xref: /openbmc/linux/mm/page_alloc.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  *  linux/mm/page_alloc.c
3  *
4  *  Manages the free list, the system allocates free pages here.
5  *  Note that kmalloc() lives in slab.c
6  *
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)
15  */
16 
17 #include <linux/stddef.h>
18 #include <linux/mm.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
43 #include <linux/fault-inject.h>
44 
45 #include <asm/tlbflush.h>
46 #include <asm/div64.h>
47 #include "internal.h"
48 
49 /*
50  * MCD - HACK: Find somewhere to initialize this EARLY, or make this
51  * initializer cleaner
52  */
53 nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
54 EXPORT_SYMBOL(node_online_map);
55 nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
56 EXPORT_SYMBOL(node_possible_map);
57 unsigned long totalram_pages __read_mostly;
58 unsigned long totalreserve_pages __read_mostly;
59 long nr_swap_pages;
60 int percpu_pagelist_fraction;
61 
62 static void __free_pages_ok(struct page *page, unsigned int order);
63 
64 /*
65  * results with 256, 32 in the lowmem_reserve sysctl:
66  *	1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
67  *	1G machine -> (16M dma, 784M normal, 224M high)
68  *	NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
69  *	HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
70  *	HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
71  *
72  * TBD: should special case ZONE_DMA32 machines here - in those we normally
73  * don't need any ZONE_NORMAL reservation
74  */
75 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
76 #ifdef CONFIG_ZONE_DMA
77 	 256,
78 #endif
79 #ifdef CONFIG_ZONE_DMA32
80 	 256,
81 #endif
82 #ifdef CONFIG_HIGHMEM
83 	 32
84 #endif
85 };
86 
87 EXPORT_SYMBOL(totalram_pages);
88 
89 static char * const zone_names[MAX_NR_ZONES] = {
90 #ifdef CONFIG_ZONE_DMA
91 	 "DMA",
92 #endif
93 #ifdef CONFIG_ZONE_DMA32
94 	 "DMA32",
95 #endif
96 	 "Normal",
97 #ifdef CONFIG_HIGHMEM
98 	 "HighMem"
99 #endif
100 };
101 
102 int min_free_kbytes = 1024;
103 
104 unsigned long __meminitdata nr_kernel_pages;
105 unsigned long __meminitdata nr_all_pages;
106 static unsigned long __meminitdata dma_reserve;
107 
108 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
109   /*
110    * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
111    * ranges of memory (RAM) that may be registered with add_active_range().
112    * Ranges passed to add_active_range() will be merged if possible
113    * so the number of times add_active_range() can be called is
114    * related to the number of nodes and the number of holes
115    */
116   #ifdef CONFIG_MAX_ACTIVE_REGIONS
117     /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
118     #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
119   #else
120     #if MAX_NUMNODES >= 32
121       /* If there can be many nodes, allow up to 50 holes per node */
122       #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
123     #else
124       /* By default, allow up to 256 distinct regions */
125       #define MAX_ACTIVE_REGIONS 256
126     #endif
127   #endif
128 
129   struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
130   int __meminitdata nr_nodemap_entries;
131   unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
132   unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
133 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
134   unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES];
135   unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES];
136 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
137 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
138 
139 #ifdef CONFIG_DEBUG_VM
140 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
141 {
142 	int ret = 0;
143 	unsigned seq;
144 	unsigned long pfn = page_to_pfn(page);
145 
146 	do {
147 		seq = zone_span_seqbegin(zone);
148 		if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
149 			ret = 1;
150 		else if (pfn < zone->zone_start_pfn)
151 			ret = 1;
152 	} while (zone_span_seqretry(zone, seq));
153 
154 	return ret;
155 }
156 
157 static int page_is_consistent(struct zone *zone, struct page *page)
158 {
159 	if (!pfn_valid_within(page_to_pfn(page)))
160 		return 0;
161 	if (zone != page_zone(page))
162 		return 0;
163 
164 	return 1;
165 }
166 /*
167  * Temporary debugging check for pages not lying within a given zone.
168  */
169 static int bad_range(struct zone *zone, struct page *page)
170 {
171 	if (page_outside_zone_boundaries(zone, page))
172 		return 1;
173 	if (!page_is_consistent(zone, page))
174 		return 1;
175 
176 	return 0;
177 }
178 #else
179 static inline int bad_range(struct zone *zone, struct page *page)
180 {
181 	return 0;
182 }
183 #endif
184 
185 static void bad_page(struct page *page)
186 {
187 	printk(KERN_EMERG "Bad page state in process '%s'\n"
188 		KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
189 		KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
190 		KERN_EMERG "Backtrace:\n",
191 		current->comm, page, (int)(2*sizeof(unsigned long)),
192 		(unsigned long)page->flags, page->mapping,
193 		page_mapcount(page), page_count(page));
194 	dump_stack();
195 	page->flags &= ~(1 << PG_lru	|
196 			1 << PG_private |
197 			1 << PG_locked	|
198 			1 << PG_active	|
199 			1 << PG_dirty	|
200 			1 << PG_reclaim |
201 			1 << PG_slab    |
202 			1 << PG_swapcache |
203 			1 << PG_writeback |
204 			1 << PG_buddy );
205 	set_page_count(page, 0);
206 	reset_page_mapcount(page);
207 	page->mapping = NULL;
208 	add_taint(TAINT_BAD_PAGE);
209 }
210 
211 /*
212  * Higher-order pages are called "compound pages".  They are structured thusly:
213  *
214  * The first PAGE_SIZE page is called the "head page".
215  *
216  * The remaining PAGE_SIZE pages are called "tail pages".
217  *
218  * All pages have PG_compound set.  All pages have their ->private pointing at
219  * the head page (even the head page has this).
220  *
221  * The first tail page's ->lru.next holds the address of the compound page's
222  * put_page() function.  Its ->lru.prev holds the order of allocation.
223  * This usage means that zero-order pages may not be compound.
224  */
225 
226 static void free_compound_page(struct page *page)
227 {
228 	__free_pages_ok(page, compound_order(page));
229 }
230 
231 static void prep_compound_page(struct page *page, unsigned long order)
232 {
233 	int i;
234 	int nr_pages = 1 << order;
235 
236 	set_compound_page_dtor(page, free_compound_page);
237 	set_compound_order(page, order);
238 	__SetPageHead(page);
239 	for (i = 1; i < nr_pages; i++) {
240 		struct page *p = page + i;
241 
242 		__SetPageTail(p);
243 		p->first_page = page;
244 	}
245 }
246 
247 static void destroy_compound_page(struct page *page, unsigned long order)
248 {
249 	int i;
250 	int nr_pages = 1 << order;
251 
252 	if (unlikely(compound_order(page) != order))
253 		bad_page(page);
254 
255 	if (unlikely(!PageHead(page)))
256 			bad_page(page);
257 	__ClearPageHead(page);
258 	for (i = 1; i < nr_pages; i++) {
259 		struct page *p = page + i;
260 
261 		if (unlikely(!PageTail(p) |
262 				(p->first_page != page)))
263 			bad_page(page);
264 		__ClearPageTail(p);
265 	}
266 }
267 
268 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
269 {
270 	int i;
271 
272 	VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
273 	/*
274 	 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
275 	 * and __GFP_HIGHMEM from hard or soft interrupt context.
276 	 */
277 	VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
278 	for (i = 0; i < (1 << order); i++)
279 		clear_highpage(page + i);
280 }
281 
282 /*
283  * function for dealing with page's order in buddy system.
284  * zone->lock is already acquired when we use these.
285  * So, we don't need atomic page->flags operations here.
286  */
287 static inline unsigned long page_order(struct page *page)
288 {
289 	return page_private(page);
290 }
291 
292 static inline void set_page_order(struct page *page, int order)
293 {
294 	set_page_private(page, order);
295 	__SetPageBuddy(page);
296 }
297 
298 static inline void rmv_page_order(struct page *page)
299 {
300 	__ClearPageBuddy(page);
301 	set_page_private(page, 0);
302 }
303 
304 /*
305  * Locate the struct page for both the matching buddy in our
306  * pair (buddy1) and the combined O(n+1) page they form (page).
307  *
308  * 1) Any buddy B1 will have an order O twin B2 which satisfies
309  * the following equation:
310  *     B2 = B1 ^ (1 << O)
311  * For example, if the starting buddy (buddy2) is #8 its order
312  * 1 buddy is #10:
313  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
314  *
315  * 2) Any buddy B will have an order O+1 parent P which
316  * satisfies the following equation:
317  *     P = B & ~(1 << O)
318  *
319  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
320  */
321 static inline struct page *
322 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
323 {
324 	unsigned long buddy_idx = page_idx ^ (1 << order);
325 
326 	return page + (buddy_idx - page_idx);
327 }
328 
329 static inline unsigned long
330 __find_combined_index(unsigned long page_idx, unsigned int order)
331 {
332 	return (page_idx & ~(1 << order));
333 }
334 
335 /*
336  * This function checks whether a page is free && is the buddy
337  * we can do coalesce a page and its buddy if
338  * (a) the buddy is not in a hole &&
339  * (b) the buddy is in the buddy system &&
340  * (c) a page and its buddy have the same order &&
341  * (d) a page and its buddy are in the same zone.
342  *
343  * For recording whether a page is in the buddy system, we use PG_buddy.
344  * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
345  *
346  * For recording page's order, we use page_private(page).
347  */
348 static inline int page_is_buddy(struct page *page, struct page *buddy,
349 								int order)
350 {
351 	if (!pfn_valid_within(page_to_pfn(buddy)))
352 		return 0;
353 
354 	if (page_zone_id(page) != page_zone_id(buddy))
355 		return 0;
356 
357 	if (PageBuddy(buddy) && page_order(buddy) == order) {
358 		BUG_ON(page_count(buddy) != 0);
359 		return 1;
360 	}
361 	return 0;
362 }
363 
364 /*
365  * Freeing function for a buddy system allocator.
366  *
367  * The concept of a buddy system is to maintain direct-mapped table
368  * (containing bit values) for memory blocks of various "orders".
369  * The bottom level table contains the map for the smallest allocatable
370  * units of memory (here, pages), and each level above it describes
371  * pairs of units from the levels below, hence, "buddies".
372  * At a high level, all that happens here is marking the table entry
373  * at the bottom level available, and propagating the changes upward
374  * as necessary, plus some accounting needed to play nicely with other
375  * parts of the VM system.
376  * At each level, we keep a list of pages, which are heads of continuous
377  * free pages of length of (1 << order) and marked with PG_buddy. Page's
378  * order is recorded in page_private(page) field.
379  * So when we are allocating or freeing one, we can derive the state of the
380  * other.  That is, if we allocate a small block, and both were
381  * free, the remainder of the region must be split into blocks.
382  * If a block is freed, and its buddy is also free, then this
383  * triggers coalescing into a block of larger size.
384  *
385  * -- wli
386  */
387 
388 static inline void __free_one_page(struct page *page,
389 		struct zone *zone, unsigned int order)
390 {
391 	unsigned long page_idx;
392 	int order_size = 1 << order;
393 
394 	if (unlikely(PageCompound(page)))
395 		destroy_compound_page(page, order);
396 
397 	page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
398 
399 	VM_BUG_ON(page_idx & (order_size - 1));
400 	VM_BUG_ON(bad_range(zone, page));
401 
402 	__mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
403 	while (order < MAX_ORDER-1) {
404 		unsigned long combined_idx;
405 		struct free_area *area;
406 		struct page *buddy;
407 
408 		buddy = __page_find_buddy(page, page_idx, order);
409 		if (!page_is_buddy(page, buddy, order))
410 			break;		/* Move the buddy up one level. */
411 
412 		list_del(&buddy->lru);
413 		area = zone->free_area + order;
414 		area->nr_free--;
415 		rmv_page_order(buddy);
416 		combined_idx = __find_combined_index(page_idx, order);
417 		page = page + (combined_idx - page_idx);
418 		page_idx = combined_idx;
419 		order++;
420 	}
421 	set_page_order(page, order);
422 	list_add(&page->lru, &zone->free_area[order].free_list);
423 	zone->free_area[order].nr_free++;
424 }
425 
426 static inline int free_pages_check(struct page *page)
427 {
428 	if (unlikely(page_mapcount(page) |
429 		(page->mapping != NULL)  |
430 		(page_count(page) != 0)  |
431 		(page->flags & (
432 			1 << PG_lru	|
433 			1 << PG_private |
434 			1 << PG_locked	|
435 			1 << PG_active	|
436 			1 << PG_slab	|
437 			1 << PG_swapcache |
438 			1 << PG_writeback |
439 			1 << PG_reserved |
440 			1 << PG_buddy ))))
441 		bad_page(page);
442 	/*
443 	 * PageReclaim == PageTail. It is only an error
444 	 * for PageReclaim to be set if PageCompound is clear.
445 	 */
446 	if (unlikely(!PageCompound(page) && PageReclaim(page)))
447 		bad_page(page);
448 	if (PageDirty(page))
449 		__ClearPageDirty(page);
450 	/*
451 	 * For now, we report if PG_reserved was found set, but do not
452 	 * clear it, and do not free the page.  But we shall soon need
453 	 * to do more, for when the ZERO_PAGE count wraps negative.
454 	 */
455 	return PageReserved(page);
456 }
457 
458 /*
459  * Frees a list of pages.
460  * Assumes all pages on list are in same zone, and of same order.
461  * count is the number of pages to free.
462  *
463  * If the zone was previously in an "all pages pinned" state then look to
464  * see if this freeing clears that state.
465  *
466  * And clear the zone's pages_scanned counter, to hold off the "all pages are
467  * pinned" detection logic.
