xref: /openbmc/linux/mm/page_alloc.c (revision 2791f47d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/mm/page_alloc.c
4  *
5  *  Manages the free list, the system allocates free pages here.
6  *  Note that kmalloc() lives in slab.c
7  *
8  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
9  *  Swap reorganised 29.12.95, Stephen Tweedie
10  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
11  *  Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
12  *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
13  *  Zone balancing, Kanoj Sarcar, SGI, Jan 2000
14  *  Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
15  *          (lots of bits borrowed from Ingo Molnar & Andrew Morton)
16  */
17 
18 #include <linux/stddef.h>
19 #include <linux/mm.h>
20 #include <linux/highmem.h>
21 #include <linux/swap.h>
22 #include <linux/interrupt.h>
23 #include <linux/pagemap.h>
24 #include <linux/jiffies.h>
25 #include <linux/memblock.h>
26 #include <linux/compiler.h>
27 #include <linux/kernel.h>
28 #include <linux/kasan.h>
29 #include <linux/module.h>
30 #include <linux/suspend.h>
31 #include <linux/pagevec.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/ratelimit.h>
35 #include <linux/oom.h>
36 #include <linux/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/memremap.h>
46 #include <linux/stop_machine.h>
47 #include <linux/random.h>
48 #include <linux/sort.h>
49 #include <linux/pfn.h>
50 #include <linux/backing-dev.h>
51 #include <linux/fault-inject.h>
52 #include <linux/page-isolation.h>
53 #include <linux/debugobjects.h>
54 #include <linux/kmemleak.h>
55 #include <linux/compaction.h>
56 #include <trace/events/kmem.h>
57 #include <trace/events/oom.h>
58 #include <linux/prefetch.h>
59 #include <linux/mm_inline.h>
60 #include <linux/migrate.h>
61 #include <linux/hugetlb.h>
62 #include <linux/sched/rt.h>
63 #include <linux/sched/mm.h>
64 #include <linux/page_owner.h>
65 #include <linux/kthread.h>
66 #include <linux/memcontrol.h>
67 #include <linux/ftrace.h>
68 #include <linux/lockdep.h>
69 #include <linux/nmi.h>
70 #include <linux/psi.h>
71 #include <linux/padata.h>
72 #include <linux/khugepaged.h>
73 
74 #include <asm/sections.h>
75 #include <asm/tlbflush.h>
76 #include <asm/div64.h>
77 #include "internal.h"
78 #include "shuffle.h"
79 #include "page_reporting.h"
80 
81 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
82 static DEFINE_MUTEX(pcp_batch_high_lock);
83 #define MIN_PERCPU_PAGELIST_FRACTION	(8)
84 
85 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
86 DEFINE_PER_CPU(int, numa_node);
87 EXPORT_PER_CPU_SYMBOL(numa_node);
88 #endif
89 
90 DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key);
91 
92 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
93 /*
94  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
95  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
96  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
97  * defined in <linux/topology.h>.
98  */
99 DEFINE_PER_CPU(int, _numa_mem_);		/* Kernel "local memory" node */
100 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
101 #endif
102 
103 /* work_structs for global per-cpu drains */
104 struct pcpu_drain {
105 	struct zone *zone;
106 	struct work_struct work;
107 };
108 static DEFINE_MUTEX(pcpu_drain_mutex);
109 static DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain);
110 
111 #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
112 volatile unsigned long latent_entropy __latent_entropy;
113 EXPORT_SYMBOL(latent_entropy);
114 #endif
115 
116 /*
117  * Array of node states.
118  */
119 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
120 	[N_POSSIBLE] = NODE_MASK_ALL,
121 	[N_ONLINE] = { { [0] = 1UL } },
122 #ifndef CONFIG_NUMA
123 	[N_NORMAL_MEMORY] = { { [0] = 1UL } },
124 #ifdef CONFIG_HIGHMEM
125 	[N_HIGH_MEMORY] = { { [0] = 1UL } },
126 #endif
127 	[N_MEMORY] = { { [0] = 1UL } },
128 	[N_CPU] = { { [0] = 1UL } },
129 #endif	/* NUMA */
130 };
131 EXPORT_SYMBOL(node_states);
132 
133 atomic_long_t _totalram_pages __read_mostly;
134 EXPORT_SYMBOL(_totalram_pages);
135 unsigned long totalreserve_pages __read_mostly;
136 unsigned long totalcma_pages __read_mostly;
137 
138 int percpu_pagelist_fraction;
139 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
140 #ifdef CONFIG_INIT_ON_ALLOC_DEFAULT_ON
141 DEFINE_STATIC_KEY_TRUE(init_on_alloc);
142 #else
143 DEFINE_STATIC_KEY_FALSE(init_on_alloc);
144 #endif
145 EXPORT_SYMBOL(init_on_alloc);
146 
147 #ifdef CONFIG_INIT_ON_FREE_DEFAULT_ON
148 DEFINE_STATIC_KEY_TRUE(init_on_free);
149 #else
150 DEFINE_STATIC_KEY_FALSE(init_on_free);
151 #endif
152 EXPORT_SYMBOL(init_on_free);
153 
154 static int __init early_init_on_alloc(char *buf)
155 {
156 	int ret;
157 	bool bool_result;
158 
159 	ret = kstrtobool(buf, &bool_result);
160 	if (ret)
161 		return ret;
162 	if (bool_result && page_poisoning_enabled())
163 		pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_alloc\n");
164 	if (bool_result)
165 		static_branch_enable(&init_on_alloc);
166 	else
167 		static_branch_disable(&init_on_alloc);
168 	return 0;
169 }
170 early_param("init_on_alloc", early_init_on_alloc);
171 
172 static int __init early_init_on_free(char *buf)
173 {
174 	int ret;
175 	bool bool_result;
176 
177 	ret = kstrtobool(buf, &bool_result);
178 	if (ret)
179 		return ret;
180 	if (bool_result && page_poisoning_enabled())
181 		pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, will take precedence over init_on_free\n");
182 	if (bool_result)
183 		static_branch_enable(&init_on_free);
184 	else
185 		static_branch_disable(&init_on_free);
186 	return 0;
187 }
188 early_param("init_on_free", early_init_on_free);
189 
190 /*
191  * A cached value of the page's pageblock's migratetype, used when the page is
192  * put on a pcplist. Used to avoid the pageblock migratetype lookup when
193  * freeing from pcplists in most cases, at the cost of possibly becoming stale.
194  * Also the migratetype set in the page does not necessarily match the pcplist
195  * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
196  * other index - this ensures that it will be put on the correct CMA freelist.
197  */
198 static inline int get_pcppage_migratetype(struct page *page)
199 {
200 	return page->index;
201 }
202 
203 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
204 {
205 	page->index = migratetype;
206 }
207 
208 #ifdef CONFIG_PM_SLEEP
209 /*
210  * The following functions are used by the suspend/hibernate code to temporarily
211  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
212  * while devices are suspended.  To avoid races with the suspend/hibernate code,
213  * they should always be called with system_transition_mutex held
214  * (gfp_allowed_mask also should only be modified with system_transition_mutex
215  * held, unless the suspend/hibernate code is guaranteed not to run in parallel
216  * with that modification).
217  */
218 
219 static gfp_t saved_gfp_mask;
220 
221 void pm_restore_gfp_mask(void)
222 {
223 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
224 	if (saved_gfp_mask) {
225 		gfp_allowed_mask = saved_gfp_mask;
226 		saved_gfp_mask = 0;
227 	}
228 }
229 
230 void pm_restrict_gfp_mask(void)
231 {
232 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
233 	WARN_ON(saved_gfp_mask);
234 	saved_gfp_mask = gfp_allowed_mask;
235 	gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
236 }
237 
238 bool pm_suspended_storage(void)
239 {
240 	if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
241 		return false;
242 	return true;
243 }
244 #endif /* CONFIG_PM_SLEEP */
245 
246 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
247 unsigned int pageblock_order __read_mostly;
248 #endif
249 
250 static void __free_pages_ok(struct page *page, unsigned int order);
251 
252 /*
253  * results with 256, 32 in the lowmem_reserve sysctl:
254  *	1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
255  *	1G machine -> (16M dma, 784M normal, 224M high)
256  *	NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
257  *	HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
258  *	HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
259  *
260  * TBD: should special case ZONE_DMA32 machines here - in those we normally
261  * don't need any ZONE_NORMAL reservation
262  */
263 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = {
264 #ifdef CONFIG_ZONE_DMA
265 	[ZONE_DMA] = 256,
266 #endif
267 #ifdef CONFIG_ZONE_DMA32
268 	[ZONE_DMA32] = 256,
269 #endif
270 	[ZONE_NORMAL] = 32,
271 #ifdef CONFIG_HIGHMEM
272 	[ZONE_HIGHMEM] = 0,
273 #endif
274 	[ZONE_MOVABLE] = 0,
275 };
276 
277 static char * const zone_names[MAX_NR_ZONES] = {
278 #ifdef CONFIG_ZONE_DMA
279 	 "DMA",
280 #endif
281 #ifdef CONFIG_ZONE_DMA32
282 	 "DMA32",
283 #endif
284 	 "Normal",
285 #ifdef CONFIG_HIGHMEM
286 	 "HighMem",
287 #endif
288 	 "Movable",
289 #ifdef CONFIG_ZONE_DEVICE
290 	 "Device",
291 #endif
292 };
293 
294 const char * const migratetype_names[MIGRATE_TYPES] = {
295 	"Unmovable",
296 	"Movable",
297 	"Reclaimable",
298 	"HighAtomic",
299 #ifdef CONFIG_CMA
300 	"CMA",
301 #endif
302 #ifdef CONFIG_MEMORY_ISOLATION
303 	"Isolate",
304 #endif
305 };
306 
307 compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = {
308 	[NULL_COMPOUND_DTOR] = NULL,
309 	[COMPOUND_PAGE_DTOR] = free_compound_page,
310 #ifdef CONFIG_HUGETLB_PAGE
311 	[HUGETLB_PAGE_DTOR] = free_huge_page,
312 #endif
313 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
314 	[TRANSHUGE_PAGE_DTOR] = free_transhuge_page,
315 #endif
316 };
317 
318 int min_free_kbytes = 1024;
319 int user_min_free_kbytes = -1;
320 #ifdef CONFIG_DISCONTIGMEM
321 /*
322  * DiscontigMem defines memory ranges as separate pg_data_t even if the ranges
323  * are not on separate NUMA nodes. Functionally this works but with
324  * watermark_boost_factor, it can reclaim prematurely as the ranges can be
325  * quite small. By default, do not boost watermarks on discontigmem as in
326  * many cases very high-order allocations like THP are likely to be
327  * unsupported and the premature reclaim offsets the advantage of long-term
328  * fragmentation avoidance.
329  */
330 int watermark_boost_factor __read_mostly;
331 #else
332 int watermark_boost_factor __read_mostly = 15000;
333 #endif
334 int watermark_scale_factor = 10;
335 
336 static unsigned long nr_kernel_pages __initdata;
337 static unsigned long nr_all_pages __initdata;
338 static unsigned long dma_reserve __initdata;
339 
340 static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata;
341 static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata;
342 static unsigned long required_kernelcore __initdata;
343 static unsigned long required_kernelcore_percent __initdata;
344 static unsigned long required_movablecore __initdata;
345 static unsigned long required_movablecore_percent __initdata;
346 static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata;
347 static bool mirrored_kernelcore __meminitdata;
348 
349 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
350 int movable_zone;
351 EXPORT_SYMBOL(movable_zone);
352 
353 #if MAX_NUMNODES > 1
354 unsigned int nr_node_ids __read_mostly = MAX_NUMNODES;
355 unsigned int nr_online_nodes __read_mostly = 1;
356 EXPORT_SYMBOL(nr_node_ids);
357 EXPORT_SYMBOL(nr_online_nodes);
358 #endif
359 
360 int page_group_by_mobility_disabled __read_mostly;
361 
362 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
363 /*
364  * During boot we initialize deferred pages on-demand, as needed, but once
365  * page_alloc_init_late() has finished, the deferred pages are all initialized,
366  * and we can permanently disable that path.
367  */
368 static DEFINE_STATIC_KEY_TRUE(deferred_pages);
369 
370 /*
371  * Calling kasan_free_pages() only after deferred memory initialization
372  * has completed. Poisoning pages during deferred memory init will greatly
373  * lengthen the process and cause problem in large memory systems as the
374  * deferred pages initialization is done with interrupt disabled.
375  *
376  * Assuming that there will be no reference to those newly initialized
377  * pages before they are ever allocated, this should have no effect on
378  * KASAN memory tracking as the poison will be properly inserted at page
379  * allocation time. The only corner case is when pages are allocated by
380  * on-demand allocation and then freed again before the deferred pages
381  * initialization is done, but this is not likely to happen.
382  */
383 static inline void kasan_free_nondeferred_pages(struct page *page, int order)
384 {
385 	if (!static_branch_unlikely(&deferred_pages))
386 		kasan_free_pages(page, order);
387 }
388 
389 /* Returns true if the struct page for the pfn is uninitialised */
390 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
391 {
392 	int nid = early_pfn_to_nid(pfn);
393 
394 	if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
395 		return true;
396 
397 	return false;
398 }
399 
400 /*
401  * Returns true when the remaining initialisation should be deferred until
402  * later in the boot cycle when it can be parallelised.
403  */
404 static bool __meminit
405 defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
406 {
407 	static unsigned long prev_end_pfn, nr_initialised;
408 
409 	/*
410 	 * prev_end_pfn static that contains the end of previous zone
411 	 * No need to protect because called very early in boot before smp_init.
412 	 */
413 	if (prev_end_pfn != end_pfn) {
414 		prev_end_pfn = end_pfn;
415 		nr_initialised = 0;
416 	}
417 
418 	/* Always populate low zones for address-constrained allocations */
419 	if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
420 		return false;
421 
422 	/*
423 	 * We start only with one section of pages, more pages are added as
424 	 * needed until the rest of deferred pages are initialized.
425 	 */
426 	nr_initialised++;
427 	if ((nr_initialised > PAGES_PER_SECTION) &&
428 	    (pfn & (PAGES_PER_SECTION - 1)) == 0) {
429 		NODE_DATA(nid)->first_deferred_pfn = pfn;
430 		return true;
431 	}
432 	return false;
433 }
434 #else
435 #define kasan_free_nondeferred_pages(p, o)	kasan_free_pages(p, o)
436 
437 static inline bool early_page_uninitialised(unsigned long pfn)
438 {
439 	return false;
440 }
441 
442 static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
443 {
444 	return false;
445 }
446 #endif
447 
448 /* Return a pointer to the bitmap storing bits affecting a block of pages */
449 static inline unsigned long *get_pageblock_bitmap(struct page *page,
450 							unsigned long pfn)
451 {
452 #ifdef CONFIG_SPARSEMEM
453 	return section_to_usemap(__pfn_to_section(pfn));
454 #else
455 	return page_zone(page)->pageblock_flags;
456 #endif /* CONFIG_SPARSEMEM */
457 }
458 
459 static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
460 {
461 #ifdef CONFIG_SPARSEMEM
462 	pfn &= (PAGES_PER_SECTION-1);
463 #else
464 	pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
465 #endif /* CONFIG_SPARSEMEM */
466 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
467 }
468 
469 /**
470  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
471  * @page: The page within the block of interest
472  * @pfn: The target page frame number
473  * @mask: mask of bits that the caller is interested in
474  *
475  * Return: pageblock_bits flags
476  */
477 static __always_inline
478 unsigned long __get_pfnblock_flags_mask(struct page *page,
479 					unsigned long pfn,
480 					unsigned long mask)
481 {
482 	unsigned long *bitmap;
483 	unsigned long bitidx, word_bitidx;
484 	unsigned long word;
485 
486 	bitmap = get_pageblock_bitmap(page, pfn);
487 	bitidx = pfn_to_bitidx(page, pfn);
488 	word_bitidx = bitidx / BITS_PER_LONG;
489 	bitidx &= (BITS_PER_LONG-1);
490 
491 	word = bitmap[word_bitidx];
492 	return (word >> bitidx) & mask;
493 }
494 
495 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
496 					unsigned long mask)
497 {
498 	return __get_pfnblock_flags_mask(page, pfn, mask);
499 }
500 
501 static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
502 {
503 	return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK);
504 }
505 
506 /**
507  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
508  * @page: The page within the block of interest
509  * @flags: The flags to set
510  * @pfn: The target page frame number
511  * @mask: mask of bits that the caller is interested in
512  */
513 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
514 					unsigned long pfn,
515 					unsigned long mask)
516 {
517 	unsigned long *bitmap;
518 	unsigned long bitidx, word_bitidx;
519 	unsigned long old_word, word;
520 
521 	BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
522 	BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits));
523 
524 	bitmap = get_pageblock_bitmap(page, pfn);
525 	bitidx = pfn_to_bitidx(page, pfn);
526 	word_bitidx = bitidx / BITS_PER_LONG;
527 	bitidx &= (BITS_PER_LONG-1);
528 
529 	VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
530 
531 	mask <<= bitidx;
532 	flags <<= bitidx;
533 
534 	word = READ_ONCE(bitmap[word_bitidx]);
535 	for (;;) {
536 		old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
537 		if (word == old_word)
538 			break;
539 		word = old_word;
540 	}
541 }
542 
543 void set_pageblock_migratetype(struct page *page, int migratetype)
544 {
545 	if (unlikely(page_group_by_mobility_disabled &&
546 		     migratetype < MIGRATE_PCPTYPES))
547 		migratetype = MIGRATE_UNMOVABLE;
548 
549 	set_pfnblock_flags_mask(page, (unsigned long)migratetype,
550 				page_to_pfn(page), MIGRATETYPE_MASK);
551 }
552 
553 #ifdef CONFIG_DEBUG_VM
554 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
555 {
556 	int ret = 0;
557 	unsigned seq;
558 	unsigned long pfn = page_to_pfn(page);
559 	unsigned long sp, start_pfn;
560 
561 	do {
562 		seq = zone_span_seqbegin(zone);
563 		start_pfn = zone->zone_start_pfn;
564 		sp = zone->spanned_pages;
565 		if (!zone_spans_pfn(zone, pfn))
566 			ret = 1;
567 	} while (zone_span_seqretry(zone, seq));
568 
569 	if (ret)
570 		pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
571 			pfn, zone_to_nid(zone), zone->name,
572 			start_pfn, start_pfn + sp);
573 
574 	return ret;
575 }
576 
577 static int page_is_consistent(struct zone *zone, struct page *page)
578 {
579 	if (!pfn_valid_within(page_to_pfn(page)))
580 		return 0;
581 	if (zone != page_zone(page))
582 		return 0;
583 
584 	return 1;
585 }
586 /*
587  * Temporary debugging check for pages not lying within a given zone.
588  */
589 static int __maybe_unused bad_range(struct zone *zone, struct page *page)
590 {
591 	if (page_outside_zone_boundaries(zone, page))
592 		return 1;
593 	if (!page_is_consistent(zone, page))
594 		return 1;
595 
596 	return 0;
597 }
598 #else
599 static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
600 {
601 	return 0;
602 }
603 #endif
604 
605 static void bad_page(struct page *page, const char *reason)
606 {
607 	static unsigned long resume;
608 	static unsigned long nr_shown;
609 	static unsigned long nr_unshown;
610 
611 	/*
612 	 * Allow a burst of 60 reports, then keep quiet for that minute;
613 	 * or allow a steady drip of one report per second.
614 	 */
615 	if (nr_shown == 60) {
616 		if (time_before(jiffies, resume)) {
617 			nr_unshown++;
618 			goto out;
619 		}
620 		if (nr_unshown) {
621 			pr_alert(
622 			      "BUG: Bad page state: %lu messages suppressed\n",
623 				nr_unshown);
624 			nr_unshown = 0;
625 		}
626 		nr_shown = 0;
627 	}
628 	if (nr_shown++ == 0)
629 		resume = jiffies + 60 * HZ;
630 
631 	pr_alert("BUG: Bad page state in process %s  pfn:%05lx\n",
632 		current->comm, page_to_pfn(page));
633 	__dump_page(page, reason);
634 	dump_page_owner(page);
635 
636 	print_modules();
637 	dump_stack();
638 out:
639 	/* Leave bad fields for debug, except PageBuddy could make trouble */
640 	page_mapcount_reset(page); /* remove PageBuddy */
641 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
642 }
643 
644 /*
645  * Higher-order pages are called "compound pages".  They are structured thusly:
646  *
647  * The first PAGE_SIZE page is called the "head page" and have PG_head set.
648  *
649  * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
650  * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
651  *
652  * The first tail page's ->compound_dtor holds the offset in array of compound
653  * page destructors. See compound_page_dtors.
654  *
655  * The first tail page's ->compound_order holds the order of allocation.
656  * This usage means that zero-order pages may not be compound.
657  */
658 
659 void free_compound_page(struct page *page)
660 {
661 	mem_cgroup_uncharge(page);
662 	__free_pages_ok(page, compound_order(page));
663 }
664 
665 void prep_compound_page(struct page *page, unsigned int order)
666 {
667 	int i;
668 	int nr_pages = 1 << order;
669 
670 	__SetPageHead(page);
671 	for (i = 1; i < nr_pages; i++) {
672 		struct page *p = page + i;
673 		set_page_count(p, 0);
674 		p->mapping = TAIL_MAPPING;
675 		set_compound_head(p, page);
676 	}
677 
678 	set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
679 	set_compound_order(page, order);
680 	atomic_set(compound_mapcount_ptr(page), -1);
681 	if (hpage_pincount_available(page))
682 		atomic_set(compound_pincount_ptr(page), 0);
683 }
684 
685 #ifdef CONFIG_DEBUG_PAGEALLOC
686 unsigned int _debug_guardpage_minorder;
687 
688 bool _debug_pagealloc_enabled_early __read_mostly
689 			= IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
690 EXPORT_SYMBOL(_debug_pagealloc_enabled_early);
691 DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled);
692 EXPORT_SYMBOL(_debug_pagealloc_enabled);
693 
694 DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled);
695 
696 static int __init early_debug_pagealloc(char *buf)
697 {
698 	return kstrtobool(buf, &_debug_pagealloc_enabled_early);
699 }
700 early_param("debug_pagealloc", early_debug_pagealloc);
701 
702 void init_debug_pagealloc(void)
703 {
704 	if (!debug_pagealloc_enabled())
705 		return;
706 
707 	static_branch_enable(&_debug_pagealloc_enabled);
708 
709 	if (!debug_guardpage_minorder())
710 		return;
711 
712 	static_branch_enable(&_debug_guardpage_enabled);
713 }
714 
715 static int __init debug_guardpage_minorder_setup(char *buf)
716 {
717 	unsigned long res;
718 
719 	if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
720 		pr_err("Bad debug_guardpage_minorder value\n");
721 		return 0;
722 	}
723 	_debug_guardpage_minorder = res;
724 	pr_info("Setting debug_guardpage_minorder to %lu\n", res);
725 	return 0;
726 }
727 early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
728 
729 static inline bool set_page_guard(struct zone *zone, struct page *page,
730 				unsigned int order, int migratetype)
731 {
732 	if (!debug_guardpage_enabled())
733 		return false;
734 
735 	if (order >= debug_guardpage_minorder())
736 		return false;
737 
738 	__SetPageGuard(page);
739 	INIT_LIST_HEAD(&page->lru);
740 	set_page_private(page, order);
741 	/* Guard pages are not available for any usage */
742 	__mod_zone_freepage_state(zone, -(1 << order), migratetype);
743 
744 	return true;
745 }
746 
747 static inline void clear_page_guard(struct zone *zone, struct page *page,
748 				unsigned int order, int migratetype)
749 {
750 	if (!debug_guardpage_enabled())
751 		return;
752 
753 	__ClearPageGuard(page);
754 
755 	set_page_private(page, 0);
756 	if (!is_migrate_isolate(migratetype))
757 		__mod_zone_freepage_state(zone, (1 << order), migratetype);
758 }
759 #else
760 static inline bool set_page_guard(struct zone *zone, struct page *page,
761 			unsigned int order, int migratetype) { return false; }
762 static inline void clear_page_guard(struct zone *zone, struct page *page,
763 				unsigned int order, int migratetype) {}
764 #endif
765 
766 static inline void set_page_order(struct page *page, unsigned int order)
767 {
768 	set_page_private(page, order);
769 	__SetPageBuddy(page);
770 }
771 
772 /*
773  * This function checks whether a page is free && is the buddy
774  * we can coalesce a page and its buddy if
775  * (a) the buddy is not in a hole (check before calling!) &&
776  * (b) the buddy is in the buddy system &&
777  * (c) a page and its buddy have the same order &&
778  * (d) a page and its buddy are in the same zone.
779  *
780  * For recording whether a page is in the buddy system, we set PageBuddy.
781  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
782  *
783  * For recording page's order, we use page_private(page).
784  */
785 static inline bool page_is_buddy(struct page *page, struct page *buddy,
786 							unsigned int order)
787 {
788 	if (!page_is_guard(buddy) && !PageBuddy(buddy))
789 		return false;
790 
791 	if (page_order(buddy) != order)
792 		return false;
793 
794 	/*
795 	 * zone check is done late to avoid uselessly calculating
796 	 * zone/node ids for pages that could never merge.
797 	 */
798 	if (page_zone_id(page) != page_zone_id(buddy))
799 		return false;
800 
801 	VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
802 
803 	return true;
804 }
805 
806 #ifdef CONFIG_COMPACTION
807 static inline struct capture_control *task_capc(struct zone *zone)
808 {
809 	struct capture_control *capc = current->capture_control;
810 
811 	return unlikely(capc) &&
812 		!(current->flags & PF_KTHREAD) &&
813 		!capc->page &&
814 		capc->cc->zone == zone ? capc : NULL;
815 }
816 
817 static inline bool
818 compaction_capture(struct capture_control *capc, struct page *page,
819 		   int order, int migratetype)
820 {
821 	if (!capc || order != capc->cc->order)
822 		return false;
823 
824 	/* Do not accidentally pollute CMA or isolated regions*/
825 	if (is_migrate_cma(migratetype) ||
826 	    is_migrate_isolate(migratetype))
827 		return false;
828 
829 	/*
830 	 * Do not let lower order allocations polluate a movable pageblock.
831 	 * This might let an unmovable request use a reclaimable pageblock
832 	 * and vice-versa but no more than normal fallback logic which can
833 	 * have trouble finding a high-order free page.
834 	 */
835 	if (order < pageblock_order && migratetype == MIGRATE_MOVABLE)
836 		return false;
837 
838 	capc->page = page;
839 	return true;
840 }
841 
842 #else
843 static inline struct capture_control *task_capc(struct zone *zone)
844 {
845 	return NULL;
846 }
847 
848 static inline bool
849 compaction_capture(struct capture_control *capc, struct page *page,
850 		   int order, int migratetype)
851 {
852 	return false;
853 }
854 #endif /* CONFIG_COMPACTION */
855 
856 /* Used for pages not on another list */
857 static inline void add_to_free_list(struct page *page, struct zone *zone,
858 				    unsigned int order, int migratetype)
859 {
860 	struct free_area *area = &zone->free_area[order];
861 
862 	list_add(&page->lru, &area->free_list[migratetype]);
863 	area->nr_free++;
864 }
865 
866 /* Used for pages not on another list */
867 static inline void add_to_free_list_tail(struct page *page, struct zone *zone,
868 					 unsigned int order, int migratetype)
869 {
870 	struct free_area *area = &zone->free_area[order];
871 
872 	list_add_tail(&page->lru, &area->free_list[migratetype]);
873 	area->nr_free++;
874 }
875 
876 /* Used for pages which are on another list */
877 static inline void move_to_free_list(struct page *page, struct zone *zone,
878 				     unsigned int order, int migratetype)
879 {
880 	struct free_area *area = &zone->free_area[order];
881 
882 	list_move(&page->lru, &area->free_list[migratetype]);
883 }
884 
885 static inline void del_page_from_free_list(struct page *page, struct zone *zone,
886 					   unsigned int order)
887 {
888 	/* clear reported state and update reported page count */
889 	if (page_reported(page))
890 		__ClearPageReported(page);
891 
892 	list_del(&page->lru);
893 	__ClearPageBuddy(page);
894 	set_page_private(page, 0);
895 	zone->free_area[order].nr_free--;
896 }
897 
898 /*
899  * If this is not the largest possible page, check if the buddy
900  * of the next-highest order is free. If it is, it's possible
901  * that pages are being freed that will coalesce soon. In case,
902  * that is happening, add the free page to the tail of the list
903  * so it's less likely to be used soon and more likely to be merged
904  * as a higher order page
905  */
906 static inline bool
907 buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn,
908 		   struct page *page, unsigned int order)
909 {
910 	struct page *higher_page, *higher_buddy;
911 	unsigned long combined_pfn;
912 
913 	if (order >= MAX_ORDER - 2)
914 		return false;
915 
916 	if (!pfn_valid_within(buddy_pfn))
917 		return false;
918 
919 	combined_pfn = buddy_pfn & pfn;
920 	higher_page = page + (combined_pfn - pfn);
921 	buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
922 	higher_buddy = higher_page + (buddy_pfn - combined_pfn);
923 
924 	return pfn_valid_within(buddy_pfn) &&
925 	       page_is_buddy(higher_page, higher_buddy, order + 1);
926 }
927 
928 /*
929  * Freeing function for a buddy system allocator.
930  *
931  * The concept of a buddy system is to maintain direct-mapped table
932  * (containing bit values) for memory blocks of various "orders".
933  * The bottom level table contains the map for the smallest allocatable
934  * units of memory (here, pages), and each level above it describes
935  * pairs of units from the levels below, hence, "buddies".
936  * At a high level, all that happens here is marking the table entry
937  * at the bottom level available, and propagating the changes upward
938  * as necessary, plus some accounting needed to play nicely with other
939  * parts of the VM system.
940  * At each level, we keep a list of pages, which are heads of continuous
941  * free pages of length of (1 << order) and marked with PageBuddy.
942  * Page's order is recorded in page_private(page) field.
943  * So when we are allocating or freeing one, we can derive the state of the
944  * other.  That is, if we allocate a small block, and both were
945  * free, the remainder of the region must be split into blocks.
946  * If a block is freed, and its buddy is also free, then this
947  * triggers coalescing into a block of larger size.
948  *
949  * -- nyc
950  */
951 
952 static inline void __free_one_page(struct page *page,
953 		unsigned long pfn,
954 		struct zone *zone, unsigned int order,
955 		int migratetype, bool report)
956 {
957 	struct capture_control *capc = task_capc(zone);
958 	unsigned long buddy_pfn;
959 	unsigned long combined_pfn;
960 	unsigned int max_order;
961 	struct page *buddy;
962 	bool to_tail;
963 
964 	max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
965 
966 	VM_BUG_ON(!zone_is_initialized(zone));
967 	VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
968 
969 	VM_BUG_ON(migratetype == -1);
970 	if (likely(!is_migrate_isolate(migratetype)))
971 		__mod_zone_freepage_state(zone, 1 << order, migratetype);
972 
973 	VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
974 	VM_BUG_ON_PAGE(bad_range(zone, page), page);
975 
976 continue_merging:
977 	while (order < max_order - 1) {
978 		if (compaction_capture(capc, page, order, migratetype)) {
979 			__mod_zone_freepage_state(zone, -(1 << order),
980 								migratetype);
981 			return;
982 		}
983 		buddy_pfn = __find_buddy_pfn(pfn, order);
984 		buddy = page + (buddy_pfn - pfn);
985 
986 		if (!pfn_valid_within(buddy_pfn))
987 			goto done_merging;
988 		if (!page_is_buddy(page, buddy, order))
989 			goto done_merging;
990 		/*
991 		 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
992 		 * merge with it and move up one order.
993 		 */
994 		if (page_is_guard(buddy))
995 			clear_page_guard(zone, buddy, order, migratetype);
996 		else
997 			del_page_from_free_list(buddy, zone, order);
998 		combined_pfn = buddy_pfn & pfn;
999 		page = page + (combined_pfn - pfn);
1000 		pfn = combined_pfn;
1001 		order++;
1002 	}
1003 	if (max_order < MAX_ORDER) {
1004 		/* If we are here, it means order is >= pageblock_order.
1005 		 * We want to prevent merge between freepages on isolate
1006 		 * pageblock and normal pageblock. Without this, pageblock
1007 		 * isolation could cause incorrect freepage or CMA accounting.
1008 		 *
1009 		 * We don't want to hit this code for the more frequent
1010 		 * low-order merging.
1011 		 */
1012 		if (unlikely(has_isolate_pageblock(zone))) {
1013 			int buddy_mt;
1014 
1015 			buddy_pfn = __find_buddy_pfn(pfn, order);
1016 			buddy = page + (buddy_pfn - pfn);
1017 			buddy_mt = get_pageblock_migratetype(buddy);
1018 
1019 			if (migratetype != buddy_mt
1020 					&& (is_migrate_isolate(migratetype) ||
1021 						is_migrate_isolate(buddy_mt)))
1022 				goto done_merging;
1023 		}
1024 		max_order++;
1025 		goto continue_merging;
1026 	}
1027 
1028 done_merging:
1029 	set_page_order(page, order);
1030 
1031 	if (is_shuffle_order(order))
1032 		to_tail = shuffle_pick_tail();
1033 	else
1034 		to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order);
1035 
1036 	if (to_tail)
1037 		add_to_free_list_tail(page, zone, order, migratetype);
1038 	else
1039 		add_to_free_list(page, zone, order, migratetype);
1040 
1041 	/* Notify page reporting subsystem of freed page */
1042 	if (report)
1043 		page_reporting_notify_free(order);
1044 }
1045 
1046 /*
1047  * A bad page could be due to a number of fields. Instead of multiple branches,
1048  * try and check multiple fields with one check. The caller must do a detailed
1049  * check if necessary.
1050  */
1051 static inline bool page_expected_state(struct page *page,
1052 					unsigned long check_flags)
1053 {
1054 	if (unlikely(atomic_read(&page->_mapcount) != -1))
1055 		return false;
1056 
1057 	if (unlikely((unsigned long)page->mapping |
1058 			page_ref_count(page) |
1059 #ifdef CONFIG_MEMCG
1060 			(unsigned long)page->mem_cgroup |
1061 #endif
1062 			(page->flags & check_flags)))
1063 		return false;
1064 
1065 	return true;
1066 }
1067 
1068 static const char *page_bad_reason(struct page *page, unsigned long flags)
1069 {
1070 	const char *bad_reason = NULL;
1071 
1072 	if (unlikely(atomic_read(&page->_mapcount) != -1))
1073 		bad_reason = "nonzero mapcount";
1074 	if (unlikely(page->mapping != NULL))
1075 		bad_reason = "non-NULL mapping";
1076 	if (unlikely(page_ref_count(page) != 0))
1077 		bad_reason = "nonzero _refcount";
1078 	if (unlikely(page->flags & flags)) {
1079 		if (flags == PAGE_FLAGS_CHECK_AT_PREP)
1080 			bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set";
1081 		else
1082 			bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
1083 	}
1084 #ifdef CONFIG_MEMCG
1085 	if (unlikely(page->mem_cgroup))
1086 		bad_reason = "page still charged to cgroup";
1087 #endif
1088 	return bad_reason;
1089 }
1090 
1091 static void check_free_page_bad(struct page *page)
1092 {
1093 	bad_page(page,
1094 		 page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE));
1095 }
1096 
1097 static inline int check_free_page(struct page *page)
1098 {
1099 	if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
1100 		return 0;
1101 
1102 	/* Something has gone sideways, find it */
1103 	check_free_page_bad(page);
1104 	return 1;
1105 }
1106 
1107 static int free_tail_pages_check(struct page *head_page, struct page *page)
1108 {
1109 	int ret = 1;
1110 
1111 	/*
1112 	 * We rely page->lru.next never has bit 0 set, unless the page
1113 	 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
1114 	 */
1115 	BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
1116 
1117 	if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
1118 		ret = 0;
1119 		goto out;
1120 	}
1121 	switch (page - head_page) {
1122 	case 1:
1123 		/* the first tail page: ->mapping may be compound_mapcount() */
1124 		if (unlikely(compound_mapcount(page))) {
1125 			bad_page(page, "nonzero compound_mapcount");
1126 			goto out;
1127 		}
1128 		break;
1129 	case 2:
1130 		/*
1131 		 * the second tail page: ->mapping is
1132 		 * deferred_list.next -- ignore value.
1133 		 */
1134 		break;
1135 	default:
1136 		if (page->mapping != TAIL_MAPPING) {
1137 			bad_page(page, "corrupted mapping in tail page");
1138 			goto out;
1139 		}
1140 		break;
1141 	}
1142 	if (unlikely(!PageTail(page))) {
1143 		bad_page(page, "PageTail not set");
1144 		goto out;
1145 	}
1146 	if (unlikely(compound_head(page) != head_page)) {
1147 		bad_page(page, "compound_head not consistent");
1148 		goto out;
1149 	}
1150 	ret = 0;
1151 out:
1152 	page->mapping = NULL;
1153 	clear_compound_head(page);
1154 	return ret;
1155 }
1156 
1157 static void kernel_init_free_pages(struct page *page, int numpages)
1158 {
1159 	int i;
1160 
1161 	/* s390's use of memset() could override KASAN redzones. */
1162 	kasan_disable_current();
1163 	for (i = 0; i < numpages; i++)
1164 		clear_highpage(page + i);
1165 	kasan_enable_current();
1166 }
1167 
1168 static __always_inline bool free_pages_prepare(struct page *page,
1169 					unsigned int order, bool check_free)
1170 {
1171 	int bad = 0;
1172 
1173 	VM_BUG_ON_PAGE(PageTail(page), page);
1174 
1175 	trace_mm_page_free(page, order);
1176 
1177 	/*
1178 	 * Check tail pages before head page information is cleared to
1179 	 * avoid checking PageCompound for order-0 pages.
