Version:  2.0.40 2.2.26 2.4.37 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8

Linux/mm/page_alloc.c

  1 /*
  2  *  linux/mm/page_alloc.c
  3  *
  4  *  Manages the free list, the system allocates free pages here.
  5  *  Note that kmalloc() lives in slab.c
  6  *
  7  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  8  *  Swap reorganised 29.12.95, Stephen Tweedie
  9  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
 10  *  Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
 11  *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
 12  *  Zone balancing, Kanoj Sarcar, SGI, Jan 2000
 13  *  Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
 14  *          (lots of bits borrowed from Ingo Molnar & Andrew Morton)
 15  */
 16 
 17 #include <linux/stddef.h>
 18 #include <linux/mm.h>
 19 #include <linux/swap.h>
 20 #include <linux/interrupt.h>
 21 #include <linux/pagemap.h>
 22 #include <linux/jiffies.h>
 23 #include <linux/bootmem.h>
 24 #include <linux/memblock.h>
 25 #include <linux/compiler.h>
 26 #include <linux/kernel.h>
 27 #include <linux/kmemcheck.h>
 28 #include <linux/kasan.h>
 29 #include <linux/module.h>
 30 #include <linux/suspend.h>
 31 #include <linux/pagevec.h>
 32 #include <linux/blkdev.h>
 33 #include <linux/slab.h>
 34 #include <linux/ratelimit.h>
 35 #include <linux/oom.h>
 36 #include <linux/notifier.h>
 37 #include <linux/topology.h>
 38 #include <linux/sysctl.h>
 39 #include <linux/cpu.h>
 40 #include <linux/cpuset.h>
 41 #include <linux/memory_hotplug.h>
 42 #include <linux/nodemask.h>
 43 #include <linux/vmalloc.h>
 44 #include <linux/vmstat.h>
 45 #include <linux/mempolicy.h>
 46 #include <linux/memremap.h>
 47 #include <linux/stop_machine.h>
 48 #include <linux/sort.h>
 49 #include <linux/pfn.h>
 50 #include <linux/backing-dev.h>
 51 #include <linux/fault-inject.h>
 52 #include <linux/page-isolation.h>
 53 #include <linux/page_ext.h>
 54 #include <linux/debugobjects.h>
 55 #include <linux/kmemleak.h>
 56 #include <linux/compaction.h>
 57 #include <trace/events/kmem.h>
 58 #include <linux/prefetch.h>
 59 #include <linux/mm_inline.h>
 60 #include <linux/migrate.h>
 61 #include <linux/page_ext.h>
 62 #include <linux/hugetlb.h>
 63 #include <linux/sched/rt.h>
 64 #include <linux/page_owner.h>
 65 #include <linux/kthread.h>
 66 #include <linux/memcontrol.h>
 67 
 68 #include <asm/sections.h>
 69 #include <asm/tlbflush.h>
 70 #include <asm/div64.h>
 71 #include "internal.h"
 72 
 73 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
 74 static DEFINE_MUTEX(pcp_batch_high_lock);
 75 #define MIN_PERCPU_PAGELIST_FRACTION    (8)
 76 
 77 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
 78 DEFINE_PER_CPU(int, numa_node);
 79 EXPORT_PER_CPU_SYMBOL(numa_node);
 80 #endif
 81 
 82 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
 83 /*
 84  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
 85  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
 86  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
 87  * defined in <linux/topology.h>.
 88  */
 89 DEFINE_PER_CPU(int, _numa_mem_);                /* Kernel "local memory" node */
 90 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
 91 int _node_numa_mem_[MAX_NUMNODES];
 92 #endif
 93 
 94 /*
 95  * Array of node states.
 96  */
 97 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
 98         [N_POSSIBLE] = NODE_MASK_ALL,
 99         [N_ONLINE] = { { [0] = 1UL } },
100 #ifndef CONFIG_NUMA
101         [N_NORMAL_MEMORY] = { { [0] = 1UL } },
102 #ifdef CONFIG_HIGHMEM
103         [N_HIGH_MEMORY] = { { [0] = 1UL } },
104 #endif
105 #ifdef CONFIG_MOVABLE_NODE
106         [N_MEMORY] = { { [0] = 1UL } },
107 #endif
108         [N_CPU] = { { [0] = 1UL } },
109 #endif  /* NUMA */
110 };
111 EXPORT_SYMBOL(node_states);
112 
113 /* Protect totalram_pages and zone->managed_pages */
114 static DEFINE_SPINLOCK(managed_page_count_lock);
115 
116 unsigned long totalram_pages __read_mostly;
117 unsigned long totalreserve_pages __read_mostly;
118 unsigned long totalcma_pages __read_mostly;
119 
120 int percpu_pagelist_fraction;
121 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
122 
123 /*
124  * A cached value of the page's pageblock's migratetype, used when the page is
125  * put on a pcplist. Used to avoid the pageblock migratetype lookup when
126  * freeing from pcplists in most cases, at the cost of possibly becoming stale.
127  * Also the migratetype set in the page does not necessarily match the pcplist
128  * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
129  * other index - this ensures that it will be put on the correct CMA freelist.
130  */
131 static inline int get_pcppage_migratetype(struct page *page)
132 {
133         return page->index;
134 }
135 
136 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
137 {
138         page->index = migratetype;
139 }
140 
141 #ifdef CONFIG_PM_SLEEP
142 /*
143  * The following functions are used by the suspend/hibernate code to temporarily
144  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
145  * while devices are suspended.  To avoid races with the suspend/hibernate code,
146  * they should always be called with pm_mutex held (gfp_allowed_mask also should
147  * only be modified with pm_mutex held, unless the suspend/hibernate code is
148  * guaranteed not to run in parallel with that modification).
149  */
150 
151 static gfp_t saved_gfp_mask;
152 
153 void pm_restore_gfp_mask(void)
154 {
155         WARN_ON(!mutex_is_locked(&pm_mutex));
156         if (saved_gfp_mask) {
157                 gfp_allowed_mask = saved_gfp_mask;
158                 saved_gfp_mask = 0;
159         }
160 }
161 
162 void pm_restrict_gfp_mask(void)
163 {
164         WARN_ON(!mutex_is_locked(&pm_mutex));
165         WARN_ON(saved_gfp_mask);
166         saved_gfp_mask = gfp_allowed_mask;
167         gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
168 }
169 
170 bool pm_suspended_storage(void)
171 {
172         if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
173                 return false;
174         return true;
175 }
176 #endif /* CONFIG_PM_SLEEP */
177 
178 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
179 unsigned int pageblock_order __read_mostly;
180 #endif
181 
182 static void __free_pages_ok(struct page *page, unsigned int order);
183 
184 /*
185  * results with 256, 32 in the lowmem_reserve sysctl:
186  *      1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
187  *      1G machine -> (16M dma, 784M normal, 224M high)
188  *      NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
189  *      HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
190  *      HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
191  *
192  * TBD: should special case ZONE_DMA32 machines here - in those we normally
193  * don't need any ZONE_NORMAL reservation
194  */
195 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
196 #ifdef CONFIG_ZONE_DMA
197          256,
198 #endif
199 #ifdef CONFIG_ZONE_DMA32
200          256,
201 #endif
202 #ifdef CONFIG_HIGHMEM
203          32,
204 #endif
205          32,
206 };
207 
208 EXPORT_SYMBOL(totalram_pages);
209 
210 static char * const zone_names[MAX_NR_ZONES] = {
211 #ifdef CONFIG_ZONE_DMA
212          "DMA",
213 #endif
214 #ifdef CONFIG_ZONE_DMA32
215          "DMA32",
216 #endif
217          "Normal",
218 #ifdef CONFIG_HIGHMEM
219          "HighMem",
220 #endif
221          "Movable",
222 #ifdef CONFIG_ZONE_DEVICE
223          "Device",
224 #endif
225 };
226 
227 char * const migratetype_names[MIGRATE_TYPES] = {
228         "Unmovable",
229         "Movable",
230         "Reclaimable",
231         "HighAtomic",
232 #ifdef CONFIG_CMA
233         "CMA",
234 #endif
235 #ifdef CONFIG_MEMORY_ISOLATION
236         "Isolate",
237 #endif
238 };
239 
240 compound_page_dtor * const compound_page_dtors[] = {
241         NULL,
242         free_compound_page,
243 #ifdef CONFIG_HUGETLB_PAGE
244         free_huge_page,
245 #endif
246 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
247         free_transhuge_page,
248 #endif
249 };
250 
251 int min_free_kbytes = 1024;
252 int user_min_free_kbytes = -1;
253 int watermark_scale_factor = 10;
254 
255 static unsigned long __meminitdata nr_kernel_pages;
256 static unsigned long __meminitdata nr_all_pages;
257 static unsigned long __meminitdata dma_reserve;
258 
259 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
260 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
261 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
262 static unsigned long __initdata required_kernelcore;
263 static unsigned long __initdata required_movablecore;
264 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
265 static bool mirrored_kernelcore;
266 
267 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
268 int movable_zone;
269 EXPORT_SYMBOL(movable_zone);
270 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
271 
272 #if MAX_NUMNODES > 1
273 int nr_node_ids __read_mostly = MAX_NUMNODES;
274 int nr_online_nodes __read_mostly = 1;
275 EXPORT_SYMBOL(nr_node_ids);
276 EXPORT_SYMBOL(nr_online_nodes);
277 #endif
278 
279 int page_group_by_mobility_disabled __read_mostly;
280 
281 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
282 static inline void reset_deferred_meminit(pg_data_t *pgdat)
283 {
284         pgdat->first_deferred_pfn = ULONG_MAX;
285 }
286 
287 /* Returns true if the struct page for the pfn is uninitialised */
288 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
289 {
290         int nid = early_pfn_to_nid(pfn);
291 
292         if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
293                 return true;
294 
295         return false;
296 }
297 
298 /*
299  * Returns false when the remaining initialisation should be deferred until
300  * later in the boot cycle when it can be parallelised.
301  */
302 static inline bool update_defer_init(pg_data_t *pgdat,
303                                 unsigned long pfn, unsigned long zone_end,
304                                 unsigned long *nr_initialised)
305 {
306         unsigned long max_initialise;
307 
308         /* Always populate low zones for address-contrained allocations */
309         if (zone_end < pgdat_end_pfn(pgdat))
310                 return true;
311         /*
312          * Initialise at least 2G of a node but also take into account that
313          * two large system hashes that can take up 1GB for 0.25TB/node.
314          */
315         max_initialise = max(2UL << (30 - PAGE_SHIFT),
316                 (pgdat->node_spanned_pages >> 8));
317 
318         (*nr_initialised)++;
319         if ((*nr_initialised > max_initialise) &&
320             (pfn & (PAGES_PER_SECTION - 1)) == 0) {
321                 pgdat->first_deferred_pfn = pfn;
322                 return false;
323         }
324 
325         return true;
326 }
327 #else
328 static inline void reset_deferred_meminit(pg_data_t *pgdat)
329 {
330 }
331 
332 static inline bool early_page_uninitialised(unsigned long pfn)
333 {
334         return false;
335 }
336 
337 static inline bool update_defer_init(pg_data_t *pgdat,
338                                 unsigned long pfn, unsigned long zone_end,
339                                 unsigned long *nr_initialised)
340 {
341         return true;
342 }
343 #endif
344 
345 /* Return a pointer to the bitmap storing bits affecting a block of pages */
346 static inline unsigned long *get_pageblock_bitmap(struct page *page,
347                                                         unsigned long pfn)
348 {
349 #ifdef CONFIG_SPARSEMEM
350         return __pfn_to_section(pfn)->pageblock_flags;
351 #else
352         return page_zone(page)->pageblock_flags;
353 #endif /* CONFIG_SPARSEMEM */
354 }
355 
356 static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
357 {
358 #ifdef CONFIG_SPARSEMEM
359         pfn &= (PAGES_PER_SECTION-1);
360         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
361 #else
362         pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
363         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
364 #endif /* CONFIG_SPARSEMEM */
365 }
366 
367 /**
368  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
369  * @page: The page within the block of interest
370  * @pfn: The target page frame number
371  * @end_bitidx: The last bit of interest to retrieve
372  * @mask: mask of bits that the caller is interested in
373  *
374  * Return: pageblock_bits flags
375  */
376 static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
377                                         unsigned long pfn,
378                                         unsigned long end_bitidx,
379                                         unsigned long mask)
380 {
381         unsigned long *bitmap;
382         unsigned long bitidx, word_bitidx;
383         unsigned long word;
384 
385         bitmap = get_pageblock_bitmap(page, pfn);
386         bitidx = pfn_to_bitidx(page, pfn);
387         word_bitidx = bitidx / BITS_PER_LONG;
388         bitidx &= (BITS_PER_LONG-1);
389 
390         word = bitmap[word_bitidx];
391         bitidx += end_bitidx;
392         return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
393 }
394 
395 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
396                                         unsigned long end_bitidx,
397                                         unsigned long mask)
398 {
399         return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
400 }
401 
402 static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
403 {
404         return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
405 }
406 
407 /**
408  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
409  * @page: The page within the block of interest
410  * @flags: The flags to set
411  * @pfn: The target page frame number
412  * @end_bitidx: The last bit of interest
413  * @mask: mask of bits that the caller is interested in
414  */
415 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
416                                         unsigned long pfn,
417                                         unsigned long end_bitidx,
418                                         unsigned long mask)
419 {
420         unsigned long *bitmap;
421         unsigned long bitidx, word_bitidx;
422         unsigned long old_word, word;
423 
424         BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
425 
426         bitmap = get_pageblock_bitmap(page, pfn);
427         bitidx = pfn_to_bitidx(page, pfn);
428         word_bitidx = bitidx / BITS_PER_LONG;
429         bitidx &= (BITS_PER_LONG-1);
430 
431         VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
432 
433         bitidx += end_bitidx;
434         mask <<= (BITS_PER_LONG - bitidx - 1);
435         flags <<= (BITS_PER_LONG - bitidx - 1);
436 
437         word = READ_ONCE(bitmap[word_bitidx]);
438         for (;;) {
439                 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
440                 if (word == old_word)
441                         break;
442                 word = old_word;
443         }
444 }
445 
446 void set_pageblock_migratetype(struct page *page, int migratetype)
447 {
448         if (unlikely(page_group_by_mobility_disabled &&
449                      migratetype < MIGRATE_PCPTYPES))
450                 migratetype = MIGRATE_UNMOVABLE;
451 
452         set_pageblock_flags_group(page, (unsigned long)migratetype,
453                                         PB_migrate, PB_migrate_end);
454 }
455 
456 #ifdef CONFIG_DEBUG_VM
457 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
458 {
459         int ret = 0;
460         unsigned seq;
461         unsigned long pfn = page_to_pfn(page);
462         unsigned long sp, start_pfn;
463 
464         do {
465                 seq = zone_span_seqbegin(zone);
466                 start_pfn = zone->zone_start_pfn;
467                 sp = zone->spanned_pages;
468                 if (!zone_spans_pfn(zone, pfn))
469                         ret = 1;
470         } while (zone_span_seqretry(zone, seq));
471 
472         if (ret)
473                 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
474                         pfn, zone_to_nid(zone), zone->name,
475                         start_pfn, start_pfn + sp);
476 
477         return ret;
478 }
479 
480 static int page_is_consistent(struct zone *zone, struct page *page)
481 {
482         if (!pfn_valid_within(page_to_pfn(page)))
483                 return 0;
484         if (zone != page_zone(page))
485                 return 0;
486 
487         return 1;
488 }
489 /*
490  * Temporary debugging check for pages not lying within a given zone.
491  */
492 static int bad_range(struct zone *zone, struct page *page)
493 {
494         if (page_outside_zone_boundaries(zone, page))
495                 return 1;
496         if (!page_is_consistent(zone, page))
497                 return 1;
498 
499         return 0;
500 }
501 #else
502 static inline int bad_range(struct zone *zone, struct page *page)
503 {
504         return 0;
505 }
506 #endif
507 
508 static void bad_page(struct page *page, const char *reason,
509                 unsigned long bad_flags)
510 {
511         static unsigned long resume;
512         static unsigned long nr_shown;
513         static unsigned long nr_unshown;
514 
515         /*
516          * Allow a burst of 60 reports, then keep quiet for that minute;
517          * or allow a steady drip of one report per second.
518          */
519         if (nr_shown == 60) {
520                 if (time_before(jiffies, resume)) {
521                         nr_unshown++;
522                         goto out;
523                 }
524                 if (nr_unshown) {
525                         pr_alert(
526                               "BUG: Bad page state: %lu messages suppressed\n",
527                                 nr_unshown);
528                         nr_unshown = 0;
529                 }
530                 nr_shown = 0;
531         }
532         if (nr_shown++ == 0)
533                 resume = jiffies + 60 * HZ;
534 
535         pr_alert("BUG: Bad page state in process %s  pfn:%05lx\n",
536                 current->comm, page_to_pfn(page));
537         __dump_page(page, reason);
538         bad_flags &= page->flags;
539         if (bad_flags)
540                 pr_alert("bad because of flags: %#lx(%pGp)\n",
541                                                 bad_flags, &bad_flags);
542         dump_page_owner(page);
543 
544         print_modules();
545         dump_stack();
546 out:
547         /* Leave bad fields for debug, except PageBuddy could make trouble */
548         page_mapcount_reset(page); /* remove PageBuddy */
549         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
550 }
551 
552 /*
553  * Higher-order pages are called "compound pages".  They are structured thusly:
554  *
555  * The first PAGE_SIZE page is called the "head page" and have PG_head set.
556  *
557  * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
558  * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
559  *
560  * The first tail page's ->compound_dtor holds the offset in array of compound
561  * page destructors. See compound_page_dtors.
562  *
563  * The first tail page's ->compound_order holds the order of allocation.
564  * This usage means that zero-order pages may not be compound.
565  */
566 
567 void free_compound_page(struct page *page)
568 {
569         __free_pages_ok(page, compound_order(page));
570 }
571 
572 void prep_compound_page(struct page *page, unsigned int order)
573 {
574         int i;
575         int nr_pages = 1 << order;
576 
577         set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
578         set_compound_order(page, order);
579         __SetPageHead(page);
580         for (i = 1; i < nr_pages; i++) {
581                 struct page *p = page + i;
582                 set_page_count(p, 0);
583                 p->mapping = TAIL_MAPPING;
584                 set_compound_head(p, page);
585         }
586         atomic_set(compound_mapcount_ptr(page), -1);
587 }
588 
589 #ifdef CONFIG_DEBUG_PAGEALLOC
590 unsigned int _debug_guardpage_minorder;
591 bool _debug_pagealloc_enabled __read_mostly
592                         = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
593 EXPORT_SYMBOL(_debug_pagealloc_enabled);
594 bool _debug_guardpage_enabled __read_mostly;
595 
596 static int __init early_debug_pagealloc(char *buf)
597 {
598         if (!buf)
599                 return -EINVAL;
600         return kstrtobool(buf, &_debug_pagealloc_enabled);
601 }
602 early_param("debug_pagealloc", early_debug_pagealloc);
603 
604 static bool need_debug_guardpage(void)
605 {
606         /* If we don't use debug_pagealloc, we don't need guard page */
607         if (!debug_pagealloc_enabled())
608                 return false;
609 
610         return true;
611 }
612 
613 static void init_debug_guardpage(void)
614 {
615         if (!debug_pagealloc_enabled())
616                 return;
617 
618         _debug_guardpage_enabled = true;
619 }
620 
621 struct page_ext_operations debug_guardpage_ops = {
622         .need = need_debug_guardpage,
623         .init = init_debug_guardpage,
624 };
625 
626 static int __init debug_guardpage_minorder_setup(char *buf)
627 {
628         unsigned long res;
629 
630         if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
631                 pr_err("Bad debug_guardpage_minorder value\n");
632                 return 0;
633         }
634         _debug_guardpage_minorder = res;
635         pr_info("Setting debug_guardpage_minorder to %lu\n", res);
636         return 0;
637 }
638 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
639 
640 static inline void set_page_guard(struct zone *zone, struct page *page,
641                                 unsigned int order, int migratetype)
642 {
643         struct page_ext *page_ext;
644 
645         if (!debug_guardpage_enabled())
646                 return;
647 
648         page_ext = lookup_page_ext(page);
649         if (unlikely(!page_ext))
650                 return;
651 
652         __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
653 
654         INIT_LIST_HEAD(&page->lru);
655         set_page_private(page, order);
656         /* Guard pages are not available for any usage */
657         __mod_zone_freepage_state(zone, -(1 << order), migratetype);
658 }
659 
660 static inline void clear_page_guard(struct zone *zone, struct page *page,
661                                 unsigned int order, int migratetype)
662 {
663         struct page_ext *page_ext;
664 
665         if (!debug_guardpage_enabled())
666                 return;
667 
668         page_ext = lookup_page_ext(page);
669         if (unlikely(!page_ext))
670                 return;
671 
672         __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
673 
674         set_page_private(page, 0);
675         if (!is_migrate_isolate(migratetype))
676                 __mod_zone_freepage_state(zone, (1 << order), migratetype);
677 }
678 #else
679 struct page_ext_operations debug_guardpage_ops = { NULL, };
680 static inline void set_page_guard(struct zone *zone, struct page *page,
681                                 unsigned int order, int migratetype) {}
682 static inline void clear_page_guard(struct zone *zone, struct page *page,
683                                 unsigned int order, int migratetype) {}
684 #endif
685 
686 static inline void set_page_order(struct page *page, unsigned int order)
687 {
688         set_page_private(page, order);
689         __SetPageBuddy(page);
690 }
691 
692 static inline void rmv_page_order(struct page *page)
693 {
694         __ClearPageBuddy(page);
695         set_page_private(page, 0);
696 }
697 
698 /*
699  * This function checks whether a page is free && is the buddy
700  * we can do coalesce a page and its buddy if
701  * (a) the buddy is not in a hole &&
702  * (b) the buddy is in the buddy system &&
703  * (c) a page and its buddy have the same order &&
704  * (d) a page and its buddy are in the same zone.
705  *
706  * For recording whether a page is in the buddy system, we set ->_mapcount
707  * PAGE_BUDDY_MAPCOUNT_VALUE.
708  * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
709  * serialized by zone->lock.
710  *
711  * For recording page's order, we use page_private(page).
712  */
713 static inline int page_is_buddy(struct page *page, struct page *buddy,
714                                                         unsigned int order)
715 {
716         if (!pfn_valid_within(page_to_pfn(buddy)))
717                 return 0;
718 
719         if (page_is_guard(buddy) && page_order(buddy) == order) {
720                 if (page_zone_id(page) != page_zone_id(buddy))
721                         return 0;
722 
723                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
724 
725                 return 1;
726         }
727 
728         if (PageBuddy(buddy) && page_order(buddy) == order) {
729                 /*
730                  * zone check is done late to avoid uselessly
731                  * calculating zone/node ids for pages that could
732                  * never merge.
733                  */
734                 if (page_zone_id(page) != page_zone_id(buddy))
735                         return 0;
736 
737                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
738 
739                 return 1;
740         }
741         return 0;
742 }
743 
744 /*
745  * Freeing function for a buddy system allocator.
746  *
747  * The concept of a buddy system is to maintain direct-mapped table
748  * (containing bit values) for memory blocks of various "orders".
749  * The bottom level table contains the map for the smallest allocatable
750  * units of memory (here, pages), and each level above it describes
751  * pairs of units from the levels below, hence, "buddies".
752  * At a high level, all that happens here is marking the table entry
753  * at the bottom level available, and propagating the changes upward
754  * as necessary, plus some accounting needed to play nicely with other
755  * parts of the VM system.
756  * At each level, we keep a list of pages, which are heads of continuous
757  * free pages of length of (1 << order) and marked with _mapcount
758  * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
759  * field.
760  * So when we are allocating or freeing one, we can derive the state of the
761  * other.  That is, if we allocate a small block, and both were
762  * free, the remainder of the region must be split into blocks.
763  * If a block is freed, and its buddy is also free, then this
764  * triggers coalescing into a block of larger size.
765  *
766  * -- nyc
767  */
768 
769 static inline void __free_one_page(struct page *page,
770                 unsigned long pfn,
771                 struct zone *zone, unsigned int order,
772                 int migratetype)
773 {
774         unsigned long page_idx;
775         unsigned long combined_idx;
776         unsigned long uninitialized_var(buddy_idx);
777         struct page *buddy;
778         unsigned int max_order;
779 
780         max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
781 
782         VM_BUG_ON(!zone_is_initialized(zone));
783         VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
784 
785         VM_BUG_ON(migratetype == -1);
786         if (likely(!is_migrate_isolate(migratetype)))
787                 __mod_zone_freepage_state(zone, 1 << order, migratetype);
788 
789         page_idx = pfn & ((1 << MAX_ORDER) - 1);
790 
791         VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
792         VM_BUG_ON_PAGE(bad_range(zone, page), page);
793 
794 continue_merging:
795         while (order < max_order - 1) {
796                 buddy_idx = __find_buddy_index(page_idx, order);
797                 buddy = page + (buddy_idx - page_idx);
798                 if (!page_is_buddy(page, buddy, order))
799                         goto done_merging;
800                 /*
801                  * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
802                  * merge with it and move up one order.
803                  */
804                 if (page_is_guard(buddy)) {
805                         clear_page_guard(zone, buddy, order, migratetype);
806                 } else {
807                         list_del(&buddy->lru);
808                         zone->free_area[order].nr_free--;
809                         rmv_page_order(buddy);
810                 }
811                 combined_idx = buddy_idx & page_idx;
812                 page = page + (combined_idx - page_idx);
813                 page_idx = combined_idx;
814                 order++;
815         }
816         if (max_order < MAX_ORDER) {
817                 /* If we are here, it means order is >= pageblock_order.
818                  * We want to prevent merge between freepages on isolate
819                  * pageblock and normal pageblock. Without this, pageblock
820                  * isolation could cause incorrect freepage or CMA accounting.
821                  *
822                  * We don't want to hit this code for the more frequent
823                  * low-order merging.
824                  */
825                 if (unlikely(has_isolate_pageblock(zone))) {
826                         int buddy_mt;
827 
828                         buddy_idx = __find_buddy_index(page_idx, order);
829                         buddy = page + (buddy_idx - page_idx);
830                         buddy_mt = get_pageblock_migratetype(buddy);
831 
832                         if (migratetype != buddy_mt
833                                         && (is_migrate_isolate(migratetype) ||
834                                                 is_migrate_isolate(buddy_mt)))
835                                 goto done_merging;
836                 }
837                 max_order++;
838                 goto continue_merging;
839         }
840 
841 done_merging:
842         set_page_order(page, order);
843 
844         /*
845          * If this is not the largest possible page, check if the buddy
846          * of the next-highest order is free. If it is, it's possible
847          * that pages are being freed that will coalesce soon. In case,
848          * that is happening, add the free page to the tail of the list
849          * so it's less likely to be used soon and more likely to be merged
850          * as a higher order page
851          */
852         if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
853                 struct page *higher_page, *higher_buddy;
854                 combined_idx = buddy_idx & page_idx;
855                 higher_page = page + (combined_idx - page_idx);
856                 buddy_idx = __find_buddy_index(combined_idx, order + 1);
857                 higher_buddy = higher_page + (buddy_idx - combined_idx);
858                 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
859                         list_add_tail(&page->lru,
860                                 &zone->free_area[order].free_list[migratetype]);
861                         goto out;
862                 }
863         }
864 
865         list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
866 out:
867         zone->free_area[order].nr_free++;
868 }
869 
870 /*
871  * A bad page could be due to a number of fields. Instead of multiple branches,
872  * try and check multiple fields with one check. The caller must do a detailed
873  * check if necessary.
874  */
875 static inline bool page_expected_state(struct page *page,
876                                         unsigned long check_flags)
877 {
878         if (unlikely(atomic_read(&page->_mapcount) != -1))
879                 return false;
880 
881         if (unlikely((unsigned long)page->mapping |
882                         page_ref_count(page) |
883 #ifdef CONFIG_MEMCG
884                         (unsigned long)page->mem_cgroup |
885 #endif
886                         (page->flags & check_flags)))
887                 return false;
888 
889         return true;
890 }
891 
892 static void free_pages_check_bad(struct page *page)
893 {
894         const char *bad_reason;
895         unsigned long bad_flags;
896 
897         bad_reason = NULL;
898         bad_flags = 0;
899 
900         if (unlikely(atomic_read(&page->_mapcount) != -1))
901                 bad_reason = "nonzero mapcount";
902         if (unlikely(page->mapping != NULL))
903                 bad_reason = "non-NULL mapping";
904         if (unlikely(page_ref_count(page) != 0))
905                 bad_reason = "nonzero _refcount";
906         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
907                 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
908                 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
909         }
910 #ifdef CONFIG_MEMCG
911         if (unlikely(page->mem_cgroup))
912                 bad_reason = "page still charged to cgroup";
913 #endif
914         bad_page(page, bad_reason, bad_flags);
915 }
916 
917 static inline int free_pages_check(struct page *page)
918 {
919         if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
920                 return 0;
921 
922         /* Something has gone sideways, find it */
923         free_pages_check_bad(page);
924         return 1;
925 }
926 
927 static int free_tail_pages_check(struct page *head_page, struct page *page)
928 {
929         int ret = 1;
930 
931         /*
932          * We rely page->lru.next never has bit 0 set, unless the page
933          * is PageTail(). Let's make sure that's true even for poisoned ->lru.
934          */
935         BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
936 
937         if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
938                 ret = 0;
939                 goto out;
940         }
941         switch (page - head_page) {
942         case 1:
943                 /* the first tail page: ->mapping is compound_mapcount() */
944                 if (unlikely(compound_mapcount(page))) {
945                         bad_page(page, "nonzero compound_mapcount", 0);
946                         goto out;
947                 }
948                 break;
949         case 2:
950                 /*
951                  * the second tail page: ->mapping is
952                  * page_deferred_list().next -- ignore value.
953                  */
954                 break;
955         default:
956                 if (page->mapping != TAIL_MAPPING) {
957                         bad_page(page, "corrupted mapping in tail page", 0);
958                         goto out;
959                 }
960                 break;
961         }
962         if (unlikely(!PageTail(page))) {
963                 bad_page(page, "PageTail not set", 0);
964                 goto out;
965         }
966         if (unlikely(compound_head(page) != head_page)) {
967                 bad_page(page, "compound_head not consistent", 0);
968                 goto out;
969         }
970         ret = 0;
971 out:
972         page->mapping = NULL;
973         clear_compound_head(page);
974         return ret;
975 }
976 
977 static __always_inline bool free_pages_prepare(struct page *page,
978                                         unsigned int order, bool check_free)
979 {
980         int bad = 0;
981 
982         VM_BUG_ON_PAGE(PageTail(page), page);
983 
984         trace_mm_page_free(page, order);
985         kmemcheck_free_shadow(page, order);
986 
987         /*
988          * Check tail pages before head page information is cleared to
989          * avoid checking PageCompound for order-0 pages.
