Version:  2.0.40 2.2.26 2.4.37 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4

Linux/mm/page_alloc.c

  1 /*
  2  *  linux/mm/page_alloc.c
  3  *
  4  *  Manages the free list, the system allocates free pages here.
  5  *  Note that kmalloc() lives in slab.c
  6  *
  7  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  8  *  Swap reorganised 29.12.95, Stephen Tweedie
  9  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
 10  *  Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
 11  *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
 12  *  Zone balancing, Kanoj Sarcar, SGI, Jan 2000
 13  *  Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
 14  *          (lots of bits borrowed from Ingo Molnar & Andrew Morton)
 15  */
 16 
 17 #include <linux/stddef.h>
 18 #include <linux/mm.h>
 19 #include <linux/swap.h>
 20 #include <linux/interrupt.h>
 21 #include <linux/pagemap.h>
 22 #include <linux/jiffies.h>
 23 #include <linux/bootmem.h>
 24 #include <linux/memblock.h>
 25 #include <linux/compiler.h>
 26 #include <linux/kernel.h>
 27 #include <linux/kmemcheck.h>
 28 #include <linux/kasan.h>
 29 #include <linux/module.h>
 30 #include <linux/suspend.h>
 31 #include <linux/pagevec.h>
 32 #include <linux/blkdev.h>
 33 #include <linux/slab.h>
 34 #include <linux/ratelimit.h>
 35 #include <linux/oom.h>
 36 #include <linux/notifier.h>
 37 #include <linux/topology.h>
 38 #include <linux/sysctl.h>
 39 #include <linux/cpu.h>
 40 #include <linux/cpuset.h>
 41 #include <linux/memory_hotplug.h>
 42 #include <linux/nodemask.h>
 43 #include <linux/vmalloc.h>
 44 #include <linux/vmstat.h>
 45 #include <linux/mempolicy.h>
 46 #include <linux/stop_machine.h>
 47 #include <linux/sort.h>
 48 #include <linux/pfn.h>
 49 #include <linux/backing-dev.h>
 50 #include <linux/fault-inject.h>
 51 #include <linux/page-isolation.h>
 52 #include <linux/page_ext.h>
 53 #include <linux/debugobjects.h>
 54 #include <linux/kmemleak.h>
 55 #include <linux/compaction.h>
 56 #include <trace/events/kmem.h>
 57 #include <linux/prefetch.h>
 58 #include <linux/mm_inline.h>
 59 #include <linux/migrate.h>
 60 #include <linux/page_ext.h>
 61 #include <linux/hugetlb.h>
 62 #include <linux/sched/rt.h>
 63 #include <linux/page_owner.h>
 64 #include <linux/kthread.h>
 65 
 66 #include <asm/sections.h>
 67 #include <asm/tlbflush.h>
 68 #include <asm/div64.h>
 69 #include "internal.h"
 70 
 71 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
 72 static DEFINE_MUTEX(pcp_batch_high_lock);
 73 #define MIN_PERCPU_PAGELIST_FRACTION    (8)
 74 
 75 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
 76 DEFINE_PER_CPU(int, numa_node);
 77 EXPORT_PER_CPU_SYMBOL(numa_node);
 78 #endif
 79 
 80 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
 81 /*
 82  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
 83  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
 84  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
 85  * defined in <linux/topology.h>.
 86  */
 87 DEFINE_PER_CPU(int, _numa_mem_);                /* Kernel "local memory" node */
 88 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
 89 int _node_numa_mem_[MAX_NUMNODES];
 90 #endif
 91 
 92 /*
 93  * Array of node states.
 94  */
 95 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
 96         [N_POSSIBLE] = NODE_MASK_ALL,
 97         [N_ONLINE] = { { [0] = 1UL } },
 98 #ifndef CONFIG_NUMA
 99         [N_NORMAL_MEMORY] = { { [0] = 1UL } },
100 #ifdef CONFIG_HIGHMEM
101         [N_HIGH_MEMORY] = { { [0] = 1UL } },
102 #endif
103 #ifdef CONFIG_MOVABLE_NODE
104         [N_MEMORY] = { { [0] = 1UL } },
105 #endif
106         [N_CPU] = { { [0] = 1UL } },
107 #endif  /* NUMA */
108 };
109 EXPORT_SYMBOL(node_states);
110 
111 /* Protect totalram_pages and zone->managed_pages */
112 static DEFINE_SPINLOCK(managed_page_count_lock);
113 
114 unsigned long totalram_pages __read_mostly;
115 unsigned long totalreserve_pages __read_mostly;
116 unsigned long totalcma_pages __read_mostly;
117 /*
118  * When calculating the number of globally allowed dirty pages, there
119  * is a certain number of per-zone reserves that should not be
120  * considered dirtyable memory.  This is the sum of those reserves
121  * over all existing zones that contribute dirtyable memory.
122  */
123 unsigned long dirty_balance_reserve __read_mostly;
124 
125 int percpu_pagelist_fraction;
126 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
127 
128 /*
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_IO | __GFP_FS);
173 }
174 
175 bool pm_suspended_storage(void)
176 {
177         if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
178                 return false;
179         return true;
180 }
181 #endif /* CONFIG_PM_SLEEP */
182 
183 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
184 unsigned 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 static void free_compound_page(struct page *page);
233 compound_page_dtor * const compound_page_dtors[] = {
234         NULL,
235         free_compound_page,
236 #ifdef CONFIG_HUGETLB_PAGE
237         free_huge_page,
238 #endif
239 };
240 
241 int min_free_kbytes = 1024;
242 int user_min_free_kbytes = -1;
243 
244 static unsigned long __meminitdata nr_kernel_pages;
245 static unsigned long __meminitdata nr_all_pages;
246 static unsigned long __meminitdata dma_reserve;
247 
248 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
249 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
250 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
251 static unsigned long __initdata required_kernelcore;
252 static unsigned long __initdata required_movablecore;
253 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
254 
255 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
256 int movable_zone;
257 EXPORT_SYMBOL(movable_zone);
258 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
259 
260 #if MAX_NUMNODES > 1
261 int nr_node_ids __read_mostly = MAX_NUMNODES;
262 int nr_online_nodes __read_mostly = 1;
263 EXPORT_SYMBOL(nr_node_ids);
264 EXPORT_SYMBOL(nr_online_nodes);
265 #endif
266 
267 int page_group_by_mobility_disabled __read_mostly;
268 
269 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
270 static inline void reset_deferred_meminit(pg_data_t *pgdat)
271 {
272         pgdat->first_deferred_pfn = ULONG_MAX;
273 }
274 
275 /* Returns true if the struct page for the pfn is uninitialised */
276 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
277 {
278         if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
279                 return true;
280 
281         return false;
282 }
283 
284 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
285 {
286         if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
287                 return true;
288 
289         return false;
290 }
291 
292 /*
293  * Returns false when the remaining initialisation should be deferred until
294  * later in the boot cycle when it can be parallelised.
295  */
296 static inline bool update_defer_init(pg_data_t *pgdat,
297                                 unsigned long pfn, unsigned long zone_end,
298                                 unsigned long *nr_initialised)
299 {
300         /* Always populate low zones for address-contrained allocations */
301         if (zone_end < pgdat_end_pfn(pgdat))
302                 return true;
303 
304         /* Initialise at least 2G of the highest zone */
305         (*nr_initialised)++;
306         if (*nr_initialised > (2UL << (30 - PAGE_SHIFT)) &&
307             (pfn & (PAGES_PER_SECTION - 1)) == 0) {
308                 pgdat->first_deferred_pfn = pfn;
309                 return false;
310         }
311 
312         return true;
313 }
314 #else
315 static inline void reset_deferred_meminit(pg_data_t *pgdat)
316 {
317 }
318 
319 static inline bool early_page_uninitialised(unsigned long pfn)
320 {
321         return false;
322 }
323 
324 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
325 {
326         return false;
327 }
328 
329 static inline bool update_defer_init(pg_data_t *pgdat,
330                                 unsigned long pfn, unsigned long zone_end,
331                                 unsigned long *nr_initialised)
332 {
333         return true;
334 }
335 #endif
336 
337 
338 void set_pageblock_migratetype(struct page *page, int migratetype)
339 {
340         if (unlikely(page_group_by_mobility_disabled &&
341                      migratetype < MIGRATE_PCPTYPES))
342                 migratetype = MIGRATE_UNMOVABLE;
343 
344         set_pageblock_flags_group(page, (unsigned long)migratetype,
345                                         PB_migrate, PB_migrate_end);
346 }
347 
348 #ifdef CONFIG_DEBUG_VM
349 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
350 {
351         int ret = 0;
352         unsigned seq;
353         unsigned long pfn = page_to_pfn(page);
354         unsigned long sp, start_pfn;
355 
356         do {
357                 seq = zone_span_seqbegin(zone);
358                 start_pfn = zone->zone_start_pfn;
359                 sp = zone->spanned_pages;
360                 if (!zone_spans_pfn(zone, pfn))
361                         ret = 1;
362         } while (zone_span_seqretry(zone, seq));
363 
364         if (ret)
365                 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
366                         pfn, zone_to_nid(zone), zone->name,
367                         start_pfn, start_pfn + sp);
368 
369         return ret;
370 }
371 
372 static int page_is_consistent(struct zone *zone, struct page *page)
373 {
374         if (!pfn_valid_within(page_to_pfn(page)))
375                 return 0;
376         if (zone != page_zone(page))
377                 return 0;
378 
379         return 1;
380 }
381 /*
382  * Temporary debugging check for pages not lying within a given zone.
383  */
384 static int bad_range(struct zone *zone, struct page *page)
385 {
386         if (page_outside_zone_boundaries(zone, page))
387                 return 1;
388         if (!page_is_consistent(zone, page))
389                 return 1;
390 
391         return 0;
392 }
393 #else
394 static inline int bad_range(struct zone *zone, struct page *page)
395 {
396         return 0;
397 }
398 #endif
399 
400 static void bad_page(struct page *page, const char *reason,
401                 unsigned long bad_flags)
402 {
403         static unsigned long resume;
404         static unsigned long nr_shown;
405         static unsigned long nr_unshown;
406 
407         /* Don't complain about poisoned pages */
408         if (PageHWPoison(page)) {
409                 page_mapcount_reset(page); /* remove PageBuddy */
410                 return;
411         }
412 
413         /*
414          * Allow a burst of 60 reports, then keep quiet for that minute;
415          * or allow a steady drip of one report per second.
416          */
417         if (nr_shown == 60) {
418                 if (time_before(jiffies, resume)) {
419                         nr_unshown++;
420                         goto out;
421                 }
422                 if (nr_unshown) {
423                         printk(KERN_ALERT
424                               "BUG: Bad page state: %lu messages suppressed\n",
425                                 nr_unshown);
426                         nr_unshown = 0;
427                 }
428                 nr_shown = 0;
429         }
430         if (nr_shown++ == 0)
431                 resume = jiffies + 60 * HZ;
432 
433         printk(KERN_ALERT "BUG: Bad page state in process %s  pfn:%05lx\n",
434                 current->comm, page_to_pfn(page));
435         dump_page_badflags(page, reason, bad_flags);
436 
437         print_modules();
438         dump_stack();
439 out:
440         /* Leave bad fields for debug, except PageBuddy could make trouble */
441         page_mapcount_reset(page); /* remove PageBuddy */
442         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
443 }
444 
445 /*
446  * Higher-order pages are called "compound pages".  They are structured thusly:
447  *
448  * The first PAGE_SIZE page is called the "head page" and have PG_head set.
449  *
450  * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
451  * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
452  *
453  * The first tail page's ->compound_dtor holds the offset in array of compound
454  * page destructors. See compound_page_dtors.
455  *
456  * The first tail page's ->compound_order holds the order of allocation.
457  * This usage means that zero-order pages may not be compound.
458  */
459 
460 static void free_compound_page(struct page *page)
461 {
462         __free_pages_ok(page, compound_order(page));
463 }
464 
465 void prep_compound_page(struct page *page, unsigned int order)
466 {
467         int i;
468         int nr_pages = 1 << order;
469 
470         set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
471         set_compound_order(page, order);
472         __SetPageHead(page);
473         for (i = 1; i < nr_pages; i++) {
474                 struct page *p = page + i;
475                 set_page_count(p, 0);
476                 set_compound_head(p, page);
477         }
478 }
479 
480 #ifdef CONFIG_DEBUG_PAGEALLOC
481 unsigned int _debug_guardpage_minorder;
482 bool _debug_pagealloc_enabled __read_mostly;
483 bool _debug_guardpage_enabled __read_mostly;
484 
485 static int __init early_debug_pagealloc(char *buf)
486 {
487         if (!buf)
488                 return -EINVAL;
489 
490         if (strcmp(buf, "on") == 0)
491                 _debug_pagealloc_enabled = true;
492 
493         return 0;
494 }
495 early_param("debug_pagealloc", early_debug_pagealloc);
496 
497 static bool need_debug_guardpage(void)
498 {
499         /* If we don't use debug_pagealloc, we don't need guard page */
500         if (!debug_pagealloc_enabled())
501                 return false;
502 
503         return true;
504 }
505 
506 static void init_debug_guardpage(void)
507 {
508         if (!debug_pagealloc_enabled())
509                 return;
510 
511         _debug_guardpage_enabled = true;
512 }
513 
514 struct page_ext_operations debug_guardpage_ops = {
515         .need = need_debug_guardpage,
516         .init = init_debug_guardpage,
517 };
518 
519 static int __init debug_guardpage_minorder_setup(char *buf)
520 {
521         unsigned long res;
522 
523         if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
524                 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
525                 return 0;
526         }
527         _debug_guardpage_minorder = res;
528         printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
529         return 0;
530 }
531 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
532 
533 static inline void set_page_guard(struct zone *zone, struct page *page,
534                                 unsigned int order, int migratetype)
535 {
536         struct page_ext *page_ext;
537 
538         if (!debug_guardpage_enabled())
539                 return;
540 
541         page_ext = lookup_page_ext(page);
542         __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
543 
544         INIT_LIST_HEAD(&page->lru);
545         set_page_private(page, order);
546         /* Guard pages are not available for any usage */
547         __mod_zone_freepage_state(zone, -(1 << order), migratetype);
548 }
549 
550 static inline void clear_page_guard(struct zone *zone, struct page *page,
551                                 unsigned int order, int migratetype)
552 {
553         struct page_ext *page_ext;
554 
555         if (!debug_guardpage_enabled())
556                 return;
557 
558         page_ext = lookup_page_ext(page);
559         __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
560 
561         set_page_private(page, 0);
562         if (!is_migrate_isolate(migratetype))
563                 __mod_zone_freepage_state(zone, (1 << order), migratetype);
564 }
565 #else
566 struct page_ext_operations debug_guardpage_ops = { NULL, };
567 static inline void set_page_guard(struct zone *zone, struct page *page,
568                                 unsigned int order, int migratetype) {}
569 static inline void clear_page_guard(struct zone *zone, struct page *page,
570                                 unsigned int order, int migratetype) {}
571 #endif
572 
573 static inline void set_page_order(struct page *page, unsigned int order)
574 {
575         set_page_private(page, order);
576         __SetPageBuddy(page);
577 }
578 
579 static inline void rmv_page_order(struct page *page)
580 {
581         __ClearPageBuddy(page);
582         set_page_private(page, 0);
583 }
584 
585 /*
586  * This function checks whether a page is free && is the buddy
587  * we can do coalesce a page and its buddy if
588  * (a) the buddy is not in a hole &&
589  * (b) the buddy is in the buddy system &&
590  * (c) a page and its buddy have the same order &&
591  * (d) a page and its buddy are in the same zone.
592  *
593  * For recording whether a page is in the buddy system, we set ->_mapcount
594  * PAGE_BUDDY_MAPCOUNT_VALUE.
595  * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
596  * serialized by zone->lock.
597  *
598  * For recording page's order, we use page_private(page).
599  */
600 static inline int page_is_buddy(struct page *page, struct page *buddy,
601                                                         unsigned int order)
602 {
603         if (!pfn_valid_within(page_to_pfn(buddy)))
604                 return 0;
605 
606         if (page_is_guard(buddy) && page_order(buddy) == order) {
607                 if (page_zone_id(page) != page_zone_id(buddy))
608                         return 0;
609 
610                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
611 
612                 return 1;
613         }
614 
615         if (PageBuddy(buddy) && page_order(buddy) == order) {
616                 /*
617                  * zone check is done late to avoid uselessly
618                  * calculating zone/node ids for pages that could
619                  * never merge.
620                  */
621                 if (page_zone_id(page) != page_zone_id(buddy))
622                         return 0;
623 
624                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
625 
626                 return 1;
627         }
628         return 0;
629 }
630 
631 /*
632  * Freeing function for a buddy system allocator.
633  *
634  * The concept of a buddy system is to maintain direct-mapped table
635  * (containing bit values) for memory blocks of various "orders".
636  * The bottom level table contains the map for the smallest allocatable
637  * units of memory (here, pages), and each level above it describes
638  * pairs of units from the levels below, hence, "buddies".
639  * At a high level, all that happens here is marking the table entry
640  * at the bottom level available, and propagating the changes upward
641  * as necessary, plus some accounting needed to play nicely with other
642  * parts of the VM system.
643  * At each level, we keep a list of pages, which are heads of continuous
644  * free pages of length of (1 << order) and marked with _mapcount
645  * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
646  * field.
647  * So when we are allocating or freeing one, we can derive the state of the
648  * other.  That is, if we allocate a small block, and both were
649  * free, the remainder of the region must be split into blocks.
650  * If a block is freed, and its buddy is also free, then this
651  * triggers coalescing into a block of larger size.
652  *
653  * -- nyc
654  */
655 
656 static inline void __free_one_page(struct page *page,
657                 unsigned long pfn,
658                 struct zone *zone, unsigned int order,
659                 int migratetype)
660 {
661         unsigned long page_idx;
662         unsigned long combined_idx;
663         unsigned long uninitialized_var(buddy_idx);
664         struct page *buddy;
665         unsigned int max_order = MAX_ORDER;
666 
667         VM_BUG_ON(!zone_is_initialized(zone));
668         VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
669 
670         VM_BUG_ON(migratetype == -1);
671         if (is_migrate_isolate(migratetype)) {
672                 /*
673                  * We restrict max order of merging to prevent merge
674                  * between freepages on isolate pageblock and normal
675                  * pageblock. Without this, pageblock isolation
676                  * could cause incorrect freepage accounting.
677                  */
678                 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
679         } else {
680                 __mod_zone_freepage_state(zone, 1 << order, migratetype);
681         }
682 
683         page_idx = pfn & ((1 << max_order) - 1);
684 
685         VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
686         VM_BUG_ON_PAGE(bad_range(zone, page), page);
687 
688         while (order < max_order - 1) {
689                 buddy_idx = __find_buddy_index(page_idx, order);
690                 buddy = page + (buddy_idx - page_idx);
691                 if (!page_is_buddy(page, buddy, order))
692                         break;
693                 /*
694                  * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
695                  * merge with it and move up one order.
696                  */
697                 if (page_is_guard(buddy)) {
698                         clear_page_guard(zone, buddy, order, migratetype);
699                 } else {
700                         list_del(&buddy->lru);
701                         zone->free_area[order].nr_free--;
702                         rmv_page_order(buddy);
703                 }
704                 combined_idx = buddy_idx & page_idx;
705                 page = page + (combined_idx - page_idx);
706                 page_idx = combined_idx;
707                 order++;
708         }
709         set_page_order(page, order);
710 
711         /*
712          * If this is not the largest possible page, check if the buddy
713          * of the next-highest order is free. If it is, it's possible
714          * that pages are being freed that will coalesce soon. In case,
715          * that is happening, add the free page to the tail of the list
716          * so it's less likely to be used soon and more likely to be merged
717          * as a higher order page
718          */
719         if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
720                 struct page *higher_page, *higher_buddy;
721                 combined_idx = buddy_idx & page_idx;
722                 higher_page = page + (combined_idx - page_idx);
723                 buddy_idx = __find_buddy_index(combined_idx, order + 1);
724                 higher_buddy = higher_page + (buddy_idx - combined_idx);
725                 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
726                         list_add_tail(&page->lru,
727                                 &zone->free_area[order].free_list[migratetype]);
728                         goto out;
729                 }
730         }
731 
732         list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
733 out:
734         zone->free_area[order].nr_free++;
735 }
736 
737 static inline int free_pages_check(struct page *page)
738 {
739         const char *bad_reason = NULL;
740         unsigned long bad_flags = 0;
741 
742         if (unlikely(page_mapcount(page)))
743                 bad_reason = "nonzero mapcount";
744         if (unlikely(page->mapping != NULL))
745                 bad_reason = "non-NULL mapping";
746         if (unlikely(atomic_read(&page->_count) != 0))
747                 bad_reason = "nonzero _count";
748         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
749                 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
750                 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
751         }
752 #ifdef CONFIG_MEMCG
753         if (unlikely(page->mem_cgroup))
754                 bad_reason = "page still charged to cgroup";
755 #endif
756         if (unlikely(bad_reason)) {
757                 bad_page(page, bad_reason, bad_flags);
758                 return 1;
759         }
760         page_cpupid_reset_last(page);
761         if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
762                 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
763         return 0;
764 }
765 
766 /*
767  * Frees a number of pages from the PCP lists
768  * Assumes all pages on list are in same zone, and of same order.
769  * count is the number of pages to free.
770  *
771  * If the zone was previously in an "all pages pinned" state then look to
772  * see if this freeing clears that state.
773  *
774  * And clear the zone's pages_scanned counter, to hold off the "all pages are
775  * pinned" detection logic.
776  */
777 static void free_pcppages_bulk(struct zone *zone, int count,
778                                         struct per_cpu_pages *pcp)
779 {
780         int migratetype = 0;
781         int batch_free = 0;
782         int to_free = count;
783         unsigned long nr_scanned;
784 
785         spin_lock(&zone->lock);
786         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
787         if (nr_scanned)
788                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
789 
790         while (to_free) {
791                 struct page *page;
792                 struct list_head *list;
793 
794                 /*
795                  * Remove pages from lists in a round-robin fashion. A
796                  * batch_free count is maintained that is incremented when an
797                  * empty list is encountered.  This is so more pages are freed
798                  * off fuller lists instead of spinning excessively around empty
799                  * lists
800                  */
801                 do {
802                         batch_free++;
803                         if (++migratetype == MIGRATE_PCPTYPES)
804                                 migratetype = 0;
805                         list = &pcp->lists[migratetype];
806                 } while (list_empty(list));
807 
808                 /* This is the only non-empty list. Free them all. */
809                 if (batch_free == MIGRATE_PCPTYPES)
810                         batch_free = to_free;
811 
812                 do {
813                         int mt; /* migratetype of the to-be-freed page */
814 
815                         page = list_entry(list->prev, struct page, lru);
816                         /* must delete as __free_one_page list manipulates */
817                         list_del(&page->lru);
818 
819                         mt = get_pcppage_migratetype(page);
820                         /* MIGRATE_ISOLATE page should not go to pcplists */
821                         VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
822                         /* Pageblock could have been isolated meanwhile */
823                         if (unlikely(has_isolate_pageblock(zone)))
824                                 mt = get_pageblock_migratetype(page);
825 
826                         __free_one_page(page, page_to_pfn(page), zone, 0, mt);
827                         trace_mm_page_pcpu_drain(page, 0, mt);
828                 } while (--to_free && --batch_free && !list_empty(list));
829         }
830         spin_unlock(&zone->lock);
831 }
832 
833 static void free_one_page(struct zone *zone,
834                                 struct page *page, unsigned long pfn,
835                                 unsigned int order,
836                                 int migratetype)
837 {
838         unsigned long nr_scanned;
839         spin_lock(&zone->lock);
840         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
841         if (nr_scanned)
842                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
843 
844         if (unlikely(has_isolate_pageblock(zone) ||
845                 is_migrate_isolate(migratetype))) {
846                 migratetype = get_pfnblock_migratetype(page, pfn);
847         }
848         __free_one_page(page, pfn, zone, order, migratetype);
849         spin_unlock(&zone->lock);
850 }
851 
852 static int free_tail_pages_check(struct page *head_page, struct page *page)
853 {
854         int ret = 1;
855 
856         /*
857          * We rely page->lru.next never has bit 0 set, unless the page
858          * is PageTail(). Let's make sure that's true even for poisoned ->lru.
859          */
860         BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
861 
862         if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
863                 ret = 0;
864                 goto out;
865         }
866         if (unlikely(!PageTail(page))) {
867                 bad_page(page, "PageTail not set", 0);
868                 goto out;
869         }
870         if (unlikely(compound_head(page) != head_page)) {
871                 bad_page(page, "compound_head not consistent", 0);
872                 goto out;
873         }
874         ret = 0;
875 out:
876         clear_compound_head(page);
877         return ret;
878 }
879 
880 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
881                                 unsigned long zone, int nid)
882 {
883         set_page_links(page, zone, nid, pfn);
884         init_page_count(page);
885         page_mapcount_reset(page);
886         page_cpupid_reset_last(page);
887 
888         INIT_LIST_HEAD(&page->lru);
889 #ifdef WANT_PAGE_VIRTUAL
890         /* The shift won't overflow because ZONE_NORMAL is below 4G. */
891         if (!is_highmem_idx(zone))
892                 set_page_address(page, __va(pfn << PAGE_SHIFT));
893 #endif
894 }
895 
896 static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
897                                         int nid)
898 {
899         return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
900 }
901 
902 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
903 static void init_reserved_page(unsigned long pfn)
904 {
905         pg_data_t *pgdat;
906         int nid, zid;
907 
908         if (!early_page_uninitialised(pfn))
909                 return;
910 
911         nid = early_pfn_to_nid(pfn);
912         pgdat = NODE_DATA(nid);
913 
914         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
915                 struct zone *zone = &pgdat->node_zones[zid];
916 
917                 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
918                         break;
919         }
920         __init_single_pfn(pfn, zid, nid);
921 }
922 #else
923 static inline void init_reserved_page(unsigned long pfn)
924 {
925 }
926 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
927 
928 /*
929  * Initialised pages do not have PageReserved set. This function is
930  * called for each range allocated by the bootmem allocator and
931  * marks the pages PageReserved. The remaining valid pages are later
932  * sent to the buddy page allocator.
