Version:  2.0.40 2.2.26 2.4.37 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0

Linux/mm/page_alloc.c

  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/memblock.h>
 25 #include <linux/compiler.h>
 26 #include <linux/kernel.h>
 27 #include <linux/kmemcheck.h>
 28 #include <linux/kasan.h>
 29 #include <linux/module.h>
 30 #include <linux/suspend.h>
 31 #include <linux/pagevec.h>
 32 #include <linux/blkdev.h>
 33 #include <linux/slab.h>
 34 #include <linux/ratelimit.h>
 35 #include <linux/oom.h>
 36 #include <linux/notifier.h>
 37 #include <linux/topology.h>
 38 #include <linux/sysctl.h>
 39 #include <linux/cpu.h>
 40 #include <linux/cpuset.h>
 41 #include <linux/memory_hotplug.h>
 42 #include <linux/nodemask.h>
 43 #include <linux/vmalloc.h>
 44 #include <linux/vmstat.h>
 45 #include <linux/mempolicy.h>
 46 #include <linux/stop_machine.h>
 47 #include <linux/sort.h>
 48 #include <linux/pfn.h>
 49 #include <linux/backing-dev.h>
 50 #include <linux/fault-inject.h>
 51 #include <linux/page-isolation.h>
 52 #include <linux/page_ext.h>
 53 #include <linux/debugobjects.h>
 54 #include <linux/kmemleak.h>
 55 #include <linux/compaction.h>
 56 #include <trace/events/kmem.h>
 57 #include <linux/prefetch.h>
 58 #include <linux/mm_inline.h>
 59 #include <linux/migrate.h>
 60 #include <linux/page_ext.h>
 61 #include <linux/hugetlb.h>
 62 #include <linux/sched/rt.h>
 63 #include <linux/page_owner.h>
 64 
 65 #include <asm/sections.h>
 66 #include <asm/tlbflush.h>
 67 #include <asm/div64.h>
 68 #include "internal.h"
 69 
 70 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
 71 static DEFINE_MUTEX(pcp_batch_high_lock);
 72 #define MIN_PERCPU_PAGELIST_FRACTION    (8)
 73 
 74 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
 75 DEFINE_PER_CPU(int, numa_node);
 76 EXPORT_PER_CPU_SYMBOL(numa_node);
 77 #endif
 78 
 79 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
 80 /*
 81  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
 82  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
 83  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
 84  * defined in <linux/topology.h>.
 85  */
 86 DEFINE_PER_CPU(int, _numa_mem_);                /* Kernel "local memory" node */
 87 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
 88 int _node_numa_mem_[MAX_NUMNODES];
 89 #endif
 90 
 91 /*
 92  * Array of node states.
 93  */
 94 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
 95         [N_POSSIBLE] = NODE_MASK_ALL,
 96         [N_ONLINE] = { { [0] = 1UL } },
 97 #ifndef CONFIG_NUMA
 98         [N_NORMAL_MEMORY] = { { [0] = 1UL } },
 99 #ifdef CONFIG_HIGHMEM
100         [N_HIGH_MEMORY] = { { [0] = 1UL } },
101 #endif
102 #ifdef CONFIG_MOVABLE_NODE
103         [N_MEMORY] = { { [0] = 1UL } },
104 #endif
105         [N_CPU] = { { [0] = 1UL } },
106 #endif  /* NUMA */
107 };
108 EXPORT_SYMBOL(node_states);
109 
110 /* Protect totalram_pages and zone->managed_pages */
111 static DEFINE_SPINLOCK(managed_page_count_lock);
112 
113 unsigned long totalram_pages __read_mostly;
114 unsigned long totalreserve_pages __read_mostly;
115 unsigned long totalcma_pages __read_mostly;
116 /*
117  * When calculating the number of globally allowed dirty pages, there
118  * is a certain number of per-zone reserves that should not be
119  * considered dirtyable memory.  This is the sum of those reserves
120  * over all existing zones that contribute dirtyable memory.
121  */
122 unsigned long dirty_balance_reserve __read_mostly;
123 
124 int percpu_pagelist_fraction;
125 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
126 
127 #ifdef CONFIG_PM_SLEEP
128 /*
129  * The following functions are used by the suspend/hibernate code to temporarily
130  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
131  * while devices are suspended.  To avoid races with the suspend/hibernate code,
132  * they should always be called with pm_mutex held (gfp_allowed_mask also should
133  * only be modified with pm_mutex held, unless the suspend/hibernate code is
134  * guaranteed not to run in parallel with that modification).
135  */
136 
137 static gfp_t saved_gfp_mask;
138 
139 void pm_restore_gfp_mask(void)
140 {
141         WARN_ON(!mutex_is_locked(&pm_mutex));
142         if (saved_gfp_mask) {
143                 gfp_allowed_mask = saved_gfp_mask;
144                 saved_gfp_mask = 0;
145         }
146 }
147 
148 void pm_restrict_gfp_mask(void)
149 {
150         WARN_ON(!mutex_is_locked(&pm_mutex));
151         WARN_ON(saved_gfp_mask);
152         saved_gfp_mask = gfp_allowed_mask;
153         gfp_allowed_mask &= ~GFP_IOFS;
154 }
155 
156 bool pm_suspended_storage(void)
157 {
158         if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
159                 return false;
160         return true;
161 }
162 #endif /* CONFIG_PM_SLEEP */
163 
164 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
165 int pageblock_order __read_mostly;
166 #endif
167 
168 static void __free_pages_ok(struct page *page, unsigned int order);
169 
170 /*
171  * results with 256, 32 in the lowmem_reserve sysctl:
172  *      1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
173  *      1G machine -> (16M dma, 784M normal, 224M high)
174  *      NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
175  *      HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
176  *      HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
177  *
178  * TBD: should special case ZONE_DMA32 machines here - in those we normally
179  * don't need any ZONE_NORMAL reservation
180  */
181 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
182 #ifdef CONFIG_ZONE_DMA
183          256,
184 #endif
185 #ifdef CONFIG_ZONE_DMA32
186          256,
187 #endif
188 #ifdef CONFIG_HIGHMEM
189          32,
190 #endif
191          32,
192 };
193 
194 EXPORT_SYMBOL(totalram_pages);
195 
196 static char * const zone_names[MAX_NR_ZONES] = {
197 #ifdef CONFIG_ZONE_DMA
198          "DMA",
199 #endif
200 #ifdef CONFIG_ZONE_DMA32
201          "DMA32",
202 #endif
203          "Normal",
204 #ifdef CONFIG_HIGHMEM
205          "HighMem",
206 #endif
207          "Movable",
208 };
209 
210 int min_free_kbytes = 1024;
211 int user_min_free_kbytes = -1;
212 
213 static unsigned long __meminitdata nr_kernel_pages;
214 static unsigned long __meminitdata nr_all_pages;
215 static unsigned long __meminitdata dma_reserve;
216 
217 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
218 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
219 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
220 static unsigned long __initdata required_kernelcore;
221 static unsigned long __initdata required_movablecore;
222 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
223 
224 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
225 int movable_zone;
226 EXPORT_SYMBOL(movable_zone);
227 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
228 
229 #if MAX_NUMNODES > 1
230 int nr_node_ids __read_mostly = MAX_NUMNODES;
231 int nr_online_nodes __read_mostly = 1;
232 EXPORT_SYMBOL(nr_node_ids);
233 EXPORT_SYMBOL(nr_online_nodes);
234 #endif
235 
236 int page_group_by_mobility_disabled __read_mostly;
237 
238 void set_pageblock_migratetype(struct page *page, int migratetype)
239 {
240         if (unlikely(page_group_by_mobility_disabled &&
241                      migratetype < MIGRATE_PCPTYPES))
242                 migratetype = MIGRATE_UNMOVABLE;
243 
244         set_pageblock_flags_group(page, (unsigned long)migratetype,
245                                         PB_migrate, PB_migrate_end);
246 }
247 
248 #ifdef CONFIG_DEBUG_VM
249 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
250 {
251         int ret = 0;
252         unsigned seq;
253         unsigned long pfn = page_to_pfn(page);
254         unsigned long sp, start_pfn;
255 
256         do {
257                 seq = zone_span_seqbegin(zone);
258                 start_pfn = zone->zone_start_pfn;
259                 sp = zone->spanned_pages;
260                 if (!zone_spans_pfn(zone, pfn))
261                         ret = 1;
262         } while (zone_span_seqretry(zone, seq));
263 
264         if (ret)
265                 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
266                         pfn, zone_to_nid(zone), zone->name,
267                         start_pfn, start_pfn + sp);
268 
269         return ret;
270 }
271 
272 static int page_is_consistent(struct zone *zone, struct page *page)
273 {
274         if (!pfn_valid_within(page_to_pfn(page)))
275                 return 0;
276         if (zone != page_zone(page))
277                 return 0;
278 
279         return 1;
280 }
281 /*
282  * Temporary debugging check for pages not lying within a given zone.
283  */
284 static int bad_range(struct zone *zone, struct page *page)
285 {
286         if (page_outside_zone_boundaries(zone, page))
287                 return 1;
288         if (!page_is_consistent(zone, page))
289                 return 1;
290 
291         return 0;
292 }
293 #else
294 static inline int bad_range(struct zone *zone, struct page *page)
295 {
296         return 0;
297 }
298 #endif
299 
300 static void bad_page(struct page *page, const char *reason,
301                 unsigned long bad_flags)
302 {
303         static unsigned long resume;
304         static unsigned long nr_shown;
305         static unsigned long nr_unshown;
306 
307         /* Don't complain about poisoned pages */
308         if (PageHWPoison(page)) {
309                 page_mapcount_reset(page); /* remove PageBuddy */
310                 return;
311         }
312 
313         /*
314          * Allow a burst of 60 reports, then keep quiet for that minute;
315          * or allow a steady drip of one report per second.
316          */
317         if (nr_shown == 60) {
318                 if (time_before(jiffies, resume)) {
319                         nr_unshown++;
320                         goto out;
321                 }
322                 if (nr_unshown) {
323                         printk(KERN_ALERT
324                               "BUG: Bad page state: %lu messages suppressed\n",
325                                 nr_unshown);
326                         nr_unshown = 0;
327                 }
328                 nr_shown = 0;
329         }
330         if (nr_shown++ == 0)
331                 resume = jiffies + 60 * HZ;
332 
333         printk(KERN_ALERT "BUG: Bad page state in process %s  pfn:%05lx\n",
334                 current->comm, page_to_pfn(page));
335         dump_page_badflags(page, reason, bad_flags);
336 
337         print_modules();
338         dump_stack();
339 out:
340         /* Leave bad fields for debug, except PageBuddy could make trouble */
341         page_mapcount_reset(page); /* remove PageBuddy */
342         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
343 }
344 
345 /*
346  * Higher-order pages are called "compound pages".  They are structured thusly:
347  *
348  * The first PAGE_SIZE page is called the "head page".
349  *
350  * The remaining PAGE_SIZE pages are called "tail pages".
351  *
352  * All pages have PG_compound set.  All tail pages have their ->first_page
353  * pointing at the head page.
354  *
355  * The first tail page's ->lru.next holds the address of the compound page's
356  * put_page() function.  Its ->lru.prev holds the order of allocation.
357  * This usage means that zero-order pages may not be compound.
358  */
359 
360 static void free_compound_page(struct page *page)
361 {
362         __free_pages_ok(page, compound_order(page));
363 }
364 
365 void prep_compound_page(struct page *page, unsigned long order)
366 {
367         int i;
368         int nr_pages = 1 << order;
369 
370         set_compound_page_dtor(page, free_compound_page);
371         set_compound_order(page, order);
372         __SetPageHead(page);
373         for (i = 1; i < nr_pages; i++) {
374                 struct page *p = page + i;
375                 set_page_count(p, 0);
376                 p->first_page = page;
377                 /* Make sure p->first_page is always valid for PageTail() */
378                 smp_wmb();
379                 __SetPageTail(p);
380         }
381 }
382 
383 static inline void prep_zero_page(struct page *page, unsigned int order,
384                                                         gfp_t gfp_flags)
385 {
386         int i;
387 
388         /*
389          * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
390          * and __GFP_HIGHMEM from hard or soft interrupt context.
391          */
392         VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
393         for (i = 0; i < (1 << order); i++)
394                 clear_highpage(page + i);
395 }
396 
397 #ifdef CONFIG_DEBUG_PAGEALLOC
398 unsigned int _debug_guardpage_minorder;
399 bool _debug_pagealloc_enabled __read_mostly;
400 bool _debug_guardpage_enabled __read_mostly;
401 
402 static int __init early_debug_pagealloc(char *buf)
403 {
404         if (!buf)
405                 return -EINVAL;
406 
407         if (strcmp(buf, "on") == 0)
408                 _debug_pagealloc_enabled = true;
409 
410         return 0;
411 }
412 early_param("debug_pagealloc", early_debug_pagealloc);
413 
414 static bool need_debug_guardpage(void)
415 {
416         /* If we don't use debug_pagealloc, we don't need guard page */
417         if (!debug_pagealloc_enabled())
418                 return false;
419 
420         return true;
421 }
422 
423 static void init_debug_guardpage(void)
424 {
425         if (!debug_pagealloc_enabled())
426                 return;
427 
428         _debug_guardpage_enabled = true;
429 }
430 
431 struct page_ext_operations debug_guardpage_ops = {
432         .need = need_debug_guardpage,
433         .init = init_debug_guardpage,
434 };
435 
436 static int __init debug_guardpage_minorder_setup(char *buf)
437 {
438         unsigned long res;
439 
440         if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
441                 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
442                 return 0;
443         }
444         _debug_guardpage_minorder = res;
445         printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
446         return 0;
447 }
448 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
449 
450 static inline void set_page_guard(struct zone *zone, struct page *page,
451                                 unsigned int order, int migratetype)
452 {
453         struct page_ext *page_ext;
454 
455         if (!debug_guardpage_enabled())
456                 return;
457 
458         page_ext = lookup_page_ext(page);
459         __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
460 
461         INIT_LIST_HEAD(&page->lru);
462         set_page_private(page, order);
463         /* Guard pages are not available for any usage */
464         __mod_zone_freepage_state(zone, -(1 << order), migratetype);
465 }
466 
467 static inline void clear_page_guard(struct zone *zone, struct page *page,
468                                 unsigned int order, int migratetype)
469 {
470         struct page_ext *page_ext;
471 
472         if (!debug_guardpage_enabled())
473                 return;
474 
475         page_ext = lookup_page_ext(page);
476         __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
477 
478         set_page_private(page, 0);
479         if (!is_migrate_isolate(migratetype))
480                 __mod_zone_freepage_state(zone, (1 << order), migratetype);
481 }
482 #else
483 struct page_ext_operations debug_guardpage_ops = { NULL, };
484 static inline void set_page_guard(struct zone *zone, struct page *page,
485                                 unsigned int order, int migratetype) {}
486 static inline void clear_page_guard(struct zone *zone, struct page *page,
487                                 unsigned int order, int migratetype) {}
488 #endif
489 
490 static inline void set_page_order(struct page *page, unsigned int order)
491 {
492         set_page_private(page, order);
493         __SetPageBuddy(page);
494 }
495 
496 static inline void rmv_page_order(struct page *page)
497 {
498         __ClearPageBuddy(page);
499         set_page_private(page, 0);
500 }
501 
502 /*
503  * This function checks whether a page is free && is the buddy
504  * we can do coalesce a page and its buddy if
505  * (a) the buddy is not in a hole &&
506  * (b) the buddy is in the buddy system &&
507  * (c) a page and its buddy have the same order &&
508  * (d) a page and its buddy are in the same zone.
509  *
510  * For recording whether a page is in the buddy system, we set ->_mapcount
511  * PAGE_BUDDY_MAPCOUNT_VALUE.
512  * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
513  * serialized by zone->lock.
514  *
515  * For recording page's order, we use page_private(page).
516  */
517 static inline int page_is_buddy(struct page *page, struct page *buddy,
518                                                         unsigned int order)
519 {
520         if (!pfn_valid_within(page_to_pfn(buddy)))
521                 return 0;
522 
523         if (page_is_guard(buddy) && page_order(buddy) == order) {
524                 if (page_zone_id(page) != page_zone_id(buddy))
525                         return 0;
526 
527                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
528 
529                 return 1;
530         }
531 
532         if (PageBuddy(buddy) && page_order(buddy) == order) {
533                 /*
534                  * zone check is done late to avoid uselessly
535                  * calculating zone/node ids for pages that could
536                  * never merge.
537                  */
538                 if (page_zone_id(page) != page_zone_id(buddy))
539                         return 0;
540 
541                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
542 
543                 return 1;
544         }
545         return 0;
546 }
547 
548 /*
549  * Freeing function for a buddy system allocator.
550  *
551  * The concept of a buddy system is to maintain direct-mapped table
552  * (containing bit values) for memory blocks of various "orders".
553  * The bottom level table contains the map for the smallest allocatable
554  * units of memory (here, pages), and each level above it describes
555  * pairs of units from the levels below, hence, "buddies".
556  * At a high level, all that happens here is marking the table entry
557  * at the bottom level available, and propagating the changes upward
558  * as necessary, plus some accounting needed to play nicely with other
559  * parts of the VM system.
560  * At each level, we keep a list of pages, which are heads of continuous
561  * free pages of length of (1 << order) and marked with _mapcount
562  * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
563  * field.
564  * So when we are allocating or freeing one, we can derive the state of the
565  * other.  That is, if we allocate a small block, and both were
566  * free, the remainder of the region must be split into blocks.
567  * If a block is freed, and its buddy is also free, then this
568  * triggers coalescing into a block of larger size.
569  *
570  * -- nyc
571  */
572 
573 static inline void __free_one_page(struct page *page,
574                 unsigned long pfn,
575                 struct zone *zone, unsigned int order,
576                 int migratetype)
577 {
578         unsigned long page_idx;
579         unsigned long combined_idx;
580         unsigned long uninitialized_var(buddy_idx);
581         struct page *buddy;
582         int max_order = MAX_ORDER;
583 
584         VM_BUG_ON(!zone_is_initialized(zone));
585         VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
586 
587         VM_BUG_ON(migratetype == -1);
588         if (is_migrate_isolate(migratetype)) {
589                 /*
590                  * We restrict max order of merging to prevent merge
591                  * between freepages on isolate pageblock and normal
592                  * pageblock. Without this, pageblock isolation
593                  * could cause incorrect freepage accounting.
594                  */
595                 max_order = min(MAX_ORDER, pageblock_order + 1);
596         } else {
597                 __mod_zone_freepage_state(zone, 1 << order, migratetype);
598         }
599 
600         page_idx = pfn & ((1 << max_order) - 1);
601 
602         VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
603         VM_BUG_ON_PAGE(bad_range(zone, page), page);
604 
605         while (order < max_order - 1) {
606                 buddy_idx = __find_buddy_index(page_idx, order);
607                 buddy = page + (buddy_idx - page_idx);
608                 if (!page_is_buddy(page, buddy, order))
609                         break;
610                 /*
611                  * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
612                  * merge with it and move up one order.
613                  */
614                 if (page_is_guard(buddy)) {
615                         clear_page_guard(zone, buddy, order, migratetype);
616                 } else {
617                         list_del(&buddy->lru);
618                         zone->free_area[order].nr_free--;
619                         rmv_page_order(buddy);
620                 }
621                 combined_idx = buddy_idx & page_idx;
622                 page = page + (combined_idx - page_idx);
623                 page_idx = combined_idx;
624                 order++;
625         }
626         set_page_order(page, order);
627 
628         /*
629          * If this is not the largest possible page, check if the buddy
630          * of the next-highest order is free. If it is, it's possible
631          * that pages are being freed that will coalesce soon. In case,
632          * that is happening, add the free page to the tail of the list
633          * so it's less likely to be used soon and more likely to be merged
634          * as a higher order page
635          */
636         if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
637                 struct page *higher_page, *higher_buddy;
638                 combined_idx = buddy_idx & page_idx;
639                 higher_page = page + (combined_idx - page_idx);
640                 buddy_idx = __find_buddy_index(combined_idx, order + 1);
641                 higher_buddy = higher_page + (buddy_idx - combined_idx);
642                 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
643                         list_add_tail(&page->lru,
644                                 &zone->free_area[order].free_list[migratetype]);
645                         goto out;
646                 }
647         }
648 
649         list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
650 out:
651         zone->free_area[order].nr_free++;
652 }
653 
654 static inline int free_pages_check(struct page *page)
655 {
656         const char *bad_reason = NULL;
657         unsigned long bad_flags = 0;
658 
659         if (unlikely(page_mapcount(page)))
660                 bad_reason = "nonzero mapcount";
661         if (unlikely(page->mapping != NULL))
662                 bad_reason = "non-NULL mapping";
663         if (unlikely(atomic_read(&page->_count) != 0))
664                 bad_reason = "nonzero _count";
665         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
666                 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
667                 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
668         }
669 #ifdef CONFIG_MEMCG
670         if (unlikely(page->mem_cgroup))
671                 bad_reason = "page still charged to cgroup";
672 #endif
673         if (unlikely(bad_reason)) {
674                 bad_page(page, bad_reason, bad_flags);
675                 return 1;
676         }
677         page_cpupid_reset_last(page);
678         if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
679                 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
680         return 0;
681 }
682 
683 /*
684  * Frees a number of pages from the PCP lists
685  * Assumes all pages on list are in same zone, and of same order.
686  * count is the number of pages to free.
687  *
688  * If the zone was previously in an "all pages pinned" state then look to
689  * see if this freeing clears that state.
690  *
691  * And clear the zone's pages_scanned counter, to hold off the "all pages are
692  * pinned" detection logic.
693  */
694 static void free_pcppages_bulk(struct zone *zone, int count,
695                                         struct per_cpu_pages *pcp)
696 {
697         int migratetype = 0;
698         int batch_free = 0;
699         int to_free = count;
700         unsigned long nr_scanned;
701 
702         spin_lock(&zone->lock);
703         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
704         if (nr_scanned)
705                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
706 
707         while (to_free) {
708                 struct page *page;
709                 struct list_head *list;
710 
711                 /*
712                  * Remove pages from lists in a round-robin fashion. A
713                  * batch_free count is maintained that is incremented when an
714                  * empty list is encountered.  This is so more pages are freed
715                  * off fuller lists instead of spinning excessively around empty
716                  * lists
717                  */
718                 do {
719                         batch_free++;
720                         if (++migratetype == MIGRATE_PCPTYPES)
721                                 migratetype = 0;
722                         list = &pcp->lists[migratetype];
723                 } while (list_empty(list));
724 
725                 /* This is the only non-empty list. Free them all. */
726                 if (batch_free == MIGRATE_PCPTYPES)
727                         batch_free = to_free;
728 
729                 do {
730                         int mt; /* migratetype of the to-be-freed page */
731 
732                         page = list_entry(list->prev, struct page, lru);
733                         /* must delete as __free_one_page list manipulates */
734                         list_del(&page->lru);
735                         mt = get_freepage_migratetype(page);
736                         if (unlikely(has_isolate_pageblock(zone)))
737                                 mt = get_pageblock_migratetype(page);
738 
739                         /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
740                         __free_one_page(page, page_to_pfn(page), zone, 0, mt);
741                         trace_mm_page_pcpu_drain(page, 0, mt);
742                 } while (--to_free && --batch_free && !list_empty(list));
743         }
744         spin_unlock(&zone->lock);
745 }
746 
747 static void free_one_page(struct zone *zone,
748                                 struct page *page, unsigned long pfn,
749                                 unsigned int order,
750                                 int migratetype)
751 {
752         unsigned long nr_scanned;
753         spin_lock(&zone->lock);
754         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
755         if (nr_scanned)
756                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
757 
758         if (unlikely(has_isolate_pageblock(zone) ||
759                 is_migrate_isolate(migratetype))) {
760                 migratetype = get_pfnblock_migratetype(page, pfn);
761         }
762         __free_one_page(page, pfn, zone, order, migratetype);
763         spin_unlock(&zone->lock);
764 }
765 
766 static int free_tail_pages_check(struct page *head_page, struct page *page)
767 {
768         if (!IS_ENABLED(CONFIG_DEBUG_VM))
769                 return 0;
770         if (unlikely(!PageTail(page))) {
771                 bad_page(page, "PageTail not set", 0);
772                 return 1;
773         }
774         if (unlikely(page->first_page != head_page)) {
775                 bad_page(page, "first_page not consistent", 0);
776                 return 1;
777         }
778         return 0;
779 }
780 
781 static bool free_pages_prepare(struct page *page, unsigned int order)
782 {
783         bool compound = PageCompound(page);
784         int i, bad = 0;
785 
786         VM_BUG_ON_PAGE(PageTail(page), page);
787         VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
788 
789         trace_mm_page_free(page, order);
790         kmemcheck_free_shadow(page, order);
791         kasan_free_pages(page, order);
792 
793         if (PageAnon(page))
794                 page->mapping = NULL;
795         bad += free_pages_check(page);
796         for (i = 1; i < (1 << order); i++) {
797                 if (compound)
798                         bad += free_tail_pages_check(page, page + i);
799                 bad += free_pages_check(page + i);
800         }
801         if (bad)
802                 return false;
803 
804         reset_page_owner(page, order);
805 
806         if (!PageHighMem(page)) {
807                 debug_check_no_locks_freed(page_address(page),
808                                            PAGE_SIZE << order);
809                 debug_check_no_obj_freed(page_address(page),
810                                            PAGE_SIZE << order);
811         }
812         arch_free_page(page, order);
813         kernel_map_pages(page, 1 << order, 0);
814 
815         return true;
816 }
817 
818 static void __free_pages_ok(struct page *page, unsigned int order)
819 {
820         unsigned long flags;
821         int migratetype;
822         unsigned long pfn = page_to_pfn(page);
823 
824         if (!free_pages_prepare(page, order))
825                 return;
826 
827         migratetype = get_pfnblock_migratetype(page, pfn);
828         local_irq_save(flags);
829         __count_vm_events(PGFREE, 1 << order);
830         set_freepage_migratetype(page, migratetype);
831         free_one_page(page_zone(page), page, pfn, order, migratetype);
832         local_irq_restore(flags);
833 }
834 
835 void __init __free_pages_bootmem(struct page *page, unsigned int order)
836 {
837         unsigned int nr_pages = 1 << order;
838         struct page *p = page;
839         unsigned int loop;
840 
841         prefetchw(p);
842         for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
843                 prefetchw(p + 1);
844                 __ClearPageReserved(p);
845                 set_page_count(p, 0);
846         }
847         __ClearPageReserved(p);
848         set_page_count(p, 0);
849 
850         page_zone(page)->managed_pages += nr_pages;
851         set_page_refcounted(page);
852         __free_pages(page, order);
853 }
854 
855 #ifdef CONFIG_CMA
856 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
857 void __init init_cma_reserved_pageblock(struct page *page)
858 {
859         unsigned i = pageblock_nr_pages;
860         struct page *p = page;
861 
862         do {
863                 __ClearPageReserved(p);
864                 set_page_count(p, 0);
865         } while (++p, --i);
866 
867         set_pageblock_migratetype(page, MIGRATE_CMA);
868 
869         if (pageblock_order >= MAX_ORDER) {
870                 i = pageblock_nr_pages;
871                 p = page;
872                 do {
873                         set_page_refcounted(p);
874                         __free_pages(p, MAX_ORDER - 1);
875                         p += MAX_ORDER_NR_PAGES;
876                 } while (i -= MAX_ORDER_NR_PAGES);
877         } else {
878                 set_page_refcounted(page);
879                 __free_pages(page, pageblock_order);
880         }
881 
882         adjust_managed_page_count(page, pageblock_nr_pages);
883 }
884 #endif
885 
886 /*
887  * The order of subdivision here is critical for the IO subsystem.