990          */
991         if (unlikely(order)) {
992                 bool compound = PageCompound(page);
993                 int i;
994 
995                 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
996 
997                 if (compound)
998                         ClearPageDoubleMap(page);
999                 for (i = 1; i < (1 << order); i++) {
1000                         if (compound)
1001                                 bad += free_tail_pages_check(page, page + i);
1002                         if (unlikely(free_pages_check(page + i))) {
1003                                 bad++;
1004                                 continue;
1005                         }
1006                         (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1007                 }
1008         }
1009         if (PageMappingFlags(page))
1010                 page->mapping = NULL;
1011         if (memcg_kmem_enabled() && PageKmemcg(page))
1012                 memcg_kmem_uncharge(page, order);
1013         if (check_free)
1014                 bad += free_pages_check(page);
1015         if (bad)
1016                 return false;
1017 
1018         page_cpupid_reset_last(page);
1019         page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1020         reset_page_owner(page, order);
1021 
1022         if (!PageHighMem(page)) {
1023                 debug_check_no_locks_freed(page_address(page),
1024                                            PAGE_SIZE << order);
1025                 debug_check_no_obj_freed(page_address(page),
1026                                            PAGE_SIZE << order);
1027         }
1028         arch_free_page(page, order);
1029         kernel_poison_pages(page, 1 << order, 0);
1030         kernel_map_pages(page, 1 << order, 0);
1031         kasan_free_pages(page, order);
1032 
1033         return true;
1034 }
1035 
1036 #ifdef CONFIG_DEBUG_VM
1037 static inline bool free_pcp_prepare(struct page *page)
1038 {
1039         return free_pages_prepare(page, 0, true);
1040 }
1041 
1042 static inline bool bulkfree_pcp_prepare(struct page *page)
1043 {
1044         return false;
1045 }
1046 #else
1047 static bool free_pcp_prepare(struct page *page)
1048 {
1049         return free_pages_prepare(page, 0, false);
1050 }
1051 
1052 static bool bulkfree_pcp_prepare(struct page *page)
1053 {
1054         return free_pages_check(page);
1055 }
1056 #endif /* CONFIG_DEBUG_VM */
1057 
1058 /*
1059  * Frees a number of pages from the PCP lists
1060  * Assumes all pages on list are in same zone, and of same order.
1061  * count is the number of pages to free.
1062  *
1063  * If the zone was previously in an "all pages pinned" state then look to
1064  * see if this freeing clears that state.
1065  *
1066  * And clear the zone's pages_scanned counter, to hold off the "all pages are
1067  * pinned" detection logic.
1068  */
1069 static void free_pcppages_bulk(struct zone *zone, int count,
1070                                         struct per_cpu_pages *pcp)
1071 {
1072         int migratetype = 0;
1073         int batch_free = 0;
1074         unsigned long nr_scanned;
1075         bool isolated_pageblocks;
1076 
1077         spin_lock(&zone->lock);
1078         isolated_pageblocks = has_isolate_pageblock(zone);
1079         nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
1080         if (nr_scanned)
1081                 __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
1082 
1083         while (count) {
1084                 struct page *page;
1085                 struct list_head *list;
1086 
1087                 /*
1088                  * Remove pages from lists in a round-robin fashion. A
1089                  * batch_free count is maintained that is incremented when an
1090                  * empty list is encountered.  This is so more pages are freed
1091                  * off fuller lists instead of spinning excessively around empty
1092                  * lists
1093                  */
1094                 do {
1095                         batch_free++;
1096                         if (++migratetype == MIGRATE_PCPTYPES)
1097                                 migratetype = 0;
1098                         list = &pcp->lists[migratetype];
1099                 } while (list_empty(list));
1100 
1101                 /* This is the only non-empty list. Free them all. */
1102                 if (batch_free == MIGRATE_PCPTYPES)
1103                         batch_free = count;
1104 
1105                 do {
1106                         int mt; /* migratetype of the to-be-freed page */
1107 
1108                         page = list_last_entry(list, struct page, lru);
1109                         /* must delete as __free_one_page list manipulates */
1110                         list_del(&page->lru);
1111 
1112                         mt = get_pcppage_migratetype(page);
1113                         /* MIGRATE_ISOLATE page should not go to pcplists */
1114                         VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1115                         /* Pageblock could have been isolated meanwhile */
1116                         if (unlikely(isolated_pageblocks))
1117                                 mt = get_pageblock_migratetype(page);
1118 
1119                         if (bulkfree_pcp_prepare(page))
1120                                 continue;
1121 
1122                         __free_one_page(page, page_to_pfn(page), zone, 0, mt);
1123                         trace_mm_page_pcpu_drain(page, 0, mt);
1124                 } while (--count && --batch_free && !list_empty(list));
1125         }
1126         spin_unlock(&zone->lock);
1127 }
1128 
1129 static void free_one_page(struct zone *zone,
1130                                 struct page *page, unsigned long pfn,
1131                                 unsigned int order,
1132                                 int migratetype)
1133 {
1134         unsigned long nr_scanned;
1135         spin_lock(&zone->lock);
1136         nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
1137         if (nr_scanned)
1138                 __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
1139 
1140         if (unlikely(has_isolate_pageblock(zone) ||
1141                 is_migrate_isolate(migratetype))) {
1142                 migratetype = get_pfnblock_migratetype(page, pfn);
1143         }
1144         __free_one_page(page, pfn, zone, order, migratetype);
1145         spin_unlock(&zone->lock);
1146 }
1147 
1148 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1149                                 unsigned long zone, int nid)
1150 {
1151         set_page_links(page, zone, nid, pfn);
1152         init_page_count(page);
1153         page_mapcount_reset(page);
1154         page_cpupid_reset_last(page);
1155 
1156         INIT_LIST_HEAD(&page->lru);
1157 #ifdef WANT_PAGE_VIRTUAL
1158         /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1159         if (!is_highmem_idx(zone))
1160                 set_page_address(page, __va(pfn << PAGE_SHIFT));
1161 #endif
1162 }
1163 
1164 static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
1165                                         int nid)
1166 {
1167         return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
1168 }
1169 
1170 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1171 static void init_reserved_page(unsigned long pfn)
1172 {
1173         pg_data_t *pgdat;
1174         int nid, zid;
1175 
1176         if (!early_page_uninitialised(pfn))
1177                 return;
1178 
1179         nid = early_pfn_to_nid(pfn);
1180         pgdat = NODE_DATA(nid);
1181 
1182         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1183                 struct zone *zone = &pgdat->node_zones[zid];
1184 
1185                 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1186                         break;
1187         }
1188         __init_single_pfn(pfn, zid, nid);
1189 }
1190 #else
1191 static inline void init_reserved_page(unsigned long pfn)
1192 {
1193 }
1194 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1195 
1196 /*
1197  * Initialised pages do not have PageReserved set. This function is
1198  * called for each range allocated by the bootmem allocator and
1199  * marks the pages PageReserved. The remaining valid pages are later
1200  * sent to the buddy page allocator.
1201  */
1202 void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1203 {
1204         unsigned long start_pfn = PFN_DOWN(start);
1205         unsigned long end_pfn = PFN_UP(end);
1206 
1207         for (; start_pfn < end_pfn; start_pfn++) {
1208                 if (pfn_valid(start_pfn)) {
1209                         struct page *page = pfn_to_page(start_pfn);
1210 
1211                         init_reserved_page(start_pfn);
1212 
1213                         /* Avoid false-positive PageTail() */
1214                         INIT_LIST_HEAD(&page->lru);
1215 
1216                         SetPageReserved(page);
1217                 }
1218         }
1219 }
1220 
1221 static void __free_pages_ok(struct page *page, unsigned int order)
1222 {
1223         unsigned long flags;
1224         int migratetype;
1225         unsigned long pfn = page_to_pfn(page);
1226 
1227         if (!free_pages_prepare(page, order, true))
1228                 return;
1229 
1230         migratetype = get_pfnblock_migratetype(page, pfn);
1231         local_irq_save(flags);
1232         __count_vm_events(PGFREE, 1 << order);
1233         free_one_page(page_zone(page), page, pfn, order, migratetype);
1234         local_irq_restore(flags);
1235 }
1236 
1237 static void __init __free_pages_boot_core(struct page *page, unsigned int order)
1238 {
1239         unsigned int nr_pages = 1 << order;
1240         struct page *p = page;
1241         unsigned int loop;
1242 
1243         prefetchw(p);
1244         for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1245                 prefetchw(p + 1);
1246                 __ClearPageReserved(p);
1247                 set_page_count(p, 0);
1248         }
1249         __ClearPageReserved(p);
1250         set_page_count(p, 0);
1251 
1252         page_zone(page)->managed_pages += nr_pages;
1253         set_page_refcounted(page);
1254         __free_pages(page, order);
1255 }
1256 
1257 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1258         defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1259 
1260 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1261 
1262 int __meminit early_pfn_to_nid(unsigned long pfn)
1263 {
1264         static DEFINE_SPINLOCK(early_pfn_lock);
1265         int nid;
1266 
1267         spin_lock(&early_pfn_lock);
1268         nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1269         if (nid < 0)
1270                 nid = first_online_node;
1271         spin_unlock(&early_pfn_lock);
1272 
1273         return nid;
1274 }
1275 #endif
1276 
1277 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1278 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1279                                         struct mminit_pfnnid_cache *state)
1280 {
1281         int nid;
1282 
1283         nid = __early_pfn_to_nid(pfn, state);
1284         if (nid >= 0 && nid != node)
1285                 return false;
1286         return true;
1287 }
1288 
1289 /* Only safe to use early in boot when initialisation is single-threaded */
1290 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1291 {
1292         return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1293 }
1294 
1295 #else
1296 
1297 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1298 {
1299         return true;
1300 }
1301 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1302                                         struct mminit_pfnnid_cache *state)
1303 {
1304         return true;
1305 }
1306 #endif
1307 
1308 
1309 void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
1310                                                         unsigned int order)
1311 {
1312         if (early_page_uninitialised(pfn))
1313                 return;
1314         return __free_pages_boot_core(page, order);
1315 }
1316 
1317 /*
1318  * Check that the whole (or subset of) a pageblock given by the interval of
1319  * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1320  * with the migration of free compaction scanner. The scanners then need to
1321  * use only pfn_valid_within() check for arches that allow holes within
1322  * pageblocks.
1323  *
1324  * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1325  *
1326  * It's possible on some configurations to have a setup like node0 node1 node0
1327  * i.e. it's possible that all pages within a zones range of pages do not
1328  * belong to a single zone. We assume that a border between node0 and node1
1329  * can occur within a single pageblock, but not a node0 node1 node0
1330  * interleaving within a single pageblock. It is therefore sufficient to check
1331  * the first and last page of a pageblock and avoid checking each individual
1332  * page in a pageblock.
1333  */
1334 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1335                                      unsigned long end_pfn, struct zone *zone)
1336 {
1337         struct page *start_page;
1338         struct page *end_page;
1339 
1340         /* end_pfn is one past the range we are checking */
1341         end_pfn--;
1342 
1343         if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1344                 return NULL;
1345 
1346         start_page = pfn_to_page(start_pfn);
1347 
1348         if (page_zone(start_page) != zone)
1349                 return NULL;
1350 
1351         end_page = pfn_to_page(end_pfn);
1352 
1353         /* This gives a shorter code than deriving page_zone(end_page) */
1354         if (page_zone_id(start_page) != page_zone_id(end_page))
1355                 return NULL;
1356 
1357         return start_page;
1358 }
1359 
1360 void set_zone_contiguous(struct zone *zone)
1361 {
1362         unsigned long block_start_pfn = zone->zone_start_pfn;
1363         unsigned long block_end_pfn;
1364 
1365         block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1366         for (; block_start_pfn < zone_end_pfn(zone);
1367                         block_start_pfn = block_end_pfn,
1368                          block_end_pfn += pageblock_nr_pages) {
1369 
1370                 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1371 
1372                 if (!__pageblock_pfn_to_page(block_start_pfn,
1373                                              block_end_pfn, zone))
1374                         return;
1375         }
1376 
1377         /* We confirm that there is no hole */
1378         zone->contiguous = true;
1379 }
1380 
1381 void clear_zone_contiguous(struct zone *zone)
1382 {
1383         zone->contiguous = false;
1384 }
1385 
1386 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1387 static void __init deferred_free_range(struct page *page,
1388                                         unsigned long pfn, int nr_pages)
1389 {
1390         int i;
1391 
1392         if (!page)
1393                 return;
1394 
1395         /* Free a large naturally-aligned chunk if possible */
1396         if (nr_pages == MAX_ORDER_NR_PAGES &&
1397             (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
1398                 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1399                 __free_pages_boot_core(page, MAX_ORDER-1);
1400                 return;
1401         }
1402 
1403         for (i = 0; i < nr_pages; i++, page++)
1404                 __free_pages_boot_core(page, 0);
1405 }
1406 
1407 /* Completion tracking for deferred_init_memmap() threads */
1408 static atomic_t pgdat_init_n_undone __initdata;
1409 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1410 
1411 static inline void __init pgdat_init_report_one_done(void)
1412 {
1413         if (atomic_dec_and_test(&pgdat_init_n_undone))
1414                 complete(&pgdat_init_all_done_comp);
1415 }
1416 
1417 /* Initialise remaining memory on a node */
1418 static int __init deferred_init_memmap(void *data)
1419 {
1420         pg_data_t *pgdat = data;
1421         int nid = pgdat->node_id;
1422         struct mminit_pfnnid_cache nid_init_state = { };
1423         unsigned long start = jiffies;
1424         unsigned long nr_pages = 0;
1425         unsigned long walk_start, walk_end;
1426         int i, zid;
1427         struct zone *zone;
1428         unsigned long first_init_pfn = pgdat->first_deferred_pfn;
1429         const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1430 
1431         if (first_init_pfn == ULONG_MAX) {
1432                 pgdat_init_report_one_done();
1433                 return 0;
1434         }
1435 
1436         /* Bind memory initialisation thread to a local node if possible */
1437         if (!cpumask_empty(cpumask))
1438                 set_cpus_allowed_ptr(current, cpumask);
1439 
1440         /* Sanity check boundaries */
1441         BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1442         BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1443         pgdat->first_deferred_pfn = ULONG_MAX;
1444 
1445         /* Only the highest zone is deferred so find it */
1446         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1447                 zone = pgdat->node_zones + zid;
1448                 if (first_init_pfn < zone_end_pfn(zone))
1449                         break;
1450         }
1451 
1452         for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1453                 unsigned long pfn, end_pfn;
1454                 struct page *page = NULL;
1455                 struct page *free_base_page = NULL;
1456                 unsigned long free_base_pfn = 0;
1457                 int nr_to_free = 0;
1458 
1459                 end_pfn = min(walk_end, zone_end_pfn(zone));
1460                 pfn = first_init_pfn;
1461                 if (pfn < walk_start)
1462                         pfn = walk_start;
1463                 if (pfn < zone->zone_start_pfn)
1464                         pfn = zone->zone_start_pfn;
1465 
1466                 for (; pfn < end_pfn; pfn++) {
1467                         if (!pfn_valid_within(pfn))
1468                                 goto free_range;
1469 
1470                         /*
1471                          * Ensure pfn_valid is checked every
1472                          * MAX_ORDER_NR_PAGES for memory holes
1473                          */
1474                         if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1475                                 if (!pfn_valid(pfn)) {
1476                                         page = NULL;
1477                                         goto free_range;
1478                                 }
1479                         }
1480 
1481                         if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1482                                 page = NULL;
1483                                 goto free_range;
1484                         }
1485 
1486                         /* Minimise pfn page lookups and scheduler checks */
1487                         if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1488                                 page++;
1489                         } else {
1490                                 nr_pages += nr_to_free;
1491                                 deferred_free_range(free_base_page,
1492                                                 free_base_pfn, nr_to_free);
1493                                 free_base_page = NULL;
1494                                 free_base_pfn = nr_to_free = 0;
1495 
1496                                 page = pfn_to_page(pfn);
1497                                 cond_resched();
1498                         }
1499 
1500                         if (page->flags) {
1501                                 VM_BUG_ON(page_zone(page) != zone);
1502                                 goto free_range;
1503                         }
1504 
1505                         __init_single_page(page, pfn, zid, nid);
1506                         if (!free_base_page) {
1507                                 free_base_page = page;
1508                                 free_base_pfn = pfn;
1509                                 nr_to_free = 0;
1510                         }
1511                         nr_to_free++;
1512 
1513                         /* Where possible, batch up pages for a single free */
1514                         continue;
1515 free_range:
1516                         /* Free the current block of pages to allocator */
1517                         nr_pages += nr_to_free;
1518                         deferred_free_range(free_base_page, free_base_pfn,
1519                                                                 nr_to_free);
1520                         free_base_page = NULL;
1521                         free_base_pfn = nr_to_free = 0;
1522                 }
1523 
1524                 first_init_pfn = max(end_pfn, first_init_pfn);
1525         }
1526 
1527         /* Sanity check that the next zone really is unpopulated */
1528         WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1529 
1530         pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1531                                         jiffies_to_msecs(jiffies - start));
1532 
1533         pgdat_init_report_one_done();
1534         return 0;
1535 }
1536 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1537 
1538 void __init page_alloc_init_late(void)
1539 {
1540         struct zone *zone;
1541 
1542 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1543         int nid;
1544 
1545         /* There will be num_node_state(N_MEMORY) threads */
1546         atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1547         for_each_node_state(nid, N_MEMORY) {
1548                 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1549         }
1550 
1551         /* Block until all are initialised */
1552         wait_for_completion(&pgdat_init_all_done_comp);
1553 
1554         /* Reinit limits that are based on free pages after the kernel is up */
1555         files_maxfiles_init();
1556 #endif
1557 
1558         for_each_populated_zone(zone)
1559                 set_zone_contiguous(zone);
1560 }
1561 
1562 #ifdef CONFIG_CMA
1563 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1564 void __init init_cma_reserved_pageblock(struct page *page)
1565 {
1566         unsigned i = pageblock_nr_pages;
1567         struct page *p = page;
1568 
1569         do {
1570                 __ClearPageReserved(p);
1571                 set_page_count(p, 0);
1572         } while (++p, --i);
1573 
1574         set_pageblock_migratetype(page, MIGRATE_CMA);
1575 
1576         if (pageblock_order >= MAX_ORDER) {
1577                 i = pageblock_nr_pages;
1578                 p = page;
1579                 do {
1580                         set_page_refcounted(p);
1581                         __free_pages(p, MAX_ORDER - 1);
1582                         p += MAX_ORDER_NR_PAGES;
1583                 } while (i -= MAX_ORDER_NR_PAGES);
1584         } else {
1585                 set_page_refcounted(page);
1586                 __free_pages(page, pageblock_order);
1587         }
1588 
1589         adjust_managed_page_count(page, pageblock_nr_pages);
1590 }
1591 #endif
1592 
1593 /*
1594  * The order of subdivision here is critical for the IO subsystem.
1595  * Please do not alter this order without good reasons and regression
1596  * testing. Specifically, as large blocks of memory are subdivided,
1597  * the order in which smaller blocks are delivered depends on the order
1598  * they're subdivided in this function. This is the primary factor
1599  * influencing the order in which pages are delivered to the IO
1600  * subsystem according to empirical testing, and this is also justified
1601  * by considering the behavior of a buddy system containing a single
1602  * large block of memory acted on by a series of small allocations.
1603  * This behavior is a critical factor in sglist merging's success.
1604  *
1605  * -- nyc
1606  */
1607 static inline void expand(struct zone *zone, struct page *page,
1608         int low, int high, struct free_area *area,
1609         int migratetype)
1610 {
1611         unsigned long size = 1 << high;
1612 
1613         while (high > low) {
1614                 area--;
1615                 high--;
1616                 size >>= 1;
1617                 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1618 
1619                 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
1620                         debug_guardpage_enabled() &&
1621                         high < debug_guardpage_minorder()) {
1622                         /*
1623                          * Mark as guard pages (or page), that will allow to
1624                          * merge back to allocator when buddy will be freed.
1625                          * Corresponding page table entries will not be touched,
1626                          * pages will stay not present in virtual address space
1627                          */
1628                         set_page_guard(zone, &page[size], high, migratetype);
1629                         continue;
1630                 }
1631                 list_add(&page[size].lru, &area->free_list[migratetype]);
1632                 area->nr_free++;
1633                 set_page_order(&page[size], high);
1634         }
1635 }
1636 
1637 static void check_new_page_bad(struct page *page)
1638 {
1639         const char *bad_reason = NULL;
1640         unsigned long bad_flags = 0;
1641 
1642         if (unlikely(atomic_read(&page->_mapcount) != -1))
1643                 bad_reason = "nonzero mapcount";
1644         if (unlikely(page->mapping != NULL))
1645                 bad_reason = "non-NULL mapping";
1646         if (unlikely(page_ref_count(page) != 0))
1647                 bad_reason = "nonzero _count";
1648         if (unlikely(page->flags & __PG_HWPOISON)) {
1649                 bad_reason = "HWPoisoned (hardware-corrupted)";
1650                 bad_flags = __PG_HWPOISON;
1651                 /* Don't complain about hwpoisoned pages */
1652                 page_mapcount_reset(page); /* remove PageBuddy */
1653                 return;
1654         }
1655         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1656                 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1657                 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1658         }
1659 #ifdef CONFIG_MEMCG
1660         if (unlikely(page->mem_cgroup))
1661                 bad_reason = "page still charged to cgroup";
1662 #endif
1663         bad_page(page, bad_reason, bad_flags);
1664 }
1665 
1666 /*
1667  * This page is about to be returned from the page allocator
1668  */
1669 static inline int check_new_page(struct page *page)
1670 {
1671         if (likely(page_expected_state(page,
1672                                 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1673                 return 0;
1674 
1675         check_new_page_bad(page);
1676         return 1;
1677 }
1678 
1679 static inline bool free_pages_prezeroed(bool poisoned)
1680 {
1681         return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1682                 page_poisoning_enabled() && poisoned;
1683 }
1684 
1685 #ifdef CONFIG_DEBUG_VM
1686 static bool check_pcp_refill(struct page *page)
1687 {
1688         return false;
1689 }
1690 
1691 static bool check_new_pcp(struct page *page)
1692 {
1693         return check_new_page(page);
1694 }
1695 #else
1696 static bool check_pcp_refill(struct page *page)
1697 {
1698         return check_new_page(page);
1699 }
1700 static bool check_new_pcp(struct page *page)
1701 {
1702         return false;
1703 }
1704 #endif /* CONFIG_DEBUG_VM */
1705 
1706 static bool check_new_pages(struct page *page, unsigned int order)
1707 {
1708         int i;
1709         for (i = 0; i < (1 << order); i++) {
1710                 struct page *p = page + i;
1711 
1712                 if (unlikely(check_new_page(p)))
1713                         return true;
1714         }
1715 
1716         return false;
1717 }
1718 
1719 inline void post_alloc_hook(struct page *page, unsigned int order,
1720                                 gfp_t gfp_flags)
1721 {
1722         set_page_private(page, 0);
1723         set_page_refcounted(page);
1724 
1725         arch_alloc_page(page, order);
1726         kernel_map_pages(page, 1 << order, 1);
1727         kernel_poison_pages(page, 1 << order, 1);
1728         kasan_alloc_pages(page, order);
1729         set_page_owner(page, order, gfp_flags);
1730 }
1731 
1732 static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1733                                                         unsigned int alloc_flags)
1734 {
1735         int i;
1736         bool poisoned = true;
1737 
1738         for (i = 0; i < (1 << order); i++) {
1739                 struct page *p = page + i;
1740                 if (poisoned)
1741                         poisoned &= page_is_poisoned(p);
1742         }
1743 
1744         post_alloc_hook(page, order, gfp_flags);
1745 
1746         if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
1747                 for (i = 0; i < (1 << order); i++)
1748                         clear_highpage(page + i);
1749 
1750         if (order && (gfp_flags & __GFP_COMP))
1751                 prep_compound_page(page, order);
1752 
1753         /*
1754          * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1755          * allocate the page. The expectation is that the caller is taking
1756          * steps that will free more memory. The caller should avoid the page
1757          * being used for !PFMEMALLOC purposes.
1758          */
1759         if (alloc_flags & ALLOC_NO_WATERMARKS)
1760                 set_page_pfmemalloc(page);
1761         else
1762                 clear_page_pfmemalloc(page);
1763 }
1764 
1765 /*
1766  * Go through the free lists for the given migratetype and remove
1767  * the smallest available page from the freelists
1768  */
1769 static inline
1770 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1771                                                 int migratetype)
1772 {
1773         unsigned int current_order;
1774         struct free_area *area;
1775         struct page *page;
1776 
1777         /* Find a page of the appropriate size in the preferred list */
1778         for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1779                 area = &(zone->free_area[current_order]);
1780                 page = list_first_entry_or_null(&area->free_list[migratetype],
1781                                                         struct page, lru);
1782                 if (!page)
1783                         continue;
1784                 list_del(&page->lru);
1785                 rmv_page_order(page);
1786                 area->nr_free--;
1787                 expand(zone, page, order, current_order, area, migratetype);
1788                 set_pcppage_migratetype(page, migratetype);
1789                 return page;
1790         }
1791 
1792         return NULL;
1793 }
1794 
1795 
1796 /*
1797  * This array describes the order lists are fallen back to when
1798  * the free lists for the desirable migrate type are depleted
1799  */
1800 static int fallbacks[MIGRATE_TYPES][4] = {
1801         [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_TYPES },
1802         [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_TYPES },
1803         [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
1804 #ifdef CONFIG_CMA
1805         [MIGRATE_CMA]         = { MIGRATE_TYPES }, /* Never used */
1806 #endif
1807 #ifdef CONFIG_MEMORY_ISOLATION
1808         [MIGRATE_ISOLATE]     = { MIGRATE_TYPES }, /* Never used */
1809 #endif
1810 };
1811 
1812 #ifdef CONFIG_CMA
1813 static struct page *__rmqueue_cma_fallback(struct zone *zone,
1814                                         unsigned int order)
1815 {
1816         return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1817 }
1818 #else
1819 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1820                                         unsigned int order) { return NULL; }
1821 #endif
1822 
1823 /*
1824  * Move the free pages in a range to the free lists of the requested type.
1825  * Note that start_page and end_pages are not aligned on a pageblock
1826  * boundary. If alignment is required, use move_freepages_block()
1827  */
1828 int move_freepages(struct zone *zone,
1829                           struct page *start_page, struct page *end_page,
1830                           int migratetype)
1831 {
1832         struct page *page;
1833         unsigned int order;
1834         int pages_moved = 0;
1835 
1836 #ifndef CONFIG_HOLES_IN_ZONE
1837         /*
1838          * page_zone is not safe to call in this context when
1839          * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1840          * anyway as we check zone boundaries in move_freepages_block().
1841          * Remove at a later date when no bug reports exist related to
1842          * grouping pages by mobility
1843          */
1844         VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1845 #endif
1846 
1847         for (page = start_page; page <= end_page;) {
1848                 /* Make sure we are not inadvertently changing nodes */
1849                 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1850 
1851                 if (!pfn_valid_within(page_to_pfn(page))) {
1852                         page++;
1853                         continue;
1854                 }
1855 
1856                 if (!PageBuddy(page)) {
1857                         page++;
1858                         continue;
1859                 }
1860 
1861                 order = page_order(page);
1862                 list_move(&page->lru,
1863                           &zone->free_area[order].free_list[migratetype]);
1864                 page += 1 << order;
1865                 pages_moved += 1 << order;
1866         }
1867 
1868         return pages_moved;
1869 }
1870 
1871 int move_freepages_block(struct zone *zone, struct page *page,
1872                                 int migratetype)
1873 {
1874         unsigned long start_pfn, end_pfn;
1875         struct page *start_page, *end_page;
1876 
1877         start_pfn = page_to_pfn(page);
1878         start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1879         start_page = pfn_to_page(start_pfn);
1880         end_page = start_page + pageblock_nr_pages - 1;
1881         end_pfn = start_pfn + pageblock_nr_pages - 1;
1882 
1883         /* Do not cross zone boundaries */
1884         if (!zone_spans_pfn(zone, start_pfn))
1885                 start_page = page;
1886         if (!zone_spans_pfn(zone, end_pfn))
1887                 return 0;
1888 
1889         return move_freepages(zone, start_page, end_page, migratetype);
1890 }
1891 
1892 static void change_pageblock_range(struct page *pageblock_page,
1893                                         int start_order, int migratetype)
1894 {
1895         int nr_pageblocks = 1 << (start_order - pageblock_order);
1896 
1897         while (nr_pageblocks--) {
1898                 set_pageblock_migratetype(pageblock_page, migratetype);
1899                 pageblock_page += pageblock_nr_pages;
1900         }
1901 }
1902 
1903 /*
1904  * When we are falling back to another migratetype during allocation, try to
1905  * steal extra free pages from the same pageblocks to satisfy further
1906  * allocations, instead of polluting multiple pageblocks.
1907  *
1908  * If we are stealing a relatively large buddy page, it is likely there will
1909  * be more free pages in the pageblock, so try to steal them all. For
1910  * reclaimable and unmovable allocations, we steal regardless of page size,
1911  * as fragmentation caused by those allocations polluting movable pageblocks
1912  * is worse than movable allocations stealing from unmovable and reclaimable
1913  * pageblocks.
1914  */
1915 static bool can_steal_fallback(unsigned int order, int start_mt)
1916 {
1917         /*
1918          * Leaving this order check is intended, although there is
1919          * relaxed order check in next check. The reason is that
1920          * we can actually steal whole pageblock if this condition met,
1921          * but, below check doesn't guarantee it and that is just heuristic
1922          * so could be changed anytime.
