Version:  2.0.40 2.2.26 2.4.37 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 4.3 4.4 4.5 4.6

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/memremap.h>
 47 #include <linux/stop_machine.h>
 48 #include <linux/sort.h>
 49 #include <linux/pfn.h>
 50 #include <linux/backing-dev.h>
 51 #include <linux/fault-inject.h>
 52 #include <linux/page-isolation.h>
 53 #include <linux/page_ext.h>
 54 #include <linux/debugobjects.h>
 55 #include <linux/kmemleak.h>
 56 #include <linux/compaction.h>
 57 #include <trace/events/kmem.h>
 58 #include <linux/prefetch.h>
 59 #include <linux/mm_inline.h>
 60 #include <linux/migrate.h>
 61 #include <linux/page_ext.h>
 62 #include <linux/hugetlb.h>
 63 #include <linux/sched/rt.h>
 64 #include <linux/page_owner.h>
 65 #include <linux/kthread.h>
 66 
 67 #include <asm/sections.h>
 68 #include <asm/tlbflush.h>
 69 #include <asm/div64.h>
 70 #include "internal.h"
 71 
 72 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
 73 static DEFINE_MUTEX(pcp_batch_high_lock);
 74 #define MIN_PERCPU_PAGELIST_FRACTION    (8)
 75 
 76 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
 77 DEFINE_PER_CPU(int, numa_node);
 78 EXPORT_PER_CPU_SYMBOL(numa_node);
 79 #endif
 80 
 81 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
 82 /*
 83  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
 84  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
 85  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
 86  * defined in <linux/topology.h>.
 87  */
 88 DEFINE_PER_CPU(int, _numa_mem_);                /* Kernel "local memory" node */
 89 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
 90 int _node_numa_mem_[MAX_NUMNODES];
 91 #endif
 92 
 93 /*
 94  * Array of node states.
 95  */
 96 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
 97         [N_POSSIBLE] = NODE_MASK_ALL,
 98         [N_ONLINE] = { { [0] = 1UL } },
 99 #ifndef CONFIG_NUMA
100         [N_NORMAL_MEMORY] = { { [0] = 1UL } },
101 #ifdef CONFIG_HIGHMEM
102         [N_HIGH_MEMORY] = { { [0] = 1UL } },
103 #endif
104 #ifdef CONFIG_MOVABLE_NODE
105         [N_MEMORY] = { { [0] = 1UL } },
106 #endif
107         [N_CPU] = { { [0] = 1UL } },
108 #endif  /* NUMA */
109 };
110 EXPORT_SYMBOL(node_states);
111 
112 /* Protect totalram_pages and zone->managed_pages */
113 static DEFINE_SPINLOCK(managed_page_count_lock);
114 
115 unsigned long totalram_pages __read_mostly;
116 unsigned long totalreserve_pages __read_mostly;
117 unsigned long totalcma_pages __read_mostly;
118 
119 int percpu_pagelist_fraction;
120 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
121 
122 /*
123  * A cached value of the page's pageblock's migratetype, used when the page is
124  * put on a pcplist. Used to avoid the pageblock migratetype lookup when
125  * freeing from pcplists in most cases, at the cost of possibly becoming stale.
126  * Also the migratetype set in the page does not necessarily match the pcplist
127  * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
128  * other index - this ensures that it will be put on the correct CMA freelist.
129  */
130 static inline int get_pcppage_migratetype(struct page *page)
131 {
132         return page->index;
133 }
134 
135 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
136 {
137         page->index = migratetype;
138 }
139 
140 #ifdef CONFIG_PM_SLEEP
141 /*
142  * The following functions are used by the suspend/hibernate code to temporarily
143  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
144  * while devices are suspended.  To avoid races with the suspend/hibernate code,
145  * they should always be called with pm_mutex held (gfp_allowed_mask also should
146  * only be modified with pm_mutex held, unless the suspend/hibernate code is
147  * guaranteed not to run in parallel with that modification).
148  */
149 
150 static gfp_t saved_gfp_mask;
151 
152 void pm_restore_gfp_mask(void)
153 {
154         WARN_ON(!mutex_is_locked(&pm_mutex));
155         if (saved_gfp_mask) {
156                 gfp_allowed_mask = saved_gfp_mask;
157                 saved_gfp_mask = 0;
158         }
159 }
160 
161 void pm_restrict_gfp_mask(void)
162 {
163         WARN_ON(!mutex_is_locked(&pm_mutex));
164         WARN_ON(saved_gfp_mask);
165         saved_gfp_mask = gfp_allowed_mask;
166         gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
167 }
168 
169 bool pm_suspended_storage(void)
170 {
171         if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
172                 return false;
173         return true;
174 }
175 #endif /* CONFIG_PM_SLEEP */
176 
177 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
178 unsigned int pageblock_order __read_mostly;
179 #endif
180 
181 static void __free_pages_ok(struct page *page, unsigned int order);
182 
183 /*
184  * results with 256, 32 in the lowmem_reserve sysctl:
185  *      1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
186  *      1G machine -> (16M dma, 784M normal, 224M high)
187  *      NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
188  *      HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
189  *      HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
190  *
191  * TBD: should special case ZONE_DMA32 machines here - in those we normally
192  * don't need any ZONE_NORMAL reservation
193  */
194 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
195 #ifdef CONFIG_ZONE_DMA
196          256,
197 #endif
198 #ifdef CONFIG_ZONE_DMA32
199          256,
200 #endif
201 #ifdef CONFIG_HIGHMEM
202          32,
203 #endif
204          32,
205 };
206 
207 EXPORT_SYMBOL(totalram_pages);
208 
209 static char * const zone_names[MAX_NR_ZONES] = {
210 #ifdef CONFIG_ZONE_DMA
211          "DMA",
212 #endif
213 #ifdef CONFIG_ZONE_DMA32
214          "DMA32",
215 #endif
216          "Normal",
217 #ifdef CONFIG_HIGHMEM
218          "HighMem",
219 #endif
220          "Movable",
221 #ifdef CONFIG_ZONE_DEVICE
222          "Device",
223 #endif
224 };
225 
226 char * const migratetype_names[MIGRATE_TYPES] = {
227         "Unmovable",
228         "Movable",
229         "Reclaimable",
230         "HighAtomic",
231 #ifdef CONFIG_CMA
232         "CMA",
233 #endif
234 #ifdef CONFIG_MEMORY_ISOLATION
235         "Isolate",
236 #endif
237 };
238 
239 compound_page_dtor * const compound_page_dtors[] = {
240         NULL,
241         free_compound_page,
242 #ifdef CONFIG_HUGETLB_PAGE
243         free_huge_page,
244 #endif
245 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
246         free_transhuge_page,
247 #endif
248 };
249 
250 int min_free_kbytes = 1024;
251 int user_min_free_kbytes = -1;
252 int watermark_scale_factor = 10;
253 
254 static unsigned long __meminitdata nr_kernel_pages;
255 static unsigned long __meminitdata nr_all_pages;
256 static unsigned long __meminitdata dma_reserve;
257 
258 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
259 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
260 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
261 static unsigned long __initdata required_kernelcore;
262 static unsigned long __initdata required_movablecore;
263 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
264 static bool mirrored_kernelcore;
265 
266 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
267 int movable_zone;
268 EXPORT_SYMBOL(movable_zone);
269 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
270 
271 #if MAX_NUMNODES > 1
272 int nr_node_ids __read_mostly = MAX_NUMNODES;
273 int nr_online_nodes __read_mostly = 1;
274 EXPORT_SYMBOL(nr_node_ids);
275 EXPORT_SYMBOL(nr_online_nodes);
276 #endif
277 
278 int page_group_by_mobility_disabled __read_mostly;
279 
280 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
281 static inline void reset_deferred_meminit(pg_data_t *pgdat)
282 {
283         pgdat->first_deferred_pfn = ULONG_MAX;
284 }
285 
286 /* Returns true if the struct page for the pfn is uninitialised */
287 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
288 {
289         if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
290                 return true;
291 
292         return false;
293 }
294 
295 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
296 {
297         if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
298                 return true;
299 
300         return false;
301 }
302 
303 /*
304  * Returns false when the remaining initialisation should be deferred until
305  * later in the boot cycle when it can be parallelised.
306  */
307 static inline bool update_defer_init(pg_data_t *pgdat,
308                                 unsigned long pfn, unsigned long zone_end,
309                                 unsigned long *nr_initialised)
310 {
311         unsigned long max_initialise;
312 
313         /* Always populate low zones for address-contrained allocations */
314         if (zone_end < pgdat_end_pfn(pgdat))
315                 return true;
316         /*
317          * Initialise at least 2G of a node but also take into account that
318          * two large system hashes that can take up 1GB for 0.25TB/node.
319          */
320         max_initialise = max(2UL << (30 - PAGE_SHIFT),
321                 (pgdat->node_spanned_pages >> 8));
322 
323         (*nr_initialised)++;
324         if ((*nr_initialised > max_initialise) &&
325             (pfn & (PAGES_PER_SECTION - 1)) == 0) {
326                 pgdat->first_deferred_pfn = pfn;
327                 return false;
328         }
329 
330         return true;
331 }
332 #else
333 static inline void reset_deferred_meminit(pg_data_t *pgdat)
334 {
335 }
336 
337 static inline bool early_page_uninitialised(unsigned long pfn)
338 {
339         return false;
340 }
341 
342 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
343 {
344         return false;
345 }
346 
347 static inline bool update_defer_init(pg_data_t *pgdat,
348                                 unsigned long pfn, unsigned long zone_end,
349                                 unsigned long *nr_initialised)
350 {
351         return true;
352 }
353 #endif
354 
355 
356 void set_pageblock_migratetype(struct page *page, int migratetype)
357 {
358         if (unlikely(page_group_by_mobility_disabled &&
359                      migratetype < MIGRATE_PCPTYPES))
360                 migratetype = MIGRATE_UNMOVABLE;
361 
362         set_pageblock_flags_group(page, (unsigned long)migratetype,
363                                         PB_migrate, PB_migrate_end);
364 }
365 
366 #ifdef CONFIG_DEBUG_VM
367 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
368 {
369         int ret = 0;
370         unsigned seq;
371         unsigned long pfn = page_to_pfn(page);
372         unsigned long sp, start_pfn;
373 
374         do {
375                 seq = zone_span_seqbegin(zone);
376                 start_pfn = zone->zone_start_pfn;
377                 sp = zone->spanned_pages;
378                 if (!zone_spans_pfn(zone, pfn))
379                         ret = 1;
380         } while (zone_span_seqretry(zone, seq));
381 
382         if (ret)
383                 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
384                         pfn, zone_to_nid(zone), zone->name,
385                         start_pfn, start_pfn + sp);
386 
387         return ret;
388 }
389 
390 static int page_is_consistent(struct zone *zone, struct page *page)
391 {
392         if (!pfn_valid_within(page_to_pfn(page)))
393                 return 0;
394         if (zone != page_zone(page))
395                 return 0;
396 
397         return 1;
398 }
399 /*
400  * Temporary debugging check for pages not lying within a given zone.
401  */
402 static int bad_range(struct zone *zone, struct page *page)
403 {
404         if (page_outside_zone_boundaries(zone, page))
405                 return 1;
406         if (!page_is_consistent(zone, page))
407                 return 1;
408 
409         return 0;
410 }
411 #else
412 static inline int bad_range(struct zone *zone, struct page *page)
413 {
414         return 0;
415 }
416 #endif
417 
418 static void bad_page(struct page *page, const char *reason,
419                 unsigned long bad_flags)
420 {
421         static unsigned long resume;
422         static unsigned long nr_shown;
423         static unsigned long nr_unshown;
424 
425         /* Don't complain about poisoned pages */
426         if (PageHWPoison(page)) {
427                 page_mapcount_reset(page); /* remove PageBuddy */
428                 return;
429         }
430 
431         /*
432          * Allow a burst of 60 reports, then keep quiet for that minute;
433          * or allow a steady drip of one report per second.
434          */
435         if (nr_shown == 60) {
436                 if (time_before(jiffies, resume)) {
437                         nr_unshown++;
438                         goto out;
439                 }
440                 if (nr_unshown) {
441                         pr_alert(
442                               "BUG: Bad page state: %lu messages suppressed\n",
443                                 nr_unshown);
444                         nr_unshown = 0;
445                 }
446                 nr_shown = 0;
447         }
448         if (nr_shown++ == 0)
449                 resume = jiffies + 60 * HZ;
450 
451         pr_alert("BUG: Bad page state in process %s  pfn:%05lx\n",
452                 current->comm, page_to_pfn(page));
453         __dump_page(page, reason);
454         bad_flags &= page->flags;
455         if (bad_flags)
456                 pr_alert("bad because of flags: %#lx(%pGp)\n",
457                                                 bad_flags, &bad_flags);
458         dump_page_owner(page);
459 
460         print_modules();
461         dump_stack();
462 out:
463         /* Leave bad fields for debug, except PageBuddy could make trouble */
464         page_mapcount_reset(page); /* remove PageBuddy */
465         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
466 }
467 
468 /*
469  * Higher-order pages are called "compound pages".  They are structured thusly:
470  *
471  * The first PAGE_SIZE page is called the "head page" and have PG_head set.
472  *
473  * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
474  * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
475  *
476  * The first tail page's ->compound_dtor holds the offset in array of compound
477  * page destructors. See compound_page_dtors.
478  *
479  * The first tail page's ->compound_order holds the order of allocation.
480  * This usage means that zero-order pages may not be compound.
481  */
482 
483 void free_compound_page(struct page *page)
484 {
485         __free_pages_ok(page, compound_order(page));
486 }
487 
488 void prep_compound_page(struct page *page, unsigned int order)
489 {
490         int i;
491         int nr_pages = 1 << order;
492 
493         set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
494         set_compound_order(page, order);
495         __SetPageHead(page);
496         for (i = 1; i < nr_pages; i++) {
497                 struct page *p = page + i;
498                 set_page_count(p, 0);
499                 p->mapping = TAIL_MAPPING;
500                 set_compound_head(p, page);
501         }
502         atomic_set(compound_mapcount_ptr(page), -1);
503 }
504 
505 #ifdef CONFIG_DEBUG_PAGEALLOC
506 unsigned int _debug_guardpage_minorder;
507 bool _debug_pagealloc_enabled __read_mostly
508                         = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
509 EXPORT_SYMBOL(_debug_pagealloc_enabled);
510 bool _debug_guardpage_enabled __read_mostly;
511 
512 static int __init early_debug_pagealloc(char *buf)
513 {
514         if (!buf)
515                 return -EINVAL;
516 
517         if (strcmp(buf, "on") == 0)
518                 _debug_pagealloc_enabled = true;
519 
520         if (strcmp(buf, "off") == 0)
521                 _debug_pagealloc_enabled = false;
522 
523         return 0;
524 }
525 early_param("debug_pagealloc", early_debug_pagealloc);
526 
527 static bool need_debug_guardpage(void)
528 {
529         /* If we don't use debug_pagealloc, we don't need guard page */
530         if (!debug_pagealloc_enabled())
531                 return false;
532 
533         return true;
534 }
535 
536 static void init_debug_guardpage(void)
537 {
538         if (!debug_pagealloc_enabled())
539                 return;
540 
541         _debug_guardpage_enabled = true;
542 }
543 
544 struct page_ext_operations debug_guardpage_ops = {
545         .need = need_debug_guardpage,
546         .init = init_debug_guardpage,
547 };
548 
549 static int __init debug_guardpage_minorder_setup(char *buf)
550 {
551         unsigned long res;
552 
553         if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
554                 pr_err("Bad debug_guardpage_minorder value\n");
555                 return 0;
556         }
557         _debug_guardpage_minorder = res;
558         pr_info("Setting debug_guardpage_minorder to %lu\n", res);
559         return 0;
560 }
561 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
562 
563 static inline void set_page_guard(struct zone *zone, struct page *page,
564                                 unsigned int order, int migratetype)
565 {
566         struct page_ext *page_ext;
567 
568         if (!debug_guardpage_enabled())
569                 return;
570 
571         page_ext = lookup_page_ext(page);
572         __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
573 
574         INIT_LIST_HEAD(&page->lru);
575         set_page_private(page, order);
576         /* Guard pages are not available for any usage */
577         __mod_zone_freepage_state(zone, -(1 << order), migratetype);
578 }
579 
580 static inline void clear_page_guard(struct zone *zone, struct page *page,
581                                 unsigned int order, int migratetype)
582 {
583         struct page_ext *page_ext;
584 
585         if (!debug_guardpage_enabled())
586                 return;
587 
588         page_ext = lookup_page_ext(page);
589         __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
590 
591         set_page_private(page, 0);
592         if (!is_migrate_isolate(migratetype))
593                 __mod_zone_freepage_state(zone, (1 << order), migratetype);
594 }
595 #else
596 struct page_ext_operations debug_guardpage_ops = { NULL, };
597 static inline void set_page_guard(struct zone *zone, struct page *page,
598                                 unsigned int order, int migratetype) {}
599 static inline void clear_page_guard(struct zone *zone, struct page *page,
600                                 unsigned int order, int migratetype) {}
601 #endif
602 
603 static inline void set_page_order(struct page *page, unsigned int order)
604 {
605         set_page_private(page, order);
606         __SetPageBuddy(page);
607 }
608 
609 static inline void rmv_page_order(struct page *page)
610 {
611         __ClearPageBuddy(page);
612         set_page_private(page, 0);
613 }
614 
615 /*
616  * This function checks whether a page is free && is the buddy
617  * we can do coalesce a page and its buddy if
618  * (a) the buddy is not in a hole &&
619  * (b) the buddy is in the buddy system &&
620  * (c) a page and its buddy have the same order &&
621  * (d) a page and its buddy are in the same zone.
622  *
623  * For recording whether a page is in the buddy system, we set ->_mapcount
624  * PAGE_BUDDY_MAPCOUNT_VALUE.
625  * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
626  * serialized by zone->lock.
627  *
628  * For recording page's order, we use page_private(page).
629  */
630 static inline int page_is_buddy(struct page *page, struct page *buddy,
631                                                         unsigned int order)
632 {
633         if (!pfn_valid_within(page_to_pfn(buddy)))
634                 return 0;
635 
636         if (page_is_guard(buddy) && page_order(buddy) == order) {
637                 if (page_zone_id(page) != page_zone_id(buddy))
638                         return 0;
639 
640                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
641 
642                 return 1;
643         }
644 
645         if (PageBuddy(buddy) && page_order(buddy) == order) {
646                 /*
647                  * zone check is done late to avoid uselessly
648                  * calculating zone/node ids for pages that could
649                  * never merge.
650                  */
651                 if (page_zone_id(page) != page_zone_id(buddy))
652                         return 0;
653 
654                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
655 
656                 return 1;
657         }
658         return 0;
659 }
660 
661 /*
662  * Freeing function for a buddy system allocator.
663  *
664  * The concept of a buddy system is to maintain direct-mapped table
665  * (containing bit values) for memory blocks of various "orders".
666  * The bottom level table contains the map for the smallest allocatable
667  * units of memory (here, pages), and each level above it describes
668  * pairs of units from the levels below, hence, "buddies".
669  * At a high level, all that happens here is marking the table entry
670  * at the bottom level available, and propagating the changes upward
671  * as necessary, plus some accounting needed to play nicely with other
672  * parts of the VM system.
673  * At each level, we keep a list of pages, which are heads of continuous
674  * free pages of length of (1 << order) and marked with _mapcount
675  * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
676  * field.
677  * So when we are allocating or freeing one, we can derive the state of the
678  * other.  That is, if we allocate a small block, and both were
679  * free, the remainder of the region must be split into blocks.
680  * If a block is freed, and its buddy is also free, then this
681  * triggers coalescing into a block of larger size.
682  *
683  * -- nyc
684  */
685 
686 static inline void __free_one_page(struct page *page,
687                 unsigned long pfn,
688                 struct zone *zone, unsigned int order,
689                 int migratetype)
690 {
691         unsigned long page_idx;
692         unsigned long combined_idx;
693         unsigned long uninitialized_var(buddy_idx);
694         struct page *buddy;
695         unsigned int max_order;
696 
697         max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
698 
699         VM_BUG_ON(!zone_is_initialized(zone));
700         VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
701 
702         VM_BUG_ON(migratetype == -1);
703         if (likely(!is_migrate_isolate(migratetype)))
704                 __mod_zone_freepage_state(zone, 1 << order, migratetype);
705 
706         page_idx = pfn & ((1 << MAX_ORDER) - 1);
707 
708         VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
709         VM_BUG_ON_PAGE(bad_range(zone, page), page);
710 
711 continue_merging:
712         while (order < max_order - 1) {
713                 buddy_idx = __find_buddy_index(page_idx, order);
714                 buddy = page + (buddy_idx - page_idx);
715                 if (!page_is_buddy(page, buddy, order))
716                         goto done_merging;
717                 /*
718                  * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
719                  * merge with it and move up one order.
720                  */
721                 if (page_is_guard(buddy)) {
722                         clear_page_guard(zone, buddy, order, migratetype);
723                 } else {
724                         list_del(&buddy->lru);
725                         zone->free_area[order].nr_free--;
726                         rmv_page_order(buddy);
727                 }
728                 combined_idx = buddy_idx & page_idx;
729                 page = page + (combined_idx - page_idx);
730                 page_idx = combined_idx;
731                 order++;
732         }
733         if (max_order < MAX_ORDER) {
734                 /* If we are here, it means order is >= pageblock_order.
735                  * We want to prevent merge between freepages on isolate
736                  * pageblock and normal pageblock. Without this, pageblock
737                  * isolation could cause incorrect freepage or CMA accounting.
738                  *
739                  * We don't want to hit this code for the more frequent
740                  * low-order merging.
741                  */
742                 if (unlikely(has_isolate_pageblock(zone))) {
743                         int buddy_mt;
744 
745                         buddy_idx = __find_buddy_index(page_idx, order);
746                         buddy = page + (buddy_idx - page_idx);
747                         buddy_mt = get_pageblock_migratetype(buddy);
748 
749                         if (migratetype != buddy_mt
750                                         && (is_migrate_isolate(migratetype) ||
751                                                 is_migrate_isolate(buddy_mt)))
752                                 goto done_merging;
753                 }
754                 max_order++;
755                 goto continue_merging;
756         }
757 
758 done_merging:
759         set_page_order(page, order);
760 
761         /*
762          * If this is not the largest possible page, check if the buddy
763          * of the next-highest order is free. If it is, it's possible
764          * that pages are being freed that will coalesce soon. In case,
765          * that is happening, add the free page to the tail of the list
766          * so it's less likely to be used soon and more likely to be merged
767          * as a higher order page
768          */
769         if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
770                 struct page *higher_page, *higher_buddy;
771                 combined_idx = buddy_idx & page_idx;
772                 higher_page = page + (combined_idx - page_idx);
773                 buddy_idx = __find_buddy_index(combined_idx, order + 1);
774                 higher_buddy = higher_page + (buddy_idx - combined_idx);
775                 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
776                         list_add_tail(&page->lru,
777                                 &zone->free_area[order].free_list[migratetype]);
778                         goto out;
779                 }
780         }
781 
782         list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
783 out:
784         zone->free_area[order].nr_free++;
785 }
786 
787 static inline int free_pages_check(struct page *page)
788 {
789         const char *bad_reason = NULL;
790         unsigned long bad_flags = 0;
791 
792         if (unlikely(atomic_read(&page->_mapcount) != -1))
793                 bad_reason = "nonzero mapcount";
794         if (unlikely(page->mapping != NULL))
795                 bad_reason = "non-NULL mapping";
796         if (unlikely(page_ref_count(page) != 0))
797                 bad_reason = "nonzero _count";
798         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
799                 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
800                 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
801         }
802 #ifdef CONFIG_MEMCG
803         if (unlikely(page->mem_cgroup))
804                 bad_reason = "page still charged to cgroup";
805 #endif
806         if (unlikely(bad_reason)) {
807                 bad_page(page, bad_reason, bad_flags);
808                 return 1;
809         }
810         page_cpupid_reset_last(page);
811         if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
812                 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
813         return 0;
814 }
815 
816 /*
817  * Frees a number of pages from the PCP lists
818  * Assumes all pages on list are in same zone, and of same order.
819  * count is the number of pages to free.
820  *
821  * If the zone was previously in an "all pages pinned" state then look to
822  * see if this freeing clears that state.
823  *
824  * And clear the zone's pages_scanned counter, to hold off the "all pages are
825  * pinned" detection logic.
826  */
827 static void free_pcppages_bulk(struct zone *zone, int count,
828                                         struct per_cpu_pages *pcp)
829 {
830         int migratetype = 0;
831         int batch_free = 0;
832         int to_free = count;
833         unsigned long nr_scanned;
834 
835         spin_lock(&zone->lock);
836         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
837         if (nr_scanned)
838                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
839 
840         while (to_free) {
841                 struct page *page;
842                 struct list_head *list;
843 
844                 /*
845                  * Remove pages from lists in a round-robin fashion. A
846                  * batch_free count is maintained that is incremented when an
847                  * empty list is encountered.  This is so more pages are freed
848                  * off fuller lists instead of spinning excessively around empty
849                  * lists
850                  */
851                 do {
852                         batch_free++;
853                         if (++migratetype == MIGRATE_PCPTYPES)
854                                 migratetype = 0;
855                         list = &pcp->lists[migratetype];
856                 } while (list_empty(list));
857 
858                 /* This is the only non-empty list. Free them all. */
859                 if (batch_free == MIGRATE_PCPTYPES)
860                         batch_free = to_free;
861 
862                 do {
863                         int mt; /* migratetype of the to-be-freed page */
864 
865                         page = list_last_entry(list, struct page, lru);
866                         /* must delete as __free_one_page list manipulates */
867                         list_del(&page->lru);
868 
869                         mt = get_pcppage_migratetype(page);
870                         /* MIGRATE_ISOLATE page should not go to pcplists */
871                         VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
872                         /* Pageblock could have been isolated meanwhile */
873                         if (unlikely(has_isolate_pageblock(zone)))
874                                 mt = get_pageblock_migratetype(page);
875 
876                         __free_one_page(page, page_to_pfn(page), zone, 0, mt);
877                         trace_mm_page_pcpu_drain(page, 0, mt);
878                 } while (--to_free && --batch_free && !list_empty(list));
879         }
880         spin_unlock(&zone->lock);
881 }
882 
883 static void free_one_page(struct zone *zone,
884                                 struct page *page, unsigned long pfn,
885                                 unsigned int order,
886                                 int migratetype)
887 {
888         unsigned long nr_scanned;
889         spin_lock(&zone->lock);
890         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
891         if (nr_scanned)
892                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
893 
894         if (unlikely(has_isolate_pageblock(zone) ||
895                 is_migrate_isolate(migratetype))) {
896                 migratetype = get_pfnblock_migratetype(page, pfn);
897         }
898         __free_one_page(page, pfn, zone, order, migratetype);
899         spin_unlock(&zone->lock);
900 }
901 
902 static int free_tail_pages_check(struct page *head_page, struct page *page)
903 {
904         int ret = 1;
905 
906         /*
907          * We rely page->lru.next never has bit 0 set, unless the page
908          * is PageTail(). Let's make sure that's true even for poisoned ->lru.
909          */
910         BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
911 
912         if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
913                 ret = 0;
914                 goto out;
915         }
916         switch (page - head_page) {
917         case 1:
918                 /* the first tail page: ->mapping is compound_mapcount() */
919                 if (unlikely(compound_mapcount(page))) {
920                         bad_page(page, "nonzero compound_mapcount", 0);
921                         goto out;
922                 }
923                 break;
924         case 2:
925                 /*
926                  * the second tail page: ->mapping is
927                  * page_deferred_list().next -- ignore value.
928                  */
929                 break;
930         default:
931                 if (page->mapping != TAIL_MAPPING) {
932                         bad_page(page, "corrupted mapping in tail page", 0);
933                         goto out;
934                 }
935                 break;
936         }
937         if (unlikely(!PageTail(page))) {
938                 bad_page(page, "PageTail not set", 0);
939                 goto out;
940         }
941         if (unlikely(compound_head(page) != head_page)) {
942                 bad_page(page, "compound_head not consistent", 0);
943                 goto out;
944         }
945         ret = 0;
946 out:
947         page->mapping = NULL;
948         clear_compound_head(page);
949         return ret;
950 }
951 
952 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
953                                 unsigned long zone, int nid)
954 {
955         set_page_links(page, zone, nid, pfn);
956         init_page_count(page);
957         page_mapcount_reset(page);
958         page_cpupid_reset_last(page);
959 
960         INIT_LIST_HEAD(&page->lru);
961 #ifdef WANT_PAGE_VIRTUAL
962         /* The shift won't overflow because ZONE_NORMAL is below 4G. */
963         if (!is_highmem_idx(zone))
964                 set_page_address(page, __va(pfn << PAGE_SHIFT));
965 #endif
966 }
967 
968 static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
969                                         int nid)
970 {
971         return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
972 }
973 
974 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
975 static void init_reserved_page(unsigned long pfn)
976 {
977         pg_data_t *pgdat;
978         int nid, zid;
979 
980         if (!early_page_uninitialised(pfn))
981                 return;
982 
983         nid = early_pfn_to_nid(pfn);
984         pgdat = NODE_DATA(nid);
985 
986         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
987                 struct zone *zone = &pgdat->node_zones[zid];
988 
989                 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
990                         break;
991         }
992         __init_single_pfn(pfn, zid, nid);
993 }
994 #else
995 static inline void init_reserved_page(unsigned long pfn)
996 {
997 }
998 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
999 
1000 /*
1001  * Initialised pages do not have PageReserved set. This function is
1002  * called for each range allocated by the bootmem allocator and
1003  * marks the pages PageReserved. The remaining valid pages are later
1004  * sent to the buddy page allocator.