468  */
469 static void free_pages_bulk(struct zone *zone, int count,
470 					struct list_head *list, int order)
471 {
472 	spin_lock(&zone->lock);
473 	zone->all_unreclaimable = 0;
474 	zone->pages_scanned = 0;
475 	while (count--) {
476 		struct page *page;
477 
478 		VM_BUG_ON(list_empty(list));
479 		page = list_entry(list->prev, struct page, lru);
480 		/* have to delete it as __free_one_page list manipulates */
481 		list_del(&page->lru);
482 		__free_one_page(page, zone, order);
483 	}
484 	spin_unlock(&zone->lock);
485 }
486 
487 static void free_one_page(struct zone *zone, struct page *page, int order)
488 {
489 	spin_lock(&zone->lock);
490 	zone->all_unreclaimable = 0;
491 	zone->pages_scanned = 0;
492 	__free_one_page(page, zone, order);
493 	spin_unlock(&zone->lock);
494 }
495 
496 static void __free_pages_ok(struct page *page, unsigned int order)
497 {
498 	unsigned long flags;
499 	int i;
500 	int reserved = 0;
501 
502 	for (i = 0 ; i < (1 << order) ; ++i)
503 		reserved += free_pages_check(page + i);
504 	if (reserved)
505 		return;
506 
507 	if (!PageHighMem(page))
508 		debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
509 	arch_free_page(page, order);
510 	kernel_map_pages(page, 1 << order, 0);
511 
512 	local_irq_save(flags);
513 	__count_vm_events(PGFREE, 1 << order);
514 	free_one_page(page_zone(page), page, order);
515 	local_irq_restore(flags);
516 }
517 
518 /*
519  * permit the bootmem allocator to evade page validation on high-order frees
520  */
521 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
522 {
523 	if (order == 0) {
524 		__ClearPageReserved(page);
525 		set_page_count(page, 0);
526 		set_page_refcounted(page);
527 		__free_page(page);
528 	} else {
529 		int loop;
530 
531 		prefetchw(page);
532 		for (loop = 0; loop < BITS_PER_LONG; loop++) {
533 			struct page *p = &page[loop];
534 
535 			if (loop + 1 < BITS_PER_LONG)
536 				prefetchw(p + 1);
537 			__ClearPageReserved(p);
538 			set_page_count(p, 0);
539 		}
540 
541 		set_page_refcounted(page);
542 		__free_pages(page, order);
543 	}
544 }
545 
546 
547 /*
548  * The order of subdivision here is critical for the IO subsystem.
549  * Please do not alter this order without good reasons and regression
550  * testing. Specifically, as large blocks of memory are subdivided,
551  * the order in which smaller blocks are delivered depends on the order
552  * they're subdivided in this function. This is the primary factor
553  * influencing the order in which pages are delivered to the IO
554  * subsystem according to empirical testing, and this is also justified
555  * by considering the behavior of a buddy system containing a single
556  * large block of memory acted on by a series of small allocations.
557  * This behavior is a critical factor in sglist merging's success.
558  *
559  * -- wli
560  */
561 static inline void expand(struct zone *zone, struct page *page,
562  	int low, int high, struct free_area *area)
563 {
564 	unsigned long size = 1 << high;
565 
566 	while (high > low) {
567 		area--;
568 		high--;
569 		size >>= 1;
570 		VM_BUG_ON(bad_range(zone, &page[size]));
571 		list_add(&page[size].lru, &area->free_list);
572 		area->nr_free++;
573 		set_page_order(&page[size], high);
574 	}
575 }
576 
577 /*
578  * This page is about to be returned from the page allocator
579  */
580 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
581 {
582 	if (unlikely(page_mapcount(page) |
583 		(page->mapping != NULL)  |
584 		(page_count(page) != 0)  |
585 		(page->flags & (
586 			1 << PG_lru	|
587 			1 << PG_private	|
588 			1 << PG_locked	|
589 			1 << PG_active	|
590 			1 << PG_dirty	|
591 			1 << PG_reclaim	|
592 			1 << PG_slab    |
593 			1 << PG_swapcache |
594 			1 << PG_writeback |
595 			1 << PG_reserved |
596 			1 << PG_buddy ))))
597 		bad_page(page);
598 
599 	/*
600 	 * For now, we report if PG_reserved was found set, but do not
601 	 * clear it, and do not allocate the page: as a safety net.
602 	 */
603 	if (PageReserved(page))
604 		return 1;
605 
606 	page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
607 			1 << PG_referenced | 1 << PG_arch_1 |
608 			1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
609 	set_page_private(page, 0);
610 	set_page_refcounted(page);
611 
612 	arch_alloc_page(page, order);
613 	kernel_map_pages(page, 1 << order, 1);
614 
615 	if (gfp_flags & __GFP_ZERO)
616 		prep_zero_page(page, order, gfp_flags);
617 
618 	if (order && (gfp_flags & __GFP_COMP))
619 		prep_compound_page(page, order);
620 
621 	return 0;
622 }
623 
624 /*
625  * Do the hard work of removing an element from the buddy allocator.
626  * Call me with the zone->lock already held.
627  */
628 static struct page *__rmqueue(struct zone *zone, unsigned int order)
629 {
630 	struct free_area * area;
631 	unsigned int current_order;
632 	struct page *page;
633 
634 	for (current_order = order; current_order < MAX_ORDER; ++current_order) {
635 		area = zone->free_area + current_order;
636 		if (list_empty(&area->free_list))
637 			continue;
638 
639 		page = list_entry(area->free_list.next, struct page, lru);
640 		list_del(&page->lru);
641 		rmv_page_order(page);
642 		area->nr_free--;
643 		__mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
644 		expand(zone, page, order, current_order, area);
645 		return page;
646 	}
647 
648 	return NULL;
649 }
650 
651 /*
652  * Obtain a specified number of elements from the buddy allocator, all under
653  * a single hold of the lock, for efficiency.  Add them to the supplied list.
654  * Returns the number of new pages which were placed at *list.
655  */
656 static int rmqueue_bulk(struct zone *zone, unsigned int order,
657 			unsigned long count, struct list_head *list)
658 {
659 	int i;
660 
661 	spin_lock(&zone->lock);
662 	for (i = 0; i < count; ++i) {
663 		struct page *page = __rmqueue(zone, order);
664 		if (unlikely(page == NULL))
665 			break;
666 		list_add_tail(&page->lru, list);
667 	}
668 	spin_unlock(&zone->lock);
669 	return i;
670 }
671 
672 #if MAX_NUMNODES > 1
673 int nr_node_ids __read_mostly = MAX_NUMNODES;
674 EXPORT_SYMBOL(nr_node_ids);
675 
676 /*
677  * Figure out the number of possible node ids.
678  */
679 static void __init setup_nr_node_ids(void)
680 {
681 	unsigned int node;
682 	unsigned int highest = 0;
683 
684 	for_each_node_mask(node, node_possible_map)
685 		highest = node;
686 	nr_node_ids = highest + 1;
687 }
688 #else
689 static void __init setup_nr_node_ids(void) {}
690 #endif
691 
692 #ifdef CONFIG_NUMA
693 /*
694  * Called from the vmstat counter updater to drain pagesets of this
695  * currently executing processor on remote nodes after they have
696  * expired.
697  *
698  * Note that this function must be called with the thread pinned to
699  * a single processor.
700  */
701 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
702 {
703 	unsigned long flags;
704 	int to_drain;
705 
706 	local_irq_save(flags);
707 	if (pcp->count >= pcp->batch)
708 		to_drain = pcp->batch;
709 	else
710 		to_drain = pcp->count;
711 	free_pages_bulk(zone, to_drain, &pcp->list, 0);
712 	pcp->count -= to_drain;
713 	local_irq_restore(flags);
714 }
715 #endif
716 
717 static void __drain_pages(unsigned int cpu)
718 {
719 	unsigned long flags;
720 	struct zone *zone;
721 	int i;
722 
723 	for_each_zone(zone) {
724 		struct per_cpu_pageset *pset;
725 
726 		if (!populated_zone(zone))
727 			continue;
728 
729 		pset = zone_pcp(zone, cpu);
730 		for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
731 			struct per_cpu_pages *pcp;
732 
733 			pcp = &pset->pcp[i];
734 			local_irq_save(flags);
735 			free_pages_bulk(zone, pcp->count, &pcp->list, 0);
736 			pcp->count = 0;
737 			local_irq_restore(flags);
738 		}
739 	}
740 }
741 
742 #ifdef CONFIG_PM
743 
744 void mark_free_pages(struct zone *zone)
745 {
746 	unsigned long pfn, max_zone_pfn;
747 	unsigned long flags;
748 	int order;
749 	struct list_head *curr;
750 
751 	if (!zone->spanned_pages)
752 		return;
753 
754 	spin_lock_irqsave(&zone->lock, flags);
755 
756 	max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
757 	for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
758 		if (pfn_valid(pfn)) {
759 			struct page *page = pfn_to_page(pfn);
760 
761 			if (!swsusp_page_is_forbidden(page))
762 				swsusp_unset_page_free(page);
763 		}
764 
765 	for (order = MAX_ORDER - 1; order >= 0; --order)
766 		list_for_each(curr, &zone->free_area[order].free_list) {
767 			unsigned long i;
768 
769 			pfn = page_to_pfn(list_entry(curr, struct page, lru));
770 			for (i = 0; i < (1UL << order); i++)
771 				swsusp_set_page_free(pfn_to_page(pfn + i));
772 		}
773 
774 	spin_unlock_irqrestore(&zone->lock, flags);
775 }
776 
777 /*
778  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
779  */
780 void drain_local_pages(void)
781 {
782 	unsigned long flags;
783 
784 	local_irq_save(flags);
785 	__drain_pages(smp_processor_id());
786 	local_irq_restore(flags);
787 }
788 #endif /* CONFIG_PM */
789 
790 /*
791  * Free a 0-order page
792  */
793 static void fastcall free_hot_cold_page(struct page *page, int cold)
794 {
795 	struct zone *zone = page_zone(page);
796 	struct per_cpu_pages *pcp;
797 	unsigned long flags;
798 
799 	if (PageAnon(page))
800 		page->mapping = NULL;
801 	if (free_pages_check(page))
802 		return;
803 
804 	if (!PageHighMem(page))
805 		debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
806 	arch_free_page(page, 0);
807 	kernel_map_pages(page, 1, 0);
808 
809 	pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
810 	local_irq_save(flags);
811 	__count_vm_event(PGFREE);
812 	list_add(&page->lru, &pcp->list);
813 	pcp->count++;
814 	if (pcp->count >= pcp->high) {
815 		free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
816 		pcp->count -= pcp->batch;
817 	}
818 	local_irq_restore(flags);
819 	put_cpu();
820 }
821 
822 void fastcall free_hot_page(struct page *page)
823 {
824 	free_hot_cold_page(page, 0);
825 }
826 
827 void fastcall free_cold_page(struct page *page)
828 {
829 	free_hot_cold_page(page, 1);
830 }
831 
832 /*
833  * split_page takes a non-compound higher-order page, and splits it into
834  * n (1<<order) sub-pages: page[0..n]
835  * Each sub-page must be freed individually.
836  *
837  * Note: this is probably too low level an operation for use in drivers.
838  * Please consult with lkml before using this in your driver.
839  */
840 void split_page(struct page *page, unsigned int order)
841 {
842 	int i;
843 
844 	VM_BUG_ON(PageCompound(page));
845 	VM_BUG_ON(!page_count(page));
846 	for (i = 1; i < (1 << order); i++)
847 		set_page_refcounted(page + i);
848 }
849 
850 /*
851  * Really, prep_compound_page() should be called from __rmqueue_bulk().  But
852  * we cheat by calling it from here, in the order > 0 path.  Saves a branch
853  * or two.
854  */
855 static struct page *buffered_rmqueue(struct zonelist *zonelist,
856 			struct zone *zone, int order, gfp_t gfp_flags)
857 {
858 	unsigned long flags;
859 	struct page *page;
860 	int cold = !!(gfp_flags & __GFP_COLD);
861 	int cpu;
862 
863 again:
864 	cpu  = get_cpu();
865 	if (likely(order == 0)) {
866 		struct per_cpu_pages *pcp;
867 
868 		pcp = &zone_pcp(zone, cpu)->pcp[cold];
869 		local_irq_save(flags);
870 		if (!pcp->count) {
871 			pcp->count = rmqueue_bulk(zone, 0,
872 						pcp->batch, &pcp->list);
873 			if (unlikely(!pcp->count))
874 				goto failed;
875 		}
876 		page = list_entry(pcp->list.next, struct page, lru);
877 		list_del(&page->lru);
878 		pcp->count--;
879 	} else {
880 		spin_lock_irqsave(&zone->lock, flags);
881 		page = __rmqueue(zone, order);
882 		spin_unlock(&zone->lock);
883 		if (!page)
884 			goto failed;
885 	}
886 
887 	__count_zone_vm_events(PGALLOC, zone, 1 << order);
888 	zone_statistics(zonelist, zone);
889 	local_irq_restore(flags);
890 	put_cpu();
891 
892 	VM_BUG_ON(bad_range(zone, page));
893 	if (prep_new_page(page, order, gfp_flags))
894 		goto again;
895 	return page;
896 
897 failed:
898 	local_irq_restore(flags);
899 	put_cpu();
900 	return NULL;
901 }
902 
903 #define ALLOC_NO_WATERMARKS	0x01 /* don't check watermarks at all */
904 #define ALLOC_WMARK_MIN		0x02 /* use pages_min watermark */
905 #define ALLOC_WMARK_LOW		0x04 /* use pages_low watermark */
906 #define ALLOC_WMARK_HIGH	0x08 /* use pages_high watermark */
907 #define ALLOC_HARDER		0x10 /* try to alloc harder */
908 #define ALLOC_HIGH		0x20 /* __GFP_HIGH set */
909 #define ALLOC_CPUSET		0x40 /* check for correct cpuset */
910 
911 #ifdef CONFIG_FAIL_PAGE_ALLOC
912 
913 static struct fail_page_alloc_attr {
914 	struct fault_attr attr;
915 
916 	u32 ignore_gfp_highmem;
917 	u32 ignore_gfp_wait;
918 
919 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
920 
921 	struct dentry *ignore_gfp_highmem_file;
922 	struct dentry *ignore_gfp_wait_file;
923 
924 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
925 
926 } fail_page_alloc = {
927 	.attr = FAULT_ATTR_INITIALIZER,
928 	.ignore_gfp_wait = 1,
929 	.ignore_gfp_highmem = 1,
930 };
931 
932 static int __init setup_fail_page_alloc(char *str)
933 {
934 	return setup_fault_attr(&fail_page_alloc.attr, str);
935 }
936 __setup("fail_page_alloc=", setup_fail_page_alloc);
937 
938 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
939 {
940 	if (gfp_mask & __GFP_NOFAIL)
941 		return 0;
942 	if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
943 		return 0;
944 	if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
945 		return 0;
946 
947 	return should_fail(&fail_page_alloc.attr, 1 << order);
948 }
949 
950 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
951 
952 static int __init fail_page_alloc_debugfs(void)
953 {
954 	mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
955 	struct dentry *dir;
956 	int err;
957 
958 	err = init_fault_attr_dentries(&fail_page_alloc.attr,
959 				       "fail_page_alloc");
960 	if (err)
961 		return err;
962 	dir = fail_page_alloc.attr.dentries.dir;
963 
964 	fail_page_alloc.ignore_gfp_wait_file =
965 		debugfs_create_bool("ignore-gfp-wait", mode, dir,
966 				      &fail_page_alloc.ignore_gfp_wait);
967 
968 	fail_page_alloc.ignore_gfp_highmem_file =
969 		debugfs_create_bool("ignore-gfp-highmem", mode, dir,
970 				      &fail_page_alloc.ignore_gfp_highmem);
971 
972 	if (!fail_page_alloc.ignore_gfp_wait_file ||
973 			!fail_page_alloc.ignore_gfp_highmem_file) {
974 		err = -ENOMEM;
975 		debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
976 		debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
977 		cleanup_fault_attr_dentries(&fail_page_alloc.attr);
978 	}
979 
980 	return err;
981 }
982 
983 late_initcall(fail_page_alloc_debugfs);
984 
985 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
986 
987 #else /* CONFIG_FAIL_PAGE_ALLOC */
988 
989 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
990 {
991 	return 0;
992 }
993 
994 #endif /* CONFIG_FAIL_PAGE_ALLOC */
995 
996 /*
997  * Return 1 if free pages are above 'mark'. This takes into account the order
998  * of the allocation.
