Version:  2.0.40 2.2.26 2.4.37 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 3.17

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         unsigned long nr_scanned;
684 
685         spin_lock(&zone->lock);
686         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
687         if (nr_scanned)
688                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
689 
690         while (to_free) {
691                 struct page *page;
692                 struct list_head *list;
693 
694                 /*
695                  * Remove pages from lists in a round-robin fashion. A
696                  * batch_free count is maintained that is incremented when an
697                  * empty list is encountered.  This is so more pages are freed
698                  * off fuller lists instead of spinning excessively around empty
699                  * lists
700                  */
701                 do {
702                         batch_free++;
703                         if (++migratetype == MIGRATE_PCPTYPES)
704                                 migratetype = 0;
705                         list = &pcp->lists[migratetype];
706                 } while (list_empty(list));
707 
708                 /* This is the only non-empty list. Free them all. */
709                 if (batch_free == MIGRATE_PCPTYPES)
710                         batch_free = to_free;
711 
712                 do {
713                         int mt; /* migratetype of the to-be-freed page */
714 
715                         page = list_entry(list->prev, struct page, lru);
716                         /* must delete as __free_one_page list manipulates */
717                         list_del(&page->lru);
718                         mt = get_freepage_migratetype(page);
719                         /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
720                         __free_one_page(page, page_to_pfn(page), zone, 0, mt);
721                         trace_mm_page_pcpu_drain(page, 0, mt);
722                         if (likely(!is_migrate_isolate_page(page))) {
723                                 __mod_zone_page_state(zone, NR_FREE_PAGES, 1);
724                                 if (is_migrate_cma(mt))
725                                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
726                         }
727                 } while (--to_free && --batch_free && !list_empty(list));
728         }
729         spin_unlock(&zone->lock);
730 }
731 
732 static void free_one_page(struct zone *zone,
733                                 struct page *page, unsigned long pfn,
734                                 unsigned int order,
735                                 int migratetype)
736 {
737         unsigned long nr_scanned;
738         spin_lock(&zone->lock);
739         nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
740         if (nr_scanned)
741                 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
742 
743         __free_one_page(page, pfn, zone, order, migratetype);
744         if (unlikely(!is_migrate_isolate(migratetype)))
745                 __mod_zone_freepage_state(zone, 1 << order, migratetype);
746         spin_unlock(&zone->lock);
747 }
748 
749 static bool free_pages_prepare(struct page *page, unsigned int order)
750 {
751         int i;
752         int bad = 0;
753 
754         trace_mm_page_free(page, order);
755         kmemcheck_free_shadow(page, order);
756 
757         if (PageAnon(page))
758                 page->mapping = NULL;
759         for (i = 0; i < (1 << order); i++)
760                 bad += free_pages_check(page + i);
761         if (bad)
762                 return false;
763 
764         if (!PageHighMem(page)) {
765                 debug_check_no_locks_freed(page_address(page),
766                                            PAGE_SIZE << order);
767                 debug_check_no_obj_freed(page_address(page),
768                                            PAGE_SIZE << order);
769         }
770         arch_free_page(page, order);
771         kernel_map_pages(page, 1 << order, 0);
772 
773         return true;
774 }
775 
776 static void __free_pages_ok(struct page *page, unsigned int order)
777 {
778         unsigned long flags;
779         int migratetype;
780         unsigned long pfn = page_to_pfn(page);
781 
782         if (!free_pages_prepare(page, order))
783                 return;
784 
785         migratetype = get_pfnblock_migratetype(page, pfn);
786         local_irq_save(flags);
787         __count_vm_events(PGFREE, 1 << order);
788         set_freepage_migratetype(page, migratetype);
789         free_one_page(page_zone(page), page, pfn, order, migratetype);
790         local_irq_restore(flags);
791 }
792 
793 void __init __free_pages_bootmem(struct page *page, unsigned int order)
794 {
795         unsigned int nr_pages = 1 << order;
796         struct page *p = page;
797         unsigned int loop;
798 
799         prefetchw(p);
800         for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
801                 prefetchw(p + 1);
802                 __ClearPageReserved(p);
803                 set_page_count(p, 0);
804         }
805         __ClearPageReserved(p);
806         set_page_count(p, 0);
807 
808         page_zone(page)->managed_pages += nr_pages;
809         set_page_refcounted(page);
810         __free_pages(page, order);
811 }
812 
813 #ifdef CONFIG_CMA
814 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
815 void __init init_cma_reserved_pageblock(struct page *page)
816 {
817         unsigned i = pageblock_nr_pages;
818         struct page *p = page;
819 
820         do {
821                 __ClearPageReserved(p);
822                 set_page_count(p, 0);
823         } while (++p, --i);
824 
825         set_pageblock_migratetype(page, MIGRATE_CMA);
826 
827         if (pageblock_order >= MAX_ORDER) {
828                 i = pageblock_nr_pages;
829                 p = page;
830                 do {
831                         set_page_refcounted(p);
832                         __free_pages(p, MAX_ORDER - 1);
833                         p += MAX_ORDER_NR_PAGES;
834                 } while (i -= MAX_ORDER_NR_PAGES);
835         } else {
836                 set_page_refcounted(page);
837                 __free_pages(page, pageblock_order);
838         }
839 
840         adjust_managed_page_count(page, pageblock_nr_pages);
841 }
842 #endif
843 
844 /*
845  * The order of subdivision here is critical for the IO subsystem.
846  * Please do not alter this order without good reasons and regression
847  * testing. Specifically, as large blocks of memory are subdivided,
848  * the order in which smaller blocks are delivered depends on the order
849  * they're subdivided in this function. This is the primary factor
850  * influencing the order in which pages are delivered to the IO
851  * subsystem according to empirical testing, and this is also justified
852  * by considering the behavior of a buddy system containing a single
853  * large block of memory acted on by a series of small allocations.
854  * This behavior is a critical factor in sglist merging's success.
855  *
856  * -- nyc
857  */
858 static inline void expand(struct zone *zone, struct page *page,
859         int low, int high, struct free_area *area,
860         int migratetype)
861 {
862         unsigned long size = 1 << high;
863 
864         while (high > low) {
865                 area--;
866                 high--;
867                 size >>= 1;
868                 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
869 
870 #ifdef CONFIG_DEBUG_PAGEALLOC
871                 if (high < debug_guardpage_minorder()) {
872                         /*
873                          * Mark as guard pages (or page), that will allow to
874                          * merge back to allocator when buddy will be freed.
875                          * Corresponding page table entries will not be touched,
876                          * pages will stay not present in virtual address space
877                          */
878                         INIT_LIST_HEAD(&page[size].lru);
879                         set_page_guard_flag(&page[size]);
880                         set_page_private(&page[size], high);
881                         /* Guard pages are not available for any usage */
882                         __mod_zone_freepage_state(zone, -(1 << high),
883                                                   migratetype);
884                         continue;
885                 }
886 #endif
887                 list_add(&page[size].lru, &area->free_list[migratetype]);
888                 area->nr_free++;
889                 set_page_order(&page[size], high);
890         }
891 }
892 
893 /*
894  * This page is about to be returned from the page allocator
895  */
896 static inline int check_new_page(struct page *page)
897 {
898         const char *bad_reason = NULL;
899         unsigned long bad_flags = 0;
900 
901         if (unlikely(page_mapcount(page)))
902                 bad_reason = "nonzero mapcount";
903         if (unlikely(page->mapping != NULL))
904                 bad_reason = "non-NULL mapping";
905         if (unlikely(atomic_read(&page->_count) != 0))
906                 bad_reason = "nonzero _count";
907         if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
908                 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
909                 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
910         }
911         if (unlikely(mem_cgroup_bad_page_check(page)))
912                 bad_reason = "cgroup check failed";
913         if (unlikely(bad_reason)) {
914                 bad_page(page, bad_reason, bad_flags);
915                 return 1;
916         }
917         return 0;
918 }
919 
920 static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags)
921 {
922         int i;
923 
924         for (i = 0; i < (1 << order); i++) {
925                 struct page *p = page + i;
926                 if (unlikely(check_new_page(p)))
927                         return 1;
928         }
929 
930         set_page_private(page, 0);
931         set_page_refcounted(page);
932 
933         arch_alloc_page(page, order);
934         kernel_map_pages(page, 1 << order, 1);
935 
936         if (gfp_flags & __GFP_ZERO)
937                 prep_zero_page(page, order, gfp_flags);
938 
939         if (order && (gfp_flags & __GFP_COMP))
940                 prep_compound_page(page, order);
941 
942         return 0;
943 }
944 
945 /*
946  * Go through the free lists for the given migratetype and remove
947  * the smallest available page from the freelists
948  */
949 static inline
950 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
951                                                 int migratetype)
952 {
953         unsigned int current_order;
954         struct free_area *area;
955         struct page *page;
956 
957         /* Find a page of the appropriate size in the preferred list */
958         for (current_order = order; current_order < MAX_ORDER; ++current_order) {
959                 area = &(zone->free_area[current_order]);
960                 if (list_empty(&area->free_list[migratetype]))
961                         continue;
962 
963                 page = list_entry(area->free_list[migratetype].next,
964                                                         struct page, lru);
965                 list_del(&page->lru);
966                 rmv_page_order(page);
967                 area->nr_free--;
968                 expand(zone, page, order, current_order, area, migratetype);
969                 set_freepage_migratetype(page, migratetype);
970                 return page;
971         }
972 
973         return NULL;
974 }
975 
976 
977 /*
978  * This array describes the order lists are fallen back to when
979  * the free lists for the desirable migrate type are depleted
980  */
981 static int fallbacks[MIGRATE_TYPES][4] = {
982         [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,     MIGRATE_RESERVE },
983         [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,     MIGRATE_RESERVE },
984 #ifdef CONFIG_CMA
985         [MIGRATE_MOVABLE]     = { MIGRATE_CMA,         MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
986         [MIGRATE_CMA]         = { MIGRATE_RESERVE }, /* Never used */
987 #else
988         [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,   MIGRATE_RESERVE },
989 #endif
990         [MIGRATE_RESERVE]     = { MIGRATE_RESERVE }, /* Never used */
991 #ifdef CONFIG_MEMORY_ISOLATION
992         [MIGRATE_ISOLATE]     = { MIGRATE_RESERVE }, /* Never used */
993 #endif
994 };
995 
996 /*
997  * Move the free pages in a range to the free lists of the requested type.
998  * Note that start_page and end_pages are not aligned on a pageblock
999  * boundary. If alignment is required, use move_freepages_block()
1000  */
1001 int move_freepages(struct zone *zone,
1002                           struct page *start_page, struct page *end_page,
1003                           int migratetype)
1004 {
1005         struct page *page;
1006         unsigned long order;
1007         int pages_moved = 0;
1008 
1009 #ifndef CONFIG_HOLES_IN_ZONE
1010         /*
1011          * page_zone is not safe to call in this context when
1012          * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1013          * anyway as we check zone boundaries in move_freepages_block().
1014          * Remove at a later date when no bug reports exist related to
1015          * grouping pages by mobility
1016          */
1017         BUG_ON(page_zone(start_page) != page_zone(end_page));
1018 #endif
1019 
1020         for (page = start_page; page <= end_page;) {
1021                 /* Make sure we are not inadvertently changing nodes */
1022                 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1023 
1024                 if (!pfn_valid_within(page_to_pfn(page))) {
1025                         page++;
1026                         continue;
1027                 }
1028 
1029                 if (!PageBuddy(page)) {
1030                         page++;
1031                         continue;
1032                 }
1033 
1034                 order = page_order(page);
1035                 list_move(&page->lru,
1036                           &zone->free_area[order].free_list[migratetype]);
1037                 set_freepage_migratetype(page, migratetype);
1038                 page += 1 << order;
1039                 pages_moved += 1 << order;
1040         }
1041 
1042         return pages_moved;
1043 }
1044 
1045 int move_freepages_block(struct zone *zone, struct page *page,
1046                                 int migratetype)
1047 {
1048         unsigned long start_pfn, end_pfn;
1049         struct page *start_page, *end_page;
1050 
1051         start_pfn = page_to_pfn(page);
1052         start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1053         start_page = pfn_to_page(start_pfn);
1054         end_page = start_page + pageblock_nr_pages - 1;
1055         end_pfn = start_pfn + pageblock_nr_pages - 1;
1056 
1057         /* Do not cross zone boundaries */
1058         if (!zone_spans_pfn(zone, start_pfn))
1059                 start_page = page;
1060         if (!zone_spans_pfn(zone, end_pfn))
1061                 return 0;
1062 
1063         return move_freepages(zone, start_page, end_page, migratetype);
1064 }
1065 
1066 static void change_pageblock_range(struct page *pageblock_page,
1067                                         int start_order, int migratetype)
1068 {
1069         int nr_pageblocks = 1 << (start_order - pageblock_order);
1070 
1071         while (nr_pageblocks--) {
1072                 set_pageblock_migratetype(pageblock_page, migratetype);
1073                 pageblock_page += pageblock_nr_pages;
1074         }
1075 }
1076 
1077 /*
1078  * If breaking a large block of pages, move all free pages to the preferred
1079  * allocation list. If falling back for a reclaimable kernel allocation, be
1080  * more aggressive about taking ownership of free pages.
1081  *
1082  * On the other hand, never change migration type of MIGRATE_CMA pageblocks
1083  * nor move CMA pages to different free lists. We don't want unmovable pages
1084  * to be allocated from MIGRATE_CMA areas.
1085  *
1086  * Returns the new migratetype of the pageblock (or the same old migratetype
1087  * if it was unchanged).
1088  */
1089 static int try_to_steal_freepages(struct zone *zone, struct page *page,
1090                                   int start_type, int fallback_type)
1091 {
1092         int current_order = page_order(page);
1093 
1094         /*
1095          * When borrowing from MIGRATE_CMA, we need to release the excess
1096          * buddy pages to CMA itself. We also ensure the freepage_migratetype
1097          * is set to CMA so it is returned to the correct freelist in case
1098          * the page ends up being not actually allocated from the pcp lists.
1099          */
1100         if (is_migrate_cma(fallback_type))
1101                 return fallback_type;
1102 
1103         /* Take ownership for orders >= pageblock_order */
1104         if (current_order >= pageblock_order) {
1105                 change_pageblock_range(page, current_order, start_type);
1106                 return start_type;
1107         }
1108 
1109         if (current_order >= pageblock_order / 2 ||
1110             start_type == MIGRATE_RECLAIMABLE ||
1111             page_group_by_mobility_disabled) {
1112                 int pages;
1113 
1114                 pages = move_freepages_block(zone, page, start_type);
1115 
1116                 /* Claim the whole block if over half of it is free */
1117                 if (pages >= (1 << (pageblock_order-1)) ||
1118                                 page_group_by_mobility_disabled) {
1119 
1120                         set_pageblock_migratetype(page, start_type);
1121                         return start_type;
1122                 }
1123 
1124         }
1125 
1126         return fallback_type;
1127 }
1128 
1129 /* Remove an element from the buddy allocator from the fallback list */
1130 static inline struct page *
1131 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
1132 {
1133         struct free_area *area;
1134         unsigned int current_order;
1135         struct page *page;
1136         int migratetype, new_type, i;
1137 
1138         /* Find the largest possible block of pages in the other list */
1139         for (current_order = MAX_ORDER-1;
1140                                 current_order >= order && current_order <= MAX_ORDER-1;
1141                                 --current_order) {
1142                 for (i = 0;; i++) {
1143                         migratetype = fallbacks[start_migratetype][i];
1144 
1145                         /* MIGRATE_RESERVE handled later if necessary */
1146                         if (migratetype == MIGRATE_RESERVE)
1147                                 break;
1148 
1149                         area = &(zone->free_area[current_order]);
1150                         if (list_empty(&area->free_list[migratetype]))
1151                                 continue;
1152 
1153                         page = list_entry(area->free_list[migratetype].next,
1154                                         struct page, lru);
1155                         area->nr_free--;
1156 
1157                         new_type = try_to_steal_freepages(zone, page,
1158                                                           start_migratetype,
1159                                                           migratetype);
1160 
1161                         /* Remove the page from the freelists */
1162                         list_del(&page->lru);
1163                         rmv_page_order(page);
1164 
1165                         expand(zone, page, order, current_order, area,
1166                                new_type);
1167                         /* The freepage_migratetype may differ from pageblock's
1168                          * migratetype depending on the decisions in
1169                          * try_to_steal_freepages. This is OK as long as it does
1170                          * not differ for MIGRATE_CMA type.
1171                          */
1172                         set_freepage_migratetype(page, new_type);
1173 
1174                         trace_mm_page_alloc_extfrag(page, order, current_order,
1175                                 start_migratetype, migratetype, new_type);
1176 
1177                         return page;
1178                 }
1179         }
1180 
1181         return NULL;
1182 }
1183 
1184 /*
1185  * Do the hard work of removing an element from the buddy allocator.
1186  * Call me with the zone->lock already held.
1187  */
1188 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1189                                                 int migratetype)
1190 {
1191         struct page *page;
1192 
1193 retry_reserve:
1194         page = __rmqueue_smallest(zone, order, migratetype);
1195 
1196         if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1197                 page = __rmqueue_fallback(zone, order, migratetype);
1198 
1199                 /*
1200                  * Use MIGRATE_RESERVE rather than fail an allocation. goto
1201                  * is used because __rmqueue_smallest is an inline function
1202                  * and we want just one call site
1203                  */
1204                 if (!page) {
1205                         migratetype = MIGRATE_RESERVE;
1206                         goto retry_reserve;
1207                 }
1208         }
1209 
1210         trace_mm_page_alloc_zone_locked(page, order, migratetype);
1211         return page;
1212 }
1213 
1214 /*
1215  * Obtain a specified number of elements from the buddy allocator, all under
1216  * a single hold of the lock, for efficiency.  Add them to the supplied list.
1217  * Returns the number of new pages which were placed at *list.
1218  */
1219 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1220                         unsigned long count, struct list_head *list,
1221                         int migratetype, bool cold)
1222 {
1223         int i;
1224 
1225         spin_lock(&zone->lock);
1226         for (i = 0; i < count; ++i) {
1227                 struct page *page = __rmqueue(zone, order, migratetype);
1228                 if (unlikely(page == NULL))
1229                         break;
1230 
1231                 /*
1232                  * Split buddy pages returned by expand() are received here
1233                  * in physical page order. The page is added to the callers and
1234                  * list and the list head then moves forward. From the callers
1235                  * perspective, the linked list is ordered by page number in
1236                  * some conditions. This is useful for IO devices that can
1237                  * merge IO requests if the physical pages are ordered
1238                  * properly.
1239                  */
1240                 if (likely(!cold))
1241                         list_add(&page->lru, list);
1242                 else
1243                         list_add_tail(&page->lru, list);
1244                 list = &page->lru;
1245                 if (is_migrate_cma(get_freepage_migratetype(page)))
1246                         __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1247                                               -(1 << order));
1248         }
1249         __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1250         spin_unlock(&zone->lock);
1251         return i;
1252 }
1253 
1254 #ifdef CONFIG_NUMA
1255 /*
1256  * Called from the vmstat counter updater to drain pagesets of this
1257  * currently executing processor on remote nodes after they have
1258  * expired.
1259  *
1260  * Note that this function must be called with the thread pinned to
1261  * a single processor.
1262  */
1263 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1264 {
1265         unsigned long flags;
1266         int to_drain, batch;
1267 
1268         local_irq_save(flags);
1269         batch = ACCESS_ONCE(pcp->batch);
1270         to_drain = min(pcp->count, batch);
1271         if (to_drain > 0) {
1272                 free_pcppages_bulk(zone, to_drain, pcp);
1273                 pcp->count -= to_drain;
1274         }
1275         local_irq_restore(flags);
1276 }
1277 #endif
1278 
1279 /*
1280  * Drain pages of the indicated processor.
1281  *
1282  * The processor must either be the current processor and the
1283  * thread pinned to the current processor or a processor that
1284  * is not online.
1285  */
1286 static void drain_pages(unsigned int cpu)
1287 {
1288         unsigned long flags;
1289         struct zone *zone;
1290 
1291         for_each_populated_zone(zone) {
1292                 struct per_cpu_pageset *pset;
1293                 struct per_cpu_pages *pcp;
1294 
1295                 local_irq_save(flags);
1296                 pset = per_cpu_ptr(zone->pageset, cpu);
1297 
1298                 pcp = &pset->pcp;
1299                 if (pcp->count) {
1300                         free_pcppages_bulk(zone, pcp->count, pcp);
1301                         pcp->count = 0;
1302                 }
1303                 local_irq_restore(flags);
1304         }
1305 }
1306 
1307 /*
1308  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1309  */
1310 void drain_local_pages(void *arg)
1311 {
1312         drain_pages(smp_processor_id());
1313 }
1314 
1315 /*
1316  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1317  *
1318  * Note that this code is protected against sending an IPI to an offline
1319  * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1320  * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1321  * nothing keeps CPUs from showing up after we populated the cpumask and
1322  * before the call to on_each_cpu_mask().
1323  */
1324 void drain_all_pages(void)
1325 {
1326         int cpu;
1327         struct per_cpu_pageset *pcp;
1328         struct zone *zone;
1329 
1330         /*
1331          * Allocate in the BSS so we wont require allocation in
1332          * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1333          */
1334         static cpumask_t cpus_with_pcps;
1335 
1336         /*
1337          * We don't care about racing with CPU hotplug event
1338          * as offline notification will cause the notified
1339          * cpu to drain that CPU pcps and on_each_cpu_mask
1340          * disables preemption as part of its processing
1341          */
1342         for_each_online_cpu(cpu) {
1343                 bool has_pcps = false;
1344                 for_each_populated_zone(zone) {
1345                         pcp = per_cpu_ptr(zone->pageset, cpu);
1346                         if (pcp->pcp.count) {
1347                                 has_pcps = true;
1348                                 break;
1349                         }
1350                 }
1351                 if (has_pcps)
1352                         cpumask_set_cpu(cpu, &cpus_with_pcps);
1353                 else
1354                         cpumask_clear_cpu(cpu, &cpus_with_pcps);
1355         }
1356         on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
1357 }
1358 
1359 #ifdef CONFIG_HIBERNATION
1360 
1361 void mark_free_pages(struct zone *zone)
1362 {
1363         unsigned long pfn, max_zone_pfn;
1364         unsigned long flags;
1365         unsigned int order, t;
1366         struct list_head *curr;
1367 
1368         if (zone_is_empty(zone))
1369                 return;
1370 
1371         spin_lock_irqsave(&zone->lock, flags);
1372 
1373         max_zone_pfn = zone_end_pfn(zone);
1374         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1375                 if (pfn_valid(pfn)) {
1376                         struct page *page = pfn_to_page(pfn);
1377 
1378                         if (!swsusp_page_is_forbidden(page))
1379                                 swsusp_unset_page_free(page);
1380                 }
1381 
1382         for_each_migratetype_order(order, t) {
1383                 list_for_each(curr, &zone->free_area[order].free_list[t]) {
1384                         unsigned long i;
1385 
1386                         pfn = page_to_pfn(list_entry(curr, struct page, lru));
1387                         for (i = 0; i < (1UL << order); i++)
1388                                 swsusp_set_page_free(pfn_to_page(pfn + i));
1389                 }
1390         }
1391         spin_unlock_irqrestore(&zone->lock, flags);
1392 }
1393 #endif /* CONFIG_PM */
1394 
1395 /*
1396  * Free a 0-order page
1397  * cold == true ? free a cold page : free a hot page
1398  */
1399 void free_hot_cold_page(struct page *page, bool cold)
1400 {
1401         struct zone *zone = page_zone(page);
1402         struct per_cpu_pages *pcp;
1403         unsigned long flags;
1404         unsigned long pfn = page_to_pfn(page);
1405         int migratetype;
1406 
1407         if (!free_pages_prepare(page, 0))
1408                 return;
1409 
1410         migratetype = get_pfnblock_migratetype(page, pfn);
1411         set_freepage_migratetype(page, migratetype);
1412         local_irq_save(flags);
1413         __count_vm_event(PGFREE);
1414 
1415         /*
1416          * We only track unmovable, reclaimable and movable on pcp lists.
1417          * Free ISOLATE pages back to the allocator because they are being
1418          * offlined but treat RESERVE as movable pages so we can get those
1419          * areas back if necessary. Otherwise, we may have to free
1420          * excessively into the page allocator
1421          */
1422         if (migratetype >= MIGRATE_PCPTYPES) {
1423                 if (unlikely(is_migrate_isolate(migratetype))) {
1424                         free_one_page(zone, page, pfn, 0, migratetype);
1425                         goto out;
1426                 }
1427                 migratetype = MIGRATE_MOVABLE;
1428         }
1429 
1430         pcp = &this_cpu_ptr(zone->pageset)->pcp;
1431         if (!cold)
1432                 list_add(&page->lru, &pcp->lists[migratetype]);
1433         else
1434                 list_add_tail(&page->lru, &pcp->lists[migratetype]);
1435         pcp->count++;
1436         if (pcp->count >= pcp->high) {
1437                 unsigned long batch = ACCESS_ONCE(pcp->batch);
1438                 free_pcppages_bulk(zone, batch, pcp);
1439                 pcp->count -= batch;
1440         }
1441 
1442 out:
1443         local_irq_restore(flags);
1444 }
1445 
1446 /*
1447  * Free a list of 0-order pages
1448  */
1449 void free_hot_cold_page_list(struct list_head *list, bool cold)
1450 {
1451         struct page *page, *next;
1452 
1453         list_for_each_entry_safe(page, next, list, lru) {
1454                 trace_mm_page_free_batched(page, cold);
1455                 free_hot_cold_page(page, cold);
1456         }
1457 }
1458 
1459 /*
1460  * split_page takes a non-compound higher-order page, and splits it into
1461  * n (1<<order) sub-pages: page[0..n]
1462  * Each sub-page must be freed individually.
