xref: /openbmc/linux/mm/page_alloc.c (revision a2fb4d78)
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, char *reason, unsigned long bad_flags)
299 {
300 	static unsigned long resume;
301 	static unsigned long nr_shown;
302 	static unsigned long nr_unshown;
303 
304 	/* Don't complain about poisoned pages */
305 	if (PageHWPoison(page)) {
306 		page_mapcount_reset(page); /* remove PageBuddy */
307 		return;
308 	}
309 
310 	/*
311 	 * Allow a burst of 60 reports, then keep quiet for that minute;
312 	 * or allow a steady drip of one report per second.
313 	 */
314 	if (nr_shown == 60) {
315 		if (time_before(jiffies, resume)) {
316 			nr_unshown++;
317 			goto out;
318 		}
319 		if (nr_unshown) {
320 			printk(KERN_ALERT
321 			      "BUG: Bad page state: %lu messages suppressed\n",
322 				nr_unshown);
323 			nr_unshown = 0;
324 		}
325 		nr_shown = 0;
326 	}
327 	if (nr_shown++ == 0)
328 		resume = jiffies + 60 * HZ;
329 
330 	printk(KERN_ALERT "BUG: Bad page state in process %s  pfn:%05lx\n",
331 		current->comm, page_to_pfn(page));
332 	dump_page_badflags(page, reason, bad_flags);
333 
334 	print_modules();
335 	dump_stack();
336 out:
337 	/* Leave bad fields for debug, except PageBuddy could make trouble */
338 	page_mapcount_reset(page); /* remove PageBuddy */
339 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
340 }
341 
342 /*
343  * Higher-order pages are called "compound pages".  They are structured thusly:
344  *
345  * The first PAGE_SIZE page is called the "head page".
346  *
347  * The remaining PAGE_SIZE pages are called "tail pages".
348  *
349  * All pages have PG_compound set.  All tail pages have their ->first_page
350  * pointing at the head page.
351  *
352  * The first tail page's ->lru.next holds the address of the compound page's
353  * put_page() function.  Its ->lru.prev holds the order of allocation.
354  * This usage means that zero-order pages may not be compound.
355  */
356 
357 static void free_compound_page(struct page *page)
358 {
359 	__free_pages_ok(page, compound_order(page));
360 }
361 
362 void prep_compound_page(struct page *page, unsigned long order)
363 {
364 	int i;
365 	int nr_pages = 1 << order;
366 
367 	set_compound_page_dtor(page, free_compound_page);
368 	set_compound_order(page, order);
369 	__SetPageHead(page);
370 	for (i = 1; i < nr_pages; i++) {
371 		struct page *p = page + i;
372 		set_page_count(p, 0);
373 		p->first_page = page;
374 		/* Make sure p->first_page is always valid for PageTail() */
375 		smp_wmb();
376 		__SetPageTail(p);
377 	}
378 }
379 
380 /* update __split_huge_page_refcount if you change this function */
381 static int destroy_compound_page(struct page *page, unsigned long order)
382 {
383 	int i;
384 	int nr_pages = 1 << order;
385 	int bad = 0;
386 
387 	if (unlikely(compound_order(page) != order)) {
388 		bad_page(page, "wrong compound order", 0);
389 		bad++;
390 	}
391 
392 	__ClearPageHead(page);
393 
394 	for (i = 1; i < nr_pages; i++) {
395 		struct page *p = page + i;
396 
397 		if (unlikely(!PageTail(p))) {
398 			bad_page(page, "PageTail not set", 0);
399 			bad++;
400 		} else if (unlikely(p->first_page != page)) {
401 			bad_page(page, "first_page not consistent", 0);
402 			bad++;
403 		}
404 		__ClearPageTail(p);
405 	}
406 
407 	return bad;
408 }
409 
410 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
411 {
412 	int i;
413 
414 	/*
415 	 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
416 	 * and __GFP_HIGHMEM from hard or soft interrupt context.
417 	 */
418 	VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
419 	for (i = 0; i < (1 << order); i++)
420 		clear_highpage(page + i);
421 }
422 
423 #ifdef CONFIG_DEBUG_PAGEALLOC
424 unsigned int _debug_guardpage_minorder;
425 
426 static int __init debug_guardpage_minorder_setup(char *buf)
427 {
428 	unsigned long res;
429 
430 	if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
431 		printk(KERN_ERR "Bad debug_guardpage_minorder value\n");
432 		return 0;
433 	}
434 	_debug_guardpage_minorder = res;
435 	printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res);
436 	return 0;
437 }
438 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
439 
440 static inline void set_page_guard_flag(struct page *page)
441 {
442 	__set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
443 }
444 
445 static inline void clear_page_guard_flag(struct page *page)
446 {
447 	__clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags);
448 }
449 #else
450 static inline void set_page_guard_flag(struct page *page) { }
451 static inline void clear_page_guard_flag(struct page *page) { }
452 #endif
453 
454 static inline void set_page_order(struct page *page, int order)
455 {
456 	set_page_private(page, order);
457 	__SetPageBuddy(page);
458 }
459 
460 static inline void rmv_page_order(struct page *page)
461 {
462 	__ClearPageBuddy(page);
463 	set_page_private(page, 0);
464 }
465 
466 /*
467  * Locate the struct page for both the matching buddy in our
468  * pair (buddy1) and the combined O(n+1) page they form (page).
469  *
470  * 1) Any buddy B1 will have an order O twin B2 which satisfies
471  * the following equation:
472  *     B2 = B1 ^ (1 << O)
473  * For example, if the starting buddy (buddy2) is #8 its order
474  * 1 buddy is #10:
475  *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
476  *
477  * 2) Any buddy B will have an order O+1 parent P which
478  * satisfies the following equation:
479  *     P = B & ~(1 << O)
480  *
481  * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
482  */
483 static inline unsigned long
484 __find_buddy_index(unsigned long page_idx, unsigned int order)
485 {
486 	return page_idx ^ (1 << order);
487 }
488 
489 /*
490  * This function checks whether a page is free && is the buddy
491  * we can do coalesce a page and its buddy if
492  * (a) the buddy is not in a hole &&
493  * (b) the buddy is in the buddy system &&
494  * (c) a page and its buddy have the same order &&
495  * (d) a page and its buddy are in the same zone.
496  *
497  * For recording whether a page is in the buddy system, we set ->_mapcount
498  * PAGE_BUDDY_MAPCOUNT_VALUE.
499  * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
500  * serialized by zone->lock.
501  *
502  * For recording page's order, we use page_private(page).
503  */
504 static inline int page_is_buddy(struct page *page, struct page *buddy,
505 								int order)
506 {
507 	if (!pfn_valid_within(page_to_pfn(buddy)))
508 		return 0;
509 
510 	if (page_zone_id(page) != page_zone_id(buddy))
511 		return 0;
512 
513 	if (page_is_guard(buddy) && page_order(buddy) == order) {
514 		VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
515 		return 1;
516 	}
517 
518 	if (PageBuddy(buddy) && page_order(buddy) == order) {
519 		VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
520 		return 1;
521 	}
522 	return 0;
523 }
524 
525 /*
526  * Freeing function for a buddy system allocator.
527  *
528  * The concept of a buddy system is to maintain direct-mapped table
529  * (containing bit values) for memory blocks of various "orders".
530  * The bottom level table contains the map for the smallest allocatable
531  * units of memory (here, pages), and each level above it describes
532  * pairs of units from the levels below, hence, "buddies".
533  * At a high level, all that happens here is marking the table entry
534  * at the bottom level available, and propagating the changes upward
535  * as necessary, plus some accounting needed to play nicely with other
536  * parts of the VM system.
537  * At each level, we keep a list of pages, which are heads of continuous
538  * free pages of length of (1 << order) and marked with _mapcount
539  * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
540  * field.
541  * So when we are allocating or freeing one, we can derive the state of the
542  * other.  That is, if we allocate a small block, and both were
543  * free, the remainder of the region must be split into blocks.
544  * If a block is freed, and its buddy is also free, then this
545  * triggers coalescing into a block of larger size.
546  *
547  * -- nyc
548  */
549 
550 static inline void __free_one_page(struct page *page,
551 		struct zone *zone, unsigned int order,
552 		int migratetype)
553 {
554 	unsigned long page_idx;
555 	unsigned long combined_idx;
556 	unsigned long uninitialized_var(buddy_idx);
557 	struct page *buddy;
558 
559 	VM_BUG_ON(!zone_is_initialized(zone));
560 
561 	if (unlikely(PageCompound(page)))
562 		if (unlikely(destroy_compound_page(page, order)))
563 			return;
564 
565 	VM_BUG_ON(migratetype == -1);
566 
567 	page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
568 
569 	VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
570 	VM_BUG_ON_PAGE(bad_range(zone, page), page);
571 
572 	while (order < MAX_ORDER-1) {
573 		buddy_idx = __find_buddy_index(page_idx, order);
574 		buddy = page + (buddy_idx - page_idx);
575 		if (!page_is_buddy(page, buddy, order))
576 			break;
577 		/*
578 		 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
579 		 * merge with it and move up one order.
580 		 */
581 		if (page_is_guard(buddy)) {
582 			clear_page_guard_flag(buddy);
583 			set_page_private(page, 0);
584 			__mod_zone_freepage_state(zone, 1 << order,
585 						  migratetype);
586 		} else {
587 			list_del(&buddy->lru);
588 			zone->free_area[order].nr_free--;
589 			rmv_page_order(buddy);
590 		}
591 		combined_idx = buddy_idx & page_idx;
592 		page = page + (combined_idx - page_idx);
593 		page_idx = combined_idx;
594 		order++;
595 	}
596 	set_page_order(page, order);
597 
598 	/*
599 	 * If this is not the largest possible page, check if the buddy
600 	 * of the next-highest order is free. If it is, it's possible
601 	 * that pages are being freed that will coalesce soon. In case,
602 	 * that is happening, add the free page to the tail of the list
603 	 * so it's less likely to be used soon and more likely to be merged
604 	 * as a higher order page
605 	 */
606 	if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
607 		struct page *higher_page, *higher_buddy;
608 		combined_idx = buddy_idx & page_idx;
609 		higher_page = page + (combined_idx - page_idx);
610 		buddy_idx = __find_buddy_index(combined_idx, order + 1);
611 		higher_buddy = higher_page + (buddy_idx - combined_idx);
612 		if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
613 			list_add_tail(&page->lru,
614 				&zone->free_area[order].free_list[migratetype]);
615 			goto out;
616 		}
617 	}
618 
619 	list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
620 out:
621 	zone->free_area[order].nr_free++;
622 }
623 
624 static inline int free_pages_check(struct page *page)
625 {
626 	char *bad_reason = NULL;
627 	unsigned long bad_flags = 0;
628 
629 	if (unlikely(page_mapcount(page)))
630 		bad_reason = "nonzero mapcount";
631 	if (unlikely(page->mapping != NULL))
632 		bad_reason = "non-NULL mapping";
633 	if (unlikely(atomic_read(&page->_count) != 0))
634 		bad_reason = "nonzero _count";
635 	if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
636 		bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
637 		bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
638 	}
639 	if (unlikely(mem_cgroup_bad_page_check(page)))
640 		bad_reason = "cgroup check failed";
641 	if (unlikely(bad_reason)) {
642 		bad_page(page, bad_reason, bad_flags);
643 		return 1;
644 	}
645 	page_cpupid_reset_last(page);
646 	if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
647 		page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
648 	return 0;
649 }
650 
651 /*
652  * Frees a number of pages from the PCP lists
653  * Assumes all pages on list are in same zone, and of same order.
654  * count is the number of pages to free.
655  *
656  * If the zone was previously in an "all pages pinned" state then look to
657  * see if this freeing clears that state.
658  *
659  * And clear the zone's pages_scanned counter, to hold off the "all pages are
660  * pinned" detection logic.
661  */
662 static void free_pcppages_bulk(struct zone *zone, int count,
663 					struct per_cpu_pages *pcp)
664 {
665 	int migratetype = 0;
666 	int batch_free = 0;
667 	int to_free = count;
668 
669 	spin_lock(&zone->lock);
670 	zone->pages_scanned = 0;
671 
672 	while (to_free) {
673 		struct page *page;
674 		struct list_head *list;
675 
676 		/*
677 		 * Remove pages from lists in a round-robin fashion. A
678 		 * batch_free count is maintained that is incremented when an
679 		 * empty list is encountered.  This is so more pages are freed
680 		 * off fuller lists instead of spinning excessively around empty
681 		 * lists
682 		 */
683 		do {
684 			batch_free++;
685 			if (++migratetype == MIGRATE_PCPTYPES)
686 				migratetype = 0;
687 			list = &pcp->lists[migratetype];
688 		} while (list_empty(list));
689 
690 		/* This is the only non-empty list. Free them all. */
691 		if (batch_free == MIGRATE_PCPTYPES)
692 			batch_free = to_free;
693 
694 		do {
695 			int mt;	/* migratetype of the to-be-freed page */
696 
697 			page = list_entry(list->prev, struct page, lru);
698 			/* must delete as __free_one_page list manipulates */
699 			list_del(&page->lru);
700 			mt = get_freepage_migratetype(page);
701 			/* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
702 			__free_one_page(page, zone, 0, mt);
703 			trace_mm_page_pcpu_drain(page, 0, mt);
704 			if (likely(!is_migrate_isolate_page(page))) {
705 				__mod_zone_page_state(zone, NR_FREE_PAGES, 1);
706 				if (is_migrate_cma(mt))
707 					__mod_zone_page_state(zone, NR_FREE_CMA_PAGES, 1);
708 			}
709 		} while (--to_free && --batch_free && !list_empty(list));
710 	}
711 	spin_unlock(&zone->lock);
712 }
713 
714 static void free_one_page(struct zone *zone, struct page *page, int order,
715 				int migratetype)
716 {
717 	spin_lock(&zone->lock);
718 	zone->pages_scanned = 0;
719 
720 	__free_one_page(page, zone, order, migratetype);
721 	if (unlikely(!is_migrate_isolate(migratetype)))
722 		__mod_zone_freepage_state(zone, 1 << order, migratetype);
723 	spin_unlock(&zone->lock);
724 }
725 
726 static bool free_pages_prepare(struct page *page, unsigned int order)
727 {
728 	int i;
729 	int bad = 0;
730 
731 	trace_mm_page_free(page, order);
732 	kmemcheck_free_shadow(page, order);
733 
734 	if (PageAnon(page))
735 		page->mapping = NULL;
736 	for (i = 0; i < (1 << order); i++)
737 		bad += free_pages_check(page + i);
738 	if (bad)
739 		return false;
740 
741 	if (!PageHighMem(page)) {
742 		debug_check_no_locks_freed(page_address(page),
743 					   PAGE_SIZE << order);
744 		debug_check_no_obj_freed(page_address(page),
745 					   PAGE_SIZE << order);
746 	}
747 	arch_free_page(page, order);
748 	kernel_map_pages(page, 1 << order, 0);
749 
750 	return true;
751 }
752 
753 static void __free_pages_ok(struct page *page, unsigned int order)
754 {
755 	unsigned long flags;
756 	int migratetype;
757 
758 	if (!free_pages_prepare(page, order))
759 		return;
760 
761 	local_irq_save(flags);
762 	__count_vm_events(PGFREE, 1 << order);
763 	migratetype = get_pageblock_migratetype(page);
764 	set_freepage_migratetype(page, migratetype);
765 	free_one_page(page_zone(page), page, order, migratetype);
766 	local_irq_restore(flags);
767 }
768 
769 void __init __free_pages_bootmem(struct page *page, unsigned int order)
770 {
771 	unsigned int nr_pages = 1 << order;
772 	struct page *p = page;
773 	unsigned int loop;
774 
775 	prefetchw(p);
776 	for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
777 		prefetchw(p + 1);
778 		__ClearPageReserved(p);
779 		set_page_count(p, 0);
780 	}
781 	__ClearPageReserved(p);
782 	set_page_count(p, 0);
783 
784 	page_zone(page)->managed_pages += nr_pages;
785 	set_page_refcounted(page);
786 	__free_pages(page, order);
787 }
788 
789 #ifdef CONFIG_CMA
790 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
791 void __init init_cma_reserved_pageblock(struct page *page)
792 {
793 	unsigned i = pageblock_nr_pages;
794 	struct page *p = page;
795 
796 	do {
797 		__ClearPageReserved(p);
798 		set_page_count(p, 0);
799 	} while (++p, --i);
800 
801 	set_page_refcounted(page);
802 	set_pageblock_migratetype(page, MIGRATE_CMA);
803 	__free_pages(page, pageblock_order);
804 	adjust_managed_page_count(page, pageblock_nr_pages);
805 }
806 #endif
807 
808 /*
809  * The order of subdivision here is critical for the IO subsystem.
810  * Please do not alter this order without good reasons and regression
811  * testing. Specifically, as large blocks of memory are subdivided,
812  * the order in which smaller blocks are delivered depends on the order
813  * they're subdivided in this function. This is the primary factor
814  * influencing the order in which pages are delivered to the IO
815  * subsystem according to empirical testing, and this is also justified
816  * by considering the behavior of a buddy system containing a single
817  * large block of memory acted on by a series of small allocations.
818  * This behavior is a critical factor in sglist merging's success.
819  *
820  * -- nyc
821  */
822 static inline void expand(struct zone *zone, struct page *page,
823 	int low, int high, struct free_area *area,
824 	int migratetype)
825 {
826 	unsigned long size = 1 << high;
827 
828 	while (high > low) {
829 		area--;
830 		high--;
831 		size >>= 1;
832 		VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
833 
834 #ifdef CONFIG_DEBUG_PAGEALLOC
835 		if (high < debug_guardpage_minorder()) {
836 			/*
837 			 * Mark as guard pages (or page), that will allow to
838 			 * merge back to allocator when buddy will be freed.
839 			 * Corresponding page table entries will not be touched,
840 			 * pages will stay not present in virtual address space
841 			 */
842 			INIT_LIST_HEAD(&page[size].lru);
843 			set_page_guard_flag(&page[size]);
844 			set_page_private(&page[size], high);
845 			/* Guard pages are not available for any usage */
846 			__mod_zone_freepage_state(zone, -(1 << high),
847 						  migratetype);
848 			continue;
849 		}
850 #endif
851 		list_add(&page[size].lru, &area->free_list[migratetype]);
852 		area->nr_free++;
853 		set_page_order(&page[size], high);
854 	}
855 }
856 
857 /*
858  * This page is about to be returned from the page allocator
859  */
860 static inline int check_new_page(struct page *page)
861 {
862 	char *bad_reason = NULL;
863 	unsigned long bad_flags = 0;
864 
865 	if (unlikely(page_mapcount(page)))
866 		bad_reason = "nonzero mapcount";
867 	if (unlikely(page->mapping != NULL))
868 		bad_reason = "non-NULL mapping";
869 	if (unlikely(atomic_read(&page->_count) != 0))
870 		bad_reason = "nonzero _count";
871 	if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
872 		bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
873 		bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
874 	}
875 	if (unlikely(mem_cgroup_bad_page_check(page)))
876 		bad_reason = "cgroup check failed";
877 	if (unlikely(bad_reason)) {
878 		bad_page(page, bad_reason, bad_flags);
879 		return 1;
880 	}
881 	return 0;
882 }
883 
884 static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
885 {
886 	int i;
887 
888 	for (i = 0; i < (1 << order); i++) {
889 		struct page *p = page + i;
890 		if (unlikely(check_new_page(p)))
891 			return 1;
892 	}
893 
894 	set_page_private(page, 0);
895 	set_page_refcounted(page);
896 
897 	arch_alloc_page(page, order);
898 	kernel_map_pages(page, 1 << order, 1);
899 
900 	if (gfp_flags & __GFP_ZERO)
901 		prep_zero_page(page, order, gfp_flags);
902 
903 	if (order && (gfp_flags & __GFP_COMP))
904 		prep_compound_page(page, order);
905 
906 	return 0;
907 }
908 
909 /*
910  * Go through the free lists for the given migratetype and remove
911  * the smallest available page from the freelists
912  */
913 static inline
914 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
915 						int migratetype)
916 {
917 	unsigned int current_order;
918 	struct free_area *area;
919 	struct page *page;
920 
921 	/* Find a page of the appropriate size in the preferred list */
922 	for (current_order = order; current_order < MAX_ORDER; ++current_order) {
923 		area = &(zone->free_area[current_order]);
924 		if (list_empty(&area->free_list[migratetype]))
925 			continue;
926 
927 		page = list_entry(area->free_list[migratetype].next,
928 							struct page, lru);
929 		list_del(&page->lru);
930 		rmv_page_order(page);
931 		area->nr_free--;
932 		expand(zone, page, order, current_order, area, migratetype);
933 		return page;
934 	}
935 
936 	return NULL;
937 }
938 
939 
940 /*
941  * This array describes the order lists are fallen back to when
942  * the free lists for the desirable migrate type are depleted
943  */
944 static int fallbacks[MIGRATE_TYPES][4] = {
945 	[MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,     MIGRATE_RESERVE },
946 	[MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,     MIGRATE_RESERVE },
947 #ifdef CONFIG_CMA
948 	[MIGRATE_MOVABLE]     = { MIGRATE_CMA,         MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
949 	[MIGRATE_CMA]         = { MIGRATE_RESERVE }, /* Never used */
950 #else
951 	[MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE,   MIGRATE_RESERVE },
952 #endif
953 	[MIGRATE_RESERVE]     = { MIGRATE_RESERVE }, /* Never used */
954 #ifdef CONFIG_MEMORY_ISOLATION
955 	[MIGRATE_ISOLATE]     = { MIGRATE_RESERVE }, /* Never used */
956 #endif
957 };
958 
959 /*
960  * Move the free pages in a range to the free lists of the requested type.
961  * Note that start_page and end_pages are not aligned on a pageblock
962  * boundary. If alignment is required, use move_freepages_block()
963  */
964 int move_freepages(struct zone *zone,
965 			  struct page *start_page, struct page *end_page,
966 			  int migratetype)
967 {
968 	struct page *page;
969 	unsigned long order;
970 	int pages_moved = 0;
971 
972 #ifndef CONFIG_HOLES_IN_ZONE
973 	/*
974 	 * page_zone is not safe to call in this context when
975 	 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
976 	 * anyway as we check zone boundaries in move_freepages_block().
977 	 * Remove at a later date when no bug reports exist related to
978 	 * grouping pages by mobility
979 	 */
980 	BUG_ON(page_zone(start_page) != page_zone(end_page));
981 #endif
982 
983 	for (page = start_page; page <= end_page;) {
984 		/* Make sure we are not inadvertently changing nodes */
985 		VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
986 
987 		if (!pfn_valid_within(page_to_pfn(page))) {
988 			page++;
989 			continue;
990 		}
991 
992 		if (!PageBuddy(page)) {
993 			page++;
994 			continue;
995 		}
996 
997 		order = page_order(page);
998 		list_move(&page->lru,
999 			  &zone->free_area[order].free_list[migratetype]);
1000 		set_freepage_migratetype(page, migratetype);
1001 		page += 1 << order;
1002 		pages_moved += 1 << order;
1003 	}
1004 
1005 	return pages_moved;
1006 }
1007 
1008 int move_freepages_block(struct zone *zone, struct page *page,
1009 				int migratetype)
1010 {
1011 	unsigned long start_pfn, end_pfn;
1012 	struct page *start_page, *end_page;
1013 
1014 	start_pfn = page_to_pfn(page);
1015 	start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1016 	start_page = pfn_to_page(start_pfn);
1017 	end_page = start_page + pageblock_nr_pages - 1;
1018 	end_pfn = start_pfn + pageblock_nr_pages - 1;
1019 
1020 	/* Do not cross zone boundaries */
1021 	if (!zone_spans_pfn(zone, start_pfn))
1022 		start_page = page;
1023 	if (!zone_spans_pfn(zone, end_pfn))
1024 		return 0;
1025 
1026 	return move_freepages(zone, start_page, end_page, migratetype);
1027 }
1028 
1029 static void change_pageblock_range(struct page *pageblock_page,
1030 					int start_order, int migratetype)
1031 {
1032 	int nr_pageblocks = 1 << (start_order - pageblock_order);
1033 
1034 	while (nr_pageblocks--) {
1035 		set_pageblock_migratetype(pageblock_page, migratetype);
1036 		pageblock_page += pageblock_nr_pages;
1037 	}
1038 }
1039 
1040 /*
1041  * If breaking a large block of pages, move all free pages to the preferred
1042  * allocation list. If falling back for a reclaimable kernel allocation, be
1043  * more aggressive about taking ownership of free pages.
1044  *
1045  * On the other hand, never change migration type of MIGRATE_CMA pageblocks
1046  * nor move CMA pages to different free lists. We don't want unmovable pages
1047  * to be allocated from MIGRATE_CMA areas.
1048  *
1049  * Returns the new migratetype of the pageblock (or the same old migratetype
1050  * if it was unchanged).
1051  */
1052 static int try_to_steal_freepages(struct zone *zone, struct page *page,
1053 				  int start_type, int fallback_type)
1054 {
1055 	int current_order = page_order(page);
1056 
1057 	/*
1058 	 * When borrowing from MIGRATE_CMA, we need to release the excess
1059 	 * buddy pages to CMA itself.
1060 	 */
1061 	if (is_migrate_cma(fallback_type))
1062 		return fallback_type;
1063 
1064 	/* Take ownership for orders >= pageblock_order */
1065 	if (current_order >= pageblock_order) {
1066 		change_pageblock_range(page, current_order, start_type);
1067 		return start_type;
1068 	}
1069 
1070 	if (current_order >= pageblock_order / 2 ||
1071 	    start_type == MIGRATE_RECLAIMABLE ||
1072 	    page_group_by_mobility_disabled) {
1073 		int pages;
1074 
1075 		pages = move_freepages_block(zone, page, start_type);
1076 
1077 		/* Claim the whole block if over half of it is free */
1078 		if (pages >= (1 << (pageblock_order-1)) ||
1079 				page_group_by_mobility_disabled) {
1080 
1081 			set_pageblock_migratetype(page, start_type);
1082 			return start_type;
1083 		}
1084 
1085 	}
1086 
1087 	return fallback_type;
1088 }
1089 
1090 /* Remove an element from the buddy allocator from the fallback list */
1091 static inline struct page *
1092 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
1093 {
1094 	struct free_area *area;
1095 	int current_order;
1096 	struct page *page;
1097 	int migratetype, new_type, i;
1098 
1099 	/* Find the largest possible block of pages in the other list */
1100 	for (current_order = MAX_ORDER-1; current_order >= order;
1101 						--current_order) {
1102 		for (i = 0;; i++) {
1103 			migratetype = fallbacks[start_migratetype][i];
1104 
1105 			/* MIGRATE_RESERVE handled later if necessary */
1106 			if (migratetype == MIGRATE_RESERVE)
1107 				break;
1108 
1109 			area = &(zone->free_area[current_order]);
1110 			if (list_empty(&area->free_list[migratetype]))
1111 				continue;
1112 
1113 			page = list_entry(area->free_list[migratetype].next,
1114 					struct page, lru);
1115 			area->nr_free--;
1116 
1117 			new_type = try_to_steal_freepages(zone, page,
1118 							  start_migratetype,
1119 							  migratetype);
1120 
1121 			/* Remove the page from the freelists */
1122 			list_del(&page->lru);
1123 			rmv_page_order(page);
1124 
1125 			expand(zone, page, order, current_order, area,
1126 			       new_type);
1127 
1128 			trace_mm_page_alloc_extfrag(page, order, current_order,
1129 				start_migratetype, migratetype, new_type);
1130 
1131 			return page;
1132 		}
1133 	}
1134 
1135 	return NULL;
1136 }
1137 
1138 /*
1139  * Do the hard work of removing an element from the buddy allocator.
