Version:  2.0.40 2.2.26 2.4.37 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 4.1 4.2

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 #include <linux/kthread.h>
 65 
 66 #include <asm/sections.h>
 67 #include <asm/tlbflush.h>
 68 #include <asm/div64.h>
 69 #include "internal.h"
 70 
 71 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
 72 static DEFINE_MUTEX(pcp_batch_high_lock);
 73 #define MIN_PERCPU_PAGELIST_FRACTION    (8)
 74 
 75 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
 76 DEFINE_PER_CPU(int, numa_node);
 77 EXPORT_PER_CPU_SYMBOL(numa_node);
 78 #endif
 79 
 80 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
 81 /*
 82  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
 83  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
 84  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
 85  * defined in <linux/topology.h>.
 86  */
 87 DEFINE_PER_CPU(int, _numa_mem_);                /* Kernel "local memory" node */
 88 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
 89 int _node_numa_mem_[MAX_NUMNODES];
 90 #endif
 91 
 92 /*
 93  * Array of node states.
 94  */
 95 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
 96         [N_POSSIBLE] = NODE_MASK_ALL,
 97         [N_ONLINE] = { { [0] = 1UL } },
 98 #ifndef CONFIG_NUMA
 99         [N_NORMAL_MEMORY] = { { [0] = 1UL } },
100 #ifdef CONFIG_HIGHMEM
101         [N_HIGH_MEMORY] = { { [0] = 1UL } },
102 #endif
103 #ifdef CONFIG_MOVABLE_NODE
104         [N_MEMORY] = { { [0] = 1UL } },
105 #endif
106         [N_CPU] = { { [0] = 1UL } },
107 #endif  /* NUMA */
108 };
109 EXPORT_SYMBOL(node_states);
110 
111 /* Protect totalram_pages and zone->managed_pages */
112 static DEFINE_SPINLOCK(managed_page_count_lock);
113 
114 unsigned long totalram_pages __read_mostly;
115 unsigned long totalreserve_pages __read_mostly;
116 unsigned long totalcma_pages __read_mostly;
117 /*
118  * When calculating the number of globally allowed dirty pages, there
119  * is a certain number of per-zone reserves that should not be
120  * considered dirtyable memory.  This is the sum of those reserves
121  * over all existing zones that contribute dirtyable memory.
122  */
123 unsigned long dirty_balance_reserve __read_mostly;
124 
125 int percpu_pagelist_fraction;
126 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
127 
128 #ifdef CONFIG_PM_SLEEP
129 /*
130  * The following functions are used by the suspend/hibernate code to temporarily
131  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
132  * while devices are suspended.  To avoid races with the suspend/hibernate code,
133  * they should always be called with pm_mutex held (gfp_allowed_mask also should
134  * only be modified with pm_mutex held, unless the suspend/hibernate code is
135  * guaranteed not to run in parallel with that modification).
136  */
137 
138 static gfp_t saved_gfp_mask;
139 
140 void pm_restore_gfp_mask(void)
141 {
142         WARN_ON(!mutex_is_locked(&pm_mutex));
143         if (saved_gfp_mask) {
144                 gfp_allowed_mask = saved_gfp_mask;
145                 saved_gfp_mask = 0;
146         }
147 }
148 
149 void pm_restrict_gfp_mask(void)
150 {
151         WARN_ON(!mutex_is_locked(&pm_mutex));
152         WARN_ON(saved_gfp_mask);
153         saved_gfp_mask = gfp_allowed_mask;
154         gfp_allowed_mask &= ~GFP_IOFS;
155 }
156 
157 bool pm_suspended_storage(void)
158 {
159         if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
160                 return false;
161         return true;
162 }
163 #endif /* CONFIG_PM_SLEEP */
164 
165 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
166 int pageblock_order __read_mostly;
167 #endif
168 
169 static void __free_pages_ok(struct page *page, unsigned int order);
170 
171 /*
172  * results with 256, 32 in the lowmem_reserve sysctl:
173  *      1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
174  *      1G machine -> (16M dma, 784M normal, 224M high)
175  *      NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
176  *      HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
177  *      HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
178  *
179  * TBD: should special case ZONE_DMA32 machines here - in those we normally
180  * don't need any ZONE_NORMAL reservation
181  */
182 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
183 #ifdef CONFIG_ZONE_DMA
184          256,
185 #endif
186 #ifdef CONFIG_ZONE_DMA32
187          256,
188 #endif
189 #ifdef CONFIG_HIGHMEM
190          32,
191 #endif
192          32,
193 };
194 
195 EXPORT_SYMBOL(totalram_pages);
196 
197 static char * const zone_names[MAX_NR_ZONES] = {
198 #ifdef CONFIG_ZONE_DMA
199          "DMA",
200 #endif
201 #ifdef CONFIG_ZONE_DMA32
202          "DMA32",
203 #endif
204          "Normal",
205 #ifdef CONFIG_HIGHMEM
206          "HighMem",
207 #endif
208          "Movable",
209 };
210 
211 int min_free_kbytes = 1024;
212 int user_min_free_kbytes = -1;
213 
214 static unsigned long __meminitdata nr_kernel_pages;
215 static unsigned long __meminitdata nr_all_pages;
216 static unsigned long __meminitdata dma_reserve;
217 
218 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
219 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
220 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
221 static unsigned long __initdata required_kernelcore;
222 static unsigned long __initdata required_movablecore;
223 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
224 
225 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
226 int movable_zone;
227 EXPORT_SYMBOL(movable_zone);
228 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
229 
230 #if MAX_NUMNODES > 1
231 int nr_node_ids __read_mostly = MAX_NUMNODES;
232 int nr_online_nodes __read_mostly = 1;
233 EXPORT_SYMBOL(nr_node_ids);
234 EXPORT_SYMBOL(nr_online_nodes);
235 #endif
236 
237 int page_group_by_mobility_disabled __read_mostly;
238 
239 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
240 static inline void reset_deferred_meminit(pg_data_t *pgdat)
241 {
242         pgdat->first_deferred_pfn = ULONG_MAX;
243 }
244 
245 /* Returns true if the struct page for the pfn is uninitialised */
246 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
247 {
248         if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
249                 return true;
250 
251         return false;
252 }
253 
254 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
255 {
256         if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
257                 return true;
258 
259         return false;
260 }
261 
262 /*
263  * Returns false when the remaining initialisation should be deferred until
264  * later in the boot cycle when it can be parallelised.
265  */
266 static inline bool update_defer_init(pg_data_t *pgdat,
267                                 unsigned long pfn, unsigned long zone_end,
268                                 unsigned long *nr_initialised)
269 {
270         /* Always populate low zones for address-contrained allocations */
271         if (zone_end < pgdat_end_pfn(pgdat))
272                 return true;
273 
274         /* Initialise at least 2G of the highest zone */
275         (*nr_initialised)++;
276         if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) &&
277             (pfn & (PAGES_PER_SECTION - 1)) == 0) {
278                 pgdat->first_deferred_pfn = pfn;
279                 return false;
280         }
281 
282         return true;
283 }
284 #else
285 static inline void reset_deferred_meminit(pg_data_t *pgdat)
286 {
287 }
288 
289 static inline bool early_page_uninitialised(unsigned long pfn)
290 {
291         return false;
292 }
293 
294 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
295 {
296         return false;
297 }
298 
299 static inline bool update_defer_init(pg_data_t *pgdat,
300                                 unsigned long pfn, unsigned long zone_end,
301                                 unsigned long *nr_initialised)
302 {
303         return true;
304 }
305 #endif
306 
307 
308 void set_pageblock_migratetype(struct page *page, int migratetype)
309 {
310         if (unlikely(page_group_by_mobility_disabled &&
311                      migratetype < MIGRATE_PCPTYPES))
312                 migratetype = MIGRATE_UNMOVABLE;
313 
314         set_pageblock_flags_group(page, (unsigned long)migratetype,
315                                         PB_migrate, PB_migrate_end);
316 }
317 
318 #ifdef CONFIG_DEBUG_VM
319 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
320 {
321         int ret = 0;
322         unsigned seq;
323         unsigned long pfn = page_to_pfn(page);
324         unsigned long sp, start_pfn;
325 
326         do {
327                 seq = zone_span_seqbegin(zone);
328                 start_pfn = zone->zone_start_pfn;
329                 sp = zone->spanned_pages;
330                 if (!zone_spans_pfn(zone, pfn))
331                         ret = 1;
332         } while (zone_span_seqretry(zone, seq));
333 
334         if (ret)
335                 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
336                         pfn, zone_to_nid(zone), zone->name,
337                         start_pfn, start_pfn + sp);
338 
339         return ret;
340 }
341 
342 static int page_is_consistent(struct zone *zone, struct page *page)
343 {
344         if (!pfn_valid_within(page_to_pfn(page)))
345                 return 0;
346         if (zone != page_zone(page))
347                 return 0;
348 
349         return 1;
350 }
351 /*
352  * Temporary debugging check for pages not lying within a given zone.
353  */
354 static int bad_range(struct zone *zone, struct page *page)
355 {
356         if (page_outside_zone_boundaries(zone, page))
357                 return 1;
358         if (!page_is_consistent(zone, page))
359                 return 1;
360 
361         return 0;
362 }
363 #else
364 static inline int bad_range(struct zone *zone, struct page *page)
365 {
366         return 0;
367 }
368 #endif
369 
370 static void bad_page(struct page *page, const char *reason,
371                 unsigned long bad_flags)
372 {
373         static unsigned long resume;
374         static unsigned long nr_shown;
375         static unsigned long nr_unshown;
376 
377         /* Don't complain about poisoned pages */
378         if (PageHWPoison(page)) {
379                 page_mapcount_reset(page); /* remove PageBuddy */
380                 return;
381         }
382 
383         /*
384          * Allow a burst of 60 reports, then keep quiet for that minute;
385          * or allow a steady drip of one report per second.
386          */
387         if (nr_shown == 60) {
388                 if (time_before(jiffies, resume)) {
389                         nr_unshown++;
390                         goto out;
391                 }
392                 if (nr_unshown) {
393                         printk(KERN_ALERT
394                               "BUG: Bad page state: %lu messages suppressed\n",
395                                 nr_unshown);
396                         nr_unshown = 0;
397                 }
398                 nr_shown = 0;
399         }
400         if (nr_shown++ == 0)
401                 resume = jiffies + 60 * HZ;
402 
403         printk(KERN_ALERT "BUG: Bad page state in process %s  pfn:%05lx\n",
404                 current->comm, page_to_pfn(page));
405         dump_page_badflags(page, reason, bad_flags);
406 
407         print_modules();
408         dump_stack();
409 out:
410         /* Leave bad fields for debug, except PageBuddy could make trouble */
411         page_mapcount_reset(page); /* remove PageBuddy */
412         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
413 }
414 
415 /*
416  * Higher-order pages are called "compound pages".  They are structured thusly:
417  *
418  * The first PAGE_SIZE page is called the "head page".
419  *
420  * The remaining PAGE_SIZE pages are called "tail pages".
421  *
422  * All pages have PG_compound set.  All tail pages have their ->first_page
423  * pointing at the head page.
424  *
425  * The first tail page's ->lru.next holds the address of the compound page's
426  * put_page() function.  Its ->lru.prev holds the order of allocation.
427  * This usage means that zero-order pages may not be compound.
428  */
429 
430 static void free_compound_page(struct page *page)
431 {
432         __free_pages_ok(page, compound_order(page));
433 }
434 
435 void prep_compound_page(struct page *page, unsigned long order)
436 {
437         int i;
438         int nr_pages = 1 << order;
439 
440         set_compound_page_dtor(page, free_compound_page);
441         set_compound_order(page, order);
442         __SetPageHead(page);
443         for (i = 1; i < nr_pages; i++) {
444                 struct page *p = page + i;
445                 set_page_count(p, 0);
446                 p->first_page = page;
447                 /* Make sure p->first_page is always valid for PageTail() */
448                 smp_wmb();
449                 __SetPageTail(p);
450         }
451 }
452 
453 #ifdef CONFIG_DEBUG_PAGEALLOC
454 unsigned int _debug_guardpage_minorder;
455 bool _debug_pagealloc_enabled __read_mostly;
456 bool _debug_guardpage_enabled __read_mostly;
457 
458 static int __init early_debug_pagealloc(char *buf)
459 {
460         if (!buf)
461                 return -EINVAL;
462 
463         if (strcmp(buf, "on") == 0)
464                 _debug_pagealloc_enabled = true;
465 
466         return 0;
467 }
468 early_param("debug_pagealloc", early_debug_pagealloc);
469 
470 static bool need_debug_guardpage(void)
471 {
472         /* If we don't use debug_pagealloc, we don't need guard page */
473         if (!debug_pagealloc_enabled())
474                 return false;
475 
476         return true;
477 }
478 
479 static void init_debug_guardpage(void)
480 {
481         if (!debug_pagealloc_enabled())
482                 return;
483 
484         _debug_guardpage_enabled = true;
485 }
486 
487 struct page_ext_operations debug_guardpage_ops = {
488         .need = need_debug_guardpage,
489         .init = init_debug_guardpage,
490 };
491 
492 static int __init debug_guardpage_minorder_setup(char *buf)
493 {
494         unsigned long res;
495 
496         if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
497                 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
498                 return 0;
499         }
500         _debug_guardpage_minorder = res;
501         printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
502         return 0;
503 }
504 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
505 
506 static inline void set_page_guard(struct zone *zone, struct page *page,
507                                 unsigned int order, int migratetype)
508 {
509         struct page_ext *page_ext;
510 
511         if (!debug_guardpage_enabled())
512                 return;
513 
514         page_ext = lookup_page_ext(page);
515         __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
516 
517         INIT_LIST_HEAD(&page->lru);
518         set_page_private(page, order);
519         /* Guard pages are not available for any usage */
520         __mod_zone_freepage_state(zone, -(1 << order), migratetype);
521 }
522 
523 static inline void clear_page_guard(struct zone *zone, struct page *page,
524                                 unsigned int order, int migratetype)
525 {
526         struct page_ext *page_ext;
527 
528         if (!debug_guardpage_enabled())
529                 return;
530 
531         page_ext = lookup_page_ext(page);
532         __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
533 
534         set_page_private(page, 0);
535         if (!is_migrate_isolate(migratetype))
536                 __mod_zone_freepage_state(zone, (1 << order), migratetype);
537 }
538 #else
539 struct page_ext_operations debug_guardpage_ops = { NULL, };
540 static inline void set_page_guard(struct zone *zone, struct page *page,
541                                 unsigned int order, int migratetype) {}
542 static inline void clear_page_guard(struct zone *zone, struct page *page,
543                                 unsigned int order, int migratetype) {}
544 #endif
545 
546 static inline void set_page_order(struct page *page, unsigned int order)
547 {
548         set_page_private(page, order);
549         __SetPageBuddy(page);
550 }
551 
552 static inline void rmv_page_order(struct page *page)
553 {
554         __ClearPageBuddy(page);
555         set_page_private(page, 0);
556 }
557 
558 /*
559  * This function checks whether a page is free && is the buddy
560  * we can do coalesce a page and its buddy if
561  * (a) the buddy is not in a hole &&
562  * (b) the buddy is in the buddy system &&
563  * (c) a page and its buddy have the same order &&
564  * (d) a page and its buddy are in the same zone.
565  *
566  * For recording whether a page is in the buddy system, we set ->_mapcount
567  * PAGE_BUDDY_MAPCOUNT_VALUE.
568  * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
569  * serialized by zone->lock.
570  *
571  * For recording page's order, we use page_private(page).
572  */
573 static inline int page_is_buddy(struct page *page, struct page *buddy,
574                                                         unsigned int order)
575 {
576         if (!pfn_valid_within(page_to_pfn(buddy)))
577                 return 0;
578 
579         if (page_is_guard(buddy) && page_order(buddy) == order) {
580                 if (page_zone_id(page) != page_zone_id(buddy))
581                         return 0;
582 
583                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
584 
585                 return 1;
586         }
587 
588         if (PageBuddy(buddy) && page_order(buddy) == order) {
589                 /*
590                  * zone check is done late to avoid uselessly
591                  * calculating zone/node ids for pages that could
592                  * never merge.
593                  */
594                 if (page_zone_id(page) != page_zone_id(buddy))
595                         return 0;
596 
597                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
598 
599                 return 1;
600         }
601         return 0;
602 }
603 
604 /*
605  * Freeing function for a buddy system allocator.
606  *
607  * The concept of a buddy system is to maintain direct-mapped table
608  * (containing bit values) for memory blocks of various "orders".
609  * The bottom level table contains the map for the smallest allocatable
610  * units of memory (here, pages), and each level above it describes
611  * pairs of units from the levels below, hence, "buddies".
612  * At a high level, all that happens here is marking the table entry
613  * at the bottom level available, and propagating the changes upward
614  * as necessary, plus some accounting needed to play nicely with other
615  * parts of the VM system.
616  * At each level, we keep a list of pages, which are heads of continuous
617  * free pages of length of (1 << order) and marked with _mapcount
618  * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
619  * field.
620  * So when we are allocating or freeing one, we can derive the state of the
621  * other.  That is, if we allocate a small block, and both were
622  * free, the remainder of the region must be split into blocks.
623  * If a block is freed, and its buddy is also free, then this
624  * triggers coalescing into a block of larger size.
625  *
626  * -- nyc
627  */
628 
629 static inline void __free_one_page(struct page *page,
630                 unsigned long pfn,
631                 struct zone *zone, unsigned int order,
632                 int migratetype)
633 {
634         unsigned long page_idx;
635         unsigned long combined_idx;
636         unsigned long uninitialized_var(buddy_idx);
637         struct page *buddy;
638         int max_order = MAX_ORDER;
639 
640         VM_BUG_ON(!zone_is_initialized(zone));
641         VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
642 
643         VM_BUG_ON(migratetype == -1);
644         if (is_migrate_isolate(migratetype)) {
645                 /*
646                  * We restrict max order of merging to prevent merge
647                  * between freepages on isolate pageblock and normal
648                  * pageblock. Without this, pageblock isolation
649                  * could cause incorrect freepage accounting.
650                  */
651                 max_order = min(MAX_ORDER, pageblock_order + 1);
652         } else {
653                 __mod_zone_freepage_state(zone, 1 << order, migratetype);
654         }
655 
656         page_idx = pfn & ((1 << max_order) - 1);
657 
658         VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
659         VM_BUG_ON_PAGE(bad_range(zone, page), page);
660 
661         while (order < max_order - 1) {
662                 buddy_idx = __find_buddy_index(page_idx, order);
663                 buddy = page + (buddy_idx - page_idx);
664                 if (!page_is_buddy(page, buddy, order))
665                         break;
666                 /*
667                  * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
668                  * merge with it and move up one order.
669                  */
670                 if (page_is_guard(buddy)) {
671                         clear_page_guard(zone, buddy, order, migratetype);
672                 } else {
673                         list_del(&buddy->lru);
674                         zone->free_area[order].nr_free--;
675                         rmv_page_order(buddy);
676                 }
677                 combined_idx = buddy_idx & page_idx;
678                 page = page + (combined_idx - page_idx);
679                 page_idx = combined_idx;
680                 order++;
681         }
682         set_page_order(page, order);
683 
684         /*
685          * If this is not the largest possible page, check if the buddy
686          * of the next-highest order is free. If it is, it's possible
687          * that pages are being freed that will coalesce soon. In case,
688          * that is happening, add the free page to the tail of the list
689          * so it's less likely to be used soon and more likely to be merged
690          * as a higher order page
691          */
692         if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
693                 struct page *higher_page, *higher_buddy;
694                 combined_idx = buddy_idx & page_idx;
695                 higher_page = page + (combined_idx - page_idx);
696                 buddy_idx = __find_buddy_index(combined_idx, order + 1);
697                 higher_buddy = higher_page + (buddy_idx - combined_idx);
698                 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
699                         list_add_tail(&page->lru,
700                                 &zone->free_area[order].free_list[migratetype]);
701                         goto out;
702                 }
703         }
704 
705         list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
706 out:
707         zone->free_area[order].nr_free++;
708 }
709 
710 static inline int free_pages_check(struct page *page)
711 {
712         const char *bad_reason = NULL;
713         unsigned long bad_flags = 0;
714 
715         if (unlikely(page_mapcount(page)))
716                 bad_reason = "nonzero mapcount";
717         if (unlikely(page->mapping != NULL))
718                 bad_reason = "non-NULL mapping";
719         if (unlikely(atomic_read(&page->_count) != 0))
720                 bad_reason = "nonzero _count";
721         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
722                 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
723                 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
724         }
725 #ifdef CONFIG_MEMCG
726         if (unlikely(page->mem_cgroup))
727                 bad_reason = "page still charged to cgroup";
728 #endif
729         if (unlikely(bad_reason)) {
730                 bad_page(page, bad_reason, bad_flags);
731                 return 1;
732         }
733         page_cpupid_reset_last(page);
734         if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
735                 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
736         return 0;
737 }
738 
739 /*
740  * Frees a number of pages from the PCP lists
741  * Assumes all pages on list are in same zone, and of same order.
742  * count is the number of pages to free.
743  *
744  * If the zone was previously in an "all pages pinned" state then look to
745  * see if this freeing clears that state.
746  *
747  * And clear the zone's pages_scanned counter, to hold off the "all pages are
748  * pinned" detection logic.
749  */
750 static void free_pcppages_bulk(struct zone *zone, int count,
751                                         struct per_cpu_pages *pcp)
752 {
753         int migratetype = 0;
754         int batch_free = 0;
755         int to_free = count;
756         unsigned long nr_scanned;
757 
758         spin_lock(&zone->lock);
759         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
760         if (nr_scanned)
761                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
762 
763         while (to_free) {
764                 struct page *page;
765                 struct list_head *list;
766 
767                 /*
768                  * Remove pages from lists in a round-robin fashion. A
769                  * batch_free count is maintained that is incremented when an
770                  * empty list is encountered.  This is so more pages are freed
771                  * off fuller lists instead of spinning excessively around empty
772                  * lists
773                  */
774                 do {
775                         batch_free++;
776                         if (++migratetype == MIGRATE_PCPTYPES)
777                                 migratetype = 0;
778                         list = &pcp->lists[migratetype];
779                 } while (list_empty(list));
780 
781                 /* This is the only non-empty list. Free them all. */
782                 if (batch_free == MIGRATE_PCPTYPES)
783                         batch_free = to_free;
784 
785                 do {
786                         int mt; /* migratetype of the to-be-freed page */
787 
788                         page = list_entry(list->prev, struct page, lru);
789                         /* must delete as __free_one_page list manipulates */
790                         list_del(&page->lru);
791                         mt = get_freepage_migratetype(page);
792                         if (unlikely(has_isolate_pageblock(zone)))
793                                 mt = get_pageblock_migratetype(page);
794 
795                         /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
796                         __free_one_page(page, page_to_pfn(page), zone, 0, mt);
797                         trace_mm_page_pcpu_drain(page, 0, mt);
798                 } while (--to_free && --batch_free && !list_empty(list));
799         }
800         spin_unlock(&zone->lock);
801 }
802 
803 static void free_one_page(struct zone *zone,
804                                 struct page *page, unsigned long pfn,
805                                 unsigned int order,
806                                 int migratetype)
807 {
808         unsigned long nr_scanned;
809         spin_lock(&zone->lock);
810         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
811         if (nr_scanned)
812                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
813 
814         if (unlikely(has_isolate_pageblock(zone) ||
815                 is_migrate_isolate(migratetype))) {
816                 migratetype = get_pfnblock_migratetype(page, pfn);
817         }
818         __free_one_page(page, pfn, zone, order, migratetype);
819         spin_unlock(&zone->lock);
820 }
821 
822 static int free_tail_pages_check(struct page *head_page, struct page *page)
823 {
824         if (!IS_ENABLED(CONFIG_DEBUG_VM))
825                 return 0;
826         if (unlikely(!PageTail(page))) {
827                 bad_page(page, "PageTail not set", 0);
828                 return 1;
829         }
830         if (unlikely(page->first_page != head_page)) {
831                 bad_page(page, "first_page not consistent", 0);
832                 return 1;
833         }
834         return 0;
835 }
836 
837 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
838                                 unsigned long zone, int nid)
839 {
840         set_page_links(page, zone, nid, pfn);
841         init_page_count(page);
842         page_mapcount_reset(page);
843         page_cpupid_reset_last(page);
844 
845         INIT_LIST_HEAD(&page->lru);
846 #ifdef WANT_PAGE_VIRTUAL
847         /* The shift won't overflow because ZONE_NORMAL is below 4G. */
848         if (!is_highmem_idx(zone))
849                 set_page_address(page, __va(pfn << PAGE_SHIFT));
850 #endif
851 }
852 
853 static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
854                                         int nid)
855 {
856         return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
857 }
858 
859 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
860 static void init_reserved_page(unsigned long pfn)
861 {
862         pg_data_t *pgdat;
863         int nid, zid;
864 
865         if (!early_page_uninitialised(pfn))
866                 return;
867 
868         nid = early_pfn_to_nid(pfn);
869         pgdat = NODE_DATA(nid);
870 
871         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
872                 struct zone *zone = &pgdat->node_zones[zid];
873 
874                 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
875                         break;
876         }
877         __init_single_pfn(pfn, zid, nid);
878 }
879 #else
880 static inline void init_reserved_page(unsigned long pfn)
881 {
882 }
883 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
884 
885 /*
886  * Initialised pages do not have PageReserved set. This function is
887  * called for each range allocated by the bootmem allocator and
888  * marks the pages PageReserved. The remaining valid pages are later
889  * sent to the buddy page allocator.
