xref: /openbmc/linux/mm/page_alloc.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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/config.h>
18 #include <linux/stddef.h>
19 #include <linux/mm.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/notifier.h>
32 #include <linux/topology.h>
33 #include <linux/sysctl.h>
34 #include <linux/cpu.h>
35 #include <linux/cpuset.h>
36 #include <linux/memory_hotplug.h>
37 #include <linux/nodemask.h>
38 #include <linux/vmalloc.h>
39 
40 #include <asm/tlbflush.h>
41 #include "internal.h"
42 
43 /*
44  * MCD - HACK: Find somewhere to initialize this EARLY, or make this
45  * initializer cleaner
46  */
47 nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
48 EXPORT_SYMBOL(node_online_map);
49 nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
50 EXPORT_SYMBOL(node_possible_map);
51 struct pglist_data *pgdat_list __read_mostly;
52 unsigned long totalram_pages __read_mostly;
53 unsigned long totalhigh_pages __read_mostly;
54 long nr_swap_pages;
55 
56 /*
57  * results with 256, 32 in the lowmem_reserve sysctl:
58  *	1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
59  *	1G machine -> (16M dma, 784M normal, 224M high)
60  *	NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
61  *	HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
62  *	HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
63  */
64 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 32 };
65 
66 EXPORT_SYMBOL(totalram_pages);
67 EXPORT_SYMBOL(nr_swap_pages);
68 
69 /*
70  * Used by page_zone() to look up the address of the struct zone whose
71  * id is encoded in the upper bits of page->flags
72  */
73 struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
74 EXPORT_SYMBOL(zone_table);
75 
76 static char *zone_names[MAX_NR_ZONES] = { "DMA", "Normal", "HighMem" };
77 int min_free_kbytes = 1024;
78 
79 unsigned long __initdata nr_kernel_pages;
80 unsigned long __initdata nr_all_pages;
81 
82 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
83 {
84 	int ret = 0;
85 	unsigned seq;
86 	unsigned long pfn = page_to_pfn(page);
87 
88 	do {
89 		seq = zone_span_seqbegin(zone);
90 		if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
91 			ret = 1;
92 		else if (pfn < zone->zone_start_pfn)
93 			ret = 1;
94 	} while (zone_span_seqretry(zone, seq));
95 
96 	return ret;
97 }
98 
99 static int page_is_consistent(struct zone *zone, struct page *page)
100 {
101 #ifdef CONFIG_HOLES_IN_ZONE
102 	if (!pfn_valid(page_to_pfn(page)))
103 		return 0;
104 #endif
105 	if (zone != page_zone(page))
106 		return 0;
107 
108 	return 1;
109 }
110 /*
111  * Temporary debugging check for pages not lying within a given zone.
112  */
113 static int bad_range(struct zone *zone, struct page *page)
114 {
115 	if (page_outside_zone_boundaries(zone, page))
116 		return 1;
117 	if (!page_is_consistent(zone, page))
118 		return 1;
119 
120 	return 0;
121 }
122 
123 static void bad_page(const char *function, struct page *page)
124 {
125 	printk(KERN_EMERG "Bad page state at %s (in process '%s', page %p)\n",
126 		function, current->comm, page);
127 	printk(KERN_EMERG "flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
128 		(int)(2*sizeof(page_flags_t)), (unsigned long)page->flags,
129 		page->mapping, page_mapcount(page), page_count(page));
130 	printk(KERN_EMERG "Backtrace:\n");
131 	dump_stack();
132 	printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n");
133 	page->flags &= ~(1 << PG_lru	|
134 			1 << PG_private |
135 			1 << PG_locked	|
136 			1 << PG_active	|
137 			1 << PG_dirty	|
138 			1 << PG_reclaim |
139 			1 << PG_slab    |
140 			1 << PG_swapcache |
141 			1 << PG_writeback |
142 			1 << PG_reserved );
143 	set_page_count(page, 0);
144 	reset_page_mapcount(page);
145 	page->mapping = NULL;
146 	add_taint(TAINT_BAD_PAGE);
147 }
148 
149 #ifndef CONFIG_HUGETLB_PAGE
150 #define prep_compound_page(page, order) do { } while (0)
151 #define destroy_compound_page(page, order) do { } while (0)
152 #else
153 /*
154  * Higher-order pages are called "compound pages".  They are structured thusly:
155  *
156  * The first PAGE_SIZE page is called the "head page".
157  *
158  * The remaining PAGE_SIZE pages are called "tail pages".
159  *
160  * All pages have PG_compound set.  All pages have their ->private pointing at
161  * the head page (even the head page has this).
162  *
163  * The first tail page's ->mapping, if non-zero, holds the address of the
164  * compound page's put_page() function.
165  *
166  * The order of the allocation is stored in the first tail page's ->index
167  * This is only for debug at present.  This usage means that zero-order pages
168  * may not be compound.
169  */
170 static void prep_compound_page(struct page *page, unsigned long order)
171 {
172 	int i;
173 	int nr_pages = 1 << order;
174 
175 	page[1].mapping = NULL;
176 	page[1].index = order;
177 	for (i = 0; i < nr_pages; i++) {
178 		struct page *p = page + i;
179 
180 		SetPageCompound(p);
181 		set_page_private(p, (unsigned long)page);
182 	}
183 }
184 
185 static void destroy_compound_page(struct page *page, unsigned long order)
186 {
187 	int i;
188 	int nr_pages = 1 << order;
189 
190 	if (!PageCompound(page))
191 		return;
192 
193 	if (page[1].index != order)
194 		bad_page(__FUNCTION__, page);
195 
196 	for (i = 0; i < nr_pages; i++) {
197 		struct page *p = page + i;
198 
199 		if (!PageCompound(p))
200 			bad_page(__FUNCTION__, page);
201 		if (page_private(p) != (unsigned long)page)
202 			bad_page(__FUNCTION__, page);
203 		ClearPageCompound(p);
204 	}
205 }
206 #endif		/* CONFIG_HUGETLB_PAGE */
207 
208 /*
209  * function for dealing with page's order in buddy system.
210  * zone->lock is already acquired when we use these.
211  * So, we don't need atomic page->flags operations here.
212  */
213 static inline unsigned long page_order(struct page *page) {
214 	return page_private(page);
215 }
216 
217 static inline void set_page_order(struct page *page, int order) {
218 	set_page_private(page, order);
219 	__SetPagePrivate(page);
220 }
221 
222 static inline void rmv_page_order(struct page *page)
223 {
224 	__ClearPagePrivate(page);
225 	set_page_private(page, 0);
226 }
227 
228 /*
229  * Locate the struct page for both the matching buddy in our
230  * pair (buddy1) and the combined O(n+1) page they form (page).
231  *
232  * 1) Any buddy B1 will have an order O twin B2 which satisfies
233  * the following equation:
234  *     B2 = B1 ^ (1 << O)
235  * For example, if the starting buddy (buddy2) is #8 its order
236  * 1 buddy is #10:
237  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
238  *
239  * 2) Any buddy B will have an order O+1 parent P which
240  * satisfies the following equation:
241  *     P = B & ~(1 << O)
242  *
243  * Assumption: *_mem_map is contigious at least up to MAX_ORDER
244  */
245 static inline struct page *
246 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
247 {
248 	unsigned long buddy_idx = page_idx ^ (1 << order);
249 
250 	return page + (buddy_idx - page_idx);
251 }
252 
253 static inline unsigned long
254 __find_combined_index(unsigned long page_idx, unsigned int order)
255 {
256 	return (page_idx & ~(1 << order));
257 }
258 
259 /*
260  * This function checks whether a page is free && is the buddy
261  * we can do coalesce a page and its buddy if
262  * (a) the buddy is free &&
263  * (b) the buddy is on the buddy system &&
264  * (c) a page and its buddy have the same order.
265  * for recording page's order, we use page_private(page) and PG_private.
266  *
267  */
268 static inline int page_is_buddy(struct page *page, int order)
269 {
270        if (PagePrivate(page)           &&
271            (page_order(page) == order) &&
272             page_count(page) == 0)
273                return 1;
274        return 0;
275 }
276 
277 /*
278  * Freeing function for a buddy system allocator.
279  *
280  * The concept of a buddy system is to maintain direct-mapped table
281  * (containing bit values) for memory blocks of various "orders".
282  * The bottom level table contains the map for the smallest allocatable
283  * units of memory (here, pages), and each level above it describes
284  * pairs of units from the levels below, hence, "buddies".
285  * At a high level, all that happens here is marking the table entry
286  * at the bottom level available, and propagating the changes upward
287  * as necessary, plus some accounting needed to play nicely with other
288  * parts of the VM system.
289  * At each level, we keep a list of pages, which are heads of continuous
290  * free pages of length of (1 << order) and marked with PG_Private.Page's
291  * order is recorded in page_private(page) field.
292  * So when we are allocating or freeing one, we can derive the state of the
293  * other.  That is, if we allocate a small block, and both were
294  * free, the remainder of the region must be split into blocks.
295  * If a block is freed, and its buddy is also free, then this
296  * triggers coalescing into a block of larger size.
297  *
298  * -- wli
299  */
300 
301 static inline void __free_pages_bulk (struct page *page,
302 		struct zone *zone, unsigned int order)
303 {
304 	unsigned long page_idx;
305 	int order_size = 1 << order;
306 
307 	if (unlikely(order))
308 		destroy_compound_page(page, order);
309 
310 	page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
311 
312 	BUG_ON(page_idx & (order_size - 1));
313 	BUG_ON(bad_range(zone, page));
314 
315 	zone->free_pages += order_size;
316 	while (order < MAX_ORDER-1) {
317 		unsigned long combined_idx;
318 		struct free_area *area;
319 		struct page *buddy;
320 
321 		combined_idx = __find_combined_index(page_idx, order);
322 		buddy = __page_find_buddy(page, page_idx, order);
323 
324 		if (bad_range(zone, buddy))
325 			break;
326 		if (!page_is_buddy(buddy, order))
327 			break;		/* Move the buddy up one level. */
328 		list_del(&buddy->lru);
329 		area = zone->free_area + order;
330 		area->nr_free--;
331 		rmv_page_order(buddy);
332 		page = page + (combined_idx - page_idx);
333 		page_idx = combined_idx;
334 		order++;
335 	}
336 	set_page_order(page, order);
337 	list_add(&page->lru, &zone->free_area[order].free_list);
338 	zone->free_area[order].nr_free++;
339 }
340 
341 static inline void free_pages_check(const char *function, struct page *page)
342 {
343 	if (	page_mapcount(page) ||
344 		page->mapping != NULL ||
345 		page_count(page) != 0 ||
346 		(page->flags & (
347 			1 << PG_lru	|
348 			1 << PG_private |
349 			1 << PG_locked	|
350 			1 << PG_active	|
351 			1 << PG_reclaim	|
352 			1 << PG_slab	|
353 			1 << PG_swapcache |
354 			1 << PG_writeback |
355 			1 << PG_reserved )))
356 		bad_page(function, page);
357 	if (PageDirty(page))
358 		__ClearPageDirty(page);
359 }
360 
361 /*
362  * Frees a list of pages.
