~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/page_alloc.c

Version: ~ [ linux-5.2-rc5 ] ~ [ linux-5.1.11 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.52 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.127 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.182 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.182 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.68 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  linux/mm/page_alloc.c
  3  *
  4  *  Manages the free list, the system allocates free pages here.
  5  *  Note that kmalloc() lives in slab.c
  6  *
  7  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  8  *  Swap reorganised 29.12.95, Stephen Tweedie
  9  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
 10  *  Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
 11  *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
 12  *  Zone balancing, Kanoj Sarcar, SGI, Jan 2000
 13  *  Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
 14  *          (lots of bits borrowed from Ingo Molnar & Andrew Morton)
 15  */
 16 
 17 #include <linux/stddef.h>
 18 #include <linux/mm.h>
 19 #include <linux/swap.h>
 20 #include <linux/interrupt.h>
 21 #include <linux/pagemap.h>
 22 #include <linux/jiffies.h>
 23 #include <linux/bootmem.h>
 24 #include <linux/memblock.h>
 25 #include <linux/compiler.h>
 26 #include <linux/kernel.h>
 27 #include <linux/kmemcheck.h>
 28 #include <linux/kasan.h>
 29 #include <linux/module.h>
 30 #include <linux/suspend.h>
 31 #include <linux/pagevec.h>
 32 #include <linux/blkdev.h>
 33 #include <linux/slab.h>
 34 #include <linux/ratelimit.h>
 35 #include <linux/oom.h>
 36 #include <linux/notifier.h>
 37 #include <linux/topology.h>
 38 #include <linux/sysctl.h>
 39 #include <linux/cpu.h>
 40 #include <linux/cpuset.h>
 41 #include <linux/memory_hotplug.h>
 42 #include <linux/nodemask.h>
 43 #include <linux/vmalloc.h>
 44 #include <linux/vmstat.h>
 45 #include <linux/mempolicy.h>
 46 #include <linux/stop_machine.h>
 47 #include <linux/sort.h>
 48 #include <linux/pfn.h>
 49 #include <linux/backing-dev.h>
 50 #include <linux/fault-inject.h>
 51 #include <linux/page-isolation.h>
 52 #include <linux/page_ext.h>
 53 #include <linux/debugobjects.h>
 54 #include <linux/kmemleak.h>
 55 #include <linux/compaction.h>
 56 #include <trace/events/kmem.h>
 57 #include <linux/prefetch.h>
 58 #include <linux/mm_inline.h>
 59 #include <linux/migrate.h>
 60 #include <linux/page_ext.h>
 61 #include <linux/hugetlb.h>
 62 #include <linux/sched/rt.h>
 63 #include <linux/page_owner.h>
 64 #include <linux/kthread.h>
 65 
 66 #include <asm/sections.h>
 67 #include <asm/tlbflush.h>
 68 #include <asm/div64.h>
 69 #include "internal.h"
 70 
 71 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
 72 static DEFINE_MUTEX(pcp_batch_high_lock);
 73 #define MIN_PERCPU_PAGELIST_FRACTION    (8)
 74 
 75 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
 76 DEFINE_PER_CPU(int, numa_node);
 77 EXPORT_PER_CPU_SYMBOL(numa_node);
 78 #endif
 79 
 80 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
 81 /*
 82  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
 83  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
 84  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
 85  * defined in <linux/topology.h>.
 86  */
 87 DEFINE_PER_CPU(int, _numa_mem_);                /* Kernel "local memory" node */
 88 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
 89 int _node_numa_mem_[MAX_NUMNODES];
 90 #endif
 91 
 92 /*
 93  * Array of node states.
 94  */
 95 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
 96         [N_POSSIBLE] = NODE_MASK_ALL,
 97         [N_ONLINE] = { { [0] = 1UL } },
 98 #ifndef CONFIG_NUMA
 99         [N_NORMAL_MEMORY] = { { [0] = 1UL } },
100 #ifdef CONFIG_HIGHMEM
101         [N_HIGH_MEMORY] = { { [0] = 1UL } },
102 #endif
103 #ifdef CONFIG_MOVABLE_NODE
104         [N_MEMORY] = { { [0] = 1UL } },
105 #endif
106         [N_CPU] = { { [0] = 1UL } },
107 #endif  /* NUMA */
108 };
109 EXPORT_SYMBOL(node_states);
110 
111 /* Protect totalram_pages and zone->managed_pages */
112 static DEFINE_SPINLOCK(managed_page_count_lock);
113 
114 unsigned long totalram_pages __read_mostly;
115 unsigned long totalreserve_pages __read_mostly;
116 unsigned long totalcma_pages __read_mostly;
117 /*
118  * When calculating the number of globally allowed dirty pages, there
119  * is a certain number of per-zone reserves that should not be
120  * considered dirtyable memory.  This is the sum of those reserves
121  * over all existing zones that contribute dirtyable memory.
122  */
123 unsigned long dirty_balance_reserve __read_mostly;
124 
125 int percpu_pagelist_fraction;
126 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
127 
128 /*
129  * A cached value of the page's pageblock's migratetype, used when the page is
130  * put on a pcplist. Used to avoid the pageblock migratetype lookup when
131  * freeing from pcplists in most cases, at the cost of possibly becoming stale.
132  * Also the migratetype set in the page does not necessarily match the pcplist
133  * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
134  * other index - this ensures that it will be put on the correct CMA freelist.
135  */
136 static inline int get_pcppage_migratetype(struct page *page)
137 {
138         return page->index;
139 }
140 
141 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
142 {
143         page->index = migratetype;
144 }
145 
146 #ifdef CONFIG_PM_SLEEP
147 /*
148  * The following functions are used by the suspend/hibernate code to temporarily
149  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
150  * while devices are suspended.  To avoid races with the suspend/hibernate code,
151  * they should always be called with pm_mutex held (gfp_allowed_mask also should
152  * only be modified with pm_mutex held, unless the suspend/hibernate code is
153  * guaranteed not to run in parallel with that modification).
154  */
155 
156 static gfp_t saved_gfp_mask;
157 
158 void pm_restore_gfp_mask(void)
159 {
160         WARN_ON(!mutex_is_locked(&pm_mutex));
161         if (saved_gfp_mask) {
162                 gfp_allowed_mask = saved_gfp_mask;
163                 saved_gfp_mask = 0;
164         }
165 }
166 
167 void pm_restrict_gfp_mask(void)
168 {
169         WARN_ON(!mutex_is_locked(&pm_mutex));
170         WARN_ON(saved_gfp_mask);
171         saved_gfp_mask = gfp_allowed_mask;
172         gfp_allowed_mask &= ~GFP_IOFS;
173 }
174 
175 bool pm_suspended_storage(void)
176 {
177         if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
178                 return false;
179         return true;
180 }
181 #endif /* CONFIG_PM_SLEEP */
182 
183 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
184 int pageblock_order __read_mostly;
185 #endif
186 
187 static void __free_pages_ok(struct page *page, unsigned int order);
188 
189 /*
190  * results with 256, 32 in the lowmem_reserve sysctl:
191  *      1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
192  *      1G machine -> (16M dma, 784M normal, 224M high)
193  *      NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
194  *      HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
195  *      HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
196  *
197  * TBD: should special case ZONE_DMA32 machines here - in those we normally
198  * don't need any ZONE_NORMAL reservation
199  */
200 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
201 #ifdef CONFIG_ZONE_DMA
202          256,
203 #endif
204 #ifdef CONFIG_ZONE_DMA32
205          256,
206 #endif
207 #ifdef CONFIG_HIGHMEM
208          32,
209 #endif
210          32,
211 };
212 
213 EXPORT_SYMBOL(totalram_pages);
214 
215 static char * const zone_names[MAX_NR_ZONES] = {
216 #ifdef CONFIG_ZONE_DMA
217          "DMA",
218 #endif
219 #ifdef CONFIG_ZONE_DMA32
220          "DMA32",
221 #endif
222          "Normal",
223 #ifdef CONFIG_HIGHMEM
224          "HighMem",
225 #endif
226          "Movable",
227 #ifdef CONFIG_ZONE_DEVICE
228          "Device",
229 #endif
230 };
231 
232 int min_free_kbytes = 1024;
233 int user_min_free_kbytes = -1;
234 
235 static unsigned long __meminitdata nr_kernel_pages;
236 static unsigned long __meminitdata nr_all_pages;
237 static unsigned long __meminitdata dma_reserve;
238 
239 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
240 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
241 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
242 static unsigned long __initdata required_kernelcore;
243 static unsigned long __initdata required_movablecore;
244 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
245 
246 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
247 int movable_zone;
248 EXPORT_SYMBOL(movable_zone);
249 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
250 
251 #if MAX_NUMNODES > 1
252 int nr_node_ids __read_mostly = MAX_NUMNODES;
253 int nr_online_nodes __read_mostly = 1;
254 EXPORT_SYMBOL(nr_node_ids);
255 EXPORT_SYMBOL(nr_online_nodes);
256 #endif
257 
258 int page_group_by_mobility_disabled __read_mostly;
259 
260 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
261 static inline void reset_deferred_meminit(pg_data_t *pgdat)
262 {
263         pgdat->first_deferred_pfn = ULONG_MAX;
264 }
265 
266 /* Returns true if the struct page for the pfn is uninitialised */
267 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
268 {
269         if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
270                 return true;
271 
272         return false;
273 }
274 
275 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
276 {
277         if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
278                 return true;
279 
280         return false;
281 }
282 
283 /*
284  * Returns false when the remaining initialisation should be deferred until
285  * later in the boot cycle when it can be parallelised.
286  */
287 static inline bool update_defer_init(pg_data_t *pgdat,
288                                 unsigned long pfn, unsigned long zone_end,
289                                 unsigned long *nr_initialised)
290 {
291         /* Always populate low zones for address-contrained allocations */
292         if (zone_end < pgdat_end_pfn(pgdat))
293                 return true;
294 
295         /* Initialise at least 2G of the highest zone */
296         (*nr_initialised)++;
297         if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) &&
298             (pfn & (PAGES_PER_SECTION - 1)) == 0) {
299                 pgdat->first_deferred_pfn = pfn;
300                 return false;
301         }
302 
303         return true;
304 }
305 #else
306 static inline void reset_deferred_meminit(pg_data_t *pgdat)
307 {
308 }
309 
310 static inline bool early_page_uninitialised(unsigned long pfn)
311 {
312         return false;
313 }
314 
315 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
316 {
317         return false;
318 }
319 
320 static inline bool update_defer_init(pg_data_t *pgdat,
321                                 unsigned long pfn, unsigned long zone_end,
322                                 unsigned long *nr_initialised)
323 {
324         return true;
325 }
326 #endif
327 
328 
329 void set_pageblock_migratetype(struct page *page, int migratetype)
330 {
331         if (unlikely(page_group_by_mobility_disabled &&
332                      migratetype < MIGRATE_PCPTYPES))
333                 migratetype = MIGRATE_UNMOVABLE;
334 
335         set_pageblock_flags_group(page, (unsigned long)migratetype,
336                                         PB_migrate, PB_migrate_end);
337 }
338 
339 #ifdef CONFIG_DEBUG_VM
340 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
341 {
342         int ret = 0;
343         unsigned seq;
344         unsigned long pfn = page_to_pfn(page);
345         unsigned long sp, start_pfn;
346 
347         do {
348                 seq = zone_span_seqbegin(zone);
349                 start_pfn = zone->zone_start_pfn;
350                 sp = zone->spanned_pages;
351                 if (!zone_spans_pfn(zone, pfn))
352                         ret = 1;
353         } while (zone_span_seqretry(zone, seq));
354 
355         if (ret)
356                 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
357                         pfn, zone_to_nid(zone), zone->name,
358                         start_pfn, start_pfn + sp);
359 
360         return ret;
361 }
362 
363 static int page_is_consistent(struct zone *zone, struct page *page)
364 {
365         if (!pfn_valid_within(page_to_pfn(page)))
366                 return 0;
367         if (zone != page_zone(page))
368                 return 0;
369 
370         return 1;
371 }
372 /*
373  * Temporary debugging check for pages not lying within a given zone.
374  */
375 static int bad_range(struct zone *zone, struct page *page)
376 {
377         if (page_outside_zone_boundaries(zone, page))
378                 return 1;
379         if (!page_is_consistent(zone, page))
380                 return 1;
381 
382         return 0;
383 }
384 #else
385 static inline int bad_range(struct zone *zone, struct page *page)
386 {
387         return 0;
388 }
389 #endif
390 
391 static void bad_page(struct page *page, const char *reason,
392                 unsigned long bad_flags)
393 {
394         static unsigned long resume;
395         static unsigned long nr_shown;
396         static unsigned long nr_unshown;
397 
398         /* Don't complain about poisoned pages */
399         if (PageHWPoison(page)) {
400                 page_mapcount_reset(page); /* remove PageBuddy */
401                 return;
402         }
403 
404         /*
405          * Allow a burst of 60 reports, then keep quiet for that minute;
406          * or allow a steady drip of one report per second.
407          */
408         if (nr_shown == 60) {
409                 if (time_before(jiffies, resume)) {
410                         nr_unshown++;
411                         goto out;
412                 }
413                 if (nr_unshown) {
414                         printk(KERN_ALERT
415                               "BUG: Bad page state: %lu messages suppressed\n",
416                                 nr_unshown);
417                         nr_unshown = 0;
418                 }
419                 nr_shown = 0;
420         }
421         if (nr_shown++ == 0)
422                 resume = jiffies + 60 * HZ;
423 
424         printk(KERN_ALERT "BUG: Bad page state in process %s  pfn:%05lx\n",
425                 current->comm, page_to_pfn(page));
426         dump_page_badflags(page, reason, bad_flags);
427 
428         print_modules();
429         dump_stack();
430 out:
431         /* Leave bad fields for debug, except PageBuddy could make trouble */
432         page_mapcount_reset(page); /* remove PageBuddy */
433         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
434 }
435 
436 /*
437  * Higher-order pages are called "compound pages".  They are structured thusly:
438  *
439  * The first PAGE_SIZE page is called the "head page".
440  *
441  * The remaining PAGE_SIZE pages are called "tail pages".
442  *
443  * All pages have PG_compound set.  All tail pages have their ->first_page
444  * pointing at the head page.
445  *
446  * The first tail page's ->lru.next holds the address of the compound page's
447  * put_page() function.  Its ->lru.prev holds the order of allocation.
448  * This usage means that zero-order pages may not be compound.
449  */
450 
451 static void free_compound_page(struct page *page)
452 {
453         __free_pages_ok(page, compound_order(page));
454 }
455 
456 void prep_compound_page(struct page *page, unsigned long order)
457 {
458         int i;
459         int nr_pages = 1 << order;
460 
461         set_compound_page_dtor(page, free_compound_page);
462         set_compound_order(page, order);
463         __SetPageHead(page);
464         for (i = 1; i < nr_pages; i++) {
465                 struct page *p = page + i;
466                 set_page_count(p, 0);
467                 p->first_page = page;
468                 /* Make sure p->first_page is always valid for PageTail() */
469                 smp_wmb();
470                 __SetPageTail(p);
471         }
472 }
473 
474 #ifdef CONFIG_DEBUG_PAGEALLOC
475 unsigned int _debug_guardpage_minorder;
476 bool _debug_pagealloc_enabled __read_mostly;
477 bool _debug_guardpage_enabled __read_mostly;
478 
479 static int __init early_debug_pagealloc(char *buf)
480 {
481         if (!buf)
482                 return -EINVAL;
483 
484         if (strcmp(buf, "on") == 0)
485                 _debug_pagealloc_enabled = true;
486 
487         return 0;
488 }
489 early_param("debug_pagealloc", early_debug_pagealloc);
490 
491 static bool need_debug_guardpage(void)
492 {
493         /* If we don't use debug_pagealloc, we don't need guard page */
494         if (!debug_pagealloc_enabled())
495                 return false;
496 
497         return true;
498 }
499 
500 static void init_debug_guardpage(void)
501 {
502         if (!debug_pagealloc_enabled())
503                 return;
504 
505         _debug_guardpage_enabled = true;
506 }
507 
508 struct page_ext_operations debug_guardpage_ops = {
509         .need = need_debug_guardpage,
510         .init = init_debug_guardpage,
511 };
512 
513 static int __init debug_guardpage_minorder_setup(char *buf)
514 {
515         unsigned long res;
516 
517         if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
518                 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
519                 return 0;
520         }
521         _debug_guardpage_minorder = res;
522         printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
523         return 0;
524 }
525 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
526 
527 static inline void set_page_guard(struct zone *zone, struct page *page,
528                                 unsigned int order, int migratetype)
529 {
530         struct page_ext *page_ext;
531 
532         if (!debug_guardpage_enabled())
533                 return;
534 
535         page_ext = lookup_page_ext(page);
536         __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
537 
538         INIT_LIST_HEAD(&page->lru);
539         set_page_private(page, order);
540         /* Guard pages are not available for any usage */
541         __mod_zone_freepage_state(zone, -(1 << order), migratetype);
542 }
543 
544 static inline void clear_page_guard(struct zone *zone, struct page *page,
545                                 unsigned int order, int migratetype)
546 {
547         struct page_ext *page_ext;
548 
549         if (!debug_guardpage_enabled())
550                 return;
551 
552         page_ext = lookup_page_ext(page);
553         __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
554 
555         set_page_private(page, 0);
556         if (!is_migrate_isolate(migratetype))
557                 __mod_zone_freepage_state(zone, (1 << order), migratetype);
558 }
559 #else
560 struct page_ext_operations debug_guardpage_ops = { NULL, };
561 static inline void set_page_guard(struct zone *zone, struct page *page,
562                                 unsigned int order, int migratetype) {}
563 static inline void clear_page_guard(struct zone *zone, struct page *page,
564                                 unsigned int order, int migratetype) {}
565 #endif
566 
567 static inline void set_page_order(struct page *page, unsigned int order)
568 {
569         set_page_private(page, order);
570         __SetPageBuddy(page);
571 }
572 
573 static inline void rmv_page_order(struct page *page)
574 {
575         __ClearPageBuddy(page);
576         set_page_private(page, 0);
577 }
578 
579 /*
580  * This function checks whether a page is free && is the buddy
581  * we can do coalesce a page and its buddy if
582  * (a) the buddy is not in a hole &&
583  * (b) the buddy is in the buddy system &&
584  * (c) a page and its buddy have the same order &&
585  * (d) a page and its buddy are in the same zone.
586  *
587  * For recording whether a page is in the buddy system, we set ->_mapcount
588  * PAGE_BUDDY_MAPCOUNT_VALUE.
589  * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
590  * serialized by zone->lock.
591  *
592  * For recording page's order, we use page_private(page).
593  */
594 static inline int page_is_buddy(struct page *page, struct page *buddy,
595                                                         unsigned int order)
596 {
597         if (!pfn_valid_within(page_to_pfn(buddy)))
598                 return 0;
599 
600         if (page_is_guard(buddy) && page_order(buddy) == order) {
601                 if (page_zone_id(page) != page_zone_id(buddy))
602                         return 0;
603 
604                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
605 
606                 return 1;
607         }
608 
609         if (PageBuddy(buddy) && page_order(buddy) == order) {
610                 /*
611                  * zone check is done late to avoid uselessly
612                  * calculating zone/node ids for pages that could
613                  * never merge.
614                  */
615                 if (page_zone_id(page) != page_zone_id(buddy))
616                         return 0;
617 
618                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
619 
620                 return 1;
621         }
622         return 0;
623 }
624 
625 /*
626  * Freeing function for a buddy system allocator.
627  *
628  * The concept of a buddy system is to maintain direct-mapped table
629  * (containing bit values) for memory blocks of various "orders".
630  * The bottom level table contains the map for the smallest allocatable
631  * units of memory (here, pages), and each level above it describes
632  * pairs of units from the levels below, hence, "buddies".
633  * At a high level, all that happens here is marking the table entry
634  * at the bottom level available, and propagating the changes upward
635  * as necessary, plus some accounting needed to play nicely with other
636  * parts of the VM system.
637  * At each level, we keep a list of pages, which are heads of continuous
638  * free pages of length of (1 << order) and marked with _mapcount
639  * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
640  * field.
641  * So when we are allocating or freeing one, we can derive the state of the
642  * other.  That is, if we allocate a small block, and both were
643  * free, the remainder of the region must be split into blocks.
644  * If a block is freed, and its buddy is also free, then this
645  * triggers coalescing into a block of larger size.
646  *
647  * -- nyc
648  */
649 
650 static inline void __free_one_page(struct page *page,
651                 unsigned long pfn,
652                 struct zone *zone, unsigned int order,
653                 int migratetype)
654 {
655         unsigned long page_idx;
656         unsigned long combined_idx;
657         unsigned long uninitialized_var(buddy_idx);
658         struct page *buddy;
659         int max_order = MAX_ORDER;
660 
661         VM_BUG_ON(!zone_is_initialized(zone));
662         VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
663 
664         VM_BUG_ON(migratetype == -1);
665         if (is_migrate_isolate(migratetype)) {
666                 /*
667                  * We restrict max order of merging to prevent merge
668                  * between freepages on isolate pageblock and normal
669                  * pageblock. Without this, pageblock isolation
670                  * could cause incorrect freepage accounting.
671                  */
672                 max_order = min(MAX_ORDER, pageblock_order + 1);
673         } else {
674                 __mod_zone_freepage_state(zone, 1 << order, migratetype);
675         }
676 
677         page_idx = pfn & ((1 << max_order) - 1);
678 
679         VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
680         VM_BUG_ON_PAGE(bad_range(zone, page), page);
681 
682         while (order < max_order - 1) {
683                 buddy_idx = __find_buddy_index(page_idx, order);
684                 buddy = page + (buddy_idx - page_idx);
685                 if (!page_is_buddy(page, buddy, order))
686                         break;
687                 /*
688                  * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
689                  * merge with it and move up one order.
690                  */
691                 if (page_is_guard(buddy)) {
692                         clear_page_guard(zone, buddy, order, migratetype);
693                 } else {
694                         list_del(&buddy->lru);
695                         zone->free_area[order].nr_free--;
696                         rmv_page_order(buddy);
697                 }
698                 combined_idx = buddy_idx & page_idx;
699                 page = page + (combined_idx - page_idx);
700                 page_idx = combined_idx;
701                 order++;
702         }
703         set_page_order(page, order);
704 
705         /*
706          * If this is not the largest possible page, check if the buddy
707          * of the next-highest order is free. If it is, it's possible
708          * that pages are being freed that will coalesce soon. In case,
709          * that is happening, add the free page to the tail of the list
710          * so it's less likely to be used soon and more likely to be merged
711          * as a higher order page
712          */
713         if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
714                 struct page *higher_page, *higher_buddy;
715                 combined_idx = buddy_idx & page_idx;
716                 higher_page = page + (combined_idx - page_idx);
717                 buddy_idx = __find_buddy_index(combined_idx, order + 1);
718                 higher_buddy = higher_page + (buddy_idx - combined_idx);
719                 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
720                         list_add_tail(&page->lru,
721                                 &zone->free_area[order].free_list[migratetype]);
722                         goto out;
723                 }
724         }
725 
726         list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
727 out:
728         zone->free_area[order].nr_free++;
729 }
730 
731 static inline int free_pages_check(struct page *page)
732 {
733         const char *bad_reason = NULL;
734         unsigned long bad_flags = 0;
735 
736         if (unlikely(page_mapcount(page)))
737                 bad_reason = "nonzero mapcount";
738         if (unlikely(page->mapping != NULL))
739                 bad_reason = "non-NULL mapping";
740         if (unlikely(atomic_read(&page->_count) != 0))
741                 bad_reason = "nonzero _count";
742         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
743                 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
744                 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
745         }
746 #ifdef CONFIG_MEMCG
747         if (unlikely(page->mem_cgroup))
748                 bad_reason = "page still charged to cgroup";
749 #endif
750         if (unlikely(bad_reason)) {
751                 bad_page(page, bad_reason, bad_flags);
752                 return 1;
753         }
754         page_cpupid_reset_last(page);
755         if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
756                 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
757         return 0;
758 }
759 
760 /*
761  * Frees a number of pages from the PCP lists
762  * Assumes all pages on list are in same zone, and of same order.
763  * count is the number of pages to free.
764  *
765  * If the zone was previously in an "all pages pinned" state then look to
766  * see if this freeing clears that state.
767  *
768  * And clear the zone's pages_scanned counter, to hold off the "all pages are
769  * pinned" detection logic.
770  */
771 static void free_pcppages_bulk(struct zone *zone, int count,
772                                         struct per_cpu_pages *pcp)
773 {
774         int migratetype = 0;
775         int batch_free = 0;
776         int to_free = count;
777         unsigned long nr_scanned;
778 
779         spin_lock(&zone->lock);
780         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
781         if (nr_scanned)
782                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
783 
784         while (to_free) {
785                 struct page *page;
786                 struct list_head *list;
787 
788                 /*
789                  * Remove pages from lists in a round-robin fashion. A
790                  * batch_free count is maintained that is incremented when an
791                  * empty list is encountered.  This is so more pages are freed
792                  * off fuller lists instead of spinning excessively around empty
793                  * lists
794                  */
795                 do {
796                         batch_free++;
797                         if (++migratetype == MIGRATE_PCPTYPES)
798                                 migratetype = 0;
799                         list = &pcp->lists[migratetype];
800                 } while (list_empty(list));
801 
802                 /* This is the only non-empty list. Free them all. */
803                 if (batch_free == MIGRATE_PCPTYPES)
804                         batch_free = to_free;
805 
806                 do {
807                         int mt; /* migratetype of the to-be-freed page */
808 
809                         page = list_entry(list->prev, struct page, lru);
810                         /* must delete as __free_one_page list manipulates */
811                         list_del(&page->lru);
812 
813                         mt = get_pcppage_migratetype(page);
814                         /* MIGRATE_ISOLATE page should not go to pcplists */
815                         VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
816                         /* Pageblock could have been isolated meanwhile */
817                         if (unlikely(has_isolate_pageblock(zone)))
818                                 mt = get_pageblock_migratetype(page);
819 
820                         /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
821                         __free_one_page(page, page_to_pfn(page), zone, 0, mt);
822                         trace_mm_page_pcpu_drain(page, 0, mt);
823                 } while (--to_free && --batch_free && !list_empty(list));
824         }
825         spin_unlock(&zone->lock);
826 }
827 
828 static void free_one_page(struct zone *zone,
829                                 struct page *page, unsigned long pfn,
830                                 unsigned int order,
831                                 int migratetype)
832 {
833         unsigned long nr_scanned;
834         spin_lock(&zone->lock);
835         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
836         if (nr_scanned)
837                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
838 
839         if (unlikely(has_isolate_pageblock(zone) ||
840                 is_migrate_isolate(migratetype))) {
841                 migratetype = get_pfnblock_migratetype(page, pfn);
842         }
843         __free_one_page(page, pfn, zone, order, migratetype);
844         spin_unlock(&zone->lock);
845 }
846 
847 static int free_tail_pages_check(struct page *head_page, struct page *page)
848 {
849         if (!IS_ENABLED(CONFIG_DEBUG_VM))
850                 return 0;
851         if (unlikely(!PageTail(page))) {
852                 bad_page(page, "PageTail not set", 0);
853                 return 1;
854         }
855         if (unlikely(page->first_page != head_page)) {
856                 bad_page(page, "first_page not consistent", 0);
857                 return 1;
858         }
859         return 0;
860 }
861 
862 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
863                                 unsigned long zone, int nid)
864 {
865         set_page_links(page, zone, nid, pfn);
866         init_page_count(page);
867         page_mapcount_reset(page);
868         page_cpupid_reset_last(page);
869 
870         INIT_LIST_HEAD(&page->lru);
871 #ifdef WANT_PAGE_VIRTUAL
872         /* The shift won't overflow because ZONE_NORMAL is below 4G. */
873         if (!is_highmem_idx(zone))
874                 set_page_address(page, __va(pfn << PAGE_SHIFT));
875 #endif
876 }
877 
878 static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
879                                         int nid)
880 {
881         return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
882 }
883 
884 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
885 static void init_reserved_page(unsigned long pfn)
886 {
887         pg_data_t *pgdat;
888         int nid, zid;
889 
890         if (!early_page_uninitialised(pfn))
891                 return;
892 
893         nid = early_pfn_to_nid(pfn);
894         pgdat = NODE_DATA(nid);
895 
896         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
897                 struct zone *zone = &pgdat->node_zones[zid];
898 
899                 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
900                         break;
901         }
902         __init_single_pfn(pfn, zid, nid);
903 }
904 #else
905 static inline void init_reserved_page(unsigned long pfn)
906 {
907 }
908 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
909 
910 /*
911  * Initialised pages do not have PageReserved set. This function is
912  * called for each range allocated by the bootmem allocator and
913  * marks the pages PageReserved. The remaining valid pages are later
914  * sent to the buddy page allocator.