1180 	 */
1181 	if (unlikely(order)) {
1182 		bool compound = PageCompound(page);
1183 		int i;
1184 
1185 		VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1186 
1187 		if (compound)
1188 			ClearPageDoubleMap(page);
1189 		for (i = 1; i < (1 << order); i++) {
1190 			if (compound)
1191 				bad += free_tail_pages_check(page, page + i);
1192 			if (unlikely(check_free_page(page + i))) {
1193 				bad++;
1194 				continue;
1195 			}
1196 			(page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1197 		}
1198 	}
1199 	if (PageMappingFlags(page))
1200 		page->mapping = NULL;
1201 	if (memcg_kmem_enabled() && PageKmemcg(page))
1202 		__memcg_kmem_uncharge_page(page, order);
1203 	if (check_free)
1204 		bad += check_free_page(page);
1205 	if (bad)
1206 		return false;
1207 
1208 	page_cpupid_reset_last(page);
1209 	page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1210 	reset_page_owner(page, order);
1211 
1212 	if (!PageHighMem(page)) {
1213 		debug_check_no_locks_freed(page_address(page),
1214 					   PAGE_SIZE << order);
1215 		debug_check_no_obj_freed(page_address(page),
1216 					   PAGE_SIZE << order);
1217 	}
1218 	if (want_init_on_free())
1219 		kernel_init_free_pages(page, 1 << order);
1220 
1221 	kernel_poison_pages(page, 1 << order, 0);
1222 	/*
1223 	 * arch_free_page() can make the page's contents inaccessible.  s390
1224 	 * does this.  So nothing which can access the page's contents should
1225 	 * happen after this.
1226 	 */
1227 	arch_free_page(page, order);
1228 
1229 	if (debug_pagealloc_enabled_static())
1230 		kernel_map_pages(page, 1 << order, 0);
1231 
1232 	kasan_free_nondeferred_pages(page, order);
1233 
1234 	return true;
1235 }
1236 
1237 #ifdef CONFIG_DEBUG_VM
1238 /*
1239  * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed
1240  * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when
1241  * moved from pcp lists to free lists.
1242  */
1243 static bool free_pcp_prepare(struct page *page)
1244 {
1245 	return free_pages_prepare(page, 0, true);
1246 }
1247 
1248 static bool bulkfree_pcp_prepare(struct page *page)
1249 {
1250 	if (debug_pagealloc_enabled_static())
1251 		return check_free_page(page);
1252 	else
1253 		return false;
1254 }
1255 #else
1256 /*
1257  * With DEBUG_VM disabled, order-0 pages being freed are checked only when
1258  * moving from pcp lists to free list in order to reduce overhead. With
1259  * debug_pagealloc enabled, they are checked also immediately when being freed
1260  * to the pcp lists.
1261  */
1262 static bool free_pcp_prepare(struct page *page)
1263 {
1264 	if (debug_pagealloc_enabled_static())
1265 		return free_pages_prepare(page, 0, true);
1266 	else
1267 		return free_pages_prepare(page, 0, false);
1268 }
1269 
1270 static bool bulkfree_pcp_prepare(struct page *page)
1271 {
1272 	return check_free_page(page);
1273 }
1274 #endif /* CONFIG_DEBUG_VM */
1275 
1276 static inline void prefetch_buddy(struct page *page)
1277 {
1278 	unsigned long pfn = page_to_pfn(page);
1279 	unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0);
1280 	struct page *buddy = page + (buddy_pfn - pfn);
1281 
1282 	prefetch(buddy);
1283 }
1284 
1285 /*
1286  * Frees a number of pages from the PCP lists
1287  * Assumes all pages on list are in same zone, and of same order.
1288  * count is the number of pages to free.
1289  *
1290  * If the zone was previously in an "all pages pinned" state then look to
1291  * see if this freeing clears that state.
1292  *
1293  * And clear the zone's pages_scanned counter, to hold off the "all pages are
1294  * pinned" detection logic.
1295  */
1296 static void free_pcppages_bulk(struct zone *zone, int count,
1297 					struct per_cpu_pages *pcp)
1298 {
1299 	int migratetype = 0;
1300 	int batch_free = 0;
1301 	int prefetch_nr = 0;
1302 	bool isolated_pageblocks;
1303 	struct page *page, *tmp;
1304 	LIST_HEAD(head);
1305 
1306 	/*
1307 	 * Ensure proper count is passed which otherwise would stuck in the
1308 	 * below while (list_empty(list)) loop.
1309 	 */
1310 	count = min(pcp->count, count);
1311 	while (count) {
1312 		struct list_head *list;
1313 
1314 		/*
1315 		 * Remove pages from lists in a round-robin fashion. A
1316 		 * batch_free count is maintained that is incremented when an
1317 		 * empty list is encountered.  This is so more pages are freed
1318 		 * off fuller lists instead of spinning excessively around empty
1319 		 * lists
1320 		 */
1321 		do {
1322 			batch_free++;
1323 			if (++migratetype == MIGRATE_PCPTYPES)
1324 				migratetype = 0;
1325 			list = &pcp->lists[migratetype];
1326 		} while (list_empty(list));
1327 
1328 		/* This is the only non-empty list. Free them all. */
1329 		if (batch_free == MIGRATE_PCPTYPES)
1330 			batch_free = count;
1331 
1332 		do {
1333 			page = list_last_entry(list, struct page, lru);
1334 			/* must delete to avoid corrupting pcp list */
1335 			list_del(&page->lru);
1336 			pcp->count--;
1337 
1338 			if (bulkfree_pcp_prepare(page))
1339 				continue;
1340 
1341 			list_add_tail(&page->lru, &head);
1342 
1343 			/*
1344 			 * We are going to put the page back to the global
1345 			 * pool, prefetch its buddy to speed up later access
1346 			 * under zone->lock. It is believed the overhead of
1347 			 * an additional test and calculating buddy_pfn here
1348 			 * can be offset by reduced memory latency later. To
1349 			 * avoid excessive prefetching due to large count, only
1350 			 * prefetch buddy for the first pcp->batch nr of pages.
1351 			 */
1352 			if (prefetch_nr++ < pcp->batch)
1353 				prefetch_buddy(page);
1354 		} while (--count && --batch_free && !list_empty(list));
1355 	}
1356 
1357 	spin_lock(&zone->lock);
1358 	isolated_pageblocks = has_isolate_pageblock(zone);
1359 
1360 	/*
1361 	 * Use safe version since after __free_one_page(),
1362 	 * page->lru.next will not point to original list.
1363 	 */
1364 	list_for_each_entry_safe(page, tmp, &head, lru) {
1365 		int mt = get_pcppage_migratetype(page);
1366 		/* MIGRATE_ISOLATE page should not go to pcplists */
1367 		VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1368 		/* Pageblock could have been isolated meanwhile */
1369 		if (unlikely(isolated_pageblocks))
1370 			mt = get_pageblock_migratetype(page);
1371 
1372 		__free_one_page(page, page_to_pfn(page), zone, 0, mt, true);
1373 		trace_mm_page_pcpu_drain(page, 0, mt);
1374 	}
1375 	spin_unlock(&zone->lock);
1376 }
1377 
1378 static void free_one_page(struct zone *zone,
1379 				struct page *page, unsigned long pfn,
1380 				unsigned int order,
1381 				int migratetype)
1382 {
1383 	spin_lock(&zone->lock);
1384 	if (unlikely(has_isolate_pageblock(zone) ||
1385 		is_migrate_isolate(migratetype))) {
1386 		migratetype = get_pfnblock_migratetype(page, pfn);
1387 	}
1388 	__free_one_page(page, pfn, zone, order, migratetype, true);
1389 	spin_unlock(&zone->lock);
1390 }
1391 
1392 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1393 				unsigned long zone, int nid)
1394 {
1395 	mm_zero_struct_page(page);
1396 	set_page_links(page, zone, nid, pfn);
1397 	init_page_count(page);
1398 	page_mapcount_reset(page);
1399 	page_cpupid_reset_last(page);
1400 	page_kasan_tag_reset(page);
1401 
1402 	INIT_LIST_HEAD(&page->lru);
1403 #ifdef WANT_PAGE_VIRTUAL
1404 	/* The shift won't overflow because ZONE_NORMAL is below 4G. */
1405 	if (!is_highmem_idx(zone))
1406 		set_page_address(page, __va(pfn << PAGE_SHIFT));
1407 #endif
1408 }
1409 
1410 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1411 static void __meminit init_reserved_page(unsigned long pfn)
1412 {
1413 	pg_data_t *pgdat;
1414 	int nid, zid;
1415 
1416 	if (!early_page_uninitialised(pfn))
1417 		return;
1418 
1419 	nid = early_pfn_to_nid(pfn);
1420 	pgdat = NODE_DATA(nid);
1421 
1422 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1423 		struct zone *zone = &pgdat->node_zones[zid];
1424 
1425 		if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1426 			break;
1427 	}
1428 	__init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1429 }
1430 #else
1431 static inline void init_reserved_page(unsigned long pfn)
1432 {
1433 }
1434 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1435 
1436 /*
1437  * Initialised pages do not have PageReserved set. This function is
1438  * called for each range allocated by the bootmem allocator and
1439  * marks the pages PageReserved. The remaining valid pages are later
1440  * sent to the buddy page allocator.
1441  */
1442 void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1443 {
1444 	unsigned long start_pfn = PFN_DOWN(start);
1445 	unsigned long end_pfn = PFN_UP(end);
1446 
1447 	for (; start_pfn < end_pfn; start_pfn++) {
1448 		if (pfn_valid(start_pfn)) {
1449 			struct page *page = pfn_to_page(start_pfn);
1450 
1451 			init_reserved_page(start_pfn);
1452 
1453 			/* Avoid false-positive PageTail() */
1454 			INIT_LIST_HEAD(&page->lru);
1455 
1456 			/*
1457 			 * no need for atomic set_bit because the struct
1458 			 * page is not visible yet so nobody should
1459 			 * access it yet.
1460 			 */
1461 			__SetPageReserved(page);
1462 		}
1463 	}
1464 }
1465 
1466 static void __free_pages_ok(struct page *page, unsigned int order)
1467 {
1468 	unsigned long flags;
1469 	int migratetype;
1470 	unsigned long pfn = page_to_pfn(page);
1471 
1472 	if (!free_pages_prepare(page, order, true))
1473 		return;
1474 
1475 	migratetype = get_pfnblock_migratetype(page, pfn);
1476 	local_irq_save(flags);
1477 	__count_vm_events(PGFREE, 1 << order);
1478 	free_one_page(page_zone(page), page, pfn, order, migratetype);
1479 	local_irq_restore(flags);
1480 }
1481 
1482 void __free_pages_core(struct page *page, unsigned int order)
1483 {
1484 	unsigned int nr_pages = 1 << order;
1485 	struct page *p = page;
1486 	unsigned int loop;
1487 
1488 	prefetchw(p);
1489 	for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1490 		prefetchw(p + 1);
1491 		__ClearPageReserved(p);
1492 		set_page_count(p, 0);
1493 	}
1494 	__ClearPageReserved(p);
1495 	set_page_count(p, 0);
1496 
1497 	atomic_long_add(nr_pages, &page_zone(page)->managed_pages);
1498 	set_page_refcounted(page);
1499 	__free_pages(page, order);
1500 }
1501 
1502 #ifdef CONFIG_NEED_MULTIPLE_NODES
1503 
1504 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1505 
1506 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
1507 
1508 /*
1509  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
1510  */
1511 int __meminit __early_pfn_to_nid(unsigned long pfn,
1512 					struct mminit_pfnnid_cache *state)
1513 {
1514 	unsigned long start_pfn, end_pfn;
1515 	int nid;
1516 
1517 	if (state->last_start <= pfn && pfn < state->last_end)
1518 		return state->last_nid;
1519 
1520 	nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
1521 	if (nid != NUMA_NO_NODE) {
1522 		state->last_start = start_pfn;
1523 		state->last_end = end_pfn;
1524 		state->last_nid = nid;
1525 	}
1526 
1527 	return nid;
1528 }
1529 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
1530 
1531 int __meminit early_pfn_to_nid(unsigned long pfn)
1532 {
1533 	static DEFINE_SPINLOCK(early_pfn_lock);
1534 	int nid;
1535 
1536 	spin_lock(&early_pfn_lock);
1537 	nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1538 	if (nid < 0)
1539 		nid = first_online_node;
1540 	spin_unlock(&early_pfn_lock);
1541 
1542 	return nid;
1543 }
1544 #endif /* CONFIG_NEED_MULTIPLE_NODES */
1545 
1546 void __init memblock_free_pages(struct page *page, unsigned long pfn,
1547 							unsigned int order)
1548 {
1549 	if (early_page_uninitialised(pfn))
1550 		return;
1551 	__free_pages_core(page, order);
1552 }
1553 
1554 /*
1555  * Check that the whole (or subset of) a pageblock given by the interval of
1556  * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1557  * with the migration of free compaction scanner. The scanners then need to
1558  * use only pfn_valid_within() check for arches that allow holes within
1559  * pageblocks.
1560  *
1561  * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1562  *
1563  * It's possible on some configurations to have a setup like node0 node1 node0
1564  * i.e. it's possible that all pages within a zones range of pages do not
1565  * belong to a single zone. We assume that a border between node0 and node1
1566  * can occur within a single pageblock, but not a node0 node1 node0
1567  * interleaving within a single pageblock. It is therefore sufficient to check
1568  * the first and last page of a pageblock and avoid checking each individual
1569  * page in a pageblock.
1570  */
1571 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1572 				     unsigned long end_pfn, struct zone *zone)
1573 {
1574 	struct page *start_page;
1575 	struct page *end_page;
1576 
1577 	/* end_pfn is one past the range we are checking */
1578 	end_pfn--;
1579 
1580 	if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1581 		return NULL;
1582 
1583 	start_page = pfn_to_online_page(start_pfn);
1584 	if (!start_page)
1585 		return NULL;
1586 
1587 	if (page_zone(start_page) != zone)
1588 		return NULL;
1589 
1590 	end_page = pfn_to_page(end_pfn);
1591 
1592 	/* This gives a shorter code than deriving page_zone(end_page) */
1593 	if (page_zone_id(start_page) != page_zone_id(end_page))
1594 		return NULL;
1595 
1596 	return start_page;
1597 }
1598 
1599 void set_zone_contiguous(struct zone *zone)
1600 {
1601 	unsigned long block_start_pfn = zone->zone_start_pfn;
1602 	unsigned long block_end_pfn;
1603 
1604 	block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1605 	for (; block_start_pfn < zone_end_pfn(zone);
1606 			block_start_pfn = block_end_pfn,
1607 			 block_end_pfn += pageblock_nr_pages) {
1608 
1609 		block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1610 
1611 		if (!__pageblock_pfn_to_page(block_start_pfn,
1612 					     block_end_pfn, zone))
1613 			return;
1614 		cond_resched();
1615 	}
1616 
1617 	/* We confirm that there is no hole */
1618 	zone->contiguous = true;
1619 }
1620 
1621 void clear_zone_contiguous(struct zone *zone)
1622 {
1623 	zone->contiguous = false;
1624 }
1625 
1626 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1627 static void __init deferred_free_range(unsigned long pfn,
1628 				       unsigned long nr_pages)
1629 {
1630 	struct page *page;
1631 	unsigned long i;
1632 
1633 	if (!nr_pages)
1634 		return;
1635 
1636 	page = pfn_to_page(pfn);
1637 
1638 	/* Free a large naturally-aligned chunk if possible */
1639 	if (nr_pages == pageblock_nr_pages &&
1640 	    (pfn & (pageblock_nr_pages - 1)) == 0) {
1641 		set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1642 		__free_pages_core(page, pageblock_order);
1643 		return;
1644 	}
1645 
1646 	for (i = 0; i < nr_pages; i++, page++, pfn++) {
1647 		if ((pfn & (pageblock_nr_pages - 1)) == 0)
1648 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1649 		__free_pages_core(page, 0);
1650 	}
1651 }
1652 
1653 /* Completion tracking for deferred_init_memmap() threads */
1654 static atomic_t pgdat_init_n_undone __initdata;
1655 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1656 
1657 static inline void __init pgdat_init_report_one_done(void)
1658 {
1659 	if (atomic_dec_and_test(&pgdat_init_n_undone))
1660 		complete(&pgdat_init_all_done_comp);
1661 }
1662 
1663 /*
1664  * Returns true if page needs to be initialized or freed to buddy allocator.
1665  *
1666  * First we check if pfn is valid on architectures where it is possible to have
1667  * holes within pageblock_nr_pages. On systems where it is not possible, this
1668  * function is optimized out.
1669  *
1670  * Then, we check if a current large page is valid by only checking the validity
1671  * of the head pfn.
1672  */
1673 static inline bool __init deferred_pfn_valid(unsigned long pfn)
1674 {
1675 	if (!pfn_valid_within(pfn))
1676 		return false;
1677 	if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1678 		return false;
1679 	return true;
1680 }
1681 
1682 /*
1683  * Free pages to buddy allocator. Try to free aligned pages in
1684  * pageblock_nr_pages sizes.
1685  */
1686 static void __init deferred_free_pages(unsigned long pfn,
1687 				       unsigned long end_pfn)
1688 {
1689 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1690 	unsigned long nr_free = 0;
1691 
1692 	for (; pfn < end_pfn; pfn++) {
1693 		if (!deferred_pfn_valid(pfn)) {
1694 			deferred_free_range(pfn - nr_free, nr_free);
1695 			nr_free = 0;
1696 		} else if (!(pfn & nr_pgmask)) {
1697 			deferred_free_range(pfn - nr_free, nr_free);
1698 			nr_free = 1;
1699 		} else {
1700 			nr_free++;
1701 		}
1702 	}
1703 	/* Free the last block of pages to allocator */
1704 	deferred_free_range(pfn - nr_free, nr_free);
1705 }
1706 
1707 /*
1708  * Initialize struct pages.  We minimize pfn page lookups and scheduler checks
1709  * by performing it only once every pageblock_nr_pages.
1710  * Return number of pages initialized.
1711  */
1712 static unsigned long  __init deferred_init_pages(struct zone *zone,
1713 						 unsigned long pfn,
1714 						 unsigned long end_pfn)
1715 {
1716 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1717 	int nid = zone_to_nid(zone);
1718 	unsigned long nr_pages = 0;
1719 	int zid = zone_idx(zone);
1720 	struct page *page = NULL;
1721 
1722 	for (; pfn < end_pfn; pfn++) {
1723 		if (!deferred_pfn_valid(pfn)) {
1724 			page = NULL;
1725 			continue;
1726 		} else if (!page || !(pfn & nr_pgmask)) {
1727 			page = pfn_to_page(pfn);
1728 		} else {
1729 			page++;
1730 		}
1731 		__init_single_page(page, pfn, zid, nid);
1732 		nr_pages++;
1733 	}
1734 	return (nr_pages);
1735 }
1736 
1737 /*
1738  * This function is meant to pre-load the iterator for the zone init.
1739  * Specifically it walks through the ranges until we are caught up to the
1740  * first_init_pfn value and exits there. If we never encounter the value we
1741  * return false indicating there are no valid ranges left.
1742  */
1743 static bool __init
1744 deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone,
1745 				    unsigned long *spfn, unsigned long *epfn,
1746 				    unsigned long first_init_pfn)
1747 {
1748 	u64 j;
1749 
1750 	/*
1751 	 * Start out by walking through the ranges in this zone that have
1752 	 * already been initialized. We don't need to do anything with them
1753 	 * so we just need to flush them out of the system.
1754 	 */
1755 	for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) {
1756 		if (*epfn <= first_init_pfn)
1757 			continue;
1758 		if (*spfn < first_init_pfn)
1759 			*spfn = first_init_pfn;
1760 		*i = j;
1761 		return true;
1762 	}
1763 
1764 	return false;
1765 }
1766 
1767 /*
1768  * Initialize and free pages. We do it in two loops: first we initialize
1769  * struct page, then free to buddy allocator, because while we are
1770  * freeing pages we can access pages that are ahead (computing buddy
1771  * page in __free_one_page()).
1772  *
1773  * In order to try and keep some memory in the cache we have the loop
1774  * broken along max page order boundaries. This way we will not cause
1775  * any issues with the buddy page computation.
1776  */
1777 static unsigned long __init
1778 deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
1779 		       unsigned long *end_pfn)
1780 {
1781 	unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
1782 	unsigned long spfn = *start_pfn, epfn = *end_pfn;
1783 	unsigned long nr_pages = 0;
1784 	u64 j = *i;
1785 
1786 	/* First we loop through and initialize the page values */
1787 	for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) {
1788 		unsigned long t;
1789 
1790 		if (mo_pfn <= *start_pfn)
1791 			break;
1792 
1793 		t = min(mo_pfn, *end_pfn);
1794 		nr_pages += deferred_init_pages(zone, *start_pfn, t);
1795 
1796 		if (mo_pfn < *end_pfn) {
1797 			*start_pfn = mo_pfn;
1798 			break;
1799 		}
1800 	}
1801 
1802 	/* Reset values and now loop through freeing pages as needed */
1803 	swap(j, *i);
1804 
1805 	for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) {
1806 		unsigned long t;
1807 
1808 		if (mo_pfn <= spfn)
1809 			break;
1810 
1811 		t = min(mo_pfn, epfn);
1812 		deferred_free_pages(spfn, t);
1813 
1814 		if (mo_pfn <= epfn)
1815 			break;
1816 	}
1817 
1818 	return nr_pages;
1819 }
1820 
1821 static void __init
1822 deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
1823 			   void *arg)
1824 {
1825 	unsigned long spfn, epfn;
1826 	struct zone *zone = arg;
1827 	u64 i;
1828 
1829 	deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn);
1830 
1831 	/*
1832 	 * Initialize and free pages in MAX_ORDER sized increments so that we
1833 	 * can avoid introducing any issues with the buddy allocator.
1834 	 */
1835 	while (spfn < end_pfn) {
1836 		deferred_init_maxorder(&i, zone, &spfn, &epfn);
1837 		cond_resched();
1838 	}
1839 }
1840 
1841 /* An arch may override for more concurrency. */
1842 __weak int __init
1843 deferred_page_init_max_threads(const struct cpumask *node_cpumask)
1844 {
1845 	return 1;
1846 }
1847 
1848 /* Initialise remaining memory on a node */
1849 static int __init deferred_init_memmap(void *data)
1850 {
1851 	pg_data_t *pgdat = data;
1852 	const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1853 	unsigned long spfn = 0, epfn = 0;
1854 	unsigned long first_init_pfn, flags;
1855 	unsigned long start = jiffies;
1856 	struct zone *zone;
1857 	int zid, max_threads;
1858 	u64 i;
1859 
1860 	/* Bind memory initialisation thread to a local node if possible */
1861 	if (!cpumask_empty(cpumask))
1862 		set_cpus_allowed_ptr(current, cpumask);
1863 
1864 	pgdat_resize_lock(pgdat, &flags);
1865 	first_init_pfn = pgdat->first_deferred_pfn;
1866 	if (first_init_pfn == ULONG_MAX) {
1867 		pgdat_resize_unlock(pgdat, &flags);
1868 		pgdat_init_report_one_done();
1869 		return 0;
1870 	}
1871 
1872 	/* Sanity check boundaries */
1873 	BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1874 	BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1875 	pgdat->first_deferred_pfn = ULONG_MAX;
1876 
1877 	/*
1878 	 * Once we unlock here, the zone cannot be grown anymore, thus if an
1879 	 * interrupt thread must allocate this early in boot, zone must be
1880 	 * pre-grown prior to start of deferred page initialization.
1881 	 */
1882 	pgdat_resize_unlock(pgdat, &flags);
1883 
1884 	/* Only the highest zone is deferred so find it */
1885 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1886 		zone = pgdat->node_zones + zid;
1887 		if (first_init_pfn < zone_end_pfn(zone))
1888 			break;
1889 	}
1890 
1891 	/* If the zone is empty somebody else may have cleared out the zone */
1892 	if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
1893 						 first_init_pfn))
1894 		goto zone_empty;
1895 
1896 	max_threads = deferred_page_init_max_threads(cpumask);
1897 
1898 	while (spfn < epfn) {
1899 		unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION);
1900 		struct padata_mt_job job = {
1901 			.thread_fn   = deferred_init_memmap_chunk,
1902 			.fn_arg      = zone,
1903 			.start       = spfn,
1904 			.size        = epfn_align - spfn,
1905 			.align       = PAGES_PER_SECTION,
1906 			.min_chunk   = PAGES_PER_SECTION,
1907 			.max_threads = max_threads,
1908 		};
1909 
1910 		padata_do_multithreaded(&job);
1911 		deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
1912 						    epfn_align);
1913 	}
1914 zone_empty:
1915 	/* Sanity check that the next zone really is unpopulated */
1916 	WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1917 
1918 	pr_info("node %d deferred pages initialised in %ums\n",
1919 		pgdat->node_id, jiffies_to_msecs(jiffies - start));
1920 
1921 	pgdat_init_report_one_done();
1922 	return 0;
1923 }
1924 
1925 /*
1926  * If this zone has deferred pages, try to grow it by initializing enough
1927  * deferred pages to satisfy the allocation specified by order, rounded up to
1928  * the nearest PAGES_PER_SECTION boundary.  So we're adding memory in increments
1929  * of SECTION_SIZE bytes by initializing struct pages in increments of
1930  * PAGES_PER_SECTION * sizeof(struct page) bytes.
1931  *
1932  * Return true when zone was grown, otherwise return false. We return true even
1933  * when we grow less than requested, to let the caller decide if there are
1934  * enough pages to satisfy the allocation.
1935  *
1936  * Note: We use noinline because this function is needed only during boot, and
1937  * it is called from a __ref function _deferred_grow_zone. This way we are
1938  * making sure that it is not inlined into permanent text section.
1939  */
1940 static noinline bool __init
1941 deferred_grow_zone(struct zone *zone, unsigned int order)
1942 {
1943 	unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
1944 	pg_data_t *pgdat = zone->zone_pgdat;
1945 	unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
1946 	unsigned long spfn, epfn, flags;
1947 	unsigned long nr_pages = 0;
1948 	u64 i;
1949 
1950 	/* Only the last zone may have deferred pages */
1951 	if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
1952 		return false;
1953 
1954 	pgdat_resize_lock(pgdat, &flags);
1955 
1956 	/*
1957 	 * If someone grew this zone while we were waiting for spinlock, return
1958 	 * true, as there might be enough pages already.
1959 	 */
1960 	if (first_deferred_pfn != pgdat->first_deferred_pfn) {
1961 		pgdat_resize_unlock(pgdat, &flags);
1962 		return true;
1963 	}
1964 
1965 	/* If the zone is empty somebody else may have cleared out the zone */
1966 	if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
1967 						 first_deferred_pfn)) {
1968 		pgdat->first_deferred_pfn = ULONG_MAX;
1969 		pgdat_resize_unlock(pgdat, &flags);
1970 		/* Retry only once. */
1971 		return first_deferred_pfn != ULONG_MAX;
1972 	}
1973 
1974 	/*
1975 	 * Initialize and free pages in MAX_ORDER sized increments so
1976 	 * that we can avoid introducing any issues with the buddy
1977 	 * allocator.
1978 	 */
1979 	while (spfn < epfn) {
1980 		/* update our first deferred PFN for this section */
1981 		first_deferred_pfn = spfn;
1982 
1983 		nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
1984 		touch_nmi_watchdog();
1985 
1986 		/* We should only stop along section boundaries */
1987 		if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
1988 			continue;
1989 
1990 		/* If our quota has been met we can stop here */
1991 		if (nr_pages >= nr_pages_needed)
1992 			break;
1993 	}
1994 
1995 	pgdat->first_deferred_pfn = spfn;
1996 	pgdat_resize_unlock(pgdat, &flags);
1997 
1998 	return nr_pages > 0;
1999 }
2000 
2001 /*
2002  * deferred_grow_zone() is __init, but it is called from
2003  * get_page_from_freelist() during early boot until deferred_pages permanently
2004  * disables this call. This is why we have refdata wrapper to avoid warning,
2005  * and to ensure that the function body gets unloaded.
2006  */
2007 static bool __ref
2008 _deferred_grow_zone(struct zone *zone, unsigned int order)
2009 {
2010 	return deferred_grow_zone(zone, order);
2011 }
2012 
2013 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
2014 
2015 void __init page_alloc_init_late(void)
2016 {
2017 	struct zone *zone;
2018 	int nid;
2019 
2020 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
2021 
2022 	/* There will be num_node_state(N_MEMORY) threads */
2023 	atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
2024 	for_each_node_state(nid, N_MEMORY) {
2025 		kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
2026 	}
2027 
2028 	/* Block until all are initialised */
2029 	wait_for_completion(&pgdat_init_all_done_comp);
2030 
2031 	/*
2032 	 * The number of managed pages has changed due to the initialisation
2033 	 * so the pcpu batch and high limits needs to be updated or the limits
2034 	 * will be artificially small.
2035 	 */
2036 	for_each_populated_zone(zone)
2037 		zone_pcp_update(zone);
2038 
2039 	/*
2040 	 * We initialized the rest of the deferred pages.  Permanently disable
2041 	 * on-demand struct page initialization.
2042 	 */
2043 	static_branch_disable(&deferred_pages);
2044 
2045 	/* Reinit limits that are based on free pages after the kernel is up */
2046 	files_maxfiles_init();
2047 #endif
2048 
2049 	/* Discard memblock private memory */
2050 	memblock_discard();
2051 
2052 	for_each_node_state(nid, N_MEMORY)
2053 		shuffle_free_memory(NODE_DATA(nid));
2054 
2055 	for_each_populated_zone(zone)
2056 		set_zone_contiguous(zone);
2057 }
2058 
2059 #ifdef CONFIG_CMA
2060 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
2061 void __init init_cma_reserved_pageblock(struct page *page)
2062 {
2063 	unsigned i = pageblock_nr_pages;
2064 	struct page *p = page;
2065 
2066 	do {
2067 		__ClearPageReserved(p);
2068 		set_page_count(p, 0);
2069 	} while (++p, --i);
2070 
2071 	set_pageblock_migratetype(page, MIGRATE_CMA);
2072 
2073 	if (pageblock_order >= MAX_ORDER) {
2074 		i = pageblock_nr_pages;
2075 		p = page;
2076 		do {
2077 			set_page_refcounted(p);
2078 			__free_pages(p, MAX_ORDER - 1);
2079 			p += MAX_ORDER_NR_PAGES;
2080 		} while (i -= MAX_ORDER_NR_PAGES);
2081 	} else {
2082 		set_page_refcounted(page);
2083 		__free_pages(page, pageblock_order);
2084 	}
2085 
2086 	adjust_managed_page_count(page, pageblock_nr_pages);
2087 }
2088 #endif
2089 
2090 /*
2091  * The order of subdivision here is critical for the IO subsystem.
2092  * Please do not alter this order without good reasons and regression
2093  * testing. Specifically, as large blocks of memory are subdivided,
2094  * the order in which smaller blocks are delivered depends on the order
2095  * they're subdivided in this function. This is the primary factor
2096  * influencing the order in which pages are delivered to the IO
2097  * subsystem according to empirical testing, and this is also justified
2098  * by considering the behavior of a buddy system containing a single
2099  * large block of memory acted on by a series of small allocations.
2100  * This behavior is a critical factor in sglist merging's success.
2101  *
2102  * -- nyc
2103  */
2104 static inline void expand(struct zone *zone, struct page *page,
2105 	int low, int high, int migratetype)
2106 {
2107 	unsigned long size = 1 << high;
2108 
2109 	while (high > low) {
2110 		high--;
2111 		size >>= 1;
2112 		VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
2113 
2114 		/*
2115 		 * Mark as guard pages (or page), that will allow to
2116 		 * merge back to allocator when buddy will be freed.
2117 		 * Corresponding page table entries will not be touched,
2118 		 * pages will stay not present in virtual address space
2119 		 */
2120 		if (set_page_guard(zone, &page[size], high, migratetype))
2121 			continue;
2122 
2123 		add_to_free_list(&page[size], zone, high, migratetype);
2124 		set_page_order(&page[size], high);
2125 	}
2126 }
2127 
2128 static void check_new_page_bad(struct page *page)
2129 {
2130 	if (unlikely(page->flags & __PG_HWPOISON)) {
2131 		/* Don't complain about hwpoisoned pages */
2132 		page_mapcount_reset(page); /* remove PageBuddy */
2133 		return;
2134 	}
2135 
2136 	bad_page(page,
2137 		 page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
2138 }
2139 
2140 /*
2141  * This page is about to be returned from the page allocator
2142  */
2143 static inline int check_new_page(struct page *page)
2144 {
2145 	if (likely(page_expected_state(page,
2146 				PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
2147 		return 0;
2148 
2149 	check_new_page_bad(page);
2150 	return 1;
2151 }
2152 
2153 static inline bool free_pages_prezeroed(void)
2154 {
2155 	return (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
2156 		page_poisoning_enabled()) || want_init_on_free();
2157 }
2158 
2159 #ifdef CONFIG_DEBUG_VM
2160 /*
2161  * With DEBUG_VM enabled, order-0 pages are checked for expected state when
2162  * being allocated from pcp lists. With debug_pagealloc also enabled, they are
2163  * also checked when pcp lists are refilled from the free lists.
2164  */
2165 static inline bool check_pcp_refill(struct page *page)
2166 {
2167 	if (debug_pagealloc_enabled_static())
2168 		return check_new_page(page);
2169 	else
2170 		return false;
2171 }
2172 
2173 static inline bool check_new_pcp(struct page *page)
2174 {
2175 	return check_new_page(page);
2176 }
2177 #else
2178 /*
2179  * With DEBUG_VM disabled, free order-0 pages are checked for expected state
2180  * when pcp lists are being refilled from the free lists. With debug_pagealloc
2181  * enabled, they are also checked when being allocated from the pcp lists.
2182  */
2183 static inline bool check_pcp_refill(struct page *page)
2184 {
2185 	return check_new_page(page);
2186 }
2187 static inline bool check_new_pcp(struct page *page)
2188 {
2189 	if (debug_pagealloc_enabled_static())
2190 		return check_new_page(page);
2191 	else
2192 		return false;
2193 }
2194 #endif /* CONFIG_DEBUG_VM */
2195 
2196 static bool check_new_pages(struct page *page, unsigned int order)
2197 {
2198 	int i;
2199 	for (i = 0; i < (1 << order); i++) {
2200 		struct page *p = page + i;
2201 
2202 		if (unlikely(check_new_page(p)))
2203 			return true;
2204 	}
2205 
2206 	return false;
2207 }
2208 
2209 inline void post_alloc_hook(struct page *page, unsigned int order,
2210 				gfp_t gfp_flags)
2211 {
2212 	set_page_private(page, 0);
2213 	set_page_refcounted(page);
2214 
2215 	arch_alloc_page(page, order);
2216 	if (debug_pagealloc_enabled_static())
2217 		kernel_map_pages(page, 1 << order, 1);
2218 	kasan_alloc_pages(page, order);
2219 	kernel_poison_pages(page, 1 << order, 1);
2220 	set_page_owner(page, order, gfp_flags);
2221 }
2222 
2223 static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
2224 							unsigned int alloc_flags)
2225 {
2226 	post_alloc_hook(page, order, gfp_flags);
2227 
2228 	if (!free_pages_prezeroed() && want_init_on_alloc(gfp_flags))
2229 		kernel_init_free_pages(page, 1 << order);
2230 
2231 	if (order && (gfp_flags & __GFP_COMP))
2232 		prep_compound_page(page, order);
2233 
2234 	/*
2235 	 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
2236 	 * allocate the page. The expectation is that the caller is taking
2237 	 * steps that will free more memory. The caller should avoid the page
2238 	 * being used for !PFMEMALLOC purposes.
2239 	 */
2240 	if (alloc_flags & ALLOC_NO_WATERMARKS)
2241 		set_page_pfmemalloc(page);
2242 	else
2243 		clear_page_pfmemalloc(page);
2244 }
2245 
2246 /*
2247  * Go through the free lists for the given migratetype and remove
2248  * the smallest available page from the freelists
2249  */
2250 static __always_inline
2251 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
2252 						int migratetype)
2253 {
2254 	unsigned int current_order;
2255 	struct free_area *area;
2256 	struct page *page;
2257 
2258 	/* Find a page of the appropriate size in the preferred list */
2259 	for (current_order = order; current_order < MAX_ORDER; ++current_order) {
2260 		area = &(zone->free_area[current_order]);
2261 		page = get_page_from_free_area(area, migratetype);
2262 		if (!page)
2263 			continue;
2264 		del_page_from_free_list(page, zone, current_order);
2265 		expand(zone, page, order, current_order, migratetype);
2266 		set_pcppage_migratetype(page, migratetype);
2267 		return page;
2268 	}
2269 
2270 	return NULL;
2271 }
2272 
2273 
2274 /*
2275  * This array describes the order lists are fallen back to when
2276  * the free lists for the desirable migrate type are depleted
2277  */
2278 static int fallbacks[MIGRATE_TYPES][3] = {
2279 	[MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_TYPES },
2280 	[MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
2281 	[MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_TYPES },
2282 #ifdef CONFIG_CMA
2283 	[MIGRATE_CMA]         = { MIGRATE_TYPES }, /* Never used */
2284 #endif
2285 #ifdef CONFIG_MEMORY_ISOLATION
2286 	[MIGRATE_ISOLATE]     = { MIGRATE_TYPES }, /* Never used */
2287 #endif
2288 };
2289 
2290 #ifdef CONFIG_CMA
2291 static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2292 					unsigned int order)
2293 {
2294 	return __rmqueue_smallest(zone, order, MIGRATE_CMA);
2295 }
2296 #else
2297 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2298 					unsigned int order) { return NULL; }
2299 #endif
2300 
2301 /*
2302  * Move the free pages in a range to the free lists of the requested type.
