ALLOC_CPUSET
) &&
1646 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1649 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
1650 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1654 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
1655 if (zone_watermark_ok(zone
, order
, mark
,
1656 classzone_idx
, alloc_flags
))
1659 if (NUMA_BUILD
&& !did_zlc_setup
&& nr_online_nodes
> 1) {
1661 * we do zlc_setup if there are multiple nodes
1662 * and before considering the first zone allowed
1665 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1670 if (zone_reclaim_mode
== 0)
1671 goto this_zone_full
;
1674 * As we may have just activated ZLC, check if the first
1675 * eligible zone has failed zone_reclaim recently.
1677 if (NUMA_BUILD
&& zlc_active
&&
1678 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1681 ret
= zone_reclaim(zone
, gfp_mask
, order
);
1683 case ZONE_RECLAIM_NOSCAN
:
1686 case ZONE_RECLAIM_FULL
:
1687 /* scanned but unreclaimable */
1690 /* did we reclaim enough */
1691 if (!zone_watermark_ok(zone
, order
, mark
,
1692 classzone_idx
, alloc_flags
))
1693 goto this_zone_full
;
1698 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1699 gfp_mask
, migratetype
);
1704 zlc_mark_zone_full(zonelist
, z
);
1707 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1708 /* Disable zlc cache for second zonelist scan */
1716 * Large machines with many possible nodes should not always dump per-node
1717 * meminfo in irq context.
1719 static inline bool should_suppress_show_mem(void)
1724 ret
= in_interrupt();
1729 static DEFINE_RATELIMIT_STATE(nopage_rs
,
1730 DEFAULT_RATELIMIT_INTERVAL
,
1731 DEFAULT_RATELIMIT_BURST
);
1733 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
1735 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
1737 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
))
1741 * This documents exceptions given to allocations in certain
1742 * contexts that are allowed to allocate outside current's set
1745 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
1746 if (test_thread_flag(TIF_MEMDIE
) ||
1747 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
1748 filter
&= ~SHOW_MEM_FILTER_NODES
;
1749 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
1750 filter
&= ~SHOW_MEM_FILTER_NODES
;
1753 struct va_format vaf
;
1756 va_start(args
, fmt
);
1761 pr_warn("%pV", &vaf
);
1766 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
1767 current
->comm
, order
, gfp_mask
);
1770 if (!should_suppress_show_mem())
1775 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1776 unsigned long pages_reclaimed
)
1778 /* Do not loop if specifically requested */
1779 if (gfp_mask
& __GFP_NORETRY
)
1783 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1784 * means __GFP_NOFAIL, but that may not be true in other
1787 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1791 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1792 * specified, then we retry until we no longer reclaim any pages
1793 * (above), or we've reclaimed an order of pages at least as
1794 * large as the allocation's order. In both cases, if the
1795 * allocation still fails, we stop retrying.
1797 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1801 * Don't let big-order allocations loop unless the caller
1802 * explicitly requests that.
1804 if (gfp_mask
& __GFP_NOFAIL
)
1810 static inline struct page
*
1811 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1812 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1813 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1818 /* Acquire the OOM killer lock for the zones in zonelist */
1819 if (!try_set_zonelist_oom(zonelist
, gfp_mask
)) {
1820 schedule_timeout_uninterruptible(1);
1825 * Go through the zonelist yet one more time, keep very high watermark
1826 * here, this is only to catch a parallel oom killing, we must fail if
1827 * we're still under heavy pressure.
1829 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1830 order
, zonelist
, high_zoneidx
,
1831 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1832 preferred_zone
, migratetype
);
1836 if (!(gfp_mask
& __GFP_NOFAIL
)) {
1837 /* The OOM killer will not help higher order allocs */
1838 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1840 /* The OOM killer does not needlessly kill tasks for lowmem */
1841 if (high_zoneidx
< ZONE_NORMAL
)
1844 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1845 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1846 * The caller should handle page allocation failure by itself if
1847 * it specifies __GFP_THISNODE.
1848 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1850 if (gfp_mask
& __GFP_THISNODE
)
1853 /* Exhausted what can be done so it's blamo time */
1854 out_of_memory(zonelist
, gfp_mask
, order
, nodemask
);
1857 clear_zonelist_oom(zonelist
, gfp_mask
);
1861 #ifdef CONFIG_COMPACTION
1862 /* Try memory compaction for high-order allocations before reclaim */
1863 static struct page
*
1864 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1865 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1866 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1867 int migratetype
, unsigned long *did_some_progress
,
1868 bool sync_migration
)
1872 if (!order
|| compaction_deferred(preferred_zone
))
1875 current
->flags
|= PF_MEMALLOC
;
1876 *did_some_progress
= try_to_compact_pages(zonelist
, order
, gfp_mask
,
1877 nodemask
, sync_migration
);
1878 current
->flags
&= ~PF_MEMALLOC
;
1879 if (*did_some_progress
!= COMPACT_SKIPPED
) {
1881 /* Page migration frees to the PCP lists but we want merging */
1882 drain_pages(get_cpu());
1885 page
= get_page_from_freelist(gfp_mask
, nodemask
,
1886 order
, zonelist
, high_zoneidx
,
1887 alloc_flags
, preferred_zone
,
1890 preferred_zone
->compact_considered
= 0;
1891 preferred_zone
->compact_defer_shift
= 0;
1892 count_vm_event(COMPACTSUCCESS
);
1897 * It's bad if compaction run occurs and fails.
1898 * The most likely reason is that pages exist,
1899 * but not enough to satisfy watermarks.
1901 count_vm_event(COMPACTFAIL
);
1902 defer_compaction(preferred_zone
);
1910 static inline struct page
*
1911 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
1912 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1913 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1914 int migratetype
, unsigned long *did_some_progress
,
1915 bool sync_migration
)
1919 #endif /* CONFIG_COMPACTION */
1921 /* The really slow allocator path where we enter direct reclaim */
1922 static inline struct page
*
1923 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1924 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1925 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1926 int migratetype
, unsigned long *did_some_progress
)
1928 struct page
*page
= NULL
;
1929 struct reclaim_state reclaim_state
;
1930 bool drained
= false;
1934 /* We now go into synchronous reclaim */
1935 cpuset_memory_pressure_bump();
1936 current
->flags
|= PF_MEMALLOC
;
1937 lockdep_set_current_reclaim_state(gfp_mask
);
1938 reclaim_state
.reclaimed_slab
= 0;
1939 current
->reclaim_state
= &reclaim_state
;
1941 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1943 current
->reclaim_state
= NULL
;
1944 lockdep_clear_current_reclaim_state();
1945 current
->flags
&= ~PF_MEMALLOC
;
1949 if (unlikely(!(*did_some_progress
)))
1952 /* After successful reclaim, reconsider all zones for allocation */
1954 zlc_clear_zones_full(zonelist
);
1957 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1958 zonelist
, high_zoneidx
,
1959 alloc_flags
, preferred_zone
,
1963 * If an allocation failed after direct reclaim, it could be because
1964 * pages are pinned on the per-cpu lists. Drain them and try again
1966 if (!page
&& !drained
) {
1976 * This is called in the allocator slow-path if the allocation request is of
1977 * sufficient urgency to ignore watermarks and take other desperate measures
1979 static inline struct page
*
1980 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1981 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1982 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1988 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1989 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1990 preferred_zone
, migratetype
);
1992 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1993 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
1994 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2000 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
2001 enum zone_type high_zoneidx
,
2002 enum zone_type classzone_idx
)
2007 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
2008 wakeup_kswapd(zone
, order
, classzone_idx
);
2012 gfp_to_alloc_flags(gfp_t gfp_mask
)
2014 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2015 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2017 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2018 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2021 * The caller may dip into page reserves a bit more if the caller
2022 * cannot run direct reclaim, or if the caller has realtime scheduling
2023 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2024 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2026 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2030 * Not worth trying to allocate harder for
2031 * __GFP_NOMEMALLOC even if it can't schedule.
2033 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2034 alloc_flags
|= ALLOC_HARDER
;
2036 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2037 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2039 alloc_flags
&= ~ALLOC_CPUSET
;
2040 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2041 alloc_flags
|= ALLOC_HARDER
;
2043 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2044 if (!in_interrupt() &&
2045 ((current
->flags
& PF_MEMALLOC
) ||
2046 unlikely(test_thread_flag(TIF_MEMDIE
))))
2047 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2053 static inline struct page
*
2054 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2055 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
2056 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
2059 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2060 struct page
*page
= NULL
;
2062 unsigned long pages_reclaimed
= 0;
2063 unsigned long did_some_progress
;
2064 bool sync_migration
= false;
2067 * In the slowpath, we sanity check order to avoid ever trying to
2068 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2069 * be using allocators in order of preference for an area that is
2072 if (order
>= MAX_ORDER
) {
2073 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2078 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2079 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2080 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2081 * using a larger set of nodes after it has established that the
2082 * allowed per node queues are empty and that nodes are
2085 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
2089 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2090 wake_all_kswapd(order
, zonelist
, high_zoneidx
,
2091 zone_idx(preferred_zone
));
2094 * OK, we're below the kswapd watermark and have kicked background
2095 * reclaim. Now things get more complex, so set up alloc_flags according
2096 * to how we want to proceed.
2098 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2101 * Find the true preferred zone if the allocation is unconstrained by
2104 if (!(alloc_flags
& ALLOC_CPUSET
) && !nodemask
)
2105 first_zones_zonelist(zonelist
, high_zoneidx
, NULL
,
2109 /* This is the last chance, in general, before the goto nopage. */
2110 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
2111 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
2112 preferred_zone
, migratetype
);
2116 /* Allocate without watermarks if the context allows */
2117 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
2118 page
= __alloc_pages_high_priority(gfp_mask
, order
,
2119 zonelist
, high_zoneidx
, nodemask
,
2120 preferred_zone
, migratetype
);
2125 /* Atomic allocations - we can't balance anything */
2129 /* Avoid recursion of direct reclaim */
2130 if (current
->flags
& PF_MEMALLOC
)
2133 /* Avoid allocations with no watermarks from looping endlessly */
2134 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
2138 * Try direct compaction. The first pass is asynchronous. Subsequent
2139 * attempts after direct reclaim are synchronous
2141 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2142 zonelist
, high_zoneidx
,
2144 alloc_flags
, preferred_zone
,
2145 migratetype
, &did_some_progress
,
2149 sync_migration
= true;
2151 /* Try direct reclaim and then allocating */
2152 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
2153 zonelist
, high_zoneidx
,
2155 alloc_flags
, preferred_zone
,
2156 migratetype
, &did_some_progress
);
2161 * If we failed to make any progress reclaiming, then we are
2162 * running out of options and have to consider going OOM
2164 if (!did_some_progress
) {
2165 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
2166 if (oom_killer_disabled
)
2168 page
= __alloc_pages_may_oom(gfp_mask
, order
,
2169 zonelist
, high_zoneidx
,
2170 nodemask
, preferred_zone
,
2175 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2177 * The oom killer is not called for high-order
2178 * allocations that may fail, so if no progress
2179 * is being made, there are no other options and
2180 * retrying is unlikely to help.