888  * Please do not alter this order without good reasons and regression
889  * testing. Specifically, as large blocks of memory are subdivided,
890  * the order in which smaller blocks are delivered depends on the order
891  * they're subdivided in this function. This is the primary factor
892  * influencing the order in which pages are delivered to the IO
893  * subsystem according to empirical testing, and this is also justified
894  * by considering the behavior of a buddy system containing a single
895  * large block of memory acted on by a series of small allocations.
896  * This behavior is a critical factor in sglist merging's success.
897  *
898  * -- nyc
899  */
900 static inline void expand(struct zone *zone, struct page *page,
901         int low, int high, struct free_area *area,
902         int migratetype)
903 {
904         unsigned long size = 1 << high;
905 
906         while (high > low) {
907                 area--;
908                 high--;
909                 size >>= 1;
910                 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
911 
912                 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
913                         debug_guardpage_enabled() &&
914                         high < debug_guardpage_minorder()) {
915                         /*
916                          * Mark as guard pages (or page), that will allow to
917                          * merge back to allocator when buddy will be freed.
918                          * Corresponding page table entries will not be touched,
919                          * pages will stay not present in virtual address space
920                          */
921                         set_page_guard(zone, &page[size], high, migratetype);
922                         continue;
923                 }
924                 list_add(&page[size].lru, &area->free_list[migratetype]);
925                 area->nr_free++;
926                 set_page_order(&page[size], high);
927         }
928 }
929 
930 /*
931  * This page is about to be returned from the page allocator
932  */
933 static inline int check_new_page(struct page *page)
934 {
935         const char *bad_reason = NULL;
936         unsigned long bad_flags = 0;
937 
938         if (unlikely(page_mapcount(page)))
939                 bad_reason = "nonzero mapcount";
940         if (unlikely(page->mapping != NULL))
941                 bad_reason = "non-NULL mapping";
942         if (unlikely(atomic_read(&page->_count) != 0))
943                 bad_reason = "nonzero _count";
944         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
945                 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
946                 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
947         }
948 #ifdef CONFIG_MEMCG
949         if (unlikely(page->mem_cgroup))
950                 bad_reason = "page still charged to cgroup";
951 #endif
952         if (unlikely(bad_reason)) {
953                 bad_page(page, bad_reason, bad_flags);
954                 return 1;
955         }
956         return 0;
957 }
958 
959 static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
960                                                                 int alloc_flags)
961 {
962         int i;
963 
964         for (i = 0; i < (1 << order); i++) {
965                 struct page *p = page + i;
966                 if (unlikely(check_new_page(p)))
967                         return 1;
968         }
969 
970         set_page_private(page, 0);
971         set_page_refcounted(page);
972 
973         arch_alloc_page(page, order);
974         kernel_map_pages(page, 1 << order, 1);
975         kasan_alloc_pages(page, order);
976 
977         if (gfp_flags & __GFP_ZERO)
978                 prep_zero_page(page, order, gfp_flags);
979 
980         if (order && (gfp_flags & __GFP_COMP))
981                 prep_compound_page(page, order);
982 
983         set_page_owner(page, order, gfp_flags);
984 
985         /*
986          * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was necessary to
987          * allocate the page. The expectation is that the caller is taking
988          * steps that will free more memory. The caller should avoid the page
989          * being used for !PFMEMALLOC purposes.
990          */
991         page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
992 
993         return 0;
994 }
995 
996 /*
997  * Go through the free lists for the given migratetype and remove
998  * the smallest available page from the freelists
999  */
1000 static inline
1001 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1002                                                 int migratetype)
1003 {
1004         unsigned int current_order;
1005         struct free_area *area;
1006         struct page *page;
1007 
1008         /* Find a page of the appropriate size in the preferred list */
1009         for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1010                 area = &(zone->free_area[current_order]);
1011                 if (list_empty(&area->free_list[migratetype]))
1012                         continue;
1013 
1014                 page = list_entry(area->free_list[migratetype].next,
1015                                                         struct page, lru);
1016                 list_del(&page->lru);
1017                 rmv_page_order(page);
1018                 area->nr_free--;
1019                 expand(zone, page, order, current_order, area, migratetype);
1020                 set_freepage_migratetype(page, migratetype);
1021                 return page;
1022         }
1023 
1024         return NULL;
1025 }
1026 
1027 
1028 /*
1029  * This array describes the order lists are fallen back to when
1030  * the free lists for the desirable migrate type are depleted
1031  */
1032 static int fallbacks[MIGRATE_TYPES][4] = {
1033         [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,     MIGRATE_RESERVE },
1034         [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,     MIGRATE_RESERVE },
1035 #ifdef CONFIG_CMA
1036         [MIGRATE_MOVABLE]     = { MIGRATE_CMA,         MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
1037         [MIGRATE_CMA]         = { MIGRATE_RESERVE }, /* Never used */
1038 #else
1039         [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,   MIGRATE_RESERVE },
1040 #endif
1041         [MIGRATE_RESERVE]     = { MIGRATE_RESERVE }, /* Never used */
1042 #ifdef CONFIG_MEMORY_ISOLATION
1043         [MIGRATE_ISOLATE]     = { MIGRATE_RESERVE }, /* Never used */
1044 #endif
1045 };
1046 
1047 /*
1048  * Move the free pages in a range to the free lists of the requested type.
1049  * Note that start_page and end_pages are not aligned on a pageblock
1050  * boundary. If alignment is required, use move_freepages_block()
1051  */
1052 int move_freepages(struct zone *zone,
1053                           struct page *start_page, struct page *end_page,
1054                           int migratetype)
1055 {
1056         struct page *page;
1057         unsigned long order;
1058         int pages_moved = 0;
1059 
1060 #ifndef CONFIG_HOLES_IN_ZONE
1061         /*
1062          * page_zone is not safe to call in this context when
1063          * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1064          * anyway as we check zone boundaries in move_freepages_block().
1065          * Remove at a later date when no bug reports exist related to
1066          * grouping pages by mobility
1067          */
1068         VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1069 #endif
1070 
1071         for (page = start_page; page <= end_page;) {
1072                 /* Make sure we are not inadvertently changing nodes */
1073                 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1074 
1075                 if (!pfn_valid_within(page_to_pfn(page))) {
1076                         page++;
1077                         continue;
1078                 }
1079 
1080                 if (!PageBuddy(page)) {
1081                         page++;
1082                         continue;
1083                 }
1084 
1085                 order = page_order(page);
1086                 list_move(&page->lru,
1087                           &zone->free_area[order].free_list[migratetype]);
1088                 set_freepage_migratetype(page, migratetype);
1089                 page += 1 << order;
1090                 pages_moved += 1 << order;
1091         }
1092 
1093         return pages_moved;
1094 }
1095 
1096 int move_freepages_block(struct zone *zone, struct page *page,
1097                                 int migratetype)
1098 {
1099         unsigned long start_pfn, end_pfn;
1100         struct page *start_page, *end_page;
1101 
1102         start_pfn = page_to_pfn(page);
1103         start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1104         start_page = pfn_to_page(start_pfn);
1105         end_page = start_page + pageblock_nr_pages - 1;
1106         end_pfn = start_pfn + pageblock_nr_pages - 1;
1107 
1108         /* Do not cross zone boundaries */
1109         if (!zone_spans_pfn(zone, start_pfn))
1110                 start_page = page;
1111         if (!zone_spans_pfn(zone, end_pfn))
1112                 return 0;
1113 
1114         return move_freepages(zone, start_page, end_page, migratetype);
1115 }
1116 
1117 static void change_pageblock_range(struct page *pageblock_page,
1118                                         int start_order, int migratetype)
1119 {
1120         int nr_pageblocks = 1 << (start_order - pageblock_order);
1121 
1122         while (nr_pageblocks--) {
1123                 set_pageblock_migratetype(pageblock_page, migratetype);
1124                 pageblock_page += pageblock_nr_pages;
1125         }
1126 }
1127 
1128 /*
1129  * When we are falling back to another migratetype during allocation, try to
1130  * steal extra free pages from the same pageblocks to satisfy further
1131  * allocations, instead of polluting multiple pageblocks.
1132  *
1133  * If we are stealing a relatively large buddy page, it is likely there will
1134  * be more free pages in the pageblock, so try to steal them all. For
1135  * reclaimable and unmovable allocations, we steal regardless of page size,
1136  * as fragmentation caused by those allocations polluting movable pageblocks
1137  * is worse than movable allocations stealing from unmovable and reclaimable
1138  * pageblocks.
1139  *
1140  * If we claim more than half of the pageblock, change pageblock's migratetype
1141  * as well.
1142  */
1143 static void try_to_steal_freepages(struct zone *zone, struct page *page,
1144                                   int start_type, int fallback_type)
1145 {
1146         int current_order = page_order(page);
1147 
1148         /* Take ownership for orders >= pageblock_order */
1149         if (current_order >= pageblock_order) {
1150                 change_pageblock_range(page, current_order, start_type);
1151                 return;
1152         }
1153 
1154         if (current_order >= pageblock_order / 2 ||
1155             start_type == MIGRATE_RECLAIMABLE ||
1156             start_type == MIGRATE_UNMOVABLE ||
1157             page_group_by_mobility_disabled) {
1158                 int pages;
1159 
1160                 pages = move_freepages_block(zone, page, start_type);
1161 
1162                 /* Claim the whole block if over half of it is free */
1163                 if (pages >= (1 << (pageblock_order-1)) ||
1164                                 page_group_by_mobility_disabled)
1165                         set_pageblock_migratetype(page, start_type);
1166         }
1167 }
1168 
1169 /* Remove an element from the buddy allocator from the fallback list */
1170 static inline struct page *
1171 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
1172 {
1173         struct free_area *area;
1174         unsigned int current_order;
1175         struct page *page;
1176 
1177         /* Find the largest possible block of pages in the other list */
1178         for (current_order = MAX_ORDER-1;
1179                                 current_order >= order && current_order <= MAX_ORDER-1;
1180                                 --current_order) {
1181                 int i;
1182                 for (i = 0;; i++) {
1183                         int migratetype = fallbacks[start_migratetype][i];
1184                         int buddy_type = start_migratetype;
1185 
1186                         /* MIGRATE_RESERVE handled later if necessary */
1187                         if (migratetype == MIGRATE_RESERVE)
1188                                 break;
1189 
1190                         area = &(zone->free_area[current_order]);
1191                         if (list_empty(&area->free_list[migratetype]))
1192                                 continue;
1193 
1194                         page = list_entry(area->free_list[migratetype].next,
1195                                         struct page, lru);
1196                         area->nr_free--;
1197 
1198                         if (!is_migrate_cma(migratetype)) {
1199                                 try_to_steal_freepages(zone, page,
1200                                                         start_migratetype,
1201                                                         migratetype);
1202                         } else {
1203                                 /*
1204                                  * When borrowing from MIGRATE_CMA, we need to
1205                                  * release the excess buddy pages to CMA
1206                                  * itself, and we do not try to steal extra
1207                                  * free pages.
1208                                  */
1209                                 buddy_type = migratetype;
1210                         }
1211 
1212                         /* Remove the page from the freelists */
1213                         list_del(&page->lru);
1214                         rmv_page_order(page);
1215 
1216                         expand(zone, page, order, current_order, area,
1217                                         buddy_type);
1218 
1219                         /*
1220                          * The freepage_migratetype may differ from pageblock's
1221                          * migratetype depending on the decisions in
1222                          * try_to_steal_freepages(). This is OK as long as it
1223                          * does not differ for MIGRATE_CMA pageblocks. For CMA
1224                          * we need to make sure unallocated pages flushed from
1225                          * pcp lists are returned to the correct freelist.
1226                          */
1227                         set_freepage_migratetype(page, buddy_type);
1228 
1229                         trace_mm_page_alloc_extfrag(page, order, current_order,
1230                                 start_migratetype, migratetype);
1231 
1232                         return page;
1233                 }
1234         }
1235 
1236         return NULL;
1237 }
1238 
1239 /*
1240  * Do the hard work of removing an element from the buddy allocator.
1241  * Call me with the zone->lock already held.
1242  */
1243 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1244                                                 int migratetype)
1245 {
1246         struct page *page;
1247 
1248 retry_reserve:
1249         page = __rmqueue_smallest(zone, order, migratetype);
1250 
1251         if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1252                 page = __rmqueue_fallback(zone, order, migratetype);
1253 
1254                 /*
1255                  * Use MIGRATE_RESERVE rather than fail an allocation. goto
1256                  * is used because __rmqueue_smallest is an inline function
1257                  * and we want just one call site
1258                  */
1259                 if (!page) {
1260                         migratetype = MIGRATE_RESERVE;
1261                         goto retry_reserve;
1262                 }
1263         }
1264 
1265         trace_mm_page_alloc_zone_locked(page, order, migratetype);
1266         return page;
1267 }
1268 
1269 /*
1270  * Obtain a specified number of elements from the buddy allocator, all under
1271  * a single hold of the lock, for efficiency.  Add them to the supplied list.
1272  * Returns the number of new pages which were placed at *list.
1273  */
1274 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1275                         unsigned long count, struct list_head *list,
1276                         int migratetype, bool cold)
1277 {
1278         int i;
1279 
1280         spin_lock(&zone->lock);
1281         for (i = 0; i < count; ++i) {
1282                 struct page *page = __rmqueue(zone, order, migratetype);
1283                 if (unlikely(page == NULL))
1284                         break;
1285 
1286                 /*
1287                  * Split buddy pages returned by expand() are received here
1288                  * in physical page order. The page is added to the callers and
1289                  * list and the list head then moves forward. From the callers
1290                  * perspective, the linked list is ordered by page number in
1291                  * some conditions. This is useful for IO devices that can
1292                  * merge IO requests if the physical pages are ordered
1293                  * properly.
1294                  */
1295                 if (likely(!cold))
1296                         list_add(&page->lru, list);
1297                 else
1298                         list_add_tail(&page->lru, list);
1299                 list = &page->lru;
1300                 if (is_migrate_cma(get_freepage_migratetype(page)))
1301                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1302                                               -(1 << order));
1303         }
1304         __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1305         spin_unlock(&zone->lock);
1306         return i;
1307 }
1308 
1309 #ifdef CONFIG_NUMA
1310 /*
1311  * Called from the vmstat counter updater to drain pagesets of this
1312  * currently executing processor on remote nodes after they have
1313  * expired.
1314  *
1315  * Note that this function must be called with the thread pinned to
1316  * a single processor.
1317  */
1318 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1319 {
1320         unsigned long flags;
1321         int to_drain, batch;
1322 
1323         local_irq_save(flags);
1324         batch = ACCESS_ONCE(pcp->batch);
1325         to_drain = min(pcp->count, batch);
1326         if (to_drain > 0) {
1327                 free_pcppages_bulk(zone, to_drain, pcp);
1328                 pcp->count -= to_drain;
1329         }
1330         local_irq_restore(flags);
1331 }
1332 #endif
1333 
1334 /*
1335  * Drain pcplists of the indicated processor and zone.
1336  *
1337  * The processor must either be the current processor and the
1338  * thread pinned to the current processor or a processor that
1339  * is not online.
1340  */
1341 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1342 {
1343         unsigned long flags;
1344         struct per_cpu_pageset *pset;
1345         struct per_cpu_pages *pcp;
1346 
1347         local_irq_save(flags);
1348         pset = per_cpu_ptr(zone->pageset, cpu);
1349 
1350         pcp = &pset->pcp;
1351         if (pcp->count) {
1352                 free_pcppages_bulk(zone, pcp->count, pcp);
1353                 pcp->count = 0;
1354         }
1355         local_irq_restore(flags);
1356 }
1357 
1358 /*
1359  * Drain pcplists of all zones on the indicated processor.
1360  *
1361  * The processor must either be the current processor and the
1362  * thread pinned to the current processor or a processor that
1363  * is not online.
1364  */
1365 static void drain_pages(unsigned int cpu)
1366 {
1367         struct zone *zone;
1368 
1369         for_each_populated_zone(zone) {
1370                 drain_pages_zone(cpu, zone);
1371         }
1372 }
1373 
1374 /*
1375  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1376  *
1377  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1378  * the single zone's pages.
1379  */
1380 void drain_local_pages(struct zone *zone)
1381 {
1382         int cpu = smp_processor_id();
1383 
1384         if (zone)
1385                 drain_pages_zone(cpu, zone);
1386         else
1387                 drain_pages(cpu);
1388 }
1389 
1390 /*
1391  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1392  *
1393  * When zone parameter is non-NULL, spill just the single zone's pages.
1394  *
1395  * Note that this code is protected against sending an IPI to an offline
1396  * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1397  * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1398  * nothing keeps CPUs from showing up after we populated the cpumask and
1399  * before the call to on_each_cpu_mask().
1400  */
1401 void drain_all_pages(struct zone *zone)
1402 {
1403         int cpu;
1404 
1405         /*
1406          * Allocate in the BSS so we wont require allocation in
1407          * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1408          */
1409         static cpumask_t cpus_with_pcps;
1410 
1411         /*
1412          * We don't care about racing with CPU hotplug event
1413          * as offline notification will cause the notified
1414          * cpu to drain that CPU pcps and on_each_cpu_mask
1415          * disables preemption as part of its processing
1416          */
1417         for_each_online_cpu(cpu) {
1418                 struct per_cpu_pageset *pcp;
1419                 struct zone *z;
1420                 bool has_pcps = false;
1421 
1422                 if (zone) {
1423                         pcp = per_cpu_ptr(zone->pageset, cpu);
1424                         if (pcp->pcp.count)
1425                                 has_pcps = true;
1426                 } else {
1427                         for_each_populated_zone(z) {
1428                                 pcp = per_cpu_ptr(z->pageset, cpu);
1429                                 if (pcp->pcp.count) {
1430                                         has_pcps = true;
1431                                         break;
1432                                 }
1433                         }
1434                 }
1435 
1436                 if (has_pcps)
1437                         cpumask_set_cpu(cpu, &cpus_with_pcps);
1438                 else
1439                         cpumask_clear_cpu(cpu, &cpus_with_pcps);
1440         }
1441         on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1442                                                                 zone, 1);
1443 }
1444 
1445 #ifdef CONFIG_HIBERNATION
1446 
1447 void mark_free_pages(struct zone *zone)
1448 {
1449         unsigned long pfn, max_zone_pfn;
1450         unsigned long flags;
1451         unsigned int order, t;
1452         struct list_head *curr;
1453 
1454         if (zone_is_empty(zone))
1455                 return;
1456 
1457         spin_lock_irqsave(&zone->lock, flags);
1458 
1459         max_zone_pfn = zone_end_pfn(zone);
1460         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1461                 if (pfn_valid(pfn)) {
1462                         struct page *page = pfn_to_page(pfn);
1463 
1464                         if (!swsusp_page_is_forbidden(page))
1465                                 swsusp_unset_page_free(page);
1466                 }
1467 
1468         for_each_migratetype_order(order, t) {
1469                 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1470                         unsigned long i;
1471 
1472                         pfn = page_to_pfn(list_entry(curr, struct page, lru));
1473                         for (i = 0; i < (1UL << order); i++)
1474                                 swsusp_set_page_free(pfn_to_page(pfn + i));
1475                 }
1476         }
1477         spin_unlock_irqrestore(&zone->lock, flags);
1478 }
1479 #endif /* CONFIG_PM */
1480 
1481 /*
1482  * Free a 0-order page
1483  * cold == true ? free a cold page : free a hot page
1484  */
1485 void free_hot_cold_page(struct page *page, bool cold)
1486 {
1487         struct zone *zone = page_zone(page);
1488         struct per_cpu_pages *pcp;
1489         unsigned long flags;
1490         unsigned long pfn = page_to_pfn(page);
1491         int migratetype;
1492 
1493         if (!free_pages_prepare(page, 0))
1494                 return;
1495 
1496         migratetype = get_pfnblock_migratetype(page, pfn);
1497         set_freepage_migratetype(page, migratetype);
1498         local_irq_save(flags);
1499         __count_vm_event(PGFREE);
1500 
1501         /*
1502          * We only track unmovable, reclaimable and movable on pcp lists.
1503          * Free ISOLATE pages back to the allocator because they are being
1504          * offlined but treat RESERVE as movable pages so we can get those
1505          * areas back if necessary. Otherwise, we may have to free
1506          * excessively into the page allocator
1507          */
1508         if (migratetype >= MIGRATE_PCPTYPES) {
1509                 if (unlikely(is_migrate_isolate(migratetype))) {
1510                         free_one_page(zone, page, pfn, 0, migratetype);
1511                         goto out;
1512                 }
1513                 migratetype = MIGRATE_MOVABLE;
1514         }
1515 
1516         pcp = &this_cpu_ptr(zone->pageset)->pcp;
1517         if (!cold)
1518                 list_add(&page->lru, &pcp->lists[migratetype]);
1519         else
1520                 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1521         pcp->count++;
1522         if (pcp->count >= pcp->high) {
1523                 unsigned long batch = ACCESS_ONCE(pcp->batch);
1524                 free_pcppages_bulk(zone, batch, pcp);
1525                 pcp->count -= batch;
1526         }
1527 
1528 out:
1529         local_irq_restore(flags);
1530 }
1531 
1532 /*
1533  * Free a list of 0-order pages
1534  */
1535 void free_hot_cold_page_list(struct list_head *list, bool cold)
1536 {
1537         struct page *page, *next;
1538 
1539         list_for_each_entry_safe(page, next, list, lru) {
1540                 trace_mm_page_free_batched(page, cold);
1541                 free_hot_cold_page(page, cold);
1542         }
1543 }
1544 
1545 /*
1546  * split_page takes a non-compound higher-order page, and splits it into
1547  * n (1<<order) sub-pages: page[0..n]
1548  * Each sub-page must be freed individually.
1549  *
1550  * Note: this is probably too low level an operation for use in drivers.
1551  * Please consult with lkml before using this in your driver.
1552  */
1553 void split_page(struct page *page, unsigned int order)
1554 {
1555         int i;
1556 
1557         VM_BUG_ON_PAGE(PageCompound(page), page);
1558         VM_BUG_ON_PAGE(!page_count(page), page);
1559 
1560 #ifdef CONFIG_KMEMCHECK
1561         /*
1562          * Split shadow pages too, because free(page[0]) would
1563          * otherwise free the whole shadow.
1564          */
1565         if (kmemcheck_page_is_tracked(page))
1566                 split_page(virt_to_page(page[0].shadow), order);
1567 #endif
1568 
1569         set_page_owner(page, 0, 0);
1570         for (i = 1; i < (1 << order); i++) {
1571                 set_page_refcounted(page + i);
1572                 set_page_owner(page + i, 0, 0);
1573         }
1574 }
1575 EXPORT_SYMBOL_GPL(split_page);
1576 
1577 int __isolate_free_page(struct page *page, unsigned int order)
1578 {
1579         unsigned long watermark;
1580         struct zone *zone;
1581         int mt;
1582 
1583         BUG_ON(!PageBuddy(page));
1584 
1585         zone = page_zone(page);
1586         mt = get_pageblock_migratetype(page);
1587 
1588         if (!is_migrate_isolate(mt)) {
1589                 /* Obey watermarks as if the page was being allocated */
1590                 watermark = low_wmark_pages(zone) + (1 << order);
1591                 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1592                         return 0;
1593 
1594                 __mod_zone_freepage_state(zone, -(1UL << order), mt);
1595         }
1596 
1597         /* Remove page from free list */
1598         list_del(&page->lru);
1599         zone->free_area[order].nr_free--;
1600         rmv_page_order(page);
1601 
1602         /* Set the pageblock if the isolated page is at least a pageblock */
1603         if (order >= pageblock_order - 1) {
1604                 struct page *endpage = page + (1 << order) - 1;
1605                 for (; page < endpage; page += pageblock_nr_pages) {
1606                         int mt = get_pageblock_migratetype(page);
1607                         if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
1608                                 set_pageblock_migratetype(page,
1609                                                           MIGRATE_MOVABLE);
1610                 }
1611         }
1612 
1613         set_page_owner(page, order, 0);
1614         return 1UL << order;
1615 }
1616 
1617 /*
1618  * Similar to split_page except the page is already free. As this is only
1619  * being used for migration, the migratetype of the block also changes.
1620  * As this is called with interrupts disabled, the caller is responsible
1621  * for calling arch_alloc_page() and kernel_map_page() after interrupts
1622  * are enabled.
1623  *
1624  * Note: this is probably too low level an operation for use in drivers.
1625  * Please consult with lkml before using this in your driver.
1626  */
1627 int split_free_page(struct page *page)
1628 {
1629         unsigned int order;
1630         int nr_pages;
1631 
1632         order = page_order(page);
1633 
1634         nr_pages = __isolate_free_page(page, order);
1635         if (!nr_pages)
1636                 return 0;
1637 
1638         /* Split into individual pages */
1639         set_page_refcounted(page);
1640         split_page(page, order);
1641         return nr_pages;
1642 }
1643 
1644 /*
1645  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
1646  */
1647 static inline
1648 struct page *buffered_rmqueue(struct zone *preferred_zone,
1649                         struct zone *zone, unsigned int order,
1650                         gfp_t gfp_flags, int migratetype)
1651 {
1652         unsigned long flags;
1653         struct page *page;
1654         bool cold = ((gfp_flags & __GFP_COLD) != 0);
1655 
1656         if (likely(order == 0)) {
1657                 struct per_cpu_pages *pcp;
1658                 struct list_head *list;
1659 
1660                 local_irq_save(flags);
1661                 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1662                 list = &pcp->lists[migratetype];
1663                 if (list_empty(list)) {
1664                         pcp->count += rmqueue_bulk(zone, 0,
1665                                         pcp->batch, list,
1666                                         migratetype, cold);
1667                         if (unlikely(list_empty(list)))
1668                                 goto failed;
1669                 }
1670 
1671                 if (cold)
1672                         page = list_entry(list->prev, struct page, lru);
1673                 else
1674                         page = list_entry(list->next, struct page, lru);
1675 
1676                 list_del(&page->lru);
1677                 pcp->count--;
1678         } else {
1679                 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1680                         /*
1681                          * __GFP_NOFAIL is not to be used in new code.
1682                          *
1683                          * All __GFP_NOFAIL callers should be fixed so that they
1684                          * properly detect and handle allocation failures.
1685                          *
1686                          * We most definitely don't want callers attempting to
1687                          * allocate greater than order-1 page units with
1688                          * __GFP_NOFAIL.