999  */
1000 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1001 		      int classzone_idx, int alloc_flags)
1002 {
1003 	/* free_pages my go negative - that's OK */
1004 	long min = mark;
1005 	long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1006 	int o;
1007 
1008 	if (alloc_flags & ALLOC_HIGH)
1009 		min -= min / 2;
1010 	if (alloc_flags & ALLOC_HARDER)
1011 		min -= min / 4;
1012 
1013 	if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1014 		return 0;
1015 	for (o = 0; o < order; o++) {
1016 		/* At the next order, this order's pages become unavailable */
1017 		free_pages -= z->free_area[o].nr_free << o;
1018 
1019 		/* Require fewer higher order pages to be free */
1020 		min >>= 1;
1021 
1022 		if (free_pages <= min)
1023 			return 0;
1024 	}
1025 	return 1;
1026 }
1027 
1028 #ifdef CONFIG_NUMA
1029 /*
1030  * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
1031  * skip over zones that are not allowed by the cpuset, or that have
1032  * been recently (in last second) found to be nearly full.  See further
1033  * comments in mmzone.h.  Reduces cache footprint of zonelist scans
1034  * that have to skip over alot of full or unallowed zones.
1035  *
1036  * If the zonelist cache is present in the passed in zonelist, then
1037  * returns a pointer to the allowed node mask (either the current
1038  * tasks mems_allowed, or node_online_map.)
1039  *
1040  * If the zonelist cache is not available for this zonelist, does
1041  * nothing and returns NULL.
1042  *
1043  * If the fullzones BITMAP in the zonelist cache is stale (more than
1044  * a second since last zap'd) then we zap it out (clear its bits.)
1045  *
1046  * We hold off even calling zlc_setup, until after we've checked the
1047  * first zone in the zonelist, on the theory that most allocations will
1048  * be satisfied from that first zone, so best to examine that zone as
1049  * quickly as we can.
1050  */
1051 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1052 {
1053 	struct zonelist_cache *zlc;	/* cached zonelist speedup info */
1054 	nodemask_t *allowednodes;	/* zonelist_cache approximation */
1055 
1056 	zlc = zonelist->zlcache_ptr;
1057 	if (!zlc)
1058 		return NULL;
1059 
1060 	if (jiffies - zlc->last_full_zap > 1 * HZ) {
1061 		bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1062 		zlc->last_full_zap = jiffies;
1063 	}
1064 
1065 	allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1066 					&cpuset_current_mems_allowed :
1067 					&node_online_map;
1068 	return allowednodes;
1069 }
1070 
1071 /*
1072  * Given 'z' scanning a zonelist, run a couple of quick checks to see
1073  * if it is worth looking at further for free memory:
1074  *  1) Check that the zone isn't thought to be full (doesn't have its
1075  *     bit set in the zonelist_cache fullzones BITMAP).
1076  *  2) Check that the zones node (obtained from the zonelist_cache
1077  *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1078  * Return true (non-zero) if zone is worth looking at further, or
1079  * else return false (zero) if it is not.
1080  *
1081  * This check -ignores- the distinction between various watermarks,
1082  * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
1083  * found to be full for any variation of these watermarks, it will
1084  * be considered full for up to one second by all requests, unless
1085  * we are so low on memory on all allowed nodes that we are forced
1086  * into the second scan of the zonelist.
1087  *
1088  * In the second scan we ignore this zonelist cache and exactly
1089  * apply the watermarks to all zones, even it is slower to do so.
1090  * We are low on memory in the second scan, and should leave no stone
1091  * unturned looking for a free page.
1092  */
1093 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1094 						nodemask_t *allowednodes)
1095 {
1096 	struct zonelist_cache *zlc;	/* cached zonelist speedup info */
1097 	int i;				/* index of *z in zonelist zones */
1098 	int n;				/* node that zone *z is on */
1099 
1100 	zlc = zonelist->zlcache_ptr;
1101 	if (!zlc)
1102 		return 1;
1103 
1104 	i = z - zonelist->zones;
1105 	n = zlc->z_to_n[i];
1106 
1107 	/* This zone is worth trying if it is allowed but not full */
1108 	return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1109 }
1110 
1111 /*
1112  * Given 'z' scanning a zonelist, set the corresponding bit in
1113  * zlc->fullzones, so that subsequent attempts to allocate a page
1114  * from that zone don't waste time re-examining it.
1115  */
1116 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1117 {
1118 	struct zonelist_cache *zlc;	/* cached zonelist speedup info */
1119 	int i;				/* index of *z in zonelist zones */
1120 
1121 	zlc = zonelist->zlcache_ptr;
1122 	if (!zlc)
1123 		return;
1124 
1125 	i = z - zonelist->zones;
1126 
1127 	set_bit(i, zlc->fullzones);
1128 }
1129 
1130 #else	/* CONFIG_NUMA */
1131 
1132 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1133 {
1134 	return NULL;
1135 }
1136 
1137 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1138 				nodemask_t *allowednodes)
1139 {
1140 	return 1;
1141 }
1142 
1143 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1144 {
1145 }
1146 #endif	/* CONFIG_NUMA */
1147 
1148 /*
1149  * get_page_from_freelist goes through the zonelist trying to allocate
1150  * a page.
1151  */
1152 static struct page *
1153 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1154 		struct zonelist *zonelist, int alloc_flags)
1155 {
1156 	struct zone **z;
1157 	struct page *page = NULL;
1158 	int classzone_idx = zone_idx(zonelist->zones[0]);
1159 	struct zone *zone;
1160 	nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1161 	int zlc_active = 0;		/* set if using zonelist_cache */
1162 	int did_zlc_setup = 0;		/* just call zlc_setup() one time */
1163 
1164 zonelist_scan:
1165 	/*
1166 	 * Scan zonelist, looking for a zone with enough free.
1167 	 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1168 	 */
1169 	z = zonelist->zones;
1170 
1171 	do {
1172 		if (NUMA_BUILD && zlc_active &&
1173 			!zlc_zone_worth_trying(zonelist, z, allowednodes))
1174 				continue;
1175 		zone = *z;
1176 		if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) &&
1177 			zone->zone_pgdat != zonelist->zones[0]->zone_pgdat))
1178 				break;
1179 		if ((alloc_flags & ALLOC_CPUSET) &&
1180 			!cpuset_zone_allowed_softwall(zone, gfp_mask))
1181 				goto try_next_zone;
1182 
1183 		if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1184 			unsigned long mark;
1185 			if (alloc_flags & ALLOC_WMARK_MIN)
1186 				mark = zone->pages_min;
1187 			else if (alloc_flags & ALLOC_WMARK_LOW)
1188 				mark = zone->pages_low;
1189 			else
1190 				mark = zone->pages_high;
1191 			if (!zone_watermark_ok(zone, order, mark,
1192 				    classzone_idx, alloc_flags)) {
1193 				if (!zone_reclaim_mode ||
1194 				    !zone_reclaim(zone, gfp_mask, order))
1195 					goto this_zone_full;
1196 			}
1197 		}
1198 
1199 		page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1200 		if (page)
1201 			break;
1202 this_zone_full:
1203 		if (NUMA_BUILD)
1204 			zlc_mark_zone_full(zonelist, z);
1205 try_next_zone:
1206 		if (NUMA_BUILD && !did_zlc_setup) {
1207 			/* we do zlc_setup after the first zone is tried */
1208 			allowednodes = zlc_setup(zonelist, alloc_flags);
1209 			zlc_active = 1;
1210 			did_zlc_setup = 1;
1211 		}
1212 	} while (*(++z) != NULL);
1213 
1214 	if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1215 		/* Disable zlc cache for second zonelist scan */
1216 		zlc_active = 0;
1217 		goto zonelist_scan;
1218 	}
1219 	return page;
1220 }
1221 
1222 /*
1223  * This is the 'heart' of the zoned buddy allocator.
1224  */
1225 struct page * fastcall
1226 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1227 		struct zonelist *zonelist)
1228 {
1229 	const gfp_t wait = gfp_mask & __GFP_WAIT;
1230 	struct zone **z;
1231 	struct page *page;
1232 	struct reclaim_state reclaim_state;
1233 	struct task_struct *p = current;
1234 	int do_retry;
1235 	int alloc_flags;
1236 	int did_some_progress;
1237 
1238 	might_sleep_if(wait);
1239 
1240 	if (should_fail_alloc_page(gfp_mask, order))
1241 		return NULL;
1242 
1243 restart:
1244 	z = zonelist->zones;  /* the list of zones suitable for gfp_mask */
1245 
1246 	if (unlikely(*z == NULL)) {
1247 		/* Should this ever happen?? */
1248 		return NULL;
1249 	}
1250 
1251 	page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1252 				zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1253 	if (page)
1254 		goto got_pg;
1255 
1256 	/*
1257 	 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1258 	 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1259 	 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1260 	 * using a larger set of nodes after it has established that the
1261 	 * allowed per node queues are empty and that nodes are
1262 	 * over allocated.
1263 	 */
1264 	if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1265 		goto nopage;
1266 
1267 	for (z = zonelist->zones; *z; z++)
1268 		wakeup_kswapd(*z, order);
1269 
1270 	/*
1271 	 * OK, we're below the kswapd watermark and have kicked background
1272 	 * reclaim. Now things get more complex, so set up alloc_flags according
1273 	 * to how we want to proceed.
1274 	 *
1275 	 * The caller may dip into page reserves a bit more if the caller
1276 	 * cannot run direct reclaim, or if the caller has realtime scheduling
1277 	 * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
1278 	 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1279 	 */
1280 	alloc_flags = ALLOC_WMARK_MIN;
1281 	if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1282 		alloc_flags |= ALLOC_HARDER;
1283 	if (gfp_mask & __GFP_HIGH)
1284 		alloc_flags |= ALLOC_HIGH;
1285 	if (wait)
1286 		alloc_flags |= ALLOC_CPUSET;
1287 
1288 	/*
1289 	 * Go through the zonelist again. Let __GFP_HIGH and allocations
1290 	 * coming from realtime tasks go deeper into reserves.
1291 	 *
1292 	 * This is the last chance, in general, before the goto nopage.
1293 	 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1294 	 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1295 	 */
1296 	page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1297 	if (page)
1298 		goto got_pg;
1299 
1300 	/* This allocation should allow future memory freeing. */
1301 
1302 rebalance:
1303 	if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1304 			&& !in_interrupt()) {
1305 		if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1306 nofail_alloc:
1307 			/* go through the zonelist yet again, ignoring mins */
1308 			page = get_page_from_freelist(gfp_mask, order,
1309 				zonelist, ALLOC_NO_WATERMARKS);
1310 			if (page)
1311 				goto got_pg;
1312 			if (gfp_mask & __GFP_NOFAIL) {
1313 				congestion_wait(WRITE, HZ/50);
1314 				goto nofail_alloc;
1315 			}
1316 		}
1317 		goto nopage;
1318 	}
1319 
1320 	/* Atomic allocations - we can't balance anything */
1321 	if (!wait)
1322 		goto nopage;
1323 
1324 	cond_resched();
1325 
1326 	/* We now go into synchronous reclaim */
1327 	cpuset_memory_pressure_bump();
1328 	p->flags |= PF_MEMALLOC;
1329 	reclaim_state.reclaimed_slab = 0;
1330 	p->reclaim_state = &reclaim_state;
1331 
1332 	did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
1333 
1334 	p->reclaim_state = NULL;
1335 	p->flags &= ~PF_MEMALLOC;
1336 
1337 	cond_resched();
1338 
1339 	if (likely(did_some_progress)) {
1340 		page = get_page_from_freelist(gfp_mask, order,
1341 						zonelist, alloc_flags);
1342 		if (page)
1343 			goto got_pg;
1344 	} else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1345 		/*
1346 		 * Go through the zonelist yet one more time, keep
1347 		 * very high watermark here, this is only to catch
1348 		 * a parallel oom killing, we must fail if we're still
1349 		 * under heavy pressure.