1140  * Call me with the zone->lock already held.
1141  */
1142 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1143 						int migratetype)
1144 {
1145 	struct page *page;
1146 
1147 retry_reserve:
1148 	page = __rmqueue_smallest(zone, order, migratetype);
1149 
1150 	if (unlikely(!page) && migratetype != MIGRATE_RESERVE) {
1151 		page = __rmqueue_fallback(zone, order, migratetype);
1152 
1153 		/*
1154 		 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1155 		 * is used because __rmqueue_smallest is an inline function
1156 		 * and we want just one call site
1157 		 */
1158 		if (!page) {
1159 			migratetype = MIGRATE_RESERVE;
1160 			goto retry_reserve;
1161 		}
1162 	}
1163 
1164 	trace_mm_page_alloc_zone_locked(page, order, migratetype);
1165 	return page;
1166 }
1167 
1168 /*
1169  * Obtain a specified number of elements from the buddy allocator, all under
1170  * a single hold of the lock, for efficiency.  Add them to the supplied list.
1171  * Returns the number of new pages which were placed at *list.
1172  */
1173 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1174 			unsigned long count, struct list_head *list,
1175 			int migratetype, int cold)
1176 {
1177 	int mt = migratetype, i;
1178 
1179 	spin_lock(&zone->lock);
1180 	for (i = 0; i < count; ++i) {
1181 		struct page *page = __rmqueue(zone, order, migratetype);
1182 		if (unlikely(page == NULL))
1183 			break;
1184 
1185 		/*
1186 		 * Split buddy pages returned by expand() are received here
1187 		 * in physical page order. The page is added to the callers and
1188 		 * list and the list head then moves forward. From the callers
1189 		 * perspective, the linked list is ordered by page number in
1190 		 * some conditions. This is useful for IO devices that can
1191 		 * merge IO requests if the physical pages are ordered
1192 		 * properly.
1193 		 */
1194 		if (likely(cold == 0))
1195 			list_add(&page->lru, list);
1196 		else
1197 			list_add_tail(&page->lru, list);
1198 		if (IS_ENABLED(CONFIG_CMA)) {
1199 			mt = get_pageblock_migratetype(page);
1200 			if (!is_migrate_cma(mt) && !is_migrate_isolate(mt))
1201 				mt = migratetype;
1202 		}
1203 		set_freepage_migratetype(page, mt);
1204 		list = &page->lru;
1205 		if (is_migrate_cma(mt))
1206 			__mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
1207 					      -(1 << order));
1208 	}
1209 	__mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
1210 	spin_unlock(&zone->lock);
1211 	return i;
1212 }
1213 
1214 #ifdef CONFIG_NUMA
1215 /*
1216  * Called from the vmstat counter updater to drain pagesets of this
1217  * currently executing processor on remote nodes after they have
1218  * expired.
1219  *
1220  * Note that this function must be called with the thread pinned to
1221  * a single processor.
1222  */
1223 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
1224 {
1225 	unsigned long flags;
1226 	int to_drain;
1227 	unsigned long batch;
1228 
1229 	local_irq_save(flags);
1230 	batch = ACCESS_ONCE(pcp->batch);
1231 	if (pcp->count >= batch)
1232 		to_drain = batch;
1233 	else
1234 		to_drain = pcp->count;
1235 	if (to_drain > 0) {
1236 		free_pcppages_bulk(zone, to_drain, pcp);
1237 		pcp->count -= to_drain;
1238 	}
1239 	local_irq_restore(flags);
1240 }
1241 static bool gfp_thisnode_allocation(gfp_t gfp_mask)
1242 {
1243 	return (gfp_mask & GFP_THISNODE) == GFP_THISNODE;
1244 }
1245 #else
1246 static bool gfp_thisnode_allocation(gfp_t gfp_mask)
1247 {
1248 	return false;
1249 }
1250 #endif
1251 
1252 /*
1253  * Drain pages of the indicated processor.
1254  *
1255  * The processor must either be the current processor and the
1256  * thread pinned to the current processor or a processor that
1257  * is not online.
1258  */
1259 static void drain_pages(unsigned int cpu)
1260 {
1261 	unsigned long flags;
1262 	struct zone *zone;
1263 
1264 	for_each_populated_zone(zone) {
1265 		struct per_cpu_pageset *pset;
1266 		struct per_cpu_pages *pcp;
1267 
1268 		local_irq_save(flags);
1269 		pset = per_cpu_ptr(zone->pageset, cpu);
1270 
1271 		pcp = &pset->pcp;
1272 		if (pcp->count) {
1273 			free_pcppages_bulk(zone, pcp->count, pcp);
1274 			pcp->count = 0;
1275 		}
1276 		local_irq_restore(flags);
1277 	}
1278 }
1279 
1280 /*
1281  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1282  */
1283 void drain_local_pages(void *arg)
1284 {
1285 	drain_pages(smp_processor_id());
1286 }
1287 
1288 /*
1289  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1290  *
1291  * Note that this code is protected against sending an IPI to an offline
1292  * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1293  * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1294  * nothing keeps CPUs from showing up after we populated the cpumask and
1295  * before the call to on_each_cpu_mask().
1296  */
1297 void drain_all_pages(void)
1298 {
1299 	int cpu;
1300 	struct per_cpu_pageset *pcp;
1301 	struct zone *zone;
1302 
1303 	/*
1304 	 * Allocate in the BSS so we wont require allocation in
1305 	 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1306 	 */
1307 	static cpumask_t cpus_with_pcps;
1308 
1309 	/*
1310 	 * We don't care about racing with CPU hotplug event
1311 	 * as offline notification will cause the notified
1312 	 * cpu to drain that CPU pcps and on_each_cpu_mask
1313 	 * disables preemption as part of its processing
1314 	 */
1315 	for_each_online_cpu(cpu) {
1316 		bool has_pcps = false;
1317 		for_each_populated_zone(zone) {
1318 			pcp = per_cpu_ptr(zone->pageset, cpu);
1319 			if (pcp->pcp.count) {
1320 				has_pcps = true;
1321 				break;
1322 			}
1323 		}
1324 		if (has_pcps)
1325 			cpumask_set_cpu(cpu, &cpus_with_pcps);
1326 		else
1327 			cpumask_clear_cpu(cpu, &cpus_with_pcps);
1328 	}
1329 	on_each_cpu_mask(&cpus_with_pcps, drain_local_pages, NULL, 1);
1330 }
1331 
1332 #ifdef CONFIG_HIBERNATION
1333 
1334 void mark_free_pages(struct zone *zone)
1335 {
1336 	unsigned long pfn, max_zone_pfn;
1337 	unsigned long flags;
1338 	int order, t;
1339 	struct list_head *curr;
1340 
1341 	if (zone_is_empty(zone))
1342 		return;
1343 
1344 	spin_lock_irqsave(&zone->lock, flags);
1345 
1346 	max_zone_pfn = zone_end_pfn(zone);
1347 	for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1348 		if (pfn_valid(pfn)) {
1349 			struct page *page = pfn_to_page(pfn);
1350 
1351 			if (!swsusp_page_is_forbidden(page))
1352 				swsusp_unset_page_free(page);
1353 		}
1354 
1355 	for_each_migratetype_order(order, t) {
1356 		list_for_each(curr, &zone->free_area[order].free_list[t]) {
1357 			unsigned long i;
1358 
1359 			pfn = page_to_pfn(list_entry(curr, struct page, lru));
1360 			for (i = 0; i < (1UL << order); i++)
1361 				swsusp_set_page_free(pfn_to_page(pfn + i));
1362 		}
1363 	}
1364 	spin_unlock_irqrestore(&zone->lock, flags);
1365 }
1366 #endif /* CONFIG_PM */
1367 
1368 /*
1369  * Free a 0-order page
1370  * cold == 1 ? free a cold page : free a hot page
1371  */
1372 void free_hot_cold_page(struct page *page, int cold)
1373 {
1374 	struct zone *zone = page_zone(page);
1375 	struct per_cpu_pages *pcp;
1376 	unsigned long flags;
1377 	int migratetype;
1378 
1379 	if (!free_pages_prepare(page, 0))
1380 		return;
1381 
1382 	migratetype = get_pageblock_migratetype(page);
1383 	set_freepage_migratetype(page, migratetype);
1384 	local_irq_save(flags);
1385 	__count_vm_event(PGFREE);
1386 
1387 	/*
1388 	 * We only track unmovable, reclaimable and movable on pcp lists.
1389 	 * Free ISOLATE pages back to the allocator because they are being
1390 	 * offlined but treat RESERVE as movable pages so we can get those
1391 	 * areas back if necessary. Otherwise, we may have to free
1392 	 * excessively into the page allocator
1393 	 */
1394 	if (migratetype >= MIGRATE_PCPTYPES) {
1395 		if (unlikely(is_migrate_isolate(migratetype))) {
1396 			free_one_page(zone, page, 0, migratetype);
1397 			goto out;
1398 		}
1399 		migratetype = MIGRATE_MOVABLE;
1400 	}
1401 
1402 	pcp = &this_cpu_ptr(zone->pageset)->pcp;
1403 	if (cold)
1404 		list_add_tail(&page->lru, &pcp->lists[migratetype]);
1405 	else
1406 		list_add(&page->lru, &pcp->lists[migratetype]);
1407 	pcp->count++;
1408 	if (pcp->count >= pcp->high) {
1409 		unsigned long batch = ACCESS_ONCE(pcp->batch);
1410 		free_pcppages_bulk(zone, batch, pcp);
1411 		pcp->count -= batch;
1412 	}
1413 
1414 out:
1415 	local_irq_restore(flags);
1416 }
1417 
1418 /*
1419  * Free a list of 0-order pages
1420  */
1421 void free_hot_cold_page_list(struct list_head *list, int cold)
1422 {
1423 	struct page *page, *next;
1424 
1425 	list_for_each_entry_safe(page, next, list, lru) {
1426 		trace_mm_page_free_batched(page, cold);
1427 		free_hot_cold_page(page, cold);
1428 	}
1429 }
1430 
1431 /*
1432  * split_page takes a non-compound higher-order page, and splits it into
1433  * n (1<<order) sub-pages: page[0..n]
1434  * Each sub-page must be freed individually.
1435  *
1436  * Note: this is probably too low level an operation for use in drivers.
1437  * Please consult with lkml before using this in your driver.
1438  */
1439 void split_page(struct page *page, unsigned int order)
1440 {
1441 	int i;
1442 
1443 	VM_BUG_ON_PAGE(PageCompound(page), page);
1444 	VM_BUG_ON_PAGE(!page_count(page), page);
1445 
1446 #ifdef CONFIG_KMEMCHECK
1447 	/*
1448 	 * Split shadow pages too, because free(page[0]) would
1449 	 * otherwise free the whole shadow.
1450 	 */
1451 	if (kmemcheck_page_is_tracked(page))
1452 		split_page(virt_to_page(page[0].shadow), order);
1453 #endif
1454 
1455 	for (i = 1; i < (1 << order); i++)
1456 		set_page_refcounted(page + i);
1457 }
1458 EXPORT_SYMBOL_GPL(split_page);
1459 
1460 static int __isolate_free_page(struct page *page, unsigned int order)
1461 {
1462 	unsigned long watermark;
1463 	struct zone *zone;
1464 	int mt;
1465 
1466 	BUG_ON(!PageBuddy(page));
1467 
1468 	zone = page_zone(page);
1469 	mt = get_pageblock_migratetype(page);
1470 
1471 	if (!is_migrate_isolate(mt)) {
1472 		/* Obey watermarks as if the page was being allocated */
1473 		watermark = low_wmark_pages(zone) + (1 << order);
1474 		if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
1475 			return 0;
1476 
1477 		__mod_zone_freepage_state(zone, -(1UL << order), mt);
1478 	}
1479 
1480 	/* Remove page from free list */
1481 	list_del(&page->lru);
1482 	zone->free_area[order].nr_free--;
1483 	rmv_page_order(page);
1484 
1485 	/* Set the pageblock if the isolated page is at least a pageblock */
1486 	if (order >= pageblock_order - 1) {
1487 		struct page *endpage = page + (1 << order) - 1;
1488 		for (; page < endpage; page += pageblock_nr_pages) {
1489 			int mt = get_pageblock_migratetype(page);
1490 			if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
1491 				set_pageblock_migratetype(page,
1492 							  MIGRATE_MOVABLE);
1493 		}
1494 	}
1495 
1496 	return 1UL << order;
1497 }
1498 
1499 /*
1500  * Similar to split_page except the page is already free. As this is only
1501  * being used for migration, the migratetype of the block also changes.
1502  * As this is called with interrupts disabled, the caller is responsible
1503  * for calling arch_alloc_page() and kernel_map_page() after interrupts
1504  * are enabled.
1505  *
1506  * Note: this is probably too low level an operation for use in drivers.
1507  * Please consult with lkml before using this in your driver.
1508  */
1509 int split_free_page(struct page *page)
1510 {
1511 	unsigned int order;
1512 	int nr_pages;
1513 
1514 	order = page_order(page);
1515 
1516 	nr_pages = __isolate_free_page(page, order);
1517 	if (!nr_pages)
1518 		return 0;
1519 
1520 	/* Split into individual pages */
1521 	set_page_refcounted(page);
1522 	split_page(page, order);
1523 	return nr_pages;
1524 }
1525 
1526 /*
1527  * Really, prep_compound_page() should be called from __rmqueue_bulk().  But
1528  * we cheat by calling it from here, in the order > 0 path.  Saves a branch
1529  * or two.
1530  */
1531 static inline
1532 struct page *buffered_rmqueue(struct zone *preferred_zone,
1533 			struct zone *zone, int order, gfp_t gfp_flags,
1534 			int migratetype)
1535 {
1536 	unsigned long flags;
1537 	struct page *page;
1538 	int cold = !!(gfp_flags & __GFP_COLD);
1539 
1540 again:
1541 	if (likely(order == 0)) {
1542 		struct per_cpu_pages *pcp;
1543 		struct list_head *list;
1544 
1545 		local_irq_save(flags);
1546 		pcp = &this_cpu_ptr(zone->pageset)->pcp;
1547 		list = &pcp->lists[migratetype];
1548 		if (list_empty(list)) {
1549 			pcp->count += rmqueue_bulk(zone, 0,
1550 					pcp->batch, list,
1551 					migratetype, cold);
1552 			if (unlikely(list_empty(list)))
1553 				goto failed;
1554 		}
1555 
1556 		if (cold)
1557 			page = list_entry(list->prev, struct page, lru);
1558 		else
1559 			page = list_entry(list->next, struct page, lru);
1560 
1561 		list_del(&page->lru);
1562 		pcp->count--;
1563 	} else {
1564 		if (unlikely(gfp_flags & __GFP_NOFAIL)) {
1565 			/*
1566 			 * __GFP_NOFAIL is not to be used in new code.
1567 			 *
1568 			 * All __GFP_NOFAIL callers should be fixed so that they
1569 			 * properly detect and handle allocation failures.
1570 			 *
1571 			 * We most definitely don't want callers attempting to
1572 			 * allocate greater than order-1 page units with
1573 			 * __GFP_NOFAIL.
1574 			 */
1575 			WARN_ON_ONCE(order > 1);
1576 		}
1577 		spin_lock_irqsave(&zone->lock, flags);
1578 		page = __rmqueue(zone, order, migratetype);
1579 		spin_unlock(&zone->lock);
1580 		if (!page)
1581 			goto failed;
1582 		__mod_zone_freepage_state(zone, -(1 << order),
1583 					  get_pageblock_migratetype(page));
1584 	}
1585 
1586 	/*
1587 	 * NOTE: GFP_THISNODE allocations do not partake in the kswapd
1588 	 * aging protocol, so they can't be fair.
1589 	 */
1590 	if (!gfp_thisnode_allocation(gfp_flags))
1591 		__mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
1592 
1593 	__count_zone_vm_events(PGALLOC, zone, 1 << order);
1594 	zone_statistics(preferred_zone, zone, gfp_flags);
1595 	local_irq_restore(flags);
1596 
1597 	VM_BUG_ON_PAGE(bad_range(zone, page), page);
1598 	if (prep_new_page(page, order, gfp_flags))
1599 		goto again;
1600 	return page;
1601 
1602 failed:
1603 	local_irq_restore(flags);
1604 	return NULL;
1605 }
1606 
1607 #ifdef CONFIG_FAIL_PAGE_ALLOC
1608 
1609 static struct {
1610 	struct fault_attr attr;
1611 
1612 	u32 ignore_gfp_highmem;
1613 	u32 ignore_gfp_wait;
1614 	u32 min_order;
1615 } fail_page_alloc = {
1616 	.attr = FAULT_ATTR_INITIALIZER,
1617 	.ignore_gfp_wait = 1,
1618 	.ignore_gfp_highmem = 1,
1619 	.min_order = 1,
1620 };
1621 
1622 static int __init setup_fail_page_alloc(char *str)
1623 {
1624 	return setup_fault_attr(&fail_page_alloc.attr, str);
1625 }
1626 __setup("fail_page_alloc=", setup_fail_page_alloc);
1627 
1628 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1629 {
1630 	if (order < fail_page_alloc.min_order)
1631 		return false;
1632 	if (gfp_mask & __GFP_NOFAIL)
1633 		return false;
1634 	if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1635 		return false;
1636 	if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1637 		return false;
1638 
1639 	return should_fail(&fail_page_alloc.attr, 1 << order);
1640 }
1641 
1642 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1643 
1644 static int __init fail_page_alloc_debugfs(void)
1645 {
1646 	umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1647 	struct dentry *dir;
1648 
1649 	dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
1650 					&fail_page_alloc.attr);
1651 	if (IS_ERR(dir))
1652 		return PTR_ERR(dir);
1653 
1654 	if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
1655 				&fail_page_alloc.ignore_gfp_wait))
1656 		goto fail;
1657 	if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1658 				&fail_page_alloc.ignore_gfp_highmem))
1659 		goto fail;
1660 	if (!debugfs_create_u32("min-order", mode, dir,
1661 				&fail_page_alloc.min_order))
1662 		goto fail;
1663 
1664 	return 0;
1665 fail:
1666 	debugfs_remove_recursive(dir);
1667 
1668 	return -ENOMEM;
1669 }
1670 
1671 late_initcall(fail_page_alloc_debugfs);
1672 
1673 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1674 
1675 #else /* CONFIG_FAIL_PAGE_ALLOC */
1676 
1677 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1678 {
1679 	return false;
1680 }
1681 
1682 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1683 
1684 /*
1685  * Return true if free pages are above 'mark'. This takes into account the order
1686  * of the allocation.
1687  */
1688 static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1689 		      int classzone_idx, int alloc_flags, long free_pages)
1690 {
1691 	/* free_pages my go negative - that's OK */
1692 	long min = mark;
1693 	long lowmem_reserve = z->lowmem_reserve[classzone_idx];
1694 	int o;
1695 	long free_cma = 0;
1696 
1697 	free_pages -= (1 << order) - 1;
1698 	if (alloc_flags & ALLOC_HIGH)
1699 		min -= min / 2;
1700 	if (alloc_flags & ALLOC_HARDER)
1701 		min -= min / 4;
1702 #ifdef CONFIG_CMA
1703 	/* If allocation can't use CMA areas don't use free CMA pages */
1704 	if (!(alloc_flags & ALLOC_CMA))
1705 		free_cma = zone_page_state(z, NR_FREE_CMA_PAGES);
1706 #endif
1707 
1708 	if (free_pages - free_cma <= min + lowmem_reserve)
1709 		return false;
1710 	for (o = 0; o < order; o++) {
1711 		/* At the next order, this order's pages become unavailable */
1712 		free_pages -= z->free_area[o].nr_free << o;
1713 
1714 		/* Require fewer higher order pages to be free */
1715 		min >>= 1;
1716 
1717 		if (free_pages <= min)
1718 			return false;
1719 	}
1720 	return true;
1721 }
1722 
1723 bool zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1724 		      int classzone_idx, int alloc_flags)
1725 {
1726 	return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1727 					zone_page_state(z, NR_FREE_PAGES));
1728 }
1729 
1730 bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark,
1731 		      int classzone_idx, int alloc_flags)
1732 {
1733 	long free_pages = zone_page_state(z, NR_FREE_PAGES);
1734 
1735 	if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
1736 		free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
1737 
1738 	return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
1739 								free_pages);
1740 }
1741 
1742 #ifdef CONFIG_NUMA
1743 /*
1744  * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
1745  * skip over zones that are not allowed by the cpuset, or that have
1746  * been recently (in last second) found to be nearly full.  See further
1747  * comments in mmzone.h.  Reduces cache footprint of zonelist scans
1748  * that have to skip over a lot of full or unallowed zones.
1749  *
1750  * If the zonelist cache is present in the passed zonelist, then
1751  * returns a pointer to the allowed node mask (either the current
1752  * tasks mems_allowed, or node_states[N_MEMORY].)
1753  *
1754  * If the zonelist cache is not available for this zonelist, does
1755  * nothing and returns NULL.
1756  *
1757  * If the fullzones BITMAP in the zonelist cache is stale (more than
1758  * a second since last zap'd) then we zap it out (clear its bits.)
1759  *
1760  * We hold off even calling zlc_setup, until after we've checked the
1761  * first zone in the zonelist, on the theory that most allocations will
1762  * be satisfied from that first zone, so best to examine that zone as
1763  * quickly as we can.
1764  */
1765 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1766 {
1767 	struct zonelist_cache *zlc;	/* cached zonelist speedup info */
1768 	nodemask_t *allowednodes;	/* zonelist_cache approximation */
1769 
1770 	zlc = zonelist->zlcache_ptr;
1771 	if (!zlc)
1772 		return NULL;
1773 
1774 	if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1775 		bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1776 		zlc->last_full_zap = jiffies;
1777 	}
1778 
1779 	allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1780 					&cpuset_current_mems_allowed :
1781 					&node_states[N_MEMORY];
1782 	return allowednodes;
1783 }
1784 
1785 /*
1786  * Given 'z' scanning a zonelist, run a couple of quick checks to see
1787  * if it is worth looking at further for free memory:
1788  *  1) Check that the zone isn't thought to be full (doesn't have its
1789  *     bit set in the zonelist_cache fullzones BITMAP).
1790  *  2) Check that the zones node (obtained from the zonelist_cache
1791  *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1792  * Return true (non-zero) if zone is worth looking at further, or
1793  * else return false (zero) if it is not.
1794  *
1795  * This check -ignores- the distinction between various watermarks,
1796  * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
1797  * found to be full for any variation of these watermarks, it will
1798  * be considered full for up to one second by all requests, unless
1799  * we are so low on memory on all allowed nodes that we are forced
1800  * into the second scan of the zonelist.
1801  *
1802  * In the second scan we ignore this zonelist cache and exactly
1803  * apply the watermarks to all zones, even it is slower to do so.
1804  * We are low on memory in the second scan, and should leave no stone
1805  * unturned looking for a free page.
1806  */
1807 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1808 						nodemask_t *allowednodes)
1809 {
1810 	struct zonelist_cache *zlc;	/* cached zonelist speedup info */
1811 	int i;				/* index of *z in zonelist zones */
1812 	int n;				/* node that zone *z is on */
1813 
1814 	zlc = zonelist->zlcache_ptr;
1815 	if (!zlc)
1816 		return 1;
1817 
1818 	i = z - zonelist->_zonerefs;
1819 	n = zlc->z_to_n[i];
1820 
1821 	/* This zone is worth trying if it is allowed but not full */
1822 	return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1823 }
1824 
1825 /*
1826  * Given 'z' scanning a zonelist, set the corresponding bit in
1827  * zlc->fullzones, so that subsequent attempts to allocate a page
1828  * from that zone don't waste time re-examining it.
1829  */
1830 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1831 {
1832 	struct zonelist_cache *zlc;	/* cached zonelist speedup info */
1833 	int i;				/* index of *z in zonelist zones */
1834 
1835 	zlc = zonelist->zlcache_ptr;
1836 	if (!zlc)
1837 		return;
1838 
1839 	i = z - zonelist->_zonerefs;
1840 
1841 	set_bit(i, zlc->fullzones);
1842 }
1843 
1844 /*
1845  * clear all zones full, called after direct reclaim makes progress so that
1846  * a zone that was recently full is not skipped over for up to a second
1847  */
1848 static void zlc_clear_zones_full(struct zonelist *zonelist)
1849 {
1850 	struct zonelist_cache *zlc;	/* cached zonelist speedup info */
1851 
1852 	zlc = zonelist->zlcache_ptr;
1853 	if (!zlc)
1854 		return;
1855 
1856 	bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1857 }
1858 
1859 static bool zone_local(struct zone *local_zone, struct zone *zone)
1860 {
1861 	return local_zone->node == zone->node;
1862 }
1863 
1864 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1865 {
1866 	return node_isset(local_zone->node, zone->zone_pgdat->reclaim_nodes);
1867 }
1868 
1869 static void __paginginit init_zone_allows_reclaim(int nid)
1870 {
1871 	int i;
1872 
1873 	for_each_online_node(i)
1874 		if (node_distance(nid, i) <= RECLAIM_DISTANCE)
1875 			node_set(i, NODE_DATA(nid)->reclaim_nodes);
1876 		else
1877 			zone_reclaim_mode = 1;
1878 }
1879 
1880 #else	/* CONFIG_NUMA */
1881 
1882 static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1883 {
1884 	return NULL;
1885 }
1886 
1887 static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1888 				nodemask_t *allowednodes)
1889 {
1890 	return 1;
1891 }
1892 
1893 static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1894 {
1895 }
1896 
1897 static void zlc_clear_zones_full(struct zonelist *zonelist)
1898 {
1899 }
1900 
1901 static bool zone_local(struct zone *local_zone, struct zone *zone)
1902 {
1903 	return true;
1904 }
1905 
1906 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
1907 {
1908 	return true;
1909 }
1910 
1911 static inline void init_zone_allows_reclaim(int nid)
1912 {
1913 }
1914 #endif	/* CONFIG_NUMA */
1915 
1916 /*
1917  * get_page_from_freelist goes through the zonelist trying to allocate
1918  * a page.
1919  */
1920 static struct page *
1921 get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1922 		struct zonelist *zonelist, int high_zoneidx, int alloc_flags,
1923 		struct zone *preferred_zone, int migratetype)
1924 {
1925 	struct zoneref *z;
1926 	struct page *page = NULL;
1927 	int classzone_idx;
1928 	struct zone *zone;
1929 	nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1930 	int zlc_active = 0;		/* set if using zonelist_cache */
1931 	int did_zlc_setup = 0;		/* just call zlc_setup() one time */
1932 
1933 	classzone_idx = zone_idx(preferred_zone);
1934 zonelist_scan:
1935 	/*
1936 	 * Scan zonelist, looking for a zone with enough free.
1937 	 * See also __cpuset_node_allowed_softwall() comment in kernel/cpuset.c.