2182 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2185 * The oom killer is not called for lowmem
2186 * allocations to prevent needlessly killing
2189 if (high_zoneidx
< ZONE_NORMAL
)
2197 /* Check if we should retry the allocation */
2198 pages_reclaimed
+= did_some_progress
;
2199 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
2200 /* Wait for some write requests to complete then retry */
2201 wait_iff_congested(preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
2205 * High-order allocations do not necessarily loop after
2206 * direct reclaim and reclaim/compaction depends on compaction
2207 * being called after reclaim so call directly if necessary
2209 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
2210 zonelist
, high_zoneidx
,
2212 alloc_flags
, preferred_zone
,
2213 migratetype
, &did_some_progress
,
2220 warn_alloc_failed(gfp_mask
, order
, NULL
);
2223 if (kmemcheck_enabled
)
2224 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
2230 * This is the 'heart' of the zoned buddy allocator.
2233 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
2234 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
2236 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
2237 struct zone
*preferred_zone
;
2239 int migratetype
= allocflags_to_migratetype(gfp_mask
);
2241 gfp_mask
&= gfp_allowed_mask
;
2243 lockdep_trace_alloc(gfp_mask
);
2245 might_sleep_if(gfp_mask
& __GFP_WAIT
);
2247 if (should_fail_alloc_page(gfp_mask
, order
))
2251 * Check the zones suitable for the gfp_mask contain at least one
2252 * valid zone. It's possible to have an empty zonelist as a result
2253 * of GFP_THISNODE and a memoryless node
2255 if (unlikely(!zonelist
->_zonerefs
->zone
))
2259 /* The preferred zone is used for statistics later */
2260 first_zones_zonelist(zonelist
, high_zoneidx
,
2261 nodemask
? : &cpuset_current_mems_allowed
,
2263 if (!preferred_zone
) {
2268 /* First allocation attempt */
2269 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
2270 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
2271 preferred_zone
, migratetype
);
2272 if (unlikely(!page
))
2273 page
= __alloc_pages_slowpath(gfp_mask
, order
,
2274 zonelist
, high_zoneidx
, nodemask
,
2275 preferred_zone
, migratetype
);
2278 trace_mm_page_alloc(page
, order
, gfp_mask
, migratetype
);
2281 EXPORT_SYMBOL(__alloc_pages_nodemask
);
2284 * Common helper functions.
2286 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
2291 * __get_free_pages() returns a 32-bit address, which cannot represent
2294 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
2296 page
= alloc_pages(gfp_mask
, order
);
2299 return (unsigned long) page_address(page
);
2301 EXPORT_SYMBOL(__get_free_pages
);
2303 unsigned long get_zeroed_page(gfp_t gfp_mask
)
2305 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
2307 EXPORT_SYMBOL(get_zeroed_page
);
2309 void __pagevec_free(struct pagevec
*pvec
)
2311 int i
= pagevec_count(pvec
);
2314 trace_mm_pagevec_free(pvec
->pages
[i
], pvec
->cold
);
2315 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
2319 void __free_pages(struct page
*page
, unsigned int order
)
2321 if (put_page_testzero(page
)) {
2323 free_hot_cold_page(page
, 0);
2325 __free_pages_ok(page
, order
);
2329 EXPORT_SYMBOL(__free_pages
);
2331 void free_pages(unsigned long addr
, unsigned int order
)
2334 VM_BUG_ON(!virt_addr_valid((void *)addr
));
2335 __free_pages(virt_to_page((void *)addr
), order
);
2339 EXPORT_SYMBOL(free_pages
);
2341 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
2344 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
2345 unsigned long used
= addr
+ PAGE_ALIGN(size
);
2347 split_page(virt_to_page((void *)addr
), order
);
2348 while (used
< alloc_end
) {
2353 return (void *)addr
;
2357 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2358 * @size: the number of bytes to allocate
2359 * @gfp_mask: GFP flags for the allocation
2361 * This function is similar to alloc_pages(), except that it allocates the
2362 * minimum number of pages to satisfy the request. alloc_pages() can only
2363 * allocate memory in power-of-two pages.
2365 * This function is also limited by MAX_ORDER.
2367 * Memory allocated by this function must be released by free_pages_exact().
2369 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
2371 unsigned int order
= get_order(size
);
2374 addr
= __get_free_pages(gfp_mask
, order
);
2375 return make_alloc_exact(addr
, order
, size
);
2377 EXPORT_SYMBOL(alloc_pages_exact
);
2380 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2382 * @nid: the preferred node ID where memory should be allocated
2383 * @size: the number of bytes to allocate
2384 * @gfp_mask: GFP flags for the allocation
2386 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2388 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2391 void *alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
2393 unsigned order
= get_order(size
);
2394 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
2397 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
2399 EXPORT_SYMBOL(alloc_pages_exact_nid
);
2402 * free_pages_exact - release memory allocated via alloc_pages_exact()
2403 * @virt: the value returned by alloc_pages_exact.
2404 * @size: size of allocation, same value as passed to alloc_pages_exact().
2406 * Release the memory allocated by a previous call to alloc_pages_exact.
2408 void free_pages_exact(void *virt
, size_t size
)
2410 unsigned long addr
= (unsigned long)virt
;
2411 unsigned long end
= addr
+ PAGE_ALIGN(size
);
2413 while (addr
< end
) {
2418 EXPORT_SYMBOL(free_pages_exact
);
2420 static unsigned int nr_free_zone_pages(int offset
)
2425 /* Just pick one node, since fallback list is circular */
2426 unsigned int sum
= 0;
2428 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
2430 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
2431 unsigned long size
= zone
->present_pages
;
2432 unsigned long high
= high_wmark_pages(zone
);
2441 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
2443 unsigned int nr_free_buffer_pages(void)
2445 return nr_free_zone_pages(gfp_zone(GFP_USER
));
2447 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
2450 * Amount of free RAM allocatable within all zones
2452 unsigned int nr_free_pagecache_pages(void)
2454 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
2457 static inline void show_node(struct zone
*zone
)
2460 printk("Node %d ", zone_to_nid(zone
));
2463 void si_meminfo(struct sysinfo
*val
)
2465 val
->totalram
= totalram_pages
;
2467 val
->freeram
= global_page_state(NR_FREE_PAGES
);
2468 val
->bufferram
= nr_blockdev_pages();
2469 val
->totalhigh
= totalhigh_pages
;
2470 val
->freehigh
= nr_free_highpages();
2471 val
->mem_unit
= PAGE_SIZE
;
2474 EXPORT_SYMBOL(si_meminfo
);
2477 void si_meminfo_node(struct sysinfo
*val
, int nid
)
2479 pg_data_t
*pgdat
= NODE_DATA(nid
);
2481 val
->totalram
= pgdat
->node_present_pages
;
2482 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
2483 #ifdef CONFIG_HIGHMEM
2484 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
2485 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
2491 val
->mem_unit
= PAGE_SIZE
;
2496 * Determine whether the node should be displayed or not, depending on whether
2497 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
2499 bool skip_free_areas_node(unsigned int flags
, int nid
)
2503 if (!(flags
& SHOW_MEM_FILTER_NODES
))
2507 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
2513 #define K(x) ((x) << (PAGE_SHIFT-10))
2516 * Show free area list (used inside shift_scroll-lock stuff)
2517 * We also calculate the percentage fragmentation. We do this by counting the
2518 * memory on each free list with the exception of the first item on the list.
2519 * Suppresses nodes that are not allowed by current's cpuset if
2520 * SHOW_MEM_FILTER_NODES is passed.
2522 void show_free_areas(unsigned int filter
)
2527 for_each_populated_zone(zone
) {
2528 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2531 printk("%s per-cpu:\n", zone
->name
);
2533 for_each_online_cpu(cpu
) {
2534 struct per_cpu_pageset
*pageset
;
2536 pageset
= per_cpu_ptr(zone
->pageset
, cpu
);
2538 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2539 cpu
, pageset
->pcp
.high
,
2540 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2544 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
2545 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
2547 " dirty:%lu writeback:%lu unstable:%lu\n"
2548 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
2549 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n",
2550 global_page_state(NR_ACTIVE_ANON
),
2551 global_page_state(NR_INACTIVE_ANON
),
2552 global_page_state(NR_ISOLATED_ANON
),
2553 global_page_state(NR_ACTIVE_FILE
),
2554 global_page_state(NR_INACTIVE_FILE
),
2555 global_page_state(NR_ISOLATED_FILE
),
2556 global_page_state(NR_UNEVICTABLE
),
2557 global_page_state(NR_FILE_DIRTY
),
2558 global_page_state(NR_WRITEBACK
),
2559 global_page_state(NR_UNSTABLE_NFS
),
2560 global_page_state(NR_FREE_PAGES
),
2561 global_page_state(NR_SLAB_RECLAIMABLE
),
2562 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2563 global_page_state(NR_FILE_MAPPED
),
2564 global_page_state(NR_SHMEM
),
2565 global_page_state(NR_PAGETABLE
),
2566 global_page_state(NR_BOUNCE
));
2568 for_each_populated_zone(zone
) {
2571 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2579 " active_anon:%lukB"
2580 " inactive_anon:%lukB"
2581 " active_file:%lukB"
2582 " inactive_file:%lukB"
2583 " unevictable:%lukB"
2584 " isolated(anon):%lukB"
2585 " isolated(file):%lukB"
2592 " slab_reclaimable:%lukB"
2593 " slab_unreclaimable:%lukB"
2594 " kernel_stack:%lukB"
2598 " writeback_tmp:%lukB"
2599 " pages_scanned:%lu"
2600 " all_unreclaimable? %s"
2603 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2604 K(min_wmark_pages(zone
)),
2605 K(low_wmark_pages(zone
)),
2606 K(high_wmark_pages(zone
)),
2607 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2608 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2609 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2610 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2611 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2612 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
2613 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
2614 K(zone
->present_pages
),
2615 K(zone_page_state(zone
, NR_MLOCK
)),
2616 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
2617 K(zone_page_state(zone
, NR_WRITEBACK
)),
2618 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
2619 K(zone_page_state(zone
, NR_SHMEM
)),
2620 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
2621 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
2622 zone_page_state(zone
, NR_KERNEL_STACK
) *
2624 K(zone_page_state(zone
, NR_PAGETABLE
)),
2625 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
2626 K(zone_page_state(zone
, NR_BOUNCE
)),
2627 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
2628 zone
->pages_scanned
,
2629 (zone
->all_unreclaimable
? "yes" : "no")
2631 printk("lowmem_reserve[]:");
2632 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2633 printk(" %lu", zone
->lowmem_reserve
[i
]);
2637 for_each_populated_zone(zone
) {
2638 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2640 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
2643 printk("%s: ", zone
->name
);
2645 spin_lock_irqsave(&zone
->lock
, flags
);
2646 for (order
= 0; order
< MAX_ORDER
; order
++) {
2647 nr
[order
] = zone
->free_area
[order
].nr_free
;
2648 total
+= nr
[order
] << order
;
2650 spin_unlock_irqrestore(&zone
->lock
, flags
);
2651 for (order
= 0; order
< MAX_ORDER
; order
++)
2652 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2653 printk("= %lukB\n", K(total
));
2656 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2658 show_swap_cache_info();
2661 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2663 zoneref
->zone
= zone
;
2664 zoneref
->zone_idx
= zone_idx(zone
);
2668 * Builds allocation fallback zone lists.
2670 * Add all populated zones of a node to the zonelist.
2672 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2673 int nr_zones
, enum zone_type zone_type
)
2677 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2682 zone
= pgdat
->node_zones
+ zone_type
;
2683 if (populated_zone(zone
)) {
2684 zoneref_set_zone(zone
,
2685 &zonelist
->_zonerefs
[nr_zones
++]);
2686 check_highest_zone(zone_type
);
2689 } while (zone_type
);
2696 * 0 = automatic detection of better ordering.
2697 * 1 = order by ([node] distance, -zonetype)
2698 * 2 = order by (-zonetype, [node] distance)
2700 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2701 * the same zonelist. So only NUMA can configure this param.
