xref: /openbmc/linux/mm/page_alloc.c (revision 18f90d372cf35b387663f1567de701e5393f6eb5)
1 /*
2  *  linux/mm/page_alloc.c
3  *
4  *  Manages the free list, the system allocates free pages here.
5  *  Note that kmalloc() lives in slab.c
6  *
7  *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
8  *  Swap reorganised 29.12.95, Stephen Tweedie
9  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10  *  Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11  *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12  *  Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13  *  Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14  *          (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15  */
16 
17 #include <linux/stddef.h>
18 #include <linux/mm.h>
19 #include <linux/highmem.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/jiffies.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kasan.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/topology.h>
36 #include <linux/sysctl.h>
37 #include <linux/cpu.h>
38 #include <linux/cpuset.h>
39 #include <linux/memory_hotplug.h>
40 #include <linux/nodemask.h>
41 #include <linux/vmalloc.h>
42 #include <linux/vmstat.h>
43 #include <linux/mempolicy.h>
44 #include <linux/memremap.h>
45 #include <linux/stop_machine.h>
46 #include <linux/sort.h>
47 #include <linux/pfn.h>
48 #include <linux/backing-dev.h>
49 #include <linux/fault-inject.h>
50 #include <linux/page-isolation.h>
51 #include <linux/page_ext.h>
52 #include <linux/debugobjects.h>
53 #include <linux/kmemleak.h>
54 #include <linux/compaction.h>
55 #include <trace/events/kmem.h>
56 #include <trace/events/oom.h>
57 #include <linux/prefetch.h>
58 #include <linux/mm_inline.h>
59 #include <linux/migrate.h>
60 #include <linux/hugetlb.h>
61 #include <linux/sched/rt.h>
62 #include <linux/sched/mm.h>
63 #include <linux/page_owner.h>
64 #include <linux/kthread.h>
65 #include <linux/memcontrol.h>
66 #include <linux/ftrace.h>
67 #include <linux/lockdep.h>
68 #include <linux/nmi.h>
69 #include <linux/psi.h>
70 
71 #include <asm/sections.h>
72 #include <asm/tlbflush.h>
73 #include <asm/div64.h>
74 #include "internal.h"
75 
76 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
77 static DEFINE_MUTEX(pcp_batch_high_lock);
78 #define MIN_PERCPU_PAGELIST_FRACTION	(8)
79 
80 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
81 DEFINE_PER_CPU(int, numa_node);
82 EXPORT_PER_CPU_SYMBOL(numa_node);
83 #endif
84 
85 DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key);
86 
87 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
88 /*
89  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
90  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
91  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
92  * defined in <linux/topology.h>.
93  */
94 DEFINE_PER_CPU(int, _numa_mem_);		/* Kernel "local memory" node */
95 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
96 int _node_numa_mem_[MAX_NUMNODES];
97 #endif
98 
99 /* work_structs for global per-cpu drains */
100 struct pcpu_drain {
101 	struct zone *zone;
102 	struct work_struct work;
103 };
104 DEFINE_MUTEX(pcpu_drain_mutex);
105 DEFINE_PER_CPU(struct pcpu_drain, pcpu_drain);
106 
107 #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
108 volatile unsigned long latent_entropy __latent_entropy;
109 EXPORT_SYMBOL(latent_entropy);
110 #endif
111 
112 /*
113  * Array of node states.
114  */
115 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
116 	[N_POSSIBLE] = NODE_MASK_ALL,
117 	[N_ONLINE] = { { [0] = 1UL } },
118 #ifndef CONFIG_NUMA
119 	[N_NORMAL_MEMORY] = { { [0] = 1UL } },
120 #ifdef CONFIG_HIGHMEM
121 	[N_HIGH_MEMORY] = { { [0] = 1UL } },
122 #endif
123 	[N_MEMORY] = { { [0] = 1UL } },
124 	[N_CPU] = { { [0] = 1UL } },
125 #endif	/* NUMA */
126 };
127 EXPORT_SYMBOL(node_states);
128 
129 atomic_long_t _totalram_pages __read_mostly;
130 EXPORT_SYMBOL(_totalram_pages);
131 unsigned long totalreserve_pages __read_mostly;
132 unsigned long totalcma_pages __read_mostly;
133 
134 int percpu_pagelist_fraction;
135 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
136 
137 /*
138  * A cached value of the page's pageblock's migratetype, used when the page is
139  * put on a pcplist. Used to avoid the pageblock migratetype lookup when
140  * freeing from pcplists in most cases, at the cost of possibly becoming stale.
141  * Also the migratetype set in the page does not necessarily match the pcplist
142  * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
143  * other index - this ensures that it will be put on the correct CMA freelist.
144  */
145 static inline int get_pcppage_migratetype(struct page *page)
146 {
147 	return page->index;
148 }
149 
150 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
151 {
152 	page->index = migratetype;
153 }
154 
155 #ifdef CONFIG_PM_SLEEP
156 /*
157  * The following functions are used by the suspend/hibernate code to temporarily
158  * change gfp_allowed_mask in order to avoid using I/O during memory allocations
159  * while devices are suspended.  To avoid races with the suspend/hibernate code,
160  * they should always be called with system_transition_mutex held
161  * (gfp_allowed_mask also should only be modified with system_transition_mutex
162  * held, unless the suspend/hibernate code is guaranteed not to run in parallel
163  * with that modification).
164  */
165 
166 static gfp_t saved_gfp_mask;
167 
168 void pm_restore_gfp_mask(void)
169 {
170 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
171 	if (saved_gfp_mask) {
172 		gfp_allowed_mask = saved_gfp_mask;
173 		saved_gfp_mask = 0;
174 	}
175 }
176 
177 void pm_restrict_gfp_mask(void)
178 {
179 	WARN_ON(!mutex_is_locked(&system_transition_mutex));
180 	WARN_ON(saved_gfp_mask);
181 	saved_gfp_mask = gfp_allowed_mask;
182 	gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
183 }
184 
185 bool pm_suspended_storage(void)
186 {
187 	if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
188 		return false;
189 	return true;
190 }
191 #endif /* CONFIG_PM_SLEEP */
192 
193 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
194 unsigned int pageblock_order __read_mostly;
195 #endif
196 
197 static void __free_pages_ok(struct page *page, unsigned int order);
198 
199 /*
200  * results with 256, 32 in the lowmem_reserve sysctl:
201  *	1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
202  *	1G machine -> (16M dma, 784M normal, 224M high)
203  *	NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
204  *	HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
205  *	HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
206  *
207  * TBD: should special case ZONE_DMA32 machines here - in those we normally
208  * don't need any ZONE_NORMAL reservation
209  */
210 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = {
211 #ifdef CONFIG_ZONE_DMA
212 	[ZONE_DMA] = 256,
213 #endif
214 #ifdef CONFIG_ZONE_DMA32
215 	[ZONE_DMA32] = 256,
216 #endif
217 	[ZONE_NORMAL] = 32,
218 #ifdef CONFIG_HIGHMEM
219 	[ZONE_HIGHMEM] = 0,
220 #endif
221 	[ZONE_MOVABLE] = 0,
222 };
223 
224 EXPORT_SYMBOL(totalram_pages);
225 
226 static char * const zone_names[MAX_NR_ZONES] = {
227 #ifdef CONFIG_ZONE_DMA
228 	 "DMA",
229 #endif
230 #ifdef CONFIG_ZONE_DMA32
231 	 "DMA32",
232 #endif
233 	 "Normal",
234 #ifdef CONFIG_HIGHMEM
235 	 "HighMem",
236 #endif
237 	 "Movable",
238 #ifdef CONFIG_ZONE_DEVICE
239 	 "Device",
240 #endif
241 };
242 
243 const char * const migratetype_names[MIGRATE_TYPES] = {
244 	"Unmovable",
245 	"Movable",
246 	"Reclaimable",
247 	"HighAtomic",
248 #ifdef CONFIG_CMA
249 	"CMA",
250 #endif
251 #ifdef CONFIG_MEMORY_ISOLATION
252 	"Isolate",
253 #endif
254 };
255 
256 compound_page_dtor * const compound_page_dtors[] = {
257 	NULL,
258 	free_compound_page,
259 #ifdef CONFIG_HUGETLB_PAGE
260 	free_huge_page,
261 #endif
262 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
263 	free_transhuge_page,
264 #endif
265 };
266 
267 int min_free_kbytes = 1024;
268 int user_min_free_kbytes = -1;
269 #ifdef CONFIG_DISCONTIGMEM
270 /*
271  * DiscontigMem defines memory ranges as separate pg_data_t even if the ranges
272  * are not on separate NUMA nodes. Functionally this works but with
273  * watermark_boost_factor, it can reclaim prematurely as the ranges can be
274  * quite small. By default, do not boost watermarks on discontigmem as in
275  * many cases very high-order allocations like THP are likely to be
276  * unsupported and the premature reclaim offsets the advantage of long-term
277  * fragmentation avoidance.
278  */
279 int watermark_boost_factor __read_mostly;
280 #else
281 int watermark_boost_factor __read_mostly = 15000;
282 #endif
283 int watermark_scale_factor = 10;
284 
285 static unsigned long nr_kernel_pages __initdata;
286 static unsigned long nr_all_pages __initdata;
287 static unsigned long dma_reserve __initdata;
288 
289 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
290 static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata;
291 static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata;
292 static unsigned long required_kernelcore __initdata;
293 static unsigned long required_kernelcore_percent __initdata;
294 static unsigned long required_movablecore __initdata;
295 static unsigned long required_movablecore_percent __initdata;
296 static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata;
297 static bool mirrored_kernelcore __meminitdata;
298 
299 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
300 int movable_zone;
301 EXPORT_SYMBOL(movable_zone);
302 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
303 
304 #if MAX_NUMNODES > 1
305 unsigned int nr_node_ids __read_mostly = MAX_NUMNODES;
306 unsigned int nr_online_nodes __read_mostly = 1;
307 EXPORT_SYMBOL(nr_node_ids);
308 EXPORT_SYMBOL(nr_online_nodes);
309 #endif
310 
311 int page_group_by_mobility_disabled __read_mostly;
312 
313 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
314 /*
315  * During boot we initialize deferred pages on-demand, as needed, but once
316  * page_alloc_init_late() has finished, the deferred pages are all initialized,
317  * and we can permanently disable that path.
318  */
319 static DEFINE_STATIC_KEY_TRUE(deferred_pages);
320 
321 /*
322  * Calling kasan_free_pages() only after deferred memory initialization
323  * has completed. Poisoning pages during deferred memory init will greatly
324  * lengthen the process and cause problem in large memory systems as the
325  * deferred pages initialization is done with interrupt disabled.
326  *
327  * Assuming that there will be no reference to those newly initialized
328  * pages before they are ever allocated, this should have no effect on
329  * KASAN memory tracking as the poison will be properly inserted at page
330  * allocation time. The only corner case is when pages are allocated by
331  * on-demand allocation and then freed again before the deferred pages
332  * initialization is done, but this is not likely to happen.
333  */
334 static inline void kasan_free_nondeferred_pages(struct page *page, int order)
335 {
336 	if (!static_branch_unlikely(&deferred_pages))
337 		kasan_free_pages(page, order);
338 }
339 
340 /* Returns true if the struct page for the pfn is uninitialised */
341 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
342 {
343 	int nid = early_pfn_to_nid(pfn);
344 
345 	if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
346 		return true;
347 
348 	return false;
349 }
350 
351 /*
352  * Returns true when the remaining initialisation should be deferred until
353  * later in the boot cycle when it can be parallelised.
354  */
355 static bool __meminit
356 defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
357 {
358 	static unsigned long prev_end_pfn, nr_initialised;
359 
360 	/*
361 	 * prev_end_pfn static that contains the end of previous zone
362 	 * No need to protect because called very early in boot before smp_init.
363 	 */
364 	if (prev_end_pfn != end_pfn) {
365 		prev_end_pfn = end_pfn;
366 		nr_initialised = 0;
367 	}
368 
369 	/* Always populate low zones for address-constrained allocations */
370 	if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
371 		return false;
372 
373 	/*
374 	 * We start only with one section of pages, more pages are added as
375 	 * needed until the rest of deferred pages are initialized.
376 	 */
377 	nr_initialised++;
378 	if ((nr_initialised > PAGES_PER_SECTION) &&
379 	    (pfn & (PAGES_PER_SECTION - 1)) == 0) {
380 		NODE_DATA(nid)->first_deferred_pfn = pfn;
381 		return true;
382 	}
383 	return false;
384 }
385 #else
386 #define kasan_free_nondeferred_pages(p, o)	kasan_free_pages(p, o)
387 
388 static inline bool early_page_uninitialised(unsigned long pfn)
389 {
390 	return false;
391 }
392 
393 static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
394 {
395 	return false;
396 }
397 #endif
398 
399 /* Return a pointer to the bitmap storing bits affecting a block of pages */
400 static inline unsigned long *get_pageblock_bitmap(struct page *page,
401 							unsigned long pfn)
402 {
403 #ifdef CONFIG_SPARSEMEM
404 	return __pfn_to_section(pfn)->pageblock_flags;
405 #else
406 	return page_zone(page)->pageblock_flags;
407 #endif /* CONFIG_SPARSEMEM */
408 }
409 
410 static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
411 {
412 #ifdef CONFIG_SPARSEMEM
413 	pfn &= (PAGES_PER_SECTION-1);
414 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
415 #else
416 	pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
417 	return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
418 #endif /* CONFIG_SPARSEMEM */
419 }
420 
421 /**
422  * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
423  * @page: The page within the block of interest
424  * @pfn: The target page frame number
425  * @end_bitidx: The last bit of interest to retrieve
426  * @mask: mask of bits that the caller is interested in
427  *
428  * Return: pageblock_bits flags
429  */
430 static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
431 					unsigned long pfn,
432 					unsigned long end_bitidx,
433 					unsigned long mask)
434 {
435 	unsigned long *bitmap;
436 	unsigned long bitidx, word_bitidx;
437 	unsigned long word;
438 
439 	bitmap = get_pageblock_bitmap(page, pfn);
440 	bitidx = pfn_to_bitidx(page, pfn);
441 	word_bitidx = bitidx / BITS_PER_LONG;
442 	bitidx &= (BITS_PER_LONG-1);
443 
444 	word = bitmap[word_bitidx];
445 	bitidx += end_bitidx;
446 	return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
447 }
448 
449 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
450 					unsigned long end_bitidx,
451 					unsigned long mask)
452 {
453 	return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
454 }
455 
456 static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
457 {
458 	return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
459 }
460 
461 /**
462  * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
463  * @page: The page within the block of interest
464  * @flags: The flags to set
465  * @pfn: The target page frame number
466  * @end_bitidx: The last bit of interest
467  * @mask: mask of bits that the caller is interested in
468  */
469 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
470 					unsigned long pfn,
471 					unsigned long end_bitidx,
472 					unsigned long mask)
473 {
474 	unsigned long *bitmap;
475 	unsigned long bitidx, word_bitidx;
476 	unsigned long old_word, word;
477 
478 	BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
479 	BUILD_BUG_ON(MIGRATE_TYPES > (1 << PB_migratetype_bits));
480 
481 	bitmap = get_pageblock_bitmap(page, pfn);
482 	bitidx = pfn_to_bitidx(page, pfn);
483 	word_bitidx = bitidx / BITS_PER_LONG;
484 	bitidx &= (BITS_PER_LONG-1);
485 
486 	VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
487 
488 	bitidx += end_bitidx;
489 	mask <<= (BITS_PER_LONG - bitidx - 1);
490 	flags <<= (BITS_PER_LONG - bitidx - 1);
491 
492 	word = READ_ONCE(bitmap[word_bitidx]);
493 	for (;;) {
494 		old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
495 		if (word == old_word)
496 			break;
497 		word = old_word;
498 	}
499 }
500 
501 void set_pageblock_migratetype(struct page *page, int migratetype)
502 {
503 	if (unlikely(page_group_by_mobility_disabled &&
504 		     migratetype < MIGRATE_PCPTYPES))
505 		migratetype = MIGRATE_UNMOVABLE;
506 
507 	set_pageblock_flags_group(page, (unsigned long)migratetype,
508 					PB_migrate, PB_migrate_end);
509 }
510 
511 #ifdef CONFIG_DEBUG_VM
512 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
513 {
514 	int ret = 0;
515 	unsigned seq;
516 	unsigned long pfn = page_to_pfn(page);
517 	unsigned long sp, start_pfn;
518 
519 	do {
520 		seq = zone_span_seqbegin(zone);
521 		start_pfn = zone->zone_start_pfn;
522 		sp = zone->spanned_pages;
523 		if (!zone_spans_pfn(zone, pfn))
524 			ret = 1;
525 	} while (zone_span_seqretry(zone, seq));
526 
527 	if (ret)
528 		pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
529 			pfn, zone_to_nid(zone), zone->name,
530 			start_pfn, start_pfn + sp);
531 
532 	return ret;
533 }
534 
535 static int page_is_consistent(struct zone *zone, struct page *page)
536 {
537 	if (!pfn_valid_within(page_to_pfn(page)))
538 		return 0;
539 	if (zone != page_zone(page))
540 		return 0;
541 
542 	return 1;
543 }
544 /*
545  * Temporary debugging check for pages not lying within a given zone.
546  */
547 static int __maybe_unused bad_range(struct zone *zone, struct page *page)
548 {
549 	if (page_outside_zone_boundaries(zone, page))
550 		return 1;
551 	if (!page_is_consistent(zone, page))
552 		return 1;
553 
554 	return 0;
555 }
556 #else
557 static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
558 {
559 	return 0;
560 }
561 #endif
562 
563 static void bad_page(struct page *page, const char *reason,
564 		unsigned long bad_flags)
565 {
566 	static unsigned long resume;
567 	static unsigned long nr_shown;
568 	static unsigned long nr_unshown;
569 
570 	/*
571 	 * Allow a burst of 60 reports, then keep quiet for that minute;
572 	 * or allow a steady drip of one report per second.
573 	 */
574 	if (nr_shown == 60) {
575 		if (time_before(jiffies, resume)) {
576 			nr_unshown++;
577 			goto out;
578 		}
579 		if (nr_unshown) {
580 			pr_alert(
581 			      "BUG: Bad page state: %lu messages suppressed\n",
582 				nr_unshown);
583 			nr_unshown = 0;
584 		}
585 		nr_shown = 0;
586 	}
587 	if (nr_shown++ == 0)
588 		resume = jiffies + 60 * HZ;
589 
590 	pr_alert("BUG: Bad page state in process %s  pfn:%05lx\n",
591 		current->comm, page_to_pfn(page));
592 	__dump_page(page, reason);
593 	bad_flags &= page->flags;
594 	if (bad_flags)
595 		pr_alert("bad because of flags: %#lx(%pGp)\n",
596 						bad_flags, &bad_flags);
597 	dump_page_owner(page);
598 
599 	print_modules();
600 	dump_stack();
601 out:
602 	/* Leave bad fields for debug, except PageBuddy could make trouble */
603 	page_mapcount_reset(page); /* remove PageBuddy */
604 	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
605 }
606 
607 /*
608  * Higher-order pages are called "compound pages".  They are structured thusly:
609  *
610  * The first PAGE_SIZE page is called the "head page" and have PG_head set.
611  *
612  * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
613  * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
614  *
615  * The first tail page's ->compound_dtor holds the offset in array of compound
616  * page destructors. See compound_page_dtors.
617  *
618  * The first tail page's ->compound_order holds the order of allocation.
619  * This usage means that zero-order pages may not be compound.
620  */
621 
622 void free_compound_page(struct page *page)
623 {
624 	__free_pages_ok(page, compound_order(page));
625 }
626 
627 void prep_compound_page(struct page *page, unsigned int order)
628 {
629 	int i;
630 	int nr_pages = 1 << order;
631 
632 	set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
633 	set_compound_order(page, order);
634 	__SetPageHead(page);
635 	for (i = 1; i < nr_pages; i++) {
636 		struct page *p = page + i;
637 		set_page_count(p, 0);
638 		p->mapping = TAIL_MAPPING;
639 		set_compound_head(p, page);
640 	}
641 	atomic_set(compound_mapcount_ptr(page), -1);
642 }
643 
644 #ifdef CONFIG_DEBUG_PAGEALLOC
645 unsigned int _debug_guardpage_minorder;
646 bool _debug_pagealloc_enabled __read_mostly
647 			= IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
648 EXPORT_SYMBOL(_debug_pagealloc_enabled);
649 bool _debug_guardpage_enabled __read_mostly;
650 
651 static int __init early_debug_pagealloc(char *buf)
652 {
653 	if (!buf)
654 		return -EINVAL;
655 	return kstrtobool(buf, &_debug_pagealloc_enabled);
656 }
657 early_param("debug_pagealloc", early_debug_pagealloc);
658 
659 static bool need_debug_guardpage(void)
660 {
661 	/* If we don't use debug_pagealloc, we don't need guard page */
662 	if (!debug_pagealloc_enabled())
663 		return false;
664 
665 	if (!debug_guardpage_minorder())
666 		return false;
667 
668 	return true;
669 }
670 
671 static void init_debug_guardpage(void)
672 {
673 	if (!debug_pagealloc_enabled())
674 		return;
675 
676 	if (!debug_guardpage_minorder())
677 		return;
678 
679 	_debug_guardpage_enabled = true;
680 }
681 
682 struct page_ext_operations debug_guardpage_ops = {
683 	.need = need_debug_guardpage,
684 	.init = init_debug_guardpage,
685 };
686 
687 static int __init debug_guardpage_minorder_setup(char *buf)
688 {
689 	unsigned long res;
690 
691 	if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
692 		pr_err("Bad debug_guardpage_minorder value\n");
693 		return 0;
694 	}
695 	_debug_guardpage_minorder = res;
696 	pr_info("Setting debug_guardpage_minorder to %lu\n", res);
697 	return 0;
698 }
699 early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
700 
701 static inline bool set_page_guard(struct zone *zone, struct page *page,
702 				unsigned int order, int migratetype)
703 {
704 	struct page_ext *page_ext;
705 
706 	if (!debug_guardpage_enabled())
707 		return false;
708 
709 	if (order >= debug_guardpage_minorder())
710 		return false;
711 
712 	page_ext = lookup_page_ext(page);
713 	if (unlikely(!page_ext))
714 		return false;
715 
716 	__set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
717 
718 	INIT_LIST_HEAD(&page->lru);
719 	set_page_private(page, order);
720 	/* Guard pages are not available for any usage */
721 	__mod_zone_freepage_state(zone, -(1 << order), migratetype);
722 
723 	return true;
724 }
725 
726 static inline void clear_page_guard(struct zone *zone, struct page *page,
727 				unsigned int order, int migratetype)
728 {
729 	struct page_ext *page_ext;
730 
731 	if (!debug_guardpage_enabled())
732 		return;
733 
734 	page_ext = lookup_page_ext(page);
735 	if (unlikely(!page_ext))
736 		return;
737 
738 	__clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
739 
740 	set_page_private(page, 0);
741 	if (!is_migrate_isolate(migratetype))
742 		__mod_zone_freepage_state(zone, (1 << order), migratetype);
743 }
744 #else
745 struct page_ext_operations debug_guardpage_ops;
746 static inline bool set_page_guard(struct zone *zone, struct page *page,
747 			unsigned int order, int migratetype) { return false; }
748 static inline void clear_page_guard(struct zone *zone, struct page *page,
749 				unsigned int order, int migratetype) {}
750 #endif
751 
752 static inline void set_page_order(struct page *page, unsigned int order)
753 {
754 	set_page_private(page, order);
755 	__SetPageBuddy(page);
756 }
757 
758 static inline void rmv_page_order(struct page *page)
759 {
760 	__ClearPageBuddy(page);
761 	set_page_private(page, 0);
762 }
763 
764 /*
765  * This function checks whether a page is free && is the buddy
766  * we can coalesce a page and its buddy if
767  * (a) the buddy is not in a hole (check before calling!) &&
768  * (b) the buddy is in the buddy system &&
769  * (c) a page and its buddy have the same order &&
770  * (d) a page and its buddy are in the same zone.
771  *
772  * For recording whether a page is in the buddy system, we set PageBuddy.
773  * Setting, clearing, and testing PageBuddy is serialized by zone->lock.
774  *
775  * For recording page's order, we use page_private(page).
776  */
777 static inline int page_is_buddy(struct page *page, struct page *buddy,
778 							unsigned int order)
779 {
780 	if (page_is_guard(buddy) && page_order(buddy) == order) {
781 		if (page_zone_id(page) != page_zone_id(buddy))
782 			return 0;
783 
784 		VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
785 
786 		return 1;
787 	}
788 
789 	if (PageBuddy(buddy) && page_order(buddy) == order) {
790 		/*
791 		 * zone check is done late to avoid uselessly
792 		 * calculating zone/node ids for pages that could
793 		 * never merge.
794 		 */
795 		if (page_zone_id(page) != page_zone_id(buddy))
796 			return 0;
797 
798 		VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
799 
800 		return 1;
801 	}
802 	return 0;
803 }
804 
805 #ifdef CONFIG_COMPACTION
806 static inline struct capture_control *task_capc(struct zone *zone)
807 {
808 	struct capture_control *capc = current->capture_control;
809 
810 	return capc &&
811 		!(current->flags & PF_KTHREAD) &&
812 		!capc->page &&
813 		capc->cc->zone == zone &&
814 		capc->cc->direct_compaction ? capc : NULL;
815 }
816 
817 static inline bool
818 compaction_capture(struct capture_control *capc, struct page *page,
819 		   int order, int migratetype)
820 {
821 	if (!capc || order != capc->cc->order)
822 		return false;
823 
824 	/* Do not accidentally pollute CMA or isolated regions*/
825 	if (is_migrate_cma(migratetype) ||
826 	    is_migrate_isolate(migratetype))
827 		return false;
828 
829 	/*
830 	 * Do not let lower order allocations polluate a movable pageblock.
831 	 * This might let an unmovable request use a reclaimable pageblock
832 	 * and vice-versa but no more than normal fallback logic which can
833 	 * have trouble finding a high-order free page.
834 	 */
835 	if (order < pageblock_order && migratetype == MIGRATE_MOVABLE)
836 		return false;
837 
838 	capc->page = page;
839 	return true;
840 }
841 
842 #else
843 static inline struct capture_control *task_capc(struct zone *zone)
844 {
845 	return NULL;
846 }
847 
848 static inline bool
849 compaction_capture(struct capture_control *capc, struct page *page,
850 		   int order, int migratetype)
851 {
852 	return false;
853 }
854 #endif /* CONFIG_COMPACTION */
855 
856 /*
857  * Freeing function for a buddy system allocator.
858  *
859  * The concept of a buddy system is to maintain direct-mapped table
860  * (containing bit values) for memory blocks of various "orders".
861  * The bottom level table contains the map for the smallest allocatable
862  * units of memory (here, pages), and each level above it describes
863  * pairs of units from the levels below, hence, "buddies".
864  * At a high level, all that happens here is marking the table entry
865  * at the bottom level available, and propagating the changes upward
866  * as necessary, plus some accounting needed to play nicely with other
867  * parts of the VM system.
868  * At each level, we keep a list of pages, which are heads of continuous
869  * free pages of length of (1 << order) and marked with PageBuddy.
870  * Page's order is recorded in page_private(page) field.
871  * So when we are allocating or freeing one, we can derive the state of the
872  * other.  That is, if we allocate a small block, and both were
873  * free, the remainder of the region must be split into blocks.
874  * If a block is freed, and its buddy is also free, then this
875  * triggers coalescing into a block of larger size.
876  *
877  * -- nyc
878  */
879 
880 static inline void __free_one_page(struct page *page,
881 		unsigned long pfn,
882 		struct zone *zone, unsigned int order,
883 		int migratetype)
884 {
885 	unsigned long combined_pfn;
886 	unsigned long uninitialized_var(buddy_pfn);
887 	struct page *buddy;
888 	unsigned int max_order;
889 	struct capture_control *capc = task_capc(zone);
890 
891 	max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
892 
893 	VM_BUG_ON(!zone_is_initialized(zone));
894 	VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
895 
896 	VM_BUG_ON(migratetype == -1);
897 	if (likely(!is_migrate_isolate(migratetype)))
898 		__mod_zone_freepage_state(zone, 1 << order, migratetype);
899 
900 	VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
901 	VM_BUG_ON_PAGE(bad_range(zone, page), page);
902 
903 continue_merging:
904 	while (order < max_order - 1) {
905 		if (compaction_capture(capc, page, order, migratetype)) {
906 			__mod_zone_freepage_state(zone, -(1 << order),
907 								migratetype);
908 			return;
909 		}
910 		buddy_pfn = __find_buddy_pfn(pfn, order);
911 		buddy = page + (buddy_pfn - pfn);
912 
913 		if (!pfn_valid_within(buddy_pfn))
914 			goto done_merging;
915 		if (!page_is_buddy(page, buddy, order))
916 			goto done_merging;
917 		/*
918 		 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
919 		 * merge with it and move up one order.
920 		 */
921 		if (page_is_guard(buddy)) {
922 			clear_page_guard(zone, buddy, order, migratetype);
923 		} else {
924 			list_del(&buddy->lru);
925 			zone->free_area[order].nr_free--;
926 			rmv_page_order(buddy);
927 		}
928 		combined_pfn = buddy_pfn & pfn;
929 		page = page + (combined_pfn - pfn);
930 		pfn = combined_pfn;
931 		order++;
932 	}
933 	if (max_order < MAX_ORDER) {
934 		/* If we are here, it means order is >= pageblock_order.
935 		 * We want to prevent merge between freepages on isolate
936 		 * pageblock and normal pageblock. Without this, pageblock
937 		 * isolation could cause incorrect freepage or CMA accounting.
938 		 *
939 		 * We don't want to hit this code for the more frequent
940 		 * low-order merging.
941 		 */
942 		if (unlikely(has_isolate_pageblock(zone))) {
943 			int buddy_mt;
944 
945 			buddy_pfn = __find_buddy_pfn(pfn, order);
946 			buddy = page + (buddy_pfn - pfn);
947 			buddy_mt = get_pageblock_migratetype(buddy);
948 
949 			if (migratetype != buddy_mt
950 					&& (is_migrate_isolate(migratetype) ||
951 						is_migrate_isolate(buddy_mt)))
952 				goto done_merging;
953 		}
954 		max_order++;
955 		goto continue_merging;
956 	}
957 
958 done_merging:
959 	set_page_order(page, order);
960 
961 	/*
962 	 * If this is not the largest possible page, check if the buddy
963 	 * of the next-highest order is free. If it is, it's possible
964 	 * that pages are being freed that will coalesce soon. In case,
965 	 * that is happening, add the free page to the tail of the list
966 	 * so it's less likely to be used soon and more likely to be merged
967 	 * as a higher order page
968 	 */
969 	if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) {
970 		struct page *higher_page, *higher_buddy;
971 		combined_pfn = buddy_pfn & pfn;
972 		higher_page = page + (combined_pfn - pfn);
973 		buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
974 		higher_buddy = higher_page + (buddy_pfn - combined_pfn);
975 		if (pfn_valid_within(buddy_pfn) &&
976 		    page_is_buddy(higher_page, higher_buddy, order + 1)) {
977 			list_add_tail(&page->lru,
978 				&zone->free_area[order].free_list[migratetype]);
979 			goto out;
980 		}
981 	}
982 
983 	list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
984 out:
985 	zone->free_area[order].nr_free++;
986 }
987 
988 /*
989  * A bad page could be due to a number of fields. Instead of multiple branches,
990  * try and check multiple fields with one check. The caller must do a detailed
991  * check if necessary.
992  */
993 static inline bool page_expected_state(struct page *page,
994 					unsigned long check_flags)
995 {
996 	if (unlikely(atomic_read(&page->_mapcount) != -1))
997 		return false;
998 
999 	if (unlikely((unsigned long)page->mapping |
1000 			page_ref_count(page) |
1001 #ifdef CONFIG_MEMCG
1002 			(unsigned long)page->mem_cgroup |
1003 #endif
1004 			(page->flags & check_flags)))
1005 		return false;
1006 
1007 	return true;
1008 }
1009 
1010 static void free_pages_check_bad(struct page *page)
1011 {
1012 	const char *bad_reason;
1013 	unsigned long bad_flags;
1014 
1015 	bad_reason = NULL;
1016 	bad_flags = 0;
1017 
1018 	if (unlikely(atomic_read(&page->_mapcount) != -1))
1019 		bad_reason = "nonzero mapcount";
1020 	if (unlikely(page->mapping != NULL))
1021 		bad_reason = "non-NULL mapping";
1022 	if (unlikely(page_ref_count(page) != 0))
1023 		bad_reason = "nonzero _refcount";
1024 	if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
1025 		bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
1026 		bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
1027 	}
1028 #ifdef CONFIG_MEMCG
1029 	if (unlikely(page->mem_cgroup))
1030 		bad_reason = "page still charged to cgroup";
1031 #endif
1032 	bad_page(page, bad_reason, bad_flags);
1033 }
1034 
1035 static inline int free_pages_check(struct page *page)
1036 {
1037 	if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
1038 		return 0;
1039 
1040 	/* Something has gone sideways, find it */
1041 	free_pages_check_bad(page);
1042 	return 1;
1043 }
1044 
1045 static int free_tail_pages_check(struct page *head_page, struct page *page)
1046 {
1047 	int ret = 1;
1048 
1049 	/*
1050 	 * We rely page->lru.next never has bit 0 set, unless the page
1051 	 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
1052 	 */
1053 	BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
1054 
1055 	if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
1056 		ret = 0;
1057 		goto out;
1058 	}
1059 	switch (page - head_page) {
1060 	case 1:
1061 		/* the first tail page: ->mapping may be compound_mapcount() */
1062 		if (unlikely(compound_mapcount(page))) {
1063 			bad_page(page, "nonzero compound_mapcount", 0);
1064 			goto out;
1065 		}
1066 		break;
1067 	case 2:
1068 		/*
1069 		 * the second tail page: ->mapping is
1070 		 * deferred_list.next -- ignore value.
1071 		 */
1072 		break;
1073 	default:
1074 		if (page->mapping != TAIL_MAPPING) {
1075 			bad_page(page, "corrupted mapping in tail page", 0);
1076 			goto out;
1077 		}
1078 		break;
1079 	}
1080 	if (unlikely(!PageTail(page))) {
1081 		bad_page(page, "PageTail not set", 0);
1082 		goto out;
1083 	}
1084 	if (unlikely(compound_head(page) != head_page)) {
1085 		bad_page(page, "compound_head not consistent", 0);
1086 		goto out;
1087 	}
1088 	ret = 0;
1089 out:
1090 	page->mapping = NULL;
1091 	clear_compound_head(page);
1092 	return ret;
1093 }
1094 
1095 static __always_inline bool free_pages_prepare(struct page *page,
1096 					unsigned int order, bool check_free)
1097 {
1098 	int bad = 0;
1099 
1100 	VM_BUG_ON_PAGE(PageTail(page), page);
1101 
1102 	trace_mm_page_free(page, order);
1103 
1104 	/*
1105 	 * Check tail pages before head page information is cleared to
1106 	 * avoid checking PageCompound for order-0 pages.
1107 	 */
1108 	if (unlikely(order)) {
1109 		bool compound = PageCompound(page);
1110 		int i;
1111 
1112 		VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1113 
1114 		if (compound)
1115 			ClearPageDoubleMap(page);
1116 		for (i = 1; i < (1 << order); i++) {
1117 			if (compound)
1118 				bad += free_tail_pages_check(page, page + i);
1119 			if (unlikely(free_pages_check(page + i))) {
1120 				bad++;
1121 				continue;
1122 			}
1123 			(page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1124 		}
1125 	}
1126 	if (PageMappingFlags(page))
1127 		page->mapping = NULL;
1128 	if (memcg_kmem_enabled() && PageKmemcg(page))
1129 		__memcg_kmem_uncharge(page, order);
1130 	if (check_free)
1131 		bad += free_pages_check(page);
1132 	if (bad)
1133 		return false;
1134 
1135 	page_cpupid_reset_last(page);
1136 	page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1137 	reset_page_owner(page, order);
1138 
1139 	if (!PageHighMem(page)) {
1140 		debug_check_no_locks_freed(page_address(page),
1141 					   PAGE_SIZE << order);
1142 		debug_check_no_obj_freed(page_address(page),
1143 					   PAGE_SIZE << order);
1144 	}
1145 	arch_free_page(page, order);
1146 	kernel_poison_pages(page, 1 << order, 0);
1147 	kernel_map_pages(page, 1 << order, 0);
1148 	kasan_free_nondeferred_pages(page, order);
1149 
1150 	return true;
1151 }
1152 
1153 #ifdef CONFIG_DEBUG_VM
1154 static inline bool free_pcp_prepare(struct page *page)
1155 {
1156 	return free_pages_prepare(page, 0, true);
1157 }
1158 
1159 static inline bool bulkfree_pcp_prepare(struct page *page)
1160 {
1161 	return false;
1162 }
1163 #else
1164 static bool free_pcp_prepare(struct page *page)
1165 {
1166 	return free_pages_prepare(page, 0, false);
1167 }
1168 
1169 static bool bulkfree_pcp_prepare(struct page *page)
1170 {
1171 	return free_pages_check(page);
1172 }
1173 #endif /* CONFIG_DEBUG_VM */
1174 
1175 static inline void prefetch_buddy(struct page *page)
1176 {
1177 	unsigned long pfn = page_to_pfn(page);
1178 	unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0);
1179 	struct page *buddy = page + (buddy_pfn - pfn);
1180 
1181 	prefetch(buddy);
1182 }
1183 
1184 /*
1185  * Frees a number of pages from the PCP lists
1186  * Assumes all pages on list are in same zone, and of same order.