363  * Assumes all pages on list are in same zone, and of same order.
364  * count is the number of pages to free.
365  *
366  * If the zone was previously in an "all pages pinned" state then look to
367  * see if this freeing clears that state.
368  *
369  * And clear the zone's pages_scanned counter, to hold off the "all pages are
370  * pinned" detection logic.
371  */
372 static int
373 free_pages_bulk(struct zone *zone, int count,
374 		struct list_head *list, unsigned int order)
375 {
376 	unsigned long flags;
377 	struct page *page = NULL;
378 	int ret = 0;
379 
380 	spin_lock_irqsave(&zone->lock, flags);
381 	zone->all_unreclaimable = 0;
382 	zone->pages_scanned = 0;
383 	while (!list_empty(list) && count--) {
384 		page = list_entry(list->prev, struct page, lru);
385 		/* have to delete it as __free_pages_bulk list manipulates */
386 		list_del(&page->lru);
387 		__free_pages_bulk(page, zone, order);
388 		ret++;
389 	}
390 	spin_unlock_irqrestore(&zone->lock, flags);
391 	return ret;
392 }
393 
394 void __free_pages_ok(struct page *page, unsigned int order)
395 {
396 	LIST_HEAD(list);
397 	int i;
398 
399 	arch_free_page(page, order);
400 
401 	mod_page_state(pgfree, 1 << order);
402 
403 #ifndef CONFIG_MMU
404 	if (order > 0)
405 		for (i = 1 ; i < (1 << order) ; ++i)
406 			__put_page(page + i);
407 #endif
408 
409 	for (i = 0 ; i < (1 << order) ; ++i)
410 		free_pages_check(__FUNCTION__, page + i);
411 	list_add(&page->lru, &list);
412 	kernel_map_pages(page, 1<<order, 0);
413 	free_pages_bulk(page_zone(page), 1, &list, order);
414 }
415 
416 
417 /*
418  * The order of subdivision here is critical for the IO subsystem.
419  * Please do not alter this order without good reasons and regression
420  * testing. Specifically, as large blocks of memory are subdivided,
421  * the order in which smaller blocks are delivered depends on the order
422  * they're subdivided in this function. This is the primary factor
423  * influencing the order in which pages are delivered to the IO
424  * subsystem according to empirical testing, and this is also justified
425  * by considering the behavior of a buddy system containing a single
426  * large block of memory acted on by a series of small allocations.
427  * This behavior is a critical factor in sglist merging's success.
428  *
429  * -- wli
430  */
431 static inline struct page *
432 expand(struct zone *zone, struct page *page,
433  	int low, int high, struct free_area *area)
434 {
435 	unsigned long size = 1 << high;
436 
437 	while (high > low) {
438 		area--;
439 		high--;
440 		size >>= 1;
441 		BUG_ON(bad_range(zone, &page[size]));
442 		list_add(&page[size].lru, &area->free_list);
443 		area->nr_free++;
444 		set_page_order(&page[size], high);
445 	}
446 	return page;
447 }
448 
449 void set_page_refs(struct page *page, int order)
450 {
451 #ifdef CONFIG_MMU
452 	set_page_count(page, 1);
453 #else
454 	int i;
455 
456 	/*
457 	 * We need to reference all the pages for this order, otherwise if
458 	 * anyone accesses one of the pages with (get/put) it will be freed.
459 	 * - eg: access_process_vm()
460 	 */
461 	for (i = 0; i < (1 << order); i++)
462 		set_page_count(page + i, 1);
463 #endif /* CONFIG_MMU */
464 }
465 
466 /*
467  * This page is about to be returned from the page allocator
468  */
469 static void prep_new_page(struct page *page, int order)
470 {
471 	if (	page_mapcount(page) ||
472 		page->mapping != NULL ||
473 		page_count(page) != 0 ||
474 		(page->flags & (
475 			1 << PG_lru	|
476 			1 << PG_private	|
477 			1 << PG_locked	|
478 			1 << PG_active	|
479 			1 << PG_dirty	|
480 			1 << PG_reclaim	|
481 			1 << PG_slab    |
482 			1 << PG_swapcache |
483 			1 << PG_writeback |
484 			1 << PG_reserved )))
485 		bad_page(__FUNCTION__, page);
486 
487 	page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
488 			1 << PG_referenced | 1 << PG_arch_1 |
489 			1 << PG_checked | 1 << PG_mappedtodisk);
490 	set_page_private(page, 0);
491 	set_page_refs(page, order);
492 	kernel_map_pages(page, 1 << order, 1);
493 }
494 
495 /*
496  * Do the hard work of removing an element from the buddy allocator.
497  * Call me with the zone->lock already held.
498  */
499 static struct page *__rmqueue(struct zone *zone, unsigned int order)
500 {
501 	struct free_area * area;
502 	unsigned int current_order;
503 	struct page *page;
504 
505 	for (current_order = order; current_order < MAX_ORDER; ++current_order) {
506 		area = zone->free_area + current_order;
507 		if (list_empty(&area->free_list))
508 			continue;
509 
510 		page = list_entry(area->free_list.next, struct page, lru);
511 		list_del(&page->lru);
512 		rmv_page_order(page);
513 		area->nr_free--;
514 		zone->free_pages -= 1UL << order;
515 		return expand(zone, page, order, current_order, area);
516 	}
517 
518 	return NULL;
519 }
520 
521 /*
522  * Obtain a specified number of elements from the buddy allocator, all under
523  * a single hold of the lock, for efficiency.  Add them to the supplied list.
524  * Returns the number of new pages which were placed at *list.
525  */
526 static int rmqueue_bulk(struct zone *zone, unsigned int order,
527 			unsigned long count, struct list_head *list)
528 {
529 	unsigned long flags;
530 	int i;
531 	int allocated = 0;
532 	struct page *page;
533 
534 	spin_lock_irqsave(&zone->lock, flags);
535 	for (i = 0; i < count; ++i) {
536 		page = __rmqueue(zone, order);
537 		if (page == NULL)
538 			break;
539 		allocated++;
540 		list_add_tail(&page->lru, list);
541 	}
542 	spin_unlock_irqrestore(&zone->lock, flags);
543 	return allocated;
544 }
545 
546 #ifdef CONFIG_NUMA
547 /* Called from the slab reaper to drain remote pagesets */
548 void drain_remote_pages(void)
549 {
550 	struct zone *zone;
551 	int i;
552 	unsigned long flags;
553 
554 	local_irq_save(flags);
555 	for_each_zone(zone) {
556 		struct per_cpu_pageset *pset;
557 
558 		/* Do not drain local pagesets */
559 		if (zone->zone_pgdat->node_id == numa_node_id())
560 			continue;
561 
562 		pset = zone->pageset[smp_processor_id()];
563 		for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
564 			struct per_cpu_pages *pcp;
565 
566 			pcp = &pset->pcp[i];
567 			if (pcp->count)
568 				pcp->count -= free_pages_bulk(zone, pcp->count,
569 						&pcp->list, 0);
570 		}
571 	}
572 	local_irq_restore(flags);
573 }
574 #endif
575 
576 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
577 static void __drain_pages(unsigned int cpu)
578 {
579 	struct zone *zone;
580 	int i;
581 
582 	for_each_zone(zone) {
583 		struct per_cpu_pageset *pset;
584 
585 		pset = zone_pcp(zone, cpu);
586 		for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
587 			struct per_cpu_pages *pcp;
588 
589 			pcp = &pset->pcp[i];
590 			pcp->count -= free_pages_bulk(zone, pcp->count,
591 						&pcp->list, 0);
592 		}
593 	}
594 }
595 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
596 
597 #ifdef CONFIG_PM
598 
599 void mark_free_pages(struct zone *zone)
600 {
601 	unsigned long zone_pfn, flags;
602 	int order;
603 	struct list_head *curr;
604 
605 	if (!zone->spanned_pages)
606 		return;
607 
608 	spin_lock_irqsave(&zone->lock, flags);
609 	for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
610 		ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
611 
612 	for (order = MAX_ORDER - 1; order >= 0; --order)
613 		list_for_each(curr, &zone->free_area[order].free_list) {
614 			unsigned long start_pfn, i;
615 
616 			start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
617 
618 			for (i=0; i < (1<<order); i++)
619 				SetPageNosaveFree(pfn_to_page(start_pfn+i));
620 	}
621 	spin_unlock_irqrestore(&zone->lock, flags);
622 }
623 
624 /*
625  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
626  */
627 void drain_local_pages(void)
628 {
629 	unsigned long flags;
630 
631 	local_irq_save(flags);
632 	__drain_pages(smp_processor_id());
633 	local_irq_restore(flags);
634 }
635 #endif /* CONFIG_PM */
636 
637 static void zone_statistics(struct zonelist *zonelist, struct zone *z)
638 {
639 #ifdef CONFIG_NUMA
640 	unsigned long flags;
641 	int cpu;
642 	pg_data_t *pg = z->zone_pgdat;
643 	pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
644 	struct per_cpu_pageset *p;
645 
646 	local_irq_save(flags);
647 	cpu = smp_processor_id();
648 	p = zone_pcp(z,cpu);
649 	if (pg == orig) {
650 		p->numa_hit++;
651 	} else {
652 		p->numa_miss++;
653 		zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
654 	}
655 	if (pg == NODE_DATA(numa_node_id()))
656 		p->local_node++;
657 	else
658 		p->other_node++;
659 	local_irq_restore(flags);
660 #endif
661 }
662 
663 /*
664  * Free a 0-order page
665  */
666 static void FASTCALL(free_hot_cold_page(struct page *page, int cold));
667 static void fastcall free_hot_cold_page(struct page *page, int cold)
668 {
669 	struct zone *zone = page_zone(page);
670 	struct per_cpu_pages *pcp;
671 	unsigned long flags;
672 
673 	arch_free_page(page, 0);
674 
675 	kernel_map_pages(page, 1, 0);
676 	inc_page_state(pgfree);
677 	if (PageAnon(page))
678 		page->mapping = NULL;
679 	free_pages_check(__FUNCTION__, page);
680 	pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
681 	local_irq_save(flags);
682 	list_add(&page->lru, &pcp->list);
683 	pcp->count++;
684 	if (pcp->count >= pcp->high)
685 		pcp->count -= free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
686 	local_irq_restore(flags);
687 	put_cpu();
688 }
689 
690 void fastcall free_hot_page(struct page *page)
691 {
692 	free_hot_cold_page(page, 0);
693 }
694 
695 void fastcall free_cold_page(struct page *page)
696 {
697 	free_hot_cold_page(page, 1);
698 }
699 
700 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
701 {
702 	int i;
703 
704 	BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
705 	for(i = 0; i < (1 << order); i++)
706 		clear_highpage(page + i);
707 }
708 
709 /*
710  * Really, prep_compound_page() should be called from __rmqueue_bulk().  But
711  * we cheat by calling it from here, in the order > 0 path.  Saves a branch
712  * or two.