1005  */
1006 void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
1007 {
1008         unsigned long start_pfn = PFN_DOWN(start);
1009         unsigned long end_pfn = PFN_UP(end);
1010 
1011         for (; start_pfn < end_pfn; start_pfn++) {
1012                 if (pfn_valid(start_pfn)) {
1013                         struct page *page = pfn_to_page(start_pfn);
1014 
1015                         init_reserved_page(start_pfn);
1016 
1017                         /* Avoid false-positive PageTail() */
1018                         INIT_LIST_HEAD(&page->lru);
1019 
1020                         SetPageReserved(page);
1021                 }
1022         }
1023 }
1024 
1025 static bool free_pages_prepare(struct page *page, unsigned int order)
1026 {
1027         bool compound = PageCompound(page);
1028         int i, bad = 0;
1029 
1030         VM_BUG_ON_PAGE(PageTail(page), page);
1031         VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1032 
1033         trace_mm_page_free(page, order);
1034         kmemcheck_free_shadow(page, order);
1035         kasan_free_pages(page, order);
1036 
1037         if (PageAnon(page))
1038                 page->mapping = NULL;
1039         bad += free_pages_check(page);
1040         for (i = 1; i < (1 << order); i++) {
1041                 if (compound)
1042                         bad += free_tail_pages_check(page, page + i);
1043                 bad += free_pages_check(page + i);
1044         }
1045         if (bad)
1046                 return false;
1047 
1048         reset_page_owner(page, order);
1049 
1050         if (!PageHighMem(page)) {
1051                 debug_check_no_locks_freed(page_address(page),
1052                                            PAGE_SIZE << order);
1053                 debug_check_no_obj_freed(page_address(page),
1054                                            PAGE_SIZE << order);
1055         }
1056         arch_free_page(page, order);
1057         kernel_poison_pages(page, 1 << order, 0);
1058         kernel_map_pages(page, 1 << order, 0);
1059 
1060         return true;
1061 }
1062 
1063 static void __free_pages_ok(struct page *page, unsigned int order)
1064 {
1065         unsigned long flags;
1066         int migratetype;
1067         unsigned long pfn = page_to_pfn(page);
1068 
1069         if (!free_pages_prepare(page, order))
1070                 return;
1071 
1072         migratetype = get_pfnblock_migratetype(page, pfn);
1073         local_irq_save(flags);
1074         __count_vm_events(PGFREE, 1 << order);
1075         free_one_page(page_zone(page), page, pfn, order, migratetype);
1076         local_irq_restore(flags);
1077 }
1078 
1079 static void __init __free_pages_boot_core(struct page *page,
1080                                         unsigned long pfn, unsigned int order)
1081 {
1082         unsigned int nr_pages = 1 << order;
1083         struct page *p = page;
1084         unsigned int loop;
1085 
1086         prefetchw(p);
1087         for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1088                 prefetchw(p + 1);
1089                 __ClearPageReserved(p);
1090                 set_page_count(p, 0);
1091         }
1092         __ClearPageReserved(p);
1093         set_page_count(p, 0);
1094 
1095         page_zone(page)->managed_pages += nr_pages;
1096         set_page_refcounted(page);
1097         __free_pages(page, order);
1098 }
1099 
1100 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1101         defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1102 
1103 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1104 
1105 int __meminit early_pfn_to_nid(unsigned long pfn)
1106 {
1107         static DEFINE_SPINLOCK(early_pfn_lock);
1108         int nid;
1109 
1110         spin_lock(&early_pfn_lock);
1111         nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1112         if (nid < 0)
1113                 nid = 0;
1114         spin_unlock(&early_pfn_lock);
1115 
1116         return nid;
1117 }
1118 #endif
1119 
1120 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1121 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1122                                         struct mminit_pfnnid_cache *state)
1123 {
1124         int nid;
1125 
1126         nid = __early_pfn_to_nid(pfn, state);
1127         if (nid >= 0 && nid != node)
1128                 return false;
1129         return true;
1130 }
1131 
1132 /* Only safe to use early in boot when initialisation is single-threaded */
1133 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1134 {
1135         return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1136 }
1137 
1138 #else
1139 
1140 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1141 {
1142         return true;
1143 }
1144 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1145                                         struct mminit_pfnnid_cache *state)
1146 {
1147         return true;
1148 }
1149 #endif
1150 
1151 
1152 void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
1153                                                         unsigned int order)
1154 {
1155         if (early_page_uninitialised(pfn))
1156                 return;
1157         return __free_pages_boot_core(page, pfn, order);
1158 }
1159 
1160 /*
1161  * Check that the whole (or subset of) a pageblock given by the interval of
1162  * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1163  * with the migration of free compaction scanner. The scanners then need to
1164  * use only pfn_valid_within() check for arches that allow holes within
1165  * pageblocks.
1166  *
1167  * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1168  *
1169  * It's possible on some configurations to have a setup like node0 node1 node0
1170  * i.e. it's possible that all pages within a zones range of pages do not
1171  * belong to a single zone. We assume that a border between node0 and node1
1172  * can occur within a single pageblock, but not a node0 node1 node0
1173  * interleaving within a single pageblock. It is therefore sufficient to check
1174  * the first and last page of a pageblock and avoid checking each individual
1175  * page in a pageblock.
1176  */
1177 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1178                                      unsigned long end_pfn, struct zone *zone)
1179 {
1180         struct page *start_page;
1181         struct page *end_page;
1182 
1183         /* end_pfn is one past the range we are checking */
1184         end_pfn--;
1185 
1186         if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1187                 return NULL;
1188 
1189         start_page = pfn_to_page(start_pfn);
1190 
1191         if (page_zone(start_page) != zone)
1192                 return NULL;
1193 
1194         end_page = pfn_to_page(end_pfn);
1195 
1196         /* This gives a shorter code than deriving page_zone(end_page) */
1197         if (page_zone_id(start_page) != page_zone_id(end_page))
1198                 return NULL;
1199 
1200         return start_page;
1201 }
1202 
1203 void set_zone_contiguous(struct zone *zone)
1204 {
1205         unsigned long block_start_pfn = zone->zone_start_pfn;
1206         unsigned long block_end_pfn;
1207 
1208         block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1209         for (; block_start_pfn < zone_end_pfn(zone);
1210                         block_start_pfn = block_end_pfn,
1211                          block_end_pfn += pageblock_nr_pages) {
1212 
1213                 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1214 
1215                 if (!__pageblock_pfn_to_page(block_start_pfn,
1216                                              block_end_pfn, zone))
1217                         return;
1218         }
1219 
1220         /* We confirm that there is no hole */
1221         zone->contiguous = true;
1222 }
1223 
1224 void clear_zone_contiguous(struct zone *zone)
1225 {
1226         zone->contiguous = false;
1227 }
1228 
1229 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1230 static void __init deferred_free_range(struct page *page,
1231                                         unsigned long pfn, int nr_pages)
1232 {
1233         int i;
1234 
1235         if (!page)
1236                 return;
1237 
1238         /* Free a large naturally-aligned chunk if possible */
1239         if (nr_pages == MAX_ORDER_NR_PAGES &&
1240             (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
1241                 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1242                 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1243                 return;
1244         }
1245 
1246         for (i = 0; i < nr_pages; i++, page++, pfn++)
1247                 __free_pages_boot_core(page, pfn, 0);
1248 }
1249 
1250 /* Completion tracking for deferred_init_memmap() threads */
1251 static atomic_t pgdat_init_n_undone __initdata;
1252 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1253 
1254 static inline void __init pgdat_init_report_one_done(void)
1255 {
1256         if (atomic_dec_and_test(&pgdat_init_n_undone))
1257                 complete(&pgdat_init_all_done_comp);
1258 }
1259 
1260 /* Initialise remaining memory on a node */
1261 static int __init deferred_init_memmap(void *data)
1262 {
1263         pg_data_t *pgdat = data;
1264         int nid = pgdat->node_id;
1265         struct mminit_pfnnid_cache nid_init_state = { };
1266         unsigned long start = jiffies;
1267         unsigned long nr_pages = 0;
1268         unsigned long walk_start, walk_end;
1269         int i, zid;
1270         struct zone *zone;
1271         unsigned long first_init_pfn = pgdat->first_deferred_pfn;
1272         const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1273 
1274         if (first_init_pfn == ULONG_MAX) {
1275                 pgdat_init_report_one_done();
1276                 return 0;
1277         }
1278 
1279         /* Bind memory initialisation thread to a local node if possible */
1280         if (!cpumask_empty(cpumask))
1281                 set_cpus_allowed_ptr(current, cpumask);
1282 
1283         /* Sanity check boundaries */
1284         BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1285         BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1286         pgdat->first_deferred_pfn = ULONG_MAX;
1287 
1288         /* Only the highest zone is deferred so find it */
1289         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1290                 zone = pgdat->node_zones + zid;
1291                 if (first_init_pfn < zone_end_pfn(zone))
1292                         break;
1293         }
1294 
1295         for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1296                 unsigned long pfn, end_pfn;
1297                 struct page *page = NULL;
1298                 struct page *free_base_page = NULL;
1299                 unsigned long free_base_pfn = 0;
1300                 int nr_to_free = 0;
1301 
1302                 end_pfn = min(walk_end, zone_end_pfn(zone));
1303                 pfn = first_init_pfn;
1304                 if (pfn < walk_start)
1305                         pfn = walk_start;
1306                 if (pfn < zone->zone_start_pfn)
1307                         pfn = zone->zone_start_pfn;
1308 
1309                 for (; pfn < end_pfn; pfn++) {
1310                         if (!pfn_valid_within(pfn))
1311                                 goto free_range;
1312 
1313                         /*
1314                          * Ensure pfn_valid is checked every
1315                          * MAX_ORDER_NR_PAGES for memory holes
1316                          */
1317                         if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1318                                 if (!pfn_valid(pfn)) {
1319                                         page = NULL;
1320                                         goto free_range;
1321                                 }
1322                         }
1323 
1324                         if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1325                                 page = NULL;
1326                                 goto free_range;
1327                         }
1328 
1329                         /* Minimise pfn page lookups and scheduler checks */
1330                         if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1331                                 page++;
1332                         } else {
1333                                 nr_pages += nr_to_free;
1334                                 deferred_free_range(free_base_page,
1335                                                 free_base_pfn, nr_to_free);
1336                                 free_base_page = NULL;
1337                                 free_base_pfn = nr_to_free = 0;
1338 
1339                                 page = pfn_to_page(pfn);
1340                                 cond_resched();
1341                         }
1342 
1343                         if (page->flags) {
1344                                 VM_BUG_ON(page_zone(page) != zone);
1345                                 goto free_range;
1346                         }
1347 
1348                         __init_single_page(page, pfn, zid, nid);
1349                         if (!free_base_page) {
1350                                 free_base_page = page;
1351                                 free_base_pfn = pfn;
1352                                 nr_to_free = 0;
1353                         }
1354                         nr_to_free++;
1355 
1356                         /* Where possible, batch up pages for a single free */
1357                         continue;
1358 free_range:
1359                         /* Free the current block of pages to allocator */
1360                         nr_pages += nr_to_free;
1361                         deferred_free_range(free_base_page, free_base_pfn,
1362                                                                 nr_to_free);
1363                         free_base_page = NULL;
1364                         free_base_pfn = nr_to_free = 0;
1365                 }
1366 
1367                 first_init_pfn = max(end_pfn, first_init_pfn);
1368         }
1369 
1370         /* Sanity check that the next zone really is unpopulated */
1371         WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1372 
1373         pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1374                                         jiffies_to_msecs(jiffies - start));
1375 
1376         pgdat_init_report_one_done();
1377         return 0;
1378 }
1379 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1380 
1381 void __init page_alloc_init_late(void)
1382 {
1383         struct zone *zone;
1384 
1385 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1386         int nid;
1387 
1388         /* There will be num_node_state(N_MEMORY) threads */
1389         atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1390         for_each_node_state(nid, N_MEMORY) {
1391                 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1392         }
1393 
1394         /* Block until all are initialised */
1395         wait_for_completion(&pgdat_init_all_done_comp);
1396 
1397         /* Reinit limits that are based on free pages after the kernel is up */
1398         files_maxfiles_init();
1399 #endif
1400 
1401         for_each_populated_zone(zone)
1402                 set_zone_contiguous(zone);
1403 }
1404 
1405 #ifdef CONFIG_CMA
1406 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1407 void __init init_cma_reserved_pageblock(struct page *page)
1408 {
1409         unsigned i = pageblock_nr_pages;
1410         struct page *p = page;
1411 
1412         do {
1413                 __ClearPageReserved(p);
1414                 set_page_count(p, 0);
1415         } while (++p, --i);
1416 
1417         set_pageblock_migratetype(page, MIGRATE_CMA);
1418 
1419         if (pageblock_order >= MAX_ORDER) {
1420                 i = pageblock_nr_pages;
1421                 p = page;
1422                 do {
1423                         set_page_refcounted(p);
1424                         __free_pages(p, MAX_ORDER - 1);
1425                         p += MAX_ORDER_NR_PAGES;
1426                 } while (i -= MAX_ORDER_NR_PAGES);
1427         } else {
1428                 set_page_refcounted(page);
1429                 __free_pages(page, pageblock_order);
1430         }
1431 
1432         adjust_managed_page_count(page, pageblock_nr_pages);
1433 }
1434 #endif
1435 
1436 /*
1437  * The order of subdivision here is critical for the IO subsystem.
1438  * Please do not alter this order without good reasons and regression
1439  * testing. Specifically, as large blocks of memory are subdivided,
1440  * the order in which smaller blocks are delivered depends on the order
1441  * they're subdivided in this function. This is the primary factor
1442  * influencing the order in which pages are delivered to the IO
1443  * subsystem according to empirical testing, and this is also justified
1444  * by considering the behavior of a buddy system containing a single
1445  * large block of memory acted on by a series of small allocations.
1446  * This behavior is a critical factor in sglist merging's success.
1447  *
1448  * -- nyc
1449  */
1450 static inline void expand(struct zone *zone, struct page *page,
1451         int low, int high, struct free_area *area,
1452         int migratetype)
1453 {
1454         unsigned long size = 1 << high;
1455 
1456         while (high > low) {
1457                 area--;
1458                 high--;
1459                 size >>= 1;
1460                 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1461 
1462                 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
1463                         debug_guardpage_enabled() &&
1464                         high < debug_guardpage_minorder()) {
1465                         /*
1466                          * Mark as guard pages (or page), that will allow to
1467                          * merge back to allocator when buddy will be freed.
1468                          * Corresponding page table entries will not be touched,
1469                          * pages will stay not present in virtual address space
1470                          */
1471                         set_page_guard(zone, &page[size], high, migratetype);
1472                         continue;
1473                 }
1474                 list_add(&page[size].lru, &area->free_list[migratetype]);
1475                 area->nr_free++;
1476                 set_page_order(&page[size], high);
1477         }
1478 }
1479 
1480 /*
1481  * This page is about to be returned from the page allocator
1482  */
1483 static inline int check_new_page(struct page *page)
1484 {
1485         const char *bad_reason = NULL;
1486         unsigned long bad_flags = 0;
1487 
1488         if (unlikely(atomic_read(&page->_mapcount) != -1))
1489                 bad_reason = "nonzero mapcount";
1490         if (unlikely(page->mapping != NULL))
1491                 bad_reason = "non-NULL mapping";
1492         if (unlikely(page_ref_count(page) != 0))
1493                 bad_reason = "nonzero _count";
1494         if (unlikely(page->flags & __PG_HWPOISON)) {
1495                 bad_reason = "HWPoisoned (hardware-corrupted)";
1496                 bad_flags = __PG_HWPOISON;
1497         }
1498         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1499                 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1500                 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1501         }
1502 #ifdef CONFIG_MEMCG
1503         if (unlikely(page->mem_cgroup))
1504                 bad_reason = "page still charged to cgroup";
1505 #endif
1506         if (unlikely(bad_reason)) {
1507                 bad_page(page, bad_reason, bad_flags);
1508                 return 1;
1509         }
1510         return 0;
1511 }
1512 
1513 static inline bool free_pages_prezeroed(bool poisoned)
1514 {
1515         return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1516                 page_poisoning_enabled() && poisoned;
1517 }
1518 
1519 static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1520                                                                 int alloc_flags)
1521 {
1522         int i;
1523         bool poisoned = true;
1524 
1525         for (i = 0; i < (1 << order); i++) {
1526                 struct page *p = page + i;
1527                 if (unlikely(check_new_page(p)))
1528                         return 1;
1529                 if (poisoned)
1530                         poisoned &= page_is_poisoned(p);
1531         }
1532 
1533         set_page_private(page, 0);
1534         set_page_refcounted(page);
1535 
1536         arch_alloc_page(page, order);
1537         kernel_map_pages(page, 1 << order, 1);
1538         kernel_poison_pages(page, 1 << order, 1);
1539         kasan_alloc_pages(page, order);
1540 
1541         if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
1542                 for (i = 0; i < (1 << order); i++)
1543                         clear_highpage(page + i);
1544 
1545         if (order && (gfp_flags & __GFP_COMP))
1546                 prep_compound_page(page, order);
1547 
1548         set_page_owner(page, order, gfp_flags);
1549 
1550         /*
1551          * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1552          * allocate the page. The expectation is that the caller is taking
1553          * steps that will free more memory. The caller should avoid the page
1554          * being used for !PFMEMALLOC purposes.
1555          */
1556         if (alloc_flags & ALLOC_NO_WATERMARKS)
1557                 set_page_pfmemalloc(page);
1558         else
1559                 clear_page_pfmemalloc(page);
1560 
1561         return 0;
1562 }
1563 
1564 /*
1565  * Go through the free lists for the given migratetype and remove
1566  * the smallest available page from the freelists
1567  */
1568 static inline
1569 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1570                                                 int migratetype)
1571 {
1572         unsigned int current_order;
1573         struct free_area *area;
1574         struct page *page;
1575 
1576         /* Find a page of the appropriate size in the preferred list */
1577         for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1578                 area = &(zone->free_area[current_order]);
1579                 page = list_first_entry_or_null(&area->free_list[migratetype],
1580                                                         struct page, lru);
1581                 if (!page)
1582                         continue;
1583                 list_del(&page->lru);
1584                 rmv_page_order(page);
1585                 area->nr_free--;
1586                 expand(zone, page, order, current_order, area, migratetype);
1587                 set_pcppage_migratetype(page, migratetype);
1588                 return page;
1589         }
1590 
1591         return NULL;
1592 }
1593 
1594 
1595 /*
1596  * This array describes the order lists are fallen back to when
1597  * the free lists for the desirable migrate type are depleted
1598  */
1599 static int fallbacks[MIGRATE_TYPES][4] = {
1600         [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_TYPES },
1601         [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_TYPES },
1602         [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
1603 #ifdef CONFIG_CMA
1604         [MIGRATE_CMA]         = { MIGRATE_TYPES }, /* Never used */
1605 #endif
1606 #ifdef CONFIG_MEMORY_ISOLATION
1607         [MIGRATE_ISOLATE]     = { MIGRATE_TYPES }, /* Never used */
1608 #endif
1609 };
1610 
1611 #ifdef CONFIG_CMA
1612 static struct page *__rmqueue_cma_fallback(struct zone *zone,
1613                                         unsigned int order)
1614 {
1615         return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1616 }
1617 #else
1618 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1619                                         unsigned int order) { return NULL; }
1620 #endif
1621 
1622 /*
1623  * Move the free pages in a range to the free lists of the requested type.
1624  * Note that start_page and end_pages are not aligned on a pageblock
1625  * boundary. If alignment is required, use move_freepages_block()
1626  */
1627 int move_freepages(struct zone *zone,
1628                           struct page *start_page, struct page *end_page,
1629                           int migratetype)
1630 {
1631         struct page *page;
1632         unsigned int order;
1633         int pages_moved = 0;
1634 
1635 #ifndef CONFIG_HOLES_IN_ZONE
1636         /*
1637          * page_zone is not safe to call in this context when
1638          * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1639          * anyway as we check zone boundaries in move_freepages_block().
1640          * Remove at a later date when no bug reports exist related to
1641          * grouping pages by mobility
1642          */
1643         VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1644 #endif
1645 
1646         for (page = start_page; page <= end_page;) {
1647                 /* Make sure we are not inadvertently changing nodes */
1648                 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1649 
1650                 if (!pfn_valid_within(page_to_pfn(page))) {
1651                         page++;
1652                         continue;
1653                 }
1654 
1655                 if (!PageBuddy(page)) {
1656                         page++;
1657                         continue;
1658                 }
1659 
1660                 order = page_order(page);
1661                 list_move(&page->lru,
1662                           &zone->free_area[order].free_list[migratetype]);
1663                 page += 1 << order;
1664                 pages_moved += 1 << order;
1665         }
1666 
1667         return pages_moved;
1668 }
1669 
1670 int move_freepages_block(struct zone *zone, struct page *page,
1671                                 int migratetype)
1672 {
1673         unsigned long start_pfn, end_pfn;
1674         struct page *start_page, *end_page;
1675 
1676         start_pfn = page_to_pfn(page);
1677         start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1678         start_page = pfn_to_page(start_pfn);
1679         end_page = start_page + pageblock_nr_pages - 1;
1680         end_pfn = start_pfn + pageblock_nr_pages - 1;
1681 
1682         /* Do not cross zone boundaries */
1683         if (!zone_spans_pfn(zone, start_pfn))
1684                 start_page = page;
1685         if (!zone_spans_pfn(zone, end_pfn))
1686                 return 0;
1687 
1688         return move_freepages(zone, start_page, end_page, migratetype);
1689 }
1690 
1691 static void change_pageblock_range(struct page *pageblock_page,
1692                                         int start_order, int migratetype)
1693 {
1694         int nr_pageblocks = 1 << (start_order - pageblock_order);
1695 
1696         while (nr_pageblocks--) {
1697                 set_pageblock_migratetype(pageblock_page, migratetype);
1698                 pageblock_page += pageblock_nr_pages;
1699         }
1700 }
1701 
1702 /*
1703  * When we are falling back to another migratetype during allocation, try to
1704  * steal extra free pages from the same pageblocks to satisfy further
1705  * allocations, instead of polluting multiple pageblocks.
1706  *
1707  * If we are stealing a relatively large buddy page, it is likely there will
1708  * be more free pages in the pageblock, so try to steal them all. For
1709  * reclaimable and unmovable allocations, we steal regardless of page size,
1710  * as fragmentation caused by those allocations polluting movable pageblocks
1711  * is worse than movable allocations stealing from unmovable and reclaimable
1712  * pageblocks.
1713  */
1714 static bool can_steal_fallback(unsigned int order, int start_mt)
1715 {
1716         /*
1717          * Leaving this order check is intended, although there is
1718          * relaxed order check in next check. The reason is that
1719          * we can actually steal whole pageblock if this condition met,
1720          * but, below check doesn't guarantee it and that is just heuristic
1721          * so could be changed anytime.
1722          */
1723         if (order >= pageblock_order)
1724                 return true;
1725 
1726         if (order >= pageblock_order / 2 ||
1727                 start_mt == MIGRATE_RECLAIMABLE ||
1728                 start_mt == MIGRATE_UNMOVABLE ||
1729                 page_group_by_mobility_disabled)
1730                 return true;
1731 
1732         return false;
1733 }
1734 
1735 /*
1736  * This function implements actual steal behaviour. If order is large enough,
1737  * we can steal whole pageblock. If not, we first move freepages in this
1738  * pageblock and check whether half of pages are moved or not. If half of
1739  * pages are moved, we can change migratetype of pageblock and permanently
1740  * use it's pages as requested migratetype in the future.
1741  */
1742 static void steal_suitable_fallback(struct zone *zone, struct page *page,
1743                                                           int start_type)
1744 {
1745         unsigned int current_order = page_order(page);
1746         int pages;
1747 
1748         /* Take ownership for orders >= pageblock_order */
1749         if (current_order >= pageblock_order) {
1750                 change_pageblock_range(page, current_order, start_type);
1751                 return;
1752         }
1753 
1754         pages = move_freepages_block(zone, page, start_type);
1755 
1756         /* Claim the whole block if over half of it is free */
1757         if (pages >= (1 << (pageblock_order-1)) ||
1758                         page_group_by_mobility_disabled)
1759                 set_pageblock_migratetype(page, start_type);
1760 }
1761 
1762 /*
1763  * Check whether there is a suitable fallback freepage with requested order.
1764  * If only_stealable is true, this function returns fallback_mt only if
1765  * we can steal other freepages all together. This would help to reduce
1766  * fragmentation due to mixed migratetype pages in one pageblock.
1767  */
1768 int find_suitable_fallback(struct free_area *area, unsigned int order,
1769                         int migratetype, bool only_stealable, bool *can_steal)
1770 {
1771         int i;
1772         int fallback_mt;
1773 
1774         if (area->nr_free == 0)
1775                 return -1;
1776 
1777         *can_steal = false;
1778         for (i = 0;; i++) {
1779                 fallback_mt = fallbacks[migratetype][i];
1780                 if (fallback_mt == MIGRATE_TYPES)
1781                         break;
1782 
1783                 if (list_empty(&area->free_list[fallback_mt]))
1784                         continue;
1785 
1786                 if (can_steal_fallback(order, migratetype))
1787                         *can_steal = true;
1788 
1789                 if (!only_stealable)
1790                         return fallback_mt;
1791 
1792                 if (*can_steal)
1793                         return fallback_mt;
1794         }
1795 
1796         return -1;
1797 }
1798 
1799 /*
1800  * Reserve a pageblock for exclusive use of high-order atomic allocations if
1801  * there are no empty page blocks that contain a page with a suitable order
1802  */
1803 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
1804                                 unsigned int alloc_order)
1805 {
1806         int mt;
1807         unsigned long max_managed, flags;
1808 
1809         /*
1810          * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1811          * Check is race-prone but harmless.
1812          */
1813         max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
1814         if (zone->nr_reserved_highatomic >= max_managed)
1815                 return;
1816 
1817         spin_lock_irqsave(&zone->lock, flags);
1818 
1819         /* Recheck the nr_reserved_highatomic limit under the lock */
1820         if (zone->nr_reserved_highatomic >= max_managed)
1821                 goto out_unlock;
1822 
1823         /* Yoink! */
1824         mt = get_pageblock_migratetype(page);
1825         if (mt != MIGRATE_HIGHATOMIC &&
1826                         !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
1827                 zone->nr_reserved_highatomic += pageblock_nr_pages;
1828                 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
1829                 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
1830         }
1831 
1832 out_unlock:
1833         spin_unlock_irqrestore(&zone->lock, flags);
1834 }
1835 
1836 /*
1837  * Used when an allocation is about to fail under memory pressure. This
1838  * potentially hurts the reliability of high-order allocations when under
1839  * intense memory pressure but failed atomic allocations should be easier
1840  * to recover from than an OOM.
1841  */
1842 static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
1843 {
1844         struct zonelist *zonelist = ac->zonelist;
1845         unsigned long flags;
1846         struct zoneref *z;
1847         struct zone *zone;
1848         struct page *page;
1849         int order;
1850 
1851         for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
1852                                                                 ac->nodemask) {
1853                 /* Preserve at least one pageblock */
1854                 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
1855                         continue;
1856 
1857                 spin_lock_irqsave(&zone->lock, flags);
1858                 for (order = 0; order < MAX_ORDER; order++) {
1859                         struct free_area *area = &(zone->free_area[order]);
1860 
1861                         page = list_first_entry_or_null(
1862                                         &area->free_list[MIGRATE_HIGHATOMIC],
1863                                         struct page, lru);
1864                         if (!page)
1865                                 continue;
1866 
1867                         /*
1868                          * It should never happen but changes to locking could
1869                          * inadvertently allow a per-cpu drain to add pages
1870                          * to MIGRATE_HIGHATOMIC while unreserving so be safe
1871                          * and watch for underflows.
1872                          */
1873                         zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
1874                                 zone->nr_reserved_highatomic);
1875 
1876                         /*
1877                          * Convert to ac->migratetype and avoid the normal
1878                          * pageblock stealing heuristics. Minimally, the caller
1879                          * is doing the work and needs the pages. More
1880                          * importantly, if the block was always converted to
1881                          * MIGRATE_UNMOVABLE or another type then the number
1882                          * of pageblocks that cannot be completely freed
1883                          * may increase.
