Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  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/jiffies.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/oom.h>
 32 #include <linux/notifier.h>
 33 #include <linux/topology.h>
 34 #include <linux/sysctl.h>
 35 #include <linux/cpu.h>
 36 #include <linux/cpuset.h>
 37 #include <linux/memory_hotplug.h>
 38 #include <linux/nodemask.h>
 39 #include <linux/vmalloc.h>
 40 #include <linux/mempolicy.h>
 41 #include <linux/stop_machine.h>
 42 #include <linux/sort.h>
 43 #include <linux/pfn.h>
 44 #include <linux/backing-dev.h>
 45 #include <linux/fault-inject.h>
 46 #include <linux/page-isolation.h>
 47 #include <linux/memcontrol.h>
 48 
 49 #include <asm/tlbflush.h>
 50 #include <asm/div64.h>
 51 #include "internal.h"
 52 
 53 /*
 54  * Array of node states.
 55  */
 56 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
 57         [N_POSSIBLE] = NODE_MASK_ALL,
 58         [N_ONLINE] = { { [0] = 1UL } },
 59 #ifndef CONFIG_NUMA
 60         [N_NORMAL_MEMORY] = { { [0] = 1UL } },
 61 #ifdef CONFIG_HIGHMEM
 62         [N_HIGH_MEMORY] = { { [0] = 1UL } },
 63 #endif
 64         [N_CPU] = { { [0] = 1UL } },
 65 #endif  /* NUMA */
 66 };
 67 EXPORT_SYMBOL(node_states);
 68 
 69 unsigned long totalram_pages __read_mostly;
 70 unsigned long totalreserve_pages __read_mostly;
 71 long nr_swap_pages;
 72 int percpu_pagelist_fraction;
 73 
 74 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
 75 int pageblock_order __read_mostly;
 76 #endif
 77 
 78 static void __free_pages_ok(struct page *page, unsigned int order);
 79 
 80 /*
 81  * results with 256, 32 in the lowmem_reserve sysctl:
 82  *      1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
 83  *      1G machine -> (16M dma, 784M normal, 224M high)
 84  *      NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
 85  *      HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
 86  *      HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
 87  *
 88  * TBD: should special case ZONE_DMA32 machines here - in those we normally
 89  * don't need any ZONE_NORMAL reservation
 90  */
 91 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
 92 #ifdef CONFIG_ZONE_DMA
 93          256,
 94 #endif
 95 #ifdef CONFIG_ZONE_DMA32
 96          256,
 97 #endif
 98 #ifdef CONFIG_HIGHMEM
 99          32,
100 #endif
101          32,
102 };
103 
104 EXPORT_SYMBOL(totalram_pages);
105 
106 static char * const zone_names[MAX_NR_ZONES] = {
107 #ifdef CONFIG_ZONE_DMA
108          "DMA",
109 #endif
110 #ifdef CONFIG_ZONE_DMA32
111          "DMA32",
112 #endif
113          "Normal",
114 #ifdef CONFIG_HIGHMEM
115          "HighMem",
116 #endif
117          "Movable",
118 };
119 
120 int min_free_kbytes = 1024;
121 
122 unsigned long __meminitdata nr_kernel_pages;
123 unsigned long __meminitdata nr_all_pages;
124 static unsigned long __meminitdata dma_reserve;
125 
126 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
127   /*
128    * MAX_ACTIVE_REGIONS determines the maximum number of distinct
129    * ranges of memory (RAM) that may be registered with add_active_range().
130    * Ranges passed to add_active_range() will be merged if possible
131    * so the number of times add_active_range() can be called is
132    * related to the number of nodes and the number of holes
133    */
134   #ifdef CONFIG_MAX_ACTIVE_REGIONS
135     /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
136     #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
137   #else
138     #if MAX_NUMNODES >= 32
139       /* If there can be many nodes, allow up to 50 holes per node */
140       #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
141     #else
142       /* By default, allow up to 256 distinct regions */
143       #define MAX_ACTIVE_REGIONS 256
144     #endif
145   #endif
146 
147   static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
148   static int __meminitdata nr_nodemap_entries;
149   static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
150   static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
151 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
152   static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
153   static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
154 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
155   unsigned long __initdata required_kernelcore;
156   static unsigned long __initdata required_movablecore;
157   unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
158 
159   /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
160   int movable_zone;
161   EXPORT_SYMBOL(movable_zone);
162 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163 
164 #ifdef CONFIG_PREEMPT_RT
165 static DEFINE_PER_CPU_LOCKED(int, pcp_locks);
166 #endif
167 
168 static inline void __lock_cpu_pcp(unsigned long *flags, int cpu)
169 {
170 #ifdef CONFIG_PREEMPT_RT
171         spin_lock(&__get_cpu_lock(pcp_locks, cpu));
172         flags = 0;
173 #else
174         local_irq_save(*flags);
175 #endif
176 }
177 
178 static inline void lock_cpu_pcp(unsigned long *flags, int *this_cpu)
179 {
180 #ifdef CONFIG_PREEMPT_RT
181         (void)get_cpu_var_locked(pcp_locks, this_cpu);
182         flags = 0;
183 #else
184         local_irq_save(*flags);
185         *this_cpu = smp_processor_id();
186 #endif
187 }
188 
189 static inline void unlock_cpu_pcp(unsigned long flags, int this_cpu)
190 {
191 #ifdef CONFIG_PREEMPT_RT
192         put_cpu_var_locked(pcp_locks, this_cpu);
193 #else
194         local_irq_restore(flags);
195 #endif
196 }
197 
198 static struct per_cpu_pageset *
199 get_zone_pcp(struct zone *zone, unsigned long *flags, int *this_cpu)
200 {
201         lock_cpu_pcp(flags, this_cpu);
202         return zone_pcp(zone, *this_cpu);
203 }
204 
205 static void
206 put_zone_pcp(struct zone *zone, unsigned long flags, int this_cpu)
207 {
208         unlock_cpu_pcp(flags, this_cpu);
209 }
210 
211 #if MAX_NUMNODES > 1
212 int nr_node_ids __read_mostly = MAX_NUMNODES;
213 EXPORT_SYMBOL(nr_node_ids);
214 #endif
215 
216 int page_group_by_mobility_disabled __read_mostly;
217 
218 static void set_pageblock_migratetype(struct page *page, int migratetype)
219 {
220         set_pageblock_flags_group(page, (unsigned long)migratetype,
221                                         PB_migrate, PB_migrate_end);
222 }
223 
224 #ifdef CONFIG_DEBUG_VM
225 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
226 {
227         int ret = 0;
228         unsigned seq;
229         unsigned long pfn = page_to_pfn(page);
230 
231         do {
232                 seq = zone_span_seqbegin(zone);
233                 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
234                         ret = 1;
235                 else if (pfn < zone->zone_start_pfn)
236                         ret = 1;
237         } while (zone_span_seqretry(zone, seq));
238 
239         return ret;
240 }
241 
242 static int page_is_consistent(struct zone *zone, struct page *page)
243 {
244         if (!pfn_valid_within(page_to_pfn(page)))
245                 return 0;
246         if (zone != page_zone(page))
247                 return 0;
248 
249         return 1;
250 }
251 /*
252  * Temporary debugging check for pages not lying within a given zone.
253  */
254 static int bad_range(struct zone *zone, struct page *page)
255 {
256         if (page_outside_zone_boundaries(zone, page))
257                 return 1;
258         if (!page_is_consistent(zone, page))
259                 return 1;
260 
261         return 0;
262 }
263 #else
264 static inline int bad_range(struct zone *zone, struct page *page)
265 {
266         return 0;
267 }
268 #endif
269 
270 static void bad_page(struct page *page)
271 {
272         void *pc = page_get_page_cgroup(page);
273 
274         printk(KERN_EMERG "Bad page state in process '%s'\n" KERN_EMERG
275                 "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
276                 current->comm, page, (int)(2*sizeof(unsigned long)),
277                 (unsigned long)page->flags, page->mapping,
278                 page_mapcount(page), page_count(page));
279         if (pc) {
280                 printk(KERN_EMERG "cgroup:%p\n", pc);
281                 page_reset_bad_cgroup(page);
282         }
283         printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
284                 KERN_EMERG "Backtrace:\n");
285         dump_stack();
286         page->flags &= ~(1 << PG_lru    |
287                         1 << PG_private |
288                         1 << PG_locked  |
289                         1 << PG_active  |
290                         1 << PG_dirty   |
291                         1 << PG_reclaim |
292                         1 << PG_slab    |
293                         1 << PG_swapcache |
294                         1 << PG_writeback |
295                         1 << PG_buddy );
296         set_page_count(page, 0);
297         reset_page_mapcount(page);
298         page->mapping = NULL;
299         add_taint(TAINT_BAD_PAGE);
300 }
301 
302 /*
303  * Higher-order pages are called "compound pages".  They are structured thusly:
304  *
305  * The first PAGE_SIZE page is called the "head page".
306  *
307  * The remaining PAGE_SIZE pages are called "tail pages".
308  *
309  * All pages have PG_compound set.  All pages have their ->private pointing at
310  * the head page (even the head page has this).
311  *
312  * The first tail page's ->lru.next holds the address of the compound page's
313  * put_page() function.  Its ->lru.prev holds the order of allocation.
314  * This usage means that zero-order pages may not be compound.
315  */
316 
317 static void free_compound_page(struct page *page)
318 {
319         __free_pages_ok(page, compound_order(page));
320 }
321 
322 static void prep_compound_page(struct page *page, unsigned long order)
323 {
324         int i;
325         int nr_pages = 1 << order;
326 
327         set_compound_page_dtor(page, free_compound_page);
328         set_compound_order(page, order);
329         __SetPageHead(page);
330         for (i = 1; i < nr_pages; i++) {
331                 struct page *p = page + i;
332 
333                 __SetPageTail(p);
334                 p->first_page = page;
335         }
336 }
337 
338 static void destroy_compound_page(struct page *page, unsigned long order)
339 {
340         int i;
341         int nr_pages = 1 << order;
342 
343         if (unlikely(compound_order(page) != order))
344                 bad_page(page);
345 
346         if (unlikely(!PageHead(page)))
347                         bad_page(page);
348         __ClearPageHead(page);
349         for (i = 1; i < nr_pages; i++) {
350                 struct page *p = page + i;
351 
352                 if (unlikely(!PageTail(p) |
353                                 (p->first_page != page)))
354                         bad_page(page);
355                 __ClearPageTail(p);
356         }
357 }
358 
359 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
360 {
361         int i;
362 
363         /*
364          * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
365          * and __GFP_HIGHMEM from hard or soft interrupt context.
366          */
367         VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
368         for (i = 0; i < (1 << order); i++)
369                 clear_highpage(page + i);
370 }
371 
372 static inline void set_page_order(struct page *page, int order)
373 {
374         set_page_private(page, order);
375         __SetPageBuddy(page);
376 }
377 
378 static inline void rmv_page_order(struct page *page)
379 {
380         __ClearPageBuddy(page);
381         set_page_private(page, 0);
382 }
383 
384 /*
385  * Locate the struct page for both the matching buddy in our
386  * pair (buddy1) and the combined O(n+1) page they form (page).
387  *
388  * 1) Any buddy B1 will have an order O twin B2 which satisfies
389  * the following equation:
390  *     B2 = B1 ^ (1 << O)
391  * For example, if the starting buddy (buddy2) is #8 its order
392  * 1 buddy is #10:
393  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
394  *
395  * 2) Any buddy B will have an order O+1 parent P which
396  * satisfies the following equation:
397  *     P = B & ~(1 << O)
398  *
399  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
400  */
401 static inline struct page *
402 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
403 {
404         unsigned long buddy_idx = page_idx ^ (1 << order);
405 
406         return page + (buddy_idx - page_idx);
407 }
408 
409 static inline unsigned long
410 __find_combined_index(unsigned long page_idx, unsigned int order)
411 {
412         return (page_idx & ~(1 << order));
413 }
414 
415 /*
416  * This function checks whether a page is free && is the buddy
417  * we can do coalesce a page and its buddy if
418  * (a) the buddy is not in a hole &&
419  * (b) the buddy is in the buddy system &&
420  * (c) a page and its buddy have the same order &&
421  * (d) a page and its buddy are in the same zone.
422  *
423  * For recording whether a page is in the buddy system, we use PG_buddy.
424  * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
425  *
426  * For recording page's order, we use page_private(page).
427  */
428 static inline int page_is_buddy(struct page *page, struct page *buddy,
429                                                                 int order)
430 {
431         if (!pfn_valid_within(page_to_pfn(buddy)))
432                 return 0;
433 
434         if (page_zone_id(page) != page_zone_id(buddy))
435                 return 0;
436 
437         if (PageBuddy(buddy) && page_order(buddy) == order) {
438                 BUG_ON(page_count(buddy) != 0);
439                 return 1;
440         }
441         return 0;
442 }
443 
444 /*
445  * Freeing function for a buddy system allocator.
446  *
447  * The concept of a buddy system is to maintain direct-mapped table
448  * (containing bit values) for memory blocks of various "orders".
449  * The bottom level table contains the map for the smallest allocatable
450  * units of memory (here, pages), and each level above it describes
451  * pairs of units from the levels below, hence, "buddies".
452  * At a high level, all that happens here is marking the table entry
453  * at the bottom level available, and propagating the changes upward
454  * as necessary, plus some accounting needed to play nicely with other
455  * parts of the VM system.
456  * At each level, we keep a list of pages, which are heads of continuous
457  * free pages of length of (1 << order) and marked with PG_buddy. Page's
458  * order is recorded in page_private(page) field.
459  * So when we are allocating or freeing one, we can derive the state of the
460  * other.  That is, if we allocate a small block, and both were   
461  * free, the remainder of the region must be split into blocks.   
462  * If a block is freed, and its buddy is also free, then this
463  * triggers coalescing into a block of larger size.            
464  *
465  * -- wli
466  */
467 
468 static inline void __free_one_page(struct page *page,
469                 struct zone *zone, unsigned int order)
470 {
471         unsigned long page_idx;
472         int order_size = 1 << order;
473         int migratetype = get_pageblock_migratetype(page);
474 
475         if (unlikely(PageCompound(page)))
476                 destroy_compound_page(page, order);
477 
478         page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
479 
480         VM_BUG_ON(page_idx & (order_size - 1));
481         VM_BUG_ON(bad_range(zone, page));
482 
483         __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
484         while (order < MAX_ORDER-1) {
485                 unsigned long combined_idx;
486                 struct page *buddy;
487 
488                 buddy = __page_find_buddy(page, page_idx, order);
489                 if (!page_is_buddy(page, buddy, order))
490                         break;          /* Move the buddy up one level. */
491 
492                 list_del(&buddy->lru);
493                 zone->free_area[order].nr_free--;
494                 rmv_page_order(buddy);
495                 combined_idx = __find_combined_index(page_idx, order);
496                 page = page + (combined_idx - page_idx);
497                 page_idx = combined_idx;
498                 order++;
499         }
500         set_page_order(page, order);
501         list_add(&page->lru,
502                 &zone->free_area[order].free_list[migratetype]);
503         zone->free_area[order].nr_free++;
504 }
505 
506 static inline int free_pages_check(struct page *page)
507 {
508         if (unlikely(page_mapcount(page) |
509                 (page->mapping != NULL)  |
510                 (page_get_page_cgroup(page) != NULL) |
511                 (page_count(page) != 0)  |
512                 (page->flags & (
513                         1 << PG_lru     |
514                         1 << PG_private |
515                         1 << PG_locked  |
516                         1 << PG_active  |
517                         1 << PG_slab    |
518                         1 << PG_swapcache |
519                         1 << PG_writeback |
520                         1 << PG_reserved |
521                         1 << PG_buddy ))))
522                 bad_page(page);
523         if (PageDirty(page))
524                 __ClearPageDirty(page);
525         /*
526          * For now, we report if PG_reserved was found set, but do not
527          * clear it, and do not free the page.  But we shall soon need
528          * to do more, for when the ZERO_PAGE count wraps negative.
529          */
530         return PageReserved(page);
531 }
532 
533 /*
534  * Frees a list of pages. 
535  * Assumes all pages on list are in same zone, and of same order.
536  * count is the number of pages to free.
537  *
538  * If the zone was previously in an "all pages pinned" state then look to
539  * see if this freeing clears that state.
540  *
541  * And clear the zone's pages_scanned counter, to hold off the "all pages are
542  * pinned" detection logic.
543  */
544 static void free_pages_bulk(struct zone *zone, int count,
545                                         struct list_head *list, int order)
546 {
547         spin_lock(&zone->lock);
548         zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
549         zone->pages_scanned = 0;
550         while (count--) {
551                 struct page *page;
552 
553                 VM_BUG_ON(list_empty(list));
554                 page = list_entry(list->prev, struct page, lru);
555                 /* have to delete it as __free_one_page list manipulates */
556                 list_del(&page->lru);
557                 __free_one_page(page, zone, order);
558         }
559         spin_unlock(&zone->lock);
560 }
561 
562 static void free_one_page(struct zone *zone, struct page *page, int order)
563 {
564         spin_lock(&zone->lock);
565         zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
566         zone->pages_scanned = 0;
567         __free_one_page(page, zone, order);
568         spin_unlock(&zone->lock);
569 }
570 
571 static void __free_pages_ok(struct page *page, unsigned int order)
572 {
573         unsigned long flags;
574         int reserved = 0;
575         int this_cpu;
576         int i;
577 
578         for (i = 0 ; i < (1 << order) ; ++i)
579                 reserved += free_pages_check(page + i);
580         if (reserved)
581                 return;
582 
583         if (!PageHighMem(page))
584                 debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
585         arch_free_page(page, order);
586         kernel_map_pages(page, 1 << order, 0);
587 
588         lock_cpu_pcp(&flags, &this_cpu);
589         count_vm_events(PGFREE, 1 << order);
590         free_one_page(page_zone(page), page, order);
591         unlock_cpu_pcp(flags, this_cpu);
592 }
593 
594 /*
595  * permit the bootmem allocator to evade page validation on high-order frees
596  */
597 void __init __free_pages_bootmem(struct page *page, unsigned int order)
598 {
599         if (order == 0) {
600                 __ClearPageReserved(page);
601                 set_page_count(page, 0);
602                 set_page_refcounted(page);
603                 __free_page(page);
604         } else {
605                 int loop;
606 
607                 prefetchw(page);
608                 for (loop = 0; loop < BITS_PER_LONG; loop++) {
609                         struct page *p = &page[loop];
610 
611                         if (loop + 1 < BITS_PER_LONG)
612                                 prefetchw(p + 1);
613                         __ClearPageReserved(p);
614                         set_page_count(p, 0);
615                 }
616 
617                 set_page_refcounted(page);
618                 __free_pages(page, order);
619         }
620 }
621 
622 
623 /*
624  * The order of subdivision here is critical for the IO subsystem.
625  * Please do not alter this order without good reasons and regression
626  * testing. Specifically, as large blocks of memory are subdivided,
627  * the order in which smaller blocks are delivered depends on the order
628  * they're subdivided in this function. This is the primary factor
629  * influencing the order in which pages are delivered to the IO
630  * subsystem according to empirical testing, and this is also justified
631  * by considering the behavior of a buddy system containing a single
632  * large block of memory acted on by a series of small allocations.