1187  * count is the number of pages to free.
1188  *
1189  * If the zone was previously in an "all pages pinned" state then look to
1190  * see if this freeing clears that state.
1191  *
1192  * And clear the zone's pages_scanned counter, to hold off the "all pages are
1193  * pinned" detection logic.
1194  */
1195 static void free_pcppages_bulk(struct zone *zone, int count,
1196 					struct per_cpu_pages *pcp)
1197 {
1198 	int migratetype = 0;
1199 	int batch_free = 0;
1200 	int prefetch_nr = 0;
1201 	bool isolated_pageblocks;
1202 	struct page *page, *tmp;
1203 	LIST_HEAD(head);
1204 
1205 	while (count) {
1206 		struct list_head *list;
1207 
1208 		/*
1209 		 * Remove pages from lists in a round-robin fashion. A
1210 		 * batch_free count is maintained that is incremented when an
1211 		 * empty list is encountered.  This is so more pages are freed
1212 		 * off fuller lists instead of spinning excessively around empty
1213 		 * lists
1214 		 */
1215 		do {
1216 			batch_free++;
1217 			if (++migratetype == MIGRATE_PCPTYPES)
1218 				migratetype = 0;
1219 			list = &pcp->lists[migratetype];
1220 		} while (list_empty(list));
1221 
1222 		/* This is the only non-empty list. Free them all. */
1223 		if (batch_free == MIGRATE_PCPTYPES)
1224 			batch_free = count;
1225 
1226 		do {
1227 			page = list_last_entry(list, struct page, lru);
1228 			/* must delete to avoid corrupting pcp list */
1229 			list_del(&page->lru);
1230 			pcp->count--;
1231 
1232 			if (bulkfree_pcp_prepare(page))
1233 				continue;
1234 
1235 			list_add_tail(&page->lru, &head);
1236 
1237 			/*
1238 			 * We are going to put the page back to the global
1239 			 * pool, prefetch its buddy to speed up later access
1240 			 * under zone->lock. It is believed the overhead of
1241 			 * an additional test and calculating buddy_pfn here
1242 			 * can be offset by reduced memory latency later. To
1243 			 * avoid excessive prefetching due to large count, only
1244 			 * prefetch buddy for the first pcp->batch nr of pages.
1245 			 */
1246 			if (prefetch_nr++ < pcp->batch)
1247 				prefetch_buddy(page);
1248 		} while (--count && --batch_free && !list_empty(list));
1249 	}
1250 
1251 	spin_lock(&zone->lock);
1252 	isolated_pageblocks = has_isolate_pageblock(zone);
1253 
1254 	/*
1255 	 * Use safe version since after __free_one_page(),
1256 	 * page->lru.next will not point to original list.
1257 	 */
1258 	list_for_each_entry_safe(page, tmp, &head, lru) {
1259 		int mt = get_pcppage_migratetype(page);
1260 		/* MIGRATE_ISOLATE page should not go to pcplists */
1261 		VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1262 		/* Pageblock could have been isolated meanwhile */
1263 		if (unlikely(isolated_pageblocks))
1264 			mt = get_pageblock_migratetype(page);
1265 
1266 		__free_one_page(page, page_to_pfn(page), zone, 0, mt);
1267 		trace_mm_page_pcpu_drain(page, 0, mt);
1268 	}
1269 	spin_unlock(&zone->lock);
1270 }
1271 
1272 static void free_one_page(struct zone *zone,
1273 				struct page *page, unsigned long pfn,
1274 				unsigned int order,
1275 				int migratetype)
1276 {
1277 	spin_lock(&zone->lock);
1278 	if (unlikely(has_isolate_pageblock(zone) ||
1279 		is_migrate_isolate(migratetype))) {
1280 		migratetype = get_pfnblock_migratetype(page, pfn);
1281 	}
1282 	__free_one_page(page, pfn, zone, order, migratetype);
1283 	spin_unlock(&zone->lock);
1284 }
1285 
1286 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1287 				unsigned long zone, int nid)
1288 {
1289 	mm_zero_struct_page(page);
1290 	set_page_links(page, zone, nid, pfn);
1291 	init_page_count(page);
1292 	page_mapcount_reset(page);
1293 	page_cpupid_reset_last(page);
1294 	page_kasan_tag_reset(page);
1295 
1296 	INIT_LIST_HEAD(&page->lru);
1297 #ifdef WANT_PAGE_VIRTUAL
1298 	/* The shift won't overflow because ZONE_NORMAL is below 4G. */
1299 	if (!is_highmem_idx(zone))
1300 		set_page_address(page, __va(pfn << PAGE_SHIFT));
1301 #endif
1302 }
1303 
1304 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1305 static void __meminit init_reserved_page(unsigned long pfn)
1306 {
1307 	pg_data_t *pgdat;
1308 	int nid, zid;
1309 
1310 	if (!early_page_uninitialised(pfn))
1311 		return;
1312 
1313 	nid = early_pfn_to_nid(pfn);
1314 	pgdat = NODE_DATA(nid);
1315 
1316 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1317 		struct zone *zone = &pgdat->node_zones[zid];
1318 
1319 		if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1320 			break;
1321 	}
1322 	__init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1323 }
1324 #else
1325 static inline void init_reserved_page(unsigned long pfn)
1326 {
1327 }
1328 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1329 
1330 /*
1331  * Initialised pages do not have PageReserved set. This function is
1332  * called for each range allocated by the bootmem allocator and
1333  * marks the pages PageReserved. The remaining valid pages are later
1334  * sent to the buddy page allocator.
1335  */
1336 void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1337 {
1338 	unsigned long start_pfn = PFN_DOWN(start);
1339 	unsigned long end_pfn = PFN_UP(end);
1340 
1341 	for (; start_pfn < end_pfn; start_pfn++) {
1342 		if (pfn_valid(start_pfn)) {
1343 			struct page *page = pfn_to_page(start_pfn);
1344 
1345 			init_reserved_page(start_pfn);
1346 
1347 			/* Avoid false-positive PageTail() */
1348 			INIT_LIST_HEAD(&page->lru);
1349 
1350 			/*
1351 			 * no need for atomic set_bit because the struct
1352 			 * page is not visible yet so nobody should
1353 			 * access it yet.
1354 			 */
1355 			__SetPageReserved(page);
1356 		}
1357 	}
1358 }
1359 
1360 static void __free_pages_ok(struct page *page, unsigned int order)
1361 {
1362 	unsigned long flags;
1363 	int migratetype;
1364 	unsigned long pfn = page_to_pfn(page);
1365 
1366 	if (!free_pages_prepare(page, order, true))
1367 		return;
1368 
1369 	migratetype = get_pfnblock_migratetype(page, pfn);
1370 	local_irq_save(flags);
1371 	__count_vm_events(PGFREE, 1 << order);
1372 	free_one_page(page_zone(page), page, pfn, order, migratetype);
1373 	local_irq_restore(flags);
1374 }
1375 
1376 void __free_pages_core(struct page *page, unsigned int order)
1377 {
1378 	unsigned int nr_pages = 1 << order;
1379 	struct page *p = page;
1380 	unsigned int loop;
1381 
1382 	prefetchw(p);
1383 	for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1384 		prefetchw(p + 1);
1385 		__ClearPageReserved(p);
1386 		set_page_count(p, 0);
1387 	}
1388 	__ClearPageReserved(p);
1389 	set_page_count(p, 0);
1390 
1391 	atomic_long_add(nr_pages, &page_zone(page)->managed_pages);
1392 	set_page_refcounted(page);
1393 	__free_pages(page, order);
1394 }
1395 
1396 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1397 	defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1398 
1399 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1400 
1401 int __meminit early_pfn_to_nid(unsigned long pfn)
1402 {
1403 	static DEFINE_SPINLOCK(early_pfn_lock);
1404 	int nid;
1405 
1406 	spin_lock(&early_pfn_lock);
1407 	nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1408 	if (nid < 0)
1409 		nid = first_online_node;
1410 	spin_unlock(&early_pfn_lock);
1411 
1412 	return nid;
1413 }
1414 #endif
1415 
1416 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1417 static inline bool __meminit __maybe_unused
1418 meminit_pfn_in_nid(unsigned long pfn, int node,
1419 		   struct mminit_pfnnid_cache *state)
1420 {
1421 	int nid;
1422 
1423 	nid = __early_pfn_to_nid(pfn, state);
1424 	if (nid >= 0 && nid != node)
1425 		return false;
1426 	return true;
1427 }
1428 
1429 /* Only safe to use early in boot when initialisation is single-threaded */
1430 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1431 {
1432 	return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1433 }
1434 
1435 #else
1436 
1437 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1438 {
1439 	return true;
1440 }
1441 static inline bool __meminit  __maybe_unused
1442 meminit_pfn_in_nid(unsigned long pfn, int node,
1443 		   struct mminit_pfnnid_cache *state)
1444 {
1445 	return true;
1446 }
1447 #endif
1448 
1449 
1450 void __init memblock_free_pages(struct page *page, unsigned long pfn,
1451 							unsigned int order)
1452 {
1453 	if (early_page_uninitialised(pfn))
1454 		return;
1455 	__free_pages_core(page, order);
1456 }
1457 
1458 /*
1459  * Check that the whole (or subset of) a pageblock given by the interval of
1460  * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1461  * with the migration of free compaction scanner. The scanners then need to
1462  * use only pfn_valid_within() check for arches that allow holes within
1463  * pageblocks.
1464  *
1465  * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1466  *
1467  * It's possible on some configurations to have a setup like node0 node1 node0
1468  * i.e. it's possible that all pages within a zones range of pages do not
1469  * belong to a single zone. We assume that a border between node0 and node1
1470  * can occur within a single pageblock, but not a node0 node1 node0
1471  * interleaving within a single pageblock. It is therefore sufficient to check
1472  * the first and last page of a pageblock and avoid checking each individual
1473  * page in a pageblock.
1474  */
1475 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1476 				     unsigned long end_pfn, struct zone *zone)
1477 {
1478 	struct page *start_page;
1479 	struct page *end_page;
1480 
1481 	/* end_pfn is one past the range we are checking */
1482 	end_pfn--;
1483 
1484 	if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1485 		return NULL;
1486 
1487 	start_page = pfn_to_online_page(start_pfn);
1488 	if (!start_page)
1489 		return NULL;
1490 
1491 	if (page_zone(start_page) != zone)
1492 		return NULL;
1493 
1494 	end_page = pfn_to_page(end_pfn);
1495 
1496 	/* This gives a shorter code than deriving page_zone(end_page) */
1497 	if (page_zone_id(start_page) != page_zone_id(end_page))
1498 		return NULL;
1499 
1500 	return start_page;
1501 }
1502 
1503 void set_zone_contiguous(struct zone *zone)
1504 {
1505 	unsigned long block_start_pfn = zone->zone_start_pfn;
1506 	unsigned long block_end_pfn;
1507 
1508 	block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1509 	for (; block_start_pfn < zone_end_pfn(zone);
1510 			block_start_pfn = block_end_pfn,
1511 			 block_end_pfn += pageblock_nr_pages) {
1512 
1513 		block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1514 
1515 		if (!__pageblock_pfn_to_page(block_start_pfn,
1516 					     block_end_pfn, zone))
1517 			return;
1518 	}
1519 
1520 	/* We confirm that there is no hole */
1521 	zone->contiguous = true;
1522 }
1523 
1524 void clear_zone_contiguous(struct zone *zone)
1525 {
1526 	zone->contiguous = false;
1527 }
1528 
1529 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1530 static void __init deferred_free_range(unsigned long pfn,
1531 				       unsigned long nr_pages)
1532 {
1533 	struct page *page;
1534 	unsigned long i;
1535 
1536 	if (!nr_pages)
1537 		return;
1538 
1539 	page = pfn_to_page(pfn);
1540 
1541 	/* Free a large naturally-aligned chunk if possible */
1542 	if (nr_pages == pageblock_nr_pages &&
1543 	    (pfn & (pageblock_nr_pages - 1)) == 0) {
1544 		set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1545 		__free_pages_core(page, pageblock_order);
1546 		return;
1547 	}
1548 
1549 	for (i = 0; i < nr_pages; i++, page++, pfn++) {
1550 		if ((pfn & (pageblock_nr_pages - 1)) == 0)
1551 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1552 		__free_pages_core(page, 0);
1553 	}
1554 }
1555 
1556 /* Completion tracking for deferred_init_memmap() threads */
1557 static atomic_t pgdat_init_n_undone __initdata;
1558 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1559 
1560 static inline void __init pgdat_init_report_one_done(void)
1561 {
1562 	if (atomic_dec_and_test(&pgdat_init_n_undone))
1563 		complete(&pgdat_init_all_done_comp);
1564 }
1565 
1566 /*
1567  * Returns true if page needs to be initialized or freed to buddy allocator.
1568  *
1569  * First we check if pfn is valid on architectures where it is possible to have
1570  * holes within pageblock_nr_pages. On systems where it is not possible, this
1571  * function is optimized out.
1572  *
1573  * Then, we check if a current large page is valid by only checking the validity
1574  * of the head pfn.
1575  *
1576  * Finally, meminit_pfn_in_nid is checked on systems where pfns can interleave
1577  * within a node: a pfn is between start and end of a node, but does not belong
1578  * to this memory node.
1579  */
1580 static inline bool __init
1581 deferred_pfn_valid(int nid, unsigned long pfn,
1582 		   struct mminit_pfnnid_cache *nid_init_state)
1583 {
1584 	if (!pfn_valid_within(pfn))
1585 		return false;
1586 	if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1587 		return false;
1588 	if (!meminit_pfn_in_nid(pfn, nid, nid_init_state))
1589 		return false;
1590 	return true;
1591 }
1592 
1593 /*
1594  * Free pages to buddy allocator. Try to free aligned pages in
1595  * pageblock_nr_pages sizes.
1596  */
1597 static void __init deferred_free_pages(int nid, int zid, unsigned long pfn,
1598 				       unsigned long end_pfn)
1599 {
1600 	struct mminit_pfnnid_cache nid_init_state = { };
1601 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1602 	unsigned long nr_free = 0;
1603 
1604 	for (; pfn < end_pfn; pfn++) {
1605 		if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
1606 			deferred_free_range(pfn - nr_free, nr_free);
1607 			nr_free = 0;
1608 		} else if (!(pfn & nr_pgmask)) {
1609 			deferred_free_range(pfn - nr_free, nr_free);
1610 			nr_free = 1;
1611 			touch_nmi_watchdog();
1612 		} else {
1613 			nr_free++;
1614 		}
1615 	}
1616 	/* Free the last block of pages to allocator */
1617 	deferred_free_range(pfn - nr_free, nr_free);
1618 }
1619 
1620 /*
1621  * Initialize struct pages.  We minimize pfn page lookups and scheduler checks
1622  * by performing it only once every pageblock_nr_pages.
1623  * Return number of pages initialized.
1624  */
1625 static unsigned long  __init deferred_init_pages(int nid, int zid,
1626 						 unsigned long pfn,
1627 						 unsigned long end_pfn)
1628 {
1629 	struct mminit_pfnnid_cache nid_init_state = { };
1630 	unsigned long nr_pgmask = pageblock_nr_pages - 1;
1631 	unsigned long nr_pages = 0;
1632 	struct page *page = NULL;
1633 
1634 	for (; pfn < end_pfn; pfn++) {
1635 		if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
1636 			page = NULL;
1637 			continue;
1638 		} else if (!page || !(pfn & nr_pgmask)) {
1639 			page = pfn_to_page(pfn);
1640 			touch_nmi_watchdog();
1641 		} else {
1642 			page++;
1643 		}
1644 		__init_single_page(page, pfn, zid, nid);
1645 		nr_pages++;
1646 	}
1647 	return (nr_pages);
1648 }
1649 
1650 /* Initialise remaining memory on a node */
1651 static int __init deferred_init_memmap(void *data)
1652 {
1653 	pg_data_t *pgdat = data;
1654 	int nid = pgdat->node_id;
1655 	unsigned long start = jiffies;
1656 	unsigned long nr_pages = 0;
1657 	unsigned long spfn, epfn, first_init_pfn, flags;
1658 	phys_addr_t spa, epa;
1659 	int zid;
1660 	struct zone *zone;
1661 	const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1662 	u64 i;
1663 
1664 	/* Bind memory initialisation thread to a local node if possible */
1665 	if (!cpumask_empty(cpumask))
1666 		set_cpus_allowed_ptr(current, cpumask);
1667 
1668 	pgdat_resize_lock(pgdat, &flags);
1669 	first_init_pfn = pgdat->first_deferred_pfn;
1670 	if (first_init_pfn == ULONG_MAX) {
1671 		pgdat_resize_unlock(pgdat, &flags);
1672 		pgdat_init_report_one_done();
1673 		return 0;
1674 	}
1675 
1676 	/* Sanity check boundaries */
1677 	BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1678 	BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1679 	pgdat->first_deferred_pfn = ULONG_MAX;
1680 
1681 	/* Only the highest zone is deferred so find it */
1682 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1683 		zone = pgdat->node_zones + zid;
1684 		if (first_init_pfn < zone_end_pfn(zone))
1685 			break;
1686 	}
1687 	first_init_pfn = max(zone->zone_start_pfn, first_init_pfn);
1688 
1689 	/*
1690 	 * Initialize and free pages. We do it in two loops: first we initialize
1691 	 * struct page, than free to buddy allocator, because while we are
1692 	 * freeing pages we can access pages that are ahead (computing buddy
1693 	 * page in __free_one_page()).
1694 	 */
1695 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1696 		spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1697 		epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1698 		nr_pages += deferred_init_pages(nid, zid, spfn, epfn);
1699 	}
1700 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1701 		spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1702 		epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1703 		deferred_free_pages(nid, zid, spfn, epfn);
1704 	}
1705 	pgdat_resize_unlock(pgdat, &flags);
1706 
1707 	/* Sanity check that the next zone really is unpopulated */
1708 	WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1709 
1710 	pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1711 					jiffies_to_msecs(jiffies - start));
1712 
1713 	pgdat_init_report_one_done();
1714 	return 0;
1715 }
1716 
1717 /*
1718  * If this zone has deferred pages, try to grow it by initializing enough
1719  * deferred pages to satisfy the allocation specified by order, rounded up to
1720  * the nearest PAGES_PER_SECTION boundary.  So we're adding memory in increments
1721  * of SECTION_SIZE bytes by initializing struct pages in increments of
1722  * PAGES_PER_SECTION * sizeof(struct page) bytes.
1723  *
1724  * Return true when zone was grown, otherwise return false. We return true even
1725  * when we grow less than requested, to let the caller decide if there are
1726  * enough pages to satisfy the allocation.
1727  *
1728  * Note: We use noinline because this function is needed only during boot, and
1729  * it is called from a __ref function _deferred_grow_zone. This way we are
1730  * making sure that it is not inlined into permanent text section.
1731  */
1732 static noinline bool __init
1733 deferred_grow_zone(struct zone *zone, unsigned int order)
1734 {
1735 	int zid = zone_idx(zone);
1736 	int nid = zone_to_nid(zone);
1737 	pg_data_t *pgdat = NODE_DATA(nid);
1738 	unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
1739 	unsigned long nr_pages = 0;
1740 	unsigned long first_init_pfn, spfn, epfn, t, flags;
1741 	unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
1742 	phys_addr_t spa, epa;
1743 	u64 i;
1744 
1745 	/* Only the last zone may have deferred pages */
1746 	if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
1747 		return false;
1748 
1749 	pgdat_resize_lock(pgdat, &flags);
1750 
1751 	/*
1752 	 * If deferred pages have been initialized while we were waiting for
1753 	 * the lock, return true, as the zone was grown.  The caller will retry
1754 	 * this zone.  We won't return to this function since the caller also
1755 	 * has this static branch.
1756 	 */
1757 	if (!static_branch_unlikely(&deferred_pages)) {
1758 		pgdat_resize_unlock(pgdat, &flags);
1759 		return true;
1760 	}
1761 
1762 	/*
1763 	 * If someone grew this zone while we were waiting for spinlock, return
1764 	 * true, as there might be enough pages already.
1765 	 */
1766 	if (first_deferred_pfn != pgdat->first_deferred_pfn) {
1767 		pgdat_resize_unlock(pgdat, &flags);
1768 		return true;
1769 	}
1770 
1771 	first_init_pfn = max(zone->zone_start_pfn, first_deferred_pfn);
1772 
1773 	if (first_init_pfn >= pgdat_end_pfn(pgdat)) {
1774 		pgdat_resize_unlock(pgdat, &flags);
1775 		return false;
1776 	}
1777 
1778 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1779 		spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1780 		epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1781 
1782 		while (spfn < epfn && nr_pages < nr_pages_needed) {
1783 			t = ALIGN(spfn + PAGES_PER_SECTION, PAGES_PER_SECTION);
1784 			first_deferred_pfn = min(t, epfn);
1785 			nr_pages += deferred_init_pages(nid, zid, spfn,
1786 							first_deferred_pfn);
1787 			spfn = first_deferred_pfn;
1788 		}
1789 
1790 		if (nr_pages >= nr_pages_needed)
1791 			break;
1792 	}
1793 
1794 	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1795 		spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1796 		epfn = min_t(unsigned long, first_deferred_pfn, PFN_DOWN(epa));
1797 		deferred_free_pages(nid, zid, spfn, epfn);
1798 
1799 		if (first_deferred_pfn == epfn)
1800 			break;
1801 	}
1802 	pgdat->first_deferred_pfn = first_deferred_pfn;
1803 	pgdat_resize_unlock(pgdat, &flags);
1804 
1805 	return nr_pages > 0;
1806 }
1807 
1808 /*
1809  * deferred_grow_zone() is __init, but it is called from
1810  * get_page_from_freelist() during early boot until deferred_pages permanently
1811  * disables this call. This is why we have refdata wrapper to avoid warning,
1812  * and to ensure that the function body gets unloaded.
1813  */
1814 static bool __ref
1815 _deferred_grow_zone(struct zone *zone, unsigned int order)
1816 {
1817 	return deferred_grow_zone(zone, order);
1818 }
1819 
1820 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1821 
1822 void __init page_alloc_init_late(void)
1823 {
1824 	struct zone *zone;
1825 
1826 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1827 	int nid;
1828 
1829 	/* There will be num_node_state(N_MEMORY) threads */
1830 	atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1831 	for_each_node_state(nid, N_MEMORY) {
1832 		kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1833 	}
1834 
1835 	/* Block until all are initialised */
1836 	wait_for_completion(&pgdat_init_all_done_comp);
1837 
1838 	/*
1839 	 * We initialized the rest of the deferred pages.  Permanently disable
1840 	 * on-demand struct page initialization.
1841 	 */
1842 	static_branch_disable(&deferred_pages);
1843 
1844 	/* Reinit limits that are based on free pages after the kernel is up */
1845 	files_maxfiles_init();
1846 #endif
1847 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
1848 	/* Discard memblock private memory */
1849 	memblock_discard();
1850 #endif
1851 
1852 	for_each_populated_zone(zone)
1853 		set_zone_contiguous(zone);
1854 }
1855 
1856 #ifdef CONFIG_CMA
1857 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1858 void __init init_cma_reserved_pageblock(struct page *page)
1859 {
1860 	unsigned i = pageblock_nr_pages;
1861 	struct page *p = page;
1862 
1863 	do {
1864 		__ClearPageReserved(p);
1865 		set_page_count(p, 0);
1866 	} while (++p, --i);
1867 
1868 	set_pageblock_migratetype(page, MIGRATE_CMA);
1869 
1870 	if (pageblock_order >= MAX_ORDER) {
1871 		i = pageblock_nr_pages;
1872 		p = page;
1873 		do {
1874 			set_page_refcounted(p);
1875 			__free_pages(p, MAX_ORDER - 1);
1876 			p += MAX_ORDER_NR_PAGES;
1877 		} while (i -= MAX_ORDER_NR_PAGES);
1878 	} else {
1879 		set_page_refcounted(page);
1880 		__free_pages(page, pageblock_order);
1881 	}
1882 
1883 	adjust_managed_page_count(page, pageblock_nr_pages);
1884 }
1885 #endif
1886 
1887 /*
1888  * The order of subdivision here is critical for the IO subsystem.
1889  * Please do not alter this order without good reasons and regression
1890  * testing. Specifically, as large blocks of memory are subdivided,
1891  * the order in which smaller blocks are delivered depends on the order
1892  * they're subdivided in this function. This is the primary factor
1893  * influencing the order in which pages are delivered to the IO
1894  * subsystem according to empirical testing, and this is also justified
1895  * by considering the behavior of a buddy system containing a single
1896  * large block of memory acted on by a series of small allocations.
1897  * This behavior is a critical factor in sglist merging's success.
1898  *
1899  * -- nyc
1900  */
1901 static inline void expand(struct zone *zone, struct page *page,
1902 	int low, int high, struct free_area *area,
1903 	int migratetype)
1904 {
1905 	unsigned long size = 1 << high;
1906 
1907 	while (high > low) {
1908 		area--;
1909 		high--;
1910 		size >>= 1;
1911 		VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1912 
1913 		/*
1914 		 * Mark as guard pages (or page), that will allow to
1915 		 * merge back to allocator when buddy will be freed.
1916 		 * Corresponding page table entries will not be touched,
1917 		 * pages will stay not present in virtual address space
1918 		 */
1919 		if (set_page_guard(zone, &page[size], high, migratetype))
1920 			continue;
1921 
1922 		list_add(&page[size].lru, &area->free_list[migratetype]);
1923 		area->nr_free++;
1924 		set_page_order(&page[size], high);
1925 	}
1926 }
1927 
1928 static void check_new_page_bad(struct page *page)
1929 {
1930 	const char *bad_reason = NULL;
1931 	unsigned long bad_flags = 0;
1932 
1933 	if (unlikely(atomic_read(&page->_mapcount) != -1))
1934 		bad_reason = "nonzero mapcount";
1935 	if (unlikely(page->mapping != NULL))
1936 		bad_reason = "non-NULL mapping";
1937 	if (unlikely(page_ref_count(page) != 0))
1938 		bad_reason = "nonzero _count";
1939 	if (unlikely(page->flags & __PG_HWPOISON)) {
1940 		bad_reason = "HWPoisoned (hardware-corrupted)";
1941 		bad_flags = __PG_HWPOISON;
1942 		/* Don't complain about hwpoisoned pages */
1943 		page_mapcount_reset(page); /* remove PageBuddy */
1944 		return;
1945 	}
1946 	if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1947 		bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1948 		bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1949 	}
1950 #ifdef CONFIG_MEMCG
1951 	if (unlikely(page->mem_cgroup))
1952 		bad_reason = "page still charged to cgroup";
1953 #endif
1954 	bad_page(page, bad_reason, bad_flags);
1955 }
1956 
1957 /*
1958  * This page is about to be returned from the page allocator
1959  */
1960 static inline int check_new_page(struct page *page)
1961 {
1962 	if (likely(page_expected_state(page,
1963 				PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1964 		return 0;
1965 
1966 	check_new_page_bad(page);
1967 	return 1;
1968 }
1969 
1970 static inline bool free_pages_prezeroed(void)
1971 {
1972 	return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1973 		page_poisoning_enabled();
1974 }
1975 
1976 #ifdef CONFIG_DEBUG_VM
1977 static bool check_pcp_refill(struct page *page)
1978 {
1979 	return false;
1980 }
1981 
1982 static bool check_new_pcp(struct page *page)
1983 {
1984 	return check_new_page(page);
1985 }
1986 #else
1987 static bool check_pcp_refill(struct page *page)
1988 {
1989 	return check_new_page(page);
1990 }
1991 static bool check_new_pcp(struct page *page)
1992 {
1993 	return false;
1994 }
1995 #endif /* CONFIG_DEBUG_VM */
1996 
1997 static bool check_new_pages(struct page *page, unsigned int order)
1998 {
1999 	int i;
2000 	for (i = 0; i < (1 << order); i++) {
2001 		struct page *p = page + i;
2002 
2003 		if (unlikely(check_new_page(p)))
2004 			return true;
2005 	}
2006 
2007 	return false;
2008 }
2009 
2010 inline void post_alloc_hook(struct page *page, unsigned int order,
2011 				gfp_t gfp_flags)
2012 {
2013 	set_page_private(page, 0);
2014 	set_page_refcounted(page);
2015 
2016 	arch_alloc_page(page, order);
2017 	kernel_map_pages(page, 1 << order, 1);
2018 	kasan_alloc_pages(page, order);
2019 	kernel_poison_pages(page, 1 << order, 1);
2020 	set_page_owner(page, order, gfp_flags);
2021 }
2022 
2023 static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
2024 							unsigned int alloc_flags)
2025 {
2026 	int i;
2027 
2028 	post_alloc_hook(page, order, gfp_flags);
2029 
2030 	if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO))
2031 		for (i = 0; i < (1 << order); i++)
2032 			clear_highpage(page + i);
2033 
2034 	if (order && (gfp_flags & __GFP_COMP))
2035 		prep_compound_page(page, order);
2036 
2037 	/*
2038 	 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
2039 	 * allocate the page. The expectation is that the caller is taking
2040 	 * steps that will free more memory. The caller should avoid the page
2041 	 * being used for !PFMEMALLOC purposes.
2042 	 */
2043 	if (alloc_flags & ALLOC_NO_WATERMARKS)
2044 		set_page_pfmemalloc(page);
2045 	else
2046 		clear_page_pfmemalloc(page);
2047 }
2048 
2049 /*
2050  * Go through the free lists for the given migratetype and remove
2051  * the smallest available page from the freelists
2052  */
2053 static __always_inline
2054 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
2055 						int migratetype)
2056 {
2057 	unsigned int current_order;
2058 	struct free_area *area;
2059 	struct page *page;
2060 
2061 	/* Find a page of the appropriate size in the preferred list */
2062 	for (current_order = order; current_order < MAX_ORDER; ++current_order) {
2063 		area = &(zone->free_area[current_order]);
2064 		page = list_first_entry_or_null(&area->free_list[migratetype],
2065 							struct page, lru);
2066 		if (!page)
2067 			continue;
2068 		list_del(&page->lru);
2069 		rmv_page_order(page);
2070 		area->nr_free--;
2071 		expand(zone, page, order, current_order, area, migratetype);
2072 		set_pcppage_migratetype(page, migratetype);
2073 		return page;
2074 	}
2075 
2076 	return NULL;
2077 }
2078 
2079 
2080 /*
2081  * This array describes the order lists are fallen back to when
2082  * the free lists for the desirable migrate type are depleted
2083  */
2084 static int fallbacks[MIGRATE_TYPES][4] = {
2085 	[MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_TYPES },
2086 	[MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
2087 	[MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_TYPES },
2088 #ifdef CONFIG_CMA
2089 	[MIGRATE_CMA]         = { MIGRATE_TYPES }, /* Never used */
2090 #endif
2091 #ifdef CONFIG_MEMORY_ISOLATION
2092 	[MIGRATE_ISOLATE]     = { MIGRATE_TYPES }, /* Never used */
2093 #endif
2094 };
2095 
2096 #ifdef CONFIG_CMA
2097 static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2098 					unsigned int order)
2099 {
2100 	return __rmqueue_smallest(zone, order, MIGRATE_CMA);
2101 }
2102 #else
2103 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
2104 					unsigned int order) { return NULL; }
2105 #endif
2106 
2107 /*
2108  * Move the free pages in a range to the free lists of the requested type.
2109  * Note that start_page and end_pages are not aligned on a pageblock
2110  * boundary. If alignment is required, use move_freepages_block()
2111  */
2112 static int move_freepages(struct zone *zone,
2113 			  struct page *start_page, struct page *end_page,
2114 			  int migratetype, int *num_movable)
2115 {
2116 	struct page *page;
2117 	unsigned int order;
2118 	int pages_moved = 0;
2119 
2120 #ifndef CONFIG_HOLES_IN_ZONE
2121 	/*
2122 	 * page_zone is not safe to call in this context when
2123 	 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
2124 	 * anyway as we check zone boundaries in move_freepages_block().
2125 	 * Remove at a later date when no bug reports exist related to
2126 	 * grouping pages by mobility
2127 	 */
2128 	VM_BUG_ON(pfn_valid(page_to_pfn(start_page)) &&
2129 	          pfn_valid(page_to_pfn(end_page)) &&
2130 	          page_zone(start_page) != page_zone(end_page));
2131 #endif
2132 	for (page = start_page; page <= end_page;) {
2133 		if (!pfn_valid_within(page_to_pfn(page))) {
2134 			page++;
2135 			continue;
2136 		}
2137 
2138 		/* Make sure we are not inadvertently changing nodes */
2139 		VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2140 
2141 		if (!PageBuddy(page)) {
2142 			/*
2143 			 * We assume that pages that could be isolated for
2144 			 * migration are movable. But we don't actually try
2145 			 * isolating, as that would be expensive.
2146 			 */
2147 			if (num_movable &&
2148 					(PageLRU(page) || __PageMovable(page)))
2149 				(*num_movable)++;
2150 
2151 			page++;
2152 			continue;
2153 		}
2154 
2155 		order = page_order(page);
2156 		list_move(&page->lru,
2157 			  &zone->free_area[order].free_list[migratetype]);
2158 		page += 1 << order;
2159 		pages_moved += 1 << order;
2160 	}
2161 
2162 	return pages_moved;
2163 }
2164 
2165 int move_freepages_block(struct zone *zone, struct page *page,
2166 				int migratetype, int *num_movable)
2167 {
2168 	unsigned long start_pfn, end_pfn;
2169 	struct page *start_page, *end_page;
2170 
2171 	if (num_movable)
2172 		*num_movable = 0;
2173 
2174 	start_pfn = page_to_pfn(page);
2175 	start_pfn = start_pfn & ~(pageblock_nr_pages-1);
2176 	start_page = pfn_to_page(start_pfn);
2177 	end_page = start_page + pageblock_nr_pages - 1;
2178 	end_pfn = start_pfn + pageblock_nr_pages - 1;
2179 
2180 	/* Do not cross zone boundaries */
2181 	if (!zone_spans_pfn(zone, start_pfn))
2182 		start_page = page;
2183 	if (!zone_spans_pfn(zone, end_pfn))
2184 		return 0;
2185 
2186 	return move_freepages(zone, start_page, end_page, migratetype,
2187 								num_movable);
2188 }
2189 
2190 static void change_pageblock_range(struct page *pageblock_page,
2191 					int start_order, int migratetype)
2192 {
2193 	int nr_pageblocks = 1 << (start_order - pageblock_order);
2194 
2195 	while (nr_pageblocks--) {
2196 		set_pageblock_migratetype(pageblock_page, migratetype);
2197 		pageblock_page += pageblock_nr_pages;
2198 	}
2199 }
2200 
2201 /*
2202  * When we are falling back to another migratetype during allocation, try to
2203  * steal extra free pages from the same pageblocks to satisfy further
2204  * allocations, instead of polluting multiple pageblocks.
2205  *
2206  * If we are stealing a relatively large buddy page, it is likely there will
2207  * be more free pages in the pageblock, so try to steal them all. For
2208  * reclaimable and unmovable allocations, we steal regardless of page size,
2209  * as fragmentation caused by those allocations polluting movable pageblocks
2210  * is worse than movable allocations stealing from unmovable and reclaimable
2211  * pageblocks.