1884                          */
1885                         set_pageblock_migratetype(page, ac->migratetype);
1886                         move_freepages_block(zone, page, ac->migratetype);
1887                         spin_unlock_irqrestore(&zone->lock, flags);
1888                         return;
1889                 }
1890                 spin_unlock_irqrestore(&zone->lock, flags);
1891         }
1892 }
1893 
1894 /* Remove an element from the buddy allocator from the fallback list */
1895 static inline struct page *
1896 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
1897 {
1898         struct free_area *area;
1899         unsigned int current_order;
1900         struct page *page;
1901         int fallback_mt;
1902         bool can_steal;
1903 
1904         /* Find the largest possible block of pages in the other list */
1905         for (current_order = MAX_ORDER-1;
1906                                 current_order >= order && current_order <= MAX_ORDER-1;
1907                                 --current_order) {
1908                 area = &(zone->free_area[current_order]);
1909                 fallback_mt = find_suitable_fallback(area, current_order,
1910                                 start_migratetype, false, &can_steal);
1911                 if (fallback_mt == -1)
1912                         continue;
1913 
1914                 page = list_first_entry(&area->free_list[fallback_mt],
1915                                                 struct page, lru);
1916                 if (can_steal)
1917                         steal_suitable_fallback(zone, page, start_migratetype);
1918 
1919                 /* Remove the page from the freelists */
1920                 area->nr_free--;
1921                 list_del(&page->lru);
1922                 rmv_page_order(page);
1923 
1924                 expand(zone, page, order, current_order, area,
1925                                         start_migratetype);
1926                 /*
1927                  * The pcppage_migratetype may differ from pageblock's
1928                  * migratetype depending on the decisions in
1929                  * find_suitable_fallback(). This is OK as long as it does not
1930                  * differ for MIGRATE_CMA pageblocks. Those can be used as
1931                  * fallback only via special __rmqueue_cma_fallback() function
1932                  */
1933                 set_pcppage_migratetype(page, start_migratetype);
1934 
1935                 trace_mm_page_alloc_extfrag(page, order, current_order,
1936                         start_migratetype, fallback_mt);
1937 
1938                 return page;
1939         }
1940 
1941         return NULL;
1942 }
1943 
1944 /*
1945  * Do the hard work of removing an element from the buddy allocator.
1946  * Call me with the zone->lock already held.
1947  */
1948 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1949                                 int migratetype)
1950 {
1951         struct page *page;
1952 
1953         page = __rmqueue_smallest(zone, order, migratetype);
1954         if (unlikely(!page)) {
1955                 if (migratetype == MIGRATE_MOVABLE)
1956                         page = __rmqueue_cma_fallback(zone, order);
1957 
1958                 if (!page)
1959                         page = __rmqueue_fallback(zone, order, migratetype);
1960         }
1961 
1962         trace_mm_page_alloc_zone_locked(page, order, migratetype);
1963         return page;
1964 }
1965 
1966 /*
1967  * Obtain a specified number of elements from the buddy allocator, all under
1968  * a single hold of the lock, for efficiency.  Add them to the supplied list.
1969  * Returns the number of new pages which were placed at *list.
1970  */
1971 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1972                         unsigned long count, struct list_head *list,
1973                         int migratetype, bool cold)
1974 {
1975         int i;
1976 
1977         spin_lock(&zone->lock);
1978         for (i = 0; i < count; ++i) {
1979                 struct page *page = __rmqueue(zone, order, migratetype);
1980                 if (unlikely(page == NULL))
1981                         break;
1982 
1983                 /*
1984                  * Split buddy pages returned by expand() are received here
1985                  * in physical page order. The page is added to the callers and
1986                  * list and the list head then moves forward. From the callers
1987                  * perspective, the linked list is ordered by page number in
1988                  * some conditions. This is useful for IO devices that can
1989                  * merge IO requests if the physical pages are ordered
1990                  * properly.
1991                  */
1992                 if (likely(!cold))
1993                         list_add(&page->lru, list);
1994                 else
1995                         list_add_tail(&page->lru, list);
1996                 list = &page->lru;
1997                 if (is_migrate_cma(get_pcppage_migratetype(page)))
1998                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1999                                               -(1 << order));
2000         }
2001         __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2002         spin_unlock(&zone->lock);
2003         return i;
2004 }
2005 
2006 #ifdef CONFIG_NUMA
2007 /*
2008  * Called from the vmstat counter updater to drain pagesets of this
2009  * currently executing processor on remote nodes after they have
2010  * expired.
2011  *
2012  * Note that this function must be called with the thread pinned to
2013  * a single processor.
2014  */
2015 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2016 {
2017         unsigned long flags;
2018         int to_drain, batch;
2019 
2020         local_irq_save(flags);
2021         batch = READ_ONCE(pcp->batch);
2022         to_drain = min(pcp->count, batch);
2023         if (to_drain > 0) {
2024                 free_pcppages_bulk(zone, to_drain, pcp);
2025                 pcp->count -= to_drain;
2026         }
2027         local_irq_restore(flags);
2028 }
2029 #endif
2030 
2031 /*
2032  * Drain pcplists of the indicated processor and zone.
2033  *
2034  * The processor must either be the current processor and the
2035  * thread pinned to the current processor or a processor that
2036  * is not online.
2037  */
2038 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2039 {
2040         unsigned long flags;
2041         struct per_cpu_pageset *pset;
2042         struct per_cpu_pages *pcp;
2043 
2044         local_irq_save(flags);
2045         pset = per_cpu_ptr(zone->pageset, cpu);
2046 
2047         pcp = &pset->pcp;
2048         if (pcp->count) {
2049                 free_pcppages_bulk(zone, pcp->count, pcp);
2050                 pcp->count = 0;
2051         }
2052         local_irq_restore(flags);
2053 }
2054 
2055 /*
2056  * Drain pcplists of all zones on the indicated processor.
2057  *
2058  * The processor must either be the current processor and the
2059  * thread pinned to the current processor or a processor that
2060  * is not online.
2061  */
2062 static void drain_pages(unsigned int cpu)
2063 {
2064         struct zone *zone;
2065 
2066         for_each_populated_zone(zone) {
2067                 drain_pages_zone(cpu, zone);
2068         }
2069 }
2070 
2071 /*
2072  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2073  *
2074  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2075  * the single zone's pages.
2076  */
2077 void drain_local_pages(struct zone *zone)
2078 {
2079         int cpu = smp_processor_id();
2080 
2081         if (zone)
2082                 drain_pages_zone(cpu, zone);
2083         else
2084                 drain_pages(cpu);
2085 }
2086 
2087 /*
2088  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2089  *
2090  * When zone parameter is non-NULL, spill just the single zone's pages.
2091  *
2092  * Note that this code is protected against sending an IPI to an offline
2093  * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2094  * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2095  * nothing keeps CPUs from showing up after we populated the cpumask and
2096  * before the call to on_each_cpu_mask().
2097  */
2098 void drain_all_pages(struct zone *zone)
2099 {
2100         int cpu;
2101 
2102         /*
2103          * Allocate in the BSS so we wont require allocation in
2104          * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2105          */
2106         static cpumask_t cpus_with_pcps;
2107 
2108         /*
2109          * We don't care about racing with CPU hotplug event
2110          * as offline notification will cause the notified
2111          * cpu to drain that CPU pcps and on_each_cpu_mask
2112          * disables preemption as part of its processing
2113          */
2114         for_each_online_cpu(cpu) {
2115                 struct per_cpu_pageset *pcp;
2116                 struct zone *z;
2117                 bool has_pcps = false;
2118 
2119                 if (zone) {
2120                         pcp = per_cpu_ptr(zone->pageset, cpu);
2121                         if (pcp->pcp.count)
2122                                 has_pcps = true;
2123                 } else {
2124                         for_each_populated_zone(z) {
2125                                 pcp = per_cpu_ptr(z->pageset, cpu);
2126                                 if (pcp->pcp.count) {
2127                                         has_pcps = true;
2128                                         break;
2129                                 }
2130                         }
2131                 }
2132 
2133                 if (has_pcps)
2134                         cpumask_set_cpu(cpu, &cpus_with_pcps);
2135                 else
2136                         cpumask_clear_cpu(cpu, &cpus_with_pcps);
2137         }
2138         on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
2139                                                                 zone, 1);
2140 }
2141 
2142 #ifdef CONFIG_HIBERNATION
2143 
2144 void mark_free_pages(struct zone *zone)
2145 {
2146         unsigned long pfn, max_zone_pfn;
2147         unsigned long flags;
2148         unsigned int order, t;
2149         struct page *page;
2150 
2151         if (zone_is_empty(zone))
2152                 return;
2153 
2154         spin_lock_irqsave(&zone->lock, flags);
2155 
2156         max_zone_pfn = zone_end_pfn(zone);
2157         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2158                 if (pfn_valid(pfn)) {
2159                         page = pfn_to_page(pfn);
2160                         if (!swsusp_page_is_forbidden(page))
2161                                 swsusp_unset_page_free(page);
2162                 }
2163 
2164         for_each_migratetype_order(order, t) {
2165                 list_for_each_entry(page,
2166                                 &zone->free_area[order].free_list[t], lru) {
2167                         unsigned long i;
2168 
2169                         pfn = page_to_pfn(page);
2170                         for (i = 0; i < (1UL << order); i++)
2171                                 swsusp_set_page_free(pfn_to_page(pfn + i));
2172                 }
2173         }
2174         spin_unlock_irqrestore(&zone->lock, flags);
2175 }
2176 #endif /* CONFIG_PM */
2177 
2178 /*
2179  * Free a 0-order page
2180  * cold == true ? free a cold page : free a hot page
2181  */
2182 void free_hot_cold_page(struct page *page, bool cold)
2183 {
2184         struct zone *zone = page_zone(page);
2185         struct per_cpu_pages *pcp;
2186         unsigned long flags;
2187         unsigned long pfn = page_to_pfn(page);
2188         int migratetype;
2189 
2190         if (!free_pages_prepare(page, 0))
2191                 return;
2192 
2193         migratetype = get_pfnblock_migratetype(page, pfn);
2194         set_pcppage_migratetype(page, migratetype);
2195         local_irq_save(flags);
2196         __count_vm_event(PGFREE);
2197 
2198         /*
2199          * We only track unmovable, reclaimable and movable on pcp lists.
2200          * Free ISOLATE pages back to the allocator because they are being
2201          * offlined but treat RESERVE as movable pages so we can get those
2202          * areas back if necessary. Otherwise, we may have to free
2203          * excessively into the page allocator
2204          */
2205         if (migratetype >= MIGRATE_PCPTYPES) {
2206                 if (unlikely(is_migrate_isolate(migratetype))) {
2207                         free_one_page(zone, page, pfn, 0, migratetype);
2208                         goto out;
2209                 }
2210                 migratetype = MIGRATE_MOVABLE;
2211         }
2212 
2213         pcp = &this_cpu_ptr(zone->pageset)->pcp;
2214         if (!cold)
2215                 list_add(&page->lru, &pcp->lists[migratetype]);
2216         else
2217                 list_add_tail(&page->lru, &pcp->lists[migratetype]);
2218         pcp->count++;
2219         if (pcp->count >= pcp->high) {
2220                 unsigned long batch = READ_ONCE(pcp->batch);
2221                 free_pcppages_bulk(zone, batch, pcp);
2222                 pcp->count -= batch;
2223         }
2224 
2225 out:
2226         local_irq_restore(flags);
2227 }
2228 
2229 /*
2230  * Free a list of 0-order pages
2231  */
2232 void free_hot_cold_page_list(struct list_head *list, bool cold)
2233 {
2234         struct page *page, *next;
2235 
2236         list_for_each_entry_safe(page, next, list, lru) {
2237                 trace_mm_page_free_batched(page, cold);
2238                 free_hot_cold_page(page, cold);
2239         }
2240 }
2241 
2242 /*
2243  * split_page takes a non-compound higher-order page, and splits it into
2244  * n (1<<order) sub-pages: page[0..n]
2245  * Each sub-page must be freed individually.
2246  *
2247  * Note: this is probably too low level an operation for use in drivers.
2248  * Please consult with lkml before using this in your driver.
2249  */
2250 void split_page(struct page *page, unsigned int order)
2251 {
2252         int i;
2253         gfp_t gfp_mask;
2254 
2255         VM_BUG_ON_PAGE(PageCompound(page), page);
2256         VM_BUG_ON_PAGE(!page_count(page), page);
2257 
2258 #ifdef CONFIG_KMEMCHECK
2259         /*
2260          * Split shadow pages too, because free(page[0]) would
2261          * otherwise free the whole shadow.
2262          */
2263         if (kmemcheck_page_is_tracked(page))
2264                 split_page(virt_to_page(page[0].shadow), order);
2265 #endif
2266 
2267         gfp_mask = get_page_owner_gfp(page);
2268         set_page_owner(page, 0, gfp_mask);
2269         for (i = 1; i < (1 << order); i++) {
2270                 set_page_refcounted(page + i);
2271                 set_page_owner(page + i, 0, gfp_mask);
2272         }
2273 }
2274 EXPORT_SYMBOL_GPL(split_page);
2275 
2276 int __isolate_free_page(struct page *page, unsigned int order)
2277 {
2278         unsigned long watermark;
2279         struct zone *zone;
2280         int mt;
2281 
2282         BUG_ON(!PageBuddy(page));
2283 
2284         zone = page_zone(page);
2285         mt = get_pageblock_migratetype(page);
2286 
2287         if (!is_migrate_isolate(mt)) {
2288                 /* Obey watermarks as if the page was being allocated */
2289                 watermark = low_wmark_pages(zone) + (1 << order);
2290                 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2291                         return 0;
2292 
2293                 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2294         }
2295 
2296         /* Remove page from free list */
2297         list_del(&page->lru);
2298         zone->free_area[order].nr_free--;
2299         rmv_page_order(page);
2300 
2301         set_page_owner(page, order, __GFP_MOVABLE);
2302 
2303         /* Set the pageblock if the isolated page is at least a pageblock */
2304         if (order >= pageblock_order - 1) {
2305                 struct page *endpage = page + (1 << order) - 1;
2306                 for (; page < endpage; page += pageblock_nr_pages) {
2307                         int mt = get_pageblock_migratetype(page);
2308                         if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
2309                                 set_pageblock_migratetype(page,
2310                                                           MIGRATE_MOVABLE);
2311                 }
2312         }
2313 
2314 
2315         return 1UL << order;
2316 }
2317 
2318 /*
2319  * Similar to split_page except the page is already free. As this is only
2320  * being used for migration, the migratetype of the block also changes.
2321  * As this is called with interrupts disabled, the caller is responsible
2322  * for calling arch_alloc_page() and kernel_map_page() after interrupts
2323  * are enabled.
2324  *
2325  * Note: this is probably too low level an operation for use in drivers.
2326  * Please consult with lkml before using this in your driver.
2327  */
2328 int split_free_page(struct page *page)
2329 {
2330         unsigned int order;
2331         int nr_pages;
2332 
2333         order = page_order(page);
2334 
2335         nr_pages = __isolate_free_page(page, order);
2336         if (!nr_pages)
2337                 return 0;
2338 
2339         /* Split into individual pages */
2340         set_page_refcounted(page);
2341         split_page(page, order);
2342         return nr_pages;
2343 }
2344 
2345 /*
2346  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2347  */
2348 static inline
2349 struct page *buffered_rmqueue(struct zone *preferred_zone,
2350                         struct zone *zone, unsigned int order,
2351                         gfp_t gfp_flags, int alloc_flags, int migratetype)
2352 {
2353         unsigned long flags;
2354         struct page *page;
2355         bool cold = ((gfp_flags & __GFP_COLD) != 0);
2356 
2357         if (likely(order == 0)) {
2358                 struct per_cpu_pages *pcp;
2359                 struct list_head *list;
2360 
2361                 local_irq_save(flags);
2362                 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2363                 list = &pcp->lists[migratetype];
2364                 if (list_empty(list)) {
2365                         pcp->count += rmqueue_bulk(zone, 0,
2366                                         pcp->batch, list,
2367                                         migratetype, cold);
2368                         if (unlikely(list_empty(list)))
2369                                 goto failed;
2370                 }
2371 
2372                 if (cold)
2373                         page = list_last_entry(list, struct page, lru);
2374                 else
2375                         page = list_first_entry(list, struct page, lru);
2376 
2377                 list_del(&page->lru);
2378                 pcp->count--;
2379         } else {
2380                 /*
2381                  * We most definitely don't want callers attempting to
2382                  * allocate greater than order-1 page units with __GFP_NOFAIL.
2383                  */
2384                 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
2385                 spin_lock_irqsave(&zone->lock, flags);
2386 
2387                 page = NULL;
2388                 if (alloc_flags & ALLOC_HARDER) {
2389                         page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2390                         if (page)
2391                                 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2392                 }
2393                 if (!page)
2394                         page = __rmqueue(zone, order, migratetype);
2395                 spin_unlock(&zone->lock);
2396                 if (!page)
2397                         goto failed;
2398                 __mod_zone_freepage_state(zone, -(1 << order),
2399                                           get_pcppage_migratetype(page));
2400         }
2401 
2402         __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
2403         if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
2404             !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2405                 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2406 
2407         __count_zone_vm_events(PGALLOC, zone, 1 << order);
2408         zone_statistics(preferred_zone, zone, gfp_flags);
2409         local_irq_restore(flags);
2410 
2411         VM_BUG_ON_PAGE(bad_range(zone, page), page);
2412         return page;
2413 
2414 failed:
2415         local_irq_restore(flags);
2416         return NULL;
2417 }
2418 
2419 #ifdef CONFIG_FAIL_PAGE_ALLOC
2420 
2421 static struct {
2422         struct fault_attr attr;
2423 
2424         bool ignore_gfp_highmem;
2425         bool ignore_gfp_reclaim;
2426         u32 min_order;
2427 } fail_page_alloc = {
2428         .attr = FAULT_ATTR_INITIALIZER,
2429         .ignore_gfp_reclaim = true,
2430         .ignore_gfp_highmem = true,
2431         .min_order = 1,
2432 };
2433 
2434 static int __init setup_fail_page_alloc(char *str)
2435 {
2436         return setup_fault_attr(&fail_page_alloc.attr, str);
2437 }
2438 __setup("fail_page_alloc=", setup_fail_page_alloc);
2439 
2440 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2441 {
2442         if (order < fail_page_alloc.min_order)
2443                 return false;
2444         if (gfp_mask & __GFP_NOFAIL)
2445                 return false;
2446         if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
2447                 return false;
2448         if (fail_page_alloc.ignore_gfp_reclaim &&
2449                         (gfp_mask & __GFP_DIRECT_RECLAIM))
2450                 return false;
2451 
2452         return should_fail(&fail_page_alloc.attr, 1 << order);
2453 }
2454 
2455 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2456 
2457 static int __init fail_page_alloc_debugfs(void)
2458 {
2459         umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
2460         struct dentry *dir;
2461 
2462         dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2463                                         &fail_page_alloc.attr);
2464         if (IS_ERR(dir))
2465                 return PTR_ERR(dir);
2466 
2467         if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2468                                 &fail_page_alloc.ignore_gfp_reclaim))
2469                 goto fail;
2470         if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2471                                 &fail_page_alloc.ignore_gfp_highmem))
2472                 goto fail;
2473         if (!debugfs_create_u32("min-order", mode, dir,
2474                                 &fail_page_alloc.min_order))
2475                 goto fail;
2476 
2477         return 0;
2478 fail:
2479         debugfs_remove_recursive(dir);
2480 
2481         return -ENOMEM;
2482 }
2483 
2484 late_initcall(fail_page_alloc_debugfs);
2485 
2486 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2487 
2488 #else /* CONFIG_FAIL_PAGE_ALLOC */
2489 
2490 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2491 {
2492         return false;
2493 }
2494 
2495 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2496 
2497 /*
2498  * Return true if free base pages are above 'mark'. For high-order checks it
2499  * will return true of the order-0 watermark is reached and there is at least
2500  * one free page of a suitable size. Checking now avoids taking the zone lock
2501  * to check in the allocation paths if no pages are free.
2502  */
2503 static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2504                         unsigned long mark, int classzone_idx, int alloc_flags,
2505                         long free_pages)
2506 {
2507         long min = mark;
2508         int o;
2509         const int alloc_harder = (alloc_flags & ALLOC_HARDER);
2510 
2511         /* free_pages may go negative - that's OK */
2512         free_pages -= (1 << order) - 1;
2513 
2514         if (alloc_flags & ALLOC_HIGH)
2515                 min -= min / 2;
2516 
2517         /*
2518          * If the caller does not have rights to ALLOC_HARDER then subtract
2519          * the high-atomic reserves. This will over-estimate the size of the
2520          * atomic reserve but it avoids a search.
2521          */
2522         if (likely(!alloc_harder))
2523                 free_pages -= z->nr_reserved_highatomic;
2524         else
2525                 min -= min / 4;
2526 
2527 #ifdef CONFIG_CMA
2528         /* If allocation can't use CMA areas don't use free CMA pages */
2529         if (!(alloc_flags & ALLOC_CMA))
2530                 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
2531 #endif
2532 
2533         /*
2534          * Check watermarks for an order-0 allocation request. If these
2535          * are not met, then a high-order request also cannot go ahead
2536          * even if a suitable page happened to be free.
2537          */
2538         if (free_pages <= min + z->lowmem_reserve[classzone_idx])
2539                 return false;
2540 
2541         /* If this is an order-0 request then the watermark is fine */
2542         if (!order)
2543                 return true;
2544 
2545         /* For a high-order request, check at least one suitable page is free */
2546         for (o = order; o < MAX_ORDER; o++) {
2547                 struct free_area *area = &z->free_area[o];
2548                 int mt;
2549 
2550                 if (!area->nr_free)
2551                         continue;
2552 
2553                 if (alloc_harder)
2554                         return true;
2555 
2556                 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2557                         if (!list_empty(&area->free_list[mt]))
2558                                 return true;
2559                 }
2560 
2561 #ifdef CONFIG_CMA
2562                 if ((alloc_flags & ALLOC_CMA) &&
2563                     !list_empty(&area->free_list[MIGRATE_CMA])) {
2564                         return true;
2565                 }
2566 #endif
2567         }
2568         return false;
2569 }
2570 
2571 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2572                       int classzone_idx, int alloc_flags)
2573 {
2574         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2575                                         zone_page_state(z, NR_FREE_PAGES));
2576 }
2577 
2578 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2579                         unsigned long mark, int classzone_idx)
2580 {
2581         long free_pages = zone_page_state(z, NR_FREE_PAGES);
2582 
2583         if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2584                 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2585 
2586         return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
2587                                                                 free_pages);
2588 }
2589 
2590 #ifdef CONFIG_NUMA
2591 static bool zone_local(struct zone *local_zone, struct zone *zone)
2592 {
2593         return local_zone->node == zone->node;
2594 }
2595 
2596 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2597 {
2598         return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2599                                 RECLAIM_DISTANCE;
2600 }
2601 #else   /* CONFIG_NUMA */
2602 static bool zone_local(struct zone *local_zone, struct zone *zone)
2603 {
2604         return true;
2605 }
2606 
2607 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2608 {
2609         return true;
2610 }
2611 #endif  /* CONFIG_NUMA */
2612 
2613 static void reset_alloc_batches(struct zone *preferred_zone)
2614 {
2615         struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2616 
2617         do {
2618                 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2619                         high_wmark_pages(zone) - low_wmark_pages(zone) -
2620                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
2621                 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2622         } while (zone++ != preferred_zone);
2623 }
2624 
2625 /*
2626  * get_page_from_freelist goes through the zonelist trying to allocate
2627  * a page.
2628  */
2629 static struct page *
2630 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2631                                                 const struct alloc_context *ac)
2632 {
2633         struct zonelist *zonelist = ac->zonelist;
2634         struct zoneref *z;
2635         struct page *page = NULL;
2636         struct zone *zone;
2637         int nr_fair_skipped = 0;
2638         bool zonelist_rescan;
2639 
2640 zonelist_scan:
2641         zonelist_rescan = false;
2642 
2643         /*
2644          * Scan zonelist, looking for a zone with enough free.
2645          * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2646          */
2647         for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2648                                                                 ac->nodemask) {
2649                 unsigned long mark;
2650 
2651                 if (cpusets_enabled() &&
2652                         (alloc_flags & ALLOC_CPUSET) &&
2653                         !cpuset_zone_allowed(zone, gfp_mask))
2654                                 continue;
2655                 /*
2656                  * Distribute pages in proportion to the individual
2657                  * zone size to ensure fair page aging.  The zone a
2658                  * page was allocated in should have no effect on the
2659                  * time the page has in memory before being reclaimed.
2660                  */
2661                 if (alloc_flags & ALLOC_FAIR) {
2662                         if (!zone_local(ac->preferred_zone, zone))
2663                                 break;
2664                         if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
2665                                 nr_fair_skipped++;
2666                                 continue;
2667                         }
2668                 }
2669                 /*
2670                  * When allocating a page cache page for writing, we
2671                  * want to get it from a zone that is within its dirty
2672                  * limit, such that no single zone holds more than its
2673                  * proportional share of globally allowed dirty pages.
2674                  * The dirty limits take into account the zone's
2675                  * lowmem reserves and high watermark so that kswapd
2676                  * should be able to balance it without having to
2677                  * write pages from its LRU list.
2678                  *
2679                  * This may look like it could increase pressure on
2680                  * lower zones by failing allocations in higher zones
2681                  * before they are full.  But the pages that do spill
2682                  * over are limited as the lower zones are protected
2683                  * by this very same mechanism.  It should not become
2684                  * a practical burden to them.
2685                  *
2686                  * XXX: For now, allow allocations to potentially
2687                  * exceed the per-zone dirty limit in the slowpath
2688                  * (spread_dirty_pages unset) before going into reclaim,
2689                  * which is important when on a NUMA setup the allowed
2690                  * zones are together not big enough to reach the
2691                  * global limit.  The proper fix for these situations
2692                  * will require awareness of zones in the
2693                  * dirty-throttling and the flusher threads.
2694                  */
2695                 if (ac->spread_dirty_pages && !zone_dirty_ok(zone))
2696                         continue;
2697 
2698                 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2699                 if (!zone_watermark_ok(zone, order, mark,
2700                                        ac->classzone_idx, alloc_flags)) {
2701                         int ret;
2702 
2703                         /* Checked here to keep the fast path fast */
2704                         BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2705                         if (alloc_flags & ALLOC_NO_WATERMARKS)
2706                                 goto try_this_zone;
2707 
2708                         if (zone_reclaim_mode == 0 ||
2709                             !zone_allows_reclaim(ac->preferred_zone, zone))
2710                                 continue;
2711 
2712                         ret = zone_reclaim(zone, gfp_mask, order);
2713                         switch (ret) {
2714                         case ZONE_RECLAIM_NOSCAN:
2715                                 /* did not scan */
2716                                 continue;
2717                         case ZONE_RECLAIM_FULL:
2718                                 /* scanned but unreclaimable */
2719                                 continue;
2720                         default:
2721                                 /* did we reclaim enough */
2722                                 if (zone_watermark_ok(zone, order, mark,
2723                                                 ac->classzone_idx, alloc_flags))
2724                                         goto try_this_zone;
2725 
2726                                 continue;
2727                         }
2728                 }
2729 
2730 try_this_zone:
2731                 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2732                                 gfp_mask, alloc_flags, ac->migratetype);
2733                 if (page) {
2734                         if (prep_new_page(page, order, gfp_mask, alloc_flags))
2735                                 goto try_this_zone;
2736 
2737                         /*
2738                          * If this is a high-order atomic allocation then check
2739                          * if the pageblock should be reserved for the future
2740                          */
2741                         if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2742                                 reserve_highatomic_pageblock(page, zone, order);
2743 
2744                         return page;
2745                 }
2746         }
2747 
2748         /*
2749          * The first pass makes sure allocations are spread fairly within the
2750          * local node.  However, the local node might have free pages left
2751          * after the fairness batches are exhausted, and remote zones haven't
2752          * even been considered yet.  Try once more without fairness, and
2753          * include remote zones now, before entering the slowpath and waking
2754          * kswapd: prefer spilling to a remote zone over swapping locally.
2755          */
2756         if (alloc_flags & ALLOC_FAIR) {
2757                 alloc_flags &= ~ALLOC_FAIR;
2758                 if (nr_fair_skipped) {
2759                         zonelist_rescan = true;
2760                         reset_alloc_batches(ac->preferred_zone);
2761                 }
2762                 if (nr_online_nodes > 1)
2763                         zonelist_rescan = true;
2764         }
2765 
2766         if (zonelist_rescan)
2767                 goto zonelist_scan;
2768 
2769         return NULL;
2770 }
2771 
2772 /*
2773  * Large machines with many possible nodes should not always dump per-node
2774  * meminfo in irq context.
2775  */
2776 static inline bool should_suppress_show_mem(void)
2777 {
2778         bool ret = false;
2779 
2780 #if NODES_SHIFT > 8
2781         ret = in_interrupt();
2782 #endif
2783         return ret;
2784 }
2785 
2786 static DEFINE_RATELIMIT_STATE(nopage_rs,
2787                 DEFAULT_RATELIMIT_INTERVAL,
2788                 DEFAULT_RATELIMIT_BURST);
2789 
2790 void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
2791 {
2792         unsigned int filter = SHOW_MEM_FILTER_NODES;
2793 
2794         if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2795             debug_guardpage_minorder() > 0)
2796                 return;
2797 
2798         /*
2799          * This documents exceptions given to allocations in certain
2800          * contexts that are allowed to allocate outside current's set
2801          * of allowed nodes.
