Version:  2.0.40 2.2.26 2.4.37 3.0 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16

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/module.h>
 29 #include <linux/suspend.h>
 30 #include <linux/pagevec.h>
 31 #include <linux/blkdev.h>
 32 #include <linux/slab.h>
 33 #include <linux/ratelimit.h>
 34 #include <linux/oom.h>
 35 #include <linux/notifier.h>
 36 #include <linux/topology.h>
 37 #include <linux/sysctl.h>
 38 #include <linux/cpu.h>
 39 #include <linux/cpuset.h>
 40 #include <linux/memory_hotplug.h>
 41 #include <linux/nodemask.h>
 42 #include <linux/vmalloc.h>
 43 #include <linux/vmstat.h>
 44 #include <linux/mempolicy.h>
 45 #include <linux/stop_machine.h>
 46 #include <linux/sort.h>
 47 #include <linux/pfn.h>
 48 #include <linux/backing-dev.h>
 49 #include <linux/fault-inject.h>
 50 #include <linux/page-isolation.h>
 51 #include <linux/page_cgroup.h>
 52 #include <linux/debugobjects.h>
 53 #include <linux/kmemleak.h>
 54 #include <linux/compaction.h>
 55 #include <trace/events/kmem.h>
 56 #include <linux/ftrace_event.h>
 57 #include <linux/memcontrol.h>
 58 #include <linux/prefetch.h>
 59 #include <linux/mm_inline.h>
 60 #include <linux/migrate.h>
 61 #include <linux/page-debug-flags.h>
 62 #include <linux/hugetlb.h>
 63 #include <linux/sched/rt.h>
 64 
 65 #include <asm/sections.h>
 66 #include <asm/tlbflush.h>
 67 #include <asm/div64.h>
 68 #include "internal.h"
 69 
 70 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
 71 static DEFINE_MUTEX(pcp_batch_high_lock);
 72 #define MIN_PERCPU_PAGELIST_FRACTION    (8)
 73 
 74 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
 75 DEFINE_PER_CPU(int, numa_node);
 76 EXPORT_PER_CPU_SYMBOL(numa_node);
 77 #endif
 78 
 79 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
 80 /*
 81  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
 82  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
 83  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
 84  * defined in <linux/topology.h>.
 85  */
 86 DEFINE_PER_CPU(int, _numa_mem_);                /* Kernel "local memory" node */
 87 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
 88 #endif
 89 
 90 /*
 91  * Array of node states.
 92  */
 93 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
 94         [N_POSSIBLE] = NODE_MASK_ALL,
 95         [N_ONLINE] = { { [0] = 1UL } },
 96 #ifndef CONFIG_NUMA
 97         [N_NORMAL_MEMORY] = { { [0] = 1UL } },
 98 #ifdef CONFIG_HIGHMEM
 99         [N_HIGH_MEMORY] = { { [0] = 1UL } },
100 #endif
101 #ifdef CONFIG_MOVABLE_NODE
102         [N_MEMORY] = { { [0] = 1UL } },
103 #endif
104         [N_CPU] = { { [0] = 1UL } },
105 #endif  /* NUMA */
106 };
107 EXPORT_SYMBOL(node_states);
108 
109 /* Protect totalram_pages and zone->managed_pages */
110 static DEFINE_SPINLOCK(managed_page_count_lock);
111 
112 unsigned long totalram_pages __read_mostly;
113 unsigned long totalreserve_pages __read_mostly;
114 /*
115  * When calculating the number of globally allowed dirty pages, there
116  * is a certain number of per-zone reserves that should not be
117  * considered dirtyable memory.  This is the sum of those reserves
118  * over all existing zones that contribute dirtyable memory.
119  */
120 unsigned long dirty_balance_reserve __read_mostly;
121 
122 int percpu_pagelist_fraction;
123 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
124 
125 #ifdef CONFIG_PM_SLEEP
126 /*
127  * The following functions are used by the suspend/hibernate code to temporarily
128  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
129  * while devices are suspended.  To avoid races with the suspend/hibernate code,
130  * they should always be called with pm_mutex held (gfp_allowed_mask also should
131  * only be modified with pm_mutex held, unless the suspend/hibernate code is
132  * guaranteed not to run in parallel with that modification).
133  */
134 
135 static gfp_t saved_gfp_mask;
136 
137 void pm_restore_gfp_mask(void)
138 {
139         WARN_ON(!mutex_is_locked(&pm_mutex));
140         if (saved_gfp_mask) {
141                 gfp_allowed_mask = saved_gfp_mask;
142                 saved_gfp_mask = 0;
143         }
144 }
145 
146 void pm_restrict_gfp_mask(void)
147 {
148         WARN_ON(!mutex_is_locked(&pm_mutex));
149         WARN_ON(saved_gfp_mask);
150         saved_gfp_mask = gfp_allowed_mask;
151         gfp_allowed_mask &= ~GFP_IOFS;
152 }
153 
154 bool pm_suspended_storage(void)
155 {
156         if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS)
157                 return false;
158         return true;
159 }
160 #endif /* CONFIG_PM_SLEEP */
161 
162 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
163 int pageblock_order __read_mostly;
164 #endif
165 
166 static void __free_pages_ok(struct page *page, unsigned int order);
167 
168 /*
169  * results with 256, 32 in the lowmem_reserve sysctl:
170  *      1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
171  *      1G machine -> (16M dma, 784M normal, 224M high)
172  *      NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
173  *      HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
174  *      HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
175  *
176  * TBD: should special case ZONE_DMA32 machines here - in those we normally
177  * don't need any ZONE_NORMAL reservation
178  */
179 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
180 #ifdef CONFIG_ZONE_DMA
181          256,
182 #endif
183 #ifdef CONFIG_ZONE_DMA32
184          256,
185 #endif
186 #ifdef CONFIG_HIGHMEM
187          32,
188 #endif
189          32,
190 };
191 
192 EXPORT_SYMBOL(totalram_pages);
193 
194 static char * const zone_names[MAX_NR_ZONES] = {
195 #ifdef CONFIG_ZONE_DMA
196          "DMA",
197 #endif
198 #ifdef CONFIG_ZONE_DMA32
199          "DMA32",
200 #endif
201          "Normal",
202 #ifdef CONFIG_HIGHMEM
203          "HighMem",
204 #endif
205          "Movable",
206 };
207 
208 int min_free_kbytes = 1024;
209 int user_min_free_kbytes = -1;
210 
211 static unsigned long __meminitdata nr_kernel_pages;
212 static unsigned long __meminitdata nr_all_pages;
213 static unsigned long __meminitdata dma_reserve;
214 
215 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
216 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
217 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
218 static unsigned long __initdata required_kernelcore;
219 static unsigned long __initdata required_movablecore;
220 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
221 
222 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
223 int movable_zone;
224 EXPORT_SYMBOL(movable_zone);
225 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
226 
227 #if MAX_NUMNODES > 1
228 int nr_node_ids __read_mostly = MAX_NUMNODES;
229 int nr_online_nodes __read_mostly = 1;
230 EXPORT_SYMBOL(nr_node_ids);
231 EXPORT_SYMBOL(nr_online_nodes);
232 #endif
233 
234 int page_group_by_mobility_disabled __read_mostly;
235 
236 void set_pageblock_migratetype(struct page *page, int migratetype)
237 {
238         if (unlikely(page_group_by_mobility_disabled &&
239                      migratetype < MIGRATE_PCPTYPES))
240                 migratetype = MIGRATE_UNMOVABLE;
241 
242         set_pageblock_flags_group(page, (unsigned long)migratetype,
243                                         PB_migrate, PB_migrate_end);
244 }
245 
246 bool oom_killer_disabled __read_mostly;
247 
248 #ifdef CONFIG_DEBUG_VM
249 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
250 {
251         int ret = 0;
252         unsigned seq;
253         unsigned long pfn = page_to_pfn(page);
254         unsigned long sp, start_pfn;
255 
256         do {
257                 seq = zone_span_seqbegin(zone);
258                 start_pfn = zone->zone_start_pfn;
259                 sp = zone->spanned_pages;
260                 if (!zone_spans_pfn(zone, pfn))
261                         ret = 1;
262         } while (zone_span_seqretry(zone, seq));
263 
264         if (ret)
265                 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
266                         pfn, zone_to_nid(zone), zone->name,
267                         start_pfn, start_pfn + sp);
268 
269         return ret;
270 }
271 
272 static int page_is_consistent(struct zone *zone, struct page *page)
273 {
274         if (!pfn_valid_within(page_to_pfn(page)))
275                 return 0;
276         if (zone != page_zone(page))
277                 return 0;
278 
279         return 1;
280 }
281 /*
282  * Temporary debugging check for pages not lying within a given zone.
283  */
284 static int bad_range(struct zone *zone, struct page *page)
285 {
286         if (page_outside_zone_boundaries(zone, page))
287                 return 1;
288         if (!page_is_consistent(zone, page))
289                 return 1;
290 
291         return 0;
292 }
293 #else
294 static inline int bad_range(struct zone *zone, struct page *page)
295 {
296         return 0;
297 }
298 #endif
299 
300 static void bad_page(struct page *page, const char *reason,
301                 unsigned long bad_flags)
302 {
303         static unsigned long resume;
304         static unsigned long nr_shown;
305         static unsigned long nr_unshown;
306 
307         /* Don't complain about poisoned pages */
308         if (PageHWPoison(page)) {
309                 page_mapcount_reset(page); /* remove PageBuddy */
310                 return;
311         }
312 
313         /*
314          * Allow a burst of 60 reports, then keep quiet for that minute;
315          * or allow a steady drip of one report per second.
316          */
317         if (nr_shown == 60) {
318                 if (time_before(jiffies, resume)) {
319                         nr_unshown++;
320                         goto out;
321                 }
322                 if (nr_unshown) {
323                         printk(KERN_ALERT
324                               "BUG: Bad page state: %lu messages suppressed\n",
325                                 nr_unshown);
326                         nr_unshown = 0;
327                 }
328                 nr_shown = 0;
329         }
330         if (nr_shown++ == 0)
331                 resume = jiffies + 60 * HZ;
332 
333         printk(KERN_ALERT "BUG: Bad page state in process %s  pfn:%05lx\n",
334                 current->comm, page_to_pfn(page));
335         dump_page_badflags(page, reason, bad_flags);
336 
337         print_modules();
338         dump_stack();
339 out:
340         /* Leave bad fields for debug, except PageBuddy could make trouble */
341         page_mapcount_reset(page); /* remove PageBuddy */
342         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
343 }
344 
345 /*
346  * Higher-order pages are called "compound pages".  They are structured thusly:
347  *
348  * The first PAGE_SIZE page is called the "head page".
349  *
350  * The remaining PAGE_SIZE pages are called "tail pages".
351  *
352  * All pages have PG_compound set.  All tail pages have their ->first_page
353  * pointing at the head page.
354  *
355  * The first tail page's ->lru.next holds the address of the compound page's
356  * put_page() function.  Its ->lru.prev holds the order of allocation.
357  * This usage means that zero-order pages may not be compound.
358  */
359 
360 static void free_compound_page(struct page *page)
361 {
362         __free_pages_ok(page, compound_order(page));
363 }
364 
365 void prep_compound_page(struct page *page, unsigned long order)
366 {
367         int i;
368         int nr_pages = 1 << order;
369 
370         set_compound_page_dtor(page, free_compound_page);
371         set_compound_order(page, order);
372         __SetPageHead(page);
373         for (i = 1; i < nr_pages; i++) {
374                 struct page *p = page + i;
375                 set_page_count(p, 0);
376                 p->first_page = page;
377                 /* Make sure p->first_page is always valid for PageTail() */
378                 smp_wmb();
379                 __SetPageTail(p);
380         }
381 }
382 
383 /* update __split_huge_page_refcount if you change this function */
384 static int destroy_compound_page(struct page *page, unsigned long order)
385 {
386         int i;
387         int nr_pages = 1 << order;
388         int bad = 0;
389 
390         if (unlikely(compound_order(page) != order)) {
391                 bad_page(page, "wrong compound order", 0);
392                 bad++;
393         }
394 
395         __ClearPageHead(page);
396 
397         for (i = 1; i < nr_pages; i++) {
398                 struct page *p = page + i;
399 
400                 if (unlikely(!PageTail(p))) {
401                         bad_page(page, "PageTail not set", 0);
402                         bad++;
403                 } else if (unlikely(p->first_page != page)) {
404                         bad_page(page, "first_page not consistent", 0);
405                         bad++;
406                 }
407                 __ClearPageTail(p);
408         }
409 
410         return bad;
411 }
412 
413 static inline void prep_zero_page(struct page *page, unsigned int order,
414                                                         gfp_t gfp_flags)
415 {
416         int i;
417 
418         /*
419          * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
420          * and __GFP_HIGHMEM from hard or soft interrupt context.
421          */
422         VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
423         for (i = 0; i < (1 << order); i++)
424                 clear_highpage(page + i);
425 }
426 
427 #ifdef CONFIG_DEBUG_PAGEALLOC
428 unsigned int _debug_guardpage_minorder;
429 
430 static int __init debug_guardpage_minorder_setup(char *buf)
431 {
432         unsigned long res;
433 
434         if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
435                 printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
436                 return 0;
437         }
438         _debug_guardpage_minorder = res;
439         printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
440         return 0;
441 }
442 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
443 
444 static inline void set_page_guard_flag(struct page *page)
445 {
446         __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
447 }
448 
449 static inline void clear_page_guard_flag(struct page *page)
450 {
451         __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
452 }
453 #else
454 static inline void set_page_guard_flag(struct page *page) { }
455 static inline void clear_page_guard_flag(struct page *page) { }
456 #endif
457 
458 static inline void set_page_order(struct page *page, unsigned int order)
459 {
460         set_page_private(page, order);
461         __SetPageBuddy(page);
462 }
463 
464 static inline void rmv_page_order(struct page *page)
465 {
466         __ClearPageBuddy(page);
467         set_page_private(page, 0);
468 }
469 
470 /*
471  * Locate the struct page for both the matching buddy in our
472  * pair (buddy1) and the combined O(n+1) page they form (page).
473  *
474  * 1) Any buddy B1 will have an order O twin B2 which satisfies
475  * the following equation:
476  *     B2 = B1 ^ (1 << O)
477  * For example, if the starting buddy (buddy2) is #8 its order
478  * 1 buddy is #10:
479  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
480  *
481  * 2) Any buddy B will have an order O+1 parent P which
482  * satisfies the following equation:
483  *     P = B & ~(1 << O)
484  *
485  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
486  */
487 static inline unsigned long
488 __find_buddy_index(unsigned long page_idx, unsigned int order)
489 {
490         return page_idx ^ (1 << order);
491 }
492 
493 /*
494  * This function checks whether a page is free && is the buddy
495  * we can do coalesce a page and its buddy if
496  * (a) the buddy is not in a hole &&
497  * (b) the buddy is in the buddy system &&
498  * (c) a page and its buddy have the same order &&
499  * (d) a page and its buddy are in the same zone.
500  *
501  * For recording whether a page is in the buddy system, we set ->_mapcount
502  * PAGE_BUDDY_MAPCOUNT_VALUE.
503  * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
504  * serialized by zone->lock.
505  *
506  * For recording page's order, we use page_private(page).
507  */
508 static inline int page_is_buddy(struct page *page, struct page *buddy,
509                                                         unsigned int order)
510 {
511         if (!pfn_valid_within(page_to_pfn(buddy)))
512                 return 0;
513 
514         if (page_is_guard(buddy) && page_order(buddy) == order) {
515                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
516 
517                 if (page_zone_id(page) != page_zone_id(buddy))
518                         return 0;
519 
520                 return 1;
521         }
522 
523         if (PageBuddy(buddy) && page_order(buddy) == order) {
524                 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
525 
526                 /*
527                  * zone check is done late to avoid uselessly
528                  * calculating zone/node ids for pages that could
529                  * never merge.
530                  */
531                 if (page_zone_id(page) != page_zone_id(buddy))
532                         return 0;
533 
534                 return 1;
535         }
536         return 0;
537 }
538 
539 /*
540  * Freeing function for a buddy system allocator.
541  *
542  * The concept of a buddy system is to maintain direct-mapped table
543  * (containing bit values) for memory blocks of various "orders".
544  * The bottom level table contains the map for the smallest allocatable
545  * units of memory (here, pages), and each level above it describes
546  * pairs of units from the levels below, hence, "buddies".
547  * At a high level, all that happens here is marking the table entry
548  * at the bottom level available, and propagating the changes upward
549  * as necessary, plus some accounting needed to play nicely with other
550  * parts of the VM system.
551  * At each level, we keep a list of pages, which are heads of continuous
552  * free pages of length of (1 << order) and marked with _mapcount
553  * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
554  * field.
555  * So when we are allocating or freeing one, we can derive the state of the
556  * other.  That is, if we allocate a small block, and both were
557  * free, the remainder of the region must be split into blocks.
558  * If a block is freed, and its buddy is also free, then this
559  * triggers coalescing into a block of larger size.
560  *
561  * -- nyc
562  */
563 
564 static inline void __free_one_page(struct page *page,
565                 unsigned long pfn,
566                 struct zone *zone, unsigned int order,
567                 int migratetype)
568 {
569         unsigned long page_idx;
570         unsigned long combined_idx;
571         unsigned long uninitialized_var(buddy_idx);
572         struct page *buddy;
573 
574         VM_BUG_ON(!zone_is_initialized(zone));
575 
576         if (unlikely(PageCompound(page)))
577                 if (unlikely(destroy_compound_page(page, order)))
578                         return;
579 
580         VM_BUG_ON(migratetype == -1);
581 
582         page_idx = pfn & ((1 << MAX_ORDER) - 1);
583 
584         VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
585         VM_BUG_ON_PAGE(bad_range(zone, page), page);
586 
587         while (order < MAX_ORDER-1) {
588                 buddy_idx = __find_buddy_index(page_idx, order);
589                 buddy = page + (buddy_idx - page_idx);
590                 if (!page_is_buddy(page, buddy, order))
591                         break;
592                 /*
593                  * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
594                  * merge with it and move up one order.
595                  */
596                 if (page_is_guard(buddy)) {
597                         clear_page_guard_flag(buddy);
598                         set_page_private(page, 0);
599                         __mod_zone_freepage_state(zone, 1 << order,
600                                                   migratetype);
601                 } else {
602                         list_del(&buddy->lru);
603                         zone->free_area[order].nr_free--;
604                         rmv_page_order(buddy);
605                 }
606                 combined_idx = buddy_idx & page_idx;
607                 page = page + (combined_idx - page_idx);
608                 page_idx = combined_idx;
609                 order++;
610         }
611         set_page_order(page, order);
612 
613         /*
614          * If this is not the largest possible page, check if the buddy
615          * of the next-highest order is free. If it is, it's possible
616          * that pages are being freed that will coalesce soon. In case,
617          * that is happening, add the free page to the tail of the list
618          * so it's less likely to be used soon and more likely to be merged
619          * as a higher order page
620          */
621         if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
622                 struct page *higher_page, *higher_buddy;
623                 combined_idx = buddy_idx & page_idx;
624                 higher_page = page + (combined_idx - page_idx);
625                 buddy_idx = __find_buddy_index(combined_idx, order + 1);
626                 higher_buddy = higher_page + (buddy_idx - combined_idx);
627                 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
628                         list_add_tail(&page->lru,
629                                 &zone->free_area[order].free_list[migratetype]);
630                         goto out;
631                 }
632         }
633 
634         list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
635 out:
636         zone->free_area[order].nr_free++;
637 }
638 
639 static inline int free_pages_check(struct page *page)
640 {
641         const char *bad_reason = NULL;
642         unsigned long bad_flags = 0;
643 
644         if (unlikely(page_mapcount(page)))
645                 bad_reason = "nonzero mapcount";
646         if (unlikely(page->mapping != NULL))
647                 bad_reason = "non-NULL mapping";
648         if (unlikely(atomic_read(&page->_count) != 0))
649                 bad_reason = "nonzero _count";
650         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
651                 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
652                 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
653         }
654         if (unlikely(mem_cgroup_bad_page_check(page)))
655                 bad_reason = "cgroup check failed";
656         if (unlikely(bad_reason)) {
657                 bad_page(page, bad_reason, bad_flags);
658                 return 1;
659         }
660         page_cpupid_reset_last(page);
661         if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
662                 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
663         return 0;
664 }
665 
666 /*
667  * Frees a number of pages from the PCP lists
668  * Assumes all pages on list are in same zone, and of same order.
669  * count is the number of pages to free.
670  *
671  * If the zone was previously in an "all pages pinned" state then look to
672  * see if this freeing clears that state.
673  *
674  * And clear the zone's pages_scanned counter, to hold off the "all pages are
675  * pinned" detection logic.
676  */
677 static void free_pcppages_bulk(struct zone *zone, int count,
678                                         struct per_cpu_pages *pcp)
679 {
680         int migratetype = 0;
681         int batch_free = 0;
682         int to_free = count;
683 
684         spin_lock(&zone->lock);
685         zone->pages_scanned = 0;
686 
687         while (to_free) {
688                 struct page *page;
689                 struct list_head *list;
690 
691                 /*
692                  * Remove pages from lists in a round-robin fashion. A
693                  * batch_free count is maintained that is incremented when an
694                  * empty list is encountered.  This is so more pages are freed
695                  * off fuller lists instead of spinning excessively around empty
696                  * lists
697                  */
698                 do {
699                         batch_free++;
700                         if (++migratetype == MIGRATE_PCPTYPES)
701                                 migratetype = 0;
702                         list = &pcp->lists[migratetype];
703                 } while (list_empty(list));
704 
705                 /* This is the only non-empty list. Free them all. */
706                 if (batch_free == MIGRATE_PCPTYPES)
707                         batch_free = to_free;
708 
709                 do {
710                         int mt; /* migratetype of the to-be-freed page */
711 
712                         page = list_entry(list->prev, struct page, lru);
713                         /* must delete as __free_one_page list manipulates */
714                         list_del(&page->lru);
715                         mt = get_freepage_migratetype(page);
716                         /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
717                         __free_one_page(page, page_to_pfn(page), zone, 0, mt);
718                         trace_mm_page_pcpu_drain(page, 0, mt);
719                         if (likely(!is_migrate_isolate_page(page))) {
720                                 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
721                                 if (is_migrate_cma(mt))
722                                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
723                         }
724                 } while (--to_free && --batch_free && !list_empty(list));
725         }
726         spin_unlock(&zone->lock);
727 }
728 
729 static void free_one_page(struct zone *zone,
730                                 struct page *page, unsigned long pfn,
731                                 unsigned int order,
732                                 int migratetype)
733 {
734         spin_lock(&zone->lock);
735         zone->pages_scanned = 0;
736 
737         __free_one_page(page, pfn, zone, order, migratetype);
738         if (unlikely(!is_migrate_isolate(migratetype)))
739                 __mod_zone_freepage_state(zone, 1 << order, migratetype);
740         spin_unlock(&zone->lock);
741 }
742 
743 static bool free_pages_prepare(struct page *page, unsigned int order)
744 {
745         int i;
746         int bad = 0;
747 
748         trace_mm_page_free(page, order);
749         kmemcheck_free_shadow(page, order);
750 
751         if (PageAnon(page))
752                 page->mapping = NULL;
753         for (i = 0; i < (1 << order); i++)
754                 bad += free_pages_check(page + i);
755         if (bad)
756                 return false;
757 
758         if (!PageHighMem(page)) {
759                 debug_check_no_locks_freed(page_address(page),
760                                            PAGE_SIZE << order);
761                 debug_check_no_obj_freed(page_address(page),
762                                            PAGE_SIZE << order);
763         }
764         arch_free_page(page, order);
765         kernel_map_pages(page, 1 << order, 0);
766 
767         return true;
768 }
769 
770 static void __free_pages_ok(struct page *page, unsigned int order)
771 {
772         unsigned long flags;
773         int migratetype;
774         unsigned long pfn = page_to_pfn(page);
775 
776         if (!free_pages_prepare(page, order))
777                 return;
778 
779         migratetype = get_pfnblock_migratetype(page, pfn);
780         local_irq_save(flags);
781         __count_vm_events(PGFREE, 1 << order);
782         set_freepage_migratetype(page, migratetype);
783         free_one_page(page_zone(page), page, pfn, order, migratetype);
784         local_irq_restore(flags);
785 }
786 
787 void __init __free_pages_bootmem(struct page *page, unsigned int order)
788 {
789         unsigned int nr_pages = 1 << order;
790         struct page *p = page;
791         unsigned int loop;
792 
793         prefetchw(p);
794         for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
795                 prefetchw(p + 1);
796                 __ClearPageReserved(p);
797                 set_page_count(p, 0);
798         }
799         __ClearPageReserved(p);
800         set_page_count(p, 0);
801 
802         page_zone(page)->managed_pages += nr_pages;
803         set_page_refcounted(page);
804         __free_pages(page, order);
805 }
806 
807 #ifdef CONFIG_CMA
808 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
809 void __init init_cma_reserved_pageblock(struct page *page)
810 {
811         unsigned i = pageblock_nr_pages;
812         struct page *p = page;
813 
814         do {
815                 __ClearPageReserved(p);
816                 set_page_count(p, 0);
817         } while (++p, --i);
818 
819         set_pageblock_migratetype(page, MIGRATE_CMA);
820 
821         if (pageblock_order >= MAX_ORDER) {
822                 i = pageblock_nr_pages;
823                 p = page;
824                 do {
825                         set_page_refcounted(p);
826                         __free_pages(p, MAX_ORDER - 1);
827                         p += MAX_ORDER_NR_PAGES;
828                 } while (i -= MAX_ORDER_NR_PAGES);
829         } else {
830                 set_page_refcounted(page);
831                 __free_pages(page, pageblock_order);
832         }
833 
834         adjust_managed_page_count(page, pageblock_nr_pages);
835 }
836 #endif
837 
838 /*
839  * The order of subdivision here is critical for the IO subsystem.
840  * Please do not alter this order without good reasons and regression
841  * testing. Specifically, as large blocks of memory are subdivided,
842  * the order in which smaller blocks are delivered depends on the order
843  * they're subdivided in this function. This is the primary factor
844  * influencing the order in which pages are delivered to the IO
845  * subsystem according to empirical testing, and this is also justified
846  * by considering the behavior of a buddy system containing a single
847  * large block of memory acted on by a series of small allocations.
848  * This behavior is a critical factor in sglist merging's success.
849  *
850  * -- nyc
851  */
852 static inline void expand(struct zone *zone, struct page *page,
853         int low, int high, struct free_area *area,
854         int migratetype)
855 {
856         unsigned long size = 1 << high;
857 
858         while (high > low) {
859                 area--;
860                 high--;
861                 size >>= 1;
862                 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
863 
864 #ifdef CONFIG_DEBUG_PAGEALLOC
865                 if (high < debug_guardpage_minorder()) {
866                         /*
867                          * Mark as guard pages (or page), that will allow to
868                          * merge back to allocator when buddy will be freed.