1463  *
1464  * Note: this is probably too low level an operation for use in drivers.
1465  * Please consult with lkml before using this in your driver.
1466  */
1467 void split_page(struct page *page, unsigned int order)
1468 {
1469         int i;
1470 
1471         VM_BUG_ON_PAGE(PageCompound(page), page);
1472         VM_BUG_ON_PAGE(!page_count(page), page);
1473 
1474 #ifdef CONFIG_KMEMCHECK
1475         /*
1476          * Split shadow pages too, because free(page[0]) would
1477          * otherwise free the whole shadow.
1478          */
1479         if (kmemcheck_page_is_tracked(page))
1480                 split_page(virt_to_page(page[0].shadow), order);
1481 #endif
1482 
1483         for (i = 1; i < (1 << order); i++)
1484                 set_page_refcounted(page + i);
1485 }
1486 EXPORT_SYMBOL_GPL(split_page);
1487 
1488 static int __isolate_free_page(struct page *page, unsigned int order)
1489 {
1490         unsigned long watermark;
1491         struct zone *zone;
1492         int mt;
1493 
1494         BUG_ON(!PageBuddy(page));
1495 
1496         zone = page_zone(page);
1497         mt = get_pageblock_migratetype(page);
1498 
1499         if (!is_migrate_isolate(mt)) {
1500                 /* Obey watermarks as if the page was being allocated */
1501                 watermark = low_wmark_pages(zone) + (1 << order);
1502                 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1503                         return 0;
1504 
1505                 __mod_zone_freepage_state(zone, -(1UL << order), mt);
1506         }
1507 
1508         /* Remove page from free list */
1509         list_del(&page->lru);
1510         zone->free_area[order].nr_free--;
1511         rmv_page_order(page);
1512 
1513         /* Set the pageblock if the isolated page is at least a pageblock */
1514         if (order >= pageblock_order - 1) {
1515                 struct page *endpage = page + (1 << order) - 1;
1516                 for (; page < endpage; page += pageblock_nr_pages) {
1517                         int mt = get_pageblock_migratetype(page);
1518                         if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
1519                                 set_pageblock_migratetype(page,
1520                                                           MIGRATE_MOVABLE);
1521                 }
1522         }
1523 
1524         return 1UL << order;
1525 }
1526 
1527 /*
1528  * Similar to split_page except the page is already free. As this is only
1529  * being used for migration, the migratetype of the block also changes.
1530  * As this is called with interrupts disabled, the caller is responsible
1531  * for calling arch_alloc_page() and kernel_map_page() after interrupts
1532  * are enabled.
1533  *
1534  * Note: this is probably too low level an operation for use in drivers.
1535  * Please consult with lkml before using this in your driver.
1536  */
1537 int split_free_page(struct page *page)
1538 {
1539         unsigned int order;
1540         int nr_pages;
1541 
1542         order = page_order(page);
1543 
1544         nr_pages = __isolate_free_page(page, order);
1545         if (!nr_pages)
1546                 return 0;
1547 
1548         /* Split into individual pages */
1549         set_page_refcounted(page);
1550         split_page(page, order);
1551         return nr_pages;
1552 }
1553 
1554 /*
1555  * Really, prep_compound_page() should be called from __rmqueue_bulk().  But
1556  * we cheat by calling it from here, in the order > 0 path.  Saves a branch
1557  * or two.
1558  */
1559 static inline
1560 struct page *buffered_rmqueue(struct zone *preferred_zone,
1561                         struct zone *zone, unsigned int order,
1562                         gfp_t gfp_flags, int migratetype)
1563 {
1564         unsigned long flags;
1565         struct page *page;
1566         bool cold = ((gfp_flags & __GFP_COLD) != 0);
1567 
1568 again:
1569         if (likely(order == 0)) {
1570                 struct per_cpu_pages *pcp;
1571                 struct list_head *list;
1572 
1573                 local_irq_save(flags);
1574                 pcp = &this_cpu_ptr(zone->pageset)->pcp;
1575                 list = &pcp->lists[migratetype];
1576                 if (list_empty(list)) {
1577                         pcp->count += rmqueue_bulk(zone, 0,
1578                                         pcp->batch, list,
1579                                         migratetype, cold);
1580                         if (unlikely(list_empty(list)))
1581                                 goto failed;
1582                 }
1583 
1584                 if (cold)
1585                         page = list_entry(list->prev, struct page, lru);
1586                 else
1587                         page = list_entry(list->next, struct page, lru);
1588 
1589                 list_del(&page->lru);
1590                 pcp->count--;
1591         } else {
1592                 if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1593                         /*
1594                          * __GFP_NOFAIL is not to be used in new code.
1595                          *
1596                          * All __GFP_NOFAIL callers should be fixed so that they
1597                          * properly detect and handle allocation failures.
1598                          *
1599                          * We most definitely don't want callers attempting to
1600                          * allocate greater than order-1 page units with
1601                          * __GFP_NOFAIL.
1602                          */
1603                         WARN_ON_ONCE(order > 1);
1604                 }
1605                 spin_lock_irqsave(&zone->lock, flags);
1606                 page = __rmqueue(zone, order, migratetype);
1607                 spin_unlock(&zone->lock);
1608                 if (!page)
1609                         goto failed;
1610                 __mod_zone_freepage_state(zone, -(1 << order),
1611                                           get_freepage_migratetype(page));
1612         }
1613 
1614         __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
1615         if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
1616             !zone_is_fair_depleted(zone))
1617                 zone_set_flag(zone, ZONE_FAIR_DEPLETED);
1618 
1619         __count_zone_vm_events(PGALLOC, zone, 1 << order);
1620         zone_statistics(preferred_zone, zone, gfp_flags);
1621         local_irq_restore(flags);
1622 
1623         VM_BUG_ON_PAGE(bad_range(zone, page), page);
1624         if (prep_new_page(page, order, gfp_flags))
1625                 goto again;
1626         return page;
1627 
1628 failed:
1629         local_irq_restore(flags);
1630         return NULL;
1631 }
1632 
1633 #ifdef CONFIG_FAIL_PAGE_ALLOC
1634 
1635 static struct {
1636         struct fault_attr attr;
1637 
1638         u32 ignore_gfp_highmem;
1639         u32 ignore_gfp_wait;
1640         u32 min_order;
1641 } fail_page_alloc = {
1642         .attr = FAULT_ATTR_INITIALIZER,
1643         .ignore_gfp_wait = 1,
1644         .ignore_gfp_highmem = 1,
1645         .min_order = 1,
1646 };
1647 
1648 static int __init setup_fail_page_alloc(char *str)
1649 {
1650         return setup_fault_attr(&fail_page_alloc.attr, str);
1651 }
1652 __setup("fail_page_alloc=", setup_fail_page_alloc);
1653 
1654 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1655 {
1656         if (order < fail_page_alloc.min_order)
1657                 return false;
1658         if (gfp_mask & __GFP_NOFAIL)
1659                 return false;
1660         if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1661                 return false;
1662         if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1663                 return false;
1664 
1665         return should_fail(&fail_page_alloc.attr, 1 << order);
1666 }
1667 
1668 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1669 
1670 static int __init fail_page_alloc_debugfs(void)
1671 {
1672         umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1673         struct dentry *dir;
1674 
1675         dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1676                                         &fail_page_alloc.attr);
1677         if (IS_ERR(dir))
1678                 return PTR_ERR(dir);
1679 
1680         if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1681                                 &fail_page_alloc.ignore_gfp_wait))
1682                 goto fail;
1683         if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1684                                 &fail_page_alloc.ignore_gfp_highmem))
1685                 goto fail;
1686         if (!debugfs_create_u32("min-order", mode, dir,
1687                                 &fail_page_alloc.min_order))
1688                 goto fail;
1689 
1690         return 0;
1691 fail:
1692         debugfs_remove_recursive(dir);
1693 
1694         return -ENOMEM;
1695 }
1696 
1697 late_initcall(fail_page_alloc_debugfs);
1698 
1699 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1700 
1701 #else /* CONFIG_FAIL_PAGE_ALLOC */
1702 
1703 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1704 {
1705         return false;
1706 }
1707 
1708 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1709 
1710 /*
1711  * Return true if free pages are above 'mark'. This takes into account the order
1712  * of the allocation.
1713  */
1714 static bool __zone_watermark_ok(struct zone *z, unsigned int order,
1715                         unsigned long mark, int classzone_idx, int alloc_flags,
1716                         long free_pages)
1717 {
1718         /* free_pages my go negative - that's OK */
1719         long min = mark;
1720         int o;
1721         long free_cma = 0;
1722 
1723         free_pages -= (1 << order) - 1;
1724         if (alloc_flags & ALLOC_HIGH)
1725                 min -= min / 2;
1726         if (alloc_flags & ALLOC_HARDER)
1727                 min -= min / 4;
1728 #ifdef CONFIG_CMA
1729         /* If allocation can't use CMA areas don't use free CMA pages */
1730         if (!(alloc_flags & ALLOC_CMA))
1731                 free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
1732 #endif
1733 
1734         if (free_pages - free_cma <= min + z->lowmem_reserve[classzone_idx])
1735                 return false;
1736         for (o = 0; o < order; o++) {
1737                 /* At the next order, this order's pages become unavailable */
1738                 free_pages -= z->free_area[o].nr_free << o;
1739 
1740                 /* Require fewer higher order pages to be free */
1741                 min >>= 1;
1742 
1743                 if (free_pages <= min)
1744                         return false;
1745         }
1746         return true;
1747 }
1748 
1749 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
1750                       int classzone_idx, int alloc_flags)
1751 {
1752         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1753                                         zone_page_state(z, NR_FREE_PAGES));
1754 }
1755 
1756 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
1757                         unsigned long mark, int classzone_idx, int alloc_flags)
1758 {
1759         long free_pages = zone_page_state(z, NR_FREE_PAGES);
1760 
1761         if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1762                 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1763 
1764         return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1765                                                                 free_pages);
1766 }
1767 
1768 #ifdef CONFIG_NUMA
1769 /*
1770  * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
1771  * skip over zones that are not allowed by the cpuset, or that have
1772  * been recently (in last second) found to be nearly full.  See further
1773  * comments in mmzone.h.  Reduces cache footprint of zonelist scans
1774  * that have to skip over a lot of full or unallowed zones.
1775  *
1776  * If the zonelist cache is present in the passed zonelist, then
1777  * returns a pointer to the allowed node mask (either the current
1778  * tasks mems_allowed, or node_states[N_MEMORY].)
1779  *
1780  * If the zonelist cache is not available for this zonelist, does
1781  * nothing and returns NULL.
1782  *
1783  * If the fullzones BITMAP in the zonelist cache is stale (more than
1784  * a second since last zap'd) then we zap it out (clear its bits.)
1785  *
1786  * We hold off even calling zlc_setup, until after we've checked the
1787  * first zone in the zonelist, on the theory that most allocations will
1788  * be satisfied from that first zone, so best to examine that zone as
1789  * quickly as we can.
1790  */
1791 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1792 {
1793         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1794         nodemask_t *allowednodes;       /* zonelist_cache approximation */
1795 
1796         zlc = zonelist->zlcache_ptr;
1797         if (!zlc)
1798                 return NULL;
1799 
1800         if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1801                 bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1802                 zlc->last_full_zap = jiffies;
1803         }
1804 
1805         allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1806                                         &cpuset_current_mems_allowed :
1807                                         &node_states[N_MEMORY];
1808         return allowednodes;
1809 }
1810 
1811 /*
1812  * Given 'z' scanning a zonelist, run a couple of quick checks to see
1813  * if it is worth looking at further for free memory:
1814  *  1) Check that the zone isn't thought to be full (doesn't have its
1815  *     bit set in the zonelist_cache fullzones BITMAP).
1816  *  2) Check that the zones node (obtained from the zonelist_cache
1817  *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1818  * Return true (non-zero) if zone is worth looking at further, or
1819  * else return false (zero) if it is not.
1820  *
1821  * This check -ignores- the distinction between various watermarks,
1822  * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
1823  * found to be full for any variation of these watermarks, it will
1824  * be considered full for up to one second by all requests, unless
1825  * we are so low on memory on all allowed nodes that we are forced
1826  * into the second scan of the zonelist.
1827  *
1828  * In the second scan we ignore this zonelist cache and exactly
1829  * apply the watermarks to all zones, even it is slower to do so.
1830  * We are low on memory in the second scan, and should leave no stone
1831  * unturned looking for a free page.
1832  */
1833 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1834                                                 nodemask_t *allowednodes)
1835 {
1836         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1837         int i;                          /* index of *z in zonelist zones */
1838         int n;                          /* node that zone *z is on */
1839 
1840         zlc = zonelist->zlcache_ptr;
1841         if (!zlc)
1842                 return 1;
1843 
1844         i = z - zonelist->_zonerefs;
1845         n = zlc->z_to_n[i];
1846 
1847         /* This zone is worth trying if it is allowed but not full */
1848         return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1849 }
1850 
1851 /*
1852  * Given 'z' scanning a zonelist, set the corresponding bit in
1853  * zlc->fullzones, so that subsequent attempts to allocate a page
1854  * from that zone don't waste time re-examining it.
1855  */
1856 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1857 {
1858         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1859         int i;                          /* index of *z in zonelist zones */
1860 
1861         zlc = zonelist->zlcache_ptr;
1862         if (!zlc)
1863                 return;
1864 
1865         i = z - zonelist->_zonerefs;
1866 
1867         set_bit(i, zlc->fullzones);
1868 }
1869 
1870 /*
1871  * clear all zones full, called after direct reclaim makes progress so that
1872  * a zone that was recently full is not skipped over for up to a second
1873  */
1874 static void zlc_clear_zones_full(struct zonelist *zonelist)
1875 {
1876         struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1877 
1878         zlc = zonelist->zlcache_ptr;
1879         if (!zlc)
1880                 return;
1881 
1882         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1883 }
1884 
1885 static bool zone_local(struct zone *local_zone, struct zone *zone)
1886 {
1887         return local_zone->node == zone->node;
1888 }
1889 
1890 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1891 {
1892         return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
1893                                 RECLAIM_DISTANCE;
1894 }
1895 
1896 #else   /* CONFIG_NUMA */
1897 
1898 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1899 {
1900         return NULL;
1901 }
1902 
1903 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1904                                 nodemask_t *allowednodes)
1905 {
1906         return 1;
1907 }
1908 
1909 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1910 {
1911 }
1912 
1913 static void zlc_clear_zones_full(struct zonelist *zonelist)
1914 {
1915 }
1916 
1917 static bool zone_local(struct zone *local_zone, struct zone *zone)
1918 {
1919         return true;
1920 }
1921 
1922 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1923 {
1924         return true;
1925 }
1926 
1927 #endif  /* CONFIG_NUMA */
1928 
1929 static void reset_alloc_batches(struct zone *preferred_zone)
1930 {
1931         struct zone *zone = preferred_zone->zone_pgdat->node_zones;
1932 
1933         do {
1934                 mod_zone_page_state(zone, NR_ALLOC_BATCH,
1935                         high_wmark_pages(zone) - low_wmark_pages(zone) -
1936                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
1937                 zone_clear_flag(zone, ZONE_FAIR_DEPLETED);
1938         } while (zone++ != preferred_zone);
1939 }
1940 
1941 /*
1942  * get_page_from_freelist goes through the zonelist trying to allocate
1943  * a page.
1944  */
1945 static struct page *
1946 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1947                 struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1948                 struct zone *preferred_zone, int classzone_idx, int migratetype)
1949 {
1950         struct zoneref *z;
1951         struct page *page = NULL;
1952         struct zone *zone;
1953         nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1954         int zlc_active = 0;             /* set if using zonelist_cache */
1955         int did_zlc_setup = 0;          /* just call zlc_setup() one time */
1956         bool consider_zone_dirty = (alloc_flags & ALLOC_WMARK_LOW) &&
1957                                 (gfp_mask & __GFP_WRITE);
1958         int nr_fair_skipped = 0;
1959         bool zonelist_rescan;
1960 
1961 zonelist_scan:
1962         zonelist_rescan = false;
1963 
1964         /*
1965          * Scan zonelist, looking for a zone with enough free.
1966          * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c.
1967          */
1968         for_each_zone_zonelist_nodemask(zone, z, zonelist,
1969                                                 high_zoneidx, nodemask) {
1970                 unsigned long mark;
1971 
1972                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
1973                         !zlc_zone_worth_trying(zonelist, z, allowednodes))
1974                                 continue;
1975                 if (cpusets_enabled() &&
1976                         (alloc_flags & ALLOC_CPUSET) &&
1977                         !cpuset_zone_allowed_softwall(zone, gfp_mask))
1978                                 continue;
1979                 /*
1980                  * Distribute pages in proportion to the individual
1981                  * zone size to ensure fair page aging.  The zone a
1982                  * page was allocated in should have no effect on the
1983                  * time the page has in memory before being reclaimed.
1984                  */
1985                 if (alloc_flags & ALLOC_FAIR) {
1986                         if (!zone_local(preferred_zone, zone))
1987                                 break;
1988                         if (zone_is_fair_depleted(zone)) {
1989                                 nr_fair_skipped++;
1990                                 continue;
1991                         }
1992                 }
1993                 /*
1994                  * When allocating a page cache page for writing, we
1995                  * want to get it from a zone that is within its dirty
1996                  * limit, such that no single zone holds more than its
1997                  * proportional share of globally allowed dirty pages.
1998                  * The dirty limits take into account the zone's
1999                  * lowmem reserves and high watermark so that kswapd
2000                  * should be able to balance it without having to
2001                  * write pages from its LRU list.
2002                  *
2003                  * This may look like it could increase pressure on
2004                  * lower zones by failing allocations in higher zones
2005                  * before they are full.  But the pages that do spill
2006                  * over are limited as the lower zones are protected
2007                  * by this very same mechanism.  It should not become
2008                  * a practical burden to them.
2009                  *
2010                  * XXX: For now, allow allocations to potentially
2011                  * exceed the per-zone dirty limit in the slowpath
2012                  * (ALLOC_WMARK_LOW unset) before going into reclaim,
2013                  * which is important when on a NUMA setup the allowed
2014                  * zones are together not big enough to reach the
2015                  * global limit.  The proper fix for these situations
2016                  * will require awareness of zones in the
2017                  * dirty-throttling and the flusher threads.
2018                  */
2019                 if (consider_zone_dirty && !zone_dirty_ok(zone))
2020                         continue;
2021 
2022                 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2023                 if (!zone_watermark_ok(zone, order, mark,
2024                                        classzone_idx, alloc_flags)) {
2025                         int ret;
2026 
2027                         /* Checked here to keep the fast path fast */
2028                         BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2029                         if (alloc_flags & ALLOC_NO_WATERMARKS)
2030                                 goto try_this_zone;
2031 
2032                         if (IS_ENABLED(CONFIG_NUMA) &&
2033                                         !did_zlc_setup && nr_online_nodes > 1) {
2034                                 /*
2035                                  * we do zlc_setup if there are multiple nodes
2036                                  * and before considering the first zone allowed
2037                                  * by the cpuset.
2038                                  */
2039                                 allowednodes = zlc_setup(zonelist, alloc_flags);
2040                                 zlc_active = 1;
2041                                 did_zlc_setup = 1;
2042                         }
2043 
2044                         if (zone_reclaim_mode == 0 ||
2045                             !zone_allows_reclaim(preferred_zone, zone))
2046                                 goto this_zone_full;
2047 
2048                         /*
2049                          * As we may have just activated ZLC, check if the first
2050                          * eligible zone has failed zone_reclaim recently.
2051                          */
2052                         if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2053                                 !zlc_zone_worth_trying(zonelist, z, allowednodes))
2054                                 continue;
2055 
2056                         ret = zone_reclaim(zone, gfp_mask, order);
2057                         switch (ret) {
2058                         case ZONE_RECLAIM_NOSCAN:
2059                                 /* did not scan */
2060                                 continue;
2061                         case ZONE_RECLAIM_FULL:
2062                                 /* scanned but unreclaimable */
2063                                 continue;
2064                         default:
2065                                 /* did we reclaim enough */
2066                                 if (zone_watermark_ok(zone, order, mark,
2067                                                 classzone_idx, alloc_flags))
2068                                         goto try_this_zone;
2069 
2070                                 /*
2071                                  * Failed to reclaim enough to meet watermark.
2072                                  * Only mark the zone full if checking the min
2073                                  * watermark or if we failed to reclaim just
2074                                  * 1<<order pages or else the page allocator
2075                                  * fastpath will prematurely mark zones full
2076                                  * when the watermark is between the low and
2077                                  * min watermarks.
2078                                  */
2079                                 if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2080                                     ret == ZONE_RECLAIM_SOME)
2081                                         goto this_zone_full;
2082 
2083                                 continue;
2084                         }
2085                 }
2086 
2087 try_this_zone:
2088                 page = buffered_rmqueue(preferred_zone, zone, order,
2089                                                 gfp_mask, migratetype);
2090                 if (page)
2091                         break;
2092 this_zone_full:
2093                 if (IS_ENABLED(CONFIG_NUMA) && zlc_active)
2094                         zlc_mark_zone_full(zonelist, z);
2095         }
2096 
2097         if (page) {
2098                 /*
2099                  * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2100                  * necessary to allocate the page. The expectation is
2101                  * that the caller is taking steps that will free more
2102                  * memory. The caller should avoid the page being used
2103                  * for !PFMEMALLOC purposes.
2104                  */
2105                 page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
2106                 return page;
2107         }
2108 
2109         /*
2110          * The first pass makes sure allocations are spread fairly within the
2111          * local node.  However, the local node might have free pages left
2112          * after the fairness batches are exhausted, and remote zones haven't
2113          * even been considered yet.  Try once more without fairness, and
2114          * include remote zones now, before entering the slowpath and waking
2115          * kswapd: prefer spilling to a remote zone over swapping locally.
2116          */
2117         if (alloc_flags & ALLOC_FAIR) {
2118                 alloc_flags &= ~ALLOC_FAIR;
2119                 if (nr_fair_skipped) {
2120                         zonelist_rescan = true;
2121                         reset_alloc_batches(preferred_zone);
2122                 }
2123                 if (nr_online_nodes > 1)
2124                         zonelist_rescan = true;
2125         }
2126 
2127         if (unlikely(IS_ENABLED(CONFIG_NUMA) && zlc_active)) {
2128                 /* Disable zlc cache for second zonelist scan */
2129                 zlc_active = 0;
2130                 zonelist_rescan = true;
2131         }
2132 
2133         if (zonelist_rescan)
2134                 goto zonelist_scan;
2135 
2136         return NULL;
2137 }
2138 
2139 /*
2140  * Large machines with many possible nodes should not always dump per-node
2141  * meminfo in irq context.
2142  */
2143 static inline bool should_suppress_show_mem(void)
2144 {
2145         bool ret = false;
2146 
2147 #if NODES_SHIFT > 8
2148         ret = in_interrupt();
2149 #endif
2150         return ret;
2151 }
2152 
2153 static DEFINE_RATELIMIT_STATE(nopage_rs,
2154                 DEFAULT_RATELIMIT_INTERVAL,
2155                 DEFAULT_RATELIMIT_BURST);
2156 
2157 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2158 {
2159         unsigned int filter = SHOW_MEM_FILTER_NODES;
2160 
2161         if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2162             debug_guardpage_minorder() > 0)
2163                 return;
2164 
2165         /*
2166          * This documents exceptions given to allocations in certain
2167          * contexts that are allowed to allocate outside current's set
2168          * of allowed nodes.
2169          */
2170         if (!(gfp_mask & __GFP_NOMEMALLOC))
2171                 if (test_thread_flag(TIF_MEMDIE) ||
2172                     (current->flags & (PF_MEMALLOC | PF_EXITING)))
2173                         filter &= ~SHOW_MEM_FILTER_NODES;
2174         if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2175                 filter &= ~SHOW_MEM_FILTER_NODES;
2176 
2177         if (fmt) {
2178                 struct va_format vaf;
2179                 va_list args;
2180 
2181                 va_start(args, fmt);
2182 
2183                 vaf.fmt = fmt;
2184                 vaf.va = &args;
2185 
2186                 pr_warn("%pV", &vaf);
2187 
2188                 va_end(args);
2189         }
2190 
2191         pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2192                 current->comm, order, gfp_mask);
2193 
2194         dump_stack();
2195         if (!should_suppress_show_mem())
2196                 show_mem(filter);
2197 }
2198 
2199 static inline int
2200 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
2201                                 unsigned long did_some_progress,
2202                                 unsigned long pages_reclaimed)
2203 {
2204         /* Do not loop if specifically requested */
2205         if (gfp_mask & __GFP_NORETRY)
2206                 return 0;
2207 
2208         /* Always retry if specifically requested */
2209         if (gfp_mask & __GFP_NOFAIL)
2210                 return 1;
2211 
2212         /*
2213          * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2214          * making forward progress without invoking OOM. Suspend also disables
2215          * storage devices so kswapd will not help. Bail if we are suspending.