2802          */
2803         if (!(gfp_mask & __GFP_NOMEMALLOC))
2804                 if (test_thread_flag(TIF_MEMDIE) ||
2805                     (current->flags & (PF_MEMALLOC | PF_EXITING)))
2806                         filter &= ~SHOW_MEM_FILTER_NODES;
2807         if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
2808                 filter &= ~SHOW_MEM_FILTER_NODES;
2809 
2810         if (fmt) {
2811                 struct va_format vaf;
2812                 va_list args;
2813 
2814                 va_start(args, fmt);
2815 
2816                 vaf.fmt = fmt;
2817                 vaf.va = &args;
2818 
2819                 pr_warn("%pV", &vaf);
2820 
2821                 va_end(args);
2822         }
2823 
2824         pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
2825                 current->comm, order, gfp_mask, &gfp_mask);
2826         dump_stack();
2827         if (!should_suppress_show_mem())
2828                 show_mem(filter);
2829 }
2830 
2831 static inline struct page *
2832 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2833         const struct alloc_context *ac, unsigned long *did_some_progress)
2834 {
2835         struct oom_control oc = {
2836                 .zonelist = ac->zonelist,
2837                 .nodemask = ac->nodemask,
2838                 .gfp_mask = gfp_mask,
2839                 .order = order,
2840         };
2841         struct page *page;
2842 
2843         *did_some_progress = 0;
2844 
2845         /*
2846          * Acquire the oom lock.  If that fails, somebody else is
2847          * making progress for us.
2848          */
2849         if (!mutex_trylock(&oom_lock)) {
2850                 *did_some_progress = 1;
2851                 schedule_timeout_uninterruptible(1);
2852                 return NULL;
2853         }
2854 
2855         /*
2856          * Go through the zonelist yet one more time, keep very high watermark
2857          * here, this is only to catch a parallel oom killing, we must fail if
2858          * we're still under heavy pressure.
2859          */
2860         page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2861                                         ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
2862         if (page)
2863                 goto out;
2864 
2865         if (!(gfp_mask & __GFP_NOFAIL)) {
2866                 /* Coredumps can quickly deplete all memory reserves */
2867                 if (current->flags & PF_DUMPCORE)
2868                         goto out;
2869                 /* The OOM killer will not help higher order allocs */
2870                 if (order > PAGE_ALLOC_COSTLY_ORDER)
2871                         goto out;
2872                 /* The OOM killer does not needlessly kill tasks for lowmem */
2873                 if (ac->high_zoneidx < ZONE_NORMAL)
2874                         goto out;
2875                 /* The OOM killer does not compensate for IO-less reclaim */
2876                 if (!(gfp_mask & __GFP_FS)) {
2877                         /*
2878                          * XXX: Page reclaim didn't yield anything,
2879                          * and the OOM killer can't be invoked, but
2880                          * keep looping as per tradition.
2881                          *
2882                          * But do not keep looping if oom_killer_disable()
2883                          * was already called, for the system is trying to
2884                          * enter a quiescent state during suspend.
2885                          */
2886                         *did_some_progress = !oom_killer_disabled;
2887                         goto out;
2888                 }
2889                 if (pm_suspended_storage())
2890                         goto out;
2891                 /* The OOM killer may not free memory on a specific node */
2892                 if (gfp_mask & __GFP_THISNODE)
2893                         goto out;
2894         }
2895         /* Exhausted what can be done so it's blamo time */
2896         if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
2897                 *did_some_progress = 1;
2898 
2899                 if (gfp_mask & __GFP_NOFAIL) {
2900                         page = get_page_from_freelist(gfp_mask, order,
2901                                         ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac);
2902                         /*
2903                          * fallback to ignore cpuset restriction if our nodes
2904                          * are depleted
2905                          */
2906                         if (!page)
2907                                 page = get_page_from_freelist(gfp_mask, order,
2908                                         ALLOC_NO_WATERMARKS, ac);
2909                 }
2910         }
2911 out:
2912         mutex_unlock(&oom_lock);
2913         return page;
2914 }
2915 
2916 #ifdef CONFIG_COMPACTION
2917 /* Try memory compaction for high-order allocations before reclaim */
2918 static struct page *
2919 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2920                 int alloc_flags, const struct alloc_context *ac,
2921                 enum migrate_mode mode, int *contended_compaction,
2922                 bool *deferred_compaction)
2923 {
2924         unsigned long compact_result;
2925         struct page *page;
2926 
2927         if (!order)
2928                 return NULL;
2929 
2930         current->flags |= PF_MEMALLOC;
2931         compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2932                                                 mode, contended_compaction);
2933         current->flags &= ~PF_MEMALLOC;
2934 
2935         switch (compact_result) {
2936         case COMPACT_DEFERRED:
2937                 *deferred_compaction = true;
2938                 /* fall-through */
2939         case COMPACT_SKIPPED:
2940                 return NULL;
2941         default:
2942                 break;
2943         }
2944 
2945         /*
2946          * At least in one zone compaction wasn't deferred or skipped, so let's
2947          * count a compaction stall
2948          */
2949         count_vm_event(COMPACTSTALL);
2950 
2951         page = get_page_from_freelist(gfp_mask, order,
2952                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2953 
2954         if (page) {
2955                 struct zone *zone = page_zone(page);
2956 
2957                 zone->compact_blockskip_flush = false;
2958                 compaction_defer_reset(zone, order, true);
2959                 count_vm_event(COMPACTSUCCESS);
2960                 return page;
2961         }
2962 
2963         /*
2964          * It's bad if compaction run occurs and fails. The most likely reason
2965          * is that pages exist, but not enough to satisfy watermarks.
2966          */
2967         count_vm_event(COMPACTFAIL);
2968 
2969         cond_resched();
2970 
2971         return NULL;
2972 }
2973 #else
2974 static inline struct page *
2975 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2976                 int alloc_flags, const struct alloc_context *ac,
2977                 enum migrate_mode mode, int *contended_compaction,
2978                 bool *deferred_compaction)
2979 {
2980         return NULL;
2981 }
2982 #endif /* CONFIG_COMPACTION */
2983 
2984 /* Perform direct synchronous page reclaim */
2985 static int
2986 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
2987                                         const struct alloc_context *ac)
2988 {
2989         struct reclaim_state reclaim_state;
2990         int progress;
2991 
2992         cond_resched();
2993 
2994         /* We now go into synchronous reclaim */
2995         cpuset_memory_pressure_bump();
2996         current->flags |= PF_MEMALLOC;
2997         lockdep_set_current_reclaim_state(gfp_mask);
2998         reclaim_state.reclaimed_slab = 0;
2999         current->reclaim_state = &reclaim_state;
3000 
3001         progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3002                                                                 ac->nodemask);
3003 
3004         current->reclaim_state = NULL;
3005         lockdep_clear_current_reclaim_state();
3006         current->flags &= ~PF_MEMALLOC;
3007 
3008         cond_resched();
3009 
3010         return progress;
3011 }
3012 
3013 /* The really slow allocator path where we enter direct reclaim */
3014 static inline struct page *
3015 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
3016                 int alloc_flags, const struct alloc_context *ac,
3017                 unsigned long *did_some_progress)
3018 {
3019         struct page *page = NULL;
3020         bool drained = false;
3021 
3022         *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
3023         if (unlikely(!(*did_some_progress)))
3024                 return NULL;
3025 
3026 retry:
3027         page = get_page_from_freelist(gfp_mask, order,
3028                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3029 
3030         /*
3031          * If an allocation failed after direct reclaim, it could be because
3032          * pages are pinned on the per-cpu lists or in high alloc reserves.
3033          * Shrink them them and try again
3034          */
3035         if (!page && !drained) {
3036                 unreserve_highatomic_pageblock(ac);
3037                 drain_all_pages(NULL);
3038                 drained = true;
3039                 goto retry;
3040         }
3041 
3042         return page;
3043 }
3044 
3045 static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
3046 {
3047         struct zoneref *z;
3048         struct zone *zone;
3049 
3050         for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
3051                                                 ac->high_zoneidx, ac->nodemask)
3052                 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
3053 }
3054 
3055 static inline int
3056 gfp_to_alloc_flags(gfp_t gfp_mask)
3057 {
3058         int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
3059 
3060         /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3061         BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
3062 
3063         /*
3064          * The caller may dip into page reserves a bit more if the caller
3065          * cannot run direct reclaim, or if the caller has realtime scheduling
3066          * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
3067          * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3068          */
3069         alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
3070 
3071         if (gfp_mask & __GFP_ATOMIC) {
3072                 /*
3073                  * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3074                  * if it can't schedule.
3075                  */
3076                 if (!(gfp_mask & __GFP_NOMEMALLOC))
3077                         alloc_flags |= ALLOC_HARDER;
3078                 /*
3079                  * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3080                  * comment for __cpuset_node_allowed().
3081                  */
3082                 alloc_flags &= ~ALLOC_CPUSET;
3083         } else if (unlikely(rt_task(current)) && !in_interrupt())
3084                 alloc_flags |= ALLOC_HARDER;
3085 
3086         if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
3087                 if (gfp_mask & __GFP_MEMALLOC)
3088                         alloc_flags |= ALLOC_NO_WATERMARKS;
3089                 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3090                         alloc_flags |= ALLOC_NO_WATERMARKS;
3091                 else if (!in_interrupt() &&
3092                                 ((current->flags & PF_MEMALLOC) ||
3093                                  unlikely(test_thread_flag(TIF_MEMDIE))))
3094                         alloc_flags |= ALLOC_NO_WATERMARKS;
3095         }
3096 #ifdef CONFIG_CMA
3097         if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
3098                 alloc_flags |= ALLOC_CMA;
3099 #endif
3100         return alloc_flags;
3101 }
3102 
3103 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3104 {
3105         return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
3106 }
3107 
3108 static inline bool is_thp_gfp_mask(gfp_t gfp_mask)
3109 {
3110         return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE;
3111 }
3112 
3113 static inline struct page *
3114 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
3115                                                 struct alloc_context *ac)
3116 {
3117         bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
3118         struct page *page = NULL;
3119         int alloc_flags;
3120         unsigned long pages_reclaimed = 0;
3121         unsigned long did_some_progress;
3122         enum migrate_mode migration_mode = MIGRATE_ASYNC;
3123         bool deferred_compaction = false;
3124         int contended_compaction = COMPACT_CONTENDED_NONE;
3125 
3126         /*
3127          * In the slowpath, we sanity check order to avoid ever trying to
3128          * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3129          * be using allocators in order of preference for an area that is
3130          * too large.
3131          */
3132         if (order >= MAX_ORDER) {
3133                 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
3134                 return NULL;
3135         }
3136 
3137         /*
3138          * We also sanity check to catch abuse of atomic reserves being used by
3139          * callers that are not in atomic context.
3140          */
3141         if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3142                                 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3143                 gfp_mask &= ~__GFP_ATOMIC;
3144 
3145 retry:
3146         if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3147                 wake_all_kswapds(order, ac);
3148 
3149         /*
3150          * OK, we're below the kswapd watermark and have kicked background
3151          * reclaim. Now things get more complex, so set up alloc_flags according
3152          * to how we want to proceed.
3153          */
3154         alloc_flags = gfp_to_alloc_flags(gfp_mask);
3155 
3156         /*
3157          * Find the true preferred zone if the allocation is unconstrained by
3158          * cpusets.
3159          */
3160         if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
3161                 struct zoneref *preferred_zoneref;
3162                 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3163                                 ac->high_zoneidx, NULL, &ac->preferred_zone);
3164                 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
3165         }
3166 
3167         /* This is the last chance, in general, before the goto nopage. */
3168         page = get_page_from_freelist(gfp_mask, order,
3169                                 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3170         if (page)
3171                 goto got_pg;
3172 
3173         /* Allocate without watermarks if the context allows */
3174         if (alloc_flags & ALLOC_NO_WATERMARKS) {
3175                 /*
3176                  * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3177                  * the allocation is high priority and these type of
3178                  * allocations are system rather than user orientated
3179                  */
3180                 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3181                 page = get_page_from_freelist(gfp_mask, order,
3182                                                 ALLOC_NO_WATERMARKS, ac);
3183                 if (page)
3184                         goto got_pg;
3185         }
3186 
3187         /* Caller is not willing to reclaim, we can't balance anything */
3188         if (!can_direct_reclaim) {
3189                 /*
3190                  * All existing users of the __GFP_NOFAIL are blockable, so warn
3191                  * of any new users that actually allow this type of allocation
3192                  * to fail.
3193                  */
3194                 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
3195                 goto nopage;
3196         }
3197 
3198         /* Avoid recursion of direct reclaim */
3199         if (current->flags & PF_MEMALLOC) {
3200                 /*
3201                  * __GFP_NOFAIL request from this context is rather bizarre
3202                  * because we cannot reclaim anything and only can loop waiting
3203                  * for somebody to do a work for us.
3204                  */
3205                 if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3206                         cond_resched();
3207                         goto retry;
3208                 }
3209                 goto nopage;
3210         }
3211 
3212         /* Avoid allocations with no watermarks from looping endlessly */
3213         if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3214                 goto nopage;
3215 
3216         /*
3217          * Try direct compaction. The first pass is asynchronous. Subsequent
3218          * attempts after direct reclaim are synchronous
3219          */
3220         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3221                                         migration_mode,
3222                                         &contended_compaction,
3223                                         &deferred_compaction);
3224         if (page)
3225                 goto got_pg;
3226 
3227         /* Checks for THP-specific high-order allocations */
3228         if (is_thp_gfp_mask(gfp_mask)) {
3229                 /*
3230                  * If compaction is deferred for high-order allocations, it is
3231                  * because sync compaction recently failed. If this is the case
3232                  * and the caller requested a THP allocation, we do not want
3233                  * to heavily disrupt the system, so we fail the allocation
3234                  * instead of entering direct reclaim.
3235                  */
3236                 if (deferred_compaction)
3237                         goto nopage;
3238 
3239                 /*
3240                  * In all zones where compaction was attempted (and not
3241                  * deferred or skipped), lock contention has been detected.
3242                  * For THP allocation we do not want to disrupt the others
3243                  * so we fallback to base pages instead.
3244                  */
3245                 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3246                         goto nopage;
3247 
3248                 /*
3249                  * If compaction was aborted due to need_resched(), we do not
3250                  * want to further increase allocation latency, unless it is
3251                  * khugepaged trying to collapse.
3252                  */
3253                 if (contended_compaction == COMPACT_CONTENDED_SCHED
3254                         && !(current->flags & PF_KTHREAD))
3255                         goto nopage;
3256         }
3257 
3258         /*
3259          * It can become very expensive to allocate transparent hugepages at
3260          * fault, so use asynchronous memory compaction for THP unless it is
3261          * khugepaged trying to collapse.
3262          */
3263         if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD))
3264                 migration_mode = MIGRATE_SYNC_LIGHT;
3265 
3266         /* Try direct reclaim and then allocating */
3267         page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3268                                                         &did_some_progress);
3269         if (page)
3270                 goto got_pg;
3271 
3272         /* Do not loop if specifically requested */
3273         if (gfp_mask & __GFP_NORETRY)
3274                 goto noretry;
3275 
3276         /* Keep reclaiming pages as long as there is reasonable progress */
3277         pages_reclaimed += did_some_progress;
3278         if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3279             ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
3280                 /* Wait for some write requests to complete then retry */
3281                 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
3282                 goto retry;
3283         }
3284 
3285         /* Reclaim has failed us, start killing things */
3286         page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3287         if (page)
3288                 goto got_pg;
3289 
3290         /* Retry as long as the OOM killer is making progress */
3291         if (did_some_progress)
3292                 goto retry;
3293 
3294 noretry:
3295         /*
3296          * High-order allocations do not necessarily loop after
3297          * direct reclaim and reclaim/compaction depends on compaction
3298          * being called after reclaim so call directly if necessary
3299          */
3300         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3301                                             ac, migration_mode,
3302                                             &contended_compaction,
3303                                             &deferred_compaction);
3304         if (page)
3305                 goto got_pg;
3306 nopage:
3307         warn_alloc_failed(gfp_mask, order, NULL);
3308 got_pg:
3309         return page;
3310 }
3311 
3312 /*
3313  * This is the 'heart' of the zoned buddy allocator.
3314  */
3315 struct page *
3316 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3317                         struct zonelist *zonelist, nodemask_t *nodemask)
3318 {
3319         struct zoneref *preferred_zoneref;
3320         struct page *page = NULL;
3321         unsigned int cpuset_mems_cookie;
3322         int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
3323         gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
3324         struct alloc_context ac = {
3325                 .high_zoneidx = gfp_zone(gfp_mask),
3326                 .nodemask = nodemask,
3327                 .migratetype = gfpflags_to_migratetype(gfp_mask),
3328         };
3329 
3330         gfp_mask &= gfp_allowed_mask;
3331 
3332         lockdep_trace_alloc(gfp_mask);
3333 
3334         might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
3335 
3336         if (should_fail_alloc_page(gfp_mask, order))
3337                 return NULL;
3338 
3339         /*
3340          * Check the zones suitable for the gfp_mask contain at least one
3341          * valid zone. It's possible to have an empty zonelist as a result
3342          * of __GFP_THISNODE and a memoryless node
3343          */
3344         if (unlikely(!zonelist->_zonerefs->zone))
3345                 return NULL;
3346 
3347         if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
3348                 alloc_flags |= ALLOC_CMA;
3349 
3350 retry_cpuset:
3351         cpuset_mems_cookie = read_mems_allowed_begin();
3352 
3353         /* We set it here, as __alloc_pages_slowpath might have changed it */
3354         ac.zonelist = zonelist;
3355 
3356         /* Dirty zone balancing only done in the fast path */
3357         ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3358 
3359         /* The preferred zone is used for statistics later */
3360         preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3361                                 ac.nodemask ? : &cpuset_current_mems_allowed,
3362                                 &ac.preferred_zone);
3363         if (!ac.preferred_zone)
3364                 goto out;
3365         ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
3366 
3367         /* First allocation attempt */
3368         alloc_mask = gfp_mask|__GFP_HARDWALL;
3369         page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
3370         if (unlikely(!page)) {
3371                 /*
3372                  * Runtime PM, block IO and its error handling path
3373                  * can deadlock because I/O on the device might not
3374                  * complete.
3375                  */
3376                 alloc_mask = memalloc_noio_flags(gfp_mask);
3377                 ac.spread_dirty_pages = false;
3378 
3379                 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
3380         }
3381 
3382         if (kmemcheck_enabled && page)
3383                 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3384 
3385         trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3386 
3387 out:
3388         /*
3389          * When updating a task's mems_allowed, it is possible to race with
3390          * parallel threads in such a way that an allocation can fail while
3391          * the mask is being updated. If a page allocation is about to fail,
3392          * check if the cpuset changed during allocation and if so, retry.
3393          */
3394         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
3395                 goto retry_cpuset;
3396 
3397         return page;
3398 }
3399 EXPORT_SYMBOL(__alloc_pages_nodemask);
3400 
3401 /*
3402  * Common helper functions.
3403  */
3404 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
3405 {
3406         struct page *page;
3407 
3408         /*
3409          * __get_free_pages() returns a 32-bit address, which cannot represent
3410          * a highmem page
3411          */
3412         VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3413 
3414         page = alloc_pages(gfp_mask, order);
3415         if (!page)
3416                 return 0;
3417         return (unsigned long) page_address(page);
3418 }
3419 EXPORT_SYMBOL(__get_free_pages);
3420 
3421 unsigned long get_zeroed_page(gfp_t gfp_mask)
3422 {
3423         return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
3424 }
3425 EXPORT_SYMBOL(get_zeroed_page);
3426 
3427 void __free_pages(struct page *page, unsigned int order)
3428 {
3429         if (put_page_testzero(page)) {
3430                 if (order == 0)
3431                         free_hot_cold_page(page, false);
3432                 else
3433                         __free_pages_ok(page, order);
3434         }
3435 }
3436 
3437 EXPORT_SYMBOL(__free_pages);
3438 
3439 void free_pages(unsigned long addr, unsigned int order)
3440 {
3441         if (addr != 0) {
3442                 VM_BUG_ON(!virt_addr_valid((void *)addr));
3443                 __free_pages(virt_to_page((void *)addr), order);
3444         }
3445 }
3446 
3447 EXPORT_SYMBOL(free_pages);
3448 
3449 /*
3450  * Page Fragment:
3451  *  An arbitrary-length arbitrary-offset area of memory which resides
3452  *  within a 0 or higher order page.  Multiple fragments within that page
3453  *  are individually refcounted, in the page's reference counter.
3454  *
3455  * The page_frag functions below provide a simple allocation framework for
3456  * page fragments.  This is used by the network stack and network device
3457  * drivers to provide a backing region of memory for use as either an
3458  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3459  */
3460 static struct page *__page_frag_refill(struct page_frag_cache *nc,
3461                                        gfp_t gfp_mask)
3462 {
3463         struct page *page = NULL;
3464         gfp_t gfp = gfp_mask;
3465 
3466 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3467         gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3468                     __GFP_NOMEMALLOC;
3469         page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3470                                 PAGE_FRAG_CACHE_MAX_ORDER);
3471         nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3472 #endif
3473         if (unlikely(!page))
3474                 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3475 
3476         nc->va = page ? page_address(page) : NULL;
3477 
3478         return page;
3479 }
3480 
3481 void *__alloc_page_frag(struct page_frag_cache *nc,
3482                         unsigned int fragsz, gfp_t gfp_mask)
3483 {
3484         unsigned int size = PAGE_SIZE;
3485         struct page *page;
3486         int offset;
3487 
3488         if (unlikely(!nc->va)) {
3489 refill:
3490                 page = __page_frag_refill(nc, gfp_mask);
3491                 if (!page)
3492                         return NULL;
3493 
3494 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3495                 /* if size can vary use size else just use PAGE_SIZE */
3496                 size = nc->size;
3497 #endif
3498                 /* Even if we own the page, we do not use atomic_set().
3499                  * This would break get_page_unless_zero() users.
3500                  */
3501                 page_ref_add(page, size - 1);
3502 
3503                 /* reset page count bias and offset to start of new frag */
3504                 nc->pfmemalloc = page_is_pfmemalloc(page);
3505                 nc->pagecnt_bias = size;
3506                 nc->offset = size;
3507         }
3508 
3509         offset = nc->offset - fragsz;
3510         if (unlikely(offset < 0)) {
3511                 page = virt_to_page(nc->va);
3512 
3513                 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
3514                         goto refill;
3515 
3516 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3517                 /* if size can vary use size else just use PAGE_SIZE */
3518                 size = nc->size;
3519 #endif
3520                 /* OK, page count is 0, we can safely set it */
3521                 set_page_count(page, size);
3522 
3523                 /* reset page count bias and offset to start of new frag */
3524                 nc->pagecnt_bias = size;
3525                 offset = size - fragsz;
3526         }
3527 
3528         nc->pagecnt_bias--;
3529         nc->offset = offset;
3530 
3531         return nc->va + offset;
3532 }
3533 EXPORT_SYMBOL(__alloc_page_frag);
3534 
3535 /*
3536  * Frees a page fragment allocated out of either a compound or order 0 page.
3537  */
3538 void __free_page_frag(void *addr)
3539 {
3540         struct page *page = virt_to_head_page(addr);
3541 
3542         if (unlikely(put_page_testzero(page)))
3543                 __free_pages_ok(page, compound_order(page));
3544 }
3545 EXPORT_SYMBOL(__free_page_frag);
3546 
3547 /*
3548  * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3549  * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3550  * equivalent to alloc_pages.
3551  *
3552  * It should be used when the caller would like to use kmalloc, but since the
3553  * allocation is large, it has to fall back to the page allocator.
3554  */
3555 struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3556 {
3557         struct page *page;
3558 
3559         page = alloc_pages(gfp_mask, order);
3560         if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3561                 __free_pages(page, order);
3562                 page = NULL;
3563         }
3564         return page;
3565 }
3566 
3567 struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3568 {
3569         struct page *page;
3570 
3571         page = alloc_pages_node(nid, gfp_mask, order);
3572         if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3573                 __free_pages(page, order);
3574                 page = NULL;
3575         }
3576         return page;
3577 }
3578 
3579 /*
3580  * __free_kmem_pages and free_kmem_pages will free pages allocated with
3581  * alloc_kmem_pages.
3582  */
3583 void __free_kmem_pages(struct page *page, unsigned int order)
3584 {
3585         memcg_kmem_uncharge(page, order);
3586         __free_pages(page, order);
3587 }
3588 
3589 void free_kmem_pages(unsigned long addr, unsigned int order)
3590 {
3591         if (addr != 0) {
3592                 VM_BUG_ON(!virt_addr_valid((void *)addr));
3593                 __free_kmem_pages(virt_to_page((void *)addr), order);
3594         }
3595 }
3596 
3597 static void *make_alloc_exact(unsigned long addr, unsigned int order,
3598                 size_t size)
3599 {
3600         if (addr) {
3601                 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3602                 unsigned long used = addr + PAGE_ALIGN(size);
3603 
3604                 split_page(virt_to_page((void *)addr), order);
3605                 while (used < alloc_end) {
3606                         free_page(used);
3607                         used += PAGE_SIZE;
3608                 }
3609         }
3610         return (void *)addr;
3611 }
3612 
3613 /**
3614  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3615  * @size: the number of bytes to allocate
3616  * @gfp_mask: GFP flags for the allocation
3617  *
3618  * This function is similar to alloc_pages(), except that it allocates the
3619  * minimum number of pages to satisfy the request.  alloc_pages() can only
3620  * allocate memory in power-of-two pages.
3621  *
3622  * This function is also limited by MAX_ORDER.
3623  *
3624  * Memory allocated by this function must be released by free_pages_exact().
3625  */
3626 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3627 {
3628         unsigned int order = get_order(size);
3629         unsigned long addr;
3630 
3631         addr = __get_free_pages(gfp_mask, order);
3632         return make_alloc_exact(addr, order, size);
3633 }
3634 EXPORT_SYMBOL(alloc_pages_exact);
3635 
3636 /**
3637  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3638  *                         pages on a node.
3639  * @nid: the preferred node ID where memory should be allocated
3640  * @size: the number of bytes to allocate
3641  * @gfp_mask: GFP flags for the allocation
3642  *
3643  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3644  * back.
3645  */
3646 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
3647 {
3648         unsigned int order = get_order(size);
3649         struct page *p = alloc_pages_node(nid, gfp_mask, order);
3650         if (!p)
3651                 return NULL;
3652         return make_alloc_exact((unsigned long)page_address(p), order, size);
3653 }
3654 
3655 /**
3656  * free_pages_exact - release memory allocated via alloc_pages_exact()
3657  * @virt: the value returned by alloc_pages_exact.
3658  * @size: size of allocation, same value as passed to alloc_pages_exact().
3659  *
3660  * Release the memory allocated by a previous call to alloc_pages_exact.
3661  */
3662 void free_pages_exact(void *virt, size_t size)
3663 {
3664         unsigned long addr = (unsigned long)virt;
3665         unsigned long end = addr + PAGE_ALIGN(size);
3666 
3667         while (addr < end) {
3668                 free_page(addr);
3669                 addr += PAGE_SIZE;
3670         }
3671 }
3672 EXPORT_SYMBOL(free_pages_exact);
3673 
3674 /**
3675  * nr_free_zone_pages - count number of pages beyond high watermark
3676  * @offset: The zone index of the highest zone
3677  *
3678  * nr_free_zone_pages() counts the number of counts pages which are beyond the
3679  * high watermark within all zones at or below a given zone index.  For each
3680  * zone, the number of pages is calculated as:
3681  *     managed_pages - high_pages
3682  */
3683 static unsigned long nr_free_zone_pages(int offset)
3684 {
3685         struct zoneref *z;
3686         struct zone *zone;
3687 
3688         /* Just pick one node, since fallback list is circular */
3689         unsigned long sum = 0;
3690 
3691         struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3692 
3693         for_each_zone_zonelist(zone, z, zonelist, offset) {
3694                 unsigned long size = zone->managed_pages;
3695                 unsigned long high = high_wmark_pages(zone);
3696                 if (size > high)
3697                         sum += size - high;
3698         }
3699 
3700         return sum;
3701 }
3702 
3703 /**
3704  * nr_free_buffer_pages - count number of pages beyond high watermark
3705  *
3706  * nr_free_buffer_pages() counts the number of pages which are beyond the high
3707  * watermark within ZONE_DMA and ZONE_NORMAL.
3708  */
3709 unsigned long nr_free_buffer_pages(void)
3710 {
3711         return nr_free_zone_pages(gfp_zone(GFP_USER));
3712 }
3713 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3714 
3715 /**
3716  * nr_free_pagecache_pages - count number of pages beyond high watermark
3717  *
3718  * nr_free_pagecache_pages() counts the number of pages which are beyond the
3719  * high watermark within all zones.
3720  */
3721 unsigned long nr_free_pagecache_pages(void)
3722 {
3723         return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3724 }
3725 
3726 static inline void show_node(struct zone *zone)
3727 {
3728         if (IS_ENABLED(CONFIG_NUMA))
3729                 printk("Node %d ", zone_to_nid(zone));
3730 }
3731 
3732 long si_mem_available(void)
3733 {
3734         long available;
3735         unsigned long pagecache;
3736         unsigned long wmark_low = 0;
3737         unsigned long pages[NR_LRU_LISTS];
3738         struct zone *zone;
3739         int lru;
3740 
3741         for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
3742                 pages[lru] = global_page_state(NR_LRU_BASE + lru);
3743 
3744         for_each_zone(zone)
3745                 wmark_low += zone->watermark[WMARK_LOW];
3746 
3747         /*
3748          * Estimate the amount of memory available for userspace allocations,
3749          * without causing swapping.