633  * This behavior is a critical factor in sglist merging's success.
634  *
635  * -- wli
636  */
637 static inline void expand(struct zone *zone, struct page *page,
638         int low, int high, struct free_area *area,
639         int migratetype)
640 {
641         unsigned long size = 1 << high;
642 
643         while (high > low) {
644                 area--;
645                 high--;
646                 size >>= 1;
647                 VM_BUG_ON(bad_range(zone, &page[size]));
648                 list_add(&page[size].lru, &area->free_list[migratetype]);
649                 area->nr_free++;
650                 set_page_order(&page[size], high);
651         }
652 }
653 
654 /*
655  * This page is about to be returned from the page allocator
656  */
657 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
658 {
659         if (unlikely(page_mapcount(page) |
660                 (page->mapping != NULL)  |
661                 (page_get_page_cgroup(page) != NULL) |
662                 (page_count(page) != 0)  |
663                 (page->flags & (
664                         1 << PG_lru     |
665                         1 << PG_private |
666                         1 << PG_locked  |
667                         1 << PG_active  |
668                         1 << PG_dirty   |
669                         1 << PG_slab    |
670                         1 << PG_swapcache |
671                         1 << PG_writeback |
672                         1 << PG_reserved |
673                         1 << PG_buddy ))))
674                 bad_page(page);
675 
676         /*
677          * For now, we report if PG_reserved was found set, but do not
678          * clear it, and do not allocate the page: as a safety net.
679          */
680         if (PageReserved(page))
681                 return 1;
682 
683         page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_readahead |
684                         1 << PG_referenced | 1 << PG_arch_1 |
685                         1 << PG_owner_priv_1 | 1 << PG_mappedtodisk);
686         set_page_private(page, 0);
687         set_page_refcounted(page);
688 
689         arch_alloc_page(page, order);
690         kernel_map_pages(page, 1 << order, 1);
691 
692         if (gfp_flags & __GFP_ZERO)
693                 prep_zero_page(page, order, gfp_flags);
694 
695         if (order && (gfp_flags & __GFP_COMP))
696                 prep_compound_page(page, order);
697 
698         return 0;
699 }
700 
701 /*
702  * Go through the free lists for the given migratetype and remove
703  * the smallest available page from the freelists
704  */
705 static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
706                                                 int migratetype)
707 {
708         unsigned int current_order;
709         struct free_area * area;
710         struct page *page;
711 
712         /* Find a page of the appropriate size in the preferred list */
713         for (current_order = order; current_order < MAX_ORDER; ++current_order) {
714                 area = &(zone->free_area[current_order]);
715                 if (list_empty(&area->free_list[migratetype]))
716                         continue;
717 
718                 page = list_entry(area->free_list[migratetype].next,
719                                                         struct page, lru);
720                 list_del(&page->lru);
721                 rmv_page_order(page);
722                 area->nr_free--;
723                 __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
724                 expand(zone, page, order, current_order, area, migratetype);
725                 return page;
726         }
727 
728         return NULL;
729 }
730 
731 
732 /*
733  * This array describes the order lists are fallen back to when
734  * the free lists for the desirable migrate type are depleted
735  */
736 static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
737         [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_RESERVE },
738         [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_RESERVE },
739         [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
740         [MIGRATE_RESERVE]     = { MIGRATE_RESERVE,     MIGRATE_RESERVE,   MIGRATE_RESERVE }, /* Never used */
741 };
742 
743 /*
744  * Move the free pages in a range to the free lists of the requested type.
745  * Note that start_page and end_pages are not aligned on a pageblock
746  * boundary. If alignment is required, use move_freepages_block()
747  */
748 int move_freepages(struct zone *zone,
749                         struct page *start_page, struct page *end_page,
750                         int migratetype)
751 {
752         struct page *page;
753         unsigned long order;
754         int pages_moved = 0;
755 
756 #ifndef CONFIG_HOLES_IN_ZONE
757         /*
758          * page_zone is not safe to call in this context when
759          * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
760          * anyway as we check zone boundaries in move_freepages_block().
761          * Remove at a later date when no bug reports exist related to
762          * grouping pages by mobility
763          */
764         BUG_ON(page_zone(start_page) != page_zone(end_page));
765 #endif
766 
767         for (page = start_page; page <= end_page;) {
768                 if (!pfn_valid_within(page_to_pfn(page))) {
769                         page++;
770                         continue;
771                 }
772 
773                 if (!PageBuddy(page)) {
774                         page++;
775                         continue;
776                 }
777 
778                 order = page_order(page);
779                 list_del(&page->lru);
780                 list_add(&page->lru,
781                         &zone->free_area[order].free_list[migratetype]);
782                 page += 1 << order;
783                 pages_moved += 1 << order;
784         }
785 
786         return pages_moved;
787 }
788 
789 int move_freepages_block(struct zone *zone, struct page *page, int migratetype)
790 {
791         unsigned long start_pfn, end_pfn;
792         struct page *start_page, *end_page;
793 
794         start_pfn = page_to_pfn(page);
795         start_pfn = start_pfn & ~(pageblock_nr_pages-1);
796         start_page = pfn_to_page(start_pfn);
797         end_page = start_page + pageblock_nr_pages - 1;
798         end_pfn = start_pfn + pageblock_nr_pages - 1;
799 
800         /* Do not cross zone boundaries */
801         if (start_pfn < zone->zone_start_pfn)
802                 start_page = page;
803         if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
804                 return 0;
805 
806         return move_freepages(zone, start_page, end_page, migratetype);
807 }
808 
809 /* Remove an element from the buddy allocator from the fallback list */
810 static struct page *__rmqueue_fallback(struct zone *zone, int order,
811                                                 int start_migratetype)
812 {
813         struct free_area * area;
814         int current_order;
815         struct page *page;
816         int migratetype, i;
817 
818         /* Find the largest possible block of pages in the other list */
819         for (current_order = MAX_ORDER-1; current_order >= order;
820                                                 --current_order) {
821                 for (i = 0; i < MIGRATE_TYPES - 1; i++) {
822                         migratetype = fallbacks[start_migratetype][i];
823 
824                         /* MIGRATE_RESERVE handled later if necessary */
825                         if (migratetype == MIGRATE_RESERVE)
826                                 continue;
827 
828                         area = &(zone->free_area[current_order]);
829                         if (list_empty(&area->free_list[migratetype]))
830                                 continue;
831 
832                         page = list_entry(area->free_list[migratetype].next,
833                                         struct page, lru);
834                         area->nr_free--;
835 
836                         /*
837                          * If breaking a large block of pages, move all free
838                          * pages to the preferred allocation list. If falling
839                          * back for a reclaimable kernel allocation, be more
840                          * agressive about taking ownership of free pages
841                          */
842                         if (unlikely(current_order >= (pageblock_order >> 1)) ||
843                                         start_migratetype == MIGRATE_RECLAIMABLE) {
844                                 unsigned long pages;
845                                 pages = move_freepages_block(zone, page,
846                                                                 start_migratetype);
847 
848                                 /* Claim the whole block if over half of it is free */
849                                 if (pages >= (1 << (pageblock_order-1)))
850                                         set_pageblock_migratetype(page,
851                                                                 start_migratetype);
852 
853                                 migratetype = start_migratetype;
854                         }
855 
856                         /* Remove the page from the freelists */
857                         list_del(&page->lru);
858                         rmv_page_order(page);
859                         __mod_zone_page_state(zone, NR_FREE_PAGES,
860                                                         -(1UL << order));
861 
862                         if (current_order == pageblock_order)
863                                 set_pageblock_migratetype(page,
864                                                         start_migratetype);
865 
866                         expand(zone, page, order, current_order, area, migratetype);
867                         return page;
868                 }
869         }
870 
871         /* Use MIGRATE_RESERVE rather than fail an allocation */
872         return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
873 }
874 
875 /*
876  * Do the hard work of removing an element from the buddy allocator.
877  * Call me with the zone->lock already held.
878  */
879 static struct page *__rmqueue(struct zone *zone, unsigned int order,
880                                                 int migratetype)
881 {
882         struct page *page;
883 
884         page = __rmqueue_smallest(zone, order, migratetype);
885 
886         if (unlikely(!page))
887                 page = __rmqueue_fallback(zone, order, migratetype);
888 
889         return page;
890 }
891 
892 /* 
893  * Obtain a specified number of elements from the buddy allocator, all under
894  * a single hold of the lock, for efficiency.  Add them to the supplied list.
895  * Returns the number of new pages which were placed at *list.
896  */
897 static int rmqueue_bulk(struct zone *zone, unsigned int order, 
898                         unsigned long count, struct list_head *list,
899                         int migratetype)
900 {
901         int i;
902         
903         spin_lock(&zone->lock);
904         for (i = 0; i < count; ++i) {
905                 struct page *page = __rmqueue(zone, order, migratetype);
906                 if (unlikely(page == NULL))
907                         break;
908 
909                 /*
910                  * Split buddy pages returned by expand() are received here
911                  * in physical page order. The page is added to the callers and
912                  * list and the list head then moves forward. From the callers
913                  * perspective, the linked list is ordered by page number in
914                  * some conditions. This is useful for IO devices that can
915                  * merge IO requests if the physical pages are ordered
916                  * properly.
917                  */
918                 list_add(&page->lru, list);
919                 set_page_private(page, migratetype);
920                 list = &page->lru;
921         }
922         spin_unlock(&zone->lock);
923         return i;
924 }
925 
926 #ifdef CONFIG_NUMA
927 /*
928  * Called from the vmstat counter updater to drain pagesets of this
929  * currently executing processor on remote nodes after they have
930  * expired.
931  *
932  * Note that this function must be called with the thread pinned to
933  * a single processor.
934  */
935 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
936 {
937         unsigned long flags;
938         int to_drain;
939         int this_cpu;
940 
941         lock_cpu_pcp(&flags, &this_cpu);
942         if (pcp->count >= pcp->batch)
943                 to_drain = pcp->batch;
944         else
945                 to_drain = pcp->count;
946         free_pages_bulk(zone, to_drain, &pcp->list, 0);
947         pcp->count -= to_drain;
948         unlock_cpu_pcp(flags, this_cpu);
949 }
950 #endif
951 
952 /*
953  * Drain pages of the indicated processor.
954  *
955  * The processor must either be the current processor and the
956  * thread pinned to the current processor or a processor that
957  * is not online.
958  */
959 static void drain_pages(unsigned int cpu)
960 {
961         unsigned long flags;
962         struct zone *zone;
963 
964         for_each_zone(zone) {
965                 struct per_cpu_pageset *pset;
966                 struct per_cpu_pages *pcp;
967 
968                 if (!populated_zone(zone))
969                         continue;
970 
971                 pset = zone_pcp(zone, cpu);
972                 if (!pset) {
973                         WARN_ON(1);
974                         continue;
975                 }
976                 pcp = &pset->pcp;
977                 lock_cpu_pcp(&flags, &cpu);
978                 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
979                 pcp->count = 0;
980                 unlock_cpu_pcp(flags, cpu);
981         }
982 }
983 
984 /*
985  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
986  */
987 void drain_local_pages(void *arg)
988 {
989         drain_pages(smp_processor_id());
990 }
991 
992 /*
993  * Spill all the per-cpu pages from all CPUs back into the buddy allocator
994  */
995 void drain_all_pages(void)
996 {
997 #ifdef CONFIG_PREEMPT_RT
998         /*
999          * HACK!!!!!
1000          *  For RT we can't use IPIs to run drain_local_pages, since
1001          *  that code will call spin_locks that will now sleep.
1002          *  But, schedule_on_each_cpu will call kzalloc, which will
1003          *  call page_alloc which was what calls this.
1004          *
1005          *  Luckily, there's a condition to get here, and that is if
1006          *  the order passed in to alloc_pages is greater than 0
1007          *  (alloced more than a page size).  The slabs only allocate
1008          *  what is needed, and the allocation made by schedule_on_each_cpu
1009          *  does an alloc of "sizeof(void *)*nr_cpu_ids".
1010          *
1011          *  So we can safely call schedule_on_each_cpu if that number
1012          *  is less than a page. Otherwise don't bother. At least warn of
1013          *  this issue.
1014          *
1015          * And yes, this is one big hack.  Please fix ;-)
1016          */
1017         if (sizeof(void *)*nr_cpu_ids < PAGE_SIZE)
1018                 schedule_on_each_cpu(drain_local_pages, NULL, 0, 1);
1019         else {
1020                 static int once;
1021                 if (!once) {
1022                         printk(KERN_ERR "Can't drain all CPUS due to possible recursion\n");
1023                         once = 1;
1024                 }
1025                 drain_local_pages(NULL);
1026         }
1027 
1028 #else
1029         on_each_cpu(drain_local_pages, NULL, 0, 1);
1030 #endif
1031 }
1032 
1033 #ifdef CONFIG_HIBERNATION
1034 
1035 void mark_free_pages(struct zone *zone)
1036 {
1037         unsigned long pfn, max_zone_pfn;
1038         unsigned long flags;
1039         int order, t;
1040         struct list_head *curr;
1041 
1042         if (!zone->spanned_pages)
1043                 return;
1044 
1045         spin_lock_irqsave(&zone->lock, flags);
1046 
1047         max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1048         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1049                 if (pfn_valid(pfn)) {
1050                         struct page *page = pfn_to_page(pfn);
1051 
1052                         if (!swsusp_page_is_forbidden(page))
1053                                 swsusp_unset_page_free(page);
1054                 }
1055 
1056         for_each_migratetype_order(order, t) {
1057                 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1058                         unsigned long i;
1059 
1060                         pfn = page_to_pfn(list_entry(curr, struct page, lru));
1061                         for (i = 0; i < (1UL << order); i++)
1062                                 swsusp_set_page_free(pfn_to_page(pfn + i));
1063                 }
1064         }
1065         spin_unlock_irqrestore(&zone->lock, flags);
1066 }
1067 #endif /* CONFIG_PM */
1068 
1069 /*
1070  * Free a 0-order page
1071  */
1072 static void free_hot_cold_page(struct page *page, int cold)
1073 {
1074         struct zone *zone = page_zone(page);
1075         struct per_cpu_pageset *pset;
1076         struct per_cpu_pages *pcp;
1077         unsigned long flags;
1078         int this_cpu;
1079 
1080         if (PageAnon(page))
1081                 page->mapping = NULL;
1082         if (free_pages_check(page))
1083                 return;
1084 
1085         if (!PageHighMem(page))
1086                 debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1087         arch_free_page(page, 0);
1088         kernel_map_pages(page, 1, 0);
1089 
1090         pset = get_zone_pcp(zone, &flags, &this_cpu);
1091         pcp = &pset->pcp;
1092 
1093         count_vm_event(PGFREE);
1094 
1095         if (cold)
1096                 list_add_tail(&page->lru, &pcp->list);
1097         else
1098                 list_add(&page->lru, &pcp->list);
1099         set_page_private(page, get_pageblock_migratetype(page));
1100         pcp->count++;
1101         if (pcp->count >= pcp->high) {
1102                 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1103                 pcp->count -= pcp->batch;
1104         }
1105         put_zone_pcp(zone, flags, this_cpu);
1106 }
1107 
1108 void free_hot_page(struct page *page)
1109 {
1110         free_hot_cold_page(page, 0);
1111 }
1112         
1113 void free_cold_page(struct page *page)
1114 {
1115         free_hot_cold_page(page, 1);
1116 }
1117 
1118 /*
1119  * split_page takes a non-compound higher-order page, and splits it into
1120  * n (1<<order) sub-pages: page[0..n]
1121  * Each sub-page must be freed individually.
1122  *
1123  * Note: this is probably too low level an operation for use in drivers.
1124  * Please consult with lkml before using this in your driver.
1125  */
1126 void split_page(struct page *page, unsigned int order)
1127 {
1128         int i;
1129 
1130         VM_BUG_ON(PageCompound(page));
1131         VM_BUG_ON(!page_count(page));
1132         for (i = 1; i < (1 << order); i++)
1133                 set_page_refcounted(page + i);
1134 }
1135 
1136 /*
1137  * Really, prep_compound_page() should be called from __rmqueue_bulk().  But
1138  * we cheat by calling it from here, in the order > 0 path.  Saves a branch
1139  * or two.
1140  */
1141 static struct page *buffered_rmqueue(struct zonelist *zonelist,
1142                         struct zone *zone, int order, gfp_t gfp_flags)
1143 {
1144         unsigned long flags;
1145         struct page *page;
1146         int cold = !!(gfp_flags & __GFP_COLD);
1147         struct per_cpu_pageset *pset;
1148         int migratetype = allocflags_to_migratetype(gfp_flags);
1149         int this_cpu;
1150 
1151 again:
1152         pset = get_zone_pcp(zone, &flags, &this_cpu);
1153         if (likely(order == 0)) {
1154                 struct per_cpu_pages *pcp = &pset->pcp;
1155 
1156                 if (!pcp->count) {
1157                         pcp->count = rmqueue_bulk(zone, 0,
1158                                         pcp->batch, &pcp->list, migratetype);
1159                         if (unlikely(!pcp->count))
1160                                 goto failed;
1161                 }
1162 
1163                 /* Find a page of the appropriate migrate type */
1164                 if (cold) {
1165                         list_for_each_entry_reverse(page, &pcp->list, lru)
1166                                 if (page_private(page) == migratetype)
1167                                         break;
1168                 } else {
1169                         list_for_each_entry(page, &pcp->list, lru)
1170                                 if (page_private(page) == migratetype)
1171                                         break;
1172                 }
1173 
1174                 /* Allocate more to the pcp list if necessary */
1175                 if (unlikely(&page->lru == &pcp->list)) {
1176                         pcp->count += rmqueue_bulk(zone, 0,
1177                                         pcp->batch, &pcp->list, migratetype);
1178                         page = list_entry(pcp->list.next, struct page, lru);
1179                 }
1180 
1181                 list_del(&page->lru);
1182                 pcp->count--;
1183         } else {
1184                 spin_lock(&zone->lock);
1185                 page = __rmqueue(zone, order, migratetype);
1186                 spin_unlock(&zone->lock);
1187                 if (!page)
1188                         goto failed;
1189         }
1190 
1191         __count_zone_vm_events(PGALLOC, zone, 1 << order);
1192         zone_statistics(zonelist, zone);
1193         put_zone_pcp(zone, flags, this_cpu);
1194 
1195         VM_BUG_ON(bad_range(zone, page));
1196         if (prep_new_page(page, order, gfp_flags))
1197                 goto again;
1198         return page;
1199 
1200 failed:
1201         put_zone_pcp(zone, flags, this_cpu);
1202         return NULL;
1203 }
1204 
1205 #define ALLOC_NO_WATERMARKS     0x01 /* don't check watermarks at all */
1206 #define ALLOC_WMARK_MIN         0x02 /* use pages_min watermark */
1207 #define ALLOC_WMARK_LOW         0x04 /* use pages_low watermark */
1208 #define ALLOC_WMARK_HIGH        0x08 /* use pages_high watermark */
1209 #define ALLOC_HARDER            0x10 /* try to alloc harder */
1210 #define ALLOC_HIGH              0x20 /* __GFP_HIGH set */
1211 #define ALLOC_CPUSET            0x40 /* check for correct cpuset */
1212 
1213 #ifdef CONFIG_FAIL_PAGE_ALLOC
1214 
1215 static struct fail_page_alloc_attr {
1216         struct fault_attr attr;
1217 
1218         u32 ignore_gfp_highmem;
1219         u32 ignore_gfp_wait;
1220         u32 min_order;
1221 
1222 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1223 
1224         struct dentry *ignore_gfp_highmem_file;
1225         struct dentry *ignore_gfp_wait_file;
1226         struct dentry *min_order_file;
1227 
1228 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1229 
1230 } fail_page_alloc = {
1231         .attr = FAULT_ATTR_INITIALIZER,
1232         .ignore_gfp_wait = 1,
1233         .ignore_gfp_highmem = 1,
1234         .min_order = 1,
1235 };
1236 
1237 static int __init setup_fail_page_alloc(char *str)
1238 {
1239         return setup_fault_attr(&fail_page_alloc.attr, str);
1240 }
1241 __setup("fail_page_alloc=", setup_fail_page_alloc);
1242 
1243 static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1244 {
1245         if (order < fail_page_alloc.min_order)
1246                 return 0;
1247         if (gfp_mask & __GFP_NOFAIL)
1248                 return 0;
1249         if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1250                 return 0;
1251         if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1252                 return 0;
1253 
1254         return should_fail(&fail_page_alloc.attr, 1 << order);
1255 }
1256 
1257 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1258 
1259 static int __init fail_page_alloc_debugfs(void)
1260 {
1261         mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1262         struct dentry *dir;
1263         int err;
1264 
1265         err = init_fault_attr_dentries(&fail_page_alloc.attr,
1266                                        "fail_page_alloc");
1267         if (err)
1268                 return err;
1269         dir = fail_page_alloc.attr.dentries.dir;
1270 
1271         fail_page_alloc.ignore_gfp_wait_file =
1272                 debugfs_create_bool("ignore-gfp-wait", mode, dir,
1273                                       &fail_page_alloc.ignore_gfp_wait);
1274 
1275         fail_page_alloc.ignore_gfp_highmem_file =
1276                 debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1277                                       &fail_page_alloc.ignore_gfp_highmem);
1278         fail_page_alloc.min_order_file =
1279                 debugfs_create_u32("min-order", mode, dir,
1280                                    &fail_page_alloc.min_order);
1281 
1282         if (!fail_page_alloc.ignore_gfp_wait_file ||
1283             !fail_page_alloc.ignore_gfp_highmem_file ||
1284             !fail_page_alloc.min_order_file) {
1285                 err = -ENOMEM;
1286                 debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1287                 debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1288                 debugfs_remove(fail_page_alloc.min_order_file);
1289                 cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1290         }
1291 
1292         return err;
1293 }
1294 
1295 late_initcall(fail_page_alloc_debugfs);
1296 
1297 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1298 
1299 #else /* CONFIG_FAIL_PAGE_ALLOC */
1300 
1301 static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1302 {
1303         return 0;
1304 }
1305 
1306 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1307 
1308 /*
1309  * Return 1 if free pages are above 'mark'. This takes into account the order
1310  * of the allocation.