1350 		 */
1351 		page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1352 				zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1353 		if (page)
1354 			goto got_pg;
1355 
1356 		out_of_memory(zonelist, gfp_mask, order);
1357 		goto restart;
1358 	}
1359 
1360 	/*
1361 	 * Don't let big-order allocations loop unless the caller explicitly
1362 	 * requests that.  Wait for some write requests to complete then retry.
1363 	 *
1364 	 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1365 	 * <= 3, but that may not be true in other implementations.
1366 	 */
1367 	do_retry = 0;
1368 	if (!(gfp_mask & __GFP_NORETRY)) {
1369 		if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1370 			do_retry = 1;
1371 		if (gfp_mask & __GFP_NOFAIL)
1372 			do_retry = 1;
1373 	}
1374 	if (do_retry) {
1375 		congestion_wait(WRITE, HZ/50);
1376 		goto rebalance;
1377 	}
1378 
1379 nopage:
1380 	if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1381 		printk(KERN_WARNING "%s: page allocation failure."
1382 			" order:%d, mode:0x%x\n",
1383 			p->comm, order, gfp_mask);
1384 		dump_stack();
1385 		show_mem();
1386 	}
1387 got_pg:
1388 	return page;
1389 }
1390 
1391 EXPORT_SYMBOL(__alloc_pages);
1392 
1393 /*
1394  * Common helper functions.
1395  */
1396 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1397 {
1398 	struct page * page;
1399 	page = alloc_pages(gfp_mask, order);
1400 	if (!page)
1401 		return 0;
1402 	return (unsigned long) page_address(page);
1403 }
1404 
1405 EXPORT_SYMBOL(__get_free_pages);
1406 
1407 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1408 {
1409 	struct page * page;
1410 
1411 	/*
1412 	 * get_zeroed_page() returns a 32-bit address, which cannot represent
1413 	 * a highmem page
1414 	 */
1415 	VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1416 
1417 	page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1418 	if (page)
1419 		return (unsigned long) page_address(page);
1420 	return 0;
1421 }
1422 
1423 EXPORT_SYMBOL(get_zeroed_page);
1424 
1425 void __pagevec_free(struct pagevec *pvec)
1426 {
1427 	int i = pagevec_count(pvec);
1428 
1429 	while (--i >= 0)
1430 		free_hot_cold_page(pvec->pages[i], pvec->cold);
1431 }
1432 
1433 fastcall void __free_pages(struct page *page, unsigned int order)
1434 {
1435 	if (put_page_testzero(page)) {
1436 		if (order == 0)
1437 			free_hot_page(page);
1438 		else
1439 			__free_pages_ok(page, order);
1440 	}
1441 }
1442 
1443 EXPORT_SYMBOL(__free_pages);
1444 
1445 fastcall void free_pages(unsigned long addr, unsigned int order)
1446 {
1447 	if (addr != 0) {
1448 		VM_BUG_ON(!virt_addr_valid((void *)addr));
1449 		__free_pages(virt_to_page((void *)addr), order);
1450 	}
1451 }
1452 
1453 EXPORT_SYMBOL(free_pages);
1454 
1455 static unsigned int nr_free_zone_pages(int offset)
1456 {
1457 	/* Just pick one node, since fallback list is circular */
1458 	pg_data_t *pgdat = NODE_DATA(numa_node_id());
1459 	unsigned int sum = 0;
1460 
1461 	struct zonelist *zonelist = pgdat->node_zonelists + offset;
1462 	struct zone **zonep = zonelist->zones;
1463 	struct zone *zone;
1464 
1465 	for (zone = *zonep++; zone; zone = *zonep++) {
1466 		unsigned long size = zone->present_pages;
1467 		unsigned long high = zone->pages_high;
1468 		if (size > high)
1469 			sum += size - high;
1470 	}
1471 
1472 	return sum;
1473 }
1474 
1475 /*
1476  * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1477  */
1478 unsigned int nr_free_buffer_pages(void)
1479 {
1480 	return nr_free_zone_pages(gfp_zone(GFP_USER));
1481 }
1482 
1483 /*
1484  * Amount of free RAM allocatable within all zones
1485  */
1486 unsigned int nr_free_pagecache_pages(void)
1487 {
1488 	return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
1489 }
1490 
1491 static inline void show_node(struct zone *zone)
1492 {
1493 	if (NUMA_BUILD)
1494 		printk("Node %d ", zone_to_nid(zone));
1495 }
1496 
1497 void si_meminfo(struct sysinfo *val)
1498 {
1499 	val->totalram = totalram_pages;
1500 	val->sharedram = 0;
1501 	val->freeram = global_page_state(NR_FREE_PAGES);
1502 	val->bufferram = nr_blockdev_pages();
1503 	val->totalhigh = totalhigh_pages;
1504 	val->freehigh = nr_free_highpages();
1505 	val->mem_unit = PAGE_SIZE;
1506 }
1507 
1508 EXPORT_SYMBOL(si_meminfo);
1509 
1510 #ifdef CONFIG_NUMA
1511 void si_meminfo_node(struct sysinfo *val, int nid)
1512 {
1513 	pg_data_t *pgdat = NODE_DATA(nid);
1514 
1515 	val->totalram = pgdat->node_present_pages;
1516 	val->freeram = node_page_state(nid, NR_FREE_PAGES);
1517 #ifdef CONFIG_HIGHMEM
1518 	val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1519 	val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1520 			NR_FREE_PAGES);
1521 #else
1522 	val->totalhigh = 0;
1523 	val->freehigh = 0;
1524 #endif
1525 	val->mem_unit = PAGE_SIZE;
1526 }
1527 #endif
1528 
1529 #define K(x) ((x) << (PAGE_SHIFT-10))
1530 
1531 /*
1532  * Show free area list (used inside shift_scroll-lock stuff)
1533  * We also calculate the percentage fragmentation. We do this by counting the
1534  * memory on each free list with the exception of the first item on the list.
1535  */
1536 void show_free_areas(void)
1537 {
1538 	int cpu;
1539 	struct zone *zone;
1540 
1541 	for_each_zone(zone) {
1542 		if (!populated_zone(zone))
1543 			continue;
1544 
1545 		show_node(zone);
1546 		printk("%s per-cpu:\n", zone->name);
1547 
1548 		for_each_online_cpu(cpu) {
1549 			struct per_cpu_pageset *pageset;
1550 
1551 			pageset = zone_pcp(zone, cpu);
1552 
1553 			printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d   "
1554 			       "Cold: hi:%5d, btch:%4d usd:%4d\n",
1555 			       cpu, pageset->pcp[0].high,
1556 			       pageset->pcp[0].batch, pageset->pcp[0].count,
1557 			       pageset->pcp[1].high, pageset->pcp[1].batch,
1558 			       pageset->pcp[1].count);
1559 		}
1560 	}
1561 
1562 	printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1563 		" free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1564 		global_page_state(NR_ACTIVE),
1565 		global_page_state(NR_INACTIVE),
1566 		global_page_state(NR_FILE_DIRTY),
1567 		global_page_state(NR_WRITEBACK),
1568 		global_page_state(NR_UNSTABLE_NFS),
1569 		global_page_state(NR_FREE_PAGES),
1570 		global_page_state(NR_SLAB_RECLAIMABLE) +
1571 			global_page_state(NR_SLAB_UNRECLAIMABLE),
1572 		global_page_state(NR_FILE_MAPPED),
1573 		global_page_state(NR_PAGETABLE),
1574 		global_page_state(NR_BOUNCE));
1575 
1576 	for_each_zone(zone) {
1577 		int i;
1578 
1579 		if (!populated_zone(zone))
1580 			continue;
1581 
1582 		show_node(zone);
1583 		printk("%s"
1584 			" free:%lukB"
1585 			" min:%lukB"
1586 			" low:%lukB"
1587 			" high:%lukB"
1588 			" active:%lukB"
1589 			" inactive:%lukB"
1590 			" present:%lukB"
1591 			" pages_scanned:%lu"
1592 			" all_unreclaimable? %s"
1593 			"\n",
1594 			zone->name,
1595 			K(zone_page_state(zone, NR_FREE_PAGES)),
1596 			K(zone->pages_min),
1597 			K(zone->pages_low),
1598 			K(zone->pages_high),
1599 			K(zone_page_state(zone, NR_ACTIVE)),
1600 			K(zone_page_state(zone, NR_INACTIVE)),
1601 			K(zone->present_pages),
1602 			zone->pages_scanned,
1603 			(zone->all_unreclaimable ? "yes" : "no")
1604 			);
1605 		printk("lowmem_reserve[]:");
1606 		for (i = 0; i < MAX_NR_ZONES; i++)
1607 			printk(" %lu", zone->lowmem_reserve[i]);
1608 		printk("\n");
1609 	}
1610 
1611 	for_each_zone(zone) {
1612  		unsigned long nr[MAX_ORDER], flags, order, total = 0;
1613 
1614 		if (!populated_zone(zone))
1615 			continue;
1616 
1617 		show_node(zone);
1618 		printk("%s: ", zone->name);
1619 
1620 		spin_lock_irqsave(&zone->lock, flags);
1621 		for (order = 0; order < MAX_ORDER; order++) {
1622 			nr[order] = zone->free_area[order].nr_free;
1623 			total += nr[order] << order;
1624 		}
1625 		spin_unlock_irqrestore(&zone->lock, flags);
1626 		for (order = 0; order < MAX_ORDER; order++)
1627 			printk("%lu*%lukB ", nr[order], K(1UL) << order);
1628 		printk("= %lukB\n", K(total));
1629 	}
1630 
1631 	show_swap_cache_info();
1632 }
1633 
1634 /*
1635  * Builds allocation fallback zone lists.
1636  *
1637  * Add all populated zones of a node to the zonelist.
1638  */
1639 static int __meminit build_zonelists_node(pg_data_t *pgdat,
1640 			struct zonelist *zonelist, int nr_zones, enum zone_type zone_type)
1641 {
1642 	struct zone *zone;
1643 
1644 	BUG_ON(zone_type >= MAX_NR_ZONES);
1645 	zone_type++;
1646 
1647 	do {
1648 		zone_type--;
1649 		zone = pgdat->node_zones + zone_type;
1650 		if (populated_zone(zone)) {
1651 			zonelist->zones[nr_zones++] = zone;
1652 			check_highest_zone(zone_type);
1653 		}
1654 
1655 	} while (zone_type);
1656 	return nr_zones;
1657 }
1658 
1659 #ifdef CONFIG_NUMA
1660 #define MAX_NODE_LOAD (num_online_nodes())
1661 static int __meminitdata node_load[MAX_NUMNODES];
1662 /**
1663  * find_next_best_node - find the next node that should appear in a given node's fallback list
1664  * @node: node whose fallback list we're appending
1665  * @used_node_mask: nodemask_t of already used nodes
1666  *
1667  * We use a number of factors to determine which is the next node that should
1668  * appear on a given node's fallback list.  The node should not have appeared
1669  * already in @node's fallback list, and it should be the next closest node
1670  * according to the distance array (which contains arbitrary distance values
1671  * from each node to each node in the system), and should also prefer nodes
1672  * with no CPUs, since presumably they'll have very little allocation pressure
1673  * on them otherwise.
1674  * It returns -1 if no node is found.
1675  */
1676 static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask)
1677 {
1678 	int n, val;
1679 	int min_val = INT_MAX;
1680 	int best_node = -1;
1681 
1682 	/* Use the local node if we haven't already */
1683 	if (!node_isset(node, *used_node_mask)) {
1684 		node_set(node, *used_node_mask);
1685 		return node;
1686 	}
1687 
1688 	for_each_online_node(n) {
1689 		cpumask_t tmp;
1690 
1691 		/* Don't want a node to appear more than once */
1692 		if (node_isset(n, *used_node_mask))
1693 			continue;
1694 
1695 		/* Use the distance array to find the distance */
1696 		val = node_distance(node, n);
1697 
1698 		/* Penalize nodes under us ("prefer the next node") */
1699 		val += (n < node);
1700 
1701 		/* Give preference to headless and unused nodes */
1702 		tmp = node_to_cpumask(n);
1703 		if (!cpus_empty(tmp))
1704 			val += PENALTY_FOR_NODE_WITH_CPUS;
1705 
1706 		/* Slight preference for less loaded node */
1707 		val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1708 		val += node_load[n];
1709 
1710 		if (val < min_val) {
1711 			min_val = val;
1712 			best_node = n;
1713 		}
1714 	}
1715 
1716 	if (best_node >= 0)
1717 		node_set(best_node, *used_node_mask);
1718 
1719 	return best_node;
1720 }
1721 
1722 static void __meminit build_zonelists(pg_data_t *pgdat)
1723 {
1724 	int j, node, local_node;
1725 	enum zone_type i;
1726 	int prev_node, load;
1727 	struct zonelist *zonelist;
1728 	nodemask_t used_mask;
1729 
1730 	/* initialize zonelists */
1731 	for (i = 0; i < MAX_NR_ZONES; i++) {
1732 		zonelist = pgdat->node_zonelists + i;
1733 		zonelist->zones[0] = NULL;
1734 	}
1735 
1736 	/* NUMA-aware ordering of nodes */
1737 	local_node = pgdat->node_id;
1738 	load = num_online_nodes();
1739 	prev_node = local_node;
1740 	nodes_clear(used_mask);
1741 	while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1742 		int distance = node_distance(local_node, node);
1743 
1744 		/*
1745 		 * If another node is sufficiently far away then it is better
1746 		 * to reclaim pages in a zone before going off node.
1747 		 */
1748 		if (distance > RECLAIM_DISTANCE)
1749 			zone_reclaim_mode = 1;
1750 
1751 		/*
1752 		 * We don't want to pressure a particular node.
1753 		 * So adding penalty to the first node in same
1754 		 * distance group to make it round-robin.