869                          * Corresponding page table entries will not be touched,
870                          * pages will stay not present in virtual address space
871                          */
872                         INIT_LIST_HEAD(&page[size].lru);
873                         set_page_guard_flag(&page[size]);
874                         set_page_private(&page[size], high);
875                         /* Guard pages are not available for any usage */
876                         __mod_zone_freepage_state(zone, -(1 << high),
877                                                   migratetype);
878                         continue;
879                 }
880 #endif
881                 list_add(&page[size].lru, &area->free_list[migratetype]);
882                 area->nr_free++;
883                 set_page_order(&page[size], high);
884         }
885 }
886 
887 /*
888  * This page is about to be returned from the page allocator
889  */
890 static inline int check_new_page(struct page *page)
891 {
892         const char *bad_reason = NULL;
893         unsigned long bad_flags = 0;
894 
895         if (unlikely(page_mapcount(page)))
896                 bad_reason = "nonzero mapcount";
897         if (unlikely(page->mapping != NULL))
898                 bad_reason = "non-NULL mapping";
899         if (unlikely(atomic_read(&page->_count) != 0))
900                 bad_reason = "nonzero _count";
901         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
902                 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
903                 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
904         }
905         if (unlikely(mem_cgroup_bad_page_check(page)))
906                 bad_reason = "cgroup check failed";
907         if (unlikely(bad_reason)) {
908                 bad_page(page, bad_reason, bad_flags);
909                 return 1;
910         }
911         return 0;
912 }
913 
914 static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags)
915 {
916         int i;
917 
918         for (i = 0; i < (1 << order); i++) {
919                 struct page *p = page + i;
920                 if (unlikely(check_new_page(p)))
921                         return 1;
922         }
923 
924         set_page_private(page, 0);
925         set_page_refcounted(page);
926 
927         arch_alloc_page(page, order);
928         kernel_map_pages(page, 1 << order, 1);
929 
930         if (gfp_flags & __GFP_ZERO)
931                 prep_zero_page(page, order, gfp_flags);
932 
933         if (order && (gfp_flags & __GFP_COMP))
934                 prep_compound_page(page, order);
935 
936         return 0;
937 }
938 
939 /*
940  * Go through the free lists for the given migratetype and remove
941  * the smallest available page from the freelists
942  */
943 static inline
944 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
945                                                 int migratetype)
946 {
947         unsigned int current_order;
948         struct free_area *area;
949         struct page *page;
950 
951         /* Find a page of the appropriate size in the preferred list */
952         for (current_order = order; current_order < MAX_ORDER; ++current_order) {
953                 area = &(zone->free_area[current_order]);
954                 if (list_empty(&area->free_list[migratetype]))
955                         continue;
956 
957                 page = list_entry(area->free_list[migratetype].next,
958                                                         struct page, lru);
959                 list_del(&page->lru);
960                 rmv_page_order(page);
961                 area->nr_free--;
962                 expand(zone, page, order, current_order, area, migratetype);
963                 set_freepage_migratetype(page, migratetype);
964                 return page;
965         }
966 
967         return NULL;
968 }
969 
970 
971 /*
972  * This array describes the order lists are fallen back to when
973  * the free lists for the desirable migrate type are depleted
974  */
975 static int fallbacks[MIGRATE_TYPES][4] = {
976         [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,     MIGRATE_RESERVE },
977         [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,     MIGRATE_RESERVE },
978 #ifdef CONFIG_CMA
979         [MIGRATE_MOVABLE]     = { MIGRATE_CMA,         MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
980         [MIGRATE_CMA]         = { MIGRATE_RESERVE }, /* Never used */
981 #else
982         [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,   MIGRATE_RESERVE },
983 #endif
984         [MIGRATE_RESERVE]     = { MIGRATE_RESERVE }, /* Never used */
985 #ifdef CONFIG_MEMORY_ISOLATION
986         [MIGRATE_ISOLATE]     = { MIGRATE_RESERVE }, /* Never used */
987 #endif
988 };
989 
990 /*
991  * Move the free pages in a range to the free lists of the requested type.
992  * Note that start_page and end_pages are not aligned on a pageblock
993  * boundary. If alignment is required, use move_freepages_block()
994  */
995 int move_freepages(struct zone *zone,
996                           struct page *start_page, struct page *end_page,
997                           int migratetype)
998 {
999         struct page *page;
1000         unsigned long order;
1001         int pages_moved = 0;
1002 
1003 #ifndef CONFIG_HOLES_IN_ZONE
1004         /*
1005          * page_zone is not safe to call in this context when
1006          * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1007          * anyway as we check zone boundaries in move_freepages_block().
1008          * Remove at a later date when no bug reports exist related to
1009          * grouping pages by mobility
1010          */
1011         BUG_ON(page_zone(start_page) != page_zone(end_page));
1012 #endif
1013 
1014         for (page = start_page; page <= end_page;) {
1015                 /* Make sure we are not inadvertently changing nodes */
1016                 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1017 
1018                 if (!pfn_valid_within(page_to_pfn(page))) {
1019                         page++;
1020                         continue;
1021                 }
1022 
1023                 if (!PageBuddy(page)) {
1024                         page++;
1025                         continue;
1026                 }
1027 
1028                 order = page_order(page);
1029                 list_move(&page->lru,
1030                           &zone->free_area[order].free_list[migratetype]);
1031                 set_freepage_migratetype(page, migratetype);
1032                 page += 1 << order;
1033                 pages_moved += 1 << order;
1034         }
1035 
1036         return pages_moved;
1037 }
1038 
1039 int move_freepages_block(struct zone *zone, struct page *page,
1040                                 int migratetype)
1041 {
1042         unsigned long start_pfn, end_pfn;
1043         struct page *start_page, *end_page;
1044 
1045         start_pfn = page_to_pfn(page);
1046         start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1047         start_page = pfn_to_page(start_pfn);
1048         end_page = start_page + pageblock_nr_pages - 1;
1049         end_pfn = start_pfn + pageblock_nr_pages - 1;
1050 
1051         /* Do not cross zone boundaries */
1052         if (!zone_spans_pfn(zone, start_pfn))
1053                 start_page = page;
1054         if (!zone_spans_pfn(zone, end_pfn))
1055                 return 0;
1056 
1057         return move_freepages(zone, start_page, end_page, migratetype);
1058 }
1059 
1060 static void change_pageblock_range(struct page *pageblock_page,
1061                                         int start_order, int migratetype)
1062 {
1063         int nr_pageblocks = 1 << (start_order - pageblock_order);
1064 
1065         while (nr_pageblocks--) {
1066                 set_pageblock_migratetype(pageblock_page, migratetype);
1067                 pageblock_page += pageblock_nr_pages;
1068         }
1069 }
1070 
1071 /*
1072  * If breaking a large block of pages, move all free pages to the preferred
1073  * allocation list. If falling back for a reclaimable kernel allocation, be
1074  * more aggressive about taking ownership of free pages.
1075  *
1076  * On the other hand, never change migration type of MIGRATE_CMA pageblocks
1077  * nor move CMA pages to different free lists. We don't want unmovable pages
1078  * to be allocated from MIGRATE_CMA areas.
1079  *
1080  * Returns the new migratetype of the pageblock (or the same old migratetype
1081  * if it was unchanged).
1082  */
1083 static int try_to_steal_freepages(struct zone *zone, struct page *page,
1084                                   int start_type, int fallback_type)
1085 {
1086         int current_order = page_order(page);
1087 
1088         /*
1089          * When borrowing from MIGRATE_CMA, we need to release the excess
1090          * buddy pages to CMA itself. We also ensure the freepage_migratetype
1091          * is set to CMA so it is returned to the correct freelist in case
1092          * the page ends up being not actually allocated from the pcp lists.
1093          */
1094         if (is_migrate_cma(fallback_type))
1095                 return fallback_type;
1096 
1097         /* Take ownership for orders >= pageblock_order */
1098         if (current_order >= pageblock_order) {
1099                 change_pageblock_range(page, current_order, start_type);
1100                 return start_type;
1101         }
1102 
1103         if (current_order >= pageblock_order / 2 ||
1104             start_type == MIGRATE_RECLAIMABLE ||
1105             page_group_by_mobility_disabled) {
1106                 int pages;
1107 
1108                 pages = move_freepages_block(zone, page, start_type);
1109 
1110                 /* Claim the whole block if over half of it is free */
1111                 if (pages >= (1 << (pageblock_order-1)) ||
1112                                 page_group_by_mobility_disabled) {
1113 
1114                         set_pageblock_migratetype(page, start_type);
1115                         return start_type;
1116                 }
1117 
1118         }
1119 
1120         return fallback_type;
1121 }
1122 
1123 /* Remove an element from the buddy allocator from the fallback list */
1124 static inline struct page *
1125 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
1126 {
1127         struct free_area *area;
1128         unsigned int current_order;
1129         struct page *page;
1130         int migratetype, new_type, i;
1131 
1132         /* Find the largest possible block of pages in the other list */
1133         for (current_order = MAX_ORDER-1;
1134                                 current_order >= order && current_order <= MAX_ORDER-1;
1135                                 --current_order) {
1136                 for (i = 0;; i++) {
1137                         migratetype = fallbacks[start_migratetype][i];
1138 
1139                         /* MIGRATE_RESERVE handled later if necessary */
1140                         if (migratetype == MIGRATE_RESERVE)
1141                                 break;
1142 
1143                         area = &(zone->free_area[current_order]);
1144                         if (list_empty(&area->free_list[migratetype]))
1145                                 continue;
1146 
1147                         page = list_entry(area->free_list[migratetype].next,
1148                                         struct page, lru);
1149                         area->nr_free--;
1150 
1151                         new_type = try_to_steal_freepages(zone, page,
1152                                                           start_migratetype,
1153                                                           migratetype);
1154 
1155                         /* Remove the page from the freelists */
1156                         list_del(&page->lru);
1157                         rmv_page_order(page);
1158 
1159                         expand(zone, page, order, current_order, area,
1160                                new_type);
1161                         /* The freepage_migratetype may differ from pageblock's
1162                          * migratetype depending on the decisions in
1163                          * try_to_steal_freepages. This is OK as long as it does
1164                          * not differ for MIGRATE_CMA type.
1165                          */
1166                         set_freepage_migratetype(page, new_type);
1167 
1168                         trace_mm_page_alloc_extfrag(page, order, current_order,
1169                                 start_migratetype, migratetype, new_type);
1170 
1171                         return page;
1172                 }
1173         }
1174 
1175         return NULL;
1176 }
1177 
1178 /*
1179  * Do the hard work of removing an element from the buddy allocator.
1180  * Call me with the zone->lock already held.
1181  */
1182 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1183                                                 int migratetype)
1184 {
1185         struct page *page;
1186 
1187 retry_reserve:
1188         page = __rmqueue_smallest(zone, order, migratetype);
1189 
1190         if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1191                 page = __rmqueue_fallback(zone, order, migratetype);
1192 
1193                 /*
1194                  * Use MIGRATE_RESERVE rather than fail an allocation. goto
1195                  * is used because __rmqueue_smallest is an inline function
1196                  * and we want just one call site
1197                  */
1198                 if (!page) {
1199                         migratetype = MIGRATE_RESERVE;
1200                         goto retry_reserve;
1201                 }
1202         }
1203 
1204         trace_mm_page_alloc_zone_locked(page, order, migratetype);
1205         return page;
1206 }
1207 
1208 /*
1209  * Obtain a specified number of elements from the buddy allocator, all under
1210  * a single hold of the lock, for efficiency.  Add them to the supplied list.
1211  * Returns the number of new pages which were placed at *list.
1212  */
1213 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1214                         unsigned long count, struct list_head *list,
1215                         int migratetype, bool cold)
1216 {
1217         int i;
1218 
1219         spin_lock(&zone->lock);
1220         for (i = 0; i < count; ++i) {
1221                 struct page *page = __rmqueue(zone, order, migratetype);
1222                 if (unlikely(page == NULL))
1223                         break;
1224 
1225                 /*
1226                  * Split buddy pages returned by expand() are received here
1227                  * in physical page order. The page is added to the callers and
1228                  * list and the list head then moves forward. From the callers
1229                  * perspective, the linked list is ordered by page number in
1230                  * some conditions. This is useful for IO devices that can
1231                  * merge IO requests if the physical pages are ordered
1232                  * properly.
1233                  */
1234                 if (likely(!cold))
1235                         list_add(&page->lru, list);
1236                 else
1237                         list_add_tail(&page->lru, list);
1238                 list = &page->lru;
1239                 if (is_migrate_cma(get_freepage_migratetype(page)))
1240                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1241                                               -(1 << order));
1242         }
1243         __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1244         spin_unlock(&zone->lock);
1245         return i;
1246 }
1247 
1248 #ifdef CONFIG_NUMA
1249 /*
1250  * Called from the vmstat counter updater to drain pagesets of this
1251  * currently executing processor on remote nodes after they have
1252  * expired.
1253  *
1254  * Note that this function must be called with the thread pinned to
1255  * a single processor.
1256  */
1257 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1258 {
1259         unsigned long flags;
1260         int to_drain;
1261         unsigned long batch;
1262 
1263         local_irq_save(flags);
1264         batch = ACCESS_ONCE(pcp->batch);
1265         if (pcp->count >= batch)
1266                 to_drain = batch;
1267         else
1268                 to_drain = pcp->count;
1269         if (to_drain > 0) {
1270                 free_pcppages_bulk(zone, to_drain, pcp);
1271                 pcp->count -= to_drain;
1272         }
1273         local_irq_restore(flags);
1274 }
1275 #endif
1276 
1277 /*
1278  * Drain pages of the indicated processor.
1279  *
1280  * The processor must either be the current processor and the
1281  * thread pinned to the current processor or a processor that
1282  * is not online.
1283  */
1284 static void drain_pages(unsigned int cpu)
1285 {
1286         unsigned long flags;
1287         struct zone *zone;
1288 
1289         for_each_populated_zone(zone) {
1290                 struct per_cpu_pageset *pset;
1291                 struct per_cpu_pages *pcp;
1292 
1293                 local_irq_save(flags);
1294                 pset = per_cpu_ptr(zone->pageset, cpu);
1295 
1296                 pcp = &pset->pcp;
1297                 if (pcp->count) {
1298                         free_pcppages_bulk(zone, pcp->count, pcp);
1299                         pcp->count = 0;
1300                 }
1301                 local_irq_restore(flags);
1302         }
1303 }
1304 
1305 /*
1306  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1307  */
1308 void drain_local_pages(void *arg)
1309 {
1310         drain_pages(smp_processor_id());
1311 }
1312 
1313 /*
1314  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1315  *
1316  * Note that this code is protected against sending an IPI to an offline
1317  * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1318  * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1319  * nothing keeps CPUs from showing up after we populated the cpumask and
1320  * before the call to on_each_cpu_mask().
1321  */
1322 void drain_all_pages(void)
1323 {
1324         int cpu;
1325         struct per_cpu_pageset *pcp;
1326         struct zone *zone;
1327 
1328         /*
1329          * Allocate in the BSS so we wont require allocation in
1330          * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1331          */
1332         static cpumask_t cpus_with_pcps;
1333 
1334         /*
1335          * We don't care about racing with CPU hotplug event
1336          * as offline notification will cause the notified
1337          * cpu to drain that CPU pcps and on_each_cpu_mask
1338          * disables preemption as part of its processing
1339          */
1340         for_each_online_cpu(cpu) {
1341                 bool has_pcps = false;
1342                 for_each_populated_zone(zone) {
1343                         pcp = per_cpu_ptr(zone->pageset, cpu);
1344                         if (pcp->pcp.count) {
1345                                 has_pcps = true;
1346                                 break;
1347                         }
1348                 }
1349                 if (has_pcps)
1350                         cpumask_set_cpu(cpu, &cpus_with_pcps);
1351                 else
1352                         cpumask_clear_cpu(cpu, &cpus_with_pcps);
1353         }
1354         on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
1355 }
1356 
1357 #ifdef CONFIG_HIBERNATION
1358 
1359 void mark_free_pages(struct zone *zone)
1360 {
1361         unsigned long pfn, max_zone_pfn;
1362         unsigned long flags;
1363         unsigned int order, t;
1364         struct list_head *curr;
1365 
1366         if (zone_is_empty(zone))
1367                 return;
1368 
1369         spin_lock_irqsave(&zone->lock, flags);
1370 
1371         max_zone_pfn = zone_end_pfn(zone);
1372         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1373                 if (pfn_valid(pfn)) {
1374                         struct page *page = pfn_to_page(pfn);
1375 
1376                         if (!swsusp_page_is_forbidden(page))
1377                                 swsusp_unset_page_free(page);
1378                 }
1379 
1380         for_each_migratetype_order(order, t) {
1381                 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1382                         unsigned long i;
1383 
1384                         pfn = page_to_pfn(list_entry(curr, struct page, lru));
1385                         for (i = 0; i < (1UL << order); i++)
1386                                 swsusp_set_page_free(pfn_to_page(pfn + i));
1387                 }
1388         }
1389         spin_unlock_irqrestore(&zone->lock, flags);
1390 }
1391 #endif /* CONFIG_PM */
1392 
1393 /*
1394  * Free a 0-order page
1395  * cold == true ? free a cold page : free a hot page
1396  */
1397 void free_hot_cold_page(struct page *page, bool cold)
1398 {
1399         struct zone *zone = page_zone(page);
1400         struct per_cpu_pages *pcp;
1401         unsigned long flags;
1402         unsigned long pfn = page_to_pfn(page);
1403         int migratetype;
1404 
1405         if (!free_pages_prepare(page, 0))
1406                 return;
1407 
1408         migratetype = get_pfnblock_migratetype(page, pfn);
1409         set_freepage_migratetype(page, migratetype);
1410         local_irq_save(flags);
1411         __count_vm_event(PGFREE);
1412 
1413         /*
1414          * We only track unmovable, reclaimable and movable on pcp lists.
1415          * Free ISOLATE pages back to the allocator because they are being
1416          * offlined but treat RESERVE as movable pages so we can get those
1417          * areas back if necessary. Otherwise, we may have to free
1418          * excessively into the page allocator
1419          */
1420         if (migratetype >= MIGRATE_PCPTYPES) {
1421                 if (unlikely(is_migrate_isolate(migratetype))) {
1422                         free_one_page(zone, page, pfn, 0, migratetype);
1423                         goto out;
1424                 }
1425                 migratetype = MIGRATE_MOVABLE;
1426         }
1427 
1428         pcp = &this_cpu_ptr(zone->pageset)->pcp;
1429         if (!cold)
1430                 list_add(&page->lru, &pcp->lists[migratetype]);
1431         else
1432                 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1433         pcp->count++;
1434         if (pcp->count >= pcp->high) {
1435                 unsigned long batch = ACCESS_ONCE(pcp->batch);
1436                 free_pcppages_bulk(zone, batch, pcp);
1437                 pcp->count -= batch;
1438         }
1439 
1440 out:
1441         local_irq_restore(flags);
1442 }
1443 
1444 /*
1445  * Free a list of 0-order pages
1446  */
1447 void free_hot_cold_page_list(struct list_head *list, bool cold)
1448 {
1449         struct page *page, *next;
1450 
1451         list_for_each_entry_safe(page, next, list, lru) {
1452                 trace_mm_page_free_batched(page, cold);
1453                 free_hot_cold_page(page, cold);
1454         }
1455 }
1456 
1457 /*
1458  * split_page takes a non-compound higher-order page, and splits it into
1459  * n (1<<order) sub-pages: page[0..n]
1460  * Each sub-page must be freed individually.
1461  *
1462  * Note: this is probably too low level an operation for use in drivers.
1463  * Please consult with lkml before using this in your driver.
1464  */
1465 void split_page(struct page *page, unsigned int order)
1466 {
1467         int i;
1468 
1469         VM_BUG_ON_PAGE(PageCompound(page), page);
1470         VM_BUG_ON_PAGE(!page_count(page), page);
1471 
1472 #ifdef CONFIG_KMEMCHECK
1473         /*
1474          * Split shadow pages too, because free(page[0]) would
1475          * otherwise free the whole shadow.
1476          */
1477         if (kmemcheck_page_is_tracked(page))
1478                 split_page(virt_to_page(page[0].shadow), order);
1479 #endif
1480 
1481         for (i = 1; i < (1 << order); i++)
1482                 set_page_refcounted(page + i);
1483 }
1484 EXPORT_SYMBOL_GPL(split_page);
1485 
1486 static int __isolate_free_page(struct page *page, unsigned int order)
1487 {
1488         unsigned long watermark;
1489         struct zone *zone;
1490         int mt;
1491 
1492         BUG_ON(!PageBuddy(page));
1493 
1494         zone = page_zone(page);
1495         mt = get_pageblock_migratetype(page);
1496 
1497         if (!is_migrate_isolate(mt)) {
1498                 /* Obey watermarks as if the page was being allocated */
1499                 watermark = low_wmark_pages(zone) + (1 << order);
1500                 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1501                         return 0;
1502 
1503                 __mod_zone_freepage_state(zone, -(1UL << order), mt);
1504         }
1505 
1506         /* Remove page from free list */
1507         list_del(&page->lru);
1508         zone->free_area[order].nr_free--;
1509         rmv_page_order(page);
1510 
1511         /* Set the pageblock if the isolated page is at least a pageblock */
1512         if (order >= pageblock_order - 1) {
1513                 struct page *endpage = page + (1 << order) - 1;
1514                 for (; page < endpage; page += pageblock_nr_pages) {
1515                         int mt = get_pageblock_migratetype(page);
1516                         if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
1517                                 set_pageblock_migratetype(page,
1518                                                           MIGRATE_MOVABLE);
1519                 }
1520         }
1521 
1522         return 1UL << order;
1523 }
1524 
1525 /*
1526  * Similar to split_page except the page is already free. As this is only
1527  * being used for migration, the migratetype of the block also changes.
1528  * As this is called with interrupts disabled, the caller is responsible
1529  * for calling arch_alloc_page() and kernel_map_page() after interrupts
1530  * are enabled.
1531  *
1532  * Note: this is probably too low level an operation for use in drivers.
1533  * Please consult with lkml before using this in your driver.
1534  */
1535 int split_free_page(struct page *page)
1536 {
1537         unsigned int order;
1538         int nr_pages;
1539 
1540         order = page_order(page);
1541 
1542         nr_pages = __isolate_free_page(page, order);
1543         if (!nr_pages)
1544                 return 0;
1545 
1546         /* Split into individual pages */
1547         set_page_refcounted(page);
1548         split_page(page, order);
1549         return nr_pages;
1550 }
1551 
1552 /*
1553  * Really, prep_compound_page() should be called from __rmqueue_bulk().  But
1554  * we cheat by calling it from here, in the order > 0 path.  Saves a branch
1555  * or two.
1556  */
1557 static inline
1558 struct page *buffered_rmqueue(struct zone *preferred_zone,
1559                         struct zone *zone, unsigned int order,
1560                         gfp_t gfp_flags, int migratetype)
1561 {
1562         unsigned long flags;
1563         struct page *page;
1564         bool cold = ((gfp_flags & __GFP_COLD) != 0);
1565 
1566 again:
1567         if (likely(order == 0)) {
1568                 struct per_cpu_pages *pcp;
1569                 struct list_head *list;
1570 
1571                 local_irq_save(flags);
1572                 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1573                 list = &pcp->lists[migratetype];
1574                 if (list_empty(list)) {
1575                         pcp->count += rmqueue_bulk(zone, 0,
1576                                         pcp->batch, list,
1577                                         migratetype, cold);
1578                         if (unlikely(list_empty(list)))
1579                                 goto failed;
1580                 }
1581 
1582                 if (cold)
1583                         page = list_entry(list->prev, struct page, lru);
1584                 else
1585                         page = list_entry(list->next, struct page, lru);
1586 
1587                 list_del(&page->lru);
1588                 pcp->count--;
1589         } else {
1590                 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1591                         /*
1592                          * __GFP_NOFAIL is not to be used in new code.
1593                          *
1594                          * All __GFP_NOFAIL callers should be fixed so that they
1595                          * properly detect and handle allocation failures.
1596                          *
1597                          * We most definitely don't want callers attempting to
1598                          * allocate greater than order-1 page units with
1599                          * __GFP_NOFAIL.
1600                          */
1601                         WARN_ON_ONCE(order > 1);
1602                 }
1603                 spin_lock_irqsave(&zone->lock, flags);
1604                 page = __rmqueue(zone, order, migratetype);
1605                 spin_unlock(&zone->lock);
1606                 if (!page)
1607                         goto failed;
1608                 __mod_zone_freepage_state(zone, -(1 << order),
1609                                           get_freepage_migratetype(page));
1610         }
1611 
1612         __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
1613 
1614         __count_zone_vm_events(PGALLOC, zone, 1 << order);
1615         zone_statistics(preferred_zone, zone, gfp_flags);
1616         local_irq_restore(flags);
1617 
1618         VM_BUG_ON_PAGE(bad_range(zone, page), page);
1619         if (prep_new_page(page, order, gfp_flags))
1620                 goto again;
1621         return page;
1622 
1623 failed:
1624         local_irq_restore(flags);
1625         return NULL;
1626 }
1627 
1628 #ifdef CONFIG_FAIL_PAGE_ALLOC
1629 
1630 static struct {
1631         struct fault_attr attr;
1632 
1633         u32 ignore_gfp_highmem;
1634         u32 ignore_gfp_wait;
1635         u32 min_order;
1636 } fail_page_alloc = {
1637         .attr = FAULT_ATTR_INITIALIZER,
1638         .ignore_gfp_wait = 1,
1639         .ignore_gfp_highmem = 1,
1640         .min_order = 1,
1641 };
1642 
1643 static int __init setup_fail_page_alloc(char *str)
1644 {
1645         return setup_fault_attr(&fail_page_alloc.attr, str);
1646 }
1647 __setup("fail_page_alloc=", setup_fail_page_alloc);
1648 
1649 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1650 {
1651         if (order < fail_page_alloc.min_order)
1652                 return false;
1653         if (gfp_mask & __GFP_NOFAIL)
1654                 return false;
1655         if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1656                 return false;
1657         if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1658                 return false;
1659 
1660         return should_fail(&fail_page_alloc.attr, 1 << order);
1661 }
1662 
1663 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1664 
1665 static int __init fail_page_alloc_debugfs(void)
1666 {
1667         umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1668         struct dentry *dir;
1669 
1670         dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1671                                         &fail_page_alloc.attr);
1672         if (IS_ERR(dir))
1673                 return PTR_ERR(dir);
1674 
1675         if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1676                                 &fail_page_alloc.ignore_gfp_wait))
1677                 goto fail;
1678         if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1679                                 &fail_page_alloc.ignore_gfp_highmem))
1680                 goto fail;
1681         if (!debugfs_create_u32("min-order", mode, dir,
1682                                 &fail_page_alloc.min_order))
1683                 goto fail;
1684 
1685         return 0;
1686 fail:
1687         debugfs_remove_recursive(dir);
1688 
1689         return -ENOMEM;
1690 }
1691 
1692 late_initcall(fail_page_alloc_debugfs);
1693 
1694 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1695 
1696 #else /* CONFIG_FAIL_PAGE_ALLOC */
1697 
1698 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1699 {
1700         return false;
1701 }
1702 
1703 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1704 
1705 /*
1706  * Return true if free pages are above 'mark'. This takes into account the order
1707  * of the allocation.