1311  */
1312 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1313                       int classzone_idx, int alloc_flags)
1314 {
1315         /* free_pages my go negative - that's OK */
1316         long min = mark;
1317         long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1318         int o;
1319 
1320         if (alloc_flags & ALLOC_HIGH)
1321                 min -= min / 2;
1322         if (alloc_flags & ALLOC_HARDER)
1323                 min -= min / 4;
1324 
1325         if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1326                 return 0;
1327         for (o = 0; o < order; o++) {
1328                 /* At the next order, this order's pages become unavailable */
1329                 free_pages -= z->free_area[o].nr_free << o;
1330 
1331                 /* Require fewer higher order pages to be free */
1332                 min >>= 1;
1333 
1334                 if (free_pages <= min)
1335                         return 0;
1336         }
1337         return 1;
1338 }
1339 
1340 #ifdef CONFIG_NUMA
1341 /*
1342  * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
1343  * skip over zones that are not allowed by the cpuset, or that have
1344  * been recently (in last second) found to be nearly full.  See further
1345  * comments in mmzone.h.  Reduces cache footprint of zonelist scans
1346  * that have to skip over a lot of full or unallowed zones.
1347  *
1348  * If the zonelist cache is present in the passed in zonelist, then
1349  * returns a pointer to the allowed node mask (either the current
1350  * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1351  *
1352  * If the zonelist cache is not available for this zonelist, does
1353  * nothing and returns NULL.
1354  *
1355  * If the fullzones BITMAP in the zonelist cache is stale (more than
1356  * a second since last zap'd) then we zap it out (clear its bits.)
1357  *
1358  * We hold off even calling zlc_setup, until after we've checked the
1359  * first zone in the zonelist, on the theory that most allocations will
1360  * be satisfied from that first zone, so best to examine that zone as
1361  * quickly as we can.
1362  */
1363 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1364 {
1365         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1366         nodemask_t *allowednodes;       /* zonelist_cache approximation */
1367 
1368         zlc = zonelist->zlcache_ptr;
1369         if (!zlc)
1370                 return NULL;
1371 
1372        if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1373                 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1374                 zlc->last_full_zap = jiffies;
1375         }
1376 
1377         allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1378                                         &cpuset_current_mems_allowed :
1379                                         &node_states[N_HIGH_MEMORY];
1380         return allowednodes;
1381 }
1382 
1383 /*
1384  * Given 'z' scanning a zonelist, run a couple of quick checks to see
1385  * if it is worth looking at further for free memory:
1386  *  1) Check that the zone isn't thought to be full (doesn't have its
1387  *     bit set in the zonelist_cache fullzones BITMAP).
1388  *  2) Check that the zones node (obtained from the zonelist_cache
1389  *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1390  * Return true (non-zero) if zone is worth looking at further, or
1391  * else return false (zero) if it is not.
1392  *
1393  * This check -ignores- the distinction between various watermarks,
1394  * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
1395  * found to be full for any variation of these watermarks, it will
1396  * be considered full for up to one second by all requests, unless
1397  * we are so low on memory on all allowed nodes that we are forced
1398  * into the second scan of the zonelist.
1399  *
1400  * In the second scan we ignore this zonelist cache and exactly
1401  * apply the watermarks to all zones, even it is slower to do so.
1402  * We are low on memory in the second scan, and should leave no stone
1403  * unturned looking for a free page.
1404  */
1405 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1406                                                 nodemask_t *allowednodes)
1407 {
1408         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1409         int i;                          /* index of *z in zonelist zones */
1410         int n;                          /* node that zone *z is on */
1411 
1412         zlc = zonelist->zlcache_ptr;
1413         if (!zlc)
1414                 return 1;
1415 
1416         i = z - zonelist->zones;
1417         n = zlc->z_to_n[i];
1418 
1419         /* This zone is worth trying if it is allowed but not full */
1420         return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1421 }
1422 
1423 /*
1424  * Given 'z' scanning a zonelist, set the corresponding bit in
1425  * zlc->fullzones, so that subsequent attempts to allocate a page
1426  * from that zone don't waste time re-examining it.
1427  */
1428 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1429 {
1430         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1431         int i;                          /* index of *z in zonelist zones */
1432 
1433         zlc = zonelist->zlcache_ptr;
1434         if (!zlc)
1435                 return;
1436 
1437         i = z - zonelist->zones;
1438 
1439         set_bit(i, zlc->fullzones);
1440 }
1441 
1442 #else   /* CONFIG_NUMA */
1443 
1444 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1445 {
1446         return NULL;
1447 }
1448 
1449 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z,
1450                                 nodemask_t *allowednodes)
1451 {
1452         return 1;
1453 }
1454 
1455 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z)
1456 {
1457 }
1458 #endif  /* CONFIG_NUMA */
1459 
1460 /*
1461  * get_page_from_freelist goes through the zonelist trying to allocate
1462  * a page.
1463  */
1464 static struct page *
1465 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
1466                 struct zonelist *zonelist, int alloc_flags)
1467 {
1468         struct zone **z;
1469         struct page *page = NULL;
1470         int classzone_idx = zone_idx(zonelist->zones[0]);
1471         struct zone *zone;
1472         nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1473         int zlc_active = 0;             /* set if using zonelist_cache */
1474         int did_zlc_setup = 0;          /* just call zlc_setup() one time */
1475         enum zone_type highest_zoneidx = -1; /* Gets set for policy zonelists */
1476 
1477 zonelist_scan:
1478         /*
1479          * Scan zonelist, looking for a zone with enough free.
1480          * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1481          */
1482         z = zonelist->zones;
1483 
1484         do {
1485                 /*
1486                  * In NUMA, this could be a policy zonelist which contains
1487                  * zones that may not be allowed by the current gfp_mask.
1488                  * Check the zone is allowed by the current flags
1489                  */
1490                 if (unlikely(alloc_should_filter_zonelist(zonelist))) {
1491                         if (highest_zoneidx == -1)
1492                                 highest_zoneidx = gfp_zone(gfp_mask);
1493                         if (zone_idx(*z) > highest_zoneidx)
1494                                 continue;
1495                 }
1496 
1497                 if (NUMA_BUILD && zlc_active &&
1498                         !zlc_zone_worth_trying(zonelist, z, allowednodes))
1499                                 continue;
1500                 zone = *z;
1501                 if ((alloc_flags & ALLOC_CPUSET) &&
1502                         !cpuset_zone_allowed_softwall(zone, gfp_mask))
1503                                 goto try_next_zone;
1504 
1505                 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1506                         unsigned long mark;
1507                         if (alloc_flags & ALLOC_WMARK_MIN)
1508                                 mark = zone->pages_min;
1509                         else if (alloc_flags & ALLOC_WMARK_LOW)
1510                                 mark = zone->pages_low;
1511                         else
1512                                 mark = zone->pages_high;
1513                         if (!zone_watermark_ok(zone, order, mark,
1514                                     classzone_idx, alloc_flags)) {
1515                                 if (!zone_reclaim_mode ||
1516                                     !zone_reclaim(zone, gfp_mask, order))
1517                                         goto this_zone_full;
1518                         }
1519                 }
1520 
1521                 page = buffered_rmqueue(zonelist, zone, order, gfp_mask);
1522                 if (page)
1523                         break;
1524 this_zone_full:
1525                 if (NUMA_BUILD)
1526                         zlc_mark_zone_full(zonelist, z);
1527 try_next_zone:
1528                 if (NUMA_BUILD && !did_zlc_setup) {
1529                         /* we do zlc_setup after the first zone is tried */
1530                         allowednodes = zlc_setup(zonelist, alloc_flags);
1531                         zlc_active = 1;
1532                         did_zlc_setup = 1;
1533                 }
1534         } while (*(++z) != NULL);
1535 
1536         if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1537                 /* Disable zlc cache for second zonelist scan */
1538                 zlc_active = 0;
1539                 goto zonelist_scan;
1540         }
1541         return page;
1542 }
1543 
1544 /*
1545  * This is the 'heart' of the zoned buddy allocator.
1546  */
1547 struct page *
1548 __alloc_pages(gfp_t gfp_mask, unsigned int order,
1549                 struct zonelist *zonelist)
1550 {
1551         const gfp_t wait = gfp_mask & __GFP_WAIT;
1552         struct zone **z;
1553         struct page *page;
1554         struct reclaim_state reclaim_state;
1555         struct task_struct *p = current;
1556         int do_retry;
1557         int alloc_flags;
1558         int did_some_progress;
1559 
1560         might_sleep_if(wait);
1561 
1562         if (should_fail_alloc_page(gfp_mask, order))
1563                 return NULL;
1564 
1565 restart:
1566         z = zonelist->zones;  /* the list of zones suitable for gfp_mask */
1567 
1568         if (unlikely(*z == NULL)) {
1569                 /*
1570                  * Happens if we have an empty zonelist as a result of
1571                  * GFP_THISNODE being used on a memoryless node
1572                  */
1573                 return NULL;
1574         }
1575 
1576         page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1577                                 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1578         if (page)
1579                 goto got_pg;
1580 
1581         /*
1582          * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1583          * __GFP_NOWARN set) should not cause reclaim since the subsystem
1584          * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1585          * using a larger set of nodes after it has established that the
1586          * allowed per node queues are empty and that nodes are
1587          * over allocated.
1588          */
1589         if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1590                 goto nopage;
1591 
1592         for (z = zonelist->zones; *z; z++)
1593                 wakeup_kswapd(*z, order);
1594 
1595         /*
1596          * OK, we're below the kswapd watermark and have kicked background
1597          * reclaim. Now things get more complex, so set up alloc_flags according
1598          * to how we want to proceed.
1599          *
1600          * The caller may dip into page reserves a bit more if the caller
1601          * cannot run direct reclaim, or if the caller has realtime scheduling
1602          * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
1603          * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1604          */
1605         alloc_flags = ALLOC_WMARK_MIN;
1606         if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1607                 alloc_flags |= ALLOC_HARDER;
1608         if (gfp_mask & __GFP_HIGH)
1609                 alloc_flags |= ALLOC_HIGH;
1610         if (wait)
1611                 alloc_flags |= ALLOC_CPUSET;
1612 
1613         /*
1614          * Go through the zonelist again. Let __GFP_HIGH and allocations
1615          * coming from realtime tasks go deeper into reserves.
1616          *
1617          * This is the last chance, in general, before the goto nopage.
1618          * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1619          * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1620          */
1621         page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
1622         if (page)
1623                 goto got_pg;
1624 
1625         /* This allocation should allow future memory freeing. */
1626 
1627 rebalance:
1628         if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1629                         && !in_interrupt()) {
1630                 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1631 nofail_alloc:
1632                         /* go through the zonelist yet again, ignoring mins */
1633                         page = get_page_from_freelist(gfp_mask, order,
1634                                 zonelist, ALLOC_NO_WATERMARKS);
1635                         if (page)
1636                                 goto got_pg;
1637                         if (gfp_mask & __GFP_NOFAIL) {
1638                                 congestion_wait(WRITE, HZ/50);
1639                                 goto nofail_alloc;
1640                         }
1641                 }
1642                 goto nopage;
1643         }
1644 
1645         /* Atomic allocations - we can't balance anything */
1646         if (!wait)
1647                 goto nopage;
1648 
1649         cond_resched();
1650 
1651         /* We now go into synchronous reclaim */
1652         cpuset_memory_pressure_bump();
1653         p->flags |= PF_MEMALLOC;
1654         reclaim_state.reclaimed_slab = 0;
1655         p->reclaim_state = &reclaim_state;
1656 
1657         did_some_progress = try_to_free_pages(zonelist->zones, order, gfp_mask);
1658 
1659         p->reclaim_state = NULL;
1660         p->flags &= ~PF_MEMALLOC;
1661 
1662         cond_resched();
1663 
1664         if (order != 0)
1665                 drain_all_pages();
1666 
1667         if (likely(did_some_progress)) {
1668                 page = get_page_from_freelist(gfp_mask, order,
1669                                                 zonelist, alloc_flags);
1670                 if (page)
1671                         goto got_pg;
1672         } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1673                 if (!try_set_zone_oom(zonelist)) {
1674                         schedule_timeout_uninterruptible(1);
1675                         goto restart;
1676                 }
1677 
1678                 /*
1679                  * Go through the zonelist yet one more time, keep
1680                  * very high watermark here, this is only to catch
1681                  * a parallel oom killing, we must fail if we're still
1682                  * under heavy pressure.
1683                  */
1684                 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1685                                 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1686                 if (page) {
1687                         clear_zonelist_oom(zonelist);
1688                         goto got_pg;
1689                 }
1690 
1691                 /* The OOM killer will not help higher order allocs so fail */
1692                 if (order > PAGE_ALLOC_COSTLY_ORDER) {
1693                         clear_zonelist_oom(zonelist);
1694                         goto nopage;
1695                 }
1696 
1697                 out_of_memory(zonelist, gfp_mask, order);
1698                 clear_zonelist_oom(zonelist);
1699                 goto restart;
1700         }
1701 
1702         /*
1703          * Don't let big-order allocations loop unless the caller explicitly
1704          * requests that.  Wait for some write requests to complete then retry.
1705          *
1706          * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1707          * <= 3, but that may not be true in other implementations.
1708          */
1709         do_retry = 0;
1710         if (!(gfp_mask & __GFP_NORETRY)) {
1711                 if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
1712                                                 (gfp_mask & __GFP_REPEAT))
1713                         do_retry = 1;
1714                 if (gfp_mask & __GFP_NOFAIL)
1715                         do_retry = 1;
1716         }
1717         if (do_retry) {
1718                 congestion_wait(WRITE, HZ/50);
1719                 goto rebalance;
1720         }
1721 
1722 nopage:
1723         if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1724                 printk(KERN_WARNING "%s: page allocation failure."
1725                         " order:%d, mode:0x%x\n",
1726                         p->comm, order, gfp_mask);
1727                 dump_stack();
1728                 show_mem();
1729         }
1730 got_pg:
1731         return page;
1732 }
1733 
1734 EXPORT_SYMBOL(__alloc_pages);
1735 
1736 /*
1737  * Common helper functions.
1738  */
1739 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1740 {
1741         struct page * page;
1742         page = alloc_pages(gfp_mask, order);
1743         if (!page)
1744                 return 0;
1745         return (unsigned long) page_address(page);
1746 }
1747 
1748 EXPORT_SYMBOL(__get_free_pages);
1749 
1750 unsigned long get_zeroed_page(gfp_t gfp_mask)
1751 {
1752         struct page * page;
1753 
1754         /*
1755          * get_zeroed_page() returns a 32-bit address, which cannot represent
1756          * a highmem page
1757          */
1758         VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1759 
1760         page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1761         if (page)
1762                 return (unsigned long) page_address(page);
1763         return 0;
1764 }
1765 
1766 EXPORT_SYMBOL(get_zeroed_page);
1767 
1768 void __pagevec_free(struct pagevec *pvec)
1769 {
1770         int i = pagevec_count(pvec);
1771 
1772         while (--i >= 0)
1773                 free_hot_cold_page(pvec->pages[i], pvec->cold);
1774 }
1775 
1776 void __free_pages(struct page *page, unsigned int order)
1777 {
1778         if (put_page_testzero(page)) {
1779                 if (order == 0)
1780                         free_hot_page(page);
1781                 else
1782                         __free_pages_ok(page, order);
1783         }
1784 }
1785 
1786 EXPORT_SYMBOL(__free_pages);
1787 
1788 void free_pages(unsigned long addr, unsigned int order)
1789 {
1790         if (addr != 0) {
1791                 VM_BUG_ON(!virt_addr_valid((void *)addr));
1792                 __free_pages(virt_to_page((void *)addr), order);
1793         }
1794 }
1795 
1796 EXPORT_SYMBOL(free_pages);
1797 
1798 static unsigned int nr_free_zone_pages(int offset)
1799 {
1800         /* Just pick one node, since fallback list is circular */
1801         pg_data_t *pgdat = NODE_DATA(numa_node_id());
1802         unsigned int sum = 0;
1803 
1804         struct zonelist *zonelist = pgdat->node_zonelists + offset;
1805         struct zone **zonep = zonelist->zones;
1806         struct zone *zone;
1807 
1808         for (zone = *zonep++; zone; zone = *zonep++) {
1809                 unsigned long size = zone->present_pages;
1810                 unsigned long high = zone->pages_high;
1811                 if (size > high)
1812                         sum += size - high;
1813         }
1814 
1815         return sum;
1816 }
1817 
1818 /*
1819  * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1820  */
1821 unsigned int nr_free_buffer_pages(void)
1822 {
1823         return nr_free_zone_pages(gfp_zone(GFP_USER));
1824 }
1825 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1826 
1827 /*
1828  * Amount of free RAM allocatable within all zones
1829  */
1830 unsigned int nr_free_pagecache_pages(void)
1831 {
1832         return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1833 }
1834 
1835 static inline void show_node(struct zone *zone)
1836 {
1837         if (NUMA_BUILD)
1838                 printk("Node %d ", zone_to_nid(zone));
1839 }
1840 
1841 void si_meminfo(struct sysinfo *val)
1842 {
1843         val->totalram = totalram_pages;
1844         val->sharedram = 0;
1845         val->freeram = global_page_state(NR_FREE_PAGES);
1846         val->bufferram = nr_blockdev_pages();
1847         val->totalhigh = totalhigh_pages;
1848         val->freehigh = nr_free_highpages();
1849         val->mem_unit = PAGE_SIZE;
1850 }
1851 
1852 EXPORT_SYMBOL(si_meminfo);
1853 
1854 #ifdef CONFIG_NUMA
1855 void si_meminfo_node(struct sysinfo *val, int nid)
1856 {
1857         pg_data_t *pgdat = NODE_DATA(nid);
1858 
1859         val->totalram = pgdat->node_present_pages;
1860         val->freeram = node_page_state(nid, NR_FREE_PAGES);
1861 #ifdef CONFIG_HIGHMEM
1862         val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1863         val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1864                         NR_FREE_PAGES);
1865 #else
1866         val->totalhigh = 0;
1867         val->freehigh = 0;
1868 #endif
1869         val->mem_unit = PAGE_SIZE;
1870 }
1871 #endif
1872 
1873 #define K(x) ((x) << (PAGE_SHIFT-10))
1874 
1875 /*
1876  * Show free area list (used inside shift_scroll-lock stuff)
1877  * We also calculate the percentage fragmentation. We do this by counting the
1878  * memory on each free list with the exception of the first item on the list.