933  */
934 void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
935 {
936         unsigned long start_pfn = PFN_DOWN(start);
937         unsigned long end_pfn = PFN_UP(end);
938 
939         for (; start_pfn < end_pfn; start_pfn++) {
940                 if (pfn_valid(start_pfn)) {
941                         struct page *page = pfn_to_page(start_pfn);
942 
943                         init_reserved_page(start_pfn);
944 
945                         /* Avoid false-positive PageTail() */
946                         INIT_LIST_HEAD(&page->lru);
947 
948                         SetPageReserved(page);
949                 }
950         }
951 }
952 
953 static bool free_pages_prepare(struct page *page, unsigned int order)
954 {
955         bool compound = PageCompound(page);
956         int i, bad = 0;
957 
958         VM_BUG_ON_PAGE(PageTail(page), page);
959         VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
960 
961         trace_mm_page_free(page, order);
962         kmemcheck_free_shadow(page, order);
963         kasan_free_pages(page, order);
964 
965         if (PageAnon(page))
966                 page->mapping = NULL;
967         bad += free_pages_check(page);
968         for (i = 1; i < (1 << order); i++) {
969                 if (compound)
970                         bad += free_tail_pages_check(page, page + i);
971                 bad += free_pages_check(page + i);
972         }
973         if (bad)
974                 return false;
975 
976         reset_page_owner(page, order);
977 
978         if (!PageHighMem(page)) {
979                 debug_check_no_locks_freed(page_address(page),
980                                            PAGE_SIZE << order);
981                 debug_check_no_obj_freed(page_address(page),
982                                            PAGE_SIZE << order);
983         }
984         arch_free_page(page, order);
985         kernel_map_pages(page, 1 << order, 0);
986 
987         return true;
988 }
989 
990 static void __free_pages_ok(struct page *page, unsigned int order)
991 {
992         unsigned long flags;
993         int migratetype;
994         unsigned long pfn = page_to_pfn(page);
995 
996         if (!free_pages_prepare(page, order))
997                 return;
998 
999         migratetype = get_pfnblock_migratetype(page, pfn);
1000         local_irq_save(flags);
1001         __count_vm_events(PGFREE, 1 << order);
1002         free_one_page(page_zone(page), page, pfn, order, migratetype);
1003         local_irq_restore(flags);
1004 }
1005 
1006 static void __init __free_pages_boot_core(struct page *page,
1007                                         unsigned long pfn, unsigned int order)
1008 {
1009         unsigned int nr_pages = 1 << order;
1010         struct page *p = page;
1011         unsigned int loop;
1012 
1013         prefetchw(p);
1014         for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1015                 prefetchw(p + 1);
1016                 __ClearPageReserved(p);
1017                 set_page_count(p, 0);
1018         }
1019         __ClearPageReserved(p);
1020         set_page_count(p, 0);
1021 
1022         page_zone(page)->managed_pages += nr_pages;
1023         set_page_refcounted(page);
1024         __free_pages(page, order);
1025 }
1026 
1027 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1028         defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1029 
1030 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1031 
1032 int __meminit early_pfn_to_nid(unsigned long pfn)
1033 {
1034         static DEFINE_SPINLOCK(early_pfn_lock);
1035         int nid;
1036 
1037         spin_lock(&early_pfn_lock);
1038         nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1039         if (nid < 0)
1040                 nid = 0;
1041         spin_unlock(&early_pfn_lock);
1042 
1043         return nid;
1044 }
1045 #endif
1046 
1047 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1048 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1049                                         struct mminit_pfnnid_cache *state)
1050 {
1051         int nid;
1052 
1053         nid = __early_pfn_to_nid(pfn, state);
1054         if (nid >= 0 && nid != node)
1055                 return false;
1056         return true;
1057 }
1058 
1059 /* Only safe to use early in boot when initialisation is single-threaded */
1060 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1061 {
1062         return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1063 }
1064 
1065 #else
1066 
1067 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1068 {
1069         return true;
1070 }
1071 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1072                                         struct mminit_pfnnid_cache *state)
1073 {
1074         return true;
1075 }
1076 #endif
1077 
1078 
1079 void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
1080                                                         unsigned int order)
1081 {
1082         if (early_page_uninitialised(pfn))
1083                 return;
1084         return __free_pages_boot_core(page, pfn, order);
1085 }
1086 
1087 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1088 static void __init deferred_free_range(struct page *page,
1089                                         unsigned long pfn, int nr_pages)
1090 {
1091         int i;
1092 
1093         if (!page)
1094                 return;
1095 
1096         /* Free a large naturally-aligned chunk if possible */
1097         if (nr_pages == MAX_ORDER_NR_PAGES &&
1098             (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
1099                 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1100                 __free_pages_boot_core(page, pfn, MAX_ORDER-1);
1101                 return;
1102         }
1103 
1104         for (i = 0; i < nr_pages; i++, page++, pfn++)
1105                 __free_pages_boot_core(page, pfn, 0);
1106 }
1107 
1108 /* Completion tracking for deferred_init_memmap() threads */
1109 static atomic_t pgdat_init_n_undone __initdata;
1110 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1111 
1112 static inline void __init pgdat_init_report_one_done(void)
1113 {
1114         if (atomic_dec_and_test(&pgdat_init_n_undone))
1115                 complete(&pgdat_init_all_done_comp);
1116 }
1117 
1118 /* Initialise remaining memory on a node */
1119 static int __init deferred_init_memmap(void *data)
1120 {
1121         pg_data_t *pgdat = data;
1122         int nid = pgdat->node_id;
1123         struct mminit_pfnnid_cache nid_init_state = { };
1124         unsigned long start = jiffies;
1125         unsigned long nr_pages = 0;
1126         unsigned long walk_start, walk_end;
1127         int i, zid;
1128         struct zone *zone;
1129         unsigned long first_init_pfn = pgdat->first_deferred_pfn;
1130         const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1131 
1132         if (first_init_pfn == ULONG_MAX) {
1133                 pgdat_init_report_one_done();
1134                 return 0;
1135         }
1136 
1137         /* Bind memory initialisation thread to a local node if possible */
1138         if (!cpumask_empty(cpumask))
1139                 set_cpus_allowed_ptr(current, cpumask);
1140 
1141         /* Sanity check boundaries */
1142         BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1143         BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1144         pgdat->first_deferred_pfn = ULONG_MAX;
1145 
1146         /* Only the highest zone is deferred so find it */
1147         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1148                 zone = pgdat->node_zones + zid;
1149                 if (first_init_pfn < zone_end_pfn(zone))
1150                         break;
1151         }
1152 
1153         for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1154                 unsigned long pfn, end_pfn;
1155                 struct page *page = NULL;
1156                 struct page *free_base_page = NULL;
1157                 unsigned long free_base_pfn = 0;
1158                 int nr_to_free = 0;
1159 
1160                 end_pfn = min(walk_end, zone_end_pfn(zone));
1161                 pfn = first_init_pfn;
1162                 if (pfn < walk_start)
1163                         pfn = walk_start;
1164                 if (pfn < zone->zone_start_pfn)
1165                         pfn = zone->zone_start_pfn;
1166 
1167                 for (; pfn < end_pfn; pfn++) {
1168                         if (!pfn_valid_within(pfn))
1169                                 goto free_range;
1170 
1171                         /*
1172                          * Ensure pfn_valid is checked every
1173                          * MAX_ORDER_NR_PAGES for memory holes
1174                          */
1175                         if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1176                                 if (!pfn_valid(pfn)) {
1177                                         page = NULL;
1178                                         goto free_range;
1179                                 }
1180                         }
1181 
1182                         if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1183                                 page = NULL;
1184                                 goto free_range;
1185                         }
1186 
1187                         /* Minimise pfn page lookups and scheduler checks */
1188                         if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1189                                 page++;
1190                         } else {
1191                                 nr_pages += nr_to_free;
1192                                 deferred_free_range(free_base_page,
1193                                                 free_base_pfn, nr_to_free);
1194                                 free_base_page = NULL;
1195                                 free_base_pfn = nr_to_free = 0;
1196 
1197                                 page = pfn_to_page(pfn);
1198                                 cond_resched();
1199                         }
1200 
1201                         if (page->flags) {
1202                                 VM_BUG_ON(page_zone(page) != zone);
1203                                 goto free_range;
1204                         }
1205 
1206                         __init_single_page(page, pfn, zid, nid);
1207                         if (!free_base_page) {
1208                                 free_base_page = page;
1209                                 free_base_pfn = pfn;
1210                                 nr_to_free = 0;
1211                         }
1212                         nr_to_free++;
1213 
1214                         /* Where possible, batch up pages for a single free */
1215                         continue;
1216 free_range:
1217                         /* Free the current block of pages to allocator */
1218                         nr_pages += nr_to_free;
1219                         deferred_free_range(free_base_page, free_base_pfn,
1220                                                                 nr_to_free);
1221                         free_base_page = NULL;
1222                         free_base_pfn = nr_to_free = 0;
1223                 }
1224 
1225                 first_init_pfn = max(end_pfn, first_init_pfn);
1226         }
1227 
1228         /* Sanity check that the next zone really is unpopulated */
1229         WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1230 
1231         pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1232                                         jiffies_to_msecs(jiffies - start));
1233 
1234         pgdat_init_report_one_done();
1235         return 0;
1236 }
1237 
1238 void __init page_alloc_init_late(void)
1239 {
1240         int nid;
1241 
1242         /* There will be num_node_state(N_MEMORY) threads */
1243         atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1244         for_each_node_state(nid, N_MEMORY) {
1245                 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1246         }
1247 
1248         /* Block until all are initialised */
1249         wait_for_completion(&pgdat_init_all_done_comp);
1250 
1251         /* Reinit limits that are based on free pages after the kernel is up */
1252         files_maxfiles_init();
1253 }
1254 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1255 
1256 #ifdef CONFIG_CMA
1257 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1258 void __init init_cma_reserved_pageblock(struct page *page)
1259 {
1260         unsigned i = pageblock_nr_pages;
1261         struct page *p = page;
1262 
1263         do {
1264                 __ClearPageReserved(p);
1265                 set_page_count(p, 0);
1266         } while (++p, --i);
1267 
1268         set_pageblock_migratetype(page, MIGRATE_CMA);
1269 
1270         if (pageblock_order >= MAX_ORDER) {
1271                 i = pageblock_nr_pages;
1272                 p = page;
1273                 do {
1274                         set_page_refcounted(p);
1275                         __free_pages(p, MAX_ORDER - 1);
1276                         p += MAX_ORDER_NR_PAGES;
1277                 } while (i -= MAX_ORDER_NR_PAGES);
1278         } else {
1279                 set_page_refcounted(page);
1280                 __free_pages(page, pageblock_order);
1281         }
1282 
1283         adjust_managed_page_count(page, pageblock_nr_pages);
1284 }
1285 #endif
1286 
1287 /*
1288  * The order of subdivision here is critical for the IO subsystem.
1289  * Please do not alter this order without good reasons and regression
1290  * testing. Specifically, as large blocks of memory are subdivided,
1291  * the order in which smaller blocks are delivered depends on the order
1292  * they're subdivided in this function. This is the primary factor
1293  * influencing the order in which pages are delivered to the IO
1294  * subsystem according to empirical testing, and this is also justified
1295  * by considering the behavior of a buddy system containing a single
1296  * large block of memory acted on by a series of small allocations.
1297  * This behavior is a critical factor in sglist merging's success.
1298  *
1299  * -- nyc
1300  */
1301 static inline void expand(struct zone *zone, struct page *page,
1302         int low, int high, struct free_area *area,
1303         int migratetype)
1304 {
1305         unsigned long size = 1 << high;
1306 
1307         while (high > low) {
1308                 area--;
1309                 high--;
1310                 size >>= 1;
1311                 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1312 
1313                 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
1314                         debug_guardpage_enabled() &&
1315                         high < debug_guardpage_minorder()) {
1316                         /*
1317                          * Mark as guard pages (or page), that will allow to
1318                          * merge back to allocator when buddy will be freed.
1319                          * Corresponding page table entries will not be touched,
1320                          * pages will stay not present in virtual address space
1321                          */
1322                         set_page_guard(zone, &page[size], high, migratetype);
1323                         continue;
1324                 }
1325                 list_add(&page[size].lru, &area->free_list[migratetype]);
1326                 area->nr_free++;
1327                 set_page_order(&page[size], high);
1328         }
1329 }
1330 
1331 /*
1332  * This page is about to be returned from the page allocator
1333  */
1334 static inline int check_new_page(struct page *page)
1335 {
1336         const char *bad_reason = NULL;
1337         unsigned long bad_flags = 0;
1338 
1339         if (unlikely(page_mapcount(page)))
1340                 bad_reason = "nonzero mapcount";
1341         if (unlikely(page->mapping != NULL))
1342                 bad_reason = "non-NULL mapping";
1343         if (unlikely(atomic_read(&page->_count) != 0))
1344                 bad_reason = "nonzero _count";
1345         if (unlikely(page->flags & __PG_HWPOISON)) {
1346                 bad_reason = "HWPoisoned (hardware-corrupted)";
1347                 bad_flags = __PG_HWPOISON;
1348         }
1349         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1350                 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1351                 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1352         }
1353 #ifdef CONFIG_MEMCG
1354         if (unlikely(page->mem_cgroup))
1355                 bad_reason = "page still charged to cgroup";
1356 #endif
1357         if (unlikely(bad_reason)) {
1358                 bad_page(page, bad_reason, bad_flags);
1359                 return 1;
1360         }
1361         return 0;
1362 }
1363 
1364 static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1365                                                                 int alloc_flags)
1366 {
1367         int i;
1368 
1369         for (i = 0; i < (1 << order); i++) {
1370                 struct page *p = page + i;
1371                 if (unlikely(check_new_page(p)))
1372                         return 1;
1373         }
1374 
1375         set_page_private(page, 0);
1376         set_page_refcounted(page);
1377 
1378         arch_alloc_page(page, order);
1379         kernel_map_pages(page, 1 << order, 1);
1380         kasan_alloc_pages(page, order);
1381 
1382         if (gfp_flags & __GFP_ZERO)
1383                 for (i = 0; i < (1 << order); i++)
1384                         clear_highpage(page + i);
1385 
1386         if (order && (gfp_flags & __GFP_COMP))
1387                 prep_compound_page(page, order);
1388 
1389         set_page_owner(page, order, gfp_flags);
1390 
1391         /*
1392          * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1393          * allocate the page. The expectation is that the caller is taking
1394          * steps that will free more memory. The caller should avoid the page
1395          * being used for !PFMEMALLOC purposes.
1396          */
1397         if (alloc_flags & ALLOC_NO_WATERMARKS)
1398                 set_page_pfmemalloc(page);
1399         else
1400                 clear_page_pfmemalloc(page);
1401 
1402         return 0;
1403 }
1404 
1405 /*
1406  * Go through the free lists for the given migratetype and remove
1407  * the smallest available page from the freelists
1408  */
1409 static inline
1410 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1411                                                 int migratetype)
1412 {
1413         unsigned int current_order;
1414         struct free_area *area;
1415         struct page *page;
1416 
1417         /* Find a page of the appropriate size in the preferred list */
1418         for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1419                 area = &(zone->free_area[current_order]);
1420                 if (list_empty(&area->free_list[migratetype]))
1421                         continue;
1422 
1423                 page = list_entry(area->free_list[migratetype].next,
1424                                                         struct page, lru);
1425                 list_del(&page->lru);
1426                 rmv_page_order(page);
1427                 area->nr_free--;
1428                 expand(zone, page, order, current_order, area, migratetype);
1429                 set_pcppage_migratetype(page, migratetype);
1430                 return page;
1431         }
1432 
1433         return NULL;
1434 }
1435 
1436 
1437 /*
1438  * This array describes the order lists are fallen back to when
1439  * the free lists for the desirable migrate type are depleted
1440  */
1441 static int fallbacks[MIGRATE_TYPES][4] = {
1442         [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_TYPES },
1443         [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_TYPES },
1444         [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
1445 #ifdef CONFIG_CMA
1446         [MIGRATE_CMA]         = { MIGRATE_TYPES }, /* Never used */
1447 #endif
1448 #ifdef CONFIG_MEMORY_ISOLATION
1449         [MIGRATE_ISOLATE]     = { MIGRATE_TYPES }, /* Never used */
1450 #endif
1451 };
1452 
1453 #ifdef CONFIG_CMA
1454 static struct page *__rmqueue_cma_fallback(struct zone *zone,
1455                                         unsigned int order)
1456 {
1457         return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1458 }
1459 #else
1460 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1461                                         unsigned int order) { return NULL; }
1462 #endif
1463 
1464 /*
1465  * Move the free pages in a range to the free lists of the requested type.
1466  * Note that start_page and end_pages are not aligned on a pageblock
1467  * boundary. If alignment is required, use move_freepages_block()
1468  */
1469 int move_freepages(struct zone *zone,
1470                           struct page *start_page, struct page *end_page,
1471                           int migratetype)
1472 {
1473         struct page *page;
1474         unsigned int order;
1475         int pages_moved = 0;
1476 
1477 #ifndef CONFIG_HOLES_IN_ZONE
1478         /*
1479          * page_zone is not safe to call in this context when
1480          * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1481          * anyway as we check zone boundaries in move_freepages_block().
1482          * Remove at a later date when no bug reports exist related to
1483          * grouping pages by mobility
1484          */
1485         VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1486 #endif
1487 
1488         for (page = start_page; page <= end_page;) {
1489                 /* Make sure we are not inadvertently changing nodes */
1490                 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1491 
1492                 if (!pfn_valid_within(page_to_pfn(page))) {
1493                         page++;
1494                         continue;
1495                 }
1496 
1497                 if (!PageBuddy(page)) {
1498                         page++;
1499                         continue;
1500                 }
1501 
1502                 order = page_order(page);
1503                 list_move(&page->lru,
1504                           &zone->free_area[order].free_list[migratetype]);
1505                 page += 1 << order;
1506                 pages_moved += 1 << order;
1507         }
1508 
1509         return pages_moved;
1510 }
1511 
1512 int move_freepages_block(struct zone *zone, struct page *page,
1513                                 int migratetype)
1514 {
1515         unsigned long start_pfn, end_pfn;
1516         struct page *start_page, *end_page;
1517 
1518         start_pfn = page_to_pfn(page);
1519         start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1520         start_page = pfn_to_page(start_pfn);
1521         end_page = start_page + pageblock_nr_pages - 1;
1522         end_pfn = start_pfn + pageblock_nr_pages - 1;
1523 
1524         /* Do not cross zone boundaries */
1525         if (!zone_spans_pfn(zone, start_pfn))
1526                 start_page = page;
1527         if (!zone_spans_pfn(zone, end_pfn))
1528                 return 0;
1529 
1530         return move_freepages(zone, start_page, end_page, migratetype);
1531 }
1532 
1533 static void change_pageblock_range(struct page *pageblock_page,
1534                                         int start_order, int migratetype)
1535 {
1536         int nr_pageblocks = 1 << (start_order - pageblock_order);
1537 
1538         while (nr_pageblocks--) {
1539                 set_pageblock_migratetype(pageblock_page, migratetype);
1540                 pageblock_page += pageblock_nr_pages;
1541         }
1542 }
1543 
1544 /*
1545  * When we are falling back to another migratetype during allocation, try to
1546  * steal extra free pages from the same pageblocks to satisfy further
1547  * allocations, instead of polluting multiple pageblocks.
1548  *
1549  * If we are stealing a relatively large buddy page, it is likely there will
1550  * be more free pages in the pageblock, so try to steal them all. For
1551  * reclaimable and unmovable allocations, we steal regardless of page size,
1552  * as fragmentation caused by those allocations polluting movable pageblocks
1553  * is worse than movable allocations stealing from unmovable and reclaimable
1554  * pageblocks.
1555  */
1556 static bool can_steal_fallback(unsigned int order, int start_mt)
1557 {
1558         /*
1559          * Leaving this order check is intended, although there is
1560          * relaxed order check in next check. The reason is that
1561          * we can actually steal whole pageblock if this condition met,
1562          * but, below check doesn't guarantee it and that is just heuristic
1563          * so could be changed anytime.
1564          */
1565         if (order >= pageblock_order)
1566                 return true;
1567 
1568         if (order >= pageblock_order / 2 ||
1569                 start_mt == MIGRATE_RECLAIMABLE ||
1570                 start_mt == MIGRATE_UNMOVABLE ||
1571                 page_group_by_mobility_disabled)
1572                 return true;
1573 
1574         return false;
1575 }
1576 
1577 /*
1578  * This function implements actual steal behaviour. If order is large enough,
1579  * we can steal whole pageblock. If not, we first move freepages in this
1580  * pageblock and check whether half of pages are moved or not. If half of
1581  * pages are moved, we can change migratetype of pageblock and permanently
1582  * use it's pages as requested migratetype in the future.
1583  */
1584 static void steal_suitable_fallback(struct zone *zone, struct page *page,
1585                                                           int start_type)
1586 {
1587         unsigned int current_order = page_order(page);
1588         int pages;
1589 
1590         /* Take ownership for orders >= pageblock_order */
1591         if (current_order >= pageblock_order) {
1592                 change_pageblock_range(page, current_order, start_type);
1593                 return;
1594         }
1595 
1596         pages = move_freepages_block(zone, page, start_type);
1597 
1598         /* Claim the whole block if over half of it is free */
1599         if (pages >= (1 << (pageblock_order-1)) ||
1600                         page_group_by_mobility_disabled)
1601                 set_pageblock_migratetype(page, start_type);
1602 }
1603 
1604 /*
1605  * Check whether there is a suitable fallback freepage with requested order.
1606  * If only_stealable is true, this function returns fallback_mt only if
1607  * we can steal other freepages all together. This would help to reduce
1608  * fragmentation due to mixed migratetype pages in one pageblock.
1609  */
1610 int find_suitable_fallback(struct free_area *area, unsigned int order,
1611                         int migratetype, bool only_stealable, bool *can_steal)
1612 {
1613         int i;
1614         int fallback_mt;
1615 
1616         if (area->nr_free == 0)
1617                 return -1;
1618 
1619         *can_steal = false;
1620         for (i = 0;; i++) {
1621                 fallback_mt = fallbacks[migratetype][i];
1622                 if (fallback_mt == MIGRATE_TYPES)
1623                         break;
1624 
1625                 if (list_empty(&area->free_list[fallback_mt]))
1626                         continue;
1627 
1628                 if (can_steal_fallback(order, migratetype))
1629                         *can_steal = true;
1630 
1631                 if (!only_stealable)
1632                         return fallback_mt;
1633 
1634                 if (*can_steal)
1635                         return fallback_mt;
1636         }
1637 
1638         return -1;
1639 }
1640 
1641 /*
1642  * Reserve a pageblock for exclusive use of high-order atomic allocations if
1643  * there are no empty page blocks that contain a page with a suitable order
1644  */
1645 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
1646                                 unsigned int alloc_order)
1647 {
1648         int mt;
1649         unsigned long max_managed, flags;
1650 
1651         /*
1652          * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1653          * Check is race-prone but harmless.
1654          */
1655         max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
1656         if (zone->nr_reserved_highatomic >= max_managed)
1657                 return;
1658 
1659         spin_lock_irqsave(&zone->lock, flags);
1660 
1661         /* Recheck the nr_reserved_highatomic limit under the lock */
1662         if (zone->nr_reserved_highatomic >= max_managed)
1663                 goto out_unlock;
1664 
1665         /* Yoink! */
1666         mt = get_pageblock_migratetype(page);
1667         if (mt != MIGRATE_HIGHATOMIC &&
1668                         !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
1669                 zone->nr_reserved_highatomic += pageblock_nr_pages;
1670                 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
1671                 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
1672         }
1673 
1674 out_unlock:
1675         spin_unlock_irqrestore(&zone->lock, flags);
1676 }
1677 
1678 /*
1679  * Used when an allocation is about to fail under memory pressure. This
1680  * potentially hurts the reliability of high-order allocations when under
1681  * intense memory pressure but failed atomic allocations should be easier
1682  * to recover from than an OOM.
1683  */
1684 static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
1685 {
1686         struct zonelist *zonelist = ac->zonelist;
1687         unsigned long flags;
1688         struct zoneref *z;
1689         struct zone *zone;
1690         struct page *page;
1691         int order;
1692 
1693         for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
1694                                                                 ac->nodemask) {
1695                 /* Preserve at least one pageblock */
1696                 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
1697                         continue;
1698 
1699                 spin_lock_irqsave(&zone->lock, flags);
1700                 for (order = 0; order < MAX_ORDER; order++) {
1701                         struct free_area *area = &(zone->free_area[order]);
1702 
1703                         if (list_empty(&area->free_list[MIGRATE_HIGHATOMIC]))
1704                                 continue;
1705 
1706                         page = list_entry(area->free_list[MIGRATE_HIGHATOMIC].next,
1707                                                 struct page, lru);
1708 
1709                         /*
1710                          * It should never happen but changes to locking could
1711                          * inadvertently allow a per-cpu drain to add pages
1712                          * to MIGRATE_HIGHATOMIC while unreserving so be safe
1713                          * and watch for underflows.
1714                          */
1715                         zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
1716                                 zone->nr_reserved_highatomic);
1717 
1718                         /*
1719                          * Convert to ac->migratetype and avoid the normal
1720                          * pageblock stealing heuristics. Minimally, the caller
1721                          * is doing the work and needs the pages. More
1722                          * importantly, if the block was always converted to
1723                          * MIGRATE_UNMOVABLE or another type then the number
1724                          * of pageblocks that cannot be completely freed
1725                          * may increase.
1726                          */
1727                         set_pageblock_migratetype(page, ac->migratetype);
1728                         move_freepages_block(zone, page, ac->migratetype);
1729                         spin_unlock_irqrestore(&zone->lock, flags);
1730                         return;
1731                 }
1732                 spin_unlock_irqrestore(&zone->lock, flags);
1733         }
1734 }
1735 
1736 /* Remove an element from the buddy allocator from the fallback list */
1737 static inline struct page *
1738 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
1739 {
1740         struct free_area *area;
1741         unsigned int current_order;
1742         struct page *page;
1743         int fallback_mt;
1744         bool can_steal;
1745 
1746         /* Find the largest possible block of pages in the other list */
1747         for (current_order = MAX_ORDER-1;
1748                                 current_order >= order && current_order <= MAX_ORDER-1;
1749                                 --current_order) {
1750                 area = &(zone->free_area[current_order]);
1751                 fallback_mt = find_suitable_fallback(area, current_order,
1752                                 start_migratetype, false, &can_steal);
1753                 if (fallback_mt == -1)
1754                         continue;
1755 
1756                 page = list_entry(area->free_list[fallback_mt].next,
1757                                                 struct page, lru);
1758                 if (can_steal)
1759                         steal_suitable_fallback(zone, page, start_migratetype);
1760 
1761                 /* Remove the page from the freelists */
1762                 area->nr_free--;
1763                 list_del(&page->lru);
1764                 rmv_page_order(page);
1765 
1766                 expand(zone, page, order, current_order, area,
1767                                         start_migratetype);
1768                 /*
1769                  * The pcppage_migratetype may differ from pageblock's
1770                  * migratetype depending on the decisions in
1771                  * find_suitable_fallback(). This is OK as long as it does not
1772                  * differ for MIGRATE_CMA pageblocks. Those can be used as
1773                  * fallback only via special __rmqueue_cma_fallback() function
1774                  */
1775                 set_pcppage_migratetype(page, start_migratetype);
1776 
1777                 trace_mm_page_alloc_extfrag(page, order, current_order,
1778                         start_migratetype, fallback_mt);
1779 
1780                 return page;
1781         }
1782 
1783         return NULL;
1784 }
1785 
1786 /*
1787  * Do the hard work of removing an element from the buddy allocator.
1788  * Call me with the zone->lock already held.
1789  */
1790 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1791                                 int migratetype, gfp_t gfp_flags)
1792 {
1793         struct page *page;
1794 
1795         page = __rmqueue_smallest(zone, order, migratetype);
1796         if (unlikely(!page)) {
1797                 if (migratetype == MIGRATE_MOVABLE)
1798                         page = __rmqueue_cma_fallback(zone, order);
1799 
1800                 if (!page)
1801                         page = __rmqueue_fallback(zone, order, migratetype);
1802         }
1803 
1804         trace_mm_page_alloc_zone_locked(page, order, migratetype);
1805         return page;
1806 }
1807 
1808 /*
1809  * Obtain a specified number of elements from the buddy allocator, all under
1810  * a single hold of the lock, for efficiency.  Add them to the supplied list.
1811  * Returns the number of new pages which were placed at *list.
1812  */
1813 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1814                         unsigned long count, struct list_head *list,
1815                         int migratetype, bool cold)
1816 {
1817         int i;
1818 
1819         spin_lock(&zone->lock);
1820         for (i = 0; i < count; ++i) {
1821                 struct page *page = __rmqueue(zone, order, migratetype, 0);
1822                 if (unlikely(page == NULL))
1823                         break;
1824 
1825                 /*
1826                  * Split buddy pages returned by expand() are received here
1827                  * in physical page order. The page is added to the callers and
1828                  * list and the list head then moves forward. From the callers
1829                  * perspective, the linked list is ordered by page number in
1830                  * some conditions. This is useful for IO devices that can
1831                  * merge IO requests if the physical pages are ordered
1832                  * properly.