1938 	 */
1939 	for_each_zone_zonelist_nodemask(zone, z, zonelist,
1940 						high_zoneidx, nodemask) {
1941 		unsigned long mark;
1942 
1943 		if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
1944 			!zlc_zone_worth_trying(zonelist, z, allowednodes))
1945 				continue;
1946 		if ((alloc_flags & ALLOC_CPUSET) &&
1947 			!cpuset_zone_allowed_softwall(zone, gfp_mask))
1948 				continue;
1949 		BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
1950 		if (unlikely(alloc_flags & ALLOC_NO_WATERMARKS))
1951 			goto try_this_zone;
1952 		/*
1953 		 * Distribute pages in proportion to the individual
1954 		 * zone size to ensure fair page aging.  The zone a
1955 		 * page was allocated in should have no effect on the
1956 		 * time the page has in memory before being reclaimed.
1957 		 *
1958 		 * Try to stay in local zones in the fastpath.  If
1959 		 * that fails, the slowpath is entered, which will do
1960 		 * another pass starting with the local zones, but
1961 		 * ultimately fall back to remote zones that do not
1962 		 * partake in the fairness round-robin cycle of this
1963 		 * zonelist.
1964 		 *
1965 		 * NOTE: GFP_THISNODE allocations do not partake in
1966 		 * the kswapd aging protocol, so they can't be fair.
1967 		 */
1968 		if ((alloc_flags & ALLOC_WMARK_LOW) &&
1969 		    !gfp_thisnode_allocation(gfp_mask)) {
1970 			if (zone_page_state(zone, NR_ALLOC_BATCH) <= 0)
1971 				continue;
1972 			if (!zone_local(preferred_zone, zone))
1973 				continue;
1974 		}
1975 		/*
1976 		 * When allocating a page cache page for writing, we
1977 		 * want to get it from a zone that is within its dirty
1978 		 * limit, such that no single zone holds more than its
1979 		 * proportional share of globally allowed dirty pages.
1980 		 * The dirty limits take into account the zone's
1981 		 * lowmem reserves and high watermark so that kswapd
1982 		 * should be able to balance it without having to
1983 		 * write pages from its LRU list.
1984 		 *
1985 		 * This may look like it could increase pressure on
1986 		 * lower zones by failing allocations in higher zones
1987 		 * before they are full.  But the pages that do spill
1988 		 * over are limited as the lower zones are protected
1989 		 * by this very same mechanism.  It should not become
1990 		 * a practical burden to them.
1991 		 *
1992 		 * XXX: For now, allow allocations to potentially
1993 		 * exceed the per-zone dirty limit in the slowpath
1994 		 * (ALLOC_WMARK_LOW unset) before going into reclaim,
1995 		 * which is important when on a NUMA setup the allowed
1996 		 * zones are together not big enough to reach the
1997 		 * global limit.  The proper fix for these situations
1998 		 * will require awareness of zones in the
1999 		 * dirty-throttling and the flusher threads.
2000 		 */
2001 		if ((alloc_flags & ALLOC_WMARK_LOW) &&
2002 		    (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone))
2003 			goto this_zone_full;
2004 
2005 		mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2006 		if (!zone_watermark_ok(zone, order, mark,
2007 				       classzone_idx, alloc_flags)) {
2008 			int ret;
2009 
2010 			if (IS_ENABLED(CONFIG_NUMA) &&
2011 					!did_zlc_setup && nr_online_nodes > 1) {
2012 				/*
2013 				 * we do zlc_setup if there are multiple nodes
2014 				 * and before considering the first zone allowed
2015 				 * by the cpuset.
2016 				 */
2017 				allowednodes = zlc_setup(zonelist, alloc_flags);
2018 				zlc_active = 1;
2019 				did_zlc_setup = 1;
2020 			}
2021 
2022 			if (zone_reclaim_mode == 0 ||
2023 			    !zone_allows_reclaim(preferred_zone, zone))
2024 				goto this_zone_full;
2025 
2026 			/*
2027 			 * As we may have just activated ZLC, check if the first
2028 			 * eligible zone has failed zone_reclaim recently.
2029 			 */
2030 			if (IS_ENABLED(CONFIG_NUMA) && zlc_active &&
2031 				!zlc_zone_worth_trying(zonelist, z, allowednodes))
2032 				continue;
2033 
2034 			ret = zone_reclaim(zone, gfp_mask, order);
2035 			switch (ret) {
2036 			case ZONE_RECLAIM_NOSCAN:
2037 				/* did not scan */
2038 				continue;
2039 			case ZONE_RECLAIM_FULL:
2040 				/* scanned but unreclaimable */
2041 				continue;
2042 			default:
2043 				/* did we reclaim enough */
2044 				if (zone_watermark_ok(zone, order, mark,
2045 						classzone_idx, alloc_flags))
2046 					goto try_this_zone;
2047 
2048 				/*
2049 				 * Failed to reclaim enough to meet watermark.
2050 				 * Only mark the zone full if checking the min
2051 				 * watermark or if we failed to reclaim just
2052 				 * 1<<order pages or else the page allocator
2053 				 * fastpath will prematurely mark zones full
2054 				 * when the watermark is between the low and
2055 				 * min watermarks.
2056 				 */
2057 				if (((alloc_flags & ALLOC_WMARK_MASK) == ALLOC_WMARK_MIN) ||
2058 				    ret == ZONE_RECLAIM_SOME)
2059 					goto this_zone_full;
2060 
2061 				continue;
2062 			}
2063 		}
2064 
2065 try_this_zone:
2066 		page = buffered_rmqueue(preferred_zone, zone, order,
2067 						gfp_mask, migratetype);
2068 		if (page)
2069 			break;
2070 this_zone_full:
2071 		if (IS_ENABLED(CONFIG_NUMA))
2072 			zlc_mark_zone_full(zonelist, z);
2073 	}
2074 
2075 	if (unlikely(IS_ENABLED(CONFIG_NUMA) && page == NULL && zlc_active)) {
2076 		/* Disable zlc cache for second zonelist scan */
2077 		zlc_active = 0;
2078 		goto zonelist_scan;
2079 	}
2080 
2081 	if (page)
2082 		/*
2083 		 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was
2084 		 * necessary to allocate the page. The expectation is
2085 		 * that the caller is taking steps that will free more
2086 		 * memory. The caller should avoid the page being used
2087 		 * for !PFMEMALLOC purposes.
2088 		 */
2089 		page->pfmemalloc = !!(alloc_flags & ALLOC_NO_WATERMARKS);
2090 
2091 	return page;
2092 }
2093 
2094 /*
2095  * Large machines with many possible nodes should not always dump per-node
2096  * meminfo in irq context.
2097  */
2098 static inline bool should_suppress_show_mem(void)
2099 {
2100 	bool ret = false;
2101 
2102 #if NODES_SHIFT > 8
2103 	ret = in_interrupt();
2104 #endif
2105 	return ret;
2106 }
2107 
2108 static DEFINE_RATELIMIT_STATE(nopage_rs,
2109 		DEFAULT_RATELIMIT_INTERVAL,
2110 		DEFAULT_RATELIMIT_BURST);
2111 
2112 void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
2113 {
2114 	unsigned int filter = SHOW_MEM_FILTER_NODES;
2115 
2116 	if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2117 	    debug_guardpage_minorder() > 0)
2118 		return;
2119 
2120 	/*
2121 	 * This documents exceptions given to allocations in certain
2122 	 * contexts that are allowed to allocate outside current's set
2123 	 * of allowed nodes.
2124 	 */
2125 	if (!(gfp_mask & __GFP_NOMEMALLOC))
2126 		if (test_thread_flag(TIF_MEMDIE) ||
2127 		    (current->flags & (PF_MEMALLOC | PF_EXITING)))
2128 			filter &= ~SHOW_MEM_FILTER_NODES;
2129 	if (in_interrupt() || !(gfp_mask & __GFP_WAIT))
2130 		filter &= ~SHOW_MEM_FILTER_NODES;
2131 
2132 	if (fmt) {
2133 		struct va_format vaf;
2134 		va_list args;
2135 
2136 		va_start(args, fmt);
2137 
2138 		vaf.fmt = fmt;
2139 		vaf.va = &args;
2140 
2141 		pr_warn("%pV", &vaf);
2142 
2143 		va_end(args);
2144 	}
2145 
2146 	pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2147 		current->comm, order, gfp_mask);
2148 
2149 	dump_stack();
2150 	if (!should_suppress_show_mem())
2151 		show_mem(filter);
2152 }
2153 
2154 static inline int
2155 should_alloc_retry(gfp_t gfp_mask, unsigned int order,
2156 				unsigned long did_some_progress,
2157 				unsigned long pages_reclaimed)
2158 {
2159 	/* Do not loop if specifically requested */
2160 	if (gfp_mask & __GFP_NORETRY)
2161 		return 0;
2162 
2163 	/* Always retry if specifically requested */
2164 	if (gfp_mask & __GFP_NOFAIL)
2165 		return 1;
2166 
2167 	/*
2168 	 * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim
2169 	 * making forward progress without invoking OOM. Suspend also disables
2170 	 * storage devices so kswapd will not help. Bail if we are suspending.
2171 	 */
2172 	if (!did_some_progress && pm_suspended_storage())
2173 		return 0;
2174 
2175 	/*
2176 	 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
2177 	 * means __GFP_NOFAIL, but that may not be true in other
2178 	 * implementations.
2179 	 */
2180 	if (order <= PAGE_ALLOC_COSTLY_ORDER)
2181 		return 1;
2182 
2183 	/*
2184 	 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
2185 	 * specified, then we retry until we no longer reclaim any pages
2186 	 * (above), or we've reclaimed an order of pages at least as
2187 	 * large as the allocation's order. In both cases, if the
2188 	 * allocation still fails, we stop retrying.
2189 	 */
2190 	if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order))
2191 		return 1;
2192 
2193 	return 0;
2194 }
2195 
2196 static inline struct page *
2197 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2198 	struct zonelist *zonelist, enum zone_type high_zoneidx,
2199 	nodemask_t *nodemask, struct zone *preferred_zone,
2200 	int migratetype)
2201 {
2202 	struct page *page;
2203 
2204 	/* Acquire the OOM killer lock for the zones in zonelist */
2205 	if (!try_set_zonelist_oom(zonelist, gfp_mask)) {
2206 		schedule_timeout_uninterruptible(1);
2207 		return NULL;
2208 	}
2209 
2210 	/*
2211 	 * Go through the zonelist yet one more time, keep very high watermark
2212 	 * here, this is only to catch a parallel oom killing, we must fail if
2213 	 * we're still under heavy pressure.
2214 	 */
2215 	page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
2216 		order, zonelist, high_zoneidx,
2217 		ALLOC_WMARK_HIGH|ALLOC_CPUSET,
2218 		preferred_zone, migratetype);
2219 	if (page)
2220 		goto out;
2221 
2222 	if (!(gfp_mask & __GFP_NOFAIL)) {
2223 		/* The OOM killer will not help higher order allocs */
2224 		if (order > PAGE_ALLOC_COSTLY_ORDER)
2225 			goto out;
2226 		/* The OOM killer does not needlessly kill tasks for lowmem */
2227 		if (high_zoneidx < ZONE_NORMAL)
2228 			goto out;
2229 		/*
2230 		 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
2231 		 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
2232 		 * The caller should handle page allocation failure by itself if
2233 		 * it specifies __GFP_THISNODE.
2234 		 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
2235 		 */
2236 		if (gfp_mask & __GFP_THISNODE)
2237 			goto out;
2238 	}
2239 	/* Exhausted what can be done so it's blamo time */
2240 	out_of_memory(zonelist, gfp_mask, order, nodemask, false);
2241 
2242 out:
2243 	clear_zonelist_oom(zonelist, gfp_mask);
2244 	return page;
2245 }
2246 
2247 #ifdef CONFIG_COMPACTION
2248 /* Try memory compaction for high-order allocations before reclaim */
2249 static struct page *
2250 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2251 	struct zonelist *zonelist, enum zone_type high_zoneidx,
2252 	nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2253 	int migratetype, bool sync_migration,
2254 	bool *contended_compaction, bool *deferred_compaction,
2255 	unsigned long *did_some_progress)
2256 {
2257 	if (!order)
2258 		return NULL;
2259 
2260 	if (compaction_deferred(preferred_zone, order)) {
2261 		*deferred_compaction = true;
2262 		return NULL;
2263 	}
2264 
2265 	current->flags |= PF_MEMALLOC;
2266 	*did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask,
2267 						nodemask, sync_migration,
2268 						contended_compaction);
2269 	current->flags &= ~PF_MEMALLOC;
2270 
2271 	if (*did_some_progress != COMPACT_SKIPPED) {
2272 		struct page *page;
2273 
2274 		/* Page migration frees to the PCP lists but we want merging */
2275 		drain_pages(get_cpu());
2276 		put_cpu();
2277 
2278 		page = get_page_from_freelist(gfp_mask, nodemask,
2279 				order, zonelist, high_zoneidx,
2280 				alloc_flags & ~ALLOC_NO_WATERMARKS,
2281 				preferred_zone, migratetype);
2282 		if (page) {
2283 			preferred_zone->compact_blockskip_flush = false;
2284 			compaction_defer_reset(preferred_zone, order, true);
2285 			count_vm_event(COMPACTSUCCESS);
2286 			return page;
2287 		}
2288 
2289 		/*
2290 		 * It's bad if compaction run occurs and fails.
2291 		 * The most likely reason is that pages exist,
2292 		 * but not enough to satisfy watermarks.
2293 		 */
2294 		count_vm_event(COMPACTFAIL);
2295 
2296 		/*
2297 		 * As async compaction considers a subset of pageblocks, only
2298 		 * defer if the failure was a sync compaction failure.
2299 		 */
2300 		if (sync_migration)
2301 			defer_compaction(preferred_zone, order);
2302 
2303 		cond_resched();
2304 	}
2305 
2306 	return NULL;
2307 }
2308 #else
2309 static inline struct page *
2310 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2311 	struct zonelist *zonelist, enum zone_type high_zoneidx,
2312 	nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2313 	int migratetype, bool sync_migration,
2314 	bool *contended_compaction, bool *deferred_compaction,
2315 	unsigned long *did_some_progress)
2316 {
2317 	return NULL;
2318 }
2319 #endif /* CONFIG_COMPACTION */
2320 
2321 /* Perform direct synchronous page reclaim */
2322 static int
2323 __perform_reclaim(gfp_t gfp_mask, unsigned int order, struct zonelist *zonelist,
2324 		  nodemask_t *nodemask)
2325 {
2326 	struct reclaim_state reclaim_state;
2327 	int progress;
2328 
2329 	cond_resched();
2330 
2331 	/* We now go into synchronous reclaim */
2332 	cpuset_memory_pressure_bump();
2333 	current->flags |= PF_MEMALLOC;
2334 	lockdep_set_current_reclaim_state(gfp_mask);
2335 	reclaim_state.reclaimed_slab = 0;
2336 	current->reclaim_state = &reclaim_state;
2337 
2338 	progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask);
2339 
2340 	current->reclaim_state = NULL;
2341 	lockdep_clear_current_reclaim_state();
2342 	current->flags &= ~PF_MEMALLOC;
2343 
2344 	cond_resched();
2345 
2346 	return progress;
2347 }
2348 
2349 /* The really slow allocator path where we enter direct reclaim */
2350 static inline struct page *
2351 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
2352 	struct zonelist *zonelist, enum zone_type high_zoneidx,
2353 	nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
2354 	int migratetype, unsigned long *did_some_progress)
2355 {
2356 	struct page *page = NULL;
2357 	bool drained = false;
2358 
2359 	*did_some_progress = __perform_reclaim(gfp_mask, order, zonelist,
2360 					       nodemask);
2361 	if (unlikely(!(*did_some_progress)))
2362 		return NULL;
2363 
2364 	/* After successful reclaim, reconsider all zones for allocation */
2365 	if (IS_ENABLED(CONFIG_NUMA))
2366 		zlc_clear_zones_full(zonelist);
2367 
2368 retry:
2369 	page = get_page_from_freelist(gfp_mask, nodemask, order,
2370 					zonelist, high_zoneidx,
2371 					alloc_flags & ~ALLOC_NO_WATERMARKS,
2372 					preferred_zone, migratetype);
2373 
2374 	/*
2375 	 * If an allocation failed after direct reclaim, it could be because
2376 	 * pages are pinned on the per-cpu lists. Drain them and try again
2377 	 */
2378 	if (!page && !drained) {
2379 		drain_all_pages();
2380 		drained = true;
2381 		goto retry;
2382 	}
2383 
2384 	return page;
2385 }
2386 
2387 /*
2388  * This is called in the allocator slow-path if the allocation request is of
2389  * sufficient urgency to ignore watermarks and take other desperate measures
2390  */
2391 static inline struct page *
2392 __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order,
2393 	struct zonelist *zonelist, enum zone_type high_zoneidx,
2394 	nodemask_t *nodemask, struct zone *preferred_zone,
2395 	int migratetype)
2396 {
2397 	struct page *page;
2398 
2399 	do {
2400 		page = get_page_from_freelist(gfp_mask, nodemask, order,
2401 			zonelist, high_zoneidx, ALLOC_NO_WATERMARKS,
2402 			preferred_zone, migratetype);
2403 
2404 		if (!page && gfp_mask & __GFP_NOFAIL)
2405 			wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2406 	} while (!page && (gfp_mask & __GFP_NOFAIL));
2407 
2408 	return page;
2409 }
2410 
2411 static void prepare_slowpath(gfp_t gfp_mask, unsigned int order,
2412 			     struct zonelist *zonelist,
2413 			     enum zone_type high_zoneidx,
2414 			     struct zone *preferred_zone)
2415 {
2416 	struct zoneref *z;
2417 	struct zone *zone;
2418 
2419 	for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
2420 		if (!(gfp_mask & __GFP_NO_KSWAPD))
2421 			wakeup_kswapd(zone, order, zone_idx(preferred_zone));
2422 		/*
2423 		 * Only reset the batches of zones that were actually
2424 		 * considered in the fast path, we don't want to
2425 		 * thrash fairness information for zones that are not
2426 		 * actually part of this zonelist's round-robin cycle.
2427 		 */
2428 		if (!zone_local(preferred_zone, zone))
2429 			continue;
2430 		mod_zone_page_state(zone, NR_ALLOC_BATCH,
2431 				    high_wmark_pages(zone) -
2432 				    low_wmark_pages(zone) -
2433 				    zone_page_state(zone, NR_ALLOC_BATCH));
2434 	}
2435 }
2436 
2437 static inline int
2438 gfp_to_alloc_flags(gfp_t gfp_mask)
2439 {
2440 	int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
2441 	const gfp_t wait = gfp_mask & __GFP_WAIT;
2442 
2443 	/* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2444 	BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
2445 
2446 	/*
2447 	 * The caller may dip into page reserves a bit more if the caller
2448 	 * cannot run direct reclaim, or if the caller has realtime scheduling
2449 	 * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
2450 	 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
2451 	 */
2452 	alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
2453 
2454 	if (!wait) {
2455 		/*
2456 		 * Not worth trying to allocate harder for
2457 		 * __GFP_NOMEMALLOC even if it can't schedule.
2458 		 */
2459 		if  (!(gfp_mask & __GFP_NOMEMALLOC))
2460 			alloc_flags |= ALLOC_HARDER;
2461 		/*
2462 		 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
2463 		 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
2464 		 */
2465 		alloc_flags &= ~ALLOC_CPUSET;
2466 	} else if (unlikely(rt_task(current)) && !in_interrupt())
2467 		alloc_flags |= ALLOC_HARDER;
2468 
2469 	if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
2470 		if (gfp_mask & __GFP_MEMALLOC)
2471 			alloc_flags |= ALLOC_NO_WATERMARKS;
2472 		else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
2473 			alloc_flags |= ALLOC_NO_WATERMARKS;
2474 		else if (!in_interrupt() &&
2475 				((current->flags & PF_MEMALLOC) ||
2476 				 unlikely(test_thread_flag(TIF_MEMDIE))))
2477 			alloc_flags |= ALLOC_NO_WATERMARKS;
2478 	}
2479 #ifdef CONFIG_CMA
2480 	if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2481 		alloc_flags |= ALLOC_CMA;
2482 #endif
2483 	return alloc_flags;
2484 }
2485 
2486 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
2487 {
2488 	return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
2489 }
2490 
2491 static inline struct page *
2492 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
2493 	struct zonelist *zonelist, enum zone_type high_zoneidx,
2494 	nodemask_t *nodemask, struct zone *preferred_zone,
2495 	int migratetype)
2496 {
2497 	const gfp_t wait = gfp_mask & __GFP_WAIT;
2498 	struct page *page = NULL;
2499 	int alloc_flags;
2500 	unsigned long pages_reclaimed = 0;
2501 	unsigned long did_some_progress;
2502 	bool sync_migration = false;
2503 	bool deferred_compaction = false;
2504 	bool contended_compaction = false;
2505 
2506 	/*
2507 	 * In the slowpath, we sanity check order to avoid ever trying to
2508 	 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2509 	 * be using allocators in order of preference for an area that is
2510 	 * too large.
2511 	 */
2512 	if (order >= MAX_ORDER) {
2513 		WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
2514 		return NULL;
2515 	}
2516 
2517 	/*
2518 	 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
2519 	 * __GFP_NOWARN set) should not cause reclaim since the subsystem
2520 	 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
2521 	 * using a larger set of nodes after it has established that the
2522 	 * allowed per node queues are empty and that nodes are
2523 	 * over allocated.
2524 	 */
2525 	if (gfp_thisnode_allocation(gfp_mask))
2526 		goto nopage;
2527 
2528 restart:
2529 	prepare_slowpath(gfp_mask, order, zonelist,
2530 			 high_zoneidx, preferred_zone);
2531 
2532 	/*
2533 	 * OK, we're below the kswapd watermark and have kicked background
2534 	 * reclaim. Now things get more complex, so set up alloc_flags according
2535 	 * to how we want to proceed.
2536 	 */
2537 	alloc_flags = gfp_to_alloc_flags(gfp_mask);
2538 
2539 	/*
2540 	 * Find the true preferred zone if the allocation is unconstrained by
2541 	 * cpusets.
2542 	 */
2543 	if (!(alloc_flags & ALLOC_CPUSET) && !nodemask)
2544 		first_zones_zonelist(zonelist, high_zoneidx, NULL,
2545 					&preferred_zone);
2546 
2547 rebalance:
2548 	/* This is the last chance, in general, before the goto nopage. */
2549 	page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
2550 			high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS,
2551 			preferred_zone, migratetype);
2552 	if (page)
2553 		goto got_pg;
2554 
2555 	/* Allocate without watermarks if the context allows */
2556 	if (alloc_flags & ALLOC_NO_WATERMARKS) {
2557 		/*
2558 		 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
2559 		 * the allocation is high priority and these type of
2560 		 * allocations are system rather than user orientated
2561 		 */
2562 		zonelist = node_zonelist(numa_node_id(), gfp_mask);
2563 
2564 		page = __alloc_pages_high_priority(gfp_mask, order,
2565 				zonelist, high_zoneidx, nodemask,
2566 				preferred_zone, migratetype);
2567 		if (page) {
2568 			goto got_pg;
2569 		}
2570 	}
2571 
2572 	/* Atomic allocations - we can't balance anything */
2573 	if (!wait) {
2574 		/*
2575 		 * All existing users of the deprecated __GFP_NOFAIL are
2576 		 * blockable, so warn of any new users that actually allow this
2577 		 * type of allocation to fail.
2578 		 */
2579 		WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
2580 		goto nopage;
2581 	}
2582 
2583 	/* Avoid recursion of direct reclaim */
2584 	if (current->flags & PF_MEMALLOC)
2585 		goto nopage;
2586 
2587 	/* Avoid allocations with no watermarks from looping endlessly */
2588 	if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
2589 		goto nopage;
2590 
2591 	/*
2592 	 * Try direct compaction. The first pass is asynchronous. Subsequent
2593 	 * attempts after direct reclaim are synchronous
2594 	 */
2595 	page = __alloc_pages_direct_compact(gfp_mask, order,
2596 					zonelist, high_zoneidx,
2597 					nodemask,
2598 					alloc_flags, preferred_zone,
2599 					migratetype, sync_migration,
2600 					&contended_compaction,
2601 					&deferred_compaction,
2602 					&did_some_progress);
2603 	if (page)
2604 		goto got_pg;
2605 	sync_migration = true;
2606 
2607 	/*
2608 	 * If compaction is deferred for high-order allocations, it is because
2609 	 * sync compaction recently failed. In this is the case and the caller
2610 	 * requested a movable allocation that does not heavily disrupt the
2611 	 * system then fail the allocation instead of entering direct reclaim.
2612 	 */
2613 	if ((deferred_compaction || contended_compaction) &&
2614 						(gfp_mask & __GFP_NO_KSWAPD))
2615 		goto nopage;
2616 
2617 	/* Try direct reclaim and then allocating */
2618 	page = __alloc_pages_direct_reclaim(gfp_mask, order,
2619 					zonelist, high_zoneidx,
2620 					nodemask,
2621 					alloc_flags, preferred_zone,
2622 					migratetype, &did_some_progress);
2623 	if (page)
2624 		goto got_pg;
2625 
2626 	/*
2627 	 * If we failed to make any progress reclaiming, then we are
2628 	 * running out of options and have to consider going OOM
2629 	 */
2630 	if (!did_some_progress) {
2631 		if (oom_gfp_allowed(gfp_mask)) {
2632 			if (oom_killer_disabled)
2633 				goto nopage;
2634 			/* Coredumps can quickly deplete all memory reserves */
2635 			if ((current->flags & PF_DUMPCORE) &&
2636 			    !(gfp_mask & __GFP_NOFAIL))
2637 				goto nopage;
2638 			page = __alloc_pages_may_oom(gfp_mask, order,
2639 					zonelist, high_zoneidx,
2640 					nodemask, preferred_zone,
2641 					migratetype);
2642 			if (page)
2643 				goto got_pg;
2644 
2645 			if (!(gfp_mask & __GFP_NOFAIL)) {
2646 				/*
2647 				 * The oom killer is not called for high-order
2648 				 * allocations that may fail, so if no progress
2649 				 * is being made, there are no other options and
2650 				 * retrying is unlikely to help.
2651 				 */
2652 				if (order > PAGE_ALLOC_COSTLY_ORDER)
2653 					goto nopage;
2654 				/*
2655 				 * The oom killer is not called for lowmem
2656 				 * allocations to prevent needlessly killing
2657 				 * innocent tasks.
2658 				 */
2659 				if (high_zoneidx < ZONE_NORMAL)
2660 					goto nopage;
2661 			}
2662 
2663 			goto restart;
2664 		}
2665 	}
2666 
2667 	/* Check if we should retry the allocation */
2668 	pages_reclaimed += did_some_progress;
2669 	if (should_alloc_retry(gfp_mask, order, did_some_progress,
2670 						pages_reclaimed)) {
2671 		/* Wait for some write requests to complete then retry */
2672 		wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50);
2673 		goto rebalance;
2674 	} else {
2675 		/*
2676 		 * High-order allocations do not necessarily loop after
2677 		 * direct reclaim and reclaim/compaction depends on compaction
2678 		 * being called after reclaim so call directly if necessary
2679 		 */
2680 		page = __alloc_pages_direct_compact(gfp_mask, order,
2681 					zonelist, high_zoneidx,
2682 					nodemask,
2683 					alloc_flags, preferred_zone,
2684 					migratetype, sync_migration,
2685 					&contended_compaction,
2686 					&deferred_compaction,
2687 					&did_some_progress);
2688 		if (page)
2689 			goto got_pg;
2690 	}
2691 
2692 nopage:
2693 	warn_alloc_failed(gfp_mask, order, NULL);
2694 	return page;
2695 got_pg:
2696 	if (kmemcheck_enabled)
2697 		kmemcheck_pagealloc_alloc(page, order, gfp_mask);
2698 
2699 	return page;
2700 }
2701 
2702 /*
2703  * This is the 'heart' of the zoned buddy allocator.