1879  */
1880 void show_free_areas(void)
1881 {
1882         int cpu;
1883         struct zone *zone;
1884 
1885         for_each_zone(zone) {
1886                 if (!populated_zone(zone))
1887                         continue;
1888 
1889                 show_node(zone);
1890                 printk("%s per-cpu:\n", zone->name);
1891 
1892                 for_each_online_cpu(cpu) {
1893                         struct per_cpu_pageset *pageset;
1894 
1895                         pageset = zone_pcp(zone, cpu);
1896 
1897                         printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1898                                cpu, pageset->pcp.high,
1899                                pageset->pcp.batch, pageset->pcp.count);
1900                 }
1901         }
1902 
1903         printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1904                 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1905                 global_page_state(NR_ACTIVE),
1906                 global_page_state(NR_INACTIVE),
1907                 global_page_state(NR_FILE_DIRTY),
1908                 global_page_state(NR_WRITEBACK),
1909                 global_page_state(NR_UNSTABLE_NFS),
1910                 global_page_state(NR_FREE_PAGES),
1911                 global_page_state(NR_SLAB_RECLAIMABLE) +
1912                         global_page_state(NR_SLAB_UNRECLAIMABLE),
1913                 global_page_state(NR_FILE_MAPPED),
1914                 global_page_state(NR_PAGETABLE),
1915                 global_page_state(NR_BOUNCE));
1916 
1917         for_each_zone(zone) {
1918                 int i;
1919 
1920                 if (!populated_zone(zone))
1921                         continue;
1922 
1923                 show_node(zone);
1924                 printk("%s"
1925                         " free:%lukB"
1926                         " min:%lukB"
1927                         " low:%lukB"
1928                         " high:%lukB"
1929                         " active:%lukB"
1930                         " inactive:%lukB"
1931                         " present:%lukB"
1932                         " pages_scanned:%lu"
1933                         " all_unreclaimable? %s"
1934                         "\n",
1935                         zone->name,
1936                         K(zone_page_state(zone, NR_FREE_PAGES)),
1937                         K(zone->pages_min),
1938                         K(zone->pages_low),
1939                         K(zone->pages_high),
1940                         K(zone_page_state(zone, NR_ACTIVE)),
1941                         K(zone_page_state(zone, NR_INACTIVE)),
1942                         K(zone->present_pages),
1943                         zone->pages_scanned,
1944                         (zone_is_all_unreclaimable(zone) ? "yes" : "no")
1945                         );
1946                 printk("lowmem_reserve[]:");
1947                 for (i = 0; i < MAX_NR_ZONES; i++)
1948                         printk(" %lu", zone->lowmem_reserve[i]);
1949                 printk("\n");
1950         }
1951 
1952         for_each_zone(zone) {
1953                 unsigned long nr[MAX_ORDER], flags, order, total = 0;
1954 
1955                 if (!populated_zone(zone))
1956                         continue;
1957 
1958                 show_node(zone);
1959                 printk("%s: ", zone->name);
1960 
1961                 spin_lock_irqsave(&zone->lock, flags);
1962                 for (order = 0; order < MAX_ORDER; order++) {
1963                         nr[order] = zone->free_area[order].nr_free;
1964                         total += nr[order] << order;
1965                 }
1966                 spin_unlock_irqrestore(&zone->lock, flags);
1967                 for (order = 0; order < MAX_ORDER; order++)
1968                         printk("%lu*%lukB ", nr[order], K(1UL) << order);
1969                 printk("= %lukB\n", K(total));
1970         }
1971 
1972         printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
1973 
1974         show_swap_cache_info();
1975 }
1976 
1977 /*
1978  * Builds allocation fallback zone lists.
1979  *
1980  * Add all populated zones of a node to the zonelist.
1981  */
1982 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
1983                                 int nr_zones, enum zone_type zone_type)
1984 {
1985         struct zone *zone;
1986 
1987         BUG_ON(zone_type >= MAX_NR_ZONES);
1988         zone_type++;
1989 
1990         do {
1991                 zone_type--;
1992                 zone = pgdat->node_zones + zone_type;
1993                 if (populated_zone(zone)) {
1994                         zonelist->zones[nr_zones++] = zone;
1995                         check_highest_zone(zone_type);
1996                 }
1997 
1998         } while (zone_type);
1999         return nr_zones;
2000 }
2001 
2002 
2003 /*
2004  *  zonelist_order:
2005  *  0 = automatic detection of better ordering.
2006  *  1 = order by ([node] distance, -zonetype)
2007  *  2 = order by (-zonetype, [node] distance)
2008  *
2009  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2010  *  the same zonelist. So only NUMA can configure this param.
2011  */
2012 #define ZONELIST_ORDER_DEFAULT  0
2013 #define ZONELIST_ORDER_NODE     1
2014 #define ZONELIST_ORDER_ZONE     2
2015 
2016 /* zonelist order in the kernel.
2017  * set_zonelist_order() will set this to NODE or ZONE.
2018  */
2019 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2020 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2021 
2022 
2023 #ifdef CONFIG_NUMA
2024 /* The value user specified ....changed by config */
2025 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2026 /* string for sysctl */
2027 #define NUMA_ZONELIST_ORDER_LEN 16
2028 char numa_zonelist_order[16] = "default";
2029 
2030 /*
2031  * interface for configure zonelist ordering.
2032  * command line option "numa_zonelist_order"
2033  *      = "[dD]efault   - default, automatic configuration.
2034  *      = "[nN]ode      - order by node locality, then by zone within node
2035  *      = "[zZ]one      - order by zone, then by locality within zone
2036  */
2037 
2038 static int __parse_numa_zonelist_order(char *s)
2039 {
2040         if (*s == 'd' || *s == 'D') {
2041                 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2042         } else if (*s == 'n' || *s == 'N') {
2043                 user_zonelist_order = ZONELIST_ORDER_NODE;
2044         } else if (*s == 'z' || *s == 'Z') {
2045                 user_zonelist_order = ZONELIST_ORDER_ZONE;
2046         } else {
2047                 printk(KERN_WARNING
2048                         "Ignoring invalid numa_zonelist_order value:  "
2049                         "%s\n", s);
2050                 return -EINVAL;
2051         }
2052         return 0;
2053 }
2054 
2055 static __init int setup_numa_zonelist_order(char *s)
2056 {
2057         if (s)
2058                 return __parse_numa_zonelist_order(s);
2059         return 0;
2060 }
2061 early_param("numa_zonelist_order", setup_numa_zonelist_order);
2062 
2063 /*
2064  * sysctl handler for numa_zonelist_order
2065  */
2066 int numa_zonelist_order_handler(ctl_table *table, int write,
2067                 struct file *file, void __user *buffer, size_t *length,
2068                 loff_t *ppos)
2069 {
2070         char saved_string[NUMA_ZONELIST_ORDER_LEN];
2071         int ret;
2072 
2073         if (write)
2074                 strncpy(saved_string, (char*)table->data,
2075                         NUMA_ZONELIST_ORDER_LEN);
2076         ret = proc_dostring(table, write, file, buffer, length, ppos);
2077         if (ret)
2078                 return ret;
2079         if (write) {
2080                 int oldval = user_zonelist_order;
2081                 if (__parse_numa_zonelist_order((char*)table->data)) {
2082                         /*
2083                          * bogus value.  restore saved string
2084                          */
2085                         strncpy((char*)table->data, saved_string,
2086                                 NUMA_ZONELIST_ORDER_LEN);
2087                         user_zonelist_order = oldval;
2088                 } else if (oldval != user_zonelist_order)
2089                         build_all_zonelists();
2090         }
2091         return 0;
2092 }
2093 
2094 
2095 #define MAX_NODE_LOAD (num_online_nodes())
2096 static int node_load[MAX_NUMNODES];
2097 
2098 /**
2099  * find_next_best_node - find the next node that should appear in a given node's fallback list
2100  * @node: node whose fallback list we're appending
2101  * @used_node_mask: nodemask_t of already used nodes
2102  *
2103  * We use a number of factors to determine which is the next node that should
2104  * appear on a given node's fallback list.  The node should not have appeared
2105  * already in @node's fallback list, and it should be the next closest node
2106  * according to the distance array (which contains arbitrary distance values
2107  * from each node to each node in the system), and should also prefer nodes
2108  * with no CPUs, since presumably they'll have very little allocation pressure
2109  * on them otherwise.
2110  * It returns -1 if no node is found.
2111  */
2112 static int find_next_best_node(int node, nodemask_t *used_node_mask)
2113 {
2114         int n, val;
2115         int min_val = INT_MAX;
2116         int best_node = -1;
2117 
2118         /* Use the local node if we haven't already */
2119         if (!node_isset(node, *used_node_mask)) {
2120                 node_set(node, *used_node_mask);
2121                 return node;
2122         }
2123 
2124         for_each_node_state(n, N_HIGH_MEMORY) {
2125                 cpumask_t tmp;
2126 
2127                 /* Don't want a node to appear more than once */
2128                 if (node_isset(n, *used_node_mask))
2129                         continue;
2130 
2131                 /* Use the distance array to find the distance */
2132                 val = node_distance(node, n);
2133 
2134                 /* Penalize nodes under us ("prefer the next node") */
2135                 val += (n < node);
2136 
2137                 /* Give preference to headless and unused nodes */
2138                 tmp = node_to_cpumask(n);
2139                 if (!cpus_empty(tmp))
2140                         val += PENALTY_FOR_NODE_WITH_CPUS;
2141 
2142                 /* Slight preference for less loaded node */
2143                 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2144                 val += node_load[n];
2145 
2146                 if (val < min_val) {
2147                         min_val = val;
2148                         best_node = n;
2149                 }
2150         }
2151 
2152         if (best_node >= 0)
2153                 node_set(best_node, *used_node_mask);
2154 
2155         return best_node;
2156 }
2157 
2158 
2159 /*
2160  * Build zonelists ordered by node and zones within node.
2161  * This results in maximum locality--normal zone overflows into local
2162  * DMA zone, if any--but risks exhausting DMA zone.
2163  */
2164 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2165 {
2166         enum zone_type i;
2167         int j;
2168         struct zonelist *zonelist;
2169 
2170         for (i = 0; i < MAX_NR_ZONES; i++) {
2171                 zonelist = pgdat->node_zonelists + i;
2172                 for (j = 0; zonelist->zones[j] != NULL; j++)
2173                         ;
2174                 j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2175                 zonelist->zones[j] = NULL;
2176         }
2177 }
2178 
2179 /*
2180  * Build gfp_thisnode zonelists
2181  */
2182 static void build_thisnode_zonelists(pg_data_t *pgdat)
2183 {
2184         enum zone_type i;
2185         int j;
2186         struct zonelist *zonelist;
2187 
2188         for (i = 0; i < MAX_NR_ZONES; i++) {
2189                 zonelist = pgdat->node_zonelists + MAX_NR_ZONES + i;
2190                 j = build_zonelists_node(pgdat, zonelist, 0, i);
2191                 zonelist->zones[j] = NULL;
2192         }
2193 }
2194 
2195 /*
2196  * Build zonelists ordered by zone and nodes within zones.
2197  * This results in conserving DMA zone[s] until all Normal memory is
2198  * exhausted, but results in overflowing to remote node while memory
2199  * may still exist in local DMA zone.
2200  */
2201 static int node_order[MAX_NUMNODES];
2202 
2203 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2204 {
2205         enum zone_type i;
2206         int pos, j, node;
2207         int zone_type;          /* needs to be signed */
2208         struct zone *z;
2209         struct zonelist *zonelist;
2210 
2211         for (i = 0; i < MAX_NR_ZONES; i++) {
2212                 zonelist = pgdat->node_zonelists + i;
2213                 pos = 0;
2214                 for (zone_type = i; zone_type >= 0; zone_type--) {
2215                         for (j = 0; j < nr_nodes; j++) {
2216                                 node = node_order[j];
2217                                 z = &NODE_DATA(node)->node_zones[zone_type];
2218                                 if (populated_zone(z)) {
2219                                         zonelist->zones[pos++] = z;
2220                                         check_highest_zone(zone_type);
2221                                 }
2222                         }
2223                 }
2224                 zonelist->zones[pos] = NULL;
2225         }
2226 }
2227 
2228 static int default_zonelist_order(void)
2229 {
2230         int nid, zone_type;
2231         unsigned long low_kmem_size,total_size;
2232         struct zone *z;
2233         int average_size;
2234         /*
2235          * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2236          * If they are really small and used heavily, the system can fall
2237          * into OOM very easily.
2238          * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2239          */
2240         /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2241         low_kmem_size = 0;
2242         total_size = 0;
2243         for_each_online_node(nid) {
2244                 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2245                         z = &NODE_DATA(nid)->node_zones[zone_type];
2246                         if (populated_zone(z)) {
2247                                 if (zone_type < ZONE_NORMAL)
2248                                         low_kmem_size += z->present_pages;
2249                                 total_size += z->present_pages;
2250                         }
2251                 }
2252         }
2253         if (!low_kmem_size ||  /* there are no DMA area. */
2254             low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2255                 return ZONELIST_ORDER_NODE;
2256         /*
2257          * look into each node's config.
2258          * If there is a node whose DMA/DMA32 memory is very big area on
2259          * local memory, NODE_ORDER may be suitable.
2260          */
2261         average_size = total_size /
2262                                 (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2263         for_each_online_node(nid) {
2264                 low_kmem_size = 0;
2265                 total_size = 0;
2266                 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2267                         z = &NODE_DATA(nid)->node_zones[zone_type];
2268                         if (populated_zone(z)) {
2269                                 if (zone_type < ZONE_NORMAL)
2270                                         low_kmem_size += z->present_pages;
2271                                 total_size += z->present_pages;
2272                         }
2273                 }
2274                 if (low_kmem_size &&
2275                     total_size > average_size && /* ignore small node */
2276                     low_kmem_size > total_size * 70/100)
2277                         return ZONELIST_ORDER_NODE;
2278         }
2279         return ZONELIST_ORDER_ZONE;
2280 }
2281 
2282 static void set_zonelist_order(void)
2283 {
2284         if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2285                 current_zonelist_order = default_zonelist_order();
2286         else
2287                 current_zonelist_order = user_zonelist_order;
2288 }
2289 
2290 static void build_zonelists(pg_data_t *pgdat)
2291 {
2292         int j, node, load;
2293         enum zone_type i;
2294         nodemask_t used_mask;
2295         int local_node, prev_node;
2296         struct zonelist *zonelist;
2297         int order = current_zonelist_order;
2298 
2299         /* initialize zonelists */
2300         for (i = 0; i < MAX_ZONELISTS; i++) {
2301                 zonelist = pgdat->node_zonelists + i;
2302                 zonelist->zones[0] = NULL;
2303         }
2304 
2305         /* NUMA-aware ordering of nodes */
2306         local_node = pgdat->node_id;
2307         load = num_online_nodes();
2308         prev_node = local_node;
2309         nodes_clear(used_mask);
2310 
2311         memset(node_load, 0, sizeof(node_load));
2312         memset(node_order, 0, sizeof(node_order));
2313         j = 0;
2314 
2315         while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2316                 int distance = node_distance(local_node, node);
2317 
2318                 /*
2319                  * If another node is sufficiently far away then it is better
2320                  * to reclaim pages in a zone before going off node.
2321                  */
2322                 if (distance > RECLAIM_DISTANCE)
2323                         zone_reclaim_mode = 1;
2324 
2325                 /*
2326                  * We don't want to pressure a particular node.
2327                  * So adding penalty to the first node in same
2328                  * distance group to make it round-robin.
2329                  */
2330                 if (distance != node_distance(local_node, prev_node))
2331                         node_load[node] = load;
2332 
2333                 prev_node = node;
2334                 load--;
2335                 if (order == ZONELIST_ORDER_NODE)
2336                         build_zonelists_in_node_order(pgdat, node);
2337                 else
2338                         node_order[j++] = node; /* remember order */
2339         }
2340 
2341         if (order == ZONELIST_ORDER_ZONE) {
2342                 /* calculate node order -- i.e., DMA last! */
2343                 build_zonelists_in_zone_order(pgdat, j);
2344         }
2345 
2346         build_thisnode_zonelists(pgdat);
2347 }
2348 
2349 /* Construct the zonelist performance cache - see further mmzone.h */
2350 static void build_zonelist_cache(pg_data_t *pgdat)
2351 {
2352         int i;
2353 
2354         for (i = 0; i < MAX_NR_ZONES; i++) {
2355                 struct zonelist *zonelist;
2356                 struct zonelist_cache *zlc;
2357                 struct zone **z;
2358 
2359                 zonelist = pgdat->node_zonelists + i;
2360                 zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2361                 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2362                 for (z = zonelist->zones; *z; z++)
2363                         zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z);
2364         }
2365 }
2366 
2367 
2368 #else   /* CONFIG_NUMA */
2369 
2370 static void set_zonelist_order(void)
2371 {
2372         current_zonelist_order = ZONELIST_ORDER_ZONE;
2373 }
2374 
2375 static void build_zonelists(pg_data_t *pgdat)
2376 {
2377         int node, local_node;
2378         enum zone_type i,j;
2379 
2380         local_node = pgdat->node_id;
2381         for (i = 0; i < MAX_NR_ZONES; i++) {
2382                 struct zonelist *zonelist;
2383 
2384                 zonelist = pgdat->node_zonelists + i;
2385 
2386                 j = build_zonelists_node(pgdat, zonelist, 0, i);
2387                 /*
2388                  * Now we build the zonelist so that it contains the zones
2389                  * of all the other nodes.