1689                          */
1690                         WARN_ON_ONCE(order > 1);
1691                 }
1692                 spin_lock_irqsave(&zone->lock, flags);
1693                 page = __rmqueue(zone, order, migratetype);
1694                 spin_unlock(&zone->lock);
1695                 if (!page)
1696                         goto failed;
1697                 __mod_zone_freepage_state(zone, -(1 << order),
1698                                           get_freepage_migratetype(page));
1699         }
1700 
1701         __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
1702         if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
1703             !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
1704                 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
1705 
1706         __count_zone_vm_events(PGALLOC, zone, 1 << order);
1707         zone_statistics(preferred_zone, zone, gfp_flags);
1708         local_irq_restore(flags);
1709 
1710         VM_BUG_ON_PAGE(bad_range(zone, page), page);
1711         return page;
1712 
1713 failed:
1714         local_irq_restore(flags);
1715         return NULL;
1716 }
1717 
1718 #ifdef CONFIG_FAIL_PAGE_ALLOC
1719 
1720 static struct {
1721         struct fault_attr attr;
1722 
1723         u32 ignore_gfp_highmem;
1724         u32 ignore_gfp_wait;
1725         u32 min_order;
1726 } fail_page_alloc = {
1727         .attr = FAULT_ATTR_INITIALIZER,
1728         .ignore_gfp_wait = 1,
1729         .ignore_gfp_highmem = 1,
1730         .min_order = 1,
1731 };
1732 
1733 static int __init setup_fail_page_alloc(char *str)
1734 {
1735         return setup_fault_attr(&fail_page_alloc.attr, str);
1736 }
1737 __setup("fail_page_alloc=", setup_fail_page_alloc);
1738 
1739 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1740 {
1741         if (order < fail_page_alloc.min_order)
1742                 return false;
1743         if (gfp_mask & __GFP_NOFAIL)
1744                 return false;
1745         if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1746                 return false;
1747         if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1748                 return false;
1749 
1750         return should_fail(&fail_page_alloc.attr, 1 << order);
1751 }
1752 
1753 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1754 
1755 static int __init fail_page_alloc_debugfs(void)
1756 {
1757         umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1758         struct dentry *dir;
1759 
1760         dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1761                                         &fail_page_alloc.attr);
1762         if (IS_ERR(dir))
1763                 return PTR_ERR(dir);
1764 
1765         if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1766                                 &fail_page_alloc.ignore_gfp_wait))
1767                 goto fail;
1768         if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1769                                 &fail_page_alloc.ignore_gfp_highmem))
1770                 goto fail;
1771         if (!debugfs_create_u32("min-order", mode, dir,
1772                                 &fail_page_alloc.min_order))
1773                 goto fail;
1774 
1775         return 0;
1776 fail:
1777         debugfs_remove_recursive(dir);
1778 
1779         return -ENOMEM;
1780 }
1781 
1782 late_initcall(fail_page_alloc_debugfs);
1783 
1784 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1785 
1786 #else /* CONFIG_FAIL_PAGE_ALLOC */
1787 
1788 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1789 {
1790         return false;
1791 }
1792 
1793 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1794 
1795 /*
1796  * Return true if free pages are above 'mark'. This takes into account the order
1797  * of the allocation.
1798  */
1799 static bool __zone_watermark_ok(struct zone *z, unsigned int order,
1800                         unsigned long mark, int classzone_idx, int alloc_flags,
1801                         long free_pages)
1802 {
1803         /* free_pages may go negative - that's OK */
1804         long min = mark;
1805         int o;
1806         long free_cma = 0;
1807 
1808         free_pages -= (1 << order) - 1;
1809         if (alloc_flags & ALLOC_HIGH)
1810                 min -= min / 2;
1811         if (alloc_flags & ALLOC_HARDER)
1812                 min -= min / 4;
1813 #ifdef CONFIG_CMA
1814         /* If allocation can't use CMA areas don't use free CMA pages */
1815         if (!(alloc_flags & ALLOC_CMA))
1816                 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
1817 #endif
1818 
1819         if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
1820                 return false;
1821         for (o = 0; o < order; o++) {
1822                 /* At the next order, this order's pages become unavailable */
1823                 free_pages -= z->free_area[o].nr_free << o;
1824 
1825                 /* Require fewer higher order pages to be free */
1826                 min >>= 1;
1827 
1828                 if (free_pages <= min)
1829                         return false;
1830         }
1831         return true;
1832 }
1833 
1834 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1835                       int classzone_idx, int alloc_flags)
1836 {
1837         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1838                                         zone_page_state(z, NR_FREE_PAGES));
1839 }
1840 
1841 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1842                         unsigned long mark, int classzone_idx, int alloc_flags)
1843 {
1844         long free_pages = zone_page_state(z, NR_FREE_PAGES);
1845 
1846         if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1847                 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1848 
1849         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1850                                                                 free_pages);
1851 }
1852 
1853 #ifdef CONFIG_NUMA
1854 /*
1855  * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
1856  * skip over zones that are not allowed by the cpuset, or that have
1857  * been recently (in last second) found to be nearly full.  See further
1858  * comments in mmzone.h.  Reduces cache footprint of zonelist scans
1859  * that have to skip over a lot of full or unallowed zones.
1860  *
1861  * If the zonelist cache is present in the passed zonelist, then
1862  * returns a pointer to the allowed node mask (either the current
1863  * tasks mems_allowed, or node_states[N_MEMORY].)
1864  *
1865  * If the zonelist cache is not available for this zonelist, does
1866  * nothing and returns NULL.
1867  *
1868  * If the fullzones BITMAP in the zonelist cache is stale (more than
1869  * a second since last zap'd) then we zap it out (clear its bits.)
1870  *
1871  * We hold off even calling zlc_setup, until after we've checked the
1872  * first zone in the zonelist, on the theory that most allocations will
1873  * be satisfied from that first zone, so best to examine that zone as
1874  * quickly as we can.
1875  */
1876 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1877 {
1878         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1879         nodemask_t *allowednodes;       /* zonelist_cache approximation */
1880 
1881         zlc = zonelist->zlcache_ptr;
1882         if (!zlc)
1883                 return NULL;
1884 
1885         if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1886                 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1887                 zlc->last_full_zap = jiffies;
1888         }
1889 
1890         allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1891                                         &cpuset_current_mems_allowed :
1892                                         &node_states[N_MEMORY];
1893         return allowednodes;
1894 }
1895 
1896 /*
1897  * Given 'z' scanning a zonelist, run a couple of quick checks to see
1898  * if it is worth looking at further for free memory:
1899  *  1) Check that the zone isn't thought to be full (doesn't have its
1900  *     bit set in the zonelist_cache fullzones BITMAP).
1901  *  2) Check that the zones node (obtained from the zonelist_cache
1902  *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1903  * Return true (non-zero) if zone is worth looking at further, or
1904  * else return false (zero) if it is not.
1905  *
1906  * This check -ignores- the distinction between various watermarks,
1907  * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
1908  * found to be full for any variation of these watermarks, it will
1909  * be considered full for up to one second by all requests, unless
1910  * we are so low on memory on all allowed nodes that we are forced
1911  * into the second scan of the zonelist.
1912  *
1913  * In the second scan we ignore this zonelist cache and exactly
1914  * apply the watermarks to all zones, even it is slower to do so.
1915  * We are low on memory in the second scan, and should leave no stone
1916  * unturned looking for a free page.
1917  */
1918 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1919                                                 nodemask_t *allowednodes)
1920 {
1921         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1922         int i;                          /* index of *z in zonelist zones */
1923         int n;                          /* node that zone *z is on */
1924 
1925         zlc = zonelist->zlcache_ptr;
1926         if (!zlc)
1927                 return 1;
1928 
1929         i = z - zonelist->_zonerefs;
1930         n = zlc->z_to_n[i];
1931 
1932         /* This zone is worth trying if it is allowed but not full */
1933         return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1934 }
1935 
1936 /*
1937  * Given 'z' scanning a zonelist, set the corresponding bit in
1938  * zlc->fullzones, so that subsequent attempts to allocate a page
1939  * from that zone don't waste time re-examining it.
1940  */
1941 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1942 {
1943         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1944         int i;                          /* index of *z in zonelist zones */
1945 
1946         zlc = zonelist->zlcache_ptr;
1947         if (!zlc)
1948                 return;
1949 
1950         i = z - zonelist->_zonerefs;
1951 
1952         set_bit(i, zlc->fullzones);
1953 }
1954 
1955 /*
1956  * clear all zones full, called after direct reclaim makes progress so that
1957  * a zone that was recently full is not skipped over for up to a second
1958  */
1959 static void zlc_clear_zones_full(struct zonelist *zonelist)
1960 {
1961         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1962 
1963         zlc = zonelist->zlcache_ptr;
1964         if (!zlc)
1965                 return;
1966 
1967         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1968 }
1969 
1970 static bool zone_local(struct zone *local_zone, struct zone *zone)
1971 {
1972         return local_zone->node == zone->node;
1973 }
1974 
1975 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1976 {
1977         return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
1978                                 RECLAIM_DISTANCE;
1979 }
1980 
1981 #else   /* CONFIG_NUMA */
1982 
1983 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1984 {
1985         return NULL;
1986 }
1987 
1988 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1989                                 nodemask_t *allowednodes)
1990 {
1991         return 1;
1992 }
1993 
1994 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1995 {
1996 }
1997 
1998 static void zlc_clear_zones_full(struct zonelist *zonelist)
1999 {
2000 }
2001 
2002 static bool zone_local(struct zone *local_zone, struct zone *zone)
2003 {
2004         return true;
2005 }
2006 
2007 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2008 {
2009         return true;
2010 }
2011 
2012 #endif  /* CONFIG_NUMA */
2013 
2014 static void reset_alloc_batches(struct zone *preferred_zone)
2015 {
2016         struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2017 
2018         do {
2019                 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2020                         high_wmark_pages(zone) - low_wmark_pages(zone) -
2021                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
2022                 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2023         } while (zone++ != preferred_zone);
2024 }
2025 
2026 /*
2027  * get_page_from_freelist goes through the zonelist trying to allocate
2028  * a page.
2029  */
2030 static struct page *
2031 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2032                                                 const struct alloc_context *ac)
2033 {
2034         struct zonelist *zonelist = ac->zonelist;
2035         struct zoneref *z;
2036         struct page *page = NULL;
2037         struct zone *zone;
2038         nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
2039         int zlc_active = 0;             /* set if using zonelist_cache */
2040         int did_zlc_setup = 0;          /* just call zlc_setup() one time */
2041         bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
2042                                 (gfp_mask & __GFP_WRITE);
2043         int nr_fair_skipped = 0;
2044         bool zonelist_rescan;
2045 
2046 zonelist_scan:
2047         zonelist_rescan = false;
2048 
2049         /*
2050          * Scan zonelist, looking for a zone with enough free.
2051          * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2052          */
2053         for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2054                                                                 ac->nodemask) {
2055                 unsigned long mark;
2056 
2057                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2058                         !zlc_zone_worth_trying(zonelist, z, allowednodes))
2059                                 continue;
2060                 if (cpusets_enabled() &&
2061                         (alloc_flags & ALLOC_CPUSET) &&
2062                         !cpuset_zone_allowed(zone, gfp_mask))
2063                                 continue;
2064                 /*
2065                  * Distribute pages in proportion to the individual
2066                  * zone size to ensure fair page aging.  The zone a
2067                  * page was allocated in should have no effect on the
2068                  * time the page has in memory before being reclaimed.
2069                  */
2070                 if (alloc_flags & ALLOC_FAIR) {
2071                         if (!zone_local(ac->preferred_zone, zone))
2072                                 break;
2073                         if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
2074                                 nr_fair_skipped++;
2075                                 continue;
2076                         }
2077                 }
2078                 /*
2079                  * When allocating a page cache page for writing, we
2080                  * want to get it from a zone that is within its dirty
2081                  * limit, such that no single zone holds more than its
2082                  * proportional share of globally allowed dirty pages.
2083                  * The dirty limits take into account the zone's
2084                  * lowmem reserves and high watermark so that kswapd
2085                  * should be able to balance it without having to
2086                  * write pages from its LRU list.
2087                  *
2088                  * This may look like it could increase pressure on
2089                  * lower zones by failing allocations in higher zones
2090                  * before they are full.  But the pages that do spill
2091                  * over are limited as the lower zones are protected
2092                  * by this very same mechanism.  It should not become
2093                  * a practical burden to them.
2094                  *
2095                  * XXX: For now, allow allocations to potentially
2096                  * exceed the per-zone dirty limit in the slowpath
2097                  * (ALLOC_WMARK_LOW unset) before going into reclaim,
2098                  * which is important when on a NUMA setup the allowed
2099                  * zones are together not big enough to reach the
2100                  * global limit.  The proper fix for these situations
2101                  * will require awareness of zones in the
2102                  * dirty-throttling and the flusher threads.
2103                  */
2104                 if (consider_zone_dirty && !zone_dirty_ok(zone))
2105                         continue;
2106 
2107                 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2108                 if (!zone_watermark_ok(zone, order, mark,
2109                                        ac->classzone_idx, alloc_flags)) {
2110                         int ret;
2111 
2112                         /* Checked here to keep the fast path fast */
2113                         BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2114                         if (alloc_flags & ALLOC_NO_WATERMARKS)
2115                                 goto try_this_zone;
2116 
2117                         if (IS_ENABLED(CONFIG_NUMA) &&
2118                                         !did_zlc_setup && nr_online_nodes > 1) {
2119                                 /*
2120                                  * we do zlc_setup if there are multiple nodes
2121                                  * and before considering the first zone allowed
2122                                  * by the cpuset.
2123                                  */
2124                                 allowednodes = zlc_setup(zonelist, alloc_flags);
2125                                 zlc_active = 1;
2126                                 did_zlc_setup = 1;
2127                         }
2128 
2129                         if (zone_reclaim_mode == 0 ||
2130                             !zone_allows_reclaim(ac->preferred_zone, zone))
2131                                 goto this_zone_full;
2132 
2133                         /*
2134                          * As we may have just activated ZLC, check if the first
2135                          * eligible zone has failed zone_reclaim recently.
2136                          */
2137                         if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2138                                 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2139                                 continue;
2140 
2141                         ret = zone_reclaim(zone, gfp_mask, order);
2142                         switch (ret) {
2143                         case ZONE_RECLAIM_NOSCAN:
2144                                 /* did not scan */
2145                                 continue;
2146                         case ZONE_RECLAIM_FULL:
2147                                 /* scanned but unreclaimable */
2148                                 continue;
2149                         default:
2150                                 /* did we reclaim enough */
2151                                 if (zone_watermark_ok(zone, order, mark,
2152                                                 ac->classzone_idx, alloc_flags))
2153                                         goto try_this_zone;
2154 
2155                                 /*
2156                                  * Failed to reclaim enough to meet watermark.
2157                                  * Only mark the zone full if checking the min
2158                                  * watermark or if we failed to reclaim just
2159                                  * 1<<order pages or else the page allocator
2160                                  * fastpath will prematurely mark zones full
2161                                  * when the watermark is between the low and
2162                                  * min watermarks.
2163                                  */
2164                                 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2165                                     ret == ZONE_RECLAIM_SOME)
2166                                         goto this_zone_full;
2167 
2168                                 continue;
2169                         }
2170                 }
2171 
2172 try_this_zone:
2173                 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2174                                                 gfp_mask, ac->migratetype);
2175                 if (page) {
2176                         if (prep_new_page(page, order, gfp_mask, alloc_flags))
2177                                 goto try_this_zone;
2178                         return page;
2179                 }
2180 this_zone_full:
2181                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
2182                         zlc_mark_zone_full(zonelist, z);
2183         }
2184 
2185         /*
2186          * The first pass makes sure allocations are spread fairly within the
2187          * local node.  However, the local node might have free pages left
2188          * after the fairness batches are exhausted, and remote zones haven't
2189          * even been considered yet.  Try once more without fairness, and
2190          * include remote zones now, before entering the slowpath and waking
2191          * kswapd: prefer spilling to a remote zone over swapping locally.
2192          */
2193         if (alloc_flags & ALLOC_FAIR) {
2194                 alloc_flags &= ~ALLOC_FAIR;
2195                 if (nr_fair_skipped) {
2196                         zonelist_rescan = true;
2197                         reset_alloc_batches(ac->preferred_zone);
2198                 }
2199                 if (nr_online_nodes > 1)
2200                         zonelist_rescan = true;
2201         }
2202 
2203         if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2204                 /* Disable zlc cache for second zonelist scan */
2205                 zlc_active = 0;
2206                 zonelist_rescan = true;
2207         }
2208 
2209         if (zonelist_rescan)
2210                 goto zonelist_scan;
2211 
2212         return NULL;
2213 }
2214 
2215 /*
2216  * Large machines with many possible nodes should not always dump per-node
2217  * meminfo in irq context.
2218  */
2219 static inline bool should_suppress_show_mem(void)
2220 {
2221         bool ret = false;
2222 
2223 #if NODES_SHIFT > 8
2224         ret = in_interrupt();
2225 #endif
2226         return ret;
2227 }
2228 
2229 static DEFINE_RATELIMIT_STATE(nopage_rs,
2230                 DEFAULT_RATELIMIT_INTERVAL,
2231                 DEFAULT_RATELIMIT_BURST);
2232 
2233 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2234 {
2235         unsigned int filter = SHOW_MEM_FILTER_NODES;
2236 
2237         if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2238             debug_guardpage_minorder() > 0)
2239                 return;
2240 
2241         /*
2242          * This documents exceptions given to allocations in certain
2243          * contexts that are allowed to allocate outside current's set
2244          * of allowed nodes.
2245          */
2246         if (!(gfp_mask & __GFP_NOMEMALLOC))
2247                 if (test_thread_flag(TIF_MEMDIE) ||
2248                     (current->flags & (PF_MEMALLOC | PF_EXITING)))
2249                         filter &= ~SHOW_MEM_FILTER_NODES;
2250         if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2251                 filter &= ~SHOW_MEM_FILTER_NODES;
2252 
2253         if (fmt) {
2254                 struct va_format vaf;
2255                 va_list args;
2256 
2257                 va_start(args, fmt);
2258 
2259                 vaf.fmt = fmt;
2260                 vaf.va = &args;
2261 
2262                 pr_warn("%pV", &vaf);
2263 
2264                 va_end(args);
2265         }
2266 
2267         pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2268                 current->comm, order, gfp_mask);
2269 
2270         dump_stack();
2271         if (!should_suppress_show_mem())
2272                 show_mem(filter);
2273 }
2274 
2275 static inline int
2276 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
2277                                 unsigned long did_some_progress,
2278                                 unsigned long pages_reclaimed)
2279 {
2280         /* Do not loop if specifically requested */
2281         if (gfp_mask & __GFP_NORETRY)
2282                 return 0;
2283 
2284         /* Always retry if specifically requested */
2285         if (gfp_mask & __GFP_NOFAIL)
2286                 return 1;
2287 
2288         /*
2289          * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2290          * making forward progress without invoking OOM. Suspend also disables
2291          * storage devices so kswapd will not help. Bail if we are suspending.
2292          */
2293         if (!did_some_progress && pm_suspended_storage())
2294                 return 0;
2295 
2296         /*
2297          * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2298          * means __GFP_NOFAIL, but that may not be true in other
2299          * implementations.
2300          */
2301         if (order <= PAGE_ALLOC_COSTLY_ORDER)
2302                 return 1;
2303 
2304         /*
2305          * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2306          * specified, then we retry until we no longer reclaim any pages
2307          * (above), or we've reclaimed an order of pages at least as
2308          * large as the allocation's order. In both cases, if the
2309          * allocation still fails, we stop retrying.
2310          */
2311         if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2312                 return 1;
2313 
2314         return 0;
2315 }
2316 
2317 static inline struct page *
2318 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2319         const struct alloc_context *ac, unsigned long *did_some_progress)
2320 {
2321         struct page *page;
2322 
2323         *did_some_progress = 0;
2324 
2325         /*
2326          * Acquire the per-zone oom lock for each zone.  If that
2327          * fails, somebody else is making progress for us.
2328          */
2329         if (!oom_zonelist_trylock(ac->zonelist, gfp_mask)) {
2330                 *did_some_progress = 1;
2331                 schedule_timeout_uninterruptible(1);
2332                 return NULL;
2333         }
2334 
2335         /*
2336          * Go through the zonelist yet one more time, keep very high watermark
2337          * here, this is only to catch a parallel oom killing, we must fail if
2338          * we're still under heavy pressure.
2339          */
2340         page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2341                                         ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
2342         if (page)
2343                 goto out;
2344 
2345         if (!(gfp_mask & __GFP_NOFAIL)) {
2346                 /* Coredumps can quickly deplete all memory reserves */
2347                 if (current->flags & PF_DUMPCORE)
2348                         goto out;
2349                 /* The OOM killer will not help higher order allocs */
2350                 if (order > PAGE_ALLOC_COSTLY_ORDER)
2351                         goto out;
2352                 /* The OOM killer does not needlessly kill tasks for lowmem */
2353                 if (ac->high_zoneidx < ZONE_NORMAL)
2354                         goto out;
2355                 /* The OOM killer does not compensate for light reclaim */
2356                 if (!(gfp_mask & __GFP_FS)) {
2357                         /*
2358                          * XXX: Page reclaim didn't yield anything,
2359                          * and the OOM killer can't be invoked, but
2360                          * keep looping as per should_alloc_retry().
2361                          */
2362                         *did_some_progress = 1;
2363                         goto out;
2364                 }
2365                 /*
2366                  * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2367                  * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2368                  * The caller should handle page allocation failure by itself if
2369                  * it specifies __GFP_THISNODE.
2370                  * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2371                  */
2372                 if (gfp_mask & __GFP_THISNODE)
2373                         goto out;
2374         }
2375         /* Exhausted what can be done so it's blamo time */
2376         if (out_of_memory(ac->zonelist, gfp_mask, order, ac->nodemask, false)
2377                         || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
2378                 *did_some_progress = 1;
2379 out:
2380         oom_zonelist_unlock(ac->zonelist, gfp_mask);
2381         return page;
2382 }
2383 
2384 #ifdef CONFIG_COMPACTION
2385 /* Try memory compaction for high-order allocations before reclaim */
2386 static struct page *
2387 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2388                 int alloc_flags, const struct alloc_context *ac,
2389                 enum migrate_mode mode, int *contended_compaction,
2390                 bool *deferred_compaction)
2391 {
2392         unsigned long compact_result;
2393         struct page *page;
2394 
2395         if (!order)
2396                 return NULL;
2397 
2398         current->flags |= PF_MEMALLOC;
2399         compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2400                                                 mode, contended_compaction);
2401         current->flags &= ~PF_MEMALLOC;
2402 
2403         switch (compact_result) {
2404         case COMPACT_DEFERRED:
2405                 *deferred_compaction = true;
2406                 /* fall-through */
2407         case COMPACT_SKIPPED:
2408                 return NULL;
2409         default:
2410                 break;
2411         }
2412 
2413         /*
2414          * At least in one zone compaction wasn't deferred or skipped, so let's
2415          * count a compaction stall
2416          */
2417         count_vm_event(COMPACTSTALL);
2418 
2419         page = get_page_from_freelist(gfp_mask, order,
2420                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2421 
2422         if (page) {
2423                 struct zone *zone = page_zone(page);
2424 
2425                 zone->compact_blockskip_flush = false;
2426                 compaction_defer_reset(zone, order, true);
2427                 count_vm_event(COMPACTSUCCESS);
2428                 return page;
2429         }
2430 
2431         /*
2432          * It's bad if compaction run occurs and fails. The most likely reason
2433          * is that pages exist, but not enough to satisfy watermarks.
2434          */
2435         count_vm_event(COMPACTFAIL);
2436 
2437         cond_resched();
2438 
2439         return NULL;
2440 }
2441 #else
2442 static inline struct page *
2443 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2444                 int alloc_flags, const struct alloc_context *ac,
2445                 enum migrate_mode mode, int *contended_compaction,
2446                 bool *deferred_compaction)
2447 {
2448         return NULL;
2449 }
2450 #endif /* CONFIG_COMPACTION */
2451 
2452 /* Perform direct synchronous page reclaim */
2453 static int
2454 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
2455                                         const struct alloc_context *ac)
2456 {
2457         struct reclaim_state reclaim_state;
2458         int progress;
2459 
2460         cond_resched();
2461 
2462         /* We now go into synchronous reclaim */
2463         cpuset_memory_pressure_bump();
2464         current->flags |= PF_MEMALLOC;
2465         lockdep_set_current_reclaim_state(gfp_mask);
2466         reclaim_state.reclaimed_slab = 0;
2467         current->reclaim_state = &reclaim_state;
2468 
2469         progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2470                                                                 ac->nodemask);
2471 
2472         current->reclaim_state = NULL;
2473         lockdep_clear_current_reclaim_state();
2474         current->flags &= ~PF_MEMALLOC;
2475 
2476         cond_resched();
2477 
2478         return progress;
2479 }
2480 
2481 /* The really slow allocator path where we enter direct reclaim */
2482 static inline struct page *
2483 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2484                 int alloc_flags, const struct alloc_context *ac,
2485                 unsigned long *did_some_progress)
2486 {
2487         struct page *page = NULL;
2488         bool drained = false;
2489 
2490         *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
2491         if (unlikely(!(*did_some_progress)))
2492                 return NULL;
2493 
2494         /* After successful reclaim, reconsider all zones for allocation */
2495         if (IS_ENABLED(CONFIG_NUMA))
2496                 zlc_clear_zones_full(ac->zonelist);
2497 
2498 retry:
2499         page = get_page_from_freelist(gfp_mask, order,
2500                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2501 
2502         /*
2503          * If an allocation failed after direct reclaim, it could be because
2504          * pages are pinned on the per-cpu lists. Drain them and try again
2505          */
2506         if (!page && !drained) {
2507                 drain_all_pages(NULL);
2508                 drained = true;
2509                 goto retry;
2510         }
2511 
2512         return page;
2513 }
2514 
2515 /*
2516  * This is called in the allocator slow-path if the allocation request is of
2517  * sufficient urgency to ignore watermarks and take other desperate measures
2518  */
2519 static inline struct page *
2520 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2521                                 const struct alloc_context *ac)
2522 {
2523         struct page *page;
2524 
2525         do {
2526                 page = get_page_from_freelist(gfp_mask, order,
2527                                                 ALLOC_NO_WATERMARKS, ac);
2528 
2529                 if (!page && gfp_mask & __GFP_NOFAIL)
2530                         wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2531                                                                         HZ/50);
2532         } while (!page && (gfp_mask & __GFP_NOFAIL));
2533 
2534         return page;
2535 }
2536 
2537 static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
2538 {
2539         struct zoneref *z;
2540         struct zone *zone;
2541 
2542         for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2543                                                 ac->high_zoneidx, ac->nodemask)
2544                 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
2545 }
2546 
2547 static inline int
2548 gfp_to_alloc_flags(gfp_t gfp_mask)
2549 {
2550         int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2551         const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
2552 
2553         /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2554         BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2555 
2556         /*
2557          * The caller may dip into page reserves a bit more if the caller
2558          * cannot run direct reclaim, or if the caller has realtime scheduling
2559          * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
2560          * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
2561          */
2562         alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2563 
2564         if (atomic) {
2565                 /*
2566                  * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2567                  * if it can't schedule.