2704  */
2705 struct page *
2706 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
2707 			struct zonelist *zonelist, nodemask_t *nodemask)
2708 {
2709 	enum zone_type high_zoneidx = gfp_zone(gfp_mask);
2710 	struct zone *preferred_zone;
2711 	struct page *page = NULL;
2712 	int migratetype = allocflags_to_migratetype(gfp_mask);
2713 	unsigned int cpuset_mems_cookie;
2714 	int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET;
2715 	struct mem_cgroup *memcg = NULL;
2716 
2717 	gfp_mask &= gfp_allowed_mask;
2718 
2719 	lockdep_trace_alloc(gfp_mask);
2720 
2721 	might_sleep_if(gfp_mask & __GFP_WAIT);
2722 
2723 	if (should_fail_alloc_page(gfp_mask, order))
2724 		return NULL;
2725 
2726 	/*
2727 	 * Check the zones suitable for the gfp_mask contain at least one
2728 	 * valid zone. It's possible to have an empty zonelist as a result
2729 	 * of GFP_THISNODE and a memoryless node
2730 	 */
2731 	if (unlikely(!zonelist->_zonerefs->zone))
2732 		return NULL;
2733 
2734 	/*
2735 	 * Will only have any effect when __GFP_KMEMCG is set.  This is
2736 	 * verified in the (always inline) callee
2737 	 */
2738 	if (!memcg_kmem_newpage_charge(gfp_mask, &memcg, order))
2739 		return NULL;
2740 
2741 retry_cpuset:
2742 	cpuset_mems_cookie = get_mems_allowed();
2743 
2744 	/* The preferred zone is used for statistics later */
2745 	first_zones_zonelist(zonelist, high_zoneidx,
2746 				nodemask ? : &cpuset_current_mems_allowed,
2747 				&preferred_zone);
2748 	if (!preferred_zone)
2749 		goto out;
2750 
2751 #ifdef CONFIG_CMA
2752 	if (allocflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
2753 		alloc_flags |= ALLOC_CMA;
2754 #endif
2755 	/* First allocation attempt */
2756 	page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
2757 			zonelist, high_zoneidx, alloc_flags,
2758 			preferred_zone, migratetype);
2759 	if (unlikely(!page)) {
2760 		/*
2761 		 * Runtime PM, block IO and its error handling path
2762 		 * can deadlock because I/O on the device might not
2763 		 * complete.
2764 		 */
2765 		gfp_mask = memalloc_noio_flags(gfp_mask);
2766 		page = __alloc_pages_slowpath(gfp_mask, order,
2767 				zonelist, high_zoneidx, nodemask,
2768 				preferred_zone, migratetype);
2769 	}
2770 
2771 	trace_mm_page_alloc(page, order, gfp_mask, migratetype);
2772 
2773 out:
2774 	/*
2775 	 * When updating a task's mems_allowed, it is possible to race with
2776 	 * parallel threads in such a way that an allocation can fail while
2777 	 * the mask is being updated. If a page allocation is about to fail,
2778 	 * check if the cpuset changed during allocation and if so, retry.
2779 	 */
2780 	if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
2781 		goto retry_cpuset;
2782 
2783 	memcg_kmem_commit_charge(page, memcg, order);
2784 
2785 	return page;
2786 }
2787 EXPORT_SYMBOL(__alloc_pages_nodemask);
2788 
2789 /*
2790  * Common helper functions.
2791  */
2792 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
2793 {
2794 	struct page *page;
2795 
2796 	/*
2797 	 * __get_free_pages() returns a 32-bit address, which cannot represent
2798 	 * a highmem page
2799 	 */
2800 	VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
2801 
2802 	page = alloc_pages(gfp_mask, order);
2803 	if (!page)
2804 		return 0;
2805 	return (unsigned long) page_address(page);
2806 }
2807 EXPORT_SYMBOL(__get_free_pages);
2808 
2809 unsigned long get_zeroed_page(gfp_t gfp_mask)
2810 {
2811 	return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
2812 }
2813 EXPORT_SYMBOL(get_zeroed_page);
2814 
2815 void __free_pages(struct page *page, unsigned int order)
2816 {
2817 	if (put_page_testzero(page)) {
2818 		if (order == 0)
2819 			free_hot_cold_page(page, 0);
2820 		else
2821 			__free_pages_ok(page, order);
2822 	}
2823 }
2824 
2825 EXPORT_SYMBOL(__free_pages);
2826 
2827 void free_pages(unsigned long addr, unsigned int order)
2828 {
2829 	if (addr != 0) {
2830 		VM_BUG_ON(!virt_addr_valid((void *)addr));
2831 		__free_pages(virt_to_page((void *)addr), order);
2832 	}
2833 }
2834 
2835 EXPORT_SYMBOL(free_pages);
2836 
2837 /*
2838  * __free_memcg_kmem_pages and free_memcg_kmem_pages will free
2839  * pages allocated with __GFP_KMEMCG.
2840  *
2841  * Those pages are accounted to a particular memcg, embedded in the
2842  * corresponding page_cgroup. To avoid adding a hit in the allocator to search
2843  * for that information only to find out that it is NULL for users who have no
2844  * interest in that whatsoever, we provide these functions.
2845  *
2846  * The caller knows better which flags it relies on.
2847  */
2848 void __free_memcg_kmem_pages(struct page *page, unsigned int order)
2849 {
2850 	memcg_kmem_uncharge_pages(page, order);
2851 	__free_pages(page, order);
2852 }
2853 
2854 void free_memcg_kmem_pages(unsigned long addr, unsigned int order)
2855 {
2856 	if (addr != 0) {
2857 		VM_BUG_ON(!virt_addr_valid((void *)addr));
2858 		__free_memcg_kmem_pages(virt_to_page((void *)addr), order);
2859 	}
2860 }
2861 
2862 static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size)
2863 {
2864 	if (addr) {
2865 		unsigned long alloc_end = addr + (PAGE_SIZE << order);
2866 		unsigned long used = addr + PAGE_ALIGN(size);
2867 
2868 		split_page(virt_to_page((void *)addr), order);
2869 		while (used < alloc_end) {
2870 			free_page(used);
2871 			used += PAGE_SIZE;
2872 		}
2873 	}
2874 	return (void *)addr;
2875 }
2876 
2877 /**
2878  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
2879  * @size: the number of bytes to allocate
2880  * @gfp_mask: GFP flags for the allocation
2881  *
2882  * This function is similar to alloc_pages(), except that it allocates the
2883  * minimum number of pages to satisfy the request.  alloc_pages() can only
2884  * allocate memory in power-of-two pages.
2885  *
2886  * This function is also limited by MAX_ORDER.
2887  *
2888  * Memory allocated by this function must be released by free_pages_exact().
2889  */
2890 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
2891 {
2892 	unsigned int order = get_order(size);
2893 	unsigned long addr;
2894 
2895 	addr = __get_free_pages(gfp_mask, order);
2896 	return make_alloc_exact(addr, order, size);
2897 }
2898 EXPORT_SYMBOL(alloc_pages_exact);
2899 
2900 /**
2901  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
2902  *			   pages on a node.
2903  * @nid: the preferred node ID where memory should be allocated
2904  * @size: the number of bytes to allocate
2905  * @gfp_mask: GFP flags for the allocation
2906  *
2907  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
2908  * back.
2909  * Note this is not alloc_pages_exact_node() which allocates on a specific node,
2910  * but is not exact.
2911  */
2912 void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
2913 {
2914 	unsigned order = get_order(size);
2915 	struct page *p = alloc_pages_node(nid, gfp_mask, order);
2916 	if (!p)
2917 		return NULL;
2918 	return make_alloc_exact((unsigned long)page_address(p), order, size);
2919 }
2920 EXPORT_SYMBOL(alloc_pages_exact_nid);
2921 
2922 /**
2923  * free_pages_exact - release memory allocated via alloc_pages_exact()
2924  * @virt: the value returned by alloc_pages_exact.
2925  * @size: size of allocation, same value as passed to alloc_pages_exact().
2926  *
2927  * Release the memory allocated by a previous call to alloc_pages_exact.
2928  */
2929 void free_pages_exact(void *virt, size_t size)
2930 {
2931 	unsigned long addr = (unsigned long)virt;
2932 	unsigned long end = addr + PAGE_ALIGN(size);
2933 
2934 	while (addr < end) {
2935 		free_page(addr);
2936 		addr += PAGE_SIZE;
2937 	}
2938 }
2939 EXPORT_SYMBOL(free_pages_exact);
2940 
2941 /**
2942  * nr_free_zone_pages - count number of pages beyond high watermark
2943  * @offset: The zone index of the highest zone
2944  *
2945  * nr_free_zone_pages() counts the number of counts pages which are beyond the
2946  * high watermark within all zones at or below a given zone index.  For each
2947  * zone, the number of pages is calculated as:
2948  *     managed_pages - high_pages
2949  */
2950 static unsigned long nr_free_zone_pages(int offset)
2951 {
2952 	struct zoneref *z;
2953 	struct zone *zone;
2954 
2955 	/* Just pick one node, since fallback list is circular */
2956 	unsigned long sum = 0;
2957 
2958 	struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
2959 
2960 	for_each_zone_zonelist(zone, z, zonelist, offset) {
2961 		unsigned long size = zone->managed_pages;
2962 		unsigned long high = high_wmark_pages(zone);
2963 		if (size > high)
2964 			sum += size - high;
2965 	}
2966 
2967 	return sum;
2968 }
2969 
2970 /**
2971  * nr_free_buffer_pages - count number of pages beyond high watermark
2972  *
2973  * nr_free_buffer_pages() counts the number of pages which are beyond the high
2974  * watermark within ZONE_DMA and ZONE_NORMAL.
2975  */
2976 unsigned long nr_free_buffer_pages(void)
2977 {
2978 	return nr_free_zone_pages(gfp_zone(GFP_USER));
2979 }
2980 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
2981 
2982 /**
2983  * nr_free_pagecache_pages - count number of pages beyond high watermark
2984  *
2985  * nr_free_pagecache_pages() counts the number of pages which are beyond the
2986  * high watermark within all zones.
2987  */
2988 unsigned long nr_free_pagecache_pages(void)
2989 {
2990 	return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
2991 }
2992 
2993 static inline void show_node(struct zone *zone)
2994 {
2995 	if (IS_ENABLED(CONFIG_NUMA))
2996 		printk("Node %d ", zone_to_nid(zone));
2997 }
2998 
2999 void si_meminfo(struct sysinfo *val)
3000 {
3001 	val->totalram = totalram_pages;
3002 	val->sharedram = 0;
3003 	val->freeram = global_page_state(NR_FREE_PAGES);
3004 	val->bufferram = nr_blockdev_pages();
3005 	val->totalhigh = totalhigh_pages;
3006 	val->freehigh = nr_free_highpages();
3007 	val->mem_unit = PAGE_SIZE;
3008 }
3009 
3010 EXPORT_SYMBOL(si_meminfo);
3011 
3012 #ifdef CONFIG_NUMA
3013 void si_meminfo_node(struct sysinfo *val, int nid)
3014 {
3015 	int zone_type;		/* needs to be signed */
3016 	unsigned long managed_pages = 0;
3017 	pg_data_t *pgdat = NODE_DATA(nid);
3018 
3019 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3020 		managed_pages += pgdat->node_zones[zone_type].managed_pages;
3021 	val->totalram = managed_pages;
3022 	val->freeram = node_page_state(nid, NR_FREE_PAGES);
3023 #ifdef CONFIG_HIGHMEM
3024 	val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].managed_pages;
3025 	val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
3026 			NR_FREE_PAGES);
3027 #else
3028 	val->totalhigh = 0;
3029 	val->freehigh = 0;
3030 #endif
3031 	val->mem_unit = PAGE_SIZE;
3032 }
3033 #endif
3034 
3035 /*
3036  * Determine whether the node should be displayed or not, depending on whether
3037  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3038  */
3039 bool skip_free_areas_node(unsigned int flags, int nid)
3040 {
3041 	bool ret = false;
3042 	unsigned int cpuset_mems_cookie;
3043 
3044 	if (!(flags & SHOW_MEM_FILTER_NODES))
3045 		goto out;
3046 
3047 	do {
3048 		cpuset_mems_cookie = get_mems_allowed();
3049 		ret = !node_isset(nid, cpuset_current_mems_allowed);
3050 	} while (!put_mems_allowed(cpuset_mems_cookie));
3051 out:
3052 	return ret;
3053 }
3054 
3055 #define K(x) ((x) << (PAGE_SHIFT-10))
3056 
3057 static void show_migration_types(unsigned char type)
3058 {
3059 	static const char types[MIGRATE_TYPES] = {
3060 		[MIGRATE_UNMOVABLE]	= 'U',
3061 		[MIGRATE_RECLAIMABLE]	= 'E',
3062 		[MIGRATE_MOVABLE]	= 'M',
3063 		[MIGRATE_RESERVE]	= 'R',
3064 #ifdef CONFIG_CMA
3065 		[MIGRATE_CMA]		= 'C',
3066 #endif
3067 #ifdef CONFIG_MEMORY_ISOLATION
3068 		[MIGRATE_ISOLATE]	= 'I',
3069 #endif
3070 	};
3071 	char tmp[MIGRATE_TYPES + 1];
3072 	char *p = tmp;
3073 	int i;
3074 
3075 	for (i = 0; i < MIGRATE_TYPES; i++) {
3076 		if (type & (1 << i))
3077 			*p++ = types[i];
3078 	}
3079 
3080 	*p = '\0';
3081 	printk("(%s) ", tmp);
3082 }
3083 
3084 /*
3085  * Show free area list (used inside shift_scroll-lock stuff)
3086  * We also calculate the percentage fragmentation. We do this by counting the
3087  * memory on each free list with the exception of the first item on the list.
3088  * Suppresses nodes that are not allowed by current's cpuset if
3089  * SHOW_MEM_FILTER_NODES is passed.
3090  */
3091 void show_free_areas(unsigned int filter)
3092 {
3093 	int cpu;
3094 	struct zone *zone;
3095 
3096 	for_each_populated_zone(zone) {
3097 		if (skip_free_areas_node(filter, zone_to_nid(zone)))
3098 			continue;
3099 		show_node(zone);
3100 		printk("%s per-cpu:\n", zone->name);
3101 
3102 		for_each_online_cpu(cpu) {
3103 			struct per_cpu_pageset *pageset;
3104 
3105 			pageset = per_cpu_ptr(zone->pageset, cpu);
3106 
3107 			printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
3108 			       cpu, pageset->pcp.high,
3109 			       pageset->pcp.batch, pageset->pcp.count);
3110 		}
3111 	}
3112 
3113 	printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3114 		" active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3115 		" unevictable:%lu"
3116 		" dirty:%lu writeback:%lu unstable:%lu\n"
3117 		" free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3118 		" mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3119 		" free_cma:%lu\n",
3120 		global_page_state(NR_ACTIVE_ANON),
3121 		global_page_state(NR_INACTIVE_ANON),
3122 		global_page_state(NR_ISOLATED_ANON),
3123 		global_page_state(NR_ACTIVE_FILE),
3124 		global_page_state(NR_INACTIVE_FILE),
3125 		global_page_state(NR_ISOLATED_FILE),
3126 		global_page_state(NR_UNEVICTABLE),
3127 		global_page_state(NR_FILE_DIRTY),
3128 		global_page_state(NR_WRITEBACK),
3129 		global_page_state(NR_UNSTABLE_NFS),
3130 		global_page_state(NR_FREE_PAGES),
3131 		global_page_state(NR_SLAB_RECLAIMABLE),
3132 		global_page_state(NR_SLAB_UNRECLAIMABLE),
3133 		global_page_state(NR_FILE_MAPPED),
3134 		global_page_state(NR_SHMEM),
3135 		global_page_state(NR_PAGETABLE),
3136 		global_page_state(NR_BOUNCE),
3137 		global_page_state(NR_FREE_CMA_PAGES));
3138 
3139 	for_each_populated_zone(zone) {
3140 		int i;
3141 
3142 		if (skip_free_areas_node(filter, zone_to_nid(zone)))
3143 			continue;
3144 		show_node(zone);
3145 		printk("%s"
3146 			" free:%lukB"
3147 			" min:%lukB"
3148 			" low:%lukB"
3149 			" high:%lukB"
3150 			" active_anon:%lukB"
3151 			" inactive_anon:%lukB"
3152 			" active_file:%lukB"
3153 			" inactive_file:%lukB"
3154 			" unevictable:%lukB"
3155 			" isolated(anon):%lukB"
3156 			" isolated(file):%lukB"
3157 			" present:%lukB"
3158 			" managed:%lukB"
3159 			" mlocked:%lukB"
3160 			" dirty:%lukB"
3161 			" writeback:%lukB"
3162 			" mapped:%lukB"
3163 			" shmem:%lukB"
3164 			" slab_reclaimable:%lukB"
3165 			" slab_unreclaimable:%lukB"
3166 			" kernel_stack:%lukB"
3167 			" pagetables:%lukB"
3168 			" unstable:%lukB"
3169 			" bounce:%lukB"
3170 			" free_cma:%lukB"
3171 			" writeback_tmp:%lukB"
3172 			" pages_scanned:%lu"
3173 			" all_unreclaimable? %s"
3174 			"\n",
3175 			zone->name,
3176 			K(zone_page_state(zone, NR_FREE_PAGES)),
3177 			K(min_wmark_pages(zone)),
3178 			K(low_wmark_pages(zone)),
3179 			K(high_wmark_pages(zone)),
3180 			K(zone_page_state(zone, NR_ACTIVE_ANON)),
3181 			K(zone_page_state(zone, NR_INACTIVE_ANON)),
3182 			K(zone_page_state(zone, NR_ACTIVE_FILE)),
3183 			K(zone_page_state(zone, NR_INACTIVE_FILE)),
3184 			K(zone_page_state(zone, NR_UNEVICTABLE)),
3185 			K(zone_page_state(zone, NR_ISOLATED_ANON)),
3186 			K(zone_page_state(zone, NR_ISOLATED_FILE)),
3187 			K(zone->present_pages),
3188 			K(zone->managed_pages),
3189 			K(zone_page_state(zone, NR_MLOCK)),
3190 			K(zone_page_state(zone, NR_FILE_DIRTY)),
3191 			K(zone_page_state(zone, NR_WRITEBACK)),
3192 			K(zone_page_state(zone, NR_FILE_MAPPED)),
3193 			K(zone_page_state(zone, NR_SHMEM)),
3194 			K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3195 			K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3196 			zone_page_state(zone, NR_KERNEL_STACK) *
3197 				THREAD_SIZE / 1024,
3198 			K(zone_page_state(zone, NR_PAGETABLE)),
3199 			K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3200 			K(zone_page_state(zone, NR_BOUNCE)),
3201 			K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3202 			K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3203 			zone->pages_scanned,
3204 			(!zone_reclaimable(zone) ? "yes" : "no")
3205 			);
3206 		printk("lowmem_reserve[]:");
3207 		for (i = 0; i < MAX_NR_ZONES; i++)
3208 			printk(" %lu", zone->lowmem_reserve[i]);
3209 		printk("\n");
3210 	}
3211 
3212 	for_each_populated_zone(zone) {
3213 		unsigned long nr[MAX_ORDER], flags, order, total = 0;
3214 		unsigned char types[MAX_ORDER];
3215 
3216 		if (skip_free_areas_node(filter, zone_to_nid(zone)))
3217 			continue;
3218 		show_node(zone);
3219 		printk("%s: ", zone->name);
3220 
3221 		spin_lock_irqsave(&zone->lock, flags);
3222 		for (order = 0; order < MAX_ORDER; order++) {
3223 			struct free_area *area = &zone->free_area[order];
3224 			int type;
3225 
3226 			nr[order] = area->nr_free;
3227 			total += nr[order] << order;
3228 
3229 			types[order] = 0;
3230 			for (type = 0; type < MIGRATE_TYPES; type++) {
3231 				if (!list_empty(&area->free_list[type]))
3232 					types[order] |= 1 << type;
3233 			}
3234 		}
3235 		spin_unlock_irqrestore(&zone->lock, flags);
3236 		for (order = 0; order < MAX_ORDER; order++) {
3237 			printk("%lu*%lukB ", nr[order], K(1UL) << order);
3238 			if (nr[order])
3239 				show_migration_types(types[order]);
3240 		}
3241 		printk("= %lukB\n", K(total));
3242 	}
3243 
3244 	hugetlb_show_meminfo();
3245 
3246 	printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
3247 
3248 	show_swap_cache_info();
3249 }
3250 
3251 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
3252 {
3253 	zoneref->zone = zone;
3254 	zoneref->zone_idx = zone_idx(zone);
3255 }
3256 
3257 /*
3258  * Builds allocation fallback zone lists.
3259  *
3260  * Add all populated zones of a node to the zonelist.
3261  */
3262 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
3263 				int nr_zones)
3264 {
3265 	struct zone *zone;
3266 	enum zone_type zone_type = MAX_NR_ZONES;
3267 
3268 	do {
3269 		zone_type--;
3270 		zone = pgdat->node_zones + zone_type;
3271 		if (populated_zone(zone)) {
3272 			zoneref_set_zone(zone,
3273 				&zonelist->_zonerefs[nr_zones++]);
3274 			check_highest_zone(zone_type);
3275 		}
3276 	} while (zone_type);
3277 
3278 	return nr_zones;
3279 }
3280 
3281 
3282 /*
3283  *  zonelist_order:
3284  *  0 = automatic detection of better ordering.
3285  *  1 = order by ([node] distance, -zonetype)
3286  *  2 = order by (-zonetype, [node] distance)
3287  *
3288  *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3289  *  the same zonelist. So only NUMA can configure this param.
3290  */
3291 #define ZONELIST_ORDER_DEFAULT  0
3292 #define ZONELIST_ORDER_NODE     1
3293 #define ZONELIST_ORDER_ZONE     2
3294 
3295 /* zonelist order in the kernel.
3296  * set_zonelist_order() will set this to NODE or ZONE.
3297  */
3298 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
3299 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
3300 
3301 
3302 #ifdef CONFIG_NUMA
3303 /* The value user specified ....changed by config */
3304 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3305 /* string for sysctl */
3306 #define NUMA_ZONELIST_ORDER_LEN	16
3307 char numa_zonelist_order[16] = "default";
3308 
3309 /*
3310  * interface for configure zonelist ordering.
3311  * command line option "numa_zonelist_order"
3312  *	= "[dD]efault	- default, automatic configuration.
3313  *	= "[nN]ode 	- order by node locality, then by zone within node
3314  *	= "[zZ]one      - order by zone, then by locality within zone
3315  */
3316 
3317 static int __parse_numa_zonelist_order(char *s)
3318 {
3319 	if (*s == 'd' || *s == 'D') {
3320 		user_zonelist_order = ZONELIST_ORDER_DEFAULT;
3321 	} else if (*s == 'n' || *s == 'N') {
3322 		user_zonelist_order = ZONELIST_ORDER_NODE;
3323 	} else if (*s == 'z' || *s == 'Z') {
3324 		user_zonelist_order = ZONELIST_ORDER_ZONE;
3325 	} else {
3326 		printk(KERN_WARNING
3327 			"Ignoring invalid numa_zonelist_order value:  "
3328 			"%s\n", s);
3329 		return -EINVAL;
3330 	}
3331 	return 0;
3332 }
3333 
3334 static __init int setup_numa_zonelist_order(char *s)
3335 {
3336 	int ret;
3337 
3338 	if (!s)
3339 		return 0;
3340 
3341 	ret = __parse_numa_zonelist_order(s);
3342 	if (ret == 0)
3343 		strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
3344 
3345 	return ret;
3346 }
3347 early_param("numa_zonelist_order", setup_numa_zonelist_order);
3348 
3349 /*
3350  * sysctl handler for numa_zonelist_order
3351  */
3352 int numa_zonelist_order_handler(ctl_table *table, int write,
3353 		void __user *buffer, size_t *length,
3354 		loff_t *ppos)
3355 {
3356 	char saved_string[NUMA_ZONELIST_ORDER_LEN];
3357 	int ret;
3358 	static DEFINE_MUTEX(zl_order_mutex);
3359 
3360 	mutex_lock(&zl_order_mutex);
3361 	if (write) {
3362 		if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
3363 			ret = -EINVAL;
3364 			goto out;
3365 		}
3366 		strcpy(saved_string, (char *)table->data);
3367 	}
3368 	ret = proc_dostring(table, write, buffer, length, ppos);
3369 	if (ret)
3370 		goto out;
3371 	if (write) {
3372 		int oldval = user_zonelist_order;
3373 
3374 		ret = __parse_numa_zonelist_order((char *)table->data);
3375 		if (ret) {
3376 			/*
3377 			 * bogus value.  restore saved string
3378 			 */
3379 			strncpy((char *)table->data, saved_string,
3380 				NUMA_ZONELIST_ORDER_LEN);
3381 			user_zonelist_order = oldval;
3382 		} else if (oldval != user_zonelist_order) {
3383 			mutex_lock(&zonelists_mutex);
3384 			build_all_zonelists(NULL, NULL);
3385 			mutex_unlock(&zonelists_mutex);
3386 		}
3387 	}
3388 out:
3389 	mutex_unlock(&zl_order_mutex);
3390 	return ret;
3391 }
3392 
3393 
3394 #define MAX_NODE_LOAD (nr_online_nodes)
3395 static int node_load[MAX_NUMNODES];
3396 
3397 /**
3398  * find_next_best_node - find the next node that should appear in a given node's fallback list
3399  * @node: node whose fallback list we're appending
3400  * @used_node_mask: nodemask_t of already used nodes
3401  *
3402  * We use a number of factors to determine which is the next node that should
3403  * appear on a given node's fallback list.  The node should not have appeared
3404  * already in @node's fallback list, and it should be the next closest node
3405  * according to the distance array (which contains arbitrary distance values
3406  * from each node to each node in the system), and should also prefer nodes
3407  * with no CPUs, since presumably they'll have very little allocation pressure
3408  * on them otherwise.
3409  * It returns -1 if no node is found.
3410  */
3411 static int find_next_best_node(int node, nodemask_t *used_node_mask)
3412 {
3413 	int n, val;
3414 	int min_val = INT_MAX;
3415 	int best_node = NUMA_NO_NODE;
3416 	const struct cpumask *tmp = cpumask_of_node(0);
3417 
3418 	/* Use the local node if we haven't already */
3419 	if (!node_isset(node, *used_node_mask)) {
3420 		node_set(node, *used_node_mask);
3421 		return node;
3422 	}
3423 
3424 	for_each_node_state(n, N_MEMORY) {
3425 
3426 		/* Don't want a node to appear more than once */
3427 		if (node_isset(n, *used_node_mask))
3428 			continue;
3429 
3430 		/* Use the distance array to find the distance */
3431 		val = node_distance(node, n);
3432 
3433 		/* Penalize nodes under us ("prefer the next node") */
3434 		val += (n < node);
3435 
3436 		/* Give preference to headless and unused nodes */
3437 		tmp = cpumask_of_node(n);
3438 		if (!cpumask_empty(tmp))
3439 			val += PENALTY_FOR_NODE_WITH_CPUS;
3440 
3441 		/* Slight preference for less loaded node */
3442 		val *= (MAX_NODE_LOAD*MAX_NUMNODES);
3443 		val += node_load[n];
3444 
3445 		if (val < min_val) {
3446 			min_val = val;
3447 			best_node = n;
3448 		}
3449 	}
3450 
3451 	if (best_node >= 0)
3452 		node_set(best_node, *used_node_mask);
3453 
3454 	return best_node;
3455 }
3456 
3457 
3458 /*
3459  * Build zonelists ordered by node and zones within node.