1708  */
1709 static bool __zone_watermark_ok(struct zone *z, unsigned int order,
1710                         unsigned long mark, int classzone_idx, int alloc_flags,
1711                         long free_pages)
1712 {
1713         /* free_pages my go negative - that's OK */
1714         long min = mark;
1715         long lowmem_reserve = z->lowmem_reserve[classzone_idx];
1716         int o;
1717         long free_cma = 0;
1718 
1719         free_pages -= (1 << order) - 1;
1720         if (alloc_flags & ALLOC_HIGH)
1721                 min -= min / 2;
1722         if (alloc_flags & ALLOC_HARDER)
1723                 min -= min / 4;
1724 #ifdef CONFIG_CMA
1725         /* If allocation can't use CMA areas don't use free CMA pages */
1726         if (!(alloc_flags & ALLOC_CMA))
1727                 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
1728 #endif
1729 
1730         if (free_pages - free_cma <= min + lowmem_reserve)
1731                 return false;
1732         for (o = 0; o < order; o++) {
1733                 /* At the next order, this order's pages become unavailable */
1734                 free_pages -= z->free_area[o].nr_free << o;
1735 
1736                 /* Require fewer higher order pages to be free */
1737                 min >>= 1;
1738 
1739                 if (free_pages <= min)
1740                         return false;
1741         }
1742         return true;
1743 }
1744 
1745 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1746                       int classzone_idx, int alloc_flags)
1747 {
1748         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1749                                         zone_page_state(z, NR_FREE_PAGES));
1750 }
1751 
1752 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1753                         unsigned long mark, int classzone_idx, int alloc_flags)
1754 {
1755         long free_pages = zone_page_state(z, NR_FREE_PAGES);
1756 
1757         if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1758                 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1759 
1760         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1761                                                                 free_pages);
1762 }
1763 
1764 #ifdef CONFIG_NUMA
1765 /*
1766  * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
1767  * skip over zones that are not allowed by the cpuset, or that have
1768  * been recently (in last second) found to be nearly full.  See further
1769  * comments in mmzone.h.  Reduces cache footprint of zonelist scans
1770  * that have to skip over a lot of full or unallowed zones.
1771  *
1772  * If the zonelist cache is present in the passed zonelist, then
1773  * returns a pointer to the allowed node mask (either the current
1774  * tasks mems_allowed, or node_states[N_MEMORY].)
1775  *
1776  * If the zonelist cache is not available for this zonelist, does
1777  * nothing and returns NULL.
1778  *
1779  * If the fullzones BITMAP in the zonelist cache is stale (more than
1780  * a second since last zap'd) then we zap it out (clear its bits.)
1781  *
1782  * We hold off even calling zlc_setup, until after we've checked the
1783  * first zone in the zonelist, on the theory that most allocations will
1784  * be satisfied from that first zone, so best to examine that zone as
1785  * quickly as we can.
1786  */
1787 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1788 {
1789         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1790         nodemask_t *allowednodes;       /* zonelist_cache approximation */
1791 
1792         zlc = zonelist->zlcache_ptr;
1793         if (!zlc)
1794                 return NULL;
1795 
1796         if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1797                 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1798                 zlc->last_full_zap = jiffies;
1799         }
1800 
1801         allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1802                                         &cpuset_current_mems_allowed :
1803                                         &node_states[N_MEMORY];
1804         return allowednodes;
1805 }
1806 
1807 /*
1808  * Given 'z' scanning a zonelist, run a couple of quick checks to see
1809  * if it is worth looking at further for free memory:
1810  *  1) Check that the zone isn't thought to be full (doesn't have its
1811  *     bit set in the zonelist_cache fullzones BITMAP).
1812  *  2) Check that the zones node (obtained from the zonelist_cache
1813  *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1814  * Return true (non-zero) if zone is worth looking at further, or
1815  * else return false (zero) if it is not.
1816  *
1817  * This check -ignores- the distinction between various watermarks,
1818  * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
1819  * found to be full for any variation of these watermarks, it will
1820  * be considered full for up to one second by all requests, unless
1821  * we are so low on memory on all allowed nodes that we are forced
1822  * into the second scan of the zonelist.
1823  *
1824  * In the second scan we ignore this zonelist cache and exactly
1825  * apply the watermarks to all zones, even it is slower to do so.
1826  * We are low on memory in the second scan, and should leave no stone
1827  * unturned looking for a free page.
1828  */
1829 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1830                                                 nodemask_t *allowednodes)
1831 {
1832         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1833         int i;                          /* index of *z in zonelist zones */
1834         int n;                          /* node that zone *z is on */
1835 
1836         zlc = zonelist->zlcache_ptr;
1837         if (!zlc)
1838                 return 1;
1839 
1840         i = z - zonelist->_zonerefs;
1841         n = zlc->z_to_n[i];
1842 
1843         /* This zone is worth trying if it is allowed but not full */
1844         return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1845 }
1846 
1847 /*
1848  * Given 'z' scanning a zonelist, set the corresponding bit in
1849  * zlc->fullzones, so that subsequent attempts to allocate a page
1850  * from that zone don't waste time re-examining it.
1851  */
1852 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1853 {
1854         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1855         int i;                          /* index of *z in zonelist zones */
1856 
1857         zlc = zonelist->zlcache_ptr;
1858         if (!zlc)
1859                 return;
1860 
1861         i = z - zonelist->_zonerefs;
1862 
1863         set_bit(i, zlc->fullzones);
1864 }
1865 
1866 /*
1867  * clear all zones full, called after direct reclaim makes progress so that
1868  * a zone that was recently full is not skipped over for up to a second
1869  */
1870 static void zlc_clear_zones_full(struct zonelist *zonelist)
1871 {
1872         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1873 
1874         zlc = zonelist->zlcache_ptr;
1875         if (!zlc)
1876                 return;
1877 
1878         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1879 }
1880 
1881 static bool zone_local(struct zone *local_zone, struct zone *zone)
1882 {
1883         return local_zone->node == zone->node;
1884 }
1885 
1886 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1887 {
1888         return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
1889                                 RECLAIM_DISTANCE;
1890 }
1891 
1892 #else   /* CONFIG_NUMA */
1893 
1894 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1895 {
1896         return NULL;
1897 }
1898 
1899 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1900                                 nodemask_t *allowednodes)
1901 {
1902         return 1;
1903 }
1904 
1905 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1906 {
1907 }
1908 
1909 static void zlc_clear_zones_full(struct zonelist *zonelist)
1910 {
1911 }
1912 
1913 static bool zone_local(struct zone *local_zone, struct zone *zone)
1914 {
1915         return true;
1916 }
1917 
1918 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1919 {
1920         return true;
1921 }
1922 
1923 #endif  /* CONFIG_NUMA */
1924 
1925 /*
1926  * get_page_from_freelist goes through the zonelist trying to allocate
1927  * a page.
1928  */
1929 static struct page *
1930 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1931                 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1932                 struct zone *preferred_zone, int classzone_idx, int migratetype)
1933 {
1934         struct zoneref *z;
1935         struct page *page = NULL;
1936         struct zone *zone;
1937         nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1938         int zlc_active = 0;             /* set if using zonelist_cache */
1939         int did_zlc_setup = 0;          /* just call zlc_setup() one time */
1940         bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
1941                                 (gfp_mask & __GFP_WRITE);
1942 
1943 zonelist_scan:
1944         /*
1945          * Scan zonelist, looking for a zone with enough free.
1946          * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c.
1947          */
1948         for_each_zone_zonelist_nodemask(zone, z, zonelist,
1949                                                 high_zoneidx, nodemask) {
1950                 unsigned long mark;
1951 
1952                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
1953                         !zlc_zone_worth_trying(zonelist, z, allowednodes))
1954                                 continue;
1955                 if (cpusets_enabled() &&
1956                         (alloc_flags & ALLOC_CPUSET) &&
1957                         !cpuset_zone_allowed_softwall(zone, gfp_mask))
1958                                 continue;
1959                 /*
1960                  * Distribute pages in proportion to the individual
1961                  * zone size to ensure fair page aging.  The zone a
1962                  * page was allocated in should have no effect on the
1963                  * time the page has in memory before being reclaimed.
1964                  */
1965                 if (alloc_flags & ALLOC_FAIR) {
1966                         if (!zone_local(preferred_zone, zone))
1967                                 continue;
1968                         if (zone_page_state(zone, NR_ALLOC_BATCH) <= 0)
1969                                 continue;
1970                 }
1971                 /*
1972                  * When allocating a page cache page for writing, we
1973                  * want to get it from a zone that is within its dirty
1974                  * limit, such that no single zone holds more than its
1975                  * proportional share of globally allowed dirty pages.
1976                  * The dirty limits take into account the zone's
1977                  * lowmem reserves and high watermark so that kswapd
1978                  * should be able to balance it without having to
1979                  * write pages from its LRU list.
1980                  *
1981                  * This may look like it could increase pressure on
1982                  * lower zones by failing allocations in higher zones
1983                  * before they are full.  But the pages that do spill
1984                  * over are limited as the lower zones are protected
1985                  * by this very same mechanism.  It should not become
1986                  * a practical burden to them.
1987                  *
1988                  * XXX: For now, allow allocations to potentially
1989                  * exceed the per-zone dirty limit in the slowpath
1990                  * (ALLOC_WMARK_LOW unset) before going into reclaim,
1991                  * which is important when on a NUMA setup the allowed
1992                  * zones are together not big enough to reach the
1993                  * global limit.  The proper fix for these situations
1994                  * will require awareness of zones in the
1995                  * dirty-throttling and the flusher threads.
1996                  */
1997                 if (consider_zone_dirty && !zone_dirty_ok(zone))
1998                         continue;
1999 
2000                 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2001                 if (!zone_watermark_ok(zone, order, mark,
2002                                        classzone_idx, alloc_flags)) {
2003                         int ret;
2004 
2005                         /* Checked here to keep the fast path fast */
2006                         BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2007                         if (alloc_flags & ALLOC_NO_WATERMARKS)
2008                                 goto try_this_zone;
2009 
2010                         if (IS_ENABLED(CONFIG_NUMA) &&
2011                                         !did_zlc_setup && nr_online_nodes > 1) {
2012                                 /*
2013                                  * we do zlc_setup if there are multiple nodes
2014                                  * and before considering the first zone allowed
2015                                  * by the cpuset.
2016                                  */
2017                                 allowednodes = zlc_setup(zonelist, alloc_flags);
2018                                 zlc_active = 1;
2019                                 did_zlc_setup = 1;
2020                         }
2021 
2022                         if (zone_reclaim_mode == 0 ||
2023                             !zone_allows_reclaim(preferred_zone, zone))
2024                                 goto this_zone_full;
2025 
2026                         /*
2027                          * As we may have just activated ZLC, check if the first
2028                          * eligible zone has failed zone_reclaim recently.
2029                          */
2030                         if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2031                                 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2032                                 continue;
2033 
2034                         ret = zone_reclaim(zone, gfp_mask, order);
2035                         switch (ret) {
2036                         case ZONE_RECLAIM_NOSCAN:
2037                                 /* did not scan */
2038                                 continue;
2039                         case ZONE_RECLAIM_FULL:
2040                                 /* scanned but unreclaimable */
2041                                 continue;
2042                         default:
2043                                 /* did we reclaim enough */
2044                                 if (zone_watermark_ok(zone, order, mark,
2045                                                 classzone_idx, alloc_flags))
2046                                         goto try_this_zone;
2047 
2048                                 /*
2049                                  * Failed to reclaim enough to meet watermark.
2050                                  * Only mark the zone full if checking the min
2051                                  * watermark or if we failed to reclaim just
2052                                  * 1<<order pages or else the page allocator
2053                                  * fastpath will prematurely mark zones full
2054                                  * when the watermark is between the low and
2055                                  * min watermarks.
2056                                  */
2057                                 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2058                                     ret == ZONE_RECLAIM_SOME)
2059                                         goto this_zone_full;
2060 
2061                                 continue;
2062                         }
2063                 }
2064 
2065 try_this_zone:
2066                 page = buffered_rmqueue(preferred_zone, zone, order,
2067                                                 gfp_mask, migratetype);
2068                 if (page)
2069                         break;
2070 this_zone_full:
2071                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
2072                         zlc_mark_zone_full(zonelist, z);
2073         }
2074 
2075         if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) {
2076                 /* Disable zlc cache for second zonelist scan */
2077                 zlc_active = 0;
2078                 goto zonelist_scan;
2079         }
2080 
2081         if (page)
2082                 /*
2083                  * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2084                  * necessary to allocate the page. The expectation is
2085                  * that the caller is taking steps that will free more
2086                  * memory. The caller should avoid the page being used
2087                  * for !PFMEMALLOC purposes.
2088                  */
2089                 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
2090 
2091         return page;
2092 }
2093 
2094 /*
2095  * Large machines with many possible nodes should not always dump per-node
2096  * meminfo in irq context.
2097  */
2098 static inline bool should_suppress_show_mem(void)
2099 {
2100         bool ret = false;
2101 
2102 #if NODES_SHIFT > 8
2103         ret = in_interrupt();
2104 #endif
2105         return ret;
2106 }
2107 
2108 static DEFINE_RATELIMIT_STATE(nopage_rs,
2109                 DEFAULT_RATELIMIT_INTERVAL,
2110                 DEFAULT_RATELIMIT_BURST);
2111 
2112 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2113 {
2114         unsigned int filter = SHOW_MEM_FILTER_NODES;
2115 
2116         if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2117             debug_guardpage_minorder() > 0)
2118                 return;
2119 
2120         /*
2121          * This documents exceptions given to allocations in certain
2122          * contexts that are allowed to allocate outside current's set
2123          * of allowed nodes.
2124          */
2125         if (!(gfp_mask & __GFP_NOMEMALLOC))
2126                 if (test_thread_flag(TIF_MEMDIE) ||
2127                     (current->flags & (PF_MEMALLOC | PF_EXITING)))
2128                         filter &= ~SHOW_MEM_FILTER_NODES;
2129         if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2130                 filter &= ~SHOW_MEM_FILTER_NODES;
2131 
2132         if (fmt) {
2133                 struct va_format vaf;
2134                 va_list args;
2135 
2136                 va_start(args, fmt);
2137 
2138                 vaf.fmt = fmt;
2139                 vaf.va = &args;
2140 
2141                 pr_warn("%pV", &vaf);
2142 
2143                 va_end(args);
2144         }
2145 
2146         pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2147                 current->comm, order, gfp_mask);
2148 
2149         dump_stack();
2150         if (!should_suppress_show_mem())
2151                 show_mem(filter);
2152 }
2153 
2154 static inline int
2155 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
2156                                 unsigned long did_some_progress,
2157                                 unsigned long pages_reclaimed)
2158 {
2159         /* Do not loop if specifically requested */
2160         if (gfp_mask & __GFP_NORETRY)
2161                 return 0;
2162 
2163         /* Always retry if specifically requested */
2164         if (gfp_mask & __GFP_NOFAIL)
2165                 return 1;
2166 
2167         /*
2168          * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2169          * making forward progress without invoking OOM. Suspend also disables
2170          * storage devices so kswapd will not help. Bail if we are suspending.
2171          */
2172         if (!did_some_progress && pm_suspended_storage())
2173                 return 0;
2174 
2175         /*
2176          * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2177          * means __GFP_NOFAIL, but that may not be true in other
2178          * implementations.
2179          */
2180         if (order <= PAGE_ALLOC_COSTLY_ORDER)
2181                 return 1;
2182 
2183         /*
2184          * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2185          * specified, then we retry until we no longer reclaim any pages
2186          * (above), or we've reclaimed an order of pages at least as
2187          * large as the allocation's order. In both cases, if the
2188          * allocation still fails, we stop retrying.
2189          */
2190         if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2191                 return 1;
2192 
2193         return 0;
2194 }
2195 
2196 static inline struct page *
2197 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2198         struct zonelist *zonelist, enum zone_type high_zoneidx,
2199         nodemask_t *nodemask, struct zone *preferred_zone,
2200         int classzone_idx, int migratetype)
2201 {
2202         struct page *page;
2203 
2204         /* Acquire the OOM killer lock for the zones in zonelist */
2205         if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
2206                 schedule_timeout_uninterruptible(1);
2207                 return NULL;
2208         }
2209 
2210         /*
2211          * Go through the zonelist yet one more time, keep very high watermark
2212          * here, this is only to catch a parallel oom killing, we must fail if
2213          * we're still under heavy pressure.
2214          */
2215         page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2216                 order, zonelist, high_zoneidx,
2217                 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
2218                 preferred_zone, classzone_idx, migratetype);
2219         if (page)
2220                 goto out;
2221 
2222         if (!(gfp_mask & __GFP_NOFAIL)) {
2223                 /* The OOM killer will not help higher order allocs */
2224                 if (order > PAGE_ALLOC_COSTLY_ORDER)
2225                         goto out;
2226                 /* The OOM killer does not needlessly kill tasks for lowmem */
2227                 if (high_zoneidx < ZONE_NORMAL)
2228                         goto out;
2229                 /*
2230                  * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2231                  * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2232                  * The caller should handle page allocation failure by itself if
2233                  * it specifies __GFP_THISNODE.
2234                  * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2235                  */
2236                 if (gfp_mask & __GFP_THISNODE)
2237                         goto out;
2238         }
2239         /* Exhausted what can be done so it's blamo time */
2240         out_of_memory(zonelist, gfp_mask, order, nodemask, false);
2241 
2242 out:
2243         clear_zonelist_oom(zonelist, gfp_mask);
2244         return page;
2245 }
2246 
2247 #ifdef CONFIG_COMPACTION
2248 /* Try memory compaction for high-order allocations before reclaim */
2249 static struct page *
2250 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2251         struct zonelist *zonelist, enum zone_type high_zoneidx,
2252         nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2253         int classzone_idx, int migratetype, enum migrate_mode mode,
2254         bool *contended_compaction, bool *deferred_compaction,
2255         unsigned long *did_some_progress)
2256 {
2257         if (!order)
2258                 return NULL;
2259 
2260         if (compaction_deferred(preferred_zone, order)) {
2261                 *deferred_compaction = true;
2262                 return NULL;
2263         }
2264 
2265         current->flags |= PF_MEMALLOC;
2266         *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
2267                                                 nodemask, mode,
2268                                                 contended_compaction);
2269         current->flags &= ~PF_MEMALLOC;
2270 
2271         if (*did_some_progress != COMPACT_SKIPPED) {
2272                 struct page *page;
2273 
2274                 /* Page migration frees to the PCP lists but we want merging */
2275                 drain_pages(get_cpu());
2276                 put_cpu();
2277 
2278                 page = get_page_from_freelist(gfp_mask, nodemask,
2279                                 order, zonelist, high_zoneidx,
2280                                 alloc_flags & ~ALLOC_NO_WATERMARKS,
2281                                 preferred_zone, classzone_idx, migratetype);
2282                 if (page) {
2283                         preferred_zone->compact_blockskip_flush = false;
2284                         compaction_defer_reset(preferred_zone, order, true);
2285                         count_vm_event(COMPACTSUCCESS);
2286                         return page;
2287                 }
2288 
2289                 /*
2290                  * It's bad if compaction run occurs and fails.
2291                  * The most likely reason is that pages exist,
2292                  * but not enough to satisfy watermarks.
2293                  */
2294                 count_vm_event(COMPACTFAIL);
2295 
2296                 /*
2297                  * As async compaction considers a subset of pageblocks, only
2298                  * defer if the failure was a sync compaction failure.
2299                  */
2300                 if (mode != MIGRATE_ASYNC)
2301                         defer_compaction(preferred_zone, order);
2302 
2303                 cond_resched();
2304         }
2305 
2306         return NULL;
2307 }
2308 #else
2309 static inline struct page *
2310 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2311         struct zonelist *zonelist, enum zone_type high_zoneidx,
2312         nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2313         int classzone_idx, int migratetype,
2314         enum migrate_mode mode, bool *contended_compaction,
2315         bool *deferred_compaction, unsigned long *did_some_progress)
2316 {
2317         return NULL;
2318 }
2319 #endif /* CONFIG_COMPACTION */
2320 
2321 /* Perform direct synchronous page reclaim */
2322 static int
2323 __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2324                   nodemask_t *nodemask)
2325 {
2326         struct reclaim_state reclaim_state;
2327         int progress;
2328 
2329         cond_resched();
2330 
2331         /* We now go into synchronous reclaim */
2332         cpuset_memory_pressure_bump();
2333         current->flags |= PF_MEMALLOC;
2334         lockdep_set_current_reclaim_state(gfp_mask);
2335         reclaim_state.reclaimed_slab = 0;
2336         current->reclaim_state = &reclaim_state;
2337 
2338         progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
2339 
2340         current->reclaim_state = NULL;
2341         lockdep_clear_current_reclaim_state();
2342         current->flags &= ~PF_MEMALLOC;
2343 
2344         cond_resched();
2345 
2346         return progress;
2347 }
2348 
2349 /* The really slow allocator path where we enter direct reclaim */
2350 static inline struct page *
2351 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2352         struct zonelist *zonelist, enum zone_type high_zoneidx,
2353         nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2354         int classzone_idx, int migratetype, unsigned long *did_some_progress)
2355 {
2356         struct page *page = NULL;
2357         bool drained = false;
2358 
2359         *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2360                                                nodemask);
2361         if (unlikely(!(*did_some_progress)))
2362                 return NULL;
2363 
2364         /* After successful reclaim, reconsider all zones for allocation */
2365         if (IS_ENABLED(CONFIG_NUMA))
2366                 zlc_clear_zones_full(zonelist);
2367 
2368 retry:
2369         page = get_page_from_freelist(gfp_mask, nodemask, order,
2370                                         zonelist, high_zoneidx,
2371                                         alloc_flags & ~ALLOC_NO_WATERMARKS,
2372                                         preferred_zone, classzone_idx,
2373                                         migratetype);
2374 
2375         /*
2376          * If an allocation failed after direct reclaim, it could be because
2377          * pages are pinned on the per-cpu lists. Drain them and try again
2378          */
2379         if (!page && !drained) {
2380                 drain_all_pages();
2381                 drained = true;
2382                 goto retry;
2383         }
2384 
2385         return page;
2386 }
2387 
2388 /*
2389  * This is called in the allocator slow-path if the allocation request is of
2390  * sufficient urgency to ignore watermarks and take other desperate measures
2391  */
2392 static inline struct page *
2393 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2394         struct zonelist *zonelist, enum zone_type high_zoneidx,
2395         nodemask_t *nodemask, struct zone *preferred_zone,
2396         int classzone_idx, int migratetype)
2397 {
2398         struct page *page;
2399 
2400         do {
2401                 page = get_page_from_freelist(gfp_mask, nodemask, order,
2402                         zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
2403                         preferred_zone, classzone_idx, migratetype);
2404 
2405                 if (!page && gfp_mask & __GFP_NOFAIL)
2406                         wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2407         } while (!page && (gfp_mask & __GFP_NOFAIL));
2408 
2409         return page;
2410 }
2411 
2412 static void reset_alloc_batches(struct zonelist *zonelist,
2413                                 enum zone_type high_zoneidx,
2414                                 struct zone *preferred_zone)
2415 {
2416         struct zoneref *z;
2417         struct zone *zone;
2418 
2419         for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
2420                 /*
2421                  * Only reset the batches of zones that were actually
2422                  * considered in the fairness pass, we don't want to
2423                  * trash fairness information for zones that are not
2424                  * actually part of this zonelist's round-robin cycle.
2425                  */
2426                 if (!zone_local(preferred_zone, zone))
2427                         continue;
2428                 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2429                         high_wmark_pages(zone) - low_wmark_pages(zone) -
2430                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
2431         }
2432 }
2433 
2434 static void wake_all_kswapds(unsigned int order,
2435                              struct zonelist *zonelist,
2436                              enum zone_type high_zoneidx,
2437                              struct zone *preferred_zone)
2438 {
2439         struct zoneref *z;
2440         struct zone *zone;
2441 
2442         for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
2443                 wakeup_kswapd(zone, order, zone_idx(preferred_zone));
2444 }
2445 
2446 static inline int
2447 gfp_to_alloc_flags(gfp_t gfp_mask)
2448 {
2449         int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2450         const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
2451 
2452         /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2453         BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2454 
2455         /*
2456          * The caller may dip into page reserves a bit more if the caller
2457          * cannot run direct reclaim, or if the caller has realtime scheduling
2458          * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
2459          * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
2460          */
2461         alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2462 
2463         if (atomic) {
2464                 /*
2465                  * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2466                  * if it can't schedule.
2467                  */
2468                 if (!(gfp_mask & __GFP_NOMEMALLOC))
2469                         alloc_flags |= ALLOC_HARDER;
2470                 /*
2471                  * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2472                  * comment for __cpuset_node_allowed_softwall().
2473                  */
2474                 alloc_flags &= ~ALLOC_CPUSET;
2475         } else if (unlikely(rt_task(current)) && !in_interrupt())
2476                 alloc_flags |= ALLOC_HARDER;
2477 
2478         if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2479                 if (gfp_mask & __GFP_MEMALLOC)
2480                         alloc_flags |= ALLOC_NO_WATERMARKS;
2481                 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2482                         alloc_flags |= ALLOC_NO_WATERMARKS;
2483                 else if (!in_interrupt() &&
2484                                 ((current->flags & PF_MEMALLOC) ||
2485                                  unlikely(test_thread_flag(TIF_MEMDIE))))
2486                         alloc_flags |= ALLOC_NO_WATERMARKS;
2487         }
2488 #ifdef CONFIG_CMA
2489         if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2490                 alloc_flags |= ALLOC_CMA;
2491 #endif
2492         return alloc_flags;
2493 }
2494 
2495 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2496 {
2497         return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
2498 }
2499 
2500 static inline struct page *
2501 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2502         struct zonelist *zonelist, enum zone_type high_zoneidx,
2503         nodemask_t *nodemask, struct zone *preferred_zone,
2504         int classzone_idx, int migratetype)
2505 {
2506         const gfp_t wait = gfp_mask & __GFP_WAIT;
2507         struct page *page = NULL;
2508         int alloc_flags;
2509         unsigned long pages_reclaimed = 0;
2510         unsigned long did_some_progress;
2511         enum migrate_mode migration_mode = MIGRATE_ASYNC;
2512         bool deferred_compaction = false;
2513         bool contended_compaction = false;
2514 
2515         /*
2516          * In the slowpath, we sanity check order to avoid ever trying to
2517          * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2518          * be using allocators in order of preference for an area that is
2519          * too large.
2520          */
2521         if (order >= MAX_ORDER) {
2522                 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2523                 return NULL;
2524         }
2525 
2526         /*
2527          * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2528          * __GFP_NOWARN set) should not cause reclaim since the subsystem
2529          * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2530          * using a larger set of nodes after it has established that the
2531          * allowed per node queues are empty and that nodes are
2532          * over allocated.
2533          */
2534         if (IS_ENABLED(CONFIG_NUMA) &&
2535             (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2536                 goto nopage;
2537 
2538 restart:
2539         if (!(gfp_mask & __GFP_NO_KSWAPD))
2540                 wake_all_kswapds(order, zonelist, high_zoneidx, preferred_zone);
2541 
2542         /*
2543          * OK, we're below the kswapd watermark and have kicked background
2544          * reclaim. Now things get more complex, so set up alloc_flags according
2545          * to how we want to proceed.
2546          */
2547         alloc_flags = gfp_to_alloc_flags(gfp_mask);
2548 
2549         /*
2550          * Find the true preferred zone if the allocation is unconstrained by
2551          * cpusets.