2216          */
2217         if (!did_some_progress && pm_suspended_storage())
2218                 return 0;
2219 
2220         /*
2221          * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2222          * means __GFP_NOFAIL, but that may not be true in other
2223          * implementations.
2224          */
2225         if (order <= PAGE_ALLOC_COSTLY_ORDER)
2226                 return 1;
2227 
2228         /*
2229          * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2230          * specified, then we retry until we no longer reclaim any pages
2231          * (above), or we've reclaimed an order of pages at least as
2232          * large as the allocation's order. In both cases, if the
2233          * allocation still fails, we stop retrying.
2234          */
2235         if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2236                 return 1;
2237 
2238         return 0;
2239 }
2240 
2241 static inline struct page *
2242 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2243         struct zonelist *zonelist, enum zone_type high_zoneidx,
2244         nodemask_t *nodemask, struct zone *preferred_zone,
2245         int classzone_idx, int migratetype)
2246 {
2247         struct page *page;
2248 
2249         /* Acquire the per-zone oom lock for each zone */
2250         if (!oom_zonelist_trylock(zonelist, gfp_mask)) {
2251                 schedule_timeout_uninterruptible(1);
2252                 return NULL;
2253         }
2254 
2255         /*
2256          * Go through the zonelist yet one more time, keep very high watermark
2257          * here, this is only to catch a parallel oom killing, we must fail if
2258          * we're still under heavy pressure.
2259          */
2260         page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2261                 order, zonelist, high_zoneidx,
2262                 ALLOC_WMARK_HIGH|ALLOC_CPUSET,
2263                 preferred_zone, classzone_idx, migratetype);
2264         if (page)
2265                 goto out;
2266 
2267         if (!(gfp_mask & __GFP_NOFAIL)) {
2268                 /* The OOM killer will not help higher order allocs */
2269                 if (order > PAGE_ALLOC_COSTLY_ORDER)
2270                         goto out;
2271                 /* The OOM killer does not needlessly kill tasks for lowmem */
2272                 if (high_zoneidx < ZONE_NORMAL)
2273                         goto out;
2274                 /*
2275                  * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2276                  * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2277                  * The caller should handle page allocation failure by itself if
2278                  * it specifies __GFP_THISNODE.
2279                  * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2280                  */
2281                 if (gfp_mask & __GFP_THISNODE)
2282                         goto out;
2283         }
2284         /* Exhausted what can be done so it's blamo time */
2285         out_of_memory(zonelist, gfp_mask, order, nodemask, false);
2286 
2287 out:
2288         oom_zonelist_unlock(zonelist, gfp_mask);
2289         return page;
2290 }
2291 
2292 #ifdef CONFIG_COMPACTION
2293 /* Try memory compaction for high-order allocations before reclaim */
2294 static struct page *
2295 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2296         struct zonelist *zonelist, enum zone_type high_zoneidx,
2297         nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2298         int classzone_idx, int migratetype, enum migrate_mode mode,
2299         bool *contended_compaction, bool *deferred_compaction,
2300         unsigned long *did_some_progress)
2301 {
2302         if (!order)
2303                 return NULL;
2304 
2305         if (compaction_deferred(preferred_zone, order)) {
2306                 *deferred_compaction = true;
2307                 return NULL;
2308         }
2309 
2310         current->flags |= PF_MEMALLOC;
2311         *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
2312                                                 nodemask, mode,
2313                                                 contended_compaction);
2314         current->flags &= ~PF_MEMALLOC;
2315 
2316         if (*did_some_progress != COMPACT_SKIPPED) {
2317                 struct page *page;
2318 
2319                 /* Page migration frees to the PCP lists but we want merging */
2320                 drain_pages(get_cpu());
2321                 put_cpu();
2322 
2323                 page = get_page_from_freelist(gfp_mask, nodemask,
2324                                 order, zonelist, high_zoneidx,
2325                                 alloc_flags & ~ALLOC_NO_WATERMARKS,
2326                                 preferred_zone, classzone_idx, migratetype);
2327                 if (page) {
2328                         preferred_zone->compact_blockskip_flush = false;
2329                         compaction_defer_reset(preferred_zone, order, true);
2330                         count_vm_event(COMPACTSUCCESS);
2331                         return page;
2332                 }
2333 
2334                 /*
2335                  * It's bad if compaction run occurs and fails.
2336                  * The most likely reason is that pages exist,
2337                  * but not enough to satisfy watermarks.
2338                  */
2339                 count_vm_event(COMPACTFAIL);
2340 
2341                 /*
2342                  * As async compaction considers a subset of pageblocks, only
2343                  * defer if the failure was a sync compaction failure.
2344                  */
2345                 if (mode != MIGRATE_ASYNC)
2346                         defer_compaction(preferred_zone, order);
2347 
2348                 cond_resched();
2349         }
2350 
2351         return NULL;
2352 }
2353 #else
2354 static inline struct page *
2355 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2356         struct zonelist *zonelist, enum zone_type high_zoneidx,
2357         nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2358         int classzone_idx, int migratetype,
2359         enum migrate_mode mode, bool *contended_compaction,
2360         bool *deferred_compaction, unsigned long *did_some_progress)
2361 {
2362         return NULL;
2363 }
2364 #endif /* CONFIG_COMPACTION */
2365 
2366 /* Perform direct synchronous page reclaim */
2367 static int
2368 __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2369                   nodemask_t *nodemask)
2370 {
2371         struct reclaim_state reclaim_state;
2372         int progress;
2373 
2374         cond_resched();
2375 
2376         /* We now go into synchronous reclaim */
2377         cpuset_memory_pressure_bump();
2378         current->flags |= PF_MEMALLOC;
2379         lockdep_set_current_reclaim_state(gfp_mask);
2380         reclaim_state.reclaimed_slab = 0;
2381         current->reclaim_state = &reclaim_state;
2382 
2383         progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
2384 
2385         current->reclaim_state = NULL;
2386         lockdep_clear_current_reclaim_state();
2387         current->flags &= ~PF_MEMALLOC;
2388 
2389         cond_resched();
2390 
2391         return progress;
2392 }
2393 
2394 /* The really slow allocator path where we enter direct reclaim */
2395 static inline struct page *
2396 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2397         struct zonelist *zonelist, enum zone_type high_zoneidx,
2398         nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2399         int classzone_idx, int migratetype, unsigned long *did_some_progress)
2400 {
2401         struct page *page = NULL;
2402         bool drained = false;
2403 
2404         *did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2405                                                nodemask);
2406         if (unlikely(!(*did_some_progress)))
2407                 return NULL;
2408 
2409         /* After successful reclaim, reconsider all zones for allocation */
2410         if (IS_ENABLED(CONFIG_NUMA))
2411                 zlc_clear_zones_full(zonelist);
2412 
2413 retry:
2414         page = get_page_from_freelist(gfp_mask, nodemask, order,
2415                                         zonelist, high_zoneidx,
2416                                         alloc_flags & ~ALLOC_NO_WATERMARKS,
2417                                         preferred_zone, classzone_idx,
2418                                         migratetype);
2419 
2420         /*
2421          * If an allocation failed after direct reclaim, it could be because
2422          * pages are pinned on the per-cpu lists. Drain them and try again
2423          */
2424         if (!page && !drained) {
2425                 drain_all_pages();
2426                 drained = true;
2427                 goto retry;
2428         }
2429 
2430         return page;
2431 }
2432 
2433 /*
2434  * This is called in the allocator slow-path if the allocation request is of
2435  * sufficient urgency to ignore watermarks and take other desperate measures
2436  */
2437 static inline struct page *
2438 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2439         struct zonelist *zonelist, enum zone_type high_zoneidx,
2440         nodemask_t *nodemask, struct zone *preferred_zone,
2441         int classzone_idx, int migratetype)
2442 {
2443         struct page *page;
2444 
2445         do {
2446                 page = get_page_from_freelist(gfp_mask, nodemask, order,
2447                         zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
2448                         preferred_zone, classzone_idx, migratetype);
2449 
2450                 if (!page && gfp_mask & __GFP_NOFAIL)
2451                         wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2452         } while (!page && (gfp_mask & __GFP_NOFAIL));
2453 
2454         return page;
2455 }
2456 
2457 static void wake_all_kswapds(unsigned int order,
2458                              struct zonelist *zonelist,
2459                              enum zone_type high_zoneidx,
2460                              struct zone *preferred_zone)
2461 {
2462         struct zoneref *z;
2463         struct zone *zone;
2464 
2465         for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
2466                 wakeup_kswapd(zone, order, zone_idx(preferred_zone));
2467 }
2468 
2469 static inline int
2470 gfp_to_alloc_flags(gfp_t gfp_mask)
2471 {
2472         int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2473         const bool atomic = !(gfp_mask & (__GFP_WAIT | __GFP_NO_KSWAPD));
2474 
2475         /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2476         BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2477 
2478         /*
2479          * The caller may dip into page reserves a bit more if the caller
2480          * cannot run direct reclaim, or if the caller has realtime scheduling
2481          * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
2482          * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
2483          */
2484         alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2485 
2486         if (atomic) {
2487                 /*
2488                  * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2489                  * if it can't schedule.
2490                  */
2491                 if (!(gfp_mask & __GFP_NOMEMALLOC))
2492                         alloc_flags |= ALLOC_HARDER;
2493                 /*
2494                  * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2495                  * comment for __cpuset_node_allowed_softwall().
2496                  */
2497                 alloc_flags &= ~ALLOC_CPUSET;
2498         } else if (unlikely(rt_task(current)) && !in_interrupt())
2499                 alloc_flags |= ALLOC_HARDER;
2500 
2501         if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2502                 if (gfp_mask & __GFP_MEMALLOC)
2503                         alloc_flags |= ALLOC_NO_WATERMARKS;
2504                 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2505                         alloc_flags |= ALLOC_NO_WATERMARKS;
2506                 else if (!in_interrupt() &&
2507                                 ((current->flags & PF_MEMALLOC) ||
2508                                  unlikely(test_thread_flag(TIF_MEMDIE))))
2509                         alloc_flags |= ALLOC_NO_WATERMARKS;
2510         }
2511 #ifdef CONFIG_CMA
2512         if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2513                 alloc_flags |= ALLOC_CMA;
2514 #endif
2515         return alloc_flags;
2516 }
2517 
2518 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2519 {
2520         return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
2521 }
2522 
2523 static inline struct page *
2524 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2525         struct zonelist *zonelist, enum zone_type high_zoneidx,
2526         nodemask_t *nodemask, struct zone *preferred_zone,
2527         int classzone_idx, int migratetype)
2528 {
2529         const gfp_t wait = gfp_mask & __GFP_WAIT;
2530         struct page *page = NULL;
2531         int alloc_flags;
2532         unsigned long pages_reclaimed = 0;
2533         unsigned long did_some_progress;
2534         enum migrate_mode migration_mode = MIGRATE_ASYNC;
2535         bool deferred_compaction = false;
2536         bool contended_compaction = false;
2537 
2538         /*
2539          * In the slowpath, we sanity check order to avoid ever trying to
2540          * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2541          * be using allocators in order of preference for an area that is
2542          * too large.
2543          */
2544         if (order >= MAX_ORDER) {
2545                 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2546                 return NULL;
2547         }
2548 
2549         /*
2550          * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2551          * __GFP_NOWARN set) should not cause reclaim since the subsystem
2552          * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2553          * using a larger set of nodes after it has established that the
2554          * allowed per node queues are empty and that nodes are
2555          * over allocated.
2556          */
2557         if (IS_ENABLED(CONFIG_NUMA) &&
2558             (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
2559                 goto nopage;
2560 
2561 restart:
2562         if (!(gfp_mask & __GFP_NO_KSWAPD))
2563                 wake_all_kswapds(order, zonelist, high_zoneidx, preferred_zone);
2564 
2565         /*
2566          * OK, we're below the kswapd watermark and have kicked background
2567          * reclaim. Now things get more complex, so set up alloc_flags according
2568          * to how we want to proceed.
2569          */
2570         alloc_flags = gfp_to_alloc_flags(gfp_mask);
2571 
2572         /*
2573          * Find the true preferred zone if the allocation is unconstrained by
2574          * cpusets.
2575          */
2576         if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) {
2577                 struct zoneref *preferred_zoneref;
2578                 preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx,
2579                                 NULL, &preferred_zone);
2580                 classzone_idx = zonelist_zone_idx(preferred_zoneref);
2581         }
2582 
2583 rebalance:
2584         /* This is the last chance, in general, before the goto nopage. */
2585         page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2586                         high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2587                         preferred_zone, classzone_idx, migratetype);
2588         if (page)
2589                 goto got_pg;
2590 
2591         /* Allocate without watermarks if the context allows */
2592         if (alloc_flags & ALLOC_NO_WATERMARKS) {
2593                 /*
2594                  * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2595                  * the allocation is high priority and these type of
2596                  * allocations are system rather than user orientated
2597                  */
2598                 zonelist = node_zonelist(numa_node_id(), gfp_mask);
2599 
2600                 page = __alloc_pages_high_priority(gfp_mask, order,
2601                                 zonelist, high_zoneidx, nodemask,
2602                                 preferred_zone, classzone_idx, migratetype);
2603                 if (page) {
2604                         goto got_pg;
2605                 }
2606         }
2607 
2608         /* Atomic allocations - we can't balance anything */
2609         if (!wait) {
2610                 /*
2611                  * All existing users of the deprecated __GFP_NOFAIL are
2612                  * blockable, so warn of any new users that actually allow this
2613                  * type of allocation to fail.
2614                  */
2615                 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
2616                 goto nopage;
2617         }
2618 
2619         /* Avoid recursion of direct reclaim */
2620         if (current->flags & PF_MEMALLOC)
2621                 goto nopage;
2622 
2623         /* Avoid allocations with no watermarks from looping endlessly */
2624         if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2625                 goto nopage;
2626 
2627         /*
2628          * Try direct compaction. The first pass is asynchronous. Subsequent
2629          * attempts after direct reclaim are synchronous
2630          */
2631         page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2632                                         high_zoneidx, nodemask, alloc_flags,
2633                                         preferred_zone,
2634                                         classzone_idx, migratetype,
2635                                         migration_mode, &contended_compaction,
2636                                         &deferred_compaction,
2637                                         &did_some_progress);
2638         if (page)
2639                 goto got_pg;
2640 
2641         /*
2642          * If compaction is deferred for high-order allocations, it is because
2643          * sync compaction recently failed. In this is the case and the caller
2644          * requested a movable allocation that does not heavily disrupt the
2645          * system then fail the allocation instead of entering direct reclaim.
2646          */
2647         if ((deferred_compaction || contended_compaction) &&
2648                                                 (gfp_mask & __GFP_NO_KSWAPD))
2649                 goto nopage;
2650 
2651         /*
2652          * It can become very expensive to allocate transparent hugepages at
2653          * fault, so use asynchronous memory compaction for THP unless it is
2654          * khugepaged trying to collapse.
2655          */
2656         if ((gfp_mask & GFP_TRANSHUGE) != GFP_TRANSHUGE ||
2657                                                 (current->flags & PF_KTHREAD))
2658                 migration_mode = MIGRATE_SYNC_LIGHT;
2659 
2660         /* Try direct reclaim and then allocating */
2661         page = __alloc_pages_direct_reclaim(gfp_mask, order,
2662                                         zonelist, high_zoneidx,
2663                                         nodemask,
2664                                         alloc_flags, preferred_zone,
2665                                         classzone_idx, migratetype,
2666                                         &did_some_progress);
2667         if (page)
2668                 goto got_pg;
2669 
2670         /*
2671          * If we failed to make any progress reclaiming, then we are
2672          * running out of options and have to consider going OOM
2673          */
2674         if (!did_some_progress) {
2675                 if (oom_gfp_allowed(gfp_mask)) {
2676                         if (oom_killer_disabled)
2677                                 goto nopage;
2678                         /* Coredumps can quickly deplete all memory reserves */
2679                         if ((current->flags & PF_DUMPCORE) &&
2680                             !(gfp_mask & __GFP_NOFAIL))
2681                                 goto nopage;
2682                         page = __alloc_pages_may_oom(gfp_mask, order,
2683                                         zonelist, high_zoneidx,
2684                                         nodemask, preferred_zone,
2685                                         classzone_idx, migratetype);
2686                         if (page)
2687                                 goto got_pg;
2688 
2689                         if (!(gfp_mask & __GFP_NOFAIL)) {
2690                                 /*
2691                                  * The oom killer is not called for high-order
2692                                  * allocations that may fail, so if no progress
2693                                  * is being made, there are no other options and
2694                                  * retrying is unlikely to help.
2695                                  */
2696                                 if (order > PAGE_ALLOC_COSTLY_ORDER)
2697                                         goto nopage;
2698                                 /*
2699                                  * The oom killer is not called for lowmem
2700                                  * allocations to prevent needlessly killing
2701                                  * innocent tasks.
2702                                  */
2703                                 if (high_zoneidx < ZONE_NORMAL)
2704                                         goto nopage;
2705                         }
2706 
2707                         goto restart;
2708                 }
2709         }
2710 
2711         /* Check if we should retry the allocation */
2712         pages_reclaimed += did_some_progress;
2713         if (should_alloc_retry(gfp_mask, order, did_some_progress,
2714                                                 pages_reclaimed)) {
2715                 /* Wait for some write requests to complete then retry */
2716                 wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2717                 goto rebalance;
2718         } else {
2719                 /*
2720                  * High-order allocations do not necessarily loop after
2721                  * direct reclaim and reclaim/compaction depends on compaction
2722                  * being called after reclaim so call directly if necessary
2723                  */
2724                 page = __alloc_pages_direct_compact(gfp_mask, order, zonelist,
2725                                         high_zoneidx, nodemask, alloc_flags,
2726                                         preferred_zone,
2727                                         classzone_idx, migratetype,
2728                                         migration_mode, &contended_compaction,
2729                                         &deferred_compaction,
2730                                         &did_some_progress);
2731                 if (page)
2732                         goto got_pg;
2733         }
2734 
2735 nopage:
2736         warn_alloc_failed(gfp_mask, order, NULL);
2737         return page;
2738 got_pg:
2739         if (kmemcheck_enabled)
2740                 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2741 
2742         return page;
2743 }
2744 
2745 /*
2746  * This is the 'heart' of the zoned buddy allocator.
2747  */
2748 struct page *
2749 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2750                         struct zonelist *zonelist, nodemask_t *nodemask)
2751 {
2752         enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2753         struct zone *preferred_zone;
2754         struct zoneref *preferred_zoneref;
2755         struct page *page = NULL;
2756         int migratetype = allocflags_to_migratetype(gfp_mask);
2757         unsigned int cpuset_mems_cookie;
2758         int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
2759         int classzone_idx;
2760 
2761         gfp_mask &= gfp_allowed_mask;
2762 
2763         lockdep_trace_alloc(gfp_mask);
2764 
2765         might_sleep_if(gfp_mask & __GFP_WAIT);
2766 
2767         if (should_fail_alloc_page(gfp_mask, order))
2768                 return NULL;
2769 
2770         /*
2771          * Check the zones suitable for the gfp_mask contain at least one
2772          * valid zone. It's possible to have an empty zonelist as a result
2773          * of GFP_THISNODE and a memoryless node
2774          */
2775         if (unlikely(!zonelist->_zonerefs->zone))
2776                 return NULL;
2777 
2778 retry_cpuset:
2779         cpuset_mems_cookie = read_mems_allowed_begin();
2780 
2781         /* The preferred zone is used for statistics later */
2782         preferred_zoneref = first_zones_zonelist(zonelist, high_zoneidx,
2783                                 nodemask ? : &cpuset_current_mems_allowed,
2784                                 &preferred_zone);
2785         if (!preferred_zone)
2786                 goto out;
2787         classzone_idx = zonelist_zone_idx(preferred_zoneref);
2788 
2789 #ifdef CONFIG_CMA
2790         if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2791                 alloc_flags |= ALLOC_CMA;
2792 #endif
2793         /* First allocation attempt */
2794         page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2795                         zonelist, high_zoneidx, alloc_flags,
2796                         preferred_zone, classzone_idx, migratetype);
2797         if (unlikely(!page)) {
2798                 /*
2799                  * Runtime PM, block IO and its error handling path
2800                  * can deadlock because I/O on the device might not
2801                  * complete.
2802                  */
2803                 gfp_mask = memalloc_noio_flags(gfp_mask);
2804                 page = __alloc_pages_slowpath(gfp_mask, order,
2805                                 zonelist, high_zoneidx, nodemask,
2806                                 preferred_zone, classzone_idx, migratetype);
2807         }
2808 
2809         trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2810 
2811 out:
2812         /*
2813          * When updating a task's mems_allowed, it is possible to race with
2814          * parallel threads in such a way that an allocation can fail while
2815          * the mask is being updated. If a page allocation is about to fail,
2816          * check if the cpuset changed during allocation and if so, retry.
2817          */
2818         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
2819                 goto retry_cpuset;
2820 
2821         return page;
2822 }
2823 EXPORT_SYMBOL(__alloc_pages_nodemask);
2824 
2825 /*
2826  * Common helper functions.
2827  */
2828 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2829 {
2830         struct page *page;
2831 
2832         /*
2833          * __get_free_pages() returns a 32-bit address, which cannot represent
2834          * a highmem page
2835          */
2836         VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2837 
2838         page = alloc_pages(gfp_mask, order);
2839         if (!page)
2840                 return 0;
2841         return (unsigned long) page_address(page);
2842 }
2843 EXPORT_SYMBOL(__get_free_pages);
2844 
2845 unsigned long get_zeroed_page(gfp_t gfp_mask)
2846 {
2847         return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2848 }
2849 EXPORT_SYMBOL(get_zeroed_page);
2850 
2851 void __free_pages(struct page *page, unsigned int order)
2852 {
2853         if (put_page_testzero(page)) {
2854                 if (order == 0)
2855                         free_hot_cold_page(page, false);
2856                 else
2857                         __free_pages_ok(page, order);
2858         }
2859 }
2860 
2861 EXPORT_SYMBOL(__free_pages);
2862 
2863 void free_pages(unsigned long addr, unsigned int order)
2864 {
2865         if (addr != 0) {
2866                 VM_BUG_ON(!virt_addr_valid((void *)addr));
2867                 __free_pages(virt_to_page((void *)addr), order);
2868         }
2869 }
2870 
2871 EXPORT_SYMBOL(free_pages);
2872 
2873 /*
2874  * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
2875  * of the current memory cgroup.
2876  *
2877  * It should be used when the caller would like to use kmalloc, but since the
2878  * allocation is large, it has to fall back to the page allocator.
2879  */
2880 struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
2881 {
2882         struct page *page;
2883         struct mem_cgroup *memcg = NULL;
2884 
2885         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2886                 return NULL;
2887         page = alloc_pages(gfp_mask, order);
2888         memcg_kmem_commit_charge(page, memcg, order);
2889         return page;
2890 }
2891 
2892 struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
2893 {
2894         struct page *page;
2895         struct mem_cgroup *memcg = NULL;
2896 
2897         if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2898                 return NULL;
2899         page = alloc_pages_node(nid, gfp_mask, order);
2900         memcg_kmem_commit_charge(page, memcg, order);
2901         return page;
2902 }
2903 
2904 /*
2905  * __free_kmem_pages and free_kmem_pages will free pages allocated with
2906  * alloc_kmem_pages.
2907  */
2908 void __free_kmem_pages(struct page *page, unsigned int order)
2909 {
2910         memcg_kmem_uncharge_pages(page, order);
2911         __free_pages(page, order);
2912 }
2913 
2914 void free_kmem_pages(unsigned long addr, unsigned int order)
2915 {
2916         if (addr != 0) {
2917                 VM_BUG_ON(!virt_addr_valid((void *)addr));
2918                 __free_kmem_pages(virt_to_page((void *)addr), order);
2919         }
2920 }
2921 
2922 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2923 {
2924         if (addr) {
2925                 unsigned long alloc_end = addr + (PAGE_SIZE << order);
2926                 unsigned long used = addr + PAGE_ALIGN(size);
2927 
2928                 split_page(virt_to_page((void *)addr), order);
2929                 while (used < alloc_end) {
2930                         free_page(used);
2931                         used += PAGE_SIZE;
2932                 }
2933         }
2934         return (void *)addr;
2935 }
2936 
2937 /**
2938  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2939  * @size: the number of bytes to allocate
2940  * @gfp_mask: GFP flags for the allocation
2941  *
2942  * This function is similar to alloc_pages(), except that it allocates the
2943  * minimum number of pages to satisfy the request.  alloc_pages() can only
2944  * allocate memory in power-of-two pages.