2303  * Note that start_page and end_pages are not aligned on a pageblock
2304  * boundary. If alignment is required, use move_freepages_block()
2305  */
2306 static int move_freepages(struct zone *zone,
2307 			  struct page *start_page, struct page *end_page,
2308 			  int migratetype, int *num_movable)
2309 {
2310 	struct page *page;
2311 	unsigned int order;
2312 	int pages_moved = 0;
2313 
2314 	for (page = start_page; page <= end_page;) {
2315 		if (!pfn_valid_within(page_to_pfn(page))) {
2316 			page++;
2317 			continue;
2318 		}
2319 
2320 		if (!PageBuddy(page)) {
2321 			/*
2322 			 * We assume that pages that could be isolated for
2323 			 * migration are movable. But we don't actually try
2324 			 * isolating, as that would be expensive.
2325 			 */
2326 			if (num_movable &&
2327 					(PageLRU(page) || __PageMovable(page)))
2328 				(*num_movable)++;
2329 
2330 			page++;
2331 			continue;
2332 		}
2333 
2334 		/* Make sure we are not inadvertently changing nodes */
2335 		VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2336 		VM_BUG_ON_PAGE(page_zone(page) != zone, page);
2337 
2338 		order = page_order(page);
2339 		move_to_free_list(page, zone, order, migratetype);
2340 		page += 1 << order;
2341 		pages_moved += 1 << order;
2342 	}
2343 
2344 	return pages_moved;
2345 }
2346 
2347 int move_freepages_block(struct zone *zone, struct page *page,
2348 				int migratetype, int *num_movable)
2349 {
2350 	unsigned long start_pfn, end_pfn;
2351 	struct page *start_page, *end_page;
2352 
2353 	if (num_movable)
2354 		*num_movable = 0;
2355 
2356 	start_pfn = page_to_pfn(page);
2357 	start_pfn = start_pfn & ~(pageblock_nr_pages-1);
2358 	start_page = pfn_to_page(start_pfn);
2359 	end_page = start_page + pageblock_nr_pages - 1;
2360 	end_pfn = start_pfn + pageblock_nr_pages - 1;
2361 
2362 	/* Do not cross zone boundaries */
2363 	if (!zone_spans_pfn(zone, start_pfn))
2364 		start_page = page;
2365 	if (!zone_spans_pfn(zone, end_pfn))
2366 		return 0;
2367 
2368 	return move_freepages(zone, start_page, end_page, migratetype,
2369 								num_movable);
2370 }
2371 
2372 static void change_pageblock_range(struct page *pageblock_page,
2373 					int start_order, int migratetype)
2374 {
2375 	int nr_pageblocks = 1 << (start_order - pageblock_order);
2376 
2377 	while (nr_pageblocks--) {
2378 		set_pageblock_migratetype(pageblock_page, migratetype);
2379 		pageblock_page += pageblock_nr_pages;
2380 	}
2381 }
2382 
2383 /*
2384  * When we are falling back to another migratetype during allocation, try to
2385  * steal extra free pages from the same pageblocks to satisfy further
2386  * allocations, instead of polluting multiple pageblocks.
2387  *
2388  * If we are stealing a relatively large buddy page, it is likely there will
2389  * be more free pages in the pageblock, so try to steal them all. For
2390  * reclaimable and unmovable allocations, we steal regardless of page size,
2391  * as fragmentation caused by those allocations polluting movable pageblocks
2392  * is worse than movable allocations stealing from unmovable and reclaimable
2393  * pageblocks.
2394  */
2395 static bool can_steal_fallback(unsigned int order, int start_mt)
2396 {
2397 	/*
2398 	 * Leaving this order check is intended, although there is
2399 	 * relaxed order check in next check. The reason is that
2400 	 * we can actually steal whole pageblock if this condition met,
2401 	 * but, below check doesn't guarantee it and that is just heuristic
2402 	 * so could be changed anytime.
2403 	 */
2404 	if (order >= pageblock_order)
2405 		return true;
2406 
2407 	if (order >= pageblock_order / 2 ||
2408 		start_mt == MIGRATE_RECLAIMABLE ||
2409 		start_mt == MIGRATE_UNMOVABLE ||
2410 		page_group_by_mobility_disabled)
2411 		return true;
2412 
2413 	return false;
2414 }
2415 
2416 static inline void boost_watermark(struct zone *zone)
2417 {
2418 	unsigned long max_boost;
2419 
2420 	if (!watermark_boost_factor)
2421 		return;
2422 	/*
2423 	 * Don't bother in zones that are unlikely to produce results.
2424 	 * On small machines, including kdump capture kernels running
2425 	 * in a small area, boosting the watermark can cause an out of
2426 	 * memory situation immediately.
2427 	 */
2428 	if ((pageblock_nr_pages * 4) > zone_managed_pages(zone))
2429 		return;
2430 
2431 	max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
2432 			watermark_boost_factor, 10000);
2433 
2434 	/*
2435 	 * high watermark may be uninitialised if fragmentation occurs
2436 	 * very early in boot so do not boost. We do not fall
2437 	 * through and boost by pageblock_nr_pages as failing
2438 	 * allocations that early means that reclaim is not going
2439 	 * to help and it may even be impossible to reclaim the
2440 	 * boosted watermark resulting in a hang.
2441 	 */
2442 	if (!max_boost)
2443 		return;
2444 
2445 	max_boost = max(pageblock_nr_pages, max_boost);
2446 
2447 	zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,
2448 		max_boost);
2449 }
2450 
2451 /*
2452  * This function implements actual steal behaviour. If order is large enough,
2453  * we can steal whole pageblock. If not, we first move freepages in this
2454  * pageblock to our migratetype and determine how many already-allocated pages
2455  * are there in the pageblock with a compatible migratetype. If at least half
2456  * of pages are free or compatible, we can change migratetype of the pageblock
2457  * itself, so pages freed in the future will be put on the correct free list.
2458  */
2459 static void steal_suitable_fallback(struct zone *zone, struct page *page,
2460 		unsigned int alloc_flags, int start_type, bool whole_block)
2461 {
2462 	unsigned int current_order = page_order(page);
2463 	int free_pages, movable_pages, alike_pages;
2464 	int old_block_type;
2465 
2466 	old_block_type = get_pageblock_migratetype(page);
2467 
2468 	/*
2469 	 * This can happen due to races and we want to prevent broken
2470 	 * highatomic accounting.
2471 	 */
2472 	if (is_migrate_highatomic(old_block_type))
2473 		goto single_page;
2474 
2475 	/* Take ownership for orders >= pageblock_order */
2476 	if (current_order >= pageblock_order) {
2477 		change_pageblock_range(page, current_order, start_type);
2478 		goto single_page;
2479 	}
2480 
2481 	/*
2482 	 * Boost watermarks to increase reclaim pressure to reduce the
2483 	 * likelihood of future fallbacks. Wake kswapd now as the node
2484 	 * may be balanced overall and kswapd will not wake naturally.
2485 	 */
2486 	boost_watermark(zone);
2487 	if (alloc_flags & ALLOC_KSWAPD)
2488 		set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
2489 
2490 	/* We are not allowed to try stealing from the whole block */
2491 	if (!whole_block)
2492 		goto single_page;
2493 
2494 	free_pages = move_freepages_block(zone, page, start_type,
2495 						&movable_pages);
2496 	/*
2497 	 * Determine how many pages are compatible with our allocation.
2498 	 * For movable allocation, it's the number of movable pages which
2499 	 * we just obtained. For other types it's a bit more tricky.
2500 	 */
2501 	if (start_type == MIGRATE_MOVABLE) {
2502 		alike_pages = movable_pages;
2503 	} else {
2504 		/*
2505 		 * If we are falling back a RECLAIMABLE or UNMOVABLE allocation
2506 		 * to MOVABLE pageblock, consider all non-movable pages as
2507 		 * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or
2508 		 * vice versa, be conservative since we can't distinguish the
2509 		 * exact migratetype of non-movable pages.
2510 		 */
2511 		if (old_block_type == MIGRATE_MOVABLE)
2512 			alike_pages = pageblock_nr_pages
2513 						- (free_pages + movable_pages);
2514 		else
2515 			alike_pages = 0;
2516 	}
2517 
2518 	/* moving whole block can fail due to zone boundary conditions */
2519 	if (!free_pages)
2520 		goto single_page;
2521 
2522 	/*
2523 	 * If a sufficient number of pages in the block are either free or of
2524 	 * comparable migratability as our allocation, claim the whole block.
2525 	 */
2526 	if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2527 			page_group_by_mobility_disabled)
2528 		set_pageblock_migratetype(page, start_type);
2529 
2530 	return;
2531 
2532 single_page:
2533 	move_to_free_list(page, zone, current_order, start_type);
2534 }
2535 
2536 /*
2537  * Check whether there is a suitable fallback freepage with requested order.
2538  * If only_stealable is true, this function returns fallback_mt only if
2539  * we can steal other freepages all together. This would help to reduce
2540  * fragmentation due to mixed migratetype pages in one pageblock.
2541  */
2542 int find_suitable_fallback(struct free_area *area, unsigned int order,
2543 			int migratetype, bool only_stealable, bool *can_steal)
2544 {
2545 	int i;
2546 	int fallback_mt;
2547 
2548 	if (area->nr_free == 0)
2549 		return -1;
2550 
2551 	*can_steal = false;
2552 	for (i = 0;; i++) {
2553 		fallback_mt = fallbacks[migratetype][i];
2554 		if (fallback_mt == MIGRATE_TYPES)
2555 			break;
2556 
2557 		if (free_area_empty(area, fallback_mt))
2558 			continue;
2559 
2560 		if (can_steal_fallback(order, migratetype))
2561 			*can_steal = true;
2562 
2563 		if (!only_stealable)
2564 			return fallback_mt;
2565 
2566 		if (*can_steal)
2567 			return fallback_mt;
2568 	}
2569 
2570 	return -1;
2571 }
2572 
2573 /*
2574  * Reserve a pageblock for exclusive use of high-order atomic allocations if
2575  * there are no empty page blocks that contain a page with a suitable order
2576  */
2577 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2578 				unsigned int alloc_order)
2579 {
2580 	int mt;
2581 	unsigned long max_managed, flags;
2582 
2583 	/*
2584 	 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2585 	 * Check is race-prone but harmless.
2586 	 */
2587 	max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
2588 	if (zone->nr_reserved_highatomic >= max_managed)
2589 		return;
2590 
2591 	spin_lock_irqsave(&zone->lock, flags);
2592 
2593 	/* Recheck the nr_reserved_highatomic limit under the lock */
2594 	if (zone->nr_reserved_highatomic >= max_managed)
2595 		goto out_unlock;
2596 
2597 	/* Yoink! */
2598 	mt = get_pageblock_migratetype(page);
2599 	if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2600 	    && !is_migrate_cma(mt)) {
2601 		zone->nr_reserved_highatomic += pageblock_nr_pages;
2602 		set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2603 		move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2604 	}
2605 
2606 out_unlock:
2607 	spin_unlock_irqrestore(&zone->lock, flags);
2608 }
2609 
2610 /*
2611  * Used when an allocation is about to fail under memory pressure. This
2612  * potentially hurts the reliability of high-order allocations when under
2613  * intense memory pressure but failed atomic allocations should be easier
2614  * to recover from than an OOM.
2615  *
2616  * If @force is true, try to unreserve a pageblock even though highatomic
2617  * pageblock is exhausted.
2618  */
2619 static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2620 						bool force)
2621 {
2622 	struct zonelist *zonelist = ac->zonelist;
2623 	unsigned long flags;
2624 	struct zoneref *z;
2625 	struct zone *zone;
2626 	struct page *page;
2627 	int order;
2628 	bool ret;
2629 
2630 	for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx,
2631 								ac->nodemask) {
2632 		/*
2633 		 * Preserve at least one pageblock unless memory pressure
2634 		 * is really high.
2635 		 */
2636 		if (!force && zone->nr_reserved_highatomic <=
2637 					pageblock_nr_pages)
2638 			continue;
2639 
2640 		spin_lock_irqsave(&zone->lock, flags);
2641 		for (order = 0; order < MAX_ORDER; order++) {
2642 			struct free_area *area = &(zone->free_area[order]);
2643 
2644 			page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC);
2645 			if (!page)
2646 				continue;
2647 
2648 			/*
2649 			 * In page freeing path, migratetype change is racy so
2650 			 * we can counter several free pages in a pageblock
2651 			 * in this loop althoug we changed the pageblock type
2652 			 * from highatomic to ac->migratetype. So we should
2653 			 * adjust the count once.
2654 			 */
2655 			if (is_migrate_highatomic_page(page)) {
2656 				/*
2657 				 * It should never happen but changes to
2658 				 * locking could inadvertently allow a per-cpu
2659 				 * drain to add pages to MIGRATE_HIGHATOMIC
2660 				 * while unreserving so be safe and watch for
2661 				 * underflows.
2662 				 */
2663 				zone->nr_reserved_highatomic -= min(
2664 						pageblock_nr_pages,
2665 						zone->nr_reserved_highatomic);
2666 			}
2667 
2668 			/*
2669 			 * Convert to ac->migratetype and avoid the normal
2670 			 * pageblock stealing heuristics. Minimally, the caller
2671 			 * is doing the work and needs the pages. More
2672 			 * importantly, if the block was always converted to
2673 			 * MIGRATE_UNMOVABLE or another type then the number
2674 			 * of pageblocks that cannot be completely freed
2675 			 * may increase.
2676 			 */
2677 			set_pageblock_migratetype(page, ac->migratetype);
2678 			ret = move_freepages_block(zone, page, ac->migratetype,
2679 									NULL);
2680 			if (ret) {
2681 				spin_unlock_irqrestore(&zone->lock, flags);
2682 				return ret;
2683 			}
2684 		}
2685 		spin_unlock_irqrestore(&zone->lock, flags);
2686 	}
2687 
2688 	return false;
2689 }
2690 
2691 /*
2692  * Try finding a free buddy page on the fallback list and put it on the free
2693  * list of requested migratetype, possibly along with other pages from the same
2694  * block, depending on fragmentation avoidance heuristics. Returns true if
2695  * fallback was found so that __rmqueue_smallest() can grab it.
2696  *
2697  * The use of signed ints for order and current_order is a deliberate
2698  * deviation from the rest of this file, to make the for loop
2699  * condition simpler.
2700  */
2701 static __always_inline bool
2702 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype,
2703 						unsigned int alloc_flags)
2704 {
2705 	struct free_area *area;
2706 	int current_order;
2707 	int min_order = order;
2708 	struct page *page;
2709 	int fallback_mt;
2710 	bool can_steal;
2711 
2712 	/*
2713 	 * Do not steal pages from freelists belonging to other pageblocks
2714 	 * i.e. orders < pageblock_order. If there are no local zones free,
2715 	 * the zonelists will be reiterated without ALLOC_NOFRAGMENT.
2716 	 */
2717 	if (alloc_flags & ALLOC_NOFRAGMENT)
2718 		min_order = pageblock_order;
2719 
2720 	/*
2721 	 * Find the largest available free page in the other list. This roughly
2722 	 * approximates finding the pageblock with the most free pages, which
2723 	 * would be too costly to do exactly.
2724 	 */
2725 	for (current_order = MAX_ORDER - 1; current_order >= min_order;
2726 				--current_order) {
2727 		area = &(zone->free_area[current_order]);
2728 		fallback_mt = find_suitable_fallback(area, current_order,
2729 				start_migratetype, false, &can_steal);
2730 		if (fallback_mt == -1)
2731 			continue;
2732 
2733 		/*
2734 		 * We cannot steal all free pages from the pageblock and the
2735 		 * requested migratetype is movable. In that case it's better to
2736 		 * steal and split the smallest available page instead of the
2737 		 * largest available page, because even if the next movable
2738 		 * allocation falls back into a different pageblock than this
2739 		 * one, it won't cause permanent fragmentation.
2740 		 */
2741 		if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2742 					&& current_order > order)
2743 			goto find_smallest;
2744 
2745 		goto do_steal;
2746 	}
2747 
2748 	return false;
2749 
2750 find_smallest:
2751 	for (current_order = order; current_order < MAX_ORDER;
2752 							current_order++) {
2753 		area = &(zone->free_area[current_order]);
2754 		fallback_mt = find_suitable_fallback(area, current_order,
2755 				start_migratetype, false, &can_steal);
2756 		if (fallback_mt != -1)
2757 			break;
2758 	}
2759 
2760 	/*
2761 	 * This should not happen - we already found a suitable fallback
2762 	 * when looking for the largest page.
2763 	 */
2764 	VM_BUG_ON(current_order == MAX_ORDER);
2765 
2766 do_steal:
2767 	page = get_page_from_free_area(area, fallback_mt);
2768 
2769 	steal_suitable_fallback(zone, page, alloc_flags, start_migratetype,
2770 								can_steal);
2771 
2772 	trace_mm_page_alloc_extfrag(page, order, current_order,
2773 		start_migratetype, fallback_mt);
2774 
2775 	return true;
2776 
2777 }
2778 
2779 /*
2780  * Do the hard work of removing an element from the buddy allocator.
2781  * Call me with the zone->lock already held.
2782  */
2783 static __always_inline struct page *
2784 __rmqueue(struct zone *zone, unsigned int order, int migratetype,
2785 						unsigned int alloc_flags)
2786 {
2787 	struct page *page;
2788 
2789 #ifdef CONFIG_CMA
2790 	/*
2791 	 * Balance movable allocations between regular and CMA areas by
2792 	 * allocating from CMA when over half of the zone's free memory
2793 	 * is in the CMA area.
2794 	 */
2795 	if (alloc_flags & ALLOC_CMA &&
2796 	    zone_page_state(zone, NR_FREE_CMA_PAGES) >
2797 	    zone_page_state(zone, NR_FREE_PAGES) / 2) {
2798 		page = __rmqueue_cma_fallback(zone, order);
2799 		if (page)
2800 			return page;
2801 	}
2802 #endif
2803 retry:
2804 	page = __rmqueue_smallest(zone, order, migratetype);
2805 	if (unlikely(!page)) {
2806 		if (alloc_flags & ALLOC_CMA)
2807 			page = __rmqueue_cma_fallback(zone, order);
2808 
2809 		if (!page && __rmqueue_fallback(zone, order, migratetype,
2810 								alloc_flags))
2811 			goto retry;
2812 	}
2813 
2814 	trace_mm_page_alloc_zone_locked(page, order, migratetype);
2815 	return page;
2816 }
2817 
2818 /*
2819  * Obtain a specified number of elements from the buddy allocator, all under
2820  * a single hold of the lock, for efficiency.  Add them to the supplied list.
2821  * Returns the number of new pages which were placed at *list.
2822  */
2823 static int rmqueue_bulk(struct zone *zone, unsigned int order,
2824 			unsigned long count, struct list_head *list,
2825 			int migratetype, unsigned int alloc_flags)
2826 {
2827 	int i, alloced = 0;
2828 
2829 	spin_lock(&zone->lock);
2830 	for (i = 0; i < count; ++i) {
2831 		struct page *page = __rmqueue(zone, order, migratetype,
2832 								alloc_flags);
2833 		if (unlikely(page == NULL))
2834 			break;
2835 
2836 		if (unlikely(check_pcp_refill(page)))
2837 			continue;
2838 
2839 		/*
2840 		 * Split buddy pages returned by expand() are received here in
2841 		 * physical page order. The page is added to the tail of
2842 		 * caller's list. From the callers perspective, the linked list
2843 		 * is ordered by page number under some conditions. This is
2844 		 * useful for IO devices that can forward direction from the
2845 		 * head, thus also in the physical page order. This is useful
2846 		 * for IO devices that can merge IO requests if the physical
2847 		 * pages are ordered properly.
2848 		 */
2849 		list_add_tail(&page->lru, list);
2850 		alloced++;
2851 		if (is_migrate_cma(get_pcppage_migratetype(page)))
2852 			__mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2853 					      -(1 << order));
2854 	}
2855 
2856 	/*
2857 	 * i pages were removed from the buddy list even if some leak due
2858 	 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2859 	 * on i. Do not confuse with 'alloced' which is the number of
2860 	 * pages added to the pcp list.
2861 	 */
2862 	__mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2863 	spin_unlock(&zone->lock);
2864 	return alloced;
2865 }
2866 
2867 #ifdef CONFIG_NUMA
2868 /*
2869  * Called from the vmstat counter updater to drain pagesets of this
2870  * currently executing processor on remote nodes after they have
2871  * expired.
2872  *
2873  * Note that this function must be called with the thread pinned to
2874  * a single processor.
2875  */
2876 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2877 {
2878 	unsigned long flags;
2879 	int to_drain, batch;
2880 
2881 	local_irq_save(flags);
2882 	batch = READ_ONCE(pcp->batch);
2883 	to_drain = min(pcp->count, batch);
2884 	if (to_drain > 0)
2885 		free_pcppages_bulk(zone, to_drain, pcp);
2886 	local_irq_restore(flags);
2887 }
2888 #endif
2889 
2890 /*
2891  * Drain pcplists of the indicated processor and zone.
2892  *
2893  * The processor must either be the current processor and the
2894  * thread pinned to the current processor or a processor that
2895  * is not online.
2896  */
2897 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2898 {
2899 	unsigned long flags;
2900 	struct per_cpu_pageset *pset;
2901 	struct per_cpu_pages *pcp;
2902 
2903 	local_irq_save(flags);
2904 	pset = per_cpu_ptr(zone->pageset, cpu);
2905 
2906 	pcp = &pset->pcp;
2907 	if (pcp->count)
2908 		free_pcppages_bulk(zone, pcp->count, pcp);
2909 	local_irq_restore(flags);
2910 }
2911 
2912 /*
2913  * Drain pcplists of all zones on the indicated processor.
2914  *
2915  * The processor must either be the current processor and the
2916  * thread pinned to the current processor or a processor that
2917  * is not online.
2918  */
2919 static void drain_pages(unsigned int cpu)
2920 {
2921 	struct zone *zone;
2922 
2923 	for_each_populated_zone(zone) {
2924 		drain_pages_zone(cpu, zone);
2925 	}
2926 }
2927 
2928 /*
2929  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2930  *
2931  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2932  * the single zone's pages.
2933  */
2934 void drain_local_pages(struct zone *zone)
2935 {
2936 	int cpu = smp_processor_id();
2937 
2938 	if (zone)
2939 		drain_pages_zone(cpu, zone);
2940 	else
2941 		drain_pages(cpu);
2942 }
2943 
2944 static void drain_local_pages_wq(struct work_struct *work)
2945 {
2946 	struct pcpu_drain *drain;
2947 
2948 	drain = container_of(work, struct pcpu_drain, work);
2949 
2950 	/*
2951 	 * drain_all_pages doesn't use proper cpu hotplug protection so
2952 	 * we can race with cpu offline when the WQ can move this from
2953 	 * a cpu pinned worker to an unbound one. We can operate on a different
2954 	 * cpu which is allright but we also have to make sure to not move to
2955 	 * a different one.
2956 	 */
2957 	preempt_disable();
2958 	drain_local_pages(drain->zone);
2959 	preempt_enable();
2960 }
2961 
2962 /*
2963  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2964  *
2965  * When zone parameter is non-NULL, spill just the single zone's pages.
2966  *
2967  * Note that this can be extremely slow as the draining happens in a workqueue.
2968  */
2969 void drain_all_pages(struct zone *zone)
2970 {
2971 	int cpu;
2972 
2973 	/*
2974 	 * Allocate in the BSS so we wont require allocation in
2975 	 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2976 	 */
2977 	static cpumask_t cpus_with_pcps;
2978 
2979 	/*
2980 	 * Make sure nobody triggers this path before mm_percpu_wq is fully
2981 	 * initialized.
2982 	 */
2983 	if (WARN_ON_ONCE(!mm_percpu_wq))
2984 		return;
2985 
2986 	/*
2987 	 * Do not drain if one is already in progress unless it's specific to
2988 	 * a zone. Such callers are primarily CMA and memory hotplug and need
2989 	 * the drain to be complete when the call returns.
2990 	 */
2991 	if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
2992 		if (!zone)
2993 			return;
2994 		mutex_lock(&pcpu_drain_mutex);
2995 	}
2996 
2997 	/*
2998 	 * We don't care about racing with CPU hotplug event
2999 	 * as offline notification will cause the notified
3000 	 * cpu to drain that CPU pcps and on_each_cpu_mask
3001 	 * disables preemption as part of its processing
3002 	 */
3003 	for_each_online_cpu(cpu) {
3004 		struct per_cpu_pageset *pcp;
3005 		struct zone *z;
3006 		bool has_pcps = false;
3007 
3008 		if (zone) {
3009 			pcp = per_cpu_ptr(zone->pageset, cpu);
3010 			if (pcp->pcp.count)
3011 				has_pcps = true;
3012 		} else {
3013 			for_each_populated_zone(z) {
3014 				pcp = per_cpu_ptr(z->pageset, cpu);
3015 				if (pcp->pcp.count) {
3016 					has_pcps = true;
3017 					break;
3018 				}
3019 			}
3020 		}
3021 
3022 		if (has_pcps)
3023 			cpumask_set_cpu(cpu, &cpus_with_pcps);
3024 		else
3025 			cpumask_clear_cpu(cpu, &cpus_with_pcps);
3026 	}
3027 
3028 	for_each_cpu(cpu, &cpus_with_pcps) {
3029 		struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu);
3030 
3031 		drain->zone = zone;
3032 		INIT_WORK(&drain->work, drain_local_pages_wq);
3033 		queue_work_on(cpu, mm_percpu_wq, &drain->work);
3034 	}
3035 	for_each_cpu(cpu, &cpus_with_pcps)
3036 		flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work);
3037 
3038 	mutex_unlock(&pcpu_drain_mutex);
3039 }
3040 
3041 #ifdef CONFIG_HIBERNATION
3042 
3043 /*
3044  * Touch the watchdog for every WD_PAGE_COUNT pages.
3045  */
3046 #define WD_PAGE_COUNT	(128*1024)
3047 
3048 void mark_free_pages(struct zone *zone)
3049 {
3050 	unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
3051 	unsigned long flags;
3052 	unsigned int order, t;
3053 	struct page *page;
3054 
3055 	if (zone_is_empty(zone))
3056 		return;
3057 
3058 	spin_lock_irqsave(&zone->lock, flags);
3059 
3060 	max_zone_pfn = zone_end_pfn(zone);
3061 	for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
3062 		if (pfn_valid(pfn)) {
3063 			page = pfn_to_page(pfn);
3064 
3065 			if (!--page_count) {
3066 				touch_nmi_watchdog();
3067 				page_count = WD_PAGE_COUNT;
3068 			}
3069 
3070 			if (page_zone(page) != zone)
3071 				continue;
3072 
3073 			if (!swsusp_page_is_forbidden(page))
3074 				swsusp_unset_page_free(page);
3075 		}
3076 
3077 	for_each_migratetype_order(order, t) {
3078 		list_for_each_entry(page,
3079 				&zone->free_area[order].free_list[t], lru) {
3080 			unsigned long i;
3081 
3082 			pfn = page_to_pfn(page);
3083 			for (i = 0; i < (1UL << order); i++) {
3084 				if (!--page_count) {
3085 					touch_nmi_watchdog();
3086 					page_count = WD_PAGE_COUNT;
3087 				}
3088 				swsusp_set_page_free(pfn_to_page(pfn + i));
3089 			}
3090 		}
3091 	}
3092 	spin_unlock_irqrestore(&zone->lock, flags);
3093 }
3094 #endif /* CONFIG_PM */
3095 
3096 static bool free_unref_page_prepare(struct page *page, unsigned long pfn)
3097 {
3098 	int migratetype;
3099 
3100 	if (!free_pcp_prepare(page))
3101 		return false;
3102 
3103 	migratetype = get_pfnblock_migratetype(page, pfn);
3104 	set_pcppage_migratetype(page, migratetype);
3105 	return true;
3106 }
3107 
3108 static void free_unref_page_commit(struct page *page, unsigned long pfn)
3109 {
3110 	struct zone *zone = page_zone(page);
3111 	struct per_cpu_pages *pcp;
3112 	int migratetype;
3113 
3114 	migratetype = get_pcppage_migratetype(page);
3115 	__count_vm_event(PGFREE);
3116 
3117 	/*
3118 	 * We only track unmovable, reclaimable and movable on pcp lists.
3119 	 * Free ISOLATE pages back to the allocator because they are being
3120 	 * offlined but treat HIGHATOMIC as movable pages so we can get those
3121 	 * areas back if necessary. Otherwise, we may have to free
3122 	 * excessively into the page allocator
3123 	 */
3124 	if (migratetype >= MIGRATE_PCPTYPES) {
3125 		if (unlikely(is_migrate_isolate(migratetype))) {
3126 			free_one_page(zone, page, pfn, 0, migratetype);
3127 			return;
3128 		}
3129 		migratetype = MIGRATE_MOVABLE;
3130 	}
3131 
3132 	pcp = &this_cpu_ptr(zone->pageset)->pcp;
3133 	list_add(&page->lru, &pcp->lists[migratetype]);
3134 	pcp->count++;
3135 	if (pcp->count >= pcp->high) {
3136 		unsigned long batch = READ_ONCE(pcp->batch);
3137 		free_pcppages_bulk(zone, batch, pcp);
3138 	}
3139 }
3140 
3141 /*
3142  * Free a 0-order page
3143  */
3144 void free_unref_page(struct page *page)
3145 {
3146 	unsigned long flags;
3147 	unsigned long pfn = page_to_pfn(page);
3148 
3149 	if (!free_unref_page_prepare(page, pfn))
3150 		return;
3151 
3152 	local_irq_save(flags);
3153 	free_unref_page_commit(page, pfn);
3154 	local_irq_restore(flags);
3155 }
3156 
3157 /*
3158  * Free a list of 0-order pages
3159  */
3160 void free_unref_page_list(struct list_head *list)
3161 {
3162 	struct page *page, *next;
3163 	unsigned long flags, pfn;
3164 	int batch_count = 0;
3165 
3166 	/* Prepare pages for freeing */
3167 	list_for_each_entry_safe(page, next, list, lru) {
3168 		pfn = page_to_pfn(page);
3169 		if (!free_unref_page_prepare(page, pfn))
3170 			list_del(&page->lru);
3171 		set_page_private(page, pfn);
3172 	}
3173 
3174 	local_irq_save(flags);
3175 	list_for_each_entry_safe(page, next, list, lru) {
3176 		unsigned long pfn = page_private(page);
3177 
3178 		set_page_private(page, 0);
3179 		trace_mm_page_free_batched(page);
3180 		free_unref_page_commit(page, pfn);
3181 
3182 		/*
3183 		 * Guard against excessive IRQ disabled times when we get
3184 		 * a large list of pages to free.
3185 		 */
3186 		if (++batch_count == SWAP_CLUSTER_MAX) {
3187 			local_irq_restore(flags);
3188 			batch_count = 0;
3189 			local_irq_save(flags);
3190 		}
3191 	}
3192 	local_irq_restore(flags);
3193 }
3194 
3195 /*
3196  * split_page takes a non-compound higher-order page, and splits it into
3197  * n (1<<order) sub-pages: page[0..n]
3198  * Each sub-page must be freed individually.
3199  *
3200  * Note: this is probably too low level an operation for use in drivers.
3201  * Please consult with lkml before using this in your driver.
3202  */
3203 void split_page(struct page *page, unsigned int order)
3204 {
3205 	int i;
3206 
3207 	VM_BUG_ON_PAGE(PageCompound(page), page);
3208 	VM_BUG_ON_PAGE(!page_count(page), page);
3209 
3210 	for (i = 1; i < (1 << order); i++)
3211 		set_page_refcounted(page + i);
3212 	split_page_owner(page, order);
3213 }
3214 EXPORT_SYMBOL_GPL(split_page);
3215 
3216 int __isolate_free_page(struct page *page, unsigned int order)
3217 {
3218 	unsigned long watermark;
3219 	struct zone *zone;
3220 	int mt;
3221 
3222 	BUG_ON(!PageBuddy(page));
3223 
3224 	zone = page_zone(page);
3225 	mt = get_pageblock_migratetype(page);
3226 
3227 	if (!is_migrate_isolate(mt)) {
3228 		/*
3229 		 * Obey watermarks as if the page was being allocated. We can
3230 		 * emulate a high-order watermark check with a raised order-0
3231 		 * watermark, because we already know our high-order page
3232 		 * exists.
3233 		 */
3234 		watermark = zone->_watermark[WMARK_MIN] + (1UL << order);
3235 		if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
3236 			return 0;
3237 
3238 		__mod_zone_freepage_state(zone, -(1UL << order), mt);
3239 	}
3240 
3241 	/* Remove page from free list */
3242 
3243 	del_page_from_free_list(page, zone, order);
3244 
3245 	/*
3246 	 * Set the pageblock if the isolated page is at least half of a
3247 	 * pageblock
3248 	 */
3249 	if (order >= pageblock_order - 1) {
3250 		struct page *endpage = page + (1 << order) - 1;
3251 		for (; page < endpage; page += pageblock_nr_pages) {
3252 			int mt = get_pageblock_migratetype(page);
3253 			if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
3254 			    && !is_migrate_highatomic(mt))
3255 				set_pageblock_migratetype(page,
3256 							  MIGRATE_MOVABLE);
3257 		}
3258 	}
3259 
3260 
3261 	return 1UL << order;
3262 }
3263 
3264 /**
3265  * __putback_isolated_page - Return a now-isolated page back where we got it
3266  * @page: Page that was isolated
3267  * @order: Order of the isolated page
3268  * @mt: The page's pageblock's migratetype
3269  *
3270  * This function is meant to return a page pulled from the free lists via
3271  * __isolate_free_page back to the free lists they were pulled from.
3272  */
3273 void __putback_isolated_page(struct page *page, unsigned int order, int mt)
3274 {
3275 	struct zone *zone = page_zone(page);
3276 
3277 	/* zone lock should be held when this function is called */
3278 	lockdep_assert_held(&zone->lock);
3279 
3280 	/* Return isolated page to tail of freelist. */
3281 	__free_one_page(page, page_to_pfn(page), zone, order, mt, false);
3282 }
3283 
3284 /*
3285  * Update NUMA hit/miss statistics
3286  *
3287  * Must be called with interrupts disabled.
3288  */
3289 static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
3290 {
3291 #ifdef CONFIG_NUMA
3292 	enum numa_stat_item local_stat = NUMA_LOCAL;
3293 
3294 	/* skip numa counters update if numa stats is disabled */
3295 	if (!static_branch_likely(&vm_numa_stat_key))
3296 		return;
3297 
3298 	if (zone_to_nid(z) != numa_node_id())
3299 		local_stat = NUMA_OTHER;
3300 
3301 	if (zone_to_nid(z) == zone_to_nid(preferred_zone))
3302 		__inc_numa_state(z, NUMA_HIT);
3303 	else {
3304 		__inc_numa_state(z, NUMA_MISS);
3305 		__inc_numa_state(preferred_zone, NUMA_FOREIGN);
3306 	}
3307 	__inc_numa_state(z, local_stat);
3308 #endif
3309 }
3310 
3311 /* Remove page from the per-cpu list, caller must protect the list */
3312 static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
3313 			unsigned int alloc_flags,
3314 			struct per_cpu_pages *pcp,
3315 			struct list_head *list)
3316 {
3317 	struct page *page;
3318 
3319 	do {
3320 		if (list_empty(list)) {
3321 			pcp->count += rmqueue_bulk(zone, 0,
3322 					pcp->batch, list,
3323 					migratetype, alloc_flags);
3324 			if (unlikely(list_empty(list)))
3325 				return NULL;
3326 		}
3327 
3328 		page = list_first_entry(list, struct page, lru);
3329 		list_del(&page->lru);
3330 		pcp->count--;
3331 	} while (check_new_pcp(page));
3332 
3333 	return page;
3334 }
3335 
3336 /* Lock and remove page from the per-cpu list */
3337 static struct page *rmqueue_pcplist(struct zone *preferred_zone,
3338 			struct zone *zone, gfp_t gfp_flags,
3339 			int migratetype, unsigned int alloc_flags)
3340 {
3341 	struct per_cpu_pages *pcp;
3342 	struct list_head *list;
3343 	struct page *page;
3344 	unsigned long flags;
3345 
3346 	local_irq_save(flags);
3347 	pcp = &this_cpu_ptr(zone->pageset)->pcp;
3348 	list = &pcp->lists[migratetype];
3349 	page = __rmqueue_pcplist(zone,  migratetype, alloc_flags, pcp, list);
3350 	if (page) {
3351 		__count_zid_vm_events(PGALLOC, page_zonenum(page), 1);
3352 		zone_statistics(preferred_zone, zone);
3353 	}
3354 	local_irq_restore(flags);
3355 	return page;
3356 }
3357 
3358 /*
3359  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
3360  */
3361 static inline
3362 struct page *rmqueue(struct zone *preferred_zone,
3363 			struct zone *zone, unsigned int order,
3364 			gfp_t gfp_flags, unsigned int alloc_flags,
3365 			int migratetype)
3366 {
3367 	unsigned long flags;
3368 	struct page *page;
3369 
3370 	if (likely(order == 0)) {
3371 		/*
3372 		 * MIGRATE_MOVABLE pcplist could have the pages on CMA area and
3373 		 * we need to skip it when CMA area isn't allowed.
3374 		 */
3375 		if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA ||
3376 				migratetype != MIGRATE_MOVABLE) {
3377 			page = rmqueue_pcplist(preferred_zone, zone, gfp_flags,
3378 					migratetype, alloc_flags);
3379 			goto out;
3380 		}
3381 	}
3382 
3383 	/*
3384 	 * We most definitely don't want callers attempting to
3385 	 * allocate greater than order-1 page units with __GFP_NOFAIL.