713  */
714 static struct page *
715 buffered_rmqueue(struct zone *zone, int order, gfp_t gfp_flags)
716 {
717 	unsigned long flags;
718 	struct page *page = NULL;
719 	int cold = !!(gfp_flags & __GFP_COLD);
720 
721 	if (order == 0) {
722 		struct per_cpu_pages *pcp;
723 
724 		pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
725 		local_irq_save(flags);
726 		if (pcp->count <= pcp->low)
727 			pcp->count += rmqueue_bulk(zone, 0,
728 						pcp->batch, &pcp->list);
729 		if (pcp->count) {
730 			page = list_entry(pcp->list.next, struct page, lru);
731 			list_del(&page->lru);
732 			pcp->count--;
733 		}
734 		local_irq_restore(flags);
735 		put_cpu();
736 	}
737 
738 	if (page == NULL) {
739 		spin_lock_irqsave(&zone->lock, flags);
740 		page = __rmqueue(zone, order);
741 		spin_unlock_irqrestore(&zone->lock, flags);
742 	}
743 
744 	if (page != NULL) {
745 		BUG_ON(bad_range(zone, page));
746 		mod_page_state_zone(zone, pgalloc, 1 << order);
747 		prep_new_page(page, order);
748 
749 		if (gfp_flags & __GFP_ZERO)
750 			prep_zero_page(page, order, gfp_flags);
751 
752 		if (order && (gfp_flags & __GFP_COMP))
753 			prep_compound_page(page, order);
754 	}
755 	return page;
756 }
757 
758 /*
759  * Return 1 if free pages are above 'mark'. This takes into account the order
760  * of the allocation.
761  */
762 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
763 		      int classzone_idx, int can_try_harder, gfp_t gfp_high)
764 {
765 	/* free_pages my go negative - that's OK */
766 	long min = mark, free_pages = z->free_pages - (1 << order) + 1;
767 	int o;
768 
769 	if (gfp_high)
770 		min -= min / 2;
771 	if (can_try_harder)
772 		min -= min / 4;
773 
774 	if (free_pages <= min + z->lowmem_reserve[classzone_idx])
775 		return 0;
776 	for (o = 0; o < order; o++) {
777 		/* At the next order, this order's pages become unavailable */
778 		free_pages -= z->free_area[o].nr_free << o;
779 
780 		/* Require fewer higher order pages to be free */
781 		min >>= 1;
782 
783 		if (free_pages <= min)
784 			return 0;
785 	}
786 	return 1;
787 }
788 
789 static inline int
790 should_reclaim_zone(struct zone *z, gfp_t gfp_mask)
791 {
792 	if (!z->reclaim_pages)
793 		return 0;
794 	if (gfp_mask & __GFP_NORECLAIM)
795 		return 0;
796 	return 1;
797 }
798 
799 /*
800  * This is the 'heart' of the zoned buddy allocator.
801  */
802 struct page * fastcall
803 __alloc_pages(gfp_t gfp_mask, unsigned int order,
804 		struct zonelist *zonelist)
805 {
806 	const gfp_t wait = gfp_mask & __GFP_WAIT;
807 	struct zone **zones, *z;
808 	struct page *page;
809 	struct reclaim_state reclaim_state;
810 	struct task_struct *p = current;
811 	int i;
812 	int classzone_idx;
813 	int do_retry;
814 	int can_try_harder;
815 	int did_some_progress;
816 
817 	might_sleep_if(wait);
818 
819 	/*
820 	 * The caller may dip into page reserves a bit more if the caller
821 	 * cannot run direct reclaim, or is the caller has realtime scheduling
822 	 * policy
823 	 */
824 	can_try_harder = (unlikely(rt_task(p)) && !in_interrupt()) || !wait;
825 
826 	zones = zonelist->zones;  /* the list of zones suitable for gfp_mask */
827 
828 	if (unlikely(zones[0] == NULL)) {
829 		/* Should this ever happen?? */
830 		return NULL;
831 	}
832 
833 	classzone_idx = zone_idx(zones[0]);
834 
835 restart:
836 	/*
837 	 * Go through the zonelist once, looking for a zone with enough free.
838 	 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
839 	 */
840 	for (i = 0; (z = zones[i]) != NULL; i++) {
841 		int do_reclaim = should_reclaim_zone(z, gfp_mask);
842 
843 		if (!cpuset_zone_allowed(z, __GFP_HARDWALL))
844 			continue;
845 
846 		/*
847 		 * If the zone is to attempt early page reclaim then this loop
848 		 * will try to reclaim pages and check the watermark a second
849 		 * time before giving up and falling back to the next zone.
850 		 */
851 zone_reclaim_retry:
852 		if (!zone_watermark_ok(z, order, z->pages_low,
853 				       classzone_idx, 0, 0)) {
854 			if (!do_reclaim)
855 				continue;
856 			else {
857 				zone_reclaim(z, gfp_mask, order);
858 				/* Only try reclaim once */
859 				do_reclaim = 0;
860 				goto zone_reclaim_retry;
861 			}
862 		}
863 
864 		page = buffered_rmqueue(z, order, gfp_mask);
865 		if (page)
866 			goto got_pg;
867 	}
868 
869 	for (i = 0; (z = zones[i]) != NULL; i++)
870 		wakeup_kswapd(z, order);
871 
872 	/*
873 	 * Go through the zonelist again. Let __GFP_HIGH and allocations
874 	 * coming from realtime tasks to go deeper into reserves
875 	 *
876 	 * This is the last chance, in general, before the goto nopage.
877 	 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
878 	 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
879 	 */
880 	for (i = 0; (z = zones[i]) != NULL; i++) {
881 		if (!zone_watermark_ok(z, order, z->pages_min,
882 				       classzone_idx, can_try_harder,
883 				       gfp_mask & __GFP_HIGH))
884 			continue;
885 
886 		if (wait && !cpuset_zone_allowed(z, gfp_mask))
887 			continue;
888 
889 		page = buffered_rmqueue(z, order, gfp_mask);
890 		if (page)
891 			goto got_pg;
892 	}
893 
894 	/* This allocation should allow future memory freeing. */
895 
896 	if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
897 			&& !in_interrupt()) {
898 		if (!(gfp_mask & __GFP_NOMEMALLOC)) {
899 			/* go through the zonelist yet again, ignoring mins */
900 			for (i = 0; (z = zones[i]) != NULL; i++) {
901 				if (!cpuset_zone_allowed(z, gfp_mask))
902 					continue;
903 				page = buffered_rmqueue(z, order, gfp_mask);
904 				if (page)
905 					goto got_pg;
906 			}
907 		}
908 		goto nopage;
909 	}
910 
911 	/* Atomic allocations - we can't balance anything */
912 	if (!wait)
913 		goto nopage;
914 
915 rebalance:
916 	cond_resched();
917 
918 	/* We now go into synchronous reclaim */
919 	p->flags |= PF_MEMALLOC;
920 	reclaim_state.reclaimed_slab = 0;
921 	p->reclaim_state = &reclaim_state;
922 
923 	did_some_progress = try_to_free_pages(zones, gfp_mask);
924 
925 	p->reclaim_state = NULL;
926 	p->flags &= ~PF_MEMALLOC;
927 
928 	cond_resched();
929 
930 	if (likely(did_some_progress)) {
931 		for (i = 0; (z = zones[i]) != NULL; i++) {
932 			if (!zone_watermark_ok(z, order, z->pages_min,
933 					       classzone_idx, can_try_harder,
934 					       gfp_mask & __GFP_HIGH))
935 				continue;
936 
937 			if (!cpuset_zone_allowed(z, gfp_mask))
938 				continue;
939 
940 			page = buffered_rmqueue(z, order, gfp_mask);
941 			if (page)
942 				goto got_pg;
943 		}
944 	} else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
945 		/*
946 		 * Go through the zonelist yet one more time, keep
947 		 * very high watermark here, this is only to catch
948 		 * a parallel oom killing, we must fail if we're still
949 		 * under heavy pressure.
950 		 */
951 		for (i = 0; (z = zones[i]) != NULL; i++) {
952 			if (!zone_watermark_ok(z, order, z->pages_high,
953 					       classzone_idx, 0, 0))
954 				continue;
955 
956 			if (!cpuset_zone_allowed(z, __GFP_HARDWALL))
957 				continue;
958 
959 			page = buffered_rmqueue(z, order, gfp_mask);
960 			if (page)
961 				goto got_pg;
962 		}
963 
964 		out_of_memory(gfp_mask, order);
965 		goto restart;
966 	}
967 
968 	/*
969 	 * Don't let big-order allocations loop unless the caller explicitly
970 	 * requests that.  Wait for some write requests to complete then retry.
971 	 *
972 	 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
973 	 * <= 3, but that may not be true in other implementations.
974 	 */
975 	do_retry = 0;
976 	if (!(gfp_mask & __GFP_NORETRY)) {
977 		if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
978 			do_retry = 1;
979 		if (gfp_mask & __GFP_NOFAIL)
980 			do_retry = 1;
981 	}
982 	if (do_retry) {
983 		blk_congestion_wait(WRITE, HZ/50);
984 		goto rebalance;
985 	}
986 
987 nopage:
988 	if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
989 		printk(KERN_WARNING "%s: page allocation failure."
990 			" order:%d, mode:0x%x\n",
991 			p->comm, order, gfp_mask);
992 		dump_stack();
993 		show_mem();
994 	}
995 	return NULL;
996 got_pg:
997 	zone_statistics(zonelist, z);
998 	return page;
999 }
1000 
1001 EXPORT_SYMBOL(__alloc_pages);
1002 
1003 /*
1004  * Common helper functions.