890  */
891 void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
892 {
893         unsigned long start_pfn = PFN_DOWN(start);
894         unsigned long end_pfn = PFN_UP(end);
895 
896         for (; start_pfn < end_pfn; start_pfn++) {
897                 if (pfn_valid(start_pfn)) {
898                         struct page *page = pfn_to_page(start_pfn);
899 
900                         init_reserved_page(start_pfn);
901                         SetPageReserved(page);
902                 }
903         }
904 }
905 
906 static bool free_pages_prepare(struct page *page, unsigned int order)
907 {
908         bool compound = PageCompound(page);
909         int i, bad = 0;
910 
911         VM_BUG_ON_PAGE(PageTail(page), page);
912         VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
913 
914         trace_mm_page_free(page, order);
915         kmemcheck_free_shadow(page, order);
916         kasan_free_pages(page, order);
917 
918         if (PageAnon(page))
919                 page->mapping = NULL;
920         bad += free_pages_check(page);
921         for (i = 1; i < (1 << order); i++) {
922                 if (compound)
923                         bad += free_tail_pages_check(page, page + i);
924                 bad += free_pages_check(page + i);
925         }
926         if (bad)
927                 return false;
928 
929         reset_page_owner(page, order);
930 
931         if (!PageHighMem(page)) {
932                 debug_check_no_locks_freed(page_address(page),
933                                            PAGE_SIZE << order);
934                 debug_check_no_obj_freed(page_address(page),
935                                            PAGE_SIZE << order);
936         }
937         arch_free_page(page, order);
938         kernel_map_pages(page, 1 << order, 0);
939 
940         return true;
941 }
942 
943 static void __free_pages_ok(struct page *page, unsigned int order)
944 {
945         unsigned long flags;
946         int migratetype;
947         unsigned long pfn = page_to_pfn(page);
948 
949         if (!free_pages_prepare(page, order))
950                 return;
951 
952         migratetype = get_pfnblock_migratetype(page, pfn);
953         local_irq_save(flags);
954         __count_vm_events(PGFREE, 1 << order);
955         set_freepage_migratetype(page, migratetype);
956         free_one_page(page_zone(page), page, pfn, order, migratetype);
957         local_irq_restore(flags);
958 }
959 
960 static void __init __free_pages_boot_core(struct page *page,
961                                         unsigned long pfn, unsigned int order)
962 {
963         unsigned int nr_pages = 1 << order;
964         struct page *p = page;
965         unsigned int loop;
966 
967         prefetchw(p);
968         for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
969                 prefetchw(p + 1);
970                 __ClearPageReserved(p);
971                 set_page_count(p, 0);
972         }
973         __ClearPageReserved(p);
974         set_page_count(p, 0);
975 
976         page_zone(page)->managed_pages += nr_pages;
977         set_page_refcounted(page);
978         __free_pages(page, order);
979 }
980 
981 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
982         defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
983 
984 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
985 
986 int __meminit early_pfn_to_nid(unsigned long pfn)
987 {
988         static DEFINE_SPINLOCK(early_pfn_lock);
989         int nid;
990 
991         spin_lock(&early_pfn_lock);
992         nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
993         if (nid < 0)
994                 nid = 0;
995         spin_unlock(&early_pfn_lock);
996 
997         return nid;
998 }
999 #endif
1000 
1001 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1002 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1003                                         struct mminit_pfnnid_cache *state)
1004 {
1005         int nid;
1006 
1007         nid = __early_pfn_to_nid(pfn, state);
1008         if (nid >= 0 && nid != node)
1009                 return false;
1010         return true;
1011 }
1012 
1013 /* Only safe to use early in boot when initialisation is single-threaded */
1014 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1015 {
1016         return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1017 }
1018 
1019 #else
1020 
1021 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1022 {
1023         return true;
1024 }
1025 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1026                                         struct mminit_pfnnid_cache *state)
1027 {
1028         return true;
1029 }
1030 #endif
1031 
1032 
1033 void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
1034                                                         unsigned int order)
1035 {
1036         if (early_page_uninitialised(pfn))
1037                 return;
1038         return __free_pages_boot_core(page, pfn, order);
1039 }
1040 
1041 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1042 static void __init deferred_free_range(struct page *page,
1043                                         unsigned long pfn, int nr_pages)
1044 {
1045         int i;
1046 
1047         if (!page)
1048                 return;
1049 
1050         /* Free a large naturally-aligned chunk if possible */
1051         if (nr_pages == MAX_ORDER_NR_PAGES &&
1052             (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
1053                 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1054                 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1055                 return;
1056         }
1057 
1058         for (i = 0; i < nr_pages; i++, page++, pfn++)
1059                 __free_pages_boot_core(page, pfn, 0);
1060 }
1061 
1062 /* Completion tracking for deferred_init_memmap() threads */
1063 static atomic_t pgdat_init_n_undone __initdata;
1064 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1065 
1066 static inline void __init pgdat_init_report_one_done(void)
1067 {
1068         if (atomic_dec_and_test(&pgdat_init_n_undone))
1069                 complete(&pgdat_init_all_done_comp);
1070 }
1071 
1072 /* Initialise remaining memory on a node */
1073 static int __init deferred_init_memmap(void *data)
1074 {
1075         pg_data_t *pgdat = data;
1076         int nid = pgdat->node_id;
1077         struct mminit_pfnnid_cache nid_init_state = { };
1078         unsigned long start = jiffies;
1079         unsigned long nr_pages = 0;
1080         unsigned long walk_start, walk_end;
1081         int i, zid;
1082         struct zone *zone;
1083         unsigned long first_init_pfn = pgdat->first_deferred_pfn;
1084         const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1085 
1086         if (first_init_pfn == ULONG_MAX) {
1087                 pgdat_init_report_one_done();
1088                 return 0;
1089         }
1090 
1091         /* Bind memory initialisation thread to a local node if possible */
1092         if (!cpumask_empty(cpumask))
1093                 set_cpus_allowed_ptr(current, cpumask);
1094 
1095         /* Sanity check boundaries */
1096         BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1097         BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1098         pgdat->first_deferred_pfn = ULONG_MAX;
1099 
1100         /* Only the highest zone is deferred so find it */
1101         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1102                 zone = pgdat->node_zones + zid;
1103                 if (first_init_pfn < zone_end_pfn(zone))
1104                         break;
1105         }
1106 
1107         for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1108                 unsigned long pfn, end_pfn;
1109                 struct page *page = NULL;
1110                 struct page *free_base_page = NULL;
1111                 unsigned long free_base_pfn = 0;
1112                 int nr_to_free = 0;
1113 
1114                 end_pfn = min(walk_end, zone_end_pfn(zone));
1115                 pfn = first_init_pfn;
1116                 if (pfn < walk_start)
1117                         pfn = walk_start;
1118                 if (pfn < zone->zone_start_pfn)
1119                         pfn = zone->zone_start_pfn;
1120 
1121                 for (; pfn < end_pfn; pfn++) {
1122                         if (!pfn_valid_within(pfn))
1123                                 goto free_range;
1124 
1125                         /*
1126                          * Ensure pfn_valid is checked every
1127                          * MAX_ORDER_NR_PAGES for memory holes
1128                          */
1129                         if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1130                                 if (!pfn_valid(pfn)) {
1131                                         page = NULL;
1132                                         goto free_range;
1133                                 }
1134                         }
1135 
1136                         if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1137                                 page = NULL;
1138                                 goto free_range;
1139                         }
1140 
1141                         /* Minimise pfn page lookups and scheduler checks */
1142                         if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1143                                 page++;
1144                         } else {
1145                                 nr_pages += nr_to_free;
1146                                 deferred_free_range(free_base_page,
1147                                                 free_base_pfn, nr_to_free);
1148                                 free_base_page = NULL;
1149                                 free_base_pfn = nr_to_free = 0;
1150 
1151                                 page = pfn_to_page(pfn);
1152                                 cond_resched();
1153                         }
1154 
1155                         if (page->flags) {
1156                                 VM_BUG_ON(page_zone(page) != zone);
1157                                 goto free_range;
1158                         }
1159 
1160                         __init_single_page(page, pfn, zid, nid);
1161                         if (!free_base_page) {
1162                                 free_base_page = page;
1163                                 free_base_pfn = pfn;
1164                                 nr_to_free = 0;
1165                         }
1166                         nr_to_free++;
1167 
1168                         /* Where possible, batch up pages for a single free */
1169                         continue;
1170 free_range:
1171                         /* Free the current block of pages to allocator */
1172                         nr_pages += nr_to_free;
1173                         deferred_free_range(free_base_page, free_base_pfn,
1174                                                                 nr_to_free);
1175                         free_base_page = NULL;
1176                         free_base_pfn = nr_to_free = 0;
1177                 }
1178 
1179                 first_init_pfn = max(end_pfn, first_init_pfn);
1180         }
1181 
1182         /* Sanity check that the next zone really is unpopulated */
1183         WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1184 
1185         pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1186                                         jiffies_to_msecs(jiffies - start));
1187 
1188         pgdat_init_report_one_done();
1189         return 0;
1190 }
1191 
1192 void __init page_alloc_init_late(void)
1193 {
1194         int nid;
1195 
1196         /* There will be num_node_state(N_MEMORY) threads */
1197         atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1198         for_each_node_state(nid, N_MEMORY) {
1199                 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1200         }
1201 
1202         /* Block until all are initialised */
1203         wait_for_completion(&pgdat_init_all_done_comp);
1204 
1205         /* Reinit limits that are based on free pages after the kernel is up */
1206         files_maxfiles_init();
1207 }
1208 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1209 
1210 #ifdef CONFIG_CMA
1211 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1212 void __init init_cma_reserved_pageblock(struct page *page)
1213 {
1214         unsigned i = pageblock_nr_pages;
1215         struct page *p = page;
1216 
1217         do {
1218                 __ClearPageReserved(p);
1219                 set_page_count(p, 0);
1220         } while (++p, --i);
1221 
1222         set_pageblock_migratetype(page, MIGRATE_CMA);
1223 
1224         if (pageblock_order >= MAX_ORDER) {
1225                 i = pageblock_nr_pages;
1226                 p = page;
1227                 do {
1228                         set_page_refcounted(p);
1229                         __free_pages(p, MAX_ORDER - 1);
1230                         p += MAX_ORDER_NR_PAGES;
1231                 } while (i -= MAX_ORDER_NR_PAGES);
1232         } else {
1233                 set_page_refcounted(page);
1234                 __free_pages(page, pageblock_order);
1235         }
1236 
1237         adjust_managed_page_count(page, pageblock_nr_pages);
1238 }
1239 #endif
1240 
1241 /*
1242  * The order of subdivision here is critical for the IO subsystem.
1243  * Please do not alter this order without good reasons and regression
1244  * testing. Specifically, as large blocks of memory are subdivided,
1245  * the order in which smaller blocks are delivered depends on the order
1246  * they're subdivided in this function. This is the primary factor
1247  * influencing the order in which pages are delivered to the IO
1248  * subsystem according to empirical testing, and this is also justified
1249  * by considering the behavior of a buddy system containing a single
1250  * large block of memory acted on by a series of small allocations.
1251  * This behavior is a critical factor in sglist merging's success.
1252  *
1253  * -- nyc
1254  */
1255 static inline void expand(struct zone *zone, struct page *page,
1256         int low, int high, struct free_area *area,
1257         int migratetype)
1258 {
1259         unsigned long size = 1 << high;
1260 
1261         while (high > low) {
1262                 area--;
1263                 high--;
1264                 size >>= 1;
1265                 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1266 
1267                 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
1268                         debug_guardpage_enabled() &&
1269                         high < debug_guardpage_minorder()) {
1270                         /*
1271                          * Mark as guard pages (or page), that will allow to
1272                          * merge back to allocator when buddy will be freed.
1273                          * Corresponding page table entries will not be touched,
1274                          * pages will stay not present in virtual address space
1275                          */
1276                         set_page_guard(zone, &page[size], high, migratetype);
1277                         continue;
1278                 }
1279                 list_add(&page[size].lru, &area->free_list[migratetype]);
1280                 area->nr_free++;
1281                 set_page_order(&page[size], high);
1282         }
1283 }
1284 
1285 /*
1286  * This page is about to be returned from the page allocator
1287  */
1288 static inline int check_new_page(struct page *page)
1289 {
1290         const char *bad_reason = NULL;
1291         unsigned long bad_flags = 0;
1292 
1293         if (unlikely(page_mapcount(page)))
1294                 bad_reason = "nonzero mapcount";
1295         if (unlikely(page->mapping != NULL))
1296                 bad_reason = "non-NULL mapping";
1297         if (unlikely(atomic_read(&page->_count) != 0))
1298                 bad_reason = "nonzero _count";
1299         if (unlikely(page->flags & __PG_HWPOISON)) {
1300                 bad_reason = "HWPoisoned (hardware-corrupted)";
1301                 bad_flags = __PG_HWPOISON;
1302         }
1303         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1304                 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1305                 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1306         }
1307 #ifdef CONFIG_MEMCG
1308         if (unlikely(page->mem_cgroup))
1309                 bad_reason = "page still charged to cgroup";
1310 #endif
1311         if (unlikely(bad_reason)) {
1312                 bad_page(page, bad_reason, bad_flags);
1313                 return 1;
1314         }
1315         return 0;
1316 }
1317 
1318 static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1319                                                                 int alloc_flags)
1320 {
1321         int i;
1322 
1323         for (i = 0; i < (1 << order); i++) {
1324                 struct page *p = page + i;
1325                 if (unlikely(check_new_page(p)))
1326                         return 1;
1327         }
1328 
1329         set_page_private(page, 0);
1330         set_page_refcounted(page);
1331 
1332         arch_alloc_page(page, order);
1333         kernel_map_pages(page, 1 << order, 1);
1334         kasan_alloc_pages(page, order);
1335 
1336         if (gfp_flags & __GFP_ZERO)
1337                 for (i = 0; i < (1 << order); i++)
1338                         clear_highpage(page + i);
1339 
1340         if (order && (gfp_flags & __GFP_COMP))
1341                 prep_compound_page(page, order);
1342 
1343         set_page_owner(page, order, gfp_flags);
1344 
1345         /*
1346          * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1347          * allocate the page. The expectation is that the caller is taking
1348          * steps that will free more memory. The caller should avoid the page
1349          * being used for !PFMEMALLOC purposes.
1350          */
1351         if (alloc_flags & ALLOC_NO_WATERMARKS)
1352                 set_page_pfmemalloc(page);
1353         else
1354                 clear_page_pfmemalloc(page);
1355 
1356         return 0;
1357 }
1358 
1359 /*
1360  * Go through the free lists for the given migratetype and remove
1361  * the smallest available page from the freelists
1362  */
1363 static inline
1364 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1365                                                 int migratetype)
1366 {
1367         unsigned int current_order;
1368         struct free_area *area;
1369         struct page *page;
1370 
1371         /* Find a page of the appropriate size in the preferred list */
1372         for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1373                 area = &(zone->free_area[current_order]);
1374                 if (list_empty(&area->free_list[migratetype]))
1375                         continue;
1376 
1377                 page = list_entry(area->free_list[migratetype].next,
1378                                                         struct page, lru);
1379                 list_del(&page->lru);
1380                 rmv_page_order(page);
1381                 area->nr_free--;
1382                 expand(zone, page, order, current_order, area, migratetype);
1383                 set_freepage_migratetype(page, migratetype);
1384                 return page;
1385         }
1386 
1387         return NULL;
1388 }
1389 
1390 
1391 /*
1392  * This array describes the order lists are fallen back to when
1393  * the free lists for the desirable migrate type are depleted
1394  */
1395 static int fallbacks[MIGRATE_TYPES][4] = {
1396         [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,     MIGRATE_RESERVE },
1397         [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,     MIGRATE_RESERVE },
1398         [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,   MIGRATE_RESERVE },
1399 #ifdef CONFIG_CMA
1400         [MIGRATE_CMA]         = { MIGRATE_RESERVE }, /* Never used */
1401 #endif
1402         [MIGRATE_RESERVE]     = { MIGRATE_RESERVE }, /* Never used */
1403 #ifdef CONFIG_MEMORY_ISOLATION
1404         [MIGRATE_ISOLATE]     = { MIGRATE_RESERVE }, /* Never used */
1405 #endif
1406 };
1407 
1408 #ifdef CONFIG_CMA
1409 static struct page *__rmqueue_cma_fallback(struct zone *zone,
1410                                         unsigned int order)
1411 {
1412         return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1413 }
1414 #else
1415 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1416                                         unsigned int order) { return NULL; }
1417 #endif
1418 
1419 /*
1420  * Move the free pages in a range to the free lists of the requested type.
1421  * Note that start_page and end_pages are not aligned on a pageblock
1422  * boundary. If alignment is required, use move_freepages_block()
1423  */
1424 int move_freepages(struct zone *zone,
1425                           struct page *start_page, struct page *end_page,
1426                           int migratetype)
1427 {
1428         struct page *page;
1429         unsigned long order;
1430         int pages_moved = 0;
1431 
1432 #ifndef CONFIG_HOLES_IN_ZONE
1433         /*
1434          * page_zone is not safe to call in this context when
1435          * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1436          * anyway as we check zone boundaries in move_freepages_block().
1437          * Remove at a later date when no bug reports exist related to
1438          * grouping pages by mobility
1439          */
1440         VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1441 #endif
1442 
1443         for (page = start_page; page <= end_page;) {
1444                 /* Make sure we are not inadvertently changing nodes */
1445                 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1446 
1447                 if (!pfn_valid_within(page_to_pfn(page))) {
1448                         page++;
1449                         continue;
1450                 }
1451 
1452                 if (!PageBuddy(page)) {
1453                         page++;
1454                         continue;
1455                 }
1456 
1457                 order = page_order(page);
1458                 list_move(&page->lru,
1459                           &zone->free_area[order].free_list[migratetype]);
1460                 set_freepage_migratetype(page, migratetype);
1461                 page += 1 << order;
1462                 pages_moved += 1 << order;
1463         }
1464 
1465         return pages_moved;
1466 }
1467 
1468 int move_freepages_block(struct zone *zone, struct page *page,
1469                                 int migratetype)
1470 {
1471         unsigned long start_pfn, end_pfn;
1472         struct page *start_page, *end_page;
1473 
1474         start_pfn = page_to_pfn(page);
1475         start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1476         start_page = pfn_to_page(start_pfn);
1477         end_page = start_page + pageblock_nr_pages - 1;
1478         end_pfn = start_pfn + pageblock_nr_pages - 1;
1479 
1480         /* Do not cross zone boundaries */
1481         if (!zone_spans_pfn(zone, start_pfn))
1482                 start_page = page;
1483         if (!zone_spans_pfn(zone, end_pfn))
1484                 return 0;
1485 
1486         return move_freepages(zone, start_page, end_page, migratetype);
1487 }
1488 
1489 static void change_pageblock_range(struct page *pageblock_page,
1490                                         int start_order, int migratetype)
1491 {
1492         int nr_pageblocks = 1 << (start_order - pageblock_order);
1493 
1494         while (nr_pageblocks--) {
1495                 set_pageblock_migratetype(pageblock_page, migratetype);
1496                 pageblock_page += pageblock_nr_pages;
1497         }
1498 }
1499 
1500 /*
1501  * When we are falling back to another migratetype during allocation, try to
1502  * steal extra free pages from the same pageblocks to satisfy further
1503  * allocations, instead of polluting multiple pageblocks.
1504  *
1505  * If we are stealing a relatively large buddy page, it is likely there will
1506  * be more free pages in the pageblock, so try to steal them all. For
1507  * reclaimable and unmovable allocations, we steal regardless of page size,
1508  * as fragmentation caused by those allocations polluting movable pageblocks
1509  * is worse than movable allocations stealing from unmovable and reclaimable
1510  * pageblocks.
1511  */
1512 static bool can_steal_fallback(unsigned int order, int start_mt)
1513 {
1514         /*
1515          * Leaving this order check is intended, although there is
1516          * relaxed order check in next check. The reason is that
1517          * we can actually steal whole pageblock if this condition met,
1518          * but, below check doesn't guarantee it and that is just heuristic
1519          * so could be changed anytime.
1520          */
1521         if (order >= pageblock_order)
1522                 return true;
1523 
1524         if (order >= pageblock_order / 2 ||
1525                 start_mt == MIGRATE_RECLAIMABLE ||
1526                 start_mt == MIGRATE_UNMOVABLE ||
1527                 page_group_by_mobility_disabled)
1528                 return true;
1529 
1530         return false;
1531 }
1532 
1533 /*
1534  * This function implements actual steal behaviour. If order is large enough,
1535  * we can steal whole pageblock. If not, we first move freepages in this
1536  * pageblock and check whether half of pages are moved or not. If half of
1537  * pages are moved, we can change migratetype of pageblock and permanently
1538  * use it's pages as requested migratetype in the future.
1539  */
1540 static void steal_suitable_fallback(struct zone *zone, struct page *page,
1541                                                           int start_type)
1542 {
1543         int current_order = page_order(page);
1544         int pages;
1545 
1546         /* Take ownership for orders >= pageblock_order */
1547         if (current_order >= pageblock_order) {
1548                 change_pageblock_range(page, current_order, start_type);
1549                 return;
1550         }
1551 
1552         pages = move_freepages_block(zone, page, start_type);
1553 
1554         /* Claim the whole block if over half of it is free */
1555         if (pages >= (1 << (pageblock_order-1)) ||
1556                         page_group_by_mobility_disabled)
1557                 set_pageblock_migratetype(page, start_type);
1558 }
1559 
1560 /*
1561  * Check whether there is a suitable fallback freepage with requested order.
1562  * If only_stealable is true, this function returns fallback_mt only if
1563  * we can steal other freepages all together. This would help to reduce
1564  * fragmentation due to mixed migratetype pages in one pageblock.
1565  */
1566 int find_suitable_fallback(struct free_area *area, unsigned int order,
1567                         int migratetype, bool only_stealable, bool *can_steal)
1568 {
1569         int i;
1570         int fallback_mt;
1571 
1572         if (area->nr_free == 0)
1573                 return -1;
1574 
1575         *can_steal = false;
1576         for (i = 0;; i++) {
1577                 fallback_mt = fallbacks[migratetype][i];
1578                 if (fallback_mt == MIGRATE_RESERVE)
1579                         break;
1580 
1581                 if (list_empty(&area->free_list[fallback_mt]))
1582                         continue;
1583 
1584                 if (can_steal_fallback(order, migratetype))
1585                         *can_steal = true;
1586 
1587                 if (!only_stealable)
1588                         return fallback_mt;
1589 
1590                 if (*can_steal)
1591                         return fallback_mt;
1592         }
1593 
1594         return -1;
1595 }
1596 
1597 /* Remove an element from the buddy allocator from the fallback list */
1598 static inline struct page *
1599 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
1600 {
1601         struct free_area *area;
1602         unsigned int current_order;
1603         struct page *page;
1604         int fallback_mt;
1605         bool can_steal;
1606 
1607         /* Find the largest possible block of pages in the other list */
1608         for (current_order = MAX_ORDER-1;
1609                                 current_order >= order && current_order <= MAX_ORDER-1;
1610                                 --current_order) {
1611                 area = &(zone->free_area[current_order]);
1612                 fallback_mt = find_suitable_fallback(area, current_order,
1613                                 start_migratetype, false, &can_steal);
1614                 if (fallback_mt == -1)
1615                         continue;
1616 
1617                 page = list_entry(area->free_list[fallback_mt].next,
1618                                                 struct page, lru);
1619                 if (can_steal)
1620                         steal_suitable_fallback(zone, page, start_migratetype);
1621 
1622                 /* Remove the page from the freelists */
1623                 area->nr_free--;
1624                 list_del(&page->lru);
1625                 rmv_page_order(page);
1626 
1627                 expand(zone, page, order, current_order, area,
1628                                         start_migratetype);
1629                 /*
1630                  * The freepage_migratetype may differ from pageblock's
1631                  * migratetype depending on the decisions in
1632                  * try_to_steal_freepages(). This is OK as long as it
1633                  * does not differ for MIGRATE_CMA pageblocks. For CMA
1634                  * we need to make sure unallocated pages flushed from
1635                  * pcp lists are returned to the correct freelist.
1636                  */
1637                 set_freepage_migratetype(page, start_migratetype);
1638 
1639                 trace_mm_page_alloc_extfrag(page, order, current_order,
1640                         start_migratetype, fallback_mt);
1641 
1642                 return page;
1643         }
1644 
1645         return NULL;
1646 }
1647 
1648 /*
1649  * Do the hard work of removing an element from the buddy allocator.
1650  * Call me with the zone->lock already held.
1651  */
1652 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1653                                                 int migratetype)
1654 {
1655         struct page *page;
1656 
1657 retry_reserve:
1658         page = __rmqueue_smallest(zone, order, migratetype);
1659 
1660         if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1661                 if (migratetype == MIGRATE_MOVABLE)
1662                         page = __rmqueue_cma_fallback(zone, order);
1663 
1664                 if (!page)
1665                         page = __rmqueue_fallback(zone, order, migratetype);
1666 
1667                 /*
1668                  * Use MIGRATE_RESERVE rather than fail an allocation. goto
1669                  * is used because __rmqueue_smallest is an inline function
1670                  * and we want just one call site
1671                  */
1672                 if (!page) {
1673                         migratetype = MIGRATE_RESERVE;
1674                         goto retry_reserve;
1675                 }
1676         }
1677 
1678         trace_mm_page_alloc_zone_locked(page, order, migratetype);
1679         return page;
1680 }
1681 
1682 /*
1683  * Obtain a specified number of elements from the buddy allocator, all under
1684  * a single hold of the lock, for efficiency.  Add them to the supplied list.
1685  * Returns the number of new pages which were placed at *list.
1686  */
1687 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1688                         unsigned long count, struct list_head *list,
1689                         int migratetype, bool cold)
1690 {
1691         int i;
1692 
1693         spin_lock(&zone->lock);
1694         for (i = 0; i < count; ++i) {
1695                 struct page *page = __rmqueue(zone, order, migratetype);
1696                 if (unlikely(page == NULL))
1697                         break;
1698 
1699                 /*
1700                  * Split buddy pages returned by expand() are received here
1701                  * in physical page order. The page is added to the callers and
1702                  * list and the list head then moves forward. From the callers
1703                  * perspective, the linked list is ordered by page number in
1704                  * some conditions. This is useful for IO devices that can
1705                  * merge IO requests if the physical pages are ordered
1706                  * properly.
1707                  */
1708                 if (likely(!cold))
1709                         list_add(&page->lru, list);
1710                 else
1711                         list_add_tail(&page->lru, list);
1712                 list = &page->lru;
1713                 if (is_migrate_cma(get_freepage_migratetype(page)))
1714                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1715                                               -(1 << order));
1716         }
1717         __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1718         spin_unlock(&zone->lock);
1719         return i;
1720 }
1721 
1722 #ifdef CONFIG_NUMA
1723 /*
1724  * Called from the vmstat counter updater to drain pagesets of this
1725  * currently executing processor on remote nodes after they have
1726  * expired.
1727  *
1728  * Note that this function must be called with the thread pinned to
1729  * a single processor.
1730  */
1731 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1732 {
1733         unsigned long flags;
1734         int to_drain, batch;
1735 
1736         local_irq_save(flags);
1737         batch = READ_ONCE(pcp->batch);
1738         to_drain = min(pcp->count, batch);
1739         if (to_drain > 0) {
1740                 free_pcppages_bulk(zone, to_drain, pcp);
1741                 pcp->count -= to_drain;
1742         }
1743         local_irq_restore(flags);
1744 }
1745 #endif
1746 
1747 /*
1748  * Drain pcplists of the indicated processor and zone.
1749  *
1750  * The processor must either be the current processor and the
1751  * thread pinned to the current processor or a processor that
1752  * is not online.
1753  */
1754 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1755 {
1756         unsigned long flags;
1757         struct per_cpu_pageset *pset;
1758         struct per_cpu_pages *pcp;
1759 
1760         local_irq_save(flags);
1761         pset = per_cpu_ptr(zone->pageset, cpu);
1762 
1763         pcp = &pset->pcp;
1764         if (pcp->count) {
1765                 free_pcppages_bulk(zone, pcp->count, pcp);
1766                 pcp->count = 0;
1767         }
1768         local_irq_restore(flags);
1769 }
1770 
1771 /*
1772  * Drain pcplists of all zones on the indicated processor.
1773  *
1774  * The processor must either be the current processor and the
1775  * thread pinned to the current processor or a processor that
1776  * is not online.
1777  */
1778 static void drain_pages(unsigned int cpu)
1779 {
1780         struct zone *zone;
1781 
1782         for_each_populated_zone(zone) {
1783                 drain_pages_zone(cpu, zone);
1784         }
1785 }
1786 
1787 /*
1788  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1789  *
1790  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1791  * the single zone's pages.
1792  */
1793 void drain_local_pages(struct zone *zone)
1794 {
1795         int cpu = smp_processor_id();
1796 
1797         if (zone)
1798                 drain_pages_zone(cpu, zone);
1799         else
1800                 drain_pages(cpu);
1801 }
1802 
1803 /*
1804  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1805  *
1806  * When zone parameter is non-NULL, spill just the single zone's pages.
1807  *
1808  * Note that this code is protected against sending an IPI to an offline
1809  * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1810  * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1811  * nothing keeps CPUs from showing up after we populated the cpumask and
1812  * before the call to on_each_cpu_mask().
1813  */
1814 void drain_all_pages(struct zone *zone)
1815 {
1816         int cpu;
1817 
1818         /*
1819          * Allocate in the BSS so we wont require allocation in
1820          * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1821          */
1822         static cpumask_t cpus_with_pcps;
1823 
1824         /*
1825          * We don't care about racing with CPU hotplug event
1826          * as offline notification will cause the notified
1827          * cpu to drain that CPU pcps and on_each_cpu_mask
1828          * disables preemption as part of its processing
1829          */
1830         for_each_online_cpu(cpu) {
1831                 struct per_cpu_pageset *pcp;
1832                 struct zone *z;
1833                 bool has_pcps = false;
1834 
1835                 if (zone) {
1836                         pcp = per_cpu_ptr(zone->pageset, cpu);
1837                         if (pcp->pcp.count)
1838                                 has_pcps = true;
1839                 } else {
1840                         for_each_populated_zone(z) {
1841                                 pcp = per_cpu_ptr(z->pageset, cpu);
1842                                 if (pcp->pcp.count) {
1843                                         has_pcps = true;
1844                                         break;
1845                                 }
1846                         }
1847                 }
1848 
1849                 if (has_pcps)
1850                         cpumask_set_cpu(cpu, &cpus_with_pcps);
1851                 else
1852                         cpumask_clear_cpu(cpu, &cpus_with_pcps);
1853         }
1854         on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1855                                                                 zone, 1);
1856 }
1857 
1858 #ifdef CONFIG_HIBERNATION
1859 
1860 void mark_free_pages(struct zone *zone)
1861 {
1862         unsigned long pfn, max_zone_pfn;
1863         unsigned long flags;
1864         unsigned int order, t;
1865         struct list_head *curr;
1866 
1867         if (zone_is_empty(zone))
1868                 return;
1869 
1870         spin_lock_irqsave(&zone->lock, flags);
1871 
1872         max_zone_pfn = zone_end_pfn(zone);
1873         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1874                 if (pfn_valid(pfn)) {
1875                         struct page *page = pfn_to_page(pfn);
1876 
1877                         if (!swsusp_page_is_forbidden(page))
1878                                 swsusp_unset_page_free(page);
1879                 }
1880 
1881         for_each_migratetype_order(order, t) {
1882                 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1883                         unsigned long i;
1884 
1885                         pfn = page_to_pfn(list_entry(curr, struct page, lru));
1886                         for (i = 0; i < (1UL << order); i++)
1887                                 swsusp_set_page_free(pfn_to_page(pfn + i));
1888                 }
1889         }
1890         spin_unlock_irqrestore(&zone->lock, flags);
1891 }
1892 #endif /* CONFIG_PM */
1893 
1894 /*
1895  * Free a 0-order page
1896  * cold == true ? free a cold page : free a hot page
1897  */
1898 void free_hot_cold_page(struct page *page, bool cold)
1899 {
1900         struct zone *zone = page_zone(page);
1901         struct per_cpu_pages *pcp;
1902         unsigned long flags;
1903         unsigned long pfn = page_to_pfn(page);
1904         int migratetype;
1905 
1906         if (!free_pages_prepare(page, 0))
1907                 return;
1908 
1909         migratetype = get_pfnblock_migratetype(page, pfn);
1910         set_freepage_migratetype(page, migratetype);
1911         local_irq_save(flags);
1912         __count_vm_event(PGFREE);
1913 
1914         /*
1915          * We only track unmovable, reclaimable and movable on pcp lists.