3386 	 */
3387 	WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
3388 	spin_lock_irqsave(&zone->lock, flags);
3389 
3390 	do {
3391 		page = NULL;
3392 		/*
3393 		 * order-0 request can reach here when the pcplist is skipped
3394 		 * due to non-CMA allocation context. HIGHATOMIC area is
3395 		 * reserved for high-order atomic allocation, so order-0
3396 		 * request should skip it.
3397 		 */
3398 		if (order > 0 && alloc_flags & ALLOC_HARDER) {
3399 			page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3400 			if (page)
3401 				trace_mm_page_alloc_zone_locked(page, order, migratetype);
3402 		}
3403 		if (!page)
3404 			page = __rmqueue(zone, order, migratetype, alloc_flags);
3405 	} while (page && check_new_pages(page, order));
3406 	spin_unlock(&zone->lock);
3407 	if (!page)
3408 		goto failed;
3409 	__mod_zone_freepage_state(zone, -(1 << order),
3410 				  get_pcppage_migratetype(page));
3411 
3412 	__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3413 	zone_statistics(preferred_zone, zone);
3414 	local_irq_restore(flags);
3415 
3416 out:
3417 	/* Separate test+clear to avoid unnecessary atomics */
3418 	if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
3419 		clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
3420 		wakeup_kswapd(zone, 0, 0, zone_idx(zone));
3421 	}
3422 
3423 	VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3424 	return page;
3425 
3426 failed:
3427 	local_irq_restore(flags);
3428 	return NULL;
3429 }
3430 
3431 #ifdef CONFIG_FAIL_PAGE_ALLOC
3432 
3433 static struct {
3434 	struct fault_attr attr;
3435 
3436 	bool ignore_gfp_highmem;
3437 	bool ignore_gfp_reclaim;
3438 	u32 min_order;
3439 } fail_page_alloc = {
3440 	.attr = FAULT_ATTR_INITIALIZER,
3441 	.ignore_gfp_reclaim = true,
3442 	.ignore_gfp_highmem = true,
3443 	.min_order = 1,
3444 };
3445 
3446 static int __init setup_fail_page_alloc(char *str)
3447 {
3448 	return setup_fault_attr(&fail_page_alloc.attr, str);
3449 }
3450 __setup("fail_page_alloc=", setup_fail_page_alloc);
3451 
3452 static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3453 {
3454 	if (order < fail_page_alloc.min_order)
3455 		return false;
3456 	if (gfp_mask & __GFP_NOFAIL)
3457 		return false;
3458 	if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3459 		return false;
3460 	if (fail_page_alloc.ignore_gfp_reclaim &&
3461 			(gfp_mask & __GFP_DIRECT_RECLAIM))
3462 		return false;
3463 
3464 	return should_fail(&fail_page_alloc.attr, 1 << order);
3465 }
3466 
3467 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3468 
3469 static int __init fail_page_alloc_debugfs(void)
3470 {
3471 	umode_t mode = S_IFREG | 0600;
3472 	struct dentry *dir;
3473 
3474 	dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3475 					&fail_page_alloc.attr);
3476 
3477 	debugfs_create_bool("ignore-gfp-wait", mode, dir,
3478 			    &fail_page_alloc.ignore_gfp_reclaim);
3479 	debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3480 			    &fail_page_alloc.ignore_gfp_highmem);
3481 	debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order);
3482 
3483 	return 0;
3484 }
3485 
3486 late_initcall(fail_page_alloc_debugfs);
3487 
3488 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
3489 
3490 #else /* CONFIG_FAIL_PAGE_ALLOC */
3491 
3492 static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3493 {
3494 	return false;
3495 }
3496 
3497 #endif /* CONFIG_FAIL_PAGE_ALLOC */
3498 
3499 static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3500 {
3501 	return __should_fail_alloc_page(gfp_mask, order);
3502 }
3503 ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE);
3504 
3505 static inline long __zone_watermark_unusable_free(struct zone *z,
3506 				unsigned int order, unsigned int alloc_flags)
3507 {
3508 	const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3509 	long unusable_free = (1 << order) - 1;
3510 
3511 	/*
3512 	 * If the caller does not have rights to ALLOC_HARDER then subtract
3513 	 * the high-atomic reserves. This will over-estimate the size of the
3514 	 * atomic reserve but it avoids a search.
3515 	 */
3516 	if (likely(!alloc_harder))
3517 		unusable_free += z->nr_reserved_highatomic;
3518 
3519 #ifdef CONFIG_CMA
3520 	/* If allocation can't use CMA areas don't use free CMA pages */
3521 	if (!(alloc_flags & ALLOC_CMA))
3522 		unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES);
3523 #endif
3524 
3525 	return unusable_free;
3526 }
3527 
3528 /*
3529  * Return true if free base pages are above 'mark'. For high-order checks it
3530  * will return true of the order-0 watermark is reached and there is at least
3531  * one free page of a suitable size. Checking now avoids taking the zone lock
3532  * to check in the allocation paths if no pages are free.
3533  */
3534 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3535 			 int highest_zoneidx, unsigned int alloc_flags,
3536 			 long free_pages)
3537 {
3538 	long min = mark;
3539 	int o;
3540 	const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3541 
3542 	/* free_pages may go negative - that's OK */
3543 	free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags);
3544 
3545 	if (alloc_flags & ALLOC_HIGH)
3546 		min -= min / 2;
3547 
3548 	if (unlikely(alloc_harder)) {
3549 		/*
3550 		 * OOM victims can try even harder than normal ALLOC_HARDER
3551 		 * users on the grounds that it's definitely going to be in
3552 		 * the exit path shortly and free memory. Any allocation it
3553 		 * makes during the free path will be small and short-lived.
3554 		 */
3555 		if (alloc_flags & ALLOC_OOM)
3556 			min -= min / 2;
3557 		else
3558 			min -= min / 4;
3559 	}
3560 
3561 	/*
3562 	 * Check watermarks for an order-0 allocation request. If these
3563 	 * are not met, then a high-order request also cannot go ahead
3564 	 * even if a suitable page happened to be free.
3565 	 */
3566 	if (free_pages <= min + z->lowmem_reserve[highest_zoneidx])
3567 		return false;
3568 
3569 	/* If this is an order-0 request then the watermark is fine */
3570 	if (!order)
3571 		return true;
3572 
3573 	/* For a high-order request, check at least one suitable page is free */
3574 	for (o = order; o < MAX_ORDER; o++) {
3575 		struct free_area *area = &z->free_area[o];
3576 		int mt;
3577 
3578 		if (!area->nr_free)
3579 			continue;
3580 
3581 		for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3582 			if (!free_area_empty(area, mt))
3583 				return true;
3584 		}
3585 
3586 #ifdef CONFIG_CMA
3587 		if ((alloc_flags & ALLOC_CMA) &&
3588 		    !free_area_empty(area, MIGRATE_CMA)) {
3589 			return true;
3590 		}
3591 #endif
3592 		if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC))
3593 			return true;
3594 	}
3595 	return false;
3596 }
3597 
3598 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3599 		      int highest_zoneidx, unsigned int alloc_flags)
3600 {
3601 	return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3602 					zone_page_state(z, NR_FREE_PAGES));
3603 }
3604 
3605 static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3606 				unsigned long mark, int highest_zoneidx,
3607 				unsigned int alloc_flags, gfp_t gfp_mask)
3608 {
3609 	long free_pages;
3610 
3611 	free_pages = zone_page_state(z, NR_FREE_PAGES);
3612 
3613 	/*
3614 	 * Fast check for order-0 only. If this fails then the reserves
3615 	 * need to be calculated.
3616 	 */
3617 	if (!order) {
3618 		long fast_free;
3619 
3620 		fast_free = free_pages;
3621 		fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags);
3622 		if (fast_free > mark + z->lowmem_reserve[highest_zoneidx])
3623 			return true;
3624 	}
3625 
3626 	if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3627 					free_pages))
3628 		return true;
3629 	/*
3630 	 * Ignore watermark boosting for GFP_ATOMIC order-0 allocations
3631 	 * when checking the min watermark. The min watermark is the
3632 	 * point where boosting is ignored so that kswapd is woken up
3633 	 * when below the low watermark.
3634 	 */
3635 	if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost
3636 		&& ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) {
3637 		mark = z->_watermark[WMARK_MIN];
3638 		return __zone_watermark_ok(z, order, mark, highest_zoneidx,
3639 					alloc_flags, free_pages);
3640 	}
3641 
3642 	return false;
3643 }
3644 
3645 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3646 			unsigned long mark, int highest_zoneidx)
3647 {
3648 	long free_pages = zone_page_state(z, NR_FREE_PAGES);
3649 
3650 	if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3651 		free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3652 
3653 	return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0,
3654 								free_pages);
3655 }
3656 
3657 #ifdef CONFIG_NUMA
3658 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3659 {
3660 	return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
3661 				node_reclaim_distance;
3662 }
3663 #else	/* CONFIG_NUMA */
3664 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3665 {
3666 	return true;
3667 }
3668 #endif	/* CONFIG_NUMA */
3669 
3670 /*
3671  * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid
3672  * fragmentation is subtle. If the preferred zone was HIGHMEM then
3673  * premature use of a lower zone may cause lowmem pressure problems that
3674  * are worse than fragmentation. If the next zone is ZONE_DMA then it is
3675  * probably too small. It only makes sense to spread allocations to avoid
3676  * fragmentation between the Normal and DMA32 zones.
3677  */
3678 static inline unsigned int
3679 alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask)
3680 {
3681 	unsigned int alloc_flags;
3682 
3683 	/*
3684 	 * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD
3685 	 * to save a branch.
3686 	 */
3687 	alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM);
3688 
3689 #ifdef CONFIG_ZONE_DMA32
3690 	if (!zone)
3691 		return alloc_flags;
3692 
3693 	if (zone_idx(zone) != ZONE_NORMAL)
3694 		return alloc_flags;
3695 
3696 	/*
3697 	 * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and
3698 	 * the pointer is within zone->zone_pgdat->node_zones[]. Also assume
3699 	 * on UMA that if Normal is populated then so is DMA32.
3700 	 */
3701 	BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1);
3702 	if (nr_online_nodes > 1 && !populated_zone(--zone))
3703 		return alloc_flags;
3704 
3705 	alloc_flags |= ALLOC_NOFRAGMENT;
3706 #endif /* CONFIG_ZONE_DMA32 */
3707 	return alloc_flags;
3708 }
3709 
3710 static inline unsigned int current_alloc_flags(gfp_t gfp_mask,
3711 					unsigned int alloc_flags)
3712 {
3713 #ifdef CONFIG_CMA
3714 	unsigned int pflags = current->flags;
3715 
3716 	if (!(pflags & PF_MEMALLOC_NOCMA) &&
3717 			gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE)
3718 		alloc_flags |= ALLOC_CMA;
3719 
3720 #endif
3721 	return alloc_flags;
3722 }
3723 
3724 /*
3725  * get_page_from_freelist goes through the zonelist trying to allocate
3726  * a page.
3727  */
3728 static struct page *
3729 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
3730 						const struct alloc_context *ac)
3731 {
3732 	struct zoneref *z;
3733 	struct zone *zone;
3734 	struct pglist_data *last_pgdat_dirty_limit = NULL;
3735 	bool no_fallback;
3736 
3737 retry:
3738 	/*
3739 	 * Scan zonelist, looking for a zone with enough free.
3740 	 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
3741 	 */
3742 	no_fallback = alloc_flags & ALLOC_NOFRAGMENT;
3743 	z = ac->preferred_zoneref;
3744 	for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx,
3745 					ac->nodemask) {
3746 		struct page *page;
3747 		unsigned long mark;
3748 
3749 		if (cpusets_enabled() &&
3750 			(alloc_flags & ALLOC_CPUSET) &&
3751 			!__cpuset_zone_allowed(zone, gfp_mask))
3752 				continue;
3753 		/*
3754 		 * When allocating a page cache page for writing, we
3755 		 * want to get it from a node that is within its dirty
3756 		 * limit, such that no single node holds more than its
3757 		 * proportional share of globally allowed dirty pages.
3758 		 * The dirty limits take into account the node's
3759 		 * lowmem reserves and high watermark so that kswapd
3760 		 * should be able to balance it without having to
3761 		 * write pages from its LRU list.
3762 		 *
3763 		 * XXX: For now, allow allocations to potentially
3764 		 * exceed the per-node dirty limit in the slowpath
3765 		 * (spread_dirty_pages unset) before going into reclaim,
3766 		 * which is important when on a NUMA setup the allowed
3767 		 * nodes are together not big enough to reach the
3768 		 * global limit.  The proper fix for these situations
3769 		 * will require awareness of nodes in the
3770 		 * dirty-throttling and the flusher threads.
3771 		 */
3772 		if (ac->spread_dirty_pages) {
3773 			if (last_pgdat_dirty_limit == zone->zone_pgdat)
3774 				continue;
3775 
3776 			if (!node_dirty_ok(zone->zone_pgdat)) {
3777 				last_pgdat_dirty_limit = zone->zone_pgdat;
3778 				continue;
3779 			}
3780 		}
3781 
3782 		if (no_fallback && nr_online_nodes > 1 &&
3783 		    zone != ac->preferred_zoneref->zone) {
3784 			int local_nid;
3785 
3786 			/*
3787 			 * If moving to a remote node, retry but allow
3788 			 * fragmenting fallbacks. Locality is more important
3789 			 * than fragmentation avoidance.
3790 			 */
3791 			local_nid = zone_to_nid(ac->preferred_zoneref->zone);
3792 			if (zone_to_nid(zone) != local_nid) {
3793 				alloc_flags &= ~ALLOC_NOFRAGMENT;
3794 				goto retry;
3795 			}
3796 		}
3797 
3798 		mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);
3799 		if (!zone_watermark_fast(zone, order, mark,
3800 				       ac->highest_zoneidx, alloc_flags,
3801 				       gfp_mask)) {
3802 			int ret;
3803 
3804 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3805 			/*
3806 			 * Watermark failed for this zone, but see if we can
3807 			 * grow this zone if it contains deferred pages.
3808 			 */
3809 			if (static_branch_unlikely(&deferred_pages)) {
3810 				if (_deferred_grow_zone(zone, order))
3811 					goto try_this_zone;
3812 			}
3813 #endif
3814 			/* Checked here to keep the fast path fast */
3815 			BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
3816 			if (alloc_flags & ALLOC_NO_WATERMARKS)
3817 				goto try_this_zone;
3818 
3819 			if (node_reclaim_mode == 0 ||
3820 			    !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
3821 				continue;
3822 
3823 			ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
3824 			switch (ret) {
3825 			case NODE_RECLAIM_NOSCAN:
3826 				/* did not scan */
3827 				continue;
3828 			case NODE_RECLAIM_FULL:
3829 				/* scanned but unreclaimable */
3830 				continue;
3831 			default:
3832 				/* did we reclaim enough */
3833 				if (zone_watermark_ok(zone, order, mark,
3834 					ac->highest_zoneidx, alloc_flags))
3835 					goto try_this_zone;
3836 
3837 				continue;
3838 			}
3839 		}
3840 
3841 try_this_zone:
3842 		page = rmqueue(ac->preferred_zoneref->zone, zone, order,
3843 				gfp_mask, alloc_flags, ac->migratetype);
3844 		if (page) {
3845 			prep_new_page(page, order, gfp_mask, alloc_flags);
3846 
3847 			/*
3848 			 * If this is a high-order atomic allocation then check
3849 			 * if the pageblock should be reserved for the future
3850 			 */
3851 			if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3852 				reserve_highatomic_pageblock(page, zone, order);
3853 
3854 			return page;
3855 		} else {
3856 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3857 			/* Try again if zone has deferred pages */
3858 			if (static_branch_unlikely(&deferred_pages)) {
3859 				if (_deferred_grow_zone(zone, order))
3860 					goto try_this_zone;
3861 			}
3862 #endif
3863 		}
3864 	}
3865 
3866 	/*
3867 	 * It's possible on a UMA machine to get through all zones that are
3868 	 * fragmented. If avoiding fragmentation, reset and try again.
3869 	 */
3870 	if (no_fallback) {
3871 		alloc_flags &= ~ALLOC_NOFRAGMENT;
3872 		goto retry;
3873 	}
3874 
3875 	return NULL;
3876 }
3877 
3878 static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
3879 {
3880 	unsigned int filter = SHOW_MEM_FILTER_NODES;
3881 
3882 	/*
3883 	 * This documents exceptions given to allocations in certain
3884 	 * contexts that are allowed to allocate outside current's set
3885 	 * of allowed nodes.
3886 	 */
3887 	if (!(gfp_mask & __GFP_NOMEMALLOC))
3888 		if (tsk_is_oom_victim(current) ||
3889 		    (current->flags & (PF_MEMALLOC | PF_EXITING)))
3890 			filter &= ~SHOW_MEM_FILTER_NODES;
3891 	if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
3892 		filter &= ~SHOW_MEM_FILTER_NODES;
3893 
3894 	show_mem(filter, nodemask);
3895 }
3896 
3897 void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
3898 {
3899 	struct va_format vaf;
3900 	va_list args;
3901 	static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1);
3902 
3903 	if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
3904 		return;
3905 
3906 	va_start(args, fmt);
3907 	vaf.fmt = fmt;
3908 	vaf.va = &args;
3909 	pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl",
3910 			current->comm, &vaf, gfp_mask, &gfp_mask,
3911 			nodemask_pr_args(nodemask));
3912 	va_end(args);
3913 
3914 	cpuset_print_current_mems_allowed();
3915 	pr_cont("\n");
3916 	dump_stack();
3917 	warn_alloc_show_mem(gfp_mask, nodemask);
3918 }
3919 
3920 static inline struct page *
3921 __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
3922 			      unsigned int alloc_flags,
3923 			      const struct alloc_context *ac)
3924 {
3925 	struct page *page;
3926 
3927 	page = get_page_from_freelist(gfp_mask, order,
3928 			alloc_flags|ALLOC_CPUSET, ac);
3929 	/*
3930 	 * fallback to ignore cpuset restriction if our nodes
3931 	 * are depleted
3932 	 */
3933 	if (!page)
3934 		page = get_page_from_freelist(gfp_mask, order,
3935 				alloc_flags, ac);
3936 
3937 	return page;
3938 }
3939 
3940 static inline struct page *
3941 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
3942 	const struct alloc_context *ac, unsigned long *did_some_progress)
3943 {
3944 	struct oom_control oc = {
3945 		.zonelist = ac->zonelist,
3946 		.nodemask = ac->nodemask,
3947 		.memcg = NULL,
3948 		.gfp_mask = gfp_mask,
3949 		.order = order,
3950 	};
3951 	struct page *page;
3952 
3953 	*did_some_progress = 0;
3954 
3955 	/*
3956 	 * Acquire the oom lock.  If that fails, somebody else is
3957 	 * making progress for us.
3958 	 */
3959 	if (!mutex_trylock(&oom_lock)) {
3960 		*did_some_progress = 1;
3961 		schedule_timeout_uninterruptible(1);
3962 		return NULL;
3963 	}
3964 
3965 	/*
3966 	 * Go through the zonelist yet one more time, keep very high watermark
3967 	 * here, this is only to catch a parallel oom killing, we must fail if
3968 	 * we're still under heavy pressure. But make sure that this reclaim
3969 	 * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
3970 	 * allocation which will never fail due to oom_lock already held.
3971 	 */
3972 	page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
3973 				      ~__GFP_DIRECT_RECLAIM, order,
3974 				      ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
3975 	if (page)
3976 		goto out;
3977 
3978 	/* Coredumps can quickly deplete all memory reserves */
3979 	if (current->flags & PF_DUMPCORE)
3980 		goto out;
3981 	/* The OOM killer will not help higher order allocs */
3982 	if (order > PAGE_ALLOC_COSTLY_ORDER)
3983 		goto out;
3984 	/*
3985 	 * We have already exhausted all our reclaim opportunities without any
3986 	 * success so it is time to admit defeat. We will skip the OOM killer
3987 	 * because it is very likely that the caller has a more reasonable
3988 	 * fallback than shooting a random task.
3989 	 *
3990 	 * The OOM killer may not free memory on a specific node.
3991 	 */
3992 	if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE))
3993 		goto out;
3994 	/* The OOM killer does not needlessly kill tasks for lowmem */
3995 	if (ac->highest_zoneidx < ZONE_NORMAL)
3996 		goto out;
3997 	if (pm_suspended_storage())
3998 		goto out;
3999 	/*
4000 	 * XXX: GFP_NOFS allocations should rather fail than rely on
4001 	 * other request to make a forward progress.
4002 	 * We are in an unfortunate situation where out_of_memory cannot
4003 	 * do much for this context but let's try it to at least get
4004 	 * access to memory reserved if the current task is killed (see
4005 	 * out_of_memory). Once filesystems are ready to handle allocation
4006 	 * failures more gracefully we should just bail out here.
4007 	 */
4008 
4009 	/* Exhausted what can be done so it's blame time */
4010 	if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
4011 		*did_some_progress = 1;
4012 
4013 		/*
4014 		 * Help non-failing allocations by giving them access to memory
4015 		 * reserves
4016 		 */
4017 		if (gfp_mask & __GFP_NOFAIL)
4018 			page = __alloc_pages_cpuset_fallback(gfp_mask, order,
4019 					ALLOC_NO_WATERMARKS, ac);
4020 	}
4021 out:
4022 	mutex_unlock(&oom_lock);
4023 	return page;
4024 }
4025 
4026 /*
4027  * Maximum number of compaction retries wit a progress before OOM
4028  * killer is consider as the only way to move forward.
4029  */
4030 #define MAX_COMPACT_RETRIES 16
4031 
4032 #ifdef CONFIG_COMPACTION
4033 /* Try memory compaction for high-order allocations before reclaim */
4034 static struct page *
4035 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4036 		unsigned int alloc_flags, const struct alloc_context *ac,
4037 		enum compact_priority prio, enum compact_result *compact_result)
4038 {
4039 	struct page *page = NULL;
4040 	unsigned long pflags;
4041 	unsigned int noreclaim_flag;
4042 
4043 	if (!order)
4044 		return NULL;
4045 
4046 	psi_memstall_enter(&pflags);
4047 	noreclaim_flag = memalloc_noreclaim_save();
4048 
4049 	*compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
4050 								prio, &page);
4051 
4052 	memalloc_noreclaim_restore(noreclaim_flag);
4053 	psi_memstall_leave(&pflags);
4054 
4055 	/*
4056 	 * At least in one zone compaction wasn't deferred or skipped, so let's
4057 	 * count a compaction stall
4058 	 */
4059 	count_vm_event(COMPACTSTALL);
4060 
4061 	/* Prep a captured page if available */
4062 	if (page)
4063 		prep_new_page(page, order, gfp_mask, alloc_flags);
4064 
4065 	/* Try get a page from the freelist if available */
4066 	if (!page)
4067 		page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4068 
4069 	if (page) {
4070 		struct zone *zone = page_zone(page);
4071 
4072 		zone->compact_blockskip_flush = false;
4073 		compaction_defer_reset(zone, order, true);
4074 		count_vm_event(COMPACTSUCCESS);
4075 		return page;
4076 	}
4077 
4078 	/*
4079 	 * It's bad if compaction run occurs and fails. The most likely reason
4080 	 * is that pages exist, but not enough to satisfy watermarks.
4081 	 */
4082 	count_vm_event(COMPACTFAIL);
4083 
4084 	cond_resched();
4085 
4086 	return NULL;
4087 }
4088 
4089 static inline bool
4090 should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
4091 		     enum compact_result compact_result,
4092 		     enum compact_priority *compact_priority,
4093 		     int *compaction_retries)
4094 {
4095 	int max_retries = MAX_COMPACT_RETRIES;
4096 	int min_priority;
4097 	bool ret = false;
4098 	int retries = *compaction_retries;
4099 	enum compact_priority priority = *compact_priority;
4100 
4101 	if (!order)
4102 		return false;
4103 
4104 	if (compaction_made_progress(compact_result))
4105 		(*compaction_retries)++;
4106 
4107 	/*
4108 	 * compaction considers all the zone as desperately out of memory
4109 	 * so it doesn't really make much sense to retry except when the
4110 	 * failure could be caused by insufficient priority
4111 	 */
4112 	if (compaction_failed(compact_result))
4113 		goto check_priority;
4114 
4115 	/*
4116 	 * compaction was skipped because there are not enough order-0 pages
4117 	 * to work with, so we retry only if it looks like reclaim can help.
4118 	 */
4119 	if (compaction_needs_reclaim(compact_result)) {
4120 		ret = compaction_zonelist_suitable(ac, order, alloc_flags);
4121 		goto out;
4122 	}
4123 
4124 	/*
4125 	 * make sure the compaction wasn't deferred or didn't bail out early
4126 	 * due to locks contention before we declare that we should give up.
4127 	 * But the next retry should use a higher priority if allowed, so
4128 	 * we don't just keep bailing out endlessly.
4129 	 */
4130 	if (compaction_withdrawn(compact_result)) {
4131 		goto check_priority;
4132 	}
4133 
4134 	/*
4135 	 * !costly requests are much more important than __GFP_RETRY_MAYFAIL
4136 	 * costly ones because they are de facto nofail and invoke OOM
4137 	 * killer to move on while costly can fail and users are ready
4138 	 * to cope with that. 1/4 retries is rather arbitrary but we
4139 	 * would need much more detailed feedback from compaction to
4140 	 * make a better decision.
4141 	 */
4142 	if (order > PAGE_ALLOC_COSTLY_ORDER)
4143 		max_retries /= 4;
4144 	if (*compaction_retries <= max_retries) {
4145 		ret = true;
4146 		goto out;
4147 	}
4148 
4149 	/*
4150 	 * Make sure there are attempts at the highest priority if we exhausted
4151 	 * all retries or failed at the lower priorities.
4152 	 */
4153 check_priority:
4154 	min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
4155 			MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
4156 
4157 	if (*compact_priority > min_priority) {
4158 		(*compact_priority)--;
4159 		*compaction_retries = 0;
4160 		ret = true;
4161 	}
4162 out:
4163 	trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
4164 	return ret;
4165 }
4166 #else
4167 static inline struct page *
4168 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4169 		unsigned int alloc_flags, const struct alloc_context *ac,
4170 		enum compact_priority prio, enum compact_result *compact_result)
4171 {
4172 	*compact_result = COMPACT_SKIPPED;
4173 	return NULL;
4174 }
4175 
4176 static inline bool
4177 should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
4178 		     enum compact_result compact_result,
4179 		     enum compact_priority *compact_priority,
4180 		     int *compaction_retries)
4181 {
4182 	struct zone *zone;
4183 	struct zoneref *z;
4184 
4185 	if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
4186 		return false;
4187 
4188 	/*
4189 	 * There are setups with compaction disabled which would prefer to loop
4190 	 * inside the allocator rather than hit the oom killer prematurely.
4191 	 * Let's give them a good hope and keep retrying while the order-0
4192 	 * watermarks are OK.
4193 	 */
4194 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4195 				ac->highest_zoneidx, ac->nodemask) {
4196 		if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
4197 					ac->highest_zoneidx, alloc_flags))
4198 			return true;
4199 	}
4200 	return false;
4201 }
4202 #endif /* CONFIG_COMPACTION */
4203 
4204 #ifdef CONFIG_LOCKDEP
4205 static struct lockdep_map __fs_reclaim_map =
4206 	STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
4207 
4208 static bool __need_fs_reclaim(gfp_t gfp_mask)
4209 {
4210 	gfp_mask = current_gfp_context(gfp_mask);
4211 
4212 	/* no reclaim without waiting on it */
4213 	if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
4214 		return false;
4215 
4216 	/* this guy won't enter reclaim */
4217 	if (current->flags & PF_MEMALLOC)
4218 		return false;
4219 
4220 	/* We're only interested __GFP_FS allocations for now */
4221 	if (!(gfp_mask & __GFP_FS))
4222 		return false;
4223 
4224 	if (gfp_mask & __GFP_NOLOCKDEP)
4225 		return false;
4226 
4227 	return true;
4228 }
4229 
4230 void __fs_reclaim_acquire(void)
4231 {
4232 	lock_map_acquire(&__fs_reclaim_map);
4233 }
4234 
4235 void __fs_reclaim_release(void)
4236 {
4237 	lock_map_release(&__fs_reclaim_map);
4238 }
4239 
4240 void fs_reclaim_acquire(gfp_t gfp_mask)
4241 {
4242 	if (__need_fs_reclaim(gfp_mask))
4243 		__fs_reclaim_acquire();
4244 }
4245 EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
4246 
4247 void fs_reclaim_release(gfp_t gfp_mask)
4248 {
4249 	if (__need_fs_reclaim(gfp_mask))
4250 		__fs_reclaim_release();
4251 }
4252 EXPORT_SYMBOL_GPL(fs_reclaim_release);
4253 #endif
4254 
4255 /* Perform direct synchronous page reclaim */
4256 static unsigned long
4257 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
4258 					const struct alloc_context *ac)
4259 {
4260 	unsigned int noreclaim_flag;
4261 	unsigned long pflags, progress;
4262 
4263 	cond_resched();
4264 
4265 	/* We now go into synchronous reclaim */
4266 	cpuset_memory_pressure_bump();
4267 	psi_memstall_enter(&pflags);
4268 	fs_reclaim_acquire(gfp_mask);
4269 	noreclaim_flag = memalloc_noreclaim_save();
4270 
4271 	progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
4272 								ac->nodemask);
4273 
4274 	memalloc_noreclaim_restore(noreclaim_flag);
4275 	fs_reclaim_release(gfp_mask);
4276 	psi_memstall_leave(&pflags);
4277 
4278 	cond_resched();
4279 
4280 	return progress;
4281 }
4282 
4283 /* The really slow allocator path where we enter direct reclaim */
4284 static inline struct page *
4285 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
4286 		unsigned int alloc_flags, const struct alloc_context *ac,
4287 		unsigned long *did_some_progress)
4288 {
4289 	struct page *page = NULL;
4290 	bool drained = false;
4291 
4292 	*did_some_progress = __perform_reclaim(gfp_mask, order, ac);
4293 	if (unlikely(!(*did_some_progress)))
4294 		return NULL;
4295 
4296 retry:
4297 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4298 
4299 	/*
4300 	 * If an allocation failed after direct reclaim, it could be because
4301 	 * pages are pinned on the per-cpu lists or in high alloc reserves.
4302 	 * Shrink them and try again
4303 	 */
4304 	if (!page && !drained) {
4305 		unreserve_highatomic_pageblock(ac, false);
4306 		drain_all_pages(NULL);
4307 		drained = true;
4308 		goto retry;
4309 	}
4310 
4311 	return page;
4312 }
4313 
4314 static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
4315 			     const struct alloc_context *ac)
4316 {
4317 	struct zoneref *z;
4318 	struct zone *zone;
4319 	pg_data_t *last_pgdat = NULL;
4320 	enum zone_type highest_zoneidx = ac->highest_zoneidx;
4321 
4322 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx,
4323 					ac->nodemask) {
4324 		if (last_pgdat != zone->zone_pgdat)
4325 			wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx);
4326 		last_pgdat = zone->zone_pgdat;
4327 	}
4328 }
4329 
4330 static inline unsigned int
4331 gfp_to_alloc_flags(gfp_t gfp_mask)
4332 {
4333 	unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
4334 
4335 	/*
4336 	 * __GFP_HIGH is assumed to be the same as ALLOC_HIGH
4337 	 * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD
4338 	 * to save two branches.
4339 	 */
4340 	BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
4341 	BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD);
4342 
4343 	/*
4344 	 * The caller may dip into page reserves a bit more if the caller
4345 	 * cannot run direct reclaim, or if the caller has realtime scheduling
4346 	 * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
4347 	 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
4348 	 */
4349 	alloc_flags |= (__force int)
4350 		(gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM));
4351 
4352 	if (gfp_mask & __GFP_ATOMIC) {
4353 		/*
4354 		 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
4355 		 * if it can't schedule.
4356 		 */
4357 		if (!(gfp_mask & __GFP_NOMEMALLOC))
4358 			alloc_flags |= ALLOC_HARDER;
4359 		/*
4360 		 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
4361 		 * comment for __cpuset_node_allowed().
4362 		 */
4363 		alloc_flags &= ~ALLOC_CPUSET;
4364 	} else if (unlikely(rt_task(current)) && !in_interrupt())
4365 		alloc_flags |= ALLOC_HARDER;
4366 
4367 	alloc_flags = current_alloc_flags(gfp_mask, alloc_flags);
4368 
4369 	return alloc_flags;
4370 }
4371 
4372 static bool oom_reserves_allowed(struct task_struct *tsk)
4373 {
4374 	if (!tsk_is_oom_victim(tsk))
4375 		return false;
4376 
4377 	/*
4378 	 * !MMU doesn't have oom reaper so give access to memory reserves
4379 	 * only to the thread with TIF_MEMDIE set
4380 	 */
4381 	if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
4382 		return false;
4383 
4384 	return true;
4385 }
4386 
4387 /*
4388  * Distinguish requests which really need access to full memory
4389  * reserves from oom victims which can live with a portion of it
4390  */
4391 static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
4392 {
4393 	if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
4394 		return 0;
4395 	if (gfp_mask & __GFP_MEMALLOC)
4396 		return ALLOC_NO_WATERMARKS;
4397 	if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
4398 		return ALLOC_NO_WATERMARKS;
4399 	if (!in_interrupt()) {
4400 		if (current->flags & PF_MEMALLOC)
4401 			return ALLOC_NO_WATERMARKS;
4402 		else if (oom_reserves_allowed(current))
4403 			return ALLOC_OOM;
4404 	}
4405 
4406 	return 0;
4407 }
4408 
4409 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
4410 {
4411 	return !!__gfp_pfmemalloc_flags(gfp_mask);
4412 }
4413 
4414 /*
4415  * Checks whether it makes sense to retry the reclaim to make a forward progress
4416  * for the given allocation request.
4417  *
4418  * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
4419  * without success, or when we couldn't even meet the watermark if we
4420  * reclaimed all remaining pages on the LRU lists.
4421  *
4422  * Returns true if a retry is viable or false to enter the oom path.
4423  */
4424 static inline bool
4425 should_reclaim_retry(gfp_t gfp_mask, unsigned order,
4426 		     struct alloc_context *ac, int alloc_flags,
4427 		     bool did_some_progress, int *no_progress_loops)
4428 {
4429 	struct zone *zone;
4430 	struct zoneref *z;
4431 	bool ret = false;
4432 
4433 	/*
4434 	 * Costly allocations might have made a progress but this doesn't mean
4435 	 * their order will become available due to high fragmentation so
4436 	 * always increment the no progress counter for them
4437 	 */
4438 	if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
4439 		*no_progress_loops = 0;
4440 	else
4441 		(*no_progress_loops)++;
4442 
4443 	/*
4444 	 * Make sure we converge to OOM if we cannot make any progress
4445 	 * several times in the row.
4446 	 */
4447 	if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
4448 		/* Before OOM, exhaust highatomic_reserve */
4449 		return unreserve_highatomic_pageblock(ac, true);
4450 	}
4451 
4452 	/*
4453 	 * Keep reclaiming pages while there is a chance this will lead
4454 	 * somewhere.  If none of the target zones can satisfy our allocation
4455 	 * request even if all reclaimable pages are considered then we are
4456 	 * screwed and have to go OOM.
4457 	 */
4458 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4459 				ac->highest_zoneidx, ac->nodemask) {
4460 		unsigned long available;
4461 		unsigned long reclaimable;
4462 		unsigned long min_wmark = min_wmark_pages(zone);
4463 		bool wmark;
4464 
4465 		available = reclaimable = zone_reclaimable_pages(zone);
4466 		available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
4467 
4468 		/*
4469 		 * Would the allocation succeed if we reclaimed all
4470 		 * reclaimable pages?
4471 		 */
4472 		wmark = __zone_watermark_ok(zone, order, min_wmark,
4473 				ac->highest_zoneidx, alloc_flags, available);
4474 		trace_reclaim_retry_zone(z, order, reclaimable,
4475 				available, min_wmark, *no_progress_loops, wmark);
4476 		if (wmark) {
4477 			/*
4478 			 * If we didn't make any progress and have a lot of
4479 			 * dirty + writeback pages then we should wait for
4480 			 * an IO to complete to slow down the reclaim and
4481 			 * prevent from pre mature OOM
4482 			 */
4483 			if (!did_some_progress) {
4484 				unsigned long write_pending;
4485 
4486 				write_pending = zone_page_state_snapshot(zone,
4487 							NR_ZONE_WRITE_PENDING);
4488 
4489 				if (2 * write_pending > reclaimable) {
4490 					congestion_wait(BLK_RW_ASYNC, HZ/10);
4491 					return true;
4492 				}
4493 			}
4494 
4495 			ret = true;
4496 			goto out;
4497 		}
4498 	}
4499 
4500 out:
4501 	/*
4502 	 * Memory allocation/reclaim might be called from a WQ context and the
4503 	 * current implementation of the WQ concurrency control doesn't
4504 	 * recognize that a particular WQ is congested if the worker thread is
4505 	 * looping without ever sleeping. Therefore we have to do a short sleep
4506 	 * here rather than calling cond_resched().
4507 	 */
4508 	if (current->flags & PF_WQ_WORKER)
4509 		schedule_timeout_uninterruptible(1);
4510 	else
4511 		cond_resched();
4512 	return ret;
4513 }
4514 
4515 static inline bool
4516 check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4517 {
4518 	/*
4519 	 * It's possible that cpuset's mems_allowed and the nodemask from
4520 	 * mempolicy don't intersect. This should be normally dealt with by
4521 	 * policy_nodemask(), but it's possible to race with cpuset update in
4522 	 * such a way the check therein was true, and then it became false
4523 	 * before we got our cpuset_mems_cookie here.