2552          */
2553         if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) {
2554                 struct zoneref *preferred_zoneref;
2555                 preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx,
2556                                 NULL, &preferred_zone);
2557                 classzone_idx = zonelist_zone_idx(preferred_zoneref);
2558         }
2559 
2560 rebalance:
2561         /* This is the last chance, in general, before the goto nopage. */
2562         page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2563                         high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2564                         preferred_zone, classzone_idx, migratetype);
2565         if (page)
2566                 goto got_pg;
2567 
2568         /* Allocate without watermarks if the context allows */
2569         if (alloc_flags & ALLOC_NO_WATERMARKS) {
2570                 /*
2571                  * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2572                  * the allocation is high priority and these type of
2573                  * allocations are system rather than user orientated
2574                  */
2575                 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2576 
2577                 page = __alloc_pages_high_priority(gfp_mask, order,
2578                                 zonelist, high_zoneidx, nodemask,
2579                                 preferred_zone, classzone_idx, migratetype);
2580                 if (page) {
2581                         goto got_pg;
2582                 }
2583         }
2584 
2585         /* Atomic allocations - we can't balance anything */
2586         if (!wait) {
2587                 /*
2588                  * All existing users of the deprecated __GFP_NOFAIL are
2589                  * blockable, so warn of any new users that actually allow this
2590                  * type of allocation to fail.
2591                  */
2592                 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
2593                 goto nopage;
2594         }
2595 
2596         /* Avoid recursion of direct reclaim */
2597         if (current->flags & PF_MEMALLOC)
2598                 goto nopage;
2599 
2600         /* Avoid allocations with no watermarks from looping endlessly */
2601         if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2602                 goto nopage;
2603 
2604         /*
2605          * Try direct compaction. The first pass is asynchronous. Subsequent
2606          * attempts after direct reclaim are synchronous
2607          */
2608         page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2609                                         high_zoneidx, nodemask, alloc_flags,
2610                                         preferred_zone,
2611                                         classzone_idx, migratetype,
2612                                         migration_mode, &contended_compaction,
2613                                         &deferred_compaction,
2614                                         &did_some_progress);
2615         if (page)
2616                 goto got_pg;
2617 
2618         /*
2619          * It can become very expensive to allocate transparent hugepages at
2620          * fault, so use asynchronous memory compaction for THP unless it is
2621          * khugepaged trying to collapse.
2622          */
2623         if (!(gfp_mask & __GFP_NO_KSWAPD) || (current->flags & PF_KTHREAD))
2624                 migration_mode = MIGRATE_SYNC_LIGHT;
2625 
2626         /*
2627          * If compaction is deferred for high-order allocations, it is because
2628          * sync compaction recently failed. In this is the case and the caller
2629          * requested a movable allocation that does not heavily disrupt the
2630          * system then fail the allocation instead of entering direct reclaim.
2631          */
2632         if ((deferred_compaction || contended_compaction) &&
2633                                                 (gfp_mask & __GFP_NO_KSWAPD))
2634                 goto nopage;
2635 
2636         /* Try direct reclaim and then allocating */
2637         page = __alloc_pages_direct_reclaim(gfp_mask, order,
2638                                         zonelist, high_zoneidx,
2639                                         nodemask,
2640                                         alloc_flags, preferred_zone,
2641                                         classzone_idx, migratetype,
2642                                         &did_some_progress);
2643         if (page)
2644                 goto got_pg;
2645 
2646         /*
2647          * If we failed to make any progress reclaiming, then we are
2648          * running out of options and have to consider going OOM
2649          */
2650         if (!did_some_progress) {
2651                 if (oom_gfp_allowed(gfp_mask)) {
2652                         if (oom_killer_disabled)
2653                                 goto nopage;
2654                         /* Coredumps can quickly deplete all memory reserves */
2655                         if ((current->flags & PF_DUMPCORE) &&
2656                             !(gfp_mask & __GFP_NOFAIL))
2657                                 goto nopage;
2658                         page = __alloc_pages_may_oom(gfp_mask, order,
2659                                         zonelist, high_zoneidx,
2660                                         nodemask, preferred_zone,
2661                                         classzone_idx, migratetype);
2662                         if (page)
2663                                 goto got_pg;
2664 
2665                         if (!(gfp_mask & __GFP_NOFAIL)) {
2666                                 /*
2667                                  * The oom killer is not called for high-order
2668                                  * allocations that may fail, so if no progress
2669                                  * is being made, there are no other options and
2670                                  * retrying is unlikely to help.
2671                                  */
2672                                 if (order > PAGE_ALLOC_COSTLY_ORDER)
2673                                         goto nopage;
2674                                 /*
2675                                  * The oom killer is not called for lowmem
2676                                  * allocations to prevent needlessly killing
2677                                  * innocent tasks.
2678                                  */
2679                                 if (high_zoneidx < ZONE_NORMAL)
2680                                         goto nopage;
2681                         }
2682 
2683                         goto restart;
2684                 }
2685         }
2686 
2687         /* Check if we should retry the allocation */
2688         pages_reclaimed += did_some_progress;
2689         if (should_alloc_retry(gfp_mask, order, did_some_progress,
2690                                                 pages_reclaimed)) {
2691                 /* Wait for some write requests to complete then retry */
2692                 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2693                 goto rebalance;
2694         } else {
2695                 /*
2696                  * High-order allocations do not necessarily loop after
2697                  * direct reclaim and reclaim/compaction depends on compaction
2698                  * being called after reclaim so call directly if necessary
2699                  */
2700                 page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2701                                         high_zoneidx, nodemask, alloc_flags,
2702                                         preferred_zone,
2703                                         classzone_idx, migratetype,
2704                                         migration_mode, &contended_compaction,
2705                                         &deferred_compaction,
2706                                         &did_some_progress);
2707                 if (page)
2708                         goto got_pg;
2709         }
2710 
2711 nopage:
2712         warn_alloc_failed(gfp_mask, order, NULL);
2713         return page;
2714 got_pg:
2715         if (kmemcheck_enabled)
2716                 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2717 
2718         return page;
2719 }
2720 
2721 /*
2722  * This is the 'heart' of the zoned buddy allocator.
2723  */
2724 struct page *
2725 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2726                         struct zonelist *zonelist, nodemask_t *nodemask)
2727 {
2728         enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2729         struct zone *preferred_zone;
2730         struct zoneref *preferred_zoneref;
2731         struct page *page = NULL;
2732         int migratetype = allocflags_to_migratetype(gfp_mask);
2733         unsigned int cpuset_mems_cookie;
2734         int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
2735         int classzone_idx;
2736 
2737         gfp_mask &= gfp_allowed_mask;
2738 
2739         lockdep_trace_alloc(gfp_mask);
2740 
2741         might_sleep_if(gfp_mask & __GFP_WAIT);
2742 
2743         if (should_fail_alloc_page(gfp_mask, order))
2744                 return NULL;
2745 
2746         /*
2747          * Check the zones suitable for the gfp_mask contain at least one
2748          * valid zone. It's possible to have an empty zonelist as a result
2749          * of GFP_THISNODE and a memoryless node
2750          */
2751         if (unlikely(!zonelist->_zonerefs->zone))
2752                 return NULL;
2753 
2754 retry_cpuset:
2755         cpuset_mems_cookie = read_mems_allowed_begin();
2756 
2757         /* The preferred zone is used for statistics later */
2758         preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx,
2759                                 nodemask ? : &cpuset_current_mems_allowed,
2760                                 &preferred_zone);
2761         if (!preferred_zone)
2762                 goto out;
2763         classzone_idx = zonelist_zone_idx(preferred_zoneref);
2764 
2765 #ifdef CONFIG_CMA
2766         if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2767                 alloc_flags |= ALLOC_CMA;
2768 #endif
2769 retry:
2770         /* First allocation attempt */
2771         page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2772                         zonelist, high_zoneidx, alloc_flags,
2773                         preferred_zone, classzone_idx, migratetype);
2774         if (unlikely(!page)) {
2775                 /*
2776                  * The first pass makes sure allocations are spread
2777                  * fairly within the local node.  However, the local
2778                  * node might have free pages left after the fairness
2779                  * batches are exhausted, and remote zones haven't
2780                  * even been considered yet.  Try once more without
2781                  * fairness, and include remote zones now, before
2782                  * entering the slowpath and waking kswapd: prefer
2783                  * spilling to a remote zone over swapping locally.
2784                  */
2785                 if (alloc_flags & ALLOC_FAIR) {
2786                         reset_alloc_batches(zonelist, high_zoneidx,
2787                                             preferred_zone);
2788                         alloc_flags &= ~ALLOC_FAIR;
2789                         goto retry;
2790                 }
2791                 /*
2792                  * Runtime PM, block IO and its error handling path
2793                  * can deadlock because I/O on the device might not
2794                  * complete.
2795                  */
2796                 gfp_mask = memalloc_noio_flags(gfp_mask);
2797                 page = __alloc_pages_slowpath(gfp_mask, order,
2798                                 zonelist, high_zoneidx, nodemask,
2799                                 preferred_zone, classzone_idx, migratetype);
2800         }
2801 
2802         trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2803 
2804 out:
2805         /*
2806          * When updating a task's mems_allowed, it is possible to race with
2807          * parallel threads in such a way that an allocation can fail while
2808          * the mask is being updated. If a page allocation is about to fail,
2809          * check if the cpuset changed during allocation and if so, retry.
2810          */
2811         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
2812                 goto retry_cpuset;
2813 
2814         return page;
2815 }
2816 EXPORT_SYMBOL(__alloc_pages_nodemask);
2817 
2818 /*
2819  * Common helper functions.
2820  */
2821 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2822 {
2823         struct page *page;
2824 
2825         /*
2826          * __get_free_pages() returns a 32-bit address, which cannot represent
2827          * a highmem page
2828          */
2829         VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2830 
2831         page = alloc_pages(gfp_mask, order);
2832         if (!page)
2833                 return 0;
2834         return (unsigned long) page_address(page);
2835 }
2836 EXPORT_SYMBOL(__get_free_pages);
2837 
2838 unsigned long get_zeroed_page(gfp_t gfp_mask)
2839 {
2840         return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2841 }
2842 EXPORT_SYMBOL(get_zeroed_page);
2843 
2844 void __free_pages(struct page *page, unsigned int order)
2845 {
2846         if (put_page_testzero(page)) {
2847                 if (order == 0)
2848                         free_hot_cold_page(page, false);
2849                 else
2850                         __free_pages_ok(page, order);
2851         }
2852 }
2853 
2854 EXPORT_SYMBOL(__free_pages);
2855 
2856 void free_pages(unsigned long addr, unsigned int order)
2857 {
2858         if (addr != 0) {
2859                 VM_BUG_ON(!virt_addr_valid((void *)addr));
2860                 __free_pages(virt_to_page((void *)addr), order);
2861         }
2862 }
2863 
2864 EXPORT_SYMBOL(free_pages);
2865 
2866 /*
2867  * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
2868  * of the current memory cgroup.
2869  *
2870  * It should be used when the caller would like to use kmalloc, but since the
2871  * allocation is large, it has to fall back to the page allocator.
2872  */
2873 struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
2874 {
2875         struct page *page;
2876         struct mem_cgroup *memcg = NULL;
2877 
2878         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2879                 return NULL;
2880         page = alloc_pages(gfp_mask, order);
2881         memcg_kmem_commit_charge(page, memcg, order);
2882         return page;
2883 }
2884 
2885 struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
2886 {
2887         struct page *page;
2888         struct mem_cgroup *memcg = NULL;
2889 
2890         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2891                 return NULL;
2892         page = alloc_pages_node(nid, gfp_mask, order);
2893         memcg_kmem_commit_charge(page, memcg, order);
2894         return page;
2895 }
2896 
2897 /*
2898  * __free_kmem_pages and free_kmem_pages will free pages allocated with
2899  * alloc_kmem_pages.
2900  */
2901 void __free_kmem_pages(struct page *page, unsigned int order)
2902 {
2903         memcg_kmem_uncharge_pages(page, order);
2904         __free_pages(page, order);
2905 }
2906 
2907 void free_kmem_pages(unsigned long addr, unsigned int order)
2908 {
2909         if (addr != 0) {
2910                 VM_BUG_ON(!virt_addr_valid((void *)addr));
2911                 __free_kmem_pages(virt_to_page((void *)addr), order);
2912         }
2913 }
2914 
2915 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2916 {
2917         if (addr) {
2918                 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2919                 unsigned long used = addr + PAGE_ALIGN(size);
2920 
2921                 split_page(virt_to_page((void *)addr), order);
2922                 while (used < alloc_end) {
2923                         free_page(used);
2924                         used += PAGE_SIZE;
2925                 }
2926         }
2927         return (void *)addr;
2928 }
2929 
2930 /**
2931  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2932  * @size: the number of bytes to allocate
2933  * @gfp_mask: GFP flags for the allocation
2934  *
2935  * This function is similar to alloc_pages(), except that it allocates the
2936  * minimum number of pages to satisfy the request.  alloc_pages() can only
2937  * allocate memory in power-of-two pages.
2938  *
2939  * This function is also limited by MAX_ORDER.
2940  *
2941  * Memory allocated by this function must be released by free_pages_exact().
2942  */
2943 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2944 {
2945         unsigned int order = get_order(size);
2946         unsigned long addr;
2947 
2948         addr = __get_free_pages(gfp_mask, order);
2949         return make_alloc_exact(addr, order, size);
2950 }
2951 EXPORT_SYMBOL(alloc_pages_exact);
2952 
2953 /**
2954  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2955  *                         pages on a node.
2956  * @nid: the preferred node ID where memory should be allocated
2957  * @size: the number of bytes to allocate
2958  * @gfp_mask: GFP flags for the allocation
2959  *
2960  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2961  * back.
2962  * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2963  * but is not exact.
2964  */
2965 void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2966 {
2967         unsigned order = get_order(size);
2968         struct page *p = alloc_pages_node(nid, gfp_mask, order);
2969         if (!p)
2970                 return NULL;
2971         return make_alloc_exact((unsigned long)page_address(p), order, size);
2972 }
2973 EXPORT_SYMBOL(alloc_pages_exact_nid);
2974 
2975 /**
2976  * free_pages_exact - release memory allocated via alloc_pages_exact()
2977  * @virt: the value returned by alloc_pages_exact.
2978  * @size: size of allocation, same value as passed to alloc_pages_exact().
2979  *
2980  * Release the memory allocated by a previous call to alloc_pages_exact.
2981  */
2982 void free_pages_exact(void *virt, size_t size)
2983 {
2984         unsigned long addr = (unsigned long)virt;
2985         unsigned long end = addr + PAGE_ALIGN(size);
2986 
2987         while (addr < end) {
2988                 free_page(addr);
2989                 addr += PAGE_SIZE;
2990         }
2991 }
2992 EXPORT_SYMBOL(free_pages_exact);
2993 
2994 /**
2995  * nr_free_zone_pages - count number of pages beyond high watermark
2996  * @offset: The zone index of the highest zone
2997  *
2998  * nr_free_zone_pages() counts the number of counts pages which are beyond the
2999  * high watermark within all zones at or below a given zone index.  For each
3000  * zone, the number of pages is calculated as:
3001  *     managed_pages - high_pages
3002  */
3003 static unsigned long nr_free_zone_pages(int offset)
3004 {
3005         struct zoneref *z;
3006         struct zone *zone;
3007 
3008         /* Just pick one node, since fallback list is circular */
3009         unsigned long sum = 0;
3010 
3011         struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3012 
3013         for_each_zone_zonelist(zone, z, zonelist, offset) {
3014                 unsigned long size = zone->managed_pages;
3015                 unsigned long high = high_wmark_pages(zone);
3016                 if (size > high)
3017                         sum += size - high;
3018         }
3019 
3020         return sum;
3021 }
3022 
3023 /**
3024  * nr_free_buffer_pages - count number of pages beyond high watermark
3025  *
3026  * nr_free_buffer_pages() counts the number of pages which are beyond the high
3027  * watermark within ZONE_DMA and ZONE_NORMAL.
3028  */
3029 unsigned long nr_free_buffer_pages(void)
3030 {
3031         return nr_free_zone_pages(gfp_zone(GFP_USER));
3032 }
3033 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3034 
3035 /**
3036  * nr_free_pagecache_pages - count number of pages beyond high watermark
3037  *
3038  * nr_free_pagecache_pages() counts the number of pages which are beyond the
3039  * high watermark within all zones.
3040  */
3041 unsigned long nr_free_pagecache_pages(void)
3042 {
3043         return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3044 }
3045 
3046 static inline void show_node(struct zone *zone)
3047 {
3048         if (IS_ENABLED(CONFIG_NUMA))
3049                 printk("Node %d ", zone_to_nid(zone));
3050 }
3051 
3052 void si_meminfo(struct sysinfo *val)
3053 {
3054         val->totalram = totalram_pages;
3055         val->sharedram = 0;
3056         val->freeram = global_page_state(NR_FREE_PAGES);
3057         val->bufferram = nr_blockdev_pages();
3058         val->totalhigh = totalhigh_pages;
3059         val->freehigh = nr_free_highpages();
3060         val->mem_unit = PAGE_SIZE;
3061 }
3062 
3063 EXPORT_SYMBOL(si_meminfo);
3064 
3065 #ifdef CONFIG_NUMA
3066 void si_meminfo_node(struct sysinfo *val, int nid)
3067 {
3068         int zone_type;          /* needs to be signed */
3069         unsigned long managed_pages = 0;
3070         pg_data_t *pgdat = NODE_DATA(nid);
3071 
3072         for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3073                 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3074         val->totalram = managed_pages;
3075         val->freeram = node_page_state(nid, NR_FREE_PAGES);
3076 #ifdef CONFIG_HIGHMEM
3077         val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
3078         val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3079                         NR_FREE_PAGES);
3080 #else
3081         val->totalhigh = 0;
3082         val->freehigh = 0;
3083 #endif
3084         val->mem_unit = PAGE_SIZE;
3085 }
3086 #endif
3087 
3088 /*
3089  * Determine whether the node should be displayed or not, depending on whether
3090  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3091  */
3092 bool skip_free_areas_node(unsigned int flags, int nid)
3093 {
3094         bool ret = false;
3095         unsigned int cpuset_mems_cookie;
3096 
3097         if (!(flags & SHOW_MEM_FILTER_NODES))
3098                 goto out;
3099 
3100         do {
3101                 cpuset_mems_cookie = read_mems_allowed_begin();
3102                 ret = !node_isset(nid, cpuset_current_mems_allowed);
3103         } while (read_mems_allowed_retry(cpuset_mems_cookie));
3104 out:
3105         return ret;
3106 }
3107 
3108 #define K(x) ((x) << (PAGE_SHIFT-10))
3109 
3110 static void show_migration_types(unsigned char type)
3111 {
3112         static const char types[MIGRATE_TYPES] = {
3113                 [MIGRATE_UNMOVABLE]     = 'U',
3114                 [MIGRATE_RECLAIMABLE]   = 'E',
3115                 [MIGRATE_MOVABLE]       = 'M',
3116                 [MIGRATE_RESERVE]       = 'R',
3117 #ifdef CONFIG_CMA
3118                 [MIGRATE_CMA]           = 'C',
3119 #endif
3120 #ifdef CONFIG_MEMORY_ISOLATION
3121                 [MIGRATE_ISOLATE]       = 'I',
3122 #endif
3123         };
3124         char tmp[MIGRATE_TYPES + 1];
3125         char *p = tmp;
3126         int i;
3127 
3128         for (i = 0; i < MIGRATE_TYPES; i++) {
3129                 if (type & (1 << i))
3130                         *p++ = types[i];
3131         }
3132 
3133         *p = '\0';
3134         printk("(%s) ", tmp);
3135 }
3136 
3137 /*
3138  * Show free area list (used inside shift_scroll-lock stuff)
3139  * We also calculate the percentage fragmentation. We do this by counting the
3140  * memory on each free list with the exception of the first item on the list.
3141  * Suppresses nodes that are not allowed by current's cpuset if
3142  * SHOW_MEM_FILTER_NODES is passed.
3143  */
3144 void show_free_areas(unsigned int filter)
3145 {
3146         int cpu;
3147         struct zone *zone;
3148 
3149         for_each_populated_zone(zone) {
3150                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3151                         continue;
3152                 show_node(zone);
3153                 printk("%s per-cpu:\n", zone->name);
3154 
3155                 for_each_online_cpu(cpu) {
3156                         struct per_cpu_pageset *pageset;
3157 
3158                         pageset = per_cpu_ptr(zone->pageset, cpu);
3159 
3160                         printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3161                                cpu, pageset->pcp.high,
3162                                pageset->pcp.batch, pageset->pcp.count);
3163                 }
3164         }
3165 
3166         printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3167                 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3168                 " unevictable:%lu"
3169                 " dirty:%lu writeback:%lu unstable:%lu\n"
3170                 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3171                 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3172                 " free_cma:%lu\n",
3173                 global_page_state(NR_ACTIVE_ANON),
3174                 global_page_state(NR_INACTIVE_ANON),
3175                 global_page_state(NR_ISOLATED_ANON),
3176                 global_page_state(NR_ACTIVE_FILE),
3177                 global_page_state(NR_INACTIVE_FILE),
3178                 global_page_state(NR_ISOLATED_FILE),
3179                 global_page_state(NR_UNEVICTABLE),
3180                 global_page_state(NR_FILE_DIRTY),
3181                 global_page_state(NR_WRITEBACK),
3182                 global_page_state(NR_UNSTABLE_NFS),
3183                 global_page_state(NR_FREE_PAGES),
3184                 global_page_state(NR_SLAB_RECLAIMABLE),
3185                 global_page_state(NR_SLAB_UNRECLAIMABLE),
3186                 global_page_state(NR_FILE_MAPPED),
3187                 global_page_state(NR_SHMEM),
3188                 global_page_state(NR_PAGETABLE),
3189                 global_page_state(NR_BOUNCE),
3190                 global_page_state(NR_FREE_CMA_PAGES));
3191 
3192         for_each_populated_zone(zone) {
3193                 int i;
3194 
3195                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3196                         continue;
3197                 show_node(zone);
3198                 printk("%s"
3199                         " free:%lukB"
3200                         " min:%lukB"
3201                         " low:%lukB"
3202                         " high:%lukB"
3203                         " active_anon:%lukB"
3204                         " inactive_anon:%lukB"
3205                         " active_file:%lukB"
3206                         " inactive_file:%lukB"
3207                         " unevictable:%lukB"
3208                         " isolated(anon):%lukB"
3209                         " isolated(file):%lukB"
3210                         " present:%lukB"
3211                         " managed:%lukB"
3212                         " mlocked:%lukB"
3213                         " dirty:%lukB"
3214                         " writeback:%lukB"
3215                         " mapped:%lukB"
3216                         " shmem:%lukB"
3217                         " slab_reclaimable:%lukB"
3218                         " slab_unreclaimable:%lukB"
3219                         " kernel_stack:%lukB"
3220                         " pagetables:%lukB"
3221                         " unstable:%lukB"
3222                         " bounce:%lukB"
3223                         " free_cma:%lukB"
3224                         " writeback_tmp:%lukB"
3225                         " pages_scanned:%lu"
3226                         " all_unreclaimable? %s"
3227                         "\n",
3228                         zone->name,
3229                         K(zone_page_state(zone, NR_FREE_PAGES)),
3230                         K(min_wmark_pages(zone)),
3231                         K(low_wmark_pages(zone)),
3232                         K(high_wmark_pages(zone)),
3233                         K(zone_page_state(zone, NR_ACTIVE_ANON)),
3234                         K(zone_page_state(zone, NR_INACTIVE_ANON)),
3235                         K(zone_page_state(zone, NR_ACTIVE_FILE)),
3236                         K(zone_page_state(zone, NR_INACTIVE_FILE)),
3237                         K(zone_page_state(zone, NR_UNEVICTABLE)),
3238                         K(zone_page_state(zone, NR_ISOLATED_ANON)),
3239                         K(zone_page_state(zone, NR_ISOLATED_FILE)),
3240                         K(zone->present_pages),
3241                         K(zone->managed_pages),
3242                         K(zone_page_state(zone, NR_MLOCK)),
3243                         K(zone_page_state(zone, NR_FILE_DIRTY)),
3244                         K(zone_page_state(zone, NR_WRITEBACK)),
3245                         K(zone_page_state(zone, NR_FILE_MAPPED)),
3246                         K(zone_page_state(zone, NR_SHMEM)),
3247                         K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3248                         K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3249                         zone_page_state(zone, NR_KERNEL_STACK) *
3250                                 THREAD_SIZE / 1024,
3251                         K(zone_page_state(zone, NR_PAGETABLE)),
3252                         K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3253                         K(zone_page_state(zone, NR_BOUNCE)),
3254                         K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3255                         K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3256                         zone->pages_scanned,
3257                         (!zone_reclaimable(zone) ? "yes" : "no")
3258                         );
3259                 printk("lowmem_reserve[]:");
3260                 for (i = 0; i < MAX_NR_ZONES; i++)
3261                         printk(" %lu", zone->lowmem_reserve[i]);
3262                 printk("\n");
3263         }
3264 
3265         for_each_populated_zone(zone) {
3266                 unsigned long nr[MAX_ORDER], flags, order, total = 0;
3267                 unsigned char types[MAX_ORDER];
3268 
3269                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3270                         continue;
3271                 show_node(zone);
3272                 printk("%s: ", zone->name);
3273 
3274                 spin_lock_irqsave(&zone->lock, flags);
3275                 for (order = 0; order < MAX_ORDER; order++) {
3276                         struct free_area *area = &zone->free_area[order];
3277                         int type;
3278 
3279                         nr[order] = area->nr_free;
3280                         total += nr[order] << order;
3281 
3282                         types[order] = 0;
3283                         for (type = 0; type < MIGRATE_TYPES; type++) {
3284                                 if (!list_empty(&area->free_list[type]))
3285                                         types[order] |= 1 << type;
3286                         }
3287                 }
3288                 spin_unlock_irqrestore(&zone->lock, flags);
3289                 for (order = 0; order < MAX_ORDER; order++) {
3290                         printk("%lu*%lukB ", nr[order], K(1UL) << order);
3291                         if (nr[order])
3292                                 show_migration_types(types[order]);
3293                 }
3294                 printk("= %lukB\n", K(total));
3295         }
3296 
3297         hugetlb_show_meminfo();
3298 
3299         printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3300 
3301         show_swap_cache_info();
3302 }
3303 
3304 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3305 {
3306         zoneref->zone = zone;
3307         zoneref->zone_idx = zone_idx(zone);
3308 }
3309 
3310 /*
3311  * Builds allocation fallback zone lists.
3312  *
3313  * Add all populated zones of a node to the zonelist.
3314  */
3315 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3316                                 int nr_zones)
3317 {
3318         struct zone *zone;
3319         enum zone_type zone_type = MAX_NR_ZONES;
3320 
3321         do {
3322                 zone_type--;
3323                 zone = pgdat->node_zones + zone_type;
3324                 if (populated_zone(zone)) {
3325                         zoneref_set_zone(zone,
3326                                 &zonelist->_zonerefs[nr_zones++]);
3327                         check_highest_zone(zone_type);
3328                 }
3329         } while (zone_type);
3330 
3331         return nr_zones;
3332 }
3333 
3334 
3335 /*
3336  *  zonelist_order:
3337  *  0 = automatic detection of better ordering.
3338  *  1 = order by ([node] distance, -zonetype)
3339  *  2 = order by (-zonetype, [node] distance)
3340  *
3341  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3342  *  the same zonelist. So only NUMA can configure this param.
3343  */
3344 #define ZONELIST_ORDER_DEFAULT  0
3345 #define ZONELIST_ORDER_NODE     1
3346 #define ZONELIST_ORDER_ZONE     2
3347 
3348 /* zonelist order in the kernel.
3349  * set_zonelist_order() will set this to NODE or ZONE.