2703 #define ZONELIST_ORDER_DEFAULT 0
2704 #define ZONELIST_ORDER_NODE 1
2705 #define ZONELIST_ORDER_ZONE 2
2707 /* zonelist order in the kernel.
2708 * set_zonelist_order() will set this to NODE or ZONE.
2710 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2711 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2715 /* The value user specified ....changed by config */
2716 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2717 /* string for sysctl */
2718 #define NUMA_ZONELIST_ORDER_LEN 16
2719 char numa_zonelist_order
[16] = "default";
2722 * interface for configure zonelist ordering.
2723 * command line option "numa_zonelist_order"
2724 * = "[dD]efault - default, automatic configuration.
2725 * = "[nN]ode - order by node locality, then by zone within node
2726 * = "[zZ]one - order by zone, then by locality within zone
2729 static int __parse_numa_zonelist_order(char *s
)
2731 if (*s
== 'd' || *s
== 'D') {
2732 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2733 } else if (*s
== 'n' || *s
== 'N') {
2734 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2735 } else if (*s
== 'z' || *s
== 'Z') {
2736 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2739 "Ignoring invalid numa_zonelist_order value: "
2746 static __init
int setup_numa_zonelist_order(char *s
)
2753 ret
= __parse_numa_zonelist_order(s
);
2755 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
2759 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2762 * sysctl handler for numa_zonelist_order
2764 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2765 void __user
*buffer
, size_t *length
,
2768 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2770 static DEFINE_MUTEX(zl_order_mutex
);
2772 mutex_lock(&zl_order_mutex
);
2774 strcpy(saved_string
, (char*)table
->data
);
2775 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
2779 int oldval
= user_zonelist_order
;
2780 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2782 * bogus value. restore saved string
2784 strncpy((char*)table
->data
, saved_string
,
2785 NUMA_ZONELIST_ORDER_LEN
);
2786 user_zonelist_order
= oldval
;
2787 } else if (oldval
!= user_zonelist_order
) {
2788 mutex_lock(&zonelists_mutex
);
2789 build_all_zonelists(NULL
);
2790 mutex_unlock(&zonelists_mutex
);
2794 mutex_unlock(&zl_order_mutex
);
2799 #define MAX_NODE_LOAD (nr_online_nodes)
2800 static int node_load
[MAX_NUMNODES
];
2803 * find_next_best_node - find the next node that should appear in a given node's fallback list
2804 * @node: node whose fallback list we're appending
2805 * @used_node_mask: nodemask_t of already used nodes
2807 * We use a number of factors to determine which is the next node that should
2808 * appear on a given node's fallback list. The node should not have appeared
2809 * already in @node's fallback list, and it should be the next closest node
2810 * according to the distance array (which contains arbitrary distance values
2811 * from each node to each node in the system), and should also prefer nodes
2812 * with no CPUs, since presumably they'll have very little allocation pressure
2813 * on them otherwise.
2814 * It returns -1 if no node is found.
2816 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2819 int min_val
= INT_MAX
;
2821 const struct cpumask
*tmp
= cpumask_of_node(0);
2823 /* Use the local node if we haven't already */
2824 if (!node_isset(node
, *used_node_mask
)) {
2825 node_set(node
, *used_node_mask
);
2829 for_each_node_state(n
, N_HIGH_MEMORY
) {
2831 /* Don't want a node to appear more than once */
2832 if (node_isset(n
, *used_node_mask
))
2835 /* Use the distance array to find the distance */
2836 val
= node_distance(node
, n
);
2838 /* Penalize nodes under us ("prefer the next node") */
2841 /* Give preference to headless and unused nodes */
2842 tmp
= cpumask_of_node(n
);
2843 if (!cpumask_empty(tmp
))
2844 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2846 /* Slight preference for less loaded node */
2847 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2848 val
+= node_load
[n
];
2850 if (val
< min_val
) {
2857 node_set(best_node
, *used_node_mask
);
2864 * Build zonelists ordered by node and zones within node.
2865 * This results in maximum locality--normal zone overflows into local
2866 * DMA zone, if any--but risks exhausting DMA zone.
2868 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2871 struct zonelist
*zonelist
;
2873 zonelist
= &pgdat
->node_zonelists
[0];
2874 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2876 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2878 zonelist
->_zonerefs
[j
].zone
= NULL
;
2879 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2883 * Build gfp_thisnode zonelists
2885 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2888 struct zonelist
*zonelist
;
2890 zonelist
= &pgdat
->node_zonelists
[1];
2891 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2892 zonelist
->_zonerefs
[j
].zone
= NULL
;
2893 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2897 * Build zonelists ordered by zone and nodes within zones.
2898 * This results in conserving DMA zone[s] until all Normal memory is
2899 * exhausted, but results in overflowing to remote node while memory
2900 * may still exist in local DMA zone.
2902 static int node_order
[MAX_NUMNODES
];
2904 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2907 int zone_type
; /* needs to be signed */
2909 struct zonelist
*zonelist
;
2911 zonelist
= &pgdat
->node_zonelists
[0];
2913 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2914 for (j
= 0; j
< nr_nodes
; j
++) {
2915 node
= node_order
[j
];
2916 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2917 if (populated_zone(z
)) {
2919 &zonelist
->_zonerefs
[pos
++]);
2920 check_highest_zone(zone_type
);
2924 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2925 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2928 static int default_zonelist_order(void)
2931 unsigned long low_kmem_size
,total_size
;
2935 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
2936 * If they are really small and used heavily, the system can fall
2937 * into OOM very easily.
2938 * This function detect ZONE_DMA/DMA32 size and configures zone order.
2940 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2943 for_each_online_node(nid
) {
2944 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2945 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2946 if (populated_zone(z
)) {
2947 if (zone_type
< ZONE_NORMAL
)
2948 low_kmem_size
+= z
->present_pages
;
2949 total_size
+= z
->present_pages
;
2950 } else if (zone_type
== ZONE_NORMAL
) {
2952 * If any node has only lowmem, then node order
2953 * is preferred to allow kernel allocations
2954 * locally; otherwise, they can easily infringe
2955 * on other nodes when there is an abundance of
2956 * lowmem available to allocate from.
2958 return ZONELIST_ORDER_NODE
;
2962 if (!low_kmem_size
|| /* there are no DMA area. */
2963 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2964 return ZONELIST_ORDER_NODE
;
2966 * look into each node's config.
2967 * If there is a node whose DMA/DMA32 memory is very big area on
2968 * local memory, NODE_ORDER may be suitable.
2970 average_size
= total_size
/
2971 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2972 for_each_online_node(nid
) {
2975 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2976 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2977 if (populated_zone(z
)) {
2978 if (zone_type
< ZONE_NORMAL
)
2979 low_kmem_size
+= z
->present_pages
;
2980 total_size
+= z
->present_pages
;
2983 if (low_kmem_size
&&
2984 total_size
> average_size
&& /* ignore small node */
2985 low_kmem_size
> total_size
* 70/100)
2986 return ZONELIST_ORDER_NODE
;
2988 return ZONELIST_ORDER_ZONE
;
2991 static void set_zonelist_order(void)
2993 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2994 current_zonelist_order
= default_zonelist_order();
2996 current_zonelist_order
= user_zonelist_order
;
2999 static void build_zonelists(pg_data_t
*pgdat
)
3003 nodemask_t used_mask
;
3004 int local_node
, prev_node
;
3005 struct zonelist
*zonelist
;
3006 int order
= current_zonelist_order
;
3008 /* initialize zonelists */
3009 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
3010 zonelist
= pgdat
->node_zonelists
+ i
;
3011 zonelist
->_zonerefs
[0].zone
= NULL
;
3012 zonelist
->_zonerefs
[0].zone_idx
= 0;
3015 /* NUMA-aware ordering of nodes */
3016 local_node
= pgdat
->node_id
;
3017 load
= nr_online_nodes
;
3018 prev_node
= local_node
;
3019 nodes_clear(used_mask
);
3021 memset(node_order
, 0, sizeof(node_order
));
3024 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
3025 int distance
= node_distance(local_node
, node
);
3028 * If another node is sufficiently far away then it is better
3029 * to reclaim pages in a zone before going off node.
3031 if (distance
> RECLAIM_DISTANCE
)
3032 zone_reclaim_mode
= 1;
3035 * We don't want to pressure a particular node.
3036 * So adding penalty to the first node in same
3037 * distance group to make it round-robin.
3039 if (distance
!= node_distance(local_node
, prev_node
))
3040 node_load
[node
] = load
;
3044 if (order
== ZONELIST_ORDER_NODE
)
3045 build_zonelists_in_node_order(pgdat
, node
);
3047 node_order
[j
++] = node
; /* remember order */
3050 if (order
== ZONELIST_ORDER_ZONE
) {
3051 /* calculate node order -- i.e., DMA last! */
3052 build_zonelists_in_zone_order(pgdat
, j
);
3055 build_thisnode_zonelists(pgdat
);
3058 /* Construct the zonelist performance cache - see further mmzone.h */
3059 static void build_zonelist_cache(pg_data_t
*pgdat
)
3061 struct zonelist
*zonelist
;
3062 struct zonelist_cache
*zlc
;
3065 zonelist
= &pgdat
->node_zonelists
[0];
3066 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
3067 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
3068 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
3069 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
3072 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3074 * Return node id of node used for "local" allocations.
3075 * I.e., first node id of first zone in arg node's generic zonelist.
3076 * Used for initializing percpu 'numa_mem', which is used primarily
3077 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3079 int local_memory_node(int node
)
3083 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
3084 gfp_zone(GFP_KERNEL
),
3091 #else /* CONFIG_NUMA */
3093 static void set_zonelist_order(void)
3095 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
3098 static void build_zonelists(pg_data_t
*pgdat
)
3100 int node
, local_node
;
3102 struct zonelist
*zonelist
;
3104 local_node
= pgdat
->node_id
;
3106 zonelist
= &pgdat
->node_zonelists
[0];
3107 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
3110 * Now we build the zonelist so that it contains the zones
3111 * of all the other nodes.
3112 * We don't want to pressure a particular node, so when
3113 * building the zones for node N, we make sure that the
3114 * zones coming right after the local ones are those from
3115 * node N+1 (modulo N)
3117 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
3118 if (!node_online(node
))
3120 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3123 for (node
= 0; node
< local_node
; node
++) {
3124 if (!node_online(node
))
3126 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
3130 zonelist
->_zonerefs
[j
].zone
= NULL
;
3131 zonelist
->_zonerefs
[j
].zone_idx
= 0;
3134 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3135 static void build_zonelist_cache(pg_data_t
*pgdat
)
3137 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
3140 #endif /* CONFIG_NUMA */
3143 * Boot pageset table. One per cpu which is going to be used for all
3144 * zones and all nodes. The parameters will be set in such a way
3145 * that an item put on a list will immediately be handed over to
3146 * the buddy list. This is safe since pageset manipulation is done
3147 * with interrupts disabled.
3149 * The boot_pagesets must be kept even after bootup is complete for
3150 * unused processors and/or zones. They do play a role for bootstrapping
3151 * hotplugged processors.
3153 * zoneinfo_show() and maybe other functions do
3154 * not check if the processor is online before following the pageset pointer.
3155 * Other parts of the kernel may not check if the zone is available.
3157 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
3158 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
3159 static void setup_zone_pageset(struct zone
*zone
);
3162 * Global mutex to protect against size modification of zonelists
3163 * as well as to serialize pageset setup for the new populated zone.