1833                  */
1834                 if (likely(!cold))
1835                         list_add(&page->lru, list);
1836                 else
1837                         list_add_tail(&page->lru, list);
1838                 list = &page->lru;
1839                 if (is_migrate_cma(get_pcppage_migratetype(page)))
1840                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1841                                               -(1 << order));
1842         }
1843         __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1844         spin_unlock(&zone->lock);
1845         return i;
1846 }
1847 
1848 #ifdef CONFIG_NUMA
1849 /*
1850  * Called from the vmstat counter updater to drain pagesets of this
1851  * currently executing processor on remote nodes after they have
1852  * expired.
1853  *
1854  * Note that this function must be called with the thread pinned to
1855  * a single processor.
1856  */
1857 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1858 {
1859         unsigned long flags;
1860         int to_drain, batch;
1861 
1862         local_irq_save(flags);
1863         batch = READ_ONCE(pcp->batch);
1864         to_drain = min(pcp->count, batch);
1865         if (to_drain > 0) {
1866                 free_pcppages_bulk(zone, to_drain, pcp);
1867                 pcp->count -= to_drain;
1868         }
1869         local_irq_restore(flags);
1870 }
1871 #endif
1872 
1873 /*
1874  * Drain pcplists of the indicated processor and zone.
1875  *
1876  * The processor must either be the current processor and the
1877  * thread pinned to the current processor or a processor that
1878  * is not online.
1879  */
1880 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
1881 {
1882         unsigned long flags;
1883         struct per_cpu_pageset *pset;
1884         struct per_cpu_pages *pcp;
1885 
1886         local_irq_save(flags);
1887         pset = per_cpu_ptr(zone->pageset, cpu);
1888 
1889         pcp = &pset->pcp;
1890         if (pcp->count) {
1891                 free_pcppages_bulk(zone, pcp->count, pcp);
1892                 pcp->count = 0;
1893         }
1894         local_irq_restore(flags);
1895 }
1896 
1897 /*
1898  * Drain pcplists of all zones on the indicated processor.
1899  *
1900  * The processor must either be the current processor and the
1901  * thread pinned to the current processor or a processor that
1902  * is not online.
1903  */
1904 static void drain_pages(unsigned int cpu)
1905 {
1906         struct zone *zone;
1907 
1908         for_each_populated_zone(zone) {
1909                 drain_pages_zone(cpu, zone);
1910         }
1911 }
1912 
1913 /*
1914  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1915  *
1916  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1917  * the single zone's pages.
1918  */
1919 void drain_local_pages(struct zone *zone)
1920 {
1921         int cpu = smp_processor_id();
1922 
1923         if (zone)
1924                 drain_pages_zone(cpu, zone);
1925         else
1926                 drain_pages(cpu);
1927 }
1928 
1929 /*
1930  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1931  *
1932  * When zone parameter is non-NULL, spill just the single zone's pages.
1933  *
1934  * Note that this code is protected against sending an IPI to an offline
1935  * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1936  * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1937  * nothing keeps CPUs from showing up after we populated the cpumask and
1938  * before the call to on_each_cpu_mask().
1939  */
1940 void drain_all_pages(struct zone *zone)
1941 {
1942         int cpu;
1943 
1944         /*
1945          * Allocate in the BSS so we wont require allocation in
1946          * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1947          */
1948         static cpumask_t cpus_with_pcps;
1949 
1950         /*
1951          * We don't care about racing with CPU hotplug event
1952          * as offline notification will cause the notified
1953          * cpu to drain that CPU pcps and on_each_cpu_mask
1954          * disables preemption as part of its processing
1955          */
1956         for_each_online_cpu(cpu) {
1957                 struct per_cpu_pageset *pcp;
1958                 struct zone *z;
1959                 bool has_pcps = false;
1960 
1961                 if (zone) {
1962                         pcp = per_cpu_ptr(zone->pageset, cpu);
1963                         if (pcp->pcp.count)
1964                                 has_pcps = true;
1965                 } else {
1966                         for_each_populated_zone(z) {
1967                                 pcp = per_cpu_ptr(z->pageset, cpu);
1968                                 if (pcp->pcp.count) {
1969                                         has_pcps = true;
1970                                         break;
1971                                 }
1972                         }
1973                 }
1974 
1975                 if (has_pcps)
1976                         cpumask_set_cpu(cpu, &cpus_with_pcps);
1977                 else
1978                         cpumask_clear_cpu(cpu, &cpus_with_pcps);
1979         }
1980         on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
1981                                                                 zone, 1);
1982 }
1983 
1984 #ifdef CONFIG_HIBERNATION
1985 
1986 void mark_free_pages(struct zone *zone)
1987 {
1988         unsigned long pfn, max_zone_pfn;
1989         unsigned long flags;
1990         unsigned int order, t;
1991         struct list_head *curr;
1992 
1993         if (zone_is_empty(zone))
1994                 return;
1995 
1996         spin_lock_irqsave(&zone->lock, flags);
1997 
1998         max_zone_pfn = zone_end_pfn(zone);
1999         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2000                 if (pfn_valid(pfn)) {
2001                         struct page *page = pfn_to_page(pfn);
2002 
2003                         if (!swsusp_page_is_forbidden(page))
2004                                 swsusp_unset_page_free(page);
2005                 }
2006 
2007         for_each_migratetype_order(order, t) {
2008                 list_for_each(curr, &zone->free_area[order].free_list[t]) {
2009                         unsigned long i;
2010 
2011                         pfn = page_to_pfn(list_entry(curr, struct page, lru));
2012                         for (i = 0; i < (1UL << order); i++)
2013                                 swsusp_set_page_free(pfn_to_page(pfn + i));
2014                 }
2015         }
2016         spin_unlock_irqrestore(&zone->lock, flags);
2017 }
2018 #endif /* CONFIG_PM */
2019 
2020 /*
2021  * Free a 0-order page
2022  * cold == true ? free a cold page : free a hot page
2023  */
2024 void free_hot_cold_page(struct page *page, bool cold)
2025 {
2026         struct zone *zone = page_zone(page);
2027         struct per_cpu_pages *pcp;
2028         unsigned long flags;
2029         unsigned long pfn = page_to_pfn(page);
2030         int migratetype;
2031 
2032         if (!free_pages_prepare(page, 0))
2033                 return;
2034 
2035         migratetype = get_pfnblock_migratetype(page, pfn);
2036         set_pcppage_migratetype(page, migratetype);
2037         local_irq_save(flags);
2038         __count_vm_event(PGFREE);
2039 
2040         /*
2041          * We only track unmovable, reclaimable and movable on pcp lists.
2042          * Free ISOLATE pages back to the allocator because they are being
2043          * offlined but treat RESERVE as movable pages so we can get those
2044          * areas back if necessary. Otherwise, we may have to free
2045          * excessively into the page allocator
2046          */
2047         if (migratetype >= MIGRATE_PCPTYPES) {
2048                 if (unlikely(is_migrate_isolate(migratetype))) {
2049                         free_one_page(zone, page, pfn, 0, migratetype);
2050                         goto out;
2051                 }
2052                 migratetype = MIGRATE_MOVABLE;
2053         }
2054 
2055         pcp = &this_cpu_ptr(zone->pageset)->pcp;
2056         if (!cold)
2057                 list_add(&page->lru, &pcp->lists[migratetype]);
2058         else
2059                 list_add_tail(&page->lru, &pcp->lists[migratetype]);
2060         pcp->count++;
2061         if (pcp->count >= pcp->high) {
2062                 unsigned long batch = READ_ONCE(pcp->batch);
2063                 free_pcppages_bulk(zone, batch, pcp);
2064                 pcp->count -= batch;
2065         }
2066 
2067 out:
2068         local_irq_restore(flags);
2069 }
2070 
2071 /*
2072  * Free a list of 0-order pages
2073  */
2074 void free_hot_cold_page_list(struct list_head *list, bool cold)
2075 {
2076         struct page *page, *next;
2077 
2078         list_for_each_entry_safe(page, next, list, lru) {
2079                 trace_mm_page_free_batched(page, cold);
2080                 free_hot_cold_page(page, cold);
2081         }
2082 }
2083 
2084 /*
2085  * split_page takes a non-compound higher-order page, and splits it into
2086  * n (1<<order) sub-pages: page[0..n]
2087  * Each sub-page must be freed individually.
2088  *
2089  * Note: this is probably too low level an operation for use in drivers.
2090  * Please consult with lkml before using this in your driver.
2091  */
2092 void split_page(struct page *page, unsigned int order)
2093 {
2094         int i;
2095         gfp_t gfp_mask;
2096 
2097         VM_BUG_ON_PAGE(PageCompound(page), page);
2098         VM_BUG_ON_PAGE(!page_count(page), page);
2099 
2100 #ifdef CONFIG_KMEMCHECK
2101         /*
2102          * Split shadow pages too, because free(page[0]) would
2103          * otherwise free the whole shadow.
2104          */
2105         if (kmemcheck_page_is_tracked(page))
2106                 split_page(virt_to_page(page[0].shadow), order);
2107 #endif
2108 
2109         gfp_mask = get_page_owner_gfp(page);
2110         set_page_owner(page, 0, gfp_mask);
2111         for (i = 1; i < (1 << order); i++) {
2112                 set_page_refcounted(page + i);
2113                 set_page_owner(page + i, 0, gfp_mask);
2114         }
2115 }
2116 EXPORT_SYMBOL_GPL(split_page);
2117 
2118 int __isolate_free_page(struct page *page, unsigned int order)
2119 {
2120         unsigned long watermark;
2121         struct zone *zone;
2122         int mt;
2123 
2124         BUG_ON(!PageBuddy(page));
2125 
2126         zone = page_zone(page);
2127         mt = get_pageblock_migratetype(page);
2128 
2129         if (!is_migrate_isolate(mt)) {
2130                 /* Obey watermarks as if the page was being allocated */
2131                 watermark = low_wmark_pages(zone) + (1 << order);
2132                 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2133                         return 0;
2134 
2135                 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2136         }
2137 
2138         /* Remove page from free list */
2139         list_del(&page->lru);
2140         zone->free_area[order].nr_free--;
2141         rmv_page_order(page);
2142 
2143         set_page_owner(page, order, __GFP_MOVABLE);
2144 
2145         /* Set the pageblock if the isolated page is at least a pageblock */
2146         if (order >= pageblock_order - 1) {
2147                 struct page *endpage = page + (1 << order) - 1;
2148                 for (; page < endpage; page += pageblock_nr_pages) {
2149                         int mt = get_pageblock_migratetype(page);
2150                         if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
2151                                 set_pageblock_migratetype(page,
2152                                                           MIGRATE_MOVABLE);
2153                 }
2154         }
2155 
2156 
2157         return 1UL << order;
2158 }
2159 
2160 /*
2161  * Similar to split_page except the page is already free. As this is only
2162  * being used for migration, the migratetype of the block also changes.
2163  * As this is called with interrupts disabled, the caller is responsible
2164  * for calling arch_alloc_page() and kernel_map_page() after interrupts
2165  * are enabled.
2166  *
2167  * Note: this is probably too low level an operation for use in drivers.
2168  * Please consult with lkml before using this in your driver.
2169  */
2170 int split_free_page(struct page *page)
2171 {
2172         unsigned int order;
2173         int nr_pages;
2174 
2175         order = page_order(page);
2176 
2177         nr_pages = __isolate_free_page(page, order);
2178         if (!nr_pages)
2179                 return 0;
2180 
2181         /* Split into individual pages */
2182         set_page_refcounted(page);
2183         split_page(page, order);
2184         return nr_pages;
2185 }
2186 
2187 /*
2188  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2189  */
2190 static inline
2191 struct page *buffered_rmqueue(struct zone *preferred_zone,
2192                         struct zone *zone, unsigned int order,
2193                         gfp_t gfp_flags, int alloc_flags, int migratetype)
2194 {
2195         unsigned long flags;
2196         struct page *page;
2197         bool cold = ((gfp_flags & __GFP_COLD) != 0);
2198 
2199         if (likely(order == 0)) {
2200                 struct per_cpu_pages *pcp;
2201                 struct list_head *list;
2202 
2203                 local_irq_save(flags);
2204                 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2205                 list = &pcp->lists[migratetype];
2206                 if (list_empty(list)) {
2207                         pcp->count += rmqueue_bulk(zone, 0,
2208                                         pcp->batch, list,
2209                                         migratetype, cold);
2210                         if (unlikely(list_empty(list)))
2211                                 goto failed;
2212                 }
2213 
2214                 if (cold)
2215                         page = list_entry(list->prev, struct page, lru);
2216                 else
2217                         page = list_entry(list->next, struct page, lru);
2218 
2219                 list_del(&page->lru);
2220                 pcp->count--;
2221         } else {
2222                 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
2223                         /*
2224                          * __GFP_NOFAIL is not to be used in new code.
2225                          *
2226                          * All __GFP_NOFAIL callers should be fixed so that they
2227                          * properly detect and handle allocation failures.
2228                          *
2229                          * We most definitely don't want callers attempting to
2230                          * allocate greater than order-1 page units with
2231                          * __GFP_NOFAIL.
2232                          */
2233                         WARN_ON_ONCE(order > 1);
2234                 }
2235                 spin_lock_irqsave(&zone->lock, flags);
2236 
2237                 page = NULL;
2238                 if (alloc_flags & ALLOC_HARDER) {
2239                         page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2240                         if (page)
2241                                 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2242                 }
2243                 if (!page)
2244                         page = __rmqueue(zone, order, migratetype, gfp_flags);
2245                 spin_unlock(&zone->lock);
2246                 if (!page)
2247                         goto failed;
2248                 __mod_zone_freepage_state(zone, -(1 << order),
2249                                           get_pcppage_migratetype(page));
2250         }
2251 
2252         __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
2253         if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
2254             !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2255                 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2256 
2257         __count_zone_vm_events(PGALLOC, zone, 1 << order);
2258         zone_statistics(preferred_zone, zone, gfp_flags);
2259         local_irq_restore(flags);
2260 
2261         VM_BUG_ON_PAGE(bad_range(zone, page), page);
2262         return page;
2263 
2264 failed:
2265         local_irq_restore(flags);
2266         return NULL;
2267 }
2268 
2269 #ifdef CONFIG_FAIL_PAGE_ALLOC
2270 
2271 static struct {
2272         struct fault_attr attr;
2273 
2274         bool ignore_gfp_highmem;
2275         bool ignore_gfp_reclaim;
2276         u32 min_order;
2277 } fail_page_alloc = {
2278         .attr = FAULT_ATTR_INITIALIZER,
2279         .ignore_gfp_reclaim = true,
2280         .ignore_gfp_highmem = true,
2281         .min_order = 1,
2282 };
2283 
2284 static int __init setup_fail_page_alloc(char *str)
2285 {
2286         return setup_fault_attr(&fail_page_alloc.attr, str);
2287 }
2288 __setup("fail_page_alloc=", setup_fail_page_alloc);
2289 
2290 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2291 {
2292         if (order < fail_page_alloc.min_order)
2293                 return false;
2294         if (gfp_mask & __GFP_NOFAIL)
2295                 return false;
2296         if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
2297                 return false;
2298         if (fail_page_alloc.ignore_gfp_reclaim &&
2299                         (gfp_mask & __GFP_DIRECT_RECLAIM))
2300                 return false;
2301 
2302         return should_fail(&fail_page_alloc.attr, 1 << order);
2303 }
2304 
2305 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2306 
2307 static int __init fail_page_alloc_debugfs(void)
2308 {
2309         umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
2310         struct dentry *dir;
2311 
2312         dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2313                                         &fail_page_alloc.attr);
2314         if (IS_ERR(dir))
2315                 return PTR_ERR(dir);
2316 
2317         if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2318                                 &fail_page_alloc.ignore_gfp_reclaim))
2319                 goto fail;
2320         if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2321                                 &fail_page_alloc.ignore_gfp_highmem))
2322                 goto fail;
2323         if (!debugfs_create_u32("min-order", mode, dir,
2324                                 &fail_page_alloc.min_order))
2325                 goto fail;
2326 
2327         return 0;
2328 fail:
2329         debugfs_remove_recursive(dir);
2330 
2331         return -ENOMEM;
2332 }
2333 
2334 late_initcall(fail_page_alloc_debugfs);
2335 
2336 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2337 
2338 #else /* CONFIG_FAIL_PAGE_ALLOC */
2339 
2340 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2341 {
2342         return false;
2343 }
2344 
2345 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2346 
2347 /*
2348  * Return true if free base pages are above 'mark'. For high-order checks it
2349  * will return true of the order-0 watermark is reached and there is at least
2350  * one free page of a suitable size. Checking now avoids taking the zone lock
2351  * to check in the allocation paths if no pages are free.
2352  */
2353 static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2354                         unsigned long mark, int classzone_idx, int alloc_flags,
2355                         long free_pages)
2356 {
2357         long min = mark;
2358         int o;
2359         const int alloc_harder = (alloc_flags & ALLOC_HARDER);
2360 
2361         /* free_pages may go negative - that's OK */
2362         free_pages -= (1 << order) - 1;
2363 
2364         if (alloc_flags & ALLOC_HIGH)
2365                 min -= min / 2;
2366 
2367         /*
2368          * If the caller does not have rights to ALLOC_HARDER then subtract
2369          * the high-atomic reserves. This will over-estimate the size of the
2370          * atomic reserve but it avoids a search.
2371          */
2372         if (likely(!alloc_harder))
2373                 free_pages -= z->nr_reserved_highatomic;
2374         else
2375                 min -= min / 4;
2376 
2377 #ifdef CONFIG_CMA
2378         /* If allocation can't use CMA areas don't use free CMA pages */
2379         if (!(alloc_flags & ALLOC_CMA))
2380                 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
2381 #endif
2382 
2383         /*
2384          * Check watermarks for an order-0 allocation request. If these
2385          * are not met, then a high-order request also cannot go ahead
2386          * even if a suitable page happened to be free.
2387          */
2388         if (free_pages <= min + z->lowmem_reserve[classzone_idx])
2389                 return false;
2390 
2391         /* If this is an order-0 request then the watermark is fine */
2392         if (!order)
2393                 return true;
2394 
2395         /* For a high-order request, check at least one suitable page is free */
2396         for (o = order; o < MAX_ORDER; o++) {
2397                 struct free_area *area = &z->free_area[o];
2398                 int mt;
2399 
2400                 if (!area->nr_free)
2401                         continue;
2402 
2403                 if (alloc_harder)
2404                         return true;
2405 
2406                 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2407                         if (!list_empty(&area->free_list[mt]))
2408                                 return true;
2409                 }
2410 
2411 #ifdef CONFIG_CMA
2412                 if ((alloc_flags & ALLOC_CMA) &&
2413                     !list_empty(&area->free_list[MIGRATE_CMA])) {
2414                         return true;
2415                 }
2416 #endif
2417         }
2418         return false;
2419 }
2420 
2421 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2422                       int classzone_idx, int alloc_flags)
2423 {
2424         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2425                                         zone_page_state(z, NR_FREE_PAGES));
2426 }
2427 
2428 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2429                         unsigned long mark, int classzone_idx)
2430 {
2431         long free_pages = zone_page_state(z, NR_FREE_PAGES);
2432 
2433         if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2434                 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2435 
2436         return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
2437                                                                 free_pages);
2438 }
2439 
2440 #ifdef CONFIG_NUMA
2441 static bool zone_local(struct zone *local_zone, struct zone *zone)
2442 {
2443         return local_zone->node == zone->node;
2444 }
2445 
2446 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2447 {
2448         return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2449                                 RECLAIM_DISTANCE;
2450 }
2451 #else   /* CONFIG_NUMA */
2452 static bool zone_local(struct zone *local_zone, struct zone *zone)
2453 {
2454         return true;
2455 }
2456 
2457 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2458 {
2459         return true;
2460 }
2461 #endif  /* CONFIG_NUMA */
2462 
2463 static void reset_alloc_batches(struct zone *preferred_zone)
2464 {
2465         struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2466 
2467         do {
2468                 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2469                         high_wmark_pages(zone) - low_wmark_pages(zone) -
2470                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
2471                 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2472         } while (zone++ != preferred_zone);
2473 }
2474 
2475 /*
2476  * get_page_from_freelist goes through the zonelist trying to allocate
2477  * a page.
2478  */
2479 static struct page *
2480 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2481                                                 const struct alloc_context *ac)
2482 {
2483         struct zonelist *zonelist = ac->zonelist;
2484         struct zoneref *z;
2485         struct page *page = NULL;
2486         struct zone *zone;
2487         int nr_fair_skipped = 0;
2488         bool zonelist_rescan;
2489 
2490 zonelist_scan:
2491         zonelist_rescan = false;
2492 
2493         /*
2494          * Scan zonelist, looking for a zone with enough free.
2495          * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2496          */
2497         for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2498                                                                 ac->nodemask) {
2499                 unsigned long mark;
2500 
2501                 if (cpusets_enabled() &&
2502                         (alloc_flags & ALLOC_CPUSET) &&
2503                         !cpuset_zone_allowed(zone, gfp_mask))
2504                                 continue;
2505                 /*
2506                  * Distribute pages in proportion to the individual
2507                  * zone size to ensure fair page aging.  The zone a
2508                  * page was allocated in should have no effect on the
2509                  * time the page has in memory before being reclaimed.
2510                  */
2511                 if (alloc_flags & ALLOC_FAIR) {
2512                         if (!zone_local(ac->preferred_zone, zone))
2513                                 break;
2514                         if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
2515                                 nr_fair_skipped++;
2516                                 continue;
2517                         }
2518                 }
2519                 /*
2520                  * When allocating a page cache page for writing, we
2521                  * want to get it from a zone that is within its dirty
2522                  * limit, such that no single zone holds more than its
2523                  * proportional share of globally allowed dirty pages.
2524                  * The dirty limits take into account the zone's
2525                  * lowmem reserves and high watermark so that kswapd
2526                  * should be able to balance it without having to
2527                  * write pages from its LRU list.
2528                  *
2529                  * This may look like it could increase pressure on
2530                  * lower zones by failing allocations in higher zones
2531                  * before they are full.  But the pages that do spill
2532                  * over are limited as the lower zones are protected
2533                  * by this very same mechanism.  It should not become
2534                  * a practical burden to them.
2535                  *
2536                  * XXX: For now, allow allocations to potentially
2537                  * exceed the per-zone dirty limit in the slowpath
2538                  * (spread_dirty_pages unset) before going into reclaim,
2539                  * which is important when on a NUMA setup the allowed
2540                  * zones are together not big enough to reach the
2541                  * global limit.  The proper fix for these situations
2542                  * will require awareness of zones in the
2543                  * dirty-throttling and the flusher threads.
2544                  */
2545                 if (ac->spread_dirty_pages && !zone_dirty_ok(zone))
2546                         continue;
2547 
2548                 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2549                 if (!zone_watermark_ok(zone, order, mark,
2550                                        ac->classzone_idx, alloc_flags)) {
2551                         int ret;
2552 
2553                         /* Checked here to keep the fast path fast */
2554                         BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2555                         if (alloc_flags & ALLOC_NO_WATERMARKS)
2556                                 goto try_this_zone;
2557 
2558                         if (zone_reclaim_mode == 0 ||
2559                             !zone_allows_reclaim(ac->preferred_zone, zone))
2560                                 continue;
2561 
2562                         ret = zone_reclaim(zone, gfp_mask, order);
2563                         switch (ret) {
2564                         case ZONE_RECLAIM_NOSCAN:
2565                                 /* did not scan */
2566                                 continue;
2567                         case ZONE_RECLAIM_FULL:
2568                                 /* scanned but unreclaimable */
2569                                 continue;
2570                         default:
2571                                 /* did we reclaim enough */
2572                                 if (zone_watermark_ok(zone, order, mark,
2573                                                 ac->classzone_idx, alloc_flags))
2574                                         goto try_this_zone;
2575 
2576                                 continue;
2577                         }
2578                 }
2579 
2580 try_this_zone:
2581                 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2582                                 gfp_mask, alloc_flags, ac->migratetype);
2583                 if (page) {
2584                         if (prep_new_page(page, order, gfp_mask, alloc_flags))
2585                                 goto try_this_zone;
2586 
2587                         /*
2588                          * If this is a high-order atomic allocation then check
2589                          * if the pageblock should be reserved for the future
2590                          */
2591                         if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2592                                 reserve_highatomic_pageblock(page, zone, order);
2593 
2594                         return page;
2595                 }
2596         }
2597 
2598         /*
2599          * The first pass makes sure allocations are spread fairly within the
2600          * local node.  However, the local node might have free pages left
2601          * after the fairness batches are exhausted, and remote zones haven't
2602          * even been considered yet.  Try once more without fairness, and
2603          * include remote zones now, before entering the slowpath and waking
2604          * kswapd: prefer spilling to a remote zone over swapping locally.
2605          */
2606         if (alloc_flags & ALLOC_FAIR) {
2607                 alloc_flags &= ~ALLOC_FAIR;
2608                 if (nr_fair_skipped) {
2609                         zonelist_rescan = true;
2610                         reset_alloc_batches(ac->preferred_zone);
2611                 }
2612                 if (nr_online_nodes > 1)
2613                         zonelist_rescan = true;
2614         }
2615 
2616         if (zonelist_rescan)
2617                 goto zonelist_scan;
2618 
2619         return NULL;
2620 }
2621 
2622 /*
2623  * Large machines with many possible nodes should not always dump per-node
2624  * meminfo in irq context.
2625  */
2626 static inline bool should_suppress_show_mem(void)
2627 {
2628         bool ret = false;
2629 
2630 #if NODES_SHIFT > 8
2631         ret = in_interrupt();
2632 #endif
2633         return ret;
2634 }
2635 
2636 static DEFINE_RATELIMIT_STATE(nopage_rs,
2637                 DEFAULT_RATELIMIT_INTERVAL,
2638                 DEFAULT_RATELIMIT_BURST);
2639 
2640 void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
2641 {
2642         unsigned int filter = SHOW_MEM_FILTER_NODES;
2643 
2644         if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2645             debug_guardpage_minorder() > 0)
2646                 return;
2647 
2648         /*
2649          * This documents exceptions given to allocations in certain
2650          * contexts that are allowed to allocate outside current's set
2651          * of allowed nodes.
2652          */
2653         if (!(gfp_mask & __GFP_NOMEMALLOC))
2654                 if (test_thread_flag(TIF_MEMDIE) ||
2655                     (current->flags & (PF_MEMALLOC | PF_EXITING)))
2656                         filter &= ~SHOW_MEM_FILTER_NODES;
2657         if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
2658                 filter &= ~SHOW_MEM_FILTER_NODES;
2659 
2660         if (fmt) {
2661                 struct va_format vaf;
2662                 va_list args;
2663 
2664                 va_start(args, fmt);
2665 
2666                 vaf.fmt = fmt;
2667                 vaf.va = &args;
2668 
2669                 pr_warn("%pV", &vaf);
2670 
2671                 va_end(args);
2672         }
2673 
2674         pr_warn("%s: page allocation failure: order:%u, mode:0x%x\n",
2675                 current->comm, order, gfp_mask);
2676 
2677         dump_stack();
2678         if (!should_suppress_show_mem())
2679                 show_mem(filter);
2680 }
2681 
2682 static inline struct page *
2683 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2684         const struct alloc_context *ac, unsigned long *did_some_progress)
2685 {
2686         struct oom_control oc = {
2687                 .zonelist = ac->zonelist,
2688                 .nodemask = ac->nodemask,
2689                 .gfp_mask = gfp_mask,
2690                 .order = order,
2691         };
2692         struct page *page;
2693 
2694         *did_some_progress = 0;
2695 
2696         /*
2697          * Acquire the oom lock.  If that fails, somebody else is
2698          * making progress for us.
2699          */
2700         if (!mutex_trylock(&oom_lock)) {
2701                 *did_some_progress = 1;
2702                 schedule_timeout_uninterruptible(1);
2703                 return NULL;
2704         }
2705 
2706         /*
2707          * Go through the zonelist yet one more time, keep very high watermark
2708          * here, this is only to catch a parallel oom killing, we must fail if
2709          * we're still under heavy pressure.
