xref: /linux/mm/page_alloc.c (revision 8eecf1c9929aef24e9e75280a39ed1ba3c64fb71)
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/swapops.h>
23 #include <linux/interrupt.h>
24 #include <linux/pagemap.h>
25 #include <linux/jiffies.h>
26 #include <linux/memblock.h>
27 #include <linux/compiler.h>
28 #include <linux/kernel.h>
29 #include <linux/kasan.h>
30 #include <linux/module.h>
31 #include <linux/suspend.h>
32 #include <linux/pagevec.h>
33 #include <linux/blkdev.h>
34 #include <linux/slab.h>
35 #include <linux/ratelimit.h>
36 #include <linux/oom.h>
37 #include <linux/topology.h>
38 #include <linux/sysctl.h>
39 #include <linux/cpu.h>
40 #include <linux/cpuset.h>
41 #include <linux/memory_hotplug.h>
42 #include <linux/nodemask.h>
43 #include <linux/vmalloc.h>
44 #include <linux/vmstat.h>
45 #include <linux/mempolicy.h>
46 #include <linux/memremap.h>
47 #include <linux/stop_machine.h>
48 #include <linux/random.h>
49 #include <linux/sort.h>
50 #include <linux/pfn.h>
51 #include <linux/backing-dev.h>
52 #include <linux/fault-inject.h>
53 #include <linux/page-isolation.h>
54 #include <linux/debugobjects.h>
55 #include <linux/kmemleak.h>
56 #include <linux/compaction.h>
57 #include <trace/events/kmem.h>
58 #include <trace/events/oom.h>
59 #include <linux/prefetch.h>
60 #include <linux/mm_inline.h>
61 #include <linux/mmu_notifier.h>
62 #include <linux/migrate.h>
63 #include <linux/hugetlb.h>
64 #include <linux/sched/rt.h>
65 #include <linux/sched/mm.h>
66 #include <linux/page_owner.h>
67 #include <linux/page_table_check.h>
68 #include <linux/kthread.h>
69 #include <linux/memcontrol.h>
70 #include <linux/ftrace.h>
71 #include <linux/lockdep.h>
72 #include <linux/nmi.h>
73 #include <linux/psi.h>
74 #include <linux/padata.h>
75 #include <linux/khugepaged.h>
76 #include <linux/buffer_head.h>
77 #include <linux/delayacct.h>
78 #include <asm/sections.h>
79 #include <asm/tlbflush.h>
80 #include <asm/div64.h>
81 #include "internal.h"
82 #include "shuffle.h"
83 #include "page_reporting.h"
84 
85 /* Free Page Internal flags: for internal, non-pcp variants of free_pages(). */
86 typedef int __bitwise fpi_t;
87 
88 /* No special request */
89 #define FPI_NONE		((__force fpi_t)0)
90 
91 /*
92  * Skip free page reporting notification for the (possibly merged) page.
93  * This does not hinder free page reporting from grabbing the page,
94  * reporting it and marking it "reported" -  it only skips notifying
95  * the free page reporting infrastructure about a newly freed page. For
96  * example, used when temporarily pulling a page from a freelist and
97  * putting it back unmodified.
98  */
99 #define FPI_SKIP_REPORT_NOTIFY	((__force fpi_t)BIT(0))
100 
101 /*
102  * Place the (possibly merged) page to the tail of the freelist. Will ignore
103  * page shuffling (relevant code - e.g., memory onlining - is expected to
104  * shuffle the whole zone).
105  *
106  * Note: No code should rely on this flag for correctness - it's purely
107  *       to allow for optimizations when handing back either fresh pages
108  *       (memory onlining) or untouched pages (page isolation, free page
109  *       reporting).
110  */
111 #define FPI_TO_TAIL		((__force fpi_t)BIT(1))
112 
113 /*
114  * Don't poison memory with KASAN (only for the tag-based modes).
115  * During boot, all non-reserved memblock memory is exposed to page_alloc.
116  * Poisoning all that memory lengthens boot time, especially on systems with
117  * large amount of RAM. This flag is used to skip that poisoning.
118  * This is only done for the tag-based KASAN modes, as those are able to
119  * detect memory corruptions with the memory tags assigned by default.
120  * All memory allocated normally after boot gets poisoned as usual.
121  */
122 #define FPI_SKIP_KASAN_POISON	((__force fpi_t)BIT(2))
123 
124 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
125 static DEFINE_MUTEX(pcp_batch_high_lock);
126 #define MIN_PERCPU_PAGELIST_HIGH_FRACTION (8)
127 
128 struct pagesets {
129 	local_lock_t lock;
130 };
131 static DEFINE_PER_CPU(struct pagesets, pagesets) = {
132 	.lock = INIT_LOCAL_LOCK(lock),
133 };
134 
135 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
136 DEFINE_PER_CPU(int, numa_node);
137 EXPORT_PER_CPU_SYMBOL(numa_node);
138 #endif
139 
140 DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key);
141 
142 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
143 /*
144  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
145  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
146  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
147  * defined in <linux/topology.h>.
148  */
149 DEFINE_PER_CPU(int, _numa_mem_);		/* Kernel "local memory" node */
150 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
151 #endif
152 
153 /* work_structs for global per-cpu drains */
154 struct pcpu_drain {
155 	struct zone *zone;
156 	struct work_struct work;
157 };
158 static DEFINE_MUTEX(pcpu_drain_mutex);
159 static DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain);
160 
161 #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
162 volatile unsigned long latent_entropy __latent_entropy;
163 EXPORT_SYMBOL(latent_entropy);
164 #endif
165 
166 /*
167  * Array of node states.
168  */
169 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
170 	[N_POSSIBLE] = NODE_MASK_ALL,
171 	[N_ONLINE] = { { [0] = 1UL } },
172 #ifndef CONFIG_NUMA
173 	[N_NORMAL_MEMORY] = { { [0] = 1UL } },
174 #ifdef CONFIG_HIGHMEM
175 	[N_HIGH_MEMORY] = { { [0] = 1UL } },
176 #endif
177 	[N_MEMORY] = { { [0] = 1UL } },
178 	[N_CPU] = { { [0] = 1UL } },
179 #endif	/* NUMA */
180 };
181 EXPORT_SYMBOL(node_states);
182 
183 atomic_long_t _totalram_pages __read_mostly;
184 EXPORT_SYMBOL(_totalram_pages);
185 unsigned long totalreserve_pages __read_mostly;
186 unsigned long totalcma_pages __read_mostly;
187 
188 int percpu_pagelist_high_fraction;
189 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
190 DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc);
191 EXPORT_SYMBOL(init_on_alloc);
192 
193 DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free);
194 EXPORT_SYMBOL(init_on_free);
195 
196 static bool _init_on_alloc_enabled_early __read_mostly
197 				= IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON);
198 static int __init early_init_on_alloc(char *buf)
199 {
200 
201 	return kstrtobool(buf, &_init_on_alloc_enabled_early);
202 }
203 early_param("init_on_alloc", early_init_on_alloc);
204 
205 static bool _init_on_free_enabled_early __read_mostly
206 				= IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON);
207 static int __init early_init_on_free(char *buf)
208 {
209 	return kstrtobool(buf, &_init_on_free_enabled_early);
210 }
211 early_param("init_on_free", early_init_on_free);
212 
213 /*
214  * A cached value of the page's pageblock's migratetype, used when the page is
215  * put on a pcplist. Used to avoid the pageblock migratetype lookup when
216  * freeing from pcplists in most cases, at the cost of possibly becoming stale.
217  * Also the migratetype set in the page does not necessarily match the pcplist
218  * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
219  * other index - this ensures that it will be put on the correct CMA freelist.
220  */
221 static inline int get_pcppage_migratetype(struct page *page)
222 {
223 	return page->index;
224 }
225 
226 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
227 {
228 	page->index = migratetype;
229 }
230 
231 #ifdef CONFIG_PM_SLEEP
232 /*
233  * The following functions are used by the suspend/hibernate code to temporarily
234  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
235  * while devices are suspended.  To avoid races with the suspend/hibernate code,
236  * they should always be called with system_transition_mutex held
237  * (gfp_allowed_mask also should only be modified with system_transition_mutex
238  * held, unless the suspend/hibernate code is guaranteed not to run in parallel
239  * with that modification).
240  */
241 
242 static gfp_t saved_gfp_mask;
243 
244 void pm_restore_gfp_mask(void)
245 {
246 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
247 	if (saved_gfp_mask) {
248 		gfp_allowed_mask = saved_gfp_mask;
249 		saved_gfp_mask = 0;
250 	}
251 }
252 
253 void pm_restrict_gfp_mask(void)
254 {
255 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
256 	WARN_ON(saved_gfp_mask);
257 	saved_gfp_mask = gfp_allowed_mask;
258 	gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
259 }
260 
261 bool pm_suspended_storage(void)
262 {
263 	if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
264 		return false;
265 	return true;
266 }
267 #endif /* CONFIG_PM_SLEEP */
268 
269 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
270 unsigned int pageblock_order __read_mostly;
271 #endif
272 
273 static void __free_pages_ok(struct page *page, unsigned int order,
274 			    fpi_t fpi_flags);
275 
276 /*
277  * results with 256, 32 in the lowmem_reserve sysctl:
278  *	1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
279  *	1G machine -> (16M dma, 784M normal, 224M high)
280  *	NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
281  *	HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
282  *	HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
283  *
284  * TBD: should special case ZONE_DMA32 machines here - in those we normally
285  * don't need any ZONE_NORMAL reservation
286  */
287 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = {
288 #ifdef CONFIG_ZONE_DMA
289 	[ZONE_DMA] = 256,
290 #endif
291 #ifdef CONFIG_ZONE_DMA32
292 	[ZONE_DMA32] = 256,
293 #endif
294 	[ZONE_NORMAL] = 32,
295 #ifdef CONFIG_HIGHMEM
296 	[ZONE_HIGHMEM] = 0,
297 #endif
298 	[ZONE_MOVABLE] = 0,
299 };
300 
301 static char * const zone_names[MAX_NR_ZONES] = {
302 #ifdef CONFIG_ZONE_DMA
303 	 "DMA",
304 #endif
305 #ifdef CONFIG_ZONE_DMA32
306 	 "DMA32",
307 #endif
308 	 "Normal",
309 #ifdef CONFIG_HIGHMEM
310 	 "HighMem",
311 #endif
312 	 "Movable",
313 #ifdef CONFIG_ZONE_DEVICE
314 	 "Device",
315 #endif
316 };
317 
318 const char * const migratetype_names[MIGRATE_TYPES] = {
319 	"Unmovable",
320 	"Movable",
321 	"Reclaimable",
322 	"HighAtomic",
323 #ifdef CONFIG_CMA
324 	"CMA",
325 #endif
326 #ifdef CONFIG_MEMORY_ISOLATION
327 	"Isolate",
328 #endif
329 };
330 
331 compound_page_dtor * const compound_page_dtors[NR_COMPOUND_DTORS] = {
332 	[NULL_COMPOUND_DTOR] = NULL,
333 	[COMPOUND_PAGE_DTOR] = free_compound_page,
334 #ifdef CONFIG_HUGETLB_PAGE
335 	[HUGETLB_PAGE_DTOR] = free_huge_page,
336 #endif
337 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
338 	[TRANSHUGE_PAGE_DTOR] = free_transhuge_page,
339 #endif
340 };
341 
342 int min_free_kbytes = 1024;
343 int user_min_free_kbytes = -1;
344 int watermark_boost_factor __read_mostly = 15000;
345 int watermark_scale_factor = 10;
346 
347 static unsigned long nr_kernel_pages __initdata;
348 static unsigned long nr_all_pages __initdata;
349 static unsigned long dma_reserve __initdata;
350 
351 static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata;
352 static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata;
353 static unsigned long required_kernelcore __initdata;
354 static unsigned long required_kernelcore_percent __initdata;
355 static unsigned long required_movablecore __initdata;
356 static unsigned long required_movablecore_percent __initdata;
357 static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata;
358 static bool mirrored_kernelcore __meminitdata;
359 
360 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
361 int movable_zone;
362 EXPORT_SYMBOL(movable_zone);
363 
364 #if MAX_NUMNODES > 1
365 unsigned int nr_node_ids __read_mostly = MAX_NUMNODES;
366 unsigned int nr_online_nodes __read_mostly = 1;
367 EXPORT_SYMBOL(nr_node_ids);
368 EXPORT_SYMBOL(nr_online_nodes);
369 #endif
370 
371 int page_group_by_mobility_disabled __read_mostly;
372 
373 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
374 /*
375  * During boot we initialize deferred pages on-demand, as needed, but once
376  * page_alloc_init_late() has finished, the deferred pages are all initialized,
377  * and we can permanently disable that path.
378  */
379 static DEFINE_STATIC_KEY_TRUE(deferred_pages);
380 
381 static inline bool deferred_pages_enabled(void)
382 {
383 	return static_branch_unlikely(&deferred_pages);
384 }
385 
386 /* Returns true if the struct page for the pfn is uninitialised */
387 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
388 {
389 	int nid = early_pfn_to_nid(pfn);
390 
391 	if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
392 		return true;
393 
394 	return false;
395 }
396 
397 /*
398  * Returns true when the remaining initialisation should be deferred until
399  * later in the boot cycle when it can be parallelised.
400  */
401 static bool __meminit
402 defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
403 {
404 	static unsigned long prev_end_pfn, nr_initialised;
405 
406 	/*
407 	 * prev_end_pfn static that contains the end of previous zone
408 	 * No need to protect because called very early in boot before smp_init.
409 	 */
410 	if (prev_end_pfn != end_pfn) {
411 		prev_end_pfn = end_pfn;
412 		nr_initialised = 0;
413 	}
414 
415 	/* Always populate low zones for address-constrained allocations */
416 	if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
417 		return false;
418 
419 	if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX)
420 		return true;
421 	/*
422 	 * We start only with one section of pages, more pages are added as
423 	 * needed until the rest of deferred pages are initialized.
424 	 */
425 	nr_initialised++;
426 	if ((nr_initialised > PAGES_PER_SECTION) &&
427 	    (pfn & (PAGES_PER_SECTION - 1)) == 0) {
428 		NODE_DATA(nid)->first_deferred_pfn = pfn;
429 		return true;
430 	}
431 	return false;
432 }
433 #else
434 static inline bool deferred_pages_enabled(void)
435 {
436 	return false;
437 }
438 
439 static inline bool early_page_uninitialised(unsigned long pfn)
440 {
441 	return false;
442 }
443 
444 static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
445 {
446 	return false;
447 }
448 #endif
449 
450 /* Return a pointer to the bitmap storing bits affecting a block of pages */
451 static inline unsigned long *get_pageblock_bitmap(const struct page *page,
452 							unsigned long pfn)
453 {
454 #ifdef CONFIG_SPARSEMEM
455 	return section_to_usemap(__pfn_to_section(pfn));
456 #else
457 	return page_zone(page)->pageblock_flags;
458 #endif /* CONFIG_SPARSEMEM */
459 }
460 
461 static inline int pfn_to_bitidx(const struct page *page, unsigned long pfn)
462 {
463 #ifdef CONFIG_SPARSEMEM
464 	pfn &= (PAGES_PER_SECTION-1);
465 #else
466 	pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
467 #endif /* CONFIG_SPARSEMEM */
468 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
469 }
470 
471 static __always_inline
472 unsigned long __get_pfnblock_flags_mask(const struct page *page,
473 					unsigned long pfn,
474 					unsigned long mask)
475 {
476 	unsigned long *bitmap;
477 	unsigned long bitidx, word_bitidx;
478 	unsigned long word;
479 
480 	bitmap = get_pageblock_bitmap(page, pfn);
481 	bitidx = pfn_to_bitidx(page, pfn);
482 	word_bitidx = bitidx / BITS_PER_LONG;
483 	bitidx &= (BITS_PER_LONG-1);
484 
485 	word = bitmap[word_bitidx];
486 	return (word >> bitidx) & mask;
487 }
488 
489 /**
490  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
491  * @page: The page within the block of interest
492  * @pfn: The target page frame number
493  * @mask: mask of bits that the caller is interested in
494  *
495  * Return: pageblock_bits flags
496  */
497 unsigned long get_pfnblock_flags_mask(const struct page *page,
498 					unsigned long pfn, unsigned long mask)
499 {
500 	return __get_pfnblock_flags_mask(page, pfn, mask);
501 }
502 
503 static __always_inline int get_pfnblock_migratetype(const struct page *page,
504 					unsigned long pfn)
505 {
506 	return __get_pfnblock_flags_mask(page, pfn, MIGRATETYPE_MASK);
507 }
508 
509 /**
510  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
511  * @page: The page within the block of interest
512  * @flags: The flags to set
513  * @pfn: The target page frame number
514  * @mask: mask of bits that the caller is interested in
515  */
516 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
517 					unsigned long pfn,
518 					unsigned long mask)
519 {
520 	unsigned long *bitmap;
521 	unsigned long bitidx, word_bitidx;
522 	unsigned long old_word, word;
523 
524 	BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
525 	BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits));
526 
527 	bitmap = get_pageblock_bitmap(page, pfn);
528 	bitidx = pfn_to_bitidx(page, pfn);
529 	word_bitidx = bitidx / BITS_PER_LONG;
530 	bitidx &= (BITS_PER_LONG-1);
531 
532 	VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
533 
534 	mask <<= bitidx;
535 	flags <<= bitidx;
536 
537 	word = READ_ONCE(bitmap[word_bitidx]);
538 	for (;;) {
539 		old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
540 		if (word == old_word)
541 			break;
542 		word = old_word;
543 	}
544 }
545 
546 void set_pageblock_migratetype(struct page *page, int migratetype)
547 {
548 	if (unlikely(page_group_by_mobility_disabled &&
549 		     migratetype < MIGRATE_PCPTYPES))
550 		migratetype = MIGRATE_UNMOVABLE;
551 
552 	set_pfnblock_flags_mask(page, (unsigned long)migratetype,
553 				page_to_pfn(page), MIGRATETYPE_MASK);
554 }
555 
556 #ifdef CONFIG_DEBUG_VM
557 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
558 {
559 	int ret = 0;
560 	unsigned seq;
561 	unsigned long pfn = page_to_pfn(page);
562 	unsigned long sp, start_pfn;
563 
564 	do {
565 		seq = zone_span_seqbegin(zone);
566 		start_pfn = zone->zone_start_pfn;
567 		sp = zone->spanned_pages;
568 		if (!zone_spans_pfn(zone, pfn))
569 			ret = 1;
570 	} while (zone_span_seqretry(zone, seq));
571 
572 	if (ret)
573 		pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
574 			pfn, zone_to_nid(zone), zone->name,
575 			start_pfn, start_pfn + sp);
576 
577 	return ret;
578 }
579 
580 static int page_is_consistent(struct zone *zone, struct page *page)
581 {
582 	if (zone != page_zone(page))
583 		return 0;
584 
585 	return 1;
586 }
587 /*
588  * Temporary debugging check for pages not lying within a given zone.
589  */
590 static int __maybe_unused bad_range(struct zone *zone, struct page *page)
591 {
592 	if (page_outside_zone_boundaries(zone, page))
593 		return 1;
594 	if (!page_is_consistent(zone, page))
595 		return 1;
596 
597 	return 0;
598 }
599 #else
600 static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
601 {
602 	return 0;
603 }
604 #endif
605 
606 static void bad_page(struct page *page, const char *reason)
607 {
608 	static unsigned long resume;
609 	static unsigned long nr_shown;
610 	static unsigned long nr_unshown;
611 
612 	/*
613 	 * Allow a burst of 60 reports, then keep quiet for that minute;
614 	 * or allow a steady drip of one report per second.
615 	 */
616 	if (nr_shown == 60) {
617 		if (time_before(jiffies, resume)) {
618 			nr_unshown++;
619 			goto out;
620 		}
621 		if (nr_unshown) {
622 			pr_alert(
623 			      "BUG: Bad page state: %lu messages suppressed\n",
624 				nr_unshown);
625 			nr_unshown = 0;
626 		}
627 		nr_shown = 0;
628 	}
629 	if (nr_shown++ == 0)
630 		resume = jiffies + 60 * HZ;
631 
632 	pr_alert("BUG: Bad page state in process %s  pfn:%05lx\n",
633 		current->comm, page_to_pfn(page));
634 	dump_page(page, reason);
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 static inline unsigned int order_to_pindex(int migratetype, int order)
645 {
646 	int base = order;
647 
648 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
649 	if (order > PAGE_ALLOC_COSTLY_ORDER) {
650 		VM_BUG_ON(order != pageblock_order);
651 		base = PAGE_ALLOC_COSTLY_ORDER + 1;
652 	}
653 #else
654 	VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER);
655 #endif
656 
657 	return (MIGRATE_PCPTYPES * base) + migratetype;
658 }
659 
660 static inline int pindex_to_order(unsigned int pindex)
661 {
662 	int order = pindex / MIGRATE_PCPTYPES;
663 
664 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
665 	if (order > PAGE_ALLOC_COSTLY_ORDER)
666 		order = pageblock_order;
667 #else
668 	VM_BUG_ON(order > PAGE_ALLOC_COSTLY_ORDER);
669 #endif
670 
671 	return order;
672 }
673 
674 static inline bool pcp_allowed_order(unsigned int order)
675 {
676 	if (order <= PAGE_ALLOC_COSTLY_ORDER)
677 		return true;
678 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
679 	if (order == pageblock_order)
680 		return true;
681 #endif
682 	return false;
683 }
684 
685 static inline void free_the_page(struct page *page, unsigned int order)
686 {
687 	if (pcp_allowed_order(order))		/* Via pcp? */
688 		free_unref_page(page, order);
689 	else
690 		__free_pages_ok(page, order, FPI_NONE);
691 }
692 
693 /*
694  * Higher-order pages are called "compound pages".  They are structured thusly:
695  *
696  * The first PAGE_SIZE page is called the "head page" and have PG_head set.
697  *
698  * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
699  * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
700  *
701  * The first tail page's ->compound_dtor holds the offset in array of compound
702  * page destructors. See compound_page_dtors.
703  *
704  * The first tail page's ->compound_order holds the order of allocation.
705  * This usage means that zero-order pages may not be compound.
706  */
707 
708 void free_compound_page(struct page *page)
709 {
710 	mem_cgroup_uncharge(page_folio(page));
711 	free_the_page(page, compound_order(page));
712 }
713 
714 static void prep_compound_head(struct page *page, unsigned int order)
715 {
716 	set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
717 	set_compound_order(page, order);
718 	atomic_set(compound_mapcount_ptr(page), -1);
719 	atomic_set(compound_pincount_ptr(page), 0);
720 }
721 
722 static void prep_compound_tail(struct page *head, int tail_idx)
723 {
724 	struct page *p = head + tail_idx;
725 
726 	p->mapping = TAIL_MAPPING;
727 	set_compound_head(p, head);
728 }
729 
730 void prep_compound_page(struct page *page, unsigned int order)
731 {
732 	int i;
733 	int nr_pages = 1 << order;
734 
735 	__SetPageHead(page);
736 	for (i = 1; i < nr_pages; i++)
737 		prep_compound_tail(page, i);
738 
739 	prep_compound_head(page, order);
740 }
741 
742 #ifdef CONFIG_DEBUG_PAGEALLOC
743 unsigned int _debug_guardpage_minorder;
744 
745 bool _debug_pagealloc_enabled_early __read_mostly
746 			= IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
747 EXPORT_SYMBOL(_debug_pagealloc_enabled_early);
748 DEFINE_STATIC_KEY_FALSE(_debug_pagealloc_enabled);
749 EXPORT_SYMBOL(_debug_pagealloc_enabled);
750 
751 DEFINE_STATIC_KEY_FALSE(_debug_guardpage_enabled);
752 
753 static int __init early_debug_pagealloc(char *buf)
754 {
755 	return kstrtobool(buf, &_debug_pagealloc_enabled_early);
756 }
757 early_param("debug_pagealloc", early_debug_pagealloc);
758 
759 static int __init debug_guardpage_minorder_setup(char *buf)
760 {
761 	unsigned long res;
762 
763 	if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
764 		pr_err("Bad debug_guardpage_minorder value\n");
765 		return 0;
766 	}
767 	_debug_guardpage_minorder = res;
768 	pr_info("Setting debug_guardpage_minorder to %lu\n", res);
769 	return 0;
770 }
771 early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
772 
773 static inline bool set_page_guard(struct zone *zone, struct page *page,
774 				unsigned int order, int migratetype)
775 {
776 	if (!debug_guardpage_enabled())
777 		return false;
778 
779 	if (order >= debug_guardpage_minorder())
780 		return false;
781 
782 	__SetPageGuard(page);
783 	INIT_LIST_HEAD(&page->lru);
784 	set_page_private(page, order);
785 	/* Guard pages are not available for any usage */
786 	__mod_zone_freepage_state(zone, -(1 << order), migratetype);
787 
788 	return true;
789 }
790 
791 static inline void clear_page_guard(struct zone *zone, struct page *page,
792 				unsigned int order, int migratetype)
793 {
794 	if (!debug_guardpage_enabled())
795 		return;
796 
797 	__ClearPageGuard(page);
798 
799 	set_page_private(page, 0);
800 	if (!is_migrate_isolate(migratetype))
801 		__mod_zone_freepage_state(zone, (1 << order), migratetype);
802 }
803 #else
804 static inline bool set_page_guard(struct zone *zone, struct page *page,
805 			unsigned int order, int migratetype) { return false; }
806 static inline void clear_page_guard(struct zone *zone, struct page *page,
807 				unsigned int order, int migratetype) {}
808 #endif
809 
810 /*
811  * Enable static keys related to various memory debugging and hardening options.
812  * Some override others, and depend on early params that are evaluated in the
813  * order of appearance. So we need to first gather the full picture of what was
814  * enabled, and then make decisions.
815  */
816 void init_mem_debugging_and_hardening(void)
817 {
818 	bool page_poisoning_requested = false;
819 
820 #ifdef CONFIG_PAGE_POISONING
821 	/*
822 	 * Page poisoning is debug page alloc for some arches. If
823 	 * either of those options are enabled, enable poisoning.
824 	 */
825 	if (page_poisoning_enabled() ||
826 	     (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) &&
827 	      debug_pagealloc_enabled())) {
828 		static_branch_enable(&_page_poisoning_enabled);
829 		page_poisoning_requested = true;
830 	}
831 #endif
832 
833 	if ((_init_on_alloc_enabled_early || _init_on_free_enabled_early) &&
834 	    page_poisoning_requested) {
835 		pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, "
836 			"will take precedence over init_on_alloc and init_on_free\n");
837 		_init_on_alloc_enabled_early = false;
838 		_init_on_free_enabled_early = false;
839 	}
840 
841 	if (_init_on_alloc_enabled_early)
842 		static_branch_enable(&init_on_alloc);
843 	else
844 		static_branch_disable(&init_on_alloc);
845 
846 	if (_init_on_free_enabled_early)
847 		static_branch_enable(&init_on_free);
848 	else
849 		static_branch_disable(&init_on_free);
850 
851 #ifdef CONFIG_DEBUG_PAGEALLOC
852 	if (!debug_pagealloc_enabled())
853 		return;
854 
855 	static_branch_enable(&_debug_pagealloc_enabled);
856 
857 	if (!debug_guardpage_minorder())
858 		return;
859 
860 	static_branch_enable(&_debug_guardpage_enabled);
861 #endif
862 }
863 
864 static inline void set_buddy_order(struct page *page, unsigned int order)
865 {
866 	set_page_private(page, order);
867 	__SetPageBuddy(page);
868 }
869 
870 /*
871  * This function checks whether a page is free && is the buddy
872  * we can coalesce a page and its buddy if
873  * (a) the buddy is not in a hole (check before calling!) &&
874  * (b) the buddy is in the buddy system &&
875  * (c) a page and its buddy have the same order &&
876  * (d) a page and its buddy are in the same zone.
877  *
878  * For recording whether a page is in the buddy system, we set PageBuddy.
879  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
880  *
881  * For recording page's order, we use page_private(page).
882  */
883 static inline bool page_is_buddy(struct page *page, struct page *buddy,
884 							unsigned int order)
885 {
886 	if (!page_is_guard(buddy) && !PageBuddy(buddy))
887 		return false;
888 
889 	if (buddy_order(buddy) != order)
890 		return false;
891 
892 	/*
893 	 * zone check is done late to avoid uselessly calculating
894 	 * zone/node ids for pages that could never merge.
895 	 */
896 	if (page_zone_id(page) != page_zone_id(buddy))
897 		return false;
898 
899 	VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
900 
901 	return true;
902 }
903 
904 #ifdef CONFIG_COMPACTION
905 static inline struct capture_control *task_capc(struct zone *zone)
906 {
907 	struct capture_control *capc = current->capture_control;
908 
909 	return unlikely(capc) &&
910 		!(current->flags & PF_KTHREAD) &&
911 		!capc->page &&
912 		capc->cc->zone == zone ? capc : NULL;
913 }
914 
915 static inline bool
916 compaction_capture(struct capture_control *capc, struct page *page,
917 		   int order, int migratetype)
918 {
919 	if (!capc || order != capc->cc->order)
920 		return false;
921 
922 	/* Do not accidentally pollute CMA or isolated regions*/
923 	if (is_migrate_cma(migratetype) ||
924 	    is_migrate_isolate(migratetype))
925 		return false;
926 
927 	/*
928 	 * Do not let lower order allocations pollute a movable pageblock.
929 	 * This might let an unmovable request use a reclaimable pageblock
930 	 * and vice-versa but no more than normal fallback logic which can
931 	 * have trouble finding a high-order free page.
932 	 */
933 	if (order < pageblock_order && migratetype == MIGRATE_MOVABLE)
934 		return false;
935 
936 	capc->page = page;
937 	return true;
938 }
939 
940 #else
941 static inline struct capture_control *task_capc(struct zone *zone)
942 {
943 	return NULL;
944 }
945 
946 static inline bool
947 compaction_capture(struct capture_control *capc, struct page *page,
948 		   int order, int migratetype)
949 {
950 	return false;
951 }
952 #endif /* CONFIG_COMPACTION */
953 
954 /* Used for pages not on another list */
955 static inline void add_to_free_list(struct page *page, struct zone *zone,
956 				    unsigned int order, int migratetype)
957 {
958 	struct free_area *area = &zone->free_area[order];
959 
960 	list_add(&page->lru, &area->free_list[migratetype]);
961 	area->nr_free++;
962 }
963 
964 /* Used for pages not on another list */
965 static inline void add_to_free_list_tail(struct page *page, struct zone *zone,
966 					 unsigned int order, int migratetype)
967 {
968 	struct free_area *area = &zone->free_area[order];
969 
970 	list_add_tail(&page->lru, &area->free_list[migratetype]);
971 	area->nr_free++;
972 }
973 
974 /*
975  * Used for pages which are on another list. Move the pages to the tail
976  * of the list - so the moved pages won't immediately be considered for
977  * allocation again (e.g., optimization for memory onlining).
978  */
979 static inline void move_to_free_list(struct page *page, struct zone *zone,
980 				     unsigned int order, int migratetype)
981 {
982 	struct free_area *area = &zone->free_area[order];
983 
984 	list_move_tail(&page->lru, &area->free_list[migratetype]);
985 }
986 
987 static inline void del_page_from_free_list(struct page *page, struct zone *zone,
988 					   unsigned int order)
989 {
990 	/* clear reported state and update reported page count */
991 	if (page_reported(page))
992 		__ClearPageReported(page);
993 
994 	list_del(&page->lru);
995 	__ClearPageBuddy(page);
996 	set_page_private(page, 0);
997 	zone->free_area[order].nr_free--;
998 }
999 
1000 /*
1001  * If this is not the largest possible page, check if the buddy
1002  * of the next-highest order is free. If it is, it's possible
1003  * that pages are being freed that will coalesce soon. In case,
1004  * that is happening, add the free page to the tail of the list
1005  * so it's less likely to be used soon and more likely to be merged
1006  * as a higher order page
1007  */
1008 static inline bool
1009 buddy_merge_likely(unsigned long pfn, unsigned long buddy_pfn,
1010 		   struct page *page, unsigned int order)
1011 {
1012 	struct page *higher_page, *higher_buddy;
1013 	unsigned long combined_pfn;
1014 
1015 	if (order >= MAX_ORDER - 2)
1016 		return false;
1017 
1018 	combined_pfn = buddy_pfn & pfn;
1019 	higher_page = page + (combined_pfn - pfn);
1020 	buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
1021 	higher_buddy = higher_page + (buddy_pfn - combined_pfn);
1022 
1023 	return page_is_buddy(higher_page, higher_buddy, order + 1);
1024 }
1025 
1026 /*
1027  * Freeing function for a buddy system allocator.
1028  *
1029  * The concept of a buddy system is to maintain direct-mapped table
1030  * (containing bit values) for memory blocks of various "orders".
1031  * The bottom level table contains the map for the smallest allocatable
1032  * units of memory (here, pages), and each level above it describes
1033  * pairs of units from the levels below, hence, "buddies".
1034  * At a high level, all that happens here is marking the table entry
1035  * at the bottom level available, and propagating the changes upward
1036  * as necessary, plus some accounting needed to play nicely with other
1037  * parts of the VM system.
1038  * At each level, we keep a list of pages, which are heads of continuous
1039  * free pages of length of (1 << order) and marked with PageBuddy.
1040  * Page's order is recorded in page_private(page) field.
1041  * So when we are allocating or freeing one, we can derive the state of the
1042  * other.  That is, if we allocate a small block, and both were
1043  * free, the remainder of the region must be split into blocks.
1044  * If a block is freed, and its buddy is also free, then this
1045  * triggers coalescing into a block of larger size.
1046  *
1047  * -- nyc
1048  */
1049 
1050 static inline void __free_one_page(struct page *page,
1051 		unsigned long pfn,
1052 		struct zone *zone, unsigned int order,
1053 		int migratetype, fpi_t fpi_flags)
1054 {
1055 	struct capture_control *capc = task_capc(zone);
1056 	unsigned int max_order = pageblock_order;
1057 	unsigned long buddy_pfn;
1058 	unsigned long combined_pfn;
1059 	struct page *buddy;
1060 	bool to_tail;
1061 
1062 	VM_BUG_ON(!zone_is_initialized(zone));
1063 	VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
1064 
1065 	VM_BUG_ON(migratetype == -1);
1066 	if (likely(!is_migrate_isolate(migratetype)))
1067 		__mod_zone_freepage_state(zone, 1 << order, migratetype);
1068 
1069 	VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
1070 	VM_BUG_ON_PAGE(bad_range(zone, page), page);
1071 
1072 continue_merging:
1073 	while (order < max_order) {
1074 		if (compaction_capture(capc, page, order, migratetype)) {
1075 			__mod_zone_freepage_state(zone, -(1 << order),
1076 								migratetype);
1077 			return;
1078 		}
1079 		buddy_pfn = __find_buddy_pfn(pfn, order);
1080 		buddy = page + (buddy_pfn - pfn);
1081 
1082 		if (!page_is_buddy(page, buddy, order))
1083 			goto done_merging;
1084 		/*
1085 		 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
1086 		 * merge with it and move up one order.
1087 		 */
1088 		if (page_is_guard(buddy))
1089 			clear_page_guard(zone, buddy, order, migratetype);
1090 		else
1091 			del_page_from_free_list(buddy, zone, order);
1092 		combined_pfn = buddy_pfn & pfn;
1093 		page = page + (combined_pfn - pfn);
1094 		pfn = combined_pfn;
1095 		order++;
1096 	}
1097 	if (order < MAX_ORDER - 1) {
1098 		/* If we are here, it means order is >= pageblock_order.
1099 		 * We want to prevent merge between freepages on pageblock
1100 		 * without fallbacks and normal pageblock. Without this,
1101 		 * pageblock isolation could cause incorrect freepage or CMA
1102 		 * accounting or HIGHATOMIC accounting.
1103 		 *
1104 		 * We don't want to hit this code for the more frequent
1105 		 * low-order merging.
1106 		 */
1107 		int buddy_mt;
1108 
1109 		buddy_pfn = __find_buddy_pfn(pfn, order);
1110 		buddy = page + (buddy_pfn - pfn);
1111 
1112 		if (!page_is_buddy(page, buddy, order))
1113 			goto done_merging;
1114 		buddy_mt = get_pageblock_migratetype(buddy);
1115 
1116 		if (migratetype != buddy_mt
1117 				&& (!migratetype_is_mergeable(migratetype) ||
1118 					!migratetype_is_mergeable(buddy_mt)))
1119 			goto done_merging;
1120 		max_order = order + 1;
1121 		goto continue_merging;
1122 	}
1123 
1124 done_merging:
1125 	set_buddy_order(page, order);
1126 
1127 	if (fpi_flags & FPI_TO_TAIL)
1128 		to_tail = true;
1129 	else if (is_shuffle_order(order))
1130 		to_tail = shuffle_pick_tail();
1131 	else
1132 		to_tail = buddy_merge_likely(pfn, buddy_pfn, page, order);
1133 
1134 	if (to_tail)
1135 		add_to_free_list_tail(page, zone, order, migratetype);
1136 	else
1137 		add_to_free_list(page, zone, order, migratetype);
1138 
1139 	/* Notify page reporting subsystem of freed page */
1140 	if (!(fpi_flags & FPI_SKIP_REPORT_NOTIFY))
1141 		page_reporting_notify_free(order);
1142 }
1143 
1144 /*
1145  * A bad page could be due to a number of fields. Instead of multiple branches,
1146  * try and check multiple fields with one check. The caller must do a detailed
1147  * check if necessary.
1148  */
1149 static inline bool page_expected_state(struct page *page,
1150 					unsigned long check_flags)
1151 {
1152 	if (unlikely(atomic_read(&page->_mapcount) != -1))
1153 		return false;
1154 
1155 	if (unlikely((unsigned long)page->mapping |
1156 			page_ref_count(page) |
1157 #ifdef CONFIG_MEMCG
1158 			page->memcg_data |
1159 #endif
1160 			(page->flags & check_flags)))
1161 		return false;
1162 
1163 	return true;
1164 }
1165 
1166 static const char *page_bad_reason(struct page *page, unsigned long flags)
1167 {
1168 	const char *bad_reason = NULL;
1169 
1170 	if (unlikely(atomic_read(&page->_mapcount) != -1))
1171 		bad_reason = "nonzero mapcount";
1172 	if (unlikely(page->mapping != NULL))
1173 		bad_reason = "non-NULL mapping";
1174 	if (unlikely(page_ref_count(page) != 0))
1175 		bad_reason = "nonzero _refcount";
1176 	if (unlikely(page->flags & flags)) {
1177 		if (flags == PAGE_FLAGS_CHECK_AT_PREP)
1178 			bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag(s) set";
1179 		else
1180 			bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
1181 	}
1182 #ifdef CONFIG_MEMCG
1183 	if (unlikely(page->memcg_data))
1184 		bad_reason = "page still charged to cgroup";
1185 #endif
1186 	return bad_reason;
1187 }
1188 
1189 static void check_free_page_bad(struct page *page)
1190 {
1191 	bad_page(page,
1192 		 page_bad_reason(page, PAGE_FLAGS_CHECK_AT_FREE));
1193 }
1194 
1195 static inline int check_free_page(struct page *page)
1196 {
1197 	if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
1198 		return 0;
1199 
1200 	/* Something has gone sideways, find it */
1201 	check_free_page_bad(page);
1202 	return 1;
1203 }
1204 
1205 static int free_tail_pages_check(struct page *head_page, struct page *page)
1206 {
1207 	int ret = 1;
1208 
1209 	/*
1210 	 * We rely page->lru.next never has bit 0 set, unless the page
1211 	 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
1212 	 */
1213 	BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
1214 
1215 	if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
1216 		ret = 0;
1217 		goto out;
1218 	}
1219 	switch (page - head_page) {
1220 	case 1:
1221 		/* the first tail page: ->mapping may be compound_mapcount() */
1222 		if (unlikely(compound_mapcount(page))) {
1223 			bad_page(page, "nonzero compound_mapcount");
1224 			goto out;
1225 		}
1226 		break;
1227 	case 2:
1228 		/*
1229 		 * the second tail page: ->mapping is
1230 		 * deferred_list.next -- ignore value.
1231 		 */
1232 		break;
1233 	default:
1234 		if (page->mapping != TAIL_MAPPING) {
1235 			bad_page(page, "corrupted mapping in tail page");
1236 			goto out;
1237 		}
1238 		break;
1239 	}
1240 	if (unlikely(!PageTail(page))) {
1241 		bad_page(page, "PageTail not set");
1242 		goto out;
1243 	}
1244 	if (unlikely(compound_head(page) != head_page)) {
1245 		bad_page(page, "compound_head not consistent");
1246 		goto out;
1247 	}
1248 	ret = 0;
1249 out:
1250 	page->mapping = NULL;
1251 	clear_compound_head(page);
1252 	return ret;
1253 }
1254 
1255 /*
1256  * Skip KASAN memory poisoning when either:
1257  *
1258  * 1. Deferred memory initialization has not yet completed,
1259  *    see the explanation below.
1260  * 2. Skipping poisoning is requested via FPI_SKIP_KASAN_POISON,
1261  *    see the comment next to it.
1262  * 3. Skipping poisoning is requested via __GFP_SKIP_KASAN_POISON,
1263  *    see the comment next to it.
1264  *
1265  * Poisoning pages during deferred memory init will greatly lengthen the
1266  * process and cause problem in large memory systems as the deferred pages
1267  * initialization is done with interrupt disabled.
1268  *
1269  * Assuming that there will be no reference to those newly initialized
1270  * pages before they are ever allocated, this should have no effect on
1271  * KASAN memory tracking as the poison will be properly inserted at page
1272  * allocation time. The only corner case is when pages are allocated by
1273  * on-demand allocation and then freed again before the deferred pages
1274  * initialization is done, but this is not likely to happen.