1755 		 */
1756 
1757 		if (distance != node_distance(local_node, prev_node))
1758 			node_load[node] += load;
1759 		prev_node = node;
1760 		load--;
1761 		for (i = 0; i < MAX_NR_ZONES; i++) {
1762 			zonelist = pgdat->node_zonelists + i;
1763 			for (j = 0; zonelist->zones[j] != NULL; j++);
1764 
1765 	 		j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1766 			zonelist->zones[j] = NULL;
1767 		}
1768 	}
1769 }
1770 
1771 /* Construct the zonelist performance cache - see further mmzone.h */
1772 static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1773 {
1774 	int i;
1775 
1776 	for (i = 0; i < MAX_NR_ZONES; i++) {
1777 		struct zonelist *zonelist;
1778 		struct zonelist_cache *zlc;
1779 		struct zone **z;
1780 
1781 		zonelist = pgdat->node_zonelists + i;
1782 		zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
1783 		bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1784 		for (z = zonelist->zones; *z; z++)
1785 			zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
1786 	}
1787 }
1788 
1789 #else	/* CONFIG_NUMA */
1790 
1791 static void __meminit build_zonelists(pg_data_t *pgdat)
1792 {
1793 	int node, local_node;
1794 	enum zone_type i,j;
1795 
1796 	local_node = pgdat->node_id;
1797 	for (i = 0; i < MAX_NR_ZONES; i++) {
1798 		struct zonelist *zonelist;
1799 
1800 		zonelist = pgdat->node_zonelists + i;
1801 
1802  		j = build_zonelists_node(pgdat, zonelist, 0, i);
1803  		/*
1804  		 * Now we build the zonelist so that it contains the zones
1805  		 * of all the other nodes.
1806  		 * We don't want to pressure a particular node, so when
1807  		 * building the zones for node N, we make sure that the
1808  		 * zones coming right after the local ones are those from
1809  		 * node N+1 (modulo N)
1810  		 */
1811 		for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1812 			if (!node_online(node))
1813 				continue;
1814 			j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1815 		}
1816 		for (node = 0; node < local_node; node++) {
1817 			if (!node_online(node))
1818 				continue;
1819 			j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
1820 		}
1821 
1822 		zonelist->zones[j] = NULL;
1823 	}
1824 }
1825 
1826 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
1827 static void __meminit build_zonelist_cache(pg_data_t *pgdat)
1828 {
1829 	int i;
1830 
1831 	for (i = 0; i < MAX_NR_ZONES; i++)
1832 		pgdat->node_zonelists[i].zlcache_ptr = NULL;
1833 }
1834 
1835 #endif	/* CONFIG_NUMA */
1836 
1837 /* return values int ....just for stop_machine_run() */
1838 static int __meminit __build_all_zonelists(void *dummy)
1839 {
1840 	int nid;
1841 
1842 	for_each_online_node(nid) {
1843 		build_zonelists(NODE_DATA(nid));
1844 		build_zonelist_cache(NODE_DATA(nid));
1845 	}
1846 	return 0;
1847 }
1848 
1849 void __meminit build_all_zonelists(void)
1850 {
1851 	if (system_state == SYSTEM_BOOTING) {
1852 		__build_all_zonelists(NULL);
1853 		cpuset_init_current_mems_allowed();
1854 	} else {
1855 		/* we have to stop all cpus to guaranntee there is no user
1856 		   of zonelist */
1857 		stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
1858 		/* cpuset refresh routine should be here */
1859 	}
1860 	vm_total_pages = nr_free_pagecache_pages();
1861 	printk("Built %i zonelists.  Total pages: %ld\n",
1862 			num_online_nodes(), vm_total_pages);
1863 }
1864 
1865 /*
1866  * Helper functions to size the waitqueue hash table.
1867  * Essentially these want to choose hash table sizes sufficiently
1868  * large so that collisions trying to wait on pages are rare.
1869  * But in fact, the number of active page waitqueues on typical
1870  * systems is ridiculously low, less than 200. So this is even
1871  * conservative, even though it seems large.
1872  *
1873  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1874  * waitqueues, i.e. the size of the waitq table given the number of pages.
1875  */
1876 #define PAGES_PER_WAITQUEUE	256
1877 
1878 #ifndef CONFIG_MEMORY_HOTPLUG
1879 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1880 {
1881 	unsigned long size = 1;
1882 
1883 	pages /= PAGES_PER_WAITQUEUE;
1884 
1885 	while (size < pages)
1886 		size <<= 1;
1887 
1888 	/*
1889 	 * Once we have dozens or even hundreds of threads sleeping
1890 	 * on IO we've got bigger problems than wait queue collision.
1891 	 * Limit the size of the wait table to a reasonable size.
1892 	 */
1893 	size = min(size, 4096UL);
1894 
1895 	return max(size, 4UL);
1896 }
1897 #else
1898 /*
1899  * A zone's size might be changed by hot-add, so it is not possible to determine
1900  * a suitable size for its wait_table.  So we use the maximum size now.
1901  *
1902  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
1903  *
1904  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
1905  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1906  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
1907  *
1908  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1909  * or more by the traditional way. (See above).  It equals:
1910  *
1911  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
1912  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
1913  *    powerpc (64K page size)             : =  (32G +16M)byte.
1914  */
1915 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
1916 {
1917 	return 4096UL;
1918 }
1919 #endif
1920 
1921 /*
1922  * This is an integer logarithm so that shifts can be used later
1923  * to extract the more random high bits from the multiplicative
1924  * hash function before the remainder is taken.
1925  */
1926 static inline unsigned long wait_table_bits(unsigned long size)
1927 {
1928 	return ffz(~size);
1929 }
1930 
1931 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1932 
1933 /*
1934  * Initially all pages are reserved - free ones are freed
1935  * up by free_all_bootmem() once the early boot process is
1936  * done. Non-atomic initialization, single-pass.
1937  */
1938 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1939 		unsigned long start_pfn, enum memmap_context context)
1940 {
1941 	struct page *page;
1942 	unsigned long end_pfn = start_pfn + size;
1943 	unsigned long pfn;
1944 
1945 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1946 		/*
1947 		 * There can be holes in boot-time mem_map[]s
1948 		 * handed to this function.  They do not
1949 		 * exist on hotplugged memory.
1950 		 */
1951 		if (context == MEMMAP_EARLY) {
1952 			if (!early_pfn_valid(pfn))
1953 				continue;
1954 			if (!early_pfn_in_nid(pfn, nid))
1955 				continue;
1956 		}
1957 		page = pfn_to_page(pfn);
1958 		set_page_links(page, zone, nid, pfn);
1959 		init_page_count(page);
1960 		reset_page_mapcount(page);
1961 		SetPageReserved(page);
1962 		INIT_LIST_HEAD(&page->lru);
1963 #ifdef WANT_PAGE_VIRTUAL
1964 		/* The shift won't overflow because ZONE_NORMAL is below 4G. */
1965 		if (!is_highmem_idx(zone))
1966 			set_page_address(page, __va(pfn << PAGE_SHIFT));
1967 #endif
1968 	}
1969 }
1970 
1971 void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1972 				unsigned long size)
1973 {
1974 	int order;
1975 	for (order = 0; order < MAX_ORDER ; order++) {
1976 		INIT_LIST_HEAD(&zone->free_area[order].free_list);
1977 		zone->free_area[order].nr_free = 0;
1978 	}
1979 }
1980 
1981 #ifndef __HAVE_ARCH_MEMMAP_INIT
1982 #define memmap_init(size, nid, zone, start_pfn) \
1983 	memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
1984 #endif
1985 
1986 static int __cpuinit zone_batchsize(struct zone *zone)
1987 {
1988 	int batch;
1989 
1990 	/*
1991 	 * The per-cpu-pages pools are set to around 1000th of the
1992 	 * size of the zone.  But no more than 1/2 of a meg.
1993 	 *
1994 	 * OK, so we don't know how big the cache is.  So guess.
1995 	 */
1996 	batch = zone->present_pages / 1024;
1997 	if (batch * PAGE_SIZE > 512 * 1024)
1998 		batch = (512 * 1024) / PAGE_SIZE;
1999 	batch /= 4;		/* We effectively *= 4 below */
2000 	if (batch < 1)
2001 		batch = 1;
2002 
2003 	/*
2004 	 * Clamp the batch to a 2^n - 1 value. Having a power
2005 	 * of 2 value was found to be more likely to have
2006 	 * suboptimal cache aliasing properties in some cases.
2007 	 *
2008 	 * For example if 2 tasks are alternately allocating
2009 	 * batches of pages, one task can end up with a lot
2010 	 * of pages of one half of the possible page colors
2011 	 * and the other with pages of the other colors.
2012 	 */
2013 	batch = (1 << (fls(batch + batch/2)-1)) - 1;
2014 
2015 	return batch;
2016 }
2017 
2018 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2019 {
2020 	struct per_cpu_pages *pcp;
2021 
2022 	memset(p, 0, sizeof(*p));
2023 
2024 	pcp = &p->pcp[0];		/* hot */
2025 	pcp->count = 0;
2026 	pcp->high = 6 * batch;
2027 	pcp->batch = max(1UL, 1 * batch);
2028 	INIT_LIST_HEAD(&pcp->list);
2029 
2030 	pcp = &p->pcp[1];		/* cold*/
2031 	pcp->count = 0;
2032 	pcp->high = 2 * batch;
2033 	pcp->batch = max(1UL, batch/2);
2034 	INIT_LIST_HEAD(&pcp->list);
2035 }
2036 
2037 /*
2038  * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2039  * to the value high for the pageset p.
2040  */
2041 
2042 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2043 				unsigned long high)
2044 {
2045 	struct per_cpu_pages *pcp;
2046 
2047 	pcp = &p->pcp[0]; /* hot list */
2048 	pcp->high = high;
2049 	pcp->batch = max(1UL, high/4);
2050 	if ((high/4) > (PAGE_SHIFT * 8))
2051 		pcp->batch = PAGE_SHIFT * 8;
2052 }
2053 
2054 
2055 #ifdef CONFIG_NUMA
2056 /*
2057  * Boot pageset table. One per cpu which is going to be used for all
2058  * zones and all nodes. The parameters will be set in such a way
2059  * that an item put on a list will immediately be handed over to
2060  * the buddy list. This is safe since pageset manipulation is done
2061  * with interrupts disabled.
2062  *
2063  * Some NUMA counter updates may also be caught by the boot pagesets.
2064  *
2065  * The boot_pagesets must be kept even after bootup is complete for
2066  * unused processors and/or zones. They do play a role for bootstrapping
2067  * hotplugged processors.
2068  *
2069  * zoneinfo_show() and maybe other functions do
2070  * not check if the processor is online before following the pageset pointer.
2071  * Other parts of the kernel may not check if the zone is available.
2072  */
2073 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2074 
2075 /*
2076  * Dynamically allocate memory for the
2077  * per cpu pageset array in struct zone.
2078  */
2079 static int __cpuinit process_zones(int cpu)
2080 {
2081 	struct zone *zone, *dzone;
2082 
2083 	for_each_zone(zone) {
2084 
2085 		if (!populated_zone(zone))
2086 			continue;
2087 
2088 		zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2089 					 GFP_KERNEL, cpu_to_node(cpu));
2090 		if (!zone_pcp(zone, cpu))
2091 			goto bad;
2092 
2093 		setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2094 
2095 		if (percpu_pagelist_fraction)
2096 			setup_pagelist_highmark(zone_pcp(zone, cpu),
2097 			 	(zone->present_pages / percpu_pagelist_fraction));
2098 	}
2099 
2100 	return 0;
2101 bad:
2102 	for_each_zone(dzone) {
2103 		if (dzone == zone)
2104 			break;
2105 		kfree(zone_pcp(dzone, cpu));
2106 		zone_pcp(dzone, cpu) = NULL;
2107 	}
2108 	return -ENOMEM;
2109 }
2110 
2111 static inline void free_zone_pagesets(int cpu)
2112 {
2113 	struct zone *zone;
2114 
2115 	for_each_zone(zone) {
2116 		struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2117 
2118 		/* Free per_cpu_pageset if it is slab allocated */
2119 		if (pset != &boot_pageset[cpu])
2120 			kfree(pset);
2121 		zone_pcp(zone, cpu) = NULL;
2122 	}
2123 }
2124 
2125 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2126 		unsigned long action,
2127 		void *hcpu)
2128 {
2129 	int cpu = (long)hcpu;
2130 	int ret = NOTIFY_OK;
2131 
2132 	switch (action) {
2133 	case CPU_UP_PREPARE:
2134 	case CPU_UP_PREPARE_FROZEN:
2135 		if (process_zones(cpu))
2136 			ret = NOTIFY_BAD;
2137 		break;
2138 	case CPU_UP_CANCELED:
2139 	case CPU_UP_CANCELED_FROZEN:
2140 	case CPU_DEAD:
2141 	case CPU_DEAD_FROZEN:
2142 		free_zone_pagesets(cpu);
2143 		break;
2144 	default:
2145 		break;
2146 	}
2147 	return ret;
2148 }
2149 
2150 static struct notifier_block __cpuinitdata pageset_notifier =
2151 	{ &pageset_cpuup_callback, NULL, 0 };
2152 
2153 void __init setup_per_cpu_pageset(void)
2154 {
2155 	int err;
2156 
2157 	/* Initialize per_cpu_pageset for cpu 0.
2158 	 * A cpuup callback will do this for every cpu
2159 	 * as it comes online
2160 	 */
2161 	err = process_zones(smp_processor_id());
2162 	BUG_ON(err);
2163 	register_cpu_notifier(&pageset_notifier);
2164 }
2165 
2166 #endif
2167 
2168 static __meminit noinline
2169 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2170 {
2171 	int i;
2172 	struct pglist_data *pgdat = zone->zone_pgdat;
2173 	size_t alloc_size;
2174 
2175 	/*
2176 	 * The per-page waitqueue mechanism uses hashed waitqueues
2177 	 * per zone.
2178 	 */
2179 	zone->wait_table_hash_nr_entries =
2180 		 wait_table_hash_nr_entries(zone_size_pages);
2181 	zone->wait_table_bits =
2182 		wait_table_bits(zone->wait_table_hash_nr_entries);
2183 	alloc_size = zone->wait_table_hash_nr_entries
2184 					* sizeof(wait_queue_head_t);
2185 
2186  	if (system_state == SYSTEM_BOOTING) {
2187 		zone->wait_table = (wait_queue_head_t *)
2188 			alloc_bootmem_node(pgdat, alloc_size);
2189 	} else {
2190 		/*
2191 		 * This case means that a zone whose size was 0 gets new memory
2192 		 * via memory hot-add.