3750          */
3751         available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
3752 
3753         /*
3754          * Not all the page cache can be freed, otherwise the system will
3755          * start swapping. Assume at least half of the page cache, or the
3756          * low watermark worth of cache, needs to stay.
3757          */
3758         pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
3759         pagecache -= min(pagecache / 2, wmark_low);
3760         available += pagecache;
3761 
3762         /*
3763          * Part of the reclaimable slab consists of items that are in use,
3764          * and cannot be freed. Cap this estimate at the low watermark.
3765          */
3766         available += global_page_state(NR_SLAB_RECLAIMABLE) -
3767                      min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
3768 
3769         if (available < 0)
3770                 available = 0;
3771         return available;
3772 }
3773 EXPORT_SYMBOL_GPL(si_mem_available);
3774 
3775 void si_meminfo(struct sysinfo *val)
3776 {
3777         val->totalram = totalram_pages;
3778         val->sharedram = global_page_state(NR_SHMEM);
3779         val->freeram = global_page_state(NR_FREE_PAGES);
3780         val->bufferram = nr_blockdev_pages();
3781         val->totalhigh = totalhigh_pages;
3782         val->freehigh = nr_free_highpages();
3783         val->mem_unit = PAGE_SIZE;
3784 }
3785 
3786 EXPORT_SYMBOL(si_meminfo);
3787 
3788 #ifdef CONFIG_NUMA
3789 void si_meminfo_node(struct sysinfo *val, int nid)
3790 {
3791         int zone_type;          /* needs to be signed */
3792         unsigned long managed_pages = 0;
3793         pg_data_t *pgdat = NODE_DATA(nid);
3794 
3795         for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3796                 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3797         val->totalram = managed_pages;
3798         val->sharedram = node_page_state(nid, NR_SHMEM);
3799         val->freeram = node_page_state(nid, NR_FREE_PAGES);
3800 #ifdef CONFIG_HIGHMEM
3801         val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
3802         val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3803                         NR_FREE_PAGES);
3804 #else
3805         val->totalhigh = 0;
3806         val->freehigh = 0;
3807 #endif
3808         val->mem_unit = PAGE_SIZE;
3809 }
3810 #endif
3811 
3812 /*
3813  * Determine whether the node should be displayed or not, depending on whether
3814  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3815  */
3816 bool skip_free_areas_node(unsigned int flags, int nid)
3817 {
3818         bool ret = false;
3819         unsigned int cpuset_mems_cookie;
3820 
3821         if (!(flags & SHOW_MEM_FILTER_NODES))
3822                 goto out;
3823 
3824         do {
3825                 cpuset_mems_cookie = read_mems_allowed_begin();
3826                 ret = !node_isset(nid, cpuset_current_mems_allowed);
3827         } while (read_mems_allowed_retry(cpuset_mems_cookie));
3828 out:
3829         return ret;
3830 }
3831 
3832 #define K(x) ((x) << (PAGE_SHIFT-10))
3833 
3834 static void show_migration_types(unsigned char type)
3835 {
3836         static const char types[MIGRATE_TYPES] = {
3837                 [MIGRATE_UNMOVABLE]     = 'U',
3838                 [MIGRATE_MOVABLE]       = 'M',
3839                 [MIGRATE_RECLAIMABLE]   = 'E',
3840                 [MIGRATE_HIGHATOMIC]    = 'H',
3841 #ifdef CONFIG_CMA
3842                 [MIGRATE_CMA]           = 'C',
3843 #endif
3844 #ifdef CONFIG_MEMORY_ISOLATION
3845                 [MIGRATE_ISOLATE]       = 'I',
3846 #endif
3847         };
3848         char tmp[MIGRATE_TYPES + 1];
3849         char *p = tmp;
3850         int i;
3851 
3852         for (i = 0; i < MIGRATE_TYPES; i++) {
3853                 if (type & (1 << i))
3854                         *p++ = types[i];
3855         }
3856 
3857         *p = '\0';
3858         printk("(%s) ", tmp);
3859 }
3860 
3861 /*
3862  * Show free area list (used inside shift_scroll-lock stuff)
3863  * We also calculate the percentage fragmentation. We do this by counting the
3864  * memory on each free list with the exception of the first item on the list.
3865  *
3866  * Bits in @filter:
3867  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3868  *   cpuset.
3869  */
3870 void show_free_areas(unsigned int filter)
3871 {
3872         unsigned long free_pcp = 0;
3873         int cpu;
3874         struct zone *zone;
3875 
3876         for_each_populated_zone(zone) {
3877                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3878                         continue;
3879 
3880                 for_each_online_cpu(cpu)
3881                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3882         }
3883 
3884         printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3885                 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3886                 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3887                 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3888                 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3889                 " free:%lu free_pcp:%lu free_cma:%lu\n",
3890                 global_page_state(NR_ACTIVE_ANON),
3891                 global_page_state(NR_INACTIVE_ANON),
3892                 global_page_state(NR_ISOLATED_ANON),
3893                 global_page_state(NR_ACTIVE_FILE),
3894                 global_page_state(NR_INACTIVE_FILE),
3895                 global_page_state(NR_ISOLATED_FILE),
3896                 global_page_state(NR_UNEVICTABLE),
3897                 global_page_state(NR_FILE_DIRTY),
3898                 global_page_state(NR_WRITEBACK),
3899                 global_page_state(NR_UNSTABLE_NFS),
3900                 global_page_state(NR_SLAB_RECLAIMABLE),
3901                 global_page_state(NR_SLAB_UNRECLAIMABLE),
3902                 global_page_state(NR_FILE_MAPPED),
3903                 global_page_state(NR_SHMEM),
3904                 global_page_state(NR_PAGETABLE),
3905                 global_page_state(NR_BOUNCE),
3906                 global_page_state(NR_FREE_PAGES),
3907                 free_pcp,
3908                 global_page_state(NR_FREE_CMA_PAGES));
3909 
3910         for_each_populated_zone(zone) {
3911                 int i;
3912 
3913                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3914                         continue;
3915 
3916                 free_pcp = 0;
3917                 for_each_online_cpu(cpu)
3918                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3919 
3920                 show_node(zone);
3921                 printk("%s"
3922                         " free:%lukB"
3923                         " min:%lukB"
3924                         " low:%lukB"
3925                         " high:%lukB"
3926                         " active_anon:%lukB"
3927                         " inactive_anon:%lukB"
3928                         " active_file:%lukB"
3929                         " inactive_file:%lukB"
3930                         " unevictable:%lukB"
3931                         " isolated(anon):%lukB"
3932                         " isolated(file):%lukB"
3933                         " present:%lukB"
3934                         " managed:%lukB"
3935                         " mlocked:%lukB"
3936                         " dirty:%lukB"
3937                         " writeback:%lukB"
3938                         " mapped:%lukB"
3939                         " shmem:%lukB"
3940                         " slab_reclaimable:%lukB"
3941                         " slab_unreclaimable:%lukB"
3942                         " kernel_stack:%lukB"
3943                         " pagetables:%lukB"
3944                         " unstable:%lukB"
3945                         " bounce:%lukB"
3946                         " free_pcp:%lukB"
3947                         " local_pcp:%ukB"
3948                         " free_cma:%lukB"
3949                         " writeback_tmp:%lukB"
3950                         " pages_scanned:%lu"
3951                         " all_unreclaimable? %s"
3952                         "\n",
3953                         zone->name,
3954                         K(zone_page_state(zone, NR_FREE_PAGES)),
3955                         K(min_wmark_pages(zone)),
3956                         K(low_wmark_pages(zone)),
3957                         K(high_wmark_pages(zone)),
3958                         K(zone_page_state(zone, NR_ACTIVE_ANON)),
3959                         K(zone_page_state(zone, NR_INACTIVE_ANON)),
3960                         K(zone_page_state(zone, NR_ACTIVE_FILE)),
3961                         K(zone_page_state(zone, NR_INACTIVE_FILE)),
3962                         K(zone_page_state(zone, NR_UNEVICTABLE)),
3963                         K(zone_page_state(zone, NR_ISOLATED_ANON)),
3964                         K(zone_page_state(zone, NR_ISOLATED_FILE)),
3965                         K(zone->present_pages),
3966                         K(zone->managed_pages),
3967                         K(zone_page_state(zone, NR_MLOCK)),
3968                         K(zone_page_state(zone, NR_FILE_DIRTY)),
3969                         K(zone_page_state(zone, NR_WRITEBACK)),
3970                         K(zone_page_state(zone, NR_FILE_MAPPED)),
3971                         K(zone_page_state(zone, NR_SHMEM)),
3972                         K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3973                         K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3974                         zone_page_state(zone, NR_KERNEL_STACK) *
3975                                 THREAD_SIZE / 1024,
3976                         K(zone_page_state(zone, NR_PAGETABLE)),
3977                         K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3978                         K(zone_page_state(zone, NR_BOUNCE)),
3979                         K(free_pcp),
3980                         K(this_cpu_read(zone->pageset->pcp.count)),
3981                         K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3982                         K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3983                         K(zone_page_state(zone, NR_PAGES_SCANNED)),
3984                         (!zone_reclaimable(zone) ? "yes" : "no")
3985                         );
3986                 printk("lowmem_reserve[]:");
3987                 for (i = 0; i < MAX_NR_ZONES; i++)
3988                         printk(" %ld", zone->lowmem_reserve[i]);
3989                 printk("\n");
3990         }
3991 
3992         for_each_populated_zone(zone) {
3993                 unsigned int order;
3994                 unsigned long nr[MAX_ORDER], flags, total = 0;
3995                 unsigned char types[MAX_ORDER];
3996 
3997                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3998                         continue;
3999                 show_node(zone);
4000                 printk("%s: ", zone->name);
4001 
4002                 spin_lock_irqsave(&zone->lock, flags);
4003                 for (order = 0; order < MAX_ORDER; order++) {
4004                         struct free_area *area = &zone->free_area[order];
4005                         int type;
4006 
4007                         nr[order] = area->nr_free;
4008                         total += nr[order] << order;
4009 
4010                         types[order] = 0;
4011                         for (type = 0; type < MIGRATE_TYPES; type++) {
4012                                 if (!list_empty(&area->free_list[type]))
4013                                         types[order] |= 1 << type;
4014                         }
4015                 }
4016                 spin_unlock_irqrestore(&zone->lock, flags);
4017                 for (order = 0; order < MAX_ORDER; order++) {
4018                         printk("%lu*%lukB ", nr[order], K(1UL) << order);
4019                         if (nr[order])
4020                                 show_migration_types(types[order]);
4021                 }
4022                 printk("= %lukB\n", K(total));
4023         }
4024 
4025         hugetlb_show_meminfo();
4026 
4027         printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
4028 
4029         show_swap_cache_info();
4030 }
4031 
4032 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4033 {
4034         zoneref->zone = zone;
4035         zoneref->zone_idx = zone_idx(zone);
4036 }
4037 
4038 /*
4039  * Builds allocation fallback zone lists.
4040  *
4041  * Add all populated zones of a node to the zonelist.
4042  */
4043 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
4044                                 int nr_zones)
4045 {
4046         struct zone *zone;
4047         enum zone_type zone_type = MAX_NR_ZONES;
4048 
4049         do {
4050                 zone_type--;
4051                 zone = pgdat->node_zones + zone_type;
4052                 if (populated_zone(zone)) {
4053                         zoneref_set_zone(zone,
4054                                 &zonelist->_zonerefs[nr_zones++]);
4055                         check_highest_zone(zone_type);
4056                 }
4057         } while (zone_type);
4058 
4059         return nr_zones;
4060 }
4061 
4062 
4063 /*
4064  *  zonelist_order:
4065  *  0 = automatic detection of better ordering.
4066  *  1 = order by ([node] distance, -zonetype)
4067  *  2 = order by (-zonetype, [node] distance)
4068  *
4069  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4070  *  the same zonelist. So only NUMA can configure this param.
4071  */
4072 #define ZONELIST_ORDER_DEFAULT  0
4073 #define ZONELIST_ORDER_NODE     1
4074 #define ZONELIST_ORDER_ZONE     2
4075 
4076 /* zonelist order in the kernel.
4077  * set_zonelist_order() will set this to NODE or ZONE.
4078  */
4079 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4080 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4081 
4082 
4083 #ifdef CONFIG_NUMA
4084 /* The value user specified ....changed by config */
4085 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4086 /* string for sysctl */
4087 #define NUMA_ZONELIST_ORDER_LEN 16
4088 char numa_zonelist_order[16] = "default";
4089 
4090 /*
4091  * interface for configure zonelist ordering.
4092  * command line option "numa_zonelist_order"
4093  *      = "[dD]efault   - default, automatic configuration.
4094  *      = "[nN]ode      - order by node locality, then by zone within node
4095  *      = "[zZ]one      - order by zone, then by locality within zone
4096  */
4097 
4098 static int __parse_numa_zonelist_order(char *s)
4099 {
4100         if (*s == 'd' || *s == 'D') {
4101                 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4102         } else if (*s == 'n' || *s == 'N') {
4103                 user_zonelist_order = ZONELIST_ORDER_NODE;
4104         } else if (*s == 'z' || *s == 'Z') {
4105                 user_zonelist_order = ZONELIST_ORDER_ZONE;
4106         } else {
4107                 pr_warn("Ignoring invalid numa_zonelist_order value:  %s\n", s);
4108                 return -EINVAL;
4109         }
4110         return 0;
4111 }
4112 
4113 static __init int setup_numa_zonelist_order(char *s)
4114 {
4115         int ret;
4116 
4117         if (!s)
4118                 return 0;
4119 
4120         ret = __parse_numa_zonelist_order(s);
4121         if (ret == 0)
4122                 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4123 
4124         return ret;
4125 }
4126 early_param("numa_zonelist_order", setup_numa_zonelist_order);
4127 
4128 /*
4129  * sysctl handler for numa_zonelist_order
4130  */
4131 int numa_zonelist_order_handler(struct ctl_table *table, int write,
4132                 void __user *buffer, size_t *length,
4133                 loff_t *ppos)
4134 {
4135         char saved_string[NUMA_ZONELIST_ORDER_LEN];
4136         int ret;
4137         static DEFINE_MUTEX(zl_order_mutex);
4138 
4139         mutex_lock(&zl_order_mutex);
4140         if (write) {
4141                 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4142                         ret = -EINVAL;
4143                         goto out;
4144                 }
4145                 strcpy(saved_string, (char *)table->data);
4146         }
4147         ret = proc_dostring(table, write, buffer, length, ppos);
4148         if (ret)
4149                 goto out;
4150         if (write) {
4151                 int oldval = user_zonelist_order;
4152 
4153                 ret = __parse_numa_zonelist_order((char *)table->data);
4154                 if (ret) {
4155                         /*
4156                          * bogus value.  restore saved string
4157                          */
4158                         strncpy((char *)table->data, saved_string,
4159                                 NUMA_ZONELIST_ORDER_LEN);
4160                         user_zonelist_order = oldval;
4161                 } else if (oldval != user_zonelist_order) {
4162                         mutex_lock(&zonelists_mutex);
4163                         build_all_zonelists(NULL, NULL);
4164                         mutex_unlock(&zonelists_mutex);
4165                 }
4166         }
4167 out:
4168         mutex_unlock(&zl_order_mutex);
4169         return ret;
4170 }
4171 
4172 
4173 #define MAX_NODE_LOAD (nr_online_nodes)
4174 static int node_load[MAX_NUMNODES];
4175 
4176 /**
4177  * find_next_best_node - find the next node that should appear in a given node's fallback list
4178  * @node: node whose fallback list we're appending
4179  * @used_node_mask: nodemask_t of already used nodes
4180  *
4181  * We use a number of factors to determine which is the next node that should
4182  * appear on a given node's fallback list.  The node should not have appeared
4183  * already in @node's fallback list, and it should be the next closest node
4184  * according to the distance array (which contains arbitrary distance values
4185  * from each node to each node in the system), and should also prefer nodes
4186  * with no CPUs, since presumably they'll have very little allocation pressure
4187  * on them otherwise.
4188  * It returns -1 if no node is found.
4189  */
4190 static int find_next_best_node(int node, nodemask_t *used_node_mask)
4191 {
4192         int n, val;
4193         int min_val = INT_MAX;
4194         int best_node = NUMA_NO_NODE;
4195         const struct cpumask *tmp = cpumask_of_node(0);
4196 
4197         /* Use the local node if we haven't already */
4198         if (!node_isset(node, *used_node_mask)) {
4199                 node_set(node, *used_node_mask);
4200                 return node;
4201         }
4202 
4203         for_each_node_state(n, N_MEMORY) {
4204 
4205                 /* Don't want a node to appear more than once */
4206                 if (node_isset(n, *used_node_mask))
4207                         continue;
4208 
4209                 /* Use the distance array to find the distance */
4210                 val = node_distance(node, n);
4211 
4212                 /* Penalize nodes under us ("prefer the next node") */
4213                 val += (n < node);
4214 
4215                 /* Give preference to headless and unused nodes */
4216                 tmp = cpumask_of_node(n);
4217                 if (!cpumask_empty(tmp))
4218                         val += PENALTY_FOR_NODE_WITH_CPUS;
4219 
4220                 /* Slight preference for less loaded node */
4221                 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4222                 val += node_load[n];
4223 
4224                 if (val < min_val) {
4225                         min_val = val;
4226                         best_node = n;
4227                 }
4228         }
4229 
4230         if (best_node >= 0)
4231                 node_set(best_node, *used_node_mask);
4232 
4233         return best_node;
4234 }
4235 
4236 
4237 /*
4238  * Build zonelists ordered by node and zones within node.
4239  * This results in maximum locality--normal zone overflows into local
4240  * DMA zone, if any--but risks exhausting DMA zone.
4241  */
4242 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
4243 {
4244         int j;
4245         struct zonelist *zonelist;
4246 
4247         zonelist = &pgdat->node_zonelists[0];
4248         for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
4249                 ;
4250         j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4251         zonelist->_zonerefs[j].zone = NULL;
4252         zonelist->_zonerefs[j].zone_idx = 0;
4253 }
4254 
4255 /*
4256  * Build gfp_thisnode zonelists
4257  */
4258 static void build_thisnode_zonelists(pg_data_t *pgdat)
4259 {
4260         int j;
4261         struct zonelist *zonelist;
4262 
4263         zonelist = &pgdat->node_zonelists[1];
4264         j = build_zonelists_node(pgdat, zonelist, 0);
4265         zonelist->_zonerefs[j].zone = NULL;
4266         zonelist->_zonerefs[j].zone_idx = 0;
4267 }
4268 
4269 /*
4270  * Build zonelists ordered by zone and nodes within zones.
4271  * This results in conserving DMA zone[s] until all Normal memory is
4272  * exhausted, but results in overflowing to remote node while memory
4273  * may still exist in local DMA zone.
4274  */
4275 static int node_order[MAX_NUMNODES];
4276 
4277 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4278 {
4279         int pos, j, node;
4280         int zone_type;          /* needs to be signed */
4281         struct zone *z;
4282         struct zonelist *zonelist;
4283 
4284         zonelist = &pgdat->node_zonelists[0];
4285         pos = 0;
4286         for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4287                 for (j = 0; j < nr_nodes; j++) {
4288                         node = node_order[j];
4289                         z = &NODE_DATA(node)->node_zones[zone_type];
4290                         if (populated_zone(z)) {
4291                                 zoneref_set_zone(z,
4292                                         &zonelist->_zonerefs[pos++]);
4293                                 check_highest_zone(zone_type);
4294                         }
4295                 }
4296         }
4297         zonelist->_zonerefs[pos].zone = NULL;
4298         zonelist->_zonerefs[pos].zone_idx = 0;
4299 }
4300 
4301 #if defined(CONFIG_64BIT)
4302 /*
4303  * Devices that require DMA32/DMA are relatively rare and do not justify a
4304  * penalty to every machine in case the specialised case applies. Default
4305  * to Node-ordering on 64-bit NUMA machines
4306  */
4307 static int default_zonelist_order(void)
4308 {
4309         return ZONELIST_ORDER_NODE;
4310 }
4311 #else
4312 /*
4313  * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4314  * by the kernel. If processes running on node 0 deplete the low memory zone
4315  * then reclaim will occur more frequency increasing stalls and potentially
4316  * be easier to OOM if a large percentage of the zone is under writeback or
4317  * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4318  * Hence, default to zone ordering on 32-bit.
4319  */
4320 static int default_zonelist_order(void)
4321 {
4322         return ZONELIST_ORDER_ZONE;
4323 }
4324 #endif /* CONFIG_64BIT */
4325 
4326 static void set_zonelist_order(void)
4327 {
4328         if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4329                 current_zonelist_order = default_zonelist_order();
4330         else
4331                 current_zonelist_order = user_zonelist_order;
4332 }
4333 
4334 static void build_zonelists(pg_data_t *pgdat)
4335 {
4336         int i, node, load;
4337         nodemask_t used_mask;
4338         int local_node, prev_node;
4339         struct zonelist *zonelist;
4340         unsigned int order = current_zonelist_order;
4341 
4342         /* initialize zonelists */
4343         for (i = 0; i < MAX_ZONELISTS; i++) {
4344                 zonelist = pgdat->node_zonelists + i;
4345                 zonelist->_zonerefs[0].zone = NULL;
4346                 zonelist->_zonerefs[0].zone_idx = 0;
4347         }
4348 
4349         /* NUMA-aware ordering of nodes */
4350         local_node = pgdat->node_id;
4351         load = nr_online_nodes;
4352         prev_node = local_node;
4353         nodes_clear(used_mask);
4354 
4355         memset(node_order, 0, sizeof(node_order));
4356         i = 0;
4357 
4358         while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4359                 /*
4360                  * We don't want to pressure a particular node.
4361                  * So adding penalty to the first node in same
4362                  * distance group to make it round-robin.
4363                  */
4364                 if (node_distance(local_node, node) !=
4365                     node_distance(local_node, prev_node))
4366                         node_load[node] = load;
4367 
4368                 prev_node = node;
4369                 load--;
4370                 if (order == ZONELIST_ORDER_NODE)
4371                         build_zonelists_in_node_order(pgdat, node);
4372                 else
4373                         node_order[i++] = node; /* remember order */
4374         }
4375 
4376         if (order == ZONELIST_ORDER_ZONE) {
4377                 /* calculate node order -- i.e., DMA last! */
4378                 build_zonelists_in_zone_order(pgdat, i);
4379         }
4380 
4381         build_thisnode_zonelists(pgdat);
4382 }
4383 
4384 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4385 /*
4386  * Return node id of node used for "local" allocations.
4387  * I.e., first node id of first zone in arg node's generic zonelist.
4388  * Used for initializing percpu 'numa_mem', which is used primarily
4389  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4390  */
4391 int local_memory_node(int node)
4392 {
4393         struct zone *zone;
4394 
4395         (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4396                                    gfp_zone(GFP_KERNEL),
4397                                    NULL,
4398                                    &zone);
4399         return zone->node;
4400 }
4401 #endif
4402 
4403 #else   /* CONFIG_NUMA */
4404 
4405 static void set_zonelist_order(void)
4406 {
4407         current_zonelist_order = ZONELIST_ORDER_ZONE;
4408 }
4409 
4410 static void build_zonelists(pg_data_t *pgdat)
4411 {
4412         int node, local_node;
4413         enum zone_type j;
4414         struct zonelist *zonelist;
4415 
4416         local_node = pgdat->node_id;
4417 
4418         zonelist = &pgdat->node_zonelists[0];
4419         j = build_zonelists_node(pgdat, zonelist, 0);
4420 
4421         /*
4422          * Now we build the zonelist so that it contains the zones
4423          * of all the other nodes.
4424          * We don't want to pressure a particular node, so when
4425          * building the zones for node N, we make sure that the
4426          * zones coming right after the local ones are those from
4427          * node N+1 (modulo N)
4428          */
4429         for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4430                 if (!node_online(node))
4431                         continue;
4432                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4433         }
4434         for (node = 0; node < local_node; node++) {
4435                 if (!node_online(node))
4436                         continue;
4437                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4438         }
4439 
4440         zonelist->_zonerefs[j].zone = NULL;
4441         zonelist->_zonerefs[j].zone_idx = 0;
4442 }
4443 
4444 #endif  /* CONFIG_NUMA */
4445 
4446 /*
4447  * Boot pageset table. One per cpu which is going to be used for all
4448  * zones and all nodes. The parameters will be set in such a way
4449  * that an item put on a list will immediately be handed over to
4450  * the buddy list. This is safe since pageset manipulation is done
4451  * with interrupts disabled.
4452  *
4453  * The boot_pagesets must be kept even after bootup is complete for
4454  * unused processors and/or zones. They do play a role for bootstrapping
4455  * hotplugged processors.
4456  *
4457  * zoneinfo_show() and maybe other functions do
4458  * not check if the processor is online before following the pageset pointer.
4459  * Other parts of the kernel may not check if the zone is available.
4460  */
4461 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4462 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
4463 static void setup_zone_pageset(struct zone *zone);
4464 
4465 /*
4466  * Global mutex to protect against size modification of zonelists
4467  * as well as to serialize pageset setup for the new populated zone.
4468  */
4469 DEFINE_MUTEX(zonelists_mutex);
4470 
4471 /* return values int ....just for stop_machine() */
4472 static int __build_all_zonelists(void *data)
4473 {
4474         int nid;
4475         int cpu;
4476         pg_data_t *self = data;
4477 
4478 #ifdef CONFIG_NUMA
4479         memset(node_load, 0, sizeof(node_load));
4480 #endif
4481 
4482         if (self && !node_online(self->node_id)) {
4483                 build_zonelists(self);
4484         }
4485 
4486         for_each_online_node(nid) {
4487                 pg_data_t *pgdat = NODE_DATA(nid);
4488 
4489                 build_zonelists(pgdat);
4490         }
4491 
4492         /*
4493          * Initialize the boot_pagesets that are going to be used
4494          * for bootstrapping processors. The real pagesets for
4495          * each zone will be allocated later when the per cpu
4496          * allocator is available.
4497          *
4498          * boot_pagesets are used also for bootstrapping offline
4499          * cpus if the system is already booted because the pagesets
4500          * are needed to initialize allocators on a specific cpu too.
4501          * F.e. the percpu allocator needs the page allocator which
4502          * needs the percpu allocator in order to allocate its pagesets
4503          * (a chicken-egg dilemma).
4504          */
4505         for_each_possible_cpu(cpu) {
4506                 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4507 
4508 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4509                 /*
4510                  * We now know the "local memory node" for each node--
4511                  * i.e., the node of the first zone in the generic zonelist.
4512                  * Set up numa_mem percpu variable for on-line cpus.  During
4513                  * boot, only the boot cpu should be on-line;  we'll init the
4514                  * secondary cpus' numa_mem as they come on-line.  During
4515                  * node/memory hotplug, we'll fixup all on-line cpus.
4516                  */
4517                 if (cpu_online(cpu))
4518                         set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4519 #endif
4520         }
4521 
4522         return 0;
4523 }
4524 
4525 static noinline void __init
4526 build_all_zonelists_init(void)
4527 {
4528         __build_all_zonelists(NULL);
4529         mminit_verify_zonelist();
4530         cpuset_init_current_mems_allowed();
4531 }
4532 
4533 /*
4534  * Called with zonelists_mutex held always
4535  * unless system_state == SYSTEM_BOOTING.
4536  *
4537  * __ref due to (1) call of __meminit annotated setup_zone_pageset
4538  * [we're only called with non-NULL zone through __meminit paths] and
4539  * (2) call of __init annotated helper build_all_zonelists_init
4540  * [protected by SYSTEM_BOOTING].
4541  */
4542 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
4543 {
4544         set_zonelist_order();
4545 
4546         if (system_state == SYSTEM_BOOTING) {
4547                 build_all_zonelists_init();
4548         } else {
4549 #ifdef CONFIG_MEMORY_HOTPLUG
4550                 if (zone)
4551                         setup_zone_pageset(zone);
4552 #endif
4553                 /* we have to stop all cpus to guarantee there is no user
4554                    of zonelist */
4555                 stop_machine(__build_all_zonelists, pgdat, NULL);
4556                 /* cpuset refresh routine should be here */
4557         }
4558         vm_total_pages = nr_free_pagecache_pages();
4559         /*
4560          * Disable grouping by mobility if the number of pages in the
4561          * system is too low to allow the mechanism to work. It would be
4562          * more accurate, but expensive to check per-zone. This check is
4563          * made on memory-hotadd so a system can start with mobility
4564          * disabled and enable it later
4565          */
4566         if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
4567                 page_group_by_mobility_disabled = 1;
4568         else
4569                 page_group_by_mobility_disabled = 0;
4570 
4571         pr_info("Built %i zonelists in %s order, mobility grouping %s.  Total pages: %ld\n",
4572                 nr_online_nodes,
4573                 zonelist_order_name[current_zonelist_order],
4574                 page_group_by_mobility_disabled ? "off" : "on",
4575                 vm_total_pages);
4576 #ifdef CONFIG_NUMA
4577         pr_info("Policy zone: %s\n", zone_names[policy_zone]);
4578 #endif
4579 }
4580 
4581 /*
4582  * Helper functions to size the waitqueue hash table.
4583  * Essentially these want to choose hash table sizes sufficiently
4584  * large so that collisions trying to wait on pages are rare.