1916          * Free ISOLATE pages back to the allocator because they are being
1917          * offlined but treat RESERVE as movable pages so we can get those
1918          * areas back if necessary. Otherwise, we may have to free
1919          * excessively into the page allocator
1920          */
1921         if (migratetype >= MIGRATE_PCPTYPES) {
1922                 if (unlikely(is_migrate_isolate(migratetype))) {
1923                         free_one_page(zone, page, pfn, 0, migratetype);
1924                         goto out;
1925                 }
1926                 migratetype = MIGRATE_MOVABLE;
1927         }
1928 
1929         pcp = &this_cpu_ptr(zone->pageset)->pcp;
1930         if (!cold)
1931                 list_add(&page->lru, &pcp->lists[migratetype]);
1932         else
1933                 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1934         pcp->count++;
1935         if (pcp->count >= pcp->high) {
1936                 unsigned long batch = READ_ONCE(pcp->batch);
1937                 free_pcppages_bulk(zone, batch, pcp);
1938                 pcp->count -= batch;
1939         }
1940 
1941 out:
1942         local_irq_restore(flags);
1943 }
1944 
1945 /*
1946  * Free a list of 0-order pages
1947  */
1948 void free_hot_cold_page_list(struct list_head *list, bool cold)
1949 {
1950         struct page *page, *next;
1951 
1952         list_for_each_entry_safe(page, next, list, lru) {
1953                 trace_mm_page_free_batched(page, cold);
1954                 free_hot_cold_page(page, cold);
1955         }
1956 }
1957 
1958 /*
1959  * split_page takes a non-compound higher-order page, and splits it into
1960  * n (1<<order) sub-pages: page[0..n]
1961  * Each sub-page must be freed individually.
1962  *
1963  * Note: this is probably too low level an operation for use in drivers.
1964  * Please consult with lkml before using this in your driver.
1965  */
1966 void split_page(struct page *page, unsigned int order)
1967 {
1968         int i;
1969         gfp_t gfp_mask;
1970 
1971         VM_BUG_ON_PAGE(PageCompound(page), page);
1972         VM_BUG_ON_PAGE(!page_count(page), page);
1973 
1974 #ifdef CONFIG_KMEMCHECK
1975         /*
1976          * Split shadow pages too, because free(page[0]) would
1977          * otherwise free the whole shadow.
1978          */
1979         if (kmemcheck_page_is_tracked(page))
1980                 split_page(virt_to_page(page[0].shadow), order);
1981 #endif
1982 
1983         gfp_mask = get_page_owner_gfp(page);
1984         set_page_owner(page, 0, gfp_mask);
1985         for (i = 1; i < (1 << order); i++) {
1986                 set_page_refcounted(page + i);
1987                 set_page_owner(page + i, 0, gfp_mask);
1988         }
1989 }
1990 EXPORT_SYMBOL_GPL(split_page);
1991 
1992 int __isolate_free_page(struct page *page, unsigned int order)
1993 {
1994         unsigned long watermark;
1995         struct zone *zone;
1996         int mt;
1997 
1998         BUG_ON(!PageBuddy(page));
1999 
2000         zone = page_zone(page);
2001         mt = get_pageblock_migratetype(page);
2002 
2003         if (!is_migrate_isolate(mt)) {
2004                 /* Obey watermarks as if the page was being allocated */
2005                 watermark = low_wmark_pages(zone) + (1 << order);
2006                 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2007                         return 0;
2008 
2009                 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2010         }
2011 
2012         /* Remove page from free list */
2013         list_del(&page->lru);
2014         zone->free_area[order].nr_free--;
2015         rmv_page_order(page);
2016 
2017         set_page_owner(page, order, __GFP_MOVABLE);
2018 
2019         /* Set the pageblock if the isolated page is at least a pageblock */
2020         if (order >= pageblock_order - 1) {
2021                 struct page *endpage = page + (1 << order) - 1;
2022                 for (; page < endpage; page += pageblock_nr_pages) {
2023                         int mt = get_pageblock_migratetype(page);
2024                         if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
2025                                 set_pageblock_migratetype(page,
2026                                                           MIGRATE_MOVABLE);
2027                 }
2028         }
2029 
2030 
2031         return 1UL << order;
2032 }
2033 
2034 /*
2035  * Similar to split_page except the page is already free. As this is only
2036  * being used for migration, the migratetype of the block also changes.
2037  * As this is called with interrupts disabled, the caller is responsible
2038  * for calling arch_alloc_page() and kernel_map_page() after interrupts
2039  * are enabled.
2040  *
2041  * Note: this is probably too low level an operation for use in drivers.
2042  * Please consult with lkml before using this in your driver.
2043  */
2044 int split_free_page(struct page *page)
2045 {
2046         unsigned int order;
2047         int nr_pages;
2048 
2049         order = page_order(page);
2050 
2051         nr_pages = __isolate_free_page(page, order);
2052         if (!nr_pages)
2053                 return 0;
2054 
2055         /* Split into individual pages */
2056         set_page_refcounted(page);
2057         split_page(page, order);
2058         return nr_pages;
2059 }
2060 
2061 /*
2062  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2063  */
2064 static inline
2065 struct page *buffered_rmqueue(struct zone *preferred_zone,
2066                         struct zone *zone, unsigned int order,
2067                         gfp_t gfp_flags, int migratetype)
2068 {
2069         unsigned long flags;
2070         struct page *page;
2071         bool cold = ((gfp_flags & __GFP_COLD) != 0);
2072 
2073         if (likely(order == 0)) {
2074                 struct per_cpu_pages *pcp;
2075                 struct list_head *list;
2076 
2077                 local_irq_save(flags);
2078                 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2079                 list = &pcp->lists[migratetype];
2080                 if (list_empty(list)) {
2081                         pcp->count += rmqueue_bulk(zone, 0,
2082                                         pcp->batch, list,
2083                                         migratetype, cold);
2084                         if (unlikely(list_empty(list)))
2085                                 goto failed;
2086                 }
2087 
2088                 if (cold)
2089                         page = list_entry(list->prev, struct page, lru);
2090                 else
2091                         page = list_entry(list->next, struct page, lru);
2092 
2093                 list_del(&page->lru);
2094                 pcp->count--;
2095         } else {
2096                 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
2097                         /*
2098                          * __GFP_NOFAIL is not to be used in new code.
2099                          *
2100                          * All __GFP_NOFAIL callers should be fixed so that they
2101                          * properly detect and handle allocation failures.
2102                          *
2103                          * We most definitely don't want callers attempting to
2104                          * allocate greater than order-1 page units with
2105                          * __GFP_NOFAIL.
2106                          */
2107                         WARN_ON_ONCE(order > 1);
2108                 }
2109                 spin_lock_irqsave(&zone->lock, flags);
2110                 page = __rmqueue(zone, order, migratetype);
2111                 spin_unlock(&zone->lock);
2112                 if (!page)
2113                         goto failed;
2114                 __mod_zone_freepage_state(zone, -(1 << order),
2115                                           get_freepage_migratetype(page));
2116         }
2117 
2118         __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
2119         if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
2120             !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2121                 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2122 
2123         __count_zone_vm_events(PGALLOC, zone, 1 << order);
2124         zone_statistics(preferred_zone, zone, gfp_flags);
2125         local_irq_restore(flags);
2126 
2127         VM_BUG_ON_PAGE(bad_range(zone, page), page);
2128         return page;
2129 
2130 failed:
2131         local_irq_restore(flags);
2132         return NULL;
2133 }
2134 
2135 #ifdef CONFIG_FAIL_PAGE_ALLOC
2136 
2137 static struct {
2138         struct fault_attr attr;
2139 
2140         u32 ignore_gfp_highmem;
2141         u32 ignore_gfp_wait;
2142         u32 min_order;
2143 } fail_page_alloc = {
2144         .attr = FAULT_ATTR_INITIALIZER,
2145         .ignore_gfp_wait = 1,
2146         .ignore_gfp_highmem = 1,
2147         .min_order = 1,
2148 };
2149 
2150 static int __init setup_fail_page_alloc(char *str)
2151 {
2152         return setup_fault_attr(&fail_page_alloc.attr, str);
2153 }
2154 __setup("fail_page_alloc=", setup_fail_page_alloc);
2155 
2156 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2157 {
2158         if (order < fail_page_alloc.min_order)
2159                 return false;
2160         if (gfp_mask & __GFP_NOFAIL)
2161                 return false;
2162         if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
2163                 return false;
2164         if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
2165                 return false;
2166 
2167         return should_fail(&fail_page_alloc.attr, 1 << order);
2168 }
2169 
2170 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2171 
2172 static int __init fail_page_alloc_debugfs(void)
2173 {
2174         umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
2175         struct dentry *dir;
2176 
2177         dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2178                                         &fail_page_alloc.attr);
2179         if (IS_ERR(dir))
2180                 return PTR_ERR(dir);
2181 
2182         if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2183                                 &fail_page_alloc.ignore_gfp_wait))
2184                 goto fail;
2185         if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2186                                 &fail_page_alloc.ignore_gfp_highmem))
2187                 goto fail;
2188         if (!debugfs_create_u32("min-order", mode, dir,
2189                                 &fail_page_alloc.min_order))
2190                 goto fail;
2191 
2192         return 0;
2193 fail:
2194         debugfs_remove_recursive(dir);
2195 
2196         return -ENOMEM;
2197 }
2198 
2199 late_initcall(fail_page_alloc_debugfs);
2200 
2201 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2202 
2203 #else /* CONFIG_FAIL_PAGE_ALLOC */
2204 
2205 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2206 {
2207         return false;
2208 }
2209 
2210 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2211 
2212 /*
2213  * Return true if free pages are above 'mark'. This takes into account the order
2214  * of the allocation.
2215  */
2216 static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2217                         unsigned long mark, int classzone_idx, int alloc_flags,
2218                         long free_pages)
2219 {
2220         /* free_pages may go negative - that's OK */
2221         long min = mark;
2222         int o;
2223         long free_cma = 0;
2224 
2225         free_pages -= (1 << order) - 1;
2226         if (alloc_flags & ALLOC_HIGH)
2227                 min -= min / 2;
2228         if (alloc_flags & ALLOC_HARDER)
2229                 min -= min / 4;
2230 #ifdef CONFIG_CMA
2231         /* If allocation can't use CMA areas don't use free CMA pages */
2232         if (!(alloc_flags & ALLOC_CMA))
2233                 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
2234 #endif
2235 
2236         if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
2237                 return false;
2238         for (o = 0; o < order; o++) {
2239                 /* At the next order, this order's pages become unavailable */
2240                 free_pages -= z->free_area[o].nr_free << o;
2241 
2242                 /* Require fewer higher order pages to be free */
2243                 min >>= 1;
2244 
2245                 if (free_pages <= min)
2246                         return false;
2247         }
2248         return true;
2249 }
2250 
2251 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2252                       int classzone_idx, int alloc_flags)
2253 {
2254         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2255                                         zone_page_state(z, NR_FREE_PAGES));
2256 }
2257 
2258 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2259                         unsigned long mark, int classzone_idx, int alloc_flags)
2260 {
2261         long free_pages = zone_page_state(z, NR_FREE_PAGES);
2262 
2263         if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2264                 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2265 
2266         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2267                                                                 free_pages);
2268 }
2269 
2270 #ifdef CONFIG_NUMA
2271 /*
2272  * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
2273  * skip over zones that are not allowed by the cpuset, or that have
2274  * been recently (in last second) found to be nearly full.  See further
2275  * comments in mmzone.h.  Reduces cache footprint of zonelist scans
2276  * that have to skip over a lot of full or unallowed zones.
2277  *
2278  * If the zonelist cache is present in the passed zonelist, then
2279  * returns a pointer to the allowed node mask (either the current
2280  * tasks mems_allowed, or node_states[N_MEMORY].)
2281  *
2282  * If the zonelist cache is not available for this zonelist, does
2283  * nothing and returns NULL.
2284  *
2285  * If the fullzones BITMAP in the zonelist cache is stale (more than
2286  * a second since last zap'd) then we zap it out (clear its bits.)
2287  *
2288  * We hold off even calling zlc_setup, until after we've checked the
2289  * first zone in the zonelist, on the theory that most allocations will
2290  * be satisfied from that first zone, so best to examine that zone as
2291  * quickly as we can.
2292  */
2293 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2294 {
2295         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
2296         nodemask_t *allowednodes;       /* zonelist_cache approximation */
2297 
2298         zlc = zonelist->zlcache_ptr;
2299         if (!zlc)
2300                 return NULL;
2301 
2302         if (time_after(jiffies, zlc->last_full_zap + HZ)) {
2303                 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2304                 zlc->last_full_zap = jiffies;
2305         }
2306 
2307         allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
2308                                         &cpuset_current_mems_allowed :
2309                                         &node_states[N_MEMORY];
2310         return allowednodes;
2311 }
2312 
2313 /*
2314  * Given 'z' scanning a zonelist, run a couple of quick checks to see
2315  * if it is worth looking at further for free memory:
2316  *  1) Check that the zone isn't thought to be full (doesn't have its
2317  *     bit set in the zonelist_cache fullzones BITMAP).
2318  *  2) Check that the zones node (obtained from the zonelist_cache
2319  *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
2320  * Return true (non-zero) if zone is worth looking at further, or
2321  * else return false (zero) if it is not.
2322  *
2323  * This check -ignores- the distinction between various watermarks,
2324  * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
2325  * found to be full for any variation of these watermarks, it will
2326  * be considered full for up to one second by all requests, unless
2327  * we are so low on memory on all allowed nodes that we are forced
2328  * into the second scan of the zonelist.
2329  *
2330  * In the second scan we ignore this zonelist cache and exactly
2331  * apply the watermarks to all zones, even it is slower to do so.
2332  * We are low on memory in the second scan, and should leave no stone
2333  * unturned looking for a free page.
2334  */
2335 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
2336                                                 nodemask_t *allowednodes)
2337 {
2338         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
2339         int i;                          /* index of *z in zonelist zones */
2340         int n;                          /* node that zone *z is on */
2341 
2342         zlc = zonelist->zlcache_ptr;
2343         if (!zlc)
2344                 return 1;
2345 
2346         i = z - zonelist->_zonerefs;
2347         n = zlc->z_to_n[i];
2348 
2349         /* This zone is worth trying if it is allowed but not full */
2350         return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
2351 }
2352 
2353 /*
2354  * Given 'z' scanning a zonelist, set the corresponding bit in
2355  * zlc->fullzones, so that subsequent attempts to allocate a page
2356  * from that zone don't waste time re-examining it.
2357  */
2358 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
2359 {
2360         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
2361         int i;                          /* index of *z in zonelist zones */
2362 
2363         zlc = zonelist->zlcache_ptr;
2364         if (!zlc)
2365                 return;
2366 
2367         i = z - zonelist->_zonerefs;
2368 
2369         set_bit(i, zlc->fullzones);
2370 }
2371 
2372 /*
2373  * clear all zones full, called after direct reclaim makes progress so that
2374  * a zone that was recently full is not skipped over for up to a second
2375  */
2376 static void zlc_clear_zones_full(struct zonelist *zonelist)
2377 {
2378         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
2379 
2380         zlc = zonelist->zlcache_ptr;
2381         if (!zlc)
2382                 return;
2383 
2384         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2385 }
2386 
2387 static bool zone_local(struct zone *local_zone, struct zone *zone)
2388 {
2389         return local_zone->node == zone->node;
2390 }
2391 
2392 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2393 {
2394         return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2395                                 RECLAIM_DISTANCE;
2396 }
2397 
2398 #else   /* CONFIG_NUMA */
2399 
2400 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2401 {
2402         return NULL;
2403 }
2404 
2405 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
2406                                 nodemask_t *allowednodes)
2407 {
2408         return 1;
2409 }
2410 
2411 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
2412 {
2413 }
2414 
2415 static void zlc_clear_zones_full(struct zonelist *zonelist)
2416 {
2417 }
2418 
2419 static bool zone_local(struct zone *local_zone, struct zone *zone)
2420 {
2421         return true;
2422 }
2423 
2424 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2425 {
2426         return true;
2427 }
2428 
2429 #endif  /* CONFIG_NUMA */
2430 
2431 static void reset_alloc_batches(struct zone *preferred_zone)
2432 {
2433         struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2434 
2435         do {
2436                 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2437                         high_wmark_pages(zone) - low_wmark_pages(zone) -
2438                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
2439                 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2440         } while (zone++ != preferred_zone);
2441 }
2442 
2443 /*
2444  * get_page_from_freelist goes through the zonelist trying to allocate
2445  * a page.
2446  */
2447 static struct page *
2448 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2449                                                 const struct alloc_context *ac)
2450 {
2451         struct zonelist *zonelist = ac->zonelist;
2452         struct zoneref *z;
2453         struct page *page = NULL;
2454         struct zone *zone;
2455         nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
2456         int zlc_active = 0;             /* set if using zonelist_cache */
2457         int did_zlc_setup = 0;          /* just call zlc_setup() one time */
2458         bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
2459                                 (gfp_mask & __GFP_WRITE);
2460         int nr_fair_skipped = 0;
2461         bool zonelist_rescan;
2462 
2463 zonelist_scan:
2464         zonelist_rescan = false;
2465 
2466         /*
2467          * Scan zonelist, looking for a zone with enough free.
2468          * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2469          */
2470         for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2471                                                                 ac->nodemask) {
2472                 unsigned long mark;
2473 
2474                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2475                         !zlc_zone_worth_trying(zonelist, z, allowednodes))
2476                                 continue;
2477                 if (cpusets_enabled() &&
2478                         (alloc_flags & ALLOC_CPUSET) &&
2479                         !cpuset_zone_allowed(zone, gfp_mask))
2480                                 continue;
2481                 /*
2482                  * Distribute pages in proportion to the individual
2483                  * zone size to ensure fair page aging.  The zone a
2484                  * page was allocated in should have no effect on the
2485                  * time the page has in memory before being reclaimed.
2486                  */
2487                 if (alloc_flags & ALLOC_FAIR) {
2488                         if (!zone_local(ac->preferred_zone, zone))
2489                                 break;
2490                         if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
2491                                 nr_fair_skipped++;
2492                                 continue;
2493                         }
2494                 }
2495                 /*
2496                  * When allocating a page cache page for writing, we
2497                  * want to get it from a zone that is within its dirty
2498                  * limit, such that no single zone holds more than its
2499                  * proportional share of globally allowed dirty pages.
2500                  * The dirty limits take into account the zone's
2501                  * lowmem reserves and high watermark so that kswapd
2502                  * should be able to balance it without having to
2503                  * write pages from its LRU list.
2504                  *
2505                  * This may look like it could increase pressure on
2506                  * lower zones by failing allocations in higher zones
2507                  * before they are full.  But the pages that do spill
2508                  * over are limited as the lower zones are protected
2509                  * by this very same mechanism.  It should not become
2510                  * a practical burden to them.
2511                  *
2512                  * XXX: For now, allow allocations to potentially
2513                  * exceed the per-zone dirty limit in the slowpath
2514                  * (ALLOC_WMARK_LOW unset) before going into reclaim,
2515                  * which is important when on a NUMA setup the allowed
2516                  * zones are together not big enough to reach the
2517                  * global limit.  The proper fix for these situations
2518                  * will require awareness of zones in the
2519                  * dirty-throttling and the flusher threads.
2520                  */
2521                 if (consider_zone_dirty && !zone_dirty_ok(zone))
2522                         continue;
2523 
2524                 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2525                 if (!zone_watermark_ok(zone, order, mark,
2526                                        ac->classzone_idx, alloc_flags)) {
2527                         int ret;
2528 
2529                         /* Checked here to keep the fast path fast */
2530                         BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2531                         if (alloc_flags & ALLOC_NO_WATERMARKS)
2532                                 goto try_this_zone;
2533 
2534                         if (IS_ENABLED(CONFIG_NUMA) &&
2535                                         !did_zlc_setup && nr_online_nodes > 1) {
2536                                 /*
2537                                  * we do zlc_setup if there are multiple nodes
2538                                  * and before considering the first zone allowed
2539                                  * by the cpuset.
2540                                  */
2541                                 allowednodes = zlc_setup(zonelist, alloc_flags);
2542                                 zlc_active = 1;
2543                                 did_zlc_setup = 1;
2544                         }
2545 
2546                         if (zone_reclaim_mode == 0 ||
2547                             !zone_allows_reclaim(ac->preferred_zone, zone))
2548                                 goto this_zone_full;
2549 
2550                         /*
2551                          * As we may have just activated ZLC, check if the first
2552                          * eligible zone has failed zone_reclaim recently.
2553                          */
2554                         if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2555                                 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2556                                 continue;
2557 
2558                         ret = zone_reclaim(zone, gfp_mask, order);
2559                         switch (ret) {
2560                         case ZONE_RECLAIM_NOSCAN:
2561                                 /* did not scan */
2562                                 continue;
2563                         case ZONE_RECLAIM_FULL:
2564                                 /* scanned but unreclaimable */
2565                                 continue;
2566                         default:
2567                                 /* did we reclaim enough */
2568                                 if (zone_watermark_ok(zone, order, mark,
2569                                                 ac->classzone_idx, alloc_flags))
2570                                         goto try_this_zone;
2571 
2572                                 /*
2573                                  * Failed to reclaim enough to meet watermark.
2574                                  * Only mark the zone full if checking the min
2575                                  * watermark or if we failed to reclaim just
2576                                  * 1<<order pages or else the page allocator
2577                                  * fastpath will prematurely mark zones full
2578                                  * when the watermark is between the low and
2579                                  * min watermarks.
2580                                  */
2581                                 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2582                                     ret == ZONE_RECLAIM_SOME)
2583                                         goto this_zone_full;
2584 
2585                                 continue;
2586                         }
2587                 }
2588 
2589 try_this_zone:
2590                 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2591                                                 gfp_mask, ac->migratetype);
2592                 if (page) {
2593                         if (prep_new_page(page, order, gfp_mask, alloc_flags))
2594                                 goto try_this_zone;
2595                         return page;
2596                 }
2597 this_zone_full:
2598                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
2599                         zlc_mark_zone_full(zonelist, z);
2600         }
2601 
2602         /*
2603          * The first pass makes sure allocations are spread fairly within the
2604          * local node.  However, the local node might have free pages left
2605          * after the fairness batches are exhausted, and remote zones haven't
2606          * even been considered yet.  Try once more without fairness, and
2607          * include remote zones now, before entering the slowpath and waking
2608          * kswapd: prefer spilling to a remote zone over swapping locally.
2609          */
2610         if (alloc_flags & ALLOC_FAIR) {
2611                 alloc_flags &= ~ALLOC_FAIR;
2612                 if (nr_fair_skipped) {
2613                         zonelist_rescan = true;
2614                         reset_alloc_batches(ac->preferred_zone);
2615                 }
2616                 if (nr_online_nodes > 1)
2617                         zonelist_rescan = true;
2618         }
2619 
2620         if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2621                 /* Disable zlc cache for second zonelist scan */
2622                 zlc_active = 0;
2623                 zonelist_rescan = true;
2624         }
2625 
2626         if (zonelist_rescan)
2627                 goto zonelist_scan;
2628 
2629         return NULL;
2630 }
2631 
2632 /*
2633  * Large machines with many possible nodes should not always dump per-node
2634  * meminfo in irq context.
2635  */
2636 static inline bool should_suppress_show_mem(void)
2637 {
2638         bool ret = false;
2639 
2640 #if NODES_SHIFT > 8
2641         ret = in_interrupt();
2642 #endif
2643         return ret;
2644 }
2645 
2646 static DEFINE_RATELIMIT_STATE(nopage_rs,
2647                 DEFAULT_RATELIMIT_INTERVAL,
2648                 DEFAULT_RATELIMIT_BURST);
2649 
2650 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2651 {
2652         unsigned int filter = SHOW_MEM_FILTER_NODES;
2653 
2654         if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2655             debug_guardpage_minorder() > 0)
2656                 return;
2657 
2658         /*
2659          * This documents exceptions given to allocations in certain
2660          * contexts that are allowed to allocate outside current's set
2661          * of allowed nodes.
2662          */
2663         if (!(gfp_mask & __GFP_NOMEMALLOC))
2664                 if (test_thread_flag(TIF_MEMDIE) ||
2665                     (current->flags & (PF_MEMALLOC | PF_EXITING)))
2666                         filter &= ~SHOW_MEM_FILTER_NODES;
2667         if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2668                 filter &= ~SHOW_MEM_FILTER_NODES;
2669 
2670         if (fmt) {
2671                 struct va_format vaf;
2672                 va_list args;
2673 
2674                 va_start(args, fmt);
2675 
2676                 vaf.fmt = fmt;
2677                 vaf.va = &args;
2678 
2679                 pr_warn("%pV", &vaf);
2680 
2681                 va_end(args);
2682         }
2683 
2684         pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2685                 current->comm, order, gfp_mask);
2686 
2687         dump_stack();
2688         if (!should_suppress_show_mem())
2689                 show_mem(filter);
2690 }
2691 
2692 static inline struct page *
2693 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2694         const struct alloc_context *ac, unsigned long *did_some_progress)
2695 {
2696         struct page *page;
2697 
2698         *did_some_progress = 0;
2699 
2700         /*
2701          * Acquire the oom lock.  If that fails, somebody else is
2702          * making progress for us.
2703          */
2704         if (!mutex_trylock(&oom_lock)) {
2705                 *did_some_progress = 1;
2706                 schedule_timeout_uninterruptible(1);
2707                 return NULL;
2708         }
2709 
2710         /*
2711          * Go through the zonelist yet one more time, keep very high watermark
2712          * here, this is only to catch a parallel oom killing, we must fail if
2713          * we're still under heavy pressure.
2714          */
2715         page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2716                                         ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
2717         if (page)
2718                 goto out;
2719 
2720         if (!(gfp_mask & __GFP_NOFAIL)) {
2721                 /* Coredumps can quickly deplete all memory reserves */
2722                 if (current->flags & PF_DUMPCORE)
2723                         goto out;
2724                 /* The OOM killer will not help higher order allocs */
2725                 if (order > PAGE_ALLOC_COSTLY_ORDER)
2726                         goto out;
2727                 /* The OOM killer does not needlessly kill tasks for lowmem */
2728                 if (ac->high_zoneidx < ZONE_NORMAL)
2729                         goto out;
2730                 /* The OOM killer does not compensate for IO-less reclaim */
2731                 if (!(gfp_mask & __GFP_FS)) {
2732                         /*
2733                          * XXX: Page reclaim didn't yield anything,
2734                          * and the OOM killer can't be invoked, but
2735                          * keep looping as per tradition.