2945  *
2946  * This function is also limited by MAX_ORDER.
2947  *
2948  * Memory allocated by this function must be released by free_pages_exact().
2949  */
2950 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2951 {
2952         unsigned int order = get_order(size);
2953         unsigned long addr;
2954 
2955         addr = __get_free_pages(gfp_mask, order);
2956         return make_alloc_exact(addr, order, size);
2957 }
2958 EXPORT_SYMBOL(alloc_pages_exact);
2959 
2960 /**
2961  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2962  *                         pages on a node.
2963  * @nid: the preferred node ID where memory should be allocated
2964  * @size: the number of bytes to allocate
2965  * @gfp_mask: GFP flags for the allocation
2966  *
2967  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2968  * back.
2969  * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2970  * but is not exact.
2971  */
2972 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2973 {
2974         unsigned order = get_order(size);
2975         struct page *p = alloc_pages_node(nid, gfp_mask, order);
2976         if (!p)
2977                 return NULL;
2978         return make_alloc_exact((unsigned long)page_address(p), order, size);
2979 }
2980 
2981 /**
2982  * free_pages_exact - release memory allocated via alloc_pages_exact()
2983  * @virt: the value returned by alloc_pages_exact.
2984  * @size: size of allocation, same value as passed to alloc_pages_exact().
2985  *
2986  * Release the memory allocated by a previous call to alloc_pages_exact.
2987  */
2988 void free_pages_exact(void *virt, size_t size)
2989 {
2990         unsigned long addr = (unsigned long)virt;
2991         unsigned long end = addr + PAGE_ALIGN(size);
2992 
2993         while (addr < end) {
2994                 free_page(addr);
2995                 addr += PAGE_SIZE;
2996         }
2997 }
2998 EXPORT_SYMBOL(free_pages_exact);
2999 
3000 /**
3001  * nr_free_zone_pages - count number of pages beyond high watermark
3002  * @offset: The zone index of the highest zone
3003  *
3004  * nr_free_zone_pages() counts the number of counts pages which are beyond the
3005  * high watermark within all zones at or below a given zone index.  For each
3006  * zone, the number of pages is calculated as:
3007  *     managed_pages - high_pages
3008  */
3009 static unsigned long nr_free_zone_pages(int offset)
3010 {
3011         struct zoneref *z;
3012         struct zone *zone;
3013 
3014         /* Just pick one node, since fallback list is circular */
3015         unsigned long sum = 0;
3016 
3017         struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3018 
3019         for_each_zone_zonelist(zone, z, zonelist, offset) {
3020                 unsigned long size = zone->managed_pages;
3021                 unsigned long high = high_wmark_pages(zone);
3022                 if (size > high)
3023                         sum += size - high;
3024         }
3025 
3026         return sum;
3027 }
3028 
3029 /**
3030  * nr_free_buffer_pages - count number of pages beyond high watermark
3031  *
3032  * nr_free_buffer_pages() counts the number of pages which are beyond the high
3033  * watermark within ZONE_DMA and ZONE_NORMAL.
3034  */
3035 unsigned long nr_free_buffer_pages(void)
3036 {
3037         return nr_free_zone_pages(gfp_zone(GFP_USER));
3038 }
3039 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3040 
3041 /**
3042  * nr_free_pagecache_pages - count number of pages beyond high watermark
3043  *
3044  * nr_free_pagecache_pages() counts the number of pages which are beyond the
3045  * high watermark within all zones.
3046  */
3047 unsigned long nr_free_pagecache_pages(void)
3048 {
3049         return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3050 }
3051 
3052 static inline void show_node(struct zone *zone)
3053 {
3054         if (IS_ENABLED(CONFIG_NUMA))
3055                 printk("Node %d ", zone_to_nid(zone));
3056 }
3057 
3058 void si_meminfo(struct sysinfo *val)
3059 {
3060         val->totalram = totalram_pages;
3061         val->sharedram = global_page_state(NR_SHMEM);
3062         val->freeram = global_page_state(NR_FREE_PAGES);
3063         val->bufferram = nr_blockdev_pages();
3064         val->totalhigh = totalhigh_pages;
3065         val->freehigh = nr_free_highpages();
3066         val->mem_unit = PAGE_SIZE;
3067 }
3068 
3069 EXPORT_SYMBOL(si_meminfo);
3070 
3071 #ifdef CONFIG_NUMA
3072 void si_meminfo_node(struct sysinfo *val, int nid)
3073 {
3074         int zone_type;          /* needs to be signed */
3075         unsigned long managed_pages = 0;
3076         pg_data_t *pgdat = NODE_DATA(nid);
3077 
3078         for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3079                 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3080         val->totalram = managed_pages;
3081         val->sharedram = node_page_state(nid, NR_SHMEM);
3082         val->freeram = node_page_state(nid, NR_FREE_PAGES);
3083 #ifdef CONFIG_HIGHMEM
3084         val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
3085         val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3086                         NR_FREE_PAGES);
3087 #else
3088         val->totalhigh = 0;
3089         val->freehigh = 0;
3090 #endif
3091         val->mem_unit = PAGE_SIZE;
3092 }
3093 #endif
3094 
3095 /*
3096  * Determine whether the node should be displayed or not, depending on whether
3097  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3098  */
3099 bool skip_free_areas_node(unsigned int flags, int nid)
3100 {
3101         bool ret = false;
3102         unsigned int cpuset_mems_cookie;
3103 
3104         if (!(flags & SHOW_MEM_FILTER_NODES))
3105                 goto out;
3106 
3107         do {
3108                 cpuset_mems_cookie = read_mems_allowed_begin();
3109                 ret = !node_isset(nid, cpuset_current_mems_allowed);
3110         } while (read_mems_allowed_retry(cpuset_mems_cookie));
3111 out:
3112         return ret;
3113 }
3114 
3115 #define K(x) ((x) << (PAGE_SHIFT-10))
3116 
3117 static void show_migration_types(unsigned char type)
3118 {
3119         static const char types[MIGRATE_TYPES] = {
3120                 [MIGRATE_UNMOVABLE]     = 'U',
3121                 [MIGRATE_RECLAIMABLE]   = 'E',
3122                 [MIGRATE_MOVABLE]       = 'M',
3123                 [MIGRATE_RESERVE]       = 'R',
3124 #ifdef CONFIG_CMA
3125                 [MIGRATE_CMA]           = 'C',
3126 #endif
3127 #ifdef CONFIG_MEMORY_ISOLATION
3128                 [MIGRATE_ISOLATE]       = 'I',
3129 #endif
3130         };
3131         char tmp[MIGRATE_TYPES + 1];
3132         char *p = tmp;
3133         int i;
3134 
3135         for (i = 0; i < MIGRATE_TYPES; i++) {
3136                 if (type & (1 << i))
3137                         *p++ = types[i];
3138         }
3139 
3140         *p = '\0';
3141         printk("(%s) ", tmp);
3142 }
3143 
3144 /*
3145  * Show free area list (used inside shift_scroll-lock stuff)
3146  * We also calculate the percentage fragmentation. We do this by counting the
3147  * memory on each free list with the exception of the first item on the list.
3148  * Suppresses nodes that are not allowed by current's cpuset if
3149  * SHOW_MEM_FILTER_NODES is passed.
3150  */
3151 void show_free_areas(unsigned int filter)
3152 {
3153         int cpu;
3154         struct zone *zone;
3155 
3156         for_each_populated_zone(zone) {
3157                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3158                         continue;
3159                 show_node(zone);
3160                 printk("%s per-cpu:\n", zone->name);
3161 
3162                 for_each_online_cpu(cpu) {
3163                         struct per_cpu_pageset *pageset;
3164 
3165                         pageset = per_cpu_ptr(zone->pageset, cpu);
3166 
3167                         printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3168                                cpu, pageset->pcp.high,
3169                                pageset->pcp.batch, pageset->pcp.count);
3170                 }
3171         }
3172 
3173         printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3174                 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3175                 " unevictable:%lu"
3176                 " dirty:%lu writeback:%lu unstable:%lu\n"
3177                 " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3178                 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3179                 " free_cma:%lu\n",
3180                 global_page_state(NR_ACTIVE_ANON),
3181                 global_page_state(NR_INACTIVE_ANON),
3182                 global_page_state(NR_ISOLATED_ANON),
3183                 global_page_state(NR_ACTIVE_FILE),
3184                 global_page_state(NR_INACTIVE_FILE),
3185                 global_page_state(NR_ISOLATED_FILE),
3186                 global_page_state(NR_UNEVICTABLE),
3187                 global_page_state(NR_FILE_DIRTY),
3188                 global_page_state(NR_WRITEBACK),
3189                 global_page_state(NR_UNSTABLE_NFS),
3190                 global_page_state(NR_FREE_PAGES),
3191                 global_page_state(NR_SLAB_RECLAIMABLE),
3192                 global_page_state(NR_SLAB_UNRECLAIMABLE),
3193                 global_page_state(NR_FILE_MAPPED),
3194                 global_page_state(NR_SHMEM),
3195                 global_page_state(NR_PAGETABLE),
3196                 global_page_state(NR_BOUNCE),
3197                 global_page_state(NR_FREE_CMA_PAGES));
3198 
3199         for_each_populated_zone(zone) {
3200                 int i;
3201 
3202                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3203                         continue;
3204                 show_node(zone);
3205                 printk("%s"
3206                         " free:%lukB"
3207                         " min:%lukB"
3208                         " low:%lukB"
3209                         " high:%lukB"
3210                         " active_anon:%lukB"
3211                         " inactive_anon:%lukB"
3212                         " active_file:%lukB"
3213                         " inactive_file:%lukB"
3214                         " unevictable:%lukB"
3215                         " isolated(anon):%lukB"
3216                         " isolated(file):%lukB"
3217                         " present:%lukB"
3218                         " managed:%lukB"
3219                         " mlocked:%lukB"
3220                         " dirty:%lukB"
3221                         " writeback:%lukB"
3222                         " mapped:%lukB"
3223                         " shmem:%lukB"
3224                         " slab_reclaimable:%lukB"
3225                         " slab_unreclaimable:%lukB"
3226                         " kernel_stack:%lukB"
3227                         " pagetables:%lukB"
3228                         " unstable:%lukB"
3229                         " bounce:%lukB"
3230                         " free_cma:%lukB"
3231                         " writeback_tmp:%lukB"
3232                         " pages_scanned:%lu"
3233                         " all_unreclaimable? %s"
3234                         "\n",
3235                         zone->name,
3236                         K(zone_page_state(zone, NR_FREE_PAGES)),
3237                         K(min_wmark_pages(zone)),
3238                         K(low_wmark_pages(zone)),
3239                         K(high_wmark_pages(zone)),
3240                         K(zone_page_state(zone, NR_ACTIVE_ANON)),
3241                         K(zone_page_state(zone, NR_INACTIVE_ANON)),
3242                         K(zone_page_state(zone, NR_ACTIVE_FILE)),
3243                         K(zone_page_state(zone, NR_INACTIVE_FILE)),
3244                         K(zone_page_state(zone, NR_UNEVICTABLE)),
3245                         K(zone_page_state(zone, NR_ISOLATED_ANON)),
3246                         K(zone_page_state(zone, NR_ISOLATED_FILE)),
3247                         K(zone->present_pages),
3248                         K(zone->managed_pages),
3249                         K(zone_page_state(zone, NR_MLOCK)),
3250                         K(zone_page_state(zone, NR_FILE_DIRTY)),
3251                         K(zone_page_state(zone, NR_WRITEBACK)),
3252                         K(zone_page_state(zone, NR_FILE_MAPPED)),
3253                         K(zone_page_state(zone, NR_SHMEM)),
3254                         K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3255                         K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3256                         zone_page_state(zone, NR_KERNEL_STACK) *
3257                                 THREAD_SIZE / 1024,
3258                         K(zone_page_state(zone, NR_PAGETABLE)),
3259                         K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3260                         K(zone_page_state(zone, NR_BOUNCE)),
3261                         K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3262                         K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3263                         K(zone_page_state(zone, NR_PAGES_SCANNED)),
3264                         (!zone_reclaimable(zone) ? "yes" : "no")
3265                         );
3266                 printk("lowmem_reserve[]:");
3267                 for (i = 0; i < MAX_NR_ZONES; i++)
3268                         printk(" %ld", zone->lowmem_reserve[i]);
3269                 printk("\n");
3270         }
3271 
3272         for_each_populated_zone(zone) {
3273                 unsigned long nr[MAX_ORDER], flags, order, total = 0;
3274                 unsigned char types[MAX_ORDER];
3275 
3276                 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3277                         continue;
3278                 show_node(zone);
3279                 printk("%s: ", zone->name);
3280 
3281                 spin_lock_irqsave(&zone->lock, flags);
3282                 for (order = 0; order < MAX_ORDER; order++) {
3283                         struct free_area *area = &zone->free_area[order];
3284                         int type;
3285 
3286                         nr[order] = area->nr_free;
3287                         total += nr[order] << order;
3288 
3289                         types[order] = 0;
3290                         for (type = 0; type < MIGRATE_TYPES; type++) {
3291                                 if (!list_empty(&area->free_list[type]))
3292                                         types[order] |= 1 << type;
3293                         }
3294                 }
3295                 spin_unlock_irqrestore(&zone->lock, flags);
3296                 for (order = 0; order < MAX_ORDER; order++) {
3297                         printk("%lu*%lukB ", nr[order], K(1UL) << order);
3298                         if (nr[order])
3299                                 show_migration_types(types[order]);
3300                 }
3301                 printk("= %lukB\n", K(total));
3302         }
3303 
3304         hugetlb_show_meminfo();
3305 
3306         printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3307 
3308         show_swap_cache_info();
3309 }
3310 
3311 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3312 {
3313         zoneref->zone = zone;
3314         zoneref->zone_idx = zone_idx(zone);
3315 }
3316 
3317 /*
3318  * Builds allocation fallback zone lists.
3319  *
3320  * Add all populated zones of a node to the zonelist.
3321  */
3322 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3323                                 int nr_zones)
3324 {
3325         struct zone *zone;
3326         enum zone_type zone_type = MAX_NR_ZONES;
3327 
3328         do {
3329                 zone_type--;
3330                 zone = pgdat->node_zones + zone_type;
3331                 if (populated_zone(zone)) {
3332                         zoneref_set_zone(zone,
3333                                 &zonelist->_zonerefs[nr_zones++]);
3334                         check_highest_zone(zone_type);
3335                 }
3336         } while (zone_type);
3337 
3338         return nr_zones;
3339 }
3340 
3341 
3342 /*
3343  *  zonelist_order:
3344  *  0 = automatic detection of better ordering.
3345  *  1 = order by ([node] distance, -zonetype)
3346  *  2 = order by (-zonetype, [node] distance)
3347  *
3348  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3349  *  the same zonelist. So only NUMA can configure this param.
3350  */
3351 #define ZONELIST_ORDER_DEFAULT  0
3352 #define ZONELIST_ORDER_NODE     1
3353 #define ZONELIST_ORDER_ZONE     2
3354 
3355 /* zonelist order in the kernel.
3356  * set_zonelist_order() will set this to NODE or ZONE.
3357  */
3358 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3359 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3360 
3361 
3362 #ifdef CONFIG_NUMA
3363 /* The value user specified ....changed by config */
3364 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3365 /* string for sysctl */
3366 #define NUMA_ZONELIST_ORDER_LEN 16
3367 char numa_zonelist_order[16] = "default";
3368 
3369 /*
3370  * interface for configure zonelist ordering.
3371  * command line option "numa_zonelist_order"
3372  *      = "[dD]efault   - default, automatic configuration.
3373  *      = "[nN]ode      - order by node locality, then by zone within node
3374  *      = "[zZ]one      - order by zone, then by locality within zone
3375  */
3376 
3377 static int __parse_numa_zonelist_order(char *s)
3378 {
3379         if (*s == 'd' || *s == 'D') {
3380                 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3381         } else if (*s == 'n' || *s == 'N') {
3382                 user_zonelist_order = ZONELIST_ORDER_NODE;
3383         } else if (*s == 'z' || *s == 'Z') {
3384                 user_zonelist_order = ZONELIST_ORDER_ZONE;
3385         } else {
3386                 printk(KERN_WARNING
3387                         "Ignoring invalid numa_zonelist_order value:  "
3388                         "%s\n", s);
3389                 return -EINVAL;
3390         }
3391         return 0;
3392 }
3393 
3394 static __init int setup_numa_zonelist_order(char *s)
3395 {
3396         int ret;
3397 
3398         if (!s)
3399                 return 0;
3400 
3401         ret = __parse_numa_zonelist_order(s);
3402         if (ret == 0)
3403                 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3404 
3405         return ret;
3406 }
3407 early_param("numa_zonelist_order", setup_numa_zonelist_order);
3408 
3409 /*
3410  * sysctl handler for numa_zonelist_order
3411  */
3412 int numa_zonelist_order_handler(struct ctl_table *table, int write,
3413                 void __user *buffer, size_t *length,
3414                 loff_t *ppos)
3415 {
3416         char saved_string[NUMA_ZONELIST_ORDER_LEN];
3417         int ret;
3418         static DEFINE_MUTEX(zl_order_mutex);
3419 
3420         mutex_lock(&zl_order_mutex);
3421         if (write) {
3422                 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3423                         ret = -EINVAL;
3424                         goto out;
3425                 }
3426                 strcpy(saved_string, (char *)table->data);
3427         }
3428         ret = proc_dostring(table, write, buffer, length, ppos);
3429         if (ret)
3430                 goto out;
3431         if (write) {
3432                 int oldval = user_zonelist_order;
3433 
3434                 ret = __parse_numa_zonelist_order((char *)table->data);
3435                 if (ret) {
3436                         /*
3437                          * bogus value.  restore saved string
3438                          */
3439                         strncpy((char *)table->data, saved_string,
3440                                 NUMA_ZONELIST_ORDER_LEN);
3441                         user_zonelist_order = oldval;
3442                 } else if (oldval != user_zonelist_order) {
3443                         mutex_lock(&zonelists_mutex);
3444                         build_all_zonelists(NULL, NULL);
3445                         mutex_unlock(&zonelists_mutex);
3446                 }
3447         }
3448 out:
3449         mutex_unlock(&zl_order_mutex);
3450         return ret;
3451 }
3452 
3453 
3454 #define MAX_NODE_LOAD (nr_online_nodes)
3455 static int node_load[MAX_NUMNODES];
3456 
3457 /**
3458  * find_next_best_node - find the next node that should appear in a given node's fallback list
3459  * @node: node whose fallback list we're appending
3460  * @used_node_mask: nodemask_t of already used nodes
3461  *
3462  * We use a number of factors to determine which is the next node that should
3463  * appear on a given node's fallback list.  The node should not have appeared
3464  * already in @node's fallback list, and it should be the next closest node
3465  * according to the distance array (which contains arbitrary distance values
3466  * from each node to each node in the system), and should also prefer nodes
3467  * with no CPUs, since presumably they'll have very little allocation pressure
3468  * on them otherwise.
3469  * It returns -1 if no node is found.
3470  */
3471 static int find_next_best_node(int node, nodemask_t *used_node_mask)
3472 {
3473         int n, val;
3474         int min_val = INT_MAX;
3475         int best_node = NUMA_NO_NODE;
3476         const struct cpumask *tmp = cpumask_of_node(0);
3477 
3478         /* Use the local node if we haven't already */
3479         if (!node_isset(node, *used_node_mask)) {
3480                 node_set(node, *used_node_mask);
3481                 return node;
3482         }
3483 
3484         for_each_node_state(n, N_MEMORY) {
3485 
3486                 /* Don't want a node to appear more than once */
3487                 if (node_isset(n, *used_node_mask))
3488                         continue;
3489 
3490                 /* Use the distance array to find the distance */
3491                 val = node_distance(node, n);
3492 
3493                 /* Penalize nodes under us ("prefer the next node") */
3494                 val += (n < node);
3495 
3496                 /* Give preference to headless and unused nodes */
3497                 tmp = cpumask_of_node(n);
3498                 if (!cpumask_empty(tmp))
3499                         val += PENALTY_FOR_NODE_WITH_CPUS;
3500 
3501                 /* Slight preference for less loaded node */
3502                 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3503                 val += node_load[n];
3504 
3505                 if (val < min_val) {
3506                         min_val = val;
3507                         best_node = n;
3508                 }
3509         }
3510 
3511         if (best_node >= 0)
3512                 node_set(best_node, *used_node_mask);
3513 
3514         return best_node;
3515 }
3516 
3517 
3518 /*
3519  * Build zonelists ordered by node and zones within node.
3520  * This results in maximum locality--normal zone overflows into local
3521  * DMA zone, if any--but risks exhausting DMA zone.
3522  */
3523 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
3524 {
3525         int j;
3526         struct zonelist *zonelist;
3527 
3528         zonelist = &pgdat->node_zonelists[0];
3529         for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
3530                 ;
3531         j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3532         zonelist->_zonerefs[j].zone = NULL;
3533         zonelist->_zonerefs[j].zone_idx = 0;
3534 }
3535 
3536 /*
3537  * Build gfp_thisnode zonelists
3538  */
3539 static void build_thisnode_zonelists(pg_data_t *pgdat)
3540 {
3541         int j;
3542         struct zonelist *zonelist;
3543 
3544         zonelist = &pgdat->node_zonelists[1];
3545         j = build_zonelists_node(pgdat, zonelist, 0);
3546         zonelist->_zonerefs[j].zone = NULL;
3547         zonelist->_zonerefs[j].zone_idx = 0;
3548 }
3549 
3550 /*
3551  * Build zonelists ordered by zone and nodes within zones.
3552  * This results in conserving DMA zone[s] until all Normal memory is
3553  * exhausted, but results in overflowing to remote node while memory
3554  * may still exist in local DMA zone.
3555  */
3556 static int node_order[MAX_NUMNODES];
3557 
3558 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3559 {
3560         int pos, j, node;
3561         int zone_type;          /* needs to be signed */
3562         struct zone *z;
3563         struct zonelist *zonelist;
3564 
3565         zonelist = &pgdat->node_zonelists[0];
3566         pos = 0;
3567         for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3568                 for (j = 0; j < nr_nodes; j++) {
3569                         node = node_order[j];
3570                         z = &NODE_DATA(node)->node_zones[zone_type];
3571                         if (populated_zone(z)) {
3572                                 zoneref_set_zone(z,
3573                                         &zonelist->_zonerefs[pos++]);
3574                                 check_highest_zone(zone_type);
3575                         }
3576                 }
3577         }
3578         zonelist->_zonerefs[pos].zone = NULL;
3579         zonelist->_zonerefs[pos].zone_idx = 0;
3580 }
3581 
3582 static int default_zonelist_order(void)
3583 {
3584         int nid, zone_type;
3585         unsigned long low_kmem_size, total_size;
3586         struct zone *z;
3587         int average_size;
3588         /*
3589          * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3590          * If they are really small and used heavily, the system can fall
3591          * into OOM very easily.
3592          * This function detect ZONE_DMA/DMA32 size and configures zone order.
3593          */
3594         /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3595         low_kmem_size = 0;
3596         total_size = 0;
3597         for_each_online_node(nid) {
3598                 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3599                         z = &NODE_DATA(nid)->node_zones[zone_type];
3600                         if (populated_zone(z)) {
3601                                 if (zone_type < ZONE_NORMAL)
3602                                         low_kmem_size += z->managed_pages;
3603                                 total_size += z->managed_pages;
3604                         } else if (zone_type == ZONE_NORMAL) {
3605                                 /*
3606                                  * If any node has only lowmem, then node order
3607                                  * is preferred to allow kernel allocations
3608                                  * locally; otherwise, they can easily infringe
3609                                  * on other nodes when there is an abundance of
3610                                  * lowmem available to allocate from.
3611                                  */
3612                                 return ZONELIST_ORDER_NODE;
3613                         }
3614                 }
3615         }
3616         if (!low_kmem_size ||  /* there are no DMA area. */
3617             low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3618                 return ZONELIST_ORDER_NODE;
3619         /*
3620          * look into each node's config.
3621          * If there is a node whose DMA/DMA32 memory is very big area on
3622          * local memory, NODE_ORDER may be suitable.