3350  */
3351 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3352 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3353 
3354 
3355 #ifdef CONFIG_NUMA
3356 /* The value user specified ....changed by config */
3357 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3358 /* string for sysctl */
3359 #define NUMA_ZONELIST_ORDER_LEN 16
3360 char numa_zonelist_order[16] = "default";
3361 
3362 /*
3363  * interface for configure zonelist ordering.
3364  * command line option "numa_zonelist_order"
3365  *      = "[dD]efault   - default, automatic configuration.
3366  *      = "[nN]ode      - order by node locality, then by zone within node
3367  *      = "[zZ]one      - order by zone, then by locality within zone
3368  */
3369 
3370 static int __parse_numa_zonelist_order(char *s)
3371 {
3372         if (*s == 'd' || *s == 'D') {
3373                 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3374         } else if (*s == 'n' || *s == 'N') {
3375                 user_zonelist_order = ZONELIST_ORDER_NODE;
3376         } else if (*s == 'z' || *s == 'Z') {
3377                 user_zonelist_order = ZONELIST_ORDER_ZONE;
3378         } else {
3379                 printk(KERN_WARNING
3380                         "Ignoring invalid numa_zonelist_order value:  "
3381                         "%s\n", s);
3382                 return -EINVAL;
3383         }
3384         return 0;
3385 }
3386 
3387 static __init int setup_numa_zonelist_order(char *s)
3388 {
3389         int ret;
3390 
3391         if (!s)
3392                 return 0;
3393 
3394         ret = __parse_numa_zonelist_order(s);
3395         if (ret == 0)
3396                 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3397 
3398         return ret;
3399 }
3400 early_param("numa_zonelist_order", setup_numa_zonelist_order);
3401 
3402 /*
3403  * sysctl handler for numa_zonelist_order
3404  */
3405 int numa_zonelist_order_handler(struct ctl_table *table, int write,
3406                 void __user *buffer, size_t *length,
3407                 loff_t *ppos)
3408 {
3409         char saved_string[NUMA_ZONELIST_ORDER_LEN];
3410         int ret;
3411         static DEFINE_MUTEX(zl_order_mutex);
3412 
3413         mutex_lock(&zl_order_mutex);
3414         if (write) {
3415                 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3416                         ret = -EINVAL;
3417                         goto out;
3418                 }
3419                 strcpy(saved_string, (char *)table->data);
3420         }
3421         ret = proc_dostring(table, write, buffer, length, ppos);
3422         if (ret)
3423                 goto out;
3424         if (write) {
3425                 int oldval = user_zonelist_order;
3426 
3427                 ret = __parse_numa_zonelist_order((char *)table->data);
3428                 if (ret) {
3429                         /*
3430                          * bogus value.  restore saved string
3431                          */
3432                         strncpy((char *)table->data, saved_string,
3433                                 NUMA_ZONELIST_ORDER_LEN);
3434                         user_zonelist_order = oldval;
3435                 } else if (oldval != user_zonelist_order) {
3436                         mutex_lock(&zonelists_mutex);
3437                         build_all_zonelists(NULL, NULL);
3438                         mutex_unlock(&zonelists_mutex);
3439                 }
3440         }
3441 out:
3442         mutex_unlock(&zl_order_mutex);
3443         return ret;
3444 }
3445 
3446 
3447 #define MAX_NODE_LOAD (nr_online_nodes)
3448 static int node_load[MAX_NUMNODES];
3449 
3450 /**
3451  * find_next_best_node - find the next node that should appear in a given node's fallback list
3452  * @node: node whose fallback list we're appending
3453  * @used_node_mask: nodemask_t of already used nodes
3454  *
3455  * We use a number of factors to determine which is the next node that should
3456  * appear on a given node's fallback list.  The node should not have appeared
3457  * already in @node's fallback list, and it should be the next closest node
3458  * according to the distance array (which contains arbitrary distance values
3459  * from each node to each node in the system), and should also prefer nodes
3460  * with no CPUs, since presumably they'll have very little allocation pressure
3461  * on them otherwise.
3462  * It returns -1 if no node is found.
3463  */
3464 static int find_next_best_node(int node, nodemask_t *used_node_mask)
3465 {
3466         int n, val;
3467         int min_val = INT_MAX;
3468         int best_node = NUMA_NO_NODE;
3469         const struct cpumask *tmp = cpumask_of_node(0);
3470 
3471         /* Use the local node if we haven't already */
3472         if (!node_isset(node, *used_node_mask)) {
3473                 node_set(node, *used_node_mask);
3474                 return node;
3475         }
3476 
3477         for_each_node_state(n, N_MEMORY) {
3478 
3479                 /* Don't want a node to appear more than once */
3480                 if (node_isset(n, *used_node_mask))
3481                         continue;
3482 
3483                 /* Use the distance array to find the distance */
3484                 val = node_distance(node, n);
3485 
3486                 /* Penalize nodes under us ("prefer the next node") */
3487                 val += (n < node);
3488 
3489                 /* Give preference to headless and unused nodes */
3490                 tmp = cpumask_of_node(n);
3491                 if (!cpumask_empty(tmp))
3492                         val += PENALTY_FOR_NODE_WITH_CPUS;
3493 
3494                 /* Slight preference for less loaded node */
3495                 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3496                 val += node_load[n];
3497 
3498                 if (val < min_val) {
3499                         min_val = val;
3500                         best_node = n;
3501                 }
3502         }
3503 
3504         if (best_node >= 0)
3505                 node_set(best_node, *used_node_mask);
3506 
3507         return best_node;
3508 }
3509 
3510 
3511 /*
3512  * Build zonelists ordered by node and zones within node.
3513  * This results in maximum locality--normal zone overflows into local
3514  * DMA zone, if any--but risks exhausting DMA zone.
3515  */
3516 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
3517 {
3518         int j;
3519         struct zonelist *zonelist;
3520 
3521         zonelist = &pgdat->node_zonelists[0];
3522         for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
3523                 ;
3524         j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3525         zonelist->_zonerefs[j].zone = NULL;
3526         zonelist->_zonerefs[j].zone_idx = 0;
3527 }
3528 
3529 /*
3530  * Build gfp_thisnode zonelists
3531  */
3532 static void build_thisnode_zonelists(pg_data_t *pgdat)
3533 {
3534         int j;
3535         struct zonelist *zonelist;
3536 
3537         zonelist = &pgdat->node_zonelists[1];
3538         j = build_zonelists_node(pgdat, zonelist, 0);
3539         zonelist->_zonerefs[j].zone = NULL;
3540         zonelist->_zonerefs[j].zone_idx = 0;
3541 }
3542 
3543 /*
3544  * Build zonelists ordered by zone and nodes within zones.
3545  * This results in conserving DMA zone[s] until all Normal memory is
3546  * exhausted, but results in overflowing to remote node while memory
3547  * may still exist in local DMA zone.
3548  */
3549 static int node_order[MAX_NUMNODES];
3550 
3551 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3552 {
3553         int pos, j, node;
3554         int zone_type;          /* needs to be signed */
3555         struct zone *z;
3556         struct zonelist *zonelist;
3557 
3558         zonelist = &pgdat->node_zonelists[0];
3559         pos = 0;
3560         for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3561                 for (j = 0; j < nr_nodes; j++) {
3562                         node = node_order[j];
3563                         z = &NODE_DATA(node)->node_zones[zone_type];
3564                         if (populated_zone(z)) {
3565                                 zoneref_set_zone(z,
3566                                         &zonelist->_zonerefs[pos++]);
3567                                 check_highest_zone(zone_type);
3568                         }
3569                 }
3570         }
3571         zonelist->_zonerefs[pos].zone = NULL;
3572         zonelist->_zonerefs[pos].zone_idx = 0;
3573 }
3574 
3575 static int default_zonelist_order(void)
3576 {
3577         int nid, zone_type;
3578         unsigned long low_kmem_size, total_size;
3579         struct zone *z;
3580         int average_size;
3581         /*
3582          * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3583          * If they are really small and used heavily, the system can fall
3584          * into OOM very easily.
3585          * This function detect ZONE_DMA/DMA32 size and configures zone order.
3586          */
3587         /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3588         low_kmem_size = 0;
3589         total_size = 0;
3590         for_each_online_node(nid) {
3591                 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3592                         z = &NODE_DATA(nid)->node_zones[zone_type];
3593                         if (populated_zone(z)) {
3594                                 if (zone_type < ZONE_NORMAL)
3595                                         low_kmem_size += z->managed_pages;
3596                                 total_size += z->managed_pages;
3597                         } else if (zone_type == ZONE_NORMAL) {
3598                                 /*
3599                                  * If any node has only lowmem, then node order
3600                                  * is preferred to allow kernel allocations
3601                                  * locally; otherwise, they can easily infringe
3602                                  * on other nodes when there is an abundance of
3603                                  * lowmem available to allocate from.
3604                                  */
3605                                 return ZONELIST_ORDER_NODE;
3606                         }
3607                 }
3608         }
3609         if (!low_kmem_size ||  /* there are no DMA area. */
3610             low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3611                 return ZONELIST_ORDER_NODE;
3612         /*
3613          * look into each node's config.
3614          * If there is a node whose DMA/DMA32 memory is very big area on
3615          * local memory, NODE_ORDER may be suitable.
3616          */
3617         average_size = total_size /
3618                                 (nodes_weight(node_states[N_MEMORY]) + 1);
3619         for_each_online_node(nid) {
3620                 low_kmem_size = 0;
3621                 total_size = 0;
3622                 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3623                         z = &NODE_DATA(nid)->node_zones[zone_type];
3624                         if (populated_zone(z)) {
3625                                 if (zone_type < ZONE_NORMAL)
3626                                         low_kmem_size += z->present_pages;
3627                                 total_size += z->present_pages;
3628                         }
3629                 }
3630                 if (low_kmem_size &&
3631                     total_size > average_size && /* ignore small node */
3632                     low_kmem_size > total_size * 70/100)
3633                         return ZONELIST_ORDER_NODE;
3634         }
3635         return ZONELIST_ORDER_ZONE;
3636 }
3637 
3638 static void set_zonelist_order(void)
3639 {
3640         if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3641                 current_zonelist_order = default_zonelist_order();
3642         else
3643                 current_zonelist_order = user_zonelist_order;
3644 }
3645 
3646 static void build_zonelists(pg_data_t *pgdat)
3647 {
3648         int j, node, load;
3649         enum zone_type i;
3650         nodemask_t used_mask;
3651         int local_node, prev_node;
3652         struct zonelist *zonelist;
3653         int order = current_zonelist_order;
3654 
3655         /* initialize zonelists */
3656         for (i = 0; i < MAX_ZONELISTS; i++) {
3657                 zonelist = pgdat->node_zonelists + i;
3658                 zonelist->_zonerefs[0].zone = NULL;
3659                 zonelist->_zonerefs[0].zone_idx = 0;
3660         }
3661 
3662         /* NUMA-aware ordering of nodes */
3663         local_node = pgdat->node_id;
3664         load = nr_online_nodes;
3665         prev_node = local_node;
3666         nodes_clear(used_mask);
3667 
3668         memset(node_order, 0, sizeof(node_order));
3669         j = 0;
3670 
3671         while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3672                 /*
3673                  * We don't want to pressure a particular node.
3674                  * So adding penalty to the first node in same
3675                  * distance group to make it round-robin.
3676                  */
3677                 if (node_distance(local_node, node) !=
3678                     node_distance(local_node, prev_node))
3679                         node_load[node] = load;
3680 
3681                 prev_node = node;
3682                 load--;
3683                 if (order == ZONELIST_ORDER_NODE)
3684                         build_zonelists_in_node_order(pgdat, node);
3685                 else
3686                         node_order[j++] = node; /* remember order */
3687         }
3688 
3689         if (order == ZONELIST_ORDER_ZONE) {
3690                 /* calculate node order -- i.e., DMA last! */
3691                 build_zonelists_in_zone_order(pgdat, j);
3692         }
3693 
3694         build_thisnode_zonelists(pgdat);
3695 }
3696 
3697 /* Construct the zonelist performance cache - see further mmzone.h */
3698 static void build_zonelist_cache(pg_data_t *pgdat)
3699 {
3700         struct zonelist *zonelist;
3701         struct zonelist_cache *zlc;
3702         struct zoneref *z;
3703 
3704         zonelist = &pgdat->node_zonelists[0];
3705         zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3706         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3707         for (z = zonelist->_zonerefs; z->zone; z++)
3708                 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3709 }
3710 
3711 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3712 /*
3713  * Return node id of node used for "local" allocations.
3714  * I.e., first node id of first zone in arg node's generic zonelist.
3715  * Used for initializing percpu 'numa_mem', which is used primarily
3716  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3717  */
3718 int local_memory_node(int node)
3719 {
3720         struct zone *zone;
3721 
3722         (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3723                                    gfp_zone(GFP_KERNEL),
3724                                    NULL,
3725                                    &zone);
3726         return zone->node;
3727 }
3728 #endif
3729 
3730 #else   /* CONFIG_NUMA */
3731 
3732 static void set_zonelist_order(void)
3733 {
3734         current_zonelist_order = ZONELIST_ORDER_ZONE;
3735 }
3736 
3737 static void build_zonelists(pg_data_t *pgdat)
3738 {
3739         int node, local_node;
3740         enum zone_type j;
3741         struct zonelist *zonelist;
3742 
3743         local_node = pgdat->node_id;
3744 
3745         zonelist = &pgdat->node_zonelists[0];
3746         j = build_zonelists_node(pgdat, zonelist, 0);
3747 
3748         /*
3749          * Now we build the zonelist so that it contains the zones
3750          * of all the other nodes.
3751          * We don't want to pressure a particular node, so when
3752          * building the zones for node N, we make sure that the
3753          * zones coming right after the local ones are those from
3754          * node N+1 (modulo N)
3755          */
3756         for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3757                 if (!node_online(node))
3758                         continue;
3759                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3760         }
3761         for (node = 0; node < local_node; node++) {
3762                 if (!node_online(node))
3763                         continue;
3764                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3765         }
3766 
3767         zonelist->_zonerefs[j].zone = NULL;
3768         zonelist->_zonerefs[j].zone_idx = 0;
3769 }
3770 
3771 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3772 static void build_zonelist_cache(pg_data_t *pgdat)
3773 {
3774         pgdat->node_zonelists[0].zlcache_ptr = NULL;
3775 }
3776 
3777 #endif  /* CONFIG_NUMA */
3778 
3779 /*
3780  * Boot pageset table. One per cpu which is going to be used for all
3781  * zones and all nodes. The parameters will be set in such a way
3782  * that an item put on a list will immediately be handed over to
3783  * the buddy list. This is safe since pageset manipulation is done
3784  * with interrupts disabled.
3785  *
3786  * The boot_pagesets must be kept even after bootup is complete for
3787  * unused processors and/or zones. They do play a role for bootstrapping
3788  * hotplugged processors.
3789  *
3790  * zoneinfo_show() and maybe other functions do
3791  * not check if the processor is online before following the pageset pointer.
3792  * Other parts of the kernel may not check if the zone is available.
3793  */
3794 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3795 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3796 static void setup_zone_pageset(struct zone *zone);
3797 
3798 /*
3799  * Global mutex to protect against size modification of zonelists
3800  * as well as to serialize pageset setup for the new populated zone.
3801  */
3802 DEFINE_MUTEX(zonelists_mutex);
3803 
3804 /* return values int ....just for stop_machine() */
3805 static int __build_all_zonelists(void *data)
3806 {
3807         int nid;
3808         int cpu;
3809         pg_data_t *self = data;
3810 
3811 #ifdef CONFIG_NUMA
3812         memset(node_load, 0, sizeof(node_load));
3813 #endif
3814 
3815         if (self && !node_online(self->node_id)) {
3816                 build_zonelists(self);
3817                 build_zonelist_cache(self);
3818         }
3819 
3820         for_each_online_node(nid) {
3821                 pg_data_t *pgdat = NODE_DATA(nid);
3822 
3823                 build_zonelists(pgdat);
3824                 build_zonelist_cache(pgdat);
3825         }
3826 
3827         /*
3828          * Initialize the boot_pagesets that are going to be used
3829          * for bootstrapping processors. The real pagesets for
3830          * each zone will be allocated later when the per cpu
3831          * allocator is available.
3832          *
3833          * boot_pagesets are used also for bootstrapping offline
3834          * cpus if the system is already booted because the pagesets
3835          * are needed to initialize allocators on a specific cpu too.
3836          * F.e. the percpu allocator needs the page allocator which
3837          * needs the percpu allocator in order to allocate its pagesets
3838          * (a chicken-egg dilemma).
3839          */
3840         for_each_possible_cpu(cpu) {
3841                 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3842 
3843 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3844                 /*
3845                  * We now know the "local memory node" for each node--
3846                  * i.e., the node of the first zone in the generic zonelist.
3847                  * Set up numa_mem percpu variable for on-line cpus.  During
3848                  * boot, only the boot cpu should be on-line;  we'll init the
3849                  * secondary cpus' numa_mem as they come on-line.  During
3850                  * node/memory hotplug, we'll fixup all on-line cpus.
3851                  */
3852                 if (cpu_online(cpu))
3853                         set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3854 #endif
3855         }
3856 
3857         return 0;
3858 }
3859 
3860 /*
3861  * Called with zonelists_mutex held always
3862  * unless system_state == SYSTEM_BOOTING.
3863  */
3864 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
3865 {
3866         set_zonelist_order();
3867 
3868         if (system_state == SYSTEM_BOOTING) {
3869                 __build_all_zonelists(NULL);
3870                 mminit_verify_zonelist();
3871                 cpuset_init_current_mems_allowed();
3872         } else {
3873 #ifdef CONFIG_MEMORY_HOTPLUG
3874                 if (zone)
3875                         setup_zone_pageset(zone);
3876 #endif
3877                 /* we have to stop all cpus to guarantee there is no user
3878                    of zonelist */
3879                 stop_machine(__build_all_zonelists, pgdat, NULL);
3880                 /* cpuset refresh routine should be here */
3881         }
3882         vm_total_pages = nr_free_pagecache_pages();
3883         /*
3884          * Disable grouping by mobility if the number of pages in the
3885          * system is too low to allow the mechanism to work. It would be
3886          * more accurate, but expensive to check per-zone. This check is
3887          * made on memory-hotadd so a system can start with mobility
3888          * disabled and enable it later
3889          */
3890         if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3891                 page_group_by_mobility_disabled = 1;
3892         else
3893                 page_group_by_mobility_disabled = 0;
3894 
3895         printk("Built %i zonelists in %s order, mobility grouping %s.  "
3896                 "Total pages: %ld\n",
3897                         nr_online_nodes,
3898                         zonelist_order_name[current_zonelist_order],
3899                         page_group_by_mobility_disabled ? "off" : "on",
3900                         vm_total_pages);
3901 #ifdef CONFIG_NUMA
3902         printk("Policy zone: %s\n", zone_names[policy_zone]);
3903 #endif
3904 }
3905 
3906 /*
3907  * Helper functions to size the waitqueue hash table.
3908  * Essentially these want to choose hash table sizes sufficiently
3909  * large so that collisions trying to wait on pages are rare.
3910  * But in fact, the number of active page waitqueues on typical
3911  * systems is ridiculously low, less than 200. So this is even
3912  * conservative, even though it seems large.
3913  *
3914  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3915  * waitqueues, i.e. the size of the waitq table given the number of pages.
3916  */
3917 #define PAGES_PER_WAITQUEUE     256
3918 
3919 #ifndef CONFIG_MEMORY_HOTPLUG
3920 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3921 {
3922         unsigned long size = 1;
3923 
3924         pages /= PAGES_PER_WAITQUEUE;
3925 
3926         while (size < pages)
3927                 size <<= 1;
3928 
3929         /*
3930          * Once we have dozens or even hundreds of threads sleeping
3931          * on IO we've got bigger problems than wait queue collision.
3932          * Limit the size of the wait table to a reasonable size.
3933          */
3934         size = min(size, 4096UL);
3935 
3936         return max(size, 4UL);
3937 }
3938 #else
3939 /*
3940  * A zone's size might be changed by hot-add, so it is not possible to determine
3941  * a suitable size for its wait_table.  So we use the maximum size now.
3942  *
3943  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
3944  *
3945  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
3946  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3947  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
3948  *
3949  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3950  * or more by the traditional way. (See above).  It equals:
3951  *
3952  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
3953  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
3954  *    powerpc (64K page size)             : =  (32G +16M)byte.
3955  */
3956 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3957 {
3958         return 4096UL;
3959 }
3960 #endif
3961 
3962 /*
3963  * This is an integer logarithm so that shifts can be used later
3964  * to extract the more random high bits from the multiplicative
3965  * hash function before the remainder is taken.
3966  */
3967 static inline unsigned long wait_table_bits(unsigned long size)
3968 {
3969         return ffz(~size);
3970 }
3971 
3972 /*
3973  * Check if a pageblock contains reserved pages
3974  */
3975 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3976 {
3977         unsigned long pfn;
3978 
3979         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3980                 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3981                         return 1;
3982         }
3983         return 0;
3984 }
3985 
3986 /*
3987  * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3988  * of blocks reserved is based on min_wmark_pages(zone). The memory within
3989  * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3990  * higher will lead to a bigger reserve which will get freed as contiguous
3991  * blocks as reclaim kicks in
3992  */
3993 static void setup_zone_migrate_reserve(struct zone *zone)
3994 {
3995         unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
3996         struct page *page;
3997         unsigned long block_migratetype;
3998         int reserve;
3999         int old_reserve;
4000 
4001         /*
4002          * Get the start pfn, end pfn and the number of blocks to reserve
4003          * We have to be careful to be aligned to pageblock_nr_pages to
4004          * make sure that we always check pfn_valid for the first page in
4005          * the block.
4006          */
4007         start_pfn = zone->zone_start_pfn;
4008         end_pfn = zone_end_pfn(zone);
4009         start_pfn = roundup(start_pfn, pageblock_nr_pages);
4010         reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
4011                                                         pageblock_order;
4012 
4013         /*
4014          * Reserve blocks are generally in place to help high-order atomic
4015          * allocations that are short-lived. A min_free_kbytes value that
4016          * would result in more than 2 reserve blocks for atomic allocations
4017          * is assumed to be in place to help anti-fragmentation for the
4018          * future allocation of hugepages at runtime.
4019          */
4020         reserve = min(2, reserve);
4021         old_reserve = zone->nr_migrate_reserve_block;
4022 
4023         /* When memory hot-add, we almost always need to do nothing */
4024         if (reserve == old_reserve)
4025                 return;
4026         zone->nr_migrate_reserve_block = reserve;
4027 
4028         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
4029                 if (!pfn_valid(pfn))
4030                         continue;
4031                 page = pfn_to_page(pfn);
4032 
4033                 /* Watch out for overlapping nodes */
4034                 if (page_to_nid(page) != zone_to_nid(zone))
4035                         continue;
4036 
4037                 block_migratetype = get_pageblock_migratetype(page);
4038 
4039                 /* Only test what is necessary when the reserves are not met */
4040                 if (reserve > 0) {
4041                         /*
4042                          * Blocks with reserved pages will never free, skip
4043                          * them.
4044                          */
4045                         block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4046                         if (pageblock_is_reserved(pfn, block_end_pfn))
4047                                 continue;
4048 
4049                         /* If this block is reserved, account for it */
4050                         if (block_migratetype == MIGRATE_RESERVE) {
4051                                 reserve--;
4052                                 continue;
4053                         }
4054 
4055                         /* Suitable for reserving if this block is movable */
4056                         if (block_migratetype == MIGRATE_MOVABLE) {
4057                                 set_pageblock_migratetype(page,
4058                                                         MIGRATE_RESERVE);
4059                                 move_freepages_block(zone, page,
4060                                                         MIGRATE_RESERVE);
4061                                 reserve--;
4062                                 continue;
4063                         }
4064                 } else if (!old_reserve) {
4065                         /*
4066                          * At boot time we don't need to scan the whole zone
4067                          * for turning off MIGRATE_RESERVE.
4068                          */
4069                         break;
4070                 }
4071 
4072                 /*
4073                  * If the reserve is met and this is a previous reserved block,
4074                  * take it back
4075                  */
4076                 if (block_migratetype == MIGRATE_RESERVE) {
4077                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4078                         move_freepages_block(zone, page, MIGRATE_MOVABLE);
4079                 }
4080         }
4081 }
4082 
4083 /*
4084  * Initially all pages are reserved - free ones are freed
4085  * up by free_all_bootmem() once the early boot process is
4086  * done. Non-atomic initialization, single-pass.
4087  */
4088 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4089                 unsigned long start_pfn, enum memmap_context context)
4090 {
4091         struct page *page;
4092         unsigned long end_pfn = start_pfn + size;
4093         unsigned long pfn;
4094         struct zone *z;
4095 
4096         if (highest_memmap_pfn < end_pfn - 1)
4097                 highest_memmap_pfn = end_pfn - 1;
4098 
4099         z = &NODE_DATA(nid)->node_zones[zone];
4100         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4101                 /*
4102                  * There can be holes in boot-time mem_map[]s
4103                  * handed to this function.  They do not
4104                  * exist on hotplugged memory.
4105                  */
4106                 if (context == MEMMAP_EARLY) {
4107                         if (!early_pfn_valid(pfn))
4108                                 continue;
4109                         if (!early_pfn_in_nid(pfn, nid))
4110                                 continue;
4111                 }
4112                 page = pfn_to_page(pfn);
4113                 set_page_links(page, zone, nid, pfn);
4114                 mminit_verify_page_links(page, zone, nid, pfn);
4115                 init_page_count(page);
4116                 page_mapcount_reset(page);
4117                 page_cpupid_reset_last(page);
4118                 SetPageReserved(page);
4119                 /*
4120                  * Mark the block movable so that blocks are reserved for
4121                  * movable at startup. This will force kernel allocations
4122                  * to reserve their blocks rather than leaking throughout
4123                  * the address space during boot when many long-lived
4124                  * kernel allocations are made. Later some blocks near
4125                  * the start are marked MIGRATE_RESERVE by
4126                  * setup_zone_migrate_reserve()
4127                  *
4128                  * bitmap is created for zone's valid pfn range. but memmap
4129                  * can be created for invalid pages (for alignment)
4130                  * check here not to call set_pageblock_migratetype() against
4131                  * pfn out of zone.
4132                  */
4133                 if ((z->zone_start_pfn <= pfn)
4134                     && (pfn < zone_end_pfn(z))
4135                     && !(pfn & (pageblock_nr_pages - 1)))
4136                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4137 
4138                 INIT_LIST_HEAD(&page->lru);
4139 #ifdef WANT_PAGE_VIRTUAL
4140                 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4141                 if (!is_highmem_idx(zone))
4142                         set_page_address(page, __va(pfn << PAGE_SHIFT));
4143 #endif
4144         }
4145 }
4146 
4147 static void __meminit zone_init_free_lists(struct zone *zone)
4148 {
4149         unsigned int order, t;
4150         for_each_migratetype_order(order, t) {
4151                 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4152                 zone->free_area[order].nr_free = 0;
4153         }
4154 }
4155 
4156 #ifndef __HAVE_ARCH_MEMMAP_INIT
4157 #define memmap_init(size, nid, zone, start_pfn) \
4158         memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4159 #endif
4160 
4161 static int zone_batchsize(struct zone *zone)
4162 {
4163 #ifdef CONFIG_MMU
4164         int batch;
4165 
4166         /*
4167          * The per-cpu-pages pools are set to around 1000th of the
4168          * size of the zone.  But no more than 1/2 of a meg.