1275  */
1276 static inline bool should_skip_kasan_poison(struct page *page, fpi_t fpi_flags)
1277 {
1278 	return deferred_pages_enabled() ||
1279 	       (!IS_ENABLED(CONFIG_KASAN_GENERIC) &&
1280 		(fpi_flags & FPI_SKIP_KASAN_POISON)) ||
1281 	       PageSkipKASanPoison(page);
1282 }
1283 
1284 static void kernel_init_free_pages(struct page *page, int numpages)
1285 {
1286 	int i;
1287 
1288 	/* s390's use of memset() could override KASAN redzones. */
1289 	kasan_disable_current();
1290 	for (i = 0; i < numpages; i++) {
1291 		u8 tag = page_kasan_tag(page + i);
1292 		page_kasan_tag_reset(page + i);
1293 		clear_highpage(page + i);
1294 		page_kasan_tag_set(page + i, tag);
1295 	}
1296 	kasan_enable_current();
1297 }
1298 
1299 static __always_inline bool free_pages_prepare(struct page *page,
1300 			unsigned int order, bool check_free, fpi_t fpi_flags)
1301 {
1302 	int bad = 0;
1303 	bool init = want_init_on_free();
1304 
1305 	VM_BUG_ON_PAGE(PageTail(page), page);
1306 
1307 	trace_mm_page_free(page, order);
1308 
1309 	if (unlikely(PageHWPoison(page)) && !order) {
1310 		/*
1311 		 * Do not let hwpoison pages hit pcplists/buddy
1312 		 * Untie memcg state and reset page's owner
1313 		 */
1314 		if (memcg_kmem_enabled() && PageMemcgKmem(page))
1315 			__memcg_kmem_uncharge_page(page, order);
1316 		reset_page_owner(page, order);
1317 		page_table_check_free(page, order);
1318 		return false;
1319 	}
1320 
1321 	/*
1322 	 * Check tail pages before head page information is cleared to
1323 	 * avoid checking PageCompound for order-0 pages.
1324 	 */
1325 	if (unlikely(order)) {
1326 		bool compound = PageCompound(page);
1327 		int i;
1328 
1329 		VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1330 
1331 		if (compound) {
1332 			ClearPageDoubleMap(page);
1333 			ClearPageHasHWPoisoned(page);
1334 		}
1335 		for (i = 1; i < (1 << order); i++) {
1336 			if (compound)
1337 				bad += free_tail_pages_check(page, page + i);
1338 			if (unlikely(check_free_page(page + i))) {
1339 				bad++;
1340 				continue;
1341 			}
1342 			(page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1343 		}
1344 	}
1345 	if (PageMappingFlags(page))
1346 		page->mapping = NULL;
1347 	if (memcg_kmem_enabled() && PageMemcgKmem(page))
1348 		__memcg_kmem_uncharge_page(page, order);
1349 	if (check_free)
1350 		bad += check_free_page(page);
1351 	if (bad)
1352 		return false;
1353 
1354 	page_cpupid_reset_last(page);
1355 	page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1356 	reset_page_owner(page, order);
1357 	page_table_check_free(page, order);
1358 
1359 	if (!PageHighMem(page)) {
1360 		debug_check_no_locks_freed(page_address(page),
1361 					   PAGE_SIZE << order);
1362 		debug_check_no_obj_freed(page_address(page),
1363 					   PAGE_SIZE << order);
1364 	}
1365 
1366 	kernel_poison_pages(page, 1 << order);
1367 
1368 	/*
1369 	 * As memory initialization might be integrated into KASAN,
1370 	 * KASAN poisoning and memory initialization code must be
1371 	 * kept together to avoid discrepancies in behavior.
1372 	 *
1373 	 * With hardware tag-based KASAN, memory tags must be set before the
1374 	 * page becomes unavailable via debug_pagealloc or arch_free_page.
1375 	 */
1376 	if (!should_skip_kasan_poison(page, fpi_flags)) {
1377 		kasan_poison_pages(page, order, init);
1378 
1379 		/* Memory is already initialized if KASAN did it internally. */
1380 		if (kasan_has_integrated_init())
1381 			init = false;
1382 	}
1383 	if (init)
1384 		kernel_init_free_pages(page, 1 << order);
1385 
1386 	/*
1387 	 * arch_free_page() can make the page's contents inaccessible.  s390
1388 	 * does this.  So nothing which can access the page's contents should
1389 	 * happen after this.
1390 	 */
1391 	arch_free_page(page, order);
1392 
1393 	debug_pagealloc_unmap_pages(page, 1 << order);
1394 
1395 	return true;
1396 }
1397 
1398 #ifdef CONFIG_DEBUG_VM
1399 /*
1400  * With DEBUG_VM enabled, order-0 pages are checked immediately when being freed
1401  * to pcp lists. With debug_pagealloc also enabled, they are also rechecked when
1402  * moved from pcp lists to free lists.
1403  */
1404 static bool free_pcp_prepare(struct page *page, unsigned int order)
1405 {
1406 	return free_pages_prepare(page, order, true, FPI_NONE);
1407 }
1408 
1409 static bool bulkfree_pcp_prepare(struct page *page)
1410 {
1411 	if (debug_pagealloc_enabled_static())
1412 		return check_free_page(page);
1413 	else
1414 		return false;
1415 }
1416 #else
1417 /*
1418  * With DEBUG_VM disabled, order-0 pages being freed are checked only when
1419  * moving from pcp lists to free list in order to reduce overhead. With
1420  * debug_pagealloc enabled, they are checked also immediately when being freed
1421  * to the pcp lists.
1422  */
1423 static bool free_pcp_prepare(struct page *page, unsigned int order)
1424 {
1425 	if (debug_pagealloc_enabled_static())
1426 		return free_pages_prepare(page, order, true, FPI_NONE);
1427 	else
1428 		return free_pages_prepare(page, order, false, FPI_NONE);
1429 }
1430 
1431 static bool bulkfree_pcp_prepare(struct page *page)
1432 {
1433 	return check_free_page(page);
1434 }
1435 #endif /* CONFIG_DEBUG_VM */
1436 
1437 /*
1438  * Frees a number of pages from the PCP lists
1439  * Assumes all pages on list are in same zone.
1440  * count is the number of pages to free.
1441  */
1442 static void free_pcppages_bulk(struct zone *zone, int count,
1443 					struct per_cpu_pages *pcp,
1444 					int pindex)
1445 {
1446 	int min_pindex = 0;
1447 	int max_pindex = NR_PCP_LISTS - 1;
1448 	unsigned int order;
1449 	bool isolated_pageblocks;
1450 	struct page *page;
1451 
1452 	/*
1453 	 * Ensure proper count is passed which otherwise would stuck in the
1454 	 * below while (list_empty(list)) loop.
1455 	 */
1456 	count = min(pcp->count, count);
1457 
1458 	/* Ensure requested pindex is drained first. */
1459 	pindex = pindex - 1;
1460 
1461 	/*
1462 	 * local_lock_irq held so equivalent to spin_lock_irqsave for
1463 	 * both PREEMPT_RT and non-PREEMPT_RT configurations.
1464 	 */
1465 	spin_lock(&zone->lock);
1466 	isolated_pageblocks = has_isolate_pageblock(zone);
1467 
1468 	while (count > 0) {
1469 		struct list_head *list;
1470 		int nr_pages;
1471 
1472 		/* Remove pages from lists in a round-robin fashion. */
1473 		do {
1474 			if (++pindex > max_pindex)
1475 				pindex = min_pindex;
1476 			list = &pcp->lists[pindex];
1477 			if (!list_empty(list))
1478 				break;
1479 
1480 			if (pindex == max_pindex)
1481 				max_pindex--;
1482 			if (pindex == min_pindex)
1483 				min_pindex++;
1484 		} while (1);
1485 
1486 		order = pindex_to_order(pindex);
1487 		nr_pages = 1 << order;
1488 		BUILD_BUG_ON(MAX_ORDER >= (1<<NR_PCP_ORDER_WIDTH));
1489 		do {
1490 			int mt;
1491 
1492 			page = list_last_entry(list, struct page, lru);
1493 			mt = get_pcppage_migratetype(page);
1494 
1495 			/* must delete to avoid corrupting pcp list */
1496 			list_del(&page->lru);
1497 			count -= nr_pages;
1498 			pcp->count -= nr_pages;
1499 
1500 			if (bulkfree_pcp_prepare(page))
1501 				continue;
1502 
1503 			/* MIGRATE_ISOLATE page should not go to pcplists */
1504 			VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1505 			/* Pageblock could have been isolated meanwhile */
1506 			if (unlikely(isolated_pageblocks))
1507 				mt = get_pageblock_migratetype(page);
1508 
1509 			__free_one_page(page, page_to_pfn(page), zone, order, mt, FPI_NONE);
1510 			trace_mm_page_pcpu_drain(page, order, mt);
1511 		} while (count > 0 && !list_empty(list));
1512 	}
1513 
1514 	spin_unlock(&zone->lock);
1515 }
1516 
1517 static void free_one_page(struct zone *zone,
1518 				struct page *page, unsigned long pfn,
1519 				unsigned int order,
1520 				int migratetype, fpi_t fpi_flags)
1521 {
1522 	unsigned long flags;
1523 
1524 	spin_lock_irqsave(&zone->lock, flags);
1525 	if (unlikely(has_isolate_pageblock(zone) ||
1526 		is_migrate_isolate(migratetype))) {
1527 		migratetype = get_pfnblock_migratetype(page, pfn);
1528 	}
1529 	__free_one_page(page, pfn, zone, order, migratetype, fpi_flags);
1530 	spin_unlock_irqrestore(&zone->lock, flags);
1531 }
1532 
1533 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1534 				unsigned long zone, int nid)
1535 {
1536 	mm_zero_struct_page(page);
1537 	set_page_links(page, zone, nid, pfn);
1538 	init_page_count(page);
1539 	page_mapcount_reset(page);
1540 	page_cpupid_reset_last(page);
1541 	page_kasan_tag_reset(page);
1542 
1543 	INIT_LIST_HEAD(&page->lru);
1544 #ifdef WANT_PAGE_VIRTUAL
1545 	/* The shift won't overflow because ZONE_NORMAL is below 4G. */
1546 	if (!is_highmem_idx(zone))
1547 		set_page_address(page, __va(pfn << PAGE_SHIFT));
1548 #endif
1549 }
1550 
1551 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1552 static void __meminit init_reserved_page(unsigned long pfn)
1553 {
1554 	pg_data_t *pgdat;
1555 	int nid, zid;
1556 
1557 	if (!early_page_uninitialised(pfn))
1558 		return;
1559 
1560 	nid = early_pfn_to_nid(pfn);
1561 	pgdat = NODE_DATA(nid);
1562 
1563 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1564 		struct zone *zone = &pgdat->node_zones[zid];
1565 
1566 		if (zone_spans_pfn(zone, pfn))
1567 			break;
1568 	}
1569 	__init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1570 }
1571 #else
1572 static inline void init_reserved_page(unsigned long pfn)
1573 {
1574 }
1575 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1576 
1577 /*
1578  * Initialised pages do not have PageReserved set. This function is
1579  * called for each range allocated by the bootmem allocator and
1580  * marks the pages PageReserved. The remaining valid pages are later
1581  * sent to the buddy page allocator.
1582  */
1583 void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1584 {
1585 	unsigned long start_pfn = PFN_DOWN(start);
1586 	unsigned long end_pfn = PFN_UP(end);
1587 
1588 	for (; start_pfn < end_pfn; start_pfn++) {
1589 		if (pfn_valid(start_pfn)) {
1590 			struct page *page = pfn_to_page(start_pfn);
1591 
1592 			init_reserved_page(start_pfn);
1593 
1594 			/* Avoid false-positive PageTail() */
1595 			INIT_LIST_HEAD(&page->lru);
1596 
1597 			/*
1598 			 * no need for atomic set_bit because the struct
1599 			 * page is not visible yet so nobody should
1600 			 * access it yet.
1601 			 */
1602 			__SetPageReserved(page);
1603 		}
1604 	}
1605 }
1606 
1607 static void __free_pages_ok(struct page *page, unsigned int order,
1608 			    fpi_t fpi_flags)
1609 {
1610 	unsigned long flags;
1611 	int migratetype;
1612 	unsigned long pfn = page_to_pfn(page);
1613 	struct zone *zone = page_zone(page);
1614 
1615 	if (!free_pages_prepare(page, order, true, fpi_flags))
1616 		return;
1617 
1618 	migratetype = get_pfnblock_migratetype(page, pfn);
1619 
1620 	spin_lock_irqsave(&zone->lock, flags);
1621 	if (unlikely(has_isolate_pageblock(zone) ||
1622 		is_migrate_isolate(migratetype))) {
1623 		migratetype = get_pfnblock_migratetype(page, pfn);
1624 	}
1625 	__free_one_page(page, pfn, zone, order, migratetype, fpi_flags);
1626 	spin_unlock_irqrestore(&zone->lock, flags);
1627 
1628 	__count_vm_events(PGFREE, 1 << order);
1629 }
1630 
1631 void __free_pages_core(struct page *page, unsigned int order)
1632 {
1633 	unsigned int nr_pages = 1 << order;
1634 	struct page *p = page;
1635 	unsigned int loop;
1636 
1637 	/*
1638 	 * When initializing the memmap, __init_single_page() sets the refcount
1639 	 * of all pages to 1 ("allocated"/"not free"). We have to set the
1640 	 * refcount of all involved pages to 0.
1641 	 */
1642 	prefetchw(p);
1643 	for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1644 		prefetchw(p + 1);
1645 		__ClearPageReserved(p);
1646 		set_page_count(p, 0);
1647 	}
1648 	__ClearPageReserved(p);
1649 	set_page_count(p, 0);
1650 
1651 	atomic_long_add(nr_pages, &page_zone(page)->managed_pages);
1652 
1653 	/*
1654 	 * Bypass PCP and place fresh pages right to the tail, primarily
1655 	 * relevant for memory onlining.
1656 	 */
1657 	__free_pages_ok(page, order, FPI_TO_TAIL | FPI_SKIP_KASAN_POISON);
1658 }
1659 
1660 #ifdef CONFIG_NUMA
1661 
1662 /*
1663  * During memory init memblocks map pfns to nids. The search is expensive and
1664  * this caches recent lookups. The implementation of __early_pfn_to_nid
1665  * treats start/end as pfns.
1666  */
1667 struct mminit_pfnnid_cache {
1668 	unsigned long last_start;
1669 	unsigned long last_end;
1670 	int last_nid;
1671 };
1672 
1673 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1674 
1675 /*
1676  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
1677  */
1678 static int __meminit __early_pfn_to_nid(unsigned long pfn,
1679 					struct mminit_pfnnid_cache *state)
1680 {
1681 	unsigned long start_pfn, end_pfn;
1682 	int nid;
1683 
1684 	if (state->last_start <= pfn && pfn < state->last_end)
1685 		return state->last_nid;
1686 
1687 	nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
1688 	if (nid != NUMA_NO_NODE) {
1689 		state->last_start = start_pfn;
1690 		state->last_end = end_pfn;
1691 		state->last_nid = nid;
1692 	}
1693 
1694 	return nid;
1695 }
1696 
1697 int __meminit early_pfn_to_nid(unsigned long pfn)
1698 {
1699 	static DEFINE_SPINLOCK(early_pfn_lock);
1700 	int nid;
1701 
1702 	spin_lock(&early_pfn_lock);
1703 	nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1704 	if (nid < 0)
1705 		nid = first_online_node;
1706 	spin_unlock(&early_pfn_lock);
1707 
1708 	return nid;
1709 }
1710 #endif /* CONFIG_NUMA */
1711 
1712 void __init memblock_free_pages(struct page *page, unsigned long pfn,
1713 							unsigned int order)
1714 {
1715 	if (early_page_uninitialised(pfn))
1716 		return;
1717 	__free_pages_core(page, order);
1718 }
1719 
1720 /*
1721  * Check that the whole (or subset of) a pageblock given by the interval of
1722  * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1723  * with the migration of free compaction scanner.
1724  *
1725  * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1726  *
1727  * It's possible on some configurations to have a setup like node0 node1 node0
1728  * i.e. it's possible that all pages within a zones range of pages do not
1729  * belong to a single zone. We assume that a border between node0 and node1
1730  * can occur within a single pageblock, but not a node0 node1 node0
1731  * interleaving within a single pageblock. It is therefore sufficient to check
1732  * the first and last page of a pageblock and avoid checking each individual
1733  * page in a pageblock.
1734  */
1735 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1736 				     unsigned long end_pfn, struct zone *zone)
1737 {
1738 	struct page *start_page;
1739 	struct page *end_page;
1740 
1741 	/* end_pfn is one past the range we are checking */
1742 	end_pfn--;
1743 
1744 	if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1745 		return NULL;
1746 
1747 	start_page = pfn_to_online_page(start_pfn);
1748 	if (!start_page)
1749 		return NULL;
1750 
1751 	if (page_zone(start_page) != zone)
1752 		return NULL;
1753 
1754 	end_page = pfn_to_page(end_pfn);
1755 
1756 	/* This gives a shorter code than deriving page_zone(end_page) */
1757 	if (page_zone_id(start_page) != page_zone_id(end_page))
1758 		return NULL;
1759 
1760 	return start_page;
1761 }
1762 
1763 void set_zone_contiguous(struct zone *zone)
1764 {
1765 	unsigned long block_start_pfn = zone->zone_start_pfn;
1766 	unsigned long block_end_pfn;
1767 
1768 	block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1769 	for (; block_start_pfn < zone_end_pfn(zone);
1770 			block_start_pfn = block_end_pfn,
1771 			 block_end_pfn += pageblock_nr_pages) {
1772 
1773 		block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1774 
1775 		if (!__pageblock_pfn_to_page(block_start_pfn,
1776 					     block_end_pfn, zone))
1777 			return;
1778 		cond_resched();
1779 	}
1780 
1781 	/* We confirm that there is no hole */
1782 	zone->contiguous = true;
1783 }
1784 
1785 void clear_zone_contiguous(struct zone *zone)
1786 {
1787 	zone->contiguous = false;
1788 }
1789 
1790 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1791 static void __init deferred_free_range(unsigned long pfn,
1792 				       unsigned long nr_pages)
1793 {
1794 	struct page *page;
1795 	unsigned long i;
1796 
1797 	if (!nr_pages)
1798 		return;
1799 
1800 	page = pfn_to_page(pfn);
1801 
1802 	/* Free a large naturally-aligned chunk if possible */
1803 	if (nr_pages == pageblock_nr_pages &&
1804 	    (pfn & (pageblock_nr_pages - 1)) == 0) {
1805 		set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1806 		__free_pages_core(page, pageblock_order);
1807 		return;
1808 	}
1809 
1810 	for (i = 0; i < nr_pages; i++, page++, pfn++) {
1811 		if ((pfn & (pageblock_nr_pages - 1)) == 0)
1812 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1813 		__free_pages_core(page, 0);
1814 	}
1815 }
1816 
1817 /* Completion tracking for deferred_init_memmap() threads */
1818 static atomic_t pgdat_init_n_undone __initdata;
1819 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1820 
1821 static inline void __init pgdat_init_report_one_done(void)
1822 {
1823 	if (atomic_dec_and_test(&pgdat_init_n_undone))
1824 		complete(&pgdat_init_all_done_comp);
1825 }
1826 
1827 /*
1828  * Returns true if page needs to be initialized or freed to buddy allocator.
1829  *
1830  * First we check if pfn is valid on architectures where it is possible to have
1831  * holes within pageblock_nr_pages. On systems where it is not possible, this
1832  * function is optimized out.
1833  *
1834  * Then, we check if a current large page is valid by only checking the validity
1835  * of the head pfn.
1836  */
1837 static inline bool __init deferred_pfn_valid(unsigned long pfn)
1838 {
1839 	if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1840 		return false;
1841 	return true;
1842 }
1843 
1844 /*
1845  * Free pages to buddy allocator. Try to free aligned pages in
1846  * pageblock_nr_pages sizes.
1847  */
1848 static void __init deferred_free_pages(unsigned long pfn,
1849 				       unsigned long end_pfn)
1850 {
1851 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1852 	unsigned long nr_free = 0;
1853 
1854 	for (; pfn < end_pfn; pfn++) {
1855 		if (!deferred_pfn_valid(pfn)) {
1856 			deferred_free_range(pfn - nr_free, nr_free);
1857 			nr_free = 0;
1858 		} else if (!(pfn & nr_pgmask)) {
1859 			deferred_free_range(pfn - nr_free, nr_free);
1860 			nr_free = 1;
1861 		} else {
1862 			nr_free++;
1863 		}
1864 	}
1865 	/* Free the last block of pages to allocator */
1866 	deferred_free_range(pfn - nr_free, nr_free);
1867 }
1868 
1869 /*
1870  * Initialize struct pages.  We minimize pfn page lookups and scheduler checks
1871  * by performing it only once every pageblock_nr_pages.
1872  * Return number of pages initialized.
1873  */
1874 static unsigned long  __init deferred_init_pages(struct zone *zone,
1875 						 unsigned long pfn,
1876 						 unsigned long end_pfn)
1877 {
1878 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1879 	int nid = zone_to_nid(zone);
1880 	unsigned long nr_pages = 0;
1881 	int zid = zone_idx(zone);
1882 	struct page *page = NULL;
1883 
1884 	for (; pfn < end_pfn; pfn++) {
1885 		if (!deferred_pfn_valid(pfn)) {
1886 			page = NULL;
1887 			continue;
1888 		} else if (!page || !(pfn & nr_pgmask)) {
1889 			page = pfn_to_page(pfn);
1890 		} else {
1891 			page++;
1892 		}
1893 		__init_single_page(page, pfn, zid, nid);
1894 		nr_pages++;
1895 	}
1896 	return (nr_pages);
1897 }
1898 
1899 /*
1900  * This function is meant to pre-load the iterator for the zone init.
1901  * Specifically it walks through the ranges until we are caught up to the
1902  * first_init_pfn value and exits there. If we never encounter the value we
1903  * return false indicating there are no valid ranges left.
1904  */
1905 static bool __init
1906 deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone,
1907 				    unsigned long *spfn, unsigned long *epfn,
1908 				    unsigned long first_init_pfn)
1909 {
1910 	u64 j;
1911 
1912 	/*
1913 	 * Start out by walking through the ranges in this zone that have
1914 	 * already been initialized. We don't need to do anything with them
1915 	 * so we just need to flush them out of the system.
1916 	 */
1917 	for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) {
1918 		if (*epfn <= first_init_pfn)
1919 			continue;
1920 		if (*spfn < first_init_pfn)
1921 			*spfn = first_init_pfn;
1922 		*i = j;
1923 		return true;
1924 	}
1925 
1926 	return false;
1927 }
1928 
1929 /*
1930  * Initialize and free pages. We do it in two loops: first we initialize
1931  * struct page, then free to buddy allocator, because while we are
1932  * freeing pages we can access pages that are ahead (computing buddy
1933  * page in __free_one_page()).
1934  *
1935  * In order to try and keep some memory in the cache we have the loop
1936  * broken along max page order boundaries. This way we will not cause
1937  * any issues with the buddy page computation.
1938  */
1939 static unsigned long __init
1940 deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
1941 		       unsigned long *end_pfn)
1942 {
1943 	unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
1944 	unsigned long spfn = *start_pfn, epfn = *end_pfn;
1945 	unsigned long nr_pages = 0;
1946 	u64 j = *i;
1947 
1948 	/* First we loop through and initialize the page values */
1949 	for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) {
1950 		unsigned long t;
1951 
1952 		if (mo_pfn <= *start_pfn)
1953 			break;
1954 
1955 		t = min(mo_pfn, *end_pfn);
1956 		nr_pages += deferred_init_pages(zone, *start_pfn, t);
1957 
1958 		if (mo_pfn < *end_pfn) {
1959 			*start_pfn = mo_pfn;
1960 			break;
1961 		}
1962 	}
1963 
1964 	/* Reset values and now loop through freeing pages as needed */
1965 	swap(j, *i);
1966 
1967 	for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) {
1968 		unsigned long t;
1969 
1970 		if (mo_pfn <= spfn)
1971 			break;
1972 
1973 		t = min(mo_pfn, epfn);
1974 		deferred_free_pages(spfn, t);
1975 
1976 		if (mo_pfn <= epfn)
1977 			break;
1978 	}
1979 
1980 	return nr_pages;
1981 }
1982 
1983 static void __init
1984 deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
1985 			   void *arg)
1986 {
1987 	unsigned long spfn, epfn;
1988 	struct zone *zone = arg;
1989 	u64 i;
1990 
1991 	deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn);
1992 
1993 	/*
1994 	 * Initialize and free pages in MAX_ORDER sized increments so that we
1995 	 * can avoid introducing any issues with the buddy allocator.
1996 	 */
1997 	while (spfn < end_pfn) {
1998 		deferred_init_maxorder(&i, zone, &spfn, &epfn);
1999 		cond_resched();
2000 	}
2001 }
2002 
2003 /* An arch may override for more concurrency. */
2004 __weak int __init
2005 deferred_page_init_max_threads(const struct cpumask *node_cpumask)
2006 {
2007 	return 1;
2008 }
2009 
2010 /* Initialise remaining memory on a node */
2011 static int __init deferred_init_memmap(void *data)
2012 {
2013 	pg_data_t *pgdat = data;
2014 	const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
2015 	unsigned long spfn = 0, epfn = 0;
2016 	unsigned long first_init_pfn, flags;
2017 	unsigned long start = jiffies;
2018 	struct zone *zone;
2019 	int zid, max_threads;
2020 	u64 i;
2021 
2022 	/* Bind memory initialisation thread to a local node if possible */
2023 	if (!cpumask_empty(cpumask))
2024 		set_cpus_allowed_ptr(current, cpumask);
2025 
2026 	pgdat_resize_lock(pgdat, &flags);
2027 	first_init_pfn = pgdat->first_deferred_pfn;
2028 	if (first_init_pfn == ULONG_MAX) {
2029 		pgdat_resize_unlock(pgdat, &flags);
2030 		pgdat_init_report_one_done();
2031 		return 0;
2032 	}
2033 
2034 	/* Sanity check boundaries */
2035 	BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
2036 	BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
2037 	pgdat->first_deferred_pfn = ULONG_MAX;
2038 
2039 	/*
2040 	 * Once we unlock here, the zone cannot be grown anymore, thus if an
2041 	 * interrupt thread must allocate this early in boot, zone must be
2042 	 * pre-grown prior to start of deferred page initialization.
2043 	 */
2044 	pgdat_resize_unlock(pgdat, &flags);
2045 
2046 	/* Only the highest zone is deferred so find it */
2047 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2048 		zone = pgdat->node_zones + zid;
2049 		if (first_init_pfn < zone_end_pfn(zone))
2050 			break;
2051 	}
2052 
2053 	/* If the zone is empty somebody else may have cleared out the zone */
2054 	if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2055 						 first_init_pfn))
2056 		goto zone_empty;
2057 
2058 	max_threads = deferred_page_init_max_threads(cpumask);
2059 
2060 	while (spfn < epfn) {
2061 		unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION);
2062 		struct padata_mt_job job = {
2063 			.thread_fn   = deferred_init_memmap_chunk,
2064 			.fn_arg      = zone,
2065 			.start       = spfn,
2066 			.size        = epfn_align - spfn,
2067 			.align       = PAGES_PER_SECTION,
2068 			.min_chunk   = PAGES_PER_SECTION,
2069 			.max_threads = max_threads,
2070 		};
2071 
2072 		padata_do_multithreaded(&job);
2073 		deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2074 						    epfn_align);
2075 	}
2076 zone_empty:
2077 	/* Sanity check that the next zone really is unpopulated */
2078 	WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
2079 
2080 	pr_info("node %d deferred pages initialised in %ums\n",
2081 		pgdat->node_id, jiffies_to_msecs(jiffies - start));
2082 
2083 	pgdat_init_report_one_done();
2084 	return 0;
2085 }
2086 
2087 /*
2088  * If this zone has deferred pages, try to grow it by initializing enough
2089  * deferred pages to satisfy the allocation specified by order, rounded up to
2090  * the nearest PAGES_PER_SECTION boundary.  So we're adding memory in increments
2091  * of SECTION_SIZE bytes by initializing struct pages in increments of
2092  * PAGES_PER_SECTION * sizeof(struct page) bytes.
2093  *
2094  * Return true when zone was grown, otherwise return false. We return true even
2095  * when we grow less than requested, to let the caller decide if there are
2096  * enough pages to satisfy the allocation.
2097  *
2098  * Note: We use noinline because this function is needed only during boot, and
2099  * it is called from a __ref function _deferred_grow_zone. This way we are
2100  * making sure that it is not inlined into permanent text section.
2101  */
2102 static noinline bool __init
2103 deferred_grow_zone(struct zone *zone, unsigned int order)
2104 {
2105 	unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
2106 	pg_data_t *pgdat = zone->zone_pgdat;
2107 	unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
2108 	unsigned long spfn, epfn, flags;
2109 	unsigned long nr_pages = 0;
2110 	u64 i;
2111 
2112 	/* Only the last zone may have deferred pages */
2113 	if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
2114 		return false;
2115 
2116 	pgdat_resize_lock(pgdat, &flags);
2117 
2118 	/*
2119 	 * If someone grew this zone while we were waiting for spinlock, return
2120 	 * true, as there might be enough pages already.
2121 	 */
2122 	if (first_deferred_pfn != pgdat->first_deferred_pfn) {
2123 		pgdat_resize_unlock(pgdat, &flags);
2124 		return true;
2125 	}
2126 
2127 	/* If the zone is empty somebody else may have cleared out the zone */
2128 	if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2129 						 first_deferred_pfn)) {
2130 		pgdat->first_deferred_pfn = ULONG_MAX;
2131 		pgdat_resize_unlock(pgdat, &flags);
2132 		/* Retry only once. */
2133 		return first_deferred_pfn != ULONG_MAX;
2134 	}
2135 
2136 	/*
2137 	 * Initialize and free pages in MAX_ORDER sized increments so
2138 	 * that we can avoid introducing any issues with the buddy
2139 	 * allocator.
2140 	 */
2141 	while (spfn < epfn) {
2142 		/* update our first deferred PFN for this section */
2143 		first_deferred_pfn = spfn;
2144 
2145 		nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
2146 		touch_nmi_watchdog();
2147 
2148 		/* We should only stop along section boundaries */
2149 		if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
2150 			continue;
2151 
2152 		/* If our quota has been met we can stop here */
2153 		if (nr_pages >= nr_pages_needed)
2154 			break;
2155 	}
2156 
2157 	pgdat->first_deferred_pfn = spfn;
2158 	pgdat_resize_unlock(pgdat, &flags);
2159 
2160 	return nr_pages > 0;
2161 }
2162 
2163 /*
2164  * deferred_grow_zone() is __init, but it is called from
2165  * get_page_from_freelist() during early boot until deferred_pages permanently
2166  * disables this call. This is why we have refdata wrapper to avoid warning,
2167  * and to ensure that the function body gets unloaded.
2168  */
2169 static bool __ref
2170 _deferred_grow_zone(struct zone *zone, unsigned int order)
2171 {
2172 	return deferred_grow_zone(zone, order);
2173 }
2174 
2175 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
2176 
2177 void __init page_alloc_init_late(void)
2178 {
2179 	struct zone *zone;
2180 	int nid;
2181 
2182 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
2183 
2184 	/* There will be num_node_state(N_MEMORY) threads */
2185 	atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
2186 	for_each_node_state(nid, N_MEMORY) {
2187 		kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
2188 	}
2189 
2190 	/* Block until all are initialised */
2191 	wait_for_completion(&pgdat_init_all_done_comp);
2192 
2193 	/*
2194 	 * We initialized the rest of the deferred pages.  Permanently disable
2195 	 * on-demand struct page initialization.
2196 	 */
2197 	static_branch_disable(&deferred_pages);
2198 
2199 	/* Reinit limits that are based on free pages after the kernel is up */
2200 	files_maxfiles_init();
2201 #endif
2202 
2203 	buffer_init();
2204 
2205 	/* Discard memblock private memory */
2206 	memblock_discard();
2207 
2208 	for_each_node_state(nid, N_MEMORY)
2209 		shuffle_free_memory(NODE_DATA(nid));
2210 
2211 	for_each_populated_zone(zone)
2212 		set_zone_contiguous(zone);
2213 }
2214 
2215 #ifdef CONFIG_CMA
2216 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
2217 void __init init_cma_reserved_pageblock(struct page *page)
2218 {
2219 	unsigned i = pageblock_nr_pages;
2220 	struct page *p = page;
2221 
2222 	do {
2223 		__ClearPageReserved(p);
2224 		set_page_count(p, 0);
2225 	} while (++p, --i);
2226 
2227 	set_pageblock_migratetype(page, MIGRATE_CMA);
2228 	set_page_refcounted(page);
2229 	__free_pages(page, pageblock_order);
2230 
2231 	adjust_managed_page_count(page, pageblock_nr_pages);
2232 	page_zone(page)->cma_pages += pageblock_nr_pages;
2233 }
2234 #endif
2235 
2236 /*
2237  * The order of subdivision here is critical for the IO subsystem.
2238  * Please do not alter this order without good reasons and regression
2239  * testing. Specifically, as large blocks of memory are subdivided,
2240  * the order in which smaller blocks are delivered depends on the order
2241  * they're subdivided in this function. This is the primary factor
2242  * influencing the order in which pages are delivered to the IO
2243  * subsystem according to empirical testing, and this is also justified
2244  * by considering the behavior of a buddy system containing a single
2245  * large block of memory acted on by a series of small allocations.
2246  * This behavior is a critical factor in sglist merging's success.
2247  *
2248  * -- nyc
2249  */
2250 static inline void expand(struct zone *zone, struct page *page,
2251 	int low, int high, int migratetype)
2252 {
2253 	unsigned long size = 1 << high;
2254 
2255 	while (high > low) {
2256 		high--;
2257 		size >>= 1;
2258 		VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
2259 
2260 		/*
2261 		 * Mark as guard pages (or page), that will allow to
2262 		 * merge back to allocator when buddy will be freed.
2263 		 * Corresponding page table entries will not be touched,
2264 		 * pages will stay not present in virtual address space
2265 		 */
2266 		if (set_page_guard(zone, &page[size], high, migratetype))
2267 			continue;
2268 
2269 		add_to_free_list(&page[size], zone, high, migratetype);
2270 		set_buddy_order(&page[size], high);
2271 	}
2272 }
2273 
2274 static void check_new_page_bad(struct page *page)
2275 {
2276 	if (unlikely(page->flags & __PG_HWPOISON)) {
2277 		/* Don't complain about hwpoisoned pages */
2278 		page_mapcount_reset(page); /* remove PageBuddy */
2279 		return;
2280 	}
2281 
2282 	bad_page(page,
2283 		 page_bad_reason(page, PAGE_FLAGS_CHECK_AT_PREP));
2284 }
2285 
2286 /*
2287  * This page is about to be returned from the page allocator
2288  */
2289 static inline int check_new_page(struct page *page)
2290 {
2291 	if (likely(page_expected_state(page,
2292 				PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
2293 		return 0;
2294 
2295 	check_new_page_bad(page);
2296 	return 1;
2297 }
2298 
2299 static bool check_new_pages(struct page *page, unsigned int order)
2300 {
2301 	int i;
2302 	for (i = 0; i < (1 << order); i++) {
2303 		struct page *p = page + i;
2304 
2305 		if (unlikely(check_new_page(p)))
2306 			return true;
2307 	}
2308 
2309 	return false;
2310 }
2311 
2312 #ifdef CONFIG_DEBUG_VM
2313 /*
2314  * With DEBUG_VM enabled, order-0 pages are checked for expected state when
2315  * being allocated from pcp lists. With debug_pagealloc also enabled, they are
2316  * also checked when pcp lists are refilled from the free lists.
2317  */
2318 static inline bool check_pcp_refill(struct page *page, unsigned int order)
2319 {
2320 	if (debug_pagealloc_enabled_static())
2321 		return check_new_pages(page, order);
2322 	else
2323 		return false;
2324 }
2325 
2326 static inline bool check_new_pcp(struct page *page, unsigned int order)
2327 {
2328 	return check_new_pages(page, order);
2329 }
2330 #else
2331 /*
2332  * With DEBUG_VM disabled, free order-0 pages are checked for expected state
2333  * when pcp lists are being refilled from the free lists. With debug_pagealloc
2334  * enabled, they are also checked when being allocated from the pcp lists.
2335  */
2336 static inline bool check_pcp_refill(struct page *page, unsigned int order)
2337 {
2338 	return check_new_pages(page, order);
2339 }
2340 static inline bool check_new_pcp(struct page *page, unsigned int order)
2341 {
2342 	if (debug_pagealloc_enabled_static())
2343 		return check_new_pages(page, order);
2344 	else
2345 		return false;
2346 }
2347 #endif /* CONFIG_DEBUG_VM */
2348 
2349 static inline bool should_skip_kasan_unpoison(gfp_t flags, bool init_tags)
2350 {
2351 	/* Don't skip if a software KASAN mode is enabled. */
2352 	if (IS_ENABLED(CONFIG_KASAN_GENERIC) ||
2353 	    IS_ENABLED(CONFIG_KASAN_SW_TAGS))
2354 		return false;
2355 
2356 	/* Skip, if hardware tag-based KASAN is not enabled. */
2357 	if (!kasan_hw_tags_enabled())
2358 		return true;
2359 
2360 	/*
2361 	 * With hardware tag-based KASAN enabled, skip if either:
2362 	 *
2363 	 * 1. Memory tags have already been cleared via tag_clear_highpage().
2364 	 * 2. Skipping has been requested via __GFP_SKIP_KASAN_UNPOISON.
2365 	 */
2366 	return init_tags || (flags & __GFP_SKIP_KASAN_UNPOISON);
2367 }
2368 
2369 static inline bool should_skip_init(gfp_t flags)
2370 {
2371 	/* Don't skip, if hardware tag-based KASAN is not enabled. */
2372 	if (!kasan_hw_tags_enabled())
2373 		return false;
2374 
2375 	/* For hardware tag-based KASAN, skip if requested. */
2376 	return (flags & __GFP_SKIP_ZERO);
2377 }
2378 
2379 inline void post_alloc_hook(struct page *page, unsigned int order,
2380 				gfp_t gfp_flags)
2381 {
2382 	bool init = !want_init_on_free() && want_init_on_alloc(gfp_flags) &&
2383 			!should_skip_init(gfp_flags);
2384 	bool init_tags = init && (gfp_flags & __GFP_ZEROTAGS);
2385 
2386 	set_page_private(page, 0);
2387 	set_page_refcounted(page);
2388 
2389 	arch_alloc_page(page, order);
2390 	debug_pagealloc_map_pages(page, 1 << order);
2391 
2392 	/*
2393 	 * Page unpoisoning must happen before memory initialization.
2394 	 * Otherwise, the poison pattern will be overwritten for __GFP_ZERO
2395 	 * allocations and the page unpoisoning code will complain.
2396 	 */
2397 	kernel_unpoison_pages(page, 1 << order);
2398 
2399 	/*
2400 	 * As memory initialization might be integrated into KASAN,
2401 	 * KASAN unpoisoning and memory initializion code must be
2402 	 * kept together to avoid discrepancies in behavior.
2403 	 */
2404 
2405 	/*
2406 	 * If memory tags should be zeroed (which happens only when memory
2407 	 * should be initialized as well).
2408 	 */
2409 	if (init_tags) {
2410 		int i;
2411 
2412 		/* Initialize both memory and tags. */
2413 		for (i = 0; i != 1 << order; ++i)
2414 			tag_clear_highpage(page + i);
2415 
2416 		/* Note that memory is already initialized by the loop above. */
2417 		init = false;
2418 	}
2419 	if (!should_skip_kasan_unpoison(gfp_flags, init_tags)) {
2420 		/* Unpoison shadow memory or set memory tags. */
2421 		kasan_unpoison_pages(page, order, init);
2422 
2423 		/* Note that memory is already initialized by KASAN. */
2424 		if (kasan_has_integrated_init())
2425 			init = false;
2426 	}
2427 	/* If memory is still not initialized, do it now. */
2428 	if (init)
2429 		kernel_init_free_pages(page, 1 << order);
2430 	/* Propagate __GFP_SKIP_KASAN_POISON to page flags. */
2431 	if (kasan_hw_tags_enabled() && (gfp_flags & __GFP_SKIP_KASAN_POISON))
2432 		SetPageSkipKASanPoison(page);
2433 
2434 	set_page_owner(page, order, gfp_flags);
2435 	page_table_check_alloc(page, order);
2436 }
2437 
2438 static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
2439 							unsigned int alloc_flags)
2440 {
2441 	post_alloc_hook(page, order, gfp_flags);
2442 
2443 	if (order && (gfp_flags & __GFP_COMP))
2444 		prep_compound_page(page, order);
2445 
2446 	/*
2447 	 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
2448 	 * allocate the page. The expectation is that the caller is taking
2449 	 * steps that will free more memory. The caller should avoid the page
2450 	 * being used for !PFMEMALLOC purposes.
2451 	 */
2452 	if (alloc_flags & ALLOC_NO_WATERMARKS)
2453 		set_page_pfmemalloc(page);
2454 	else
2455 		clear_page_pfmemalloc(page);
2456 }
2457 
2458 /*
2459  * Go through the free lists for the given migratetype and remove
2460  * the smallest available page from the freelists
2461  */
2462 static __always_inline
2463 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
2464 						int migratetype)
2465 {
2466 	unsigned int current_order;
2467 	struct free_area *area;
2468 	struct page *page;
2469 
2470 	/* Find a page of the appropriate size in the preferred list */
2471 	for (current_order = order; current_order < MAX_ORDER; ++current_order) {
2472 		area = &(zone->free_area[current_order]);
2473 		page = get_page_from_free_area(area, migratetype);
2474 		if (!page)
2475 			continue;
2476 		del_page_from_free_list(page, zone, current_order);
2477 		expand(zone, page, order, current_order, migratetype);
2478 		set_pcppage_migratetype(page, migratetype);
2479 		return page;
2480 	}
2481 
2482 	return NULL;
2483 }
2484 
2485 
2486 /*
2487  * This array describes the order lists are fallen back to when
2488  * the free lists for the desirable migrate type are depleted
2489  *
2490  * The other migratetypes do not have fallbacks.
2491  */
2492 static int fallbacks[MIGRATE_TYPES][3] = {
2493 	[MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_TYPES },
2494 	[MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
2495 	[MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_TYPES },
2496 };
2497 
2498 #ifdef CONFIG_CMA
2499 static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2500 					unsigned int order)
2501 {
2502 	return __rmqueue_smallest(zone, order, MIGRATE_CMA);
2503 }
2504 #else
2505 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2506 					unsigned int order) { return NULL; }
2507 #endif
2508 
2509 /*
2510  * Move the free pages in a range to the freelist tail of the requested type.
2511  * Note that start_page and end_pages are not aligned on a pageblock
2512  * boundary. If alignment is required, use move_freepages_block()
2513  */
2514 static int move_freepages(struct zone *zone,
2515 			  unsigned long start_pfn, unsigned long end_pfn,
2516 			  int migratetype, int *num_movable)
2517 {
2518 	struct page *page;
2519 	unsigned long pfn;
2520 	unsigned int order;
2521 	int pages_moved = 0;
2522 
2523 	for (pfn = start_pfn; pfn <= end_pfn;) {
2524 		page = pfn_to_page(pfn);
2525 		if (!PageBuddy(page)) {
2526 			/*
2527 			 * We assume that pages that could be isolated for
2528 			 * migration are movable. But we don't actually try
2529 			 * isolating, as that would be expensive.