3623          */
3624         average_size = total_size /
3625                                 (nodes_weight(node_states[N_MEMORY]) + 1);
3626         for_each_online_node(nid) {
3627                 low_kmem_size = 0;
3628                 total_size = 0;
3629                 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3630                         z = &NODE_DATA(nid)->node_zones[zone_type];
3631                         if (populated_zone(z)) {
3632                                 if (zone_type < ZONE_NORMAL)
3633                                         low_kmem_size += z->present_pages;
3634                                 total_size += z->present_pages;
3635                         }
3636                 }
3637                 if (low_kmem_size &&
3638                     total_size > average_size && /* ignore small node */
3639                     low_kmem_size > total_size * 70/100)
3640                         return ZONELIST_ORDER_NODE;
3641         }
3642         return ZONELIST_ORDER_ZONE;
3643 }
3644 
3645 static void set_zonelist_order(void)
3646 {
3647         if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3648                 current_zonelist_order = default_zonelist_order();
3649         else
3650                 current_zonelist_order = user_zonelist_order;
3651 }
3652 
3653 static void build_zonelists(pg_data_t *pgdat)
3654 {
3655         int j, node, load;
3656         enum zone_type i;
3657         nodemask_t used_mask;
3658         int local_node, prev_node;
3659         struct zonelist *zonelist;
3660         int order = current_zonelist_order;
3661 
3662         /* initialize zonelists */
3663         for (i = 0; i < MAX_ZONELISTS; i++) {
3664                 zonelist = pgdat->node_zonelists + i;
3665                 zonelist->_zonerefs[0].zone = NULL;
3666                 zonelist->_zonerefs[0].zone_idx = 0;
3667         }
3668 
3669         /* NUMA-aware ordering of nodes */
3670         local_node = pgdat->node_id;
3671         load = nr_online_nodes;
3672         prev_node = local_node;
3673         nodes_clear(used_mask);
3674 
3675         memset(node_order, 0, sizeof(node_order));
3676         j = 0;
3677 
3678         while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3679                 /*
3680                  * We don't want to pressure a particular node.
3681                  * So adding penalty to the first node in same
3682                  * distance group to make it round-robin.
3683                  */
3684                 if (node_distance(local_node, node) !=
3685                     node_distance(local_node, prev_node))
3686                         node_load[node] = load;
3687 
3688                 prev_node = node;
3689                 load--;
3690                 if (order == ZONELIST_ORDER_NODE)
3691                         build_zonelists_in_node_order(pgdat, node);
3692                 else
3693                         node_order[j++] = node; /* remember order */
3694         }
3695 
3696         if (order == ZONELIST_ORDER_ZONE) {
3697                 /* calculate node order -- i.e., DMA last! */
3698                 build_zonelists_in_zone_order(pgdat, j);
3699         }
3700 
3701         build_thisnode_zonelists(pgdat);
3702 }
3703 
3704 /* Construct the zonelist performance cache - see further mmzone.h */
3705 static void build_zonelist_cache(pg_data_t *pgdat)
3706 {
3707         struct zonelist *zonelist;
3708         struct zonelist_cache *zlc;
3709         struct zoneref *z;
3710 
3711         zonelist = &pgdat->node_zonelists[0];
3712         zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3713         bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3714         for (z = zonelist->_zonerefs; z->zone; z++)
3715                 zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3716 }
3717 
3718 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3719 /*
3720  * Return node id of node used for "local" allocations.
3721  * I.e., first node id of first zone in arg node's generic zonelist.
3722  * Used for initializing percpu 'numa_mem', which is used primarily
3723  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3724  */
3725 int local_memory_node(int node)
3726 {
3727         struct zone *zone;
3728 
3729         (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3730                                    gfp_zone(GFP_KERNEL),
3731                                    NULL,
3732                                    &zone);
3733         return zone->node;
3734 }
3735 #endif
3736 
3737 #else   /* CONFIG_NUMA */
3738 
3739 static void set_zonelist_order(void)
3740 {
3741         current_zonelist_order = ZONELIST_ORDER_ZONE;
3742 }
3743 
3744 static void build_zonelists(pg_data_t *pgdat)
3745 {
3746         int node, local_node;
3747         enum zone_type j;
3748         struct zonelist *zonelist;
3749 
3750         local_node = pgdat->node_id;
3751 
3752         zonelist = &pgdat->node_zonelists[0];
3753         j = build_zonelists_node(pgdat, zonelist, 0);
3754 
3755         /*
3756          * Now we build the zonelist so that it contains the zones
3757          * of all the other nodes.
3758          * We don't want to pressure a particular node, so when
3759          * building the zones for node N, we make sure that the
3760          * zones coming right after the local ones are those from
3761          * node N+1 (modulo N)
3762          */
3763         for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3764                 if (!node_online(node))
3765                         continue;
3766                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3767         }
3768         for (node = 0; node < local_node; node++) {
3769                 if (!node_online(node))
3770                         continue;
3771                 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3772         }
3773 
3774         zonelist->_zonerefs[j].zone = NULL;
3775         zonelist->_zonerefs[j].zone_idx = 0;
3776 }
3777 
3778 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3779 static void build_zonelist_cache(pg_data_t *pgdat)
3780 {
3781         pgdat->node_zonelists[0].zlcache_ptr = NULL;
3782 }
3783 
3784 #endif  /* CONFIG_NUMA */
3785 
3786 /*
3787  * Boot pageset table. One per cpu which is going to be used for all
3788  * zones and all nodes. The parameters will be set in such a way
3789  * that an item put on a list will immediately be handed over to
3790  * the buddy list. This is safe since pageset manipulation is done
3791  * with interrupts disabled.
3792  *
3793  * The boot_pagesets must be kept even after bootup is complete for
3794  * unused processors and/or zones. They do play a role for bootstrapping
3795  * hotplugged processors.
3796  *
3797  * zoneinfo_show() and maybe other functions do
3798  * not check if the processor is online before following the pageset pointer.
3799  * Other parts of the kernel may not check if the zone is available.
3800  */
3801 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3802 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3803 static void setup_zone_pageset(struct zone *zone);
3804 
3805 /*
3806  * Global mutex to protect against size modification of zonelists
3807  * as well as to serialize pageset setup for the new populated zone.
3808  */
3809 DEFINE_MUTEX(zonelists_mutex);
3810 
3811 /* return values int ....just for stop_machine() */
3812 static int __build_all_zonelists(void *data)
3813 {
3814         int nid;
3815         int cpu;
3816         pg_data_t *self = data;
3817 
3818 #ifdef CONFIG_NUMA
3819         memset(node_load, 0, sizeof(node_load));
3820 #endif
3821 
3822         if (self && !node_online(self->node_id)) {
3823                 build_zonelists(self);
3824                 build_zonelist_cache(self);
3825         }
3826 
3827         for_each_online_node(nid) {
3828                 pg_data_t *pgdat = NODE_DATA(nid);
3829 
3830                 build_zonelists(pgdat);
3831                 build_zonelist_cache(pgdat);
3832         }
3833 
3834         /*
3835          * Initialize the boot_pagesets that are going to be used
3836          * for bootstrapping processors. The real pagesets for
3837          * each zone will be allocated later when the per cpu
3838          * allocator is available.
3839          *
3840          * boot_pagesets are used also for bootstrapping offline
3841          * cpus if the system is already booted because the pagesets
3842          * are needed to initialize allocators on a specific cpu too.
3843          * F.e. the percpu allocator needs the page allocator which
3844          * needs the percpu allocator in order to allocate its pagesets
3845          * (a chicken-egg dilemma).
3846          */
3847         for_each_possible_cpu(cpu) {
3848                 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3849 
3850 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3851                 /*
3852                  * We now know the "local memory node" for each node--
3853                  * i.e., the node of the first zone in the generic zonelist.
3854                  * Set up numa_mem percpu variable for on-line cpus.  During
3855                  * boot, only the boot cpu should be on-line;  we'll init the
3856                  * secondary cpus' numa_mem as they come on-line.  During
3857                  * node/memory hotplug, we'll fixup all on-line cpus.
3858                  */
3859                 if (cpu_online(cpu))
3860                         set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3861 #endif
3862         }
3863 
3864         return 0;
3865 }
3866 
3867 /*
3868  * Called with zonelists_mutex held always
3869  * unless system_state == SYSTEM_BOOTING.
3870  */
3871 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
3872 {
3873         set_zonelist_order();
3874 
3875         if (system_state == SYSTEM_BOOTING) {
3876                 __build_all_zonelists(NULL);
3877                 mminit_verify_zonelist();
3878                 cpuset_init_current_mems_allowed();
3879         } else {
3880 #ifdef CONFIG_MEMORY_HOTPLUG
3881                 if (zone)
3882                         setup_zone_pageset(zone);
3883 #endif
3884                 /* we have to stop all cpus to guarantee there is no user
3885                    of zonelist */
3886                 stop_machine(__build_all_zonelists, pgdat, NULL);
3887                 /* cpuset refresh routine should be here */
3888         }
3889         vm_total_pages = nr_free_pagecache_pages();
3890         /*
3891          * Disable grouping by mobility if the number of pages in the
3892          * system is too low to allow the mechanism to work. It would be
3893          * more accurate, but expensive to check per-zone. This check is
3894          * made on memory-hotadd so a system can start with mobility
3895          * disabled and enable it later
3896          */
3897         if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3898                 page_group_by_mobility_disabled = 1;
3899         else
3900                 page_group_by_mobility_disabled = 0;
3901 
3902         printk("Built %i zonelists in %s order, mobility grouping %s.  "
3903                 "Total pages: %ld\n",
3904                         nr_online_nodes,
3905                         zonelist_order_name[current_zonelist_order],
3906                         page_group_by_mobility_disabled ? "off" : "on",
3907                         vm_total_pages);
3908 #ifdef CONFIG_NUMA
3909         printk("Policy zone: %s\n", zone_names[policy_zone]);
3910 #endif
3911 }
3912 
3913 /*
3914  * Helper functions to size the waitqueue hash table.
3915  * Essentially these want to choose hash table sizes sufficiently
3916  * large so that collisions trying to wait on pages are rare.
3917  * But in fact, the number of active page waitqueues on typical
3918  * systems is ridiculously low, less than 200. So this is even
3919  * conservative, even though it seems large.
3920  *
3921  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3922  * waitqueues, i.e. the size of the waitq table given the number of pages.
3923  */
3924 #define PAGES_PER_WAITQUEUE     256
3925 
3926 #ifndef CONFIG_MEMORY_HOTPLUG
3927 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3928 {
3929         unsigned long size = 1;
3930 
3931         pages /= PAGES_PER_WAITQUEUE;
3932 
3933         while (size < pages)
3934                 size <<= 1;
3935 
3936         /*
3937          * Once we have dozens or even hundreds of threads sleeping
3938          * on IO we've got bigger problems than wait queue collision.
3939          * Limit the size of the wait table to a reasonable size.
3940          */
3941         size = min(size, 4096UL);
3942 
3943         return max(size, 4UL);
3944 }
3945 #else
3946 /*
3947  * A zone's size might be changed by hot-add, so it is not possible to determine
3948  * a suitable size for its wait_table.  So we use the maximum size now.
3949  *
3950  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
3951  *
3952  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
3953  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3954  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
3955  *
3956  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3957  * or more by the traditional way. (See above).  It equals:
3958  *
3959  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
3960  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
3961  *    powerpc (64K page size)             : =  (32G +16M)byte.
3962  */
3963 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3964 {
3965         return 4096UL;
3966 }
3967 #endif
3968 
3969 /*
3970  * This is an integer logarithm so that shifts can be used later
3971  * to extract the more random high bits from the multiplicative
3972  * hash function before the remainder is taken.
3973  */
3974 static inline unsigned long wait_table_bits(unsigned long size)
3975 {
3976         return ffz(~size);
3977 }
3978 
3979 /*
3980  * Check if a pageblock contains reserved pages
3981  */
3982 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3983 {
3984         unsigned long pfn;
3985 
3986         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3987                 if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3988                         return 1;
3989         }
3990         return 0;
3991 }
3992 
3993 /*
3994  * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3995  * of blocks reserved is based on min_wmark_pages(zone). The memory within
3996  * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3997  * higher will lead to a bigger reserve which will get freed as contiguous
3998  * blocks as reclaim kicks in
3999  */
4000 static void setup_zone_migrate_reserve(struct zone *zone)
4001 {
4002         unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
4003         struct page *page;
4004         unsigned long block_migratetype;
4005         int reserve;
4006         int old_reserve;
4007 
4008         /*
4009          * Get the start pfn, end pfn and the number of blocks to reserve
4010          * We have to be careful to be aligned to pageblock_nr_pages to
4011          * make sure that we always check pfn_valid for the first page in
4012          * the block.
4013          */
4014         start_pfn = zone->zone_start_pfn;
4015         end_pfn = zone_end_pfn(zone);
4016         start_pfn = roundup(start_pfn, pageblock_nr_pages);
4017         reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
4018                                                         pageblock_order;
4019 
4020         /*
4021          * Reserve blocks are generally in place to help high-order atomic
4022          * allocations that are short-lived. A min_free_kbytes value that
4023          * would result in more than 2 reserve blocks for atomic allocations
4024          * is assumed to be in place to help anti-fragmentation for the
4025          * future allocation of hugepages at runtime.
4026          */
4027         reserve = min(2, reserve);
4028         old_reserve = zone->nr_migrate_reserve_block;
4029 
4030         /* When memory hot-add, we almost always need to do nothing */
4031         if (reserve == old_reserve)
4032                 return;
4033         zone->nr_migrate_reserve_block = reserve;
4034 
4035         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
4036                 if (!pfn_valid(pfn))
4037                         continue;
4038                 page = pfn_to_page(pfn);
4039 
4040                 /* Watch out for overlapping nodes */
4041                 if (page_to_nid(page) != zone_to_nid(zone))
4042                         continue;
4043 
4044                 block_migratetype = get_pageblock_migratetype(page);
4045 
4046                 /* Only test what is necessary when the reserves are not met */
4047                 if (reserve > 0) {
4048                         /*
4049                          * Blocks with reserved pages will never free, skip
4050                          * them.
4051                          */
4052                         block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
4053                         if (pageblock_is_reserved(pfn, block_end_pfn))
4054                                 continue;
4055 
4056                         /* If this block is reserved, account for it */
4057                         if (block_migratetype == MIGRATE_RESERVE) {
4058                                 reserve--;
4059                                 continue;
4060                         }
4061 
4062                         /* Suitable for reserving if this block is movable */
4063                         if (block_migratetype == MIGRATE_MOVABLE) {
4064                                 set_pageblock_migratetype(page,
4065                                                         MIGRATE_RESERVE);
4066                                 move_freepages_block(zone, page,
4067                                                         MIGRATE_RESERVE);
4068                                 reserve--;
4069                                 continue;
4070                         }
4071                 } else if (!old_reserve) {
4072                         /*
4073                          * At boot time we don't need to scan the whole zone
4074                          * for turning off MIGRATE_RESERVE.
4075                          */
4076                         break;
4077                 }
4078 
4079                 /*
4080                  * If the reserve is met and this is a previous reserved block,
4081                  * take it back
4082                  */
4083                 if (block_migratetype == MIGRATE_RESERVE) {
4084                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4085                         move_freepages_block(zone, page, MIGRATE_MOVABLE);
4086                 }
4087         }
4088 }
4089 
4090 /*
4091  * Initially all pages are reserved - free ones are freed
4092  * up by free_all_bootmem() once the early boot process is
4093  * done. Non-atomic initialization, single-pass.
4094  */
4095 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4096                 unsigned long start_pfn, enum memmap_context context)
4097 {
4098         struct page *page;
4099         unsigned long end_pfn = start_pfn + size;
4100         unsigned long pfn;
4101         struct zone *z;
4102 
4103         if (highest_memmap_pfn < end_pfn - 1)
4104                 highest_memmap_pfn = end_pfn - 1;
4105 
4106         z = &NODE_DATA(nid)->node_zones[zone];
4107         for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4108                 /*
4109                  * There can be holes in boot-time mem_map[]s
4110                  * handed to this function.  They do not
4111                  * exist on hotplugged memory.
4112                  */
4113                 if (context == MEMMAP_EARLY) {
4114                         if (!early_pfn_valid(pfn))
4115                                 continue;
4116                         if (!early_pfn_in_nid(pfn, nid))
4117                                 continue;
4118                 }
4119                 page = pfn_to_page(pfn);
4120                 set_page_links(page, zone, nid, pfn);
4121                 mminit_verify_page_links(page, zone, nid, pfn);
4122                 init_page_count(page);
4123                 page_mapcount_reset(page);
4124                 page_cpupid_reset_last(page);
4125                 SetPageReserved(page);
4126                 /*
4127                  * Mark the block movable so that blocks are reserved for
4128                  * movable at startup. This will force kernel allocations
4129                  * to reserve their blocks rather than leaking throughout
4130                  * the address space during boot when many long-lived
4131                  * kernel allocations are made. Later some blocks near
4132                  * the start are marked MIGRATE_RESERVE by
4133                  * setup_zone_migrate_reserve()
4134                  *
4135                  * bitmap is created for zone's valid pfn range. but memmap
4136                  * can be created for invalid pages (for alignment)
4137                  * check here not to call set_pageblock_migratetype() against
4138                  * pfn out of zone.
4139                  */
4140                 if ((z->zone_start_pfn <= pfn)
4141                     && (pfn < zone_end_pfn(z))
4142                     && !(pfn & (pageblock_nr_pages - 1)))
4143                         set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4144 
4145                 INIT_LIST_HEAD(&page->lru);
4146 #ifdef WANT_PAGE_VIRTUAL
4147                 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
4148                 if (!is_highmem_idx(zone))
4149                         set_page_address(page, __va(pfn << PAGE_SHIFT));
4150 #endif
4151         }
4152 }
4153 
4154 static void __meminit zone_init_free_lists(struct zone *zone)
4155 {
4156         unsigned int order, t;
4157         for_each_migratetype_order(order, t) {
4158                 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4159                 zone->free_area[order].nr_free = 0;
4160         }
4161 }
4162 
4163 #ifndef __HAVE_ARCH_MEMMAP_INIT
4164 #define memmap_init(size, nid, zone, start_pfn) \
4165         memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4166 #endif
4167 
4168 static int zone_batchsize(struct zone *zone)
4169 {
4170 #ifdef CONFIG_MMU
4171         int batch;
4172 
4173         /*
4174          * The per-cpu-pages pools are set to around 1000th of the
4175          * size of the zone.  But no more than 1/2 of a meg.
4176          *
4177          * OK, so we don't know how big the cache is.  So guess.
4178          */
4179         batch = zone->managed_pages / 1024;
4180         if (batch * PAGE_SIZE > 512 * 1024)
4181                 batch = (512 * 1024) / PAGE_SIZE;
4182         batch /= 4;             /* We effectively *= 4 below */
4183         if (batch < 1)
4184                 batch = 1;
4185 
4186         /*
4187          * Clamp the batch to a 2^n - 1 value. Having a power
4188          * of 2 value was found to be more likely to have
4189          * suboptimal cache aliasing properties in some cases.
4190          *
4191          * For example if 2 tasks are alternately allocating
4192          * batches of pages, one task can end up with a lot
4193          * of pages of one half of the possible page colors
4194          * and the other with pages of the other colors.
4195          */
4196         batch = rounddown_pow_of_two(batch + batch/2) - 1;
4197 
4198         return batch;
4199 
4200 #else
4201         /* The deferral and batching of frees should be suppressed under NOMMU
4202          * conditions.
4203          *
4204          * The problem is that NOMMU needs to be able to allocate large chunks
4205          * of contiguous memory as there's no hardware page translation to
4206          * assemble apparent contiguous memory from discontiguous pages.
4207          *
4208          * Queueing large contiguous runs of pages for batching, however,
4209          * causes the pages to actually be freed in smaller chunks.  As there
4210          * can be a significant delay between the individual batches being
4211          * recycled, this leads to the once large chunks of space being
4212          * fragmented and becoming unavailable for high-order allocations.
4213          */
4214         return 0;
4215 #endif
4216 }
4217 
4218 /*
4219  * pcp->high and pcp->batch values are related and dependent on one another:
4220  * ->batch must never be higher then ->high.
4221  * The following function updates them in a safe manner without read side
4222  * locking.
4223  *
4224  * Any new users of pcp->batch and pcp->high should ensure they can cope with
4225  * those fields changing asynchronously (acording the the above rule).
4226  *
4227  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4228  * outside of boot time (or some other assurance that no concurrent updaters
4229  * exist).
4230  */
4231 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4232                 unsigned long batch)
4233 {
4234        /* start with a fail safe value for batch */
4235         pcp->batch = 1;
4236         smp_wmb();
4237 
4238        /* Update high, then batch, in order */
4239         pcp->high = high;
4240         smp_wmb();
4241 
4242         pcp->batch = batch;
4243 }
4244 
4245 /* a companion to pageset_set_high() */
4246 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4247 {
4248         pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4249 }
4250 
4251 static void pageset_init(struct per_cpu_pageset *p)
4252 {
4253         struct per_cpu_pages *pcp;
4254         int migratetype;
4255 
4256         memset(p, 0, sizeof(*p));
4257 
4258         pcp = &p->pcp;
4259         pcp->count = 0;
4260         for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4261                 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4262 }
4263 
4264 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4265 {
4266         pageset_init(p);
4267         pageset_set_batch(p, batch);
4268 }
4269 
4270 /*
4271  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4272  * to the value high for the pageset p.
4273  */
4274 static void pageset_set_high(struct per_cpu_pageset *p,
4275                                 unsigned long high)
4276 {
4277         unsigned long batch = max(1UL, high / 4);
4278         if ((high / 4) > (PAGE_SHIFT * 8))
4279                 batch = PAGE_SHIFT * 8;
4280 
4281         pageset_update(&p->pcp, high, batch);
4282 }
4283 
4284 static void pageset_set_high_and_batch(struct zone *zone,
4285                                        struct per_cpu_pageset *pcp)
4286 {
4287         if (percpu_pagelist_fraction)
4288                 pageset_set_high(pcp,
4289                         (zone->managed_pages /
4290                                 percpu_pagelist_fraction));
4291         else
4292                 pageset_set_batch(pcp, zone_batchsize(zone));
4293 }
4294 
4295 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4296 {
4297         struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4298 
4299         pageset_init(pcp);
4300         pageset_set_high_and_batch(zone, pcp);
4301 }
4302 
4303 static void __meminit setup_zone_pageset(struct zone *zone)
4304 {
4305         int cpu;
4306         zone->pageset = alloc_percpu(struct per_cpu_pageset);
4307         for_each_possible_cpu(cpu)
4308                 zone_pageset_init(zone, cpu);
4309 }
4310 
4311 /*
4312  * Allocate per cpu pagesets and initialize them.
4313  * Before this call only boot pagesets were available.
4314  */
4315 void __init setup_per_cpu_pageset(void)
4316 {
4317         struct zone *zone;
4318 
4319         for_each_populated_zone(zone)
4320                 setup_zone_pageset(zone);
4321 }
4322 
4323 static noinline __init_refok
4324 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4325 {
4326         int i;
4327         size_t alloc_size;
4328 
4329         /*
4330          * The per-page waitqueue mechanism uses hashed waitqueues
4331          * per zone.
4332          */
4333         zone->wait_table_hash_nr_entries =
4334                  wait_table_hash_nr_entries(zone_size_pages);
4335         zone->wait_table_bits =
4336                 wait_table_bits(zone->wait_table_hash_nr_entries);
4337         alloc_size = zone->wait_table_hash_nr_entries
4338                                         * sizeof(wait_queue_head_t);
4339 
4340         if (!slab_is_available()) {
4341                 zone->wait_table = (wait_queue_head_t *)
4342                         memblock_virt_alloc_node_nopanic(
4343                                 alloc_size, zone->zone_pgdat->node_id);
4344         } else {
4345                 /*
4346                  * This case means that a zone whose size was 0 gets new memory
4347                  * via memory hot-add.
4348                  * But it may be the case that a new node was hot-added.  In
4349                  * this case vmalloc() will not be able to use this new node's
4350                  * memory - this wait_table must be initialized to use this new
4351                  * node itself as well.
4352                  * To use this new node's memory, further consideration will be
4353                  * necessary.
4354                  */
4355                 zone->wait_table = vmalloc(alloc_size);
4356         }
4357         if (!zone->wait_table)
4358                 return -ENOMEM;
4359 
4360         for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4361                 init_waitqueue_head(zone->wait_table + i);
4362 
4363         return 0;
4364 }
4365 
4366 static __meminit void zone_pcp_init(struct zone *zone)
4367 {
4368         /*
4369          * per cpu subsystem is not up at this point. The following code
4370          * relies on the ability of the linker to provide the
4371          * offset of a (static) per cpu variable into the per cpu area.
4372          */
4373         zone->pageset = &boot_pageset;
4374 
4375         if (populated_zone(zone))
4376                 printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
4377                         zone->name, zone->present_pages,
4378                                          zone_batchsize(zone));
4379 }
4380 
4381 int __meminit init_currently_empty_zone(struct zone *zone,
4382                                         unsigned long zone_start_pfn,
4383                                         unsigned long size,
4384                                         enum memmap_context context)
4385 {
4386         struct pglist_data *pgdat = zone->zone_pgdat;
4387         int ret;
4388         ret = zone_wait_table_init(zone, size);
4389         if (ret)
4390                 return ret;
4391         pgdat->nr_zones = zone_idx(zone) + 1;
4392 
4393         zone->zone_start_pfn = zone_start_pfn;
4394 
4395         mminit_dprintk(MMINIT_TRACE, "memmap_init",
4396                         "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4397                         pgdat->node_id,
4398                         (unsigned long)zone_idx(zone),
4399                         zone_start_pfn, (zone_start_pfn + size));
4400 
4401         zone_init_free_lists(zone);
4402 
4403         return 0;
4404 }
4405 
4406 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4407 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4408 /*
4409  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4410  */
4411 int __meminit __early_pfn_to_nid(unsigned long pfn)
4412 {
4413         unsigned long start_pfn, end_pfn;
4414         int nid;
4415         /*
4416          * NOTE: The following SMP-unsafe globals are only used early in boot
4417          * when the kernel is running single-threaded.