2212  */
2213 static bool can_steal_fallback(unsigned int order, int start_mt)
2214 {
2215 	/*
2216 	 * Leaving this order check is intended, although there is
2217 	 * relaxed order check in next check. The reason is that
2218 	 * we can actually steal whole pageblock if this condition met,
2219 	 * but, below check doesn't guarantee it and that is just heuristic
2220 	 * so could be changed anytime.
2221 	 */
2222 	if (order >= pageblock_order)
2223 		return true;
2224 
2225 	if (order >= pageblock_order / 2 ||
2226 		start_mt == MIGRATE_RECLAIMABLE ||
2227 		start_mt == MIGRATE_UNMOVABLE ||
2228 		page_group_by_mobility_disabled)
2229 		return true;
2230 
2231 	return false;
2232 }
2233 
2234 static inline void boost_watermark(struct zone *zone)
2235 {
2236 	unsigned long max_boost;
2237 
2238 	if (!watermark_boost_factor)
2239 		return;
2240 
2241 	max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
2242 			watermark_boost_factor, 10000);
2243 
2244 	/*
2245 	 * high watermark may be uninitialised if fragmentation occurs
2246 	 * very early in boot so do not boost. We do not fall
2247 	 * through and boost by pageblock_nr_pages as failing
2248 	 * allocations that early means that reclaim is not going
2249 	 * to help and it may even be impossible to reclaim the
2250 	 * boosted watermark resulting in a hang.
2251 	 */
2252 	if (!max_boost)
2253 		return;
2254 
2255 	max_boost = max(pageblock_nr_pages, max_boost);
2256 
2257 	zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,
2258 		max_boost);
2259 }
2260 
2261 /*
2262  * This function implements actual steal behaviour. If order is large enough,
2263  * we can steal whole pageblock. If not, we first move freepages in this
2264  * pageblock to our migratetype and determine how many already-allocated pages
2265  * are there in the pageblock with a compatible migratetype. If at least half
2266  * of pages are free or compatible, we can change migratetype of the pageblock
2267  * itself, so pages freed in the future will be put on the correct free list.
2268  */
2269 static void steal_suitable_fallback(struct zone *zone, struct page *page,
2270 		unsigned int alloc_flags, int start_type, bool whole_block)
2271 {
2272 	unsigned int current_order = page_order(page);
2273 	struct free_area *area;
2274 	int free_pages, movable_pages, alike_pages;
2275 	int old_block_type;
2276 
2277 	old_block_type = get_pageblock_migratetype(page);
2278 
2279 	/*
2280 	 * This can happen due to races and we want to prevent broken
2281 	 * highatomic accounting.
2282 	 */
2283 	if (is_migrate_highatomic(old_block_type))
2284 		goto single_page;
2285 
2286 	/* Take ownership for orders >= pageblock_order */
2287 	if (current_order >= pageblock_order) {
2288 		change_pageblock_range(page, current_order, start_type);
2289 		goto single_page;
2290 	}
2291 
2292 	/*
2293 	 * Boost watermarks to increase reclaim pressure to reduce the
2294 	 * likelihood of future fallbacks. Wake kswapd now as the node
2295 	 * may be balanced overall and kswapd will not wake naturally.
2296 	 */
2297 	boost_watermark(zone);
2298 	if (alloc_flags & ALLOC_KSWAPD)
2299 		set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
2300 
2301 	/* We are not allowed to try stealing from the whole block */
2302 	if (!whole_block)
2303 		goto single_page;
2304 
2305 	free_pages = move_freepages_block(zone, page, start_type,
2306 						&movable_pages);
2307 	/*
2308 	 * Determine how many pages are compatible with our allocation.
2309 	 * For movable allocation, it's the number of movable pages which
2310 	 * we just obtained. For other types it's a bit more tricky.
2311 	 */
2312 	if (start_type == MIGRATE_MOVABLE) {
2313 		alike_pages = movable_pages;
2314 	} else {
2315 		/*
2316 		 * If we are falling back a RECLAIMABLE or UNMOVABLE allocation
2317 		 * to MOVABLE pageblock, consider all non-movable pages as
2318 		 * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or
2319 		 * vice versa, be conservative since we can't distinguish the
2320 		 * exact migratetype of non-movable pages.
2321 		 */
2322 		if (old_block_type == MIGRATE_MOVABLE)
2323 			alike_pages = pageblock_nr_pages
2324 						- (free_pages + movable_pages);
2325 		else
2326 			alike_pages = 0;
2327 	}
2328 
2329 	/* moving whole block can fail due to zone boundary conditions */
2330 	if (!free_pages)
2331 		goto single_page;
2332 
2333 	/*
2334 	 * If a sufficient number of pages in the block are either free or of
2335 	 * comparable migratability as our allocation, claim the whole block.
2336 	 */
2337 	if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2338 			page_group_by_mobility_disabled)
2339 		set_pageblock_migratetype(page, start_type);
2340 
2341 	return;
2342 
2343 single_page:
2344 	area = &zone->free_area[current_order];
2345 	list_move(&page->lru, &area->free_list[start_type]);
2346 }
2347 
2348 /*
2349  * Check whether there is a suitable fallback freepage with requested order.
2350  * If only_stealable is true, this function returns fallback_mt only if
2351  * we can steal other freepages all together. This would help to reduce
2352  * fragmentation due to mixed migratetype pages in one pageblock.
2353  */
2354 int find_suitable_fallback(struct free_area *area, unsigned int order,
2355 			int migratetype, bool only_stealable, bool *can_steal)
2356 {
2357 	int i;
2358 	int fallback_mt;
2359 
2360 	if (area->nr_free == 0)
2361 		return -1;
2362 
2363 	*can_steal = false;
2364 	for (i = 0;; i++) {
2365 		fallback_mt = fallbacks[migratetype][i];
2366 		if (fallback_mt == MIGRATE_TYPES)
2367 			break;
2368 
2369 		if (list_empty(&area->free_list[fallback_mt]))
2370 			continue;
2371 
2372 		if (can_steal_fallback(order, migratetype))
2373 			*can_steal = true;
2374 
2375 		if (!only_stealable)
2376 			return fallback_mt;
2377 
2378 		if (*can_steal)
2379 			return fallback_mt;
2380 	}
2381 
2382 	return -1;
2383 }
2384 
2385 /*
2386  * Reserve a pageblock for exclusive use of high-order atomic allocations if
2387  * there are no empty page blocks that contain a page with a suitable order
2388  */
2389 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2390 				unsigned int alloc_order)
2391 {
2392 	int mt;
2393 	unsigned long max_managed, flags;
2394 
2395 	/*
2396 	 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2397 	 * Check is race-prone but harmless.
2398 	 */
2399 	max_managed = (zone_managed_pages(zone) / 100) + pageblock_nr_pages;
2400 	if (zone->nr_reserved_highatomic >= max_managed)
2401 		return;
2402 
2403 	spin_lock_irqsave(&zone->lock, flags);
2404 
2405 	/* Recheck the nr_reserved_highatomic limit under the lock */
2406 	if (zone->nr_reserved_highatomic >= max_managed)
2407 		goto out_unlock;
2408 
2409 	/* Yoink! */
2410 	mt = get_pageblock_migratetype(page);
2411 	if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2412 	    && !is_migrate_cma(mt)) {
2413 		zone->nr_reserved_highatomic += pageblock_nr_pages;
2414 		set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2415 		move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2416 	}
2417 
2418 out_unlock:
2419 	spin_unlock_irqrestore(&zone->lock, flags);
2420 }
2421 
2422 /*
2423  * Used when an allocation is about to fail under memory pressure. This
2424  * potentially hurts the reliability of high-order allocations when under
2425  * intense memory pressure but failed atomic allocations should be easier
2426  * to recover from than an OOM.
2427  *
2428  * If @force is true, try to unreserve a pageblock even though highatomic
2429  * pageblock is exhausted.
2430  */
2431 static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2432 						bool force)
2433 {
2434 	struct zonelist *zonelist = ac->zonelist;
2435 	unsigned long flags;
2436 	struct zoneref *z;
2437 	struct zone *zone;
2438 	struct page *page;
2439 	int order;
2440 	bool ret;
2441 
2442 	for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2443 								ac->nodemask) {
2444 		/*
2445 		 * Preserve at least one pageblock unless memory pressure
2446 		 * is really high.
2447 		 */
2448 		if (!force && zone->nr_reserved_highatomic <=
2449 					pageblock_nr_pages)
2450 			continue;
2451 
2452 		spin_lock_irqsave(&zone->lock, flags);
2453 		for (order = 0; order < MAX_ORDER; order++) {
2454 			struct free_area *area = &(zone->free_area[order]);
2455 
2456 			page = list_first_entry_or_null(
2457 					&area->free_list[MIGRATE_HIGHATOMIC],
2458 					struct page, lru);
2459 			if (!page)
2460 				continue;
2461 
2462 			/*
2463 			 * In page freeing path, migratetype change is racy so
2464 			 * we can counter several free pages in a pageblock
2465 			 * in this loop althoug we changed the pageblock type
2466 			 * from highatomic to ac->migratetype. So we should
2467 			 * adjust the count once.
2468 			 */
2469 			if (is_migrate_highatomic_page(page)) {
2470 				/*
2471 				 * It should never happen but changes to
2472 				 * locking could inadvertently allow a per-cpu
2473 				 * drain to add pages to MIGRATE_HIGHATOMIC
2474 				 * while unreserving so be safe and watch for
2475 				 * underflows.
2476 				 */
2477 				zone->nr_reserved_highatomic -= min(
2478 						pageblock_nr_pages,
2479 						zone->nr_reserved_highatomic);
2480 			}
2481 
2482 			/*
2483 			 * Convert to ac->migratetype and avoid the normal
2484 			 * pageblock stealing heuristics. Minimally, the caller
2485 			 * is doing the work and needs the pages. More
2486 			 * importantly, if the block was always converted to
2487 			 * MIGRATE_UNMOVABLE or another type then the number
2488 			 * of pageblocks that cannot be completely freed
2489 			 * may increase.
2490 			 */
2491 			set_pageblock_migratetype(page, ac->migratetype);
2492 			ret = move_freepages_block(zone, page, ac->migratetype,
2493 									NULL);
2494 			if (ret) {
2495 				spin_unlock_irqrestore(&zone->lock, flags);
2496 				return ret;
2497 			}
2498 		}
2499 		spin_unlock_irqrestore(&zone->lock, flags);
2500 	}
2501 
2502 	return false;
2503 }
2504 
2505 /*
2506  * Try finding a free buddy page on the fallback list and put it on the free
2507  * list of requested migratetype, possibly along with other pages from the same
2508  * block, depending on fragmentation avoidance heuristics. Returns true if
2509  * fallback was found so that __rmqueue_smallest() can grab it.
2510  *
2511  * The use of signed ints for order and current_order is a deliberate
2512  * deviation from the rest of this file, to make the for loop
2513  * condition simpler.
2514  */
2515 static __always_inline bool
2516 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype,
2517 						unsigned int alloc_flags)
2518 {
2519 	struct free_area *area;
2520 	int current_order;
2521 	int min_order = order;
2522 	struct page *page;
2523 	int fallback_mt;
2524 	bool can_steal;
2525 
2526 	/*
2527 	 * Do not steal pages from freelists belonging to other pageblocks
2528 	 * i.e. orders < pageblock_order. If there are no local zones free,
2529 	 * the zonelists will be reiterated without ALLOC_NOFRAGMENT.
2530 	 */
2531 	if (alloc_flags & ALLOC_NOFRAGMENT)
2532 		min_order = pageblock_order;
2533 
2534 	/*
2535 	 * Find the largest available free page in the other list. This roughly
2536 	 * approximates finding the pageblock with the most free pages, which
2537 	 * would be too costly to do exactly.
2538 	 */
2539 	for (current_order = MAX_ORDER - 1; current_order >= min_order;
2540 				--current_order) {
2541 		area = &(zone->free_area[current_order]);
2542 		fallback_mt = find_suitable_fallback(area, current_order,
2543 				start_migratetype, false, &can_steal);
2544 		if (fallback_mt == -1)
2545 			continue;
2546 
2547 		/*
2548 		 * We cannot steal all free pages from the pageblock and the
2549 		 * requested migratetype is movable. In that case it's better to
2550 		 * steal and split the smallest available page instead of the
2551 		 * largest available page, because even if the next movable
2552 		 * allocation falls back into a different pageblock than this
2553 		 * one, it won't cause permanent fragmentation.
2554 		 */
2555 		if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2556 					&& current_order > order)
2557 			goto find_smallest;
2558 
2559 		goto do_steal;
2560 	}
2561 
2562 	return false;
2563 
2564 find_smallest:
2565 	for (current_order = order; current_order < MAX_ORDER;
2566 							current_order++) {
2567 		area = &(zone->free_area[current_order]);
2568 		fallback_mt = find_suitable_fallback(area, current_order,
2569 				start_migratetype, false, &can_steal);
2570 		if (fallback_mt != -1)
2571 			break;
2572 	}
2573 
2574 	/*
2575 	 * This should not happen - we already found a suitable fallback
2576 	 * when looking for the largest page.
2577 	 */
2578 	VM_BUG_ON(current_order == MAX_ORDER);
2579 
2580 do_steal:
2581 	page = list_first_entry(&area->free_list[fallback_mt],
2582 							struct page, lru);
2583 
2584 	steal_suitable_fallback(zone, page, alloc_flags, start_migratetype,
2585 								can_steal);
2586 
2587 	trace_mm_page_alloc_extfrag(page, order, current_order,
2588 		start_migratetype, fallback_mt);
2589 
2590 	return true;
2591 
2592 }
2593 
2594 /*
2595  * Do the hard work of removing an element from the buddy allocator.
2596  * Call me with the zone->lock already held.
2597  */
2598 static __always_inline struct page *
2599 __rmqueue(struct zone *zone, unsigned int order, int migratetype,
2600 						unsigned int alloc_flags)
2601 {
2602 	struct page *page;
2603 
2604 retry:
2605 	page = __rmqueue_smallest(zone, order, migratetype);
2606 	if (unlikely(!page)) {
2607 		if (migratetype == MIGRATE_MOVABLE)
2608 			page = __rmqueue_cma_fallback(zone, order);
2609 
2610 		if (!page && __rmqueue_fallback(zone, order, migratetype,
2611 								alloc_flags))
2612 			goto retry;
2613 	}
2614 
2615 	trace_mm_page_alloc_zone_locked(page, order, migratetype);
2616 	return page;
2617 }
2618 
2619 /*
2620  * Obtain a specified number of elements from the buddy allocator, all under
2621  * a single hold of the lock, for efficiency.  Add them to the supplied list.
2622  * Returns the number of new pages which were placed at *list.
2623  */
2624 static int rmqueue_bulk(struct zone *zone, unsigned int order,
2625 			unsigned long count, struct list_head *list,
2626 			int migratetype, unsigned int alloc_flags)
2627 {
2628 	int i, alloced = 0;
2629 
2630 	spin_lock(&zone->lock);
2631 	for (i = 0; i < count; ++i) {
2632 		struct page *page = __rmqueue(zone, order, migratetype,
2633 								alloc_flags);
2634 		if (unlikely(page == NULL))
2635 			break;
2636 
2637 		if (unlikely(check_pcp_refill(page)))
2638 			continue;
2639 
2640 		/*
2641 		 * Split buddy pages returned by expand() are received here in
2642 		 * physical page order. The page is added to the tail of
2643 		 * caller's list. From the callers perspective, the linked list
2644 		 * is ordered by page number under some conditions. This is
2645 		 * useful for IO devices that can forward direction from the
2646 		 * head, thus also in the physical page order. This is useful
2647 		 * for IO devices that can merge IO requests if the physical
2648 		 * pages are ordered properly.
2649 		 */
2650 		list_add_tail(&page->lru, list);
2651 		alloced++;
2652 		if (is_migrate_cma(get_pcppage_migratetype(page)))
2653 			__mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2654 					      -(1 << order));
2655 	}
2656 
2657 	/*
2658 	 * i pages were removed from the buddy list even if some leak due
2659 	 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2660 	 * on i. Do not confuse with 'alloced' which is the number of
2661 	 * pages added to the pcp list.
2662 	 */
2663 	__mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2664 	spin_unlock(&zone->lock);
2665 	return alloced;
2666 }
2667 
2668 #ifdef CONFIG_NUMA
2669 /*
2670  * Called from the vmstat counter updater to drain pagesets of this
2671  * currently executing processor on remote nodes after they have
2672  * expired.
2673  *
2674  * Note that this function must be called with the thread pinned to
2675  * a single processor.
2676  */
2677 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2678 {
2679 	unsigned long flags;
2680 	int to_drain, batch;
2681 
2682 	local_irq_save(flags);
2683 	batch = READ_ONCE(pcp->batch);
2684 	to_drain = min(pcp->count, batch);
2685 	if (to_drain > 0)
2686 		free_pcppages_bulk(zone, to_drain, pcp);
2687 	local_irq_restore(flags);
2688 }
2689 #endif
2690 
2691 /*
2692  * Drain pcplists of the indicated processor and zone.
2693  *
2694  * The processor must either be the current processor and the
2695  * thread pinned to the current processor or a processor that
2696  * is not online.
2697  */
2698 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2699 {
2700 	unsigned long flags;
2701 	struct per_cpu_pageset *pset;
2702 	struct per_cpu_pages *pcp;
2703 
2704 	local_irq_save(flags);
2705 	pset = per_cpu_ptr(zone->pageset, cpu);
2706 
2707 	pcp = &pset->pcp;
2708 	if (pcp->count)
2709 		free_pcppages_bulk(zone, pcp->count, pcp);
2710 	local_irq_restore(flags);
2711 }
2712 
2713 /*
2714  * Drain pcplists of all zones on the indicated processor.
2715  *
2716  * The processor must either be the current processor and the
2717  * thread pinned to the current processor or a processor that
2718  * is not online.
2719  */
2720 static void drain_pages(unsigned int cpu)
2721 {
2722 	struct zone *zone;
2723 
2724 	for_each_populated_zone(zone) {
2725 		drain_pages_zone(cpu, zone);
2726 	}
2727 }
2728 
2729 /*
2730  * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2731  *
2732  * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2733  * the single zone's pages.
2734  */
2735 void drain_local_pages(struct zone *zone)
2736 {
2737 	int cpu = smp_processor_id();
2738 
2739 	if (zone)
2740 		drain_pages_zone(cpu, zone);
2741 	else
2742 		drain_pages(cpu);
2743 }
2744 
2745 static void drain_local_pages_wq(struct work_struct *work)
2746 {
2747 	struct pcpu_drain *drain;
2748 
2749 	drain = container_of(work, struct pcpu_drain, work);
2750 
2751 	/*
2752 	 * drain_all_pages doesn't use proper cpu hotplug protection so
2753 	 * we can race with cpu offline when the WQ can move this from
2754 	 * a cpu pinned worker to an unbound one. We can operate on a different
2755 	 * cpu which is allright but we also have to make sure to not move to
2756 	 * a different one.
2757 	 */
2758 	preempt_disable();
2759 	drain_local_pages(drain->zone);
2760 	preempt_enable();
2761 }
2762 
2763 /*
2764  * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2765  *
2766  * When zone parameter is non-NULL, spill just the single zone's pages.
2767  *
2768  * Note that this can be extremely slow as the draining happens in a workqueue.
2769  */
2770 void drain_all_pages(struct zone *zone)
2771 {
2772 	int cpu;
2773 
2774 	/*
2775 	 * Allocate in the BSS so we wont require allocation in
2776 	 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2777 	 */
2778 	static cpumask_t cpus_with_pcps;
2779 
2780 	/*
2781 	 * Make sure nobody triggers this path before mm_percpu_wq is fully
2782 	 * initialized.
2783 	 */
2784 	if (WARN_ON_ONCE(!mm_percpu_wq))
2785 		return;
2786 
2787 	/*
2788 	 * Do not drain if one is already in progress unless it's specific to
2789 	 * a zone. Such callers are primarily CMA and memory hotplug and need
2790 	 * the drain to be complete when the call returns.
2791 	 */
2792 	if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
2793 		if (!zone)
2794 			return;
2795 		mutex_lock(&pcpu_drain_mutex);
2796 	}
2797 
2798 	/*
2799 	 * We don't care about racing with CPU hotplug event
2800 	 * as offline notification will cause the notified
2801 	 * cpu to drain that CPU pcps and on_each_cpu_mask
2802 	 * disables preemption as part of its processing
2803 	 */
2804 	for_each_online_cpu(cpu) {
2805 		struct per_cpu_pageset *pcp;
2806 		struct zone *z;
2807 		bool has_pcps = false;
2808 
2809 		if (zone) {
2810 			pcp = per_cpu_ptr(zone->pageset, cpu);
2811 			if (pcp->pcp.count)
2812 				has_pcps = true;
2813 		} else {
2814 			for_each_populated_zone(z) {
2815 				pcp = per_cpu_ptr(z->pageset, cpu);
2816 				if (pcp->pcp.count) {
2817 					has_pcps = true;
2818 					break;
2819 				}
2820 			}
2821 		}
2822 
2823 		if (has_pcps)
2824 			cpumask_set_cpu(cpu, &cpus_with_pcps);
2825 		else
2826 			cpumask_clear_cpu(cpu, &cpus_with_pcps);
2827 	}
2828 
2829 	for_each_cpu(cpu, &cpus_with_pcps) {
2830 		struct pcpu_drain *drain = per_cpu_ptr(&pcpu_drain, cpu);
2831 
2832 		drain->zone = zone;
2833 		INIT_WORK(&drain->work, drain_local_pages_wq);
2834 		queue_work_on(cpu, mm_percpu_wq, &drain->work);
2835 	}
2836 	for_each_cpu(cpu, &cpus_with_pcps)
2837 		flush_work(&per_cpu_ptr(&pcpu_drain, cpu)->work);
2838 
2839 	mutex_unlock(&pcpu_drain_mutex);
2840 }
2841 
2842 #ifdef CONFIG_HIBERNATION
2843 
2844 /*
2845  * Touch the watchdog for every WD_PAGE_COUNT pages.
2846  */
2847 #define WD_PAGE_COUNT	(128*1024)
2848 
2849 void mark_free_pages(struct zone *zone)
2850 {
2851 	unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
2852 	unsigned long flags;
2853 	unsigned int order, t;
2854 	struct page *page;
2855 
2856 	if (zone_is_empty(zone))
2857 		return;
2858 
2859 	spin_lock_irqsave(&zone->lock, flags);
2860 
2861 	max_zone_pfn = zone_end_pfn(zone);
2862 	for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2863 		if (pfn_valid(pfn)) {
2864 			page = pfn_to_page(pfn);
2865 
2866 			if (!--page_count) {
2867 				touch_nmi_watchdog();
2868 				page_count = WD_PAGE_COUNT;
2869 			}
2870 
2871 			if (page_zone(page) != zone)
2872 				continue;
2873 
2874 			if (!swsusp_page_is_forbidden(page))
2875 				swsusp_unset_page_free(page);
2876 		}
2877 
2878 	for_each_migratetype_order(order, t) {
2879 		list_for_each_entry(page,
2880 				&zone->free_area[order].free_list[t], lru) {
2881 			unsigned long i;
2882 
2883 			pfn = page_to_pfn(page);
2884 			for (i = 0; i < (1UL << order); i++) {
2885 				if (!--page_count) {
2886 					touch_nmi_watchdog();
2887 					page_count = WD_PAGE_COUNT;
2888 				}
2889 				swsusp_set_page_free(pfn_to_page(pfn + i));
2890 			}
2891 		}
2892 	}
2893 	spin_unlock_irqrestore(&zone->lock, flags);
2894 }
2895 #endif /* CONFIG_PM */
2896 
2897 static bool free_unref_page_prepare(struct page *page, unsigned long pfn)
2898 {
2899 	int migratetype;
2900 
2901 	if (!free_pcp_prepare(page))
2902 		return false;
2903 
2904 	migratetype = get_pfnblock_migratetype(page, pfn);
2905 	set_pcppage_migratetype(page, migratetype);
2906 	return true;
2907 }
2908 
2909 static void free_unref_page_commit(struct page *page, unsigned long pfn)
2910 {
2911 	struct zone *zone = page_zone(page);
2912 	struct per_cpu_pages *pcp;
2913 	int migratetype;
2914 
2915 	migratetype = get_pcppage_migratetype(page);
2916 	__count_vm_event(PGFREE);
2917 
2918 	/*
2919 	 * We only track unmovable, reclaimable and movable on pcp lists.
2920 	 * Free ISOLATE pages back to the allocator because they are being
2921 	 * offlined but treat HIGHATOMIC as movable pages so we can get those
2922 	 * areas back if necessary. Otherwise, we may have to free
2923 	 * excessively into the page allocator
2924 	 */
2925 	if (migratetype >= MIGRATE_PCPTYPES) {
2926 		if (unlikely(is_migrate_isolate(migratetype))) {
2927 			free_one_page(zone, page, pfn, 0, migratetype);
2928 			return;
2929 		}
2930 		migratetype = MIGRATE_MOVABLE;
2931 	}
2932 
2933 	pcp = &this_cpu_ptr(zone->pageset)->pcp;
2934 	list_add(&page->lru, &pcp->lists[migratetype]);
2935 	pcp->count++;
2936 	if (pcp->count >= pcp->high) {
2937 		unsigned long batch = READ_ONCE(pcp->batch);
2938 		free_pcppages_bulk(zone, batch, pcp);
2939 	}
2940 }
2941 
2942 /*
2943  * Free a 0-order page
2944  */
2945 void free_unref_page(struct page *page)
2946 {
2947 	unsigned long flags;
2948 	unsigned long pfn = page_to_pfn(page);
2949 
2950 	if (!free_unref_page_prepare(page, pfn))
2951 		return;
2952 
2953 	local_irq_save(flags);
2954 	free_unref_page_commit(page, pfn);
2955 	local_irq_restore(flags);
2956 }
2957 
2958 /*
2959  * Free a list of 0-order pages
2960  */
2961 void free_unref_page_list(struct list_head *list)
2962 {
2963 	struct page *page, *next;
2964 	unsigned long flags, pfn;
2965 	int batch_count = 0;
2966 
2967 	/* Prepare pages for freeing */
2968 	list_for_each_entry_safe(page, next, list, lru) {
2969 		pfn = page_to_pfn(page);
2970 		if (!free_unref_page_prepare(page, pfn))
2971 			list_del(&page->lru);
2972 		set_page_private(page, pfn);
2973 	}
2974 
2975 	local_irq_save(flags);
2976 	list_for_each_entry_safe(page, next, list, lru) {
2977 		unsigned long pfn = page_private(page);
2978 
2979 		set_page_private(page, 0);
2980 		trace_mm_page_free_batched(page);
2981 		free_unref_page_commit(page, pfn);
2982 
2983 		/*
2984 		 * Guard against excessive IRQ disabled times when we get
2985 		 * a large list of pages to free.
2986 		 */
2987 		if (++batch_count == SWAP_CLUSTER_MAX) {
2988 			local_irq_restore(flags);
2989 			batch_count = 0;
2990 			local_irq_save(flags);
2991 		}
2992 	}
2993 	local_irq_restore(flags);
2994 }
2995 
2996 /*
2997  * split_page takes a non-compound higher-order page, and splits it into
2998  * n (1<<order) sub-pages: page[0..n]
2999  * Each sub-page must be freed individually.
3000  *
3001  * Note: this is probably too low level an operation for use in drivers.
3002  * Please consult with lkml before using this in your driver.
3003  */
3004 void split_page(struct page *page, unsigned int order)
3005 {
3006 	int i;
3007 
3008 	VM_BUG_ON_PAGE(PageCompound(page), page);
3009 	VM_BUG_ON_PAGE(!page_count(page), page);
3010 
3011 	for (i = 1; i < (1 << order); i++)
3012 		set_page_refcounted(page + i);
3013 	split_page_owner(page, order);
3014 }
3015 EXPORT_SYMBOL_GPL(split_page);
3016 
3017 int __isolate_free_page(struct page *page, unsigned int order)
3018 {
3019 	unsigned long watermark;
3020 	struct zone *zone;
3021 	int mt;
3022 
3023 	BUG_ON(!PageBuddy(page));
3024 
3025 	zone = page_zone(page);
3026 	mt = get_pageblock_migratetype(page);
3027 
3028 	if (!is_migrate_isolate(mt)) {
3029 		/*
3030 		 * Obey watermarks as if the page was being allocated. We can
3031 		 * emulate a high-order watermark check with a raised order-0
3032 		 * watermark, because we already know our high-order page
3033 		 * exists.
3034 		 */
3035 		watermark = zone->_watermark[WMARK_MIN] + (1UL << order);
3036 		if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
3037 			return 0;
3038 
3039 		__mod_zone_freepage_state(zone, -(1UL << order), mt);
3040 	}
3041 
3042 	/* Remove page from free list */
3043 	list_del(&page->lru);
3044 	zone->free_area[order].nr_free--;
3045 	rmv_page_order(page);
3046 
3047 	/*
3048 	 * Set the pageblock if the isolated page is at least half of a
3049 	 * pageblock
3050 	 */
3051 	if (order >= pageblock_order - 1) {
3052 		struct page *endpage = page + (1 << order) - 1;
3053 		for (; page < endpage; page += pageblock_nr_pages) {
3054 			int mt = get_pageblock_migratetype(page);
3055 			if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
3056 			    && !is_migrate_highatomic(mt))
3057 				set_pageblock_migratetype(page,
3058 							  MIGRATE_MOVABLE);
3059 		}
3060 	}
3061 
3062 
3063 	return 1UL << order;
3064 }
3065 
3066 /*
3067  * Update NUMA hit/miss statistics
3068  *
3069  * Must be called with interrupts disabled.
3070  */
3071 static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
3072 {
3073 #ifdef CONFIG_NUMA
3074 	enum numa_stat_item local_stat = NUMA_LOCAL;
3075 
3076 	/* skip numa counters update if numa stats is disabled */
3077 	if (!static_branch_likely(&vm_numa_stat_key))
3078 		return;
3079 
3080 	if (zone_to_nid(z) != numa_node_id())
3081 		local_stat = NUMA_OTHER;
3082 
3083 	if (zone_to_nid(z) == zone_to_nid(preferred_zone))
3084 		__inc_numa_state(z, NUMA_HIT);
3085 	else {
3086 		__inc_numa_state(z, NUMA_MISS);
3087 		__inc_numa_state(preferred_zone, NUMA_FOREIGN);
3088 	}
3089 	__inc_numa_state(z, local_stat);
3090 #endif
3091 }
3092 
3093 /* Remove page from the per-cpu list, caller must protect the list */
3094 static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
3095 			unsigned int alloc_flags,
3096 			struct per_cpu_pages *pcp,
3097 			struct list_head *list)
3098 {
3099 	struct page *page;
3100 
3101 	do {
3102 		if (list_empty(list)) {
3103 			pcp->count += rmqueue_bulk(zone, 0,
3104 					pcp->batch, list,
3105 					migratetype, alloc_flags);
3106 			if (unlikely(list_empty(list)))
3107 				return NULL;
3108 		}
3109 
3110 		page = list_first_entry(list, struct page, lru);
3111 		list_del(&page->lru);
3112 		pcp->count--;
3113 	} while (check_new_pcp(page));
3114 
3115 	return page;
3116 }
3117 
3118 /* Lock and remove page from the per-cpu list */
3119 static struct page *rmqueue_pcplist(struct zone *preferred_zone,
3120 			struct zone *zone, unsigned int order,
3121 			gfp_t gfp_flags, int migratetype,
3122 			unsigned int alloc_flags)
3123 {
3124 	struct per_cpu_pages *pcp;
3125 	struct list_head *list;
3126 	struct page *page;
3127 	unsigned long flags;
3128 
3129 	local_irq_save(flags);
3130 	pcp = &this_cpu_ptr(zone->pageset)->pcp;
3131 	list = &pcp->lists[migratetype];
3132 	page = __rmqueue_pcplist(zone,  migratetype, alloc_flags, pcp, list);
3133 	if (page) {
3134 		__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3135 		zone_statistics(preferred_zone, zone);
3136 	}
3137 	local_irq_restore(flags);
3138 	return page;
3139 }
3140 
3141 /*
3142  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
3143  */
3144 static inline
3145 struct page *rmqueue(struct zone *preferred_zone,
3146 			struct zone *zone, unsigned int order,
3147 			gfp_t gfp_flags, unsigned int alloc_flags,
3148 			int migratetype)
3149 {
3150 	unsigned long flags;
3151 	struct page *page;
3152 
3153 	if (likely(order == 0)) {
3154 		page = rmqueue_pcplist(preferred_zone, zone, order,
3155 				gfp_flags, migratetype, alloc_flags);
3156 		goto out;
3157 	}
3158 
3159 	/*
3160 	 * We most definitely don't want callers attempting to
3161 	 * allocate greater than order-1 page units with __GFP_NOFAIL.
3162 	 */
3163 	WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
3164 	spin_lock_irqsave(&zone->lock, flags);
3165 
3166 	do {
3167 		page = NULL;
3168 		if (alloc_flags & ALLOC_HARDER) {
3169 			page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3170 			if (page)
3171 				trace_mm_page_alloc_zone_locked(page, order, migratetype);
3172 		}
3173 		if (!page)
3174 			page = __rmqueue(zone, order, migratetype, alloc_flags);
3175 	} while (page && check_new_pages(page, order));
3176 	spin_unlock(&zone->lock);
3177 	if (!page)
3178 		goto failed;
3179 	__mod_zone_freepage_state(zone, -(1 << order),
3180 				  get_pcppage_migratetype(page));
3181 
3182 	__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3183 	zone_statistics(preferred_zone, zone);
3184 	local_irq_restore(flags);
3185 
3186 out:
3187 	/* Separate test+clear to avoid unnecessary atomics */
3188 	if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
3189 		clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
3190 		wakeup_kswapd(zone, 0, 0, zone_idx(zone));
3191 	}
3192 
3193 	VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3194 	return page;
3195 
3196 failed:
3197 	local_irq_restore(flags);
3198 	return NULL;
3199 }
3200 
3201 #ifdef CONFIG_FAIL_PAGE_ALLOC
3202 
3203 static struct {
3204 	struct fault_attr attr;
3205 
3206 	bool ignore_gfp_highmem;
3207 	bool ignore_gfp_reclaim;
3208 	u32 min_order;
3209 } fail_page_alloc = {
3210 	.attr = FAULT_ATTR_INITIALIZER,
3211 	.ignore_gfp_reclaim = true,
3212 	.ignore_gfp_highmem = true,
3213 	.min_order = 1,
3214 };
3215 
3216 static int __init setup_fail_page_alloc(char *str)
3217 {
3218 	return setup_fault_attr(&fail_page_alloc.attr, str);
3219 }
3220 __setup("fail_page_alloc=", setup_fail_page_alloc);
3221 
3222 static bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3223 {
3224 	if (order < fail_page_alloc.min_order)
3225 		return false;
3226 	if (gfp_mask & __GFP_NOFAIL)
3227 		return false;
3228 	if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3229 		return false;
3230 	if (fail_page_alloc.ignore_gfp_reclaim &&
3231 			(gfp_mask & __GFP_DIRECT_RECLAIM))
3232 		return false;
3233 
3234 	return should_fail(&fail_page_alloc.attr, 1 << order);
3235 }
3236 
3237 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3238 
3239 static int __init fail_page_alloc_debugfs(void)
3240 {
3241 	umode_t mode = S_IFREG | 0600;
3242 	struct dentry *dir;
3243 
3244 	dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3245 					&fail_page_alloc.attr);
3246 
3247 	debugfs_create_bool("ignore-gfp-wait", mode, dir,
3248 			    &fail_page_alloc.ignore_gfp_reclaim);
3249 	debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3250 			    &fail_page_alloc.ignore_gfp_highmem);
3251 	debugfs_create_u32("min-order", mode, dir, &fail_page_alloc.min_order);
3252 
3253 	return 0;
3254 }
3255 
3256 late_initcall(fail_page_alloc_debugfs);
3257 
3258 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
3259 
3260 #else /* CONFIG_FAIL_PAGE_ALLOC */
3261 
3262 static inline bool __should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3263 {
3264 	return false;
3265 }
3266 
3267 #endif /* CONFIG_FAIL_PAGE_ALLOC */
3268 
3269 static noinline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3270 {
3271 	return __should_fail_alloc_page(gfp_mask, order);
3272 }
3273 ALLOW_ERROR_INJECTION(should_fail_alloc_page, TRUE);
3274 
3275 /*
3276  * Return true if free base pages are above 'mark'. For high-order checks it
3277  * will return true of the order-0 watermark is reached and there is at least
3278  * one free page of a suitable size. Checking now avoids taking the zone lock
3279  * to check in the allocation paths if no pages are free.