1923          */
1924         if (order >= pageblock_order)
1925                 return true;
1926 
1927         if (order >= pageblock_order / 2 ||
1928                 start_mt == MIGRATE_RECLAIMABLE ||
1929                 start_mt == MIGRATE_UNMOVABLE ||
1930                 page_group_by_mobility_disabled)
1931                 return true;
1932 
1933         return false;
1934 }
1935 
1936 /*
1937  * This function implements actual steal behaviour. If order is large enough,
1938  * we can steal whole pageblock. If not, we first move freepages in this
1939  * pageblock and check whether half of pages are moved or not. If half of
1940  * pages are moved, we can change migratetype of pageblock and permanently
1941  * use it's pages as requested migratetype in the future.
1942  */
1943 static void steal_suitable_fallback(struct zone *zone, struct page *page,
1944                                                           int start_type)
1945 {
1946         unsigned int current_order = page_order(page);
1947         int pages;
1948 
1949         /* Take ownership for orders >= pageblock_order */
1950         if (current_order >= pageblock_order) {
1951                 change_pageblock_range(page, current_order, start_type);
1952                 return;
1953         }
1954 
1955         pages = move_freepages_block(zone, page, start_type);
1956 
1957         /* Claim the whole block if over half of it is free */
1958         if (pages >= (1 << (pageblock_order-1)) ||
1959                         page_group_by_mobility_disabled)
1960                 set_pageblock_migratetype(page, start_type);
1961 }
1962 
1963 /*
1964  * Check whether there is a suitable fallback freepage with requested order.
1965  * If only_stealable is true, this function returns fallback_mt only if
1966  * we can steal other freepages all together. This would help to reduce
1967  * fragmentation due to mixed migratetype pages in one pageblock.
1968  */
1969 int find_suitable_fallback(struct free_area *area, unsigned int order,
1970                         int migratetype, bool only_stealable, bool *can_steal)
1971 {
1972         int i;
1973         int fallback_mt;
1974 
1975         if (area->nr_free == 0)
1976                 return -1;
1977 
1978         *can_steal = false;
1979         for (i = 0;; i++) {
1980                 fallback_mt = fallbacks[migratetype][i];
1981                 if (fallback_mt == MIGRATE_TYPES)
1982                         break;
1983 
1984                 if (list_empty(&area->free_list[fallback_mt]))
1985                         continue;
1986 
1987                 if (can_steal_fallback(order, migratetype))
1988                         *can_steal = true;
1989 
1990                 if (!only_stealable)
1991                         return fallback_mt;
1992 
1993                 if (*can_steal)
1994                         return fallback_mt;
1995         }
1996 
1997         return -1;
1998 }
1999 
2000 /*
2001  * Reserve a pageblock for exclusive use of high-order atomic allocations if
2002  * there are no empty page blocks that contain a page with a suitable order
2003  */
2004 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2005                                 unsigned int alloc_order)
2006 {
2007         int mt;
2008         unsigned long max_managed, flags;
2009 
2010         /*
2011          * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2012          * Check is race-prone but harmless.
2013          */
2014         max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
2015         if (zone->nr_reserved_highatomic >= max_managed)
2016                 return;
2017 
2018         spin_lock_irqsave(&zone->lock, flags);
2019 
2020         /* Recheck the nr_reserved_highatomic limit under the lock */
2021         if (zone->nr_reserved_highatomic >= max_managed)
2022                 goto out_unlock;
2023 
2024         /* Yoink! */
2025         mt = get_pageblock_migratetype(page);
2026         if (mt != MIGRATE_HIGHATOMIC &&
2027                         !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
2028                 zone->nr_reserved_highatomic += pageblock_nr_pages;
2029                 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2030                 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
2031         }
2032 
2033 out_unlock:
2034         spin_unlock_irqrestore(&zone->lock, flags);
2035 }
2036 
2037 /*
2038  * Used when an allocation is about to fail under memory pressure. This
2039  * potentially hurts the reliability of high-order allocations when under
2040  * intense memory pressure but failed atomic allocations should be easier
2041  * to recover from than an OOM.
2042  */
2043 static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
2044 {
2045         struct zonelist *zonelist = ac->zonelist;
2046         unsigned long flags;
2047         struct zoneref *z;
2048         struct zone *zone;
2049         struct page *page;
2050         int order;
2051 
2052         for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2053                                                                 ac->nodemask) {
2054                 /* Preserve at least one pageblock */
2055                 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
2056                         continue;
2057 
2058                 spin_lock_irqsave(&zone->lock, flags);
2059                 for (order = 0; order < MAX_ORDER; order++) {
2060                         struct free_area *area = &(zone->free_area[order]);
2061 
2062                         page = list_first_entry_or_null(
2063                                         &area->free_list[MIGRATE_HIGHATOMIC],
2064                                         struct page, lru);
2065                         if (!page)
2066                                 continue;
2067 
2068                         /*
2069                          * It should never happen but changes to locking could
2070                          * inadvertently allow a per-cpu drain to add pages
2071                          * to MIGRATE_HIGHATOMIC while unreserving so be safe
2072                          * and watch for underflows.
2073                          */
2074                         zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
2075                                 zone->nr_reserved_highatomic);
2076 
2077                         /*
2078                          * Convert to ac->migratetype and avoid the normal
2079                          * pageblock stealing heuristics. Minimally, the caller
2080                          * is doing the work and needs the pages. More
2081                          * importantly, if the block was always converted to
2082                          * MIGRATE_UNMOVABLE or another type then the number
2083                          * of pageblocks that cannot be completely freed
2084                          * may increase.
2085                          */
2086                         set_pageblock_migratetype(page, ac->migratetype);
2087                         move_freepages_block(zone, page, ac->migratetype);
2088                         spin_unlock_irqrestore(&zone->lock, flags);
2089                         return;
2090                 }
2091                 spin_unlock_irqrestore(&zone->lock, flags);
2092         }
2093 }
2094 
2095 /* Remove an element from the buddy allocator from the fallback list */
2096 static inline struct page *
2097 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
2098 {
2099         struct free_area *area;
2100         unsigned int current_order;
2101         struct page *page;
2102         int fallback_mt;
2103         bool can_steal;
2104 
2105         /* Find the largest possible block of pages in the other list */
2106         for (current_order = MAX_ORDER-1;
2107                                 current_order >= order && current_order <= MAX_ORDER-1;
2108                                 --current_order) {
2109                 area = &(zone->free_area[current_order]);
2110                 fallback_mt = find_suitable_fallback(area, current_order,
2111                                 start_migratetype, false, &can_steal);
2112                 if (fallback_mt == -1)
2113                         continue;
2114 
2115                 page = list_first_entry(&area->free_list[fallback_mt],
2116                                                 struct page, lru);
2117                 if (can_steal)
2118                         steal_suitable_fallback(zone, page, start_migratetype);
2119 
2120                 /* Remove the page from the freelists */
2121                 area->nr_free--;
2122                 list_del(&page->lru);
2123                 rmv_page_order(page);
2124 
2125                 expand(zone, page, order, current_order, area,
2126                                         start_migratetype);
2127                 /*
2128                  * The pcppage_migratetype may differ from pageblock's
2129                  * migratetype depending on the decisions in
2130                  * find_suitable_fallback(). This is OK as long as it does not
2131                  * differ for MIGRATE_CMA pageblocks. Those can be used as
2132                  * fallback only via special __rmqueue_cma_fallback() function
2133                  */
2134                 set_pcppage_migratetype(page, start_migratetype);
2135 
2136                 trace_mm_page_alloc_extfrag(page, order, current_order,
2137                         start_migratetype, fallback_mt);
2138 
2139                 return page;
2140         }
2141 
2142         return NULL;
2143 }
2144 
2145 /*
2146  * Do the hard work of removing an element from the buddy allocator.
2147  * Call me with the zone->lock already held.
2148  */
2149 static struct page *__rmqueue(struct zone *zone, unsigned int order,
2150                                 int migratetype)
2151 {
2152         struct page *page;
2153 
2154         page = __rmqueue_smallest(zone, order, migratetype);
2155         if (unlikely(!page)) {
2156                 if (migratetype == MIGRATE_MOVABLE)
2157                         page = __rmqueue_cma_fallback(zone, order);
2158 
2159                 if (!page)
2160                         page = __rmqueue_fallback(zone, order, migratetype);
2161         }
2162 
2163         trace_mm_page_alloc_zone_locked(page, order, migratetype);
2164         return page;
2165 }
2166 
2167 /*
2168  * Obtain a specified number of elements from the buddy allocator, all under
2169  * a single hold of the lock, for efficiency.  Add them to the supplied list.
2170  * Returns the number of new pages which were placed at *list.
2171  */
2172 static int rmqueue_bulk(struct zone *zone, unsigned int order,
2173                         unsigned long count, struct list_head *list,
2174                         int migratetype, bool cold)
2175 {
2176         int i;
2177 
2178         spin_lock(&zone->lock);
2179         for (i = 0; i < count; ++i) {
2180                 struct page *page = __rmqueue(zone, order, migratetype);
2181                 if (unlikely(page == NULL))
2182                         break;
2183 
2184                 if (unlikely(check_pcp_refill(page)))
2185                         continue;
2186 
2187                 /*
2188                  * Split buddy pages returned by expand() are received here
2189                  * in physical page order. The page is added to the callers and
2190                  * list and the list head then moves forward. From the callers
2191                  * perspective, the linked list is ordered by page number in
2192                  * some conditions. This is useful for IO devices that can
2193                  * merge IO requests if the physical pages are ordered
2194                  * properly.
2195                  */
2196                 if (likely(!cold))
2197                         list_add(&page->lru, list);
2198                 else
2199                         list_add_tail(&page->lru, list);
2200                 list = &page->lru;
2201                 if (is_migrate_cma(get_pcppage_migratetype(page)))
2202                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2203                                               -(1 << order));
2204         }
2205         __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2206         spin_unlock(&zone->lock);
2207         return i;
2208 }
2209 
2210 #ifdef CONFIG_NUMA
2211 /*
2212  * Called from the vmstat counter updater to drain pagesets of this
2213  * currently executing processor on remote nodes after they have
2214  * expired.
2215  *
2216  * Note that this function must be called with the thread pinned to
2217  * a single processor.
2218  */
2219 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2220 {
2221         unsigned long flags;
2222         int to_drain, batch;
2223 
2224         local_irq_save(flags);
2225         batch = READ_ONCE(pcp->batch);
2226         to_drain = min(pcp->count, batch);
2227         if (to_drain > 0) {
2228                 free_pcppages_bulk(zone, to_drain, pcp);
2229                 pcp->count -= to_drain;
2230         }
2231         local_irq_restore(flags);
2232 }
2233 #endif
2234 
2235 /*
2236  * Drain pcplists of the indicated processor and zone.
2237  *
2238  * The processor must either be the current processor and the
2239  * thread pinned to the current processor or a processor that
2240  * is not online.
2241  */
2242 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2243 {
2244         unsigned long flags;
2245         struct per_cpu_pageset *pset;
2246         struct per_cpu_pages *pcp;
2247 
2248         local_irq_save(flags);
2249         pset = per_cpu_ptr(zone->pageset, cpu);
2250 
2251         pcp = &pset->pcp;
2252         if (pcp->count) {
2253                 free_pcppages_bulk(zone, pcp->count, pcp);
2254                 pcp->count = 0;
2255         }
2256         local_irq_restore(flags);
2257 }
2258 
2259 /*
2260  * Drain pcplists of all zones on the indicated processor.
2261  *
2262  * The processor must either be the current processor and the
2263  * thread pinned to the current processor or a processor that
2264  * is not online.
2265  */
2266 static void drain_pages(unsigned int cpu)
2267 {
2268         struct zone *zone;
2269 
2270         for_each_populated_zone(zone) {
2271                 drain_pages_zone(cpu, zone);
2272         }
2273 }
2274 
2275 /*
2276  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2277  *
2278  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2279  * the single zone's pages.
2280  */
2281 void drain_local_pages(struct zone *zone)
2282 {
2283         int cpu = smp_processor_id();
2284 
2285         if (zone)
2286                 drain_pages_zone(cpu, zone);
2287         else
2288                 drain_pages(cpu);
2289 }
2290 
2291 /*
2292  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2293  *
2294  * When zone parameter is non-NULL, spill just the single zone's pages.
2295  *
2296  * Note that this code is protected against sending an IPI to an offline
2297  * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2298  * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2299  * nothing keeps CPUs from showing up after we populated the cpumask and
2300  * before the call to on_each_cpu_mask().
2301  */
2302 void drain_all_pages(struct zone *zone)
2303 {
2304         int cpu;
2305 
2306         /*
2307          * Allocate in the BSS so we wont require allocation in
2308          * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2309          */
2310         static cpumask_t cpus_with_pcps;
2311 
2312         /*
2313          * We don't care about racing with CPU hotplug event
2314          * as offline notification will cause the notified
2315          * cpu to drain that CPU pcps and on_each_cpu_mask
2316          * disables preemption as part of its processing
2317          */
2318         for_each_online_cpu(cpu) {
2319                 struct per_cpu_pageset *pcp;
2320                 struct zone *z;
2321                 bool has_pcps = false;
2322 
2323                 if (zone) {
2324                         pcp = per_cpu_ptr(zone->pageset, cpu);
2325                         if (pcp->pcp.count)
2326                                 has_pcps = true;
2327                 } else {
2328                         for_each_populated_zone(z) {
2329                                 pcp = per_cpu_ptr(z->pageset, cpu);
2330                                 if (pcp->pcp.count) {
2331                                         has_pcps = true;
2332                                         break;
2333                                 }
2334                         }
2335                 }
2336 
2337                 if (has_pcps)
2338                         cpumask_set_cpu(cpu, &cpus_with_pcps);
2339                 else
2340                         cpumask_clear_cpu(cpu, &cpus_with_pcps);
2341         }
2342         on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
2343                                                                 zone, 1);
2344 }
2345 
2346 #ifdef CONFIG_HIBERNATION
2347 
2348 void mark_free_pages(struct zone *zone)
2349 {
2350         unsigned long pfn, max_zone_pfn;
2351         unsigned long flags;
2352         unsigned int order, t;
2353         struct page *page;
2354 
2355         if (zone_is_empty(zone))
2356                 return;
2357 
2358         spin_lock_irqsave(&zone->lock, flags);
2359 
2360         max_zone_pfn = zone_end_pfn(zone);
2361         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2362                 if (pfn_valid(pfn)) {
2363                         page = pfn_to_page(pfn);
2364 
2365                         if (page_zone(page) != zone)
2366                                 continue;
2367 
2368                         if (!swsusp_page_is_forbidden(page))
2369                                 swsusp_unset_page_free(page);
2370                 }
2371 
2372         for_each_migratetype_order(order, t) {
2373                 list_for_each_entry(page,
2374                                 &zone->free_area[order].free_list[t], lru) {
2375                         unsigned long i;
2376 
2377                         pfn = page_to_pfn(page);
2378                         for (i = 0; i < (1UL << order); i++)
2379                                 swsusp_set_page_free(pfn_to_page(pfn + i));
2380                 }
2381         }
2382         spin_unlock_irqrestore(&zone->lock, flags);
2383 }
2384 #endif /* CONFIG_PM */
2385 
2386 /*
2387  * Free a 0-order page
2388  * cold == true ? free a cold page : free a hot page
2389  */
2390 void free_hot_cold_page(struct page *page, bool cold)
2391 {
2392         struct zone *zone = page_zone(page);
2393         struct per_cpu_pages *pcp;
2394         unsigned long flags;
2395         unsigned long pfn = page_to_pfn(page);
2396         int migratetype;
2397 
2398         if (!free_pcp_prepare(page))
2399                 return;
2400 
2401         migratetype = get_pfnblock_migratetype(page, pfn);
2402         set_pcppage_migratetype(page, migratetype);
2403         local_irq_save(flags);
2404         __count_vm_event(PGFREE);
2405 
2406         /*
2407          * We only track unmovable, reclaimable and movable on pcp lists.
2408          * Free ISOLATE pages back to the allocator because they are being
2409          * offlined but treat RESERVE as movable pages so we can get those
2410          * areas back if necessary. Otherwise, we may have to free
2411          * excessively into the page allocator
2412          */
2413         if (migratetype >= MIGRATE_PCPTYPES) {
2414                 if (unlikely(is_migrate_isolate(migratetype))) {
2415                         free_one_page(zone, page, pfn, 0, migratetype);
2416                         goto out;
2417                 }
2418                 migratetype = MIGRATE_MOVABLE;
2419         }
2420 
2421         pcp = &this_cpu_ptr(zone->pageset)->pcp;
2422         if (!cold)
2423                 list_add(&page->lru, &pcp->lists[migratetype]);
2424         else
2425                 list_add_tail(&page->lru, &pcp->lists[migratetype]);
2426         pcp->count++;
2427         if (pcp->count >= pcp->high) {
2428                 unsigned long batch = READ_ONCE(pcp->batch);
2429                 free_pcppages_bulk(zone, batch, pcp);
2430                 pcp->count -= batch;
2431         }
2432 
2433 out:
2434         local_irq_restore(flags);
2435 }
2436 
2437 /*
2438  * Free a list of 0-order pages
2439  */
2440 void free_hot_cold_page_list(struct list_head *list, bool cold)
2441 {
2442         struct page *page, *next;
2443 
2444         list_for_each_entry_safe(page, next, list, lru) {
2445                 trace_mm_page_free_batched(page, cold);
2446                 free_hot_cold_page(page, cold);
2447         }
2448 }
2449 
2450 /*
2451  * split_page takes a non-compound higher-order page, and splits it into
2452  * n (1<<order) sub-pages: page[0..n]
2453  * Each sub-page must be freed individually.
2454  *
2455  * Note: this is probably too low level an operation for use in drivers.
2456  * Please consult with lkml before using this in your driver.
2457  */
2458 void split_page(struct page *page, unsigned int order)
2459 {
2460         int i;
2461 
2462         VM_BUG_ON_PAGE(PageCompound(page), page);
2463         VM_BUG_ON_PAGE(!page_count(page), page);
2464 
2465 #ifdef CONFIG_KMEMCHECK
2466         /*
2467          * Split shadow pages too, because free(page[0]) would
2468          * otherwise free the whole shadow.
2469          */
2470         if (kmemcheck_page_is_tracked(page))
2471                 split_page(virt_to_page(page[0].shadow), order);
2472 #endif
2473 
2474         for (i = 1; i < (1 << order); i++)
2475                 set_page_refcounted(page + i);
2476         split_page_owner(page, order);
2477 }
2478 EXPORT_SYMBOL_GPL(split_page);
2479 
2480 int __isolate_free_page(struct page *page, unsigned int order)
2481 {
2482         unsigned long watermark;
2483         struct zone *zone;
2484         int mt;
2485 
2486         BUG_ON(!PageBuddy(page));
2487 
2488         zone = page_zone(page);
2489         mt = get_pageblock_migratetype(page);
2490 
2491         if (!is_migrate_isolate(mt)) {
2492                 /* Obey watermarks as if the page was being allocated */
2493                 watermark = low_wmark_pages(zone) + (1 << order);
2494                 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2495                         return 0;
2496 
2497                 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2498         }
2499 
2500         /* Remove page from free list */
2501         list_del(&page->lru);
2502         zone->free_area[order].nr_free--;
2503         rmv_page_order(page);
2504 
2505         /*
2506          * Set the pageblock if the isolated page is at least half of a
2507          * pageblock
2508          */
2509         if (order >= pageblock_order - 1) {
2510                 struct page *endpage = page + (1 << order) - 1;
2511                 for (; page < endpage; page += pageblock_nr_pages) {
2512                         int mt = get_pageblock_migratetype(page);
2513                         if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
2514                                 set_pageblock_migratetype(page,
2515                                                           MIGRATE_MOVABLE);
2516                 }
2517         }
2518 
2519 
2520         return 1UL << order;
2521 }
2522 
2523 /*
2524  * Update NUMA hit/miss statistics
2525  *
2526  * Must be called with interrupts disabled.
2527  *
2528  * When __GFP_OTHER_NODE is set assume the node of the preferred
2529  * zone is the local node. This is useful for daemons who allocate
2530  * memory on behalf of other processes.
2531  */
2532 static inline void zone_statistics(struct zone *preferred_zone, struct zone *z,
2533                                                                 gfp_t flags)
2534 {
2535 #ifdef CONFIG_NUMA
2536         int local_nid = numa_node_id();
2537         enum zone_stat_item local_stat = NUMA_LOCAL;
2538 
2539         if (unlikely(flags & __GFP_OTHER_NODE)) {
2540                 local_stat = NUMA_OTHER;
2541                 local_nid = preferred_zone->node;
2542         }
2543 
2544         if (z->node == local_nid) {
2545                 __inc_zone_state(z, NUMA_HIT);
2546                 __inc_zone_state(z, local_stat);
2547         } else {
2548                 __inc_zone_state(z, NUMA_MISS);
2549                 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
2550         }
2551 #endif
2552 }
2553 
2554 /*
2555  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2556  */
2557 static inline
2558 struct page *buffered_rmqueue(struct zone *preferred_zone,
2559                         struct zone *zone, unsigned int order,
2560                         gfp_t gfp_flags, unsigned int alloc_flags,
2561                         int migratetype)
2562 {
2563         unsigned long flags;
2564         struct page *page;
2565         bool cold = ((gfp_flags & __GFP_COLD) != 0);
2566 
2567         if (likely(order == 0)) {
2568                 struct per_cpu_pages *pcp;
2569                 struct list_head *list;
2570 
2571                 local_irq_save(flags);
2572                 do {
2573                         pcp = &this_cpu_ptr(zone->pageset)->pcp;
2574                         list = &pcp->lists[migratetype];
2575                         if (list_empty(list)) {
2576                                 pcp->count += rmqueue_bulk(zone, 0,
2577                                                 pcp->batch, list,
2578                                                 migratetype, cold);
2579                                 if (unlikely(list_empty(list)))
2580                                         goto failed;
2581                         }
2582 
2583                         if (cold)
2584                                 page = list_last_entry(list, struct page, lru);
2585                         else
2586                                 page = list_first_entry(list, struct page, lru);
2587 
2588                         list_del(&page->lru);
2589                         pcp->count--;
2590 
2591                 } while (check_new_pcp(page));
2592         } else {
2593                 /*
2594                  * We most definitely don't want callers attempting to
2595                  * allocate greater than order-1 page units with __GFP_NOFAIL.
2596                  */
2597                 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
2598                 spin_lock_irqsave(&zone->lock, flags);
2599 
2600                 do {
2601                         page = NULL;
2602                         if (alloc_flags & ALLOC_HARDER) {
2603                                 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2604                                 if (page)
2605                                         trace_mm_page_alloc_zone_locked(page, order, migratetype);
2606                         }
2607                         if (!page)
2608                                 page = __rmqueue(zone, order, migratetype);
2609                 } while (page && check_new_pages(page, order));
2610                 spin_unlock(&zone->lock);
2611                 if (!page)
2612                         goto failed;
2613                 __mod_zone_freepage_state(zone, -(1 << order),
2614                                           get_pcppage_migratetype(page));
2615         }
2616 
2617         __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
2618         zone_statistics(preferred_zone, zone, gfp_flags);
2619         local_irq_restore(flags);
2620 
2621         VM_BUG_ON_PAGE(bad_range(zone, page), page);
2622         return page;
2623 
2624 failed:
2625         local_irq_restore(flags);
2626         return NULL;
2627 }
2628 
2629 #ifdef CONFIG_FAIL_PAGE_ALLOC
2630 
2631 static struct {
2632         struct fault_attr attr;
2633 
2634         bool ignore_gfp_highmem;
2635         bool ignore_gfp_reclaim;
2636         u32 min_order;
2637 } fail_page_alloc = {
2638         .attr = FAULT_ATTR_INITIALIZER,
2639         .ignore_gfp_reclaim = true,
2640         .ignore_gfp_highmem = true,
2641         .min_order = 1,
2642 };
2643 
2644 static int __init setup_fail_page_alloc(char *str)
2645 {
2646         return setup_fault_attr(&fail_page_alloc.attr, str);
2647 }
2648 __setup("fail_page_alloc=", setup_fail_page_alloc);
2649 
2650 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2651 {
2652         if (order < fail_page_alloc.min_order)
2653                 return false;
2654         if (gfp_mask & __GFP_NOFAIL)
2655                 return false;
2656         if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
2657                 return false;
2658         if (fail_page_alloc.ignore_gfp_reclaim &&
2659                         (gfp_mask & __GFP_DIRECT_RECLAIM))
2660                 return false;
2661 
2662         return should_fail(&fail_page_alloc.attr, 1 << order);
2663 }
2664 
2665 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2666 
2667 static int __init fail_page_alloc_debugfs(void)
2668 {
2669         umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
2670         struct dentry *dir;
2671 
2672         dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2673                                         &fail_page_alloc.attr);
2674         if (IS_ERR(dir))
2675                 return PTR_ERR(dir);
2676 
2677         if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2678                                 &fail_page_alloc.ignore_gfp_reclaim))
2679                 goto fail;
2680         if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2681                                 &fail_page_alloc.ignore_gfp_highmem))
2682                 goto fail;
2683         if (!debugfs_create_u32("min-order", mode, dir,
2684                                 &fail_page_alloc.min_order))
2685                 goto fail;
2686 
2687         return 0;
2688 fail:
2689         debugfs_remove_recursive(dir);
2690 
2691         return -ENOMEM;
2692 }
2693 
2694 late_initcall(fail_page_alloc_debugfs);
2695 
2696 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2697 
2698 #else /* CONFIG_FAIL_PAGE_ALLOC */
2699 
2700 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2701 {
2702         return false;
2703 }
2704 
2705 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2706 
2707 /*
2708  * Return true if free base pages are above 'mark'. For high-order checks it
2709  * will return true of the order-0 watermark is reached and there is at least
2710  * one free page of a suitable size. Checking now avoids taking the zone lock
2711  * to check in the allocation paths if no pages are free.
2712  */
2713 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2714                          int classzone_idx, unsigned int alloc_flags,
2715                          long free_pages)
2716 {
2717         long min = mark;
2718         int o;
2719         const bool alloc_harder = (alloc_flags & ALLOC_HARDER);
2720 
2721         /* free_pages may go negative - that's OK */
2722         free_pages -= (1 << order) - 1;
2723 
2724         if (alloc_flags & ALLOC_HIGH)
2725                 min -= min / 2;
2726 
2727         /*
2728          * If the caller does not have rights to ALLOC_HARDER then subtract
2729          * the high-atomic reserves. This will over-estimate the size of the
2730          * atomic reserve but it avoids a search.
2731          */
2732         if (likely(!alloc_harder))
2733                 free_pages -= z->nr_reserved_highatomic;
2734         else
2735                 min -= min / 4;
2736 
2737 #ifdef CONFIG_CMA
2738         /* If allocation can't use CMA areas don't use free CMA pages */
2739         if (!(alloc_flags & ALLOC_CMA))
2740                 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
2741 #endif
2742 
2743         /*
2744          * Check watermarks for an order-0 allocation request. If these
2745          * are not met, then a high-order request also cannot go ahead
2746          * even if a suitable page happened to be free.
2747          */
2748         if (free_pages <= min + z->lowmem_reserve[classzone_idx])
2749                 return false;
2750 
2751         /* If this is an order-0 request then the watermark is fine */
2752         if (!order)
2753                 return true;
2754 
2755         /* For a high-order request, check at least one suitable page is free */
2756         for (o = order; o < MAX_ORDER; o++) {
2757                 struct free_area *area = &z->free_area[o];
2758                 int mt;
2759 
2760                 if (!area->nr_free)
2761                         continue;
2762 
2763                 if (alloc_harder)
2764                         return true;
2765 
2766                 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2767                         if (!list_empty(&area->free_list[mt]))
2768                                 return true;
2769                 }
2770 
2771 #ifdef CONFIG_CMA
2772                 if ((alloc_flags & ALLOC_CMA) &&
2773                     !list_empty(&area->free_list[MIGRATE_CMA])) {
2774                         return true;
2775                 }
2776 #endif
2777         }
2778         return false;
2779 }
2780 
2781 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2782                       int classzone_idx, unsigned int alloc_flags)
2783 {
2784         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2785                                         zone_page_state(z, NR_FREE_PAGES));
2786 }
2787 
2788 static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
2789                 unsigned long mark, int classzone_idx, unsigned int alloc_flags)
2790 {
2791         long free_pages = zone_page_state(z, NR_FREE_PAGES);
2792         long cma_pages = 0;
2793 
2794 #ifdef CONFIG_CMA
2795         /* If allocation can't use CMA areas don't use free CMA pages */
2796         if (!(alloc_flags & ALLOC_CMA))
2797                 cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
2798 #endif
2799 
2800         /*
2801          * Fast check for order-0 only. If this fails then the reserves
2802          * need to be calculated. There is a corner case where the check
2803          * passes but only the high-order atomic reserve are free. If
2804          * the caller is !atomic then it'll uselessly search the free
2805          * list. That corner case is then slower but it is harmless.
2806          */
2807         if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
2808                 return true;
2809 
2810         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2811                                         free_pages);
2812 }
2813 
2814 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2815                         unsigned long mark, int classzone_idx)
2816 {
2817         long free_pages = zone_page_state(z, NR_FREE_PAGES);
2818 
2819         if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2820                 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2821 
2822         return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
2823                                                                 free_pages);
2824 }
2825 
2826 #ifdef CONFIG_NUMA
2827 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2828 {
2829         return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2830                                 RECLAIM_DISTANCE;
2831 }
2832 #else   /* CONFIG_NUMA */
2833 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2834 {
2835         return true;
2836 }
2837 #endif  /* CONFIG_NUMA */
2838 
2839 /*
2840  * get_page_from_freelist goes through the zonelist trying to allocate
2841  * a page.
2842  */
2843 static struct page *
2844 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2845                                                 const struct alloc_context *ac)
2846 {
2847         struct zoneref *z = ac->preferred_zoneref;
2848         struct zone *zone;
2849         struct pglist_data *last_pgdat_dirty_limit = NULL;
2850 
2851         /*
2852          * Scan zonelist, looking for a zone with enough free.
2853          * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2854          */
2855         for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
2856                                                                 ac->nodemask) {
2857                 struct page *page;
2858                 unsigned long mark;
2859 
2860                 if (cpusets_enabled() &&
2861                         (alloc_flags & ALLOC_CPUSET) &&
2862                         !__cpuset_zone_allowed(zone, gfp_mask))
2863                                 continue;
2864                 /*
2865                  * When allocating a page cache page for writing, we
2866                  * want to get it from a node that is within its dirty
2867                  * limit, such that no single node holds more than its
2868                  * proportional share of globally allowed dirty pages.