2193 		 * But it may be the case that a new node was hot-added.  In
2194 		 * this case vmalloc() will not be able to use this new node's
2195 		 * memory - this wait_table must be initialized to use this new
2196 		 * node itself as well.
2197 		 * To use this new node's memory, further consideration will be
2198 		 * necessary.
2199 		 */
2200 		zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size);
2201 	}
2202 	if (!zone->wait_table)
2203 		return -ENOMEM;
2204 
2205 	for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2206 		init_waitqueue_head(zone->wait_table + i);
2207 
2208 	return 0;
2209 }
2210 
2211 static __meminit void zone_pcp_init(struct zone *zone)
2212 {
2213 	int cpu;
2214 	unsigned long batch = zone_batchsize(zone);
2215 
2216 	for (cpu = 0; cpu < NR_CPUS; cpu++) {
2217 #ifdef CONFIG_NUMA
2218 		/* Early boot. Slab allocator not functional yet */
2219 		zone_pcp(zone, cpu) = &boot_pageset[cpu];
2220 		setup_pageset(&boot_pageset[cpu],0);
2221 #else
2222 		setup_pageset(zone_pcp(zone,cpu), batch);
2223 #endif
2224 	}
2225 	if (zone->present_pages)
2226 		printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%lu\n",
2227 			zone->name, zone->present_pages, batch);
2228 }
2229 
2230 __meminit int init_currently_empty_zone(struct zone *zone,
2231 					unsigned long zone_start_pfn,
2232 					unsigned long size,
2233 					enum memmap_context context)
2234 {
2235 	struct pglist_data *pgdat = zone->zone_pgdat;
2236 	int ret;
2237 	ret = zone_wait_table_init(zone, size);
2238 	if (ret)
2239 		return ret;
2240 	pgdat->nr_zones = zone_idx(zone) + 1;
2241 
2242 	zone->zone_start_pfn = zone_start_pfn;
2243 
2244 	memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2245 
2246 	zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2247 
2248 	return 0;
2249 }
2250 
2251 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2252 /*
2253  * Basic iterator support. Return the first range of PFNs for a node
2254  * Note: nid == MAX_NUMNODES returns first region regardless of node
2255  */
2256 static int __meminit first_active_region_index_in_nid(int nid)
2257 {
2258 	int i;
2259 
2260 	for (i = 0; i < nr_nodemap_entries; i++)
2261 		if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2262 			return i;
2263 
2264 	return -1;
2265 }
2266 
2267 /*
2268  * Basic iterator support. Return the next active range of PFNs for a node
2269  * Note: nid == MAX_NUMNODES returns next region regardles of node
2270  */
2271 static int __meminit next_active_region_index_in_nid(int index, int nid)
2272 {
2273 	for (index = index + 1; index < nr_nodemap_entries; index++)
2274 		if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2275 			return index;
2276 
2277 	return -1;
2278 }
2279 
2280 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2281 /*
2282  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2283  * Architectures may implement their own version but if add_active_range()
2284  * was used and there are no special requirements, this is a convenient
2285  * alternative
2286  */
2287 int __meminit early_pfn_to_nid(unsigned long pfn)
2288 {
2289 	int i;
2290 
2291 	for (i = 0; i < nr_nodemap_entries; i++) {
2292 		unsigned long start_pfn = early_node_map[i].start_pfn;
2293 		unsigned long end_pfn = early_node_map[i].end_pfn;
2294 
2295 		if (start_pfn <= pfn && pfn < end_pfn)
2296 			return early_node_map[i].nid;
2297 	}
2298 
2299 	return 0;
2300 }
2301 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2302 
2303 /* Basic iterator support to walk early_node_map[] */
2304 #define for_each_active_range_index_in_nid(i, nid) \
2305 	for (i = first_active_region_index_in_nid(nid); i != -1; \
2306 				i = next_active_region_index_in_nid(i, nid))
2307 
2308 /**
2309  * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2310  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2311  * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2312  *
2313  * If an architecture guarantees that all ranges registered with
2314  * add_active_ranges() contain no holes and may be freed, this
2315  * this function may be used instead of calling free_bootmem() manually.
2316  */
2317 void __init free_bootmem_with_active_regions(int nid,
2318 						unsigned long max_low_pfn)
2319 {
2320 	int i;
2321 
2322 	for_each_active_range_index_in_nid(i, nid) {
2323 		unsigned long size_pages = 0;
2324 		unsigned long end_pfn = early_node_map[i].end_pfn;
2325 
2326 		if (early_node_map[i].start_pfn >= max_low_pfn)
2327 			continue;
2328 
2329 		if (end_pfn > max_low_pfn)
2330 			end_pfn = max_low_pfn;
2331 
2332 		size_pages = end_pfn - early_node_map[i].start_pfn;
2333 		free_bootmem_node(NODE_DATA(early_node_map[i].nid),
2334 				PFN_PHYS(early_node_map[i].start_pfn),
2335 				size_pages << PAGE_SHIFT);
2336 	}
2337 }
2338 
2339 /**
2340  * sparse_memory_present_with_active_regions - Call memory_present for each active range
2341  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2342  *
2343  * If an architecture guarantees that all ranges registered with
2344  * add_active_ranges() contain no holes and may be freed, this
2345  * function may be used instead of calling memory_present() manually.
2346  */
2347 void __init sparse_memory_present_with_active_regions(int nid)
2348 {
2349 	int i;
2350 
2351 	for_each_active_range_index_in_nid(i, nid)
2352 		memory_present(early_node_map[i].nid,
2353 				early_node_map[i].start_pfn,
2354 				early_node_map[i].end_pfn);
2355 }
2356 
2357 /**
2358  * push_node_boundaries - Push node boundaries to at least the requested boundary
2359  * @nid: The nid of the node to push the boundary for
2360  * @start_pfn: The start pfn of the node
2361  * @end_pfn: The end pfn of the node
2362  *
2363  * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2364  * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2365  * be hotplugged even though no physical memory exists. This function allows
2366  * an arch to push out the node boundaries so mem_map is allocated that can
2367  * be used later.
2368  */
2369 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2370 void __init push_node_boundaries(unsigned int nid,
2371 		unsigned long start_pfn, unsigned long end_pfn)
2372 {
2373 	printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
2374 			nid, start_pfn, end_pfn);
2375 
2376 	/* Initialise the boundary for this node if necessary */
2377 	if (node_boundary_end_pfn[nid] == 0)
2378 		node_boundary_start_pfn[nid] = -1UL;
2379 
2380 	/* Update the boundaries */
2381 	if (node_boundary_start_pfn[nid] > start_pfn)
2382 		node_boundary_start_pfn[nid] = start_pfn;
2383 	if (node_boundary_end_pfn[nid] < end_pfn)
2384 		node_boundary_end_pfn[nid] = end_pfn;
2385 }
2386 
2387 /* If necessary, push the node boundary out for reserve hotadd */
2388 static void __init account_node_boundary(unsigned int nid,
2389 		unsigned long *start_pfn, unsigned long *end_pfn)
2390 {
2391 	printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
2392 			nid, *start_pfn, *end_pfn);
2393 
2394 	/* Return if boundary information has not been provided */
2395 	if (node_boundary_end_pfn[nid] == 0)
2396 		return;
2397 
2398 	/* Check the boundaries and update if necessary */
2399 	if (node_boundary_start_pfn[nid] < *start_pfn)
2400 		*start_pfn = node_boundary_start_pfn[nid];
2401 	if (node_boundary_end_pfn[nid] > *end_pfn)
2402 		*end_pfn = node_boundary_end_pfn[nid];
2403 }
2404 #else
2405 void __init push_node_boundaries(unsigned int nid,
2406 		unsigned long start_pfn, unsigned long end_pfn) {}
2407 
2408 static void __init account_node_boundary(unsigned int nid,
2409 		unsigned long *start_pfn, unsigned long *end_pfn) {}
2410 #endif
2411 
2412 
2413 /**
2414  * get_pfn_range_for_nid - Return the start and end page frames for a node
2415  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
2416  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
2417  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
2418  *
2419  * It returns the start and end page frame of a node based on information
2420  * provided by an arch calling add_active_range(). If called for a node
2421  * with no available memory, a warning is printed and the start and end
2422  * PFNs will be 0.
2423  */
2424 void __meminit get_pfn_range_for_nid(unsigned int nid,
2425 			unsigned long *start_pfn, unsigned long *end_pfn)
2426 {
2427 	int i;
2428 	*start_pfn = -1UL;
2429 	*end_pfn = 0;
2430 
2431 	for_each_active_range_index_in_nid(i, nid) {
2432 		*start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
2433 		*end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
2434 	}
2435 
2436 	if (*start_pfn == -1UL) {
2437 		printk(KERN_WARNING "Node %u active with no memory\n", nid);
2438 		*start_pfn = 0;
2439 	}
2440 
2441 	/* Push the node boundaries out if requested */
2442 	account_node_boundary(nid, start_pfn, end_pfn);
2443 }
2444 
2445 /*
2446  * Return the number of pages a zone spans in a node, including holes
2447  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
2448  */
2449 unsigned long __meminit zone_spanned_pages_in_node(int nid,
2450 					unsigned long zone_type,
2451 					unsigned long *ignored)
2452 {
2453 	unsigned long node_start_pfn, node_end_pfn;
2454 	unsigned long zone_start_pfn, zone_end_pfn;
2455 
2456 	/* Get the start and end of the node and zone */
2457 	get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2458 	zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
2459 	zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
2460 
2461 	/* Check that this node has pages within the zone's required range */
2462 	if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
2463 		return 0;
2464 
2465 	/* Move the zone boundaries inside the node if necessary */
2466 	zone_end_pfn = min(zone_end_pfn, node_end_pfn);
2467 	zone_start_pfn = max(zone_start_pfn, node_start_pfn);
2468 
2469 	/* Return the spanned pages */
2470 	return zone_end_pfn - zone_start_pfn;
2471 }
2472 
2473 /*
2474  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
2475  * then all holes in the requested range will be accounted for.
2476  */
2477 unsigned long __meminit __absent_pages_in_range(int nid,
2478 				unsigned long range_start_pfn,
2479 				unsigned long range_end_pfn)
2480 {
2481 	int i = 0;
2482 	unsigned long prev_end_pfn = 0, hole_pages = 0;
2483 	unsigned long start_pfn;
2484 
2485 	/* Find the end_pfn of the first active range of pfns in the node */
2486 	i = first_active_region_index_in_nid(nid);
2487 	if (i == -1)
2488 		return 0;
2489 
2490 	/* Account for ranges before physical memory on this node */
2491 	if (early_node_map[i].start_pfn > range_start_pfn)
2492 		hole_pages = early_node_map[i].start_pfn - range_start_pfn;
2493 
2494 	prev_end_pfn = early_node_map[i].start_pfn;
2495 
2496 	/* Find all holes for the zone within the node */
2497 	for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
2498 
2499 		/* No need to continue if prev_end_pfn is outside the zone */
2500 		if (prev_end_pfn >= range_end_pfn)
2501 			break;
2502 
2503 		/* Make sure the end of the zone is not within the hole */
2504 		start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
2505 		prev_end_pfn = max(prev_end_pfn, range_start_pfn);
2506 
2507 		/* Update the hole size cound and move on */
2508 		if (start_pfn > range_start_pfn) {
2509 			BUG_ON(prev_end_pfn > start_pfn);
2510 			hole_pages += start_pfn - prev_end_pfn;
2511 		}
2512 		prev_end_pfn = early_node_map[i].end_pfn;
2513 	}
2514 
2515 	/* Account for ranges past physical memory on this node */
2516 	if (range_end_pfn > prev_end_pfn)
2517 		hole_pages += range_end_pfn -
2518 				max(range_start_pfn, prev_end_pfn);
2519 
2520 	return hole_pages;
2521 }
2522 
2523 /**
2524  * absent_pages_in_range - Return number of page frames in holes within a range
2525  * @start_pfn: The start PFN to start searching for holes
2526  * @end_pfn: The end PFN to stop searching for holes
2527  *
2528  * It returns the number of pages frames in memory holes within a range.