4418          */
4419         static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4420         static int __meminitdata last_nid;
4421 
4422         if (last_start_pfn <= pfn && pfn < last_end_pfn)
4423                 return last_nid;
4424 
4425         nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4426         if (nid != -1) {
4427                 last_start_pfn = start_pfn;
4428                 last_end_pfn = end_pfn;
4429                 last_nid = nid;
4430         }
4431 
4432         return nid;
4433 }
4434 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4435 
4436 int __meminit early_pfn_to_nid(unsigned long pfn)
4437 {
4438         int nid;
4439 
4440         nid = __early_pfn_to_nid(pfn);
4441         if (nid >= 0)
4442                 return nid;
4443         /* just returns 0 */
4444         return 0;
4445 }
4446 
4447 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4448 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4449 {
4450         int nid;
4451 
4452         nid = __early_pfn_to_nid(pfn);
4453         if (nid >= 0 && nid != node)
4454                 return false;
4455         return true;
4456 }
4457 #endif
4458 
4459 /**
4460  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4461  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4462  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4463  *
4464  * If an architecture guarantees that all ranges registered contain no holes
4465  * and may be freed, this this function may be used instead of calling
4466  * memblock_free_early_nid() manually.
4467  */
4468 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
4469 {
4470         unsigned long start_pfn, end_pfn;
4471         int i, this_nid;
4472 
4473         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4474                 start_pfn = min(start_pfn, max_low_pfn);
4475                 end_pfn = min(end_pfn, max_low_pfn);
4476 
4477                 if (start_pfn < end_pfn)
4478                         memblock_free_early_nid(PFN_PHYS(start_pfn),
4479                                         (end_pfn - start_pfn) << PAGE_SHIFT,
4480                                         this_nid);
4481         }
4482 }
4483 
4484 /**
4485  * sparse_memory_present_with_active_regions - Call memory_present for each active range
4486  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4487  *
4488  * If an architecture guarantees that all ranges registered contain no holes and may
4489  * be freed, this function may be used instead of calling memory_present() manually.
4490  */
4491 void __init sparse_memory_present_with_active_regions(int nid)
4492 {
4493         unsigned long start_pfn, end_pfn;
4494         int i, this_nid;
4495 
4496         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4497                 memory_present(this_nid, start_pfn, end_pfn);
4498 }
4499 
4500 /**
4501  * get_pfn_range_for_nid - Return the start and end page frames for a node
4502  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4503  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4504  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4505  *
4506  * It returns the start and end page frame of a node based on information
4507  * provided by memblock_set_node(). If called for a node
4508  * with no available memory, a warning is printed and the start and end
4509  * PFNs will be 0.
4510  */
4511 void __meminit get_pfn_range_for_nid(unsigned int nid,
4512                         unsigned long *start_pfn, unsigned long *end_pfn)
4513 {
4514         unsigned long this_start_pfn, this_end_pfn;
4515         int i;
4516 
4517         *start_pfn = -1UL;
4518         *end_pfn = 0;
4519 
4520         for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4521                 *start_pfn = min(*start_pfn, this_start_pfn);
4522                 *end_pfn = max(*end_pfn, this_end_pfn);
4523         }
4524 
4525         if (*start_pfn == -1UL)
4526                 *start_pfn = 0;
4527 }
4528 
4529 /*
4530  * This finds a zone that can be used for ZONE_MOVABLE pages. The
4531  * assumption is made that zones within a node are ordered in monotonic
4532  * increasing memory addresses so that the "highest" populated zone is used
4533  */
4534 static void __init find_usable_zone_for_movable(void)
4535 {
4536         int zone_index;
4537         for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4538                 if (zone_index == ZONE_MOVABLE)
4539                         continue;
4540 
4541                 if (arch_zone_highest_possible_pfn[zone_index] >
4542                                 arch_zone_lowest_possible_pfn[zone_index])
4543                         break;
4544         }
4545 
4546         VM_BUG_ON(zone_index == -1);
4547         movable_zone = zone_index;
4548 }
4549 
4550 /*
4551  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4552  * because it is sized independent of architecture. Unlike the other zones,
4553  * the starting point for ZONE_MOVABLE is not fixed. It may be different
4554  * in each node depending on the size of each node and how evenly kernelcore
4555  * is distributed. This helper function adjusts the zone ranges
4556  * provided by the architecture for a given node by using the end of the
4557  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4558  * zones within a node are in order of monotonic increases memory addresses
4559  */
4560 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4561                                         unsigned long zone_type,
4562                                         unsigned long node_start_pfn,
4563                                         unsigned long node_end_pfn,
4564                                         unsigned long *zone_start_pfn,
4565                                         unsigned long *zone_end_pfn)
4566 {
4567         /* Only adjust if ZONE_MOVABLE is on this node */
4568         if (zone_movable_pfn[nid]) {
4569                 /* Size ZONE_MOVABLE */
4570                 if (zone_type == ZONE_MOVABLE) {
4571                         *zone_start_pfn = zone_movable_pfn[nid];
4572                         *zone_end_pfn = min(node_end_pfn,
4573                                 arch_zone_highest_possible_pfn[movable_zone]);
4574 
4575                 /* Adjust for ZONE_MOVABLE starting within this range */
4576                 } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4577                                 *zone_end_pfn > zone_movable_pfn[nid]) {
4578                         *zone_end_pfn = zone_movable_pfn[nid];
4579 
4580                 /* Check if this whole range is within ZONE_MOVABLE */
4581                 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
4582                         *zone_start_pfn = *zone_end_pfn;
4583         }
4584 }
4585 
4586 /*
4587  * Return the number of pages a zone spans in a node, including holes
4588  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4589  */
4590 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4591                                         unsigned long zone_type,
4592                                         unsigned long node_start_pfn,
4593                                         unsigned long node_end_pfn,
4594                                         unsigned long *ignored)
4595 {
4596         unsigned long zone_start_pfn, zone_end_pfn;
4597 
4598         /* Get the start and end of the zone */
4599         zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4600         zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4601         adjust_zone_range_for_zone_movable(nid, zone_type,
4602                                 node_start_pfn, node_end_pfn,
4603                                 &zone_start_pfn, &zone_end_pfn);
4604 
4605         /* Check that this node has pages within the zone's required range */
4606         if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4607                 return 0;
4608 
4609         /* Move the zone boundaries inside the node if necessary */
4610         zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4611         zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4612 
4613         /* Return the spanned pages */
4614         return zone_end_pfn - zone_start_pfn;
4615 }
4616 
4617 /*
4618  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4619  * then all holes in the requested range will be accounted for.
4620  */
4621 unsigned long __meminit __absent_pages_in_range(int nid,
4622                                 unsigned long range_start_pfn,
4623                                 unsigned long range_end_pfn)
4624 {
4625         unsigned long nr_absent = range_end_pfn - range_start_pfn;
4626         unsigned long start_pfn, end_pfn;
4627         int i;
4628 
4629         for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4630                 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4631                 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4632                 nr_absent -= end_pfn - start_pfn;
4633         }
4634         return nr_absent;
4635 }
4636 
4637 /**
4638  * absent_pages_in_range - Return number of page frames in holes within a range
4639  * @start_pfn: The start PFN to start searching for holes
4640  * @end_pfn: The end PFN to stop searching for holes
4641  *
4642  * It returns the number of pages frames in memory holes within a range.
4643  */
4644 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4645                                                         unsigned long end_pfn)
4646 {
4647         return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4648 }
4649 
4650 /* Return the number of page frames in holes in a zone on a node */
4651 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4652                                         unsigned long zone_type,
4653                                         unsigned long node_start_pfn,
4654                                         unsigned long node_end_pfn,
4655                                         unsigned long *ignored)
4656 {
4657         unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4658         unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
4659         unsigned long zone_start_pfn, zone_end_pfn;
4660 
4661         zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4662         zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
4663 
4664         adjust_zone_range_for_zone_movable(nid, zone_type,
4665                         node_start_pfn, node_end_pfn,
4666                         &zone_start_pfn, &zone_end_pfn);
4667         return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4668 }
4669 
4670 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4671 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4672                                         unsigned long zone_type,
4673                                         unsigned long node_start_pfn,
4674                                         unsigned long node_end_pfn,
4675                                         unsigned long *zones_size)
4676 {
4677         return zones_size[zone_type];
4678 }
4679 
4680 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4681                                                 unsigned long zone_type,
4682                                                 unsigned long node_start_pfn,
4683                                                 unsigned long node_end_pfn,
4684                                                 unsigned long *zholes_size)
4685 {
4686         if (!zholes_size)
4687                 return 0;
4688 
4689         return zholes_size[zone_type];
4690 }
4691 
4692 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4693 
4694 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4695                                                 unsigned long node_start_pfn,
4696                                                 unsigned long node_end_pfn,
4697                                                 unsigned long *zones_size,
4698                                                 unsigned long *zholes_size)
4699 {
4700         unsigned long realtotalpages, totalpages = 0;
4701         enum zone_type i;
4702 
4703         for (i = 0; i < MAX_NR_ZONES; i++)
4704                 totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4705                                                          node_start_pfn,
4706                                                          node_end_pfn,
4707                                                          zones_size);
4708         pgdat->node_spanned_pages = totalpages;
4709 
4710         realtotalpages = totalpages;
4711         for (i = 0; i < MAX_NR_ZONES; i++)
4712                 realtotalpages -=
4713                         zone_absent_pages_in_node(pgdat->node_id, i,
4714                                                   node_start_pfn, node_end_pfn,
4715                                                   zholes_size);
4716         pgdat->node_present_pages = realtotalpages;
4717         printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4718                                                         realtotalpages);
4719 }
4720 
4721 #ifndef CONFIG_SPARSEMEM
4722 /*
4723  * Calculate the size of the zone->blockflags rounded to an unsigned long
4724  * Start by making sure zonesize is a multiple of pageblock_order by rounding
4725  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4726  * round what is now in bits to nearest long in bits, then return it in
4727  * bytes.
4728  */
4729 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
4730 {
4731         unsigned long usemapsize;
4732 
4733         zonesize += zone_start_pfn & (pageblock_nr_pages-1);
4734         usemapsize = roundup(zonesize, pageblock_nr_pages);
4735         usemapsize = usemapsize >> pageblock_order;
4736         usemapsize *= NR_PAGEBLOCK_BITS;
4737         usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4738 
4739         return usemapsize / 8;
4740 }
4741 
4742 static void __init setup_usemap(struct pglist_data *pgdat,
4743                                 struct zone *zone,
4744                                 unsigned long zone_start_pfn,
4745                                 unsigned long zonesize)
4746 {
4747         unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
4748         zone->pageblock_flags = NULL;
4749         if (usemapsize)
4750                 zone->pageblock_flags =
4751                         memblock_virt_alloc_node_nopanic(usemapsize,
4752                                                          pgdat->node_id);
4753 }
4754 #else
4755 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4756                                 unsigned long zone_start_pfn, unsigned long zonesize) {}
4757 #endif /* CONFIG_SPARSEMEM */
4758 
4759 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4760 
4761 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4762 void __paginginit set_pageblock_order(void)
4763 {
4764         unsigned int order;
4765 
4766         /* Check that pageblock_nr_pages has not already been setup */
4767         if (pageblock_order)
4768                 return;
4769 
4770         if (HPAGE_SHIFT > PAGE_SHIFT)
4771                 order = HUGETLB_PAGE_ORDER;
4772         else
4773                 order = MAX_ORDER - 1;
4774 
4775         /*
4776          * Assume the largest contiguous order of interest is a huge page.
4777          * This value may be variable depending on boot parameters on IA64 and
4778          * powerpc.
4779          */
4780         pageblock_order = order;
4781 }
4782 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4783 
4784 /*
4785  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4786  * is unused as pageblock_order is set at compile-time. See
4787  * include/linux/pageblock-flags.h for the values of pageblock_order based on
4788  * the kernel config
4789  */
4790 void __paginginit set_pageblock_order(void)
4791 {
4792 }
4793 
4794 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4795 
4796 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4797                                                    unsigned long present_pages)
4798 {
4799         unsigned long pages = spanned_pages;
4800 
4801         /*
4802          * Provide a more accurate estimation if there are holes within
4803          * the zone and SPARSEMEM is in use. If there are holes within the
4804          * zone, each populated memory region may cost us one or two extra
4805          * memmap pages due to alignment because memmap pages for each
4806          * populated regions may not naturally algined on page boundary.
4807          * So the (present_pages >> 4) heuristic is a tradeoff for that.
4808          */
4809         if (spanned_pages > present_pages + (present_pages >> 4) &&
4810             IS_ENABLED(CONFIG_SPARSEMEM))
4811                 pages = present_pages;
4812 
4813         return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4814 }
4815 
4816 /*
4817  * Set up the zone data structures:
4818  *   - mark all pages reserved
4819  *   - mark all memory queues empty
4820  *   - clear the memory bitmaps
4821  *
4822  * NOTE: pgdat should get zeroed by caller.
4823  */
4824 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4825                 unsigned long node_start_pfn, unsigned long node_end_pfn,
4826                 unsigned long *zones_size, unsigned long *zholes_size)
4827 {
4828         enum zone_type j;
4829         int nid = pgdat->node_id;
4830         unsigned long zone_start_pfn = pgdat->node_start_pfn;
4831         int ret;
4832 
4833         pgdat_resize_init(pgdat);
4834 #ifdef CONFIG_NUMA_BALANCING
4835         spin_lock_init(&pgdat->numabalancing_migrate_lock);
4836         pgdat->numabalancing_migrate_nr_pages = 0;
4837         pgdat->numabalancing_migrate_next_window = jiffies;
4838 #endif
4839         init_waitqueue_head(&pgdat->kswapd_wait);
4840         init_waitqueue_head(&pgdat->pfmemalloc_wait);
4841         pgdat_page_cgroup_init(pgdat);
4842 
4843         for (j = 0; j < MAX_NR_ZONES; j++) {
4844                 struct zone *zone = pgdat->node_zones + j;
4845                 unsigned long size, realsize, freesize, memmap_pages;
4846 
4847                 size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4848                                                   node_end_pfn, zones_size);
4849                 realsize = freesize = size - zone_absent_pages_in_node(nid, j,
4850                                                                 node_start_pfn,
4851                                                                 node_end_pfn,
4852                                                                 zholes_size);
4853 
4854                 /*
4855                  * Adjust freesize so that it accounts for how much memory
4856                  * is used by this zone for memmap. This affects the watermark
4857                  * and per-cpu initialisations
4858                  */
4859                 memmap_pages = calc_memmap_size(size, realsize);
4860                 if (freesize >= memmap_pages) {
4861                         freesize -= memmap_pages;
4862                         if (memmap_pages)
4863                                 printk(KERN_DEBUG
4864                                        "  %s zone: %lu pages used for memmap\n",
4865                                        zone_names[j], memmap_pages);
4866                 } else
4867                         printk(KERN_WARNING
4868                                 "  %s zone: %lu pages exceeds freesize %lu\n",
4869                                 zone_names[j], memmap_pages, freesize);
4870 
4871                 /* Account for reserved pages */
4872                 if (j == 0 && freesize > dma_reserve) {
4873                         freesize -= dma_reserve;
4874                         printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
4875                                         zone_names[0], dma_reserve);
4876                 }
4877 
4878                 if (!is_highmem_idx(j))
4879                         nr_kernel_pages += freesize;
4880                 /* Charge for highmem memmap if there are enough kernel pages */
4881                 else if (nr_kernel_pages > memmap_pages * 2)
4882                         nr_kernel_pages -= memmap_pages;
4883                 nr_all_pages += freesize;
4884 
4885                 zone->spanned_pages = size;
4886                 zone->present_pages = realsize;
4887                 /*
4888                  * Set an approximate value for lowmem here, it will be adjusted
4889                  * when the bootmem allocator frees pages into the buddy system.
4890                  * And all highmem pages will be managed by the buddy system.
4891                  */
4892                 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
4893 #ifdef CONFIG_NUMA
4894                 zone->node = nid;
4895                 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
4896                                                 / 100;
4897                 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
4898 #endif
4899                 zone->name = zone_names[j];
4900                 spin_lock_init(&zone->lock);
4901                 spin_lock_init(&zone->lru_lock);
4902                 zone_seqlock_init(zone);
4903                 zone->zone_pgdat = pgdat;
4904                 zone_pcp_init(zone);
4905 
4906                 /* For bootup, initialized properly in watermark setup */
4907                 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4908 
4909                 lruvec_init(&zone->lruvec);
4910                 if (!size)
4911                         continue;
4912 
4913                 set_pageblock_order();
4914                 setup_usemap(pgdat, zone, zone_start_pfn, size);
4915                 ret = init_currently_empty_zone(zone, zone_start_pfn,
4916                                                 size, MEMMAP_EARLY);
4917                 BUG_ON(ret);
4918                 memmap_init(size, nid, j, zone_start_pfn);
4919                 zone_start_pfn += size;
4920         }
4921 }
4922 
4923 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4924 {
4925         /* Skip empty nodes */
4926         if (!pgdat->node_spanned_pages)
4927                 return;
4928 
4929 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4930         /* ia64 gets its own node_mem_map, before this, without bootmem */
4931         if (!pgdat->node_mem_map) {
4932                 unsigned long size, start, end;
4933                 struct page *map;
4934 
4935                 /*
4936                  * The zone's endpoints aren't required to be MAX_ORDER
4937                  * aligned but the node_mem_map endpoints must be in order
4938                  * for the buddy allocator to function correctly.
4939                  */
4940                 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4941                 end = pgdat_end_pfn(pgdat);
4942                 end = ALIGN(end, MAX_ORDER_NR_PAGES);
4943                 size =  (end - start) * sizeof(struct page);
4944                 map = alloc_remap(pgdat->node_id, size);
4945                 if (!map)
4946                         map = memblock_virt_alloc_node_nopanic(size,
4947                                                                pgdat->node_id);
4948                 pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4949         }
4950 #ifndef CONFIG_NEED_MULTIPLE_NODES
4951         /*
4952          * With no DISCONTIG, the global mem_map is just set as node 0's
4953          */
4954         if (pgdat == NODE_DATA(0)) {
4955                 mem_map = NODE_DATA(0)->node_mem_map;
4956 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4957                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4958                         mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4959 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4960         }
4961 #endif
4962 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4963 }
4964 
4965 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4966                 unsigned long node_start_pfn, unsigned long *zholes_size)
4967 {
4968         pg_data_t *pgdat = NODE_DATA(nid);
4969         unsigned long start_pfn = 0;
4970         unsigned long end_pfn = 0;
4971 
4972         /* pg_data_t should be reset to zero when it's allocated */
4973         WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
4974 
4975         pgdat->node_id = nid;
4976         pgdat->node_start_pfn = node_start_pfn;
4977 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4978         get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
4979 #endif
4980         calculate_node_totalpages(pgdat, start_pfn, end_pfn,
4981                                   zones_size, zholes_size);
4982 
4983         alloc_node_mem_map(pgdat);
4984 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4985         printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4986                 nid, (unsigned long)pgdat,
4987                 (unsigned long)pgdat->node_mem_map);
4988 #endif
4989 
4990         free_area_init_core(pgdat, start_pfn, end_pfn,
4991                             zones_size, zholes_size);
4992 }
4993 
4994 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4995 
4996 #if MAX_NUMNODES > 1
4997 /*
4998  * Figure out the number of possible node ids.
4999  */
5000 void __init setup_nr_node_ids(void)
5001 {
5002         unsigned int node;
5003         unsigned int highest = 0;
5004 
5005         for_each_node_mask(node, node_possible_map)
5006                 highest = node;
5007         nr_node_ids = highest + 1;
5008 }
5009 #endif
5010 
5011 /**
5012  * node_map_pfn_alignment - determine the maximum internode alignment
5013  *
5014  * This function should be called after node map is populated and sorted.
5015  * It calculates the maximum power of two alignment which can distinguish
5016  * all the nodes.
5017  *
5018  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5019  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
5020  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
5021  * shifted, 1GiB is enough and this function will indicate so.
5022  *
5023  * This is used to test whether pfn -> nid mapping of the chosen memory
5024  * model has fine enough granularity to avoid incorrect mapping for the
5025  * populated node map.
5026  *
5027  * Returns the determined alignment in pfn's.  0 if there is no alignment
5028  * requirement (single node).
5029  */
5030 unsigned long __init node_map_pfn_alignment(void)
5031 {
5032         unsigned long accl_mask = 0, last_end = 0;
5033         unsigned long start, end, mask;
5034         int last_nid = -1;
5035         int i, nid;
5036 
5037         for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5038                 if (!start || last_nid < 0 || last_nid == nid) {
5039                         last_nid = nid;
5040                         last_end = end;
5041                         continue;
5042                 }
5043 
5044                 /*
5045                  * Start with a mask granular enough to pin-point to the
5046                  * start pfn and tick off bits one-by-one until it becomes
5047                  * too coarse to separate the current node from the last.
5048                  */
5049                 mask = ~((1 << __ffs(start)) - 1);
5050                 while (mask && last_end <= (start & (mask << 1)))
5051                         mask <<= 1;
5052 
5053                 /* accumulate all internode masks */
5054                 accl_mask |= mask;
5055         }
5056 
5057         /* convert mask to number of pages */
5058         return ~accl_mask + 1;
5059 }
5060 
5061 /* Find the lowest pfn for a node */
5062 static unsigned long __init find_min_pfn_for_node(int nid)
5063 {
5064         unsigned long min_pfn = ULONG_MAX;
5065         unsigned long start_pfn;
5066         int i;
5067 
5068         for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5069                 min_pfn = min(min_pfn, start_pfn);
5070 
5071         if (min_pfn == ULONG_MAX) {
5072                 printk(KERN_WARNING
5073                         "Could not find start_pfn for node %d\n", nid);
5074                 return 0;
5075         }
5076 
5077         return min_pfn;
5078 }
5079 
5080 /**
5081  * find_min_pfn_with_active_regions - Find the minimum PFN registered
5082  *
5083  * It returns the minimum PFN based on information provided via
5084  * memblock_set_node().
5085  */
5086 unsigned long __init find_min_pfn_with_active_regions(void)
5087 {
5088         return find_min_pfn_for_node(MAX_NUMNODES);
5089 }
5090 
5091 /*
5092  * early_calculate_totalpages()
5093  * Sum pages in active regions for movable zone.
5094  * Populate N_MEMORY for calculating usable_nodes.
5095  */
5096 static unsigned long __init early_calculate_totalpages(void)
5097 {
5098         unsigned long totalpages = 0;
5099         unsigned long start_pfn, end_pfn;
5100         int i, nid;
5101 
5102         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5103                 unsigned long pages = end_pfn - start_pfn;
5104 
5105                 totalpages += pages;
5106                 if (pages)
5107                         node_set_state(nid, N_MEMORY);
5108         }
5109         return totalpages;
5110 }
5111 
5112 /*
5113  * Find the PFN the Movable zone begins in each node. Kernel memory
5114  * is spread evenly between nodes as long as the nodes have enough
5115  * memory. When they don't, some nodes will have more kernelcore than
5116  * others
5117  */
5118 static void __init find_zone_movable_pfns_for_nodes(void)
5119 {
5120         int i, nid;
5121         unsigned long usable_startpfn;
5122         unsigned long kernelcore_node, kernelcore_remaining;
5123         /* save the state before borrow the nodemask */
5124         nodemask_t saved_node_state = node_states[N_MEMORY];
5125         unsigned long totalpages = early_calculate_totalpages();
5126         int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5127         struct memblock_region *r;
5128 
5129         /* Need to find movable_zone earlier when movable_node is specified. */
5130         find_usable_zone_for_movable();
5131 
5132         /*
5133          * If movable_node is specified, ignore kernelcore and movablecore
5134          * options.
5135          */
5136         if (movable_node_is_enabled()) {
5137                 for_each_memblock(memory, r) {
5138                         if (!memblock_is_hotpluggable(r))
5139                                 continue;
5140 
5141                         nid = r->nid;
5142 
5143                         usable_startpfn = PFN_DOWN(r->base);
5144                         zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5145                                 min(usable_startpfn, zone_movable_pfn[nid]) :
5146                                 usable_startpfn;
5147                 }
5148 
5149                 goto out2;
5150         }
5151 
5152         /*
5153          * If movablecore=nn[KMG] was specified, calculate what size of
5154          * kernelcore that corresponds so that memory usable for
5155          * any allocation type is evenly spread. If both kernelcore
5156          * and movablecore are specified, then the value of kernelcore
5157          * will be used for required_kernelcore if it's greater than
5158          * what movablecore would have allowed.