2568                  */
2569                 if (!(gfp_mask & __GFP_NOMEMALLOC))
2570                         alloc_flags |= ALLOC_HARDER;
2571                 /*
2572                  * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2573                  * comment for __cpuset_node_allowed().
2574                  */
2575                 alloc_flags &= ~ALLOC_CPUSET;
2576         } else if (unlikely(rt_task(current)) && !in_interrupt())
2577                 alloc_flags |= ALLOC_HARDER;
2578 
2579         if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2580                 if (gfp_mask & __GFP_MEMALLOC)
2581                         alloc_flags |= ALLOC_NO_WATERMARKS;
2582                 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2583                         alloc_flags |= ALLOC_NO_WATERMARKS;
2584                 else if (!in_interrupt() &&
2585                                 ((current->flags & PF_MEMALLOC) ||
2586                                  unlikely(test_thread_flag(TIF_MEMDIE))))
2587                         alloc_flags |= ALLOC_NO_WATERMARKS;
2588         }
2589 #ifdef CONFIG_CMA
2590         if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2591                 alloc_flags |= ALLOC_CMA;
2592 #endif
2593         return alloc_flags;
2594 }
2595 
2596 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2597 {
2598         return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
2599 }
2600 
2601 static inline struct page *
2602 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2603                                                 struct alloc_context *ac)
2604 {
2605         const gfp_t wait = gfp_mask & __GFP_WAIT;
2606         struct page *page = NULL;
2607         int alloc_flags;
2608         unsigned long pages_reclaimed = 0;
2609         unsigned long did_some_progress;
2610         enum migrate_mode migration_mode = MIGRATE_ASYNC;
2611         bool deferred_compaction = false;
2612         int contended_compaction = COMPACT_CONTENDED_NONE;
2613 
2614         /*
2615          * In the slowpath, we sanity check order to avoid ever trying to
2616          * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2617          * be using allocators in order of preference for an area that is
2618          * too large.
2619          */
2620         if (order >= MAX_ORDER) {
2621                 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2622                 return NULL;
2623         }
2624 
2625         /*
2626          * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2627          * __GFP_NOWARN set) should not cause reclaim since the subsystem
2628          * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2629          * using a larger set of nodes after it has established that the
2630          * allowed per node queues are empty and that nodes are
2631          * over allocated.
2632          */
2633         if (IS_ENABLED(CONFIG_NUMA) &&
2634             (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2635                 goto nopage;
2636 
2637 retry:
2638         if (!(gfp_mask & __GFP_NO_KSWAPD))
2639                 wake_all_kswapds(order, ac);
2640 
2641         /*
2642          * OK, we're below the kswapd watermark and have kicked background
2643          * reclaim. Now things get more complex, so set up alloc_flags according
2644          * to how we want to proceed.
2645          */
2646         alloc_flags = gfp_to_alloc_flags(gfp_mask);
2647 
2648         /*
2649          * Find the true preferred zone if the allocation is unconstrained by
2650          * cpusets.
2651          */
2652         if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
2653                 struct zoneref *preferred_zoneref;
2654                 preferred_zoneref = first_zones_zonelist(ac->zonelist,
2655                                 ac->high_zoneidx, NULL, &ac->preferred_zone);
2656                 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
2657         }
2658 
2659         /* This is the last chance, in general, before the goto nopage. */
2660         page = get_page_from_freelist(gfp_mask, order,
2661                                 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2662         if (page)
2663                 goto got_pg;
2664 
2665         /* Allocate without watermarks if the context allows */
2666         if (alloc_flags & ALLOC_NO_WATERMARKS) {
2667                 /*
2668                  * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2669                  * the allocation is high priority and these type of
2670                  * allocations are system rather than user orientated
2671                  */
2672                 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
2673 
2674                 page = __alloc_pages_high_priority(gfp_mask, order, ac);
2675 
2676                 if (page) {
2677                         goto got_pg;
2678                 }
2679         }
2680 
2681         /* Atomic allocations - we can't balance anything */
2682         if (!wait) {
2683                 /*
2684                  * All existing users of the deprecated __GFP_NOFAIL are
2685                  * blockable, so warn of any new users that actually allow this
2686                  * type of allocation to fail.
2687                  */
2688                 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
2689                 goto nopage;
2690         }
2691 
2692         /* Avoid recursion of direct reclaim */
2693         if (current->flags & PF_MEMALLOC)
2694                 goto nopage;
2695 
2696         /* Avoid allocations with no watermarks from looping endlessly */
2697         if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2698                 goto nopage;
2699 
2700         /*
2701          * Try direct compaction. The first pass is asynchronous. Subsequent
2702          * attempts after direct reclaim are synchronous
2703          */
2704         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
2705                                         migration_mode,
2706                                         &contended_compaction,
2707                                         &deferred_compaction);
2708         if (page)
2709                 goto got_pg;
2710 
2711         /* Checks for THP-specific high-order allocations */
2712         if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
2713                 /*
2714                  * If compaction is deferred for high-order allocations, it is
2715                  * because sync compaction recently failed. If this is the case
2716                  * and the caller requested a THP allocation, we do not want
2717                  * to heavily disrupt the system, so we fail the allocation
2718                  * instead of entering direct reclaim.
2719                  */
2720                 if (deferred_compaction)
2721                         goto nopage;
2722 
2723                 /*
2724                  * In all zones where compaction was attempted (and not
2725                  * deferred or skipped), lock contention has been detected.
2726                  * For THP allocation we do not want to disrupt the others
2727                  * so we fallback to base pages instead.
2728                  */
2729                 if (contended_compaction == COMPACT_CONTENDED_LOCK)
2730                         goto nopage;
2731 
2732                 /*
2733                  * If compaction was aborted due to need_resched(), we do not
2734                  * want to further increase allocation latency, unless it is
2735                  * khugepaged trying to collapse.
2736                  */
2737                 if (contended_compaction == COMPACT_CONTENDED_SCHED
2738                         && !(current->flags & PF_KTHREAD))
2739                         goto nopage;
2740         }
2741 
2742         /*
2743          * It can become very expensive to allocate transparent hugepages at
2744          * fault, so use asynchronous memory compaction for THP unless it is
2745          * khugepaged trying to collapse.
2746          */
2747         if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
2748                                                 (current->flags & PF_KTHREAD))
2749                 migration_mode = MIGRATE_SYNC_LIGHT;
2750 
2751         /* Try direct reclaim and then allocating */
2752         page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
2753                                                         &did_some_progress);
2754         if (page)
2755                 goto got_pg;
2756 
2757         /* Check if we should retry the allocation */
2758         pages_reclaimed += did_some_progress;
2759         if (should_alloc_retry(gfp_mask, order, did_some_progress,
2760                                                 pages_reclaimed)) {
2761                 /*
2762                  * If we fail to make progress by freeing individual
2763                  * pages, but the allocation wants us to keep going,
2764                  * start OOM killing tasks.
2765                  */
2766                 if (!did_some_progress) {
2767                         page = __alloc_pages_may_oom(gfp_mask, order, ac,
2768                                                         &did_some_progress);
2769                         if (page)
2770                                 goto got_pg;
2771                         if (!did_some_progress)
2772                                 goto nopage;
2773                 }
2774                 /* Wait for some write requests to complete then retry */
2775                 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
2776                 goto retry;
2777         } else {
2778                 /*
2779                  * High-order allocations do not necessarily loop after
2780                  * direct reclaim and reclaim/compaction depends on compaction
2781                  * being called after reclaim so call directly if necessary
2782                  */
2783                 page = __alloc_pages_direct_compact(gfp_mask, order,
2784                                         alloc_flags, ac, migration_mode,
2785                                         &contended_compaction,
2786                                         &deferred_compaction);
2787                 if (page)
2788                         goto got_pg;
2789         }
2790 
2791 nopage:
2792         warn_alloc_failed(gfp_mask, order, NULL);
2793 got_pg:
2794         return page;
2795 }
2796 
2797 /*
2798  * This is the 'heart' of the zoned buddy allocator.
2799  */
2800 struct page *
2801 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2802                         struct zonelist *zonelist, nodemask_t *nodemask)
2803 {
2804         struct zoneref *preferred_zoneref;
2805         struct page *page = NULL;
2806         unsigned int cpuset_mems_cookie;
2807         int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
2808         gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
2809         struct alloc_context ac = {
2810                 .high_zoneidx = gfp_zone(gfp_mask),
2811                 .nodemask = nodemask,
2812                 .migratetype = gfpflags_to_migratetype(gfp_mask),
2813         };
2814 
2815         gfp_mask &= gfp_allowed_mask;
2816 
2817         lockdep_trace_alloc(gfp_mask);
2818 
2819         might_sleep_if(gfp_mask & __GFP_WAIT);
2820 
2821         if (should_fail_alloc_page(gfp_mask, order))
2822                 return NULL;
2823 
2824         /*
2825          * Check the zones suitable for the gfp_mask contain at least one
2826          * valid zone. It's possible to have an empty zonelist as a result
2827          * of GFP_THISNODE and a memoryless node
2828          */
2829         if (unlikely(!zonelist->_zonerefs->zone))
2830                 return NULL;
2831 
2832         if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
2833                 alloc_flags |= ALLOC_CMA;
2834 
2835 retry_cpuset:
2836         cpuset_mems_cookie = read_mems_allowed_begin();
2837 
2838         /* We set it here, as __alloc_pages_slowpath might have changed it */
2839         ac.zonelist = zonelist;
2840         /* The preferred zone is used for statistics later */
2841         preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
2842                                 ac.nodemask ? : &cpuset_current_mems_allowed,
2843                                 &ac.preferred_zone);
2844         if (!ac.preferred_zone)
2845                 goto out;
2846         ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
2847 
2848         /* First allocation attempt */
2849         alloc_mask = gfp_mask|__GFP_HARDWALL;
2850         page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
2851         if (unlikely(!page)) {
2852                 /*
2853                  * Runtime PM, block IO and its error handling path
2854                  * can deadlock because I/O on the device might not
2855                  * complete.
2856                  */
2857                 alloc_mask = memalloc_noio_flags(gfp_mask);
2858 
2859                 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
2860         }
2861 
2862         if (kmemcheck_enabled && page)
2863                 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2864 
2865         trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
2866 
2867 out:
2868         /*
2869          * When updating a task's mems_allowed, it is possible to race with
2870          * parallel threads in such a way that an allocation can fail while
2871          * the mask is being updated. If a page allocation is about to fail,
2872          * check if the cpuset changed during allocation and if so, retry.
2873          */
2874         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
2875                 goto retry_cpuset;
2876 
2877         return page;
2878 }
2879 EXPORT_SYMBOL(__alloc_pages_nodemask);
2880 
2881 /*
2882  * Common helper functions.
2883  */
2884 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2885 {
2886         struct page *page;
2887 
2888         /*
2889          * __get_free_pages() returns a 32-bit address, which cannot represent
2890          * a highmem page
2891          */
2892         VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2893 
2894         page = alloc_pages(gfp_mask, order);
2895         if (!page)
2896                 return 0;
2897         return (unsigned long) page_address(page);
2898 }
2899 EXPORT_SYMBOL(__get_free_pages);
2900 
2901 unsigned long get_zeroed_page(gfp_t gfp_mask)
2902 {
2903         return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2904 }
2905 EXPORT_SYMBOL(get_zeroed_page);
2906 
2907 void __free_pages(struct page *page, unsigned int order)
2908 {
2909         if (put_page_testzero(page)) {
2910                 if (order == 0)
2911                         free_hot_cold_page(page, false);
2912                 else
2913                         __free_pages_ok(page, order);
2914         }
2915 }
2916 
2917 EXPORT_SYMBOL(__free_pages);
2918 
2919 void free_pages(unsigned long addr, unsigned int order)
2920 {
2921         if (addr != 0) {
2922                 VM_BUG_ON(!virt_addr_valid((void *)addr));
2923                 __free_pages(virt_to_page((void *)addr), order);
2924         }
2925 }
2926 
2927 EXPORT_SYMBOL(free_pages);
2928 
2929 /*
2930  * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
2931  * of the current memory cgroup.
2932  *
2933  * It should be used when the caller would like to use kmalloc, but since the
2934  * allocation is large, it has to fall back to the page allocator.
2935  */
2936 struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
2937 {
2938         struct page *page;
2939         struct mem_cgroup *memcg = NULL;
2940 
2941         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2942                 return NULL;
2943         page = alloc_pages(gfp_mask, order);
2944         memcg_kmem_commit_charge(page, memcg, order);
2945         return page;
2946 }
2947 
2948 struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
2949 {
2950         struct page *page;
2951         struct mem_cgroup *memcg = NULL;
2952 
2953         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2954                 return NULL;
2955         page = alloc_pages_node(nid, gfp_mask, order);
2956         memcg_kmem_commit_charge(page, memcg, order);
2957         return page;
2958 }
2959 
2960 /*
2961  * __free_kmem_pages and free_kmem_pages will free pages allocated with
2962  * alloc_kmem_pages.
2963  */
2964 void __free_kmem_pages(struct page *page, unsigned int order)
2965 {
2966         memcg_kmem_uncharge_pages(page, order);
2967         __free_pages(page, order);
2968 }
2969 
2970 void free_kmem_pages(unsigned long addr, unsigned int order)
2971 {
2972         if (addr != 0) {
2973                 VM_BUG_ON(!virt_addr_valid((void *)addr));
2974                 __free_kmem_pages(virt_to_page((void *)addr), order);
2975         }
2976 }
2977 
2978 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2979 {
2980         if (addr) {
2981                 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2982                 unsigned long used = addr + PAGE_ALIGN(size);
2983 
2984                 split_page(virt_to_page((void *)addr), order);
2985                 while (used < alloc_end) {
2986                         free_page(used);
2987                         used += PAGE_SIZE;
2988                 }
2989         }
2990         return (void *)addr;
2991 }
2992 
2993 /**
2994  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2995  * @size: the number of bytes to allocate
2996  * @gfp_mask: GFP flags for the allocation
2997  *
2998  * This function is similar to alloc_pages(), except that it allocates the
2999  * minimum number of pages to satisfy the request.  alloc_pages() can only
3000  * allocate memory in power-of-two pages.
3001  *
3002  * This function is also limited by MAX_ORDER.
3003  *
3004  * Memory allocated by this function must be released by free_pages_exact().
3005  */
3006 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3007 {
3008         unsigned int order = get_order(size);
3009         unsigned long addr;
3010 
3011         addr = __get_free_pages(gfp_mask, order);
3012         return make_alloc_exact(addr, order, size);
3013 }
3014 EXPORT_SYMBOL(alloc_pages_exact);
3015 
3016 /**
3017  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3018  *                         pages on a node.
3019  * @nid: the preferred node ID where memory should be allocated
3020  * @size: the number of bytes to allocate
3021  * @gfp_mask: GFP flags for the allocation
3022  *
3023  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3024  * back.
3025  * Note this is not alloc_pages_exact_node() which allocates on a specific node,
3026  * but is not exact.
3027  */
3028 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
3029 {
3030         unsigned order = get_order(size);
3031         struct page *p = alloc_pages_node(nid, gfp_mask, order);
3032         if (!p)
3033                 return NULL;
3034         return make_alloc_exact((unsigned long)page_address(p), order, size);
3035 }
3036 
3037 /**
3038  * free_pages_exact - release memory allocated via alloc_pages_exact()
3039  * @virt: the value returned by alloc_pages_exact.
3040  * @size: size of allocation, same value as passed to alloc_pages_exact().
3041  *
3042  * Release the memory allocated by a previous call to alloc_pages_exact.
3043  */
3044 void free_pages_exact(void *virt, size_t size)
3045 {
3046         unsigned long addr = (unsigned long)virt;
3047         unsigned long end = addr + PAGE_ALIGN(size);
3048 
3049         while (addr < end) {
3050                 free_page(addr);
3051                 addr += PAGE_SIZE;
3052         }
3053 }
3054 EXPORT_SYMBOL(free_pages_exact);
3055 
3056 /**
3057  * nr_free_zone_pages - count number of pages beyond high watermark
3058  * @offset: The zone index of the highest zone
3059  *
3060  * nr_free_zone_pages() counts the number of counts pages which are beyond the
3061  * high watermark within all zones at or below a given zone index.  For each
3062  * zone, the number of pages is calculated as:
3063  *     managed_pages - high_pages
3064  */
3065 static unsigned long nr_free_zone_pages(int offset)
3066 {
3067         struct zoneref *z;
3068         struct zone *zone;
3069 
3070         /* Just pick one node, since fallback list is circular */
3071         unsigned long sum = 0;
3072 
3073         struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3074 
3075         for_each_zone_zonelist(zone, z, zonelist, offset) {
3076                 unsigned long size = zone->managed_pages;
3077                 unsigned long high = high_wmark_pages(zone);
3078                 if (size > high)
3079                         sum += size - high;
3080         }
3081 
3082         return sum;
3083 }
3084 
3085 /**
3086  * nr_free_buffer_pages - count number of pages beyond high watermark
3087  *
3088  * nr_free_buffer_pages() counts the number of pages which are beyond the high
3089  * watermark within ZONE_DMA and ZONE_NORMAL.
3090  */
3091 unsigned long nr_free_buffer_pages(void)
3092 {
3093         return nr_free_zone_pages(gfp_zone(GFP_USER));
3094 }
3095 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3096 
3097 /**
3098  * nr_free_pagecache_pages - count number of pages beyond high watermark
3099  *
3100  * nr_free_pagecache_pages() counts the number of pages which are beyond the
3101  * high watermark within all zones.
3102  */
3103 unsigned long nr_free_pagecache_pages(void)
3104 {
3105         return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3106 }
3107 
3108 static inline void show_node(struct zone *zone)
3109 {
3110         if (IS_ENABLED(CONFIG_NUMA))
3111                 printk("Node %d ", zone_to_nid(zone));
3112 }
3113 
3114 void si_meminfo(struct sysinfo *val)
3115 {
3116         val->totalram = totalram_pages;
3117         val->sharedram = global_page_state(NR_SHMEM);
3118         val->freeram = global_page_state(NR_FREE_PAGES);
3119         val->bufferram = nr_blockdev_pages();
3120         val->totalhigh = totalhigh_pages;
3121         val->freehigh = nr_free_highpages();
3122         val->mem_unit = PAGE_SIZE;
3123 }
3124 
3125 EXPORT_SYMBOL(si_meminfo);
3126 
3127 #ifdef CONFIG_NUMA
3128 void si_meminfo_node(struct sysinfo *val, int nid)
3129 {
3130         int zone_type;          /* needs to be signed */
3131         unsigned long managed_pages = 0;
3132         pg_data_t *pgdat = NODE_DATA(nid);
3133 
3134         for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3135                 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3136         val->totalram = managed_pages;
3137         val->sharedram = node_page_state(nid, NR_SHMEM);
3138         val->freeram = node_page_state(nid, NR_FREE_PAGES);
3139 #ifdef CONFIG_HIGHMEM
3140         val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
3141         val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3142                         NR_FREE_PAGES);
3143 #else
3144         val->totalhigh = 0;
3145         val->freehigh = 0;
3146 #endif
3147         val->mem_unit = PAGE_SIZE;
3148 }
3149 #endif
3150 
3151 /*
3152  * Determine whether the node should be displayed or not, depending on whether
3153  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3154  */
3155 bool skip_free_areas_node(unsigned int flags, int nid)
3156 {
3157         bool ret = false;
3158         unsigned int cpuset_mems_cookie;
3159 
3160         if (!(flags & SHOW_MEM_FILTER_NODES))
3161                 goto out;
3162 
3163         do {
3164                 cpuset_mems_cookie = read_mems_allowed_begin();
3165                 ret = !node_isset(nid, cpuset_current_mems_allowed);
3166         } while (read_mems_allowed_retry(cpuset_mems_cookie));
3167 out:
3168         return ret;
3169 }
3170 
3171 #define K(x) ((x) << (PAGE_SHIFT-10))
3172 
3173 static void show_migration_types(unsigned char type)
3174 {
3175         static const char types[MIGRATE_TYPES] = {
3176                 [MIGRATE_UNMOVABLE]     = 'U',
3177                 [MIGRATE_RECLAIMABLE]   = 'E',
3178                 [MIGRATE_MOVABLE]       = 'M',
3179                 [MIGRATE_RESERVE]       = 'R',
3180 #ifdef CONFIG_CMA
3181                 [MIGRATE_CMA]           = 'C',
3182 #endif
3183 #ifdef CONFIG_MEMORY_ISOLATION
3184                 [MIGRATE_ISOLATE]       = 'I',
3185 #endif
3186         };
3187         char tmp[MIGRATE_TYPES + 1];
3188         char *p = tmp;
3189         int i;
3190 
3191         for (i = 0; i < MIGRATE_TYPES; i++) {
3192                 if (type & (1 << i))
3193                         *p++ = types[i];
3194         }
3195 
3196         *p = '\0';
3197         printk("(%s) ", tmp);
3198 }
3199 
3200 /*
3201  * Show free area list (used inside shift_scroll-lock stuff)
3202  * We also calculate the percentage fragmentation. We do this by counting the
3203  * memory on each free list with the exception of the first item on the list.
3204  * Suppresses nodes that are not allowed by current's cpuset if
3205  * SHOW_MEM_FILTER_NODES is passed.
3206  */
3207 void show_free_areas(unsigned int filter)
3208 {
3209         int cpu;
3210         struct zone *zone;
3211 
3212         for_each_populated_zone(zone) {
3213                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3214                         continue;
3215                 show_node(zone);
3216                 printk("%s per-cpu:\n", zone->name);
3217 
3218                 for_each_online_cpu(cpu) {
3219                         struct per_cpu_pageset *pageset;
3220 
3221                         pageset = per_cpu_ptr(zone->pageset, cpu);
3222 
3223                         printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3224                                cpu, pageset->pcp.high,
3225                                pageset->pcp.batch, pageset->pcp.count);
3226                 }
3227         }
3228 
3229         printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3230                 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3231                 " unevictable:%lu"
3232                 " dirty:%lu writeback:%lu unstable:%lu\n"
3233                 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3234                 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3235                 " free_cma:%lu\n",
3236                 global_page_state(NR_ACTIVE_ANON),
3237                 global_page_state(NR_INACTIVE_ANON),
3238                 global_page_state(NR_ISOLATED_ANON),
3239                 global_page_state(NR_ACTIVE_FILE),
3240                 global_page_state(NR_INACTIVE_FILE),
3241                 global_page_state(NR_ISOLATED_FILE),
3242                 global_page_state(NR_UNEVICTABLE),
3243                 global_page_state(NR_FILE_DIRTY),
3244                 global_page_state(NR_WRITEBACK),
3245                 global_page_state(NR_UNSTABLE_NFS),
3246                 global_page_state(NR_FREE_PAGES),
3247                 global_page_state(NR_SLAB_RECLAIMABLE),
3248                 global_page_state(NR_SLAB_UNRECLAIMABLE),
3249                 global_page_state(NR_FILE_MAPPED),
3250                 global_page_state(NR_SHMEM),
3251                 global_page_state(NR_PAGETABLE),
3252                 global_page_state(NR_BOUNCE),
3253                 global_page_state(NR_FREE_CMA_PAGES));
3254 
3255         for_each_populated_zone(zone) {
3256                 int i;
3257 
3258                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3259                         continue;
3260                 show_node(zone);
3261                 printk("%s"
3262                         " free:%lukB"
3263                         " min:%lukB"
3264                         " low:%lukB"
3265                         " high:%lukB"
3266                         " active_anon:%lukB"
3267                         " inactive_anon:%lukB"
3268                         " active_file:%lukB"
3269                         " inactive_file:%lukB"
3270                         " unevictable:%lukB"
3271                         " isolated(anon):%lukB"
3272                         " isolated(file):%lukB"
3273                         " present:%lukB"
3274                         " managed:%lukB"
3275                         " mlocked:%lukB"
3276                         " dirty:%lukB"
3277                         " writeback:%lukB"
3278                         " mapped:%lukB"
3279                         " shmem:%lukB"
3280                         " slab_reclaimable:%lukB"
3281                         " slab_unreclaimable:%lukB"
3282                         " kernel_stack:%lukB"
3283                         " pagetables:%lukB"
3284                         " unstable:%lukB"
3285                         " bounce:%lukB"
3286                         " free_cma:%lukB"
3287                         " writeback_tmp:%lukB"
3288                         " pages_scanned:%lu"
3289                         " all_unreclaimable? %s"
3290                         "\n",
3291                         zone->name,
3292                         K(zone_page_state(zone, NR_FREE_PAGES)),
3293                         K(min_wmark_pages(zone)),
3294                         K(low_wmark_pages(zone)),
3295                         K(high_wmark_pages(zone)),
3296                         K(zone_page_state(zone, NR_ACTIVE_ANON)),
3297                         K(zone_page_state(zone, NR_INACTIVE_ANON)),
3298                         K(zone_page_state(zone, NR_ACTIVE_FILE)),
3299                         K(zone_page_state(zone, NR_INACTIVE_FILE)),
3300                         K(zone_page_state(zone, NR_UNEVICTABLE)),
3301                         K(zone_page_state(zone, NR_ISOLATED_ANON)),
3302                         K(zone_page_state(zone, NR_ISOLATED_FILE)),
3303                         K(zone->present_pages),
3304                         K(zone->managed_pages),
3305                         K(zone_page_state(zone, NR_MLOCK)),
3306                         K(zone_page_state(zone, NR_FILE_DIRTY)),
3307                         K(zone_page_state(zone, NR_WRITEBACK)),
3308                         K(zone_page_state(zone, NR_FILE_MAPPED)),
3309                         K(zone_page_state(zone, NR_SHMEM)),
3310                         K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3311                         K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3312                         zone_page_state(zone, NR_KERNEL_STACK) *
3313                                 THREAD_SIZE / 1024,
3314                         K(zone_page_state(zone, NR_PAGETABLE)),
3315                         K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3316                         K(zone_page_state(zone, NR_BOUNCE)),
3317                         K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3318                         K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3319                         K(zone_page_state(zone, NR_PAGES_SCANNED)),
3320                         (!zone_reclaimable(zone) ? "yes" : "no")
3321                         );
3322                 printk("lowmem_reserve[]:");
3323                 for (i = 0; i < MAX_NR_ZONES; i++)
3324                         printk(" %ld", zone->lowmem_reserve[i]);
3325                 printk("\n");
3326         }
3327 
3328         for_each_populated_zone(zone) {
3329                 unsigned long nr[MAX_ORDER], flags, order, total = 0;
3330                 unsigned char types[MAX_ORDER];
3331 
3332                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3333                         continue;
3334                 show_node(zone);
3335                 printk("%s: ", zone->name);
3336 
3337                 spin_lock_irqsave(&zone->lock, flags);
3338                 for (order = 0; order < MAX_ORDER; order++) {
3339                         struct free_area *area = &zone->free_area[order];
3340                         int type;
3341 
3342                         nr[order] = area->nr_free;
3343                         total += nr[order] << order;
3344 
3345                         types[order] = 0;
3346                         for (type = 0; type < MIGRATE_TYPES; type++) {
3347                                 if (!list_empty(&area->free_list[type]))
3348                                         types[order] |= 1 << type;
3349                         }
3350                 }
3351                 spin_unlock_irqrestore(&zone->lock, flags);
3352                 for (order = 0; order < MAX_ORDER; order++) {
3353                         printk("%lu*%lukB ", nr[order], K(1UL) << order);
3354                         if (nr[order])
3355                                 show_migration_types(types[order]);
3356                 }
3357                 printk("= %lukB\n", K(total));
3358         }
3359 
3360         hugetlb_show_meminfo();
3361 
3362         printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3363 
3364         show_swap_cache_info();
3365 }
3366 
3367 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3368 {
3369         zoneref->zone = zone;
3370         zoneref->zone_idx = zone_idx(zone);
3371 }
3372 
3373 /*
3374  * Builds allocation fallback zone lists.