2529  */
2530 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
2531 							unsigned long end_pfn)
2532 {
2533 	return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
2534 }
2535 
2536 /* Return the number of page frames in holes in a zone on a node */
2537 unsigned long __meminit zone_absent_pages_in_node(int nid,
2538 					unsigned long zone_type,
2539 					unsigned long *ignored)
2540 {
2541 	unsigned long node_start_pfn, node_end_pfn;
2542 	unsigned long zone_start_pfn, zone_end_pfn;
2543 
2544 	get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
2545 	zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
2546 							node_start_pfn);
2547 	zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
2548 							node_end_pfn);
2549 
2550 	return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
2551 }
2552 
2553 #else
2554 static inline unsigned long zone_spanned_pages_in_node(int nid,
2555 					unsigned long zone_type,
2556 					unsigned long *zones_size)
2557 {
2558 	return zones_size[zone_type];
2559 }
2560 
2561 static inline unsigned long zone_absent_pages_in_node(int nid,
2562 						unsigned long zone_type,
2563 						unsigned long *zholes_size)
2564 {
2565 	if (!zholes_size)
2566 		return 0;
2567 
2568 	return zholes_size[zone_type];
2569 }
2570 
2571 #endif
2572 
2573 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
2574 		unsigned long *zones_size, unsigned long *zholes_size)
2575 {
2576 	unsigned long realtotalpages, totalpages = 0;
2577 	enum zone_type i;
2578 
2579 	for (i = 0; i < MAX_NR_ZONES; i++)
2580 		totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
2581 								zones_size);
2582 	pgdat->node_spanned_pages = totalpages;
2583 
2584 	realtotalpages = totalpages;
2585 	for (i = 0; i < MAX_NR_ZONES; i++)
2586 		realtotalpages -=
2587 			zone_absent_pages_in_node(pgdat->node_id, i,
2588 								zholes_size);
2589 	pgdat->node_present_pages = realtotalpages;
2590 	printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
2591 							realtotalpages);
2592 }
2593 
2594 /*
2595  * Set up the zone data structures:
2596  *   - mark all pages reserved
2597  *   - mark all memory queues empty
2598  *   - clear the memory bitmaps
2599  */
2600 static void __meminit free_area_init_core(struct pglist_data *pgdat,
2601 		unsigned long *zones_size, unsigned long *zholes_size)
2602 {
2603 	enum zone_type j;
2604 	int nid = pgdat->node_id;
2605 	unsigned long zone_start_pfn = pgdat->node_start_pfn;
2606 	int ret;
2607 
2608 	pgdat_resize_init(pgdat);
2609 	pgdat->nr_zones = 0;
2610 	init_waitqueue_head(&pgdat->kswapd_wait);
2611 	pgdat->kswapd_max_order = 0;
2612 
2613 	for (j = 0; j < MAX_NR_ZONES; j++) {
2614 		struct zone *zone = pgdat->node_zones + j;
2615 		unsigned long size, realsize, memmap_pages;
2616 
2617 		size = zone_spanned_pages_in_node(nid, j, zones_size);
2618 		realsize = size - zone_absent_pages_in_node(nid, j,
2619 								zholes_size);
2620 
2621 		/*
2622 		 * Adjust realsize so that it accounts for how much memory
2623 		 * is used by this zone for memmap. This affects the watermark
2624 		 * and per-cpu initialisations
2625 		 */
2626 		memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
2627 		if (realsize >= memmap_pages) {
2628 			realsize -= memmap_pages;
2629 			printk(KERN_DEBUG
2630 				"  %s zone: %lu pages used for memmap\n",
2631 				zone_names[j], memmap_pages);
2632 		} else
2633 			printk(KERN_WARNING
2634 				"  %s zone: %lu pages exceeds realsize %lu\n",
2635 				zone_names[j], memmap_pages, realsize);
2636 
2637 		/* Account for reserved pages */
2638 		if (j == 0 && realsize > dma_reserve) {
2639 			realsize -= dma_reserve;
2640 			printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
2641 					zone_names[0], dma_reserve);
2642 		}
2643 
2644 		if (!is_highmem_idx(j))
2645 			nr_kernel_pages += realsize;
2646 		nr_all_pages += realsize;
2647 
2648 		zone->spanned_pages = size;
2649 		zone->present_pages = realsize;
2650 #ifdef CONFIG_NUMA
2651 		zone->node = nid;
2652 		zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
2653 						/ 100;
2654 		zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
2655 #endif
2656 		zone->name = zone_names[j];
2657 		spin_lock_init(&zone->lock);
2658 		spin_lock_init(&zone->lru_lock);
2659 		zone_seqlock_init(zone);
2660 		zone->zone_pgdat = pgdat;
2661 
2662 		zone->prev_priority = DEF_PRIORITY;
2663 
2664 		zone_pcp_init(zone);
2665 		INIT_LIST_HEAD(&zone->active_list);
2666 		INIT_LIST_HEAD(&zone->inactive_list);
2667 		zone->nr_scan_active = 0;
2668 		zone->nr_scan_inactive = 0;
2669 		zap_zone_vm_stats(zone);
2670 		atomic_set(&zone->reclaim_in_progress, 0);
2671 		if (!size)
2672 			continue;
2673 
2674 		ret = init_currently_empty_zone(zone, zone_start_pfn,
2675 						size, MEMMAP_EARLY);
2676 		BUG_ON(ret);
2677 		zone_start_pfn += size;
2678 	}
2679 }
2680 
2681 static void __meminit alloc_node_mem_map(struct pglist_data *pgdat)
2682 {
2683 	/* Skip empty nodes */
2684 	if (!pgdat->node_spanned_pages)
2685 		return;
2686 
2687 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2688 	/* ia64 gets its own node_mem_map, before this, without bootmem */
2689 	if (!pgdat->node_mem_map) {
2690 		unsigned long size, start, end;
2691 		struct page *map;
2692 
2693 		/*
2694 		 * The zone's endpoints aren't required to be MAX_ORDER
2695 		 * aligned but the node_mem_map endpoints must be in order
2696 		 * for the buddy allocator to function correctly.
2697 		 */
2698 		start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
2699 		end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
2700 		end = ALIGN(end, MAX_ORDER_NR_PAGES);
2701 		size =  (end - start) * sizeof(struct page);
2702 		map = alloc_remap(pgdat->node_id, size);
2703 		if (!map)
2704 			map = alloc_bootmem_node(pgdat, size);
2705 		pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
2706 	}
2707 #ifdef CONFIG_FLATMEM
2708 	/*
2709 	 * With no DISCONTIG, the global mem_map is just set as node 0's
2710 	 */
2711 	if (pgdat == NODE_DATA(0)) {
2712 		mem_map = NODE_DATA(0)->node_mem_map;
2713 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2714 		if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
2715 			mem_map -= pgdat->node_start_pfn;
2716 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2717 	}
2718 #endif
2719 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2720 }
2721 
2722 void __meminit free_area_init_node(int nid, struct pglist_data *pgdat,
2723 		unsigned long *zones_size, unsigned long node_start_pfn,
2724 		unsigned long *zholes_size)
2725 {
2726 	pgdat->node_id = nid;
2727 	pgdat->node_start_pfn = node_start_pfn;
2728 	calculate_node_totalpages(pgdat, zones_size, zholes_size);
2729 
2730 	alloc_node_mem_map(pgdat);
2731 
2732 	free_area_init_core(pgdat, zones_size, zholes_size);
2733 }
2734 
2735 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2736 /**
2737  * add_active_range - Register a range of PFNs backed by physical memory
2738  * @nid: The node ID the range resides on
2739  * @start_pfn: The start PFN of the available physical memory
2740  * @end_pfn: The end PFN of the available physical memory
2741  *
2742  * These ranges are stored in an early_node_map[] and later used by
2743  * free_area_init_nodes() to calculate zone sizes and holes. If the
2744  * range spans a memory hole, it is up to the architecture to ensure
2745  * the memory is not freed by the bootmem allocator. If possible
2746  * the range being registered will be merged with existing ranges.
2747  */
2748 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
2749 						unsigned long end_pfn)
2750 {
2751 	int i;
2752 
2753 	printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
2754 			  "%d entries of %d used\n",
2755 			  nid, start_pfn, end_pfn,
2756 			  nr_nodemap_entries, MAX_ACTIVE_REGIONS);
2757 
2758 	/* Merge with existing active regions if possible */
2759 	for (i = 0; i < nr_nodemap_entries; i++) {
2760 		if (early_node_map[i].nid != nid)
2761 			continue;
2762 
2763 		/* Skip if an existing region covers this new one */
2764 		if (start_pfn >= early_node_map[i].start_pfn &&
2765 				end_pfn <= early_node_map[i].end_pfn)
2766 			return;
2767 
2768 		/* Merge forward if suitable */
2769 		if (start_pfn <= early_node_map[i].end_pfn &&
2770 				end_pfn > early_node_map[i].end_pfn) {
2771 			early_node_map[i].end_pfn = end_pfn;
2772 			return;
2773 		}
2774 
2775 		/* Merge backward if suitable */
2776 		if (start_pfn < early_node_map[i].end_pfn &&
2777 				end_pfn >= early_node_map[i].start_pfn) {
2778 			early_node_map[i].start_pfn = start_pfn;
2779 			return;
2780 		}
2781 	}
2782 
2783 	/* Check that early_node_map is large enough */
2784 	if (i >= MAX_ACTIVE_REGIONS) {
2785 		printk(KERN_CRIT "More than %d memory regions, truncating\n",
2786 							MAX_ACTIVE_REGIONS);
2787 		return;
2788 	}
2789 
2790 	early_node_map[i].nid = nid;
2791 	early_node_map[i].start_pfn = start_pfn;
2792 	early_node_map[i].end_pfn = end_pfn;
2793 	nr_nodemap_entries = i + 1;
2794 }
2795 
2796 /**
2797  * shrink_active_range - Shrink an existing registered range of PFNs
2798  * @nid: The node id the range is on that should be shrunk
2799  * @old_end_pfn: The old end PFN of the range
2800  * @new_end_pfn: The new PFN of the range
2801  *
2802  * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
2803  * The map is kept at the end physical page range that has already been
2804  * registered with add_active_range(). This function allows an arch to shrink
2805  * an existing registered range.
2806  */
2807 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
2808 						unsigned long new_end_pfn)
2809 {
2810 	int i;
2811 
2812 	/* Find the old active region end and shrink */
2813 	for_each_active_range_index_in_nid(i, nid)
2814 		if (early_node_map[i].end_pfn == old_end_pfn) {
2815 			early_node_map[i].end_pfn = new_end_pfn;
2816 			break;
2817 		}
2818 }
2819 
2820 /**
2821  * remove_all_active_ranges - Remove all currently registered regions
2822  *
2823  * During discovery, it may be found that a table like SRAT is invalid
2824  * and an alternative discovery method must be used. This function removes
2825  * all currently registered regions.
2826  */
2827 void __init remove_all_active_ranges(void)
2828 {
2829 	memset(early_node_map, 0, sizeof(early_node_map));
2830 	nr_nodemap_entries = 0;
2831 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2832 	memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
2833 	memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
2834 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
2835 }
2836 
2837 /* Compare two active node_active_regions */
2838 static int __init cmp_node_active_region(const void *a, const void *b)
2839 {
2840 	struct node_active_region *arange = (struct node_active_region *)a;
2841 	struct node_active_region *brange = (struct node_active_region *)b;
2842 
2843 	/* Done this way to avoid overflows */
2844 	if (arange->start_pfn > brange->start_pfn)
2845 		return 1;
2846 	if (arange->start_pfn < brange->start_pfn)
2847 		return -1;
2848 
2849 	return 0;
2850 }
2851 
2852 /* sort the node_map by start_pfn */
2853 static void __init sort_node_map(void)
2854 {
2855 	sort(early_node_map, (size_t)nr_nodemap_entries,
2856 			sizeof(struct node_active_region),
2857 			cmp_node_active_region, NULL);
2858 }
2859 
2860 /* Find the lowest pfn for a node */
2861 unsigned long __init find_min_pfn_for_node(unsigned long nid)
2862 {
2863 	int i;
2864 	unsigned long min_pfn = ULONG_MAX;
2865 
2866 	/* Assuming a sorted map, the first range found has the starting pfn */
2867 	for_each_active_range_index_in_nid(i, nid)
2868 		min_pfn = min(min_pfn, early_node_map[i].start_pfn);
2869 
2870 	if (min_pfn == ULONG_MAX) {
2871 		printk(KERN_WARNING
2872 			"Could not find start_pfn for node %lu\n", nid);
2873 		return 0;
2874 	}
2875 
2876 	return min_pfn;
2877 }
2878 
2879 /**
2880  * find_min_pfn_with_active_regions - Find the minimum PFN registered
2881  *
2882  * It returns the minimum PFN based on information provided via
2883  * add_active_range().
2884  */
2885 unsigned long __init find_min_pfn_with_active_regions(void)
2886 {
2887 	return find_min_pfn_for_node(MAX_NUMNODES);
2888 }
2889 
2890 /**
2891  * find_max_pfn_with_active_regions - Find the maximum PFN registered
2892  *
2893  * It returns the maximum PFN based on information provided via
2894  * add_active_range().
2895  */
2896 unsigned long __init find_max_pfn_with_active_regions(void)
2897 {
2898 	int i;
2899 	unsigned long max_pfn = 0;
2900 
2901 	for (i = 0; i < nr_nodemap_entries; i++)
2902 		max_pfn = max(max_pfn, early_node_map[i].end_pfn);
2903 
2904 	return max_pfn;
2905 }
2906 
2907 /**
2908  * free_area_init_nodes - Initialise all pg_data_t and zone data
2909  * @max_zone_pfn: an array of max PFNs for each zone
2910  *
2911  * This will call free_area_init_node() for each active node in the system.
2912  * Using the page ranges provided by add_active_range(), the size of each
2913  * zone in each node and their holes is calculated. If the maximum PFN
2914  * between two adjacent zones match, it is assumed that the zone is empty.
2915  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
2916  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
2917  * starts where the previous one ended. For example, ZONE_DMA32 starts
2918  * at arch_max_dma_pfn.
2919  */
2920 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
2921 {
2922 	unsigned long nid;
2923 	enum zone_type i;
2924 
2925 	/* Sort early_node_map as initialisation assumes it is sorted */
2926 	sort_node_map();
2927 
2928 	/* Record where the zone boundaries are */
2929 	memset(arch_zone_lowest_possible_pfn, 0,
2930 				sizeof(arch_zone_lowest_possible_pfn));
2931 	memset(arch_zone_highest_possible_pfn, 0,
2932 				sizeof(arch_zone_highest_possible_pfn));
2933 	arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
2934 	arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
2935 	for (i = 1; i < MAX_NR_ZONES; i++) {
2936 		arch_zone_lowest_possible_pfn[i] =
2937 			arch_zone_highest_possible_pfn[i-1];
2938 		arch_zone_highest_possible_pfn[i] =
2939 			max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
2940 	}
2941 
2942 	/* Print out the zone ranges */
2943 	printk("Zone PFN ranges:\n");
2944 	for (i = 0; i < MAX_NR_ZONES; i++)
2945 		printk("  %-8s %8lu -> %8lu\n",
2946 				zone_names[i],
2947 				arch_zone_lowest_possible_pfn[i],
2948 				arch_zone_highest_possible_pfn[i]);
2949 
2950 	/* Print out the early_node_map[] */
2951 	printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
2952 	for (i = 0; i < nr_nodemap_entries; i++)
2953 		printk("  %3d: %8lu -> %8lu\n", early_node_map[i].nid,
2954 						early_node_map[i].start_pfn,
2955 						early_node_map[i].end_pfn);
2956 
2957 	/* Initialise every node */
2958 	setup_nr_node_ids();
2959 	for_each_online_node(nid) {
2960 		pg_data_t *pgdat = NODE_DATA(nid);
2961 		free_area_init_node(nid, pgdat, NULL,
2962 				find_min_pfn_for_node(nid), NULL);
2963 	}
2964 }
2965 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
2966 
2967 /**
2968  * set_dma_reserve - set the specified number of pages reserved in the first zone
2969  * @new_dma_reserve: The number of pages to mark reserved
2970  *
2971  * The per-cpu batchsize and zone watermarks are determined by present_pages.