2869                  * The dirty limits take into account the node's
2870                  * lowmem reserves and high watermark so that kswapd
2871                  * should be able to balance it without having to
2872                  * write pages from its LRU list.
2873                  *
2874                  * XXX: For now, allow allocations to potentially
2875                  * exceed the per-node dirty limit in the slowpath
2876                  * (spread_dirty_pages unset) before going into reclaim,
2877                  * which is important when on a NUMA setup the allowed
2878                  * nodes are together not big enough to reach the
2879                  * global limit.  The proper fix for these situations
2880                  * will require awareness of nodes in the
2881                  * dirty-throttling and the flusher threads.
2882                  */
2883                 if (ac->spread_dirty_pages) {
2884                         if (last_pgdat_dirty_limit == zone->zone_pgdat)
2885                                 continue;
2886 
2887                         if (!node_dirty_ok(zone->zone_pgdat)) {
2888                                 last_pgdat_dirty_limit = zone->zone_pgdat;
2889                                 continue;
2890                         }
2891                 }
2892 
2893                 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2894                 if (!zone_watermark_fast(zone, order, mark,
2895                                        ac_classzone_idx(ac), alloc_flags)) {
2896                         int ret;
2897 
2898                         /* Checked here to keep the fast path fast */
2899                         BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2900                         if (alloc_flags & ALLOC_NO_WATERMARKS)
2901                                 goto try_this_zone;
2902 
2903                         if (node_reclaim_mode == 0 ||
2904                             !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
2905                                 continue;
2906 
2907                         ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
2908                         switch (ret) {
2909                         case NODE_RECLAIM_NOSCAN:
2910                                 /* did not scan */
2911                                 continue;
2912                         case NODE_RECLAIM_FULL:
2913                                 /* scanned but unreclaimable */
2914                                 continue;
2915                         default:
2916                                 /* did we reclaim enough */
2917                                 if (zone_watermark_ok(zone, order, mark,
2918                                                 ac_classzone_idx(ac), alloc_flags))
2919                                         goto try_this_zone;
2920 
2921                                 continue;
2922                         }
2923                 }
2924 
2925 try_this_zone:
2926                 page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order,
2927                                 gfp_mask, alloc_flags, ac->migratetype);
2928                 if (page) {
2929                         prep_new_page(page, order, gfp_mask, alloc_flags);
2930 
2931                         /*
2932                          * If this is a high-order atomic allocation then check
2933                          * if the pageblock should be reserved for the future
2934                          */
2935                         if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2936                                 reserve_highatomic_pageblock(page, zone, order);
2937 
2938                         return page;
2939                 }
2940         }
2941 
2942         return NULL;
2943 }
2944 
2945 /*
2946  * Large machines with many possible nodes should not always dump per-node
2947  * meminfo in irq context.
2948  */
2949 static inline bool should_suppress_show_mem(void)
2950 {
2951         bool ret = false;
2952 
2953 #if NODES_SHIFT > 8
2954         ret = in_interrupt();
2955 #endif
2956         return ret;
2957 }
2958 
2959 static DEFINE_RATELIMIT_STATE(nopage_rs,
2960                 DEFAULT_RATELIMIT_INTERVAL,
2961                 DEFAULT_RATELIMIT_BURST);
2962 
2963 void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
2964 {
2965         unsigned int filter = SHOW_MEM_FILTER_NODES;
2966 
2967         if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2968             debug_guardpage_minorder() > 0)
2969                 return;
2970 
2971         /*
2972          * This documents exceptions given to allocations in certain
2973          * contexts that are allowed to allocate outside current's set
2974          * of allowed nodes.
2975          */
2976         if (!(gfp_mask & __GFP_NOMEMALLOC))
2977                 if (test_thread_flag(TIF_MEMDIE) ||
2978                     (current->flags & (PF_MEMALLOC | PF_EXITING)))
2979                         filter &= ~SHOW_MEM_FILTER_NODES;
2980         if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
2981                 filter &= ~SHOW_MEM_FILTER_NODES;
2982 
2983         if (fmt) {
2984                 struct va_format vaf;
2985                 va_list args;
2986 
2987                 va_start(args, fmt);
2988 
2989                 vaf.fmt = fmt;
2990                 vaf.va = &args;
2991 
2992                 pr_warn("%pV", &vaf);
2993 
2994                 va_end(args);
2995         }
2996 
2997         pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
2998                 current->comm, order, gfp_mask, &gfp_mask);
2999         dump_stack();
3000         if (!should_suppress_show_mem())
3001                 show_mem(filter);
3002 }
3003 
3004 static inline struct page *
3005 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
3006         const struct alloc_context *ac, unsigned long *did_some_progress)
3007 {
3008         struct oom_control oc = {
3009                 .zonelist = ac->zonelist,
3010                 .nodemask = ac->nodemask,
3011                 .memcg = NULL,
3012                 .gfp_mask = gfp_mask,
3013                 .order = order,
3014         };
3015         struct page *page;
3016 
3017         *did_some_progress = 0;
3018 
3019         /*
3020          * Acquire the oom lock.  If that fails, somebody else is
3021          * making progress for us.
3022          */
3023         if (!mutex_trylock(&oom_lock)) {
3024                 *did_some_progress = 1;
3025                 schedule_timeout_uninterruptible(1);
3026                 return NULL;
3027         }
3028 
3029         /*
3030          * Go through the zonelist yet one more time, keep very high watermark
3031          * here, this is only to catch a parallel oom killing, we must fail if
3032          * we're still under heavy pressure.
3033          */
3034         page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
3035                                         ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
3036         if (page)
3037                 goto out;
3038 
3039         if (!(gfp_mask & __GFP_NOFAIL)) {
3040                 /* Coredumps can quickly deplete all memory reserves */
3041                 if (current->flags & PF_DUMPCORE)
3042                         goto out;
3043                 /* The OOM killer will not help higher order allocs */
3044                 if (order > PAGE_ALLOC_COSTLY_ORDER)
3045                         goto out;
3046                 /* The OOM killer does not needlessly kill tasks for lowmem */
3047                 if (ac->high_zoneidx < ZONE_NORMAL)
3048                         goto out;
3049                 if (pm_suspended_storage())
3050                         goto out;
3051                 /*
3052                  * XXX: GFP_NOFS allocations should rather fail than rely on
3053                  * other request to make a forward progress.
3054                  * We are in an unfortunate situation where out_of_memory cannot
3055                  * do much for this context but let's try it to at least get
3056                  * access to memory reserved if the current task is killed (see
3057                  * out_of_memory). Once filesystems are ready to handle allocation
3058                  * failures more gracefully we should just bail out here.
3059                  */
3060 
3061                 /* The OOM killer may not free memory on a specific node */
3062                 if (gfp_mask & __GFP_THISNODE)
3063                         goto out;
3064         }
3065         /* Exhausted what can be done so it's blamo time */
3066         if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3067                 *did_some_progress = 1;
3068 
3069                 if (gfp_mask & __GFP_NOFAIL) {
3070                         page = get_page_from_freelist(gfp_mask, order,
3071                                         ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac);
3072                         /*
3073                          * fallback to ignore cpuset restriction if our nodes
3074                          * are depleted
3075                          */
3076                         if (!page)
3077                                 page = get_page_from_freelist(gfp_mask, order,
3078                                         ALLOC_NO_WATERMARKS, ac);
3079                 }
3080         }
3081 out:
3082         mutex_unlock(&oom_lock);
3083         return page;
3084 }
3085 
3086 /*
3087  * Maximum number of compaction retries wit a progress before OOM
3088  * killer is consider as the only way to move forward.
3089  */
3090 #define MAX_COMPACT_RETRIES 16
3091 
3092 #ifdef CONFIG_COMPACTION
3093 /* Try memory compaction for high-order allocations before reclaim */
3094 static struct page *
3095 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3096                 unsigned int alloc_flags, const struct alloc_context *ac,
3097                 enum compact_priority prio, enum compact_result *compact_result)
3098 {
3099         struct page *page;
3100 
3101         if (!order)
3102                 return NULL;
3103 
3104         current->flags |= PF_MEMALLOC;
3105         *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3106                                                                         prio);
3107         current->flags &= ~PF_MEMALLOC;
3108 
3109         if (*compact_result <= COMPACT_INACTIVE)
3110                 return NULL;
3111 
3112         /*
3113          * At least in one zone compaction wasn't deferred or skipped, so let's
3114          * count a compaction stall
3115          */
3116         count_vm_event(COMPACTSTALL);
3117 
3118         page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3119 
3120         if (page) {
3121                 struct zone *zone = page_zone(page);
3122 
3123                 zone->compact_blockskip_flush = false;
3124                 compaction_defer_reset(zone, order, true);
3125                 count_vm_event(COMPACTSUCCESS);
3126                 return page;
3127         }
3128 
3129         /*
3130          * It's bad if compaction run occurs and fails. The most likely reason
3131          * is that pages exist, but not enough to satisfy watermarks.
3132          */
3133         count_vm_event(COMPACTFAIL);
3134 
3135         cond_resched();
3136 
3137         return NULL;
3138 }
3139 
3140 #else
3141 static inline struct page *
3142 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3143                 unsigned int alloc_flags, const struct alloc_context *ac,
3144                 enum compact_priority prio, enum compact_result *compact_result)
3145 {
3146         *compact_result = COMPACT_SKIPPED;
3147         return NULL;
3148 }
3149 
3150 #endif /* CONFIG_COMPACTION */
3151 
3152 static inline bool
3153 should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3154                      enum compact_result compact_result,
3155                      enum compact_priority *compact_priority,
3156                      int compaction_retries)
3157 {
3158         struct zone *zone;
3159         struct zoneref *z;
3160 
3161         if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3162                 return false;
3163 
3164         /*
3165          * There are setups with compaction disabled which would prefer to loop
3166          * inside the allocator rather than hit the oom killer prematurely.
3167          * Let's give them a good hope and keep retrying while the order-0
3168          * watermarks are OK.
3169          */
3170         for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3171                                         ac->nodemask) {
3172                 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3173                                         ac_classzone_idx(ac), alloc_flags))
3174                         return true;
3175         }
3176         return false;
3177 }
3178 
3179 /* Perform direct synchronous page reclaim */
3180 static int
3181 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
3182                                         const struct alloc_context *ac)
3183 {
3184         struct reclaim_state reclaim_state;
3185         int progress;
3186 
3187         cond_resched();
3188 
3189         /* We now go into synchronous reclaim */
3190         cpuset_memory_pressure_bump();
3191         current->flags |= PF_MEMALLOC;
3192         lockdep_set_current_reclaim_state(gfp_mask);
3193         reclaim_state.reclaimed_slab = 0;
3194         current->reclaim_state = &reclaim_state;
3195 
3196         progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3197                                                                 ac->nodemask);
3198 
3199         current->reclaim_state = NULL;
3200         lockdep_clear_current_reclaim_state();
3201         current->flags &= ~PF_MEMALLOC;
3202 
3203         cond_resched();
3204 
3205         return progress;
3206 }
3207 
3208 /* The really slow allocator path where we enter direct reclaim */
3209 static inline struct page *
3210 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
3211                 unsigned int alloc_flags, const struct alloc_context *ac,
3212                 unsigned long *did_some_progress)
3213 {
3214         struct page *page = NULL;
3215         bool drained = false;
3216 
3217         *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
3218         if (unlikely(!(*did_some_progress)))
3219                 return NULL;
3220 
3221 retry:
3222         page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3223 
3224         /*
3225          * If an allocation failed after direct reclaim, it could be because
3226          * pages are pinned on the per-cpu lists or in high alloc reserves.
3227          * Shrink them them and try again
3228          */
3229         if (!page && !drained) {
3230                 unreserve_highatomic_pageblock(ac);
3231                 drain_all_pages(NULL);
3232                 drained = true;
3233                 goto retry;
3234         }
3235 
3236         return page;
3237 }
3238 
3239 static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
3240 {
3241         struct zoneref *z;
3242         struct zone *zone;
3243         pg_data_t *last_pgdat = NULL;
3244 
3245         for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
3246                                         ac->high_zoneidx, ac->nodemask) {
3247                 if (last_pgdat != zone->zone_pgdat)
3248                         wakeup_kswapd(zone, order, ac->high_zoneidx);
3249                 last_pgdat = zone->zone_pgdat;
3250         }
3251 }
3252 
3253 static inline unsigned int
3254 gfp_to_alloc_flags(gfp_t gfp_mask)
3255 {
3256         unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
3257 
3258         /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3259         BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
3260 
3261         /*
3262          * The caller may dip into page reserves a bit more if the caller
3263          * cannot run direct reclaim, or if the caller has realtime scheduling
3264          * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
3265          * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3266          */
3267         alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
3268 
3269         if (gfp_mask & __GFP_ATOMIC) {
3270                 /*
3271                  * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3272                  * if it can't schedule.
3273                  */
3274                 if (!(gfp_mask & __GFP_NOMEMALLOC))
3275                         alloc_flags |= ALLOC_HARDER;
3276                 /*
3277                  * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3278                  * comment for __cpuset_node_allowed().
3279                  */
3280                 alloc_flags &= ~ALLOC_CPUSET;
3281         } else if (unlikely(rt_task(current)) && !in_interrupt())
3282                 alloc_flags |= ALLOC_HARDER;
3283 
3284 #ifdef CONFIG_CMA
3285         if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
3286                 alloc_flags |= ALLOC_CMA;
3287 #endif
3288         return alloc_flags;
3289 }
3290 
3291 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3292 {
3293         if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
3294                 return false;
3295 
3296         if (gfp_mask & __GFP_MEMALLOC)
3297                 return true;
3298         if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3299                 return true;
3300         if (!in_interrupt() &&
3301                         ((current->flags & PF_MEMALLOC) ||
3302                          unlikely(test_thread_flag(TIF_MEMDIE))))
3303                 return true;
3304 
3305         return false;
3306 }
3307 
3308 /*
3309  * Maximum number of reclaim retries without any progress before OOM killer
3310  * is consider as the only way to move forward.
3311  */
3312 #define MAX_RECLAIM_RETRIES 16
3313 
3314 /*
3315  * Checks whether it makes sense to retry the reclaim to make a forward progress
3316  * for the given allocation request.
3317  * The reclaim feedback represented by did_some_progress (any progress during
3318  * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3319  * any progress in a row) is considered as well as the reclaimable pages on the
3320  * applicable zone list (with a backoff mechanism which is a function of
3321  * no_progress_loops).
3322  *
3323  * Returns true if a retry is viable or false to enter the oom path.
3324  */
3325 static inline bool
3326 should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3327                      struct alloc_context *ac, int alloc_flags,
3328                      bool did_some_progress, int no_progress_loops)
3329 {
3330         struct zone *zone;
3331         struct zoneref *z;
3332 
3333         /*
3334          * Make sure we converge to OOM if we cannot make any progress
3335          * several times in the row.
3336          */
3337         if (no_progress_loops > MAX_RECLAIM_RETRIES)
3338                 return false;
3339 
3340         /*
3341          * Keep reclaiming pages while there is a chance this will lead
3342          * somewhere.  If none of the target zones can satisfy our allocation
3343          * request even if all reclaimable pages are considered then we are
3344          * screwed and have to go OOM.
3345          */
3346         for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3347                                         ac->nodemask) {
3348                 unsigned long available;
3349                 unsigned long reclaimable;
3350 
3351                 available = reclaimable = zone_reclaimable_pages(zone);
3352                 available -= DIV_ROUND_UP(no_progress_loops * available,
3353                                           MAX_RECLAIM_RETRIES);
3354                 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
3355 
3356                 /*
3357                  * Would the allocation succeed if we reclaimed the whole
3358                  * available?
3359                  */
3360                 if (__zone_watermark_ok(zone, order, min_wmark_pages(zone),
3361                                 ac_classzone_idx(ac), alloc_flags, available)) {
3362                         /*
3363                          * If we didn't make any progress and have a lot of
3364                          * dirty + writeback pages then we should wait for
3365                          * an IO to complete to slow down the reclaim and
3366                          * prevent from pre mature OOM
3367                          */
3368                         if (!did_some_progress) {
3369                                 unsigned long write_pending;
3370 
3371                                 write_pending = zone_page_state_snapshot(zone,
3372                                                         NR_ZONE_WRITE_PENDING);
3373 
3374                                 if (2 * write_pending > reclaimable) {
3375                                         congestion_wait(BLK_RW_ASYNC, HZ/10);
3376                                         return true;
3377                                 }
3378                         }
3379 
3380                         /*
3381                          * Memory allocation/reclaim might be called from a WQ
3382                          * context and the current implementation of the WQ
3383                          * concurrency control doesn't recognize that
3384                          * a particular WQ is congested if the worker thread is
3385                          * looping without ever sleeping. Therefore we have to
3386                          * do a short sleep here rather than calling
3387                          * cond_resched().
3388                          */
3389                         if (current->flags & PF_WQ_WORKER)
3390                                 schedule_timeout_uninterruptible(1);
3391                         else
3392                                 cond_resched();
3393 
3394                         return true;
3395                 }
3396         }
3397 
3398         return false;
3399 }
3400 
3401 static inline struct page *
3402 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
3403                                                 struct alloc_context *ac)
3404 {
3405         bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
3406         struct page *page = NULL;
3407         unsigned int alloc_flags;
3408         unsigned long did_some_progress;
3409         enum compact_priority compact_priority = DEF_COMPACT_PRIORITY;
3410         enum compact_result compact_result;
3411         int compaction_retries = 0;
3412         int no_progress_loops = 0;
3413 
3414         /*
3415          * In the slowpath, we sanity check order to avoid ever trying to
3416          * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3417          * be using allocators in order of preference for an area that is
3418          * too large.
3419          */
3420         if (order >= MAX_ORDER) {
3421                 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
3422                 return NULL;
3423         }
3424 
3425         /*
3426          * We also sanity check to catch abuse of atomic reserves being used by
3427          * callers that are not in atomic context.
3428          */
3429         if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3430                                 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3431                 gfp_mask &= ~__GFP_ATOMIC;
3432 
3433         /*
3434          * The fast path uses conservative alloc_flags to succeed only until
3435          * kswapd needs to be woken up, and to avoid the cost of setting up
3436          * alloc_flags precisely. So we do that now.
3437          */
3438         alloc_flags = gfp_to_alloc_flags(gfp_mask);
3439 
3440         if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3441                 wake_all_kswapds(order, ac);
3442 
3443         /*
3444          * The adjusted alloc_flags might result in immediate success, so try
3445          * that first
3446          */
3447         page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3448         if (page)
3449                 goto got_pg;
3450 
3451         /*
3452          * For costly allocations, try direct compaction first, as it's likely
3453          * that we have enough base pages and don't need to reclaim. Don't try
3454          * that for allocations that are allowed to ignore watermarks, as the
3455          * ALLOC_NO_WATERMARKS attempt didn't yet happen.
3456          */
3457         if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER &&
3458                 !gfp_pfmemalloc_allowed(gfp_mask)) {
3459                 page = __alloc_pages_direct_compact(gfp_mask, order,
3460                                                 alloc_flags, ac,
3461                                                 INIT_COMPACT_PRIORITY,
3462                                                 &compact_result);
3463                 if (page)
3464                         goto got_pg;
3465 
3466                 /*
3467                  * Checks for costly allocations with __GFP_NORETRY, which
3468                  * includes THP page fault allocations
3469                  */
3470                 if (gfp_mask & __GFP_NORETRY) {
3471                         /*
3472                          * If compaction is deferred for high-order allocations,
3473                          * it is because sync compaction recently failed. If
3474                          * this is the case and the caller requested a THP
3475                          * allocation, we do not want to heavily disrupt the
3476                          * system, so we fail the allocation instead of entering
3477                          * direct reclaim.
3478                          */
3479                         if (compact_result == COMPACT_DEFERRED)
3480                                 goto nopage;
3481 
3482                         /*
3483                          * Looks like reclaim/compaction is worth trying, but
3484                          * sync compaction could be very expensive, so keep
3485                          * using async compaction.
3486                          */
3487                         compact_priority = INIT_COMPACT_PRIORITY;
3488                 }
3489         }
3490 
3491 retry:
3492         /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
3493         if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3494                 wake_all_kswapds(order, ac);
3495 
3496         if (gfp_pfmemalloc_allowed(gfp_mask))
3497                 alloc_flags = ALLOC_NO_WATERMARKS;
3498 
3499         /*
3500          * Reset the zonelist iterators if memory policies can be ignored.
3501          * These allocations are high priority and system rather than user
3502          * orientated.
3503          */
3504         if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) {
3505                 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3506                 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
3507                                         ac->high_zoneidx, ac->nodemask);
3508         }
3509 
3510         /* Attempt with potentially adjusted zonelist and alloc_flags */
3511         page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3512         if (page)
3513                 goto got_pg;
3514 
3515         /* Caller is not willing to reclaim, we can't balance anything */
3516         if (!can_direct_reclaim) {
3517                 /*
3518                  * All existing users of the __GFP_NOFAIL are blockable, so warn
3519                  * of any new users that actually allow this type of allocation
3520                  * to fail.
3521                  */
3522                 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
3523                 goto nopage;
3524         }
3525 
3526         /* Avoid recursion of direct reclaim */
3527         if (current->flags & PF_MEMALLOC) {
3528                 /*
3529                  * __GFP_NOFAIL request from this context is rather bizarre
3530                  * because we cannot reclaim anything and only can loop waiting
3531                  * for somebody to do a work for us.
3532                  */
3533                 if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3534                         cond_resched();
3535                         goto retry;
3536                 }
3537                 goto nopage;
3538         }
3539 
3540         /* Avoid allocations with no watermarks from looping endlessly */
3541         if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3542                 goto nopage;
3543 
3544 
3545         /* Try direct reclaim and then allocating */
3546         page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3547                                                         &did_some_progress);
3548         if (page)
3549                 goto got_pg;
3550 
3551         /* Try direct compaction and then allocating */
3552         page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3553                                         compact_priority, &compact_result);
3554         if (page)
3555                 goto got_pg;
3556 
3557         if (order && compaction_made_progress(compact_result))
3558                 compaction_retries++;
3559 
3560         /* Do not loop if specifically requested */
3561         if (gfp_mask & __GFP_NORETRY)
3562                 goto nopage;
3563 
3564         /*
3565          * Do not retry costly high order allocations unless they are
3566          * __GFP_REPEAT
3567          */
3568         if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT))
3569                 goto nopage;
3570 
3571         /*
3572          * Costly allocations might have made a progress but this doesn't mean
3573          * their order will become available due to high fragmentation so
3574          * always increment the no progress counter for them
3575          */
3576         if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
3577                 no_progress_loops = 0;
3578         else
3579                 no_progress_loops++;
3580 
3581         if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
3582                                  did_some_progress > 0, no_progress_loops))
3583                 goto retry;
3584 
3585         /*
3586          * It doesn't make any sense to retry for the compaction if the order-0
3587          * reclaim is not able to make any progress because the current
3588          * implementation of the compaction depends on the sufficient amount
3589          * of free memory (see __compaction_suitable)
3590          */
3591         if (did_some_progress > 0 &&
3592                         should_compact_retry(ac, order, alloc_flags,
3593                                 compact_result, &compact_priority,
3594                                 compaction_retries))
3595                 goto retry;
3596 
3597         /* Reclaim has failed us, start killing things */
3598         page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3599         if (page)
3600                 goto got_pg;
3601 
3602         /* Retry as long as the OOM killer is making progress */
3603         if (did_some_progress) {
3604                 no_progress_loops = 0;
3605                 goto retry;
3606         }
3607 
3608 nopage:
3609         warn_alloc_failed(gfp_mask, order, NULL);
3610 got_pg:
3611         return page;
3612 }
3613 
3614 /*
3615  * This is the 'heart' of the zoned buddy allocator.
3616  */
3617 struct page *
3618 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3619                         struct zonelist *zonelist, nodemask_t *nodemask)
3620 {
3621         struct page *page;
3622         unsigned int cpuset_mems_cookie;
3623         unsigned int alloc_flags = ALLOC_WMARK_LOW;
3624         gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */
3625         struct alloc_context ac = {
3626                 .high_zoneidx = gfp_zone(gfp_mask),
3627                 .zonelist = zonelist,
3628                 .nodemask = nodemask,
3629                 .migratetype = gfpflags_to_migratetype(gfp_mask),
3630         };
3631 
3632         if (cpusets_enabled()) {
3633                 alloc_mask |= __GFP_HARDWALL;
3634                 alloc_flags |= ALLOC_CPUSET;
3635                 if (!ac.nodemask)
3636                         ac.nodemask = &cpuset_current_mems_allowed;
3637         }
3638 
3639         gfp_mask &= gfp_allowed_mask;
3640 
3641         lockdep_trace_alloc(gfp_mask);
3642 
3643         might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
3644 
3645         if (should_fail_alloc_page(gfp_mask, order))
3646                 return NULL;
3647 
3648         /*
3649          * Check the zones suitable for the gfp_mask contain at least one
3650          * valid zone. It's possible to have an empty zonelist as a result
3651          * of __GFP_THISNODE and a memoryless node
3652          */
3653         if (unlikely(!zonelist->_zonerefs->zone))
3654                 return NULL;
3655 
3656         if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
3657                 alloc_flags |= ALLOC_CMA;
3658 
3659 retry_cpuset:
3660         cpuset_mems_cookie = read_mems_allowed_begin();
3661 
3662         /* Dirty zone balancing only done in the fast path */
3663         ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3664 
3665         /*
3666          * The preferred zone is used for statistics but crucially it is
3667          * also used as the starting point for the zonelist iterator. It
3668          * may get reset for allocations that ignore memory policies.
3669          */
3670         ac.preferred_zoneref = first_zones_zonelist(ac.zonelist,
3671                                         ac.high_zoneidx, ac.nodemask);
3672         if (!ac.preferred_zoneref) {
3673                 page = NULL;
3674                 goto no_zone;
3675         }
3676 
3677         /* First allocation attempt */
3678         page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
3679         if (likely(page))
3680                 goto out;
3681 
3682         /*
3683          * Runtime PM, block IO and its error handling path can deadlock
3684          * because I/O on the device might not complete.
3685          */
3686         alloc_mask = memalloc_noio_flags(gfp_mask);
3687         ac.spread_dirty_pages = false;
3688 
3689         /*
3690          * Restore the original nodemask if it was potentially replaced with
3691          * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3692          */
3693         if (cpusets_enabled())
3694                 ac.nodemask = nodemask;
3695         page = __alloc_pages_slowpath(alloc_mask, order, &ac);
3696 
3697 no_zone:
3698         /*
3699          * When updating a task's mems_allowed, it is possible to race with
3700          * parallel threads in such a way that an allocation can fail while
3701          * the mask is being updated. If a page allocation is about to fail,
3702          * check if the cpuset changed during allocation and if so, retry.
3703          */
3704         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) {
3705                 alloc_mask = gfp_mask;
3706                 goto retry_cpuset;
3707         }
3708 
3709 out:
3710         if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
3711             unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) {
3712                 __free_pages(page, order);
3713                 page = NULL;
3714         }
3715 
3716         if (kmemcheck_enabled && page)
3717                 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3718 
3719         trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3720 
3721         return page;
3722 }
3723 EXPORT_SYMBOL(__alloc_pages_nodemask);
3724 
3725 /*
3726  * Common helper functions.
3727  */
3728 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
3729 {
3730         struct page *page;
3731 
3732         /*
3733          * __get_free_pages() returns a 32-bit address, which cannot represent
3734          * a highmem page
3735          */
3736         VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3737 
3738         page = alloc_pages(gfp_mask, order);
3739         if (!page)
3740                 return 0;
3741         return (unsigned long) page_address(page);
3742 }
3743 EXPORT_SYMBOL(__get_free_pages);
3744 
3745 unsigned long get_zeroed_page(gfp_t gfp_mask)
3746 {
3747         return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
3748 }
3749 EXPORT_SYMBOL(get_zeroed_page);
3750 
3751 void __free_pages(struct page *page, unsigned int order)
3752 {
3753         if (put_page_testzero(page)) {
3754                 if (order == 0)
3755                         free_hot_cold_page(page, false);
3756                 else
3757                         __free_pages_ok(page, order);
3758         }
3759 }
3760 
3761 EXPORT_SYMBOL(__free_pages);
3762 
3763 void free_pages(unsigned long addr, unsigned int order)
3764 {
3765         if (addr != 0) {
3766                 VM_BUG_ON(!virt_addr_valid((void *)addr));
3767                 __free_pages(virt_to_page((void *)addr), order);
3768         }
3769 }
3770 
3771 EXPORT_SYMBOL(free_pages);
3772 
3773 /*
3774  * Page Fragment:
3775  *  An arbitrary-length arbitrary-offset area of memory which resides
3776  *  within a 0 or higher order page.  Multiple fragments within that page
3777  *  are individually refcounted, in the page's reference counter.
3778  *
3779  * The page_frag functions below provide a simple allocation framework for
3780  * page fragments.  This is used by the network stack and network device
3781  * drivers to provide a backing region of memory for use as either an
3782  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3783  */
3784 static struct page *__page_frag_refill(struct page_frag_cache *nc,
3785                                        gfp_t gfp_mask)
3786 {
3787         struct page *page = NULL;
3788         gfp_t gfp = gfp_mask;
3789 
3790 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3791         gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3792                     __GFP_NOMEMALLOC;
3793         page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3794                                 PAGE_FRAG_CACHE_MAX_ORDER);
3795         nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3796 #endif
3797         if (unlikely(!page))
3798                 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3799 
3800         nc->va = page ? page_address(page) : NULL;
3801 
3802         return page;
3803 }
3804 
3805 void *__alloc_page_frag(struct page_frag_cache *nc,
3806                         unsigned int fragsz, gfp_t gfp_mask)
3807 {
3808         unsigned int size = PAGE_SIZE;
3809         struct page *page;
3810         int offset;
3811 
3812         if (unlikely(!nc->va)) {
3813 refill:
3814                 page = __page_frag_refill(nc, gfp_mask);
3815                 if (!page)
3816                         return NULL;
3817 
3818 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3819                 /* if size can vary use size else just use PAGE_SIZE */
3820                 size = nc->size;
3821 #endif
3822                 /* Even if we own the page, we do not use atomic_set().
3823                  * This would break get_page_unless_zero() users.