4585  * But in fact, the number of active page waitqueues on typical
4586  * systems is ridiculously low, less than 200. So this is even
4587  * conservative, even though it seems large.
4588  *
4589  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4590  * waitqueues, i.e. the size of the waitq table given the number of pages.
4591  */
4592 #define PAGES_PER_WAITQUEUE     256
4593 
4594 #ifndef CONFIG_MEMORY_HOTPLUG
4595 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4596 {
4597         unsigned long size = 1;
4598 
4599         pages /= PAGES_PER_WAITQUEUE;
4600 
4601         while (size < pages)
4602                 size <<= 1;
4603 
4604         /*
4605          * Once we have dozens or even hundreds of threads sleeping
4606          * on IO we've got bigger problems than wait queue collision.
4607          * Limit the size of the wait table to a reasonable size.
4608          */
4609         size = min(size, 4096UL);
4610 
4611         return max(size, 4UL);
4612 }
4613 #else
4614 /*
4615  * A zone's size might be changed by hot-add, so it is not possible to determine
4616  * a suitable size for its wait_table.  So we use the maximum size now.
4617  *
4618  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
4619  *
4620  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
4621  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4622  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
4623  *
4624  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4625  * or more by the traditional way. (See above).  It equals:
4626  *
4627  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
4628  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
4629  *    powerpc (64K page size)             : =  (32G +16M)byte.
4630  */
4631 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4632 {
4633         return 4096UL;
4634 }
4635 #endif
4636 
4637 /*
4638  * This is an integer logarithm so that shifts can be used later
4639  * to extract the more random high bits from the multiplicative
4640  * hash function before the remainder is taken.
4641  */
4642 static inline unsigned long wait_table_bits(unsigned long size)
4643 {
4644         return ffz(~size);
4645 }
4646 
4647 /*
4648  * Initially all pages are reserved - free ones are freed
4649  * up by free_all_bootmem() once the early boot process is
4650  * done. Non-atomic initialization, single-pass.
4651  */
4652 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4653                 unsigned long start_pfn, enum memmap_context context)
4654 {
4655         struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
4656         unsigned long end_pfn = start_pfn + size;
4657         pg_data_t *pgdat = NODE_DATA(nid);
4658         unsigned long pfn;
4659         unsigned long nr_initialised = 0;
4660 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4661         struct memblock_region *r = NULL, *tmp;
4662 #endif
4663 
4664         if (highest_memmap_pfn < end_pfn - 1)
4665                 highest_memmap_pfn = end_pfn - 1;
4666 
4667         /*
4668          * Honor reservation requested by the driver for this ZONE_DEVICE
4669          * memory
4670          */
4671         if (altmap && start_pfn == altmap->base_pfn)
4672                 start_pfn += altmap->reserve;
4673 
4674         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4675                 /*
4676                  * There can be holes in boot-time mem_map[]s handed to this
4677                  * function.  They do not exist on hotplugged memory.
4678                  */
4679                 if (context != MEMMAP_EARLY)
4680                         goto not_early;
4681 
4682                 if (!early_pfn_valid(pfn))
4683                         continue;
4684                 if (!early_pfn_in_nid(pfn, nid))
4685                         continue;
4686                 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
4687                         break;
4688 
4689 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4690                 /*
4691                  * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
4692                  * from zone_movable_pfn[nid] to end of each node should be
4693                  * ZONE_MOVABLE not ZONE_NORMAL. skip it.
4694                  */
4695                 if (!mirrored_kernelcore && zone_movable_pfn[nid])
4696                         if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid])
4697                                 continue;
4698 
4699                 /*
4700                  * Check given memblock attribute by firmware which can affect
4701                  * kernel memory layout.  If zone==ZONE_MOVABLE but memory is
4702                  * mirrored, it's an overlapped memmap init. skip it.
4703                  */
4704                 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
4705                         if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
4706                                 for_each_memblock(memory, tmp)
4707                                         if (pfn < memblock_region_memory_end_pfn(tmp))
4708                                                 break;
4709                                 r = tmp;
4710                         }
4711                         if (pfn >= memblock_region_memory_base_pfn(r) &&
4712                             memblock_is_mirror(r)) {
4713                                 /* already initialized as NORMAL */
4714                                 pfn = memblock_region_memory_end_pfn(r);
4715                                 continue;
4716                         }
4717                 }
4718 #endif
4719 
4720 not_early:
4721                 /*
4722                  * Mark the block movable so that blocks are reserved for
4723                  * movable at startup. This will force kernel allocations
4724                  * to reserve their blocks rather than leaking throughout
4725                  * the address space during boot when many long-lived
4726                  * kernel allocations are made.
4727                  *
4728                  * bitmap is created for zone's valid pfn range. but memmap
4729                  * can be created for invalid pages (for alignment)
4730                  * check here not to call set_pageblock_migratetype() against
4731                  * pfn out of zone.
4732                  */
4733                 if (!(pfn & (pageblock_nr_pages - 1))) {
4734                         struct page *page = pfn_to_page(pfn);
4735 
4736                         __init_single_page(page, pfn, zone, nid);
4737                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4738                 } else {
4739                         __init_single_pfn(pfn, zone, nid);
4740                 }
4741         }
4742 }
4743 
4744 static void __meminit zone_init_free_lists(struct zone *zone)
4745 {
4746         unsigned int order, t;
4747         for_each_migratetype_order(order, t) {
4748                 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4749                 zone->free_area[order].nr_free = 0;
4750         }
4751 }
4752 
4753 #ifndef __HAVE_ARCH_MEMMAP_INIT
4754 #define memmap_init(size, nid, zone, start_pfn) \
4755         memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4756 #endif
4757 
4758 static int zone_batchsize(struct zone *zone)
4759 {
4760 #ifdef CONFIG_MMU
4761         int batch;
4762 
4763         /*
4764          * The per-cpu-pages pools are set to around 1000th of the
4765          * size of the zone.  But no more than 1/2 of a meg.
4766          *
4767          * OK, so we don't know how big the cache is.  So guess.
4768          */
4769         batch = zone->managed_pages / 1024;
4770         if (batch * PAGE_SIZE > 512 * 1024)
4771                 batch = (512 * 1024) / PAGE_SIZE;
4772         batch /= 4;             /* We effectively *= 4 below */
4773         if (batch < 1)
4774                 batch = 1;
4775 
4776         /*
4777          * Clamp the batch to a 2^n - 1 value. Having a power
4778          * of 2 value was found to be more likely to have
4779          * suboptimal cache aliasing properties in some cases.
4780          *
4781          * For example if 2 tasks are alternately allocating
4782          * batches of pages, one task can end up with a lot
4783          * of pages of one half of the possible page colors
4784          * and the other with pages of the other colors.
4785          */
4786         batch = rounddown_pow_of_two(batch + batch/2) - 1;
4787 
4788         return batch;
4789 
4790 #else
4791         /* The deferral and batching of frees should be suppressed under NOMMU
4792          * conditions.
4793          *
4794          * The problem is that NOMMU needs to be able to allocate large chunks
4795          * of contiguous memory as there's no hardware page translation to
4796          * assemble apparent contiguous memory from discontiguous pages.
4797          *
4798          * Queueing large contiguous runs of pages for batching, however,
4799          * causes the pages to actually be freed in smaller chunks.  As there
4800          * can be a significant delay between the individual batches being
4801          * recycled, this leads to the once large chunks of space being
4802          * fragmented and becoming unavailable for high-order allocations.
4803          */
4804         return 0;
4805 #endif
4806 }
4807 
4808 /*
4809  * pcp->high and pcp->batch values are related and dependent on one another:
4810  * ->batch must never be higher then ->high.
4811  * The following function updates them in a safe manner without read side
4812  * locking.
4813  *
4814  * Any new users of pcp->batch and pcp->high should ensure they can cope with
4815  * those fields changing asynchronously (acording the the above rule).
4816  *
4817  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4818  * outside of boot time (or some other assurance that no concurrent updaters
4819  * exist).
4820  */
4821 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4822                 unsigned long batch)
4823 {
4824        /* start with a fail safe value for batch */
4825         pcp->batch = 1;
4826         smp_wmb();
4827 
4828        /* Update high, then batch, in order */
4829         pcp->high = high;
4830         smp_wmb();
4831 
4832         pcp->batch = batch;
4833 }
4834 
4835 /* a companion to pageset_set_high() */
4836 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4837 {
4838         pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4839 }
4840 
4841 static void pageset_init(struct per_cpu_pageset *p)
4842 {
4843         struct per_cpu_pages *pcp;
4844         int migratetype;
4845 
4846         memset(p, 0, sizeof(*p));
4847 
4848         pcp = &p->pcp;
4849         pcp->count = 0;
4850         for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4851                 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4852 }
4853 
4854 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4855 {
4856         pageset_init(p);
4857         pageset_set_batch(p, batch);
4858 }
4859 
4860 /*
4861  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4862  * to the value high for the pageset p.
4863  */
4864 static void pageset_set_high(struct per_cpu_pageset *p,
4865                                 unsigned long high)
4866 {
4867         unsigned long batch = max(1UL, high / 4);
4868         if ((high / 4) > (PAGE_SHIFT * 8))
4869                 batch = PAGE_SHIFT * 8;
4870 
4871         pageset_update(&p->pcp, high, batch);
4872 }
4873 
4874 static void pageset_set_high_and_batch(struct zone *zone,
4875                                        struct per_cpu_pageset *pcp)
4876 {
4877         if (percpu_pagelist_fraction)
4878                 pageset_set_high(pcp,
4879                         (zone->managed_pages /
4880                                 percpu_pagelist_fraction));
4881         else
4882                 pageset_set_batch(pcp, zone_batchsize(zone));
4883 }
4884 
4885 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4886 {
4887         struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4888 
4889         pageset_init(pcp);
4890         pageset_set_high_and_batch(zone, pcp);
4891 }
4892 
4893 static void __meminit setup_zone_pageset(struct zone *zone)
4894 {
4895         int cpu;
4896         zone->pageset = alloc_percpu(struct per_cpu_pageset);
4897         for_each_possible_cpu(cpu)
4898                 zone_pageset_init(zone, cpu);
4899 }
4900 
4901 /*
4902  * Allocate per cpu pagesets and initialize them.
4903  * Before this call only boot pagesets were available.
4904  */
4905 void __init setup_per_cpu_pageset(void)
4906 {
4907         struct zone *zone;
4908 
4909         for_each_populated_zone(zone)
4910                 setup_zone_pageset(zone);
4911 }
4912 
4913 static noinline __init_refok
4914 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4915 {
4916         int i;
4917         size_t alloc_size;
4918 
4919         /*
4920          * The per-page waitqueue mechanism uses hashed waitqueues
4921          * per zone.
4922          */
4923         zone->wait_table_hash_nr_entries =
4924                  wait_table_hash_nr_entries(zone_size_pages);
4925         zone->wait_table_bits =
4926                 wait_table_bits(zone->wait_table_hash_nr_entries);
4927         alloc_size = zone->wait_table_hash_nr_entries
4928                                         * sizeof(wait_queue_head_t);
4929 
4930         if (!slab_is_available()) {
4931                 zone->wait_table = (wait_queue_head_t *)
4932                         memblock_virt_alloc_node_nopanic(
4933                                 alloc_size, zone->zone_pgdat->node_id);
4934         } else {
4935                 /*
4936                  * This case means that a zone whose size was 0 gets new memory
4937                  * via memory hot-add.
4938                  * But it may be the case that a new node was hot-added.  In
4939                  * this case vmalloc() will not be able to use this new node's
4940                  * memory - this wait_table must be initialized to use this new
4941                  * node itself as well.
4942                  * To use this new node's memory, further consideration will be
4943                  * necessary.
4944                  */
4945                 zone->wait_table = vmalloc(alloc_size);
4946         }
4947         if (!zone->wait_table)
4948                 return -ENOMEM;
4949 
4950         for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4951                 init_waitqueue_head(zone->wait_table + i);
4952 
4953         return 0;
4954 }
4955 
4956 static __meminit void zone_pcp_init(struct zone *zone)
4957 {
4958         /*
4959          * per cpu subsystem is not up at this point. The following code
4960          * relies on the ability of the linker to provide the
4961          * offset of a (static) per cpu variable into the per cpu area.
4962          */
4963         zone->pageset = &boot_pageset;
4964 
4965         if (populated_zone(zone))
4966                 printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
4967                         zone->name, zone->present_pages,
4968                                          zone_batchsize(zone));
4969 }
4970 
4971 int __meminit init_currently_empty_zone(struct zone *zone,
4972                                         unsigned long zone_start_pfn,
4973                                         unsigned long size)
4974 {
4975         struct pglist_data *pgdat = zone->zone_pgdat;
4976         int ret;
4977         ret = zone_wait_table_init(zone, size);
4978         if (ret)
4979                 return ret;
4980         pgdat->nr_zones = zone_idx(zone) + 1;
4981 
4982         zone->zone_start_pfn = zone_start_pfn;
4983 
4984         mminit_dprintk(MMINIT_TRACE, "memmap_init",
4985                         "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4986                         pgdat->node_id,
4987                         (unsigned long)zone_idx(zone),
4988                         zone_start_pfn, (zone_start_pfn + size));
4989 
4990         zone_init_free_lists(zone);
4991 
4992         return 0;
4993 }
4994 
4995 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4996 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4997 
4998 /*
4999  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5000  */
5001 int __meminit __early_pfn_to_nid(unsigned long pfn,
5002                                         struct mminit_pfnnid_cache *state)
5003 {
5004         unsigned long start_pfn, end_pfn;
5005         int nid;
5006 
5007         if (state->last_start <= pfn && pfn < state->last_end)
5008                 return state->last_nid;
5009 
5010         nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5011         if (nid != -1) {
5012                 state->last_start = start_pfn;
5013                 state->last_end = end_pfn;
5014                 state->last_nid = nid;
5015         }
5016 
5017         return nid;
5018 }
5019 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5020 
5021 /**
5022  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5023  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5024  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5025  *
5026  * If an architecture guarantees that all ranges registered contain no holes
5027  * and may be freed, this this function may be used instead of calling
5028  * memblock_free_early_nid() manually.
5029  */
5030 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
5031 {
5032         unsigned long start_pfn, end_pfn;
5033         int i, this_nid;
5034 
5035         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5036                 start_pfn = min(start_pfn, max_low_pfn);
5037                 end_pfn = min(end_pfn, max_low_pfn);
5038 
5039                 if (start_pfn < end_pfn)
5040                         memblock_free_early_nid(PFN_PHYS(start_pfn),
5041                                         (end_pfn - start_pfn) << PAGE_SHIFT,
5042                                         this_nid);
5043         }
5044 }
5045 
5046 /**
5047  * sparse_memory_present_with_active_regions - Call memory_present for each active range
5048  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5049  *
5050  * If an architecture guarantees that all ranges registered contain no holes and may
5051  * be freed, this function may be used instead of calling memory_present() manually.
5052  */
5053 void __init sparse_memory_present_with_active_regions(int nid)
5054 {
5055         unsigned long start_pfn, end_pfn;
5056         int i, this_nid;
5057 
5058         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5059                 memory_present(this_nid, start_pfn, end_pfn);
5060 }
5061 
5062 /**
5063  * get_pfn_range_for_nid - Return the start and end page frames for a node
5064  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5065  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5066  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5067  *
5068  * It returns the start and end page frame of a node based on information
5069  * provided by memblock_set_node(). If called for a node
5070  * with no available memory, a warning is printed and the start and end
5071  * PFNs will be 0.
5072  */
5073 void __meminit get_pfn_range_for_nid(unsigned int nid,
5074                         unsigned long *start_pfn, unsigned long *end_pfn)
5075 {
5076         unsigned long this_start_pfn, this_end_pfn;
5077         int i;
5078 
5079         *start_pfn = -1UL;
5080         *end_pfn = 0;
5081 
5082         for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5083                 *start_pfn = min(*start_pfn, this_start_pfn);
5084                 *end_pfn = max(*end_pfn, this_end_pfn);
5085         }
5086 
5087         if (*start_pfn == -1UL)
5088                 *start_pfn = 0;
5089 }
5090 
5091 /*
5092  * This finds a zone that can be used for ZONE_MOVABLE pages. The
5093  * assumption is made that zones within a node are ordered in monotonic
5094  * increasing memory addresses so that the "highest" populated zone is used
5095  */
5096 static void __init find_usable_zone_for_movable(void)
5097 {
5098         int zone_index;
5099         for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5100                 if (zone_index == ZONE_MOVABLE)
5101                         continue;
5102 
5103                 if (arch_zone_highest_possible_pfn[zone_index] >
5104                                 arch_zone_lowest_possible_pfn[zone_index])
5105                         break;
5106         }
5107 
5108         VM_BUG_ON(zone_index == -1);
5109         movable_zone = zone_index;
5110 }
5111 
5112 /*
5113  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5114  * because it is sized independent of architecture. Unlike the other zones,
5115  * the starting point for ZONE_MOVABLE is not fixed. It may be different
5116  * in each node depending on the size of each node and how evenly kernelcore
5117  * is distributed. This helper function adjusts the zone ranges
5118  * provided by the architecture for a given node by using the end of the
5119  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5120  * zones within a node are in order of monotonic increases memory addresses
5121  */
5122 static void __meminit adjust_zone_range_for_zone_movable(int nid,
5123                                         unsigned long zone_type,
5124                                         unsigned long node_start_pfn,
5125                                         unsigned long node_end_pfn,
5126                                         unsigned long *zone_start_pfn,
5127                                         unsigned long *zone_end_pfn)
5128 {
5129         /* Only adjust if ZONE_MOVABLE is on this node */
5130         if (zone_movable_pfn[nid]) {
5131                 /* Size ZONE_MOVABLE */
5132                 if (zone_type == ZONE_MOVABLE) {
5133                         *zone_start_pfn = zone_movable_pfn[nid];
5134                         *zone_end_pfn = min(node_end_pfn,
5135                                 arch_zone_highest_possible_pfn[movable_zone]);
5136 
5137                 /* Check if this whole range is within ZONE_MOVABLE */
5138                 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5139                         *zone_start_pfn = *zone_end_pfn;
5140         }
5141 }
5142 
5143 /*
5144  * Return the number of pages a zone spans in a node, including holes
5145  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5146  */
5147 static 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 *zone_start_pfn,
5152                                         unsigned long *zone_end_pfn,
5153                                         unsigned long *ignored)
5154 {
5155         /* When hotadd a new node from cpu_up(), the node should be empty */
5156         if (!node_start_pfn && !node_end_pfn)
5157                 return 0;
5158 
5159         /* Get the start and end of the zone */
5160         *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5161         *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
5162         adjust_zone_range_for_zone_movable(nid, zone_type,
5163                                 node_start_pfn, node_end_pfn,
5164                                 zone_start_pfn, zone_end_pfn);
5165 
5166         /* Check that this node has pages within the zone's required range */
5167         if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
5168                 return 0;
5169 
5170         /* Move the zone boundaries inside the node if necessary */
5171         *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5172         *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
5173 
5174         /* Return the spanned pages */
5175         return *zone_end_pfn - *zone_start_pfn;
5176 }
5177 
5178 /*
5179  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5180  * then all holes in the requested range will be accounted for.
5181  */
5182 unsigned long __meminit __absent_pages_in_range(int nid,
5183                                 unsigned long range_start_pfn,
5184                                 unsigned long range_end_pfn)
5185 {
5186         unsigned long nr_absent = range_end_pfn - range_start_pfn;
5187         unsigned long start_pfn, end_pfn;
5188         int i;
5189 
5190         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5191                 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5192                 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5193                 nr_absent -= end_pfn - start_pfn;
5194         }
5195         return nr_absent;
5196 }
5197 
5198 /**
5199  * absent_pages_in_range - Return number of page frames in holes within a range
5200  * @start_pfn: The start PFN to start searching for holes
5201  * @end_pfn: The end PFN to stop searching for holes
5202  *
5203  * It returns the number of pages frames in memory holes within a range.
5204  */
5205 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5206                                                         unsigned long end_pfn)
5207 {
5208         return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5209 }
5210 
5211 /* Return the number of page frames in holes in a zone on a node */
5212 static unsigned long __meminit zone_absent_pages_in_node(int nid,
5213                                         unsigned long zone_type,
5214                                         unsigned long node_start_pfn,
5215                                         unsigned long node_end_pfn,
5216                                         unsigned long *ignored)
5217 {
5218         unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5219         unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
5220         unsigned long zone_start_pfn, zone_end_pfn;
5221         unsigned long nr_absent;
5222 
5223         /* When hotadd a new node from cpu_up(), the node should be empty */
5224         if (!node_start_pfn && !node_end_pfn)
5225                 return 0;
5226 
5227         zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5228         zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
5229 
5230         adjust_zone_range_for_zone_movable(nid, zone_type,
5231                         node_start_pfn, node_end_pfn,
5232                         &zone_start_pfn, &zone_end_pfn);
5233         nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5234 
5235         /*
5236          * ZONE_MOVABLE handling.
5237          * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5238          * and vice versa.
5239          */
5240         if (zone_movable_pfn[nid]) {
5241                 if (mirrored_kernelcore) {
5242                         unsigned long start_pfn, end_pfn;
5243                         struct memblock_region *r;
5244 
5245                         for_each_memblock(memory, r) {
5246                                 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5247                                                   zone_start_pfn, zone_end_pfn);
5248                                 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5249                                                 zone_start_pfn, zone_end_pfn);
5250 
5251                                 if (zone_type == ZONE_MOVABLE &&
5252                                     memblock_is_mirror(r))
5253                                         nr_absent += end_pfn - start_pfn;
5254 
5255                                 if (zone_type == ZONE_NORMAL &&
5256                                     !memblock_is_mirror(r))
5257                                         nr_absent += end_pfn - start_pfn;
5258                         }
5259                 } else {
5260                         if (zone_type == ZONE_NORMAL)
5261                                 nr_absent += node_end_pfn - zone_movable_pfn[nid];
5262                 }
5263         }
5264 
5265         return nr_absent;
5266 }
5267 
5268 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5269 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
5270                                         unsigned long zone_type,
5271                                         unsigned long node_start_pfn,
5272                                         unsigned long node_end_pfn,
5273                                         unsigned long *zone_start_pfn,
5274                                         unsigned long *zone_end_pfn,
5275                                         unsigned long *zones_size)
5276 {
5277         unsigned int zone;
5278 
5279         *zone_start_pfn = node_start_pfn;
5280         for (zone = 0; zone < zone_type; zone++)
5281                 *zone_start_pfn += zones_size[zone];
5282 
5283         *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5284 
5285         return zones_size[zone_type];
5286 }
5287 
5288 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
5289                                                 unsigned long zone_type,
5290                                                 unsigned long node_start_pfn,
5291                                                 unsigned long node_end_pfn,
5292                                                 unsigned long *zholes_size)
5293 {
5294         if (!zholes_size)
5295                 return 0;
5296 
5297         return zholes_size[zone_type];
5298 }
5299 
5300 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5301 
5302 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
5303                                                 unsigned long node_start_pfn,
5304                                                 unsigned long node_end_pfn,
5305                                                 unsigned long *zones_size,
5306                                                 unsigned long *zholes_size)
5307 {
5308         unsigned long realtotalpages = 0, totalpages = 0;
5309         enum zone_type i;
5310 
5311         for (i = 0; i < MAX_NR_ZONES; i++) {
5312                 struct zone *zone = pgdat->node_zones + i;
5313                 unsigned long zone_start_pfn, zone_end_pfn;
5314                 unsigned long size, real_size;
5315 
5316                 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5317                                                   node_start_pfn,
5318                                                   node_end_pfn,
5319                                                   &zone_start_pfn,
5320                                                   &zone_end_pfn,
5321                                                   zones_size);
5322                 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
5323                                                   node_start_pfn, node_end_pfn,
5324                                                   zholes_size);
5325                 if (size)
5326                         zone->zone_start_pfn = zone_start_pfn;
5327                 else
5328                         zone->zone_start_pfn = 0;
5329                 zone->spanned_pages = size;
5330                 zone->present_pages = real_size;
5331 
5332                 totalpages += size;
5333                 realtotalpages += real_size;
5334         }
5335 
5336         pgdat->node_spanned_pages = totalpages;
5337         pgdat->node_present_pages = realtotalpages;
5338         printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5339                                                         realtotalpages);
5340 }
5341 
5342 #ifndef CONFIG_SPARSEMEM
5343 /*
5344  * Calculate the size of the zone->blockflags rounded to an unsigned long
5345  * Start by making sure zonesize is a multiple of pageblock_order by rounding
5346  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5347  * round what is now in bits to nearest long in bits, then return it in
5348  * bytes.
5349  */
5350 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
5351 {
5352         unsigned long usemapsize;
5353 
5354         zonesize += zone_start_pfn & (pageblock_nr_pages-1);
5355         usemapsize = roundup(zonesize, pageblock_nr_pages);
5356         usemapsize = usemapsize >> pageblock_order;
5357         usemapsize *= NR_PAGEBLOCK_BITS;
5358         usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5359 
5360         return usemapsize / 8;
5361 }
5362 
5363 static void __init setup_usemap(struct pglist_data *pgdat,
5364                                 struct zone *zone,
5365                                 unsigned long zone_start_pfn,
5366                                 unsigned long zonesize)
5367 {
5368         unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
5369         zone->pageblock_flags = NULL;
5370         if (usemapsize)
5371                 zone->pageblock_flags =
5372                         memblock_virt_alloc_node_nopanic(usemapsize,
5373                                                          pgdat->node_id);
5374 }
5375 #else
5376 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5377                                 unsigned long zone_start_pfn, unsigned long zonesize) {}
5378 #endif /* CONFIG_SPARSEMEM */
5379 
5380 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5381 
5382 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5383 void __paginginit set_pageblock_order(void)
5384 {
5385         unsigned int order;
5386 
5387         /* Check that pageblock_nr_pages has not already been setup */
5388         if (pageblock_order)
5389                 return;
5390 
5391         if (HPAGE_SHIFT > PAGE_SHIFT)
5392                 order = HUGETLB_PAGE_ORDER;
5393         else
5394                 order = MAX_ORDER - 1;
5395 
5396         /*
5397          * Assume the largest contiguous order of interest is a huge page.
5398          * This value may be variable depending on boot parameters on IA64 and
5399          * powerpc.
5400          */
5401         pageblock_order = order;
5402 }
5403 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5404 
5405 /*
5406  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5407  * is unused as pageblock_order is set at compile-time. See
5408  * include/linux/pageblock-flags.h for the values of pageblock_order based on
5409  * the kernel config
5410  */
5411 void __paginginit set_pageblock_order(void)
5412 {
5413 }
5414 
5415 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5416 
5417 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5418                                                    unsigned long present_pages)
5419 {
5420         unsigned long pages = spanned_pages;
5421 
5422         /*
5423          * Provide a more accurate estimation if there are holes within
5424          * the zone and SPARSEMEM is in use. If there are holes within the
5425          * zone, each populated memory region may cost us one or two extra
5426          * memmap pages due to alignment because memmap pages for each
5427          * populated regions may not naturally algined on page boundary.
5428          * So the (present_pages >> 4) heuristic is a tradeoff for that.
5429          */
5430         if (spanned_pages > present_pages + (present_pages >> 4) &&
5431             IS_ENABLED(CONFIG_SPARSEMEM))
5432                 pages = present_pages;
5433 
5434         return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5435 }
5436 
5437 /*
5438  * Set up the zone data structures:
5439  *   - mark all pages reserved
5440  *   - mark all memory queues empty
5441  *   - clear the memory bitmaps
5442  *
5443  * NOTE: pgdat should get zeroed by caller.
5444  */
5445 static void __paginginit free_area_init_core(struct pglist_data *pgdat)
5446 {
5447         enum zone_type j;
5448         int nid = pgdat->node_id;
5449         int ret;
5450 
5451         pgdat_resize_init(pgdat);
5452 #ifdef CONFIG_NUMA_BALANCING
5453         spin_lock_init(&pgdat->numabalancing_migrate_lock);
5454         pgdat->numabalancing_migrate_nr_pages = 0;
5455         pgdat->numabalancing_migrate_next_window = jiffies;
5456 #endif
5457 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5458         spin_lock_init(&pgdat->split_queue_lock);
5459         INIT_LIST_HEAD(&pgdat->split_queue);
5460         pgdat->split_queue_len = 0;
5461 #endif
5462         init_waitqueue_head(&pgdat->kswapd_wait);
5463         init_waitqueue_head(&pgdat->pfmemalloc_wait);
5464 #ifdef CONFIG_COMPACTION
5465         init_waitqueue_head(&pgdat->kcompactd_wait);
5466 #endif
5467         pgdat_page_ext_init(pgdat);
5468 
5469         for (j = 0; j < MAX_NR_ZONES; j++) {
5470                 struct zone *zone = pgdat->node_zones + j;
5471                 unsigned long size, realsize, freesize, memmap_pages;
5472                 unsigned long zone_start_pfn = zone->zone_start_pfn;
5473 
5474                 size = zone->spanned_pages;
5475                 realsize = freesize = zone->present_pages;
5476 
5477                 /*
5478                  * Adjust freesize so that it accounts for how much memory
5479                  * is used by this zone for memmap. This affects the watermark
5480                  * and per-cpu initialisations
5481                  */
5482                 memmap_pages = calc_memmap_size(size, realsize);
5483                 if (!is_highmem_idx(j)) {
5484                         if (freesize >= memmap_pages) {
5485                                 freesize -= memmap_pages;
5486                                 if (memmap_pages)
5487                                         printk(KERN_DEBUG
5488                                                "  %s zone: %lu pages used for memmap\n",
5489                                                zone_names[j], memmap_pages);
5490                         } else
5491                                 pr_warn("  %s zone: %lu pages exceeds freesize %lu\n",
5492                                         zone_names[j], memmap_pages, freesize);
5493                 }
5494 
5495                 /* Account for reserved pages */
5496                 if (j == 0 && freesize > dma_reserve) {
5497                         freesize -= dma_reserve;
5498                         printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
5499                                         zone_names[0], dma_reserve);
5500                 }
5501 
5502                 if (!is_highmem_idx(j))
5503                         nr_kernel_pages += freesize;
5504                 /* Charge for highmem memmap if there are enough kernel pages */
5505                 else if (nr_kernel_pages > memmap_pages * 2)
5506                         nr_kernel_pages -= memmap_pages;
5507                 nr_all_pages += freesize;
5508 
5509                 /*
5510                  * Set an approximate value for lowmem here, it will be adjusted
5511                  * when the bootmem allocator frees pages into the buddy system.