2390                  * We don't want to pressure a particular node, so when
2391                  * building the zones for node N, we make sure that the
2392                  * zones coming right after the local ones are those from
2393                  * node N+1 (modulo N)
2394                  */
2395                 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2396                         if (!node_online(node))
2397                                 continue;
2398                         j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2399                 }
2400                 for (node = 0; node < local_node; node++) {
2401                         if (!node_online(node))
2402                                 continue;
2403                         j = build_zonelists_node(NODE_DATA(node), zonelist, j, i);
2404                 }
2405 
2406                 zonelist->zones[j] = NULL;
2407         }
2408 }
2409 
2410 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2411 static void build_zonelist_cache(pg_data_t *pgdat)
2412 {
2413         int i;
2414 
2415         for (i = 0; i < MAX_NR_ZONES; i++)
2416                 pgdat->node_zonelists[i].zlcache_ptr = NULL;
2417 }
2418 
2419 #endif  /* CONFIG_NUMA */
2420 
2421 /* return values int ....just for stop_machine_run() */
2422 static int __build_all_zonelists(void *dummy)
2423 {
2424         int nid;
2425 
2426         for_each_online_node(nid) {
2427                 pg_data_t *pgdat = NODE_DATA(nid);
2428 
2429                 build_zonelists(pgdat);
2430                 build_zonelist_cache(pgdat);
2431         }
2432         return 0;
2433 }
2434 
2435 void build_all_zonelists(void)
2436 {
2437         set_zonelist_order();
2438 
2439         if (system_state == SYSTEM_BOOTING) {
2440                 __build_all_zonelists(NULL);
2441                 cpuset_init_current_mems_allowed();
2442         } else {
2443                 /* we have to stop all cpus to guarantee there is no user
2444                    of zonelist */
2445                 stop_machine_run(__build_all_zonelists, NULL, NR_CPUS);
2446                 /* cpuset refresh routine should be here */
2447         }
2448         vm_total_pages = nr_free_pagecache_pages();
2449         /*
2450          * Disable grouping by mobility if the number of pages in the
2451          * system is too low to allow the mechanism to work. It would be
2452          * more accurate, but expensive to check per-zone. This check is
2453          * made on memory-hotadd so a system can start with mobility
2454          * disabled and enable it later
2455          */
2456         if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2457                 page_group_by_mobility_disabled = 1;
2458         else
2459                 page_group_by_mobility_disabled = 0;
2460 
2461         printk("Built %i zonelists in %s order, mobility grouping %s.  "
2462                 "Total pages: %ld\n",
2463                         num_online_nodes(),
2464                         zonelist_order_name[current_zonelist_order],
2465                         page_group_by_mobility_disabled ? "off" : "on",
2466                         vm_total_pages);
2467 #ifdef CONFIG_NUMA
2468         printk("Policy zone: %s\n", zone_names[policy_zone]);
2469 #endif
2470 }
2471 
2472 /*
2473  * Helper functions to size the waitqueue hash table.
2474  * Essentially these want to choose hash table sizes sufficiently
2475  * large so that collisions trying to wait on pages are rare.
2476  * But in fact, the number of active page waitqueues on typical
2477  * systems is ridiculously low, less than 200. So this is even
2478  * conservative, even though it seems large.
2479  *
2480  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2481  * waitqueues, i.e. the size of the waitq table given the number of pages.
2482  */
2483 #define PAGES_PER_WAITQUEUE     256
2484 
2485 #ifndef CONFIG_MEMORY_HOTPLUG
2486 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2487 {
2488         unsigned long size = 1;
2489 
2490         pages /= PAGES_PER_WAITQUEUE;
2491 
2492         while (size < pages)
2493                 size <<= 1;
2494 
2495         /*
2496          * Once we have dozens or even hundreds of threads sleeping
2497          * on IO we've got bigger problems than wait queue collision.
2498          * Limit the size of the wait table to a reasonable size.
2499          */
2500         size = min(size, 4096UL);
2501 
2502         return max(size, 4UL);
2503 }
2504 #else
2505 /*
2506  * A zone's size might be changed by hot-add, so it is not possible to determine
2507  * a suitable size for its wait_table.  So we use the maximum size now.
2508  *
2509  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
2510  *
2511  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
2512  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2513  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
2514  *
2515  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2516  * or more by the traditional way. (See above).  It equals:
2517  *
2518  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
2519  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
2520  *    powerpc (64K page size)             : =  (32G +16M)byte.
2521  */
2522 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2523 {
2524         return 4096UL;
2525 }
2526 #endif
2527 
2528 /*
2529  * This is an integer logarithm so that shifts can be used later
2530  * to extract the more random high bits from the multiplicative
2531  * hash function before the remainder is taken.
2532  */
2533 static inline unsigned long wait_table_bits(unsigned long size)
2534 {
2535         return ffz(~size);
2536 }
2537 
2538 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2539 
2540 /*
2541  * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2542  * of blocks reserved is based on zone->pages_min. The memory within the
2543  * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2544  * higher will lead to a bigger reserve which will get freed as contiguous
2545  * blocks as reclaim kicks in
2546  */
2547 static void setup_zone_migrate_reserve(struct zone *zone)
2548 {
2549         unsigned long start_pfn, pfn, end_pfn;
2550         struct page *page;
2551         unsigned long reserve, block_migratetype;
2552 
2553         /* Get the start pfn, end pfn and the number of blocks to reserve */
2554         start_pfn = zone->zone_start_pfn;
2555         end_pfn = start_pfn + zone->spanned_pages;
2556         reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2557                                                         pageblock_order;
2558 
2559         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2560                 if (!pfn_valid(pfn))
2561                         continue;
2562                 page = pfn_to_page(pfn);
2563 
2564                 /* Blocks with reserved pages will never free, skip them. */
2565                 if (PageReserved(page))
2566                         continue;
2567 
2568                 block_migratetype = get_pageblock_migratetype(page);
2569 
2570                 /* If this block is reserved, account for it */
2571                 if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2572                         reserve--;
2573                         continue;
2574                 }
2575 
2576                 /* Suitable for reserving if this block is movable */
2577                 if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2578                         set_pageblock_migratetype(page, MIGRATE_RESERVE);
2579                         move_freepages_block(zone, page, MIGRATE_RESERVE);
2580                         reserve--;
2581                         continue;
2582                 }
2583 
2584                 /*
2585                  * If the reserve is met and this is a previous reserved block,
2586                  * take it back
2587                  */
2588                 if (block_migratetype == MIGRATE_RESERVE) {
2589                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2590                         move_freepages_block(zone, page, MIGRATE_MOVABLE);
2591                 }
2592         }
2593 }
2594 
2595 /*
2596  * Initially all pages are reserved - free ones are freed
2597  * up by free_all_bootmem() once the early boot process is
2598  * done. Non-atomic initialization, single-pass.
2599  */
2600 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2601                 unsigned long start_pfn, enum memmap_context context)
2602 {
2603         struct page *page;
2604         unsigned long end_pfn = start_pfn + size;
2605         unsigned long pfn;
2606         struct zone *z;
2607 
2608         z = &NODE_DATA(nid)->node_zones[zone];
2609         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2610                 /*
2611                  * There can be holes in boot-time mem_map[]s
2612                  * handed to this function.  They do not
2613                  * exist on hotplugged memory.
2614                  */
2615                 if (context == MEMMAP_EARLY) {
2616                         if (!early_pfn_valid(pfn))
2617                                 continue;
2618                         if (!early_pfn_in_nid(pfn, nid))
2619                                 continue;
2620                 }
2621                 page = pfn_to_page(pfn);
2622                 set_page_links(page, zone, nid, pfn);
2623                 init_page_count(page);
2624                 reset_page_mapcount(page);
2625                 SetPageReserved(page);
2626                 /*
2627                  * Mark the block movable so that blocks are reserved for
2628                  * movable at startup. This will force kernel allocations
2629                  * to reserve their blocks rather than leaking throughout
2630                  * the address space during boot when many long-lived
2631                  * kernel allocations are made. Later some blocks near
2632                  * the start are marked MIGRATE_RESERVE by
2633                  * setup_zone_migrate_reserve()
2634                  *
2635                  * bitmap is created for zone's valid pfn range. but memmap
2636                  * can be created for invalid pages (for alignment)
2637                  * check here not to call set_pageblock_migratetype() against
2638                  * pfn out of zone.
2639                  */
2640                 if ((z->zone_start_pfn <= pfn)
2641                     && (pfn < z->zone_start_pfn + z->spanned_pages)
2642                     && !(pfn & (pageblock_nr_pages - 1)))
2643                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2644 
2645                 INIT_LIST_HEAD(&page->lru);
2646 #ifdef WANT_PAGE_VIRTUAL
2647                 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2648                 if (!is_highmem_idx(zone))
2649                         set_page_address(page, __va(pfn << PAGE_SHIFT));
2650 #endif
2651         }
2652 }
2653 
2654 static void __meminit zone_init_free_lists(struct zone *zone)
2655 {
2656         int order, t;
2657         for_each_migratetype_order(order, t) {
2658                 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2659                 zone->free_area[order].nr_free = 0;
2660         }
2661 }
2662 
2663 #ifndef __HAVE_ARCH_MEMMAP_INIT
2664 #define memmap_init(size, nid, zone, start_pfn) \
2665         memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2666 #endif
2667 
2668 static int zone_batchsize(struct zone *zone)
2669 {
2670         int batch;
2671 
2672         /*
2673          * The per-cpu-pages pools are set to around 1000th of the
2674          * size of the zone.  But no more than 1/2 of a meg.
2675          *
2676          * OK, so we don't know how big the cache is.  So guess.
2677          */
2678         batch = zone->present_pages / 1024;
2679         if (batch * PAGE_SIZE > 512 * 1024)
2680                 batch = (512 * 1024) / PAGE_SIZE;
2681         batch /= 4;             /* We effectively *= 4 below */
2682         if (batch < 1)
2683                 batch = 1;
2684 
2685         /*
2686          * Clamp the batch to a 2^n - 1 value. Having a power
2687          * of 2 value was found to be more likely to have
2688          * suboptimal cache aliasing properties in some cases.
2689          *
2690          * For example if 2 tasks are alternately allocating
2691          * batches of pages, one task can end up with a lot
2692          * of pages of one half of the possible page colors
2693          * and the other with pages of the other colors.
2694          */
2695         batch = (1 << (fls(batch + batch/2)-1)) - 1;
2696 
2697         return batch;
2698 }
2699 
2700 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2701 {
2702         struct per_cpu_pages *pcp;
2703 
2704         memset(p, 0, sizeof(*p));
2705 
2706         pcp = &p->pcp;
2707         pcp->count = 0;
2708         pcp->high = 6 * batch;
2709         pcp->batch = max(1UL, 1 * batch);
2710         INIT_LIST_HEAD(&pcp->list);
2711 }
2712 
2713 /*
2714  * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2715  * to the value high for the pageset p.
2716  */
2717 
2718 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2719                                 unsigned long high)
2720 {
2721         struct per_cpu_pages *pcp;
2722 
2723         pcp = &p->pcp;
2724         pcp->high = high;
2725         pcp->batch = max(1UL, high/4);
2726         if ((high/4) > (PAGE_SHIFT * 8))
2727                 pcp->batch = PAGE_SHIFT * 8;
2728 }
2729 
2730 
2731 #ifdef CONFIG_NUMA
2732 /*
2733  * Boot pageset table. One per cpu which is going to be used for all
2734  * zones and all nodes. The parameters will be set in such a way
2735  * that an item put on a list will immediately be handed over to
2736  * the buddy list. This is safe since pageset manipulation is done
2737  * with interrupts disabled.
2738  *
2739  * Some NUMA counter updates may also be caught by the boot pagesets.
2740  *
2741  * The boot_pagesets must be kept even after bootup is complete for
2742  * unused processors and/or zones. They do play a role for bootstrapping
2743  * hotplugged processors.
2744  *
2745  * zoneinfo_show() and maybe other functions do
2746  * not check if the processor is online before following the pageset pointer.
2747  * Other parts of the kernel may not check if the zone is available.
2748  */
2749 static struct per_cpu_pageset boot_pageset[NR_CPUS];
2750 
2751 /*
2752  * Dynamically allocate memory for the
2753  * per cpu pageset array in struct zone.
2754  */
2755 static int __cpuinit process_zones(int cpu)
2756 {
2757         struct zone *zone, *dzone;
2758         int node = cpu_to_node(cpu);
2759 
2760         node_set_state(node, N_CPU);    /* this node has a cpu */
2761 
2762         for_each_zone(zone) {
2763 
2764                 if (!populated_zone(zone))
2765                         continue;
2766 
2767                 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2768                                          GFP_KERNEL, node);
2769                 if (!zone_pcp(zone, cpu))
2770                         goto bad;
2771 
2772                 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2773 
2774                 if (percpu_pagelist_fraction)
2775                         setup_pagelist_highmark(zone_pcp(zone, cpu),
2776                                 (zone->present_pages / percpu_pagelist_fraction));
2777         }
2778 
2779         return 0;
2780 bad:
2781         for_each_zone(dzone) {
2782                 if (!populated_zone(dzone))
2783                         continue;
2784                 if (dzone == zone)
2785                         break;
2786                 kfree(zone_pcp(dzone, cpu));
2787                 zone_pcp(dzone, cpu) = NULL;
2788         }
2789         return -ENOMEM;
2790 }
2791 
2792 static inline void free_zone_pagesets(int cpu)
2793 {
2794         struct zone *zone;
2795 
2796         for_each_zone(zone) {
2797                 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2798 
2799                 /* Free per_cpu_pageset if it is slab allocated */
2800                 if (pset != &boot_pageset[cpu])
2801                         kfree(pset);
2802                 zone_pcp(zone, cpu) = NULL;
2803         }
2804 }
2805 
2806 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2807                 unsigned long action,
2808                 void *hcpu)
2809 {
2810         int cpu = (long)hcpu;
2811         int ret = NOTIFY_OK;
2812 
2813         switch (action) {
2814         case CPU_UP_PREPARE:
2815         case CPU_UP_PREPARE_FROZEN:
2816                 if (process_zones(cpu))
2817                         ret = NOTIFY_BAD;
2818                 break;
2819         case CPU_UP_CANCELED:
2820         case CPU_UP_CANCELED_FROZEN:
2821         case CPU_DEAD:
2822         case CPU_DEAD_FROZEN:
2823                 free_zone_pagesets(cpu);
2824                 break;
2825         default:
2826                 break;
2827         }
2828         return ret;
2829 }
2830 
2831 static struct notifier_block __cpuinitdata pageset_notifier =
2832         { &pageset_cpuup_callback, NULL, 0 };
2833 
2834 void __init setup_per_cpu_pageset(void)
2835 {
2836         int err;
2837 
2838         /* Initialize per_cpu_pageset for cpu 0.
2839          * A cpuup callback will do this for every cpu
2840          * as it comes online
2841          */
2842         err = process_zones(smp_processor_id());
2843         BUG_ON(err);
2844         register_cpu_notifier(&pageset_notifier);
2845 }
2846 
2847 #endif
2848 
2849 static noinline __init_refok
2850 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2851 {
2852         int i;
2853         struct pglist_data *pgdat = zone->zone_pgdat;
2854         size_t alloc_size;
2855 
2856         /*
2857          * The per-page waitqueue mechanism uses hashed waitqueues
2858          * per zone.
2859          */
2860         zone->wait_table_hash_nr_entries =
2861                  wait_table_hash_nr_entries(zone_size_pages);
2862         zone->wait_table_bits =
2863                 wait_table_bits(zone->wait_table_hash_nr_entries);
2864         alloc_size = zone->wait_table_hash_nr_entries
2865                                         * sizeof(wait_queue_head_t);
2866 
2867         if (system_state == SYSTEM_BOOTING) {
2868                 zone->wait_table = (wait_queue_head_t *)
2869                         alloc_bootmem_node(pgdat, alloc_size);
2870         } else {
2871                 /*
2872                  * This case means that a zone whose size was 0 gets new memory
2873                  * via memory hot-add.
2874                  * But it may be the case that a new node was hot-added.  In
2875                  * this case vmalloc() will not be able to use this new node's
2876                  * memory - this wait_table must be initialized to use this new
2877                  * node itself as well.
2878                  * To use this new node's memory, further consideration will be
2879                  * necessary.
2880                  */
2881                 zone->wait_table = vmalloc(alloc_size);
2882         }
2883         if (!zone->wait_table)
2884                 return -ENOMEM;
2885 
2886         for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2887                 init_waitqueue_head(zone->wait_table + i);
2888 
2889         return 0;
2890 }
2891 
2892 static __meminit void zone_pcp_init(struct zone *zone)
2893 {
2894         int cpu;
2895         unsigned long batch = zone_batchsize(zone);
2896 
2897         for (cpu = 0; cpu < NR_CPUS; cpu++) {
2898 #ifdef CONFIG_NUMA
2899                 /* Early boot. Slab allocator not functional yet */
2900                 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2901                 setup_pageset(&boot_pageset[cpu],0);
2902 #else
2903                 setup_pageset(zone_pcp(zone,cpu), batch);
2904 #endif
2905         }
2906         if (zone->present_pages)
2907                 printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%lu\n",
2908                         zone->name, zone->present_pages, batch);
2909 }
2910 
2911 __meminit int init_currently_empty_zone(struct zone *zone,
2912                                         unsigned long zone_start_pfn,
2913                                         unsigned long size,
2914                                         enum memmap_context context)
2915 {
2916         struct pglist_data *pgdat = zone->zone_pgdat;
2917         int ret;
2918         ret = zone_wait_table_init(zone, size);
2919         if (ret)
2920                 return ret;
2921         pgdat->nr_zones = zone_idx(zone) + 1;
2922 
2923         zone->zone_start_pfn = zone_start_pfn;
2924 
2925         zone_init_free_lists(zone);
2926 
2927         return 0;
2928 }
2929 
2930 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2931 /*
2932  * Basic iterator support. Return the first range of PFNs for a node
2933  * Note: nid == MAX_NUMNODES returns first region regardless of node
2934  */
2935 static int __meminit first_active_region_index_in_nid(int nid)
2936 {
2937         int i;
2938 
2939         for (i = 0; i < nr_nodemap_entries; i++)
2940                 if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2941                         return i;
2942 
2943         return -1;
2944 }
2945 
2946 /*
2947  * Basic iterator support. Return the next active range of PFNs for a node
2948  * Note: nid == MAX_NUMNODES returns next region regardless of node
2949  */
2950 static int __meminit next_active_region_index_in_nid(int index, int nid)
2951 {
2952         for (index = index + 1; index < nr_nodemap_entries; index++)
2953                 if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2954                         return index;
2955 
2956         return -1;
2957 }
2958 
2959 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2960 /*
2961  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2962  * Architectures may implement their own version but if add_active_range()
2963  * was used and there are no special requirements, this is a convenient
2964  * alternative
2965  */
2966 int __meminit early_pfn_to_nid(unsigned long pfn)
2967 {
2968         int i;
2969 
2970         for (i = 0; i < nr_nodemap_entries; i++) {
2971                 unsigned long start_pfn = early_node_map[i].start_pfn;
2972                 unsigned long end_pfn = early_node_map[i].end_pfn;
2973 
2974                 if (start_pfn <= pfn && pfn < end_pfn)
2975                         return early_node_map[i].nid;
2976         }
2977 
2978         return 0;
2979 }
2980 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2981 
2982 /* Basic iterator support to walk early_node_map[] */
2983 #define for_each_active_range_index_in_nid(i, nid) \
2984         for (i = first_active_region_index_in_nid(nid); i != -1; \
2985                                 i = next_active_region_index_in_nid(i, nid))
2986 
2987 /**
2988  * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2989  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2990  * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2991  *
2992  * If an architecture guarantees that all ranges registered with
2993  * add_active_ranges() contain no holes and may be freed, this
2994  * this function may be used instead of calling free_bootmem() manually.