2710          */
2711         page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2712                                         ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
2713         if (page)
2714                 goto out;
2715 
2716         if (!(gfp_mask & __GFP_NOFAIL)) {
2717                 /* Coredumps can quickly deplete all memory reserves */
2718                 if (current->flags & PF_DUMPCORE)
2719                         goto out;
2720                 /* The OOM killer will not help higher order allocs */
2721                 if (order > PAGE_ALLOC_COSTLY_ORDER)
2722                         goto out;
2723                 /* The OOM killer does not needlessly kill tasks for lowmem */
2724                 if (ac->high_zoneidx < ZONE_NORMAL)
2725                         goto out;
2726                 /* The OOM killer does not compensate for IO-less reclaim */
2727                 if (!(gfp_mask & __GFP_FS)) {
2728                         /*
2729                          * XXX: Page reclaim didn't yield anything,
2730                          * and the OOM killer can't be invoked, but
2731                          * keep looping as per tradition.
2732                          */
2733                         *did_some_progress = 1;
2734                         goto out;
2735                 }
2736                 if (pm_suspended_storage())
2737                         goto out;
2738                 /* The OOM killer may not free memory on a specific node */
2739                 if (gfp_mask & __GFP_THISNODE)
2740                         goto out;
2741         }
2742         /* Exhausted what can be done so it's blamo time */
2743         if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL))
2744                 *did_some_progress = 1;
2745 out:
2746         mutex_unlock(&oom_lock);
2747         return page;
2748 }
2749 
2750 #ifdef CONFIG_COMPACTION
2751 /* Try memory compaction for high-order allocations before reclaim */
2752 static struct page *
2753 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2754                 int alloc_flags, const struct alloc_context *ac,
2755                 enum migrate_mode mode, int *contended_compaction,
2756                 bool *deferred_compaction)
2757 {
2758         unsigned long compact_result;
2759         struct page *page;
2760 
2761         if (!order)
2762                 return NULL;
2763 
2764         current->flags |= PF_MEMALLOC;
2765         compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2766                                                 mode, contended_compaction);
2767         current->flags &= ~PF_MEMALLOC;
2768 
2769         switch (compact_result) {
2770         case COMPACT_DEFERRED:
2771                 *deferred_compaction = true;
2772                 /* fall-through */
2773         case COMPACT_SKIPPED:
2774                 return NULL;
2775         default:
2776                 break;
2777         }
2778 
2779         /*
2780          * At least in one zone compaction wasn't deferred or skipped, so let's
2781          * count a compaction stall
2782          */
2783         count_vm_event(COMPACTSTALL);
2784 
2785         page = get_page_from_freelist(gfp_mask, order,
2786                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2787 
2788         if (page) {
2789                 struct zone *zone = page_zone(page);
2790 
2791                 zone->compact_blockskip_flush = false;
2792                 compaction_defer_reset(zone, order, true);
2793                 count_vm_event(COMPACTSUCCESS);
2794                 return page;
2795         }
2796 
2797         /*
2798          * It's bad if compaction run occurs and fails. The most likely reason
2799          * is that pages exist, but not enough to satisfy watermarks.
2800          */
2801         count_vm_event(COMPACTFAIL);
2802 
2803         cond_resched();
2804 
2805         return NULL;
2806 }
2807 #else
2808 static inline struct page *
2809 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2810                 int alloc_flags, const struct alloc_context *ac,
2811                 enum migrate_mode mode, int *contended_compaction,
2812                 bool *deferred_compaction)
2813 {
2814         return NULL;
2815 }
2816 #endif /* CONFIG_COMPACTION */
2817 
2818 /* Perform direct synchronous page reclaim */
2819 static int
2820 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
2821                                         const struct alloc_context *ac)
2822 {
2823         struct reclaim_state reclaim_state;
2824         int progress;
2825 
2826         cond_resched();
2827 
2828         /* We now go into synchronous reclaim */
2829         cpuset_memory_pressure_bump();
2830         current->flags |= PF_MEMALLOC;
2831         lockdep_set_current_reclaim_state(gfp_mask);
2832         reclaim_state.reclaimed_slab = 0;
2833         current->reclaim_state = &reclaim_state;
2834 
2835         progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
2836                                                                 ac->nodemask);
2837 
2838         current->reclaim_state = NULL;
2839         lockdep_clear_current_reclaim_state();
2840         current->flags &= ~PF_MEMALLOC;
2841 
2842         cond_resched();
2843 
2844         return progress;
2845 }
2846 
2847 /* The really slow allocator path where we enter direct reclaim */
2848 static inline struct page *
2849 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2850                 int alloc_flags, const struct alloc_context *ac,
2851                 unsigned long *did_some_progress)
2852 {
2853         struct page *page = NULL;
2854         bool drained = false;
2855 
2856         *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
2857         if (unlikely(!(*did_some_progress)))
2858                 return NULL;
2859 
2860 retry:
2861         page = get_page_from_freelist(gfp_mask, order,
2862                                         alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2863 
2864         /*
2865          * If an allocation failed after direct reclaim, it could be because
2866          * pages are pinned on the per-cpu lists or in high alloc reserves.
2867          * Shrink them them and try again
2868          */
2869         if (!page && !drained) {
2870                 unreserve_highatomic_pageblock(ac);
2871                 drain_all_pages(NULL);
2872                 drained = true;
2873                 goto retry;
2874         }
2875 
2876         return page;
2877 }
2878 
2879 /*
2880  * This is called in the allocator slow-path if the allocation request is of
2881  * sufficient urgency to ignore watermarks and take other desperate measures
2882  */
2883 static inline struct page *
2884 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2885                                 const struct alloc_context *ac)
2886 {
2887         struct page *page;
2888 
2889         do {
2890                 page = get_page_from_freelist(gfp_mask, order,
2891                                                 ALLOC_NO_WATERMARKS, ac);
2892 
2893                 if (!page && gfp_mask & __GFP_NOFAIL)
2894                         wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC,
2895                                                                         HZ/50);
2896         } while (!page && (gfp_mask & __GFP_NOFAIL));
2897 
2898         return page;
2899 }
2900 
2901 static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
2902 {
2903         struct zoneref *z;
2904         struct zone *zone;
2905 
2906         for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
2907                                                 ac->high_zoneidx, ac->nodemask)
2908                 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
2909 }
2910 
2911 static inline int
2912 gfp_to_alloc_flags(gfp_t gfp_mask)
2913 {
2914         int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2915 
2916         /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2917         BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2918 
2919         /*
2920          * The caller may dip into page reserves a bit more if the caller
2921          * cannot run direct reclaim, or if the caller has realtime scheduling
2922          * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
2923          * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
2924          */
2925         alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2926 
2927         if (gfp_mask & __GFP_ATOMIC) {
2928                 /*
2929                  * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2930                  * if it can't schedule.
2931                  */
2932                 if (!(gfp_mask & __GFP_NOMEMALLOC))
2933                         alloc_flags |= ALLOC_HARDER;
2934                 /*
2935                  * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2936                  * comment for __cpuset_node_allowed().
2937                  */
2938                 alloc_flags &= ~ALLOC_CPUSET;
2939         } else if (unlikely(rt_task(current)) && !in_interrupt())
2940                 alloc_flags |= ALLOC_HARDER;
2941 
2942         if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2943                 if (gfp_mask & __GFP_MEMALLOC)
2944                         alloc_flags |= ALLOC_NO_WATERMARKS;
2945                 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2946                         alloc_flags |= ALLOC_NO_WATERMARKS;
2947                 else if (!in_interrupt() &&
2948                                 ((current->flags & PF_MEMALLOC) ||
2949                                  unlikely(test_thread_flag(TIF_MEMDIE))))
2950                         alloc_flags |= ALLOC_NO_WATERMARKS;
2951         }
2952 #ifdef CONFIG_CMA
2953         if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2954                 alloc_flags |= ALLOC_CMA;
2955 #endif
2956         return alloc_flags;
2957 }
2958 
2959 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2960 {
2961         return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
2962 }
2963 
2964 static inline bool is_thp_gfp_mask(gfp_t gfp_mask)
2965 {
2966         return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE;
2967 }
2968 
2969 static inline struct page *
2970 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2971                                                 struct alloc_context *ac)
2972 {
2973         bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
2974         struct page *page = NULL;
2975         int alloc_flags;
2976         unsigned long pages_reclaimed = 0;
2977         unsigned long did_some_progress;
2978         enum migrate_mode migration_mode = MIGRATE_ASYNC;
2979         bool deferred_compaction = false;
2980         int contended_compaction = COMPACT_CONTENDED_NONE;
2981 
2982         /*
2983          * In the slowpath, we sanity check order to avoid ever trying to
2984          * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2985          * be using allocators in order of preference for an area that is
2986          * too large.
2987          */
2988         if (order >= MAX_ORDER) {
2989                 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2990                 return NULL;
2991         }
2992 
2993         /*
2994          * We also sanity check to catch abuse of atomic reserves being used by
2995          * callers that are not in atomic context.
2996          */
2997         if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
2998                                 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
2999                 gfp_mask &= ~__GFP_ATOMIC;
3000 
3001         /*
3002          * If this allocation cannot block and it is for a specific node, then
3003          * fail early.  There's no need to wakeup kswapd or retry for a
3004          * speculative node-specific allocation.
3005          */
3006         if (IS_ENABLED(CONFIG_NUMA) && (gfp_mask & __GFP_THISNODE) && !can_direct_reclaim)
3007                 goto nopage;
3008 
3009 retry:
3010         if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3011                 wake_all_kswapds(order, ac);
3012 
3013         /*
3014          * OK, we're below the kswapd watermark and have kicked background
3015          * reclaim. Now things get more complex, so set up alloc_flags according
3016          * to how we want to proceed.
3017          */
3018         alloc_flags = gfp_to_alloc_flags(gfp_mask);
3019 
3020         /*
3021          * Find the true preferred zone if the allocation is unconstrained by
3022          * cpusets.
3023          */
3024         if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
3025                 struct zoneref *preferred_zoneref;
3026                 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3027                                 ac->high_zoneidx, NULL, &ac->preferred_zone);
3028                 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
3029         }
3030 
3031         /* This is the last chance, in general, before the goto nopage. */
3032         page = get_page_from_freelist(gfp_mask, order,
3033                                 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3034         if (page)
3035                 goto got_pg;
3036 
3037         /* Allocate without watermarks if the context allows */
3038         if (alloc_flags & ALLOC_NO_WATERMARKS) {
3039                 /*
3040                  * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3041                  * the allocation is high priority and these type of
3042                  * allocations are system rather than user orientated
3043                  */
3044                 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3045 
3046                 page = __alloc_pages_high_priority(gfp_mask, order, ac);
3047 
3048                 if (page) {
3049                         goto got_pg;
3050                 }
3051         }
3052 
3053         /* Caller is not willing to reclaim, we can't balance anything */
3054         if (!can_direct_reclaim) {
3055                 /*
3056                  * All existing users of the deprecated __GFP_NOFAIL are
3057                  * blockable, so warn of any new users that actually allow this
3058                  * type of allocation to fail.
3059                  */
3060                 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
3061                 goto nopage;
3062         }
3063 
3064         /* Avoid recursion of direct reclaim */
3065         if (current->flags & PF_MEMALLOC)
3066                 goto nopage;
3067 
3068         /* Avoid allocations with no watermarks from looping endlessly */
3069         if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3070                 goto nopage;
3071 
3072         /*
3073          * Try direct compaction. The first pass is asynchronous. Subsequent
3074          * attempts after direct reclaim are synchronous
3075          */
3076         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3077                                         migration_mode,
3078                                         &contended_compaction,
3079                                         &deferred_compaction);
3080         if (page)
3081                 goto got_pg;
3082 
3083         /* Checks for THP-specific high-order allocations */
3084         if (is_thp_gfp_mask(gfp_mask)) {
3085                 /*
3086                  * If compaction is deferred for high-order allocations, it is
3087                  * because sync compaction recently failed. If this is the case
3088                  * and the caller requested a THP allocation, we do not want
3089                  * to heavily disrupt the system, so we fail the allocation
3090                  * instead of entering direct reclaim.
3091                  */
3092                 if (deferred_compaction)
3093                         goto nopage;
3094 
3095                 /*
3096                  * In all zones where compaction was attempted (and not
3097                  * deferred or skipped), lock contention has been detected.
3098                  * For THP allocation we do not want to disrupt the others
3099                  * so we fallback to base pages instead.
3100                  */
3101                 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3102                         goto nopage;
3103 
3104                 /*
3105                  * If compaction was aborted due to need_resched(), we do not
3106                  * want to further increase allocation latency, unless it is
3107                  * khugepaged trying to collapse.
3108                  */
3109                 if (contended_compaction == COMPACT_CONTENDED_SCHED
3110                         && !(current->flags & PF_KTHREAD))
3111                         goto nopage;
3112         }
3113 
3114         /*
3115          * It can become very expensive to allocate transparent hugepages at
3116          * fault, so use asynchronous memory compaction for THP unless it is
3117          * khugepaged trying to collapse.
3118          */
3119         if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD))
3120                 migration_mode = MIGRATE_SYNC_LIGHT;
3121 
3122         /* Try direct reclaim and then allocating */
3123         page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3124                                                         &did_some_progress);
3125         if (page)
3126                 goto got_pg;
3127 
3128         /* Do not loop if specifically requested */
3129         if (gfp_mask & __GFP_NORETRY)
3130                 goto noretry;
3131 
3132         /* Keep reclaiming pages as long as there is reasonable progress */
3133         pages_reclaimed += did_some_progress;
3134         if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3135             ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
3136                 /* Wait for some write requests to complete then retry */
3137                 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
3138                 goto retry;
3139         }
3140 
3141         /* Reclaim has failed us, start killing things */
3142         page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3143         if (page)
3144                 goto got_pg;
3145 
3146         /* Retry as long as the OOM killer is making progress */
3147         if (did_some_progress)
3148                 goto retry;
3149 
3150 noretry:
3151         /*
3152          * High-order allocations do not necessarily loop after
3153          * direct reclaim and reclaim/compaction depends on compaction
3154          * being called after reclaim so call directly if necessary
3155          */
3156         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3157                                             ac, migration_mode,
3158                                             &contended_compaction,
3159                                             &deferred_compaction);
3160         if (page)
3161                 goto got_pg;
3162 nopage:
3163         warn_alloc_failed(gfp_mask, order, NULL);
3164 got_pg:
3165         return page;
3166 }
3167 
3168 /*
3169  * This is the 'heart' of the zoned buddy allocator.
3170  */
3171 struct page *
3172 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3173                         struct zonelist *zonelist, nodemask_t *nodemask)
3174 {
3175         struct zoneref *preferred_zoneref;
3176         struct page *page = NULL;
3177         unsigned int cpuset_mems_cookie;
3178         int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
3179         gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
3180         struct alloc_context ac = {
3181                 .high_zoneidx = gfp_zone(gfp_mask),
3182                 .nodemask = nodemask,
3183                 .migratetype = gfpflags_to_migratetype(gfp_mask),
3184         };
3185 
3186         gfp_mask &= gfp_allowed_mask;
3187 
3188         lockdep_trace_alloc(gfp_mask);
3189 
3190         might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
3191 
3192         if (should_fail_alloc_page(gfp_mask, order))
3193                 return NULL;
3194 
3195         /*
3196          * Check the zones suitable for the gfp_mask contain at least one
3197          * valid zone. It's possible to have an empty zonelist as a result
3198          * of __GFP_THISNODE and a memoryless node
3199          */
3200         if (unlikely(!zonelist->_zonerefs->zone))
3201                 return NULL;
3202 
3203         if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
3204                 alloc_flags |= ALLOC_CMA;
3205 
3206 retry_cpuset:
3207         cpuset_mems_cookie = read_mems_allowed_begin();
3208 
3209         /* We set it here, as __alloc_pages_slowpath might have changed it */
3210         ac.zonelist = zonelist;
3211 
3212         /* Dirty zone balancing only done in the fast path */
3213         ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3214 
3215         /* The preferred zone is used for statistics later */
3216         preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3217                                 ac.nodemask ? : &cpuset_current_mems_allowed,
3218                                 &ac.preferred_zone);
3219         if (!ac.preferred_zone)
3220                 goto out;
3221         ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
3222 
3223         /* First allocation attempt */
3224         alloc_mask = gfp_mask|__GFP_HARDWALL;
3225         page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
3226         if (unlikely(!page)) {
3227                 /*
3228                  * Runtime PM, block IO and its error handling path
3229                  * can deadlock because I/O on the device might not
3230                  * complete.
3231                  */
3232                 alloc_mask = memalloc_noio_flags(gfp_mask);
3233                 ac.spread_dirty_pages = false;
3234 
3235                 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
3236         }
3237 
3238         if (kmemcheck_enabled && page)
3239                 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3240 
3241         trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3242 
3243 out:
3244         /*
3245          * When updating a task's mems_allowed, it is possible to race with
3246          * parallel threads in such a way that an allocation can fail while
3247          * the mask is being updated. If a page allocation is about to fail,
3248          * check if the cpuset changed during allocation and if so, retry.
3249          */
3250         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
3251                 goto retry_cpuset;
3252 
3253         return page;
3254 }
3255 EXPORT_SYMBOL(__alloc_pages_nodemask);
3256 
3257 /*
3258  * Common helper functions.
3259  */
3260 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
3261 {
3262         struct page *page;
3263 
3264         /*
3265          * __get_free_pages() returns a 32-bit address, which cannot represent
3266          * a highmem page
3267          */
3268         VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3269 
3270         page = alloc_pages(gfp_mask, order);
3271         if (!page)
3272                 return 0;
3273         return (unsigned long) page_address(page);
3274 }
3275 EXPORT_SYMBOL(__get_free_pages);
3276 
3277 unsigned long get_zeroed_page(gfp_t gfp_mask)
3278 {
3279         return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
3280 }
3281 EXPORT_SYMBOL(get_zeroed_page);
3282 
3283 void __free_pages(struct page *page, unsigned int order)
3284 {
3285         if (put_page_testzero(page)) {
3286                 if (order == 0)
3287                         free_hot_cold_page(page, false);
3288                 else
3289                         __free_pages_ok(page, order);
3290         }
3291 }
3292 
3293 EXPORT_SYMBOL(__free_pages);
3294 
3295 void free_pages(unsigned long addr, unsigned int order)
3296 {
3297         if (addr != 0) {
3298                 VM_BUG_ON(!virt_addr_valid((void *)addr));
3299                 __free_pages(virt_to_page((void *)addr), order);
3300         }
3301 }
3302 
3303 EXPORT_SYMBOL(free_pages);
3304 
3305 /*
3306  * Page Fragment:
3307  *  An arbitrary-length arbitrary-offset area of memory which resides
3308  *  within a 0 or higher order page.  Multiple fragments within that page
3309  *  are individually refcounted, in the page's reference counter.
3310  *
3311  * The page_frag functions below provide a simple allocation framework for
3312  * page fragments.  This is used by the network stack and network device
3313  * drivers to provide a backing region of memory for use as either an
3314  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3315  */
3316 static struct page *__page_frag_refill(struct page_frag_cache *nc,
3317                                        gfp_t gfp_mask)
3318 {
3319         struct page *page = NULL;
3320         gfp_t gfp = gfp_mask;
3321 
3322 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3323         gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3324                     __GFP_NOMEMALLOC;
3325         page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3326                                 PAGE_FRAG_CACHE_MAX_ORDER);
3327         nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3328 #endif
3329         if (unlikely(!page))
3330                 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3331 
3332         nc->va = page ? page_address(page) : NULL;
3333 
3334         return page;
3335 }
3336 
3337 void *__alloc_page_frag(struct page_frag_cache *nc,
3338                         unsigned int fragsz, gfp_t gfp_mask)
3339 {
3340         unsigned int size = PAGE_SIZE;
3341         struct page *page;
3342         int offset;
3343 
3344         if (unlikely(!nc->va)) {
3345 refill:
3346                 page = __page_frag_refill(nc, gfp_mask);
3347                 if (!page)
3348                         return NULL;
3349 
3350 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3351                 /* if size can vary use size else just use PAGE_SIZE */
3352                 size = nc->size;
3353 #endif
3354                 /* Even if we own the page, we do not use atomic_set().
3355                  * This would break get_page_unless_zero() users.
3356                  */
3357                 atomic_add(size - 1, &page->_count);
3358 
3359                 /* reset page count bias and offset to start of new frag */
3360                 nc->pfmemalloc = page_is_pfmemalloc(page);
3361                 nc->pagecnt_bias = size;
3362                 nc->offset = size;
3363         }
3364 
3365         offset = nc->offset - fragsz;
3366         if (unlikely(offset < 0)) {
3367                 page = virt_to_page(nc->va);
3368 
3369                 if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
3370                         goto refill;
3371 
3372 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3373                 /* if size can vary use size else just use PAGE_SIZE */
3374                 size = nc->size;
3375 #endif
3376                 /* OK, page count is 0, we can safely set it */
3377                 atomic_set(&page->_count, size);
3378 
3379                 /* reset page count bias and offset to start of new frag */
3380                 nc->pagecnt_bias = size;
3381                 offset = size - fragsz;
3382         }
3383 
3384         nc->pagecnt_bias--;
3385         nc->offset = offset;
3386 
3387         return nc->va + offset;
3388 }
3389 EXPORT_SYMBOL(__alloc_page_frag);
3390 
3391 /*
3392  * Frees a page fragment allocated out of either a compound or order 0 page.
3393  */
3394 void __free_page_frag(void *addr)
3395 {
3396         struct page *page = virt_to_head_page(addr);
3397 
3398         if (unlikely(put_page_testzero(page)))
3399                 __free_pages_ok(page, compound_order(page));
3400 }
3401 EXPORT_SYMBOL(__free_page_frag);
3402 
3403 /*
3404  * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3405  * of the current memory cgroup.
3406  *
3407  * It should be used when the caller would like to use kmalloc, but since the
3408  * allocation is large, it has to fall back to the page allocator.
3409  */
3410 struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3411 {
3412         struct page *page;
3413 
3414         page = alloc_pages(gfp_mask, order);
3415         if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3416                 __free_pages(page, order);
3417                 page = NULL;
3418         }
3419         return page;
3420 }
3421 
3422 struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3423 {
3424         struct page *page;
3425 
3426         page = alloc_pages_node(nid, gfp_mask, order);
3427         if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3428                 __free_pages(page, order);
3429                 page = NULL;
3430         }
3431         return page;
3432 }
3433 
3434 /*
3435  * __free_kmem_pages and free_kmem_pages will free pages allocated with
3436  * alloc_kmem_pages.
3437  */
3438 void __free_kmem_pages(struct page *page, unsigned int order)
3439 {
3440         memcg_kmem_uncharge(page, order);
3441         __free_pages(page, order);
3442 }
3443 
3444 void free_kmem_pages(unsigned long addr, unsigned int order)
3445 {
3446         if (addr != 0) {
3447                 VM_BUG_ON(!virt_addr_valid((void *)addr));
3448                 __free_kmem_pages(virt_to_page((void *)addr), order);
3449         }
3450 }
3451 
3452 static void *make_alloc_exact(unsigned long addr, unsigned int order,
3453                 size_t size)
3454 {
3455         if (addr) {
3456                 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3457                 unsigned long used = addr + PAGE_ALIGN(size);
3458 
3459                 split_page(virt_to_page((void *)addr), order);
3460                 while (used < alloc_end) {
3461                         free_page(used);
3462                         used += PAGE_SIZE;
3463                 }
3464         }
3465         return (void *)addr;
3466 }
3467 
3468 /**
3469  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3470  * @size: the number of bytes to allocate
3471  * @gfp_mask: GFP flags for the allocation
3472  *
3473  * This function is similar to alloc_pages(), except that it allocates the
3474  * minimum number of pages to satisfy the request.  alloc_pages() can only
3475  * allocate memory in power-of-two pages.
3476  *
3477  * This function is also limited by MAX_ORDER.
3478  *
3479  * Memory allocated by this function must be released by free_pages_exact().
3480  */
3481 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3482 {
3483         unsigned int order = get_order(size);
3484         unsigned long addr;
3485 
3486         addr = __get_free_pages(gfp_mask, order);
3487         return make_alloc_exact(addr, order, size);
3488 }
3489 EXPORT_SYMBOL(alloc_pages_exact);
3490 
3491 /**
3492  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3493  *                         pages on a node.
3494  * @nid: the preferred node ID where memory should be allocated
3495  * @size: the number of bytes to allocate
3496  * @gfp_mask: GFP flags for the allocation
3497  *
3498  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3499  * back.
3500  */
3501 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
3502 {
3503         unsigned int order = get_order(size);
3504         struct page *p = alloc_pages_node(nid, gfp_mask, order);
3505         if (!p)
3506                 return NULL;
3507         return make_alloc_exact((unsigned long)page_address(p), order, size);
3508 }
3509 
3510 /**
3511  * free_pages_exact - release memory allocated via alloc_pages_exact()
3512  * @virt: the value returned by alloc_pages_exact.
3513  * @size: size of allocation, same value as passed to alloc_pages_exact().
3514  *
3515  * Release the memory allocated by a previous call to alloc_pages_exact.
3516  */
3517 void free_pages_exact(void *virt, size_t size)
3518 {
3519         unsigned long addr = (unsigned long)virt;
3520         unsigned long end = addr + PAGE_ALIGN(size);
3521 
3522         while (addr < end) {
3523                 free_page(addr);
3524                 addr += PAGE_SIZE;
3525         }
3526 }
3527 EXPORT_SYMBOL(free_pages_exact);
3528 
3529 /**
3530  * nr_free_zone_pages - count number of pages beyond high watermark
3531  * @offset: The zone index of the highest zone
3532  *
3533  * nr_free_zone_pages() counts the number of counts pages which are beyond the
3534  * high watermark within all zones at or below a given zone index.  For each
3535  * zone, the number of pages is calculated as:
3536  *     managed_pages - high_pages
3537  */
3538 static unsigned long nr_free_zone_pages(int offset)
3539 {
3540         struct zoneref *z;
3541         struct zone *zone;
3542 
3543         /* Just pick one node, since fallback list is circular */
3544         unsigned long sum = 0;
3545 
3546         struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3547 
3548         for_each_zone_zonelist(zone, z, zonelist, offset) {
3549                 unsigned long size = zone->managed_pages;
3550                 unsigned long high = high_wmark_pages(zone);
3551                 if (size > high)
3552                         sum += size - high;
3553         }
3554 
3555         return sum;
3556 }
3557 
3558 /**
3559  * nr_free_buffer_pages - count number of pages beyond high watermark
3560  *
3561  * nr_free_buffer_pages() counts the number of pages which are beyond the high
3562  * watermark within ZONE_DMA and ZONE_NORMAL.
3563  */
3564 unsigned long nr_free_buffer_pages(void)
3565 {
3566         return nr_free_zone_pages(gfp_zone(GFP_USER));
3567 }
3568 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3569 
3570 /**
3571  * nr_free_pagecache_pages - count number of pages beyond high watermark
3572  *
3573  * nr_free_pagecache_pages() counts the number of pages which are beyond the
3574  * high watermark within all zones.