3824                  */
3825                 page_ref_add(page, size - 1);
3826 
3827                 /* reset page count bias and offset to start of new frag */
3828                 nc->pfmemalloc = page_is_pfmemalloc(page);
3829                 nc->pagecnt_bias = size;
3830                 nc->offset = size;
3831         }
3832 
3833         offset = nc->offset - fragsz;
3834         if (unlikely(offset < 0)) {
3835                 page = virt_to_page(nc->va);
3836 
3837                 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
3838                         goto refill;
3839 
3840 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3841                 /* if size can vary use size else just use PAGE_SIZE */
3842                 size = nc->size;
3843 #endif
3844                 /* OK, page count is 0, we can safely set it */
3845                 set_page_count(page, size);
3846 
3847                 /* reset page count bias and offset to start of new frag */
3848                 nc->pagecnt_bias = size;
3849                 offset = size - fragsz;
3850         }
3851 
3852         nc->pagecnt_bias--;
3853         nc->offset = offset;
3854 
3855         return nc->va + offset;
3856 }
3857 EXPORT_SYMBOL(__alloc_page_frag);
3858 
3859 /*
3860  * Frees a page fragment allocated out of either a compound or order 0 page.
3861  */
3862 void __free_page_frag(void *addr)
3863 {
3864         struct page *page = virt_to_head_page(addr);
3865 
3866         if (unlikely(put_page_testzero(page)))
3867                 __free_pages_ok(page, compound_order(page));
3868 }
3869 EXPORT_SYMBOL(__free_page_frag);
3870 
3871 static void *make_alloc_exact(unsigned long addr, unsigned int order,
3872                 size_t size)
3873 {
3874         if (addr) {
3875                 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3876                 unsigned long used = addr + PAGE_ALIGN(size);
3877 
3878                 split_page(virt_to_page((void *)addr), order);
3879                 while (used < alloc_end) {
3880                         free_page(used);
3881                         used += PAGE_SIZE;
3882                 }
3883         }
3884         return (void *)addr;
3885 }
3886 
3887 /**
3888  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3889  * @size: the number of bytes to allocate
3890  * @gfp_mask: GFP flags for the allocation
3891  *
3892  * This function is similar to alloc_pages(), except that it allocates the
3893  * minimum number of pages to satisfy the request.  alloc_pages() can only
3894  * allocate memory in power-of-two pages.
3895  *
3896  * This function is also limited by MAX_ORDER.
3897  *
3898  * Memory allocated by this function must be released by free_pages_exact().
3899  */
3900 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3901 {
3902         unsigned int order = get_order(size);
3903         unsigned long addr;
3904 
3905         addr = __get_free_pages(gfp_mask, order);
3906         return make_alloc_exact(addr, order, size);
3907 }
3908 EXPORT_SYMBOL(alloc_pages_exact);
3909 
3910 /**
3911  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3912  *                         pages on a node.
3913  * @nid: the preferred node ID where memory should be allocated
3914  * @size: the number of bytes to allocate
3915  * @gfp_mask: GFP flags for the allocation
3916  *
3917  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3918  * back.
3919  */
3920 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
3921 {
3922         unsigned int order = get_order(size);
3923         struct page *p = alloc_pages_node(nid, gfp_mask, order);
3924         if (!p)
3925                 return NULL;
3926         return make_alloc_exact((unsigned long)page_address(p), order, size);
3927 }
3928 
3929 /**
3930  * free_pages_exact - release memory allocated via alloc_pages_exact()
3931  * @virt: the value returned by alloc_pages_exact.
3932  * @size: size of allocation, same value as passed to alloc_pages_exact().
3933  *
3934  * Release the memory allocated by a previous call to alloc_pages_exact.
3935  */
3936 void free_pages_exact(void *virt, size_t size)
3937 {
3938         unsigned long addr = (unsigned long)virt;
3939         unsigned long end = addr + PAGE_ALIGN(size);
3940 
3941         while (addr < end) {
3942                 free_page(addr);
3943                 addr += PAGE_SIZE;
3944         }
3945 }
3946 EXPORT_SYMBOL(free_pages_exact);
3947 
3948 /**
3949  * nr_free_zone_pages - count number of pages beyond high watermark
3950  * @offset: The zone index of the highest zone
3951  *
3952  * nr_free_zone_pages() counts the number of counts pages which are beyond the
3953  * high watermark within all zones at or below a given zone index.  For each
3954  * zone, the number of pages is calculated as:
3955  *     managed_pages - high_pages
3956  */
3957 static unsigned long nr_free_zone_pages(int offset)
3958 {
3959         struct zoneref *z;
3960         struct zone *zone;
3961 
3962         /* Just pick one node, since fallback list is circular */
3963         unsigned long sum = 0;
3964 
3965         struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3966 
3967         for_each_zone_zonelist(zone, z, zonelist, offset) {
3968                 unsigned long size = zone->managed_pages;
3969                 unsigned long high = high_wmark_pages(zone);
3970                 if (size > high)
3971                         sum += size - high;
3972         }
3973 
3974         return sum;
3975 }
3976 
3977 /**
3978  * nr_free_buffer_pages - count number of pages beyond high watermark
3979  *
3980  * nr_free_buffer_pages() counts the number of pages which are beyond the high
3981  * watermark within ZONE_DMA and ZONE_NORMAL.
3982  */
3983 unsigned long nr_free_buffer_pages(void)
3984 {
3985         return nr_free_zone_pages(gfp_zone(GFP_USER));
3986 }
3987 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3988 
3989 /**
3990  * nr_free_pagecache_pages - count number of pages beyond high watermark
3991  *
3992  * nr_free_pagecache_pages() counts the number of pages which are beyond the
3993  * high watermark within all zones.
3994  */
3995 unsigned long nr_free_pagecache_pages(void)
3996 {
3997         return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3998 }
3999 
4000 static inline void show_node(struct zone *zone)
4001 {
4002         if (IS_ENABLED(CONFIG_NUMA))
4003                 printk("Node %d ", zone_to_nid(zone));
4004 }
4005 
4006 long si_mem_available(void)
4007 {
4008         long available;
4009         unsigned long pagecache;
4010         unsigned long wmark_low = 0;
4011         unsigned long pages[NR_LRU_LISTS];
4012         struct zone *zone;
4013         int lru;
4014 
4015         for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
4016                 pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
4017 
4018         for_each_zone(zone)
4019                 wmark_low += zone->watermark[WMARK_LOW];
4020 
4021         /*
4022          * Estimate the amount of memory available for userspace allocations,
4023          * without causing swapping.
4024          */
4025         available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
4026 
4027         /*
4028          * Not all the page cache can be freed, otherwise the system will
4029          * start swapping. Assume at least half of the page cache, or the
4030          * low watermark worth of cache, needs to stay.
4031          */
4032         pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4033         pagecache -= min(pagecache / 2, wmark_low);
4034         available += pagecache;
4035 
4036         /*
4037          * Part of the reclaimable slab consists of items that are in use,
4038          * and cannot be freed. Cap this estimate at the low watermark.
4039          */
4040         available += global_page_state(NR_SLAB_RECLAIMABLE) -
4041                      min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
4042 
4043         if (available < 0)
4044                 available = 0;
4045         return available;
4046 }
4047 EXPORT_SYMBOL_GPL(si_mem_available);
4048 
4049 void si_meminfo(struct sysinfo *val)
4050 {
4051         val->totalram = totalram_pages;
4052         val->sharedram = global_node_page_state(NR_SHMEM);
4053         val->freeram = global_page_state(NR_FREE_PAGES);
4054         val->bufferram = nr_blockdev_pages();
4055         val->totalhigh = totalhigh_pages;
4056         val->freehigh = nr_free_highpages();
4057         val->mem_unit = PAGE_SIZE;
4058 }
4059 
4060 EXPORT_SYMBOL(si_meminfo);
4061 
4062 #ifdef CONFIG_NUMA
4063 void si_meminfo_node(struct sysinfo *val, int nid)
4064 {
4065         int zone_type;          /* needs to be signed */
4066         unsigned long managed_pages = 0;
4067         unsigned long managed_highpages = 0;
4068         unsigned long free_highpages = 0;
4069         pg_data_t *pgdat = NODE_DATA(nid);
4070 
4071         for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4072                 managed_pages += pgdat->node_zones[zone_type].managed_pages;
4073         val->totalram = managed_pages;
4074         val->sharedram = node_page_state(pgdat, NR_SHMEM);
4075         val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
4076 #ifdef CONFIG_HIGHMEM
4077         for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4078                 struct zone *zone = &pgdat->node_zones[zone_type];
4079 
4080                 if (is_highmem(zone)) {
4081                         managed_highpages += zone->managed_pages;
4082                         free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4083                 }
4084         }
4085         val->totalhigh = managed_highpages;
4086         val->freehigh = free_highpages;
4087 #else
4088         val->totalhigh = managed_highpages;
4089         val->freehigh = free_highpages;
4090 #endif
4091         val->mem_unit = PAGE_SIZE;
4092 }
4093 #endif
4094 
4095 /*
4096  * Determine whether the node should be displayed or not, depending on whether
4097  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4098  */
4099 bool skip_free_areas_node(unsigned int flags, int nid)
4100 {
4101         bool ret = false;
4102         unsigned int cpuset_mems_cookie;
4103 
4104         if (!(flags & SHOW_MEM_FILTER_NODES))
4105                 goto out;
4106 
4107         do {
4108                 cpuset_mems_cookie = read_mems_allowed_begin();
4109                 ret = !node_isset(nid, cpuset_current_mems_allowed);
4110         } while (read_mems_allowed_retry(cpuset_mems_cookie));
4111 out:
4112         return ret;
4113 }
4114 
4115 #define K(x) ((x) << (PAGE_SHIFT-10))
4116 
4117 static void show_migration_types(unsigned char type)
4118 {
4119         static const char types[MIGRATE_TYPES] = {
4120                 [MIGRATE_UNMOVABLE]     = 'U',
4121                 [MIGRATE_MOVABLE]       = 'M',
4122                 [MIGRATE_RECLAIMABLE]   = 'E',
4123                 [MIGRATE_HIGHATOMIC]    = 'H',
4124 #ifdef CONFIG_CMA
4125                 [MIGRATE_CMA]           = 'C',
4126 #endif
4127 #ifdef CONFIG_MEMORY_ISOLATION
4128                 [MIGRATE_ISOLATE]       = 'I',
4129 #endif
4130         };
4131         char tmp[MIGRATE_TYPES + 1];
4132         char *p = tmp;
4133         int i;
4134 
4135         for (i = 0; i < MIGRATE_TYPES; i++) {
4136                 if (type & (1 << i))
4137                         *p++ = types[i];
4138         }
4139 
4140         *p = '\0';
4141         printk("(%s) ", tmp);
4142 }
4143 
4144 /*
4145  * Show free area list (used inside shift_scroll-lock stuff)
4146  * We also calculate the percentage fragmentation. We do this by counting the
4147  * memory on each free list with the exception of the first item on the list.
4148  *
4149  * Bits in @filter:
4150  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4151  *   cpuset.
4152  */
4153 void show_free_areas(unsigned int filter)
4154 {
4155         unsigned long free_pcp = 0;
4156         int cpu;
4157         struct zone *zone;
4158         pg_data_t *pgdat;
4159 
4160         for_each_populated_zone(zone) {
4161                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
4162                         continue;
4163 
4164                 for_each_online_cpu(cpu)
4165                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4166         }
4167 
4168         printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4169                 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4170                 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4171                 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4172                 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4173                 " free:%lu free_pcp:%lu free_cma:%lu\n",
4174                 global_node_page_state(NR_ACTIVE_ANON),
4175                 global_node_page_state(NR_INACTIVE_ANON),
4176                 global_node_page_state(NR_ISOLATED_ANON),
4177                 global_node_page_state(NR_ACTIVE_FILE),
4178                 global_node_page_state(NR_INACTIVE_FILE),
4179                 global_node_page_state(NR_ISOLATED_FILE),
4180                 global_node_page_state(NR_UNEVICTABLE),
4181                 global_node_page_state(NR_FILE_DIRTY),
4182                 global_node_page_state(NR_WRITEBACK),
4183                 global_node_page_state(NR_UNSTABLE_NFS),
4184                 global_page_state(NR_SLAB_RECLAIMABLE),
4185                 global_page_state(NR_SLAB_UNRECLAIMABLE),
4186                 global_node_page_state(NR_FILE_MAPPED),
4187                 global_node_page_state(NR_SHMEM),
4188                 global_page_state(NR_PAGETABLE),
4189                 global_page_state(NR_BOUNCE),
4190                 global_page_state(NR_FREE_PAGES),
4191                 free_pcp,
4192                 global_page_state(NR_FREE_CMA_PAGES));
4193 
4194         for_each_online_pgdat(pgdat) {
4195                 printk("Node %d"
4196                         " active_anon:%lukB"
4197                         " inactive_anon:%lukB"
4198                         " active_file:%lukB"
4199                         " inactive_file:%lukB"
4200                         " unevictable:%lukB"
4201                         " isolated(anon):%lukB"
4202                         " isolated(file):%lukB"
4203                         " mapped:%lukB"
4204                         " dirty:%lukB"
4205                         " writeback:%lukB"
4206                         " shmem:%lukB"
4207 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4208                         " shmem_thp: %lukB"
4209                         " shmem_pmdmapped: %lukB"
4210                         " anon_thp: %lukB"
4211 #endif
4212                         " writeback_tmp:%lukB"
4213                         " unstable:%lukB"
4214                         " pages_scanned:%lu"
4215                         " all_unreclaimable? %s"
4216                         "\n",
4217                         pgdat->node_id,
4218                         K(node_page_state(pgdat, NR_ACTIVE_ANON)),
4219                         K(node_page_state(pgdat, NR_INACTIVE_ANON)),
4220                         K(node_page_state(pgdat, NR_ACTIVE_FILE)),
4221                         K(node_page_state(pgdat, NR_INACTIVE_FILE)),
4222                         K(node_page_state(pgdat, NR_UNEVICTABLE)),
4223                         K(node_page_state(pgdat, NR_ISOLATED_ANON)),
4224                         K(node_page_state(pgdat, NR_ISOLATED_FILE)),
4225                         K(node_page_state(pgdat, NR_FILE_MAPPED)),
4226                         K(node_page_state(pgdat, NR_FILE_DIRTY)),
4227                         K(node_page_state(pgdat, NR_WRITEBACK)),
4228 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4229                         K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
4230                         K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
4231                                         * HPAGE_PMD_NR),
4232                         K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
4233 #endif
4234                         K(node_page_state(pgdat, NR_SHMEM)),
4235                         K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
4236                         K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
4237                         node_page_state(pgdat, NR_PAGES_SCANNED),
4238                         !pgdat_reclaimable(pgdat) ? "yes" : "no");
4239         }
4240 
4241         for_each_populated_zone(zone) {
4242                 int i;
4243 
4244                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
4245                         continue;
4246 
4247                 free_pcp = 0;
4248                 for_each_online_cpu(cpu)
4249                         free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4250 
4251                 show_node(zone);
4252                 printk("%s"
4253                         " free:%lukB"
4254                         " min:%lukB"
4255                         " low:%lukB"
4256                         " high:%lukB"
4257                         " active_anon:%lukB"
4258                         " inactive_anon:%lukB"
4259                         " active_file:%lukB"
4260                         " inactive_file:%lukB"
4261                         " unevictable:%lukB"
4262                         " writepending:%lukB"
4263                         " present:%lukB"
4264                         " managed:%lukB"
4265                         " mlocked:%lukB"
4266                         " slab_reclaimable:%lukB"
4267                         " slab_unreclaimable:%lukB"
4268                         " kernel_stack:%lukB"
4269                         " pagetables:%lukB"
4270                         " bounce:%lukB"
4271                         " free_pcp:%lukB"
4272                         " local_pcp:%ukB"
4273                         " free_cma:%lukB"
4274                         "\n",
4275                         zone->name,
4276                         K(zone_page_state(zone, NR_FREE_PAGES)),
4277                         K(min_wmark_pages(zone)),
4278                         K(low_wmark_pages(zone)),
4279                         K(high_wmark_pages(zone)),
4280                         K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
4281                         K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
4282                         K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
4283                         K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
4284                         K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
4285                         K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
4286                         K(zone->present_pages),
4287                         K(zone->managed_pages),
4288                         K(zone_page_state(zone, NR_MLOCK)),
4289                         K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
4290                         K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
4291                         zone_page_state(zone, NR_KERNEL_STACK_KB),
4292                         K(zone_page_state(zone, NR_PAGETABLE)),
4293                         K(zone_page_state(zone, NR_BOUNCE)),
4294                         K(free_pcp),
4295                         K(this_cpu_read(zone->pageset->pcp.count)),
4296                         K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
4297                 printk("lowmem_reserve[]:");
4298                 for (i = 0; i < MAX_NR_ZONES; i++)
4299                         printk(" %ld", zone->lowmem_reserve[i]);
4300                 printk("\n");
4301         }
4302 
4303         for_each_populated_zone(zone) {
4304                 unsigned int order;
4305                 unsigned long nr[MAX_ORDER], flags, total = 0;
4306                 unsigned char types[MAX_ORDER];
4307 
4308                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
4309                         continue;
4310                 show_node(zone);
4311                 printk("%s: ", zone->name);
4312 
4313                 spin_lock_irqsave(&zone->lock, flags);
4314                 for (order = 0; order < MAX_ORDER; order++) {
4315                         struct free_area *area = &zone->free_area[order];
4316                         int type;
4317 
4318                         nr[order] = area->nr_free;
4319                         total += nr[order] << order;
4320 
4321                         types[order] = 0;
4322                         for (type = 0; type < MIGRATE_TYPES; type++) {
4323                                 if (!list_empty(&area->free_list[type]))
4324                                         types[order] |= 1 << type;
4325                         }
4326                 }
4327                 spin_unlock_irqrestore(&zone->lock, flags);
4328                 for (order = 0; order < MAX_ORDER; order++) {
4329                         printk("%lu*%lukB ", nr[order], K(1UL) << order);
4330                         if (nr[order])
4331                                 show_migration_types(types[order]);
4332                 }
4333                 printk("= %lukB\n", K(total));
4334         }
4335 
4336         hugetlb_show_meminfo();
4337 
4338         printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
4339 
4340         show_swap_cache_info();
4341 }
4342 
4343 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4344 {
4345         zoneref->zone = zone;
4346         zoneref->zone_idx = zone_idx(zone);
4347 }
4348 
4349 /*
4350  * Builds allocation fallback zone lists.
4351  *
4352  * Add all populated zones of a node to the zonelist.
4353  */
4354 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
4355                                 int nr_zones)
4356 {
4357         struct zone *zone;
4358         enum zone_type zone_type = MAX_NR_ZONES;
4359 
4360         do {
4361                 zone_type--;
4362                 zone = pgdat->node_zones + zone_type;
4363                 if (managed_zone(zone)) {
4364                         zoneref_set_zone(zone,
4365                                 &zonelist->_zonerefs[nr_zones++]);
4366                         check_highest_zone(zone_type);
4367                 }
4368         } while (zone_type);
4369 
4370         return nr_zones;
4371 }
4372 
4373 
4374 /*
4375  *  zonelist_order:
4376  *  0 = automatic detection of better ordering.
4377  *  1 = order by ([node] distance, -zonetype)
4378  *  2 = order by (-zonetype, [node] distance)
4379  *
4380  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4381  *  the same zonelist. So only NUMA can configure this param.
4382  */
4383 #define ZONELIST_ORDER_DEFAULT  0
4384 #define ZONELIST_ORDER_NODE     1
4385 #define ZONELIST_ORDER_ZONE     2
4386 
4387 /* zonelist order in the kernel.
4388  * set_zonelist_order() will set this to NODE or ZONE.
4389  */
4390 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4391 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4392 
4393 
4394 #ifdef CONFIG_NUMA
4395 /* The value user specified ....changed by config */
4396 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4397 /* string for sysctl */
4398 #define NUMA_ZONELIST_ORDER_LEN 16
4399 char numa_zonelist_order[16] = "default";
4400 
4401 /*
4402  * interface for configure zonelist ordering.
4403  * command line option "numa_zonelist_order"
4404  *      = "[dD]efault   - default, automatic configuration.
4405  *      = "[nN]ode      - order by node locality, then by zone within node
4406  *      = "[zZ]one      - order by zone, then by locality within zone
4407  */
4408 
4409 static int __parse_numa_zonelist_order(char *s)
4410 {
4411         if (*s == 'd' || *s == 'D') {
4412                 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4413         } else if (*s == 'n' || *s == 'N') {
4414                 user_zonelist_order = ZONELIST_ORDER_NODE;
4415         } else if (*s == 'z' || *s == 'Z') {
4416                 user_zonelist_order = ZONELIST_ORDER_ZONE;
4417         } else {
4418                 pr_warn("Ignoring invalid numa_zonelist_order value:  %s\n", s);
4419                 return -EINVAL;
4420         }
4421         return 0;
4422 }
4423 
4424 static __init int setup_numa_zonelist_order(char *s)
4425 {
4426         int ret;
4427 
4428         if (!s)
4429                 return 0;
4430 
4431         ret = __parse_numa_zonelist_order(s);
4432         if (ret == 0)
4433                 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4434 
4435         return ret;
4436 }
4437 early_param("numa_zonelist_order", setup_numa_zonelist_order);
4438 
4439 /*
4440  * sysctl handler for numa_zonelist_order
4441  */
4442 int numa_zonelist_order_handler(struct ctl_table *table, int write,
4443                 void __user *buffer, size_t *length,
4444                 loff_t *ppos)
4445 {
4446         char saved_string[NUMA_ZONELIST_ORDER_LEN];
4447         int ret;
4448         static DEFINE_MUTEX(zl_order_mutex);
4449 
4450         mutex_lock(&zl_order_mutex);
4451         if (write) {
4452                 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4453                         ret = -EINVAL;
4454                         goto out;
4455                 }
4456                 strcpy(saved_string, (char *)table->data);
4457         }
4458         ret = proc_dostring(table, write, buffer, length, ppos);
4459         if (ret)
4460                 goto out;
4461         if (write) {
4462                 int oldval = user_zonelist_order;
4463 
4464                 ret = __parse_numa_zonelist_order((char *)table->data);
4465                 if (ret) {
4466                         /*
4467                          * bogus value.  restore saved string
4468                          */
4469                         strncpy((char *)table->data, saved_string,
4470                                 NUMA_ZONELIST_ORDER_LEN);
4471                         user_zonelist_order = oldval;
4472                 } else if (oldval != user_zonelist_order) {
4473                         mutex_lock(&zonelists_mutex);
4474                         build_all_zonelists(NULL, NULL);
4475                         mutex_unlock(&zonelists_mutex);
4476                 }
4477         }
4478 out:
4479         mutex_unlock(&zl_order_mutex);
4480         return ret;
4481 }
4482 
4483 
4484 #define MAX_NODE_LOAD (nr_online_nodes)
4485 static int node_load[MAX_NUMNODES];
4486 
4487 /**
4488  * find_next_best_node - find the next node that should appear in a given node's fallback list
4489  * @node: node whose fallback list we're appending
4490  * @used_node_mask: nodemask_t of already used nodes
4491  *
4492  * We use a number of factors to determine which is the next node that should
4493  * appear on a given node's fallback list.  The node should not have appeared
4494  * already in @node's fallback list, and it should be the next closest node
4495  * according to the distance array (which contains arbitrary distance values
4496  * from each node to each node in the system), and should also prefer nodes
4497  * with no CPUs, since presumably they'll have very little allocation pressure
4498  * on them otherwise.
4499  * It returns -1 if no node is found.
4500  */
4501 static int find_next_best_node(int node, nodemask_t *used_node_mask)
4502 {
4503         int n, val;
4504         int min_val = INT_MAX;
4505         int best_node = NUMA_NO_NODE;
4506         const struct cpumask *tmp = cpumask_of_node(0);
4507 
4508         /* Use the local node if we haven't already */
4509         if (!node_isset(node, *used_node_mask)) {
4510                 node_set(node, *used_node_mask);
4511                 return node;
4512         }
4513 
4514         for_each_node_state(n, N_MEMORY) {
4515 
4516                 /* Don't want a node to appear more than once */
4517                 if (node_isset(n, *used_node_mask))
4518                         continue;
4519 
4520                 /* Use the distance array to find the distance */
4521                 val = node_distance(node, n);
4522 
4523                 /* Penalize nodes under us ("prefer the next node") */
4524                 val += (n < node);
4525 
4526                 /* Give preference to headless and unused nodes */
4527                 tmp = cpumask_of_node(n);
4528                 if (!cpumask_empty(tmp))
4529                         val += PENALTY_FOR_NODE_WITH_CPUS;
4530 
4531                 /* Slight preference for less loaded node */
4532                 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4533                 val += node_load[n];
4534 
4535                 if (val < min_val) {
4536                         min_val = val;
4537                         best_node = n;
4538                 }
4539         }
4540 
4541         if (best_node >= 0)
4542                 node_set(best_node, *used_node_mask);
4543 
4544         return best_node;
4545 }
4546 
4547 
4548 /*
4549  * Build zonelists ordered by node and zones within node.
4550  * This results in maximum locality--normal zone overflows into local
4551  * DMA zone, if any--but risks exhausting DMA zone.
4552  */
4553 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
4554 {
4555         int j;
4556         struct zonelist *zonelist;
4557 
4558         zonelist = &pgdat->node_zonelists[0];
4559         for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
4560                 ;
4561         j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4562         zonelist->_zonerefs[j].zone = NULL;
4563         zonelist->_zonerefs[j].zone_idx = 0;
4564 }
4565 
4566 /*
4567  * Build gfp_thisnode zonelists
4568  */
4569 static void build_thisnode_zonelists(pg_data_t *pgdat)
4570 {
4571         int j;
4572         struct zonelist *zonelist;
4573 
4574         zonelist = &pgdat->node_zonelists[1];
4575         j = build_zonelists_node(pgdat, zonelist, 0);
4576         zonelist->_zonerefs[j].zone = NULL;
4577         zonelist->_zonerefs[j].zone_idx = 0;
4578 }
4579 
4580 /*
4581  * Build zonelists ordered by zone and nodes within zones.
4582  * This results in conserving DMA zone[s] until all Normal memory is
4583  * exhausted, but results in overflowing to remote node while memory
4584  * may still exist in local DMA zone.
4585  */
4586 static int node_order[MAX_NUMNODES];
4587 
4588 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4589 {
4590         int pos, j, node;
4591         int zone_type;          /* needs to be signed */
4592         struct zone *z;
4593         struct zonelist *zonelist;
4594 
4595         zonelist = &pgdat->node_zonelists[0];
4596         pos = 0;
4597         for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4598                 for (j = 0; j < nr_nodes; j++) {
4599                         node = node_order[j];
4600                         z = &NODE_DATA(node)->node_zones[zone_type];
4601                         if (managed_zone(z)) {
4602                                 zoneref_set_zone(z,
4603                                         &zonelist->_zonerefs[pos++]);
4604                                 check_highest_zone(zone_type);
4605                         }
4606                 }
4607         }
4608         zonelist->_zonerefs[pos].zone = NULL;
4609         zonelist->_zonerefs[pos].zone_idx = 0;
4610 }
4611 
4612 #if defined(CONFIG_64BIT)
4613 /*
4614  * Devices that require DMA32/DMA are relatively rare and do not justify a
4615  * penalty to every machine in case the specialised case applies. Default
4616  * to Node-ordering on 64-bit NUMA machines
4617  */
4618 static int default_zonelist_order(void)
4619 {
4620         return ZONELIST_ORDER_NODE;
4621 }
4622 #else
4623 /*
4624  * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4625  * by the kernel. If processes running on node 0 deplete the low memory zone
4626  * then reclaim will occur more frequency increasing stalls and potentially
4627  * be easier to OOM if a large percentage of the zone is under writeback or
4628  * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4629  * Hence, default to zone ordering on 32-bit.
4630  */
4631 static int default_zonelist_order(void)
4632 {
4633         return ZONELIST_ORDER_ZONE;
4634 }
4635 #endif /* CONFIG_64BIT */
4636 
4637 static void set_zonelist_order(void)
4638 {
4639         if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4640                 current_zonelist_order = default_zonelist_order();
4641         else
4642                 current_zonelist_order = user_zonelist_order;
4643 }
4644 
4645 static void build_zonelists(pg_data_t *pgdat)
4646 {
4647         int i, node, load;
4648         nodemask_t used_mask;
4649         int local_node, prev_node;
4650         struct zonelist *zonelist;
4651         unsigned int order = current_zonelist_order;
4652 
4653         /* initialize zonelists */
4654         for (i = 0; i < MAX_ZONELISTS; i++) {
4655                 zonelist = pgdat->node_zonelists + i;
4656                 zonelist->_zonerefs[0].zone = NULL;
4657                 zonelist->_zonerefs[0].zone_idx = 0;
4658         }
4659 
4660         /* NUMA-aware ordering of nodes */
4661         local_node = pgdat->node_id;
4662         load = nr_online_nodes;
4663         prev_node = local_node;
4664         nodes_clear(used_mask);
4665 
4666         memset(node_order, 0, sizeof(node_order));
4667         i = 0;
4668 
4669         while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4670                 /*
4671                  * We don't want to pressure a particular node.
4672                  * So adding penalty to the first node in same
4673                  * distance group to make it round-robin.
4674                  */
4675                 if (node_distance(local_node, node) !=
4676                     node_distance(local_node, prev_node))
4677                         node_load[node] = load;
4678 
4679                 prev_node = node;
4680                 load--;
4681                 if (order == ZONELIST_ORDER_NODE)
4682                         build_zonelists_in_node_order(pgdat, node);
4683                 else
4684                         node_order[i++] = node; /* remember order */
4685         }
4686 
4687         if (order == ZONELIST_ORDER_ZONE) {
4688                 /* calculate node order -- i.e., DMA last! */
4689                 build_zonelists_in_zone_order(pgdat, i);
4690         }
4691 
4692         build_thisnode_zonelists(pgdat);
4693 }
4694 
4695 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4696 /*
4697  * Return node id of node used for "local" allocations.
4698  * I.e., first node id of first zone in arg node's generic zonelist.
4699  * Used for initializing percpu 'numa_mem', which is used primarily
4700  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4701  */
4702 int local_memory_node(int node)
4703 {
4704         struct zoneref *z;
4705 
4706         z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4707                                    gfp_zone(GFP_KERNEL),
4708                                    NULL);
4709         return z->zone->node;
4710 }
4711 #endif
4712 
4713 static void setup_min_unmapped_ratio(void);
4714 static void setup_min_slab_ratio(void);
4715 #else   /* CONFIG_NUMA */
4716 
4717 static void set_zonelist_order(void)
4718 {
4719         current_zonelist_order = ZONELIST_ORDER_ZONE;
4720 }
4721 
4722 static void build_zonelists(pg_data_t *pgdat)
4723 {
4724         int node, local_node;
4725         enum zone_type j;
4726         struct zonelist *zonelist;
4727 
4728         local_node = pgdat->node_id;
4729 
4730         zonelist = &pgdat->node_zonelists[0];
4731         j = build_zonelists_node(pgdat, zonelist, 0);
4732 
4733         /*
4734          * Now we build the zonelist so that it contains the zones
4735          * of all the other nodes.