3280  */
3281 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3282 			 int classzone_idx, unsigned int alloc_flags,
3283 			 long free_pages)
3284 {
3285 	long min = mark;
3286 	int o;
3287 	const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3288 
3289 	/* free_pages may go negative - that's OK */
3290 	free_pages -= (1 << order) - 1;
3291 
3292 	if (alloc_flags & ALLOC_HIGH)
3293 		min -= min / 2;
3294 
3295 	/*
3296 	 * If the caller does not have rights to ALLOC_HARDER then subtract
3297 	 * the high-atomic reserves. This will over-estimate the size of the
3298 	 * atomic reserve but it avoids a search.
3299 	 */
3300 	if (likely(!alloc_harder)) {
3301 		free_pages -= z->nr_reserved_highatomic;
3302 	} else {
3303 		/*
3304 		 * OOM victims can try even harder than normal ALLOC_HARDER
3305 		 * users on the grounds that it's definitely going to be in
3306 		 * the exit path shortly and free memory. Any allocation it
3307 		 * makes during the free path will be small and short-lived.
3308 		 */
3309 		if (alloc_flags & ALLOC_OOM)
3310 			min -= min / 2;
3311 		else
3312 			min -= min / 4;
3313 	}
3314 
3315 
3316 #ifdef CONFIG_CMA
3317 	/* If allocation can't use CMA areas don't use free CMA pages */
3318 	if (!(alloc_flags & ALLOC_CMA))
3319 		free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
3320 #endif
3321 
3322 	/*
3323 	 * Check watermarks for an order-0 allocation request. If these
3324 	 * are not met, then a high-order request also cannot go ahead
3325 	 * even if a suitable page happened to be free.
3326 	 */
3327 	if (free_pages <= min + z->lowmem_reserve[classzone_idx])
3328 		return false;
3329 
3330 	/* If this is an order-0 request then the watermark is fine */
3331 	if (!order)
3332 		return true;
3333 
3334 	/* For a high-order request, check at least one suitable page is free */
3335 	for (o = order; o < MAX_ORDER; o++) {
3336 		struct free_area *area = &z->free_area[o];
3337 		int mt;
3338 
3339 		if (!area->nr_free)
3340 			continue;
3341 
3342 		for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3343 			if (!list_empty(&area->free_list[mt]))
3344 				return true;
3345 		}
3346 
3347 #ifdef CONFIG_CMA
3348 		if ((alloc_flags & ALLOC_CMA) &&
3349 		    !list_empty(&area->free_list[MIGRATE_CMA])) {
3350 			return true;
3351 		}
3352 #endif
3353 		if (alloc_harder &&
3354 			!list_empty(&area->free_list[MIGRATE_HIGHATOMIC]))
3355 			return true;
3356 	}
3357 	return false;
3358 }
3359 
3360 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3361 		      int classzone_idx, unsigned int alloc_flags)
3362 {
3363 	return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3364 					zone_page_state(z, NR_FREE_PAGES));
3365 }
3366 
3367 static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3368 		unsigned long mark, int classzone_idx, unsigned int alloc_flags)
3369 {
3370 	long free_pages = zone_page_state(z, NR_FREE_PAGES);
3371 	long cma_pages = 0;
3372 
3373 #ifdef CONFIG_CMA
3374 	/* If allocation can't use CMA areas don't use free CMA pages */
3375 	if (!(alloc_flags & ALLOC_CMA))
3376 		cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
3377 #endif
3378 
3379 	/*
3380 	 * Fast check for order-0 only. If this fails then the reserves
3381 	 * need to be calculated. There is a corner case where the check
3382 	 * passes but only the high-order atomic reserve are free. If
3383 	 * the caller is !atomic then it'll uselessly search the free
3384 	 * list. That corner case is then slower but it is harmless.
3385 	 */
3386 	if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
3387 		return true;
3388 
3389 	return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3390 					free_pages);
3391 }
3392 
3393 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3394 			unsigned long mark, int classzone_idx)
3395 {
3396 	long free_pages = zone_page_state(z, NR_FREE_PAGES);
3397 
3398 	if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3399 		free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3400 
3401 	return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
3402 								free_pages);
3403 }
3404 
3405 #ifdef CONFIG_NUMA
3406 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3407 {
3408 	return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
3409 				RECLAIM_DISTANCE;
3410 }
3411 #else	/* CONFIG_NUMA */
3412 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3413 {
3414 	return true;
3415 }
3416 #endif	/* CONFIG_NUMA */
3417 
3418 /*
3419  * The restriction on ZONE_DMA32 as being a suitable zone to use to avoid
3420  * fragmentation is subtle. If the preferred zone was HIGHMEM then
3421  * premature use of a lower zone may cause lowmem pressure problems that
3422  * are worse than fragmentation. If the next zone is ZONE_DMA then it is
3423  * probably too small. It only makes sense to spread allocations to avoid
3424  * fragmentation between the Normal and DMA32 zones.
3425  */
3426 static inline unsigned int
3427 alloc_flags_nofragment(struct zone *zone, gfp_t gfp_mask)
3428 {
3429 	unsigned int alloc_flags = 0;
3430 
3431 	if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3432 		alloc_flags |= ALLOC_KSWAPD;
3433 
3434 #ifdef CONFIG_ZONE_DMA32
3435 	if (!zone)
3436 		return alloc_flags;
3437 
3438 	if (zone_idx(zone) != ZONE_NORMAL)
3439 		return alloc_flags;
3440 
3441 	/*
3442 	 * If ZONE_DMA32 exists, assume it is the one after ZONE_NORMAL and
3443 	 * the pointer is within zone->zone_pgdat->node_zones[]. Also assume
3444 	 * on UMA that if Normal is populated then so is DMA32.
3445 	 */
3446 	BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1);
3447 	if (nr_online_nodes > 1 && !populated_zone(--zone))
3448 		return alloc_flags;
3449 
3450 	alloc_flags |= ALLOC_NOFRAGMENT;
3451 #endif /* CONFIG_ZONE_DMA32 */
3452 	return alloc_flags;
3453 }
3454 
3455 /*
3456  * get_page_from_freelist goes through the zonelist trying to allocate
3457  * a page.
3458  */
3459 static struct page *
3460 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
3461 						const struct alloc_context *ac)
3462 {
3463 	struct zoneref *z;
3464 	struct zone *zone;
3465 	struct pglist_data *last_pgdat_dirty_limit = NULL;
3466 	bool no_fallback;
3467 
3468 retry:
3469 	/*
3470 	 * Scan zonelist, looking for a zone with enough free.
3471 	 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
3472 	 */
3473 	no_fallback = alloc_flags & ALLOC_NOFRAGMENT;
3474 	z = ac->preferred_zoneref;
3475 	for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3476 								ac->nodemask) {
3477 		struct page *page;
3478 		unsigned long mark;
3479 
3480 		if (cpusets_enabled() &&
3481 			(alloc_flags & ALLOC_CPUSET) &&
3482 			!__cpuset_zone_allowed(zone, gfp_mask))
3483 				continue;
3484 		/*
3485 		 * When allocating a page cache page for writing, we
3486 		 * want to get it from a node that is within its dirty
3487 		 * limit, such that no single node holds more than its
3488 		 * proportional share of globally allowed dirty pages.
3489 		 * The dirty limits take into account the node's
3490 		 * lowmem reserves and high watermark so that kswapd
3491 		 * should be able to balance it without having to
3492 		 * write pages from its LRU list.
3493 		 *
3494 		 * XXX: For now, allow allocations to potentially
3495 		 * exceed the per-node dirty limit in the slowpath
3496 		 * (spread_dirty_pages unset) before going into reclaim,
3497 		 * which is important when on a NUMA setup the allowed
3498 		 * nodes are together not big enough to reach the
3499 		 * global limit.  The proper fix for these situations
3500 		 * will require awareness of nodes in the
3501 		 * dirty-throttling and the flusher threads.
3502 		 */
3503 		if (ac->spread_dirty_pages) {
3504 			if (last_pgdat_dirty_limit == zone->zone_pgdat)
3505 				continue;
3506 
3507 			if (!node_dirty_ok(zone->zone_pgdat)) {
3508 				last_pgdat_dirty_limit = zone->zone_pgdat;
3509 				continue;
3510 			}
3511 		}
3512 
3513 		if (no_fallback && nr_online_nodes > 1 &&
3514 		    zone != ac->preferred_zoneref->zone) {
3515 			int local_nid;
3516 
3517 			/*
3518 			 * If moving to a remote node, retry but allow
3519 			 * fragmenting fallbacks. Locality is more important
3520 			 * than fragmentation avoidance.
3521 			 */
3522 			local_nid = zone_to_nid(ac->preferred_zoneref->zone);
3523 			if (zone_to_nid(zone) != local_nid) {
3524 				alloc_flags &= ~ALLOC_NOFRAGMENT;
3525 				goto retry;
3526 			}
3527 		}
3528 
3529 		mark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);
3530 		if (!zone_watermark_fast(zone, order, mark,
3531 				       ac_classzone_idx(ac), alloc_flags)) {
3532 			int ret;
3533 
3534 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3535 			/*
3536 			 * Watermark failed for this zone, but see if we can
3537 			 * grow this zone if it contains deferred pages.
3538 			 */
3539 			if (static_branch_unlikely(&deferred_pages)) {
3540 				if (_deferred_grow_zone(zone, order))
3541 					goto try_this_zone;
3542 			}
3543 #endif
3544 			/* Checked here to keep the fast path fast */
3545 			BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
3546 			if (alloc_flags & ALLOC_NO_WATERMARKS)
3547 				goto try_this_zone;
3548 
3549 			if (node_reclaim_mode == 0 ||
3550 			    !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
3551 				continue;
3552 
3553 			ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
3554 			switch (ret) {
3555 			case NODE_RECLAIM_NOSCAN:
3556 				/* did not scan */
3557 				continue;
3558 			case NODE_RECLAIM_FULL:
3559 				/* scanned but unreclaimable */
3560 				continue;
3561 			default:
3562 				/* did we reclaim enough */
3563 				if (zone_watermark_ok(zone, order, mark,
3564 						ac_classzone_idx(ac), alloc_flags))
3565 					goto try_this_zone;
3566 
3567 				continue;
3568 			}
3569 		}
3570 
3571 try_this_zone:
3572 		page = rmqueue(ac->preferred_zoneref->zone, zone, order,
3573 				gfp_mask, alloc_flags, ac->migratetype);
3574 		if (page) {
3575 			prep_new_page(page, order, gfp_mask, alloc_flags);
3576 
3577 			/*
3578 			 * If this is a high-order atomic allocation then check
3579 			 * if the pageblock should be reserved for the future
3580 			 */
3581 			if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3582 				reserve_highatomic_pageblock(page, zone, order);
3583 
3584 			return page;
3585 		} else {
3586 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3587 			/* Try again if zone has deferred pages */
3588 			if (static_branch_unlikely(&deferred_pages)) {
3589 				if (_deferred_grow_zone(zone, order))
3590 					goto try_this_zone;
3591 			}
3592 #endif
3593 		}
3594 	}
3595 
3596 	/*
3597 	 * It's possible on a UMA machine to get through all zones that are
3598 	 * fragmented. If avoiding fragmentation, reset and try again.
3599 	 */
3600 	if (no_fallback) {
3601 		alloc_flags &= ~ALLOC_NOFRAGMENT;
3602 		goto retry;
3603 	}
3604 
3605 	return NULL;
3606 }
3607 
3608 static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
3609 {
3610 	unsigned int filter = SHOW_MEM_FILTER_NODES;
3611 	static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1);
3612 
3613 	if (!__ratelimit(&show_mem_rs))
3614 		return;
3615 
3616 	/*
3617 	 * This documents exceptions given to allocations in certain
3618 	 * contexts that are allowed to allocate outside current's set
3619 	 * of allowed nodes.
3620 	 */
3621 	if (!(gfp_mask & __GFP_NOMEMALLOC))
3622 		if (tsk_is_oom_victim(current) ||
3623 		    (current->flags & (PF_MEMALLOC | PF_EXITING)))
3624 			filter &= ~SHOW_MEM_FILTER_NODES;
3625 	if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
3626 		filter &= ~SHOW_MEM_FILTER_NODES;
3627 
3628 	show_mem(filter, nodemask);
3629 }
3630 
3631 void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
3632 {
3633 	struct va_format vaf;
3634 	va_list args;
3635 	static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL,
3636 				      DEFAULT_RATELIMIT_BURST);
3637 
3638 	if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
3639 		return;
3640 
3641 	va_start(args, fmt);
3642 	vaf.fmt = fmt;
3643 	vaf.va = &args;
3644 	pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl",
3645 			current->comm, &vaf, gfp_mask, &gfp_mask,
3646 			nodemask_pr_args(nodemask));
3647 	va_end(args);
3648 
3649 	cpuset_print_current_mems_allowed();
3650 	pr_cont("\n");
3651 	dump_stack();
3652 	warn_alloc_show_mem(gfp_mask, nodemask);
3653 }
3654 
3655 static inline struct page *
3656 __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
3657 			      unsigned int alloc_flags,
3658 			      const struct alloc_context *ac)
3659 {
3660 	struct page *page;
3661 
3662 	page = get_page_from_freelist(gfp_mask, order,
3663 			alloc_flags|ALLOC_CPUSET, ac);
3664 	/*
3665 	 * fallback to ignore cpuset restriction if our nodes
3666 	 * are depleted
3667 	 */
3668 	if (!page)
3669 		page = get_page_from_freelist(gfp_mask, order,
3670 				alloc_flags, ac);
3671 
3672 	return page;
3673 }
3674 
3675 static inline struct page *
3676 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
3677 	const struct alloc_context *ac, unsigned long *did_some_progress)
3678 {
3679 	struct oom_control oc = {
3680 		.zonelist = ac->zonelist,
3681 		.nodemask = ac->nodemask,
3682 		.memcg = NULL,
3683 		.gfp_mask = gfp_mask,
3684 		.order = order,
3685 	};
3686 	struct page *page;
3687 
3688 	*did_some_progress = 0;
3689 
3690 	/*
3691 	 * Acquire the oom lock.  If that fails, somebody else is
3692 	 * making progress for us.
3693 	 */
3694 	if (!mutex_trylock(&oom_lock)) {
3695 		*did_some_progress = 1;
3696 		schedule_timeout_uninterruptible(1);
3697 		return NULL;
3698 	}
3699 
3700 	/*
3701 	 * Go through the zonelist yet one more time, keep very high watermark
3702 	 * here, this is only to catch a parallel oom killing, we must fail if
3703 	 * we're still under heavy pressure. But make sure that this reclaim
3704 	 * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
3705 	 * allocation which will never fail due to oom_lock already held.
3706 	 */
3707 	page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
3708 				      ~__GFP_DIRECT_RECLAIM, order,
3709 				      ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
3710 	if (page)
3711 		goto out;
3712 
3713 	/* Coredumps can quickly deplete all memory reserves */
3714 	if (current->flags & PF_DUMPCORE)
3715 		goto out;
3716 	/* The OOM killer will not help higher order allocs */
3717 	if (order > PAGE_ALLOC_COSTLY_ORDER)
3718 		goto out;
3719 	/*
3720 	 * We have already exhausted all our reclaim opportunities without any
3721 	 * success so it is time to admit defeat. We will skip the OOM killer
3722 	 * because it is very likely that the caller has a more reasonable
3723 	 * fallback than shooting a random task.
3724 	 */
3725 	if (gfp_mask & __GFP_RETRY_MAYFAIL)
3726 		goto out;
3727 	/* The OOM killer does not needlessly kill tasks for lowmem */
3728 	if (ac->high_zoneidx < ZONE_NORMAL)
3729 		goto out;
3730 	if (pm_suspended_storage())
3731 		goto out;
3732 	/*
3733 	 * XXX: GFP_NOFS allocations should rather fail than rely on
3734 	 * other request to make a forward progress.
3735 	 * We are in an unfortunate situation where out_of_memory cannot
3736 	 * do much for this context but let's try it to at least get
3737 	 * access to memory reserved if the current task is killed (see
3738 	 * out_of_memory). Once filesystems are ready to handle allocation
3739 	 * failures more gracefully we should just bail out here.
3740 	 */
3741 
3742 	/* The OOM killer may not free memory on a specific node */
3743 	if (gfp_mask & __GFP_THISNODE)
3744 		goto out;
3745 
3746 	/* Exhausted what can be done so it's blame time */
3747 	if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3748 		*did_some_progress = 1;
3749 
3750 		/*
3751 		 * Help non-failing allocations by giving them access to memory
3752 		 * reserves
3753 		 */
3754 		if (gfp_mask & __GFP_NOFAIL)
3755 			page = __alloc_pages_cpuset_fallback(gfp_mask, order,
3756 					ALLOC_NO_WATERMARKS, ac);
3757 	}
3758 out:
3759 	mutex_unlock(&oom_lock);
3760 	return page;
3761 }
3762 
3763 /*
3764  * Maximum number of compaction retries wit a progress before OOM
3765  * killer is consider as the only way to move forward.
3766  */
3767 #define MAX_COMPACT_RETRIES 16
3768 
3769 #ifdef CONFIG_COMPACTION
3770 /* Try memory compaction for high-order allocations before reclaim */
3771 static struct page *
3772 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3773 		unsigned int alloc_flags, const struct alloc_context *ac,
3774 		enum compact_priority prio, enum compact_result *compact_result)
3775 {
3776 	struct page *page = NULL;
3777 	unsigned long pflags;
3778 	unsigned int noreclaim_flag;
3779 
3780 	if (!order)
3781 		return NULL;
3782 
3783 	psi_memstall_enter(&pflags);
3784 	noreclaim_flag = memalloc_noreclaim_save();
3785 
3786 	*compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3787 								prio, &page);
3788 
3789 	memalloc_noreclaim_restore(noreclaim_flag);
3790 	psi_memstall_leave(&pflags);
3791 
3792 	/*
3793 	 * At least in one zone compaction wasn't deferred or skipped, so let's
3794 	 * count a compaction stall
3795 	 */
3796 	count_vm_event(COMPACTSTALL);
3797 
3798 	/* Prep a captured page if available */
3799 	if (page)
3800 		prep_new_page(page, order, gfp_mask, alloc_flags);
3801 
3802 	/* Try get a page from the freelist if available */
3803 	if (!page)
3804 		page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3805 
3806 	if (page) {
3807 		struct zone *zone = page_zone(page);
3808 
3809 		zone->compact_blockskip_flush = false;
3810 		compaction_defer_reset(zone, order, true);
3811 		count_vm_event(COMPACTSUCCESS);
3812 		return page;
3813 	}
3814 
3815 	/*
3816 	 * It's bad if compaction run occurs and fails. The most likely reason
3817 	 * is that pages exist, but not enough to satisfy watermarks.
3818 	 */
3819 	count_vm_event(COMPACTFAIL);
3820 
3821 	cond_resched();
3822 
3823 	return NULL;
3824 }
3825 
3826 static inline bool
3827 should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3828 		     enum compact_result compact_result,
3829 		     enum compact_priority *compact_priority,
3830 		     int *compaction_retries)
3831 {
3832 	int max_retries = MAX_COMPACT_RETRIES;
3833 	int min_priority;
3834 	bool ret = false;
3835 	int retries = *compaction_retries;
3836 	enum compact_priority priority = *compact_priority;
3837 
3838 	if (!order)
3839 		return false;
3840 
3841 	if (compaction_made_progress(compact_result))
3842 		(*compaction_retries)++;
3843 
3844 	/*
3845 	 * compaction considers all the zone as desperately out of memory
3846 	 * so it doesn't really make much sense to retry except when the
3847 	 * failure could be caused by insufficient priority
3848 	 */
3849 	if (compaction_failed(compact_result))
3850 		goto check_priority;
3851 
3852 	/*
3853 	 * make sure the compaction wasn't deferred or didn't bail out early
3854 	 * due to locks contention before we declare that we should give up.
3855 	 * But do not retry if the given zonelist is not suitable for
3856 	 * compaction.
3857 	 */
3858 	if (compaction_withdrawn(compact_result)) {
3859 		ret = compaction_zonelist_suitable(ac, order, alloc_flags);
3860 		goto out;
3861 	}
3862 
3863 	/*
3864 	 * !costly requests are much more important than __GFP_RETRY_MAYFAIL
3865 	 * costly ones because they are de facto nofail and invoke OOM
3866 	 * killer to move on while costly can fail and users are ready
3867 	 * to cope with that. 1/4 retries is rather arbitrary but we
3868 	 * would need much more detailed feedback from compaction to
3869 	 * make a better decision.
3870 	 */
3871 	if (order > PAGE_ALLOC_COSTLY_ORDER)
3872 		max_retries /= 4;
3873 	if (*compaction_retries <= max_retries) {
3874 		ret = true;
3875 		goto out;
3876 	}
3877 
3878 	/*
3879 	 * Make sure there are attempts at the highest priority if we exhausted
3880 	 * all retries or failed at the lower priorities.
3881 	 */
3882 check_priority:
3883 	min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
3884 			MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
3885 
3886 	if (*compact_priority > min_priority) {
3887 		(*compact_priority)--;
3888 		*compaction_retries = 0;
3889 		ret = true;
3890 	}
3891 out:
3892 	trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
3893 	return ret;
3894 }
3895 #else
3896 static inline struct page *
3897 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3898 		unsigned int alloc_flags, const struct alloc_context *ac,
3899 		enum compact_priority prio, enum compact_result *compact_result)
3900 {
3901 	*compact_result = COMPACT_SKIPPED;
3902 	return NULL;
3903 }
3904 
3905 static inline bool
3906 should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3907 		     enum compact_result compact_result,
3908 		     enum compact_priority *compact_priority,
3909 		     int *compaction_retries)
3910 {
3911 	struct zone *zone;
3912 	struct zoneref *z;
3913 
3914 	if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3915 		return false;
3916 
3917 	/*
3918 	 * There are setups with compaction disabled which would prefer to loop
3919 	 * inside the allocator rather than hit the oom killer prematurely.
3920 	 * Let's give them a good hope and keep retrying while the order-0
3921 	 * watermarks are OK.
3922 	 */
3923 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3924 					ac->nodemask) {
3925 		if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3926 					ac_classzone_idx(ac), alloc_flags))
3927 			return true;
3928 	}
3929 	return false;
3930 }
3931 #endif /* CONFIG_COMPACTION */
3932 
3933 #ifdef CONFIG_LOCKDEP
3934 static struct lockdep_map __fs_reclaim_map =
3935 	STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
3936 
3937 static bool __need_fs_reclaim(gfp_t gfp_mask)
3938 {
3939 	gfp_mask = current_gfp_context(gfp_mask);
3940 
3941 	/* no reclaim without waiting on it */
3942 	if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
3943 		return false;
3944 
3945 	/* this guy won't enter reclaim */
3946 	if (current->flags & PF_MEMALLOC)
3947 		return false;
3948 
3949 	/* We're only interested __GFP_FS allocations for now */
3950 	if (!(gfp_mask & __GFP_FS))
3951 		return false;
3952 
3953 	if (gfp_mask & __GFP_NOLOCKDEP)
3954 		return false;
3955 
3956 	return true;
3957 }
3958 
3959 void __fs_reclaim_acquire(void)
3960 {
3961 	lock_map_acquire(&__fs_reclaim_map);
3962 }
3963 
3964 void __fs_reclaim_release(void)
3965 {
3966 	lock_map_release(&__fs_reclaim_map);
3967 }
3968 
3969 void fs_reclaim_acquire(gfp_t gfp_mask)
3970 {
3971 	if (__need_fs_reclaim(gfp_mask))
3972 		__fs_reclaim_acquire();
3973 }
3974 EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
3975 
3976 void fs_reclaim_release(gfp_t gfp_mask)
3977 {
3978 	if (__need_fs_reclaim(gfp_mask))
3979 		__fs_reclaim_release();
3980 }
3981 EXPORT_SYMBOL_GPL(fs_reclaim_release);
3982 #endif
3983 
3984 /* Perform direct synchronous page reclaim */
3985 static int
3986 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
3987 					const struct alloc_context *ac)
3988 {
3989 	struct reclaim_state reclaim_state;
3990 	int progress;
3991 	unsigned int noreclaim_flag;
3992 	unsigned long pflags;
3993 
3994 	cond_resched();
3995 
3996 	/* We now go into synchronous reclaim */
3997 	cpuset_memory_pressure_bump();
3998 	psi_memstall_enter(&pflags);
3999 	fs_reclaim_acquire(gfp_mask);
4000 	noreclaim_flag = memalloc_noreclaim_save();
4001 	reclaim_state.reclaimed_slab = 0;
4002 	current->reclaim_state = &reclaim_state;
4003 
4004 	progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
4005 								ac->nodemask);
4006 
4007 	current->reclaim_state = NULL;
4008 	memalloc_noreclaim_restore(noreclaim_flag);
4009 	fs_reclaim_release(gfp_mask);
4010 	psi_memstall_leave(&pflags);
4011 
4012 	cond_resched();
4013 
4014 	return progress;
4015 }
4016 
4017 /* The really slow allocator path where we enter direct reclaim */
4018 static inline struct page *
4019 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
4020 		unsigned int alloc_flags, const struct alloc_context *ac,
4021 		unsigned long *did_some_progress)
4022 {
4023 	struct page *page = NULL;
4024 	bool drained = false;
4025 
4026 	*did_some_progress = __perform_reclaim(gfp_mask, order, ac);
4027 	if (unlikely(!(*did_some_progress)))
4028 		return NULL;
4029 
4030 retry:
4031 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4032 
4033 	/*
4034 	 * If an allocation failed after direct reclaim, it could be because
4035 	 * pages are pinned on the per-cpu lists or in high alloc reserves.
4036 	 * Shrink them them and try again
4037 	 */
4038 	if (!page && !drained) {
4039 		unreserve_highatomic_pageblock(ac, false);
4040 		drain_all_pages(NULL);
4041 		drained = true;
4042 		goto retry;
4043 	}
4044 
4045 	return page;
4046 }
4047 
4048 static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
4049 			     const struct alloc_context *ac)
4050 {
4051 	struct zoneref *z;
4052 	struct zone *zone;
4053 	pg_data_t *last_pgdat = NULL;
4054 	enum zone_type high_zoneidx = ac->high_zoneidx;
4055 
4056 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx,
4057 					ac->nodemask) {
4058 		if (last_pgdat != zone->zone_pgdat)
4059 			wakeup_kswapd(zone, gfp_mask, order, high_zoneidx);
4060 		last_pgdat = zone->zone_pgdat;
4061 	}
4062 }
4063 
4064 static inline unsigned int
4065 gfp_to_alloc_flags(gfp_t gfp_mask)
4066 {
4067 	unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
4068 
4069 	/* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
4070 	BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
4071 
4072 	/*
4073 	 * The caller may dip into page reserves a bit more if the caller
4074 	 * cannot run direct reclaim, or if the caller has realtime scheduling
4075 	 * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
4076 	 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
4077 	 */
4078 	alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
4079 
4080 	if (gfp_mask & __GFP_ATOMIC) {
4081 		/*
4082 		 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
4083 		 * if it can't schedule.
4084 		 */
4085 		if (!(gfp_mask & __GFP_NOMEMALLOC))
4086 			alloc_flags |= ALLOC_HARDER;
4087 		/*
4088 		 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
4089 		 * comment for __cpuset_node_allowed().
4090 		 */
4091 		alloc_flags &= ~ALLOC_CPUSET;
4092 	} else if (unlikely(rt_task(current)) && !in_interrupt())
4093 		alloc_flags |= ALLOC_HARDER;
4094 
4095 	if (gfp_mask & __GFP_KSWAPD_RECLAIM)
4096 		alloc_flags |= ALLOC_KSWAPD;
4097 
4098 #ifdef CONFIG_CMA
4099 	if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
4100 		alloc_flags |= ALLOC_CMA;
4101 #endif
4102 	return alloc_flags;
4103 }
4104 
4105 static bool oom_reserves_allowed(struct task_struct *tsk)
4106 {
4107 	if (!tsk_is_oom_victim(tsk))
4108 		return false;
4109 
4110 	/*
4111 	 * !MMU doesn't have oom reaper so give access to memory reserves
4112 	 * only to the thread with TIF_MEMDIE set
4113 	 */
4114 	if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
4115 		return false;
4116 
4117 	return true;
4118 }
4119 
4120 /*
4121  * Distinguish requests which really need access to full memory
4122  * reserves from oom victims which can live with a portion of it
4123  */
4124 static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
4125 {
4126 	if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
4127 		return 0;
4128 	if (gfp_mask & __GFP_MEMALLOC)
4129 		return ALLOC_NO_WATERMARKS;
4130 	if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
4131 		return ALLOC_NO_WATERMARKS;
4132 	if (!in_interrupt()) {
4133 		if (current->flags & PF_MEMALLOC)
4134 			return ALLOC_NO_WATERMARKS;
4135 		else if (oom_reserves_allowed(current))
4136 			return ALLOC_OOM;
4137 	}
4138 
4139 	return 0;
4140 }
4141 
4142 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
4143 {
4144 	return !!__gfp_pfmemalloc_flags(gfp_mask);
4145 }
4146 
4147 /*
4148  * Checks whether it makes sense to retry the reclaim to make a forward progress
4149  * for the given allocation request.
4150  *
4151  * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
4152  * without success, or when we couldn't even meet the watermark if we
4153  * reclaimed all remaining pages on the LRU lists.
4154  *
4155  * Returns true if a retry is viable or false to enter the oom path.
4156  */
4157 static inline bool
4158 should_reclaim_retry(gfp_t gfp_mask, unsigned order,
4159 		     struct alloc_context *ac, int alloc_flags,
4160 		     bool did_some_progress, int *no_progress_loops)
4161 {
4162 	struct zone *zone;
4163 	struct zoneref *z;
4164 	bool ret = false;
4165 
4166 	/*
4167 	 * Costly allocations might have made a progress but this doesn't mean
4168 	 * their order will become available due to high fragmentation so
4169 	 * always increment the no progress counter for them
4170 	 */
4171 	if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
4172 		*no_progress_loops = 0;
4173 	else
4174 		(*no_progress_loops)++;
4175 
4176 	/*
4177 	 * Make sure we converge to OOM if we cannot make any progress
4178 	 * several times in the row.
4179 	 */
4180 	if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
4181 		/* Before OOM, exhaust highatomic_reserve */
4182 		return unreserve_highatomic_pageblock(ac, true);
4183 	}
4184 
4185 	/*
4186 	 * Keep reclaiming pages while there is a chance this will lead
4187 	 * somewhere.  If none of the target zones can satisfy our allocation
4188 	 * request even if all reclaimable pages are considered then we are
4189 	 * screwed and have to go OOM.
4190 	 */
4191 	for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
4192 					ac->nodemask) {
4193 		unsigned long available;
4194 		unsigned long reclaimable;
4195 		unsigned long min_wmark = min_wmark_pages(zone);
4196 		bool wmark;
4197 
4198 		available = reclaimable = zone_reclaimable_pages(zone);
4199 		available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
4200 
4201 		/*
4202 		 * Would the allocation succeed if we reclaimed all
4203 		 * reclaimable pages?
4204 		 */
4205 		wmark = __zone_watermark_ok(zone, order, min_wmark,
4206 				ac_classzone_idx(ac), alloc_flags, available);
4207 		trace_reclaim_retry_zone(z, order, reclaimable,
4208 				available, min_wmark, *no_progress_loops, wmark);
4209 		if (wmark) {
4210 			/*
4211 			 * If we didn't make any progress and have a lot of
4212 			 * dirty + writeback pages then we should wait for
4213 			 * an IO to complete to slow down the reclaim and
4214 			 * prevent from pre mature OOM
4215 			 */
4216 			if (!did_some_progress) {
4217 				unsigned long write_pending;
4218 
4219 				write_pending = zone_page_state_snapshot(zone,
4220 							NR_ZONE_WRITE_PENDING);
4221 
4222 				if (2 * write_pending > reclaimable) {
4223 					congestion_wait(BLK_RW_ASYNC, HZ/10);
4224 					return true;
4225 				}
4226 			}
4227 
4228 			ret = true;
4229 			goto out;
4230 		}
4231 	}
4232 
4233 out:
4234 	/*
4235 	 * Memory allocation/reclaim might be called from a WQ context and the
4236 	 * current implementation of the WQ concurrency control doesn't
4237 	 * recognize that a particular WQ is congested if the worker thread is
4238 	 * looping without ever sleeping. Therefore we have to do a short sleep
4239 	 * here rather than calling cond_resched().
4240 	 */
4241 	if (current->flags & PF_WQ_WORKER)
4242 		schedule_timeout_uninterruptible(1);
4243 	else
4244 		cond_resched();
4245 	return ret;
4246 }
4247 
4248 static inline bool
4249 check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4250 {
4251 	/*
4252 	 * It's possible that cpuset's mems_allowed and the nodemask from
4253 	 * mempolicy don't intersect. This should be normally dealt with by
4254 	 * policy_nodemask(), but it's possible to race with cpuset update in
4255 	 * such a way the check therein was true, and then it became false
4256 	 * before we got our cpuset_mems_cookie here.
4257 	 * This assumes that for all allocations, ac->nodemask can come only
4258 	 * from MPOL_BIND mempolicy (whose documented semantics is to be ignored
4259 	 * when it does not intersect with the cpuset restrictions) or the
4260 	 * caller can deal with a violated nodemask.
4261 	 */
4262 	if (cpusets_enabled() && ac->nodemask &&
4263 			!cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4264 		ac->nodemask = NULL;
4265 		return true;
4266 	}
4267 
4268 	/*
4269 	 * When updating a task's mems_allowed or mempolicy nodemask, it is
4270 	 * possible to race with parallel threads in such a way that our
4271 	 * allocation can fail while the mask is being updated. If we are about
4272 	 * to fail, check if the cpuset changed during allocation and if so,
4273 	 * retry.
4274 	 */
4275 	if (read_mems_allowed_retry(cpuset_mems_cookie))
4276 		return true;
4277 
4278 	return false;
4279 }
4280 
4281 static inline struct page *
4282 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4283 						struct alloc_context *ac)
4284 {
4285 	bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4286 	const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4287 	struct page *page = NULL;
4288 	unsigned int alloc_flags;
4289 	unsigned long did_some_progress;
4290 	enum compact_priority compact_priority;
4291 	enum compact_result compact_result;
4292 	int compaction_retries;
4293 	int no_progress_loops;
4294 	unsigned int cpuset_mems_cookie;
4295 	int reserve_flags;
4296 
4297 	/*
4298 	 * We also sanity check to catch abuse of atomic reserves being used by
4299 	 * callers that are not in atomic context.
4300 	 */
4301 	if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4302 				(__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4303 		gfp_mask &= ~__GFP_ATOMIC;
4304 
4305 retry_cpuset:
4306 	compaction_retries = 0;
4307 	no_progress_loops = 0;
4308 	compact_priority = DEF_COMPACT_PRIORITY;
4309 	cpuset_mems_cookie = read_mems_allowed_begin();
4310 
4311 	/*
4312 	 * The fast path uses conservative alloc_flags to succeed only until
4313 	 * kswapd needs to be woken up, and to avoid the cost of setting up
4314 	 * alloc_flags precisely. So we do that now.
4315 	 */
4316 	alloc_flags = gfp_to_alloc_flags(gfp_mask);
4317 
4318 	/*
4319 	 * We need to recalculate the starting point for the zonelist iterator
4320 	 * because we might have used different nodemask in the fast path, or
4321 	 * there was a cpuset modification and we are retrying - otherwise we
4322 	 * could end up iterating over non-eligible zones endlessly.
4323 	 */
4324 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4325 					ac->high_zoneidx, ac->nodemask);
4326 	if (!ac->preferred_zoneref->zone)
4327 		goto nopage;
4328 
4329 	if (alloc_flags & ALLOC_KSWAPD)
4330 		wake_all_kswapds(order, gfp_mask, ac);
4331 
4332 	/*
4333 	 * The adjusted alloc_flags might result in immediate success, so try
4334 	 * that first
4335 	 */
4336 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4337 	if (page)
4338 		goto got_pg;
4339 
4340 	/*
4341 	 * For costly allocations, try direct compaction first, as it's likely
4342 	 * that we have enough base pages and don't need to reclaim. For non-
4343 	 * movable high-order allocations, do that as well, as compaction will
4344 	 * try prevent permanent fragmentation by migrating from blocks of the
4345 	 * same migratetype.