3165 DEFINE_MUTEX(zonelists_mutex
);
3167 /* return values int ....just for stop_machine() */
3168 static __init_refok
int __build_all_zonelists(void *data
)
3174 memset(node_load
, 0, sizeof(node_load
));
3176 for_each_online_node(nid
) {
3177 pg_data_t
*pgdat
= NODE_DATA(nid
);
3179 build_zonelists(pgdat
);
3180 build_zonelist_cache(pgdat
);
3184 * Initialize the boot_pagesets that are going to be used
3185 * for bootstrapping processors. The real pagesets for
3186 * each zone will be allocated later when the per cpu
3187 * allocator is available.
3189 * boot_pagesets are used also for bootstrapping offline
3190 * cpus if the system is already booted because the pagesets
3191 * are needed to initialize allocators on a specific cpu too.
3192 * F.e. the percpu allocator needs the page allocator which
3193 * needs the percpu allocator in order to allocate its pagesets
3194 * (a chicken-egg dilemma).
3196 for_each_possible_cpu(cpu
) {
3197 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
3199 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3201 * We now know the "local memory node" for each node--
3202 * i.e., the node of the first zone in the generic zonelist.
3203 * Set up numa_mem percpu variable for on-line cpus. During
3204 * boot, only the boot cpu should be on-line; we'll init the
3205 * secondary cpus' numa_mem as they come on-line. During
3206 * node/memory hotplug, we'll fixup all on-line cpus.
3208 if (cpu_online(cpu
))
3209 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
3217 * Called with zonelists_mutex held always
3218 * unless system_state == SYSTEM_BOOTING.
3220 void __ref
build_all_zonelists(void *data
)
3222 set_zonelist_order();
3224 if (system_state
== SYSTEM_BOOTING
) {
3225 __build_all_zonelists(NULL
);
3226 mminit_verify_zonelist();
3227 cpuset_init_current_mems_allowed();
3229 /* we have to stop all cpus to guarantee there is no user
3231 #ifdef CONFIG_MEMORY_HOTPLUG
3233 setup_zone_pageset((struct zone
*)data
);
3235 stop_machine(__build_all_zonelists
, NULL
, NULL
);
3236 /* cpuset refresh routine should be here */
3238 vm_total_pages
= nr_free_pagecache_pages();
3240 * Disable grouping by mobility if the number of pages in the
3241 * system is too low to allow the mechanism to work. It would be
3242 * more accurate, but expensive to check per-zone. This check is
3243 * made on memory-hotadd so a system can start with mobility
3244 * disabled and enable it later
3246 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
3247 page_group_by_mobility_disabled
= 1;
3249 page_group_by_mobility_disabled
= 0;
3251 printk("Built %i zonelists in %s order, mobility grouping %s. "
3252 "Total pages: %ld\n",
3254 zonelist_order_name
[current_zonelist_order
],
3255 page_group_by_mobility_disabled
? "off" : "on",
3258 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
3263 * Helper functions to size the waitqueue hash table.
3264 * Essentially these want to choose hash table sizes sufficiently
3265 * large so that collisions trying to wait on pages are rare.
3266 * But in fact, the number of active page waitqueues on typical
3267 * systems is ridiculously low, less than 200. So this is even
3268 * conservative, even though it seems large.
3270 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3271 * waitqueues, i.e. the size of the waitq table given the number of pages.
3273 #define PAGES_PER_WAITQUEUE 256
3275 #ifndef CONFIG_MEMORY_HOTPLUG
3276 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3278 unsigned long size
= 1;
3280 pages
/= PAGES_PER_WAITQUEUE
;
3282 while (size
< pages
)
3286 * Once we have dozens or even hundreds of threads sleeping
3287 * on IO we've got bigger problems than wait queue collision.
3288 * Limit the size of the wait table to a reasonable size.
3290 size
= min(size
, 4096UL);
3292 return max(size
, 4UL);
3296 * A zone's size might be changed by hot-add, so it is not possible to determine
3297 * a suitable size for its wait_table. So we use the maximum size now.
3299 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
3301 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
3302 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3303 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
3305 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3306 * or more by the traditional way. (See above). It equals:
3308 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
3309 * ia64(16K page size) : = ( 8G + 4M)byte.
3310 * powerpc (64K page size) : = (32G +16M)byte.
3312 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
3319 * This is an integer logarithm so that shifts can be used later
3320 * to extract the more random high bits from the multiplicative
3321 * hash function before the remainder is taken.
3323 static inline unsigned long wait_table_bits(unsigned long size
)
3328 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
3331 * Check if a pageblock contains reserved pages
3333 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
3337 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3338 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
3345 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3346 * of blocks reserved is based on min_wmark_pages(zone). The memory within
3347 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3348 * higher will lead to a bigger reserve which will get freed as contiguous
3349 * blocks as reclaim kicks in
3351 static void setup_zone_migrate_reserve(struct zone
*zone
)
3353 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
3355 unsigned long block_migratetype
;
3359 * Get the start pfn, end pfn and the number of blocks to reserve
3360 * We have to be careful to be aligned to pageblock_nr_pages to
3361 * make sure that we always check pfn_valid for the first page in
3364 start_pfn
= zone
->zone_start_pfn
;
3365 end_pfn
= start_pfn
+ zone
->spanned_pages
;
3366 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
3367 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
3371 * Reserve blocks are generally in place to help high-order atomic
3372 * allocations that are short-lived. A min_free_kbytes value that
3373 * would result in more than 2 reserve blocks for atomic allocations
3374 * is assumed to be in place to help anti-fragmentation for the
3375 * future allocation of hugepages at runtime.
3377 reserve
= min(2, reserve
);
3379 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
3380 if (!pfn_valid(pfn
))
3382 page
= pfn_to_page(pfn
);
3384 /* Watch out for overlapping nodes */
3385 if (page_to_nid(page
) != zone_to_nid(zone
))
3388 /* Blocks with reserved pages will never free, skip them. */
3389 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
3390 if (pageblock_is_reserved(pfn
, block_end_pfn
))
3393 block_migratetype
= get_pageblock_migratetype(page
);
3395 /* If this block is reserved, account for it */
3396 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
3401 /* Suitable for reserving if this block is movable */
3402 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
3403 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
3404 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
3410 * If the reserve is met and this is a previous reserved block,
3413 if (block_migratetype
== MIGRATE_RESERVE
) {
3414 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3415 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
3421 * Initially all pages are reserved - free ones are freed
3422 * up by free_all_bootmem() once the early boot process is
3423 * done. Non-atomic initialization, single-pass.
3425 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
3426 unsigned long start_pfn
, enum memmap_context context
)
3429 unsigned long end_pfn
= start_pfn
+ size
;
3433 if (highest_memmap_pfn
< end_pfn
- 1)
3434 highest_memmap_pfn
= end_pfn
- 1;
3436 z
= &NODE_DATA(nid
)->node_zones
[zone
];
3437 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
3439 * There can be holes in boot-time mem_map[]s
3440 * handed to this function. They do not
3441 * exist on hotplugged memory.
3443 if (context
== MEMMAP_EARLY
) {
3444 if (!early_pfn_valid(pfn
))
3446 if (!early_pfn_in_nid(pfn
, nid
))
3449 page
= pfn_to_page(pfn
);
3450 set_page_links(page
, zone
, nid
, pfn
);
3451 mminit_verify_page_links(page
, zone
, nid
, pfn
);
3452 init_page_count(page
);
3453 reset_page_mapcount(page
);
3454 SetPageReserved(page
);
3456 * Mark the block movable so that blocks are reserved for
3457 * movable at startup. This will force kernel allocations
3458 * to reserve their blocks rather than leaking throughout
3459 * the address space during boot when many long-lived
3460 * kernel allocations are made. Later some blocks near
3461 * the start are marked MIGRATE_RESERVE by
3462 * setup_zone_migrate_reserve()
3464 * bitmap is created for zone's valid pfn range. but memmap
3465 * can be created for invalid pages (for alignment)
3466 * check here not to call set_pageblock_migratetype() against
3469 if ((z
->zone_start_pfn
<= pfn
)
3470 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
3471 && !(pfn
& (pageblock_nr_pages
- 1)))
3472 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
3474 INIT_LIST_HEAD(&page
->lru
);
3475 #ifdef WANT_PAGE_VIRTUAL
3476 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
3477 if (!is_highmem_idx(zone
))
3478 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
3483 static void __meminit
zone_init_free_lists(struct zone
*zone
)
3486 for_each_migratetype_order(order
, t
) {
3487 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
3488 zone
->free_area
[order
].nr_free
= 0;
3492 #ifndef __HAVE_ARCH_MEMMAP_INIT
3493 #define memmap_init(size, nid, zone, start_pfn) \
3494 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
3497 static int zone_batchsize(struct zone
*zone
)
3503 * The per-cpu-pages pools are set to around 1000th of the
3504 * size of the zone. But no more than 1/2 of a meg.
3506 * OK, so we don't know how big the cache is. So guess.
3508 batch
= zone
->present_pages
/ 1024;
3509 if (batch
* PAGE_SIZE
> 512 * 1024)
3510 batch
= (512 * 1024) / PAGE_SIZE
;
3511 batch
/= 4; /* We effectively *= 4 below */
3516 * Clamp the batch to a 2^n - 1 value. Having a power
3517 * of 2 value was found to be more likely to have
3518 * suboptimal cache aliasing properties in some cases.
3520 * For example if 2 tasks are alternately allocating
3521 * batches of pages, one task can end up with a lot
3522 * of pages of one half of the possible page colors
3523 * and the other with pages of the other colors.
3525 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
3530 /* The deferral and batching of frees should be suppressed under NOMMU
3533 * The problem is that NOMMU needs to be able to allocate large chunks
3534 * of contiguous memory as there's no hardware page translation to
3535 * assemble apparent contiguous memory from discontiguous pages.
3537 * Queueing large contiguous runs of pages for batching, however,
3538 * causes the pages to actually be freed in smaller chunks. As there
3539 * can be a significant delay between the individual batches being
3540 * recycled, this leads to the once large chunks of space being
3541 * fragmented and becoming unavailable for high-order allocations.
3547 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
3549 struct per_cpu_pages
*pcp
;
3552 memset(p
, 0, sizeof(*p
));
3556 pcp
->high
= 6 * batch
;
3557 pcp
->batch
= max(1UL, 1 * batch
);
3558 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
3559 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
3563 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
3564 * to the value high for the pageset p.
3567 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
3570 struct per_cpu_pages
*pcp
;
3574 pcp
->batch
= max(1UL, high
/4);
3575 if ((high
/4) > (PAGE_SHIFT
* 8))
3576 pcp
->batch
= PAGE_SHIFT
* 8;
3579 static void setup_zone_pageset(struct zone
*zone
)
3583 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
3585 for_each_possible_cpu(cpu
) {
3586 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
3588 setup_pageset(pcp
, zone_batchsize(zone
));
3590 if (percpu_pagelist_fraction
)
3591 setup_pagelist_highmark(pcp
,
3592 (zone
->present_pages
/
3593 percpu_pagelist_fraction
));
3598 * Allocate per cpu pagesets and initialize them.
3599 * Before this call only boot pagesets were available.