2736                          */
2737                         *did_some_progress = 1;
2738                         goto out;
2739                 }
2740                 if (pm_suspended_storage())
2741                         goto out;
2742                 /* The OOM killer may not free memory on a specific node */
2743                 if (gfp_mask & __GFP_THISNODE)
2744                         goto out;
2745         }
2746         /* Exhausted what can be done so it's blamo time */
2747         if (out_of_memory(ac->zonelist, gfp_mask, order, ac->nodemask, false)
2748                         || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
2749                 *did_some_progress = 1;
2750 out:
2751         mutex_unlock(&oom_lock);
2752         return page;
2753 }
2754 
2755 #ifdef CONFIG_COMPACTION
2756 /* Try memory compaction for high-order allocations before reclaim */
2757 static struct page *
2758 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2759                 int alloc_flags, const struct alloc_context *ac,
2760                 enum migrate_mode mode, int *contended_compaction,
2761                 bool *deferred_compaction)
2762 {
2763         unsigned long compact_result;
2764         struct page *page;
2765 
2766         if (!order)
2767                 return NULL;
2768 
2769         current->flags |= PF_MEMALLOC;
2770         compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2771                                                 mode, contended_compaction);
2772         current->flags &= ~PF_MEMALLOC;
2773 
2774         switch (compact_result) {
2775         case COMPACT_DEFERRED:
2776                 *deferred_compaction = true;
2777                 /* fall-through */
2778         case COMPACT_SKIPPED:
2779                 return NULL;
2780         default:
2781                 break;
2782         }
2783 
2784         /*
2785          * At least in one zone compaction wasn't deferred or skipped, so let's
2786          * count a compaction stall
2787          */
2788         count_vm_event(COMPACTSTALL);
2789 
2790         page = get_page_from_freelist(gfp_mask, order,
2791                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2792 
2793         if (page) {
2794                 struct zone *zone = page_zone(page);
2795 
2796                 zone->compact_blockskip_flush = false;
2797                 compaction_defer_reset(zone, order, true);
2798                 count_vm_event(COMPACTSUCCESS);
2799                 return page;
2800         }
2801 
2802         /*
2803          * It's bad if compaction run occurs and fails. The most likely reason
2804          * is that pages exist, but not enough to satisfy watermarks.
2805          */
2806         count_vm_event(COMPACTFAIL);
2807 
2808         cond_resched();
2809 
2810         return NULL;
2811 }
2812 #else
2813 static inline struct page *
2814 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2815                 int alloc_flags, const struct alloc_context *ac,
2816                 enum migrate_mode mode, int *contended_compaction,
2817                 bool *deferred_compaction)
2818 {
2819         return NULL;
2820 }
2821 #endif /* CONFIG_COMPACTION */
2822 
2823 /* Perform direct synchronous page reclaim */
2824 static int
2825 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
2826                                         const struct alloc_context *ac)
2827 {
2828         struct reclaim_state reclaim_state;
2829         int progress;
2830 
2831         cond_resched();
2832 
2833         /* We now go into synchronous reclaim */
2834         cpuset_memory_pressure_bump();
2835         current->flags |= PF_MEMALLOC;
2836         lockdep_set_current_reclaim_state(gfp_mask);
2837         reclaim_state.reclaimed_slab = 0;
2838         current->reclaim_state = &reclaim_state;
2839 
2840         progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2841                                                                 ac->nodemask);
2842 
2843         current->reclaim_state = NULL;
2844         lockdep_clear_current_reclaim_state();
2845         current->flags &= ~PF_MEMALLOC;
2846 
2847         cond_resched();
2848 
2849         return progress;
2850 }
2851 
2852 /* The really slow allocator path where we enter direct reclaim */
2853 static inline struct page *
2854 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2855                 int alloc_flags, const struct alloc_context *ac,
2856                 unsigned long *did_some_progress)
2857 {
2858         struct page *page = NULL;
2859         bool drained = false;
2860 
2861         *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
2862         if (unlikely(!(*did_some_progress)))
2863                 return NULL;
2864 
2865         /* After successful reclaim, reconsider all zones for allocation */
2866         if (IS_ENABLED(CONFIG_NUMA))
2867                 zlc_clear_zones_full(ac->zonelist);
2868 
2869 retry:
2870         page = get_page_from_freelist(gfp_mask, order,
2871                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2872 
2873         /*
2874          * If an allocation failed after direct reclaim, it could be because
2875          * pages are pinned on the per-cpu lists. Drain them and try again
2876          */
2877         if (!page && !drained) {
2878                 drain_all_pages(NULL);
2879                 drained = true;
2880                 goto retry;
2881         }
2882 
2883         return page;
2884 }
2885 
2886 /*
2887  * This is called in the allocator slow-path if the allocation request is of
2888  * sufficient urgency to ignore watermarks and take other desperate measures
2889  */
2890 static inline struct page *
2891 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2892                                 const struct alloc_context *ac)
2893 {
2894         struct page *page;
2895 
2896         do {
2897                 page = get_page_from_freelist(gfp_mask, order,
2898                                                 ALLOC_NO_WATERMARKS, ac);
2899 
2900                 if (!page && gfp_mask & __GFP_NOFAIL)
2901                         wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2902                                                                         HZ/50);
2903         } while (!page && (gfp_mask & __GFP_NOFAIL));
2904 
2905         return page;
2906 }
2907 
2908 static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
2909 {
2910         struct zoneref *z;
2911         struct zone *zone;
2912 
2913         for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2914                                                 ac->high_zoneidx, ac->nodemask)
2915                 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
2916 }
2917 
2918 static inline int
2919 gfp_to_alloc_flags(gfp_t gfp_mask)
2920 {
2921         int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2922         const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
2923 
2924         /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2925         BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2926 
2927         /*
2928          * The caller may dip into page reserves a bit more if the caller
2929          * cannot run direct reclaim, or if the caller has realtime scheduling
2930          * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
2931          * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
2932          */
2933         alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2934 
2935         if (atomic) {
2936                 /*
2937                  * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2938                  * if it can't schedule.
2939                  */
2940                 if (!(gfp_mask & __GFP_NOMEMALLOC))
2941                         alloc_flags |= ALLOC_HARDER;
2942                 /*
2943                  * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2944                  * comment for __cpuset_node_allowed().
2945                  */
2946                 alloc_flags &= ~ALLOC_CPUSET;
2947         } else if (unlikely(rt_task(current)) && !in_interrupt())
2948                 alloc_flags |= ALLOC_HARDER;
2949 
2950         if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2951                 if (gfp_mask & __GFP_MEMALLOC)
2952                         alloc_flags |= ALLOC_NO_WATERMARKS;
2953                 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2954                         alloc_flags |= ALLOC_NO_WATERMARKS;
2955                 else if (!in_interrupt() &&
2956                                 ((current->flags & PF_MEMALLOC) ||
2957                                  unlikely(test_thread_flag(TIF_MEMDIE))))
2958                         alloc_flags |= ALLOC_NO_WATERMARKS;
2959         }
2960 #ifdef CONFIG_CMA
2961         if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2962                 alloc_flags |= ALLOC_CMA;
2963 #endif
2964         return alloc_flags;
2965 }
2966 
2967 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2968 {
2969         return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
2970 }
2971 
2972 static inline struct page *
2973 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2974                                                 struct alloc_context *ac)
2975 {
2976         const gfp_t wait = gfp_mask & __GFP_WAIT;
2977         struct page *page = NULL;
2978         int alloc_flags;
2979         unsigned long pages_reclaimed = 0;
2980         unsigned long did_some_progress;
2981         enum migrate_mode migration_mode = MIGRATE_ASYNC;
2982         bool deferred_compaction = false;
2983         int contended_compaction = COMPACT_CONTENDED_NONE;
2984 
2985         /*
2986          * In the slowpath, we sanity check order to avoid ever trying to
2987          * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2988          * be using allocators in order of preference for an area that is
2989          * too large.
2990          */
2991         if (order >= MAX_ORDER) {
2992                 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2993                 return NULL;
2994         }
2995 
2996         /*
2997          * If this allocation cannot block and it is for a specific node, then
2998          * fail early.  There's no need to wakeup kswapd or retry for a
2999          * speculative node-specific allocation.
3000          */
3001         if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !wait)
3002                 goto nopage;
3003 
3004 retry:
3005         if (!(gfp_mask & __GFP_NO_KSWAPD))
3006                 wake_all_kswapds(order, ac);
3007 
3008         /*
3009          * OK, we're below the kswapd watermark and have kicked background
3010          * reclaim. Now things get more complex, so set up alloc_flags according
3011          * to how we want to proceed.
3012          */
3013         alloc_flags = gfp_to_alloc_flags(gfp_mask);
3014 
3015         /*
3016          * Find the true preferred zone if the allocation is unconstrained by
3017          * cpusets.
3018          */
3019         if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
3020                 struct zoneref *preferred_zoneref;
3021                 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3022                                 ac->high_zoneidx, NULL, &ac->preferred_zone);
3023                 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
3024         }
3025 
3026         /* This is the last chance, in general, before the goto nopage. */
3027         page = get_page_from_freelist(gfp_mask, order,
3028                                 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3029         if (page)
3030                 goto got_pg;
3031 
3032         /* Allocate without watermarks if the context allows */
3033         if (alloc_flags & ALLOC_NO_WATERMARKS) {
3034                 /*
3035                  * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3036                  * the allocation is high priority and these type of
3037                  * allocations are system rather than user orientated
3038                  */
3039                 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3040 
3041                 page = __alloc_pages_high_priority(gfp_mask, order, ac);
3042 
3043                 if (page) {
3044                         goto got_pg;
3045                 }
3046         }
3047 
3048         /* Atomic allocations - we can't balance anything */
3049         if (!wait) {
3050                 /*
3051                  * All existing users of the deprecated __GFP_NOFAIL are
3052                  * blockable, so warn of any new users that actually allow this
3053                  * type of allocation to fail.
3054                  */
3055                 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
3056                 goto nopage;
3057         }
3058 
3059         /* Avoid recursion of direct reclaim */
3060         if (current->flags & PF_MEMALLOC)
3061                 goto nopage;
3062 
3063         /* Avoid allocations with no watermarks from looping endlessly */
3064         if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3065                 goto nopage;
3066 
3067         /*
3068          * Try direct compaction. The first pass is asynchronous. Subsequent
3069          * attempts after direct reclaim are synchronous
3070          */
3071         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3072                                         migration_mode,
3073                                         &contended_compaction,
3074                                         &deferred_compaction);
3075         if (page)
3076                 goto got_pg;
3077 
3078         /* Checks for THP-specific high-order allocations */
3079         if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
3080                 /*
3081                  * If compaction is deferred for high-order allocations, it is
3082                  * because sync compaction recently failed. If this is the case
3083                  * and the caller requested a THP allocation, we do not want
3084                  * to heavily disrupt the system, so we fail the allocation
3085                  * instead of entering direct reclaim.
3086                  */
3087                 if (deferred_compaction)
3088                         goto nopage;
3089 
3090                 /*
3091                  * In all zones where compaction was attempted (and not
3092                  * deferred or skipped), lock contention has been detected.
3093                  * For THP allocation we do not want to disrupt the others
3094                  * so we fallback to base pages instead.
3095                  */
3096                 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3097                         goto nopage;
3098 
3099                 /*
3100                  * If compaction was aborted due to need_resched(), we do not
3101                  * want to further increase allocation latency, unless it is
3102                  * khugepaged trying to collapse.
3103                  */
3104                 if (contended_compaction == COMPACT_CONTENDED_SCHED
3105                         && !(current->flags & PF_KTHREAD))
3106                         goto nopage;
3107         }
3108 
3109         /*
3110          * It can become very expensive to allocate transparent hugepages at
3111          * fault, so use asynchronous memory compaction for THP unless it is
3112          * khugepaged trying to collapse.
3113          */
3114         if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
3115                                                 (current->flags & PF_KTHREAD))
3116                 migration_mode = MIGRATE_SYNC_LIGHT;
3117 
3118         /* Try direct reclaim and then allocating */
3119         page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3120                                                         &did_some_progress);
3121         if (page)
3122                 goto got_pg;
3123 
3124         /* Do not loop if specifically requested */
3125         if (gfp_mask & __GFP_NORETRY)
3126                 goto noretry;
3127 
3128         /* Keep reclaiming pages as long as there is reasonable progress */
3129         pages_reclaimed += did_some_progress;
3130         if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3131             ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
3132                 /* Wait for some write requests to complete then retry */
3133                 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
3134                 goto retry;
3135         }
3136 
3137         /* Reclaim has failed us, start killing things */
3138         page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3139         if (page)
3140                 goto got_pg;
3141 
3142         /* Retry as long as the OOM killer is making progress */
3143         if (did_some_progress)
3144                 goto retry;
3145 
3146 noretry:
3147         /*
3148          * High-order allocations do not necessarily loop after
3149          * direct reclaim and reclaim/compaction depends on compaction
3150          * being called after reclaim so call directly if necessary
3151          */
3152         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3153                                             ac, migration_mode,
3154                                             &contended_compaction,
3155                                             &deferred_compaction);
3156         if (page)
3157                 goto got_pg;
3158 nopage:
3159         warn_alloc_failed(gfp_mask, order, NULL);
3160 got_pg:
3161         return page;
3162 }
3163 
3164 /*
3165  * This is the 'heart' of the zoned buddy allocator.
3166  */
3167 struct page *
3168 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3169                         struct zonelist *zonelist, nodemask_t *nodemask)
3170 {
3171         struct zoneref *preferred_zoneref;
3172         struct page *page = NULL;
3173         unsigned int cpuset_mems_cookie;
3174         int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
3175         gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
3176         struct alloc_context ac = {
3177                 .high_zoneidx = gfp_zone(gfp_mask),
3178                 .nodemask = nodemask,
3179                 .migratetype = gfpflags_to_migratetype(gfp_mask),
3180         };
3181 
3182         gfp_mask &= gfp_allowed_mask;
3183 
3184         lockdep_trace_alloc(gfp_mask);
3185 
3186         might_sleep_if(gfp_mask & __GFP_WAIT);
3187 
3188         if (should_fail_alloc_page(gfp_mask, order))
3189                 return NULL;
3190 
3191         /*
3192          * Check the zones suitable for the gfp_mask contain at least one
3193          * valid zone. It's possible to have an empty zonelist as a result
3194          * of __GFP_THISNODE and a memoryless node
3195          */
3196         if (unlikely(!zonelist->_zonerefs->zone))
3197                 return NULL;
3198 
3199         if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
3200                 alloc_flags |= ALLOC_CMA;
3201 
3202 retry_cpuset:
3203         cpuset_mems_cookie = read_mems_allowed_begin();
3204 
3205         /* We set it here, as __alloc_pages_slowpath might have changed it */
3206         ac.zonelist = zonelist;
3207         /* The preferred zone is used for statistics later */
3208         preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3209                                 ac.nodemask ? : &cpuset_current_mems_allowed,
3210                                 &ac.preferred_zone);
3211         if (!ac.preferred_zone)
3212                 goto out;
3213         ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
3214 
3215         /* First allocation attempt */
3216         alloc_mask = gfp_mask|__GFP_HARDWALL;
3217         page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
3218         if (unlikely(!page)) {
3219                 /*
3220                  * Runtime PM, block IO and its error handling path
3221                  * can deadlock because I/O on the device might not
3222                  * complete.
3223                  */
3224                 alloc_mask = memalloc_noio_flags(gfp_mask);
3225 
3226                 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
3227         }
3228 
3229         if (kmemcheck_enabled && page)
3230                 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3231 
3232         trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3233 
3234 out:
3235         /*
3236          * When updating a task's mems_allowed, it is possible to race with
3237          * parallel threads in such a way that an allocation can fail while
3238          * the mask is being updated. If a page allocation is about to fail,
3239          * check if the cpuset changed during allocation and if so, retry.
3240          */
3241         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
3242                 goto retry_cpuset;
3243 
3244         return page;
3245 }
3246 EXPORT_SYMBOL(__alloc_pages_nodemask);
3247 
3248 /*
3249  * Common helper functions.
3250  */
3251 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
3252 {
3253         struct page *page;
3254 
3255         /*
3256          * __get_free_pages() returns a 32-bit address, which cannot represent
3257          * a highmem page
3258          */
3259         VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3260 
3261         page = alloc_pages(gfp_mask, order);
3262         if (!page)
3263                 return 0;
3264         return (unsigned long) page_address(page);
3265 }
3266 EXPORT_SYMBOL(__get_free_pages);
3267 
3268 unsigned long get_zeroed_page(gfp_t gfp_mask)
3269 {
3270         return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
3271 }
3272 EXPORT_SYMBOL(get_zeroed_page);
3273 
3274 void __free_pages(struct page *page, unsigned int order)
3275 {
3276         if (put_page_testzero(page)) {
3277                 if (order == 0)
3278                         free_hot_cold_page(page, false);
3279                 else
3280                         __free_pages_ok(page, order);
3281         }
3282 }
3283 
3284 EXPORT_SYMBOL(__free_pages);
3285 
3286 void free_pages(unsigned long addr, unsigned int order)
3287 {
3288         if (addr != 0) {
3289                 VM_BUG_ON(!virt_addr_valid((void *)addr));
3290                 __free_pages(virt_to_page((void *)addr), order);
3291         }
3292 }
3293 
3294 EXPORT_SYMBOL(free_pages);
3295 
3296 /*
3297  * Page Fragment:
3298  *  An arbitrary-length arbitrary-offset area of memory which resides
3299  *  within a 0 or higher order page.  Multiple fragments within that page
3300  *  are individually refcounted, in the page's reference counter.
3301  *
3302  * The page_frag functions below provide a simple allocation framework for
3303  * page fragments.  This is used by the network stack and network device
3304  * drivers to provide a backing region of memory for use as either an
3305  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3306  */
3307 static struct page *__page_frag_refill(struct page_frag_cache *nc,
3308                                        gfp_t gfp_mask)
3309 {
3310         struct page *page = NULL;
3311         gfp_t gfp = gfp_mask;
3312 
3313 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3314         gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3315                     __GFP_NOMEMALLOC;
3316         page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3317                                 PAGE_FRAG_CACHE_MAX_ORDER);
3318         nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3319 #endif
3320         if (unlikely(!page))
3321                 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3322 
3323         nc->va = page ? page_address(page) : NULL;
3324 
3325         return page;
3326 }
3327 
3328 void *__alloc_page_frag(struct page_frag_cache *nc,
3329                         unsigned int fragsz, gfp_t gfp_mask)
3330 {
3331         unsigned int size = PAGE_SIZE;
3332         struct page *page;
3333         int offset;
3334 
3335         if (unlikely(!nc->va)) {
3336 refill:
3337                 page = __page_frag_refill(nc, gfp_mask);
3338                 if (!page)
3339                         return NULL;
3340 
3341 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3342                 /* if size can vary use size else just use PAGE_SIZE */
3343                 size = nc->size;
3344 #endif
3345                 /* Even if we own the page, we do not use atomic_set().
3346                  * This would break get_page_unless_zero() users.
3347                  */
3348                 atomic_add(size - 1, &page->_count);
3349 
3350                 /* reset page count bias and offset to start of new frag */
3351                 nc->pfmemalloc = page_is_pfmemalloc(page);
3352                 nc->pagecnt_bias = size;
3353                 nc->offset = size;
3354         }
3355 
3356         offset = nc->offset - fragsz;
3357         if (unlikely(offset < 0)) {
3358                 page = virt_to_page(nc->va);
3359 
3360                 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
3361                         goto refill;
3362 
3363 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3364                 /* if size can vary use size else just use PAGE_SIZE */
3365                 size = nc->size;
3366 #endif
3367                 /* OK, page count is 0, we can safely set it */
3368                 atomic_set(&page->_count, size);
3369 
3370                 /* reset page count bias and offset to start of new frag */
3371                 nc->pagecnt_bias = size;
3372                 offset = size - fragsz;
3373         }
3374 
3375         nc->pagecnt_bias--;
3376         nc->offset = offset;
3377 
3378         return nc->va + offset;
3379 }
3380 EXPORT_SYMBOL(__alloc_page_frag);
3381 
3382 /*
3383  * Frees a page fragment allocated out of either a compound or order 0 page.
3384  */
3385 void __free_page_frag(void *addr)
3386 {
3387         struct page *page = virt_to_head_page(addr);
3388 
3389         if (unlikely(put_page_testzero(page)))
3390                 __free_pages_ok(page, compound_order(page));
3391 }
3392 EXPORT_SYMBOL(__free_page_frag);
3393 
3394 /*
3395  * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3396  * of the current memory cgroup.
3397  *
3398  * It should be used when the caller would like to use kmalloc, but since the
3399  * allocation is large, it has to fall back to the page allocator.
3400  */
3401 struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3402 {
3403         struct page *page;
3404         struct mem_cgroup *memcg = NULL;
3405 
3406         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3407                 return NULL;
3408         page = alloc_pages(gfp_mask, order);
3409         memcg_kmem_commit_charge(page, memcg, order);
3410         return page;
3411 }
3412 
3413 struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3414 {
3415         struct page *page;
3416         struct mem_cgroup *memcg = NULL;
3417 
3418         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3419                 return NULL;
3420         page = alloc_pages_node(nid, gfp_mask, order);
3421         memcg_kmem_commit_charge(page, memcg, order);
3422         return page;
3423 }
3424 
3425 /*
3426  * __free_kmem_pages and free_kmem_pages will free pages allocated with
3427  * alloc_kmem_pages.
3428  */
3429 void __free_kmem_pages(struct page *page, unsigned int order)
3430 {
3431         memcg_kmem_uncharge_pages(page, order);
3432         __free_pages(page, order);
3433 }
3434 
3435 void free_kmem_pages(unsigned long addr, unsigned int order)
3436 {
3437         if (addr != 0) {
3438                 VM_BUG_ON(!virt_addr_valid((void *)addr));
3439                 __free_kmem_pages(virt_to_page((void *)addr), order);
3440         }
3441 }
3442 
3443 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
3444 {
3445         if (addr) {
3446                 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3447                 unsigned long used = addr + PAGE_ALIGN(size);
3448 
3449                 split_page(virt_to_page((void *)addr), order);
3450                 while (used < alloc_end) {
3451                         free_page(used);
3452                         used += PAGE_SIZE;
3453                 }
3454         }
3455         return (void *)addr;
3456 }
3457 
3458 /**
3459  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3460  * @size: the number of bytes to allocate
3461  * @gfp_mask: GFP flags for the allocation
3462  *
3463  * This function is similar to alloc_pages(), except that it allocates the
3464  * minimum number of pages to satisfy the request.  alloc_pages() can only
3465  * allocate memory in power-of-two pages.
3466  *
3467  * This function is also limited by MAX_ORDER.
3468  *
3469  * Memory allocated by this function must be released by free_pages_exact().
3470  */
3471 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3472 {
3473         unsigned int order = get_order(size);
3474         unsigned long addr;
3475 
3476         addr = __get_free_pages(gfp_mask, order);
3477         return make_alloc_exact(addr, order, size);
3478 }
3479 EXPORT_SYMBOL(alloc_pages_exact);
3480 
3481 /**
3482  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3483  *                         pages on a node.
3484  * @nid: the preferred node ID where memory should be allocated
3485  * @size: the number of bytes to allocate
3486  * @gfp_mask: GFP flags for the allocation
3487  *
3488  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3489  * back.
3490  * Note this is not alloc_pages_exact_node() which allocates on a specific node,
3491  * but is not exact.
3492  */
3493 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
3494 {
3495         unsigned order = get_order(size);
3496         struct page *p = alloc_pages_node(nid, gfp_mask, order);
3497         if (!p)
3498                 return NULL;
3499         return make_alloc_exact((unsigned long)page_address(p), order, size);
3500 }
3501 
3502 /**
3503  * free_pages_exact - release memory allocated via alloc_pages_exact()
3504  * @virt: the value returned by alloc_pages_exact.
3505  * @size: size of allocation, same value as passed to alloc_pages_exact().
3506  *
3507  * Release the memory allocated by a previous call to alloc_pages_exact.
3508  */
3509 void free_pages_exact(void *virt, size_t size)
3510 {
3511         unsigned long addr = (unsigned long)virt;
3512         unsigned long end = addr + PAGE_ALIGN(size);
3513 
3514         while (addr < end) {
3515                 free_page(addr);
3516                 addr += PAGE_SIZE;
3517         }
3518 }
3519 EXPORT_SYMBOL(free_pages_exact);
3520 
3521 /**
3522  * nr_free_zone_pages - count number of pages beyond high watermark
3523  * @offset: The zone index of the highest zone
3524  *
3525  * nr_free_zone_pages() counts the number of counts pages which are beyond the
3526  * high watermark within all zones at or below a given zone index.  For each
3527  * zone, the number of pages is calculated as:
3528  *     managed_pages - high_pages
3529  */
3530 static unsigned long nr_free_zone_pages(int offset)
3531 {
3532         struct zoneref *z;
3533         struct zone *zone;
3534 
3535         /* Just pick one node, since fallback list is circular */
3536         unsigned long sum = 0;
3537 
3538         struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3539 
3540         for_each_zone_zonelist(zone, z, zonelist, offset) {
3541                 unsigned long size = zone->managed_pages;
3542                 unsigned long high = high_wmark_pages(zone);
3543                 if (size > high)
3544                         sum += size - high;
3545         }
3546 
3547         return sum;
3548 }
3549 
3550 /**
3551  * nr_free_buffer_pages - count number of pages beyond high watermark
3552  *
3553  * nr_free_buffer_pages() counts the number of pages which are beyond the high
3554  * watermark within ZONE_DMA and ZONE_NORMAL.
3555  */
3556 unsigned long nr_free_buffer_pages(void)
3557 {
3558         return nr_free_zone_pages(gfp_zone(GFP_USER));
3559 }
3560 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3561 
3562 /**
3563  * nr_free_pagecache_pages - count number of pages beyond high watermark
3564  *
3565  * nr_free_pagecache_pages() counts the number of pages which are beyond the
3566  * high watermark within all zones.
3567  */
3568 unsigned long nr_free_pagecache_pages(void)
3569 {
3570         return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3571 }
3572 
3573 static inline void show_node(struct zone *zone)
3574 {
3575         if (IS_ENABLED(CONFIG_NUMA))
3576                 printk("Node %d ", zone_to_nid(zone));
3577 }
3578 
3579 void si_meminfo(struct sysinfo *val)
3580 {
3581         val->totalram = totalram_pages;
3582         val->sharedram = global_page_state(NR_SHMEM);
3583         val->freeram = global_page_state(NR_FREE_PAGES);
3584         val->bufferram = nr_blockdev_pages();
3585         val->totalhigh = totalhigh_pages;
3586         val->freehigh = nr_free_highpages();
3587         val->mem_unit = PAGE_SIZE;
3588 }
3589 
3590 EXPORT_SYMBOL(si_meminfo);
3591 
3592 #ifdef CONFIG_NUMA
3593 void si_meminfo_node(struct sysinfo *val, int nid)
3594 {
3595         int zone_type;          /* needs to be signed */
3596         unsigned long managed_pages = 0;
3597         pg_data_t *pgdat = NODE_DATA(nid);
3598 
3599         for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3600                 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3601         val->totalram = managed_pages;
3602         val->sharedram = node_page_state(nid, NR_SHMEM);
3603         val->freeram = node_page_state(nid, NR_FREE_PAGES);
3604 #ifdef CONFIG_HIGHMEM
3605         val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
3606         val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3607                         NR_FREE_PAGES);
3608 #else
3609         val->totalhigh = 0;
3610         val->freehigh = 0;
3611 #endif
3612         val->mem_unit = PAGE_SIZE;
3613 }
3614 #endif
3615 
3616 /*
3617  * Determine whether the node should be displayed or not, depending on whether
3618  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3619  */
3620 bool skip_free_areas_node(unsigned int flags, int nid)
3621 {
3622         bool ret = false;
3623         unsigned int cpuset_mems_cookie;
3624 
3625         if (!(flags & SHOW_MEM_FILTER_NODES))
3626                 goto out;
3627 
3628         do {
3629                 cpuset_mems_cookie = read_mems_allowed_begin();
3630                 ret = !node_isset(nid, cpuset_current_mems_allowed);
3631         } while (read_mems_allowed_retry(cpuset_mems_cookie));
3632 out:
3633         return ret;
3634 }
3635 
3636 #define K(x) ((x) << (PAGE_SHIFT-10))
3637 
3638 static void show_migration_types(unsigned char type)
3639 {
3640         static const char types[MIGRATE_TYPES] = {
3641                 [MIGRATE_UNMOVABLE]     = 'U',
3642                 [MIGRATE_RECLAIMABLE]   = 'E',
3643                 [MIGRATE_MOVABLE]       = 'M',
3644                 [MIGRATE_RESERVE]       = 'R',
3645 #ifdef CONFIG_CMA
3646                 [MIGRATE_CMA]           = 'C',
3647 #endif
3648 #ifdef CONFIG_MEMORY_ISOLATION
3649                 [MIGRATE_ISOLATE]       = 'I',
3650 #endif
3651         };
3652         char tmp[MIGRATE_TYPES + 1];
3653         char *p = tmp;
3654         int i;
3655 
3656         for (i = 0; i < MIGRATE_TYPES; i++) {
3657                 if (type & (1 << i))
3658                         *p++ = types[i];
3659         }
3660 
3661         *p = '\0';
3662         printk("(%s) ", tmp);
3663 }
3664 
3665 /*
3666  * Show free area list (used inside shift_scroll-lock stuff)
3667  * We also calculate the percentage fragmentation. We do this by counting the
3668  * memory on each free list with the exception of the first item on the list.