4346 	 * Don't try this for allocations that are allowed to ignore
4347 	 * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen.
4348 	 */
4349 	if (can_direct_reclaim &&
4350 			(costly_order ||
4351 			   (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4352 			&& !gfp_pfmemalloc_allowed(gfp_mask)) {
4353 		page = __alloc_pages_direct_compact(gfp_mask, order,
4354 						alloc_flags, ac,
4355 						INIT_COMPACT_PRIORITY,
4356 						&compact_result);
4357 		if (page)
4358 			goto got_pg;
4359 
4360 		/*
4361 		 * Checks for costly allocations with __GFP_NORETRY, which
4362 		 * includes THP page fault allocations
4363 		 */
4364 		if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4365 			/*
4366 			 * If compaction is deferred for high-order allocations,
4367 			 * it is because sync compaction recently failed. If
4368 			 * this is the case and the caller requested a THP
4369 			 * allocation, we do not want to heavily disrupt the
4370 			 * system, so we fail the allocation instead of entering
4371 			 * direct reclaim.
4372 			 */
4373 			if (compact_result == COMPACT_DEFERRED)
4374 				goto nopage;
4375 
4376 			/*
4377 			 * Looks like reclaim/compaction is worth trying, but
4378 			 * sync compaction could be very expensive, so keep
4379 			 * using async compaction.
4380 			 */
4381 			compact_priority = INIT_COMPACT_PRIORITY;
4382 		}
4383 	}
4384 
4385 retry:
4386 	/* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
4387 	if (alloc_flags & ALLOC_KSWAPD)
4388 		wake_all_kswapds(order, gfp_mask, ac);
4389 
4390 	reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
4391 	if (reserve_flags)
4392 		alloc_flags = reserve_flags;
4393 
4394 	/*
4395 	 * Reset the nodemask and zonelist iterators if memory policies can be
4396 	 * ignored. These allocations are high priority and system rather than
4397 	 * user oriented.
4398 	 */
4399 	if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
4400 		ac->nodemask = NULL;
4401 		ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4402 					ac->high_zoneidx, ac->nodemask);
4403 	}
4404 
4405 	/* Attempt with potentially adjusted zonelist and alloc_flags */
4406 	page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4407 	if (page)
4408 		goto got_pg;
4409 
4410 	/* Caller is not willing to reclaim, we can't balance anything */
4411 	if (!can_direct_reclaim)
4412 		goto nopage;
4413 
4414 	/* Avoid recursion of direct reclaim */
4415 	if (current->flags & PF_MEMALLOC)
4416 		goto nopage;
4417 
4418 	/* Try direct reclaim and then allocating */
4419 	page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
4420 							&did_some_progress);
4421 	if (page)
4422 		goto got_pg;
4423 
4424 	/* Try direct compaction and then allocating */
4425 	page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
4426 					compact_priority, &compact_result);
4427 	if (page)
4428 		goto got_pg;
4429 
4430 	/* Do not loop if specifically requested */
4431 	if (gfp_mask & __GFP_NORETRY)
4432 		goto nopage;
4433 
4434 	/*
4435 	 * Do not retry costly high order allocations unless they are
4436 	 * __GFP_RETRY_MAYFAIL
4437 	 */
4438 	if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
4439 		goto nopage;
4440 
4441 	if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
4442 				 did_some_progress > 0, &no_progress_loops))
4443 		goto retry;
4444 
4445 	/*
4446 	 * It doesn't make any sense to retry for the compaction if the order-0
4447 	 * reclaim is not able to make any progress because the current
4448 	 * implementation of the compaction depends on the sufficient amount
4449 	 * of free memory (see __compaction_suitable)
4450 	 */
4451 	if (did_some_progress > 0 &&
4452 			should_compact_retry(ac, order, alloc_flags,
4453 				compact_result, &compact_priority,
4454 				&compaction_retries))
4455 		goto retry;
4456 
4457 
4458 	/* Deal with possible cpuset update races before we start OOM killing */
4459 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
4460 		goto retry_cpuset;
4461 
4462 	/* Reclaim has failed us, start killing things */
4463 	page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
4464 	if (page)
4465 		goto got_pg;
4466 
4467 	/* Avoid allocations with no watermarks from looping endlessly */
4468 	if (tsk_is_oom_victim(current) &&
4469 	    (alloc_flags == ALLOC_OOM ||
4470 	     (gfp_mask & __GFP_NOMEMALLOC)))
4471 		goto nopage;
4472 
4473 	/* Retry as long as the OOM killer is making progress */
4474 	if (did_some_progress) {
4475 		no_progress_loops = 0;
4476 		goto retry;
4477 	}
4478 
4479 nopage:
4480 	/* Deal with possible cpuset update races before we fail */
4481 	if (check_retry_cpuset(cpuset_mems_cookie, ac))
4482 		goto retry_cpuset;
4483 
4484 	/*
4485 	 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
4486 	 * we always retry
4487 	 */
4488 	if (gfp_mask & __GFP_NOFAIL) {
4489 		/*
4490 		 * All existing users of the __GFP_NOFAIL are blockable, so warn
4491 		 * of any new users that actually require GFP_NOWAIT
4492 		 */
4493 		if (WARN_ON_ONCE(!can_direct_reclaim))
4494 			goto fail;
4495 
4496 		/*
4497 		 * PF_MEMALLOC request from this context is rather bizarre
4498 		 * because we cannot reclaim anything and only can loop waiting
4499 		 * for somebody to do a work for us
4500 		 */
4501 		WARN_ON_ONCE(current->flags & PF_MEMALLOC);
4502 
4503 		/*
4504 		 * non failing costly orders are a hard requirement which we
4505 		 * are not prepared for much so let's warn about these users
4506 		 * so that we can identify them and convert them to something
4507 		 * else.
4508 		 */
4509 		WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
4510 
4511 		/*
4512 		 * Help non-failing allocations by giving them access to memory
4513 		 * reserves but do not use ALLOC_NO_WATERMARKS because this
4514 		 * could deplete whole memory reserves which would just make
4515 		 * the situation worse
4516 		 */
4517 		page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
4518 		if (page)
4519 			goto got_pg;
4520 
4521 		cond_resched();
4522 		goto retry;
4523 	}
4524 fail:
4525 	warn_alloc(gfp_mask, ac->nodemask,
4526 			"page allocation failure: order:%u", order);
4527 got_pg:
4528 	return page;
4529 }
4530 
4531 static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
4532 		int preferred_nid, nodemask_t *nodemask,
4533 		struct alloc_context *ac, gfp_t *alloc_mask,
4534 		unsigned int *alloc_flags)
4535 {
4536 	ac->high_zoneidx = gfp_zone(gfp_mask);
4537 	ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
4538 	ac->nodemask = nodemask;
4539 	ac->migratetype = gfpflags_to_migratetype(gfp_mask);
4540 
4541 	if (cpusets_enabled()) {
4542 		*alloc_mask |= __GFP_HARDWALL;
4543 		if (!ac->nodemask)
4544 			ac->nodemask = &cpuset_current_mems_allowed;
4545 		else
4546 			*alloc_flags |= ALLOC_CPUSET;
4547 	}
4548 
4549 	fs_reclaim_acquire(gfp_mask);
4550 	fs_reclaim_release(gfp_mask);
4551 
4552 	might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
4553 
4554 	if (should_fail_alloc_page(gfp_mask, order))
4555 		return false;
4556 
4557 	if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE)
4558 		*alloc_flags |= ALLOC_CMA;
4559 
4560 	return true;
4561 }
4562 
4563 /* Determine whether to spread dirty pages and what the first usable zone */
4564 static inline void finalise_ac(gfp_t gfp_mask, struct alloc_context *ac)
4565 {
4566 	/* Dirty zone balancing only done in the fast path */
4567 	ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
4568 
4569 	/*
4570 	 * The preferred zone is used for statistics but crucially it is
4571 	 * also used as the starting point for the zonelist iterator. It
4572 	 * may get reset for allocations that ignore memory policies.
4573 	 */
4574 	ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4575 					ac->high_zoneidx, ac->nodemask);
4576 }
4577 
4578 /*
4579  * This is the 'heart' of the zoned buddy allocator.
4580  */
4581 struct page *
4582 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
4583 							nodemask_t *nodemask)
4584 {
4585 	struct page *page;
4586 	unsigned int alloc_flags = ALLOC_WMARK_LOW;
4587 	gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
4588 	struct alloc_context ac = { };
4589 
4590 	/*
4591 	 * There are several places where we assume that the order value is sane
4592 	 * so bail out early if the request is out of bound.
4593 	 */
4594 	if (unlikely(order >= MAX_ORDER)) {
4595 		WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
4596 		return NULL;
4597 	}
4598 
4599 	gfp_mask &= gfp_allowed_mask;
4600 	alloc_mask = gfp_mask;
4601 	if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))
4602 		return NULL;
4603 
4604 	finalise_ac(gfp_mask, &ac);
4605 
4606 	/*
4607 	 * Forbid the first pass from falling back to types that fragment
4608 	 * memory until all local zones are considered.
4609 	 */
4610 	alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp_mask);
4611 
4612 	/* First allocation attempt */
4613 	page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
4614 	if (likely(page))
4615 		goto out;
4616 
4617 	/*
4618 	 * Apply scoped allocation constraints. This is mainly about GFP_NOFS
4619 	 * resp. GFP_NOIO which has to be inherited for all allocation requests
4620 	 * from a particular context which has been marked by
4621 	 * memalloc_no{fs,io}_{save,restore}.
4622 	 */
4623 	alloc_mask = current_gfp_context(gfp_mask);
4624 	ac.spread_dirty_pages = false;
4625 
4626 	/*
4627 	 * Restore the original nodemask if it was potentially replaced with
4628 	 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
4629 	 */
4630 	if (unlikely(ac.nodemask != nodemask))
4631 		ac.nodemask = nodemask;
4632 
4633 	page = __alloc_pages_slowpath(alloc_mask, order, &ac);
4634 
4635 out:
4636 	if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
4637 	    unlikely(__memcg_kmem_charge(page, gfp_mask, order) != 0)) {
4638 		__free_pages(page, order);
4639 		page = NULL;
4640 	}
4641 
4642 	trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
4643 
4644 	return page;
4645 }
4646 EXPORT_SYMBOL(__alloc_pages_nodemask);
4647 
4648 /*
4649  * Common helper functions. Never use with __GFP_HIGHMEM because the returned
4650  * address cannot represent highmem pages. Use alloc_pages and then kmap if
4651  * you need to access high mem.
4652  */
4653 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
4654 {
4655 	struct page *page;
4656 
4657 	page = alloc_pages(gfp_mask & ~__GFP_HIGHMEM, order);
4658 	if (!page)
4659 		return 0;
4660 	return (unsigned long) page_address(page);
4661 }
4662 EXPORT_SYMBOL(__get_free_pages);
4663 
4664 unsigned long get_zeroed_page(gfp_t gfp_mask)
4665 {
4666 	return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
4667 }
4668 EXPORT_SYMBOL(get_zeroed_page);
4669 
4670 static inline void free_the_page(struct page *page, unsigned int order)
4671 {
4672 	if (order == 0)		/* Via pcp? */
4673 		free_unref_page(page);
4674 	else
4675 		__free_pages_ok(page, order);
4676 }
4677 
4678 void __free_pages(struct page *page, unsigned int order)
4679 {
4680 	if (put_page_testzero(page))
4681 		free_the_page(page, order);
4682 }
4683 EXPORT_SYMBOL(__free_pages);
4684 
4685 void free_pages(unsigned long addr, unsigned int order)
4686 {
4687 	if (addr != 0) {
4688 		VM_BUG_ON(!virt_addr_valid((void *)addr));
4689 		__free_pages(virt_to_page((void *)addr), order);
4690 	}
4691 }
4692 
4693 EXPORT_SYMBOL(free_pages);
4694 
4695 /*
4696  * Page Fragment:
4697  *  An arbitrary-length arbitrary-offset area of memory which resides
4698  *  within a 0 or higher order page.  Multiple fragments within that page
4699  *  are individually refcounted, in the page's reference counter.
4700  *
4701  * The page_frag functions below provide a simple allocation framework for
4702  * page fragments.  This is used by the network stack and network device
4703  * drivers to provide a backing region of memory for use as either an
4704  * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4705  */
4706 static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
4707 					     gfp_t gfp_mask)
4708 {
4709 	struct page *page = NULL;
4710 	gfp_t gfp = gfp_mask;
4711 
4712 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4713 	gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
4714 		    __GFP_NOMEMALLOC;
4715 	page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
4716 				PAGE_FRAG_CACHE_MAX_ORDER);
4717 	nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
4718 #endif
4719 	if (unlikely(!page))
4720 		page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
4721 
4722 	nc->va = page ? page_address(page) : NULL;
4723 
4724 	return page;
4725 }
4726 
4727 void __page_frag_cache_drain(struct page *page, unsigned int count)
4728 {
4729 	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
4730 
4731 	if (page_ref_sub_and_test(page, count))
4732 		free_the_page(page, compound_order(page));
4733 }
4734 EXPORT_SYMBOL(__page_frag_cache_drain);
4735 
4736 void *page_frag_alloc(struct page_frag_cache *nc,
4737 		      unsigned int fragsz, gfp_t gfp_mask)
4738 {
4739 	unsigned int size = PAGE_SIZE;
4740 	struct page *page;
4741 	int offset;
4742 
4743 	if (unlikely(!nc->va)) {
4744 refill:
4745 		page = __page_frag_cache_refill(nc, gfp_mask);
4746 		if (!page)
4747 			return NULL;
4748 
4749 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4750 		/* if size can vary use size else just use PAGE_SIZE */
4751 		size = nc->size;
4752 #endif
4753 		/* Even if we own the page, we do not use atomic_set().
4754 		 * This would break get_page_unless_zero() users.
4755 		 */
4756 		page_ref_add(page, PAGE_FRAG_CACHE_MAX_SIZE);
4757 
4758 		/* reset page count bias and offset to start of new frag */
4759 		nc->pfmemalloc = page_is_pfmemalloc(page);
4760 		nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
4761 		nc->offset = size;
4762 	}
4763 
4764 	offset = nc->offset - fragsz;
4765 	if (unlikely(offset < 0)) {
4766 		page = virt_to_page(nc->va);
4767 
4768 		if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
4769 			goto refill;
4770 
4771 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4772 		/* if size can vary use size else just use PAGE_SIZE */
4773 		size = nc->size;
4774 #endif
4775 		/* OK, page count is 0, we can safely set it */
4776 		set_page_count(page, PAGE_FRAG_CACHE_MAX_SIZE + 1);
4777 
4778 		/* reset page count bias and offset to start of new frag */
4779 		nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1;
4780 		offset = size - fragsz;
4781 	}
4782 
4783 	nc->pagecnt_bias--;
4784 	nc->offset = offset;
4785 
4786 	return nc->va + offset;
4787 }
4788 EXPORT_SYMBOL(page_frag_alloc);
4789 
4790 /*
4791  * Frees a page fragment allocated out of either a compound or order 0 page.
4792  */
4793 void page_frag_free(void *addr)
4794 {
4795 	struct page *page = virt_to_head_page(addr);
4796 
4797 	if (unlikely(put_page_testzero(page)))
4798 		free_the_page(page, compound_order(page));
4799 }
4800 EXPORT_SYMBOL(page_frag_free);
4801 
4802 static void *make_alloc_exact(unsigned long addr, unsigned int order,
4803 		size_t size)
4804 {
4805 	if (addr) {
4806 		unsigned long alloc_end = addr + (PAGE_SIZE << order);
4807 		unsigned long used = addr + PAGE_ALIGN(size);
4808 
4809 		split_page(virt_to_page((void *)addr), order);
4810 		while (used < alloc_end) {
4811 			free_page(used);
4812 			used += PAGE_SIZE;
4813 		}
4814 	}
4815 	return (void *)addr;
4816 }
4817 
4818 /**
4819  * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4820  * @size: the number of bytes to allocate
4821  * @gfp_mask: GFP flags for the allocation
4822  *
4823  * This function is similar to alloc_pages(), except that it allocates the
4824  * minimum number of pages to satisfy the request.  alloc_pages() can only
4825  * allocate memory in power-of-two pages.
4826  *
4827  * This function is also limited by MAX_ORDER.
4828  *
4829  * Memory allocated by this function must be released by free_pages_exact().
4830  *
4831  * Return: pointer to the allocated area or %NULL in case of error.
4832  */
4833 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
4834 {
4835 	unsigned int order = get_order(size);
4836 	unsigned long addr;
4837 
4838 	addr = __get_free_pages(gfp_mask, order);
4839 	return make_alloc_exact(addr, order, size);
4840 }
4841 EXPORT_SYMBOL(alloc_pages_exact);
4842 
4843 /**
4844  * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4845  *			   pages on a node.
4846  * @nid: the preferred node ID where memory should be allocated
4847  * @size: the number of bytes to allocate
4848  * @gfp_mask: GFP flags for the allocation
4849  *
4850  * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4851  * back.
4852  *
4853  * Return: pointer to the allocated area or %NULL in case of error.
4854  */
4855 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
4856 {
4857 	unsigned int order = get_order(size);
4858 	struct page *p = alloc_pages_node(nid, gfp_mask, order);
4859 	if (!p)
4860 		return NULL;
4861 	return make_alloc_exact((unsigned long)page_address(p), order, size);
4862 }
4863 
4864 /**
4865  * free_pages_exact - release memory allocated via alloc_pages_exact()
4866  * @virt: the value returned by alloc_pages_exact.
4867  * @size: size of allocation, same value as passed to alloc_pages_exact().
4868  *
4869  * Release the memory allocated by a previous call to alloc_pages_exact.
4870  */
4871 void free_pages_exact(void *virt, size_t size)
4872 {
4873 	unsigned long addr = (unsigned long)virt;
4874 	unsigned long end = addr + PAGE_ALIGN(size);
4875 
4876 	while (addr < end) {
4877 		free_page(addr);
4878 		addr += PAGE_SIZE;
4879 	}
4880 }
4881 EXPORT_SYMBOL(free_pages_exact);
4882 
4883 /**
4884  * nr_free_zone_pages - count number of pages beyond high watermark
4885  * @offset: The zone index of the highest zone
4886  *
4887  * nr_free_zone_pages() counts the number of pages which are beyond the
4888  * high watermark within all zones at or below a given zone index.  For each
4889  * zone, the number of pages is calculated as:
4890  *
4891  *     nr_free_zone_pages = managed_pages - high_pages
4892  *
4893  * Return: number of pages beyond high watermark.
4894  */
4895 static unsigned long nr_free_zone_pages(int offset)
4896 {
4897 	struct zoneref *z;
4898 	struct zone *zone;
4899 
4900 	/* Just pick one node, since fallback list is circular */
4901 	unsigned long sum = 0;
4902 
4903 	struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
4904 
4905 	for_each_zone_zonelist(zone, z, zonelist, offset) {
4906 		unsigned long size = zone_managed_pages(zone);
4907 		unsigned long high = high_wmark_pages(zone);
4908 		if (size > high)
4909 			sum += size - high;
4910 	}
4911 
4912 	return sum;
4913 }
4914 
4915 /**
4916  * nr_free_buffer_pages - count number of pages beyond high watermark
4917  *
4918  * nr_free_buffer_pages() counts the number of pages which are beyond the high
4919  * watermark within ZONE_DMA and ZONE_NORMAL.
4920  *
4921  * Return: number of pages beyond high watermark within ZONE_DMA and
4922  * ZONE_NORMAL.
4923  */
4924 unsigned long nr_free_buffer_pages(void)
4925 {
4926 	return nr_free_zone_pages(gfp_zone(GFP_USER));
4927 }
4928 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
4929 
4930 /**
4931  * nr_free_pagecache_pages - count number of pages beyond high watermark
4932  *
4933  * nr_free_pagecache_pages() counts the number of pages which are beyond the
4934  * high watermark within all zones.
4935  *
4936  * Return: number of pages beyond high watermark within all zones.
4937  */
4938 unsigned long nr_free_pagecache_pages(void)
4939 {
4940 	return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
4941 }
4942 
4943 static inline void show_node(struct zone *zone)
4944 {
4945 	if (IS_ENABLED(CONFIG_NUMA))
4946 		printk("Node %d ", zone_to_nid(zone));
4947 }
4948 
4949 long si_mem_available(void)
4950 {
4951 	long available;
4952 	unsigned long pagecache;
4953 	unsigned long wmark_low = 0;
4954 	unsigned long pages[NR_LRU_LISTS];
4955 	unsigned long reclaimable;
4956 	struct zone *zone;
4957 	int lru;
4958 
4959 	for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
4960 		pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
4961 
4962 	for_each_zone(zone)
4963 		wmark_low += low_wmark_pages(zone);
4964 
4965 	/*
4966 	 * Estimate the amount of memory available for userspace allocations,
4967 	 * without causing swapping.
4968 	 */
4969 	available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
4970 
4971 	/*
4972 	 * Not all the page cache can be freed, otherwise the system will
4973 	 * start swapping. Assume at least half of the page cache, or the
4974 	 * low watermark worth of cache, needs to stay.
4975 	 */
4976 	pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4977 	pagecache -= min(pagecache / 2, wmark_low);
4978 	available += pagecache;
4979 
4980 	/*
4981 	 * Part of the reclaimable slab and other kernel memory consists of
4982 	 * items that are in use, and cannot be freed. Cap this estimate at the
4983 	 * low watermark.
4984 	 */
4985 	reclaimable = global_node_page_state(NR_SLAB_RECLAIMABLE) +
4986 			global_node_page_state(NR_KERNEL_MISC_RECLAIMABLE);
4987 	available += reclaimable - min(reclaimable / 2, wmark_low);
4988 
4989 	if (available < 0)
4990 		available = 0;
4991 	return available;
4992 }
4993 EXPORT_SYMBOL_GPL(si_mem_available);
4994 
4995 void si_meminfo(struct sysinfo *val)
4996 {
4997 	val->totalram = totalram_pages();
4998 	val->sharedram = global_node_page_state(NR_SHMEM);
4999 	val->freeram = global_zone_page_state(NR_FREE_PAGES);
5000 	val->bufferram = nr_blockdev_pages();
5001 	val->totalhigh = totalhigh_pages();
5002 	val->freehigh = nr_free_highpages();
5003 	val->mem_unit = PAGE_SIZE;
5004 }
5005 
5006 EXPORT_SYMBOL(si_meminfo);
5007 
5008 #ifdef CONFIG_NUMA
5009 void si_meminfo_node(struct sysinfo *val, int nid)
5010 {
5011 	int zone_type;		/* needs to be signed */
5012 	unsigned long managed_pages = 0;
5013 	unsigned long managed_highpages = 0;
5014 	unsigned long free_highpages = 0;
5015 	pg_data_t *pgdat = NODE_DATA(nid);
5016 
5017 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
5018 		managed_pages += zone_managed_pages(&pgdat->node_zones[zone_type]);
5019 	val->totalram = managed_pages;
5020 	val->sharedram = node_page_state(pgdat, NR_SHMEM);
5021 	val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
5022 #ifdef CONFIG_HIGHMEM
5023 	for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
5024 		struct zone *zone = &pgdat->node_zones[zone_type];
5025 
5026 		if (is_highmem(zone)) {
5027 			managed_highpages += zone_managed_pages(zone);
5028 			free_highpages += zone_page_state(zone, NR_FREE_PAGES);
5029 		}
5030 	}
5031 	val->totalhigh = managed_highpages;
5032 	val->freehigh = free_highpages;
5033 #else
5034 	val->totalhigh = managed_highpages;
5035 	val->freehigh = free_highpages;
5036 #endif
5037 	val->mem_unit = PAGE_SIZE;
5038 }
5039 #endif
5040 
5041 /*
5042  * Determine whether the node should be displayed or not, depending on whether
5043  * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
5044  */
5045 static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
5046 {
5047 	if (!(flags & SHOW_MEM_FILTER_NODES))
5048 		return false;
5049 
5050 	/*
5051 	 * no node mask - aka implicit memory numa policy. Do not bother with
5052 	 * the synchronization - read_mems_allowed_begin - because we do not
5053 	 * have to be precise here.
5054 	 */
5055 	if (!nodemask)
5056 		nodemask = &cpuset_current_mems_allowed;
5057 
5058 	return !node_isset(nid, *nodemask);
5059 }
5060 
5061 #define K(x) ((x) << (PAGE_SHIFT-10))
5062 
5063 static void show_migration_types(unsigned char type)
5064 {
5065 	static const char types[MIGRATE_TYPES] = {
5066 		[MIGRATE_UNMOVABLE]	= 'U',
5067 		[MIGRATE_MOVABLE]	= 'M',
5068 		[MIGRATE_RECLAIMABLE]	= 'E',
5069 		[MIGRATE_HIGHATOMIC]	= 'H',
5070 #ifdef CONFIG_CMA
5071 		[MIGRATE_CMA]		= 'C',
5072 #endif
5073 #ifdef CONFIG_MEMORY_ISOLATION
5074 		[MIGRATE_ISOLATE]	= 'I',
5075 #endif
5076 	};
5077 	char tmp[MIGRATE_TYPES + 1];
5078 	char *p = tmp;
5079 	int i;
5080 
5081 	for (i = 0; i < MIGRATE_TYPES; i++) {
5082 		if (type & (1 << i))
5083 			*p++ = types[i];
5084 	}
5085 
5086 	*p = '\0';
5087 	printk(KERN_CONT "(%s) ", tmp);
5088 }
5089 
5090 /*
5091  * Show free area list (used inside shift_scroll-lock stuff)
5092  * We also calculate the percentage fragmentation. We do this by counting the
5093  * memory on each free list with the exception of the first item on the list.
5094  *
5095  * Bits in @filter:
5096  * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
5097  *   cpuset.
5098  */
5099 void show_free_areas(unsigned int filter, nodemask_t *nodemask)
5100 {
5101 	unsigned long free_pcp = 0;
5102 	int cpu;
5103 	struct zone *zone;
5104 	pg_data_t *pgdat;
5105 
5106 	for_each_populated_zone(zone) {
5107 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5108 			continue;
5109 
5110 		for_each_online_cpu(cpu)
5111 			free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5112 	}
5113 
5114 	printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
5115 		" active_file:%lu inactive_file:%lu isolated_file:%lu\n"
5116 		" unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
5117 		" slab_reclaimable:%lu slab_unreclaimable:%lu\n"
5118 		" mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
5119 		" free:%lu free_pcp:%lu free_cma:%lu\n",
5120 		global_node_page_state(NR_ACTIVE_ANON),
5121 		global_node_page_state(NR_INACTIVE_ANON),
5122 		global_node_page_state(NR_ISOLATED_ANON),
5123 		global_node_page_state(NR_ACTIVE_FILE),
5124 		global_node_page_state(NR_INACTIVE_FILE),
5125 		global_node_page_state(NR_ISOLATED_FILE),
5126 		global_node_page_state(NR_UNEVICTABLE),
5127 		global_node_page_state(NR_FILE_DIRTY),
5128 		global_node_page_state(NR_WRITEBACK),
5129 		global_node_page_state(NR_UNSTABLE_NFS),
5130 		global_node_page_state(NR_SLAB_RECLAIMABLE),
5131 		global_node_page_state(NR_SLAB_UNRECLAIMABLE),
5132 		global_node_page_state(NR_FILE_MAPPED),
5133 		global_node_page_state(NR_SHMEM),
5134 		global_zone_page_state(NR_PAGETABLE),
5135 		global_zone_page_state(NR_BOUNCE),
5136 		global_zone_page_state(NR_FREE_PAGES),
5137 		free_pcp,
5138 		global_zone_page_state(NR_FREE_CMA_PAGES));
5139 
5140 	for_each_online_pgdat(pgdat) {
5141 		if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
5142 			continue;
5143 
5144 		printk("Node %d"
5145 			" active_anon:%lukB"
5146 			" inactive_anon:%lukB"
5147 			" active_file:%lukB"
5148 			" inactive_file:%lukB"
5149 			" unevictable:%lukB"
5150 			" isolated(anon):%lukB"
5151 			" isolated(file):%lukB"
5152 			" mapped:%lukB"
5153 			" dirty:%lukB"
5154 			" writeback:%lukB"
5155 			" shmem:%lukB"
5156 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5157 			" shmem_thp: %lukB"
5158 			" shmem_pmdmapped: %lukB"
5159 			" anon_thp: %lukB"
5160 #endif
5161 			" writeback_tmp:%lukB"
5162 			" unstable:%lukB"
5163 			" all_unreclaimable? %s"
5164 			"\n",
5165 			pgdat->node_id,
5166 			K(node_page_state(pgdat, NR_ACTIVE_ANON)),
5167 			K(node_page_state(pgdat, NR_INACTIVE_ANON)),
5168 			K(node_page_state(pgdat, NR_ACTIVE_FILE)),
5169 			K(node_page_state(pgdat, NR_INACTIVE_FILE)),
5170 			K(node_page_state(pgdat, NR_UNEVICTABLE)),
5171 			K(node_page_state(pgdat, NR_ISOLATED_ANON)),
5172 			K(node_page_state(pgdat, NR_ISOLATED_FILE)),
5173 			K(node_page_state(pgdat, NR_FILE_MAPPED)),
5174 			K(node_page_state(pgdat, NR_FILE_DIRTY)),
5175 			K(node_page_state(pgdat, NR_WRITEBACK)),
5176 			K(node_page_state(pgdat, NR_SHMEM)),
5177 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5178 			K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
5179 			K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
5180 					* HPAGE_PMD_NR),
5181 			K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
5182 #endif
5183 			K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
5184 			K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
5185 			pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
5186 				"yes" : "no");
5187 	}
5188 
5189 	for_each_populated_zone(zone) {
5190 		int i;
5191 
5192 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5193 			continue;
5194 
5195 		free_pcp = 0;
5196 		for_each_online_cpu(cpu)
5197 			free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
5198 
5199 		show_node(zone);
5200 		printk(KERN_CONT
5201 			"%s"
5202 			" free:%lukB"
5203 			" min:%lukB"
5204 			" low:%lukB"
5205 			" high:%lukB"
5206 			" active_anon:%lukB"
5207 			" inactive_anon:%lukB"
5208 			" active_file:%lukB"
5209 			" inactive_file:%lukB"
5210 			" unevictable:%lukB"
5211 			" writepending:%lukB"
5212 			" present:%lukB"
5213 			" managed:%lukB"
5214 			" mlocked:%lukB"
5215 			" kernel_stack:%lukB"
5216 			" pagetables:%lukB"
5217 			" bounce:%lukB"
5218 			" free_pcp:%lukB"
5219 			" local_pcp:%ukB"
5220 			" free_cma:%lukB"
5221 			"\n",
5222 			zone->name,
5223 			K(zone_page_state(zone, NR_FREE_PAGES)),
5224 			K(min_wmark_pages(zone)),
5225 			K(low_wmark_pages(zone)),
5226 			K(high_wmark_pages(zone)),
5227 			K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
5228 			K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
5229 			K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
5230 			K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
5231 			K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
5232 			K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
5233 			K(zone->present_pages),
5234 			K(zone_managed_pages(zone)),
5235 			K(zone_page_state(zone, NR_MLOCK)),
5236 			zone_page_state(zone, NR_KERNEL_STACK_KB),
5237 			K(zone_page_state(zone, NR_PAGETABLE)),
5238 			K(zone_page_state(zone, NR_BOUNCE)),
5239 			K(free_pcp),
5240 			K(this_cpu_read(zone->pageset->pcp.count)),
5241 			K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
5242 		printk("lowmem_reserve[]:");
5243 		for (i = 0; i < MAX_NR_ZONES; i++)
5244 			printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
5245 		printk(KERN_CONT "\n");
5246 	}
5247 
5248 	for_each_populated_zone(zone) {
5249 		unsigned int order;
5250 		unsigned long nr[MAX_ORDER], flags, total = 0;
5251 		unsigned char types[MAX_ORDER];
5252 
5253 		if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5254 			continue;
5255 		show_node(zone);
5256 		printk(KERN_CONT "%s: ", zone->name);
5257 
5258 		spin_lock_irqsave(&zone->lock, flags);
5259 		for (order = 0; order < MAX_ORDER; order++) {
5260 			struct free_area *area = &zone->free_area[order];
5261 			int type;
5262 
5263 			nr[order] = area->nr_free;
5264 			total += nr[order] << order;
5265 
5266 			types[order] = 0;
5267 			for (type = 0; type < MIGRATE_TYPES; type++) {
5268 				if (!list_empty(&area->free_list[type]))
5269 					types[order] |= 1 << type;
5270 			}
5271 		}
5272 		spin_unlock_irqrestore(&zone->lock, flags);
5273 		for (order = 0; order < MAX_ORDER; order++) {
5274 			printk(KERN_CONT "%lu*%lukB ",
5275 			       nr[order], K(1UL) << order);
5276 			if (nr[order])
5277 				show_migration_types(types[order]);
5278 		}
5279 		printk(KERN_CONT "= %lukB\n", K(total));
5280 	}
5281 
5282 	hugetlb_show_meminfo();
5283 
5284 	printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
5285 
5286 	show_swap_cache_info();
5287 }
5288 
5289 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
5290 {
5291 	zoneref->zone = zone;
5292 	zoneref->zone_idx = zone_idx(zone);
5293 }
5294 
5295 /*
5296  * Builds allocation fallback zone lists.
5297  *
5298  * Add all populated zones of a node to the zonelist.
5299  */
5300 static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
5301 {
5302 	struct zone *zone;
5303 	enum zone_type zone_type = MAX_NR_ZONES;
5304 	int nr_zones = 0;
5305 
5306 	do {
5307 		zone_type--;
5308 		zone = pgdat->node_zones + zone_type;
5309 		if (managed_zone(zone)) {
5310 			zoneref_set_zone(zone, &zonerefs[nr_zones++]);
5311 			check_highest_zone(zone_type);
5312 		}
5313 	} while (zone_type);
5314 
5315 	return nr_zones;
5316 }
5317 
5318 #ifdef CONFIG_NUMA
5319 
5320 static int __parse_numa_zonelist_order(char *s)
5321 {
5322 	/*
5323 	 * We used to support different zonlists modes but they turned
5324 	 * out to be just not useful. Let's keep the warning in place
5325 	 * if somebody still use the cmd line parameter so that we do
5326 	 * not fail it silently
5327 	 */
5328 	if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
5329 		pr_warn("Ignoring unsupported numa_zonelist_order value:  %s\n", s);
5330 		return -EINVAL;
5331 	}
5332 	return 0;
5333 }
5334 
5335 static __init int setup_numa_zonelist_order(char *s)
5336 {
5337 	if (!s)
5338 		return 0;
5339 
5340 	return __parse_numa_zonelist_order(s);
5341 }
5342 early_param("numa_zonelist_order", setup_numa_zonelist_order);
5343 
5344 char numa_zonelist_order[] = "Node";
5345 
5346 /*
5347  * sysctl handler for numa_zonelist_order
5348  */
5349 int numa_zonelist_order_handler(struct ctl_table *table, int write,
5350 		void __user *buffer, size_t *length,
5351 		loff_t *ppos)
5352 {
5353 	char *str;
5354 	int ret;
5355 
5356 	if (!write)
5357 		return proc_dostring(table, write, buffer, length, ppos);
5358 	str = memdup_user_nul(buffer, 16);
5359 	if (IS_ERR(str))
5360 		return PTR_ERR(str);
5361 
5362 	ret = __parse_numa_zonelist_order(str);
5363 	kfree(str);
5364 	return ret;
5365 }
5366 
5367 
5368 #define MAX_NODE_LOAD (nr_online_nodes)
5369 static int node_load[MAX_NUMNODES];
5370 
5371 /**
5372  * find_next_best_node - find the next node that should appear in a given node's fallback list
5373  * @node: node whose fallback list we're appending
5374  * @used_node_mask: nodemask_t of already used nodes
5375  *
5376  * We use a number of factors to determine which is the next node that should
5377  * appear on a given node's fallback list.  The node should not have appeared
5378  * already in @node's fallback list, and it should be the next closest node
5379  * according to the distance array (which contains arbitrary distance values
5380  * from each node to each node in the system), and should also prefer nodes
5381  * with no CPUs, since presumably they'll have very little allocation pressure
5382  * on them otherwise.