1005  */
1006 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1007 {
1008 	struct page * page;
1009 	page = alloc_pages(gfp_mask, order);
1010 	if (!page)
1011 		return 0;
1012 	return (unsigned long) page_address(page);
1013 }
1014 
1015 EXPORT_SYMBOL(__get_free_pages);
1016 
1017 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1018 {
1019 	struct page * page;
1020 
1021 	/*
1022 	 * get_zeroed_page() returns a 32-bit address, which cannot represent
1023 	 * a highmem page
1024 	 */
1025 	BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1026 
1027 	page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1028 	if (page)
1029 		return (unsigned long) page_address(page);
1030 	return 0;
1031 }
1032 
1033 EXPORT_SYMBOL(get_zeroed_page);
1034 
1035 void __pagevec_free(struct pagevec *pvec)
1036 {
1037 	int i = pagevec_count(pvec);
1038 
1039 	while (--i >= 0)
1040 		free_hot_cold_page(pvec->pages[i], pvec->cold);
1041 }
1042 
1043 fastcall void __free_pages(struct page *page, unsigned int order)
1044 {
1045 	if (put_page_testzero(page)) {
1046 		if (order == 0)
1047 			free_hot_page(page);
1048 		else
1049 			__free_pages_ok(page, order);
1050 	}
1051 }
1052 
1053 EXPORT_SYMBOL(__free_pages);
1054 
1055 fastcall void free_pages(unsigned long addr, unsigned int order)
1056 {
1057 	if (addr != 0) {
1058 		BUG_ON(!virt_addr_valid((void *)addr));
1059 		__free_pages(virt_to_page((void *)addr), order);
1060 	}
1061 }
1062 
1063 EXPORT_SYMBOL(free_pages);
1064 
1065 /*
1066  * Total amount of free (allocatable) RAM:
1067  */
1068 unsigned int nr_free_pages(void)
1069 {
1070 	unsigned int sum = 0;
1071 	struct zone *zone;
1072 
1073 	for_each_zone(zone)
1074 		sum += zone->free_pages;
1075 
1076 	return sum;
1077 }
1078 
1079 EXPORT_SYMBOL(nr_free_pages);
1080 
1081 #ifdef CONFIG_NUMA
1082 unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1083 {
1084 	unsigned int i, sum = 0;
1085 
1086 	for (i = 0; i < MAX_NR_ZONES; i++)
1087 		sum += pgdat->node_zones[i].free_pages;
1088 
1089 	return sum;
1090 }
1091 #endif
1092 
1093 static unsigned int nr_free_zone_pages(int offset)
1094 {
1095 	/* Just pick one node, since fallback list is circular */
1096 	pg_data_t *pgdat = NODE_DATA(numa_node_id());
1097 	unsigned int sum = 0;
1098 
1099 	struct zonelist *zonelist = pgdat->node_zonelists + offset;
1100 	struct zone **zonep = zonelist->zones;
1101 	struct zone *zone;
1102 
1103 	for (zone = *zonep++; zone; zone = *zonep++) {
1104 		unsigned long size = zone->present_pages;
1105 		unsigned long high = zone->pages_high;
1106 		if (size > high)
1107 			sum += size - high;
1108 	}
1109 
1110 	return sum;
1111 }
1112 
1113 /*
1114  * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1115  */
1116 unsigned int nr_free_buffer_pages(void)
1117 {
1118 	return nr_free_zone_pages(gfp_zone(GFP_USER));
1119 }
1120 
1121 /*
1122  * Amount of free RAM allocatable within all zones
1123  */
1124 unsigned int nr_free_pagecache_pages(void)
1125 {
1126 	return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
1127 }
1128 
1129 #ifdef CONFIG_HIGHMEM
1130 unsigned int nr_free_highpages (void)
1131 {
1132 	pg_data_t *pgdat;
1133 	unsigned int pages = 0;
1134 
1135 	for_each_pgdat(pgdat)
1136 		pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1137 
1138 	return pages;
1139 }
1140 #endif
1141 
1142 #ifdef CONFIG_NUMA
1143 static void show_node(struct zone *zone)
1144 {
1145 	printk("Node %d ", zone->zone_pgdat->node_id);
1146 }
1147 #else
1148 #define show_node(zone)	do { } while (0)
1149 #endif
1150 
1151 /*
1152  * Accumulate the page_state information across all CPUs.
1153  * The result is unavoidably approximate - it can change
1154  * during and after execution of this function.
1155  */
1156 static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1157 
1158 atomic_t nr_pagecache = ATOMIC_INIT(0);
1159 EXPORT_SYMBOL(nr_pagecache);
1160 #ifdef CONFIG_SMP
1161 DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1162 #endif
1163 
1164 void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
1165 {
1166 	int cpu = 0;
1167 
1168 	memset(ret, 0, sizeof(*ret));
1169 	cpus_and(*cpumask, *cpumask, cpu_online_map);
1170 
1171 	cpu = first_cpu(*cpumask);
1172 	while (cpu < NR_CPUS) {
1173 		unsigned long *in, *out, off;
1174 
1175 		in = (unsigned long *)&per_cpu(page_states, cpu);
1176 
1177 		cpu = next_cpu(cpu, *cpumask);
1178 
1179 		if (cpu < NR_CPUS)
1180 			prefetch(&per_cpu(page_states, cpu));
1181 
1182 		out = (unsigned long *)ret;
1183 		for (off = 0; off < nr; off++)
1184 			*out++ += *in++;
1185 	}
1186 }
1187 
1188 void get_page_state_node(struct page_state *ret, int node)
1189 {
1190 	int nr;
1191 	cpumask_t mask = node_to_cpumask(node);
1192 
1193 	nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1194 	nr /= sizeof(unsigned long);
1195 
1196 	__get_page_state(ret, nr+1, &mask);
1197 }
1198 
1199 void get_page_state(struct page_state *ret)
1200 {
1201 	int nr;
1202 	cpumask_t mask = CPU_MASK_ALL;
1203 
1204 	nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1205 	nr /= sizeof(unsigned long);
1206 
1207 	__get_page_state(ret, nr + 1, &mask);
1208 }
1209 
1210 void get_full_page_state(struct page_state *ret)
1211 {
1212 	cpumask_t mask = CPU_MASK_ALL;
1213 
1214 	__get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
1215 }
1216 
1217 unsigned long __read_page_state(unsigned long offset)
1218 {
1219 	unsigned long ret = 0;
1220 	int cpu;
1221 
1222 	for_each_online_cpu(cpu) {
1223 		unsigned long in;
1224 
1225 		in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1226 		ret += *((unsigned long *)in);
1227 	}
1228 	return ret;
1229 }
1230 
1231 void __mod_page_state(unsigned long offset, unsigned long delta)
1232 {
1233 	unsigned long flags;
1234 	void* ptr;
1235 
1236 	local_irq_save(flags);
1237 	ptr = &__get_cpu_var(page_states);
1238 	*(unsigned long*)(ptr + offset) += delta;
1239 	local_irq_restore(flags);
1240 }
1241 
1242 EXPORT_SYMBOL(__mod_page_state);
1243 
1244 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1245 			unsigned long *free, struct pglist_data *pgdat)
1246 {
1247 	struct zone *zones = pgdat->node_zones;
1248 	int i;
1249 
1250 	*active = 0;
1251 	*inactive = 0;
1252 	*free = 0;
1253 	for (i = 0; i < MAX_NR_ZONES; i++) {
1254 		*active += zones[i].nr_active;
1255 		*inactive += zones[i].nr_inactive;
1256 		*free += zones[i].free_pages;
1257 	}
1258 }
1259 
1260 void get_zone_counts(unsigned long *active,
1261 		unsigned long *inactive, unsigned long *free)
1262 {
1263 	struct pglist_data *pgdat;
1264 
1265 	*active = 0;
1266 	*inactive = 0;
1267 	*free = 0;
1268 	for_each_pgdat(pgdat) {
1269 		unsigned long l, m, n;
1270 		__get_zone_counts(&l, &m, &n, pgdat);
1271 		*active += l;
1272 		*inactive += m;
1273 		*free += n;
1274 	}
1275 }
1276 
1277 void si_meminfo(struct sysinfo *val)
1278 {
1279 	val->totalram = totalram_pages;
1280 	val->sharedram = 0;
1281 	val->freeram = nr_free_pages();
1282 	val->bufferram = nr_blockdev_pages();
1283 #ifdef CONFIG_HIGHMEM
1284 	val->totalhigh = totalhigh_pages;
1285 	val->freehigh = nr_free_highpages();
1286 #else
1287 	val->totalhigh = 0;
1288 	val->freehigh = 0;
1289 #endif
1290 	val->mem_unit = PAGE_SIZE;
1291 }
1292 
1293 EXPORT_SYMBOL(si_meminfo);
1294 
1295 #ifdef CONFIG_NUMA
1296 void si_meminfo_node(struct sysinfo *val, int nid)
1297 {
1298 	pg_data_t *pgdat = NODE_DATA(nid);
1299 
1300 	val->totalram = pgdat->node_present_pages;
1301 	val->freeram = nr_free_pages_pgdat(pgdat);
1302 	val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1303 	val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1304 	val->mem_unit = PAGE_SIZE;
1305 }
1306 #endif
1307 
1308 #define K(x) ((x) << (PAGE_SHIFT-10))
1309 
1310 /*
1311  * Show free area list (used inside shift_scroll-lock stuff)
1312  * We also calculate the percentage fragmentation. We do this by counting the
1313  * memory on each free list with the exception of the first item on the list.
1314  */
1315 void show_free_areas(void)
1316 {
1317 	struct page_state ps;
1318 	int cpu, temperature;
1319 	unsigned long active;
1320 	unsigned long inactive;
1321 	unsigned long free;
1322 	struct zone *zone;
1323 
1324 	for_each_zone(zone) {
1325 		show_node(zone);
1326 		printk("%s per-cpu:", zone->name);
1327 
1328 		if (!zone->present_pages) {
1329 			printk(" empty\n");
1330 			continue;
1331 		} else
1332 			printk("\n");
1333 
1334 		for_each_cpu(cpu) {
1335 			struct per_cpu_pageset *pageset;
1336 
1337 			pageset = zone_pcp(zone, cpu);
1338 
1339 			for (temperature = 0; temperature < 2; temperature++)
1340 				printk("cpu %d %s: low %d, high %d, batch %d used:%d\n",
1341 					cpu,
1342 					temperature ? "cold" : "hot",
1343 					pageset->pcp[temperature].low,
1344 					pageset->pcp[temperature].high,
1345 					pageset->pcp[temperature].batch,
1346 					pageset->pcp[temperature].count);
1347 		}
1348 	}
1349 
1350 	get_page_state(&ps);
1351 	get_zone_counts(&active, &inactive, &free);
1352 
1353 	printk("Free pages: %11ukB (%ukB HighMem)\n",
1354 		K(nr_free_pages()),
1355 		K(nr_free_highpages()));
1356 
1357 	printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1358 		"unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1359 		active,
1360 		inactive,
1361 		ps.nr_dirty,
1362 		ps.nr_writeback,
1363 		ps.nr_unstable,
1364 		nr_free_pages(),
1365 		ps.nr_slab,
1366 		ps.nr_mapped,
1367 		ps.nr_page_table_pages);
1368 
1369 	for_each_zone(zone) {
1370 		int i;
1371 
1372 		show_node(zone);
1373 		printk("%s"
1374 			" free:%lukB"
1375 			" min:%lukB"
1376 			" low:%lukB"
1377 			" high:%lukB"
1378 			" active:%lukB"
1379 			" inactive:%lukB"
1380 			" present:%lukB"
1381 			" pages_scanned:%lu"
1382 			" all_unreclaimable? %s"
1383 			"\n",
1384 			zone->name,
1385 			K(zone->free_pages),
1386 			K(zone->pages_min),
1387 			K(zone->pages_low),
1388 			K(zone->pages_high),
1389 			K(zone->nr_active),
1390 			K(zone->nr_inactive),
1391 			K(zone->present_pages),
1392 			zone->pages_scanned,
1393 			(zone->all_unreclaimable ? "yes" : "no")
1394 			);
1395 		printk("lowmem_reserve[]:");
1396 		for (i = 0; i < MAX_NR_ZONES; i++)
1397 			printk(" %lu", zone->lowmem_reserve[i]);
1398 		printk("\n");
1399 	}
1400 
1401 	for_each_zone(zone) {
1402  		unsigned long nr, flags, order, total = 0;
1403 
1404 		show_node(zone);
1405 		printk("%s: ", zone->name);
1406 		if (!zone->present_pages) {
1407 			printk("empty\n");
1408 			continue;
1409 		}
1410 
1411 		spin_lock_irqsave(&zone->lock, flags);
1412 		for (order = 0; order < MAX_ORDER; order++) {
1413 			nr = zone->free_area[order].nr_free;
1414 			total += nr << order;
1415 			printk("%lu*%lukB ", nr, K(1UL) << order);
1416 		}
1417 		spin_unlock_irqrestore(&zone->lock, flags);
1418 		printk("= %lukB\n", K(total));
1419 	}
1420 
1421 	show_swap_cache_info();
1422 }
1423 
1424 /*
1425  * Builds allocation fallback zone lists.