3575  */
3576 unsigned long nr_free_pagecache_pages(void)
3577 {
3578         return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3579 }
3580 
3581 static inline void show_node(struct zone *zone)
3582 {
3583         if (IS_ENABLED(CONFIG_NUMA))
3584                 printk("Node %d ", zone_to_nid(zone));
3585 }
3586 
3587 void si_meminfo(struct sysinfo *val)
3588 {
3589         val->totalram = totalram_pages;
3590         val->sharedram = global_page_state(NR_SHMEM);
3591         val->freeram = global_page_state(NR_FREE_PAGES);
3592         val->bufferram = nr_blockdev_pages();
3593         val->totalhigh = totalhigh_pages;
3594         val->freehigh = nr_free_highpages();
3595         val->mem_unit = PAGE_SIZE;
3596 }
3597 
3598 EXPORT_SYMBOL(si_meminfo);
3599 
3600 #ifdef CONFIG_NUMA
3601 void si_meminfo_node(struct sysinfo *val, int nid)
3602 {
3603         int zone_type;          /* needs to be signed */
3604         unsigned long managed_pages = 0;
3605         pg_data_t *pgdat = NODE_DATA(nid);
3606 
3607         for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3608                 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3609         val->totalram = managed_pages;
3610         val->sharedram = node_page_state(nid, NR_SHMEM);
3611         val->freeram = node_page_state(nid, NR_FREE_PAGES);
3612 #ifdef CONFIG_HIGHMEM
3613         val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
3614         val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3615                         NR_FREE_PAGES);
3616 #else
3617         val->totalhigh = 0;
3618         val->freehigh = 0;
3619 #endif
3620         val->mem_unit = PAGE_SIZE;
3621 }
3622 #endif
3623 
3624 /*
3625  * Determine whether the node should be displayed or not, depending on whether
3626  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3627  */
3628 bool skip_free_areas_node(unsigned int flags, int nid)
3629 {
3630         bool ret = false;
3631         unsigned int cpuset_mems_cookie;
3632 
3633         if (!(flags & SHOW_MEM_FILTER_NODES))
3634                 goto out;
3635 
3636         do {
3637                 cpuset_mems_cookie = read_mems_allowed_begin();
3638                 ret = !node_isset(nid, cpuset_current_mems_allowed);
3639         } while (read_mems_allowed_retry(cpuset_mems_cookie));
3640 out:
3641         return ret;
3642 }
3643 
3644 #define K(x) ((x) << (PAGE_SHIFT-10))
3645 
3646 static void show_migration_types(unsigned char type)
3647 {
3648         static const char types[MIGRATE_TYPES] = {
3649                 [MIGRATE_UNMOVABLE]     = 'U',
3650                 [MIGRATE_MOVABLE]       = 'M',
3651                 [MIGRATE_RECLAIMABLE]   = 'E',
3652                 [MIGRATE_HIGHATOMIC]    = 'H',
3653 #ifdef CONFIG_CMA
3654                 [MIGRATE_CMA]           = 'C',
3655 #endif
3656 #ifdef CONFIG_MEMORY_ISOLATION
3657                 [MIGRATE_ISOLATE]       = 'I',
3658 #endif
3659         };
3660         char tmp[MIGRATE_TYPES + 1];
3661         char *p = tmp;
3662         int i;
3663 
3664         for (i = 0; i < MIGRATE_TYPES; i++) {
3665                 if (type & (1 << i))
3666                         *p++ = types[i];
3667         }
3668 
3669         *p = '\0';
3670         printk("(%s) ", tmp);
3671 }
3672 
3673 /*
3674  * Show free area list (used inside shift_scroll-lock stuff)
3675  * We also calculate the percentage fragmentation. We do this by counting the
3676  * memory on each free list with the exception of the first item on the list.
3677  *
3678  * Bits in @filter:
3679  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3680  *   cpuset.
3681  */
3682 void show_free_areas(unsigned int filter)
3683 {
3684         unsigned long free_pcp = 0;
3685         int cpu;
3686         struct zone *zone;
3687 
3688         for_each_populated_zone(zone) {
3689                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3690                         continue;
3691 
3692                 for_each_online_cpu(cpu)
3693                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3694         }
3695 
3696         printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3697                 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3698                 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3699                 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3700                 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3701                 " free:%lu free_pcp:%lu free_cma:%lu\n",
3702                 global_page_state(NR_ACTIVE_ANON),
3703                 global_page_state(NR_INACTIVE_ANON),
3704                 global_page_state(NR_ISOLATED_ANON),
3705                 global_page_state(NR_ACTIVE_FILE),
3706                 global_page_state(NR_INACTIVE_FILE),
3707                 global_page_state(NR_ISOLATED_FILE),
3708                 global_page_state(NR_UNEVICTABLE),
3709                 global_page_state(NR_FILE_DIRTY),
3710                 global_page_state(NR_WRITEBACK),
3711                 global_page_state(NR_UNSTABLE_NFS),
3712                 global_page_state(NR_SLAB_RECLAIMABLE),
3713                 global_page_state(NR_SLAB_UNRECLAIMABLE),
3714                 global_page_state(NR_FILE_MAPPED),
3715                 global_page_state(NR_SHMEM),
3716                 global_page_state(NR_PAGETABLE),
3717                 global_page_state(NR_BOUNCE),
3718                 global_page_state(NR_FREE_PAGES),
3719                 free_pcp,
3720                 global_page_state(NR_FREE_CMA_PAGES));
3721 
3722         for_each_populated_zone(zone) {
3723                 int i;
3724 
3725                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3726                         continue;
3727 
3728                 free_pcp = 0;
3729                 for_each_online_cpu(cpu)
3730                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3731 
3732                 show_node(zone);
3733                 printk("%s"
3734                         " free:%lukB"
3735                         " min:%lukB"
3736                         " low:%lukB"
3737                         " high:%lukB"
3738                         " active_anon:%lukB"
3739                         " inactive_anon:%lukB"
3740                         " active_file:%lukB"
3741                         " inactive_file:%lukB"
3742                         " unevictable:%lukB"
3743                         " isolated(anon):%lukB"
3744                         " isolated(file):%lukB"
3745                         " present:%lukB"
3746                         " managed:%lukB"
3747                         " mlocked:%lukB"
3748                         " dirty:%lukB"
3749                         " writeback:%lukB"
3750                         " mapped:%lukB"
3751                         " shmem:%lukB"
3752                         " slab_reclaimable:%lukB"
3753                         " slab_unreclaimable:%lukB"
3754                         " kernel_stack:%lukB"
3755                         " pagetables:%lukB"
3756                         " unstable:%lukB"
3757                         " bounce:%lukB"
3758                         " free_pcp:%lukB"
3759                         " local_pcp:%ukB"
3760                         " free_cma:%lukB"
3761                         " writeback_tmp:%lukB"
3762                         " pages_scanned:%lu"
3763                         " all_unreclaimable? %s"
3764                         "\n",
3765                         zone->name,
3766                         K(zone_page_state(zone, NR_FREE_PAGES)),
3767                         K(min_wmark_pages(zone)),
3768                         K(low_wmark_pages(zone)),
3769                         K(high_wmark_pages(zone)),
3770                         K(zone_page_state(zone, NR_ACTIVE_ANON)),
3771                         K(zone_page_state(zone, NR_INACTIVE_ANON)),
3772                         K(zone_page_state(zone, NR_ACTIVE_FILE)),
3773                         K(zone_page_state(zone, NR_INACTIVE_FILE)),
3774                         K(zone_page_state(zone, NR_UNEVICTABLE)),
3775                         K(zone_page_state(zone, NR_ISOLATED_ANON)),
3776                         K(zone_page_state(zone, NR_ISOLATED_FILE)),
3777                         K(zone->present_pages),
3778                         K(zone->managed_pages),
3779                         K(zone_page_state(zone, NR_MLOCK)),
3780                         K(zone_page_state(zone, NR_FILE_DIRTY)),
3781                         K(zone_page_state(zone, NR_WRITEBACK)),
3782                         K(zone_page_state(zone, NR_FILE_MAPPED)),
3783                         K(zone_page_state(zone, NR_SHMEM)),
3784                         K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3785                         K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3786                         zone_page_state(zone, NR_KERNEL_STACK) *
3787                                 THREAD_SIZE / 1024,
3788                         K(zone_page_state(zone, NR_PAGETABLE)),
3789                         K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3790                         K(zone_page_state(zone, NR_BOUNCE)),
3791                         K(free_pcp),
3792                         K(this_cpu_read(zone->pageset->pcp.count)),
3793                         K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3794                         K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3795                         K(zone_page_state(zone, NR_PAGES_SCANNED)),
3796                         (!zone_reclaimable(zone) ? "yes" : "no")
3797                         );
3798                 printk("lowmem_reserve[]:");
3799                 for (i = 0; i < MAX_NR_ZONES; i++)
3800                         printk(" %ld", zone->lowmem_reserve[i]);
3801                 printk("\n");
3802         }
3803 
3804         for_each_populated_zone(zone) {
3805                 unsigned int order;
3806                 unsigned long nr[MAX_ORDER], flags, total = 0;
3807                 unsigned char types[MAX_ORDER];
3808 
3809                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3810                         continue;
3811                 show_node(zone);
3812                 printk("%s: ", zone->name);
3813 
3814                 spin_lock_irqsave(&zone->lock, flags);
3815                 for (order = 0; order < MAX_ORDER; order++) {
3816                         struct free_area *area = &zone->free_area[order];
3817                         int type;
3818 
3819                         nr[order] = area->nr_free;
3820                         total += nr[order] << order;
3821 
3822                         types[order] = 0;
3823                         for (type = 0; type < MIGRATE_TYPES; type++) {
3824                                 if (!list_empty(&area->free_list[type]))
3825                                         types[order] |= 1 << type;
3826                         }
3827                 }
3828                 spin_unlock_irqrestore(&zone->lock, flags);
3829                 for (order = 0; order < MAX_ORDER; order++) {
3830                         printk("%lu*%lukB ", nr[order], K(1UL) << order);
3831                         if (nr[order])
3832                                 show_migration_types(types[order]);
3833                 }
3834                 printk("= %lukB\n", K(total));
3835         }
3836 
3837         hugetlb_show_meminfo();
3838 
3839         printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3840 
3841         show_swap_cache_info();
3842 }
3843 
3844 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3845 {
3846         zoneref->zone = zone;
3847         zoneref->zone_idx = zone_idx(zone);
3848 }
3849 
3850 /*
3851  * Builds allocation fallback zone lists.
3852  *
3853  * Add all populated zones of a node to the zonelist.
3854  */
3855 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3856                                 int nr_zones)
3857 {
3858         struct zone *zone;
3859         enum zone_type zone_type = MAX_NR_ZONES;
3860 
3861         do {
3862                 zone_type--;
3863                 zone = pgdat->node_zones + zone_type;
3864                 if (populated_zone(zone)) {
3865                         zoneref_set_zone(zone,
3866                                 &zonelist->_zonerefs[nr_zones++]);
3867                         check_highest_zone(zone_type);
3868                 }
3869         } while (zone_type);
3870 
3871         return nr_zones;
3872 }
3873 
3874 
3875 /*
3876  *  zonelist_order:
3877  *  0 = automatic detection of better ordering.
3878  *  1 = order by ([node] distance, -zonetype)
3879  *  2 = order by (-zonetype, [node] distance)
3880  *
3881  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3882  *  the same zonelist. So only NUMA can configure this param.
3883  */
3884 #define ZONELIST_ORDER_DEFAULT  0
3885 #define ZONELIST_ORDER_NODE     1
3886 #define ZONELIST_ORDER_ZONE     2
3887 
3888 /* zonelist order in the kernel.
3889  * set_zonelist_order() will set this to NODE or ZONE.
3890  */
3891 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3892 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3893 
3894 
3895 #ifdef CONFIG_NUMA
3896 /* The value user specified ....changed by config */
3897 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3898 /* string for sysctl */
3899 #define NUMA_ZONELIST_ORDER_LEN 16
3900 char numa_zonelist_order[16] = "default";
3901 
3902 /*
3903  * interface for configure zonelist ordering.
3904  * command line option "numa_zonelist_order"
3905  *      = "[dD]efault   - default, automatic configuration.
3906  *      = "[nN]ode      - order by node locality, then by zone within node
3907  *      = "[zZ]one      - order by zone, then by locality within zone
3908  */
3909 
3910 static int __parse_numa_zonelist_order(char *s)
3911 {
3912         if (*s == 'd' || *s == 'D') {
3913                 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3914         } else if (*s == 'n' || *s == 'N') {
3915                 user_zonelist_order = ZONELIST_ORDER_NODE;
3916         } else if (*s == 'z' || *s == 'Z') {
3917                 user_zonelist_order = ZONELIST_ORDER_ZONE;
3918         } else {
3919                 printk(KERN_WARNING
3920                         "Ignoring invalid numa_zonelist_order value:  "
3921                         "%s\n", s);
3922                 return -EINVAL;
3923         }
3924         return 0;
3925 }
3926 
3927 static __init int setup_numa_zonelist_order(char *s)
3928 {
3929         int ret;
3930 
3931         if (!s)
3932                 return 0;
3933 
3934         ret = __parse_numa_zonelist_order(s);
3935         if (ret == 0)
3936                 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3937 
3938         return ret;
3939 }
3940 early_param("numa_zonelist_order", setup_numa_zonelist_order);
3941 
3942 /*
3943  * sysctl handler for numa_zonelist_order
3944  */
3945 int numa_zonelist_order_handler(struct ctl_table *table, int write,
3946                 void __user *buffer, size_t *length,
3947                 loff_t *ppos)
3948 {
3949         char saved_string[NUMA_ZONELIST_ORDER_LEN];
3950         int ret;
3951         static DEFINE_MUTEX(zl_order_mutex);
3952 
3953         mutex_lock(&zl_order_mutex);
3954         if (write) {
3955                 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3956                         ret = -EINVAL;
3957                         goto out;
3958                 }
3959                 strcpy(saved_string, (char *)table->data);
3960         }
3961         ret = proc_dostring(table, write, buffer, length, ppos);
3962         if (ret)
3963                 goto out;
3964         if (write) {
3965                 int oldval = user_zonelist_order;
3966 
3967                 ret = __parse_numa_zonelist_order((char *)table->data);
3968                 if (ret) {
3969                         /*
3970                          * bogus value.  restore saved string
3971                          */
3972                         strncpy((char *)table->data, saved_string,
3973                                 NUMA_ZONELIST_ORDER_LEN);
3974                         user_zonelist_order = oldval;
3975                 } else if (oldval != user_zonelist_order) {
3976                         mutex_lock(&zonelists_mutex);
3977                         build_all_zonelists(NULL, NULL);
3978                         mutex_unlock(&zonelists_mutex);
3979                 }
3980         }
3981 out:
3982         mutex_unlock(&zl_order_mutex);
3983         return ret;
3984 }
3985 
3986 
3987 #define MAX_NODE_LOAD (nr_online_nodes)
3988 static int node_load[MAX_NUMNODES];
3989 
3990 /**
3991  * find_next_best_node - find the next node that should appear in a given node's fallback list
3992  * @node: node whose fallback list we're appending
3993  * @used_node_mask: nodemask_t of already used nodes
3994  *
3995  * We use a number of factors to determine which is the next node that should
3996  * appear on a given node's fallback list.  The node should not have appeared
3997  * already in @node's fallback list, and it should be the next closest node
3998  * according to the distance array (which contains arbitrary distance values
3999  * from each node to each node in the system), and should also prefer nodes
4000  * with no CPUs, since presumably they'll have very little allocation pressure
4001  * on them otherwise.
4002  * It returns -1 if no node is found.
4003  */
4004 static int find_next_best_node(int node, nodemask_t *used_node_mask)
4005 {
4006         int n, val;
4007         int min_val = INT_MAX;
4008         int best_node = NUMA_NO_NODE;
4009         const struct cpumask *tmp = cpumask_of_node(0);
4010 
4011         /* Use the local node if we haven't already */
4012         if (!node_isset(node, *used_node_mask)) {
4013                 node_set(node, *used_node_mask);
4014                 return node;
4015         }
4016 
4017         for_each_node_state(n, N_MEMORY) {
4018 
4019                 /* Don't want a node to appear more than once */
4020                 if (node_isset(n, *used_node_mask))
4021                         continue;
4022 
4023                 /* Use the distance array to find the distance */
4024                 val = node_distance(node, n);
4025 
4026                 /* Penalize nodes under us ("prefer the next node") */
4027                 val += (n < node);
4028 
4029                 /* Give preference to headless and unused nodes */
4030                 tmp = cpumask_of_node(n);
4031                 if (!cpumask_empty(tmp))
4032                         val += PENALTY_FOR_NODE_WITH_CPUS;
4033 
4034                 /* Slight preference for less loaded node */
4035                 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4036                 val += node_load[n];
4037 
4038                 if (val < min_val) {
4039                         min_val = val;
4040                         best_node = n;
4041                 }
4042         }
4043 
4044         if (best_node >= 0)
4045                 node_set(best_node, *used_node_mask);
4046 
4047         return best_node;
4048 }
4049 
4050 
4051 /*
4052  * Build zonelists ordered by node and zones within node.
4053  * This results in maximum locality--normal zone overflows into local
4054  * DMA zone, if any--but risks exhausting DMA zone.
4055  */
4056 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
4057 {
4058         int j;
4059         struct zonelist *zonelist;
4060 
4061         zonelist = &pgdat->node_zonelists[0];
4062         for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
4063                 ;
4064         j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4065         zonelist->_zonerefs[j].zone = NULL;
4066         zonelist->_zonerefs[j].zone_idx = 0;
4067 }
4068 
4069 /*
4070  * Build gfp_thisnode zonelists
4071  */
4072 static void build_thisnode_zonelists(pg_data_t *pgdat)
4073 {
4074         int j;
4075         struct zonelist *zonelist;
4076 
4077         zonelist = &pgdat->node_zonelists[1];
4078         j = build_zonelists_node(pgdat, zonelist, 0);
4079         zonelist->_zonerefs[j].zone = NULL;
4080         zonelist->_zonerefs[j].zone_idx = 0;
4081 }
4082 
4083 /*
4084  * Build zonelists ordered by zone and nodes within zones.
4085  * This results in conserving DMA zone[s] until all Normal memory is
4086  * exhausted, but results in overflowing to remote node while memory
4087  * may still exist in local DMA zone.
4088  */
4089 static int node_order[MAX_NUMNODES];
4090 
4091 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4092 {
4093         int pos, j, node;
4094         int zone_type;          /* needs to be signed */
4095         struct zone *z;
4096         struct zonelist *zonelist;
4097 
4098         zonelist = &pgdat->node_zonelists[0];
4099         pos = 0;
4100         for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4101                 for (j = 0; j < nr_nodes; j++) {
4102                         node = node_order[j];
4103                         z = &NODE_DATA(node)->node_zones[zone_type];
4104                         if (populated_zone(z)) {
4105                                 zoneref_set_zone(z,
4106                                         &zonelist->_zonerefs[pos++]);
4107                                 check_highest_zone(zone_type);
4108                         }
4109                 }
4110         }
4111         zonelist->_zonerefs[pos].zone = NULL;
4112         zonelist->_zonerefs[pos].zone_idx = 0;
4113 }
4114 
4115 #if defined(CONFIG_64BIT)
4116 /*
4117  * Devices that require DMA32/DMA are relatively rare and do not justify a
4118  * penalty to every machine in case the specialised case applies. Default
4119  * to Node-ordering on 64-bit NUMA machines
4120  */
4121 static int default_zonelist_order(void)
4122 {
4123         return ZONELIST_ORDER_NODE;
4124 }
4125 #else
4126 /*
4127  * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4128  * by the kernel. If processes running on node 0 deplete the low memory zone
4129  * then reclaim will occur more frequency increasing stalls and potentially
4130  * be easier to OOM if a large percentage of the zone is under writeback or
4131  * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4132  * Hence, default to zone ordering on 32-bit.
4133  */
4134 static int default_zonelist_order(void)
4135 {
4136         return ZONELIST_ORDER_ZONE;
4137 }
4138 #endif /* CONFIG_64BIT */
4139 
4140 static void set_zonelist_order(void)
4141 {
4142         if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4143                 current_zonelist_order = default_zonelist_order();
4144         else
4145                 current_zonelist_order = user_zonelist_order;
4146 }
4147 
4148 static void build_zonelists(pg_data_t *pgdat)
4149 {
4150         int j, node, load;
4151         enum zone_type i;
4152         nodemask_t used_mask;
4153         int local_node, prev_node;
4154         struct zonelist *zonelist;
4155         unsigned int order = current_zonelist_order;
4156 
4157         /* initialize zonelists */
4158         for (i = 0; i < MAX_ZONELISTS; i++) {
4159                 zonelist = pgdat->node_zonelists + i;
4160                 zonelist->_zonerefs[0].zone = NULL;
4161                 zonelist->_zonerefs[0].zone_idx = 0;
4162         }
4163 
4164         /* NUMA-aware ordering of nodes */
4165         local_node = pgdat->node_id;
4166         load = nr_online_nodes;
4167         prev_node = local_node;
4168         nodes_clear(used_mask);
4169 
4170         memset(node_order, 0, sizeof(node_order));
4171         j = 0;
4172 
4173         while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4174                 /*
4175                  * We don't want to pressure a particular node.
4176                  * So adding penalty to the first node in same
4177                  * distance group to make it round-robin.
4178                  */
4179                 if (node_distance(local_node, node) !=
4180                     node_distance(local_node, prev_node))
4181                         node_load[node] = load;
4182 
4183                 prev_node = node;
4184                 load--;
4185                 if (order == ZONELIST_ORDER_NODE)
4186                         build_zonelists_in_node_order(pgdat, node);
4187                 else
4188                         node_order[j++] = node; /* remember order */
4189         }
4190 
4191         if (order == ZONELIST_ORDER_ZONE) {
4192                 /* calculate node order -- i.e., DMA last! */
4193                 build_zonelists_in_zone_order(pgdat, j);
4194         }
4195 
4196         build_thisnode_zonelists(pgdat);
4197 }
4198 
4199 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4200 /*
4201  * Return node id of node used for "local" allocations.
4202  * I.e., first node id of first zone in arg node's generic zonelist.
4203  * Used for initializing percpu 'numa_mem', which is used primarily
4204  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4205  */
4206 int local_memory_node(int node)
4207 {
4208         struct zone *zone;
4209 
4210         (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4211                                    gfp_zone(GFP_KERNEL),
4212                                    NULL,
4213                                    &zone);
4214         return zone->node;
4215 }
4216 #endif
4217 
4218 #else   /* CONFIG_NUMA */
4219 
4220 static void set_zonelist_order(void)
4221 {
4222         current_zonelist_order = ZONELIST_ORDER_ZONE;
4223 }
4224 
4225 static void build_zonelists(pg_data_t *pgdat)
4226 {
4227         int node, local_node;
4228         enum zone_type j;
4229         struct zonelist *zonelist;
4230 
4231         local_node = pgdat->node_id;
4232 
4233         zonelist = &pgdat->node_zonelists[0];
4234         j = build_zonelists_node(pgdat, zonelist, 0);
4235 
4236         /*
4237          * Now we build the zonelist so that it contains the zones
4238          * of all the other nodes.
4239          * We don't want to pressure a particular node, so when
4240          * building the zones for node N, we make sure that the
4241          * zones coming right after the local ones are those from
4242          * node N+1 (modulo N)
4243          */
4244         for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4245                 if (!node_online(node))
4246                         continue;
4247                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4248         }
4249         for (node = 0; node < local_node; node++) {
4250                 if (!node_online(node))
4251                         continue;
4252                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4253         }
4254 
4255         zonelist->_zonerefs[j].zone = NULL;
4256         zonelist->_zonerefs[j].zone_idx = 0;
4257 }
4258 
4259 #endif  /* CONFIG_NUMA */
4260 
4261 /*
4262  * Boot pageset table. One per cpu which is going to be used for all
4263  * zones and all nodes. The parameters will be set in such a way
4264  * that an item put on a list will immediately be handed over to
4265  * the buddy list. This is safe since pageset manipulation is done
4266  * with interrupts disabled.
4267  *
4268  * The boot_pagesets must be kept even after bootup is complete for
4269  * unused processors and/or zones. They do play a role for bootstrapping
4270  * hotplugged processors.
4271  *
4272  * zoneinfo_show() and maybe other functions do
4273  * not check if the processor is online before following the pageset pointer.
4274  * Other parts of the kernel may not check if the zone is available.
4275  */
4276 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4277 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
4278 static void setup_zone_pageset(struct zone *zone);
4279 
4280 /*
4281  * Global mutex to protect against size modification of zonelists
4282  * as well as to serialize pageset setup for the new populated zone.
4283  */
4284 DEFINE_MUTEX(zonelists_mutex);
4285 
4286 /* return values int ....just for stop_machine() */
4287 static int __build_all_zonelists(void *data)
4288 {
4289         int nid;
4290         int cpu;
4291         pg_data_t *self = data;
4292 
4293 #ifdef CONFIG_NUMA
4294         memset(node_load, 0, sizeof(node_load));
4295 #endif
4296 
4297         if (self && !node_online(self->node_id)) {
4298                 build_zonelists(self);
4299         }
4300 
4301         for_each_online_node(nid) {
4302                 pg_data_t *pgdat = NODE_DATA(nid);
4303 
4304                 build_zonelists(pgdat);
4305         }
4306 
4307         /*
4308          * Initialize the boot_pagesets that are going to be used
4309          * for bootstrapping processors. The real pagesets for
4310          * each zone will be allocated later when the per cpu
4311          * allocator is available.
4312          *
4313          * boot_pagesets are used also for bootstrapping offline
4314          * cpus if the system is already booted because the pagesets
4315          * are needed to initialize allocators on a specific cpu too.
4316          * F.e. the percpu allocator needs the page allocator which
4317          * needs the percpu allocator in order to allocate its pagesets
4318          * (a chicken-egg dilemma).
4319          */
4320         for_each_possible_cpu(cpu) {
4321                 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4322 
4323 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4324                 /*
4325                  * We now know the "local memory node" for each node--
4326                  * i.e., the node of the first zone in the generic zonelist.
4327                  * Set up numa_mem percpu variable for on-line cpus.  During
4328                  * boot, only the boot cpu should be on-line;  we'll init the
4329                  * secondary cpus' numa_mem as they come on-line.  During
4330                  * node/memory hotplug, we'll fixup all on-line cpus.
4331                  */
4332                 if (cpu_online(cpu))
4333                         set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4334 #endif
4335         }
4336 
4337         return 0;
4338 }
4339 
4340 static noinline void __init
4341 build_all_zonelists_init(void)
4342 {
4343         __build_all_zonelists(NULL);
4344         mminit_verify_zonelist();
4345         cpuset_init_current_mems_allowed();
4346 }
4347 
4348 /*
4349  * Called with zonelists_mutex held always
4350  * unless system_state == SYSTEM_BOOTING.
4351  *
4352  * __ref due to (1) call of __meminit annotated setup_zone_pageset
4353  * [we're only called with non-NULL zone through __meminit paths] and
4354  * (2) call of __init annotated helper build_all_zonelists_init
4355  * [protected by SYSTEM_BOOTING].
4356  */
4357 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
4358 {
4359         set_zonelist_order();
4360 
4361         if (system_state == SYSTEM_BOOTING) {
4362                 build_all_zonelists_init();
4363         } else {
4364 #ifdef CONFIG_MEMORY_HOTPLUG
4365                 if (zone)
4366                         setup_zone_pageset(zone);
4367 #endif
4368                 /* we have to stop all cpus to guarantee there is no user
4369                    of zonelist */
4370                 stop_machine(__build_all_zonelists, pgdat, NULL);
4371                 /* cpuset refresh routine should be here */
4372         }
4373         vm_total_pages = nr_free_pagecache_pages();
4374         /*
4375          * Disable grouping by mobility if the number of pages in the
4376          * system is too low to allow the mechanism to work. It would be
4377          * more accurate, but expensive to check per-zone. This check is
4378          * made on memory-hotadd so a system can start with mobility
4379          * disabled and enable it later
4380          */
4381         if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
4382                 page_group_by_mobility_disabled = 1;
4383         else
4384                 page_group_by_mobility_disabled = 0;
4385 
4386         pr_info("Built %i zonelists in %s order, mobility grouping %s.  "
4387                 "Total pages: %ld\n",
4388                         nr_online_nodes,
4389                         zonelist_order_name[current_zonelist_order],
4390                         page_group_by_mobility_disabled ? "off" : "on",
4391                         vm_total_pages);
4392 #ifdef CONFIG_NUMA
4393         pr_info("Policy zone: %s\n", zone_names[policy_zone]);
4394 #endif
4395 }
4396 
4397 /*
4398  * Helper functions to size the waitqueue hash table.