4736          * We don't want to pressure a particular node, so when
4737          * building the zones for node N, we make sure that the
4738          * zones coming right after the local ones are those from
4739          * node N+1 (modulo N)
4740          */
4741         for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4742                 if (!node_online(node))
4743                         continue;
4744                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4745         }
4746         for (node = 0; node < local_node; node++) {
4747                 if (!node_online(node))
4748                         continue;
4749                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4750         }
4751 
4752         zonelist->_zonerefs[j].zone = NULL;
4753         zonelist->_zonerefs[j].zone_idx = 0;
4754 }
4755 
4756 #endif  /* CONFIG_NUMA */
4757 
4758 /*
4759  * Boot pageset table. One per cpu which is going to be used for all
4760  * zones and all nodes. The parameters will be set in such a way
4761  * that an item put on a list will immediately be handed over to
4762  * the buddy list. This is safe since pageset manipulation is done
4763  * with interrupts disabled.
4764  *
4765  * The boot_pagesets must be kept even after bootup is complete for
4766  * unused processors and/or zones. They do play a role for bootstrapping
4767  * hotplugged processors.
4768  *
4769  * zoneinfo_show() and maybe other functions do
4770  * not check if the processor is online before following the pageset pointer.
4771  * Other parts of the kernel may not check if the zone is available.
4772  */
4773 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4774 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
4775 static void setup_zone_pageset(struct zone *zone);
4776 
4777 /*
4778  * Global mutex to protect against size modification of zonelists
4779  * as well as to serialize pageset setup for the new populated zone.
4780  */
4781 DEFINE_MUTEX(zonelists_mutex);
4782 
4783 /* return values int ....just for stop_machine() */
4784 static int __build_all_zonelists(void *data)
4785 {
4786         int nid;
4787         int cpu;
4788         pg_data_t *self = data;
4789 
4790 #ifdef CONFIG_NUMA
4791         memset(node_load, 0, sizeof(node_load));
4792 #endif
4793 
4794         if (self && !node_online(self->node_id)) {
4795                 build_zonelists(self);
4796         }
4797 
4798         for_each_online_node(nid) {
4799                 pg_data_t *pgdat = NODE_DATA(nid);
4800 
4801                 build_zonelists(pgdat);
4802         }
4803 
4804         /*
4805          * Initialize the boot_pagesets that are going to be used
4806          * for bootstrapping processors. The real pagesets for
4807          * each zone will be allocated later when the per cpu
4808          * allocator is available.
4809          *
4810          * boot_pagesets are used also for bootstrapping offline
4811          * cpus if the system is already booted because the pagesets
4812          * are needed to initialize allocators on a specific cpu too.
4813          * F.e. the percpu allocator needs the page allocator which
4814          * needs the percpu allocator in order to allocate its pagesets
4815          * (a chicken-egg dilemma).
4816          */
4817         for_each_possible_cpu(cpu) {
4818                 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4819 
4820 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4821                 /*
4822                  * We now know the "local memory node" for each node--
4823                  * i.e., the node of the first zone in the generic zonelist.
4824                  * Set up numa_mem percpu variable for on-line cpus.  During
4825                  * boot, only the boot cpu should be on-line;  we'll init the
4826                  * secondary cpus' numa_mem as they come on-line.  During
4827                  * node/memory hotplug, we'll fixup all on-line cpus.
4828                  */
4829                 if (cpu_online(cpu))
4830                         set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4831 #endif
4832         }
4833 
4834         return 0;
4835 }
4836 
4837 static noinline void __init
4838 build_all_zonelists_init(void)
4839 {
4840         __build_all_zonelists(NULL);
4841         mminit_verify_zonelist();
4842         cpuset_init_current_mems_allowed();
4843 }
4844 
4845 /*
4846  * Called with zonelists_mutex held always
4847  * unless system_state == SYSTEM_BOOTING.
4848  *
4849  * __ref due to (1) call of __meminit annotated setup_zone_pageset
4850  * [we're only called with non-NULL zone through __meminit paths] and
4851  * (2) call of __init annotated helper build_all_zonelists_init
4852  * [protected by SYSTEM_BOOTING].
4853  */
4854 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
4855 {
4856         set_zonelist_order();
4857 
4858         if (system_state == SYSTEM_BOOTING) {
4859                 build_all_zonelists_init();
4860         } else {
4861 #ifdef CONFIG_MEMORY_HOTPLUG
4862                 if (zone)
4863                         setup_zone_pageset(zone);
4864 #endif
4865                 /* we have to stop all cpus to guarantee there is no user
4866                    of zonelist */
4867                 stop_machine(__build_all_zonelists, pgdat, NULL);
4868                 /* cpuset refresh routine should be here */
4869         }
4870         vm_total_pages = nr_free_pagecache_pages();
4871         /*
4872          * Disable grouping by mobility if the number of pages in the
4873          * system is too low to allow the mechanism to work. It would be
4874          * more accurate, but expensive to check per-zone. This check is
4875          * made on memory-hotadd so a system can start with mobility
4876          * disabled and enable it later
4877          */
4878         if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
4879                 page_group_by_mobility_disabled = 1;
4880         else
4881                 page_group_by_mobility_disabled = 0;
4882 
4883         pr_info("Built %i zonelists in %s order, mobility grouping %s.  Total pages: %ld\n",
4884                 nr_online_nodes,
4885                 zonelist_order_name[current_zonelist_order],
4886                 page_group_by_mobility_disabled ? "off" : "on",
4887                 vm_total_pages);
4888 #ifdef CONFIG_NUMA
4889         pr_info("Policy zone: %s\n", zone_names[policy_zone]);
4890 #endif
4891 }
4892 
4893 /*
4894  * Helper functions to size the waitqueue hash table.
4895  * Essentially these want to choose hash table sizes sufficiently
4896  * large so that collisions trying to wait on pages are rare.
4897  * But in fact, the number of active page waitqueues on typical
4898  * systems is ridiculously low, less than 200. So this is even
4899  * conservative, even though it seems large.
4900  *
4901  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4902  * waitqueues, i.e. the size of the waitq table given the number of pages.
4903  */
4904 #define PAGES_PER_WAITQUEUE     256
4905 
4906 #ifndef CONFIG_MEMORY_HOTPLUG
4907 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4908 {
4909         unsigned long size = 1;
4910 
4911         pages /= PAGES_PER_WAITQUEUE;
4912 
4913         while (size < pages)
4914                 size <<= 1;
4915 
4916         /*
4917          * Once we have dozens or even hundreds of threads sleeping
4918          * on IO we've got bigger problems than wait queue collision.
4919          * Limit the size of the wait table to a reasonable size.
4920          */
4921         size = min(size, 4096UL);
4922 
4923         return max(size, 4UL);
4924 }
4925 #else
4926 /*
4927  * A zone's size might be changed by hot-add, so it is not possible to determine
4928  * a suitable size for its wait_table.  So we use the maximum size now.
4929  *
4930  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
4931  *
4932  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
4933  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4934  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
4935  *
4936  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4937  * or more by the traditional way. (See above).  It equals:
4938  *
4939  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
4940  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
4941  *    powerpc (64K page size)             : =  (32G +16M)byte.
4942  */
4943 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4944 {
4945         return 4096UL;
4946 }
4947 #endif
4948 
4949 /*
4950  * This is an integer logarithm so that shifts can be used later
4951  * to extract the more random high bits from the multiplicative
4952  * hash function before the remainder is taken.
4953  */
4954 static inline unsigned long wait_table_bits(unsigned long size)
4955 {
4956         return ffz(~size);
4957 }
4958 
4959 /*
4960  * Initially all pages are reserved - free ones are freed
4961  * up by free_all_bootmem() once the early boot process is
4962  * done. Non-atomic initialization, single-pass.
4963  */
4964 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4965                 unsigned long start_pfn, enum memmap_context context)
4966 {
4967         struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
4968         unsigned long end_pfn = start_pfn + size;
4969         pg_data_t *pgdat = NODE_DATA(nid);
4970         unsigned long pfn;
4971         unsigned long nr_initialised = 0;
4972 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4973         struct memblock_region *r = NULL, *tmp;
4974 #endif
4975 
4976         if (highest_memmap_pfn < end_pfn - 1)
4977                 highest_memmap_pfn = end_pfn - 1;
4978 
4979         /*
4980          * Honor reservation requested by the driver for this ZONE_DEVICE
4981          * memory
4982          */
4983         if (altmap && start_pfn == altmap->base_pfn)
4984                 start_pfn += altmap->reserve;
4985 
4986         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4987                 /*
4988                  * There can be holes in boot-time mem_map[]s handed to this
4989                  * function.  They do not exist on hotplugged memory.
4990                  */
4991                 if (context != MEMMAP_EARLY)
4992                         goto not_early;
4993 
4994                 if (!early_pfn_valid(pfn))
4995                         continue;
4996                 if (!early_pfn_in_nid(pfn, nid))
4997                         continue;
4998                 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
4999                         break;
5000 
5001 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5002                 /*
5003                  * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
5004                  * from zone_movable_pfn[nid] to end of each node should be
5005                  * ZONE_MOVABLE not ZONE_NORMAL. skip it.
5006                  */
5007                 if (!mirrored_kernelcore && zone_movable_pfn[nid])
5008                         if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid])
5009                                 continue;
5010 
5011                 /*
5012                  * Check given memblock attribute by firmware which can affect
5013                  * kernel memory layout.  If zone==ZONE_MOVABLE but memory is
5014                  * mirrored, it's an overlapped memmap init. skip it.
5015                  */
5016                 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5017                         if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
5018                                 for_each_memblock(memory, tmp)
5019                                         if (pfn < memblock_region_memory_end_pfn(tmp))
5020                                                 break;
5021                                 r = tmp;
5022                         }
5023                         if (pfn >= memblock_region_memory_base_pfn(r) &&
5024                             memblock_is_mirror(r)) {
5025                                 /* already initialized as NORMAL */
5026                                 pfn = memblock_region_memory_end_pfn(r);
5027                                 continue;
5028                         }
5029                 }
5030 #endif
5031 
5032 not_early:
5033                 /*
5034                  * Mark the block movable so that blocks are reserved for
5035                  * movable at startup. This will force kernel allocations
5036                  * to reserve their blocks rather than leaking throughout
5037                  * the address space during boot when many long-lived
5038                  * kernel allocations are made.
5039                  *
5040                  * bitmap is created for zone's valid pfn range. but memmap
5041                  * can be created for invalid pages (for alignment)
5042                  * check here not to call set_pageblock_migratetype() against
5043                  * pfn out of zone.
5044                  */
5045                 if (!(pfn & (pageblock_nr_pages - 1))) {
5046                         struct page *page = pfn_to_page(pfn);
5047 
5048                         __init_single_page(page, pfn, zone, nid);
5049                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5050                 } else {
5051                         __init_single_pfn(pfn, zone, nid);
5052                 }
5053         }
5054 }
5055 
5056 static void __meminit zone_init_free_lists(struct zone *zone)
5057 {
5058         unsigned int order, t;
5059         for_each_migratetype_order(order, t) {
5060                 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
5061                 zone->free_area[order].nr_free = 0;
5062         }
5063 }
5064 
5065 #ifndef __HAVE_ARCH_MEMMAP_INIT
5066 #define memmap_init(size, nid, zone, start_pfn) \
5067         memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
5068 #endif
5069 
5070 static int zone_batchsize(struct zone *zone)
5071 {
5072 #ifdef CONFIG_MMU
5073         int batch;
5074 
5075         /*
5076          * The per-cpu-pages pools are set to around 1000th of the
5077          * size of the zone.  But no more than 1/2 of a meg.
5078          *
5079          * OK, so we don't know how big the cache is.  So guess.
5080          */
5081         batch = zone->managed_pages / 1024;
5082         if (batch * PAGE_SIZE > 512 * 1024)
5083                 batch = (512 * 1024) / PAGE_SIZE;
5084         batch /= 4;             /* We effectively *= 4 below */
5085         if (batch < 1)
5086                 batch = 1;
5087 
5088         /*
5089          * Clamp the batch to a 2^n - 1 value. Having a power
5090          * of 2 value was found to be more likely to have
5091          * suboptimal cache aliasing properties in some cases.
5092          *
5093          * For example if 2 tasks are alternately allocating
5094          * batches of pages, one task can end up with a lot
5095          * of pages of one half of the possible page colors
5096          * and the other with pages of the other colors.
5097          */
5098         batch = rounddown_pow_of_two(batch + batch/2) - 1;
5099 
5100         return batch;
5101 
5102 #else
5103         /* The deferral and batching of frees should be suppressed under NOMMU
5104          * conditions.
5105          *
5106          * The problem is that NOMMU needs to be able to allocate large chunks
5107          * of contiguous memory as there's no hardware page translation to
5108          * assemble apparent contiguous memory from discontiguous pages.
5109          *
5110          * Queueing large contiguous runs of pages for batching, however,
5111          * causes the pages to actually be freed in smaller chunks.  As there
5112          * can be a significant delay between the individual batches being
5113          * recycled, this leads to the once large chunks of space being
5114          * fragmented and becoming unavailable for high-order allocations.
5115          */
5116         return 0;
5117 #endif
5118 }
5119 
5120 /*
5121  * pcp->high and pcp->batch values are related and dependent on one another:
5122  * ->batch must never be higher then ->high.
5123  * The following function updates them in a safe manner without read side
5124  * locking.
5125  *
5126  * Any new users of pcp->batch and pcp->high should ensure they can cope with
5127  * those fields changing asynchronously (acording the the above rule).
5128  *
5129  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5130  * outside of boot time (or some other assurance that no concurrent updaters
5131  * exist).
5132  */
5133 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5134                 unsigned long batch)
5135 {
5136        /* start with a fail safe value for batch */
5137         pcp->batch = 1;
5138         smp_wmb();
5139 
5140        /* Update high, then batch, in order */
5141         pcp->high = high;
5142         smp_wmb();
5143 
5144         pcp->batch = batch;
5145 }
5146 
5147 /* a companion to pageset_set_high() */
5148 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5149 {
5150         pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
5151 }
5152 
5153 static void pageset_init(struct per_cpu_pageset *p)
5154 {
5155         struct per_cpu_pages *pcp;
5156         int migratetype;
5157 
5158         memset(p, 0, sizeof(*p));
5159 
5160         pcp = &p->pcp;
5161         pcp->count = 0;
5162         for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5163                 INIT_LIST_HEAD(&pcp->lists[migratetype]);
5164 }
5165 
5166 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5167 {
5168         pageset_init(p);
5169         pageset_set_batch(p, batch);
5170 }
5171 
5172 /*
5173  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5174  * to the value high for the pageset p.
5175  */
5176 static void pageset_set_high(struct per_cpu_pageset *p,
5177                                 unsigned long high)
5178 {
5179         unsigned long batch = max(1UL, high / 4);
5180         if ((high / 4) > (PAGE_SHIFT * 8))
5181                 batch = PAGE_SHIFT * 8;
5182 
5183         pageset_update(&p->pcp, high, batch);
5184 }
5185 
5186 static void pageset_set_high_and_batch(struct zone *zone,
5187                                        struct per_cpu_pageset *pcp)
5188 {
5189         if (percpu_pagelist_fraction)
5190                 pageset_set_high(pcp,
5191                         (zone->managed_pages /
5192                                 percpu_pagelist_fraction));
5193         else
5194                 pageset_set_batch(pcp, zone_batchsize(zone));
5195 }
5196 
5197 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5198 {
5199         struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5200 
5201         pageset_init(pcp);
5202         pageset_set_high_and_batch(zone, pcp);
5203 }
5204 
5205 static void __meminit setup_zone_pageset(struct zone *zone)
5206 {
5207         int cpu;
5208         zone->pageset = alloc_percpu(struct per_cpu_pageset);
5209         for_each_possible_cpu(cpu)
5210                 zone_pageset_init(zone, cpu);
5211 }
5212 
5213 /*
5214  * Allocate per cpu pagesets and initialize them.
5215  * Before this call only boot pagesets were available.
5216  */
5217 void __init setup_per_cpu_pageset(void)
5218 {
5219         struct pglist_data *pgdat;
5220         struct zone *zone;
5221 
5222         for_each_populated_zone(zone)
5223                 setup_zone_pageset(zone);
5224 
5225         for_each_online_pgdat(pgdat)
5226                 pgdat->per_cpu_nodestats =
5227                         alloc_percpu(struct per_cpu_nodestat);
5228 }
5229 
5230 static noinline __ref
5231 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
5232 {
5233         int i;
5234         size_t alloc_size;
5235 
5236         /*
5237          * The per-page waitqueue mechanism uses hashed waitqueues
5238          * per zone.
5239          */
5240         zone->wait_table_hash_nr_entries =
5241                  wait_table_hash_nr_entries(zone_size_pages);
5242         zone->wait_table_bits =
5243                 wait_table_bits(zone->wait_table_hash_nr_entries);
5244         alloc_size = zone->wait_table_hash_nr_entries
5245                                         * sizeof(wait_queue_head_t);
5246 
5247         if (!slab_is_available()) {
5248                 zone->wait_table = (wait_queue_head_t *)
5249                         memblock_virt_alloc_node_nopanic(
5250                                 alloc_size, zone->zone_pgdat->node_id);
5251         } else {
5252                 /*
5253                  * This case means that a zone whose size was 0 gets new memory
5254                  * via memory hot-add.
5255                  * But it may be the case that a new node was hot-added.  In
5256                  * this case vmalloc() will not be able to use this new node's
5257                  * memory - this wait_table must be initialized to use this new
5258                  * node itself as well.
5259                  * To use this new node's memory, further consideration will be
5260                  * necessary.
5261                  */
5262                 zone->wait_table = vmalloc(alloc_size);
5263         }
5264         if (!zone->wait_table)
5265                 return -ENOMEM;
5266 
5267         for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
5268                 init_waitqueue_head(zone->wait_table + i);
5269 
5270         return 0;
5271 }
5272 
5273 static __meminit void zone_pcp_init(struct zone *zone)
5274 {
5275         /*
5276          * per cpu subsystem is not up at this point. The following code
5277          * relies on the ability of the linker to provide the
5278          * offset of a (static) per cpu variable into the per cpu area.
5279          */
5280         zone->pageset = &boot_pageset;
5281 
5282         if (populated_zone(zone))
5283                 printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
5284                         zone->name, zone->present_pages,
5285                                          zone_batchsize(zone));
5286 }
5287 
5288 int __meminit init_currently_empty_zone(struct zone *zone,
5289                                         unsigned long zone_start_pfn,
5290                                         unsigned long size)
5291 {
5292         struct pglist_data *pgdat = zone->zone_pgdat;
5293         int ret;
5294         ret = zone_wait_table_init(zone, size);
5295         if (ret)
5296                 return ret;
5297         pgdat->nr_zones = zone_idx(zone) + 1;
5298 
5299         zone->zone_start_pfn = zone_start_pfn;
5300 
5301         mminit_dprintk(MMINIT_TRACE, "memmap_init",
5302                         "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5303                         pgdat->node_id,
5304                         (unsigned long)zone_idx(zone),
5305                         zone_start_pfn, (zone_start_pfn + size));
5306 
5307         zone_init_free_lists(zone);
5308 
5309         return 0;
5310 }
5311 
5312 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5313 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5314 
5315 /*
5316  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5317  */
5318 int __meminit __early_pfn_to_nid(unsigned long pfn,
5319                                         struct mminit_pfnnid_cache *state)
5320 {
5321         unsigned long start_pfn, end_pfn;
5322         int nid;
5323 
5324         if (state->last_start <= pfn && pfn < state->last_end)
5325                 return state->last_nid;
5326 
5327         nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5328         if (nid != -1) {
5329                 state->last_start = start_pfn;
5330                 state->last_end = end_pfn;
5331                 state->last_nid = nid;
5332         }
5333 
5334         return nid;
5335 }
5336 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5337 
5338 /**
5339  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5340  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5341  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5342  *
5343  * If an architecture guarantees that all ranges registered contain no holes
5344  * and may be freed, this this function may be used instead of calling
5345  * memblock_free_early_nid() manually.
5346  */
5347 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
5348 {
5349         unsigned long start_pfn, end_pfn;
5350         int i, this_nid;
5351 
5352         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5353                 start_pfn = min(start_pfn, max_low_pfn);
5354                 end_pfn = min(end_pfn, max_low_pfn);
5355 
5356                 if (start_pfn < end_pfn)
5357                         memblock_free_early_nid(PFN_PHYS(start_pfn),
5358                                         (end_pfn - start_pfn) << PAGE_SHIFT,
5359                                         this_nid);
5360         }
5361 }
5362 
5363 /**
5364  * sparse_memory_present_with_active_regions - Call memory_present for each active range
5365  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5366  *
5367  * If an architecture guarantees that all ranges registered contain no holes and may
5368  * be freed, this function may be used instead of calling memory_present() manually.
5369  */
5370 void __init sparse_memory_present_with_active_regions(int nid)
5371 {
5372         unsigned long start_pfn, end_pfn;
5373         int i, this_nid;
5374 
5375         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5376                 memory_present(this_nid, start_pfn, end_pfn);
5377 }
5378 
5379 /**
5380  * get_pfn_range_for_nid - Return the start and end page frames for a node
5381  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5382  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5383  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5384  *
5385  * It returns the start and end page frame of a node based on information
5386  * provided by memblock_set_node(). If called for a node
5387  * with no available memory, a warning is printed and the start and end
5388  * PFNs will be 0.
5389  */
5390 void __meminit get_pfn_range_for_nid(unsigned int nid,
5391                         unsigned long *start_pfn, unsigned long *end_pfn)
5392 {
5393         unsigned long this_start_pfn, this_end_pfn;
5394         int i;
5395 
5396         *start_pfn = -1UL;
5397         *end_pfn = 0;
5398 
5399         for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5400                 *start_pfn = min(*start_pfn, this_start_pfn);
5401                 *end_pfn = max(*end_pfn, this_end_pfn);
5402         }
5403 
5404         if (*start_pfn == -1UL)
5405                 *start_pfn = 0;
5406 }
5407 
5408 /*
5409  * This finds a zone that can be used for ZONE_MOVABLE pages. The
5410  * assumption is made that zones within a node are ordered in monotonic
5411  * increasing memory addresses so that the "highest" populated zone is used
5412  */
5413 static void __init find_usable_zone_for_movable(void)
5414 {
5415         int zone_index;
5416         for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5417                 if (zone_index == ZONE_MOVABLE)
5418                         continue;
5419 
5420                 if (arch_zone_highest_possible_pfn[zone_index] >
5421                                 arch_zone_lowest_possible_pfn[zone_index])
5422                         break;
5423         }
5424 
5425         VM_BUG_ON(zone_index == -1);
5426         movable_zone = zone_index;
5427 }
5428 
5429 /*
5430  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5431  * because it is sized independent of architecture. Unlike the other zones,
5432  * the starting point for ZONE_MOVABLE is not fixed. It may be different
5433  * in each node depending on the size of each node and how evenly kernelcore
5434  * is distributed. This helper function adjusts the zone ranges
5435  * provided by the architecture for a given node by using the end of the
5436  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5437  * zones within a node are in order of monotonic increases memory addresses
5438  */
5439 static void __meminit adjust_zone_range_for_zone_movable(int nid,
5440                                         unsigned long zone_type,
5441                                         unsigned long node_start_pfn,
5442                                         unsigned long node_end_pfn,
5443                                         unsigned long *zone_start_pfn,
5444                                         unsigned long *zone_end_pfn)
5445 {
5446         /* Only adjust if ZONE_MOVABLE is on this node */
5447         if (zone_movable_pfn[nid]) {
5448                 /* Size ZONE_MOVABLE */
5449                 if (zone_type == ZONE_MOVABLE) {
5450                         *zone_start_pfn = zone_movable_pfn[nid];
5451                         *zone_end_pfn = min(node_end_pfn,
5452                                 arch_zone_highest_possible_pfn[movable_zone]);
5453 
5454                 /* Check if this whole range is within ZONE_MOVABLE */
5455                 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5456                         *zone_start_pfn = *zone_end_pfn;
5457         }
5458 }
5459 
5460 /*
5461  * Return the number of pages a zone spans in a node, including holes
5462  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5463  */
5464 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
5465                                         unsigned long zone_type,
5466                                         unsigned long node_start_pfn,
5467                                         unsigned long node_end_pfn,
5468                                         unsigned long *zone_start_pfn,
5469                                         unsigned long *zone_end_pfn,
5470                                         unsigned long *ignored)
5471 {
5472         /* When hotadd a new node from cpu_up(), the node should be empty */
5473         if (!node_start_pfn && !node_end_pfn)
5474                 return 0;
5475 
5476         /* Get the start and end of the zone */
5477         *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5478         *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
5479         adjust_zone_range_for_zone_movable(nid, zone_type,
5480                                 node_start_pfn, node_end_pfn,
5481                                 zone_start_pfn, zone_end_pfn);
5482 
5483         /* Check that this node has pages within the zone's required range */
5484         if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
5485                 return 0;
5486 
5487         /* Move the zone boundaries inside the node if necessary */
5488         *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5489         *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
5490 
5491         /* Return the spanned pages */
5492         return *zone_end_pfn - *zone_start_pfn;
5493 }
5494 
5495 /*
5496  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5497  * then all holes in the requested range will be accounted for.
5498  */
5499 unsigned long __meminit __absent_pages_in_range(int nid,
5500                                 unsigned long range_start_pfn,
5501                                 unsigned long range_end_pfn)
5502 {
5503         unsigned long nr_absent = range_end_pfn - range_start_pfn;
5504         unsigned long start_pfn, end_pfn;
5505         int i;
5506 
5507         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5508                 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5509                 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5510                 nr_absent -= end_pfn - start_pfn;
5511         }
5512         return nr_absent;
5513 }
5514 
5515 /**
5516  * absent_pages_in_range - Return number of page frames in holes within a range
5517  * @start_pfn: The start PFN to start searching for holes
5518  * @end_pfn: The end PFN to stop searching for holes
5519  *
5520  * It returns the number of pages frames in memory holes within a range.
5521  */
5522 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5523                                                         unsigned long end_pfn)
5524 {
5525         return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5526 }
5527 
5528 /* Return the number of page frames in holes in a zone on a node */
5529 static unsigned long __meminit zone_absent_pages_in_node(int nid,
5530                                         unsigned long zone_type,
5531                                         unsigned long node_start_pfn,
5532                                         unsigned long node_end_pfn,
5533                                         unsigned long *ignored)
5534 {
5535         unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5536         unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
5537         unsigned long zone_start_pfn, zone_end_pfn;
5538         unsigned long nr_absent;
5539 
5540         /* When hotadd a new node from cpu_up(), the node should be empty */
5541         if (!node_start_pfn && !node_end_pfn)
5542                 return 0;
5543 
5544         zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5545         zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
5546 
5547         adjust_zone_range_for_zone_movable(nid, zone_type,
5548                         node_start_pfn, node_end_pfn,
5549                         &zone_start_pfn, &zone_end_pfn);
5550         nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5551 
5552         /*
5553          * ZONE_MOVABLE handling.
5554          * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5555          * and vice versa.
5556          */
5557         if (zone_movable_pfn[nid]) {
5558                 if (mirrored_kernelcore) {
5559                         unsigned long start_pfn, end_pfn;
5560                         struct memblock_region *r;
5561 
5562                         for_each_memblock(memory, r) {
5563                                 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5564                                                   zone_start_pfn, zone_end_pfn);
5565                                 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5566                                                 zone_start_pfn, zone_end_pfn);
5567 
5568                                 if (zone_type == ZONE_MOVABLE &&
5569                                     memblock_is_mirror(r))
5570                                         nr_absent += end_pfn - start_pfn;
5571 
5572                                 if (zone_type == ZONE_NORMAL &&
5573                                     !memblock_is_mirror(r))
5574                                         nr_absent += end_pfn - start_pfn;
5575                         }
5576                 } else {
5577                         if (zone_type == ZONE_NORMAL)
5578                                 nr_absent += node_end_pfn - zone_movable_pfn[nid];
5579                 }
5580         }
5581 
5582         return nr_absent;
5583 }
5584 
5585 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5586 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
5587                                         unsigned long zone_type,
5588                                         unsigned long node_start_pfn,
5589                                         unsigned long node_end_pfn,
5590                                         unsigned long *zone_start_pfn,
5591                                         unsigned long *zone_end_pfn,
5592                                         unsigned long *zones_size)
5593 {
5594         unsigned int zone;
5595 
5596         *zone_start_pfn = node_start_pfn;
5597         for (zone = 0; zone < zone_type; zone++)
5598                 *zone_start_pfn += zones_size[zone];
5599 
5600         *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5601 
5602         return zones_size[zone_type];
5603 }
5604 
5605 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
5606                                                 unsigned long zone_type,
5607                                                 unsigned long node_start_pfn,
5608                                                 unsigned long node_end_pfn,
5609                                                 unsigned long *zholes_size)
5610 {
5611         if (!zholes_size)
5612                 return 0;
5613 
5614         return zholes_size[zone_type];
5615 }
5616 
5617 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5618 
5619 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
5620                                                 unsigned long node_start_pfn,
5621                                                 unsigned long node_end_pfn,
5622                                                 unsigned long *zones_size,
5623                                                 unsigned long *zholes_size)
5624 {
5625         unsigned long realtotalpages = 0, totalpages = 0;
5626         enum zone_type i;
5627 
5628         for (i = 0; i < MAX_NR_ZONES; i++) {
5629                 struct zone *zone = pgdat->node_zones + i;
5630                 unsigned long zone_start_pfn, zone_end_pfn;
5631                 unsigned long size, real_size;
5632 
5633                 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5634                                                   node_start_pfn,
5635                                                   node_end_pfn,
5636                                                   &zone_start_pfn,
5637                                                   &zone_end_pfn,
5638                                                   zones_size);
5639                 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
5640                                                   node_start_pfn, node_end_pfn,
5641                                                   zholes_size);
5642                 if (size)
5643                         zone->zone_start_pfn = zone_start_pfn;
5644                 else
5645                         zone->zone_start_pfn = 0;
5646                 zone->spanned_pages = size;
5647                 zone->present_pages = real_size;
5648 
5649                 totalpages += size;
5650                 realtotalpages += real_size;
5651         }
5652 
5653         pgdat->node_spanned_pages = totalpages;
5654         pgdat->node_present_pages = realtotalpages;
5655         printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5656                                                         realtotalpages);
5657 }
5658 
5659 #ifndef CONFIG_SPARSEMEM
5660 /*
5661  * Calculate the size of the zone->blockflags rounded to an unsigned long
5662  * Start by making sure zonesize is a multiple of pageblock_order by rounding
5663  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5664  * round what is now in bits to nearest long in bits, then return it in
5665  * bytes.