1426  */
1427 static int __init build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, int j, int k)
1428 {
1429 	switch (k) {
1430 		struct zone *zone;
1431 	default:
1432 		BUG();
1433 	case ZONE_HIGHMEM:
1434 		zone = pgdat->node_zones + ZONE_HIGHMEM;
1435 		if (zone->present_pages) {
1436 #ifndef CONFIG_HIGHMEM
1437 			BUG();
1438 #endif
1439 			zonelist->zones[j++] = zone;
1440 		}
1441 	case ZONE_NORMAL:
1442 		zone = pgdat->node_zones + ZONE_NORMAL;
1443 		if (zone->present_pages)
1444 			zonelist->zones[j++] = zone;
1445 	case ZONE_DMA:
1446 		zone = pgdat->node_zones + ZONE_DMA;
1447 		if (zone->present_pages)
1448 			zonelist->zones[j++] = zone;
1449 	}
1450 
1451 	return j;
1452 }
1453 
1454 static inline int highest_zone(int zone_bits)
1455 {
1456 	int res = ZONE_NORMAL;
1457 	if (zone_bits & (__force int)__GFP_HIGHMEM)
1458 		res = ZONE_HIGHMEM;
1459 	if (zone_bits & (__force int)__GFP_DMA)
1460 		res = ZONE_DMA;
1461 	return res;
1462 }
1463 
1464 #ifdef CONFIG_NUMA
1465 #define MAX_NODE_LOAD (num_online_nodes())
1466 static int __initdata node_load[MAX_NUMNODES];
1467 /**
1468  * find_next_best_node - find the next node that should appear in a given node's fallback list
1469  * @node: node whose fallback list we're appending
1470  * @used_node_mask: nodemask_t of already used nodes
1471  *
1472  * We use a number of factors to determine which is the next node that should
1473  * appear on a given node's fallback list.  The node should not have appeared
1474  * already in @node's fallback list, and it should be the next closest node
1475  * according to the distance array (which contains arbitrary distance values
1476  * from each node to each node in the system), and should also prefer nodes
1477  * with no CPUs, since presumably they'll have very little allocation pressure
1478  * on them otherwise.
1479  * It returns -1 if no node is found.
1480  */
1481 static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1482 {
1483 	int i, n, val;
1484 	int min_val = INT_MAX;
1485 	int best_node = -1;
1486 
1487 	for_each_online_node(i) {
1488 		cpumask_t tmp;
1489 
1490 		/* Start from local node */
1491 		n = (node+i) % num_online_nodes();
1492 
1493 		/* Don't want a node to appear more than once */
1494 		if (node_isset(n, *used_node_mask))
1495 			continue;
1496 
1497 		/* Use the local node if we haven't already */
1498 		if (!node_isset(node, *used_node_mask)) {
1499 			best_node = node;
1500 			break;
1501 		}
1502 
1503 		/* Use the distance array to find the distance */
1504 		val = node_distance(node, n);
1505 
1506 		/* Give preference to headless and unused nodes */
1507 		tmp = node_to_cpumask(n);
1508 		if (!cpus_empty(tmp))
1509 			val += PENALTY_FOR_NODE_WITH_CPUS;
1510 
1511 		/* Slight preference for less loaded node */
1512 		val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1513 		val += node_load[n];
1514 
1515 		if (val < min_val) {
1516 			min_val = val;
1517 			best_node = n;
1518 		}
1519 	}
1520 
1521 	if (best_node >= 0)
1522 		node_set(best_node, *used_node_mask);
1523 
1524 	return best_node;
1525 }
1526 
1527 static void __init build_zonelists(pg_data_t *pgdat)
1528 {
1529 	int i, j, k, node, local_node;
1530 	int prev_node, load;
1531 	struct zonelist *zonelist;
1532 	nodemask_t used_mask;
1533 
1534 	/* initialize zonelists */
1535 	for (i = 0; i < GFP_ZONETYPES; i++) {
1536 		zonelist = pgdat->node_zonelists + i;
1537 		zonelist->zones[0] = NULL;
1538 	}
1539 
1540 	/* NUMA-aware ordering of nodes */
1541 	local_node = pgdat->node_id;
1542 	load = num_online_nodes();
1543 	prev_node = local_node;
1544 	nodes_clear(used_mask);
1545 	while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1546 		/*
1547 		 * We don't want to pressure a particular node.
1548 		 * So adding penalty to the first node in same
1549 		 * distance group to make it round-robin.
1550 		 */
1551 		if (node_distance(local_node, node) !=
1552 				node_distance(local_node, prev_node))
1553 			node_load[node] += load;
1554 		prev_node = node;
1555 		load--;
1556 		for (i = 0; i < GFP_ZONETYPES; i++) {
1557 			zonelist = pgdat->node_zonelists + i;
1558 			for (j = 0; zonelist->zones[j] != NULL; j++);
1559 
1560 			k = highest_zone(i);
1561 
1562 	 		j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1563 			zonelist->zones[j] = NULL;
1564 		}
1565 	}
1566 }
1567 
1568 #else	/* CONFIG_NUMA */
1569 
1570 static void __init build_zonelists(pg_data_t *pgdat)
1571 {
1572 	int i, j, k, node, local_node;
1573 
1574 	local_node = pgdat->node_id;
1575 	for (i = 0; i < GFP_ZONETYPES; i++) {
1576 		struct zonelist *zonelist;
1577 
1578 		zonelist = pgdat->node_zonelists + i;
1579 
1580 		j = 0;
1581 		k = highest_zone(i);
1582  		j = build_zonelists_node(pgdat, zonelist, j, k);
1583  		/*
1584  		 * Now we build the zonelist so that it contains the zones
1585  		 * of all the other nodes.
1586  		 * We don't want to pressure a particular node, so when
1587  		 * building the zones for node N, we make sure that the
1588  		 * zones coming right after the local ones are those from
1589  		 * node N+1 (modulo N)
1590  		 */
1591 		for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1592 			if (!node_online(node))
1593 				continue;
1594 			j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1595 		}
1596 		for (node = 0; node < local_node; node++) {
1597 			if (!node_online(node))
1598 				continue;
1599 			j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1600 		}
1601 
1602 		zonelist->zones[j] = NULL;
1603 	}
1604 }
1605 
1606 #endif	/* CONFIG_NUMA */
1607 
1608 void __init build_all_zonelists(void)
1609 {
1610 	int i;
1611 
1612 	for_each_online_node(i)
1613 		build_zonelists(NODE_DATA(i));
1614 	printk("Built %i zonelists\n", num_online_nodes());
1615 	cpuset_init_current_mems_allowed();
1616 }
1617 
1618 /*
1619  * Helper functions to size the waitqueue hash table.
1620  * Essentially these want to choose hash table sizes sufficiently
1621  * large so that collisions trying to wait on pages are rare.
1622  * But in fact, the number of active page waitqueues on typical
1623  * systems is ridiculously low, less than 200. So this is even
1624  * conservative, even though it seems large.
1625  *
1626  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1627  * waitqueues, i.e. the size of the waitq table given the number of pages.
1628  */
1629 #define PAGES_PER_WAITQUEUE	256
1630 
1631 static inline unsigned long wait_table_size(unsigned long pages)
1632 {
1633 	unsigned long size = 1;
1634 
1635 	pages /= PAGES_PER_WAITQUEUE;
1636 
1637 	while (size < pages)
1638 		size <<= 1;
1639 
1640 	/*
1641 	 * Once we have dozens or even hundreds of threads sleeping
1642 	 * on IO we've got bigger problems than wait queue collision.
1643 	 * Limit the size of the wait table to a reasonable size.
1644 	 */
1645 	size = min(size, 4096UL);
1646 
1647 	return max(size, 4UL);
1648 }
1649 
1650 /*
1651  * This is an integer logarithm so that shifts can be used later
1652  * to extract the more random high bits from the multiplicative
1653  * hash function before the remainder is taken.
1654  */
1655 static inline unsigned long wait_table_bits(unsigned long size)
1656 {
1657 	return ffz(~size);
1658 }
1659 
1660 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1661 
1662 static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1663 		unsigned long *zones_size, unsigned long *zholes_size)
1664 {
1665 	unsigned long realtotalpages, totalpages = 0;
1666 	int i;
1667 
1668 	for (i = 0; i < MAX_NR_ZONES; i++)
1669 		totalpages += zones_size[i];
1670 	pgdat->node_spanned_pages = totalpages;
1671 
1672 	realtotalpages = totalpages;
1673 	if (zholes_size)
1674 		for (i = 0; i < MAX_NR_ZONES; i++)
1675 			realtotalpages -= zholes_size[i];
1676 	pgdat->node_present_pages = realtotalpages;
1677 	printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1678 }
1679 
1680 
1681 /*
1682  * Initially all pages are reserved - free ones are freed
1683  * up by free_all_bootmem() once the early boot process is
1684  * done. Non-atomic initialization, single-pass.