4399  * Essentially these want to choose hash table sizes sufficiently
4400  * large so that collisions trying to wait on pages are rare.
4401  * But in fact, the number of active page waitqueues on typical
4402  * systems is ridiculously low, less than 200. So this is even
4403  * conservative, even though it seems large.
4404  *
4405  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4406  * waitqueues, i.e. the size of the waitq table given the number of pages.
4407  */
4408 #define PAGES_PER_WAITQUEUE     256
4409 
4410 #ifndef CONFIG_MEMORY_HOTPLUG
4411 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4412 {
4413         unsigned long size = 1;
4414 
4415         pages /= PAGES_PER_WAITQUEUE;
4416 
4417         while (size < pages)
4418                 size <<= 1;
4419 
4420         /*
4421          * Once we have dozens or even hundreds of threads sleeping
4422          * on IO we've got bigger problems than wait queue collision.
4423          * Limit the size of the wait table to a reasonable size.
4424          */
4425         size = min(size, 4096UL);
4426 
4427         return max(size, 4UL);
4428 }
4429 #else
4430 /*
4431  * A zone's size might be changed by hot-add, so it is not possible to determine
4432  * a suitable size for its wait_table.  So we use the maximum size now.
4433  *
4434  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
4435  *
4436  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
4437  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4438  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
4439  *
4440  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4441  * or more by the traditional way. (See above).  It equals:
4442  *
4443  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
4444  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
4445  *    powerpc (64K page size)             : =  (32G +16M)byte.
4446  */
4447 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4448 {
4449         return 4096UL;
4450 }
4451 #endif
4452 
4453 /*
4454  * This is an integer logarithm so that shifts can be used later
4455  * to extract the more random high bits from the multiplicative
4456  * hash function before the remainder is taken.
4457  */
4458 static inline unsigned long wait_table_bits(unsigned long size)
4459 {
4460         return ffz(~size);
4461 }
4462 
4463 /*
4464  * Initially all pages are reserved - free ones are freed
4465  * up by free_all_bootmem() once the early boot process is
4466  * done. Non-atomic initialization, single-pass.
4467  */
4468 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4469                 unsigned long start_pfn, enum memmap_context context)
4470 {
4471         pg_data_t *pgdat = NODE_DATA(nid);
4472         unsigned long end_pfn = start_pfn + size;
4473         unsigned long pfn;
4474         struct zone *z;
4475         unsigned long nr_initialised = 0;
4476 
4477         if (highest_memmap_pfn < end_pfn - 1)
4478                 highest_memmap_pfn = end_pfn - 1;
4479 
4480         z = &pgdat->node_zones[zone];
4481         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4482                 /*
4483                  * There can be holes in boot-time mem_map[]s
4484                  * handed to this function.  They do not
4485                  * exist on hotplugged memory.
4486                  */
4487                 if (context == MEMMAP_EARLY) {
4488                         if (!early_pfn_valid(pfn))
4489                                 continue;
4490                         if (!early_pfn_in_nid(pfn, nid))
4491                                 continue;
4492                         if (!update_defer_init(pgdat, pfn, end_pfn,
4493                                                 &nr_initialised))
4494                                 break;
4495                 }
4496 
4497                 /*
4498                  * Mark the block movable so that blocks are reserved for
4499                  * movable at startup. This will force kernel allocations
4500                  * to reserve their blocks rather than leaking throughout
4501                  * the address space during boot when many long-lived
4502                  * kernel allocations are made.
4503                  *
4504                  * bitmap is created for zone's valid pfn range. but memmap
4505                  * can be created for invalid pages (for alignment)
4506                  * check here not to call set_pageblock_migratetype() against
4507                  * pfn out of zone.
4508                  */
4509                 if (!(pfn & (pageblock_nr_pages - 1))) {
4510                         struct page *page = pfn_to_page(pfn);
4511 
4512                         __init_single_page(page, pfn, zone, nid);
4513                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4514                 } else {
4515                         __init_single_pfn(pfn, zone, nid);
4516                 }
4517         }
4518 }
4519 
4520 static void __meminit zone_init_free_lists(struct zone *zone)
4521 {
4522         unsigned int order, t;
4523         for_each_migratetype_order(order, t) {
4524                 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4525                 zone->free_area[order].nr_free = 0;
4526         }
4527 }
4528 
4529 #ifndef __HAVE_ARCH_MEMMAP_INIT
4530 #define memmap_init(size, nid, zone, start_pfn) \
4531         memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4532 #endif
4533 
4534 static int zone_batchsize(struct zone *zone)
4535 {
4536 #ifdef CONFIG_MMU
4537         int batch;
4538 
4539         /*
4540          * The per-cpu-pages pools are set to around 1000th of the
4541          * size of the zone.  But no more than 1/2 of a meg.
4542          *
4543          * OK, so we don't know how big the cache is.  So guess.
4544          */
4545         batch = zone->managed_pages / 1024;
4546         if (batch * PAGE_SIZE > 512 * 1024)
4547                 batch = (512 * 1024) / PAGE_SIZE;
4548         batch /= 4;             /* We effectively *= 4 below */
4549         if (batch < 1)
4550                 batch = 1;
4551 
4552         /*
4553          * Clamp the batch to a 2^n - 1 value. Having a power
4554          * of 2 value was found to be more likely to have
4555          * suboptimal cache aliasing properties in some cases.
4556          *
4557          * For example if 2 tasks are alternately allocating
4558          * batches of pages, one task can end up with a lot
4559          * of pages of one half of the possible page colors
4560          * and the other with pages of the other colors.
4561          */
4562         batch = rounddown_pow_of_two(batch + batch/2) - 1;
4563 
4564         return batch;
4565 
4566 #else
4567         /* The deferral and batching of frees should be suppressed under NOMMU
4568          * conditions.
4569          *
4570          * The problem is that NOMMU needs to be able to allocate large chunks
4571          * of contiguous memory as there's no hardware page translation to
4572          * assemble apparent contiguous memory from discontiguous pages.
4573          *
4574          * Queueing large contiguous runs of pages for batching, however,
4575          * causes the pages to actually be freed in smaller chunks.  As there
4576          * can be a significant delay between the individual batches being
4577          * recycled, this leads to the once large chunks of space being
4578          * fragmented and becoming unavailable for high-order allocations.
4579          */
4580         return 0;
4581 #endif
4582 }
4583 
4584 /*
4585  * pcp->high and pcp->batch values are related and dependent on one another:
4586  * ->batch must never be higher then ->high.
4587  * The following function updates them in a safe manner without read side
4588  * locking.
4589  *
4590  * Any new users of pcp->batch and pcp->high should ensure they can cope with
4591  * those fields changing asynchronously (acording the the above rule).
4592  *
4593  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4594  * outside of boot time (or some other assurance that no concurrent updaters
4595  * exist).
4596  */
4597 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4598                 unsigned long batch)
4599 {
4600        /* start with a fail safe value for batch */
4601         pcp->batch = 1;
4602         smp_wmb();
4603 
4604        /* Update high, then batch, in order */
4605         pcp->high = high;
4606         smp_wmb();
4607 
4608         pcp->batch = batch;
4609 }
4610 
4611 /* a companion to pageset_set_high() */
4612 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4613 {
4614         pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4615 }
4616 
4617 static void pageset_init(struct per_cpu_pageset *p)
4618 {
4619         struct per_cpu_pages *pcp;
4620         int migratetype;
4621 
4622         memset(p, 0, sizeof(*p));
4623 
4624         pcp = &p->pcp;
4625         pcp->count = 0;
4626         for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4627                 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4628 }
4629 
4630 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4631 {
4632         pageset_init(p);
4633         pageset_set_batch(p, batch);
4634 }
4635 
4636 /*
4637  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4638  * to the value high for the pageset p.
4639  */
4640 static void pageset_set_high(struct per_cpu_pageset *p,
4641                                 unsigned long high)
4642 {
4643         unsigned long batch = max(1UL, high / 4);
4644         if ((high / 4) > (PAGE_SHIFT * 8))
4645                 batch = PAGE_SHIFT * 8;
4646 
4647         pageset_update(&p->pcp, high, batch);
4648 }
4649 
4650 static void pageset_set_high_and_batch(struct zone *zone,
4651                                        struct per_cpu_pageset *pcp)
4652 {
4653         if (percpu_pagelist_fraction)
4654                 pageset_set_high(pcp,
4655                         (zone->managed_pages /
4656                                 percpu_pagelist_fraction));
4657         else
4658                 pageset_set_batch(pcp, zone_batchsize(zone));
4659 }
4660 
4661 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4662 {
4663         struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4664 
4665         pageset_init(pcp);
4666         pageset_set_high_and_batch(zone, pcp);
4667 }
4668 
4669 static void __meminit setup_zone_pageset(struct zone *zone)
4670 {
4671         int cpu;
4672         zone->pageset = alloc_percpu(struct per_cpu_pageset);
4673         for_each_possible_cpu(cpu)
4674                 zone_pageset_init(zone, cpu);
4675 }
4676 
4677 /*
4678  * Allocate per cpu pagesets and initialize them.
4679  * Before this call only boot pagesets were available.
4680  */
4681 void __init setup_per_cpu_pageset(void)
4682 {
4683         struct zone *zone;
4684 
4685         for_each_populated_zone(zone)
4686                 setup_zone_pageset(zone);
4687 }
4688 
4689 static noinline __init_refok
4690 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4691 {
4692         int i;
4693         size_t alloc_size;
4694 
4695         /*
4696          * The per-page waitqueue mechanism uses hashed waitqueues
4697          * per zone.
4698          */
4699         zone->wait_table_hash_nr_entries =
4700                  wait_table_hash_nr_entries(zone_size_pages);
4701         zone->wait_table_bits =
4702                 wait_table_bits(zone->wait_table_hash_nr_entries);
4703         alloc_size = zone->wait_table_hash_nr_entries
4704                                         * sizeof(wait_queue_head_t);
4705 
4706         if (!slab_is_available()) {
4707                 zone->wait_table = (wait_queue_head_t *)
4708                         memblock_virt_alloc_node_nopanic(
4709                                 alloc_size, zone->zone_pgdat->node_id);
4710         } else {
4711                 /*
4712                  * This case means that a zone whose size was 0 gets new memory
4713                  * via memory hot-add.
4714                  * But it may be the case that a new node was hot-added.  In
4715                  * this case vmalloc() will not be able to use this new node's
4716                  * memory - this wait_table must be initialized to use this new
4717                  * node itself as well.
4718                  * To use this new node's memory, further consideration will be
4719                  * necessary.
4720                  */
4721                 zone->wait_table = vmalloc(alloc_size);
4722         }
4723         if (!zone->wait_table)
4724                 return -ENOMEM;
4725 
4726         for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4727                 init_waitqueue_head(zone->wait_table + i);
4728 
4729         return 0;
4730 }
4731 
4732 static __meminit void zone_pcp_init(struct zone *zone)
4733 {
4734         /*
4735          * per cpu subsystem is not up at this point. The following code
4736          * relies on the ability of the linker to provide the
4737          * offset of a (static) per cpu variable into the per cpu area.
4738          */
4739         zone->pageset = &boot_pageset;
4740 
4741         if (populated_zone(zone))
4742                 printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
4743                         zone->name, zone->present_pages,
4744                                          zone_batchsize(zone));
4745 }
4746 
4747 int __meminit init_currently_empty_zone(struct zone *zone,
4748                                         unsigned long zone_start_pfn,
4749                                         unsigned long size)
4750 {
4751         struct pglist_data *pgdat = zone->zone_pgdat;
4752         int ret;
4753         ret = zone_wait_table_init(zone, size);
4754         if (ret)
4755                 return ret;
4756         pgdat->nr_zones = zone_idx(zone) + 1;
4757 
4758         zone->zone_start_pfn = zone_start_pfn;
4759 
4760         mminit_dprintk(MMINIT_TRACE, "memmap_init",
4761                         "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4762                         pgdat->node_id,
4763                         (unsigned long)zone_idx(zone),
4764                         zone_start_pfn, (zone_start_pfn + size));
4765 
4766         zone_init_free_lists(zone);
4767 
4768         return 0;
4769 }
4770 
4771 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4772 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4773 
4774 /*
4775  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4776  */
4777 int __meminit __early_pfn_to_nid(unsigned long pfn,
4778                                         struct mminit_pfnnid_cache *state)
4779 {
4780         unsigned long start_pfn, end_pfn;
4781         int nid;
4782 
4783         if (state->last_start <= pfn && pfn < state->last_end)
4784                 return state->last_nid;
4785 
4786         nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4787         if (nid != -1) {
4788                 state->last_start = start_pfn;
4789                 state->last_end = end_pfn;
4790                 state->last_nid = nid;
4791         }
4792 
4793         return nid;
4794 }
4795 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4796 
4797 /**
4798  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4799  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4800  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4801  *
4802  * If an architecture guarantees that all ranges registered contain no holes
4803  * and may be freed, this this function may be used instead of calling
4804  * memblock_free_early_nid() manually.
4805  */
4806 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
4807 {
4808         unsigned long start_pfn, end_pfn;
4809         int i, this_nid;
4810 
4811         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4812                 start_pfn = min(start_pfn, max_low_pfn);
4813                 end_pfn = min(end_pfn, max_low_pfn);
4814 
4815                 if (start_pfn < end_pfn)
4816                         memblock_free_early_nid(PFN_PHYS(start_pfn),
4817                                         (end_pfn - start_pfn) << PAGE_SHIFT,
4818                                         this_nid);
4819         }
4820 }
4821 
4822 /**
4823  * sparse_memory_present_with_active_regions - Call memory_present for each active range
4824  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4825  *
4826  * If an architecture guarantees that all ranges registered contain no holes and may
4827  * be freed, this function may be used instead of calling memory_present() manually.
4828  */
4829 void __init sparse_memory_present_with_active_regions(int nid)
4830 {
4831         unsigned long start_pfn, end_pfn;
4832         int i, this_nid;
4833 
4834         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4835                 memory_present(this_nid, start_pfn, end_pfn);
4836 }
4837 
4838 /**
4839  * get_pfn_range_for_nid - Return the start and end page frames for a node
4840  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4841  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4842  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4843  *
4844  * It returns the start and end page frame of a node based on information
4845  * provided by memblock_set_node(). If called for a node
4846  * with no available memory, a warning is printed and the start and end
4847  * PFNs will be 0.
4848  */
4849 void __meminit get_pfn_range_for_nid(unsigned int nid,
4850                         unsigned long *start_pfn, unsigned long *end_pfn)
4851 {
4852         unsigned long this_start_pfn, this_end_pfn;
4853         int i;
4854 
4855         *start_pfn = -1UL;
4856         *end_pfn = 0;
4857 
4858         for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4859                 *start_pfn = min(*start_pfn, this_start_pfn);
4860                 *end_pfn = max(*end_pfn, this_end_pfn);
4861         }
4862 
4863         if (*start_pfn == -1UL)
4864                 *start_pfn = 0;
4865 }
4866 
4867 /*
4868  * This finds a zone that can be used for ZONE_MOVABLE pages. The
4869  * assumption is made that zones within a node are ordered in monotonic
4870  * increasing memory addresses so that the "highest" populated zone is used
4871  */
4872 static void __init find_usable_zone_for_movable(void)
4873 {
4874         int zone_index;
4875         for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4876                 if (zone_index == ZONE_MOVABLE)
4877                         continue;
4878 
4879                 if (arch_zone_highest_possible_pfn[zone_index] >
4880                                 arch_zone_lowest_possible_pfn[zone_index])
4881                         break;
4882         }
4883 
4884         VM_BUG_ON(zone_index == -1);
4885         movable_zone = zone_index;
4886 }
4887 
4888 /*
4889  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4890  * because it is sized independent of architecture. Unlike the other zones,
4891  * the starting point for ZONE_MOVABLE is not fixed. It may be different
4892  * in each node depending on the size of each node and how evenly kernelcore
4893  * is distributed. This helper function adjusts the zone ranges
4894  * provided by the architecture for a given node by using the end of the
4895  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4896  * zones within a node are in order of monotonic increases memory addresses
4897  */
4898 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4899                                         unsigned long zone_type,
4900                                         unsigned long node_start_pfn,
4901                                         unsigned long node_end_pfn,
4902                                         unsigned long *zone_start_pfn,
4903                                         unsigned long *zone_end_pfn)
4904 {
4905         /* Only adjust if ZONE_MOVABLE is on this node */
4906         if (zone_movable_pfn[nid]) {
4907                 /* Size ZONE_MOVABLE */
4908                 if (zone_type == ZONE_MOVABLE) {
4909                         *zone_start_pfn = zone_movable_pfn[nid];
4910                         *zone_end_pfn = min(node_end_pfn,
4911                                 arch_zone_highest_possible_pfn[movable_zone]);
4912 
4913                 /* Adjust for ZONE_MOVABLE starting within this range */
4914                 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4915                                 *zone_end_pfn > zone_movable_pfn[nid]) {
4916                         *zone_end_pfn = zone_movable_pfn[nid];
4917 
4918                 /* Check if this whole range is within ZONE_MOVABLE */
4919                 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4920                         *zone_start_pfn = *zone_end_pfn;
4921         }
4922 }
4923 
4924 /*
4925  * Return the number of pages a zone spans in a node, including holes
4926  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4927  */
4928 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4929                                         unsigned long zone_type,
4930                                         unsigned long node_start_pfn,
4931                                         unsigned long node_end_pfn,
4932                                         unsigned long *ignored)
4933 {
4934         unsigned long zone_start_pfn, zone_end_pfn;
4935 
4936         /* When hotadd a new node from cpu_up(), the node should be empty */
4937         if (!node_start_pfn && !node_end_pfn)
4938                 return 0;
4939 
4940         /* Get the start and end of the zone */
4941         zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4942         zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4943         adjust_zone_range_for_zone_movable(nid, zone_type,
4944                                 node_start_pfn, node_end_pfn,
4945                                 &zone_start_pfn, &zone_end_pfn);
4946 
4947         /* Check that this node has pages within the zone's required range */
4948         if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4949                 return 0;
4950 
4951         /* Move the zone boundaries inside the node if necessary */
4952         zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4953         zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4954 
4955         /* Return the spanned pages */
4956         return zone_end_pfn - zone_start_pfn;
4957 }
4958 
4959 /*
4960  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4961  * then all holes in the requested range will be accounted for.
4962  */
4963 unsigned long __meminit __absent_pages_in_range(int nid,
4964                                 unsigned long range_start_pfn,
4965                                 unsigned long range_end_pfn)
4966 {
4967         unsigned long nr_absent = range_end_pfn - range_start_pfn;
4968         unsigned long start_pfn, end_pfn;
4969         int i;
4970 
4971         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4972                 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4973                 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4974                 nr_absent -= end_pfn - start_pfn;
4975         }
4976         return nr_absent;
4977 }
4978 
4979 /**
4980  * absent_pages_in_range - Return number of page frames in holes within a range
4981  * @start_pfn: The start PFN to start searching for holes
4982  * @end_pfn: The end PFN to stop searching for holes
4983  *
4984  * It returns the number of pages frames in memory holes within a range.
4985  */
4986 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4987                                                         unsigned long end_pfn)
4988 {
4989         return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4990 }
4991 
4992 /* Return the number of page frames in holes in a zone on a node */
4993 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4994                                         unsigned long zone_type,
4995                                         unsigned long node_start_pfn,
4996                                         unsigned long node_end_pfn,
4997                                         unsigned long *ignored)
4998 {
4999         unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5000         unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
5001         unsigned long zone_start_pfn, zone_end_pfn;
5002 
5003         /* When hotadd a new node from cpu_up(), the node should be empty */
5004         if (!node_start_pfn && !node_end_pfn)
5005                 return 0;
5006 
5007         zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5008         zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
5009 
5010         adjust_zone_range_for_zone_movable(nid, zone_type,
5011                         node_start_pfn, node_end_pfn,
5012                         &zone_start_pfn, &zone_end_pfn);
5013         return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5014 }
5015 
5016 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5017 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
5018                                         unsigned long zone_type,
5019                                         unsigned long node_start_pfn,
5020                                         unsigned long node_end_pfn,
5021                                         unsigned long *zones_size)
5022 {
5023         return zones_size[zone_type];
5024 }
5025 
5026 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
5027                                                 unsigned long zone_type,
5028                                                 unsigned long node_start_pfn,
5029                                                 unsigned long node_end_pfn,
5030                                                 unsigned long *zholes_size)
5031 {
5032         if (!zholes_size)
5033                 return 0;
5034 
5035         return zholes_size[zone_type];
5036 }
5037 
5038 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5039 
5040 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
5041                                                 unsigned long node_start_pfn,
5042                                                 unsigned long node_end_pfn,
5043                                                 unsigned long *zones_size,
5044                                                 unsigned long *zholes_size)
5045 {
5046         unsigned long realtotalpages = 0, totalpages = 0;
5047         enum zone_type i;
5048 
5049         for (i = 0; i < MAX_NR_ZONES; i++) {
5050                 struct zone *zone = pgdat->node_zones + i;
5051                 unsigned long size, real_size;
5052 
5053                 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5054                                                   node_start_pfn,
5055                                                   node_end_pfn,
5056                                                   zones_size);
5057                 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
5058                                                   node_start_pfn, node_end_pfn,
5059                                                   zholes_size);
5060                 zone->spanned_pages = size;
5061                 zone->present_pages = real_size;
5062 
5063                 totalpages += size;
5064                 realtotalpages += real_size;
5065         }
5066 
5067         pgdat->node_spanned_pages = totalpages;
5068         pgdat->node_present_pages = realtotalpages;
5069         printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5070                                                         realtotalpages);
5071 }
5072 
5073 #ifndef CONFIG_SPARSEMEM
5074 /*
5075  * Calculate the size of the zone->blockflags rounded to an unsigned long
5076  * Start by making sure zonesize is a multiple of pageblock_order by rounding
5077  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5078  * round what is now in bits to nearest long in bits, then return it in
5079  * bytes.
5080  */
5081 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
5082 {
5083         unsigned long usemapsize;
5084 
5085         zonesize += zone_start_pfn & (pageblock_nr_pages-1);
5086         usemapsize = roundup(zonesize, pageblock_nr_pages);
5087         usemapsize = usemapsize >> pageblock_order;
5088         usemapsize *= NR_PAGEBLOCK_BITS;
5089         usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5090 
5091         return usemapsize / 8;
5092 }
5093 
5094 static void __init setup_usemap(struct pglist_data *pgdat,
5095                                 struct zone *zone,
5096                                 unsigned long zone_start_pfn,
5097                                 unsigned long zonesize)
5098 {
5099         unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
5100         zone->pageblock_flags = NULL;
5101         if (usemapsize)
5102                 zone->pageblock_flags =
5103                         memblock_virt_alloc_node_nopanic(usemapsize,
5104                                                          pgdat->node_id);
5105 }
5106 #else
5107 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5108                                 unsigned long zone_start_pfn, unsigned long zonesize) {}
5109 #endif /* CONFIG_SPARSEMEM */
5110 
5111 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5112 
5113 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5114 void __paginginit set_pageblock_order(void)
5115 {
5116         unsigned int order;
5117 
5118         /* Check that pageblock_nr_pages has not already been setup */
5119         if (pageblock_order)
5120                 return;
5121 
5122         if (HPAGE_SHIFT > PAGE_SHIFT)
5123                 order = HUGETLB_PAGE_ORDER;
5124         else
5125                 order = MAX_ORDER - 1;
5126 
5127         /*
5128          * Assume the largest contiguous order of interest is a huge page.
5129          * This value may be variable depending on boot parameters on IA64 and
5130          * powerpc.
5131          */
5132         pageblock_order = order;
5133 }
5134 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5135 
5136 /*
5137  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5138  * is unused as pageblock_order is set at compile-time. See
5139  * include/linux/pageblock-flags.h for the values of pageblock_order based on
5140  * the kernel config
5141  */
5142 void __paginginit set_pageblock_order(void)
5143 {
5144 }
5145 
5146 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5147 
5148 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5149                                                    unsigned long present_pages)
5150 {
5151         unsigned long pages = spanned_pages;
5152 
5153         /*
5154          * Provide a more accurate estimation if there are holes within
5155          * the zone and SPARSEMEM is in use. If there are holes within the
5156          * zone, each populated memory region may cost us one or two extra
5157          * memmap pages due to alignment because memmap pages for each
5158          * populated regions may not naturally algined on page boundary.
5159          * So the (present_pages >> 4) heuristic is a tradeoff for that.
5160          */
5161         if (spanned_pages > present_pages + (present_pages >> 4) &&
5162             IS_ENABLED(CONFIG_SPARSEMEM))
5163                 pages = present_pages;
5164 
5165         return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5166 }
5167 
5168 /*
5169  * Set up the zone data structures:
5170  *   - mark all pages reserved
5171  *   - mark all memory queues empty
5172  *   - clear the memory bitmaps
5173  *
5174  * NOTE: pgdat should get zeroed by caller.
5175  */
5176 static void __paginginit free_area_init_core(struct pglist_data *pgdat)
5177 {
5178         enum zone_type j;
5179         int nid = pgdat->node_id;
5180         unsigned long zone_start_pfn = pgdat->node_start_pfn;
5181         int ret;
5182 
5183         pgdat_resize_init(pgdat);
5184 #ifdef CONFIG_NUMA_BALANCING
5185         spin_lock_init(&pgdat->numabalancing_migrate_lock);
5186         pgdat->numabalancing_migrate_nr_pages = 0;
5187         pgdat->numabalancing_migrate_next_window = jiffies;
5188 #endif
5189         init_waitqueue_head(&pgdat->kswapd_wait);
5190         init_waitqueue_head(&pgdat->pfmemalloc_wait);
5191         pgdat_page_ext_init(pgdat);
5192 
5193         for (j = 0; j < MAX_NR_ZONES; j++) {
5194                 struct zone *zone = pgdat->node_zones + j;
5195                 unsigned long size, realsize, freesize, memmap_pages;
5196 
5197                 size = zone->spanned_pages;
5198                 realsize = freesize = zone->present_pages;
5199 
5200                 /*
5201                  * Adjust freesize so that it accounts for how much memory
5202                  * is used by this zone for memmap. This affects the watermark
5203                  * and per-cpu initialisations
5204                  */
5205                 memmap_pages = calc_memmap_size(size, realsize);
5206                 if (!is_highmem_idx(j)) {
5207                         if (freesize >= memmap_pages) {
5208                                 freesize -= memmap_pages;
5209                                 if (memmap_pages)
5210                                         printk(KERN_DEBUG
5211                                                "  %s zone: %lu pages used for memmap\n",
5212                                                zone_names[j], memmap_pages);
5213                         } else
5214                                 printk(KERN_WARNING
5215                                         "  %s zone: %lu pages exceeds freesize %lu\n",
5216                                         zone_names[j], memmap_pages, freesize);
5217                 }
5218 
5219                 /* Account for reserved pages */
5220                 if (j == 0 && freesize > dma_reserve) {
5221                         freesize -= dma_reserve;
5222                         printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
5223                                         zone_names[0], dma_reserve);
5224                 }
5225 
5226                 if (!is_highmem_idx(j))
5227                         nr_kernel_pages += freesize;
5228                 /* Charge for highmem memmap if there are enough kernel pages */
5229                 else if (nr_kernel_pages > memmap_pages * 2)
5230                         nr_kernel_pages -= memmap_pages;
5231                 nr_all_pages += freesize;
5232 
5233                 /*
5234                  * Set an approximate value for lowmem here, it will be adjusted
5235                  * when the bootmem allocator frees pages into the buddy system.