3460  * This results in maximum locality--normal zone overflows into local
3461  * DMA zone, if any--but risks exhausting DMA zone.
3462  */
3463 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
3464 {
3465 	int j;
3466 	struct zonelist *zonelist;
3467 
3468 	zonelist = &pgdat->node_zonelists[0];
3469 	for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
3470 		;
3471 	j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3472 	zonelist->_zonerefs[j].zone = NULL;
3473 	zonelist->_zonerefs[j].zone_idx = 0;
3474 }
3475 
3476 /*
3477  * Build gfp_thisnode zonelists
3478  */
3479 static void build_thisnode_zonelists(pg_data_t *pgdat)
3480 {
3481 	int j;
3482 	struct zonelist *zonelist;
3483 
3484 	zonelist = &pgdat->node_zonelists[1];
3485 	j = build_zonelists_node(pgdat, zonelist, 0);
3486 	zonelist->_zonerefs[j].zone = NULL;
3487 	zonelist->_zonerefs[j].zone_idx = 0;
3488 }
3489 
3490 /*
3491  * Build zonelists ordered by zone and nodes within zones.
3492  * This results in conserving DMA zone[s] until all Normal memory is
3493  * exhausted, but results in overflowing to remote node while memory
3494  * may still exist in local DMA zone.
3495  */
3496 static int node_order[MAX_NUMNODES];
3497 
3498 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
3499 {
3500 	int pos, j, node;
3501 	int zone_type;		/* needs to be signed */
3502 	struct zone *z;
3503 	struct zonelist *zonelist;
3504 
3505 	zonelist = &pgdat->node_zonelists[0];
3506 	pos = 0;
3507 	for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
3508 		for (j = 0; j < nr_nodes; j++) {
3509 			node = node_order[j];
3510 			z = &NODE_DATA(node)->node_zones[zone_type];
3511 			if (populated_zone(z)) {
3512 				zoneref_set_zone(z,
3513 					&zonelist->_zonerefs[pos++]);
3514 				check_highest_zone(zone_type);
3515 			}
3516 		}
3517 	}
3518 	zonelist->_zonerefs[pos].zone = NULL;
3519 	zonelist->_zonerefs[pos].zone_idx = 0;
3520 }
3521 
3522 static int default_zonelist_order(void)
3523 {
3524 	int nid, zone_type;
3525 	unsigned long low_kmem_size, total_size;
3526 	struct zone *z;
3527 	int average_size;
3528 	/*
3529 	 * ZONE_DMA and ZONE_DMA32 can be very small area in the system.
3530 	 * If they are really small and used heavily, the system can fall
3531 	 * into OOM very easily.
3532 	 * This function detect ZONE_DMA/DMA32 size and configures zone order.
3533 	 */
3534 	/* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
3535 	low_kmem_size = 0;
3536 	total_size = 0;
3537 	for_each_online_node(nid) {
3538 		for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3539 			z = &NODE_DATA(nid)->node_zones[zone_type];
3540 			if (populated_zone(z)) {
3541 				if (zone_type < ZONE_NORMAL)
3542 					low_kmem_size += z->managed_pages;
3543 				total_size += z->managed_pages;
3544 			} else if (zone_type == ZONE_NORMAL) {
3545 				/*
3546 				 * If any node has only lowmem, then node order
3547 				 * is preferred to allow kernel allocations
3548 				 * locally; otherwise, they can easily infringe
3549 				 * on other nodes when there is an abundance of
3550 				 * lowmem available to allocate from.
3551 				 */
3552 				return ZONELIST_ORDER_NODE;
3553 			}
3554 		}
3555 	}
3556 	if (!low_kmem_size ||  /* there are no DMA area. */
3557 	    low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
3558 		return ZONELIST_ORDER_NODE;
3559 	/*
3560 	 * look into each node's config.
3561 	 * If there is a node whose DMA/DMA32 memory is very big area on
3562 	 * local memory, NODE_ORDER may be suitable.
3563 	 */
3564 	average_size = total_size /
3565 				(nodes_weight(node_states[N_MEMORY]) + 1);
3566 	for_each_online_node(nid) {
3567 		low_kmem_size = 0;
3568 		total_size = 0;
3569 		for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3570 			z = &NODE_DATA(nid)->node_zones[zone_type];
3571 			if (populated_zone(z)) {
3572 				if (zone_type < ZONE_NORMAL)
3573 					low_kmem_size += z->present_pages;
3574 				total_size += z->present_pages;
3575 			}
3576 		}
3577 		if (low_kmem_size &&
3578 		    total_size > average_size && /* ignore small node */
3579 		    low_kmem_size > total_size * 70/100)
3580 			return ZONELIST_ORDER_NODE;
3581 	}
3582 	return ZONELIST_ORDER_ZONE;
3583 }
3584 
3585 static void set_zonelist_order(void)
3586 {
3587 	if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
3588 		current_zonelist_order = default_zonelist_order();
3589 	else
3590 		current_zonelist_order = user_zonelist_order;
3591 }
3592 
3593 static void build_zonelists(pg_data_t *pgdat)
3594 {
3595 	int j, node, load;
3596 	enum zone_type i;
3597 	nodemask_t used_mask;
3598 	int local_node, prev_node;
3599 	struct zonelist *zonelist;
3600 	int order = current_zonelist_order;
3601 
3602 	/* initialize zonelists */
3603 	for (i = 0; i < MAX_ZONELISTS; i++) {
3604 		zonelist = pgdat->node_zonelists + i;
3605 		zonelist->_zonerefs[0].zone = NULL;
3606 		zonelist->_zonerefs[0].zone_idx = 0;
3607 	}
3608 
3609 	/* NUMA-aware ordering of nodes */
3610 	local_node = pgdat->node_id;
3611 	load = nr_online_nodes;
3612 	prev_node = local_node;
3613 	nodes_clear(used_mask);
3614 
3615 	memset(node_order, 0, sizeof(node_order));
3616 	j = 0;
3617 
3618 	while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
3619 		/*
3620 		 * We don't want to pressure a particular node.
3621 		 * So adding penalty to the first node in same
3622 		 * distance group to make it round-robin.
3623 		 */
3624 		if (node_distance(local_node, node) !=
3625 		    node_distance(local_node, prev_node))
3626 			node_load[node] = load;
3627 
3628 		prev_node = node;
3629 		load--;
3630 		if (order == ZONELIST_ORDER_NODE)
3631 			build_zonelists_in_node_order(pgdat, node);
3632 		else
3633 			node_order[j++] = node;	/* remember order */
3634 	}
3635 
3636 	if (order == ZONELIST_ORDER_ZONE) {
3637 		/* calculate node order -- i.e., DMA last! */
3638 		build_zonelists_in_zone_order(pgdat, j);
3639 	}
3640 
3641 	build_thisnode_zonelists(pgdat);
3642 }
3643 
3644 /* Construct the zonelist performance cache - see further mmzone.h */
3645 static void build_zonelist_cache(pg_data_t *pgdat)
3646 {
3647 	struct zonelist *zonelist;
3648 	struct zonelist_cache *zlc;
3649 	struct zoneref *z;
3650 
3651 	zonelist = &pgdat->node_zonelists[0];
3652 	zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
3653 	bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
3654 	for (z = zonelist->_zonerefs; z->zone; z++)
3655 		zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
3656 }
3657 
3658 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3659 /*
3660  * Return node id of node used for "local" allocations.
3661  * I.e., first node id of first zone in arg node's generic zonelist.
3662  * Used for initializing percpu 'numa_mem', which is used primarily
3663  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
3664  */
3665 int local_memory_node(int node)
3666 {
3667 	struct zone *zone;
3668 
3669 	(void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
3670 				   gfp_zone(GFP_KERNEL),
3671 				   NULL,
3672 				   &zone);
3673 	return zone->node;
3674 }
3675 #endif
3676 
3677 #else	/* CONFIG_NUMA */
3678 
3679 static void set_zonelist_order(void)
3680 {
3681 	current_zonelist_order = ZONELIST_ORDER_ZONE;
3682 }
3683 
3684 static void build_zonelists(pg_data_t *pgdat)
3685 {
3686 	int node, local_node;
3687 	enum zone_type j;
3688 	struct zonelist *zonelist;
3689 
3690 	local_node = pgdat->node_id;
3691 
3692 	zonelist = &pgdat->node_zonelists[0];
3693 	j = build_zonelists_node(pgdat, zonelist, 0);
3694 
3695 	/*
3696 	 * Now we build the zonelist so that it contains the zones
3697 	 * of all the other nodes.
3698 	 * We don't want to pressure a particular node, so when
3699 	 * building the zones for node N, we make sure that the
3700 	 * zones coming right after the local ones are those from
3701 	 * node N+1 (modulo N)
3702 	 */
3703 	for (node = local_node + 1; node < MAX_NUMNODES; node++) {
3704 		if (!node_online(node))
3705 			continue;
3706 		j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3707 	}
3708 	for (node = 0; node < local_node; node++) {
3709 		if (!node_online(node))
3710 			continue;
3711 		j = build_zonelists_node(NODE_DATA(node), zonelist, j);
3712 	}
3713 
3714 	zonelist->_zonerefs[j].zone = NULL;
3715 	zonelist->_zonerefs[j].zone_idx = 0;
3716 }
3717 
3718 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
3719 static void build_zonelist_cache(pg_data_t *pgdat)
3720 {
3721 	pgdat->node_zonelists[0].zlcache_ptr = NULL;
3722 }
3723 
3724 #endif	/* CONFIG_NUMA */
3725 
3726 /*
3727  * Boot pageset table. One per cpu which is going to be used for all
3728  * zones and all nodes. The parameters will be set in such a way
3729  * that an item put on a list will immediately be handed over to
3730  * the buddy list. This is safe since pageset manipulation is done
3731  * with interrupts disabled.
3732  *
3733  * The boot_pagesets must be kept even after bootup is complete for
3734  * unused processors and/or zones. They do play a role for bootstrapping
3735  * hotplugged processors.
3736  *
3737  * zoneinfo_show() and maybe other functions do
3738  * not check if the processor is online before following the pageset pointer.
3739  * Other parts of the kernel may not check if the zone is available.
3740  */
3741 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
3742 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
3743 static void setup_zone_pageset(struct zone *zone);
3744 
3745 /*
3746  * Global mutex to protect against size modification of zonelists
3747  * as well as to serialize pageset setup for the new populated zone.
3748  */
3749 DEFINE_MUTEX(zonelists_mutex);
3750 
3751 /* return values int ....just for stop_machine() */
3752 static int __build_all_zonelists(void *data)
3753 {
3754 	int nid;
3755 	int cpu;
3756 	pg_data_t *self = data;
3757 
3758 #ifdef CONFIG_NUMA
3759 	memset(node_load, 0, sizeof(node_load));
3760 #endif
3761 
3762 	if (self && !node_online(self->node_id)) {
3763 		build_zonelists(self);
3764 		build_zonelist_cache(self);
3765 	}
3766 
3767 	for_each_online_node(nid) {
3768 		pg_data_t *pgdat = NODE_DATA(nid);
3769 
3770 		build_zonelists(pgdat);
3771 		build_zonelist_cache(pgdat);
3772 	}
3773 
3774 	/*
3775 	 * Initialize the boot_pagesets that are going to be used
3776 	 * for bootstrapping processors. The real pagesets for
3777 	 * each zone will be allocated later when the per cpu
3778 	 * allocator is available.
3779 	 *
3780 	 * boot_pagesets are used also for bootstrapping offline
3781 	 * cpus if the system is already booted because the pagesets
3782 	 * are needed to initialize allocators on a specific cpu too.
3783 	 * F.e. the percpu allocator needs the page allocator which
3784 	 * needs the percpu allocator in order to allocate its pagesets
3785 	 * (a chicken-egg dilemma).
3786 	 */
3787 	for_each_possible_cpu(cpu) {
3788 		setup_pageset(&per_cpu(boot_pageset, cpu), 0);
3789 
3790 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
3791 		/*
3792 		 * We now know the "local memory node" for each node--
3793 		 * i.e., the node of the first zone in the generic zonelist.
3794 		 * Set up numa_mem percpu variable for on-line cpus.  During
3795 		 * boot, only the boot cpu should be on-line;  we'll init the
3796 		 * secondary cpus' numa_mem as they come on-line.  During
3797 		 * node/memory hotplug, we'll fixup all on-line cpus.
3798 		 */
3799 		if (cpu_online(cpu))
3800 			set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
3801 #endif
3802 	}
3803 
3804 	return 0;
3805 }
3806 
3807 /*
3808  * Called with zonelists_mutex held always
3809  * unless system_state == SYSTEM_BOOTING.
3810  */
3811 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
3812 {
3813 	set_zonelist_order();
3814 
3815 	if (system_state == SYSTEM_BOOTING) {
3816 		__build_all_zonelists(NULL);
3817 		mminit_verify_zonelist();
3818 		cpuset_init_current_mems_allowed();
3819 	} else {
3820 #ifdef CONFIG_MEMORY_HOTPLUG
3821 		if (zone)
3822 			setup_zone_pageset(zone);
3823 #endif
3824 		/* we have to stop all cpus to guarantee there is no user
3825 		   of zonelist */
3826 		stop_machine(__build_all_zonelists, pgdat, NULL);
3827 		/* cpuset refresh routine should be here */
3828 	}
3829 	vm_total_pages = nr_free_pagecache_pages();
3830 	/*
3831 	 * Disable grouping by mobility if the number of pages in the
3832 	 * system is too low to allow the mechanism to work. It would be
3833 	 * more accurate, but expensive to check per-zone. This check is
3834 	 * made on memory-hotadd so a system can start with mobility
3835 	 * disabled and enable it later
3836 	 */
3837 	if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
3838 		page_group_by_mobility_disabled = 1;
3839 	else
3840 		page_group_by_mobility_disabled = 0;
3841 
3842 	printk("Built %i zonelists in %s order, mobility grouping %s.  "
3843 		"Total pages: %ld\n",
3844 			nr_online_nodes,
3845 			zonelist_order_name[current_zonelist_order],
3846 			page_group_by_mobility_disabled ? "off" : "on",
3847 			vm_total_pages);
3848 #ifdef CONFIG_NUMA
3849 	printk("Policy zone: %s\n", zone_names[policy_zone]);
3850 #endif
3851 }
3852 
3853 /*
3854  * Helper functions to size the waitqueue hash table.
3855  * Essentially these want to choose hash table sizes sufficiently
3856  * large so that collisions trying to wait on pages are rare.
3857  * But in fact, the number of active page waitqueues on typical
3858  * systems is ridiculously low, less than 200. So this is even
3859  * conservative, even though it seems large.
3860  *
3861  * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
3862  * waitqueues, i.e. the size of the waitq table given the number of pages.
3863  */
3864 #define PAGES_PER_WAITQUEUE	256
3865 
3866 #ifndef CONFIG_MEMORY_HOTPLUG
3867 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3868 {
3869 	unsigned long size = 1;
3870 
3871 	pages /= PAGES_PER_WAITQUEUE;
3872 
3873 	while (size < pages)
3874 		size <<= 1;
3875 
3876 	/*
3877 	 * Once we have dozens or even hundreds of threads sleeping
3878 	 * on IO we've got bigger problems than wait queue collision.
3879 	 * Limit the size of the wait table to a reasonable size.
3880 	 */
3881 	size = min(size, 4096UL);
3882 
3883 	return max(size, 4UL);
3884 }
3885 #else
3886 /*
3887  * A zone's size might be changed by hot-add, so it is not possible to determine
3888  * a suitable size for its wait_table.  So we use the maximum size now.
3889  *
3890  * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
3891  *
3892  *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
3893  *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
3894  *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
3895  *
3896  * The maximum entries are prepared when a zone's memory is (512K + 256) pages
3897  * or more by the traditional way. (See above).  It equals:
3898  *
3899  *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
3900  *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
3901  *    powerpc (64K page size)             : =  (32G +16M)byte.
3902  */
3903 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
3904 {
3905 	return 4096UL;
3906 }
3907 #endif
3908 
3909 /*
3910  * This is an integer logarithm so that shifts can be used later
3911  * to extract the more random high bits from the multiplicative
3912  * hash function before the remainder is taken.
3913  */
3914 static inline unsigned long wait_table_bits(unsigned long size)
3915 {
3916 	return ffz(~size);
3917 }
3918 
3919 /*
3920  * Check if a pageblock contains reserved pages
3921  */
3922 static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn)
3923 {
3924 	unsigned long pfn;
3925 
3926 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
3927 		if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn)))
3928 			return 1;
3929 	}
3930 	return 0;
3931 }
3932 
3933 /*
3934  * Mark a number of pageblocks as MIGRATE_RESERVE. The number
3935  * of blocks reserved is based on min_wmark_pages(zone). The memory within
3936  * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
3937  * higher will lead to a bigger reserve which will get freed as contiguous
3938  * blocks as reclaim kicks in
3939  */
3940 static void setup_zone_migrate_reserve(struct zone *zone)
3941 {
3942 	unsigned long start_pfn, pfn, end_pfn, block_end_pfn;
3943 	struct page *page;
3944 	unsigned long block_migratetype;
3945 	int reserve;
3946 	int old_reserve;
3947 
3948 	/*
3949 	 * Get the start pfn, end pfn and the number of blocks to reserve
3950 	 * We have to be careful to be aligned to pageblock_nr_pages to
3951 	 * make sure that we always check pfn_valid for the first page in
3952 	 * the block.
3953 	 */
3954 	start_pfn = zone->zone_start_pfn;
3955 	end_pfn = zone_end_pfn(zone);
3956 	start_pfn = roundup(start_pfn, pageblock_nr_pages);
3957 	reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >>
3958 							pageblock_order;
3959 
3960 	/*
3961 	 * Reserve blocks are generally in place to help high-order atomic
3962 	 * allocations that are short-lived. A min_free_kbytes value that
3963 	 * would result in more than 2 reserve blocks for atomic allocations
3964 	 * is assumed to be in place to help anti-fragmentation for the
3965 	 * future allocation of hugepages at runtime.
3966 	 */
3967 	reserve = min(2, reserve);
3968 	old_reserve = zone->nr_migrate_reserve_block;
3969 
3970 	/* When memory hot-add, we almost always need to do nothing */
3971 	if (reserve == old_reserve)
3972 		return;
3973 	zone->nr_migrate_reserve_block = reserve;
3974 
3975 	for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
3976 		if (!pfn_valid(pfn))
3977 			continue;
3978 		page = pfn_to_page(pfn);
3979 
3980 		/* Watch out for overlapping nodes */
3981 		if (page_to_nid(page) != zone_to_nid(zone))
3982 			continue;
3983 
3984 		block_migratetype = get_pageblock_migratetype(page);
3985 
3986 		/* Only test what is necessary when the reserves are not met */
3987 		if (reserve > 0) {
3988 			/*
3989 			 * Blocks with reserved pages will never free, skip
3990 			 * them.
3991 			 */
3992 			block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn);
3993 			if (pageblock_is_reserved(pfn, block_end_pfn))
3994 				continue;
3995 
3996 			/* If this block is reserved, account for it */
3997 			if (block_migratetype == MIGRATE_RESERVE) {
3998 				reserve--;
3999 				continue;
4000 			}
4001 
4002 			/* Suitable for reserving if this block is movable */
4003 			if (block_migratetype == MIGRATE_MOVABLE) {
4004 				set_pageblock_migratetype(page,
4005 							MIGRATE_RESERVE);
4006 				move_freepages_block(zone, page,
4007 							MIGRATE_RESERVE);
4008 				reserve--;
4009 				continue;
4010 			}
4011 		} else if (!old_reserve) {
4012 			/*
4013 			 * At boot time we don't need to scan the whole zone
4014 			 * for turning off MIGRATE_RESERVE.
4015 			 */
4016 			break;
4017 		}
4018 
4019 		/*
4020 		 * If the reserve is met and this is a previous reserved block,
4021 		 * take it back
4022 		 */
4023 		if (block_migratetype == MIGRATE_RESERVE) {
4024 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4025 			move_freepages_block(zone, page, MIGRATE_MOVABLE);
4026 		}
4027 	}
4028 }
4029 
4030 /*
4031  * Initially all pages are reserved - free ones are freed
4032  * up by free_all_bootmem() once the early boot process is
4033  * done. Non-atomic initialization, single-pass.
4034  */
4035 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4036 		unsigned long start_pfn, enum memmap_context context)
4037 {
4038 	struct page *page;
4039 	unsigned long end_pfn = start_pfn + size;
4040 	unsigned long pfn;
4041 	struct zone *z;
4042 
4043 	if (highest_memmap_pfn < end_pfn - 1)
4044 		highest_memmap_pfn = end_pfn - 1;
4045 
4046 	z = &NODE_DATA(nid)->node_zones[zone];
4047 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4048 		/*
4049 		 * There can be holes in boot-time mem_map[]s
4050 		 * handed to this function.  They do not
4051 		 * exist on hotplugged memory.
4052 		 */
4053 		if (context == MEMMAP_EARLY) {
4054 			if (!early_pfn_valid(pfn))
4055 				continue;
4056 			if (!early_pfn_in_nid(pfn, nid))
4057 				continue;
4058 		}
4059 		page = pfn_to_page(pfn);
4060 		set_page_links(page, zone, nid, pfn);
4061 		mminit_verify_page_links(page, zone, nid, pfn);
4062 		init_page_count(page);
4063 		page_mapcount_reset(page);
4064 		page_cpupid_reset_last(page);
4065 		SetPageReserved(page);
4066 		/*
4067 		 * Mark the block movable so that blocks are reserved for
4068 		 * movable at startup. This will force kernel allocations
4069 		 * to reserve their blocks rather than leaking throughout
4070 		 * the address space during boot when many long-lived
4071 		 * kernel allocations are made. Later some blocks near
4072 		 * the start are marked MIGRATE_RESERVE by
4073 		 * setup_zone_migrate_reserve()
4074 		 *
4075 		 * bitmap is created for zone's valid pfn range. but memmap
4076 		 * can be created for invalid pages (for alignment)
4077 		 * check here not to call set_pageblock_migratetype() against
4078 		 * pfn out of zone.
4079 		 */
4080 		if ((z->zone_start_pfn <= pfn)
4081 		    && (pfn < zone_end_pfn(z))
4082 		    && !(pfn & (pageblock_nr_pages - 1)))
4083 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4084 
4085 		INIT_LIST_HEAD(&page->lru);
4086 #ifdef WANT_PAGE_VIRTUAL
4087 		/* The shift won't overflow because ZONE_NORMAL is below 4G. */
4088 		if (!is_highmem_idx(zone))
4089 			set_page_address(page, __va(pfn << PAGE_SHIFT));
4090 #endif
4091 	}
4092 }
4093 
4094 static void __meminit zone_init_free_lists(struct zone *zone)
4095 {
4096 	int order, t;
4097 	for_each_migratetype_order(order, t) {
4098 		INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4099 		zone->free_area[order].nr_free = 0;
4100 	}
4101 }
4102 
4103 #ifndef __HAVE_ARCH_MEMMAP_INIT
4104 #define memmap_init(size, nid, zone, start_pfn) \
4105 	memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4106 #endif
4107 
4108 static int __meminit zone_batchsize(struct zone *zone)
4109 {
4110 #ifdef CONFIG_MMU
4111 	int batch;
4112 
4113 	/*
4114 	 * The per-cpu-pages pools are set to around 1000th of the
4115 	 * size of the zone.  But no more than 1/2 of a meg.
4116 	 *
4117 	 * OK, so we don't know how big the cache is.  So guess.
4118 	 */
4119 	batch = zone->managed_pages / 1024;
4120 	if (batch * PAGE_SIZE > 512 * 1024)
4121 		batch = (512 * 1024) / PAGE_SIZE;
4122 	batch /= 4;		/* We effectively *= 4 below */
4123 	if (batch < 1)
4124 		batch = 1;
4125 
4126 	/*
4127 	 * Clamp the batch to a 2^n - 1 value. Having a power
4128 	 * of 2 value was found to be more likely to have
4129 	 * suboptimal cache aliasing properties in some cases.
4130 	 *
4131 	 * For example if 2 tasks are alternately allocating
4132 	 * batches of pages, one task can end up with a lot
4133 	 * of pages of one half of the possible page colors
4134 	 * and the other with pages of the other colors.
4135 	 */
4136 	batch = rounddown_pow_of_two(batch + batch/2) - 1;
4137 
4138 	return batch;
4139 
4140 #else
4141 	/* The deferral and batching of frees should be suppressed under NOMMU
4142 	 * conditions.
4143 	 *
4144 	 * The problem is that NOMMU needs to be able to allocate large chunks
4145 	 * of contiguous memory as there's no hardware page translation to
4146 	 * assemble apparent contiguous memory from discontiguous pages.
4147 	 *
4148 	 * Queueing large contiguous runs of pages for batching, however,
4149 	 * causes the pages to actually be freed in smaller chunks.  As there
4150 	 * can be a significant delay between the individual batches being
4151 	 * recycled, this leads to the once large chunks of space being
4152 	 * fragmented and becoming unavailable for high-order allocations.
4153 	 */
4154 	return 0;
4155 #endif
4156 }
4157 
4158 /*
4159  * pcp->high and pcp->batch values are related and dependent on one another:
4160  * ->batch must never be higher then ->high.
4161  * The following function updates them in a safe manner without read side
4162  * locking.
4163  *
4164  * Any new users of pcp->batch and pcp->high should ensure they can cope with
4165  * those fields changing asynchronously (acording the the above rule).
4166  *
4167  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4168  * outside of boot time (or some other assurance that no concurrent updaters
4169  * exist).
4170  */
4171 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4172 		unsigned long batch)
4173 {
4174        /* start with a fail safe value for batch */
4175 	pcp->batch = 1;
4176 	smp_wmb();
4177 
4178        /* Update high, then batch, in order */
4179 	pcp->high = high;
4180 	smp_wmb();
4181 
4182 	pcp->batch = batch;
4183 }
4184 
4185 /* a companion to pageset_set_high() */
4186 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4187 {
4188 	pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4189 }
4190 
4191 static void pageset_init(struct per_cpu_pageset *p)
4192 {
4193 	struct per_cpu_pages *pcp;
4194 	int migratetype;
4195 
4196 	memset(p, 0, sizeof(*p));
4197 
4198 	pcp = &p->pcp;
4199 	pcp->count = 0;
4200 	for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4201 		INIT_LIST_HEAD(&pcp->lists[migratetype]);
4202 }
4203 
4204 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4205 {
4206 	pageset_init(p);
4207 	pageset_set_batch(p, batch);
4208 }
4209 
4210 /*
4211  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4212  * to the value high for the pageset p.