4524 	 * This assumes that for all allocations, ac->nodemask can come only
4525 	 * from MPOL_BIND mempolicy (whose documented semantics is to be ignored
4526 	 * when it does not intersect with the cpuset restrictions) or the
4527 	 * caller can deal with a violated nodemask.
4528 	 */
4529 	if (cpusets_enabled() && ac->nodemask &&
4530 			!cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4531 		ac->nodemask = NULL;
4532 		return true;
4533 	}
4534 
4535 	/*
4536 	 * When updating a task's mems_allowed or mempolicy nodemask, it is
4537 	 * possible to race with parallel threads in such a way that our
4538 	 * allocation can fail while the mask is being updated. If we are about
4539 	 * to fail, check if the cpuset changed during allocation and if so,
4540 	 * retry.
4541 	 */
4542 	if (read_mems_allowed_retry(cpuset_mems_cookie))
4543 		return true;
4544 
4545 	return false;
4546 }
4547 
4548 static inline struct page *
4549 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4550 						struct alloc_context *ac)
4551 {
4552 	bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4553 	const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4554 	struct page *page = NULL;
4555 	unsigned int alloc_flags;
4556 	unsigned long did_some_progress;
4557 	enum compact_priority compact_priority;
4558 	enum compact_result compact_result;
4559 	int compaction_retries;
4560 	int no_progress_loops;
4561 	unsigned int cpuset_mems_cookie;
4562 	int reserve_flags;
4563 
4564 	/*
4565 	 * We also sanity check to catch abuse of atomic reserves being used by
4566 	 * callers that are not in atomic context.
4567 	 */
4568 	if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4569 				(__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4570 		gfp_mask &= ~__GFP_ATOMIC;
4571 
4572 retry_cpuset:
4573 	compaction_retries = 0;
4574 	no_progress_loops = 0;
4575 	compact_priority = DEF_COMPACT_PRIORITY;
4576 	cpuset_mems_cookie = read_mems_allowed_begin();
4577 
4578 	/*
4579 	 * The fast path uses conservative alloc_flags to succeed only until
4580 	 * kswapd needs to be woken up, and to avoid the cost of setting up
4581 	 * alloc_flags precisely. So we do that now.
4582 	 */
4583 	alloc_flags = gfp_to_alloc_flags(gfp_mask);
4584 
4585 	/*
4586 	 * We need to recalculate the starting point for the zonelist iterator
4587 	 * because we might have used different nodemask in the fast path, or
4588 	 * there was a cpuset modification and we are retrying - otherwise we
4589 	 * could end up iterating over non-eligible zones endlessly.
4590 	 */
4591 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4592 					ac->highest_zoneidx, ac->nodemask);
4593 	if (!ac->preferred_zoneref->zone)
4594 		goto nopage;
4595 
4596 	if (alloc_flags & ALLOC_KSWAPD)
4597 		wake_all_kswapds(order, gfp_mask, ac);
4598 
4599 	/*
4600 	 * The adjusted alloc_flags might result in immediate success, so try
4601 	 * that first
4602 	 */
4603 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4604 	if (page)
4605 		goto got_pg;
4606 
4607 	/*
4608 	 * For costly allocations, try direct compaction first, as it's likely
4609 	 * that we have enough base pages and don't need to reclaim. For non-
4610 	 * movable high-order allocations, do that as well, as compaction will
4611 	 * try prevent permanent fragmentation by migrating from blocks of the
4612 	 * same migratetype.
4613 	 * Don't try this for allocations that are allowed to ignore
4614 	 * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen.
4615 	 */
4616 	if (can_direct_reclaim &&
4617 			(costly_order ||
4618 			   (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4619 			&& !gfp_pfmemalloc_allowed(gfp_mask)) {
4620 		page = __alloc_pages_direct_compact(gfp_mask, order,
4621 						alloc_flags, ac,
4622 						INIT_COMPACT_PRIORITY,
4623 						&compact_result);
4624 		if (page)
4625 			goto got_pg;
4626 
4627 		/*
4628 		 * Checks for costly allocations with __GFP_NORETRY, which
4629 		 * includes some THP page fault allocations
4630 		 */
4631 		if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4632 			/*
4633 			 * If allocating entire pageblock(s) and compaction
4634 			 * failed because all zones are below low watermarks
4635 			 * or is prohibited because it recently failed at this
4636 			 * order, fail immediately unless the allocator has
4637 			 * requested compaction and reclaim retry.
4638 			 *
4639 			 * Reclaim is
4640 			 *  - potentially very expensive because zones are far
4641 			 *    below their low watermarks or this is part of very
4642 			 *    bursty high order allocations,
4643 			 *  - not guaranteed to help because isolate_freepages()
4644 			 *    may not iterate over freed pages as part of its
4645 			 *    linear scan, and
4646 			 *  - unlikely to make entire pageblocks free on its
4647 			 *    own.
4648 			 */
4649 			if (compact_result == COMPACT_SKIPPED ||
4650 			    compact_result == COMPACT_DEFERRED)
4651 				goto nopage;
4652 
4653 			/*
4654 			 * Looks like reclaim/compaction is worth trying, but
4655 			 * sync compaction could be very expensive, so keep
4656 			 * using async compaction.
4657 			 */
4658 			compact_priority = INIT_COMPACT_PRIORITY;
4659 		}
4660 	}
4661 
4662 retry:
4663 	/* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
4664 	if (alloc_flags & ALLOC_KSWAPD)
4665 		wake_all_kswapds(order, gfp_mask, ac);
4666 
4667 	reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
4668 	if (reserve_flags)
4669 		alloc_flags = current_alloc_flags(gfp_mask, reserve_flags);
4670 
4671 	/*
4672 	 * Reset the nodemask and zonelist iterators if memory policies can be
4673 	 * ignored. These allocations are high priority and system rather than
4674 	 * user oriented.
4675 	 */
4676 	if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
4677 		ac->nodemask = NULL;
4678 		ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4679 					ac->highest_zoneidx, ac->nodemask);
4680 	}
4681 
4682 	/* Attempt with potentially adjusted zonelist and alloc_flags */
4683 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4684 	if (page)
4685 		goto got_pg;
4686 
4687 	/* Caller is not willing to reclaim, we can't balance anything */
4688 	if (!can_direct_reclaim)
4689 		goto nopage;
4690 
4691 	/* Avoid recursion of direct reclaim */
4692 	if (current->flags & PF_MEMALLOC)
4693 		goto nopage;
4694 
4695 	/* Try direct reclaim and then allocating */
4696 	page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
4697 							&did_some_progress);
4698 	if (page)
4699 		goto got_pg;
4700 
4701 	/* Try direct compaction and then allocating */
4702 	page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
4703 					compact_priority, &compact_result);
4704 	if (page)
4705 		goto got_pg;
4706 
4707 	/* Do not loop if specifically requested */
4708 	if (gfp_mask & __GFP_NORETRY)
4709 		goto nopage;
4710 
4711 	/*
4712 	 * Do not retry costly high order allocations unless they are
4713 	 * __GFP_RETRY_MAYFAIL
4714 	 */
4715 	if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
4716 		goto nopage;
4717 
4718 	if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
4719 				 did_some_progress > 0, &no_progress_loops))
4720 		goto retry;
4721 
4722 	/*
4723 	 * It doesn't make any sense to retry for the compaction if the order-0
4724 	 * reclaim is not able to make any progress because the current
4725 	 * implementation of the compaction depends on the sufficient amount
4726 	 * of free memory (see __compaction_suitable)
4727 	 */
4728 	if (did_some_progress > 0 &&
4729 			should_compact_retry(ac, order, alloc_flags,
4730 				compact_result, &compact_priority,
4731 				&compaction_retries))
4732 		goto retry;
4733 
4734 
4735 	/* Deal with possible cpuset update races before we start OOM killing */
4736 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
4737 		goto retry_cpuset;
4738 
4739 	/* Reclaim has failed us, start killing things */
4740 	page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
4741 	if (page)
4742 		goto got_pg;
4743 
4744 	/* Avoid allocations with no watermarks from looping endlessly */
4745 	if (tsk_is_oom_victim(current) &&
4746 	    (alloc_flags & ALLOC_OOM ||
4747 	     (gfp_mask & __GFP_NOMEMALLOC)))
4748 		goto nopage;
4749 
4750 	/* Retry as long as the OOM killer is making progress */
4751 	if (did_some_progress) {
4752 		no_progress_loops = 0;
4753 		goto retry;
4754 	}
4755 
4756 nopage:
4757 	/* Deal with possible cpuset update races before we fail */
4758 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
4759 		goto retry_cpuset;
4760 
4761 	/*
4762 	 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
4763 	 * we always retry
4764 	 */
4765 	if (gfp_mask & __GFP_NOFAIL) {
4766 		/*
4767 		 * All existing users of the __GFP_NOFAIL are blockable, so warn
4768 		 * of any new users that actually require GFP_NOWAIT
4769 		 */
4770 		if (WARN_ON_ONCE(!can_direct_reclaim))
4771 			goto fail;
4772 
4773 		/*
4774 		 * PF_MEMALLOC request from this context is rather bizarre
4775 		 * because we cannot reclaim anything and only can loop waiting
4776 		 * for somebody to do a work for us
4777 		 */
4778 		WARN_ON_ONCE(current->flags & PF_MEMALLOC);
4779 
4780 		/*
4781 		 * non failing costly orders are a hard requirement which we
4782 		 * are not prepared for much so let's warn about these users
4783 		 * so that we can identify them and convert them to something
4784 		 * else.
4785 		 */
4786 		WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
4787 
4788 		/*
4789 		 * Help non-failing allocations by giving them access to memory
4790 		 * reserves but do not use ALLOC_NO_WATERMARKS because this
4791 		 * could deplete whole memory reserves which would just make
4792 		 * the situation worse
4793 		 */
4794 		page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
4795 		if (page)
4796 			goto got_pg;
4797 
4798 		cond_resched();
4799 		goto retry;
4800 	}
4801 fail:
4802 	warn_alloc(gfp_mask, ac->nodemask,
4803 			"page allocation failure: order:%u", order);
4804 got_pg:
4805 	return page;
4806 }
4807 
4808 static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
4809 		int preferred_nid, nodemask_t *nodemask,
4810 		struct alloc_context *ac, gfp_t *alloc_mask,
4811 		unsigned int *alloc_flags)
4812 {
4813 	ac->highest_zoneidx = gfp_zone(gfp_mask);
4814 	ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
4815 	ac->nodemask = nodemask;
4816 	ac->migratetype = gfp_migratetype(gfp_mask);
4817 
4818 	if (cpusets_enabled()) {
4819 		*alloc_mask |= __GFP_HARDWALL;
4820 		/*
4821 		 * When we are in the interrupt context, it is irrelevant
4822 		 * to the current task context. It means that any node ok.
4823 		 */
4824 		if (!in_interrupt() && !ac->nodemask)
4825 			ac->nodemask = &cpuset_current_mems_allowed;
4826 		else
4827 			*alloc_flags |= ALLOC_CPUSET;
4828 	}
4829 
4830 	fs_reclaim_acquire(gfp_mask);
4831 	fs_reclaim_release(gfp_mask);
4832 
4833 	might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
4834 
4835 	if (should_fail_alloc_page(gfp_mask, order))
4836 		return false;
4837 
4838 	*alloc_flags = current_alloc_flags(gfp_mask, *alloc_flags);
4839 
4840 	/* Dirty zone balancing only done in the fast path */
4841 	ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
4842 
4843 	/*
4844 	 * The preferred zone is used for statistics but crucially it is
4845 	 * also used as the starting point for the zonelist iterator. It
4846 	 * may get reset for allocations that ignore memory policies.
4847 	 */
4848 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4849 					ac->highest_zoneidx, ac->nodemask);
4850 
4851 	return true;
4852 }
4853 
4854 /*
4855  * This is the 'heart' of the zoned buddy allocator.
4856  */
4857 struct page *
4858 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
4859 							nodemask_t *nodemask)
4860 {
4861 	struct page *page;
4862 	unsigned int alloc_flags = ALLOC_WMARK_LOW;
4863 	gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
4864 	struct alloc_context ac = { };
4865 
4866 	/*
4867 	 * There are several places where we assume that the order value is sane
4868 	 * so bail out early if the request is out of bound.
4869 	 */
4870 	if (unlikely(order >= MAX_ORDER)) {
4871 		WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
4872 		return NULL;
4873 	}
4874 
4875 	gfp_mask &= gfp_allowed_mask;
4876 	alloc_mask = gfp_mask;
4877 	if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))
4878 		return NULL;
4879 
4880 	/*
4881 	 * Forbid the first pass from falling back to types that fragment
4882 	 * memory until all local zones are considered.
4883 	 */
4884 	alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp_mask);
4885 
4886 	/* First allocation attempt */
4887 	page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
4888 	if (likely(page))
4889 		goto out;
4890 
4891 	/*
4892 	 * Apply scoped allocation constraints. This is mainly about GFP_NOFS
4893 	 * resp. GFP_NOIO which has to be inherited for all allocation requests
4894 	 * from a particular context which has been marked by
4895 	 * memalloc_no{fs,io}_{save,restore}.
4896 	 */
4897 	alloc_mask = current_gfp_context(gfp_mask);
4898 	ac.spread_dirty_pages = false;
4899 
4900 	/*
4901 	 * Restore the original nodemask if it was potentially replaced with
4902 	 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
4903 	 */
4904 	ac.nodemask = nodemask;
4905 
4906 	page = __alloc_pages_slowpath(alloc_mask, order, &ac);
4907 
4908 out:
4909 	if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
4910 	    unlikely(__memcg_kmem_charge_page(page, gfp_mask, order) != 0)) {
4911 		__free_pages(page, order);
4912 		page = NULL;
4913 	}
4914 
4915 	trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
4916 
4917 	return page;
4918 }
4919 EXPORT_SYMBOL(__alloc_pages_nodemask);
4920 
4921 /*
4922  * Common helper functions. Never use with __GFP_HIGHMEM because the returned
4923  * address cannot represent highmem pages. Use alloc_pages and then kmap if
4924  * you need to access high mem.
4925  */
4926 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
4927 {
4928 	struct page *page;
4929 
4930 	page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order);
4931 	if (!page)
4932 		return 0;
4933 	return (unsigned long) page_address(page);
4934 }
4935 EXPORT_SYMBOL(__get_free_pages);
4936 
4937 unsigned long get_zeroed_page(gfp_t gfp_mask)
4938 {
4939 	return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
4940 }
4941 EXPORT_SYMBOL(get_zeroed_page);
4942 
4943 static inline void free_the_page(struct page *page, unsigned int order)
4944 {
4945 	if (order == 0)		/* Via pcp? */
4946 		free_unref_page(page);
4947 	else
4948 		__free_pages_ok(page, order);
4949 }
4950 
4951 void __free_pages(struct page *page, unsigned int order)
4952 {
4953 	if (put_page_testzero(page))
4954 		free_the_page(page, order);
4955 	else if (!PageHead(page))
4956 		while (order-- > 0)
4957 			free_the_page(page + (1 << order), order);
4958 }
4959 EXPORT_SYMBOL(__free_pages);
4960 
4961 void free_pages(unsigned long addr, unsigned int order)
4962 {
4963 	if (addr != 0) {
4964 		VM_BUG_ON(!virt_addr_valid((void *)addr));
4965 		__free_pages(virt_to_page((void *)addr), order);
4966 	}
4967 }
4968 
4969 EXPORT_SYMBOL(free_pages);
4970 
4971 /*
4972  * Page Fragment:
4973  *  An arbitrary-length arbitrary-offset area of memory which resides
4974  *  within a 0 or higher order page.  Multiple fragments within that page
4975  *  are individually refcounted, in the page's reference counter.
4976  *
4977  * The page_frag functions below provide a simple allocation framework for
4978  * page fragments.  This is used by the network stack and network device
4979  * drivers to provide a backing region of memory for use as either an
4980  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4981  */
4982 static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
4983 					     gfp_t gfp_mask)
4984 {
4985 	struct page *page = NULL;
4986 	gfp_t gfp = gfp_mask;
4987 
4988 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4989 	gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
4990 		    __GFP_NOMEMALLOC;
4991 	page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
4992 				PAGE_FRAG_CACHE_MAX_ORDER);
4993 	nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
4994 #endif
4995 	if (unlikely(!page))
4996 		page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
4997 
4998 	nc->va = page ? page_address(page) : NULL;
4999 
5000 	return page;
5001 }
5002 
5003 void __page_frag_cache_drain(struct page *page, unsigned int count)
5004 {
5005 	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
5006 
5007 	if (page_ref_sub_and_test(page, count))
5008 		free_the_page(page, compound_order(page));
5009 }
5010 EXPORT_SYMBOL(__page_frag_cache_drain);
5011 
5012 void *page_frag_alloc(struct page_frag_cache *nc,
5013 		      unsigned int fragsz, gfp_t gfp_mask)
5014 {
5015 	unsigned int size = PAGE_SIZE;
5016 	struct page *page;
5017 	int offset;
5018 
5019 	if (unlikely(!nc->va)) {
5020 refill:
5021 		page = __page_frag_cache_refill(nc, gfp_mask);
5022 		if (!page)
5023 			return NULL;
5024 
5025 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5026 		/* if size can vary use size else just use PAGE_SIZE */
5027 		size = nc->size;
5028 #endif
5029 		/* Even if we own the page, we do not use atomic_set().
5030 		 * This would break get_page_unless_zero() users.
5031 		 */
5032 		page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
5033 
5034 		/* reset page count bias and offset to start of new frag */
5035 		nc->pfmemalloc = page_is_pfmemalloc(page);
5036 		nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5037 		nc->offset = size;
5038 	}
5039 
5040 	offset = nc->offset - fragsz;
5041 	if (unlikely(offset < 0)) {
5042 		page = virt_to_page(nc->va);
5043 
5044 		if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
5045 			goto refill;
5046 
5047 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5048 		/* if size can vary use size else just use PAGE_SIZE */
5049 		size = nc->size;
5050 #endif
5051 		/* OK, page count is 0, we can safely set it */
5052 		set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1);
5053 
5054 		/* reset page count bias and offset to start of new frag */
5055 		nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5056 		offset = size - fragsz;
5057 	}
5058 
5059 	nc->pagecnt_bias--;
5060 	nc->offset = offset;
5061 
5062 	return nc->va + offset;
5063 }
5064 EXPORT_SYMBOL(page_frag_alloc);
5065 
5066 /*
5067  * Frees a page fragment allocated out of either a compound or order 0 page.
5068  */
5069 void page_frag_free(void *addr)
5070 {
5071 	struct page *page = virt_to_head_page(addr);
5072 
5073 	if (unlikely(put_page_testzero(page)))
5074 		free_the_page(page, compound_order(page));
5075 }
5076 EXPORT_SYMBOL(page_frag_free);
5077 
5078 static void *make_alloc_exact(unsigned long addr, unsigned int order,
5079 		size_t size)
5080 {
5081 	if (addr) {
5082 		unsigned long alloc_end = addr + (PAGE_SIZE << order);
5083 		unsigned long used = addr + PAGE_ALIGN(size);
5084 
5085 		split_page(virt_to_page((void *)addr), order);
5086 		while (used < alloc_end) {
5087 			free_page(used);
5088 			used += PAGE_SIZE;
5089 		}
5090 	}
5091 	return (void *)addr;
5092 }
5093 
5094 /**
5095  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
5096  * @size: the number of bytes to allocate
5097  * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP
5098  *
5099  * This function is similar to alloc_pages(), except that it allocates the
5100  * minimum number of pages to satisfy the request.  alloc_pages() can only
5101  * allocate memory in power-of-two pages.
5102  *
5103  * This function is also limited by MAX_ORDER.
5104  *
5105  * Memory allocated by this function must be released by free_pages_exact().
5106  *
5107  * Return: pointer to the allocated area or %NULL in case of error.
5108  */
5109 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
5110 {
5111 	unsigned int order = get_order(size);
5112 	unsigned long addr;
5113 
5114 	if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
5115 		gfp_mask &= ~__GFP_COMP;
5116 
5117 	addr = __get_free_pages(gfp_mask, order);
5118 	return make_alloc_exact(addr, order, size);
5119 }
5120 EXPORT_SYMBOL(alloc_pages_exact);
5121 
5122 /**
5123  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
5124  *			   pages on a node.
5125  * @nid: the preferred node ID where memory should be allocated
5126  * @size: the number of bytes to allocate
5127  * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP
5128  *
5129  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
5130  * back.
5131  *
5132  * Return: pointer to the allocated area or %NULL in case of error.
5133  */
5134 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
5135 {
5136 	unsigned int order = get_order(size);
5137 	struct page *p;
5138 
5139 	if (WARN_ON_ONCE(gfp_mask & __GFP_COMP))
5140 		gfp_mask &= ~__GFP_COMP;
5141 
5142 	p = alloc_pages_node(nid, gfp_mask, order);
5143 	if (!p)
5144 		return NULL;
5145 	return make_alloc_exact((unsigned long)page_address(p), order, size);
5146 }
5147 
5148 /**
5149  * free_pages_exact - release memory allocated via alloc_pages_exact()
5150  * @virt: the value returned by alloc_pages_exact.
5151  * @size: size of allocation, same value as passed to alloc_pages_exact().
5152  *
5153  * Release the memory allocated by a previous call to alloc_pages_exact.
5154  */
5155 void free_pages_exact(void *virt, size_t size)
5156 {
5157 	unsigned long addr = (unsigned long)virt;
5158 	unsigned long end = addr + PAGE_ALIGN(size);
5159 
5160 	while (addr < end) {
5161 		free_page(addr);
5162 		addr += PAGE_SIZE;
5163 	}
5164 }
5165 EXPORT_SYMBOL(free_pages_exact);
5166 
5167 /**
5168  * nr_free_zone_pages - count number of pages beyond high watermark
5169  * @offset: The zone index of the highest zone
5170  *
5171  * nr_free_zone_pages() counts the number of pages which are beyond the
5172  * high watermark within all zones at or below a given zone index.  For each
5173  * zone, the number of pages is calculated as:
5174  *
5175  *     nr_free_zone_pages = managed_pages - high_pages
5176  *
5177  * Return: number of pages beyond high watermark.
5178  */
5179 static unsigned long nr_free_zone_pages(int offset)
5180 {
5181 	struct zoneref *z;
5182 	struct zone *zone;
5183 
5184 	/* Just pick one node, since fallback list is circular */
5185 	unsigned long sum = 0;
5186 
5187 	struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
5188 
5189 	for_each_zone_zonelist(zone, z, zonelist, offset) {
5190 		unsigned long size = zone_managed_pages(zone);
5191 		unsigned long high = high_wmark_pages(zone);
5192 		if (size > high)
5193 			sum += size - high;
5194 	}
5195 
5196 	return sum;
5197 }
5198 
5199 /**
5200  * nr_free_buffer_pages - count number of pages beyond high watermark
5201  *
5202  * nr_free_buffer_pages() counts the number of pages which are beyond the high
5203  * watermark within ZONE_DMA and ZONE_NORMAL.
5204  *
5205  * Return: number of pages beyond high watermark within ZONE_DMA and
5206  * ZONE_NORMAL.
5207  */
5208 unsigned long nr_free_buffer_pages(void)
5209 {
5210 	return nr_free_zone_pages(gfp_zone(GFP_USER));
5211 }
5212 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
5213 
5214 static inline void show_node(struct zone *zone)
5215 {
5216 	if (IS_ENABLED(CONFIG_NUMA))
5217 		printk("Node %d ", zone_to_nid(zone));
5218 }
5219 
5220 long si_mem_available(void)
5221 {
5222 	long available;
5223 	unsigned long pagecache;
5224 	unsigned long wmark_low = 0;
5225 	unsigned long pages[NR_LRU_LISTS];
5226 	unsigned long reclaimable;
5227 	struct zone *zone;
5228 	int lru;
5229 
5230 	for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
5231 		pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
5232 
5233 	for_each_zone(zone)
5234 		wmark_low += low_wmark_pages(zone);
5235 
5236 	/*
5237 	 * Estimate the amount of memory available for userspace allocations,
5238 	 * without causing swapping.
5239 	 */
5240 	available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
5241 
5242 	/*
5243 	 * Not all the page cache can be freed, otherwise the system will
5244 	 * start swapping. Assume at least half of the page cache, or the
5245 	 * low watermark worth of cache, needs to stay.
5246 	 */
5247 	pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
5248 	pagecache -= min(pagecache / 2, wmark_low);
5249 	available += pagecache;
5250 
5251 	/*
5252 	 * Part of the reclaimable slab and other kernel memory consists of
5253 	 * items that are in use, and cannot be freed. Cap this estimate at the
5254 	 * low watermark.
5255 	 */
5256 	reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) +
5257 		global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
5258 	available += reclaimable - min(reclaimable / 2, wmark_low);
5259 
5260 	if (available < 0)
5261 		available = 0;
5262 	return available;
5263 }
5264 EXPORT_SYMBOL_GPL(si_mem_available);
5265 
5266 void si_meminfo(struct sysinfo *val)
5267 {
5268 	val->totalram = totalram_pages();
5269 	val->sharedram = global_node_page_state(NR_SHMEM);
5270 	val->freeram = global_zone_page_state(NR_FREE_PAGES);
5271 	val->bufferram = nr_blockdev_pages();
5272 	val->totalhigh = totalhigh_pages();
5273 	val->freehigh = nr_free_highpages();
5274 	val->mem_unit = PAGE_SIZE;
5275 }
5276 
5277 EXPORT_SYMBOL(si_meminfo);
5278 
5279 #ifdef CONFIG_NUMA
5280 void si_meminfo_node(struct sysinfo *val, int nid)
5281 {
5282 	int zone_type;		/* needs to be signed */
5283 	unsigned long managed_pages = 0;
5284 	unsigned long managed_highpages = 0;
5285 	unsigned long free_highpages = 0;
5286 	pg_data_t *pgdat = NODE_DATA(nid);
5287 
5288 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
5289 		managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]);
5290 	val->totalram = managed_pages;
5291 	val->sharedram = node_page_state(pgdat, NR_SHMEM);
5292 	val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
5293 #ifdef CONFIG_HIGHMEM
5294 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
5295 		struct zone *zone = &pgdat->node_zones[zone_type];
5296 
5297 		if (is_highmem(zone)) {
5298 			managed_highpages += zone_managed_pages(zone);
5299 			free_highpages += zone_page_state(zone, NR_FREE_PAGES);
5300 		}
5301 	}
5302 	val->totalhigh = managed_highpages;
5303 	val->freehigh = free_highpages;
5304 #else
5305 	val->totalhigh = managed_highpages;
5306 	val->freehigh = free_highpages;
5307 #endif
5308 	val->mem_unit = PAGE_SIZE;
5309 }
5310 #endif
5311 
5312 /*
5313  * Determine whether the node should be displayed or not, depending on whether
5314  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
5315  */
5316 static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
5317 {
5318 	if (!(flags & SHOW_MEM_FILTER_NODES))
5319 		return false;
5320 
5321 	/*
5322 	 * no node mask - aka implicit memory numa policy. Do not bother with
5323 	 * the synchronization - read_mems_allowed_begin - because we do not
5324 	 * have to be precise here.
5325 	 */
5326 	if (!nodemask)
5327 		nodemask = &cpuset_current_mems_allowed;
5328 
5329 	return !node_isset(nid, *nodemask);
5330 }
5331 
5332 #define K(x) ((x) << (PAGE_SHIFT-10))
5333 
5334 static void show_migration_types(unsigned char type)
5335 {
5336 	static const char types[MIGRATE_TYPES] = {
5337 		[MIGRATE_UNMOVABLE]	= 'U',
5338 		[MIGRATE_MOVABLE]	= 'M',
5339 		[MIGRATE_RECLAIMABLE]	= 'E',
5340 		[MIGRATE_HIGHATOMIC]	= 'H',
5341 #ifdef CONFIG_CMA
5342 		[MIGRATE_CMA]		= 'C',
5343 #endif
5344 #ifdef CONFIG_MEMORY_ISOLATION
5345 		[MIGRATE_ISOLATE]	= 'I',
5346 #endif
5347 	};
5348 	char tmp[MIGRATE_TYPES + 1];
5349 	char *p = tmp;
5350 	int i;
5351 
5352 	for (i = 0; i < MIGRATE_TYPES; i++) {
5353 		if (type & (1 << i))
5354 			*p++ = types[i];
5355 	}
5356 
5357 	*p = '\0';
5358 	printk(KERN_CONT "(%s) ", tmp);
5359 }
5360 
5361 /*
5362  * Show free area list (used inside shift_scroll-lock stuff)
5363  * We also calculate the percentage fragmentation. We do this by counting the
5364  * memory on each free list with the exception of the first item on the list.
5365  *
5366  * Bits in @filter:
5367  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
5368  *   cpuset.
5369  */
5370 void show_free_areas(unsigned int filter, nodemask_t *nodemask)
5371 {
5372 	unsigned long free_pcp = 0;
5373 	int cpu;
5374 	struct zone *zone;
5375 	pg_data_t *pgdat;
5376 
5377 	for_each_populated_zone(zone) {
5378 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5379 			continue;
5380 
5381 		for_each_online_cpu(cpu)
5382 			free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5383 	}
5384 
5385 	printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
5386 		" active_file:%lu inactive_file:%lu isolated_file:%lu\n"
5387 		" unevictable:%lu dirty:%lu writeback:%lu\n"
5388 		" slab_reclaimable:%lu slab_unreclaimable:%lu\n"
5389 		" mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
5390 		" free:%lu free_pcp:%lu free_cma:%lu\n",
5391 		global_node_page_state(NR_ACTIVE_ANON),
5392 		global_node_page_state(NR_INACTIVE_ANON),
5393 		global_node_page_state(NR_ISOLATED_ANON),
5394 		global_node_page_state(NR_ACTIVE_FILE),
5395 		global_node_page_state(NR_INACTIVE_FILE),
5396 		global_node_page_state(NR_ISOLATED_FILE),
5397 		global_node_page_state(NR_UNEVICTABLE),
5398 		global_node_page_state(NR_FILE_DIRTY),
5399 		global_node_page_state(NR_WRITEBACK),
5400 		global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B),
5401 		global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B),
5402 		global_node_page_state(NR_FILE_MAPPED),
5403 		global_node_page_state(NR_SHMEM),
5404 		global_zone_page_state(NR_PAGETABLE),
5405 		global_zone_page_state(NR_BOUNCE),
5406 		global_zone_page_state(NR_FREE_PAGES),
5407 		free_pcp,
5408 		global_zone_page_state(NR_FREE_CMA_PAGES));
5409 
5410 	for_each_online_pgdat(pgdat) {
5411 		if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
5412 			continue;
5413 
5414 		printk("Node %d"
5415 			" active_anon:%lukB"
5416 			" inactive_anon:%lukB"
5417 			" active_file:%lukB"
5418 			" inactive_file:%lukB"
5419 			" unevictable:%lukB"
5420 			" isolated(anon):%lukB"
5421 			" isolated(file):%lukB"
5422 			" mapped:%lukB"
5423 			" dirty:%lukB"
5424 			" writeback:%lukB"
5425 			" shmem:%lukB"
5426 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5427 			" shmem_thp: %lukB"
5428 			" shmem_pmdmapped: %lukB"
5429 			" anon_thp: %lukB"
5430 #endif
5431 			" writeback_tmp:%lukB"
5432 			" kernel_stack:%lukB"
5433 #ifdef CONFIG_SHADOW_CALL_STACK
5434 			" shadow_call_stack:%lukB"
5435 #endif
5436 			" all_unreclaimable? %s"
5437 			"\n",
5438 			pgdat->node_id,
5439 			K(node_page_state(pgdat, NR_ACTIVE_ANON)),
5440 			K(node_page_state(pgdat, NR_INACTIVE_ANON)),
5441 			K(node_page_state(pgdat, NR_ACTIVE_FILE)),
5442 			K(node_page_state(pgdat, NR_INACTIVE_FILE)),
5443 			K(node_page_state(pgdat, NR_UNEVICTABLE)),
5444 			K(node_page_state(pgdat, NR_ISOLATED_ANON)),
5445 			K(node_page_state(pgdat, NR_ISOLATED_FILE)),
5446 			K(node_page_state(pgdat, NR_FILE_MAPPED)),
5447 			K(node_page_state(pgdat, NR_FILE_DIRTY)),
5448 			K(node_page_state(pgdat, NR_WRITEBACK)),
5449 			K(node_page_state(pgdat, NR_SHMEM)),
5450 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5451 			K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
5452 			K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
5453 					* HPAGE_PMD_NR),
5454 			K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
5455 #endif
5456 			K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
5457 			node_page_state(pgdat, NR_KERNEL_STACK_KB),
5458 #ifdef CONFIG_SHADOW_CALL_STACK
5459 			node_page_state(pgdat, NR_KERNEL_SCS_KB),
5460 #endif
5461 			pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
5462 				"yes" : "no");
5463 	}
5464 
5465 	for_each_populated_zone(zone) {
5466 		int i;
5467 
5468 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5469 			continue;
5470 
5471 		free_pcp = 0;
5472 		for_each_online_cpu(cpu)
5473 			free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5474 
5475 		show_node(zone);
5476 		printk(KERN_CONT
5477 			"%s"
5478 			" free:%lukB"
5479 			" min:%lukB"
5480 			" low:%lukB"
5481 			" high:%lukB"
5482 			" reserved_highatomic:%luKB"
5483 			" active_anon:%lukB"
5484 			" inactive_anon:%lukB"
5485 			" active_file:%lukB"
5486 			" inactive_file:%lukB"
5487 			" unevictable:%lukB"
5488 			" writepending:%lukB"
5489 			" present:%lukB"
5490 			" managed:%lukB"
5491 			" mlocked:%lukB"
5492 			" pagetables:%lukB"
5493 			" bounce:%lukB"
5494 			" free_pcp:%lukB"
5495 			" local_pcp:%ukB"
5496 			" free_cma:%lukB"
5497 			"\n",
5498 			zone->name,
5499 			K(zone_page_state(zone, NR_FREE_PAGES)),
5500 			K(min_wmark_pages(zone)),
5501 			K(low_wmark_pages(zone)),
5502 			K(high_wmark_pages(zone)),
5503 			K(zone->nr_reserved_highatomic),
5504 			K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
5505 			K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
5506 			K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
5507 			K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
5508 			K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
5509 			K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
5510 			K(zone->present_pages),
5511 			K(zone_managed_pages(zone)),
5512 			K(zone_page_state(zone, NR_MLOCK)),
5513 			K(zone_page_state(zone, NR_PAGETABLE)),
5514 			K(zone_page_state(zone, NR_BOUNCE)),
5515 			K(free_pcp),
5516 			K(this_cpu_read(zone->pageset->pcp.count)),
5517 			K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
5518 		printk("lowmem_reserve[]:");
5519 		for (i = 0; i < MAX_NR_ZONES; i++)
5520 			printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
5521 		printk(KERN_CONT "\n");
5522 	}
5523 
5524 	for_each_populated_zone(zone) {
5525 		unsigned int order;
5526 		unsigned long nr[MAX_ORDER], flags, total = 0;
5527 		unsigned char types[MAX_ORDER];
5528 
5529 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5530 			continue;
5531 		show_node(zone);
5532 		printk(KERN_CONT "%s: ", zone->name);
5533 
5534 		spin_lock_irqsave(&zone->lock, flags);
5535 		for (order = 0; order < MAX_ORDER; order++) {
5536 			struct free_area *area = &zone->free_area[order];
5537 			int type;
5538 
5539 			nr[order] = area->nr_free;
5540 			total += nr[order] << order;
5541 
5542 			types[order] = 0;
5543 			for (type = 0; type < MIGRATE_TYPES; type++) {
5544 				if (!free_area_empty(area, type))
5545 					types[order] |= 1 << type;
5546 			}
5547 		}
5548 		spin_unlock_irqrestore(&zone->lock, flags);
5549 		for (order = 0; order < MAX_ORDER; order++) {
5550 			printk(KERN_CONT "%lu*%lukB ",
5551 			       nr[order], K(1UL) << order);
5552 			if (nr[order])
5553 				show_migration_types(types[order]);
5554 		}
5555 		printk(KERN_CONT "= %lukB\n", K(total));
5556 	}
5557 
5558 	hugetlb_show_meminfo();
5559 
5560 	printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
5561 
5562 	show_swap_cache_info();
5563 }
5564 
5565 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
5566 {
5567 	zoneref->zone = zone;
5568 	zoneref->zone_idx = zone_idx(zone);
5569 }
5570 
5571 /*
5572  * Builds allocation fallback zone lists.
5573  *
5574  * Add all populated zones of a node to the zonelist.