5159          */
5160         if (required_movablecore) {
5161                 unsigned long corepages;
5162 
5163                 /*
5164                  * Round-up so that ZONE_MOVABLE is at least as large as what
5165                  * was requested by the user
5166                  */
5167                 required_movablecore =
5168                         roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5169                 corepages = totalpages - required_movablecore;
5170 
5171                 required_kernelcore = max(required_kernelcore, corepages);
5172         }
5173 
5174         /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5175         if (!required_kernelcore)
5176                 goto out;
5177 
5178         /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5179         usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5180 
5181 restart:
5182         /* Spread kernelcore memory as evenly as possible throughout nodes */
5183         kernelcore_node = required_kernelcore / usable_nodes;
5184         for_each_node_state(nid, N_MEMORY) {
5185                 unsigned long start_pfn, end_pfn;
5186 
5187                 /*
5188                  * Recalculate kernelcore_node if the division per node
5189                  * now exceeds what is necessary to satisfy the requested
5190                  * amount of memory for the kernel
5191                  */
5192                 if (required_kernelcore < kernelcore_node)
5193                         kernelcore_node = required_kernelcore / usable_nodes;
5194 
5195                 /*
5196                  * As the map is walked, we track how much memory is usable
5197                  * by the kernel using kernelcore_remaining. When it is
5198                  * 0, the rest of the node is usable by ZONE_MOVABLE
5199                  */
5200                 kernelcore_remaining = kernelcore_node;
5201 
5202                 /* Go through each range of PFNs within this node */
5203                 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5204                         unsigned long size_pages;
5205 
5206                         start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5207                         if (start_pfn >= end_pfn)
5208                                 continue;
5209 
5210                         /* Account for what is only usable for kernelcore */
5211                         if (start_pfn < usable_startpfn) {
5212                                 unsigned long kernel_pages;
5213                                 kernel_pages = min(end_pfn, usable_startpfn)
5214                                                                 - start_pfn;
5215 
5216                                 kernelcore_remaining -= min(kernel_pages,
5217                                                         kernelcore_remaining);
5218                                 required_kernelcore -= min(kernel_pages,
5219                                                         required_kernelcore);
5220 
5221                                 /* Continue if range is now fully accounted */
5222                                 if (end_pfn <= usable_startpfn) {
5223 
5224                                         /*
5225                                          * Push zone_movable_pfn to the end so
5226                                          * that if we have to rebalance
5227                                          * kernelcore across nodes, we will
5228                                          * not double account here
5229                                          */
5230                                         zone_movable_pfn[nid] = end_pfn;
5231                                         continue;
5232                                 }
5233                                 start_pfn = usable_startpfn;
5234                         }
5235 
5236                         /*
5237                          * The usable PFN range for ZONE_MOVABLE is from
5238                          * start_pfn->end_pfn. Calculate size_pages as the
5239                          * number of pages used as kernelcore
5240                          */
5241                         size_pages = end_pfn - start_pfn;
5242                         if (size_pages > kernelcore_remaining)
5243                                 size_pages = kernelcore_remaining;
5244                         zone_movable_pfn[nid] = start_pfn + size_pages;
5245 
5246                         /*
5247                          * Some kernelcore has been met, update counts and
5248                          * break if the kernelcore for this node has been
5249                          * satisfied
5250                          */
5251                         required_kernelcore -= min(required_kernelcore,
5252                                                                 size_pages);
5253                         kernelcore_remaining -= size_pages;
5254                         if (!kernelcore_remaining)
5255                                 break;
5256                 }
5257         }
5258 
5259         /*
5260          * If there is still required_kernelcore, we do another pass with one
5261          * less node in the count. This will push zone_movable_pfn[nid] further
5262          * along on the nodes that still have memory until kernelcore is
5263          * satisfied
5264          */
5265         usable_nodes--;
5266         if (usable_nodes && required_kernelcore > usable_nodes)
5267                 goto restart;
5268 
5269 out2:
5270         /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5271         for (nid = 0; nid < MAX_NUMNODES; nid++)
5272                 zone_movable_pfn[nid] =
5273                         roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5274 
5275 out:
5276         /* restore the node_state */
5277         node_states[N_MEMORY] = saved_node_state;
5278 }
5279 
5280 /* Any regular or high memory on that node ? */
5281 static void check_for_memory(pg_data_t *pgdat, int nid)
5282 {
5283         enum zone_type zone_type;
5284 
5285         if (N_MEMORY == N_NORMAL_MEMORY)
5286                 return;
5287 
5288         for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5289                 struct zone *zone = &pgdat->node_zones[zone_type];
5290                 if (populated_zone(zone)) {
5291                         node_set_state(nid, N_HIGH_MEMORY);
5292                         if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5293                             zone_type <= ZONE_NORMAL)
5294                                 node_set_state(nid, N_NORMAL_MEMORY);
5295                         break;
5296                 }
5297         }
5298 }
5299 
5300 /**
5301  * free_area_init_nodes - Initialise all pg_data_t and zone data
5302  * @max_zone_pfn: an array of max PFNs for each zone
5303  *
5304  * This will call free_area_init_node() for each active node in the system.
5305  * Using the page ranges provided by memblock_set_node(), the size of each
5306  * zone in each node and their holes is calculated. If the maximum PFN
5307  * between two adjacent zones match, it is assumed that the zone is empty.
5308  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5309  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5310  * starts where the previous one ended. For example, ZONE_DMA32 starts
5311  * at arch_max_dma_pfn.
5312  */
5313 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5314 {
5315         unsigned long start_pfn, end_pfn;
5316         int i, nid;
5317 
5318         /* Record where the zone boundaries are */
5319         memset(arch_zone_lowest_possible_pfn, 0,
5320                                 sizeof(arch_zone_lowest_possible_pfn));
5321         memset(arch_zone_highest_possible_pfn, 0,
5322                                 sizeof(arch_zone_highest_possible_pfn));
5323         arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5324         arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5325         for (i = 1; i < MAX_NR_ZONES; i++) {
5326                 if (i == ZONE_MOVABLE)
5327                         continue;
5328                 arch_zone_lowest_possible_pfn[i] =
5329                         arch_zone_highest_possible_pfn[i-1];
5330                 arch_zone_highest_possible_pfn[i] =
5331                         max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5332         }
5333         arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5334         arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5335 
5336         /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5337         memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5338         find_zone_movable_pfns_for_nodes();
5339 
5340         /* Print out the zone ranges */
5341         printk("Zone ranges:\n");
5342         for (i = 0; i < MAX_NR_ZONES; i++) {
5343                 if (i == ZONE_MOVABLE)
5344                         continue;
5345                 printk(KERN_CONT "  %-8s ", zone_names[i]);
5346                 if (arch_zone_lowest_possible_pfn[i] ==
5347                                 arch_zone_highest_possible_pfn[i])
5348                         printk(KERN_CONT "empty\n");
5349                 else
5350                         printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5351                                 arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5352                                 (arch_zone_highest_possible_pfn[i]
5353                                         << PAGE_SHIFT) - 1);
5354         }
5355 
5356         /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5357         printk("Movable zone start for each node\n");
5358         for (i = 0; i < MAX_NUMNODES; i++) {
5359                 if (zone_movable_pfn[i])
5360                         printk("  Node %d: %#010lx\n", i,
5361                                zone_movable_pfn[i] << PAGE_SHIFT);
5362         }
5363 
5364         /* Print out the early node map */
5365         printk("Early memory node ranges\n");
5366         for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5367                 printk("  node %3d: [mem %#010lx-%#010lx]\n", nid,
5368                        start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
5369 
5370         /* Initialise every node */
5371         mminit_verify_pageflags_layout();
5372         setup_nr_node_ids();
5373         for_each_online_node(nid) {
5374                 pg_data_t *pgdat = NODE_DATA(nid);
5375                 free_area_init_node(nid, NULL,
5376                                 find_min_pfn_for_node(nid), NULL);
5377 
5378                 /* Any memory on that node */
5379                 if (pgdat->node_present_pages)
5380                         node_set_state(nid, N_MEMORY);
5381                 check_for_memory(pgdat, nid);
5382         }
5383 }
5384 
5385 static int __init cmdline_parse_core(char *p, unsigned long *core)
5386 {
5387         unsigned long long coremem;
5388         if (!p)
5389                 return -EINVAL;
5390 
5391         coremem = memparse(p, &p);
5392         *core = coremem >> PAGE_SHIFT;
5393 
5394         /* Paranoid check that UL is enough for the coremem value */
5395         WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5396 
5397         return 0;
5398 }
5399 
5400 /*
5401  * kernelcore=size sets the amount of memory for use for allocations that
5402  * cannot be reclaimed or migrated.
5403  */
5404 static int __init cmdline_parse_kernelcore(char *p)
5405 {
5406         return cmdline_parse_core(p, &required_kernelcore);
5407 }
5408 
5409 /*
5410  * movablecore=size sets the amount of memory for use for allocations that
5411  * can be reclaimed or migrated.
5412  */
5413 static int __init cmdline_parse_movablecore(char *p)
5414 {
5415         return cmdline_parse_core(p, &required_movablecore);
5416 }
5417 
5418 early_param("kernelcore", cmdline_parse_kernelcore);
5419 early_param("movablecore", cmdline_parse_movablecore);
5420 
5421 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5422 
5423 void adjust_managed_page_count(struct page *page, long count)
5424 {
5425         spin_lock(&managed_page_count_lock);
5426         page_zone(page)->managed_pages += count;
5427         totalram_pages += count;
5428 #ifdef CONFIG_HIGHMEM
5429         if (PageHighMem(page))
5430                 totalhigh_pages += count;
5431 #endif
5432         spin_unlock(&managed_page_count_lock);
5433 }
5434 EXPORT_SYMBOL(adjust_managed_page_count);
5435 
5436 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
5437 {
5438         void *pos;
5439         unsigned long pages = 0;
5440 
5441         start = (void *)PAGE_ALIGN((unsigned long)start);
5442         end = (void *)((unsigned long)end & PAGE_MASK);
5443         for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
5444                 if ((unsigned int)poison <= 0xFF)
5445                         memset(pos, poison, PAGE_SIZE);
5446                 free_reserved_page(virt_to_page(pos));
5447         }
5448 
5449         if (pages && s)
5450                 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5451                         s, pages << (PAGE_SHIFT - 10), start, end);
5452 
5453         return pages;
5454 }
5455 EXPORT_SYMBOL(free_reserved_area);
5456 
5457 #ifdef  CONFIG_HIGHMEM
5458 void free_highmem_page(struct page *page)
5459 {
5460         __free_reserved_page(page);
5461         totalram_pages++;
5462         page_zone(page)->managed_pages++;
5463         totalhigh_pages++;
5464 }
5465 #endif
5466 
5467 
5468 void __init mem_init_print_info(const char *str)
5469 {
5470         unsigned long physpages, codesize, datasize, rosize, bss_size;
5471         unsigned long init_code_size, init_data_size;
5472 
5473         physpages = get_num_physpages();
5474         codesize = _etext - _stext;
5475         datasize = _edata - _sdata;
5476         rosize = __end_rodata - __start_rodata;
5477         bss_size = __bss_stop - __bss_start;
5478         init_data_size = __init_end - __init_begin;
5479         init_code_size = _einittext - _sinittext;
5480 
5481         /*
5482          * Detect special cases and adjust section sizes accordingly:
5483          * 1) .init.* may be embedded into .data sections
5484          * 2) .init.text.* may be out of [__init_begin, __init_end],
5485          *    please refer to arch/tile/kernel/vmlinux.lds.S.
5486          * 3) .rodata.* may be embedded into .text or .data sections.
5487          */
5488 #define adj_init_size(start, end, size, pos, adj) \
5489         do { \
5490                 if (start <= pos && pos < end && size > adj) \
5491                         size -= adj; \
5492         } while (0)
5493 
5494         adj_init_size(__init_begin, __init_end, init_data_size,
5495                      _sinittext, init_code_size);
5496         adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5497         adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5498         adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5499         adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5500 
5501 #undef  adj_init_size
5502 
5503         printk("Memory: %luK/%luK available "
5504                "(%luK kernel code, %luK rwdata, %luK rodata, "
5505                "%luK init, %luK bss, %luK reserved"
5506 #ifdef  CONFIG_HIGHMEM
5507                ", %luK highmem"
5508 #endif
5509                "%s%s)\n",
5510                nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5511                codesize >> 10, datasize >> 10, rosize >> 10,
5512                (init_data_size + init_code_size) >> 10, bss_size >> 10,
5513                (physpages - totalram_pages) << (PAGE_SHIFT-10),
5514 #ifdef  CONFIG_HIGHMEM
5515                totalhigh_pages << (PAGE_SHIFT-10),
5516 #endif
5517                str ? ", " : "", str ? str : "");
5518 }
5519 
5520 /**
5521  * set_dma_reserve - set the specified number of pages reserved in the first zone
5522  * @new_dma_reserve: The number of pages to mark reserved
5523  *
5524  * The per-cpu batchsize and zone watermarks are determined by present_pages.
5525  * In the DMA zone, a significant percentage may be consumed by kernel image
5526  * and other unfreeable allocations which can skew the watermarks badly. This
5527  * function may optionally be used to account for unfreeable pages in the
5528  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5529  * smaller per-cpu batchsize.
5530  */
5531 void __init set_dma_reserve(unsigned long new_dma_reserve)
5532 {
5533         dma_reserve = new_dma_reserve;
5534 }
5535 
5536 void __init free_area_init(unsigned long *zones_size)
5537 {
5538         free_area_init_node(0, zones_size,
5539                         __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5540 }
5541 
5542 static int page_alloc_cpu_notify(struct notifier_block *self,
5543                                  unsigned long action, void *hcpu)
5544 {
5545         int cpu = (unsigned long)hcpu;
5546 
5547         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
5548                 lru_add_drain_cpu(cpu);
5549                 drain_pages(cpu);
5550 
5551                 /*
5552                  * Spill the event counters of the dead processor
5553                  * into the current processors event counters.
5554                  * This artificially elevates the count of the current
5555                  * processor.
5556                  */
5557                 vm_events_fold_cpu(cpu);
5558 
5559                 /*
5560                  * Zero the differential counters of the dead processor
5561                  * so that the vm statistics are consistent.
5562                  *
5563                  * This is only okay since the processor is dead and cannot
5564                  * race with what we are doing.
5565                  */
5566                 cpu_vm_stats_fold(cpu);
5567         }
5568         return NOTIFY_OK;
5569 }
5570 
5571 void __init page_alloc_init(void)
5572 {
5573         hotcpu_notifier(page_alloc_cpu_notify, 0);
5574 }
5575 
5576 /*
5577  * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5578  *      or min_free_kbytes changes.
5579  */
5580 static void calculate_totalreserve_pages(void)
5581 {
5582         struct pglist_data *pgdat;
5583         unsigned long reserve_pages = 0;
5584         enum zone_type i, j;
5585 
5586         for_each_online_pgdat(pgdat) {
5587                 for (i = 0; i < MAX_NR_ZONES; i++) {
5588                         struct zone *zone = pgdat->node_zones + i;
5589                         long max = 0;
5590 
5591                         /* Find valid and maximum lowmem_reserve in the zone */
5592                         for (j = i; j < MAX_NR_ZONES; j++) {
5593                                 if (zone->lowmem_reserve[j] > max)
5594                                         max = zone->lowmem_reserve[j];
5595                         }
5596 
5597                         /* we treat the high watermark as reserved pages. */
5598                         max += high_wmark_pages(zone);
5599 
5600                         if (max > zone->managed_pages)
5601                                 max = zone->managed_pages;
5602                         reserve_pages += max;
5603                         /*
5604                          * Lowmem reserves are not available to
5605                          * GFP_HIGHUSER page cache allocations and
5606                          * kswapd tries to balance zones to their high
5607                          * watermark.  As a result, neither should be
5608                          * regarded as dirtyable memory, to prevent a
5609                          * situation where reclaim has to clean pages
5610                          * in order to balance the zones.
5611                          */
5612                         zone->dirty_balance_reserve = max;
5613                 }
5614         }
5615         dirty_balance_reserve = reserve_pages;
5616         totalreserve_pages = reserve_pages;
5617 }
5618 
5619 /*
5620  * setup_per_zone_lowmem_reserve - called whenever
5621  *      sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
5622  *      has a correct pages reserved value, so an adequate number of
5623  *      pages are left in the zone after a successful __alloc_pages().
5624  */
5625 static void setup_per_zone_lowmem_reserve(void)
5626 {
5627         struct pglist_data *pgdat;
5628         enum zone_type j, idx;
5629 
5630         for_each_online_pgdat(pgdat) {
5631                 for (j = 0; j < MAX_NR_ZONES; j++) {
5632                         struct zone *zone = pgdat->node_zones + j;
5633                         unsigned long managed_pages = zone->managed_pages;
5634 
5635                         zone->lowmem_reserve[j] = 0;
5636 
5637                         idx = j;
5638                         while (idx) {
5639                                 struct zone *lower_zone;
5640 
5641                                 idx--;
5642 
5643                                 if (sysctl_lowmem_reserve_ratio[idx] < 1)
5644                                         sysctl_lowmem_reserve_ratio[idx] = 1;
5645 
5646                                 lower_zone = pgdat->node_zones + idx;
5647                                 lower_zone->lowmem_reserve[j] = managed_pages /
5648                                         sysctl_lowmem_reserve_ratio[idx];
5649                                 managed_pages += lower_zone->managed_pages;
5650                         }
5651                 }
5652         }
5653 
5654         /* update totalreserve_pages */
5655         calculate_totalreserve_pages();
5656 }
5657 
5658 static void __setup_per_zone_wmarks(void)
5659 {
5660         unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5661         unsigned long lowmem_pages = 0;
5662         struct zone *zone;
5663         unsigned long flags;
5664 
5665         /* Calculate total number of !ZONE_HIGHMEM pages */
5666         for_each_zone(zone) {
5667                 if (!is_highmem(zone))
5668                         lowmem_pages += zone->managed_pages;
5669         }
5670 
5671         for_each_zone(zone) {
5672                 u64 tmp;
5673 
5674                 spin_lock_irqsave(&zone->lock, flags);
5675                 tmp = (u64)pages_min * zone->managed_pages;
5676                 do_div(tmp, lowmem_pages);
5677                 if (is_highmem(zone)) {
5678                         /*
5679                          * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5680                          * need highmem pages, so cap pages_min to a small
5681                          * value here.
5682                          *
5683                          * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5684                          * deltas controls asynch page reclaim, and so should
5685                          * not be capped for highmem.
5686                          */
5687                         unsigned long min_pages;
5688 
5689                         min_pages = zone->managed_pages / 1024;
5690                         min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
5691                         zone->watermark[WMARK_MIN] = min_pages;
5692                 } else {
5693                         /*
5694                          * If it's a lowmem zone, reserve a number of pages
5695                          * proportionate to the zone's size.
5696                          */
5697                         zone->watermark[WMARK_MIN] = tmp;
5698                 }
5699 
5700                 zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + (tmp >> 2);
5701                 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
5702 
5703                 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
5704                         high_wmark_pages(zone) - low_wmark_pages(zone) -
5705                         atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
5706 
5707                 setup_zone_migrate_reserve(zone);
5708                 spin_unlock_irqrestore(&zone->lock, flags);
5709         }
5710 
5711         /* update totalreserve_pages */
5712         calculate_totalreserve_pages();
5713 }
5714 
5715 /**
5716  * setup_per_zone_wmarks - called when min_free_kbytes changes
5717  * or when memory is hot-{added|removed}
5718  *
5719  * Ensures that the watermark[min,low,high] values for each zone are set
5720  * correctly with respect to min_free_kbytes.
5721  */
5722 void setup_per_zone_wmarks(void)
5723 {
5724         mutex_lock(&zonelists_mutex);
5725         __setup_per_zone_wmarks();
5726         mutex_unlock(&zonelists_mutex);
5727 }
5728 
5729 /*
5730  * The inactive anon list should be small enough that the VM never has to
5731  * do too much work, but large enough that each inactive page has a chance
5732  * to be referenced again before it is swapped out.
5733  *
5734  * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5735  * INACTIVE_ANON pages on this zone's LRU, maintained by the
5736  * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5737  * the anonymous pages are kept on the inactive list.
5738  *
5739  * total     target    max
5740  * memory    ratio     inactive anon
5741  * -------------------------------------
5742  *   10MB       1         5MB
5743  *  100MB       1        50MB
5744  *    1GB       3       250MB
5745  *   10GB      10       0.9GB
5746  *  100GB      31         3GB
5747  *    1TB     101        10GB
5748  *   10TB     320        32GB
5749  */
5750 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
5751 {
5752         unsigned int gb, ratio;
5753 
5754         /* Zone size in gigabytes */
5755         gb = zone->managed_pages >> (30 - PAGE_SHIFT);
5756         if (gb)
5757                 ratio = int_sqrt(10 * gb);
5758         else
5759                 ratio = 1;
5760 
5761         zone->inactive_ratio = ratio;
5762 }
5763 
5764 static void __meminit setup_per_zone_inactive_ratio(void)
5765 {
5766         struct zone *zone;
5767 
5768         for_each_zone(zone)
5769                 calculate_zone_inactive_ratio(zone);
5770 }
5771 
5772 /*
5773  * Initialise min_free_kbytes.
5774  *
5775  * For small machines we want it small (128k min).  For large machines
5776  * we want it large (64MB max).  But it is not linear, because network
5777  * bandwidth does not increase linearly with machine size.  We use
5778  *
5779  *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5780  *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
5781  *
5782  * which yields
5783  *
5784  * 16MB:        512k
5785  * 32MB:        724k
5786  * 64MB:        1024k
5787  * 128MB:       1448k
5788  * 256MB:       2048k
5789  * 512MB:       2896k
5790  * 1024MB:      4096k
5791  * 2048MB:      5792k
5792  * 4096MB:      8192k
5793  * 8192MB:      11584k
5794  * 16384MB:     16384k
5795  */
5796 int __meminit init_per_zone_wmark_min(void)
5797 {
5798         unsigned long lowmem_kbytes;
5799         int new_min_free_kbytes;
5800 
5801         lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5802         new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5803 
5804         if (new_min_free_kbytes > user_min_free_kbytes) {
5805                 min_free_kbytes = new_min_free_kbytes;
5806                 if (min_free_kbytes < 128)
5807                         min_free_kbytes = 128;
5808                 if (min_free_kbytes > 65536)
5809                         min_free_kbytes = 65536;
5810         } else {
5811                 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5812                                 new_min_free_kbytes, user_min_free_kbytes);
5813         }
5814         setup_per_zone_wmarks();
5815         refresh_zone_stat_thresholds();
5816         setup_per_zone_lowmem_reserve();
5817         setup_per_zone_inactive_ratio();
5818         return 0;
5819 }
5820 module_init(init_per_zone_wmark_min)
5821 
5822 /*
5823  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5824  *      that we can call two helper functions whenever min_free_kbytes
5825  *      changes.
5826  */
5827 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
5828         void __user *buffer, size_t *length, loff_t *ppos)
5829 {
5830         int rc;
5831 
5832         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5833         if (rc)
5834                 return rc;
5835 
5836         if (write) {
5837                 user_min_free_kbytes = min_free_kbytes;
5838                 setup_per_zone_wmarks();
5839         }
5840         return 0;
5841 }
5842 
5843 #ifdef CONFIG_NUMA
5844 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
5845         void __user *buffer, size_t *length, loff_t *ppos)
5846 {
5847         struct zone *zone;
5848         int rc;
5849 
5850         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5851         if (rc)
5852                 return rc;
5853 
5854         for_each_zone(zone)
5855                 zone->min_unmapped_pages = (zone->managed_pages *
5856                                 sysctl_min_unmapped_ratio) / 100;
5857         return 0;
5858 }
5859 
5860 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
5861         void __user *buffer, size_t *length, loff_t *ppos)
5862 {
5863         struct zone *zone;
5864         int rc;
5865 
5866         rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5867         if (rc)
5868                 return rc;
5869 
5870         for_each_zone(zone)
5871                 zone->min_slab_pages = (zone->managed_pages *
5872                                 sysctl_min_slab_ratio) / 100;
5873         return 0;
5874 }
5875 #endif
5876 
5877 /*
5878  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5879  *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5880  *      whenever sysctl_lowmem_reserve_ratio changes.
5881  *
5882  * The reserve ratio obviously has absolutely no relation with the
5883  * minimum watermarks. The lowmem reserve ratio can only make sense
5884  * if in function of the boot time zone sizes.
5885  */
5886 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
5887         void __user *buffer, size_t *length, loff_t *ppos)
5888 {
5889         proc_dointvec_minmax(table, write, buffer, length, ppos);
5890         setup_per_zone_lowmem_reserve();
5891         return 0;
5892 }
5893 
5894 /*
5895  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5896  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
5897  * pagelist can have before it gets flushed back to buddy allocator.