2530 			 */
2531 			if (num_movable &&
2532 					(PageLRU(page) || __PageMovable(page)))
2533 				(*num_movable)++;
2534 			pfn++;
2535 			continue;
2536 		}
2537 
2538 		/* Make sure we are not inadvertently changing nodes */
2539 		VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2540 		VM_BUG_ON_PAGE(page_zone(page) != zone, page);
2541 
2542 		order = buddy_order(page);
2543 		move_to_free_list(page, zone, order, migratetype);
2544 		pfn += 1 << order;
2545 		pages_moved += 1 << order;
2546 	}
2547 
2548 	return pages_moved;
2549 }
2550 
2551 int move_freepages_block(struct zone *zone, struct page *page,
2552 				int migratetype, int *num_movable)
2553 {
2554 	unsigned long start_pfn, end_pfn, pfn;
2555 
2556 	if (num_movable)
2557 		*num_movable = 0;
2558 
2559 	pfn = page_to_pfn(page);
2560 	start_pfn = pfn & ~(pageblock_nr_pages - 1);
2561 	end_pfn = start_pfn + pageblock_nr_pages - 1;
2562 
2563 	/* Do not cross zone boundaries */
2564 	if (!zone_spans_pfn(zone, start_pfn))
2565 		start_pfn = pfn;
2566 	if (!zone_spans_pfn(zone, end_pfn))
2567 		return 0;
2568 
2569 	return move_freepages(zone, start_pfn, end_pfn, migratetype,
2570 								num_movable);
2571 }
2572 
2573 static void change_pageblock_range(struct page *pageblock_page,
2574 					int start_order, int migratetype)
2575 {
2576 	int nr_pageblocks = 1 << (start_order - pageblock_order);
2577 
2578 	while (nr_pageblocks--) {
2579 		set_pageblock_migratetype(pageblock_page, migratetype);
2580 		pageblock_page += pageblock_nr_pages;
2581 	}
2582 }
2583 
2584 /*
2585  * When we are falling back to another migratetype during allocation, try to
2586  * steal extra free pages from the same pageblocks to satisfy further
2587  * allocations, instead of polluting multiple pageblocks.
2588  *
2589  * If we are stealing a relatively large buddy page, it is likely there will
2590  * be more free pages in the pageblock, so try to steal them all. For
2591  * reclaimable and unmovable allocations, we steal regardless of page size,
2592  * as fragmentation caused by those allocations polluting movable pageblocks
2593  * is worse than movable allocations stealing from unmovable and reclaimable
2594  * pageblocks.
2595  */
2596 static bool can_steal_fallback(unsigned int order, int start_mt)
2597 {
2598 	/*
2599 	 * Leaving this order check is intended, although there is
2600 	 * relaxed order check in next check. The reason is that
2601 	 * we can actually steal whole pageblock if this condition met,
2602 	 * but, below check doesn't guarantee it and that is just heuristic
2603 	 * so could be changed anytime.
2604 	 */
2605 	if (order >= pageblock_order)
2606 		return true;
2607 
2608 	if (order >= pageblock_order / 2 ||
2609 		start_mt == MIGRATE_RECLAIMABLE ||
2610 		start_mt == MIGRATE_UNMOVABLE ||
2611 		page_group_by_mobility_disabled)
2612 		return true;
2613 
2614 	return false;
2615 }
2616 
2617 static inline bool boost_watermark(struct zone *zone)
2618 {
2619 	unsigned long max_boost;
2620 
2621 	if (!watermark_boost_factor)
2622 		return false;
2623 	/*
2624 	 * Don't bother in zones that are unlikely to produce results.
2625 	 * On small machines, including kdump capture kernels running
2626 	 * in a small area, boosting the watermark can cause an out of
2627 	 * memory situation immediately.
2628 	 */
2629 	if ((pageblock_nr_pages * 4) > zone_managed_pages(zone))
2630 		return false;
2631 
2632 	max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
2633 			watermark_boost_factor, 10000);
2634 
2635 	/*
2636 	 * high watermark may be uninitialised if fragmentation occurs
2637 	 * very early in boot so do not boost. We do not fall
2638 	 * through and boost by pageblock_nr_pages as failing
2639 	 * allocations that early means that reclaim is not going
2640 	 * to help and it may even be impossible to reclaim the
2641 	 * boosted watermark resulting in a hang.
2642 	 */
2643 	if (!max_boost)
2644 		return false;
2645 
2646 	max_boost = max(pageblock_nr_pages, max_boost);
2647 
2648 	zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,
2649 		max_boost);
2650 
2651 	return true;
2652 }
2653 
2654 /*
2655  * This function implements actual steal behaviour. If order is large enough,
2656  * we can steal whole pageblock. If not, we first move freepages in this
2657  * pageblock to our migratetype and determine how many already-allocated pages
2658  * are there in the pageblock with a compatible migratetype. If at least half
2659  * of pages are free or compatible, we can change migratetype of the pageblock
2660  * itself, so pages freed in the future will be put on the correct free list.
2661  */
2662 static void steal_suitable_fallback(struct zone *zone, struct page *page,
2663 		unsigned int alloc_flags, int start_type, bool whole_block)
2664 {
2665 	unsigned int current_order = buddy_order(page);
2666 	int free_pages, movable_pages, alike_pages;
2667 	int old_block_type;
2668 
2669 	old_block_type = get_pageblock_migratetype(page);
2670 
2671 	/*
2672 	 * This can happen due to races and we want to prevent broken
2673 	 * highatomic accounting.
2674 	 */
2675 	if (is_migrate_highatomic(old_block_type))
2676 		goto single_page;
2677 
2678 	/* Take ownership for orders >= pageblock_order */
2679 	if (current_order >= pageblock_order) {
2680 		change_pageblock_range(page, current_order, start_type);
2681 		goto single_page;
2682 	}
2683 
2684 	/*
2685 	 * Boost watermarks to increase reclaim pressure to reduce the
2686 	 * likelihood of future fallbacks. Wake kswapd now as the node
2687 	 * may be balanced overall and kswapd will not wake naturally.
2688 	 */
2689 	if (boost_watermark(zone) && (alloc_flags & ALLOC_KSWAPD))
2690 		set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
2691 
2692 	/* We are not allowed to try stealing from the whole block */
2693 	if (!whole_block)
2694 		goto single_page;
2695 
2696 	free_pages = move_freepages_block(zone, page, start_type,
2697 						&movable_pages);
2698 	/*
2699 	 * Determine how many pages are compatible with our allocation.
2700 	 * For movable allocation, it's the number of movable pages which
2701 	 * we just obtained. For other types it's a bit more tricky.
2702 	 */
2703 	if (start_type == MIGRATE_MOVABLE) {
2704 		alike_pages = movable_pages;
2705 	} else {
2706 		/*
2707 		 * If we are falling back a RECLAIMABLE or UNMOVABLE allocation
2708 		 * to MOVABLE pageblock, consider all non-movable pages as
2709 		 * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or
2710 		 * vice versa, be conservative since we can't distinguish the
2711 		 * exact migratetype of non-movable pages.
2712 		 */
2713 		if (old_block_type == MIGRATE_MOVABLE)
2714 			alike_pages = pageblock_nr_pages
2715 						- (free_pages + movable_pages);
2716 		else
2717 			alike_pages = 0;
2718 	}
2719 
2720 	/* moving whole block can fail due to zone boundary conditions */
2721 	if (!free_pages)
2722 		goto single_page;
2723 
2724 	/*
2725 	 * If a sufficient number of pages in the block are either free or of
2726 	 * comparable migratability as our allocation, claim the whole block.
2727 	 */
2728 	if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2729 			page_group_by_mobility_disabled)
2730 		set_pageblock_migratetype(page, start_type);
2731 
2732 	return;
2733 
2734 single_page:
2735 	move_to_free_list(page, zone, current_order, start_type);
2736 }
2737 
2738 /*
2739  * Check whether there is a suitable fallback freepage with requested order.
2740  * If only_stealable is true, this function returns fallback_mt only if
2741  * we can steal other freepages all together. This would help to reduce
2742  * fragmentation due to mixed migratetype pages in one pageblock.
2743  */
2744 int find_suitable_fallback(struct free_area *area, unsigned int order,
2745 			int migratetype, bool only_stealable, bool *can_steal)
2746 {
2747 	int i;
2748 	int fallback_mt;
2749 
2750 	if (area->nr_free == 0)
2751 		return -1;
2752 
2753 	*can_steal = false;
2754 	for (i = 0;; i++) {
2755 		fallback_mt = fallbacks[migratetype][i];
2756 		if (fallback_mt == MIGRATE_TYPES)
2757 			break;
2758 
2759 		if (free_area_empty(area, fallback_mt))
2760 			continue;
2761 
2762 		if (can_steal_fallback(order, migratetype))
2763 			*can_steal = true;
2764 
2765 		if (!only_stealable)
2766 			return fallback_mt;
2767 
2768 		if (*can_steal)
2769 			return fallback_mt;
2770 	}
2771 
2772 	return -1;
2773 }
2774 
2775 /*
2776  * Reserve a pageblock for exclusive use of high-order atomic allocations if
2777  * there are no empty page blocks that contain a page with a suitable order
2778  */
2779 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2780 				unsigned int alloc_order)
2781 {
2782 	int mt;
2783 	unsigned long max_managed, flags;
2784 
2785 	/*
2786 	 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2787 	 * Check is race-prone but harmless.
2788 	 */
2789 	max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
2790 	if (zone->nr_reserved_highatomic >= max_managed)
2791 		return;
2792 
2793 	spin_lock_irqsave(&zone->lock, flags);
2794 
2795 	/* Recheck the nr_reserved_highatomic limit under the lock */
2796 	if (zone->nr_reserved_highatomic >= max_managed)
2797 		goto out_unlock;
2798 
2799 	/* Yoink! */
2800 	mt = get_pageblock_migratetype(page);
2801 	/* Only reserve normal pageblocks (i.e., they can merge with others) */
2802 	if (migratetype_is_mergeable(mt)) {
2803 		zone->nr_reserved_highatomic += pageblock_nr_pages;
2804 		set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2805 		move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2806 	}
2807 
2808 out_unlock:
2809 	spin_unlock_irqrestore(&zone->lock, flags);
2810 }
2811 
2812 /*
2813  * Used when an allocation is about to fail under memory pressure. This
2814  * potentially hurts the reliability of high-order allocations when under
2815  * intense memory pressure but failed atomic allocations should be easier
2816  * to recover from than an OOM.
2817  *
2818  * If @force is true, try to unreserve a pageblock even though highatomic
2819  * pageblock is exhausted.
2820  */
2821 static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2822 						bool force)
2823 {
2824 	struct zonelist *zonelist = ac->zonelist;
2825 	unsigned long flags;
2826 	struct zoneref *z;
2827 	struct zone *zone;
2828 	struct page *page;
2829 	int order;
2830 	bool ret;
2831 
2832 	for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx,
2833 								ac->nodemask) {
2834 		/*
2835 		 * Preserve at least one pageblock unless memory pressure
2836 		 * is really high.
2837 		 */
2838 		if (!force && zone->nr_reserved_highatomic <=
2839 					pageblock_nr_pages)
2840 			continue;
2841 
2842 		spin_lock_irqsave(&zone->lock, flags);
2843 		for (order = 0; order < MAX_ORDER; order++) {
2844 			struct free_area *area = &(zone->free_area[order]);
2845 
2846 			page = get_page_from_free_area(area, MIGRATE_HIGHATOMIC);
2847 			if (!page)
2848 				continue;
2849 
2850 			/*
2851 			 * In page freeing path, migratetype change is racy so
2852 			 * we can counter several free pages in a pageblock
2853 			 * in this loop although we changed the pageblock type
2854 			 * from highatomic to ac->migratetype. So we should
2855 			 * adjust the count once.
2856 			 */
2857 			if (is_migrate_highatomic_page(page)) {
2858 				/*
2859 				 * It should never happen but changes to
2860 				 * locking could inadvertently allow a per-cpu
2861 				 * drain to add pages to MIGRATE_HIGHATOMIC
2862 				 * while unreserving so be safe and watch for
2863 				 * underflows.
2864 				 */
2865 				zone->nr_reserved_highatomic -= min(
2866 						pageblock_nr_pages,
2867 						zone->nr_reserved_highatomic);
2868 			}
2869 
2870 			/*
2871 			 * Convert to ac->migratetype and avoid the normal
2872 			 * pageblock stealing heuristics. Minimally, the caller
2873 			 * is doing the work and needs the pages. More
2874 			 * importantly, if the block was always converted to
2875 			 * MIGRATE_UNMOVABLE or another type then the number
2876 			 * of pageblocks that cannot be completely freed
2877 			 * may increase.
2878 			 */
2879 			set_pageblock_migratetype(page, ac->migratetype);
2880 			ret = move_freepages_block(zone, page, ac->migratetype,
2881 									NULL);
2882 			if (ret) {
2883 				spin_unlock_irqrestore(&zone->lock, flags);
2884 				return ret;
2885 			}
2886 		}
2887 		spin_unlock_irqrestore(&zone->lock, flags);
2888 	}
2889 
2890 	return false;
2891 }
2892 
2893 /*
2894  * Try finding a free buddy page on the fallback list and put it on the free
2895  * list of requested migratetype, possibly along with other pages from the same
2896  * block, depending on fragmentation avoidance heuristics. Returns true if
2897  * fallback was found so that __rmqueue_smallest() can grab it.
2898  *
2899  * The use of signed ints for order and current_order is a deliberate
2900  * deviation from the rest of this file, to make the for loop
2901  * condition simpler.
2902  */
2903 static __always_inline bool
2904 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype,
2905 						unsigned int alloc_flags)
2906 {
2907 	struct free_area *area;
2908 	int current_order;
2909 	int min_order = order;
2910 	struct page *page;
2911 	int fallback_mt;
2912 	bool can_steal;
2913 
2914 	/*
2915 	 * Do not steal pages from freelists belonging to other pageblocks
2916 	 * i.e. orders < pageblock_order. If there are no local zones free,
2917 	 * the zonelists will be reiterated without ALLOC_NOFRAGMENT.
2918 	 */
2919 	if (alloc_flags & ALLOC_NOFRAGMENT)
2920 		min_order = pageblock_order;
2921 
2922 	/*
2923 	 * Find the largest available free page in the other list. This roughly
2924 	 * approximates finding the pageblock with the most free pages, which
2925 	 * would be too costly to do exactly.
2926 	 */
2927 	for (current_order = MAX_ORDER - 1; current_order >= min_order;
2928 				--current_order) {
2929 		area = &(zone->free_area[current_order]);
2930 		fallback_mt = find_suitable_fallback(area, current_order,
2931 				start_migratetype, false, &can_steal);
2932 		if (fallback_mt == -1)
2933 			continue;
2934 
2935 		/*
2936 		 * We cannot steal all free pages from the pageblock and the
2937 		 * requested migratetype is movable. In that case it's better to
2938 		 * steal and split the smallest available page instead of the
2939 		 * largest available page, because even if the next movable
2940 		 * allocation falls back into a different pageblock than this
2941 		 * one, it won't cause permanent fragmentation.
2942 		 */
2943 		if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2944 					&& current_order > order)
2945 			goto find_smallest;
2946 
2947 		goto do_steal;
2948 	}
2949 
2950 	return false;
2951 
2952 find_smallest:
2953 	for (current_order = order; current_order < MAX_ORDER;
2954 							current_order++) {
2955 		area = &(zone->free_area[current_order]);
2956 		fallback_mt = find_suitable_fallback(area, current_order,
2957 				start_migratetype, false, &can_steal);
2958 		if (fallback_mt != -1)
2959 			break;
2960 	}
2961 
2962 	/*
2963 	 * This should not happen - we already found a suitable fallback
2964 	 * when looking for the largest page.
2965 	 */
2966 	VM_BUG_ON(current_order == MAX_ORDER);
2967 
2968 do_steal:
2969 	page = get_page_from_free_area(area, fallback_mt);
2970 
2971 	steal_suitable_fallback(zone, page, alloc_flags, start_migratetype,
2972 								can_steal);
2973 
2974 	trace_mm_page_alloc_extfrag(page, order, current_order,
2975 		start_migratetype, fallback_mt);
2976 
2977 	return true;
2978 
2979 }
2980 
2981 /*
2982  * Do the hard work of removing an element from the buddy allocator.
2983  * Call me with the zone->lock already held.
2984  */
2985 static __always_inline struct page *
2986 __rmqueue(struct zone *zone, unsigned int order, int migratetype,
2987 						unsigned int alloc_flags)
2988 {
2989 	struct page *page;
2990 
2991 	if (IS_ENABLED(CONFIG_CMA)) {
2992 		/*
2993 		 * Balance movable allocations between regular and CMA areas by
2994 		 * allocating from CMA when over half of the zone's free memory
2995 		 * is in the CMA area.
2996 		 */
2997 		if (alloc_flags & ALLOC_CMA &&
2998 		    zone_page_state(zone, NR_FREE_CMA_PAGES) >
2999 		    zone_page_state(zone, NR_FREE_PAGES) / 2) {
3000 			page = __rmqueue_cma_fallback(zone, order);
3001 			if (page)
3002 				goto out;
3003 		}
3004 	}
3005 retry:
3006 	page = __rmqueue_smallest(zone, order, migratetype);
3007 	if (unlikely(!page)) {
3008 		if (alloc_flags & ALLOC_CMA)
3009 			page = __rmqueue_cma_fallback(zone, order);
3010 
3011 		if (!page && __rmqueue_fallback(zone, order, migratetype,
3012 								alloc_flags))
3013 			goto retry;
3014 	}
3015 out:
3016 	if (page)
3017 		trace_mm_page_alloc_zone_locked(page, order, migratetype);
3018 	return page;
3019 }
3020 
3021 /*
3022  * Obtain a specified number of elements from the buddy allocator, all under
3023  * a single hold of the lock, for efficiency.  Add them to the supplied list.
3024  * Returns the number of new pages which were placed at *list.
3025  */
3026 static int rmqueue_bulk(struct zone *zone, unsigned int order,
3027 			unsigned long count, struct list_head *list,
3028 			int migratetype, unsigned int alloc_flags)
3029 {
3030 	int i, allocated = 0;
3031 
3032 	/*
3033 	 * local_lock_irq held so equivalent to spin_lock_irqsave for
3034 	 * both PREEMPT_RT and non-PREEMPT_RT configurations.
3035 	 */
3036 	spin_lock(&zone->lock);
3037 	for (i = 0; i < count; ++i) {
3038 		struct page *page = __rmqueue(zone, order, migratetype,
3039 								alloc_flags);
3040 		if (unlikely(page == NULL))
3041 			break;
3042 
3043 		if (unlikely(check_pcp_refill(page, order)))
3044 			continue;
3045 
3046 		/*
3047 		 * Split buddy pages returned by expand() are received here in
3048 		 * physical page order. The page is added to the tail of
3049 		 * caller's list. From the callers perspective, the linked list
3050 		 * is ordered by page number under some conditions. This is
3051 		 * useful for IO devices that can forward direction from the
3052 		 * head, thus also in the physical page order. This is useful
3053 		 * for IO devices that can merge IO requests if the physical
3054 		 * pages are ordered properly.
3055 		 */
3056 		list_add_tail(&page->lru, list);
3057 		allocated++;
3058 		if (is_migrate_cma(get_pcppage_migratetype(page)))
3059 			__mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
3060 					      -(1 << order));
3061 	}
3062 
3063 	/*
3064 	 * i pages were removed from the buddy list even if some leak due
3065 	 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
3066 	 * on i. Do not confuse with 'allocated' which is the number of
3067 	 * pages added to the pcp list.
3068 	 */
3069 	__mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
3070 	spin_unlock(&zone->lock);
3071 	return allocated;
3072 }
3073 
3074 #ifdef CONFIG_NUMA
3075 /*
3076  * Called from the vmstat counter updater to drain pagesets of this
3077  * currently executing processor on remote nodes after they have
3078  * expired.
3079  *
3080  * Note that this function must be called with the thread pinned to
3081  * a single processor.
3082  */
3083 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
3084 {
3085 	unsigned long flags;
3086 	int to_drain, batch;
3087 
3088 	local_lock_irqsave(&pagesets.lock, flags);
3089 	batch = READ_ONCE(pcp->batch);
3090 	to_drain = min(pcp->count, batch);
3091 	if (to_drain > 0)
3092 		free_pcppages_bulk(zone, to_drain, pcp, 0);
3093 	local_unlock_irqrestore(&pagesets.lock, flags);
3094 }
3095 #endif
3096 
3097 /*
3098  * Drain pcplists of the indicated processor and zone.
3099  *
3100  * The processor must either be the current processor and the
3101  * thread pinned to the current processor or a processor that
3102  * is not online.
3103  */
3104 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
3105 {
3106 	unsigned long flags;
3107 	struct per_cpu_pages *pcp;
3108 
3109 	local_lock_irqsave(&pagesets.lock, flags);
3110 
3111 	pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
3112 	if (pcp->count)
3113 		free_pcppages_bulk(zone, pcp->count, pcp, 0);
3114 
3115 	local_unlock_irqrestore(&pagesets.lock, flags);
3116 }
3117 
3118 /*
3119  * Drain pcplists of all zones on the indicated processor.
3120  *
3121  * The processor must either be the current processor and the
3122  * thread pinned to the current processor or a processor that
3123  * is not online.
3124  */
3125 static void drain_pages(unsigned int cpu)
3126 {
3127 	struct zone *zone;
3128 
3129 	for_each_populated_zone(zone) {
3130 		drain_pages_zone(cpu, zone);
3131 	}
3132 }
3133 
3134 /*
3135  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
3136  *
3137  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
3138  * the single zone's pages.
3139  */
3140 void drain_local_pages(struct zone *zone)
3141 {
3142 	int cpu = smp_processor_id();
3143 
3144 	if (zone)
3145 		drain_pages_zone(cpu, zone);
3146 	else
3147 		drain_pages(cpu);
3148 }
3149 
3150 static void drain_local_pages_wq(struct work_struct *work)
3151 {
3152 	struct pcpu_drain *drain;
3153 
3154 	drain = container_of(work, struct pcpu_drain, work);
3155 
3156 	/*
3157 	 * drain_all_pages doesn't use proper cpu hotplug protection so
3158 	 * we can race with cpu offline when the WQ can move this from
3159 	 * a cpu pinned worker to an unbound one. We can operate on a different
3160 	 * cpu which is alright but we also have to make sure to not move to
3161 	 * a different one.
3162 	 */
3163 	migrate_disable();
3164 	drain_local_pages(drain->zone);
3165 	migrate_enable();
3166 }
3167 
3168 /*
3169  * The implementation of drain_all_pages(), exposing an extra parameter to
3170  * drain on all cpus.
3171  *
3172  * drain_all_pages() is optimized to only execute on cpus where pcplists are
3173  * not empty. The check for non-emptiness can however race with a free to
3174  * pcplist that has not yet increased the pcp->count from 0 to 1. Callers
3175  * that need the guarantee that every CPU has drained can disable the
3176  * optimizing racy check.
3177  */
3178 static void __drain_all_pages(struct zone *zone, bool force_all_cpus)
3179 {
3180 	int cpu;
3181 
3182 	/*
3183 	 * Allocate in the BSS so we won't require allocation in
3184 	 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
3185 	 */
3186 	static cpumask_t cpus_with_pcps;
3187 
3188 	/*
3189 	 * Make sure nobody triggers this path before mm_percpu_wq is fully
3190 	 * initialized.
3191 	 */
3192 	if (WARN_ON_ONCE(!mm_percpu_wq))
3193 		return;
3194 
3195 	/*
3196 	 * Do not drain if one is already in progress unless it's specific to
3197 	 * a zone. Such callers are primarily CMA and memory hotplug and need
3198 	 * the drain to be complete when the call returns.
3199 	 */
3200 	if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
3201 		if (!zone)
3202 			return;
3203 		mutex_lock(&pcpu_drain_mutex);
3204 	}
3205 
3206 	/*
3207 	 * We don't care about racing with CPU hotplug event
3208 	 * as offline notification will cause the notified
3209 	 * cpu to drain that CPU pcps and on_each_cpu_mask
3210 	 * disables preemption as part of its processing
3211 	 */
3212 	for_each_online_cpu(cpu) {
3213 		struct per_cpu_pages *pcp;
3214 		struct zone *z;
3215 		bool has_pcps = false;
3216 
3217 		if (force_all_cpus) {
3218 			/*
3219 			 * The pcp.count check is racy, some callers need a
3220 			 * guarantee that no cpu is missed.
3221 			 */
3222 			has_pcps = true;
3223 		} else if (zone) {
3224 			pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
3225 			if (pcp->count)
3226 				has_pcps = true;
3227 		} else {
3228 			for_each_populated_zone(z) {
3229 				pcp = per_cpu_ptr(z->per_cpu_pageset, cpu);
3230 				if (pcp->count) {
3231 					has_pcps = true;
3232 					break;
3233 				}
3234 			}
3235 		}
3236 
3237 		if (has_pcps)
3238 			cpumask_set_cpu(cpu, &cpus_with_pcps);
3239 		else
3240 			cpumask_clear_cpu(cpu, &cpus_with_pcps);
3241 	}
3242 
3243 	for_each_cpu(cpu, &cpus_with_pcps) {
3244 		struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu);
3245 
3246 		drain->zone = zone;
3247 		INIT_WORK(&drain->work, drain_local_pages_wq);
3248 		queue_work_on(cpu, mm_percpu_wq, &drain->work);
3249 	}
3250 	for_each_cpu(cpu, &cpus_with_pcps)
3251 		flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work);
3252 
3253 	mutex_unlock(&pcpu_drain_mutex);
3254 }
3255 
3256 /*
3257  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
3258  *
3259  * When zone parameter is non-NULL, spill just the single zone's pages.
3260  *
3261  * Note that this can be extremely slow as the draining happens in a workqueue.
3262  */
3263 void drain_all_pages(struct zone *zone)
3264 {
3265 	__drain_all_pages(zone, false);
3266 }
3267 
3268 #ifdef CONFIG_HIBERNATION
3269 
3270 /*
3271  * Touch the watchdog for every WD_PAGE_COUNT pages.
3272  */
3273 #define WD_PAGE_COUNT	(128*1024)
3274 
3275 void mark_free_pages(struct zone *zone)
3276 {
3277 	unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
3278 	unsigned long flags;
3279 	unsigned int order, t;
3280 	struct page *page;
3281 
3282 	if (zone_is_empty(zone))
3283 		return;
3284 
3285 	spin_lock_irqsave(&zone->lock, flags);
3286 
3287 	max_zone_pfn = zone_end_pfn(zone);
3288 	for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
3289 		if (pfn_valid(pfn)) {
3290 			page = pfn_to_page(pfn);
3291 
3292 			if (!--page_count) {
3293 				touch_nmi_watchdog();
3294 				page_count = WD_PAGE_COUNT;
3295 			}
3296 
3297 			if (page_zone(page) != zone)
3298 				continue;
3299 
3300 			if (!swsusp_page_is_forbidden(page))
3301 				swsusp_unset_page_free(page);
3302 		}
3303 
3304 	for_each_migratetype_order(order, t) {
3305 		list_for_each_entry(page,
3306 				&zone->free_area[order].free_list[t], lru) {
3307 			unsigned long i;
3308 
3309 			pfn = page_to_pfn(page);
3310 			for (i = 0; i < (1UL << order); i++) {
3311 				if (!--page_count) {
3312 					touch_nmi_watchdog();
3313 					page_count = WD_PAGE_COUNT;
3314 				}
3315 				swsusp_set_page_free(pfn_to_page(pfn + i));
3316 			}
3317 		}
3318 	}
3319 	spin_unlock_irqrestore(&zone->lock, flags);
3320 }
3321 #endif /* CONFIG_PM */
3322 
3323 static bool free_unref_page_prepare(struct page *page, unsigned long pfn,
3324 							unsigned int order)
3325 {
3326 	int migratetype;
3327 
3328 	if (!free_pcp_prepare(page, order))
3329 		return false;
3330 
3331 	migratetype = get_pfnblock_migratetype(page, pfn);
3332 	set_pcppage_migratetype(page, migratetype);
3333 	return true;
3334 }
3335 
3336 static int nr_pcp_free(struct per_cpu_pages *pcp, int high, int batch,
3337 		       bool free_high)
3338 {
3339 	int min_nr_free, max_nr_free;
3340 
3341 	/* Free everything if batch freeing high-order pages. */
3342 	if (unlikely(free_high))
3343 		return pcp->count;
3344 
3345 	/* Check for PCP disabled or boot pageset */
3346 	if (unlikely(high < batch))
3347 		return 1;
3348 
3349 	/* Leave at least pcp->batch pages on the list */
3350 	min_nr_free = batch;
3351 	max_nr_free = high - batch;
3352 
3353 	/*
3354 	 * Double the number of pages freed each time there is subsequent
3355 	 * freeing of pages without any allocation.
3356 	 */
3357 	batch <<= pcp->free_factor;
3358 	if (batch < max_nr_free)
3359 		pcp->free_factor++;
3360 	batch = clamp(batch, min_nr_free, max_nr_free);
3361 
3362 	return batch;
3363 }
3364 
3365 static int nr_pcp_high(struct per_cpu_pages *pcp, struct zone *zone,
3366 		       bool free_high)
3367 {
3368 	int high = READ_ONCE(pcp->high);
3369 
3370 	if (unlikely(!high || free_high))
3371 		return 0;
3372 
3373 	if (!test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags))
3374 		return high;
3375 
3376 	/*
3377 	 * If reclaim is active, limit the number of pages that can be
3378 	 * stored on pcp lists
3379 	 */
3380 	return min(READ_ONCE(pcp->batch) << 2, high);
3381 }
3382 
3383 static void free_unref_page_commit(struct page *page, int migratetype,
3384 				   unsigned int order)
3385 {
3386 	struct zone *zone = page_zone(page);
3387 	struct per_cpu_pages *pcp;
3388 	int high;
3389 	int pindex;
3390 	bool free_high;
3391 
3392 	__count_vm_event(PGFREE);
3393 	pcp = this_cpu_ptr(zone->per_cpu_pageset);
3394 	pindex = order_to_pindex(migratetype, order);
3395 	list_add(&page->lru, &pcp->lists[pindex]);
3396 	pcp->count += 1 << order;
3397 
3398 	/*
3399 	 * As high-order pages other than THP's stored on PCP can contribute
3400 	 * to fragmentation, limit the number stored when PCP is heavily
3401 	 * freeing without allocation. The remainder after bulk freeing
3402 	 * stops will be drained from vmstat refresh context.
3403 	 */
3404 	free_high = (pcp->free_factor && order && order <= PAGE_ALLOC_COSTLY_ORDER);
3405 
3406 	high = nr_pcp_high(pcp, zone, free_high);
3407 	if (pcp->count >= high) {
3408 		int batch = READ_ONCE(pcp->batch);
3409 
3410 		free_pcppages_bulk(zone, nr_pcp_free(pcp, high, batch, free_high), pcp, pindex);
3411 	}
3412 }
3413 
3414 /*
3415  * Free a pcp page
3416  */
3417 void free_unref_page(struct page *page, unsigned int order)
3418 {
3419 	unsigned long flags;
3420 	unsigned long pfn = page_to_pfn(page);
3421 	int migratetype;
3422 
3423 	if (!free_unref_page_prepare(page, pfn, order))
3424 		return;
3425 
3426 	/*
3427 	 * We only track unmovable, reclaimable and movable on pcp lists.
3428 	 * Place ISOLATE pages on the isolated list because they are being
3429 	 * offlined but treat HIGHATOMIC as movable pages so we can get those
3430 	 * areas back if necessary. Otherwise, we may have to free
3431 	 * excessively into the page allocator
3432 	 */
3433 	migratetype = get_pcppage_migratetype(page);
3434 	if (unlikely(migratetype >= MIGRATE_PCPTYPES)) {
3435 		if (unlikely(is_migrate_isolate(migratetype))) {
3436 			free_one_page(page_zone(page), page, pfn, order, migratetype, FPI_NONE);
3437 			return;
3438 		}
3439 		migratetype = MIGRATE_MOVABLE;
3440 	}
3441 
3442 	local_lock_irqsave(&pagesets.lock, flags);
3443 	free_unref_page_commit(page, migratetype, order);
3444 	local_unlock_irqrestore(&pagesets.lock, flags);
3445 }
3446 
3447 /*
3448  * Free a list of 0-order pages
3449  */
3450 void free_unref_page_list(struct list_head *list)
3451 {
3452 	struct page *page, *next;
3453 	unsigned long flags;
3454 	int batch_count = 0;
3455 	int migratetype;
3456 
3457 	/* Prepare pages for freeing */
3458 	list_for_each_entry_safe(page, next, list, lru) {
3459 		unsigned long pfn = page_to_pfn(page);
3460 		if (!free_unref_page_prepare(page, pfn, 0)) {
3461 			list_del(&page->lru);
3462 			continue;
3463 		}
3464 
3465 		/*
3466 		 * Free isolated pages directly to the allocator, see
3467 		 * comment in free_unref_page.
3468 		 */
3469 		migratetype = get_pcppage_migratetype(page);
3470 		if (unlikely(is_migrate_isolate(migratetype))) {
3471 			list_del(&page->lru);
3472 			free_one_page(page_zone(page), page, pfn, 0, migratetype, FPI_NONE);
3473 			continue;
3474 		}
3475 	}
3476 
3477 	local_lock_irqsave(&pagesets.lock, flags);
3478 	list_for_each_entry_safe(page, next, list, lru) {
3479 		/*
3480 		 * Non-isolated types over MIGRATE_PCPTYPES get added
3481 		 * to the MIGRATE_MOVABLE pcp list.
3482 		 */
3483 		migratetype = get_pcppage_migratetype(page);
3484 		if (unlikely(migratetype >= MIGRATE_PCPTYPES))
3485 			migratetype = MIGRATE_MOVABLE;
3486 
3487 		trace_mm_page_free_batched(page);
3488 		free_unref_page_commit(page, migratetype, 0);
3489 
3490 		/*
3491 		 * Guard against excessive IRQ disabled times when we get
3492 		 * a large list of pages to free.
3493 		 */
3494 		if (++batch_count == SWAP_CLUSTER_MAX) {
3495 			local_unlock_irqrestore(&pagesets.lock, flags);
3496 			batch_count = 0;
3497 			local_lock_irqsave(&pagesets.lock, flags);
3498 		}
3499 	}
3500 	local_unlock_irqrestore(&pagesets.lock, flags);
3501 }
3502 
3503 /*
3504  * split_page takes a non-compound higher-order page, and splits it into
3505  * n (1<<order) sub-pages: page[0..n]
3506  * Each sub-page must be freed individually.
3507  *
3508  * Note: this is probably too low level an operation for use in drivers.
3509  * Please consult with lkml before using this in your driver.
3510  */
3511 void split_page(struct page *page, unsigned int order)
3512 {
3513 	int i;
3514 
3515 	VM_BUG_ON_PAGE(PageCompound(page), page);
3516 	VM_BUG_ON_PAGE(!page_count(page), page);
3517 
3518 	for (i = 1; i < (1 << order); i++)
3519 		set_page_refcounted(page + i);
3520 	split_page_owner(page, 1 << order);
3521 	split_page_memcg(page, 1 << order);
3522 }
3523 EXPORT_SYMBOL_GPL(split_page);
3524 
3525 int __isolate_free_page(struct page *page, unsigned int order)
3526 {
3527 	unsigned long watermark;
3528 	struct zone *zone;
3529 	int mt;
3530 
3531 	BUG_ON(!PageBuddy(page));
3532 
3533 	zone = page_zone(page);
3534 	mt = get_pageblock_migratetype(page);
3535 
3536 	if (!is_migrate_isolate(mt)) {
3537 		/*
3538 		 * Obey watermarks as if the page was being allocated. We can
3539 		 * emulate a high-order watermark check with a raised order-0
3540 		 * watermark, because we already know our high-order page
3541 		 * exists.
3542 		 */
3543 		watermark = zone->_watermark[WMARK_MIN] + (1UL << order);
3544 		if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
3545 			return 0;
3546 
3547 		__mod_zone_freepage_state(zone, -(1UL << order), mt);
3548 	}
3549 
3550 	/* Remove page from free list */
3551 
3552 	del_page_from_free_list(page, zone, order);
3553 
3554 	/*
3555 	 * Set the pageblock if the isolated page is at least half of a
3556 	 * pageblock
3557 	 */
3558 	if (order >= pageblock_order - 1) {
3559 		struct page *endpage = page + (1 << order) - 1;
3560 		for (; page < endpage; page += pageblock_nr_pages) {
3561 			int mt = get_pageblock_migratetype(page);
3562 			/*
3563 			 * Only change normal pageblocks (i.e., they can merge
3564 			 * with others)
3565 			 */
3566 			if (migratetype_is_mergeable(mt))
3567 				set_pageblock_migratetype(page,
3568 							  MIGRATE_MOVABLE);
3569 		}
3570 	}
3571 
3572 
3573 	return 1UL << order;
3574 }
3575 
3576 /**
3577  * __putback_isolated_page - Return a now-isolated page back where we got it
3578  * @page: Page that was isolated
3579  * @order: Order of the isolated page
3580  * @mt: The page's pageblock's migratetype
3581  *
3582  * This function is meant to return a page pulled from the free lists via
3583  * __isolate_free_page back to the free lists they were pulled from.
3584  */
3585 void __putback_isolated_page(struct page *page, unsigned int order, int mt)
3586 {
3587 	struct zone *zone = page_zone(page);
3588 
3589 	/* zone lock should be held when this function is called */
3590 	lockdep_assert_held(&zone->lock);
3591 
3592 	/* Return isolated page to tail of freelist. */
3593 	__free_one_page(page, page_to_pfn(page), zone, order, mt,
3594 			FPI_SKIP_REPORT_NOTIFY | FPI_TO_TAIL);
3595 }
3596 
3597 /*
3598  * Update NUMA hit/miss statistics
3599  *
3600  * Must be called with interrupts disabled.
3601  */
3602 static inline void zone_statistics(struct zone *preferred_zone, struct zone *z,
3603 				   long nr_account)
3604 {
3605 #ifdef CONFIG_NUMA
3606 	enum numa_stat_item local_stat = NUMA_LOCAL;
3607 
3608 	/* skip numa counters update if numa stats is disabled */
3609 	if (!static_branch_likely(&vm_numa_stat_key))
3610 		return;
3611 
3612 	if (zone_to_nid(z) != numa_node_id())
3613 		local_stat = NUMA_OTHER;
3614 
3615 	if (zone_to_nid(z) == zone_to_nid(preferred_zone))
3616 		__count_numa_events(z, NUMA_HIT, nr_account);
3617 	else {
3618 		__count_numa_events(z, NUMA_MISS, nr_account);
3619 		__count_numa_events(preferred_zone, NUMA_FOREIGN, nr_account);
3620 	}
3621 	__count_numa_events(z, local_stat, nr_account);
3622 #endif
3623 }
3624 
3625 /* Remove page from the per-cpu list, caller must protect the list */
3626 static inline
3627 struct page *__rmqueue_pcplist(struct zone *zone, unsigned int order,
3628 			int migratetype,
3629 			unsigned int alloc_flags,
3630 			struct per_cpu_pages *pcp,
3631 			struct list_head *list)
3632 {
3633 	struct page *page;
3634 
3635 	do {
3636 		if (list_empty(list)) {
3637 			int batch = READ_ONCE(pcp->batch);
3638 			int alloced;
3639 
3640 			/*
3641 			 * Scale batch relative to order if batch implies
3642 			 * free pages can be stored on the PCP. Batch can
3643 			 * be 1 for small zones or for boot pagesets which
3644 			 * should never store free pages as the pages may
3645 			 * belong to arbitrary zones.
3646 			 */
3647 			if (batch > 1)
3648 				batch = max(batch >> order, 2);
3649 			alloced = rmqueue_bulk(zone, order,
3650 					batch, list,
3651 					migratetype, alloc_flags);
3652 
3653 			pcp->count += alloced << order;
3654 			if (unlikely(list_empty(list)))
3655 				return NULL;
3656 		}
3657 
3658 		page = list_first_entry(list, struct page, lru);
3659 		list_del(&page->lru);
3660 		pcp->count -= 1 << order;
3661 	} while (check_new_pcp(page, order));
3662 
3663 	return page;
3664 }
3665 
3666 /* Lock and remove page from the per-cpu list */
3667 static struct page *rmqueue_pcplist(struct zone *preferred_zone,
3668 			struct zone *zone, unsigned int order,
3669 			gfp_t gfp_flags, int migratetype,
3670 			unsigned int alloc_flags)
3671 {
3672 	struct per_cpu_pages *pcp;
3673 	struct list_head *list;
3674 	struct page *page;
3675 	unsigned long flags;
3676 
3677 	local_lock_irqsave(&pagesets.lock, flags);
3678 
3679 	/*
3680 	 * On allocation, reduce the number of pages that are batch freed.
3681 	 * See nr_pcp_free() where free_factor is increased for subsequent
3682 	 * frees.
3683 	 */
3684 	pcp = this_cpu_ptr(zone->per_cpu_pageset);
3685 	pcp->free_factor >>= 1;
3686 	list = &pcp->lists[order_to_pindex(migratetype, order)];
3687 	page = __rmqueue_pcplist(zone, order, migratetype, alloc_flags, pcp, list);
3688 	local_unlock_irqrestore(&pagesets.lock, flags);
3689 	if (page) {
3690 		__count_zid_vm_events(PGALLOC, page_zonenum(page), 1);
3691 		zone_statistics(preferred_zone, zone, 1);
3692 	}
3693 	return page;
3694 }
3695 
3696 /*
3697  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
3698  */
3699 static inline
3700 struct page *rmqueue(struct zone *preferred_zone,
3701 			struct zone *zone, unsigned int order,
3702 			gfp_t gfp_flags, unsigned int alloc_flags,
3703 			int migratetype)
3704 {
3705 	unsigned long flags;
3706 	struct page *page;
3707 
3708 	if (likely(pcp_allowed_order(order))) {
3709 		/*
3710 		 * MIGRATE_MOVABLE pcplist could have the pages on CMA area and
3711 		 * we need to skip it when CMA area isn't allowed.