5575  */
5576 static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
5577 {
5578 	struct zone *zone;
5579 	enum zone_type zone_type = MAX_NR_ZONES;
5580 	int nr_zones = 0;
5581 
5582 	do {
5583 		zone_type--;
5584 		zone = pgdat->node_zones + zone_type;
5585 		if (managed_zone(zone)) {
5586 			zoneref_set_zone(zone, &zonerefs[nr_zones++]);
5587 			check_highest_zone(zone_type);
5588 		}
5589 	} while (zone_type);
5590 
5591 	return nr_zones;
5592 }
5593 
5594 #ifdef CONFIG_NUMA
5595 
5596 static int __parse_numa_zonelist_order(char *s)
5597 {
5598 	/*
5599 	 * We used to support different zonlists modes but they turned
5600 	 * out to be just not useful. Let's keep the warning in place
5601 	 * if somebody still use the cmd line parameter so that we do
5602 	 * not fail it silently
5603 	 */
5604 	if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
5605 		pr_warn("Ignoring unsupported numa_zonelist_order value:  %s\n", s);
5606 		return -EINVAL;
5607 	}
5608 	return 0;
5609 }
5610 
5611 char numa_zonelist_order[] = "Node";
5612 
5613 /*
5614  * sysctl handler for numa_zonelist_order
5615  */
5616 int numa_zonelist_order_handler(struct ctl_table *table, int write,
5617 		void *buffer, size_t *length, loff_t *ppos)
5618 {
5619 	if (write)
5620 		return __parse_numa_zonelist_order(buffer);
5621 	return proc_dostring(table, write, buffer, length, ppos);
5622 }
5623 
5624 
5625 #define MAX_NODE_LOAD (nr_online_nodes)
5626 static int node_load[MAX_NUMNODES];
5627 
5628 /**
5629  * find_next_best_node - find the next node that should appear in a given node's fallback list
5630  * @node: node whose fallback list we're appending
5631  * @used_node_mask: nodemask_t of already used nodes
5632  *
5633  * We use a number of factors to determine which is the next node that should
5634  * appear on a given node's fallback list.  The node should not have appeared
5635  * already in @node's fallback list, and it should be the next closest node
5636  * according to the distance array (which contains arbitrary distance values
5637  * from each node to each node in the system), and should also prefer nodes
5638  * with no CPUs, since presumably they'll have very little allocation pressure
5639  * on them otherwise.
5640  *
5641  * Return: node id of the found node or %NUMA_NO_NODE if no node is found.
5642  */
5643 static int find_next_best_node(int node, nodemask_t *used_node_mask)
5644 {
5645 	int n, val;
5646 	int min_val = INT_MAX;
5647 	int best_node = NUMA_NO_NODE;
5648 
5649 	/* Use the local node if we haven't already */
5650 	if (!node_isset(node, *used_node_mask)) {
5651 		node_set(node, *used_node_mask);
5652 		return node;
5653 	}
5654 
5655 	for_each_node_state(n, N_MEMORY) {
5656 
5657 		/* Don't want a node to appear more than once */
5658 		if (node_isset(n, *used_node_mask))
5659 			continue;
5660 
5661 		/* Use the distance array to find the distance */
5662 		val = node_distance(node, n);
5663 
5664 		/* Penalize nodes under us ("prefer the next node") */
5665 		val += (n < node);
5666 
5667 		/* Give preference to headless and unused nodes */
5668 		if (!cpumask_empty(cpumask_of_node(n)))
5669 			val += PENALTY_FOR_NODE_WITH_CPUS;
5670 
5671 		/* Slight preference for less loaded node */
5672 		val *= (MAX_NODE_LOAD*MAX_NUMNODES);
5673 		val += node_load[n];
5674 
5675 		if (val < min_val) {
5676 			min_val = val;
5677 			best_node = n;
5678 		}
5679 	}
5680 
5681 	if (best_node >= 0)
5682 		node_set(best_node, *used_node_mask);
5683 
5684 	return best_node;
5685 }
5686 
5687 
5688 /*
5689  * Build zonelists ordered by node and zones within node.
5690  * This results in maximum locality--normal zone overflows into local
5691  * DMA zone, if any--but risks exhausting DMA zone.
5692  */
5693 static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
5694 		unsigned nr_nodes)
5695 {
5696 	struct zoneref *zonerefs;
5697 	int i;
5698 
5699 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5700 
5701 	for (i = 0; i < nr_nodes; i++) {
5702 		int nr_zones;
5703 
5704 		pg_data_t *node = NODE_DATA(node_order[i]);
5705 
5706 		nr_zones = build_zonerefs_node(node, zonerefs);
5707 		zonerefs += nr_zones;
5708 	}
5709 	zonerefs->zone = NULL;
5710 	zonerefs->zone_idx = 0;
5711 }
5712 
5713 /*
5714  * Build gfp_thisnode zonelists
5715  */
5716 static void build_thisnode_zonelists(pg_data_t *pgdat)
5717 {
5718 	struct zoneref *zonerefs;
5719 	int nr_zones;
5720 
5721 	zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
5722 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
5723 	zonerefs += nr_zones;
5724 	zonerefs->zone = NULL;
5725 	zonerefs->zone_idx = 0;
5726 }
5727 
5728 /*
5729  * Build zonelists ordered by zone and nodes within zones.
5730  * This results in conserving DMA zone[s] until all Normal memory is
5731  * exhausted, but results in overflowing to remote node while memory
5732  * may still exist in local DMA zone.
5733  */
5734 
5735 static void build_zonelists(pg_data_t *pgdat)
5736 {
5737 	static int node_order[MAX_NUMNODES];
5738 	int node, load, nr_nodes = 0;
5739 	nodemask_t used_mask = NODE_MASK_NONE;
5740 	int local_node, prev_node;
5741 
5742 	/* NUMA-aware ordering of nodes */
5743 	local_node = pgdat->node_id;
5744 	load = nr_online_nodes;
5745 	prev_node = local_node;
5746 
5747 	memset(node_order, 0, sizeof(node_order));
5748 	while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5749 		/*
5750 		 * We don't want to pressure a particular node.
5751 		 * So adding penalty to the first node in same
5752 		 * distance group to make it round-robin.
5753 		 */
5754 		if (node_distance(local_node, node) !=
5755 		    node_distance(local_node, prev_node))
5756 			node_load[node] = load;
5757 
5758 		node_order[nr_nodes++] = node;
5759 		prev_node = node;
5760 		load--;
5761 	}
5762 
5763 	build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
5764 	build_thisnode_zonelists(pgdat);
5765 }
5766 
5767 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5768 /*
5769  * Return node id of node used for "local" allocations.
5770  * I.e., first node id of first zone in arg node's generic zonelist.
5771  * Used for initializing percpu 'numa_mem', which is used primarily
5772  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
5773  */
5774 int local_memory_node(int node)
5775 {
5776 	struct zoneref *z;
5777 
5778 	z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
5779 				   gfp_zone(GFP_KERNEL),
5780 				   NULL);
5781 	return zone_to_nid(z->zone);
5782 }
5783 #endif
5784 
5785 static void setup_min_unmapped_ratio(void);
5786 static void setup_min_slab_ratio(void);
5787 #else	/* CONFIG_NUMA */
5788 
5789 static void build_zonelists(pg_data_t *pgdat)
5790 {
5791 	int node, local_node;
5792 	struct zoneref *zonerefs;
5793 	int nr_zones;
5794 
5795 	local_node = pgdat->node_id;
5796 
5797 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5798 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
5799 	zonerefs += nr_zones;
5800 
5801 	/*
5802 	 * Now we build the zonelist so that it contains the zones
5803 	 * of all the other nodes.
5804 	 * We don't want to pressure a particular node, so when
5805 	 * building the zones for node N, we make sure that the
5806 	 * zones coming right after the local ones are those from
5807 	 * node N+1 (modulo N)
5808 	 */
5809 	for (node = local_node + 1; node < MAX_NUMNODES; node++) {
5810 		if (!node_online(node))
5811 			continue;
5812 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5813 		zonerefs += nr_zones;
5814 	}
5815 	for (node = 0; node < local_node; node++) {
5816 		if (!node_online(node))
5817 			continue;
5818 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5819 		zonerefs += nr_zones;
5820 	}
5821 
5822 	zonerefs->zone = NULL;
5823 	zonerefs->zone_idx = 0;
5824 }
5825 
5826 #endif	/* CONFIG_NUMA */
5827 
5828 /*
5829  * Boot pageset table. One per cpu which is going to be used for all
5830  * zones and all nodes. The parameters will be set in such a way
5831  * that an item put on a list will immediately be handed over to
5832  * the buddy list. This is safe since pageset manipulation is done
5833  * with interrupts disabled.
5834  *
5835  * The boot_pagesets must be kept even after bootup is complete for
5836  * unused processors and/or zones. They do play a role for bootstrapping
5837  * hotplugged processors.
5838  *
5839  * zoneinfo_show() and maybe other functions do
5840  * not check if the processor is online before following the pageset pointer.
5841  * Other parts of the kernel may not check if the zone is available.
5842  */
5843 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
5844 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
5845 static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
5846 
5847 static void __build_all_zonelists(void *data)
5848 {
5849 	int nid;
5850 	int __maybe_unused cpu;
5851 	pg_data_t *self = data;
5852 	static DEFINE_SPINLOCK(lock);
5853 
5854 	spin_lock(&lock);
5855 
5856 #ifdef CONFIG_NUMA
5857 	memset(node_load, 0, sizeof(node_load));
5858 #endif
5859 
5860 	/*
5861 	 * This node is hotadded and no memory is yet present.   So just
5862 	 * building zonelists is fine - no need to touch other nodes.
5863 	 */
5864 	if (self && !node_online(self->node_id)) {
5865 		build_zonelists(self);
5866 	} else {
5867 		for_each_online_node(nid) {
5868 			pg_data_t *pgdat = NODE_DATA(nid);
5869 
5870 			build_zonelists(pgdat);
5871 		}
5872 
5873 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5874 		/*
5875 		 * We now know the "local memory node" for each node--
5876 		 * i.e., the node of the first zone in the generic zonelist.
5877 		 * Set up numa_mem percpu variable for on-line cpus.  During
5878 		 * boot, only the boot cpu should be on-line;  we'll init the
5879 		 * secondary cpus' numa_mem as they come on-line.  During
5880 		 * node/memory hotplug, we'll fixup all on-line cpus.
5881 		 */
5882 		for_each_online_cpu(cpu)
5883 			set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
5884 #endif
5885 	}
5886 
5887 	spin_unlock(&lock);
5888 }
5889 
5890 static noinline void __init
5891 build_all_zonelists_init(void)
5892 {
5893 	int cpu;
5894 
5895 	__build_all_zonelists(NULL);
5896 
5897 	/*
5898 	 * Initialize the boot_pagesets that are going to be used
5899 	 * for bootstrapping processors. The real pagesets for
5900 	 * each zone will be allocated later when the per cpu
5901 	 * allocator is available.
5902 	 *
5903 	 * boot_pagesets are used also for bootstrapping offline
5904 	 * cpus if the system is already booted because the pagesets
5905 	 * are needed to initialize allocators on a specific cpu too.
5906 	 * F.e. the percpu allocator needs the page allocator which
5907 	 * needs the percpu allocator in order to allocate its pagesets
5908 	 * (a chicken-egg dilemma).
5909 	 */
5910 	for_each_possible_cpu(cpu)
5911 		setup_pageset(&per_cpu(boot_pageset, cpu), 0);
5912 
5913 	mminit_verify_zonelist();
5914 	cpuset_init_current_mems_allowed();
5915 }
5916 
5917 /*
5918  * unless system_state == SYSTEM_BOOTING.
5919  *
5920  * __ref due to call of __init annotated helper build_all_zonelists_init
5921  * [protected by SYSTEM_BOOTING].
5922  */
5923 void __ref build_all_zonelists(pg_data_t *pgdat)
5924 {
5925 	unsigned long vm_total_pages;
5926 
5927 	if (system_state == SYSTEM_BOOTING) {
5928 		build_all_zonelists_init();
5929 	} else {
5930 		__build_all_zonelists(pgdat);
5931 		/* cpuset refresh routine should be here */
5932 	}
5933 	/* Get the number of free pages beyond high watermark in all zones. */
5934 	vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
5935 	/*
5936 	 * Disable grouping by mobility if the number of pages in the
5937 	 * system is too low to allow the mechanism to work. It would be
5938 	 * more accurate, but expensive to check per-zone. This check is
5939 	 * made on memory-hotadd so a system can start with mobility
5940 	 * disabled and enable it later
5941 	 */
5942 	if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
5943 		page_group_by_mobility_disabled = 1;
5944 	else
5945 		page_group_by_mobility_disabled = 0;
5946 
5947 	pr_info("Built %u zonelists, mobility grouping %s.  Total pages: %ld\n",
5948 		nr_online_nodes,
5949 		page_group_by_mobility_disabled ? "off" : "on",
5950 		vm_total_pages);
5951 #ifdef CONFIG_NUMA
5952 	pr_info("Policy zone: %s\n", zone_names[policy_zone]);
5953 #endif
5954 }
5955 
5956 /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */
5957 static bool __meminit
5958 overlap_memmap_init(unsigned long zone, unsigned long *pfn)
5959 {
5960 	static struct memblock_region *r;
5961 
5962 	if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5963 		if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
5964 			for_each_mem_region(r) {
5965 				if (*pfn < memblock_region_memory_end_pfn(r))
5966 					break;
5967 			}
5968 		}
5969 		if (*pfn >= memblock_region_memory_base_pfn(r) &&
5970 		    memblock_is_mirror(r)) {
5971 			*pfn = memblock_region_memory_end_pfn(r);
5972 			return true;
5973 		}
5974 	}
5975 	return false;
5976 }
5977 
5978 /*
5979  * Initially all pages are reserved - free ones are freed
5980  * up by memblock_free_all() once the early boot process is
5981  * done. Non-atomic initialization, single-pass.
5982  */
5983 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
5984 		unsigned long start_pfn, enum meminit_context context,
5985 		struct vmem_altmap *altmap)
5986 {
5987 	unsigned long pfn, end_pfn = start_pfn + size;
5988 	struct page *page;
5989 
5990 	if (highest_memmap_pfn < end_pfn - 1)
5991 		highest_memmap_pfn = end_pfn - 1;
5992 
5993 #ifdef CONFIG_ZONE_DEVICE
5994 	/*
5995 	 * Honor reservation requested by the driver for this ZONE_DEVICE
5996 	 * memory. We limit the total number of pages to initialize to just
5997 	 * those that might contain the memory mapping. We will defer the
5998 	 * ZONE_DEVICE page initialization until after we have released
5999 	 * the hotplug lock.
6000 	 */
6001 	if (zone == ZONE_DEVICE) {
6002 		if (!altmap)
6003 			return;
6004 
6005 		if (start_pfn == altmap->base_pfn)
6006 			start_pfn += altmap->reserve;
6007 		end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6008 	}
6009 #endif
6010 
6011 	for (pfn = start_pfn; pfn < end_pfn; ) {
6012 		/*
6013 		 * There can be holes in boot-time mem_map[]s handed to this
6014 		 * function.  They do not exist on hotplugged memory.
6015 		 */
6016 		if (context == MEMINIT_EARLY) {
6017 			if (overlap_memmap_init(zone, &pfn))
6018 				continue;
6019 			if (defer_init(nid, pfn, end_pfn))
6020 				break;
6021 		}
6022 
6023 		page = pfn_to_page(pfn);
6024 		__init_single_page(page, pfn, zone, nid);
6025 		if (context == MEMINIT_HOTPLUG)
6026 			__SetPageReserved(page);
6027 
6028 		/*
6029 		 * Mark the block movable so that blocks are reserved for
6030 		 * movable at startup. This will force kernel allocations
6031 		 * to reserve their blocks rather than leaking throughout
6032 		 * the address space during boot when many long-lived
6033 		 * kernel allocations are made.
6034 		 *
6035 		 * bitmap is created for zone's valid pfn range. but memmap
6036 		 * can be created for invalid pages (for alignment)
6037 		 * check here not to call set_pageblock_migratetype() against
6038 		 * pfn out of zone.
6039 		 */
6040 		if (!(pfn & (pageblock_nr_pages - 1))) {
6041 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
6042 			cond_resched();
6043 		}
6044 		pfn++;
6045 	}
6046 }
6047 
6048 #ifdef CONFIG_ZONE_DEVICE
6049 void __ref memmap_init_zone_device(struct zone *zone,
6050 				   unsigned long start_pfn,
6051 				   unsigned long nr_pages,
6052 				   struct dev_pagemap *pgmap)
6053 {
6054 	unsigned long pfn, end_pfn = start_pfn + nr_pages;
6055 	struct pglist_data *pgdat = zone->zone_pgdat;
6056 	struct vmem_altmap *altmap = pgmap_altmap(pgmap);
6057 	unsigned long zone_idx = zone_idx(zone);
6058 	unsigned long start = jiffies;
6059 	int nid = pgdat->node_id;
6060 
6061 	if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE))
6062 		return;
6063 
6064 	/*
6065 	 * The call to memmap_init_zone should have already taken care
6066 	 * of the pages reserved for the memmap, so we can just jump to
6067 	 * the end of that region and start processing the device pages.
6068 	 */
6069 	if (altmap) {
6070 		start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6071 		nr_pages = end_pfn - start_pfn;
6072 	}
6073 
6074 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
6075 		struct page *page = pfn_to_page(pfn);
6076 
6077 		__init_single_page(page, pfn, zone_idx, nid);
6078 
6079 		/*
6080 		 * Mark page reserved as it will need to wait for onlining
6081 		 * phase for it to be fully associated with a zone.
6082 		 *
6083 		 * We can use the non-atomic __set_bit operation for setting
6084 		 * the flag as we are still initializing the pages.
6085 		 */
6086 		__SetPageReserved(page);
6087 
6088 		/*
6089 		 * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer
6090 		 * and zone_device_data.  It is a bug if a ZONE_DEVICE page is
6091 		 * ever freed or placed on a driver-private list.
6092 		 */
6093 		page->pgmap = pgmap;
6094 		page->zone_device_data = NULL;
6095 
6096 		/*
6097 		 * Mark the block movable so that blocks are reserved for
6098 		 * movable at startup. This will force kernel allocations
6099 		 * to reserve their blocks rather than leaking throughout
6100 		 * the address space during boot when many long-lived
6101 		 * kernel allocations are made.
6102 		 *
6103 		 * bitmap is created for zone's valid pfn range. but memmap
6104 		 * can be created for invalid pages (for alignment)
6105 		 * check here not to call set_pageblock_migratetype() against
6106 		 * pfn out of zone.
6107 		 *
6108 		 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
6109 		 * because this is done early in section_activate()
6110 		 */
6111 		if (!(pfn & (pageblock_nr_pages - 1))) {
6112 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
6113 			cond_resched();
6114 		}
6115 	}
6116 
6117 	pr_info("%s initialised %lu pages in %ums\n", __func__,
6118 		nr_pages, jiffies_to_msecs(jiffies - start));
6119 }
6120 
6121 #endif
6122 static void __meminit zone_init_free_lists(struct zone *zone)
6123 {
6124 	unsigned int order, t;
6125 	for_each_migratetype_order(order, t) {
6126 		INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
6127 		zone->free_area[order].nr_free = 0;
6128 	}
6129 }
6130 
6131 void __meminit __weak memmap_init(unsigned long size, int nid,
6132 				  unsigned long zone,
6133 				  unsigned long range_start_pfn)
6134 {
6135 	unsigned long start_pfn, end_pfn;
6136 	unsigned long range_end_pfn = range_start_pfn + size;
6137 	int i;
6138 
6139 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6140 		start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
6141 		end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
6142 
6143 		if (end_pfn > start_pfn) {
6144 			size = end_pfn - start_pfn;
6145 			memmap_init_zone(size, nid, zone, start_pfn,
6146 					 MEMINIT_EARLY, NULL);
6147 		}
6148 	}
6149 }
6150 
6151 static int zone_batchsize(struct zone *zone)
6152 {
6153 #ifdef CONFIG_MMU
6154 	int batch;
6155 
6156 	/*
6157 	 * The per-cpu-pages pools are set to around 1000th of the
6158 	 * size of the zone.
6159 	 */
6160 	batch = zone_managed_pages(zone) / 1024;
6161 	/* But no more than a meg. */
6162 	if (batch * PAGE_SIZE > 1024 * 1024)
6163 		batch = (1024 * 1024) / PAGE_SIZE;
6164 	batch /= 4;		/* We effectively *= 4 below */
6165 	if (batch < 1)
6166 		batch = 1;
6167 
6168 	/*
6169 	 * Clamp the batch to a 2^n - 1 value. Having a power
6170 	 * of 2 value was found to be more likely to have
6171 	 * suboptimal cache aliasing properties in some cases.
6172 	 *
6173 	 * For example if 2 tasks are alternately allocating
6174 	 * batches of pages, one task can end up with a lot
6175 	 * of pages of one half of the possible page colors
6176 	 * and the other with pages of the other colors.
6177 	 */
6178 	batch = rounddown_pow_of_two(batch + batch/2) - 1;
6179 
6180 	return batch;
6181 
6182 #else
6183 	/* The deferral and batching of frees should be suppressed under NOMMU
6184 	 * conditions.
6185 	 *
6186 	 * The problem is that NOMMU needs to be able to allocate large chunks
6187 	 * of contiguous memory as there's no hardware page translation to
6188 	 * assemble apparent contiguous memory from discontiguous pages.
6189 	 *
6190 	 * Queueing large contiguous runs of pages for batching, however,
6191 	 * causes the pages to actually be freed in smaller chunks.  As there
6192 	 * can be a significant delay between the individual batches being
6193 	 * recycled, this leads to the once large chunks of space being
6194 	 * fragmented and becoming unavailable for high-order allocations.
6195 	 */
6196 	return 0;
6197 #endif
6198 }
6199 
6200 /*
6201  * pcp->high and pcp->batch values are related and dependent on one another:
6202  * ->batch must never be higher then ->high.
6203  * The following function updates them in a safe manner without read side
6204  * locking.
6205  *
6206  * Any new users of pcp->batch and pcp->high should ensure they can cope with
6207  * those fields changing asynchronously (acording to the above rule).
6208  *
6209  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
6210  * outside of boot time (or some other assurance that no concurrent updaters
6211  * exist).
6212  */
6213 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
6214 		unsigned long batch)
6215 {
6216        /* start with a fail safe value for batch */
6217 	pcp->batch = 1;
6218 	smp_wmb();
6219 
6220        /* Update high, then batch, in order */
6221 	pcp->high = high;
6222 	smp_wmb();
6223 
6224 	pcp->batch = batch;
6225 }
6226 
6227 /* a companion to pageset_set_high() */
6228 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
6229 {
6230 	pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
6231 }
6232 
6233 static void pageset_init(struct per_cpu_pageset *p)
6234 {
6235 	struct per_cpu_pages *pcp;
6236 	int migratetype;
6237 
6238 	memset(p, 0, sizeof(*p));
6239 
6240 	pcp = &p->pcp;
6241 	for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
6242 		INIT_LIST_HEAD(&pcp->lists[migratetype]);
6243 }
6244 
6245 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
6246 {
6247 	pageset_init(p);
6248 	pageset_set_batch(p, batch);
6249 }
6250 
6251 /*
6252  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
6253  * to the value high for the pageset p.
6254  */
6255 static void pageset_set_high(struct per_cpu_pageset *p,
6256 				unsigned long high)
6257 {
6258 	unsigned long batch = max(1UL, high / 4);
6259 	if ((high / 4) > (PAGE_SHIFT * 8))
6260 		batch = PAGE_SHIFT * 8;
6261 
6262 	pageset_update(&p->pcp, high, batch);
6263 }
6264 
6265 static void pageset_set_high_and_batch(struct zone *zone,
6266 				       struct per_cpu_pageset *pcp)
6267 {
6268 	if (percpu_pagelist_fraction)
6269 		pageset_set_high(pcp,
6270 			(zone_managed_pages(zone) /
6271 				percpu_pagelist_fraction));
6272 	else
6273 		pageset_set_batch(pcp, zone_batchsize(zone));
6274 }
6275 
6276 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
6277 {
6278 	struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
6279 
6280 	pageset_init(pcp);
6281 	pageset_set_high_and_batch(zone, pcp);
6282 }
6283 
6284 void __meminit setup_zone_pageset(struct zone *zone)
6285 {
6286 	int cpu;
6287 	zone->pageset = alloc_percpu(struct per_cpu_pageset);
6288 	for_each_possible_cpu(cpu)
6289 		zone_pageset_init(zone, cpu);
6290 }
6291 
6292 /*
6293  * Allocate per cpu pagesets and initialize them.
6294  * Before this call only boot pagesets were available.
6295  */
6296 void __init setup_per_cpu_pageset(void)
6297 {
6298 	struct pglist_data *pgdat;
6299 	struct zone *zone;
6300 	int __maybe_unused cpu;
6301 
6302 	for_each_populated_zone(zone)
6303 		setup_zone_pageset(zone);
6304 
6305 #ifdef CONFIG_NUMA
6306 	/*
6307 	 * Unpopulated zones continue using the boot pagesets.
6308 	 * The numa stats for these pagesets need to be reset.
6309 	 * Otherwise, they will end up skewing the stats of
6310 	 * the nodes these zones are associated with.
6311 	 */
6312 	for_each_possible_cpu(cpu) {
6313 		struct per_cpu_pageset *pcp = &per_cpu(boot_pageset, cpu);
6314 		memset(pcp->vm_numa_stat_diff, 0,
6315 		       sizeof(pcp->vm_numa_stat_diff));
6316 	}
6317 #endif
6318 
6319 	for_each_online_pgdat(pgdat)
6320 		pgdat->per_cpu_nodestats =
6321 			alloc_percpu(struct per_cpu_nodestat);
6322 }
6323 
6324 static __meminit void zone_pcp_init(struct zone *zone)
6325 {
6326 	/*
6327 	 * per cpu subsystem is not up at this point. The following code
6328 	 * relies on the ability of the linker to provide the
6329 	 * offset of a (static) per cpu variable into the per cpu area.
6330 	 */
6331 	zone->pageset = &boot_pageset;
6332 
6333 	if (populated_zone(zone))
6334 		printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
6335 			zone->name, zone->present_pages,
6336 					 zone_batchsize(zone));
6337 }
6338 
6339 void __meminit init_currently_empty_zone(struct zone *zone,
6340 					unsigned long zone_start_pfn,
6341 					unsigned long size)
6342 {
6343 	struct pglist_data *pgdat = zone->zone_pgdat;
6344 	int zone_idx = zone_idx(zone) + 1;
6345 
6346 	if (zone_idx > pgdat->nr_zones)
6347 		pgdat->nr_zones = zone_idx;
6348 
6349 	zone->zone_start_pfn = zone_start_pfn;
6350 
6351 	mminit_dprintk(MMINIT_TRACE, "memmap_init",
6352 			"Initialising map node %d zone %lu pfns %lu -> %lu\n",
6353 			pgdat->node_id,
6354 			(unsigned long)zone_idx(zone),
6355 			zone_start_pfn, (zone_start_pfn + size));
6356 
6357 	zone_init_free_lists(zone);
6358 	zone->initialized = 1;
6359 }
6360 
6361 /**
6362  * get_pfn_range_for_nid - Return the start and end page frames for a node
6363  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
6364  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
6365  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
6366  *
6367  * It returns the start and end page frame of a node based on information
6368  * provided by memblock_set_node(). If called for a node
6369  * with no available memory, a warning is printed and the start and end
6370  * PFNs will be 0.
6371  */
6372 void __init get_pfn_range_for_nid(unsigned int nid,
6373 			unsigned long *start_pfn, unsigned long *end_pfn)
6374 {
6375 	unsigned long this_start_pfn, this_end_pfn;
6376 	int i;
6377 
6378 	*start_pfn = -1UL;
6379 	*end_pfn = 0;
6380 
6381 	for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
6382 		*start_pfn = min(*start_pfn, this_start_pfn);
6383 		*end_pfn = max(*end_pfn, this_end_pfn);
6384 	}
6385 
6386 	if (*start_pfn == -1UL)
6387 		*start_pfn = 0;
6388 }
6389 
6390 /*
6391  * This finds a zone that can be used for ZONE_MOVABLE pages. The
6392  * assumption is made that zones within a node are ordered in monotonic
6393  * increasing memory addresses so that the "highest" populated zone is used
6394  */
6395 static void __init find_usable_zone_for_movable(void)
6396 {
6397 	int zone_index;
6398 	for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
6399 		if (zone_index == ZONE_MOVABLE)
6400 			continue;
6401 
6402 		if (arch_zone_highest_possible_pfn[zone_index] >
6403 				arch_zone_lowest_possible_pfn[zone_index])
6404 			break;
6405 	}
6406 
6407 	VM_BUG_ON(zone_index == -1);
6408 	movable_zone = zone_index;
6409 }
6410 
6411 /*
6412  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
6413  * because it is sized independent of architecture. Unlike the other zones,
6414  * the starting point for ZONE_MOVABLE is not fixed. It may be different
6415  * in each node depending on the size of each node and how evenly kernelcore
6416  * is distributed. This helper function adjusts the zone ranges
6417  * provided by the architecture for a given node by using the end of the
6418  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
6419  * zones within a node are in order of monotonic increases memory addresses
6420  */
6421 static void __init adjust_zone_range_for_zone_movable(int nid,
6422 					unsigned long zone_type,
6423 					unsigned long node_start_pfn,
6424 					unsigned long node_end_pfn,
6425 					unsigned long *zone_start_pfn,
6426 					unsigned long *zone_end_pfn)
6427 {
6428 	/* Only adjust if ZONE_MOVABLE is on this node */
6429 	if (zone_movable_pfn[nid]) {
6430 		/* Size ZONE_MOVABLE */
6431 		if (zone_type == ZONE_MOVABLE) {
6432 			*zone_start_pfn = zone_movable_pfn[nid];
6433 			*zone_end_pfn = min(node_end_pfn,
6434 				arch_zone_highest_possible_pfn[movable_zone]);
6435 
6436 		/* Adjust for ZONE_MOVABLE starting within this range */
6437 		} else if (!mirrored_kernelcore &&
6438 			*zone_start_pfn < zone_movable_pfn[nid] &&
6439 			*zone_end_pfn > zone_movable_pfn[nid]) {
6440 			*zone_end_pfn = zone_movable_pfn[nid];
6441 
6442 		/* Check if this whole range is within ZONE_MOVABLE */
6443 		} else if (*zone_start_pfn >= zone_movable_pfn[nid])
6444 			*zone_start_pfn = *zone_end_pfn;
6445 	}
6446 }
6447 
6448 /*
6449  * Return the number of pages a zone spans in a node, including holes
6450  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
6451  */
6452 static unsigned long __init zone_spanned_pages_in_node(int nid,
6453 					unsigned long zone_type,
6454 					unsigned long node_start_pfn,
6455 					unsigned long node_end_pfn,
6456 					unsigned long *zone_start_pfn,
6457 					unsigned long *zone_end_pfn)
6458 {
6459 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
6460 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
6461 	/* When hotadd a new node from cpu_up(), the node should be empty */
6462 	if (!node_start_pfn && !node_end_pfn)
6463 		return 0;
6464 
6465 	/* Get the start and end of the zone */
6466 	*zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
6467 	*zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
6468 	adjust_zone_range_for_zone_movable(nid, zone_type,
6469 				node_start_pfn, node_end_pfn,
6470 				zone_start_pfn, zone_end_pfn);
6471 
6472 	/* Check that this node has pages within the zone's required range */
6473 	if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
6474 		return 0;
6475 
6476 	/* Move the zone boundaries inside the node if necessary */
6477 	*zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
6478 	*zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
6479 
6480 	/* Return the spanned pages */
6481 	return *zone_end_pfn - *zone_start_pfn;
6482 }
6483 
6484 /*
6485  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
6486  * then all holes in the requested range will be accounted for.
6487  */
6488 unsigned long __init __absent_pages_in_range(int nid,
6489 				unsigned long range_start_pfn,
6490 				unsigned long range_end_pfn)
6491 {
6492 	unsigned long nr_absent = range_end_pfn - range_start_pfn;
6493 	unsigned long start_pfn, end_pfn;
6494 	int i;
6495 
6496 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6497 		start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
6498 		end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
6499 		nr_absent -= end_pfn - start_pfn;
6500 	}
6501 	return nr_absent;
6502 }
6503 
6504 /**
6505  * absent_pages_in_range - Return number of page frames in holes within a range
6506  * @start_pfn: The start PFN to start searching for holes
6507  * @end_pfn: The end PFN to stop searching for holes
6508  *
6509  * Return: the number of pages frames in memory holes within a range.
6510  */
6511 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
6512 							unsigned long end_pfn)
6513 {
6514 	return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
6515 }
6516 
6517 /* Return the number of page frames in holes in a zone on a node */
6518 static unsigned long __init zone_absent_pages_in_node(int nid,
6519 					unsigned long zone_type,
6520 					unsigned long node_start_pfn,
6521 					unsigned long node_end_pfn)
6522 {
6523 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
6524 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
6525 	unsigned long zone_start_pfn, zone_end_pfn;
6526 	unsigned long nr_absent;
6527 
6528 	/* When hotadd a new node from cpu_up(), the node should be empty */
6529 	if (!node_start_pfn && !node_end_pfn)
6530 		return 0;
6531 
6532 	zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
6533 	zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
6534 
6535 	adjust_zone_range_for_zone_movable(nid, zone_type,
6536 			node_start_pfn, node_end_pfn,
6537 			&zone_start_pfn, &zone_end_pfn);
6538 	nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
6539 
6540 	/*
6541 	 * ZONE_MOVABLE handling.
6542 	 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
6543 	 * and vice versa.
6544 	 */
6545 	if (mirrored_kernelcore && zone_movable_pfn[nid]) {
6546 		unsigned long start_pfn, end_pfn;
6547 		struct memblock_region *r;
6548 
6549 		for_each_mem_region(r) {
6550 			start_pfn = clamp(memblock_region_memory_base_pfn(r),
6551 					  zone_start_pfn, zone_end_pfn);
6552 			end_pfn = clamp(memblock_region_memory_end_pfn(r),
6553 					zone_start_pfn, zone_end_pfn);
6554 
6555 			if (zone_type == ZONE_MOVABLE &&
6556 			    memblock_is_mirror(r))
6557 				nr_absent += end_pfn - start_pfn;
6558 
6559 			if (zone_type == ZONE_NORMAL &&
6560 			    !memblock_is_mirror(r))
6561 				nr_absent += end_pfn - start_pfn;
6562 		}
6563 	}
6564 
6565 	return nr_absent;
6566 }
6567 
6568 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
6569 						unsigned long node_start_pfn,
6570 						unsigned long node_end_pfn)
6571 {
6572 	unsigned long realtotalpages = 0, totalpages = 0;
6573 	enum zone_type i;
6574 
6575 	for (i = 0; i < MAX_NR_ZONES; i++) {
6576 		struct zone *zone = pgdat->node_zones + i;
6577 		unsigned long zone_start_pfn, zone_end_pfn;
6578 		unsigned long spanned, absent;
6579 		unsigned long size, real_size;
6580 
6581 		spanned = zone_spanned_pages_in_node(pgdat->node_id, i,
6582 						     node_start_pfn,
6583 						     node_end_pfn,
6584 						     &zone_start_pfn,
6585 						     &zone_end_pfn);
6586 		absent = zone_absent_pages_in_node(pgdat->node_id, i,
6587 						   node_start_pfn,
6588 						   node_end_pfn);
6589 
6590 		size = spanned;
6591 		real_size = size - absent;
6592 
6593 		if (size)
6594 			zone->zone_start_pfn = zone_start_pfn;
6595 		else
6596 			zone->zone_start_pfn = 0;
6597 		zone->spanned_pages = size;
6598 		zone->present_pages = real_size;
6599 
6600 		totalpages += size;
6601 		realtotalpages += real_size;
6602 	}
6603 
6604 	pgdat->node_spanned_pages = totalpages;
6605 	pgdat->node_present_pages = realtotalpages;
6606 	printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
6607 							realtotalpages);
6608 }
6609 
6610 #ifndef CONFIG_SPARSEMEM
6611 /*
6612  * Calculate the size of the zone->blockflags rounded to an unsigned long
6613  * Start by making sure zonesize is a multiple of pageblock_order by rounding
6614  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
6615  * round what is now in bits to nearest long in bits, then return it in
6616  * bytes.
6617  */
6618 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
6619 {
6620 	unsigned long usemapsize;
6621 
6622 	zonesize += zone_start_pfn & (pageblock_nr_pages-1);
6623 	usemapsize = roundup(zonesize, pageblock_nr_pages);
6624 	usemapsize = usemapsize >> pageblock_order;
6625 	usemapsize *= NR_PAGEBLOCK_BITS;
6626 	usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
6627 
6628 	return usemapsize / 8;
6629 }
6630 
6631 static void __ref setup_usemap(struct pglist_data *pgdat,
6632 				struct zone *zone,
6633 				unsigned long zone_start_pfn,
6634 				unsigned long zonesize)
6635 {
6636 	unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
6637 	zone->pageblock_flags = NULL;
6638 	if (usemapsize) {
6639 		zone->pageblock_flags =
6640 			memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
6641 					    pgdat->node_id);
6642 		if (!zone->pageblock_flags)
6643 			panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
6644 			      usemapsize, zone->name, pgdat->node_id);
6645 	}
6646 }
6647 #else
6648 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
6649 				unsigned long zone_start_pfn, unsigned long zonesize) {}
6650 #endif /* CONFIG_SPARSEMEM */
6651 
6652 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
6653 
6654 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
6655 void __init set_pageblock_order(void)
6656 {
6657 	unsigned int order;
6658 
6659 	/* Check that pageblock_nr_pages has not already been setup */
6660 	if (pageblock_order)
6661 		return;
6662 
6663 	if (HPAGE_SHIFT > PAGE_SHIFT)
6664 		order = HUGETLB_PAGE_ORDER;
6665 	else
6666 		order = MAX_ORDER - 1;
6667 
6668 	/*
6669 	 * Assume the largest contiguous order of interest is a huge page.
6670 	 * This value may be variable depending on boot parameters on IA64 and
6671 	 * powerpc.