915  */
916 void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
917 {
918         unsigned long start_pfn = PFN_DOWN(start);
919         unsigned long end_pfn = PFN_UP(end);
920 
921         for (; start_pfn < end_pfn; start_pfn++) {
922                 if (pfn_valid(start_pfn)) {
923                         struct page *page = pfn_to_page(start_pfn);
924 
925                         init_reserved_page(start_pfn);
926                         SetPageReserved(page);
927                 }
928         }
929 }
930 
931 static bool free_pages_prepare(struct page *page, unsigned int order)
932 {
933         bool compound = PageCompound(page);
934         int i, bad = 0;
935 
936         VM_BUG_ON_PAGE(PageTail(page), page);
937         VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
938 
939         trace_mm_page_free(page, order);
940         kmemcheck_free_shadow(page, order);
941         kasan_free_pages(page, order);
942 
943         if (PageAnon(page))
944                 page->mapping = NULL;
945         bad += free_pages_check(page);
946         for (i = 1; i < (1 << order); i++) {
947                 if (compound)
948                         bad += free_tail_pages_check(page, page + i);
949                 bad += free_pages_check(page + i);
950         }
951         if (bad)
952                 return false;
953 
954         reset_page_owner(page, order);
955 
956         if (!PageHighMem(page)) {
957                 debug_check_no_locks_freed(page_address(page),
958                                            PAGE_SIZE << order);
959                 debug_check_no_obj_freed(page_address(page),
960                                            PAGE_SIZE << order);
961         }
962         arch_free_page(page, order);
963         kernel_map_pages(page, 1 << order, 0);
964 
965         return true;
966 }
967 
968 static void __free_pages_ok(struct page *page, unsigned int order)
969 {
970         unsigned long flags;
971         int migratetype;
972         unsigned long pfn = page_to_pfn(page);
973 
974         if (!free_pages_prepare(page, order))
975                 return;
976 
977         migratetype = get_pfnblock_migratetype(page, pfn);
978         local_irq_save(flags);
979         __count_vm_events(PGFREE, 1 << order);
980         free_one_page(page_zone(page), page, pfn, order, migratetype);
981         local_irq_restore(flags);
982 }
983 
984 static void __init __free_pages_boot_core(struct page *page,
985                                         unsigned long pfn, unsigned int order)
986 {
987         unsigned int nr_pages = 1 << order;
988         struct page *p = page;
989         unsigned int loop;
990 
991         prefetchw(p);
992         for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
993                 prefetchw(p + 1);
994                 __ClearPageReserved(p);
995                 set_page_count(p, 0);
996         }
997         __ClearPageReserved(p);
998         set_page_count(p, 0);
999 
1000         page_zone(page)->managed_pages += nr_pages;
1001         set_page_refcounted(page);
1002         __free_pages(page, order);
1003 }
1004 
1005 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1006         defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1007 
1008 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1009 
1010 int __meminit early_pfn_to_nid(unsigned long pfn)
1011 {
1012         static DEFINE_SPINLOCK(early_pfn_lock);
1013         int nid;
1014 
1015         spin_lock(&early_pfn_lock);
1016         nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1017         if (nid < 0)
1018                 nid = 0;
1019         spin_unlock(&early_pfn_lock);
1020 
1021         return nid;
1022 }
1023 #endif
1024 
1025 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1026 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1027                                         struct mminit_pfnnid_cache *state)
1028 {
1029         int nid;
1030 
1031         nid = __early_pfn_to_nid(pfn, state);
1032         if (nid >= 0 && nid != node)
1033                 return false;
1034         return true;
1035 }
1036 
1037 /* Only safe to use early in boot when initialisation is single-threaded */
1038 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1039 {
1040         return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1041 }
1042 
1043 #else
1044 
1045 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1046 {
1047         return true;
1048 }
1049 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1050                                         struct mminit_pfnnid_cache *state)
1051 {
1052         return true;
1053 }
1054 #endif
1055 
1056 
1057 void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
1058                                                         unsigned int order)
1059 {
1060         if (early_page_uninitialised(pfn))
1061                 return;
1062         return __free_pages_boot_core(page, pfn, order);
1063 }
1064 
1065 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1066 static void __init deferred_free_range(struct page *page,
1067                                         unsigned long pfn, int nr_pages)
1068 {
1069         int i;
1070 
1071         if (!page)
1072                 return;
1073 
1074         /* Free a large naturally-aligned chunk if possible */
1075         if (nr_pages == MAX_ORDER_NR_PAGES &&
1076             (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
1077                 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1078                 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1079                 return;
1080         }
1081 
1082         for (i = 0; i < nr_pages; i++, page++, pfn++)
1083                 __free_pages_boot_core(page, pfn, 0);
1084 }
1085 
1086 /* Completion tracking for deferred_init_memmap() threads */
1087 static atomic_t pgdat_init_n_undone __initdata;
1088 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1089 
1090 static inline void __init pgdat_init_report_one_done(void)
1091 {
1092         if (atomic_dec_and_test(&pgdat_init_n_undone))
1093                 complete(&pgdat_init_all_done_comp);
1094 }
1095 
1096 /* Initialise remaining memory on a node */
1097 static int __init deferred_init_memmap(void *data)
1098 {
1099         pg_data_t *pgdat = data;
1100         int nid = pgdat->node_id;
1101         struct mminit_pfnnid_cache nid_init_state = { };
1102         unsigned long start = jiffies;
1103         unsigned long nr_pages = 0;
1104         unsigned long walk_start, walk_end;
1105         int i, zid;
1106         struct zone *zone;
1107         unsigned long first_init_pfn = pgdat->first_deferred_pfn;
1108         const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1109 
1110         if (first_init_pfn == ULONG_MAX) {
1111                 pgdat_init_report_one_done();
1112                 return 0;
1113         }
1114 
1115         /* Bind memory initialisation thread to a local node if possible */
1116         if (!cpumask_empty(cpumask))
1117                 set_cpus_allowed_ptr(current, cpumask);
1118 
1119         /* Sanity check boundaries */
1120         BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1121         BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1122         pgdat->first_deferred_pfn = ULONG_MAX;
1123 
1124         /* Only the highest zone is deferred so find it */
1125         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1126                 zone = pgdat->node_zones + zid;
1127                 if (first_init_pfn < zone_end_pfn(zone))
1128                         break;
1129         }
1130 
1131         for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1132                 unsigned long pfn, end_pfn;
1133                 struct page *page = NULL;
1134                 struct page *free_base_page = NULL;
1135                 unsigned long free_base_pfn = 0;
1136                 int nr_to_free = 0;
1137 
1138                 end_pfn = min(walk_end, zone_end_pfn(zone));
1139                 pfn = first_init_pfn;
1140                 if (pfn < walk_start)
1141                         pfn = walk_start;
1142                 if (pfn < zone->zone_start_pfn)
1143                         pfn = zone->zone_start_pfn;
1144 
1145                 for (; pfn < end_pfn; pfn++) {
1146                         if (!pfn_valid_within(pfn))
1147                                 goto free_range;
1148 
1149                         /*
1150                          * Ensure pfn_valid is checked every
1151                          * MAX_ORDER_NR_PAGES for memory holes
1152                          */
1153                         if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1154                                 if (!pfn_valid(pfn)) {
1155                                         page = NULL;
1156                                         goto free_range;
1157                                 }
1158                         }
1159 
1160                         if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1161                                 page = NULL;
1162                                 goto free_range;
1163                         }
1164 
1165                         /* Minimise pfn page lookups and scheduler checks */
1166                         if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1167                                 page++;
1168                         } else {
1169                                 nr_pages += nr_to_free;
1170                                 deferred_free_range(free_base_page,
1171                                                 free_base_pfn, nr_to_free);
1172                                 free_base_page = NULL;
1173                                 free_base_pfn = nr_to_free = 0;
1174 
1175                                 page = pfn_to_page(pfn);
1176                                 cond_resched();
1177                         }
1178 
1179                         if (page->flags) {
1180                                 VM_BUG_ON(page_zone(page) != zone);
1181                                 goto free_range;
1182                         }
1183 
1184                         __init_single_page(page, pfn, zid, nid);
1185                         if (!free_base_page) {
1186                                 free_base_page = page;
1187                                 free_base_pfn = pfn;
1188                                 nr_to_free = 0;
1189                         }
1190                         nr_to_free++;
1191 
1192                         /* Where possible, batch up pages for a single free */
1193                         continue;
1194 free_range:
1195                         /* Free the current block of pages to allocator */
1196                         nr_pages += nr_to_free;
1197                         deferred_free_range(free_base_page, free_base_pfn,
1198                                                                 nr_to_free);
1199                         free_base_page = NULL;
1200                         free_base_pfn = nr_to_free = 0;
1201                 }
1202 
1203                 first_init_pfn = max(end_pfn, first_init_pfn);
1204         }
1205 
1206         /* Sanity check that the next zone really is unpopulated */
1207         WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1208 
1209         pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1210                                         jiffies_to_msecs(jiffies - start));
1211 
1212         pgdat_init_report_one_done();
1213         return 0;
1214 }
1215 
1216 void __init page_alloc_init_late(void)
1217 {
1218         int nid;
1219 
1220         /* There will be num_node_state(N_MEMORY) threads */
1221         atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1222         for_each_node_state(nid, N_MEMORY) {
1223                 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1224         }
1225 
1226         /* Block until all are initialised */
1227         wait_for_completion(&pgdat_init_all_done_comp);
1228 
1229         /* Reinit limits that are based on free pages after the kernel is up */
1230         files_maxfiles_init();
1231 }
1232 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1233 
1234 #ifdef CONFIG_CMA
1235 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1236 void __init init_cma_reserved_pageblock(struct page *page)
1237 {
1238         unsigned i = pageblock_nr_pages;
1239         struct page *p = page;
1240 
1241         do {
1242                 __ClearPageReserved(p);
1243                 set_page_count(p, 0);
1244         } while (++p, --i);
1245 
1246         set_pageblock_migratetype(page, MIGRATE_CMA);
1247 
1248         if (pageblock_order >= MAX_ORDER) {
1249                 i = pageblock_nr_pages;
1250                 p = page;
1251                 do {
1252                         set_page_refcounted(p);
1253                         __free_pages(p, MAX_ORDER - 1);
1254                         p += MAX_ORDER_NR_PAGES;
1255                 } while (i -= MAX_ORDER_NR_PAGES);
1256         } else {
1257                 set_page_refcounted(page);
1258                 __free_pages(page, pageblock_order);
1259         }
1260 
1261         adjust_managed_page_count(page, pageblock_nr_pages);
1262 }
1263 #endif
1264 
1265 /*
1266  * The order of subdivision here is critical for the IO subsystem.
1267  * Please do not alter this order without good reasons and regression
1268  * testing. Specifically, as large blocks of memory are subdivided,
1269  * the order in which smaller blocks are delivered depends on the order
1270  * they're subdivided in this function. This is the primary factor
1271  * influencing the order in which pages are delivered to the IO
1272  * subsystem according to empirical testing, and this is also justified
1273  * by considering the behavior of a buddy system containing a single
1274  * large block of memory acted on by a series of small allocations.
1275  * This behavior is a critical factor in sglist merging's success.
1276  *
1277  * -- nyc
1278  */
1279 static inline void expand(struct zone *zone, struct page *page,
1280         int low, int high, struct free_area *area,
1281         int migratetype)
1282 {
1283         unsigned long size = 1 << high;
1284 
1285         while (high > low) {
1286                 area--;
1287                 high--;
1288                 size >>= 1;
1289                 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1290 
1291                 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
1292                         debug_guardpage_enabled() &&
1293                         high < debug_guardpage_minorder()) {
1294                         /*
1295                          * Mark as guard pages (or page), that will allow to
1296                          * merge back to allocator when buddy will be freed.
1297                          * Corresponding page table entries will not be touched,
1298                          * pages will stay not present in virtual address space
1299                          */
1300                         set_page_guard(zone, &page[size], high, migratetype);
1301                         continue;
1302                 }
1303                 list_add(&page[size].lru, &area->free_list[migratetype]);
1304                 area->nr_free++;
1305                 set_page_order(&page[size], high);
1306         }
1307 }
1308 
1309 /*
1310  * This page is about to be returned from the page allocator
1311  */
1312 static inline int check_new_page(struct page *page)
1313 {
1314         const char *bad_reason = NULL;
1315         unsigned long bad_flags = 0;
1316 
1317         if (unlikely(page_mapcount(page)))
1318                 bad_reason = "nonzero mapcount";
1319         if (unlikely(page->mapping != NULL))
1320                 bad_reason = "non-NULL mapping";
1321         if (unlikely(atomic_read(&page->_count) != 0))
1322                 bad_reason = "nonzero _count";
1323         if (unlikely(page->flags & __PG_HWPOISON)) {
1324                 bad_reason = "HWPoisoned (hardware-corrupted)";
1325                 bad_flags = __PG_HWPOISON;
1326         }
1327         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1328                 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1329                 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1330         }
1331 #ifdef CONFIG_MEMCG
1332         if (unlikely(page->mem_cgroup))
1333                 bad_reason = "page still charged to cgroup";
1334 #endif
1335         if (unlikely(bad_reason)) {
1336                 bad_page(page, bad_reason, bad_flags);
1337                 return 1;
1338         }
1339         return 0;
1340 }
1341 
1342 static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1343                                                                 int alloc_flags)
1344 {
1345         int i;
1346 
1347         for (i = 0; i < (1 << order); i++) {
1348                 struct page *p = page + i;
1349                 if (unlikely(check_new_page(p)))
1350                         return 1;
1351         }
1352 
1353         set_page_private(page, 0);
1354         set_page_refcounted(page);
1355 
1356         arch_alloc_page(page, order);
1357         kernel_map_pages(page, 1 << order, 1);
1358         kasan_alloc_pages(page, order);
1359 
1360         if (gfp_flags & __GFP_ZERO)
1361                 for (i = 0; i < (1 << order); i++)
1362                         clear_highpage(page + i);
1363 
1364         if (order && (gfp_flags & __GFP_COMP))
1365                 prep_compound_page(page, order);
1366 
1367         set_page_owner(page, order, gfp_flags);
1368 
1369         /*
1370          * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1371          * allocate the page. The expectation is that the caller is taking
1372          * steps that will free more memory. The caller should avoid the page
1373          * being used for !PFMEMALLOC purposes.
1374          */
1375         if (alloc_flags & ALLOC_NO_WATERMARKS)
1376                 set_page_pfmemalloc(page);
1377         else
1378                 clear_page_pfmemalloc(page);
1379 
1380         return 0;
1381 }
1382 
1383 /*
1384  * Go through the free lists for the given migratetype and remove
1385  * the smallest available page from the freelists
1386  */
1387 static inline
1388 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1389                                                 int migratetype)
1390 {
1391         unsigned int current_order;
1392         struct free_area *area;
1393         struct page *page;
1394 
1395         /* Find a page of the appropriate size in the preferred list */
1396         for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1397                 area = &(zone->free_area[current_order]);
1398                 if (list_empty(&area->free_list[migratetype]))
1399                         continue;
1400 
1401                 page = list_entry(area->free_list[migratetype].next,
1402                                                         struct page, lru);
1403                 list_del(&page->lru);
1404                 rmv_page_order(page);
1405                 area->nr_free--;
1406                 expand(zone, page, order, current_order, area, migratetype);
1407                 set_pcppage_migratetype(page, migratetype);
1408                 return page;
1409         }
1410 
1411         return NULL;
1412 }
1413 
1414 
1415 /*
1416  * This array describes the order lists are fallen back to when
1417  * the free lists for the desirable migrate type are depleted
1418  */
1419 static int fallbacks[MIGRATE_TYPES][4] = {
1420         [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,     MIGRATE_RESERVE },
1421         [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,     MIGRATE_RESERVE },
1422         [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,   MIGRATE_RESERVE },
1423 #ifdef CONFIG_CMA
1424         [MIGRATE_CMA]         = { MIGRATE_RESERVE }, /* Never used */
1425 #endif
1426         [MIGRATE_RESERVE]     = { MIGRATE_RESERVE }, /* Never used */
1427 #ifdef CONFIG_MEMORY_ISOLATION
1428         [MIGRATE_ISOLATE]     = { MIGRATE_RESERVE }, /* Never used */
1429 #endif
1430 };
1431 
1432 #ifdef CONFIG_CMA
1433 static struct page *__rmqueue_cma_fallback(struct zone *zone,
1434                                         unsigned int order)
1435 {
1436         return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1437 }
1438 #else
1439 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1440                                         unsigned int order) { return NULL; }
1441 #endif
1442 
1443 /*
1444  * Move the free pages in a range to the free lists of the requested type.
1445  * Note that start_page and end_pages are not aligned on a pageblock
1446  * boundary. If alignment is required, use move_freepages_block()
1447  */
1448 int move_freepages(struct zone *zone,
1449                           struct page *start_page, struct page *end_page,
1450                           int migratetype)
1451 {
1452         struct page *page;
1453         unsigned long order;
1454         int pages_moved = 0;
1455 
1456 #ifndef CONFIG_HOLES_IN_ZONE
1457         /*
1458          * page_zone is not safe to call in this context when
1459          * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1460          * anyway as we check zone boundaries in move_freepages_block().
1461          * Remove at a later date when no bug reports exist related to
1462          * grouping pages by mobility
1463          */
1464         VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1465 #endif
1466 
1467         for (page = start_page; page <= end_page;) {
1468                 /* Make sure we are not inadvertently changing nodes */
1469                 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1470 
1471                 if (!pfn_valid_within(page_to_pfn(page))) {
1472                         page++;
1473                         continue;
1474                 }
1475 
1476                 if (!PageBuddy(page)) {
1477                         page++;
1478                         continue;
1479                 }
1480 
1481                 order = page_order(page);
1482                 list_move(&page->lru,
1483                           &zone->free_area[order].free_list[migratetype]);
1484                 page += 1 << order;
1485                 pages_moved += 1 << order;
1486         }
1487 
1488         return pages_moved;
1489 }
1490 
1491 int move_freepages_block(struct zone *zone, struct page *page,
1492                                 int migratetype)
1493 {
1494         unsigned long start_pfn, end_pfn;
1495         struct page *start_page, *end_page;
1496 
1497         start_pfn = page_to_pfn(page);
1498         start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1499         start_page = pfn_to_page(start_pfn);
1500         end_page = start_page + pageblock_nr_pages - 1;
1501         end_pfn = start_pfn + pageblock_nr_pages - 1;
1502 
1503         /* Do not cross zone boundaries */
1504         if (!zone_spans_pfn(zone, start_pfn))
1505                 start_page = page;
1506         if (!zone_spans_pfn(zone, end_pfn))
1507                 return 0;
1508 
1509         return move_freepages(zone, start_page, end_page, migratetype);
1510 }
1511 
1512 static void change_pageblock_range(struct page *pageblock_page,
1513                                         int start_order, int migratetype)
1514 {
1515         int nr_pageblocks = 1 << (start_order - pageblock_order);
1516 
1517         while (nr_pageblocks--) {
1518                 set_pageblock_migratetype(pageblock_page, migratetype);
1519                 pageblock_page += pageblock_nr_pages;
1520         }
1521 }
1522 
1523 /*
1524  * When we are falling back to another migratetype during allocation, try to
1525  * steal extra free pages from the same pageblocks to satisfy further
1526  * allocations, instead of polluting multiple pageblocks.
1527  *
1528  * If we are stealing a relatively large buddy page, it is likely there will
1529  * be more free pages in the pageblock, so try to steal them all. For
1530  * reclaimable and unmovable allocations, we steal regardless of page size,
1531  * as fragmentation caused by those allocations polluting movable pageblocks
1532  * is worse than movable allocations stealing from unmovable and reclaimable
1533  * pageblocks.
1534  */
1535 static bool can_steal_fallback(unsigned int order, int start_mt)
1536 {
1537         /*
1538          * Leaving this order check is intended, although there is
1539          * relaxed order check in next check. The reason is that
1540          * we can actually steal whole pageblock if this condition met,
1541          * but, below check doesn't guarantee it and that is just heuristic
1542          * so could be changed anytime.
1543          */
1544         if (order >= pageblock_order)
1545                 return true;
1546 
1547         if (order >= pageblock_order / 2 ||
1548                 start_mt == MIGRATE_RECLAIMABLE ||
1549                 start_mt == MIGRATE_UNMOVABLE ||
1550                 page_group_by_mobility_disabled)
1551                 return true;
1552 
1553         return false;
1554 }
1555 
1556 /*
1557  * This function implements actual steal behaviour. If order is large enough,
1558  * we can steal whole pageblock. If not, we first move freepages in this
1559  * pageblock and check whether half of pages are moved or not. If half of
1560  * pages are moved, we can change migratetype of pageblock and permanently
1561  * use it's pages as requested migratetype in the future.
1562  */
1563 static void steal_suitable_fallback(struct zone *zone, struct page *page,
1564                                                           int start_type)
1565 {
1566         int current_order = page_order(page);
1567         int pages;
1568 
1569         /* Take ownership for orders >= pageblock_order */
1570         if (current_order >= pageblock_order) {
1571                 change_pageblock_range(page, current_order, start_type);
1572                 return;
1573         }
1574 
1575         pages = move_freepages_block(zone, page, start_type);
1576 
1577         /* Claim the whole block if over half of it is free */
1578         if (pages >= (1 << (pageblock_order-1)) ||
1579                         page_group_by_mobility_disabled)
1580                 set_pageblock_migratetype(page, start_type);
1581 }
1582 
1583 /*
1584  * Check whether there is a suitable fallback freepage with requested order.
1585  * If only_stealable is true, this function returns fallback_mt only if
1586  * we can steal other freepages all together. This would help to reduce
1587  * fragmentation due to mixed migratetype pages in one pageblock.
1588  */
1589 int find_suitable_fallback(struct free_area *area, unsigned int order,
1590                         int migratetype, bool only_stealable, bool *can_steal)
1591 {
1592         int i;
1593         int fallback_mt;
1594 
1595         if (area->nr_free == 0)
1596                 return -1;
1597 
1598         *can_steal = false;
1599         for (i = 0;; i++) {
1600                 fallback_mt = fallbacks[migratetype][i];
1601                 if (fallback_mt == MIGRATE_RESERVE)
1602                         break;
1603 
1604                 if (list_empty(&area->free_list[fallback_mt]))
1605                         continue;
1606 
1607                 if (can_steal_fallback(order, migratetype))
1608                         *can_steal = true;
1609 
1610                 if (!only_stealable)
1611                         return fallback_mt;
1612 
1613                 if (*can_steal)
1614                         return fallback_mt;
1615         }
1616 
1617         return -1;
1618 }
1619 
1620 /* Remove an element from the buddy allocator from the fallback list */
1621 static inline struct page *
1622 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
1623 {
1624         struct free_area *area;
1625         unsigned int current_order;
1626         struct page *page;
1627         int fallback_mt;
1628         bool can_steal;
1629 
1630         /* Find the largest possible block of pages in the other list */
1631         for (current_order = MAX_ORDER-1;
1632                                 current_order >= order && current_order <= MAX_ORDER-1;
1633                                 --current_order) {
1634                 area = &(zone->free_area[current_order]);
1635                 fallback_mt = find_suitable_fallback(area, current_order,
1636                                 start_migratetype, false, &can_steal);
1637                 if (fallback_mt == -1)
1638                         continue;
1639 
1640                 page = list_entry(area->free_list[fallback_mt].next,
1641                                                 struct page, lru);
1642                 if (can_steal)
1643                         steal_suitable_fallback(zone, page, start_migratetype);
1644 
1645                 /* Remove the page from the freelists */
1646                 area->nr_free--;
1647                 list_del(&page->lru);
1648                 rmv_page_order(page);
1649 
1650                 expand(zone, page, order, current_order, area,
1651                                         start_migratetype);
1652                 /*
1653                  * The pcppage_migratetype may differ from pageblock's
1654                  * migratetype depending on the decisions in
1655                  * find_suitable_fallback(). This is OK as long as it does not
1656                  * differ for MIGRATE_CMA pageblocks. Those can be used as
1657                  * fallback only via special __rmqueue_cma_fallback() function
1658                  */
1659                 set_pcppage_migratetype(page, start_migratetype);
1660 
1661                 trace_mm_page_alloc_extfrag(page, order, current_order,
1662                         start_migratetype, fallback_mt);
1663 
1664                 return page;
1665         }
1666 
1667         return NULL;
1668 }
1669 
1670 /*
1671  * Do the hard work of removing an element from the buddy allocator.
1672  * Call me with the zone->lock already held.
1673  */
1674 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1675                                                 int migratetype)
1676 {
1677         struct page *page;
1678 
1679 retry_reserve:
1680         page = __rmqueue_smallest(zone, order, migratetype);
1681 
1682         if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1683                 if (migratetype == MIGRATE_MOVABLE)
1684                         page = __rmqueue_cma_fallback(zone, order);
1685 
1686                 if (!page)
1687                         page = __rmqueue_fallback(zone, order, migratetype);
1688 
1689                 /*
1690                  * Use MIGRATE_RESERVE rather than fail an allocation. goto
1691                  * is used because __rmqueue_smallest is an inline function
1692                  * and we want just one call site
1693                  */
1694                 if (!page) {
1695                         migratetype = MIGRATE_RESERVE;
1696                         goto retry_reserve;
1697                 }
1698         }
1699 
1700         trace_mm_page_alloc_zone_locked(page, order, migratetype);
1701         return page;
1702 }
1703 
1704 /*
1705  * Obtain a specified number of elements from the buddy allocator, all under
1706  * a single hold of the lock, for efficiency.  Add them to the supplied list.
1707  * Returns the number of new pages which were placed at *list.
1708  */
1709 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1710                         unsigned long count, struct list_head *list,
1711                         int migratetype, bool cold)
1712 {
1713         int i;
1714 
1715         spin_lock(&zone->lock);
1716         for (i = 0; i < count; ++i) {
1717                 struct page *page = __rmqueue(zone, order, migratetype);
1718                 if (unlikely(page == NULL))
1719                         break;
1720 
1721                 /*
1722                  * Split buddy pages returned by expand() are received here
1723                  * in physical page order. The page is added to the callers and
1724                  * list and the list head then moves forward. From the callers
1725                  * perspective, the linked list is ordered by page number in
1726                  * some conditions. This is useful for IO devices that can
1727                  * merge IO requests if the physical pages are ordered
1728                  * properly.
1729                  */
1730                 if (likely(!cold))
1731                         list_add(&page->lru, list);
1732                 else
1733                         list_add_tail(&page->lru, list);
1734                 list = &page->lru;
1735                 if (is_migrate_cma(get_pcppage_migratetype(page)))
1736                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1737                                               -(1 << order));
1738         }
1739         __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1740         spin_unlock(&zone->lock);
1741         return i;
1742 }
1743 
1744 #ifdef CONFIG_NUMA
1745 /*
1746  * Called from the vmstat counter updater to drain pagesets of this
1747  * currently executing processor on remote nodes after they have
1748  * expired.
1749  *
1750  * Note that this function must be called with the thread pinned to
1751  * a single processor.
1752  */
1753 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1754 {
1755         unsigned long flags;
1756         int to_drain, batch;
1757 
1758         local_irq_save(flags);
1759         batch = READ_ONCE(pcp->batch);
1760         to_drain = min(pcp->count, batch);
1761         if (to_drain > 0) {
1762                 free_pcppages_bulk(zone, to_drain, pcp);
1763                 pcp->count -= to_drain;
1764         }
1765         local_irq_restore(flags);
1766 }
1767 #endif
1768 
1769 /*
1770  * Drain pcplists of the indicated processor and zone.
1771  *
1772  * The processor must either be the current processor and the
1773  * thread pinned to the current processor or a processor that
1774  * is not online.
1775  */
1776 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1777 {
1778         unsigned long flags;
1779         struct per_cpu_pageset *pset;
1780         struct per_cpu_pages *pcp;
1781 
1782         local_irq_save(flags);
1783         pset = per_cpu_ptr(zone->pageset, cpu);
1784 
1785         pcp = &pset->pcp;
1786         if (pcp->count) {
1787                 free_pcppages_bulk(zone, pcp->count, pcp);
1788                 pcp->count = 0;
1789         }
1790         local_irq_restore(flags);
1791 }
1792 
1793 /*
1794  * Drain pcplists of all zones on the indicated processor.
1795  *
1796  * The processor must either be the current processor and the
1797  * thread pinned to the current processor or a processor that
1798  * is not online.
1799  */
1800 static void drain_pages(unsigned int cpu)
1801 {
1802         struct zone *zone;
1803 
1804         for_each_populated_zone(zone) {
1805                 drain_pages_zone(cpu, zone);
1806         }
1807 }
1808 
1809 /*
1810  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1811  *
1812  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1813  * the single zone's pages.
