Version:  2.0.40 2.2.26 2.4.37 2.6.39 3.0 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15

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

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