2995  */
2996 void __init free_bootmem_with_active_regions(int nid,
2997                                                 unsigned long max_low_pfn)
2998 {
2999         int i;
3000 
3001         for_each_active_range_index_in_nid(i, nid) {
3002                 unsigned long size_pages = 0;
3003                 unsigned long end_pfn = early_node_map[i].end_pfn;
3004 
3005                 if (early_node_map[i].start_pfn >= max_low_pfn)
3006                         continue;
3007 
3008                 if (end_pfn > max_low_pfn)
3009                         end_pfn = max_low_pfn;
3010 
3011                 size_pages = end_pfn - early_node_map[i].start_pfn;
3012                 free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3013                                 PFN_PHYS(early_node_map[i].start_pfn),
3014                                 size_pages << PAGE_SHIFT);
3015         }
3016 }
3017 
3018 /**
3019  * sparse_memory_present_with_active_regions - Call memory_present for each active range
3020  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3021  *
3022  * If an architecture guarantees that all ranges registered with
3023  * add_active_ranges() contain no holes and may be freed, this
3024  * function may be used instead of calling memory_present() manually.
3025  */
3026 void __init sparse_memory_present_with_active_regions(int nid)
3027 {
3028         int i;
3029 
3030         for_each_active_range_index_in_nid(i, nid)
3031                 memory_present(early_node_map[i].nid,
3032                                 early_node_map[i].start_pfn,
3033                                 early_node_map[i].end_pfn);
3034 }
3035 
3036 /**
3037  * push_node_boundaries - Push node boundaries to at least the requested boundary
3038  * @nid: The nid of the node to push the boundary for
3039  * @start_pfn: The start pfn of the node
3040  * @end_pfn: The end pfn of the node
3041  *
3042  * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3043  * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3044  * be hotplugged even though no physical memory exists. This function allows
3045  * an arch to push out the node boundaries so mem_map is allocated that can
3046  * be used later.
3047  */
3048 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3049 void __init push_node_boundaries(unsigned int nid,
3050                 unsigned long start_pfn, unsigned long end_pfn)
3051 {
3052         printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n",
3053                         nid, start_pfn, end_pfn);
3054 
3055         /* Initialise the boundary for this node if necessary */
3056         if (node_boundary_end_pfn[nid] == 0)
3057                 node_boundary_start_pfn[nid] = -1UL;
3058 
3059         /* Update the boundaries */
3060         if (node_boundary_start_pfn[nid] > start_pfn)
3061                 node_boundary_start_pfn[nid] = start_pfn;
3062         if (node_boundary_end_pfn[nid] < end_pfn)
3063                 node_boundary_end_pfn[nid] = end_pfn;
3064 }
3065 
3066 /* If necessary, push the node boundary out for reserve hotadd */
3067 static void __meminit account_node_boundary(unsigned int nid,
3068                 unsigned long *start_pfn, unsigned long *end_pfn)
3069 {
3070         printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n",
3071                         nid, *start_pfn, *end_pfn);
3072 
3073         /* Return if boundary information has not been provided */
3074         if (node_boundary_end_pfn[nid] == 0)
3075                 return;
3076 
3077         /* Check the boundaries and update if necessary */
3078         if (node_boundary_start_pfn[nid] < *start_pfn)
3079                 *start_pfn = node_boundary_start_pfn[nid];
3080         if (node_boundary_end_pfn[nid] > *end_pfn)
3081                 *end_pfn = node_boundary_end_pfn[nid];
3082 }
3083 #else
3084 void __init push_node_boundaries(unsigned int nid,
3085                 unsigned long start_pfn, unsigned long end_pfn) {}
3086 
3087 static void __meminit account_node_boundary(unsigned int nid,
3088                 unsigned long *start_pfn, unsigned long *end_pfn) {}
3089 #endif
3090 
3091 
3092 /**
3093  * get_pfn_range_for_nid - Return the start and end page frames for a node
3094  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3095  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3096  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3097  *
3098  * It returns the start and end page frame of a node based on information
3099  * provided by an arch calling add_active_range(). If called for a node
3100  * with no available memory, a warning is printed and the start and end
3101  * PFNs will be 0.
3102  */
3103 void __meminit get_pfn_range_for_nid(unsigned int nid,
3104                         unsigned long *start_pfn, unsigned long *end_pfn)
3105 {
3106         int i;
3107         *start_pfn = -1UL;
3108         *end_pfn = 0;
3109 
3110         for_each_active_range_index_in_nid(i, nid) {
3111                 *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3112                 *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3113         }
3114 
3115         if (*start_pfn == -1UL)
3116                 *start_pfn = 0;
3117 
3118         /* Push the node boundaries out if requested */
3119         account_node_boundary(nid, start_pfn, end_pfn);
3120 }
3121 
3122 /*
3123  * This finds a zone that can be used for ZONE_MOVABLE pages. The
3124  * assumption is made that zones within a node are ordered in monotonic
3125  * increasing memory addresses so that the "highest" populated zone is used
3126  */
3127 void __init find_usable_zone_for_movable(void)
3128 {
3129         int zone_index;
3130         for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3131                 if (zone_index == ZONE_MOVABLE)
3132                         continue;
3133 
3134                 if (arch_zone_highest_possible_pfn[zone_index] >
3135                                 arch_zone_lowest_possible_pfn[zone_index])
3136                         break;
3137         }
3138 
3139         VM_BUG_ON(zone_index == -1);
3140         movable_zone = zone_index;
3141 }
3142 
3143 /*
3144  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3145  * because it is sized independant of architecture. Unlike the other zones,
3146  * the starting point for ZONE_MOVABLE is not fixed. It may be different
3147  * in each node depending on the size of each node and how evenly kernelcore
3148  * is distributed. This helper function adjusts the zone ranges
3149  * provided by the architecture for a given node by using the end of the
3150  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3151  * zones within a node are in order of monotonic increases memory addresses
3152  */
3153 void __meminit adjust_zone_range_for_zone_movable(int nid,
3154                                         unsigned long zone_type,
3155                                         unsigned long node_start_pfn,
3156                                         unsigned long node_end_pfn,
3157                                         unsigned long *zone_start_pfn,
3158                                         unsigned long *zone_end_pfn)
3159 {
3160         /* Only adjust if ZONE_MOVABLE is on this node */
3161         if (zone_movable_pfn[nid]) {
3162                 /* Size ZONE_MOVABLE */
3163                 if (zone_type == ZONE_MOVABLE) {
3164                         *zone_start_pfn = zone_movable_pfn[nid];
3165                         *zone_end_pfn = min(node_end_pfn,
3166                                 arch_zone_highest_possible_pfn[movable_zone]);
3167 
3168                 /* Adjust for ZONE_MOVABLE starting within this range */
3169                 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3170                                 *zone_end_pfn > zone_movable_pfn[nid]) {
3171                         *zone_end_pfn = zone_movable_pfn[nid];
3172 
3173                 /* Check if this whole range is within ZONE_MOVABLE */
3174                 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3175                         *zone_start_pfn = *zone_end_pfn;
3176         }
3177 }
3178 
3179 /*
3180  * Return the number of pages a zone spans in a node, including holes
3181  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3182  */
3183 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3184                                         unsigned long zone_type,
3185                                         unsigned long *ignored)
3186 {
3187         unsigned long node_start_pfn, node_end_pfn;
3188         unsigned long zone_start_pfn, zone_end_pfn;
3189 
3190         /* Get the start and end of the node and zone */
3191         get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3192         zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3193         zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3194         adjust_zone_range_for_zone_movable(nid, zone_type,
3195                                 node_start_pfn, node_end_pfn,
3196                                 &zone_start_pfn, &zone_end_pfn);
3197 
3198         /* Check that this node has pages within the zone's required range */
3199         if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3200                 return 0;
3201 
3202         /* Move the zone boundaries inside the node if necessary */
3203         zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3204         zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3205 
3206         /* Return the spanned pages */
3207         return zone_end_pfn - zone_start_pfn;
3208 }
3209 
3210 /*
3211  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3212  * then all holes in the requested range will be accounted for.
3213  */
3214 unsigned long __meminit __absent_pages_in_range(int nid,
3215                                 unsigned long range_start_pfn,
3216                                 unsigned long range_end_pfn)
3217 {
3218         int i = 0;
3219         unsigned long prev_end_pfn = 0, hole_pages = 0;
3220         unsigned long start_pfn;
3221 
3222         /* Find the end_pfn of the first active range of pfns in the node */
3223         i = first_active_region_index_in_nid(nid);
3224         if (i == -1)
3225                 return 0;
3226 
3227         prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3228 
3229         /* Account for ranges before physical memory on this node */
3230         if (early_node_map[i].start_pfn > range_start_pfn)
3231                 hole_pages = prev_end_pfn - range_start_pfn;
3232 
3233         /* Find all holes for the zone within the node */
3234         for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3235 
3236                 /* No need to continue if prev_end_pfn is outside the zone */
3237                 if (prev_end_pfn >= range_end_pfn)
3238                         break;
3239 
3240                 /* Make sure the end of the zone is not within the hole */
3241                 start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3242                 prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3243 
3244                 /* Update the hole size cound and move on */
3245                 if (start_pfn > range_start_pfn) {
3246                         BUG_ON(prev_end_pfn > start_pfn);
3247                         hole_pages += start_pfn - prev_end_pfn;
3248                 }
3249                 prev_end_pfn = early_node_map[i].end_pfn;
3250         }
3251 
3252         /* Account for ranges past physical memory on this node */
3253         if (range_end_pfn > prev_end_pfn)
3254                 hole_pages += range_end_pfn -
3255                                 max(range_start_pfn, prev_end_pfn);
3256 
3257         return hole_pages;
3258 }
3259 
3260 /**
3261  * absent_pages_in_range - Return number of page frames in holes within a range
3262  * @start_pfn: The start PFN to start searching for holes
3263  * @end_pfn: The end PFN to stop searching for holes
3264  *
3265  * It returns the number of pages frames in memory holes within a range.
3266  */
3267 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3268                                                         unsigned long end_pfn)
3269 {
3270         return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3271 }
3272 
3273 /* Return the number of page frames in holes in a zone on a node */
3274 static unsigned long __meminit zone_absent_pages_in_node(int nid,
3275                                         unsigned long zone_type,
3276                                         unsigned long *ignored)
3277 {
3278         unsigned long node_start_pfn, node_end_pfn;
3279         unsigned long zone_start_pfn, zone_end_pfn;
3280 
3281         get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3282         zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3283                                                         node_start_pfn);
3284         zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3285                                                         node_end_pfn);
3286 
3287         adjust_zone_range_for_zone_movable(nid, zone_type,
3288                         node_start_pfn, node_end_pfn,
3289                         &zone_start_pfn, &zone_end_pfn);
3290         return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3291 }
3292 
3293 #else
3294 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3295                                         unsigned long zone_type,
3296                                         unsigned long *zones_size)
3297 {
3298         return zones_size[zone_type];
3299 }
3300 
3301 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3302                                                 unsigned long zone_type,
3303                                                 unsigned long *zholes_size)
3304 {
3305         if (!zholes_size)
3306                 return 0;
3307 
3308         return zholes_size[zone_type];
3309 }
3310 
3311 #endif
3312 
3313 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3314                 unsigned long *zones_size, unsigned long *zholes_size)
3315 {
3316         unsigned long realtotalpages, totalpages = 0;
3317         enum zone_type i;
3318 
3319         for (i = 0; i < MAX_NR_ZONES; i++)
3320                 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3321                                                                 zones_size);
3322         pgdat->node_spanned_pages = totalpages;
3323 
3324         realtotalpages = totalpages;
3325         for (i = 0; i < MAX_NR_ZONES; i++)
3326                 realtotalpages -=
3327                         zone_absent_pages_in_node(pgdat->node_id, i,
3328                                                                 zholes_size);
3329         pgdat->node_present_pages = realtotalpages;
3330         printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3331                                                         realtotalpages);
3332 }
3333 
3334 #ifndef CONFIG_SPARSEMEM
3335 /*
3336  * Calculate the size of the zone->blockflags rounded to an unsigned long
3337  * Start by making sure zonesize is a multiple of pageblock_order by rounding
3338  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3339  * round what is now in bits to nearest long in bits, then return it in
3340  * bytes.
3341  */
3342 static unsigned long __init usemap_size(unsigned long zonesize)
3343 {
3344         unsigned long usemapsize;
3345 
3346         usemapsize = roundup(zonesize, pageblock_nr_pages);
3347         usemapsize = usemapsize >> pageblock_order;
3348         usemapsize *= NR_PAGEBLOCK_BITS;
3349         usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3350 
3351         return usemapsize / 8;
3352 }
3353 
3354 static void __init setup_usemap(struct pglist_data *pgdat,
3355                                 struct zone *zone, unsigned long zonesize)
3356 {
3357         unsigned long usemapsize = usemap_size(zonesize);
3358         zone->pageblock_flags = NULL;
3359         if (usemapsize) {
3360                 zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3361                 memset(zone->pageblock_flags, 0, usemapsize);
3362         }
3363 }
3364 #else
3365 static void inline setup_usemap(struct pglist_data *pgdat,
3366                                 struct zone *zone, unsigned long zonesize) {}
3367 #endif /* CONFIG_SPARSEMEM */
3368 
3369 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3370 
3371 /* Return a sensible default order for the pageblock size. */
3372 static inline int pageblock_default_order(void)
3373 {
3374         if (HPAGE_SHIFT > PAGE_SHIFT)
3375                 return HUGETLB_PAGE_ORDER;
3376 
3377         return MAX_ORDER-1;
3378 }
3379 
3380 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3381 static inline void __init set_pageblock_order(unsigned int order)
3382 {
3383         /* Check that pageblock_nr_pages has not already been setup */
3384         if (pageblock_order)
3385                 return;
3386 
3387         /*
3388          * Assume the largest contiguous order of interest is a huge page.
3389          * This value may be variable depending on boot parameters on IA64
3390          */
3391         pageblock_order = order;
3392 }
3393 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3394 
3395 /*
3396  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3397  * and pageblock_default_order() are unused as pageblock_order is set
3398  * at compile-time. See include/linux/pageblock-flags.h for the values of
3399  * pageblock_order based on the kernel config
3400  */
3401 static inline int pageblock_default_order(unsigned int order)
3402 {
3403         return MAX_ORDER-1;
3404 }
3405 #define set_pageblock_order(x)  do {} while (0)
3406 
3407 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3408 
3409 /*
3410  * Set up the zone data structures:
3411  *   - mark all pages reserved
3412  *   - mark all memory queues empty
3413  *   - clear the memory bitmaps
3414  */
3415 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3416                 unsigned long *zones_size, unsigned long *zholes_size)
3417 {
3418         enum zone_type j;
3419         int nid = pgdat->node_id;
3420         unsigned long zone_start_pfn = pgdat->node_start_pfn;
3421         int ret;
3422 
3423         pgdat_resize_init(pgdat);
3424         pgdat->nr_zones = 0;
3425         init_waitqueue_head(&pgdat->kswapd_wait);
3426         pgdat->kswapd_max_order = 0;
3427         
3428         for (j = 0; j < MAX_NR_ZONES; j++) {
3429                 struct zone *zone = pgdat->node_zones + j;
3430                 unsigned long size, realsize, memmap_pages;
3431 
3432                 size = zone_spanned_pages_in_node(nid, j, zones_size);
3433                 realsize = size - zone_absent_pages_in_node(nid, j,
3434                                                                 zholes_size);
3435 
3436                 /*
3437                  * Adjust realsize so that it accounts for how much memory
3438                  * is used by this zone for memmap. This affects the watermark
3439                  * and per-cpu initialisations
3440                  */
3441                 memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT;
3442                 if (realsize >= memmap_pages) {
3443                         realsize -= memmap_pages;
3444                         printk(KERN_DEBUG
3445                                 "  %s zone: %lu pages used for memmap\n",
3446                                 zone_names[j], memmap_pages);
3447                 } else
3448                         printk(KERN_WARNING
3449                                 "  %s zone: %lu pages exceeds realsize %lu\n",
3450                                 zone_names[j], memmap_pages, realsize);
3451 
3452                 /* Account for reserved pages */
3453                 if (j == 0 && realsize > dma_reserve) {
3454                         realsize -= dma_reserve;
3455                         printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
3456                                         zone_names[0], dma_reserve);
3457                 }
3458 
3459                 if (!is_highmem_idx(j))
3460                         nr_kernel_pages += realsize;
3461                 nr_all_pages += realsize;
3462 
3463                 zone->spanned_pages = size;
3464                 zone->present_pages = realsize;
3465 #ifdef CONFIG_NUMA
3466                 zone->node = nid;
3467                 zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3468                                                 / 100;
3469                 zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3470 #endif
3471                 zone->name = zone_names[j];
3472                 spin_lock_init(&zone->lock);
3473                 spin_lock_init(&zone->lru_lock);
3474                 zone_seqlock_init(zone);
3475                 zone->zone_pgdat = pgdat;
3476 
3477                 zone->prev_priority = DEF_PRIORITY;
3478 
3479                 zone_pcp_init(zone);
3480                 INIT_LIST_HEAD(&zone->active_list);
3481                 INIT_LIST_HEAD(&zone->inactive_list);
3482                 zone->nr_scan_active = 0;
3483                 zone->nr_scan_inactive = 0;
3484                 zap_zone_vm_stats(zone);
3485                 zone->flags = 0;
3486                 if (!size)
3487                         continue;
3488 
3489                 set_pageblock_order(pageblock_default_order());
3490                 setup_usemap(pgdat, zone, size);
3491                 ret = init_currently_empty_zone(zone, zone_start_pfn,
3492                                                 size, MEMMAP_EARLY);
3493                 BUG_ON(ret);
3494                 memmap_init(size, nid, j, zone_start_pfn);
3495                 zone_start_pfn += size;
3496         }
3497 }
3498 
3499 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3500 {
3501         /* Skip empty nodes */
3502         if (!pgdat->node_spanned_pages)
3503                 return;
3504 
3505 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3506         /* ia64 gets its own node_mem_map, before this, without bootmem */
3507         if (!pgdat->node_mem_map) {
3508                 unsigned long size, start, end;
3509                 struct page *map;
3510 
3511                 /*
3512                  * The zone's endpoints aren't required to be MAX_ORDER
3513                  * aligned but the node_mem_map endpoints must be in order
3514                  * for the buddy allocator to function correctly.