6672 	 */
6673 	pageblock_order = order;
6674 }
6675 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6676 
6677 /*
6678  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
6679  * is unused as pageblock_order is set at compile-time. See
6680  * include/linux/pageblock-flags.h for the values of pageblock_order based on
6681  * the kernel config
6682  */
6683 void __init set_pageblock_order(void)
6684 {
6685 }
6686 
6687 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6688 
6689 static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
6690 						unsigned long present_pages)
6691 {
6692 	unsigned long pages = spanned_pages;
6693 
6694 	/*
6695 	 * Provide a more accurate estimation if there are holes within
6696 	 * the zone and SPARSEMEM is in use. If there are holes within the
6697 	 * zone, each populated memory region may cost us one or two extra
6698 	 * memmap pages due to alignment because memmap pages for each
6699 	 * populated regions may not be naturally aligned on page boundary.
6700 	 * So the (present_pages >> 4) heuristic is a tradeoff for that.
6701 	 */
6702 	if (spanned_pages > present_pages + (present_pages >> 4) &&
6703 	    IS_ENABLED(CONFIG_SPARSEMEM))
6704 		pages = present_pages;
6705 
6706 	return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
6707 }
6708 
6709 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6710 static void pgdat_init_split_queue(struct pglist_data *pgdat)
6711 {
6712 	struct deferred_split *ds_queue = &pgdat->deferred_split_queue;
6713 
6714 	spin_lock_init(&ds_queue->split_queue_lock);
6715 	INIT_LIST_HEAD(&ds_queue->split_queue);
6716 	ds_queue->split_queue_len = 0;
6717 }
6718 #else
6719 static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
6720 #endif
6721 
6722 #ifdef CONFIG_COMPACTION
6723 static void pgdat_init_kcompactd(struct pglist_data *pgdat)
6724 {
6725 	init_waitqueue_head(&pgdat->kcompactd_wait);
6726 }
6727 #else
6728 static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
6729 #endif
6730 
6731 static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
6732 {
6733 	pgdat_resize_init(pgdat);
6734 
6735 	pgdat_init_split_queue(pgdat);
6736 	pgdat_init_kcompactd(pgdat);
6737 
6738 	init_waitqueue_head(&pgdat->kswapd_wait);
6739 	init_waitqueue_head(&pgdat->pfmemalloc_wait);
6740 
6741 	pgdat_page_ext_init(pgdat);
6742 	spin_lock_init(&pgdat->lru_lock);
6743 	lruvec_init(&pgdat->__lruvec);
6744 }
6745 
6746 static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
6747 							unsigned long remaining_pages)
6748 {
6749 	atomic_long_set(&zone->managed_pages, remaining_pages);
6750 	zone_set_nid(zone, nid);
6751 	zone->name = zone_names[idx];
6752 	zone->zone_pgdat = NODE_DATA(nid);
6753 	spin_lock_init(&zone->lock);
6754 	zone_seqlock_init(zone);
6755 	zone_pcp_init(zone);
6756 }
6757 
6758 /*
6759  * Set up the zone data structures
6760  * - init pgdat internals
6761  * - init all zones belonging to this node
6762  *
6763  * NOTE: this function is only called during memory hotplug
6764  */
6765 #ifdef CONFIG_MEMORY_HOTPLUG
6766 void __ref free_area_init_core_hotplug(int nid)
6767 {
6768 	enum zone_type z;
6769 	pg_data_t *pgdat = NODE_DATA(nid);
6770 
6771 	pgdat_init_internals(pgdat);
6772 	for (z = 0; z < MAX_NR_ZONES; z++)
6773 		zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
6774 }
6775 #endif
6776 
6777 /*
6778  * Set up the zone data structures:
6779  *   - mark all pages reserved
6780  *   - mark all memory queues empty
6781  *   - clear the memory bitmaps
6782  *
6783  * NOTE: pgdat should get zeroed by caller.
6784  * NOTE: this function is only called during early init.
6785  */
6786 static void __init free_area_init_core(struct pglist_data *pgdat)
6787 {
6788 	enum zone_type j;
6789 	int nid = pgdat->node_id;
6790 
6791 	pgdat_init_internals(pgdat);
6792 	pgdat->per_cpu_nodestats = &boot_nodestats;
6793 
6794 	for (j = 0; j < MAX_NR_ZONES; j++) {
6795 		struct zone *zone = pgdat->node_zones + j;
6796 		unsigned long size, freesize, memmap_pages;
6797 		unsigned long zone_start_pfn = zone->zone_start_pfn;
6798 
6799 		size = zone->spanned_pages;
6800 		freesize = zone->present_pages;
6801 
6802 		/*
6803 		 * Adjust freesize so that it accounts for how much memory
6804 		 * is used by this zone for memmap. This affects the watermark
6805 		 * and per-cpu initialisations
6806 		 */
6807 		memmap_pages = calc_memmap_size(size, freesize);
6808 		if (!is_highmem_idx(j)) {
6809 			if (freesize >= memmap_pages) {
6810 				freesize -= memmap_pages;
6811 				if (memmap_pages)
6812 					printk(KERN_DEBUG
6813 					       "  %s zone: %lu pages used for memmap\n",
6814 					       zone_names[j], memmap_pages);
6815 			} else
6816 				pr_warn("  %s zone: %lu pages exceeds freesize %lu\n",
6817 					zone_names[j], memmap_pages, freesize);
6818 		}
6819 
6820 		/* Account for reserved pages */
6821 		if (j == 0 && freesize > dma_reserve) {
6822 			freesize -= dma_reserve;
6823 			printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
6824 					zone_names[0], dma_reserve);
6825 		}
6826 
6827 		if (!is_highmem_idx(j))
6828 			nr_kernel_pages += freesize;
6829 		/* Charge for highmem memmap if there are enough kernel pages */
6830 		else if (nr_kernel_pages > memmap_pages * 2)
6831 			nr_kernel_pages -= memmap_pages;
6832 		nr_all_pages += freesize;
6833 
6834 		/*
6835 		 * Set an approximate value for lowmem here, it will be adjusted
6836 		 * when the bootmem allocator frees pages into the buddy system.
6837 		 * And all highmem pages will be managed by the buddy system.
6838 		 */
6839 		zone_init_internals(zone, j, nid, freesize);
6840 
6841 		if (!size)
6842 			continue;
6843 
6844 		set_pageblock_order();
6845 		setup_usemap(pgdat, zone, zone_start_pfn, size);
6846 		init_currently_empty_zone(zone, zone_start_pfn, size);
6847 		memmap_init(size, nid, j, zone_start_pfn);
6848 	}
6849 }
6850 
6851 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6852 static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
6853 {
6854 	unsigned long __maybe_unused start = 0;
6855 	unsigned long __maybe_unused offset = 0;
6856 
6857 	/* Skip empty nodes */
6858 	if (!pgdat->node_spanned_pages)
6859 		return;
6860 
6861 	start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
6862 	offset = pgdat->node_start_pfn - start;
6863 	/* ia64 gets its own node_mem_map, before this, without bootmem */
6864 	if (!pgdat->node_mem_map) {
6865 		unsigned long size, end;
6866 		struct page *map;
6867 
6868 		/*
6869 		 * The zone's endpoints aren't required to be MAX_ORDER
6870 		 * aligned but the node_mem_map endpoints must be in order
6871 		 * for the buddy allocator to function correctly.
6872 		 */
6873 		end = pgdat_end_pfn(pgdat);
6874 		end = ALIGN(end, MAX_ORDER_NR_PAGES);
6875 		size =  (end - start) * sizeof(struct page);
6876 		map = memblock_alloc_node(size, SMP_CACHE_BYTES,
6877 					  pgdat->node_id);
6878 		if (!map)
6879 			panic("Failed to allocate %ld bytes for node %d memory map\n",
6880 			      size, pgdat->node_id);
6881 		pgdat->node_mem_map = map + offset;
6882 	}
6883 	pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
6884 				__func__, pgdat->node_id, (unsigned long)pgdat,
6885 				(unsigned long)pgdat->node_mem_map);
6886 #ifndef CONFIG_NEED_MULTIPLE_NODES
6887 	/*
6888 	 * With no DISCONTIG, the global mem_map is just set as node 0's
6889 	 */
6890 	if (pgdat == NODE_DATA(0)) {
6891 		mem_map = NODE_DATA(0)->node_mem_map;
6892 		if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
6893 			mem_map -= offset;
6894 	}
6895 #endif
6896 }
6897 #else
6898 static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { }
6899 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6900 
6901 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
6902 static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
6903 {
6904 	pgdat->first_deferred_pfn = ULONG_MAX;
6905 }
6906 #else
6907 static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
6908 #endif
6909 
6910 static void __init free_area_init_node(int nid)
6911 {
6912 	pg_data_t *pgdat = NODE_DATA(nid);
6913 	unsigned long start_pfn = 0;
6914 	unsigned long end_pfn = 0;
6915 
6916 	/* pg_data_t should be reset to zero when it's allocated */
6917 	WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx);
6918 
6919 	get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
6920 
6921 	pgdat->node_id = nid;
6922 	pgdat->node_start_pfn = start_pfn;
6923 	pgdat->per_cpu_nodestats = NULL;
6924 
6925 	pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
6926 		(u64)start_pfn << PAGE_SHIFT,
6927 		end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
6928 	calculate_node_totalpages(pgdat, start_pfn, end_pfn);
6929 
6930 	alloc_node_mem_map(pgdat);
6931 	pgdat_set_deferred_range(pgdat);
6932 
6933 	free_area_init_core(pgdat);
6934 }
6935 
6936 void __init free_area_init_memoryless_node(int nid)
6937 {
6938 	free_area_init_node(nid);
6939 }
6940 
6941 #if !defined(CONFIG_FLAT_NODE_MEM_MAP)
6942 /*
6943  * Initialize all valid struct pages in the range [spfn, epfn) and mark them
6944  * PageReserved(). Return the number of struct pages that were initialized.
6945  */
6946 static u64 __init init_unavailable_range(unsigned long spfn, unsigned long epfn)
6947 {
6948 	unsigned long pfn;
6949 	u64 pgcnt = 0;
6950 
6951 	for (pfn = spfn; pfn < epfn; pfn++) {
6952 		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
6953 			pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
6954 				+ pageblock_nr_pages - 1;
6955 			continue;
6956 		}
6957 		/*
6958 		 * Use a fake node/zone (0) for now. Some of these pages
6959 		 * (in memblock.reserved but not in memblock.memory) will
6960 		 * get re-initialized via reserve_bootmem_region() later.
6961 		 */
6962 		__init_single_page(pfn_to_page(pfn), pfn, 0, 0);
6963 		__SetPageReserved(pfn_to_page(pfn));
6964 		pgcnt++;
6965 	}
6966 
6967 	return pgcnt;
6968 }
6969 
6970 /*
6971  * Only struct pages that are backed by physical memory are zeroed and
6972  * initialized by going through __init_single_page(). But, there are some
6973  * struct pages which are reserved in memblock allocator and their fields
6974  * may be accessed (for example page_to_pfn() on some configuration accesses
6975  * flags). We must explicitly initialize those struct pages.
6976  *
6977  * This function also addresses a similar issue where struct pages are left
6978  * uninitialized because the physical address range is not covered by
6979  * memblock.memory or memblock.reserved. That could happen when memblock
6980  * layout is manually configured via memmap=, or when the highest physical
6981  * address (max_pfn) does not end on a section boundary.
6982  */
6983 static void __init init_unavailable_mem(void)
6984 {
6985 	phys_addr_t start, end;
6986 	u64 i, pgcnt;
6987 	phys_addr_t next = 0;
6988 
6989 	/*
6990 	 * Loop through unavailable ranges not covered by memblock.memory.
6991 	 */
6992 	pgcnt = 0;
6993 	for_each_mem_range(i, &start, &end) {
6994 		if (next < start)
6995 			pgcnt += init_unavailable_range(PFN_DOWN(next),
6996 							PFN_UP(start));
6997 		next = end;
6998 	}
6999 
7000 	/*
7001 	 * Early sections always have a fully populated memmap for the whole
7002 	 * section - see pfn_valid(). If the last section has holes at the
7003 	 * end and that section is marked "online", the memmap will be
7004 	 * considered initialized. Make sure that memmap has a well defined
7005 	 * state.
7006 	 */
7007 	pgcnt += init_unavailable_range(PFN_DOWN(next),
7008 					round_up(max_pfn, PAGES_PER_SECTION));
7009 
7010 	/*
7011 	 * Struct pages that do not have backing memory. This could be because
7012 	 * firmware is using some of this memory, or for some other reasons.
7013 	 */
7014 	if (pgcnt)
7015 		pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt);
7016 }
7017 #else
7018 static inline void __init init_unavailable_mem(void)
7019 {
7020 }
7021 #endif /* !CONFIG_FLAT_NODE_MEM_MAP */
7022 
7023 #if MAX_NUMNODES > 1
7024 /*
7025  * Figure out the number of possible node ids.
7026  */
7027 void __init setup_nr_node_ids(void)
7028 {
7029 	unsigned int highest;
7030 
7031 	highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
7032 	nr_node_ids = highest + 1;
7033 }
7034 #endif
7035 
7036 /**
7037  * node_map_pfn_alignment - determine the maximum internode alignment
7038  *
7039  * This function should be called after node map is populated and sorted.
7040  * It calculates the maximum power of two alignment which can distinguish
7041  * all the nodes.
7042  *
7043  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
7044  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
7045  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
7046  * shifted, 1GiB is enough and this function will indicate so.
7047  *
7048  * This is used to test whether pfn -> nid mapping of the chosen memory
7049  * model has fine enough granularity to avoid incorrect mapping for the
7050  * populated node map.
7051  *
7052  * Return: the determined alignment in pfn's.  0 if there is no alignment
7053  * requirement (single node).
7054  */
7055 unsigned long __init node_map_pfn_alignment(void)
7056 {
7057 	unsigned long accl_mask = 0, last_end = 0;
7058 	unsigned long start, end, mask;
7059 	int last_nid = NUMA_NO_NODE;
7060 	int i, nid;
7061 
7062 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
7063 		if (!start || last_nid < 0 || last_nid == nid) {
7064 			last_nid = nid;
7065 			last_end = end;
7066 			continue;
7067 		}
7068 
7069 		/*
7070 		 * Start with a mask granular enough to pin-point to the
7071 		 * start pfn and tick off bits one-by-one until it becomes
7072 		 * too coarse to separate the current node from the last.
7073 		 */
7074 		mask = ~((1 << __ffs(start)) - 1);
7075 		while (mask && last_end <= (start & (mask << 1)))
7076 			mask <<= 1;
7077 
7078 		/* accumulate all internode masks */
7079 		accl_mask |= mask;
7080 	}
7081 
7082 	/* convert mask to number of pages */
7083 	return ~accl_mask + 1;
7084 }
7085 
7086 /**
7087  * find_min_pfn_with_active_regions - Find the minimum PFN registered
7088  *
7089  * Return: the minimum PFN based on information provided via
7090  * memblock_set_node().
7091  */
7092 unsigned long __init find_min_pfn_with_active_regions(void)
7093 {
7094 	return PHYS_PFN(memblock_start_of_DRAM());
7095 }
7096 
7097 /*
7098  * early_calculate_totalpages()
7099  * Sum pages in active regions for movable zone.
7100  * Populate N_MEMORY for calculating usable_nodes.
7101  */
7102 static unsigned long __init early_calculate_totalpages(void)
7103 {
7104 	unsigned long totalpages = 0;
7105 	unsigned long start_pfn, end_pfn;
7106 	int i, nid;
7107 
7108 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
7109 		unsigned long pages = end_pfn - start_pfn;
7110 
7111 		totalpages += pages;
7112 		if (pages)
7113 			node_set_state(nid, N_MEMORY);
7114 	}
7115 	return totalpages;
7116 }
7117 
7118 /*
7119  * Find the PFN the Movable zone begins in each node. Kernel memory
7120  * is spread evenly between nodes as long as the nodes have enough
7121  * memory. When they don't, some nodes will have more kernelcore than
7122  * others
7123  */
7124 static void __init find_zone_movable_pfns_for_nodes(void)
7125 {
7126 	int i, nid;
7127 	unsigned long usable_startpfn;
7128 	unsigned long kernelcore_node, kernelcore_remaining;
7129 	/* save the state before borrow the nodemask */
7130 	nodemask_t saved_node_state = node_states[N_MEMORY];
7131 	unsigned long totalpages = early_calculate_totalpages();
7132 	int usable_nodes = nodes_weight(node_states[N_MEMORY]);
7133 	struct memblock_region *r;
7134 
7135 	/* Need to find movable_zone earlier when movable_node is specified. */
7136 	find_usable_zone_for_movable();
7137 
7138 	/*
7139 	 * If movable_node is specified, ignore kernelcore and movablecore
7140 	 * options.
7141 	 */
7142 	if (movable_node_is_enabled()) {
7143 		for_each_mem_region(r) {
7144 			if (!memblock_is_hotpluggable(r))
7145 				continue;
7146 
7147 			nid = memblock_get_region_node(r);
7148 
7149 			usable_startpfn = PFN_DOWN(r->base);
7150 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7151 				min(usable_startpfn, zone_movable_pfn[nid]) :
7152 				usable_startpfn;
7153 		}
7154 
7155 		goto out2;
7156 	}
7157 
7158 	/*
7159 	 * If kernelcore=mirror is specified, ignore movablecore option
7160 	 */
7161 	if (mirrored_kernelcore) {
7162 		bool mem_below_4gb_not_mirrored = false;
7163 
7164 		for_each_mem_region(r) {
7165 			if (memblock_is_mirror(r))
7166 				continue;
7167 
7168 			nid = memblock_get_region_node(r);
7169 
7170 			usable_startpfn = memblock_region_memory_base_pfn(r);
7171 
7172 			if (usable_startpfn < 0x100000) {
7173 				mem_below_4gb_not_mirrored = true;
7174 				continue;
7175 			}
7176 
7177 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7178 				min(usable_startpfn, zone_movable_pfn[nid]) :
7179 				usable_startpfn;
7180 		}
7181 
7182 		if (mem_below_4gb_not_mirrored)
7183 			pr_warn("This configuration results in unmirrored kernel memory.\n");
7184 
7185 		goto out2;
7186 	}
7187 
7188 	/*
7189 	 * If kernelcore=nn% or movablecore=nn% was specified, calculate the
7190 	 * amount of necessary memory.
7191 	 */
7192 	if (required_kernelcore_percent)
7193 		required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
7194 				       10000UL;
7195 	if (required_movablecore_percent)
7196 		required_movablecore = (totalpages * 100 * required_movablecore_percent) /
7197 					10000UL;
7198 
7199 	/*
7200 	 * If movablecore= was specified, calculate what size of
7201 	 * kernelcore that corresponds so that memory usable for
7202 	 * any allocation type is evenly spread. If both kernelcore
7203 	 * and movablecore are specified, then the value of kernelcore
7204 	 * will be used for required_kernelcore if it's greater than
7205 	 * what movablecore would have allowed.
7206 	 */
7207 	if (required_movablecore) {
7208 		unsigned long corepages;
7209 
7210 		/*
7211 		 * Round-up so that ZONE_MOVABLE is at least as large as what
7212 		 * was requested by the user
7213 		 */
7214 		required_movablecore =
7215 			roundup(required_movablecore, MAX_ORDER_NR_PAGES);
7216 		required_movablecore = min(totalpages, required_movablecore);
7217 		corepages = totalpages - required_movablecore;
7218 
7219 		required_kernelcore = max(required_kernelcore, corepages);
7220 	}
7221 
7222 	/*
7223 	 * If kernelcore was not specified or kernelcore size is larger
7224 	 * than totalpages, there is no ZONE_MOVABLE.
7225 	 */
7226 	if (!required_kernelcore || required_kernelcore >= totalpages)
7227 		goto out;
7228 
7229 	/* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
7230 	usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
7231 
7232 restart:
7233 	/* Spread kernelcore memory as evenly as possible throughout nodes */
7234 	kernelcore_node = required_kernelcore / usable_nodes;
7235 	for_each_node_state(nid, N_MEMORY) {
7236 		unsigned long start_pfn, end_pfn;
7237 
7238 		/*
7239 		 * Recalculate kernelcore_node if the division per node
7240 		 * now exceeds what is necessary to satisfy the requested
7241 		 * amount of memory for the kernel
7242 		 */
7243 		if (required_kernelcore < kernelcore_node)
7244 			kernelcore_node = required_kernelcore / usable_nodes;
7245 
7246 		/*
7247 		 * As the map is walked, we track how much memory is usable
7248 		 * by the kernel using kernelcore_remaining. When it is
7249 		 * 0, the rest of the node is usable by ZONE_MOVABLE
7250 		 */
7251 		kernelcore_remaining = kernelcore_node;
7252 
7253 		/* Go through each range of PFNs within this node */
7254 		for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7255 			unsigned long size_pages;
7256 
7257 			start_pfn = max(start_pfn, zone_movable_pfn[nid]);
7258 			if (start_pfn >= end_pfn)
7259 				continue;
7260 
7261 			/* Account for what is only usable for kernelcore */
7262 			if (start_pfn < usable_startpfn) {
7263 				unsigned long kernel_pages;
7264 				kernel_pages = min(end_pfn, usable_startpfn)
7265 								- start_pfn;
7266 
7267 				kernelcore_remaining -= min(kernel_pages,
7268 							kernelcore_remaining);
7269 				required_kernelcore -= min(kernel_pages,
7270 							required_kernelcore);
7271 
7272 				/* Continue if range is now fully accounted */
7273 				if (end_pfn <= usable_startpfn) {
7274 
7275 					/*
7276 					 * Push zone_movable_pfn to the end so
7277 					 * that if we have to rebalance
7278 					 * kernelcore across nodes, we will
7279 					 * not double account here
7280 					 */
7281 					zone_movable_pfn[nid] = end_pfn;
7282 					continue;
7283 				}
7284 				start_pfn = usable_startpfn;
7285 			}
7286 
7287 			/*
7288 			 * The usable PFN range for ZONE_MOVABLE is from
7289 			 * start_pfn->end_pfn. Calculate size_pages as the
7290 			 * number of pages used as kernelcore
7291 			 */
7292 			size_pages = end_pfn - start_pfn;
7293 			if (size_pages > kernelcore_remaining)
7294 				size_pages = kernelcore_remaining;
7295 			zone_movable_pfn[nid] = start_pfn + size_pages;
7296 
7297 			/*
7298 			 * Some kernelcore has been met, update counts and
7299 			 * break if the kernelcore for this node has been
7300 			 * satisfied
7301 			 */
7302 			required_kernelcore -= min(required_kernelcore,
7303 								size_pages);
7304 			kernelcore_remaining -= size_pages;
7305 			if (!kernelcore_remaining)
7306 				break;
7307 		}
7308 	}
7309 
7310 	/*
7311 	 * If there is still required_kernelcore, we do another pass with one
7312 	 * less node in the count. This will push zone_movable_pfn[nid] further
7313 	 * along on the nodes that still have memory until kernelcore is
7314 	 * satisfied
7315 	 */
7316 	usable_nodes--;
7317 	if (usable_nodes && required_kernelcore > usable_nodes)
7318 		goto restart;
7319 
7320 out2:
7321 	/* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
7322 	for (nid = 0; nid < MAX_NUMNODES; nid++)
7323 		zone_movable_pfn[nid] =
7324 			roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
7325 
7326 out:
7327 	/* restore the node_state */
7328 	node_states[N_MEMORY] = saved_node_state;
7329 }
7330 
7331 /* Any regular or high memory on that node ? */
7332 static void check_for_memory(pg_data_t *pgdat, int nid)
7333 {
7334 	enum zone_type zone_type;
7335 
7336 	for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
7337 		struct zone *zone = &pgdat->node_zones[zone_type];
7338 		if (populated_zone(zone)) {
7339 			if (IS_ENABLED(CONFIG_HIGHMEM))
7340 				node_set_state(nid, N_HIGH_MEMORY);
7341 			if (zone_type <= ZONE_NORMAL)
7342 				node_set_state(nid, N_NORMAL_MEMORY);
7343 			break;
7344 		}
7345 	}
7346 }
7347 
7348 /*
7349  * Some architecturs, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For
7350  * such cases we allow max_zone_pfn sorted in the descending order
7351  */
7352 bool __weak arch_has_descending_max_zone_pfns(void)
7353 {
7354 	return false;
7355 }
7356 
7357 /**
7358  * free_area_init - Initialise all pg_data_t and zone data
7359  * @max_zone_pfn: an array of max PFNs for each zone
7360  *
7361  * This will call free_area_init_node() for each active node in the system.
7362  * Using the page ranges provided by memblock_set_node(), the size of each
7363  * zone in each node and their holes is calculated. If the maximum PFN
7364  * between two adjacent zones match, it is assumed that the zone is empty.
7365  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
7366  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
7367  * starts where the previous one ended. For example, ZONE_DMA32 starts
7368  * at arch_max_dma_pfn.
7369  */
7370 void __init free_area_init(unsigned long *max_zone_pfn)
7371 {
7372 	unsigned long start_pfn, end_pfn;
7373 	int i, nid, zone;
7374 	bool descending;
7375 
7376 	/* Record where the zone boundaries are */
7377 	memset(arch_zone_lowest_possible_pfn, 0,
7378 				sizeof(arch_zone_lowest_possible_pfn));
7379 	memset(arch_zone_highest_possible_pfn, 0,
7380 				sizeof(arch_zone_highest_possible_pfn));
7381 
7382 	start_pfn = find_min_pfn_with_active_regions();
7383 	descending = arch_has_descending_max_zone_pfns();
7384 
7385 	for (i = 0; i < MAX_NR_ZONES; i++) {
7386 		if (descending)
7387 			zone = MAX_NR_ZONES - i - 1;
7388 		else
7389 			zone = i;
7390 
7391 		if (zone == ZONE_MOVABLE)
7392 			continue;
7393 
7394 		end_pfn = max(max_zone_pfn[zone], start_pfn);
7395 		arch_zone_lowest_possible_pfn[zone] = start_pfn;
7396 		arch_zone_highest_possible_pfn[zone] = end_pfn;
7397 
7398 		start_pfn = end_pfn;
7399 	}
7400 
7401 	/* Find the PFNs that ZONE_MOVABLE begins at in each node */
7402 	memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
7403 	find_zone_movable_pfns_for_nodes();
7404 
7405 	/* Print out the zone ranges */
7406 	pr_info("Zone ranges:\n");
7407 	for (i = 0; i < MAX_NR_ZONES; i++) {
7408 		if (i == ZONE_MOVABLE)
7409 			continue;
7410 		pr_info("  %-8s ", zone_names[i]);
7411 		if (arch_zone_lowest_possible_pfn[i] ==
7412 				arch_zone_highest_possible_pfn[i])
7413 			pr_cont("empty\n");
7414 		else
7415 			pr_cont("[mem %#018Lx-%#018Lx]\n",
7416 				(u64)arch_zone_lowest_possible_pfn[i]
7417 					<< PAGE_SHIFT,
7418 				((u64)arch_zone_highest_possible_pfn[i]
7419 					<< PAGE_SHIFT) - 1);
7420 	}
7421 
7422 	/* Print out the PFNs ZONE_MOVABLE begins at in each node */
7423 	pr_info("Movable zone start for each node\n");
7424 	for (i = 0; i < MAX_NUMNODES; i++) {
7425 		if (zone_movable_pfn[i])
7426 			pr_info("  Node %d: %#018Lx\n", i,
7427 			       (u64)zone_movable_pfn[i] << PAGE_SHIFT);
7428 	}
7429 
7430 	/*
7431 	 * Print out the early node map, and initialize the
7432 	 * subsection-map relative to active online memory ranges to
7433 	 * enable future "sub-section" extensions of the memory map.
7434 	 */
7435 	pr_info("Early memory node ranges\n");
7436 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
7437 		pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
7438 			(u64)start_pfn << PAGE_SHIFT,
7439 			((u64)end_pfn << PAGE_SHIFT) - 1);
7440 		subsection_map_init(start_pfn, end_pfn - start_pfn);
7441 	}
7442 
7443 	/* Initialise every node */
7444 	mminit_verify_pageflags_layout();
7445 	setup_nr_node_ids();
7446 	init_unavailable_mem();
7447 	for_each_online_node(nid) {
7448 		pg_data_t *pgdat = NODE_DATA(nid);
7449 		free_area_init_node(nid);
7450 
7451 		/* Any memory on that node */
7452 		if (pgdat->node_present_pages)
7453 			node_set_state(nid, N_MEMORY);
7454 		check_for_memory(pgdat, nid);
7455 	}
7456 }
7457 
7458 static int __init cmdline_parse_core(char *p, unsigned long *core,
7459 				     unsigned long *percent)
7460 {
7461 	unsigned long long coremem;
7462 	char *endptr;
7463 
7464 	if (!p)
7465 		return -EINVAL;
7466 
7467 	/* Value may be a percentage of total memory, otherwise bytes */
7468 	coremem = simple_strtoull(p, &endptr, 0);
7469 	if (*endptr == '%') {
7470 		/* Paranoid check for percent values greater than 100 */
7471 		WARN_ON(coremem > 100);
7472 
7473 		*percent = coremem;
7474 	} else {
7475 		coremem = memparse(p, &p);
7476 		/* Paranoid check that UL is enough for the coremem value */
7477 		WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
7478 
7479 		*core = coremem >> PAGE_SHIFT;
7480 		*percent = 0UL;
7481 	}
7482 	return 0;
7483 }
7484 
7485 /*
7486  * kernelcore=size sets the amount of memory for use for allocations that
7487  * cannot be reclaimed or migrated.
7488  */
7489 static int __init cmdline_parse_kernelcore(char *p)
7490 {
7491 	/* parse kernelcore=mirror */
7492 	if (parse_option_str(p, "mirror")) {
7493 		mirrored_kernelcore = true;
7494 		return 0;
7495 	}
7496 
7497 	return cmdline_parse_core(p, &required_kernelcore,
7498 				  &required_kernelcore_percent);
7499 }
7500 
7501 /*
7502  * movablecore=size sets the amount of memory for use for allocations that
7503  * can be reclaimed or migrated.
7504  */
7505 static int __init cmdline_parse_movablecore(char *p)
7506 {
7507 	return cmdline_parse_core(p, &required_movablecore,
7508 				  &required_movablecore_percent);
7509 }
7510 
7511 early_param("kernelcore", cmdline_parse_kernelcore);
7512 early_param("movablecore", cmdline_parse_movablecore);
7513 
7514 void adjust_managed_page_count(struct page *page, long count)
7515 {
7516 	atomic_long_add(count, &page_zone(page)->managed_pages);
7517 	totalram_pages_add(count);
7518 #ifdef CONFIG_HIGHMEM
7519 	if (PageHighMem(page))
7520 		totalhigh_pages_add(count);
7521 #endif
7522 }
7523 EXPORT_SYMBOL(adjust_managed_page_count);
7524 
7525 unsigned long free_reserved_area(void *start, void *end, int poison, const char *s)
7526 {
7527 	void *pos;
7528 	unsigned long pages = 0;
7529 
7530 	start = (void *)PAGE_ALIGN((unsigned long)start);
7531 	end = (void *)((unsigned long)end & PAGE_MASK);
7532 	for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
7533 		struct page *page = virt_to_page(pos);
7534 		void *direct_map_addr;
7535 
7536 		/*
7537 		 * 'direct_map_addr' might be different from 'pos'
7538 		 * because some architectures' virt_to_page()
7539 		 * work with aliases.  Getting the direct map
7540 		 * address ensures that we get a _writeable_
7541 		 * alias for the memset().
7542 		 */
7543 		direct_map_addr = page_address(page);
7544 		if ((unsigned int)poison <= 0xFF)
7545 			memset(direct_map_addr, poison, PAGE_SIZE);
7546 
7547 		free_reserved_page(page);
7548 	}
7549 
7550 	if (pages && s)
7551 		pr_info("Freeing %s memory: %ldK\n",
7552 			s, pages << (PAGE_SHIFT - 10));
7553 
7554 	return pages;
7555 }
7556 
7557 #ifdef	CONFIG_HIGHMEM
7558 void free_highmem_page(struct page *page)
7559 {
7560 	__free_reserved_page(page);
7561 	totalram_pages_inc();
7562 	atomic_long_inc(&page_zone(page)->managed_pages);
7563 	totalhigh_pages_inc();
7564 }
7565 #endif
7566 
7567 
7568 void __init mem_init_print_info(const char *str)
7569 {
7570 	unsigned long physpages, codesize, datasize, rosize, bss_size;
7571 	unsigned long init_code_size, init_data_size;
7572 
7573 	physpages = get_num_physpages();
7574 	codesize = _etext - _stext;
7575 	datasize = _edata - _sdata;
7576 	rosize = __end_rodata - __start_rodata;
7577 	bss_size = __bss_stop - __bss_start;
7578 	init_data_size = __init_end - __init_begin;
7579 	init_code_size = _einittext - _sinittext;
7580 
7581 	/*
7582 	 * Detect special cases and adjust section sizes accordingly:
7583 	 * 1) .init.* may be embedded into .data sections
7584 	 * 2) .init.text.* may be out of [__init_begin, __init_end],
7585 	 *    please refer to arch/tile/kernel/vmlinux.lds.S.
7586 	 * 3) .rodata.* may be embedded into .text or .data sections.
7587 	 */
7588 #define adj_init_size(start, end, size, pos, adj) \
7589 	do { \
7590 		if (start <= pos && pos < end && size > adj) \
7591 			size -= adj; \
7592 	} while (0)
7593 
7594 	adj_init_size(__init_begin, __init_end, init_data_size,
7595 		     _sinittext, init_code_size);
7596 	adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
7597 	adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
7598 	adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
7599 	adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
7600 
7601 #undef	adj_init_size
7602 
7603 	pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
7604 #ifdef	CONFIG_HIGHMEM
7605 		", %luK highmem"
7606 #endif
7607 		"%s%s)\n",
7608 		nr_free_pages() << (PAGE_SHIFT - 10),
7609 		physpages << (PAGE_SHIFT - 10),
7610 		codesize >> 10, datasize >> 10, rosize >> 10,
7611 		(init_data_size + init_code_size) >> 10, bss_size >> 10,
7612 		(physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10),
7613 		totalcma_pages << (PAGE_SHIFT - 10),
7614 #ifdef	CONFIG_HIGHMEM
7615 		totalhigh_pages() << (PAGE_SHIFT - 10),
7616 #endif
7617 		str ? ", " : "", str ? str : "");
7618 }
7619 
7620 /**
7621  * set_dma_reserve - set the specified number of pages reserved in the first zone
7622  * @new_dma_reserve: The number of pages to mark reserved
7623  *
7624  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
7625  * In the DMA zone, a significant percentage may be consumed by kernel image
7626  * and other unfreeable allocations which can skew the watermarks badly. This
7627  * function may optionally be used to account for unfreeable pages in the
7628  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
7629  * smaller per-cpu batchsize.
7630  */
7631 void __init set_dma_reserve(unsigned long new_dma_reserve)
7632 {
7633 	dma_reserve = new_dma_reserve;
7634 }
7635 
7636 static int page_alloc_cpu_dead(unsigned int cpu)
7637 {
7638 
7639 	lru_add_drain_cpu(cpu);
7640 	drain_pages(cpu);
7641 
7642 	/*
7643 	 * Spill the event counters of the dead processor
7644 	 * into the current processors event counters.
7645 	 * This artificially elevates the count of the current
7646 	 * processor.
7647 	 */
7648 	vm_events_fold_cpu(cpu);
7649 
7650 	/*
7651 	 * Zero the differential counters of the dead processor
7652 	 * so that the vm statistics are consistent.
7653 	 *
7654 	 * This is only okay since the processor is dead and cannot
7655 	 * race with what we are doing.
7656 	 */
7657 	cpu_vm_stats_fold(cpu);
7658 	return 0;
7659 }
7660 
7661 #ifdef CONFIG_NUMA
7662 int hashdist = HASHDIST_DEFAULT;
7663 
7664 static int __init set_hashdist(char *str)
7665 {
7666 	if (!str)
7667 		return 0;
7668 	hashdist = simple_strtoul(str, &str, 0);
7669 	return 1;
7670 }
7671 __setup("hashdist=", set_hashdist);
7672 #endif
7673 
7674 void __init page_alloc_init(void)
7675 {
7676 	int ret;
7677 
7678 #ifdef CONFIG_NUMA
7679 	if (num_node_state(N_MEMORY) == 1)
7680 		hashdist = 0;
7681 #endif
7682 
7683 	ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
7684 					"mm/page_alloc:dead", NULL,
7685 					page_alloc_cpu_dead);
7686 	WARN_ON(ret < 0);
7687 }
7688 
7689 /*
7690  * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
7691  *	or min_free_kbytes changes.
7692  */
7693 static void calculate_totalreserve_pages(void)
7694 {
7695 	struct pglist_data *pgdat;
7696 	unsigned long reserve_pages = 0;
7697 	enum zone_type i, j;
7698 
7699 	for_each_online_pgdat(pgdat) {
7700 
7701 		pgdat->totalreserve_pages = 0;
7702 
7703 		for (i = 0; i < MAX_NR_ZONES; i++) {
7704 			struct zone *zone = pgdat->node_zones + i;
7705 			long max = 0;
7706 			unsigned long managed_pages = zone_managed_pages(zone);
7707 
7708 			/* Find valid and maximum lowmem_reserve in the zone */
7709 			for (j = i; j < MAX_NR_ZONES; j++) {
7710 				if (zone->lowmem_reserve[j] > max)
7711 					max = zone->lowmem_reserve[j];
7712 			}
7713 
7714 			/* we treat the high watermark as reserved pages. */
7715 			max += high_wmark_pages(zone);
7716 
7717 			if (max > managed_pages)
7718 				max = managed_pages;
7719 
7720 			pgdat->totalreserve_pages += max;
7721 
7722 			reserve_pages += max;
7723 		}
7724 	}
7725 	totalreserve_pages = reserve_pages;
7726 }
7727 
7728 /*
7729  * setup_per_zone_lowmem_reserve - called whenever
7730  *	sysctl_lowmem_reserve_ratio changes.  Ensures that each zone
7731  *	has a correct pages reserved value, so an adequate number of
7732  *	pages are left in the zone after a successful __alloc_pages().