3601 void __init
setup_per_cpu_pageset(void)
3605 for_each_populated_zone(zone
)
3606 setup_zone_pageset(zone
);
3609 static noinline __init_refok
3610 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3613 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3617 * The per-page waitqueue mechanism uses hashed waitqueues
3620 zone
->wait_table_hash_nr_entries
=
3621 wait_table_hash_nr_entries(zone_size_pages
);
3622 zone
->wait_table_bits
=
3623 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3624 alloc_size
= zone
->wait_table_hash_nr_entries
3625 * sizeof(wait_queue_head_t
);
3627 if (!slab_is_available()) {
3628 zone
->wait_table
= (wait_queue_head_t
*)
3629 alloc_bootmem_node_nopanic(pgdat
, alloc_size
);
3632 * This case means that a zone whose size was 0 gets new memory
3633 * via memory hot-add.
3634 * But it may be the case that a new node was hot-added. In
3635 * this case vmalloc() will not be able to use this new node's
3636 * memory - this wait_table must be initialized to use this new
3637 * node itself as well.
3638 * To use this new node's memory, further consideration will be
3641 zone
->wait_table
= vmalloc(alloc_size
);
3643 if (!zone
->wait_table
)
3646 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3647 init_waitqueue_head(zone
->wait_table
+ i
);
3652 static int __zone_pcp_update(void *data
)
3654 struct zone
*zone
= data
;
3656 unsigned long batch
= zone_batchsize(zone
), flags
;
3658 for_each_possible_cpu(cpu
) {
3659 struct per_cpu_pageset
*pset
;
3660 struct per_cpu_pages
*pcp
;
3662 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
3665 local_irq_save(flags
);
3666 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
3667 setup_pageset(pset
, batch
);
3668 local_irq_restore(flags
);
3673 void zone_pcp_update(struct zone
*zone
)
3675 stop_machine(__zone_pcp_update
, zone
, NULL
);
3678 static __meminit
void zone_pcp_init(struct zone
*zone
)
3681 * per cpu subsystem is not up at this point. The following code
3682 * relies on the ability of the linker to provide the
3683 * offset of a (static) per cpu variable into the per cpu area.
3685 zone
->pageset
= &boot_pageset
;
3687 if (zone
->present_pages
)
3688 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
3689 zone
->name
, zone
->present_pages
,
3690 zone_batchsize(zone
));
3693 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3694 unsigned long zone_start_pfn
,
3696 enum memmap_context context
)
3698 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3700 ret
= zone_wait_table_init(zone
, size
);
3703 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3705 zone
->zone_start_pfn
= zone_start_pfn
;
3707 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3708 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3710 (unsigned long)zone_idx(zone
),
3711 zone_start_pfn
, (zone_start_pfn
+ size
));
3713 zone_init_free_lists(zone
);
3718 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
3719 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3721 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3722 * Architectures may implement their own version but if add_active_range()
3723 * was used and there are no special requirements, this is a convenient
3726 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3728 unsigned long start_pfn
, end_pfn
;
3731 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
3732 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3734 /* This is a memory hole */
3737 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3739 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3743 nid
= __early_pfn_to_nid(pfn
);
3746 /* just returns 0 */
3750 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3751 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3755 nid
= __early_pfn_to_nid(pfn
);
3756 if (nid
>= 0 && nid
!= node
)
3763 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3764 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3765 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3767 * If an architecture guarantees that all ranges registered with
3768 * add_active_ranges() contain no holes and may be freed, this
3769 * this function may be used instead of calling free_bootmem() manually.
3771 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
3773 unsigned long start_pfn
, end_pfn
;
3776 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
3777 start_pfn
= min(start_pfn
, max_low_pfn
);
3778 end_pfn
= min(end_pfn
, max_low_pfn
);
3780 if (start_pfn
< end_pfn
)
3781 free_bootmem_node(NODE_DATA(this_nid
),
3782 PFN_PHYS(start_pfn
),
3783 (end_pfn
- start_pfn
) << PAGE_SHIFT
);
3787 int __init
add_from_early_node_map(struct range
*range
, int az
,
3788 int nr_range
, int nid
)
3790 unsigned long start_pfn
, end_pfn
;
3793 /* need to go over early_node_map to find out good range for node */
3794 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
)
3795 nr_range
= add_range(range
, az
, nr_range
, start_pfn
, end_pfn
);
3800 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3801 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3803 * If an architecture guarantees that all ranges registered with
3804 * add_active_ranges() contain no holes and may be freed, this
3805 * function may be used instead of calling memory_present() manually.
3807 void __init
sparse_memory_present_with_active_regions(int nid
)
3809 unsigned long start_pfn
, end_pfn
;
3812 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
3813 memory_present(this_nid
, start_pfn
, end_pfn
);
3817 * get_pfn_range_for_nid - Return the start and end page frames for a node
3818 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3819 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3820 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3822 * It returns the start and end page frame of a node based on information
3823 * provided by an arch calling add_active_range(). If called for a node
3824 * with no available memory, a warning is printed and the start and end
3827 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3828 unsigned long *start_pfn
, unsigned long *end_pfn
)
3830 unsigned long this_start_pfn
, this_end_pfn
;
3836 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
3837 *start_pfn
= min(*start_pfn
, this_start_pfn
);
3838 *end_pfn
= max(*end_pfn
, this_end_pfn
);
3841 if (*start_pfn
== -1UL)
3846 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3847 * assumption is made that zones within a node are ordered in monotonic
3848 * increasing memory addresses so that the "highest" populated zone is used
3850 static void __init
find_usable_zone_for_movable(void)
3853 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3854 if (zone_index
== ZONE_MOVABLE
)
3857 if (arch_zone_highest_possible_pfn
[zone_index
] >
3858 arch_zone_lowest_possible_pfn
[zone_index
])
3862 VM_BUG_ON(zone_index
== -1);
3863 movable_zone
= zone_index
;
3867 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3868 * because it is sized independent of architecture. Unlike the other zones,
3869 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3870 * in each node depending on the size of each node and how evenly kernelcore
3871 * is distributed. This helper function adjusts the zone ranges
3872 * provided by the architecture for a given node by using the end of the
3873 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3874 * zones within a node are in order of monotonic increases memory addresses
3876 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3877 unsigned long zone_type
,
3878 unsigned long node_start_pfn
,
3879 unsigned long node_end_pfn
,
3880 unsigned long *zone_start_pfn
,
3881 unsigned long *zone_end_pfn
)
3883 /* Only adjust if ZONE_MOVABLE is on this node */
3884 if (zone_movable_pfn
[nid
]) {
3885 /* Size ZONE_MOVABLE */
3886 if (zone_type
== ZONE_MOVABLE
) {
3887 *zone_start_pfn
= zone_movable_pfn
[nid
];
3888 *zone_end_pfn
= min(node_end_pfn
,
3889 arch_zone_highest_possible_pfn
[movable_zone
]);
3891 /* Adjust for ZONE_MOVABLE starting within this range */
3892 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3893 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3894 *zone_end_pfn
= zone_movable_pfn
[nid
];
3896 /* Check if this whole range is within ZONE_MOVABLE */
3897 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3898 *zone_start_pfn
= *zone_end_pfn
;
3903 * Return the number of pages a zone spans in a node, including holes
3904 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3906 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3907 unsigned long zone_type
,
3908 unsigned long *ignored
)
3910 unsigned long node_start_pfn
, node_end_pfn
;
3911 unsigned long zone_start_pfn
, zone_end_pfn
;
3913 /* Get the start and end of the node and zone */
3914 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3915 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3916 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3917 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3918 node_start_pfn
, node_end_pfn
,
3919 &zone_start_pfn
, &zone_end_pfn
);
3921 /* Check that this node has pages within the zone's required range */
3922 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3925 /* Move the zone boundaries inside the node if necessary */
3926 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3927 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3929 /* Return the spanned pages */
3930 return zone_end_pfn
- zone_start_pfn
;
3934 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3935 * then all holes in the requested range will be accounted for.
3937 unsigned long __meminit
__absent_pages_in_range(int nid
,
3938 unsigned long range_start_pfn
,
3939 unsigned long range_end_pfn
)
3941 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
3942 unsigned long start_pfn
, end_pfn
;
3945 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
3946 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
3947 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
3948 nr_absent
-= end_pfn
- start_pfn
;
3954 * absent_pages_in_range - Return number of page frames in holes within a range
3955 * @start_pfn: The start PFN to start searching for holes
3956 * @end_pfn: The end PFN to stop searching for holes
3958 * It returns the number of pages frames in memory holes within a range.
3960 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3961 unsigned long end_pfn
)
3963 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3966 /* Return the number of page frames in holes in a zone on a node */
3967 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3968 unsigned long zone_type
,
3969 unsigned long *ignored
)
3971 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
3972 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
3973 unsigned long node_start_pfn
, node_end_pfn
;
3974 unsigned long zone_start_pfn
, zone_end_pfn
;
3976 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3977 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
3978 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
3980 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3981 node_start_pfn
, node_end_pfn
,
3982 &zone_start_pfn
, &zone_end_pfn
);
3983 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3986 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
3987 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3988 unsigned long zone_type
,
3989 unsigned long *zones_size
)
3991 return zones_size
[zone_type
];
3994 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3995 unsigned long zone_type
,
3996 unsigned long *zholes_size
)
4001 return zholes_size
[zone_type
];
4004 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4006 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
4007 unsigned long *zones_size
, unsigned long *zholes_size
)
4009 unsigned long realtotalpages
, totalpages
= 0;
4012 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4013 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
4015 pgdat
->node_spanned_pages
= totalpages
;
4017 realtotalpages
= totalpages
;
4018 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4020 zone_absent_pages_in_node(pgdat
->node_id
, i
,
4022 pgdat
->node_present_pages
= realtotalpages
;
4023 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
4027 #ifndef CONFIG_SPARSEMEM
4029 * Calculate the size of the zone->blockflags rounded to an unsigned long
4030 * Start by making sure zonesize is a multiple of pageblock_order by rounding
4031 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4032 * round what is now in bits to nearest long in bits, then return it in
4035 static unsigned long __init
usemap_size(unsigned long zonesize
)
4037 unsigned long usemapsize
;
4039 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
4040 usemapsize
= usemapsize
>> pageblock_order
;
4041 usemapsize
*= NR_PAGEBLOCK_BITS
;
4042 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
4044 return usemapsize
/ 8;
4047 static void __init
setup_usemap(struct pglist_data
*pgdat
,
4048 struct zone
*zone
, unsigned long zonesize
)
4050 unsigned long usemapsize
= usemap_size(zonesize
);
4051 zone
->pageblock_flags
= NULL
;
4053 zone
->pageblock_flags
= alloc_bootmem_node_nopanic(pgdat
,
4057 static inline void setup_usemap(struct pglist_data
*pgdat
,
4058 struct zone
*zone
, unsigned long zonesize
) {}
4059 #endif /* CONFIG_SPARSEMEM */
4061 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4063 /* Return a sensible default order for the pageblock size. */
4064 static inline int pageblock_default_order(void)
4066 if (HPAGE_SHIFT
> PAGE_SHIFT
)
4067 return HUGETLB_PAGE_ORDER
;
4072 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4073 static inline void __init
set_pageblock_order(unsigned int order
)
4075 /* Check that pageblock_nr_pages has not already been setup */
4076 if (pageblock_order
)
4080 * Assume the largest contiguous order of interest is a huge page.