3712 		 */
3713 		if (!IS_ENABLED(CONFIG_CMA) || alloc_flags & ALLOC_CMA ||
3714 				migratetype != MIGRATE_MOVABLE) {
3715 			page = rmqueue_pcplist(preferred_zone, zone, order,
3716 					gfp_flags, migratetype, alloc_flags);
3717 			goto out;
3718 		}
3719 	}
3720 
3721 	/*
3722 	 * We most definitely don't want callers attempting to
3723 	 * allocate greater than order-1 page units with __GFP_NOFAIL.
3724 	 */
3725 	WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
3726 
3727 	do {
3728 		page = NULL;
3729 		spin_lock_irqsave(&zone->lock, flags);
3730 		/*
3731 		 * order-0 request can reach here when the pcplist is skipped
3732 		 * due to non-CMA allocation context. HIGHATOMIC area is
3733 		 * reserved for high-order atomic allocation, so order-0
3734 		 * request should skip it.
3735 		 */
3736 		if (order > 0 && alloc_flags & ALLOC_HARDER) {
3737 			page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3738 			if (page)
3739 				trace_mm_page_alloc_zone_locked(page, order, migratetype);
3740 		}
3741 		if (!page) {
3742 			page = __rmqueue(zone, order, migratetype, alloc_flags);
3743 			if (!page)
3744 				goto failed;
3745 		}
3746 		__mod_zone_freepage_state(zone, -(1 << order),
3747 					  get_pcppage_migratetype(page));
3748 		spin_unlock_irqrestore(&zone->lock, flags);
3749 	} while (check_new_pages(page, order));
3750 
3751 	__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3752 	zone_statistics(preferred_zone, zone, 1);
3753 
3754 out:
3755 	/* Separate test+clear to avoid unnecessary atomics */
3756 	if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
3757 		clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
3758 		wakeup_kswapd(zone, 0, 0, zone_idx(zone));
3759 	}
3760 
3761 	VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3762 	return page;
3763 
3764 failed:
3765 	spin_unlock_irqrestore(&zone->lock, flags);
3766 	return NULL;
3767 }
3768 
3769 #ifdef CONFIG_FAIL_PAGE_ALLOC
3770 
3771 static struct {
3772 	struct fault_attr attr;
3773 
3774 	bool ignore_gfp_highmem;
3775 	bool ignore_gfp_reclaim;
3776 	u32 min_order;
3777 } fail_page_alloc = {
3778 	.attr = FAULT_ATTR_INITIALIZER,
3779 	.ignore_gfp_reclaim = true,
3780 	.ignore_gfp_highmem = true,
3781 	.min_order = 1,
3782 };
3783 
3784 static int __init setup_fail_page_alloc(char *str)
3785 {
3786 	return setup_fault_attr(&fail_page_alloc.attr, str);
3787 }
3788 __setup("fail_page_alloc=", setup_fail_page_alloc);
3789 
3790 static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3791 {
3792 	if (order < fail_page_alloc.min_order)
3793 		return false;
3794 	if (gfp_mask & __GFP_NOFAIL)
3795 		return false;
3796 	if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3797 		return false;
3798 	if (fail_page_alloc.ignore_gfp_reclaim &&
3799 			(gfp_mask & __GFP_DIRECT_RECLAIM))
3800 		return false;
3801 
3802 	return should_fail(&fail_page_alloc.attr, 1 << order);
3803 }
3804 
3805 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3806 
3807 static int __init fail_page_alloc_debugfs(void)
3808 {
3809 	umode_t mode = S_IFREG | 0600;
3810 	struct dentry *dir;
3811 
3812 	dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3813 					&fail_page_alloc.attr);
3814 
3815 	debugfs_create_bool("ignore-gfp-wait", mode, dir,
3816 			    &fail_page_alloc.ignore_gfp_reclaim);
3817 	debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3818 			    &fail_page_alloc.ignore_gfp_highmem);
3819 	debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order);
3820 
3821 	return 0;
3822 }
3823 
3824 late_initcall(fail_page_alloc_debugfs);
3825 
3826 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
3827 
3828 #else /* CONFIG_FAIL_PAGE_ALLOC */
3829 
3830 static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3831 {
3832 	return false;
3833 }
3834 
3835 #endif /* CONFIG_FAIL_PAGE_ALLOC */
3836 
3837 noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3838 {
3839 	return __should_fail_alloc_page(gfp_mask, order);
3840 }
3841 ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE);
3842 
3843 static inline long __zone_watermark_unusable_free(struct zone *z,
3844 				unsigned int order, unsigned int alloc_flags)
3845 {
3846 	const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3847 	long unusable_free = (1 << order) - 1;
3848 
3849 	/*
3850 	 * If the caller does not have rights to ALLOC_HARDER then subtract
3851 	 * the high-atomic reserves. This will over-estimate the size of the
3852 	 * atomic reserve but it avoids a search.
3853 	 */
3854 	if (likely(!alloc_harder))
3855 		unusable_free += z->nr_reserved_highatomic;
3856 
3857 #ifdef CONFIG_CMA
3858 	/* If allocation can't use CMA areas don't use free CMA pages */
3859 	if (!(alloc_flags & ALLOC_CMA))
3860 		unusable_free += zone_page_state(z, NR_FREE_CMA_PAGES);
3861 #endif
3862 
3863 	return unusable_free;
3864 }
3865 
3866 /*
3867  * Return true if free base pages are above 'mark'. For high-order checks it
3868  * will return true of the order-0 watermark is reached and there is at least
3869  * one free page of a suitable size. Checking now avoids taking the zone lock
3870  * to check in the allocation paths if no pages are free.
3871  */
3872 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3873 			 int highest_zoneidx, unsigned int alloc_flags,
3874 			 long free_pages)
3875 {
3876 	long min = mark;
3877 	int o;
3878 	const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3879 
3880 	/* free_pages may go negative - that's OK */
3881 	free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags);
3882 
3883 	if (alloc_flags & ALLOC_HIGH)
3884 		min -= min / 2;
3885 
3886 	if (unlikely(alloc_harder)) {
3887 		/*
3888 		 * OOM victims can try even harder than normal ALLOC_HARDER
3889 		 * users on the grounds that it's definitely going to be in
3890 		 * the exit path shortly and free memory. Any allocation it
3891 		 * makes during the free path will be small and short-lived.
3892 		 */
3893 		if (alloc_flags & ALLOC_OOM)
3894 			min -= min / 2;
3895 		else
3896 			min -= min / 4;
3897 	}
3898 
3899 	/*
3900 	 * Check watermarks for an order-0 allocation request. If these
3901 	 * are not met, then a high-order request also cannot go ahead
3902 	 * even if a suitable page happened to be free.
3903 	 */
3904 	if (free_pages <= min + z->lowmem_reserve[highest_zoneidx])
3905 		return false;
3906 
3907 	/* If this is an order-0 request then the watermark is fine */
3908 	if (!order)
3909 		return true;
3910 
3911 	/* For a high-order request, check at least one suitable page is free */
3912 	for (o = order; o < MAX_ORDER; o++) {
3913 		struct free_area *area = &z->free_area[o];
3914 		int mt;
3915 
3916 		if (!area->nr_free)
3917 			continue;
3918 
3919 		for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3920 			if (!free_area_empty(area, mt))
3921 				return true;
3922 		}
3923 
3924 #ifdef CONFIG_CMA
3925 		if ((alloc_flags & ALLOC_CMA) &&
3926 		    !free_area_empty(area, MIGRATE_CMA)) {
3927 			return true;
3928 		}
3929 #endif
3930 		if (alloc_harder && !free_area_empty(area, MIGRATE_HIGHATOMIC))
3931 			return true;
3932 	}
3933 	return false;
3934 }
3935 
3936 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3937 		      int highest_zoneidx, unsigned int alloc_flags)
3938 {
3939 	return __zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3940 					zone_page_state(z, NR_FREE_PAGES));
3941 }
3942 
3943 static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3944 				unsigned long mark, int highest_zoneidx,
3945 				unsigned int alloc_flags, gfp_t gfp_mask)
3946 {
3947 	long free_pages;
3948 
3949 	free_pages = zone_page_state(z, NR_FREE_PAGES);
3950 
3951 	/*
3952 	 * Fast check for order-0 only. If this fails then the reserves
3953 	 * need to be calculated.
3954 	 */
3955 	if (!order) {
3956 		long fast_free;
3957 
3958 		fast_free = free_pages;
3959 		fast_free -= __zone_watermark_unusable_free(z, 0, alloc_flags);
3960 		if (fast_free > mark + z->lowmem_reserve[highest_zoneidx])
3961 			return true;
3962 	}
3963 
3964 	if (__zone_watermark_ok(z, order, mark, highest_zoneidx, alloc_flags,
3965 					free_pages))
3966 		return true;
3967 	/*
3968 	 * Ignore watermark boosting for GFP_ATOMIC order-0 allocations
3969 	 * when checking the min watermark. The min watermark is the
3970 	 * point where boosting is ignored so that kswapd is woken up
3971 	 * when below the low watermark.
3972 	 */
3973 	if (unlikely(!order && (gfp_mask & __GFP_ATOMIC) && z->watermark_boost
3974 		&& ((alloc_flags & ALLOC_WMARK_MASK) == WMARK_MIN))) {
3975 		mark = z->_watermark[WMARK_MIN];
3976 		return __zone_watermark_ok(z, order, mark, highest_zoneidx,
3977 					alloc_flags, free_pages);
3978 	}
3979 
3980 	return false;
3981 }
3982 
3983 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3984 			unsigned long mark, int highest_zoneidx)
3985 {
3986 	long free_pages = zone_page_state(z, NR_FREE_PAGES);
3987 
3988 	if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3989 		free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3990 
3991 	return __zone_watermark_ok(z, order, mark, highest_zoneidx, 0,
3992 								free_pages);
3993 }
3994 
3995 #ifdef CONFIG_NUMA
3996 int __read_mostly node_reclaim_distance = RECLAIM_DISTANCE;
3997 
3998 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3999 {
4000 	return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
4001 				node_reclaim_distance;
4002 }
4003 #else	/* CONFIG_NUMA */
4004 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
4005 {
4006 	return true;
4007 }
4008 #endif	/* CONFIG_NUMA */
4009 
4010 /*
4011  * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid
4012  * fragmentation is subtle. If the preferred zone was HIGHMEM then
4013  * premature use of a lower zone may cause lowmem pressure problems that
4014  * are worse than fragmentation. If the next zone is ZONE_DMA then it is
4015  * probably too small. It only makes sense to spread allocations to avoid
4016  * fragmentation between the Normal and DMA32 zones.
4017  */
4018 static inline unsigned int
4019 alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask)
4020 {
4021 	unsigned int alloc_flags;
4022 
4023 	/*
4024 	 * __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD
4025 	 * to save a branch.
4026 	 */
4027 	alloc_flags = (__force int) (gfp_mask & __GFP_KSWAPD_RECLAIM);
4028 
4029 #ifdef CONFIG_ZONE_DMA32
4030 	if (!zone)
4031 		return alloc_flags;
4032 
4033 	if (zone_idx(zone) != ZONE_NORMAL)
4034 		return alloc_flags;
4035 
4036 	/*
4037 	 * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and
4038 	 * the pointer is within zone->zone_pgdat->node_zones[]. Also assume
4039 	 * on UMA that if Normal is populated then so is DMA32.
4040 	 */
4041 	BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1);
4042 	if (nr_online_nodes > 1 && !populated_zone(--zone))
4043 		return alloc_flags;
4044 
4045 	alloc_flags |= ALLOC_NOFRAGMENT;
4046 #endif /* CONFIG_ZONE_DMA32 */
4047 	return alloc_flags;
4048 }
4049 
4050 /* Must be called after current_gfp_context() which can change gfp_mask */
4051 static inline unsigned int gfp_to_alloc_flags_cma(gfp_t gfp_mask,
4052 						  unsigned int alloc_flags)
4053 {
4054 #ifdef CONFIG_CMA
4055 	if (gfp_migratetype(gfp_mask) == MIGRATE_MOVABLE)
4056 		alloc_flags |= ALLOC_CMA;
4057 #endif
4058 	return alloc_flags;
4059 }
4060 
4061 /*
4062  * get_page_from_freelist goes through the zonelist trying to allocate
4063  * a page.
4064  */
4065 static struct page *
4066 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
4067 						const struct alloc_context *ac)
4068 {
4069 	struct zoneref *z;
4070 	struct zone *zone;
4071 	struct pglist_data *last_pgdat_dirty_limit = NULL;
4072 	bool no_fallback;
4073 
4074 retry:
4075 	/*
4076 	 * Scan zonelist, looking for a zone with enough free.
4077 	 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
4078 	 */
4079 	no_fallback = alloc_flags & ALLOC_NOFRAGMENT;
4080 	z = ac->preferred_zoneref;
4081 	for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx,
4082 					ac->nodemask) {
4083 		struct page *page;
4084 		unsigned long mark;
4085 
4086 		if (cpusets_enabled() &&
4087 			(alloc_flags & ALLOC_CPUSET) &&
4088 			!__cpuset_zone_allowed(zone, gfp_mask))
4089 				continue;
4090 		/*
4091 		 * When allocating a page cache page for writing, we
4092 		 * want to get it from a node that is within its dirty
4093 		 * limit, such that no single node holds more than its
4094 		 * proportional share of globally allowed dirty pages.
4095 		 * The dirty limits take into account the node's
4096 		 * lowmem reserves and high watermark so that kswapd
4097 		 * should be able to balance it without having to
4098 		 * write pages from its LRU list.
4099 		 *
4100 		 * XXX: For now, allow allocations to potentially
4101 		 * exceed the per-node dirty limit in the slowpath
4102 		 * (spread_dirty_pages unset) before going into reclaim,
4103 		 * which is important when on a NUMA setup the allowed
4104 		 * nodes are together not big enough to reach the
4105 		 * global limit.  The proper fix for these situations
4106 		 * will require awareness of nodes in the
4107 		 * dirty-throttling and the flusher threads.
4108 		 */
4109 		if (ac->spread_dirty_pages) {
4110 			if (last_pgdat_dirty_limit == zone->zone_pgdat)
4111 				continue;
4112 
4113 			if (!node_dirty_ok(zone->zone_pgdat)) {
4114 				last_pgdat_dirty_limit = zone->zone_pgdat;
4115 				continue;
4116 			}
4117 		}
4118 
4119 		if (no_fallback && nr_online_nodes > 1 &&
4120 		    zone != ac->preferred_zoneref->zone) {
4121 			int local_nid;
4122 
4123 			/*
4124 			 * If moving to a remote node, retry but allow
4125 			 * fragmenting fallbacks. Locality is more important
4126 			 * than fragmentation avoidance.
4127 			 */
4128 			local_nid = zone_to_nid(ac->preferred_zoneref->zone);
4129 			if (zone_to_nid(zone) != local_nid) {
4130 				alloc_flags &= ~ALLOC_NOFRAGMENT;
4131 				goto retry;
4132 			}
4133 		}
4134 
4135 		mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);
4136 		if (!zone_watermark_fast(zone, order, mark,
4137 				       ac->highest_zoneidx, alloc_flags,
4138 				       gfp_mask)) {
4139 			int ret;
4140 
4141 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
4142 			/*
4143 			 * Watermark failed for this zone, but see if we can
4144 			 * grow this zone if it contains deferred pages.
4145 			 */
4146 			if (static_branch_unlikely(&deferred_pages)) {
4147 				if (_deferred_grow_zone(zone, order))
4148 					goto try_this_zone;
4149 			}
4150 #endif
4151 			/* Checked here to keep the fast path fast */
4152 			BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
4153 			if (alloc_flags & ALLOC_NO_WATERMARKS)
4154 				goto try_this_zone;
4155 
4156 			if (!node_reclaim_enabled() ||
4157 			    !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
4158 				continue;
4159 
4160 			ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
4161 			switch (ret) {
4162 			case NODE_RECLAIM_NOSCAN:
4163 				/* did not scan */
4164 				continue;
4165 			case NODE_RECLAIM_FULL:
4166 				/* scanned but unreclaimable */
4167 				continue;
4168 			default:
4169 				/* did we reclaim enough */
4170 				if (zone_watermark_ok(zone, order, mark,
4171 					ac->highest_zoneidx, alloc_flags))
4172 					goto try_this_zone;
4173 
4174 				continue;
4175 			}
4176 		}
4177 
4178 try_this_zone:
4179 		page = rmqueue(ac->preferred_zoneref->zone, zone, order,
4180 				gfp_mask, alloc_flags, ac->migratetype);
4181 		if (page) {
4182 			prep_new_page(page, order, gfp_mask, alloc_flags);
4183 
4184 			/*
4185 			 * If this is a high-order atomic allocation then check
4186 			 * if the pageblock should be reserved for the future
4187 			 */
4188 			if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
4189 				reserve_highatomic_pageblock(page, zone, order);
4190 
4191 			return page;
4192 		} else {
4193 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
4194 			/* Try again if zone has deferred pages */
4195 			if (static_branch_unlikely(&deferred_pages)) {
4196 				if (_deferred_grow_zone(zone, order))
4197 					goto try_this_zone;
4198 			}
4199 #endif
4200 		}
4201 	}
4202 
4203 	/*
4204 	 * It's possible on a UMA machine to get through all zones that are
4205 	 * fragmented. If avoiding fragmentation, reset and try again.
4206 	 */
4207 	if (no_fallback) {
4208 		alloc_flags &= ~ALLOC_NOFRAGMENT;
4209 		goto retry;
4210 	}
4211 
4212 	return NULL;
4213 }
4214 
4215 static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
4216 {
4217 	unsigned int filter = SHOW_MEM_FILTER_NODES;
4218 
4219 	/*
4220 	 * This documents exceptions given to allocations in certain
4221 	 * contexts that are allowed to allocate outside current's set
4222 	 * of allowed nodes.
4223 	 */
4224 	if (!(gfp_mask & __GFP_NOMEMALLOC))
4225 		if (tsk_is_oom_victim(current) ||
4226 		    (current->flags & (PF_MEMALLOC | PF_EXITING)))
4227 			filter &= ~SHOW_MEM_FILTER_NODES;
4228 	if (!in_task() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
4229 		filter &= ~SHOW_MEM_FILTER_NODES;
4230 
4231 	show_mem(filter, nodemask);
4232 }
4233 
4234 void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
4235 {
4236 	struct va_format vaf;
4237 	va_list args;
4238 	static DEFINE_RATELIMIT_STATE(nopage_rs, 10*HZ, 1);
4239 
4240 	if ((gfp_mask & __GFP_NOWARN) ||
4241 	     !__ratelimit(&nopage_rs) ||
4242 	     ((gfp_mask & __GFP_DMA) && !has_managed_dma()))
4243 		return;
4244 
4245 	va_start(args, fmt);
4246 	vaf.fmt = fmt;
4247 	vaf.va = &args;
4248 	pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl",
4249 			current->comm, &vaf, gfp_mask, &gfp_mask,
4250 			nodemask_pr_args(nodemask));
4251 	va_end(args);
4252 
4253 	cpuset_print_current_mems_allowed();
4254 	pr_cont("\n");
4255 	dump_stack();
4256 	warn_alloc_show_mem(gfp_mask, nodemask);
4257 }
4258 
4259 static inline struct page *
4260 __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
4261 			      unsigned int alloc_flags,
4262 			      const struct alloc_context *ac)
4263 {
4264 	struct page *page;
4265 
4266 	page = get_page_from_freelist(gfp_mask, order,
4267 			alloc_flags|ALLOC_CPUSET, ac);
4268 	/*
4269 	 * fallback to ignore cpuset restriction if our nodes
4270 	 * are depleted
4271 	 */
4272 	if (!page)
4273 		page = get_page_from_freelist(gfp_mask, order,
4274 				alloc_flags, ac);
4275 
4276 	return page;
4277 }
4278 
4279 static inline struct page *
4280 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
4281 	const struct alloc_context *ac, unsigned long *did_some_progress)
4282 {
4283 	struct oom_control oc = {
4284 		.zonelist = ac->zonelist,
4285 		.nodemask = ac->nodemask,
4286 		.memcg = NULL,
4287 		.gfp_mask = gfp_mask,
4288 		.order = order,
4289 	};
4290 	struct page *page;
4291 
4292 	*did_some_progress = 0;
4293 
4294 	/*
4295 	 * Acquire the oom lock.  If that fails, somebody else is
4296 	 * making progress for us.
4297 	 */
4298 	if (!mutex_trylock(&oom_lock)) {
4299 		*did_some_progress = 1;
4300 		schedule_timeout_uninterruptible(1);
4301 		return NULL;
4302 	}
4303 
4304 	/*
4305 	 * Go through the zonelist yet one more time, keep very high watermark
4306 	 * here, this is only to catch a parallel oom killing, we must fail if
4307 	 * we're still under heavy pressure. But make sure that this reclaim
4308 	 * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
4309 	 * allocation which will never fail due to oom_lock already held.
4310 	 */
4311 	page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
4312 				      ~__GFP_DIRECT_RECLAIM, order,
4313 				      ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
4314 	if (page)
4315 		goto out;
4316 
4317 	/* Coredumps can quickly deplete all memory reserves */
4318 	if (current->flags & PF_DUMPCORE)
4319 		goto out;
4320 	/* The OOM killer will not help higher order allocs */
4321 	if (order > PAGE_ALLOC_COSTLY_ORDER)
4322 		goto out;
4323 	/*
4324 	 * We have already exhausted all our reclaim opportunities without any
4325 	 * success so it is time to admit defeat. We will skip the OOM killer
4326 	 * because it is very likely that the caller has a more reasonable
4327 	 * fallback than shooting a random task.
4328 	 *
4329 	 * The OOM killer may not free memory on a specific node.
4330 	 */
4331 	if (gfp_mask & (__GFP_RETRY_MAYFAIL | __GFP_THISNODE))
4332 		goto out;
4333 	/* The OOM killer does not needlessly kill tasks for lowmem */
4334 	if (ac->highest_zoneidx < ZONE_NORMAL)
4335 		goto out;
4336 	if (pm_suspended_storage())
4337 		goto out;
4338 	/*
4339 	 * XXX: GFP_NOFS allocations should rather fail than rely on
4340 	 * other request to make a forward progress.
4341 	 * We are in an unfortunate situation where out_of_memory cannot
4342 	 * do much for this context but let's try it to at least get
4343 	 * access to memory reserved if the current task is killed (see
4344 	 * out_of_memory). Once filesystems are ready to handle allocation
4345 	 * failures more gracefully we should just bail out here.
4346 	 */
4347 
4348 	/* Exhausted what can be done so it's blame time */
4349 	if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
4350 		*did_some_progress = 1;
4351 
4352 		/*
4353 		 * Help non-failing allocations by giving them access to memory
4354 		 * reserves
4355 		 */
4356 		if (gfp_mask & __GFP_NOFAIL)
4357 			page = __alloc_pages_cpuset_fallback(gfp_mask, order,
4358 					ALLOC_NO_WATERMARKS, ac);
4359 	}
4360 out:
4361 	mutex_unlock(&oom_lock);
4362 	return page;
4363 }
4364 
4365 /*
4366  * Maximum number of compaction retries with a progress before OOM
4367  * killer is consider as the only way to move forward.
4368  */
4369 #define MAX_COMPACT_RETRIES 16
4370 
4371 #ifdef CONFIG_COMPACTION
4372 /* Try memory compaction for high-order allocations before reclaim */
4373 static struct page *
4374 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4375 		unsigned int alloc_flags, const struct alloc_context *ac,
4376 		enum compact_priority prio, enum compact_result *compact_result)
4377 {
4378 	struct page *page = NULL;
4379 	unsigned long pflags;
4380 	unsigned int noreclaim_flag;
4381 
4382 	if (!order)
4383 		return NULL;
4384 
4385 	psi_memstall_enter(&pflags);
4386 	delayacct_compact_start();
4387 	noreclaim_flag = memalloc_noreclaim_save();
4388 
4389 	*compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
4390 								prio, &page);
4391 
4392 	memalloc_noreclaim_restore(noreclaim_flag);
4393 	psi_memstall_leave(&pflags);
4394 	delayacct_compact_end();
4395 
4396 	if (*compact_result == COMPACT_SKIPPED)
4397 		return NULL;
4398 	/*
4399 	 * At least in one zone compaction wasn't deferred or skipped, so let's
4400 	 * count a compaction stall
4401 	 */
4402 	count_vm_event(COMPACTSTALL);
4403 
4404 	/* Prep a captured page if available */
4405 	if (page)
4406 		prep_new_page(page, order, gfp_mask, alloc_flags);
4407 
4408 	/* Try get a page from the freelist if available */
4409 	if (!page)
4410 		page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4411 
4412 	if (page) {
4413 		struct zone *zone = page_zone(page);
4414 
4415 		zone->compact_blockskip_flush = false;
4416 		compaction_defer_reset(zone, order, true);
4417 		count_vm_event(COMPACTSUCCESS);
4418 		return page;
4419 	}
4420 
4421 	/*
4422 	 * It's bad if compaction run occurs and fails. The most likely reason
4423 	 * is that pages exist, but not enough to satisfy watermarks.
4424 	 */
4425 	count_vm_event(COMPACTFAIL);
4426 
4427 	cond_resched();
4428 
4429 	return NULL;
4430 }
4431 
4432 static inline bool
4433 should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
4434 		     enum compact_result compact_result,
4435 		     enum compact_priority *compact_priority,
4436 		     int *compaction_retries)
4437 {
4438 	int max_retries = MAX_COMPACT_RETRIES;
4439 	int min_priority;
4440 	bool ret = false;
4441 	int retries = *compaction_retries;
4442 	enum compact_priority priority = *compact_priority;
4443 
4444 	if (!order)
4445 		return false;
4446 
4447 	if (fatal_signal_pending(current))
4448 		return false;
4449 
4450 	if (compaction_made_progress(compact_result))
4451 		(*compaction_retries)++;
4452 
4453 	/*
4454 	 * compaction considers all the zone as desperately out of memory
4455 	 * so it doesn't really make much sense to retry except when the
4456 	 * failure could be caused by insufficient priority
4457 	 */
4458 	if (compaction_failed(compact_result))
4459 		goto check_priority;
4460 
4461 	/*
4462 	 * compaction was skipped because there are not enough order-0 pages
4463 	 * to work with, so we retry only if it looks like reclaim can help.
4464 	 */
4465 	if (compaction_needs_reclaim(compact_result)) {
4466 		ret = compaction_zonelist_suitable(ac, order, alloc_flags);
4467 		goto out;
4468 	}
4469 
4470 	/*
4471 	 * make sure the compaction wasn't deferred or didn't bail out early
4472 	 * due to locks contention before we declare that we should give up.
4473 	 * But the next retry should use a higher priority if allowed, so
4474 	 * we don't just keep bailing out endlessly.
4475 	 */
4476 	if (compaction_withdrawn(compact_result)) {
4477 		goto check_priority;
4478 	}
4479 
4480 	/*
4481 	 * !costly requests are much more important than __GFP_RETRY_MAYFAIL
4482 	 * costly ones because they are de facto nofail and invoke OOM
4483 	 * killer to move on while costly can fail and users are ready
4484 	 * to cope with that. 1/4 retries is rather arbitrary but we
4485 	 * would need much more detailed feedback from compaction to
4486 	 * make a better decision.
4487 	 */
4488 	if (order > PAGE_ALLOC_COSTLY_ORDER)
4489 		max_retries /= 4;
4490 	if (*compaction_retries <= max_retries) {
4491 		ret = true;
4492 		goto out;
4493 	}
4494 
4495 	/*
4496 	 * Make sure there are attempts at the highest priority if we exhausted
4497 	 * all retries or failed at the lower priorities.
4498 	 */
4499 check_priority:
4500 	min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
4501 			MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
4502 
4503 	if (*compact_priority > min_priority) {
4504 		(*compact_priority)--;
4505 		*compaction_retries = 0;
4506 		ret = true;
4507 	}
4508 out:
4509 	trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
4510 	return ret;
4511 }
4512 #else
4513 static inline struct page *
4514 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
4515 		unsigned int alloc_flags, const struct alloc_context *ac,
4516 		enum compact_priority prio, enum compact_result *compact_result)
4517 {
4518 	*compact_result = COMPACT_SKIPPED;
4519 	return NULL;
4520 }
4521 
4522 static inline bool
4523 should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
4524 		     enum compact_result compact_result,
4525 		     enum compact_priority *compact_priority,
4526 		     int *compaction_retries)
4527 {
4528 	struct zone *zone;
4529 	struct zoneref *z;
4530 
4531 	if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
4532 		return false;
4533 
4534 	/*
4535 	 * There are setups with compaction disabled which would prefer to loop
4536 	 * inside the allocator rather than hit the oom killer prematurely.
4537 	 * Let's give them a good hope and keep retrying while the order-0
4538 	 * watermarks are OK.
4539 	 */
4540 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4541 				ac->highest_zoneidx, ac->nodemask) {
4542 		if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
4543 					ac->highest_zoneidx, alloc_flags))
4544 			return true;
4545 	}
4546 	return false;
4547 }
4548 #endif /* CONFIG_COMPACTION */
4549 
4550 #ifdef CONFIG_LOCKDEP
4551 static struct lockdep_map __fs_reclaim_map =
4552 	STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
4553 
4554 static bool __need_reclaim(gfp_t gfp_mask)
4555 {
4556 	/* no reclaim without waiting on it */
4557 	if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
4558 		return false;
4559 
4560 	/* this guy won't enter reclaim */
4561 	if (current->flags & PF_MEMALLOC)
4562 		return false;
4563 
4564 	if (gfp_mask & __GFP_NOLOCKDEP)
4565 		return false;
4566 
4567 	return true;
4568 }
4569 
4570 void __fs_reclaim_acquire(unsigned long ip)
4571 {
4572 	lock_acquire_exclusive(&__fs_reclaim_map, 0, 0, NULL, ip);
4573 }
4574 
4575 void __fs_reclaim_release(unsigned long ip)
4576 {
4577 	lock_release(&__fs_reclaim_map, ip);
4578 }
4579 
4580 void fs_reclaim_acquire(gfp_t gfp_mask)
4581 {
4582 	gfp_mask = current_gfp_context(gfp_mask);
4583 
4584 	if (__need_reclaim(gfp_mask)) {
4585 		if (gfp_mask & __GFP_FS)
4586 			__fs_reclaim_acquire(_RET_IP_);
4587 
4588 #ifdef CONFIG_MMU_NOTIFIER
4589 		lock_map_acquire(&__mmu_notifier_invalidate_range_start_map);
4590 		lock_map_release(&__mmu_notifier_invalidate_range_start_map);
4591 #endif
4592 
4593 	}
4594 }
4595 EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
4596 
4597 void fs_reclaim_release(gfp_t gfp_mask)
4598 {
4599 	gfp_mask = current_gfp_context(gfp_mask);
4600 
4601 	if (__need_reclaim(gfp_mask)) {
4602 		if (gfp_mask & __GFP_FS)
4603 			__fs_reclaim_release(_RET_IP_);
4604 	}
4605 }
4606 EXPORT_SYMBOL_GPL(fs_reclaim_release);
4607 #endif
4608 
4609 /* Perform direct synchronous page reclaim */
4610 static unsigned long
4611 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
4612 					const struct alloc_context *ac)
4613 {
4614 	unsigned int noreclaim_flag;
4615 	unsigned long progress;
4616 
4617 	cond_resched();
4618 
4619 	/* We now go into synchronous reclaim */
4620 	cpuset_memory_pressure_bump();
4621 	fs_reclaim_acquire(gfp_mask);
4622 	noreclaim_flag = memalloc_noreclaim_save();
4623 
4624 	progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
4625 								ac->nodemask);
4626 
4627 	memalloc_noreclaim_restore(noreclaim_flag);
4628 	fs_reclaim_release(gfp_mask);
4629 
4630 	cond_resched();
4631 
4632 	return progress;
4633 }
4634 
4635 /* The really slow allocator path where we enter direct reclaim */
4636 static inline struct page *
4637 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
4638 		unsigned int alloc_flags, const struct alloc_context *ac,
4639 		unsigned long *did_some_progress)
4640 {
4641 	struct page *page = NULL;
4642 	unsigned long pflags;
4643 	bool drained = false;
4644 
4645 	psi_memstall_enter(&pflags);
4646 	*did_some_progress = __perform_reclaim(gfp_mask, order, ac);
4647 	if (unlikely(!(*did_some_progress)))
4648 		goto out;
4649 
4650 retry:
4651 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4652 
4653 	/*
4654 	 * If an allocation failed after direct reclaim, it could be because
4655 	 * pages are pinned on the per-cpu lists or in high alloc reserves.
4656 	 * Shrink them and try again
4657 	 */
4658 	if (!page && !drained) {
4659 		unreserve_highatomic_pageblock(ac, false);
4660 		drain_all_pages(NULL);
4661 		drained = true;
4662 		goto retry;
4663 	}
4664 out:
4665 	psi_memstall_leave(&pflags);
4666 
4667 	return page;
4668 }
4669 
4670 static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
4671 			     const struct alloc_context *ac)
4672 {
4673 	struct zoneref *z;
4674 	struct zone *zone;
4675 	pg_data_t *last_pgdat = NULL;
4676 	enum zone_type highest_zoneidx = ac->highest_zoneidx;
4677 
4678 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx,
4679 					ac->nodemask) {
4680 		if (last_pgdat != zone->zone_pgdat)
4681 			wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx);
4682 		last_pgdat = zone->zone_pgdat;
4683 	}
4684 }
4685 
4686 static inline unsigned int
4687 gfp_to_alloc_flags(gfp_t gfp_mask)
4688 {
4689 	unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
4690 
4691 	/*
4692 	 * __GFP_HIGH is assumed to be the same as ALLOC_HIGH
4693 	 * and __GFP_KSWAPD_RECLAIM is assumed to be the same as ALLOC_KSWAPD
4694 	 * to save two branches.
4695 	 */
4696 	BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
4697 	BUILD_BUG_ON(__GFP_KSWAPD_RECLAIM != (__force gfp_t) ALLOC_KSWAPD);
4698 
4699 	/*
4700 	 * The caller may dip into page reserves a bit more if the caller
4701 	 * cannot run direct reclaim, or if the caller has realtime scheduling
4702 	 * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
4703 	 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
4704 	 */
4705 	alloc_flags |= (__force int)
4706 		(gfp_mask & (__GFP_HIGH | __GFP_KSWAPD_RECLAIM));
4707 
4708 	if (gfp_mask & __GFP_ATOMIC) {
4709 		/*
4710 		 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
4711 		 * if it can't schedule.
4712 		 */
4713 		if (!(gfp_mask & __GFP_NOMEMALLOC))
4714 			alloc_flags |= ALLOC_HARDER;
4715 		/*
4716 		 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
4717 		 * comment for __cpuset_node_allowed().
4718 		 */
4719 		alloc_flags &= ~ALLOC_CPUSET;
4720 	} else if (unlikely(rt_task(current)) && in_task())
4721 		alloc_flags |= ALLOC_HARDER;
4722 
4723 	alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, alloc_flags);
4724 
4725 	return alloc_flags;
4726 }
4727 
4728 static bool oom_reserves_allowed(struct task_struct *tsk)
4729 {
4730 	if (!tsk_is_oom_victim(tsk))
4731 		return false;
4732 
4733 	/*
4734 	 * !MMU doesn't have oom reaper so give access to memory reserves
4735 	 * only to the thread with TIF_MEMDIE set
4736 	 */
4737 	if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
4738 		return false;
4739 
4740 	return true;
4741 }
4742 
4743 /*
4744  * Distinguish requests which really need access to full memory
4745  * reserves from oom victims which can live with a portion of it
4746  */
4747 static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
4748 {
4749 	if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
4750 		return 0;
4751 	if (gfp_mask & __GFP_MEMALLOC)
4752 		return ALLOC_NO_WATERMARKS;
4753 	if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
4754 		return ALLOC_NO_WATERMARKS;
4755 	if (!in_interrupt()) {
4756 		if (current->flags & PF_MEMALLOC)
4757 			return ALLOC_NO_WATERMARKS;
4758 		else if (oom_reserves_allowed(current))
4759 			return ALLOC_OOM;
4760 	}
4761 
4762 	return 0;
4763 }
4764 
4765 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
4766 {
4767 	return !!__gfp_pfmemalloc_flags(gfp_mask);
4768 }
4769 
4770 /*
4771  * Checks whether it makes sense to retry the reclaim to make a forward progress
4772  * for the given allocation request.
4773  *
4774  * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
4775  * without success, or when we couldn't even meet the watermark if we
4776  * reclaimed all remaining pages on the LRU lists.
4777  *
4778  * Returns true if a retry is viable or false to enter the oom path.
4779  */
4780 static inline bool
4781 should_reclaim_retry(gfp_t gfp_mask, unsigned order,
4782 		     struct alloc_context *ac, int alloc_flags,
4783 		     bool did_some_progress, int *no_progress_loops)
4784 {
4785 	struct zone *zone;
4786 	struct zoneref *z;
4787 	bool ret = false;
4788 
4789 	/*
4790 	 * Costly allocations might have made a progress but this doesn't mean
4791 	 * their order will become available due to high fragmentation so
4792 	 * always increment the no progress counter for them
4793 	 */
4794 	if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
4795 		*no_progress_loops = 0;
4796 	else
4797 		(*no_progress_loops)++;
4798 
4799 	/*
4800 	 * Make sure we converge to OOM if we cannot make any progress
4801 	 * several times in the row.
4802 	 */
4803 	if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
4804 		/* Before OOM, exhaust highatomic_reserve */
4805 		return unreserve_highatomic_pageblock(ac, true);
4806 	}
4807 
4808 	/*
4809 	 * Keep reclaiming pages while there is a chance this will lead
4810 	 * somewhere.  If none of the target zones can satisfy our allocation
4811 	 * request even if all reclaimable pages are considered then we are
4812 	 * screwed and have to go OOM.
4813 	 */
4814 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
4815 				ac->highest_zoneidx, ac->nodemask) {
4816 		unsigned long available;
4817 		unsigned long reclaimable;
4818 		unsigned long min_wmark = min_wmark_pages(zone);
4819 		bool wmark;
4820 
4821 		available = reclaimable = zone_reclaimable_pages(zone);
4822 		available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
4823 
4824 		/*
4825 		 * Would the allocation succeed if we reclaimed all
4826 		 * reclaimable pages?
4827 		 */
4828 		wmark = __zone_watermark_ok(zone, order, min_wmark,
4829 				ac->highest_zoneidx, alloc_flags, available);
4830 		trace_reclaim_retry_zone(z, order, reclaimable,
4831 				available, min_wmark, *no_progress_loops, wmark);
4832 		if (wmark) {
4833 			ret = true;
4834 			break;
4835 		}
4836 	}
4837 
4838 	/*
4839 	 * Memory allocation/reclaim might be called from a WQ context and the
4840 	 * current implementation of the WQ concurrency control doesn't
4841 	 * recognize that a particular WQ is congested if the worker thread is
4842 	 * looping without ever sleeping. Therefore we have to do a short sleep
4843 	 * here rather than calling cond_resched().
4844 	 */
4845 	if (current->flags & PF_WQ_WORKER)
4846 		schedule_timeout_uninterruptible(1);
4847 	else
4848 		cond_resched();
4849 	return ret;
4850 }
4851 
4852 static inline bool
4853 check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4854 {
4855 	/*
4856 	 * It's possible that cpuset's mems_allowed and the nodemask from
4857 	 * mempolicy don't intersect. This should be normally dealt with by
4858 	 * policy_nodemask(), but it's possible to race with cpuset update in
4859 	 * such a way the check therein was true, and then it became false
4860 	 * before we got our cpuset_mems_cookie here.
4861 	 * This assumes that for all allocations, ac->nodemask can come only
4862 	 * from MPOL_BIND mempolicy (whose documented semantics is to be ignored
4863 	 * when it does not intersect with the cpuset restrictions) or the
4864 	 * caller can deal with a violated nodemask.
4865 	 */
4866 	if (cpusets_enabled() && ac->nodemask &&
4867 			!cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4868 		ac->nodemask = NULL;
4869 		return true;
4870 	}
4871 
4872 	/*
4873 	 * When updating a task's mems_allowed or mempolicy nodemask, it is
4874 	 * possible to race with parallel threads in such a way that our
4875 	 * allocation can fail while the mask is being updated. If we are about
4876 	 * to fail, check if the cpuset changed during allocation and if so,
4877 	 * retry.
4878 	 */
4879 	if (read_mems_allowed_retry(cpuset_mems_cookie))
4880 		return true;
4881 
4882 	return false;
4883 }
4884 
4885 static inline struct page *
4886 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4887 						struct alloc_context *ac)
4888 {
4889 	bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4890 	const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4891 	struct page *page = NULL;
4892 	unsigned int alloc_flags;
4893 	unsigned long did_some_progress;
4894 	enum compact_priority compact_priority;
4895 	enum compact_result compact_result;
4896 	int compaction_retries;
4897 	int no_progress_loops;
4898 	unsigned int cpuset_mems_cookie;
4899 	int reserve_flags;
4900 
4901 	/*
4902 	 * We also sanity check to catch abuse of atomic reserves being used by
4903 	 * callers that are not in atomic context.
4904 	 */
4905 	if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4906 				(__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4907 		gfp_mask &= ~__GFP_ATOMIC;
4908 
4909 retry_cpuset:
4910 	compaction_retries = 0;
4911 	no_progress_loops = 0;
4912 	compact_priority = DEF_COMPACT_PRIORITY;
4913 	cpuset_mems_cookie = read_mems_allowed_begin();
4914 
4915 	/*
4916 	 * The fast path uses conservative alloc_flags to succeed only until
4917 	 * kswapd needs to be woken up, and to avoid the cost of setting up
4918 	 * alloc_flags precisely. So we do that now.
4919 	 */
4920 	alloc_flags = gfp_to_alloc_flags(gfp_mask);
4921 
4922 	/*
4923 	 * We need to recalculate the starting point for the zonelist iterator
4924 	 * because we might have used different nodemask in the fast path, or
4925 	 * there was a cpuset modification and we are retrying - otherwise we
4926 	 * could end up iterating over non-eligible zones endlessly.
4927 	 */
4928 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4929 					ac->highest_zoneidx, ac->nodemask);
4930 	if (!ac->preferred_zoneref->zone)
4931 		goto nopage;
4932 
4933 	/*
4934 	 * Check for insane configurations where the cpuset doesn't contain
4935 	 * any suitable zone to satisfy the request - e.g. non-movable
4936 	 * GFP_HIGHUSER allocations from MOVABLE nodes only.