3375  *
3376  * Add all populated zones of a node to the zonelist.
3377  */
3378 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3379                                 int nr_zones)
3380 {
3381         struct zone *zone;
3382         enum zone_type zone_type = MAX_NR_ZONES;
3383 
3384         do {
3385                 zone_type--;
3386                 zone = pgdat->node_zones + zone_type;
3387                 if (populated_zone(zone)) {
3388                         zoneref_set_zone(zone,
3389                                 &zonelist->_zonerefs[nr_zones++]);
3390                         check_highest_zone(zone_type);
3391                 }
3392         } while (zone_type);
3393 
3394         return nr_zones;
3395 }
3396 
3397 
3398 /*
3399  *  zonelist_order:
3400  *  0 = automatic detection of better ordering.
3401  *  1 = order by ([node] distance, -zonetype)
3402  *  2 = order by (-zonetype, [node] distance)
3403  *
3404  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3405  *  the same zonelist. So only NUMA can configure this param.
3406  */
3407 #define ZONELIST_ORDER_DEFAULT  0
3408 #define ZONELIST_ORDER_NODE     1
3409 #define ZONELIST_ORDER_ZONE     2
3410 
3411 /* zonelist order in the kernel.
3412  * set_zonelist_order() will set this to NODE or ZONE.
3413  */
3414 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3415 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3416 
3417 
3418 #ifdef CONFIG_NUMA
3419 /* The value user specified ....changed by config */
3420 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3421 /* string for sysctl */
3422 #define NUMA_ZONELIST_ORDER_LEN 16
3423 char numa_zonelist_order[16] = "default";
3424 
3425 /*
3426  * interface for configure zonelist ordering.
3427  * command line option "numa_zonelist_order"
3428  *      = "[dD]efault   - default, automatic configuration.
3429  *      = "[nN]ode      - order by node locality, then by zone within node
3430  *      = "[zZ]one      - order by zone, then by locality within zone
3431  */
3432 
3433 static int __parse_numa_zonelist_order(char *s)
3434 {
3435         if (*s == 'd' || *s == 'D') {
3436                 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3437         } else if (*s == 'n' || *s == 'N') {
3438                 user_zonelist_order = ZONELIST_ORDER_NODE;
3439         } else if (*s == 'z' || *s == 'Z') {
3440                 user_zonelist_order = ZONELIST_ORDER_ZONE;
3441         } else {
3442                 printk(KERN_WARNING
3443                         "Ignoring invalid numa_zonelist_order value:  "
3444                         "%s\n", s);
3445                 return -EINVAL;
3446         }
3447         return 0;
3448 }
3449 
3450 static __init int setup_numa_zonelist_order(char *s)
3451 {
3452         int ret;
3453 
3454         if (!s)
3455                 return 0;
3456 
3457         ret = __parse_numa_zonelist_order(s);
3458         if (ret == 0)
3459                 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3460 
3461         return ret;
3462 }
3463 early_param("numa_zonelist_order", setup_numa_zonelist_order);
3464 
3465 /*
3466  * sysctl handler for numa_zonelist_order
3467  */
3468 int numa_zonelist_order_handler(struct ctl_table *table, int write,
3469                 void __user *buffer, size_t *length,
3470                 loff_t *ppos)
3471 {
3472         char saved_string[NUMA_ZONELIST_ORDER_LEN];
3473         int ret;
3474         static DEFINE_MUTEX(zl_order_mutex);
3475 
3476         mutex_lock(&zl_order_mutex);
3477         if (write) {
3478                 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3479                         ret = -EINVAL;
3480                         goto out;
3481                 }
3482                 strcpy(saved_string, (char *)table->data);
3483         }
3484         ret = proc_dostring(table, write, buffer, length, ppos);
3485         if (ret)
3486                 goto out;
3487         if (write) {
3488                 int oldval = user_zonelist_order;
3489 
3490                 ret = __parse_numa_zonelist_order((char *)table->data);
3491                 if (ret) {
3492                         /*
3493                          * bogus value.  restore saved string
3494                          */
3495                         strncpy((char *)table->data, saved_string,
3496                                 NUMA_ZONELIST_ORDER_LEN);
3497                         user_zonelist_order = oldval;
3498                 } else if (oldval != user_zonelist_order) {
3499                         mutex_lock(&zonelists_mutex);
3500                         build_all_zonelists(NULL, NULL);
3501                         mutex_unlock(&zonelists_mutex);
3502                 }
3503         }
3504 out:
3505         mutex_unlock(&zl_order_mutex);
3506         return ret;
3507 }
3508 
3509 
3510 #define MAX_NODE_LOAD (nr_online_nodes)
3511 static int node_load[MAX_NUMNODES];
3512 
3513 /**
3514  * find_next_best_node - find the next node that should appear in a given node's fallback list
3515  * @node: node whose fallback list we're appending
3516  * @used_node_mask: nodemask_t of already used nodes
3517  *
3518  * We use a number of factors to determine which is the next node that should
3519  * appear on a given node's fallback list.  The node should not have appeared
3520  * already in @node's fallback list, and it should be the next closest node
3521  * according to the distance array (which contains arbitrary distance values
3522  * from each node to each node in the system), and should also prefer nodes
3523  * with no CPUs, since presumably they'll have very little allocation pressure
3524  * on them otherwise.
3525  * It returns -1 if no node is found.
3526  */
3527 static int find_next_best_node(int node, nodemask_t *used_node_mask)
3528 {
3529         int n, val;
3530         int min_val = INT_MAX;
3531         int best_node = NUMA_NO_NODE;
3532         const struct cpumask *tmp = cpumask_of_node(0);
3533 
3534         /* Use the local node if we haven't already */
3535         if (!node_isset(node, *used_node_mask)) {
3536                 node_set(node, *used_node_mask);
3537                 return node;
3538         }
3539 
3540         for_each_node_state(n, N_MEMORY) {
3541 
3542                 /* Don't want a node to appear more than once */
3543                 if (node_isset(n, *used_node_mask))
3544                         continue;
3545 
3546                 /* Use the distance array to find the distance */
3547                 val = node_distance(node, n);
3548 
3549                 /* Penalize nodes under us ("prefer the next node") */
3550                 val += (n < node);
3551 
3552                 /* Give preference to headless and unused nodes */
3553                 tmp = cpumask_of_node(n);
3554                 if (!cpumask_empty(tmp))
3555                         val += PENALTY_FOR_NODE_WITH_CPUS;
3556 
3557                 /* Slight preference for less loaded node */
3558                 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3559                 val += node_load[n];
3560 
3561                 if (val < min_val) {
3562                         min_val = val;
3563                         best_node = n;
3564                 }
3565         }
3566 
3567         if (best_node >= 0)
3568                 node_set(best_node, *used_node_mask);
3569 
3570         return best_node;
3571 }
3572 
3573 
3574 /*
3575  * Build zonelists ordered by node and zones within node.
3576  * This results in maximum locality--normal zone overflows into local
3577  * DMA zone, if any--but risks exhausting DMA zone.
3578  */
3579 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
3580 {
3581         int j;
3582         struct zonelist *zonelist;
3583 
3584         zonelist = &pgdat->node_zonelists[0];
3585         for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
3586                 ;
3587         j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3588         zonelist->_zonerefs[j].zone = NULL;
3589         zonelist->_zonerefs[j].zone_idx = 0;
3590 }
3591 
3592 /*
3593  * Build gfp_thisnode zonelists
3594  */
3595 static void build_thisnode_zonelists(pg_data_t *pgdat)
3596 {
3597         int j;
3598         struct zonelist *zonelist;
3599 
3600         zonelist = &pgdat->node_zonelists[1];
3601         j = build_zonelists_node(pgdat, zonelist, 0);
3602         zonelist->_zonerefs[j].zone = NULL;
3603         zonelist->_zonerefs[j].zone_idx = 0;
3604 }
3605 
3606 /*
3607  * Build zonelists ordered by zone and nodes within zones.
3608  * This results in conserving DMA zone[s] until all Normal memory is
3609  * exhausted, but results in overflowing to remote node while memory
3610  * may still exist in local DMA zone.
3611  */
3612 static int node_order[MAX_NUMNODES];
3613 
3614 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3615 {
3616         int pos, j, node;
3617         int zone_type;          /* needs to be signed */
3618         struct zone *z;
3619         struct zonelist *zonelist;
3620 
3621         zonelist = &pgdat->node_zonelists[0];
3622         pos = 0;
3623         for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3624                 for (j = 0; j < nr_nodes; j++) {
3625                         node = node_order[j];
3626                         z = &NODE_DATA(node)->node_zones[zone_type];
3627                         if (populated_zone(z)) {
3628                                 zoneref_set_zone(z,
3629                                         &zonelist->_zonerefs[pos++]);
3630                                 check_highest_zone(zone_type);
3631                         }
3632                 }
3633         }
3634         zonelist->_zonerefs[pos].zone = NULL;
3635         zonelist->_zonerefs[pos].zone_idx = 0;
3636 }
3637 
3638 #if defined(CONFIG_64BIT)
3639 /*
3640  * Devices that require DMA32/DMA are relatively rare and do not justify a
3641  * penalty to every machine in case the specialised case applies. Default
3642  * to Node-ordering on 64-bit NUMA machines
3643  */
3644 static int default_zonelist_order(void)
3645 {
3646         return ZONELIST_ORDER_NODE;
3647 }
3648 #else
3649 /*
3650  * On 32-bit, the Normal zone needs to be preserved for allocations accessible
3651  * by the kernel. If processes running on node 0 deplete the low memory zone
3652  * then reclaim will occur more frequency increasing stalls and potentially
3653  * be easier to OOM if a large percentage of the zone is under writeback or
3654  * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
3655  * Hence, default to zone ordering on 32-bit.
3656  */
3657 static int default_zonelist_order(void)
3658 {
3659         return ZONELIST_ORDER_ZONE;
3660 }
3661 #endif /* CONFIG_64BIT */
3662 
3663 static void set_zonelist_order(void)
3664 {
3665         if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3666                 current_zonelist_order = default_zonelist_order();
3667         else
3668                 current_zonelist_order = user_zonelist_order;
3669 }
3670 
3671 static void build_zonelists(pg_data_t *pgdat)
3672 {
3673         int j, node, load;
3674         enum zone_type i;
3675         nodemask_t used_mask;
3676         int local_node, prev_node;
3677         struct zonelist *zonelist;
3678         int order = current_zonelist_order;
3679 
3680         /* initialize zonelists */
3681         for (i = 0; i < MAX_ZONELISTS; i++) {
3682                 zonelist = pgdat->node_zonelists + i;
3683                 zonelist->_zonerefs[0].zone = NULL;
3684                 zonelist->_zonerefs[0].zone_idx = 0;
3685         }
3686 
3687         /* NUMA-aware ordering of nodes */
3688         local_node = pgdat->node_id;
3689         load = nr_online_nodes;
3690         prev_node = local_node;
3691         nodes_clear(used_mask);
3692 
3693         memset(node_order, 0, sizeof(node_order));
3694         j = 0;
3695 
3696         while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3697                 /*
3698                  * We don't want to pressure a particular node.
3699                  * So adding penalty to the first node in same
3700                  * distance group to make it round-robin.
3701                  */
3702                 if (node_distance(local_node, node) !=
3703                     node_distance(local_node, prev_node))
3704                         node_load[node] = load;
3705 
3706                 prev_node = node;
3707                 load--;
3708                 if (order == ZONELIST_ORDER_NODE)
3709                         build_zonelists_in_node_order(pgdat, node);
3710                 else
3711                         node_order[j++] = node; /* remember order */
3712         }
3713 
3714         if (order == ZONELIST_ORDER_ZONE) {
3715                 /* calculate node order -- i.e., DMA last! */
3716                 build_zonelists_in_zone_order(pgdat, j);
3717         }
3718 
3719         build_thisnode_zonelists(pgdat);
3720 }
3721 
3722 /* Construct the zonelist performance cache - see further mmzone.h */
3723 static void build_zonelist_cache(pg_data_t *pgdat)
3724 {
3725         struct zonelist *zonelist;
3726         struct zonelist_cache *zlc;
3727         struct zoneref *z;
3728 
3729         zonelist = &pgdat->node_zonelists[0];
3730         zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3731         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3732         for (z = zonelist->_zonerefs; z->zone; z++)
3733                 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3734 }
3735 
3736 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3737 /*
3738  * Return node id of node used for "local" allocations.
3739  * I.e., first node id of first zone in arg node's generic zonelist.
3740  * Used for initializing percpu 'numa_mem', which is used primarily
3741  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3742  */
3743 int local_memory_node(int node)
3744 {
3745         struct zone *zone;
3746 
3747         (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3748                                    gfp_zone(GFP_KERNEL),
3749                                    NULL,
3750                                    &zone);
3751         return zone->node;
3752 }
3753 #endif
3754 
3755 #else   /* CONFIG_NUMA */
3756 
3757 static void set_zonelist_order(void)
3758 {
3759         current_zonelist_order = ZONELIST_ORDER_ZONE;
3760 }
3761 
3762 static void build_zonelists(pg_data_t *pgdat)
3763 {
3764         int node, local_node;
3765         enum zone_type j;
3766         struct zonelist *zonelist;
3767 
3768         local_node = pgdat->node_id;
3769 
3770         zonelist = &pgdat->node_zonelists[0];
3771         j = build_zonelists_node(pgdat, zonelist, 0);
3772 
3773         /*
3774          * Now we build the zonelist so that it contains the zones
3775          * of all the other nodes.
3776          * We don't want to pressure a particular node, so when
3777          * building the zones for node N, we make sure that the
3778          * zones coming right after the local ones are those from
3779          * node N+1 (modulo N)
3780          */
3781         for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3782                 if (!node_online(node))
3783                         continue;
3784                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3785         }
3786         for (node = 0; node < local_node; node++) {
3787                 if (!node_online(node))
3788                         continue;
3789                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3790         }
3791 
3792         zonelist->_zonerefs[j].zone = NULL;
3793         zonelist->_zonerefs[j].zone_idx = 0;
3794 }
3795 
3796 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3797 static void build_zonelist_cache(pg_data_t *pgdat)
3798 {
3799         pgdat->node_zonelists[0].zlcache_ptr = NULL;
3800 }
3801 
3802 #endif  /* CONFIG_NUMA */
3803 
3804 /*
3805  * Boot pageset table. One per cpu which is going to be used for all
3806  * zones and all nodes. The parameters will be set in such a way
3807  * that an item put on a list will immediately be handed over to
3808  * the buddy list. This is safe since pageset manipulation is done
3809  * with interrupts disabled.
3810  *
3811  * The boot_pagesets must be kept even after bootup is complete for
3812  * unused processors and/or zones. They do play a role for bootstrapping
3813  * hotplugged processors.
3814  *
3815  * zoneinfo_show() and maybe other functions do
3816  * not check if the processor is online before following the pageset pointer.
3817  * Other parts of the kernel may not check if the zone is available.
3818  */
3819 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3820 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3821 static void setup_zone_pageset(struct zone *zone);
3822 
3823 /*
3824  * Global mutex to protect against size modification of zonelists
3825  * as well as to serialize pageset setup for the new populated zone.
3826  */
3827 DEFINE_MUTEX(zonelists_mutex);
3828 
3829 /* return values int ....just for stop_machine() */
3830 static int __build_all_zonelists(void *data)
3831 {
3832         int nid;
3833         int cpu;
3834         pg_data_t *self = data;
3835 
3836 #ifdef CONFIG_NUMA
3837         memset(node_load, 0, sizeof(node_load));
3838 #endif
3839 
3840         if (self && !node_online(self->node_id)) {
3841                 build_zonelists(self);
3842                 build_zonelist_cache(self);
3843         }
3844 
3845         for_each_online_node(nid) {
3846                 pg_data_t *pgdat = NODE_DATA(nid);
3847 
3848                 build_zonelists(pgdat);
3849                 build_zonelist_cache(pgdat);
3850         }
3851 
3852         /*
3853          * Initialize the boot_pagesets that are going to be used
3854          * for bootstrapping processors. The real pagesets for
3855          * each zone will be allocated later when the per cpu
3856          * allocator is available.
3857          *
3858          * boot_pagesets are used also for bootstrapping offline
3859          * cpus if the system is already booted because the pagesets
3860          * are needed to initialize allocators on a specific cpu too.
3861          * F.e. the percpu allocator needs the page allocator which
3862          * needs the percpu allocator in order to allocate its pagesets
3863          * (a chicken-egg dilemma).
3864          */
3865         for_each_possible_cpu(cpu) {
3866                 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3867 
3868 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3869                 /*
3870                  * We now know the "local memory node" for each node--
3871                  * i.e., the node of the first zone in the generic zonelist.
3872                  * Set up numa_mem percpu variable for on-line cpus.  During
3873                  * boot, only the boot cpu should be on-line;  we'll init the
3874                  * secondary cpus' numa_mem as they come on-line.  During
3875                  * node/memory hotplug, we'll fixup all on-line cpus.
3876                  */
3877                 if (cpu_online(cpu))
3878                         set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3879 #endif
3880         }
3881 
3882         return 0;
3883 }
3884 
3885 static noinline void __init
3886 build_all_zonelists_init(void)
3887 {
3888         __build_all_zonelists(NULL);
3889         mminit_verify_zonelist();
3890         cpuset_init_current_mems_allowed();
3891 }
3892 
3893 /*
3894  * Called with zonelists_mutex held always
3895  * unless system_state == SYSTEM_BOOTING.
3896  *
3897  * __ref due to (1) call of __meminit annotated setup_zone_pageset
3898  * [we're only called with non-NULL zone through __meminit paths] and
3899  * (2) call of __init annotated helper build_all_zonelists_init
3900  * [protected by SYSTEM_BOOTING].
3901  */
3902 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
3903 {
3904         set_zonelist_order();
3905 
3906         if (system_state == SYSTEM_BOOTING) {
3907                 build_all_zonelists_init();
3908         } else {
3909 #ifdef CONFIG_MEMORY_HOTPLUG
3910                 if (zone)
3911                         setup_zone_pageset(zone);
3912 #endif
3913                 /* we have to stop all cpus to guarantee there is no user
3914                    of zonelist */
3915                 stop_machine(__build_all_zonelists, pgdat, NULL);
3916                 /* cpuset refresh routine should be here */
3917         }
3918         vm_total_pages = nr_free_pagecache_pages();
3919         /*
3920          * Disable grouping by mobility if the number of pages in the
3921          * system is too low to allow the mechanism to work. It would be
3922          * more accurate, but expensive to check per-zone. This check is
3923          * made on memory-hotadd so a system can start with mobility
3924          * disabled and enable it later
3925          */
3926         if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3927                 page_group_by_mobility_disabled = 1;
3928         else
3929                 page_group_by_mobility_disabled = 0;
3930 
3931         pr_info("Built %i zonelists in %s order, mobility grouping %s.  "
3932                 "Total pages: %ld\n",
3933                         nr_online_nodes,
3934                         zonelist_order_name[current_zonelist_order],
3935                         page_group_by_mobility_disabled ? "off" : "on",
3936                         vm_total_pages);
3937 #ifdef CONFIG_NUMA
3938         pr_info("Policy zone: %s\n", zone_names[policy_zone]);
3939 #endif
3940 }
3941 
3942 /*
3943  * Helper functions to size the waitqueue hash table.
3944  * Essentially these want to choose hash table sizes sufficiently
3945  * large so that collisions trying to wait on pages are rare.
3946  * But in fact, the number of active page waitqueues on typical
3947  * systems is ridiculously low, less than 200. So this is even
3948  * conservative, even though it seems large.
3949  *
3950  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3951  * waitqueues, i.e. the size of the waitq table given the number of pages.
3952  */
3953 #define PAGES_PER_WAITQUEUE     256
3954 
3955 #ifndef CONFIG_MEMORY_HOTPLUG
3956 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3957 {
3958         unsigned long size = 1;
3959 
3960         pages /= PAGES_PER_WAITQUEUE;
3961 
3962         while (size < pages)
3963                 size <<= 1;
3964 
3965         /*
3966          * Once we have dozens or even hundreds of threads sleeping
3967          * on IO we've got bigger problems than wait queue collision.
3968          * Limit the size of the wait table to a reasonable size.
3969          */
3970         size = min(size, 4096UL);
3971 
3972         return max(size, 4UL);
3973 }
3974 #else
3975 /*
3976  * A zone's size might be changed by hot-add, so it is not possible to determine
3977  * a suitable size for its wait_table.  So we use the maximum size now.
3978  *
3979  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
3980  *
3981  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
3982  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3983  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
3984  *
3985  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3986  * or more by the traditional way. (See above).  It equals:
3987  *
3988  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
3989  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
3990  *    powerpc (64K page size)             : =  (32G +16M)byte.
3991  */
3992 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3993 {
3994         return 4096UL;
3995 }
3996 #endif
3997 
3998 /*
3999  * This is an integer logarithm so that shifts can be used later
4000  * to extract the more random high bits from the multiplicative
4001  * hash function before the remainder is taken.
4002  */
4003 static inline unsigned long wait_table_bits(unsigned long size)
4004 {
4005         return ffz(~size);
4006 }
4007 
4008 /*
4009  * Check if a pageblock contains reserved pages
4010  */
4011 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4012 {
4013         unsigned long pfn;
4014 
4015         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4016                 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4017                         return 1;
4018         }
4019         return 0;
4020 }
4021 
4022 /*
4023  * Mark a number of pageblocks as MIGRATE_RESERVE. The number
4024  * of blocks reserved is based on min_wmark_pages(zone). The memory within
4025  * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
4026  * higher will lead to a bigger reserve which will get freed as contiguous
4027  * blocks as reclaim kicks in
4028  */
4029 static void setup_zone_migrate_reserve(struct zone *zone)
4030 {
4031         unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
4032         struct page *page;
4033         unsigned long block_migratetype;
4034         int reserve;
4035         int old_reserve;
4036 
4037         /*
4038          * Get the start pfn, end pfn and the number of blocks to reserve
4039          * We have to be careful to be aligned to pageblock_nr_pages to
4040          * make sure that we always check pfn_valid for the first page in
4041          * the block.
4042          */
4043         start_pfn = zone->zone_start_pfn;
4044         end_pfn = zone_end_pfn(zone);
4045         start_pfn = roundup(start_pfn, pageblock_nr_pages);
4046         reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
4047                                                         pageblock_order;
4048 
4049         /*
4050          * Reserve blocks are generally in place to help high-order atomic
4051          * allocations that are short-lived. A min_free_kbytes value that
4052          * would result in more than 2 reserve blocks for atomic allocations
4053          * is assumed to be in place to help anti-fragmentation for the
4054          * future allocation of hugepages at runtime.
4055          */
4056         reserve = min(2, reserve);
4057         old_reserve = zone->nr_migrate_reserve_block;
4058 
4059         /* When memory hot-add, we almost always need to do nothing */
4060         if (reserve == old_reserve)
4061                 return;
4062         zone->nr_migrate_reserve_block = reserve;
4063 
4064         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
4065                 if (!pfn_valid(pfn))
4066                         continue;
4067                 page = pfn_to_page(pfn);
4068 
4069                 /* Watch out for overlapping nodes */
4070                 if (page_to_nid(page) != zone_to_nid(zone))
4071                         continue;
4072 
4073                 block_migratetype = get_pageblock_migratetype(page);
4074 
4075                 /* Only test what is necessary when the reserves are not met */
4076                 if (reserve > 0) {
4077                         /*
4078                          * Blocks with reserved pages will never free, skip
4079                          * them.
4080                          */
4081                         block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4082                         if (pageblock_is_reserved(pfn, block_end_pfn))
4083                                 continue;
4084 
4085                         /* If this block is reserved, account for it */
4086                         if (block_migratetype == MIGRATE_RESERVE) {
4087                                 reserve--;
4088                                 continue;
4089                         }
4090 
4091                         /* Suitable for reserving if this block is movable */
4092                         if (block_migratetype == MIGRATE_MOVABLE) {
4093                                 set_pageblock_migratetype(page,
4094                                                         MIGRATE_RESERVE);
4095                                 move_freepages_block(zone, page,
4096                                                         MIGRATE_RESERVE);
4097                                 reserve--;
4098                                 continue;
4099                         }
4100                 } else if (!old_reserve) {
4101                         /*
4102                          * At boot time we don't need to scan the whole zone
4103                          * for turning off MIGRATE_RESERVE.
4104                          */
4105                         break;
4106                 }
4107 
4108                 /*
4109                  * If the reserve is met and this is a previous reserved block,
4110                  * take it back
4111                  */
4112                 if (block_migratetype == MIGRATE_RESERVE) {
4113                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4114                         move_freepages_block(zone, page, MIGRATE_MOVABLE);
4115                 }
4116         }
4117 }
4118 
4119 /*
4120  * Initially all pages are reserved - free ones are freed
4121  * up by free_all_bootmem() once the early boot process is
4122  * done. Non-atomic initialization, single-pass.
4123  */
4124 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4125                 unsigned long start_pfn, enum memmap_context context)
4126 {
4127         struct page *page;
4128         unsigned long end_pfn = start_pfn + size;
4129         unsigned long pfn;
4130         struct zone *z;
4131 
4132         if (highest_memmap_pfn < end_pfn - 1)
4133                 highest_memmap_pfn = end_pfn - 1;
4134 
4135         z = &NODE_DATA(nid)->node_zones[zone];
4136         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4137                 /*
4138                  * There can be holes in boot-time mem_map[]s
4139                  * handed to this function.  They do not
4140                  * exist on hotplugged memory.