1814  */
1815 void drain_local_pages(struct zone *zone)
1816 {
1817         int cpu = smp_processor_id();
1818 
1819         if (zone)
1820                 drain_pages_zone(cpu, zone);
1821         else
1822                 drain_pages(cpu);
1823 }
1824 
1825 /*
1826  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1827  *
1828  * When zone parameter is non-NULL, spill just the single zone's pages.
1829  *
1830  * Note that this code is protected against sending an IPI to an offline
1831  * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1832  * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1833  * nothing keeps CPUs from showing up after we populated the cpumask and
1834  * before the call to on_each_cpu_mask().
1835  */
1836 void drain_all_pages(struct zone *zone)
1837 {
1838         int cpu;
1839 
1840         /*
1841          * Allocate in the BSS so we wont require allocation in
1842          * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1843          */
1844         static cpumask_t cpus_with_pcps;
1845 
1846         /*
1847          * We don't care about racing with CPU hotplug event
1848          * as offline notification will cause the notified
1849          * cpu to drain that CPU pcps and on_each_cpu_mask
1850          * disables preemption as part of its processing
1851          */
1852         for_each_online_cpu(cpu) {
1853                 struct per_cpu_pageset *pcp;
1854                 struct zone *z;
1855                 bool has_pcps = false;
1856 
1857                 if (zone) {
1858                         pcp = per_cpu_ptr(zone->pageset, cpu);
1859                         if (pcp->pcp.count)
1860                                 has_pcps = true;
1861                 } else {
1862                         for_each_populated_zone(z) {
1863                                 pcp = per_cpu_ptr(z->pageset, cpu);
1864                                 if (pcp->pcp.count) {
1865                                         has_pcps = true;
1866                                         break;
1867                                 }
1868                         }
1869                 }
1870 
1871                 if (has_pcps)
1872                         cpumask_set_cpu(cpu, &cpus_with_pcps);
1873                 else
1874                         cpumask_clear_cpu(cpu, &cpus_with_pcps);
1875         }
1876         on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1877                                                                 zone, 1);
1878 }
1879 
1880 #ifdef CONFIG_HIBERNATION
1881 
1882 void mark_free_pages(struct zone *zone)
1883 {
1884         unsigned long pfn, max_zone_pfn;
1885         unsigned long flags;
1886         unsigned int order, t;
1887         struct list_head *curr;
1888 
1889         if (zone_is_empty(zone))
1890                 return;
1891 
1892         spin_lock_irqsave(&zone->lock, flags);
1893 
1894         max_zone_pfn = zone_end_pfn(zone);
1895         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1896                 if (pfn_valid(pfn)) {
1897                         struct page *page = pfn_to_page(pfn);
1898 
1899                         if (!swsusp_page_is_forbidden(page))
1900                                 swsusp_unset_page_free(page);
1901                 }
1902 
1903         for_each_migratetype_order(order, t) {
1904                 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1905                         unsigned long i;
1906 
1907                         pfn = page_to_pfn(list_entry(curr, struct page, lru));
1908                         for (i = 0; i < (1UL << order); i++)
1909                                 swsusp_set_page_free(pfn_to_page(pfn + i));
1910                 }
1911         }
1912         spin_unlock_irqrestore(&zone->lock, flags);
1913 }
1914 #endif /* CONFIG_PM */
1915 
1916 /*
1917  * Free a 0-order page
1918  * cold == true ? free a cold page : free a hot page
1919  */
1920 void free_hot_cold_page(struct page *page, bool cold)
1921 {
1922         struct zone *zone = page_zone(page);
1923         struct per_cpu_pages *pcp;
1924         unsigned long flags;
1925         unsigned long pfn = page_to_pfn(page);
1926         int migratetype;
1927 
1928         if (!free_pages_prepare(page, 0))
1929                 return;
1930 
1931         migratetype = get_pfnblock_migratetype(page, pfn);
1932         set_pcppage_migratetype(page, migratetype);
1933         local_irq_save(flags);
1934         __count_vm_event(PGFREE);
1935 
1936         /*
1937          * We only track unmovable, reclaimable and movable on pcp lists.
1938          * Free ISOLATE pages back to the allocator because they are being
1939          * offlined but treat RESERVE as movable pages so we can get those
1940          * areas back if necessary. Otherwise, we may have to free
1941          * excessively into the page allocator
1942          */
1943         if (migratetype >= MIGRATE_PCPTYPES) {
1944                 if (unlikely(is_migrate_isolate(migratetype))) {
1945                         free_one_page(zone, page, pfn, 0, migratetype);
1946                         goto out;
1947                 }
1948                 migratetype = MIGRATE_MOVABLE;
1949         }
1950 
1951         pcp = &this_cpu_ptr(zone->pageset)->pcp;
1952         if (!cold)
1953                 list_add(&page->lru, &pcp->lists[migratetype]);
1954         else
1955                 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1956         pcp->count++;
1957         if (pcp->count >= pcp->high) {
1958                 unsigned long batch = READ_ONCE(pcp->batch);
1959                 free_pcppages_bulk(zone, batch, pcp);
1960                 pcp->count -= batch;
1961         }
1962 
1963 out:
1964         local_irq_restore(flags);
1965 }
1966 
1967 /*
1968  * Free a list of 0-order pages
1969  */
1970 void free_hot_cold_page_list(struct list_head *list, bool cold)
1971 {
1972         struct page *page, *next;
1973 
1974         list_for_each_entry_safe(page, next, list, lru) {
1975                 trace_mm_page_free_batched(page, cold);
1976                 free_hot_cold_page(page, cold);
1977         }
1978 }
1979 
1980 /*
1981  * split_page takes a non-compound higher-order page, and splits it into
1982  * n (1<<order) sub-pages: page[0..n]
1983  * Each sub-page must be freed individually.
1984  *
1985  * Note: this is probably too low level an operation for use in drivers.
1986  * Please consult with lkml before using this in your driver.
1987  */
1988 void split_page(struct page *page, unsigned int order)
1989 {
1990         int i;
1991         gfp_t gfp_mask;
1992 
1993         VM_BUG_ON_PAGE(PageCompound(page), page);
1994         VM_BUG_ON_PAGE(!page_count(page), page);
1995 
1996 #ifdef CONFIG_KMEMCHECK
1997         /*
1998          * Split shadow pages too, because free(page[0]) would
1999          * otherwise free the whole shadow.
2000          */
2001         if (kmemcheck_page_is_tracked(page))
2002                 split_page(virt_to_page(page[0].shadow), order);
2003 #endif
2004 
2005         gfp_mask = get_page_owner_gfp(page);
2006         set_page_owner(page, 0, gfp_mask);
2007         for (i = 1; i < (1 << order); i++) {
2008                 set_page_refcounted(page + i);
2009                 set_page_owner(page + i, 0, gfp_mask);
2010         }
2011 }
2012 EXPORT_SYMBOL_GPL(split_page);
2013 
2014 int __isolate_free_page(struct page *page, unsigned int order)
2015 {
2016         unsigned long watermark;
2017         struct zone *zone;
2018         int mt;
2019 
2020         BUG_ON(!PageBuddy(page));
2021 
2022         zone = page_zone(page);
2023         mt = get_pageblock_migratetype(page);
2024 
2025         if (!is_migrate_isolate(mt)) {
2026                 /* Obey watermarks as if the page was being allocated */
2027                 watermark = low_wmark_pages(zone) + (1 << order);
2028                 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2029                         return 0;
2030 
2031                 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2032         }
2033 
2034         /* Remove page from free list */
2035         list_del(&page->lru);
2036         zone->free_area[order].nr_free--;
2037         rmv_page_order(page);
2038 
2039         set_page_owner(page, order, __GFP_MOVABLE);
2040 
2041         /* Set the pageblock if the isolated page is at least a pageblock */
2042         if (order >= pageblock_order - 1) {
2043                 struct page *endpage = page + (1 << order) - 1;
2044                 for (; page < endpage; page += pageblock_nr_pages) {
2045                         int mt = get_pageblock_migratetype(page);
2046                         if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
2047                                 set_pageblock_migratetype(page,
2048                                                           MIGRATE_MOVABLE);
2049                 }
2050         }
2051 
2052 
2053         return 1UL << order;
2054 }
2055 
2056 /*
2057  * Similar to split_page except the page is already free. As this is only
2058  * being used for migration, the migratetype of the block also changes.
2059  * As this is called with interrupts disabled, the caller is responsible
2060  * for calling arch_alloc_page() and kernel_map_page() after interrupts
2061  * are enabled.
2062  *
2063  * Note: this is probably too low level an operation for use in drivers.
2064  * Please consult with lkml before using this in your driver.
2065  */
2066 int split_free_page(struct page *page)
2067 {
2068         unsigned int order;
2069         int nr_pages;
2070 
2071         order = page_order(page);
2072 
2073         nr_pages = __isolate_free_page(page, order);
2074         if (!nr_pages)
2075                 return 0;
2076 
2077         /* Split into individual pages */
2078         set_page_refcounted(page);
2079         split_page(page, order);
2080         return nr_pages;
2081 }
2082 
2083 /*
2084  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2085  */
2086 static inline
2087 struct page *buffered_rmqueue(struct zone *preferred_zone,
2088                         struct zone *zone, unsigned int order,
2089                         gfp_t gfp_flags, int migratetype)
2090 {
2091         unsigned long flags;
2092         struct page *page;
2093         bool cold = ((gfp_flags & __GFP_COLD) != 0);
2094 
2095         if (likely(order == 0)) {
2096                 struct per_cpu_pages *pcp;
2097                 struct list_head *list;
2098 
2099                 local_irq_save(flags);
2100                 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2101                 list = &pcp->lists[migratetype];
2102                 if (list_empty(list)) {
2103                         pcp->count += rmqueue_bulk(zone, 0,
2104                                         pcp->batch, list,
2105                                         migratetype, cold);
2106                         if (unlikely(list_empty(list)))
2107                                 goto failed;
2108                 }
2109 
2110                 if (cold)
2111                         page = list_entry(list->prev, struct page, lru);
2112                 else
2113                         page = list_entry(list->next, struct page, lru);
2114 
2115                 list_del(&page->lru);
2116                 pcp->count--;
2117         } else {
2118                 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
2119                         /*
2120                          * __GFP_NOFAIL is not to be used in new code.
2121                          *
2122                          * All __GFP_NOFAIL callers should be fixed so that they
2123                          * properly detect and handle allocation failures.
2124                          *
2125                          * We most definitely don't want callers attempting to
2126                          * allocate greater than order-1 page units with
2127                          * __GFP_NOFAIL.
2128                          */
2129                         WARN_ON_ONCE(order > 1);
2130                 }
2131                 spin_lock_irqsave(&zone->lock, flags);
2132                 page = __rmqueue(zone, order, migratetype);
2133                 spin_unlock(&zone->lock);
2134                 if (!page)
2135                         goto failed;
2136                 __mod_zone_freepage_state(zone, -(1 << order),
2137                                           get_pcppage_migratetype(page));
2138         }
2139 
2140         __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
2141         if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
2142             !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2143                 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2144 
2145         __count_zone_vm_events(PGALLOC, zone, 1 << order);
2146         zone_statistics(preferred_zone, zone, gfp_flags);
2147         local_irq_restore(flags);
2148 
2149         VM_BUG_ON_PAGE(bad_range(zone, page), page);
2150         return page;
2151 
2152 failed:
2153         local_irq_restore(flags);
2154         return NULL;
2155 }
2156 
2157 #ifdef CONFIG_FAIL_PAGE_ALLOC
2158 
2159 static struct {
2160         struct fault_attr attr;
2161 
2162         u32 ignore_gfp_highmem;
2163         u32 ignore_gfp_wait;
2164         u32 min_order;
2165 } fail_page_alloc = {
2166         .attr = FAULT_ATTR_INITIALIZER,
2167         .ignore_gfp_wait = 1,
2168         .ignore_gfp_highmem = 1,
2169         .min_order = 1,
2170 };
2171 
2172 static int __init setup_fail_page_alloc(char *str)
2173 {
2174         return setup_fault_attr(&fail_page_alloc.attr, str);
2175 }
2176 __setup("fail_page_alloc=", setup_fail_page_alloc);
2177 
2178 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2179 {
2180         if (order < fail_page_alloc.min_order)
2181                 return false;
2182         if (gfp_mask & __GFP_NOFAIL)
2183                 return false;
2184         if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
2185                 return false;
2186         if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
2187                 return false;
2188 
2189         return should_fail(&fail_page_alloc.attr, 1 << order);
2190 }
2191 
2192 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2193 
2194 static int __init fail_page_alloc_debugfs(void)
2195 {
2196         umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
2197         struct dentry *dir;
2198 
2199         dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2200                                         &fail_page_alloc.attr);
2201         if (IS_ERR(dir))
2202                 return PTR_ERR(dir);
2203 
2204         if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2205                                 &fail_page_alloc.ignore_gfp_wait))
2206                 goto fail;
2207         if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2208                                 &fail_page_alloc.ignore_gfp_highmem))
2209                 goto fail;
2210         if (!debugfs_create_u32("min-order", mode, dir,
2211                                 &fail_page_alloc.min_order))
2212                 goto fail;
2213 
2214         return 0;
2215 fail:
2216         debugfs_remove_recursive(dir);
2217 
2218         return -ENOMEM;
2219 }
2220 
2221 late_initcall(fail_page_alloc_debugfs);
2222 
2223 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2224 
2225 #else /* CONFIG_FAIL_PAGE_ALLOC */
2226 
2227 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2228 {
2229         return false;
2230 }
2231 
2232 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2233 
2234 /*
2235  * Return true if free pages are above 'mark'. This takes into account the order
2236  * of the allocation.
2237  */
2238 static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2239                         unsigned long mark, int classzone_idx, int alloc_flags,
2240                         long free_pages)
2241 {
2242         /* free_pages may go negative - that's OK */
2243         long min = mark;
2244         int o;
2245         long free_cma = 0;
2246 
2247         free_pages -= (1 << order) - 1;
2248         if (alloc_flags & ALLOC_HIGH)
2249                 min -= min / 2;
2250         if (alloc_flags & ALLOC_HARDER)
2251                 min -= min / 4;
2252 #ifdef CONFIG_CMA
2253         /* If allocation can't use CMA areas don't use free CMA pages */
2254         if (!(alloc_flags & ALLOC_CMA))
2255                 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
2256 #endif
2257 
2258         if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
2259                 return false;
2260         for (o = 0; o < order; o++) {
2261                 /* At the next order, this order's pages become unavailable */
2262                 free_pages -= z->free_area[o].nr_free << o;
2263 
2264                 /* Require fewer higher order pages to be free */
2265                 min >>= 1;
2266 
2267                 if (free_pages <= min)
2268                         return false;
2269         }
2270         return true;
2271 }
2272 
2273 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2274                       int classzone_idx, int alloc_flags)
2275 {
2276         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2277                                         zone_page_state(z, NR_FREE_PAGES));
2278 }
2279 
2280 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2281                         unsigned long mark, int classzone_idx, int alloc_flags)
2282 {
2283         long free_pages = zone_page_state(z, NR_FREE_PAGES);
2284 
2285         if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2286                 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2287 
2288         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2289                                                                 free_pages);
2290 }
2291 
2292 #ifdef CONFIG_NUMA
2293 /*
2294  * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
2295  * skip over zones that are not allowed by the cpuset, or that have
2296  * been recently (in last second) found to be nearly full.  See further
2297  * comments in mmzone.h.  Reduces cache footprint of zonelist scans
2298  * that have to skip over a lot of full or unallowed zones.
2299  *
2300  * If the zonelist cache is present in the passed zonelist, then
2301  * returns a pointer to the allowed node mask (either the current
2302  * tasks mems_allowed, or node_states[N_MEMORY].)
2303  *
2304  * If the zonelist cache is not available for this zonelist, does
2305  * nothing and returns NULL.
2306  *
2307  * If the fullzones BITMAP in the zonelist cache is stale (more than
2308  * a second since last zap'd) then we zap it out (clear its bits.)
2309  *
2310  * We hold off even calling zlc_setup, until after we've checked the
2311  * first zone in the zonelist, on the theory that most allocations will
2312  * be satisfied from that first zone, so best to examine that zone as
2313  * quickly as we can.
2314  */
2315 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2316 {
2317         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
2318         nodemask_t *allowednodes;       /* zonelist_cache approximation */
2319 
2320         zlc = zonelist->zlcache_ptr;
2321         if (!zlc)
2322                 return NULL;
2323 
2324         if (time_after(jiffies, zlc->last_full_zap + HZ)) {
2325                 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2326                 zlc->last_full_zap = jiffies;
2327         }
2328 
2329         allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
2330                                         &cpuset_current_mems_allowed :
2331                                         &node_states[N_MEMORY];
2332         return allowednodes;
2333 }
2334 
2335 /*
2336  * Given 'z' scanning a zonelist, run a couple of quick checks to see
2337  * if it is worth looking at further for free memory:
2338  *  1) Check that the zone isn't thought to be full (doesn't have its
2339  *     bit set in the zonelist_cache fullzones BITMAP).
2340  *  2) Check that the zones node (obtained from the zonelist_cache
2341  *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
2342  * Return true (non-zero) if zone is worth looking at further, or
2343  * else return false (zero) if it is not.
2344  *
2345  * This check -ignores- the distinction between various watermarks,
2346  * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
2347  * found to be full for any variation of these watermarks, it will
2348  * be considered full for up to one second by all requests, unless
2349  * we are so low on memory on all allowed nodes that we are forced
2350  * into the second scan of the zonelist.
2351  *
2352  * In the second scan we ignore this zonelist cache and exactly
2353  * apply the watermarks to all zones, even it is slower to do so.
2354  * We are low on memory in the second scan, and should leave no stone
2355  * unturned looking for a free page.
2356  */
2357 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
2358                                                 nodemask_t *allowednodes)
2359 {
2360         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
2361         int i;                          /* index of *z in zonelist zones */
2362         int n;                          /* node that zone *z is on */
2363 
2364         zlc = zonelist->zlcache_ptr;
2365         if (!zlc)
2366                 return 1;
2367 
2368         i = z - zonelist->_zonerefs;
2369         n = zlc->z_to_n[i];
2370 
2371         /* This zone is worth trying if it is allowed but not full */
2372         return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
2373 }
2374 
2375 /*
2376  * Given 'z' scanning a zonelist, set the corresponding bit in
2377  * zlc->fullzones, so that subsequent attempts to allocate a page
2378  * from that zone don't waste time re-examining it.
2379  */
2380 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
2381 {
2382         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
2383         int i;                          /* index of *z in zonelist zones */
2384 
2385         zlc = zonelist->zlcache_ptr;
2386         if (!zlc)
2387                 return;
2388 
2389         i = z - zonelist->_zonerefs;
2390 
2391         set_bit(i, zlc->fullzones);
2392 }
2393 
2394 /*
2395  * clear all zones full, called after direct reclaim makes progress so that
2396  * a zone that was recently full is not skipped over for up to a second
2397  */
2398 static void zlc_clear_zones_full(struct zonelist *zonelist)
2399 {
2400         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
2401 
2402         zlc = zonelist->zlcache_ptr;
2403         if (!zlc)
2404                 return;
2405 
2406         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2407 }
2408 
2409 static bool zone_local(struct zone *local_zone, struct zone *zone)
2410 {
2411         return local_zone->node == zone->node;
2412 }
2413 
2414 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2415 {
2416         return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2417                                 RECLAIM_DISTANCE;
2418 }
2419 
2420 #else   /* CONFIG_NUMA */
2421 
2422 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
2423 {
2424         return NULL;
2425 }
2426 
2427 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
2428                                 nodemask_t *allowednodes)
2429 {
2430         return 1;
2431 }
2432 
2433 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
2434 {
2435 }
2436 
2437 static void zlc_clear_zones_full(struct zonelist *zonelist)
2438 {
2439 }
2440 
2441 static bool zone_local(struct zone *local_zone, struct zone *zone)
2442 {
2443         return true;
2444 }
2445 
2446 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2447 {
2448         return true;
2449 }
2450 
2451 #endif  /* CONFIG_NUMA */
2452 
2453 static void reset_alloc_batches(struct zone *preferred_zone)
2454 {
2455         struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2456 
2457         do {
2458                 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2459                         high_wmark_pages(zone) - low_wmark_pages(zone) -
2460                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
2461                 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2462         } while (zone++ != preferred_zone);
2463 }
2464 
2465 /*
2466  * get_page_from_freelist goes through the zonelist trying to allocate
2467  * a page.
2468  */
2469 static struct page *
2470 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2471                                                 const struct alloc_context *ac)
2472 {
2473         struct zonelist *zonelist = ac->zonelist;
2474         struct zoneref *z;
2475         struct page *page = NULL;
2476         struct zone *zone;
2477         nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
2478         int zlc_active = 0;             /* set if using zonelist_cache */
2479         int did_zlc_setup = 0;          /* just call zlc_setup() one time */
2480         bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
2481                                 (gfp_mask & __GFP_WRITE);
2482         int nr_fair_skipped = 0;
2483         bool zonelist_rescan;
2484 
2485 zonelist_scan:
2486         zonelist_rescan = false;
2487 
2488         /*
2489          * Scan zonelist, looking for a zone with enough free.
2490          * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2491          */
2492         for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2493                                                                 ac->nodemask) {
2494                 unsigned long mark;
2495 
2496                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2497                         !zlc_zone_worth_trying(zonelist, z, allowednodes))
2498                                 continue;
2499                 if (cpusets_enabled() &&
2500                         (alloc_flags & ALLOC_CPUSET) &&
2501                         !cpuset_zone_allowed(zone, gfp_mask))
2502                                 continue;
2503                 /*
2504                  * Distribute pages in proportion to the individual
2505                  * zone size to ensure fair page aging.  The zone a
2506                  * page was allocated in should have no effect on the
2507                  * time the page has in memory before being reclaimed.
2508                  */
2509                 if (alloc_flags & ALLOC_FAIR) {
2510                         if (!zone_local(ac->preferred_zone, zone))
2511                                 break;
2512                         if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
2513                                 nr_fair_skipped++;
2514                                 continue;
2515                         }
2516                 }
2517                 /*
2518                  * When allocating a page cache page for writing, we
2519                  * want to get it from a zone that is within its dirty
2520                  * limit, such that no single zone holds more than its
2521                  * proportional share of globally allowed dirty pages.
2522                  * The dirty limits take into account the zone's
2523                  * lowmem reserves and high watermark so that kswapd
2524                  * should be able to balance it without having to
2525                  * write pages from its LRU list.
2526                  *
2527                  * This may look like it could increase pressure on
2528                  * lower zones by failing allocations in higher zones
2529                  * before they are full.  But the pages that do spill
2530                  * over are limited as the lower zones are protected
2531                  * by this very same mechanism.  It should not become
2532                  * a practical burden to them.
2533                  *
2534                  * XXX: For now, allow allocations to potentially
2535                  * exceed the per-zone dirty limit in the slowpath
2536                  * (ALLOC_WMARK_LOW unset) before going into reclaim,
2537                  * which is important when on a NUMA setup the allowed
2538                  * zones are together not big enough to reach the
2539                  * global limit.  The proper fix for these situations
2540                  * will require awareness of zones in the
2541                  * dirty-throttling and the flusher threads.
2542                  */
2543                 if (consider_zone_dirty && !zone_dirty_ok(zone))
2544                         continue;
2545 
2546                 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2547                 if (!zone_watermark_ok(zone, order, mark,
2548                                        ac->classzone_idx, alloc_flags)) {
2549                         int ret;
2550 
2551                         /* Checked here to keep the fast path fast */
2552                         BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2553                         if (alloc_flags & ALLOC_NO_WATERMARKS)
2554                                 goto try_this_zone;
2555 
2556                         if (IS_ENABLED(CONFIG_NUMA) &&
2557                                         !did_zlc_setup && nr_online_nodes > 1) {
2558                                 /*
2559                                  * we do zlc_setup if there are multiple nodes
2560                                  * and before considering the first zone allowed
2561                                  * by the cpuset.
2562                                  */
2563                                 allowednodes = zlc_setup(zonelist, alloc_flags);
2564                                 zlc_active = 1;
2565                                 did_zlc_setup = 1;
2566                         }
2567 
2568                         if (zone_reclaim_mode == 0 ||
2569                             !zone_allows_reclaim(ac->preferred_zone, zone))
2570                                 goto this_zone_full;
2571 
2572                         /*
2573                          * As we may have just activated ZLC, check if the first
2574                          * eligible zone has failed zone_reclaim recently.
2575                          */
2576                         if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2577                                 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2578                                 continue;
2579 
2580                         ret = zone_reclaim(zone, gfp_mask, order);
2581                         switch (ret) {
2582                         case ZONE_RECLAIM_NOSCAN:
2583                                 /* did not scan */
2584                                 continue;
2585                         case ZONE_RECLAIM_FULL:
2586                                 /* scanned but unreclaimable */
2587                                 continue;
2588                         default:
2589                                 /* did we reclaim enough */
2590                                 if (zone_watermark_ok(zone, order, mark,
2591                                                 ac->classzone_idx, alloc_flags))
2592                                         goto try_this_zone;
2593 
2594                                 /*
2595                                  * Failed to reclaim enough to meet watermark.
2596                                  * Only mark the zone full if checking the min
2597                                  * watermark or if we failed to reclaim just
2598                                  * 1<<order pages or else the page allocator
2599                                  * fastpath will prematurely mark zones full
2600                                  * when the watermark is between the low and
2601                                  * min watermarks.
2602                                  */
2603                                 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2604                                     ret == ZONE_RECLAIM_SOME)
2605                                         goto this_zone_full;
2606 
2607                                 continue;
2608                         }
2609                 }
2610 
2611 try_this_zone:
2612                 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2613                                                 gfp_mask, ac->migratetype);
2614                 if (page) {
2615                         if (prep_new_page(page, order, gfp_mask, alloc_flags))
2616                                 goto try_this_zone;
2617                         return page;
2618                 }
2619 this_zone_full:
2620                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
2621                         zlc_mark_zone_full(zonelist, z);
2622         }
2623 
2624         /*
2625          * The first pass makes sure allocations are spread fairly within the
2626          * local node.  However, the local node might have free pages left
2627          * after the fairness batches are exhausted, and remote zones haven't
2628          * even been considered yet.  Try once more without fairness, and
2629          * include remote zones now, before entering the slowpath and waking
2630          * kswapd: prefer spilling to a remote zone over swapping locally.
2631          */
2632         if (alloc_flags & ALLOC_FAIR) {
2633                 alloc_flags &= ~ALLOC_FAIR;
2634                 if (nr_fair_skipped) {
2635                         zonelist_rescan = true;
2636                         reset_alloc_batches(ac->preferred_zone);
2637                 }
2638                 if (nr_online_nodes > 1)
2639                         zonelist_rescan = true;
2640         }
2641 
2642         if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2643                 /* Disable zlc cache for second zonelist scan */
2644                 zlc_active = 0;
2645                 zonelist_rescan = true;
2646         }
2647 
2648         if (zonelist_rescan)
2649                 goto zonelist_scan;
2650 
2651         return NULL;
2652 }
2653 
2654 /*
2655  * Large machines with many possible nodes should not always dump per-node
2656  * meminfo in irq context.
2657  */
2658 static inline bool should_suppress_show_mem(void)
2659 {
2660         bool ret = false;
2661 
2662 #if NODES_SHIFT > 8
2663         ret = in_interrupt();
2664 #endif
2665         return ret;
2666 }
2667 
2668 static DEFINE_RATELIMIT_STATE(nopage_rs,
2669                 DEFAULT_RATELIMIT_INTERVAL,
2670                 DEFAULT_RATELIMIT_BURST);
2671 
2672 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2673 {
2674         unsigned int filter = SHOW_MEM_FILTER_NODES;
2675 
2676         if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2677             debug_guardpage_minorder() > 0)
2678                 return;
2679 
2680         /*
2681          * This documents exceptions given to allocations in certain
2682          * contexts that are allowed to allocate outside current's set
2683          * of allowed nodes.
2684          */
2685         if (!(gfp_mask & __GFP_NOMEMALLOC))
2686                 if (test_thread_flag(TIF_MEMDIE) ||
2687                     (current->flags & (PF_MEMALLOC | PF_EXITING)))
2688                         filter &= ~SHOW_MEM_FILTER_NODES;
2689         if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2690                 filter &= ~SHOW_MEM_FILTER_NODES;
2691 
2692         if (fmt) {
2693                 struct va_format vaf;
2694                 va_list args;
2695 
2696                 va_start(args, fmt);
2697 
2698                 vaf.fmt = fmt;
2699                 vaf.va = &args;
2700 
2701                 pr_warn("%pV", &vaf);
2702 
2703                 va_end(args);
2704         }
2705 
2706         pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2707                 current->comm, order, gfp_mask);
2708 
2709         dump_stack();
2710         if (!should_suppress_show_mem())
2711                 show_mem(filter);
2712 }
2713 
2714 static inline struct page *
2715 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2716         const struct alloc_context *ac, unsigned long *did_some_progress)
2717 {
2718         struct oom_control oc = {
2719                 .zonelist = ac->zonelist,
2720                 .nodemask = ac->nodemask,
2721                 .gfp_mask = gfp_mask,
2722                 .order = order,
2723         };
2724         struct page *page;
2725 
2726         *did_some_progress = 0;
2727 
2728         /*
2729          * Acquire the oom lock.  If that fails, somebody else is
2730          * making progress for us.