4937 	 */
4938 	if (cpusets_insane_config() && (gfp_mask & __GFP_HARDWALL)) {
4939 		struct zoneref *z = first_zones_zonelist(ac->zonelist,
4940 					ac->highest_zoneidx,
4941 					&cpuset_current_mems_allowed);
4942 		if (!z->zone)
4943 			goto nopage;
4944 	}
4945 
4946 	if (alloc_flags & ALLOC_KSWAPD)
4947 		wake_all_kswapds(order, gfp_mask, ac);
4948 
4949 	/*
4950 	 * The adjusted alloc_flags might result in immediate success, so try
4951 	 * that first
4952 	 */
4953 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4954 	if (page)
4955 		goto got_pg;
4956 
4957 	/*
4958 	 * For costly allocations, try direct compaction first, as it's likely
4959 	 * that we have enough base pages and don't need to reclaim. For non-
4960 	 * movable high-order allocations, do that as well, as compaction will
4961 	 * try prevent permanent fragmentation by migrating from blocks of the
4962 	 * same migratetype.
4963 	 * Don't try this for allocations that are allowed to ignore
4964 	 * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen.
4965 	 */
4966 	if (can_direct_reclaim &&
4967 			(costly_order ||
4968 			   (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4969 			&& !gfp_pfmemalloc_allowed(gfp_mask)) {
4970 		page = __alloc_pages_direct_compact(gfp_mask, order,
4971 						alloc_flags, ac,
4972 						INIT_COMPACT_PRIORITY,
4973 						&compact_result);
4974 		if (page)
4975 			goto got_pg;
4976 
4977 		/*
4978 		 * Checks for costly allocations with __GFP_NORETRY, which
4979 		 * includes some THP page fault allocations
4980 		 */
4981 		if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4982 			/*
4983 			 * If allocating entire pageblock(s) and compaction
4984 			 * failed because all zones are below low watermarks
4985 			 * or is prohibited because it recently failed at this
4986 			 * order, fail immediately unless the allocator has
4987 			 * requested compaction and reclaim retry.
4988 			 *
4989 			 * Reclaim is
4990 			 *  - potentially very expensive because zones are far
4991 			 *    below their low watermarks or this is part of very
4992 			 *    bursty high order allocations,
4993 			 *  - not guaranteed to help because isolate_freepages()
4994 			 *    may not iterate over freed pages as part of its
4995 			 *    linear scan, and
4996 			 *  - unlikely to make entire pageblocks free on its
4997 			 *    own.
4998 			 */
4999 			if (compact_result == COMPACT_SKIPPED ||
5000 			    compact_result == COMPACT_DEFERRED)
5001 				goto nopage;
5002 
5003 			/*
5004 			 * Looks like reclaim/compaction is worth trying, but
5005 			 * sync compaction could be very expensive, so keep
5006 			 * using async compaction.
5007 			 */
5008 			compact_priority = INIT_COMPACT_PRIORITY;
5009 		}
5010 	}
5011 
5012 retry:
5013 	/* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
5014 	if (alloc_flags & ALLOC_KSWAPD)
5015 		wake_all_kswapds(order, gfp_mask, ac);
5016 
5017 	reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
5018 	if (reserve_flags)
5019 		alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, reserve_flags);
5020 
5021 	/*
5022 	 * Reset the nodemask and zonelist iterators if memory policies can be
5023 	 * ignored. These allocations are high priority and system rather than
5024 	 * user oriented.
5025 	 */
5026 	if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
5027 		ac->nodemask = NULL;
5028 		ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
5029 					ac->highest_zoneidx, ac->nodemask);
5030 	}
5031 
5032 	/* Attempt with potentially adjusted zonelist and alloc_flags */
5033 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
5034 	if (page)
5035 		goto got_pg;
5036 
5037 	/* Caller is not willing to reclaim, we can't balance anything */
5038 	if (!can_direct_reclaim)
5039 		goto nopage;
5040 
5041 	/* Avoid recursion of direct reclaim */
5042 	if (current->flags & PF_MEMALLOC)
5043 		goto nopage;
5044 
5045 	/* Try direct reclaim and then allocating */
5046 	page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
5047 							&did_some_progress);
5048 	if (page)
5049 		goto got_pg;
5050 
5051 	/* Try direct compaction and then allocating */
5052 	page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
5053 					compact_priority, &compact_result);
5054 	if (page)
5055 		goto got_pg;
5056 
5057 	/* Do not loop if specifically requested */
5058 	if (gfp_mask & __GFP_NORETRY)
5059 		goto nopage;
5060 
5061 	/*
5062 	 * Do not retry costly high order allocations unless they are
5063 	 * __GFP_RETRY_MAYFAIL
5064 	 */
5065 	if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
5066 		goto nopage;
5067 
5068 	if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
5069 				 did_some_progress > 0, &no_progress_loops))
5070 		goto retry;
5071 
5072 	/*
5073 	 * It doesn't make any sense to retry for the compaction if the order-0
5074 	 * reclaim is not able to make any progress because the current
5075 	 * implementation of the compaction depends on the sufficient amount
5076 	 * of free memory (see __compaction_suitable)
5077 	 */
5078 	if (did_some_progress > 0 &&
5079 			should_compact_retry(ac, order, alloc_flags,
5080 				compact_result, &compact_priority,
5081 				&compaction_retries))
5082 		goto retry;
5083 
5084 
5085 	/* Deal with possible cpuset update races before we start OOM killing */
5086 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
5087 		goto retry_cpuset;
5088 
5089 	/* Reclaim has failed us, start killing things */
5090 	page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
5091 	if (page)
5092 		goto got_pg;
5093 
5094 	/* Avoid allocations with no watermarks from looping endlessly */
5095 	if (tsk_is_oom_victim(current) &&
5096 	    (alloc_flags & ALLOC_OOM ||
5097 	     (gfp_mask & __GFP_NOMEMALLOC)))
5098 		goto nopage;
5099 
5100 	/* Retry as long as the OOM killer is making progress */
5101 	if (did_some_progress) {
5102 		no_progress_loops = 0;
5103 		goto retry;
5104 	}
5105 
5106 nopage:
5107 	/* Deal with possible cpuset update races before we fail */
5108 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
5109 		goto retry_cpuset;
5110 
5111 	/*
5112 	 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
5113 	 * we always retry
5114 	 */
5115 	if (gfp_mask & __GFP_NOFAIL) {
5116 		/*
5117 		 * All existing users of the __GFP_NOFAIL are blockable, so warn
5118 		 * of any new users that actually require GFP_NOWAIT
5119 		 */
5120 		if (WARN_ON_ONCE(!can_direct_reclaim))
5121 			goto fail;
5122 
5123 		/*
5124 		 * PF_MEMALLOC request from this context is rather bizarre
5125 		 * because we cannot reclaim anything and only can loop waiting
5126 		 * for somebody to do a work for us
5127 		 */
5128 		WARN_ON_ONCE(current->flags & PF_MEMALLOC);
5129 
5130 		/*
5131 		 * non failing costly orders are a hard requirement which we
5132 		 * are not prepared for much so let's warn about these users
5133 		 * so that we can identify them and convert them to something
5134 		 * else.
5135 		 */
5136 		WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
5137 
5138 		/*
5139 		 * Help non-failing allocations by giving them access to memory
5140 		 * reserves but do not use ALLOC_NO_WATERMARKS because this
5141 		 * could deplete whole memory reserves which would just make
5142 		 * the situation worse
5143 		 */
5144 		page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
5145 		if (page)
5146 			goto got_pg;
5147 
5148 		cond_resched();
5149 		goto retry;
5150 	}
5151 fail:
5152 	warn_alloc(gfp_mask, ac->nodemask,
5153 			"page allocation failure: order:%u", order);
5154 got_pg:
5155 	return page;
5156 }
5157 
5158 static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
5159 		int preferred_nid, nodemask_t *nodemask,
5160 		struct alloc_context *ac, gfp_t *alloc_gfp,
5161 		unsigned int *alloc_flags)
5162 {
5163 	ac->highest_zoneidx = gfp_zone(gfp_mask);
5164 	ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
5165 	ac->nodemask = nodemask;
5166 	ac->migratetype = gfp_migratetype(gfp_mask);
5167 
5168 	if (cpusets_enabled()) {
5169 		*alloc_gfp |= __GFP_HARDWALL;
5170 		/*
5171 		 * When we are in the interrupt context, it is irrelevant
5172 		 * to the current task context. It means that any node ok.
5173 		 */
5174 		if (in_task() && !ac->nodemask)
5175 			ac->nodemask = &cpuset_current_mems_allowed;
5176 		else
5177 			*alloc_flags |= ALLOC_CPUSET;
5178 	}
5179 
5180 	fs_reclaim_acquire(gfp_mask);
5181 	fs_reclaim_release(gfp_mask);
5182 
5183 	might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
5184 
5185 	if (should_fail_alloc_page(gfp_mask, order))
5186 		return false;
5187 
5188 	*alloc_flags = gfp_to_alloc_flags_cma(gfp_mask, *alloc_flags);
5189 
5190 	/* Dirty zone balancing only done in the fast path */
5191 	ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
5192 
5193 	/*
5194 	 * The preferred zone is used for statistics but crucially it is
5195 	 * also used as the starting point for the zonelist iterator. It
5196 	 * may get reset for allocations that ignore memory policies.
5197 	 */
5198 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
5199 					ac->highest_zoneidx, ac->nodemask);
5200 
5201 	return true;
5202 }
5203 
5204 /*
5205  * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array
5206  * @gfp: GFP flags for the allocation
5207  * @preferred_nid: The preferred NUMA node ID to allocate from
5208  * @nodemask: Set of nodes to allocate from, may be NULL
5209  * @nr_pages: The number of pages desired on the list or array
5210  * @page_list: Optional list to store the allocated pages
5211  * @page_array: Optional array to store the pages
5212  *
5213  * This is a batched version of the page allocator that attempts to
5214  * allocate nr_pages quickly. Pages are added to page_list if page_list
5215  * is not NULL, otherwise it is assumed that the page_array is valid.
5216  *
5217  * For lists, nr_pages is the number of pages that should be allocated.
5218  *
5219  * For arrays, only NULL elements are populated with pages and nr_pages
5220  * is the maximum number of pages that will be stored in the array.
5221  *
5222  * Returns the number of pages on the list or array.
5223  */
5224 unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
5225 			nodemask_t *nodemask, int nr_pages,
5226 			struct list_head *page_list,
5227 			struct page **page_array)
5228 {
5229 	struct page *page;
5230 	unsigned long flags;
5231 	struct zone *zone;
5232 	struct zoneref *z;
5233 	struct per_cpu_pages *pcp;
5234 	struct list_head *pcp_list;
5235 	struct alloc_context ac;
5236 	gfp_t alloc_gfp;
5237 	unsigned int alloc_flags = ALLOC_WMARK_LOW;
5238 	int nr_populated = 0, nr_account = 0;
5239 
5240 	/*
5241 	 * Skip populated array elements to determine if any pages need
5242 	 * to be allocated before disabling IRQs.
5243 	 */
5244 	while (page_array && nr_populated < nr_pages && page_array[nr_populated])
5245 		nr_populated++;
5246 
5247 	/* No pages requested? */
5248 	if (unlikely(nr_pages <= 0))
5249 		goto out;
5250 
5251 	/* Already populated array? */
5252 	if (unlikely(page_array && nr_pages - nr_populated == 0))
5253 		goto out;
5254 
5255 	/* Bulk allocator does not support memcg accounting. */
5256 	if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT))
5257 		goto failed;
5258 
5259 	/* Use the single page allocator for one page. */
5260 	if (nr_pages - nr_populated == 1)
5261 		goto failed;
5262 
5263 #ifdef CONFIG_PAGE_OWNER
5264 	/*
5265 	 * PAGE_OWNER may recurse into the allocator to allocate space to
5266 	 * save the stack with pagesets.lock held. Releasing/reacquiring
5267 	 * removes much of the performance benefit of bulk allocation so
5268 	 * force the caller to allocate one page at a time as it'll have
5269 	 * similar performance to added complexity to the bulk allocator.
5270 	 */
5271 	if (static_branch_unlikely(&page_owner_inited))
5272 		goto failed;
5273 #endif
5274 
5275 	/* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */
5276 	gfp &= gfp_allowed_mask;
5277 	alloc_gfp = gfp;
5278 	if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags))
5279 		goto out;
5280 	gfp = alloc_gfp;
5281 
5282 	/* Find an allowed local zone that meets the low watermark. */
5283 	for_each_zone_zonelist_nodemask(zone, z, ac.zonelist, ac.highest_zoneidx, ac.nodemask) {
5284 		unsigned long mark;
5285 
5286 		if (cpusets_enabled() && (alloc_flags & ALLOC_CPUSET) &&
5287 		    !__cpuset_zone_allowed(zone, gfp)) {
5288 			continue;
5289 		}
5290 
5291 		if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone &&
5292 		    zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) {
5293 			goto failed;
5294 		}
5295 
5296 		mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK) + nr_pages;
5297 		if (zone_watermark_fast(zone, 0,  mark,
5298 				zonelist_zone_idx(ac.preferred_zoneref),
5299 				alloc_flags, gfp)) {
5300 			break;
5301 		}
5302 	}
5303 
5304 	/*
5305 	 * If there are no allowed local zones that meets the watermarks then
5306 	 * try to allocate a single page and reclaim if necessary.
5307 	 */
5308 	if (unlikely(!zone))
5309 		goto failed;
5310 
5311 	/* Attempt the batch allocation */
5312 	local_lock_irqsave(&pagesets.lock, flags);
5313 	pcp = this_cpu_ptr(zone->per_cpu_pageset);
5314 	pcp_list = &pcp->lists[order_to_pindex(ac.migratetype, 0)];
5315 
5316 	while (nr_populated < nr_pages) {
5317 
5318 		/* Skip existing pages */
5319 		if (page_array && page_array[nr_populated]) {
5320 			nr_populated++;
5321 			continue;
5322 		}
5323 
5324 		page = __rmqueue_pcplist(zone, 0, ac.migratetype, alloc_flags,
5325 								pcp, pcp_list);
5326 		if (unlikely(!page)) {
5327 			/* Try and get at least one page */
5328 			if (!nr_populated)
5329 				goto failed_irq;
5330 			break;
5331 		}
5332 		nr_account++;
5333 
5334 		prep_new_page(page, 0, gfp, 0);
5335 		if (page_list)
5336 			list_add(&page->lru, page_list);
5337 		else
5338 			page_array[nr_populated] = page;
5339 		nr_populated++;
5340 	}
5341 
5342 	local_unlock_irqrestore(&pagesets.lock, flags);
5343 
5344 	__count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account);
5345 	zone_statistics(ac.preferred_zoneref->zone, zone, nr_account);
5346 
5347 out:
5348 	return nr_populated;
5349 
5350 failed_irq:
5351 	local_unlock_irqrestore(&pagesets.lock, flags);
5352 
5353 failed:
5354 	page = __alloc_pages(gfp, 0, preferred_nid, nodemask);
5355 	if (page) {
5356 		if (page_list)
5357 			list_add(&page->lru, page_list);
5358 		else
5359 			page_array[nr_populated] = page;
5360 		nr_populated++;
5361 	}
5362 
5363 	goto out;
5364 }
5365 EXPORT_SYMBOL_GPL(__alloc_pages_bulk);
5366 
5367 /*
5368  * This is the 'heart' of the zoned buddy allocator.
5369  */
5370 struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid,
5371 							nodemask_t *nodemask)
5372 {
5373 	struct page *page;
5374 	unsigned int alloc_flags = ALLOC_WMARK_LOW;
5375 	gfp_t alloc_gfp; /* The gfp_t that was actually used for allocation */
5376 	struct alloc_context ac = { };
5377 
5378 	/*
5379 	 * There are several places where we assume that the order value is sane
5380 	 * so bail out early if the request is out of bound.
5381 	 */
5382 	if (unlikely(order >= MAX_ORDER)) {
5383 		WARN_ON_ONCE(!(gfp & __GFP_NOWARN));
5384 		return NULL;
5385 	}
5386 
5387 	gfp &= gfp_allowed_mask;
5388 	/*
5389 	 * Apply scoped allocation constraints. This is mainly about GFP_NOFS
5390 	 * resp. GFP_NOIO which has to be inherited for all allocation requests
5391 	 * from a particular context which has been marked by
5392 	 * memalloc_no{fs,io}_{save,restore}. And PF_MEMALLOC_PIN which ensures
5393 	 * movable zones are not used during allocation.
5394 	 */
5395 	gfp = current_gfp_context(gfp);
5396 	alloc_gfp = gfp;
5397 	if (!prepare_alloc_pages(gfp, order, preferred_nid, nodemask, &ac,
5398 			&alloc_gfp, &alloc_flags))
5399 		return NULL;
5400 
5401 	/*
5402 	 * Forbid the first pass from falling back to types that fragment
5403 	 * memory until all local zones are considered.
5404 	 */
5405 	alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp);
5406 
5407 	/* First allocation attempt */
5408 	page = get_page_from_freelist(alloc_gfp, order, alloc_flags, &ac);
5409 	if (likely(page))
5410 		goto out;
5411 
5412 	alloc_gfp = gfp;
5413 	ac.spread_dirty_pages = false;
5414 
5415 	/*
5416 	 * Restore the original nodemask if it was potentially replaced with
5417 	 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
5418 	 */
5419 	ac.nodemask = nodemask;
5420 
5421 	page = __alloc_pages_slowpath(alloc_gfp, order, &ac);
5422 
5423 out:
5424 	if (memcg_kmem_enabled() && (gfp & __GFP_ACCOUNT) && page &&
5425 	    unlikely(__memcg_kmem_charge_page(page, gfp, order) != 0)) {
5426 		__free_pages(page, order);
5427 		page = NULL;
5428 	}
5429 
5430 	trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype);
5431 
5432 	return page;
5433 }
5434 EXPORT_SYMBOL(__alloc_pages);
5435 
5436 struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid,
5437 		nodemask_t *nodemask)
5438 {
5439 	struct page *page = __alloc_pages(gfp | __GFP_COMP, order,
5440 			preferred_nid, nodemask);
5441 
5442 	if (page && order > 1)
5443 		prep_transhuge_page(page);
5444 	return (struct folio *)page;
5445 }
5446 EXPORT_SYMBOL(__folio_alloc);
5447 
5448 /*
5449  * Common helper functions. Never use with __GFP_HIGHMEM because the returned
5450  * address cannot represent highmem pages. Use alloc_pages and then kmap if
5451  * you need to access high mem.
5452  */
5453 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
5454 {
5455 	struct page *page;
5456 
5457 	page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order);
5458 	if (!page)
5459 		return 0;
5460 	return (unsigned long) page_address(page);
5461 }
5462 EXPORT_SYMBOL(__get_free_pages);
5463 
5464 unsigned long get_zeroed_page(gfp_t gfp_mask)
5465 {
5466 	return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
5467 }
5468 EXPORT_SYMBOL(get_zeroed_page);
5469 
5470 /**
5471  * __free_pages - Free pages allocated with alloc_pages().
5472  * @page: The page pointer returned from alloc_pages().
5473  * @order: The order of the allocation.
5474  *
5475  * This function can free multi-page allocations that are not compound
5476  * pages.  It does not check that the @order passed in matches that of
5477  * the allocation, so it is easy to leak memory.  Freeing more memory
5478  * than was allocated will probably emit a warning.
5479  *
5480  * If the last reference to this page is speculative, it will be released
5481  * by put_page() which only frees the first page of a non-compound
5482  * allocation.  To prevent the remaining pages from being leaked, we free
5483  * the subsequent pages here.  If you want to use the page's reference
5484  * count to decide when to free the allocation, you should allocate a
5485  * compound page, and use put_page() instead of __free_pages().
5486  *
5487  * Context: May be called in interrupt context or while holding a normal
5488  * spinlock, but not in NMI context or while holding a raw spinlock.
5489  */
5490 void __free_pages(struct page *page, unsigned int order)
5491 {
5492 	if (put_page_testzero(page))
5493 		free_the_page(page, order);
5494 	else if (!PageHead(page))
5495 		while (order-- > 0)
5496 			free_the_page(page + (1 << order), order);
5497 }
5498 EXPORT_SYMBOL(__free_pages);
5499 
5500 void free_pages(unsigned long addr, unsigned int order)
5501 {
5502 	if (addr != 0) {
5503 		VM_BUG_ON(!virt_addr_valid((void *)addr));
5504 		__free_pages(virt_to_page((void *)addr), order);
5505 	}
5506 }
5507 
5508 EXPORT_SYMBOL(free_pages);
5509 
5510 /*
5511  * Page Fragment:
5512  *  An arbitrary-length arbitrary-offset area of memory which resides
5513  *  within a 0 or higher order page.  Multiple fragments within that page
5514  *  are individually refcounted, in the page's reference counter.
5515  *
5516  * The page_frag functions below provide a simple allocation framework for
5517  * page fragments.  This is used by the network stack and network device
5518  * drivers to provide a backing region of memory for use as either an
5519  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
5520  */
5521 static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
5522 					     gfp_t gfp_mask)
5523 {
5524 	struct page *page = NULL;
5525 	gfp_t gfp = gfp_mask;
5526 
5527 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5528 	gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
5529 		    __GFP_NOMEMALLOC;
5530 	page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
5531 				PAGE_FRAG_CACHE_MAX_ORDER);
5532 	nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
5533 #endif
5534 	if (unlikely(!page))
5535 		page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
5536 
5537 	nc->va = page ? page_address(page) : NULL;
5538 
5539 	return page;
5540 }
5541 
5542 void __page_frag_cache_drain(struct page *page, unsigned int count)
5543 {
5544 	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
5545 
5546 	if (page_ref_sub_and_test(page, count))
5547 		free_the_page(page, compound_order(page));
5548 }
5549 EXPORT_SYMBOL(__page_frag_cache_drain);
5550 
5551 void *page_frag_alloc_align(struct page_frag_cache *nc,
5552 		      unsigned int fragsz, gfp_t gfp_mask,
5553 		      unsigned int align_mask)
5554 {
5555 	unsigned int size = PAGE_SIZE;
5556 	struct page *page;
5557 	int offset;
5558 
5559 	if (unlikely(!nc->va)) {
5560 refill:
5561 		page = __page_frag_cache_refill(nc, gfp_mask);
5562 		if (!page)
5563 			return NULL;
5564 
5565 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5566 		/* if size can vary use size else just use PAGE_SIZE */
5567 		size = nc->size;
5568 #endif
5569 		/* Even if we own the page, we do not use atomic_set().
5570 		 * This would break get_page_unless_zero() users.
5571 		 */
5572 		page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
5573 
5574 		/* reset page count bias and offset to start of new frag */
5575 		nc->pfmemalloc = page_is_pfmemalloc(page);
5576 		nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5577 		nc->offset = size;
5578 	}
5579 
5580 	offset = nc->offset - fragsz;
5581 	if (unlikely(offset < 0)) {
5582 		page = virt_to_page(nc->va);
5583 
5584 		if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
5585 			goto refill;
5586 
5587 		if (unlikely(nc->pfmemalloc)) {
5588 			free_the_page(page, compound_order(page));
5589 			goto refill;
5590 		}
5591 
5592 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
5593 		/* if size can vary use size else just use PAGE_SIZE */
5594 		size = nc->size;
5595 #endif
5596 		/* OK, page count is 0, we can safely set it */
5597 		set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1);
5598 
5599 		/* reset page count bias and offset to start of new frag */
5600 		nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
5601 		offset = size - fragsz;
5602 	}
5603 
5604 	nc->pagecnt_bias--;
5605 	offset &= align_mask;
5606 	nc->offset = offset;
5607 
5608 	return nc->va + offset;
5609 }
5610 EXPORT_SYMBOL(page_frag_alloc_align);
5611 
5612 /*
5613  * Frees a page fragment allocated out of either a compound or order 0 page.
5614  */
5615 void page_frag_free(void *addr)
5616 {
5617 	struct page *page = virt_to_head_page(addr);
5618 
5619 	if (unlikely(put_page_testzero(page)))
5620 		free_the_page(page, compound_order(page));
5621 }
5622 EXPORT_SYMBOL(page_frag_free);
5623 
5624 static void *make_alloc_exact(unsigned long addr, unsigned int order,
5625 		size_t size)
5626 {
5627 	if (addr) {
5628 		unsigned long alloc_end = addr + (PAGE_SIZE << order);
5629 		unsigned long used = addr + PAGE_ALIGN(size);
5630 
5631 		split_page(virt_to_page((void *)addr), order);
5632 		while (used < alloc_end) {
5633 			free_page(used);
5634 			used += PAGE_SIZE;
5635 		}
5636 	}
5637 	return (void *)addr;
5638 }
5639 
5640 /**
5641  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
5642  * @size: the number of bytes to allocate
5643  * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP
5644  *
5645  * This function is similar to alloc_pages(), except that it allocates the
5646  * minimum number of pages to satisfy the request.  alloc_pages() can only
5647  * allocate memory in power-of-two pages.
5648  *
5649  * This function is also limited by MAX_ORDER.
5650  *
5651  * Memory allocated by this function must be released by free_pages_exact().
5652  *
5653  * Return: pointer to the allocated area or %NULL in case of error.
5654  */
5655 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
5656 {
5657 	unsigned int order = get_order(size);
5658 	unsigned long addr;
5659 
5660 	if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM)))
5661 		gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM);
5662 
5663 	addr = __get_free_pages(gfp_mask, order);
5664 	return make_alloc_exact(addr, order, size);
5665 }
5666 EXPORT_SYMBOL(alloc_pages_exact);
5667 
5668 /**
5669  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
5670  *			   pages on a node.
5671  * @nid: the preferred node ID where memory should be allocated
5672  * @size: the number of bytes to allocate
5673  * @gfp_mask: GFP flags for the allocation, must not contain __GFP_COMP
5674  *
5675  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
5676  * back.
5677  *
5678  * Return: pointer to the allocated area or %NULL in case of error.
5679  */
5680 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
5681 {
5682 	unsigned int order = get_order(size);
5683 	struct page *p;
5684 
5685 	if (WARN_ON_ONCE(gfp_mask & (__GFP_COMP | __GFP_HIGHMEM)))
5686 		gfp_mask &= ~(__GFP_COMP | __GFP_HIGHMEM);
5687 
5688 	p = alloc_pages_node(nid, gfp_mask, order);
5689 	if (!p)
5690 		return NULL;
5691 	return make_alloc_exact((unsigned long)page_address(p), order, size);
5692 }
5693 
5694 /**
5695  * free_pages_exact - release memory allocated via alloc_pages_exact()
5696  * @virt: the value returned by alloc_pages_exact.
5697  * @size: size of allocation, same value as passed to alloc_pages_exact().
5698  *
5699  * Release the memory allocated by a previous call to alloc_pages_exact.
5700  */
5701 void free_pages_exact(void *virt, size_t size)
5702 {
5703 	unsigned long addr = (unsigned long)virt;
5704 	unsigned long end = addr + PAGE_ALIGN(size);
5705 
5706 	while (addr < end) {
5707 		free_page(addr);
5708 		addr += PAGE_SIZE;
5709 	}
5710 }
5711 EXPORT_SYMBOL(free_pages_exact);
5712 
5713 /**
5714  * nr_free_zone_pages - count number of pages beyond high watermark
5715  * @offset: The zone index of the highest zone
5716  *
5717  * nr_free_zone_pages() counts the number of pages which are beyond the
5718  * high watermark within all zones at or below a given zone index.  For each
5719  * zone, the number of pages is calculated as:
5720  *
5721  *     nr_free_zone_pages = managed_pages - high_pages
5722  *
5723  * Return: number of pages beyond high watermark.
5724  */
5725 static unsigned long nr_free_zone_pages(int offset)
5726 {
5727 	struct zoneref *z;
5728 	struct zone *zone;
5729 
5730 	/* Just pick one node, since fallback list is circular */
5731 	unsigned long sum = 0;
5732 
5733 	struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
5734 
5735 	for_each_zone_zonelist(zone, z, zonelist, offset) {
5736 		unsigned long size = zone_managed_pages(zone);
5737 		unsigned long high = high_wmark_pages(zone);
5738 		if (size > high)
5739 			sum += size - high;
5740 	}
5741 
5742 	return sum;
5743 }
5744 
5745 /**
5746  * nr_free_buffer_pages - count number of pages beyond high watermark
5747  *
5748  * nr_free_buffer_pages() counts the number of pages which are beyond the high
5749  * watermark within ZONE_DMA and ZONE_NORMAL.
5750  *
5751  * Return: number of pages beyond high watermark within ZONE_DMA and
5752  * ZONE_NORMAL.
5753  */
5754 unsigned long nr_free_buffer_pages(void)
5755 {
5756 	return nr_free_zone_pages(gfp_zone(GFP_USER));
5757 }
5758 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
5759 
5760 static inline void show_node(struct zone *zone)
5761 {
5762 	if (IS_ENABLED(CONFIG_NUMA))
5763 		printk("Node %d ", zone_to_nid(zone));
5764 }
5765 
5766 long si_mem_available(void)
5767 {
5768 	long available;
5769 	unsigned long pagecache;
5770 	unsigned long wmark_low = 0;
5771 	unsigned long pages[NR_LRU_LISTS];
5772 	unsigned long reclaimable;
5773 	struct zone *zone;
5774 	int lru;
5775 
5776 	for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
5777 		pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
5778 
5779 	for_each_zone(zone)
5780 		wmark_low += low_wmark_pages(zone);
5781 
5782 	/*
5783 	 * Estimate the amount of memory available for userspace allocations,
5784 	 * without causing swapping.
5785 	 */
5786 	available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
5787 
5788 	/*
5789 	 * Not all the page cache can be freed, otherwise the system will
5790 	 * start swapping. Assume at least half of the page cache, or the
5791 	 * low watermark worth of cache, needs to stay.
5792 	 */
5793 	pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
5794 	pagecache -= min(pagecache / 2, wmark_low);
5795 	available += pagecache;
5796 
5797 	/*
5798 	 * Part of the reclaimable slab and other kernel memory consists of
5799 	 * items that are in use, and cannot be freed. Cap this estimate at the
5800 	 * low watermark.
5801 	 */
5802 	reclaimable = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B) +
5803 		global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
5804 	available += reclaimable - min(reclaimable / 2, wmark_low);
5805 
5806 	if (available < 0)
5807 		available = 0;
5808 	return available;
5809 }
5810 EXPORT_SYMBOL_GPL(si_mem_available);
5811 
5812 void si_meminfo(struct sysinfo *val)
5813 {
5814 	val->totalram = totalram_pages();
5815 	val->sharedram = global_node_page_state(NR_SHMEM);
5816 	val->freeram = global_zone_page_state(NR_FREE_PAGES);
5817 	val->bufferram = nr_blockdev_pages();
5818 	val->totalhigh = totalhigh_pages();
5819 	val->freehigh = nr_free_highpages();
5820 	val->mem_unit = PAGE_SIZE;
5821 }
5822 
5823 EXPORT_SYMBOL(si_meminfo);
5824 
5825 #ifdef CONFIG_NUMA
5826 void si_meminfo_node(struct sysinfo *val, int nid)
5827 {
5828 	int zone_type;		/* needs to be signed */
5829 	unsigned long managed_pages = 0;
5830 	unsigned long managed_highpages = 0;
5831 	unsigned long free_highpages = 0;
5832 	pg_data_t *pgdat = NODE_DATA(nid);
5833 
5834 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
5835 		managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]);
5836 	val->totalram = managed_pages;
5837 	val->sharedram = node_page_state(pgdat, NR_SHMEM);
5838 	val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
5839 #ifdef CONFIG_HIGHMEM
5840 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
5841 		struct zone *zone = &pgdat->node_zones[zone_type];
5842 
5843 		if (is_highmem(zone)) {
5844 			managed_highpages += zone_managed_pages(zone);
5845 			free_highpages += zone_page_state(zone, NR_FREE_PAGES);
5846 		}
5847 	}
5848 	val->totalhigh = managed_highpages;
5849 	val->freehigh = free_highpages;
5850 #else
5851 	val->totalhigh = managed_highpages;
5852 	val->freehigh = free_highpages;
5853 #endif
5854 	val->mem_unit = PAGE_SIZE;
5855 }
5856 #endif
5857 
5858 /*
5859  * Determine whether the node should be displayed or not, depending on whether
5860  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
5861  */
5862 static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
5863 {
5864 	if (!(flags & SHOW_MEM_FILTER_NODES))
5865 		return false;
5866 
5867 	/*
5868 	 * no node mask - aka implicit memory numa policy. Do not bother with
5869 	 * the synchronization - read_mems_allowed_begin - because we do not
5870 	 * have to be precise here.
5871 	 */
5872 	if (!nodemask)
5873 		nodemask = &cpuset_current_mems_allowed;
5874 
5875 	return !node_isset(nid, *nodemask);
5876 }
5877 
5878 #define K(x) ((x) << (PAGE_SHIFT-10))
5879 
5880 static void show_migration_types(unsigned char type)
5881 {
5882 	static const char types[MIGRATE_TYPES] = {
5883 		[MIGRATE_UNMOVABLE]	= 'U',
5884 		[MIGRATE_MOVABLE]	= 'M',
5885 		[MIGRATE_RECLAIMABLE]	= 'E',
5886 		[MIGRATE_HIGHATOMIC]	= 'H',
5887 #ifdef CONFIG_CMA
5888 		[MIGRATE_CMA]		= 'C',
5889 #endif
5890 #ifdef CONFIG_MEMORY_ISOLATION
5891 		[MIGRATE_ISOLATE]	= 'I',
5892 #endif
5893 	};
5894 	char tmp[MIGRATE_TYPES + 1];
5895 	char *p = tmp;
5896 	int i;
5897 
5898 	for (i = 0; i < MIGRATE_TYPES; i++) {
5899 		if (type & (1 << i))
5900 			*p++ = types[i];
5901 	}
5902 
5903 	*p = '\0';
5904 	printk(KERN_CONT "(%s) ", tmp);
5905 }
5906 
5907 /*
5908  * Show free area list (used inside shift_scroll-lock stuff)
5909  * We also calculate the percentage fragmentation. We do this by counting the
5910  * memory on each free list with the exception of the first item on the list.
5911  *
5912  * Bits in @filter:
5913  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
5914  *   cpuset.
5915  */
5916 void show_free_areas(unsigned int filter, nodemask_t *nodemask)
5917 {
5918 	unsigned long free_pcp = 0;
5919 	int cpu;
5920 	struct zone *zone;
5921 	pg_data_t *pgdat;
5922 
5923 	for_each_populated_zone(zone) {
5924 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5925 			continue;
5926 
5927 		for_each_online_cpu(cpu)
5928 			free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count;
5929 	}
5930 
5931 	printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
5932 		" active_file:%lu inactive_file:%lu isolated_file:%lu\n"
5933 		" unevictable:%lu dirty:%lu writeback:%lu\n"
5934 		" slab_reclaimable:%lu slab_unreclaimable:%lu\n"
5935 		" mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
5936 		" kernel_misc_reclaimable:%lu\n"
5937 		" free:%lu free_pcp:%lu free_cma:%lu\n",
5938 		global_node_page_state(NR_ACTIVE_ANON),
5939 		global_node_page_state(NR_INACTIVE_ANON),
5940 		global_node_page_state(NR_ISOLATED_ANON),
5941 		global_node_page_state(NR_ACTIVE_FILE),
5942 		global_node_page_state(NR_INACTIVE_FILE),
5943 		global_node_page_state(NR_ISOLATED_FILE),
5944 		global_node_page_state(NR_UNEVICTABLE),
5945 		global_node_page_state(NR_FILE_DIRTY),
5946 		global_node_page_state(NR_WRITEBACK),
5947 		global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B),
5948 		global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B),
5949 		global_node_page_state(NR_FILE_MAPPED),
5950 		global_node_page_state(NR_SHMEM),
5951 		global_node_page_state(NR_PAGETABLE),
5952 		global_zone_page_state(NR_BOUNCE),
5953 		global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE),
5954 		global_zone_page_state(NR_FREE_PAGES),
5955 		free_pcp,
5956 		global_zone_page_state(NR_FREE_CMA_PAGES));
5957 
5958 	for_each_online_pgdat(pgdat) {
5959 		if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
5960 			continue;
5961 
5962 		printk("Node %d"
5963 			" active_anon:%lukB"
5964 			" inactive_anon:%lukB"
5965 			" active_file:%lukB"
5966 			" inactive_file:%lukB"
5967 			" unevictable:%lukB"
5968 			" isolated(anon):%lukB"
5969 			" isolated(file):%lukB"
5970 			" mapped:%lukB"
5971 			" dirty:%lukB"
5972 			" writeback:%lukB"
5973 			" shmem:%lukB"
5974 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5975 			" shmem_thp: %lukB"
5976 			" shmem_pmdmapped: %lukB"
5977 			" anon_thp: %lukB"
5978 #endif
5979 			" writeback_tmp:%lukB"
5980 			" kernel_stack:%lukB"
5981 #ifdef CONFIG_SHADOW_CALL_STACK
5982 			" shadow_call_stack:%lukB"
5983 #endif
5984 			" pagetables:%lukB"
5985 			" all_unreclaimable? %s"
5986 			"\n",
5987 			pgdat->node_id,
5988 			K(node_page_state(pgdat, NR_ACTIVE_ANON)),
5989 			K(node_page_state(pgdat, NR_INACTIVE_ANON)),
5990 			K(node_page_state(pgdat, NR_ACTIVE_FILE)),
5991 			K(node_page_state(pgdat, NR_INACTIVE_FILE)),
5992 			K(node_page_state(pgdat, NR_UNEVICTABLE)),
5993 			K(node_page_state(pgdat, NR_ISOLATED_ANON)),
5994 			K(node_page_state(pgdat, NR_ISOLATED_FILE)),
5995 			K(node_page_state(pgdat, NR_FILE_MAPPED)),
5996 			K(node_page_state(pgdat, NR_FILE_DIRTY)),
5997 			K(node_page_state(pgdat, NR_WRITEBACK)),
5998 			K(node_page_state(pgdat, NR_SHMEM)),
5999 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6000 			K(node_page_state(pgdat, NR_SHMEM_THPS)),
6001 			K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)),
6002 			K(node_page_state(pgdat, NR_ANON_THPS)),
6003 #endif
6004 			K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
6005 			node_page_state(pgdat, NR_KERNEL_STACK_KB),
6006 #ifdef CONFIG_SHADOW_CALL_STACK
6007 			node_page_state(pgdat, NR_KERNEL_SCS_KB),
6008 #endif
6009 			K(node_page_state(pgdat, NR_PAGETABLE)),
6010 			pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
6011 				"yes" : "no");
6012 	}
6013 
6014 	for_each_populated_zone(zone) {
6015 		int i;
6016 
6017 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
6018 			continue;
6019 
6020 		free_pcp = 0;
6021 		for_each_online_cpu(cpu)
6022 			free_pcp += per_cpu_ptr(zone->per_cpu_pageset, cpu)->count;
6023 
6024 		show_node(zone);
6025 		printk(KERN_CONT
6026 			"%s"
6027 			" free:%lukB"
6028 			" boost:%lukB"
6029 			" min:%lukB"
6030 			" low:%lukB"
6031 			" high:%lukB"
6032 			" reserved_highatomic:%luKB"
6033 			" active_anon:%lukB"
6034 			" inactive_anon:%lukB"
6035 			" active_file:%lukB"
6036 			" inactive_file:%lukB"
6037 			" unevictable:%lukB"
6038 			" writepending:%lukB"
6039 			" present:%lukB"
6040 			" managed:%lukB"
6041 			" mlocked:%lukB"
6042 			" bounce:%lukB"
6043 			" free_pcp:%lukB"
6044 			" local_pcp:%ukB"
6045 			" free_cma:%lukB"
6046 			"\n",
6047 			zone->name,
6048 			K(zone_page_state(zone, NR_FREE_PAGES)),
6049 			K(zone->watermark_boost),
6050 			K(min_wmark_pages(zone)),
6051 			K(low_wmark_pages(zone)),
6052 			K(high_wmark_pages(zone)),
6053 			K(zone->nr_reserved_highatomic),
6054 			K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
6055 			K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
6056 			K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
6057 			K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
6058 			K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
6059 			K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
6060 			K(zone->present_pages),
6061 			K(zone_managed_pages(zone)),
6062 			K(zone_page_state(zone, NR_MLOCK)),
6063 			K(zone_page_state(zone, NR_BOUNCE)),
6064 			K(free_pcp),
6065 			K(this_cpu_read(zone->per_cpu_pageset->count)),
6066 			K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
6067 		printk("lowmem_reserve[]:");
6068 		for (i = 0; i < MAX_NR_ZONES; i++)
6069 			printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
6070 		printk(KERN_CONT "\n");
6071 	}
6072 
6073 	for_each_populated_zone(zone) {
6074 		unsigned int order;
6075 		unsigned long nr[MAX_ORDER], flags, total = 0;
6076 		unsigned char types[MAX_ORDER];
6077 
6078 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
6079 			continue;
6080 		show_node(zone);
6081 		printk(KERN_CONT "%s: ", zone->name);
6082 
6083 		spin_lock_irqsave(&zone->lock, flags);
6084 		for (order = 0; order < MAX_ORDER; order++) {
6085 			struct free_area *area = &zone->free_area[order];
6086 			int type;
6087 
6088 			nr[order] = area->nr_free;
6089 			total += nr[order] << order;
6090 
6091 			types[order] = 0;
6092 			for (type = 0; type < MIGRATE_TYPES; type++) {
6093 				if (!free_area_empty(area, type))
6094 					types[order] |= 1 << type;
6095 			}
6096 		}
6097 		spin_unlock_irqrestore(&zone->lock, flags);
6098 		for (order = 0; order < MAX_ORDER; order++) {
6099 			printk(KERN_CONT "%lu*%lukB ",
6100 			       nr[order], K(1UL) << order);
6101 			if (nr[order])
6102 				show_migration_types(types[order]);
6103 		}
6104 		printk(KERN_CONT "= %lukB\n", K(total));
6105 	}
6106 
6107 	hugetlb_show_meminfo();
6108 
6109 	printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
6110 
6111 	show_swap_cache_info();
6112 }
6113 
6114 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
6115 {
6116 	zoneref->zone = zone;
6117 	zoneref->zone_idx = zone_idx(zone);
6118 }
6119 
6120 /*
6121  * Builds allocation fallback zone lists.
6122  *
6123  * Add all populated zones of a node to the zonelist.