5666  */
5667 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
5668 {
5669         unsigned long usemapsize;
5670 
5671         zonesize += zone_start_pfn & (pageblock_nr_pages-1);
5672         usemapsize = roundup(zonesize, pageblock_nr_pages);
5673         usemapsize = usemapsize >> pageblock_order;
5674         usemapsize *= NR_PAGEBLOCK_BITS;
5675         usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5676 
5677         return usemapsize / 8;
5678 }
5679 
5680 static void __init setup_usemap(struct pglist_data *pgdat,
5681                                 struct zone *zone,
5682                                 unsigned long zone_start_pfn,
5683                                 unsigned long zonesize)
5684 {
5685         unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
5686         zone->pageblock_flags = NULL;
5687         if (usemapsize)
5688                 zone->pageblock_flags =
5689                         memblock_virt_alloc_node_nopanic(usemapsize,
5690                                                          pgdat->node_id);
5691 }
5692 #else
5693 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5694                                 unsigned long zone_start_pfn, unsigned long zonesize) {}
5695 #endif /* CONFIG_SPARSEMEM */
5696 
5697 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5698 
5699 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5700 void __paginginit set_pageblock_order(void)
5701 {
5702         unsigned int order;
5703 
5704         /* Check that pageblock_nr_pages has not already been setup */
5705         if (pageblock_order)
5706                 return;
5707 
5708         if (HPAGE_SHIFT > PAGE_SHIFT)
5709                 order = HUGETLB_PAGE_ORDER;
5710         else
5711                 order = MAX_ORDER - 1;
5712 
5713         /*
5714          * Assume the largest contiguous order of interest is a huge page.
5715          * This value may be variable depending on boot parameters on IA64 and
5716          * powerpc.
5717          */
5718         pageblock_order = order;
5719 }
5720 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5721 
5722 /*
5723  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5724  * is unused as pageblock_order is set at compile-time. See
5725  * include/linux/pageblock-flags.h for the values of pageblock_order based on
5726  * the kernel config
5727  */
5728 void __paginginit set_pageblock_order(void)
5729 {
5730 }
5731 
5732 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5733 
5734 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5735                                                    unsigned long present_pages)
5736 {
5737         unsigned long pages = spanned_pages;
5738 
5739         /*
5740          * Provide a more accurate estimation if there are holes within
5741          * the zone and SPARSEMEM is in use. If there are holes within the
5742          * zone, each populated memory region may cost us one or two extra
5743          * memmap pages due to alignment because memmap pages for each
5744          * populated regions may not naturally algined on page boundary.
5745          * So the (present_pages >> 4) heuristic is a tradeoff for that.
5746          */
5747         if (spanned_pages > present_pages + (present_pages >> 4) &&
5748             IS_ENABLED(CONFIG_SPARSEMEM))
5749                 pages = present_pages;
5750 
5751         return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5752 }
5753 
5754 /*
5755  * Set up the zone data structures:
5756  *   - mark all pages reserved
5757  *   - mark all memory queues empty
5758  *   - clear the memory bitmaps
5759  *
5760  * NOTE: pgdat should get zeroed by caller.
5761  */
5762 static void __paginginit free_area_init_core(struct pglist_data *pgdat)
5763 {
5764         enum zone_type j;
5765         int nid = pgdat->node_id;
5766         int ret;
5767 
5768         pgdat_resize_init(pgdat);
5769 #ifdef CONFIG_NUMA_BALANCING
5770         spin_lock_init(&pgdat->numabalancing_migrate_lock);
5771         pgdat->numabalancing_migrate_nr_pages = 0;
5772         pgdat->numabalancing_migrate_next_window = jiffies;
5773 #endif
5774 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5775         spin_lock_init(&pgdat->split_queue_lock);
5776         INIT_LIST_HEAD(&pgdat->split_queue);
5777         pgdat->split_queue_len = 0;
5778 #endif
5779         init_waitqueue_head(&pgdat->kswapd_wait);
5780         init_waitqueue_head(&pgdat->pfmemalloc_wait);
5781 #ifdef CONFIG_COMPACTION
5782         init_waitqueue_head(&pgdat->kcompactd_wait);
5783 #endif
5784         pgdat_page_ext_init(pgdat);
5785         spin_lock_init(&pgdat->lru_lock);
5786         lruvec_init(node_lruvec(pgdat));
5787 
5788         for (j = 0; j < MAX_NR_ZONES; j++) {
5789                 struct zone *zone = pgdat->node_zones + j;
5790                 unsigned long size, realsize, freesize, memmap_pages;
5791                 unsigned long zone_start_pfn = zone->zone_start_pfn;
5792 
5793                 size = zone->spanned_pages;
5794                 realsize = freesize = zone->present_pages;
5795 
5796                 /*
5797                  * Adjust freesize so that it accounts for how much memory
5798                  * is used by this zone for memmap. This affects the watermark
5799                  * and per-cpu initialisations
5800                  */
5801                 memmap_pages = calc_memmap_size(size, realsize);
5802                 if (!is_highmem_idx(j)) {
5803                         if (freesize >= memmap_pages) {
5804                                 freesize -= memmap_pages;
5805                                 if (memmap_pages)
5806                                         printk(KERN_DEBUG
5807                                                "  %s zone: %lu pages used for memmap\n",
5808                                                zone_names[j], memmap_pages);
5809                         } else
5810                                 pr_warn("  %s zone: %lu pages exceeds freesize %lu\n",
5811                                         zone_names[j], memmap_pages, freesize);
5812                 }
5813 
5814                 /* Account for reserved pages */
5815                 if (j == 0 && freesize > dma_reserve) {
5816                         freesize -= dma_reserve;
5817                         printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
5818                                         zone_names[0], dma_reserve);
5819                 }
5820 
5821                 if (!is_highmem_idx(j))
5822                         nr_kernel_pages += freesize;
5823                 /* Charge for highmem memmap if there are enough kernel pages */
5824                 else if (nr_kernel_pages > memmap_pages * 2)
5825                         nr_kernel_pages -= memmap_pages;
5826                 nr_all_pages += freesize;
5827 
5828                 /*
5829                  * Set an approximate value for lowmem here, it will be adjusted
5830                  * when the bootmem allocator frees pages into the buddy system.
5831                  * And all highmem pages will be managed by the buddy system.
5832                  */
5833                 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
5834 #ifdef CONFIG_NUMA
5835                 zone->node = nid;
5836 #endif
5837                 zone->name = zone_names[j];
5838                 zone->zone_pgdat = pgdat;
5839                 spin_lock_init(&zone->lock);
5840                 zone_seqlock_init(zone);
5841                 zone_pcp_init(zone);
5842 
5843                 if (!size)
5844                         continue;
5845 
5846                 set_pageblock_order();
5847                 setup_usemap(pgdat, zone, zone_start_pfn, size);
5848                 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
5849                 BUG_ON(ret);
5850                 memmap_init(size, nid, j, zone_start_pfn);
5851         }
5852 }
5853 
5854 static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
5855 {
5856         unsigned long __maybe_unused start = 0;
5857         unsigned long __maybe_unused offset = 0;
5858 
5859         /* Skip empty nodes */
5860         if (!pgdat->node_spanned_pages)
5861                 return;
5862 
5863 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5864         start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5865         offset = pgdat->node_start_pfn - start;
5866         /* ia64 gets its own node_mem_map, before this, without bootmem */
5867         if (!pgdat->node_mem_map) {
5868                 unsigned long size, end;
5869                 struct page *map;
5870 
5871                 /*
5872                  * The zone's endpoints aren't required to be MAX_ORDER
5873                  * aligned but the node_mem_map endpoints must be in order
5874                  * for the buddy allocator to function correctly.
5875                  */
5876                 end = pgdat_end_pfn(pgdat);
5877                 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5878                 size =  (end - start) * sizeof(struct page);
5879                 map = alloc_remap(pgdat->node_id, size);
5880                 if (!map)
5881                         map = memblock_virt_alloc_node_nopanic(size,
5882                                                                pgdat->node_id);
5883                 pgdat->node_mem_map = map + offset;
5884         }
5885 #ifndef CONFIG_NEED_MULTIPLE_NODES
5886         /*
5887          * With no DISCONTIG, the global mem_map is just set as node 0's
5888          */
5889         if (pgdat == NODE_DATA(0)) {
5890                 mem_map = NODE_DATA(0)->node_mem_map;
5891 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5892                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
5893                         mem_map -= offset;
5894 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5895         }
5896 #endif
5897 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5898 }
5899 
5900 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5901                 unsigned long node_start_pfn, unsigned long *zholes_size)
5902 {
5903         pg_data_t *pgdat = NODE_DATA(nid);
5904         unsigned long start_pfn = 0;
5905         unsigned long end_pfn = 0;
5906 
5907         /* pg_data_t should be reset to zero when it's allocated */
5908         WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
5909 
5910         reset_deferred_meminit(pgdat);
5911         pgdat->node_id = nid;
5912         pgdat->node_start_pfn = node_start_pfn;
5913         pgdat->per_cpu_nodestats = NULL;
5914 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5915         get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
5916         pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5917                 (u64)start_pfn << PAGE_SHIFT,
5918                 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
5919 #else
5920         start_pfn = node_start_pfn;
5921 #endif
5922         calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5923                                   zones_size, zholes_size);
5924 
5925         alloc_node_mem_map(pgdat);
5926 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5927         printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5928                 nid, (unsigned long)pgdat,
5929                 (unsigned long)pgdat->node_mem_map);
5930 #endif
5931 
5932         free_area_init_core(pgdat);
5933 }
5934 
5935 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5936 
5937 #if MAX_NUMNODES > 1
5938 /*
5939  * Figure out the number of possible node ids.
5940  */
5941 void __init setup_nr_node_ids(void)
5942 {
5943         unsigned int highest;
5944 
5945         highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
5946         nr_node_ids = highest + 1;
5947 }
5948 #endif
5949 
5950 /**
5951  * node_map_pfn_alignment - determine the maximum internode alignment
5952  *
5953  * This function should be called after node map is populated and sorted.
5954  * It calculates the maximum power of two alignment which can distinguish
5955  * all the nodes.
5956  *
5957  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5958  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
5959  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
5960  * shifted, 1GiB is enough and this function will indicate so.
5961  *
5962  * This is used to test whether pfn -> nid mapping of the chosen memory
5963  * model has fine enough granularity to avoid incorrect mapping for the
5964  * populated node map.
5965  *
5966  * Returns the determined alignment in pfn's.  0 if there is no alignment
5967  * requirement (single node).
5968  */
5969 unsigned long __init node_map_pfn_alignment(void)
5970 {
5971         unsigned long accl_mask = 0, last_end = 0;
5972         unsigned long start, end, mask;
5973         int last_nid = -1;
5974         int i, nid;
5975 
5976         for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5977                 if (!start || last_nid < 0 || last_nid == nid) {
5978                         last_nid = nid;
5979                         last_end = end;
5980                         continue;
5981                 }
5982 
5983                 /*
5984                  * Start with a mask granular enough to pin-point to the
5985                  * start pfn and tick off bits one-by-one until it becomes
5986                  * too coarse to separate the current node from the last.
5987                  */
5988                 mask = ~((1 << __ffs(start)) - 1);
5989                 while (mask && last_end <= (start & (mask << 1)))
5990                         mask <<= 1;
5991 
5992                 /* accumulate all internode masks */
5993                 accl_mask |= mask;
5994         }
5995 
5996         /* convert mask to number of pages */
5997         return ~accl_mask + 1;
5998 }
5999 
6000 /* Find the lowest pfn for a node */
6001 static unsigned long __init find_min_pfn_for_node(int nid)
6002 {
6003         unsigned long min_pfn = ULONG_MAX;
6004         unsigned long start_pfn;
6005         int i;
6006 
6007         for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6008                 min_pfn = min(min_pfn, start_pfn);
6009 
6010         if (min_pfn == ULONG_MAX) {
6011                 pr_warn("Could not find start_pfn for node %d\n", nid);
6012                 return 0;
6013         }
6014 
6015         return min_pfn;
6016 }
6017 
6018 /**
6019  * find_min_pfn_with_active_regions - Find the minimum PFN registered
6020  *
6021  * It returns the minimum PFN based on information provided via
6022  * memblock_set_node().
6023  */
6024 unsigned long __init find_min_pfn_with_active_regions(void)
6025 {
6026         return find_min_pfn_for_node(MAX_NUMNODES);
6027 }
6028 
6029 /*
6030  * early_calculate_totalpages()
6031  * Sum pages in active regions for movable zone.
6032  * Populate N_MEMORY for calculating usable_nodes.
6033  */
6034 static unsigned long __init early_calculate_totalpages(void)
6035 {
6036         unsigned long totalpages = 0;
6037         unsigned long start_pfn, end_pfn;
6038         int i, nid;
6039 
6040         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6041                 unsigned long pages = end_pfn - start_pfn;
6042 
6043                 totalpages += pages;
6044                 if (pages)
6045                         node_set_state(nid, N_MEMORY);
6046         }
6047         return totalpages;
6048 }
6049 
6050 /*
6051  * Find the PFN the Movable zone begins in each node. Kernel memory
6052  * is spread evenly between nodes as long as the nodes have enough
6053  * memory. When they don't, some nodes will have more kernelcore than
6054  * others
6055  */
6056 static void __init find_zone_movable_pfns_for_nodes(void)
6057 {
6058         int i, nid;
6059         unsigned long usable_startpfn;
6060         unsigned long kernelcore_node, kernelcore_remaining;
6061         /* save the state before borrow the nodemask */
6062         nodemask_t saved_node_state = node_states[N_MEMORY];
6063         unsigned long totalpages = early_calculate_totalpages();
6064         int usable_nodes = nodes_weight(node_states[N_MEMORY]);
6065         struct memblock_region *r;
6066 
6067         /* Need to find movable_zone earlier when movable_node is specified. */
6068         find_usable_zone_for_movable();
6069 
6070         /*
6071          * If movable_node is specified, ignore kernelcore and movablecore
6072          * options.
6073          */
6074         if (movable_node_is_enabled()) {
6075                 for_each_memblock(memory, r) {
6076                         if (!memblock_is_hotpluggable(r))
6077                                 continue;
6078 
6079                         nid = r->nid;
6080 
6081                         usable_startpfn = PFN_DOWN(r->base);
6082                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6083                                 min(usable_startpfn, zone_movable_pfn[nid]) :
6084                                 usable_startpfn;
6085                 }
6086 
6087                 goto out2;
6088         }
6089 
6090         /*
6091          * If kernelcore=mirror is specified, ignore movablecore option
6092          */
6093         if (mirrored_kernelcore) {
6094                 bool mem_below_4gb_not_mirrored = false;
6095 
6096                 for_each_memblock(memory, r) {
6097                         if (memblock_is_mirror(r))
6098                                 continue;
6099 
6100                         nid = r->nid;
6101 
6102                         usable_startpfn = memblock_region_memory_base_pfn(r);
6103 
6104                         if (usable_startpfn < 0x100000) {
6105                                 mem_below_4gb_not_mirrored = true;
6106                                 continue;
6107                         }
6108 
6109                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6110                                 min(usable_startpfn, zone_movable_pfn[nid]) :
6111                                 usable_startpfn;
6112                 }
6113 
6114                 if (mem_below_4gb_not_mirrored)
6115                         pr_warn("This configuration results in unmirrored kernel memory.");
6116 
6117                 goto out2;
6118         }
6119 
6120         /*
6121          * If movablecore=nn[KMG] was specified, calculate what size of
6122          * kernelcore that corresponds so that memory usable for
6123          * any allocation type is evenly spread. If both kernelcore
6124          * and movablecore are specified, then the value of kernelcore
6125          * will be used for required_kernelcore if it's greater than
6126          * what movablecore would have allowed.
6127          */
6128         if (required_movablecore) {
6129                 unsigned long corepages;
6130 
6131                 /*
6132                  * Round-up so that ZONE_MOVABLE is at least as large as what
6133                  * was requested by the user
6134                  */
6135                 required_movablecore =
6136                         roundup(required_movablecore, MAX_ORDER_NR_PAGES);
6137                 required_movablecore = min(totalpages, required_movablecore);
6138                 corepages = totalpages - required_movablecore;
6139 
6140                 required_kernelcore = max(required_kernelcore, corepages);
6141         }
6142 
6143         /*
6144          * If kernelcore was not specified or kernelcore size is larger
6145          * than totalpages, there is no ZONE_MOVABLE.
6146          */
6147         if (!required_kernelcore || required_kernelcore >= totalpages)
6148                 goto out;
6149 
6150         /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
6151         usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
6152 
6153 restart:
6154         /* Spread kernelcore memory as evenly as possible throughout nodes */
6155         kernelcore_node = required_kernelcore / usable_nodes;
6156         for_each_node_state(nid, N_MEMORY) {
6157                 unsigned long start_pfn, end_pfn;
6158 
6159                 /*
6160                  * Recalculate kernelcore_node if the division per node
6161                  * now exceeds what is necessary to satisfy the requested
6162                  * amount of memory for the kernel
6163                  */
6164                 if (required_kernelcore < kernelcore_node)
6165                         kernelcore_node = required_kernelcore / usable_nodes;
6166 
6167                 /*
6168                  * As the map is walked, we track how much memory is usable
6169                  * by the kernel using kernelcore_remaining. When it is
6170                  * 0, the rest of the node is usable by ZONE_MOVABLE
6171                  */
6172                 kernelcore_remaining = kernelcore_node;
6173 
6174                 /* Go through each range of PFNs within this node */
6175                 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6176                         unsigned long size_pages;
6177 
6178                         start_pfn = max(start_pfn, zone_movable_pfn[nid]);
6179                         if (start_pfn >= end_pfn)
6180                                 continue;
6181 
6182                         /* Account for what is only usable for kernelcore */
6183                         if (start_pfn < usable_startpfn) {
6184                                 unsigned long kernel_pages;
6185                                 kernel_pages = min(end_pfn, usable_startpfn)
6186                                                                 - start_pfn;
6187 
6188                                 kernelcore_remaining -= min(kernel_pages,
6189                                                         kernelcore_remaining);
6190                                 required_kernelcore -= min(kernel_pages,
6191                                                         required_kernelcore);
6192 
6193                                 /* Continue if range is now fully accounted */
6194                                 if (end_pfn <= usable_startpfn) {
6195 
6196                                         /*
6197                                          * Push zone_movable_pfn to the end so
6198                                          * that if we have to rebalance
6199                                          * kernelcore across nodes, we will
6200                                          * not double account here
6201                                          */
6202                                         zone_movable_pfn[nid] = end_pfn;
6203                                         continue;
6204                                 }
6205                                 start_pfn = usable_startpfn;
6206                         }
6207 
6208                         /*
6209                          * The usable PFN range for ZONE_MOVABLE is from
6210                          * start_pfn->end_pfn. Calculate size_pages as the
6211                          * number of pages used as kernelcore
6212                          */
6213                         size_pages = end_pfn - start_pfn;
6214                         if (size_pages > kernelcore_remaining)
6215                                 size_pages = kernelcore_remaining;
6216                         zone_movable_pfn[nid] = start_pfn + size_pages;
6217 
6218                         /*
6219                          * Some kernelcore has been met, update counts and
6220                          * break if the kernelcore for this node has been
6221                          * satisfied
6222                          */
6223                         required_kernelcore -= min(required_kernelcore,
6224                                                                 size_pages);
6225                         kernelcore_remaining -= size_pages;
6226                         if (!kernelcore_remaining)
6227                                 break;
6228                 }
6229         }
6230 
6231         /*
6232          * If there is still required_kernelcore, we do another pass with one
6233          * less node in the count. This will push zone_movable_pfn[nid] further
6234          * along on the nodes that still have memory until kernelcore is
6235          * satisfied
6236          */
6237         usable_nodes--;
6238         if (usable_nodes && required_kernelcore > usable_nodes)
6239                 goto restart;
6240 
6241 out2:
6242         /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6243         for (nid = 0; nid < MAX_NUMNODES; nid++)
6244                 zone_movable_pfn[nid] =
6245                         roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
6246 
6247 out:
6248         /* restore the node_state */
6249         node_states[N_MEMORY] = saved_node_state;
6250 }
6251 
6252 /* Any regular or high memory on that node ? */
6253 static void check_for_memory(pg_data_t *pgdat, int nid)
6254 {
6255         enum zone_type zone_type;
6256 
6257         if (N_MEMORY == N_NORMAL_MEMORY)
6258                 return;
6259 
6260         for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
6261                 struct zone *zone = &pgdat->node_zones[zone_type];
6262                 if (populated_zone(zone)) {
6263                         node_set_state(nid, N_HIGH_MEMORY);
6264                         if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
6265                             zone_type <= ZONE_NORMAL)
6266                                 node_set_state(nid, N_NORMAL_MEMORY);
6267                         break;
6268                 }
6269         }
6270 }
6271 
6272 /**
6273  * free_area_init_nodes - Initialise all pg_data_t and zone data
6274  * @max_zone_pfn: an array of max PFNs for each zone
6275  *
6276  * This will call free_area_init_node() for each active node in the system.
6277  * Using the page ranges provided by memblock_set_node(), the size of each
6278  * zone in each node and their holes is calculated. If the maximum PFN
6279  * between two adjacent zones match, it is assumed that the zone is empty.
6280  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6281  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6282  * starts where the previous one ended. For example, ZONE_DMA32 starts
6283  * at arch_max_dma_pfn.
6284  */
6285 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
6286 {
6287         unsigned long start_pfn, end_pfn;
6288         int i, nid;
6289 
6290         /* Record where the zone boundaries are */
6291         memset(arch_zone_lowest_possible_pfn, 0,
6292                                 sizeof(arch_zone_lowest_possible_pfn));
6293         memset(arch_zone_highest_possible_pfn, 0,
6294                                 sizeof(arch_zone_highest_possible_pfn));
6295 
6296         start_pfn = find_min_pfn_with_active_regions();
6297 
6298         for (i = 0; i < MAX_NR_ZONES; i++) {
6299                 if (i == ZONE_MOVABLE)
6300                         continue;
6301 
6302                 end_pfn = max(max_zone_pfn[i], start_pfn);
6303                 arch_zone_lowest_possible_pfn[i] = start_pfn;
6304                 arch_zone_highest_possible_pfn[i] = end_pfn;
6305 
6306                 start_pfn = end_pfn;
6307         }
6308         arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
6309         arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
6310 
6311         /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6312         memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
6313         find_zone_movable_pfns_for_nodes();
6314 
6315         /* Print out the zone ranges */
6316         pr_info("Zone ranges:\n");
6317         for (i = 0; i < MAX_NR_ZONES; i++) {
6318                 if (i == ZONE_MOVABLE)
6319                         continue;
6320                 pr_info("  %-8s ", zone_names[i]);
6321                 if (arch_zone_lowest_possible_pfn[i] ==
6322                                 arch_zone_highest_possible_pfn[i])
6323                         pr_cont("empty\n");
6324                 else
6325                         pr_cont("[mem %#018Lx-%#018Lx]\n",
6326                                 (u64)arch_zone_lowest_possible_pfn[i]
6327                                         << PAGE_SHIFT,
6328                                 ((u64)arch_zone_highest_possible_pfn[i]
6329                                         << PAGE_SHIFT) - 1);
6330         }
6331 
6332         /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6333         pr_info("Movable zone start for each node\n");
6334         for (i = 0; i < MAX_NUMNODES; i++) {
6335                 if (zone_movable_pfn[i])
6336                         pr_info("  Node %d: %#018Lx\n", i,
6337                                (u64)zone_movable_pfn[i] << PAGE_SHIFT);
6338         }
6339 
6340         /* Print out the early node map */
6341         pr_info("Early memory node ranges\n");
6342         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
6343                 pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6344                         (u64)start_pfn << PAGE_SHIFT,
6345                         ((u64)end_pfn << PAGE_SHIFT) - 1);
6346 
6347         /* Initialise every node */
6348         mminit_verify_pageflags_layout();
6349         setup_nr_node_ids();
6350         for_each_online_node(nid) {
6351                 pg_data_t *pgdat = NODE_DATA(nid);
6352                 free_area_init_node(nid, NULL,
6353                                 find_min_pfn_for_node(nid), NULL);
6354 
6355                 /* Any memory on that node */
6356                 if (pgdat->node_present_pages)
6357                         node_set_state(nid, N_MEMORY);
6358                 check_for_memory(pgdat, nid);
6359         }
6360 }
6361 
6362 static int __init cmdline_parse_core(char *p, unsigned long *core)
6363 {
6364         unsigned long long coremem;
6365         if (!p)
6366                 return -EINVAL;
6367 
6368         coremem = memparse(p, &p);
6369         *core = coremem >> PAGE_SHIFT;
6370 
6371         /* Paranoid check that UL is enough for the coremem value */
6372         WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6373 
6374         return 0;
6375 }
6376 
6377 /*
6378  * kernelcore=size sets the amount of memory for use for allocations that
6379  * cannot be reclaimed or migrated.
6380  */
6381 static int __init cmdline_parse_kernelcore(char *p)
6382 {
6383         /* parse kernelcore=mirror */
6384         if (parse_option_str(p, "mirror")) {
6385                 mirrored_kernelcore = true;
6386                 return 0;
6387         }
6388 
6389         return cmdline_parse_core(p, &required_kernelcore);
6390 }
6391 
6392 /*
6393  * movablecore=size sets the amount of memory for use for allocations that
6394  * can be reclaimed or migrated.
6395  */
6396 static int __init cmdline_parse_movablecore(char *p)
6397 {
6398         return cmdline_parse_core(p, &required_movablecore);
6399 }
6400 
6401 early_param("kernelcore", cmdline_parse_kernelcore);
6402 early_param("movablecore", cmdline_parse_movablecore);
6403 
6404 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6405 
6406 void adjust_managed_page_count(struct page *page, long count)
6407 {
6408         spin_lock(&managed_page_count_lock);
6409         page_zone(page)->managed_pages += count;
6410         totalram_pages += count;
6411 #ifdef CONFIG_HIGHMEM
6412         if (PageHighMem(page))
6413                 totalhigh_pages += count;
6414 #endif
6415         spin_unlock(&managed_page_count_lock);
6416 }
6417 EXPORT_SYMBOL(adjust_managed_page_count);
6418 
6419 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
6420 {
6421         void *pos;
6422         unsigned long pages = 0;
6423 
6424         start = (void *)PAGE_ALIGN((unsigned long)start);
6425         end = (void *)((unsigned long)end & PAGE_MASK);
6426         for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
6427                 if ((unsigned int)poison <= 0xFF)
6428                         memset(pos, poison, PAGE_SIZE);
6429                 free_reserved_page(virt_to_page(pos));
6430         }
6431 
6432         if (pages && s)
6433                 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
6434                         s, pages << (PAGE_SHIFT - 10), start, end);
6435 
6436         return pages;
6437 }
6438 EXPORT_SYMBOL(free_reserved_area);
6439 
6440 #ifdef  CONFIG_HIGHMEM
6441 void free_highmem_page(struct page *page)
6442 {
6443         __free_reserved_page(page);
6444         totalram_pages++;
6445         page_zone(page)->managed_pages++;
6446         totalhigh_pages++;
6447 }
6448 #endif
6449 
6450 
6451 void __init mem_init_print_info(const char *str)
6452 {
6453         unsigned long physpages, codesize, datasize, rosize, bss_size;
6454         unsigned long init_code_size, init_data_size;
6455 
6456         physpages = get_num_physpages();
6457         codesize = _etext - _stext;
6458         datasize = _edata - _sdata;
6459         rosize = __end_rodata - __start_rodata;
6460         bss_size = __bss_stop - __bss_start;
6461         init_data_size = __init_end - __init_begin;
6462         init_code_size = _einittext - _sinittext;
6463 
6464         /*
6465          * Detect special cases and adjust section sizes accordingly:
6466          * 1) .init.* may be embedded into .data sections
6467          * 2) .init.text.* may be out of [__init_begin, __init_end],
6468          *    please refer to arch/tile/kernel/vmlinux.lds.S.
6469          * 3) .rodata.* may be embedded into .text or .data sections.
6470          */
6471 #define adj_init_size(start, end, size, pos, adj) \
6472         do { \
6473                 if (start <= pos && pos < end && size > adj) \
6474                         size -= adj; \
6475         } while (0)
6476 
6477         adj_init_size(__init_begin, __init_end, init_data_size,
6478                      _sinittext, init_code_size);
6479         adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6480         adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6481         adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6482         adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6483 
6484 #undef  adj_init_size
6485 
6486         pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6487 #ifdef  CONFIG_HIGHMEM
6488                 ", %luK highmem"
6489 #endif
6490                 "%s%s)\n",
6491                 nr_free_pages() << (PAGE_SHIFT - 10),
6492                 physpages << (PAGE_SHIFT - 10),
6493                 codesize >> 10, datasize >> 10, rosize >> 10,
6494                 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6495                 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6496                 totalcma_pages << (PAGE_SHIFT - 10),
6497 #ifdef  CONFIG_HIGHMEM
6498                 totalhigh_pages << (PAGE_SHIFT - 10),
6499 #endif
6500                 str ? ", " : "", str ? str : "");
6501 }
6502 
6503 /**
6504  * set_dma_reserve - set the specified number of pages reserved in the first zone
6505  * @new_dma_reserve: The number of pages to mark reserved
6506  *
6507  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6508  * In the DMA zone, a significant percentage may be consumed by kernel image
6509  * and other unfreeable allocations which can skew the watermarks badly. This
6510  * function may optionally be used to account for unfreeable pages in the
6511  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6512  * smaller per-cpu batchsize.