2731          */
2732         if (!mutex_trylock(&oom_lock)) {
2733                 *did_some_progress = 1;
2734                 schedule_timeout_uninterruptible(1);
2735                 return NULL;
2736         }
2737 
2738         /*
2739          * Go through the zonelist yet one more time, keep very high watermark
2740          * here, this is only to catch a parallel oom killing, we must fail if
2741          * we're still under heavy pressure.
2742          */
2743         page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2744                                         ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
2745         if (page)
2746                 goto out;
2747 
2748         if (!(gfp_mask & __GFP_NOFAIL)) {
2749                 /* Coredumps can quickly deplete all memory reserves */
2750                 if (current->flags & PF_DUMPCORE)
2751                         goto out;
2752                 /* The OOM killer will not help higher order allocs */
2753                 if (order > PAGE_ALLOC_COSTLY_ORDER)
2754                         goto out;
2755                 /* The OOM killer does not needlessly kill tasks for lowmem */
2756                 if (ac->high_zoneidx < ZONE_NORMAL)
2757                         goto out;
2758                 /* The OOM killer does not compensate for IO-less reclaim */
2759                 if (!(gfp_mask & __GFP_FS)) {
2760                         /*
2761                          * XXX: Page reclaim didn't yield anything,
2762                          * and the OOM killer can't be invoked, but
2763                          * keep looping as per tradition.
2764                          */
2765                         *did_some_progress = 1;
2766                         goto out;
2767                 }
2768                 if (pm_suspended_storage())
2769                         goto out;
2770                 /* The OOM killer may not free memory on a specific node */
2771                 if (gfp_mask & __GFP_THISNODE)
2772                         goto out;
2773         }
2774         /* Exhausted what can be done so it's blamo time */
2775         if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
2776                 *did_some_progress = 1;
2777 out:
2778         mutex_unlock(&oom_lock);
2779         return page;
2780 }
2781 
2782 #ifdef CONFIG_COMPACTION
2783 /* Try memory compaction for high-order allocations before reclaim */
2784 static struct page *
2785 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2786                 int alloc_flags, const struct alloc_context *ac,
2787                 enum migrate_mode mode, int *contended_compaction,
2788                 bool *deferred_compaction)
2789 {
2790         unsigned long compact_result;
2791         struct page *page;
2792 
2793         if (!order)
2794                 return NULL;
2795 
2796         current->flags |= PF_MEMALLOC;
2797         compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2798                                                 mode, contended_compaction);
2799         current->flags &= ~PF_MEMALLOC;
2800 
2801         switch (compact_result) {
2802         case COMPACT_DEFERRED:
2803                 *deferred_compaction = true;
2804                 /* fall-through */
2805         case COMPACT_SKIPPED:
2806                 return NULL;
2807         default:
2808                 break;
2809         }
2810 
2811         /*
2812          * At least in one zone compaction wasn't deferred or skipped, so let's
2813          * count a compaction stall
2814          */
2815         count_vm_event(COMPACTSTALL);
2816 
2817         page = get_page_from_freelist(gfp_mask, order,
2818                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2819 
2820         if (page) {
2821                 struct zone *zone = page_zone(page);
2822 
2823                 zone->compact_blockskip_flush = false;
2824                 compaction_defer_reset(zone, order, true);
2825                 count_vm_event(COMPACTSUCCESS);
2826                 return page;
2827         }
2828 
2829         /*
2830          * It's bad if compaction run occurs and fails. The most likely reason
2831          * is that pages exist, but not enough to satisfy watermarks.
2832          */
2833         count_vm_event(COMPACTFAIL);
2834 
2835         cond_resched();
2836 
2837         return NULL;
2838 }
2839 #else
2840 static inline struct page *
2841 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2842                 int alloc_flags, const struct alloc_context *ac,
2843                 enum migrate_mode mode, int *contended_compaction,
2844                 bool *deferred_compaction)
2845 {
2846         return NULL;
2847 }
2848 #endif /* CONFIG_COMPACTION */
2849 
2850 /* Perform direct synchronous page reclaim */
2851 static int
2852 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
2853                                         const struct alloc_context *ac)
2854 {
2855         struct reclaim_state reclaim_state;
2856         int progress;
2857 
2858         cond_resched();
2859 
2860         /* We now go into synchronous reclaim */
2861         cpuset_memory_pressure_bump();
2862         current->flags |= PF_MEMALLOC;
2863         lockdep_set_current_reclaim_state(gfp_mask);
2864         reclaim_state.reclaimed_slab = 0;
2865         current->reclaim_state = &reclaim_state;
2866 
2867         progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2868                                                                 ac->nodemask);
2869 
2870         current->reclaim_state = NULL;
2871         lockdep_clear_current_reclaim_state();
2872         current->flags &= ~PF_MEMALLOC;
2873 
2874         cond_resched();
2875 
2876         return progress;
2877 }
2878 
2879 /* The really slow allocator path where we enter direct reclaim */
2880 static inline struct page *
2881 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2882                 int alloc_flags, const struct alloc_context *ac,
2883                 unsigned long *did_some_progress)
2884 {
2885         struct page *page = NULL;
2886         bool drained = false;
2887 
2888         *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
2889         if (unlikely(!(*did_some_progress)))
2890                 return NULL;
2891 
2892         /* After successful reclaim, reconsider all zones for allocation */
2893         if (IS_ENABLED(CONFIG_NUMA))
2894                 zlc_clear_zones_full(ac->zonelist);
2895 
2896 retry:
2897         page = get_page_from_freelist(gfp_mask, order,
2898                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2899 
2900         /*
2901          * If an allocation failed after direct reclaim, it could be because
2902          * pages are pinned on the per-cpu lists. Drain them and try again
2903          */
2904         if (!page && !drained) {
2905                 drain_all_pages(NULL);
2906                 drained = true;
2907                 goto retry;
2908         }
2909 
2910         return page;
2911 }
2912 
2913 /*
2914  * This is called in the allocator slow-path if the allocation request is of
2915  * sufficient urgency to ignore watermarks and take other desperate measures
2916  */
2917 static inline struct page *
2918 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2919                                 const struct alloc_context *ac)
2920 {
2921         struct page *page;
2922 
2923         do {
2924                 page = get_page_from_freelist(gfp_mask, order,
2925                                                 ALLOC_NO_WATERMARKS, ac);
2926 
2927                 if (!page && gfp_mask & __GFP_NOFAIL)
2928                         wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2929                                                                         HZ/50);
2930         } while (!page && (gfp_mask & __GFP_NOFAIL));
2931 
2932         return page;
2933 }
2934 
2935 static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
2936 {
2937         struct zoneref *z;
2938         struct zone *zone;
2939 
2940         for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2941                                                 ac->high_zoneidx, ac->nodemask)
2942                 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
2943 }
2944 
2945 static inline int
2946 gfp_to_alloc_flags(gfp_t gfp_mask)
2947 {
2948         int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2949         const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
2950 
2951         /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2952         BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2953 
2954         /*
2955          * The caller may dip into page reserves a bit more if the caller
2956          * cannot run direct reclaim, or if the caller has realtime scheduling
2957          * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
2958          * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
2959          */
2960         alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2961 
2962         if (atomic) {
2963                 /*
2964                  * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2965                  * if it can't schedule.
2966                  */
2967                 if (!(gfp_mask & __GFP_NOMEMALLOC))
2968                         alloc_flags |= ALLOC_HARDER;
2969                 /*
2970                  * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2971                  * comment for __cpuset_node_allowed().
2972                  */
2973                 alloc_flags &= ~ALLOC_CPUSET;
2974         } else if (unlikely(rt_task(current)) && !in_interrupt())
2975                 alloc_flags |= ALLOC_HARDER;
2976 
2977         if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2978                 if (gfp_mask & __GFP_MEMALLOC)
2979                         alloc_flags |= ALLOC_NO_WATERMARKS;
2980                 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2981                         alloc_flags |= ALLOC_NO_WATERMARKS;
2982                 else if (!in_interrupt() &&
2983                                 ((current->flags & PF_MEMALLOC) ||
2984                                  unlikely(test_thread_flag(TIF_MEMDIE))))
2985                         alloc_flags |= ALLOC_NO_WATERMARKS;
2986         }
2987 #ifdef CONFIG_CMA
2988         if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2989                 alloc_flags |= ALLOC_CMA;
2990 #endif
2991         return alloc_flags;
2992 }
2993 
2994 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2995 {
2996         return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
2997 }
2998 
2999 static inline struct page *
3000 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
3001                                                 struct alloc_context *ac)
3002 {
3003         const gfp_t wait = gfp_mask & __GFP_WAIT;
3004         struct page *page = NULL;
3005         int alloc_flags;
3006         unsigned long pages_reclaimed = 0;
3007         unsigned long did_some_progress;
3008         enum migrate_mode migration_mode = MIGRATE_ASYNC;
3009         bool deferred_compaction = false;
3010         int contended_compaction = COMPACT_CONTENDED_NONE;
3011 
3012         /*
3013          * In the slowpath, we sanity check order to avoid ever trying to
3014          * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3015          * be using allocators in order of preference for an area that is
3016          * too large.
3017          */
3018         if (order >= MAX_ORDER) {
3019                 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
3020                 return NULL;
3021         }
3022 
3023         /*
3024          * If this allocation cannot block and it is for a specific node, then
3025          * fail early.  There's no need to wakeup kswapd or retry for a
3026          * speculative node-specific allocation.
3027          */
3028         if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !wait)
3029                 goto nopage;
3030 
3031 retry:
3032         if (!(gfp_mask & __GFP_NO_KSWAPD))
3033                 wake_all_kswapds(order, ac);
3034 
3035         /*
3036          * OK, we're below the kswapd watermark and have kicked background
3037          * reclaim. Now things get more complex, so set up alloc_flags according
3038          * to how we want to proceed.
3039          */
3040         alloc_flags = gfp_to_alloc_flags(gfp_mask);
3041 
3042         /*
3043          * Find the true preferred zone if the allocation is unconstrained by
3044          * cpusets.
3045          */
3046         if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
3047                 struct zoneref *preferred_zoneref;
3048                 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3049                                 ac->high_zoneidx, NULL, &ac->preferred_zone);
3050                 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
3051         }
3052 
3053         /* This is the last chance, in general, before the goto nopage. */
3054         page = get_page_from_freelist(gfp_mask, order,
3055                                 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3056         if (page)
3057                 goto got_pg;
3058 
3059         /* Allocate without watermarks if the context allows */
3060         if (alloc_flags & ALLOC_NO_WATERMARKS) {
3061                 /*
3062                  * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3063                  * the allocation is high priority and these type of
3064                  * allocations are system rather than user orientated
3065                  */
3066                 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3067 
3068                 page = __alloc_pages_high_priority(gfp_mask, order, ac);
3069 
3070                 if (page) {
3071                         goto got_pg;
3072                 }
3073         }
3074 
3075         /* Atomic allocations - we can't balance anything */
3076         if (!wait) {
3077                 /*
3078                  * All existing users of the deprecated __GFP_NOFAIL are
3079                  * blockable, so warn of any new users that actually allow this
3080                  * type of allocation to fail.
3081                  */
3082                 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
3083                 goto nopage;
3084         }
3085 
3086         /* Avoid recursion of direct reclaim */
3087         if (current->flags & PF_MEMALLOC)
3088                 goto nopage;
3089 
3090         /* Avoid allocations with no watermarks from looping endlessly */
3091         if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3092                 goto nopage;
3093 
3094         /*
3095          * Try direct compaction. The first pass is asynchronous. Subsequent
3096          * attempts after direct reclaim are synchronous
3097          */
3098         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3099                                         migration_mode,
3100                                         &contended_compaction,
3101                                         &deferred_compaction);
3102         if (page)
3103                 goto got_pg;
3104 
3105         /* Checks for THP-specific high-order allocations */
3106         if ((gfp_mask & GFP_TRANSHUGE) == GFP_TRANSHUGE) {
3107                 /*
3108                  * If compaction is deferred for high-order allocations, it is
3109                  * because sync compaction recently failed. If this is the case
3110                  * and the caller requested a THP allocation, we do not want
3111                  * to heavily disrupt the system, so we fail the allocation
3112                  * instead of entering direct reclaim.
3113                  */
3114                 if (deferred_compaction)
3115                         goto nopage;
3116 
3117                 /*
3118                  * In all zones where compaction was attempted (and not
3119                  * deferred or skipped), lock contention has been detected.
3120                  * For THP allocation we do not want to disrupt the others
3121                  * so we fallback to base pages instead.
3122                  */
3123                 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3124                         goto nopage;
3125 
3126                 /*
3127                  * If compaction was aborted due to need_resched(), we do not
3128                  * want to further increase allocation latency, unless it is
3129                  * khugepaged trying to collapse.
3130                  */
3131                 if (contended_compaction == COMPACT_CONTENDED_SCHED
3132                         && !(current->flags & PF_KTHREAD))
3133                         goto nopage;
3134         }
3135 
3136         /*
3137          * It can become very expensive to allocate transparent hugepages at
3138          * fault, so use asynchronous memory compaction for THP unless it is
3139          * khugepaged trying to collapse.
3140          */
3141         if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
3142                                                 (current->flags & PF_KTHREAD))
3143                 migration_mode = MIGRATE_SYNC_LIGHT;
3144 
3145         /* Try direct reclaim and then allocating */
3146         page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3147                                                         &did_some_progress);
3148         if (page)
3149                 goto got_pg;
3150 
3151         /* Do not loop if specifically requested */
3152         if (gfp_mask & __GFP_NORETRY)
3153                 goto noretry;
3154 
3155         /* Keep reclaiming pages as long as there is reasonable progress */
3156         pages_reclaimed += did_some_progress;
3157         if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3158             ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
3159                 /* Wait for some write requests to complete then retry */
3160                 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
3161                 goto retry;
3162         }
3163 
3164         /* Reclaim has failed us, start killing things */
3165         page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3166         if (page)
3167                 goto got_pg;
3168 
3169         /* Retry as long as the OOM killer is making progress */
3170         if (did_some_progress)
3171                 goto retry;
3172 
3173 noretry:
3174         /*
3175          * High-order allocations do not necessarily loop after
3176          * direct reclaim and reclaim/compaction depends on compaction
3177          * being called after reclaim so call directly if necessary
3178          */
3179         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3180                                             ac, migration_mode,
3181                                             &contended_compaction,
3182                                             &deferred_compaction);
3183         if (page)
3184                 goto got_pg;
3185 nopage:
3186         warn_alloc_failed(gfp_mask, order, NULL);
3187 got_pg:
3188         return page;
3189 }
3190 
3191 /*
3192  * This is the 'heart' of the zoned buddy allocator.
3193  */
3194 struct page *
3195 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3196                         struct zonelist *zonelist, nodemask_t *nodemask)
3197 {
3198         struct zoneref *preferred_zoneref;
3199         struct page *page = NULL;
3200         unsigned int cpuset_mems_cookie;
3201         int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
3202         gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
3203         struct alloc_context ac = {
3204                 .high_zoneidx = gfp_zone(gfp_mask),
3205                 .nodemask = nodemask,
3206                 .migratetype = gfpflags_to_migratetype(gfp_mask),
3207         };
3208 
3209         gfp_mask &= gfp_allowed_mask;
3210 
3211         lockdep_trace_alloc(gfp_mask);
3212 
3213         might_sleep_if(gfp_mask & __GFP_WAIT);
3214 
3215         if (should_fail_alloc_page(gfp_mask, order))
3216                 return NULL;
3217 
3218         /*
3219          * Check the zones suitable for the gfp_mask contain at least one
3220          * valid zone. It's possible to have an empty zonelist as a result
3221          * of __GFP_THISNODE and a memoryless node
3222          */
3223         if (unlikely(!zonelist->_zonerefs->zone))
3224                 return NULL;
3225 
3226         if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
3227                 alloc_flags |= ALLOC_CMA;
3228 
3229 retry_cpuset:
3230         cpuset_mems_cookie = read_mems_allowed_begin();
3231 
3232         /* We set it here, as __alloc_pages_slowpath might have changed it */
3233         ac.zonelist = zonelist;
3234         /* The preferred zone is used for statistics later */
3235         preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3236                                 ac.nodemask ? : &cpuset_current_mems_allowed,
3237                                 &ac.preferred_zone);
3238         if (!ac.preferred_zone)
3239                 goto out;
3240         ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
3241 
3242         /* First allocation attempt */
3243         alloc_mask = gfp_mask|__GFP_HARDWALL;
3244         page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
3245         if (unlikely(!page)) {
3246                 /*
3247                  * Runtime PM, block IO and its error handling path
3248                  * can deadlock because I/O on the device might not
3249                  * complete.
3250                  */
3251                 alloc_mask = memalloc_noio_flags(gfp_mask);
3252 
3253                 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
3254         }
3255 
3256         if (kmemcheck_enabled && page)
3257                 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3258 
3259         trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3260 
3261 out:
3262         /*
3263          * When updating a task's mems_allowed, it is possible to race with
3264          * parallel threads in such a way that an allocation can fail while
3265          * the mask is being updated. If a page allocation is about to fail,
3266          * check if the cpuset changed during allocation and if so, retry.
3267          */
3268         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
3269                 goto retry_cpuset;
3270 
3271         return page;
3272 }
3273 EXPORT_SYMBOL(__alloc_pages_nodemask);
3274 
3275 /*
3276  * Common helper functions.
3277  */
3278 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
3279 {
3280         struct page *page;
3281 
3282         /*
3283          * __get_free_pages() returns a 32-bit address, which cannot represent
3284          * a highmem page
3285          */
3286         VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3287 
3288         page = alloc_pages(gfp_mask, order);
3289         if (!page)
3290                 return 0;
3291         return (unsigned long) page_address(page);
3292 }
3293 EXPORT_SYMBOL(__get_free_pages);
3294 
3295 unsigned long get_zeroed_page(gfp_t gfp_mask)
3296 {
3297         return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
3298 }
3299 EXPORT_SYMBOL(get_zeroed_page);
3300 
3301 void __free_pages(struct page *page, unsigned int order)
3302 {
3303         if (put_page_testzero(page)) {
3304                 if (order == 0)
3305                         free_hot_cold_page(page, false);
3306                 else
3307                         __free_pages_ok(page, order);
3308         }
3309 }
3310 
3311 EXPORT_SYMBOL(__free_pages);
3312 
3313 void free_pages(unsigned long addr, unsigned int order)
3314 {
3315         if (addr != 0) {
3316                 VM_BUG_ON(!virt_addr_valid((void *)addr));
3317                 __free_pages(virt_to_page((void *)addr), order);
3318         }
3319 }
3320 
3321 EXPORT_SYMBOL(free_pages);
3322 
3323 /*
3324  * Page Fragment:
3325  *  An arbitrary-length arbitrary-offset area of memory which resides
3326  *  within a 0 or higher order page.  Multiple fragments within that page
3327  *  are individually refcounted, in the page's reference counter.
3328  *
3329  * The page_frag functions below provide a simple allocation framework for
3330  * page fragments.  This is used by the network stack and network device
3331  * drivers to provide a backing region of memory for use as either an
3332  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3333  */
3334 static struct page *__page_frag_refill(struct page_frag_cache *nc,
3335                                        gfp_t gfp_mask)
3336 {
3337         struct page *page = NULL;
3338         gfp_t gfp = gfp_mask;
3339 
3340 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3341         gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3342                     __GFP_NOMEMALLOC;
3343         page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3344                                 PAGE_FRAG_CACHE_MAX_ORDER);
3345         nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3346 #endif
3347         if (unlikely(!page))
3348                 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3349 
3350         nc->va = page ? page_address(page) : NULL;
3351 
3352         return page;
3353 }
3354 
3355 void *__alloc_page_frag(struct page_frag_cache *nc,
3356                         unsigned int fragsz, gfp_t gfp_mask)
3357 {
3358         unsigned int size = PAGE_SIZE;
3359         struct page *page;
3360         int offset;
3361 
3362         if (unlikely(!nc->va)) {
3363 refill:
3364                 page = __page_frag_refill(nc, gfp_mask);
3365                 if (!page)
3366                         return NULL;
3367 
3368 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3369                 /* if size can vary use size else just use PAGE_SIZE */
3370                 size = nc->size;
3371 #endif
3372                 /* Even if we own the page, we do not use atomic_set().
3373                  * This would break get_page_unless_zero() users.
3374                  */
3375                 atomic_add(size - 1, &page->_count);
3376 
3377                 /* reset page count bias and offset to start of new frag */
3378                 nc->pfmemalloc = page_is_pfmemalloc(page);
3379                 nc->pagecnt_bias = size;
3380                 nc->offset = size;
3381         }
3382 
3383         offset = nc->offset - fragsz;
3384         if (unlikely(offset < 0)) {
3385                 page = virt_to_page(nc->va);
3386 
3387                 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
3388                         goto refill;
3389 
3390 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3391                 /* if size can vary use size else just use PAGE_SIZE */
3392                 size = nc->size;
3393 #endif
3394                 /* OK, page count is 0, we can safely set it */
3395                 atomic_set(&page->_count, size);
3396 
3397                 /* reset page count bias and offset to start of new frag */
3398                 nc->pagecnt_bias = size;
3399                 offset = size - fragsz;
3400         }
3401 
3402         nc->pagecnt_bias--;
3403         nc->offset = offset;
3404 
3405         return nc->va + offset;
3406 }
3407 EXPORT_SYMBOL(__alloc_page_frag);
3408 
3409 /*
3410  * Frees a page fragment allocated out of either a compound or order 0 page.
3411  */
3412 void __free_page_frag(void *addr)
3413 {
3414         struct page *page = virt_to_head_page(addr);
3415 
3416         if (unlikely(put_page_testzero(page)))
3417                 __free_pages_ok(page, compound_order(page));
3418 }
3419 EXPORT_SYMBOL(__free_page_frag);
3420 
3421 /*
3422  * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3423  * of the current memory cgroup.
3424  *
3425  * It should be used when the caller would like to use kmalloc, but since the
3426  * allocation is large, it has to fall back to the page allocator.
3427  */
3428 struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3429 {
3430         struct page *page;
3431         struct mem_cgroup *memcg = NULL;
3432 
3433         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3434                 return NULL;
3435         page = alloc_pages(gfp_mask, order);
3436         memcg_kmem_commit_charge(page, memcg, order);
3437         return page;
3438 }
3439 
3440 struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3441 {
3442         struct page *page;
3443         struct mem_cgroup *memcg = NULL;
3444 
3445         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
3446                 return NULL;
3447         page = alloc_pages_node(nid, gfp_mask, order);
3448         memcg_kmem_commit_charge(page, memcg, order);
3449         return page;
3450 }
3451 
3452 /*
3453  * __free_kmem_pages and free_kmem_pages will free pages allocated with
3454  * alloc_kmem_pages.
3455  */
3456 void __free_kmem_pages(struct page *page, unsigned int order)
3457 {
3458         memcg_kmem_uncharge_pages(page, order);
3459         __free_pages(page, order);
3460 }
3461 
3462 void free_kmem_pages(unsigned long addr, unsigned int order)
3463 {
3464         if (addr != 0) {
3465                 VM_BUG_ON(!virt_addr_valid((void *)addr));
3466                 __free_kmem_pages(virt_to_page((void *)addr), order);
3467         }
3468 }
3469 
3470 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
3471 {
3472         if (addr) {
3473                 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3474                 unsigned long used = addr + PAGE_ALIGN(size);
3475 
3476                 split_page(virt_to_page((void *)addr), order);
3477                 while (used < alloc_end) {
3478                         free_page(used);
3479                         used += PAGE_SIZE;
3480                 }
3481         }
3482         return (void *)addr;
3483 }
3484 
3485 /**
3486  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3487  * @size: the number of bytes to allocate
3488  * @gfp_mask: GFP flags for the allocation
3489  *
3490  * This function is similar to alloc_pages(), except that it allocates the
3491  * minimum number of pages to satisfy the request.  alloc_pages() can only
3492  * allocate memory in power-of-two pages.
3493  *
3494  * This function is also limited by MAX_ORDER.
3495  *
3496  * Memory allocated by this function must be released by free_pages_exact().
3497  */
3498 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3499 {
3500         unsigned int order = get_order(size);
3501         unsigned long addr;
3502 
3503         addr = __get_free_pages(gfp_mask, order);
3504         return make_alloc_exact(addr, order, size);
3505 }
3506 EXPORT_SYMBOL(alloc_pages_exact);
3507 
3508 /**
3509  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3510  *                         pages on a node.
3511  * @nid: the preferred node ID where memory should be allocated
3512  * @size: the number of bytes to allocate
3513  * @gfp_mask: GFP flags for the allocation
3514  *
3515  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3516  * back.
3517  */
3518 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
3519 {
3520         unsigned order = get_order(size);
3521         struct page *p = alloc_pages_node(nid, gfp_mask, order);
3522         if (!p)
3523                 return NULL;
3524         return make_alloc_exact((unsigned long)page_address(p), order, size);
3525 }
3526 
3527 /**
3528  * free_pages_exact - release memory allocated via alloc_pages_exact()
3529  * @virt: the value returned by alloc_pages_exact.
3530  * @size: size of allocation, same value as passed to alloc_pages_exact().
3531  *
3532  * Release the memory allocated by a previous call to alloc_pages_exact.
3533  */
3534 void free_pages_exact(void *virt, size_t size)
3535 {
3536         unsigned long addr = (unsigned long)virt;
3537         unsigned long end = addr + PAGE_ALIGN(size);
3538 
3539         while (addr < end) {
3540                 free_page(addr);
3541                 addr += PAGE_SIZE;
3542         }
3543 }
3544 EXPORT_SYMBOL(free_pages_exact);
3545 
3546 /**
3547  * nr_free_zone_pages - count number of pages beyond high watermark
3548  * @offset: The zone index of the highest zone
3549  *
3550  * nr_free_zone_pages() counts the number of counts pages which are beyond the
3551  * high watermark within all zones at or below a given zone index.  For each
3552  * zone, the number of pages is calculated as:
3553  *     managed_pages - high_pages
3554  */
3555 static unsigned long nr_free_zone_pages(int offset)
3556 {
3557         struct zoneref *z;
3558         struct zone *zone;
3559 
3560         /* Just pick one node, since fallback list is circular */
3561         unsigned long sum = 0;
3562 
3563         struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3564 
3565         for_each_zone_zonelist(zone, z, zonelist, offset) {
3566                 unsigned long size = zone->managed_pages;
3567                 unsigned long high = high_wmark_pages(zone);
3568                 if (size > high)
3569                         sum += size - high;
3570         }
3571 
3572         return sum;
3573 }
3574 
3575 /**
3576  * nr_free_buffer_pages - count number of pages beyond high watermark
3577  *
3578  * nr_free_buffer_pages() counts the number of pages which are beyond the high
3579  * watermark within ZONE_DMA and ZONE_NORMAL.
3580  */
3581 unsigned long nr_free_buffer_pages(void)
3582 {
3583         return nr_free_zone_pages(gfp_zone(GFP_USER));
3584 }
3585 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3586 
3587 /**
3588  * nr_free_pagecache_pages - count number of pages beyond high watermark
3589  *
3590  * nr_free_pagecache_pages() counts the number of pages which are beyond the
3591  * high watermark within all zones.