6124  */
6125 static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
6126 {
6127 	struct zone *zone;
6128 	enum zone_type zone_type = MAX_NR_ZONES;
6129 	int nr_zones = 0;
6130 
6131 	do {
6132 		zone_type--;
6133 		zone = pgdat->node_zones + zone_type;
6134 		if (populated_zone(zone)) {
6135 			zoneref_set_zone(zone, &zonerefs[nr_zones++]);
6136 			check_highest_zone(zone_type);
6137 		}
6138 	} while (zone_type);
6139 
6140 	return nr_zones;
6141 }
6142 
6143 #ifdef CONFIG_NUMA
6144 
6145 static int __parse_numa_zonelist_order(char *s)
6146 {
6147 	/*
6148 	 * We used to support different zonelists modes but they turned
6149 	 * out to be just not useful. Let's keep the warning in place
6150 	 * if somebody still use the cmd line parameter so that we do
6151 	 * not fail it silently
6152 	 */
6153 	if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
6154 		pr_warn("Ignoring unsupported numa_zonelist_order value:  %s\n", s);
6155 		return -EINVAL;
6156 	}
6157 	return 0;
6158 }
6159 
6160 char numa_zonelist_order[] = "Node";
6161 
6162 /*
6163  * sysctl handler for numa_zonelist_order
6164  */
6165 int numa_zonelist_order_handler(struct ctl_table *table, int write,
6166 		void *buffer, size_t *length, loff_t *ppos)
6167 {
6168 	if (write)
6169 		return __parse_numa_zonelist_order(buffer);
6170 	return proc_dostring(table, write, buffer, length, ppos);
6171 }
6172 
6173 
6174 #define MAX_NODE_LOAD (nr_online_nodes)
6175 static int node_load[MAX_NUMNODES];
6176 
6177 /**
6178  * find_next_best_node - find the next node that should appear in a given node's fallback list
6179  * @node: node whose fallback list we're appending
6180  * @used_node_mask: nodemask_t of already used nodes
6181  *
6182  * We use a number of factors to determine which is the next node that should
6183  * appear on a given node's fallback list.  The node should not have appeared
6184  * already in @node's fallback list, and it should be the next closest node
6185  * according to the distance array (which contains arbitrary distance values
6186  * from each node to each node in the system), and should also prefer nodes
6187  * with no CPUs, since presumably they'll have very little allocation pressure
6188  * on them otherwise.
6189  *
6190  * Return: node id of the found node or %NUMA_NO_NODE if no node is found.
6191  */
6192 int find_next_best_node(int node, nodemask_t *used_node_mask)
6193 {
6194 	int n, val;
6195 	int min_val = INT_MAX;
6196 	int best_node = NUMA_NO_NODE;
6197 
6198 	/* Use the local node if we haven't already */
6199 	if (!node_isset(node, *used_node_mask)) {
6200 		node_set(node, *used_node_mask);
6201 		return node;
6202 	}
6203 
6204 	for_each_node_state(n, N_MEMORY) {
6205 
6206 		/* Don't want a node to appear more than once */
6207 		if (node_isset(n, *used_node_mask))
6208 			continue;
6209 
6210 		/* Use the distance array to find the distance */
6211 		val = node_distance(node, n);
6212 
6213 		/* Penalize nodes under us ("prefer the next node") */
6214 		val += (n < node);
6215 
6216 		/* Give preference to headless and unused nodes */
6217 		if (!cpumask_empty(cpumask_of_node(n)))
6218 			val += PENALTY_FOR_NODE_WITH_CPUS;
6219 
6220 		/* Slight preference for less loaded node */
6221 		val *= (MAX_NODE_LOAD*MAX_NUMNODES);
6222 		val += node_load[n];
6223 
6224 		if (val < min_val) {
6225 			min_val = val;
6226 			best_node = n;
6227 		}
6228 	}
6229 
6230 	if (best_node >= 0)
6231 		node_set(best_node, *used_node_mask);
6232 
6233 	return best_node;
6234 }
6235 
6236 
6237 /*
6238  * Build zonelists ordered by node and zones within node.
6239  * This results in maximum locality--normal zone overflows into local
6240  * DMA zone, if any--but risks exhausting DMA zone.
6241  */
6242 static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
6243 		unsigned nr_nodes)
6244 {
6245 	struct zoneref *zonerefs;
6246 	int i;
6247 
6248 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
6249 
6250 	for (i = 0; i < nr_nodes; i++) {
6251 		int nr_zones;
6252 
6253 		pg_data_t *node = NODE_DATA(node_order[i]);
6254 
6255 		nr_zones = build_zonerefs_node(node, zonerefs);
6256 		zonerefs += nr_zones;
6257 	}
6258 	zonerefs->zone = NULL;
6259 	zonerefs->zone_idx = 0;
6260 }
6261 
6262 /*
6263  * Build gfp_thisnode zonelists
6264  */
6265 static void build_thisnode_zonelists(pg_data_t *pgdat)
6266 {
6267 	struct zoneref *zonerefs;
6268 	int nr_zones;
6269 
6270 	zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
6271 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
6272 	zonerefs += nr_zones;
6273 	zonerefs->zone = NULL;
6274 	zonerefs->zone_idx = 0;
6275 }
6276 
6277 /*
6278  * Build zonelists ordered by zone and nodes within zones.
6279  * This results in conserving DMA zone[s] until all Normal memory is
6280  * exhausted, but results in overflowing to remote node while memory
6281  * may still exist in local DMA zone.
6282  */
6283 
6284 static void build_zonelists(pg_data_t *pgdat)
6285 {
6286 	static int node_order[MAX_NUMNODES];
6287 	int node, load, nr_nodes = 0;
6288 	nodemask_t used_mask = NODE_MASK_NONE;
6289 	int local_node, prev_node;
6290 
6291 	/* NUMA-aware ordering of nodes */
6292 	local_node = pgdat->node_id;
6293 	load = nr_online_nodes;
6294 	prev_node = local_node;
6295 
6296 	memset(node_order, 0, sizeof(node_order));
6297 	while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
6298 		/*
6299 		 * We don't want to pressure a particular node.
6300 		 * So adding penalty to the first node in same
6301 		 * distance group to make it round-robin.
6302 		 */
6303 		if (node_distance(local_node, node) !=
6304 		    node_distance(local_node, prev_node))
6305 			node_load[node] += load;
6306 
6307 		node_order[nr_nodes++] = node;
6308 		prev_node = node;
6309 		load--;
6310 	}
6311 
6312 	build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
6313 	build_thisnode_zonelists(pgdat);
6314 	pr_info("Fallback order for Node %d: ", local_node);
6315 	for (node = 0; node < nr_nodes; node++)
6316 		pr_cont("%d ", node_order[node]);
6317 	pr_cont("\n");
6318 }
6319 
6320 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
6321 /*
6322  * Return node id of node used for "local" allocations.
6323  * I.e., first node id of first zone in arg node's generic zonelist.
6324  * Used for initializing percpu 'numa_mem', which is used primarily
6325  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
6326  */
6327 int local_memory_node(int node)
6328 {
6329 	struct zoneref *z;
6330 
6331 	z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
6332 				   gfp_zone(GFP_KERNEL),
6333 				   NULL);
6334 	return zone_to_nid(z->zone);
6335 }
6336 #endif
6337 
6338 static void setup_min_unmapped_ratio(void);
6339 static void setup_min_slab_ratio(void);
6340 #else	/* CONFIG_NUMA */
6341 
6342 static void build_zonelists(pg_data_t *pgdat)
6343 {
6344 	int node, local_node;
6345 	struct zoneref *zonerefs;
6346 	int nr_zones;
6347 
6348 	local_node = pgdat->node_id;
6349 
6350 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
6351 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
6352 	zonerefs += nr_zones;
6353 
6354 	/*
6355 	 * Now we build the zonelist so that it contains the zones
6356 	 * of all the other nodes.
6357 	 * We don't want to pressure a particular node, so when
6358 	 * building the zones for node N, we make sure that the
6359 	 * zones coming right after the local ones are those from
6360 	 * node N+1 (modulo N)
6361 	 */
6362 	for (node = local_node + 1; node < MAX_NUMNODES; node++) {
6363 		if (!node_online(node))
6364 			continue;
6365 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
6366 		zonerefs += nr_zones;
6367 	}
6368 	for (node = 0; node < local_node; node++) {
6369 		if (!node_online(node))
6370 			continue;
6371 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
6372 		zonerefs += nr_zones;
6373 	}
6374 
6375 	zonerefs->zone = NULL;
6376 	zonerefs->zone_idx = 0;
6377 }
6378 
6379 #endif	/* CONFIG_NUMA */
6380 
6381 /*
6382  * Boot pageset table. One per cpu which is going to be used for all
6383  * zones and all nodes. The parameters will be set in such a way
6384  * that an item put on a list will immediately be handed over to
6385  * the buddy list. This is safe since pageset manipulation is done
6386  * with interrupts disabled.
6387  *
6388  * The boot_pagesets must be kept even after bootup is complete for
6389  * unused processors and/or zones. They do play a role for bootstrapping
6390  * hotplugged processors.
6391  *
6392  * zoneinfo_show() and maybe other functions do
6393  * not check if the processor is online before following the pageset pointer.
6394  * Other parts of the kernel may not check if the zone is available.
6395  */
6396 static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats);
6397 /* These effectively disable the pcplists in the boot pageset completely */
6398 #define BOOT_PAGESET_HIGH	0
6399 #define BOOT_PAGESET_BATCH	1
6400 static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset);
6401 static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats);
6402 DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
6403 
6404 static void __build_all_zonelists(void *data)
6405 {
6406 	int nid;
6407 	int __maybe_unused cpu;
6408 	pg_data_t *self = data;
6409 	static DEFINE_SPINLOCK(lock);
6410 
6411 	spin_lock(&lock);
6412 
6413 #ifdef CONFIG_NUMA
6414 	memset(node_load, 0, sizeof(node_load));
6415 #endif
6416 
6417 	/*
6418 	 * This node is hotadded and no memory is yet present.   So just
6419 	 * building zonelists is fine - no need to touch other nodes.
6420 	 */
6421 	if (self && !node_online(self->node_id)) {
6422 		build_zonelists(self);
6423 	} else {
6424 		/*
6425 		 * All possible nodes have pgdat preallocated
6426 		 * in free_area_init
6427 		 */
6428 		for_each_node(nid) {
6429 			pg_data_t *pgdat = NODE_DATA(nid);
6430 
6431 			build_zonelists(pgdat);
6432 		}
6433 
6434 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
6435 		/*
6436 		 * We now know the "local memory node" for each node--
6437 		 * i.e., the node of the first zone in the generic zonelist.
6438 		 * Set up numa_mem percpu variable for on-line cpus.  During
6439 		 * boot, only the boot cpu should be on-line;  we'll init the
6440 		 * secondary cpus' numa_mem as they come on-line.  During
6441 		 * node/memory hotplug, we'll fixup all on-line cpus.
6442 		 */
6443 		for_each_online_cpu(cpu)
6444 			set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
6445 #endif
6446 	}
6447 
6448 	spin_unlock(&lock);
6449 }
6450 
6451 static noinline void __init
6452 build_all_zonelists_init(void)
6453 {
6454 	int cpu;
6455 
6456 	__build_all_zonelists(NULL);
6457 
6458 	/*
6459 	 * Initialize the boot_pagesets that are going to be used
6460 	 * for bootstrapping processors. The real pagesets for
6461 	 * each zone will be allocated later when the per cpu
6462 	 * allocator is available.
6463 	 *
6464 	 * boot_pagesets are used also for bootstrapping offline
6465 	 * cpus if the system is already booted because the pagesets
6466 	 * are needed to initialize allocators on a specific cpu too.
6467 	 * F.e. the percpu allocator needs the page allocator which
6468 	 * needs the percpu allocator in order to allocate its pagesets
6469 	 * (a chicken-egg dilemma).
6470 	 */
6471 	for_each_possible_cpu(cpu)
6472 		per_cpu_pages_init(&per_cpu(boot_pageset, cpu), &per_cpu(boot_zonestats, cpu));
6473 
6474 	mminit_verify_zonelist();
6475 	cpuset_init_current_mems_allowed();
6476 }
6477 
6478 /*
6479  * unless system_state == SYSTEM_BOOTING.
6480  *
6481  * __ref due to call of __init annotated helper build_all_zonelists_init
6482  * [protected by SYSTEM_BOOTING].
6483  */
6484 void __ref build_all_zonelists(pg_data_t *pgdat)
6485 {
6486 	unsigned long vm_total_pages;
6487 
6488 	if (system_state == SYSTEM_BOOTING) {
6489 		build_all_zonelists_init();
6490 	} else {
6491 		__build_all_zonelists(pgdat);
6492 		/* cpuset refresh routine should be here */
6493 	}
6494 	/* Get the number of free pages beyond high watermark in all zones. */
6495 	vm_total_pages = nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
6496 	/*
6497 	 * Disable grouping by mobility if the number of pages in the
6498 	 * system is too low to allow the mechanism to work. It would be
6499 	 * more accurate, but expensive to check per-zone. This check is
6500 	 * made on memory-hotadd so a system can start with mobility
6501 	 * disabled and enable it later
6502 	 */
6503 	if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
6504 		page_group_by_mobility_disabled = 1;
6505 	else
6506 		page_group_by_mobility_disabled = 0;
6507 
6508 	pr_info("Built %u zonelists, mobility grouping %s.  Total pages: %ld\n",
6509 		nr_online_nodes,
6510 		page_group_by_mobility_disabled ? "off" : "on",
6511 		vm_total_pages);
6512 #ifdef CONFIG_NUMA
6513 	pr_info("Policy zone: %s\n", zone_names[policy_zone]);
6514 #endif
6515 }
6516 
6517 /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */
6518 static bool __meminit
6519 overlap_memmap_init(unsigned long zone, unsigned long *pfn)
6520 {
6521 	static struct memblock_region *r;
6522 
6523 	if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
6524 		if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
6525 			for_each_mem_region(r) {
6526 				if (*pfn < memblock_region_memory_end_pfn(r))
6527 					break;
6528 			}
6529 		}
6530 		if (*pfn >= memblock_region_memory_base_pfn(r) &&
6531 		    memblock_is_mirror(r)) {
6532 			*pfn = memblock_region_memory_end_pfn(r);
6533 			return true;
6534 		}
6535 	}
6536 	return false;
6537 }
6538 
6539 /*
6540  * Initially all pages are reserved - free ones are freed
6541  * up by memblock_free_all() once the early boot process is
6542  * done. Non-atomic initialization, single-pass.
6543  *
6544  * All aligned pageblocks are initialized to the specified migratetype
6545  * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
6546  * zone stats (e.g., nr_isolate_pageblock) are touched.
6547  */
6548 void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone,
6549 		unsigned long start_pfn, unsigned long zone_end_pfn,
6550 		enum meminit_context context,
6551 		struct vmem_altmap *altmap, int migratetype)
6552 {
6553 	unsigned long pfn, end_pfn = start_pfn + size;
6554 	struct page *page;
6555 
6556 	if (highest_memmap_pfn < end_pfn - 1)
6557 		highest_memmap_pfn = end_pfn - 1;
6558 
6559 #ifdef CONFIG_ZONE_DEVICE
6560 	/*
6561 	 * Honor reservation requested by the driver for this ZONE_DEVICE
6562 	 * memory. We limit the total number of pages to initialize to just
6563 	 * those that might contain the memory mapping. We will defer the
6564 	 * ZONE_DEVICE page initialization until after we have released
6565 	 * the hotplug lock.
6566 	 */
6567 	if (zone == ZONE_DEVICE) {
6568 		if (!altmap)
6569 			return;
6570 
6571 		if (start_pfn == altmap->base_pfn)
6572 			start_pfn += altmap->reserve;
6573 		end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6574 	}
6575 #endif
6576 
6577 	for (pfn = start_pfn; pfn < end_pfn; ) {
6578 		/*
6579 		 * There can be holes in boot-time mem_map[]s handed to this
6580 		 * function.  They do not exist on hotplugged memory.
6581 		 */
6582 		if (context == MEMINIT_EARLY) {
6583 			if (overlap_memmap_init(zone, &pfn))
6584 				continue;
6585 			if (defer_init(nid, pfn, zone_end_pfn))
6586 				break;
6587 		}
6588 
6589 		page = pfn_to_page(pfn);
6590 		__init_single_page(page, pfn, zone, nid);
6591 		if (context == MEMINIT_HOTPLUG)
6592 			__SetPageReserved(page);
6593 
6594 		/*
6595 		 * Usually, we want to mark the pageblock MIGRATE_MOVABLE,
6596 		 * such that unmovable allocations won't be scattered all
6597 		 * over the place during system boot.
6598 		 */
6599 		if (IS_ALIGNED(pfn, pageblock_nr_pages)) {
6600 			set_pageblock_migratetype(page, migratetype);
6601 			cond_resched();
6602 		}
6603 		pfn++;
6604 	}
6605 }
6606 
6607 #ifdef CONFIG_ZONE_DEVICE
6608 static void __ref __init_zone_device_page(struct page *page, unsigned long pfn,
6609 					  unsigned long zone_idx, int nid,
6610 					  struct dev_pagemap *pgmap)
6611 {
6612 
6613 	__init_single_page(page, pfn, zone_idx, nid);
6614 
6615 	/*
6616 	 * Mark page reserved as it will need to wait for onlining
6617 	 * phase for it to be fully associated with a zone.
6618 	 *
6619 	 * We can use the non-atomic __set_bit operation for setting
6620 	 * the flag as we are still initializing the pages.
6621 	 */
6622 	__SetPageReserved(page);
6623 
6624 	/*
6625 	 * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer
6626 	 * and zone_device_data.  It is a bug if a ZONE_DEVICE page is
6627 	 * ever freed or placed on a driver-private list.
6628 	 */
6629 	page->pgmap = pgmap;
6630 	page->zone_device_data = NULL;
6631 
6632 	/*
6633 	 * Mark the block movable so that blocks are reserved for
6634 	 * movable at startup. This will force kernel allocations
6635 	 * to reserve their blocks rather than leaking throughout
6636 	 * the address space during boot when many long-lived
6637 	 * kernel allocations are made.
6638 	 *
6639 	 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
6640 	 * because this is done early in section_activate()
6641 	 */
6642 	if (IS_ALIGNED(pfn, pageblock_nr_pages)) {
6643 		set_pageblock_migratetype(page, MIGRATE_MOVABLE);
6644 		cond_resched();
6645 	}
6646 }
6647 
6648 static void __ref memmap_init_compound(struct page *head,
6649 				       unsigned long head_pfn,
6650 				       unsigned long zone_idx, int nid,
6651 				       struct dev_pagemap *pgmap,
6652 				       unsigned long nr_pages)
6653 {
6654 	unsigned long pfn, end_pfn = head_pfn + nr_pages;
6655 	unsigned int order = pgmap->vmemmap_shift;
6656 
6657 	__SetPageHead(head);
6658 	for (pfn = head_pfn + 1; pfn < end_pfn; pfn++) {
6659 		struct page *page = pfn_to_page(pfn);
6660 
6661 		__init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
6662 		prep_compound_tail(head, pfn - head_pfn);
6663 		set_page_count(page, 0);
6664 
6665 		/*
6666 		 * The first tail page stores compound_mapcount_ptr() and
6667 		 * compound_order() and the second tail page stores
6668 		 * compound_pincount_ptr(). Call prep_compound_head() after
6669 		 * the first and second tail pages have been initialized to
6670 		 * not have the data overwritten.
6671 		 */
6672 		if (pfn == head_pfn + 2)
6673 			prep_compound_head(head, order);
6674 	}
6675 }
6676 
6677 void __ref memmap_init_zone_device(struct zone *zone,
6678 				   unsigned long start_pfn,
6679 				   unsigned long nr_pages,
6680 				   struct dev_pagemap *pgmap)
6681 {
6682 	unsigned long pfn, end_pfn = start_pfn + nr_pages;
6683 	struct pglist_data *pgdat = zone->zone_pgdat;
6684 	struct vmem_altmap *altmap = pgmap_altmap(pgmap);
6685 	unsigned int pfns_per_compound = pgmap_vmemmap_nr(pgmap);
6686 	unsigned long zone_idx = zone_idx(zone);
6687 	unsigned long start = jiffies;
6688 	int nid = pgdat->node_id;
6689 
6690 	if (WARN_ON_ONCE(!pgmap || zone_idx(zone) != ZONE_DEVICE))
6691 		return;
6692 
6693 	/*
6694 	 * The call to memmap_init should have already taken care
6695 	 * of the pages reserved for the memmap, so we can just jump to
6696 	 * the end of that region and start processing the device pages.
6697 	 */
6698 	if (altmap) {
6699 		start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
6700 		nr_pages = end_pfn - start_pfn;
6701 	}
6702 
6703 	for (pfn = start_pfn; pfn < end_pfn; pfn += pfns_per_compound) {
6704 		struct page *page = pfn_to_page(pfn);
6705 
6706 		__init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
6707 
6708 		if (pfns_per_compound == 1)
6709 			continue;
6710 
6711 		memmap_init_compound(page, pfn, zone_idx, nid, pgmap,
6712 				     pfns_per_compound);
6713 	}
6714 
6715 	pr_info("%s initialised %lu pages in %ums\n", __func__,
6716 		nr_pages, jiffies_to_msecs(jiffies - start));
6717 }
6718 
6719 #endif
6720 static void __meminit zone_init_free_lists(struct zone *zone)
6721 {
6722 	unsigned int order, t;
6723 	for_each_migratetype_order(order, t) {
6724 		INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
6725 		zone->free_area[order].nr_free = 0;
6726 	}
6727 }
6728 
6729 /*
6730  * Only struct pages that correspond to ranges defined by memblock.memory
6731  * are zeroed and initialized by going through __init_single_page() during
6732  * memmap_init_zone_range().
6733  *
6734  * But, there could be struct pages that correspond to holes in
6735  * memblock.memory. This can happen because of the following reasons:
6736  * - physical memory bank size is not necessarily the exact multiple of the
6737  *   arbitrary section size
6738  * - early reserved memory may not be listed in memblock.memory
6739  * - memory layouts defined with memmap= kernel parameter may not align
6740  *   nicely with memmap sections
6741  *
6742  * Explicitly initialize those struct pages so that:
6743  * - PG_Reserved is set
6744  * - zone and node links point to zone and node that span the page if the
6745  *   hole is in the middle of a zone
6746  * - zone and node links point to adjacent zone/node if the hole falls on
6747  *   the zone boundary; the pages in such holes will be prepended to the
6748  *   zone/node above the hole except for the trailing pages in the last
6749  *   section that will be appended to the zone/node below.
6750  */
6751 static void __init init_unavailable_range(unsigned long spfn,
6752 					  unsigned long epfn,
6753 					  int zone, int node)
6754 {
6755 	unsigned long pfn;
6756 	u64 pgcnt = 0;
6757 
6758 	for (pfn = spfn; pfn < epfn; pfn++) {
6759 		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
6760 			pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
6761 				+ pageblock_nr_pages - 1;
6762 			continue;
6763 		}
6764 		__init_single_page(pfn_to_page(pfn), pfn, zone, node);
6765 		__SetPageReserved(pfn_to_page(pfn));
6766 		pgcnt++;
6767 	}
6768 
6769 	if (pgcnt)
6770 		pr_info("On node %d, zone %s: %lld pages in unavailable ranges",
6771 			node, zone_names[zone], pgcnt);
6772 }
6773 
6774 static void __init memmap_init_zone_range(struct zone *zone,
6775 					  unsigned long start_pfn,
6776 					  unsigned long end_pfn,
6777 					  unsigned long *hole_pfn)
6778 {
6779 	unsigned long zone_start_pfn = zone->zone_start_pfn;
6780 	unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages;
6781 	int nid = zone_to_nid(zone), zone_id = zone_idx(zone);
6782 
6783 	start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn);
6784 	end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn);
6785 
6786 	if (start_pfn >= end_pfn)
6787 		return;
6788 
6789 	memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn,
6790 			  zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
6791 
6792 	if (*hole_pfn < start_pfn)
6793 		init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid);
6794 
6795 	*hole_pfn = end_pfn;
6796 }
6797 
6798 static void __init memmap_init(void)
6799 {
6800 	unsigned long start_pfn, end_pfn;
6801 	unsigned long hole_pfn = 0;
6802 	int i, j, zone_id = 0, nid;
6803 
6804 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6805 		struct pglist_data *node = NODE_DATA(nid);
6806 
6807 		for (j = 0; j < MAX_NR_ZONES; j++) {
6808 			struct zone *zone = node->node_zones + j;
6809 
6810 			if (!populated_zone(zone))
6811 				continue;
6812 
6813 			memmap_init_zone_range(zone, start_pfn, end_pfn,
6814 					       &hole_pfn);
6815 			zone_id = j;
6816 		}
6817 	}
6818 
6819 #ifdef CONFIG_SPARSEMEM
6820 	/*
6821 	 * Initialize the memory map for hole in the range [memory_end,
6822 	 * section_end].
6823 	 * Append the pages in this hole to the highest zone in the last
6824 	 * node.
6825 	 * The call to init_unavailable_range() is outside the ifdef to
6826 	 * silence the compiler warining about zone_id set but not used;
6827 	 * for FLATMEM it is a nop anyway
6828 	 */
6829 	end_pfn = round_up(end_pfn, PAGES_PER_SECTION);
6830 	if (hole_pfn < end_pfn)
6831 #endif
6832 		init_unavailable_range(hole_pfn, end_pfn, zone_id, nid);
6833 }
6834 
6835 void __init *memmap_alloc(phys_addr_t size, phys_addr_t align,
6836 			  phys_addr_t min_addr, int nid, bool exact_nid)
6837 {
6838 	void *ptr;
6839 
6840 	if (exact_nid)
6841 		ptr = memblock_alloc_exact_nid_raw(size, align, min_addr,
6842 						   MEMBLOCK_ALLOC_ACCESSIBLE,
6843 						   nid);
6844 	else
6845 		ptr = memblock_alloc_try_nid_raw(size, align, min_addr,
6846 						 MEMBLOCK_ALLOC_ACCESSIBLE,
6847 						 nid);
6848 
6849 	if (ptr && size > 0)
6850 		page_init_poison(ptr, size);
6851 
6852 	return ptr;
6853 }
6854 
6855 static int zone_batchsize(struct zone *zone)
6856 {
6857 #ifdef CONFIG_MMU
6858 	int batch;
6859 
6860 	/*
6861 	 * The number of pages to batch allocate is either ~0.1%
6862 	 * of the zone or 1MB, whichever is smaller. The batch
6863 	 * size is striking a balance between allocation latency
6864 	 * and zone lock contention.
6865 	 */
6866 	batch = min(zone_managed_pages(zone) >> 10, (1024 * 1024) / PAGE_SIZE);
6867 	batch /= 4;		/* We effectively *= 4 below */
6868 	if (batch < 1)
6869 		batch = 1;
6870 
6871 	/*
6872 	 * Clamp the batch to a 2^n - 1 value. Having a power
6873 	 * of 2 value was found to be more likely to have
6874 	 * suboptimal cache aliasing properties in some cases.
6875 	 *
6876 	 * For example if 2 tasks are alternately allocating
6877 	 * batches of pages, one task can end up with a lot
6878 	 * of pages of one half of the possible page colors
6879 	 * and the other with pages of the other colors.
6880 	 */
6881 	batch = rounddown_pow_of_two(batch + batch/2) - 1;
6882 
6883 	return batch;
6884 
6885 #else
6886 	/* The deferral and batching of frees should be suppressed under NOMMU
6887 	 * conditions.
6888 	 *
6889 	 * The problem is that NOMMU needs to be able to allocate large chunks
6890 	 * of contiguous memory as there's no hardware page translation to
6891 	 * assemble apparent contiguous memory from discontiguous pages.
6892 	 *
6893 	 * Queueing large contiguous runs of pages for batching, however,
6894 	 * causes the pages to actually be freed in smaller chunks.  As there
6895 	 * can be a significant delay between the individual batches being
6896 	 * recycled, this leads to the once large chunks of space being
6897 	 * fragmented and becoming unavailable for high-order allocations.
6898 	 */
6899 	return 0;
6900 #endif
6901 }
6902 
6903 static int zone_highsize(struct zone *zone, int batch, int cpu_online)
6904 {
6905 #ifdef CONFIG_MMU
6906 	int high;
6907 	int nr_split_cpus;
6908 	unsigned long total_pages;
6909 
6910 	if (!percpu_pagelist_high_fraction) {
6911 		/*
6912 		 * By default, the high value of the pcp is based on the zone
6913 		 * low watermark so that if they are full then background
6914 		 * reclaim will not be started prematurely.
6915 		 */
6916 		total_pages = low_wmark_pages(zone);
6917 	} else {
6918 		/*
6919 		 * If percpu_pagelist_high_fraction is configured, the high
6920 		 * value is based on a fraction of the managed pages in the
6921 		 * zone.
6922 		 */
6923 		total_pages = zone_managed_pages(zone) / percpu_pagelist_high_fraction;
6924 	}
6925 
6926 	/*
6927 	 * Split the high value across all online CPUs local to the zone. Note
6928 	 * that early in boot that CPUs may not be online yet and that during
6929 	 * CPU hotplug that the cpumask is not yet updated when a CPU is being
6930 	 * onlined. For memory nodes that have no CPUs, split pcp->high across
6931 	 * all online CPUs to mitigate the risk that reclaim is triggered
6932 	 * prematurely due to pages stored on pcp lists.
6933 	 */
6934 	nr_split_cpus = cpumask_weight(cpumask_of_node(zone_to_nid(zone))) + cpu_online;
6935 	if (!nr_split_cpus)
6936 		nr_split_cpus = num_online_cpus();
6937 	high = total_pages / nr_split_cpus;
6938 
6939 	/*
6940 	 * Ensure high is at least batch*4. The multiple is based on the
6941 	 * historical relationship between high and batch.
6942 	 */
6943 	high = max(high, batch << 2);
6944 
6945 	return high;
6946 #else
6947 	return 0;
6948 #endif
6949 }
6950 
6951 /*
6952  * pcp->high and pcp->batch values are related and generally batch is lower
6953  * than high. They are also related to pcp->count such that count is lower
6954  * than high, and as soon as it reaches high, the pcplist is flushed.
6955  *
6956  * However, guaranteeing these relations at all times would require e.g. write
6957  * barriers here but also careful usage of read barriers at the read side, and
6958  * thus be prone to error and bad for performance. Thus the update only prevents
6959  * store tearing. Any new users of pcp->batch and pcp->high should ensure they
6960  * can cope with those fields changing asynchronously, and fully trust only the
6961  * pcp->count field on the local CPU with interrupts disabled.
6962  *
6963  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
6964  * outside of boot time (or some other assurance that no concurrent updaters
6965  * exist).
6966  */
6967 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
6968 		unsigned long batch)
6969 {
6970 	WRITE_ONCE(pcp->batch, batch);
6971 	WRITE_ONCE(pcp->high, high);
6972 }
6973 
6974 static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonestat *pzstats)
6975 {
6976 	int pindex;
6977 
6978 	memset(pcp, 0, sizeof(*pcp));
6979 	memset(pzstats, 0, sizeof(*pzstats));
6980 
6981 	for (pindex = 0; pindex < NR_PCP_LISTS; pindex++)
6982 		INIT_LIST_HEAD(&pcp->lists[pindex]);
6983 
6984 	/*
6985 	 * Set batch and high values safe for a boot pageset. A true percpu
6986 	 * pageset's initialization will update them subsequently. Here we don't
6987 	 * need to be as careful as pageset_update() as nobody can access the
6988 	 * pageset yet.
6989 	 */
6990 	pcp->high = BOOT_PAGESET_HIGH;
6991 	pcp->batch = BOOT_PAGESET_BATCH;
6992 	pcp->free_factor = 0;
6993 }
6994 
6995 static void __zone_set_pageset_high_and_batch(struct zone *zone, unsigned long high,
6996 		unsigned long batch)
6997 {
6998 	struct per_cpu_pages *pcp;
6999 	int cpu;
7000 
7001 	for_each_possible_cpu(cpu) {
7002 		pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
7003 		pageset_update(pcp, high, batch);
7004 	}
7005 }
7006 
7007 /*
7008  * Calculate and set new high and batch values for all per-cpu pagesets of a
7009  * zone based on the zone's size.
7010  */
7011 static void zone_set_pageset_high_and_batch(struct zone *zone, int cpu_online)
7012 {
7013 	int new_high, new_batch;
7014 
7015 	new_batch = max(1, zone_batchsize(zone));
7016 	new_high = zone_highsize(zone, new_batch, cpu_online);
7017 
7018 	if (zone->pageset_high == new_high &&
7019 	    zone->pageset_batch == new_batch)
7020 		return;
7021 
7022 	zone->pageset_high = new_high;
7023 	zone->pageset_batch = new_batch;
7024 
7025 	__zone_set_pageset_high_and_batch(zone, new_high, new_batch);
7026 }
7027 
7028 void __meminit setup_zone_pageset(struct zone *zone)
7029 {
7030 	int cpu;
7031 
7032 	/* Size may be 0 on !SMP && !NUMA */
7033 	if (sizeof(struct per_cpu_zonestat) > 0)
7034 		zone->per_cpu_zonestats = alloc_percpu(struct per_cpu_zonestat);
7035 
7036 	zone->per_cpu_pageset = alloc_percpu(struct per_cpu_pages);
7037 	for_each_possible_cpu(cpu) {
7038 		struct per_cpu_pages *pcp;
7039 		struct per_cpu_zonestat *pzstats;
7040 
7041 		pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu);
7042 		pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
7043 		per_cpu_pages_init(pcp, pzstats);
7044 	}
7045 
7046 	zone_set_pageset_high_and_batch(zone, 0);
7047 }
7048 
7049 /*
7050  * Allocate per cpu pagesets and initialize them.
7051  * Before this call only boot pagesets were available.
7052  */
7053 void __init setup_per_cpu_pageset(void)
7054 {
7055 	struct pglist_data *pgdat;
7056 	struct zone *zone;
7057 	int __maybe_unused cpu;
7058 
7059 	for_each_populated_zone(zone)
7060 		setup_zone_pageset(zone);
7061 
7062 #ifdef CONFIG_NUMA
7063 	/*
7064 	 * Unpopulated zones continue using the boot pagesets.
7065 	 * The numa stats for these pagesets need to be reset.
7066 	 * Otherwise, they will end up skewing the stats of
7067 	 * the nodes these zones are associated with.
7068 	 */
7069 	for_each_possible_cpu(cpu) {
7070 		struct per_cpu_zonestat *pzstats = &per_cpu(boot_zonestats, cpu);
7071 		memset(pzstats->vm_numa_event, 0,
7072 		       sizeof(pzstats->vm_numa_event));
7073 	}
7074 #endif
7075 
7076 	for_each_online_pgdat(pgdat)
7077 		pgdat->per_cpu_nodestats =
7078 			alloc_percpu(struct per_cpu_nodestat);
7079 }
7080 
7081 static __meminit void zone_pcp_init(struct zone *zone)
7082 {
7083 	/*
7084 	 * per cpu subsystem is not up at this point. The following code
7085 	 * relies on the ability of the linker to provide the
7086 	 * offset of a (static) per cpu variable into the per cpu area.
7087 	 */
7088 	zone->per_cpu_pageset = &boot_pageset;
7089 	zone->per_cpu_zonestats = &boot_zonestats;
7090 	zone->pageset_high = BOOT_PAGESET_HIGH;
7091 	zone->pageset_batch = BOOT_PAGESET_BATCH;
7092 
7093 	if (populated_zone(zone))
7094 		pr_debug("  %s zone: %lu pages, LIFO batch:%u\n", zone->name,
7095 			 zone->present_pages, zone_batchsize(zone));
7096 }
7097 
7098 void __meminit init_currently_empty_zone(struct zone *zone,
7099 					unsigned long zone_start_pfn,
7100 					unsigned long size)
7101 {
7102 	struct pglist_data *pgdat = zone->zone_pgdat;
7103 	int zone_idx = zone_idx(zone) + 1;
7104 
7105 	if (zone_idx > pgdat->nr_zones)
7106 		pgdat->nr_zones = zone_idx;
7107 
7108 	zone->zone_start_pfn = zone_start_pfn;
7109 
7110 	mminit_dprintk(MMINIT_TRACE, "memmap_init",
7111 			"Initialising map node %d zone %lu pfns %lu -> %lu\n",
7112 			pgdat->node_id,
7113 			(unsigned long)zone_idx(zone),
7114 			zone_start_pfn, (zone_start_pfn + size));
7115 
7116 	zone_init_free_lists(zone);
7117 	zone->initialized = 1;
7118 }
7119 
7120 /**
7121  * get_pfn_range_for_nid - Return the start and end page frames for a node
7122  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
7123  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
7124  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
7125  *
7126  * It returns the start and end page frame of a node based on information
7127  * provided by memblock_set_node(). If called for a node
7128  * with no available memory, a warning is printed and the start and end
7129  * PFNs will be 0.
7130  */
7131 void __init get_pfn_range_for_nid(unsigned int nid,
7132 			unsigned long *start_pfn, unsigned long *end_pfn)
7133 {
7134 	unsigned long this_start_pfn, this_end_pfn;
7135 	int i;
7136 
7137 	*start_pfn = -1UL;
7138 	*end_pfn = 0;
7139 
7140 	for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
7141 		*start_pfn = min(*start_pfn, this_start_pfn);
7142 		*end_pfn = max(*end_pfn, this_end_pfn);
7143 	}
7144 
7145 	if (*start_pfn == -1UL)
7146 		*start_pfn = 0;
7147 }
7148 
7149 /*
7150  * This finds a zone that can be used for ZONE_MOVABLE pages. The
7151  * assumption is made that zones within a node are ordered in monotonic
7152  * increasing memory addresses so that the "highest" populated zone is used
7153  */
7154 static void __init find_usable_zone_for_movable(void)
7155 {
7156 	int zone_index;
7157 	for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
7158 		if (zone_index == ZONE_MOVABLE)
7159 			continue;
7160 
7161 		if (arch_zone_highest_possible_pfn[zone_index] >
7162 				arch_zone_lowest_possible_pfn[zone_index])
7163 			break;
7164 	}
7165 
7166 	VM_BUG_ON(zone_index == -1);
7167 	movable_zone = zone_index;
7168 }
7169 
7170 /*
7171  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
7172  * because it is sized independent of architecture. Unlike the other zones,
7173  * the starting point for ZONE_MOVABLE is not fixed. It may be different
7174  * in each node depending on the size of each node and how evenly kernelcore
7175  * is distributed. This helper function adjusts the zone ranges
7176  * provided by the architecture for a given node by using the end of the
7177  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
7178  * zones within a node are in order of monotonic increases memory addresses
7179  */
7180 static void __init adjust_zone_range_for_zone_movable(int nid,
7181 					unsigned long zone_type,
7182 					unsigned long node_start_pfn,
7183 					unsigned long node_end_pfn,
7184 					unsigned long *zone_start_pfn,
7185 					unsigned long *zone_end_pfn)
7186 {
7187 	/* Only adjust if ZONE_MOVABLE is on this node */
7188 	if (zone_movable_pfn[nid]) {
7189 		/* Size ZONE_MOVABLE */
7190 		if (zone_type == ZONE_MOVABLE) {
7191 			*zone_start_pfn = zone_movable_pfn[nid];
7192 			*zone_end_pfn = min(node_end_pfn,
7193 				arch_zone_highest_possible_pfn[movable_zone]);
7194 
7195 		/* Adjust for ZONE_MOVABLE starting within this range */
7196 		} else if (!mirrored_kernelcore &&
7197 			*zone_start_pfn < zone_movable_pfn[nid] &&
7198 			*zone_end_pfn > zone_movable_pfn[nid]) {
7199 			*zone_end_pfn = zone_movable_pfn[nid];
7200 
7201 		/* Check if this whole range is within ZONE_MOVABLE */
7202 		} else if (*zone_start_pfn >= zone_movable_pfn[nid])
7203 			*zone_start_pfn = *zone_end_pfn;
7204 	}
7205 }
7206 
7207 /*
7208  * Return the number of pages a zone spans in a node, including holes
7209  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
7210  */
7211 static unsigned long __init zone_spanned_pages_in_node(int nid,
7212 					unsigned long zone_type,
7213 					unsigned long node_start_pfn,
7214 					unsigned long node_end_pfn,
7215 					unsigned long *zone_start_pfn,
7216 					unsigned long *zone_end_pfn)
7217 {
7218 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
7219 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
7220 	/* When hotadd a new node from cpu_up(), the node should be empty */
7221 	if (!node_start_pfn && !node_end_pfn)
7222 		return 0;
7223 
7224 	/* Get the start and end of the zone */
7225 	*zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
7226 	*zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
7227 	adjust_zone_range_for_zone_movable(nid, zone_type,
7228 				node_start_pfn, node_end_pfn,
7229 				zone_start_pfn, zone_end_pfn);
7230 
7231 	/* Check that this node has pages within the zone's required range */
7232 	if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
7233 		return 0;
7234 
7235 	/* Move the zone boundaries inside the node if necessary */
7236 	*zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
7237 	*zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
7238 
7239 	/* Return the spanned pages */
7240 	return *zone_end_pfn - *zone_start_pfn;
7241 }
7242 
7243 /*
7244  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
7245  * then all holes in the requested range will be accounted for.
7246  */
7247 unsigned long __init __absent_pages_in_range(int nid,
7248 				unsigned long range_start_pfn,
7249 				unsigned long range_end_pfn)
7250 {
7251 	unsigned long nr_absent = range_end_pfn - range_start_pfn;
7252 	unsigned long start_pfn, end_pfn;
7253 	int i;
7254 
7255 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7256 		start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
7257 		end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
7258 		nr_absent -= end_pfn - start_pfn;
7259 	}
7260 	return nr_absent;
7261 }
7262 
7263 /**
7264  * absent_pages_in_range - Return number of page frames in holes within a range
7265  * @start_pfn: The start PFN to start searching for holes
7266  * @end_pfn: The end PFN to stop searching for holes
7267  *
7268  * Return: the number of pages frames in memory holes within a range.
7269  */
7270 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
7271 							unsigned long end_pfn)
7272 {
7273 	return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
7274 }
7275 
7276 /* Return the number of page frames in holes in a zone on a node */
7277 static unsigned long __init zone_absent_pages_in_node(int nid,
7278 					unsigned long zone_type,
7279 					unsigned long node_start_pfn,
7280 					unsigned long node_end_pfn)
7281 {
7282 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
7283 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
7284 	unsigned long zone_start_pfn, zone_end_pfn;
7285 	unsigned long nr_absent;
7286 
7287 	/* When hotadd a new node from cpu_up(), the node should be empty */
7288 	if (!node_start_pfn && !node_end_pfn)
7289 		return 0;
7290 
7291 	zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
7292 	zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
7293 
7294 	adjust_zone_range_for_zone_movable(nid, zone_type,
7295 			node_start_pfn, node_end_pfn,
7296 			&zone_start_pfn, &zone_end_pfn);
7297 	nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
7298 
7299 	/*
7300 	 * ZONE_MOVABLE handling.