5383  *
5384  * Return: node id of the found node or %NUMA_NO_NODE if no node is found.
5385  */
5386 static int find_next_best_node(int node, nodemask_t *used_node_mask)
5387 {
5388 	int n, val;
5389 	int min_val = INT_MAX;
5390 	int best_node = NUMA_NO_NODE;
5391 	const struct cpumask *tmp = cpumask_of_node(0);
5392 
5393 	/* Use the local node if we haven't already */
5394 	if (!node_isset(node, *used_node_mask)) {
5395 		node_set(node, *used_node_mask);
5396 		return node;
5397 	}
5398 
5399 	for_each_node_state(n, N_MEMORY) {
5400 
5401 		/* Don't want a node to appear more than once */
5402 		if (node_isset(n, *used_node_mask))
5403 			continue;
5404 
5405 		/* Use the distance array to find the distance */
5406 		val = node_distance(node, n);
5407 
5408 		/* Penalize nodes under us ("prefer the next node") */
5409 		val += (n < node);
5410 
5411 		/* Give preference to headless and unused nodes */
5412 		tmp = cpumask_of_node(n);
5413 		if (!cpumask_empty(tmp))
5414 			val += PENALTY_FOR_NODE_WITH_CPUS;
5415 
5416 		/* Slight preference for less loaded node */
5417 		val *= (MAX_NODE_LOAD*MAX_NUMNODES);
5418 		val += node_load[n];
5419 
5420 		if (val < min_val) {
5421 			min_val = val;
5422 			best_node = n;
5423 		}
5424 	}
5425 
5426 	if (best_node >= 0)
5427 		node_set(best_node, *used_node_mask);
5428 
5429 	return best_node;
5430 }
5431 
5432 
5433 /*
5434  * Build zonelists ordered by node and zones within node.
5435  * This results in maximum locality--normal zone overflows into local
5436  * DMA zone, if any--but risks exhausting DMA zone.
5437  */
5438 static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
5439 		unsigned nr_nodes)
5440 {
5441 	struct zoneref *zonerefs;
5442 	int i;
5443 
5444 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5445 
5446 	for (i = 0; i < nr_nodes; i++) {
5447 		int nr_zones;
5448 
5449 		pg_data_t *node = NODE_DATA(node_order[i]);
5450 
5451 		nr_zones = build_zonerefs_node(node, zonerefs);
5452 		zonerefs += nr_zones;
5453 	}
5454 	zonerefs->zone = NULL;
5455 	zonerefs->zone_idx = 0;
5456 }
5457 
5458 /*
5459  * Build gfp_thisnode zonelists
5460  */
5461 static void build_thisnode_zonelists(pg_data_t *pgdat)
5462 {
5463 	struct zoneref *zonerefs;
5464 	int nr_zones;
5465 
5466 	zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
5467 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
5468 	zonerefs += nr_zones;
5469 	zonerefs->zone = NULL;
5470 	zonerefs->zone_idx = 0;
5471 }
5472 
5473 /*
5474  * Build zonelists ordered by zone and nodes within zones.
5475  * This results in conserving DMA zone[s] until all Normal memory is
5476  * exhausted, but results in overflowing to remote node while memory
5477  * may still exist in local DMA zone.
5478  */
5479 
5480 static void build_zonelists(pg_data_t *pgdat)
5481 {
5482 	static int node_order[MAX_NUMNODES];
5483 	int node, load, nr_nodes = 0;
5484 	nodemask_t used_mask;
5485 	int local_node, prev_node;
5486 
5487 	/* NUMA-aware ordering of nodes */
5488 	local_node = pgdat->node_id;
5489 	load = nr_online_nodes;
5490 	prev_node = local_node;
5491 	nodes_clear(used_mask);
5492 
5493 	memset(node_order, 0, sizeof(node_order));
5494 	while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5495 		/*
5496 		 * We don't want to pressure a particular node.
5497 		 * So adding penalty to the first node in same
5498 		 * distance group to make it round-robin.
5499 		 */
5500 		if (node_distance(local_node, node) !=
5501 		    node_distance(local_node, prev_node))
5502 			node_load[node] = load;
5503 
5504 		node_order[nr_nodes++] = node;
5505 		prev_node = node;
5506 		load--;
5507 	}
5508 
5509 	build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
5510 	build_thisnode_zonelists(pgdat);
5511 }
5512 
5513 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5514 /*
5515  * Return node id of node used for "local" allocations.
5516  * I.e., first node id of first zone in arg node's generic zonelist.
5517  * Used for initializing percpu 'numa_mem', which is used primarily
5518  * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
5519  */
5520 int local_memory_node(int node)
5521 {
5522 	struct zoneref *z;
5523 
5524 	z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
5525 				   gfp_zone(GFP_KERNEL),
5526 				   NULL);
5527 	return zone_to_nid(z->zone);
5528 }
5529 #endif
5530 
5531 static void setup_min_unmapped_ratio(void);
5532 static void setup_min_slab_ratio(void);
5533 #else	/* CONFIG_NUMA */
5534 
5535 static void build_zonelists(pg_data_t *pgdat)
5536 {
5537 	int node, local_node;
5538 	struct zoneref *zonerefs;
5539 	int nr_zones;
5540 
5541 	local_node = pgdat->node_id;
5542 
5543 	zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5544 	nr_zones = build_zonerefs_node(pgdat, zonerefs);
5545 	zonerefs += nr_zones;
5546 
5547 	/*
5548 	 * Now we build the zonelist so that it contains the zones
5549 	 * of all the other nodes.
5550 	 * We don't want to pressure a particular node, so when
5551 	 * building the zones for node N, we make sure that the
5552 	 * zones coming right after the local ones are those from
5553 	 * node N+1 (modulo N)
5554 	 */
5555 	for (node = local_node + 1; node < MAX_NUMNODES; node++) {
5556 		if (!node_online(node))
5557 			continue;
5558 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5559 		zonerefs += nr_zones;
5560 	}
5561 	for (node = 0; node < local_node; node++) {
5562 		if (!node_online(node))
5563 			continue;
5564 		nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5565 		zonerefs += nr_zones;
5566 	}
5567 
5568 	zonerefs->zone = NULL;
5569 	zonerefs->zone_idx = 0;
5570 }
5571 
5572 #endif	/* CONFIG_NUMA */
5573 
5574 /*
5575  * Boot pageset table. One per cpu which is going to be used for all
5576  * zones and all nodes. The parameters will be set in such a way
5577  * that an item put on a list will immediately be handed over to
5578  * the buddy list. This is safe since pageset manipulation is done
5579  * with interrupts disabled.
5580  *
5581  * The boot_pagesets must be kept even after bootup is complete for
5582  * unused processors and/or zones. They do play a role for bootstrapping
5583  * hotplugged processors.
5584  *
5585  * zoneinfo_show() and maybe other functions do
5586  * not check if the processor is online before following the pageset pointer.
5587  * Other parts of the kernel may not check if the zone is available.
5588  */
5589 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
5590 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
5591 static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
5592 
5593 static void __build_all_zonelists(void *data)
5594 {
5595 	int nid;
5596 	int __maybe_unused cpu;
5597 	pg_data_t *self = data;
5598 	static DEFINE_SPINLOCK(lock);
5599 
5600 	spin_lock(&lock);
5601 
5602 #ifdef CONFIG_NUMA
5603 	memset(node_load, 0, sizeof(node_load));
5604 #endif
5605 
5606 	/*
5607 	 * This node is hotadded and no memory is yet present.   So just
5608 	 * building zonelists is fine - no need to touch other nodes.
5609 	 */
5610 	if (self && !node_online(self->node_id)) {
5611 		build_zonelists(self);
5612 	} else {
5613 		for_each_online_node(nid) {
5614 			pg_data_t *pgdat = NODE_DATA(nid);
5615 
5616 			build_zonelists(pgdat);
5617 		}
5618 
5619 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5620 		/*
5621 		 * We now know the "local memory node" for each node--
5622 		 * i.e., the node of the first zone in the generic zonelist.
5623 		 * Set up numa_mem percpu variable for on-line cpus.  During
5624 		 * boot, only the boot cpu should be on-line;  we'll init the
5625 		 * secondary cpus' numa_mem as they come on-line.  During
5626 		 * node/memory hotplug, we'll fixup all on-line cpus.
5627 		 */
5628 		for_each_online_cpu(cpu)
5629 			set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
5630 #endif
5631 	}
5632 
5633 	spin_unlock(&lock);
5634 }
5635 
5636 static noinline void __init
5637 build_all_zonelists_init(void)
5638 {
5639 	int cpu;
5640 
5641 	__build_all_zonelists(NULL);
5642 
5643 	/*
5644 	 * Initialize the boot_pagesets that are going to be used
5645 	 * for bootstrapping processors. The real pagesets for
5646 	 * each zone will be allocated later when the per cpu
5647 	 * allocator is available.
5648 	 *
5649 	 * boot_pagesets are used also for bootstrapping offline
5650 	 * cpus if the system is already booted because the pagesets
5651 	 * are needed to initialize allocators on a specific cpu too.
5652 	 * F.e. the percpu allocator needs the page allocator which
5653 	 * needs the percpu allocator in order to allocate its pagesets
5654 	 * (a chicken-egg dilemma).
5655 	 */
5656 	for_each_possible_cpu(cpu)
5657 		setup_pageset(&per_cpu(boot_pageset, cpu), 0);
5658 
5659 	mminit_verify_zonelist();
5660 	cpuset_init_current_mems_allowed();
5661 }
5662 
5663 /*
5664  * unless system_state == SYSTEM_BOOTING.
5665  *
5666  * __ref due to call of __init annotated helper build_all_zonelists_init
5667  * [protected by SYSTEM_BOOTING].
5668  */
5669 void __ref build_all_zonelists(pg_data_t *pgdat)
5670 {
5671 	if (system_state == SYSTEM_BOOTING) {
5672 		build_all_zonelists_init();
5673 	} else {
5674 		__build_all_zonelists(pgdat);
5675 		/* cpuset refresh routine should be here */
5676 	}
5677 	vm_total_pages = nr_free_pagecache_pages();
5678 	/*
5679 	 * Disable grouping by mobility if the number of pages in the
5680 	 * system is too low to allow the mechanism to work. It would be
5681 	 * more accurate, but expensive to check per-zone. This check is
5682 	 * made on memory-hotadd so a system can start with mobility
5683 	 * disabled and enable it later
5684 	 */
5685 	if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
5686 		page_group_by_mobility_disabled = 1;
5687 	else
5688 		page_group_by_mobility_disabled = 0;
5689 
5690 	pr_info("Built %u zonelists, mobility grouping %s.  Total pages: %ld\n",
5691 		nr_online_nodes,
5692 		page_group_by_mobility_disabled ? "off" : "on",
5693 		vm_total_pages);
5694 #ifdef CONFIG_NUMA
5695 	pr_info("Policy zone: %s\n", zone_names[policy_zone]);
5696 #endif
5697 }
5698 
5699 /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */
5700 static bool __meminit
5701 overlap_memmap_init(unsigned long zone, unsigned long *pfn)
5702 {
5703 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5704 	static struct memblock_region *r;
5705 
5706 	if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5707 		if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
5708 			for_each_memblock(memory, r) {
5709 				if (*pfn < memblock_region_memory_end_pfn(r))
5710 					break;
5711 			}
5712 		}
5713 		if (*pfn >= memblock_region_memory_base_pfn(r) &&
5714 		    memblock_is_mirror(r)) {
5715 			*pfn = memblock_region_memory_end_pfn(r);
5716 			return true;
5717 		}
5718 	}
5719 #endif
5720 	return false;
5721 }
5722 
5723 /*
5724  * Initially all pages are reserved - free ones are freed
5725  * up by memblock_free_all() once the early boot process is
5726  * done. Non-atomic initialization, single-pass.
5727  */
5728 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
5729 		unsigned long start_pfn, enum memmap_context context,
5730 		struct vmem_altmap *altmap)
5731 {
5732 	unsigned long pfn, end_pfn = start_pfn + size;
5733 	struct page *page;
5734 
5735 	if (highest_memmap_pfn < end_pfn - 1)
5736 		highest_memmap_pfn = end_pfn - 1;
5737 
5738 #ifdef CONFIG_ZONE_DEVICE
5739 	/*
5740 	 * Honor reservation requested by the driver for this ZONE_DEVICE
5741 	 * memory. We limit the total number of pages to initialize to just
5742 	 * those that might contain the memory mapping. We will defer the
5743 	 * ZONE_DEVICE page initialization until after we have released
5744 	 * the hotplug lock.
5745 	 */
5746 	if (zone == ZONE_DEVICE) {
5747 		if (!altmap)
5748 			return;
5749 
5750 		if (start_pfn == altmap->base_pfn)
5751 			start_pfn += altmap->reserve;
5752 		end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
5753 	}
5754 #endif
5755 
5756 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5757 		/*
5758 		 * There can be holes in boot-time mem_map[]s handed to this
5759 		 * function.  They do not exist on hotplugged memory.
5760 		 */
5761 		if (context == MEMMAP_EARLY) {
5762 			if (!early_pfn_valid(pfn))
5763 				continue;
5764 			if (!early_pfn_in_nid(pfn, nid))
5765 				continue;
5766 			if (overlap_memmap_init(zone, &pfn))
5767 				continue;
5768 			if (defer_init(nid, pfn, end_pfn))
5769 				break;
5770 		}
5771 
5772 		page = pfn_to_page(pfn);
5773 		__init_single_page(page, pfn, zone, nid);
5774 		if (context == MEMMAP_HOTPLUG)
5775 			__SetPageReserved(page);
5776 
5777 		/*
5778 		 * Mark the block movable so that blocks are reserved for
5779 		 * movable at startup. This will force kernel allocations
5780 		 * to reserve their blocks rather than leaking throughout
5781 		 * the address space during boot when many long-lived
5782 		 * kernel allocations are made.
5783 		 *
5784 		 * bitmap is created for zone's valid pfn range. but memmap
5785 		 * can be created for invalid pages (for alignment)
5786 		 * check here not to call set_pageblock_migratetype() against
5787 		 * pfn out of zone.
5788 		 */
5789 		if (!(pfn & (pageblock_nr_pages - 1))) {
5790 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5791 			cond_resched();
5792 		}
5793 	}
5794 }
5795 
5796 #ifdef CONFIG_ZONE_DEVICE
5797 void __ref memmap_init_zone_device(struct zone *zone,
5798 				   unsigned long start_pfn,
5799 				   unsigned long size,
5800 				   struct dev_pagemap *pgmap)
5801 {
5802 	unsigned long pfn, end_pfn = start_pfn + size;
5803 	struct pglist_data *pgdat = zone->zone_pgdat;
5804 	unsigned long zone_idx = zone_idx(zone);
5805 	unsigned long start = jiffies;
5806 	int nid = pgdat->node_id;
5807 
5808 	if (WARN_ON_ONCE(!pgmap || !is_dev_zone(zone)))
5809 		return;
5810 
5811 	/*
5812 	 * The call to memmap_init_zone should have already taken care
5813 	 * of the pages reserved for the memmap, so we can just jump to
5814 	 * the end of that region and start processing the device pages.
5815 	 */
5816 	if (pgmap->altmap_valid) {
5817 		struct vmem_altmap *altmap = &pgmap->altmap;
5818 
5819 		start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
5820 		size = end_pfn - start_pfn;
5821 	}
5822 
5823 	for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5824 		struct page *page = pfn_to_page(pfn);
5825 
5826 		__init_single_page(page, pfn, zone_idx, nid);
5827 
5828 		/*
5829 		 * Mark page reserved as it will need to wait for onlining
5830 		 * phase for it to be fully associated with a zone.
5831 		 *
5832 		 * We can use the non-atomic __set_bit operation for setting
5833 		 * the flag as we are still initializing the pages.
5834 		 */
5835 		__SetPageReserved(page);
5836 
5837 		/*
5838 		 * ZONE_DEVICE pages union ->lru with a ->pgmap back
5839 		 * pointer and hmm_data.  It is a bug if a ZONE_DEVICE
5840 		 * page is ever freed or placed on a driver-private list.
5841 		 */
5842 		page->pgmap = pgmap;
5843 		page->hmm_data = 0;
5844 
5845 		/*
5846 		 * Mark the block movable so that blocks are reserved for
5847 		 * movable at startup. This will force kernel allocations
5848 		 * to reserve their blocks rather than leaking throughout
5849 		 * the address space during boot when many long-lived
5850 		 * kernel allocations are made.
5851 		 *
5852 		 * bitmap is created for zone's valid pfn range. but memmap
5853 		 * can be created for invalid pages (for alignment)
5854 		 * check here not to call set_pageblock_migratetype() against
5855 		 * pfn out of zone.
5856 		 *
5857 		 * Please note that MEMMAP_HOTPLUG path doesn't clear memmap
5858 		 * because this is done early in sparse_add_one_section
5859 		 */
5860 		if (!(pfn & (pageblock_nr_pages - 1))) {
5861 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5862 			cond_resched();
5863 		}
5864 	}
5865 
5866 	pr_info("%s initialised, %lu pages in %ums\n", dev_name(pgmap->dev),
5867 		size, jiffies_to_msecs(jiffies - start));
5868 }
5869 
5870 #endif
5871 static void __meminit zone_init_free_lists(struct zone *zone)
5872 {
5873 	unsigned int order, t;
5874 	for_each_migratetype_order(order, t) {
5875 		INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
5876 		zone->free_area[order].nr_free = 0;
5877 	}
5878 }
5879 
5880 void __meminit __weak memmap_init(unsigned long size, int nid,
5881 				  unsigned long zone, unsigned long start_pfn)
5882 {
5883 	memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY, NULL);
5884 }
5885 
5886 static int zone_batchsize(struct zone *zone)
5887 {
5888 #ifdef CONFIG_MMU
5889 	int batch;
5890 
5891 	/*
5892 	 * The per-cpu-pages pools are set to around 1000th of the
5893 	 * size of the zone.
5894 	 */
5895 	batch = zone_managed_pages(zone) / 1024;
5896 	/* But no more than a meg. */
5897 	if (batch * PAGE_SIZE > 1024 * 1024)
5898 		batch = (1024 * 1024) / PAGE_SIZE;
5899 	batch /= 4;		/* We effectively *= 4 below */
5900 	if (batch < 1)
5901 		batch = 1;
5902 
5903 	/*
5904 	 * Clamp the batch to a 2^n - 1 value. Having a power
5905 	 * of 2 value was found to be more likely to have
5906 	 * suboptimal cache aliasing properties in some cases.
5907 	 *
5908 	 * For example if 2 tasks are alternately allocating
5909 	 * batches of pages, one task can end up with a lot
5910 	 * of pages of one half of the possible page colors
5911 	 * and the other with pages of the other colors.
5912 	 */
5913 	batch = rounddown_pow_of_two(batch + batch/2) - 1;
5914 
5915 	return batch;
5916 
5917 #else
5918 	/* The deferral and batching of frees should be suppressed under NOMMU
5919 	 * conditions.
5920 	 *
5921 	 * The problem is that NOMMU needs to be able to allocate large chunks
5922 	 * of contiguous memory as there's no hardware page translation to
5923 	 * assemble apparent contiguous memory from discontiguous pages.
5924 	 *
5925 	 * Queueing large contiguous runs of pages for batching, however,
5926 	 * causes the pages to actually be freed in smaller chunks.  As there
5927 	 * can be a significant delay between the individual batches being
5928 	 * recycled, this leads to the once large chunks of space being
5929 	 * fragmented and becoming unavailable for high-order allocations.
5930 	 */
5931 	return 0;
5932 #endif
5933 }
5934 
5935 /*
5936  * pcp->high and pcp->batch values are related and dependent on one another:
5937  * ->batch must never be higher then ->high.
5938  * The following function updates them in a safe manner without read side
5939  * locking.
5940  *
5941  * Any new users of pcp->batch and pcp->high should ensure they can cope with
5942  * those fields changing asynchronously (acording the the above rule).
5943  *
5944  * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5945  * outside of boot time (or some other assurance that no concurrent updaters
5946  * exist).
5947  */
5948 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5949 		unsigned long batch)
5950 {
5951        /* start with a fail safe value for batch */
5952 	pcp->batch = 1;
5953 	smp_wmb();
5954 
5955        /* Update high, then batch, in order */
5956 	pcp->high = high;
5957 	smp_wmb();
5958 
5959 	pcp->batch = batch;
5960 }
5961 
5962 /* a companion to pageset_set_high() */
5963 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5964 {
5965 	pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
5966 }
5967 
5968 static void pageset_init(struct per_cpu_pageset *p)
5969 {
5970 	struct per_cpu_pages *pcp;
5971 	int migratetype;
5972 
5973 	memset(p, 0, sizeof(*p));
5974 
5975 	pcp = &p->pcp;
5976 	for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5977 		INIT_LIST_HEAD(&pcp->lists[migratetype]);
5978 }
5979 
5980 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5981 {
5982 	pageset_init(p);
5983 	pageset_set_batch(p, batch);
5984 }
5985 
5986 /*
5987  * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5988  * to the value high for the pageset p.
5989  */
5990 static void pageset_set_high(struct per_cpu_pageset *p,
5991 				unsigned long high)
5992 {
5993 	unsigned long batch = max(1UL, high / 4);
5994 	if ((high / 4) > (PAGE_SHIFT * 8))
5995 		batch = PAGE_SHIFT * 8;
5996 
5997 	pageset_update(&p->pcp, high, batch);
5998 }
5999 
6000 static void pageset_set_high_and_batch(struct zone *zone,
6001 				       struct per_cpu_pageset *pcp)
6002 {
6003 	if (percpu_pagelist_fraction)
6004 		pageset_set_high(pcp,
6005 			(zone_managed_pages(zone) /
6006 				percpu_pagelist_fraction));
6007 	else
6008 		pageset_set_batch(pcp, zone_batchsize(zone));
6009 }
6010 
6011 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
6012 {
6013 	struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
6014 
6015 	pageset_init(pcp);
6016 	pageset_set_high_and_batch(zone, pcp);
6017 }
6018 
6019 void __meminit setup_zone_pageset(struct zone *zone)
6020 {
6021 	int cpu;
6022 	zone->pageset = alloc_percpu(struct per_cpu_pageset);
6023 	for_each_possible_cpu(cpu)
6024 		zone_pageset_init(zone, cpu);
6025 }
6026 
6027 /*
6028  * Allocate per cpu pagesets and initialize them.
6029  * Before this call only boot pagesets were available.
6030  */
6031 void __init setup_per_cpu_pageset(void)
6032 {
6033 	struct pglist_data *pgdat;
6034 	struct zone *zone;
6035 
6036 	for_each_populated_zone(zone)
6037 		setup_zone_pageset(zone);
6038 
6039 	for_each_online_pgdat(pgdat)
6040 		pgdat->per_cpu_nodestats =
6041 			alloc_percpu(struct per_cpu_nodestat);
6042 }
6043 
6044 static __meminit void zone_pcp_init(struct zone *zone)
6045 {
6046 	/*
6047 	 * per cpu subsystem is not up at this point. The following code
6048 	 * relies on the ability of the linker to provide the
6049 	 * offset of a (static) per cpu variable into the per cpu area.
6050 	 */
6051 	zone->pageset = &boot_pageset;
6052 
6053 	if (populated_zone(zone))
6054 		printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
6055 			zone->name, zone->present_pages,
6056 					 zone_batchsize(zone));
6057 }
6058 
6059 void __meminit init_currently_empty_zone(struct zone *zone,
6060 					unsigned long zone_start_pfn,
6061 					unsigned long size)
6062 {
6063 	struct pglist_data *pgdat = zone->zone_pgdat;
6064 	int zone_idx = zone_idx(zone) + 1;
6065 
6066 	if (zone_idx > pgdat->nr_zones)
6067 		pgdat->nr_zones = zone_idx;
6068 
6069 	zone->zone_start_pfn = zone_start_pfn;
6070 
6071 	mminit_dprintk(MMINIT_TRACE, "memmap_init",
6072 			"Initialising map node %d zone %lu pfns %lu -> %lu\n",
6073 			pgdat->node_id,
6074 			(unsigned long)zone_idx(zone),
6075 			zone_start_pfn, (zone_start_pfn + size));
6076 
6077 	zone_init_free_lists(zone);
6078 	zone->initialized = 1;
6079 }
6080 
6081 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6082 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
6083 
6084 /*
6085  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
6086  */
6087 int __meminit __early_pfn_to_nid(unsigned long pfn,
6088 					struct mminit_pfnnid_cache *state)
6089 {
6090 	unsigned long start_pfn, end_pfn;
6091 	int nid;
6092 
6093 	if (state->last_start <= pfn && pfn < state->last_end)
6094 		return state->last_nid;
6095 
6096 	nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
6097 	if (nid != NUMA_NO_NODE) {
6098 		state->last_start = start_pfn;
6099 		state->last_end = end_pfn;
6100 		state->last_nid = nid;
6101 	}
6102 
6103 	return nid;
6104 }
6105 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
6106 
6107 /**
6108  * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
6109  * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
6110  * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
6111  *
6112  * If an architecture guarantees that all ranges registered contain no holes
6113  * and may be freed, this this function may be used instead of calling
6114  * memblock_free_early_nid() manually.
6115  */
6116 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
6117 {
6118 	unsigned long start_pfn, end_pfn;
6119 	int i, this_nid;
6120 
6121 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
6122 		start_pfn = min(start_pfn, max_low_pfn);
6123 		end_pfn = min(end_pfn, max_low_pfn);
6124 
6125 		if (start_pfn < end_pfn)
6126 			memblock_free_early_nid(PFN_PHYS(start_pfn),
6127 					(end_pfn - start_pfn) << PAGE_SHIFT,
6128 					this_nid);
6129 	}
6130 }
6131 
6132 /**
6133  * sparse_memory_present_with_active_regions - Call memory_present for each active range
6134  * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
6135  *
6136  * If an architecture guarantees that all ranges registered contain no holes and may
6137  * be freed, this function may be used instead of calling memory_present() manually.
6138  */
6139 void __init sparse_memory_present_with_active_regions(int nid)
6140 {
6141 	unsigned long start_pfn, end_pfn;
6142 	int i, this_nid;
6143 
6144 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
6145 		memory_present(this_nid, start_pfn, end_pfn);
6146 }
6147 
6148 /**
6149  * get_pfn_range_for_nid - Return the start and end page frames for a node
6150  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
6151  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
6152  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
6153  *
6154  * It returns the start and end page frame of a node based on information
6155  * provided by memblock_set_node(). If called for a node
6156  * with no available memory, a warning is printed and the start and end
6157  * PFNs will be 0.
6158  */
6159 void __init get_pfn_range_for_nid(unsigned int nid,
6160 			unsigned long *start_pfn, unsigned long *end_pfn)
6161 {
6162 	unsigned long this_start_pfn, this_end_pfn;
6163 	int i;
6164 
6165 	*start_pfn = -1UL;
6166 	*end_pfn = 0;
6167 
6168 	for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
6169 		*start_pfn = min(*start_pfn, this_start_pfn);
6170 		*end_pfn = max(*end_pfn, this_end_pfn);
6171 	}
6172 
6173 	if (*start_pfn == -1UL)
6174 		*start_pfn = 0;
6175 }
6176 
6177 /*
6178  * This finds a zone that can be used for ZONE_MOVABLE pages. The
6179  * assumption is made that zones within a node are ordered in monotonic
6180  * increasing memory addresses so that the "highest" populated zone is used
6181  */
6182 static void __init find_usable_zone_for_movable(void)
6183 {
6184 	int zone_index;
6185 	for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
6186 		if (zone_index == ZONE_MOVABLE)
6187 			continue;
6188 
6189 		if (arch_zone_highest_possible_pfn[zone_index] >
6190 				arch_zone_lowest_possible_pfn[zone_index])
6191 			break;
6192 	}
6193 
6194 	VM_BUG_ON(zone_index == -1);
6195 	movable_zone = zone_index;
6196 }
6197 
6198 /*
6199  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
6200  * because it is sized independent of architecture. Unlike the other zones,
6201  * the starting point for ZONE_MOVABLE is not fixed. It may be different
6202  * in each node depending on the size of each node and how evenly kernelcore
6203  * is distributed. This helper function adjusts the zone ranges
6204  * provided by the architecture for a given node by using the end of the
6205  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
6206  * zones within a node are in order of monotonic increases memory addresses
6207  */
6208 static void __init adjust_zone_range_for_zone_movable(int nid,
6209 					unsigned long zone_type,
6210 					unsigned long node_start_pfn,
6211 					unsigned long node_end_pfn,
6212 					unsigned long *zone_start_pfn,
6213 					unsigned long *zone_end_pfn)
6214 {
6215 	/* Only adjust if ZONE_MOVABLE is on this node */
6216 	if (zone_movable_pfn[nid]) {
6217 		/* Size ZONE_MOVABLE */
6218 		if (zone_type == ZONE_MOVABLE) {
6219 			*zone_start_pfn = zone_movable_pfn[nid];
6220 			*zone_end_pfn = min(node_end_pfn,
6221 				arch_zone_highest_possible_pfn[movable_zone]);
6222 
6223 		/* Adjust for ZONE_MOVABLE starting within this range */
6224 		} else if (!mirrored_kernelcore &&
6225 			*zone_start_pfn < zone_movable_pfn[nid] &&
6226 			*zone_end_pfn > zone_movable_pfn[nid]) {
6227 			*zone_end_pfn = zone_movable_pfn[nid];
6228 
6229 		/* Check if this whole range is within ZONE_MOVABLE */
6230 		} else if (*zone_start_pfn >= zone_movable_pfn[nid])
6231 			*zone_start_pfn = *zone_end_pfn;
6232 	}
6233 }
6234 
6235 /*
6236  * Return the number of pages a zone spans in a node, including holes
6237  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
6238  */
6239 static unsigned long __init zone_spanned_pages_in_node(int nid,
6240 					unsigned long zone_type,
6241 					unsigned long node_start_pfn,
6242 					unsigned long node_end_pfn,
6243 					unsigned long *zone_start_pfn,
6244 					unsigned long *zone_end_pfn,
6245 					unsigned long *ignored)
6246 {
6247 	/* When hotadd a new node from cpu_up(), the node should be empty */
6248 	if (!node_start_pfn && !node_end_pfn)
6249 		return 0;
6250 
6251 	/* Get the start and end of the zone */
6252 	*zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
6253 	*zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
6254 	adjust_zone_range_for_zone_movable(nid, zone_type,
6255 				node_start_pfn, node_end_pfn,
6256 				zone_start_pfn, zone_end_pfn);
6257 
6258 	/* Check that this node has pages within the zone's required range */
6259 	if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
6260 		return 0;
6261 
6262 	/* Move the zone boundaries inside the node if necessary */
6263 	*zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
6264 	*zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
6265 
6266 	/* Return the spanned pages */
6267 	return *zone_end_pfn - *zone_start_pfn;
6268 }
6269 
6270 /*
6271  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
6272  * then all holes in the requested range will be accounted for.
6273  */
6274 unsigned long __init __absent_pages_in_range(int nid,
6275 				unsigned long range_start_pfn,
6276 				unsigned long range_end_pfn)
6277 {
6278 	unsigned long nr_absent = range_end_pfn - range_start_pfn;
6279 	unsigned long start_pfn, end_pfn;
6280 	int i;
6281 
6282 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6283 		start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
6284 		end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
6285 		nr_absent -= end_pfn - start_pfn;
6286 	}
6287 	return nr_absent;
6288 }
6289 
6290 /**
6291  * absent_pages_in_range - Return number of page frames in holes within a range
6292  * @start_pfn: The start PFN to start searching for holes
6293  * @end_pfn: The end PFN to stop searching for holes
6294  *
6295  * Return: the number of pages frames in memory holes within a range.
6296  */
6297 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
6298 							unsigned long end_pfn)
6299 {
6300 	return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
6301 }
6302 
6303 /* Return the number of page frames in holes in a zone on a node */
6304 static unsigned long __init zone_absent_pages_in_node(int nid,
6305 					unsigned long zone_type,
6306 					unsigned long node_start_pfn,
6307 					unsigned long node_end_pfn,
6308 					unsigned long *ignored)
6309 {
6310 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
6311 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
6312 	unsigned long zone_start_pfn, zone_end_pfn;
6313 	unsigned long nr_absent;
6314 
6315 	/* When hotadd a new node from cpu_up(), the node should be empty */
6316 	if (!node_start_pfn && !node_end_pfn)
6317 		return 0;
6318 
6319 	zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
6320 	zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
6321 
6322 	adjust_zone_range_for_zone_movable(nid, zone_type,
6323 			node_start_pfn, node_end_pfn,
6324 			&zone_start_pfn, &zone_end_pfn);
6325 	nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
6326 
6327 	/*
6328 	 * ZONE_MOVABLE handling.
6329 	 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
6330 	 * and vice versa.
6331 	 */
6332 	if (mirrored_kernelcore && zone_movable_pfn[nid]) {
6333 		unsigned long start_pfn, end_pfn;
6334 		struct memblock_region *r;
6335 
6336 		for_each_memblock(memory, r) {
6337 			start_pfn = clamp(memblock_region_memory_base_pfn(r),
6338 					  zone_start_pfn, zone_end_pfn);
6339 			end_pfn = clamp(memblock_region_memory_end_pfn(r),
6340 					zone_start_pfn, zone_end_pfn);
6341 
6342 			if (zone_type == ZONE_MOVABLE &&
6343 			    memblock_is_mirror(r))
6344 				nr_absent += end_pfn - start_pfn;
6345 
6346 			if (zone_type == ZONE_NORMAL &&
6347 			    !memblock_is_mirror(r))
6348 				nr_absent += end_pfn - start_pfn;
6349 		}
6350 	}
6351 
6352 	return nr_absent;
6353 }
6354 
6355 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6356 static inline unsigned long __init zone_spanned_pages_in_node(int nid,
6357 					unsigned long zone_type,
6358 					unsigned long node_start_pfn,
6359 					unsigned long node_end_pfn,
6360 					unsigned long *zone_start_pfn,
6361 					unsigned long *zone_end_pfn,
6362 					unsigned long *zones_size)
6363 {
6364 	unsigned int zone;
6365 
6366 	*zone_start_pfn = node_start_pfn;
6367 	for (zone = 0; zone < zone_type; zone++)
6368 		*zone_start_pfn += zones_size[zone];
6369 
6370 	*zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
6371 
6372 	return zones_size[zone_type];
6373 }
6374 
6375 static inline unsigned long __init zone_absent_pages_in_node(int nid,
6376 						unsigned long zone_type,
6377 						unsigned long node_start_pfn,
6378 						unsigned long node_end_pfn,
6379 						unsigned long *zholes_size)
6380 {
6381 	if (!zholes_size)
6382 		return 0;
6383 
6384 	return zholes_size[zone_type];
6385 }
6386 
6387 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6388 
6389 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
6390 						unsigned long node_start_pfn,
6391 						unsigned long node_end_pfn,
6392 						unsigned long *zones_size,
6393 						unsigned long *zholes_size)
6394 {
6395 	unsigned long realtotalpages = 0, totalpages = 0;
6396 	enum zone_type i;
6397 
6398 	for (i = 0; i < MAX_NR_ZONES; i++) {
6399 		struct zone *zone = pgdat->node_zones + i;
6400 		unsigned long zone_start_pfn, zone_end_pfn;
6401 		unsigned long size, real_size;
6402 
6403 		size = zone_spanned_pages_in_node(pgdat->node_id, i,
6404 						  node_start_pfn,
6405 						  node_end_pfn,
6406 						  &zone_start_pfn,
6407 						  &zone_end_pfn,
6408 						  zones_size);
6409 		real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
6410 						  node_start_pfn, node_end_pfn,
6411 						  zholes_size);
6412 		if (size)
6413 			zone->zone_start_pfn = zone_start_pfn;
6414 		else
6415 			zone->zone_start_pfn = 0;
6416 		zone->spanned_pages = size;
6417 		zone->present_pages = real_size;
6418 
6419 		totalpages += size;
6420 		realtotalpages += real_size;
6421 	}
6422 
6423 	pgdat->node_spanned_pages = totalpages;
6424 	pgdat->node_present_pages = realtotalpages;
6425 	printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
6426 							realtotalpages);
6427 }
6428 
6429 #ifndef CONFIG_SPARSEMEM
6430 /*
6431  * Calculate the size of the zone->blockflags rounded to an unsigned long
6432  * Start by making sure zonesize is a multiple of pageblock_order by rounding
6433  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
6434  * round what is now in bits to nearest long in bits, then return it in
6435  * bytes.