4141                  */
4142                 if (context == MEMMAP_EARLY) {
4143                         if (!early_pfn_valid(pfn))
4144                                 continue;
4145                         if (!early_pfn_in_nid(pfn, nid))
4146                                 continue;
4147                 }
4148                 page = pfn_to_page(pfn);
4149                 set_page_links(page, zone, nid, pfn);
4150                 mminit_verify_page_links(page, zone, nid, pfn);
4151                 init_page_count(page);
4152                 page_mapcount_reset(page);
4153                 page_cpupid_reset_last(page);
4154                 SetPageReserved(page);
4155                 /*
4156                  * Mark the block movable so that blocks are reserved for
4157                  * movable at startup. This will force kernel allocations
4158                  * to reserve their blocks rather than leaking throughout
4159                  * the address space during boot when many long-lived
4160                  * kernel allocations are made. Later some blocks near
4161                  * the start are marked MIGRATE_RESERVE by
4162                  * setup_zone_migrate_reserve()
4163                  *
4164                  * bitmap is created for zone's valid pfn range. but memmap
4165                  * can be created for invalid pages (for alignment)
4166                  * check here not to call set_pageblock_migratetype() against
4167                  * pfn out of zone.
4168                  */
4169                 if ((z->zone_start_pfn <= pfn)
4170                     && (pfn < zone_end_pfn(z))
4171                     && !(pfn & (pageblock_nr_pages - 1)))
4172                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4173 
4174                 INIT_LIST_HEAD(&page->lru);
4175 #ifdef WANT_PAGE_VIRTUAL
4176                 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4177                 if (!is_highmem_idx(zone))
4178                         set_page_address(page, __va(pfn << PAGE_SHIFT));
4179 #endif
4180         }
4181 }
4182 
4183 static void __meminit zone_init_free_lists(struct zone *zone)
4184 {
4185         unsigned int order, t;
4186         for_each_migratetype_order(order, t) {
4187                 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4188                 zone->free_area[order].nr_free = 0;
4189         }
4190 }
4191 
4192 #ifndef __HAVE_ARCH_MEMMAP_INIT
4193 #define memmap_init(size, nid, zone, start_pfn) \
4194         memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4195 #endif
4196 
4197 static int zone_batchsize(struct zone *zone)
4198 {
4199 #ifdef CONFIG_MMU
4200         int batch;
4201 
4202         /*
4203          * The per-cpu-pages pools are set to around 1000th of the
4204          * size of the zone.  But no more than 1/2 of a meg.
4205          *
4206          * OK, so we don't know how big the cache is.  So guess.
4207          */
4208         batch = zone->managed_pages / 1024;
4209         if (batch * PAGE_SIZE > 512 * 1024)
4210                 batch = (512 * 1024) / PAGE_SIZE;
4211         batch /= 4;             /* We effectively *= 4 below */
4212         if (batch < 1)
4213                 batch = 1;
4214 
4215         /*
4216          * Clamp the batch to a 2^n - 1 value. Having a power
4217          * of 2 value was found to be more likely to have
4218          * suboptimal cache aliasing properties in some cases.
4219          *
4220          * For example if 2 tasks are alternately allocating
4221          * batches of pages, one task can end up with a lot
4222          * of pages of one half of the possible page colors
4223          * and the other with pages of the other colors.
4224          */
4225         batch = rounddown_pow_of_two(batch + batch/2) - 1;
4226 
4227         return batch;
4228 
4229 #else
4230         /* The deferral and batching of frees should be suppressed under NOMMU
4231          * conditions.
4232          *
4233          * The problem is that NOMMU needs to be able to allocate large chunks
4234          * of contiguous memory as there's no hardware page translation to
4235          * assemble apparent contiguous memory from discontiguous pages.
4236          *
4237          * Queueing large contiguous runs of pages for batching, however,
4238          * causes the pages to actually be freed in smaller chunks.  As there
4239          * can be a significant delay between the individual batches being
4240          * recycled, this leads to the once large chunks of space being
4241          * fragmented and becoming unavailable for high-order allocations.
4242          */
4243         return 0;
4244 #endif
4245 }
4246 
4247 /*
4248  * pcp->high and pcp->batch values are related and dependent on one another:
4249  * ->batch must never be higher then ->high.
4250  * The following function updates them in a safe manner without read side
4251  * locking.
4252  *
4253  * Any new users of pcp->batch and pcp->high should ensure they can cope with
4254  * those fields changing asynchronously (acording the the above rule).
4255  *
4256  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4257  * outside of boot time (or some other assurance that no concurrent updaters
4258  * exist).
4259  */
4260 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4261                 unsigned long batch)
4262 {
4263        /* start with a fail safe value for batch */
4264         pcp->batch = 1;
4265         smp_wmb();
4266 
4267        /* Update high, then batch, in order */
4268         pcp->high = high;
4269         smp_wmb();
4270 
4271         pcp->batch = batch;
4272 }
4273 
4274 /* a companion to pageset_set_high() */
4275 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4276 {
4277         pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4278 }
4279 
4280 static void pageset_init(struct per_cpu_pageset *p)
4281 {
4282         struct per_cpu_pages *pcp;
4283         int migratetype;
4284 
4285         memset(p, 0, sizeof(*p));
4286 
4287         pcp = &p->pcp;
4288         pcp->count = 0;
4289         for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4290                 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4291 }
4292 
4293 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4294 {
4295         pageset_init(p);
4296         pageset_set_batch(p, batch);
4297 }
4298 
4299 /*
4300  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4301  * to the value high for the pageset p.
4302  */
4303 static void pageset_set_high(struct per_cpu_pageset *p,
4304                                 unsigned long high)
4305 {
4306         unsigned long batch = max(1UL, high / 4);
4307         if ((high / 4) > (PAGE_SHIFT * 8))
4308                 batch = PAGE_SHIFT * 8;
4309 
4310         pageset_update(&p->pcp, high, batch);
4311 }
4312 
4313 static void pageset_set_high_and_batch(struct zone *zone,
4314                                        struct per_cpu_pageset *pcp)
4315 {
4316         if (percpu_pagelist_fraction)
4317                 pageset_set_high(pcp,
4318                         (zone->managed_pages /
4319                                 percpu_pagelist_fraction));
4320         else
4321                 pageset_set_batch(pcp, zone_batchsize(zone));
4322 }
4323 
4324 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4325 {
4326         struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4327 
4328         pageset_init(pcp);
4329         pageset_set_high_and_batch(zone, pcp);
4330 }
4331 
4332 static void __meminit setup_zone_pageset(struct zone *zone)
4333 {
4334         int cpu;
4335         zone->pageset = alloc_percpu(struct per_cpu_pageset);
4336         for_each_possible_cpu(cpu)
4337                 zone_pageset_init(zone, cpu);
4338 }
4339 
4340 /*
4341  * Allocate per cpu pagesets and initialize them.
4342  * Before this call only boot pagesets were available.
4343  */
4344 void __init setup_per_cpu_pageset(void)
4345 {
4346         struct zone *zone;
4347 
4348         for_each_populated_zone(zone)
4349                 setup_zone_pageset(zone);
4350 }
4351 
4352 static noinline __init_refok
4353 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4354 {
4355         int i;
4356         size_t alloc_size;
4357 
4358         /*
4359          * The per-page waitqueue mechanism uses hashed waitqueues
4360          * per zone.
4361          */
4362         zone->wait_table_hash_nr_entries =
4363                  wait_table_hash_nr_entries(zone_size_pages);
4364         zone->wait_table_bits =
4365                 wait_table_bits(zone->wait_table_hash_nr_entries);
4366         alloc_size = zone->wait_table_hash_nr_entries
4367                                         * sizeof(wait_queue_head_t);
4368 
4369         if (!slab_is_available()) {
4370                 zone->wait_table = (wait_queue_head_t *)
4371                         memblock_virt_alloc_node_nopanic(
4372                                 alloc_size, zone->zone_pgdat->node_id);
4373         } else {
4374                 /*
4375                  * This case means that a zone whose size was 0 gets new memory
4376                  * via memory hot-add.
4377                  * But it may be the case that a new node was hot-added.  In
4378                  * this case vmalloc() will not be able to use this new node's
4379                  * memory - this wait_table must be initialized to use this new
4380                  * node itself as well.
4381                  * To use this new node's memory, further consideration will be
4382                  * necessary.
4383                  */
4384                 zone->wait_table = vmalloc(alloc_size);
4385         }
4386         if (!zone->wait_table)
4387                 return -ENOMEM;
4388 
4389         for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4390                 init_waitqueue_head(zone->wait_table + i);
4391 
4392         return 0;
4393 }
4394 
4395 static __meminit void zone_pcp_init(struct zone *zone)
4396 {
4397         /*
4398          * per cpu subsystem is not up at this point. The following code
4399          * relies on the ability of the linker to provide the
4400          * offset of a (static) per cpu variable into the per cpu area.
4401          */
4402         zone->pageset = &boot_pageset;
4403 
4404         if (populated_zone(zone))
4405                 printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
4406                         zone->name, zone->present_pages,
4407                                          zone_batchsize(zone));
4408 }
4409 
4410 int __meminit init_currently_empty_zone(struct zone *zone,
4411                                         unsigned long zone_start_pfn,
4412                                         unsigned long size,
4413                                         enum memmap_context context)
4414 {
4415         struct pglist_data *pgdat = zone->zone_pgdat;
4416         int ret;
4417         ret = zone_wait_table_init(zone, size);
4418         if (ret)
4419                 return ret;
4420         pgdat->nr_zones = zone_idx(zone) + 1;
4421 
4422         zone->zone_start_pfn = zone_start_pfn;
4423 
4424         mminit_dprintk(MMINIT_TRACE, "memmap_init",
4425                         "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4426                         pgdat->node_id,
4427                         (unsigned long)zone_idx(zone),
4428                         zone_start_pfn, (zone_start_pfn + size));
4429 
4430         zone_init_free_lists(zone);
4431 
4432         return 0;
4433 }
4434 
4435 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4436 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4437 /*
4438  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4439  */
4440 int __meminit __early_pfn_to_nid(unsigned long pfn)
4441 {
4442         unsigned long start_pfn, end_pfn;
4443         int nid;
4444         /*
4445          * NOTE: The following SMP-unsafe globals are only used early in boot
4446          * when the kernel is running single-threaded.
4447          */
4448         static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4449         static int __meminitdata last_nid;
4450 
4451         if (last_start_pfn <= pfn && pfn < last_end_pfn)
4452                 return last_nid;
4453 
4454         nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4455         if (nid != -1) {
4456                 last_start_pfn = start_pfn;
4457                 last_end_pfn = end_pfn;
4458                 last_nid = nid;
4459         }
4460 
4461         return nid;
4462 }
4463 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4464 
4465 int __meminit early_pfn_to_nid(unsigned long pfn)
4466 {
4467         int nid;
4468 
4469         nid = __early_pfn_to_nid(pfn);
4470         if (nid >= 0)
4471                 return nid;
4472         /* just returns 0 */
4473         return 0;
4474 }
4475 
4476 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4477 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4478 {
4479         int nid;
4480 
4481         nid = __early_pfn_to_nid(pfn);
4482         if (nid >= 0 && nid != node)
4483                 return false;
4484         return true;
4485 }
4486 #endif
4487 
4488 /**
4489  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4490  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4491  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4492  *
4493  * If an architecture guarantees that all ranges registered contain no holes
4494  * and may be freed, this this function may be used instead of calling
4495  * memblock_free_early_nid() manually.
4496  */
4497 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
4498 {
4499         unsigned long start_pfn, end_pfn;
4500         int i, this_nid;
4501 
4502         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4503                 start_pfn = min(start_pfn, max_low_pfn);
4504                 end_pfn = min(end_pfn, max_low_pfn);
4505 
4506                 if (start_pfn < end_pfn)
4507                         memblock_free_early_nid(PFN_PHYS(start_pfn),
4508                                         (end_pfn - start_pfn) << PAGE_SHIFT,
4509                                         this_nid);
4510         }
4511 }
4512 
4513 /**
4514  * sparse_memory_present_with_active_regions - Call memory_present for each active range
4515  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4516  *
4517  * If an architecture guarantees that all ranges registered contain no holes and may
4518  * be freed, this function may be used instead of calling memory_present() manually.
4519  */
4520 void __init sparse_memory_present_with_active_regions(int nid)
4521 {
4522         unsigned long start_pfn, end_pfn;
4523         int i, this_nid;
4524 
4525         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4526                 memory_present(this_nid, start_pfn, end_pfn);
4527 }
4528 
4529 /**
4530  * get_pfn_range_for_nid - Return the start and end page frames for a node
4531  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4532  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4533  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4534  *
4535  * It returns the start and end page frame of a node based on information
4536  * provided by memblock_set_node(). If called for a node
4537  * with no available memory, a warning is printed and the start and end
4538  * PFNs will be 0.
4539  */
4540 void __meminit get_pfn_range_for_nid(unsigned int nid,
4541                         unsigned long *start_pfn, unsigned long *end_pfn)
4542 {
4543         unsigned long this_start_pfn, this_end_pfn;
4544         int i;
4545 
4546         *start_pfn = -1UL;
4547         *end_pfn = 0;
4548 
4549         for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4550                 *start_pfn = min(*start_pfn, this_start_pfn);
4551                 *end_pfn = max(*end_pfn, this_end_pfn);
4552         }
4553 
4554         if (*start_pfn == -1UL)
4555                 *start_pfn = 0;
4556 }
4557 
4558 /*
4559  * This finds a zone that can be used for ZONE_MOVABLE pages. The
4560  * assumption is made that zones within a node are ordered in monotonic
4561  * increasing memory addresses so that the "highest" populated zone is used
4562  */
4563 static void __init find_usable_zone_for_movable(void)
4564 {
4565         int zone_index;
4566         for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4567                 if (zone_index == ZONE_MOVABLE)
4568                         continue;
4569 
4570                 if (arch_zone_highest_possible_pfn[zone_index] >
4571                                 arch_zone_lowest_possible_pfn[zone_index])
4572                         break;
4573         }
4574 
4575         VM_BUG_ON(zone_index == -1);
4576         movable_zone = zone_index;
4577 }
4578 
4579 /*
4580  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4581  * because it is sized independent of architecture. Unlike the other zones,
4582  * the starting point for ZONE_MOVABLE is not fixed. It may be different
4583  * in each node depending on the size of each node and how evenly kernelcore
4584  * is distributed. This helper function adjusts the zone ranges
4585  * provided by the architecture for a given node by using the end of the
4586  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4587  * zones within a node are in order of monotonic increases memory addresses
4588  */
4589 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4590                                         unsigned long zone_type,
4591                                         unsigned long node_start_pfn,
4592                                         unsigned long node_end_pfn,
4593                                         unsigned long *zone_start_pfn,
4594                                         unsigned long *zone_end_pfn)
4595 {
4596         /* Only adjust if ZONE_MOVABLE is on this node */
4597         if (zone_movable_pfn[nid]) {
4598                 /* Size ZONE_MOVABLE */
4599                 if (zone_type == ZONE_MOVABLE) {
4600                         *zone_start_pfn = zone_movable_pfn[nid];
4601                         *zone_end_pfn = min(node_end_pfn,
4602                                 arch_zone_highest_possible_pfn[movable_zone]);
4603 
4604                 /* Adjust for ZONE_MOVABLE starting within this range */
4605                 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4606                                 *zone_end_pfn > zone_movable_pfn[nid]) {
4607                         *zone_end_pfn = zone_movable_pfn[nid];
4608 
4609                 /* Check if this whole range is within ZONE_MOVABLE */
4610                 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4611                         *zone_start_pfn = *zone_end_pfn;
4612         }
4613 }
4614 
4615 /*
4616  * Return the number of pages a zone spans in a node, including holes
4617  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4618  */
4619 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4620                                         unsigned long zone_type,
4621                                         unsigned long node_start_pfn,
4622                                         unsigned long node_end_pfn,
4623                                         unsigned long *ignored)
4624 {
4625         unsigned long zone_start_pfn, zone_end_pfn;
4626 
4627         /* Get the start and end of the zone */
4628         zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4629         zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4630         adjust_zone_range_for_zone_movable(nid, zone_type,
4631                                 node_start_pfn, node_end_pfn,
4632                                 &zone_start_pfn, &zone_end_pfn);
4633 
4634         /* Check that this node has pages within the zone's required range */
4635         if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4636                 return 0;
4637 
4638         /* Move the zone boundaries inside the node if necessary */
4639         zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4640         zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4641 
4642         /* Return the spanned pages */
4643         return zone_end_pfn - zone_start_pfn;
4644 }
4645 
4646 /*
4647  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4648  * then all holes in the requested range will be accounted for.
4649  */
4650 unsigned long __meminit __absent_pages_in_range(int nid,
4651                                 unsigned long range_start_pfn,
4652                                 unsigned long range_end_pfn)
4653 {
4654         unsigned long nr_absent = range_end_pfn - range_start_pfn;
4655         unsigned long start_pfn, end_pfn;
4656         int i;
4657 
4658         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4659                 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4660                 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4661                 nr_absent -= end_pfn - start_pfn;
4662         }
4663         return nr_absent;
4664 }
4665 
4666 /**
4667  * absent_pages_in_range - Return number of page frames in holes within a range
4668  * @start_pfn: The start PFN to start searching for holes
4669  * @end_pfn: The end PFN to stop searching for holes
4670  *
4671  * It returns the number of pages frames in memory holes within a range.
4672  */
4673 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4674                                                         unsigned long end_pfn)
4675 {
4676         return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4677 }
4678 
4679 /* Return the number of page frames in holes in a zone on a node */
4680 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4681                                         unsigned long zone_type,
4682                                         unsigned long node_start_pfn,
4683                                         unsigned long node_end_pfn,
4684                                         unsigned long *ignored)
4685 {
4686         unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4687         unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
4688         unsigned long zone_start_pfn, zone_end_pfn;
4689 
4690         zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4691         zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
4692 
4693         adjust_zone_range_for_zone_movable(nid, zone_type,
4694                         node_start_pfn, node_end_pfn,
4695                         &zone_start_pfn, &zone_end_pfn);
4696         return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4697 }
4698 
4699 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4700 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4701                                         unsigned long zone_type,
4702                                         unsigned long node_start_pfn,
4703                                         unsigned long node_end_pfn,
4704                                         unsigned long *zones_size)
4705 {
4706         return zones_size[zone_type];
4707 }
4708 
4709 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4710                                                 unsigned long zone_type,
4711                                                 unsigned long node_start_pfn,
4712                                                 unsigned long node_end_pfn,
4713                                                 unsigned long *zholes_size)
4714 {
4715         if (!zholes_size)
4716                 return 0;
4717 
4718         return zholes_size[zone_type];
4719 }
4720 
4721 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4722 
4723 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4724                                                 unsigned long node_start_pfn,
4725                                                 unsigned long node_end_pfn,
4726                                                 unsigned long *zones_size,
4727                                                 unsigned long *zholes_size)
4728 {
4729         unsigned long realtotalpages, totalpages = 0;
4730         enum zone_type i;
4731 
4732         for (i = 0; i < MAX_NR_ZONES; i++)
4733                 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4734                                                          node_start_pfn,
4735                                                          node_end_pfn,
4736                                                          zones_size);
4737         pgdat->node_spanned_pages = totalpages;
4738 
4739         realtotalpages = totalpages;
4740         for (i = 0; i < MAX_NR_ZONES; i++)
4741                 realtotalpages -=
4742                         zone_absent_pages_in_node(pgdat->node_id, i,
4743                                                   node_start_pfn, node_end_pfn,
4744                                                   zholes_size);
4745         pgdat->node_present_pages = realtotalpages;
4746         printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4747                                                         realtotalpages);
4748 }
4749 
4750 #ifndef CONFIG_SPARSEMEM
4751 /*
4752  * Calculate the size of the zone->blockflags rounded to an unsigned long
4753  * Start by making sure zonesize is a multiple of pageblock_order by rounding
4754  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4755  * round what is now in bits to nearest long in bits, then return it in
4756  * bytes.
4757  */
4758 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
4759 {
4760         unsigned long usemapsize;
4761 
4762         zonesize += zone_start_pfn & (pageblock_nr_pages-1);
4763         usemapsize = roundup(zonesize, pageblock_nr_pages);
4764         usemapsize = usemapsize >> pageblock_order;
4765         usemapsize *= NR_PAGEBLOCK_BITS;
4766         usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4767 
4768         return usemapsize / 8;
4769 }
4770 
4771 static void __init setup_usemap(struct pglist_data *pgdat,
4772                                 struct zone *zone,
4773                                 unsigned long zone_start_pfn,
4774                                 unsigned long zonesize)
4775 {
4776         unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
4777         zone->pageblock_flags = NULL;
4778         if (usemapsize)
4779                 zone->pageblock_flags =
4780                         memblock_virt_alloc_node_nopanic(usemapsize,
4781                                                          pgdat->node_id);
4782 }
4783 #else
4784 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4785                                 unsigned long zone_start_pfn, unsigned long zonesize) {}
4786 #endif /* CONFIG_SPARSEMEM */
4787 
4788 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4789 
4790 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4791 void __paginginit set_pageblock_order(void)
4792 {
4793         unsigned int order;
4794 
4795         /* Check that pageblock_nr_pages has not already been setup */
4796         if (pageblock_order)
4797                 return;
4798 
4799         if (HPAGE_SHIFT > PAGE_SHIFT)
4800                 order = HUGETLB_PAGE_ORDER;
4801         else
4802                 order = MAX_ORDER - 1;
4803 
4804         /*
4805          * Assume the largest contiguous order of interest is a huge page.
4806          * This value may be variable depending on boot parameters on IA64 and
4807          * powerpc.
4808          */
4809         pageblock_order = order;
4810 }
4811 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4812 
4813 /*
4814  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4815  * is unused as pageblock_order is set at compile-time. See
4816  * include/linux/pageblock-flags.h for the values of pageblock_order based on
4817  * the kernel config
4818  */
4819 void __paginginit set_pageblock_order(void)
4820 {
4821 }
4822 
4823 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4824 
4825 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4826                                                    unsigned long present_pages)
4827 {
4828         unsigned long pages = spanned_pages;
4829 
4830         /*
4831          * Provide a more accurate estimation if there are holes within
4832          * the zone and SPARSEMEM is in use. If there are holes within the
4833          * zone, each populated memory region may cost us one or two extra
4834          * memmap pages due to alignment because memmap pages for each
4835          * populated regions may not naturally algined on page boundary.
4836          * So the (present_pages >> 4) heuristic is a tradeoff for that.
4837          */
4838         if (spanned_pages > present_pages + (present_pages >> 4) &&
4839             IS_ENABLED(CONFIG_SPARSEMEM))
4840                 pages = present_pages;
4841 
4842         return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4843 }
4844 
4845 /*
4846  * Set up the zone data structures:
4847  *   - mark all pages reserved
4848  *   - mark all memory queues empty
4849  *   - clear the memory bitmaps
4850  *
4851  * NOTE: pgdat should get zeroed by caller.
4852  */
4853 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4854                 unsigned long node_start_pfn, unsigned long node_end_pfn,
4855                 unsigned long *zones_size, unsigned long *zholes_size)
4856 {
4857         enum zone_type j;
4858         int nid = pgdat->node_id;
4859         unsigned long zone_start_pfn = pgdat->node_start_pfn;
4860         int ret;
4861 
4862         pgdat_resize_init(pgdat);
4863 #ifdef CONFIG_NUMA_BALANCING
4864         spin_lock_init(&pgdat->numabalancing_migrate_lock);
4865         pgdat->numabalancing_migrate_nr_pages = 0;
4866         pgdat->numabalancing_migrate_next_window = jiffies;
4867 #endif
4868         init_waitqueue_head(&pgdat->kswapd_wait);
4869         init_waitqueue_head(&pgdat->pfmemalloc_wait);
4870         pgdat_page_ext_init(pgdat);
4871 
4872         for (j = 0; j < MAX_NR_ZONES; j++) {
4873                 struct zone *zone = pgdat->node_zones + j;
4874                 unsigned long size, realsize, freesize, memmap_pages;
4875 
4876                 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4877                                                   node_end_pfn, zones_size);
4878                 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
4879                                                                 node_start_pfn,
4880                                                                 node_end_pfn,
4881                                                                 zholes_size);
4882 
4883                 /*
4884                  * Adjust freesize so that it accounts for how much memory
4885                  * is used by this zone for memmap. This affects the watermark
4886                  * and per-cpu initialisations
4887                  */
4888                 memmap_pages = calc_memmap_size(size, realsize);
4889                 if (!is_highmem_idx(j)) {
4890                         if (freesize >= memmap_pages) {
4891                                 freesize -= memmap_pages;
4892                                 if (memmap_pages)
4893                                         printk(KERN_DEBUG
4894                                                "  %s zone: %lu pages used for memmap\n",
4895                                                zone_names[j], memmap_pages);
4896                         } else
4897                                 printk(KERN_WARNING
4898                                         "  %s zone: %lu pages exceeds freesize %lu\n",
4899                                         zone_names[j], memmap_pages, freesize);
4900                 }
4901 
4902                 /* Account for reserved pages */
4903                 if (j == 0 && freesize > dma_reserve) {
4904                         freesize -= dma_reserve;
4905                         printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
4906                                         zone_names[0], dma_reserve);
4907                 }
4908 
4909                 if (!is_highmem_idx(j))
4910                         nr_kernel_pages += freesize;
4911                 /* Charge for highmem memmap if there are enough kernel pages */
4912                 else if (nr_kernel_pages > memmap_pages * 2)
4913                         nr_kernel_pages -= memmap_pages;
4914                 nr_all_pages += freesize;
4915 
4916                 zone->spanned_pages = size;
4917                 zone->present_pages = realsize;
4918                 /*
4919                  * Set an approximate value for lowmem here, it will be adjusted
4920                  * when the bootmem allocator frees pages into the buddy system.
4921                  * And all highmem pages will be managed by the buddy system.
4922                  */
4923                 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
4924 #ifdef CONFIG_NUMA
4925                 zone->node = nid;
4926                 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
4927                                                 / 100;
4928                 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
4929 #endif
4930                 zone->name = zone_names[j];
4931                 spin_lock_init(&zone->lock);
4932                 spin_lock_init(&zone->lru_lock);
4933                 zone_seqlock_init(zone);
4934                 zone->zone_pgdat = pgdat;
4935                 zone_pcp_init(zone);
4936 
4937                 /* For bootup, initialized properly in watermark setup */
4938                 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4939 
4940                 lruvec_init(&zone->lruvec);
4941                 if (!size)
4942                         continue;
4943 
4944                 set_pageblock_order();
4945                 setup_usemap(pgdat, zone, zone_start_pfn, size);
4946                 ret = init_currently_empty_zone(zone, zone_start_pfn,
4947                                                 size, MEMMAP_EARLY);
4948                 BUG_ON(ret);
4949                 memmap_init(size, nid, j, zone_start_pfn);
4950                 zone_start_pfn += size;
4951         }
4952 }
4953 
4954 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4955 {
4956         /* Skip empty nodes */
4957         if (!pgdat->node_spanned_pages)
4958                 return;
4959 
4960 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4961         /* ia64 gets its own node_mem_map, before this, without bootmem */
4962         if (!pgdat->node_mem_map) {
4963                 unsigned long size, start, end;
4964                 struct page *map;
4965 
4966                 /*
4967                  * The zone's endpoints aren't required to be MAX_ORDER
4968                  * aligned but the node_mem_map endpoints must be in order
4969                  * for the buddy allocator to function correctly.