1685  */
1686 void __devinit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1687 		unsigned long start_pfn)
1688 {
1689 	struct page *page;
1690 	unsigned long end_pfn = start_pfn + size;
1691 	unsigned long pfn;
1692 
1693 	for (pfn = start_pfn; pfn < end_pfn; pfn++, page++) {
1694 		if (!early_pfn_valid(pfn))
1695 			continue;
1696 		if (!early_pfn_in_nid(pfn, nid))
1697 			continue;
1698 		page = pfn_to_page(pfn);
1699 		set_page_links(page, zone, nid, pfn);
1700 		set_page_count(page, 1);
1701 		reset_page_mapcount(page);
1702 		SetPageReserved(page);
1703 		INIT_LIST_HEAD(&page->lru);
1704 #ifdef WANT_PAGE_VIRTUAL
1705 		/* The shift won't overflow because ZONE_NORMAL is below 4G. */
1706 		if (!is_highmem_idx(zone))
1707 			set_page_address(page, __va(pfn << PAGE_SHIFT));
1708 #endif
1709 	}
1710 }
1711 
1712 void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1713 				unsigned long size)
1714 {
1715 	int order;
1716 	for (order = 0; order < MAX_ORDER ; order++) {
1717 		INIT_LIST_HEAD(&zone->free_area[order].free_list);
1718 		zone->free_area[order].nr_free = 0;
1719 	}
1720 }
1721 
1722 #define ZONETABLE_INDEX(x, zone_nr)	((x << ZONES_SHIFT) | zone_nr)
1723 void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1724 		unsigned long size)
1725 {
1726 	unsigned long snum = pfn_to_section_nr(pfn);
1727 	unsigned long end = pfn_to_section_nr(pfn + size);
1728 
1729 	if (FLAGS_HAS_NODE)
1730 		zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1731 	else
1732 		for (; snum <= end; snum++)
1733 			zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1734 }
1735 
1736 #ifndef __HAVE_ARCH_MEMMAP_INIT
1737 #define memmap_init(size, nid, zone, start_pfn) \
1738 	memmap_init_zone((size), (nid), (zone), (start_pfn))
1739 #endif
1740 
1741 static int __devinit zone_batchsize(struct zone *zone)
1742 {
1743 	int batch;
1744 
1745 	/*
1746 	 * The per-cpu-pages pools are set to around 1000th of the
1747 	 * size of the zone.  But no more than 1/2 of a meg.
1748 	 *
1749 	 * OK, so we don't know how big the cache is.  So guess.
1750 	 */
1751 	batch = zone->present_pages / 1024;
1752 	if (batch * PAGE_SIZE > 512 * 1024)
1753 		batch = (512 * 1024) / PAGE_SIZE;
1754 	batch /= 4;		/* We effectively *= 4 below */
1755 	if (batch < 1)
1756 		batch = 1;
1757 
1758 	/*
1759 	 * We will be trying to allcoate bigger chunks of contiguous
1760 	 * memory of the order of fls(batch).  This should result in
1761 	 * better cache coloring.
1762 	 *
1763 	 * A sanity check also to ensure that batch is still in limits.
1764 	 */
1765 	batch = (1 << fls(batch + batch/2));
1766 
1767 	if (fls(batch) >= (PAGE_SHIFT + MAX_ORDER - 2))
1768 		batch = PAGE_SHIFT + ((MAX_ORDER - 1 - PAGE_SHIFT)/2);
1769 
1770 	return batch;
1771 }
1772 
1773 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1774 {
1775 	struct per_cpu_pages *pcp;
1776 
1777 	memset(p, 0, sizeof(*p));
1778 
1779 	pcp = &p->pcp[0];		/* hot */
1780 	pcp->count = 0;
1781 	pcp->low = 0;
1782 	pcp->high = 6 * batch;
1783 	pcp->batch = max(1UL, 1 * batch);
1784 	INIT_LIST_HEAD(&pcp->list);
1785 
1786 	pcp = &p->pcp[1];		/* cold*/
1787 	pcp->count = 0;
1788 	pcp->low = 0;
1789 	pcp->high = 2 * batch;
1790 	pcp->batch = max(1UL, batch/2);
1791 	INIT_LIST_HEAD(&pcp->list);
1792 }
1793 
1794 #ifdef CONFIG_NUMA
1795 /*
1796  * Boot pageset table. One per cpu which is going to be used for all
1797  * zones and all nodes. The parameters will be set in such a way
1798  * that an item put on a list will immediately be handed over to
1799  * the buddy list. This is safe since pageset manipulation is done
1800  * with interrupts disabled.
1801  *
1802  * Some NUMA counter updates may also be caught by the boot pagesets.
1803  *
1804  * The boot_pagesets must be kept even after bootup is complete for
1805  * unused processors and/or zones. They do play a role for bootstrapping
1806  * hotplugged processors.
1807  *
1808  * zoneinfo_show() and maybe other functions do
1809  * not check if the processor is online before following the pageset pointer.
1810  * Other parts of the kernel may not check if the zone is available.
1811  */
1812 static struct per_cpu_pageset
1813 	boot_pageset[NR_CPUS];
1814 
1815 /*
1816  * Dynamically allocate memory for the
1817  * per cpu pageset array in struct zone.
1818  */
1819 static int __devinit process_zones(int cpu)
1820 {
1821 	struct zone *zone, *dzone;
1822 
1823 	for_each_zone(zone) {
1824 
1825 		zone->pageset[cpu] = kmalloc_node(sizeof(struct per_cpu_pageset),
1826 					 GFP_KERNEL, cpu_to_node(cpu));
1827 		if (!zone->pageset[cpu])
1828 			goto bad;
1829 
1830 		setup_pageset(zone->pageset[cpu], zone_batchsize(zone));
1831 	}
1832 
1833 	return 0;
1834 bad:
1835 	for_each_zone(dzone) {
1836 		if (dzone == zone)
1837 			break;
1838 		kfree(dzone->pageset[cpu]);
1839 		dzone->pageset[cpu] = NULL;
1840 	}
1841 	return -ENOMEM;
1842 }
1843 
1844 static inline void free_zone_pagesets(int cpu)
1845 {
1846 #ifdef CONFIG_NUMA
1847 	struct zone *zone;
1848 
1849 	for_each_zone(zone) {
1850 		struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1851 
1852 		zone_pcp(zone, cpu) = NULL;
1853 		kfree(pset);
1854 	}
1855 #endif
1856 }
1857 
1858 static int __devinit pageset_cpuup_callback(struct notifier_block *nfb,
1859 		unsigned long action,
1860 		void *hcpu)
1861 {
1862 	int cpu = (long)hcpu;
1863 	int ret = NOTIFY_OK;
1864 
1865 	switch (action) {
1866 		case CPU_UP_PREPARE:
1867 			if (process_zones(cpu))
1868 				ret = NOTIFY_BAD;
1869 			break;
1870 #ifdef CONFIG_HOTPLUG_CPU
1871 		case CPU_DEAD:
1872 			free_zone_pagesets(cpu);
1873 			break;
1874 #endif
1875 		default:
1876 			break;
1877 	}
1878 	return ret;
1879 }
1880 
1881 static struct notifier_block pageset_notifier =
1882 	{ &pageset_cpuup_callback, NULL, 0 };
1883 
1884 void __init setup_per_cpu_pageset()
1885 {
1886 	int err;
1887 
1888 	/* Initialize per_cpu_pageset for cpu 0.
1889 	 * A cpuup callback will do this for every cpu
1890 	 * as it comes online
1891 	 */
1892 	err = process_zones(smp_processor_id());
1893 	BUG_ON(err);
1894 	register_cpu_notifier(&pageset_notifier);
1895 }
1896 
1897 #endif
1898 
1899 static __devinit
1900 void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
1901 {
1902 	int i;
1903 	struct pglist_data *pgdat = zone->zone_pgdat;
1904 
1905 	/*
1906 	 * The per-page waitqueue mechanism uses hashed waitqueues
1907 	 * per zone.