5898  */
5899 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
5900         void __user *buffer, size_t *length, loff_t *ppos)
5901 {
5902         struct zone *zone;
5903         int old_percpu_pagelist_fraction;
5904         int ret;
5905 
5906         mutex_lock(&pcp_batch_high_lock);
5907         old_percpu_pagelist_fraction = percpu_pagelist_fraction;
5908 
5909         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5910         if (!write || ret < 0)
5911                 goto out;
5912 
5913         /* Sanity checking to avoid pcp imbalance */
5914         if (percpu_pagelist_fraction &&
5915             percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
5916                 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
5917                 ret = -EINVAL;
5918                 goto out;
5919         }
5920 
5921         /* No change? */
5922         if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
5923                 goto out;
5924 
5925         for_each_populated_zone(zone) {
5926                 unsigned int cpu;
5927 
5928                 for_each_possible_cpu(cpu)
5929                         pageset_set_high_and_batch(zone,
5930                                         per_cpu_ptr(zone->pageset, cpu));
5931         }
5932 out:
5933         mutex_unlock(&pcp_batch_high_lock);
5934         return ret;
5935 }
5936 
5937 int hashdist = HASHDIST_DEFAULT;
5938 
5939 #ifdef CONFIG_NUMA
5940 static int __init set_hashdist(char *str)
5941 {
5942         if (!str)
5943                 return 0;
5944         hashdist = simple_strtoul(str, &str, 0);
5945         return 1;
5946 }
5947 __setup("hashdist=", set_hashdist);
5948 #endif
5949 
5950 /*
5951  * allocate a large system hash table from bootmem
5952  * - it is assumed that the hash table must contain an exact power-of-2
5953  *   quantity of entries
5954  * - limit is the number of hash buckets, not the total allocation size
5955  */
5956 void *__init alloc_large_system_hash(const char *tablename,
5957                                      unsigned long bucketsize,
5958                                      unsigned long numentries,
5959                                      int scale,
5960                                      int flags,
5961                                      unsigned int *_hash_shift,
5962                                      unsigned int *_hash_mask,
5963                                      unsigned long low_limit,
5964                                      unsigned long high_limit)
5965 {
5966         unsigned long long max = high_limit;
5967         unsigned long log2qty, size;
5968         void *table = NULL;
5969 
5970         /* allow the kernel cmdline to have a say */
5971         if (!numentries) {
5972                 /* round applicable memory size up to nearest megabyte */
5973                 numentries = nr_kernel_pages;
5974 
5975                 /* It isn't necessary when PAGE_SIZE >= 1MB */
5976                 if (PAGE_SHIFT < 20)
5977                         numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
5978 
5979                 /* limit to 1 bucket per 2^scale bytes of low memory */
5980                 if (scale > PAGE_SHIFT)
5981                         numentries >>= (scale - PAGE_SHIFT);
5982                 else
5983                         numentries <<= (PAGE_SHIFT - scale);
5984 
5985                 /* Make sure we've got at least a 0-order allocation.. */
5986                 if (unlikely(flags & HASH_SMALL)) {
5987                         /* Makes no sense without HASH_EARLY */
5988                         WARN_ON(!(flags & HASH_EARLY));
5989                         if (!(numentries >> *_hash_shift)) {
5990                                 numentries = 1UL << *_hash_shift;
5991                                 BUG_ON(!numentries);
5992                         }
5993                 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5994                         numentries = PAGE_SIZE / bucketsize;
5995         }
5996         numentries = roundup_pow_of_two(numentries);
5997 
5998         /* limit allocation size to 1/16 total memory by default */
5999         if (max == 0) {
6000                 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6001                 do_div(max, bucketsize);
6002         }
6003         max = min(max, 0x80000000ULL);
6004 
6005         if (numentries < low_limit)
6006                 numentries = low_limit;
6007         if (numentries > max)
6008                 numentries = max;
6009 
6010         log2qty = ilog2(numentries);
6011 
6012         do {
6013                 size = bucketsize << log2qty;
6014                 if (flags & HASH_EARLY)
6015                         table = memblock_virt_alloc_nopanic(size, 0);
6016                 else if (hashdist)
6017                         table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6018                 else {
6019                         /*
6020                          * If bucketsize is not a power-of-two, we may free
6021                          * some pages at the end of hash table which
6022                          * alloc_pages_exact() automatically does
6023                          */
6024                         if (get_order(size) < MAX_ORDER) {
6025                                 table = alloc_pages_exact(size, GFP_ATOMIC);
6026                                 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6027                         }
6028                 }
6029         } while (!table && size > PAGE_SIZE && --log2qty);
6030 
6031         if (!table)
6032                 panic("Failed to allocate %s hash table\n", tablename);
6033 
6034         printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
6035                tablename,
6036                (1UL << log2qty),
6037                ilog2(size) - PAGE_SHIFT,
6038                size);
6039 
6040         if (_hash_shift)
6041                 *_hash_shift = log2qty;
6042         if (_hash_mask)
6043                 *_hash_mask = (1 << log2qty) - 1;
6044 
6045         return table;
6046 }
6047 
6048 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6049 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6050                                                         unsigned long pfn)
6051 {
6052 #ifdef CONFIG_SPARSEMEM
6053         return __pfn_to_section(pfn)->pageblock_flags;
6054 #else
6055         return zone->pageblock_flags;
6056 #endif /* CONFIG_SPARSEMEM */
6057 }
6058 
6059 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6060 {
6061 #ifdef CONFIG_SPARSEMEM
6062         pfn &= (PAGES_PER_SECTION-1);
6063         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6064 #else
6065         pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
6066         return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6067 #endif /* CONFIG_SPARSEMEM */
6068 }
6069 
6070 /**
6071  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6072  * @page: The page within the block of interest
6073  * @pfn: The target page frame number
6074  * @end_bitidx: The last bit of interest to retrieve
6075  * @mask: mask of bits that the caller is interested in
6076  *
6077  * Return: pageblock_bits flags
6078  */
6079 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
6080                                         unsigned long end_bitidx,
6081                                         unsigned long mask)
6082 {
6083         struct zone *zone;
6084         unsigned long *bitmap;
6085         unsigned long bitidx, word_bitidx;
6086         unsigned long word;
6087 
6088         zone = page_zone(page);
6089         bitmap = get_pageblock_bitmap(zone, pfn);
6090         bitidx = pfn_to_bitidx(zone, pfn);
6091         word_bitidx = bitidx / BITS_PER_LONG;
6092         bitidx &= (BITS_PER_LONG-1);
6093 
6094         word = bitmap[word_bitidx];
6095         bitidx += end_bitidx;
6096         return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
6097 }
6098 
6099 /**
6100  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6101  * @page: The page within the block of interest
6102  * @flags: The flags to set
6103  * @pfn: The target page frame number
6104  * @end_bitidx: The last bit of interest
6105  * @mask: mask of bits that the caller is interested in
6106  */
6107 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6108                                         unsigned long pfn,
6109                                         unsigned long end_bitidx,
6110                                         unsigned long mask)
6111 {
6112         struct zone *zone;
6113         unsigned long *bitmap;
6114         unsigned long bitidx, word_bitidx;
6115         unsigned long old_word, word;
6116 
6117         BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
6118 
6119         zone = page_zone(page);
6120         bitmap = get_pageblock_bitmap(zone, pfn);
6121         bitidx = pfn_to_bitidx(zone, pfn);
6122         word_bitidx = bitidx / BITS_PER_LONG;
6123         bitidx &= (BITS_PER_LONG-1);
6124 
6125         VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
6126 
6127         bitidx += end_bitidx;
6128         mask <<= (BITS_PER_LONG - bitidx - 1);
6129         flags <<= (BITS_PER_LONG - bitidx - 1);
6130 
6131         word = ACCESS_ONCE(bitmap[word_bitidx]);
6132         for (;;) {
6133                 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6134                 if (word == old_word)
6135                         break;
6136                 word = old_word;
6137         }
6138 }
6139 
6140 /*
6141  * This function checks whether pageblock includes unmovable pages or not.
6142  * If @count is not zero, it is okay to include less @count unmovable pages
6143  *
6144  * PageLRU check without isolation or lru_lock could race so that
6145  * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6146  * expect this function should be exact.
6147  */
6148 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6149                          bool skip_hwpoisoned_pages)
6150 {
6151         unsigned long pfn, iter, found;
6152         int mt;
6153 
6154         /*
6155          * For avoiding noise data, lru_add_drain_all() should be called
6156          * If ZONE_MOVABLE, the zone never contains unmovable pages
6157          */
6158         if (zone_idx(zone) == ZONE_MOVABLE)
6159                 return false;
6160         mt = get_pageblock_migratetype(page);
6161         if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6162                 return false;
6163 
6164         pfn = page_to_pfn(page);
6165         for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6166                 unsigned long check = pfn + iter;
6167 
6168                 if (!pfn_valid_within(check))
6169                         continue;
6170 
6171                 page = pfn_to_page(check);
6172 
6173                 /*
6174                  * Hugepages are not in LRU lists, but they're movable.
6175                  * We need not scan over tail pages bacause we don't
6176                  * handle each tail page individually in migration.
6177                  */
6178                 if (PageHuge(page)) {
6179                         iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6180                         continue;
6181                 }
6182 
6183                 /*
6184                  * We can't use page_count without pin a page
6185                  * because another CPU can free compound page.
6186                  * This check already skips compound tails of THP
6187                  * because their page->_count is zero at all time.
6188                  */
6189                 if (!atomic_read(&page->_count)) {
6190                         if (PageBuddy(page))
6191                                 iter += (1 << page_order(page)) - 1;
6192                         continue;
6193                 }
6194 
6195                 /*
6196                  * The HWPoisoned page may be not in buddy system, and
6197                  * page_count() is not 0.
6198                  */
6199                 if (skip_hwpoisoned_pages && PageHWPoison(page))
6200                         continue;
6201 
6202                 if (!PageLRU(page))
6203                         found++;
6204                 /*
6205                  * If there are RECLAIMABLE pages, we need to check it.
6206                  * But now, memory offline itself doesn't call shrink_slab()
6207                  * and it still to be fixed.
6208                  */
6209                 /*
6210                  * If the page is not RAM, page_count()should be 0.
6211                  * we don't need more check. This is an _used_ not-movable page.
6212                  *
6213                  * The problematic thing here is PG_reserved pages. PG_reserved
6214                  * is set to both of a memory hole page and a _used_ kernel
6215                  * page at boot.
6216                  */
6217                 if (found > count)
6218                         return true;
6219         }
6220         return false;
6221 }
6222 
6223 bool is_pageblock_removable_nolock(struct page *page)
6224 {
6225         struct zone *zone;
6226         unsigned long pfn;
6227 
6228         /*
6229          * We have to be careful here because we are iterating over memory
6230          * sections which are not zone aware so we might end up outside of
6231          * the zone but still within the section.
6232          * We have to take care about the node as well. If the node is offline
6233          * its NODE_DATA will be NULL - see page_zone.
6234          */
6235         if (!node_online(page_to_nid(page)))
6236                 return false;
6237 
6238         zone = page_zone(page);
6239         pfn = page_to_pfn(page);
6240         if (!zone_spans_pfn(zone, pfn))
6241                 return false;
6242 
6243         return !has_unmovable_pages(zone, page, 0, true);
6244 }
6245 
6246 #ifdef CONFIG_CMA
6247 
6248 static unsigned long pfn_max_align_down(unsigned long pfn)
6249 {
6250         return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6251                              pageblock_nr_pages) - 1);
6252 }
6253 
6254 static unsigned long pfn_max_align_up(unsigned long pfn)
6255 {
6256         return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6257                                 pageblock_nr_pages));
6258 }
6259 
6260 /* [start, end) must belong to a single zone. */
6261 static int __alloc_contig_migrate_range(struct compact_control *cc,
6262                                         unsigned long start, unsigned long end)
6263 {
6264         /* This function is based on compact_zone() from compaction.c. */
6265         unsigned long nr_reclaimed;
6266         unsigned long pfn = start;
6267         unsigned int tries = 0;
6268         int ret = 0;
6269 
6270         migrate_prep();
6271 
6272         while (pfn < end || !list_empty(&cc->migratepages)) {
6273                 if (fatal_signal_pending(current)) {
6274                         ret = -EINTR;
6275                         break;
6276                 }
6277 
6278                 if (list_empty(&cc->migratepages)) {
6279                         cc->nr_migratepages = 0;
6280                         pfn = isolate_migratepages_range(cc->zone, cc,
6281                                                          pfn, end, true);
6282                         if (!pfn) {
6283                                 ret = -EINTR;
6284                                 break;
6285                         }
6286                         tries = 0;
6287                 } else if (++tries == 5) {
6288                         ret = ret < 0 ? ret : -EBUSY;
6289                         break;
6290                 }
6291 
6292                 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6293                                                         &cc->migratepages);
6294                 cc->nr_migratepages -= nr_reclaimed;
6295 
6296                 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6297                                     NULL, 0, cc->mode, MR_CMA);
6298         }
6299         if (ret < 0) {
6300                 putback_movable_pages(&cc->migratepages);
6301                 return ret;
6302         }
6303         return 0;
6304 }
6305 
6306 /**
6307  * alloc_contig_range() -- tries to allocate given range of pages
6308  * @start:      start PFN to allocate
6309  * @end:        one-past-the-last PFN to allocate
6310  * @migratetype:        migratetype of the underlaying pageblocks (either
6311  *                      #MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
6312  *                      in range must have the same migratetype and it must
6313  *                      be either of the two.
6314  *
6315  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6316  * aligned, however it's the caller's responsibility to guarantee that
6317  * we are the only thread that changes migrate type of pageblocks the
6318  * pages fall in.
6319  *
6320  * The PFN range must belong to a single zone.
6321  *
6322  * Returns zero on success or negative error code.  On success all
6323  * pages which PFN is in [start, end) are allocated for the caller and
6324  * need to be freed with free_contig_range().
6325  */
6326 int alloc_contig_range(unsigned long start, unsigned long end,
6327                        unsigned migratetype)
6328 {
6329         unsigned long outer_start, outer_end;
6330         int ret = 0, order;
6331 
6332         struct compact_control cc = {
6333                 .nr_migratepages = 0,
6334                 .order = -1,
6335                 .zone = page_zone(pfn_to_page(start)),
6336                 .mode = MIGRATE_SYNC,
6337                 .ignore_skip_hint = true,
6338         };
6339         INIT_LIST_HEAD(&cc.migratepages);
6340 
6341         /*
6342          * What we do here is we mark all pageblocks in range as
6343          * MIGRATE_ISOLATE.  Because pageblock and max order pages may
6344          * have different sizes, and due to the way page allocator
6345          * work, we align the range to biggest of the two pages so
6346          * that page allocator won't try to merge buddies from
6347          * different pageblocks and change MIGRATE_ISOLATE to some
6348          * other migration type.
6349          *
6350          * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6351          * migrate the pages from an unaligned range (ie. pages that
6352          * we are interested in).  This will put all the pages in
6353          * range back to page allocator as MIGRATE_ISOLATE.
6354          *
6355          * When this is done, we take the pages in range from page
6356          * allocator removing them from the buddy system.  This way
6357          * page allocator will never consider using them.
6358          *
6359          * This lets us mark the pageblocks back as
6360          * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6361          * aligned range but not in the unaligned, original range are
6362          * put back to page allocator so that buddy can use them.
6363          */
6364 
6365         ret = start_isolate_page_range(pfn_max_align_down(start),
6366                                        pfn_max_align_up(end), migratetype,
6367                                        false);
6368         if (ret)
6369                 return ret;
6370 
6371         ret = __alloc_contig_migrate_range(&cc, start, end);
6372         if (ret)
6373                 goto done;
6374 
6375         /*
6376          * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6377          * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
6378          * more, all pages in [start, end) are free in page allocator.
6379          * What we are going to do is to allocate all pages from
6380          * [start, end) (that is remove them from page allocator).
6381          *
6382          * The only problem is that pages at the beginning and at the
6383          * end of interesting range may be not aligned with pages that
6384          * page allocator holds, ie. they can be part of higher order
6385          * pages.  Because of this, we reserve the bigger range and
6386          * once this is done free the pages we are not interested in.
6387          *
6388          * We don't have to hold zone->lock here because the pages are
6389          * isolated thus they won't get removed from buddy.
6390          */
6391 
6392         lru_add_drain_all();
6393         drain_all_pages();
6394 
6395         order = 0;
6396         outer_start = start;
6397         while (!PageBuddy(pfn_to_page(outer_start))) {
6398                 if (++order >= MAX_ORDER) {
6399                         ret = -EBUSY;
6400                         goto done;
6401                 }
6402                 outer_start &= ~0UL << order;
6403         }
6404 
6405         /* Make sure the range is really isolated. */
6406         if (test_pages_isolated(outer_start, end, false)) {
6407                 pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6408                        outer_start, end);
6409                 ret = -EBUSY;
6410                 goto done;
6411         }
6412 
6413 
6414         /* Grab isolated pages from freelists. */
6415         outer_end = isolate_freepages_range(&cc, outer_start, end);
6416         if (!outer_end) {
6417                 ret = -EBUSY;
6418                 goto done;
6419         }
6420 
6421         /* Free head and tail (if any) */
6422         if (start != outer_start)
6423                 free_contig_range(outer_start, start - outer_start);
6424         if (end != outer_end)
6425                 free_contig_range(end, outer_end - end);
6426 
6427 done:
6428         undo_isolate_page_range(pfn_max_align_down(start),
6429                                 pfn_max_align_up(end), migratetype);
6430         return ret;
6431 }
6432 
6433 void free_contig_range(unsigned long pfn, unsigned nr_pages)
6434 {
6435         unsigned int count = 0;
6436 
6437         for (; nr_pages--; pfn++) {
6438                 struct page *page = pfn_to_page(pfn);
6439 
6440                 count += page_count(page) != 1;
6441                 __free_page(page);
6442         }
6443         WARN(count != 0, "%d pages are still in use!\n", count);
6444 }
6445 #endif
6446 
6447 #ifdef CONFIG_MEMORY_HOTPLUG
6448 /*
6449  * The zone indicated has a new number of managed_pages; batch sizes and percpu
6450  * page high values need to be recalulated.
6451  */
6452 void __meminit zone_pcp_update(struct zone *zone)
6453 {
6454         unsigned cpu;
6455         mutex_lock(&pcp_batch_high_lock);
6456         for_each_possible_cpu(cpu)
6457                 pageset_set_high_and_batch(zone,
6458                                 per_cpu_ptr(zone->pageset, cpu));
6459         mutex_unlock(&pcp_batch_high_lock);
6460 }
6461 #endif
6462 
6463 void zone_pcp_reset(struct zone *zone)
6464 {
6465         unsigned long flags;
6466         int cpu;
6467         struct per_cpu_pageset *pset;
6468 
6469         /* avoid races with drain_pages()  */
6470         local_irq_save(flags);
6471         if (zone->pageset != &boot_pageset) {
6472                 for_each_online_cpu(cpu) {
6473                         pset = per_cpu_ptr(zone->pageset, cpu);
6474                         drain_zonestat(zone, pset);
6475                 }
6476                 free_percpu(zone->pageset);
6477                 zone->pageset = &boot_pageset;
6478         }
6479         local_irq_restore(flags);
6480 }
6481 
6482 #ifdef CONFIG_MEMORY_HOTREMOVE
6483 /*
6484  * All pages in the range must be isolated before calling this.
6485  */
6486 void
6487 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6488 {
6489         struct page *page;
6490         struct zone *zone;
6491         unsigned int order, i;
6492         unsigned long pfn;
6493         unsigned long flags;
6494         /* find the first valid pfn */
6495         for (pfn = start_pfn; pfn < end_pfn; pfn++)
6496                 if (pfn_valid(pfn))
6497                         break;
6498         if (pfn == end_pfn)
6499                 return;
6500         zone = page_zone(pfn_to_page(pfn));
6501         spin_lock_irqsave(&zone->lock, flags);
6502         pfn = start_pfn;
6503         while (pfn < end_pfn) {
6504                 if (!pfn_valid(pfn)) {
6505                         pfn++;
6506                         continue;
6507                 }
6508                 page = pfn_to_page(pfn);
6509                 /*
6510                  * The HWPoisoned page may be not in buddy system, and
6511                  * page_count() is not 0.
6512                  */
6513                 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6514                         pfn++;
6515                         SetPageReserved(page);
6516                         continue;
6517                 }
6518 
6519                 BUG_ON(page_count(page));
6520                 BUG_ON(!PageBuddy(page));
6521                 order = page_order(page);
6522 #ifdef CONFIG_DEBUG_VM
6523                 printk(KERN_INFO "remove from free list %lx %d %lx\n",
6524                        pfn, 1 << order, end_pfn);
6525 #endif
6526                 list_del(&page->lru);
6527                 rmv_page_order(page);
6528                 zone->free_area[order].nr_free--;
6529                 for (i = 0; i < (1 << order); i++)
6530                         SetPageReserved((page+i));
6531                 pfn += (1 << order);
6532         }
6533         spin_unlock_irqrestore(&zone->lock, flags);
6534 }
6535 #endif
6536 
6537 #ifdef CONFIG_MEMORY_FAILURE
6538 bool is_free_buddy_page(struct page *page)
6539 {
6540         struct zone *zone = page_zone(page);
6541         unsigned long pfn = page_to_pfn(page);
6542         unsigned long flags;
6543         unsigned int order;
6544 
6545         spin_lock_irqsave(&zone->lock, flags);
6546         for (order = 0; order < MAX_ORDER; order++) {
6547                 struct page *page_head = page - (pfn & ((1 << order) - 1));
6548 
6549                 if (PageBuddy(page_head) && page_order(page_head) >= order)
6550                         break;
6551         }
6552         spin_unlock_irqrestore(&zone->lock, flags);
6553 
6554         return order < MAX_ORDER;
6555 }
6556 #endif
6557 
6558 static const struct trace_print_flags pageflag_names[] = {
6559         {1UL << PG_locked,              "locked"        },
6560         {1UL << PG_error,               "error"         },
6561         {1UL << PG_referenced,          "referenced"    },
6562         {1UL << PG_uptodate,            "uptodate"      },
6563         {1UL << PG_dirty,               "dirty"         },
6564         {1UL << PG_lru,                 "lru"           },
6565         {1UL << PG_active,              "active"        },
6566         {1UL << PG_slab,                "slab"          },
6567         {1UL << PG_owner_priv_1,        "owner_priv_1"  },
6568         {1UL << PG_arch_1,              "arch_1"        },
6569         {1UL << PG_reserved,            "reserved"      },
6570         {1UL << PG_private,             "private"       },
6571         {1UL << PG_private_2,           "private_2"     },
6572         {1UL << PG_writeback,           "writeback"     },
6573 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6574         {1UL << PG_head,                "head"          },
6575         {1UL << PG_tail,                "tail"          },
6576 #else
6577         {1UL << PG_compound,            "compound"      },
6578 #endif
6579         {1UL << PG_swapcache,           "swapcache"     },
6580         {1UL << PG_mappedtodisk,        "mappedtodisk"  },
6581         {1UL << PG_reclaim,             "reclaim"       },
6582         {1UL << PG_swapbacked,          "swapbacked"    },
6583         {1UL << PG_unevictable,         "unevictable"   },
6584 #ifdef CONFIG_MMU
6585         {1UL << PG_mlocked,             "mlocked"       },
6586 #endif
6587 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6588         {1UL << PG_uncached,            "uncached"      },
6589 #endif
6590 #ifdef CONFIG_MEMORY_FAILURE
6591         {1UL << PG_hwpoison,            "hwpoison"      },
6592 #endif
6593 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6594         {1UL << PG_compound_lock,       "compound_lock" },
6595 #endif
6596 };
6597 
6598 static void dump_page_flags(unsigned long flags)
6599 {
6600         const char *delim = "";
6601         unsigned long mask;
6602         int i;
6603 
6604         BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
6605 
6606         printk(KERN_ALERT "page flags: %#lx(", flags);
6607 
6608         /* remove zone id */
6609         flags &= (1UL << NR_PAGEFLAGS) - 1;
6610 
6611         for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
6612 
6613                 mask = pageflag_names[i].mask;
6614                 if ((flags & mask) != mask)
6615                         continue;
6616 
6617                 flags &= ~mask;
6618                 printk("%s%s", delim, pageflag_names[i].name);
6619                 delim = "|";
6620         }
6621 
6622         /* check for left over flags */
6623         if (flags)
6624                 printk("%s%#lx", delim, flags);
6625 
6626         printk(")\n");
6627 }
6628 
6629 void dump_page_badflags(struct page *page, const char *reason,
6630                 unsigned long badflags)
6631 {
6632         printk(KERN_ALERT
6633                "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6634                 page, atomic_read(&page->_count), page_mapcount(page),
6635                 page->mapping, page->index);
6636         dump_page_flags(page->flags);
6637         if (reason)
6638                 pr_alert("page dumped because: %s\n", reason);
6639         if (page->flags & badflags) {
6640                 pr_alert("bad because of flags:\n");
6641                 dump_page_flags(page->flags & badflags);
6642         }
6643         mem_cgroup_print_bad_page(page);
6644 }
6645 
6646 void dump_page(struct page *page, const char *reason)
6647 {
6648         dump_page_badflags(page, reason, 0);
6649 }
6650 EXPORT_SYMBOL(dump_page);
6651 

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