3592  */
3593 unsigned long nr_free_pagecache_pages(void)
3594 {
3595         return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3596 }
3597 
3598 static inline void show_node(struct zone *zone)
3599 {
3600         if (IS_ENABLED(CONFIG_NUMA))
3601                 printk("Node %d ", zone_to_nid(zone));
3602 }
3603 
3604 void si_meminfo(struct sysinfo *val)
3605 {
3606         val->totalram = totalram_pages;
3607         val->sharedram = global_page_state(NR_SHMEM);
3608         val->freeram = global_page_state(NR_FREE_PAGES);
3609         val->bufferram = nr_blockdev_pages();
3610         val->totalhigh = totalhigh_pages;
3611         val->freehigh = nr_free_highpages();
3612         val->mem_unit = PAGE_SIZE;
3613 }
3614 
3615 EXPORT_SYMBOL(si_meminfo);
3616 
3617 #ifdef CONFIG_NUMA
3618 void si_meminfo_node(struct sysinfo *val, int nid)
3619 {
3620         int zone_type;          /* needs to be signed */
3621         unsigned long managed_pages = 0;
3622         pg_data_t *pgdat = NODE_DATA(nid);
3623 
3624         for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3625                 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3626         val->totalram = managed_pages;
3627         val->sharedram = node_page_state(nid, NR_SHMEM);
3628         val->freeram = node_page_state(nid, NR_FREE_PAGES);
3629 #ifdef CONFIG_HIGHMEM
3630         val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
3631         val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3632                         NR_FREE_PAGES);
3633 #else
3634         val->totalhigh = 0;
3635         val->freehigh = 0;
3636 #endif
3637         val->mem_unit = PAGE_SIZE;
3638 }
3639 #endif
3640 
3641 /*
3642  * Determine whether the node should be displayed or not, depending on whether
3643  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3644  */
3645 bool skip_free_areas_node(unsigned int flags, int nid)
3646 {
3647         bool ret = false;
3648         unsigned int cpuset_mems_cookie;
3649 
3650         if (!(flags & SHOW_MEM_FILTER_NODES))
3651                 goto out;
3652 
3653         do {
3654                 cpuset_mems_cookie = read_mems_allowed_begin();
3655                 ret = !node_isset(nid, cpuset_current_mems_allowed);
3656         } while (read_mems_allowed_retry(cpuset_mems_cookie));
3657 out:
3658         return ret;
3659 }
3660 
3661 #define K(x) ((x) << (PAGE_SHIFT-10))
3662 
3663 static void show_migration_types(unsigned char type)
3664 {
3665         static const char types[MIGRATE_TYPES] = {
3666                 [MIGRATE_UNMOVABLE]     = 'U',
3667                 [MIGRATE_RECLAIMABLE]   = 'E',
3668                 [MIGRATE_MOVABLE]       = 'M',
3669                 [MIGRATE_RESERVE]       = 'R',
3670 #ifdef CONFIG_CMA
3671                 [MIGRATE_CMA]           = 'C',
3672 #endif
3673 #ifdef CONFIG_MEMORY_ISOLATION
3674                 [MIGRATE_ISOLATE]       = 'I',
3675 #endif
3676         };
3677         char tmp[MIGRATE_TYPES + 1];
3678         char *p = tmp;
3679         int i;
3680 
3681         for (i = 0; i < MIGRATE_TYPES; i++) {
3682                 if (type & (1 << i))
3683                         *p++ = types[i];
3684         }
3685 
3686         *p = '\0';
3687         printk("(%s) ", tmp);
3688 }
3689 
3690 /*
3691  * Show free area list (used inside shift_scroll-lock stuff)
3692  * We also calculate the percentage fragmentation. We do this by counting the
3693  * memory on each free list with the exception of the first item on the list.
3694  *
3695  * Bits in @filter:
3696  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3697  *   cpuset.
3698  */
3699 void show_free_areas(unsigned int filter)
3700 {
3701         unsigned long free_pcp = 0;
3702         int cpu;
3703         struct zone *zone;
3704 
3705         for_each_populated_zone(zone) {
3706                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3707                         continue;
3708 
3709                 for_each_online_cpu(cpu)
3710                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3711         }
3712 
3713         printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3714                 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3715                 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3716                 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3717                 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3718                 " free:%lu free_pcp:%lu free_cma:%lu\n",
3719                 global_page_state(NR_ACTIVE_ANON),
3720                 global_page_state(NR_INACTIVE_ANON),
3721                 global_page_state(NR_ISOLATED_ANON),
3722                 global_page_state(NR_ACTIVE_FILE),
3723                 global_page_state(NR_INACTIVE_FILE),
3724                 global_page_state(NR_ISOLATED_FILE),
3725                 global_page_state(NR_UNEVICTABLE),
3726                 global_page_state(NR_FILE_DIRTY),
3727                 global_page_state(NR_WRITEBACK),
3728                 global_page_state(NR_UNSTABLE_NFS),
3729                 global_page_state(NR_SLAB_RECLAIMABLE),
3730                 global_page_state(NR_SLAB_UNRECLAIMABLE),
3731                 global_page_state(NR_FILE_MAPPED),
3732                 global_page_state(NR_SHMEM),
3733                 global_page_state(NR_PAGETABLE),
3734                 global_page_state(NR_BOUNCE),
3735                 global_page_state(NR_FREE_PAGES),
3736                 free_pcp,
3737                 global_page_state(NR_FREE_CMA_PAGES));
3738 
3739         for_each_populated_zone(zone) {
3740                 int i;
3741 
3742                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3743                         continue;
3744 
3745                 free_pcp = 0;
3746                 for_each_online_cpu(cpu)
3747                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3748 
3749                 show_node(zone);
3750                 printk("%s"
3751                         " free:%lukB"
3752                         " min:%lukB"
3753                         " low:%lukB"
3754                         " high:%lukB"
3755                         " active_anon:%lukB"
3756                         " inactive_anon:%lukB"
3757                         " active_file:%lukB"
3758                         " inactive_file:%lukB"
3759                         " unevictable:%lukB"
3760                         " isolated(anon):%lukB"
3761                         " isolated(file):%lukB"
3762                         " present:%lukB"
3763                         " managed:%lukB"
3764                         " mlocked:%lukB"
3765                         " dirty:%lukB"
3766                         " writeback:%lukB"
3767                         " mapped:%lukB"
3768                         " shmem:%lukB"
3769                         " slab_reclaimable:%lukB"
3770                         " slab_unreclaimable:%lukB"
3771                         " kernel_stack:%lukB"
3772                         " pagetables:%lukB"
3773                         " unstable:%lukB"
3774                         " bounce:%lukB"
3775                         " free_pcp:%lukB"
3776                         " local_pcp:%ukB"
3777                         " free_cma:%lukB"
3778                         " writeback_tmp:%lukB"
3779                         " pages_scanned:%lu"
3780                         " all_unreclaimable? %s"
3781                         "\n",
3782                         zone->name,
3783                         K(zone_page_state(zone, NR_FREE_PAGES)),
3784                         K(min_wmark_pages(zone)),
3785                         K(low_wmark_pages(zone)),
3786                         K(high_wmark_pages(zone)),
3787                         K(zone_page_state(zone, NR_ACTIVE_ANON)),
3788                         K(zone_page_state(zone, NR_INACTIVE_ANON)),
3789                         K(zone_page_state(zone, NR_ACTIVE_FILE)),
3790                         K(zone_page_state(zone, NR_INACTIVE_FILE)),
3791                         K(zone_page_state(zone, NR_UNEVICTABLE)),
3792                         K(zone_page_state(zone, NR_ISOLATED_ANON)),
3793                         K(zone_page_state(zone, NR_ISOLATED_FILE)),
3794                         K(zone->present_pages),
3795                         K(zone->managed_pages),
3796                         K(zone_page_state(zone, NR_MLOCK)),
3797                         K(zone_page_state(zone, NR_FILE_DIRTY)),
3798                         K(zone_page_state(zone, NR_WRITEBACK)),
3799                         K(zone_page_state(zone, NR_FILE_MAPPED)),
3800                         K(zone_page_state(zone, NR_SHMEM)),
3801                         K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3802                         K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3803                         zone_page_state(zone, NR_KERNEL_STACK) *
3804                                 THREAD_SIZE / 1024,
3805                         K(zone_page_state(zone, NR_PAGETABLE)),
3806                         K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3807                         K(zone_page_state(zone, NR_BOUNCE)),
3808                         K(free_pcp),
3809                         K(this_cpu_read(zone->pageset->pcp.count)),
3810                         K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3811                         K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3812                         K(zone_page_state(zone, NR_PAGES_SCANNED)),
3813                         (!zone_reclaimable(zone) ? "yes" : "no")
3814                         );
3815                 printk("lowmem_reserve[]:");
3816                 for (i = 0; i < MAX_NR_ZONES; i++)
3817                         printk(" %ld", zone->lowmem_reserve[i]);
3818                 printk("\n");
3819         }
3820 
3821         for_each_populated_zone(zone) {
3822                 unsigned long nr[MAX_ORDER], flags, order, total = 0;
3823                 unsigned char types[MAX_ORDER];
3824 
3825                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3826                         continue;
3827                 show_node(zone);
3828                 printk("%s: ", zone->name);
3829 
3830                 spin_lock_irqsave(&zone->lock, flags);
3831                 for (order = 0; order < MAX_ORDER; order++) {
3832                         struct free_area *area = &zone->free_area[order];
3833                         int type;
3834 
3835                         nr[order] = area->nr_free;
3836                         total += nr[order] << order;
3837 
3838                         types[order] = 0;
3839                         for (type = 0; type < MIGRATE_TYPES; type++) {
3840                                 if (!list_empty(&area->free_list[type]))
3841                                         types[order] |= 1 << type;
3842                         }
3843                 }
3844                 spin_unlock_irqrestore(&zone->lock, flags);
3845                 for (order = 0; order < MAX_ORDER; order++) {
3846                         printk("%lu*%lukB ", nr[order], K(1UL) << order);
3847                         if (nr[order])
3848                                 show_migration_types(types[order]);
3849                 }
3850                 printk("= %lukB\n", K(total));
3851         }
3852 
3853         hugetlb_show_meminfo();
3854 
3855         printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3856 
3857         show_swap_cache_info();
3858 }
3859 
3860 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3861 {
3862         zoneref->zone = zone;
3863         zoneref->zone_idx = zone_idx(zone);
3864 }
3865 
3866 /*
3867  * Builds allocation fallback zone lists.
3868  *
3869  * Add all populated zones of a node to the zonelist.
3870  */
3871 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3872                                 int nr_zones)
3873 {
3874         struct zone *zone;
3875         enum zone_type zone_type = MAX_NR_ZONES;
3876 
3877         do {
3878                 zone_type--;
3879                 zone = pgdat->node_zones + zone_type;
3880                 if (populated_zone(zone)) {
3881                         zoneref_set_zone(zone,
3882                                 &zonelist->_zonerefs[nr_zones++]);
3883                         check_highest_zone(zone_type);
3884                 }
3885         } while (zone_type);
3886 
3887         return nr_zones;
3888 }
3889 
3890 
3891 /*
3892  *  zonelist_order:
3893  *  0 = automatic detection of better ordering.
3894  *  1 = order by ([node] distance, -zonetype)
3895  *  2 = order by (-zonetype, [node] distance)
3896  *
3897  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3898  *  the same zonelist. So only NUMA can configure this param.
3899  */
3900 #define ZONELIST_ORDER_DEFAULT  0
3901 #define ZONELIST_ORDER_NODE     1
3902 #define ZONELIST_ORDER_ZONE     2
3903 
3904 /* zonelist order in the kernel.
3905  * set_zonelist_order() will set this to NODE or ZONE.
3906  */
3907 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3908 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3909 
3910 
3911 #ifdef CONFIG_NUMA
3912 /* The value user specified ....changed by config */
3913 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3914 /* string for sysctl */
3915 #define NUMA_ZONELIST_ORDER_LEN 16
3916 char numa_zonelist_order[16] = "default";
3917 
3918 /*
3919  * interface for configure zonelist ordering.
3920  * command line option "numa_zonelist_order"
3921  *      = "[dD]efault   - default, automatic configuration.
3922  *      = "[nN]ode      - order by node locality, then by zone within node
3923  *      = "[zZ]one      - order by zone, then by locality within zone
3924  */
3925 
3926 static int __parse_numa_zonelist_order(char *s)
3927 {
3928         if (*s == 'd' || *s == 'D') {
3929                 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3930         } else if (*s == 'n' || *s == 'N') {
3931                 user_zonelist_order = ZONELIST_ORDER_NODE;
3932         } else if (*s == 'z' || *s == 'Z') {
3933                 user_zonelist_order = ZONELIST_ORDER_ZONE;
3934         } else {
3935                 printk(KERN_WARNING
3936                         "Ignoring invalid numa_zonelist_order value:  "
3937                         "%s\n", s);
3938                 return -EINVAL;
3939         }
3940         return 0;
3941 }
3942 
3943 static __init int setup_numa_zonelist_order(char *s)
3944 {
3945         int ret;
3946 
3947         if (!s)
3948                 return 0;
3949 
3950         ret = __parse_numa_zonelist_order(s);
3951         if (ret == 0)
3952                 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3953 
3954         return ret;
3955 }
3956 early_param("numa_zonelist_order", setup_numa_zonelist_order);
3957 
3958 /*
3959  * sysctl handler for numa_zonelist_order
3960  */
3961 int numa_zonelist_order_handler(struct ctl_table *table, int write,
3962                 void __user *buffer, size_t *length,
3963                 loff_t *ppos)
3964 {
3965         char saved_string[NUMA_ZONELIST_ORDER_LEN];
3966         int ret;
3967         static DEFINE_MUTEX(zl_order_mutex);
3968 
3969         mutex_lock(&zl_order_mutex);
3970         if (write) {
3971                 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3972                         ret = -EINVAL;
3973                         goto out;
3974                 }
3975                 strcpy(saved_string, (char *)table->data);
3976         }
3977         ret = proc_dostring(table, write, buffer, length, ppos);
3978         if (ret)
3979                 goto out;
3980         if (write) {
3981                 int oldval = user_zonelist_order;
3982 
3983                 ret = __parse_numa_zonelist_order((char *)table->data);
3984                 if (ret) {
3985                         /*
3986                          * bogus value.  restore saved string
3987                          */
3988                         strncpy((char *)table->data, saved_string,
3989                                 NUMA_ZONELIST_ORDER_LEN);
3990                         user_zonelist_order = oldval;
3991                 } else if (oldval != user_zonelist_order) {
3992                         mutex_lock(&zonelists_mutex);
3993                         build_all_zonelists(NULL, NULL);
3994                         mutex_unlock(&zonelists_mutex);
3995                 }
3996         }
3997 out:
3998         mutex_unlock(&zl_order_mutex);
3999         return ret;
4000 }
4001 
4002 
4003 #define MAX_NODE_LOAD (nr_online_nodes)
4004 static int node_load[MAX_NUMNODES];
4005 
4006 /**
4007  * find_next_best_node - find the next node that should appear in a given node's fallback list
4008  * @node: node whose fallback list we're appending
4009  * @used_node_mask: nodemask_t of already used nodes
4010  *
4011  * We use a number of factors to determine which is the next node that should
4012  * appear on a given node's fallback list.  The node should not have appeared
4013  * already in @node's fallback list, and it should be the next closest node
4014  * according to the distance array (which contains arbitrary distance values
4015  * from each node to each node in the system), and should also prefer nodes
4016  * with no CPUs, since presumably they'll have very little allocation pressure
4017  * on them otherwise.
4018  * It returns -1 if no node is found.
4019  */
4020 static int find_next_best_node(int node, nodemask_t *used_node_mask)
4021 {
4022         int n, val;
4023         int min_val = INT_MAX;
4024         int best_node = NUMA_NO_NODE;
4025         const struct cpumask *tmp = cpumask_of_node(0);
4026 
4027         /* Use the local node if we haven't already */
4028         if (!node_isset(node, *used_node_mask)) {
4029                 node_set(node, *used_node_mask);
4030                 return node;
4031         }
4032 
4033         for_each_node_state(n, N_MEMORY) {
4034 
4035                 /* Don't want a node to appear more than once */
4036                 if (node_isset(n, *used_node_mask))
4037                         continue;
4038 
4039                 /* Use the distance array to find the distance */
4040                 val = node_distance(node, n);
4041 
4042                 /* Penalize nodes under us ("prefer the next node") */
4043                 val += (n < node);
4044 
4045                 /* Give preference to headless and unused nodes */
4046                 tmp = cpumask_of_node(n);
4047                 if (!cpumask_empty(tmp))
4048                         val += PENALTY_FOR_NODE_WITH_CPUS;
4049 
4050                 /* Slight preference for less loaded node */
4051                 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4052                 val += node_load[n];
4053 
4054                 if (val < min_val) {
4055                         min_val = val;
4056                         best_node = n;
4057                 }
4058         }
4059 
4060         if (best_node >= 0)
4061                 node_set(best_node, *used_node_mask);
4062 
4063         return best_node;
4064 }
4065 
4066 
4067 /*
4068  * Build zonelists ordered by node and zones within node.
4069  * This results in maximum locality--normal zone overflows into local
4070  * DMA zone, if any--but risks exhausting DMA zone.
4071  */
4072 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
4073 {
4074         int j;
4075         struct zonelist *zonelist;
4076 
4077         zonelist = &pgdat->node_zonelists[0];
4078         for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
4079                 ;
4080         j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4081         zonelist->_zonerefs[j].zone = NULL;
4082         zonelist->_zonerefs[j].zone_idx = 0;
4083 }
4084 
4085 /*
4086  * Build gfp_thisnode zonelists
4087  */
4088 static void build_thisnode_zonelists(pg_data_t *pgdat)
4089 {
4090         int j;
4091         struct zonelist *zonelist;
4092 
4093         zonelist = &pgdat->node_zonelists[1];
4094         j = build_zonelists_node(pgdat, zonelist, 0);
4095         zonelist->_zonerefs[j].zone = NULL;
4096         zonelist->_zonerefs[j].zone_idx = 0;
4097 }
4098 
4099 /*
4100  * Build zonelists ordered by zone and nodes within zones.
4101  * This results in conserving DMA zone[s] until all Normal memory is
4102  * exhausted, but results in overflowing to remote node while memory
4103  * may still exist in local DMA zone.
4104  */
4105 static int node_order[MAX_NUMNODES];
4106 
4107 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4108 {
4109         int pos, j, node;
4110         int zone_type;          /* needs to be signed */
4111         struct zone *z;
4112         struct zonelist *zonelist;
4113 
4114         zonelist = &pgdat->node_zonelists[0];
4115         pos = 0;
4116         for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4117                 for (j = 0; j < nr_nodes; j++) {
4118                         node = node_order[j];
4119                         z = &NODE_DATA(node)->node_zones[zone_type];
4120                         if (populated_zone(z)) {
4121                                 zoneref_set_zone(z,
4122                                         &zonelist->_zonerefs[pos++]);
4123                                 check_highest_zone(zone_type);
4124                         }
4125                 }
4126         }
4127         zonelist->_zonerefs[pos].zone = NULL;
4128         zonelist->_zonerefs[pos].zone_idx = 0;
4129 }
4130 
4131 #if defined(CONFIG_64BIT)
4132 /*
4133  * Devices that require DMA32/DMA are relatively rare and do not justify a
4134  * penalty to every machine in case the specialised case applies. Default
4135  * to Node-ordering on 64-bit NUMA machines
4136  */
4137 static int default_zonelist_order(void)
4138 {
4139         return ZONELIST_ORDER_NODE;
4140 }
4141 #else
4142 /*
4143  * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4144  * by the kernel. If processes running on node 0 deplete the low memory zone
4145  * then reclaim will occur more frequency increasing stalls and potentially
4146  * be easier to OOM if a large percentage of the zone is under writeback or
4147  * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4148  * Hence, default to zone ordering on 32-bit.
4149  */
4150 static int default_zonelist_order(void)
4151 {
4152         return ZONELIST_ORDER_ZONE;
4153 }
4154 #endif /* CONFIG_64BIT */
4155 
4156 static void set_zonelist_order(void)
4157 {
4158         if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4159                 current_zonelist_order = default_zonelist_order();
4160         else
4161                 current_zonelist_order = user_zonelist_order;
4162 }
4163 
4164 static void build_zonelists(pg_data_t *pgdat)
4165 {
4166         int j, node, load;
4167         enum zone_type i;
4168         nodemask_t used_mask;
4169         int local_node, prev_node;
4170         struct zonelist *zonelist;
4171         int order = current_zonelist_order;
4172 
4173         /* initialize zonelists */
4174         for (i = 0; i < MAX_ZONELISTS; i++) {
4175                 zonelist = pgdat->node_zonelists + i;
4176                 zonelist->_zonerefs[0].zone = NULL;
4177                 zonelist->_zonerefs[0].zone_idx = 0;
4178         }
4179 
4180         /* NUMA-aware ordering of nodes */
4181         local_node = pgdat->node_id;
4182         load = nr_online_nodes;
4183         prev_node = local_node;
4184         nodes_clear(used_mask);
4185 
4186         memset(node_order, 0, sizeof(node_order));
4187         j = 0;
4188 
4189         while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4190                 /*
4191                  * We don't want to pressure a particular node.
4192                  * So adding penalty to the first node in same
4193                  * distance group to make it round-robin.
4194                  */
4195                 if (node_distance(local_node, node) !=
4196                     node_distance(local_node, prev_node))
4197                         node_load[node] = load;
4198 
4199                 prev_node = node;
4200                 load--;
4201                 if (order == ZONELIST_ORDER_NODE)
4202                         build_zonelists_in_node_order(pgdat, node);
4203                 else
4204                         node_order[j++] = node; /* remember order */
4205         }
4206 
4207         if (order == ZONELIST_ORDER_ZONE) {
4208                 /* calculate node order -- i.e., DMA last! */
4209                 build_zonelists_in_zone_order(pgdat, j);
4210         }
4211 
4212         build_thisnode_zonelists(pgdat);
4213 }
4214 
4215 /* Construct the zonelist performance cache - see further mmzone.h */
4216 static void build_zonelist_cache(pg_data_t *pgdat)
4217 {
4218         struct zonelist *zonelist;
4219         struct zonelist_cache *zlc;
4220         struct zoneref *z;
4221 
4222         zonelist = &pgdat->node_zonelists[0];
4223         zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
4224         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
4225         for (z = zonelist->_zonerefs; z->zone; z++)
4226                 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
4227 }
4228 
4229 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4230 /*
4231  * Return node id of node used for "local" allocations.
4232  * I.e., first node id of first zone in arg node's generic zonelist.
4233  * Used for initializing percpu 'numa_mem', which is used primarily
4234  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4235  */
4236 int local_memory_node(int node)
4237 {
4238         struct zone *zone;
4239 
4240         (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4241                                    gfp_zone(GFP_KERNEL),
4242                                    NULL,
4243                                    &zone);
4244         return zone->node;
4245 }
4246 #endif
4247 
4248 #else   /* CONFIG_NUMA */
4249 
4250 static void set_zonelist_order(void)
4251 {
4252         current_zonelist_order = ZONELIST_ORDER_ZONE;
4253 }
4254 
4255 static void build_zonelists(pg_data_t *pgdat)
4256 {
4257         int node, local_node;
4258         enum zone_type j;
4259         struct zonelist *zonelist;
4260 
4261         local_node = pgdat->node_id;
4262 
4263         zonelist = &pgdat->node_zonelists[0];
4264         j = build_zonelists_node(pgdat, zonelist, 0);
4265 
4266         /*
4267          * Now we build the zonelist so that it contains the zones
4268          * of all the other nodes.
4269          * We don't want to pressure a particular node, so when
4270          * building the zones for node N, we make sure that the
4271          * zones coming right after the local ones are those from
4272          * node N+1 (modulo N)
4273          */
4274         for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4275                 if (!node_online(node))
4276                         continue;
4277                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4278         }
4279         for (node = 0; node < local_node; node++) {
4280                 if (!node_online(node))
4281                         continue;
4282                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4283         }
4284 
4285         zonelist->_zonerefs[j].zone = NULL;
4286         zonelist->_zonerefs[j].zone_idx = 0;
4287 }
4288 
4289 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
4290 static void build_zonelist_cache(pg_data_t *pgdat)
4291 {
4292         pgdat->node_zonelists[0].zlcache_ptr = NULL;
4293 }
4294 
4295 #endif  /* CONFIG_NUMA */
4296 
4297 /*
4298  * Boot pageset table. One per cpu which is going to be used for all
4299  * zones and all nodes. The parameters will be set in such a way
4300  * that an item put on a list will immediately be handed over to
4301  * the buddy list. This is safe since pageset manipulation is done
4302  * with interrupts disabled.
4303  *
4304  * The boot_pagesets must be kept even after bootup is complete for
4305  * unused processors and/or zones. They do play a role for bootstrapping
4306  * hotplugged processors.
4307  *
4308  * zoneinfo_show() and maybe other functions do
4309  * not check if the processor is online before following the pageset pointer.
4310  * Other parts of the kernel may not check if the zone is available.
4311  */
4312 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4313 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
4314 static void setup_zone_pageset(struct zone *zone);
4315 
4316 /*
4317  * Global mutex to protect against size modification of zonelists
4318  * as well as to serialize pageset setup for the new populated zone.
4319  */
4320 DEFINE_MUTEX(zonelists_mutex);
4321 
4322 /* return values int ....just for stop_machine() */
4323 static int __build_all_zonelists(void *data)
4324 {
4325         int nid;
4326         int cpu;
4327         pg_data_t *self = data;
4328 
4329 #ifdef CONFIG_NUMA
4330         memset(node_load, 0, sizeof(node_load));
4331 #endif
4332 
4333         if (self && !node_online(self->node_id)) {
4334                 build_zonelists(self);
4335                 build_zonelist_cache(self);
4336         }
4337 
4338         for_each_online_node(nid) {
4339                 pg_data_t *pgdat = NODE_DATA(nid);
4340 
4341                 build_zonelists(pgdat);
4342                 build_zonelist_cache(pgdat);
4343         }
4344 
4345         /*
4346          * Initialize the boot_pagesets that are going to be used
4347          * for bootstrapping processors. The real pagesets for
4348          * each zone will be allocated later when the per cpu
4349          * allocator is available.
4350          *
4351          * boot_pagesets are used also for bootstrapping offline
4352          * cpus if the system is already booted because the pagesets
4353          * are needed to initialize allocators on a specific cpu too.
4354          * F.e. the percpu allocator needs the page allocator which
4355          * needs the percpu allocator in order to allocate its pagesets
4356          * (a chicken-egg dilemma).
4357          */
4358         for_each_possible_cpu(cpu) {
4359                 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4360 
4361 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4362                 /*
4363                  * We now know the "local memory node" for each node--
4364                  * i.e., the node of the first zone in the generic zonelist.
4365                  * Set up numa_mem percpu variable for on-line cpus.  During
4366                  * boot, only the boot cpu should be on-line;  we'll init the
4367                  * secondary cpus' numa_mem as they come on-line.  During
4368                  * node/memory hotplug, we'll fixup all on-line cpus.
4369                  */
4370                 if (cpu_online(cpu))
4371                         set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4372 #endif
4373         }
4374 
4375         return 0;
4376 }
4377 
4378 static noinline void __init
4379 build_all_zonelists_init(void)
4380 {
4381         __build_all_zonelists(NULL);
4382         mminit_verify_zonelist();
4383         cpuset_init_current_mems_allowed();
4384 }
4385 
4386 /*
4387  * Called with zonelists_mutex held always
4388  * unless system_state == SYSTEM_BOOTING.
4389  *
4390  * __ref due to (1) call of __meminit annotated setup_zone_pageset
4391  * [we're only called with non-NULL zone through __meminit paths] and
4392  * (2) call of __init annotated helper build_all_zonelists_init
4393  * [protected by SYSTEM_BOOTING].
4394  */
4395 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
4396 {
4397         set_zonelist_order();
4398 
4399         if (system_state == SYSTEM_BOOTING) {
4400                 build_all_zonelists_init();
4401         } else {
4402 #ifdef CONFIG_MEMORY_HOTPLUG
4403                 if (zone)
4404                         setup_zone_pageset(zone);
4405 #endif
4406                 /* we have to stop all cpus to guarantee there is no user
4407                    of zonelist */
4408                 stop_machine(__build_all_zonelists, pgdat, NULL);
4409                 /* cpuset refresh routine should be here */
4410         }
4411         vm_total_pages = nr_free_pagecache_pages();
4412         /*
4413          * Disable grouping by mobility if the number of pages in the
4414          * system is too low to allow the mechanism to work. It would be
4415          * more accurate, but expensive to check per-zone. This check is
4416          * made on memory-hotadd so a system can start with mobility
4417          * disabled and enable it later
4418          */
4419         if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
4420                 page_group_by_mobility_disabled = 1;
4421         else
4422                 page_group_by_mobility_disabled = 0;
4423 
4424         pr_info("Built %i zonelists in %s order, mobility grouping %s.  "
4425                 "Total pages: %ld\n",
4426                         nr_online_nodes,
4427                         zonelist_order_name[current_zonelist_order],
4428                         page_group_by_mobility_disabled ? "off" : "on",
4429                         vm_total_pages);
4430 #ifdef CONFIG_NUMA
4431         pr_info("Policy zone: %s\n", zone_names[policy_zone]);
4432 #endif
4433 }
4434 
4435 /*
4436  * Helper functions to size the waitqueue hash table.
4437  * Essentially these want to choose hash table sizes sufficiently
4438  * large so that collisions trying to wait on pages are rare.
4439  * But in fact, the number of active page waitqueues on typical
4440  * systems is ridiculously low, less than 200. So this is even
4441  * conservative, even though it seems large.
4442  *
4443  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4444  * waitqueues, i.e. the size of the waitq table given the number of pages.