6436  */
6437 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
6438 {
6439 	unsigned long usemapsize;
6440 
6441 	zonesize += zone_start_pfn & (pageblock_nr_pages-1);
6442 	usemapsize = roundup(zonesize, pageblock_nr_pages);
6443 	usemapsize = usemapsize >> pageblock_order;
6444 	usemapsize *= NR_PAGEBLOCK_BITS;
6445 	usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
6446 
6447 	return usemapsize / 8;
6448 }
6449 
6450 static void __ref setup_usemap(struct pglist_data *pgdat,
6451 				struct zone *zone,
6452 				unsigned long zone_start_pfn,
6453 				unsigned long zonesize)
6454 {
6455 	unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
6456 	zone->pageblock_flags = NULL;
6457 	if (usemapsize) {
6458 		zone->pageblock_flags =
6459 			memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
6460 					    pgdat->node_id);
6461 		if (!zone->pageblock_flags)
6462 			panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
6463 			      usemapsize, zone->name, pgdat->node_id);
6464 	}
6465 }
6466 #else
6467 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
6468 				unsigned long zone_start_pfn, unsigned long zonesize) {}
6469 #endif /* CONFIG_SPARSEMEM */
6470 
6471 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
6472 
6473 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
6474 void __init set_pageblock_order(void)
6475 {
6476 	unsigned int order;
6477 
6478 	/* Check that pageblock_nr_pages has not already been setup */
6479 	if (pageblock_order)
6480 		return;
6481 
6482 	if (HPAGE_SHIFT > PAGE_SHIFT)
6483 		order = HUGETLB_PAGE_ORDER;
6484 	else
6485 		order = MAX_ORDER - 1;
6486 
6487 	/*
6488 	 * Assume the largest contiguous order of interest is a huge page.
6489 	 * This value may be variable depending on boot parameters on IA64 and
6490 	 * powerpc.
6491 	 */
6492 	pageblock_order = order;
6493 }
6494 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6495 
6496 /*
6497  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
6498  * is unused as pageblock_order is set at compile-time. See
6499  * include/linux/pageblock-flags.h for the values of pageblock_order based on
6500  * the kernel config
6501  */
6502 void __init set_pageblock_order(void)
6503 {
6504 }
6505 
6506 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6507 
6508 static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
6509 						unsigned long present_pages)
6510 {
6511 	unsigned long pages = spanned_pages;
6512 
6513 	/*
6514 	 * Provide a more accurate estimation if there are holes within
6515 	 * the zone and SPARSEMEM is in use. If there are holes within the
6516 	 * zone, each populated memory region may cost us one or two extra
6517 	 * memmap pages due to alignment because memmap pages for each
6518 	 * populated regions may not be naturally aligned on page boundary.
6519 	 * So the (present_pages >> 4) heuristic is a tradeoff for that.
6520 	 */
6521 	if (spanned_pages > present_pages + (present_pages >> 4) &&
6522 	    IS_ENABLED(CONFIG_SPARSEMEM))
6523 		pages = present_pages;
6524 
6525 	return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
6526 }
6527 
6528 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6529 static void pgdat_init_split_queue(struct pglist_data *pgdat)
6530 {
6531 	spin_lock_init(&pgdat->split_queue_lock);
6532 	INIT_LIST_HEAD(&pgdat->split_queue);
6533 	pgdat->split_queue_len = 0;
6534 }
6535 #else
6536 static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
6537 #endif
6538 
6539 #ifdef CONFIG_COMPACTION
6540 static void pgdat_init_kcompactd(struct pglist_data *pgdat)
6541 {
6542 	init_waitqueue_head(&pgdat->kcompactd_wait);
6543 }
6544 #else
6545 static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
6546 #endif
6547 
6548 static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
6549 {
6550 	pgdat_resize_init(pgdat);
6551 
6552 	pgdat_init_split_queue(pgdat);
6553 	pgdat_init_kcompactd(pgdat);
6554 
6555 	init_waitqueue_head(&pgdat->kswapd_wait);
6556 	init_waitqueue_head(&pgdat->pfmemalloc_wait);
6557 
6558 	pgdat_page_ext_init(pgdat);
6559 	spin_lock_init(&pgdat->lru_lock);
6560 	lruvec_init(node_lruvec(pgdat));
6561 }
6562 
6563 static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
6564 							unsigned long remaining_pages)
6565 {
6566 	atomic_long_set(&zone->managed_pages, remaining_pages);
6567 	zone_set_nid(zone, nid);
6568 	zone->name = zone_names[idx];
6569 	zone->zone_pgdat = NODE_DATA(nid);
6570 	spin_lock_init(&zone->lock);
6571 	zone_seqlock_init(zone);
6572 	zone_pcp_init(zone);
6573 }
6574 
6575 /*
6576  * Set up the zone data structures
6577  * - init pgdat internals
6578  * - init all zones belonging to this node
6579  *
6580  * NOTE: this function is only called during memory hotplug
6581  */
6582 #ifdef CONFIG_MEMORY_HOTPLUG
6583 void __ref free_area_init_core_hotplug(int nid)
6584 {
6585 	enum zone_type z;
6586 	pg_data_t *pgdat = NODE_DATA(nid);
6587 
6588 	pgdat_init_internals(pgdat);
6589 	for (z = 0; z < MAX_NR_ZONES; z++)
6590 		zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
6591 }
6592 #endif
6593 
6594 /*
6595  * Set up the zone data structures:
6596  *   - mark all pages reserved
6597  *   - mark all memory queues empty
6598  *   - clear the memory bitmaps
6599  *
6600  * NOTE: pgdat should get zeroed by caller.
6601  * NOTE: this function is only called during early init.
6602  */
6603 static void __init free_area_init_core(struct pglist_data *pgdat)
6604 {
6605 	enum zone_type j;
6606 	int nid = pgdat->node_id;
6607 
6608 	pgdat_init_internals(pgdat);
6609 	pgdat->per_cpu_nodestats = &boot_nodestats;
6610 
6611 	for (j = 0; j < MAX_NR_ZONES; j++) {
6612 		struct zone *zone = pgdat->node_zones + j;
6613 		unsigned long size, freesize, memmap_pages;
6614 		unsigned long zone_start_pfn = zone->zone_start_pfn;
6615 
6616 		size = zone->spanned_pages;
6617 		freesize = zone->present_pages;
6618 
6619 		/*
6620 		 * Adjust freesize so that it accounts for how much memory
6621 		 * is used by this zone for memmap. This affects the watermark
6622 		 * and per-cpu initialisations
6623 		 */
6624 		memmap_pages = calc_memmap_size(size, freesize);
6625 		if (!is_highmem_idx(j)) {
6626 			if (freesize >= memmap_pages) {
6627 				freesize -= memmap_pages;
6628 				if (memmap_pages)
6629 					printk(KERN_DEBUG
6630 					       "  %s zone: %lu pages used for memmap\n",
6631 					       zone_names[j], memmap_pages);
6632 			} else
6633 				pr_warn("  %s zone: %lu pages exceeds freesize %lu\n",
6634 					zone_names[j], memmap_pages, freesize);
6635 		}
6636 
6637 		/* Account for reserved pages */
6638 		if (j == 0 && freesize > dma_reserve) {
6639 			freesize -= dma_reserve;
6640 			printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
6641 					zone_names[0], dma_reserve);
6642 		}
6643 
6644 		if (!is_highmem_idx(j))
6645 			nr_kernel_pages += freesize;
6646 		/* Charge for highmem memmap if there are enough kernel pages */
6647 		else if (nr_kernel_pages > memmap_pages * 2)
6648 			nr_kernel_pages -= memmap_pages;
6649 		nr_all_pages += freesize;
6650 
6651 		/*
6652 		 * Set an approximate value for lowmem here, it will be adjusted
6653 		 * when the bootmem allocator frees pages into the buddy system.
6654 		 * And all highmem pages will be managed by the buddy system.
6655 		 */
6656 		zone_init_internals(zone, j, nid, freesize);
6657 
6658 		if (!size)
6659 			continue;
6660 
6661 		set_pageblock_order();
6662 		setup_usemap(pgdat, zone, zone_start_pfn, size);
6663 		init_currently_empty_zone(zone, zone_start_pfn, size);
6664 		memmap_init(size, nid, j, zone_start_pfn);
6665 	}
6666 }
6667 
6668 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6669 static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
6670 {
6671 	unsigned long __maybe_unused start = 0;
6672 	unsigned long __maybe_unused offset = 0;
6673 
6674 	/* Skip empty nodes */
6675 	if (!pgdat->node_spanned_pages)
6676 		return;
6677 
6678 	start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
6679 	offset = pgdat->node_start_pfn - start;
6680 	/* ia64 gets its own node_mem_map, before this, without bootmem */
6681 	if (!pgdat->node_mem_map) {
6682 		unsigned long size, end;
6683 		struct page *map;
6684 
6685 		/*
6686 		 * The zone's endpoints aren't required to be MAX_ORDER
6687 		 * aligned but the node_mem_map endpoints must be in order
6688 		 * for the buddy allocator to function correctly.
6689 		 */
6690 		end = pgdat_end_pfn(pgdat);
6691 		end = ALIGN(end, MAX_ORDER_NR_PAGES);
6692 		size =  (end - start) * sizeof(struct page);
6693 		map = memblock_alloc_node(size, SMP_CACHE_BYTES,
6694 					  pgdat->node_id);
6695 		if (!map)
6696 			panic("Failed to allocate %ld bytes for node %d memory map\n",
6697 			      size, pgdat->node_id);
6698 		pgdat->node_mem_map = map + offset;
6699 	}
6700 	pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
6701 				__func__, pgdat->node_id, (unsigned long)pgdat,
6702 				(unsigned long)pgdat->node_mem_map);
6703 #ifndef CONFIG_NEED_MULTIPLE_NODES
6704 	/*
6705 	 * With no DISCONTIG, the global mem_map is just set as node 0's
6706 	 */
6707 	if (pgdat == NODE_DATA(0)) {
6708 		mem_map = NODE_DATA(0)->node_mem_map;
6709 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6710 		if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
6711 			mem_map -= offset;
6712 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6713 	}
6714 #endif
6715 }
6716 #else
6717 static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { }
6718 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6719 
6720 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
6721 static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
6722 {
6723 	pgdat->first_deferred_pfn = ULONG_MAX;
6724 }
6725 #else
6726 static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
6727 #endif
6728 
6729 void __init free_area_init_node(int nid, unsigned long *zones_size,
6730 				   unsigned long node_start_pfn,
6731 				   unsigned long *zholes_size)
6732 {
6733 	pg_data_t *pgdat = NODE_DATA(nid);
6734 	unsigned long start_pfn = 0;
6735 	unsigned long end_pfn = 0;
6736 
6737 	/* pg_data_t should be reset to zero when it's allocated */
6738 	WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
6739 
6740 	pgdat->node_id = nid;
6741 	pgdat->node_start_pfn = node_start_pfn;
6742 	pgdat->per_cpu_nodestats = NULL;
6743 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6744 	get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
6745 	pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
6746 		(u64)start_pfn << PAGE_SHIFT,
6747 		end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
6748 #else
6749 	start_pfn = node_start_pfn;
6750 #endif
6751 	calculate_node_totalpages(pgdat, start_pfn, end_pfn,
6752 				  zones_size, zholes_size);
6753 
6754 	alloc_node_mem_map(pgdat);
6755 	pgdat_set_deferred_range(pgdat);
6756 
6757 	free_area_init_core(pgdat);
6758 }
6759 
6760 #if !defined(CONFIG_FLAT_NODE_MEM_MAP)
6761 /*
6762  * Zero all valid struct pages in range [spfn, epfn), return number of struct
6763  * pages zeroed
6764  */
6765 static u64 zero_pfn_range(unsigned long spfn, unsigned long epfn)
6766 {
6767 	unsigned long pfn;
6768 	u64 pgcnt = 0;
6769 
6770 	for (pfn = spfn; pfn < epfn; pfn++) {
6771 		if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages))) {
6772 			pfn = ALIGN_DOWN(pfn, pageblock_nr_pages)
6773 				+ pageblock_nr_pages - 1;
6774 			continue;
6775 		}
6776 		mm_zero_struct_page(pfn_to_page(pfn));
6777 		pgcnt++;
6778 	}
6779 
6780 	return pgcnt;
6781 }
6782 
6783 /*
6784  * Only struct pages that are backed by physical memory are zeroed and
6785  * initialized by going through __init_single_page(). But, there are some
6786  * struct pages which are reserved in memblock allocator and their fields
6787  * may be accessed (for example page_to_pfn() on some configuration accesses
6788  * flags). We must explicitly zero those struct pages.
6789  *
6790  * This function also addresses a similar issue where struct pages are left
6791  * uninitialized because the physical address range is not covered by
6792  * memblock.memory or memblock.reserved. That could happen when memblock
6793  * layout is manually configured via memmap=.
6794  */
6795 void __init zero_resv_unavail(void)
6796 {
6797 	phys_addr_t start, end;
6798 	u64 i, pgcnt;
6799 	phys_addr_t next = 0;
6800 
6801 	/*
6802 	 * Loop through unavailable ranges not covered by memblock.memory.
6803 	 */
6804 	pgcnt = 0;
6805 	for_each_mem_range(i, &memblock.memory, NULL,
6806 			NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end, NULL) {
6807 		if (next < start)
6808 			pgcnt += zero_pfn_range(PFN_DOWN(next), PFN_UP(start));
6809 		next = end;
6810 	}
6811 	pgcnt += zero_pfn_range(PFN_DOWN(next), max_pfn);
6812 
6813 	/*
6814 	 * Struct pages that do not have backing memory. This could be because
6815 	 * firmware is using some of this memory, or for some other reasons.
6816 	 */
6817 	if (pgcnt)
6818 		pr_info("Zeroed struct page in unavailable ranges: %lld pages", pgcnt);
6819 }
6820 #endif /* !CONFIG_FLAT_NODE_MEM_MAP */
6821 
6822 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6823 
6824 #if MAX_NUMNODES > 1
6825 /*
6826  * Figure out the number of possible node ids.
6827  */
6828 void __init setup_nr_node_ids(void)
6829 {
6830 	unsigned int highest;
6831 
6832 	highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
6833 	nr_node_ids = highest + 1;
6834 }
6835 #endif
6836 
6837 /**
6838  * node_map_pfn_alignment - determine the maximum internode alignment
6839  *
6840  * This function should be called after node map is populated and sorted.
6841  * It calculates the maximum power of two alignment which can distinguish
6842  * all the nodes.
6843  *
6844  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6845  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
6846  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
6847  * shifted, 1GiB is enough and this function will indicate so.
6848  *
6849  * This is used to test whether pfn -> nid mapping of the chosen memory
6850  * model has fine enough granularity to avoid incorrect mapping for the
6851  * populated node map.
6852  *
6853  * Return: the determined alignment in pfn's.  0 if there is no alignment
6854  * requirement (single node).
6855  */
6856 unsigned long __init node_map_pfn_alignment(void)
6857 {
6858 	unsigned long accl_mask = 0, last_end = 0;
6859 	unsigned long start, end, mask;
6860 	int last_nid = NUMA_NO_NODE;
6861 	int i, nid;
6862 
6863 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
6864 		if (!start || last_nid < 0 || last_nid == nid) {
6865 			last_nid = nid;
6866 			last_end = end;
6867 			continue;
6868 		}
6869 
6870 		/*
6871 		 * Start with a mask granular enough to pin-point to the
6872 		 * start pfn and tick off bits one-by-one until it becomes
6873 		 * too coarse to separate the current node from the last.
6874 		 */
6875 		mask = ~((1 << __ffs(start)) - 1);
6876 		while (mask && last_end <= (start & (mask << 1)))
6877 			mask <<= 1;
6878 
6879 		/* accumulate all internode masks */
6880 		accl_mask |= mask;
6881 	}
6882 
6883 	/* convert mask to number of pages */
6884 	return ~accl_mask + 1;
6885 }
6886 
6887 /* Find the lowest pfn for a node */
6888 static unsigned long __init find_min_pfn_for_node(int nid)
6889 {
6890 	unsigned long min_pfn = ULONG_MAX;
6891 	unsigned long start_pfn;
6892 	int i;
6893 
6894 	for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6895 		min_pfn = min(min_pfn, start_pfn);
6896 
6897 	if (min_pfn == ULONG_MAX) {
6898 		pr_warn("Could not find start_pfn for node %d\n", nid);
6899 		return 0;
6900 	}
6901 
6902 	return min_pfn;
6903 }
6904 
6905 /**
6906  * find_min_pfn_with_active_regions - Find the minimum PFN registered
6907  *
6908  * Return: the minimum PFN based on information provided via
6909  * memblock_set_node().
6910  */
6911 unsigned long __init find_min_pfn_with_active_regions(void)
6912 {
6913 	return find_min_pfn_for_node(MAX_NUMNODES);
6914 }
6915 
6916 /*
6917  * early_calculate_totalpages()
6918  * Sum pages in active regions for movable zone.
6919  * Populate N_MEMORY for calculating usable_nodes.
6920  */
6921 static unsigned long __init early_calculate_totalpages(void)
6922 {
6923 	unsigned long totalpages = 0;
6924 	unsigned long start_pfn, end_pfn;
6925 	int i, nid;
6926 
6927 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6928 		unsigned long pages = end_pfn - start_pfn;
6929 
6930 		totalpages += pages;
6931 		if (pages)
6932 			node_set_state(nid, N_MEMORY);
6933 	}
6934 	return totalpages;
6935 }
6936 
6937 /*
6938  * Find the PFN the Movable zone begins in each node. Kernel memory
6939  * is spread evenly between nodes as long as the nodes have enough
6940  * memory. When they don't, some nodes will have more kernelcore than
6941  * others
6942  */
6943 static void __init find_zone_movable_pfns_for_nodes(void)
6944 {
6945 	int i, nid;
6946 	unsigned long usable_startpfn;
6947 	unsigned long kernelcore_node, kernelcore_remaining;
6948 	/* save the state before borrow the nodemask */
6949 	nodemask_t saved_node_state = node_states[N_MEMORY];
6950 	unsigned long totalpages = early_calculate_totalpages();
6951 	int usable_nodes = nodes_weight(node_states[N_MEMORY]);
6952 	struct memblock_region *r;
6953 
6954 	/* Need to find movable_zone earlier when movable_node is specified. */
6955 	find_usable_zone_for_movable();
6956 
6957 	/*
6958 	 * If movable_node is specified, ignore kernelcore and movablecore
6959 	 * options.
6960 	 */
6961 	if (movable_node_is_enabled()) {
6962 		for_each_memblock(memory, r) {
6963 			if (!memblock_is_hotpluggable(r))
6964 				continue;
6965 
6966 			nid = r->nid;
6967 
6968 			usable_startpfn = PFN_DOWN(r->base);
6969 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6970 				min(usable_startpfn, zone_movable_pfn[nid]) :
6971 				usable_startpfn;
6972 		}
6973 
6974 		goto out2;
6975 	}
6976 
6977 	/*
6978 	 * If kernelcore=mirror is specified, ignore movablecore option
6979 	 */
6980 	if (mirrored_kernelcore) {
6981 		bool mem_below_4gb_not_mirrored = false;
6982 
6983 		for_each_memblock(memory, r) {
6984 			if (memblock_is_mirror(r))
6985 				continue;
6986 
6987 			nid = r->nid;
6988 
6989 			usable_startpfn = memblock_region_memory_base_pfn(r);
6990 
6991 			if (usable_startpfn < 0x100000) {
6992 				mem_below_4gb_not_mirrored = true;
6993 				continue;
6994 			}
6995 
6996 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6997 				min(usable_startpfn, zone_movable_pfn[nid]) :
6998 				usable_startpfn;
6999 		}
7000 
7001 		if (mem_below_4gb_not_mirrored)
7002 			pr_warn("This configuration results in unmirrored kernel memory.");
7003 
7004 		goto out2;
7005 	}
7006 
7007 	/*
7008 	 * If kernelcore=nn% or movablecore=nn% was specified, calculate the
7009 	 * amount of necessary memory.
7010 	 */
7011 	if (required_kernelcore_percent)
7012 		required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
7013 				       10000UL;
7014 	if (required_movablecore_percent)
7015 		required_movablecore = (totalpages * 100 * required_movablecore_percent) /
7016 					10000UL;
7017 
7018 	/*
7019 	 * If movablecore= was specified, calculate what size of
7020 	 * kernelcore that corresponds so that memory usable for
7021 	 * any allocation type is evenly spread. If both kernelcore
7022 	 * and movablecore are specified, then the value of kernelcore
7023 	 * will be used for required_kernelcore if it's greater than
7024 	 * what movablecore would have allowed.
7025 	 */
7026 	if (required_movablecore) {
7027 		unsigned long corepages;
7028 
7029 		/*
7030 		 * Round-up so that ZONE_MOVABLE is at least as large as what
7031 		 * was requested by the user
7032 		 */
7033 		required_movablecore =
7034 			roundup(required_movablecore, MAX_ORDER_NR_PAGES);
7035 		required_movablecore = min(totalpages, required_movablecore);
7036 		corepages = totalpages - required_movablecore;
7037 
7038 		required_kernelcore = max(required_kernelcore, corepages);
7039 	}
7040 
7041 	/*
7042 	 * If kernelcore was not specified or kernelcore size is larger
7043 	 * than totalpages, there is no ZONE_MOVABLE.
7044 	 */
7045 	if (!required_kernelcore || required_kernelcore >= totalpages)
7046 		goto out;
7047 
7048 	/* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
7049 	usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
7050 
7051 restart:
7052 	/* Spread kernelcore memory as evenly as possible throughout nodes */
7053 	kernelcore_node = required_kernelcore / usable_nodes;
7054 	for_each_node_state(nid, N_MEMORY) {
7055 		unsigned long start_pfn, end_pfn;
7056 
7057 		/*
7058 		 * Recalculate kernelcore_node if the division per node
7059 		 * now exceeds what is necessary to satisfy the requested
7060 		 * amount of memory for the kernel
7061 		 */
7062 		if (required_kernelcore < kernelcore_node)
7063 			kernelcore_node = required_kernelcore / usable_nodes;
7064 
7065 		/*
7066 		 * As the map is walked, we track how much memory is usable
7067 		 * by the kernel using kernelcore_remaining. When it is
7068 		 * 0, the rest of the node is usable by ZONE_MOVABLE
7069 		 */
7070 		kernelcore_remaining = kernelcore_node;
7071 
7072 		/* Go through each range of PFNs within this node */
7073 		for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
7074 			unsigned long size_pages;
7075 
7076 			start_pfn = max(start_pfn, zone_movable_pfn[nid]);
7077 			if (start_pfn >= end_pfn)
7078 				continue;
7079 
7080 			/* Account for what is only usable for kernelcore */
7081 			if (start_pfn < usable_startpfn) {
7082 				unsigned long kernel_pages;
7083 				kernel_pages = min(end_pfn, usable_startpfn)
7084 								- start_pfn;
7085 
7086 				kernelcore_remaining -= min(kernel_pages,
7087 							kernelcore_remaining);
7088 				required_kernelcore -= min(kernel_pages,
7089 							required_kernelcore);
7090 
7091 				/* Continue if range is now fully accounted */
7092 				if (end_pfn <= usable_startpfn) {
7093 
7094 					/*
7095 					 * Push zone_movable_pfn to the end so
7096 					 * that if we have to rebalance
7097 					 * kernelcore across nodes, we will
7098 					 * not double account here
7099 					 */
7100 					zone_movable_pfn[nid] = end_pfn;
7101 					continue;
7102 				}
7103 				start_pfn = usable_startpfn;
7104 			}
7105 
7106 			/*
7107 			 * The usable PFN range for ZONE_MOVABLE is from
7108 			 * start_pfn->end_pfn. Calculate size_pages as the
7109 			 * number of pages used as kernelcore
7110 			 */
7111 			size_pages = end_pfn - start_pfn;
7112 			if (size_pages > kernelcore_remaining)
7113 				size_pages = kernelcore_remaining;
7114 			zone_movable_pfn[nid] = start_pfn + size_pages;
7115 
7116 			/*
7117 			 * Some kernelcore has been met, update counts and
7118 			 * break if the kernelcore for this node has been
7119 			 * satisfied
7120 			 */
7121 			required_kernelcore -= min(required_kernelcore,
7122 								size_pages);
7123 			kernelcore_remaining -= size_pages;
7124 			if (!kernelcore_remaining)
7125 				break;
7126 		}
7127 	}
7128 
7129 	/*
7130 	 * If there is still required_kernelcore, we do another pass with one
7131 	 * less node in the count. This will push zone_movable_pfn[nid] further
7132 	 * along on the nodes that still have memory until kernelcore is
7133 	 * satisfied
7134 	 */
7135 	usable_nodes--;
7136 	if (usable_nodes && required_kernelcore > usable_nodes)
7137 		goto restart;
7138 
7139 out2:
7140 	/* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
7141 	for (nid = 0; nid < MAX_NUMNODES; nid++)
7142 		zone_movable_pfn[nid] =
7143 			roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
7144 
7145 out:
7146 	/* restore the node_state */
7147 	node_states[N_MEMORY] = saved_node_state;
7148 }
7149 
7150 /* Any regular or high memory on that node ? */
7151 static void check_for_memory(pg_data_t *pgdat, int nid)
7152 {
7153 	enum zone_type zone_type;
7154 
7155 	for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
7156 		struct zone *zone = &pgdat->node_zones[zone_type];
7157 		if (populated_zone(zone)) {
7158 			if (IS_ENABLED(CONFIG_HIGHMEM))
7159 				node_set_state(nid, N_HIGH_MEMORY);
7160 			if (zone_type <= ZONE_NORMAL)
7161 				node_set_state(nid, N_NORMAL_MEMORY);
7162 			break;
7163 		}
7164 	}
7165 }
7166 
7167 /**
7168  * free_area_init_nodes - Initialise all pg_data_t and zone data
7169  * @max_zone_pfn: an array of max PFNs for each zone
7170  *
7171  * This will call free_area_init_node() for each active node in the system.
7172  * Using the page ranges provided by memblock_set_node(), the size of each
7173  * zone in each node and their holes is calculated. If the maximum PFN
7174  * between two adjacent zones match, it is assumed that the zone is empty.
7175  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
7176  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
7177  * starts where the previous one ended. For example, ZONE_DMA32 starts
7178  * at arch_max_dma_pfn.
7179  */
7180 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
7181 {
7182 	unsigned long start_pfn, end_pfn;
7183 	int i, nid;
7184 
7185 	/* Record where the zone boundaries are */
7186 	memset(arch_zone_lowest_possible_pfn, 0,
7187 				sizeof(arch_zone_lowest_possible_pfn));
7188 	memset(arch_zone_highest_possible_pfn, 0,
7189 				sizeof(arch_zone_highest_possible_pfn));
7190 
7191 	start_pfn = find_min_pfn_with_active_regions();
7192 
7193 	for (i = 0; i < MAX_NR_ZONES; i++) {
7194 		if (i == ZONE_MOVABLE)
7195 			continue;
7196 
7197 		end_pfn = max(max_zone_pfn[i], start_pfn);
7198 		arch_zone_lowest_possible_pfn[i] = start_pfn;
7199 		arch_zone_highest_possible_pfn[i] = end_pfn;
7200 
7201 		start_pfn = end_pfn;
7202 	}
7203 
7204 	/* Find the PFNs that ZONE_MOVABLE begins at in each node */
7205 	memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
7206 	find_zone_movable_pfns_for_nodes();
7207 
7208 	/* Print out the zone ranges */
7209 	pr_info("Zone ranges:\n");
7210 	for (i = 0; i < MAX_NR_ZONES; i++) {
7211 		if (i == ZONE_MOVABLE)
7212 			continue;
7213 		pr_info("  %-8s ", zone_names[i]);
7214 		if (arch_zone_lowest_possible_pfn[i] ==
7215 				arch_zone_highest_possible_pfn[i])
7216 			pr_cont("empty\n");
7217 		else
7218 			pr_cont("[mem %#018Lx-%#018Lx]\n",
7219 				(u64)arch_zone_lowest_possible_pfn[i]
7220 					<< PAGE_SHIFT,
7221 				((u64)arch_zone_highest_possible_pfn[i]
7222 					<< PAGE_SHIFT) - 1);
7223 	}
7224 
7225 	/* Print out the PFNs ZONE_MOVABLE begins at in each node */
7226 	pr_info("Movable zone start for each node\n");
7227 	for (i = 0; i < MAX_NUMNODES; i++) {
7228 		if (zone_movable_pfn[i])
7229 			pr_info("  Node %d: %#018Lx\n", i,
7230 			       (u64)zone_movable_pfn[i] << PAGE_SHIFT);
7231 	}
7232 
7233 	/* Print out the early node map */
7234 	pr_info("Early memory node ranges\n");
7235 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
7236 		pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
7237 			(u64)start_pfn << PAGE_SHIFT,
7238 			((u64)end_pfn << PAGE_SHIFT) - 1);
7239 
7240 	/* Initialise every node */
7241 	mminit_verify_pageflags_layout();
7242 	setup_nr_node_ids();
7243 	zero_resv_unavail();
7244 	for_each_online_node(nid) {
7245 		pg_data_t *pgdat = NODE_DATA(nid);
7246 		free_area_init_node(nid, NULL,
7247 				find_min_pfn_for_node(nid), NULL);
7248 
7249 		/* Any memory on that node */
7250 		if (pgdat->node_present_pages)
7251 			node_set_state(nid, N_MEMORY);
7252 		check_for_memory(pgdat, nid);
7253 	}
7254 }
7255 
7256 static int __init cmdline_parse_core(char *p, unsigned long *core,
7257 				     unsigned long *percent)
7258 {
7259 	unsigned long long coremem;
7260 	char *endptr;
7261 
7262 	if (!p)
7263 		return -EINVAL;
7264 
7265 	/* Value may be a percentage of total memory, otherwise bytes */
7266 	coremem = simple_strtoull(p, &endptr, 0);
7267 	if (*endptr == '%') {
7268 		/* Paranoid check for percent values greater than 100 */
7269 		WARN_ON(coremem > 100);
7270 
7271 		*percent = coremem;
7272 	} else {
7273 		coremem = memparse(p, &p);
7274 		/* Paranoid check that UL is enough for the coremem value */
7275 		WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
7276 
7277 		*core = coremem >> PAGE_SHIFT;
7278 		*percent = 0UL;
7279 	}
7280 	return 0;
7281 }
7282 
7283 /*
7284  * kernelcore=size sets the amount of memory for use for allocations that
7285  * cannot be reclaimed or migrated.
7286  */
7287 static int __init cmdline_parse_kernelcore(char *p)
7288 {
7289 	/* parse kernelcore=mirror */
7290 	if (parse_option_str(p, "mirror")) {
7291 		mirrored_kernelcore = true;
7292 		return 0;
7293 	}
7294 
7295 	return cmdline_parse_core(p, &required_kernelcore,
7296 				  &required_kernelcore_percent);
7297 }
7298 
7299 /*
7300  * movablecore=size sets the amount of memory for use for allocations that
7301  * can be reclaimed or migrated.
7302  */
7303 static int __init cmdline_parse_movablecore(char *p)
7304 {
7305 	return cmdline_parse_core(p, &required_movablecore,
7306 				  &required_movablecore_percent);
7307 }
7308 
7309 early_param("kernelcore", cmdline_parse_kernelcore);
7310 early_param("movablecore", cmdline_parse_movablecore);
7311 
7312 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
7313 
7314 void adjust_managed_page_count(struct page *page, long count)
7315 {
7316 	atomic_long_add(count, &page_zone(page)->managed_pages);
7317 	totalram_pages_add(count);
7318 #ifdef CONFIG_HIGHMEM
7319 	if (PageHighMem(page))
7320 		totalhigh_pages_add(count);
7321 #endif
7322 }
7323 EXPORT_SYMBOL(adjust_managed_page_count);
7324 
7325 unsigned long free_reserved_area(void *start, void *end, int poison, const char *s)
7326 {
7327 	void *pos;
7328 	unsigned long pages = 0;
7329 
7330 	start = (void *)PAGE_ALIGN((unsigned long)start);
7331 	end = (void *)((unsigned long)end & PAGE_MASK);
7332 	for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
7333 		struct page *page = virt_to_page(pos);
7334 		void *direct_map_addr;
7335 
7336 		/*
7337 		 * 'direct_map_addr' might be different from 'pos'
7338 		 * because some architectures' virt_to_page()
7339 		 * work with aliases.  Getting the direct map
7340 		 * address ensures that we get a _writeable_
7341 		 * alias for the memset().
7342 		 */
7343 		direct_map_addr = page_address(page);
7344 		if ((unsigned int)poison <= 0xFF)
7345 			memset(direct_map_addr, poison, PAGE_SIZE);
7346 
7347 		free_reserved_page(page);
7348 	}
7349 
7350 	if (pages && s)
7351 		pr_info("Freeing %s memory: %ldK\n",
7352 			s, pages << (PAGE_SHIFT - 10));
7353 
7354 	return pages;
7355 }
7356 
7357 #ifdef	CONFIG_HIGHMEM
7358 void free_highmem_page(struct page *page)
7359 {
7360 	__free_reserved_page(page);
7361 	totalram_pages_inc();
7362 	atomic_long_inc(&page_zone(page)->managed_pages);
7363 	totalhigh_pages_inc();
7364 }
7365 #endif
7366 
7367 
7368 void __init mem_init_print_info(const char *str)
7369 {
7370 	unsigned long physpages, codesize, datasize, rosize, bss_size;
7371 	unsigned long init_code_size, init_data_size;
7372 
7373 	physpages = get_num_physpages();
7374 	codesize = _etext - _stext;
7375 	datasize = _edata - _sdata;
7376 	rosize = __end_rodata - __start_rodata;
7377 	bss_size = __bss_stop - __bss_start;
7378 	init_data_size = __init_end - __init_begin;
7379 	init_code_size = _einittext - _sinittext;
7380 
7381 	/*
7382 	 * Detect special cases and adjust section sizes accordingly:
7383 	 * 1) .init.* may be embedded into .data sections
7384 	 * 2) .init.text.* may be out of [__init_begin, __init_end],
7385 	 *    please refer to arch/tile/kernel/vmlinux.lds.S.
7386 	 * 3) .rodata.* may be embedded into .text or .data sections.
7387 	 */
7388 #define adj_init_size(start, end, size, pos, adj) \
7389 	do { \
7390 		if (start <= pos && pos < end && size > adj) \
7391 			size -= adj; \
7392 	} while (0)
7393 
7394 	adj_init_size(__init_begin, __init_end, init_data_size,
7395 		     _sinittext, init_code_size);
7396 	adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
7397 	adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
7398 	adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
7399 	adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
7400 
7401 #undef	adj_init_size
7402 
7403 	pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
7404 #ifdef	CONFIG_HIGHMEM
7405 		", %luK highmem"
7406 #endif
7407 		"%s%s)\n",
7408 		nr_free_pages() << (PAGE_SHIFT - 10),
7409 		physpages << (PAGE_SHIFT - 10),
7410 		codesize >> 10, datasize >> 10, rosize >> 10,
7411 		(init_data_size + init_code_size) >> 10, bss_size >> 10,
7412 		(physpages - totalram_pages() - totalcma_pages) << (PAGE_SHIFT - 10),
7413 		totalcma_pages << (PAGE_SHIFT - 10),
7414 #ifdef	CONFIG_HIGHMEM
7415 		totalhigh_pages() << (PAGE_SHIFT - 10),
7416 #endif
7417 		str ? ", " : "", str ? str : "");
7418 }
7419 
7420 /**
7421  * set_dma_reserve - set the specified number of pages reserved in the first zone
7422  * @new_dma_reserve: The number of pages to mark reserved
7423  *
7424  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
7425  * In the DMA zone, a significant percentage may be consumed by kernel image
7426  * and other unfreeable allocations which can skew the watermarks badly. This
7427  * function may optionally be used to account for unfreeable pages in the
7428  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
7429  * smaller per-cpu batchsize.