5236                  * And all highmem pages will be managed by the buddy system.
5237                  */
5238                 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
5239 #ifdef CONFIG_NUMA
5240                 zone->node = nid;
5241                 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
5242                                                 / 100;
5243                 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
5244 #endif
5245                 zone->name = zone_names[j];
5246                 spin_lock_init(&zone->lock);
5247                 spin_lock_init(&zone->lru_lock);
5248                 zone_seqlock_init(zone);
5249                 zone->zone_pgdat = pgdat;
5250                 zone_pcp_init(zone);
5251 
5252                 /* For bootup, initialized properly in watermark setup */
5253                 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5254 
5255                 lruvec_init(&zone->lruvec);
5256                 if (!size)
5257                         continue;
5258 
5259                 set_pageblock_order();
5260                 setup_usemap(pgdat, zone, zone_start_pfn, size);
5261                 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
5262                 BUG_ON(ret);
5263                 memmap_init(size, nid, j, zone_start_pfn);
5264                 zone_start_pfn += size;
5265         }
5266 }
5267 
5268 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
5269 {
5270         unsigned long __maybe_unused start = 0;
5271         unsigned long __maybe_unused offset = 0;
5272 
5273         /* Skip empty nodes */
5274         if (!pgdat->node_spanned_pages)
5275                 return;
5276 
5277 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5278         start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5279         offset = pgdat->node_start_pfn - start;
5280         /* ia64 gets its own node_mem_map, before this, without bootmem */
5281         if (!pgdat->node_mem_map) {
5282                 unsigned long size, end;
5283                 struct page *map;
5284 
5285                 /*
5286                  * The zone's endpoints aren't required to be MAX_ORDER
5287                  * aligned but the node_mem_map endpoints must be in order
5288                  * for the buddy allocator to function correctly.
5289                  */
5290                 end = pgdat_end_pfn(pgdat);
5291                 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5292                 size =  (end - start) * sizeof(struct page);
5293                 map = alloc_remap(pgdat->node_id, size);
5294                 if (!map)
5295                         map = memblock_virt_alloc_node_nopanic(size,
5296                                                                pgdat->node_id);
5297                 pgdat->node_mem_map = map + offset;
5298         }
5299 #ifndef CONFIG_NEED_MULTIPLE_NODES
5300         /*
5301          * With no DISCONTIG, the global mem_map is just set as node 0's
5302          */
5303         if (pgdat == NODE_DATA(0)) {
5304                 mem_map = NODE_DATA(0)->node_mem_map;
5305 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5306                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
5307                         mem_map -= offset;
5308 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5309         }
5310 #endif
5311 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5312 }
5313 
5314 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5315                 unsigned long node_start_pfn, unsigned long *zholes_size)
5316 {
5317         pg_data_t *pgdat = NODE_DATA(nid);
5318         unsigned long start_pfn = 0;
5319         unsigned long end_pfn = 0;
5320 
5321         /* pg_data_t should be reset to zero when it's allocated */
5322         WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
5323 
5324         reset_deferred_meminit(pgdat);
5325         pgdat->node_id = nid;
5326         pgdat->node_start_pfn = node_start_pfn;
5327 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5328         get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
5329         pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5330                 (u64)start_pfn << PAGE_SHIFT,
5331                 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
5332 #endif
5333         calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5334                                   zones_size, zholes_size);
5335 
5336         alloc_node_mem_map(pgdat);
5337 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5338         printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5339                 nid, (unsigned long)pgdat,
5340                 (unsigned long)pgdat->node_mem_map);
5341 #endif
5342 
5343         free_area_init_core(pgdat);
5344 }
5345 
5346 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5347 
5348 #if MAX_NUMNODES > 1
5349 /*
5350  * Figure out the number of possible node ids.
5351  */
5352 void __init setup_nr_node_ids(void)
5353 {
5354         unsigned int highest;
5355 
5356         highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
5357         nr_node_ids = highest + 1;
5358 }
5359 #endif
5360 
5361 /**
5362  * node_map_pfn_alignment - determine the maximum internode alignment
5363  *
5364  * This function should be called after node map is populated and sorted.
5365  * It calculates the maximum power of two alignment which can distinguish
5366  * all the nodes.
5367  *
5368  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5369  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
5370  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
5371  * shifted, 1GiB is enough and this function will indicate so.
5372  *
5373  * This is used to test whether pfn -> nid mapping of the chosen memory
5374  * model has fine enough granularity to avoid incorrect mapping for the
5375  * populated node map.
5376  *
5377  * Returns the determined alignment in pfn's.  0 if there is no alignment
5378  * requirement (single node).
5379  */
5380 unsigned long __init node_map_pfn_alignment(void)
5381 {
5382         unsigned long accl_mask = 0, last_end = 0;
5383         unsigned long start, end, mask;
5384         int last_nid = -1;
5385         int i, nid;
5386 
5387         for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5388                 if (!start || last_nid < 0 || last_nid == nid) {
5389                         last_nid = nid;
5390                         last_end = end;
5391                         continue;
5392                 }
5393 
5394                 /*
5395                  * Start with a mask granular enough to pin-point to the
5396                  * start pfn and tick off bits one-by-one until it becomes
5397                  * too coarse to separate the current node from the last.
5398                  */
5399                 mask = ~((1 << __ffs(start)) - 1);
5400                 while (mask && last_end <= (start & (mask << 1)))
5401                         mask <<= 1;
5402 
5403                 /* accumulate all internode masks */
5404                 accl_mask |= mask;
5405         }
5406 
5407         /* convert mask to number of pages */
5408         return ~accl_mask + 1;
5409 }
5410 
5411 /* Find the lowest pfn for a node */
5412 static unsigned long __init find_min_pfn_for_node(int nid)
5413 {
5414         unsigned long min_pfn = ULONG_MAX;
5415         unsigned long start_pfn;
5416         int i;
5417 
5418         for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5419                 min_pfn = min(min_pfn, start_pfn);
5420 
5421         if (min_pfn == ULONG_MAX) {
5422                 printk(KERN_WARNING
5423                         "Could not find start_pfn for node %d\n", nid);
5424                 return 0;
5425         }
5426 
5427         return min_pfn;
5428 }
5429 
5430 /**
5431  * find_min_pfn_with_active_regions - Find the minimum PFN registered
5432  *
5433  * It returns the minimum PFN based on information provided via
5434  * memblock_set_node().
5435  */
5436 unsigned long __init find_min_pfn_with_active_regions(void)
5437 {
5438         return find_min_pfn_for_node(MAX_NUMNODES);
5439 }
5440 
5441 /*
5442  * early_calculate_totalpages()
5443  * Sum pages in active regions for movable zone.
5444  * Populate N_MEMORY for calculating usable_nodes.
5445  */
5446 static unsigned long __init early_calculate_totalpages(void)
5447 {
5448         unsigned long totalpages = 0;
5449         unsigned long start_pfn, end_pfn;
5450         int i, nid;
5451 
5452         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5453                 unsigned long pages = end_pfn - start_pfn;
5454 
5455                 totalpages += pages;
5456                 if (pages)
5457                         node_set_state(nid, N_MEMORY);
5458         }
5459         return totalpages;
5460 }
5461 
5462 /*
5463  * Find the PFN the Movable zone begins in each node. Kernel memory
5464  * is spread evenly between nodes as long as the nodes have enough
5465  * memory. When they don't, some nodes will have more kernelcore than
5466  * others
5467  */
5468 static void __init find_zone_movable_pfns_for_nodes(void)
5469 {
5470         int i, nid;
5471         unsigned long usable_startpfn;
5472         unsigned long kernelcore_node, kernelcore_remaining;
5473         /* save the state before borrow the nodemask */
5474         nodemask_t saved_node_state = node_states[N_MEMORY];
5475         unsigned long totalpages = early_calculate_totalpages();
5476         int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5477         struct memblock_region *r;
5478 
5479         /* Need to find movable_zone earlier when movable_node is specified. */
5480         find_usable_zone_for_movable();
5481 
5482         /*
5483          * If movable_node is specified, ignore kernelcore and movablecore
5484          * options.
5485          */
5486         if (movable_node_is_enabled()) {
5487                 for_each_memblock(memory, r) {
5488                         if (!memblock_is_hotpluggable(r))
5489                                 continue;
5490 
5491                         nid = r->nid;
5492 
5493                         usable_startpfn = PFN_DOWN(r->base);
5494                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5495                                 min(usable_startpfn, zone_movable_pfn[nid]) :
5496                                 usable_startpfn;
5497                 }
5498 
5499                 goto out2;
5500         }
5501 
5502         /*
5503          * If movablecore=nn[KMG] was specified, calculate what size of
5504          * kernelcore that corresponds so that memory usable for
5505          * any allocation type is evenly spread. If both kernelcore
5506          * and movablecore are specified, then the value of kernelcore
5507          * will be used for required_kernelcore if it's greater than
5508          * what movablecore would have allowed.
5509          */
5510         if (required_movablecore) {
5511                 unsigned long corepages;
5512 
5513                 /*
5514                  * Round-up so that ZONE_MOVABLE is at least as large as what
5515                  * was requested by the user
5516                  */
5517                 required_movablecore =
5518                         roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5519                 required_movablecore = min(totalpages, required_movablecore);
5520                 corepages = totalpages - required_movablecore;
5521 
5522                 required_kernelcore = max(required_kernelcore, corepages);
5523         }
5524 
5525         /*
5526          * If kernelcore was not specified or kernelcore size is larger
5527          * than totalpages, there is no ZONE_MOVABLE.
5528          */
5529         if (!required_kernelcore || required_kernelcore >= totalpages)
5530                 goto out;
5531 
5532         /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5533         usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5534 
5535 restart:
5536         /* Spread kernelcore memory as evenly as possible throughout nodes */
5537         kernelcore_node = required_kernelcore / usable_nodes;
5538         for_each_node_state(nid, N_MEMORY) {
5539                 unsigned long start_pfn, end_pfn;
5540 
5541                 /*
5542                  * Recalculate kernelcore_node if the division per node
5543                  * now exceeds what is necessary to satisfy the requested
5544                  * amount of memory for the kernel
5545                  */
5546                 if (required_kernelcore < kernelcore_node)
5547                         kernelcore_node = required_kernelcore / usable_nodes;
5548 
5549                 /*
5550                  * As the map is walked, we track how much memory is usable
5551                  * by the kernel using kernelcore_remaining. When it is
5552                  * 0, the rest of the node is usable by ZONE_MOVABLE
5553                  */
5554                 kernelcore_remaining = kernelcore_node;
5555 
5556                 /* Go through each range of PFNs within this node */
5557                 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5558                         unsigned long size_pages;
5559 
5560                         start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5561                         if (start_pfn >= end_pfn)
5562                                 continue;
5563 
5564                         /* Account for what is only usable for kernelcore */
5565                         if (start_pfn < usable_startpfn) {
5566                                 unsigned long kernel_pages;
5567                                 kernel_pages = min(end_pfn, usable_startpfn)
5568                                                                 - start_pfn;
5569 
5570                                 kernelcore_remaining -= min(kernel_pages,
5571                                                         kernelcore_remaining);
5572                                 required_kernelcore -= min(kernel_pages,
5573                                                         required_kernelcore);
5574 
5575                                 /* Continue if range is now fully accounted */
5576                                 if (end_pfn <= usable_startpfn) {
5577 
5578                                         /*
5579                                          * Push zone_movable_pfn to the end so
5580                                          * that if we have to rebalance
5581                                          * kernelcore across nodes, we will
5582                                          * not double account here
5583                                          */
5584                                         zone_movable_pfn[nid] = end_pfn;
5585                                         continue;
5586                                 }
5587                                 start_pfn = usable_startpfn;
5588                         }
5589 
5590                         /*
5591                          * The usable PFN range for ZONE_MOVABLE is from
5592                          * start_pfn->end_pfn. Calculate size_pages as the
5593                          * number of pages used as kernelcore
5594                          */
5595                         size_pages = end_pfn - start_pfn;
5596                         if (size_pages > kernelcore_remaining)
5597                                 size_pages = kernelcore_remaining;
5598                         zone_movable_pfn[nid] = start_pfn + size_pages;
5599 
5600                         /*
5601                          * Some kernelcore has been met, update counts and
5602                          * break if the kernelcore for this node has been
5603                          * satisfied
5604                          */
5605                         required_kernelcore -= min(required_kernelcore,
5606                                                                 size_pages);
5607                         kernelcore_remaining -= size_pages;
5608                         if (!kernelcore_remaining)
5609                                 break;
5610                 }
5611         }
5612 
5613         /*
5614          * If there is still required_kernelcore, we do another pass with one
5615          * less node in the count. This will push zone_movable_pfn[nid] further
5616          * along on the nodes that still have memory until kernelcore is
5617          * satisfied
5618          */
5619         usable_nodes--;
5620         if (usable_nodes && required_kernelcore > usable_nodes)
5621                 goto restart;
5622 
5623 out2:
5624         /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5625         for (nid = 0; nid < MAX_NUMNODES; nid++)
5626                 zone_movable_pfn[nid] =
5627                         roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5628 
5629 out:
5630         /* restore the node_state */
5631         node_states[N_MEMORY] = saved_node_state;
5632 }
5633 
5634 /* Any regular or high memory on that node ? */
5635 static void check_for_memory(pg_data_t *pgdat, int nid)
5636 {
5637         enum zone_type zone_type;
5638 
5639         if (N_MEMORY == N_NORMAL_MEMORY)
5640                 return;
5641 
5642         for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5643                 struct zone *zone = &pgdat->node_zones[zone_type];
5644                 if (populated_zone(zone)) {
5645                         node_set_state(nid, N_HIGH_MEMORY);
5646                         if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5647                             zone_type <= ZONE_NORMAL)
5648                                 node_set_state(nid, N_NORMAL_MEMORY);
5649                         break;
5650                 }
5651         }
5652 }
5653 
5654 /**
5655  * free_area_init_nodes - Initialise all pg_data_t and zone data
5656  * @max_zone_pfn: an array of max PFNs for each zone
5657  *
5658  * This will call free_area_init_node() for each active node in the system.
5659  * Using the page ranges provided by memblock_set_node(), the size of each
5660  * zone in each node and their holes is calculated. If the maximum PFN
5661  * between two adjacent zones match, it is assumed that the zone is empty.
5662  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5663  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5664  * starts where the previous one ended. For example, ZONE_DMA32 starts
5665  * at arch_max_dma_pfn.
5666  */
5667 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5668 {
5669         unsigned long start_pfn, end_pfn;
5670         int i, nid;
5671 
5672         /* Record where the zone boundaries are */
5673         memset(arch_zone_lowest_possible_pfn, 0,
5674                                 sizeof(arch_zone_lowest_possible_pfn));
5675         memset(arch_zone_highest_possible_pfn, 0,
5676                                 sizeof(arch_zone_highest_possible_pfn));
5677         arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5678         arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5679         for (i = 1; i < MAX_NR_ZONES; i++) {
5680                 if (i == ZONE_MOVABLE)
5681                         continue;
5682                 arch_zone_lowest_possible_pfn[i] =
5683                         arch_zone_highest_possible_pfn[i-1];
5684                 arch_zone_highest_possible_pfn[i] =
5685                         max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5686         }
5687         arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5688         arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5689 
5690         /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5691         memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5692         find_zone_movable_pfns_for_nodes();
5693 
5694         /* Print out the zone ranges */
5695         pr_info("Zone ranges:\n");
5696         for (i = 0; i < MAX_NR_ZONES; i++) {
5697                 if (i == ZONE_MOVABLE)
5698                         continue;
5699                 pr_info("  %-8s ", zone_names[i]);
5700                 if (arch_zone_lowest_possible_pfn[i] ==
5701                                 arch_zone_highest_possible_pfn[i])
5702                         pr_cont("empty\n");
5703                 else
5704                         pr_cont("[mem %#018Lx-%#018Lx]\n",
5705                                 (u64)arch_zone_lowest_possible_pfn[i]
5706                                         << PAGE_SHIFT,
5707                                 ((u64)arch_zone_highest_possible_pfn[i]
5708                                         << PAGE_SHIFT) - 1);
5709         }
5710 
5711         /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5712         pr_info("Movable zone start for each node\n");
5713         for (i = 0; i < MAX_NUMNODES; i++) {
5714                 if (zone_movable_pfn[i])
5715                         pr_info("  Node %d: %#018Lx\n", i,
5716                                (u64)zone_movable_pfn[i] << PAGE_SHIFT);
5717         }
5718 
5719         /* Print out the early node map */
5720         pr_info("Early memory node ranges\n");
5721         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5722                 pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
5723                         (u64)start_pfn << PAGE_SHIFT,
5724                         ((u64)end_pfn << PAGE_SHIFT) - 1);
5725 
5726         /* Initialise every node */
5727         mminit_verify_pageflags_layout();
5728         setup_nr_node_ids();
5729         for_each_online_node(nid) {
5730                 pg_data_t *pgdat = NODE_DATA(nid);
5731                 free_area_init_node(nid, NULL,
5732                                 find_min_pfn_for_node(nid), NULL);
5733 
5734                 /* Any memory on that node */
5735                 if (pgdat->node_present_pages)
5736                         node_set_state(nid, N_MEMORY);
5737                 check_for_memory(pgdat, nid);
5738         }
5739 }
5740 
5741 static int __init cmdline_parse_core(char *p, unsigned long *core)
5742 {
5743         unsigned long long coremem;
5744         if (!p)
5745                 return -EINVAL;
5746 
5747         coremem = memparse(p, &p);
5748         *core = coremem >> PAGE_SHIFT;
5749 
5750         /* Paranoid check that UL is enough for the coremem value */
5751         WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5752 
5753         return 0;
5754 }
5755 
5756 /*
5757  * kernelcore=size sets the amount of memory for use for allocations that
5758  * cannot be reclaimed or migrated.
5759  */
5760 static int __init cmdline_parse_kernelcore(char *p)
5761 {
5762         return cmdline_parse_core(p, &required_kernelcore);
5763 }
5764 
5765 /*
5766  * movablecore=size sets the amount of memory for use for allocations that
5767  * can be reclaimed or migrated.
5768  */
5769 static int __init cmdline_parse_movablecore(char *p)
5770 {
5771         return cmdline_parse_core(p, &required_movablecore);
5772 }
5773 
5774 early_param("kernelcore", cmdline_parse_kernelcore);
5775 early_param("movablecore", cmdline_parse_movablecore);
5776 
5777 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5778 
5779 void adjust_managed_page_count(struct page *page, long count)
5780 {
5781         spin_lock(&managed_page_count_lock);
5782         page_zone(page)->managed_pages += count;
5783         totalram_pages += count;
5784 #ifdef CONFIG_HIGHMEM
5785         if (PageHighMem(page))
5786                 totalhigh_pages += count;
5787 #endif
5788         spin_unlock(&managed_page_count_lock);
5789 }
5790 EXPORT_SYMBOL(adjust_managed_page_count);
5791 
5792 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
5793 {
5794         void *pos;
5795         unsigned long pages = 0;
5796 
5797         start = (void *)PAGE_ALIGN((unsigned long)start);
5798         end = (void *)((unsigned long)end & PAGE_MASK);
5799         for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
5800                 if ((unsigned int)poison <= 0xFF)
5801                         memset(pos, poison, PAGE_SIZE);
5802                 free_reserved_page(virt_to_page(pos));
5803         }
5804 
5805         if (pages && s)
5806                 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5807                         s, pages << (PAGE_SHIFT - 10), start, end);
5808 
5809         return pages;
5810 }
5811 EXPORT_SYMBOL(free_reserved_area);
5812 
5813 #ifdef  CONFIG_HIGHMEM
5814 void free_highmem_page(struct page *page)
5815 {
5816         __free_reserved_page(page);
5817         totalram_pages++;
5818         page_zone(page)->managed_pages++;
5819         totalhigh_pages++;
5820 }
5821 #endif
5822 
5823 
5824 void __init mem_init_print_info(const char *str)
5825 {
5826         unsigned long physpages, codesize, datasize, rosize, bss_size;
5827         unsigned long init_code_size, init_data_size;
5828 
5829         physpages = get_num_physpages();
5830         codesize = _etext - _stext;
5831         datasize = _edata - _sdata;
5832         rosize = __end_rodata - __start_rodata;
5833         bss_size = __bss_stop - __bss_start;
5834         init_data_size = __init_end - __init_begin;
5835         init_code_size = _einittext - _sinittext;
5836 
5837         /*
5838          * Detect special cases and adjust section sizes accordingly:
5839          * 1) .init.* may be embedded into .data sections
5840          * 2) .init.text.* may be out of [__init_begin, __init_end],
5841          *    please refer to arch/tile/kernel/vmlinux.lds.S.
5842          * 3) .rodata.* may be embedded into .text or .data sections.
5843          */
5844 #define adj_init_size(start, end, size, pos, adj) \
5845         do { \
5846                 if (start <= pos && pos < end && size > adj) \
5847                         size -= adj; \
5848         } while (0)
5849 
5850         adj_init_size(__init_begin, __init_end, init_data_size,
5851                      _sinittext, init_code_size);
5852         adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5853         adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5854         adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5855         adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5856 
5857 #undef  adj_init_size
5858 
5859         pr_info("Memory: %luK/%luK available "
5860                "(%luK kernel code, %luK rwdata, %luK rodata, "
5861                "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
5862 #ifdef  CONFIG_HIGHMEM
5863                ", %luK highmem"
5864 #endif
5865                "%s%s)\n",
5866                nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5867                codesize >> 10, datasize >> 10, rosize >> 10,
5868                (init_data_size + init_code_size) >> 10, bss_size >> 10,
5869                (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT-10),
5870                totalcma_pages << (PAGE_SHIFT-10),
5871 #ifdef  CONFIG_HIGHMEM
5872                totalhigh_pages << (PAGE_SHIFT-10),
5873 #endif
5874                str ? ", " : "", str ? str : "");
5875 }
5876 
5877 /**
5878  * set_dma_reserve - set the specified number of pages reserved in the first zone
5879  * @new_dma_reserve: The number of pages to mark reserved
5880  *
5881  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
5882  * In the DMA zone, a significant percentage may be consumed by kernel image
5883  * and other unfreeable allocations which can skew the watermarks badly. This
5884  * function may optionally be used to account for unfreeable pages in the
5885  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5886  * smaller per-cpu batchsize.
5887  */
5888 void __init set_dma_reserve(unsigned long new_dma_reserve)
5889 {
5890         dma_reserve = new_dma_reserve;
5891 }
5892 
5893 void __init free_area_init(unsigned long *zones_size)
5894 {
5895         free_area_init_node(0, zones_size,
5896                         __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5897 }
5898 
5899 static int page_alloc_cpu_notify(struct notifier_block *self,
5900                                  unsigned long action, void *hcpu)
5901 {
5902         int cpu = (unsigned long)hcpu;
5903 
5904         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
5905                 lru_add_drain_cpu(cpu);
5906                 drain_pages(cpu);
5907 
5908                 /*
5909                  * Spill the event counters of the dead processor
5910                  * into the current processors event counters.
5911                  * This artificially elevates the count of the current
5912                  * processor.
5913                  */
5914                 vm_events_fold_cpu(cpu);
5915 
5916                 /*
5917                  * Zero the differential counters of the dead processor
5918                  * so that the vm statistics are consistent.
5919                  *
5920                  * This is only okay since the processor is dead and cannot
5921                  * race with what we are doing.
5922                  */
5923                 cpu_vm_stats_fold(cpu);
5924         }
5925         return NOTIFY_OK;
5926 }
5927 
5928 void __init page_alloc_init(void)
5929 {
5930         hotcpu_notifier(page_alloc_cpu_notify, 0);
5931 }
5932 
5933 /*
5934  * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
5935  *      or min_free_kbytes changes.
5936  */
5937 static void calculate_totalreserve_pages(void)
5938 {
5939         struct pglist_data *pgdat;
5940         unsigned long reserve_pages = 0;
5941         enum zone_type i, j;
5942 
5943         for_each_online_pgdat(pgdat) {
5944                 for (i = 0; i < MAX_NR_ZONES; i++) {
5945                         struct zone *zone = pgdat->node_zones + i;
5946                         long max = 0;
5947 
5948                         /* Find valid and maximum lowmem_reserve in the zone */
5949                         for (j = i; j < MAX_NR_ZONES; j++) {
5950                                 if (zone->lowmem_reserve[j] > max)
5951                                         max = zone->lowmem_reserve[j];
5952                         }
5953 
5954                         /* we treat the high watermark as reserved pages. */
5955                         max += high_wmark_pages(zone);
5956 
5957                         if (max > zone->managed_pages)
5958                                 max = zone->managed_pages;
5959                         reserve_pages += max;
5960                         /*
5961                          * Lowmem reserves are not available to
5962                          * GFP_HIGHUSER page cache allocations and
5963                          * kswapd tries to balance zones to their high
5964                          * watermark.  As a result, neither should be
5965                          * regarded as dirtyable memory, to prevent a
5966                          * situation where reclaim has to clean pages
5967                          * in order to balance the zones.
5968                          */
5969                         zone->dirty_balance_reserve = max;
5970                 }
5971         }
5972         dirty_balance_reserve = reserve_pages;
5973         totalreserve_pages = reserve_pages;
5974 }
5975 
5976 /*
5977  * setup_per_zone_lowmem_reserve - called whenever
5978  *      sysctl_lowmem_reserve_ratio changes.  Ensures that each zone
5979  *      has a correct pages reserved value, so an adequate number of
5980  *      pages are left in the zone after a successful __alloc_pages().
5981  */
5982 static void setup_per_zone_lowmem_reserve(void)
5983 {
5984         struct pglist_data *pgdat;
5985         enum zone_type j, idx;
5986 
5987         for_each_online_pgdat(pgdat) {
5988                 for (j = 0; j < MAX_NR_ZONES; j++) {
5989                         struct zone *zone = pgdat->node_zones + j;
5990                         unsigned long managed_pages = zone->managed_pages;
5991 
5992                         zone->lowmem_reserve[j] = 0;
5993 
5994                         idx = j;
5995                         while (idx) {
5996                                 struct zone *lower_zone;
5997 
5998                                 idx--;
5999 
6000                                 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6001                                         sysctl_lowmem_reserve_ratio[idx] = 1;
6002 
6003                                 lower_zone = pgdat->node_zones + idx;
6004                                 lower_zone->lowmem_reserve[j] = managed_pages /
6005                                         sysctl_lowmem_reserve_ratio[idx];
6006                                 managed_pages += lower_zone->managed_pages;
6007                         }
6008                 }
6009         }
6010 
6011         /* update totalreserve_pages */
6012         calculate_totalreserve_pages();
6013 }
6014 
6015 static void __setup_per_zone_wmarks(void)
6016 {
6017         unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6018         unsigned long lowmem_pages = 0;
6019         struct zone *zone;
6020         unsigned long flags;
6021 
6022         /* Calculate total number of !ZONE_HIGHMEM pages */
6023         for_each_zone(zone) {
6024                 if (!is_highmem(zone))
6025                         lowmem_pages += zone->managed_pages;
6026         }
6027 
6028         for_each_zone(zone) {
6029                 u64 tmp;
6030 
6031                 spin_lock_irqsave(&zone->lock, flags);
6032                 tmp = (u64)pages_min * zone->managed_pages;
6033                 do_div(tmp, lowmem_pages);
6034                 if (is_highmem(zone)) {
6035                         /*
6036                          * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6037                          * need highmem pages, so cap pages_min to a small
6038                          * value here.
6039                          *
6040                          * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6041                          * deltas control asynch page reclaim, and so should
6042                          * not be capped for highmem.
6043                          */
6044                         unsigned long min_pages;
6045 
6046                         min_pages = zone->managed_pages / 1024;
6047                         min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
6048                         zone->watermark[WMARK_MIN] = min_pages;
6049                 } else {
6050                         /*
6051                          * If it's a lowmem zone, reserve a number of pages
6052                          * proportionate to the zone's size.