3515                  */
3516                 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3517                 end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3518                 end = ALIGN(end, MAX_ORDER_NR_PAGES);
3519                 size =  (end - start) * sizeof(struct page);
3520                 map = alloc_remap(pgdat->node_id, size);
3521                 if (!map)
3522                         map = alloc_bootmem_node(pgdat, size);
3523                 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3524         }
3525 #ifndef CONFIG_NEED_MULTIPLE_NODES
3526         /*
3527          * With no DISCONTIG, the global mem_map is just set as node 0's
3528          */
3529         if (pgdat == NODE_DATA(0)) {
3530                 mem_map = NODE_DATA(0)->node_mem_map;
3531 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3532                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3533                         mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3534 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3535         }
3536 #endif
3537 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3538 }
3539 
3540 void __paginginit free_area_init_node(int nid, struct pglist_data *pgdat,
3541                 unsigned long *zones_size, unsigned long node_start_pfn,
3542                 unsigned long *zholes_size)
3543 {
3544         pgdat->node_id = nid;
3545         pgdat->node_start_pfn = node_start_pfn;
3546         calculate_node_totalpages(pgdat, zones_size, zholes_size);
3547 
3548         alloc_node_mem_map(pgdat);
3549 
3550         free_area_init_core(pgdat, zones_size, zholes_size);
3551 }
3552 
3553 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3554 
3555 #if MAX_NUMNODES > 1
3556 /*
3557  * Figure out the number of possible node ids.
3558  */
3559 static void __init setup_nr_node_ids(void)
3560 {
3561         unsigned int node;
3562         unsigned int highest = 0;
3563 
3564         for_each_node_mask(node, node_possible_map)
3565                 highest = node;
3566         nr_node_ids = highest + 1;
3567 }
3568 #else
3569 static inline void setup_nr_node_ids(void)
3570 {
3571 }
3572 #endif
3573 
3574 /**
3575  * add_active_range - Register a range of PFNs backed by physical memory
3576  * @nid: The node ID the range resides on
3577  * @start_pfn: The start PFN of the available physical memory
3578  * @end_pfn: The end PFN of the available physical memory
3579  *
3580  * These ranges are stored in an early_node_map[] and later used by
3581  * free_area_init_nodes() to calculate zone sizes and holes. If the
3582  * range spans a memory hole, it is up to the architecture to ensure
3583  * the memory is not freed by the bootmem allocator. If possible
3584  * the range being registered will be merged with existing ranges.
3585  */
3586 void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3587                                                 unsigned long end_pfn)
3588 {
3589         int i;
3590 
3591         printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) "
3592                           "%d entries of %d used\n",
3593                           nid, start_pfn, end_pfn,
3594                           nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3595 
3596         /* Merge with existing active regions if possible */
3597         for (i = 0; i < nr_nodemap_entries; i++) {
3598                 if (early_node_map[i].nid != nid)
3599                         continue;
3600 
3601                 /* Skip if an existing region covers this new one */
3602                 if (start_pfn >= early_node_map[i].start_pfn &&
3603                                 end_pfn <= early_node_map[i].end_pfn)
3604                         return;
3605 
3606                 /* Merge forward if suitable */
3607                 if (start_pfn <= early_node_map[i].end_pfn &&
3608                                 end_pfn > early_node_map[i].end_pfn) {
3609                         early_node_map[i].end_pfn = end_pfn;
3610                         return;
3611                 }
3612 
3613                 /* Merge backward if suitable */
3614                 if (start_pfn < early_node_map[i].end_pfn &&
3615                                 end_pfn >= early_node_map[i].start_pfn) {
3616                         early_node_map[i].start_pfn = start_pfn;
3617                         return;
3618                 }
3619         }
3620 
3621         /* Check that early_node_map is large enough */
3622         if (i >= MAX_ACTIVE_REGIONS) {
3623                 printk(KERN_CRIT "More than %d memory regions, truncating\n",
3624                                                         MAX_ACTIVE_REGIONS);
3625                 return;
3626         }
3627 
3628         early_node_map[i].nid = nid;
3629         early_node_map[i].start_pfn = start_pfn;
3630         early_node_map[i].end_pfn = end_pfn;
3631         nr_nodemap_entries = i + 1;
3632 }
3633 
3634 /**
3635  * shrink_active_range - Shrink an existing registered range of PFNs
3636  * @nid: The node id the range is on that should be shrunk
3637  * @old_end_pfn: The old end PFN of the range
3638  * @new_end_pfn: The new PFN of the range
3639  *
3640  * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3641  * The map is kept at the end physical page range that has already been
3642  * registered with add_active_range(). This function allows an arch to shrink
3643  * an existing registered range.
3644  */
3645 void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn,
3646                                                 unsigned long new_end_pfn)
3647 {
3648         int i;
3649 
3650         /* Find the old active region end and shrink */
3651         for_each_active_range_index_in_nid(i, nid)
3652                 if (early_node_map[i].end_pfn == old_end_pfn) {
3653                         early_node_map[i].end_pfn = new_end_pfn;
3654                         break;
3655                 }
3656 }
3657 
3658 /**
3659  * remove_all_active_ranges - Remove all currently registered regions
3660  *
3661  * During discovery, it may be found that a table like SRAT is invalid
3662  * and an alternative discovery method must be used. This function removes
3663  * all currently registered regions.
3664  */
3665 void __init remove_all_active_ranges(void)
3666 {
3667         memset(early_node_map, 0, sizeof(early_node_map));
3668         nr_nodemap_entries = 0;
3669 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3670         memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3671         memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3672 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3673 }
3674 
3675 /* Compare two active node_active_regions */
3676 static int __init cmp_node_active_region(const void *a, const void *b)
3677 {
3678         struct node_active_region *arange = (struct node_active_region *)a;
3679         struct node_active_region *brange = (struct node_active_region *)b;
3680 
3681         /* Done this way to avoid overflows */
3682         if (arange->start_pfn > brange->start_pfn)
3683                 return 1;
3684         if (arange->start_pfn < brange->start_pfn)
3685                 return -1;
3686 
3687         return 0;
3688 }
3689 
3690 /* sort the node_map by start_pfn */
3691 static void __init sort_node_map(void)
3692 {
3693         sort(early_node_map, (size_t)nr_nodemap_entries,
3694                         sizeof(struct node_active_region),
3695                         cmp_node_active_region, NULL);
3696 }
3697 
3698 /* Find the lowest pfn for a node */
3699 unsigned long __init find_min_pfn_for_node(unsigned long nid)
3700 {
3701         int i;
3702         unsigned long min_pfn = ULONG_MAX;
3703 
3704         /* Assuming a sorted map, the first range found has the starting pfn */
3705         for_each_active_range_index_in_nid(i, nid)
3706                 min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3707 
3708         if (min_pfn == ULONG_MAX) {
3709                 printk(KERN_WARNING
3710                         "Could not find start_pfn for node %lu\n", nid);
3711                 return 0;
3712         }
3713 
3714         return min_pfn;
3715 }
3716 
3717 /**
3718  * find_min_pfn_with_active_regions - Find the minimum PFN registered
3719  *
3720  * It returns the minimum PFN based on information provided via
3721  * add_active_range().
3722  */
3723 unsigned long __init find_min_pfn_with_active_regions(void)
3724 {
3725         return find_min_pfn_for_node(MAX_NUMNODES);
3726 }
3727 
3728 /**
3729  * find_max_pfn_with_active_regions - Find the maximum PFN registered
3730  *
3731  * It returns the maximum PFN based on information provided via
3732  * add_active_range().
3733  */
3734 unsigned long __init find_max_pfn_with_active_regions(void)
3735 {
3736         int i;
3737         unsigned long max_pfn = 0;
3738 
3739         for (i = 0; i < nr_nodemap_entries; i++)
3740                 max_pfn = max(max_pfn, early_node_map[i].end_pfn);
3741 
3742         return max_pfn;
3743 }
3744 
3745 /*
3746  * early_calculate_totalpages()
3747  * Sum pages in active regions for movable zone.
3748  * Populate N_HIGH_MEMORY for calculating usable_nodes.
3749  */
3750 static unsigned long __init early_calculate_totalpages(void)
3751 {
3752         int i;
3753         unsigned long totalpages = 0;
3754 
3755         for (i = 0; i < nr_nodemap_entries; i++) {
3756                 unsigned long pages = early_node_map[i].end_pfn -
3757                                                 early_node_map[i].start_pfn;
3758                 totalpages += pages;
3759                 if (pages)
3760                         node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3761         }
3762         return totalpages;
3763 }
3764 
3765 /*
3766  * Find the PFN the Movable zone begins in each node. Kernel memory
3767  * is spread evenly between nodes as long as the nodes have enough
3768  * memory. When they don't, some nodes will have more kernelcore than
3769  * others
3770  */
3771 void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3772 {
3773         int i, nid;
3774         unsigned long usable_startpfn;
3775         unsigned long kernelcore_node, kernelcore_remaining;
3776         unsigned long totalpages = early_calculate_totalpages();
3777         int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3778 
3779         /*
3780          * If movablecore was specified, calculate what size of
3781          * kernelcore that corresponds so that memory usable for
3782          * any allocation type is evenly spread. If both kernelcore
3783          * and movablecore are specified, then the value of kernelcore
3784          * will be used for required_kernelcore if it's greater than
3785          * what movablecore would have allowed.
3786          */
3787         if (required_movablecore) {
3788                 unsigned long corepages;
3789 
3790                 /*
3791                  * Round-up so that ZONE_MOVABLE is at least as large as what
3792                  * was requested by the user
3793                  */
3794                 required_movablecore =
3795                         roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3796                 corepages = totalpages - required_movablecore;
3797 
3798                 required_kernelcore = max(required_kernelcore, corepages);
3799         }
3800 
3801         /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3802         if (!required_kernelcore)
3803                 return;
3804 
3805         /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3806         find_usable_zone_for_movable();
3807         usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3808 
3809 restart:
3810         /* Spread kernelcore memory as evenly as possible throughout nodes */
3811         kernelcore_node = required_kernelcore / usable_nodes;
3812         for_each_node_state(nid, N_HIGH_MEMORY) {
3813                 /*
3814                  * Recalculate kernelcore_node if the division per node
3815                  * now exceeds what is necessary to satisfy the requested
3816                  * amount of memory for the kernel
3817                  */
3818                 if (required_kernelcore < kernelcore_node)
3819                         kernelcore_node = required_kernelcore / usable_nodes;
3820 
3821                 /*
3822                  * As the map is walked, we track how much memory is usable
3823                  * by the kernel using kernelcore_remaining. When it is
3824                  * 0, the rest of the node is usable by ZONE_MOVABLE
3825                  */
3826                 kernelcore_remaining = kernelcore_node;
3827 
3828                 /* Go through each range of PFNs within this node */
3829                 for_each_active_range_index_in_nid(i, nid) {
3830                         unsigned long start_pfn, end_pfn;
3831                         unsigned long size_pages;
3832 
3833                         start_pfn = max(early_node_map[i].start_pfn,
3834                                                 zone_movable_pfn[nid]);
3835                         end_pfn = early_node_map[i].end_pfn;
3836                         if (start_pfn >= end_pfn)
3837                                 continue;
3838 
3839                         /* Account for what is only usable for kernelcore */
3840                         if (start_pfn < usable_startpfn) {
3841                                 unsigned long kernel_pages;
3842                                 kernel_pages = min(end_pfn, usable_startpfn)
3843                                                                 - start_pfn;
3844 
3845                                 kernelcore_remaining -= min(kernel_pages,
3846                                                         kernelcore_remaining);
3847                                 required_kernelcore -= min(kernel_pages,
3848                                                         required_kernelcore);
3849 
3850                                 /* Continue if range is now fully accounted */
3851                                 if (end_pfn <= usable_startpfn) {
3852 
3853                                         /*
3854                                          * Push zone_movable_pfn to the end so
3855                                          * that if we have to rebalance
3856                                          * kernelcore across nodes, we will
3857                                          * not double account here
3858                                          */
3859                                         zone_movable_pfn[nid] = end_pfn;
3860                                         continue;
3861                                 }
3862                                 start_pfn = usable_startpfn;
3863                         }
3864 
3865                         /*
3866                          * The usable PFN range for ZONE_MOVABLE is from
3867                          * start_pfn->end_pfn. Calculate size_pages as the
3868                          * number of pages used as kernelcore
3869                          */
3870                         size_pages = end_pfn - start_pfn;
3871                         if (size_pages > kernelcore_remaining)
3872                                 size_pages = kernelcore_remaining;
3873                         zone_movable_pfn[nid] = start_pfn + size_pages;
3874 
3875                         /*
3876                          * Some kernelcore has been met, update counts and
3877                          * break if the kernelcore for this node has been
3878                          * satisified
3879                          */
3880                         required_kernelcore -= min(required_kernelcore,
3881                                                                 size_pages);
3882                         kernelcore_remaining -= size_pages;
3883                         if (!kernelcore_remaining)
3884                                 break;
3885                 }
3886         }
3887 
3888         /*
3889          * If there is still required_kernelcore, we do another pass with one
3890          * less node in the count. This will push zone_movable_pfn[nid] further
3891          * along on the nodes that still have memory until kernelcore is
3892          * satisified
3893          */
3894         usable_nodes--;
3895         if (usable_nodes && required_kernelcore > usable_nodes)
3896                 goto restart;
3897 
3898         /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3899         for (nid = 0; nid < MAX_NUMNODES; nid++)
3900                 zone_movable_pfn[nid] =
3901                         roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
3902 }
3903 
3904 /* Any regular memory on that node ? */
3905 static void check_for_regular_memory(pg_data_t *pgdat)
3906 {
3907 #ifdef CONFIG_HIGHMEM
3908         enum zone_type zone_type;
3909 
3910         for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
3911                 struct zone *zone = &pgdat->node_zones[zone_type];
3912                 if (zone->present_pages)
3913                         node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
3914         }
3915 #endif
3916 }
3917 
3918 /**
3919  * free_area_init_nodes - Initialise all pg_data_t and zone data
3920  * @max_zone_pfn: an array of max PFNs for each zone
3921  *
3922  * This will call free_area_init_node() for each active node in the system.
3923  * Using the page ranges provided by add_active_range(), the size of each
3924  * zone in each node and their holes is calculated. If the maximum PFN
3925  * between two adjacent zones match, it is assumed that the zone is empty.
3926  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3927  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3928  * starts where the previous one ended. For example, ZONE_DMA32 starts
3929  * at arch_max_dma_pfn.
3930  */
3931 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
3932 {
3933         unsigned long nid;
3934         enum zone_type i;
3935 
3936         /* Sort early_node_map as initialisation assumes it is sorted */
3937         sort_node_map();
3938 
3939         /* Record where the zone boundaries are */
3940         memset(arch_zone_lowest_possible_pfn, 0,
3941                                 sizeof(arch_zone_lowest_possible_pfn));
3942         memset(arch_zone_highest_possible_pfn, 0,
3943                                 sizeof(arch_zone_highest_possible_pfn));
3944         arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
3945         arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
3946         for (i = 1; i < MAX_NR_ZONES; i++) {
3947                 if (i == ZONE_MOVABLE)
3948                         continue;
3949                 arch_zone_lowest_possible_pfn[i] =
3950                         arch_zone_highest_possible_pfn[i-1];
3951                 arch_zone_highest_possible_pfn[i] =
3952                         max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
3953         }
3954         arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
3955         arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
3956 
3957         /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3958         memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
3959         find_zone_movable_pfns_for_nodes(zone_movable_pfn);
3960 
3961         /* Print out the zone ranges */
3962         printk("Zone PFN ranges:\n");
3963         for (i = 0; i < MAX_NR_ZONES; i++) {
3964                 if (i == ZONE_MOVABLE)
3965                         continue;
3966                 printk("  %-8s %8lu -> %8lu\n",
3967                                 zone_names[i],
3968                                 arch_zone_lowest_possible_pfn[i],
3969                                 arch_zone_highest_possible_pfn[i]);
3970         }
3971 
3972         /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3973         printk("Movable zone start PFN for each node\n");
3974         for (i = 0; i < MAX_NUMNODES; i++) {
3975                 if (zone_movable_pfn[i])
3976                         printk("  Node %d: %lu\n", i, zone_movable_pfn[i]);
3977         }
3978 
3979         /* Print out the early_node_map[] */
3980         printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
3981         for (i = 0; i < nr_nodemap_entries; i++)
3982                 printk("  %3d: %8lu -> %8lu\n", early_node_map[i].nid,
3983                                                 early_node_map[i].start_pfn,
3984                                                 early_node_map[i].end_pfn);
3985 
3986         /* Initialise every node */
3987         setup_nr_node_ids();
3988         for_each_online_node(nid) {
3989                 pg_data_t *pgdat = NODE_DATA(nid);
3990                 free_area_init_node(nid, pgdat, NULL,
3991                                 find_min_pfn_for_node(nid), NULL);
3992 
3993                 /* Any memory on that node */
3994                 if (pgdat->node_present_pages)
3995                         node_set_state(nid, N_HIGH_MEMORY);
3996                 check_for_regular_memory(pgdat);
3997         }
3998 }
3999 
4000 static int __init cmdline_parse_core(char *p, unsigned long *core)
4001 {
4002         unsigned long long coremem;
4003         if (!p)
4004                 return -EINVAL;
4005 
4006         coremem = memparse(p, &p);
4007         *core = coremem >> PAGE_SHIFT;
4008 
4009         /* Paranoid check that UL is enough for the coremem value */
4010         WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4011 
4012         return 0;
4013 }
4014 
4015 /*
4016  * kernelcore=size sets the amount of memory for use for allocations that
4017  * cannot be reclaimed or migrated.
4018  */
4019 static int __init cmdline_parse_kernelcore(char *p)
4020 {
4021         return cmdline_parse_core(p, &required_kernelcore);
4022 }
4023 
4024 /*
4025  * movablecore=size sets the amount of memory for use for allocations that
4026  * can be reclaimed or migrated.
4027  */
4028 static int __init cmdline_parse_movablecore(char *p)
4029 {
4030         return cmdline_parse_core(p, &required_movablecore);
4031 }
4032 
4033 early_param("kernelcore", cmdline_parse_kernelcore);
4034 early_param("movablecore", cmdline_parse_movablecore);
4035 
4036 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4037 
4038 /**
4039  * set_dma_reserve - set the specified number of pages reserved in the first zone
4040  * @new_dma_reserve: The number of pages to mark reserved
4041  *
4042  * The per-cpu batchsize and zone watermarks are determined by present_pages.
4043  * In the DMA zone, a significant percentage may be consumed by kernel image
4044  * and other unfreeable allocations which can skew the watermarks badly. This
4045  * function may optionally be used to account for unfreeable pages in the
4046  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4047  * smaller per-cpu batchsize.