7301 	 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
7302 	 * and vice versa.
7303 	 */
7304 	if (mirrored_kernelcore && zone_movable_pfn[nid]) {
7305 		unsigned long start_pfn, end_pfn;
7306 		struct memblock_region *r;
7307 
7308 		for_each_mem_region(r) {
7309 			start_pfn = clamp(memblock_region_memory_base_pfn(r),
7310 					  zone_start_pfn, zone_end_pfn);
7311 			end_pfn = clamp(memblock_region_memory_end_pfn(r),
7312 					zone_start_pfn, zone_end_pfn);
7313 
7314 			if (zone_type == ZONE_MOVABLE &&
7315 			    memblock_is_mirror(r))
7316 				nr_absent += end_pfn - start_pfn;
7317 
7318 			if (zone_type == ZONE_NORMAL &&
7319 			    !memblock_is_mirror(r))
7320 				nr_absent += end_pfn - start_pfn;
7321 		}
7322 	}
7323 
7324 	return nr_absent;
7325 }
7326 
7327 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
7328 						unsigned long node_start_pfn,
7329 						unsigned long node_end_pfn)
7330 {
7331 	unsigned long realtotalpages = 0, totalpages = 0;
7332 	enum zone_type i;
7333 
7334 	for (i = 0; i < MAX_NR_ZONES; i++) {
7335 		struct zone *zone = pgdat->node_zones + i;
7336 		unsigned long zone_start_pfn, zone_end_pfn;
7337 		unsigned long spanned, absent;
7338 		unsigned long size, real_size;
7339 
7340 		spanned = zone_spanned_pages_in_node(pgdat->node_id, i,
7341 						     node_start_pfn,
7342 						     node_end_pfn,
7343 						     &zone_start_pfn,
7344 						     &zone_end_pfn);
7345 		absent = zone_absent_pages_in_node(pgdat->node_id, i,
7346 						   node_start_pfn,
7347 						   node_end_pfn);
7348 
7349 		size = spanned;
7350 		real_size = size - absent;
7351 
7352 		if (size)
7353 			zone->zone_start_pfn = zone_start_pfn;
7354 		else
7355 			zone->zone_start_pfn = 0;
7356 		zone->spanned_pages = size;
7357 		zone->present_pages = real_size;
7358 #if defined(CONFIG_MEMORY_HOTPLUG)
7359 		zone->present_early_pages = real_size;
7360 #endif
7361 
7362 		totalpages += size;
7363 		realtotalpages += real_size;
7364 	}
7365 
7366 	pgdat->node_spanned_pages = totalpages;
7367 	pgdat->node_present_pages = realtotalpages;
7368 	pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
7369 }
7370 
7371 #ifndef CONFIG_SPARSEMEM
7372 /*
7373  * Calculate the size of the zone->blockflags rounded to an unsigned long
7374  * Start by making sure zonesize is a multiple of pageblock_order by rounding
7375  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
7376  * round what is now in bits to nearest long in bits, then return it in
7377  * bytes.
7378  */
7379 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
7380 {
7381 	unsigned long usemapsize;
7382 
7383 	zonesize += zone_start_pfn & (pageblock_nr_pages-1);
7384 	usemapsize = roundup(zonesize, pageblock_nr_pages);
7385 	usemapsize = usemapsize >> pageblock_order;
7386 	usemapsize *= NR_PAGEBLOCK_BITS;
7387 	usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
7388 
7389 	return usemapsize / 8;
7390 }
7391 
7392 static void __ref setup_usemap(struct zone *zone)
7393 {
7394 	unsigned long usemapsize = usemap_size(zone->zone_start_pfn,
7395 					       zone->spanned_pages);
7396 	zone->pageblock_flags = NULL;
7397 	if (usemapsize) {
7398 		zone->pageblock_flags =
7399 			memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
7400 					    zone_to_nid(zone));
7401 		if (!zone->pageblock_flags)
7402 			panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
7403 			      usemapsize, zone->name, zone_to_nid(zone));
7404 	}
7405 }
7406 #else
7407 static inline void setup_usemap(struct zone *zone) {}
7408 #endif /* CONFIG_SPARSEMEM */
7409 
7410 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
7411 
7412 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
7413 void __init set_pageblock_order(void)
7414 {
7415 	unsigned int order = MAX_ORDER - 1;
7416 
7417 	/* Check that pageblock_nr_pages has not already been setup */
7418 	if (pageblock_order)
7419 		return;
7420 
7421 	/* Don't let pageblocks exceed the maximum allocation granularity. */
7422 	if (HPAGE_SHIFT > PAGE_SHIFT && HUGETLB_PAGE_ORDER < order)
7423 		order = HUGETLB_PAGE_ORDER;
7424 
7425 	/*
7426 	 * Assume the largest contiguous order of interest is a huge page.
7427 	 * This value may be variable depending on boot parameters on IA64 and
7428 	 * powerpc.
7429 	 */
7430 	pageblock_order = order;
7431 }
7432 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
7433 
7434 /*
7435  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
7436  * is unused as pageblock_order is set at compile-time. See
7437  * include/linux/pageblock-flags.h for the values of pageblock_order based on
7438  * the kernel config
7439  */
7440 void __init set_pageblock_order(void)
7441 {
7442 }
7443 
7444 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
7445 
7446 static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
7447 						unsigned long present_pages)
7448 {
7449 	unsigned long pages = spanned_pages;
7450 
7451 	/*
7452 	 * Provide a more accurate estimation if there are holes within
7453 	 * the zone and SPARSEMEM is in use. If there are holes within the
7454 	 * zone, each populated memory region may cost us one or two extra
7455 	 * memmap pages due to alignment because memmap pages for each
7456 	 * populated regions may not be naturally aligned on page boundary.
7457 	 * So the (present_pages >> 4) heuristic is a tradeoff for that.
7458 	 */
7459 	if (spanned_pages > present_pages + (present_pages >> 4) &&
7460 	    IS_ENABLED(CONFIG_SPARSEMEM))
7461 		pages = present_pages;
7462 
7463 	return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
7464 }
7465 
7466 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
7467 static void pgdat_init_split_queue(struct pglist_data *pgdat)
7468 {
7469 	struct deferred_split *ds_queue = &pgdat->deferred_split_queue;
7470 
7471 	spin_lock_init(&ds_queue->split_queue_lock);
7472 	INIT_LIST_HEAD(&ds_queue->split_queue);
7473 	ds_queue->split_queue_len = 0;
7474 }
7475 #else
7476 static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
7477 #endif
7478 
7479 #ifdef CONFIG_COMPACTION
7480 static void pgdat_init_kcompactd(struct pglist_data *pgdat)
7481 {
7482 	init_waitqueue_head(&pgdat->kcompactd_wait);
7483 }
7484 #else
7485 static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
7486 #endif
7487 
7488 static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
7489 {
7490 	int i;
7491 
7492 	pgdat_resize_init(pgdat);
7493 
7494 	pgdat_init_split_queue(pgdat);
7495 	pgdat_init_kcompactd(pgdat);
7496 
7497 	init_waitqueue_head(&pgdat->kswapd_wait);
7498 	init_waitqueue_head(&pgdat->pfmemalloc_wait);
7499 
7500 	for (i = 0; i < NR_VMSCAN_THROTTLE; i++)
7501 		init_waitqueue_head(&pgdat->reclaim_wait[i]);
7502 
7503 	pgdat_page_ext_init(pgdat);
7504 	lruvec_init(&pgdat->__lruvec);
7505 }
7506 
7507 static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
7508 							unsigned long remaining_pages)
7509 {
7510 	atomic_long_set(&zone->managed_pages, remaining_pages);
7511 	zone_set_nid(zone, nid);
7512 	zone->name = zone_names[idx];
7513 	zone->zone_pgdat = NODE_DATA(nid);
7514 	spin_lock_init(&zone->lock);
7515 	zone_seqlock_init(zone);
7516 	zone_pcp_init(zone);
7517 }
7518 
7519 /*
7520  * Set up the zone data structures
7521  * - init pgdat internals
7522  * - init all zones belonging to this node
7523  *
7524  * NOTE: this function is only called during memory hotplug
7525  */
7526 #ifdef CONFIG_MEMORY_HOTPLUG
7527 void __ref free_area_init_core_hotplug(struct pglist_data *pgdat)
7528 {
7529 	int nid = pgdat->node_id;
7530 	enum zone_type z;
7531 	int cpu;
7532 
7533 	pgdat_init_internals(pgdat);
7534 
7535 	if (pgdat->per_cpu_nodestats == &boot_nodestats)
7536 		pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
7537 
7538 	/*
7539 	 * Reset the nr_zones, order and highest_zoneidx before reuse.
7540 	 * Note that kswapd will init kswapd_highest_zoneidx properly
7541 	 * when it starts in the near future.
7542 	 */
7543 	pgdat->nr_zones = 0;
7544 	pgdat->kswapd_order = 0;
7545 	pgdat->kswapd_highest_zoneidx = 0;
7546 	pgdat->node_start_pfn = 0;
7547 	for_each_online_cpu(cpu) {
7548 		struct per_cpu_nodestat *p;
7549 
7550 		p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
7551 		memset(p, 0, sizeof(*p));
7552 	}
7553 
7554 	for (z = 0; z < MAX_NR_ZONES; z++)
7555 		zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
7556 }
7557 #endif
7558 
7559 /*
7560  * Set up the zone data structures:
7561  *   - mark all pages reserved
7562  *   - mark all memory queues empty
7563  *   - clear the memory bitmaps
7564  *
7565  * NOTE: pgdat should get zeroed by caller.
7566  * NOTE: this function is only called during early init.
7567  */
7568 static void __init free_area_init_core(struct pglist_data *pgdat)
7569 {
7570 	enum zone_type j;
7571 	int nid = pgdat->node_id;
7572 
7573 	pgdat_init_internals(pgdat);
7574 	pgdat->per_cpu_nodestats = &boot_nodestats;
7575 
7576 	for (j = 0; j < MAX_NR_ZONES; j++) {
7577 		struct zone *zone = pgdat->node_zones + j;
7578 		unsigned long size, freesize, memmap_pages;
7579 
7580 		size = zone->spanned_pages;
7581 		freesize = zone->present_pages;
7582 
7583 		/*
7584 		 * Adjust freesize so that it accounts for how much memory
7585 		 * is used by this zone for memmap. This affects the watermark
7586 		 * and per-cpu initialisations
7587 		 */
7588 		memmap_pages = calc_memmap_size(size, freesize);
7589 		if (!is_highmem_idx(j)) {
7590 			if (freesize >= memmap_pages) {
7591 				freesize -= memmap_pages;
7592 				if (memmap_pages)
7593 					pr_debug("  %s zone: %lu pages used for memmap\n",
7594 						 zone_names[j], memmap_pages);
7595 			} else
7596 				pr_warn("  %s zone: %lu memmap pages exceeds freesize %lu\n",
7597 					zone_names[j], memmap_pages, freesize);
7598 		}
7599 
7600 		/* Account for reserved pages */
7601 		if (j == 0 && freesize > dma_reserve) {
7602 			freesize -= dma_reserve;
7603 			pr_debug("  %s zone: %lu pages reserved\n", zone_names[0], dma_reserve);
7604 		}
7605 
7606 		if (!is_highmem_idx(j))
7607 			nr_kernel_pages += freesize;
7608 		/* Charge for highmem memmap if there are enough kernel pages */
7609 		else if (nr_kernel_pages > memmap_pages * 2)
7610 			nr_kernel_pages -= memmap_pages;
7611 		nr_all_pages += freesize;
7612 
7613 		/*
7614 		 * Set an approximate value for lowmem here, it will be adjusted
7615 		 * when the bootmem allocator frees pages into the buddy system.
7616 		 * And all highmem pages will be managed by the buddy system.
7617 		 */
7618 		zone_init_internals(zone, j, nid, freesize);
7619 
7620 		if (!size)
7621 			continue;
7622 
7623 		set_pageblock_order();
7624 		setup_usemap(zone);
7625 		init_currently_empty_zone(zone, zone->zone_start_pfn, size);
7626 	}
7627 }
7628 
7629 #ifdef CONFIG_FLATMEM
7630 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
7631 {
7632 	unsigned long __maybe_unused start = 0;
7633 	unsigned long __maybe_unused offset = 0;
7634 
7635 	/* Skip empty nodes */
7636 	if (!pgdat->node_spanned_pages)
7637 		return;
7638 
7639 	start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
7640 	offset = pgdat->node_start_pfn - start;
7641 	/* ia64 gets its own node_mem_map, before this, without bootmem */
7642 	if (!pgdat->node_mem_map) {
7643 		unsigned long size, end;
7644 		struct page *map;
7645 
7646 		/*
7647 		 * The zone's endpoints aren't required to be MAX_ORDER
7648 		 * aligned but the node_mem_map endpoints must be in order
7649 		 * for the buddy allocator to function correctly.
7650 		 */
7651 		end = pgdat_end_pfn(pgdat);
7652 		end = ALIGN(end, MAX_ORDER_NR_PAGES);
7653 		size =  (end - start) * sizeof(struct page);
7654 		map = memmap_alloc(size, SMP_CACHE_BYTES, MEMBLOCK_LOW_LIMIT,
7655 				   pgdat->node_id, false);
7656 		if (!map)
7657 			panic("Failed to allocate %ld bytes for node %d memory map\n",
7658 			      size, pgdat->node_id);
7659 		pgdat->node_mem_map = map + offset;
7660 	}
7661 	pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
7662 				__func__, pgdat->node_id, (unsigned long)pgdat,
7663 				(unsigned long)pgdat->node_mem_map);
7664 #ifndef CONFIG_NUMA
7665 	/*
7666 	 * With no DISCONTIG, the global mem_map is just set as node 0's
7667 	 */
7668 	if (pgdat == NODE_DATA(0)) {
7669 		mem_map = NODE_DATA(0)->node_mem_map;
7670 		if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
7671 			mem_map -= offset;
7672 	}
7673 #endif
7674 }
7675 #else
7676 static inline void alloc_node_mem_map(struct pglist_data *pgdat) { }
7677 #endif /* CONFIG_FLATMEM */
7678 
7679 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
7680 static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
7681 {
7682 	pgdat->first_deferred_pfn = ULONG_MAX;
7683 }
7684 #else
7685 static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
7686 #endif
7687 
7688 static void __init free_area_init_node(int nid)
7689 {
7690 	pg_data_t *pgdat = NODE_DATA(nid);
7691 	unsigned long start_pfn = 0;
7692 	unsigned long end_pfn = 0;
7693 
7694 	/* pg_data_t should be reset to zero when it's allocated */
7695 	WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx);
7696 
7697 	get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
7698 
7699 	pgdat->node_id = nid;
7700 	pgdat->node_start_pfn = start_pfn;
7701 	pgdat->per_cpu_nodestats = NULL;
7702 
7703 	if (start_pfn != end_pfn) {
7704 		pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
7705 			(u64)start_pfn << PAGE_SHIFT,
7706 			end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
7707 	} else {
7708 		pr_info("Initmem setup node %d as memoryless\n", nid);
7709 	}
7710 
7711 	calculate_node_totalpages(pgdat, start_pfn, end_pfn);
7712 
7713 	alloc_node_mem_map(pgdat);
7714 	pgdat_set_deferred_range(pgdat);
7715 
7716 	free_area_init_core(pgdat);
7717 }
7718 
7719 static void __init free_area_init_memoryless_node(int nid)
7720 {
7721 	free_area_init_node(nid);
7722 }
7723 
7724 #if MAX_NUMNODES > 1
7725 /*
7726  * Figure out the number of possible node ids.
7727  */
7728 void __init setup_nr_node_ids(void)
7729 {
7730 	unsigned int highest;
7731 
7732 	highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
7733 	nr_node_ids = highest + 1;
7734 }
7735 #endif
7736 
7737 /**
7738  * node_map_pfn_alignment - determine the maximum internode alignment
7739  *
7740  * This function should be called after node map is populated and sorted.
7741  * It calculates the maximum power of two alignment which can distinguish
7742  * all the nodes.
7743  *
7744  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
7745  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
7746  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
7747  * shifted, 1GiB is enough and this function will indicate so.
7748  *
7749  * This is used to test whether pfn -> nid mapping of the chosen memory
7750  * model has fine enough granularity to avoid incorrect mapping for the
7751  * populated node map.
7752  *
7753  * Return: the determined alignment in pfn's.  0 if there is no alignment
7754  * requirement (single node).
7755  */
7756 unsigned long __init node_map_pfn_alignment(void)
7757 {
7758 	unsigned long accl_mask = 0, last_end = 0;
7759 	unsigned long start, end, mask;
7760 	int last_nid = NUMA_NO_NODE;
7761 	int i, nid;
7762 
7763 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
7764 		if (!start || last_nid < 0 || last_nid == nid) {
7765 			last_nid = nid;
7766 			last_end = end;
7767 			continue;
7768 		}
7769 
7770 		/*
7771 		 * Start with a mask granular enough to pin-point to the
7772 		 * start pfn and tick off bits one-by-one until it becomes
7773 		 * too coarse to separate the current node from the last.
7774 		 */
7775 		mask = ~((1 << __ffs(start)) - 1);
7776 		while (mask && last_end <= (start & (mask << 1)))
7777 			mask <<= 1;
7778 
7779 		/* accumulate all internode masks */
7780 		accl_mask |= mask;
7781 	}
7782 
7783 	/* convert mask to number of pages */
7784 	return ~accl_mask + 1;
7785 }
7786 
7787 /**
7788  * find_min_pfn_with_active_regions - Find the minimum PFN registered
7789  *
7790  * Return: the minimum PFN based on information provided via
7791  * memblock_set_node().
7792  */
7793 unsigned long __init find_min_pfn_with_active_regions(void)
7794 {
7795 	return PHYS_PFN(memblock_start_of_DRAM());
7796 }
7797 
7798 /*
7799  * early_calculate_totalpages()
7800  * Sum pages in active regions for movable zone.
7801  * Populate N_MEMORY for calculating usable_nodes.
7802  */
7803 static unsigned long __init early_calculate_totalpages(void)
7804 {
7805 	unsigned long totalpages = 0;
7806 	unsigned long start_pfn, end_pfn;
7807 	int i, nid;
7808 
7809 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
7810 		unsigned long pages = end_pfn - start_pfn;
7811 
7812 		totalpages += pages;
7813 		if (pages)
7814 			node_set_state(nid, N_MEMORY);
7815 	}
7816 	return totalpages;
7817 }
7818 
7819 /*
7820  * Find the PFN the Movable zone begins in each node. Kernel memory
7821  * is spread evenly between nodes as long as the nodes have enough
7822  * memory. When they don't, some nodes will have more kernelcore than
7823  * others
7824  */
7825 static void __init find_zone_movable_pfns_for_nodes(void)
7826 {
7827 	int i, nid;
7828 	unsigned long usable_startpfn;
7829 	unsigned long kernelcore_node, kernelcore_remaining;
7830 	/* save the state before borrow the nodemask */
7831 	nodemask_t saved_node_state = node_states[N_MEMORY];
7832 	unsigned long totalpages = early_calculate_totalpages();
7833 	int usable_nodes = nodes_weight(node_states[N_MEMORY]);
7834 	struct memblock_region *r;
7835 
7836 	/* Need to find movable_zone earlier when movable_node is specified. */
7837 	find_usable_zone_for_movable();
7838 
7839 	/*
7840 	 * If movable_node is specified, ignore kernelcore and movablecore
7841 	 * options.
7842 	 */
7843 	if (movable_node_is_enabled()) {
7844 		for_each_mem_region(r) {
7845 			if (!memblock_is_hotpluggable(r))
7846 				continue;
7847 
7848 			nid = memblock_get_region_node(r);
7849 
7850 			usable_startpfn = PFN_DOWN(r->base);
7851 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7852 				min(usable_startpfn, zone_movable_pfn[nid]) :
7853 				usable_startpfn;
7854 		}
7855 
7856 		goto out2;
7857 	}
7858 
7859 	/*
7860 	 * If kernelcore=mirror is specified, ignore movablecore option
7861 	 */
7862 	if (mirrored_kernelcore) {
7863 		bool mem_below_4gb_not_mirrored = false;
7864 
7865 		for_each_mem_region(r) {
7866 			if (memblock_is_mirror(r))
7867 				continue;
7868 
7869 			nid = memblock_get_region_node(r);
7870 
7871 			usable_startpfn = memblock_region_memory_base_pfn(r);
7872 
7873 			if (usable_startpfn < 0x100000) {
7874 				mem_below_4gb_not_mirrored = true;
7875 				continue;
7876 			}
7877 
7878 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
7879 				min(usable_startpfn, zone_movable_pfn[nid]) :
7880 				usable_startpfn;
7881 		}
7882 
7883 		if (mem_below_4gb_not_mirrored)
7884 			pr_warn("This configuration results in unmirrored kernel memory.\n");
7885 
7886 		goto out2;
7887 	}
7888 
7889 	/*
7890 	 * If kernelcore=nn% or movablecore=nn% was specified, calculate the
7891 	 * amount of necessary memory.
7892 	 */
7893 	if (required_kernelcore_percent)
7894 		required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
7895 				       10000UL;
7896 	if (required_movablecore_percent)
7897 		required_movablecore = (totalpages * 100 * required_movablecore_percent) /
7898 					10000UL;
7899 
7900 	/*
7901 	 * If movablecore= was specified, calculate what size of
7902 	 * kernelcore that corresponds so that memory usable for
7903 	 * any allocation type is evenly spread. If both kernelcore
7904 	 * and movablecore are specified, then the value of kernelcore
7905 	 * will be used for required_kernelcore if it's greater than
7906 	 * what movablecore would have allowed.
7907 	 */
7908 	if (required_movablecore) {
7909 		unsigned long corepages;
7910 
7911 		/*
7912 		 * Round-up so that ZONE_MOVABLE is at least as large as what
7913 		 * was requested by the user
7914 		 */
7915 		required_movablecore =
7916 			roundup(required_movablecore, MAX_ORDER_NR_PAGES);
7917 		required_movablecore = min(totalpages, required_movablecore);
7918 		corepages = totalpages - required_movablecore;
7919 
7920 		required_kernelcore = max(required_kernelcore, corepages);
7921 	}
7922 
7923 	/*
7924 	 * If kernelcore was not specified or kernelcore size is larger
7925 	 * than totalpages, there is no ZONE_MOVABLE.
7926 	 */
7927 	if (!required_kernelcore || required_kernelcore >= totalpages)
7928 		goto out;
7929 
7930 	/* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
7931 	usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
7932 
7933 restart:
7934 	/* Spread kernelcore memory as evenly as possible throughout nodes */
7935 	kernelcore_node = required_kernelcore / usable_nodes;
7936 	for_each_node_state(nid, N_MEMORY) {
7937 		unsigned long start_pfn, end_pfn;
7938 
7939 		/*
7940 		 * Recalculate kernelcore_node if the division per node
7941 		 * now exceeds what is necessary to satisfy the requested
7942 		 * amount of memory for the kernel
7943 		 */
7944 		if (required_kernelcore < kernelcore_node)
7945 			kernelcore_node = required_kernelcore / usable_nodes;
7946 
7947 		/*
7948 		 * As the map is walked, we track how much memory is usable
7949 		 * by the kernel using kernelcore_remaining. When it is
7950 		 * 0, the rest of the node is usable by ZONE_MOVABLE
7951 		 */
7952 		kernelcore_remaining = kernelcore_node;
7953 
7954 		/* Go through each range of PFNs within this node */
7955 		for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7956 			unsigned long size_pages;
7957 
7958 			start_pfn = max(start_pfn, zone_movable_pfn[nid]);
7959 			if (start_pfn >= end_pfn)
7960 				continue;
7961 
7962 			/* Account for what is only usable for kernelcore */
7963 			if (start_pfn < usable_startpfn) {
7964 				unsigned long kernel_pages;
7965 				kernel_pages = min(end_pfn, usable_startpfn)
7966 								- start_pfn;
7967 
7968 				kernelcore_remaining -= min(kernel_pages,
7969 							kernelcore_remaining);
7970 				required_kernelcore -= min(kernel_pages,
7971 							required_kernelcore);
7972 
7973 				/* Continue if range is now fully accounted */
7974 				if (end_pfn <= usable_startpfn) {
7975 
7976 					/*
7977 					 * Push zone_movable_pfn to the end so
7978 					 * that if we have to rebalance
7979 					 * kernelcore across nodes, we will
7980 					 * not double account here
7981 					 */
7982 					zone_movable_pfn[nid] = end_pfn;
7983 					continue;
7984 				}
7985 				start_pfn = usable_startpfn;
7986 			}
7987 
7988 			/*
7989 			 * The usable PFN range for ZONE_MOVABLE is from
7990 			 * start_pfn->end_pfn. Calculate size_pages as the
7991 			 * number of pages used as kernelcore
7992 			 */
7993 			size_pages = end_pfn - start_pfn;
7994 			if (size_pages > kernelcore_remaining)
7995 				size_pages = kernelcore_remaining;
7996 			zone_movable_pfn[nid] = start_pfn + size_pages;
7997 
7998 			/*
7999 			 * Some kernelcore has been met, update counts and
8000 			 * break if the kernelcore for this node has been
8001 			 * satisfied
8002 			 */
8003 			required_kernelcore -= min(required_kernelcore,
8004 								size_pages);
8005 			kernelcore_remaining -= size_pages;
8006 			if (!kernelcore_remaining)
8007 				break;
8008 		}
8009 	}
8010 
8011 	/*
8012 	 * If there is still required_kernelcore, we do another pass with one
8013 	 * less node in the count. This will push zone_movable_pfn[nid] further
8014 	 * along on the nodes that still have memory until kernelcore is
8015 	 * satisfied
8016 	 */
8017 	usable_nodes--;
8018 	if (usable_nodes && required_kernelcore > usable_nodes)
8019 		goto restart;
8020 
8021 out2:
8022 	/* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
8023 	for (nid = 0; nid < MAX_NUMNODES; nid++) {
8024 		unsigned long start_pfn, end_pfn;
8025 
8026 		zone_movable_pfn[nid] =
8027 			roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
8028 
8029 		get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
8030 		if (zone_movable_pfn[nid] >= end_pfn)
8031 			zone_movable_pfn[nid] = 0;
8032 	}
8033 
8034 out:
8035 	/* restore the node_state */
8036 	node_states[N_MEMORY] = saved_node_state;
8037 }
8038 
8039 /* Any regular or high memory on that node ? */
8040 static void check_for_memory(pg_data_t *pgdat, int nid)
8041 {
8042 	enum zone_type zone_type;
8043 
8044 	for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
8045 		struct zone *zone = &pgdat->node_zones[zone_type];
8046 		if (populated_zone(zone)) {
8047 			if (IS_ENABLED(CONFIG_HIGHMEM))
8048 				node_set_state(nid, N_HIGH_MEMORY);
8049 			if (zone_type <= ZONE_NORMAL)
8050 				node_set_state(nid, N_NORMAL_MEMORY);
8051 			break;
8052 		}
8053 	}
8054 }
8055 
8056 /*
8057  * Some architectures, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For
8058  * such cases we allow max_zone_pfn sorted in the descending order
8059  */
8060 bool __weak arch_has_descending_max_zone_pfns(void)
8061 {
8062 	return false;
8063 }
8064 
8065 /**
8066  * free_area_init - Initialise all pg_data_t and zone data
8067  * @max_zone_pfn: an array of max PFNs for each zone
8068  *
8069  * This will call free_area_init_node() for each active node in the system.
8070  * Using the page ranges provided by memblock_set_node(), the size of each
8071  * zone in each node and their holes is calculated. If the maximum PFN
8072  * between two adjacent zones match, it is assumed that the zone is empty.
8073  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
8074  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
8075  * starts where the previous one ended. For example, ZONE_DMA32 starts
8076  * at arch_max_dma_pfn.
8077  */
8078 void __init free_area_init(unsigned long *max_zone_pfn)
8079 {
8080 	unsigned long start_pfn, end_pfn;
8081 	int i, nid, zone;
8082 	bool descending;
8083 
8084 	/* Record where the zone boundaries are */
8085 	memset(arch_zone_lowest_possible_pfn, 0,
8086 				sizeof(arch_zone_lowest_possible_pfn));
8087 	memset(arch_zone_highest_possible_pfn, 0,
8088 				sizeof(arch_zone_highest_possible_pfn));
8089 
8090 	start_pfn = find_min_pfn_with_active_regions();
8091 	descending = arch_has_descending_max_zone_pfns();
8092 
8093 	for (i = 0; i < MAX_NR_ZONES; i++) {
8094 		if (descending)
8095 			zone = MAX_NR_ZONES - i - 1;
8096 		else
8097 			zone = i;
8098 
8099 		if (zone == ZONE_MOVABLE)
8100 			continue;
8101 
8102 		end_pfn = max(max_zone_pfn[zone], start_pfn);
8103 		arch_zone_lowest_possible_pfn[zone] = start_pfn;
8104 		arch_zone_highest_possible_pfn[zone] = end_pfn;
8105 
8106 		start_pfn = end_pfn;
8107 	}
8108 
8109 	/* Find the PFNs that ZONE_MOVABLE begins at in each node */
8110 	memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
8111 	find_zone_movable_pfns_for_nodes();
8112 
8113 	/* Print out the zone ranges */
8114 	pr_info("Zone ranges:\n");
8115 	for (i = 0; i < MAX_NR_ZONES; i++) {
8116 		if (i == ZONE_MOVABLE)
8117 			continue;
8118 		pr_info("  %-8s ", zone_names[i]);
8119 		if (arch_zone_lowest_possible_pfn[i] ==
8120 				arch_zone_highest_possible_pfn[i])
8121 			pr_cont("empty\n");
8122 		else
8123 			pr_cont("[mem %#018Lx-%#018Lx]\n",
8124 				(u64)arch_zone_lowest_possible_pfn[i]
8125 					<< PAGE_SHIFT,
8126 				((u64)arch_zone_highest_possible_pfn[i]
8127 					<< PAGE_SHIFT) - 1);
8128 	}
8129 
8130 	/* Print out the PFNs ZONE_MOVABLE begins at in each node */
8131 	pr_info("Movable zone start for each node\n");
8132 	for (i = 0; i < MAX_NUMNODES; i++) {
8133 		if (zone_movable_pfn[i])
8134 			pr_info("  Node %d: %#018Lx\n", i,
8135 			       (u64)zone_movable_pfn[i] << PAGE_SHIFT);
8136 	}
8137 
8138 	/*
8139 	 * Print out the early node map, and initialize the
8140 	 * subsection-map relative to active online memory ranges to
8141 	 * enable future "sub-section" extensions of the memory map.
8142 	 */
8143 	pr_info("Early memory node ranges\n");
8144 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
8145 		pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
8146 			(u64)start_pfn << PAGE_SHIFT,
8147 			((u64)end_pfn << PAGE_SHIFT) - 1);
8148 		subsection_map_init(start_pfn, end_pfn - start_pfn);
8149 	}
8150 
8151 	/* Initialise every node */
8152 	mminit_verify_pageflags_layout();
8153 	setup_nr_node_ids();
8154 	for_each_node(nid) {
8155 		pg_data_t *pgdat;
8156 
8157 		if (!node_online(nid)) {
8158 			pr_info("Initializing node %d as memoryless\n", nid);
8159 
8160 			/* Allocator not initialized yet */
8161 			pgdat = arch_alloc_nodedata(nid);
8162 			if (!pgdat) {
8163 				pr_err("Cannot allocate %zuB for node %d.\n",
8164 						sizeof(*pgdat), nid);
8165 				continue;
8166 			}
8167 			arch_refresh_nodedata(nid, pgdat);
8168 			free_area_init_memoryless_node(nid);
8169 
8170 			/*
8171 			 * We do not want to confuse userspace by sysfs
8172 			 * files/directories for node without any memory
8173 			 * attached to it, so this node is not marked as
8174 			 * N_MEMORY and not marked online so that no sysfs
8175 			 * hierarchy will be created via register_one_node for
8176 			 * it. The pgdat will get fully initialized by
8177 			 * hotadd_init_pgdat() when memory is hotplugged into
8178 			 * this node.
8179 			 */
8180 			continue;
8181 		}
8182 
8183 		pgdat = NODE_DATA(nid);
8184 		free_area_init_node(nid);
8185 
8186 		/* Any memory on that node */
8187 		if (pgdat->node_present_pages)
8188 			node_set_state(nid, N_MEMORY);
8189 		check_for_memory(pgdat, nid);
8190 	}
8191 
8192 	memmap_init();
8193 }
8194 
8195 static int __init cmdline_parse_core(char *p, unsigned long *core,
8196 				     unsigned long *percent)
8197 {
8198 	unsigned long long coremem;
8199 	char *endptr;
8200 
8201 	if (!p)
8202 		return -EINVAL;
8203 
8204 	/* Value may be a percentage of total memory, otherwise bytes */
8205 	coremem = simple_strtoull(p, &endptr, 0);
8206 	if (*endptr == '%') {
8207 		/* Paranoid check for percent values greater than 100 */
8208 		WARN_ON(coremem > 100);
8209 
8210 		*percent = coremem;
8211 	} else {
8212 		coremem = memparse(p, &p);
8213 		/* Paranoid check that UL is enough for the coremem value */
8214 		WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
8215 
8216 		*core = coremem >> PAGE_SHIFT;
8217 		*percent = 0UL;
8218 	}
8219 	return 0;
8220 }
8221 
8222 /*
8223  * kernelcore=size sets the amount of memory for use for allocations that
8224  * cannot be reclaimed or migrated.
8225  */
8226 static int __init cmdline_parse_kernelcore(char *p)
8227 {
8228 	/* parse kernelcore=mirror */
8229 	if (parse_option_str(p, "mirror")) {
8230 		mirrored_kernelcore = true;
8231 		return 0;
8232 	}
8233 
8234 	return cmdline_parse_core(p, &required_kernelcore,
8235 				  &required_kernelcore_percent);
8236 }
8237 
8238 /*
8239  * movablecore=size sets the amount of memory for use for allocations that
8240  * can be reclaimed or migrated.
8241  */
8242 static int __init cmdline_parse_movablecore(char *p)
8243 {
8244 	return cmdline_parse_core(p, &required_movablecore,
8245 				  &required_movablecore_percent);
8246 }
8247 
8248 early_param("kernelcore", cmdline_parse_kernelcore);
8249 early_param("movablecore", cmdline_parse_movablecore);
8250 
8251 void adjust_managed_page_count(struct page *page, long count)
8252 {
8253 	atomic_long_add(count, &page_zone(page)->managed_pages);
8254 	totalram_pages_add(count);
8255 #ifdef CONFIG_HIGHMEM
8256 	if (PageHighMem(page))
8257 		totalhigh_pages_add(count);
8258 #endif
8259 }
8260 EXPORT_SYMBOL(adjust_managed_page_count);
8261 
8262 unsigned long free_reserved_area(void *start, void *end, int poison, const char *s)
8263 {
8264 	void *pos;
8265 	unsigned long pages = 0;
8266 
8267 	start = (void *)PAGE_ALIGN((unsigned long)start);
8268 	end = (void *)((unsigned long)end & PAGE_MASK);
8269 	for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
8270 		struct page *page = virt_to_page(pos);
8271 		void *direct_map_addr;
8272 
8273 		/*
8274 		 * 'direct_map_addr' might be different from 'pos'
8275 		 * because some architectures' virt_to_page()
8276 		 * work with aliases.  Getting the direct map
8277 		 * address ensures that we get a _writeable_
8278 		 * alias for the memset().
8279 		 */
8280 		direct_map_addr = page_address(page);
8281 		/*
8282 		 * Perform a kasan-unchecked memset() since this memory
8283 		 * has not been initialized.
8284 		 */
8285 		direct_map_addr = kasan_reset_tag(direct_map_addr);
8286 		if ((unsigned int)poison <= 0xFF)
8287 			memset(direct_map_addr, poison, PAGE_SIZE);
8288 
8289 		free_reserved_page(page);
8290 	}
8291 
8292 	if (pages && s)
8293 		pr_info("Freeing %s memory: %ldK\n", s, K(pages));
8294 
8295 	return pages;
8296 }
8297 
8298 void __init mem_init_print_info(void)
8299 {
8300 	unsigned long physpages, codesize, datasize, rosize, bss_size;
8301 	unsigned long init_code_size, init_data_size;
8302 
8303 	physpages = get_num_physpages();
8304 	codesize = _etext - _stext;
8305 	datasize = _edata - _sdata;
8306 	rosize = __end_rodata - __start_rodata;
8307 	bss_size = __bss_stop - __bss_start;
8308 	init_data_size = __init_end - __init_begin;
8309 	init_code_size = _einittext - _sinittext;
8310 
8311 	/*
8312 	 * Detect special cases and adjust section sizes accordingly:
8313 	 * 1) .init.* may be embedded into .data sections
8314 	 * 2) .init.text.* may be out of [__init_begin, __init_end],
8315 	 *    please refer to arch/tile/kernel/vmlinux.lds.S.
8316 	 * 3) .rodata.* may be embedded into .text or .data sections.
8317 	 */
8318 #define adj_init_size(start, end, size, pos, adj) \
8319 	do { \
8320 		if (&start[0] <= &pos[0] && &pos[0] < &end[0] && size > adj) \
8321 			size -= adj; \
8322 	} while (0)
8323 
8324 	adj_init_size(__init_begin, __init_end, init_data_size,
8325 		     _sinittext, init_code_size);
8326 	adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
8327 	adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
8328 	adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
8329 	adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
8330 
8331 #undef	adj_init_size
8332 
8333 	pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
8334 #ifdef	CONFIG_HIGHMEM
8335 		", %luK highmem"
8336 #endif
8337 		")\n",
8338 		K(nr_free_pages()), K(physpages),
8339 		codesize >> 10, datasize >> 10, rosize >> 10,
8340 		(init_data_size + init_code_size) >> 10, bss_size >> 10,
8341 		K(physpages - totalram_pages() - totalcma_pages),
8342 		K(totalcma_pages)
8343 #ifdef	CONFIG_HIGHMEM
8344 		, K(totalhigh_pages())
8345 #endif
8346 		);
8347 }
8348 
8349 /**
8350  * set_dma_reserve - set the specified number of pages reserved in the first zone
8351  * @new_dma_reserve: The number of pages to mark reserved
8352  *
8353  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
8354  * In the DMA zone, a significant percentage may be consumed by kernel image
8355  * and other unfreeable allocations which can skew the watermarks badly. This
8356  * function may optionally be used to account for unfreeable pages in the
8357  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
8358  * smaller per-cpu batchsize.
8359  */
8360 void __init set_dma_reserve(unsigned long new_dma_reserve)
8361 {
8362 	dma_reserve = new_dma_reserve;
8363 }
8364 
8365 static int page_alloc_cpu_dead(unsigned int cpu)
8366 {
8367 	struct zone *zone;
8368 
8369 	lru_add_drain_cpu(cpu);
8370 	mlock_page_drain_remote(cpu);
8371 	drain_pages(cpu);
8372 
8373 	/*
8374 	 * Spill the event counters of the dead processor
8375 	 * into the current processors event counters.
8376 	 * This artificially elevates the count of the current
8377 	 * processor.
8378 	 */
8379 	vm_events_fold_cpu(cpu);
8380 
8381 	/*
8382 	 * Zero the differential counters of the dead processor
8383 	 * so that the vm statistics are consistent.
8384 	 *
8385 	 * This is only okay since the processor is dead and cannot
8386 	 * race with what we are doing.
8387 	 */
8388 	cpu_vm_stats_fold(cpu);
8389 
8390 	for_each_populated_zone(zone)
8391 		zone_pcp_update(zone, 0);
8392 
8393 	return 0;
8394 }
8395 
8396 static int page_alloc_cpu_online(unsigned int cpu)
8397 {
8398 	struct zone *zone;
8399 
8400 	for_each_populated_zone(zone)
8401 		zone_pcp_update(zone, 1);
8402 	return 0;
8403 }
8404 
8405 #ifdef CONFIG_NUMA
8406 int hashdist = HASHDIST_DEFAULT;
8407 
8408 static int __init set_hashdist(char *str)
8409 {
8410 	if (!str)
8411 		return 0;
8412 	hashdist = simple_strtoul(str, &str, 0);
8413 	return 1;
8414 }
8415 __setup("hashdist=", set_hashdist);
8416 #endif
8417 
8418 void __init page_alloc_init(void)
8419 {
8420 	int ret;
8421 
8422 #ifdef CONFIG_NUMA
8423 	if (num_node_state(N_MEMORY) == 1)
8424 		hashdist = 0;
8425 #endif
8426 
8427 	ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC,
8428 					"mm/page_alloc:pcp",
8429 					page_alloc_cpu_online,
8430 					page_alloc_cpu_dead);
8431 	WARN_ON(ret < 0);
8432 }
8433 
8434 /*
8435  * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
8436  *	or min_free_kbytes changes.
8437  */
8438 static void calculate_totalreserve_pages(void)
8439 {
8440 	struct pglist_data *pgdat;
8441 	unsigned long reserve_pages = 0;
8442 	enum zone_type i, j;
8443 
8444 	for_each_online_pgdat(pgdat) {
8445 
8446 		pgdat->totalreserve_pages = 0;
8447 
8448 		for (i = 0; i < MAX_NR_ZONES; i++) {
8449 			struct zone *zone = pgdat->node_zones + i;
8450 			long max = 0;
8451 			unsigned long managed_pages = zone_managed_pages(zone);
8452 
8453 			/* Find valid and maximum lowmem_reserve in the zone */
8454 			for (j = i; j < MAX_NR_ZONES; j++) {
8455 				if (zone->lowmem_reserve[j] > max)
8456 					max = zone->lowmem_reserve[j];
8457 			}
8458 
8459 			/* we treat the high watermark as reserved pages. */
8460 			max += high_wmark_pages(zone);
8461 
8462 			if (max > managed_pages)
8463 				max = managed_pages;
8464 
8465 			pgdat->totalreserve_pages += max;
8466 
8467 			reserve_pages += max;
8468 		}
8469 	}
8470 	totalreserve_pages = reserve_pages;
8471 }
8472 
8473 /*
8474  * setup_per_zone_lowmem_reserve - called whenever
8475  *	sysctl_lowmem_reserve_ratio changes.  Ensures that each zone
8476  *	has a correct pages reserved value, so an adequate number of
8477  *	pages are left in the zone after a successful __alloc_pages().