4169          *
4170          * OK, so we don't know how big the cache is.  So guess.
4171          */
4172         batch = zone->managed_pages / 1024;
4173         if (batch * PAGE_SIZE > 512 * 1024)
4174                 batch = (512 * 1024) / PAGE_SIZE;
4175         batch /= 4;             /* We effectively *= 4 below */
4176         if (batch < 1)
4177                 batch = 1;
4178 
4179         /*
4180          * Clamp the batch to a 2^n - 1 value. Having a power
4181          * of 2 value was found to be more likely to have
4182          * suboptimal cache aliasing properties in some cases.
4183          *
4184          * For example if 2 tasks are alternately allocating
4185          * batches of pages, one task can end up with a lot
4186          * of pages of one half of the possible page colors
4187          * and the other with pages of the other colors.
4188          */
4189         batch = rounddown_pow_of_two(batch + batch/2) - 1;
4190 
4191         return batch;
4192 
4193 #else
4194         /* The deferral and batching of frees should be suppressed under NOMMU
4195          * conditions.
4196          *
4197          * The problem is that NOMMU needs to be able to allocate large chunks
4198          * of contiguous memory as there's no hardware page translation to
4199          * assemble apparent contiguous memory from discontiguous pages.
4200          *
4201          * Queueing large contiguous runs of pages for batching, however,
4202          * causes the pages to actually be freed in smaller chunks.  As there
4203          * can be a significant delay between the individual batches being
4204          * recycled, this leads to the once large chunks of space being
4205          * fragmented and becoming unavailable for high-order allocations.
4206          */
4207         return 0;
4208 #endif
4209 }
4210 
4211 /*
4212  * pcp->high and pcp->batch values are related and dependent on one another:
4213  * ->batch must never be higher then ->high.
4214  * The following function updates them in a safe manner without read side
4215  * locking.
4216  *
4217  * Any new users of pcp->batch and pcp->high should ensure they can cope with
4218  * those fields changing asynchronously (acording the the above rule).
4219  *
4220  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4221  * outside of boot time (or some other assurance that no concurrent updaters
4222  * exist).
4223  */
4224 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4225                 unsigned long batch)
4226 {
4227        /* start with a fail safe value for batch */
4228         pcp->batch = 1;
4229         smp_wmb();
4230 
4231        /* Update high, then batch, in order */
4232         pcp->high = high;
4233         smp_wmb();
4234 
4235         pcp->batch = batch;
4236 }
4237 
4238 /* a companion to pageset_set_high() */
4239 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4240 {
4241         pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4242 }
4243 
4244 static void pageset_init(struct per_cpu_pageset *p)
4245 {
4246         struct per_cpu_pages *pcp;
4247         int migratetype;
4248 
4249         memset(p, 0, sizeof(*p));
4250 
4251         pcp = &p->pcp;
4252         pcp->count = 0;
4253         for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4254                 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4255 }
4256 
4257 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4258 {
4259         pageset_init(p);
4260         pageset_set_batch(p, batch);
4261 }
4262 
4263 /*
4264  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4265  * to the value high for the pageset p.
4266  */
4267 static void pageset_set_high(struct per_cpu_pageset *p,
4268                                 unsigned long high)
4269 {
4270         unsigned long batch = max(1UL, high / 4);
4271         if ((high / 4) > (PAGE_SHIFT * 8))
4272                 batch = PAGE_SHIFT * 8;
4273 
4274         pageset_update(&p->pcp, high, batch);
4275 }
4276 
4277 static void pageset_set_high_and_batch(struct zone *zone,
4278                                        struct per_cpu_pageset *pcp)
4279 {
4280         if (percpu_pagelist_fraction)
4281                 pageset_set_high(pcp,
4282                         (zone->managed_pages /
4283                                 percpu_pagelist_fraction));
4284         else
4285                 pageset_set_batch(pcp, zone_batchsize(zone));
4286 }
4287 
4288 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4289 {
4290         struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4291 
4292         pageset_init(pcp);
4293         pageset_set_high_and_batch(zone, pcp);
4294 }
4295 
4296 static void __meminit setup_zone_pageset(struct zone *zone)
4297 {
4298         int cpu;
4299         zone->pageset = alloc_percpu(struct per_cpu_pageset);
4300         for_each_possible_cpu(cpu)
4301                 zone_pageset_init(zone, cpu);
4302 }
4303 
4304 /*
4305  * Allocate per cpu pagesets and initialize them.
4306  * Before this call only boot pagesets were available.
4307  */
4308 void __init setup_per_cpu_pageset(void)
4309 {
4310         struct zone *zone;
4311 
4312         for_each_populated_zone(zone)
4313                 setup_zone_pageset(zone);
4314 }
4315 
4316 static noinline __init_refok
4317 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4318 {
4319         int i;
4320         size_t alloc_size;
4321 
4322         /*
4323          * The per-page waitqueue mechanism uses hashed waitqueues
4324          * per zone.
4325          */
4326         zone->wait_table_hash_nr_entries =
4327                  wait_table_hash_nr_entries(zone_size_pages);
4328         zone->wait_table_bits =
4329                 wait_table_bits(zone->wait_table_hash_nr_entries);
4330         alloc_size = zone->wait_table_hash_nr_entries
4331                                         * sizeof(wait_queue_head_t);
4332 
4333         if (!slab_is_available()) {
4334                 zone->wait_table = (wait_queue_head_t *)
4335                         memblock_virt_alloc_node_nopanic(
4336                                 alloc_size, zone->zone_pgdat->node_id);
4337         } else {
4338                 /*
4339                  * This case means that a zone whose size was 0 gets new memory
4340                  * via memory hot-add.
4341                  * But it may be the case that a new node was hot-added.  In
4342                  * this case vmalloc() will not be able to use this new node's
4343                  * memory - this wait_table must be initialized to use this new
4344                  * node itself as well.
4345                  * To use this new node's memory, further consideration will be
4346                  * necessary.
4347                  */
4348                 zone->wait_table = vmalloc(alloc_size);
4349         }
4350         if (!zone->wait_table)
4351                 return -ENOMEM;
4352 
4353         for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4354                 init_waitqueue_head(zone->wait_table + i);
4355 
4356         return 0;
4357 }
4358 
4359 static __meminit void zone_pcp_init(struct zone *zone)
4360 {
4361         /*
4362          * per cpu subsystem is not up at this point. The following code
4363          * relies on the ability of the linker to provide the
4364          * offset of a (static) per cpu variable into the per cpu area.
4365          */
4366         zone->pageset = &boot_pageset;
4367 
4368         if (populated_zone(zone))
4369                 printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
4370                         zone->name, zone->present_pages,
4371                                          zone_batchsize(zone));
4372 }
4373 
4374 int __meminit init_currently_empty_zone(struct zone *zone,
4375                                         unsigned long zone_start_pfn,
4376                                         unsigned long size,
4377                                         enum memmap_context context)
4378 {
4379         struct pglist_data *pgdat = zone->zone_pgdat;
4380         int ret;
4381         ret = zone_wait_table_init(zone, size);
4382         if (ret)
4383                 return ret;
4384         pgdat->nr_zones = zone_idx(zone) + 1;
4385 
4386         zone->zone_start_pfn = zone_start_pfn;
4387 
4388         mminit_dprintk(MMINIT_TRACE, "memmap_init",
4389                         "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4390                         pgdat->node_id,
4391                         (unsigned long)zone_idx(zone),
4392                         zone_start_pfn, (zone_start_pfn + size));
4393 
4394         zone_init_free_lists(zone);
4395 
4396         return 0;
4397 }
4398 
4399 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4400 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4401 /*
4402  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4403  */
4404 int __meminit __early_pfn_to_nid(unsigned long pfn)
4405 {
4406         unsigned long start_pfn, end_pfn;
4407         int nid;
4408         /*
4409          * NOTE: The following SMP-unsafe globals are only used early in boot
4410          * when the kernel is running single-threaded.
4411          */
4412         static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4413         static int __meminitdata last_nid;
4414 
4415         if (last_start_pfn <= pfn && pfn < last_end_pfn)
4416                 return last_nid;
4417 
4418         nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4419         if (nid != -1) {
4420                 last_start_pfn = start_pfn;
4421                 last_end_pfn = end_pfn;
4422                 last_nid = nid;
4423         }
4424 
4425         return nid;
4426 }
4427 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4428 
4429 int __meminit early_pfn_to_nid(unsigned long pfn)
4430 {
4431         int nid;
4432 
4433         nid = __early_pfn_to_nid(pfn);
4434         if (nid >= 0)
4435                 return nid;
4436         /* just returns 0 */
4437         return 0;
4438 }
4439 
4440 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4441 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4442 {
4443         int nid;
4444 
4445         nid = __early_pfn_to_nid(pfn);
4446         if (nid >= 0 && nid != node)
4447                 return false;
4448         return true;
4449 }
4450 #endif
4451 
4452 /**
4453  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4454  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4455  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4456  *
4457  * If an architecture guarantees that all ranges registered contain no holes
4458  * and may be freed, this this function may be used instead of calling
4459  * memblock_free_early_nid() manually.
4460  */
4461 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
4462 {
4463         unsigned long start_pfn, end_pfn;
4464         int i, this_nid;
4465 
4466         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4467                 start_pfn = min(start_pfn, max_low_pfn);
4468                 end_pfn = min(end_pfn, max_low_pfn);
4469 
4470                 if (start_pfn < end_pfn)
4471                         memblock_free_early_nid(PFN_PHYS(start_pfn),
4472                                         (end_pfn - start_pfn) << PAGE_SHIFT,
4473                                         this_nid);
4474         }
4475 }
4476 
4477 /**
4478  * sparse_memory_present_with_active_regions - Call memory_present for each active range
4479  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4480  *
4481  * If an architecture guarantees that all ranges registered contain no holes and may
4482  * be freed, this function may be used instead of calling memory_present() manually.
4483  */
4484 void __init sparse_memory_present_with_active_regions(int nid)
4485 {
4486         unsigned long start_pfn, end_pfn;
4487         int i, this_nid;
4488 
4489         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4490                 memory_present(this_nid, start_pfn, end_pfn);
4491 }
4492 
4493 /**
4494  * get_pfn_range_for_nid - Return the start and end page frames for a node
4495  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4496  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4497  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4498  *
4499  * It returns the start and end page frame of a node based on information
4500  * provided by memblock_set_node(). If called for a node
4501  * with no available memory, a warning is printed and the start and end
4502  * PFNs will be 0.
4503  */
4504 void __meminit get_pfn_range_for_nid(unsigned int nid,
4505                         unsigned long *start_pfn, unsigned long *end_pfn)
4506 {
4507         unsigned long this_start_pfn, this_end_pfn;
4508         int i;
4509 
4510         *start_pfn = -1UL;
4511         *end_pfn = 0;
4512 
4513         for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4514                 *start_pfn = min(*start_pfn, this_start_pfn);
4515                 *end_pfn = max(*end_pfn, this_end_pfn);
4516         }
4517 
4518         if (*start_pfn == -1UL)
4519                 *start_pfn = 0;
4520 }
4521 
4522 /*
4523  * This finds a zone that can be used for ZONE_MOVABLE pages. The
4524  * assumption is made that zones within a node are ordered in monotonic
4525  * increasing memory addresses so that the "highest" populated zone is used
4526  */
4527 static void __init find_usable_zone_for_movable(void)
4528 {
4529         int zone_index;
4530         for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4531                 if (zone_index == ZONE_MOVABLE)
4532                         continue;
4533 
4534                 if (arch_zone_highest_possible_pfn[zone_index] >
4535                                 arch_zone_lowest_possible_pfn[zone_index])
4536                         break;
4537         }
4538 
4539         VM_BUG_ON(zone_index == -1);
4540         movable_zone = zone_index;
4541 }
4542 
4543 /*
4544  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4545  * because it is sized independent of architecture. Unlike the other zones,
4546  * the starting point for ZONE_MOVABLE is not fixed. It may be different
4547  * in each node depending on the size of each node and how evenly kernelcore
4548  * is distributed. This helper function adjusts the zone ranges
4549  * provided by the architecture for a given node by using the end of the
4550  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4551  * zones within a node are in order of monotonic increases memory addresses
4552  */
4553 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4554                                         unsigned long zone_type,
4555                                         unsigned long node_start_pfn,
4556                                         unsigned long node_end_pfn,
4557                                         unsigned long *zone_start_pfn,
4558                                         unsigned long *zone_end_pfn)
4559 {
4560         /* Only adjust if ZONE_MOVABLE is on this node */
4561         if (zone_movable_pfn[nid]) {
4562                 /* Size ZONE_MOVABLE */
4563                 if (zone_type == ZONE_MOVABLE) {
4564                         *zone_start_pfn = zone_movable_pfn[nid];
4565                         *zone_end_pfn = min(node_end_pfn,
4566                                 arch_zone_highest_possible_pfn[movable_zone]);
4567 
4568                 /* Adjust for ZONE_MOVABLE starting within this range */
4569                 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4570                                 *zone_end_pfn > zone_movable_pfn[nid]) {
4571                         *zone_end_pfn = zone_movable_pfn[nid];
4572 
4573                 /* Check if this whole range is within ZONE_MOVABLE */
4574                 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4575                         *zone_start_pfn = *zone_end_pfn;
4576         }
4577 }
4578 
4579 /*
4580  * Return the number of pages a zone spans in a node, including holes
4581  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4582  */
4583 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4584                                         unsigned long zone_type,
4585                                         unsigned long node_start_pfn,
4586                                         unsigned long node_end_pfn,
4587                                         unsigned long *ignored)
4588 {
4589         unsigned long zone_start_pfn, zone_end_pfn;
4590 
4591         /* Get the start and end of the zone */
4592         zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4593         zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4594         adjust_zone_range_for_zone_movable(nid, zone_type,
4595                                 node_start_pfn, node_end_pfn,
4596                                 &zone_start_pfn, &zone_end_pfn);
4597 
4598         /* Check that this node has pages within the zone's required range */
4599         if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4600                 return 0;
4601 
4602         /* Move the zone boundaries inside the node if necessary */
4603         zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4604         zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4605 
4606         /* Return the spanned pages */
4607         return zone_end_pfn - zone_start_pfn;
4608 }
4609 
4610 /*
4611  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4612  * then all holes in the requested range will be accounted for.
4613  */
4614 unsigned long __meminit __absent_pages_in_range(int nid,
4615                                 unsigned long range_start_pfn,
4616                                 unsigned long range_end_pfn)
4617 {
4618         unsigned long nr_absent = range_end_pfn - range_start_pfn;
4619         unsigned long start_pfn, end_pfn;
4620         int i;
4621 
4622         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4623                 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4624                 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4625                 nr_absent -= end_pfn - start_pfn;
4626         }
4627         return nr_absent;
4628 }
4629 
4630 /**
4631  * absent_pages_in_range - Return number of page frames in holes within a range
4632  * @start_pfn: The start PFN to start searching for holes
4633  * @end_pfn: The end PFN to stop searching for holes
4634  *
4635  * It returns the number of pages frames in memory holes within a range.
4636  */
4637 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4638                                                         unsigned long end_pfn)
4639 {
4640         return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4641 }
4642 
4643 /* Return the number of page frames in holes in a zone on a node */
4644 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4645                                         unsigned long zone_type,
4646                                         unsigned long node_start_pfn,
4647                                         unsigned long node_end_pfn,
4648                                         unsigned long *ignored)
4649 {
4650         unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4651         unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
4652         unsigned long zone_start_pfn, zone_end_pfn;
4653 
4654         zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4655         zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
4656 
4657         adjust_zone_range_for_zone_movable(nid, zone_type,
4658                         node_start_pfn, node_end_pfn,
4659                         &zone_start_pfn, &zone_end_pfn);
4660         return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4661 }
4662 
4663 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4664 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4665                                         unsigned long zone_type,
4666                                         unsigned long node_start_pfn,
4667                                         unsigned long node_end_pfn,
4668                                         unsigned long *zones_size)
4669 {
4670         return zones_size[zone_type];
4671 }
4672 
4673 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4674                                                 unsigned long zone_type,
4675                                                 unsigned long node_start_pfn,
4676                                                 unsigned long node_end_pfn,
4677                                                 unsigned long *zholes_size)
4678 {
4679         if (!zholes_size)
4680                 return 0;
4681 
4682         return zholes_size[zone_type];
4683 }
4684 
4685 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4686 
4687 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4688                                                 unsigned long node_start_pfn,
4689                                                 unsigned long node_end_pfn,
4690                                                 unsigned long *zones_size,
4691                                                 unsigned long *zholes_size)
4692 {
4693         unsigned long realtotalpages, totalpages = 0;
4694         enum zone_type i;
4695 
4696         for (i = 0; i < MAX_NR_ZONES; i++)
4697                 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4698                                                          node_start_pfn,
4699                                                          node_end_pfn,
4700                                                          zones_size);
4701         pgdat->node_spanned_pages = totalpages;
4702 
4703         realtotalpages = totalpages;
4704         for (i = 0; i < MAX_NR_ZONES; i++)
4705                 realtotalpages -=
4706                         zone_absent_pages_in_node(pgdat->node_id, i,
4707                                                   node_start_pfn, node_end_pfn,
4708                                                   zholes_size);
4709         pgdat->node_present_pages = realtotalpages;
4710         printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4711                                                         realtotalpages);
4712 }
4713 
4714 #ifndef CONFIG_SPARSEMEM
4715 /*
4716  * Calculate the size of the zone->blockflags rounded to an unsigned long
4717  * Start by making sure zonesize is a multiple of pageblock_order by rounding
4718  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4719  * round what is now in bits to nearest long in bits, then return it in
4720  * bytes.
4721  */
4722 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
4723 {
4724         unsigned long usemapsize;
4725 
4726         zonesize += zone_start_pfn & (pageblock_nr_pages-1);
4727         usemapsize = roundup(zonesize, pageblock_nr_pages);
4728         usemapsize = usemapsize >> pageblock_order;
4729         usemapsize *= NR_PAGEBLOCK_BITS;
4730         usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4731 
4732         return usemapsize / 8;
4733 }
4734 
4735 static void __init setup_usemap(struct pglist_data *pgdat,
4736                                 struct zone *zone,
4737                                 unsigned long zone_start_pfn,
4738                                 unsigned long zonesize)
4739 {
4740         unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
4741         zone->pageblock_flags = NULL;
4742         if (usemapsize)
4743                 zone->pageblock_flags =
4744                         memblock_virt_alloc_node_nopanic(usemapsize,
4745                                                          pgdat->node_id);
4746 }
4747 #else
4748 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4749                                 unsigned long zone_start_pfn, unsigned long zonesize) {}
4750 #endif /* CONFIG_SPARSEMEM */
4751 
4752 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4753 
4754 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4755 void __paginginit set_pageblock_order(void)
4756 {
4757         unsigned int order;
4758 
4759         /* Check that pageblock_nr_pages has not already been setup */
4760         if (pageblock_order)
4761                 return;
4762 
4763         if (HPAGE_SHIFT > PAGE_SHIFT)
4764                 order = HUGETLB_PAGE_ORDER;
4765         else
4766                 order = MAX_ORDER - 1;
4767 
4768         /*
4769          * Assume the largest contiguous order of interest is a huge page.
4770          * This value may be variable depending on boot parameters on IA64 and
4771          * powerpc.
4772          */
4773         pageblock_order = order;
4774 }
4775 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4776 
4777 /*
4778  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4779  * is unused as pageblock_order is set at compile-time. See
4780  * include/linux/pageblock-flags.h for the values of pageblock_order based on
4781  * the kernel config
4782  */
4783 void __paginginit set_pageblock_order(void)
4784 {
4785 }
4786 
4787 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4788 
4789 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4790                                                    unsigned long present_pages)
4791 {
4792         unsigned long pages = spanned_pages;
4793 
4794         /*
4795          * Provide a more accurate estimation if there are holes within
4796          * the zone and SPARSEMEM is in use. If there are holes within the
4797          * zone, each populated memory region may cost us one or two extra
4798          * memmap pages due to alignment because memmap pages for each
4799          * populated regions may not naturally algined on page boundary.
4800          * So the (present_pages >> 4) heuristic is a tradeoff for that.
4801          */
4802         if (spanned_pages > present_pages + (present_pages >> 4) &&
4803             IS_ENABLED(CONFIG_SPARSEMEM))
4804                 pages = present_pages;
4805 
4806         return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4807 }
4808 
4809 /*
4810  * Set up the zone data structures:
4811  *   - mark all pages reserved
4812  *   - mark all memory queues empty
4813  *   - clear the memory bitmaps
4814  *
4815  * NOTE: pgdat should get zeroed by caller.
4816  */
4817 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4818                 unsigned long node_start_pfn, unsigned long node_end_pfn,
4819                 unsigned long *zones_size, unsigned long *zholes_size)
4820 {
4821         enum zone_type j;
4822         int nid = pgdat->node_id;
4823         unsigned long zone_start_pfn = pgdat->node_start_pfn;
4824         int ret;
4825 
4826         pgdat_resize_init(pgdat);
4827 #ifdef CONFIG_NUMA_BALANCING
4828         spin_lock_init(&pgdat->numabalancing_migrate_lock);
4829         pgdat->numabalancing_migrate_nr_pages = 0;
4830         pgdat->numabalancing_migrate_next_window = jiffies;
4831 #endif
4832         init_waitqueue_head(&pgdat->kswapd_wait);
4833         init_waitqueue_head(&pgdat->pfmemalloc_wait);
4834         pgdat_page_cgroup_init(pgdat);
4835 
4836         for (j = 0; j < MAX_NR_ZONES; j++) {
4837                 struct zone *zone = pgdat->node_zones + j;
4838                 unsigned long size, realsize, freesize, memmap_pages;
4839 
4840                 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4841                                                   node_end_pfn, zones_size);
4842                 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
4843                                                                 node_start_pfn,
4844                                                                 node_end_pfn,
4845                                                                 zholes_size);
4846 
4847                 /*
4848                  * Adjust freesize so that it accounts for how much memory
4849                  * is used by this zone for memmap. This affects the watermark
4850                  * and per-cpu initialisations
4851                  */
4852                 memmap_pages = calc_memmap_size(size, realsize);
4853                 if (freesize >= memmap_pages) {
4854                         freesize -= memmap_pages;
4855                         if (memmap_pages)
4856                                 printk(KERN_DEBUG
4857                                        "  %s zone: %lu pages used for memmap\n",
4858                                        zone_names[j], memmap_pages);
4859                 } else
4860                         printk(KERN_WARNING
4861                                 "  %s zone: %lu pages exceeds freesize %lu\n",
4862                                 zone_names[j], memmap_pages, freesize);
4863 
4864                 /* Account for reserved pages */
4865                 if (j == 0 && freesize > dma_reserve) {
4866                         freesize -= dma_reserve;
4867                         printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
4868                                         zone_names[0], dma_reserve);
4869                 }
4870 
4871                 if (!is_highmem_idx(j))
4872                         nr_kernel_pages += freesize;
4873                 /* Charge for highmem memmap if there are enough kernel pages */
4874                 else if (nr_kernel_pages > memmap_pages * 2)
4875                         nr_kernel_pages -= memmap_pages;
4876                 nr_all_pages += freesize;
4877 
4878                 zone->spanned_pages = size;
4879                 zone->present_pages = realsize;
4880                 /*
4881                  * Set an approximate value for lowmem here, it will be adjusted
4882                  * when the bootmem allocator frees pages into the buddy system.
4883                  * And all highmem pages will be managed by the buddy system.
4884                  */
4885                 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
4886 #ifdef CONFIG_NUMA
4887                 zone->node = nid;
4888                 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
4889                                                 / 100;
4890                 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
4891 #endif
4892                 zone->name = zone_names[j];
4893                 spin_lock_init(&zone->lock);
4894                 spin_lock_init(&zone->lru_lock);
4895                 zone_seqlock_init(zone);
4896                 zone->zone_pgdat = pgdat;
4897                 zone_pcp_init(zone);
4898 
4899                 /* For bootup, initialized properly in watermark setup */
4900                 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4901 
4902                 lruvec_init(&zone->lruvec);
4903                 if (!size)
4904                         continue;
4905 
4906                 set_pageblock_order();
4907                 setup_usemap(pgdat, zone, zone_start_pfn, size);
4908                 ret = init_currently_empty_zone(zone, zone_start_pfn,
4909                                                 size, MEMMAP_EARLY);
4910                 BUG_ON(ret);
4911                 memmap_init(size, nid, j, zone_start_pfn);
4912                 zone_start_pfn += size;
4913         }
4914 }
4915 
4916 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4917 {
4918         /* Skip empty nodes */
4919         if (!pgdat->node_spanned_pages)
4920                 return;
4921 
4922 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4923         /* ia64 gets its own node_mem_map, before this, without bootmem */
4924         if (!pgdat->node_mem_map) {
4925                 unsigned long size, start, end;
4926                 struct page *map;
4927 
4928                 /*
4929                  * The zone's endpoints aren't required to be MAX_ORDER
4930                  * aligned but the node_mem_map endpoints must be in order
4931                  * for the buddy allocator to function correctly.
4932                  */
4933                 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4934                 end = pgdat_end_pfn(pgdat);
4935                 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4936                 size =  (end - start) * sizeof(struct page);
4937                 map = alloc_remap(pgdat->node_id, size);
4938                 if (!map)
4939                         map = memblock_virt_alloc_node_nopanic(size,
4940                                                                pgdat->node_id);
4941                 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4942         }
4943 #ifndef CONFIG_NEED_MULTIPLE_NODES
4944         /*
4945          * With no DISCONTIG, the global mem_map is just set as node 0's
4946          */
4947         if (pgdat == NODE_DATA(0)) {
4948                 mem_map = NODE_DATA(0)->node_mem_map;
4949 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4950                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4951                         mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4952 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4953         }
4954 #endif
4955 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4956 }
4957 
4958 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4959                 unsigned long node_start_pfn, unsigned long *zholes_size)
4960 {
4961         pg_data_t *pgdat = NODE_DATA(nid);
4962         unsigned long start_pfn = 0;
4963         unsigned long end_pfn = 0;
4964 
4965         /* pg_data_t should be reset to zero when it's allocated */
4966         WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
4967 
4968         pgdat->node_id = nid;
4969         pgdat->node_start_pfn = node_start_pfn;
4970 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4971         get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
4972 #endif
4973         calculate_node_totalpages(pgdat, start_pfn, end_pfn,
4974                                   zones_size, zholes_size);
4975 
4976         alloc_node_mem_map(pgdat);
4977 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4978         printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4979                 nid, (unsigned long)pgdat,
4980                 (unsigned long)pgdat->node_mem_map);
4981 #endif
4982 
4983         free_area_init_core(pgdat, start_pfn, end_pfn,
4984                             zones_size, zholes_size);
4985 }
4986 
4987 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4988 
4989 #if MAX_NUMNODES > 1
4990 /*
4991  * Figure out the number of possible node ids.
4992  */
4993 void __init setup_nr_node_ids(void)
4994 {
4995         unsigned int node;
4996         unsigned int highest = 0;
4997 
4998         for_each_node_mask(node, node_possible_map)
4999                 highest = node;
5000         nr_node_ids = highest + 1;
5001 }
5002 #endif
5003 
5004 /**
5005  * node_map_pfn_alignment - determine the maximum internode alignment
5006  *
5007  * This function should be called after node map is populated and sorted.