5512                  * And all highmem pages will be managed by the buddy system.
5513                  */
5514                 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
5515 #ifdef CONFIG_NUMA
5516                 zone->node = nid;
5517                 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
5518                                                 / 100;
5519                 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
5520 #endif
5521                 zone->name = zone_names[j];
5522                 spin_lock_init(&zone->lock);
5523                 spin_lock_init(&zone->lru_lock);
5524                 zone_seqlock_init(zone);
5525                 zone->zone_pgdat = pgdat;
5526                 zone_pcp_init(zone);
5527 
5528                 /* For bootup, initialized properly in watermark setup */
5529                 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5530 
5531                 lruvec_init(&zone->lruvec);
5532                 if (!size)
5533                         continue;
5534 
5535                 set_pageblock_order();
5536                 setup_usemap(pgdat, zone, zone_start_pfn, size);
5537                 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
5538                 BUG_ON(ret);
5539                 memmap_init(size, nid, j, zone_start_pfn);
5540         }
5541 }
5542 
5543 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
5544 {
5545         unsigned long __maybe_unused start = 0;
5546         unsigned long __maybe_unused offset = 0;
5547 
5548         /* Skip empty nodes */
5549         if (!pgdat->node_spanned_pages)
5550                 return;
5551 
5552 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5553         start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5554         offset = pgdat->node_start_pfn - start;
5555         /* ia64 gets its own node_mem_map, before this, without bootmem */
5556         if (!pgdat->node_mem_map) {
5557                 unsigned long size, end;
5558                 struct page *map;
5559 
5560                 /*
5561                  * The zone's endpoints aren't required to be MAX_ORDER
5562                  * aligned but the node_mem_map endpoints must be in order
5563                  * for the buddy allocator to function correctly.
5564                  */
5565                 end = pgdat_end_pfn(pgdat);
5566                 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5567                 size =  (end - start) * sizeof(struct page);
5568                 map = alloc_remap(pgdat->node_id, size);
5569                 if (!map)
5570                         map = memblock_virt_alloc_node_nopanic(size,
5571                                                                pgdat->node_id);
5572                 pgdat->node_mem_map = map + offset;
5573         }
5574 #ifndef CONFIG_NEED_MULTIPLE_NODES
5575         /*
5576          * With no DISCONTIG, the global mem_map is just set as node 0's
5577          */
5578         if (pgdat == NODE_DATA(0)) {
5579                 mem_map = NODE_DATA(0)->node_mem_map;
5580 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5581                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
5582                         mem_map -= offset;
5583 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5584         }
5585 #endif
5586 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5587 }
5588 
5589 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5590                 unsigned long node_start_pfn, unsigned long *zholes_size)
5591 {
5592         pg_data_t *pgdat = NODE_DATA(nid);
5593         unsigned long start_pfn = 0;
5594         unsigned long end_pfn = 0;
5595 
5596         /* pg_data_t should be reset to zero when it's allocated */
5597         WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
5598 
5599         reset_deferred_meminit(pgdat);
5600         pgdat->node_id = nid;
5601         pgdat->node_start_pfn = node_start_pfn;
5602 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5603         get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
5604         pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5605                 (u64)start_pfn << PAGE_SHIFT,
5606                 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
5607 #else
5608         start_pfn = node_start_pfn;
5609 #endif
5610         calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5611                                   zones_size, zholes_size);
5612 
5613         alloc_node_mem_map(pgdat);
5614 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5615         printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5616                 nid, (unsigned long)pgdat,
5617                 (unsigned long)pgdat->node_mem_map);
5618 #endif
5619 
5620         free_area_init_core(pgdat);
5621 }
5622 
5623 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5624 
5625 #if MAX_NUMNODES > 1
5626 /*
5627  * Figure out the number of possible node ids.
5628  */
5629 void __init setup_nr_node_ids(void)
5630 {
5631         unsigned int highest;
5632 
5633         highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
5634         nr_node_ids = highest + 1;
5635 }
5636 #endif
5637 
5638 /**
5639  * node_map_pfn_alignment - determine the maximum internode alignment
5640  *
5641  * This function should be called after node map is populated and sorted.
5642  * It calculates the maximum power of two alignment which can distinguish
5643  * all the nodes.
5644  *
5645  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5646  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
5647  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
5648  * shifted, 1GiB is enough and this function will indicate so.
5649  *
5650  * This is used to test whether pfn -> nid mapping of the chosen memory
5651  * model has fine enough granularity to avoid incorrect mapping for the
5652  * populated node map.
5653  *
5654  * Returns the determined alignment in pfn's.  0 if there is no alignment
5655  * requirement (single node).
5656  */
5657 unsigned long __init node_map_pfn_alignment(void)
5658 {
5659         unsigned long accl_mask = 0, last_end = 0;
5660         unsigned long start, end, mask;
5661         int last_nid = -1;
5662         int i, nid;
5663 
5664         for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5665                 if (!start || last_nid < 0 || last_nid == nid) {
5666                         last_nid = nid;
5667                         last_end = end;
5668                         continue;
5669                 }
5670 
5671                 /*
5672                  * Start with a mask granular enough to pin-point to the
5673                  * start pfn and tick off bits one-by-one until it becomes
5674                  * too coarse to separate the current node from the last.
5675                  */
5676                 mask = ~((1 << __ffs(start)) - 1);
5677                 while (mask && last_end <= (start & (mask << 1)))
5678                         mask <<= 1;
5679 
5680                 /* accumulate all internode masks */
5681                 accl_mask |= mask;
5682         }
5683 
5684         /* convert mask to number of pages */
5685         return ~accl_mask + 1;
5686 }
5687 
5688 /* Find the lowest pfn for a node */
5689 static unsigned long __init find_min_pfn_for_node(int nid)
5690 {
5691         unsigned long min_pfn = ULONG_MAX;
5692         unsigned long start_pfn;
5693         int i;
5694 
5695         for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5696                 min_pfn = min(min_pfn, start_pfn);
5697 
5698         if (min_pfn == ULONG_MAX) {
5699                 pr_warn("Could not find start_pfn for node %d\n", nid);
5700                 return 0;
5701         }
5702 
5703         return min_pfn;
5704 }
5705 
5706 /**
5707  * find_min_pfn_with_active_regions - Find the minimum PFN registered
5708  *
5709  * It returns the minimum PFN based on information provided via
5710  * memblock_set_node().
5711  */
5712 unsigned long __init find_min_pfn_with_active_regions(void)
5713 {
5714         return find_min_pfn_for_node(MAX_NUMNODES);
5715 }
5716 
5717 /*
5718  * early_calculate_totalpages()
5719  * Sum pages in active regions for movable zone.
5720  * Populate N_MEMORY for calculating usable_nodes.
5721  */
5722 static unsigned long __init early_calculate_totalpages(void)
5723 {
5724         unsigned long totalpages = 0;
5725         unsigned long start_pfn, end_pfn;
5726         int i, nid;
5727 
5728         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5729                 unsigned long pages = end_pfn - start_pfn;
5730 
5731                 totalpages += pages;
5732                 if (pages)
5733                         node_set_state(nid, N_MEMORY);
5734         }
5735         return totalpages;
5736 }
5737 
5738 /*
5739  * Find the PFN the Movable zone begins in each node. Kernel memory
5740  * is spread evenly between nodes as long as the nodes have enough
5741  * memory. When they don't, some nodes will have more kernelcore than
5742  * others
5743  */
5744 static void __init find_zone_movable_pfns_for_nodes(void)
5745 {
5746         int i, nid;
5747         unsigned long usable_startpfn;
5748         unsigned long kernelcore_node, kernelcore_remaining;
5749         /* save the state before borrow the nodemask */
5750         nodemask_t saved_node_state = node_states[N_MEMORY];
5751         unsigned long totalpages = early_calculate_totalpages();
5752         int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5753         struct memblock_region *r;
5754 
5755         /* Need to find movable_zone earlier when movable_node is specified. */
5756         find_usable_zone_for_movable();
5757 
5758         /*
5759          * If movable_node is specified, ignore kernelcore and movablecore
5760          * options.
5761          */
5762         if (movable_node_is_enabled()) {
5763                 for_each_memblock(memory, r) {
5764                         if (!memblock_is_hotpluggable(r))
5765                                 continue;
5766 
5767                         nid = r->nid;
5768 
5769                         usable_startpfn = PFN_DOWN(r->base);
5770                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5771                                 min(usable_startpfn, zone_movable_pfn[nid]) :
5772                                 usable_startpfn;
5773                 }
5774 
5775                 goto out2;
5776         }
5777 
5778         /*
5779          * If kernelcore=mirror is specified, ignore movablecore option
5780          */
5781         if (mirrored_kernelcore) {
5782                 bool mem_below_4gb_not_mirrored = false;
5783 
5784                 for_each_memblock(memory, r) {
5785                         if (memblock_is_mirror(r))
5786                                 continue;
5787 
5788                         nid = r->nid;
5789 
5790                         usable_startpfn = memblock_region_memory_base_pfn(r);
5791 
5792                         if (usable_startpfn < 0x100000) {
5793                                 mem_below_4gb_not_mirrored = true;
5794                                 continue;
5795                         }
5796 
5797                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5798                                 min(usable_startpfn, zone_movable_pfn[nid]) :
5799                                 usable_startpfn;
5800                 }
5801 
5802                 if (mem_below_4gb_not_mirrored)
5803                         pr_warn("This configuration results in unmirrored kernel memory.");
5804 
5805                 goto out2;
5806         }
5807 
5808         /*
5809          * If movablecore=nn[KMG] was specified, calculate what size of
5810          * kernelcore that corresponds so that memory usable for
5811          * any allocation type is evenly spread. If both kernelcore
5812          * and movablecore are specified, then the value of kernelcore
5813          * will be used for required_kernelcore if it's greater than
5814          * what movablecore would have allowed.
5815          */
5816         if (required_movablecore) {
5817                 unsigned long corepages;
5818 
5819                 /*
5820                  * Round-up so that ZONE_MOVABLE is at least as large as what
5821                  * was requested by the user
5822                  */
5823                 required_movablecore =
5824                         roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5825                 required_movablecore = min(totalpages, required_movablecore);
5826                 corepages = totalpages - required_movablecore;
5827 
5828                 required_kernelcore = max(required_kernelcore, corepages);
5829         }
5830 
5831         /*
5832          * If kernelcore was not specified or kernelcore size is larger
5833          * than totalpages, there is no ZONE_MOVABLE.
5834          */
5835         if (!required_kernelcore || required_kernelcore >= totalpages)
5836                 goto out;
5837 
5838         /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5839         usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5840 
5841 restart:
5842         /* Spread kernelcore memory as evenly as possible throughout nodes */
5843         kernelcore_node = required_kernelcore / usable_nodes;
5844         for_each_node_state(nid, N_MEMORY) {
5845                 unsigned long start_pfn, end_pfn;
5846 
5847                 /*
5848                  * Recalculate kernelcore_node if the division per node
5849                  * now exceeds what is necessary to satisfy the requested
5850                  * amount of memory for the kernel
5851                  */
5852                 if (required_kernelcore < kernelcore_node)
5853                         kernelcore_node = required_kernelcore / usable_nodes;
5854 
5855                 /*
5856                  * As the map is walked, we track how much memory is usable
5857                  * by the kernel using kernelcore_remaining. When it is
5858                  * 0, the rest of the node is usable by ZONE_MOVABLE
5859                  */
5860                 kernelcore_remaining = kernelcore_node;
5861 
5862                 /* Go through each range of PFNs within this node */
5863                 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5864                         unsigned long size_pages;
5865 
5866                         start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5867                         if (start_pfn >= end_pfn)
5868                                 continue;
5869 
5870                         /* Account for what is only usable for kernelcore */
5871                         if (start_pfn < usable_startpfn) {
5872                                 unsigned long kernel_pages;
5873                                 kernel_pages = min(end_pfn, usable_startpfn)
5874                                                                 - start_pfn;
5875 
5876                                 kernelcore_remaining -= min(kernel_pages,
5877                                                         kernelcore_remaining);
5878                                 required_kernelcore -= min(kernel_pages,
5879                                                         required_kernelcore);
5880 
5881                                 /* Continue if range is now fully accounted */
5882                                 if (end_pfn <= usable_startpfn) {
5883 
5884                                         /*
5885                                          * Push zone_movable_pfn to the end so
5886                                          * that if we have to rebalance
5887                                          * kernelcore across nodes, we will
5888                                          * not double account here
5889                                          */
5890                                         zone_movable_pfn[nid] = end_pfn;
5891                                         continue;
5892                                 }
5893                                 start_pfn = usable_startpfn;
5894                         }
5895 
5896                         /*
5897                          * The usable PFN range for ZONE_MOVABLE is from
5898                          * start_pfn->end_pfn. Calculate size_pages as the
5899                          * number of pages used as kernelcore
5900                          */
5901                         size_pages = end_pfn - start_pfn;
5902                         if (size_pages > kernelcore_remaining)
5903                                 size_pages = kernelcore_remaining;
5904                         zone_movable_pfn[nid] = start_pfn + size_pages;
5905 
5906                         /*
5907                          * Some kernelcore has been met, update counts and
5908                          * break if the kernelcore for this node has been
5909                          * satisfied
5910                          */
5911                         required_kernelcore -= min(required_kernelcore,
5912                                                                 size_pages);
5913                         kernelcore_remaining -= size_pages;
5914                         if (!kernelcore_remaining)
5915                                 break;
5916                 }
5917         }
5918 
5919         /*
5920          * If there is still required_kernelcore, we do another pass with one
5921          * less node in the count. This will push zone_movable_pfn[nid] further
5922          * along on the nodes that still have memory until kernelcore is
5923          * satisfied
5924          */
5925         usable_nodes--;
5926         if (usable_nodes && required_kernelcore > usable_nodes)
5927                 goto restart;
5928 
5929 out2:
5930         /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5931         for (nid = 0; nid < MAX_NUMNODES; nid++)
5932                 zone_movable_pfn[nid] =
5933                         roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5934 
5935 out:
5936         /* restore the node_state */
5937         node_states[N_MEMORY] = saved_node_state;
5938 }
5939 
5940 /* Any regular or high memory on that node ? */
5941 static void check_for_memory(pg_data_t *pgdat, int nid)
5942 {
5943         enum zone_type zone_type;
5944 
5945         if (N_MEMORY == N_NORMAL_MEMORY)
5946                 return;
5947 
5948         for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5949                 struct zone *zone = &pgdat->node_zones[zone_type];
5950                 if (populated_zone(zone)) {
5951                         node_set_state(nid, N_HIGH_MEMORY);
5952                         if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5953                             zone_type <= ZONE_NORMAL)
5954                                 node_set_state(nid, N_NORMAL_MEMORY);
5955                         break;
5956                 }
5957         }
5958 }
5959 
5960 /**
5961  * free_area_init_nodes - Initialise all pg_data_t and zone data
5962  * @max_zone_pfn: an array of max PFNs for each zone
5963  *
5964  * This will call free_area_init_node() for each active node in the system.
5965  * Using the page ranges provided by memblock_set_node(), the size of each
5966  * zone in each node and their holes is calculated. If the maximum PFN
5967  * between two adjacent zones match, it is assumed that the zone is empty.
5968  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5969  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5970  * starts where the previous one ended. For example, ZONE_DMA32 starts
5971  * at arch_max_dma_pfn.
5972  */
5973 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5974 {
5975         unsigned long start_pfn, end_pfn;
5976         int i, nid;
5977 
5978         /* Record where the zone boundaries are */
5979         memset(arch_zone_lowest_possible_pfn, 0,
5980                                 sizeof(arch_zone_lowest_possible_pfn));
5981         memset(arch_zone_highest_possible_pfn, 0,
5982                                 sizeof(arch_zone_highest_possible_pfn));
5983         arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5984         arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5985         for (i = 1; i < MAX_NR_ZONES; i++) {
5986                 if (i == ZONE_MOVABLE)
5987                         continue;
5988                 arch_zone_lowest_possible_pfn[i] =
5989                         arch_zone_highest_possible_pfn[i-1];
5990                 arch_zone_highest_possible_pfn[i] =
5991                         max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5992         }
5993         arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5994         arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5995 
5996         /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5997         memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5998         find_zone_movable_pfns_for_nodes();
5999 
6000         /* Print out the zone ranges */
6001         pr_info("Zone ranges:\n");
6002         for (i = 0; i < MAX_NR_ZONES; i++) {
6003                 if (i == ZONE_MOVABLE)
6004                         continue;
6005                 pr_info("  %-8s ", zone_names[i]);
6006                 if (arch_zone_lowest_possible_pfn[i] ==
6007                                 arch_zone_highest_possible_pfn[i])
6008                         pr_cont("empty\n");
6009                 else
6010                         pr_cont("[mem %#018Lx-%#018Lx]\n",
6011                                 (u64)arch_zone_lowest_possible_pfn[i]
6012                                         << PAGE_SHIFT,
6013                                 ((u64)arch_zone_highest_possible_pfn[i]
6014                                         << PAGE_SHIFT) - 1);
6015         }
6016 
6017         /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6018         pr_info("Movable zone start for each node\n");
6019         for (i = 0; i < MAX_NUMNODES; i++) {
6020                 if (zone_movable_pfn[i])
6021                         pr_info("  Node %d: %#018Lx\n", i,
6022                                (u64)zone_movable_pfn[i] << PAGE_SHIFT);
6023         }
6024 
6025         /* Print out the early node map */
6026         pr_info("Early memory node ranges\n");
6027         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
6028                 pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6029                         (u64)start_pfn << PAGE_SHIFT,
6030                         ((u64)end_pfn << PAGE_SHIFT) - 1);
6031 
6032         /* Initialise every node */
6033         mminit_verify_pageflags_layout();
6034         setup_nr_node_ids();
6035         for_each_online_node(nid) {
6036                 pg_data_t *pgdat = NODE_DATA(nid);
6037                 free_area_init_node(nid, NULL,
6038                                 find_min_pfn_for_node(nid), NULL);
6039 
6040                 /* Any memory on that node */
6041                 if (pgdat->node_present_pages)
6042                         node_set_state(nid, N_MEMORY);
6043                 check_for_memory(pgdat, nid);
6044         }
6045 }
6046 
6047 static int __init cmdline_parse_core(char *p, unsigned long *core)
6048 {
6049         unsigned long long coremem;
6050         if (!p)
6051                 return -EINVAL;
6052 
6053         coremem = memparse(p, &p);
6054         *core = coremem >> PAGE_SHIFT;
6055 
6056         /* Paranoid check that UL is enough for the coremem value */
6057         WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6058 
6059         return 0;
6060 }
6061 
6062 /*
6063  * kernelcore=size sets the amount of memory for use for allocations that
6064  * cannot be reclaimed or migrated.
6065  */
6066 static int __init cmdline_parse_kernelcore(char *p)
6067 {
6068         /* parse kernelcore=mirror */
6069         if (parse_option_str(p, "mirror")) {
6070                 mirrored_kernelcore = true;
6071                 return 0;
6072         }
6073 
6074         return cmdline_parse_core(p, &required_kernelcore);
6075 }
6076 
6077 /*
6078  * movablecore=size sets the amount of memory for use for allocations that
6079  * can be reclaimed or migrated.
6080  */
6081 static int __init cmdline_parse_movablecore(char *p)
6082 {
6083         return cmdline_parse_core(p, &required_movablecore);
6084 }
6085 
6086 early_param("kernelcore", cmdline_parse_kernelcore);
6087 early_param("movablecore", cmdline_parse_movablecore);
6088 
6089 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6090 
6091 void adjust_managed_page_count(struct page *page, long count)
6092 {
6093         spin_lock(&managed_page_count_lock);
6094         page_zone(page)->managed_pages += count;
6095         totalram_pages += count;
6096 #ifdef CONFIG_HIGHMEM
6097         if (PageHighMem(page))
6098                 totalhigh_pages += count;
6099 #endif
6100         spin_unlock(&managed_page_count_lock);
6101 }
6102 EXPORT_SYMBOL(adjust_managed_page_count);
6103 
6104 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
6105 {
6106         void *pos;
6107         unsigned long pages = 0;
6108 
6109         start = (void *)PAGE_ALIGN((unsigned long)start);
6110         end = (void *)((unsigned long)end & PAGE_MASK);
6111         for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
6112                 if ((unsigned int)poison <= 0xFF)
6113                         memset(pos, poison, PAGE_SIZE);
6114                 free_reserved_page(virt_to_page(pos));
6115         }
6116 
6117         if (pages && s)
6118                 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
6119                         s, pages << (PAGE_SHIFT - 10), start, end);
6120 
6121         return pages;
6122 }
6123 EXPORT_SYMBOL(free_reserved_area);
6124 
6125 #ifdef  CONFIG_HIGHMEM
6126 void free_highmem_page(struct page *page)
6127 {
6128         __free_reserved_page(page);
6129         totalram_pages++;
6130         page_zone(page)->managed_pages++;
6131         totalhigh_pages++;
6132 }
6133 #endif
6134 
6135 
6136 void __init mem_init_print_info(const char *str)
6137 {
6138         unsigned long physpages, codesize, datasize, rosize, bss_size;
6139         unsigned long init_code_size, init_data_size;
6140 
6141         physpages = get_num_physpages();
6142         codesize = _etext - _stext;
6143         datasize = _edata - _sdata;
6144         rosize = __end_rodata - __start_rodata;
6145         bss_size = __bss_stop - __bss_start;
6146         init_data_size = __init_end - __init_begin;
6147         init_code_size = _einittext - _sinittext;
6148 
6149         /*
6150          * Detect special cases and adjust section sizes accordingly:
6151          * 1) .init.* may be embedded into .data sections
6152          * 2) .init.text.* may be out of [__init_begin, __init_end],
6153          *    please refer to arch/tile/kernel/vmlinux.lds.S.
6154          * 3) .rodata.* may be embedded into .text or .data sections.
6155          */
6156 #define adj_init_size(start, end, size, pos, adj) \
6157         do { \
6158                 if (start <= pos && pos < end && size > adj) \
6159                         size -= adj; \
6160         } while (0)
6161 
6162         adj_init_size(__init_begin, __init_end, init_data_size,
6163                      _sinittext, init_code_size);
6164         adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6165         adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6166         adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6167         adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6168 
6169 #undef  adj_init_size
6170 
6171         pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6172 #ifdef  CONFIG_HIGHMEM
6173                 ", %luK highmem"
6174 #endif
6175                 "%s%s)\n",
6176                 nr_free_pages() << (PAGE_SHIFT - 10),
6177                 physpages << (PAGE_SHIFT - 10),
6178                 codesize >> 10, datasize >> 10, rosize >> 10,
6179                 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6180                 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6181                 totalcma_pages << (PAGE_SHIFT - 10),
6182 #ifdef  CONFIG_HIGHMEM
6183                 totalhigh_pages << (PAGE_SHIFT - 10),
6184 #endif
6185                 str ? ", " : "", str ? str : "");
6186 }
6187 
6188 /**
6189  * set_dma_reserve - set the specified number of pages reserved in the first zone
6190  * @new_dma_reserve: The number of pages to mark reserved
6191  *
6192  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6193  * In the DMA zone, a significant percentage may be consumed by kernel image
6194  * and other unfreeable allocations which can skew the watermarks badly. This
6195  * function may optionally be used to account for unfreeable pages in the
6196  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6197  * smaller per-cpu batchsize.
6198  */
6199 void __init set_dma_reserve(unsigned long new_dma_reserve)
6200 {
6201         dma_reserve = new_dma_reserve;
6202 }
6203 
6204 void __init free_area_init(unsigned long *zones_size)
6205 {
6206         free_area_init_node(0, zones_size,
6207                         __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6208 }
6209 
6210 static int page_alloc_cpu_notify(struct notifier_block *self,
6211                                  unsigned long action, void *hcpu)
6212 {
6213         int cpu = (unsigned long)hcpu;
6214 
6215         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
6216                 lru_add_drain_cpu(cpu);
6217                 drain_pages(cpu);
6218 
6219                 /*
6220                  * Spill the event counters of the dead processor
6221                  * into the current processors event counters.
6222                  * This artificially elevates the count of the current
6223                  * processor.
6224                  */
6225                 vm_events_fold_cpu(cpu);
6226 
6227                 /*
6228                  * Zero the differential counters of the dead processor
6229                  * so that the vm statistics are consistent.
6230                  *
6231                  * This is only okay since the processor is dead and cannot
6232                  * race with what we are doing.
6233                  */
6234                 cpu_vm_stats_fold(cpu);
6235         }
6236         return NOTIFY_OK;
6237 }
6238 
6239 void __init page_alloc_init(void)
6240 {
6241         hotcpu_notifier(page_alloc_cpu_notify, 0);
6242 }
6243 
6244 /*
6245  * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6246  *      or min_free_kbytes changes.
6247  */
6248 static void calculate_totalreserve_pages(void)
6249 {
6250         struct pglist_data *pgdat;
6251         unsigned long reserve_pages = 0;
6252         enum zone_type i, j;
6253 
6254         for_each_online_pgdat(pgdat) {
6255                 for (i = 0; i < MAX_NR_ZONES; i++) {
6256                         struct zone *zone = pgdat->node_zones + i;
6257                         long max = 0;
6258 
6259                         /* Find valid and maximum lowmem_reserve in the zone */
6260                         for (j = i; j < MAX_NR_ZONES; j++) {
6261                                 if (zone->lowmem_reserve[j] > max)
6262                                         max = zone->lowmem_reserve[j];
6263                         }
6264 
6265                         /* we treat the high watermark as reserved pages. */
6266                         max += high_wmark_pages(zone);
6267 
6268                         if (max > zone->managed_pages)
6269                                 max = zone->managed_pages;
6270 
6271                         zone->totalreserve_pages = max;
6272 
6273                         reserve_pages += max;
6274                 }
6275         }
6276         totalreserve_pages = reserve_pages;
6277 }
6278 
6279 /*
6280  * setup_per_zone_lowmem_reserve - called whenever
6281  *      sysctl_lowmem_reserve_ratio changes.  Ensures that each zone
6282  *      has a correct pages reserved value, so an adequate number of
6283  *      pages are left in the zone after a successful __alloc_pages().
6284  */
6285 static void setup_per_zone_lowmem_reserve(void)
6286 {
6287         struct pglist_data *pgdat;
6288         enum zone_type j, idx;
6289 
6290         for_each_online_pgdat(pgdat) {
6291                 for (j = 0; j < MAX_NR_ZONES; j++) {
6292                         struct zone *zone = pgdat->node_zones + j;
6293                         unsigned long managed_pages = zone->managed_pages;
6294 
6295                         zone->lowmem_reserve[j] = 0;
6296 
6297                         idx = j;
6298                         while (idx) {
6299                                 struct zone *lower_zone;
6300 
6301                                 idx--;
6302 
6303                                 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6304                                         sysctl_lowmem_reserve_ratio[idx] = 1;
6305 
6306                                 lower_zone = pgdat->node_zones + idx;
6307                                 lower_zone->lowmem_reserve[j] = managed_pages /
6308                                         sysctl_lowmem_reserve_ratio[idx];
6309                                 managed_pages += lower_zone->managed_pages;
6310                         }
6311                 }
6312         }
6313 
6314         /* update totalreserve_pages */
6315         calculate_totalreserve_pages();
6316 }
6317 
6318 static void __setup_per_zone_wmarks(void)
6319 {
6320         unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6321         unsigned long lowmem_pages = 0;
6322         struct zone *zone;
6323         unsigned long flags;
6324 
6325         /* Calculate total number of !ZONE_HIGHMEM pages */
6326         for_each_zone(zone) {
6327                 if (!is_highmem(zone))
6328                         lowmem_pages += zone->managed_pages;
6329         }
6330 
6331         for_each_zone(zone) {
6332                 u64 tmp;
6333 
6334                 spin_lock_irqsave(&zone->lock, flags);
6335                 tmp = (u64)pages_min * zone->managed_pages;
6336                 do_div(tmp, lowmem_pages);
6337                 if (is_highmem(zone)) {
6338                         /*
6339                          * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6340                          * need highmem pages, so cap pages_min to a small
6341                          * value here.