4081 * This value may be variable depending on boot parameters on IA64
4083 pageblock_order
= order
;
4085 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4088 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4089 * and pageblock_default_order() are unused as pageblock_order is set
4090 * at compile-time. See include/linux/pageblock-flags.h for the values of
4091 * pageblock_order based on the kernel config
4093 static inline int pageblock_default_order(unsigned int order
)
4097 #define set_pageblock_order(x) do {} while (0)
4099 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4102 * Set up the zone data structures:
4103 * - mark all pages reserved
4104 * - mark all memory queues empty
4105 * - clear the memory bitmaps
4107 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
4108 unsigned long *zones_size
, unsigned long *zholes_size
)
4111 int nid
= pgdat
->node_id
;
4112 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
4115 pgdat_resize_init(pgdat
);
4116 pgdat
->nr_zones
= 0;
4117 init_waitqueue_head(&pgdat
->kswapd_wait
);
4118 pgdat
->kswapd_max_order
= 0;
4119 pgdat_page_cgroup_init(pgdat
);
4121 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4122 struct zone
*zone
= pgdat
->node_zones
+ j
;
4123 unsigned long size
, realsize
, memmap_pages
;
4126 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
4127 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
4131 * Adjust realsize so that it accounts for how much memory
4132 * is used by this zone for memmap. This affects the watermark
4133 * and per-cpu initialisations
4136 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
4137 if (realsize
>= memmap_pages
) {
4138 realsize
-= memmap_pages
;
4141 " %s zone: %lu pages used for memmap\n",
4142 zone_names
[j
], memmap_pages
);
4145 " %s zone: %lu pages exceeds realsize %lu\n",
4146 zone_names
[j
], memmap_pages
, realsize
);
4148 /* Account for reserved pages */
4149 if (j
== 0 && realsize
> dma_reserve
) {
4150 realsize
-= dma_reserve
;
4151 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
4152 zone_names
[0], dma_reserve
);
4155 if (!is_highmem_idx(j
))
4156 nr_kernel_pages
+= realsize
;
4157 nr_all_pages
+= realsize
;
4159 zone
->spanned_pages
= size
;
4160 zone
->present_pages
= realsize
;
4163 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
4165 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
4167 zone
->name
= zone_names
[j
];
4168 spin_lock_init(&zone
->lock
);
4169 spin_lock_init(&zone
->lru_lock
);
4170 zone_seqlock_init(zone
);
4171 zone
->zone_pgdat
= pgdat
;
4173 zone_pcp_init(zone
);
4175 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
4176 zone
->reclaim_stat
.recent_rotated
[0] = 0;
4177 zone
->reclaim_stat
.recent_rotated
[1] = 0;
4178 zone
->reclaim_stat
.recent_scanned
[0] = 0;
4179 zone
->reclaim_stat
.recent_scanned
[1] = 0;
4180 zap_zone_vm_stats(zone
);
4185 set_pageblock_order(pageblock_default_order());
4186 setup_usemap(pgdat
, zone
, size
);
4187 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
4188 size
, MEMMAP_EARLY
);
4190 memmap_init(size
, nid
, j
, zone_start_pfn
);
4191 zone_start_pfn
+= size
;
4195 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
4197 /* Skip empty nodes */
4198 if (!pgdat
->node_spanned_pages
)
4201 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4202 /* ia64 gets its own node_mem_map, before this, without bootmem */
4203 if (!pgdat
->node_mem_map
) {
4204 unsigned long size
, start
, end
;
4208 * The zone's endpoints aren't required to be MAX_ORDER
4209 * aligned but the node_mem_map endpoints must be in order
4210 * for the buddy allocator to function correctly.
4212 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
4213 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
4214 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
4215 size
= (end
- start
) * sizeof(struct page
);
4216 map
= alloc_remap(pgdat
->node_id
, size
);
4218 map
= alloc_bootmem_node_nopanic(pgdat
, size
);
4219 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
4221 #ifndef CONFIG_NEED_MULTIPLE_NODES
4223 * With no DISCONTIG, the global mem_map is just set as node 0's
4225 if (pgdat
== NODE_DATA(0)) {
4226 mem_map
= NODE_DATA(0)->node_mem_map
;
4227 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4228 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
4229 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
4230 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4233 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4236 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
4237 unsigned long node_start_pfn
, unsigned long *zholes_size
)
4239 pg_data_t
*pgdat
= NODE_DATA(nid
);
4241 pgdat
->node_id
= nid
;
4242 pgdat
->node_start_pfn
= node_start_pfn
;
4243 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
4245 alloc_node_mem_map(pgdat
);
4246 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4247 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4248 nid
, (unsigned long)pgdat
,
4249 (unsigned long)pgdat
->node_mem_map
);
4252 free_area_init_core(pgdat
, zones_size
, zholes_size
);
4255 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4257 #if MAX_NUMNODES > 1
4259 * Figure out the number of possible node ids.
4261 static void __init
setup_nr_node_ids(void)
4264 unsigned int highest
= 0;
4266 for_each_node_mask(node
, node_possible_map
)
4268 nr_node_ids
= highest
+ 1;
4271 static inline void setup_nr_node_ids(void)
4277 * node_map_pfn_alignment - determine the maximum internode alignment
4279 * This function should be called after node map is populated and sorted.
4280 * It calculates the maximum power of two alignment which can distinguish
4283 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4284 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
4285 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
4286 * shifted, 1GiB is enough and this function will indicate so.
4288 * This is used to test whether pfn -> nid mapping of the chosen memory
4289 * model has fine enough granularity to avoid incorrect mapping for the
4290 * populated node map.
4292 * Returns the determined alignment in pfn's. 0 if there is no alignment
4293 * requirement (single node).
4295 unsigned long __init
node_map_pfn_alignment(void)
4297 unsigned long accl_mask
= 0, last_end
= 0;
4298 unsigned long start
, end
, mask
;
4302 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
4303 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
4310 * Start with a mask granular enough to pin-point to the
4311 * start pfn and tick off bits one-by-one until it becomes
4312 * too coarse to separate the current node from the last.
4314 mask
= ~((1 << __ffs(start
)) - 1);
4315 while (mask
&& last_end
<= (start
& (mask
<< 1)))
4318 /* accumulate all internode masks */
4322 /* convert mask to number of pages */
4323 return ~accl_mask
+ 1;
4326 /* Find the lowest pfn for a node */
4327 static unsigned long __init
find_min_pfn_for_node(int nid
)
4329 unsigned long min_pfn
= ULONG_MAX
;
4330 unsigned long start_pfn
;
4333 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
4334 min_pfn
= min(min_pfn
, start_pfn
);
4336 if (min_pfn
== ULONG_MAX
) {
4338 "Could not find start_pfn for node %d\n", nid
);
4346 * find_min_pfn_with_active_regions - Find the minimum PFN registered
4348 * It returns the minimum PFN based on information provided via
4349 * add_active_range().
4351 unsigned long __init
find_min_pfn_with_active_regions(void)
4353 return find_min_pfn_for_node(MAX_NUMNODES
);
4357 * early_calculate_totalpages()
4358 * Sum pages in active regions for movable zone.
4359 * Populate N_HIGH_MEMORY for calculating usable_nodes.
4361 static unsigned long __init
early_calculate_totalpages(void)
4363 unsigned long totalpages
= 0;
4364 unsigned long start_pfn
, end_pfn
;
4367 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
4368 unsigned long pages
= end_pfn
- start_pfn
;
4370 totalpages
+= pages
;
4372 node_set_state(nid
, N_HIGH_MEMORY
);
4378 * Find the PFN the Movable zone begins in each node. Kernel memory
4379 * is spread evenly between nodes as long as the nodes have enough
4380 * memory. When they don't, some nodes will have more kernelcore than
4383 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
4386 unsigned long usable_startpfn
;
4387 unsigned long kernelcore_node
, kernelcore_remaining
;
4388 /* save the state before borrow the nodemask */
4389 nodemask_t saved_node_state
= node_states
[N_HIGH_MEMORY
];
4390 unsigned long totalpages
= early_calculate_totalpages();
4391 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
4394 * If movablecore was specified, calculate what size of
4395 * kernelcore that corresponds so that memory usable for
4396 * any allocation type is evenly spread. If both kernelcore
4397 * and movablecore are specified, then the value of kernelcore
4398 * will be used for required_kernelcore if it's greater than
4399 * what movablecore would have allowed.
4401 if (required_movablecore
) {
4402 unsigned long corepages
;
4405 * Round-up so that ZONE_MOVABLE is at least as large as what
4406 * was requested by the user
4408 required_movablecore
=
4409 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
4410 corepages
= totalpages
- required_movablecore
;
4412 required_kernelcore
= max(required_kernelcore
, corepages
);
4415 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
4416 if (!required_kernelcore
)
4419 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
4420 find_usable_zone_for_movable();
4421 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
4424 /* Spread kernelcore memory as evenly as possible throughout nodes */
4425 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4426 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4427 unsigned long start_pfn
, end_pfn
;
4430 * Recalculate kernelcore_node if the division per node
4431 * now exceeds what is necessary to satisfy the requested
4432 * amount of memory for the kernel
4434 if (required_kernelcore
< kernelcore_node
)
4435 kernelcore_node
= required_kernelcore
/ usable_nodes
;
4438 * As the map is walked, we track how much memory is usable
4439 * by the kernel using kernelcore_remaining. When it is
4440 * 0, the rest of the node is usable by ZONE_MOVABLE
4442 kernelcore_remaining
= kernelcore_node
;
4444 /* Go through each range of PFNs within this node */
4445 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4446 unsigned long size_pages
;
4448 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
4449 if (start_pfn
>= end_pfn
)
4452 /* Account for what is only usable for kernelcore */
4453 if (start_pfn
< usable_startpfn
) {
4454 unsigned long kernel_pages
;
4455 kernel_pages
= min(end_pfn
, usable_startpfn
)
4458 kernelcore_remaining
-= min(kernel_pages
,
4459 kernelcore_remaining
);
4460 required_kernelcore
-= min(kernel_pages
,
4461 required_kernelcore
);
4463 /* Continue if range is now fully accounted */
4464 if (end_pfn
<= usable_startpfn
) {
4467 * Push zone_movable_pfn to the end so
4468 * that if we have to rebalance
4469 * kernelcore across nodes, we will
4470 * not double account here
4472 zone_movable_pfn
[nid
] = end_pfn
;
4475 start_pfn
= usable_startpfn
;
4479 * The usable PFN range for ZONE_MOVABLE is from
4480 * start_pfn->end_pfn. Calculate size_pages as the
4481 * number of pages used as kernelcore
4483 size_pages
= end_pfn
- start_pfn
;
4484 if (size_pages
> kernelcore_remaining
)
4485 size_pages
= kernelcore_remaining
;
4486 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4489 * Some kernelcore has been met, update counts and
4490 * break if the kernelcore for this node has been
4493 required_kernelcore
-= min(required_kernelcore
,
4495 kernelcore_remaining
-= size_pages
;
4496 if (!kernelcore_remaining
)
4502 * If there is still required_kernelcore, we do another pass with one
4503 * less node in the count. This will push zone_movable_pfn[nid] further
4504 * along on the nodes that still have memory until kernelcore is
4508 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4511 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4512 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4513 zone_movable_pfn
[nid
] =
4514 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4517 /* restore the node_state */
4518 node_states
[N_HIGH_MEMORY
] = saved_node_state
;
4521 /* Any regular memory on that node ? */
4522 static void check_for_regular_memory(pg_data_t
*pgdat
)
4524 #ifdef CONFIG_HIGHMEM
4525 enum zone_type zone_type
;
4527 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4528 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4529 if (zone
->present_pages
)
4530 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4536 * free_area_init_nodes - Initialise all pg_data_t and zone data
4537 * @max_zone_pfn: an array of max PFNs for each zone
4539 * This will call free_area_init_node() for each active node in the system.
4540 * Using the page ranges provided by add_active_range(), the size of each
4541 * zone in each node and their holes is calculated. If the maximum PFN
4542 * between two adjacent zones match, it is assumed that the zone is empty.