2972  * In the DMA zone, a significant percentage may be consumed by kernel image
2973  * and other unfreeable allocations which can skew the watermarks badly. This
2974  * function may optionally be used to account for unfreeable pages in the
2975  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2976  * smaller per-cpu batchsize.
2977  */
2978 void __init set_dma_reserve(unsigned long new_dma_reserve)
2979 {
2980 	dma_reserve = new_dma_reserve;
2981 }
2982 
2983 #ifndef CONFIG_NEED_MULTIPLE_NODES
2984 static bootmem_data_t contig_bootmem_data;
2985 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2986 
2987 EXPORT_SYMBOL(contig_page_data);
2988 #endif
2989 
2990 void __init free_area_init(unsigned long *zones_size)
2991 {
2992 	free_area_init_node(0, NODE_DATA(0), zones_size,
2993 			__pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2994 }
2995 
2996 static int page_alloc_cpu_notify(struct notifier_block *self,
2997 				 unsigned long action, void *hcpu)
2998 {
2999 	int cpu = (unsigned long)hcpu;
3000 
3001 	if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
3002 		local_irq_disable();
3003 		__drain_pages(cpu);
3004 		vm_events_fold_cpu(cpu);
3005 		local_irq_enable();
3006 		refresh_cpu_vm_stats(cpu);
3007 	}
3008 	return NOTIFY_OK;
3009 }
3010 
3011 void __init page_alloc_init(void)
3012 {
3013 	hotcpu_notifier(page_alloc_cpu_notify, 0);
3014 }
3015 
3016 /*
3017  * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3018  *	or min_free_kbytes changes.
3019  */
3020 static void calculate_totalreserve_pages(void)
3021 {
3022 	struct pglist_data *pgdat;
3023 	unsigned long reserve_pages = 0;
3024 	enum zone_type i, j;
3025 
3026 	for_each_online_pgdat(pgdat) {
3027 		for (i = 0; i < MAX_NR_ZONES; i++) {
3028 			struct zone *zone = pgdat->node_zones + i;
3029 			unsigned long max = 0;
3030 
3031 			/* Find valid and maximum lowmem_reserve in the zone */
3032 			for (j = i; j < MAX_NR_ZONES; j++) {
3033 				if (zone->lowmem_reserve[j] > max)
3034 					max = zone->lowmem_reserve[j];
3035 			}
3036 
3037 			/* we treat pages_high as reserved pages. */
3038 			max += zone->pages_high;
3039 
3040 			if (max > zone->present_pages)
3041 				max = zone->present_pages;
3042 			reserve_pages += max;
3043 		}
3044 	}
3045 	totalreserve_pages = reserve_pages;
3046 }
3047 
3048 /*
3049  * setup_per_zone_lowmem_reserve - called whenever
3050  *	sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
3051  *	has a correct pages reserved value, so an adequate number of
3052  *	pages are left in the zone after a successful __alloc_pages().
3053  */
3054 static void setup_per_zone_lowmem_reserve(void)
3055 {
3056 	struct pglist_data *pgdat;
3057 	enum zone_type j, idx;
3058 
3059 	for_each_online_pgdat(pgdat) {
3060 		for (j = 0; j < MAX_NR_ZONES; j++) {
3061 			struct zone *zone = pgdat->node_zones + j;
3062 			unsigned long present_pages = zone->present_pages;
3063 
3064 			zone->lowmem_reserve[j] = 0;
3065 
3066 			idx = j;
3067 			while (idx) {
3068 				struct zone *lower_zone;
3069 
3070 				idx--;
3071 
3072 				if (sysctl_lowmem_reserve_ratio[idx] < 1)
3073 					sysctl_lowmem_reserve_ratio[idx] = 1;
3074 
3075 				lower_zone = pgdat->node_zones + idx;
3076 				lower_zone->lowmem_reserve[j] = present_pages /
3077 					sysctl_lowmem_reserve_ratio[idx];
3078 				present_pages += lower_zone->present_pages;
3079 			}
3080 		}
3081 	}
3082 
3083 	/* update totalreserve_pages */
3084 	calculate_totalreserve_pages();
3085 }
3086 
3087 /**
3088  * setup_per_zone_pages_min - called when min_free_kbytes changes.
3089  *
3090  * Ensures that the pages_{min,low,high} values for each zone are set correctly
3091  * with respect to min_free_kbytes.
3092  */
3093 void setup_per_zone_pages_min(void)
3094 {
3095 	unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
3096 	unsigned long lowmem_pages = 0;
3097 	struct zone *zone;
3098 	unsigned long flags;
3099 
3100 	/* Calculate total number of !ZONE_HIGHMEM pages */
3101 	for_each_zone(zone) {
3102 		if (!is_highmem(zone))
3103 			lowmem_pages += zone->present_pages;
3104 	}
3105 
3106 	for_each_zone(zone) {
3107 		u64 tmp;
3108 
3109 		spin_lock_irqsave(&zone->lru_lock, flags);
3110 		tmp = (u64)pages_min * zone->present_pages;
3111 		do_div(tmp, lowmem_pages);
3112 		if (is_highmem(zone)) {
3113 			/*
3114 			 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
3115 			 * need highmem pages, so cap pages_min to a small
3116 			 * value here.
3117 			 *
3118 			 * The (pages_high-pages_low) and (pages_low-pages_min)
3119 			 * deltas controls asynch page reclaim, and so should
3120 			 * not be capped for highmem.
3121 			 */
3122 			int min_pages;
3123 
3124 			min_pages = zone->present_pages / 1024;
3125 			if (min_pages < SWAP_CLUSTER_MAX)
3126 				min_pages = SWAP_CLUSTER_MAX;
3127 			if (min_pages > 128)
3128 				min_pages = 128;
3129 			zone->pages_min = min_pages;
3130 		} else {
3131 			/*
3132 			 * If it's a lowmem zone, reserve a number of pages
3133 			 * proportionate to the zone's size.
3134 			 */
3135 			zone->pages_min = tmp;
3136 		}
3137 
3138 		zone->pages_low   = zone->pages_min + (tmp >> 2);
3139 		zone->pages_high  = zone->pages_min + (tmp >> 1);
3140 		spin_unlock_irqrestore(&zone->lru_lock, flags);
3141 	}
3142 
3143 	/* update totalreserve_pages */
3144 	calculate_totalreserve_pages();
3145 }
3146 
3147 /*
3148  * Initialise min_free_kbytes.
3149  *
3150  * For small machines we want it small (128k min).  For large machines
3151  * we want it large (64MB max).  But it is not linear, because network
3152  * bandwidth does not increase linearly with machine size.  We use
3153  *
3154  * 	min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
3155  *	min_free_kbytes = sqrt(lowmem_kbytes * 16)
3156  *
3157  * which yields
3158  *
3159  * 16MB:	512k
3160  * 32MB:	724k
3161  * 64MB:	1024k
3162  * 128MB:	1448k
3163  * 256MB:	2048k
3164  * 512MB:	2896k
3165  * 1024MB:	4096k
3166  * 2048MB:	5792k
3167  * 4096MB:	8192k
3168  * 8192MB:	11584k
3169  * 16384MB:	16384k
3170  */
3171 static int __init init_per_zone_pages_min(void)
3172 {
3173 	unsigned long lowmem_kbytes;
3174 
3175 	lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
3176 
3177 	min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
3178 	if (min_free_kbytes < 128)
3179 		min_free_kbytes = 128;
3180 	if (min_free_kbytes > 65536)
3181 		min_free_kbytes = 65536;
3182 	setup_per_zone_pages_min();
3183 	setup_per_zone_lowmem_reserve();
3184 	return 0;
3185 }
3186 module_init(init_per_zone_pages_min)
3187 
3188 /*
3189  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
3190  *	that we can call two helper functions whenever min_free_kbytes
3191  *	changes.
3192  */
3193 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
3194 	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3195 {
3196 	proc_dointvec(table, write, file, buffer, length, ppos);
3197 	if (write)
3198 		setup_per_zone_pages_min();
3199 	return 0;
3200 }
3201 
3202 #ifdef CONFIG_NUMA
3203 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
3204 	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3205 {
3206 	struct zone *zone;
3207 	int rc;
3208 
3209 	rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3210 	if (rc)
3211 		return rc;
3212 
3213 	for_each_zone(zone)
3214 		zone->min_unmapped_pages = (zone->present_pages *
3215 				sysctl_min_unmapped_ratio) / 100;
3216 	return 0;
3217 }
3218 
3219 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
3220 	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3221 {
3222 	struct zone *zone;
3223 	int rc;
3224 
3225 	rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3226 	if (rc)
3227 		return rc;
3228 
3229 	for_each_zone(zone)
3230 		zone->min_slab_pages = (zone->present_pages *
3231 				sysctl_min_slab_ratio) / 100;
3232 	return 0;
3233 }
3234 #endif
3235 
3236 /*
3237  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
3238  *	proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
3239  *	whenever sysctl_lowmem_reserve_ratio changes.
3240  *
3241  * The reserve ratio obviously has absolutely no relation with the
3242  * pages_min watermarks. The lowmem reserve ratio can only make sense
3243  * if in function of the boot time zone sizes.
3244  */
3245 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
3246 	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3247 {
3248 	proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3249 	setup_per_zone_lowmem_reserve();
3250 	return 0;
3251 }
3252 
3253 /*
3254  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
3255  * cpu.  It is the fraction of total pages in each zone that a hot per cpu pagelist
3256  * can have before it gets flushed back to buddy allocator.
3257  */
3258 
3259 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
3260 	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
3261 {
3262 	struct zone *zone;
3263 	unsigned int cpu;
3264 	int ret;
3265 
3266 	ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
3267 	if (!write || (ret == -EINVAL))
3268 		return ret;
3269 	for_each_zone(zone) {
3270 		for_each_online_cpu(cpu) {
3271 			unsigned long  high;
3272 			high = zone->present_pages / percpu_pagelist_fraction;
3273 			setup_pagelist_highmark(zone_pcp(zone, cpu), high);
3274 		}
3275 	}
3276 	return 0;
3277 }
3278 
3279 int hashdist = HASHDIST_DEFAULT;
3280 
3281 #ifdef CONFIG_NUMA
3282 static int __init set_hashdist(char *str)
3283 {
3284 	if (!str)
3285 		return 0;
3286 	hashdist = simple_strtoul(str, &str, 0);
3287 	return 1;
3288 }
3289 __setup("hashdist=", set_hashdist);
3290 #endif
3291 
3292 /*
3293  * allocate a large system hash table from bootmem
3294  * - it is assumed that the hash table must contain an exact power-of-2
3295  *   quantity of entries
3296  * - limit is the number of hash buckets, not the total allocation size
3297  */
3298 void *__init alloc_large_system_hash(const char *tablename,
3299 				     unsigned long bucketsize,
3300 				     unsigned long numentries,
3301 				     int scale,
3302 				     int flags,
3303 				     unsigned int *_hash_shift,
3304 				     unsigned int *_hash_mask,
3305 				     unsigned long limit)
3306 {
3307 	unsigned long long max = limit;
3308 	unsigned long log2qty, size;
3309 	void *table = NULL;
3310 
3311 	/* allow the kernel cmdline to have a say */
3312 	if (!numentries) {
3313 		/* round applicable memory size up to nearest megabyte */
3314 		numentries = nr_kernel_pages;
3315 		numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
3316 		numentries >>= 20 - PAGE_SHIFT;
3317 		numentries <<= 20 - PAGE_SHIFT;
3318 
3319 		/* limit to 1 bucket per 2^scale bytes of low memory */
3320 		if (scale > PAGE_SHIFT)
3321 			numentries >>= (scale - PAGE_SHIFT);
3322 		else
3323 			numentries <<= (PAGE_SHIFT - scale);
3324 
3325 		/* Make sure we've got at least a 0-order allocation.. */
3326 		if (unlikely((numentries * bucketsize) < PAGE_SIZE))
3327 			numentries = PAGE_SIZE / bucketsize;
3328 	}
3329 	numentries = roundup_pow_of_two(numentries);
3330 
3331 	/* limit allocation size to 1/16 total memory by default */
3332 	if (max == 0) {
3333 		max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
3334 		do_div(max, bucketsize);
3335 	}
3336 
3337 	if (numentries > max)
3338 		numentries = max;
3339 
3340 	log2qty = ilog2(numentries);
3341 
3342 	do {
3343 		size = bucketsize << log2qty;
3344 		if (flags & HASH_EARLY)
3345 			table = alloc_bootmem(size);
3346 		else if (hashdist)
3347 			table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
3348 		else {
3349 			unsigned long order;
3350 			for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
3351 				;
3352 			table = (void*) __get_free_pages(GFP_ATOMIC, order);
3353 		}
3354 	} while (!table && size > PAGE_SIZE && --log2qty);
3355 
3356 	if (!table)
3357 		panic("Failed to allocate %s hash table\n", tablename);
3358 
3359 	printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
3360 	       tablename,
3361 	       (1U << log2qty),
3362 	       ilog2(size) - PAGE_SHIFT,
3363 	       size);
3364 
3365 	if (_hash_shift)
3366 		*_hash_shift = log2qty;
3367 	if (_hash_mask)
3368 		*_hash_mask = (1 << log2qty) - 1;
3369 
3370 	return table;
3371 }
3372 
3373 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
3374 struct page *pfn_to_page(unsigned long pfn)
3375 {
3376 	return __pfn_to_page(pfn);
3377 }
3378 unsigned long page_to_pfn(struct page *page)
3379 {
3380 	return __page_to_pfn(page);
3381 }
3382 EXPORT_SYMBOL(pfn_to_page);
3383 EXPORT_SYMBOL(page_to_pfn);
3384 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
3385 
3386 
3387