6513  */
6514 void __init set_dma_reserve(unsigned long new_dma_reserve)
6515 {
6516         dma_reserve = new_dma_reserve;
6517 }
6518 
6519 void __init free_area_init(unsigned long *zones_size)
6520 {
6521         free_area_init_node(0, zones_size,
6522                         __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6523 }
6524 
6525 static int page_alloc_cpu_notify(struct notifier_block *self,
6526                                  unsigned long action, void *hcpu)
6527 {
6528         int cpu = (unsigned long)hcpu;
6529 
6530         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
6531                 lru_add_drain_cpu(cpu);
6532                 drain_pages(cpu);
6533 
6534                 /*
6535                  * Spill the event counters of the dead processor
6536                  * into the current processors event counters.
6537                  * This artificially elevates the count of the current
6538                  * processor.
6539                  */
6540                 vm_events_fold_cpu(cpu);
6541 
6542                 /*
6543                  * Zero the differential counters of the dead processor
6544                  * so that the vm statistics are consistent.
6545                  *
6546                  * This is only okay since the processor is dead and cannot
6547                  * race with what we are doing.
6548                  */
6549                 cpu_vm_stats_fold(cpu);
6550         }
6551         return NOTIFY_OK;
6552 }
6553 
6554 void __init page_alloc_init(void)
6555 {
6556         hotcpu_notifier(page_alloc_cpu_notify, 0);
6557 }
6558 
6559 /*
6560  * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6561  *      or min_free_kbytes changes.
6562  */
6563 static void calculate_totalreserve_pages(void)
6564 {
6565         struct pglist_data *pgdat;
6566         unsigned long reserve_pages = 0;
6567         enum zone_type i, j;
6568 
6569         for_each_online_pgdat(pgdat) {
6570 
6571                 pgdat->totalreserve_pages = 0;
6572 
6573                 for (i = 0; i < MAX_NR_ZONES; i++) {
6574                         struct zone *zone = pgdat->node_zones + i;
6575                         long max = 0;
6576 
6577                         /* Find valid and maximum lowmem_reserve in the zone */
6578                         for (j = i; j < MAX_NR_ZONES; j++) {
6579                                 if (zone->lowmem_reserve[j] > max)
6580                                         max = zone->lowmem_reserve[j];
6581                         }
6582 
6583                         /* we treat the high watermark as reserved pages. */
6584                         max += high_wmark_pages(zone);
6585 
6586                         if (max > zone->managed_pages)
6587                                 max = zone->managed_pages;
6588 
6589                         pgdat->totalreserve_pages += max;
6590 
6591                         reserve_pages += max;
6592                 }
6593         }
6594         totalreserve_pages = reserve_pages;
6595 }
6596 
6597 /*
6598  * setup_per_zone_lowmem_reserve - called whenever
6599  *      sysctl_lowmem_reserve_ratio changes.  Ensures that each zone
6600  *      has a correct pages reserved value, so an adequate number of
6601  *      pages are left in the zone after a successful __alloc_pages().
6602  */
6603 static void setup_per_zone_lowmem_reserve(void)
6604 {
6605         struct pglist_data *pgdat;
6606         enum zone_type j, idx;
6607 
6608         for_each_online_pgdat(pgdat) {
6609                 for (j = 0; j < MAX_NR_ZONES; j++) {
6610                         struct zone *zone = pgdat->node_zones + j;
6611                         unsigned long managed_pages = zone->managed_pages;
6612 
6613                         zone->lowmem_reserve[j] = 0;
6614 
6615                         idx = j;
6616                         while (idx) {
6617                                 struct zone *lower_zone;
6618 
6619                                 idx--;
6620 
6621                                 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6622                                         sysctl_lowmem_reserve_ratio[idx] = 1;
6623 
6624                                 lower_zone = pgdat->node_zones + idx;
6625                                 lower_zone->lowmem_reserve[j] = managed_pages /
6626                                         sysctl_lowmem_reserve_ratio[idx];
6627                                 managed_pages += lower_zone->managed_pages;
6628                         }
6629                 }
6630         }
6631 
6632         /* update totalreserve_pages */
6633         calculate_totalreserve_pages();
6634 }
6635 
6636 static void __setup_per_zone_wmarks(void)
6637 {
6638         unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6639         unsigned long lowmem_pages = 0;
6640         struct zone *zone;
6641         unsigned long flags;
6642 
6643         /* Calculate total number of !ZONE_HIGHMEM pages */
6644         for_each_zone(zone) {
6645                 if (!is_highmem(zone))
6646                         lowmem_pages += zone->managed_pages;
6647         }
6648 
6649         for_each_zone(zone) {
6650                 u64 tmp;
6651 
6652                 spin_lock_irqsave(&zone->lock, flags);
6653                 tmp = (u64)pages_min * zone->managed_pages;
6654                 do_div(tmp, lowmem_pages);
6655                 if (is_highmem(zone)) {
6656                         /*
6657                          * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6658                          * need highmem pages, so cap pages_min to a small
6659                          * value here.
6660                          *
6661                          * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6662                          * deltas control asynch page reclaim, and so should
6663                          * not be capped for highmem.
6664                          */
6665                         unsigned long min_pages;
6666 
6667                         min_pages = zone->managed_pages / 1024;
6668                         min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
6669                         zone->watermark[WMARK_MIN] = min_pages;
6670                 } else {
6671                         /*
6672                          * If it's a lowmem zone, reserve a number of pages
6673                          * proportionate to the zone's size.
6674                          */
6675                         zone->watermark[WMARK_MIN] = tmp;
6676                 }
6677 
6678                 /*
6679                  * Set the kswapd watermarks distance according to the
6680                  * scale factor in proportion to available memory, but
6681                  * ensure a minimum size on small systems.
6682                  */
6683                 tmp = max_t(u64, tmp >> 2,
6684                             mult_frac(zone->managed_pages,
6685                                       watermark_scale_factor, 10000));
6686 
6687                 zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + tmp;
6688                 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
6689 
6690                 spin_unlock_irqrestore(&zone->lock, flags);
6691         }
6692 
6693         /* update totalreserve_pages */
6694         calculate_totalreserve_pages();
6695 }
6696 
6697 /**
6698  * setup_per_zone_wmarks - called when min_free_kbytes changes
6699  * or when memory is hot-{added|removed}
6700  *
6701  * Ensures that the watermark[min,low,high] values for each zone are set
6702  * correctly with respect to min_free_kbytes.
6703  */
6704 void setup_per_zone_wmarks(void)
6705 {
6706         mutex_lock(&zonelists_mutex);
6707         __setup_per_zone_wmarks();
6708         mutex_unlock(&zonelists_mutex);
6709 }
6710 
6711 /*
6712  * Initialise min_free_kbytes.
6713  *
6714  * For small machines we want it small (128k min).  For large machines
6715  * we want it large (64MB max).  But it is not linear, because network
6716  * bandwidth does not increase linearly with machine size.  We use
6717  *
6718  *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6719  *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
6720  *
6721  * which yields
6722  *
6723  * 16MB:        512k
6724  * 32MB:        724k
6725  * 64MB:        1024k
6726  * 128MB:       1448k
6727  * 256MB:       2048k
6728  * 512MB:       2896k
6729  * 1024MB:      4096k
6730  * 2048MB:      5792k
6731  * 4096MB:      8192k
6732  * 8192MB:      11584k
6733  * 16384MB:     16384k
6734  */
6735 int __meminit init_per_zone_wmark_min(void)
6736 {
6737         unsigned long lowmem_kbytes;
6738         int new_min_free_kbytes;
6739 
6740         lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
6741         new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
6742 
6743         if (new_min_free_kbytes > user_min_free_kbytes) {
6744                 min_free_kbytes = new_min_free_kbytes;
6745                 if (min_free_kbytes < 128)
6746                         min_free_kbytes = 128;
6747                 if (min_free_kbytes > 65536)
6748                         min_free_kbytes = 65536;
6749         } else {
6750                 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6751                                 new_min_free_kbytes, user_min_free_kbytes);
6752         }
6753         setup_per_zone_wmarks();
6754         refresh_zone_stat_thresholds();
6755         setup_per_zone_lowmem_reserve();
6756 
6757 #ifdef CONFIG_NUMA
6758         setup_min_unmapped_ratio();
6759         setup_min_slab_ratio();
6760 #endif
6761 
6762         return 0;
6763 }
6764 core_initcall(init_per_zone_wmark_min)
6765 
6766 /*
6767  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6768  *      that we can call two helper functions whenever min_free_kbytes
6769  *      changes.
6770  */
6771 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
6772         void __user *buffer, size_t *length, loff_t *ppos)
6773 {
6774         int rc;
6775 
6776         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6777         if (rc)
6778                 return rc;
6779 
6780         if (write) {
6781                 user_min_free_kbytes = min_free_kbytes;
6782                 setup_per_zone_wmarks();
6783         }
6784         return 0;
6785 }
6786 
6787 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
6788         void __user *buffer, size_t *length, loff_t *ppos)
6789 {
6790         int rc;
6791 
6792         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6793         if (rc)
6794                 return rc;
6795 
6796         if (write)
6797                 setup_per_zone_wmarks();
6798 
6799         return 0;
6800 }
6801 
6802 #ifdef CONFIG_NUMA
6803 static void setup_min_unmapped_ratio(void)
6804 {
6805         pg_data_t *pgdat;
6806         struct zone *zone;
6807 
6808         for_each_online_pgdat(pgdat)
6809                 pgdat->min_unmapped_pages = 0;
6810 
6811         for_each_zone(zone)
6812                 zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages *
6813                                 sysctl_min_unmapped_ratio) / 100;
6814 }
6815 
6816 
6817 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
6818         void __user *buffer, size_t *length, loff_t *ppos)
6819 {
6820         int rc;
6821 
6822         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6823         if (rc)
6824                 return rc;
6825 
6826         setup_min_unmapped_ratio();
6827 
6828         return 0;
6829 }
6830 
6831 static void setup_min_slab_ratio(void)
6832 {
6833         pg_data_t *pgdat;
6834         struct zone *zone;
6835 
6836         for_each_online_pgdat(pgdat)
6837                 pgdat->min_slab_pages = 0;
6838 
6839         for_each_zone(zone)
6840                 zone->zone_pgdat->min_slab_pages += (zone->managed_pages *
6841                                 sysctl_min_slab_ratio) / 100;
6842 }
6843 
6844 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
6845         void __user *buffer, size_t *length, loff_t *ppos)
6846 {
6847         int rc;
6848 
6849         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6850         if (rc)
6851                 return rc;
6852 
6853         setup_min_slab_ratio();
6854 
6855         return 0;
6856 }
6857 #endif
6858 
6859 /*
6860  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6861  *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6862  *      whenever sysctl_lowmem_reserve_ratio changes.
6863  *
6864  * The reserve ratio obviously has absolutely no relation with the
6865  * minimum watermarks. The lowmem reserve ratio can only make sense
6866  * if in function of the boot time zone sizes.
6867  */
6868 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
6869         void __user *buffer, size_t *length, loff_t *ppos)
6870 {
6871         proc_dointvec_minmax(table, write, buffer, length, ppos);
6872         setup_per_zone_lowmem_reserve();
6873         return 0;
6874 }
6875 
6876 /*
6877  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6878  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
6879  * pagelist can have before it gets flushed back to buddy allocator.
6880  */
6881 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
6882         void __user *buffer, size_t *length, loff_t *ppos)
6883 {
6884         struct zone *zone;
6885         int old_percpu_pagelist_fraction;
6886         int ret;
6887 
6888         mutex_lock(&pcp_batch_high_lock);
6889         old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6890 
6891         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6892         if (!write || ret < 0)
6893                 goto out;
6894 
6895         /* Sanity checking to avoid pcp imbalance */
6896         if (percpu_pagelist_fraction &&
6897             percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6898                 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6899                 ret = -EINVAL;
6900                 goto out;
6901         }
6902 
6903         /* No change? */
6904         if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6905                 goto out;
6906 
6907         for_each_populated_zone(zone) {
6908                 unsigned int cpu;
6909 
6910                 for_each_possible_cpu(cpu)
6911                         pageset_set_high_and_batch(zone,
6912                                         per_cpu_ptr(zone->pageset, cpu));
6913         }
6914 out:
6915         mutex_unlock(&pcp_batch_high_lock);
6916         return ret;
6917 }
6918 
6919 #ifdef CONFIG_NUMA
6920 int hashdist = HASHDIST_DEFAULT;
6921 
6922 static int __init set_hashdist(char *str)
6923 {
6924         if (!str)
6925                 return 0;
6926         hashdist = simple_strtoul(str, &str, 0);
6927         return 1;
6928 }
6929 __setup("hashdist=", set_hashdist);
6930 #endif
6931 
6932 /*
6933  * allocate a large system hash table from bootmem
6934  * - it is assumed that the hash table must contain an exact power-of-2
6935  *   quantity of entries
6936  * - limit is the number of hash buckets, not the total allocation size
6937  */
6938 void *__init alloc_large_system_hash(const char *tablename,
6939                                      unsigned long bucketsize,
6940                                      unsigned long numentries,
6941                                      int scale,
6942                                      int flags,
6943                                      unsigned int *_hash_shift,
6944                                      unsigned int *_hash_mask,
6945                                      unsigned long low_limit,
6946                                      unsigned long high_limit)
6947 {
6948         unsigned long long max = high_limit;
6949         unsigned long log2qty, size;
6950         void *table = NULL;
6951 
6952         /* allow the kernel cmdline to have a say */
6953         if (!numentries) {
6954                 /* round applicable memory size up to nearest megabyte */
6955                 numentries = nr_kernel_pages;
6956 
6957                 /* It isn't necessary when PAGE_SIZE >= 1MB */
6958                 if (PAGE_SHIFT < 20)
6959                         numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
6960 
6961                 /* limit to 1 bucket per 2^scale bytes of low memory */
6962                 if (scale > PAGE_SHIFT)
6963                         numentries >>= (scale - PAGE_SHIFT);
6964                 else
6965                         numentries <<= (PAGE_SHIFT - scale);
6966 
6967                 /* Make sure we've got at least a 0-order allocation.. */
6968                 if (unlikely(flags & HASH_SMALL)) {
6969                         /* Makes no sense without HASH_EARLY */
6970                         WARN_ON(!(flags & HASH_EARLY));
6971                         if (!(numentries >> *_hash_shift)) {
6972                                 numentries = 1UL << *_hash_shift;
6973                                 BUG_ON(!numentries);
6974                         }
6975                 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
6976                         numentries = PAGE_SIZE / bucketsize;
6977         }
6978         numentries = roundup_pow_of_two(numentries);
6979 
6980         /* limit allocation size to 1/16 total memory by default */
6981         if (max == 0) {
6982                 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6983                 do_div(max, bucketsize);
6984         }
6985         max = min(max, 0x80000000ULL);
6986 
6987         if (numentries < low_limit)
6988                 numentries = low_limit;
6989         if (numentries > max)
6990                 numentries = max;
6991 
6992         log2qty = ilog2(numentries);
6993 
6994         do {
6995                 size = bucketsize << log2qty;
6996                 if (flags & HASH_EARLY)
6997                         table = memblock_virt_alloc_nopanic(size, 0);
6998                 else if (hashdist)
6999                         table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
7000                 else {
7001                         /*
7002                          * If bucketsize is not a power-of-two, we may free
7003                          * some pages at the end of hash table which
7004                          * alloc_pages_exact() automatically does
7005                          */
7006                         if (get_order(size) < MAX_ORDER) {
7007                                 table = alloc_pages_exact(size, GFP_ATOMIC);
7008                                 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
7009                         }
7010                 }
7011         } while (!table && size > PAGE_SIZE && --log2qty);
7012 
7013         if (!table)
7014                 panic("Failed to allocate %s hash table\n", tablename);
7015 
7016         pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7017                 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
7018 
7019         if (_hash_shift)
7020                 *_hash_shift = log2qty;
7021         if (_hash_mask)
7022                 *_hash_mask = (1 << log2qty) - 1;
7023 
7024         return table;
7025 }
7026 
7027 /*
7028  * This function checks whether pageblock includes unmovable pages or not.
7029  * If @count is not zero, it is okay to include less @count unmovable pages
7030  *
7031  * PageLRU check without isolation or lru_lock could race so that
7032  * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
7033  * expect this function should be exact.
7034  */
7035 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
7036                          bool skip_hwpoisoned_pages)
7037 {
7038         unsigned long pfn, iter, found;
7039         int mt;
7040 
7041         /*
7042          * For avoiding noise data, lru_add_drain_all() should be called
7043          * If ZONE_MOVABLE, the zone never contains unmovable pages
7044          */
7045         if (zone_idx(zone) == ZONE_MOVABLE)
7046                 return false;
7047         mt = get_pageblock_migratetype(page);
7048         if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
7049                 return false;
7050 
7051         pfn = page_to_pfn(page);
7052         for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
7053                 unsigned long check = pfn + iter;
7054 
7055                 if (!pfn_valid_within(check))
7056                         continue;
7057 
7058                 page = pfn_to_page(check);
7059 
7060                 /*
7061                  * Hugepages are not in LRU lists, but they're movable.
7062                  * We need not scan over tail pages bacause we don't
7063                  * handle each tail page individually in migration.
7064                  */
7065                 if (PageHuge(page)) {
7066                         iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
7067                         continue;
7068                 }
7069 
7070                 /*
7071                  * We can't use page_count without pin a page
7072                  * because another CPU can free compound page.
7073                  * This check already skips compound tails of THP
7074                  * because their page->_refcount is zero at all time.
7075                  */
7076                 if (!page_ref_count(page)) {
7077                         if (PageBuddy(page))
7078                                 iter += (1 << page_order(page)) - 1;
7079                         continue;
7080                 }
7081 
7082                 /*
7083                  * The HWPoisoned page may be not in buddy system, and
7084                  * page_count() is not 0.
7085                  */
7086                 if (skip_hwpoisoned_pages && PageHWPoison(page))
7087                         continue;
7088 
7089                 if (!PageLRU(page))
7090                         found++;
7091                 /*
7092                  * If there are RECLAIMABLE pages, we need to check
7093                  * it.  But now, memory offline itself doesn't call
7094                  * shrink_node_slabs() and it still to be fixed.
7095                  */
7096                 /*
7097                  * If the page is not RAM, page_count()should be 0.
7098                  * we don't need more check. This is an _used_ not-movable page.
7099                  *
7100                  * The problematic thing here is PG_reserved pages. PG_reserved
7101                  * is set to both of a memory hole page and a _used_ kernel
7102                  * page at boot.
7103                  */
7104                 if (found > count)
7105                         return true;
7106         }
7107         return false;
7108 }
7109 
7110 bool is_pageblock_removable_nolock(struct page *page)
7111 {
7112         struct zone *zone;
7113         unsigned long pfn;
7114 
7115         /*
7116          * We have to be careful here because we are iterating over memory
7117          * sections which are not zone aware so we might end up outside of
7118          * the zone but still within the section.
7119          * We have to take care about the node as well. If the node is offline
7120          * its NODE_DATA will be NULL - see page_zone.
7121          */
7122         if (!node_online(page_to_nid(page)))
7123                 return false;
7124 
7125         zone = page_zone(page);
7126         pfn = page_to_pfn(page);
7127         if (!zone_spans_pfn(zone, pfn))
7128                 return false;
7129 
7130         return !has_unmovable_pages(zone, page, 0, true);
7131 }
7132 
7133 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7134 
7135 static unsigned long pfn_max_align_down(unsigned long pfn)
7136 {
7137         return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
7138                              pageblock_nr_pages) - 1);
7139 }
7140 
7141 static unsigned long pfn_max_align_up(unsigned long pfn)
7142 {
7143         return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
7144                                 pageblock_nr_pages));
7145 }
7146 
7147 /* [start, end) must belong to a single zone. */
7148 static int __alloc_contig_migrate_range(struct compact_control *cc,
7149                                         unsigned long start, unsigned long end)
7150 {
7151         /* This function is based on compact_zone() from compaction.c. */
7152         unsigned long nr_reclaimed;
7153         unsigned long pfn = start;
7154         unsigned int tries = 0;
7155         int ret = 0;
7156 
7157         migrate_prep();
7158 
7159         while (pfn < end || !list_empty(&cc->migratepages)) {
7160                 if (fatal_signal_pending(current)) {
7161                         ret = -EINTR;
7162                         break;
7163                 }
7164 
7165                 if (list_empty(&cc->migratepages)) {
7166                         cc->nr_migratepages = 0;
7167                         pfn = isolate_migratepages_range(cc, pfn, end);
7168                         if (!pfn) {
7169                                 ret = -EINTR;
7170                                 break;
7171                         }
7172                         tries = 0;
7173                 } else if (++tries == 5) {
7174                         ret = ret < 0 ? ret : -EBUSY;
7175                         break;
7176                 }
7177 
7178                 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
7179                                                         &cc->migratepages);
7180                 cc->nr_migratepages -= nr_reclaimed;
7181 
7182                 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
7183                                     NULL, 0, cc->mode, MR_CMA);
7184         }
7185         if (ret < 0) {
7186                 putback_movable_pages(&cc->migratepages);
7187                 return ret;
7188         }
7189         return 0;
7190 }
7191 
7192 /**
7193  * alloc_contig_range() -- tries to allocate given range of pages
7194  * @start:      start PFN to allocate
7195  * @end:        one-past-the-last PFN to allocate
7196  * @migratetype:        migratetype of the underlaying pageblocks (either
7197  *                      #MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
7198  *                      in range must have the same migratetype and it must
7199  *                      be either of the two.
7200  *
7201  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7202  * aligned, however it's the caller's responsibility to guarantee that
7203  * we are the only thread that changes migrate type of pageblocks the
7204  * pages fall in.
7205  *
7206  * The PFN range must belong to a single zone.
7207  *
7208  * Returns zero on success or negative error code.  On success all
7209  * pages which PFN is in [start, end) are allocated for the caller and
7210  * need to be freed with free_contig_range().
7211  */
7212 int alloc_contig_range(unsigned long start, unsigned long end,
7213                        unsigned migratetype)
7214 {
7215         unsigned long outer_start, outer_end;
7216         unsigned int order;
7217         int ret = 0;
7218 
7219         struct compact_control cc = {
7220                 .nr_migratepages = 0,
7221                 .order = -1,
7222                 .zone = page_zone(pfn_to_page(start)),
7223                 .mode = MIGRATE_SYNC,
7224                 .ignore_skip_hint = true,
7225         };
7226         INIT_LIST_HEAD(&cc.migratepages);
7227 
7228         /*
7229          * What we do here is we mark all pageblocks in range as
7230          * MIGRATE_ISOLATE.  Because pageblock and max order pages may
7231          * have different sizes, and due to the way page allocator
7232          * work, we align the range to biggest of the two pages so
7233          * that page allocator won't try to merge buddies from
7234          * different pageblocks and change MIGRATE_ISOLATE to some
7235          * other migration type.
7236          *
7237          * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7238          * migrate the pages from an unaligned range (ie. pages that
7239          * we are interested in).  This will put all the pages in
7240          * range back to page allocator as MIGRATE_ISOLATE.
7241          *
7242          * When this is done, we take the pages in range from page
7243          * allocator removing them from the buddy system.  This way
7244          * page allocator will never consider using them.
7245          *
7246          * This lets us mark the pageblocks back as
7247          * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7248          * aligned range but not in the unaligned, original range are
7249          * put back to page allocator so that buddy can use them.
7250          */
7251 
7252         ret = start_isolate_page_range(pfn_max_align_down(start),
7253                                        pfn_max_align_up(end), migratetype,
7254                                        false);
7255         if (ret)
7256                 return ret;
7257 
7258         /*
7259          * In case of -EBUSY, we'd like to know which page causes problem.
7260          * So, just fall through. We will check it in test_pages_isolated().
7261          */
7262         ret = __alloc_contig_migrate_range(&cc, start, end);
7263         if (ret && ret != -EBUSY)
7264                 goto done;
7265 
7266         /*
7267          * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7268          * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
7269          * more, all pages in [start, end) are free in page allocator.
7270          * What we are going to do is to allocate all pages from
7271          * [start, end) (that is remove them from page allocator).
7272          *
7273          * The only problem is that pages at the beginning and at the
7274          * end of interesting range may be not aligned with pages that
7275          * page allocator holds, ie. they can be part of higher order
7276          * pages.  Because of this, we reserve the bigger range and
7277          * once this is done free the pages we are not interested in.
7278          *
7279          * We don't have to hold zone->lock here because the pages are
7280          * isolated thus they won't get removed from buddy.
7281          */
7282 
7283         lru_add_drain_all();
7284         drain_all_pages(cc.zone);
7285 
7286         order = 0;
7287         outer_start = start;
7288         while (!PageBuddy(pfn_to_page(outer_start))) {
7289                 if (++order >= MAX_ORDER) {
7290                         outer_start = start;
7291                         break;
7292                 }
7293                 outer_start &= ~0UL << order;
7294         }
7295 
7296         if (outer_start != start) {
7297                 order = page_order(pfn_to_page(outer_start));
7298 
7299                 /*
7300                  * outer_start page could be small order buddy page and
7301                  * it doesn't include start page. Adjust outer_start
7302                  * in this case to report failed page properly
7303                  * on tracepoint in test_pages_isolated()
7304                  */
7305                 if (outer_start + (1UL << order) <= start)
7306                         outer_start = start;
7307         }
7308 
7309         /* Make sure the range is really isolated. */
7310         if (test_pages_isolated(outer_start, end, false)) {
7311                 pr_info("%s: [%lx, %lx) PFNs busy\n",
7312                         __func__, outer_start, end);
7313                 ret = -EBUSY;
7314                 goto done;
7315         }
7316 
7317         /* Grab isolated pages from freelists. */
7318         outer_end = isolate_freepages_range(&cc, outer_start, end);
7319         if (!outer_end) {
7320                 ret = -EBUSY;
7321                 goto done;
7322         }
7323 
7324         /* Free head and tail (if any) */
7325         if (start != outer_start)
7326                 free_contig_range(outer_start, start - outer_start);
7327         if (end != outer_end)
7328                 free_contig_range(end, outer_end - end);
7329 
7330 done:
7331         undo_isolate_page_range(pfn_max_align_down(start),
7332                                 pfn_max_align_up(end), migratetype);
7333         return ret;
7334 }
7335 
7336 void free_contig_range(unsigned long pfn, unsigned nr_pages)
7337 {
7338         unsigned int count = 0;
7339 
7340         for (; nr_pages--; pfn++) {
7341                 struct page *page = pfn_to_page(pfn);
7342 
7343                 count += page_count(page) != 1;
7344                 __free_page(page);
7345         }
7346         WARN(count != 0, "%d pages are still in use!\n", count);
7347 }
7348 #endif
7349 
7350 #ifdef CONFIG_MEMORY_HOTPLUG
7351 /*
7352  * The zone indicated has a new number of managed_pages; batch sizes and percpu
7353  * page high values need to be recalulated.
7354  */
7355 void __meminit zone_pcp_update(struct zone *zone)
7356 {
7357         unsigned cpu;
7358         mutex_lock(&pcp_batch_high_lock);
7359         for_each_possible_cpu(cpu)
7360                 pageset_set_high_and_batch(zone,
7361                                 per_cpu_ptr(zone->pageset, cpu));
7362         mutex_unlock(&pcp_batch_high_lock);
7363 }
7364 #endif
7365 
7366 void zone_pcp_reset(struct zone *zone)
7367 {
7368         unsigned long flags;
7369         int cpu;
7370         struct per_cpu_pageset *pset;
7371 
7372         /* avoid races with drain_pages()  */
7373         local_irq_save(flags);
7374         if (zone->pageset != &boot_pageset) {
7375                 for_each_online_cpu(cpu) {
7376                         pset = per_cpu_ptr(zone->pageset, cpu);
7377                         drain_zonestat(zone, pset);
7378                 }
7379                 free_percpu(zone->pageset);
7380                 zone->pageset = &boot_pageset;
7381         }
7382         local_irq_restore(flags);
7383 }
7384 
7385 #ifdef CONFIG_MEMORY_HOTREMOVE
7386 /*
7387  * All pages in the range must be in a single zone and isolated
7388  * before calling this.
7389  */
7390 void
7391 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7392 {
7393         struct page *page;
7394         struct zone *zone;
7395         unsigned int order, i;
7396         unsigned long pfn;
7397         unsigned long flags;
7398         /* find the first valid pfn */
7399         for (pfn = start_pfn; pfn < end_pfn; pfn++)
7400                 if (pfn_valid(pfn))
7401                         break;
7402         if (pfn == end_pfn)
7403                 return;
7404         zone = page_zone(pfn_to_page(pfn));
7405         spin_lock_irqsave(&zone->lock, flags);
7406         pfn = start_pfn;
7407         while (pfn < end_pfn) {
7408                 if (!pfn_valid(pfn)) {
7409                         pfn++;
7410                         continue;
7411                 }
7412                 page = pfn_to_page(pfn);
7413                 /*
7414                  * The HWPoisoned page may be not in buddy system, and
7415                  * page_count() is not 0.
7416                  */
7417                 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7418                         pfn++;
7419                         SetPageReserved(page);
7420                         continue;
7421                 }
7422 
7423                 BUG_ON(page_count(page));
7424                 BUG_ON(!PageBuddy(page));
7425                 order = page_order(page);
7426 #ifdef CONFIG_DEBUG_VM
7427                 pr_info("remove from free list %lx %d %lx\n",
7428                         pfn, 1 << order, end_pfn);
7429 #endif
7430                 list_del(&page->lru);
7431                 rmv_page_order(page);
7432                 zone->free_area[order].nr_free--;
7433                 for (i = 0; i < (1 << order); i++)
7434                         SetPageReserved((page+i));
7435                 pfn += (1 << order);
7436         }
7437         spin_unlock_irqrestore(&zone->lock, flags);
7438 }
7439 #endif
7440 
7441 bool is_free_buddy_page(struct page *page)
7442 {
7443         struct zone *zone = page_zone(page);
7444         unsigned long pfn = page_to_pfn(page);
7445         unsigned long flags;
7446         unsigned int order;
7447 
7448         spin_lock_irqsave(&zone->lock, flags);
7449         for (order = 0; order < MAX_ORDER; order++) {
7450                 struct page *page_head = page - (pfn & ((1 << order) - 1));
7451 
7452                 if (PageBuddy(page_head) && page_order(page_head) >= order)
7453                         break;
7454         }
7455         spin_unlock_irqrestore(&zone->lock, flags);
7456 
7457         return order < MAX_ORDER;
7458 }
7459 

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