3669  *
3670  * Bits in @filter:
3671  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3672  *   cpuset.
3673  */
3674 void show_free_areas(unsigned int filter)
3675 {
3676         unsigned long free_pcp = 0;
3677         int cpu;
3678         struct zone *zone;
3679 
3680         for_each_populated_zone(zone) {
3681                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3682                         continue;
3683 
3684                 for_each_online_cpu(cpu)
3685                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3686         }
3687 
3688         printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3689                 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3690                 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3691                 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3692                 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3693                 " free:%lu free_pcp:%lu free_cma:%lu\n",
3694                 global_page_state(NR_ACTIVE_ANON),
3695                 global_page_state(NR_INACTIVE_ANON),
3696                 global_page_state(NR_ISOLATED_ANON),
3697                 global_page_state(NR_ACTIVE_FILE),
3698                 global_page_state(NR_INACTIVE_FILE),
3699                 global_page_state(NR_ISOLATED_FILE),
3700                 global_page_state(NR_UNEVICTABLE),
3701                 global_page_state(NR_FILE_DIRTY),
3702                 global_page_state(NR_WRITEBACK),
3703                 global_page_state(NR_UNSTABLE_NFS),
3704                 global_page_state(NR_SLAB_RECLAIMABLE),
3705                 global_page_state(NR_SLAB_UNRECLAIMABLE),
3706                 global_page_state(NR_FILE_MAPPED),
3707                 global_page_state(NR_SHMEM),
3708                 global_page_state(NR_PAGETABLE),
3709                 global_page_state(NR_BOUNCE),
3710                 global_page_state(NR_FREE_PAGES),
3711                 free_pcp,
3712                 global_page_state(NR_FREE_CMA_PAGES));
3713 
3714         for_each_populated_zone(zone) {
3715                 int i;
3716 
3717                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3718                         continue;
3719 
3720                 free_pcp = 0;
3721                 for_each_online_cpu(cpu)
3722                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3723 
3724                 show_node(zone);
3725                 printk("%s"
3726                         " free:%lukB"
3727                         " min:%lukB"
3728                         " low:%lukB"
3729                         " high:%lukB"
3730                         " active_anon:%lukB"
3731                         " inactive_anon:%lukB"
3732                         " active_file:%lukB"
3733                         " inactive_file:%lukB"
3734                         " unevictable:%lukB"
3735                         " isolated(anon):%lukB"
3736                         " isolated(file):%lukB"
3737                         " present:%lukB"
3738                         " managed:%lukB"
3739                         " mlocked:%lukB"
3740                         " dirty:%lukB"
3741                         " writeback:%lukB"
3742                         " mapped:%lukB"
3743                         " shmem:%lukB"
3744                         " slab_reclaimable:%lukB"
3745                         " slab_unreclaimable:%lukB"
3746                         " kernel_stack:%lukB"
3747                         " pagetables:%lukB"
3748                         " unstable:%lukB"
3749                         " bounce:%lukB"
3750                         " free_pcp:%lukB"
3751                         " local_pcp:%ukB"
3752                         " free_cma:%lukB"
3753                         " writeback_tmp:%lukB"
3754                         " pages_scanned:%lu"
3755                         " all_unreclaimable? %s"
3756                         "\n",
3757                         zone->name,
3758                         K(zone_page_state(zone, NR_FREE_PAGES)),
3759                         K(min_wmark_pages(zone)),
3760                         K(low_wmark_pages(zone)),
3761                         K(high_wmark_pages(zone)),
3762                         K(zone_page_state(zone, NR_ACTIVE_ANON)),
3763                         K(zone_page_state(zone, NR_INACTIVE_ANON)),
3764                         K(zone_page_state(zone, NR_ACTIVE_FILE)),
3765                         K(zone_page_state(zone, NR_INACTIVE_FILE)),
3766                         K(zone_page_state(zone, NR_UNEVICTABLE)),
3767                         K(zone_page_state(zone, NR_ISOLATED_ANON)),
3768                         K(zone_page_state(zone, NR_ISOLATED_FILE)),
3769                         K(zone->present_pages),
3770                         K(zone->managed_pages),
3771                         K(zone_page_state(zone, NR_MLOCK)),
3772                         K(zone_page_state(zone, NR_FILE_DIRTY)),
3773                         K(zone_page_state(zone, NR_WRITEBACK)),
3774                         K(zone_page_state(zone, NR_FILE_MAPPED)),
3775                         K(zone_page_state(zone, NR_SHMEM)),
3776                         K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3777                         K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3778                         zone_page_state(zone, NR_KERNEL_STACK) *
3779                                 THREAD_SIZE / 1024,
3780                         K(zone_page_state(zone, NR_PAGETABLE)),
3781                         K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3782                         K(zone_page_state(zone, NR_BOUNCE)),
3783                         K(free_pcp),
3784                         K(this_cpu_read(zone->pageset->pcp.count)),
3785                         K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3786                         K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3787                         K(zone_page_state(zone, NR_PAGES_SCANNED)),
3788                         (!zone_reclaimable(zone) ? "yes" : "no")
3789                         );
3790                 printk("lowmem_reserve[]:");
3791                 for (i = 0; i < MAX_NR_ZONES; i++)
3792                         printk(" %ld", zone->lowmem_reserve[i]);
3793                 printk("\n");
3794         }
3795 
3796         for_each_populated_zone(zone) {
3797                 unsigned long nr[MAX_ORDER], flags, order, total = 0;
3798                 unsigned char types[MAX_ORDER];
3799 
3800                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3801                         continue;
3802                 show_node(zone);
3803                 printk("%s: ", zone->name);
3804 
3805                 spin_lock_irqsave(&zone->lock, flags);
3806                 for (order = 0; order < MAX_ORDER; order++) {
3807                         struct free_area *area = &zone->free_area[order];
3808                         int type;
3809 
3810                         nr[order] = area->nr_free;
3811                         total += nr[order] << order;
3812 
3813                         types[order] = 0;
3814                         for (type = 0; type < MIGRATE_TYPES; type++) {
3815                                 if (!list_empty(&area->free_list[type]))
3816                                         types[order] |= 1 << type;
3817                         }
3818                 }
3819                 spin_unlock_irqrestore(&zone->lock, flags);
3820                 for (order = 0; order < MAX_ORDER; order++) {
3821                         printk("%lu*%lukB ", nr[order], K(1UL) << order);
3822                         if (nr[order])
3823                                 show_migration_types(types[order]);
3824                 }
3825                 printk("= %lukB\n", K(total));
3826         }
3827 
3828         hugetlb_show_meminfo();
3829 
3830         printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3831 
3832         show_swap_cache_info();
3833 }
3834 
3835 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3836 {
3837         zoneref->zone = zone;
3838         zoneref->zone_idx = zone_idx(zone);
3839 }
3840 
3841 /*
3842  * Builds allocation fallback zone lists.
3843  *
3844  * Add all populated zones of a node to the zonelist.
3845  */
3846 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3847                                 int nr_zones)
3848 {
3849         struct zone *zone;
3850         enum zone_type zone_type = MAX_NR_ZONES;
3851 
3852         do {
3853                 zone_type--;
3854                 zone = pgdat->node_zones + zone_type;
3855                 if (populated_zone(zone)) {
3856                         zoneref_set_zone(zone,
3857                                 &zonelist->_zonerefs[nr_zones++]);
3858                         check_highest_zone(zone_type);
3859                 }
3860         } while (zone_type);
3861 
3862         return nr_zones;
3863 }
3864 
3865 
3866 /*
3867  *  zonelist_order:
3868  *  0 = automatic detection of better ordering.
3869  *  1 = order by ([node] distance, -zonetype)
3870  *  2 = order by (-zonetype, [node] distance)
3871  *
3872  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3873  *  the same zonelist. So only NUMA can configure this param.
3874  */
3875 #define ZONELIST_ORDER_DEFAULT  0
3876 #define ZONELIST_ORDER_NODE     1
3877 #define ZONELIST_ORDER_ZONE     2
3878 
3879 /* zonelist order in the kernel.
3880  * set_zonelist_order() will set this to NODE or ZONE.
3881  */
3882 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3883 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3884 
3885 
3886 #ifdef CONFIG_NUMA
3887 /* The value user specified ....changed by config */
3888 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3889 /* string for sysctl */
3890 #define NUMA_ZONELIST_ORDER_LEN 16
3891 char numa_zonelist_order[16] = "default";
3892 
3893 /*
3894  * interface for configure zonelist ordering.
3895  * command line option "numa_zonelist_order"
3896  *      = "[dD]efault   - default, automatic configuration.
3897  *      = "[nN]ode      - order by node locality, then by zone within node
3898  *      = "[zZ]one      - order by zone, then by locality within zone
3899  */
3900 
3901 static int __parse_numa_zonelist_order(char *s)
3902 {
3903         if (*s == 'd' || *s == 'D') {
3904                 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3905         } else if (*s == 'n' || *s == 'N') {
3906                 user_zonelist_order = ZONELIST_ORDER_NODE;
3907         } else if (*s == 'z' || *s == 'Z') {
3908                 user_zonelist_order = ZONELIST_ORDER_ZONE;
3909         } else {
3910                 printk(KERN_WARNING
3911                         "Ignoring invalid numa_zonelist_order value:  "
3912                         "%s\n", s);
3913                 return -EINVAL;
3914         }
3915         return 0;
3916 }
3917 
3918 static __init int setup_numa_zonelist_order(char *s)
3919 {
3920         int ret;
3921 
3922         if (!s)
3923                 return 0;
3924 
3925         ret = __parse_numa_zonelist_order(s);
3926         if (ret == 0)
3927                 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3928 
3929         return ret;
3930 }
3931 early_param("numa_zonelist_order", setup_numa_zonelist_order);
3932 
3933 /*
3934  * sysctl handler for numa_zonelist_order
3935  */
3936 int numa_zonelist_order_handler(struct ctl_table *table, int write,
3937                 void __user *buffer, size_t *length,
3938                 loff_t *ppos)
3939 {
3940         char saved_string[NUMA_ZONELIST_ORDER_LEN];
3941         int ret;
3942         static DEFINE_MUTEX(zl_order_mutex);
3943 
3944         mutex_lock(&zl_order_mutex);
3945         if (write) {
3946                 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3947                         ret = -EINVAL;
3948                         goto out;
3949                 }
3950                 strcpy(saved_string, (char *)table->data);
3951         }
3952         ret = proc_dostring(table, write, buffer, length, ppos);
3953         if (ret)
3954                 goto out;
3955         if (write) {
3956                 int oldval = user_zonelist_order;
3957 
3958                 ret = __parse_numa_zonelist_order((char *)table->data);
3959                 if (ret) {
3960                         /*
3961                          * bogus value.  restore saved string
3962                          */
3963                         strncpy((char *)table->data, saved_string,
3964                                 NUMA_ZONELIST_ORDER_LEN);
3965                         user_zonelist_order = oldval;
3966                 } else if (oldval != user_zonelist_order) {
3967                         mutex_lock(&zonelists_mutex);
3968                         build_all_zonelists(NULL, NULL);
3969                         mutex_unlock(&zonelists_mutex);
3970                 }
3971         }
3972 out:
3973         mutex_unlock(&zl_order_mutex);
3974         return ret;
3975 }
3976 
3977 
3978 #define MAX_NODE_LOAD (nr_online_nodes)
3979 static int node_load[MAX_NUMNODES];
3980 
3981 /**
3982  * find_next_best_node - find the next node that should appear in a given node's fallback list
3983  * @node: node whose fallback list we're appending
3984  * @used_node_mask: nodemask_t of already used nodes
3985  *
3986  * We use a number of factors to determine which is the next node that should
3987  * appear on a given node's fallback list.  The node should not have appeared
3988  * already in @node's fallback list, and it should be the next closest node
3989  * according to the distance array (which contains arbitrary distance values
3990  * from each node to each node in the system), and should also prefer nodes
3991  * with no CPUs, since presumably they'll have very little allocation pressure
3992  * on them otherwise.
3993  * It returns -1 if no node is found.
3994  */
3995 static int find_next_best_node(int node, nodemask_t *used_node_mask)
3996 {
3997         int n, val;
3998         int min_val = INT_MAX;
3999         int best_node = NUMA_NO_NODE;
4000         const struct cpumask *tmp = cpumask_of_node(0);
4001 
4002         /* Use the local node if we haven't already */
4003         if (!node_isset(node, *used_node_mask)) {
4004                 node_set(node, *used_node_mask);
4005                 return node;
4006         }
4007 
4008         for_each_node_state(n, N_MEMORY) {
4009 
4010                 /* Don't want a node to appear more than once */
4011                 if (node_isset(n, *used_node_mask))
4012                         continue;
4013 
4014                 /* Use the distance array to find the distance */
4015                 val = node_distance(node, n);
4016 
4017                 /* Penalize nodes under us ("prefer the next node") */
4018                 val += (n < node);
4019 
4020                 /* Give preference to headless and unused nodes */
4021                 tmp = cpumask_of_node(n);
4022                 if (!cpumask_empty(tmp))
4023                         val += PENALTY_FOR_NODE_WITH_CPUS;
4024 
4025                 /* Slight preference for less loaded node */
4026                 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4027                 val += node_load[n];
4028 
4029                 if (val < min_val) {
4030                         min_val = val;
4031                         best_node = n;
4032                 }
4033         }
4034 
4035         if (best_node >= 0)
4036                 node_set(best_node, *used_node_mask);
4037 
4038         return best_node;
4039 }
4040 
4041 
4042 /*
4043  * Build zonelists ordered by node and zones within node.
4044  * This results in maximum locality--normal zone overflows into local
4045  * DMA zone, if any--but risks exhausting DMA zone.
4046  */
4047 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
4048 {
4049         int j;
4050         struct zonelist *zonelist;
4051 
4052         zonelist = &pgdat->node_zonelists[0];
4053         for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
4054                 ;
4055         j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4056         zonelist->_zonerefs[j].zone = NULL;
4057         zonelist->_zonerefs[j].zone_idx = 0;
4058 }
4059 
4060 /*
4061  * Build gfp_thisnode zonelists
4062  */
4063 static void build_thisnode_zonelists(pg_data_t *pgdat)
4064 {
4065         int j;
4066         struct zonelist *zonelist;
4067 
4068         zonelist = &pgdat->node_zonelists[1];
4069         j = build_zonelists_node(pgdat, zonelist, 0);
4070         zonelist->_zonerefs[j].zone = NULL;
4071         zonelist->_zonerefs[j].zone_idx = 0;
4072 }
4073 
4074 /*
4075  * Build zonelists ordered by zone and nodes within zones.
4076  * This results in conserving DMA zone[s] until all Normal memory is
4077  * exhausted, but results in overflowing to remote node while memory
4078  * may still exist in local DMA zone.
4079  */
4080 static int node_order[MAX_NUMNODES];
4081 
4082 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4083 {
4084         int pos, j, node;
4085         int zone_type;          /* needs to be signed */
4086         struct zone *z;
4087         struct zonelist *zonelist;
4088 
4089         zonelist = &pgdat->node_zonelists[0];
4090         pos = 0;
4091         for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4092                 for (j = 0; j < nr_nodes; j++) {
4093                         node = node_order[j];
4094                         z = &NODE_DATA(node)->node_zones[zone_type];
4095                         if (populated_zone(z)) {
4096                                 zoneref_set_zone(z,
4097                                         &zonelist->_zonerefs[pos++]);
4098                                 check_highest_zone(zone_type);
4099                         }
4100                 }
4101         }
4102         zonelist->_zonerefs[pos].zone = NULL;
4103         zonelist->_zonerefs[pos].zone_idx = 0;
4104 }
4105 
4106 #if defined(CONFIG_64BIT)
4107 /*
4108  * Devices that require DMA32/DMA are relatively rare and do not justify a
4109  * penalty to every machine in case the specialised case applies. Default
4110  * to Node-ordering on 64-bit NUMA machines
4111  */
4112 static int default_zonelist_order(void)
4113 {
4114         return ZONELIST_ORDER_NODE;
4115 }
4116 #else
4117 /*
4118  * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4119  * by the kernel. If processes running on node 0 deplete the low memory zone
4120  * then reclaim will occur more frequency increasing stalls and potentially
4121  * be easier to OOM if a large percentage of the zone is under writeback or
4122  * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4123  * Hence, default to zone ordering on 32-bit.
4124  */
4125 static int default_zonelist_order(void)
4126 {
4127         return ZONELIST_ORDER_ZONE;
4128 }
4129 #endif /* CONFIG_64BIT */
4130 
4131 static void set_zonelist_order(void)
4132 {
4133         if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4134                 current_zonelist_order = default_zonelist_order();
4135         else
4136                 current_zonelist_order = user_zonelist_order;
4137 }
4138 
4139 static void build_zonelists(pg_data_t *pgdat)
4140 {
4141         int j, node, load;
4142         enum zone_type i;
4143         nodemask_t used_mask;
4144         int local_node, prev_node;
4145         struct zonelist *zonelist;
4146         int order = current_zonelist_order;
4147 
4148         /* initialize zonelists */
4149         for (i = 0; i < MAX_ZONELISTS; i++) {
4150                 zonelist = pgdat->node_zonelists + i;
4151                 zonelist->_zonerefs[0].zone = NULL;
4152                 zonelist->_zonerefs[0].zone_idx = 0;
4153         }
4154 
4155         /* NUMA-aware ordering of nodes */
4156         local_node = pgdat->node_id;
4157         load = nr_online_nodes;
4158         prev_node = local_node;
4159         nodes_clear(used_mask);
4160 
4161         memset(node_order, 0, sizeof(node_order));
4162         j = 0;
4163 
4164         while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4165                 /*
4166                  * We don't want to pressure a particular node.
4167                  * So adding penalty to the first node in same
4168                  * distance group to make it round-robin.
4169                  */
4170                 if (node_distance(local_node, node) !=
4171                     node_distance(local_node, prev_node))
4172                         node_load[node] = load;
4173 
4174                 prev_node = node;
4175                 load--;
4176                 if (order == ZONELIST_ORDER_NODE)
4177                         build_zonelists_in_node_order(pgdat, node);
4178                 else
4179                         node_order[j++] = node; /* remember order */
4180         }
4181 
4182         if (order == ZONELIST_ORDER_ZONE) {
4183                 /* calculate node order -- i.e., DMA last! */
4184                 build_zonelists_in_zone_order(pgdat, j);
4185         }
4186 
4187         build_thisnode_zonelists(pgdat);
4188 }
4189 
4190 /* Construct the zonelist performance cache - see further mmzone.h */
4191 static void build_zonelist_cache(pg_data_t *pgdat)
4192 {
4193         struct zonelist *zonelist;
4194         struct zonelist_cache *zlc;
4195         struct zoneref *z;
4196 
4197         zonelist = &pgdat->node_zonelists[0];
4198         zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
4199         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
4200         for (z = zonelist->_zonerefs; z->zone; z++)
4201                 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
4202 }
4203 
4204 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4205 /*
4206  * Return node id of node used for "local" allocations.
4207  * I.e., first node id of first zone in arg node's generic zonelist.
4208  * Used for initializing percpu 'numa_mem', which is used primarily
4209  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4210  */
4211 int local_memory_node(int node)
4212 {
4213         struct zone *zone;
4214 
4215         (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4216                                    gfp_zone(GFP_KERNEL),
4217                                    NULL,
4218                                    &zone);
4219         return zone->node;
4220 }
4221 #endif
4222 
4223 #else   /* CONFIG_NUMA */
4224 
4225 static void set_zonelist_order(void)
4226 {
4227         current_zonelist_order = ZONELIST_ORDER_ZONE;
4228 }
4229 
4230 static void build_zonelists(pg_data_t *pgdat)
4231 {
4232         int node, local_node;
4233         enum zone_type j;
4234         struct zonelist *zonelist;
4235 
4236         local_node = pgdat->node_id;
4237 
4238         zonelist = &pgdat->node_zonelists[0];
4239         j = build_zonelists_node(pgdat, zonelist, 0);
4240 
4241         /*
4242          * Now we build the zonelist so that it contains the zones
4243          * of all the other nodes.
4244          * We don't want to pressure a particular node, so when
4245          * building the zones for node N, we make sure that the
4246          * zones coming right after the local ones are those from
4247          * node N+1 (modulo N)
4248          */
4249         for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4250                 if (!node_online(node))
4251                         continue;
4252                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4253         }
4254         for (node = 0; node < local_node; node++) {
4255                 if (!node_online(node))
4256                         continue;
4257                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4258         }
4259 
4260         zonelist->_zonerefs[j].zone = NULL;
4261         zonelist->_zonerefs[j].zone_idx = 0;
4262 }
4263 
4264 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
4265 static void build_zonelist_cache(pg_data_t *pgdat)
4266 {
4267         pgdat->node_zonelists[0].zlcache_ptr = NULL;
4268 }
4269 
4270 #endif  /* CONFIG_NUMA */
4271 
4272 /*
4273  * Boot pageset table. One per cpu which is going to be used for all
4274  * zones and all nodes. The parameters will be set in such a way
4275  * that an item put on a list will immediately be handed over to
4276  * the buddy list. This is safe since pageset manipulation is done
4277  * with interrupts disabled.
4278  *
4279  * The boot_pagesets must be kept even after bootup is complete for
4280  * unused processors and/or zones. They do play a role for bootstrapping
4281  * hotplugged processors.
4282  *
4283  * zoneinfo_show() and maybe other functions do
4284  * not check if the processor is online before following the pageset pointer.
4285  * Other parts of the kernel may not check if the zone is available.
4286  */
4287 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4288 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
4289 static void setup_zone_pageset(struct zone *zone);
4290 
4291 /*
4292  * Global mutex to protect against size modification of zonelists
4293  * as well as to serialize pageset setup for the new populated zone.
4294  */
4295 DEFINE_MUTEX(zonelists_mutex);
4296 
4297 /* return values int ....just for stop_machine() */
4298 static int __build_all_zonelists(void *data)
4299 {
4300         int nid;
4301         int cpu;
4302         pg_data_t *self = data;
4303 
4304 #ifdef CONFIG_NUMA
4305         memset(node_load, 0, sizeof(node_load));
4306 #endif
4307 
4308         if (self && !node_online(self->node_id)) {
4309                 build_zonelists(self);
4310                 build_zonelist_cache(self);
4311         }
4312 
4313         for_each_online_node(nid) {
4314                 pg_data_t *pgdat = NODE_DATA(nid);
4315 
4316                 build_zonelists(pgdat);
4317                 build_zonelist_cache(pgdat);
4318         }
4319 
4320         /*
4321          * Initialize the boot_pagesets that are going to be used
4322          * for bootstrapping processors. The real pagesets for
4323          * each zone will be allocated later when the per cpu
4324          * allocator is available.
4325          *
4326          * boot_pagesets are used also for bootstrapping offline
4327          * cpus if the system is already booted because the pagesets
4328          * are needed to initialize allocators on a specific cpu too.
4329          * F.e. the percpu allocator needs the page allocator which
4330          * needs the percpu allocator in order to allocate its pagesets
4331          * (a chicken-egg dilemma).
4332          */
4333         for_each_possible_cpu(cpu) {
4334                 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4335 
4336 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4337                 /*
4338                  * We now know the "local memory node" for each node--
4339                  * i.e., the node of the first zone in the generic zonelist.
4340                  * Set up numa_mem percpu variable for on-line cpus.  During
4341                  * boot, only the boot cpu should be on-line;  we'll init the
4342                  * secondary cpus' numa_mem as they come on-line.  During
4343                  * node/memory hotplug, we'll fixup all on-line cpus.
4344                  */
4345                 if (cpu_online(cpu))
4346                         set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4347 #endif
4348         }
4349 
4350         return 0;
4351 }
4352 
4353 static noinline void __init
4354 build_all_zonelists_init(void)
4355 {
4356         __build_all_zonelists(NULL);
4357         mminit_verify_zonelist();
4358         cpuset_init_current_mems_allowed();
4359 }
4360 
4361 /*
4362  * Called with zonelists_mutex held always
4363  * unless system_state == SYSTEM_BOOTING.
4364  *
4365  * __ref due to (1) call of __meminit annotated setup_zone_pageset
4366  * [we're only called with non-NULL zone through __meminit paths] and
4367  * (2) call of __init annotated helper build_all_zonelists_init
4368  * [protected by SYSTEM_BOOTING].
4369  */
4370 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
4371 {
4372         set_zonelist_order();
4373 
4374         if (system_state == SYSTEM_BOOTING) {
4375                 build_all_zonelists_init();
4376         } else {
4377 #ifdef CONFIG_MEMORY_HOTPLUG
4378                 if (zone)
4379                         setup_zone_pageset(zone);
4380 #endif
4381                 /* we have to stop all cpus to guarantee there is no user
4382                    of zonelist */
4383                 stop_machine(__build_all_zonelists, pgdat, NULL);
4384                 /* cpuset refresh routine should be here */
4385         }
4386         vm_total_pages = nr_free_pagecache_pages();
4387         /*
4388          * Disable grouping by mobility if the number of pages in the
4389          * system is too low to allow the mechanism to work. It would be
4390          * more accurate, but expensive to check per-zone. This check is
4391          * made on memory-hotadd so a system can start with mobility
4392          * disabled and enable it later
4393          */
4394         if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
4395                 page_group_by_mobility_disabled = 1;
4396         else
4397                 page_group_by_mobility_disabled = 0;
4398 
4399         pr_info("Built %i zonelists in %s order, mobility grouping %s.  "
4400                 "Total pages: %ld\n",
4401                         nr_online_nodes,
4402                         zonelist_order_name[current_zonelist_order],
4403                         page_group_by_mobility_disabled ? "off" : "on",
4404                         vm_total_pages);
4405 #ifdef CONFIG_NUMA
4406         pr_info("Policy zone: %s\n", zone_names[policy_zone]);
4407 #endif
4408 }
4409 
4410 /*
4411  * Helper functions to size the waitqueue hash table.
4412  * Essentially these want to choose hash table sizes sufficiently
4413  * large so that collisions trying to wait on pages are rare.
4414  * But in fact, the number of active page waitqueues on typical
4415  * systems is ridiculously low, less than 200. So this is even
4416  * conservative, even though it seems large.
4417  *
4418  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4419  * waitqueues, i.e. the size of the waitq table given the number of pages.
4420  */
4421 #define PAGES_PER_WAITQUEUE     256
4422 
4423 #ifndef CONFIG_MEMORY_HOTPLUG
4424 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4425 {
4426         unsigned long size = 1;
4427 
4428         pages /= PAGES_PER_WAITQUEUE;
4429 
4430         while (size < pages)
4431                 size <<= 1;
4432 
4433         /*
4434          * Once we have dozens or even hundreds of threads sleeping
4435          * on IO we've got bigger problems than wait queue collision.
4436          * Limit the size of the wait table to a reasonable size.
4437          */
4438         size = min(size, 4096UL);
4439 
4440         return max(size, 4UL);
4441 }
4442 #else
4443 /*
4444  * A zone's size might be changed by hot-add, so it is not possible to determine
4445  * a suitable size for its wait_table.  So we use the maximum size now.
4446  *
4447  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
4448  *
4449  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
4450  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4451  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
4452  *
4453  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4454  * or more by the traditional way. (See above).  It equals:
4455  *
4456  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
4457  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
4458  *    powerpc (64K page size)             : =  (32G +16M)byte.
4459  */
4460 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4461 {
4462         return 4096UL;
4463 }
4464 #endif
4465 
4466 /*
4467  * This is an integer logarithm so that shifts can be used later
4468  * to extract the more random high bits from the multiplicative
4469  * hash function before the remainder is taken.