4445  */
4446 #define PAGES_PER_WAITQUEUE     256
4447 
4448 #ifndef CONFIG_MEMORY_HOTPLUG
4449 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4450 {
4451         unsigned long size = 1;
4452 
4453         pages /= PAGES_PER_WAITQUEUE;
4454 
4455         while (size < pages)
4456                 size <<= 1;
4457 
4458         /*
4459          * Once we have dozens or even hundreds of threads sleeping
4460          * on IO we've got bigger problems than wait queue collision.
4461          * Limit the size of the wait table to a reasonable size.
4462          */
4463         size = min(size, 4096UL);
4464 
4465         return max(size, 4UL);
4466 }
4467 #else
4468 /*
4469  * A zone's size might be changed by hot-add, so it is not possible to determine
4470  * a suitable size for its wait_table.  So we use the maximum size now.
4471  *
4472  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
4473  *
4474  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
4475  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4476  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
4477  *
4478  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4479  * or more by the traditional way. (See above).  It equals:
4480  *
4481  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
4482  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
4483  *    powerpc (64K page size)             : =  (32G +16M)byte.
4484  */
4485 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4486 {
4487         return 4096UL;
4488 }
4489 #endif
4490 
4491 /*
4492  * This is an integer logarithm so that shifts can be used later
4493  * to extract the more random high bits from the multiplicative
4494  * hash function before the remainder is taken.
4495  */
4496 static inline unsigned long wait_table_bits(unsigned long size)
4497 {
4498         return ffz(~size);
4499 }
4500 
4501 /*
4502  * Check if a pageblock contains reserved pages
4503  */
4504 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
4505 {
4506         unsigned long pfn;
4507 
4508         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4509                 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
4510                         return 1;
4511         }
4512         return 0;
4513 }
4514 
4515 /*
4516  * Mark a number of pageblocks as MIGRATE_RESERVE. The number
4517  * of blocks reserved is based on min_wmark_pages(zone). The memory within
4518  * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
4519  * higher will lead to a bigger reserve which will get freed as contiguous
4520  * blocks as reclaim kicks in
4521  */
4522 static void setup_zone_migrate_reserve(struct zone *zone)
4523 {
4524         unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
4525         struct page *page;
4526         unsigned long block_migratetype;
4527         int reserve;
4528         int old_reserve;
4529 
4530         /*
4531          * Get the start pfn, end pfn and the number of blocks to reserve
4532          * We have to be careful to be aligned to pageblock_nr_pages to
4533          * make sure that we always check pfn_valid for the first page in
4534          * the block.
4535          */
4536         start_pfn = zone->zone_start_pfn;
4537         end_pfn = zone_end_pfn(zone);
4538         start_pfn = roundup(start_pfn, pageblock_nr_pages);
4539         reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
4540                                                         pageblock_order;
4541 
4542         /*
4543          * Reserve blocks are generally in place to help high-order atomic
4544          * allocations that are short-lived. A min_free_kbytes value that
4545          * would result in more than 2 reserve blocks for atomic allocations
4546          * is assumed to be in place to help anti-fragmentation for the
4547          * future allocation of hugepages at runtime.
4548          */
4549         reserve = min(2, reserve);
4550         old_reserve = zone->nr_migrate_reserve_block;
4551 
4552         /* When memory hot-add, we almost always need to do nothing */
4553         if (reserve == old_reserve)
4554                 return;
4555         zone->nr_migrate_reserve_block = reserve;
4556 
4557         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
4558                 if (!early_page_nid_uninitialised(pfn, zone_to_nid(zone)))
4559                         return;
4560 
4561                 if (!pfn_valid(pfn))
4562                         continue;
4563                 page = pfn_to_page(pfn);
4564 
4565                 /* Watch out for overlapping nodes */
4566                 if (page_to_nid(page) != zone_to_nid(zone))
4567                         continue;
4568 
4569                 block_migratetype = get_pageblock_migratetype(page);
4570 
4571                 /* Only test what is necessary when the reserves are not met */
4572                 if (reserve > 0) {
4573                         /*
4574                          * Blocks with reserved pages will never free, skip
4575                          * them.
4576                          */
4577                         block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4578                         if (pageblock_is_reserved(pfn, block_end_pfn))
4579                                 continue;
4580 
4581                         /* If this block is reserved, account for it */
4582                         if (block_migratetype == MIGRATE_RESERVE) {
4583                                 reserve--;
4584                                 continue;
4585                         }
4586 
4587                         /* Suitable for reserving if this block is movable */
4588                         if (block_migratetype == MIGRATE_MOVABLE) {
4589                                 set_pageblock_migratetype(page,
4590                                                         MIGRATE_RESERVE);
4591                                 move_freepages_block(zone, page,
4592                                                         MIGRATE_RESERVE);
4593                                 reserve--;
4594                                 continue;
4595                         }
4596                 } else if (!old_reserve) {
4597                         /*
4598                          * At boot time we don't need to scan the whole zone
4599                          * for turning off MIGRATE_RESERVE.
4600                          */
4601                         break;
4602                 }
4603 
4604                 /*
4605                  * If the reserve is met and this is a previous reserved block,
4606                  * take it back
4607                  */
4608                 if (block_migratetype == MIGRATE_RESERVE) {
4609                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4610                         move_freepages_block(zone, page, MIGRATE_MOVABLE);
4611                 }
4612         }
4613 }
4614 
4615 /*
4616  * Initially all pages are reserved - free ones are freed
4617  * up by free_all_bootmem() once the early boot process is
4618  * done. Non-atomic initialization, single-pass.
4619  */
4620 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4621                 unsigned long start_pfn, enum memmap_context context)
4622 {
4623         pg_data_t *pgdat = NODE_DATA(nid);
4624         unsigned long end_pfn = start_pfn + size;
4625         unsigned long pfn;
4626         struct zone *z;
4627         unsigned long nr_initialised = 0;
4628 
4629         if (highest_memmap_pfn < end_pfn - 1)
4630                 highest_memmap_pfn = end_pfn - 1;
4631 
4632         z = &pgdat->node_zones[zone];
4633         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4634                 /*
4635                  * There can be holes in boot-time mem_map[]s
4636                  * handed to this function.  They do not
4637                  * exist on hotplugged memory.
4638                  */
4639                 if (context == MEMMAP_EARLY) {
4640                         if (!early_pfn_valid(pfn))
4641                                 continue;
4642                         if (!early_pfn_in_nid(pfn, nid))
4643                                 continue;
4644                         if (!update_defer_init(pgdat, pfn, end_pfn,
4645                                                 &nr_initialised))
4646                                 break;
4647                 }
4648 
4649                 /*
4650                  * Mark the block movable so that blocks are reserved for
4651                  * movable at startup. This will force kernel allocations
4652                  * to reserve their blocks rather than leaking throughout
4653                  * the address space during boot when many long-lived
4654                  * kernel allocations are made. Later some blocks near
4655                  * the start are marked MIGRATE_RESERVE by
4656                  * setup_zone_migrate_reserve()
4657                  *
4658                  * bitmap is created for zone's valid pfn range. but memmap
4659                  * can be created for invalid pages (for alignment)
4660                  * check here not to call set_pageblock_migratetype() against
4661                  * pfn out of zone.
4662                  */
4663                 if (!(pfn & (pageblock_nr_pages - 1))) {
4664                         struct page *page = pfn_to_page(pfn);
4665 
4666                         __init_single_page(page, pfn, zone, nid);
4667                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4668                 } else {
4669                         __init_single_pfn(pfn, zone, nid);
4670                 }
4671         }
4672 }
4673 
4674 static void __meminit zone_init_free_lists(struct zone *zone)
4675 {
4676         unsigned int order, t;
4677         for_each_migratetype_order(order, t) {
4678                 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4679                 zone->free_area[order].nr_free = 0;
4680         }
4681 }
4682 
4683 #ifndef __HAVE_ARCH_MEMMAP_INIT
4684 #define memmap_init(size, nid, zone, start_pfn) \
4685         memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4686 #endif
4687 
4688 static int zone_batchsize(struct zone *zone)
4689 {
4690 #ifdef CONFIG_MMU
4691         int batch;
4692 
4693         /*
4694          * The per-cpu-pages pools are set to around 1000th of the
4695          * size of the zone.  But no more than 1/2 of a meg.
4696          *
4697          * OK, so we don't know how big the cache is.  So guess.
4698          */
4699         batch = zone->managed_pages / 1024;
4700         if (batch * PAGE_SIZE > 512 * 1024)
4701                 batch = (512 * 1024) / PAGE_SIZE;
4702         batch /= 4;             /* We effectively *= 4 below */
4703         if (batch < 1)
4704                 batch = 1;
4705 
4706         /*
4707          * Clamp the batch to a 2^n - 1 value. Having a power
4708          * of 2 value was found to be more likely to have
4709          * suboptimal cache aliasing properties in some cases.
4710          *
4711          * For example if 2 tasks are alternately allocating
4712          * batches of pages, one task can end up with a lot
4713          * of pages of one half of the possible page colors
4714          * and the other with pages of the other colors.
4715          */
4716         batch = rounddown_pow_of_two(batch + batch/2) - 1;
4717 
4718         return batch;
4719 
4720 #else
4721         /* The deferral and batching of frees should be suppressed under NOMMU
4722          * conditions.
4723          *
4724          * The problem is that NOMMU needs to be able to allocate large chunks
4725          * of contiguous memory as there's no hardware page translation to
4726          * assemble apparent contiguous memory from discontiguous pages.
4727          *
4728          * Queueing large contiguous runs of pages for batching, however,
4729          * causes the pages to actually be freed in smaller chunks.  As there
4730          * can be a significant delay between the individual batches being
4731          * recycled, this leads to the once large chunks of space being
4732          * fragmented and becoming unavailable for high-order allocations.
4733          */
4734         return 0;
4735 #endif
4736 }
4737 
4738 /*
4739  * pcp->high and pcp->batch values are related and dependent on one another:
4740  * ->batch must never be higher then ->high.
4741  * The following function updates them in a safe manner without read side
4742  * locking.
4743  *
4744  * Any new users of pcp->batch and pcp->high should ensure they can cope with
4745  * those fields changing asynchronously (acording the the above rule).
4746  *
4747  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4748  * outside of boot time (or some other assurance that no concurrent updaters
4749  * exist).
4750  */
4751 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4752                 unsigned long batch)
4753 {
4754        /* start with a fail safe value for batch */
4755         pcp->batch = 1;
4756         smp_wmb();
4757 
4758        /* Update high, then batch, in order */
4759         pcp->high = high;
4760         smp_wmb();
4761 
4762         pcp->batch = batch;
4763 }
4764 
4765 /* a companion to pageset_set_high() */
4766 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4767 {
4768         pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4769 }
4770 
4771 static void pageset_init(struct per_cpu_pageset *p)
4772 {
4773         struct per_cpu_pages *pcp;
4774         int migratetype;
4775 
4776         memset(p, 0, sizeof(*p));
4777 
4778         pcp = &p->pcp;
4779         pcp->count = 0;
4780         for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4781                 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4782 }
4783 
4784 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4785 {
4786         pageset_init(p);
4787         pageset_set_batch(p, batch);
4788 }
4789 
4790 /*
4791  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4792  * to the value high for the pageset p.
4793  */
4794 static void pageset_set_high(struct per_cpu_pageset *p,
4795                                 unsigned long high)
4796 {
4797         unsigned long batch = max(1UL, high / 4);
4798         if ((high / 4) > (PAGE_SHIFT * 8))
4799                 batch = PAGE_SHIFT * 8;
4800 
4801         pageset_update(&p->pcp, high, batch);
4802 }
4803 
4804 static void pageset_set_high_and_batch(struct zone *zone,
4805                                        struct per_cpu_pageset *pcp)
4806 {
4807         if (percpu_pagelist_fraction)
4808                 pageset_set_high(pcp,
4809                         (zone->managed_pages /
4810                                 percpu_pagelist_fraction));
4811         else
4812                 pageset_set_batch(pcp, zone_batchsize(zone));
4813 }
4814 
4815 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4816 {
4817         struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4818 
4819         pageset_init(pcp);
4820         pageset_set_high_and_batch(zone, pcp);
4821 }
4822 
4823 static void __meminit setup_zone_pageset(struct zone *zone)
4824 {
4825         int cpu;
4826         zone->pageset = alloc_percpu(struct per_cpu_pageset);
4827         for_each_possible_cpu(cpu)
4828                 zone_pageset_init(zone, cpu);
4829 }
4830 
4831 /*
4832  * Allocate per cpu pagesets and initialize them.
4833  * Before this call only boot pagesets were available.
4834  */
4835 void __init setup_per_cpu_pageset(void)
4836 {
4837         struct zone *zone;
4838 
4839         for_each_populated_zone(zone)
4840                 setup_zone_pageset(zone);
4841 }
4842 
4843 static noinline __init_refok
4844 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4845 {
4846         int i;
4847         size_t alloc_size;
4848 
4849         /*
4850          * The per-page waitqueue mechanism uses hashed waitqueues
4851          * per zone.
4852          */
4853         zone->wait_table_hash_nr_entries =
4854                  wait_table_hash_nr_entries(zone_size_pages);
4855         zone->wait_table_bits =
4856                 wait_table_bits(zone->wait_table_hash_nr_entries);
4857         alloc_size = zone->wait_table_hash_nr_entries
4858                                         * sizeof(wait_queue_head_t);
4859 
4860         if (!slab_is_available()) {
4861                 zone->wait_table = (wait_queue_head_t *)
4862                         memblock_virt_alloc_node_nopanic(
4863                                 alloc_size, zone->zone_pgdat->node_id);
4864         } else {
4865                 /*
4866                  * This case means that a zone whose size was 0 gets new memory
4867                  * via memory hot-add.
4868                  * But it may be the case that a new node was hot-added.  In
4869                  * this case vmalloc() will not be able to use this new node's
4870                  * memory - this wait_table must be initialized to use this new
4871                  * node itself as well.
4872                  * To use this new node's memory, further consideration will be
4873                  * necessary.
4874                  */
4875                 zone->wait_table = vmalloc(alloc_size);
4876         }
4877         if (!zone->wait_table)
4878                 return -ENOMEM;
4879 
4880         for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4881                 init_waitqueue_head(zone->wait_table + i);
4882 
4883         return 0;
4884 }
4885 
4886 static __meminit void zone_pcp_init(struct zone *zone)
4887 {
4888         /*
4889          * per cpu subsystem is not up at this point. The following code
4890          * relies on the ability of the linker to provide the
4891          * offset of a (static) per cpu variable into the per cpu area.
4892          */
4893         zone->pageset = &boot_pageset;
4894 
4895         if (populated_zone(zone))
4896                 printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
4897                         zone->name, zone->present_pages,
4898                                          zone_batchsize(zone));
4899 }
4900 
4901 int __meminit init_currently_empty_zone(struct zone *zone,
4902                                         unsigned long zone_start_pfn,
4903                                         unsigned long size,
4904                                         enum memmap_context context)
4905 {
4906         struct pglist_data *pgdat = zone->zone_pgdat;
4907         int ret;
4908         ret = zone_wait_table_init(zone, size);
4909         if (ret)
4910                 return ret;
4911         pgdat->nr_zones = zone_idx(zone) + 1;
4912 
4913         zone->zone_start_pfn = zone_start_pfn;
4914 
4915         mminit_dprintk(MMINIT_TRACE, "memmap_init",
4916                         "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4917                         pgdat->node_id,
4918                         (unsigned long)zone_idx(zone),
4919                         zone_start_pfn, (zone_start_pfn + size));
4920 
4921         zone_init_free_lists(zone);
4922 
4923         return 0;
4924 }
4925 
4926 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4927 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4928 
4929 /*
4930  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4931  */
4932 int __meminit __early_pfn_to_nid(unsigned long pfn,
4933                                         struct mminit_pfnnid_cache *state)
4934 {
4935         unsigned long start_pfn, end_pfn;
4936         int nid;
4937 
4938         if (state->last_start <= pfn && pfn < state->last_end)
4939                 return state->last_nid;
4940 
4941         nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4942         if (nid != -1) {
4943                 state->last_start = start_pfn;
4944                 state->last_end = end_pfn;
4945                 state->last_nid = nid;
4946         }
4947 
4948         return nid;
4949 }
4950 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4951 
4952 /**
4953  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4954  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4955  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4956  *
4957  * If an architecture guarantees that all ranges registered contain no holes
4958  * and may be freed, this this function may be used instead of calling
4959  * memblock_free_early_nid() manually.
4960  */
4961 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
4962 {
4963         unsigned long start_pfn, end_pfn;
4964         int i, this_nid;
4965 
4966         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4967                 start_pfn = min(start_pfn, max_low_pfn);
4968                 end_pfn = min(end_pfn, max_low_pfn);
4969 
4970                 if (start_pfn < end_pfn)
4971                         memblock_free_early_nid(PFN_PHYS(start_pfn),
4972                                         (end_pfn - start_pfn) << PAGE_SHIFT,
4973                                         this_nid);
4974         }
4975 }
4976 
4977 /**
4978  * sparse_memory_present_with_active_regions - Call memory_present for each active range
4979  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4980  *
4981  * If an architecture guarantees that all ranges registered contain no holes and may
4982  * be freed, this function may be used instead of calling memory_present() manually.
4983  */
4984 void __init sparse_memory_present_with_active_regions(int nid)
4985 {
4986         unsigned long start_pfn, end_pfn;
4987         int i, this_nid;
4988 
4989         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4990                 memory_present(this_nid, start_pfn, end_pfn);
4991 }
4992 
4993 /**
4994  * get_pfn_range_for_nid - Return the start and end page frames for a node
4995  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4996  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4997  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4998  *
4999  * It returns the start and end page frame of a node based on information
5000  * provided by memblock_set_node(). If called for a node
5001  * with no available memory, a warning is printed and the start and end
5002  * PFNs will be 0.
5003  */
5004 void __meminit get_pfn_range_for_nid(unsigned int nid,
5005                         unsigned long *start_pfn, unsigned long *end_pfn)
5006 {
5007         unsigned long this_start_pfn, this_end_pfn;
5008         int i;
5009 
5010         *start_pfn = -1UL;
5011         *end_pfn = 0;
5012 
5013         for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5014                 *start_pfn = min(*start_pfn, this_start_pfn);
5015                 *end_pfn = max(*end_pfn, this_end_pfn);
5016         }
5017 
5018         if (*start_pfn == -1UL)
5019                 *start_pfn = 0;
5020 }
5021 
5022 /*
5023  * This finds a zone that can be used for ZONE_MOVABLE pages. The
5024  * assumption is made that zones within a node are ordered in monotonic
5025  * increasing memory addresses so that the "highest" populated zone is used
5026  */
5027 static void __init find_usable_zone_for_movable(void)
5028 {
5029         int zone_index;
5030         for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5031                 if (zone_index == ZONE_MOVABLE)
5032                         continue;
5033 
5034                 if (arch_zone_highest_possible_pfn[zone_index] >
5035                                 arch_zone_lowest_possible_pfn[zone_index])
5036                         break;
5037         }
5038 
5039         VM_BUG_ON(zone_index == -1);
5040         movable_zone = zone_index;
5041 }
5042 
5043 /*
5044  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5045  * because it is sized independent of architecture. Unlike the other zones,
5046  * the starting point for ZONE_MOVABLE is not fixed. It may be different
5047  * in each node depending on the size of each node and how evenly kernelcore
5048  * is distributed. This helper function adjusts the zone ranges
5049  * provided by the architecture for a given node by using the end of the
5050  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5051  * zones within a node are in order of monotonic increases memory addresses
5052  */
5053 static void __meminit adjust_zone_range_for_zone_movable(int nid,
5054                                         unsigned long zone_type,
5055                                         unsigned long node_start_pfn,
5056                                         unsigned long node_end_pfn,
5057                                         unsigned long *zone_start_pfn,
5058                                         unsigned long *zone_end_pfn)
5059 {
5060         /* Only adjust if ZONE_MOVABLE is on this node */
5061         if (zone_movable_pfn[nid]) {
5062                 /* Size ZONE_MOVABLE */
5063                 if (zone_type == ZONE_MOVABLE) {
5064                         *zone_start_pfn = zone_movable_pfn[nid];
5065                         *zone_end_pfn = min(node_end_pfn,
5066                                 arch_zone_highest_possible_pfn[movable_zone]);
5067 
5068                 /* Adjust for ZONE_MOVABLE starting within this range */
5069                 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
5070                                 *zone_end_pfn > zone_movable_pfn[nid]) {
5071                         *zone_end_pfn = zone_movable_pfn[nid];
5072 
5073                 /* Check if this whole range is within ZONE_MOVABLE */
5074                 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5075                         *zone_start_pfn = *zone_end_pfn;
5076         }
5077 }
5078 
5079 /*
5080  * Return the number of pages a zone spans in a node, including holes
5081  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5082  */
5083 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
5084                                         unsigned long zone_type,
5085                                         unsigned long node_start_pfn,
5086                                         unsigned long node_end_pfn,
5087                                         unsigned long *ignored)
5088 {
5089         unsigned long zone_start_pfn, zone_end_pfn;
5090 
5091         /* When hotadd a new node from cpu_up(), the node should be empty */
5092         if (!node_start_pfn && !node_end_pfn)
5093                 return 0;
5094 
5095         /* Get the start and end of the zone */
5096         zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5097         zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
5098         adjust_zone_range_for_zone_movable(nid, zone_type,
5099                                 node_start_pfn, node_end_pfn,
5100                                 &zone_start_pfn, &zone_end_pfn);
5101 
5102         /* Check that this node has pages within the zone's required range */
5103         if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
5104                 return 0;
5105 
5106         /* Move the zone boundaries inside the node if necessary */
5107         zone_end_pfn = min(zone_end_pfn, node_end_pfn);
5108         zone_start_pfn = max(zone_start_pfn, node_start_pfn);
5109 
5110         /* Return the spanned pages */
5111         return zone_end_pfn - zone_start_pfn;
5112 }
5113 
5114 /*
5115  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5116  * then all holes in the requested range will be accounted for.
5117  */
5118 unsigned long __meminit __absent_pages_in_range(int nid,
5119                                 unsigned long range_start_pfn,
5120                                 unsigned long range_end_pfn)
5121 {
5122         unsigned long nr_absent = range_end_pfn - range_start_pfn;
5123         unsigned long start_pfn, end_pfn;
5124         int i;
5125 
5126         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5127                 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5128                 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5129                 nr_absent -= end_pfn - start_pfn;
5130         }
5131         return nr_absent;
5132 }
5133 
5134 /**
5135  * absent_pages_in_range - Return number of page frames in holes within a range
5136  * @start_pfn: The start PFN to start searching for holes
5137  * @end_pfn: The end PFN to stop searching for holes
5138  *
5139  * It returns the number of pages frames in memory holes within a range.
5140  */
5141 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5142                                                         unsigned long end_pfn)
5143 {
5144         return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5145 }
5146 
5147 /* Return the number of page frames in holes in a zone on a node */
5148 static unsigned long __meminit zone_absent_pages_in_node(int nid,
5149                                         unsigned long zone_type,
5150                                         unsigned long node_start_pfn,
5151                                         unsigned long node_end_pfn,
5152                                         unsigned long *ignored)
5153 {
5154         unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5155         unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
5156         unsigned long zone_start_pfn, zone_end_pfn;
5157 
5158         /* When hotadd a new node from cpu_up(), the node should be empty */
5159         if (!node_start_pfn && !node_end_pfn)
5160                 return 0;
5161 
5162         zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5163         zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
5164 
5165         adjust_zone_range_for_zone_movable(nid, zone_type,
5166                         node_start_pfn, node_end_pfn,
5167                         &zone_start_pfn, &zone_end_pfn);
5168         return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5169 }
5170 
5171 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5172 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
5173                                         unsigned long zone_type,
5174                                         unsigned long node_start_pfn,
5175                                         unsigned long node_end_pfn,
5176                                         unsigned long *zones_size)
5177 {
5178         return zones_size[zone_type];
5179 }
5180 
5181 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
5182                                                 unsigned long zone_type,
5183                                                 unsigned long node_start_pfn,
5184                                                 unsigned long node_end_pfn,
5185                                                 unsigned long *zholes_size)
5186 {
5187         if (!zholes_size)
5188                 return 0;
5189 
5190         return zholes_size[zone_type];
5191 }
5192 
5193 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5194 
5195 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
5196                                                 unsigned long node_start_pfn,
5197                                                 unsigned long node_end_pfn,
5198                                                 unsigned long *zones_size,
5199                                                 unsigned long *zholes_size)
5200 {
5201         unsigned long realtotalpages = 0, totalpages = 0;
5202         enum zone_type i;
5203 
5204         for (i = 0; i < MAX_NR_ZONES; i++) {
5205                 struct zone *zone = pgdat->node_zones + i;
5206                 unsigned long size, real_size;
5207 
5208                 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5209                                                   node_start_pfn,
5210                                                   node_end_pfn,
5211                                                   zones_size);
5212                 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
5213                                                   node_start_pfn, node_end_pfn,
5214                                                   zholes_size);
5215                 zone->spanned_pages = size;
5216                 zone->present_pages = real_size;
5217 
5218                 totalpages += size;
5219                 realtotalpages += real_size;
5220         }
5221 
5222         pgdat->node_spanned_pages = totalpages;
5223         pgdat->node_present_pages = realtotalpages;
5224         printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5225                                                         realtotalpages);
5226 }
5227 
5228 #ifndef CONFIG_SPARSEMEM
5229 /*
5230  * Calculate the size of the zone->blockflags rounded to an unsigned long
5231  * Start by making sure zonesize is a multiple of pageblock_order by rounding
5232  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5233  * round what is now in bits to nearest long in bits, then return it in
5234  * bytes.
5235  */
5236 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
5237 {
5238         unsigned long usemapsize;
5239 
5240         zonesize += zone_start_pfn & (pageblock_nr_pages-1);
5241         usemapsize = roundup(zonesize, pageblock_nr_pages);
5242         usemapsize = usemapsize >> pageblock_order;
5243         usemapsize *= NR_PAGEBLOCK_BITS;
5244         usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5245 
5246         return usemapsize / 8;
5247 }
5248 
5249 static void __init setup_usemap(struct pglist_data *pgdat,
5250                                 struct zone *zone,
5251                                 unsigned long zone_start_pfn,
5252                                 unsigned long zonesize)
5253 {
5254         unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
5255         zone->pageblock_flags = NULL;
5256         if (usemapsize)
5257                 zone->pageblock_flags =
5258                         memblock_virt_alloc_node_nopanic(usemapsize,
5259                                                          pgdat->node_id);
5260 }
5261 #else
5262 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5263                                 unsigned long zone_start_pfn, unsigned long zonesize) {}
5264 #endif /* CONFIG_SPARSEMEM */
5265 
5266 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5267 
5268 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5269 void __paginginit set_pageblock_order(void)
5270 {
5271         unsigned int order;
5272 
5273         /* Check that pageblock_nr_pages has not already been setup */
5274         if (pageblock_order)
5275                 return;
5276 
5277         if (HPAGE_SHIFT > PAGE_SHIFT)
5278                 order = HUGETLB_PAGE_ORDER;
5279         else
5280                 order = MAX_ORDER - 1;
5281 
5282         /*
5283          * Assume the largest contiguous order of interest is a huge page.
5284          * This value may be variable depending on boot parameters on IA64 and
5285          * powerpc.
5286          */
5287         pageblock_order = order;
5288 }
5289 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5290 
5291 /*
5292  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5293  * is unused as pageblock_order is set at compile-time. See
5294  * include/linux/pageblock-flags.h for the values of pageblock_order based on
5295  * the kernel config
5296  */
5297 void __paginginit set_pageblock_order(void)
5298 {
5299 }
5300 
5301 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5302 
5303 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5304                                                    unsigned long present_pages)
5305 {
5306         unsigned long pages = spanned_pages;
5307 
5308         /*
5309          * Provide a more accurate estimation if there are holes within
5310          * the zone and SPARSEMEM is in use. If there are holes within the
5311          * zone, each populated memory region may cost us one or two extra
5312          * memmap pages due to alignment because memmap pages for each
5313          * populated regions may not naturally algined on page boundary.
5314          * So the (present_pages >> 4) heuristic is a tradeoff for that.
5315          */
5316         if (spanned_pages > present_pages + (present_pages >> 4) &&
5317             IS_ENABLED(CONFIG_SPARSEMEM))
5318                 pages = present_pages;
5319 
5320         return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5321 }
5322 
5323 /*
5324  * Set up the zone data structures:
5325  *   - mark all pages reserved
5326  *   - mark all memory queues empty
5327  *   - clear the memory bitmaps
5328  *
5329  * NOTE: pgdat should get zeroed by caller.