4543 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4544 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4545 * starts where the previous one ended. For example, ZONE_DMA32 starts
4546 * at arch_max_dma_pfn.
4548 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4550 unsigned long start_pfn
, end_pfn
;
4553 /* Record where the zone boundaries are */
4554 memset(arch_zone_lowest_possible_pfn
, 0,
4555 sizeof(arch_zone_lowest_possible_pfn
));
4556 memset(arch_zone_highest_possible_pfn
, 0,
4557 sizeof(arch_zone_highest_possible_pfn
));
4558 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4559 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4560 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4561 if (i
== ZONE_MOVABLE
)
4563 arch_zone_lowest_possible_pfn
[i
] =
4564 arch_zone_highest_possible_pfn
[i
-1];
4565 arch_zone_highest_possible_pfn
[i
] =
4566 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4568 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4569 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4571 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4572 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4573 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4575 /* Print out the zone ranges */
4576 printk("Zone PFN ranges:\n");
4577 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4578 if (i
== ZONE_MOVABLE
)
4580 printk(" %-8s ", zone_names
[i
]);
4581 if (arch_zone_lowest_possible_pfn
[i
] ==
4582 arch_zone_highest_possible_pfn
[i
])
4585 printk("%0#10lx -> %0#10lx\n",
4586 arch_zone_lowest_possible_pfn
[i
],
4587 arch_zone_highest_possible_pfn
[i
]);
4590 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4591 printk("Movable zone start PFN for each node\n");
4592 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4593 if (zone_movable_pfn
[i
])
4594 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4597 /* Print out the early_node_map[] */
4598 printk("Early memory PFN ranges\n");
4599 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
4600 printk(" %3d: %0#10lx -> %0#10lx\n", nid
, start_pfn
, end_pfn
);
4602 /* Initialise every node */
4603 mminit_verify_pageflags_layout();
4604 setup_nr_node_ids();
4605 for_each_online_node(nid
) {
4606 pg_data_t
*pgdat
= NODE_DATA(nid
);
4607 free_area_init_node(nid
, NULL
,
4608 find_min_pfn_for_node(nid
), NULL
);
4610 /* Any memory on that node */
4611 if (pgdat
->node_present_pages
)
4612 node_set_state(nid
, N_HIGH_MEMORY
);
4613 check_for_regular_memory(pgdat
);
4617 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4619 unsigned long long coremem
;
4623 coremem
= memparse(p
, &p
);
4624 *core
= coremem
>> PAGE_SHIFT
;
4626 /* Paranoid check that UL is enough for the coremem value */
4627 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4633 * kernelcore=size sets the amount of memory for use for allocations that
4634 * cannot be reclaimed or migrated.
4636 static int __init
cmdline_parse_kernelcore(char *p
)
4638 return cmdline_parse_core(p
, &required_kernelcore
);
4642 * movablecore=size sets the amount of memory for use for allocations that
4643 * can be reclaimed or migrated.
4645 static int __init
cmdline_parse_movablecore(char *p
)
4647 return cmdline_parse_core(p
, &required_movablecore
);
4650 early_param("kernelcore", cmdline_parse_kernelcore
);
4651 early_param("movablecore", cmdline_parse_movablecore
);
4653 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4656 * set_dma_reserve - set the specified number of pages reserved in the first zone
4657 * @new_dma_reserve: The number of pages to mark reserved
4659 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4660 * In the DMA zone, a significant percentage may be consumed by kernel image
4661 * and other unfreeable allocations which can skew the watermarks badly. This
4662 * function may optionally be used to account for unfreeable pages in the
4663 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4664 * smaller per-cpu batchsize.
4666 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4668 dma_reserve
= new_dma_reserve
;
4671 void __init
free_area_init(unsigned long *zones_size
)
4673 free_area_init_node(0, zones_size
,
4674 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4677 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4678 unsigned long action
, void *hcpu
)
4680 int cpu
= (unsigned long)hcpu
;
4682 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4686 * Spill the event counters of the dead processor
4687 * into the current processors event counters.
4688 * This artificially elevates the count of the current
4691 vm_events_fold_cpu(cpu
);
4694 * Zero the differential counters of the dead processor
4695 * so that the vm statistics are consistent.
4697 * This is only okay since the processor is dead and cannot
4698 * race with what we are doing.
4700 refresh_cpu_vm_stats(cpu
);
4705 void __init
page_alloc_init(void)
4707 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4711 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4712 * or min_free_kbytes changes.
4714 static void calculate_totalreserve_pages(void)
4716 struct pglist_data
*pgdat
;
4717 unsigned long reserve_pages
= 0;
4718 enum zone_type i
, j
;
4720 for_each_online_pgdat(pgdat
) {
4721 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4722 struct zone
*zone
= pgdat
->node_zones
+ i
;
4723 unsigned long max
= 0;
4725 /* Find valid and maximum lowmem_reserve in the zone */
4726 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4727 if (zone
->lowmem_reserve
[j
] > max
)
4728 max
= zone
->lowmem_reserve
[j
];
4731 /* we treat the high watermark as reserved pages. */
4732 max
+= high_wmark_pages(zone
);
4734 if (max
> zone
->present_pages
)
4735 max
= zone
->present_pages
;
4736 reserve_pages
+= max
;
4739 totalreserve_pages
= reserve_pages
;
4743 * setup_per_zone_lowmem_reserve - called whenever
4744 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4745 * has a correct pages reserved value, so an adequate number of
4746 * pages are left in the zone after a successful __alloc_pages().
4748 static void setup_per_zone_lowmem_reserve(void)
4750 struct pglist_data
*pgdat
;
4751 enum zone_type j
, idx
;
4753 for_each_online_pgdat(pgdat
) {
4754 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4755 struct zone
*zone
= pgdat
->node_zones
+ j
;
4756 unsigned long present_pages
= zone
->present_pages
;
4758 zone
->lowmem_reserve
[j
] = 0;
4762 struct zone
*lower_zone
;
4766 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4767 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4769 lower_zone
= pgdat
->node_zones
+ idx
;
4770 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4771 sysctl_lowmem_reserve_ratio
[idx
];
4772 present_pages
+= lower_zone
->present_pages
;
4777 /* update totalreserve_pages */
4778 calculate_totalreserve_pages();
4782 * setup_per_zone_wmarks - called when min_free_kbytes changes
4783 * or when memory is hot-{added|removed}
4785 * Ensures that the watermark[min,low,high] values for each zone are set
4786 * correctly with respect to min_free_kbytes.
4788 void setup_per_zone_wmarks(void)
4790 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4791 unsigned long lowmem_pages
= 0;
4793 unsigned long flags
;
4795 /* Calculate total number of !ZONE_HIGHMEM pages */
4796 for_each_zone(zone
) {
4797 if (!is_highmem(zone
))
4798 lowmem_pages
+= zone
->present_pages
;
4801 for_each_zone(zone
) {
4804 spin_lock_irqsave(&zone
->lock
, flags
);
4805 tmp
= (u64
)pages_min
* zone
->present_pages
;
4806 do_div(tmp
, lowmem_pages
);
4807 if (is_highmem(zone
)) {
4809 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4810 * need highmem pages, so cap pages_min to a small
4813 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
4814 * deltas controls asynch page reclaim, and so should
4815 * not be capped for highmem.
4819 min_pages
= zone
->present_pages
/ 1024;
4820 if (min_pages
< SWAP_CLUSTER_MAX
)
4821 min_pages
= SWAP_CLUSTER_MAX
;
4822 if (min_pages
> 128)
4824 zone
->watermark
[WMARK_MIN
] = min_pages
;
4827 * If it's a lowmem zone, reserve a number of pages
4828 * proportionate to the zone's size.
4830 zone
->watermark
[WMARK_MIN
] = tmp
;
4833 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
4834 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
4835 setup_zone_migrate_reserve(zone
);
4836 spin_unlock_irqrestore(&zone
->lock
, flags
);
4839 /* update totalreserve_pages */
4840 calculate_totalreserve_pages();
4844 * The inactive anon list should be small enough that the VM never has to
4845 * do too much work, but large enough that each inactive page has a chance
4846 * to be referenced again before it is swapped out.
4848 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4849 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4850 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4851 * the anonymous pages are kept on the inactive list.
4854 * memory ratio inactive anon
4855 * -------------------------------------
4864 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
4866 unsigned int gb
, ratio
;
4868 /* Zone size in gigabytes */
4869 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4871 ratio
= int_sqrt(10 * gb
);
4875 zone
->inactive_ratio
= ratio
;
4878 static void __meminit
setup_per_zone_inactive_ratio(void)
4883 calculate_zone_inactive_ratio(zone
);
4887 * Initialise min_free_kbytes.
4889 * For small machines we want it small (128k min). For large machines
4890 * we want it large (64MB max). But it is not linear, because network
4891 * bandwidth does not increase linearly with machine size. We use
4893 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4894 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4910 int __meminit
init_per_zone_wmark_min(void)
4912 unsigned long lowmem_kbytes
;
4914 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4916 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4917 if (min_free_kbytes
< 128)
4918 min_free_kbytes
= 128;
4919 if (min_free_kbytes
> 65536)
4920 min_free_kbytes
= 65536;
4921 setup_per_zone_wmarks();
4922 refresh_zone_stat_thresholds();
4923 setup_per_zone_lowmem_reserve();
4924 setup_per_zone_inactive_ratio();
4927 module_init(init_per_zone_wmark_min
)
4930 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4931 * that we can call two helper functions whenever min_free_kbytes
4934 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4935 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4937 proc_dointvec(table
, write
, buffer
, length
, ppos
);
4939 setup_per_zone_wmarks();
4944 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4945 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4950 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
4955 zone
->min_unmapped_pages
= (zone
->present_pages
*
4956 sysctl_min_unmapped_ratio
) / 100;
4960 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4961 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4966 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
4971 zone
->min_slab_pages
= (zone
->present_pages
*
4972 sysctl_min_slab_ratio
) / 100;
4978 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4979 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4980 * whenever sysctl_lowmem_reserve_ratio changes.
4982 * The reserve ratio obviously has absolutely no relation with the
4983 * minimum watermarks. The lowmem reserve ratio can only make sense
4984 * if in function of the boot time zone sizes.
4986 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4987 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4989 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
4990 setup_per_zone_lowmem_reserve();
4995 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4996 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4997 * can have before it gets flushed back to buddy allocator.