7733  */
7734 static void setup_per_zone_lowmem_reserve(void)
7735 {
7736 	struct pglist_data *pgdat;
7737 	enum zone_type j, idx;
7738 
7739 	for_each_online_pgdat(pgdat) {
7740 		for (j = 0; j < MAX_NR_ZONES; j++) {
7741 			struct zone *zone = pgdat->node_zones + j;
7742 			unsigned long managed_pages = zone_managed_pages(zone);
7743 
7744 			zone->lowmem_reserve[j] = 0;
7745 
7746 			idx = j;
7747 			while (idx) {
7748 				struct zone *lower_zone;
7749 
7750 				idx--;
7751 				lower_zone = pgdat->node_zones + idx;
7752 
7753 				if (!sysctl_lowmem_reserve_ratio[idx] ||
7754 				    !zone_managed_pages(lower_zone)) {
7755 					lower_zone->lowmem_reserve[j] = 0;
7756 					continue;
7757 				} else {
7758 					lower_zone->lowmem_reserve[j] =
7759 						managed_pages / sysctl_lowmem_reserve_ratio[idx];
7760 				}
7761 				managed_pages += zone_managed_pages(lower_zone);
7762 			}
7763 		}
7764 	}
7765 
7766 	/* update totalreserve_pages */
7767 	calculate_totalreserve_pages();
7768 }
7769 
7770 static void __setup_per_zone_wmarks(void)
7771 {
7772 	unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
7773 	unsigned long lowmem_pages = 0;
7774 	struct zone *zone;
7775 	unsigned long flags;
7776 
7777 	/* Calculate total number of !ZONE_HIGHMEM pages */
7778 	for_each_zone(zone) {
7779 		if (!is_highmem(zone))
7780 			lowmem_pages += zone_managed_pages(zone);
7781 	}
7782 
7783 	for_each_zone(zone) {
7784 		u64 tmp;
7785 
7786 		spin_lock_irqsave(&zone->lock, flags);
7787 		tmp = (u64)pages_min * zone_managed_pages(zone);
7788 		do_div(tmp, lowmem_pages);
7789 		if (is_highmem(zone)) {
7790 			/*
7791 			 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
7792 			 * need highmem pages, so cap pages_min to a small
7793 			 * value here.
7794 			 *
7795 			 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
7796 			 * deltas control async page reclaim, and so should
7797 			 * not be capped for highmem.
7798 			 */
7799 			unsigned long min_pages;
7800 
7801 			min_pages = zone_managed_pages(zone) / 1024;
7802 			min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
7803 			zone->_watermark[WMARK_MIN] = min_pages;
7804 		} else {
7805 			/*
7806 			 * If it's a lowmem zone, reserve a number of pages
7807 			 * proportionate to the zone's size.
7808 			 */
7809 			zone->_watermark[WMARK_MIN] = tmp;
7810 		}
7811 
7812 		/*
7813 		 * Set the kswapd watermarks distance according to the
7814 		 * scale factor in proportion to available memory, but
7815 		 * ensure a minimum size on small systems.
7816 		 */
7817 		tmp = max_t(u64, tmp >> 2,
7818 			    mult_frac(zone_managed_pages(zone),
7819 				      watermark_scale_factor, 10000));
7820 
7821 		zone->watermark_boost = 0;
7822 		zone->_watermark[WMARK_LOW]  = min_wmark_pages(zone) + tmp;
7823 		zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
7824 
7825 		spin_unlock_irqrestore(&zone->lock, flags);
7826 	}
7827 
7828 	/* update totalreserve_pages */
7829 	calculate_totalreserve_pages();
7830 }
7831 
7832 /**
7833  * setup_per_zone_wmarks - called when min_free_kbytes changes
7834  * or when memory is hot-{added|removed}
7835  *
7836  * Ensures that the watermark[min,low,high] values for each zone are set
7837  * correctly with respect to min_free_kbytes.
7838  */
7839 void setup_per_zone_wmarks(void)
7840 {
7841 	static DEFINE_SPINLOCK(lock);
7842 
7843 	spin_lock(&lock);
7844 	__setup_per_zone_wmarks();
7845 	spin_unlock(&lock);
7846 }
7847 
7848 /*
7849  * Initialise min_free_kbytes.
7850  *
7851  * For small machines we want it small (128k min).  For large machines
7852  * we want it large (256MB max).  But it is not linear, because network
7853  * bandwidth does not increase linearly with machine size.  We use
7854  *
7855  *	min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
7856  *	min_free_kbytes = sqrt(lowmem_kbytes * 16)
7857  *
7858  * which yields
7859  *
7860  * 16MB:	512k
7861  * 32MB:	724k
7862  * 64MB:	1024k
7863  * 128MB:	1448k
7864  * 256MB:	2048k
7865  * 512MB:	2896k
7866  * 1024MB:	4096k
7867  * 2048MB:	5792k
7868  * 4096MB:	8192k
7869  * 8192MB:	11584k
7870  * 16384MB:	16384k
7871  */
7872 int __meminit init_per_zone_wmark_min(void)
7873 {
7874 	unsigned long lowmem_kbytes;
7875 	int new_min_free_kbytes;
7876 
7877 	lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
7878 	new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
7879 
7880 	if (new_min_free_kbytes > user_min_free_kbytes) {
7881 		min_free_kbytes = new_min_free_kbytes;
7882 		if (min_free_kbytes < 128)
7883 			min_free_kbytes = 128;
7884 		if (min_free_kbytes > 262144)
7885 			min_free_kbytes = 262144;
7886 	} else {
7887 		pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
7888 				new_min_free_kbytes, user_min_free_kbytes);
7889 	}
7890 	setup_per_zone_wmarks();
7891 	refresh_zone_stat_thresholds();
7892 	setup_per_zone_lowmem_reserve();
7893 
7894 #ifdef CONFIG_NUMA
7895 	setup_min_unmapped_ratio();
7896 	setup_min_slab_ratio();
7897 #endif
7898 
7899 	khugepaged_min_free_kbytes_update();
7900 
7901 	return 0;
7902 }
7903 postcore_initcall(init_per_zone_wmark_min)
7904 
7905 /*
7906  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
7907  *	that we can call two helper functions whenever min_free_kbytes
7908  *	changes.
7909  */
7910 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
7911 		void *buffer, size_t *length, loff_t *ppos)
7912 {
7913 	int rc;
7914 
7915 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7916 	if (rc)
7917 		return rc;
7918 
7919 	if (write) {
7920 		user_min_free_kbytes = min_free_kbytes;
7921 		setup_per_zone_wmarks();
7922 	}
7923 	return 0;
7924 }
7925 
7926 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
7927 		void *buffer, size_t *length, loff_t *ppos)
7928 {
7929 	int rc;
7930 
7931 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7932 	if (rc)
7933 		return rc;
7934 
7935 	if (write)
7936 		setup_per_zone_wmarks();
7937 
7938 	return 0;
7939 }
7940 
7941 #ifdef CONFIG_NUMA
7942 static void setup_min_unmapped_ratio(void)
7943 {
7944 	pg_data_t *pgdat;
7945 	struct zone *zone;
7946 
7947 	for_each_online_pgdat(pgdat)
7948 		pgdat->min_unmapped_pages = 0;
7949 
7950 	for_each_zone(zone)
7951 		zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) *
7952 						         sysctl_min_unmapped_ratio) / 100;
7953 }
7954 
7955 
7956 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
7957 		void *buffer, size_t *length, loff_t *ppos)
7958 {
7959 	int rc;
7960 
7961 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7962 	if (rc)
7963 		return rc;
7964 
7965 	setup_min_unmapped_ratio();
7966 
7967 	return 0;
7968 }
7969 
7970 static void setup_min_slab_ratio(void)
7971 {
7972 	pg_data_t *pgdat;
7973 	struct zone *zone;
7974 
7975 	for_each_online_pgdat(pgdat)
7976 		pgdat->min_slab_pages = 0;
7977 
7978 	for_each_zone(zone)
7979 		zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) *
7980 						     sysctl_min_slab_ratio) / 100;
7981 }
7982 
7983 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
7984 		void *buffer, size_t *length, loff_t *ppos)
7985 {
7986 	int rc;
7987 
7988 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7989 	if (rc)
7990 		return rc;
7991 
7992 	setup_min_slab_ratio();
7993 
7994 	return 0;
7995 }
7996 #endif
7997 
7998 /*
7999  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
8000  *	proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
8001  *	whenever sysctl_lowmem_reserve_ratio changes.
8002  *
8003  * The reserve ratio obviously has absolutely no relation with the
8004  * minimum watermarks. The lowmem reserve ratio can only make sense
8005  * if in function of the boot time zone sizes.
8006  */
8007 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
8008 		void *buffer, size_t *length, loff_t *ppos)
8009 {
8010 	int i;
8011 
8012 	proc_dointvec_minmax(table, write, buffer, length, ppos);
8013 
8014 	for (i = 0; i < MAX_NR_ZONES; i++) {
8015 		if (sysctl_lowmem_reserve_ratio[i] < 1)
8016 			sysctl_lowmem_reserve_ratio[i] = 0;
8017 	}
8018 
8019 	setup_per_zone_lowmem_reserve();
8020 	return 0;
8021 }
8022 
8023 static void __zone_pcp_update(struct zone *zone)
8024 {
8025 	unsigned int cpu;
8026 
8027 	for_each_possible_cpu(cpu)
8028 		pageset_set_high_and_batch(zone,
8029 				per_cpu_ptr(zone->pageset, cpu));
8030 }
8031 
8032 /*
8033  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
8034  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
8035  * pagelist can have before it gets flushed back to buddy allocator.
8036  */
8037 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
8038 		void *buffer, size_t *length, loff_t *ppos)
8039 {
8040 	struct zone *zone;
8041 	int old_percpu_pagelist_fraction;
8042 	int ret;
8043 
8044 	mutex_lock(&pcp_batch_high_lock);
8045 	old_percpu_pagelist_fraction = percpu_pagelist_fraction;
8046 
8047 	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
8048 	if (!write || ret < 0)
8049 		goto out;
8050 
8051 	/* Sanity checking to avoid pcp imbalance */
8052 	if (percpu_pagelist_fraction &&
8053 	    percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
8054 		percpu_pagelist_fraction = old_percpu_pagelist_fraction;
8055 		ret = -EINVAL;
8056 		goto out;
8057 	}
8058 
8059 	/* No change? */
8060 	if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
8061 		goto out;
8062 
8063 	for_each_populated_zone(zone)
8064 		__zone_pcp_update(zone);
8065 out:
8066 	mutex_unlock(&pcp_batch_high_lock);
8067 	return ret;
8068 }
8069 
8070 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
8071 /*
8072  * Returns the number of pages that arch has reserved but
8073  * is not known to alloc_large_system_hash().
8074  */
8075 static unsigned long __init arch_reserved_kernel_pages(void)
8076 {
8077 	return 0;
8078 }
8079 #endif
8080 
8081 /*
8082  * Adaptive scale is meant to reduce sizes of hash tables on large memory
8083  * machines. As memory size is increased the scale is also increased but at
8084  * slower pace.  Starting from ADAPT_SCALE_BASE (64G), every time memory
8085  * quadruples the scale is increased by one, which means the size of hash table
8086  * only doubles, instead of quadrupling as well.
8087  * Because 32-bit systems cannot have large physical memory, where this scaling
8088  * makes sense, it is disabled on such platforms.
8089  */
8090 #if __BITS_PER_LONG > 32
8091 #define ADAPT_SCALE_BASE	(64ul << 30)
8092 #define ADAPT_SCALE_SHIFT	2
8093 #define ADAPT_SCALE_NPAGES	(ADAPT_SCALE_BASE >> PAGE_SHIFT)
8094 #endif
8095 
8096 /*
8097  * allocate a large system hash table from bootmem
8098  * - it is assumed that the hash table must contain an exact power-of-2
8099  *   quantity of entries
8100  * - limit is the number of hash buckets, not the total allocation size
8101  */
8102 void *__init alloc_large_system_hash(const char *tablename,
8103 				     unsigned long bucketsize,
8104 				     unsigned long numentries,
8105 				     int scale,
8106 				     int flags,
8107 				     unsigned int *_hash_shift,
8108 				     unsigned int *_hash_mask,
8109 				     unsigned long low_limit,
8110 				     unsigned long high_limit)
8111 {
8112 	unsigned long long max = high_limit;
8113 	unsigned long log2qty, size;
8114 	void *table = NULL;
8115 	gfp_t gfp_flags;
8116 	bool virt;
8117 
8118 	/* allow the kernel cmdline to have a say */
8119 	if (!numentries) {
8120 		/* round applicable memory size up to nearest megabyte */
8121 		numentries = nr_kernel_pages;
8122 		numentries -= arch_reserved_kernel_pages();
8123 
8124 		/* It isn't necessary when PAGE_SIZE >= 1MB */
8125 		if (PAGE_SHIFT < 20)
8126 			numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
8127 
8128 #if __BITS_PER_LONG > 32
8129 		if (!high_limit) {
8130 			unsigned long adapt;
8131 
8132 			for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
8133 			     adapt <<= ADAPT_SCALE_SHIFT)
8134 				scale++;
8135 		}
8136 #endif
8137 
8138 		/* limit to 1 bucket per 2^scale bytes of low memory */
8139 		if (scale > PAGE_SHIFT)
8140 			numentries >>= (scale - PAGE_SHIFT);
8141 		else
8142 			numentries <<= (PAGE_SHIFT - scale);
8143 
8144 		/* Make sure we've got at least a 0-order allocation.. */
8145 		if (unlikely(flags & HASH_SMALL)) {
8146 			/* Makes no sense without HASH_EARLY */
8147 			WARN_ON(!(flags & HASH_EARLY));
8148 			if (!(numentries >> *_hash_shift)) {
8149 				numentries = 1UL << *_hash_shift;
8150 				BUG_ON(!numentries);
8151 			}
8152 		} else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
8153 			numentries = PAGE_SIZE / bucketsize;
8154 	}
8155 	numentries = roundup_pow_of_two(numentries);
8156 
8157 	/* limit allocation size to 1/16 total memory by default */
8158 	if (max == 0) {
8159 		max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
8160 		do_div(max, bucketsize);
8161 	}
8162 	max = min(max, 0x80000000ULL);
8163 
8164 	if (numentries < low_limit)
8165 		numentries = low_limit;
8166 	if (numentries > max)
8167 		numentries = max;
8168 
8169 	log2qty = ilog2(numentries);
8170 
8171 	gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
8172 	do {
8173 		virt = false;
8174 		size = bucketsize << log2qty;
8175 		if (flags & HASH_EARLY) {
8176 			if (flags & HASH_ZERO)
8177 				table = memblock_alloc(size, SMP_CACHE_BYTES);
8178 			else
8179 				table = memblock_alloc_raw(size,
8180 							   SMP_CACHE_BYTES);
8181 		} else if (get_order(size) >= MAX_ORDER || hashdist) {
8182 			table = __vmalloc(size, gfp_flags);
8183 			virt = true;
8184 		} else {
8185 			/*
8186 			 * If bucketsize is not a power-of-two, we may free
8187 			 * some pages at the end of hash table which
8188 			 * alloc_pages_exact() automatically does
8189 			 */
8190 			table = alloc_pages_exact(size, gfp_flags);
8191 			kmemleak_alloc(table, size, 1, gfp_flags);
8192 		}
8193 	} while (!table && size > PAGE_SIZE && --log2qty);
8194 
8195 	if (!table)
8196 		panic("Failed to allocate %s hash table\n", tablename);
8197 
8198 	pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n",
8199 		tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size,
8200 		virt ? "vmalloc" : "linear");
8201 
8202 	if (_hash_shift)
8203 		*_hash_shift = log2qty;
8204 	if (_hash_mask)
8205 		*_hash_mask = (1 << log2qty) - 1;
8206 
8207 	return table;
8208 }
8209 
8210 /*
8211  * This function checks whether pageblock includes unmovable pages or not.
8212  *
8213  * PageLRU check without isolation or lru_lock could race so that
8214  * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
8215  * check without lock_page also may miss some movable non-lru pages at
8216  * race condition. So you can't expect this function should be exact.
8217  *
8218  * Returns a page without holding a reference. If the caller wants to
8219  * dereference that page (e.g., dumping), it has to make sure that it
8220  * cannot get removed (e.g., via memory unplug) concurrently.
8221  *
8222  */
8223 struct page *has_unmovable_pages(struct zone *zone, struct page *page,
8224 				 int migratetype, int flags)
8225 {
8226 	unsigned long iter = 0;
8227 	unsigned long pfn = page_to_pfn(page);
8228 	unsigned long offset = pfn % pageblock_nr_pages;
8229 
8230 	if (is_migrate_cma_page(page)) {
8231 		/*
8232 		 * CMA allocations (alloc_contig_range) really need to mark
8233 		 * isolate CMA pageblocks even when they are not movable in fact
8234 		 * so consider them movable here.
8235 		 */
8236 		if (is_migrate_cma(migratetype))
8237 			return NULL;
8238 
8239 		return page;
8240 	}
8241 
8242 	for (; iter < pageblock_nr_pages - offset; iter++) {
8243 		if (!pfn_valid_within(pfn + iter))
8244 			continue;
8245 
8246 		page = pfn_to_page(pfn + iter);
8247 
8248 		/*
8249 		 * Both, bootmem allocations and memory holes are marked
8250 		 * PG_reserved and are unmovable. We can even have unmovable
8251 		 * allocations inside ZONE_MOVABLE, for example when
8252 		 * specifying "movablecore".
8253 		 */
8254 		if (PageReserved(page))
8255 			return page;
8256 
8257 		/*
8258 		 * If the zone is movable and we have ruled out all reserved
8259 		 * pages then it should be reasonably safe to assume the rest
8260 		 * is movable.
8261 		 */
8262 		if (zone_idx(zone) == ZONE_MOVABLE)
8263 			continue;
8264 
8265 		/*
8266 		 * Hugepages are not in LRU lists, but they're movable.
8267 		 * THPs are on the LRU, but need to be counted as #small pages.
8268 		 * We need not scan over tail pages because we don't
8269 		 * handle each tail page individually in migration.
8270 		 */
8271 		if (PageHuge(page) || PageTransCompound(page)) {
8272 			struct page *head = compound_head(page);
8273 			unsigned int skip_pages;
8274 
8275 			if (PageHuge(page)) {
8276 				if (!hugepage_migration_supported(page_hstate(head)))
8277 					return page;
8278 			} else if (!PageLRU(head) && !__PageMovable(head)) {
8279 				return page;
8280 			}
8281 
8282 			skip_pages = compound_nr(head) - (page - head);
8283 			iter += skip_pages - 1;
8284 			continue;
8285 		}
8286 
8287 		/*
8288 		 * We can't use page_count without pin a page
8289 		 * because another CPU can free compound page.
8290 		 * This check already skips compound tails of THP
8291 		 * because their page->_refcount is zero at all time.
8292 		 */
8293 		if (!page_ref_count(page)) {
8294 			if (PageBuddy(page))
8295 				iter += (1 << page_order(page)) - 1;
8296 			continue;
8297 		}
8298 
8299 		/*
8300 		 * The HWPoisoned page may be not in buddy system, and
8301 		 * page_count() is not 0.
8302 		 */
8303 		if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
8304 			continue;
8305 
8306 		/*
8307 		 * We treat all PageOffline() pages as movable when offlining
8308 		 * to give drivers a chance to decrement their reference count
8309 		 * in MEM_GOING_OFFLINE in order to indicate that these pages
8310 		 * can be offlined as there are no direct references anymore.
8311 		 * For actually unmovable PageOffline() where the driver does
8312 		 * not support this, we will fail later when trying to actually
8313 		 * move these pages that still have a reference count > 0.
8314 		 * (false negatives in this function only)
8315 		 */
8316 		if ((flags & MEMORY_OFFLINE) && PageOffline(page))
8317 			continue;
8318 
8319 		if (__PageMovable(page) || PageLRU(page))
8320 			continue;
8321 
8322 		/*
8323 		 * If there are RECLAIMABLE pages, we need to check
8324 		 * it.  But now, memory offline itself doesn't call
8325 		 * shrink_node_slabs() and it still to be fixed.
8326 		 */
8327 		return page;
8328 	}
8329 	return NULL;
8330 }
8331 
8332 #ifdef CONFIG_CONTIG_ALLOC
8333 static unsigned long pfn_max_align_down(unsigned long pfn)
8334 {
8335 	return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
8336 			     pageblock_nr_pages) - 1);
8337 }
8338 
8339 static unsigned long pfn_max_align_up(unsigned long pfn)
8340 {
8341 	return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
8342 				pageblock_nr_pages));
8343 }
8344 
8345 /* [start, end) must belong to a single zone. */
8346 static int __alloc_contig_migrate_range(struct compact_control *cc,
8347 					unsigned long start, unsigned long end)
8348 {
8349 	/* This function is based on compact_zone() from compaction.c. */
8350 	unsigned int nr_reclaimed;
8351 	unsigned long pfn = start;
8352 	unsigned int tries = 0;
8353 	int ret = 0;
8354 	struct migration_target_control mtc = {
8355 		.nid = zone_to_nid(cc->zone),
8356 		.gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
8357 	};
8358 
8359 	migrate_prep();
8360 
8361 	while (pfn < end || !list_empty(&cc->migratepages)) {
8362 		if (fatal_signal_pending(current)) {
8363 			ret = -EINTR;
8364 			break;
8365 		}
8366 
8367 		if (list_empty(&cc->migratepages)) {
8368 			cc->nr_migratepages = 0;
8369 			pfn = isolate_migratepages_range(cc, pfn, end);
8370 			if (!pfn) {
8371 				ret = -EINTR;
8372 				break;
8373 			}
8374 			tries = 0;
8375 		} else if (++tries == 5) {
8376 			ret = ret < 0 ? ret : -EBUSY;
8377 			break;
8378 		}
8379 
8380 		nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
8381 							&cc->migratepages);
8382 		cc->nr_migratepages -= nr_reclaimed;
8383 
8384 		ret = migrate_pages(&cc->migratepages, alloc_migration_target,
8385 				NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE);
8386 	}
8387 	if (ret < 0) {
8388 		putback_movable_pages(&cc->migratepages);
8389 		return ret;
8390 	}
8391 	return 0;
8392 }
8393 
8394 /**
8395  * alloc_contig_range() -- tries to allocate given range of pages
8396  * @start:	start PFN to allocate
8397  * @end:	one-past-the-last PFN to allocate
8398  * @migratetype:	migratetype of the underlaying pageblocks (either
8399  *			#MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
8400  *			in range must have the same migratetype and it must
8401  *			be either of the two.
8402  * @gfp_mask:	GFP mask to use during compaction
8403  *
8404  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
8405  * aligned.  The PFN range must belong to a single zone.
8406  *
8407  * The first thing this routine does is attempt to MIGRATE_ISOLATE all
8408  * pageblocks in the range.  Once isolated, the pageblocks should not
8409  * be modified by others.
8410  *
8411  * Return: zero on success or negative error code.  On success all
8412  * pages which PFN is in [start, end) are allocated for the caller and
8413  * need to be freed with free_contig_range().
8414  */
8415 int alloc_contig_range(unsigned long start, unsigned long end,
8416 		       unsigned migratetype, gfp_t gfp_mask)
8417 {
8418 	unsigned long outer_start, outer_end;
8419 	unsigned int order;
8420 	int ret = 0;
8421 
8422 	struct compact_control cc = {
8423 		.nr_migratepages = 0,
8424 		.order = -1,
8425 		.zone = page_zone(pfn_to_page(start)),
8426 		.mode = MIGRATE_SYNC,
8427 		.ignore_skip_hint = true,
8428 		.no_set_skip_hint = true,
8429 		.gfp_mask = current_gfp_context(gfp_mask),
8430 		.alloc_contig = true,
8431 	};
8432 	INIT_LIST_HEAD(&cc.migratepages);
8433 
8434 	/*
8435 	 * What we do here is we mark all pageblocks in range as
8436 	 * MIGRATE_ISOLATE.  Because pageblock and max order pages may
8437 	 * have different sizes, and due to the way page allocator
8438 	 * work, we align the range to biggest of the two pages so
8439 	 * that page allocator won't try to merge buddies from
8440 	 * different pageblocks and change MIGRATE_ISOLATE to some
8441 	 * other migration type.
8442 	 *
8443 	 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
8444 	 * migrate the pages from an unaligned range (ie. pages that
8445 	 * we are interested in).  This will put all the pages in
8446 	 * range back to page allocator as MIGRATE_ISOLATE.
8447 	 *
8448 	 * When this is done, we take the pages in range from page
8449 	 * allocator removing them from the buddy system.  This way
8450 	 * page allocator will never consider using them.
8451 	 *
8452 	 * This lets us mark the pageblocks back as
8453 	 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
8454 	 * aligned range but not in the unaligned, original range are
8455 	 * put back to page allocator so that buddy can use them.
8456 	 */
8457 
8458 	ret = start_isolate_page_range(pfn_max_align_down(start),
8459 				       pfn_max_align_up(end), migratetype, 0);
8460 	if (ret < 0)
8461 		return ret;
8462 
8463 	/*
8464 	 * In case of -EBUSY, we'd like to know which page causes problem.
8465 	 * So, just fall through. test_pages_isolated() has a tracepoint
8466 	 * which will report the busy page.
8467 	 *
8468 	 * It is possible that busy pages could become available before
8469 	 * the call to test_pages_isolated, and the range will actually be
8470 	 * allocated.  So, if we fall through be sure to clear ret so that
8471 	 * -EBUSY is not accidentally used or returned to caller.
8472 	 */
8473 	ret = __alloc_contig_migrate_range(&cc, start, end);
8474 	if (ret && ret != -EBUSY)
8475 		goto done;
8476 	ret =0;
8477 
8478 	/*
8479 	 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
8480 	 * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
8481 	 * more, all pages in [start, end) are free in page allocator.
8482 	 * What we are going to do is to allocate all pages from
8483 	 * [start, end) (that is remove them from page allocator).
8484 	 *
8485 	 * The only problem is that pages at the beginning and at the
8486 	 * end of interesting range may be not aligned with pages that
8487 	 * page allocator holds, ie. they can be part of higher order
8488 	 * pages.  Because of this, we reserve the bigger range and
8489 	 * once this is done free the pages we are not interested in.
8490 	 *
8491 	 * We don't have to hold zone->lock here because the pages are
8492 	 * isolated thus they won't get removed from buddy.
8493 	 */
8494 
8495 	lru_add_drain_all();
8496 
8497 	order = 0;
8498 	outer_start = start;
8499 	while (!PageBuddy(pfn_to_page(outer_start))) {
8500 		if (++order >= MAX_ORDER) {
8501 			outer_start = start;
8502 			break;
8503 		}
8504 		outer_start &= ~0UL << order;
8505 	}
8506 
8507 	if (outer_start != start) {
8508 		order = page_order(pfn_to_page(outer_start));
8509 
8510 		/*
8511 		 * outer_start page could be small order buddy page and
8512 		 * it doesn't include start page. Adjust outer_start
8513 		 * in this case to report failed page properly
8514 		 * on tracepoint in test_pages_isolated()
8515 		 */
8516 		if (outer_start + (1UL << order) <= start)
8517 			outer_start = start;
8518 	}
8519 
8520 	/* Make sure the range is really isolated. */
8521 	if (test_pages_isolated(outer_start, end, 0)) {
8522 		pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n",
8523 			__func__, outer_start, end);
8524 		ret = -EBUSY;
8525 		goto done;
8526 	}
8527 
8528 	/* Grab isolated pages from freelists. */
8529 	outer_end = isolate_freepages_range(&cc, outer_start, end);
8530 	if (!outer_end) {
8531 		ret = -EBUSY;
8532 		goto done;
8533 	}
8534 
8535 	/* Free head and tail (if any) */
8536 	if (start != outer_start)
8537 		free_contig_range(outer_start, start - outer_start);
8538 	if (end != outer_end)
8539 		free_contig_range(end, outer_end - end);
8540 
8541 done:
8542 	undo_isolate_page_range(pfn_max_align_down(start),
8543 				pfn_max_align_up(end), migratetype);
8544 	return ret;
8545 }
8546 EXPORT_SYMBOL(alloc_contig_range);
8547 
8548 static int __alloc_contig_pages(unsigned long start_pfn,
8549 				unsigned long nr_pages, gfp_t gfp_mask)
8550 {
8551 	unsigned long end_pfn = start_pfn + nr_pages;
8552 
8553 	return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE,
8554 				  gfp_mask);
8555 }
8556 
8557 static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn,
8558 				   unsigned long nr_pages)
8559 {
8560 	unsigned long i, end_pfn = start_pfn + nr_pages;
8561 	struct page *page;
8562 
8563 	for (i = start_pfn; i < end_pfn; i++) {
8564 		page = pfn_to_online_page(i);
8565 		if (!page)
8566 			return false;
8567 
8568 		if (page_zone(page) != z)
8569 			return false;
8570 
8571 		if (PageReserved(page))
8572 			return false;
8573 
8574 		if (page_count(page) > 0)
8575 			return false;
8576 
8577 		if (PageHuge(page))
8578 			return false;
8579 	}
8580 	return true;
8581 }
8582 
8583 static bool zone_spans_last_pfn(const struct zone *zone,
8584 				unsigned long start_pfn, unsigned long nr_pages)
8585 {
8586 	unsigned long last_pfn = start_pfn + nr_pages - 1;
8587 
8588 	return zone_spans_pfn(zone, last_pfn);
8589 }
8590 
8591 /**
8592  * alloc_contig_pages() -- tries to find and allocate contiguous range of pages
8593  * @nr_pages:	Number of contiguous pages to allocate
8594  * @gfp_mask:	GFP mask to limit search and used during compaction
8595  * @nid:	Target node
8596  * @nodemask:	Mask for other possible nodes
8597  *
8598  * This routine is a wrapper around alloc_contig_range(). It scans over zones
8599  * on an applicable zonelist to find a contiguous pfn range which can then be
8600  * tried for allocation with alloc_contig_range(). This routine is intended
8601  * for allocation requests which can not be fulfilled with the buddy allocator.
8602  *
8603  * The allocated memory is always aligned to a page boundary. If nr_pages is a
8604  * power of two then the alignment is guaranteed to be to the given nr_pages
8605  * (e.g. 1GB request would be aligned to 1GB).
8606  *
8607  * Allocated pages can be freed with free_contig_range() or by manually calling
8608  * __free_page() on each allocated page.
8609  *
8610  * Return: pointer to contiguous pages on success, or NULL if not successful.
8611  */
8612 struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
8613 				int nid, nodemask_t *nodemask)
8614 {
8615 	unsigned long ret, pfn, flags;
8616 	struct zonelist *zonelist;
8617 	struct zone *zone;
8618 	struct zoneref *z;
8619 
8620 	zonelist = node_zonelist(nid, gfp_mask);
8621 	for_each_zone_zonelist_nodemask(zone, z, zonelist,
8622 					gfp_zone(gfp_mask), nodemask) {
8623 		spin_lock_irqsave(&zone->lock, flags);
8624 
8625 		pfn = ALIGN(zone->zone_start_pfn, nr_pages);
8626 		while (zone_spans_last_pfn(zone, pfn, nr_pages)) {
8627 			if (pfn_range_valid_contig(zone, pfn, nr_pages)) {
8628 				/*
8629 				 * We release the zone lock here because
8630 				 * alloc_contig_range() will also lock the zone
8631 				 * at some point. If there's an allocation
8632 				 * spinning on this lock, it may win the race
8633 				 * and cause alloc_contig_range() to fail...
8634 				 */
8635 				spin_unlock_irqrestore(&zone->lock, flags);
8636 				ret = __alloc_contig_pages(pfn, nr_pages,
8637 							gfp_mask);
8638 				if (!ret)
8639 					return pfn_to_page(pfn);
8640 				spin_lock_irqsave(&zone->lock, flags);
8641 			}
8642 			pfn += nr_pages;
8643 		}
8644 		spin_unlock_irqrestore(&zone->lock, flags);
8645 	}
8646 	return NULL;
8647 }
8648 #endif /* CONFIG_CONTIG_ALLOC */
8649 
8650 void free_contig_range(unsigned long pfn, unsigned int nr_pages)
8651 {
8652 	unsigned int count = 0;
8653 
8654 	for (; nr_pages--; pfn++) {
8655 		struct page *page = pfn_to_page(pfn);
8656 
8657 		count += page_count(page) != 1;
8658 		__free_page(page);
8659 	}
8660 	WARN(count != 0, "%d pages are still in use!\n", count);
8661 }
8662 EXPORT_SYMBOL(free_contig_range);
8663 
8664 /*
8665  * The zone indicated has a new number of managed_pages; batch sizes and percpu
8666  * page high values need to be recalulated.
8667  */
8668 void __meminit zone_pcp_update(struct zone *zone)
8669 {
8670 	mutex_lock(&pcp_batch_high_lock);
8671 	__zone_pcp_update(zone);
8672 	mutex_unlock(&pcp_batch_high_lock);
8673 }
8674 
8675 void zone_pcp_reset(struct zone *zone)
8676 {
8677 	unsigned long flags;
8678 	int cpu;
8679 	struct per_cpu_pageset *pset;
8680 
8681 	/* avoid races with drain_pages()  */
8682 	local_irq_save(flags);
8683 	if (zone->pageset != &boot_pageset) {
8684 		for_each_online_cpu(cpu) {
8685 			pset = per_cpu_ptr(zone->pageset, cpu);
8686 			drain_zonestat(zone, pset);
8687 		}
8688 		free_percpu(zone->pageset);
8689 		zone->pageset = &boot_pageset;
8690 	}
8691 	local_irq_restore(flags);
8692 }
8693 
8694 #ifdef CONFIG_MEMORY_HOTREMOVE
8695 /*
8696  * All pages in the range must be in a single zone and isolated
8697  * before calling this.
8698  */
8699 unsigned long
8700 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
8701 {
8702 	struct page *page;
8703 	struct zone *zone;
8704 	unsigned int order;
8705 	unsigned long pfn;
8706 	unsigned long flags;
8707 	unsigned long offlined_pages = 0;
8708 
8709 	/* find the first valid pfn */
8710 	for (pfn = start_pfn; pfn < end_pfn; pfn++)
8711 		if (pfn_valid(pfn))
8712 			break;
8713 	if (pfn == end_pfn)
8714 		return offlined_pages;
8715 
8716 	offline_mem_sections(pfn, end_pfn);
8717 	zone = page_zone(pfn_to_page(pfn));
8718 	spin_lock_irqsave(&zone->lock, flags);
8719 	pfn = start_pfn;
8720 	while (pfn < end_pfn) {
8721 		if (!pfn_valid(pfn)) {
8722 			pfn++;
8723 			continue;
8724 		}
8725 		page = pfn_to_page(pfn);
8726 		/*
8727 		 * The HWPoisoned page may be not in buddy system, and
8728 		 * page_count() is not 0.
8729 		 */
8730 		if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
8731 			pfn++;
8732 			offlined_pages++;
8733 			continue;
8734 		}
8735 		/*
8736 		 * At this point all remaining PageOffline() pages have a
8737 		 * reference count of 0 and can simply be skipped.
8738 		 */
8739 		if (PageOffline(page)) {
8740 			BUG_ON(page_count(page));
8741 			BUG_ON(PageBuddy(page));
8742 			pfn++;
8743 			offlined_pages++;
8744 			continue;
8745 		}
8746 
8747 		BUG_ON(page_count(page));
8748 		BUG_ON(!PageBuddy(page));
8749 		order = page_order(page);
8750 		offlined_pages += 1 << order;
8751 		del_page_from_free_list(page, zone, order);
8752 		pfn += (1 << order);
8753 	}
8754 	spin_unlock_irqrestore(&zone->lock, flags);
8755 
8756 	return offlined_pages;
8757 }
8758 #endif
8759 
8760 bool is_free_buddy_page(struct page *page)
8761 {
8762 	struct zone *zone = page_zone(page);
8763 	unsigned long pfn = page_to_pfn(page);
8764 	unsigned long flags;
8765 	unsigned int order;
8766 
8767 	spin_lock_irqsave(&zone->lock, flags);
8768 	for (order = 0; order < MAX_ORDER; order++) {
8769 		struct page *page_head = page - (pfn & ((1 << order) - 1));
8770 
8771 		if (PageBuddy(page_head) && page_order(page_head) >= order)
8772 			break;
8773 	}
8774 	spin_unlock_irqrestore(&zone->lock, flags);
8775 
8776 	return order < MAX_ORDER;
8777 }
8778 
8779 #ifdef CONFIG_MEMORY_FAILURE
8780 /*
8781  * Set PG_hwpoison flag if a given page is confirmed to be a free page.  This
8782  * test is performed under the zone lock to prevent a race against page
8783  * allocation.
8784  */
8785 bool set_hwpoison_free_buddy_page(struct page *page)
8786 {
8787 	struct zone *zone = page_zone(page);
8788 	unsigned long pfn = page_to_pfn(page);
8789 	unsigned long flags;
8790 	unsigned int order;
8791 	bool hwpoisoned = false;
8792 
8793 	spin_lock_irqsave(&zone->lock, flags);
8794 	for (order = 0; order < MAX_ORDER; order++) {
8795 		struct page *page_head = page - (pfn & ((1 << order) - 1));
8796 
8797 		if (PageBuddy(page_head) && page_order(page_head) >= order) {
8798 			if (!TestSetPageHWPoison(page))
8799 				hwpoisoned = true;
8800 			break;
8801 		}
8802 	}
8803 	spin_unlock_irqrestore(&zone->lock, flags);
8804 
8805 	return hwpoisoned;
8806 }
8807 #endif
8808