6053                          */
6054                         zone->watermark[WMARK_MIN] = tmp;
6055                 }
6056 
6057                 zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + (tmp >> 2);
6058                 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
6059 
6060                 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
6061                         high_wmark_pages(zone) - low_wmark_pages(zone) -
6062                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
6063 
6064                 spin_unlock_irqrestore(&zone->lock, flags);
6065         }
6066 
6067         /* update totalreserve_pages */
6068         calculate_totalreserve_pages();
6069 }
6070 
6071 /**
6072  * setup_per_zone_wmarks - called when min_free_kbytes changes
6073  * or when memory is hot-{added|removed}
6074  *
6075  * Ensures that the watermark[min,low,high] values for each zone are set
6076  * correctly with respect to min_free_kbytes.
6077  */
6078 void setup_per_zone_wmarks(void)
6079 {
6080         mutex_lock(&zonelists_mutex);
6081         __setup_per_zone_wmarks();
6082         mutex_unlock(&zonelists_mutex);
6083 }
6084 
6085 /*
6086  * The inactive anon list should be small enough that the VM never has to
6087  * do too much work, but large enough that each inactive page has a chance
6088  * to be referenced again before it is swapped out.
6089  *
6090  * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6091  * INACTIVE_ANON pages on this zone's LRU, maintained by the
6092  * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6093  * the anonymous pages are kept on the inactive list.
6094  *
6095  * total     target    max
6096  * memory    ratio     inactive anon
6097  * -------------------------------------
6098  *   10MB       1         5MB
6099  *  100MB       1        50MB
6100  *    1GB       3       250MB
6101  *   10GB      10       0.9GB
6102  *  100GB      31         3GB
6103  *    1TB     101        10GB
6104  *   10TB     320        32GB
6105  */
6106 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
6107 {
6108         unsigned int gb, ratio;
6109 
6110         /* Zone size in gigabytes */
6111         gb = zone->managed_pages >> (30 - PAGE_SHIFT);
6112         if (gb)
6113                 ratio = int_sqrt(10 * gb);
6114         else
6115                 ratio = 1;
6116 
6117         zone->inactive_ratio = ratio;
6118 }
6119 
6120 static void __meminit setup_per_zone_inactive_ratio(void)
6121 {
6122         struct zone *zone;
6123 
6124         for_each_zone(zone)
6125                 calculate_zone_inactive_ratio(zone);
6126 }
6127 
6128 /*
6129  * Initialise min_free_kbytes.
6130  *
6131  * For small machines we want it small (128k min).  For large machines
6132  * we want it large (64MB max).  But it is not linear, because network
6133  * bandwidth does not increase linearly with machine size.  We use
6134  *
6135  *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6136  *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
6137  *
6138  * which yields
6139  *
6140  * 16MB:        512k
6141  * 32MB:        724k
6142  * 64MB:        1024k
6143  * 128MB:       1448k
6144  * 256MB:       2048k
6145  * 512MB:       2896k
6146  * 1024MB:      4096k
6147  * 2048MB:      5792k
6148  * 4096MB:      8192k
6149  * 8192MB:      11584k
6150  * 16384MB:     16384k
6151  */
6152 int __meminit init_per_zone_wmark_min(void)
6153 {
6154         unsigned long lowmem_kbytes;
6155         int new_min_free_kbytes;
6156 
6157         lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
6158         new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
6159 
6160         if (new_min_free_kbytes > user_min_free_kbytes) {
6161                 min_free_kbytes = new_min_free_kbytes;
6162                 if (min_free_kbytes < 128)
6163                         min_free_kbytes = 128;
6164                 if (min_free_kbytes > 65536)
6165                         min_free_kbytes = 65536;
6166         } else {
6167                 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6168                                 new_min_free_kbytes, user_min_free_kbytes);
6169         }
6170         setup_per_zone_wmarks();
6171         refresh_zone_stat_thresholds();
6172         setup_per_zone_lowmem_reserve();
6173         setup_per_zone_inactive_ratio();
6174         return 0;
6175 }
6176 module_init(init_per_zone_wmark_min)
6177 
6178 /*
6179  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6180  *      that we can call two helper functions whenever min_free_kbytes
6181  *      changes.
6182  */
6183 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
6184         void __user *buffer, size_t *length, loff_t *ppos)
6185 {
6186         int rc;
6187 
6188         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6189         if (rc)
6190                 return rc;
6191 
6192         if (write) {
6193                 user_min_free_kbytes = min_free_kbytes;
6194                 setup_per_zone_wmarks();
6195         }
6196         return 0;
6197 }
6198 
6199 #ifdef CONFIG_NUMA
6200 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
6201         void __user *buffer, size_t *length, loff_t *ppos)
6202 {
6203         struct zone *zone;
6204         int rc;
6205 
6206         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6207         if (rc)
6208                 return rc;
6209 
6210         for_each_zone(zone)
6211                 zone->min_unmapped_pages = (zone->managed_pages *
6212                                 sysctl_min_unmapped_ratio) / 100;
6213         return 0;
6214 }
6215 
6216 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
6217         void __user *buffer, size_t *length, loff_t *ppos)
6218 {
6219         struct zone *zone;
6220         int rc;
6221 
6222         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6223         if (rc)
6224                 return rc;
6225 
6226         for_each_zone(zone)
6227                 zone->min_slab_pages = (zone->managed_pages *
6228                                 sysctl_min_slab_ratio) / 100;
6229         return 0;
6230 }
6231 #endif
6232 
6233 /*
6234  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6235  *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6236  *      whenever sysctl_lowmem_reserve_ratio changes.
6237  *
6238  * The reserve ratio obviously has absolutely no relation with the
6239  * minimum watermarks. The lowmem reserve ratio can only make sense
6240  * if in function of the boot time zone sizes.
6241  */
6242 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
6243         void __user *buffer, size_t *length, loff_t *ppos)
6244 {
6245         proc_dointvec_minmax(table, write, buffer, length, ppos);
6246         setup_per_zone_lowmem_reserve();
6247         return 0;
6248 }
6249 
6250 /*
6251  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6252  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
6253  * pagelist can have before it gets flushed back to buddy allocator.
6254  */
6255 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
6256         void __user *buffer, size_t *length, loff_t *ppos)
6257 {
6258         struct zone *zone;
6259         int old_percpu_pagelist_fraction;
6260         int ret;
6261 
6262         mutex_lock(&pcp_batch_high_lock);
6263         old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6264 
6265         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6266         if (!write || ret < 0)
6267                 goto out;
6268 
6269         /* Sanity checking to avoid pcp imbalance */
6270         if (percpu_pagelist_fraction &&
6271             percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6272                 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6273                 ret = -EINVAL;
6274                 goto out;
6275         }
6276 
6277         /* No change? */
6278         if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6279                 goto out;
6280 
6281         for_each_populated_zone(zone) {
6282                 unsigned int cpu;
6283 
6284                 for_each_possible_cpu(cpu)
6285                         pageset_set_high_and_batch(zone,
6286                                         per_cpu_ptr(zone->pageset, cpu));
6287         }
6288 out:
6289         mutex_unlock(&pcp_batch_high_lock);
6290         return ret;
6291 }
6292 
6293 #ifdef CONFIG_NUMA
6294 int hashdist = HASHDIST_DEFAULT;
6295 
6296 static int __init set_hashdist(char *str)
6297 {
6298         if (!str)
6299                 return 0;
6300         hashdist = simple_strtoul(str, &str, 0);
6301         return 1;
6302 }
6303 __setup("hashdist=", set_hashdist);
6304 #endif
6305 
6306 /*
6307  * allocate a large system hash table from bootmem
6308  * - it is assumed that the hash table must contain an exact power-of-2
6309  *   quantity of entries
6310  * - limit is the number of hash buckets, not the total allocation size
6311  */
6312 void *__init alloc_large_system_hash(const char *tablename,
6313                                      unsigned long bucketsize,
6314                                      unsigned long numentries,
6315                                      int scale,
6316                                      int flags,
6317                                      unsigned int *_hash_shift,
6318                                      unsigned int *_hash_mask,
6319                                      unsigned long low_limit,
6320                                      unsigned long high_limit)
6321 {
6322         unsigned long long max = high_limit;
6323         unsigned long log2qty, size;
6324         void *table = NULL;
6325 
6326         /* allow the kernel cmdline to have a say */
6327         if (!numentries) {
6328                 /* round applicable memory size up to nearest megabyte */
6329                 numentries = nr_kernel_pages;
6330 
6331                 /* It isn't necessary when PAGE_SIZE >= 1MB */
6332                 if (PAGE_SHIFT < 20)
6333                         numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
6334 
6335                 /* limit to 1 bucket per 2^scale bytes of low memory */
6336                 if (scale > PAGE_SHIFT)
6337                         numentries >>= (scale - PAGE_SHIFT);
6338                 else
6339                         numentries <<= (PAGE_SHIFT - scale);
6340 
6341                 /* Make sure we've got at least a 0-order allocation.. */
6342                 if (unlikely(flags & HASH_SMALL)) {
6343                         /* Makes no sense without HASH_EARLY */
6344                         WARN_ON(!(flags & HASH_EARLY));
6345                         if (!(numentries >> *_hash_shift)) {
6346                                 numentries = 1UL << *_hash_shift;
6347                                 BUG_ON(!numentries);
6348                         }
6349                 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
6350                         numentries = PAGE_SIZE / bucketsize;
6351         }
6352         numentries = roundup_pow_of_two(numentries);
6353 
6354         /* limit allocation size to 1/16 total memory by default */
6355         if (max == 0) {
6356                 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6357                 do_div(max, bucketsize);
6358         }
6359         max = min(max, 0x80000000ULL);
6360 
6361         if (numentries < low_limit)
6362                 numentries = low_limit;
6363         if (numentries > max)
6364                 numentries = max;
6365 
6366         log2qty = ilog2(numentries);
6367 
6368         do {
6369                 size = bucketsize << log2qty;
6370                 if (flags & HASH_EARLY)
6371                         table = memblock_virt_alloc_nopanic(size, 0);
6372                 else if (hashdist)
6373                         table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6374                 else {
6375                         /*
6376                          * If bucketsize is not a power-of-two, we may free
6377                          * some pages at the end of hash table which
6378                          * alloc_pages_exact() automatically does
6379                          */
6380                         if (get_order(size) < MAX_ORDER) {
6381                                 table = alloc_pages_exact(size, GFP_ATOMIC);
6382                                 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6383                         }
6384                 }
6385         } while (!table && size > PAGE_SIZE && --log2qty);
6386 
6387         if (!table)
6388                 panic("Failed to allocate %s hash table\n", tablename);
6389 
6390         printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
6391                tablename,
6392                (1UL << log2qty),
6393                ilog2(size) - PAGE_SHIFT,
6394                size);
6395 
6396         if (_hash_shift)
6397                 *_hash_shift = log2qty;
6398         if (_hash_mask)
6399                 *_hash_mask = (1 << log2qty) - 1;
6400 
6401         return table;
6402 }
6403 
6404 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6405 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6406                                                         unsigned long pfn)
6407 {
6408 #ifdef CONFIG_SPARSEMEM
6409         return __pfn_to_section(pfn)->pageblock_flags;
6410 #else
6411         return zone->pageblock_flags;
6412 #endif /* CONFIG_SPARSEMEM */
6413 }
6414 
6415 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6416 {
6417 #ifdef CONFIG_SPARSEMEM
6418         pfn &= (PAGES_PER_SECTION-1);
6419         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6420 #else
6421         pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
6422         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6423 #endif /* CONFIG_SPARSEMEM */
6424 }
6425 
6426 /**
6427  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6428  * @page: The page within the block of interest
6429  * @pfn: The target page frame number
6430  * @end_bitidx: The last bit of interest to retrieve
6431  * @mask: mask of bits that the caller is interested in
6432  *
6433  * Return: pageblock_bits flags
6434  */
6435 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
6436                                         unsigned long end_bitidx,
6437                                         unsigned long mask)
6438 {
6439         struct zone *zone;
6440         unsigned long *bitmap;
6441         unsigned long bitidx, word_bitidx;
6442         unsigned long word;
6443 
6444         zone = page_zone(page);
6445         bitmap = get_pageblock_bitmap(zone, pfn);
6446         bitidx = pfn_to_bitidx(zone, pfn);
6447         word_bitidx = bitidx / BITS_PER_LONG;
6448         bitidx &= (BITS_PER_LONG-1);
6449 
6450         word = bitmap[word_bitidx];
6451         bitidx += end_bitidx;
6452         return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
6453 }
6454 
6455 /**
6456  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6457  * @page: The page within the block of interest
6458  * @flags: The flags to set
6459  * @pfn: The target page frame number
6460  * @end_bitidx: The last bit of interest
6461  * @mask: mask of bits that the caller is interested in
6462  */
6463 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6464                                         unsigned long pfn,
6465                                         unsigned long end_bitidx,
6466                                         unsigned long mask)
6467 {
6468         struct zone *zone;
6469         unsigned long *bitmap;
6470         unsigned long bitidx, word_bitidx;
6471         unsigned long old_word, word;
6472 
6473         BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
6474 
6475         zone = page_zone(page);
6476         bitmap = get_pageblock_bitmap(zone, pfn);
6477         bitidx = pfn_to_bitidx(zone, pfn);
6478         word_bitidx = bitidx / BITS_PER_LONG;
6479         bitidx &= (BITS_PER_LONG-1);
6480 
6481         VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
6482 
6483         bitidx += end_bitidx;
6484         mask <<= (BITS_PER_LONG - bitidx - 1);
6485         flags <<= (BITS_PER_LONG - bitidx - 1);
6486 
6487         word = READ_ONCE(bitmap[word_bitidx]);
6488         for (;;) {
6489                 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6490                 if (word == old_word)
6491                         break;
6492                 word = old_word;
6493         }
6494 }
6495 
6496 /*
6497  * This function checks whether pageblock includes unmovable pages or not.
6498  * If @count is not zero, it is okay to include less @count unmovable pages
6499  *
6500  * PageLRU check without isolation or lru_lock could race so that
6501  * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6502  * expect this function should be exact.
6503  */
6504 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6505                          bool skip_hwpoisoned_pages)
6506 {
6507         unsigned long pfn, iter, found;
6508         int mt;
6509 
6510         /*
6511          * For avoiding noise data, lru_add_drain_all() should be called
6512          * If ZONE_MOVABLE, the zone never contains unmovable pages
6513          */
6514         if (zone_idx(zone) == ZONE_MOVABLE)
6515                 return false;
6516         mt = get_pageblock_migratetype(page);
6517         if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6518                 return false;
6519 
6520         pfn = page_to_pfn(page);
6521         for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6522                 unsigned long check = pfn + iter;
6523 
6524                 if (!pfn_valid_within(check))
6525                         continue;
6526 
6527                 page = pfn_to_page(check);
6528 
6529                 /*
6530                  * Hugepages are not in LRU lists, but they're movable.
6531                  * We need not scan over tail pages bacause we don't
6532                  * handle each tail page individually in migration.
6533                  */
6534                 if (PageHuge(page)) {
6535                         iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6536                         continue;
6537                 }
6538 
6539                 /*
6540                  * We can't use page_count without pin a page
6541                  * because another CPU can free compound page.
6542                  * This check already skips compound tails of THP
6543                  * because their page->_count is zero at all time.
6544                  */
6545                 if (!atomic_read(&page->_count)) {
6546                         if (PageBuddy(page))
6547                                 iter += (1 << page_order(page)) - 1;
6548                         continue;
6549                 }
6550 
6551                 /*
6552                  * The HWPoisoned page may be not in buddy system, and
6553                  * page_count() is not 0.
6554                  */
6555                 if (skip_hwpoisoned_pages && PageHWPoison(page))
6556                         continue;
6557 
6558                 if (!PageLRU(page))
6559                         found++;
6560                 /*
6561                  * If there are RECLAIMABLE pages, we need to check
6562                  * it.  But now, memory offline itself doesn't call
6563                  * shrink_node_slabs() and it still to be fixed.
6564                  */
6565                 /*
6566                  * If the page is not RAM, page_count()should be 0.
6567                  * we don't need more check. This is an _used_ not-movable page.
6568                  *
6569                  * The problematic thing here is PG_reserved pages. PG_reserved
6570                  * is set to both of a memory hole page and a _used_ kernel
6571                  * page at boot.
6572                  */
6573                 if (found > count)
6574                         return true;
6575         }
6576         return false;
6577 }
6578 
6579 bool is_pageblock_removable_nolock(struct page *page)
6580 {
6581         struct zone *zone;
6582         unsigned long pfn;
6583 
6584         /*
6585          * We have to be careful here because we are iterating over memory
6586          * sections which are not zone aware so we might end up outside of
6587          * the zone but still within the section.
6588          * We have to take care about the node as well. If the node is offline
6589          * its NODE_DATA will be NULL - see page_zone.
6590          */
6591         if (!node_online(page_to_nid(page)))
6592                 return false;
6593 
6594         zone = page_zone(page);
6595         pfn = page_to_pfn(page);
6596         if (!zone_spans_pfn(zone, pfn))
6597                 return false;
6598 
6599         return !has_unmovable_pages(zone, page, 0, true);
6600 }
6601 
6602 #ifdef CONFIG_CMA
6603 
6604 static unsigned long pfn_max_align_down(unsigned long pfn)
6605 {
6606         return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6607                              pageblock_nr_pages) - 1);
6608 }
6609 
6610 static unsigned long pfn_max_align_up(unsigned long pfn)
6611 {
6612         return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6613                                 pageblock_nr_pages));
6614 }
6615 
6616 /* [start, end) must belong to a single zone. */
6617 static int __alloc_contig_migrate_range(struct compact_control *cc,
6618                                         unsigned long start, unsigned long end)
6619 {
6620         /* This function is based on compact_zone() from compaction.c. */
6621         unsigned long nr_reclaimed;
6622         unsigned long pfn = start;
6623         unsigned int tries = 0;
6624         int ret = 0;
6625 
6626         migrate_prep();
6627 
6628         while (pfn < end || !list_empty(&cc->migratepages)) {
6629                 if (fatal_signal_pending(current)) {
6630                         ret = -EINTR;
6631                         break;
6632                 }
6633 
6634                 if (list_empty(&cc->migratepages)) {
6635                         cc->nr_migratepages = 0;
6636                         pfn = isolate_migratepages_range(cc, pfn, end);
6637                         if (!pfn) {
6638                                 ret = -EINTR;
6639                                 break;
6640                         }
6641                         tries = 0;
6642                 } else if (++tries == 5) {
6643                         ret = ret < 0 ? ret : -EBUSY;
6644                         break;
6645                 }
6646 
6647                 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6648                                                         &cc->migratepages);
6649                 cc->nr_migratepages -= nr_reclaimed;
6650 
6651                 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6652                                     NULL, 0, cc->mode, MR_CMA);
6653         }
6654         if (ret < 0) {
6655                 putback_movable_pages(&cc->migratepages);
6656                 return ret;
6657         }
6658         return 0;
6659 }
6660 
6661 /**
6662  * alloc_contig_range() -- tries to allocate given range of pages
6663  * @start:      start PFN to allocate
6664  * @end:        one-past-the-last PFN to allocate
6665  * @migratetype:        migratetype of the underlaying pageblocks (either
6666  *                      #MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
6667  *                      in range must have the same migratetype and it must
6668  *                      be either of the two.
6669  *
6670  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6671  * aligned, however it's the caller's responsibility to guarantee that
6672  * we are the only thread that changes migrate type of pageblocks the
6673  * pages fall in.
6674  *
6675  * The PFN range must belong to a single zone.
6676  *
6677  * Returns zero on success or negative error code.  On success all
6678  * pages which PFN is in [start, end) are allocated for the caller and
6679  * need to be freed with free_contig_range().
6680  */
6681 int alloc_contig_range(unsigned long start, unsigned long end,
6682                        unsigned migratetype)
6683 {
6684         unsigned long outer_start, outer_end;
6685         unsigned int order;
6686         int ret = 0;
6687 
6688         struct compact_control cc = {
6689                 .nr_migratepages = 0,
6690                 .order = -1,
6691                 .zone = page_zone(pfn_to_page(start)),
6692                 .mode = MIGRATE_SYNC,
6693                 .ignore_skip_hint = true,
6694         };
6695         INIT_LIST_HEAD(&cc.migratepages);
6696 
6697         /*
6698          * What we do here is we mark all pageblocks in range as
6699          * MIGRATE_ISOLATE.  Because pageblock and max order pages may
6700          * have different sizes, and due to the way page allocator
6701          * work, we align the range to biggest of the two pages so
6702          * that page allocator won't try to merge buddies from
6703          * different pageblocks and change MIGRATE_ISOLATE to some
6704          * other migration type.
6705          *
6706          * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6707          * migrate the pages from an unaligned range (ie. pages that
6708          * we are interested in).  This will put all the pages in
6709          * range back to page allocator as MIGRATE_ISOLATE.
6710          *
6711          * When this is done, we take the pages in range from page
6712          * allocator removing them from the buddy system.  This way
6713          * page allocator will never consider using them.
6714          *
6715          * This lets us mark the pageblocks back as
6716          * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6717          * aligned range but not in the unaligned, original range are
6718          * put back to page allocator so that buddy can use them.
6719          */
6720 
6721         ret = start_isolate_page_range(pfn_max_align_down(start),
6722                                        pfn_max_align_up(end), migratetype,
6723                                        false);
6724         if (ret)
6725                 return ret;
6726 
6727         ret = __alloc_contig_migrate_range(&cc, start, end);
6728         if (ret)
6729                 goto done;
6730 
6731         /*
6732          * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6733          * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
6734          * more, all pages in [start, end) are free in page allocator.
6735          * What we are going to do is to allocate all pages from
6736          * [start, end) (that is remove them from page allocator).
6737          *
6738          * The only problem is that pages at the beginning and at the
6739          * end of interesting range may be not aligned with pages that
6740          * page allocator holds, ie. they can be part of higher order
6741          * pages.  Because of this, we reserve the bigger range and
6742          * once this is done free the pages we are not interested in.
6743          *
6744          * We don't have to hold zone->lock here because the pages are
6745          * isolated thus they won't get removed from buddy.
6746          */
6747 
6748         lru_add_drain_all();
6749         drain_all_pages(cc.zone);
6750 
6751         order = 0;
6752         outer_start = start;
6753         while (!PageBuddy(pfn_to_page(outer_start))) {
6754                 if (++order >= MAX_ORDER) {
6755                         ret = -EBUSY;
6756                         goto done;
6757                 }
6758                 outer_start &= ~0UL << order;
6759         }
6760 
6761         /* Make sure the range is really isolated. */
6762         if (test_pages_isolated(outer_start, end, false)) {
6763                 pr_info("%s: [%lx, %lx) PFNs busy\n",
6764                         __func__, outer_start, end);
6765                 ret = -EBUSY;
6766                 goto done;
6767         }
6768 
6769         /* Grab isolated pages from freelists. */
6770         outer_end = isolate_freepages_range(&cc, outer_start, end);
6771         if (!outer_end) {
6772                 ret = -EBUSY;
6773                 goto done;
6774         }
6775 
6776         /* Free head and tail (if any) */
6777         if (start != outer_start)
6778                 free_contig_range(outer_start, start - outer_start);
6779         if (end != outer_end)
6780                 free_contig_range(end, outer_end - end);
6781 
6782 done:
6783         undo_isolate_page_range(pfn_max_align_down(start),
6784                                 pfn_max_align_up(end), migratetype);
6785         return ret;
6786 }
6787 
6788 void free_contig_range(unsigned long pfn, unsigned nr_pages)
6789 {
6790         unsigned int count = 0;
6791 
6792         for (; nr_pages--; pfn++) {
6793                 struct page *page = pfn_to_page(pfn);
6794 
6795                 count += page_count(page) != 1;
6796                 __free_page(page);
6797         }
6798         WARN(count != 0, "%d pages are still in use!\n", count);
6799 }
6800 #endif
6801 
6802 #ifdef CONFIG_MEMORY_HOTPLUG
6803 /*
6804  * The zone indicated has a new number of managed_pages; batch sizes and percpu
6805  * page high values need to be recalulated.
6806  */
6807 void __meminit zone_pcp_update(struct zone *zone)
6808 {
6809         unsigned cpu;
6810         mutex_lock(&pcp_batch_high_lock);
6811         for_each_possible_cpu(cpu)
6812                 pageset_set_high_and_batch(zone,
6813                                 per_cpu_ptr(zone->pageset, cpu));
6814         mutex_unlock(&pcp_batch_high_lock);
6815 }
6816 #endif
6817 
6818 void zone_pcp_reset(struct zone *zone)
6819 {
6820         unsigned long flags;
6821         int cpu;
6822         struct per_cpu_pageset *pset;
6823 
6824         /* avoid races with drain_pages()  */
6825         local_irq_save(flags);
6826         if (zone->pageset != &boot_pageset) {
6827                 for_each_online_cpu(cpu) {
6828                         pset = per_cpu_ptr(zone->pageset, cpu);
6829                         drain_zonestat(zone, pset);
6830                 }
6831                 free_percpu(zone->pageset);
6832                 zone->pageset = &boot_pageset;
6833         }
6834         local_irq_restore(flags);
6835 }
6836 
6837 #ifdef CONFIG_MEMORY_HOTREMOVE
6838 /*
6839  * All pages in the range must be isolated before calling this.
6840  */
6841 void
6842 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6843 {
6844         struct page *page;
6845         struct zone *zone;
6846         unsigned int order, i;
6847         unsigned long pfn;
6848         unsigned long flags;
6849         /* find the first valid pfn */
6850         for (pfn = start_pfn; pfn < end_pfn; pfn++)
6851                 if (pfn_valid(pfn))
6852                         break;
6853         if (pfn == end_pfn)
6854                 return;
6855         zone = page_zone(pfn_to_page(pfn));
6856         spin_lock_irqsave(&zone->lock, flags);
6857         pfn = start_pfn;
6858         while (pfn < end_pfn) {
6859                 if (!pfn_valid(pfn)) {
6860                         pfn++;
6861                         continue;
6862                 }
6863                 page = pfn_to_page(pfn);
6864                 /*
6865                  * The HWPoisoned page may be not in buddy system, and
6866                  * page_count() is not 0.
6867                  */
6868                 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6869                         pfn++;
6870                         SetPageReserved(page);
6871                         continue;
6872                 }
6873 
6874                 BUG_ON(page_count(page));
6875                 BUG_ON(!PageBuddy(page));
6876                 order = page_order(page);
6877 #ifdef CONFIG_DEBUG_VM
6878                 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6879                        pfn, 1 << order, end_pfn);
6880 #endif
6881                 list_del(&page->lru);
6882                 rmv_page_order(page);
6883                 zone->free_area[order].nr_free--;
6884                 for (i = 0; i < (1 << order); i++)
6885                         SetPageReserved((page+i));
6886                 pfn += (1 << order);
6887         }
6888         spin_unlock_irqrestore(&zone->lock, flags);
6889 }
6890 #endif
6891 
6892 #ifdef CONFIG_MEMORY_FAILURE
6893 bool is_free_buddy_page(struct page *page)
6894 {
6895         struct zone *zone = page_zone(page);
6896         unsigned long pfn = page_to_pfn(page);
6897         unsigned long flags;
6898         unsigned int order;
6899 
6900         spin_lock_irqsave(&zone->lock, flags);
6901         for (order = 0; order < MAX_ORDER; order++) {
6902                 struct page *page_head = page - (pfn & ((1 << order) - 1));
6903 
6904                 if (PageBuddy(page_head) && page_order(page_head) >= order)
6905                         break;
6906         }
6907         spin_unlock_irqrestore(&zone->lock, flags);
6908 
6909         return order < MAX_ORDER;
6910 }
6911 #endif
6912 

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