4470  */
4471 static inline unsigned long wait_table_bits(unsigned long size)
4472 {
4473         return ffz(~size);
4474 }
4475 
4476 /*
4477  * Check if a pageblock contains reserved pages
4478  */
4479 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4480 {
4481         unsigned long pfn;
4482 
4483         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4484                 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4485                         return 1;
4486         }
4487         return 0;
4488 }
4489 
4490 /*
4491  * Mark a number of pageblocks as MIGRATE_RESERVE. The number
4492  * of blocks reserved is based on min_wmark_pages(zone). The memory within
4493  * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
4494  * higher will lead to a bigger reserve which will get freed as contiguous
4495  * blocks as reclaim kicks in
4496  */
4497 static void setup_zone_migrate_reserve(struct zone *zone)
4498 {
4499         unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
4500         struct page *page;
4501         unsigned long block_migratetype;
4502         int reserve;
4503         int old_reserve;
4504 
4505         /*
4506          * Get the start pfn, end pfn and the number of blocks to reserve
4507          * We have to be careful to be aligned to pageblock_nr_pages to
4508          * make sure that we always check pfn_valid for the first page in
4509          * the block.
4510          */
4511         start_pfn = zone->zone_start_pfn;
4512         end_pfn = zone_end_pfn(zone);
4513         start_pfn = roundup(start_pfn, pageblock_nr_pages);
4514         reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
4515                                                         pageblock_order;
4516 
4517         /*
4518          * Reserve blocks are generally in place to help high-order atomic
4519          * allocations that are short-lived. A min_free_kbytes value that
4520          * would result in more than 2 reserve blocks for atomic allocations
4521          * is assumed to be in place to help anti-fragmentation for the
4522          * future allocation of hugepages at runtime.
4523          */
4524         reserve = min(2, reserve);
4525         old_reserve = zone->nr_migrate_reserve_block;
4526 
4527         /* When memory hot-add, we almost always need to do nothing */
4528         if (reserve == old_reserve)
4529                 return;
4530         zone->nr_migrate_reserve_block = reserve;
4531 
4532         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
4533                 if (!early_page_nid_uninitialised(pfn, zone_to_nid(zone)))
4534                         return;
4535 
4536                 if (!pfn_valid(pfn))
4537                         continue;
4538                 page = pfn_to_page(pfn);
4539 
4540                 /* Watch out for overlapping nodes */
4541                 if (page_to_nid(page) != zone_to_nid(zone))
4542                         continue;
4543 
4544                 block_migratetype = get_pageblock_migratetype(page);
4545 
4546                 /* Only test what is necessary when the reserves are not met */
4547                 if (reserve > 0) {
4548                         /*
4549                          * Blocks with reserved pages will never free, skip
4550                          * them.
4551                          */
4552                         block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4553                         if (pageblock_is_reserved(pfn, block_end_pfn))
4554                                 continue;
4555 
4556                         /* If this block is reserved, account for it */
4557                         if (block_migratetype == MIGRATE_RESERVE) {
4558                                 reserve--;
4559                                 continue;
4560                         }
4561 
4562                         /* Suitable for reserving if this block is movable */
4563                         if (block_migratetype == MIGRATE_MOVABLE) {
4564                                 set_pageblock_migratetype(page,
4565                                                         MIGRATE_RESERVE);
4566                                 move_freepages_block(zone, page,
4567                                                         MIGRATE_RESERVE);
4568                                 reserve--;
4569                                 continue;
4570                         }
4571                 } else if (!old_reserve) {
4572                         /*
4573                          * At boot time we don't need to scan the whole zone
4574                          * for turning off MIGRATE_RESERVE.
4575                          */
4576                         break;
4577                 }
4578 
4579                 /*
4580                  * If the reserve is met and this is a previous reserved block,
4581                  * take it back
4582                  */
4583                 if (block_migratetype == MIGRATE_RESERVE) {
4584                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4585                         move_freepages_block(zone, page, MIGRATE_MOVABLE);
4586                 }
4587         }
4588 }
4589 
4590 /*
4591  * Initially all pages are reserved - free ones are freed
4592  * up by free_all_bootmem() once the early boot process is
4593  * done. Non-atomic initialization, single-pass.
4594  */
4595 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4596                 unsigned long start_pfn, enum memmap_context context)
4597 {
4598         pg_data_t *pgdat = NODE_DATA(nid);
4599         unsigned long end_pfn = start_pfn + size;
4600         unsigned long pfn;
4601         struct zone *z;
4602         unsigned long nr_initialised = 0;
4603 
4604         if (highest_memmap_pfn < end_pfn - 1)
4605                 highest_memmap_pfn = end_pfn - 1;
4606 
4607         z = &pgdat->node_zones[zone];
4608         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4609                 /*
4610                  * There can be holes in boot-time mem_map[]s
4611                  * handed to this function.  They do not
4612                  * exist on hotplugged memory.
4613                  */
4614                 if (context == MEMMAP_EARLY) {
4615                         if (!early_pfn_valid(pfn))
4616                                 continue;
4617                         if (!early_pfn_in_nid(pfn, nid))
4618                                 continue;
4619                         if (!update_defer_init(pgdat, pfn, end_pfn,
4620                                                 &nr_initialised))
4621                                 break;
4622                 }
4623 
4624                 /*
4625                  * Mark the block movable so that blocks are reserved for
4626                  * movable at startup. This will force kernel allocations
4627                  * to reserve their blocks rather than leaking throughout
4628                  * the address space during boot when many long-lived
4629                  * kernel allocations are made. Later some blocks near
4630                  * the start are marked MIGRATE_RESERVE by
4631                  * setup_zone_migrate_reserve()
4632                  *
4633                  * bitmap is created for zone's valid pfn range. but memmap
4634                  * can be created for invalid pages (for alignment)
4635                  * check here not to call set_pageblock_migratetype() against
4636                  * pfn out of zone.
4637                  */
4638                 if (!(pfn & (pageblock_nr_pages - 1))) {
4639                         struct page *page = pfn_to_page(pfn);
4640 
4641                         __init_single_page(page, pfn, zone, nid);
4642                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4643                 } else {
4644                         __init_single_pfn(pfn, zone, nid);
4645                 }
4646         }
4647 }
4648 
4649 static void __meminit zone_init_free_lists(struct zone *zone)
4650 {
4651         unsigned int order, t;
4652         for_each_migratetype_order(order, t) {
4653                 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4654                 zone->free_area[order].nr_free = 0;
4655         }
4656 }
4657 
4658 #ifndef __HAVE_ARCH_MEMMAP_INIT
4659 #define memmap_init(size, nid, zone, start_pfn) \
4660         memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4661 #endif
4662 
4663 static int zone_batchsize(struct zone *zone)
4664 {
4665 #ifdef CONFIG_MMU
4666         int batch;
4667 
4668         /*
4669          * The per-cpu-pages pools are set to around 1000th of the
4670          * size of the zone.  But no more than 1/2 of a meg.
4671          *
4672          * OK, so we don't know how big the cache is.  So guess.
4673          */
4674         batch = zone->managed_pages / 1024;
4675         if (batch * PAGE_SIZE > 512 * 1024)
4676                 batch = (512 * 1024) / PAGE_SIZE;
4677         batch /= 4;             /* We effectively *= 4 below */
4678         if (batch < 1)
4679                 batch = 1;
4680 
4681         /*
4682          * Clamp the batch to a 2^n - 1 value. Having a power
4683          * of 2 value was found to be more likely to have
4684          * suboptimal cache aliasing properties in some cases.
4685          *
4686          * For example if 2 tasks are alternately allocating
4687          * batches of pages, one task can end up with a lot
4688          * of pages of one half of the possible page colors
4689          * and the other with pages of the other colors.
4690          */
4691         batch = rounddown_pow_of_two(batch + batch/2) - 1;
4692 
4693         return batch;
4694 
4695 #else
4696         /* The deferral and batching of frees should be suppressed under NOMMU
4697          * conditions.
4698          *
4699          * The problem is that NOMMU needs to be able to allocate large chunks
4700          * of contiguous memory as there's no hardware page translation to
4701          * assemble apparent contiguous memory from discontiguous pages.
4702          *
4703          * Queueing large contiguous runs of pages for batching, however,
4704          * causes the pages to actually be freed in smaller chunks.  As there
4705          * can be a significant delay between the individual batches being
4706          * recycled, this leads to the once large chunks of space being
4707          * fragmented and becoming unavailable for high-order allocations.
4708          */
4709         return 0;
4710 #endif
4711 }
4712 
4713 /*
4714  * pcp->high and pcp->batch values are related and dependent on one another:
4715  * ->batch must never be higher then ->high.
4716  * The following function updates them in a safe manner without read side
4717  * locking.
4718  *
4719  * Any new users of pcp->batch and pcp->high should ensure they can cope with
4720  * those fields changing asynchronously (acording the the above rule).
4721  *
4722  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4723  * outside of boot time (or some other assurance that no concurrent updaters
4724  * exist).
4725  */
4726 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4727                 unsigned long batch)
4728 {
4729        /* start with a fail safe value for batch */
4730         pcp->batch = 1;
4731         smp_wmb();
4732 
4733        /* Update high, then batch, in order */
4734         pcp->high = high;
4735         smp_wmb();
4736 
4737         pcp->batch = batch;
4738 }
4739 
4740 /* a companion to pageset_set_high() */
4741 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4742 {
4743         pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4744 }
4745 
4746 static void pageset_init(struct per_cpu_pageset *p)
4747 {
4748         struct per_cpu_pages *pcp;
4749         int migratetype;
4750 
4751         memset(p, 0, sizeof(*p));
4752 
4753         pcp = &p->pcp;
4754         pcp->count = 0;
4755         for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4756                 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4757 }
4758 
4759 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4760 {
4761         pageset_init(p);
4762         pageset_set_batch(p, batch);
4763 }
4764 
4765 /*
4766  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4767  * to the value high for the pageset p.
4768  */
4769 static void pageset_set_high(struct per_cpu_pageset *p,
4770                                 unsigned long high)
4771 {
4772         unsigned long batch = max(1UL, high / 4);
4773         if ((high / 4) > (PAGE_SHIFT * 8))
4774                 batch = PAGE_SHIFT * 8;
4775 
4776         pageset_update(&p->pcp, high, batch);
4777 }
4778 
4779 static void pageset_set_high_and_batch(struct zone *zone,
4780                                        struct per_cpu_pageset *pcp)
4781 {
4782         if (percpu_pagelist_fraction)
4783                 pageset_set_high(pcp,
4784                         (zone->managed_pages /
4785                                 percpu_pagelist_fraction));
4786         else
4787                 pageset_set_batch(pcp, zone_batchsize(zone));
4788 }
4789 
4790 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4791 {
4792         struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4793 
4794         pageset_init(pcp);
4795         pageset_set_high_and_batch(zone, pcp);
4796 }
4797 
4798 static void __meminit setup_zone_pageset(struct zone *zone)
4799 {
4800         int cpu;
4801         zone->pageset = alloc_percpu(struct per_cpu_pageset);
4802         for_each_possible_cpu(cpu)
4803                 zone_pageset_init(zone, cpu);
4804 }
4805 
4806 /*
4807  * Allocate per cpu pagesets and initialize them.
4808  * Before this call only boot pagesets were available.
4809  */
4810 void __init setup_per_cpu_pageset(void)
4811 {
4812         struct zone *zone;
4813 
4814         for_each_populated_zone(zone)
4815                 setup_zone_pageset(zone);
4816 }
4817 
4818 static noinline __init_refok
4819 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4820 {
4821         int i;
4822         size_t alloc_size;
4823 
4824         /*
4825          * The per-page waitqueue mechanism uses hashed waitqueues
4826          * per zone.
4827          */
4828         zone->wait_table_hash_nr_entries =
4829                  wait_table_hash_nr_entries(zone_size_pages);
4830         zone->wait_table_bits =
4831                 wait_table_bits(zone->wait_table_hash_nr_entries);
4832         alloc_size = zone->wait_table_hash_nr_entries
4833                                         * sizeof(wait_queue_head_t);
4834 
4835         if (!slab_is_available()) {
4836                 zone->wait_table = (wait_queue_head_t *)
4837                         memblock_virt_alloc_node_nopanic(
4838                                 alloc_size, zone->zone_pgdat->node_id);
4839         } else {
4840                 /*
4841                  * This case means that a zone whose size was 0 gets new memory
4842                  * via memory hot-add.
4843                  * But it may be the case that a new node was hot-added.  In
4844                  * this case vmalloc() will not be able to use this new node's
4845                  * memory - this wait_table must be initialized to use this new
4846                  * node itself as well.
4847                  * To use this new node's memory, further consideration will be
4848                  * necessary.
4849                  */
4850                 zone->wait_table = vmalloc(alloc_size);
4851         }
4852         if (!zone->wait_table)
4853                 return -ENOMEM;
4854 
4855         for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4856                 init_waitqueue_head(zone->wait_table + i);
4857 
4858         return 0;
4859 }
4860 
4861 static __meminit void zone_pcp_init(struct zone *zone)
4862 {
4863         /*
4864          * per cpu subsystem is not up at this point. The following code
4865          * relies on the ability of the linker to provide the
4866          * offset of a (static) per cpu variable into the per cpu area.
4867          */
4868         zone->pageset = &boot_pageset;
4869 
4870         if (populated_zone(zone))
4871                 printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
4872                         zone->name, zone->present_pages,
4873                                          zone_batchsize(zone));
4874 }
4875 
4876 int __meminit init_currently_empty_zone(struct zone *zone,
4877                                         unsigned long zone_start_pfn,
4878                                         unsigned long size,
4879                                         enum memmap_context context)
4880 {
4881         struct pglist_data *pgdat = zone->zone_pgdat;
4882         int ret;
4883         ret = zone_wait_table_init(zone, size);
4884         if (ret)
4885                 return ret;
4886         pgdat->nr_zones = zone_idx(zone) + 1;
4887 
4888         zone->zone_start_pfn = zone_start_pfn;
4889 
4890         mminit_dprintk(MMINIT_TRACE, "memmap_init",
4891                         "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4892                         pgdat->node_id,
4893                         (unsigned long)zone_idx(zone),
4894                         zone_start_pfn, (zone_start_pfn + size));
4895 
4896         zone_init_free_lists(zone);
4897 
4898         return 0;
4899 }
4900 
4901 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4902 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4903 
4904 /*
4905  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4906  */
4907 int __meminit __early_pfn_to_nid(unsigned long pfn,
4908                                         struct mminit_pfnnid_cache *state)
4909 {
4910         unsigned long start_pfn, end_pfn;
4911         int nid;
4912 
4913         if (state->last_start <= pfn && pfn < state->last_end)
4914                 return state->last_nid;
4915 
4916         nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4917         if (nid != -1) {
4918                 state->last_start = start_pfn;
4919                 state->last_end = end_pfn;
4920                 state->last_nid = nid;
4921         }
4922 
4923         return nid;
4924 }
4925 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4926 
4927 /**
4928  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4929  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4930  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4931  *
4932  * If an architecture guarantees that all ranges registered contain no holes
4933  * and may be freed, this this function may be used instead of calling
4934  * memblock_free_early_nid() manually.
4935  */
4936 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
4937 {
4938         unsigned long start_pfn, end_pfn;
4939         int i, this_nid;
4940 
4941         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4942                 start_pfn = min(start_pfn, max_low_pfn);
4943                 end_pfn = min(end_pfn, max_low_pfn);
4944 
4945                 if (start_pfn < end_pfn)
4946                         memblock_free_early_nid(PFN_PHYS(start_pfn),
4947                                         (end_pfn - start_pfn) << PAGE_SHIFT,
4948                                         this_nid);
4949         }
4950 }
4951 
4952 /**
4953  * sparse_memory_present_with_active_regions - Call memory_present for each active range
4954  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4955  *
4956  * If an architecture guarantees that all ranges registered contain no holes and may
4957  * be freed, this function may be used instead of calling memory_present() manually.
4958  */
4959 void __init sparse_memory_present_with_active_regions(int nid)
4960 {
4961         unsigned long start_pfn, end_pfn;
4962         int i, this_nid;
4963 
4964         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4965                 memory_present(this_nid, start_pfn, end_pfn);
4966 }
4967 
4968 /**
4969  * get_pfn_range_for_nid - Return the start and end page frames for a node
4970  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4971  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4972  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4973  *
4974  * It returns the start and end page frame of a node based on information
4975  * provided by memblock_set_node(). If called for a node
4976  * with no available memory, a warning is printed and the start and end
4977  * PFNs will be 0.
4978  */
4979 void __meminit get_pfn_range_for_nid(unsigned int nid,
4980                         unsigned long *start_pfn, unsigned long *end_pfn)
4981 {
4982         unsigned long this_start_pfn, this_end_pfn;
4983         int i;
4984 
4985         *start_pfn = -1UL;
4986         *end_pfn = 0;
4987 
4988         for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4989                 *start_pfn = min(*start_pfn, this_start_pfn);
4990                 *end_pfn = max(*end_pfn, this_end_pfn);
4991         }
4992 
4993         if (*start_pfn == -1UL)
4994                 *start_pfn = 0;
4995 }
4996 
4997 /*
4998  * This finds a zone that can be used for ZONE_MOVABLE pages. The
4999  * assumption is made that zones within a node are ordered in monotonic
5000  * increasing memory addresses so that the "highest" populated zone is used
5001  */
5002 static void __init find_usable_zone_for_movable(void)
5003 {
5004         int zone_index;
5005         for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5006                 if (zone_index == ZONE_MOVABLE)
5007                         continue;
5008 
5009                 if (arch_zone_highest_possible_pfn[zone_index] >
5010                                 arch_zone_lowest_possible_pfn[zone_index])
5011                         break;
5012         }
5013 
5014         VM_BUG_ON(zone_index == -1);
5015         movable_zone = zone_index;
5016 }
5017 
5018 /*
5019  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5020  * because it is sized independent of architecture. Unlike the other zones,
5021  * the starting point for ZONE_MOVABLE is not fixed. It may be different
5022  * in each node depending on the size of each node and how evenly kernelcore
5023  * is distributed. This helper function adjusts the zone ranges
5024  * provided by the architecture for a given node by using the end of the
5025  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5026  * zones within a node are in order of monotonic increases memory addresses
5027  */
5028 static void __meminit adjust_zone_range_for_zone_movable(int nid,
5029                                         unsigned long zone_type,
5030                                         unsigned long node_start_pfn,
5031                                         unsigned long node_end_pfn,
5032                                         unsigned long *zone_start_pfn,
5033                                         unsigned long *zone_end_pfn)
5034 {
5035         /* Only adjust if ZONE_MOVABLE is on this node */
5036         if (zone_movable_pfn[nid]) {
5037                 /* Size ZONE_MOVABLE */
5038                 if (zone_type == ZONE_MOVABLE) {
5039                         *zone_start_pfn = zone_movable_pfn[nid];
5040                         *zone_end_pfn = min(node_end_pfn,
5041                                 arch_zone_highest_possible_pfn[movable_zone]);
5042 
5043                 /* Adjust for ZONE_MOVABLE starting within this range */
5044                 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
5045                                 *zone_end_pfn > zone_movable_pfn[nid]) {
5046                         *zone_end_pfn = zone_movable_pfn[nid];
5047 
5048                 /* Check if this whole range is within ZONE_MOVABLE */
5049                 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5050                         *zone_start_pfn = *zone_end_pfn;
5051         }
5052 }
5053 
5054 /*
5055  * Return the number of pages a zone spans in a node, including holes
5056  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5057  */
5058 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
5059                                         unsigned long zone_type,
5060                                         unsigned long node_start_pfn,
5061                                         unsigned long node_end_pfn,
5062                                         unsigned long *ignored)
5063 {
5064         unsigned long zone_start_pfn, zone_end_pfn;
5065 
5066         /* When hotadd a new node, the node should be empty */
5067         if (!node_start_pfn && !node_end_pfn)
5068                 return 0;
5069 
5070         /* Get the start and end of the zone */
5071         zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5072         zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
5073         adjust_zone_range_for_zone_movable(nid, zone_type,
5074                                 node_start_pfn, node_end_pfn,
5075                                 &zone_start_pfn, &zone_end_pfn);
5076 
5077         /* Check that this node has pages within the zone's required range */
5078         if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
5079                 return 0;
5080 
5081         /* Move the zone boundaries inside the node if necessary */
5082         zone_end_pfn = min(zone_end_pfn, node_end_pfn);
5083         zone_start_pfn = max(zone_start_pfn, node_start_pfn);
5084 
5085         /* Return the spanned pages */
5086         return zone_end_pfn - zone_start_pfn;
5087 }
5088 
5089 /*
5090  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5091  * then all holes in the requested range will be accounted for.
5092  */
5093 unsigned long __meminit __absent_pages_in_range(int nid,
5094                                 unsigned long range_start_pfn,
5095                                 unsigned long range_end_pfn)
5096 {
5097         unsigned long nr_absent = range_end_pfn - range_start_pfn;
5098         unsigned long start_pfn, end_pfn;
5099         int i;
5100 
5101         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5102                 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5103                 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5104                 nr_absent -= end_pfn - start_pfn;
5105         }
5106         return nr_absent;
5107 }
5108 
5109 /**
5110  * absent_pages_in_range - Return number of page frames in holes within a range
5111  * @start_pfn: The start PFN to start searching for holes
5112  * @end_pfn: The end PFN to stop searching for holes
5113  *
5114  * It returns the number of pages frames in memory holes within a range.
5115  */
5116 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5117                                                         unsigned long end_pfn)
5118 {
5119         return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5120 }
5121 
5122 /* Return the number of page frames in holes in a zone on a node */
5123 static unsigned long __meminit zone_absent_pages_in_node(int nid,
5124                                         unsigned long zone_type,
5125                                         unsigned long node_start_pfn,
5126                                         unsigned long node_end_pfn,
5127                                         unsigned long *ignored)
5128 {
5129         unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5130         unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
5131         unsigned long zone_start_pfn, zone_end_pfn;
5132 
5133         /* When hotadd a new node, the node should be empty */
5134         if (!node_start_pfn && !node_end_pfn)
5135                 return 0;
5136 
5137         zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5138         zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
5139 
5140         adjust_zone_range_for_zone_movable(nid, zone_type,
5141                         node_start_pfn, node_end_pfn,
5142                         &zone_start_pfn, &zone_end_pfn);
5143         return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5144 }
5145 
5146 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5147 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
5148                                         unsigned long zone_type,
5149                                         unsigned long node_start_pfn,
5150                                         unsigned long node_end_pfn,
5151                                         unsigned long *zones_size)
5152 {
5153         return zones_size[zone_type];
5154 }
5155 
5156 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
5157                                                 unsigned long zone_type,
5158                                                 unsigned long node_start_pfn,
5159                                                 unsigned long node_end_pfn,
5160                                                 unsigned long *zholes_size)
5161 {
5162         if (!zholes_size)
5163                 return 0;
5164 
5165         return zholes_size[zone_type];
5166 }
5167 
5168 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5169 
5170 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
5171                                                 unsigned long node_start_pfn,
5172                                                 unsigned long node_end_pfn,
5173                                                 unsigned long *zones_size,
5174                                                 unsigned long *zholes_size)
5175 {
5176         unsigned long realtotalpages = 0, totalpages = 0;
5177         enum zone_type i;
5178 
5179         for (i = 0; i < MAX_NR_ZONES; i++) {
5180                 struct zone *zone = pgdat->node_zones + i;
5181                 unsigned long size, real_size;
5182 
5183                 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5184                                                   node_start_pfn,
5185                                                   node_end_pfn,
5186                                                   zones_size);
5187                 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
5188                                                   node_start_pfn, node_end_pfn,
5189                                                   zholes_size);
5190                 zone->spanned_pages = size;
5191                 zone->present_pages = real_size;
5192 
5193                 totalpages += size;
5194                 realtotalpages += real_size;
5195         }
5196 
5197         pgdat->node_spanned_pages = totalpages;
5198         pgdat->node_present_pages = realtotalpages;
5199         printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5200                                                         realtotalpages);
5201 }
5202 
5203 #ifndef CONFIG_SPARSEMEM
5204 /*
5205  * Calculate the size of the zone->blockflags rounded to an unsigned long
5206  * Start by making sure zonesize is a multiple of pageblock_order by rounding
5207  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5208  * round what is now in bits to nearest long in bits, then return it in
5209  * bytes.
5210  */
5211 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
5212 {
5213         unsigned long usemapsize;
5214 
5215         zonesize += zone_start_pfn & (pageblock_nr_pages-1);
5216         usemapsize = roundup(zonesize, pageblock_nr_pages);
5217         usemapsize = usemapsize >> pageblock_order;
5218         usemapsize *= NR_PAGEBLOCK_BITS;
5219         usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5220 
5221         return usemapsize / 8;
5222 }
5223 
5224 static void __init setup_usemap(struct pglist_data *pgdat,
5225                                 struct zone *zone,
5226                                 unsigned long zone_start_pfn,
5227                                 unsigned long zonesize)
5228 {
5229         unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
5230         zone->pageblock_flags = NULL;
5231         if (usemapsize)
5232                 zone->pageblock_flags =
5233                         memblock_virt_alloc_node_nopanic(usemapsize,
5234                                                          pgdat->node_id);
5235 }
5236 #else
5237 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5238                                 unsigned long zone_start_pfn, unsigned long zonesize) {}
5239 #endif /* CONFIG_SPARSEMEM */
5240 
5241 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5242 
5243 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5244 void __paginginit set_pageblock_order(void)
5245 {
5246         unsigned int order;
5247 
5248         /* Check that pageblock_nr_pages has not already been setup */
5249         if (pageblock_order)
5250                 return;
5251 
5252         if (HPAGE_SHIFT > PAGE_SHIFT)
5253                 order = HUGETLB_PAGE_ORDER;
5254         else
5255                 order = MAX_ORDER - 1;
5256 
5257         /*
5258          * Assume the largest contiguous order of interest is a huge page.
5259          * This value may be variable depending on boot parameters on IA64 and
5260          * powerpc.
5261          */
5262         pageblock_order = order;
5263 }
5264 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5265 
5266 /*
5267  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5268  * is unused as pageblock_order is set at compile-time. See
5269  * include/linux/pageblock-flags.h for the values of pageblock_order based on
5270  * the kernel config
5271  */
5272 void __paginginit set_pageblock_order(void)
5273 {
5274 }
5275 
5276 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5277 
5278 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5279                                                    unsigned long present_pages)
5280 {
5281         unsigned long pages = spanned_pages;
5282 
5283         /*
5284          * Provide a more accurate estimation if there are holes within
5285          * the zone and SPARSEMEM is in use. If there are holes within the
5286          * zone, each populated memory region may cost us one or two extra
5287          * memmap pages due to alignment because memmap pages for each
5288          * populated regions may not naturally algined on page boundary.
5289          * So the (present_pages >> 4) heuristic is a tradeoff for that.
5290          */
5291         if (spanned_pages > present_pages + (present_pages >> 4) &&
5292             IS_ENABLED(CONFIG_SPARSEMEM))
5293                 pages = present_pages;
5294 
5295         return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5296 }
5297 
5298 /*
5299  * Set up the zone data structures:
5300  *   - mark all pages reserved
5301  *   - mark all memory queues empty
5302  *   - clear the memory bitmaps
5303  *
5304  * NOTE: pgdat should get zeroed by caller.