4213  */
4214 static void pageset_set_high(struct per_cpu_pageset *p,
4215 				unsigned long high)
4216 {
4217 	unsigned long batch = max(1UL, high / 4);
4218 	if ((high / 4) > (PAGE_SHIFT * 8))
4219 		batch = PAGE_SHIFT * 8;
4220 
4221 	pageset_update(&p->pcp, high, batch);
4222 }
4223 
4224 static void __meminit pageset_set_high_and_batch(struct zone *zone,
4225 		struct per_cpu_pageset *pcp)
4226 {
4227 	if (percpu_pagelist_fraction)
4228 		pageset_set_high(pcp,
4229 			(zone->managed_pages /
4230 				percpu_pagelist_fraction));
4231 	else
4232 		pageset_set_batch(pcp, zone_batchsize(zone));
4233 }
4234 
4235 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4236 {
4237 	struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4238 
4239 	pageset_init(pcp);
4240 	pageset_set_high_and_batch(zone, pcp);
4241 }
4242 
4243 static void __meminit setup_zone_pageset(struct zone *zone)
4244 {
4245 	int cpu;
4246 	zone->pageset = alloc_percpu(struct per_cpu_pageset);
4247 	for_each_possible_cpu(cpu)
4248 		zone_pageset_init(zone, cpu);
4249 }
4250 
4251 /*
4252  * Allocate per cpu pagesets and initialize them.
4253  * Before this call only boot pagesets were available.
4254  */
4255 void __init setup_per_cpu_pageset(void)
4256 {
4257 	struct zone *zone;
4258 
4259 	for_each_populated_zone(zone)
4260 		setup_zone_pageset(zone);
4261 }
4262 
4263 static noinline __init_refok
4264 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4265 {
4266 	int i;
4267 	size_t alloc_size;
4268 
4269 	/*
4270 	 * The per-page waitqueue mechanism uses hashed waitqueues
4271 	 * per zone.
4272 	 */
4273 	zone->wait_table_hash_nr_entries =
4274 		 wait_table_hash_nr_entries(zone_size_pages);
4275 	zone->wait_table_bits =
4276 		wait_table_bits(zone->wait_table_hash_nr_entries);
4277 	alloc_size = zone->wait_table_hash_nr_entries
4278 					* sizeof(wait_queue_head_t);
4279 
4280 	if (!slab_is_available()) {
4281 		zone->wait_table = (wait_queue_head_t *)
4282 			memblock_virt_alloc_node_nopanic(
4283 				alloc_size, zone->zone_pgdat->node_id);
4284 	} else {
4285 		/*
4286 		 * This case means that a zone whose size was 0 gets new memory
4287 		 * via memory hot-add.
4288 		 * But it may be the case that a new node was hot-added.  In
4289 		 * this case vmalloc() will not be able to use this new node's
4290 		 * memory - this wait_table must be initialized to use this new
4291 		 * node itself as well.
4292 		 * To use this new node's memory, further consideration will be
4293 		 * necessary.
4294 		 */
4295 		zone->wait_table = vmalloc(alloc_size);
4296 	}
4297 	if (!zone->wait_table)
4298 		return -ENOMEM;
4299 
4300 	for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4301 		init_waitqueue_head(zone->wait_table + i);
4302 
4303 	return 0;
4304 }
4305 
4306 static __meminit void zone_pcp_init(struct zone *zone)
4307 {
4308 	/*
4309 	 * per cpu subsystem is not up at this point. The following code
4310 	 * relies on the ability of the linker to provide the
4311 	 * offset of a (static) per cpu variable into the per cpu area.
4312 	 */
4313 	zone->pageset = &boot_pageset;
4314 
4315 	if (populated_zone(zone))
4316 		printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
4317 			zone->name, zone->present_pages,
4318 					 zone_batchsize(zone));
4319 }
4320 
4321 int __meminit init_currently_empty_zone(struct zone *zone,
4322 					unsigned long zone_start_pfn,
4323 					unsigned long size,
4324 					enum memmap_context context)
4325 {
4326 	struct pglist_data *pgdat = zone->zone_pgdat;
4327 	int ret;
4328 	ret = zone_wait_table_init(zone, size);
4329 	if (ret)
4330 		return ret;
4331 	pgdat->nr_zones = zone_idx(zone) + 1;
4332 
4333 	zone->zone_start_pfn = zone_start_pfn;
4334 
4335 	mminit_dprintk(MMINIT_TRACE, "memmap_init",
4336 			"Initialising map node %d zone %lu pfns %lu -> %lu\n",
4337 			pgdat->node_id,
4338 			(unsigned long)zone_idx(zone),
4339 			zone_start_pfn, (zone_start_pfn + size));
4340 
4341 	zone_init_free_lists(zone);
4342 
4343 	return 0;
4344 }
4345 
4346 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4347 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4348 /*
4349  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4350  * Architectures may implement their own version but if add_active_range()
4351  * was used and there are no special requirements, this is a convenient
4352  * alternative
4353  */
4354 int __meminit __early_pfn_to_nid(unsigned long pfn)
4355 {
4356 	unsigned long start_pfn, end_pfn;
4357 	int nid;
4358 	/*
4359 	 * NOTE: The following SMP-unsafe globals are only used early in boot
4360 	 * when the kernel is running single-threaded.
4361 	 */
4362 	static unsigned long __meminitdata last_start_pfn, last_end_pfn;
4363 	static int __meminitdata last_nid;
4364 
4365 	if (last_start_pfn <= pfn && pfn < last_end_pfn)
4366 		return last_nid;
4367 
4368 	nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
4369 	if (nid != -1) {
4370 		last_start_pfn = start_pfn;
4371 		last_end_pfn = end_pfn;
4372 		last_nid = nid;
4373 	}
4374 
4375 	return nid;
4376 }
4377 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4378 
4379 int __meminit early_pfn_to_nid(unsigned long pfn)
4380 {
4381 	int nid;
4382 
4383 	nid = __early_pfn_to_nid(pfn);
4384 	if (nid >= 0)
4385 		return nid;
4386 	/* just returns 0 */
4387 	return 0;
4388 }
4389 
4390 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
4391 bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
4392 {
4393 	int nid;
4394 
4395 	nid = __early_pfn_to_nid(pfn);
4396 	if (nid >= 0 && nid != node)
4397 		return false;
4398 	return true;
4399 }
4400 #endif
4401 
4402 /**
4403  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4404  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4405  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4406  *
4407  * If an architecture guarantees that all ranges registered with
4408  * add_active_ranges() contain no holes and may be freed, this
4409  * this function may be used instead of calling memblock_free_early_nid()
4410  * manually.
4411  */
4412 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
4413 {
4414 	unsigned long start_pfn, end_pfn;
4415 	int i, this_nid;
4416 
4417 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
4418 		start_pfn = min(start_pfn, max_low_pfn);
4419 		end_pfn = min(end_pfn, max_low_pfn);
4420 
4421 		if (start_pfn < end_pfn)
4422 			memblock_free_early_nid(PFN_PHYS(start_pfn),
4423 					(end_pfn - start_pfn) << PAGE_SHIFT,
4424 					this_nid);
4425 	}
4426 }
4427 
4428 /**
4429  * sparse_memory_present_with_active_regions - Call memory_present for each active range
4430  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4431  *
4432  * If an architecture guarantees that all ranges registered with
4433  * add_active_ranges() contain no holes and may be freed, this
4434  * function may be used instead of calling memory_present() manually.
4435  */
4436 void __init sparse_memory_present_with_active_regions(int nid)
4437 {
4438 	unsigned long start_pfn, end_pfn;
4439 	int i, this_nid;
4440 
4441 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
4442 		memory_present(this_nid, start_pfn, end_pfn);
4443 }
4444 
4445 /**
4446  * get_pfn_range_for_nid - Return the start and end page frames for a node
4447  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4448  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4449  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4450  *
4451  * It returns the start and end page frame of a node based on information
4452  * provided by an arch calling add_active_range(). If called for a node
4453  * with no available memory, a warning is printed and the start and end
4454  * PFNs will be 0.
4455  */
4456 void __meminit get_pfn_range_for_nid(unsigned int nid,
4457 			unsigned long *start_pfn, unsigned long *end_pfn)
4458 {
4459 	unsigned long this_start_pfn, this_end_pfn;
4460 	int i;
4461 
4462 	*start_pfn = -1UL;
4463 	*end_pfn = 0;
4464 
4465 	for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
4466 		*start_pfn = min(*start_pfn, this_start_pfn);
4467 		*end_pfn = max(*end_pfn, this_end_pfn);
4468 	}
4469 
4470 	if (*start_pfn == -1UL)
4471 		*start_pfn = 0;
4472 }
4473 
4474 /*
4475  * This finds a zone that can be used for ZONE_MOVABLE pages. The
4476  * assumption is made that zones within a node are ordered in monotonic
4477  * increasing memory addresses so that the "highest" populated zone is used
4478  */
4479 static void __init find_usable_zone_for_movable(void)
4480 {
4481 	int zone_index;
4482 	for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
4483 		if (zone_index == ZONE_MOVABLE)
4484 			continue;
4485 
4486 		if (arch_zone_highest_possible_pfn[zone_index] >
4487 				arch_zone_lowest_possible_pfn[zone_index])
4488 			break;
4489 	}
4490 
4491 	VM_BUG_ON(zone_index == -1);
4492 	movable_zone = zone_index;
4493 }
4494 
4495 /*
4496  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4497  * because it is sized independent of architecture. Unlike the other zones,
4498  * the starting point for ZONE_MOVABLE is not fixed. It may be different
4499  * in each node depending on the size of each node and how evenly kernelcore
4500  * is distributed. This helper function adjusts the zone ranges
4501  * provided by the architecture for a given node by using the end of the
4502  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4503  * zones within a node are in order of monotonic increases memory addresses
4504  */
4505 static void __meminit adjust_zone_range_for_zone_movable(int nid,
4506 					unsigned long zone_type,
4507 					unsigned long node_start_pfn,
4508 					unsigned long node_end_pfn,
4509 					unsigned long *zone_start_pfn,
4510 					unsigned long *zone_end_pfn)
4511 {
4512 	/* Only adjust if ZONE_MOVABLE is on this node */
4513 	if (zone_movable_pfn[nid]) {
4514 		/* Size ZONE_MOVABLE */
4515 		if (zone_type == ZONE_MOVABLE) {
4516 			*zone_start_pfn = zone_movable_pfn[nid];
4517 			*zone_end_pfn = min(node_end_pfn,
4518 				arch_zone_highest_possible_pfn[movable_zone]);
4519 
4520 		/* Adjust for ZONE_MOVABLE starting within this range */
4521 		} else if (*zone_start_pfn < zone_movable_pfn[nid] &&
4522 				*zone_end_pfn > zone_movable_pfn[nid]) {
4523 			*zone_end_pfn = zone_movable_pfn[nid];
4524 
4525 		/* Check if this whole range is within ZONE_MOVABLE */
4526 		} else if (*zone_start_pfn >= zone_movable_pfn[nid])
4527 			*zone_start_pfn = *zone_end_pfn;
4528 	}
4529 }
4530 
4531 /*
4532  * Return the number of pages a zone spans in a node, including holes
4533  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4534  */
4535 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
4536 					unsigned long zone_type,
4537 					unsigned long node_start_pfn,
4538 					unsigned long node_end_pfn,
4539 					unsigned long *ignored)
4540 {
4541 	unsigned long zone_start_pfn, zone_end_pfn;
4542 
4543 	/* Get the start and end of the zone */
4544 	zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
4545 	zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
4546 	adjust_zone_range_for_zone_movable(nid, zone_type,
4547 				node_start_pfn, node_end_pfn,
4548 				&zone_start_pfn, &zone_end_pfn);
4549 
4550 	/* Check that this node has pages within the zone's required range */
4551 	if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
4552 		return 0;
4553 
4554 	/* Move the zone boundaries inside the node if necessary */
4555 	zone_end_pfn = min(zone_end_pfn, node_end_pfn);
4556 	zone_start_pfn = max(zone_start_pfn, node_start_pfn);
4557 
4558 	/* Return the spanned pages */
4559 	return zone_end_pfn - zone_start_pfn;
4560 }
4561 
4562 /*
4563  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4564  * then all holes in the requested range will be accounted for.
4565  */
4566 unsigned long __meminit __absent_pages_in_range(int nid,
4567 				unsigned long range_start_pfn,
4568 				unsigned long range_end_pfn)
4569 {
4570 	unsigned long nr_absent = range_end_pfn - range_start_pfn;
4571 	unsigned long start_pfn, end_pfn;
4572 	int i;
4573 
4574 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
4575 		start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
4576 		end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
4577 		nr_absent -= end_pfn - start_pfn;
4578 	}
4579 	return nr_absent;
4580 }
4581 
4582 /**
4583  * absent_pages_in_range - Return number of page frames in holes within a range
4584  * @start_pfn: The start PFN to start searching for holes
4585  * @end_pfn: The end PFN to stop searching for holes
4586  *
4587  * It returns the number of pages frames in memory holes within a range.
4588  */
4589 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
4590 							unsigned long end_pfn)
4591 {
4592 	return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
4593 }
4594 
4595 /* Return the number of page frames in holes in a zone on a node */
4596 static unsigned long __meminit zone_absent_pages_in_node(int nid,
4597 					unsigned long zone_type,
4598 					unsigned long node_start_pfn,
4599 					unsigned long node_end_pfn,
4600 					unsigned long *ignored)
4601 {
4602 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
4603 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
4604 	unsigned long zone_start_pfn, zone_end_pfn;
4605 
4606 	zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
4607 	zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
4608 
4609 	adjust_zone_range_for_zone_movable(nid, zone_type,
4610 			node_start_pfn, node_end_pfn,
4611 			&zone_start_pfn, &zone_end_pfn);
4612 	return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
4613 }
4614 
4615 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4616 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
4617 					unsigned long zone_type,
4618 					unsigned long node_start_pfn,
4619 					unsigned long node_end_pfn,
4620 					unsigned long *zones_size)
4621 {
4622 	return zones_size[zone_type];
4623 }
4624 
4625 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
4626 						unsigned long zone_type,
4627 						unsigned long node_start_pfn,
4628 						unsigned long node_end_pfn,
4629 						unsigned long *zholes_size)
4630 {
4631 	if (!zholes_size)
4632 		return 0;
4633 
4634 	return zholes_size[zone_type];
4635 }
4636 
4637 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4638 
4639 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
4640 						unsigned long node_start_pfn,
4641 						unsigned long node_end_pfn,
4642 						unsigned long *zones_size,
4643 						unsigned long *zholes_size)
4644 {
4645 	unsigned long realtotalpages, totalpages = 0;
4646 	enum zone_type i;
4647 
4648 	for (i = 0; i < MAX_NR_ZONES; i++)
4649 		totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
4650 							 node_start_pfn,
4651 							 node_end_pfn,
4652 							 zones_size);
4653 	pgdat->node_spanned_pages = totalpages;
4654 
4655 	realtotalpages = totalpages;
4656 	for (i = 0; i < MAX_NR_ZONES; i++)
4657 		realtotalpages -=
4658 			zone_absent_pages_in_node(pgdat->node_id, i,
4659 						  node_start_pfn, node_end_pfn,
4660 						  zholes_size);
4661 	pgdat->node_present_pages = realtotalpages;
4662 	printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
4663 							realtotalpages);
4664 }
4665 
4666 #ifndef CONFIG_SPARSEMEM
4667 /*
4668  * Calculate the size of the zone->blockflags rounded to an unsigned long
4669  * Start by making sure zonesize is a multiple of pageblock_order by rounding
4670  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
4671  * round what is now in bits to nearest long in bits, then return it in
4672  * bytes.
4673  */
4674 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
4675 {
4676 	unsigned long usemapsize;
4677 
4678 	zonesize += zone_start_pfn & (pageblock_nr_pages-1);
4679 	usemapsize = roundup(zonesize, pageblock_nr_pages);
4680 	usemapsize = usemapsize >> pageblock_order;
4681 	usemapsize *= NR_PAGEBLOCK_BITS;
4682 	usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
4683 
4684 	return usemapsize / 8;
4685 }
4686 
4687 static void __init setup_usemap(struct pglist_data *pgdat,
4688 				struct zone *zone,
4689 				unsigned long zone_start_pfn,
4690 				unsigned long zonesize)
4691 {
4692 	unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
4693 	zone->pageblock_flags = NULL;
4694 	if (usemapsize)
4695 		zone->pageblock_flags =
4696 			memblock_virt_alloc_node_nopanic(usemapsize,
4697 							 pgdat->node_id);
4698 }
4699 #else
4700 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
4701 				unsigned long zone_start_pfn, unsigned long zonesize) {}
4702 #endif /* CONFIG_SPARSEMEM */
4703 
4704 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
4705 
4706 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
4707 void __paginginit set_pageblock_order(void)
4708 {
4709 	unsigned int order;
4710 
4711 	/* Check that pageblock_nr_pages has not already been setup */
4712 	if (pageblock_order)
4713 		return;
4714 
4715 	if (HPAGE_SHIFT > PAGE_SHIFT)
4716 		order = HUGETLB_PAGE_ORDER;
4717 	else
4718 		order = MAX_ORDER - 1;
4719 
4720 	/*
4721 	 * Assume the largest contiguous order of interest is a huge page.
4722 	 * This value may be variable depending on boot parameters on IA64 and
4723 	 * powerpc.
4724 	 */
4725 	pageblock_order = order;
4726 }
4727 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4728 
4729 /*
4730  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
4731  * is unused as pageblock_order is set at compile-time. See
4732  * include/linux/pageblock-flags.h for the values of pageblock_order based on
4733  * the kernel config
4734  */
4735 void __paginginit set_pageblock_order(void)
4736 {
4737 }
4738 
4739 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
4740 
4741 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
4742 						   unsigned long present_pages)
4743 {
4744 	unsigned long pages = spanned_pages;
4745 
4746 	/*
4747 	 * Provide a more accurate estimation if there are holes within
4748 	 * the zone and SPARSEMEM is in use. If there are holes within the
4749 	 * zone, each populated memory region may cost us one or two extra
4750 	 * memmap pages due to alignment because memmap pages for each
4751 	 * populated regions may not naturally algined on page boundary.
4752 	 * So the (present_pages >> 4) heuristic is a tradeoff for that.
4753 	 */
4754 	if (spanned_pages > present_pages + (present_pages >> 4) &&
4755 	    IS_ENABLED(CONFIG_SPARSEMEM))
4756 		pages = present_pages;
4757 
4758 	return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
4759 }
4760 
4761 /*
4762  * Set up the zone data structures:
4763  *   - mark all pages reserved
4764  *   - mark all memory queues empty
4765  *   - clear the memory bitmaps
4766  *
4767  * NOTE: pgdat should get zeroed by caller.
4768  */
4769 static void __paginginit free_area_init_core(struct pglist_data *pgdat,
4770 		unsigned long node_start_pfn, unsigned long node_end_pfn,
4771 		unsigned long *zones_size, unsigned long *zholes_size)
4772 {
4773 	enum zone_type j;
4774 	int nid = pgdat->node_id;
4775 	unsigned long zone_start_pfn = pgdat->node_start_pfn;
4776 	int ret;
4777 
4778 	pgdat_resize_init(pgdat);
4779 #ifdef CONFIG_NUMA_BALANCING
4780 	spin_lock_init(&pgdat->numabalancing_migrate_lock);
4781 	pgdat->numabalancing_migrate_nr_pages = 0;
4782 	pgdat->numabalancing_migrate_next_window = jiffies;
4783 #endif
4784 	init_waitqueue_head(&pgdat->kswapd_wait);
4785 	init_waitqueue_head(&pgdat->pfmemalloc_wait);
4786 	pgdat_page_cgroup_init(pgdat);
4787 
4788 	for (j = 0; j < MAX_NR_ZONES; j++) {
4789 		struct zone *zone = pgdat->node_zones + j;
4790 		unsigned long size, realsize, freesize, memmap_pages;
4791 
4792 		size = zone_spanned_pages_in_node(nid, j, node_start_pfn,
4793 						  node_end_pfn, zones_size);
4794 		realsize = freesize = size - zone_absent_pages_in_node(nid, j,
4795 								node_start_pfn,
4796 								node_end_pfn,
4797 								zholes_size);
4798 
4799 		/*
4800 		 * Adjust freesize so that it accounts for how much memory
4801 		 * is used by this zone for memmap. This affects the watermark
4802 		 * and per-cpu initialisations
4803 		 */
4804 		memmap_pages = calc_memmap_size(size, realsize);
4805 		if (freesize >= memmap_pages) {
4806 			freesize -= memmap_pages;
4807 			if (memmap_pages)
4808 				printk(KERN_DEBUG
4809 				       "  %s zone: %lu pages used for memmap\n",
4810 				       zone_names[j], memmap_pages);
4811 		} else
4812 			printk(KERN_WARNING
4813 				"  %s zone: %lu pages exceeds freesize %lu\n",
4814 				zone_names[j], memmap_pages, freesize);
4815 
4816 		/* Account for reserved pages */
4817 		if (j == 0 && freesize > dma_reserve) {
4818 			freesize -= dma_reserve;
4819 			printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
4820 					zone_names[0], dma_reserve);
4821 		}
4822 
4823 		if (!is_highmem_idx(j))
4824 			nr_kernel_pages += freesize;
4825 		/* Charge for highmem memmap if there are enough kernel pages */
4826 		else if (nr_kernel_pages > memmap_pages * 2)
4827 			nr_kernel_pages -= memmap_pages;
4828 		nr_all_pages += freesize;
4829 
4830 		zone->spanned_pages = size;
4831 		zone->present_pages = realsize;
4832 		/*
4833 		 * Set an approximate value for lowmem here, it will be adjusted
4834 		 * when the bootmem allocator frees pages into the buddy system.
4835 		 * And all highmem pages will be managed by the buddy system.
4836 		 */
4837 		zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
4838 #ifdef CONFIG_NUMA
4839 		zone->node = nid;
4840 		zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
4841 						/ 100;
4842 		zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
4843 #endif
4844 		zone->name = zone_names[j];
4845 		spin_lock_init(&zone->lock);
4846 		spin_lock_init(&zone->lru_lock);
4847 		zone_seqlock_init(zone);
4848 		zone->zone_pgdat = pgdat;
4849 		zone_pcp_init(zone);
4850 
4851 		/* For bootup, initialized properly in watermark setup */
4852 		mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
4853 
4854 		lruvec_init(&zone->lruvec);
4855 		if (!size)
4856 			continue;
4857 
4858 		set_pageblock_order();
4859 		setup_usemap(pgdat, zone, zone_start_pfn, size);
4860 		ret = init_currently_empty_zone(zone, zone_start_pfn,
4861 						size, MEMMAP_EARLY);
4862 		BUG_ON(ret);
4863 		memmap_init(size, nid, j, zone_start_pfn);
4864 		zone_start_pfn += size;
4865 	}
4866 }
4867 
4868 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
4869 {
4870 	/* Skip empty nodes */
4871 	if (!pgdat->node_spanned_pages)
4872 		return;
4873 
4874 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4875 	/* ia64 gets its own node_mem_map, before this, without bootmem */
4876 	if (!pgdat->node_mem_map) {
4877 		unsigned long size, start, end;
4878 		struct page *map;
4879 
4880 		/*
4881 		 * The zone's endpoints aren't required to be MAX_ORDER
4882 		 * aligned but the node_mem_map endpoints must be in order
4883 		 * for the buddy allocator to function correctly.
4884 		 */
4885 		start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
4886 		end = pgdat_end_pfn(pgdat);
4887 		end = ALIGN(end, MAX_ORDER_NR_PAGES);
4888 		size =  (end - start) * sizeof(struct page);
4889 		map = alloc_remap(pgdat->node_id, size);
4890 		if (!map)
4891 			map = memblock_virt_alloc_node_nopanic(size,
4892 							       pgdat->node_id);
4893 		pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
4894 	}
4895 #ifndef CONFIG_NEED_MULTIPLE_NODES
4896 	/*
4897 	 * With no DISCONTIG, the global mem_map is just set as node 0's
4898 	 */
4899 	if (pgdat == NODE_DATA(0)) {
4900 		mem_map = NODE_DATA(0)->node_mem_map;
4901 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4902 		if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
4903 			mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
4904 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4905 	}
4906 #endif
4907 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
4908 }
4909 
4910 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
4911 		unsigned long node_start_pfn, unsigned long *zholes_size)
4912 {
4913 	pg_data_t *pgdat = NODE_DATA(nid);
4914 	unsigned long start_pfn = 0;
4915 	unsigned long end_pfn = 0;
4916 
4917 	/* pg_data_t should be reset to zero when it's allocated */
4918 	WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
4919 
4920 	pgdat->node_id = nid;
4921 	pgdat->node_start_pfn = node_start_pfn;
4922 	init_zone_allows_reclaim(nid);
4923 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4924 	get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
4925 #endif
4926 	calculate_node_totalpages(pgdat, start_pfn, end_pfn,
4927 				  zones_size, zholes_size);
4928 
4929 	alloc_node_mem_map(pgdat);
4930 #ifdef CONFIG_FLAT_NODE_MEM_MAP
4931 	printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
4932 		nid, (unsigned long)pgdat,
4933 		(unsigned long)pgdat->node_mem_map);
4934 #endif
4935 
4936 	free_area_init_core(pgdat, start_pfn, end_pfn,
4937 			    zones_size, zholes_size);
4938 }
4939 
4940 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4941 
4942 #if MAX_NUMNODES > 1
4943 /*
4944  * Figure out the number of possible node ids.
4945  */
4946 void __init setup_nr_node_ids(void)
4947 {
4948 	unsigned int node;
4949 	unsigned int highest = 0;
4950 
4951 	for_each_node_mask(node, node_possible_map)
4952 		highest = node;
4953 	nr_node_ids = highest + 1;
4954 }
4955 #endif
4956 
4957 /**
4958  * node_map_pfn_alignment - determine the maximum internode alignment
4959  *
4960  * This function should be called after node map is populated and sorted.
4961  * It calculates the maximum power of two alignment which can distinguish
4962  * all the nodes.
4963  *
4964  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
4965  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
4966  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
4967  * shifted, 1GiB is enough and this function will indicate so.
4968  *
4969  * This is used to test whether pfn -> nid mapping of the chosen memory
4970  * model has fine enough granularity to avoid incorrect mapping for the
4971  * populated node map.
4972  *
4973  * Returns the determined alignment in pfn's.  0 if there is no alignment
4974  * requirement (single node).
4975  */
4976 unsigned long __init node_map_pfn_alignment(void)
4977 {
4978 	unsigned long accl_mask = 0, last_end = 0;
4979 	unsigned long start, end, mask;
4980 	int last_nid = -1;
4981 	int i, nid;
4982 
4983 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
4984 		if (!start || last_nid < 0 || last_nid == nid) {
4985 			last_nid = nid;
4986 			last_end = end;
4987 			continue;
4988 		}
4989 
4990 		/*
4991 		 * Start with a mask granular enough to pin-point to the
4992 		 * start pfn and tick off bits one-by-one until it becomes
4993 		 * too coarse to separate the current node from the last.
4994 		 */
4995 		mask = ~((1 << __ffs(start)) - 1);
4996 		while (mask && last_end <= (start & (mask << 1)))
4997 			mask <<= 1;
4998 
4999 		/* accumulate all internode masks */
5000 		accl_mask |= mask;
5001 	}
5002 
5003 	/* convert mask to number of pages */
5004 	return ~accl_mask + 1;
5005 }
5006 
5007 /* Find the lowest pfn for a node */
5008 static unsigned long __init find_min_pfn_for_node(int nid)
5009 {
5010 	unsigned long min_pfn = ULONG_MAX;
5011 	unsigned long start_pfn;
5012 	int i;
5013 
5014 	for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5015 		min_pfn = min(min_pfn, start_pfn);
5016 
5017 	if (min_pfn == ULONG_MAX) {
5018 		printk(KERN_WARNING
5019 			"Could not find start_pfn for node %d\n", nid);
5020 		return 0;
5021 	}
5022 
5023 	return min_pfn;
5024 }
5025 
5026 /**
5027  * find_min_pfn_with_active_regions - Find the minimum PFN registered
5028  *
5029  * It returns the minimum PFN based on information provided via
5030  * add_active_range().