5008  * It calculates the maximum power of two alignment which can distinguish
5009  * all the nodes.
5010  *
5011  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5012  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
5013  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
5014  * shifted, 1GiB is enough and this function will indicate so.
5015  *
5016  * This is used to test whether pfn -> nid mapping of the chosen memory
5017  * model has fine enough granularity to avoid incorrect mapping for the
5018  * populated node map.
5019  *
5020  * Returns the determined alignment in pfn's.  0 if there is no alignment
5021  * requirement (single node).
5022  */
5023 unsigned long __init node_map_pfn_alignment(void)
5024 {
5025         unsigned long accl_mask = 0, last_end = 0;
5026         unsigned long start, end, mask;
5027         int last_nid = -1;
5028         int i, nid;
5029 
5030         for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5031                 if (!start || last_nid < 0 || last_nid == nid) {
5032                         last_nid = nid;
5033                         last_end = end;
5034                         continue;
5035                 }
5036 
5037                 /*
5038                  * Start with a mask granular enough to pin-point to the
5039                  * start pfn and tick off bits one-by-one until it becomes
5040                  * too coarse to separate the current node from the last.
5041                  */
5042                 mask = ~((1 << __ffs(start)) - 1);
5043                 while (mask && last_end <= (start & (mask << 1)))
5044                         mask <<= 1;
5045 
5046                 /* accumulate all internode masks */
5047                 accl_mask |= mask;
5048         }
5049 
5050         /* convert mask to number of pages */
5051         return ~accl_mask + 1;
5052 }
5053 
5054 /* Find the lowest pfn for a node */
5055 static unsigned long __init find_min_pfn_for_node(int nid)
5056 {
5057         unsigned long min_pfn = ULONG_MAX;
5058         unsigned long start_pfn;
5059         int i;
5060 
5061         for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5062                 min_pfn = min(min_pfn, start_pfn);
5063 
5064         if (min_pfn == ULONG_MAX) {
5065                 printk(KERN_WARNING
5066                         "Could not find start_pfn for node %d\n", nid);
5067                 return 0;
5068         }
5069 
5070         return min_pfn;
5071 }
5072 
5073 /**
5074  * find_min_pfn_with_active_regions - Find the minimum PFN registered
5075  *
5076  * It returns the minimum PFN based on information provided via
5077  * memblock_set_node().
5078  */
5079 unsigned long __init find_min_pfn_with_active_regions(void)
5080 {
5081         return find_min_pfn_for_node(MAX_NUMNODES);
5082 }
5083 
5084 /*
5085  * early_calculate_totalpages()
5086  * Sum pages in active regions for movable zone.
5087  * Populate N_MEMORY for calculating usable_nodes.
5088  */
5089 static unsigned long __init early_calculate_totalpages(void)
5090 {
5091         unsigned long totalpages = 0;
5092         unsigned long start_pfn, end_pfn;
5093         int i, nid;
5094 
5095         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5096                 unsigned long pages = end_pfn - start_pfn;
5097 
5098                 totalpages += pages;
5099                 if (pages)
5100                         node_set_state(nid, N_MEMORY);
5101         }
5102         return totalpages;
5103 }
5104 
5105 /*
5106  * Find the PFN the Movable zone begins in each node. Kernel memory
5107  * is spread evenly between nodes as long as the nodes have enough
5108  * memory. When they don't, some nodes will have more kernelcore than
5109  * others
5110  */
5111 static void __init find_zone_movable_pfns_for_nodes(void)
5112 {
5113         int i, nid;
5114         unsigned long usable_startpfn;
5115         unsigned long kernelcore_node, kernelcore_remaining;
5116         /* save the state before borrow the nodemask */
5117         nodemask_t saved_node_state = node_states[N_MEMORY];
5118         unsigned long totalpages = early_calculate_totalpages();
5119         int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5120         struct memblock_region *r;
5121 
5122         /* Need to find movable_zone earlier when movable_node is specified. */
5123         find_usable_zone_for_movable();
5124 
5125         /*
5126          * If movable_node is specified, ignore kernelcore and movablecore
5127          * options.
5128          */
5129         if (movable_node_is_enabled()) {
5130                 for_each_memblock(memory, r) {
5131                         if (!memblock_is_hotpluggable(r))
5132                                 continue;
5133 
5134                         nid = r->nid;
5135 
5136                         usable_startpfn = PFN_DOWN(r->base);
5137                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5138                                 min(usable_startpfn, zone_movable_pfn[nid]) :
5139                                 usable_startpfn;
5140                 }
5141 
5142                 goto out2;
5143         }
5144 
5145         /*
5146          * If movablecore=nn[KMG] was specified, calculate what size of
5147          * kernelcore that corresponds so that memory usable for
5148          * any allocation type is evenly spread. If both kernelcore
5149          * and movablecore are specified, then the value of kernelcore
5150          * will be used for required_kernelcore if it's greater than
5151          * what movablecore would have allowed.
5152          */
5153         if (required_movablecore) {
5154                 unsigned long corepages;
5155 
5156                 /*
5157                  * Round-up so that ZONE_MOVABLE is at least as large as what
5158                  * was requested by the user
5159                  */
5160                 required_movablecore =
5161                         roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5162                 corepages = totalpages - required_movablecore;
5163 
5164                 required_kernelcore = max(required_kernelcore, corepages);
5165         }
5166 
5167         /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5168         if (!required_kernelcore)
5169                 goto out;
5170 
5171         /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5172         usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5173 
5174 restart:
5175         /* Spread kernelcore memory as evenly as possible throughout nodes */
5176         kernelcore_node = required_kernelcore / usable_nodes;
5177         for_each_node_state(nid, N_MEMORY) {
5178                 unsigned long start_pfn, end_pfn;
5179 
5180                 /*
5181                  * Recalculate kernelcore_node if the division per node
5182                  * now exceeds what is necessary to satisfy the requested
5183                  * amount of memory for the kernel
5184                  */
5185                 if (required_kernelcore < kernelcore_node)
5186                         kernelcore_node = required_kernelcore / usable_nodes;
5187 
5188                 /*
5189                  * As the map is walked, we track how much memory is usable
5190                  * by the kernel using kernelcore_remaining. When it is
5191                  * 0, the rest of the node is usable by ZONE_MOVABLE
5192                  */
5193                 kernelcore_remaining = kernelcore_node;
5194 
5195                 /* Go through each range of PFNs within this node */
5196                 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5197                         unsigned long size_pages;
5198 
5199                         start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5200                         if (start_pfn >= end_pfn)
5201                                 continue;
5202 
5203                         /* Account for what is only usable for kernelcore */
5204                         if (start_pfn < usable_startpfn) {
5205                                 unsigned long kernel_pages;
5206                                 kernel_pages = min(end_pfn, usable_startpfn)
5207                                                                 - start_pfn;
5208 
5209                                 kernelcore_remaining -= min(kernel_pages,
5210                                                         kernelcore_remaining);
5211                                 required_kernelcore -= min(kernel_pages,
5212                                                         required_kernelcore);
5213 
5214                                 /* Continue if range is now fully accounted */
5215                                 if (end_pfn <= usable_startpfn) {
5216 
5217                                         /*
5218                                          * Push zone_movable_pfn to the end so
5219                                          * that if we have to rebalance
5220                                          * kernelcore across nodes, we will
5221                                          * not double account here
5222                                          */
5223                                         zone_movable_pfn[nid] = end_pfn;
5224                                         continue;
5225                                 }
5226                                 start_pfn = usable_startpfn;
5227                         }
5228 
5229                         /*
5230                          * The usable PFN range for ZONE_MOVABLE is from
5231                          * start_pfn->end_pfn. Calculate size_pages as the
5232                          * number of pages used as kernelcore
5233                          */
5234                         size_pages = end_pfn - start_pfn;
5235                         if (size_pages > kernelcore_remaining)
5236                                 size_pages = kernelcore_remaining;
5237                         zone_movable_pfn[nid] = start_pfn + size_pages;
5238 
5239                         /*
5240                          * Some kernelcore has been met, update counts and
5241                          * break if the kernelcore for this node has been
5242                          * satisfied
5243                          */
5244                         required_kernelcore -= min(required_kernelcore,
5245                                                                 size_pages);
5246                         kernelcore_remaining -= size_pages;
5247                         if (!kernelcore_remaining)
5248                                 break;
5249                 }
5250         }
5251 
5252         /*
5253          * If there is still required_kernelcore, we do another pass with one
5254          * less node in the count. This will push zone_movable_pfn[nid] further
5255          * along on the nodes that still have memory until kernelcore is
5256          * satisfied
5257          */
5258         usable_nodes--;
5259         if (usable_nodes && required_kernelcore > usable_nodes)
5260                 goto restart;
5261 
5262 out2:
5263         /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5264         for (nid = 0; nid < MAX_NUMNODES; nid++)
5265                 zone_movable_pfn[nid] =
5266                         roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5267 
5268 out:
5269         /* restore the node_state */
5270         node_states[N_MEMORY] = saved_node_state;
5271 }
5272 
5273 /* Any regular or high memory on that node ? */
5274 static void check_for_memory(pg_data_t *pgdat, int nid)
5275 {
5276         enum zone_type zone_type;
5277 
5278         if (N_MEMORY == N_NORMAL_MEMORY)
5279                 return;
5280 
5281         for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5282                 struct zone *zone = &pgdat->node_zones[zone_type];
5283                 if (populated_zone(zone)) {
5284                         node_set_state(nid, N_HIGH_MEMORY);
5285                         if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5286                             zone_type <= ZONE_NORMAL)
5287                                 node_set_state(nid, N_NORMAL_MEMORY);
5288                         break;
5289                 }
5290         }
5291 }
5292 
5293 /**
5294  * free_area_init_nodes - Initialise all pg_data_t and zone data
5295  * @max_zone_pfn: an array of max PFNs for each zone
5296  *
5297  * This will call free_area_init_node() for each active node in the system.
5298  * Using the page ranges provided by memblock_set_node(), the size of each
5299  * zone in each node and their holes is calculated. If the maximum PFN
5300  * between two adjacent zones match, it is assumed that the zone is empty.
5301  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5302  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5303  * starts where the previous one ended. For example, ZONE_DMA32 starts
5304  * at arch_max_dma_pfn.
5305  */
5306 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5307 {
5308         unsigned long start_pfn, end_pfn;
5309         int i, nid;
5310 
5311         /* Record where the zone boundaries are */
5312         memset(arch_zone_lowest_possible_pfn, 0,
5313                                 sizeof(arch_zone_lowest_possible_pfn));
5314         memset(arch_zone_highest_possible_pfn, 0,
5315                                 sizeof(arch_zone_highest_possible_pfn));
5316         arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5317         arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5318         for (i = 1; i < MAX_NR_ZONES; i++) {
5319                 if (i == ZONE_MOVABLE)
5320                         continue;
5321                 arch_zone_lowest_possible_pfn[i] =
5322                         arch_zone_highest_possible_pfn[i-1];
5323                 arch_zone_highest_possible_pfn[i] =
5324                         max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5325         }
5326         arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5327         arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5328 
5329         /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5330         memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5331         find_zone_movable_pfns_for_nodes();
5332 
5333         /* Print out the zone ranges */
5334         printk("Zone ranges:\n");
5335         for (i = 0; i < MAX_NR_ZONES; i++) {
5336                 if (i == ZONE_MOVABLE)
5337                         continue;
5338                 printk(KERN_CONT "  %-8s ", zone_names[i]);
5339                 if (arch_zone_lowest_possible_pfn[i] ==
5340                                 arch_zone_highest_possible_pfn[i])
5341                         printk(KERN_CONT "empty\n");
5342                 else
5343                         printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5344                                 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5345                                 (arch_zone_highest_possible_pfn[i]
5346                                         << PAGE_SHIFT) - 1);
5347         }
5348 
5349         /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5350         printk("Movable zone start for each node\n");
5351         for (i = 0; i < MAX_NUMNODES; i++) {
5352                 if (zone_movable_pfn[i])
5353                         printk("  Node %d: %#010lx\n", i,
5354                                zone_movable_pfn[i] << PAGE_SHIFT);
5355         }
5356 
5357         /* Print out the early node map */
5358         printk("Early memory node ranges\n");
5359         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5360                 printk("  node %3d: [mem %#010lx-%#010lx]\n", nid,
5361                        start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
5362 
5363         /* Initialise every node */
5364         mminit_verify_pageflags_layout();
5365         setup_nr_node_ids();
5366         for_each_online_node(nid) {
5367                 pg_data_t *pgdat = NODE_DATA(nid);
5368                 free_area_init_node(nid, NULL,
5369                                 find_min_pfn_for_node(nid), NULL);
5370 
5371                 /* Any memory on that node */
5372                 if (pgdat->node_present_pages)
5373                         node_set_state(nid, N_MEMORY);
5374                 check_for_memory(pgdat, nid);
5375         }
5376 }
5377 
5378 static int __init cmdline_parse_core(char *p, unsigned long *core)
5379 {
5380         unsigned long long coremem;
5381         if (!p)
5382                 return -EINVAL;
5383 
5384         coremem = memparse(p, &p);
5385         *core = coremem >> PAGE_SHIFT;
5386 
5387         /* Paranoid check that UL is enough for the coremem value */
5388         WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5389 
5390         return 0;
5391 }
5392 
5393 /*
5394  * kernelcore=size sets the amount of memory for use for allocations that
5395  * cannot be reclaimed or migrated.
5396  */
5397 static int __init cmdline_parse_kernelcore(char *p)
5398 {
5399         return cmdline_parse_core(p, &required_kernelcore);
5400 }
5401 
5402 /*
5403  * movablecore=size sets the amount of memory for use for allocations that
5404  * can be reclaimed or migrated.
5405  */
5406 static int __init cmdline_parse_movablecore(char *p)
5407 {
5408         return cmdline_parse_core(p, &required_movablecore);
5409 }
5410 
5411 early_param("kernelcore", cmdline_parse_kernelcore);
5412 early_param("movablecore", cmdline_parse_movablecore);
5413 
5414 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5415 
5416 void adjust_managed_page_count(struct page *page, long count)
5417 {
5418         spin_lock(&managed_page_count_lock);
5419         page_zone(page)->managed_pages += count;
5420         totalram_pages += count;
5421 #ifdef CONFIG_HIGHMEM
5422         if (PageHighMem(page))
5423                 totalhigh_pages += count;
5424 #endif
5425         spin_unlock(&managed_page_count_lock);
5426 }
5427 EXPORT_SYMBOL(adjust_managed_page_count);
5428 
5429 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
5430 {
5431         void *pos;
5432         unsigned long pages = 0;
5433 
5434         start = (void *)PAGE_ALIGN((unsigned long)start);
5435         end = (void *)((unsigned long)end & PAGE_MASK);
5436         for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
5437                 if ((unsigned int)poison <= 0xFF)
5438                         memset(pos, poison, PAGE_SIZE);
5439                 free_reserved_page(virt_to_page(pos));
5440         }
5441 
5442         if (pages && s)
5443                 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5444                         s, pages << (PAGE_SHIFT - 10), start, end);
5445 
5446         return pages;
5447 }
5448 EXPORT_SYMBOL(free_reserved_area);
5449 
5450 #ifdef  CONFIG_HIGHMEM
5451 void free_highmem_page(struct page *page)
5452 {
5453         __free_reserved_page(page);
5454         totalram_pages++;
5455         page_zone(page)->managed_pages++;
5456         totalhigh_pages++;
5457 }
5458 #endif
5459 
5460 
5461 void __init mem_init_print_info(const char *str)
5462 {
5463         unsigned long physpages, codesize, datasize, rosize, bss_size;
5464         unsigned long init_code_size, init_data_size;
5465 
5466         physpages = get_num_physpages();
5467         codesize = _etext - _stext;
5468         datasize = _edata - _sdata;
5469         rosize = __end_rodata - __start_rodata;
5470         bss_size = __bss_stop - __bss_start;
5471         init_data_size = __init_end - __init_begin;
5472         init_code_size = _einittext - _sinittext;
5473 
5474         /*
5475          * Detect special cases and adjust section sizes accordingly:
5476          * 1) .init.* may be embedded into .data sections
5477          * 2) .init.text.* may be out of [__init_begin, __init_end],
5478          *    please refer to arch/tile/kernel/vmlinux.lds.S.
5479          * 3) .rodata.* may be embedded into .text or .data sections.
5480          */
5481 #define adj_init_size(start, end, size, pos, adj) \
5482         do { \
5483                 if (start <= pos && pos < end && size > adj) \
5484                         size -= adj; \
5485         } while (0)
5486 
5487         adj_init_size(__init_begin, __init_end, init_data_size,
5488                      _sinittext, init_code_size);
5489         adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5490         adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5491         adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5492         adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5493 
5494 #undef  adj_init_size
5495 
5496         printk("Memory: %luK/%luK available "
5497                "(%luK kernel code, %luK rwdata, %luK rodata, "
5498                "%luK init, %luK bss, %luK reserved"
5499 #ifdef  CONFIG_HIGHMEM
5500                ", %luK highmem"
5501 #endif
5502                "%s%s)\n",
5503                nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5504                codesize >> 10, datasize >> 10, rosize >> 10,
5505                (init_data_size + init_code_size) >> 10, bss_size >> 10,
5506                (physpages - totalram_pages) << (PAGE_SHIFT-10),
5507 #ifdef  CONFIG_HIGHMEM
5508                totalhigh_pages << (PAGE_SHIFT-10),
5509 #endif
5510                str ? ", " : "", str ? str : "");
5511 }
5512 
5513 /**
5514  * set_dma_reserve - set the specified number of pages reserved in the first zone
5515  * @new_dma_reserve: The number of pages to mark reserved
5516  *
5517  * The per-cpu batchsize and zone watermarks are determined by present_pages.
5518  * In the DMA zone, a significant percentage may be consumed by kernel image
5519  * and other unfreeable allocations which can skew the watermarks badly. This
5520  * function may optionally be used to account for unfreeable pages in the
5521  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5522  * smaller per-cpu batchsize.
5523  */
5524 void __init set_dma_reserve(unsigned long new_dma_reserve)
5525 {
5526         dma_reserve = new_dma_reserve;
5527 }
5528 
5529 void __init free_area_init(unsigned long *zones_size)
5530 {
5531         free_area_init_node(0, zones_size,
5532                         __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5533 }
5534 
5535 static int page_alloc_cpu_notify(struct notifier_block *self,
5536                                  unsigned long action, void *hcpu)
5537 {
5538         int cpu = (unsigned long)hcpu;
5539 
5540         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
5541                 lru_add_drain_cpu(cpu);
5542                 drain_pages(cpu);
5543 
5544                 /*
5545                  * Spill the event counters of the dead processor
5546                  * into the current processors event counters.
5547                  * This artificially elevates the count of the current
5548                  * processor.
5549                  */
5550                 vm_events_fold_cpu(cpu);
5551 
5552                 /*
5553                  * Zero the differential counters of the dead processor
5554                  * so that the vm statistics are consistent.
5555                  *
5556                  * This is only okay since the processor is dead and cannot
5557                  * race with what we are doing.
5558                  */
5559                 cpu_vm_stats_fold(cpu);
5560         }
5561         return NOTIFY_OK;
5562 }
5563 
5564 void __init page_alloc_init(void)
5565 {
5566         hotcpu_notifier(page_alloc_cpu_notify, 0);
5567 }
5568 
5569 /*
5570  * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5571  *      or min_free_kbytes changes.
5572  */
5573 static void calculate_totalreserve_pages(void)
5574 {
5575         struct pglist_data *pgdat;
5576         unsigned long reserve_pages = 0;
5577         enum zone_type i, j;
5578 
5579         for_each_online_pgdat(pgdat) {
5580                 for (i = 0; i < MAX_NR_ZONES; i++) {
5581                         struct zone *zone = pgdat->node_zones + i;
5582                         unsigned long max = 0;
5583 
5584                         /* Find valid and maximum lowmem_reserve in the zone */
5585                         for (j = i; j < MAX_NR_ZONES; j++) {
5586                                 if (zone->lowmem_reserve[j] > max)
5587                                         max = zone->lowmem_reserve[j];
5588                         }
5589 
5590                         /* we treat the high watermark as reserved pages. */
5591                         max += high_wmark_pages(zone);
5592 
5593                         if (max > zone->managed_pages)
5594                                 max = zone->managed_pages;
5595                         reserve_pages += max;
5596                         /*
5597                          * Lowmem reserves are not available to
5598                          * GFP_HIGHUSER page cache allocations and
5599                          * kswapd tries to balance zones to their high
5600                          * watermark.  As a result, neither should be
5601                          * regarded as dirtyable memory, to prevent a
5602                          * situation where reclaim has to clean pages
5603                          * in order to balance the zones.
5604                          */
5605                         zone->dirty_balance_reserve = max;
5606                 }
5607         }
5608         dirty_balance_reserve = reserve_pages;
5609         totalreserve_pages = reserve_pages;
5610 }
5611 
5612 /*
5613  * setup_per_zone_lowmem_reserve - called whenever
5614  *      sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
5615  *      has a correct pages reserved value, so an adequate number of
5616  *      pages are left in the zone after a successful __alloc_pages().
5617  */
5618 static void setup_per_zone_lowmem_reserve(void)
5619 {
5620         struct pglist_data *pgdat;
5621         enum zone_type j, idx;
5622 
5623         for_each_online_pgdat(pgdat) {
5624                 for (j = 0; j < MAX_NR_ZONES; j++) {
5625                         struct zone *zone = pgdat->node_zones + j;
5626                         unsigned long managed_pages = zone->managed_pages;
5627 
5628                         zone->lowmem_reserve[j] = 0;
5629 
5630                         idx = j;
5631                         while (idx) {
5632                                 struct zone *lower_zone;
5633 
5634                                 idx--;
5635 
5636                                 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5637                                         sysctl_lowmem_reserve_ratio[idx] = 1;
5638 
5639                                 lower_zone = pgdat->node_zones + idx;
5640                                 lower_zone->lowmem_reserve[j] = managed_pages /
5641                                         sysctl_lowmem_reserve_ratio[idx];
5642                                 managed_pages += lower_zone->managed_pages;
5643                         }
5644                 }
5645         }
5646 
5647         /* update totalreserve_pages */
5648         calculate_totalreserve_pages();
5649 }
5650 
5651 static void __setup_per_zone_wmarks(void)
5652 {
5653         unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5654         unsigned long lowmem_pages = 0;
5655         struct zone *zone;
5656         unsigned long flags;
5657 
5658         /* Calculate total number of !ZONE_HIGHMEM pages */
5659         for_each_zone(zone) {
5660                 if (!is_highmem(zone))
5661                         lowmem_pages += zone->managed_pages;
5662         }
5663 
5664         for_each_zone(zone) {
5665                 u64 tmp;
5666 
5667                 spin_lock_irqsave(&zone->lock, flags);
5668                 tmp = (u64)pages_min * zone->managed_pages;
5669                 do_div(tmp, lowmem_pages);
5670                 if (is_highmem(zone)) {
5671                         /*
5672                          * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5673                          * need highmem pages, so cap pages_min to a small
5674                          * value here.
5675                          *
5676                          * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5677                          * deltas controls asynch page reclaim, and so should
5678                          * not be capped for highmem.
5679                          */
5680                         unsigned long min_pages;
5681 
5682                         min_pages = zone->managed_pages / 1024;
5683                         min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
5684                         zone->watermark[WMARK_MIN] = min_pages;
5685                 } else {
5686                         /*
5687                          * If it's a lowmem zone, reserve a number of pages
5688                          * proportionate to the zone's size.
5689                          */
5690                         zone->watermark[WMARK_MIN] = tmp;
5691                 }
5692 
5693                 zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + (tmp >> 2);
5694                 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
5695 
5696                 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
5697                                       high_wmark_pages(zone) -
5698                                       low_wmark_pages(zone) -
5699                                       zone_page_state(zone, NR_ALLOC_BATCH));
5700 
5701                 setup_zone_migrate_reserve(zone);
5702                 spin_unlock_irqrestore(&zone->lock, flags);
5703         }
5704 
5705         /* update totalreserve_pages */
5706         calculate_totalreserve_pages();
5707 }
5708 
5709 /**
5710  * setup_per_zone_wmarks - called when min_free_kbytes changes
5711  * or when memory is hot-{added|removed}
5712  *
5713  * Ensures that the watermark[min,low,high] values for each zone are set
5714  * correctly with respect to min_free_kbytes.
5715  */
5716 void setup_per_zone_wmarks(void)
5717 {
5718         mutex_lock(&zonelists_mutex);
5719         __setup_per_zone_wmarks();
5720         mutex_unlock(&zonelists_mutex);
5721 }
5722 
5723 /*
5724  * The inactive anon list should be small enough that the VM never has to
5725  * do too much work, but large enough that each inactive page has a chance
5726  * to be referenced again before it is swapped out.
5727  *
5728  * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5729  * INACTIVE_ANON pages on this zone's LRU, maintained by the
5730  * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5731  * the anonymous pages are kept on the inactive list.
5732  *
5733  * total     target    max
5734  * memory    ratio     inactive anon
5735  * -------------------------------------
5736  *   10MB       1         5MB
5737  *  100MB       1        50MB
5738  *    1GB       3       250MB
5739  *   10GB      10       0.9GB
5740  *  100GB      31         3GB
5741  *    1TB     101        10GB
5742  *   10TB     320        32GB
5743  */
5744 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
5745 {
5746         unsigned int gb, ratio;
5747 
5748         /* Zone size in gigabytes */
5749         gb = zone->managed_pages >> (30 - PAGE_SHIFT);
5750         if (gb)
5751                 ratio = int_sqrt(10 * gb);
5752         else
5753                 ratio = 1;
5754 
5755         zone->inactive_ratio = ratio;
5756 }
5757 
5758 static void __meminit setup_per_zone_inactive_ratio(void)
5759 {
5760         struct zone *zone;
5761 
5762         for_each_zone(zone)
5763                 calculate_zone_inactive_ratio(zone);
5764 }
5765 
5766 /*
5767  * Initialise min_free_kbytes.
5768  *
5769  * For small machines we want it small (128k min).  For large machines
5770  * we want it large (64MB max).  But it is not linear, because network
5771  * bandwidth does not increase linearly with machine size.  We use
5772  *
5773  *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5774  *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
5775  *
5776  * which yields
5777  *
5778  * 16MB:        512k
5779  * 32MB:        724k
5780  * 64MB:        1024k
5781  * 128MB:       1448k
5782  * 256MB:       2048k
5783  * 512MB:       2896k
5784  * 1024MB:      4096k
5785  * 2048MB:      5792k
5786  * 4096MB:      8192k
5787  * 8192MB:      11584k
5788  * 16384MB:     16384k
5789  */
5790 int __meminit init_per_zone_wmark_min(void)
5791 {
5792         unsigned long lowmem_kbytes;
5793         int new_min_free_kbytes;
5794 
5795         lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5796         new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5797 
5798         if (new_min_free_kbytes > user_min_free_kbytes) {
5799                 min_free_kbytes = new_min_free_kbytes;
5800                 if (min_free_kbytes < 128)
5801                         min_free_kbytes = 128;
5802                 if (min_free_kbytes > 65536)
5803                         min_free_kbytes = 65536;
5804         } else {
5805                 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5806                                 new_min_free_kbytes, user_min_free_kbytes);
5807         }
5808         setup_per_zone_wmarks();
5809         refresh_zone_stat_thresholds();
5810         setup_per_zone_lowmem_reserve();
5811         setup_per_zone_inactive_ratio();
5812         return 0;
5813 }
5814 module_init(init_per_zone_wmark_min)
5815 
5816 /*
5817  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5818  *      that we can call two helper functions whenever min_free_kbytes
5819  *      changes.