5305  */
5306 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
5307                 unsigned long node_start_pfn, unsigned long node_end_pfn)
5308 {
5309         enum zone_type j;
5310         int nid = pgdat->node_id;
5311         unsigned long zone_start_pfn = pgdat->node_start_pfn;
5312         int ret;
5313 
5314         pgdat_resize_init(pgdat);
5315 #ifdef CONFIG_NUMA_BALANCING
5316         spin_lock_init(&pgdat->numabalancing_migrate_lock);
5317         pgdat->numabalancing_migrate_nr_pages = 0;
5318         pgdat->numabalancing_migrate_next_window = jiffies;
5319 #endif
5320         init_waitqueue_head(&pgdat->kswapd_wait);
5321         init_waitqueue_head(&pgdat->pfmemalloc_wait);
5322         pgdat_page_ext_init(pgdat);
5323 
5324         for (j = 0; j < MAX_NR_ZONES; j++) {
5325                 struct zone *zone = pgdat->node_zones + j;
5326                 unsigned long size, realsize, freesize, memmap_pages;
5327 
5328                 size = zone->spanned_pages;
5329                 realsize = freesize = zone->present_pages;
5330 
5331                 /*
5332                  * Adjust freesize so that it accounts for how much memory
5333                  * is used by this zone for memmap. This affects the watermark
5334                  * and per-cpu initialisations
5335                  */
5336                 memmap_pages = calc_memmap_size(size, realsize);
5337                 if (!is_highmem_idx(j)) {
5338                         if (freesize >= memmap_pages) {
5339                                 freesize -= memmap_pages;
5340                                 if (memmap_pages)
5341                                         printk(KERN_DEBUG
5342                                                "  %s zone: %lu pages used for memmap\n",
5343                                                zone_names[j], memmap_pages);
5344                         } else
5345                                 printk(KERN_WARNING
5346                                         "  %s zone: %lu pages exceeds freesize %lu\n",
5347                                         zone_names[j], memmap_pages, freesize);
5348                 }
5349 
5350                 /* Account for reserved pages */
5351                 if (j == 0 && freesize > dma_reserve) {
5352                         freesize -= dma_reserve;
5353                         printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
5354                                         zone_names[0], dma_reserve);
5355                 }
5356 
5357                 if (!is_highmem_idx(j))
5358                         nr_kernel_pages += freesize;
5359                 /* Charge for highmem memmap if there are enough kernel pages */
5360                 else if (nr_kernel_pages > memmap_pages * 2)
5361                         nr_kernel_pages -= memmap_pages;
5362                 nr_all_pages += freesize;
5363 
5364                 /*
5365                  * Set an approximate value for lowmem here, it will be adjusted
5366                  * when the bootmem allocator frees pages into the buddy system.
5367                  * And all highmem pages will be managed by the buddy system.
5368                  */
5369                 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
5370 #ifdef CONFIG_NUMA
5371                 zone->node = nid;
5372                 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
5373                                                 / 100;
5374                 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
5375 #endif
5376                 zone->name = zone_names[j];
5377                 spin_lock_init(&zone->lock);
5378                 spin_lock_init(&zone->lru_lock);
5379                 zone_seqlock_init(zone);
5380                 zone->zone_pgdat = pgdat;
5381                 zone_pcp_init(zone);
5382 
5383                 /* For bootup, initialized properly in watermark setup */
5384                 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5385 
5386                 lruvec_init(&zone->lruvec);
5387                 if (!size)
5388                         continue;
5389 
5390                 set_pageblock_order();
5391                 setup_usemap(pgdat, zone, zone_start_pfn, size);
5392                 ret = init_currently_empty_zone(zone, zone_start_pfn,
5393                                                 size, MEMMAP_EARLY);
5394                 BUG_ON(ret);
5395                 memmap_init(size, nid, j, zone_start_pfn);
5396                 zone_start_pfn += size;
5397         }
5398 }
5399 
5400 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
5401 {
5402         /* Skip empty nodes */
5403         if (!pgdat->node_spanned_pages)
5404                 return;
5405 
5406 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5407         /* ia64 gets its own node_mem_map, before this, without bootmem */
5408         if (!pgdat->node_mem_map) {
5409                 unsigned long size, start, end;
5410                 struct page *map;
5411 
5412                 /*
5413                  * The zone's endpoints aren't required to be MAX_ORDER
5414                  * aligned but the node_mem_map endpoints must be in order
5415                  * for the buddy allocator to function correctly.
5416                  */
5417                 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5418                 end = pgdat_end_pfn(pgdat);
5419                 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5420                 size =  (end - start) * sizeof(struct page);
5421                 map = alloc_remap(pgdat->node_id, size);
5422                 if (!map)
5423                         map = memblock_virt_alloc_node_nopanic(size,
5424                                                                pgdat->node_id);
5425                 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
5426         }
5427 #ifndef CONFIG_NEED_MULTIPLE_NODES
5428         /*
5429          * With no DISCONTIG, the global mem_map is just set as node 0's
5430          */
5431         if (pgdat == NODE_DATA(0)) {
5432                 mem_map = NODE_DATA(0)->node_mem_map;
5433 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5434                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
5435                         mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
5436 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5437         }
5438 #endif
5439 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5440 }
5441 
5442 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5443                 unsigned long node_start_pfn, unsigned long *zholes_size)
5444 {
5445         pg_data_t *pgdat = NODE_DATA(nid);
5446         unsigned long start_pfn = 0;
5447         unsigned long end_pfn = 0;
5448 
5449         /* pg_data_t should be reset to zero when it's allocated */
5450         WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
5451 
5452         reset_deferred_meminit(pgdat);
5453         pgdat->node_id = nid;
5454         pgdat->node_start_pfn = node_start_pfn;
5455 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5456         get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
5457         pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5458                 (u64)start_pfn << PAGE_SHIFT, ((u64)end_pfn << PAGE_SHIFT) - 1);
5459 #endif
5460         calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5461                                   zones_size, zholes_size);
5462 
5463         alloc_node_mem_map(pgdat);
5464 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5465         printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5466                 nid, (unsigned long)pgdat,
5467                 (unsigned long)pgdat->node_mem_map);
5468 #endif
5469 
5470         free_area_init_core(pgdat, start_pfn, end_pfn);
5471 }
5472 
5473 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5474 
5475 #if MAX_NUMNODES > 1
5476 /*
5477  * Figure out the number of possible node ids.
5478  */
5479 void __init setup_nr_node_ids(void)
5480 {
5481         unsigned int node;
5482         unsigned int highest = 0;
5483 
5484         for_each_node_mask(node, node_possible_map)
5485                 highest = node;
5486         nr_node_ids = highest + 1;
5487 }
5488 #endif
5489 
5490 /**
5491  * node_map_pfn_alignment - determine the maximum internode alignment
5492  *
5493  * This function should be called after node map is populated and sorted.
5494  * It calculates the maximum power of two alignment which can distinguish
5495  * all the nodes.
5496  *
5497  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5498  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
5499  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
5500  * shifted, 1GiB is enough and this function will indicate so.
5501  *
5502  * This is used to test whether pfn -> nid mapping of the chosen memory
5503  * model has fine enough granularity to avoid incorrect mapping for the
5504  * populated node map.
5505  *
5506  * Returns the determined alignment in pfn's.  0 if there is no alignment
5507  * requirement (single node).
5508  */
5509 unsigned long __init node_map_pfn_alignment(void)
5510 {
5511         unsigned long accl_mask = 0, last_end = 0;
5512         unsigned long start, end, mask;
5513         int last_nid = -1;
5514         int i, nid;
5515 
5516         for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5517                 if (!start || last_nid < 0 || last_nid == nid) {
5518                         last_nid = nid;
5519                         last_end = end;
5520                         continue;
5521                 }
5522 
5523                 /*
5524                  * Start with a mask granular enough to pin-point to the
5525                  * start pfn and tick off bits one-by-one until it becomes
5526                  * too coarse to separate the current node from the last.
5527                  */
5528                 mask = ~((1 << __ffs(start)) - 1);
5529                 while (mask && last_end <= (start & (mask << 1)))
5530                         mask <<= 1;
5531 
5532                 /* accumulate all internode masks */
5533                 accl_mask |= mask;
5534         }
5535 
5536         /* convert mask to number of pages */
5537         return ~accl_mask + 1;
5538 }
5539 
5540 /* Find the lowest pfn for a node */
5541 static unsigned long __init find_min_pfn_for_node(int nid)
5542 {
5543         unsigned long min_pfn = ULONG_MAX;
5544         unsigned long start_pfn;
5545         int i;
5546 
5547         for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5548                 min_pfn = min(min_pfn, start_pfn);
5549 
5550         if (min_pfn == ULONG_MAX) {
5551                 printk(KERN_WARNING
5552                         "Could not find start_pfn for node %d\n", nid);
5553                 return 0;
5554         }
5555 
5556         return min_pfn;
5557 }
5558 
5559 /**
5560  * find_min_pfn_with_active_regions - Find the minimum PFN registered
5561  *
5562  * It returns the minimum PFN based on information provided via
5563  * memblock_set_node().
5564  */
5565 unsigned long __init find_min_pfn_with_active_regions(void)
5566 {
5567         return find_min_pfn_for_node(MAX_NUMNODES);
5568 }
5569 
5570 /*
5571  * early_calculate_totalpages()
5572  * Sum pages in active regions for movable zone.
5573  * Populate N_MEMORY for calculating usable_nodes.
5574  */
5575 static unsigned long __init early_calculate_totalpages(void)
5576 {
5577         unsigned long totalpages = 0;
5578         unsigned long start_pfn, end_pfn;
5579         int i, nid;
5580 
5581         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5582                 unsigned long pages = end_pfn - start_pfn;
5583 
5584                 totalpages += pages;
5585                 if (pages)
5586                         node_set_state(nid, N_MEMORY);
5587         }
5588         return totalpages;
5589 }
5590 
5591 /*
5592  * Find the PFN the Movable zone begins in each node. Kernel memory
5593  * is spread evenly between nodes as long as the nodes have enough
5594  * memory. When they don't, some nodes will have more kernelcore than
5595  * others
5596  */
5597 static void __init find_zone_movable_pfns_for_nodes(void)
5598 {
5599         int i, nid;
5600         unsigned long usable_startpfn;
5601         unsigned long kernelcore_node, kernelcore_remaining;
5602         /* save the state before borrow the nodemask */
5603         nodemask_t saved_node_state = node_states[N_MEMORY];
5604         unsigned long totalpages = early_calculate_totalpages();
5605         int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5606         struct memblock_region *r;
5607 
5608         /* Need to find movable_zone earlier when movable_node is specified. */
5609         find_usable_zone_for_movable();
5610 
5611         /*
5612          * If movable_node is specified, ignore kernelcore and movablecore
5613          * options.
5614          */
5615         if (movable_node_is_enabled()) {
5616                 for_each_memblock(memory, r) {
5617                         if (!memblock_is_hotpluggable(r))
5618                                 continue;
5619 
5620                         nid = r->nid;
5621 
5622                         usable_startpfn = PFN_DOWN(r->base);
5623                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5624                                 min(usable_startpfn, zone_movable_pfn[nid]) :
5625                                 usable_startpfn;
5626                 }
5627 
5628                 goto out2;
5629         }
5630 
5631         /*
5632          * If movablecore=nn[KMG] was specified, calculate what size of
5633          * kernelcore that corresponds so that memory usable for
5634          * any allocation type is evenly spread. If both kernelcore
5635          * and movablecore are specified, then the value of kernelcore
5636          * will be used for required_kernelcore if it's greater than
5637          * what movablecore would have allowed.
5638          */
5639         if (required_movablecore) {
5640                 unsigned long corepages;
5641 
5642                 /*
5643                  * Round-up so that ZONE_MOVABLE is at least as large as what
5644                  * was requested by the user
5645                  */
5646                 required_movablecore =
5647                         roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5648                 corepages = totalpages - required_movablecore;
5649 
5650                 required_kernelcore = max(required_kernelcore, corepages);
5651         }
5652 
5653         /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5654         if (!required_kernelcore)
5655                 goto out;
5656 
5657         /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5658         usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5659 
5660 restart:
5661         /* Spread kernelcore memory as evenly as possible throughout nodes */
5662         kernelcore_node = required_kernelcore / usable_nodes;
5663         for_each_node_state(nid, N_MEMORY) {
5664                 unsigned long start_pfn, end_pfn;
5665 
5666                 /*
5667                  * Recalculate kernelcore_node if the division per node
5668                  * now exceeds what is necessary to satisfy the requested
5669                  * amount of memory for the kernel
5670                  */
5671                 if (required_kernelcore < kernelcore_node)
5672                         kernelcore_node = required_kernelcore / usable_nodes;
5673 
5674                 /*
5675                  * As the map is walked, we track how much memory is usable
5676                  * by the kernel using kernelcore_remaining. When it is
5677                  * 0, the rest of the node is usable by ZONE_MOVABLE
5678                  */
5679                 kernelcore_remaining = kernelcore_node;
5680 
5681                 /* Go through each range of PFNs within this node */
5682                 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5683                         unsigned long size_pages;
5684 
5685                         start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5686                         if (start_pfn >= end_pfn)
5687                                 continue;
5688 
5689                         /* Account for what is only usable for kernelcore */
5690                         if (start_pfn < usable_startpfn) {
5691                                 unsigned long kernel_pages;
5692                                 kernel_pages = min(end_pfn, usable_startpfn)
5693                                                                 - start_pfn;
5694 
5695                                 kernelcore_remaining -= min(kernel_pages,
5696                                                         kernelcore_remaining);
5697                                 required_kernelcore -= min(kernel_pages,
5698                                                         required_kernelcore);
5699 
5700                                 /* Continue if range is now fully accounted */
5701                                 if (end_pfn <= usable_startpfn) {
5702 
5703                                         /*
5704                                          * Push zone_movable_pfn to the end so
5705                                          * that if we have to rebalance
5706                                          * kernelcore across nodes, we will
5707                                          * not double account here
5708                                          */
5709                                         zone_movable_pfn[nid] = end_pfn;
5710                                         continue;
5711                                 }
5712                                 start_pfn = usable_startpfn;
5713                         }
5714 
5715                         /*
5716                          * The usable PFN range for ZONE_MOVABLE is from
5717                          * start_pfn->end_pfn. Calculate size_pages as the
5718                          * number of pages used as kernelcore
5719                          */
5720                         size_pages = end_pfn - start_pfn;
5721                         if (size_pages > kernelcore_remaining)
5722                                 size_pages = kernelcore_remaining;
5723                         zone_movable_pfn[nid] = start_pfn + size_pages;
5724 
5725                         /*
5726                          * Some kernelcore has been met, update counts and
5727                          * break if the kernelcore for this node has been
5728                          * satisfied
5729                          */
5730                         required_kernelcore -= min(required_kernelcore,
5731                                                                 size_pages);
5732                         kernelcore_remaining -= size_pages;
5733                         if (!kernelcore_remaining)
5734                                 break;
5735                 }
5736         }
5737 
5738         /*
5739          * If there is still required_kernelcore, we do another pass with one
5740          * less node in the count. This will push zone_movable_pfn[nid] further
5741          * along on the nodes that still have memory until kernelcore is
5742          * satisfied
5743          */
5744         usable_nodes--;
5745         if (usable_nodes && required_kernelcore > usable_nodes)
5746                 goto restart;
5747 
5748 out2:
5749         /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5750         for (nid = 0; nid < MAX_NUMNODES; nid++)
5751                 zone_movable_pfn[nid] =
5752                         roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5753 
5754 out:
5755         /* restore the node_state */
5756         node_states[N_MEMORY] = saved_node_state;
5757 }
5758 
5759 /* Any regular or high memory on that node ? */
5760 static void check_for_memory(pg_data_t *pgdat, int nid)
5761 {
5762         enum zone_type zone_type;
5763 
5764         if (N_MEMORY == N_NORMAL_MEMORY)
5765                 return;
5766 
5767         for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5768                 struct zone *zone = &pgdat->node_zones[zone_type];
5769                 if (populated_zone(zone)) {
5770                         node_set_state(nid, N_HIGH_MEMORY);
5771                         if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5772                             zone_type <= ZONE_NORMAL)
5773                                 node_set_state(nid, N_NORMAL_MEMORY);
5774                         break;
5775                 }
5776         }
5777 }
5778 
5779 /**
5780  * free_area_init_nodes - Initialise all pg_data_t and zone data
5781  * @max_zone_pfn: an array of max PFNs for each zone
5782  *
5783  * This will call free_area_init_node() for each active node in the system.
5784  * Using the page ranges provided by memblock_set_node(), the size of each
5785  * zone in each node and their holes is calculated. If the maximum PFN
5786  * between two adjacent zones match, it is assumed that the zone is empty.
5787  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5788  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5789  * starts where the previous one ended. For example, ZONE_DMA32 starts
5790  * at arch_max_dma_pfn.
5791  */
5792 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5793 {
5794         unsigned long start_pfn, end_pfn;
5795         int i, nid;
5796 
5797         /* Record where the zone boundaries are */
5798         memset(arch_zone_lowest_possible_pfn, 0,
5799                                 sizeof(arch_zone_lowest_possible_pfn));
5800         memset(arch_zone_highest_possible_pfn, 0,
5801                                 sizeof(arch_zone_highest_possible_pfn));
5802         arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5803         arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5804         for (i = 1; i < MAX_NR_ZONES; i++) {
5805                 if (i == ZONE_MOVABLE)
5806                         continue;
5807                 arch_zone_lowest_possible_pfn[i] =
5808                         arch_zone_highest_possible_pfn[i-1];
5809                 arch_zone_highest_possible_pfn[i] =
5810                         max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5811         }
5812         arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5813         arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5814 
5815         /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5816         memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5817         find_zone_movable_pfns_for_nodes();
5818 
5819         /* Print out the zone ranges */
5820         pr_info("Zone ranges:\n");
5821         for (i = 0; i < MAX_NR_ZONES; i++) {
5822                 if (i == ZONE_MOVABLE)
5823                         continue;
5824                 pr_info("  %-8s ", zone_names[i]);
5825                 if (arch_zone_lowest_possible_pfn[i] ==
5826                                 arch_zone_highest_possible_pfn[i])
5827                         pr_cont("empty\n");
5828                 else
5829                         pr_cont("[mem %#018Lx-%#018Lx]\n",
5830                                 (u64)arch_zone_lowest_possible_pfn[i]
5831                                         << PAGE_SHIFT,
5832                                 ((u64)arch_zone_highest_possible_pfn[i]
5833                                         << PAGE_SHIFT) - 1);
5834         }
5835 
5836         /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5837         pr_info("Movable zone start for each node\n");
5838         for (i = 0; i < MAX_NUMNODES; i++) {
5839                 if (zone_movable_pfn[i])
5840                         pr_info("  Node %d: %#018Lx\n", i,
5841                                (u64)zone_movable_pfn[i] << PAGE_SHIFT);
5842         }
5843 
5844         /* Print out the early node map */
5845         pr_info("Early memory node ranges\n");
5846         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5847                 pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5848                         (u64)start_pfn << PAGE_SHIFT,
5849                         ((u64)end_pfn << PAGE_SHIFT) - 1);
5850 
5851         /* Initialise every node */
5852         mminit_verify_pageflags_layout();
5853         setup_nr_node_ids();
5854         for_each_online_node(nid) {
5855                 pg_data_t *pgdat = NODE_DATA(nid);
5856                 free_area_init_node(nid, NULL,
5857                                 find_min_pfn_for_node(nid), NULL);
5858 
5859                 /* Any memory on that node */
5860                 if (pgdat->node_present_pages)
5861                         node_set_state(nid, N_MEMORY);
5862                 check_for_memory(pgdat, nid);
5863         }
5864 }
5865 
5866 static int __init cmdline_parse_core(char *p, unsigned long *core)
5867 {
5868         unsigned long long coremem;
5869         if (!p)
5870                 return -EINVAL;
5871 
5872         coremem = memparse(p, &p);
5873         *core = coremem >> PAGE_SHIFT;
5874 
5875         /* Paranoid check that UL is enough for the coremem value */
5876         WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5877 
5878         return 0;
5879 }
5880 
5881 /*
5882  * kernelcore=size sets the amount of memory for use for allocations that
5883  * cannot be reclaimed or migrated.
5884  */
5885 static int __init cmdline_parse_kernelcore(char *p)
5886 {
5887         return cmdline_parse_core(p, &required_kernelcore);
5888 }
5889 
5890 /*
5891  * movablecore=size sets the amount of memory for use for allocations that
5892  * can be reclaimed or migrated.
5893  */
5894 static int __init cmdline_parse_movablecore(char *p)
5895 {
5896         return cmdline_parse_core(p, &required_movablecore);
5897 }
5898 
5899 early_param("kernelcore", cmdline_parse_kernelcore);
5900 early_param("movablecore", cmdline_parse_movablecore);
5901 
5902 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5903 
5904 void adjust_managed_page_count(struct page *page, long count)
5905 {
5906         spin_lock(&managed_page_count_lock);
5907         page_zone(page)->managed_pages += count;
5908         totalram_pages += count;
5909 #ifdef CONFIG_HIGHMEM
5910         if (PageHighMem(page))
5911                 totalhigh_pages += count;
5912 #endif
5913         spin_unlock(&managed_page_count_lock);
5914 }
5915 EXPORT_SYMBOL(adjust_managed_page_count);
5916 
5917 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
5918 {
5919         void *pos;
5920         unsigned long pages = 0;
5921 
5922         start = (void *)PAGE_ALIGN((unsigned long)start);
5923         end = (void *)((unsigned long)end & PAGE_MASK);
5924         for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
5925                 if ((unsigned int)poison <= 0xFF)
5926                         memset(pos, poison, PAGE_SIZE);
5927                 free_reserved_page(virt_to_page(pos));
5928         }
5929 
5930         if (pages && s)
5931                 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5932                         s, pages << (PAGE_SHIFT - 10), start, end);
5933 
5934         return pages;
5935 }
5936 EXPORT_SYMBOL(free_reserved_area);
5937 
5938 #ifdef  CONFIG_HIGHMEM
5939 void free_highmem_page(struct page *page)
5940 {
5941         __free_reserved_page(page);
5942         totalram_pages++;
5943         page_zone(page)->managed_pages++;
5944         totalhigh_pages++;
5945 }
5946 #endif
5947 
5948 
5949 void __init mem_init_print_info(const char *str)
5950 {
5951         unsigned long physpages, codesize, datasize, rosize, bss_size;
5952         unsigned long init_code_size, init_data_size;
5953 
5954         physpages = get_num_physpages();
5955         codesize = _etext - _stext;
5956         datasize = _edata - _sdata;
5957         rosize = __end_rodata - __start_rodata;
5958         bss_size = __bss_stop - __bss_start;
5959         init_data_size = __init_end - __init_begin;
5960         init_code_size = _einittext - _sinittext;
5961 
5962         /*
5963          * Detect special cases and adjust section sizes accordingly:
5964          * 1) .init.* may be embedded into .data sections
5965          * 2) .init.text.* may be out of [__init_begin, __init_end],
5966          *    please refer to arch/tile/kernel/vmlinux.lds.S.
5967          * 3) .rodata.* may be embedded into .text or .data sections.
5968          */
5969 #define adj_init_size(start, end, size, pos, adj) \
5970         do { \
5971                 if (start <= pos && pos < end && size > adj) \
5972                         size -= adj; \
5973         } while (0)
5974 
5975         adj_init_size(__init_begin, __init_end, init_data_size,
5976                      _sinittext, init_code_size);
5977         adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5978         adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5979         adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5980         adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5981 
5982 #undef  adj_init_size
5983 
5984         pr_info("Memory: %luK/%luK available "
5985                "(%luK kernel code, %luK rwdata, %luK rodata, "
5986                "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
5987 #ifdef  CONFIG_HIGHMEM
5988                ", %luK highmem"
5989 #endif
5990                "%s%s)\n",
5991                nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5992                codesize >> 10, datasize >> 10, rosize >> 10,
5993                (init_data_size + init_code_size) >> 10, bss_size >> 10,
5994                (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5995                totalcma_pages << (PAGE_SHIFT-10),
5996 #ifdef  CONFIG_HIGHMEM
5997                totalhigh_pages << (PAGE_SHIFT-10),
5998 #endif
5999                str ? ", " : "", str ? str : "");
6000 }
6001 
6002 /**
6003  * set_dma_reserve - set the specified number of pages reserved in the first zone
6004  * @new_dma_reserve: The number of pages to mark reserved
6005  *
6006  * The per-cpu batchsize and zone watermarks are determined by present_pages.
6007  * In the DMA zone, a significant percentage may be consumed by kernel image
6008  * and other unfreeable allocations which can skew the watermarks badly. This
6009  * function may optionally be used to account for unfreeable pages in the
6010  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6011  * smaller per-cpu batchsize.
6012  */
6013 void __init set_dma_reserve(unsigned long new_dma_reserve)
6014 {
6015         dma_reserve = new_dma_reserve;
6016 }
6017 
6018 void __init free_area_init(unsigned long *zones_size)
6019 {
6020         free_area_init_node(0, zones_size,
6021                         __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6022 }
6023 
6024 static int page_alloc_cpu_notify(struct notifier_block *self,
6025                                  unsigned long action, void *hcpu)
6026 {
6027         int cpu = (unsigned long)hcpu;
6028 
6029         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
6030                 lru_add_drain_cpu(cpu);
6031                 drain_pages(cpu);
6032 
6033                 /*
6034                  * Spill the event counters of the dead processor
6035                  * into the current processors event counters.
6036                  * This artificially elevates the count of the current
6037                  * processor.
6038                  */
6039                 vm_events_fold_cpu(cpu);
6040 
6041                 /*
6042                  * Zero the differential counters of the dead processor
6043                  * so that the vm statistics are consistent.
6044                  *
6045                  * This is only okay since the processor is dead and cannot
6046                  * race with what we are doing.
6047                  */
6048                 cpu_vm_stats_fold(cpu);
6049         }
6050         return NOTIFY_OK;
6051 }
6052 
6053 void __init page_alloc_init(void)
6054 {
6055         hotcpu_notifier(page_alloc_cpu_notify, 0);
6056 }
6057 
6058 /*
6059  * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
6060  *      or min_free_kbytes changes.
6061  */
6062 static void calculate_totalreserve_pages(void)
6063 {
6064         struct pglist_data *pgdat;
6065         unsigned long reserve_pages = 0;
6066         enum zone_type i, j;
6067 
6068         for_each_online_pgdat(pgdat) {
6069                 for (i = 0; i < MAX_NR_ZONES; i++) {
6070                         struct zone *zone = pgdat->node_zones + i;
6071                         long max = 0;
6072 
6073                         /* Find valid and maximum lowmem_reserve in the zone */
6074                         for (j = i; j < MAX_NR_ZONES; j++) {
6075                                 if (zone->lowmem_reserve[j] > max)
6076                                         max = zone->lowmem_reserve[j];
6077                         }
6078 
6079                         /* we treat the high watermark as reserved pages. */
6080                         max += high_wmark_pages(zone);
6081 
6082                         if (max > zone->managed_pages)
6083                                 max = zone->managed_pages;
6084                         reserve_pages += max;
6085                         /*
6086                          * Lowmem reserves are not available to
6087                          * GFP_HIGHUSER page cache allocations and
6088                          * kswapd tries to balance zones to their high
6089                          * watermark.  As a result, neither should be
6090                          * regarded as dirtyable memory, to prevent a
6091                          * situation where reclaim has to clean pages
6092                          * in order to balance the zones.
6093                          */
6094                         zone->dirty_balance_reserve = max;
6095                 }
6096         }
6097         dirty_balance_reserve = reserve_pages;
6098         totalreserve_pages = reserve_pages;
6099 }
6100 
6101 /*
6102  * setup_per_zone_lowmem_reserve - called whenever
6103  *      sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
6104  *      has a correct pages reserved value, so an adequate number of
6105  *      pages are left in the zone after a successful __alloc_pages().
6106  */
6107 static void setup_per_zone_lowmem_reserve(void)
6108 {
6109         struct pglist_data *pgdat;
6110         enum zone_type j, idx;
6111 
6112         for_each_online_pgdat(pgdat) {
6113                 for (j = 0; j < MAX_NR_ZONES; j++) {
6114                         struct zone *zone = pgdat->node_zones + j;
6115                         unsigned long managed_pages = zone->managed_pages;
6116 
6117                         zone->lowmem_reserve[j] = 0;
6118 
6119                         idx = j;
6120                         while (idx) {
6121                                 struct zone *lower_zone;
6122 
6123                                 idx--;
6124 
6125                                 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6126                                         sysctl_lowmem_reserve_ratio[idx] = 1;
6127 
6128                                 lower_zone = pgdat->node_zones + idx;
6129                                 lower_zone->lowmem_reserve[j] = managed_pages /
6130                                         sysctl_lowmem_reserve_ratio[idx];
6131                                 managed_pages += lower_zone->managed_pages;
6132                         }
6133                 }
6134         }
6135 
6136         /* update totalreserve_pages */
6137         calculate_totalreserve_pages();
6138 }
6139 
6140 static void __setup_per_zone_wmarks(void)
6141 {
6142         unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6143         unsigned long lowmem_pages = 0;
6144         struct zone *zone;
6145         unsigned long flags;
6146 
6147         /* Calculate total number of !ZONE_HIGHMEM pages */
6148         for_each_zone(zone) {
6149                 if (!is_highmem(zone))
6150                         lowmem_pages += zone->managed_pages;
6151         }
6152 
6153         for_each_zone(zone) {
6154                 u64 tmp;
6155 
6156                 spin_lock_irqsave(&zone->lock, flags);
6157                 tmp = (u64)pages_min * zone->managed_pages;
6158                 do_div(tmp, lowmem_pages);
6159                 if (is_highmem(zone)) {
6160                         /*
6161                          * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6162                          * need highmem pages, so cap pages_min to a small
6163                          * value here.