7430  */
7431 void __init set_dma_reserve(unsigned long new_dma_reserve)
7432 {
7433 	dma_reserve = new_dma_reserve;
7434 }
7435 
7436 void __init free_area_init(unsigned long *zones_size)
7437 {
7438 	zero_resv_unavail();
7439 	free_area_init_node(0, zones_size,
7440 			__pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
7441 }
7442 
7443 static int page_alloc_cpu_dead(unsigned int cpu)
7444 {
7445 
7446 	lru_add_drain_cpu(cpu);
7447 	drain_pages(cpu);
7448 
7449 	/*
7450 	 * Spill the event counters of the dead processor
7451 	 * into the current processors event counters.
7452 	 * This artificially elevates the count of the current
7453 	 * processor.
7454 	 */
7455 	vm_events_fold_cpu(cpu);
7456 
7457 	/*
7458 	 * Zero the differential counters of the dead processor
7459 	 * so that the vm statistics are consistent.
7460 	 *
7461 	 * This is only okay since the processor is dead and cannot
7462 	 * race with what we are doing.
7463 	 */
7464 	cpu_vm_stats_fold(cpu);
7465 	return 0;
7466 }
7467 
7468 void __init page_alloc_init(void)
7469 {
7470 	int ret;
7471 
7472 	ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
7473 					"mm/page_alloc:dead", NULL,
7474 					page_alloc_cpu_dead);
7475 	WARN_ON(ret < 0);
7476 }
7477 
7478 /*
7479  * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
7480  *	or min_free_kbytes changes.
7481  */
7482 static void calculate_totalreserve_pages(void)
7483 {
7484 	struct pglist_data *pgdat;
7485 	unsigned long reserve_pages = 0;
7486 	enum zone_type i, j;
7487 
7488 	for_each_online_pgdat(pgdat) {
7489 
7490 		pgdat->totalreserve_pages = 0;
7491 
7492 		for (i = 0; i < MAX_NR_ZONES; i++) {
7493 			struct zone *zone = pgdat->node_zones + i;
7494 			long max = 0;
7495 			unsigned long managed_pages = zone_managed_pages(zone);
7496 
7497 			/* Find valid and maximum lowmem_reserve in the zone */
7498 			for (j = i; j < MAX_NR_ZONES; j++) {
7499 				if (zone->lowmem_reserve[j] > max)
7500 					max = zone->lowmem_reserve[j];
7501 			}
7502 
7503 			/* we treat the high watermark as reserved pages. */
7504 			max += high_wmark_pages(zone);
7505 
7506 			if (max > managed_pages)
7507 				max = managed_pages;
7508 
7509 			pgdat->totalreserve_pages += max;
7510 
7511 			reserve_pages += max;
7512 		}
7513 	}
7514 	totalreserve_pages = reserve_pages;
7515 }
7516 
7517 /*
7518  * setup_per_zone_lowmem_reserve - called whenever
7519  *	sysctl_lowmem_reserve_ratio changes.  Ensures that each zone
7520  *	has a correct pages reserved value, so an adequate number of
7521  *	pages are left in the zone after a successful __alloc_pages().
7522  */
7523 static void setup_per_zone_lowmem_reserve(void)
7524 {
7525 	struct pglist_data *pgdat;
7526 	enum zone_type j, idx;
7527 
7528 	for_each_online_pgdat(pgdat) {
7529 		for (j = 0; j < MAX_NR_ZONES; j++) {
7530 			struct zone *zone = pgdat->node_zones + j;
7531 			unsigned long managed_pages = zone_managed_pages(zone);
7532 
7533 			zone->lowmem_reserve[j] = 0;
7534 
7535 			idx = j;
7536 			while (idx) {
7537 				struct zone *lower_zone;
7538 
7539 				idx--;
7540 				lower_zone = pgdat->node_zones + idx;
7541 
7542 				if (sysctl_lowmem_reserve_ratio[idx] < 1) {
7543 					sysctl_lowmem_reserve_ratio[idx] = 0;
7544 					lower_zone->lowmem_reserve[j] = 0;
7545 				} else {
7546 					lower_zone->lowmem_reserve[j] =
7547 						managed_pages / sysctl_lowmem_reserve_ratio[idx];
7548 				}
7549 				managed_pages += zone_managed_pages(lower_zone);
7550 			}
7551 		}
7552 	}
7553 
7554 	/* update totalreserve_pages */
7555 	calculate_totalreserve_pages();
7556 }
7557 
7558 static void __setup_per_zone_wmarks(void)
7559 {
7560 	unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
7561 	unsigned long lowmem_pages = 0;
7562 	struct zone *zone;
7563 	unsigned long flags;
7564 
7565 	/* Calculate total number of !ZONE_HIGHMEM pages */
7566 	for_each_zone(zone) {
7567 		if (!is_highmem(zone))
7568 			lowmem_pages += zone_managed_pages(zone);
7569 	}
7570 
7571 	for_each_zone(zone) {
7572 		u64 tmp;
7573 
7574 		spin_lock_irqsave(&zone->lock, flags);
7575 		tmp = (u64)pages_min * zone_managed_pages(zone);
7576 		do_div(tmp, lowmem_pages);
7577 		if (is_highmem(zone)) {
7578 			/*
7579 			 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
7580 			 * need highmem pages, so cap pages_min to a small
7581 			 * value here.
7582 			 *
7583 			 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
7584 			 * deltas control async page reclaim, and so should
7585 			 * not be capped for highmem.
7586 			 */
7587 			unsigned long min_pages;
7588 
7589 			min_pages = zone_managed_pages(zone) / 1024;
7590 			min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
7591 			zone->_watermark[WMARK_MIN] = min_pages;
7592 		} else {
7593 			/*
7594 			 * If it's a lowmem zone, reserve a number of pages
7595 			 * proportionate to the zone's size.
7596 			 */
7597 			zone->_watermark[WMARK_MIN] = tmp;
7598 		}
7599 
7600 		/*
7601 		 * Set the kswapd watermarks distance according to the
7602 		 * scale factor in proportion to available memory, but
7603 		 * ensure a minimum size on small systems.
7604 		 */
7605 		tmp = max_t(u64, tmp >> 2,
7606 			    mult_frac(zone_managed_pages(zone),
7607 				      watermark_scale_factor, 10000));
7608 
7609 		zone->_watermark[WMARK_LOW]  = min_wmark_pages(zone) + tmp;
7610 		zone->_watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
7611 		zone->watermark_boost = 0;
7612 
7613 		spin_unlock_irqrestore(&zone->lock, flags);
7614 	}
7615 
7616 	/* update totalreserve_pages */
7617 	calculate_totalreserve_pages();
7618 }
7619 
7620 /**
7621  * setup_per_zone_wmarks - called when min_free_kbytes changes
7622  * or when memory is hot-{added|removed}
7623  *
7624  * Ensures that the watermark[min,low,high] values for each zone are set
7625  * correctly with respect to min_free_kbytes.
7626  */
7627 void setup_per_zone_wmarks(void)
7628 {
7629 	static DEFINE_SPINLOCK(lock);
7630 
7631 	spin_lock(&lock);
7632 	__setup_per_zone_wmarks();
7633 	spin_unlock(&lock);
7634 }
7635 
7636 /*
7637  * Initialise min_free_kbytes.
7638  *
7639  * For small machines we want it small (128k min).  For large machines
7640  * we want it large (64MB max).  But it is not linear, because network
7641  * bandwidth does not increase linearly with machine size.  We use
7642  *
7643  *	min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
7644  *	min_free_kbytes = sqrt(lowmem_kbytes * 16)
7645  *
7646  * which yields
7647  *
7648  * 16MB:	512k
7649  * 32MB:	724k
7650  * 64MB:	1024k
7651  * 128MB:	1448k
7652  * 256MB:	2048k
7653  * 512MB:	2896k
7654  * 1024MB:	4096k
7655  * 2048MB:	5792k
7656  * 4096MB:	8192k
7657  * 8192MB:	11584k
7658  * 16384MB:	16384k
7659  */
7660 int __meminit init_per_zone_wmark_min(void)
7661 {
7662 	unsigned long lowmem_kbytes;
7663 	int new_min_free_kbytes;
7664 
7665 	lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
7666 	new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
7667 
7668 	if (new_min_free_kbytes > user_min_free_kbytes) {
7669 		min_free_kbytes = new_min_free_kbytes;
7670 		if (min_free_kbytes < 128)
7671 			min_free_kbytes = 128;
7672 		if (min_free_kbytes > 65536)
7673 			min_free_kbytes = 65536;
7674 	} else {
7675 		pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
7676 				new_min_free_kbytes, user_min_free_kbytes);
7677 	}
7678 	setup_per_zone_wmarks();
7679 	refresh_zone_stat_thresholds();
7680 	setup_per_zone_lowmem_reserve();
7681 
7682 #ifdef CONFIG_NUMA
7683 	setup_min_unmapped_ratio();
7684 	setup_min_slab_ratio();
7685 #endif
7686 
7687 	return 0;
7688 }
7689 core_initcall(init_per_zone_wmark_min)
7690 
7691 /*
7692  * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
7693  *	that we can call two helper functions whenever min_free_kbytes
7694  *	changes.
7695  */
7696 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
7697 	void __user *buffer, size_t *length, loff_t *ppos)
7698 {
7699 	int rc;
7700 
7701 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7702 	if (rc)
7703 		return rc;
7704 
7705 	if (write) {
7706 		user_min_free_kbytes = min_free_kbytes;
7707 		setup_per_zone_wmarks();
7708 	}
7709 	return 0;
7710 }
7711 
7712 int watermark_boost_factor_sysctl_handler(struct ctl_table *table, int write,
7713 	void __user *buffer, size_t *length, loff_t *ppos)
7714 {
7715 	int rc;
7716 
7717 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7718 	if (rc)
7719 		return rc;
7720 
7721 	return 0;
7722 }
7723 
7724 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
7725 	void __user *buffer, size_t *length, loff_t *ppos)
7726 {
7727 	int rc;
7728 
7729 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7730 	if (rc)
7731 		return rc;
7732 
7733 	if (write)
7734 		setup_per_zone_wmarks();
7735 
7736 	return 0;
7737 }
7738 
7739 #ifdef CONFIG_NUMA
7740 static void setup_min_unmapped_ratio(void)
7741 {
7742 	pg_data_t *pgdat;
7743 	struct zone *zone;
7744 
7745 	for_each_online_pgdat(pgdat)
7746 		pgdat->min_unmapped_pages = 0;
7747 
7748 	for_each_zone(zone)
7749 		zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) *
7750 						         sysctl_min_unmapped_ratio) / 100;
7751 }
7752 
7753 
7754 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
7755 	void __user *buffer, size_t *length, loff_t *ppos)
7756 {
7757 	int rc;
7758 
7759 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7760 	if (rc)
7761 		return rc;
7762 
7763 	setup_min_unmapped_ratio();
7764 
7765 	return 0;
7766 }
7767 
7768 static void setup_min_slab_ratio(void)
7769 {
7770 	pg_data_t *pgdat;
7771 	struct zone *zone;
7772 
7773 	for_each_online_pgdat(pgdat)
7774 		pgdat->min_slab_pages = 0;
7775 
7776 	for_each_zone(zone)
7777 		zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) *
7778 						     sysctl_min_slab_ratio) / 100;
7779 }
7780 
7781 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
7782 	void __user *buffer, size_t *length, loff_t *ppos)
7783 {
7784 	int rc;
7785 
7786 	rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7787 	if (rc)
7788 		return rc;
7789 
7790 	setup_min_slab_ratio();
7791 
7792 	return 0;
7793 }
7794 #endif
7795 
7796 /*
7797  * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
7798  *	proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
7799  *	whenever sysctl_lowmem_reserve_ratio changes.
7800  *
7801  * The reserve ratio obviously has absolutely no relation with the
7802  * minimum watermarks. The lowmem reserve ratio can only make sense
7803  * if in function of the boot time zone sizes.
7804  */
7805 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
7806 	void __user *buffer, size_t *length, loff_t *ppos)
7807 {
7808 	proc_dointvec_minmax(table, write, buffer, length, ppos);
7809 	setup_per_zone_lowmem_reserve();
7810 	return 0;
7811 }
7812 
7813 /*
7814  * percpu_pagelist_fraction - changes the pcp->high for each zone on each
7815  * cpu.  It is the fraction of total pages in each zone that a hot per cpu
7816  * pagelist can have before it gets flushed back to buddy allocator.
7817  */
7818 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
7819 	void __user *buffer, size_t *length, loff_t *ppos)
7820 {
7821 	struct zone *zone;
7822 	int old_percpu_pagelist_fraction;
7823 	int ret;
7824 
7825 	mutex_lock(&pcp_batch_high_lock);
7826 	old_percpu_pagelist_fraction = percpu_pagelist_fraction;
7827 
7828 	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
7829 	if (!write || ret < 0)
7830 		goto out;
7831 
7832 	/* Sanity checking to avoid pcp imbalance */
7833 	if (percpu_pagelist_fraction &&
7834 	    percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
7835 		percpu_pagelist_fraction = old_percpu_pagelist_fraction;
7836 		ret = -EINVAL;
7837 		goto out;
7838 	}
7839 
7840 	/* No change? */
7841 	if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
7842 		goto out;
7843 
7844 	for_each_populated_zone(zone) {
7845 		unsigned int cpu;
7846 
7847 		for_each_possible_cpu(cpu)
7848 			pageset_set_high_and_batch(zone,
7849 					per_cpu_ptr(zone->pageset, cpu));
7850 	}
7851 out:
7852 	mutex_unlock(&pcp_batch_high_lock);
7853 	return ret;
7854 }
7855 
7856 #ifdef CONFIG_NUMA
7857 int hashdist = HASHDIST_DEFAULT;
7858 
7859 static int __init set_hashdist(char *str)
7860 {
7861 	if (!str)
7862 		return 0;
7863 	hashdist = simple_strtoul(str, &str, 0);
7864 	return 1;
7865 }
7866 __setup("hashdist=", set_hashdist);
7867 #endif
7868 
7869 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7870 /*
7871  * Returns the number of pages that arch has reserved but
7872  * is not known to alloc_large_system_hash().
7873  */
7874 static unsigned long __init arch_reserved_kernel_pages(void)
7875 {
7876 	return 0;
7877 }
7878 #endif
7879 
7880 /*
7881  * Adaptive scale is meant to reduce sizes of hash tables on large memory
7882  * machines. As memory size is increased the scale is also increased but at
7883  * slower pace.  Starting from ADAPT_SCALE_BASE (64G), every time memory
7884  * quadruples the scale is increased by one, which means the size of hash table
7885  * only doubles, instead of quadrupling as well.
7886  * Because 32-bit systems cannot have large physical memory, where this scaling
7887  * makes sense, it is disabled on such platforms.
7888  */
7889 #if __BITS_PER_LONG > 32
7890 #define ADAPT_SCALE_BASE	(64ul << 30)
7891 #define ADAPT_SCALE_SHIFT	2
7892 #define ADAPT_SCALE_NPAGES	(ADAPT_SCALE_BASE >> PAGE_SHIFT)
7893 #endif
7894 
7895 /*
7896  * allocate a large system hash table from bootmem
7897  * - it is assumed that the hash table must contain an exact power-of-2
7898  *   quantity of entries
7899  * - limit is the number of hash buckets, not the total allocation size
7900  */
7901 void *__init alloc_large_system_hash(const char *tablename,
7902 				     unsigned long bucketsize,
7903 				     unsigned long numentries,
7904 				     int scale,
7905 				     int flags,
7906 				     unsigned int *_hash_shift,
7907 				     unsigned int *_hash_mask,
7908 				     unsigned long low_limit,
7909 				     unsigned long high_limit)
7910 {
7911 	unsigned long long max = high_limit;
7912 	unsigned long log2qty, size;
7913 	void *table = NULL;
7914 	gfp_t gfp_flags;
7915 
7916 	/* allow the kernel cmdline to have a say */
7917 	if (!numentries) {
7918 		/* round applicable memory size up to nearest megabyte */
7919 		numentries = nr_kernel_pages;
7920 		numentries -= arch_reserved_kernel_pages();
7921 
7922 		/* It isn't necessary when PAGE_SIZE >= 1MB */
7923 		if (PAGE_SHIFT < 20)
7924 			numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
7925 
7926 #if __BITS_PER_LONG > 32
7927 		if (!high_limit) {
7928 			unsigned long adapt;
7929 
7930 			for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
7931 			     adapt <<= ADAPT_SCALE_SHIFT)
7932 				scale++;
7933 		}
7934 #endif
7935 
7936 		/* limit to 1 bucket per 2^scale bytes of low memory */
7937 		if (scale > PAGE_SHIFT)
7938 			numentries >>= (scale - PAGE_SHIFT);
7939 		else
7940 			numentries <<= (PAGE_SHIFT - scale);
7941 
7942 		/* Make sure we've got at least a 0-order allocation.. */
7943 		if (unlikely(flags & HASH_SMALL)) {
7944 			/* Makes no sense without HASH_EARLY */
7945 			WARN_ON(!(flags & HASH_EARLY));
7946 			if (!(numentries >> *_hash_shift)) {
7947 				numentries = 1UL << *_hash_shift;
7948 				BUG_ON(!numentries);
7949 			}
7950 		} else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
7951 			numentries = PAGE_SIZE / bucketsize;
7952 	}
7953 	numentries = roundup_pow_of_two(numentries);
7954 
7955 	/* limit allocation size to 1/16 total memory by default */
7956 	if (max == 0) {
7957 		max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
7958 		do_div(max, bucketsize);
7959 	}
7960 	max = min(max, 0x80000000ULL);
7961 
7962 	if (numentries < low_limit)
7963 		numentries = low_limit;
7964 	if (numentries > max)
7965 		numentries = max;
7966 
7967 	log2qty = ilog2(numentries);
7968 
7969 	gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
7970 	do {
7971 		size = bucketsize << log2qty;
7972 		if (flags & HASH_EARLY) {
7973 			if (flags & HASH_ZERO)
7974 				table = memblock_alloc(size, SMP_CACHE_BYTES);
7975 			else
7976 				table = memblock_alloc_raw(size,
7977 							   SMP_CACHE_BYTES);
7978 		} else if (hashdist) {
7979 			table = __vmalloc(size, gfp_flags, PAGE_KERNEL);
7980 		} else {
7981 			/*
7982 			 * If bucketsize is not a power-of-two, we may free
7983 			 * some pages at the end of hash table which
7984 			 * alloc_pages_exact() automatically does
7985 			 */
7986 			if (get_order(size) < MAX_ORDER) {
7987 				table = alloc_pages_exact(size, gfp_flags);
7988 				kmemleak_alloc(table, size, 1, gfp_flags);
7989 			}
7990 		}
7991 	} while (!table && size > PAGE_SIZE && --log2qty);
7992 
7993 	if (!table)
7994 		panic("Failed to allocate %s hash table\n", tablename);
7995 
7996 	pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7997 		tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
7998 
7999 	if (_hash_shift)
8000 		*_hash_shift = log2qty;
8001 	if (_hash_mask)
8002 		*_hash_mask = (1 << log2qty) - 1;
8003 
8004 	return table;
8005 }
8006 
8007 /*
8008  * This function checks whether pageblock includes unmovable pages or not.
8009  * If @count is not zero, it is okay to include less @count unmovable pages
8010  *
8011  * PageLRU check without isolation or lru_lock could race so that
8012  * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
8013  * check without lock_page also may miss some movable non-lru pages at
8014  * race condition. So you can't expect this function should be exact.
8015  */
8016 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
8017 			 int migratetype, int flags)
8018 {
8019 	unsigned long found;
8020 	unsigned long iter = 0;
8021 	unsigned long pfn = page_to_pfn(page);
8022 	const char *reason = "unmovable page";
8023 
8024 	/*
8025 	 * TODO we could make this much more efficient by not checking every
8026 	 * page in the range if we know all of them are in MOVABLE_ZONE and
8027 	 * that the movable zone guarantees that pages are migratable but
8028 	 * the later is not the case right now unfortunatelly. E.g. movablecore
8029 	 * can still lead to having bootmem allocations in zone_movable.
8030 	 */
8031 
8032 	if (is_migrate_cma_page(page)) {
8033 		/*
8034 		 * CMA allocations (alloc_contig_range) really need to mark
8035 		 * isolate CMA pageblocks even when they are not movable in fact
8036 		 * so consider them movable here.
8037 		 */
8038 		if (is_migrate_cma(migratetype))
8039 			return false;
8040 
8041 		reason = "CMA page";
8042 		goto unmovable;
8043 	}
8044 
8045 	for (found = 0; iter < pageblock_nr_pages; iter++) {
8046 		unsigned long check = pfn + iter;
8047 
8048 		if (!pfn_valid_within(check))
8049 			continue;
8050 
8051 		page = pfn_to_page(check);
8052 
8053 		if (PageReserved(page))
8054 			goto unmovable;
8055 
8056 		/*
8057 		 * If the zone is movable and we have ruled out all reserved
8058 		 * pages then it should be reasonably safe to assume the rest
8059 		 * is movable.
8060 		 */
8061 		if (zone_idx(zone) == ZONE_MOVABLE)
8062 			continue;
8063 
8064 		/*
8065 		 * Hugepages are not in LRU lists, but they're movable.
8066 		 * We need not scan over tail pages because we don't
8067 		 * handle each tail page individually in migration.
8068 		 */
8069 		if (PageHuge(page)) {
8070 			struct page *head = compound_head(page);
8071 			unsigned int skip_pages;
8072 
8073 			if (!hugepage_migration_supported(page_hstate(head)))
8074 				goto unmovable;
8075 
8076 			skip_pages = (1 << compound_order(head)) - (page - head);
8077 			iter += skip_pages - 1;
8078 			continue;
8079 		}
8080 
8081 		/*
8082 		 * We can't use page_count without pin a page
8083 		 * because another CPU can free compound page.
8084 		 * This check already skips compound tails of THP
8085 		 * because their page->_refcount is zero at all time.
8086 		 */
8087 		if (!page_ref_count(page)) {
8088 			if (PageBuddy(page))
8089 				iter += (1 << page_order(page)) - 1;
8090 			continue;
8091 		}
8092 
8093 		/*
8094 		 * The HWPoisoned page may be not in buddy system, and
8095 		 * page_count() is not 0.
8096 		 */
8097 		if ((flags & SKIP_HWPOISON) && PageHWPoison(page))
8098 			continue;
8099 
8100 		if (__PageMovable(page))
8101 			continue;
8102 
8103 		if (!PageLRU(page))
8104 			found++;
8105 		/*
8106 		 * If there are RECLAIMABLE pages, we need to check
8107 		 * it.  But now, memory offline itself doesn't call
8108 		 * shrink_node_slabs() and it still to be fixed.
8109 		 */
8110 		/*
8111 		 * If the page is not RAM, page_count()should be 0.
8112 		 * we don't need more check. This is an _used_ not-movable page.
8113 		 *
8114 		 * The problematic thing here is PG_reserved pages. PG_reserved
8115 		 * is set to both of a memory hole page and a _used_ kernel
8116 		 * page at boot.
8117 		 */
8118 		if (found > count)
8119 			goto unmovable;
8120 	}
8121 	return false;
8122 unmovable:
8123 	WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE);
8124 	if (flags & REPORT_FAILURE)
8125 		dump_page(pfn_to_page(pfn + iter), reason);
8126 	return true;
8127 }
8128 
8129 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
8130 
8131 static unsigned long pfn_max_align_down(unsigned long pfn)
8132 {
8133 	return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
8134 			     pageblock_nr_pages) - 1);
8135 }
8136 
8137 static unsigned long pfn_max_align_up(unsigned long pfn)
8138 {
8139 	return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
8140 				pageblock_nr_pages));
8141 }
8142 
8143 /* [start, end) must belong to a single zone. */
8144 static int __alloc_contig_migrate_range(struct compact_control *cc,
8145 					unsigned long start, unsigned long end)
8146 {
8147 	/* This function is based on compact_zone() from compaction.c. */
8148 	unsigned long nr_reclaimed;
8149 	unsigned long pfn = start;
8150 	unsigned int tries = 0;
8151 	int ret = 0;
8152 
8153 	migrate_prep();
8154 
8155 	while (pfn < end || !list_empty(&cc->migratepages)) {
8156 		if (fatal_signal_pending(current)) {
8157 			ret = -EINTR;
8158 			break;
8159 		}
8160 
8161 		if (list_empty(&cc->migratepages)) {
8162 			cc->nr_migratepages = 0;
8163 			pfn = isolate_migratepages_range(cc, pfn, end);
8164 			if (!pfn) {
8165 				ret = -EINTR;
8166 				break;
8167 			}
8168 			tries = 0;
8169 		} else if (++tries == 5) {
8170 			ret = ret < 0 ? ret : -EBUSY;
8171 			break;
8172 		}
8173 
8174 		nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
8175 							&cc->migratepages);
8176 		cc->nr_migratepages -= nr_reclaimed;
8177 
8178 		ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
8179 				    NULL, 0, cc->mode, MR_CONTIG_RANGE);
8180 	}
8181 	if (ret < 0) {
8182 		putback_movable_pages(&cc->migratepages);
8183 		return ret;
8184 	}
8185 	return 0;
8186 }
8187 
8188 /**
8189  * alloc_contig_range() -- tries to allocate given range of pages
8190  * @start:	start PFN to allocate
8191  * @end:	one-past-the-last PFN to allocate
8192  * @migratetype:	migratetype of the underlaying pageblocks (either
8193  *			#MIGRATE_MOVABLE or #MIGRATE_CMA).  All pageblocks
8194  *			in range must have the same migratetype and it must
8195  *			be either of the two.
8196  * @gfp_mask:	GFP mask to use during compaction
8197  *
8198  * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
8199  * aligned.  The PFN range must belong to a single zone.
8200  *
8201  * The first thing this routine does is attempt to MIGRATE_ISOLATE all
8202  * pageblocks in the range.  Once isolated, the pageblocks should not
8203  * be modified by others.
8204  *
8205  * Return: zero on success or negative error code.  On success all
8206  * pages which PFN is in [start, end) are allocated for the caller and
8207  * need to be freed with free_contig_range().
8208  */
8209 int alloc_contig_range(unsigned long start, unsigned long end,
8210 		       unsigned migratetype, gfp_t gfp_mask)
8211 {
8212 	unsigned long outer_start, outer_end;
8213 	unsigned int order;
8214 	int ret = 0;
8215 
8216 	struct compact_control cc = {
8217 		.nr_migratepages = 0,
8218 		.order = -1,
8219 		.zone = page_zone(pfn_to_page(start)),
8220 		.mode = MIGRATE_SYNC,
8221 		.ignore_skip_hint = true,
8222 		.no_set_skip_hint = true,
8223 		.gfp_mask = current_gfp_context(gfp_mask),
8224 	};
8225 	INIT_LIST_HEAD(&cc.migratepages);
8226 
8227 	/*
8228 	 * What we do here is we mark all pageblocks in range as
8229 	 * MIGRATE_ISOLATE.  Because pageblock and max order pages may
8230 	 * have different sizes, and due to the way page allocator
8231 	 * work, we align the range to biggest of the two pages so
8232 	 * that page allocator won't try to merge buddies from
8233 	 * different pageblocks and change MIGRATE_ISOLATE to some
8234 	 * other migration type.
8235 	 *
8236 	 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
8237 	 * migrate the pages from an unaligned range (ie. pages that
8238 	 * we are interested in).  This will put all the pages in
8239 	 * range back to page allocator as MIGRATE_ISOLATE.
8240 	 *
8241 	 * When this is done, we take the pages in range from page
8242 	 * allocator removing them from the buddy system.  This way
8243 	 * page allocator will never consider using them.
8244 	 *
8245 	 * This lets us mark the pageblocks back as
8246 	 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
8247 	 * aligned range but not in the unaligned, original range are
8248 	 * put back to page allocator so that buddy can use them.
8249 	 */
8250 
8251 	ret = start_isolate_page_range(pfn_max_align_down(start),
8252 				       pfn_max_align_up(end), migratetype, 0);
8253 	if (ret < 0)
8254 		return ret;
8255 
8256 	/*
8257 	 * In case of -EBUSY, we'd like to know which page causes problem.
8258 	 * So, just fall through. test_pages_isolated() has a tracepoint
8259 	 * which will report the busy page.
8260 	 *
8261 	 * It is possible that busy pages could become available before
8262 	 * the call to test_pages_isolated, and the range will actually be
8263 	 * allocated.  So, if we fall through be sure to clear ret so that
8264 	 * -EBUSY is not accidentally used or returned to caller.
8265 	 */
8266 	ret = __alloc_contig_migrate_range(&cc, start, end);
8267 	if (ret && ret != -EBUSY)
8268 		goto done;
8269 	ret =0;
8270 
8271 	/*
8272 	 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
8273 	 * aligned blocks that are marked as MIGRATE_ISOLATE.  What's
8274 	 * more, all pages in [start, end) are free in page allocator.
8275 	 * What we are going to do is to allocate all pages from
8276 	 * [start, end) (that is remove them from page allocator).
8277 	 *
8278 	 * The only problem is that pages at the beginning and at the
8279 	 * end of interesting range may be not aligned with pages that
8280 	 * page allocator holds, ie. they can be part of higher order
8281 	 * pages.  Because of this, we reserve the bigger range and
8282 	 * once this is done free the pages we are not interested in.
8283 	 *
8284 	 * We don't have to hold zone->lock here because the pages are
8285 	 * isolated thus they won't get removed from buddy.
8286 	 */
8287 
8288 	lru_add_drain_all();
8289 
8290 	order = 0;
8291 	outer_start = start;
8292 	while (!PageBuddy(pfn_to_page(outer_start))) {
8293 		if (++order >= MAX_ORDER) {
8294 			outer_start = start;
8295 			break;
8296 		}
8297 		outer_start &= ~0UL << order;
8298 	}
8299 
8300 	if (outer_start != start) {
8301 		order = page_order(pfn_to_page(outer_start));
8302 
8303 		/*
8304 		 * outer_start page could be small order buddy page and
8305 		 * it doesn't include start page. Adjust outer_start
8306 		 * in this case to report failed page properly
8307 		 * on tracepoint in test_pages_isolated()
8308 		 */
8309 		if (outer_start + (1UL << order) <= start)
8310 			outer_start = start;
8311 	}
8312 
8313 	/* Make sure the range is really isolated. */
8314 	if (test_pages_isolated(outer_start, end, false)) {
8315 		pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n",
8316 			__func__, outer_start, end);
8317 		ret = -EBUSY;
8318 		goto done;
8319 	}
8320 
8321 	/* Grab isolated pages from freelists. */
8322 	outer_end = isolate_freepages_range(&cc, outer_start, end);
8323 	if (!outer_end) {
8324 		ret = -EBUSY;
8325 		goto done;
8326 	}
8327 
8328 	/* Free head and tail (if any) */
8329 	if (start != outer_start)
8330 		free_contig_range(outer_start, start - outer_start);
8331 	if (end != outer_end)
8332 		free_contig_range(end, outer_end - end);
8333 
8334 done:
8335 	undo_isolate_page_range(pfn_max_align_down(start),
8336 				pfn_max_align_up(end), migratetype);
8337 	return ret;
8338 }
8339 
8340 void free_contig_range(unsigned long pfn, unsigned nr_pages)
8341 {
8342 	unsigned int count = 0;
8343 
8344 	for (; nr_pages--; pfn++) {
8345 		struct page *page = pfn_to_page(pfn);
8346 
8347 		count += page_count(page) != 1;
8348 		__free_page(page);
8349 	}
8350 	WARN(count != 0, "%d pages are still in use!\n", count);
8351 }
8352 #endif
8353 
8354 #ifdef CONFIG_MEMORY_HOTPLUG
8355 /*
8356  * The zone indicated has a new number of managed_pages; batch sizes and percpu
8357  * page high values need to be recalulated.
8358  */
8359 void __meminit zone_pcp_update(struct zone *zone)
8360 {
8361 	unsigned cpu;
8362 	mutex_lock(&pcp_batch_high_lock);
8363 	for_each_possible_cpu(cpu)
8364 		pageset_set_high_and_batch(zone,
8365 				per_cpu_ptr(zone->pageset, cpu));
8366 	mutex_unlock(&pcp_batch_high_lock);
8367 }
8368 #endif
8369 
8370 void zone_pcp_reset(struct zone *zone)
8371 {
8372 	unsigned long flags;
8373 	int cpu;
8374 	struct per_cpu_pageset *pset;
8375 
8376 	/* avoid races with drain_pages()  */
8377 	local_irq_save(flags);
8378 	if (zone->pageset != &boot_pageset) {
8379 		for_each_online_cpu(cpu) {
8380 			pset = per_cpu_ptr(zone->pageset, cpu);
8381 			drain_zonestat(zone, pset);
8382 		}
8383 		free_percpu(zone->pageset);
8384 		zone->pageset = &boot_pageset;
8385 	}
8386 	local_irq_restore(flags);
8387 }
8388 
8389 #ifdef CONFIG_MEMORY_HOTREMOVE
8390 /*
8391  * All pages in the range must be in a single zone and isolated
8392  * before calling this.
8393  */
8394 void
8395 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
8396 {
8397 	struct page *page;
8398 	struct zone *zone;
8399 	unsigned int order, i;
8400 	unsigned long pfn;
8401 	unsigned long flags;
8402 	/* find the first valid pfn */
8403 	for (pfn = start_pfn; pfn < end_pfn; pfn++)
8404 		if (pfn_valid(pfn))
8405 			break;
8406 	if (pfn == end_pfn)
8407 		return;
8408 	offline_mem_sections(pfn, end_pfn);
8409 	zone = page_zone(pfn_to_page(pfn));
8410 	spin_lock_irqsave(&zone->lock, flags);
8411 	pfn = start_pfn;
8412 	while (pfn < end_pfn) {
8413 		if (!pfn_valid(pfn)) {
8414 			pfn++;
8415 			continue;
8416 		}
8417 		page = pfn_to_page(pfn);
8418 		/*
8419 		 * The HWPoisoned page may be not in buddy system, and
8420 		 * page_count() is not 0.
8421 		 */
8422 		if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
8423 			pfn++;
8424 			SetPageReserved(page);
8425 			continue;
8426 		}
8427 
8428 		BUG_ON(page_count(page));
8429 		BUG_ON(!PageBuddy(page));
8430 		order = page_order(page);
8431 #ifdef CONFIG_DEBUG_VM
8432 		pr_info("remove from free list %lx %d %lx\n",
8433 			pfn, 1 << order, end_pfn);
8434 #endif
8435 		list_del(&page->lru);
8436 		rmv_page_order(page);
8437 		zone->free_area[order].nr_free--;
8438 		for (i = 0; i < (1 << order); i++)
8439 			SetPageReserved((page+i));
8440 		pfn += (1 << order);
8441 	}
8442 	spin_unlock_irqrestore(&zone->lock, flags);
8443 }
8444 #endif
8445 
8446 bool is_free_buddy_page(struct page *page)
8447 {
8448 	struct zone *zone = page_zone(page);
8449 	unsigned long pfn = page_to_pfn(page);
8450 	unsigned long flags;
8451 	unsigned int order;
8452 
8453 	spin_lock_irqsave(&zone->lock, flags);
8454 	for (order = 0; order < MAX_ORDER; order++) {
8455 		struct page *page_head = page - (pfn & ((1 << order) - 1));
8456 
8457 		if (PageBuddy(page_head) && page_order(page_head) >= order)
8458 			break;
8459 	}
8460 	spin_unlock_irqrestore(&zone->lock, flags);
8461 
8462 	return order < MAX_ORDER;
8463 }
8464 
8465 #ifdef CONFIG_MEMORY_FAILURE
8466 /*
8467  * Set PG_hwpoison flag if a given page is confirmed to be a free page.  This
8468  * test is performed under the zone lock to prevent a race against page
8469  * allocation.
8470  */
8471 bool set_hwpoison_free_buddy_page(struct page *page)
8472 {
8473 	struct zone *zone = page_zone(page);
8474 	unsigned long pfn = page_to_pfn(page);
8475 	unsigned long flags;
8476 	unsigned int order;
8477 	bool hwpoisoned = false;
8478 
8479 	spin_lock_irqsave(&zone->lock, flags);
8480 	for (order = 0; order < MAX_ORDER; order++) {
8481 		struct page *page_head = page - (pfn & ((1 << order) - 1));
8482 
8483 		if (PageBuddy(page_head) && page_order(page_head) >= order) {
8484 			if (!TestSetPageHWPoison(page))
8485 				hwpoisoned = true;
8486 			break;
8487 		}
8488 	}
8489 	spin_unlock_irqrestore(&zone->lock, flags);
8490 
8491 	return hwpoisoned;
8492 }
8493 #endif
8494