5000 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
5001 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
5007 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
5008 if (!write
|| (ret
== -EINVAL
))
5010 for_each_populated_zone(zone
) {
5011 for_each_possible_cpu(cpu
) {
5013 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
5014 setup_pagelist_highmark(
5015 per_cpu_ptr(zone
->pageset
, cpu
), high
);
5021 int hashdist
= HASHDIST_DEFAULT
;
5024 static int __init
set_hashdist(char *str
)
5028 hashdist
= simple_strtoul(str
, &str
, 0);
5031 __setup("hashdist=", set_hashdist
);
5035 * allocate a large system hash table from bootmem
5036 * - it is assumed that the hash table must contain an exact power-of-2
5037 * quantity of entries
5038 * - limit is the number of hash buckets, not the total allocation size
5040 void *__init
alloc_large_system_hash(const char *tablename
,
5041 unsigned long bucketsize
,
5042 unsigned long numentries
,
5045 unsigned int *_hash_shift
,
5046 unsigned int *_hash_mask
,
5047 unsigned long limit
)
5049 unsigned long long max
= limit
;
5050 unsigned long log2qty
, size
;
5053 /* allow the kernel cmdline to have a say */
5055 /* round applicable memory size up to nearest megabyte */
5056 numentries
= nr_kernel_pages
;
5057 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
5058 numentries
>>= 20 - PAGE_SHIFT
;
5059 numentries
<<= 20 - PAGE_SHIFT
;
5061 /* limit to 1 bucket per 2^scale bytes of low memory */
5062 if (scale
> PAGE_SHIFT
)
5063 numentries
>>= (scale
- PAGE_SHIFT
);
5065 numentries
<<= (PAGE_SHIFT
- scale
);
5067 /* Make sure we've got at least a 0-order allocation.. */
5068 if (unlikely(flags
& HASH_SMALL
)) {
5069 /* Makes no sense without HASH_EARLY */
5070 WARN_ON(!(flags
& HASH_EARLY
));
5071 if (!(numentries
>> *_hash_shift
)) {
5072 numentries
= 1UL << *_hash_shift
;
5073 BUG_ON(!numentries
);
5075 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
5076 numentries
= PAGE_SIZE
/ bucketsize
;
5078 numentries
= roundup_pow_of_two(numentries
);
5080 /* limit allocation size to 1/16 total memory by default */
5082 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
5083 do_div(max
, bucketsize
);
5086 if (numentries
> max
)
5089 log2qty
= ilog2(numentries
);
5092 size
= bucketsize
<< log2qty
;
5093 if (flags
& HASH_EARLY
)
5094 table
= alloc_bootmem_nopanic(size
);
5096 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
5099 * If bucketsize is not a power-of-two, we may free
5100 * some pages at the end of hash table which
5101 * alloc_pages_exact() automatically does
5103 if (get_order(size
) < MAX_ORDER
) {
5104 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
5105 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
5108 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
5111 panic("Failed to allocate %s hash table\n", tablename
);
5113 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
5116 ilog2(size
) - PAGE_SHIFT
,
5120 *_hash_shift
= log2qty
;
5122 *_hash_mask
= (1 << log2qty
) - 1;
5127 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5128 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
5131 #ifdef CONFIG_SPARSEMEM
5132 return __pfn_to_section(pfn
)->pageblock_flags
;
5134 return zone
->pageblock_flags
;
5135 #endif /* CONFIG_SPARSEMEM */
5138 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
5140 #ifdef CONFIG_SPARSEMEM
5141 pfn
&= (PAGES_PER_SECTION
-1);
5142 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5144 pfn
= pfn
- zone
->zone_start_pfn
;
5145 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
5146 #endif /* CONFIG_SPARSEMEM */
5150 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
5151 * @page: The page within the block of interest
5152 * @start_bitidx: The first bit of interest to retrieve
5153 * @end_bitidx: The last bit of interest
5154 * returns pageblock_bits flags
5156 unsigned long get_pageblock_flags_group(struct page
*page
,
5157 int start_bitidx
, int end_bitidx
)
5160 unsigned long *bitmap
;
5161 unsigned long pfn
, bitidx
;
5162 unsigned long flags
= 0;
5163 unsigned long value
= 1;
5165 zone
= page_zone(page
);
5166 pfn
= page_to_pfn(page
);
5167 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5168 bitidx
= pfn_to_bitidx(zone
, pfn
);
5170 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5171 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
5178 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
5179 * @page: The page within the block of interest
5180 * @start_bitidx: The first bit of interest
5181 * @end_bitidx: The last bit of interest
5182 * @flags: The flags to set
5184 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
5185 int start_bitidx
, int end_bitidx
)
5188 unsigned long *bitmap
;
5189 unsigned long pfn
, bitidx
;
5190 unsigned long value
= 1;
5192 zone
= page_zone(page
);
5193 pfn
= page_to_pfn(page
);
5194 bitmap
= get_pageblock_bitmap(zone
, pfn
);
5195 bitidx
= pfn_to_bitidx(zone
, pfn
);
5196 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
5197 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
5199 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
5201 __set_bit(bitidx
+ start_bitidx
, bitmap
);
5203 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
5207 * This is designed as sub function...plz see page_isolation.c also.
5208 * set/clear page block's type to be ISOLATE.
5209 * page allocater never alloc memory from ISOLATE block.
5213 __count_immobile_pages(struct zone
*zone
, struct page
*page
, int count
)
5215 unsigned long pfn
, iter
, found
;
5217 * For avoiding noise data, lru_add_drain_all() should be called
5218 * If ZONE_MOVABLE, the zone never contains immobile pages
5220 if (zone_idx(zone
) == ZONE_MOVABLE
)
5223 if (get_pageblock_migratetype(page
) == MIGRATE_MOVABLE
)
5226 pfn
= page_to_pfn(page
);
5227 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
5228 unsigned long check
= pfn
+ iter
;
5230 if (!pfn_valid_within(check
))
5233 page
= pfn_to_page(check
);
5234 if (!page_count(page
)) {
5235 if (PageBuddy(page
))
5236 iter
+= (1 << page_order(page
)) - 1;
5242 * If there are RECLAIMABLE pages, we need to check it.
5243 * But now, memory offline itself doesn't call shrink_slab()
5244 * and it still to be fixed.
5247 * If the page is not RAM, page_count()should be 0.
5248 * we don't need more check. This is an _used_ not-movable page.
5250 * The problematic thing here is PG_reserved pages. PG_reserved
5251 * is set to both of a memory hole page and a _used_ kernel
5260 bool is_pageblock_removable_nolock(struct page
*page
)
5262 struct zone
*zone
= page_zone(page
);
5263 return __count_immobile_pages(zone
, page
, 0);
5266 int set_migratetype_isolate(struct page
*page
)
5269 unsigned long flags
, pfn
;
5270 struct memory_isolate_notify arg
;
5274 zone
= page_zone(page
);
5276 spin_lock_irqsave(&zone
->lock
, flags
);
5278 pfn
= page_to_pfn(page
);
5279 arg
.start_pfn
= pfn
;
5280 arg
.nr_pages
= pageblock_nr_pages
;
5281 arg
.pages_found
= 0;
5284 * It may be possible to isolate a pageblock even if the
5285 * migratetype is not MIGRATE_MOVABLE. The memory isolation
5286 * notifier chain is used by balloon drivers to return the
5287 * number of pages in a range that are held by the balloon
5288 * driver to shrink memory. If all the pages are accounted for
5289 * by balloons, are free, or on the LRU, isolation can continue.
5290 * Later, for example, when memory hotplug notifier runs, these
5291 * pages reported as "can be isolated" should be isolated(freed)
5292 * by the balloon driver through the memory notifier chain.
5294 notifier_ret
= memory_isolate_notify(MEM_ISOLATE_COUNT
, &arg
);
5295 notifier_ret
= notifier_to_errno(notifier_ret
);
5299 * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself.
5300 * We just check MOVABLE pages.
5302 if (__count_immobile_pages(zone
, page
, arg
.pages_found
))
5306 * immobile means "not-on-lru" paes. If immobile is larger than
5307 * removable-by-driver pages reported by notifier, we'll fail.
5312 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
5313 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
5316 spin_unlock_irqrestore(&zone
->lock
, flags
);
5322 void unset_migratetype_isolate(struct page
*page
)
5325 unsigned long flags
;
5326 zone
= page_zone(page
);
5327 spin_lock_irqsave(&zone
->lock
, flags
);
5328 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
5330 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5331 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
5333 spin_unlock_irqrestore(&zone
->lock
, flags
);
5336 #ifdef CONFIG_MEMORY_HOTREMOVE
5338 * All pages in the range must be isolated before calling this.
5341 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
5347 unsigned long flags
;
5348 /* find the first valid pfn */
5349 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
5354 zone
= page_zone(pfn_to_page(pfn
));
5355 spin_lock_irqsave(&zone
->lock
, flags
);
5357 while (pfn
< end_pfn
) {
5358 if (!pfn_valid(pfn
)) {
5362 page
= pfn_to_page(pfn
);
5363 BUG_ON(page_count(page
));
5364 BUG_ON(!PageBuddy(page
));
5365 order
= page_order(page
);
5366 #ifdef CONFIG_DEBUG_VM
5367 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
5368 pfn
, 1 << order
, end_pfn
);
5370 list_del(&page
->lru
);
5371 rmv_page_order(page
);
5372 zone
->free_area
[order
].nr_free
--;
5373 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
5375 for (i
= 0; i
< (1 << order
); i
++)
5376 SetPageReserved((page
+i
));
5377 pfn
+= (1 << order
);
5379 spin_unlock_irqrestore(&zone
->lock
, flags
);
5383 #ifdef CONFIG_MEMORY_FAILURE
5384 bool is_free_buddy_page(struct page
*page
)
5386 struct zone
*zone
= page_zone(page
);
5387 unsigned long pfn
= page_to_pfn(page
);
5388 unsigned long flags
;
5391 spin_lock_irqsave(&zone
->lock
, flags
);
5392 for (order
= 0; order
< MAX_ORDER
; order
++) {
5393 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
5395 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
5398 spin_unlock_irqrestore(&zone
->lock
, flags
);
5400 return order
< MAX_ORDER
;
5404 static struct trace_print_flags pageflag_names
[] = {
5405 {1UL << PG_locked
, "locked" },
5406 {1UL << PG_error
, "error" },
5407 {1UL << PG_referenced
, "referenced" },
5408 {1UL << PG_uptodate
, "uptodate" },
5409 {1UL << PG_dirty
, "dirty" },
5410 {1UL << PG_lru
, "lru" },
5411 {1UL << PG_active
, "active" },
5412 {1UL << PG_slab
, "slab" },
5413 {1UL << PG_owner_priv_1
, "owner_priv_1" },
5414 {1UL << PG_arch_1
, "arch_1" },
5415 {1UL << PG_reserved
, "reserved" },
5416 {1UL << PG_private
, "private" },
5417 {1UL << PG_private_2
, "private_2" },
5418 {1UL << PG_writeback
, "writeback" },
5419 #ifdef CONFIG_PAGEFLAGS_EXTENDED
5420 {1UL << PG_head
, "head" },
5421 {1UL << PG_tail
, "tail" },
5423 {1UL << PG_compound
, "compound" },
5425 {1UL << PG_swapcache
, "swapcache" },
5426 {1UL << PG_mappedtodisk
, "mappedtodisk" },
5427 {1UL << PG_reclaim
, "reclaim" },
5428 {1UL << PG_swapbacked
, "swapbacked" },
5429 {1UL << PG_unevictable
, "unevictable" },
5431 {1UL << PG_mlocked
, "mlocked" },
5433 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
5434 {1UL << PG_uncached
, "uncached" },
5436 #ifdef CONFIG_MEMORY_FAILURE
5437 {1UL << PG_hwpoison
, "hwpoison" },
5442 static void dump_page_flags(unsigned long flags
)
5444 const char *delim
= "";
5448 printk(KERN_ALERT
"page flags: %#lx(", flags
);
5450 /* remove zone id */
5451 flags
&= (1UL << NR_PAGEFLAGS
) - 1;
5453 for (i
= 0; pageflag_names
[i
].name
&& flags
; i
++) {
5455 mask
= pageflag_names
[i
].mask
;
5456 if ((flags
& mask
) != mask
)
5460 printk("%s%s", delim
, pageflag_names
[i
].name
);
5464 /* check for left over flags */
5466 printk("%s%#lx", delim
, flags
);
5471 void dump_page(struct page
*page
)
5474 "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
5475 page
, atomic_read(&page
->_count
), page_mapcount(page
),
5476 page
->mapping
, page
->index
);
5477 dump_page_flags(page
->flags
);
5478 mem_cgroup_print_bad_page(page
);