1908 	 */
1909 	zone->wait_table_size = wait_table_size(zone_size_pages);
1910 	zone->wait_table_bits =	wait_table_bits(zone->wait_table_size);
1911 	zone->wait_table = (wait_queue_head_t *)
1912 		alloc_bootmem_node(pgdat, zone->wait_table_size
1913 					* sizeof(wait_queue_head_t));
1914 
1915 	for(i = 0; i < zone->wait_table_size; ++i)
1916 		init_waitqueue_head(zone->wait_table + i);
1917 }
1918 
1919 static __devinit void zone_pcp_init(struct zone *zone)
1920 {
1921 	int cpu;
1922 	unsigned long batch = zone_batchsize(zone);
1923 
1924 	for (cpu = 0; cpu < NR_CPUS; cpu++) {
1925 #ifdef CONFIG_NUMA
1926 		/* Early boot. Slab allocator not functional yet */
1927 		zone->pageset[cpu] = &boot_pageset[cpu];
1928 		setup_pageset(&boot_pageset[cpu],0);
1929 #else
1930 		setup_pageset(zone_pcp(zone,cpu), batch);
1931 #endif
1932 	}
1933 	printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%lu\n",
1934 		zone->name, zone->present_pages, batch);
1935 }
1936 
1937 static __devinit void init_currently_empty_zone(struct zone *zone,
1938 		unsigned long zone_start_pfn, unsigned long size)
1939 {
1940 	struct pglist_data *pgdat = zone->zone_pgdat;
1941 
1942 	zone_wait_table_init(zone, size);
1943 	pgdat->nr_zones = zone_idx(zone) + 1;
1944 
1945 	zone->zone_mem_map = pfn_to_page(zone_start_pfn);
1946 	zone->zone_start_pfn = zone_start_pfn;
1947 
1948 	memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
1949 
1950 	zone_init_free_lists(pgdat, zone, zone->spanned_pages);
1951 }
1952 
1953 /*
1954  * Set up the zone data structures:
1955  *   - mark all pages reserved
1956  *   - mark all memory queues empty
1957  *   - clear the memory bitmaps
1958  */
1959 static void __init free_area_init_core(struct pglist_data *pgdat,
1960 		unsigned long *zones_size, unsigned long *zholes_size)
1961 {
1962 	unsigned long j;
1963 	int nid = pgdat->node_id;
1964 	unsigned long zone_start_pfn = pgdat->node_start_pfn;
1965 
1966 	pgdat_resize_init(pgdat);
1967 	pgdat->nr_zones = 0;
1968 	init_waitqueue_head(&pgdat->kswapd_wait);
1969 	pgdat->kswapd_max_order = 0;
1970 
1971 	for (j = 0; j < MAX_NR_ZONES; j++) {
1972 		struct zone *zone = pgdat->node_zones + j;
1973 		unsigned long size, realsize;
1974 
1975 		realsize = size = zones_size[j];
1976 		if (zholes_size)
1977 			realsize -= zholes_size[j];
1978 
1979 		if (j == ZONE_DMA || j == ZONE_NORMAL)
1980 			nr_kernel_pages += realsize;
1981 		nr_all_pages += realsize;
1982 
1983 		zone->spanned_pages = size;
1984 		zone->present_pages = realsize;
1985 		zone->name = zone_names[j];
1986 		spin_lock_init(&zone->lock);
1987 		spin_lock_init(&zone->lru_lock);
1988 		zone_seqlock_init(zone);
1989 		zone->zone_pgdat = pgdat;
1990 		zone->free_pages = 0;
1991 
1992 		zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
1993 
1994 		zone_pcp_init(zone);
1995 		INIT_LIST_HEAD(&zone->active_list);
1996 		INIT_LIST_HEAD(&zone->inactive_list);
1997 		zone->nr_scan_active = 0;
1998 		zone->nr_scan_inactive = 0;
1999 		zone->nr_active = 0;
2000 		zone->nr_inactive = 0;
2001 		atomic_set(&zone->reclaim_in_progress, 0);
2002 		if (!size)
2003 			continue;
2004 
2005 		zonetable_add(zone, nid, j, zone_start_pfn, size);
2006 		init_currently_empty_zone(zone, zone_start_pfn, size);
2007 		zone_start_pfn += size;
2008 	}
2009 }
2010 
2011 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2012 {
2013 	/* Skip empty nodes */
2014 	if (!pgdat->node_spanned_pages)
2015 		return;
2016 
2017 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2018 	/* ia64 gets its own node_mem_map, before this, without bootmem */
2019 	if (!pgdat->node_mem_map) {
2020 		unsigned long size;
2021 		struct page *map;
2022 
2023 		size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
2024 		map = alloc_remap(pgdat->node_id, size);
2025 		if (!map)
2026 			map = alloc_bootmem_node(pgdat, size);
2027 		pgdat->node_mem_map = map;
2028 	}
2029 #ifdef CONFIG_FLATMEM
2030 	/*
2031 	 * With no DISCONTIG, the global mem_map is just set as node 0's
2032 	 */
2033 	if (pgdat == NODE_DATA(0))
2034 		mem_map = NODE_DATA(0)->node_mem_map;
2035 #endif
2036 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2037 }
2038 
2039 void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2040 		unsigned long *zones_size, unsigned long node_start_pfn,
2041 		unsigned long *zholes_size)
2042 {
2043 	pgdat->node_id = nid;
2044 	pgdat->node_start_pfn = node_start_pfn;
2045 	calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2046 
2047 	alloc_node_mem_map(pgdat);
2048 
2049 	free_area_init_core(pgdat, zones_size, zholes_size);
2050 }
2051 
2052 #ifndef CONFIG_NEED_MULTIPLE_NODES
2053 static bootmem_data_t contig_bootmem_data;
2054 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2055 
2056 EXPORT_SYMBOL(contig_page_data);
2057 #endif
2058 
2059 void __init free_area_init(unsigned long *zones_size)
2060 {
2061 	free_area_init_node(0, NODE_DATA(0), zones_size,
2062 			__pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2063 }
2064 
2065 #ifdef CONFIG_PROC_FS
2066 
2067 #include <linux/seq_file.h>
2068 
2069 static void *frag_start(struct seq_file *m, loff_t *pos)
2070 {
2071 	pg_data_t *pgdat;
2072 	loff_t node = *pos;
2073 
2074 	for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
2075 		--node;
2076 
2077 	return pgdat;
2078 }
2079 
2080 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2081 {
2082 	pg_data_t *pgdat = (pg_data_t *)arg;
2083 
2084 	(*pos)++;
2085 	return pgdat->pgdat_next;
2086 }
2087 
2088 static void frag_stop(struct seq_file *m, void *arg)
2089 {
2090 }
2091 
2092 /*
2093  * This walks the free areas for each zone.
2094  */
2095 static int frag_show(struct seq_file *m, void *arg)
2096 {
2097 	pg_data_t *pgdat = (pg_data_t *)arg;
2098 	struct zone *zone;
2099 	struct zone *node_zones = pgdat->node_zones;
2100 	unsigned long flags;
2101 	int order;
2102 
2103 	for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2104 		if (!zone->present_pages)
2105 			continue;
2106 
2107 		spin_lock_irqsave(&zone->lock, flags);
2108 		seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2109 		for (order = 0; order < MAX_ORDER; ++order)
2110 			seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2111 		spin_unlock_irqrestore(&zone->lock, flags);
2112 		seq_putc(m, '\n');
2113 	}
2114 	return 0;
2115 }
2116 
2117 struct seq_operations fragmentation_op = {
2118 	.start	= frag_start,
2119 	.next	= frag_next,
2120 	.stop	= frag_stop,
2121 	.show	= frag_show,
2122 };
2123 
2124 /*
2125  * Output information about zones in @pgdat.
2126  */
2127 static int zoneinfo_show(struct seq_file *m, void *arg)
2128 {
2129 	pg_data_t *pgdat = arg;
2130 	struct zone *zone;
2131 	struct zone *node_zones = pgdat->node_zones;
2132 	unsigned long flags;
2133 
2134 	for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2135 		int i;
2136 
2137 		if (!zone->present_pages)
2138 			continue;
2139 
2140 		spin_lock_irqsave(&zone->lock, flags);
2141 		seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2142 		seq_printf(m,
2143 			   "\n  pages free     %lu"
2144 			   "\n        min      %lu"
2145 			   "\n        low      %lu"
2146 			   "\n        high     %lu"
2147 			   "\n        active   %lu"
2148 			   "\n        inactive %lu"
2149 			   "\n        scanned  %lu (a: %lu i: %lu)"
2150 			   "\n        spanned  %lu"
2151 			   "\n        present  %lu",
2152 			   zone->free_pages,
2153 			   zone->pages_min,
2154 			   zone->pages_low,
2155 			   zone->pages_high,
2156 			   zone->nr_active,
2157 			   zone->nr_inactive,
2158 			   zone->pages_scanned,
2159 			   zone->nr_scan_active, zone->nr_scan_inactive,
2160 			   zone->spanned_pages,
2161 			   zone->present_pages);
2162 		seq_printf(m,
2163 			   "\n        protection: (%lu",
2164 			   zone->lowmem_reserve[0]);
2165 		for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2166 			seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2167 		seq_printf(m,
2168 			   ")"
2169 			   "\n  pagesets");
2170 		for (i = 0; i < ARRAY_SIZE(zone->pageset); i++) {
2171 			struct per_cpu_pageset *pageset;
2172 			int j;
2173 
2174 			pageset = zone_pcp(zone, i);
2175 			for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2176 				if (pageset->pcp[j].count)
2177 					break;
2178 			}
2179 			if (j == ARRAY_SIZE(pageset->pcp))
2180 				continue;
2181 			for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2182 				seq_printf(m,
2183 					   "\n    cpu: %i pcp: %i"
2184 					   "\n              count: %i"
2185 					   "\n              low:   %i"
2186 					   "\n              high:  %i"
2187 					   "\n              batch: %i",
2188 					   i, j,
2189 					   pageset->pcp[j].count,
2190 					   pageset->pcp[j].low,
2191 					   pageset->pcp[j].high,
2192 					   pageset->pcp[j].batch);
2193 			}
2194 #ifdef CONFIG_NUMA
2195 			seq_printf(m,
2196 				   "\n            numa_hit:       %lu"
2197 				   "\n            numa_miss:      %lu"
2198 				   "\n            numa_foreign:   %lu"
2199 				   "\n            interleave_hit: %lu"
2200 				   "\n            local_node:     %lu"
2201 				   "\n            other_node:     %lu",
2202 				   pageset->numa_hit,
2203 				   pageset->numa_miss,
2204 				   pageset->numa_foreign,
2205 				   pageset->interleave_hit,
2206 				   pageset->local_node,
2207 				   pageset->other_node);
2208 #endif
2209 		}
2210 		seq_printf(m,
2211 			   "\n  all_unreclaimable: %u"
2212 			   "\n  prev_priority:     %i"
2213 			   "\n  temp_priority:     %i"
2214 			   "\n  start_pfn:         %lu",
2215 			   zone->all_unreclaimable,
2216 			   zone->prev_priority,
2217 			   zone->temp_priority,
2218 			   zone->zone_start_pfn);
2219 		spin_unlock_irqrestore(&zone->lock, flags);
2220 		seq_putc(m, '\n');
2221 	}
2222 	return 0;
2223 }
2224 
2225 struct seq_operations zoneinfo_op = {
2226 	.start	= frag_start, /* iterate over all zones. The same as in
2227 			       * fragmentation. */
2228 	.next	= frag_next,
2229 	.stop	= frag_stop,
2230 	.show	= zoneinfo_show,
2231 };
2232 
2233 static char *vmstat_text[] = {
2234 	"nr_dirty",
2235 	"nr_writeback",
2236 	"nr_unstable",
2237 	"nr_page_table_pages",
2238 	"nr_mapped",
2239 	"nr_slab",
2240 
2241 	"pgpgin",
2242 	"pgpgout",
2243 	"pswpin",
2244 	"pswpout",
2245 	"pgalloc_high",
2246 
2247 	"pgalloc_normal",
2248 	"pgalloc_dma",
2249 	"pgfree",
2250 	"pgactivate",
2251 	"pgdeactivate",
2252 
2253 	"pgfault",
2254 	"pgmajfault",
2255 	"pgrefill_high",
2256 	"pgrefill_normal",
2257 	"pgrefill_dma",
2258 
2259 	"pgsteal_high",
2260 	"pgsteal_normal",
2261 	"pgsteal_dma",
2262 	"pgscan_kswapd_high",
2263 	"pgscan_kswapd_normal",
2264 
2265 	"pgscan_kswapd_dma",
2266 	"pgscan_direct_high",
2267 	"pgscan_direct_normal",
2268 	"pgscan_direct_dma",
2269 	"pginodesteal",
2270 
2271 	"slabs_scanned",
2272 	"kswapd_steal",
2273 	"kswapd_inodesteal",
2274 	"pageoutrun",
2275 	"allocstall",
2276 
2277 	"pgrotated",
2278 	"nr_bounce",
2279 };
2280 
2281 static void *vmstat_start(struct seq_file *m, loff_t *pos)
2282 {
2283 	struct page_state *ps;
2284 
2285 	if (*pos >= ARRAY_SIZE(vmstat_text))
2286 		return NULL;
2287 
2288 	ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2289 	m->private = ps;
2290 	if (!ps)
2291 		return ERR_PTR(-ENOMEM);
2292 	get_full_page_state(ps);
2293 	ps->pgpgin /= 2;		/* sectors -> kbytes */
2294 	ps->pgpgout /= 2;
2295 	return (unsigned long *)ps + *pos;
2296 }
2297 
2298 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2299 {
2300 	(*pos)++;
2301 	if (*pos >= ARRAY_SIZE(vmstat_text))
2302 		return NULL;
2303 	return (unsigned long *)m->private + *pos;
2304 }
2305 
2306 static int vmstat_show(struct seq_file *m, void *arg)
2307 {
2308 	unsigned long *l = arg;
2309 	unsigned long off = l - (unsigned long *)m->private;
2310 
2311 	seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2312 	return 0;
2313 }
2314 
2315 static void vmstat_stop(struct seq_file *m, void *arg)
2316 {
2317 	kfree(m->private);
2318 	m->private = NULL;
2319 }
2320 
2321 struct seq_operations vmstat_op = {
2322 	.start	= vmstat_start,
2323 	.next	= vmstat_next,
2324 	.stop	= vmstat_stop,
2325 	.show	= vmstat_show,
2326 };
2327 
2328 #endif /* CONFIG_PROC_FS */
2329 
2330 #ifdef CONFIG_HOTPLUG_CPU
2331 static int page_alloc_cpu_notify(struct notifier_block *self,
2332 				 unsigned long action, void *hcpu)
2333 {
2334 	int cpu = (unsigned long)hcpu;
2335 	long *count;
2336 	unsigned long *src, *dest;
2337 
2338 	if (action == CPU_DEAD) {
2339 		int i;
2340 
2341 		/* Drain local pagecache count. */
2342 		count = &per_cpu(nr_pagecache_local, cpu);
2343 		atomic_add(*count, &nr_pagecache);
2344 		*count = 0;
2345 		local_irq_disable();
2346 		__drain_pages(cpu);
2347 
2348 		/* Add dead cpu's page_states to our own. */
2349 		dest = (unsigned long *)&__get_cpu_var(page_states);
2350 		src = (unsigned long *)&per_cpu(page_states, cpu);
2351 
2352 		for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2353 				i++) {
2354 			dest[i] += src[i];
2355 			src[i] = 0;
2356 		}
2357 
2358 		local_irq_enable();
2359 	}
2360 	return NOTIFY_OK;
2361 }
2362 #endif /* CONFIG_HOTPLUG_CPU */
2363 
2364 void __init page_alloc_init(void)
2365 {
2366 	hotcpu_notifier(page_alloc_cpu_notify, 0);
2367 }
2368 
2369 /*
2370  * setup_per_zone_lowmem_reserve - called whenever
2371  *	sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
2372  *	has a correct pages reserved value, so an adequate number of
2373  *	pages are left in the zone after a successful __alloc_pages().