4048  */
4049 void __init set_dma_reserve(unsigned long new_dma_reserve)
4050 {
4051         dma_reserve = new_dma_reserve;
4052 }
4053 
4054 #ifndef CONFIG_NEED_MULTIPLE_NODES
4055 static bootmem_data_t contig_bootmem_data;
4056 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
4057 
4058 EXPORT_SYMBOL(contig_page_data);
4059 #endif
4060 
4061 void __init free_area_init(unsigned long *zones_size)
4062 {
4063         free_area_init_node(0, NODE_DATA(0), zones_size,
4064                         __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4065 }
4066 
4067 static int page_alloc_cpu_notify(struct notifier_block *self,
4068                                  unsigned long action, void *hcpu)
4069 {
4070         int cpu = (unsigned long)hcpu;
4071 
4072         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4073                 drain_pages(cpu);
4074 
4075                 /*
4076                  * Spill the event counters of the dead processor
4077                  * into the current processors event counters.
4078                  * This artificially elevates the count of the current
4079                  * processor.
4080                  */
4081                 vm_events_fold_cpu(cpu);
4082 
4083                 /*
4084                  * Zero the differential counters of the dead processor
4085                  * so that the vm statistics are consistent.
4086                  *
4087                  * This is only okay since the processor is dead and cannot
4088                  * race with what we are doing.
4089                  */
4090                 refresh_cpu_vm_stats(cpu);
4091         }
4092         return NOTIFY_OK;
4093 }
4094 
4095 void __init page_alloc_init(void)
4096 {
4097         hotcpu_notifier(page_alloc_cpu_notify, 0);
4098 }
4099 
4100 /*
4101  * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4102  *      or min_free_kbytes changes.
4103  */
4104 static void calculate_totalreserve_pages(void)
4105 {
4106         struct pglist_data *pgdat;
4107         unsigned long reserve_pages = 0;
4108         enum zone_type i, j;
4109 
4110         for_each_online_pgdat(pgdat) {
4111                 for (i = 0; i < MAX_NR_ZONES; i++) {
4112                         struct zone *zone = pgdat->node_zones + i;
4113                         unsigned long max = 0;
4114 
4115                         /* Find valid and maximum lowmem_reserve in the zone */
4116                         for (j = i; j < MAX_NR_ZONES; j++) {
4117                                 if (zone->lowmem_reserve[j] > max)
4118                                         max = zone->lowmem_reserve[j];
4119                         }
4120 
4121                         /* we treat pages_high as reserved pages. */
4122                         max += zone->pages_high;
4123 
4124                         if (max > zone->present_pages)
4125                                 max = zone->present_pages;
4126                         reserve_pages += max;
4127                 }
4128         }
4129         totalreserve_pages = reserve_pages;
4130 }
4131 
4132 /*
4133  * setup_per_zone_lowmem_reserve - called whenever
4134  *      sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
4135  *      has a correct pages reserved value, so an adequate number of
4136  *      pages are left in the zone after a successful __alloc_pages().
4137  */
4138 static void setup_per_zone_lowmem_reserve(void)
4139 {
4140         struct pglist_data *pgdat;
4141         enum zone_type j, idx;
4142 
4143         for_each_online_pgdat(pgdat) {
4144                 for (j = 0; j < MAX_NR_ZONES; j++) {
4145                         struct zone *zone = pgdat->node_zones + j;
4146                         unsigned long present_pages = zone->present_pages;
4147 
4148                         zone->lowmem_reserve[j] = 0;
4149 
4150                         idx = j;
4151                         while (idx) {
4152                                 struct zone *lower_zone;
4153 
4154                                 idx--;
4155 
4156                                 if (sysctl_lowmem_reserve_ratio[idx] < 1)
4157                                         sysctl_lowmem_reserve_ratio[idx] = 1;
4158 
4159                                 lower_zone = pgdat->node_zones + idx;
4160                                 lower_zone->lowmem_reserve[j] = present_pages /
4161                                         sysctl_lowmem_reserve_ratio[idx];
4162                                 present_pages += lower_zone->present_pages;
4163                         }
4164                 }
4165         }
4166 
4167         /* update totalreserve_pages */
4168         calculate_totalreserve_pages();
4169 }
4170 
4171 /**
4172  * setup_per_zone_pages_min - called when min_free_kbytes changes.
4173  *
4174  * Ensures that the pages_{min,low,high} values for each zone are set correctly
4175  * with respect to min_free_kbytes.
4176  */
4177 void setup_per_zone_pages_min(void)
4178 {
4179         unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4180         unsigned long lowmem_pages = 0;
4181         struct zone *zone;
4182         unsigned long flags;
4183 
4184         /* Calculate total number of !ZONE_HIGHMEM pages */
4185         for_each_zone(zone) {
4186                 if (!is_highmem(zone))
4187                         lowmem_pages += zone->present_pages;
4188         }
4189 
4190         for_each_zone(zone) {
4191                 u64 tmp;
4192 
4193                 spin_lock_irqsave(&zone->lru_lock, flags);
4194                 tmp = (u64)pages_min * zone->present_pages;
4195                 do_div(tmp, lowmem_pages);
4196                 if (is_highmem(zone)) {
4197                         /*
4198                          * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4199                          * need highmem pages, so cap pages_min to a small
4200                          * value here.
4201                          *
4202                          * The (pages_high-pages_low) and (pages_low-pages_min)
4203                          * deltas controls asynch page reclaim, and so should
4204                          * not be capped for highmem.
4205                          */
4206                         int min_pages;
4207 
4208                         min_pages = zone->present_pages / 1024;
4209                         if (min_pages < SWAP_CLUSTER_MAX)
4210                                 min_pages = SWAP_CLUSTER_MAX;
4211                         if (min_pages > 128)
4212                                 min_pages = 128;
4213                         zone->pages_min = min_pages;
4214                 } else {
4215                         /*
4216                          * If it's a lowmem zone, reserve a number of pages
4217                          * proportionate to the zone's size.
4218                          */
4219                         zone->pages_min = tmp;
4220                 }
4221 
4222                 zone->pages_low   = zone->pages_min + (tmp >> 2);
4223                 zone->pages_high  = zone->pages_min + (tmp >> 1);
4224                 setup_zone_migrate_reserve(zone);
4225                 spin_unlock_irqrestore(&zone->lru_lock, flags);
4226         }
4227 
4228         /* update totalreserve_pages */
4229         calculate_totalreserve_pages();
4230 }
4231 
4232 /*
4233  * Initialise min_free_kbytes.
4234  *
4235  * For small machines we want it small (128k min).  For large machines
4236  * we want it large (64MB max).  But it is not linear, because network
4237  * bandwidth does not increase linearly with machine size.  We use
4238  *
4239  *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4240  *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
4241  *
4242  * which yields
4243  *
4244  * 16MB:        512k
4245  * 32MB:        724k
4246  * 64MB:        1024k
4247  * 128MB:       1448k
4248  * 256MB:       2048k
4249  * 512MB:       2896k
4250  * 1024MB:      4096k
4251  * 2048MB:      5792k
4252  * 4096MB:      8192k
4253  * 8192MB:      11584k
4254  * 16384MB:     16384k
4255  */
4256 static int __init init_per_zone_pages_min(void)
4257 {
4258         unsigned long lowmem_kbytes;
4259 
4260         lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4261 
4262         min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4263         if (min_free_kbytes < 128)
4264                 min_free_kbytes = 128;
4265         if (min_free_kbytes > 65536)
4266                 min_free_kbytes = 65536;
4267         setup_per_zone_pages_min();
4268         setup_per_zone_lowmem_reserve();
4269         return 0;
4270 }
4271 module_init(init_per_zone_pages_min)
4272 
4273 /*
4274  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so 
4275  *      that we can call two helper functions whenever min_free_kbytes
4276  *      changes.
4277  */
4278 int min_free_kbytes_sysctl_handler(ctl_table *table, int write, 
4279         struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4280 {
4281         proc_dointvec(table, write, file, buffer, length, ppos);
4282         if (write)
4283                 setup_per_zone_pages_min();
4284         return 0;
4285 }
4286 
4287 #ifdef CONFIG_NUMA
4288 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4289         struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4290 {
4291         struct zone *zone;
4292         int rc;
4293 
4294         rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4295         if (rc)
4296                 return rc;
4297 
4298         for_each_zone(zone)
4299                 zone->min_unmapped_pages = (zone->present_pages *
4300                                 sysctl_min_unmapped_ratio) / 100;
4301         return 0;
4302 }
4303 
4304 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4305         struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4306 {
4307         struct zone *zone;
4308         int rc;
4309 
4310         rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4311         if (rc)
4312                 return rc;
4313 
4314         for_each_zone(zone)
4315                 zone->min_slab_pages = (zone->present_pages *
4316                                 sysctl_min_slab_ratio) / 100;
4317         return 0;
4318 }
4319 #endif
4320 
4321 /*
4322  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4323  *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4324  *      whenever sysctl_lowmem_reserve_ratio changes.
4325  *
4326  * The reserve ratio obviously has absolutely no relation with the
4327  * pages_min watermarks. The lowmem reserve ratio can only make sense
4328  * if in function of the boot time zone sizes.
4329  */
4330 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4331         struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4332 {
4333         proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4334         setup_per_zone_lowmem_reserve();
4335         return 0;
4336 }
4337 
4338 /*
4339  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4340  * cpu.  It is the fraction of total pages in each zone that a hot per cpu pagelist
4341  * can have before it gets flushed back to buddy allocator.
4342  */
4343 
4344 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4345         struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4346 {
4347         struct zone *zone;
4348         unsigned int cpu;
4349         int ret;
4350 
4351         ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4352         if (!write || (ret == -EINVAL))
4353                 return ret;
4354         for_each_zone(zone) {
4355                 for_each_online_cpu(cpu) {
4356                         unsigned long  high;
4357                         high = zone->present_pages / percpu_pagelist_fraction;
4358                         setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4359                 }
4360         }
4361         return 0;
4362 }
4363 
4364 int hashdist = HASHDIST_DEFAULT;
4365 
4366 #ifdef CONFIG_NUMA
4367 static int __init set_hashdist(char *str)
4368 {
4369         if (!str)
4370                 return 0;
4371         hashdist = simple_strtoul(str, &str, 0);
4372         return 1;
4373 }
4374 __setup("hashdist=", set_hashdist);
4375 #endif
4376 
4377 /*
4378  * allocate a large system hash table from bootmem
4379  * - it is assumed that the hash table must contain an exact power-of-2
4380  *   quantity of entries
4381  * - limit is the number of hash buckets, not the total allocation size
4382  */
4383 void *__init alloc_large_system_hash(const char *tablename,
4384                                      unsigned long bucketsize,
4385                                      unsigned long numentries,
4386                                      int scale,
4387                                      int flags,
4388                                      unsigned int *_hash_shift,
4389                                      unsigned int *_hash_mask,
4390                                      unsigned long limit)
4391 {
4392         unsigned long long max = limit;
4393         unsigned long log2qty, size;
4394         void *table = NULL;
4395 
4396         /* allow the kernel cmdline to have a say */
4397         if (!numentries) {
4398                 /* round applicable memory size up to nearest megabyte */
4399                 numentries = nr_kernel_pages;
4400                 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4401                 numentries >>= 20 - PAGE_SHIFT;
4402                 numentries <<= 20 - PAGE_SHIFT;
4403 
4404                 /* limit to 1 bucket per 2^scale bytes of low memory */
4405                 if (scale > PAGE_SHIFT)
4406                         numentries >>= (scale - PAGE_SHIFT);
4407                 else
4408                         numentries <<= (PAGE_SHIFT - scale);
4409 
4410                 /* Make sure we've got at least a 0-order allocation.. */
4411                 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4412                         numentries = PAGE_SIZE / bucketsize;
4413         }
4414         numentries = roundup_pow_of_two(numentries);
4415 
4416         /* limit allocation size to 1/16 total memory by default */
4417         if (max == 0) {
4418                 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4419                 do_div(max, bucketsize);
4420         }
4421 
4422         if (numentries > max)
4423                 numentries = max;
4424 
4425         log2qty = ilog2(numentries);
4426 
4427         do {
4428                 size = bucketsize << log2qty;
4429                 if (flags & HASH_EARLY)
4430                         table = alloc_bootmem(size);
4431                 else if (hashdist)
4432                         table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4433                 else {
4434                         unsigned long order;
4435                         for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
4436                                 ;
4437                         table = (void*) __get_free_pages(GFP_ATOMIC, order);
4438                         /*
4439                          * If bucketsize is not a power-of-two, we may free
4440                          * some pages at the end of hash table.
4441                          */
4442                         if (table) {
4443                                 unsigned long alloc_end = (unsigned long)table +
4444                                                 (PAGE_SIZE << order);
4445                                 unsigned long used = (unsigned long)table +
4446                                                 PAGE_ALIGN(size);
4447                                 split_page(virt_to_page(table), order);
4448                                 while (used < alloc_end) {
4449                                         free_page(used);
4450                                         used += PAGE_SIZE;
4451                                 }
4452                         }
4453                 }
4454         } while (!table && size > PAGE_SIZE && --log2qty);
4455 
4456         if (!table)
4457                 panic("Failed to allocate %s hash table\n", tablename);
4458 
4459         printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4460                tablename,
4461                (1U << log2qty),
4462                ilog2(size) - PAGE_SHIFT,
4463                size);
4464 
4465         if (_hash_shift)
4466                 *_hash_shift = log2qty;
4467         if (_hash_mask)
4468                 *_hash_mask = (1 << log2qty) - 1;
4469 
4470         return table;
4471 }
4472 
4473 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4474 struct page *pfn_to_page(unsigned long pfn)
4475 {
4476         return __pfn_to_page(pfn);
4477 }
4478 unsigned long page_to_pfn(struct page *page)
4479 {
4480         return __page_to_pfn(page);
4481 }
4482 EXPORT_SYMBOL(pfn_to_page);
4483 EXPORT_SYMBOL(page_to_pfn);
4484 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4485 
4486 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4487 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4488                                                         unsigned long pfn)
4489 {
4490 #ifdef CONFIG_SPARSEMEM
4491         return __pfn_to_section(pfn)->pageblock_flags;
4492 #else
4493         return zone->pageblock_flags;
4494 #endif /* CONFIG_SPARSEMEM */
4495 }
4496 
4497 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4498 {
4499 #ifdef CONFIG_SPARSEMEM
4500         pfn &= (PAGES_PER_SECTION-1);
4501         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4502 #else
4503         pfn = pfn - zone->zone_start_pfn;
4504         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4505 #endif /* CONFIG_SPARSEMEM */
4506 }
4507 
4508 /**
4509  * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4510  * @page: The page within the block of interest
4511  * @start_bitidx: The first bit of interest to retrieve
4512  * @end_bitidx: The last bit of interest
4513  * returns pageblock_bits flags
4514  */
4515 unsigned long get_pageblock_flags_group(struct page *page,
4516                                         int start_bitidx, int end_bitidx)
4517 {
4518         struct zone *zone;
4519         unsigned long *bitmap;
4520         unsigned long pfn, bitidx;
4521         unsigned long flags = 0;
4522         unsigned long value = 1;
4523 
4524         zone = page_zone(page);
4525         pfn = page_to_pfn(page);
4526         bitmap = get_pageblock_bitmap(zone, pfn);
4527         bitidx = pfn_to_bitidx(zone, pfn);
4528 
4529         for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4530                 if (test_bit(bitidx + start_bitidx, bitmap))
4531                         flags |= value;
4532 
4533         return flags;
4534 }
4535 
4536 /**
4537  * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4538  * @page: The page within the block of interest
4539  * @start_bitidx: The first bit of interest
4540  * @end_bitidx: The last bit of interest
4541  * @flags: The flags to set
4542  */
4543 void set_pageblock_flags_group(struct page *page, unsigned long flags,
4544                                         int start_bitidx, int end_bitidx)
4545 {
4546         struct zone *zone;
4547         unsigned long *bitmap;
4548         unsigned long pfn, bitidx;
4549         unsigned long value = 1;
4550 
4551         zone = page_zone(page);
4552         pfn = page_to_pfn(page);
4553         bitmap = get_pageblock_bitmap(zone, pfn);
4554         bitidx = pfn_to_bitidx(zone, pfn);
4555         VM_BUG_ON(pfn < zone->zone_start_pfn);
4556         VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4557 
4558         for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4559                 if (flags & value)
4560                         __set_bit(bitidx + start_bitidx, bitmap);
4561                 else
4562                         __clear_bit(bitidx + start_bitidx, bitmap);
4563 }
4564 
4565 /*
4566  * This is designed as sub function...plz see page_isolation.c also.
4567  * set/clear page block's type to be ISOLATE.
4568  * page allocater never alloc memory from ISOLATE block.
4569  */
4570 
4571 int set_migratetype_isolate(struct page *page)
4572 {
4573         struct zone *zone;
4574         unsigned long flags;
4575         int ret = -EBUSY;
4576 
4577         zone = page_zone(page);
4578         spin_lock_irqsave(&zone->lock, flags);
4579         /*
4580          * In future, more migrate types will be able to be isolation target.
4581          */
4582         if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4583                 goto out;
4584         set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4585         move_freepages_block(zone, page, MIGRATE_ISOLATE);
4586         ret = 0;
4587 out:
4588         spin_unlock_irqrestore(&zone->lock, flags);
4589         if (!ret)
4590                 drain_all_pages();
4591         return ret;
4592 }
4593 
4594 void unset_migratetype_isolate(struct page *page)
4595 {
4596         struct zone *zone;
4597         unsigned long flags;
4598         zone = page_zone(page);
4599         spin_lock_irqsave(&zone->lock, flags);
4600         if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4601                 goto out;
4602         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4603         move_freepages_block(zone, page, MIGRATE_MOVABLE);
4604 out:
4605         spin_unlock_irqrestore(&zone->lock, flags);
4606 }
4607 
4608 #ifdef CONFIG_MEMORY_HOTREMOVE
4609 /*
4610  * All pages in the range must be isolated before calling this.
4611  */
4612 void
4613 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4614 {
4615         struct page *page;
4616         struct zone *zone;
4617         int order, i;
4618         unsigned long pfn;
4619         unsigned long flags;
4620         /* find the first valid pfn */
4621         for (pfn = start_pfn; pfn < end_pfn; pfn++)
4622                 if (pfn_valid(pfn))
4623                         break;
4624         if (pfn == end_pfn)
4625                 return;
4626         zone = page_zone(pfn_to_page(pfn));
4627         spin_lock_irqsave(&zone->lock, flags);
4628         pfn = start_pfn;
4629         while (pfn < end_pfn) {
4630                 if (!pfn_valid(pfn)) {
4631                         pfn++;
4632                         continue;
4633                 }
4634                 page = pfn_to_page(pfn);
4635                 BUG_ON(page_count(page));
4636                 BUG_ON(!PageBuddy(page));
4637                 order = page_order(page);
4638 #ifdef CONFIG_DEBUG_VM
4639                 printk(KERN_INFO "remove from free list %lx %d %lx\n",
4640                        pfn, 1 << order, end_pfn);
4641 #endif
4642                 list_del(&page->lru);
4643                 rmv_page_order(page);
4644                 zone->free_area[order].nr_free--;
4645                 __mod_zone_page_state(zone, NR_FREE_PAGES,
4646                                       - (1UL << order));
4647                 for (i = 0; i < (1 << order); i++)
4648                         SetPageReserved((page+i));
4649                 pfn += (1 << order);
4650         }
4651         spin_unlock_irqrestore(&zone->lock, flags);
4652 }
4653 #endif
4654 
  This page was automatically generated by the LXR engine.