5031  */
5032 unsigned long __init find_min_pfn_with_active_regions(void)
5033 {
5034 	return find_min_pfn_for_node(MAX_NUMNODES);
5035 }
5036 
5037 /*
5038  * early_calculate_totalpages()
5039  * Sum pages in active regions for movable zone.
5040  * Populate N_MEMORY for calculating usable_nodes.
5041  */
5042 static unsigned long __init early_calculate_totalpages(void)
5043 {
5044 	unsigned long totalpages = 0;
5045 	unsigned long start_pfn, end_pfn;
5046 	int i, nid;
5047 
5048 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5049 		unsigned long pages = end_pfn - start_pfn;
5050 
5051 		totalpages += pages;
5052 		if (pages)
5053 			node_set_state(nid, N_MEMORY);
5054 	}
5055 	return totalpages;
5056 }
5057 
5058 /*
5059  * Find the PFN the Movable zone begins in each node. Kernel memory
5060  * is spread evenly between nodes as long as the nodes have enough
5061  * memory. When they don't, some nodes will have more kernelcore than
5062  * others
5063  */
5064 static void __init find_zone_movable_pfns_for_nodes(void)
5065 {
5066 	int i, nid;
5067 	unsigned long usable_startpfn;
5068 	unsigned long kernelcore_node, kernelcore_remaining;
5069 	/* save the state before borrow the nodemask */
5070 	nodemask_t saved_node_state = node_states[N_MEMORY];
5071 	unsigned long totalpages = early_calculate_totalpages();
5072 	int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5073 	struct memblock_type *type = &memblock.memory;
5074 
5075 	/* Need to find movable_zone earlier when movable_node is specified. */
5076 	find_usable_zone_for_movable();
5077 
5078 	/*
5079 	 * If movable_node is specified, ignore kernelcore and movablecore
5080 	 * options.
5081 	 */
5082 	if (movable_node_is_enabled()) {
5083 		for (i = 0; i < type->cnt; i++) {
5084 			if (!memblock_is_hotpluggable(&type->regions[i]))
5085 				continue;
5086 
5087 			nid = type->regions[i].nid;
5088 
5089 			usable_startpfn = PFN_DOWN(type->regions[i].base);
5090 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5091 				min(usable_startpfn, zone_movable_pfn[nid]) :
5092 				usable_startpfn;
5093 		}
5094 
5095 		goto out2;
5096 	}
5097 
5098 	/*
5099 	 * If movablecore=nn[KMG] was specified, calculate what size of
5100 	 * kernelcore that corresponds so that memory usable for
5101 	 * any allocation type is evenly spread. If both kernelcore
5102 	 * and movablecore are specified, then the value of kernelcore
5103 	 * will be used for required_kernelcore if it's greater than
5104 	 * what movablecore would have allowed.
5105 	 */
5106 	if (required_movablecore) {
5107 		unsigned long corepages;
5108 
5109 		/*
5110 		 * Round-up so that ZONE_MOVABLE is at least as large as what
5111 		 * was requested by the user
5112 		 */
5113 		required_movablecore =
5114 			roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5115 		corepages = totalpages - required_movablecore;
5116 
5117 		required_kernelcore = max(required_kernelcore, corepages);
5118 	}
5119 
5120 	/* If kernelcore was not specified, there is no ZONE_MOVABLE */
5121 	if (!required_kernelcore)
5122 		goto out;
5123 
5124 	/* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5125 	usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5126 
5127 restart:
5128 	/* Spread kernelcore memory as evenly as possible throughout nodes */
5129 	kernelcore_node = required_kernelcore / usable_nodes;
5130 	for_each_node_state(nid, N_MEMORY) {
5131 		unsigned long start_pfn, end_pfn;
5132 
5133 		/*
5134 		 * Recalculate kernelcore_node if the division per node
5135 		 * now exceeds what is necessary to satisfy the requested
5136 		 * amount of memory for the kernel
5137 		 */
5138 		if (required_kernelcore < kernelcore_node)
5139 			kernelcore_node = required_kernelcore / usable_nodes;
5140 
5141 		/*
5142 		 * As the map is walked, we track how much memory is usable
5143 		 * by the kernel using kernelcore_remaining. When it is
5144 		 * 0, the rest of the node is usable by ZONE_MOVABLE
5145 		 */
5146 		kernelcore_remaining = kernelcore_node;
5147 
5148 		/* Go through each range of PFNs within this node */
5149 		for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5150 			unsigned long size_pages;
5151 
5152 			start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5153 			if (start_pfn >= end_pfn)
5154 				continue;
5155 
5156 			/* Account for what is only usable for kernelcore */
5157 			if (start_pfn < usable_startpfn) {
5158 				unsigned long kernel_pages;
5159 				kernel_pages = min(end_pfn, usable_startpfn)
5160 								- start_pfn;
5161 
5162 				kernelcore_remaining -= min(kernel_pages,
5163 							kernelcore_remaining);
5164 				required_kernelcore -= min(kernel_pages,
5165 							required_kernelcore);
5166 
5167 				/* Continue if range is now fully accounted */
5168 				if (end_pfn <= usable_startpfn) {
5169 
5170 					/*
5171 					 * Push zone_movable_pfn to the end so
5172 					 * that if we have to rebalance
5173 					 * kernelcore across nodes, we will
5174 					 * not double account here
5175 					 */
5176 					zone_movable_pfn[nid] = end_pfn;
5177 					continue;
5178 				}
5179 				start_pfn = usable_startpfn;
5180 			}
5181 
5182 			/*
5183 			 * The usable PFN range for ZONE_MOVABLE is from
5184 			 * start_pfn->end_pfn. Calculate size_pages as the
5185 			 * number of pages used as kernelcore
5186 			 */
5187 			size_pages = end_pfn - start_pfn;
5188 			if (size_pages > kernelcore_remaining)
5189 				size_pages = kernelcore_remaining;
5190 			zone_movable_pfn[nid] = start_pfn + size_pages;
5191 
5192 			/*
5193 			 * Some kernelcore has been met, update counts and
5194 			 * break if the kernelcore for this node has been
5195 			 * satisfied
5196 			 */
5197 			required_kernelcore -= min(required_kernelcore,
5198 								size_pages);
5199 			kernelcore_remaining -= size_pages;
5200 			if (!kernelcore_remaining)
5201 				break;
5202 		}
5203 	}
5204 
5205 	/*
5206 	 * If there is still required_kernelcore, we do another pass with one
5207 	 * less node in the count. This will push zone_movable_pfn[nid] further
5208 	 * along on the nodes that still have memory until kernelcore is
5209 	 * satisfied
5210 	 */
5211 	usable_nodes--;
5212 	if (usable_nodes && required_kernelcore > usable_nodes)
5213 		goto restart;
5214 
5215 out2:
5216 	/* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5217 	for (nid = 0; nid < MAX_NUMNODES; nid++)
5218 		zone_movable_pfn[nid] =
5219 			roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5220 
5221 out:
5222 	/* restore the node_state */
5223 	node_states[N_MEMORY] = saved_node_state;
5224 }
5225 
5226 /* Any regular or high memory on that node ? */
5227 static void check_for_memory(pg_data_t *pgdat, int nid)
5228 {
5229 	enum zone_type zone_type;
5230 
5231 	if (N_MEMORY == N_NORMAL_MEMORY)
5232 		return;
5233 
5234 	for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5235 		struct zone *zone = &pgdat->node_zones[zone_type];
5236 		if (populated_zone(zone)) {
5237 			node_set_state(nid, N_HIGH_MEMORY);
5238 			if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5239 			    zone_type <= ZONE_NORMAL)
5240 				node_set_state(nid, N_NORMAL_MEMORY);
5241 			break;
5242 		}
5243 	}
5244 }
5245 
5246 /**
5247  * free_area_init_nodes - Initialise all pg_data_t and zone data
5248  * @max_zone_pfn: an array of max PFNs for each zone
5249  *
5250  * This will call free_area_init_node() for each active node in the system.
5251  * Using the page ranges provided by add_active_range(), the size of each
5252  * zone in each node and their holes is calculated. If the maximum PFN
5253  * between two adjacent zones match, it is assumed that the zone is empty.
5254  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5255  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5256  * starts where the previous one ended. For example, ZONE_DMA32 starts
5257  * at arch_max_dma_pfn.
5258  */
5259 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5260 {
5261 	unsigned long start_pfn, end_pfn;
5262 	int i, nid;
5263 
5264 	/* Record where the zone boundaries are */
5265 	memset(arch_zone_lowest_possible_pfn, 0,
5266 				sizeof(arch_zone_lowest_possible_pfn));
5267 	memset(arch_zone_highest_possible_pfn, 0,
5268 				sizeof(arch_zone_highest_possible_pfn));
5269 	arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5270 	arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5271 	for (i = 1; i < MAX_NR_ZONES; i++) {
5272 		if (i == ZONE_MOVABLE)
5273 			continue;
5274 		arch_zone_lowest_possible_pfn[i] =
5275 			arch_zone_highest_possible_pfn[i-1];
5276 		arch_zone_highest_possible_pfn[i] =
5277 			max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
5278 	}
5279 	arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
5280 	arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
5281 
5282 	/* Find the PFNs that ZONE_MOVABLE begins at in each node */
5283 	memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
5284 	find_zone_movable_pfns_for_nodes();
5285 
5286 	/* Print out the zone ranges */
5287 	printk("Zone ranges:\n");
5288 	for (i = 0; i < MAX_NR_ZONES; i++) {
5289 		if (i == ZONE_MOVABLE)
5290 			continue;
5291 		printk(KERN_CONT "  %-8s ", zone_names[i]);
5292 		if (arch_zone_lowest_possible_pfn[i] ==
5293 				arch_zone_highest_possible_pfn[i])
5294 			printk(KERN_CONT "empty\n");
5295 		else
5296 			printk(KERN_CONT "[mem %0#10lx-%0#10lx]\n",
5297 				arch_zone_lowest_possible_pfn[i] << PAGE_SHIFT,
5298 				(arch_zone_highest_possible_pfn[i]
5299 					<< PAGE_SHIFT) - 1);
5300 	}
5301 
5302 	/* Print out the PFNs ZONE_MOVABLE begins at in each node */
5303 	printk("Movable zone start for each node\n");
5304 	for (i = 0; i < MAX_NUMNODES; i++) {
5305 		if (zone_movable_pfn[i])
5306 			printk("  Node %d: %#010lx\n", i,
5307 			       zone_movable_pfn[i] << PAGE_SHIFT);
5308 	}
5309 
5310 	/* Print out the early node map */
5311 	printk("Early memory node ranges\n");
5312 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
5313 		printk("  node %3d: [mem %#010lx-%#010lx]\n", nid,
5314 		       start_pfn << PAGE_SHIFT, (end_pfn << PAGE_SHIFT) - 1);
5315 
5316 	/* Initialise every node */
5317 	mminit_verify_pageflags_layout();
5318 	setup_nr_node_ids();
5319 	for_each_online_node(nid) {
5320 		pg_data_t *pgdat = NODE_DATA(nid);
5321 		free_area_init_node(nid, NULL,
5322 				find_min_pfn_for_node(nid), NULL);
5323 
5324 		/* Any memory on that node */
5325 		if (pgdat->node_present_pages)
5326 			node_set_state(nid, N_MEMORY);
5327 		check_for_memory(pgdat, nid);
5328 	}
5329 }
5330 
5331 static int __init cmdline_parse_core(char *p, unsigned long *core)
5332 {
5333 	unsigned long long coremem;
5334 	if (!p)
5335 		return -EINVAL;
5336 
5337 	coremem = memparse(p, &p);
5338 	*core = coremem >> PAGE_SHIFT;
5339 
5340 	/* Paranoid check that UL is enough for the coremem value */
5341 	WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
5342 
5343 	return 0;
5344 }
5345 
5346 /*
5347  * kernelcore=size sets the amount of memory for use for allocations that
5348  * cannot be reclaimed or migrated.
5349  */
5350 static int __init cmdline_parse_kernelcore(char *p)
5351 {
5352 	return cmdline_parse_core(p, &required_kernelcore);
5353 }
5354 
5355 /*
5356  * movablecore=size sets the amount of memory for use for allocations that
5357  * can be reclaimed or migrated.
5358  */
5359 static int __init cmdline_parse_movablecore(char *p)
5360 {
5361 	return cmdline_parse_core(p, &required_movablecore);
5362 }
5363 
5364 early_param("kernelcore", cmdline_parse_kernelcore);
5365 early_param("movablecore", cmdline_parse_movablecore);
5366 
5367 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5368 
5369 void adjust_managed_page_count(struct page *page, long count)
5370 {
5371 	spin_lock(&managed_page_count_lock);
5372 	page_zone(page)->managed_pages += count;
5373 	totalram_pages += count;
5374 #ifdef CONFIG_HIGHMEM
5375 	if (PageHighMem(page))
5376 		totalhigh_pages += count;
5377 #endif
5378 	spin_unlock(&managed_page_count_lock);
5379 }
5380 EXPORT_SYMBOL(adjust_managed_page_count);
5381 
5382 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
5383 {
5384 	void *pos;
5385 	unsigned long pages = 0;
5386 
5387 	start = (void *)PAGE_ALIGN((unsigned long)start);
5388 	end = (void *)((unsigned long)end & PAGE_MASK);
5389 	for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
5390 		if ((unsigned int)poison <= 0xFF)
5391 			memset(pos, poison, PAGE_SIZE);
5392 		free_reserved_page(virt_to_page(pos));
5393 	}
5394 
5395 	if (pages && s)
5396 		pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5397 			s, pages << (PAGE_SHIFT - 10), start, end);
5398 
5399 	return pages;
5400 }
5401 EXPORT_SYMBOL(free_reserved_area);
5402 
5403 #ifdef	CONFIG_HIGHMEM
5404 void free_highmem_page(struct page *page)
5405 {
5406 	__free_reserved_page(page);
5407 	totalram_pages++;
5408 	page_zone(page)->managed_pages++;
5409 	totalhigh_pages++;
5410 }
5411 #endif
5412 
5413 
5414 void __init mem_init_print_info(const char *str)
5415 {
5416 	unsigned long physpages, codesize, datasize, rosize, bss_size;
5417 	unsigned long init_code_size, init_data_size;
5418 
5419 	physpages = get_num_physpages();
5420 	codesize = _etext - _stext;
5421 	datasize = _edata - _sdata;
5422 	rosize = __end_rodata - __start_rodata;
5423 	bss_size = __bss_stop - __bss_start;
5424 	init_data_size = __init_end - __init_begin;
5425 	init_code_size = _einittext - _sinittext;
5426 
5427 	/*
5428 	 * Detect special cases and adjust section sizes accordingly:
5429 	 * 1) .init.* may be embedded into .data sections
5430 	 * 2) .init.text.* may be out of [__init_begin, __init_end],
5431 	 *    please refer to arch/tile/kernel/vmlinux.lds.S.
5432 	 * 3) .rodata.* may be embedded into .text or .data sections.
5433 	 */
5434 #define adj_init_size(start, end, size, pos, adj) \
5435 	do { \
5436 		if (start <= pos && pos < end && size > adj) \
5437 			size -= adj; \
5438 	} while (0)
5439 
5440 	adj_init_size(__init_begin, __init_end, init_data_size,
5441 		     _sinittext, init_code_size);
5442 	adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
5443 	adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
5444 	adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
5445 	adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
5446 
5447 #undef	adj_init_size
5448 
5449 	printk("Memory: %luK/%luK available "
5450 	       "(%luK kernel code, %luK rwdata, %luK rodata, "
5451 	       "%luK init, %luK bss, %luK reserved"
5452 #ifdef	CONFIG_HIGHMEM
5453 	       ", %luK highmem"
5454 #endif
5455 	       "%s%s)\n",
5456 	       nr_free_pages() << (PAGE_SHIFT-10), physpages << (PAGE_SHIFT-10),
5457 	       codesize >> 10, datasize >> 10, rosize >> 10,
5458 	       (init_data_size + init_code_size) >> 10, bss_size >> 10,
5459 	       (physpages - totalram_pages) << (PAGE_SHIFT-10),
5460 #ifdef	CONFIG_HIGHMEM
5461 	       totalhigh_pages << (PAGE_SHIFT-10),
5462 #endif
5463 	       str ? ", " : "", str ? str : "");
5464 }
5465 
5466 /**
5467  * set_dma_reserve - set the specified number of pages reserved in the first zone
5468  * @new_dma_reserve: The number of pages to mark reserved
5469  *
5470  * The per-cpu batchsize and zone watermarks are determined by present_pages.
5471  * In the DMA zone, a significant percentage may be consumed by kernel image
5472  * and other unfreeable allocations which can skew the watermarks badly. This
5473  * function may optionally be used to account for unfreeable pages in the
5474  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5475  * smaller per-cpu batchsize.
5476  */
5477 void __init set_dma_reserve(unsigned long new_dma_reserve)
5478 {
5479 	dma_reserve = new_dma_reserve;
5480 }
5481 
5482 void __init free_area_init(unsigned long *zones_size)
5483 {
5484 	free_area_init_node(0, zones_size,
5485 			__pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
5486 }
5487 
5488 static int page_alloc_cpu_notify(struct notifier_block *self,
5489 				 unsigned long action, void *hcpu)
5490 {
5491 	int cpu = (unsigned long)hcpu;
5492 
5493 	if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
5494 		lru_add_drain_cpu(cpu);
5495 		drain_pages(cpu);
5496 
5497 		/*
5498 		 * Spill the event counters of the dead processor
5499 		 * into the current processors event counters.
5500 		 * This artificially elevates the count of the current
5501 		 * processor.
5502 		 */
5503 		vm_events_fold_cpu(cpu);
5504 
5505 		/*
5506 		 * Zero the differential counters of the dead processor
5507 		 * so that the vm statistics are consistent.
5508 		 *
5509 		 * This is only okay since the processor is dead and cannot
5510 		 * race with what we are doing.
5511 		 */
5512 		cpu_vm_stats_fold(cpu);
5513 	}
5514 	return NOTIFY_OK;
5515 }
5516 
5517 void __init page_alloc_init(void)
5518 {
5519 	hotcpu_notifier(page_alloc_cpu_notify, 0);
5520 }
5521 
5522 /*
5523  * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
5524  *	or min_free_kbytes changes.
5525  */
5526 static void calculate_totalreserve_pages(void)
5527 {
5528 	struct pglist_data *pgdat;
5529 	unsigned long reserve_pages = 0;
5530 	enum zone_type i, j;
5531 
5532 	for_each_online_pgdat(pgdat) {
5533 		for (i = 0; i < MAX_NR_ZONES; i++) {
5534 			struct zone *zone = pgdat->node_zones + i;
5535 			unsigned long max = 0;
5536 
5537 			/* Find valid and maximum lowmem_reserve in the zone */
5538 			for (j = i; j < MAX_NR_ZONES; j++) {
5539 				if (zone->lowmem_reserve[j] > max)
5540 					max = zone->lowmem_reserve[j];
5541 			}
5542 
5543 			/* we treat the high watermark as reserved pages. */
5544 			max += high_wmark_pages(zone);
5545 
5546 			if (max > zone->managed_pages)
5547 				max = zone->managed_pages;
5548 			reserve_pages += max;
5549 			/*
5550 			 * Lowmem reserves are not available to
5551 			 * GFP_HIGHUSER page cache allocations and
5552 			 * kswapd tries to balance zones to their high
5553 			 * watermark.  As a result, neither should be
5554 			 * regarded as dirtyable memory, to prevent a
5555 			 * situation where reclaim has to clean pages
5556 			 * in order to balance the zones.
5557 			 */
5558 			zone->dirty_balance_reserve = max;
5559 		}
5560 	}
5561 	dirty_balance_reserve = reserve_pages;
5562 	totalreserve_pages = reserve_pages;
5563 }
5564 
5565 /*
5566  * setup_per_zone_lowmem_reserve - called whenever
5567  *	sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
5568  *	has a correct pages reserved value, so an adequate number of
5569  *	pages are left in the zone after a successful __alloc_pages().
5570  */
5571 static void setup_per_zone_lowmem_reserve(void)
5572 {
5573 	struct pglist_data *pgdat;
5574 	enum zone_type j, idx;
5575 
5576 	for_each_online_pgdat(pgdat) {
5577 		for (j = 0; j < MAX_NR_ZONES; j++) {
5578 			struct zone *zone = pgdat->node_zones + j;
5579 			unsigned long managed_pages = zone->managed_pages;
5580 
5581 			zone->lowmem_reserve[j] = 0;
5582 
5583 			idx = j;
5584 			while (idx) {
5585 				struct zone *lower_zone;
5586 
5587 				idx--;
5588 
5589 				if (sysctl_lowmem_reserve_ratio[idx] < 1)
5590 					sysctl_lowmem_reserve_ratio[idx] = 1;
5591 
5592 				lower_zone = pgdat->node_zones + idx;
5593 				lower_zone->lowmem_reserve[j] = managed_pages /
5594 					sysctl_lowmem_reserve_ratio[idx];
5595 				managed_pages += lower_zone->managed_pages;
5596 			}
5597 		}
5598 	}
5599 
5600 	/* update totalreserve_pages */
5601 	calculate_totalreserve_pages();
5602 }
5603 
5604 static void __setup_per_zone_wmarks(void)
5605 {
5606 	unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
5607 	unsigned long lowmem_pages = 0;
5608 	struct zone *zone;
5609 	unsigned long flags;
5610 
5611 	/* Calculate total number of !ZONE_HIGHMEM pages */
5612 	for_each_zone(zone) {
5613 		if (!is_highmem(zone))
5614 			lowmem_pages += zone->managed_pages;
5615 	}
5616 
5617 	for_each_zone(zone) {
5618 		u64 tmp;
5619 
5620 		spin_lock_irqsave(&zone->lock, flags);
5621 		tmp = (u64)pages_min * zone->managed_pages;
5622 		do_div(tmp, lowmem_pages);
5623 		if (is_highmem(zone)) {
5624 			/*
5625 			 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
5626 			 * need highmem pages, so cap pages_min to a small
5627 			 * value here.
5628 			 *
5629 			 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
5630 			 * deltas controls asynch page reclaim, and so should
5631 			 * not be capped for highmem.
5632 			 */
5633 			unsigned long min_pages;
5634 
5635 			min_pages = zone->managed_pages / 1024;
5636 			min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
5637 			zone->watermark[WMARK_MIN] = min_pages;
5638 		} else {
5639 			/*
5640 			 * If it's a lowmem zone, reserve a number of pages
5641 			 * proportionate to the zone's size.
5642 			 */
5643 			zone->watermark[WMARK_MIN] = tmp;
5644 		}
5645 
5646 		zone->watermark[WMARK_LOW]  = min_wmark_pages(zone) + (tmp >> 2);
5647 		zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1);
5648 
5649 		__mod_zone_page_state(zone, NR_ALLOC_BATCH,
5650 				      high_wmark_pages(zone) -
5651 				      low_wmark_pages(zone) -
5652 				      zone_page_state(zone, NR_ALLOC_BATCH));
5653 
5654 		setup_zone_migrate_reserve(zone);
5655 		spin_unlock_irqrestore(&zone->lock, flags);
5656 	}
5657 
5658 	/* update totalreserve_pages */
5659 	calculate_totalreserve_pages();
5660 }
5661 
5662 /**
5663  * setup_per_zone_wmarks - called when min_free_kbytes changes
5664  * or when memory is hot-{added|removed}
5665  *
5666  * Ensures that the watermark[min,low,high] values for each zone are set
5667  * correctly with respect to min_free_kbytes.
5668  */
5669 void setup_per_zone_wmarks(void)
5670 {
5671 	mutex_lock(&zonelists_mutex);
5672 	__setup_per_zone_wmarks();
5673 	mutex_unlock(&zonelists_mutex);
5674 }
5675 
5676 /*
5677  * The inactive anon list should be small enough that the VM never has to
5678  * do too much work, but large enough that each inactive page has a chance
5679  * to be referenced again before it is swapped out.
5680  *
5681  * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
5682  * INACTIVE_ANON pages on this zone's LRU, maintained by the
5683  * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
5684  * the anonymous pages are kept on the inactive list.
5685  *
5686  * total     target    max
5687  * memory    ratio     inactive anon
5688  * -------------------------------------
5689  *   10MB       1         5MB
5690  *  100MB       1        50MB
5691  *    1GB       3       250MB
5692  *   10GB      10       0.9GB
5693  *  100GB      31         3GB
5694  *    1TB     101        10GB
5695  *   10TB     320        32GB
5696  */
5697 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
5698 {
5699 	unsigned int gb, ratio;
5700 
5701 	/* Zone size in gigabytes */
5702 	gb = zone->managed_pages >> (30 - PAGE_SHIFT);
5703 	if (gb)
5704 		ratio = int_sqrt(10 * gb);
5705 	else
5706 		ratio = 1;
5707 
5708 	zone->inactive_ratio = ratio;
5709 }
5710 
5711 static void __meminit setup_per_zone_inactive_ratio(void)
5712 {
5713 	struct zone *zone;
5714 
5715 	for_each_zone(zone)
5716 		calculate_zone_inactive_ratio(zone);
5717 }
5718 
5719 /*
5720  * Initialise min_free_kbytes.
5721  *
5722  * For small machines we want it small (128k min).  For large machines
5723  * we want it large (64MB max).  But it is not linear, because network
5724  * bandwidth does not increase linearly with machine size.  We use
5725  *
5726  *	min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
5727  *	min_free_kbytes = sqrt(lowmem_kbytes * 16)
5728  *
5729  * which yields
5730  *
5731  * 16MB:	512k
5732  * 32MB:	724k
5733  * 64MB:	1024k
5734  * 128MB:	1448k
5735  * 256MB:	2048k
5736  * 512MB:	2896k
5737  * 1024MB:	4096k
5738  * 2048MB:	5792k
5739  * 4096MB:	8192k
5740  * 8192MB:	11584k
5741  * 16384MB:	16384k
5742  */
5743 int __meminit init_per_zone_wmark_min(void)
5744 {
5745 	unsigned long lowmem_kbytes;
5746 	int new_min_free_kbytes;
5747 
5748 	lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
5749 	new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
5750 
5751 	if (new_min_free_kbytes > user_min_free_kbytes) {
5752 		min_free_kbytes = new_min_free_kbytes;
5753 		if (min_free_kbytes < 128)
5754 			min_free_kbytes = 128;
5755 		if (min_free_kbytes > 65536)
5756 			min_free_kbytes = 65536;
5757 	} else {
5758 		pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
5759 				new_min_free_kbytes, user_min_free_kbytes);
5760 	}
5761 	setup_per_zone_wmarks();
5762 	refresh_zone_stat_thresholds();
5763 	setup_per_zone_lowmem_reserve();
5764 	setup_per_zone_inactive_ratio();
5765 	return 0;
5766 }
5767 module_init(init_per_zone_wmark_min)
5768 
5769 /*
5770  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
5771  *	that we can call two helper functions whenever min_free_kbytes
5772  *	changes.