4970                  */
4971                 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4972                 end = pgdat_end_pfn(pgdat);
4973                 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4974                 size =  (end - start) * sizeof(struct page);
4975                 map = alloc_remap(pgdat->node_id, size);
4976                 if (!map)
4977                         map = memblock_virt_alloc_node_nopanic(size,
4978                                                                pgdat->node_id);
4979                 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4980         }
4981 #ifndef CONFIG_NEED_MULTIPLE_NODES
4982         /*
4983          * With no DISCONTIG, the global mem_map is just set as node 0's
4984          */
4985         if (pgdat == NODE_DATA(0)) {
4986                 mem_map = NODE_DATA(0)->node_mem_map;
4987 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4988                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4989                         mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4990 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4991         }
4992 #endif
4993 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4994 }
4995 
4996 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4997                 unsigned long node_start_pfn, unsigned long *zholes_size)
4998 {
4999         pg_data_t *pgdat = NODE_DATA(nid);
5000         unsigned long start_pfn = 0;
5001         unsigned long end_pfn = 0;
5002 
5003         /* pg_data_t should be reset to zero when it's allocated */
5004         WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
5005 
5006         pgdat->node_id = nid;
5007         pgdat->node_start_pfn = node_start_pfn;
5008 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5009         get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
5010         pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5011                 (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
5012 #endif
5013         calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5014                                   zones_size, zholes_size);
5015 
5016         alloc_node_mem_map(pgdat);
5017 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5018         printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5019                 nid, (unsigned long)pgdat,
5020                 (unsigned long)pgdat->node_mem_map);
5021 #endif
5022 
5023         free_area_init_core(pgdat, start_pfn, end_pfn,
5024                             zones_size, zholes_size);
5025 }
5026 
5027 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5028 
5029 #if MAX_NUMNODES > 1
5030 /*
5031  * Figure out the number of possible node ids.
5032  */
5033 void __init setup_nr_node_ids(void)
5034 {
5035         unsigned int node;
5036         unsigned int highest = 0;
5037 
5038         for_each_node_mask(node, node_possible_map)
5039                 highest = node;
5040         nr_node_ids = highest + 1;
5041 }
5042 #endif
5043 
5044 /**
5045  * node_map_pfn_alignment - determine the maximum internode alignment
5046  *
5047  * This function should be called after node map is populated and sorted.
5048  * It calculates the maximum power of two alignment which can distinguish
5049  * all the nodes.
5050  *
5051  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5052  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
5053  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
5054  * shifted, 1GiB is enough and this function will indicate so.
5055  *
5056  * This is used to test whether pfn -> nid mapping of the chosen memory
5057  * model has fine enough granularity to avoid incorrect mapping for the
5058  * populated node map.
5059  *
5060  * Returns the determined alignment in pfn's.  0 if there is no alignment
5061  * requirement (single node).
5062  */
5063 unsigned long __init node_map_pfn_alignment(void)
5064 {
5065         unsigned long accl_mask = 0, last_end = 0;
5066         unsigned long start, end, mask;
5067         int last_nid = -1;
5068         int i, nid;
5069 
5070         for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5071                 if (!start || last_nid < 0 || last_nid == nid) {
5072                         last_nid = nid;
5073                         last_end = end;
5074                         continue;
5075                 }
5076 
5077                 /*
5078                  * Start with a mask granular enough to pin-point to the
5079                  * start pfn and tick off bits one-by-one until it becomes
5080                  * too coarse to separate the current node from the last.
5081                  */
5082                 mask = ~((1 << __ffs(start)) - 1);
5083                 while (mask && last_end <= (start & (mask << 1)))
5084                         mask <<= 1;
5085 
5086                 /* accumulate all internode masks */
5087                 accl_mask |= mask;
5088         }
5089 
5090         /* convert mask to number of pages */
5091         return ~accl_mask + 1;
5092 }
5093 
5094 /* Find the lowest pfn for a node */
5095 static unsigned long __init find_min_pfn_for_node(int nid)
5096 {
5097         unsigned long min_pfn = ULONG_MAX;
5098         unsigned long start_pfn;
5099         int i;
5100 
5101         for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5102                 min_pfn = min(min_pfn, start_pfn);
5103 
5104         if (min_pfn == ULONG_MAX) {
5105                 printk(KERN_WARNING
5106                         "Could not find start_pfn for node %d\n", nid);
5107                 return 0;
5108         }
5109 
5110         return min_pfn;
5111 }
5112 
5113 /**
5114  * find_min_pfn_with_active_regions - Find the minimum PFN registered
5115  *
5116  * It returns the minimum PFN based on information provided via
5117  * memblock_set_node().
5118  */
5119 unsigned long __init find_min_pfn_with_active_regions(void)
5120 {
5121         return find_min_pfn_for_node(MAX_NUMNODES);
5122 }
5123 
5124 /*
5125  * early_calculate_totalpages()
5126  * Sum pages in active regions for movable zone.
5127  * Populate N_MEMORY for calculating usable_nodes.
5128  */
5129 static unsigned long __init early_calculate_totalpages(void)
5130 {
5131         unsigned long totalpages = 0;
5132         unsigned long start_pfn, end_pfn;
5133         int i, nid;
5134 
5135         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5136                 unsigned long pages = end_pfn - start_pfn;
5137 
5138                 totalpages += pages;
5139                 if (pages)
5140                         node_set_state(nid, N_MEMORY);
5141         }
5142         return totalpages;
5143 }
5144 
5145 /*
5146  * Find the PFN the Movable zone begins in each node. Kernel memory
5147  * is spread evenly between nodes as long as the nodes have enough
5148  * memory. When they don't, some nodes will have more kernelcore than
5149  * others
5150  */
5151 static void __init find_zone_movable_pfns_for_nodes(void)
5152 {
5153         int i, nid;
5154         unsigned long usable_startpfn;
5155         unsigned long kernelcore_node, kernelcore_remaining;
5156         /* save the state before borrow the nodemask */
5157         nodemask_t saved_node_state = node_states[N_MEMORY];
5158         unsigned long totalpages = early_calculate_totalpages();
5159         int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5160         struct memblock_region *r;
5161 
5162         /* Need to find movable_zone earlier when movable_node is specified. */
5163         find_usable_zone_for_movable();
5164 
5165         /*
5166          * If movable_node is specified, ignore kernelcore and movablecore
5167          * options.
5168          */
5169         if (movable_node_is_enabled()) {
5170                 for_each_memblock(memory, r) {
5171                         if (!memblock_is_hotpluggable(r))
5172                                 continue;
5173 
5174                         nid = r->nid;
5175 
5176                         usable_startpfn = PFN_DOWN(r->base);
5177                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5178                                 min(usable_startpfn, zone_movable_pfn[nid]) :
5179                                 usable_startpfn;
5180                 }
5181 
5182                 goto out2;
5183         }
5184 
5185         /*
5186          * If movablecore=nn[KMG] was specified, calculate what size of
5187          * kernelcore that corresponds so that memory usable for
5188          * any allocation type is evenly spread. If both kernelcore
5189          * and movablecore are specified, then the value of kernelcore
5190          * will be used for required_kernelcore if it's greater than
5191          * what movablecore would have allowed.
5192          */
5193         if (required_movablecore) {
5194                 unsigned long corepages;
5195 
5196                 /*
5197                  * Round-up so that ZONE_MOVABLE is at least as large as what
5198                  * was requested by the user
5199                  */
5200                 required_movablecore =
5201                         roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5202                 corepages = totalpages - required_movablecore;
5203 
5204                 required_kernelcore = max(required_kernelcore, corepages);
5205         }
5206 
5207         /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5208         if (!required_kernelcore)
5209                 goto out;
5210 
5211         /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5212         usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5213 
5214 restart:
5215         /* Spread kernelcore memory as evenly as possible throughout nodes */
5216         kernelcore_node = required_kernelcore / usable_nodes;
5217         for_each_node_state(nid, N_MEMORY) {
5218                 unsigned long start_pfn, end_pfn;
5219 
5220                 /*
5221                  * Recalculate kernelcore_node if the division per node
5222                  * now exceeds what is necessary to satisfy the requested
5223                  * amount of memory for the kernel
5224                  */
5225                 if (required_kernelcore < kernelcore_node)
5226                         kernelcore_node = required_kernelcore / usable_nodes;
5227 
5228                 /*
5229                  * As the map is walked, we track how much memory is usable
5230                  * by the kernel using kernelcore_remaining. When it is
5231                  * 0, the rest of the node is usable by ZONE_MOVABLE
5232                  */
5233                 kernelcore_remaining = kernelcore_node;
5234 
5235                 /* Go through each range of PFNs within this node */
5236                 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5237                         unsigned long size_pages;
5238 
5239                         start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5240                         if (start_pfn >= end_pfn)
5241                                 continue;
5242 
5243                         /* Account for what is only usable for kernelcore */
5244                         if (start_pfn < usable_startpfn) {
5245                                 unsigned long kernel_pages;
5246                                 kernel_pages = min(end_pfn, usable_startpfn)
5247                                                                 - start_pfn;
5248 
5249                                 kernelcore_remaining -= min(kernel_pages,
5250                                                         kernelcore_remaining);
5251                                 required_kernelcore -= min(kernel_pages,
5252                                                         required_kernelcore);
5253 
5254                                 /* Continue if range is now fully accounted */
5255                                 if (end_pfn <= usable_startpfn) {
5256 
5257                                         /*
5258                                          * Push zone_movable_pfn to the end so
5259                                          * that if we have to rebalance
5260                                          * kernelcore across nodes, we will
5261                                          * not double account here
5262                                          */
5263                                         zone_movable_pfn[nid] = end_pfn;
5264                                         continue;
5265                                 }
5266                                 start_pfn = usable_startpfn;
5267                         }
5268 
5269                         /*
5270                          * The usable PFN range for ZONE_MOVABLE is from
5271                          * start_pfn->end_pfn. Calculate size_pages as the
5272                          * number of pages used as kernelcore
5273                          */
5274                         size_pages = end_pfn - start_pfn;
5275                         if (size_pages > kernelcore_remaining)
5276                                 size_pages = kernelcore_remaining;
5277                         zone_movable_pfn[nid] = start_pfn + size_pages;
5278 
5279                         /*
5280                          * Some kernelcore has been met, update counts and
5281                          * break if the kernelcore for this node has been
5282                          * satisfied
5283                          */
5284                         required_kernelcore -= min(required_kernelcore,
5285                                                                 size_pages);
5286                         kernelcore_remaining -= size_pages;
5287                         if (!kernelcore_remaining)
5288                                 break;
5289                 }
5290         }
5291 
5292         /*
5293          * If there is still required_kernelcore, we do another pass with one
5294          * less node in the count. This will push zone_movable_pfn[nid] further
5295          * along on the nodes that still have memory until kernelcore is
5296          * satisfied
5297          */
5298         usable_nodes--;
5299         if (usable_nodes && required_kernelcore > usable_nodes)
5300                 goto restart;
5301 
5302 out2:
5303         /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5304         for (nid = 0; nid < MAX_NUMNODES; nid++)
5305                 zone_movable_pfn[nid] =
5306                         roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5307 
5308 out:
5309         /* restore the node_state */
5310         node_states[N_MEMORY] = saved_node_state;
5311 }
5312 
5313 /* Any regular or high memory on that node ? */
5314 static void check_for_memory(pg_data_t *pgdat, int nid)
5315 {
5316         enum zone_type zone_type;
5317 
5318         if (N_MEMORY == N_NORMAL_MEMORY)
5319                 return;
5320 
5321         for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5322                 struct zone *zone = &pgdat->node_zones[zone_type];
5323                 if (populated_zone(zone)) {
5324                         node_set_state(nid, N_HIGH_MEMORY);
5325                         if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5326                             zone_type <= ZONE_NORMAL)
5327                                 node_set_state(nid, N_NORMAL_MEMORY);
5328                         break;
5329                 }
5330         }
5331 }
5332 
5333 /**
5334  * free_area_init_nodes - Initialise all pg_data_t and zone data
5335  * @max_zone_pfn: an array of max PFNs for each zone
5336  *
5337  * This will call free_area_init_node() for each active node in the system.
5338  * Using the page ranges provided by memblock_set_node(), the size of each
5339  * zone in each node and their holes is calculated. If the maximum PFN
5340  * between two adjacent zones match, it is assumed that the zone is empty.
5341  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5342  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5343  * starts where the previous one ended. For example, ZONE_DMA32 starts
5344  * at arch_max_dma_pfn.
5345  */
5346 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5347 {
5348         unsigned long start_pfn, end_pfn;
5349         int i, nid;
5350 
5351         /* Record where the zone boundaries are */
5352         memset(arch_zone_lowest_possible_pfn, 0,
5353                                 sizeof(arch_zone_lowest_possible_pfn));
5354         memset(arch_zone_highest_possible_pfn, 0,
5355                                 sizeof(arch_zone_highest_possible_pfn));
5356         arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5357         arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5358         for (i = 1; i < MAX_NR_ZONES; i++) {
5359                 if (i == ZONE_MOVABLE)
5360                         continue;
5361                 arch_zone_lowest_possible_pfn[i] =
5362                         arch_zone_highest_possible_pfn[i-1];
5363                 arch_zone_highest_possible_pfn[i] =
5364                         max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5365         }
5366         arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5367         arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5368 
5369         /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5370         memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5371         find_zone_movable_pfns_for_nodes();
5372 
5373         /* Print out the zone ranges */
5374         pr_info("Zone ranges:\n");
5375         for (i = 0; i < MAX_NR_ZONES; i++) {
5376                 if (i == ZONE_MOVABLE)
5377                         continue;
5378                 pr_info("  %-8s ", zone_names[i]);
5379                 if (arch_zone_lowest_possible_pfn[i] ==
5380                                 arch_zone_highest_possible_pfn[i])
5381                         pr_cont("empty\n");
5382                 else
5383                         pr_cont("[mem %#018Lx-%#018Lx]\n",
5384                                 (u64)arch_zone_lowest_possible_pfn[i]
5385                                         << PAGE_SHIFT,
5386                                 ((u64)arch_zone_highest_possible_pfn[i]
5387                                         << PAGE_SHIFT) - 1);
5388         }
5389 
5390         /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5391         pr_info("Movable zone start for each node\n");
5392         for (i = 0; i < MAX_NUMNODES; i++) {
5393                 if (zone_movable_pfn[i])
5394                         pr_info("  Node %d: %#018Lx\n", i,
5395                                (u64)zone_movable_pfn[i] << PAGE_SHIFT);
5396         }
5397 
5398         /* Print out the early node map */
5399         pr_info("Early memory node ranges\n");
5400         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5401                 pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5402                         (u64)start_pfn << PAGE_SHIFT,
5403                         ((u64)end_pfn << PAGE_SHIFT) - 1);
5404 
5405         /* Initialise every node */
5406         mminit_verify_pageflags_layout();
5407         setup_nr_node_ids();
5408         for_each_online_node(nid) {
5409                 pg_data_t *pgdat = NODE_DATA(nid);
5410                 free_area_init_node(nid, NULL,
5411                                 find_min_pfn_for_node(nid), NULL);
5412 
5413                 /* Any memory on that node */
5414                 if (pgdat->node_present_pages)
5415                         node_set_state(nid, N_MEMORY);
5416                 check_for_memory(pgdat, nid);
5417         }
5418 }
5419 
5420 static int __init cmdline_parse_core(char *p, unsigned long *core)
5421 {
5422         unsigned long long coremem;
5423         if (!p)
5424                 return -EINVAL;
5425 
5426         coremem = memparse(p, &p);
5427         *core = coremem >> PAGE_SHIFT;
5428 
5429         /* Paranoid check that UL is enough for the coremem value */
5430         WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5431 
5432         return 0;
5433 }
5434 
5435 /*
5436  * kernelcore=size sets the amount of memory for use for allocations that
5437  * cannot be reclaimed or migrated.
5438  */
5439 static int __init cmdline_parse_kernelcore(char *p)
5440 {
5441         return cmdline_parse_core(p, &required_kernelcore);
5442 }
5443 
5444 /*
5445  * movablecore=size sets the amount of memory for use for allocations that
5446  * can be reclaimed or migrated.
5447  */
5448 static int __init cmdline_parse_movablecore(char *p)
5449 {
5450         return cmdline_parse_core(p, &required_movablecore);
5451 }
5452 
5453 early_param("kernelcore", cmdline_parse_kernelcore);
5454 early_param("movablecore", cmdline_parse_movablecore);
5455 
5456 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5457 
5458 void adjust_managed_page_count(struct page *page, long count)
5459 {
5460         spin_lock(&managed_page_count_lock);
5461         page_zone(page)->managed_pages += count;
5462         totalram_pages += count;
5463 #ifdef CONFIG_HIGHMEM
5464         if (PageHighMem(page))
5465                 totalhigh_pages += count;
5466 #endif
5467         spin_unlock(&managed_page_count_lock);
5468 }
5469 EXPORT_SYMBOL(adjust_managed_page_count);
5470 
5471 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
5472 {
5473         void *pos;
5474         unsigned long pages = 0;
5475 
5476         start = (void *)PAGE_ALIGN((unsigned long)start);
5477         end = (void *)((unsigned long)end & PAGE_MASK);
5478         for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
5479                 if ((unsigned int)poison <= 0xFF)
5480                         memset(pos, poison, PAGE_SIZE);
5481                 free_reserved_page(virt_to_page(pos));
5482         }
5483 
5484         if (pages && s)
5485                 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5486                         s, pages << (PAGE_SHIFT - 10), start, end);
5487 
5488         return pages;
5489 }
5490 EXPORT_SYMBOL(free_reserved_area);
5491 
5492 #ifdef  CONFIG_HIGHMEM
5493 void free_highmem_page(struct page *page)
5494 {
5495         __free_reserved_page(page);
5496         totalram_pages++;
5497         page_zone(page)->managed_pages++;
5498         totalhigh_pages++;
5499 }
5500 #endif
5501 
5502 
5503 void __init mem_init_print_info(const char *str)
5504 {
5505         unsigned long physpages, codesize, datasize, rosize, bss_size;
5506         unsigned long init_code_size, init_data_size;
5507 
5508         physpages = get_num_physpages();
5509         codesize = _etext - _stext;
5510         datasize = _edata - _sdata;
5511         rosize = __end_rodata - __start_rodata;
5512         bss_size = __bss_stop - __bss_start;
5513         init_data_size = __init_end - __init_begin;
5514         init_code_size = _einittext - _sinittext;
5515 
5516         /*
5517          * Detect special cases and adjust section sizes accordingly:
5518          * 1) .init.* may be embedded into .data sections
5519          * 2) .init.text.* may be out of [__init_begin, __init_end],
5520          *    please refer to arch/tile/kernel/vmlinux.lds.S.
5521          * 3) .rodata.* may be embedded into .text or .data sections.
5522          */
5523 #define adj_init_size(start, end, size, pos, adj) \
5524         do { \
5525                 if (start <= pos && pos < end && size > adj) \
5526                         size -= adj; \
5527         } while (0)
5528 
5529         adj_init_size(__init_begin, __init_end, init_data_size,
5530                      _sinittext, init_code_size);
5531         adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5532         adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5533         adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5534         adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5535 
5536 #undef  adj_init_size
5537 
5538         pr_info("Memory: %luK/%luK available "
5539                "(%luK kernel code, %luK rwdata, %luK rodata, "
5540                "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
5541 #ifdef  CONFIG_HIGHMEM
5542                ", %luK highmem"
5543 #endif
5544                "%s%s)\n",
5545                nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5546                codesize >> 10, datasize >> 10, rosize >> 10,
5547                (init_data_size + init_code_size) >> 10, bss_size >> 10,
5548                (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5549                totalcma_pages << (PAGE_SHIFT-10),
5550 #ifdef  CONFIG_HIGHMEM
5551                totalhigh_pages << (PAGE_SHIFT-10),
5552 #endif
5553                str ? ", " : "", str ? str : "");
5554 }
5555 
5556 /**
5557  * set_dma_reserve - set the specified number of pages reserved in the first zone
5558  * @new_dma_reserve: The number of pages to mark reserved
5559  *
5560  * The per-cpu batchsize and zone watermarks are determined by present_pages.
5561  * In the DMA zone, a significant percentage may be consumed by kernel image
5562  * and other unfreeable allocations which can skew the watermarks badly. This
5563  * function may optionally be used to account for unfreeable pages in the
5564  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5565  * smaller per-cpu batchsize.
5566  */
5567 void __init set_dma_reserve(unsigned long new_dma_reserve)
5568 {
5569         dma_reserve = new_dma_reserve;
5570 }
5571 
5572 void __init free_area_init(unsigned long *zones_size)
5573 {
5574         free_area_init_node(0, zones_size,
5575                         __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5576 }
5577 
5578 static int page_alloc_cpu_notify(struct notifier_block *self,
5579                                  unsigned long action, void *hcpu)
5580 {
5581         int cpu = (unsigned long)hcpu;
5582 
5583         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
5584                 lru_add_drain_cpu(cpu);
5585                 drain_pages(cpu);
5586 
5587                 /*
5588                  * Spill the event counters of the dead processor
5589                  * into the current processors event counters.
5590                  * This artificially elevates the count of the current
5591                  * processor.
5592                  */
5593                 vm_events_fold_cpu(cpu);
5594 
5595                 /*
5596                  * Zero the differential counters of the dead processor
5597                  * so that the vm statistics are consistent.
5598                  *
5599                  * This is only okay since the processor is dead and cannot
5600                  * race with what we are doing.
5601                  */
5602                 cpu_vm_stats_fold(cpu);
5603         }
5604         return NOTIFY_OK;
5605 }
5606 
5607 void __init page_alloc_init(void)
5608 {
5609         hotcpu_notifier(page_alloc_cpu_notify, 0);
5610 }
5611 
5612 /*
5613  * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5614  *      or min_free_kbytes changes.
5615  */
5616 static void calculate_totalreserve_pages(void)
5617 {
5618         struct pglist_data *pgdat;
5619         unsigned long reserve_pages = 0;
5620         enum zone_type i, j;
5621 
5622         for_each_online_pgdat(pgdat) {
5623                 for (i = 0; i < MAX_NR_ZONES; i++) {
5624                         struct zone *zone = pgdat->node_zones + i;
5625                         long max = 0;
5626 
5627                         /* Find valid and maximum lowmem_reserve in the zone */
5628                         for (j = i; j < MAX_NR_ZONES; j++) {
5629                                 if (zone->lowmem_reserve[j] > max)
5630                                         max = zone->lowmem_reserve[j];
5631                         }
5632 
5633                         /* we treat the high watermark as reserved pages. */
5634                         max += high_wmark_pages(zone);
5635 
5636                         if (max > zone->managed_pages)
5637                                 max = zone->managed_pages;
5638                         reserve_pages += max;
5639                         /*
5640                          * Lowmem reserves are not available to
5641                          * GFP_HIGHUSER page cache allocations and
5642                          * kswapd tries to balance zones to their high
5643                          * watermark.  As a result, neither should be
5644                          * regarded as dirtyable memory, to prevent a
5645                          * situation where reclaim has to clean pages
5646                          * in order to balance the zones.
5647                          */
5648                         zone->dirty_balance_reserve = max;
5649                 }
5650         }
5651         dirty_balance_reserve = reserve_pages;
5652         totalreserve_pages = reserve_pages;
5653 }
5654 
5655 /*
5656  * setup_per_zone_lowmem_reserve - called whenever
5657  *      sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
5658  *      has a correct pages reserved value, so an adequate number of
5659  *      pages are left in the zone after a successful __alloc_pages().
5660  */
5661 static void setup_per_zone_lowmem_reserve(void)
5662 {
5663         struct pglist_data *pgdat;
5664         enum zone_type j, idx;
5665 
5666         for_each_online_pgdat(pgdat) {
5667                 for (j = 0; j < MAX_NR_ZONES; j++) {
5668                         struct zone *zone = pgdat->node_zones + j;
5669                         unsigned long managed_pages = zone->managed_pages;
5670 
5671                         zone->lowmem_reserve[j] = 0;
5672 
5673                         idx = j;
5674                         while (idx) {
5675                                 struct zone *lower_zone;
5676 
5677                                 idx--;
5678 
5679                                 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5680                                         sysctl_lowmem_reserve_ratio[idx] = 1;
5681 
5682                                 lower_zone = pgdat->node_zones + idx;
5683                                 lower_zone->lowmem_reserve[j] = managed_pages /
5684                                         sysctl_lowmem_reserve_ratio[idx];
5685                                 managed_pages += lower_zone->managed_pages;
5686                         }
5687                 }
5688         }
5689 
5690         /* update totalreserve_pages */
5691         calculate_totalreserve_pages();
5692 }
5693 
5694 static void __setup_per_zone_wmarks(void)
5695 {
5696         unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5697         unsigned long lowmem_pages = 0;
5698         struct zone *zone;
5699         unsigned long flags;
5700 
5701         /* Calculate total number of !ZONE_HIGHMEM pages */
5702         for_each_zone(zone) {
5703                 if (!is_highmem(zone))
5704                         lowmem_pages += zone->managed_pages;
5705         }
5706 
5707         for_each_zone(zone) {
5708                 u64 tmp;
5709 
5710                 spin_lock_irqsave(&zone->lock, flags);
5711                 tmp = (u64)pages_min * zone->managed_pages;
5712                 do_div(tmp, lowmem_pages);
5713                 if (is_highmem(zone)) {
5714                         /*
5715                          * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5716                          * need highmem pages, so cap pages_min to a small
5717                          * value here.
5718                          *
5719                          * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5720                          * deltas controls asynch page reclaim, and so should
5721                          * not be capped for highmem.
5722                          */
5723                         unsigned long min_pages;
5724 
5725                         min_pages = zone->managed_pages / 1024;
5726                         min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
5727                         zone->watermark[WMARK_MIN] = min_pages;
5728                 } else {
5729                         /*
5730                          * If it's a lowmem zone, reserve a number of pages
5731                          * proportionate to the zone's size.
5732                          */
5733                         zone->watermark[WMARK_MIN] = tmp;
5734                 }
5735 
5736                 zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + (tmp >> 2);
5737                 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
5738 
5739                 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
5740                         high_wmark_pages(zone) - low_wmark_pages(zone) -
5741                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
5742 
5743                 setup_zone_migrate_reserve(zone);
5744                 spin_unlock_irqrestore(&zone->lock, flags);
5745         }
5746 
5747         /* update totalreserve_pages */
5748         calculate_totalreserve_pages();
5749 }
5750 
5751 /**
5752  * setup_per_zone_wmarks - called when min_free_kbytes changes
5753  * or when memory is hot-{added|removed}
5754  *
5755  * Ensures that the watermark[min,low,high] values for each zone are set
5756  * correctly with respect to min_free_kbytes.