8478  */
8479 static void setup_per_zone_lowmem_reserve(void)
8480 {
8481 	struct pglist_data *pgdat;
8482 	enum zone_type i, j;
8483 
8484 	for_each_online_pgdat(pgdat) {
8485 		for (i = 0; i < MAX_NR_ZONES - 1; i++) {
8486 			struct zone *zone = &pgdat->node_zones[i];
8487 			int ratio = sysctl_lowmem_reserve_ratio[i];
8488 			bool clear = !ratio || !zone_managed_pages(zone);
8489 			unsigned long managed_pages = 0;
8490 
8491 			for (j = i + 1; j < MAX_NR_ZONES; j++) {
8492 				struct zone *upper_zone = &pgdat->node_zones[j];
8493 
8494 				managed_pages += zone_managed_pages(upper_zone);
8495 
8496 				if (clear)
8497 					zone->lowmem_reserve[j] = 0;
8498 				else
8499 					zone->lowmem_reserve[j] = managed_pages / ratio;
8500 			}
8501 		}
8502 	}
8503 
8504 	/* update totalreserve_pages */
8505 	calculate_totalreserve_pages();
8506 }
8507 
8508 static void __setup_per_zone_wmarks(void)
8509 {
8510 	unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
8511 	unsigned long lowmem_pages = 0;
8512 	struct zone *zone;
8513 	unsigned long flags;
8514 
8515 	/* Calculate total number of !ZONE_HIGHMEM pages */
8516 	for_each_zone(zone) {
8517 		if (!is_highmem(zone))
8518 			lowmem_pages += zone_managed_pages(zone);
8519 	}
8520 
8521 	for_each_zone(zone) {
8522 		u64 tmp;
8523 
8524 		spin_lock_irqsave(&zone->lock, flags);
8525 		tmp = (u64)pages_min * zone_managed_pages(zone);
8526 		do_div(tmp, lowmem_pages);
8527 		if (is_highmem(zone)) {
8528 			/*
8529 			 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
8530 			 * need highmem pages, so cap pages_min to a small
8531 			 * value here.
8532 			 *
8533 			 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
8534 			 * deltas control async page reclaim, and so should
8535 			 * not be capped for highmem.
8536 			 */
8537 			unsigned long min_pages;
8538 
8539 			min_pages = zone_managed_pages(zone) / 1024;
8540 			min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
8541 			zone->_watermark[WMARK_MIN] = min_pages;
8542 		} else {
8543 			/*
8544 			 * If it's a lowmem zone, reserve a number of pages
8545 			 * proportionate to the zone's size.
8546 			 */
8547 			zone->_watermark[WMARK_MIN] = tmp;
8548 		}
8549 
8550 		/*
8551 		 * Set the kswapd watermarks distance according to the
8552 		 * scale factor in proportion to available memory, but
8553 		 * ensure a minimum size on small systems.
8554 		 */
8555 		tmp = max_t(u64, tmp >> 2,
8556 			    mult_frac(zone_managed_pages(zone),
8557 				      watermark_scale_factor, 10000));
8558 
8559 		zone->watermark_boost = 0;
8560 		zone->_watermark[WMARK_LOW]  = min_wmark_pages(zone) + tmp;
8561 		zone->_watermark[WMARK_HIGH] = low_wmark_pages(zone) + tmp;
8562 		zone->_watermark[WMARK_PROMO] = high_wmark_pages(zone) + tmp;
8563 
8564 		spin_unlock_irqrestore(&zone->lock, flags);
8565 	}
8566 
8567 	/* update totalreserve_pages */
8568 	calculate_totalreserve_pages();
8569 }
8570 
8571 /**
8572  * setup_per_zone_wmarks - called when min_free_kbytes changes
8573  * or when memory is hot-{added|removed}
8574  *
8575  * Ensures that the watermark[min,low,high] values for each zone are set
8576  * correctly with respect to min_free_kbytes.
8577  */
8578 void setup_per_zone_wmarks(void)
8579 {
8580 	struct zone *zone;
8581 	static DEFINE_SPINLOCK(lock);
8582 
8583 	spin_lock(&lock);
8584 	__setup_per_zone_wmarks();
8585 	spin_unlock(&lock);
8586 
8587 	/*
8588 	 * The watermark size have changed so update the pcpu batch
8589 	 * and high limits or the limits may be inappropriate.
8590 	 */
8591 	for_each_zone(zone)
8592 		zone_pcp_update(zone, 0);
8593 }
8594 
8595 /*
8596  * Initialise min_free_kbytes.
8597  *
8598  * For small machines we want it small (128k min).  For large machines
8599  * we want it large (256MB max).  But it is not linear, because network
8600  * bandwidth does not increase linearly with machine size.  We use
8601  *
8602  *	min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
8603  *	min_free_kbytes = sqrt(lowmem_kbytes * 16)
8604  *
8605  * which yields
8606  *
8607  * 16MB:	512k
8608  * 32MB:	724k
8609  * 64MB:	1024k
8610  * 128MB:	1448k
8611  * 256MB:	2048k
8612  * 512MB:	2896k
8613  * 1024MB:	4096k
8614  * 2048MB:	5792k
8615  * 4096MB:	8192k
8616  * 8192MB:	11584k
8617  * 16384MB:	16384k
8618  */
8619 void calculate_min_free_kbytes(void)
8620 {
8621 	unsigned long lowmem_kbytes;
8622 	int new_min_free_kbytes;
8623 
8624 	lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
8625 	new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
8626 
8627 	if (new_min_free_kbytes > user_min_free_kbytes)
8628 		min_free_kbytes = clamp(new_min_free_kbytes, 128, 262144);
8629 	else
8630 		pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
8631 				new_min_free_kbytes, user_min_free_kbytes);
8632 
8633 }
8634 
8635 int __meminit init_per_zone_wmark_min(void)
8636 {
8637 	calculate_min_free_kbytes();
8638 	setup_per_zone_wmarks();
8639 	refresh_zone_stat_thresholds();
8640 	setup_per_zone_lowmem_reserve();
8641 
8642 #ifdef CONFIG_NUMA
8643 	setup_min_unmapped_ratio();
8644 	setup_min_slab_ratio();
8645 #endif
8646 
8647 	khugepaged_min_free_kbytes_update();
8648 
8649 	return 0;
8650 }
8651 postcore_initcall(init_per_zone_wmark_min)
8652 
8653 /*
8654  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
8655  *	that we can call two helper functions whenever min_free_kbytes
8656  *	changes.
8657  */
8658 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
8659 		void *buffer, size_t *length, loff_t *ppos)
8660 {
8661 	int rc;
8662 
8663 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8664 	if (rc)
8665 		return rc;
8666 
8667 	if (write) {
8668 		user_min_free_kbytes = min_free_kbytes;
8669 		setup_per_zone_wmarks();
8670 	}
8671 	return 0;
8672 }
8673 
8674 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
8675 		void *buffer, size_t *length, loff_t *ppos)
8676 {
8677 	int rc;
8678 
8679 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8680 	if (rc)
8681 		return rc;
8682 
8683 	if (write)
8684 		setup_per_zone_wmarks();
8685 
8686 	return 0;
8687 }
8688 
8689 #ifdef CONFIG_NUMA
8690 static void setup_min_unmapped_ratio(void)
8691 {
8692 	pg_data_t *pgdat;
8693 	struct zone *zone;
8694 
8695 	for_each_online_pgdat(pgdat)
8696 		pgdat->min_unmapped_pages = 0;
8697 
8698 	for_each_zone(zone)
8699 		zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) *
8700 						         sysctl_min_unmapped_ratio) / 100;
8701 }
8702 
8703 
8704 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
8705 		void *buffer, size_t *length, loff_t *ppos)
8706 {
8707 	int rc;
8708 
8709 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8710 	if (rc)
8711 		return rc;
8712 
8713 	setup_min_unmapped_ratio();
8714 
8715 	return 0;
8716 }
8717 
8718 static void setup_min_slab_ratio(void)
8719 {
8720 	pg_data_t *pgdat;
8721 	struct zone *zone;
8722 
8723 	for_each_online_pgdat(pgdat)
8724 		pgdat->min_slab_pages = 0;
8725 
8726 	for_each_zone(zone)
8727 		zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) *
8728 						     sysctl_min_slab_ratio) / 100;
8729 }
8730 
8731 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
8732 		void *buffer, size_t *length, loff_t *ppos)
8733 {
8734 	int rc;
8735 
8736 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
8737 	if (rc)
8738 		return rc;
8739 
8740 	setup_min_slab_ratio();
8741 
8742 	return 0;
8743 }
8744 #endif
8745 
8746 /*
8747  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
8748  *	proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
8749  *	whenever sysctl_lowmem_reserve_ratio changes.
8750  *
8751  * The reserve ratio obviously has absolutely no relation with the
8752  * minimum watermarks. The lowmem reserve ratio can only make sense
8753  * if in function of the boot time zone sizes.
8754  */
8755 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
8756 		void *buffer, size_t *length, loff_t *ppos)
8757 {
8758 	int i;
8759 
8760 	proc_dointvec_minmax(table, write, buffer, length, ppos);
8761 
8762 	for (i = 0; i < MAX_NR_ZONES; i++) {
8763 		if (sysctl_lowmem_reserve_ratio[i] < 1)
8764 			sysctl_lowmem_reserve_ratio[i] = 0;
8765 	}
8766 
8767 	setup_per_zone_lowmem_reserve();
8768 	return 0;
8769 }
8770 
8771 /*
8772  * percpu_pagelist_high_fraction - changes the pcp->high for each zone on each
8773  * cpu. It is the fraction of total pages in each zone that a hot per cpu
8774  * pagelist can have before it gets flushed back to buddy allocator.
8775  */
8776 int percpu_pagelist_high_fraction_sysctl_handler(struct ctl_table *table,
8777 		int write, void *buffer, size_t *length, loff_t *ppos)
8778 {
8779 	struct zone *zone;
8780 	int old_percpu_pagelist_high_fraction;
8781 	int ret;
8782 
8783 	mutex_lock(&pcp_batch_high_lock);
8784 	old_percpu_pagelist_high_fraction = percpu_pagelist_high_fraction;
8785 
8786 	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
8787 	if (!write || ret < 0)
8788 		goto out;
8789 
8790 	/* Sanity checking to avoid pcp imbalance */
8791 	if (percpu_pagelist_high_fraction &&
8792 	    percpu_pagelist_high_fraction < MIN_PERCPU_PAGELIST_HIGH_FRACTION) {
8793 		percpu_pagelist_high_fraction = old_percpu_pagelist_high_fraction;
8794 		ret = -EINVAL;
8795 		goto out;
8796 	}
8797 
8798 	/* No change? */
8799 	if (percpu_pagelist_high_fraction == old_percpu_pagelist_high_fraction)
8800 		goto out;
8801 
8802 	for_each_populated_zone(zone)
8803 		zone_set_pageset_high_and_batch(zone, 0);
8804 out:
8805 	mutex_unlock(&pcp_batch_high_lock);
8806 	return ret;
8807 }
8808 
8809 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
8810 /*
8811  * Returns the number of pages that arch has reserved but
8812  * is not known to alloc_large_system_hash().
8813  */
8814 static unsigned long __init arch_reserved_kernel_pages(void)
8815 {
8816 	return 0;
8817 }
8818 #endif
8819 
8820 /*
8821  * Adaptive scale is meant to reduce sizes of hash tables on large memory
8822  * machines. As memory size is increased the scale is also increased but at
8823  * slower pace.  Starting from ADAPT_SCALE_BASE (64G), every time memory
8824  * quadruples the scale is increased by one, which means the size of hash table
8825  * only doubles, instead of quadrupling as well.
8826  * Because 32-bit systems cannot have large physical memory, where this scaling
8827  * makes sense, it is disabled on such platforms.
8828  */
8829 #if __BITS_PER_LONG > 32
8830 #define ADAPT_SCALE_BASE	(64ul << 30)
8831 #define ADAPT_SCALE_SHIFT	2
8832 #define ADAPT_SCALE_NPAGES	(ADAPT_SCALE_BASE >> PAGE_SHIFT)
8833 #endif
8834 
8835 /*
8836  * allocate a large system hash table from bootmem
8837  * - it is assumed that the hash table must contain an exact power-of-2
8838  *   quantity of entries
8839  * - limit is the number of hash buckets, not the total allocation size
8840  */
8841 void *__init alloc_large_system_hash(const char *tablename,
8842 				     unsigned long bucketsize,
8843 				     unsigned long numentries,
8844 				     int scale,
8845 				     int flags,
8846 				     unsigned int *_hash_shift,
8847 				     unsigned int *_hash_mask,
8848 				     unsigned long low_limit,
8849 				     unsigned long high_limit)
8850 {
8851 	unsigned long long max = high_limit;
8852 	unsigned long log2qty, size;
8853 	void *table = NULL;
8854 	gfp_t gfp_flags;
8855 	bool virt;
8856 	bool huge;
8857 
8858 	/* allow the kernel cmdline to have a say */
8859 	if (!numentries) {
8860 		/* round applicable memory size up to nearest megabyte */
8861 		numentries = nr_kernel_pages;
8862 		numentries -= arch_reserved_kernel_pages();
8863 
8864 		/* It isn't necessary when PAGE_SIZE >= 1MB */
8865 		if (PAGE_SHIFT < 20)
8866 			numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
8867 
8868 #if __BITS_PER_LONG > 32
8869 		if (!high_limit) {
8870 			unsigned long adapt;
8871 
8872 			for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
8873 			     adapt <<= ADAPT_SCALE_SHIFT)
8874 				scale++;
8875 		}
8876 #endif
8877 
8878 		/* limit to 1 bucket per 2^scale bytes of low memory */
8879 		if (scale > PAGE_SHIFT)
8880 			numentries >>= (scale - PAGE_SHIFT);
8881 		else
8882 			numentries <<= (PAGE_SHIFT - scale);
8883 
8884 		/* Make sure we've got at least a 0-order allocation.. */
8885 		if (unlikely(flags & HASH_SMALL)) {
8886 			/* Makes no sense without HASH_EARLY */
8887 			WARN_ON(!(flags & HASH_EARLY));
8888 			if (!(numentries >> *_hash_shift)) {
8889 				numentries = 1UL << *_hash_shift;
8890 				BUG_ON(!numentries);
8891 			}
8892 		} else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
8893 			numentries = PAGE_SIZE / bucketsize;
8894 	}
8895 	numentries = roundup_pow_of_two(numentries);
8896 
8897 	/* limit allocation size to 1/16 total memory by default */
8898 	if (max == 0) {
8899 		max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
8900 		do_div(max, bucketsize);
8901 	}
8902 	max = min(max, 0x80000000ULL);
8903 
8904 	if (numentries < low_limit)
8905 		numentries = low_limit;
8906 	if (numentries > max)
8907 		numentries = max;
8908 
8909 	log2qty = ilog2(numentries);
8910 
8911 	gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
8912 	do {
8913 		virt = false;
8914 		size = bucketsize << log2qty;
8915 		if (flags & HASH_EARLY) {
8916 			if (flags & HASH_ZERO)
8917 				table = memblock_alloc(size, SMP_CACHE_BYTES);
8918 			else
8919 				table = memblock_alloc_raw(size,
8920 							   SMP_CACHE_BYTES);
8921 		} else if (get_order(size) >= MAX_ORDER || hashdist) {
8922 			table = vmalloc_huge(size, gfp_flags);
8923 			virt = true;
8924 			if (table)
8925 				huge = is_vm_area_hugepages(table);
8926 		} else {
8927 			/*
8928 			 * If bucketsize is not a power-of-two, we may free
8929 			 * some pages at the end of hash table which
8930 			 * alloc_pages_exact() automatically does
8931 			 */
8932 			table = alloc_pages_exact(size, gfp_flags);
8933 			kmemleak_alloc(table, size, 1, gfp_flags);
8934 		}
8935 	} while (!table && size > PAGE_SIZE && --log2qty);
8936 
8937 	if (!table)
8938 		panic("Failed to allocate %s hash table\n", tablename);
8939 
8940 	pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n",
8941 		tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size,
8942 		virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear");
8943 
8944 	if (_hash_shift)
8945 		*_hash_shift = log2qty;
8946 	if (_hash_mask)
8947 		*_hash_mask = (1 << log2qty) - 1;
8948 
8949 	return table;
8950 }
8951 
8952 /*
8953  * This function checks whether pageblock includes unmovable pages or not.
8954  *
8955  * PageLRU check without isolation or lru_lock could race so that
8956  * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
8957  * check without lock_page also may miss some movable non-lru pages at
8958  * race condition. So you can't expect this function should be exact.
8959  *
8960  * Returns a page without holding a reference. If the caller wants to
8961  * dereference that page (e.g., dumping), it has to make sure that it
8962  * cannot get removed (e.g., via memory unplug) concurrently.
8963  *
8964  */
8965 struct page *has_unmovable_pages(struct zone *zone, struct page *page,
8966 				 int migratetype, int flags)
8967 {
8968 	unsigned long iter = 0;
8969 	unsigned long pfn = page_to_pfn(page);
8970 	unsigned long offset = pfn % pageblock_nr_pages;
8971 
8972 	if (is_migrate_cma_page(page)) {
8973 		/*
8974 		 * CMA allocations (alloc_contig_range) really need to mark
8975 		 * isolate CMA pageblocks even when they are not movable in fact
8976 		 * so consider them movable here.
8977 		 */
8978 		if (is_migrate_cma(migratetype))
8979 			return NULL;
8980 
8981 		return page;
8982 	}
8983 
8984 	for (; iter < pageblock_nr_pages - offset; iter++) {
8985 		page = pfn_to_page(pfn + iter);
8986 
8987 		/*
8988 		 * Both, bootmem allocations and memory holes are marked
8989 		 * PG_reserved and are unmovable. We can even have unmovable
8990 		 * allocations inside ZONE_MOVABLE, for example when
8991 		 * specifying "movablecore".
8992 		 */
8993 		if (PageReserved(page))
8994 			return page;
8995 
8996 		/*
8997 		 * If the zone is movable and we have ruled out all reserved
8998 		 * pages then it should be reasonably safe to assume the rest
8999 		 * is movable.
9000 		 */
9001 		if (zone_idx(zone) == ZONE_MOVABLE)
9002 			continue;
9003 
9004 		/*
9005 		 * Hugepages are not in LRU lists, but they're movable.
9006 		 * THPs are on the LRU, but need to be counted as #small pages.
9007 		 * We need not scan over tail pages because we don't
9008 		 * handle each tail page individually in migration.
9009 		 */
9010 		if (PageHuge(page) || PageTransCompound(page)) {
9011 			struct page *head = compound_head(page);
9012 			unsigned int skip_pages;
9013 
9014 			if (PageHuge(page)) {
9015 				if (!hugepage_migration_supported(page_hstate(head)))
9016 					return page;
9017 			} else if (!PageLRU(head) && !__PageMovable(head)) {
9018 				return page;
9019 			}
9020 
9021 			skip_pages = compound_nr(head) - (page - head);
9022 			iter += skip_pages - 1;
9023 			continue;
9024 		}
9025 
9026 		/*
9027 		 * We can't use page_count without pin a page
9028 		 * because another CPU can free compound page.
9029 		 * This check already skips compound tails of THP
9030 		 * because their page->_refcount is zero at all time.
9031 		 */
9032 		if (!page_ref_count(page)) {
9033 			if (PageBuddy(page))
9034 				iter += (1 << buddy_order(page)) - 1;
9035 			continue;
9036 		}
9037 
9038 		/*
9039 		 * The HWPoisoned page may be not in buddy system, and
9040 		 * page_count() is not 0.
9041 		 */
9042 		if ((flags & MEMORY_OFFLINE) && PageHWPoison(page))
9043 			continue;
9044 
9045 		/*
9046 		 * We treat all PageOffline() pages as movable when offlining
9047 		 * to give drivers a chance to decrement their reference count
9048 		 * in MEM_GOING_OFFLINE in order to indicate that these pages
9049 		 * can be offlined as there are no direct references anymore.
9050 		 * For actually unmovable PageOffline() where the driver does
9051 		 * not support this, we will fail later when trying to actually
9052 		 * move these pages that still have a reference count > 0.
9053 		 * (false negatives in this function only)
9054 		 */
9055 		if ((flags & MEMORY_OFFLINE) && PageOffline(page))
9056 			continue;
9057 
9058 		if (__PageMovable(page) || PageLRU(page))
9059 			continue;
9060 
9061 		/*
9062 		 * If there are RECLAIMABLE pages, we need to check
9063 		 * it.  But now, memory offline itself doesn't call
9064 		 * shrink_node_slabs() and it still to be fixed.
9065 		 */
9066 		return page;
9067 	}
9068 	return NULL;
9069 }
9070 
9071 #ifdef CONFIG_CONTIG_ALLOC
9072 static unsigned long pfn_max_align_down(unsigned long pfn)
9073 {
9074 	return ALIGN_DOWN(pfn, MAX_ORDER_NR_PAGES);
9075 }
9076 
9077 static unsigned long pfn_max_align_up(unsigned long pfn)
9078 {
9079 	return ALIGN(pfn, MAX_ORDER_NR_PAGES);
9080 }
9081 
9082 #if defined(CONFIG_DYNAMIC_DEBUG) || \
9083 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
9084 /* Usage: See admin-guide/dynamic-debug-howto.rst */
9085 static void alloc_contig_dump_pages(struct list_head *page_list)
9086 {
9087 	DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, "migrate failure");
9088 
9089 	if (DYNAMIC_DEBUG_BRANCH(descriptor)) {
9090 		struct page *page;
9091 
9092 		dump_stack();
9093 		list_for_each_entry(page, page_list, lru)
9094 			dump_page(page, "migration failure");
9095 	}
9096 }
9097 #else
9098 static inline void alloc_contig_dump_pages(struct list_head *page_list)
9099 {
9100 }
9101 #endif
9102 
9103 /* [start, end) must belong to a single zone. */
9104 static int __alloc_contig_migrate_range(struct compact_control *cc,
9105 					unsigned long start, unsigned long end)
9106 {
9107 	/* This function is based on compact_zone() from compaction.c. */
9108 	unsigned int nr_reclaimed;
9109 	unsigned long pfn = start;
9110 	unsigned int tries = 0;
9111 	int ret = 0;
9112 	struct migration_target_control mtc = {
9113 		.nid = zone_to_nid(cc->zone),
9114 		.gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
9115 	};
9116 
9117 	lru_cache_disable();
9118 
9119 	while (pfn < end || !list_empty(&cc->migratepages)) {
9120 		if (fatal_signal_pending(current)) {
9121 			ret = -EINTR;
9122 			break;
9123 		}
9124 
9125 		if (list_empty(&cc->migratepages)) {
9126 			cc->nr_migratepages = 0;
9127 			ret = isolate_migratepages_range(cc, pfn, end);
9128 			if (ret && ret != -EAGAIN)
9129 				break;
9130 			pfn = cc->migrate_pfn;
9131 			tries = 0;
9132 		} else if (++tries == 5) {
9133 			ret = -EBUSY;
9134 			break;
9135 		}
9136 
9137 		nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
9138 							&cc->migratepages);
9139 		cc->nr_migratepages -= nr_reclaimed;
9140 
9141 		ret = migrate_pages(&cc->migratepages, alloc_migration_target,
9142 			NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE, NULL);
9143 
9144 		/*
9145 		 * On -ENOMEM, migrate_pages() bails out right away. It is pointless
9146 		 * to retry again over this error, so do the same here.
9147 		 */
9148 		if (ret == -ENOMEM)
9149 			break;
9150 	}
9151 
9152 	lru_cache_enable();
9153 	if (ret < 0) {
9154 		if (ret == -EBUSY)
9155 			alloc_contig_dump_pages(&cc->migratepages);
9156 		putback_movable_pages(&cc->migratepages);
9157 		return ret;
9158 	}
9159 	return 0;
9160 }
9161 
9162 /**
9163  * alloc_contig_range() -- tries to allocate given range of pages
9164  * @start:	start PFN to allocate
9165  * @end:	one-past-the-last PFN to allocate
9166  * @migratetype:	migratetype of the underlying pageblocks (either
9167  *			#MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
9168  *			in range must have the same migratetype and it must
9169  *			be either of the two.
9170  * @gfp_mask:	GFP mask to use during compaction
9171  *
9172  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
9173  * aligned.  The PFN range must belong to a single zone.
9174  *
9175  * The first thing this routine does is attempt to MIGRATE_ISOLATE all
9176  * pageblocks in the range.  Once isolated, the pageblocks should not
9177  * be modified by others.
9178  *
9179  * Return: zero on success or negative error code.  On success all
9180  * pages which PFN is in [start, end) are allocated for the caller and
9181  * need to be freed with free_contig_range().
9182  */
9183 int alloc_contig_range(unsigned long start, unsigned long end,
9184 		       unsigned migratetype, gfp_t gfp_mask)
9185 {
9186 	unsigned long outer_start, outer_end;
9187 	unsigned int order;
9188 	int ret = 0;
9189 
9190 	struct compact_control cc = {
9191 		.nr_migratepages = 0,
9192 		.order = -1,
9193 		.zone = page_zone(pfn_to_page(start)),
9194 		.mode = MIGRATE_SYNC,
9195 		.ignore_skip_hint = true,
9196 		.no_set_skip_hint = true,
9197 		.gfp_mask = current_gfp_context(gfp_mask),
9198 		.alloc_contig = true,
9199 	};
9200 	INIT_LIST_HEAD(&cc.migratepages);
9201 
9202 	/*
9203 	 * What we do here is we mark all pageblocks in range as
9204 	 * MIGRATE_ISOLATE.  Because pageblock and max order pages may
9205 	 * have different sizes, and due to the way page allocator
9206 	 * work, we align the range to biggest of the two pages so
9207 	 * that page allocator won't try to merge buddies from
9208 	 * different pageblocks and change MIGRATE_ISOLATE to some
9209 	 * other migration type.
9210 	 *
9211 	 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
9212 	 * migrate the pages from an unaligned range (ie. pages that
9213 	 * we are interested in).  This will put all the pages in
9214 	 * range back to page allocator as MIGRATE_ISOLATE.
9215 	 *
9216 	 * When this is done, we take the pages in range from page
9217 	 * allocator removing them from the buddy system.  This way
9218 	 * page allocator will never consider using them.
9219 	 *
9220 	 * This lets us mark the pageblocks back as
9221 	 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
9222 	 * aligned range but not in the unaligned, original range are
9223 	 * put back to page allocator so that buddy can use them.
9224 	 */
9225 
9226 	ret = start_isolate_page_range(pfn_max_align_down(start),
9227 				       pfn_max_align_up(end), migratetype, 0);
9228 	if (ret)
9229 		return ret;
9230 
9231 	drain_all_pages(cc.zone);
9232 
9233 	/*
9234 	 * In case of -EBUSY, we'd like to know which page causes problem.
9235 	 * So, just fall through. test_pages_isolated() has a tracepoint
9236 	 * which will report the busy page.
9237 	 *
9238 	 * It is possible that busy pages could become available before
9239 	 * the call to test_pages_isolated, and the range will actually be
9240 	 * allocated.  So, if we fall through be sure to clear ret so that
9241 	 * -EBUSY is not accidentally used or returned to caller.
9242 	 */
9243 	ret = __alloc_contig_migrate_range(&cc, start, end);
9244 	if (ret && ret != -EBUSY)
9245 		goto done;
9246 	ret = 0;
9247 
9248 	/*
9249 	 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
9250 	 * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
9251 	 * more, all pages in [start, end) are free in page allocator.
9252 	 * What we are going to do is to allocate all pages from
9253 	 * [start, end) (that is remove them from page allocator).
9254 	 *
9255 	 * The only problem is that pages at the beginning and at the
9256 	 * end of interesting range may be not aligned with pages that
9257 	 * page allocator holds, ie. they can be part of higher order
9258 	 * pages.  Because of this, we reserve the bigger range and
9259 	 * once this is done free the pages we are not interested in.
9260 	 *
9261 	 * We don't have to hold zone->lock here because the pages are
9262 	 * isolated thus they won't get removed from buddy.
9263 	 */
9264 
9265 	order = 0;
9266 	outer_start = start;
9267 	while (!PageBuddy(pfn_to_page(outer_start))) {
9268 		if (++order >= MAX_ORDER) {
9269 			outer_start = start;
9270 			break;
9271 		}
9272 		outer_start &= ~0UL << order;
9273 	}
9274 
9275 	if (outer_start != start) {
9276 		order = buddy_order(pfn_to_page(outer_start));
9277 
9278 		/*
9279 		 * outer_start page could be small order buddy page and
9280 		 * it doesn't include start page. Adjust outer_start
9281 		 * in this case to report failed page properly
9282 		 * on tracepoint in test_pages_isolated()
9283 		 */
9284 		if (outer_start + (1UL << order) <= start)
9285 			outer_start = start;
9286 	}
9287 
9288 	/* Make sure the range is really isolated. */
9289 	if (test_pages_isolated(outer_start, end, 0)) {
9290 		ret = -EBUSY;
9291 		goto done;
9292 	}
9293 
9294 	/* Grab isolated pages from freelists. */
9295 	outer_end = isolate_freepages_range(&cc, outer_start, end);
9296 	if (!outer_end) {
9297 		ret = -EBUSY;
9298 		goto done;
9299 	}
9300 
9301 	/* Free head and tail (if any) */
9302 	if (start != outer_start)
9303 		free_contig_range(outer_start, start - outer_start);
9304 	if (end != outer_end)
9305 		free_contig_range(end, outer_end - end);
9306 
9307 done:
9308 	undo_isolate_page_range(pfn_max_align_down(start),
9309 				pfn_max_align_up(end), migratetype);
9310 	return ret;
9311 }
9312 EXPORT_SYMBOL(alloc_contig_range);
9313 
9314 static int __alloc_contig_pages(unsigned long start_pfn,
9315 				unsigned long nr_pages, gfp_t gfp_mask)
9316 {
9317 	unsigned long end_pfn = start_pfn + nr_pages;
9318 
9319 	return alloc_contig_range(start_pfn, end_pfn, MIGRATE_MOVABLE,
9320 				  gfp_mask);
9321 }
9322 
9323 static bool pfn_range_valid_contig(struct zone *z, unsigned long start_pfn,
9324 				   unsigned long nr_pages)
9325 {
9326 	unsigned long i, end_pfn = start_pfn + nr_pages;
9327 	struct page *page;
9328 
9329 	for (i = start_pfn; i < end_pfn; i++) {
9330 		page = pfn_to_online_page(i);
9331 		if (!page)
9332 			return false;
9333 
9334 		if (page_zone(page) != z)
9335 			return false;
9336 
9337 		if (PageReserved(page))
9338 			return false;
9339 	}
9340 	return true;
9341 }
9342 
9343 static bool zone_spans_last_pfn(const struct zone *zone,
9344 				unsigned long start_pfn, unsigned long nr_pages)
9345 {
9346 	unsigned long last_pfn = start_pfn + nr_pages - 1;
9347 
9348 	return zone_spans_pfn(zone, last_pfn);
9349 }
9350 
9351 /**
9352  * alloc_contig_pages() -- tries to find and allocate contiguous range of pages
9353  * @nr_pages:	Number of contiguous pages to allocate
9354  * @gfp_mask:	GFP mask to limit search and used during compaction
9355  * @nid:	Target node
9356  * @nodemask:	Mask for other possible nodes
9357  *
9358  * This routine is a wrapper around alloc_contig_range(). It scans over zones
9359  * on an applicable zonelist to find a contiguous pfn range which can then be
9360  * tried for allocation with alloc_contig_range(). This routine is intended
9361  * for allocation requests which can not be fulfilled with the buddy allocator.
9362  *
9363  * The allocated memory is always aligned to a page boundary. If nr_pages is a
9364  * power of two, then allocated range is also guaranteed to be aligned to same
9365  * nr_pages (e.g. 1GB request would be aligned to 1GB).
9366  *
9367  * Allocated pages can be freed with free_contig_range() or by manually calling
9368  * __free_page() on each allocated page.
9369  *
9370  * Return: pointer to contiguous pages on success, or NULL if not successful.
9371  */
9372 struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
9373 				int nid, nodemask_t *nodemask)
9374 {
9375 	unsigned long ret, pfn, flags;
9376 	struct zonelist *zonelist;
9377 	struct zone *zone;
9378 	struct zoneref *z;
9379 
9380 	zonelist = node_zonelist(nid, gfp_mask);
9381 	for_each_zone_zonelist_nodemask(zone, z, zonelist,
9382 					gfp_zone(gfp_mask), nodemask) {
9383 		spin_lock_irqsave(&zone->lock, flags);
9384 
9385 		pfn = ALIGN(zone->zone_start_pfn, nr_pages);
9386 		while (zone_spans_last_pfn(zone, pfn, nr_pages)) {
9387 			if (pfn_range_valid_contig(zone, pfn, nr_pages)) {
9388 				/*
9389 				 * We release the zone lock here because
9390 				 * alloc_contig_range() will also lock the zone
9391 				 * at some point. If there's an allocation
9392 				 * spinning on this lock, it may win the race
9393 				 * and cause alloc_contig_range() to fail...
9394 				 */
9395 				spin_unlock_irqrestore(&zone->lock, flags);
9396 				ret = __alloc_contig_pages(pfn, nr_pages,
9397 							gfp_mask);
9398 				if (!ret)
9399 					return pfn_to_page(pfn);
9400 				spin_lock_irqsave(&zone->lock, flags);
9401 			}
9402 			pfn += nr_pages;
9403 		}
9404 		spin_unlock_irqrestore(&zone->lock, flags);
9405 	}
9406 	return NULL;
9407 }
9408 #endif /* CONFIG_CONTIG_ALLOC */
9409 
9410 void free_contig_range(unsigned long pfn, unsigned long nr_pages)
9411 {
9412 	unsigned long count = 0;
9413 
9414 	for (; nr_pages--; pfn++) {
9415 		struct page *page = pfn_to_page(pfn);
9416 
9417 		count += page_count(page) != 1;
9418 		__free_page(page);
9419 	}
9420 	WARN(count != 0, "%lu pages are still in use!\n", count);
9421 }
9422 EXPORT_SYMBOL(free_contig_range);
9423 
9424 /*
9425  * The zone indicated has a new number of managed_pages; batch sizes and percpu
9426  * page high values need to be recalculated.
9427  */
9428 void zone_pcp_update(struct zone *zone, int cpu_online)
9429 {
9430 	mutex_lock(&pcp_batch_high_lock);
9431 	zone_set_pageset_high_and_batch(zone, cpu_online);
9432 	mutex_unlock(&pcp_batch_high_lock);
9433 }
9434 
9435 /*
9436  * Effectively disable pcplists for the zone by setting the high limit to 0
9437  * and draining all cpus. A concurrent page freeing on another CPU that's about
9438  * to put the page on pcplist will either finish before the drain and the page
9439  * will be drained, or observe the new high limit and skip the pcplist.
9440  *
9441  * Must be paired with a call to zone_pcp_enable().
9442  */
9443 void zone_pcp_disable(struct zone *zone)
9444 {
9445 	mutex_lock(&pcp_batch_high_lock);
9446 	__zone_set_pageset_high_and_batch(zone, 0, 1);
9447 	__drain_all_pages(zone, true);
9448 }
9449 
9450 void zone_pcp_enable(struct zone *zone)
9451 {
9452 	__zone_set_pageset_high_and_batch(zone, zone->pageset_high, zone->pageset_batch);
9453 	mutex_unlock(&pcp_batch_high_lock);
9454 }
9455 
9456 void zone_pcp_reset(struct zone *zone)
9457 {
9458 	int cpu;
9459 	struct per_cpu_zonestat *pzstats;
9460 
9461 	if (zone->per_cpu_pageset != &boot_pageset) {
9462 		for_each_online_cpu(cpu) {
9463 			pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu);
9464 			drain_zonestat(zone, pzstats);
9465 		}
9466 		free_percpu(zone->per_cpu_pageset);
9467 		free_percpu(zone->per_cpu_zonestats);
9468 		zone->per_cpu_pageset = &boot_pageset;
9469 		zone->per_cpu_zonestats = &boot_zonestats;
9470 	}
9471 }
9472 
9473 #ifdef CONFIG_MEMORY_HOTREMOVE
9474 /*
9475  * All pages in the range must be in a single zone, must not contain holes,
9476  * must span full sections, and must be isolated before calling this function.
9477  */
9478 void __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
9479 {
9480 	unsigned long pfn = start_pfn;
9481 	struct page *page;
9482 	struct zone *zone;
9483 	unsigned int order;
9484 	unsigned long flags;
9485 
9486 	offline_mem_sections(pfn, end_pfn);
9487 	zone = page_zone(pfn_to_page(pfn));
9488 	spin_lock_irqsave(&zone->lock, flags);
9489 	while (pfn < end_pfn) {
9490 		page = pfn_to_page(pfn);
9491 		/*
9492 		 * The HWPoisoned page may be not in buddy system, and
9493 		 * page_count() is not 0.
9494 		 */
9495 		if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
9496 			pfn++;
9497 			continue;
9498 		}
9499 		/*
9500 		 * At this point all remaining PageOffline() pages have a
9501 		 * reference count of 0 and can simply be skipped.
9502 		 */
9503 		if (PageOffline(page)) {
9504 			BUG_ON(page_count(page));
9505 			BUG_ON(PageBuddy(page));
9506 			pfn++;
9507 			continue;
9508 		}
9509 
9510 		BUG_ON(page_count(page));
9511 		BUG_ON(!PageBuddy(page));
9512 		order = buddy_order(page);
9513 		del_page_from_free_list(page, zone, order);
9514 		pfn += (1 << order);
9515 	}
9516 	spin_unlock_irqrestore(&zone->lock, flags);
9517 }
9518 #endif
9519 
9520 /*
9521  * This function returns a stable result only if called under zone lock.
9522  */
9523 bool is_free_buddy_page(struct page *page)
9524 {
9525 	unsigned long pfn = page_to_pfn(page);
9526 	unsigned int order;
9527 
9528 	for (order = 0; order < MAX_ORDER; order++) {
9529 		struct page *page_head = page - (pfn & ((1 << order) - 1));
9530 
9531 		if (PageBuddy(page_head) &&
9532 		    buddy_order_unsafe(page_head) >= order)
9533 			break;
9534 	}
9535 
9536 	return order < MAX_ORDER;
9537 }
9538 EXPORT_SYMBOL(is_free_buddy_page);
9539 
9540 #ifdef CONFIG_MEMORY_FAILURE
9541 /*
9542  * Break down a higher-order page in sub-pages, and keep our target out of
9543  * buddy allocator.
9544  */
9545 static void break_down_buddy_pages(struct zone *zone, struct page *page,
9546 				   struct page *target, int low, int high,
9547 				   int migratetype)
9548 {
9549 	unsigned long size = 1 << high;
9550 	struct page *current_buddy, *next_page;
9551 
9552 	while (high > low) {
9553 		high--;
9554 		size >>= 1;
9555 
9556 		if (target >= &page[size]) {
9557 			next_page = page + size;
9558 			current_buddy = page;
9559 		} else {
9560 			next_page = page;
9561 			current_buddy = page + size;
9562 		}
9563 
9564 		if (set_page_guard(zone, current_buddy, high, migratetype))
9565 			continue;
9566 
9567 		if (current_buddy != target) {
9568 			add_to_free_list(current_buddy, zone, high, migratetype);
9569 			set_buddy_order(current_buddy, high);
9570 			page = next_page;
9571 		}
9572 	}
9573 }
9574 
9575 /*
9576  * Take a page that will be marked as poisoned off the buddy allocator.
9577  */
9578 bool take_page_off_buddy(struct page *page)
9579 {
9580 	struct zone *zone = page_zone(page);
9581 	unsigned long pfn = page_to_pfn(page);
9582 	unsigned long flags;
9583 	unsigned int order;
9584 	bool ret = false;
9585 
9586 	spin_lock_irqsave(&zone->lock, flags);
9587 	for (order = 0; order < MAX_ORDER; order++) {
9588 		struct page *page_head = page - (pfn & ((1 << order) - 1));
9589 		int page_order = buddy_order(page_head);
9590 
9591 		if (PageBuddy(page_head) && page_order >= order) {
9592 			unsigned long pfn_head = page_to_pfn(page_head);
9593 			int migratetype = get_pfnblock_migratetype(page_head,
9594 								   pfn_head);
9595 
9596 			del_page_from_free_list(page_head, zone, page_order);
9597 			break_down_buddy_pages(zone, page_head, page, 0,
9598 						page_order, migratetype);
9599 			SetPageHWPoisonTakenOff(page);
9600 			if (!is_migrate_isolate(migratetype))
9601 				__mod_zone_freepage_state(zone, -1, migratetype);
9602 			ret = true;
9603 			break;
9604 		}
9605 		if (page_count(page_head) > 0)
9606 			break;
9607 	}
9608 	spin_unlock_irqrestore(&zone->lock, flags);
9609 	return ret;
9610 }
9611 
9612 /*
9613  * Cancel takeoff done by take_page_off_buddy().
9614  */
9615 bool put_page_back_buddy(struct page *page)
9616 {
9617 	struct zone *zone = page_zone(page);
9618 	unsigned long pfn = page_to_pfn(page);
9619 	unsigned long flags;
9620 	int migratetype = get_pfnblock_migratetype(page, pfn);
9621 	bool ret = false;
9622 
9623 	spin_lock_irqsave(&zone->lock, flags);
9624 	if (put_page_testzero(page)) {
9625 		ClearPageHWPoisonTakenOff(page);
9626 		__free_one_page(page, pfn, zone, 0, migratetype, FPI_NONE);
9627 		if (TestClearPageHWPoison(page)) {
9628 			num_poisoned_pages_dec();
9629 			ret = true;
9630 		}
9631 	}
9632 	spin_unlock_irqrestore(&zone->lock, flags);
9633 
9634 	return ret;
9635 }
9636 #endif
9637 
9638 #ifdef CONFIG_ZONE_DMA
9639 bool has_managed_dma(void)
9640 {
9641 	struct pglist_data *pgdat;
9642 
9643 	for_each_online_pgdat(pgdat) {
9644 		struct zone *zone = &pgdat->node_zones[ZONE_DMA];
9645 
9646 		if (managed_zone(zone))
9647 			return true;
9648 	}
9649 	return false;
9650 }
9651 #endif /* CONFIG_ZONE_DMA */
9652