5330  */
5331 static void __paginginit free_area_init_core(struct pglist_data *pgdat)
5332 {
5333         enum zone_type j;
5334         int nid = pgdat->node_id;
5335         unsigned long zone_start_pfn = pgdat->node_start_pfn;
5336         int ret;
5337 
5338         pgdat_resize_init(pgdat);
5339 #ifdef CONFIG_NUMA_BALANCING
5340         spin_lock_init(&pgdat->numabalancing_migrate_lock);
5341         pgdat->numabalancing_migrate_nr_pages = 0;
5342         pgdat->numabalancing_migrate_next_window = jiffies;
5343 #endif
5344         init_waitqueue_head(&pgdat->kswapd_wait);
5345         init_waitqueue_head(&pgdat->pfmemalloc_wait);
5346         pgdat_page_ext_init(pgdat);
5347 
5348         for (j = 0; j < MAX_NR_ZONES; j++) {
5349                 struct zone *zone = pgdat->node_zones + j;
5350                 unsigned long size, realsize, freesize, memmap_pages;
5351 
5352                 size = zone->spanned_pages;
5353                 realsize = freesize = zone->present_pages;
5354 
5355                 /*
5356                  * Adjust freesize so that it accounts for how much memory
5357                  * is used by this zone for memmap. This affects the watermark
5358                  * and per-cpu initialisations
5359                  */
5360                 memmap_pages = calc_memmap_size(size, realsize);
5361                 if (!is_highmem_idx(j)) {
5362                         if (freesize >= memmap_pages) {
5363                                 freesize -= memmap_pages;
5364                                 if (memmap_pages)
5365                                         printk(KERN_DEBUG
5366                                                "  %s zone: %lu pages used for memmap\n",
5367                                                zone_names[j], memmap_pages);
5368                         } else
5369                                 printk(KERN_WARNING
5370                                         "  %s zone: %lu pages exceeds freesize %lu\n",
5371                                         zone_names[j], memmap_pages, freesize);
5372                 }
5373 
5374                 /* Account for reserved pages */
5375                 if (j == 0 && freesize > dma_reserve) {
5376                         freesize -= dma_reserve;
5377                         printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
5378                                         zone_names[0], dma_reserve);
5379                 }
5380 
5381                 if (!is_highmem_idx(j))
5382                         nr_kernel_pages += freesize;
5383                 /* Charge for highmem memmap if there are enough kernel pages */
5384                 else if (nr_kernel_pages > memmap_pages * 2)
5385                         nr_kernel_pages -= memmap_pages;
5386                 nr_all_pages += freesize;
5387 
5388                 /*
5389                  * Set an approximate value for lowmem here, it will be adjusted
5390                  * when the bootmem allocator frees pages into the buddy system.
5391                  * And all highmem pages will be managed by the buddy system.
5392                  */
5393                 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
5394 #ifdef CONFIG_NUMA
5395                 zone->node = nid;
5396                 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
5397                                                 / 100;
5398                 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
5399 #endif
5400                 zone->name = zone_names[j];
5401                 spin_lock_init(&zone->lock);
5402                 spin_lock_init(&zone->lru_lock);
5403                 zone_seqlock_init(zone);
5404                 zone->zone_pgdat = pgdat;
5405                 zone_pcp_init(zone);
5406 
5407                 /* For bootup, initialized properly in watermark setup */
5408                 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5409 
5410                 lruvec_init(&zone->lruvec);
5411                 if (!size)
5412                         continue;
5413 
5414                 set_pageblock_order();
5415                 setup_usemap(pgdat, zone, zone_start_pfn, size);
5416                 ret = init_currently_empty_zone(zone, zone_start_pfn,
5417                                                 size, MEMMAP_EARLY);
5418                 BUG_ON(ret);
5419                 memmap_init(size, nid, j, zone_start_pfn);
5420                 zone_start_pfn += size;
5421         }
5422 }
5423 
5424 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
5425 {
5426         /* Skip empty nodes */
5427         if (!pgdat->node_spanned_pages)
5428                 return;
5429 
5430 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5431         /* ia64 gets its own node_mem_map, before this, without bootmem */
5432         if (!pgdat->node_mem_map) {
5433                 unsigned long size, start, end;
5434                 struct page *map;
5435 
5436                 /*
5437                  * The zone's endpoints aren't required to be MAX_ORDER
5438                  * aligned but the node_mem_map endpoints must be in order
5439                  * for the buddy allocator to function correctly.
5440                  */
5441                 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5442                 end = pgdat_end_pfn(pgdat);
5443                 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5444                 size =  (end - start) * sizeof(struct page);
5445                 map = alloc_remap(pgdat->node_id, size);
5446                 if (!map)
5447                         map = memblock_virt_alloc_node_nopanic(size,
5448                                                                pgdat->node_id);
5449                 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
5450         }
5451 #ifndef CONFIG_NEED_MULTIPLE_NODES
5452         /*
5453          * With no DISCONTIG, the global mem_map is just set as node 0's
5454          */
5455         if (pgdat == NODE_DATA(0)) {
5456                 mem_map = NODE_DATA(0)->node_mem_map;
5457 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5458                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
5459                         mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
5460 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5461         }
5462 #endif
5463 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5464 }
5465 
5466 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5467                 unsigned long node_start_pfn, unsigned long *zholes_size)
5468 {
5469         pg_data_t *pgdat = NODE_DATA(nid);
5470         unsigned long start_pfn = 0;
5471         unsigned long end_pfn = 0;
5472 
5473         /* pg_data_t should be reset to zero when it's allocated */
5474         WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
5475 
5476         reset_deferred_meminit(pgdat);
5477         pgdat->node_id = nid;
5478         pgdat->node_start_pfn = node_start_pfn;
5479 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5480         get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
5481         pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5482                 (u64)start_pfn << PAGE_SHIFT,
5483                 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
5484 #endif
5485         calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5486                                   zones_size, zholes_size);
5487 
5488         alloc_node_mem_map(pgdat);
5489 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5490         printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5491                 nid, (unsigned long)pgdat,
5492                 (unsigned long)pgdat->node_mem_map);
5493 #endif
5494 
5495         free_area_init_core(pgdat);
5496 }
5497 
5498 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5499 
5500 #if MAX_NUMNODES > 1
5501 /*
5502  * Figure out the number of possible node ids.
5503  */
5504 void __init setup_nr_node_ids(void)
5505 {
5506         unsigned int highest;
5507 
5508         highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
5509         nr_node_ids = highest + 1;
5510 }
5511 #endif
5512 
5513 /**
5514  * node_map_pfn_alignment - determine the maximum internode alignment
5515  *
5516  * This function should be called after node map is populated and sorted.
5517  * It calculates the maximum power of two alignment which can distinguish
5518  * all the nodes.
5519  *
5520  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5521  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
5522  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
5523  * shifted, 1GiB is enough and this function will indicate so.
5524  *
5525  * This is used to test whether pfn -> nid mapping of the chosen memory
5526  * model has fine enough granularity to avoid incorrect mapping for the
5527  * populated node map.
5528  *
5529  * Returns the determined alignment in pfn's.  0 if there is no alignment
5530  * requirement (single node).
5531  */
5532 unsigned long __init node_map_pfn_alignment(void)
5533 {
5534         unsigned long accl_mask = 0, last_end = 0;
5535         unsigned long start, end, mask;
5536         int last_nid = -1;
5537         int i, nid;
5538 
5539         for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5540                 if (!start || last_nid < 0 || last_nid == nid) {
5541                         last_nid = nid;
5542                         last_end = end;
5543                         continue;
5544                 }
5545 
5546                 /*
5547                  * Start with a mask granular enough to pin-point to the
5548                  * start pfn and tick off bits one-by-one until it becomes
5549                  * too coarse to separate the current node from the last.
5550                  */
5551                 mask = ~((1 << __ffs(start)) - 1);
5552                 while (mask && last_end <= (start & (mask << 1)))
5553                         mask <<= 1;
5554 
5555                 /* accumulate all internode masks */
5556                 accl_mask |= mask;
5557         }
5558 
5559         /* convert mask to number of pages */
5560         return ~accl_mask + 1;
5561 }
5562 
5563 /* Find the lowest pfn for a node */
5564 static unsigned long __init find_min_pfn_for_node(int nid)
5565 {
5566         unsigned long min_pfn = ULONG_MAX;
5567         unsigned long start_pfn;
5568         int i;
5569 
5570         for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5571                 min_pfn = min(min_pfn, start_pfn);
5572 
5573         if (min_pfn == ULONG_MAX) {
5574                 printk(KERN_WARNING
5575                         "Could not find start_pfn for node %d\n", nid);
5576                 return 0;
5577         }
5578 
5579         return min_pfn;
5580 }
5581 
5582 /**
5583  * find_min_pfn_with_active_regions - Find the minimum PFN registered
5584  *
5585  * It returns the minimum PFN based on information provided via
5586  * memblock_set_node().
5587  */
5588 unsigned long __init find_min_pfn_with_active_regions(void)
5589 {
5590         return find_min_pfn_for_node(MAX_NUMNODES);
5591 }
5592 
5593 /*
5594  * early_calculate_totalpages()
5595  * Sum pages in active regions for movable zone.
5596  * Populate N_MEMORY for calculating usable_nodes.
5597  */
5598 static unsigned long __init early_calculate_totalpages(void)
5599 {
5600         unsigned long totalpages = 0;
5601         unsigned long start_pfn, end_pfn;
5602         int i, nid;
5603 
5604         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5605                 unsigned long pages = end_pfn - start_pfn;
5606 
5607                 totalpages += pages;
5608                 if (pages)
5609                         node_set_state(nid, N_MEMORY);
5610         }
5611         return totalpages;
5612 }
5613 
5614 /*
5615  * Find the PFN the Movable zone begins in each node. Kernel memory
5616  * is spread evenly between nodes as long as the nodes have enough
5617  * memory. When they don't, some nodes will have more kernelcore than
5618  * others
5619  */
5620 static void __init find_zone_movable_pfns_for_nodes(void)
5621 {
5622         int i, nid;
5623         unsigned long usable_startpfn;
5624         unsigned long kernelcore_node, kernelcore_remaining;
5625         /* save the state before borrow the nodemask */
5626         nodemask_t saved_node_state = node_states[N_MEMORY];
5627         unsigned long totalpages = early_calculate_totalpages();
5628         int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5629         struct memblock_region *r;
5630 
5631         /* Need to find movable_zone earlier when movable_node is specified. */
5632         find_usable_zone_for_movable();
5633 
5634         /*
5635          * If movable_node is specified, ignore kernelcore and movablecore
5636          * options.
5637          */
5638         if (movable_node_is_enabled()) {
5639                 for_each_memblock(memory, r) {
5640                         if (!memblock_is_hotpluggable(r))
5641                                 continue;
5642 
5643                         nid = r->nid;
5644 
5645                         usable_startpfn = PFN_DOWN(r->base);
5646                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5647                                 min(usable_startpfn, zone_movable_pfn[nid]) :
5648                                 usable_startpfn;
5649                 }
5650 
5651                 goto out2;
5652         }
5653 
5654         /*
5655          * If movablecore=nn[KMG] was specified, calculate what size of
5656          * kernelcore that corresponds so that memory usable for
5657          * any allocation type is evenly spread. If both kernelcore
5658          * and movablecore are specified, then the value of kernelcore
5659          * will be used for required_kernelcore if it's greater than
5660          * what movablecore would have allowed.
5661          */
5662         if (required_movablecore) {
5663                 unsigned long corepages;
5664 
5665                 /*
5666                  * Round-up so that ZONE_MOVABLE is at least as large as what
5667                  * was requested by the user
5668                  */
5669                 required_movablecore =
5670                         roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5671                 corepages = totalpages - required_movablecore;
5672 
5673                 required_kernelcore = max(required_kernelcore, corepages);
5674         }
5675 
5676         /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5677         if (!required_kernelcore)
5678                 goto out;
5679 
5680         /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5681         usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5682 
5683 restart:
5684         /* Spread kernelcore memory as evenly as possible throughout nodes */
5685         kernelcore_node = required_kernelcore / usable_nodes;
5686         for_each_node_state(nid, N_MEMORY) {
5687                 unsigned long start_pfn, end_pfn;
5688 
5689                 /*
5690                  * Recalculate kernelcore_node if the division per node
5691                  * now exceeds what is necessary to satisfy the requested
5692                  * amount of memory for the kernel
5693                  */
5694                 if (required_kernelcore < kernelcore_node)
5695                         kernelcore_node = required_kernelcore / usable_nodes;
5696 
5697                 /*
5698                  * As the map is walked, we track how much memory is usable
5699                  * by the kernel using kernelcore_remaining. When it is
5700                  * 0, the rest of the node is usable by ZONE_MOVABLE
5701                  */
5702                 kernelcore_remaining = kernelcore_node;
5703 
5704                 /* Go through each range of PFNs within this node */
5705                 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5706                         unsigned long size_pages;
5707 
5708                         start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5709                         if (start_pfn >= end_pfn)
5710                                 continue;
5711 
5712                         /* Account for what is only usable for kernelcore */
5713                         if (start_pfn < usable_startpfn) {
5714                                 unsigned long kernel_pages;
5715                                 kernel_pages = min(end_pfn, usable_startpfn)
5716                                                                 - start_pfn;
5717 
5718                                 kernelcore_remaining -= min(kernel_pages,
5719                                                         kernelcore_remaining);
5720                                 required_kernelcore -= min(kernel_pages,
5721                                                         required_kernelcore);
5722 
5723                                 /* Continue if range is now fully accounted */
5724                                 if (end_pfn <= usable_startpfn) {
5725 
5726                                         /*
5727                                          * Push zone_movable_pfn to the end so
5728                                          * that if we have to rebalance
5729                                          * kernelcore across nodes, we will
5730                                          * not double account here
5731                                          */
5732                                         zone_movable_pfn[nid] = end_pfn;
5733                                         continue;
5734                                 }
5735                                 start_pfn = usable_startpfn;
5736                         }
5737 
5738                         /*
5739                          * The usable PFN range for ZONE_MOVABLE is from
5740                          * start_pfn->end_pfn. Calculate size_pages as the
5741                          * number of pages used as kernelcore
5742                          */
5743                         size_pages = end_pfn - start_pfn;
5744                         if (size_pages > kernelcore_remaining)
5745                                 size_pages = kernelcore_remaining;
5746                         zone_movable_pfn[nid] = start_pfn + size_pages;
5747 
5748                         /*
5749                          * Some kernelcore has been met, update counts and
5750                          * break if the kernelcore for this node has been
5751                          * satisfied
5752                          */
5753                         required_kernelcore -= min(required_kernelcore,
5754                                                                 size_pages);
5755                         kernelcore_remaining -= size_pages;
5756                         if (!kernelcore_remaining)
5757                                 break;
5758                 }
5759         }
5760 
5761         /*
5762          * If there is still required_kernelcore, we do another pass with one
5763          * less node in the count. This will push zone_movable_pfn[nid] further
5764          * along on the nodes that still have memory until kernelcore is
5765          * satisfied
5766          */
5767         usable_nodes--;
5768         if (usable_nodes && required_kernelcore > usable_nodes)
5769                 goto restart;
5770 
5771 out2:
5772         /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5773         for (nid = 0; nid < MAX_NUMNODES; nid++)
5774                 zone_movable_pfn[nid] =
5775                         roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5776 
5777 out:
5778         /* restore the node_state */
5779         node_states[N_MEMORY] = saved_node_state;
5780 }
5781 
5782 /* Any regular or high memory on that node ? */
5783 static void check_for_memory(pg_data_t *pgdat, int nid)
5784 {
5785         enum zone_type zone_type;
5786 
5787         if (N_MEMORY == N_NORMAL_MEMORY)
5788                 return;
5789 
5790         for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5791                 struct zone *zone = &pgdat->node_zones[zone_type];
5792                 if (populated_zone(zone)) {
5793                         node_set_state(nid, N_HIGH_MEMORY);
5794                         if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5795                             zone_type <= ZONE_NORMAL)
5796                                 node_set_state(nid, N_NORMAL_MEMORY);
5797                         break;
5798                 }
5799         }
5800 }
5801 
5802 /**
5803  * free_area_init_nodes - Initialise all pg_data_t and zone data
5804  * @max_zone_pfn: an array of max PFNs for each zone
5805  *
5806  * This will call free_area_init_node() for each active node in the system.
5807  * Using the page ranges provided by memblock_set_node(), the size of each
5808  * zone in each node and their holes is calculated. If the maximum PFN
5809  * between two adjacent zones match, it is assumed that the zone is empty.
5810  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5811  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5812  * starts where the previous one ended. For example, ZONE_DMA32 starts
5813  * at arch_max_dma_pfn.
5814  */
5815 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5816 {
5817         unsigned long start_pfn, end_pfn;
5818         int i, nid;
5819 
5820         /* Record where the zone boundaries are */
5821         memset(arch_zone_lowest_possible_pfn, 0,
5822                                 sizeof(arch_zone_lowest_possible_pfn));
5823         memset(arch_zone_highest_possible_pfn, 0,
5824                                 sizeof(arch_zone_highest_possible_pfn));
5825         arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5826         arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5827         for (i = 1; i < MAX_NR_ZONES; i++) {
5828                 if (i == ZONE_MOVABLE)
5829                         continue;
5830                 arch_zone_lowest_possible_pfn[i] =
5831                         arch_zone_highest_possible_pfn[i-1];
5832                 arch_zone_highest_possible_pfn[i] =
5833                         max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5834         }
5835         arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5836         arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5837 
5838         /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5839         memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5840         find_zone_movable_pfns_for_nodes();
5841 
5842         /* Print out the zone ranges */
5843         pr_info("Zone ranges:\n");
5844         for (i = 0; i < MAX_NR_ZONES; i++) {
5845                 if (i == ZONE_MOVABLE)
5846                         continue;
5847                 pr_info("  %-8s ", zone_names[i]);
5848                 if (arch_zone_lowest_possible_pfn[i] ==
5849                                 arch_zone_highest_possible_pfn[i])
5850                         pr_cont("empty\n");
5851                 else
5852                         pr_cont("[mem %#018Lx-%#018Lx]\n",
5853                                 (u64)arch_zone_lowest_possible_pfn[i]
5854                                         << PAGE_SHIFT,
5855                                 ((u64)arch_zone_highest_possible_pfn[i]
5856                                         << PAGE_SHIFT) - 1);
5857         }
5858 
5859         /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5860         pr_info("Movable zone start for each node\n");
5861         for (i = 0; i < MAX_NUMNODES; i++) {
5862                 if (zone_movable_pfn[i])
5863                         pr_info("  Node %d: %#018Lx\n", i,
5864                                (u64)zone_movable_pfn[i] << PAGE_SHIFT);
5865         }
5866 
5867         /* Print out the early node map */
5868         pr_info("Early memory node ranges\n");
5869         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5870                 pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5871                         (u64)start_pfn << PAGE_SHIFT,
5872                         ((u64)end_pfn << PAGE_SHIFT) - 1);
5873 
5874         /* Initialise every node */
5875         mminit_verify_pageflags_layout();
5876         setup_nr_node_ids();
5877         for_each_online_node(nid) {
5878                 pg_data_t *pgdat = NODE_DATA(nid);
5879                 free_area_init_node(nid, NULL,
5880                                 find_min_pfn_for_node(nid), NULL);
5881 
5882                 /* Any memory on that node */
5883                 if (pgdat->node_present_pages)
5884                         node_set_state(nid, N_MEMORY);
5885                 check_for_memory(pgdat, nid);
5886         }
5887 }
5888 
5889 static int __init cmdline_parse_core(char *p, unsigned long *core)
5890 {
5891         unsigned long long coremem;
5892         if (!p)
5893                 return -EINVAL;
5894 
5895         coremem = memparse(p, &p);
5896         *core = coremem >> PAGE_SHIFT;
5897 
5898         /* Paranoid check that UL is enough for the coremem value */
5899         WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5900 
5901         return 0;
5902 }
5903 
5904 /*
5905  * kernelcore=size sets the amount of memory for use for allocations that
5906  * cannot be reclaimed or migrated.
5907  */
5908 static int __init cmdline_parse_kernelcore(char *p)
5909 {
5910         return cmdline_parse_core(p, &required_kernelcore);
5911 }
5912 
5913 /*
5914  * movablecore=size sets the amount of memory for use for allocations that
5915  * can be reclaimed or migrated.
5916  */
5917 static int __init cmdline_parse_movablecore(char *p)
5918 {
5919         return cmdline_parse_core(p, &required_movablecore);
5920 }
5921 
5922 early_param("kernelcore", cmdline_parse_kernelcore);
5923 early_param("movablecore", cmdline_parse_movablecore);
5924 
5925 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5926 
5927 void adjust_managed_page_count(struct page *page, long count)
5928 {
5929         spin_lock(&managed_page_count_lock);
5930         page_zone(page)->managed_pages += count;
5931         totalram_pages += count;
5932 #ifdef CONFIG_HIGHMEM
5933         if (PageHighMem(page))
5934                 totalhigh_pages += count;
5935 #endif
5936         spin_unlock(&managed_page_count_lock);
5937 }
5938 EXPORT_SYMBOL(adjust_managed_page_count);
5939 
5940 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
5941 {
5942         void *pos;
5943         unsigned long pages = 0;
5944 
5945         start = (void *)PAGE_ALIGN((unsigned long)start);
5946         end = (void *)((unsigned long)end & PAGE_MASK);
5947         for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
5948                 if ((unsigned int)poison <= 0xFF)
5949                         memset(pos, poison, PAGE_SIZE);
5950                 free_reserved_page(virt_to_page(pos));
5951         }
5952 
5953         if (pages && s)
5954                 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5955                         s, pages << (PAGE_SHIFT - 10), start, end);
5956 
5957         return pages;
5958 }
5959 EXPORT_SYMBOL(free_reserved_area);
5960 
5961 #ifdef  CONFIG_HIGHMEM
5962 void free_highmem_page(struct page *page)
5963 {
5964         __free_reserved_page(page);
5965         totalram_pages++;
5966         page_zone(page)->managed_pages++;
5967         totalhigh_pages++;
5968 }
5969 #endif
5970 
5971 
5972 void __init mem_init_print_info(const char *str)
5973 {
5974         unsigned long physpages, codesize, datasize, rosize, bss_size;
5975         unsigned long init_code_size, init_data_size;
5976 
5977         physpages = get_num_physpages();
5978         codesize = _etext - _stext;
5979         datasize = _edata - _sdata;
5980         rosize = __end_rodata - __start_rodata;
5981         bss_size = __bss_stop - __bss_start;
5982         init_data_size = __init_end - __init_begin;
5983         init_code_size = _einittext - _sinittext;
5984 
5985         /*
5986          * Detect special cases and adjust section sizes accordingly:
5987          * 1) .init.* may be embedded into .data sections
5988          * 2) .init.text.* may be out of [__init_begin, __init_end],
5989          *    please refer to arch/tile/kernel/vmlinux.lds.S.
5990          * 3) .rodata.* may be embedded into .text or .data sections.
5991          */
5992 #define adj_init_size(start, end, size, pos, adj) \
5993         do { \
5994                 if (start <= pos && pos < end && size > adj) \
5995                         size -= adj; \
5996         } while (0)
5997 
5998         adj_init_size(__init_begin, __init_end, init_data_size,
5999                      _sinittext, init_code_size);
6000         adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6001         adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6002         adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6003         adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6004 
6005 #undef  adj_init_size
6006 
6007         pr_info("Memory: %luK/%luK available "
6008                "(%luK kernel code, %luK rwdata, %luK rodata, "
6009                "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
6010 #ifdef  CONFIG_HIGHMEM
6011                ", %luK highmem"
6012 #endif
6013                "%s%s)\n",
6014                nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
6015                codesize >> 10, datasize >> 10, rosize >> 10,
6016                (init_data_size + init_code_size) >> 10, bss_size >> 10,
6017                (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
6018                totalcma_pages << (PAGE_SHIFT-10),
6019 #ifdef  CONFIG_HIGHMEM
6020                totalhigh_pages << (PAGE_SHIFT-10),
6021 #endif
6022                str ? ", " : "", str ? str : "");
6023 }
6024 
6025 /**
6026  * set_dma_reserve - set the specified number of pages reserved in the first zone
6027  * @new_dma_reserve: The number of pages to mark reserved
6028  *
6029  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6030  * In the DMA zone, a significant percentage may be consumed by kernel image
6031  * and other unfreeable allocations which can skew the watermarks badly. This
6032  * function may optionally be used to account for unfreeable pages in the
6033  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6034  * smaller per-cpu batchsize.
6035  */
6036 void __init set_dma_reserve(unsigned long new_dma_reserve)
6037 {
6038         dma_reserve = new_dma_reserve;
6039 }
6040 
6041 void __init free_area_init(unsigned long *zones_size)
6042 {
6043         free_area_init_node(0, zones_size,
6044                         __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6045 }
6046 
6047 static int page_alloc_cpu_notify(struct notifier_block *self,
6048                                  unsigned long action, void *hcpu)
6049 {
6050         int cpu = (unsigned long)hcpu;
6051 
6052         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
6053                 lru_add_drain_cpu(cpu);
6054                 drain_pages(cpu);
6055 
6056                 /*
6057                  * Spill the event counters of the dead processor
6058                  * into the current processors event counters.
6059                  * This artificially elevates the count of the current
6060                  * processor.
6061                  */
6062                 vm_events_fold_cpu(cpu);
6063 
6064                 /*
6065                  * Zero the differential counters of the dead processor
6066                  * so that the vm statistics are consistent.
6067                  *
6068                  * This is only okay since the processor is dead and cannot
6069                  * race with what we are doing.
6070                  */
6071                 cpu_vm_stats_fold(cpu);
6072         }
6073         return NOTIFY_OK;
6074 }
6075 
6076 void __init page_alloc_init(void)
6077 {
6078         hotcpu_notifier(page_alloc_cpu_notify, 0);
6079 }
6080 
6081 /*
6082  * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6083  *      or min_free_kbytes changes.
6084  */
6085 static void calculate_totalreserve_pages(void)
6086 {
6087         struct pglist_data *pgdat;
6088         unsigned long reserve_pages = 0;
6089         enum zone_type i, j;
6090 
6091         for_each_online_pgdat(pgdat) {
6092                 for (i = 0; i < MAX_NR_ZONES; i++) {
6093                         struct zone *zone = pgdat->node_zones + i;
6094                         long max = 0;
6095 
6096                         /* Find valid and maximum lowmem_reserve in the zone */
6097                         for (j = i; j < MAX_NR_ZONES; j++) {
6098                                 if (zone->lowmem_reserve[j] > max)
6099                                         max = zone->lowmem_reserve[j];
6100                         }
6101 
6102                         /* we treat the high watermark as reserved pages. */
6103                         max += high_wmark_pages(zone);
6104 
6105                         if (max > zone->managed_pages)
6106                                 max = zone->managed_pages;
6107                         reserve_pages += max;
6108                         /*
6109                          * Lowmem reserves are not available to
6110                          * GFP_HIGHUSER page cache allocations and
6111                          * kswapd tries to balance zones to their high
6112                          * watermark.  As a result, neither should be
6113                          * regarded as dirtyable memory, to prevent a
6114                          * situation where reclaim has to clean pages
6115                          * in order to balance the zones.
6116                          */
6117                         zone->dirty_balance_reserve = max;
6118                 }
6119         }
6120         dirty_balance_reserve = reserve_pages;
6121         totalreserve_pages = reserve_pages;
6122 }
6123 
6124 /*
6125  * setup_per_zone_lowmem_reserve - called whenever
6126  *      sysctl_lowmem_reserve_ratio changes.  Ensures that each zone
6127  *      has a correct pages reserved value, so an adequate number of
6128  *      pages are left in the zone after a successful __alloc_pages().
6129  */
6130 static void setup_per_zone_lowmem_reserve(void)
6131 {
6132         struct pglist_data *pgdat;
6133         enum zone_type j, idx;
6134 
6135         for_each_online_pgdat(pgdat) {
6136                 for (j = 0; j < MAX_NR_ZONES; j++) {
6137                         struct zone *zone = pgdat->node_zones + j;
6138                         unsigned long managed_pages = zone->managed_pages;
6139 
6140                         zone->lowmem_reserve[j] = 0;
6141 
6142                         idx = j;
6143                         while (idx) {
6144                                 struct zone *lower_zone;
6145 
6146                                 idx--;
6147 
6148                                 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6149                                         sysctl_lowmem_reserve_ratio[idx] = 1;
6150 
6151                                 lower_zone = pgdat->node_zones + idx;
6152                                 lower_zone->lowmem_reserve[j] = managed_pages /
6153                                         sysctl_lowmem_reserve_ratio[idx];
6154                                 managed_pages += lower_zone->managed_pages;
6155                         }
6156                 }
6157         }
6158 
6159         /* update totalreserve_pages */
6160         calculate_totalreserve_pages();
6161 }
6162 
6163 static void __setup_per_zone_wmarks(void)
6164 {
6165         unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6166         unsigned long lowmem_pages = 0;
6167         struct zone *zone;
6168         unsigned long flags;
6169 
6170         /* Calculate total number of !ZONE_HIGHMEM pages */
6171         for_each_zone(zone) {
6172                 if (!is_highmem(zone))
6173                         lowmem_pages += zone->managed_pages;
6174         }
6175 
6176         for_each_zone(zone) {
6177                 u64 tmp;
6178 
6179                 spin_lock_irqsave(&zone->lock, flags);
6180                 tmp = (u64)pages_min * zone->managed_pages;
6181                 do_div(tmp, lowmem_pages);
6182                 if (is_highmem(zone)) {
6183                         /*
6184                          * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6185                          * need highmem pages, so cap pages_min to a small
6186                          * value here.