6342                          *
6343                          * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6344                          * deltas control asynch page reclaim, and so should
6345                          * not be capped for highmem.
6346                          */
6347                         unsigned long min_pages;
6348 
6349                         min_pages = zone->managed_pages / 1024;
6350                         min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
6351                         zone->watermark[WMARK_MIN] = min_pages;
6352                 } else {
6353                         /*
6354                          * If it's a lowmem zone, reserve a number of pages
6355                          * proportionate to the zone's size.
6356                          */
6357                         zone->watermark[WMARK_MIN] = tmp;
6358                 }
6359 
6360                 /*
6361                  * Set the kswapd watermarks distance according to the
6362                  * scale factor in proportion to available memory, but
6363                  * ensure a minimum size on small systems.
6364                  */
6365                 tmp = max_t(u64, tmp >> 2,
6366                             mult_frac(zone->managed_pages,
6367                                       watermark_scale_factor, 10000));
6368 
6369                 zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + tmp;
6370                 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
6371 
6372                 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
6373                         high_wmark_pages(zone) - low_wmark_pages(zone) -
6374                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
6375 
6376                 spin_unlock_irqrestore(&zone->lock, flags);
6377         }
6378 
6379         /* update totalreserve_pages */
6380         calculate_totalreserve_pages();
6381 }
6382 
6383 /**
6384  * setup_per_zone_wmarks - called when min_free_kbytes changes
6385  * or when memory is hot-{added|removed}
6386  *
6387  * Ensures that the watermark[min,low,high] values for each zone are set
6388  * correctly with respect to min_free_kbytes.
6389  */
6390 void setup_per_zone_wmarks(void)
6391 {
6392         mutex_lock(&zonelists_mutex);
6393         __setup_per_zone_wmarks();
6394         mutex_unlock(&zonelists_mutex);
6395 }
6396 
6397 /*
6398  * The inactive anon list should be small enough that the VM never has to
6399  * do too much work, but large enough that each inactive page has a chance
6400  * to be referenced again before it is swapped out.
6401  *
6402  * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6403  * INACTIVE_ANON pages on this zone's LRU, maintained by the
6404  * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6405  * the anonymous pages are kept on the inactive list.
6406  *
6407  * total     target    max
6408  * memory    ratio     inactive anon
6409  * -------------------------------------
6410  *   10MB       1         5MB
6411  *  100MB       1        50MB
6412  *    1GB       3       250MB
6413  *   10GB      10       0.9GB
6414  *  100GB      31         3GB
6415  *    1TB     101        10GB
6416  *   10TB     320        32GB
6417  */
6418 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
6419 {
6420         unsigned int gb, ratio;
6421 
6422         /* Zone size in gigabytes */
6423         gb = zone->managed_pages >> (30 - PAGE_SHIFT);
6424         if (gb)
6425                 ratio = int_sqrt(10 * gb);
6426         else
6427                 ratio = 1;
6428 
6429         zone->inactive_ratio = ratio;
6430 }
6431 
6432 static void __meminit setup_per_zone_inactive_ratio(void)
6433 {
6434         struct zone *zone;
6435 
6436         for_each_zone(zone)
6437                 calculate_zone_inactive_ratio(zone);
6438 }
6439 
6440 /*
6441  * Initialise min_free_kbytes.
6442  *
6443  * For small machines we want it small (128k min).  For large machines
6444  * we want it large (64MB max).  But it is not linear, because network
6445  * bandwidth does not increase linearly with machine size.  We use
6446  *
6447  *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6448  *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
6449  *
6450  * which yields
6451  *
6452  * 16MB:        512k
6453  * 32MB:        724k
6454  * 64MB:        1024k
6455  * 128MB:       1448k
6456  * 256MB:       2048k
6457  * 512MB:       2896k
6458  * 1024MB:      4096k
6459  * 2048MB:      5792k
6460  * 4096MB:      8192k
6461  * 8192MB:      11584k
6462  * 16384MB:     16384k
6463  */
6464 int __meminit init_per_zone_wmark_min(void)
6465 {
6466         unsigned long lowmem_kbytes;
6467         int new_min_free_kbytes;
6468 
6469         lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
6470         new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
6471 
6472         if (new_min_free_kbytes > user_min_free_kbytes) {
6473                 min_free_kbytes = new_min_free_kbytes;
6474                 if (min_free_kbytes < 128)
6475                         min_free_kbytes = 128;
6476                 if (min_free_kbytes > 65536)
6477                         min_free_kbytes = 65536;
6478         } else {
6479                 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6480                                 new_min_free_kbytes, user_min_free_kbytes);
6481         }
6482         setup_per_zone_wmarks();
6483         refresh_zone_stat_thresholds();
6484         setup_per_zone_lowmem_reserve();
6485         setup_per_zone_inactive_ratio();
6486         return 0;
6487 }
6488 core_initcall(init_per_zone_wmark_min)
6489 
6490 /*
6491  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6492  *      that we can call two helper functions whenever min_free_kbytes
6493  *      changes.
6494  */
6495 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
6496         void __user *buffer, size_t *length, loff_t *ppos)
6497 {
6498         int rc;
6499 
6500         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6501         if (rc)
6502                 return rc;
6503 
6504         if (write) {
6505                 user_min_free_kbytes = min_free_kbytes;
6506                 setup_per_zone_wmarks();
6507         }
6508         return 0;
6509 }
6510 
6511 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
6512         void __user *buffer, size_t *length, loff_t *ppos)
6513 {
6514         int rc;
6515 
6516         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6517         if (rc)
6518                 return rc;
6519 
6520         if (write)
6521                 setup_per_zone_wmarks();
6522 
6523         return 0;
6524 }
6525 
6526 #ifdef CONFIG_NUMA
6527 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
6528         void __user *buffer, size_t *length, loff_t *ppos)
6529 {
6530         struct zone *zone;
6531         int rc;
6532 
6533         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6534         if (rc)
6535                 return rc;
6536 
6537         for_each_zone(zone)
6538                 zone->min_unmapped_pages = (zone->managed_pages *
6539                                 sysctl_min_unmapped_ratio) / 100;
6540         return 0;
6541 }
6542 
6543 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
6544         void __user *buffer, size_t *length, loff_t *ppos)
6545 {
6546         struct zone *zone;
6547         int rc;
6548 
6549         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6550         if (rc)
6551                 return rc;
6552 
6553         for_each_zone(zone)
6554                 zone->min_slab_pages = (zone->managed_pages *
6555                                 sysctl_min_slab_ratio) / 100;
6556         return 0;
6557 }
6558 #endif
6559 
6560 /*
6561  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6562  *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6563  *      whenever sysctl_lowmem_reserve_ratio changes.
6564  *
6565  * The reserve ratio obviously has absolutely no relation with the
6566  * minimum watermarks. The lowmem reserve ratio can only make sense
6567  * if in function of the boot time zone sizes.
6568  */
6569 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
6570         void __user *buffer, size_t *length, loff_t *ppos)
6571 {
6572         proc_dointvec_minmax(table, write, buffer, length, ppos);
6573         setup_per_zone_lowmem_reserve();
6574         return 0;
6575 }
6576 
6577 /*
6578  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6579  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
6580  * pagelist can have before it gets flushed back to buddy allocator.
6581  */
6582 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
6583         void __user *buffer, size_t *length, loff_t *ppos)
6584 {
6585         struct zone *zone;
6586         int old_percpu_pagelist_fraction;
6587         int ret;
6588 
6589         mutex_lock(&pcp_batch_high_lock);
6590         old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6591 
6592         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6593         if (!write || ret < 0)
6594                 goto out;
6595 
6596         /* Sanity checking to avoid pcp imbalance */
6597         if (percpu_pagelist_fraction &&
6598             percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6599                 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6600                 ret = -EINVAL;
6601                 goto out;
6602         }
6603 
6604         /* No change? */
6605         if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6606                 goto out;
6607 
6608         for_each_populated_zone(zone) {
6609                 unsigned int cpu;
6610 
6611                 for_each_possible_cpu(cpu)
6612                         pageset_set_high_and_batch(zone,
6613                                         per_cpu_ptr(zone->pageset, cpu));
6614         }
6615 out:
6616         mutex_unlock(&pcp_batch_high_lock);
6617         return ret;
6618 }
6619 
6620 #ifdef CONFIG_NUMA
6621 int hashdist = HASHDIST_DEFAULT;
6622 
6623 static int __init set_hashdist(char *str)
6624 {
6625         if (!str)
6626                 return 0;
6627         hashdist = simple_strtoul(str, &str, 0);
6628         return 1;
6629 }
6630 __setup("hashdist=", set_hashdist);
6631 #endif
6632 
6633 /*
6634  * allocate a large system hash table from bootmem
6635  * - it is assumed that the hash table must contain an exact power-of-2
6636  *   quantity of entries
6637  * - limit is the number of hash buckets, not the total allocation size
6638  */
6639 void *__init alloc_large_system_hash(const char *tablename,
6640                                      unsigned long bucketsize,
6641                                      unsigned long numentries,
6642                                      int scale,
6643                                      int flags,
6644                                      unsigned int *_hash_shift,
6645                                      unsigned int *_hash_mask,
6646                                      unsigned long low_limit,
6647                                      unsigned long high_limit)
6648 {
6649         unsigned long long max = high_limit;
6650         unsigned long log2qty, size;
6651         void *table = NULL;
6652 
6653         /* allow the kernel cmdline to have a say */
6654         if (!numentries) {
6655                 /* round applicable memory size up to nearest megabyte */
6656                 numentries = nr_kernel_pages;
6657 
6658                 /* It isn't necessary when PAGE_SIZE >= 1MB */
6659                 if (PAGE_SHIFT < 20)
6660                         numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
6661 
6662                 /* limit to 1 bucket per 2^scale bytes of low memory */
6663                 if (scale > PAGE_SHIFT)
6664                         numentries >>= (scale - PAGE_SHIFT);
6665                 else
6666                         numentries <<= (PAGE_SHIFT - scale);
6667 
6668                 /* Make sure we've got at least a 0-order allocation.. */
6669                 if (unlikely(flags & HASH_SMALL)) {
6670                         /* Makes no sense without HASH_EARLY */
6671                         WARN_ON(!(flags & HASH_EARLY));
6672                         if (!(numentries >> *_hash_shift)) {
6673                                 numentries = 1UL << *_hash_shift;
6674                                 BUG_ON(!numentries);
6675                         }
6676                 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
6677                         numentries = PAGE_SIZE / bucketsize;
6678         }
6679         numentries = roundup_pow_of_two(numentries);
6680 
6681         /* limit allocation size to 1/16 total memory by default */
6682         if (max == 0) {
6683                 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6684                 do_div(max, bucketsize);
6685         }
6686         max = min(max, 0x80000000ULL);
6687 
6688         if (numentries < low_limit)
6689                 numentries = low_limit;
6690         if (numentries > max)
6691                 numentries = max;
6692 
6693         log2qty = ilog2(numentries);
6694 
6695         do {
6696                 size = bucketsize << log2qty;
6697                 if (flags & HASH_EARLY)
6698                         table = memblock_virt_alloc_nopanic(size, 0);
6699                 else if (hashdist)
6700                         table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6701                 else {
6702                         /*
6703                          * If bucketsize is not a power-of-two, we may free
6704                          * some pages at the end of hash table which
6705                          * alloc_pages_exact() automatically does
6706                          */
6707                         if (get_order(size) < MAX_ORDER) {
6708                                 table = alloc_pages_exact(size, GFP_ATOMIC);
6709                                 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6710                         }
6711                 }
6712         } while (!table && size > PAGE_SIZE && --log2qty);
6713 
6714         if (!table)
6715                 panic("Failed to allocate %s hash table\n", tablename);
6716 
6717         pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
6718                 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
6719 
6720         if (_hash_shift)
6721                 *_hash_shift = log2qty;
6722         if (_hash_mask)
6723                 *_hash_mask = (1 << log2qty) - 1;
6724 
6725         return table;
6726 }
6727 
6728 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6729 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6730                                                         unsigned long pfn)
6731 {
6732 #ifdef CONFIG_SPARSEMEM
6733         return __pfn_to_section(pfn)->pageblock_flags;
6734 #else
6735         return zone->pageblock_flags;
6736 #endif /* CONFIG_SPARSEMEM */
6737 }
6738 
6739 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6740 {
6741 #ifdef CONFIG_SPARSEMEM
6742         pfn &= (PAGES_PER_SECTION-1);
6743         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6744 #else
6745         pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
6746         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6747 #endif /* CONFIG_SPARSEMEM */
6748 }
6749 
6750 /**
6751  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6752  * @page: The page within the block of interest
6753  * @pfn: The target page frame number
6754  * @end_bitidx: The last bit of interest to retrieve
6755  * @mask: mask of bits that the caller is interested in
6756  *
6757  * Return: pageblock_bits flags
6758  */
6759 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
6760                                         unsigned long end_bitidx,
6761                                         unsigned long mask)
6762 {
6763         struct zone *zone;
6764         unsigned long *bitmap;
6765         unsigned long bitidx, word_bitidx;
6766         unsigned long word;
6767 
6768         zone = page_zone(page);
6769         bitmap = get_pageblock_bitmap(zone, pfn);
6770         bitidx = pfn_to_bitidx(zone, pfn);
6771         word_bitidx = bitidx / BITS_PER_LONG;
6772         bitidx &= (BITS_PER_LONG-1);
6773 
6774         word = bitmap[word_bitidx];
6775         bitidx += end_bitidx;
6776         return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
6777 }
6778 
6779 /**
6780  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6781  * @page: The page within the block of interest
6782  * @flags: The flags to set
6783  * @pfn: The target page frame number
6784  * @end_bitidx: The last bit of interest
6785  * @mask: mask of bits that the caller is interested in
6786  */
6787 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6788                                         unsigned long pfn,
6789                                         unsigned long end_bitidx,
6790                                         unsigned long mask)
6791 {
6792         struct zone *zone;
6793         unsigned long *bitmap;
6794         unsigned long bitidx, word_bitidx;
6795         unsigned long old_word, word;
6796 
6797         BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
6798 
6799         zone = page_zone(page);
6800         bitmap = get_pageblock_bitmap(zone, pfn);
6801         bitidx = pfn_to_bitidx(zone, pfn);
6802         word_bitidx = bitidx / BITS_PER_LONG;
6803         bitidx &= (BITS_PER_LONG-1);
6804 
6805         VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
6806 
6807         bitidx += end_bitidx;
6808         mask <<= (BITS_PER_LONG - bitidx - 1);
6809         flags <<= (BITS_PER_LONG - bitidx - 1);
6810 
6811         word = READ_ONCE(bitmap[word_bitidx]);
6812         for (;;) {
6813                 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6814                 if (word == old_word)
6815                         break;
6816                 word = old_word;
6817         }
6818 }
6819 
6820 /*
6821  * This function checks whether pageblock includes unmovable pages or not.
6822  * If @count is not zero, it is okay to include less @count unmovable pages
6823  *
6824  * PageLRU check without isolation or lru_lock could race so that
6825  * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6826  * expect this function should be exact.
6827  */
6828 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6829                          bool skip_hwpoisoned_pages)
6830 {
6831         unsigned long pfn, iter, found;
6832         int mt;
6833 
6834         /*
6835          * For avoiding noise data, lru_add_drain_all() should be called
6836          * If ZONE_MOVABLE, the zone never contains unmovable pages
6837          */
6838         if (zone_idx(zone) == ZONE_MOVABLE)
6839                 return false;
6840         mt = get_pageblock_migratetype(page);
6841         if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6842                 return false;
6843 
6844         pfn = page_to_pfn(page);
6845         for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6846                 unsigned long check = pfn + iter;
6847 
6848                 if (!pfn_valid_within(check))
6849                         continue;
6850 
6851                 page = pfn_to_page(check);
6852 
6853                 /*
6854                  * Hugepages are not in LRU lists, but they're movable.
6855                  * We need not scan over tail pages bacause we don't
6856                  * handle each tail page individually in migration.
6857                  */
6858                 if (PageHuge(page)) {
6859                         iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6860                         continue;
6861                 }
6862 
6863                 /*
6864                  * We can't use page_count without pin a page
6865                  * because another CPU can free compound page.
6866                  * This check already skips compound tails of THP
6867                  * because their page->_count is zero at all time.
6868                  */
6869                 if (!page_ref_count(page)) {
6870                         if (PageBuddy(page))
6871                                 iter += (1 << page_order(page)) - 1;
6872                         continue;
6873                 }
6874 
6875                 /*
6876                  * The HWPoisoned page may be not in buddy system, and
6877                  * page_count() is not 0.
6878                  */
6879                 if (skip_hwpoisoned_pages && PageHWPoison(page))
6880                         continue;
6881 
6882                 if (!PageLRU(page))
6883                         found++;
6884                 /*
6885                  * If there are RECLAIMABLE pages, we need to check
6886                  * it.  But now, memory offline itself doesn't call
6887                  * shrink_node_slabs() and it still to be fixed.
6888                  */
6889                 /*
6890                  * If the page is not RAM, page_count()should be 0.
6891                  * we don't need more check. This is an _used_ not-movable page.
6892                  *
6893                  * The problematic thing here is PG_reserved pages. PG_reserved
6894                  * is set to both of a memory hole page and a _used_ kernel
6895                  * page at boot.
6896                  */
6897                 if (found > count)
6898                         return true;
6899         }
6900         return false;
6901 }
6902 
6903 bool is_pageblock_removable_nolock(struct page *page)
6904 {
6905         struct zone *zone;
6906         unsigned long pfn;
6907 
6908         /*
6909          * We have to be careful here because we are iterating over memory
6910          * sections which are not zone aware so we might end up outside of
6911          * the zone but still within the section.
6912          * We have to take care about the node as well. If the node is offline
6913          * its NODE_DATA will be NULL - see page_zone.
6914          */
6915         if (!node_online(page_to_nid(page)))
6916                 return false;
6917 
6918         zone = page_zone(page);
6919         pfn = page_to_pfn(page);
6920         if (!zone_spans_pfn(zone, pfn))
6921                 return false;
6922 
6923         return !has_unmovable_pages(zone, page, 0, true);
6924 }
6925 
6926 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
6927 
6928 static unsigned long pfn_max_align_down(unsigned long pfn)
6929 {
6930         return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6931                              pageblock_nr_pages) - 1);
6932 }
6933 
6934 static unsigned long pfn_max_align_up(unsigned long pfn)
6935 {
6936         return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6937                                 pageblock_nr_pages));
6938 }
6939 
6940 /* [start, end) must belong to a single zone. */
6941 static int __alloc_contig_migrate_range(struct compact_control *cc,
6942                                         unsigned long start, unsigned long end)
6943 {
6944         /* This function is based on compact_zone() from compaction.c. */
6945         unsigned long nr_reclaimed;
6946         unsigned long pfn = start;
6947         unsigned int tries = 0;
6948         int ret = 0;
6949 
6950         migrate_prep();
6951 
6952         while (pfn < end || !list_empty(&cc->migratepages)) {
6953                 if (fatal_signal_pending(current)) {
6954                         ret = -EINTR;
6955                         break;
6956                 }
6957 
6958                 if (list_empty(&cc->migratepages)) {
6959                         cc->nr_migratepages = 0;
6960                         pfn = isolate_migratepages_range(cc, pfn, end);
6961                         if (!pfn) {
6962                                 ret = -EINTR;
6963                                 break;
6964                         }
6965                         tries = 0;
6966                 } else if (++tries == 5) {
6967                         ret = ret < 0 ? ret : -EBUSY;
6968                         break;
6969                 }
6970 
6971                 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6972                                                         &cc->migratepages);
6973                 cc->nr_migratepages -= nr_reclaimed;
6974 
6975                 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6976                                     NULL, 0, cc->mode, MR_CMA);
6977         }
6978         if (ret < 0) {
6979                 putback_movable_pages(&cc->migratepages);
6980                 return ret;
6981         }
6982         return 0;
6983 }
6984 
6985 /**
6986  * alloc_contig_range() -- tries to allocate given range of pages
6987  * @start:      start PFN to allocate
6988  * @end:        one-past-the-last PFN to allocate
6989  * @migratetype:        migratetype of the underlaying pageblocks (either
6990  *                      #MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
6991  *                      in range must have the same migratetype and it must
6992  *                      be either of the two.
6993  *
6994  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6995  * aligned, however it's the caller's responsibility to guarantee that
6996  * we are the only thread that changes migrate type of pageblocks the
6997  * pages fall in.
6998  *
6999  * The PFN range must belong to a single zone.
7000  *
7001  * Returns zero on success or negative error code.  On success all
7002  * pages which PFN is in [start, end) are allocated for the caller and
7003  * need to be freed with free_contig_range().
7004  */
7005 int alloc_contig_range(unsigned long start, unsigned long end,
7006                        unsigned migratetype)
7007 {
7008         unsigned long outer_start, outer_end;
7009         unsigned int order;
7010         int ret = 0;
7011 
7012         struct compact_control cc = {
7013                 .nr_migratepages = 0,
7014                 .order = -1,
7015                 .zone = page_zone(pfn_to_page(start)),
7016                 .mode = MIGRATE_SYNC,
7017                 .ignore_skip_hint = true,
7018         };
7019         INIT_LIST_HEAD(&cc.migratepages);
7020 
7021         /*
7022          * What we do here is we mark all pageblocks in range as
7023          * MIGRATE_ISOLATE.  Because pageblock and max order pages may
7024          * have different sizes, and due to the way page allocator
7025          * work, we align the range to biggest of the two pages so
7026          * that page allocator won't try to merge buddies from
7027          * different pageblocks and change MIGRATE_ISOLATE to some
7028          * other migration type.
7029          *
7030          * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7031          * migrate the pages from an unaligned range (ie. pages that
7032          * we are interested in).  This will put all the pages in
7033          * range back to page allocator as MIGRATE_ISOLATE.
7034          *
7035          * When this is done, we take the pages in range from page
7036          * allocator removing them from the buddy system.  This way
7037          * page allocator will never consider using them.
7038          *
7039          * This lets us mark the pageblocks back as
7040          * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7041          * aligned range but not in the unaligned, original range are
7042          * put back to page allocator so that buddy can use them.
7043          */
7044 
7045         ret = start_isolate_page_range(pfn_max_align_down(start),
7046                                        pfn_max_align_up(end), migratetype,
7047                                        false);
7048         if (ret)
7049                 return ret;
7050 
7051         /*
7052          * In case of -EBUSY, we'd like to know which page causes problem.
7053          * So, just fall through. We will check it in test_pages_isolated().
7054          */
7055         ret = __alloc_contig_migrate_range(&cc, start, end);
7056         if (ret && ret != -EBUSY)
7057                 goto done;
7058 
7059         /*
7060          * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7061          * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
7062          * more, all pages in [start, end) are free in page allocator.
7063          * What we are going to do is to allocate all pages from
7064          * [start, end) (that is remove them from page allocator).
7065          *
7066          * The only problem is that pages at the beginning and at the
7067          * end of interesting range may be not aligned with pages that
7068          * page allocator holds, ie. they can be part of higher order
7069          * pages.  Because of this, we reserve the bigger range and
7070          * once this is done free the pages we are not interested in.
7071          *
7072          * We don't have to hold zone->lock here because the pages are
7073          * isolated thus they won't get removed from buddy.
7074          */
7075 
7076         lru_add_drain_all();
7077         drain_all_pages(cc.zone);
7078 
7079         order = 0;
7080         outer_start = start;
7081         while (!PageBuddy(pfn_to_page(outer_start))) {
7082                 if (++order >= MAX_ORDER) {
7083                         outer_start = start;
7084                         break;
7085                 }
7086                 outer_start &= ~0UL << order;
7087         }
7088 
7089         if (outer_start != start) {
7090                 order = page_order(pfn_to_page(outer_start));
7091 
7092                 /*
7093                  * outer_start page could be small order buddy page and
7094                  * it doesn't include start page. Adjust outer_start
7095                  * in this case to report failed page properly
7096                  * on tracepoint in test_pages_isolated()
7097                  */
7098                 if (outer_start + (1UL << order) <= start)
7099                         outer_start = start;
7100         }
7101 
7102         /* Make sure the range is really isolated. */
7103         if (test_pages_isolated(outer_start, end, false)) {
7104                 pr_info("%s: [%lx, %lx) PFNs busy\n",
7105                         __func__, outer_start, end);
7106                 ret = -EBUSY;
7107                 goto done;
7108         }
7109 
7110         /* Grab isolated pages from freelists. */
7111         outer_end = isolate_freepages_range(&cc, outer_start, end);
7112         if (!outer_end) {
7113                 ret = -EBUSY;
7114                 goto done;
7115         }
7116 
7117         /* Free head and tail (if any) */
7118         if (start != outer_start)
7119                 free_contig_range(outer_start, start - outer_start);
7120         if (end != outer_end)
7121                 free_contig_range(end, outer_end - end);
7122 
7123 done:
7124         undo_isolate_page_range(pfn_max_align_down(start),
7125                                 pfn_max_align_up(end), migratetype);
7126         return ret;
7127 }
7128 
7129 void free_contig_range(unsigned long pfn, unsigned nr_pages)
7130 {
7131         unsigned int count = 0;
7132 
7133         for (; nr_pages--; pfn++) {
7134                 struct page *page = pfn_to_page(pfn);
7135 
7136                 count += page_count(page) != 1;
7137                 __free_page(page);
7138         }
7139         WARN(count != 0, "%d pages are still in use!\n", count);
7140 }
7141 #endif
7142 
7143 #ifdef CONFIG_MEMORY_HOTPLUG
7144 /*
7145  * The zone indicated has a new number of managed_pages; batch sizes and percpu
7146  * page high values need to be recalulated.
7147  */
7148 void __meminit zone_pcp_update(struct zone *zone)
7149 {
7150         unsigned cpu;
7151         mutex_lock(&pcp_batch_high_lock);
7152         for_each_possible_cpu(cpu)
7153                 pageset_set_high_and_batch(zone,
7154                                 per_cpu_ptr(zone->pageset, cpu));
7155         mutex_unlock(&pcp_batch_high_lock);
7156 }
7157 #endif
7158 
7159 void zone_pcp_reset(struct zone *zone)
7160 {
7161         unsigned long flags;
7162         int cpu;
7163         struct per_cpu_pageset *pset;
7164 
7165         /* avoid races with drain_pages()  */
7166         local_irq_save(flags);
7167         if (zone->pageset != &boot_pageset) {
7168                 for_each_online_cpu(cpu) {
7169                         pset = per_cpu_ptr(zone->pageset, cpu);
7170                         drain_zonestat(zone, pset);
7171                 }
7172                 free_percpu(zone->pageset);
7173                 zone->pageset = &boot_pageset;
7174         }
7175         local_irq_restore(flags);
7176 }
7177 
7178 #ifdef CONFIG_MEMORY_HOTREMOVE
7179 /*
7180  * All pages in the range must be isolated before calling this.
7181  */
7182 void
7183 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7184 {
7185         struct page *page;
7186         struct zone *zone;
7187         unsigned int order, i;
7188         unsigned long pfn;
7189         unsigned long flags;
7190         /* find the first valid pfn */
7191         for (pfn = start_pfn; pfn < end_pfn; pfn++)
7192                 if (pfn_valid(pfn))
7193                         break;
7194         if (pfn == end_pfn)
7195                 return;
7196         zone = page_zone(pfn_to_page(pfn));
7197         spin_lock_irqsave(&zone->lock, flags);
7198         pfn = start_pfn;
7199         while (pfn < end_pfn) {
7200                 if (!pfn_valid(pfn)) {
7201                         pfn++;
7202                         continue;
7203                 }
7204                 page = pfn_to_page(pfn);
7205                 /*
7206                  * The HWPoisoned page may be not in buddy system, and
7207                  * page_count() is not 0.
7208                  */
7209                 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7210                         pfn++;
7211                         SetPageReserved(page);
7212                         continue;
7213                 }
7214 
7215                 BUG_ON(page_count(page));
7216                 BUG_ON(!PageBuddy(page));
7217                 order = page_order(page);
7218 #ifdef CONFIG_DEBUG_VM
7219                 pr_info("remove from free list %lx %d %lx\n",
7220                         pfn, 1 << order, end_pfn);
7221 #endif
7222                 list_del(&page->lru);
7223                 rmv_page_order(page);
7224                 zone->free_area[order].nr_free--;
7225                 for (i = 0; i < (1 << order); i++)
7226                         SetPageReserved((page+i));
7227                 pfn += (1 << order);
7228         }
7229         spin_unlock_irqrestore(&zone->lock, flags);
7230 }
7231 #endif
7232 
7233 bool is_free_buddy_page(struct page *page)
7234 {
7235         struct zone *zone = page_zone(page);
7236         unsigned long pfn = page_to_pfn(page);
7237         unsigned long flags;
7238         unsigned int order;
7239 
7240         spin_lock_irqsave(&zone->lock, flags);
7241         for (order = 0; order < MAX_ORDER; order++) {
7242                 struct page *page_head = page - (pfn & ((1 << order) - 1));
7243 
7244                 if (PageBuddy(page_head) && page_order(page_head) >= order)
7245                         break;
7246         }
7247         spin_unlock_irqrestore(&zone->lock, flags);
7248 
7249         return order < MAX_ORDER;
7250 }
7251 

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