5757  */
5758 void setup_per_zone_wmarks(void)
5759 {
5760         mutex_lock(&zonelists_mutex);
5761         __setup_per_zone_wmarks();
5762         mutex_unlock(&zonelists_mutex);
5763 }
5764 
5765 /*
5766  * The inactive anon list should be small enough that the VM never has to
5767  * do too much work, but large enough that each inactive page has a chance
5768  * to be referenced again before it is swapped out.
5769  *
5770  * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5771  * INACTIVE_ANON pages on this zone's LRU, maintained by the
5772  * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5773  * the anonymous pages are kept on the inactive list.
5774  *
5775  * total     target    max
5776  * memory    ratio     inactive anon
5777  * -------------------------------------
5778  *   10MB       1         5MB
5779  *  100MB       1        50MB
5780  *    1GB       3       250MB
5781  *   10GB      10       0.9GB
5782  *  100GB      31         3GB
5783  *    1TB     101        10GB
5784  *   10TB     320        32GB
5785  */
5786 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
5787 {
5788         unsigned int gb, ratio;
5789 
5790         /* Zone size in gigabytes */
5791         gb = zone->managed_pages >> (30 - PAGE_SHIFT);
5792         if (gb)
5793                 ratio = int_sqrt(10 * gb);
5794         else
5795                 ratio = 1;
5796 
5797         zone->inactive_ratio = ratio;
5798 }
5799 
5800 static void __meminit setup_per_zone_inactive_ratio(void)
5801 {
5802         struct zone *zone;
5803 
5804         for_each_zone(zone)
5805                 calculate_zone_inactive_ratio(zone);
5806 }
5807 
5808 /*
5809  * Initialise min_free_kbytes.
5810  *
5811  * For small machines we want it small (128k min).  For large machines
5812  * we want it large (64MB max).  But it is not linear, because network
5813  * bandwidth does not increase linearly with machine size.  We use
5814  *
5815  *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5816  *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
5817  *
5818  * which yields
5819  *
5820  * 16MB:        512k
5821  * 32MB:        724k
5822  * 64MB:        1024k
5823  * 128MB:       1448k
5824  * 256MB:       2048k
5825  * 512MB:       2896k
5826  * 1024MB:      4096k
5827  * 2048MB:      5792k
5828  * 4096MB:      8192k
5829  * 8192MB:      11584k
5830  * 16384MB:     16384k
5831  */
5832 int __meminit init_per_zone_wmark_min(void)
5833 {
5834         unsigned long lowmem_kbytes;
5835         int new_min_free_kbytes;
5836 
5837         lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5838         new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5839 
5840         if (new_min_free_kbytes > user_min_free_kbytes) {
5841                 min_free_kbytes = new_min_free_kbytes;
5842                 if (min_free_kbytes < 128)
5843                         min_free_kbytes = 128;
5844                 if (min_free_kbytes > 65536)
5845                         min_free_kbytes = 65536;
5846         } else {
5847                 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5848                                 new_min_free_kbytes, user_min_free_kbytes);
5849         }
5850         setup_per_zone_wmarks();
5851         refresh_zone_stat_thresholds();
5852         setup_per_zone_lowmem_reserve();
5853         setup_per_zone_inactive_ratio();
5854         return 0;
5855 }
5856 module_init(init_per_zone_wmark_min)
5857 
5858 /*
5859  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5860  *      that we can call two helper functions whenever min_free_kbytes
5861  *      changes.
5862  */
5863 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
5864         void __user *buffer, size_t *length, loff_t *ppos)
5865 {
5866         int rc;
5867 
5868         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5869         if (rc)
5870                 return rc;
5871 
5872         if (write) {
5873                 user_min_free_kbytes = min_free_kbytes;
5874                 setup_per_zone_wmarks();
5875         }
5876         return 0;
5877 }
5878 
5879 #ifdef CONFIG_NUMA
5880 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
5881         void __user *buffer, size_t *length, loff_t *ppos)
5882 {
5883         struct zone *zone;
5884         int rc;
5885 
5886         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5887         if (rc)
5888                 return rc;
5889 
5890         for_each_zone(zone)
5891                 zone->min_unmapped_pages = (zone->managed_pages *
5892                                 sysctl_min_unmapped_ratio) / 100;
5893         return 0;
5894 }
5895 
5896 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
5897         void __user *buffer, size_t *length, loff_t *ppos)
5898 {
5899         struct zone *zone;
5900         int rc;
5901 
5902         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5903         if (rc)
5904                 return rc;
5905 
5906         for_each_zone(zone)
5907                 zone->min_slab_pages = (zone->managed_pages *
5908                                 sysctl_min_slab_ratio) / 100;
5909         return 0;
5910 }
5911 #endif
5912 
5913 /*
5914  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5915  *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5916  *      whenever sysctl_lowmem_reserve_ratio changes.
5917  *
5918  * The reserve ratio obviously has absolutely no relation with the
5919  * minimum watermarks. The lowmem reserve ratio can only make sense
5920  * if in function of the boot time zone sizes.
5921  */
5922 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
5923         void __user *buffer, size_t *length, loff_t *ppos)
5924 {
5925         proc_dointvec_minmax(table, write, buffer, length, ppos);
5926         setup_per_zone_lowmem_reserve();
5927         return 0;
5928 }
5929 
5930 /*
5931  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5932  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
5933  * pagelist can have before it gets flushed back to buddy allocator.
5934  */
5935 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
5936         void __user *buffer, size_t *length, loff_t *ppos)
5937 {
5938         struct zone *zone;
5939         int old_percpu_pagelist_fraction;
5940         int ret;
5941 
5942         mutex_lock(&pcp_batch_high_lock);
5943         old_percpu_pagelist_fraction = percpu_pagelist_fraction;
5944 
5945         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5946         if (!write || ret < 0)
5947                 goto out;
5948 
5949         /* Sanity checking to avoid pcp imbalance */
5950         if (percpu_pagelist_fraction &&
5951             percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
5952                 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
5953                 ret = -EINVAL;
5954                 goto out;
5955         }
5956 
5957         /* No change? */
5958         if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
5959                 goto out;
5960 
5961         for_each_populated_zone(zone) {
5962                 unsigned int cpu;
5963 
5964                 for_each_possible_cpu(cpu)
5965                         pageset_set_high_and_batch(zone,
5966                                         per_cpu_ptr(zone->pageset, cpu));
5967         }
5968 out:
5969         mutex_unlock(&pcp_batch_high_lock);
5970         return ret;
5971 }
5972 
5973 int hashdist = HASHDIST_DEFAULT;
5974 
5975 #ifdef CONFIG_NUMA
5976 static int __init set_hashdist(char *str)
5977 {
5978         if (!str)
5979                 return 0;
5980         hashdist = simple_strtoul(str, &str, 0);
5981         return 1;
5982 }
5983 __setup("hashdist=", set_hashdist);
5984 #endif
5985 
5986 /*
5987  * allocate a large system hash table from bootmem
5988  * - it is assumed that the hash table must contain an exact power-of-2
5989  *   quantity of entries
5990  * - limit is the number of hash buckets, not the total allocation size
5991  */
5992 void *__init alloc_large_system_hash(const char *tablename,
5993                                      unsigned long bucketsize,
5994                                      unsigned long numentries,
5995                                      int scale,
5996                                      int flags,
5997                                      unsigned int *_hash_shift,
5998                                      unsigned int *_hash_mask,
5999                                      unsigned long low_limit,
6000                                      unsigned long high_limit)
6001 {
6002         unsigned long long max = high_limit;
6003         unsigned long log2qty, size;
6004         void *table = NULL;
6005 
6006         /* allow the kernel cmdline to have a say */
6007         if (!numentries) {
6008                 /* round applicable memory size up to nearest megabyte */
6009                 numentries = nr_kernel_pages;
6010 
6011                 /* It isn't necessary when PAGE_SIZE >= 1MB */
6012                 if (PAGE_SHIFT < 20)
6013                         numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
6014 
6015                 /* limit to 1 bucket per 2^scale bytes of low memory */
6016                 if (scale > PAGE_SHIFT)
6017                         numentries >>= (scale - PAGE_SHIFT);
6018                 else
6019                         numentries <<= (PAGE_SHIFT - scale);
6020 
6021                 /* Make sure we've got at least a 0-order allocation.. */
6022                 if (unlikely(flags & HASH_SMALL)) {
6023                         /* Makes no sense without HASH_EARLY */
6024                         WARN_ON(!(flags & HASH_EARLY));
6025                         if (!(numentries >> *_hash_shift)) {
6026                                 numentries = 1UL << *_hash_shift;
6027                                 BUG_ON(!numentries);
6028                         }
6029                 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
6030                         numentries = PAGE_SIZE / bucketsize;
6031         }
6032         numentries = roundup_pow_of_two(numentries);
6033 
6034         /* limit allocation size to 1/16 total memory by default */
6035         if (max == 0) {
6036                 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6037                 do_div(max, bucketsize);
6038         }
6039         max = min(max, 0x80000000ULL);
6040 
6041         if (numentries < low_limit)
6042                 numentries = low_limit;
6043         if (numentries > max)
6044                 numentries = max;
6045 
6046         log2qty = ilog2(numentries);
6047 
6048         do {
6049                 size = bucketsize << log2qty;
6050                 if (flags & HASH_EARLY)
6051                         table = memblock_virt_alloc_nopanic(size, 0);
6052                 else if (hashdist)
6053                         table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6054                 else {
6055                         /*
6056                          * If bucketsize is not a power-of-two, we may free
6057                          * some pages at the end of hash table which
6058                          * alloc_pages_exact() automatically does
6059                          */
6060                         if (get_order(size) < MAX_ORDER) {
6061                                 table = alloc_pages_exact(size, GFP_ATOMIC);
6062                                 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6063                         }
6064                 }
6065         } while (!table && size > PAGE_SIZE && --log2qty);
6066 
6067         if (!table)
6068                 panic("Failed to allocate %s hash table\n", tablename);
6069 
6070         printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
6071                tablename,
6072                (1UL << log2qty),
6073                ilog2(size) - PAGE_SHIFT,
6074                size);
6075 
6076         if (_hash_shift)
6077                 *_hash_shift = log2qty;
6078         if (_hash_mask)
6079                 *_hash_mask = (1 << log2qty) - 1;
6080 
6081         return table;
6082 }
6083 
6084 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6085 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6086                                                         unsigned long pfn)
6087 {
6088 #ifdef CONFIG_SPARSEMEM
6089         return __pfn_to_section(pfn)->pageblock_flags;
6090 #else
6091         return zone->pageblock_flags;
6092 #endif /* CONFIG_SPARSEMEM */
6093 }
6094 
6095 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6096 {
6097 #ifdef CONFIG_SPARSEMEM
6098         pfn &= (PAGES_PER_SECTION-1);
6099         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6100 #else
6101         pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
6102         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6103 #endif /* CONFIG_SPARSEMEM */
6104 }
6105 
6106 /**
6107  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6108  * @page: The page within the block of interest
6109  * @pfn: The target page frame number
6110  * @end_bitidx: The last bit of interest to retrieve
6111  * @mask: mask of bits that the caller is interested in
6112  *
6113  * Return: pageblock_bits flags
6114  */
6115 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
6116                                         unsigned long end_bitidx,
6117                                         unsigned long mask)
6118 {
6119         struct zone *zone;
6120         unsigned long *bitmap;
6121         unsigned long bitidx, word_bitidx;
6122         unsigned long word;
6123 
6124         zone = page_zone(page);
6125         bitmap = get_pageblock_bitmap(zone, pfn);
6126         bitidx = pfn_to_bitidx(zone, pfn);
6127         word_bitidx = bitidx / BITS_PER_LONG;
6128         bitidx &= (BITS_PER_LONG-1);
6129 
6130         word = bitmap[word_bitidx];
6131         bitidx += end_bitidx;
6132         return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
6133 }
6134 
6135 /**
6136  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6137  * @page: The page within the block of interest
6138  * @flags: The flags to set
6139  * @pfn: The target page frame number
6140  * @end_bitidx: The last bit of interest
6141  * @mask: mask of bits that the caller is interested in
6142  */
6143 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6144                                         unsigned long pfn,
6145                                         unsigned long end_bitidx,
6146                                         unsigned long mask)
6147 {
6148         struct zone *zone;
6149         unsigned long *bitmap;
6150         unsigned long bitidx, word_bitidx;
6151         unsigned long old_word, word;
6152 
6153         BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
6154 
6155         zone = page_zone(page);
6156         bitmap = get_pageblock_bitmap(zone, pfn);
6157         bitidx = pfn_to_bitidx(zone, pfn);
6158         word_bitidx = bitidx / BITS_PER_LONG;
6159         bitidx &= (BITS_PER_LONG-1);
6160 
6161         VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
6162 
6163         bitidx += end_bitidx;
6164         mask <<= (BITS_PER_LONG - bitidx - 1);
6165         flags <<= (BITS_PER_LONG - bitidx - 1);
6166 
6167         word = ACCESS_ONCE(bitmap[word_bitidx]);
6168         for (;;) {
6169                 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6170                 if (word == old_word)
6171                         break;
6172                 word = old_word;
6173         }
6174 }
6175 
6176 /*
6177  * This function checks whether pageblock includes unmovable pages or not.
6178  * If @count is not zero, it is okay to include less @count unmovable pages
6179  *
6180  * PageLRU check without isolation or lru_lock could race so that
6181  * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6182  * expect this function should be exact.
6183  */
6184 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6185                          bool skip_hwpoisoned_pages)
6186 {
6187         unsigned long pfn, iter, found;
6188         int mt;
6189 
6190         /*
6191          * For avoiding noise data, lru_add_drain_all() should be called
6192          * If ZONE_MOVABLE, the zone never contains unmovable pages
6193          */
6194         if (zone_idx(zone) == ZONE_MOVABLE)
6195                 return false;
6196         mt = get_pageblock_migratetype(page);
6197         if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6198                 return false;
6199 
6200         pfn = page_to_pfn(page);
6201         for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6202                 unsigned long check = pfn + iter;
6203 
6204                 if (!pfn_valid_within(check))
6205                         continue;
6206 
6207                 page = pfn_to_page(check);
6208 
6209                 /*
6210                  * Hugepages are not in LRU lists, but they're movable.
6211                  * We need not scan over tail pages bacause we don't
6212                  * handle each tail page individually in migration.
6213                  */
6214                 if (PageHuge(page)) {
6215                         iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6216                         continue;
6217                 }
6218 
6219                 /*
6220                  * We can't use page_count without pin a page
6221                  * because another CPU can free compound page.
6222                  * This check already skips compound tails of THP
6223                  * because their page->_count is zero at all time.
6224                  */
6225                 if (!atomic_read(&page->_count)) {
6226                         if (PageBuddy(page))
6227                                 iter += (1 << page_order(page)) - 1;
6228                         continue;
6229                 }
6230 
6231                 /*
6232                  * The HWPoisoned page may be not in buddy system, and
6233                  * page_count() is not 0.
6234                  */
6235                 if (skip_hwpoisoned_pages && PageHWPoison(page))
6236                         continue;
6237 
6238                 if (!PageLRU(page))
6239                         found++;
6240                 /*
6241                  * If there are RECLAIMABLE pages, we need to check
6242                  * it.  But now, memory offline itself doesn't call
6243                  * shrink_node_slabs() and it still to be fixed.
6244                  */
6245                 /*
6246                  * If the page is not RAM, page_count()should be 0.
6247                  * we don't need more check. This is an _used_ not-movable page.
6248                  *
6249                  * The problematic thing here is PG_reserved pages. PG_reserved
6250                  * is set to both of a memory hole page and a _used_ kernel
6251                  * page at boot.
6252                  */
6253                 if (found > count)
6254                         return true;
6255         }
6256         return false;
6257 }
6258 
6259 bool is_pageblock_removable_nolock(struct page *page)
6260 {
6261         struct zone *zone;
6262         unsigned long pfn;
6263 
6264         /*
6265          * We have to be careful here because we are iterating over memory
6266          * sections which are not zone aware so we might end up outside of
6267          * the zone but still within the section.
6268          * We have to take care about the node as well. If the node is offline
6269          * its NODE_DATA will be NULL - see page_zone.
6270          */
6271         if (!node_online(page_to_nid(page)))
6272                 return false;
6273 
6274         zone = page_zone(page);
6275         pfn = page_to_pfn(page);
6276         if (!zone_spans_pfn(zone, pfn))
6277                 return false;
6278 
6279         return !has_unmovable_pages(zone, page, 0, true);
6280 }
6281 
6282 #ifdef CONFIG_CMA
6283 
6284 static unsigned long pfn_max_align_down(unsigned long pfn)
6285 {
6286         return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6287                              pageblock_nr_pages) - 1);
6288 }
6289 
6290 static unsigned long pfn_max_align_up(unsigned long pfn)
6291 {
6292         return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6293                                 pageblock_nr_pages));
6294 }
6295 
6296 /* [start, end) must belong to a single zone. */
6297 static int __alloc_contig_migrate_range(struct compact_control *cc,
6298                                         unsigned long start, unsigned long end)
6299 {
6300         /* This function is based on compact_zone() from compaction.c. */
6301         unsigned long nr_reclaimed;
6302         unsigned long pfn = start;
6303         unsigned int tries = 0;
6304         int ret = 0;
6305 
6306         migrate_prep();
6307 
6308         while (pfn < end || !list_empty(&cc->migratepages)) {
6309                 if (fatal_signal_pending(current)) {
6310                         ret = -EINTR;
6311                         break;
6312                 }
6313 
6314                 if (list_empty(&cc->migratepages)) {
6315                         cc->nr_migratepages = 0;
6316                         pfn = isolate_migratepages_range(cc, pfn, end);
6317                         if (!pfn) {
6318                                 ret = -EINTR;
6319                                 break;
6320                         }
6321                         tries = 0;
6322                 } else if (++tries == 5) {
6323                         ret = ret < 0 ? ret : -EBUSY;
6324                         break;
6325                 }
6326 
6327                 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6328                                                         &cc->migratepages);
6329                 cc->nr_migratepages -= nr_reclaimed;
6330 
6331                 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6332                                     NULL, 0, cc->mode, MR_CMA);
6333         }
6334         if (ret < 0) {
6335                 putback_movable_pages(&cc->migratepages);
6336                 return ret;
6337         }
6338         return 0;
6339 }
6340 
6341 /**
6342  * alloc_contig_range() -- tries to allocate given range of pages
6343  * @start:      start PFN to allocate
6344  * @end:        one-past-the-last PFN to allocate
6345  * @migratetype:        migratetype of the underlaying pageblocks (either
6346  *                      #MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
6347  *                      in range must have the same migratetype and it must
6348  *                      be either of the two.
6349  *
6350  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6351  * aligned, however it's the caller's responsibility to guarantee that
6352  * we are the only thread that changes migrate type of pageblocks the
6353  * pages fall in.
6354  *
6355  * The PFN range must belong to a single zone.
6356  *
6357  * Returns zero on success or negative error code.  On success all
6358  * pages which PFN is in [start, end) are allocated for the caller and
6359  * need to be freed with free_contig_range().
6360  */
6361 int alloc_contig_range(unsigned long start, unsigned long end,
6362                        unsigned migratetype)
6363 {
6364         unsigned long outer_start, outer_end;
6365         int ret = 0, order;
6366 
6367         struct compact_control cc = {
6368                 .nr_migratepages = 0,
6369                 .order = -1,
6370                 .zone = page_zone(pfn_to_page(start)),
6371                 .mode = MIGRATE_SYNC,
6372                 .ignore_skip_hint = true,
6373         };
6374         INIT_LIST_HEAD(&cc.migratepages);
6375 
6376         /*
6377          * What we do here is we mark all pageblocks in range as
6378          * MIGRATE_ISOLATE.  Because pageblock and max order pages may
6379          * have different sizes, and due to the way page allocator
6380          * work, we align the range to biggest of the two pages so
6381          * that page allocator won't try to merge buddies from
6382          * different pageblocks and change MIGRATE_ISOLATE to some
6383          * other migration type.
6384          *
6385          * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6386          * migrate the pages from an unaligned range (ie. pages that
6387          * we are interested in).  This will put all the pages in
6388          * range back to page allocator as MIGRATE_ISOLATE.
6389          *
6390          * When this is done, we take the pages in range from page
6391          * allocator removing them from the buddy system.  This way
6392          * page allocator will never consider using them.
6393          *
6394          * This lets us mark the pageblocks back as
6395          * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6396          * aligned range but not in the unaligned, original range are
6397          * put back to page allocator so that buddy can use them.
6398          */
6399 
6400         ret = start_isolate_page_range(pfn_max_align_down(start),
6401                                        pfn_max_align_up(end), migratetype,
6402                                        false);
6403         if (ret)
6404                 return ret;
6405 
6406         ret = __alloc_contig_migrate_range(&cc, start, end);
6407         if (ret)
6408                 goto done;
6409 
6410         /*
6411          * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6412          * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
6413          * more, all pages in [start, end) are free in page allocator.
6414          * What we are going to do is to allocate all pages from
6415          * [start, end) (that is remove them from page allocator).
6416          *
6417          * The only problem is that pages at the beginning and at the
6418          * end of interesting range may be not aligned with pages that
6419          * page allocator holds, ie. they can be part of higher order
6420          * pages.  Because of this, we reserve the bigger range and
6421          * once this is done free the pages we are not interested in.
6422          *
6423          * We don't have to hold zone->lock here because the pages are
6424          * isolated thus they won't get removed from buddy.
6425          */
6426 
6427         lru_add_drain_all();
6428         drain_all_pages(cc.zone);
6429 
6430         order = 0;
6431         outer_start = start;
6432         while (!PageBuddy(pfn_to_page(outer_start))) {
6433                 if (++order >= MAX_ORDER) {
6434                         ret = -EBUSY;
6435                         goto done;
6436                 }
6437                 outer_start &= ~0UL << order;
6438         }
6439 
6440         /* Make sure the range is really isolated. */
6441         if (test_pages_isolated(outer_start, end, false)) {
6442                 pr_info("%s: [%lx, %lx) PFNs busy\n",
6443                         __func__, outer_start, end);
6444                 ret = -EBUSY;
6445                 goto done;
6446         }
6447 
6448         /* Grab isolated pages from freelists. */
6449         outer_end = isolate_freepages_range(&cc, outer_start, end);
6450         if (!outer_end) {
6451                 ret = -EBUSY;
6452                 goto done;
6453         }
6454 
6455         /* Free head and tail (if any) */
6456         if (start != outer_start)
6457                 free_contig_range(outer_start, start - outer_start);
6458         if (end != outer_end)
6459                 free_contig_range(end, outer_end - end);
6460 
6461 done:
6462         undo_isolate_page_range(pfn_max_align_down(start),
6463                                 pfn_max_align_up(end), migratetype);
6464         return ret;
6465 }
6466 
6467 void free_contig_range(unsigned long pfn, unsigned nr_pages)
6468 {
6469         unsigned int count = 0;
6470 
6471         for (; nr_pages--; pfn++) {
6472                 struct page *page = pfn_to_page(pfn);
6473 
6474                 count += page_count(page) != 1;
6475                 __free_page(page);
6476         }
6477         WARN(count != 0, "%d pages are still in use!\n", count);
6478 }
6479 #endif
6480 
6481 #ifdef CONFIG_MEMORY_HOTPLUG
6482 /*
6483  * The zone indicated has a new number of managed_pages; batch sizes and percpu
6484  * page high values need to be recalulated.
6485  */
6486 void __meminit zone_pcp_update(struct zone *zone)
6487 {
6488         unsigned cpu;
6489         mutex_lock(&pcp_batch_high_lock);
6490         for_each_possible_cpu(cpu)
6491                 pageset_set_high_and_batch(zone,
6492                                 per_cpu_ptr(zone->pageset, cpu));
6493         mutex_unlock(&pcp_batch_high_lock);
6494 }
6495 #endif
6496 
6497 void zone_pcp_reset(struct zone *zone)
6498 {
6499         unsigned long flags;
6500         int cpu;
6501         struct per_cpu_pageset *pset;
6502 
6503         /* avoid races with drain_pages()  */
6504         local_irq_save(flags);
6505         if (zone->pageset != &boot_pageset) {
6506                 for_each_online_cpu(cpu) {
6507                         pset = per_cpu_ptr(zone->pageset, cpu);
6508                         drain_zonestat(zone, pset);
6509                 }
6510                 free_percpu(zone->pageset);
6511                 zone->pageset = &boot_pageset;
6512         }
6513         local_irq_restore(flags);
6514 }
6515 
6516 #ifdef CONFIG_MEMORY_HOTREMOVE
6517 /*
6518  * All pages in the range must be isolated before calling this.
6519  */
6520 void
6521 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6522 {
6523         struct page *page;
6524         struct zone *zone;
6525         unsigned int order, i;
6526         unsigned long pfn;
6527         unsigned long flags;
6528         /* find the first valid pfn */
6529         for (pfn = start_pfn; pfn < end_pfn; pfn++)
6530                 if (pfn_valid(pfn))
6531                         break;
6532         if (pfn == end_pfn)
6533                 return;
6534         zone = page_zone(pfn_to_page(pfn));
6535         spin_lock_irqsave(&zone->lock, flags);
6536         pfn = start_pfn;
6537         while (pfn < end_pfn) {
6538                 if (!pfn_valid(pfn)) {
6539                         pfn++;
6540                         continue;
6541                 }
6542                 page = pfn_to_page(pfn);
6543                 /*
6544                  * The HWPoisoned page may be not in buddy system, and
6545                  * page_count() is not 0.
6546                  */
6547                 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6548                         pfn++;
6549                         SetPageReserved(page);
6550                         continue;
6551                 }
6552 
6553                 BUG_ON(page_count(page));
6554                 BUG_ON(!PageBuddy(page));
6555                 order = page_order(page);
6556 #ifdef CONFIG_DEBUG_VM
6557                 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6558                        pfn, 1 << order, end_pfn);
6559 #endif
6560                 list_del(&page->lru);
6561                 rmv_page_order(page);
6562                 zone->free_area[order].nr_free--;
6563                 for (i = 0; i < (1 << order); i++)
6564                         SetPageReserved((page+i));
6565                 pfn += (1 << order);
6566         }
6567         spin_unlock_irqrestore(&zone->lock, flags);
6568 }
6569 #endif
6570 
6571 #ifdef CONFIG_MEMORY_FAILURE
6572 bool is_free_buddy_page(struct page *page)
6573 {
6574         struct zone *zone = page_zone(page);
6575         unsigned long pfn = page_to_pfn(page);
6576         unsigned long flags;
6577         unsigned int order;
6578 
6579         spin_lock_irqsave(&zone->lock, flags);
6580         for (order = 0; order < MAX_ORDER; order++) {
6581                 struct page *page_head = page - (pfn & ((1 << order) - 1));
6582 
6583                 if (PageBuddy(page_head) && page_order(page_head) >= order)
6584                         break;
6585         }
6586         spin_unlock_irqrestore(&zone->lock, flags);
6587 
6588         return order < MAX_ORDER;
6589 }
6590 #endif
6591 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us