2374  */
2375 static void setup_per_zone_lowmem_reserve(void)
2376 {
2377 	struct pglist_data *pgdat;
2378 	int j, idx;
2379 
2380 	for_each_pgdat(pgdat) {
2381 		for (j = 0; j < MAX_NR_ZONES; j++) {
2382 			struct zone *zone = pgdat->node_zones + j;
2383 			unsigned long present_pages = zone->present_pages;
2384 
2385 			zone->lowmem_reserve[j] = 0;
2386 
2387 			for (idx = j-1; idx >= 0; idx--) {
2388 				struct zone *lower_zone;
2389 
2390 				if (sysctl_lowmem_reserve_ratio[idx] < 1)
2391 					sysctl_lowmem_reserve_ratio[idx] = 1;
2392 
2393 				lower_zone = pgdat->node_zones + idx;
2394 				lower_zone->lowmem_reserve[j] = present_pages /
2395 					sysctl_lowmem_reserve_ratio[idx];
2396 				present_pages += lower_zone->present_pages;
2397 			}
2398 		}
2399 	}
2400 }
2401 
2402 /*
2403  * setup_per_zone_pages_min - called when min_free_kbytes changes.  Ensures
2404  *	that the pages_{min,low,high} values for each zone are set correctly
2405  *	with respect to min_free_kbytes.
2406  */
2407 void setup_per_zone_pages_min(void)
2408 {
2409 	unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2410 	unsigned long lowmem_pages = 0;
2411 	struct zone *zone;
2412 	unsigned long flags;
2413 
2414 	/* Calculate total number of !ZONE_HIGHMEM pages */
2415 	for_each_zone(zone) {
2416 		if (!is_highmem(zone))
2417 			lowmem_pages += zone->present_pages;
2418 	}
2419 
2420 	for_each_zone(zone) {
2421 		spin_lock_irqsave(&zone->lru_lock, flags);
2422 		if (is_highmem(zone)) {
2423 			/*
2424 			 * Often, highmem doesn't need to reserve any pages.
2425 			 * But the pages_min/low/high values are also used for
2426 			 * batching up page reclaim activity so we need a
2427 			 * decent value here.
2428 			 */
2429 			int min_pages;
2430 
2431 			min_pages = zone->present_pages / 1024;
2432 			if (min_pages < SWAP_CLUSTER_MAX)
2433 				min_pages = SWAP_CLUSTER_MAX;
2434 			if (min_pages > 128)
2435 				min_pages = 128;
2436 			zone->pages_min = min_pages;
2437 		} else {
2438 			/* if it's a lowmem zone, reserve a number of pages
2439 			 * proportionate to the zone's size.
2440 			 */
2441 			zone->pages_min = (pages_min * zone->present_pages) /
2442 			                   lowmem_pages;
2443 		}
2444 
2445 		/*
2446 		 * When interpreting these watermarks, just keep in mind that:
2447 		 * zone->pages_min == (zone->pages_min * 4) / 4;
2448 		 */
2449 		zone->pages_low   = (zone->pages_min * 5) / 4;
2450 		zone->pages_high  = (zone->pages_min * 6) / 4;
2451 		spin_unlock_irqrestore(&zone->lru_lock, flags);
2452 	}
2453 }
2454 
2455 /*
2456  * Initialise min_free_kbytes.
2457  *
2458  * For small machines we want it small (128k min).  For large machines
2459  * we want it large (64MB max).  But it is not linear, because network
2460  * bandwidth does not increase linearly with machine size.  We use
2461  *
2462  * 	min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2463  *	min_free_kbytes = sqrt(lowmem_kbytes * 16)
2464  *
2465  * which yields
2466  *
2467  * 16MB:	512k
2468  * 32MB:	724k
2469  * 64MB:	1024k
2470  * 128MB:	1448k
2471  * 256MB:	2048k
2472  * 512MB:	2896k
2473  * 1024MB:	4096k
2474  * 2048MB:	5792k
2475  * 4096MB:	8192k
2476  * 8192MB:	11584k
2477  * 16384MB:	16384k
2478  */
2479 static int __init init_per_zone_pages_min(void)
2480 {
2481 	unsigned long lowmem_kbytes;
2482 
2483 	lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2484 
2485 	min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2486 	if (min_free_kbytes < 128)
2487 		min_free_kbytes = 128;
2488 	if (min_free_kbytes > 65536)
2489 		min_free_kbytes = 65536;
2490 	setup_per_zone_pages_min();
2491 	setup_per_zone_lowmem_reserve();
2492 	return 0;
2493 }
2494 module_init(init_per_zone_pages_min)
2495 
2496 /*
2497  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2498  *	that we can call two helper functions whenever min_free_kbytes
2499  *	changes.
2500  */
2501 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2502 	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2503 {
2504 	proc_dointvec(table, write, file, buffer, length, ppos);
2505 	setup_per_zone_pages_min();
2506 	return 0;
2507 }
2508 
2509 /*
2510  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2511  *	proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2512  *	whenever sysctl_lowmem_reserve_ratio changes.
2513  *
2514  * The reserve ratio obviously has absolutely no relation with the
2515  * pages_min watermarks. The lowmem reserve ratio can only make sense
2516  * if in function of the boot time zone sizes.
2517  */
2518 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2519 	struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2520 {
2521 	proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2522 	setup_per_zone_lowmem_reserve();
2523 	return 0;
2524 }
2525 
2526 __initdata int hashdist = HASHDIST_DEFAULT;
2527 
2528 #ifdef CONFIG_NUMA
2529 static int __init set_hashdist(char *str)
2530 {
2531 	if (!str)
2532 		return 0;
2533 	hashdist = simple_strtoul(str, &str, 0);
2534 	return 1;
2535 }
2536 __setup("hashdist=", set_hashdist);
2537 #endif
2538 
2539 /*
2540  * allocate a large system hash table from bootmem
2541  * - it is assumed that the hash table must contain an exact power-of-2
2542  *   quantity of entries
2543  * - limit is the number of hash buckets, not the total allocation size
2544  */
2545 void *__init alloc_large_system_hash(const char *tablename,
2546 				     unsigned long bucketsize,
2547 				     unsigned long numentries,
2548 				     int scale,
2549 				     int flags,
2550 				     unsigned int *_hash_shift,
2551 				     unsigned int *_hash_mask,
2552 				     unsigned long limit)
2553 {
2554 	unsigned long long max = limit;
2555 	unsigned long log2qty, size;
2556 	void *table = NULL;
2557 
2558 	/* allow the kernel cmdline to have a say */
2559 	if (!numentries) {
2560 		/* round applicable memory size up to nearest megabyte */
2561 		numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2562 		numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2563 		numentries >>= 20 - PAGE_SHIFT;
2564 		numentries <<= 20 - PAGE_SHIFT;
2565 
2566 		/* limit to 1 bucket per 2^scale bytes of low memory */
2567 		if (scale > PAGE_SHIFT)
2568 			numentries >>= (scale - PAGE_SHIFT);
2569 		else
2570 			numentries <<= (PAGE_SHIFT - scale);
2571 	}
2572 	/* rounded up to nearest power of 2 in size */
2573 	numentries = 1UL << (long_log2(numentries) + 1);
2574 
2575 	/* limit allocation size to 1/16 total memory by default */
2576 	if (max == 0) {
2577 		max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2578 		do_div(max, bucketsize);
2579 	}
2580 
2581 	if (numentries > max)
2582 		numentries = max;
2583 
2584 	log2qty = long_log2(numentries);
2585 
2586 	do {
2587 		size = bucketsize << log2qty;
2588 		if (flags & HASH_EARLY)
2589 			table = alloc_bootmem(size);
2590 		else if (hashdist)
2591 			table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2592 		else {
2593 			unsigned long order;
2594 			for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2595 				;
2596 			table = (void*) __get_free_pages(GFP_ATOMIC, order);
2597 		}
2598 	} while (!table && size > PAGE_SIZE && --log2qty);
2599 
2600 	if (!table)
2601 		panic("Failed to allocate %s hash table\n", tablename);
2602 
2603 	printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2604 	       tablename,
2605 	       (1U << log2qty),
2606 	       long_log2(size) - PAGE_SHIFT,
2607 	       size);
2608 
2609 	if (_hash_shift)
2610 		*_hash_shift = log2qty;
2611 	if (_hash_mask)
2612 		*_hash_mask = (1 << log2qty) - 1;
2613 
2614 	return table;
2615 }
2616