5820  */
5821 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
5822         void __user *buffer, size_t *length, loff_t *ppos)
5823 {
5824         int rc;
5825 
5826         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5827         if (rc)
5828                 return rc;
5829 
5830         if (write) {
5831                 user_min_free_kbytes = min_free_kbytes;
5832                 setup_per_zone_wmarks();
5833         }
5834         return 0;
5835 }
5836 
5837 #ifdef CONFIG_NUMA
5838 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
5839         void __user *buffer, size_t *length, loff_t *ppos)
5840 {
5841         struct zone *zone;
5842         int rc;
5843 
5844         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5845         if (rc)
5846                 return rc;
5847 
5848         for_each_zone(zone)
5849                 zone->min_unmapped_pages = (zone->managed_pages *
5850                                 sysctl_min_unmapped_ratio) / 100;
5851         return 0;
5852 }
5853 
5854 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
5855         void __user *buffer, size_t *length, loff_t *ppos)
5856 {
5857         struct zone *zone;
5858         int rc;
5859 
5860         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5861         if (rc)
5862                 return rc;
5863 
5864         for_each_zone(zone)
5865                 zone->min_slab_pages = (zone->managed_pages *
5866                                 sysctl_min_slab_ratio) / 100;
5867         return 0;
5868 }
5869 #endif
5870 
5871 /*
5872  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5873  *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5874  *      whenever sysctl_lowmem_reserve_ratio changes.
5875  *
5876  * The reserve ratio obviously has absolutely no relation with the
5877  * minimum watermarks. The lowmem reserve ratio can only make sense
5878  * if in function of the boot time zone sizes.
5879  */
5880 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
5881         void __user *buffer, size_t *length, loff_t *ppos)
5882 {
5883         proc_dointvec_minmax(table, write, buffer, length, ppos);
5884         setup_per_zone_lowmem_reserve();
5885         return 0;
5886 }
5887 
5888 /*
5889  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5890  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
5891  * pagelist can have before it gets flushed back to buddy allocator.
5892  */
5893 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
5894         void __user *buffer, size_t *length, loff_t *ppos)
5895 {
5896         struct zone *zone;
5897         int old_percpu_pagelist_fraction;
5898         int ret;
5899 
5900         mutex_lock(&pcp_batch_high_lock);
5901         old_percpu_pagelist_fraction = percpu_pagelist_fraction;
5902 
5903         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5904         if (!write || ret < 0)
5905                 goto out;
5906 
5907         /* Sanity checking to avoid pcp imbalance */
5908         if (percpu_pagelist_fraction &&
5909             percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
5910                 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
5911                 ret = -EINVAL;
5912                 goto out;
5913         }
5914 
5915         /* No change? */
5916         if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
5917                 goto out;
5918 
5919         for_each_populated_zone(zone) {
5920                 unsigned int cpu;
5921 
5922                 for_each_possible_cpu(cpu)
5923                         pageset_set_high_and_batch(zone,
5924                                         per_cpu_ptr(zone->pageset, cpu));
5925         }
5926 out:
5927         mutex_unlock(&pcp_batch_high_lock);
5928         return ret;
5929 }
5930 
5931 int hashdist = HASHDIST_DEFAULT;
5932 
5933 #ifdef CONFIG_NUMA
5934 static int __init set_hashdist(char *str)
5935 {
5936         if (!str)
5937                 return 0;
5938         hashdist = simple_strtoul(str, &str, 0);
5939         return 1;
5940 }
5941 __setup("hashdist=", set_hashdist);
5942 #endif
5943 
5944 /*
5945  * allocate a large system hash table from bootmem
5946  * - it is assumed that the hash table must contain an exact power-of-2
5947  *   quantity of entries
5948  * - limit is the number of hash buckets, not the total allocation size
5949  */
5950 void *__init alloc_large_system_hash(const char *tablename,
5951                                      unsigned long bucketsize,
5952                                      unsigned long numentries,
5953                                      int scale,
5954                                      int flags,
5955                                      unsigned int *_hash_shift,
5956                                      unsigned int *_hash_mask,
5957                                      unsigned long low_limit,
5958                                      unsigned long high_limit)
5959 {
5960         unsigned long long max = high_limit;
5961         unsigned long log2qty, size;
5962         void *table = NULL;
5963 
5964         /* allow the kernel cmdline to have a say */
5965         if (!numentries) {
5966                 /* round applicable memory size up to nearest megabyte */
5967                 numentries = nr_kernel_pages;
5968 
5969                 /* It isn't necessary when PAGE_SIZE >= 1MB */
5970                 if (PAGE_SHIFT < 20)
5971                         numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
5972 
5973                 /* limit to 1 bucket per 2^scale bytes of low memory */
5974                 if (scale > PAGE_SHIFT)
5975                         numentries >>= (scale - PAGE_SHIFT);
5976                 else
5977                         numentries <<= (PAGE_SHIFT - scale);
5978 
5979                 /* Make sure we've got at least a 0-order allocation.. */
5980                 if (unlikely(flags & HASH_SMALL)) {
5981                         /* Makes no sense without HASH_EARLY */
5982                         WARN_ON(!(flags & HASH_EARLY));
5983                         if (!(numentries >> *_hash_shift)) {
5984                                 numentries = 1UL << *_hash_shift;
5985                                 BUG_ON(!numentries);
5986                         }
5987                 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5988                         numentries = PAGE_SIZE / bucketsize;
5989         }
5990         numentries = roundup_pow_of_two(numentries);
5991 
5992         /* limit allocation size to 1/16 total memory by default */
5993         if (max == 0) {
5994                 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5995                 do_div(max, bucketsize);
5996         }
5997         max = min(max, 0x80000000ULL);
5998 
5999         if (numentries < low_limit)
6000                 numentries = low_limit;
6001         if (numentries > max)
6002                 numentries = max;
6003 
6004         log2qty = ilog2(numentries);
6005 
6006         do {
6007                 size = bucketsize << log2qty;
6008                 if (flags & HASH_EARLY)
6009                         table = memblock_virt_alloc_nopanic(size, 0);
6010                 else if (hashdist)
6011                         table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6012                 else {
6013                         /*
6014                          * If bucketsize is not a power-of-two, we may free
6015                          * some pages at the end of hash table which
6016                          * alloc_pages_exact() automatically does
6017                          */
6018                         if (get_order(size) < MAX_ORDER) {
6019                                 table = alloc_pages_exact(size, GFP_ATOMIC);
6020                                 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6021                         }
6022                 }
6023         } while (!table && size > PAGE_SIZE && --log2qty);
6024 
6025         if (!table)
6026                 panic("Failed to allocate %s hash table\n", tablename);
6027 
6028         printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
6029                tablename,
6030                (1UL << log2qty),
6031                ilog2(size) - PAGE_SHIFT,
6032                size);
6033 
6034         if (_hash_shift)
6035                 *_hash_shift = log2qty;
6036         if (_hash_mask)
6037                 *_hash_mask = (1 << log2qty) - 1;
6038 
6039         return table;
6040 }
6041 
6042 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6043 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6044                                                         unsigned long pfn)
6045 {
6046 #ifdef CONFIG_SPARSEMEM
6047         return __pfn_to_section(pfn)->pageblock_flags;
6048 #else
6049         return zone->pageblock_flags;
6050 #endif /* CONFIG_SPARSEMEM */
6051 }
6052 
6053 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6054 {
6055 #ifdef CONFIG_SPARSEMEM
6056         pfn &= (PAGES_PER_SECTION-1);
6057         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6058 #else
6059         pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
6060         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6061 #endif /* CONFIG_SPARSEMEM */
6062 }
6063 
6064 /**
6065  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6066  * @page: The page within the block of interest
6067  * @pfn: The target page frame number
6068  * @end_bitidx: The last bit of interest to retrieve
6069  * @mask: mask of bits that the caller is interested in
6070  *
6071  * Return: pageblock_bits flags
6072  */
6073 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
6074                                         unsigned long end_bitidx,
6075                                         unsigned long mask)
6076 {
6077         struct zone *zone;
6078         unsigned long *bitmap;
6079         unsigned long bitidx, word_bitidx;
6080         unsigned long word;
6081 
6082         zone = page_zone(page);
6083         bitmap = get_pageblock_bitmap(zone, pfn);
6084         bitidx = pfn_to_bitidx(zone, pfn);
6085         word_bitidx = bitidx / BITS_PER_LONG;
6086         bitidx &= (BITS_PER_LONG-1);
6087 
6088         word = bitmap[word_bitidx];
6089         bitidx += end_bitidx;
6090         return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
6091 }
6092 
6093 /**
6094  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6095  * @page: The page within the block of interest
6096  * @flags: The flags to set
6097  * @pfn: The target page frame number
6098  * @end_bitidx: The last bit of interest
6099  * @mask: mask of bits that the caller is interested in
6100  */
6101 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6102                                         unsigned long pfn,
6103                                         unsigned long end_bitidx,
6104                                         unsigned long mask)
6105 {
6106         struct zone *zone;
6107         unsigned long *bitmap;
6108         unsigned long bitidx, word_bitidx;
6109         unsigned long old_word, word;
6110 
6111         BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
6112 
6113         zone = page_zone(page);
6114         bitmap = get_pageblock_bitmap(zone, pfn);
6115         bitidx = pfn_to_bitidx(zone, pfn);
6116         word_bitidx = bitidx / BITS_PER_LONG;
6117         bitidx &= (BITS_PER_LONG-1);
6118 
6119         VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
6120 
6121         bitidx += end_bitidx;
6122         mask <<= (BITS_PER_LONG - bitidx - 1);
6123         flags <<= (BITS_PER_LONG - bitidx - 1);
6124 
6125         word = ACCESS_ONCE(bitmap[word_bitidx]);
6126         for (;;) {
6127                 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6128                 if (word == old_word)
6129                         break;
6130                 word = old_word;
6131         }
6132 }
6133 
6134 /*
6135  * This function checks whether pageblock includes unmovable pages or not.
6136  * If @count is not zero, it is okay to include less @count unmovable pages
6137  *
6138  * PageLRU check without isolation or lru_lock could race so that
6139  * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6140  * expect this function should be exact.
6141  */
6142 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6143                          bool skip_hwpoisoned_pages)
6144 {
6145         unsigned long pfn, iter, found;
6146         int mt;
6147 
6148         /*
6149          * For avoiding noise data, lru_add_drain_all() should be called
6150          * If ZONE_MOVABLE, the zone never contains unmovable pages
6151          */
6152         if (zone_idx(zone) == ZONE_MOVABLE)
6153                 return false;
6154         mt = get_pageblock_migratetype(page);
6155         if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6156                 return false;
6157 
6158         pfn = page_to_pfn(page);
6159         for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6160                 unsigned long check = pfn + iter;
6161 
6162                 if (!pfn_valid_within(check))
6163                         continue;
6164 
6165                 page = pfn_to_page(check);
6166 
6167                 /*
6168                  * Hugepages are not in LRU lists, but they're movable.
6169                  * We need not scan over tail pages bacause we don't
6170                  * handle each tail page individually in migration.
6171                  */
6172                 if (PageHuge(page)) {
6173                         iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6174                         continue;
6175                 }
6176 
6177                 /*
6178                  * We can't use page_count without pin a page
6179                  * because another CPU can free compound page.
6180                  * This check already skips compound tails of THP
6181                  * because their page->_count is zero at all time.
6182                  */
6183                 if (!atomic_read(&page->_count)) {
6184                         if (PageBuddy(page))
6185                                 iter += (1 << page_order(page)) - 1;
6186                         continue;
6187                 }
6188 
6189                 /*
6190                  * The HWPoisoned page may be not in buddy system, and
6191                  * page_count() is not 0.
6192                  */
6193                 if (skip_hwpoisoned_pages && PageHWPoison(page))
6194                         continue;
6195 
6196                 if (!PageLRU(page))
6197                         found++;
6198                 /*
6199                  * If there are RECLAIMABLE pages, we need to check it.
6200                  * But now, memory offline itself doesn't call shrink_slab()
6201                  * and it still to be fixed.
6202                  */
6203                 /*
6204                  * If the page is not RAM, page_count()should be 0.
6205                  * we don't need more check. This is an _used_ not-movable page.
6206                  *
6207                  * The problematic thing here is PG_reserved pages. PG_reserved
6208                  * is set to both of a memory hole page and a _used_ kernel
6209                  * page at boot.
6210                  */
6211                 if (found > count)
6212                         return true;
6213         }
6214         return false;
6215 }
6216 
6217 bool is_pageblock_removable_nolock(struct page *page)
6218 {
6219         struct zone *zone;
6220         unsigned long pfn;
6221 
6222         /*
6223          * We have to be careful here because we are iterating over memory
6224          * sections which are not zone aware so we might end up outside of
6225          * the zone but still within the section.
6226          * We have to take care about the node as well. If the node is offline
6227          * its NODE_DATA will be NULL - see page_zone.
6228          */
6229         if (!node_online(page_to_nid(page)))
6230                 return false;
6231 
6232         zone = page_zone(page);
6233         pfn = page_to_pfn(page);
6234         if (!zone_spans_pfn(zone, pfn))
6235                 return false;
6236 
6237         return !has_unmovable_pages(zone, page, 0, true);
6238 }
6239 
6240 #ifdef CONFIG_CMA
6241 
6242 static unsigned long pfn_max_align_down(unsigned long pfn)
6243 {
6244         return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6245                              pageblock_nr_pages) - 1);
6246 }
6247 
6248 static unsigned long pfn_max_align_up(unsigned long pfn)
6249 {
6250         return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6251                                 pageblock_nr_pages));
6252 }
6253 
6254 /* [start, end) must belong to a single zone. */
6255 static int __alloc_contig_migrate_range(struct compact_control *cc,
6256                                         unsigned long start, unsigned long end)
6257 {
6258         /* This function is based on compact_zone() from compaction.c. */
6259         unsigned long nr_reclaimed;
6260         unsigned long pfn = start;
6261         unsigned int tries = 0;
6262         int ret = 0;
6263 
6264         migrate_prep();
6265 
6266         while (pfn < end || !list_empty(&cc->migratepages)) {
6267                 if (fatal_signal_pending(current)) {
6268                         ret = -EINTR;
6269                         break;
6270                 }
6271 
6272                 if (list_empty(&cc->migratepages)) {
6273                         cc->nr_migratepages = 0;
6274                         pfn = isolate_migratepages_range(cc->zone, cc,
6275                                                          pfn, end, true);
6276                         if (!pfn) {
6277                                 ret = -EINTR;
6278                                 break;
6279                         }
6280                         tries = 0;
6281                 } else if (++tries == 5) {
6282                         ret = ret < 0 ? ret : -EBUSY;
6283                         break;
6284                 }
6285 
6286                 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6287                                                         &cc->migratepages);
6288                 cc->nr_migratepages -= nr_reclaimed;
6289 
6290                 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6291                                     NULL, 0, cc->mode, MR_CMA);
6292         }
6293         if (ret < 0) {
6294                 putback_movable_pages(&cc->migratepages);
6295                 return ret;
6296         }
6297         return 0;
6298 }
6299 
6300 /**
6301  * alloc_contig_range() -- tries to allocate given range of pages
6302  * @start:      start PFN to allocate
6303  * @end:        one-past-the-last PFN to allocate
6304  * @migratetype:        migratetype of the underlaying pageblocks (either
6305  *                      #MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
6306  *                      in range must have the same migratetype and it must
6307  *                      be either of the two.
6308  *
6309  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6310  * aligned, however it's the caller's responsibility to guarantee that
6311  * we are the only thread that changes migrate type of pageblocks the
6312  * pages fall in.
6313  *
6314  * The PFN range must belong to a single zone.
6315  *
6316  * Returns zero on success or negative error code.  On success all
6317  * pages which PFN is in [start, end) are allocated for the caller and
6318  * need to be freed with free_contig_range().
6319  */
6320 int alloc_contig_range(unsigned long start, unsigned long end,
6321                        unsigned migratetype)
6322 {
6323         unsigned long outer_start, outer_end;
6324         int ret = 0, order;
6325 
6326         struct compact_control cc = {
6327                 .nr_migratepages = 0,
6328                 .order = -1,
6329                 .zone = page_zone(pfn_to_page(start)),
6330                 .mode = MIGRATE_SYNC,
6331                 .ignore_skip_hint = true,
6332         };
6333         INIT_LIST_HEAD(&cc.migratepages);
6334 
6335         /*
6336          * What we do here is we mark all pageblocks in range as
6337          * MIGRATE_ISOLATE.  Because pageblock and max order pages may
6338          * have different sizes, and due to the way page allocator
6339          * work, we align the range to biggest of the two pages so
6340          * that page allocator won't try to merge buddies from
6341          * different pageblocks and change MIGRATE_ISOLATE to some
6342          * other migration type.
6343          *
6344          * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6345          * migrate the pages from an unaligned range (ie. pages that
6346          * we are interested in).  This will put all the pages in
6347          * range back to page allocator as MIGRATE_ISOLATE.
6348          *
6349          * When this is done, we take the pages in range from page
6350          * allocator removing them from the buddy system.  This way
6351          * page allocator will never consider using them.
6352          *
6353          * This lets us mark the pageblocks back as
6354          * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6355          * aligned range but not in the unaligned, original range are
6356          * put back to page allocator so that buddy can use them.
6357          */
6358 
6359         ret = start_isolate_page_range(pfn_max_align_down(start),
6360                                        pfn_max_align_up(end), migratetype,
6361                                        false);
6362         if (ret)
6363                 return ret;
6364 
6365         ret = __alloc_contig_migrate_range(&cc, start, end);
6366         if (ret)
6367                 goto done;
6368 
6369         /*
6370          * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6371          * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
6372          * more, all pages in [start, end) are free in page allocator.
6373          * What we are going to do is to allocate all pages from
6374          * [start, end) (that is remove them from page allocator).
6375          *
6376          * The only problem is that pages at the beginning and at the
6377          * end of interesting range may be not aligned with pages that
6378          * page allocator holds, ie. they can be part of higher order
6379          * pages.  Because of this, we reserve the bigger range and
6380          * once this is done free the pages we are not interested in.
6381          *
6382          * We don't have to hold zone->lock here because the pages are
6383          * isolated thus they won't get removed from buddy.
6384          */
6385 
6386         lru_add_drain_all();
6387         drain_all_pages();
6388 
6389         order = 0;
6390         outer_start = start;
6391         while (!PageBuddy(pfn_to_page(outer_start))) {
6392                 if (++order >= MAX_ORDER) {
6393                         ret = -EBUSY;
6394                         goto done;
6395                 }
6396                 outer_start &= ~0UL << order;
6397         }
6398 
6399         /* Make sure the range is really isolated. */
6400         if (test_pages_isolated(outer_start, end, false)) {
6401                 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6402                        outer_start, end);
6403                 ret = -EBUSY;
6404                 goto done;
6405         }
6406 
6407 
6408         /* Grab isolated pages from freelists. */
6409         outer_end = isolate_freepages_range(&cc, outer_start, end);
6410         if (!outer_end) {
6411                 ret = -EBUSY;
6412                 goto done;
6413         }
6414 
6415         /* Free head and tail (if any) */
6416         if (start != outer_start)
6417                 free_contig_range(outer_start, start - outer_start);
6418         if (end != outer_end)
6419                 free_contig_range(end, outer_end - end);
6420 
6421 done:
6422         undo_isolate_page_range(pfn_max_align_down(start),
6423                                 pfn_max_align_up(end), migratetype);
6424         return ret;
6425 }
6426 
6427 void free_contig_range(unsigned long pfn, unsigned nr_pages)
6428 {
6429         unsigned int count = 0;
6430 
6431         for (; nr_pages--; pfn++) {
6432                 struct page *page = pfn_to_page(pfn);
6433 
6434                 count += page_count(page) != 1;
6435                 __free_page(page);
6436         }
6437         WARN(count != 0, "%d pages are still in use!\n", count);
6438 }
6439 #endif
6440 
6441 #ifdef CONFIG_MEMORY_HOTPLUG
6442 /*
6443  * The zone indicated has a new number of managed_pages; batch sizes and percpu
6444  * page high values need to be recalulated.
6445  */
6446 void __meminit zone_pcp_update(struct zone *zone)
6447 {
6448         unsigned cpu;
6449         mutex_lock(&pcp_batch_high_lock);
6450         for_each_possible_cpu(cpu)
6451                 pageset_set_high_and_batch(zone,
6452                                 per_cpu_ptr(zone->pageset, cpu));
6453         mutex_unlock(&pcp_batch_high_lock);
6454 }
6455 #endif
6456 
6457 void zone_pcp_reset(struct zone *zone)
6458 {
6459         unsigned long flags;
6460         int cpu;
6461         struct per_cpu_pageset *pset;
6462 
6463         /* avoid races with drain_pages()  */
6464         local_irq_save(flags);
6465         if (zone->pageset != &boot_pageset) {
6466                 for_each_online_cpu(cpu) {
6467                         pset = per_cpu_ptr(zone->pageset, cpu);
6468                         drain_zonestat(zone, pset);
6469                 }
6470                 free_percpu(zone->pageset);
6471                 zone->pageset = &boot_pageset;
6472         }
6473         local_irq_restore(flags);
6474 }
6475 
6476 #ifdef CONFIG_MEMORY_HOTREMOVE
6477 /*
6478  * All pages in the range must be isolated before calling this.
6479  */
6480 void
6481 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6482 {
6483         struct page *page;
6484         struct zone *zone;
6485         unsigned int order, i;
6486         unsigned long pfn;
6487         unsigned long flags;
6488         /* find the first valid pfn */
6489         for (pfn = start_pfn; pfn < end_pfn; pfn++)
6490                 if (pfn_valid(pfn))
6491                         break;
6492         if (pfn == end_pfn)
6493                 return;
6494         zone = page_zone(pfn_to_page(pfn));
6495         spin_lock_irqsave(&zone->lock, flags);
6496         pfn = start_pfn;
6497         while (pfn < end_pfn) {
6498                 if (!pfn_valid(pfn)) {
6499                         pfn++;
6500                         continue;
6501                 }
6502                 page = pfn_to_page(pfn);
6503                 /*
6504                  * The HWPoisoned page may be not in buddy system, and
6505                  * page_count() is not 0.
6506                  */
6507                 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6508                         pfn++;
6509                         SetPageReserved(page);
6510                         continue;
6511                 }
6512 
6513                 BUG_ON(page_count(page));
6514                 BUG_ON(!PageBuddy(page));
6515                 order = page_order(page);
6516 #ifdef CONFIG_DEBUG_VM
6517                 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6518                        pfn, 1 << order, end_pfn);
6519 #endif
6520                 list_del(&page->lru);
6521                 rmv_page_order(page);
6522                 zone->free_area[order].nr_free--;
6523                 for (i = 0; i < (1 << order); i++)
6524                         SetPageReserved((page+i));
6525                 pfn += (1 << order);
6526         }
6527         spin_unlock_irqrestore(&zone->lock, flags);
6528 }
6529 #endif
6530 
6531 #ifdef CONFIG_MEMORY_FAILURE
6532 bool is_free_buddy_page(struct page *page)
6533 {
6534         struct zone *zone = page_zone(page);
6535         unsigned long pfn = page_to_pfn(page);
6536         unsigned long flags;
6537         unsigned int order;
6538 
6539         spin_lock_irqsave(&zone->lock, flags);
6540         for (order = 0; order < MAX_ORDER; order++) {
6541                 struct page *page_head = page - (pfn & ((1 << order) - 1));
6542 
6543                 if (PageBuddy(page_head) && page_order(page_head) >= order)
6544                         break;
6545         }
6546         spin_unlock_irqrestore(&zone->lock, flags);
6547 
6548         return order < MAX_ORDER;
6549 }
6550 #endif
6551 
6552 static const struct trace_print_flags pageflag_names[] = {
6553         {1UL << PG_locked,              "locked"        },
6554         {1UL << PG_error,               "error"         },
6555         {1UL << PG_referenced,          "referenced"    },
6556         {1UL << PG_uptodate,            "uptodate"      },
6557         {1UL << PG_dirty,               "dirty"         },
6558         {1UL << PG_lru,                 "lru"           },
6559         {1UL << PG_active,              "active"        },
6560         {1UL << PG_slab,                "slab"          },
6561         {1UL << PG_owner_priv_1,        "owner_priv_1"  },
6562         {1UL << PG_arch_1,              "arch_1"        },
6563         {1UL << PG_reserved,            "reserved"      },
6564         {1UL << PG_private,             "private"       },
6565         {1UL << PG_private_2,           "private_2"     },
6566         {1UL << PG_writeback,           "writeback"     },
6567 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6568         {1UL << PG_head,                "head"          },
6569         {1UL << PG_tail,                "tail"          },
6570 #else
6571         {1UL << PG_compound,            "compound"      },
6572 #endif
6573         {1UL << PG_swapcache,           "swapcache"     },
6574         {1UL << PG_mappedtodisk,        "mappedtodisk"  },
6575         {1UL << PG_reclaim,             "reclaim"       },
6576         {1UL << PG_swapbacked,          "swapbacked"    },
6577         {1UL << PG_unevictable,         "unevictable"   },
6578 #ifdef CONFIG_MMU
6579         {1UL << PG_mlocked,             "mlocked"       },
6580 #endif
6581 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6582         {1UL << PG_uncached,            "uncached"      },
6583 #endif
6584 #ifdef CONFIG_MEMORY_FAILURE
6585         {1UL << PG_hwpoison,            "hwpoison"      },
6586 #endif
6587 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6588         {1UL << PG_compound_lock,       "compound_lock" },
6589 #endif
6590 };
6591 
6592 static void dump_page_flags(unsigned long flags)
6593 {
6594         const char *delim = "";
6595         unsigned long mask;
6596         int i;
6597 
6598         BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
6599 
6600         printk(KERN_ALERT "page flags: %#lx(", flags);
6601 
6602         /* remove zone id */
6603         flags &= (1UL << NR_PAGEFLAGS) - 1;
6604 
6605         for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
6606 
6607                 mask = pageflag_names[i].mask;
6608                 if ((flags & mask) != mask)
6609                         continue;
6610 
6611                 flags &= ~mask;
6612                 printk("%s%s", delim, pageflag_names[i].name);
6613                 delim = "|";
6614         }
6615 
6616         /* check for left over flags */
6617         if (flags)
6618                 printk("%s%#lx", delim, flags);
6619 
6620         printk(")\n");
6621 }
6622 
6623 void dump_page_badflags(struct page *page, const char *reason,
6624                 unsigned long badflags)
6625 {
6626         printk(KERN_ALERT
6627                "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6628                 page, atomic_read(&page->_count), page_mapcount(page),
6629                 page->mapping, page->index);
6630         dump_page_flags(page->flags);
6631         if (reason)
6632                 pr_alert("page dumped because: %s\n", reason);
6633         if (page->flags & badflags) {
6634                 pr_alert("bad because of flags:\n");
6635                 dump_page_flags(page->flags & badflags);
6636         }
6637         mem_cgroup_print_bad_page(page);
6638 }
6639 
6640 void dump_page(struct page *page, const char *reason)
6641 {
6642         dump_page_badflags(page, reason, 0);
6643 }
6644 EXPORT_SYMBOL(dump_page);
6645 

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