5773  */
5774 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
5775 	void __user *buffer, size_t *length, loff_t *ppos)
5776 {
5777 	int rc;
5778 
5779 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5780 	if (rc)
5781 		return rc;
5782 
5783 	if (write) {
5784 		user_min_free_kbytes = min_free_kbytes;
5785 		setup_per_zone_wmarks();
5786 	}
5787 	return 0;
5788 }
5789 
5790 #ifdef CONFIG_NUMA
5791 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
5792 	void __user *buffer, size_t *length, loff_t *ppos)
5793 {
5794 	struct zone *zone;
5795 	int rc;
5796 
5797 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5798 	if (rc)
5799 		return rc;
5800 
5801 	for_each_zone(zone)
5802 		zone->min_unmapped_pages = (zone->managed_pages *
5803 				sysctl_min_unmapped_ratio) / 100;
5804 	return 0;
5805 }
5806 
5807 int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
5808 	void __user *buffer, size_t *length, loff_t *ppos)
5809 {
5810 	struct zone *zone;
5811 	int rc;
5812 
5813 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
5814 	if (rc)
5815 		return rc;
5816 
5817 	for_each_zone(zone)
5818 		zone->min_slab_pages = (zone->managed_pages *
5819 				sysctl_min_slab_ratio) / 100;
5820 	return 0;
5821 }
5822 #endif
5823 
5824 /*
5825  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
5826  *	proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
5827  *	whenever sysctl_lowmem_reserve_ratio changes.
5828  *
5829  * The reserve ratio obviously has absolutely no relation with the
5830  * minimum watermarks. The lowmem reserve ratio can only make sense
5831  * if in function of the boot time zone sizes.
5832  */
5833 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
5834 	void __user *buffer, size_t *length, loff_t *ppos)
5835 {
5836 	proc_dointvec_minmax(table, write, buffer, length, ppos);
5837 	setup_per_zone_lowmem_reserve();
5838 	return 0;
5839 }
5840 
5841 /*
5842  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
5843  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
5844  * pagelist can have before it gets flushed back to buddy allocator.
5845  */
5846 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
5847 	void __user *buffer, size_t *length, loff_t *ppos)
5848 {
5849 	struct zone *zone;
5850 	unsigned int cpu;
5851 	int ret;
5852 
5853 	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
5854 	if (!write || (ret < 0))
5855 		return ret;
5856 
5857 	mutex_lock(&pcp_batch_high_lock);
5858 	for_each_populated_zone(zone) {
5859 		unsigned long  high;
5860 		high = zone->managed_pages / percpu_pagelist_fraction;
5861 		for_each_possible_cpu(cpu)
5862 			pageset_set_high(per_cpu_ptr(zone->pageset, cpu),
5863 					 high);
5864 	}
5865 	mutex_unlock(&pcp_batch_high_lock);
5866 	return 0;
5867 }
5868 
5869 int hashdist = HASHDIST_DEFAULT;
5870 
5871 #ifdef CONFIG_NUMA
5872 static int __init set_hashdist(char *str)
5873 {
5874 	if (!str)
5875 		return 0;
5876 	hashdist = simple_strtoul(str, &str, 0);
5877 	return 1;
5878 }
5879 __setup("hashdist=", set_hashdist);
5880 #endif
5881 
5882 /*
5883  * allocate a large system hash table from bootmem
5884  * - it is assumed that the hash table must contain an exact power-of-2
5885  *   quantity of entries
5886  * - limit is the number of hash buckets, not the total allocation size
5887  */
5888 void *__init alloc_large_system_hash(const char *tablename,
5889 				     unsigned long bucketsize,
5890 				     unsigned long numentries,
5891 				     int scale,
5892 				     int flags,
5893 				     unsigned int *_hash_shift,
5894 				     unsigned int *_hash_mask,
5895 				     unsigned long low_limit,
5896 				     unsigned long high_limit)
5897 {
5898 	unsigned long long max = high_limit;
5899 	unsigned long log2qty, size;
5900 	void *table = NULL;
5901 
5902 	/* allow the kernel cmdline to have a say */
5903 	if (!numentries) {
5904 		/* round applicable memory size up to nearest megabyte */
5905 		numentries = nr_kernel_pages;
5906 
5907 		/* It isn't necessary when PAGE_SIZE >= 1MB */
5908 		if (PAGE_SHIFT < 20)
5909 			numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
5910 
5911 		/* limit to 1 bucket per 2^scale bytes of low memory */
5912 		if (scale > PAGE_SHIFT)
5913 			numentries >>= (scale - PAGE_SHIFT);
5914 		else
5915 			numentries <<= (PAGE_SHIFT - scale);
5916 
5917 		/* Make sure we've got at least a 0-order allocation.. */
5918 		if (unlikely(flags & HASH_SMALL)) {
5919 			/* Makes no sense without HASH_EARLY */
5920 			WARN_ON(!(flags & HASH_EARLY));
5921 			if (!(numentries >> *_hash_shift)) {
5922 				numentries = 1UL << *_hash_shift;
5923 				BUG_ON(!numentries);
5924 			}
5925 		} else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
5926 			numentries = PAGE_SIZE / bucketsize;
5927 	}
5928 	numentries = roundup_pow_of_two(numentries);
5929 
5930 	/* limit allocation size to 1/16 total memory by default */
5931 	if (max == 0) {
5932 		max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
5933 		do_div(max, bucketsize);
5934 	}
5935 	max = min(max, 0x80000000ULL);
5936 
5937 	if (numentries < low_limit)
5938 		numentries = low_limit;
5939 	if (numentries > max)
5940 		numentries = max;
5941 
5942 	log2qty = ilog2(numentries);
5943 
5944 	do {
5945 		size = bucketsize << log2qty;
5946 		if (flags & HASH_EARLY)
5947 			table = memblock_virt_alloc_nopanic(size, 0);
5948 		else if (hashdist)
5949 			table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
5950 		else {
5951 			/*
5952 			 * If bucketsize is not a power-of-two, we may free
5953 			 * some pages at the end of hash table which
5954 			 * alloc_pages_exact() automatically does
5955 			 */
5956 			if (get_order(size) < MAX_ORDER) {
5957 				table = alloc_pages_exact(size, GFP_ATOMIC);
5958 				kmemleak_alloc(table, size, 1, GFP_ATOMIC);
5959 			}
5960 		}
5961 	} while (!table && size > PAGE_SIZE && --log2qty);
5962 
5963 	if (!table)
5964 		panic("Failed to allocate %s hash table\n", tablename);
5965 
5966 	printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n",
5967 	       tablename,
5968 	       (1UL << log2qty),
5969 	       ilog2(size) - PAGE_SHIFT,
5970 	       size);
5971 
5972 	if (_hash_shift)
5973 		*_hash_shift = log2qty;
5974 	if (_hash_mask)
5975 		*_hash_mask = (1 << log2qty) - 1;
5976 
5977 	return table;
5978 }
5979 
5980 /* Return a pointer to the bitmap storing bits affecting a block of pages */
5981 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
5982 							unsigned long pfn)
5983 {
5984 #ifdef CONFIG_SPARSEMEM
5985 	return __pfn_to_section(pfn)->pageblock_flags;
5986 #else
5987 	return zone->pageblock_flags;
5988 #endif /* CONFIG_SPARSEMEM */
5989 }
5990 
5991 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
5992 {
5993 #ifdef CONFIG_SPARSEMEM
5994 	pfn &= (PAGES_PER_SECTION-1);
5995 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5996 #else
5997 	pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
5998 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
5999 #endif /* CONFIG_SPARSEMEM */
6000 }
6001 
6002 /**
6003  * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
6004  * @page: The page within the block of interest
6005  * @start_bitidx: The first bit of interest to retrieve
6006  * @end_bitidx: The last bit of interest
6007  * returns pageblock_bits flags
6008  */
6009 unsigned long get_pageblock_flags_group(struct page *page,
6010 					int start_bitidx, int end_bitidx)
6011 {
6012 	struct zone *zone;
6013 	unsigned long *bitmap;
6014 	unsigned long pfn, bitidx;
6015 	unsigned long flags = 0;
6016 	unsigned long value = 1;
6017 
6018 	zone = page_zone(page);
6019 	pfn = page_to_pfn(page);
6020 	bitmap = get_pageblock_bitmap(zone, pfn);
6021 	bitidx = pfn_to_bitidx(zone, pfn);
6022 
6023 	for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
6024 		if (test_bit(bitidx + start_bitidx, bitmap))
6025 			flags |= value;
6026 
6027 	return flags;
6028 }
6029 
6030 /**
6031  * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
6032  * @page: The page within the block of interest
6033  * @start_bitidx: The first bit of interest
6034  * @end_bitidx: The last bit of interest
6035  * @flags: The flags to set
6036  */
6037 void set_pageblock_flags_group(struct page *page, unsigned long flags,
6038 					int start_bitidx, int end_bitidx)
6039 {
6040 	struct zone *zone;
6041 	unsigned long *bitmap;
6042 	unsigned long pfn, bitidx;
6043 	unsigned long value = 1;
6044 
6045 	zone = page_zone(page);
6046 	pfn = page_to_pfn(page);
6047 	bitmap = get_pageblock_bitmap(zone, pfn);
6048 	bitidx = pfn_to_bitidx(zone, pfn);
6049 	VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
6050 
6051 	for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
6052 		if (flags & value)
6053 			__set_bit(bitidx + start_bitidx, bitmap);
6054 		else
6055 			__clear_bit(bitidx + start_bitidx, bitmap);
6056 }
6057 
6058 /*
6059  * This function checks whether pageblock includes unmovable pages or not.
6060  * If @count is not zero, it is okay to include less @count unmovable pages
6061  *
6062  * PageLRU check without isolation or lru_lock could race so that
6063  * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6064  * expect this function should be exact.
6065  */
6066 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6067 			 bool skip_hwpoisoned_pages)
6068 {
6069 	unsigned long pfn, iter, found;
6070 	int mt;
6071 
6072 	/*
6073 	 * For avoiding noise data, lru_add_drain_all() should be called
6074 	 * If ZONE_MOVABLE, the zone never contains unmovable pages
6075 	 */
6076 	if (zone_idx(zone) == ZONE_MOVABLE)
6077 		return false;
6078 	mt = get_pageblock_migratetype(page);
6079 	if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6080 		return false;
6081 
6082 	pfn = page_to_pfn(page);
6083 	for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6084 		unsigned long check = pfn + iter;
6085 
6086 		if (!pfn_valid_within(check))
6087 			continue;
6088 
6089 		page = pfn_to_page(check);
6090 
6091 		/*
6092 		 * Hugepages are not in LRU lists, but they're movable.
6093 		 * We need not scan over tail pages bacause we don't
6094 		 * handle each tail page individually in migration.
6095 		 */
6096 		if (PageHuge(page)) {
6097 			iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6098 			continue;
6099 		}
6100 
6101 		/*
6102 		 * We can't use page_count without pin a page
6103 		 * because another CPU can free compound page.
6104 		 * This check already skips compound tails of THP
6105 		 * because their page->_count is zero at all time.
6106 		 */
6107 		if (!atomic_read(&page->_count)) {
6108 			if (PageBuddy(page))
6109 				iter += (1 << page_order(page)) - 1;
6110 			continue;
6111 		}
6112 
6113 		/*
6114 		 * The HWPoisoned page may be not in buddy system, and
6115 		 * page_count() is not 0.
6116 		 */
6117 		if (skip_hwpoisoned_pages && PageHWPoison(page))
6118 			continue;
6119 
6120 		if (!PageLRU(page))
6121 			found++;
6122 		/*
6123 		 * If there are RECLAIMABLE pages, we need to check it.
6124 		 * But now, memory offline itself doesn't call shrink_slab()
6125 		 * and it still to be fixed.
6126 		 */
6127 		/*
6128 		 * If the page is not RAM, page_count()should be 0.
6129 		 * we don't need more check. This is an _used_ not-movable page.
6130 		 *
6131 		 * The problematic thing here is PG_reserved pages. PG_reserved
6132 		 * is set to both of a memory hole page and a _used_ kernel
6133 		 * page at boot.
6134 		 */
6135 		if (found > count)
6136 			return true;
6137 	}
6138 	return false;
6139 }
6140 
6141 bool is_pageblock_removable_nolock(struct page *page)
6142 {
6143 	struct zone *zone;
6144 	unsigned long pfn;
6145 
6146 	/*
6147 	 * We have to be careful here because we are iterating over memory
6148 	 * sections which are not zone aware so we might end up outside of
6149 	 * the zone but still within the section.
6150 	 * We have to take care about the node as well. If the node is offline
6151 	 * its NODE_DATA will be NULL - see page_zone.
6152 	 */
6153 	if (!node_online(page_to_nid(page)))
6154 		return false;
6155 
6156 	zone = page_zone(page);
6157 	pfn = page_to_pfn(page);
6158 	if (!zone_spans_pfn(zone, pfn))
6159 		return false;
6160 
6161 	return !has_unmovable_pages(zone, page, 0, true);
6162 }
6163 
6164 #ifdef CONFIG_CMA
6165 
6166 static unsigned long pfn_max_align_down(unsigned long pfn)
6167 {
6168 	return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6169 			     pageblock_nr_pages) - 1);
6170 }
6171 
6172 static unsigned long pfn_max_align_up(unsigned long pfn)
6173 {
6174 	return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6175 				pageblock_nr_pages));
6176 }
6177 
6178 /* [start, end) must belong to a single zone. */
6179 static int __alloc_contig_migrate_range(struct compact_control *cc,
6180 					unsigned long start, unsigned long end)
6181 {
6182 	/* This function is based on compact_zone() from compaction.c. */
6183 	unsigned long nr_reclaimed;
6184 	unsigned long pfn = start;
6185 	unsigned int tries = 0;
6186 	int ret = 0;
6187 
6188 	migrate_prep();
6189 
6190 	while (pfn < end || !list_empty(&cc->migratepages)) {
6191 		if (fatal_signal_pending(current)) {
6192 			ret = -EINTR;
6193 			break;
6194 		}
6195 
6196 		if (list_empty(&cc->migratepages)) {
6197 			cc->nr_migratepages = 0;
6198 			pfn = isolate_migratepages_range(cc->zone, cc,
6199 							 pfn, end, true);
6200 			if (!pfn) {
6201 				ret = -EINTR;
6202 				break;
6203 			}
6204 			tries = 0;
6205 		} else if (++tries == 5) {
6206 			ret = ret < 0 ? ret : -EBUSY;
6207 			break;
6208 		}
6209 
6210 		nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6211 							&cc->migratepages);
6212 		cc->nr_migratepages -= nr_reclaimed;
6213 
6214 		ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6215 				    0, MIGRATE_SYNC, MR_CMA);
6216 	}
6217 	if (ret < 0) {
6218 		putback_movable_pages(&cc->migratepages);
6219 		return ret;
6220 	}
6221 	return 0;
6222 }
6223 
6224 /**
6225  * alloc_contig_range() -- tries to allocate given range of pages
6226  * @start:	start PFN to allocate
6227  * @end:	one-past-the-last PFN to allocate
6228  * @migratetype:	migratetype of the underlaying pageblocks (either
6229  *			#MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
6230  *			in range must have the same migratetype and it must
6231  *			be either of the two.
6232  *
6233  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6234  * aligned, however it's the caller's responsibility to guarantee that
6235  * we are the only thread that changes migrate type of pageblocks the
6236  * pages fall in.
6237  *
6238  * The PFN range must belong to a single zone.
6239  *
6240  * Returns zero on success or negative error code.  On success all
6241  * pages which PFN is in [start, end) are allocated for the caller and
6242  * need to be freed with free_contig_range().
6243  */
6244 int alloc_contig_range(unsigned long start, unsigned long end,
6245 		       unsigned migratetype)
6246 {
6247 	unsigned long outer_start, outer_end;
6248 	int ret = 0, order;
6249 
6250 	struct compact_control cc = {
6251 		.nr_migratepages = 0,
6252 		.order = -1,
6253 		.zone = page_zone(pfn_to_page(start)),
6254 		.sync = true,
6255 		.ignore_skip_hint = true,
6256 	};
6257 	INIT_LIST_HEAD(&cc.migratepages);
6258 
6259 	/*
6260 	 * What we do here is we mark all pageblocks in range as
6261 	 * MIGRATE_ISOLATE.  Because pageblock and max order pages may
6262 	 * have different sizes, and due to the way page allocator
6263 	 * work, we align the range to biggest of the two pages so
6264 	 * that page allocator won't try to merge buddies from
6265 	 * different pageblocks and change MIGRATE_ISOLATE to some
6266 	 * other migration type.
6267 	 *
6268 	 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6269 	 * migrate the pages from an unaligned range (ie. pages that
6270 	 * we are interested in).  This will put all the pages in
6271 	 * range back to page allocator as MIGRATE_ISOLATE.
6272 	 *
6273 	 * When this is done, we take the pages in range from page
6274 	 * allocator removing them from the buddy system.  This way
6275 	 * page allocator will never consider using them.
6276 	 *
6277 	 * This lets us mark the pageblocks back as
6278 	 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6279 	 * aligned range but not in the unaligned, original range are
6280 	 * put back to page allocator so that buddy can use them.
6281 	 */
6282 
6283 	ret = start_isolate_page_range(pfn_max_align_down(start),
6284 				       pfn_max_align_up(end), migratetype,
6285 				       false);
6286 	if (ret)
6287 		return ret;
6288 
6289 	ret = __alloc_contig_migrate_range(&cc, start, end);
6290 	if (ret)
6291 		goto done;
6292 
6293 	/*
6294 	 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6295 	 * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
6296 	 * more, all pages in [start, end) are free in page allocator.
6297 	 * What we are going to do is to allocate all pages from
6298 	 * [start, end) (that is remove them from page allocator).
6299 	 *
6300 	 * The only problem is that pages at the beginning and at the
6301 	 * end of interesting range may be not aligned with pages that
6302 	 * page allocator holds, ie. they can be part of higher order
6303 	 * pages.  Because of this, we reserve the bigger range and
6304 	 * once this is done free the pages we are not interested in.
6305 	 *
6306 	 * We don't have to hold zone->lock here because the pages are
6307 	 * isolated thus they won't get removed from buddy.
6308 	 */
6309 
6310 	lru_add_drain_all();
6311 	drain_all_pages();
6312 
6313 	order = 0;
6314 	outer_start = start;
6315 	while (!PageBuddy(pfn_to_page(outer_start))) {
6316 		if (++order >= MAX_ORDER) {
6317 			ret = -EBUSY;
6318 			goto done;
6319 		}
6320 		outer_start &= ~0UL << order;
6321 	}
6322 
6323 	/* Make sure the range is really isolated. */
6324 	if (test_pages_isolated(outer_start, end, false)) {
6325 		pr_warn("alloc_contig_range test_pages_isolated(%lx, %lx) failed\n",
6326 		       outer_start, end);
6327 		ret = -EBUSY;
6328 		goto done;
6329 	}
6330 
6331 
6332 	/* Grab isolated pages from freelists. */
6333 	outer_end = isolate_freepages_range(&cc, outer_start, end);
6334 	if (!outer_end) {
6335 		ret = -EBUSY;
6336 		goto done;
6337 	}
6338 
6339 	/* Free head and tail (if any) */
6340 	if (start != outer_start)
6341 		free_contig_range(outer_start, start - outer_start);
6342 	if (end != outer_end)
6343 		free_contig_range(end, outer_end - end);
6344 
6345 done:
6346 	undo_isolate_page_range(pfn_max_align_down(start),
6347 				pfn_max_align_up(end), migratetype);
6348 	return ret;
6349 }
6350 
6351 void free_contig_range(unsigned long pfn, unsigned nr_pages)
6352 {
6353 	unsigned int count = 0;
6354 
6355 	for (; nr_pages--; pfn++) {
6356 		struct page *page = pfn_to_page(pfn);
6357 
6358 		count += page_count(page) != 1;
6359 		__free_page(page);
6360 	}
6361 	WARN(count != 0, "%d pages are still in use!\n", count);
6362 }
6363 #endif
6364 
6365 #ifdef CONFIG_MEMORY_HOTPLUG
6366 /*
6367  * The zone indicated has a new number of managed_pages; batch sizes and percpu
6368  * page high values need to be recalulated.
6369  */
6370 void __meminit zone_pcp_update(struct zone *zone)
6371 {
6372 	unsigned cpu;
6373 	mutex_lock(&pcp_batch_high_lock);
6374 	for_each_possible_cpu(cpu)
6375 		pageset_set_high_and_batch(zone,
6376 				per_cpu_ptr(zone->pageset, cpu));
6377 	mutex_unlock(&pcp_batch_high_lock);
6378 }
6379 #endif
6380 
6381 void zone_pcp_reset(struct zone *zone)
6382 {
6383 	unsigned long flags;
6384 	int cpu;
6385 	struct per_cpu_pageset *pset;
6386 
6387 	/* avoid races with drain_pages()  */
6388 	local_irq_save(flags);
6389 	if (zone->pageset != &boot_pageset) {
6390 		for_each_online_cpu(cpu) {
6391 			pset = per_cpu_ptr(zone->pageset, cpu);
6392 			drain_zonestat(zone, pset);
6393 		}
6394 		free_percpu(zone->pageset);
6395 		zone->pageset = &boot_pageset;
6396 	}
6397 	local_irq_restore(flags);
6398 }
6399 
6400 #ifdef CONFIG_MEMORY_HOTREMOVE
6401 /*
6402  * All pages in the range must be isolated before calling this.
6403  */
6404 void
6405 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
6406 {
6407 	struct page *page;
6408 	struct zone *zone;
6409 	int order, i;
6410 	unsigned long pfn;
6411 	unsigned long flags;
6412 	/* find the first valid pfn */
6413 	for (pfn = start_pfn; pfn < end_pfn; pfn++)
6414 		if (pfn_valid(pfn))
6415 			break;
6416 	if (pfn == end_pfn)
6417 		return;
6418 	zone = page_zone(pfn_to_page(pfn));
6419 	spin_lock_irqsave(&zone->lock, flags);
6420 	pfn = start_pfn;
6421 	while (pfn < end_pfn) {
6422 		if (!pfn_valid(pfn)) {
6423 			pfn++;
6424 			continue;
6425 		}
6426 		page = pfn_to_page(pfn);
6427 		/*
6428 		 * The HWPoisoned page may be not in buddy system, and
6429 		 * page_count() is not 0.
6430 		 */
6431 		if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
6432 			pfn++;
6433 			SetPageReserved(page);
6434 			continue;
6435 		}
6436 
6437 		BUG_ON(page_count(page));
6438 		BUG_ON(!PageBuddy(page));
6439 		order = page_order(page);
6440 #ifdef CONFIG_DEBUG_VM
6441 		printk(KERN_INFO "remove from free list %lx %d %lx\n",
6442 		       pfn, 1 << order, end_pfn);
6443 #endif
6444 		list_del(&page->lru);
6445 		rmv_page_order(page);
6446 		zone->free_area[order].nr_free--;
6447 		for (i = 0; i < (1 << order); i++)
6448 			SetPageReserved((page+i));
6449 		pfn += (1 << order);
6450 	}
6451 	spin_unlock_irqrestore(&zone->lock, flags);
6452 }
6453 #endif
6454 
6455 #ifdef CONFIG_MEMORY_FAILURE
6456 bool is_free_buddy_page(struct page *page)
6457 {
6458 	struct zone *zone = page_zone(page);
6459 	unsigned long pfn = page_to_pfn(page);
6460 	unsigned long flags;
6461 	int order;
6462 
6463 	spin_lock_irqsave(&zone->lock, flags);
6464 	for (order = 0; order < MAX_ORDER; order++) {
6465 		struct page *page_head = page - (pfn & ((1 << order) - 1));
6466 
6467 		if (PageBuddy(page_head) && page_order(page_head) >= order)
6468 			break;
6469 	}
6470 	spin_unlock_irqrestore(&zone->lock, flags);
6471 
6472 	return order < MAX_ORDER;
6473 }
6474 #endif
6475 
6476 static const struct trace_print_flags pageflag_names[] = {
6477 	{1UL << PG_locked,		"locked"	},
6478 	{1UL << PG_error,		"error"		},
6479 	{1UL << PG_referenced,		"referenced"	},
6480 	{1UL << PG_uptodate,		"uptodate"	},
6481 	{1UL << PG_dirty,		"dirty"		},
6482 	{1UL << PG_lru,			"lru"		},
6483 	{1UL << PG_active,		"active"	},
6484 	{1UL << PG_slab,		"slab"		},
6485 	{1UL << PG_owner_priv_1,	"owner_priv_1"	},
6486 	{1UL << PG_arch_1,		"arch_1"	},
6487 	{1UL << PG_reserved,		"reserved"	},
6488 	{1UL << PG_private,		"private"	},
6489 	{1UL << PG_private_2,		"private_2"	},
6490 	{1UL << PG_writeback,		"writeback"	},
6491 #ifdef CONFIG_PAGEFLAGS_EXTENDED
6492 	{1UL << PG_head,		"head"		},
6493 	{1UL << PG_tail,		"tail"		},
6494 #else
6495 	{1UL << PG_compound,		"compound"	},
6496 #endif
6497 	{1UL << PG_swapcache,		"swapcache"	},
6498 	{1UL << PG_mappedtodisk,	"mappedtodisk"	},
6499 	{1UL << PG_reclaim,		"reclaim"	},
6500 	{1UL << PG_swapbacked,		"swapbacked"	},
6501 	{1UL << PG_unevictable,		"unevictable"	},
6502 #ifdef CONFIG_MMU
6503 	{1UL << PG_mlocked,		"mlocked"	},
6504 #endif
6505 #ifdef CONFIG_ARCH_USES_PG_UNCACHED
6506 	{1UL << PG_uncached,		"uncached"	},
6507 #endif
6508 #ifdef CONFIG_MEMORY_FAILURE
6509 	{1UL << PG_hwpoison,		"hwpoison"	},
6510 #endif
6511 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6512 	{1UL << PG_compound_lock,	"compound_lock"	},
6513 #endif
6514 };
6515 
6516 static void dump_page_flags(unsigned long flags)
6517 {
6518 	const char *delim = "";
6519 	unsigned long mask;
6520 	int i;
6521 
6522 	BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS);
6523 
6524 	printk(KERN_ALERT "page flags: %#lx(", flags);
6525 
6526 	/* remove zone id */
6527 	flags &= (1UL << NR_PAGEFLAGS) - 1;
6528 
6529 	for (i = 0; i < ARRAY_SIZE(pageflag_names) && flags; i++) {
6530 
6531 		mask = pageflag_names[i].mask;
6532 		if ((flags & mask) != mask)
6533 			continue;
6534 
6535 		flags &= ~mask;
6536 		printk("%s%s", delim, pageflag_names[i].name);
6537 		delim = "|";
6538 	}
6539 
6540 	/* check for left over flags */
6541 	if (flags)
6542 		printk("%s%#lx", delim, flags);
6543 
6544 	printk(")\n");
6545 }
6546 
6547 void dump_page_badflags(struct page *page, char *reason, unsigned long badflags)
6548 {
6549 	printk(KERN_ALERT
6550 	       "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n",
6551 		page, atomic_read(&page->_count), page_mapcount(page),
6552 		page->mapping, page->index);
6553 	dump_page_flags(page->flags);
6554 	if (reason)
6555 		pr_alert("page dumped because: %s\n", reason);
6556 	if (page->flags & badflags) {
6557 		pr_alert("bad because of flags:\n");
6558 		dump_page_flags(page->flags & badflags);
6559 	}
6560 	mem_cgroup_print_bad_page(page);
6561 }
6562 
6563 void dump_page(struct page *page, char *reason)
6564 {
6565 	dump_page_badflags(page, reason, 0);
6566 }
6567 EXPORT_SYMBOL_GPL(dump_page);
6568