6164                          *
6165                          * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6166                          * deltas control asynch page reclaim, and so should
6167                          * not be capped for highmem.
6168                          */
6169                         unsigned long min_pages;
6170 
6171                         min_pages = zone->managed_pages / 1024;
6172                         min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
6173                         zone->watermark[WMARK_MIN] = min_pages;
6174                 } else {
6175                         /*
6176                          * If it's a lowmem zone, reserve a number of pages
6177                          * proportionate to the zone's size.
6178                          */
6179                         zone->watermark[WMARK_MIN] = tmp;
6180                 }
6181 
6182                 zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + (tmp >> 2);
6183                 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
6184 
6185                 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
6186                         high_wmark_pages(zone) - low_wmark_pages(zone) -
6187                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
6188 
6189                 setup_zone_migrate_reserve(zone);
6190                 spin_unlock_irqrestore(&zone->lock, flags);
6191         }
6192 
6193         /* update totalreserve_pages */
6194         calculate_totalreserve_pages();
6195 }
6196 
6197 /**
6198  * setup_per_zone_wmarks - called when min_free_kbytes changes
6199  * or when memory is hot-{added|removed}
6200  *
6201  * Ensures that the watermark[min,low,high] values for each zone are set
6202  * correctly with respect to min_free_kbytes.
6203  */
6204 void setup_per_zone_wmarks(void)
6205 {
6206         mutex_lock(&zonelists_mutex);
6207         __setup_per_zone_wmarks();
6208         mutex_unlock(&zonelists_mutex);
6209 }
6210 
6211 /*
6212  * The inactive anon list should be small enough that the VM never has to
6213  * do too much work, but large enough that each inactive page has a chance
6214  * to be referenced again before it is swapped out.
6215  *
6216  * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6217  * INACTIVE_ANON pages on this zone's LRU, maintained by the
6218  * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6219  * the anonymous pages are kept on the inactive list.
6220  *
6221  * total     target    max
6222  * memory    ratio     inactive anon
6223  * -------------------------------------
6224  *   10MB       1         5MB
6225  *  100MB       1        50MB
6226  *    1GB       3       250MB
6227  *   10GB      10       0.9GB
6228  *  100GB      31         3GB
6229  *    1TB     101        10GB
6230  *   10TB     320        32GB
6231  */
6232 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
6233 {
6234         unsigned int gb, ratio;
6235 
6236         /* Zone size in gigabytes */
6237         gb = zone->managed_pages >> (30 - PAGE_SHIFT);
6238         if (gb)
6239                 ratio = int_sqrt(10 * gb);
6240         else
6241                 ratio = 1;
6242 
6243         zone->inactive_ratio = ratio;
6244 }
6245 
6246 static void __meminit setup_per_zone_inactive_ratio(void)
6247 {
6248         struct zone *zone;
6249 
6250         for_each_zone(zone)
6251                 calculate_zone_inactive_ratio(zone);
6252 }
6253 
6254 /*
6255  * Initialise min_free_kbytes.
6256  *
6257  * For small machines we want it small (128k min).  For large machines
6258  * we want it large (64MB max).  But it is not linear, because network
6259  * bandwidth does not increase linearly with machine size.  We use
6260  *
6261  *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6262  *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
6263  *
6264  * which yields
6265  *
6266  * 16MB:        512k
6267  * 32MB:        724k
6268  * 64MB:        1024k
6269  * 128MB:       1448k
6270  * 256MB:       2048k
6271  * 512MB:       2896k
6272  * 1024MB:      4096k
6273  * 2048MB:      5792k
6274  * 4096MB:      8192k
6275  * 8192MB:      11584k
6276  * 16384MB:     16384k
6277  */
6278 int __meminit init_per_zone_wmark_min(void)
6279 {
6280         unsigned long lowmem_kbytes;
6281         int new_min_free_kbytes;
6282 
6283         lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
6284         new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
6285 
6286         if (new_min_free_kbytes > user_min_free_kbytes) {
6287                 min_free_kbytes = new_min_free_kbytes;
6288                 if (min_free_kbytes < 128)
6289                         min_free_kbytes = 128;
6290                 if (min_free_kbytes > 65536)
6291                         min_free_kbytes = 65536;
6292         } else {
6293                 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6294                                 new_min_free_kbytes, user_min_free_kbytes);
6295         }
6296         setup_per_zone_wmarks();
6297         refresh_zone_stat_thresholds();
6298         setup_per_zone_lowmem_reserve();
6299         setup_per_zone_inactive_ratio();
6300         return 0;
6301 }
6302 module_init(init_per_zone_wmark_min)
6303 
6304 /*
6305  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6306  *      that we can call two helper functions whenever min_free_kbytes
6307  *      changes.
6308  */
6309 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
6310         void __user *buffer, size_t *length, loff_t *ppos)
6311 {
6312         int rc;
6313 
6314         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6315         if (rc)
6316                 return rc;
6317 
6318         if (write) {
6319                 user_min_free_kbytes = min_free_kbytes;
6320                 setup_per_zone_wmarks();
6321         }
6322         return 0;
6323 }
6324 
6325 #ifdef CONFIG_NUMA
6326 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
6327         void __user *buffer, size_t *length, loff_t *ppos)
6328 {
6329         struct zone *zone;
6330         int rc;
6331 
6332         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6333         if (rc)
6334                 return rc;
6335 
6336         for_each_zone(zone)
6337                 zone->min_unmapped_pages = (zone->managed_pages *
6338                                 sysctl_min_unmapped_ratio) / 100;
6339         return 0;
6340 }
6341 
6342 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
6343         void __user *buffer, size_t *length, loff_t *ppos)
6344 {
6345         struct zone *zone;
6346         int rc;
6347 
6348         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6349         if (rc)
6350                 return rc;
6351 
6352         for_each_zone(zone)
6353                 zone->min_slab_pages = (zone->managed_pages *
6354                                 sysctl_min_slab_ratio) / 100;
6355         return 0;
6356 }
6357 #endif
6358 
6359 /*
6360  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6361  *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6362  *      whenever sysctl_lowmem_reserve_ratio changes.
6363  *
6364  * The reserve ratio obviously has absolutely no relation with the
6365  * minimum watermarks. The lowmem reserve ratio can only make sense
6366  * if in function of the boot time zone sizes.
6367  */
6368 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
6369         void __user *buffer, size_t *length, loff_t *ppos)
6370 {
6371         proc_dointvec_minmax(table, write, buffer, length, ppos);
6372         setup_per_zone_lowmem_reserve();
6373         return 0;
6374 }
6375 
6376 /*
6377  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6378  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
6379  * pagelist can have before it gets flushed back to buddy allocator.
6380  */
6381 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
6382         void __user *buffer, size_t *length, loff_t *ppos)
6383 {
6384         struct zone *zone;
6385         int old_percpu_pagelist_fraction;
6386         int ret;
6387 
6388         mutex_lock(&pcp_batch_high_lock);
6389         old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6390 
6391         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6392         if (!write || ret < 0)
6393                 goto out;
6394 
6395         /* Sanity checking to avoid pcp imbalance */
6396         if (percpu_pagelist_fraction &&
6397             percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6398                 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6399                 ret = -EINVAL;
6400                 goto out;
6401         }
6402 
6403         /* No change? */
6404         if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6405                 goto out;
6406 
6407         for_each_populated_zone(zone) {
6408                 unsigned int cpu;
6409 
6410                 for_each_possible_cpu(cpu)
6411                         pageset_set_high_and_batch(zone,
6412                                         per_cpu_ptr(zone->pageset, cpu));
6413         }
6414 out:
6415         mutex_unlock(&pcp_batch_high_lock);
6416         return ret;
6417 }
6418 
6419 #ifdef CONFIG_NUMA
6420 int hashdist = HASHDIST_DEFAULT;
6421 
6422 static int __init set_hashdist(char *str)
6423 {
6424         if (!str)
6425                 return 0;
6426         hashdist = simple_strtoul(str, &str, 0);
6427         return 1;
6428 }
6429 __setup("hashdist=", set_hashdist);
6430 #endif
6431 
6432 /*
6433  * allocate a large system hash table from bootmem
6434  * - it is assumed that the hash table must contain an exact power-of-2
6435  *   quantity of entries
6436  * - limit is the number of hash buckets, not the total allocation size
6437  */
6438 void *__init alloc_large_system_hash(const char *tablename,
6439                                      unsigned long bucketsize,
6440                                      unsigned long numentries,
6441                                      int scale,
6442                                      int flags,
6443                                      unsigned int *_hash_shift,
6444                                      unsigned int *_hash_mask,
6445                                      unsigned long low_limit,
6446                                      unsigned long high_limit)
6447 {
6448         unsigned long long max = high_limit;
6449         unsigned long log2qty, size;
6450         void *table = NULL;
6451 
6452         /* allow the kernel cmdline to have a say */
6453         if (!numentries) {
6454                 /* round applicable memory size up to nearest megabyte */
6455                 numentries = nr_kernel_pages;
6456 
6457                 /* It isn't necessary when PAGE_SIZE >= 1MB */
6458                 if (PAGE_SHIFT < 20)
6459                         numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
6460 
6461                 /* limit to 1 bucket per 2^scale bytes of low memory */
6462                 if (scale > PAGE_SHIFT)
6463                         numentries >>= (scale - PAGE_SHIFT);
6464                 else
6465                         numentries <<= (PAGE_SHIFT - scale);
6466 
6467                 /* Make sure we've got at least a 0-order allocation.. */
6468                 if (unlikely(flags & HASH_SMALL)) {
6469                         /* Makes no sense without HASH_EARLY */
6470                         WARN_ON(!(flags & HASH_EARLY));
6471                         if (!(numentries >> *_hash_shift)) {
6472                                 numentries = 1UL << *_hash_shift;
6473                                 BUG_ON(!numentries);
6474                         }
6475                 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
6476                         numentries = PAGE_SIZE / bucketsize;
6477         }
6478         numentries = roundup_pow_of_two(numentries);
6479 
6480         /* limit allocation size to 1/16 total memory by default */
6481         if (max == 0) {
6482                 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6483                 do_div(max, bucketsize);
6484         }
6485         max = min(max, 0x80000000ULL);
6486 
6487         if (numentries < low_limit)
6488                 numentries = low_limit;
6489         if (numentries > max)
6490                 numentries = max;
6491 
6492         log2qty = ilog2(numentries);
6493 
6494         do {
6495                 size = bucketsize << log2qty;
6496                 if (flags & HASH_EARLY)
6497                         table = memblock_virt_alloc_nopanic(size, 0);
6498                 else if (hashdist)
6499                         table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6500                 else {
6501                         /*
6502                          * If bucketsize is not a power-of-two, we may free
6503                          * some pages at the end of hash table which
6504                          * alloc_pages_exact() automatically does
6505                          */
6506                         if (get_order(size) < MAX_ORDER) {
6507                                 table = alloc_pages_exact(size, GFP_ATOMIC);
6508                                 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6509                         }
6510                 }
6511         } while (!table && size > PAGE_SIZE && --log2qty);
6512 
6513         if (!table)
6514                 panic("Failed to allocate %s hash table\n", tablename);
6515 
6516         printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
6517                tablename,
6518                (1UL << log2qty),
6519                ilog2(size) - PAGE_SHIFT,
6520                size);
6521 
6522         if (_hash_shift)
6523                 *_hash_shift = log2qty;
6524         if (_hash_mask)
6525                 *_hash_mask = (1 << log2qty) - 1;
6526 
6527         return table;
6528 }
6529 
6530 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6531 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6532                                                         unsigned long pfn)
6533 {
6534 #ifdef CONFIG_SPARSEMEM
6535         return __pfn_to_section(pfn)->pageblock_flags;
6536 #else
6537         return zone->pageblock_flags;
6538 #endif /* CONFIG_SPARSEMEM */
6539 }
6540 
6541 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6542 {
6543 #ifdef CONFIG_SPARSEMEM
6544         pfn &= (PAGES_PER_SECTION-1);
6545         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6546 #else
6547         pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
6548         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6549 #endif /* CONFIG_SPARSEMEM */
6550 }
6551 
6552 /**
6553  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6554  * @page: The page within the block of interest
6555  * @pfn: The target page frame number
6556  * @end_bitidx: The last bit of interest to retrieve
6557  * @mask: mask of bits that the caller is interested in
6558  *
6559  * Return: pageblock_bits flags
6560  */
6561 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
6562                                         unsigned long end_bitidx,
6563                                         unsigned long mask)
6564 {
6565         struct zone *zone;
6566         unsigned long *bitmap;
6567         unsigned long bitidx, word_bitidx;
6568         unsigned long word;
6569 
6570         zone = page_zone(page);
6571         bitmap = get_pageblock_bitmap(zone, pfn);
6572         bitidx = pfn_to_bitidx(zone, pfn);
6573         word_bitidx = bitidx / BITS_PER_LONG;
6574         bitidx &= (BITS_PER_LONG-1);
6575 
6576         word = bitmap[word_bitidx];
6577         bitidx += end_bitidx;
6578         return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
6579 }
6580 
6581 /**
6582  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6583  * @page: The page within the block of interest
6584  * @flags: The flags to set
6585  * @pfn: The target page frame number
6586  * @end_bitidx: The last bit of interest
6587  * @mask: mask of bits that the caller is interested in
6588  */
6589 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6590                                         unsigned long pfn,
6591                                         unsigned long end_bitidx,
6592                                         unsigned long mask)
6593 {
6594         struct zone *zone;
6595         unsigned long *bitmap;
6596         unsigned long bitidx, word_bitidx;
6597         unsigned long old_word, word;
6598 
6599         BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
6600 
6601         zone = page_zone(page);
6602         bitmap = get_pageblock_bitmap(zone, pfn);
6603         bitidx = pfn_to_bitidx(zone, pfn);
6604         word_bitidx = bitidx / BITS_PER_LONG;
6605         bitidx &= (BITS_PER_LONG-1);
6606 
6607         VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
6608 
6609         bitidx += end_bitidx;
6610         mask <<= (BITS_PER_LONG - bitidx - 1);
6611         flags <<= (BITS_PER_LONG - bitidx - 1);
6612 
6613         word = READ_ONCE(bitmap[word_bitidx]);
6614         for (;;) {
6615                 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6616                 if (word == old_word)
6617                         break;
6618                 word = old_word;
6619         }
6620 }
6621 
6622 /*
6623  * This function checks whether pageblock includes unmovable pages or not.
6624  * If @count is not zero, it is okay to include less @count unmovable pages
6625  *
6626  * PageLRU check without isolation or lru_lock could race so that
6627  * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6628  * expect this function should be exact.
6629  */
6630 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6631                          bool skip_hwpoisoned_pages)
6632 {
6633         unsigned long pfn, iter, found;
6634         int mt;
6635 
6636         /*
6637          * For avoiding noise data, lru_add_drain_all() should be called
6638          * If ZONE_MOVABLE, the zone never contains unmovable pages
6639          */
6640         if (zone_idx(zone) == ZONE_MOVABLE)
6641                 return false;
6642         mt = get_pageblock_migratetype(page);
6643         if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6644                 return false;
6645 
6646         pfn = page_to_pfn(page);
6647         for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6648                 unsigned long check = pfn + iter;
6649 
6650                 if (!pfn_valid_within(check))
6651                         continue;
6652 
6653                 page = pfn_to_page(check);
6654 
6655                 /*
6656                  * Hugepages are not in LRU lists, but they're movable.
6657                  * We need not scan over tail pages bacause we don't
6658                  * handle each tail page individually in migration.
6659                  */
6660                 if (PageHuge(page)) {
6661                         iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6662                         continue;
6663                 }
6664 
6665                 /*
6666                  * We can't use page_count without pin a page
6667                  * because another CPU can free compound page.
6668                  * This check already skips compound tails of THP
6669                  * because their page->_count is zero at all time.
6670                  */
6671                 if (!atomic_read(&page->_count)) {
6672                         if (PageBuddy(page))
6673                                 iter += (1 << page_order(page)) - 1;
6674                         continue;
6675                 }
6676 
6677                 /*
6678                  * The HWPoisoned page may be not in buddy system, and
6679                  * page_count() is not 0.
6680                  */
6681                 if (skip_hwpoisoned_pages && PageHWPoison(page))
6682                         continue;
6683 
6684                 if (!PageLRU(page))
6685                         found++;
6686                 /*
6687                  * If there are RECLAIMABLE pages, we need to check
6688                  * it.  But now, memory offline itself doesn't call
6689                  * shrink_node_slabs() and it still to be fixed.
6690                  */
6691                 /*
6692                  * If the page is not RAM, page_count()should be 0.
6693                  * we don't need more check. This is an _used_ not-movable page.
6694                  *
6695                  * The problematic thing here is PG_reserved pages. PG_reserved
6696                  * is set to both of a memory hole page and a _used_ kernel
6697                  * page at boot.
6698                  */
6699                 if (found > count)
6700                         return true;
6701         }
6702         return false;
6703 }
6704 
6705 bool is_pageblock_removable_nolock(struct page *page)
6706 {
6707         struct zone *zone;
6708         unsigned long pfn;
6709 
6710         /*
6711          * We have to be careful here because we are iterating over memory
6712          * sections which are not zone aware so we might end up outside of
6713          * the zone but still within the section.
6714          * We have to take care about the node as well. If the node is offline
6715          * its NODE_DATA will be NULL - see page_zone.
6716          */
6717         if (!node_online(page_to_nid(page)))
6718                 return false;
6719 
6720         zone = page_zone(page);
6721         pfn = page_to_pfn(page);
6722         if (!zone_spans_pfn(zone, pfn))
6723                 return false;
6724 
6725         return !has_unmovable_pages(zone, page, 0, true);
6726 }
6727 
6728 #ifdef CONFIG_CMA
6729 
6730 static unsigned long pfn_max_align_down(unsigned long pfn)
6731 {
6732         return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6733                              pageblock_nr_pages) - 1);
6734 }
6735 
6736 static unsigned long pfn_max_align_up(unsigned long pfn)
6737 {
6738         return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6739                                 pageblock_nr_pages));
6740 }
6741 
6742 /* [start, end) must belong to a single zone. */
6743 static int __alloc_contig_migrate_range(struct compact_control *cc,
6744                                         unsigned long start, unsigned long end)
6745 {
6746         /* This function is based on compact_zone() from compaction.c. */
6747         unsigned long nr_reclaimed;
6748         unsigned long pfn = start;
6749         unsigned int tries = 0;
6750         int ret = 0;
6751 
6752         migrate_prep();
6753 
6754         while (pfn < end || !list_empty(&cc->migratepages)) {
6755                 if (fatal_signal_pending(current)) {
6756                         ret = -EINTR;
6757                         break;
6758                 }
6759 
6760                 if (list_empty(&cc->migratepages)) {
6761                         cc->nr_migratepages = 0;
6762                         pfn = isolate_migratepages_range(cc, pfn, end);
6763                         if (!pfn) {
6764                                 ret = -EINTR;
6765                                 break;
6766                         }
6767                         tries = 0;
6768                 } else if (++tries == 5) {
6769                         ret = ret < 0 ? ret : -EBUSY;
6770                         break;
6771                 }
6772 
6773                 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6774                                                         &cc->migratepages);
6775                 cc->nr_migratepages -= nr_reclaimed;
6776 
6777                 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6778                                     NULL, 0, cc->mode, MR_CMA);
6779         }
6780         if (ret < 0) {
6781                 putback_movable_pages(&cc->migratepages);
6782                 return ret;
6783         }
6784         return 0;
6785 }
6786 
6787 /**
6788  * alloc_contig_range() -- tries to allocate given range of pages
6789  * @start:      start PFN to allocate
6790  * @end:        one-past-the-last PFN to allocate
6791  * @migratetype:        migratetype of the underlaying pageblocks (either
6792  *                      #MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
6793  *                      in range must have the same migratetype and it must
6794  *                      be either of the two.
6795  *
6796  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6797  * aligned, however it's the caller's responsibility to guarantee that
6798  * we are the only thread that changes migrate type of pageblocks the
6799  * pages fall in.
6800  *
6801  * The PFN range must belong to a single zone.
6802  *
6803  * Returns zero on success or negative error code.  On success all
6804  * pages which PFN is in [start, end) are allocated for the caller and
6805  * need to be freed with free_contig_range().
6806  */
6807 int alloc_contig_range(unsigned long start, unsigned long end,
6808                        unsigned migratetype)
6809 {
6810         unsigned long outer_start, outer_end;
6811         int ret = 0, order;
6812 
6813         struct compact_control cc = {
6814                 .nr_migratepages = 0,
6815                 .order = -1,
6816                 .zone = page_zone(pfn_to_page(start)),
6817                 .mode = MIGRATE_SYNC,
6818                 .ignore_skip_hint = true,
6819         };
6820         INIT_LIST_HEAD(&cc.migratepages);
6821 
6822         /*
6823          * What we do here is we mark all pageblocks in range as
6824          * MIGRATE_ISOLATE.  Because pageblock and max order pages may
6825          * have different sizes, and due to the way page allocator
6826          * work, we align the range to biggest of the two pages so
6827          * that page allocator won't try to merge buddies from
6828          * different pageblocks and change MIGRATE_ISOLATE to some
6829          * other migration type.
6830          *
6831          * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6832          * migrate the pages from an unaligned range (ie. pages that
6833          * we are interested in).  This will put all the pages in
6834          * range back to page allocator as MIGRATE_ISOLATE.
6835          *
6836          * When this is done, we take the pages in range from page
6837          * allocator removing them from the buddy system.  This way
6838          * page allocator will never consider using them.
6839          *
6840          * This lets us mark the pageblocks back as
6841          * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6842          * aligned range but not in the unaligned, original range are
6843          * put back to page allocator so that buddy can use them.
6844          */
6845 
6846         ret = start_isolate_page_range(pfn_max_align_down(start),
6847                                        pfn_max_align_up(end), migratetype,
6848                                        false);
6849         if (ret)
6850                 return ret;
6851 
6852         ret = __alloc_contig_migrate_range(&cc, start, end);
6853         if (ret)
6854                 goto done;
6855 
6856         /*
6857          * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6858          * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
6859          * more, all pages in [start, end) are free in page allocator.
6860          * What we are going to do is to allocate all pages from
6861          * [start, end) (that is remove them from page allocator).
6862          *
6863          * The only problem is that pages at the beginning and at the
6864          * end of interesting range may be not aligned with pages that
6865          * page allocator holds, ie. they can be part of higher order
6866          * pages.  Because of this, we reserve the bigger range and
6867          * once this is done free the pages we are not interested in.
6868          *
6869          * We don't have to hold zone->lock here because the pages are
6870          * isolated thus they won't get removed from buddy.
6871          */
6872 
6873         lru_add_drain_all();
6874         drain_all_pages(cc.zone);
6875 
6876         order = 0;
6877         outer_start = start;
6878         while (!PageBuddy(pfn_to_page(outer_start))) {
6879                 if (++order >= MAX_ORDER) {
6880                         ret = -EBUSY;
6881                         goto done;
6882                 }
6883                 outer_start &= ~0UL << order;
6884         }
6885 
6886         /* Make sure the range is really isolated. */
6887         if (test_pages_isolated(outer_start, end, false)) {
6888                 pr_info("%s: [%lx, %lx) PFNs busy\n",
6889                         __func__, outer_start, end);
6890                 ret = -EBUSY;
6891                 goto done;
6892         }
6893 
6894         /* Grab isolated pages from freelists. */
6895         outer_end = isolate_freepages_range(&cc, outer_start, end);
6896         if (!outer_end) {
6897                 ret = -EBUSY;
6898                 goto done;
6899         }
6900 
6901         /* Free head and tail (if any) */
6902         if (start != outer_start)
6903                 free_contig_range(outer_start, start - outer_start);
6904         if (end != outer_end)
6905                 free_contig_range(end, outer_end - end);
6906 
6907 done:
6908         undo_isolate_page_range(pfn_max_align_down(start),
6909                                 pfn_max_align_up(end), migratetype);
6910         return ret;
6911 }
6912 
6913 void free_contig_range(unsigned long pfn, unsigned nr_pages)
6914 {
6915         unsigned int count = 0;
6916 
6917         for (; nr_pages--; pfn++) {
6918                 struct page *page = pfn_to_page(pfn);
6919 
6920                 count += page_count(page) != 1;
6921                 __free_page(page);
6922         }
6923         WARN(count != 0, "%d pages are still in use!\n", count);
6924 }
6925 #endif
6926 
6927 #ifdef CONFIG_MEMORY_HOTPLUG
6928 /*
6929  * The zone indicated has a new number of managed_pages; batch sizes and percpu
6930  * page high values need to be recalulated.
6931  */
6932 void __meminit zone_pcp_update(struct zone *zone)
6933 {
6934         unsigned cpu;
6935         mutex_lock(&pcp_batch_high_lock);
6936         for_each_possible_cpu(cpu)
6937                 pageset_set_high_and_batch(zone,
6938                                 per_cpu_ptr(zone->pageset, cpu));
6939         mutex_unlock(&pcp_batch_high_lock);
6940 }
6941 #endif
6942 
6943 void zone_pcp_reset(struct zone *zone)
6944 {
6945         unsigned long flags;
6946         int cpu;
6947         struct per_cpu_pageset *pset;
6948 
6949         /* avoid races with drain_pages()  */
6950         local_irq_save(flags);
6951         if (zone->pageset != &boot_pageset) {
6952                 for_each_online_cpu(cpu) {
6953                         pset = per_cpu_ptr(zone->pageset, cpu);
6954                         drain_zonestat(zone, pset);
6955                 }
6956                 free_percpu(zone->pageset);
6957                 zone->pageset = &boot_pageset;
6958         }
6959         local_irq_restore(flags);
6960 }
6961 
6962 #ifdef CONFIG_MEMORY_HOTREMOVE
6963 /*
6964  * All pages in the range must be isolated before calling this.
6965  */
6966 void
6967 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6968 {
6969         struct page *page;
6970         struct zone *zone;
6971         unsigned int order, i;
6972         unsigned long pfn;
6973         unsigned long flags;
6974         /* find the first valid pfn */
6975         for (pfn = start_pfn; pfn < end_pfn; pfn++)
6976                 if (pfn_valid(pfn))
6977                         break;
6978         if (pfn == end_pfn)
6979                 return;
6980         zone = page_zone(pfn_to_page(pfn));
6981         spin_lock_irqsave(&zone->lock, flags);
6982         pfn = start_pfn;
6983         while (pfn < end_pfn) {
6984                 if (!pfn_valid(pfn)) {
6985                         pfn++;
6986                         continue;
6987                 }
6988                 page = pfn_to_page(pfn);
6989                 /*
6990                  * The HWPoisoned page may be not in buddy system, and
6991                  * page_count() is not 0.
6992                  */
6993                 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6994                         pfn++;
6995                         SetPageReserved(page);
6996                         continue;
6997                 }
6998 
6999                 BUG_ON(page_count(page));
7000                 BUG_ON(!PageBuddy(page));
7001                 order = page_order(page);
7002 #ifdef CONFIG_DEBUG_VM
7003                 printk(KERN_INFO "remove from free list %lx %d %lx\n",
7004                        pfn, 1 << order, end_pfn);
7005 #endif
7006                 list_del(&page->lru);
7007                 rmv_page_order(page);
7008                 zone->free_area[order].nr_free--;
7009                 for (i = 0; i < (1 << order); i++)
7010                         SetPageReserved((page+i));
7011                 pfn += (1 << order);
7012         }
7013         spin_unlock_irqrestore(&zone->lock, flags);
7014 }
7015 #endif
7016 
7017 #ifdef CONFIG_MEMORY_FAILURE
7018 bool is_free_buddy_page(struct page *page)
7019 {
7020         struct zone *zone = page_zone(page);
7021         unsigned long pfn = page_to_pfn(page);
7022         unsigned long flags;
7023         unsigned int order;
7024 
7025         spin_lock_irqsave(&zone->lock, flags);
7026         for (order = 0; order < MAX_ORDER; order++) {
7027                 struct page *page_head = page - (pfn & ((1 << order) - 1));
7028 
7029                 if (PageBuddy(page_head) && page_order(page_head) >= order)
7030                         break;
7031         }
7032         spin_unlock_irqrestore(&zone->lock, flags);
7033 
7034         return order < MAX_ORDER;
7035 }
7036 #endif
7037 

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