6187                          *
6188                          * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6189                          * deltas control asynch page reclaim, and so should
6190                          * not be capped for highmem.
6191                          */
6192                         unsigned long min_pages;
6193 
6194                         min_pages = zone->managed_pages / 1024;
6195                         min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
6196                         zone->watermark[WMARK_MIN] = min_pages;
6197                 } else {
6198                         /*
6199                          * If it's a lowmem zone, reserve a number of pages
6200                          * proportionate to the zone's size.
6201                          */
6202                         zone->watermark[WMARK_MIN] = tmp;
6203                 }
6204 
6205                 zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + (tmp >> 2);
6206                 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
6207 
6208                 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
6209                         high_wmark_pages(zone) - low_wmark_pages(zone) -
6210                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
6211 
6212                 setup_zone_migrate_reserve(zone);
6213                 spin_unlock_irqrestore(&zone->lock, flags);
6214         }
6215 
6216         /* update totalreserve_pages */
6217         calculate_totalreserve_pages();
6218 }
6219 
6220 /**
6221  * setup_per_zone_wmarks - called when min_free_kbytes changes
6222  * or when memory is hot-{added|removed}
6223  *
6224  * Ensures that the watermark[min,low,high] values for each zone are set
6225  * correctly with respect to min_free_kbytes.
6226  */
6227 void setup_per_zone_wmarks(void)
6228 {
6229         mutex_lock(&zonelists_mutex);
6230         __setup_per_zone_wmarks();
6231         mutex_unlock(&zonelists_mutex);
6232 }
6233 
6234 /*
6235  * The inactive anon list should be small enough that the VM never has to
6236  * do too much work, but large enough that each inactive page has a chance
6237  * to be referenced again before it is swapped out.
6238  *
6239  * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6240  * INACTIVE_ANON pages on this zone's LRU, maintained by the
6241  * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6242  * the anonymous pages are kept on the inactive list.
6243  *
6244  * total     target    max
6245  * memory    ratio     inactive anon
6246  * -------------------------------------
6247  *   10MB       1         5MB
6248  *  100MB       1        50MB
6249  *    1GB       3       250MB
6250  *   10GB      10       0.9GB
6251  *  100GB      31         3GB
6252  *    1TB     101        10GB
6253  *   10TB     320        32GB
6254  */
6255 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
6256 {
6257         unsigned int gb, ratio;
6258 
6259         /* Zone size in gigabytes */
6260         gb = zone->managed_pages >> (30 - PAGE_SHIFT);
6261         if (gb)
6262                 ratio = int_sqrt(10 * gb);
6263         else
6264                 ratio = 1;
6265 
6266         zone->inactive_ratio = ratio;
6267 }
6268 
6269 static void __meminit setup_per_zone_inactive_ratio(void)
6270 {
6271         struct zone *zone;
6272 
6273         for_each_zone(zone)
6274                 calculate_zone_inactive_ratio(zone);
6275 }
6276 
6277 /*
6278  * Initialise min_free_kbytes.
6279  *
6280  * For small machines we want it small (128k min).  For large machines
6281  * we want it large (64MB max).  But it is not linear, because network
6282  * bandwidth does not increase linearly with machine size.  We use
6283  *
6284  *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6285  *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
6286  *
6287  * which yields
6288  *
6289  * 16MB:        512k
6290  * 32MB:        724k
6291  * 64MB:        1024k
6292  * 128MB:       1448k
6293  * 256MB:       2048k
6294  * 512MB:       2896k
6295  * 1024MB:      4096k
6296  * 2048MB:      5792k
6297  * 4096MB:      8192k
6298  * 8192MB:      11584k
6299  * 16384MB:     16384k
6300  */
6301 int __meminit init_per_zone_wmark_min(void)
6302 {
6303         unsigned long lowmem_kbytes;
6304         int new_min_free_kbytes;
6305 
6306         lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
6307         new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
6308 
6309         if (new_min_free_kbytes > user_min_free_kbytes) {
6310                 min_free_kbytes = new_min_free_kbytes;
6311                 if (min_free_kbytes < 128)
6312                         min_free_kbytes = 128;
6313                 if (min_free_kbytes > 65536)
6314                         min_free_kbytes = 65536;
6315         } else {
6316                 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6317                                 new_min_free_kbytes, user_min_free_kbytes);
6318         }
6319         setup_per_zone_wmarks();
6320         refresh_zone_stat_thresholds();
6321         setup_per_zone_lowmem_reserve();
6322         setup_per_zone_inactive_ratio();
6323         return 0;
6324 }
6325 module_init(init_per_zone_wmark_min)
6326 
6327 /*
6328  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6329  *      that we can call two helper functions whenever min_free_kbytes
6330  *      changes.
6331  */
6332 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
6333         void __user *buffer, size_t *length, loff_t *ppos)
6334 {
6335         int rc;
6336 
6337         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6338         if (rc)
6339                 return rc;
6340 
6341         if (write) {
6342                 user_min_free_kbytes = min_free_kbytes;
6343                 setup_per_zone_wmarks();
6344         }
6345         return 0;
6346 }
6347 
6348 #ifdef CONFIG_NUMA
6349 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
6350         void __user *buffer, size_t *length, loff_t *ppos)
6351 {
6352         struct zone *zone;
6353         int rc;
6354 
6355         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6356         if (rc)
6357                 return rc;
6358 
6359         for_each_zone(zone)
6360                 zone->min_unmapped_pages = (zone->managed_pages *
6361                                 sysctl_min_unmapped_ratio) / 100;
6362         return 0;
6363 }
6364 
6365 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
6366         void __user *buffer, size_t *length, loff_t *ppos)
6367 {
6368         struct zone *zone;
6369         int rc;
6370 
6371         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6372         if (rc)
6373                 return rc;
6374 
6375         for_each_zone(zone)
6376                 zone->min_slab_pages = (zone->managed_pages *
6377                                 sysctl_min_slab_ratio) / 100;
6378         return 0;
6379 }
6380 #endif
6381 
6382 /*
6383  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6384  *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6385  *      whenever sysctl_lowmem_reserve_ratio changes.
6386  *
6387  * The reserve ratio obviously has absolutely no relation with the
6388  * minimum watermarks. The lowmem reserve ratio can only make sense
6389  * if in function of the boot time zone sizes.
6390  */
6391 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
6392         void __user *buffer, size_t *length, loff_t *ppos)
6393 {
6394         proc_dointvec_minmax(table, write, buffer, length, ppos);
6395         setup_per_zone_lowmem_reserve();
6396         return 0;
6397 }
6398 
6399 /*
6400  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6401  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
6402  * pagelist can have before it gets flushed back to buddy allocator.
6403  */
6404 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
6405         void __user *buffer, size_t *length, loff_t *ppos)
6406 {
6407         struct zone *zone;
6408         int old_percpu_pagelist_fraction;
6409         int ret;
6410 
6411         mutex_lock(&pcp_batch_high_lock);
6412         old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6413 
6414         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6415         if (!write || ret < 0)
6416                 goto out;
6417 
6418         /* Sanity checking to avoid pcp imbalance */
6419         if (percpu_pagelist_fraction &&
6420             percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6421                 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6422                 ret = -EINVAL;
6423                 goto out;
6424         }
6425 
6426         /* No change? */
6427         if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6428                 goto out;
6429 
6430         for_each_populated_zone(zone) {
6431                 unsigned int cpu;
6432 
6433                 for_each_possible_cpu(cpu)
6434                         pageset_set_high_and_batch(zone,
6435                                         per_cpu_ptr(zone->pageset, cpu));
6436         }
6437 out:
6438         mutex_unlock(&pcp_batch_high_lock);
6439         return ret;
6440 }
6441 
6442 #ifdef CONFIG_NUMA
6443 int hashdist = HASHDIST_DEFAULT;
6444 
6445 static int __init set_hashdist(char *str)
6446 {
6447         if (!str)
6448                 return 0;
6449         hashdist = simple_strtoul(str, &str, 0);
6450         return 1;
6451 }
6452 __setup("hashdist=", set_hashdist);
6453 #endif
6454 
6455 /*
6456  * allocate a large system hash table from bootmem
6457  * - it is assumed that the hash table must contain an exact power-of-2
6458  *   quantity of entries
6459  * - limit is the number of hash buckets, not the total allocation size
6460  */
6461 void *__init alloc_large_system_hash(const char *tablename,
6462                                      unsigned long bucketsize,
6463                                      unsigned long numentries,
6464                                      int scale,
6465                                      int flags,
6466                                      unsigned int *_hash_shift,
6467                                      unsigned int *_hash_mask,
6468                                      unsigned long low_limit,
6469                                      unsigned long high_limit)
6470 {
6471         unsigned long long max = high_limit;
6472         unsigned long log2qty, size;
6473         void *table = NULL;
6474 
6475         /* allow the kernel cmdline to have a say */
6476         if (!numentries) {
6477                 /* round applicable memory size up to nearest megabyte */
6478                 numentries = nr_kernel_pages;
6479 
6480                 /* It isn't necessary when PAGE_SIZE >= 1MB */
6481                 if (PAGE_SHIFT < 20)
6482                         numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
6483 
6484                 /* limit to 1 bucket per 2^scale bytes of low memory */
6485                 if (scale > PAGE_SHIFT)
6486                         numentries >>= (scale - PAGE_SHIFT);
6487                 else
6488                         numentries <<= (PAGE_SHIFT - scale);
6489 
6490                 /* Make sure we've got at least a 0-order allocation.. */
6491                 if (unlikely(flags & HASH_SMALL)) {
6492                         /* Makes no sense without HASH_EARLY */
6493                         WARN_ON(!(flags & HASH_EARLY));
6494                         if (!(numentries >> *_hash_shift)) {
6495                                 numentries = 1UL << *_hash_shift;
6496                                 BUG_ON(!numentries);
6497                         }
6498                 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
6499                         numentries = PAGE_SIZE / bucketsize;
6500         }
6501         numentries = roundup_pow_of_two(numentries);
6502 
6503         /* limit allocation size to 1/16 total memory by default */
6504         if (max == 0) {
6505                 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6506                 do_div(max, bucketsize);
6507         }
6508         max = min(max, 0x80000000ULL);
6509 
6510         if (numentries < low_limit)
6511                 numentries = low_limit;
6512         if (numentries > max)
6513                 numentries = max;
6514 
6515         log2qty = ilog2(numentries);
6516 
6517         do {
6518                 size = bucketsize << log2qty;
6519                 if (flags & HASH_EARLY)
6520                         table = memblock_virt_alloc_nopanic(size, 0);
6521                 else if (hashdist)
6522                         table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6523                 else {
6524                         /*
6525                          * If bucketsize is not a power-of-two, we may free
6526                          * some pages at the end of hash table which
6527                          * alloc_pages_exact() automatically does
6528                          */
6529                         if (get_order(size) < MAX_ORDER) {
6530                                 table = alloc_pages_exact(size, GFP_ATOMIC);
6531                                 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6532                         }
6533                 }
6534         } while (!table && size > PAGE_SIZE && --log2qty);
6535 
6536         if (!table)
6537                 panic("Failed to allocate %s hash table\n", tablename);
6538 
6539         printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
6540                tablename,
6541                (1UL << log2qty),
6542                ilog2(size) - PAGE_SHIFT,
6543                size);
6544 
6545         if (_hash_shift)
6546                 *_hash_shift = log2qty;
6547         if (_hash_mask)
6548                 *_hash_mask = (1 << log2qty) - 1;
6549 
6550         return table;
6551 }
6552 
6553 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6554 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6555                                                         unsigned long pfn)
6556 {
6557 #ifdef CONFIG_SPARSEMEM
6558         return __pfn_to_section(pfn)->pageblock_flags;
6559 #else
6560         return zone->pageblock_flags;
6561 #endif /* CONFIG_SPARSEMEM */
6562 }
6563 
6564 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6565 {
6566 #ifdef CONFIG_SPARSEMEM
6567         pfn &= (PAGES_PER_SECTION-1);
6568         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6569 #else
6570         pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
6571         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6572 #endif /* CONFIG_SPARSEMEM */
6573 }
6574 
6575 /**
6576  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6577  * @page: The page within the block of interest
6578  * @pfn: The target page frame number
6579  * @end_bitidx: The last bit of interest to retrieve
6580  * @mask: mask of bits that the caller is interested in
6581  *
6582  * Return: pageblock_bits flags
6583  */
6584 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
6585                                         unsigned long end_bitidx,
6586                                         unsigned long mask)
6587 {
6588         struct zone *zone;
6589         unsigned long *bitmap;
6590         unsigned long bitidx, word_bitidx;
6591         unsigned long word;
6592 
6593         zone = page_zone(page);
6594         bitmap = get_pageblock_bitmap(zone, pfn);
6595         bitidx = pfn_to_bitidx(zone, pfn);
6596         word_bitidx = bitidx / BITS_PER_LONG;
6597         bitidx &= (BITS_PER_LONG-1);
6598 
6599         word = bitmap[word_bitidx];
6600         bitidx += end_bitidx;
6601         return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
6602 }
6603 
6604 /**
6605  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6606  * @page: The page within the block of interest
6607  * @flags: The flags to set
6608  * @pfn: The target page frame number
6609  * @end_bitidx: The last bit of interest
6610  * @mask: mask of bits that the caller is interested in
6611  */
6612 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6613                                         unsigned long pfn,
6614                                         unsigned long end_bitidx,
6615                                         unsigned long mask)
6616 {
6617         struct zone *zone;
6618         unsigned long *bitmap;
6619         unsigned long bitidx, word_bitidx;
6620         unsigned long old_word, word;
6621 
6622         BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
6623 
6624         zone = page_zone(page);
6625         bitmap = get_pageblock_bitmap(zone, pfn);
6626         bitidx = pfn_to_bitidx(zone, pfn);
6627         word_bitidx = bitidx / BITS_PER_LONG;
6628         bitidx &= (BITS_PER_LONG-1);
6629 
6630         VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
6631 
6632         bitidx += end_bitidx;
6633         mask <<= (BITS_PER_LONG - bitidx - 1);
6634         flags <<= (BITS_PER_LONG - bitidx - 1);
6635 
6636         word = READ_ONCE(bitmap[word_bitidx]);
6637         for (;;) {
6638                 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6639                 if (word == old_word)
6640                         break;
6641                 word = old_word;
6642         }
6643 }
6644 
6645 /*
6646  * This function checks whether pageblock includes unmovable pages or not.
6647  * If @count is not zero, it is okay to include less @count unmovable pages
6648  *
6649  * PageLRU check without isolation or lru_lock could race so that
6650  * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6651  * expect this function should be exact.
6652  */
6653 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6654                          bool skip_hwpoisoned_pages)
6655 {
6656         unsigned long pfn, iter, found;
6657         int mt;
6658 
6659         /*
6660          * For avoiding noise data, lru_add_drain_all() should be called
6661          * If ZONE_MOVABLE, the zone never contains unmovable pages
6662          */
6663         if (zone_idx(zone) == ZONE_MOVABLE)
6664                 return false;
6665         mt = get_pageblock_migratetype(page);
6666         if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6667                 return false;
6668 
6669         pfn = page_to_pfn(page);
6670         for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6671                 unsigned long check = pfn + iter;
6672 
6673                 if (!pfn_valid_within(check))
6674                         continue;
6675 
6676                 page = pfn_to_page(check);
6677 
6678                 /*
6679                  * Hugepages are not in LRU lists, but they're movable.
6680                  * We need not scan over tail pages bacause we don't
6681                  * handle each tail page individually in migration.
6682                  */
6683                 if (PageHuge(page)) {
6684                         iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6685                         continue;
6686                 }
6687 
6688                 /*
6689                  * We can't use page_count without pin a page
6690                  * because another CPU can free compound page.
6691                  * This check already skips compound tails of THP
6692                  * because their page->_count is zero at all time.
6693                  */
6694                 if (!atomic_read(&page->_count)) {
6695                         if (PageBuddy(page))
6696                                 iter += (1 << page_order(page)) - 1;
6697                         continue;
6698                 }
6699 
6700                 /*
6701                  * The HWPoisoned page may be not in buddy system, and
6702                  * page_count() is not 0.
6703                  */
6704                 if (skip_hwpoisoned_pages && PageHWPoison(page))
6705                         continue;
6706 
6707                 if (!PageLRU(page))
6708                         found++;
6709                 /*
6710                  * If there are RECLAIMABLE pages, we need to check
6711                  * it.  But now, memory offline itself doesn't call
6712                  * shrink_node_slabs() and it still to be fixed.
6713                  */
6714                 /*
6715                  * If the page is not RAM, page_count()should be 0.
6716                  * we don't need more check. This is an _used_ not-movable page.
6717                  *
6718                  * The problematic thing here is PG_reserved pages. PG_reserved
6719                  * is set to both of a memory hole page and a _used_ kernel
6720                  * page at boot.
6721                  */
6722                 if (found > count)
6723                         return true;
6724         }
6725         return false;
6726 }
6727 
6728 bool is_pageblock_removable_nolock(struct page *page)
6729 {
6730         struct zone *zone;
6731         unsigned long pfn;
6732 
6733         /*
6734          * We have to be careful here because we are iterating over memory
6735          * sections which are not zone aware so we might end up outside of
6736          * the zone but still within the section.
6737          * We have to take care about the node as well. If the node is offline
6738          * its NODE_DATA will be NULL - see page_zone.
6739          */
6740         if (!node_online(page_to_nid(page)))
6741                 return false;
6742 
6743         zone = page_zone(page);
6744         pfn = page_to_pfn(page);
6745         if (!zone_spans_pfn(zone, pfn))
6746                 return false;
6747 
6748         return !has_unmovable_pages(zone, page, 0, true);
6749 }
6750 
6751 #ifdef CONFIG_CMA
6752 
6753 static unsigned long pfn_max_align_down(unsigned long pfn)
6754 {
6755         return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6756                              pageblock_nr_pages) - 1);
6757 }
6758 
6759 static unsigned long pfn_max_align_up(unsigned long pfn)
6760 {
6761         return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6762                                 pageblock_nr_pages));
6763 }
6764 
6765 /* [start, end) must belong to a single zone. */
6766 static int __alloc_contig_migrate_range(struct compact_control *cc,
6767                                         unsigned long start, unsigned long end)
6768 {
6769         /* This function is based on compact_zone() from compaction.c. */
6770         unsigned long nr_reclaimed;
6771         unsigned long pfn = start;
6772         unsigned int tries = 0;
6773         int ret = 0;
6774 
6775         migrate_prep();
6776 
6777         while (pfn < end || !list_empty(&cc->migratepages)) {
6778                 if (fatal_signal_pending(current)) {
6779                         ret = -EINTR;
6780                         break;
6781                 }
6782 
6783                 if (list_empty(&cc->migratepages)) {
6784                         cc->nr_migratepages = 0;
6785                         pfn = isolate_migratepages_range(cc, pfn, end);
6786                         if (!pfn) {
6787                                 ret = -EINTR;
6788                                 break;
6789                         }
6790                         tries = 0;
6791                 } else if (++tries == 5) {
6792                         ret = ret < 0 ? ret : -EBUSY;
6793                         break;
6794                 }
6795 
6796                 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6797                                                         &cc->migratepages);
6798                 cc->nr_migratepages -= nr_reclaimed;
6799 
6800                 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6801                                     NULL, 0, cc->mode, MR_CMA);
6802         }
6803         if (ret < 0) {
6804                 putback_movable_pages(&cc->migratepages);
6805                 return ret;
6806         }
6807         return 0;
6808 }
6809 
6810 /**
6811  * alloc_contig_range() -- tries to allocate given range of pages
6812  * @start:      start PFN to allocate
6813  * @end:        one-past-the-last PFN to allocate
6814  * @migratetype:        migratetype of the underlaying pageblocks (either
6815  *                      #MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
6816  *                      in range must have the same migratetype and it must
6817  *                      be either of the two.
6818  *
6819  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6820  * aligned, however it's the caller's responsibility to guarantee that
6821  * we are the only thread that changes migrate type of pageblocks the
6822  * pages fall in.
6823  *
6824  * The PFN range must belong to a single zone.
6825  *
6826  * Returns zero on success or negative error code.  On success all
6827  * pages which PFN is in [start, end) are allocated for the caller and
6828  * need to be freed with free_contig_range().
6829  */
6830 int alloc_contig_range(unsigned long start, unsigned long end,
6831                        unsigned migratetype)
6832 {
6833         unsigned long outer_start, outer_end;
6834         int ret = 0, order;
6835 
6836         struct compact_control cc = {
6837                 .nr_migratepages = 0,
6838                 .order = -1,
6839                 .zone = page_zone(pfn_to_page(start)),
6840                 .mode = MIGRATE_SYNC,
6841                 .ignore_skip_hint = true,
6842         };
6843         INIT_LIST_HEAD(&cc.migratepages);
6844 
6845         /*
6846          * What we do here is we mark all pageblocks in range as
6847          * MIGRATE_ISOLATE.  Because pageblock and max order pages may
6848          * have different sizes, and due to the way page allocator
6849          * work, we align the range to biggest of the two pages so
6850          * that page allocator won't try to merge buddies from
6851          * different pageblocks and change MIGRATE_ISOLATE to some
6852          * other migration type.
6853          *
6854          * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6855          * migrate the pages from an unaligned range (ie. pages that
6856          * we are interested in).  This will put all the pages in
6857          * range back to page allocator as MIGRATE_ISOLATE.
6858          *
6859          * When this is done, we take the pages in range from page
6860          * allocator removing them from the buddy system.  This way
6861          * page allocator will never consider using them.
6862          *
6863          * This lets us mark the pageblocks back as
6864          * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6865          * aligned range but not in the unaligned, original range are
6866          * put back to page allocator so that buddy can use them.
6867          */
6868 
6869         ret = start_isolate_page_range(pfn_max_align_down(start),
6870                                        pfn_max_align_up(end), migratetype,
6871                                        false);
6872         if (ret)
6873                 return ret;
6874 
6875         ret = __alloc_contig_migrate_range(&cc, start, end);
6876         if (ret)
6877                 goto done;
6878 
6879         /*
6880          * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6881          * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
6882          * more, all pages in [start, end) are free in page allocator.
6883          * What we are going to do is to allocate all pages from
6884          * [start, end) (that is remove them from page allocator).
6885          *
6886          * The only problem is that pages at the beginning and at the
6887          * end of interesting range may be not aligned with pages that
6888          * page allocator holds, ie. they can be part of higher order
6889          * pages.  Because of this, we reserve the bigger range and
6890          * once this is done free the pages we are not interested in.
6891          *
6892          * We don't have to hold zone->lock here because the pages are
6893          * isolated thus they won't get removed from buddy.
6894          */
6895 
6896         lru_add_drain_all();
6897         drain_all_pages(cc.zone);
6898 
6899         order = 0;
6900         outer_start = start;
6901         while (!PageBuddy(pfn_to_page(outer_start))) {
6902                 if (++order >= MAX_ORDER) {
6903                         ret = -EBUSY;
6904                         goto done;
6905                 }
6906                 outer_start &= ~0UL << order;
6907         }
6908 
6909         /* Make sure the range is really isolated. */
6910         if (test_pages_isolated(outer_start, end, false)) {
6911                 pr_info("%s: [%lx, %lx) PFNs busy\n",
6912                         __func__, outer_start, end);
6913                 ret = -EBUSY;
6914                 goto done;
6915         }
6916 
6917         /* Grab isolated pages from freelists. */
6918         outer_end = isolate_freepages_range(&cc, outer_start, end);
6919         if (!outer_end) {
6920                 ret = -EBUSY;
6921                 goto done;
6922         }
6923 
6924         /* Free head and tail (if any) */
6925         if (start != outer_start)
6926                 free_contig_range(outer_start, start - outer_start);
6927         if (end != outer_end)
6928                 free_contig_range(end, outer_end - end);
6929 
6930 done:
6931         undo_isolate_page_range(pfn_max_align_down(start),
6932                                 pfn_max_align_up(end), migratetype);
6933         return ret;
6934 }
6935 
6936 void free_contig_range(unsigned long pfn, unsigned nr_pages)
6937 {
6938         unsigned int count = 0;
6939 
6940         for (; nr_pages--; pfn++) {
6941                 struct page *page = pfn_to_page(pfn);
6942 
6943                 count += page_count(page) != 1;
6944                 __free_page(page);
6945         }
6946         WARN(count != 0, "%d pages are still in use!\n", count);
6947 }
6948 #endif
6949 
6950 #ifdef CONFIG_MEMORY_HOTPLUG
6951 /*
6952  * The zone indicated has a new number of managed_pages; batch sizes and percpu
6953  * page high values need to be recalulated.
6954  */
6955 void __meminit zone_pcp_update(struct zone *zone)
6956 {
6957         unsigned cpu;
6958         mutex_lock(&pcp_batch_high_lock);
6959         for_each_possible_cpu(cpu)
6960                 pageset_set_high_and_batch(zone,
6961                                 per_cpu_ptr(zone->pageset, cpu));
6962         mutex_unlock(&pcp_batch_high_lock);
6963 }
6964 #endif
6965 
6966 void zone_pcp_reset(struct zone *zone)
6967 {
6968         unsigned long flags;
6969         int cpu;
6970         struct per_cpu_pageset *pset;
6971 
6972         /* avoid races with drain_pages()  */
6973         local_irq_save(flags);
6974         if (zone->pageset != &boot_pageset) {
6975                 for_each_online_cpu(cpu) {
6976                         pset = per_cpu_ptr(zone->pageset, cpu);
6977                         drain_zonestat(zone, pset);
6978                 }
6979                 free_percpu(zone->pageset);
6980                 zone->pageset = &boot_pageset;
6981         }
6982         local_irq_restore(flags);
6983 }
6984 
6985 #ifdef CONFIG_MEMORY_HOTREMOVE
6986 /*
6987  * All pages in the range must be isolated before calling this.
6988  */
6989 void
6990 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6991 {
6992         struct page *page;
6993         struct zone *zone;
6994         unsigned int order, i;
6995         unsigned long pfn;
6996         unsigned long flags;
6997         /* find the first valid pfn */
6998         for (pfn = start_pfn; pfn < end_pfn; pfn++)
6999                 if (pfn_valid(pfn))
7000                         break;
7001         if (pfn == end_pfn)
7002                 return;
7003         zone = page_zone(pfn_to_page(pfn));
7004         spin_lock_irqsave(&zone->lock, flags);
7005         pfn = start_pfn;
7006         while (pfn < end_pfn) {
7007                 if (!pfn_valid(pfn)) {
7008                         pfn++;
7009                         continue;
7010                 }
7011                 page = pfn_to_page(pfn);
7012                 /*
7013                  * The HWPoisoned page may be not in buddy system, and
7014                  * page_count() is not 0.
7015                  */
7016                 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7017                         pfn++;
7018                         SetPageReserved(page);
7019                         continue;
7020                 }
7021 
7022                 BUG_ON(page_count(page));
7023                 BUG_ON(!PageBuddy(page));
7024                 order = page_order(page);
7025 #ifdef CONFIG_DEBUG_VM
7026                 printk(KERN_INFO "remove from free list %lx %d %lx\n",
7027                        pfn, 1 << order, end_pfn);
7028 #endif
7029                 list_del(&page->lru);
7030                 rmv_page_order(page);
7031                 zone->free_area[order].nr_free--;
7032                 for (i = 0; i < (1 << order); i++)
7033                         SetPageReserved((page+i));
7034                 pfn += (1 << order);
7035         }
7036         spin_unlock_irqrestore(&zone->lock, flags);
7037 }
7038 #endif
7039 
7040 #ifdef CONFIG_MEMORY_FAILURE
7041 bool is_free_buddy_page(struct page *page)
7042 {
7043         struct zone *zone = page_zone(page);
7044         unsigned long pfn = page_to_pfn(page);
7045         unsigned long flags;
7046         unsigned int order;
7047 
7048         spin_lock_irqsave(&zone->lock, flags);
7049         for (order = 0; order < MAX_ORDER; order++) {
7050                 struct page *page_head = page - (pfn & ((1 << order) - 1));
7051 
7052                 if (PageBuddy(page_head) && page_order(page_head) >= order)
7053                         break;
7054         }
7055         spin_unlock_irqrestore(&zone->lock, flags);
7056 
7057         return order < MAX_ORDER;
7058 }
7059 #endif
7060 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp